Index: head/sys/cam/ctl/ctl.c =================================================================== --- head/sys/cam/ctl/ctl.c (revision 333445) +++ head/sys/cam/ctl/ctl.c (revision 333446) @@ -1,13592 +1,13613 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003-2009 Silicon Graphics International Corp. * Copyright (c) 2012 The FreeBSD Foundation * Copyright (c) 2014-2017 Alexander Motin + * Copyright (c) 2017 Jakub Wojciech Klama + * Copyright (c) 2018 Marcelo Araujo * All rights reserved. * * Portions of this software were developed by Edward Tomasz Napierala * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id$ */ /* * CAM Target Layer, a SCSI device emulation subsystem. * * Author: Ken Merry */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct ctl_softc *control_softc = NULL; /* * Template mode pages. */ /* * Note that these are default values only. The actual values will be * filled in when the user does a mode sense. */ const static struct scsi_da_rw_recovery_page rw_er_page_default = { /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE, /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2, /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE, /*read_retry_count*/0, /*correction_span*/0, /*head_offset_count*/0, /*data_strobe_offset_cnt*/0, /*byte8*/SMS_RWER_LBPERE, /*write_retry_count*/0, /*reserved2*/0, /*recovery_time_limit*/{0, 0}, }; const static struct scsi_da_rw_recovery_page rw_er_page_changeable = { /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE, /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2, /*byte3*/SMS_RWER_PER, /*read_retry_count*/0, /*correction_span*/0, /*head_offset_count*/0, /*data_strobe_offset_cnt*/0, /*byte8*/SMS_RWER_LBPERE, /*write_retry_count*/0, /*reserved2*/0, /*recovery_time_limit*/{0, 0}, }; const static struct scsi_format_page format_page_default = { /*page_code*/SMS_FORMAT_DEVICE_PAGE, /*page_length*/sizeof(struct scsi_format_page) - 2, /*tracks_per_zone*/ {0, 0}, /*alt_sectors_per_zone*/ {0, 0}, /*alt_tracks_per_zone*/ {0, 0}, /*alt_tracks_per_lun*/ {0, 0}, /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff, CTL_DEFAULT_SECTORS_PER_TRACK & 0xff}, /*bytes_per_sector*/ {0, 0}, /*interleave*/ {0, 0}, /*track_skew*/ {0, 0}, /*cylinder_skew*/ {0, 0}, /*flags*/ SFP_HSEC, /*reserved*/ {0, 0, 0} }; const static struct scsi_format_page format_page_changeable = { /*page_code*/SMS_FORMAT_DEVICE_PAGE, /*page_length*/sizeof(struct scsi_format_page) - 2, /*tracks_per_zone*/ {0, 0}, /*alt_sectors_per_zone*/ {0, 0}, /*alt_tracks_per_zone*/ {0, 0}, /*alt_tracks_per_lun*/ {0, 0}, /*sectors_per_track*/ {0, 0}, /*bytes_per_sector*/ {0, 0}, /*interleave*/ {0, 0}, /*track_skew*/ {0, 0}, /*cylinder_skew*/ {0, 0}, /*flags*/ 0, /*reserved*/ {0, 0, 0} }; const static struct scsi_rigid_disk_page rigid_disk_page_default = { /*page_code*/SMS_RIGID_DISK_PAGE, /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2, /*cylinders*/ {0, 0, 0}, /*heads*/ CTL_DEFAULT_HEADS, /*start_write_precomp*/ {0, 0, 0}, /*start_reduced_current*/ {0, 0, 0}, /*step_rate*/ {0, 0}, /*landing_zone_cylinder*/ {0, 0, 0}, /*rpl*/ SRDP_RPL_DISABLED, /*rotational_offset*/ 0, /*reserved1*/ 0, /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff, CTL_DEFAULT_ROTATION_RATE & 0xff}, /*reserved2*/ {0, 0} }; const static struct scsi_rigid_disk_page rigid_disk_page_changeable = { /*page_code*/SMS_RIGID_DISK_PAGE, /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2, /*cylinders*/ {0, 0, 0}, /*heads*/ 0, /*start_write_precomp*/ {0, 0, 0}, /*start_reduced_current*/ {0, 0, 0}, /*step_rate*/ {0, 0}, /*landing_zone_cylinder*/ {0, 0, 0}, /*rpl*/ 0, /*rotational_offset*/ 0, /*reserved1*/ 0, /*rotation_rate*/ {0, 0}, /*reserved2*/ {0, 0} }; const static struct scsi_da_verify_recovery_page verify_er_page_default = { /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE, /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2, /*byte3*/0, /*read_retry_count*/0, /*reserved*/{ 0, 0, 0, 0, 0, 0 }, /*recovery_time_limit*/{0, 0}, }; const static struct scsi_da_verify_recovery_page verify_er_page_changeable = { /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE, /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2, /*byte3*/SMS_VER_PER, /*read_retry_count*/0, /*reserved*/{ 0, 0, 0, 0, 0, 0 }, /*recovery_time_limit*/{0, 0}, }; const static struct scsi_caching_page caching_page_default = { /*page_code*/SMS_CACHING_PAGE, /*page_length*/sizeof(struct scsi_caching_page) - 2, /*flags1*/ SCP_DISC | SCP_WCE, /*ret_priority*/ 0, /*disable_pf_transfer_len*/ {0xff, 0xff}, /*min_prefetch*/ {0, 0}, /*max_prefetch*/ {0xff, 0xff}, /*max_pf_ceiling*/ {0xff, 0xff}, /*flags2*/ 0, /*cache_segments*/ 0, /*cache_seg_size*/ {0, 0}, /*reserved*/ 0, /*non_cache_seg_size*/ {0, 0, 0} }; const static struct scsi_caching_page caching_page_changeable = { /*page_code*/SMS_CACHING_PAGE, /*page_length*/sizeof(struct scsi_caching_page) - 2, /*flags1*/ SCP_WCE | SCP_RCD, /*ret_priority*/ 0, /*disable_pf_transfer_len*/ {0, 0}, /*min_prefetch*/ {0, 0}, /*max_prefetch*/ {0, 0}, /*max_pf_ceiling*/ {0, 0}, /*flags2*/ 0, /*cache_segments*/ 0, /*cache_seg_size*/ {0, 0}, /*reserved*/ 0, /*non_cache_seg_size*/ {0, 0, 0} }; const static struct scsi_control_page control_page_default = { /*page_code*/SMS_CONTROL_MODE_PAGE, /*page_length*/sizeof(struct scsi_control_page) - 2, /*rlec*/0, /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED, /*eca_and_aen*/0, /*flags4*/SCP_TAS, /*aen_holdoff_period*/{0, 0}, /*busy_timeout_period*/{0, 0}, /*extended_selftest_completion_time*/{0, 0} }; const static struct scsi_control_page control_page_changeable = { /*page_code*/SMS_CONTROL_MODE_PAGE, /*page_length*/sizeof(struct scsi_control_page) - 2, /*rlec*/SCP_DSENSE, /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR, /*eca_and_aen*/SCP_SWP, /*flags4*/0, /*aen_holdoff_period*/{0, 0}, /*busy_timeout_period*/{0, 0}, /*extended_selftest_completion_time*/{0, 0} }; #define CTL_CEM_LEN (sizeof(struct scsi_control_ext_page) - 4) const static struct scsi_control_ext_page control_ext_page_default = { /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF, /*subpage_code*/0x01, /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN}, /*flags*/0, /*prio*/0, /*max_sense*/0 }; const static struct scsi_control_ext_page control_ext_page_changeable = { /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF, /*subpage_code*/0x01, /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN}, /*flags*/0, /*prio*/0, /*max_sense*/0xff }; const static struct scsi_info_exceptions_page ie_page_default = { /*page_code*/SMS_INFO_EXCEPTIONS_PAGE, /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2, /*info_flags*/SIEP_FLAGS_EWASC, /*mrie*/SIEP_MRIE_NO, /*interval_timer*/{0, 0, 0, 0}, /*report_count*/{0, 0, 0, 1} }; const static struct scsi_info_exceptions_page ie_page_changeable = { /*page_code*/SMS_INFO_EXCEPTIONS_PAGE, /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2, /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST | SIEP_FLAGS_LOGERR, /*mrie*/0x0f, /*interval_timer*/{0xff, 0xff, 0xff, 0xff}, /*report_count*/{0xff, 0xff, 0xff, 0xff} }; #define CTL_LBPM_LEN (sizeof(struct ctl_logical_block_provisioning_page) - 4) const static struct ctl_logical_block_provisioning_page lbp_page_default = {{ /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF, /*subpage_code*/0x02, /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN}, /*flags*/0, /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, /*descr*/{}}, {{/*flags*/0, /*resource*/0x01, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}}, {/*flags*/0, /*resource*/0x02, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}}, {/*flags*/0, /*resource*/0xf1, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}}, {/*flags*/0, /*resource*/0xf2, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}} } }; const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{ /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF, /*subpage_code*/0x02, /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN}, /*flags*/SLBPP_SITUA, /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, /*descr*/{}}, {{/*flags*/0, /*resource*/0, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}}, {/*flags*/0, /*resource*/0, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}}, {/*flags*/0, /*resource*/0, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}}, {/*flags*/0, /*resource*/0, /*reserved*/{0, 0}, /*count*/{0, 0, 0, 0}} } }; const static struct scsi_cddvd_capabilities_page cddvd_page_default = { /*page_code*/SMS_CDDVD_CAPS_PAGE, /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2, /*caps1*/0x3f, /*caps2*/0x00, /*caps3*/0xf0, /*caps4*/0x00, /*caps5*/0x29, /*caps6*/0x00, /*obsolete*/{0, 0}, /*nvol_levels*/{0, 0}, /*buffer_size*/{8, 0}, /*obsolete2*/{0, 0}, /*reserved*/0, /*digital*/0, /*obsolete3*/0, /*copy_management*/0, /*reserved2*/0, /*rotation_control*/0, /*cur_write_speed*/0, /*num_speed_descr*/0, }; const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = { /*page_code*/SMS_CDDVD_CAPS_PAGE, /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2, /*caps1*/0, /*caps2*/0, /*caps3*/0, /*caps4*/0, /*caps5*/0, /*caps6*/0, /*obsolete*/{0, 0}, /*nvol_levels*/{0, 0}, /*buffer_size*/{0, 0}, /*obsolete2*/{0, 0}, /*reserved*/0, /*digital*/0, /*obsolete3*/0, /*copy_management*/0, /*reserved2*/0, /*rotation_control*/0, /*cur_write_speed*/0, /*num_speed_descr*/0, }; SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer"); static int worker_threads = -1; SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN, &worker_threads, 1, "Number of worker threads"); static int ctl_debug = CTL_DEBUG_NONE; SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN, &ctl_debug, 0, "Enabled debug flags"); static int ctl_lun_map_size = 1024; SYSCTL_INT(_kern_cam_ctl, OID_AUTO, lun_map_size, CTLFLAG_RWTUN, &ctl_lun_map_size, 0, "Size of per-port LUN map (max LUN + 1)"); #ifdef CTL_TIME_IO static int ctl_time_io_secs = CTL_TIME_IO_DEFAULT_SECS; SYSCTL_INT(_kern_cam_ctl, OID_AUTO, time_io_secs, CTLFLAG_RWTUN, &ctl_time_io_secs, 0, "Log requests taking more seconds"); #endif /* * Maximum number of LUNs we support. MUST be a power of 2. */ #define CTL_DEFAULT_MAX_LUNS 1024 static int ctl_max_luns = CTL_DEFAULT_MAX_LUNS; TUNABLE_INT("kern.cam.ctl.max_luns", &ctl_max_luns); SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_luns, CTLFLAG_RDTUN, &ctl_max_luns, CTL_DEFAULT_MAX_LUNS, "Maximum number of LUNs"); /* * Maximum number of ports registered at one time. */ #define CTL_DEFAULT_MAX_PORTS 256 static int ctl_max_ports = CTL_DEFAULT_MAX_PORTS; TUNABLE_INT("kern.cam.ctl.max_ports", &ctl_max_ports); SYSCTL_INT(_kern_cam_ctl, OID_AUTO, max_ports, CTLFLAG_RDTUN, &ctl_max_ports, CTL_DEFAULT_MAX_LUNS, "Maximum number of ports"); /* * Maximum number of initiators we support. */ #define CTL_MAX_INITIATORS (CTL_MAX_INIT_PER_PORT * ctl_max_ports) /* * Supported pages (0x00), Serial number (0x80), Device ID (0x83), * Extended INQUIRY Data (0x86), Mode Page Policy (0x87), * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0), * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2) */ #define SCSI_EVPD_NUM_SUPPORTED_PAGES 10 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event, int param); static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest); static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest); static int ctl_init(void); static int ctl_shutdown(void); static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td); static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td); static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio); static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num, struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries); static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun, struct ctl_be_lun *be_lun); static int ctl_free_lun(struct ctl_lun *lun); static void ctl_create_lun(struct ctl_be_lun *be_lun); static int ctl_do_mode_select(union ctl_io *io); static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key, uint64_t sa_res_key, uint8_t type, uint32_t residx, struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb, struct scsi_per_res_out_parms* param); static void ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg); static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io); static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len); static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio); static int ctl_inquiry_std(struct ctl_scsiio *ctsio); static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len); static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq); static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2); static ctl_action ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io, union ctl_io *ooa_io); static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io, union ctl_io *starting_io); static int ctl_check_blocked(struct ctl_lun *lun); static int ctl_scsiio_lun_check(struct ctl_lun *lun, const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio); static void ctl_failover_lun(union ctl_io *io); static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio); static int ctl_scsiio(struct ctl_scsiio *ctsio); static int ctl_target_reset(union ctl_io *io); static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type); static int ctl_lun_reset(union ctl_io *io); static int ctl_abort_task(union ctl_io *io); static int ctl_abort_task_set(union ctl_io *io); static int ctl_query_task(union ctl_io *io, int task_set); static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx, ctl_ua_type ua_type); static int ctl_i_t_nexus_reset(union ctl_io *io); static int ctl_query_async_event(union ctl_io *io); static void ctl_run_task(union ctl_io *io); #ifdef CTL_IO_DELAY static void ctl_datamove_timer_wakeup(void *arg); static void ctl_done_timer_wakeup(void *arg); #endif /* CTL_IO_DELAY */ static void ctl_send_datamove_done(union ctl_io *io, int have_lock); static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq); static int ctl_datamove_remote_dm_write_cb(union ctl_io *io); static void ctl_datamove_remote_write(union ctl_io *io); static int ctl_datamove_remote_dm_read_cb(union ctl_io *io); static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq); static int ctl_datamove_remote_sgl_setup(union ctl_io *io); static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command, ctl_ha_dt_cb callback); static void ctl_datamove_remote_read(union ctl_io *io); static void ctl_datamove_remote(union ctl_io *io); static void ctl_process_done(union ctl_io *io); static void ctl_lun_thread(void *arg); static void ctl_thresh_thread(void *arg); static void ctl_work_thread(void *arg); static void ctl_enqueue_incoming(union ctl_io *io); static void ctl_enqueue_rtr(union ctl_io *io); static void ctl_enqueue_done(union ctl_io *io); static void ctl_enqueue_isc(union ctl_io *io); static const struct ctl_cmd_entry * ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa); static const struct ctl_cmd_entry * ctl_validate_command(struct ctl_scsiio *ctsio); static int ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry); static int ctl_ha_init(void); static int ctl_ha_shutdown(void); static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx); static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx); static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx); static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key); /* * Load the serialization table. This isn't very pretty, but is probably * the easiest way to do it. */ #include "ctl_ser_table.c" /* * We only need to define open, close and ioctl routines for this driver. */ static struct cdevsw ctl_cdevsw = { .d_version = D_VERSION, .d_flags = 0, .d_open = ctl_open, .d_close = ctl_close, .d_ioctl = ctl_ioctl, .d_name = "ctl", }; MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL"); static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *); static moduledata_t ctl_moduledata = { "ctl", ctl_module_event_handler, NULL }; DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD); MODULE_VERSION(ctl, 1); static struct ctl_frontend ha_frontend = { .name = "ha", .init = ctl_ha_init, .shutdown = ctl_ha_shutdown, }; static int ctl_ha_init(void) { struct ctl_softc *softc = control_softc; if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC, &softc->othersc_pool) != 0) return (ENOMEM); if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) { ctl_pool_free(softc->othersc_pool); return (EIO); } if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler) != CTL_HA_STATUS_SUCCESS) { ctl_ha_msg_destroy(softc); ctl_pool_free(softc->othersc_pool); return (EIO); } return (0); }; static int ctl_ha_shutdown(void) { struct ctl_softc *softc = control_softc; struct ctl_port *port; ctl_ha_msg_shutdown(softc); if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL) != CTL_HA_STATUS_SUCCESS) return (EIO); if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS) return (EIO); ctl_pool_free(softc->othersc_pool); while ((port = STAILQ_FIRST(&ha_frontend.port_list)) != NULL) { ctl_port_deregister(port); free(port->port_name, M_CTL); free(port, M_CTL); } return (0); }; static void ctl_ha_datamove(union ctl_io *io) { struct ctl_lun *lun = CTL_LUN(io); struct ctl_sg_entry *sgl; union ctl_ha_msg msg; uint32_t sg_entries_sent; int do_sg_copy, i, j; memset(&msg.dt, 0, sizeof(msg.dt)); msg.hdr.msg_type = CTL_MSG_DATAMOVE; msg.hdr.original_sc = io->io_hdr.original_sc; msg.hdr.serializing_sc = io; msg.hdr.nexus = io->io_hdr.nexus; msg.hdr.status = io->io_hdr.status; msg.dt.flags = io->io_hdr.flags; /* * We convert everything into a S/G list here. We can't * pass by reference, only by value between controllers. * So we can't pass a pointer to the S/G list, only as many * S/G entries as we can fit in here. If it's possible for * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries, * then we need to break this up into multiple transfers. */ if (io->scsiio.kern_sg_entries == 0) { msg.dt.kern_sg_entries = 1; #if 0 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) { msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr; } else { /* XXX KDM use busdma here! */ msg.dt.sg_list[0].addr = (void *)vtophys(io->scsiio.kern_data_ptr); } #else KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0, ("HA does not support BUS_ADDR")); msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr; #endif msg.dt.sg_list[0].len = io->scsiio.kern_data_len; do_sg_copy = 0; } else { msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries; do_sg_copy = 1; } msg.dt.kern_data_len = io->scsiio.kern_data_len; msg.dt.kern_total_len = io->scsiio.kern_total_len; msg.dt.kern_data_resid = io->scsiio.kern_data_resid; msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset; msg.dt.sg_sequence = 0; /* * Loop until we've sent all of the S/G entries. On the * other end, we'll recompose these S/G entries into one * contiguous list before processing. */ for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries; msg.dt.sg_sequence++) { msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) / sizeof(msg.dt.sg_list[0])), msg.dt.kern_sg_entries - sg_entries_sent); if (do_sg_copy != 0) { sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; for (i = sg_entries_sent, j = 0; i < msg.dt.cur_sg_entries; i++, j++) { #if 0 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) { msg.dt.sg_list[j].addr = sgl[i].addr; } else { /* XXX KDM use busdma here! */ msg.dt.sg_list[j].addr = (void *)vtophys(sgl[i].addr); } #else KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0, ("HA does not support BUS_ADDR")); msg.dt.sg_list[j].addr = sgl[i].addr; #endif msg.dt.sg_list[j].len = sgl[i].len; } } sg_entries_sent += msg.dt.cur_sg_entries; msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries); if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.dt) - sizeof(msg.dt.sg_list) + sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries, M_WAITOK) > CTL_HA_STATUS_SUCCESS) { io->io_hdr.port_status = 31341; io->scsiio.be_move_done(io); return; } msg.dt.sent_sg_entries = sg_entries_sent; } /* * Officially handover the request from us to peer. * If failover has just happened, then we must return error. * If failover happen just after, then it is not our problem. */ if (lun) mtx_lock(&lun->lun_lock); if (io->io_hdr.flags & CTL_FLAG_FAILOVER) { if (lun) mtx_unlock(&lun->lun_lock); io->io_hdr.port_status = 31342; io->scsiio.be_move_done(io); return; } io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; io->io_hdr.flags |= CTL_FLAG_DMA_INPROG; if (lun) mtx_unlock(&lun->lun_lock); } static void ctl_ha_done(union ctl_io *io) { union ctl_ha_msg msg; if (io->io_hdr.io_type == CTL_IO_SCSI) { memset(&msg, 0, sizeof(msg)); msg.hdr.msg_type = CTL_MSG_FINISH_IO; msg.hdr.original_sc = io->io_hdr.original_sc; msg.hdr.nexus = io->io_hdr.nexus; msg.hdr.status = io->io_hdr.status; msg.scsi.scsi_status = io->scsiio.scsi_status; msg.scsi.tag_num = io->scsiio.tag_num; msg.scsi.tag_type = io->scsiio.tag_type; msg.scsi.sense_len = io->scsiio.sense_len; memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data, io->scsiio.sense_len); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) + msg.scsi.sense_len, M_WAITOK); } ctl_free_io(io); } static void ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc, union ctl_ha_msg *msg_info) { struct ctl_scsiio *ctsio; if (msg_info->hdr.original_sc == NULL) { printf("%s: original_sc == NULL!\n", __func__); /* XXX KDM now what? */ return; } ctsio = &msg_info->hdr.original_sc->scsiio; ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE; ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO; ctsio->io_hdr.status = msg_info->hdr.status; ctsio->scsi_status = msg_info->scsi.scsi_status; ctsio->sense_len = msg_info->scsi.sense_len; memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data, msg_info->scsi.sense_len); ctl_enqueue_isc((union ctl_io *)ctsio); } static void ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc, union ctl_ha_msg *msg_info) { struct ctl_scsiio *ctsio; if (msg_info->hdr.serializing_sc == NULL) { printf("%s: serializing_sc == NULL!\n", __func__); /* XXX KDM now what? */ return; } ctsio = &msg_info->hdr.serializing_sc->scsiio; ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO; ctl_enqueue_isc((union ctl_io *)ctsio); } void ctl_isc_announce_lun(struct ctl_lun *lun) { struct ctl_softc *softc = lun->ctl_softc; union ctl_ha_msg *msg; struct ctl_ha_msg_lun_pr_key pr_key; int i, k; if (softc->ha_link != CTL_HA_LINK_ONLINE) return; mtx_lock(&lun->lun_lock); i = sizeof(msg->lun); if (lun->lun_devid) i += lun->lun_devid->len; i += sizeof(pr_key) * lun->pr_key_count; alloc: mtx_unlock(&lun->lun_lock); msg = malloc(i, M_CTL, M_WAITOK); mtx_lock(&lun->lun_lock); k = sizeof(msg->lun); if (lun->lun_devid) k += lun->lun_devid->len; k += sizeof(pr_key) * lun->pr_key_count; if (i < k) { free(msg, M_CTL); i = k; goto alloc; } bzero(&msg->lun, sizeof(msg->lun)); msg->hdr.msg_type = CTL_MSG_LUN_SYNC; msg->hdr.nexus.targ_lun = lun->lun; msg->hdr.nexus.targ_mapped_lun = lun->lun; msg->lun.flags = lun->flags; msg->lun.pr_generation = lun->pr_generation; msg->lun.pr_res_idx = lun->pr_res_idx; msg->lun.pr_res_type = lun->pr_res_type; msg->lun.pr_key_count = lun->pr_key_count; i = 0; if (lun->lun_devid) { msg->lun.lun_devid_len = lun->lun_devid->len; memcpy(&msg->lun.data[i], lun->lun_devid->data, msg->lun.lun_devid_len); i += msg->lun.lun_devid_len; } for (k = 0; k < CTL_MAX_INITIATORS; k++) { if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0) continue; pr_key.pr_iid = k; memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key)); i += sizeof(pr_key); } mtx_unlock(&lun->lun_lock); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i, M_WAITOK); free(msg, M_CTL); if (lun->flags & CTL_LUN_PRIMARY_SC) { for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { ctl_isc_announce_mode(lun, -1, lun->mode_pages.index[i].page_code & SMPH_PC_MASK, lun->mode_pages.index[i].subpage); } } } void ctl_isc_announce_port(struct ctl_port *port) { struct ctl_softc *softc = port->ctl_softc; union ctl_ha_msg *msg; int i; if (port->targ_port < softc->port_min || port->targ_port >= softc->port_max || softc->ha_link != CTL_HA_LINK_ONLINE) return; i = sizeof(msg->port) + strlen(port->port_name) + 1; if (port->lun_map) i += port->lun_map_size * sizeof(uint32_t); if (port->port_devid) i += port->port_devid->len; if (port->target_devid) i += port->target_devid->len; if (port->init_devid) i += port->init_devid->len; msg = malloc(i, M_CTL, M_WAITOK); bzero(&msg->port, sizeof(msg->port)); msg->hdr.msg_type = CTL_MSG_PORT_SYNC; msg->hdr.nexus.targ_port = port->targ_port; msg->port.port_type = port->port_type; msg->port.physical_port = port->physical_port; msg->port.virtual_port = port->virtual_port; msg->port.status = port->status; i = 0; msg->port.name_len = sprintf(&msg->port.data[i], "%d:%s", softc->ha_id, port->port_name) + 1; i += msg->port.name_len; if (port->lun_map) { msg->port.lun_map_len = port->lun_map_size * sizeof(uint32_t); memcpy(&msg->port.data[i], port->lun_map, msg->port.lun_map_len); i += msg->port.lun_map_len; } if (port->port_devid) { msg->port.port_devid_len = port->port_devid->len; memcpy(&msg->port.data[i], port->port_devid->data, msg->port.port_devid_len); i += msg->port.port_devid_len; } if (port->target_devid) { msg->port.target_devid_len = port->target_devid->len; memcpy(&msg->port.data[i], port->target_devid->data, msg->port.target_devid_len); i += msg->port.target_devid_len; } if (port->init_devid) { msg->port.init_devid_len = port->init_devid->len; memcpy(&msg->port.data[i], port->init_devid->data, msg->port.init_devid_len); i += msg->port.init_devid_len; } ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i, M_WAITOK); free(msg, M_CTL); } void ctl_isc_announce_iid(struct ctl_port *port, int iid) { struct ctl_softc *softc = port->ctl_softc; union ctl_ha_msg *msg; int i, l; if (port->targ_port < softc->port_min || port->targ_port >= softc->port_max || softc->ha_link != CTL_HA_LINK_ONLINE) return; mtx_lock(&softc->ctl_lock); i = sizeof(msg->iid); l = 0; if (port->wwpn_iid[iid].name) l = strlen(port->wwpn_iid[iid].name) + 1; i += l; msg = malloc(i, M_CTL, M_NOWAIT); if (msg == NULL) { mtx_unlock(&softc->ctl_lock); return; } bzero(&msg->iid, sizeof(msg->iid)); msg->hdr.msg_type = CTL_MSG_IID_SYNC; msg->hdr.nexus.targ_port = port->targ_port; msg->hdr.nexus.initid = iid; msg->iid.in_use = port->wwpn_iid[iid].in_use; msg->iid.name_len = l; msg->iid.wwpn = port->wwpn_iid[iid].wwpn; if (port->wwpn_iid[iid].name) strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l); mtx_unlock(&softc->ctl_lock); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT); free(msg, M_CTL); } void ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx, uint8_t page, uint8_t subpage) { struct ctl_softc *softc = lun->ctl_softc; union ctl_ha_msg msg; u_int i; if (softc->ha_link != CTL_HA_LINK_ONLINE) return; for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) == page && lun->mode_pages.index[i].subpage == subpage) break; } if (i == CTL_NUM_MODE_PAGES) return; /* Don't try to replicate pages not present on this device. */ if (lun->mode_pages.index[i].page_data == NULL) return; bzero(&msg.mode, sizeof(msg.mode)); msg.hdr.msg_type = CTL_MSG_MODE_SYNC; msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT; msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT; msg.hdr.nexus.targ_lun = lun->lun; msg.hdr.nexus.targ_mapped_lun = lun->lun; msg.mode.page_code = page; msg.mode.subpage = subpage; msg.mode.page_len = lun->mode_pages.index[i].page_len; memcpy(msg.mode.data, lun->mode_pages.index[i].page_data, msg.mode.page_len); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode), M_WAITOK); } static void ctl_isc_ha_link_up(struct ctl_softc *softc) { struct ctl_port *port; struct ctl_lun *lun; union ctl_ha_msg msg; int i; /* Announce this node parameters to peer for validation. */ msg.login.msg_type = CTL_MSG_LOGIN; msg.login.version = CTL_HA_VERSION; msg.login.ha_mode = softc->ha_mode; msg.login.ha_id = softc->ha_id; msg.login.max_luns = ctl_max_luns; msg.login.max_ports = ctl_max_ports; msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login), M_WAITOK); STAILQ_FOREACH(port, &softc->port_list, links) { ctl_isc_announce_port(port); for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) { if (port->wwpn_iid[i].in_use) ctl_isc_announce_iid(port, i); } } STAILQ_FOREACH(lun, &softc->lun_list, links) ctl_isc_announce_lun(lun); } static void ctl_isc_ha_link_down(struct ctl_softc *softc) { struct ctl_port *port; struct ctl_lun *lun; union ctl_io *io; int i; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_lock(&lun->lun_lock); if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) { lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY; ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE); } mtx_unlock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); io = ctl_alloc_io(softc->othersc_pool); mtx_lock(&softc->ctl_lock); ctl_zero_io(io); io->io_hdr.msg_type = CTL_MSG_FAILOVER; io->io_hdr.nexus.targ_mapped_lun = lun->lun; ctl_enqueue_isc(io); } STAILQ_FOREACH(port, &softc->port_list, links) { if (port->targ_port >= softc->port_min && port->targ_port < softc->port_max) continue; port->status &= ~CTL_PORT_STATUS_ONLINE; for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) { port->wwpn_iid[i].in_use = 0; free(port->wwpn_iid[i].name, M_CTL); port->wwpn_iid[i].name = NULL; } } mtx_unlock(&softc->ctl_lock); } static void ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len) { struct ctl_lun *lun; uint32_t iid = ctl_get_initindex(&msg->hdr.nexus); mtx_lock(&softc->ctl_lock); if (msg->hdr.nexus.targ_mapped_lun >= ctl_max_luns || (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); return; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES && msg->ua.ua_set) memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8); if (msg->ua.ua_all) { if (msg->ua.ua_set) ctl_est_ua_all(lun, iid, msg->ua.ua_type); else ctl_clr_ua_all(lun, iid, msg->ua.ua_type); } else { if (msg->ua.ua_set) ctl_est_ua(lun, iid, msg->ua.ua_type); else ctl_clr_ua(lun, iid, msg->ua.ua_type); } mtx_unlock(&lun->lun_lock); } static void ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len) { struct ctl_lun *lun; struct ctl_ha_msg_lun_pr_key pr_key; int i, k; ctl_lun_flags oflags; uint32_t targ_lun; targ_lun = msg->hdr.nexus.targ_mapped_lun; mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); return; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); return; } i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0; if (msg->lun.lun_devid_len != i || (i > 0 && memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) { mtx_unlock(&lun->lun_lock); printf("%s: Received conflicting HA LUN %d\n", __func__, targ_lun); return; } else { /* Record whether peer is primary. */ oflags = lun->flags; if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) && (msg->lun.flags & CTL_LUN_DISABLED) == 0) lun->flags |= CTL_LUN_PEER_SC_PRIMARY; else lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY; if (oflags != lun->flags) ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE); /* If peer is primary and we are not -- use data */ if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 && (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) { lun->pr_generation = msg->lun.pr_generation; lun->pr_res_idx = msg->lun.pr_res_idx; lun->pr_res_type = msg->lun.pr_res_type; lun->pr_key_count = msg->lun.pr_key_count; for (k = 0; k < CTL_MAX_INITIATORS; k++) ctl_clr_prkey(lun, k); for (k = 0; k < msg->lun.pr_key_count; k++) { memcpy(&pr_key, &msg->lun.data[i], sizeof(pr_key)); ctl_alloc_prkey(lun, pr_key.pr_iid); ctl_set_prkey(lun, pr_key.pr_iid, pr_key.pr_key); i += sizeof(pr_key); } } mtx_unlock(&lun->lun_lock); CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n", __func__, targ_lun, (msg->lun.flags & CTL_LUN_PRIMARY_SC) ? "primary" : "secondary")); /* If we are primary but peer doesn't know -- notify */ if ((lun->flags & CTL_LUN_PRIMARY_SC) && (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0) ctl_isc_announce_lun(lun); } } static void ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len) { struct ctl_port *port; struct ctl_lun *lun; int i, new; port = softc->ctl_ports[msg->hdr.nexus.targ_port]; if (port == NULL) { CTL_DEBUG_PRINT(("%s: New port %d\n", __func__, msg->hdr.nexus.targ_port)); new = 1; port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO); port->frontend = &ha_frontend; port->targ_port = msg->hdr.nexus.targ_port; port->fe_datamove = ctl_ha_datamove; port->fe_done = ctl_ha_done; } else if (port->frontend == &ha_frontend) { CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__, msg->hdr.nexus.targ_port)); new = 0; } else { printf("%s: Received conflicting HA port %d\n", __func__, msg->hdr.nexus.targ_port); return; } port->port_type = msg->port.port_type; port->physical_port = msg->port.physical_port; port->virtual_port = msg->port.virtual_port; port->status = msg->port.status; i = 0; free(port->port_name, M_CTL); port->port_name = strndup(&msg->port.data[i], msg->port.name_len, M_CTL); i += msg->port.name_len; if (msg->port.lun_map_len != 0) { if (port->lun_map == NULL || port->lun_map_size * sizeof(uint32_t) < msg->port.lun_map_len) { port->lun_map_size = 0; free(port->lun_map, M_CTL); port->lun_map = malloc(msg->port.lun_map_len, M_CTL, M_WAITOK); } memcpy(port->lun_map, &msg->port.data[i], msg->port.lun_map_len); port->lun_map_size = msg->port.lun_map_len / sizeof(uint32_t); i += msg->port.lun_map_len; } else { port->lun_map_size = 0; free(port->lun_map, M_CTL); port->lun_map = NULL; } if (msg->port.port_devid_len != 0) { if (port->port_devid == NULL || port->port_devid->len < msg->port.port_devid_len) { free(port->port_devid, M_CTL); port->port_devid = malloc(sizeof(struct ctl_devid) + msg->port.port_devid_len, M_CTL, M_WAITOK); } memcpy(port->port_devid->data, &msg->port.data[i], msg->port.port_devid_len); port->port_devid->len = msg->port.port_devid_len; i += msg->port.port_devid_len; } else { free(port->port_devid, M_CTL); port->port_devid = NULL; } if (msg->port.target_devid_len != 0) { if (port->target_devid == NULL || port->target_devid->len < msg->port.target_devid_len) { free(port->target_devid, M_CTL); port->target_devid = malloc(sizeof(struct ctl_devid) + msg->port.target_devid_len, M_CTL, M_WAITOK); } memcpy(port->target_devid->data, &msg->port.data[i], msg->port.target_devid_len); port->target_devid->len = msg->port.target_devid_len; i += msg->port.target_devid_len; } else { free(port->target_devid, M_CTL); port->target_devid = NULL; } if (msg->port.init_devid_len != 0) { if (port->init_devid == NULL || port->init_devid->len < msg->port.init_devid_len) { free(port->init_devid, M_CTL); port->init_devid = malloc(sizeof(struct ctl_devid) + msg->port.init_devid_len, M_CTL, M_WAITOK); } memcpy(port->init_devid->data, &msg->port.data[i], msg->port.init_devid_len); port->init_devid->len = msg->port.init_devid_len; i += msg->port.init_devid_len; } else { free(port->init_devid, M_CTL); port->init_devid = NULL; } if (new) { if (ctl_port_register(port) != 0) { printf("%s: ctl_port_register() failed with error\n", __func__); } } mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; mtx_lock(&lun->lun_lock); ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE); mtx_unlock(&lun->lun_lock); } mtx_unlock(&softc->ctl_lock); } static void ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len) { struct ctl_port *port; int iid; port = softc->ctl_ports[msg->hdr.nexus.targ_port]; if (port == NULL) { printf("%s: Received IID for unknown port %d\n", __func__, msg->hdr.nexus.targ_port); return; } iid = msg->hdr.nexus.initid; if (port->wwpn_iid[iid].in_use != 0 && msg->iid.in_use == 0) ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON); port->wwpn_iid[iid].in_use = msg->iid.in_use; port->wwpn_iid[iid].wwpn = msg->iid.wwpn; free(port->wwpn_iid[iid].name, M_CTL); if (msg->iid.name_len) { port->wwpn_iid[iid].name = strndup(&msg->iid.data[0], msg->iid.name_len, M_CTL); } else port->wwpn_iid[iid].name = NULL; } static void ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len) { if (msg->login.version != CTL_HA_VERSION) { printf("CTL HA peers have different versions %d != %d\n", msg->login.version, CTL_HA_VERSION); ctl_ha_msg_abort(CTL_HA_CHAN_CTL); return; } if (msg->login.ha_mode != softc->ha_mode) { printf("CTL HA peers have different ha_mode %d != %d\n", msg->login.ha_mode, softc->ha_mode); ctl_ha_msg_abort(CTL_HA_CHAN_CTL); return; } if (msg->login.ha_id == softc->ha_id) { printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id); ctl_ha_msg_abort(CTL_HA_CHAN_CTL); return; } if (msg->login.max_luns != ctl_max_luns || msg->login.max_ports != ctl_max_ports || msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) { printf("CTL HA peers have different limits\n"); ctl_ha_msg_abort(CTL_HA_CHAN_CTL); return; } } static void ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len) { struct ctl_lun *lun; u_int i; uint32_t initidx, targ_lun; targ_lun = msg->hdr.nexus.targ_mapped_lun; mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); return; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); return; } for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) == msg->mode.page_code && lun->mode_pages.index[i].subpage == msg->mode.subpage) break; } if (i == CTL_NUM_MODE_PAGES) { mtx_unlock(&lun->lun_lock); return; } memcpy(lun->mode_pages.index[i].page_data, msg->mode.data, lun->mode_pages.index[i].page_len); initidx = ctl_get_initindex(&msg->hdr.nexus); if (initidx != -1) ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE); mtx_unlock(&lun->lun_lock); } /* * ISC (Inter Shelf Communication) event handler. Events from the HA * subsystem come in here. */ static void ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param) { struct ctl_softc *softc = control_softc; union ctl_io *io; struct ctl_prio *presio; ctl_ha_status isc_status; CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event)); if (event == CTL_HA_EVT_MSG_RECV) { union ctl_ha_msg *msg, msgbuf; if (param > sizeof(msgbuf)) msg = malloc(param, M_CTL, M_WAITOK); else msg = &msgbuf; isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param, M_WAITOK); if (isc_status != CTL_HA_STATUS_SUCCESS) { printf("%s: Error receiving message: %d\n", __func__, isc_status); if (msg != &msgbuf) free(msg, M_CTL); return; } CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type)); switch (msg->hdr.msg_type) { case CTL_MSG_SERIALIZE: io = ctl_alloc_io(softc->othersc_pool); ctl_zero_io(io); // populate ctsio from msg io->io_hdr.io_type = CTL_IO_SCSI; io->io_hdr.msg_type = CTL_MSG_SERIALIZE; io->io_hdr.original_sc = msg->hdr.original_sc; io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC | CTL_FLAG_IO_ACTIVE; /* * If we're in serialization-only mode, we don't * want to go through full done processing. Thus * the COPY flag. * * XXX KDM add another flag that is more specific. */ if (softc->ha_mode != CTL_HA_MODE_XFER) io->io_hdr.flags |= CTL_FLAG_INT_COPY; io->io_hdr.nexus = msg->hdr.nexus; #if 0 printf("port %u, iid %u, lun %u\n", io->io_hdr.nexus.targ_port, io->io_hdr.nexus.initid, io->io_hdr.nexus.targ_lun); #endif io->scsiio.tag_num = msg->scsi.tag_num; io->scsiio.tag_type = msg->scsi.tag_type; #ifdef CTL_TIME_IO io->io_hdr.start_time = time_uptime; getbinuptime(&io->io_hdr.start_bt); #endif /* CTL_TIME_IO */ io->scsiio.cdb_len = msg->scsi.cdb_len; memcpy(io->scsiio.cdb, msg->scsi.cdb, CTL_MAX_CDBLEN); if (softc->ha_mode == CTL_HA_MODE_XFER) { const struct ctl_cmd_entry *entry; entry = ctl_get_cmd_entry(&io->scsiio, NULL); io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK; io->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK; } ctl_enqueue_isc(io); break; /* Performed on the Originating SC, XFER mode only */ case CTL_MSG_DATAMOVE: { struct ctl_sg_entry *sgl; int i, j; io = msg->hdr.original_sc; if (io == NULL) { printf("%s: original_sc == NULL!\n", __func__); /* XXX KDM do something here */ break; } io->io_hdr.msg_type = CTL_MSG_DATAMOVE; io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE; /* * Keep track of this, we need to send it back over * when the datamove is complete. */ io->io_hdr.serializing_sc = msg->hdr.serializing_sc; if (msg->hdr.status == CTL_SUCCESS) io->io_hdr.status = msg->hdr.status; if (msg->dt.sg_sequence == 0) { #ifdef CTL_TIME_IO getbinuptime(&io->io_hdr.dma_start_bt); #endif i = msg->dt.kern_sg_entries + msg->dt.kern_data_len / CTL_HA_DATAMOVE_SEGMENT + 1; sgl = malloc(sizeof(*sgl) * i, M_CTL, M_WAITOK | M_ZERO); io->io_hdr.remote_sglist = sgl; io->io_hdr.local_sglist = &sgl[msg->dt.kern_sg_entries]; io->scsiio.kern_data_ptr = (uint8_t *)sgl; io->scsiio.kern_sg_entries = msg->dt.kern_sg_entries; io->scsiio.rem_sg_entries = msg->dt.kern_sg_entries; io->scsiio.kern_data_len = msg->dt.kern_data_len; io->scsiio.kern_total_len = msg->dt.kern_total_len; io->scsiio.kern_data_resid = msg->dt.kern_data_resid; io->scsiio.kern_rel_offset = msg->dt.kern_rel_offset; io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR; io->io_hdr.flags |= msg->dt.flags & CTL_FLAG_BUS_ADDR; } else sgl = (struct ctl_sg_entry *) io->scsiio.kern_data_ptr; for (i = msg->dt.sent_sg_entries, j = 0; i < (msg->dt.sent_sg_entries + msg->dt.cur_sg_entries); i++, j++) { sgl[i].addr = msg->dt.sg_list[j].addr; sgl[i].len = msg->dt.sg_list[j].len; #if 0 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n", __func__, sgl[i].addr, sgl[i].len, j, i); #endif } /* * If this is the last piece of the I/O, we've got * the full S/G list. Queue processing in the thread. * Otherwise wait for the next piece. */ if (msg->dt.sg_last != 0) ctl_enqueue_isc(io); break; } /* Performed on the Serializing (primary) SC, XFER mode only */ case CTL_MSG_DATAMOVE_DONE: { if (msg->hdr.serializing_sc == NULL) { printf("%s: serializing_sc == NULL!\n", __func__); /* XXX KDM now what? */ break; } /* * We grab the sense information here in case * there was a failure, so we can return status * back to the initiator. */ io = msg->hdr.serializing_sc; io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE; io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE; io->io_hdr.port_status = msg->scsi.port_status; io->scsiio.kern_data_resid = msg->scsi.kern_data_resid; if (msg->hdr.status != CTL_STATUS_NONE) { io->io_hdr.status = msg->hdr.status; io->scsiio.scsi_status = msg->scsi.scsi_status; io->scsiio.sense_len = msg->scsi.sense_len; memcpy(&io->scsiio.sense_data, &msg->scsi.sense_data, msg->scsi.sense_len); if (msg->hdr.status == CTL_SUCCESS) io->io_hdr.flags |= CTL_FLAG_STATUS_SENT; } ctl_enqueue_isc(io); break; } /* Preformed on Originating SC, SER_ONLY mode */ case CTL_MSG_R2R: io = msg->hdr.original_sc; if (io == NULL) { printf("%s: original_sc == NULL!\n", __func__); break; } io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE; io->io_hdr.msg_type = CTL_MSG_R2R; io->io_hdr.serializing_sc = msg->hdr.serializing_sc; ctl_enqueue_isc(io); break; /* * Performed on Serializing(i.e. primary SC) SC in SER_ONLY * mode. * Performed on the Originating (i.e. secondary) SC in XFER * mode */ case CTL_MSG_FINISH_IO: if (softc->ha_mode == CTL_HA_MODE_XFER) ctl_isc_handler_finish_xfer(softc, msg); else ctl_isc_handler_finish_ser_only(softc, msg); break; /* Preformed on Originating SC */ case CTL_MSG_BAD_JUJU: io = msg->hdr.original_sc; if (io == NULL) { printf("%s: Bad JUJU!, original_sc is NULL!\n", __func__); break; } ctl_copy_sense_data(msg, io); /* * IO should have already been cleaned up on other * SC so clear this flag so we won't send a message * back to finish the IO there. */ io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC; io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE; /* io = msg->hdr.serializing_sc; */ io->io_hdr.msg_type = CTL_MSG_BAD_JUJU; ctl_enqueue_isc(io); break; /* Handle resets sent from the other side */ case CTL_MSG_MANAGE_TASKS: { struct ctl_taskio *taskio; taskio = (struct ctl_taskio *)ctl_alloc_io( softc->othersc_pool); ctl_zero_io((union ctl_io *)taskio); taskio->io_hdr.io_type = CTL_IO_TASK; taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC; taskio->io_hdr.nexus = msg->hdr.nexus; taskio->task_action = msg->task.task_action; taskio->tag_num = msg->task.tag_num; taskio->tag_type = msg->task.tag_type; #ifdef CTL_TIME_IO taskio->io_hdr.start_time = time_uptime; getbinuptime(&taskio->io_hdr.start_bt); #endif /* CTL_TIME_IO */ ctl_run_task((union ctl_io *)taskio); break; } /* Persistent Reserve action which needs attention */ case CTL_MSG_PERS_ACTION: presio = (struct ctl_prio *)ctl_alloc_io( softc->othersc_pool); ctl_zero_io((union ctl_io *)presio); presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION; presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC; presio->io_hdr.nexus = msg->hdr.nexus; presio->pr_msg = msg->pr; ctl_enqueue_isc((union ctl_io *)presio); break; case CTL_MSG_UA: ctl_isc_ua(softc, msg, param); break; case CTL_MSG_PORT_SYNC: ctl_isc_port_sync(softc, msg, param); break; case CTL_MSG_LUN_SYNC: ctl_isc_lun_sync(softc, msg, param); break; case CTL_MSG_IID_SYNC: ctl_isc_iid_sync(softc, msg, param); break; case CTL_MSG_LOGIN: ctl_isc_login(softc, msg, param); break; case CTL_MSG_MODE_SYNC: ctl_isc_mode_sync(softc, msg, param); break; default: printf("Received HA message of unknown type %d\n", msg->hdr.msg_type); ctl_ha_msg_abort(CTL_HA_CHAN_CTL); break; } if (msg != &msgbuf) free(msg, M_CTL); } else if (event == CTL_HA_EVT_LINK_CHANGE) { printf("CTL: HA link status changed from %d to %d\n", softc->ha_link, param); if (param == softc->ha_link) return; if (softc->ha_link == CTL_HA_LINK_ONLINE) { softc->ha_link = param; ctl_isc_ha_link_down(softc); } else { softc->ha_link = param; if (softc->ha_link == CTL_HA_LINK_ONLINE) ctl_isc_ha_link_up(softc); } return; } else { printf("ctl_isc_event_handler: Unknown event %d\n", event); return; } } static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest) { memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data, src->scsi.sense_len); dest->scsiio.scsi_status = src->scsi.scsi_status; dest->scsiio.sense_len = src->scsi.sense_len; dest->io_hdr.status = src->hdr.status; } static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest) { memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data, src->scsiio.sense_len); dest->scsi.scsi_status = src->scsiio.scsi_status; dest->scsi.sense_len = src->scsiio.sense_len; dest->hdr.status = src->io_hdr.status; } void ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua) { struct ctl_softc *softc = lun->ctl_softc; ctl_ua_type *pu; if (initidx < softc->init_min || initidx >= softc->init_max) return; mtx_assert(&lun->lun_lock, MA_OWNED); pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT]; if (pu == NULL) return; pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua; } void ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua) { int i; mtx_assert(&lun->lun_lock, MA_OWNED); if (lun->pending_ua[port] == NULL) return; for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) { if (port * CTL_MAX_INIT_PER_PORT + i == except) continue; lun->pending_ua[port][i] |= ua; } } void ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua) { struct ctl_softc *softc = lun->ctl_softc; int i; mtx_assert(&lun->lun_lock, MA_OWNED); for (i = softc->port_min; i < softc->port_max; i++) ctl_est_ua_port(lun, i, except, ua); } void ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua) { struct ctl_softc *softc = lun->ctl_softc; ctl_ua_type *pu; if (initidx < softc->init_min || initidx >= softc->init_max) return; mtx_assert(&lun->lun_lock, MA_OWNED); pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT]; if (pu == NULL) return; pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua; } void ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua) { struct ctl_softc *softc = lun->ctl_softc; int i, j; mtx_assert(&lun->lun_lock, MA_OWNED); for (i = softc->port_min; i < softc->port_max; i++) { if (lun->pending_ua[i] == NULL) continue; for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) { if (i * CTL_MAX_INIT_PER_PORT + j == except) continue; lun->pending_ua[i][j] &= ~ua; } } } void ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx, ctl_ua_type ua_type) { struct ctl_lun *lun; mtx_assert(&ctl_softc->ctl_lock, MA_OWNED); STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) { mtx_lock(&lun->lun_lock); ctl_clr_ua(lun, initidx, ua_type); mtx_unlock(&lun->lun_lock); } } static int ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS) { struct ctl_softc *softc = (struct ctl_softc *)arg1; struct ctl_lun *lun; struct ctl_lun_req ireq; int error, value; value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1; error = sysctl_handle_int(oidp, &value, 0, req); if ((error != 0) || (req->newptr == NULL)) return (error); mtx_lock(&softc->ctl_lock); if (value == 0) softc->flags |= CTL_FLAG_ACTIVE_SHELF; else softc->flags &= ~CTL_FLAG_ACTIVE_SHELF; STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_unlock(&softc->ctl_lock); bzero(&ireq, sizeof(ireq)); ireq.reqtype = CTL_LUNREQ_MODIFY; ireq.reqdata.modify.lun_id = lun->lun; lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0, curthread); if (ireq.status != CTL_LUN_OK) { printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n", __func__, ireq.status, ireq.error_str); } mtx_lock(&softc->ctl_lock); } mtx_unlock(&softc->ctl_lock); return (0); } static int ctl_init(void) { struct make_dev_args args; struct ctl_softc *softc; int i, error; softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF, M_WAITOK | M_ZERO); make_dev_args_init(&args); args.mda_devsw = &ctl_cdevsw; args.mda_uid = UID_ROOT; args.mda_gid = GID_OPERATOR; args.mda_mode = 0600; args.mda_si_drv1 = softc; + args.mda_si_drv2 = NULL; error = make_dev_s(&args, &softc->dev, "cam/ctl"); if (error != 0) { free(softc, M_DEVBUF); control_softc = NULL; return (error); } sysctl_ctx_init(&softc->sysctl_ctx); softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl", CTLFLAG_RD, 0, "CAM Target Layer"); if (softc->sysctl_tree == NULL) { printf("%s: unable to allocate sysctl tree\n", __func__); destroy_dev(softc->dev); free(softc, M_DEVBUF); control_softc = NULL; return (ENOMEM); } mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF); softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); softc->flags = 0; SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0, "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)"); if (ctl_max_luns <= 0 || powerof2(ctl_max_luns) == 0) { printf("Bad value %d for kern.cam.ctl.max_luns, must be a power of two, using %d\n", ctl_max_luns, CTL_DEFAULT_MAX_LUNS); ctl_max_luns = CTL_DEFAULT_MAX_LUNS; } softc->ctl_luns = malloc(sizeof(struct ctl_lun *) * ctl_max_luns, M_DEVBUF, M_WAITOK | M_ZERO); softc->ctl_lun_mask = malloc(sizeof(uint32_t) * ((ctl_max_luns + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO); if (ctl_max_ports <= 0 || powerof2(ctl_max_ports) == 0) { printf("Bad value %d for kern.cam.ctl.max_ports, must be a power of two, using %d\n", ctl_max_ports, CTL_DEFAULT_MAX_PORTS); ctl_max_ports = CTL_DEFAULT_MAX_PORTS; } softc->ctl_port_mask = malloc(sizeof(uint32_t) * ((ctl_max_ports + 31) / 32), M_DEVBUF, M_WAITOK | M_ZERO); softc->ctl_ports = malloc(sizeof(struct ctl_port *) * ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO); /* * In Copan's HA scheme, the "master" and "slave" roles are * figured out through the slot the controller is in. Although it * is an active/active system, someone has to be in charge. */ SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0, "HA head ID (0 - no HA)"); if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) { softc->flags |= CTL_FLAG_ACTIVE_SHELF; softc->is_single = 1; softc->port_cnt = ctl_max_ports; softc->port_min = 0; } else { softc->port_cnt = ctl_max_ports / NUM_HA_SHELVES; softc->port_min = (softc->ha_id - 1) * softc->port_cnt; } softc->port_max = softc->port_min + softc->port_cnt; softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT; softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT; SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0, "HA link state (0 - offline, 1 - unknown, 2 - online)"); STAILQ_INIT(&softc->lun_list); STAILQ_INIT(&softc->pending_lun_queue); STAILQ_INIT(&softc->fe_list); STAILQ_INIT(&softc->port_list); STAILQ_INIT(&softc->be_list); ctl_tpc_init(softc); if (worker_threads <= 0) worker_threads = max(1, mp_ncpus / 4); if (worker_threads > CTL_MAX_THREADS) worker_threads = CTL_MAX_THREADS; for (i = 0; i < worker_threads; i++) { struct ctl_thread *thr = &softc->threads[i]; mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF); thr->ctl_softc = softc; STAILQ_INIT(&thr->incoming_queue); STAILQ_INIT(&thr->rtr_queue); STAILQ_INIT(&thr->done_queue); STAILQ_INIT(&thr->isc_queue); error = kproc_kthread_add(ctl_work_thread, thr, &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i); if (error != 0) { printf("error creating CTL work thread!\n"); return (error); } } error = kproc_kthread_add(ctl_lun_thread, softc, &softc->ctl_proc, &softc->lun_thread, 0, 0, "ctl", "lun"); if (error != 0) { printf("error creating CTL lun thread!\n"); return (error); } error = kproc_kthread_add(ctl_thresh_thread, softc, &softc->ctl_proc, &softc->thresh_thread, 0, 0, "ctl", "thresh"); if (error != 0) { printf("error creating CTL threshold thread!\n"); return (error); } SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN, softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head"); if (softc->is_single == 0) { if (ctl_frontend_register(&ha_frontend) != 0) softc->is_single = 1; } return (0); } static int ctl_shutdown(void) { struct ctl_softc *softc = control_softc; int i; if (softc->is_single == 0) ctl_frontend_deregister(&ha_frontend); destroy_dev(softc->dev); /* Shutdown CTL threads. */ softc->shutdown = 1; for (i = 0; i < worker_threads; i++) { struct ctl_thread *thr = &softc->threads[i]; while (thr->thread != NULL) { wakeup(thr); if (thr->thread != NULL) pause("CTL thr shutdown", 1); } mtx_destroy(&thr->queue_lock); } while (softc->lun_thread != NULL) { wakeup(&softc->pending_lun_queue); if (softc->lun_thread != NULL) pause("CTL thr shutdown", 1); } while (softc->thresh_thread != NULL) { wakeup(softc->thresh_thread); if (softc->thresh_thread != NULL) pause("CTL thr shutdown", 1); } ctl_tpc_shutdown(softc); uma_zdestroy(softc->io_zone); mtx_destroy(&softc->ctl_lock); free(softc->ctl_luns, M_DEVBUF); free(softc->ctl_lun_mask, M_DEVBUF); free(softc->ctl_port_mask, M_DEVBUF); free(softc->ctl_ports, M_DEVBUF); sysctl_ctx_free(&softc->sysctl_ctx); free(softc, M_DEVBUF); control_softc = NULL; return (0); } static int ctl_module_event_handler(module_t mod, int what, void *arg) { switch (what) { case MOD_LOAD: return (ctl_init()); case MOD_UNLOAD: return (ctl_shutdown()); default: return (EOPNOTSUPP); } } /* * XXX KDM should we do some access checks here? Bump a reference count to * prevent a CTL module from being unloaded while someone has it open? */ static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td) { return (0); } static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td) { return (0); } /* * Remove an initiator by port number and initiator ID. * Returns 0 for success, -1 for failure. */ int ctl_remove_initiator(struct ctl_port *port, int iid) { struct ctl_softc *softc = port->ctl_softc; int last; mtx_assert(&softc->ctl_lock, MA_NOTOWNED); if (iid > CTL_MAX_INIT_PER_PORT) { printf("%s: initiator ID %u > maximun %u!\n", __func__, iid, CTL_MAX_INIT_PER_PORT); return (-1); } mtx_lock(&softc->ctl_lock); last = (--port->wwpn_iid[iid].in_use == 0); port->wwpn_iid[iid].last_use = time_uptime; mtx_unlock(&softc->ctl_lock); if (last) ctl_i_t_nexus_loss(softc, iid, CTL_UA_POWERON); ctl_isc_announce_iid(port, iid); return (0); } /* * Add an initiator to the initiator map. * Returns iid for success, < 0 for failure. */ int ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name) { struct ctl_softc *softc = port->ctl_softc; time_t best_time; int i, best; mtx_assert(&softc->ctl_lock, MA_NOTOWNED); if (iid >= CTL_MAX_INIT_PER_PORT) { printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n", __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT); free(name, M_CTL); return (-1); } mtx_lock(&softc->ctl_lock); if (iid < 0 && (wwpn != 0 || name != NULL)) { for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) { if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) { iid = i; break; } if (name != NULL && port->wwpn_iid[i].name != NULL && strcmp(name, port->wwpn_iid[i].name) == 0) { iid = i; break; } } } if (iid < 0) { for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) { if (port->wwpn_iid[i].in_use == 0 && port->wwpn_iid[i].wwpn == 0 && port->wwpn_iid[i].name == NULL) { iid = i; break; } } } if (iid < 0) { best = -1; best_time = INT32_MAX; for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) { if (port->wwpn_iid[i].in_use == 0) { if (port->wwpn_iid[i].last_use < best_time) { best = i; best_time = port->wwpn_iid[i].last_use; } } } iid = best; } if (iid < 0) { mtx_unlock(&softc->ctl_lock); free(name, M_CTL); return (-2); } if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) { /* * This is not an error yet. */ if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) { #if 0 printf("%s: port %d iid %u WWPN %#jx arrived" " again\n", __func__, port->targ_port, iid, (uintmax_t)wwpn); #endif goto take; } if (name != NULL && port->wwpn_iid[iid].name != NULL && strcmp(name, port->wwpn_iid[iid].name) == 0) { #if 0 printf("%s: port %d iid %u name '%s' arrived" " again\n", __func__, port->targ_port, iid, name); #endif goto take; } /* * This is an error, but what do we do about it? The * driver is telling us we have a new WWPN for this * initiator ID, so we pretty much need to use it. */ printf("%s: port %d iid %u WWPN %#jx '%s' arrived," " but WWPN %#jx '%s' is still at that address\n", __func__, port->targ_port, iid, wwpn, name, (uintmax_t)port->wwpn_iid[iid].wwpn, port->wwpn_iid[iid].name); } take: free(port->wwpn_iid[iid].name, M_CTL); port->wwpn_iid[iid].name = name; port->wwpn_iid[iid].wwpn = wwpn; port->wwpn_iid[iid].in_use++; mtx_unlock(&softc->ctl_lock); ctl_isc_announce_iid(port, iid); return (iid); } static int ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf) { int len; switch (port->port_type) { case CTL_PORT_FC: { struct scsi_transportid_fcp *id = (struct scsi_transportid_fcp *)buf; if (port->wwpn_iid[iid].wwpn == 0) return (0); memset(id, 0, sizeof(*id)); id->format_protocol = SCSI_PROTO_FC; scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name); return (sizeof(*id)); } case CTL_PORT_ISCSI: { struct scsi_transportid_iscsi_port *id = (struct scsi_transportid_iscsi_port *)buf; if (port->wwpn_iid[iid].name == NULL) return (0); memset(id, 0, 256); id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT | SCSI_PROTO_ISCSI; len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1; len = roundup2(min(len, 252), 4); scsi_ulto2b(len, id->additional_length); return (sizeof(*id) + len); } case CTL_PORT_SAS: { struct scsi_transportid_sas *id = (struct scsi_transportid_sas *)buf; if (port->wwpn_iid[iid].wwpn == 0) return (0); memset(id, 0, sizeof(*id)); id->format_protocol = SCSI_PROTO_SAS; scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address); return (sizeof(*id)); } default: { struct scsi_transportid_spi *id = (struct scsi_transportid_spi *)buf; memset(id, 0, sizeof(*id)); id->format_protocol = SCSI_PROTO_SPI; scsi_ulto2b(iid, id->scsi_addr); scsi_ulto2b(port->targ_port, id->rel_trgt_port_id); return (sizeof(*id)); } } } /* * Serialize a command that went down the "wrong" side, and so was sent to * this controller for execution. The logic is a little different than the * standard case in ctl_scsiio_precheck(). Errors in this case need to get * sent back to the other side, but in the success case, we execute the * command on this side (XFER mode) or tell the other side to execute it * (SER_ONLY mode). */ static void ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_port *port = CTL_PORT(ctsio); union ctl_ha_msg msg_info; struct ctl_lun *lun; const struct ctl_cmd_entry *entry; uint32_t targ_lun; targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun; /* Make sure that we know about this port. */ if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) { ctl_set_internal_failure(ctsio, /*sks_valid*/ 0, /*retry_count*/ 1); goto badjuju; } /* Make sure that we know about this LUN. */ mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); /* * The other node would not send this request to us unless * received announce that we are primary node for this LUN. * If this LUN does not exist now, it is probably result of * a race, so respond to initiator in the most opaque way. */ ctl_set_busy(ctsio); goto badjuju; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); /* * If the LUN is invalid, pretend that it doesn't exist. * It will go away as soon as all pending I/Os completed. */ if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); ctl_set_busy(ctsio); goto badjuju; } entry = ctl_get_cmd_entry(ctsio, NULL); if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) { mtx_unlock(&lun->lun_lock); goto badjuju; } CTL_LUN(ctsio) = lun; CTL_BACKEND_LUN(ctsio) = lun->be_lun; /* * Every I/O goes into the OOA queue for a * particular LUN, and stays there until completion. */ #ifdef CTL_TIME_IO if (TAILQ_EMPTY(&lun->ooa_queue)) lun->idle_time += getsbinuptime() - lun->last_busy; #endif TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq, ooa_links))) { case CTL_ACTION_BLOCK: ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED; TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr, blocked_links); mtx_unlock(&lun->lun_lock); break; case CTL_ACTION_PASS: case CTL_ACTION_SKIP: if (softc->ha_mode == CTL_HA_MODE_XFER) { ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; ctl_enqueue_rtr((union ctl_io *)ctsio); mtx_unlock(&lun->lun_lock); } else { ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; mtx_unlock(&lun->lun_lock); /* send msg back to other side */ msg_info.hdr.original_sc = ctsio->io_hdr.original_sc; msg_info.hdr.serializing_sc = (union ctl_io *)ctsio; msg_info.hdr.msg_type = CTL_MSG_R2R; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.hdr), M_WAITOK); } break; case CTL_ACTION_OVERLAP: TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); mtx_unlock(&lun->lun_lock); ctl_set_overlapped_cmd(ctsio); goto badjuju; case CTL_ACTION_OVERLAP_TAG: TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); mtx_unlock(&lun->lun_lock); ctl_set_overlapped_tag(ctsio, ctsio->tag_num); goto badjuju; case CTL_ACTION_ERROR: default: TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); mtx_unlock(&lun->lun_lock); ctl_set_internal_failure(ctsio, /*sks_valid*/ 0, /*retry_count*/ 0); badjuju: ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info); msg_info.hdr.original_sc = ctsio->io_hdr.original_sc; msg_info.hdr.serializing_sc = NULL; msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.scsi), M_WAITOK); ctl_free_io((union ctl_io *)ctsio); break; } } /* * Returns 0 for success, errno for failure. */ static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num, struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries) { union ctl_io *io; mtx_lock(&lun->lun_lock); for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL); (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links)) { struct ctl_ooa_entry *entry; /* * If we've got more than we can fit, just count the * remaining entries. */ if (*cur_fill_num >= ooa_hdr->alloc_num) continue; entry = &kern_entries[*cur_fill_num]; entry->tag_num = io->scsiio.tag_num; entry->lun_num = lun->lun; #ifdef CTL_TIME_IO entry->start_bt = io->io_hdr.start_bt; #endif bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len); entry->cdb_len = io->scsiio.cdb_len; if (io->io_hdr.flags & CTL_FLAG_BLOCKED) entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED; if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG) entry->cmd_flags |= CTL_OOACMD_FLAG_DMA; if (io->io_hdr.flags & CTL_FLAG_ABORT) entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT; if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR) entry->cmd_flags |= CTL_OOACMD_FLAG_RTR; if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED) entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED; } mtx_unlock(&lun->lun_lock); } static void * ctl_copyin_alloc(void *user_addr, unsigned int len, char *error_str, size_t error_str_len) { void *kptr; kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO); if (copyin(user_addr, kptr, len) != 0) { snprintf(error_str, error_str_len, "Error copying %d bytes " "from user address %p to kernel address %p", len, user_addr, kptr); free(kptr, M_CTL); return (NULL); } return (kptr); } -static void -ctl_free_args(int num_args, struct ctl_be_arg *args) -{ - int i; - - if (args == NULL) - return; - - for (i = 0; i < num_args; i++) { - free(args[i].kname, M_CTL); - free(args[i].kvalue, M_CTL); - } - - free(args, M_CTL); -} - -static struct ctl_be_arg * -ctl_copyin_args(int num_args, struct ctl_be_arg *uargs, - char *error_str, size_t error_str_len) -{ - struct ctl_be_arg *args; - int i; - - args = ctl_copyin_alloc(uargs, num_args * sizeof(*args), - error_str, error_str_len); - - if (args == NULL) - goto bailout; - - for (i = 0; i < num_args; i++) { - args[i].kname = NULL; - args[i].kvalue = NULL; - } - - for (i = 0; i < num_args; i++) { - uint8_t *tmpptr; - - if (args[i].namelen == 0) { - snprintf(error_str, error_str_len, "Argument %d " - "name length is zero", i); - goto bailout; - } - - args[i].kname = ctl_copyin_alloc(args[i].name, - args[i].namelen, error_str, error_str_len); - if (args[i].kname == NULL) - goto bailout; - - if (args[i].kname[args[i].namelen - 1] != '\0') { - snprintf(error_str, error_str_len, "Argument %d " - "name is not NUL-terminated", i); - goto bailout; - } - - if (args[i].flags & CTL_BEARG_RD) { - if (args[i].vallen == 0) { - snprintf(error_str, error_str_len, "Argument %d " - "value length is zero", i); - goto bailout; - } - - tmpptr = ctl_copyin_alloc(args[i].value, - args[i].vallen, error_str, error_str_len); - if (tmpptr == NULL) - goto bailout; - - if ((args[i].flags & CTL_BEARG_ASCII) - && (tmpptr[args[i].vallen - 1] != '\0')) { - snprintf(error_str, error_str_len, "Argument " - "%d value is not NUL-terminated", i); - free(tmpptr, M_CTL); - goto bailout; - } - args[i].kvalue = tmpptr; - } else { - args[i].kvalue = malloc(args[i].vallen, - M_CTL, M_WAITOK | M_ZERO); - } - } - - return (args); -bailout: - - ctl_free_args(num_args, args); - - return (NULL); -} - -static void -ctl_copyout_args(int num_args, struct ctl_be_arg *args) -{ - int i; - - for (i = 0; i < num_args; i++) { - if (args[i].flags & CTL_BEARG_WR) - copyout(args[i].kvalue, args[i].value, args[i].vallen); - } -} - /* * Escape characters that are illegal or not recommended in XML. */ int ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size) { char *end = str + size; int retval; retval = 0; for (; *str && str < end; str++) { switch (*str) { case '&': retval = sbuf_printf(sb, "&"); break; case '>': retval = sbuf_printf(sb, ">"); break; case '<': retval = sbuf_printf(sb, "<"); break; default: retval = sbuf_putc(sb, *str); break; } if (retval != 0) break; } return (retval); } static void ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb) { struct scsi_vpd_id_descriptor *desc; int i; if (id == NULL || id->len < 4) return; desc = (struct scsi_vpd_id_descriptor *)id->data; switch (desc->id_type & SVPD_ID_TYPE_MASK) { case SVPD_ID_TYPE_T10: sbuf_printf(sb, "t10."); break; case SVPD_ID_TYPE_EUI64: sbuf_printf(sb, "eui."); break; case SVPD_ID_TYPE_NAA: sbuf_printf(sb, "naa."); break; case SVPD_ID_TYPE_SCSI_NAME: break; } switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) { case SVPD_ID_CODESET_BINARY: for (i = 0; i < desc->length; i++) sbuf_printf(sb, "%02x", desc->identifier[i]); break; case SVPD_ID_CODESET_ASCII: sbuf_printf(sb, "%.*s", (int)desc->length, (char *)desc->identifier); break; case SVPD_ID_CODESET_UTF8: sbuf_printf(sb, "%s", (char *)desc->identifier); break; } } static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td) { struct ctl_softc *softc = dev->si_drv1; struct ctl_port *port; struct ctl_lun *lun; int retval; retval = 0; switch (cmd) { case CTL_IO: retval = ctl_ioctl_io(dev, cmd, addr, flag, td); break; case CTL_ENABLE_PORT: case CTL_DISABLE_PORT: case CTL_SET_PORT_WWNS: { struct ctl_port *port; struct ctl_port_entry *entry; entry = (struct ctl_port_entry *)addr; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { int action, done; if (port->targ_port < softc->port_min || port->targ_port >= softc->port_max) continue; action = 0; done = 0; if ((entry->port_type == CTL_PORT_NONE) && (entry->targ_port == port->targ_port)) { /* * If the user only wants to enable or * disable or set WWNs on a specific port, * do the operation and we're done. */ action = 1; done = 1; } else if (entry->port_type & port->port_type) { /* * Compare the user's type mask with the * particular frontend type to see if we * have a match. */ action = 1; done = 0; /* * Make sure the user isn't trying to set * WWNs on multiple ports at the same time. */ if (cmd == CTL_SET_PORT_WWNS) { printf("%s: Can't set WWNs on " "multiple ports\n", __func__); retval = EINVAL; break; } } if (action == 0) continue; /* * XXX KDM we have to drop the lock here, because * the online/offline operations can potentially * block. We need to reference count the frontends * so they can't go away, */ if (cmd == CTL_ENABLE_PORT) { mtx_unlock(&softc->ctl_lock); ctl_port_online(port); mtx_lock(&softc->ctl_lock); } else if (cmd == CTL_DISABLE_PORT) { mtx_unlock(&softc->ctl_lock); ctl_port_offline(port); mtx_lock(&softc->ctl_lock); } else if (cmd == CTL_SET_PORT_WWNS) { ctl_port_set_wwns(port, (entry->flags & CTL_PORT_WWNN_VALID) ? 1 : 0, entry->wwnn, (entry->flags & CTL_PORT_WWPN_VALID) ? 1 : 0, entry->wwpn); } if (done != 0) break; } mtx_unlock(&softc->ctl_lock); break; } case CTL_GET_OOA: { struct ctl_ooa *ooa_hdr; struct ctl_ooa_entry *entries; uint32_t cur_fill_num; ooa_hdr = (struct ctl_ooa *)addr; if ((ooa_hdr->alloc_len == 0) || (ooa_hdr->alloc_num == 0)) { printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u " "must be non-zero\n", __func__, ooa_hdr->alloc_len, ooa_hdr->alloc_num); retval = EINVAL; break; } if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num * sizeof(struct ctl_ooa_entry))) { printf("%s: CTL_GET_OOA: alloc len %u must be alloc " "num %d * sizeof(struct ctl_ooa_entry) %zd\n", __func__, ooa_hdr->alloc_len, ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry)); retval = EINVAL; break; } entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO); if (entries == NULL) { printf("%s: could not allocate %d bytes for OOA " "dump\n", __func__, ooa_hdr->alloc_len); retval = ENOMEM; break; } mtx_lock(&softc->ctl_lock); if ((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0 && (ooa_hdr->lun_num >= ctl_max_luns || softc->ctl_luns[ooa_hdr->lun_num] == NULL)) { mtx_unlock(&softc->ctl_lock); free(entries, M_CTL); printf("%s: CTL_GET_OOA: invalid LUN %ju\n", __func__, (uintmax_t)ooa_hdr->lun_num); retval = EINVAL; break; } cur_fill_num = 0; if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) { STAILQ_FOREACH(lun, &softc->lun_list, links) { ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr, entries); } } else { lun = softc->ctl_luns[ooa_hdr->lun_num]; ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr, entries); } mtx_unlock(&softc->ctl_lock); ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num); ooa_hdr->fill_len = ooa_hdr->fill_num * sizeof(struct ctl_ooa_entry); retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len); if (retval != 0) { printf("%s: error copying out %d bytes for OOA dump\n", __func__, ooa_hdr->fill_len); } getbinuptime(&ooa_hdr->cur_bt); if (cur_fill_num > ooa_hdr->alloc_num) { ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num; ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE; } else { ooa_hdr->dropped_num = 0; ooa_hdr->status = CTL_OOA_OK; } free(entries, M_CTL); break; } case CTL_DELAY_IO: { struct ctl_io_delay_info *delay_info; delay_info = (struct ctl_io_delay_info *)addr; #ifdef CTL_IO_DELAY mtx_lock(&softc->ctl_lock); if (delay_info->lun_id >= ctl_max_luns || (lun = softc->ctl_luns[delay_info->lun_id]) == NULL) { mtx_unlock(&softc->ctl_lock); delay_info->status = CTL_DELAY_STATUS_INVALID_LUN; break; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); delay_info->status = CTL_DELAY_STATUS_OK; switch (delay_info->delay_type) { case CTL_DELAY_TYPE_CONT: case CTL_DELAY_TYPE_ONESHOT: break; default: delay_info->status = CTL_DELAY_STATUS_INVALID_TYPE; break; } switch (delay_info->delay_loc) { case CTL_DELAY_LOC_DATAMOVE: lun->delay_info.datamove_type = delay_info->delay_type; lun->delay_info.datamove_delay = delay_info->delay_secs; break; case CTL_DELAY_LOC_DONE: lun->delay_info.done_type = delay_info->delay_type; lun->delay_info.done_delay = delay_info->delay_secs; break; default: delay_info->status = CTL_DELAY_STATUS_INVALID_LOC; break; } mtx_unlock(&lun->lun_lock); #else delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED; #endif /* CTL_IO_DELAY */ break; } #ifdef CTL_LEGACY_STATS case CTL_GETSTATS: { struct ctl_stats *stats = (struct ctl_stats *)addr; int i; /* * XXX KDM no locking here. If the LUN list changes, * things can blow up. */ i = 0; stats->status = CTL_SS_OK; stats->fill_len = 0; STAILQ_FOREACH(lun, &softc->lun_list, links) { if (stats->fill_len + sizeof(lun->legacy_stats) > stats->alloc_len) { stats->status = CTL_SS_NEED_MORE_SPACE; break; } retval = copyout(&lun->legacy_stats, &stats->lun_stats[i++], sizeof(lun->legacy_stats)); if (retval != 0) break; stats->fill_len += sizeof(lun->legacy_stats); } stats->num_luns = softc->num_luns; stats->flags = CTL_STATS_FLAG_NONE; #ifdef CTL_TIME_IO stats->flags |= CTL_STATS_FLAG_TIME_VALID; #endif getnanouptime(&stats->timestamp); break; } #endif /* CTL_LEGACY_STATS */ case CTL_ERROR_INJECT: { struct ctl_error_desc *err_desc, *new_err_desc; err_desc = (struct ctl_error_desc *)addr; new_err_desc = malloc(sizeof(*new_err_desc), M_CTL, M_WAITOK | M_ZERO); bcopy(err_desc, new_err_desc, sizeof(*new_err_desc)); mtx_lock(&softc->ctl_lock); if (err_desc->lun_id >= ctl_max_luns || (lun = softc->ctl_luns[err_desc->lun_id]) == NULL) { mtx_unlock(&softc->ctl_lock); free(new_err_desc, M_CTL); printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n", __func__, (uintmax_t)err_desc->lun_id); retval = EINVAL; break; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); /* * We could do some checking here to verify the validity * of the request, but given the complexity of error * injection requests, the checking logic would be fairly * complex. * * For now, if the request is invalid, it just won't get * executed and might get deleted. */ STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links); /* * XXX KDM check to make sure the serial number is unique, * in case we somehow manage to wrap. That shouldn't * happen for a very long time, but it's the right thing to * do. */ new_err_desc->serial = lun->error_serial; err_desc->serial = lun->error_serial; lun->error_serial++; mtx_unlock(&lun->lun_lock); break; } case CTL_ERROR_INJECT_DELETE: { struct ctl_error_desc *delete_desc, *desc, *desc2; int delete_done; delete_desc = (struct ctl_error_desc *)addr; delete_done = 0; mtx_lock(&softc->ctl_lock); if (delete_desc->lun_id >= ctl_max_luns || (lun = softc->ctl_luns[delete_desc->lun_id]) == NULL) { mtx_unlock(&softc->ctl_lock); printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n", __func__, (uintmax_t)delete_desc->lun_id); retval = EINVAL; break; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) { if (desc->serial != delete_desc->serial) continue; STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links); free(desc, M_CTL); delete_done = 1; } mtx_unlock(&lun->lun_lock); if (delete_done == 0) { printf("%s: CTL_ERROR_INJECT_DELETE: can't find " "error serial %ju on LUN %u\n", __func__, delete_desc->serial, delete_desc->lun_id); retval = EINVAL; break; } break; } case CTL_DUMP_STRUCTS: { int j, k; struct ctl_port *port; struct ctl_frontend *fe; mtx_lock(&softc->ctl_lock); printf("CTL Persistent Reservation information start:\n"); STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_lock(&lun->lun_lock); if ((lun->flags & CTL_LUN_DISABLED) != 0) { mtx_unlock(&lun->lun_lock); continue; } for (j = 0; j < ctl_max_ports; j++) { if (lun->pr_keys[j] == NULL) continue; for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){ if (lun->pr_keys[j][k] == 0) continue; printf(" LUN %ju port %d iid %d key " "%#jx\n", lun->lun, j, k, (uintmax_t)lun->pr_keys[j][k]); } } mtx_unlock(&lun->lun_lock); } printf("CTL Persistent Reservation information end\n"); printf("CTL Ports:\n"); STAILQ_FOREACH(port, &softc->port_list, links) { printf(" Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN " "%#jx WWPN %#jx\n", port->targ_port, port->port_name, port->frontend->name, port->port_type, port->physical_port, port->virtual_port, (uintmax_t)port->wwnn, (uintmax_t)port->wwpn); for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) { if (port->wwpn_iid[j].in_use == 0 && port->wwpn_iid[j].wwpn == 0 && port->wwpn_iid[j].name == NULL) continue; printf(" iid %u use %d WWPN %#jx '%s'\n", j, port->wwpn_iid[j].in_use, (uintmax_t)port->wwpn_iid[j].wwpn, port->wwpn_iid[j].name); } } printf("CTL Port information end\n"); mtx_unlock(&softc->ctl_lock); /* * XXX KDM calling this without a lock. We'd likely want * to drop the lock before calling the frontend's dump * routine anyway. */ printf("CTL Frontends:\n"); STAILQ_FOREACH(fe, &softc->fe_list, links) { printf(" Frontend '%s'\n", fe->name); if (fe->fe_dump != NULL) fe->fe_dump(); } printf("CTL Frontend information end\n"); break; } case CTL_LUN_REQ: { struct ctl_lun_req *lun_req; struct ctl_backend_driver *backend; + void *packed; + nvlist_t *tmp_args_nvl; + size_t packed_len; lun_req = (struct ctl_lun_req *)addr; + tmp_args_nvl = lun_req->args_nvl; backend = ctl_backend_find(lun_req->backend); if (backend == NULL) { lun_req->status = CTL_LUN_ERROR; snprintf(lun_req->error_str, sizeof(lun_req->error_str), "Backend \"%s\" not found.", lun_req->backend); break; } - if (lun_req->num_be_args > 0) { - lun_req->kern_be_args = ctl_copyin_args( - lun_req->num_be_args, - lun_req->be_args, - lun_req->error_str, - sizeof(lun_req->error_str)); - if (lun_req->kern_be_args == NULL) { + + if (lun_req->args != NULL) { + lun_req->args_nvl = nvlist_unpack(lun_req->args, + lun_req->args_len, 0); + + if (lun_req->args_nvl == NULL) { lun_req->status = CTL_LUN_ERROR; + snprintf(lun_req->error_str, sizeof(lun_req->error_str), + "Cannot unpack args nvlist."); break; } - } + } else + lun_req->args_nvl = nvlist_create(0); retval = backend->ioctl(dev, cmd, addr, flag, td); + nvlist_destroy(lun_req->args_nvl); + lun_req->args_nvl = tmp_args_nvl; - if (lun_req->num_be_args > 0) { - ctl_copyout_args(lun_req->num_be_args, - lun_req->kern_be_args); - ctl_free_args(lun_req->num_be_args, - lun_req->kern_be_args); + if (lun_req->result_nvl != NULL) { + if (lun_req->result != NULL) { + packed = nvlist_pack(lun_req->result_nvl, + &packed_len); + if (packed == NULL) { + lun_req->status = CTL_LUN_ERROR; + snprintf(lun_req->error_str, + sizeof(lun_req->error_str), + "Cannot pack result nvlist."); + break; + } + + if (packed_len > lun_req->result_len) { + lun_req->status = CTL_LUN_ERROR; + snprintf(lun_req->error_str, + sizeof(lun_req->error_str), + "Result nvlist too large."); + free(packed, M_NVLIST); + break; + } + + if (copyout(packed, lun_req->result, packed_len)) { + lun_req->status = CTL_LUN_ERROR; + snprintf(lun_req->error_str, + sizeof(lun_req->error_str), + "Cannot copyout() the result."); + free(packed, M_NVLIST); + break; + } + + lun_req->result_len = packed_len; + free(packed, M_NVLIST); + } + + nvlist_destroy(lun_req->result_nvl); } break; } case CTL_LUN_LIST: { struct sbuf *sb; struct ctl_lun_list *list; - struct ctl_option *opt; + const char *name, *value; + void *cookie; + int type; list = (struct ctl_lun_list *)addr; /* * Allocate a fixed length sbuf here, based on the length * of the user's buffer. We could allocate an auto-extending * buffer, and then tell the user how much larger our * amount of data is than his buffer, but that presents * some problems: * * 1. The sbuf(9) routines use a blocking malloc, and so * we can't hold a lock while calling them with an * auto-extending buffer. * * 2. There is not currently a LUN reference counting * mechanism, outside of outstanding transactions on * the LUN's OOA queue. So a LUN could go away on us * while we're getting the LUN number, backend-specific * information, etc. Thus, given the way things * currently work, we need to hold the CTL lock while * grabbing LUN information. * * So, from the user's standpoint, the best thing to do is * allocate what he thinks is a reasonable buffer length, * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error, * double the buffer length and try again. (And repeat * that until he succeeds.) */ sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN); if (sb == NULL) { list->status = CTL_LUN_LIST_ERROR; snprintf(list->error_str, sizeof(list->error_str), "Unable to allocate %d bytes for LUN list", list->alloc_len); break; } sbuf_printf(sb, "\n"); mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_lock(&lun->lun_lock); retval = sbuf_printf(sb, "\n", (uintmax_t)lun->lun); /* * Bail out as soon as we see that we've overfilled * the buffer. */ if (retval != 0) break; retval = sbuf_printf(sb, "\t%s" "\n", (lun->backend == NULL) ? "none" : lun->backend->name); if (retval != 0) break; retval = sbuf_printf(sb, "\t%d\n", lun->be_lun->lun_type); if (retval != 0) break; if (lun->backend == NULL) { retval = sbuf_printf(sb, "\n"); if (retval != 0) break; continue; } retval = sbuf_printf(sb, "\t%ju\n", (lun->be_lun->maxlba > 0) ? lun->be_lun->maxlba + 1 : 0); if (retval != 0) break; retval = sbuf_printf(sb, "\t%u\n", lun->be_lun->blocksize); if (retval != 0) break; retval = sbuf_printf(sb, "\t"); if (retval != 0) break; retval = ctl_sbuf_printf_esc(sb, lun->be_lun->serial_num, sizeof(lun->be_lun->serial_num)); if (retval != 0) break; retval = sbuf_printf(sb, "\n"); if (retval != 0) break; retval = sbuf_printf(sb, "\t"); if (retval != 0) break; retval = ctl_sbuf_printf_esc(sb, lun->be_lun->device_id, sizeof(lun->be_lun->device_id)); if (retval != 0) break; retval = sbuf_printf(sb, "\n"); if (retval != 0) break; if (lun->backend->lun_info != NULL) { retval = lun->backend->lun_info(lun->be_lun->be_lun, sb); if (retval != 0) break; } - STAILQ_FOREACH(opt, &lun->be_lun->options, links) { - retval = sbuf_printf(sb, "\t<%s>%s\n", - opt->name, opt->value, opt->name); - if (retval != 0) - break; + + cookie = NULL; + while ((name = nvlist_next(lun->be_lun->options, &type, + &cookie)) != NULL) { + sbuf_printf(sb, "\t<%s>", name); + + if (type == NV_TYPE_STRING) { + value = dnvlist_get_string( + lun->be_lun->options, name, NULL); + if (value != NULL) + sbuf_printf(sb, "%s", value); + } + + sbuf_printf(sb, "\n", name); } retval = sbuf_printf(sb, "\n"); if (retval != 0) break; mtx_unlock(&lun->lun_lock); } if (lun != NULL) mtx_unlock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if ((retval != 0) || ((retval = sbuf_printf(sb, "\n")) != 0)) { retval = 0; sbuf_delete(sb); list->status = CTL_LUN_LIST_NEED_MORE_SPACE; snprintf(list->error_str, sizeof(list->error_str), "Out of space, %d bytes is too small", list->alloc_len); break; } sbuf_finish(sb); retval = copyout(sbuf_data(sb), list->lun_xml, sbuf_len(sb) + 1); list->fill_len = sbuf_len(sb) + 1; list->status = CTL_LUN_LIST_OK; sbuf_delete(sb); break; } case CTL_ISCSI: { struct ctl_iscsi *ci; struct ctl_frontend *fe; ci = (struct ctl_iscsi *)addr; fe = ctl_frontend_find("iscsi"); if (fe == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "Frontend \"iscsi\" not found."); break; } retval = fe->ioctl(dev, cmd, addr, flag, td); break; } case CTL_PORT_REQ: { struct ctl_req *req; struct ctl_frontend *fe; + void *packed; + nvlist_t *tmp_args_nvl; + size_t packed_len; req = (struct ctl_req *)addr; + tmp_args_nvl = req->args_nvl; fe = ctl_frontend_find(req->driver); if (fe == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Frontend \"%s\" not found.", req->driver); break; } - if (req->num_args > 0) { - req->kern_args = ctl_copyin_args(req->num_args, - req->args, req->error_str, sizeof(req->error_str)); - if (req->kern_args == NULL) { + + if (req->args != NULL) { + req->args_nvl = nvlist_unpack(req->args, + req->args_len, 0); + + if (req->args_nvl == NULL) { req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "Cannot unpack args nvlist."); break; } - } + } else + req->args_nvl = nvlist_create(0); if (fe->ioctl) retval = fe->ioctl(dev, cmd, addr, flag, td); else retval = ENODEV; - if (req->num_args > 0) { - ctl_copyout_args(req->num_args, req->kern_args); - ctl_free_args(req->num_args, req->kern_args); + nvlist_destroy(req->args_nvl); + req->args_nvl = tmp_args_nvl; + + if (req->result_nvl != NULL) { + if (req->result != NULL) { + packed = nvlist_pack(req->result_nvl, + &packed_len); + if (packed == NULL) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, + sizeof(req->error_str), + "Cannot pack result nvlist."); + break; + } + + if (packed_len > req->result_len) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, + sizeof(req->error_str), + "Result nvlist too large."); + free(packed, M_NVLIST); + break; + } + + if (copyout(packed, req->result, packed_len)) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, + sizeof(req->error_str), + "Cannot copyout() the result."); + free(packed, M_NVLIST); + break; + } + + req->result_len = packed_len; + free(packed, M_NVLIST); + } + + nvlist_destroy(req->result_nvl); } break; } case CTL_PORT_LIST: { struct sbuf *sb; struct ctl_port *port; struct ctl_lun_list *list; - struct ctl_option *opt; - int j; + const char *name, *value; + void *cookie; + int j, type; uint32_t plun; list = (struct ctl_lun_list *)addr; sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN); if (sb == NULL) { list->status = CTL_LUN_LIST_ERROR; snprintf(list->error_str, sizeof(list->error_str), "Unable to allocate %d bytes for LUN list", list->alloc_len); break; } sbuf_printf(sb, "\n"); mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { retval = sbuf_printf(sb, "\n", (uintmax_t)port->targ_port); /* * Bail out as soon as we see that we've overfilled * the buffer. */ if (retval != 0) break; retval = sbuf_printf(sb, "\t%s" "\n", port->frontend->name); if (retval != 0) break; retval = sbuf_printf(sb, "\t%d\n", port->port_type); if (retval != 0) break; retval = sbuf_printf(sb, "\t%s\n", (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO"); if (retval != 0) break; retval = sbuf_printf(sb, "\t%s\n", port->port_name); if (retval != 0) break; retval = sbuf_printf(sb, "\t%d\n", port->physical_port); if (retval != 0) break; retval = sbuf_printf(sb, "\t%d\n", port->virtual_port); if (retval != 0) break; if (port->target_devid != NULL) { sbuf_printf(sb, "\t"); ctl_id_sbuf(port->target_devid, sb); sbuf_printf(sb, "\n"); } if (port->port_devid != NULL) { sbuf_printf(sb, "\t"); ctl_id_sbuf(port->port_devid, sb); sbuf_printf(sb, "\n"); } if (port->port_info != NULL) { retval = port->port_info(port->onoff_arg, sb); if (retval != 0) break; } - STAILQ_FOREACH(opt, &port->options, links) { - retval = sbuf_printf(sb, "\t<%s>%s\n", - opt->name, opt->value, opt->name); - if (retval != 0) - break; + + cookie = NULL; + while ((name = nvlist_next(port->options, &type, + &cookie)) != NULL) { + sbuf_printf(sb, "\t<%s>", name); + + if (type == NV_TYPE_STRING) { + value = dnvlist_get_string(port->options, + name, NULL); + if (value != NULL) + sbuf_printf(sb, "%s", value); + } + + sbuf_printf(sb, "\n", name); } if (port->lun_map != NULL) { sbuf_printf(sb, "\ton\n"); for (j = 0; j < port->lun_map_size; j++) { plun = ctl_lun_map_from_port(port, j); if (plun == UINT32_MAX) continue; sbuf_printf(sb, "\t%u\n", j, plun); } } for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) { if (port->wwpn_iid[j].in_use == 0 || (port->wwpn_iid[j].wwpn == 0 && port->wwpn_iid[j].name == NULL)) continue; if (port->wwpn_iid[j].name != NULL) retval = sbuf_printf(sb, "\t%s\n", j, port->wwpn_iid[j].name); else retval = sbuf_printf(sb, "\tnaa.%08jx\n", j, port->wwpn_iid[j].wwpn); if (retval != 0) break; } if (retval != 0) break; retval = sbuf_printf(sb, "\n"); if (retval != 0) break; } mtx_unlock(&softc->ctl_lock); if ((retval != 0) || ((retval = sbuf_printf(sb, "\n")) != 0)) { retval = 0; sbuf_delete(sb); list->status = CTL_LUN_LIST_NEED_MORE_SPACE; snprintf(list->error_str, sizeof(list->error_str), "Out of space, %d bytes is too small", list->alloc_len); break; } sbuf_finish(sb); retval = copyout(sbuf_data(sb), list->lun_xml, sbuf_len(sb) + 1); list->fill_len = sbuf_len(sb) + 1; list->status = CTL_LUN_LIST_OK; sbuf_delete(sb); break; } case CTL_LUN_MAP: { struct ctl_lun_map *lm = (struct ctl_lun_map *)addr; struct ctl_port *port; mtx_lock(&softc->ctl_lock); if (lm->port < softc->port_min || lm->port >= softc->port_max || (port = softc->ctl_ports[lm->port]) == NULL) { mtx_unlock(&softc->ctl_lock); return (ENXIO); } if (port->status & CTL_PORT_STATUS_ONLINE) { STAILQ_FOREACH(lun, &softc->lun_list, links) { if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; mtx_lock(&lun->lun_lock); ctl_est_ua_port(lun, lm->port, -1, CTL_UA_LUN_CHANGE); mtx_unlock(&lun->lun_lock); } } mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps if (lm->plun != UINT32_MAX) { if (lm->lun == UINT32_MAX) retval = ctl_lun_map_unset(port, lm->plun); else if (lm->lun < ctl_max_luns && softc->ctl_luns[lm->lun] != NULL) retval = ctl_lun_map_set(port, lm->plun, lm->lun); else return (ENXIO); } else { if (lm->lun == UINT32_MAX) retval = ctl_lun_map_deinit(port); else retval = ctl_lun_map_init(port); } if (port->status & CTL_PORT_STATUS_ONLINE) ctl_isc_announce_port(port); break; } case CTL_GET_LUN_STATS: { struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr; int i; /* * XXX KDM no locking here. If the LUN list changes, * things can blow up. */ i = 0; stats->status = CTL_SS_OK; stats->fill_len = 0; STAILQ_FOREACH(lun, &softc->lun_list, links) { if (lun->lun < stats->first_item) continue; if (stats->fill_len + sizeof(lun->stats) > stats->alloc_len) { stats->status = CTL_SS_NEED_MORE_SPACE; break; } retval = copyout(&lun->stats, &stats->stats[i++], sizeof(lun->stats)); if (retval != 0) break; stats->fill_len += sizeof(lun->stats); } stats->num_items = softc->num_luns; stats->flags = CTL_STATS_FLAG_NONE; #ifdef CTL_TIME_IO stats->flags |= CTL_STATS_FLAG_TIME_VALID; #endif getnanouptime(&stats->timestamp); break; } case CTL_GET_PORT_STATS: { struct ctl_get_io_stats *stats = (struct ctl_get_io_stats *)addr; int i; /* * XXX KDM no locking here. If the LUN list changes, * things can blow up. */ i = 0; stats->status = CTL_SS_OK; stats->fill_len = 0; STAILQ_FOREACH(port, &softc->port_list, links) { if (port->targ_port < stats->first_item) continue; if (stats->fill_len + sizeof(port->stats) > stats->alloc_len) { stats->status = CTL_SS_NEED_MORE_SPACE; break; } retval = copyout(&port->stats, &stats->stats[i++], sizeof(port->stats)); if (retval != 0) break; stats->fill_len += sizeof(port->stats); } stats->num_items = softc->num_ports; stats->flags = CTL_STATS_FLAG_NONE; #ifdef CTL_TIME_IO stats->flags |= CTL_STATS_FLAG_TIME_VALID; #endif getnanouptime(&stats->timestamp); break; } default: { /* XXX KDM should we fix this? */ #if 0 struct ctl_backend_driver *backend; unsigned int type; int found; found = 0; /* * We encode the backend type as the ioctl type for backend * ioctls. So parse it out here, and then search for a * backend of this type. */ type = _IOC_TYPE(cmd); STAILQ_FOREACH(backend, &softc->be_list, links) { if (backend->type == type) { found = 1; break; } } if (found == 0) { printf("ctl: unknown ioctl command %#lx or backend " "%d\n", cmd, type); retval = EINVAL; break; } retval = backend->ioctl(dev, cmd, addr, flag, td); #endif retval = ENOTTY; break; } } return (retval); } uint32_t ctl_get_initindex(struct ctl_nexus *nexus) { return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT)); } int ctl_lun_map_init(struct ctl_port *port) { struct ctl_softc *softc = port->ctl_softc; struct ctl_lun *lun; int size = ctl_lun_map_size; uint32_t i; if (port->lun_map == NULL || port->lun_map_size < size) { port->lun_map_size = 0; free(port->lun_map, M_CTL); port->lun_map = malloc(size * sizeof(uint32_t), M_CTL, M_NOWAIT); } if (port->lun_map == NULL) return (ENOMEM); for (i = 0; i < size; i++) port->lun_map[i] = UINT32_MAX; port->lun_map_size = size; if (port->status & CTL_PORT_STATUS_ONLINE) { if (port->lun_disable != NULL) { STAILQ_FOREACH(lun, &softc->lun_list, links) port->lun_disable(port->targ_lun_arg, lun->lun); } ctl_isc_announce_port(port); } return (0); } int ctl_lun_map_deinit(struct ctl_port *port) { struct ctl_softc *softc = port->ctl_softc; struct ctl_lun *lun; if (port->lun_map == NULL) return (0); port->lun_map_size = 0; free(port->lun_map, M_CTL); port->lun_map = NULL; if (port->status & CTL_PORT_STATUS_ONLINE) { if (port->lun_enable != NULL) { STAILQ_FOREACH(lun, &softc->lun_list, links) port->lun_enable(port->targ_lun_arg, lun->lun); } ctl_isc_announce_port(port); } return (0); } int ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun) { int status; uint32_t old; if (port->lun_map == NULL) { status = ctl_lun_map_init(port); if (status != 0) return (status); } if (plun >= port->lun_map_size) return (EINVAL); old = port->lun_map[plun]; port->lun_map[plun] = glun; if ((port->status & CTL_PORT_STATUS_ONLINE) && old == UINT32_MAX) { if (port->lun_enable != NULL) port->lun_enable(port->targ_lun_arg, plun); ctl_isc_announce_port(port); } return (0); } int ctl_lun_map_unset(struct ctl_port *port, uint32_t plun) { uint32_t old; if (port->lun_map == NULL || plun >= port->lun_map_size) return (0); old = port->lun_map[plun]; port->lun_map[plun] = UINT32_MAX; if ((port->status & CTL_PORT_STATUS_ONLINE) && old != UINT32_MAX) { if (port->lun_disable != NULL) port->lun_disable(port->targ_lun_arg, plun); ctl_isc_announce_port(port); } return (0); } uint32_t ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id) { if (port == NULL) return (UINT32_MAX); if (port->lun_map == NULL) return (lun_id); if (lun_id > port->lun_map_size) return (UINT32_MAX); return (port->lun_map[lun_id]); } uint32_t ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id) { uint32_t i; if (port == NULL) return (UINT32_MAX); if (port->lun_map == NULL) return (lun_id); for (i = 0; i < port->lun_map_size; i++) { if (port->lun_map[i] == lun_id) return (i); } return (UINT32_MAX); } uint32_t ctl_decode_lun(uint64_t encoded) { uint8_t lun[8]; uint32_t result = 0xffffffff; be64enc(lun, encoded); switch (lun[0] & RPL_LUNDATA_ATYP_MASK) { case RPL_LUNDATA_ATYP_PERIPH: if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0) result = lun[1]; break; case RPL_LUNDATA_ATYP_FLAT: if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0) result = ((lun[0] & 0x3f) << 8) + lun[1]; break; case RPL_LUNDATA_ATYP_EXTLUN: switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) { case 0x02: switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) { case 0x00: result = lun[1]; break; case 0x10: result = (lun[1] << 16) + (lun[2] << 8) + lun[3]; break; case 0x20: if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0) result = (lun[2] << 24) + (lun[3] << 16) + (lun[4] << 8) + lun[5]; break; } break; case RPL_LUNDATA_EXT_EAM_NOT_SPEC: result = 0xffffffff; break; } break; } return (result); } uint64_t ctl_encode_lun(uint32_t decoded) { uint64_t l = decoded; if (l <= 0xff) return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48)); if (l <= 0x3fff) return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48)); if (l <= 0xffffff) return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) | (l << 32)); return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16)); } int ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last) { int i; for (i = first; i < last; i++) { if ((mask[i / 32] & (1 << (i % 32))) == 0) return (i); } return (-1); } int ctl_set_mask(uint32_t *mask, uint32_t bit) { uint32_t chunk, piece; chunk = bit >> 5; piece = bit % (sizeof(uint32_t) * 8); if ((mask[chunk] & (1 << piece)) != 0) return (-1); else mask[chunk] |= (1 << piece); return (0); } int ctl_clear_mask(uint32_t *mask, uint32_t bit) { uint32_t chunk, piece; chunk = bit >> 5; piece = bit % (sizeof(uint32_t) * 8); if ((mask[chunk] & (1 << piece)) == 0) return (-1); else mask[chunk] &= ~(1 << piece); return (0); } int ctl_is_set(uint32_t *mask, uint32_t bit) { uint32_t chunk, piece; chunk = bit >> 5; piece = bit % (sizeof(uint32_t) * 8); if ((mask[chunk] & (1 << piece)) == 0) return (0); else return (1); } static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx) { uint64_t *t; t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT]; if (t == NULL) return (0); return (t[residx % CTL_MAX_INIT_PER_PORT]); } static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx) { uint64_t *t; t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT]; if (t == NULL) return; t[residx % CTL_MAX_INIT_PER_PORT] = 0; } static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx) { uint64_t *p; u_int i; i = residx/CTL_MAX_INIT_PER_PORT; if (lun->pr_keys[i] != NULL) return; mtx_unlock(&lun->lun_lock); p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL, M_WAITOK | M_ZERO); mtx_lock(&lun->lun_lock); if (lun->pr_keys[i] == NULL) lun->pr_keys[i] = p; else free(p, M_CTL); } static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key) { uint64_t *t; t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT]; KASSERT(t != NULL, ("prkey %d is not allocated", residx)); t[residx % CTL_MAX_INIT_PER_PORT] = key; } /* * ctl_softc, pool_name, total_ctl_io are passed in. * npool is passed out. */ int ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name, uint32_t total_ctl_io, void **npool) { struct ctl_io_pool *pool; pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL, M_NOWAIT | M_ZERO); if (pool == NULL) return (ENOMEM); snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name); pool->ctl_softc = ctl_softc; #ifdef IO_POOLS pool->zone = uma_zsecond_create(pool->name, NULL, NULL, NULL, NULL, ctl_softc->io_zone); /* uma_prealloc(pool->zone, total_ctl_io); */ #else pool->zone = ctl_softc->io_zone; #endif *npool = pool; return (0); } void ctl_pool_free(struct ctl_io_pool *pool) { if (pool == NULL) return; #ifdef IO_POOLS uma_zdestroy(pool->zone); #endif free(pool, M_CTL); } union ctl_io * ctl_alloc_io(void *pool_ref) { struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref; union ctl_io *io; io = uma_zalloc(pool->zone, M_WAITOK); if (io != NULL) { io->io_hdr.pool = pool_ref; CTL_SOFTC(io) = pool->ctl_softc; } return (io); } union ctl_io * ctl_alloc_io_nowait(void *pool_ref) { struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref; union ctl_io *io; io = uma_zalloc(pool->zone, M_NOWAIT); if (io != NULL) { io->io_hdr.pool = pool_ref; CTL_SOFTC(io) = pool->ctl_softc; } return (io); } void ctl_free_io(union ctl_io *io) { struct ctl_io_pool *pool; if (io == NULL) return; pool = (struct ctl_io_pool *)io->io_hdr.pool; uma_zfree(pool->zone, io); } void ctl_zero_io(union ctl_io *io) { struct ctl_io_pool *pool; if (io == NULL) return; /* * May need to preserve linked list pointers at some point too. */ pool = io->io_hdr.pool; memset(io, 0, sizeof(*io)); io->io_hdr.pool = pool; CTL_SOFTC(io) = pool->ctl_softc; } int ctl_expand_number(const char *buf, uint64_t *num) { char *endptr; uint64_t number; unsigned shift; number = strtoq(buf, &endptr, 0); switch (tolower((unsigned char)*endptr)) { case 'e': shift = 60; break; case 'p': shift = 50; break; case 't': shift = 40; break; case 'g': shift = 30; break; case 'm': shift = 20; break; case 'k': shift = 10; break; case 'b': case '\0': /* No unit. */ *num = number; return (0); default: /* Unrecognized unit. */ return (-1); } if ((number << shift) >> shift != number) { /* Overflow */ return (-1); } *num = number << shift; return (0); } /* * This routine could be used in the future to load default and/or saved * mode page parameters for a particuar lun. */ static int ctl_init_page_index(struct ctl_lun *lun) { int i, page_code; struct ctl_page_index *page_index; const char *value; uint64_t ival; memcpy(&lun->mode_pages.index, page_index_template, sizeof(page_index_template)); for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { page_index = &lun->mode_pages.index[i]; if (lun->be_lun->lun_type == T_DIRECT && (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0) continue; if (lun->be_lun->lun_type == T_PROCESSOR && (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0) continue; if (lun->be_lun->lun_type == T_CDROM && (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0) continue; page_code = page_index->page_code & SMPH_PC_MASK; switch (page_code) { case SMS_RW_ERROR_RECOVERY_PAGE: { KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0, ("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code)); memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT], &rw_er_page_default, sizeof(rw_er_page_default)); memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE], &rw_er_page_changeable, sizeof(rw_er_page_changeable)); memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT], &rw_er_page_default, sizeof(rw_er_page_default)); memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED], &rw_er_page_default, sizeof(rw_er_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.rw_er_page; break; } case SMS_FORMAT_DEVICE_PAGE: { struct scsi_format_page *format_page; KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0, ("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code)); /* * Sectors per track are set above. Bytes per * sector need to be set here on a per-LUN basis. */ memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT], &format_page_default, sizeof(format_page_default)); memcpy(&lun->mode_pages.format_page[ CTL_PAGE_CHANGEABLE], &format_page_changeable, sizeof(format_page_changeable)); memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT], &format_page_default, sizeof(format_page_default)); memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED], &format_page_default, sizeof(format_page_default)); format_page = &lun->mode_pages.format_page[ CTL_PAGE_CURRENT]; scsi_ulto2b(lun->be_lun->blocksize, format_page->bytes_per_sector); format_page = &lun->mode_pages.format_page[ CTL_PAGE_DEFAULT]; scsi_ulto2b(lun->be_lun->blocksize, format_page->bytes_per_sector); format_page = &lun->mode_pages.format_page[ CTL_PAGE_SAVED]; scsi_ulto2b(lun->be_lun->blocksize, format_page->bytes_per_sector); page_index->page_data = (uint8_t *)lun->mode_pages.format_page; break; } case SMS_RIGID_DISK_PAGE: { struct scsi_rigid_disk_page *rigid_disk_page; uint32_t sectors_per_cylinder; uint64_t cylinders; #ifndef __XSCALE__ int shift; #endif /* !__XSCALE__ */ KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0, ("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code)); /* * Rotation rate and sectors per track are set * above. We calculate the cylinders here based on * capacity. Due to the number of heads and * sectors per track we're using, smaller arrays * may turn out to have 0 cylinders. Linux and * FreeBSD don't pay attention to these mode pages * to figure out capacity, but Solaris does. It * seems to deal with 0 cylinders just fine, and * works out a fake geometry based on the capacity. */ memcpy(&lun->mode_pages.rigid_disk_page[ CTL_PAGE_DEFAULT], &rigid_disk_page_default, sizeof(rigid_disk_page_default)); memcpy(&lun->mode_pages.rigid_disk_page[ CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable, sizeof(rigid_disk_page_changeable)); sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK * CTL_DEFAULT_HEADS; /* * The divide method here will be more accurate, * probably, but results in floating point being * used in the kernel on i386 (__udivdi3()). On the * XScale, though, __udivdi3() is implemented in * software. * * The shift method for cylinder calculation is * accurate if sectors_per_cylinder is a power of * 2. Otherwise it might be slightly off -- you * might have a bit of a truncation problem. */ #ifdef __XSCALE__ cylinders = (lun->be_lun->maxlba + 1) / sectors_per_cylinder; #else for (shift = 31; shift > 0; shift--) { if (sectors_per_cylinder & (1 << shift)) break; } cylinders = (lun->be_lun->maxlba + 1) >> shift; #endif /* * We've basically got 3 bytes, or 24 bits for the * cylinder size in the mode page. If we're over, * just round down to 2^24. */ if (cylinders > 0xffffff) cylinders = 0xffffff; rigid_disk_page = &lun->mode_pages.rigid_disk_page[ CTL_PAGE_DEFAULT]; scsi_ulto3b(cylinders, rigid_disk_page->cylinders); - if ((value = ctl_get_opt(&lun->be_lun->options, - "rpm")) != NULL) { + if ((value = dnvlist_get_string(lun->be_lun->options, + "rpm", NULL)) != NULL) { scsi_ulto2b(strtol(value, NULL, 0), rigid_disk_page->rotation_rate); } memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT], &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT], sizeof(rigid_disk_page_default)); memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED], &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT], sizeof(rigid_disk_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.rigid_disk_page; break; } case SMS_VERIFY_ERROR_RECOVERY_PAGE: { KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0, ("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code)); memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT], &verify_er_page_default, sizeof(verify_er_page_default)); memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE], &verify_er_page_changeable, sizeof(verify_er_page_changeable)); memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT], &verify_er_page_default, sizeof(verify_er_page_default)); memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED], &verify_er_page_default, sizeof(verify_er_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.verify_er_page; break; } case SMS_CACHING_PAGE: { struct scsi_caching_page *caching_page; KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0, ("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code)); memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT], &caching_page_default, sizeof(caching_page_default)); memcpy(&lun->mode_pages.caching_page[ CTL_PAGE_CHANGEABLE], &caching_page_changeable, sizeof(caching_page_changeable)); memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED], &caching_page_default, sizeof(caching_page_default)); caching_page = &lun->mode_pages.caching_page[ CTL_PAGE_SAVED]; - value = ctl_get_opt(&lun->be_lun->options, "writecache"); + value = dnvlist_get_string(lun->be_lun->options, + "writecache", NULL); if (value != NULL && strcmp(value, "off") == 0) caching_page->flags1 &= ~SCP_WCE; - value = ctl_get_opt(&lun->be_lun->options, "readcache"); + value = dnvlist_get_string(lun->be_lun->options, + "readcache", NULL); if (value != NULL && strcmp(value, "off") == 0) caching_page->flags1 |= SCP_RCD; memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT], &lun->mode_pages.caching_page[CTL_PAGE_SAVED], sizeof(caching_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.caching_page; break; } case SMS_CONTROL_MODE_PAGE: { switch (page_index->subpage) { case SMS_SUBPAGE_PAGE_0: { struct scsi_control_page *control_page; memcpy(&lun->mode_pages.control_page[ CTL_PAGE_DEFAULT], &control_page_default, sizeof(control_page_default)); memcpy(&lun->mode_pages.control_page[ CTL_PAGE_CHANGEABLE], &control_page_changeable, sizeof(control_page_changeable)); memcpy(&lun->mode_pages.control_page[ CTL_PAGE_SAVED], &control_page_default, sizeof(control_page_default)); control_page = &lun->mode_pages.control_page[ CTL_PAGE_SAVED]; - value = ctl_get_opt(&lun->be_lun->options, - "reordering"); + value = dnvlist_get_string(lun->be_lun->options, + "reordering", NULL); if (value != NULL && strcmp(value, "unrestricted") == 0) { control_page->queue_flags &= ~SCP_QUEUE_ALG_MASK; control_page->queue_flags |= SCP_QUEUE_ALG_UNRESTRICTED; } memcpy(&lun->mode_pages.control_page[ CTL_PAGE_CURRENT], &lun->mode_pages.control_page[ CTL_PAGE_SAVED], sizeof(control_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.control_page; break; } case 0x01: memcpy(&lun->mode_pages.control_ext_page[ CTL_PAGE_DEFAULT], &control_ext_page_default, sizeof(control_ext_page_default)); memcpy(&lun->mode_pages.control_ext_page[ CTL_PAGE_CHANGEABLE], &control_ext_page_changeable, sizeof(control_ext_page_changeable)); memcpy(&lun->mode_pages.control_ext_page[ CTL_PAGE_SAVED], &control_ext_page_default, sizeof(control_ext_page_default)); memcpy(&lun->mode_pages.control_ext_page[ CTL_PAGE_CURRENT], &lun->mode_pages.control_ext_page[ CTL_PAGE_SAVED], sizeof(control_ext_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.control_ext_page; break; default: panic("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code); } break; } case SMS_INFO_EXCEPTIONS_PAGE: { switch (page_index->subpage) { case SMS_SUBPAGE_PAGE_0: memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT], &ie_page_default, sizeof(ie_page_default)); memcpy(&lun->mode_pages.ie_page[ CTL_PAGE_CHANGEABLE], &ie_page_changeable, sizeof(ie_page_changeable)); memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT], &ie_page_default, sizeof(ie_page_default)); memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED], &ie_page_default, sizeof(ie_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.ie_page; break; case 0x02: { struct ctl_logical_block_provisioning_page *page; memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT], &lbp_page_default, sizeof(lbp_page_default)); memcpy(&lun->mode_pages.lbp_page[ CTL_PAGE_CHANGEABLE], &lbp_page_changeable, sizeof(lbp_page_changeable)); memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED], &lbp_page_default, sizeof(lbp_page_default)); page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED]; - value = ctl_get_opt(&lun->be_lun->options, - "avail-threshold"); + value = dnvlist_get_string(lun->be_lun->options, + "avail-threshold", NULL); if (value != NULL && ctl_expand_number(value, &ival) == 0) { page->descr[0].flags |= SLBPPD_ENABLED | SLBPPD_ARMING_DEC; if (lun->be_lun->blocksize) ival /= lun->be_lun->blocksize; else ival /= 512; scsi_ulto4b(ival >> CTL_LBP_EXPONENT, page->descr[0].count); } - value = ctl_get_opt(&lun->be_lun->options, - "used-threshold"); + value = dnvlist_get_string(lun->be_lun->options, + "used-threshold", NULL); if (value != NULL && ctl_expand_number(value, &ival) == 0) { page->descr[1].flags |= SLBPPD_ENABLED | SLBPPD_ARMING_INC; if (lun->be_lun->blocksize) ival /= lun->be_lun->blocksize; else ival /= 512; scsi_ulto4b(ival >> CTL_LBP_EXPONENT, page->descr[1].count); } - value = ctl_get_opt(&lun->be_lun->options, - "pool-avail-threshold"); + value = dnvlist_get_string(lun->be_lun->options, + "pool-avail-threshold", NULL); if (value != NULL && ctl_expand_number(value, &ival) == 0) { page->descr[2].flags |= SLBPPD_ENABLED | SLBPPD_ARMING_DEC; if (lun->be_lun->blocksize) ival /= lun->be_lun->blocksize; else ival /= 512; scsi_ulto4b(ival >> CTL_LBP_EXPONENT, page->descr[2].count); } - value = ctl_get_opt(&lun->be_lun->options, - "pool-used-threshold"); + value = dnvlist_get_string(lun->be_lun->options, + "pool-used-threshold", NULL); if (value != NULL && ctl_expand_number(value, &ival) == 0) { page->descr[3].flags |= SLBPPD_ENABLED | SLBPPD_ARMING_INC; if (lun->be_lun->blocksize) ival /= lun->be_lun->blocksize; else ival /= 512; scsi_ulto4b(ival >> CTL_LBP_EXPONENT, page->descr[3].count); } memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT], &lun->mode_pages.lbp_page[CTL_PAGE_SAVED], sizeof(lbp_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.lbp_page; break; } default: panic("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code); } break; } case SMS_CDDVD_CAPS_PAGE:{ KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0, ("subpage %#x for page %#x is incorrect!", page_index->subpage, page_code)); memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT], &cddvd_page_default, sizeof(cddvd_page_default)); memcpy(&lun->mode_pages.cddvd_page[ CTL_PAGE_CHANGEABLE], &cddvd_page_changeable, sizeof(cddvd_page_changeable)); memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED], &cddvd_page_default, sizeof(cddvd_page_default)); memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT], &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED], sizeof(cddvd_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.cddvd_page; break; } default: panic("invalid page code value %#x", page_code); } } return (CTL_RETVAL_COMPLETE); } static int ctl_init_log_page_index(struct ctl_lun *lun) { struct ctl_page_index *page_index; int i, j, k, prev; memcpy(&lun->log_pages.index, log_page_index_template, sizeof(log_page_index_template)); prev = -1; for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) { page_index = &lun->log_pages.index[i]; if (lun->be_lun->lun_type == T_DIRECT && (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0) continue; if (lun->be_lun->lun_type == T_PROCESSOR && (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0) continue; if (lun->be_lun->lun_type == T_CDROM && (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0) continue; if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING && lun->backend->lun_attr == NULL) continue; if (page_index->page_code != prev) { lun->log_pages.pages_page[j] = page_index->page_code; prev = page_index->page_code; j++; } lun->log_pages.subpages_page[k*2] = page_index->page_code; lun->log_pages.subpages_page[k*2+1] = page_index->subpage; k++; } lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0]; lun->log_pages.index[0].page_len = j; lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0]; lun->log_pages.index[1].page_len = k * 2; lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0]; lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS; lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page; lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page); lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page; lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page); return (CTL_RETVAL_COMPLETE); } static int hex2bin(const char *str, uint8_t *buf, int buf_size) { int i; u_char c; memset(buf, 0, buf_size); while (isspace(str[0])) str++; if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X')) str += 2; buf_size *= 2; for (i = 0; str[i] != 0 && i < buf_size; i++) { while (str[i] == '-') /* Skip dashes in UUIDs. */ str++; c = str[i]; if (isdigit(c)) c -= '0'; else if (isalpha(c)) c -= isupper(c) ? 'A' - 10 : 'a' - 10; else break; if (c >= 16) break; if ((i & 1) == 0) buf[i / 2] |= (c << 4); else buf[i / 2] |= c; } return ((i + 1) / 2); } /* * LUN allocation. * * Requirements: * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he * wants us to allocate the LUN and he can block. * - ctl_softc is always set * - be_lun is set if the LUN has a backend (needed for disk LUNs) * * Returns 0 for success, non-zero (errno) for failure. */ static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun, struct ctl_be_lun *const be_lun) { struct ctl_lun *nlun, *lun; struct scsi_vpd_id_descriptor *desc; struct scsi_vpd_id_t10 *t10id; const char *eui, *naa, *scsiname, *uuid, *vendor, *value; int lun_number, lun_malloced; int devidlen, idlen1, idlen2 = 0, len; if (be_lun == NULL) return (EINVAL); /* * We currently only support Direct Access or Processor LUN types. */ switch (be_lun->lun_type) { case T_DIRECT: case T_PROCESSOR: case T_CDROM: break; case T_SEQUENTIAL: case T_CHANGER: default: be_lun->lun_config_status(be_lun->be_lun, CTL_LUN_CONFIG_FAILURE); break; } if (ctl_lun == NULL) { lun = malloc(sizeof(*lun), M_CTL, M_WAITOK); lun_malloced = 1; } else { lun_malloced = 0; lun = ctl_lun; } memset(lun, 0, sizeof(*lun)); if (lun_malloced) lun->flags = CTL_LUN_MALLOCED; lun->pending_sense = malloc(sizeof(struct scsi_sense_data *) * ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO); lun->pending_ua = malloc(sizeof(ctl_ua_type *) * ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO); lun->pr_keys = malloc(sizeof(uint64_t *) * ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO); /* Generate LUN ID. */ devidlen = max(CTL_DEVID_MIN_LEN, strnlen(be_lun->device_id, CTL_DEVID_LEN)); idlen1 = sizeof(*t10id) + devidlen; len = sizeof(struct scsi_vpd_id_descriptor) + idlen1; - scsiname = ctl_get_opt(&be_lun->options, "scsiname"); + scsiname = dnvlist_get_string(be_lun->options, "scsiname", NULL); if (scsiname != NULL) { idlen2 = roundup2(strlen(scsiname) + 1, 4); len += sizeof(struct scsi_vpd_id_descriptor) + idlen2; } - eui = ctl_get_opt(&be_lun->options, "eui"); + eui = dnvlist_get_string(be_lun->options, "eui", NULL); if (eui != NULL) { len += sizeof(struct scsi_vpd_id_descriptor) + 16; } - naa = ctl_get_opt(&be_lun->options, "naa"); + naa = dnvlist_get_string(be_lun->options, "naa", NULL); if (naa != NULL) { len += sizeof(struct scsi_vpd_id_descriptor) + 16; } - uuid = ctl_get_opt(&be_lun->options, "uuid"); + uuid = dnvlist_get_string(be_lun->options, "uuid", NULL); if (uuid != NULL) { len += sizeof(struct scsi_vpd_id_descriptor) + 18; } lun->lun_devid = malloc(sizeof(struct ctl_devid) + len, M_CTL, M_WAITOK | M_ZERO); desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data; desc->proto_codeset = SVPD_ID_CODESET_ASCII; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10; desc->length = idlen1; t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0]; memset(t10id->vendor, ' ', sizeof(t10id->vendor)); - if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) { + if ((vendor = dnvlist_get_string(be_lun->options, "vendor", NULL)) == NULL) { strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor)); } else { strncpy(t10id->vendor, vendor, min(sizeof(t10id->vendor), strlen(vendor))); } strncpy((char *)t10id->vendor_spec_id, (char *)be_lun->device_id, devidlen); if (scsiname != NULL) { desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] + desc->length); desc->proto_codeset = SVPD_ID_CODESET_UTF8; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_SCSI_NAME; desc->length = idlen2; strlcpy(desc->identifier, scsiname, idlen2); } if (eui != NULL) { desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] + desc->length); desc->proto_codeset = SVPD_ID_CODESET_BINARY; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_EUI64; desc->length = hex2bin(eui, desc->identifier, 16); desc->length = desc->length > 12 ? 16 : (desc->length > 8 ? 12 : 8); len -= 16 - desc->length; } if (naa != NULL) { desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] + desc->length); desc->proto_codeset = SVPD_ID_CODESET_BINARY; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_NAA; desc->length = hex2bin(naa, desc->identifier, 16); desc->length = desc->length > 8 ? 16 : 8; len -= 16 - desc->length; } if (uuid != NULL) { desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] + desc->length); desc->proto_codeset = SVPD_ID_CODESET_BINARY; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_UUID; desc->identifier[0] = 0x10; hex2bin(uuid, &desc->identifier[2], 16); desc->length = 18; } lun->lun_devid->len = len; mtx_lock(&ctl_softc->ctl_lock); /* * See if the caller requested a particular LUN number. If so, see * if it is available. Otherwise, allocate the first available LUN. */ if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) { if ((be_lun->req_lun_id > (ctl_max_luns - 1)) || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) { mtx_unlock(&ctl_softc->ctl_lock); if (be_lun->req_lun_id > (ctl_max_luns - 1)) { printf("ctl: requested LUN ID %d is higher " "than ctl_max_luns - 1 (%d)\n", be_lun->req_lun_id, ctl_max_luns - 1); } else { /* * XXX KDM return an error, or just assign * another LUN ID in this case?? */ printf("ctl: requested LUN ID %d is already " "in use\n", be_lun->req_lun_id); } fail: free(lun->lun_devid, M_CTL); if (lun->flags & CTL_LUN_MALLOCED) free(lun, M_CTL); be_lun->lun_config_status(be_lun->be_lun, CTL_LUN_CONFIG_FAILURE); return (ENOSPC); } lun_number = be_lun->req_lun_id; } else { lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, ctl_max_luns); if (lun_number == -1) { mtx_unlock(&ctl_softc->ctl_lock); printf("ctl: can't allocate LUN, out of LUNs\n"); goto fail; } } ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number); mtx_unlock(&ctl_softc->ctl_lock); mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF); lun->lun = lun_number; lun->be_lun = be_lun; /* * The processor LUN is always enabled. Disk LUNs come on line * disabled, and must be enabled by the backend. */ lun->flags |= CTL_LUN_DISABLED; lun->backend = be_lun->be; be_lun->ctl_lun = lun; be_lun->lun_id = lun_number; atomic_add_int(&be_lun->be->num_luns, 1); if (be_lun->flags & CTL_LUN_FLAG_EJECTED) lun->flags |= CTL_LUN_EJECTED; if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA) lun->flags |= CTL_LUN_NO_MEDIA; if (be_lun->flags & CTL_LUN_FLAG_STOPPED) lun->flags |= CTL_LUN_STOPPED; if (be_lun->flags & CTL_LUN_FLAG_PRIMARY) lun->flags |= CTL_LUN_PRIMARY_SC; - value = ctl_get_opt(&be_lun->options, "removable"); + value = dnvlist_get_string(be_lun->options, "removable", NULL); if (value != NULL) { if (strcmp(value, "on") == 0) lun->flags |= CTL_LUN_REMOVABLE; } else if (be_lun->lun_type == T_CDROM) lun->flags |= CTL_LUN_REMOVABLE; lun->ctl_softc = ctl_softc; #ifdef CTL_TIME_IO lun->last_busy = getsbinuptime(); #endif TAILQ_INIT(&lun->ooa_queue); TAILQ_INIT(&lun->blocked_queue); STAILQ_INIT(&lun->error_list); lun->ie_reported = 1; callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0); ctl_tpc_lun_init(lun); if (lun->flags & CTL_LUN_REMOVABLE) { lun->prevent = malloc((CTL_MAX_INITIATORS + 31) / 32 * 4, M_CTL, M_WAITOK); } /* * Initialize the mode and log page index. */ ctl_init_page_index(lun); ctl_init_log_page_index(lun); /* Setup statistics gathering */ #ifdef CTL_LEGACY_STATS lun->legacy_stats.device_type = be_lun->lun_type; lun->legacy_stats.lun_number = lun_number; lun->legacy_stats.blocksize = be_lun->blocksize; if (be_lun->blocksize == 0) lun->legacy_stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE; lun->legacy_stats.ports = malloc(sizeof(struct ctl_lun_io_port_stats) * ctl_max_ports, M_DEVBUF, M_WAITOK | M_ZERO); for (len = 0; len < ctl_max_ports; len++) lun->legacy_stats.ports[len].targ_port = len; #endif /* CTL_LEGACY_STATS */ lun->stats.item = lun_number; /* * Now, before we insert this lun on the lun list, set the lun * inventory changed UA for all other luns. */ mtx_lock(&ctl_softc->ctl_lock); STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) { mtx_lock(&nlun->lun_lock); ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE); mtx_unlock(&nlun->lun_lock); } STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links); ctl_softc->ctl_luns[lun_number] = lun; ctl_softc->num_luns++; mtx_unlock(&ctl_softc->ctl_lock); lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK); return (0); } /* * Delete a LUN. * Assumptions: * - LUN has already been marked invalid and any pending I/O has been taken * care of. */ static int ctl_free_lun(struct ctl_lun *lun) { struct ctl_softc *softc = lun->ctl_softc; struct ctl_lun *nlun; int i; KASSERT(TAILQ_EMPTY(&lun->ooa_queue), ("Freeing a LUN %p with outstanding I/O!\n", lun)); mtx_lock(&softc->ctl_lock); STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links); ctl_clear_mask(softc->ctl_lun_mask, lun->lun); softc->ctl_luns[lun->lun] = NULL; softc->num_luns--; STAILQ_FOREACH(nlun, &softc->lun_list, links) { mtx_lock(&nlun->lun_lock); ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE); mtx_unlock(&nlun->lun_lock); } mtx_unlock(&softc->ctl_lock); /* * Tell the backend to free resources, if this LUN has a backend. */ atomic_subtract_int(&lun->be_lun->be->num_luns, 1); lun->be_lun->lun_shutdown(lun->be_lun->be_lun); lun->ie_reportcnt = UINT32_MAX; callout_drain(&lun->ie_callout); ctl_tpc_lun_shutdown(lun); mtx_destroy(&lun->lun_lock); free(lun->lun_devid, M_CTL); for (i = 0; i < ctl_max_ports; i++) free(lun->pending_ua[i], M_CTL); free(lun->pending_ua, M_DEVBUF); for (i = 0; i < ctl_max_ports; i++) free(lun->pr_keys[i], M_CTL); free(lun->pr_keys, M_DEVBUF); free(lun->write_buffer, M_CTL); free(lun->prevent, M_CTL); if (lun->flags & CTL_LUN_MALLOCED) free(lun, M_CTL); return (0); } static void ctl_create_lun(struct ctl_be_lun *be_lun) { /* * ctl_alloc_lun() should handle all potential failure cases. */ ctl_alloc_lun(control_softc, NULL, be_lun); } int ctl_add_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *softc = control_softc; mtx_lock(&softc->ctl_lock); STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links); mtx_unlock(&softc->ctl_lock); wakeup(&softc->pending_lun_queue); return (0); } int ctl_enable_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *softc; struct ctl_port *port, *nport; struct ctl_lun *lun; int retval; lun = (struct ctl_lun *)be_lun->ctl_lun; softc = lun->ctl_softc; mtx_lock(&softc->ctl_lock); mtx_lock(&lun->lun_lock); if ((lun->flags & CTL_LUN_DISABLED) == 0) { /* * eh? Why did we get called if the LUN is already * enabled? */ mtx_unlock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); return (0); } lun->flags &= ~CTL_LUN_DISABLED; mtx_unlock(&lun->lun_lock); STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 || port->lun_map != NULL || port->lun_enable == NULL) continue; /* * Drop the lock while we call the FETD's enable routine. * This can lead to a callback into CTL (at least in the * case of the internal initiator frontend. */ mtx_unlock(&softc->ctl_lock); retval = port->lun_enable(port->targ_lun_arg, lun->lun); mtx_lock(&softc->ctl_lock); if (retval != 0) { printf("%s: FETD %s port %d returned error " "%d for lun_enable on lun %jd\n", __func__, port->port_name, port->targ_port, retval, (intmax_t)lun->lun); } } mtx_unlock(&softc->ctl_lock); ctl_isc_announce_lun(lun); return (0); } int ctl_disable_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *softc; struct ctl_port *port; struct ctl_lun *lun; int retval; lun = (struct ctl_lun *)be_lun->ctl_lun; softc = lun->ctl_softc; mtx_lock(&softc->ctl_lock); mtx_lock(&lun->lun_lock); if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); return (0); } lun->flags |= CTL_LUN_DISABLED; mtx_unlock(&lun->lun_lock); STAILQ_FOREACH(port, &softc->port_list, links) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 || port->lun_map != NULL || port->lun_disable == NULL) continue; /* * Drop the lock before we call the frontend's disable * routine, to avoid lock order reversals. * * XXX KDM what happens if the frontend list changes while * we're traversing it? It's unlikely, but should be handled. */ mtx_unlock(&softc->ctl_lock); retval = port->lun_disable(port->targ_lun_arg, lun->lun); mtx_lock(&softc->ctl_lock); if (retval != 0) { printf("%s: FETD %s port %d returned error " "%d for lun_disable on lun %jd\n", __func__, port->port_name, port->targ_port, retval, (intmax_t)lun->lun); } } mtx_unlock(&softc->ctl_lock); ctl_isc_announce_lun(lun); return (0); } int ctl_start_lun(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags &= ~CTL_LUN_STOPPED; mtx_unlock(&lun->lun_lock); return (0); } int ctl_stop_lun(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags |= CTL_LUN_STOPPED; mtx_unlock(&lun->lun_lock); return (0); } int ctl_lun_no_media(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags |= CTL_LUN_NO_MEDIA; mtx_unlock(&lun->lun_lock); return (0); } int ctl_lun_has_media(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; union ctl_ha_msg msg; mtx_lock(&lun->lun_lock); lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED); if (lun->flags & CTL_LUN_REMOVABLE) ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE); mtx_unlock(&lun->lun_lock); if ((lun->flags & CTL_LUN_REMOVABLE) && lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) { bzero(&msg.ua, sizeof(msg.ua)); msg.hdr.msg_type = CTL_MSG_UA; msg.hdr.nexus.initid = -1; msg.hdr.nexus.targ_port = -1; msg.hdr.nexus.targ_lun = lun->lun; msg.hdr.nexus.targ_mapped_lun = lun->lun; msg.ua.ua_all = 1; msg.ua.ua_set = 1; msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua), M_WAITOK); } return (0); } int ctl_lun_ejected(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags |= CTL_LUN_EJECTED; mtx_unlock(&lun->lun_lock); return (0); } int ctl_lun_primary(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags |= CTL_LUN_PRIMARY_SC; ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE); mtx_unlock(&lun->lun_lock); ctl_isc_announce_lun(lun); return (0); } int ctl_lun_secondary(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags &= ~CTL_LUN_PRIMARY_SC; ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE); mtx_unlock(&lun->lun_lock); ctl_isc_announce_lun(lun); return (0); } int ctl_invalidate_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *softc; struct ctl_lun *lun; lun = (struct ctl_lun *)be_lun->ctl_lun; softc = lun->ctl_softc; mtx_lock(&lun->lun_lock); /* * The LUN needs to be disabled before it can be marked invalid. */ if ((lun->flags & CTL_LUN_DISABLED) == 0) { mtx_unlock(&lun->lun_lock); return (-1); } /* * Mark the LUN invalid. */ lun->flags |= CTL_LUN_INVALID; /* * If there is nothing in the OOA queue, go ahead and free the LUN. * If we have something in the OOA queue, we'll free it when the * last I/O completes. */ if (TAILQ_EMPTY(&lun->ooa_queue)) { mtx_unlock(&lun->lun_lock); ctl_free_lun(lun); } else mtx_unlock(&lun->lun_lock); return (0); } void ctl_lun_capacity_changed(struct ctl_be_lun *be_lun) { struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun; union ctl_ha_msg msg; mtx_lock(&lun->lun_lock); ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE); mtx_unlock(&lun->lun_lock); if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) { /* Send msg to other side. */ bzero(&msg.ua, sizeof(msg.ua)); msg.hdr.msg_type = CTL_MSG_UA; msg.hdr.nexus.initid = -1; msg.hdr.nexus.targ_port = -1; msg.hdr.nexus.targ_lun = lun->lun; msg.hdr.nexus.targ_mapped_lun = lun->lun; msg.ua.ua_all = 1; msg.ua.ua_set = 1; msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua), M_WAITOK); } } /* * Backend "memory move is complete" callback for requests that never * make it down to say RAIDCore's configuration code. */ int ctl_config_move_done(union ctl_io *io) { int retval; CTL_DEBUG_PRINT(("ctl_config_move_done\n")); KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type)); if ((io->io_hdr.port_status != 0) && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ io->io_hdr.port_status); } else if (io->scsiio.kern_data_resid != 0 && (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { ctl_set_invalid_field_ciu(&io->scsiio); } if (ctl_debug & CTL_DEBUG_CDB_DATA) ctl_data_print(io); if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) { /* * XXX KDM just assuming a single pointer here, and not a * S/G list. If we start using S/G lists for config data, * we'll need to know how to clean them up here as well. */ if (io->io_hdr.flags & CTL_FLAG_ALLOCATED) free(io->scsiio.kern_data_ptr, M_CTL); ctl_done(io); retval = CTL_RETVAL_COMPLETE; } else { /* * XXX KDM now we need to continue data movement. Some * options: * - call ctl_scsiio() again? We don't do this for data * writes, because for those at least we know ahead of * time where the write will go and how long it is. For * config writes, though, that information is largely * contained within the write itself, thus we need to * parse out the data again. * * - Call some other function once the data is in? */ /* * XXX KDM call ctl_scsiio() again for now, and check flag * bits to see whether we're allocated or not. */ retval = ctl_scsiio(&io->scsiio); } return (retval); } /* * This gets called by a backend driver when it is done with a * data_submit method. */ void ctl_data_submit_done(union ctl_io *io) { /* * If the IO_CONT flag is set, we need to call the supplied * function to continue processing the I/O, instead of completing * the I/O just yet. * * If there is an error, though, we don't want to keep processing. * Instead, just send status back to the initiator. */ if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) && (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { io->scsiio.io_cont(io); return; } ctl_done(io); } /* * This gets called by a backend driver when it is done with a * configuration write. */ void ctl_config_write_done(union ctl_io *io) { uint8_t *buf; /* * If the IO_CONT flag is set, we need to call the supplied * function to continue processing the I/O, instead of completing * the I/O just yet. * * If there is an error, though, we don't want to keep processing. * Instead, just send status back to the initiator. */ if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) && (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { io->scsiio.io_cont(io); return; } /* * Since a configuration write can be done for commands that actually * have data allocated, like write buffer, and commands that have * no data, like start/stop unit, we need to check here. */ if (io->io_hdr.flags & CTL_FLAG_ALLOCATED) buf = io->scsiio.kern_data_ptr; else buf = NULL; ctl_done(io); if (buf) free(buf, M_CTL); } void ctl_config_read_done(union ctl_io *io) { uint8_t *buf; /* * If there is some error -- we are done, skip data transfer. */ if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { if (io->io_hdr.flags & CTL_FLAG_ALLOCATED) buf = io->scsiio.kern_data_ptr; else buf = NULL; ctl_done(io); if (buf) free(buf, M_CTL); return; } /* * If the IO_CONT flag is set, we need to call the supplied * function to continue processing the I/O, instead of completing * the I/O just yet. */ if (io->io_hdr.flags & CTL_FLAG_IO_CONT) { io->scsiio.io_cont(io); return; } ctl_datamove(io); } /* * SCSI release command. */ int ctl_scsi_release(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); uint32_t residx; CTL_DEBUG_PRINT(("ctl_scsi_release\n")); residx = ctl_get_initindex(&ctsio->io_hdr.nexus); /* * XXX KDM right now, we only support LUN reservation. We don't * support 3rd party reservations, or extent reservations, which * might actually need the parameter list. If we've gotten this * far, we've got a LUN reservation. Anything else got kicked out * above. So, according to SPC, ignore the length. */ mtx_lock(&lun->lun_lock); /* * According to SPC, it is not an error for an intiator to attempt * to release a reservation on a LUN that isn't reserved, or that * is reserved by another initiator. The reservation can only be * released, though, by the initiator who made it or by one of * several reset type events. */ if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx)) lun->flags &= ~CTL_LUN_RESERVED; mtx_unlock(&lun->lun_lock); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_scsi_reserve(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); uint32_t residx; CTL_DEBUG_PRINT(("ctl_reserve\n")); residx = ctl_get_initindex(&ctsio->io_hdr.nexus); /* * XXX KDM right now, we only support LUN reservation. We don't * support 3rd party reservations, or extent reservations, which * might actually need the parameter list. If we've gotten this * far, we've got a LUN reservation. Anything else got kicked out * above. So, according to SPC, ignore the length. */ mtx_lock(&lun->lun_lock); if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) { ctl_set_reservation_conflict(ctsio); goto bailout; } /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */ if (lun->flags & CTL_LUN_PR_RESERVED) { ctl_set_success(ctsio); goto bailout; } lun->flags |= CTL_LUN_RESERVED; lun->res_idx = residx; ctl_set_success(ctsio); bailout: mtx_unlock(&lun->lun_lock); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_start_stop(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_start_stop_unit *cdb; int retval; CTL_DEBUG_PRINT(("ctl_start_stop\n")); cdb = (struct scsi_start_stop_unit *)ctsio->cdb; if ((cdb->how & SSS_PC_MASK) == 0) { if ((lun->flags & CTL_LUN_PR_RESERVED) && (cdb->how & SSS_START) == 0) { uint32_t residx; residx = ctl_get_initindex(&ctsio->io_hdr.nexus); if (ctl_get_prkey(lun, residx) == 0 || (lun->pr_res_idx != residx && lun->pr_res_type < 4)) { ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } } if ((cdb->how & SSS_LOEJ) && (lun->flags & CTL_LUN_REMOVABLE) == 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 4, /*bit_valid*/ 1, /*bit*/ 1); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) && lun->prevent_count > 0) { /* "Medium removal prevented" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ? SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } } retval = lun->backend->config_write((union ctl_io *)ctsio); return (retval); } int ctl_prevent_allow(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_prevent *cdb; int retval; uint32_t initidx; CTL_DEBUG_PRINT(("ctl_prevent_allow\n")); cdb = (struct scsi_prevent *)ctsio->cdb; if ((lun->flags & CTL_LUN_REMOVABLE) == 0 || lun->prevent == NULL) { ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); mtx_lock(&lun->lun_lock); if ((cdb->how & PR_PREVENT) && ctl_is_set(lun->prevent, initidx) == 0) { ctl_set_mask(lun->prevent, initidx); lun->prevent_count++; } else if ((cdb->how & PR_PREVENT) == 0 && ctl_is_set(lun->prevent, initidx)) { ctl_clear_mask(lun->prevent, initidx); lun->prevent_count--; } mtx_unlock(&lun->lun_lock); retval = lun->backend->config_write((union ctl_io *)ctsio); return (retval); } /* * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but * we don't really do anything with the LBA and length fields if the user * passes them in. Instead we'll just flush out the cache for the entire * LUN. */ int ctl_sync_cache(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct ctl_lba_len_flags *lbalen; uint64_t starting_lba; uint32_t block_count; int retval; uint8_t byte2; CTL_DEBUG_PRINT(("ctl_sync_cache\n")); retval = 0; switch (ctsio->cdb[0]) { case SYNCHRONIZE_CACHE: { struct scsi_sync_cache *cdb; cdb = (struct scsi_sync_cache *)ctsio->cdb; starting_lba = scsi_4btoul(cdb->begin_lba); block_count = scsi_2btoul(cdb->lb_count); byte2 = cdb->byte2; break; } case SYNCHRONIZE_CACHE_16: { struct scsi_sync_cache_16 *cdb; cdb = (struct scsi_sync_cache_16 *)ctsio->cdb; starting_lba = scsi_8btou64(cdb->begin_lba); block_count = scsi_4btoul(cdb->lb_count); byte2 = cdb->byte2; break; } default: ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); goto bailout; break; /* NOTREACHED */ } /* * We check the LBA and length, but don't do anything with them. * A SYNCHRONIZE CACHE will cause the entire cache for this lun to * get flushed. This check will just help satisfy anyone who wants * to see an error for an out of range LBA. */ if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) { ctl_set_lba_out_of_range(ctsio, MAX(starting_lba, lun->be_lun->maxlba + 1)); ctl_done((union ctl_io *)ctsio); goto bailout; } lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; lbalen->lba = starting_lba; lbalen->len = block_count; lbalen->flags = byte2; retval = lun->backend->config_write((union ctl_io *)ctsio); bailout: return (retval); } int ctl_format(struct ctl_scsiio *ctsio) { struct scsi_format *cdb; int length, defect_list_len; CTL_DEBUG_PRINT(("ctl_format\n")); cdb = (struct scsi_format *)ctsio->cdb; length = 0; if (cdb->byte2 & SF_FMTDATA) { if (cdb->byte2 & SF_LONGLIST) length = sizeof(struct scsi_format_header_long); else length = sizeof(struct scsi_format_header_short); } if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) && (length > 0)) { ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK); ctsio->kern_data_len = length; ctsio->kern_total_len = length; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } defect_list_len = 0; if (cdb->byte2 & SF_FMTDATA) { if (cdb->byte2 & SF_LONGLIST) { struct scsi_format_header_long *header; header = (struct scsi_format_header_long *) ctsio->kern_data_ptr; defect_list_len = scsi_4btoul(header->defect_list_len); if (defect_list_len != 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); goto bailout; } } else { struct scsi_format_header_short *header; header = (struct scsi_format_header_short *) ctsio->kern_data_ptr; defect_list_len = scsi_2btoul(header->defect_list_len); if (defect_list_len != 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); goto bailout; } } } ctl_set_success(ctsio); bailout: if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) { free(ctsio->kern_data_ptr, M_CTL); ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED; } ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_read_buffer(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); uint64_t buffer_offset; uint32_t len; uint8_t byte2; static uint8_t descr[4]; static uint8_t echo_descr[4] = { 0 }; CTL_DEBUG_PRINT(("ctl_read_buffer\n")); switch (ctsio->cdb[0]) { case READ_BUFFER: { struct scsi_read_buffer *cdb; cdb = (struct scsi_read_buffer *)ctsio->cdb; buffer_offset = scsi_3btoul(cdb->offset); len = scsi_3btoul(cdb->length); byte2 = cdb->byte2; break; } case READ_BUFFER_16: { struct scsi_read_buffer_16 *cdb; cdb = (struct scsi_read_buffer_16 *)ctsio->cdb; buffer_offset = scsi_8btou64(cdb->offset); len = scsi_4btoul(cdb->length); byte2 = cdb->byte2; break; } default: /* This shouldn't happen. */ ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if (buffer_offset > CTL_WRITE_BUFFER_SIZE || buffer_offset + len > CTL_WRITE_BUFFER_SIZE) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 6, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) { descr[0] = 0; scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]); ctsio->kern_data_ptr = descr; len = min(len, sizeof(descr)); } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) { ctsio->kern_data_ptr = echo_descr; len = min(len, sizeof(echo_descr)); } else { if (lun->write_buffer == NULL) { lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE, M_CTL, M_WAITOK); } ctsio->kern_data_ptr = lun->write_buffer + buffer_offset; } ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctl_set_success(ctsio); ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_write_buffer(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_write_buffer *cdb; int buffer_offset, len; CTL_DEBUG_PRINT(("ctl_write_buffer\n")); cdb = (struct scsi_write_buffer *)ctsio->cdb; len = scsi_3btoul(cdb->length); buffer_offset = scsi_3btoul(cdb->offset); if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 6, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * If we've got a kernel request that hasn't been malloced yet, * malloc it and tell the caller the data buffer is here. */ if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { if (lun->write_buffer == NULL) { lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE, M_CTL, M_WAITOK); } ctsio->kern_data_ptr = lun->write_buffer + buffer_offset; ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_write_same(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int len, retval; uint8_t byte2; CTL_DEBUG_PRINT(("ctl_write_same\n")); switch (ctsio->cdb[0]) { case WRITE_SAME_10: { struct scsi_write_same_10 *cdb; cdb = (struct scsi_write_same_10 *)ctsio->cdb; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_2btoul(cdb->length); byte2 = cdb->byte2; break; } case WRITE_SAME_16: { struct scsi_write_same_16 *cdb; cdb = (struct scsi_write_same_16 *)ctsio->cdb; lba = scsi_8btou64(cdb->addr); num_blocks = scsi_4btoul(cdb->length); byte2 = cdb->byte2; break; } default: /* * We got a command we don't support. This shouldn't * happen, commands should be filtered out above us. */ ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); break; /* NOTREACHED */ } /* ANCHOR flag can be used only together with UNMAP */ if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * The first check is to make sure we're in bounds, the second * check is to catch wrap-around problems. If the lba + num blocks * is less than the lba, then we've wrapped around and the block * range is invalid anyway. */ if (((lba + num_blocks) > (lun->be_lun->maxlba + 1)) || ((lba + num_blocks) < lba)) { ctl_set_lba_out_of_range(ctsio, MAX(lba, lun->be_lun->maxlba + 1)); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* Zero number of blocks means "to the last logical block" */ if (num_blocks == 0) { if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 0, /*command*/ 1, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } num_blocks = (lun->be_lun->maxlba + 1) - lba; } len = lun->be_lun->blocksize; /* * If we've got a kernel request that hasn't been malloced yet, * malloc it and tell the caller the data buffer is here. */ if ((byte2 & SWS_NDOB) == 0 && (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK); ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; lbalen->lba = lba; lbalen->len = num_blocks; lbalen->flags = byte2; retval = lun->backend->config_write((union ctl_io *)ctsio); return (retval); } int ctl_unmap(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_unmap *cdb; struct ctl_ptr_len_flags *ptrlen; struct scsi_unmap_header *hdr; struct scsi_unmap_desc *buf, *end, *endnz, *range; uint64_t lba; uint32_t num_blocks; int len, retval; uint8_t byte2; CTL_DEBUG_PRINT(("ctl_unmap\n")); cdb = (struct scsi_unmap *)ctsio->cdb; len = scsi_2btoul(cdb->length); byte2 = cdb->byte2; /* * If we've got a kernel request that hasn't been malloced yet, * malloc it and tell the caller the data buffer is here. */ if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK); ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } len = ctsio->kern_total_len - ctsio->kern_data_resid; hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr; if (len < sizeof (*hdr) || len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) || len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) || scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } len = scsi_2btoul(hdr->desc_length); buf = (struct scsi_unmap_desc *)(hdr + 1); end = buf + len / sizeof(*buf); endnz = buf; for (range = buf; range < end; range++) { lba = scsi_8btou64(range->lba); num_blocks = scsi_4btoul(range->length); if (((lba + num_blocks) > (lun->be_lun->maxlba + 1)) || ((lba + num_blocks) < lba)) { ctl_set_lba_out_of_range(ctsio, MAX(lba, lun->be_lun->maxlba + 1)); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if (num_blocks != 0) endnz = range + 1; } /* * Block backend can not handle zero last range. * Filter it out and return if there is nothing left. */ len = (uint8_t *)endnz - (uint8_t *)buf; if (len == 0) { ctl_set_success(ctsio); goto done; } mtx_lock(&lun->lun_lock); ptrlen = (struct ctl_ptr_len_flags *) &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; ptrlen->ptr = (void *)buf; ptrlen->len = len; ptrlen->flags = byte2; ctl_check_blocked(lun); mtx_unlock(&lun->lun_lock); retval = lun->backend->config_write((union ctl_io *)ctsio); return (retval); done: if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) { free(ctsio->kern_data_ptr, M_CTL); ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED; } ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_default_page_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr) { struct ctl_lun *lun = CTL_LUN(ctsio); uint8_t *current_cp; int set_ua; uint32_t initidx; initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); set_ua = 0; current_cp = (page_index->page_data + (page_index->page_len * CTL_PAGE_CURRENT)); mtx_lock(&lun->lun_lock); if (memcmp(current_cp, page_ptr, page_index->page_len)) { memcpy(current_cp, page_ptr, page_index->page_len); set_ua = 1; } if (set_ua != 0) ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE); mtx_unlock(&lun->lun_lock); if (set_ua) { ctl_isc_announce_mode(lun, ctl_get_initindex(&ctsio->io_hdr.nexus), page_index->page_code, page_index->subpage); } return (CTL_RETVAL_COMPLETE); } static void ctl_ie_timer(void *arg) { struct ctl_lun *lun = arg; uint64_t t; if (lun->ie_asc == 0) return; if (lun->MODE_IE.mrie == SIEP_MRIE_UA) ctl_est_ua_all(lun, -1, CTL_UA_IE); else lun->ie_reported = 0; if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) { lun->ie_reportcnt++; t = scsi_4btoul(lun->MODE_IE.interval_timer); if (t == 0 || t == UINT32_MAX) t = 3000; /* 5 min */ callout_schedule(&lun->ie_callout, t * hz / 10); } } int ctl_ie_page_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_info_exceptions_page *pg; uint64_t t; (void)ctl_default_page_handler(ctsio, page_index, page_ptr); pg = (struct scsi_info_exceptions_page *)page_ptr; mtx_lock(&lun->lun_lock); if (pg->info_flags & SIEP_FLAGS_TEST) { lun->ie_asc = 0x5d; lun->ie_ascq = 0xff; if (pg->mrie == SIEP_MRIE_UA) { ctl_est_ua_all(lun, -1, CTL_UA_IE); lun->ie_reported = 1; } else { ctl_clr_ua_all(lun, -1, CTL_UA_IE); lun->ie_reported = -1; } lun->ie_reportcnt = 1; if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) { lun->ie_reportcnt++; t = scsi_4btoul(pg->interval_timer); if (t == 0 || t == UINT32_MAX) t = 3000; /* 5 min */ callout_reset(&lun->ie_callout, t * hz / 10, ctl_ie_timer, lun); } } else { lun->ie_asc = 0; lun->ie_ascq = 0; lun->ie_reported = 1; ctl_clr_ua_all(lun, -1, CTL_UA_IE); lun->ie_reportcnt = UINT32_MAX; callout_stop(&lun->ie_callout); } mtx_unlock(&lun->lun_lock); return (CTL_RETVAL_COMPLETE); } static int ctl_do_mode_select(union ctl_io *io) { struct ctl_lun *lun = CTL_LUN(io); struct scsi_mode_page_header *page_header; struct ctl_page_index *page_index; struct ctl_scsiio *ctsio; int page_len, page_len_offset, page_len_size; union ctl_modepage_info *modepage_info; uint16_t *len_left, *len_used; int retval, i; ctsio = &io->scsiio; page_index = NULL; page_len = 0; modepage_info = (union ctl_modepage_info *) ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes; len_left = &modepage_info->header.len_left; len_used = &modepage_info->header.len_used; do_next_page: page_header = (struct scsi_mode_page_header *) (ctsio->kern_data_ptr + *len_used); if (*len_left == 0) { free(ctsio->kern_data_ptr, M_CTL); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } else if (*len_left < sizeof(struct scsi_mode_page_header)) { free(ctsio->kern_data_ptr, M_CTL); ctl_set_param_len_error(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } else if ((page_header->page_code & SMPH_SPF) && (*len_left < sizeof(struct scsi_mode_page_header_sp))) { free(ctsio->kern_data_ptr, M_CTL); ctl_set_param_len_error(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * XXX KDM should we do something with the block descriptor? */ for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { page_index = &lun->mode_pages.index[i]; if (lun->be_lun->lun_type == T_DIRECT && (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0) continue; if (lun->be_lun->lun_type == T_PROCESSOR && (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0) continue; if (lun->be_lun->lun_type == T_CDROM && (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0) continue; if ((page_index->page_code & SMPH_PC_MASK) != (page_header->page_code & SMPH_PC_MASK)) continue; /* * If neither page has a subpage code, then we've got a * match. */ if (((page_index->page_code & SMPH_SPF) == 0) && ((page_header->page_code & SMPH_SPF) == 0)) { page_len = page_header->page_length; break; } /* * If both pages have subpages, then the subpage numbers * have to match. */ if ((page_index->page_code & SMPH_SPF) && (page_header->page_code & SMPH_SPF)) { struct scsi_mode_page_header_sp *sph; sph = (struct scsi_mode_page_header_sp *)page_header; if (page_index->subpage == sph->subpage) { page_len = scsi_2btoul(sph->page_length); break; } } } /* * If we couldn't find the page, or if we don't have a mode select * handler for it, send back an error to the user. */ if ((i >= CTL_NUM_MODE_PAGES) || (page_index->select_handler == NULL)) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ *len_used, /*bit_valid*/ 0, /*bit*/ 0); free(ctsio->kern_data_ptr, M_CTL); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if (page_index->page_code & SMPH_SPF) { page_len_offset = 2; page_len_size = 2; } else { page_len_size = 1; page_len_offset = 1; } /* * If the length the initiator gives us isn't the one we specify in * the mode page header, or if they didn't specify enough data in * the CDB to avoid truncating this page, kick out the request. */ if (page_len != page_index->page_len - page_len_offset - page_len_size) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ *len_used + page_len_offset, /*bit_valid*/ 0, /*bit*/ 0); free(ctsio->kern_data_ptr, M_CTL); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if (*len_left < page_index->page_len) { free(ctsio->kern_data_ptr, M_CTL); ctl_set_param_len_error(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * Run through the mode page, checking to make sure that the bits * the user changed are actually legal for him to change. */ for (i = 0; i < page_index->page_len; i++) { uint8_t *user_byte, *change_mask, *current_byte; int bad_bit; int j; user_byte = (uint8_t *)page_header + i; change_mask = page_index->page_data + (page_index->page_len * CTL_PAGE_CHANGEABLE) + i; current_byte = page_index->page_data + (page_index->page_len * CTL_PAGE_CURRENT) + i; /* * Check to see whether the user set any bits in this byte * that he is not allowed to set. */ if ((*user_byte & ~(*change_mask)) == (*current_byte & ~(*change_mask))) continue; /* * Go through bit by bit to determine which one is illegal. */ bad_bit = 0; for (j = 7; j >= 0; j--) { if ((((1 << i) & ~(*change_mask)) & *user_byte) != (((1 << i) & ~(*change_mask)) & *current_byte)) { bad_bit = i; break; } } ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ *len_used + i, /*bit_valid*/ 1, /*bit*/ bad_bit); free(ctsio->kern_data_ptr, M_CTL); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * Decrement these before we call the page handler, since we may * end up getting called back one way or another before the handler * returns to this context. */ *len_left -= page_index->page_len; *len_used += page_index->page_len; retval = page_index->select_handler(ctsio, page_index, (uint8_t *)page_header); /* * If the page handler returns CTL_RETVAL_QUEUED, then we need to * wait until this queued command completes to finish processing * the mode page. If it returns anything other than * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have * already set the sense information, freed the data pointer, and * completed the io for us. */ if (retval != CTL_RETVAL_COMPLETE) goto bailout_no_done; /* * If the initiator sent us more than one page, parse the next one. */ if (*len_left > 0) goto do_next_page; ctl_set_success(ctsio); free(ctsio->kern_data_ptr, M_CTL); ctl_done((union ctl_io *)ctsio); bailout_no_done: return (CTL_RETVAL_COMPLETE); } int ctl_mode_select(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); union ctl_modepage_info *modepage_info; int bd_len, i, header_size, param_len, pf, rtd, sp; uint32_t initidx; initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); switch (ctsio->cdb[0]) { case MODE_SELECT_6: { struct scsi_mode_select_6 *cdb; cdb = (struct scsi_mode_select_6 *)ctsio->cdb; pf = (cdb->byte2 & SMS_PF) ? 1 : 0; rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0; sp = (cdb->byte2 & SMS_SP) ? 1 : 0; param_len = cdb->length; header_size = sizeof(struct scsi_mode_header_6); break; } case MODE_SELECT_10: { struct scsi_mode_select_10 *cdb; cdb = (struct scsi_mode_select_10 *)ctsio->cdb; pf = (cdb->byte2 & SMS_PF) ? 1 : 0; rtd = (cdb->byte2 & SMS_RTD) ? 1 : 0; sp = (cdb->byte2 & SMS_SP) ? 1 : 0; param_len = scsi_2btoul(cdb->length); header_size = sizeof(struct scsi_mode_header_10); break; } default: ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if (rtd) { if (param_len != 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 0, /*command*/ 1, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* Revert to defaults. */ ctl_init_page_index(lun); mtx_lock(&lun->lun_lock); ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE); mtx_unlock(&lun->lun_lock); for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { ctl_isc_announce_mode(lun, -1, lun->mode_pages.index[i].page_code & SMPH_PC_MASK, lun->mode_pages.index[i].subpage); } ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * From SPC-3: * "A parameter list length of zero indicates that the Data-Out Buffer * shall be empty. This condition shall not be considered as an error." */ if (param_len == 0) { ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * Since we'll hit this the first time through, prior to * allocation, we don't need to free a data buffer here. */ if (param_len < header_size) { ctl_set_param_len_error(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * Allocate the data buffer and grab the user's data. In theory, * we shouldn't have to sanity check the parameter list length here * because the maximum size is 64K. We should be able to malloc * that much without too many problems. */ if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK); ctsio->kern_data_len = param_len; ctsio->kern_total_len = param_len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } switch (ctsio->cdb[0]) { case MODE_SELECT_6: { struct scsi_mode_header_6 *mh6; mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr; bd_len = mh6->blk_desc_len; break; } case MODE_SELECT_10: { struct scsi_mode_header_10 *mh10; mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr; bd_len = scsi_2btoul(mh10->blk_desc_len); break; } default: panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]); } if (param_len < (header_size + bd_len)) { free(ctsio->kern_data_ptr, M_CTL); ctl_set_param_len_error(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * Set the IO_CONT flag, so that if this I/O gets passed to * ctl_config_write_done(), it'll get passed back to * ctl_do_mode_select() for further processing, or completion if * we're all done. */ ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT; ctsio->io_cont = ctl_do_mode_select; modepage_info = (union ctl_modepage_info *) ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes; memset(modepage_info, 0, sizeof(*modepage_info)); modepage_info->header.len_left = param_len - header_size - bd_len; modepage_info->header.len_used = header_size + bd_len; return (ctl_do_mode_select((union ctl_io *)ctsio)); } int ctl_mode_sense(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); int pc, page_code, dbd, llba, subpage; int alloc_len, page_len, header_len, total_len; struct scsi_mode_block_descr *block_desc; struct ctl_page_index *page_index; dbd = 0; llba = 0; block_desc = NULL; CTL_DEBUG_PRINT(("ctl_mode_sense\n")); switch (ctsio->cdb[0]) { case MODE_SENSE_6: { struct scsi_mode_sense_6 *cdb; cdb = (struct scsi_mode_sense_6 *)ctsio->cdb; header_len = sizeof(struct scsi_mode_hdr_6); if (cdb->byte2 & SMS_DBD) dbd = 1; else header_len += sizeof(struct scsi_mode_block_descr); pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6; page_code = cdb->page & SMS_PAGE_CODE; subpage = cdb->subpage; alloc_len = cdb->length; break; } case MODE_SENSE_10: { struct scsi_mode_sense_10 *cdb; cdb = (struct scsi_mode_sense_10 *)ctsio->cdb; header_len = sizeof(struct scsi_mode_hdr_10); if (cdb->byte2 & SMS_DBD) dbd = 1; else header_len += sizeof(struct scsi_mode_block_descr); if (cdb->byte2 & SMS10_LLBAA) llba = 1; pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6; page_code = cdb->page & SMS_PAGE_CODE; subpage = cdb->subpage; alloc_len = scsi_2btoul(cdb->length); break; } default: ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); break; /* NOTREACHED */ } /* * We have to make a first pass through to calculate the size of * the pages that match the user's query. Then we allocate enough * memory to hold it, and actually copy the data into the buffer. */ switch (page_code) { case SMS_ALL_PAGES_PAGE: { u_int i; page_len = 0; /* * At the moment, values other than 0 and 0xff here are * reserved according to SPC-3. */ if ((subpage != SMS_SUBPAGE_PAGE_0) && (subpage != SMS_SUBPAGE_ALL)) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 3, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { page_index = &lun->mode_pages.index[i]; /* Make sure the page is supported for this dev type */ if (lun->be_lun->lun_type == T_DIRECT && (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0) continue; if (lun->be_lun->lun_type == T_PROCESSOR && (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0) continue; if (lun->be_lun->lun_type == T_CDROM && (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0) continue; /* * We don't use this subpage if the user didn't * request all subpages. */ if ((page_index->subpage != 0) && (subpage == SMS_SUBPAGE_PAGE_0)) continue; #if 0 printf("found page %#x len %d\n", page_index->page_code & SMPH_PC_MASK, page_index->page_len); #endif page_len += page_index->page_len; } break; } default: { u_int i; page_len = 0; for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { page_index = &lun->mode_pages.index[i]; /* Make sure the page is supported for this dev type */ if (lun->be_lun->lun_type == T_DIRECT && (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0) continue; if (lun->be_lun->lun_type == T_PROCESSOR && (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0) continue; if (lun->be_lun->lun_type == T_CDROM && (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0) continue; /* Look for the right page code */ if ((page_index->page_code & SMPH_PC_MASK) != page_code) continue; /* Look for the right subpage or the subpage wildcard*/ if ((page_index->subpage != subpage) && (subpage != SMS_SUBPAGE_ALL)) continue; #if 0 printf("found page %#x len %d\n", page_index->page_code & SMPH_PC_MASK, page_index->page_len); #endif page_len += page_index->page_len; } if (page_len == 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 5); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } break; } } total_len = header_len + page_len; #if 0 printf("header_len = %d, page_len = %d, total_len = %d\n", header_len, page_len, total_len); #endif ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; switch (ctsio->cdb[0]) { case MODE_SENSE_6: { struct scsi_mode_hdr_6 *header; header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr; header->datalen = MIN(total_len - 1, 254); if (lun->be_lun->lun_type == T_DIRECT) { header->dev_specific = 0x10; /* DPOFUA */ if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) || (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) header->dev_specific |= 0x80; /* WP */ } if (dbd) header->block_descr_len = 0; else header->block_descr_len = sizeof(struct scsi_mode_block_descr); block_desc = (struct scsi_mode_block_descr *)&header[1]; break; } case MODE_SENSE_10: { struct scsi_mode_hdr_10 *header; int datalen; header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr; datalen = MIN(total_len - 2, 65533); scsi_ulto2b(datalen, header->datalen); if (lun->be_lun->lun_type == T_DIRECT) { header->dev_specific = 0x10; /* DPOFUA */ if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) || (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) header->dev_specific |= 0x80; /* WP */ } if (dbd) scsi_ulto2b(0, header->block_descr_len); else scsi_ulto2b(sizeof(struct scsi_mode_block_descr), header->block_descr_len); block_desc = (struct scsi_mode_block_descr *)&header[1]; break; } default: panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]); } /* * If we've got a disk, use its blocksize in the block * descriptor. Otherwise, just set it to 0. */ if (dbd == 0) { if (lun->be_lun->lun_type == T_DIRECT) scsi_ulto3b(lun->be_lun->blocksize, block_desc->block_len); else scsi_ulto3b(0, block_desc->block_len); } switch (page_code) { case SMS_ALL_PAGES_PAGE: { int i, data_used; data_used = header_len; for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { struct ctl_page_index *page_index; page_index = &lun->mode_pages.index[i]; if (lun->be_lun->lun_type == T_DIRECT && (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0) continue; if (lun->be_lun->lun_type == T_PROCESSOR && (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0) continue; if (lun->be_lun->lun_type == T_CDROM && (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0) continue; /* * We don't use this subpage if the user didn't * request all subpages. We already checked (above) * to make sure the user only specified a subpage * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case. */ if ((page_index->subpage != 0) && (subpage == SMS_SUBPAGE_PAGE_0)) continue; /* * Call the handler, if it exists, to update the * page to the latest values. */ if (page_index->sense_handler != NULL) page_index->sense_handler(ctsio, page_index,pc); memcpy(ctsio->kern_data_ptr + data_used, page_index->page_data + (page_index->page_len * pc), page_index->page_len); data_used += page_index->page_len; } break; } default: { int i, data_used; data_used = header_len; for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { struct ctl_page_index *page_index; page_index = &lun->mode_pages.index[i]; /* Look for the right page code */ if ((page_index->page_code & SMPH_PC_MASK) != page_code) continue; /* Look for the right subpage or the subpage wildcard*/ if ((page_index->subpage != subpage) && (subpage != SMS_SUBPAGE_ALL)) continue; /* Make sure the page is supported for this dev type */ if (lun->be_lun->lun_type == T_DIRECT && (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0) continue; if (lun->be_lun->lun_type == T_PROCESSOR && (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0) continue; if (lun->be_lun->lun_type == T_CDROM && (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0) continue; /* * Call the handler, if it exists, to update the * page to the latest values. */ if (page_index->sense_handler != NULL) page_index->sense_handler(ctsio, page_index,pc); memcpy(ctsio->kern_data_ptr + data_used, page_index->page_data + (page_index->page_len * pc), page_index->page_len); data_used += page_index->page_len; } break; } } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_log_param_header *phdr; uint8_t *data; uint64_t val; data = page_index->page_data; if (lun->backend->lun_attr != NULL && (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail")) != UINT64_MAX) { phdr = (struct scsi_log_param_header *)data; scsi_ulto2b(0x0001, phdr->param_code); phdr->param_control = SLP_LBIN | SLP_LP; phdr->param_len = 8; data = (uint8_t *)(phdr + 1); scsi_ulto4b(val >> CTL_LBP_EXPONENT, data); data[4] = 0x02; /* per-pool */ data += phdr->param_len; } if (lun->backend->lun_attr != NULL && (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused")) != UINT64_MAX) { phdr = (struct scsi_log_param_header *)data; scsi_ulto2b(0x0002, phdr->param_code); phdr->param_control = SLP_LBIN | SLP_LP; phdr->param_len = 8; data = (uint8_t *)(phdr + 1); scsi_ulto4b(val >> CTL_LBP_EXPONENT, data); data[4] = 0x01; /* per-LUN */ data += phdr->param_len; } if (lun->backend->lun_attr != NULL && (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail")) != UINT64_MAX) { phdr = (struct scsi_log_param_header *)data; scsi_ulto2b(0x00f1, phdr->param_code); phdr->param_control = SLP_LBIN | SLP_LP; phdr->param_len = 8; data = (uint8_t *)(phdr + 1); scsi_ulto4b(val >> CTL_LBP_EXPONENT, data); data[4] = 0x02; /* per-pool */ data += phdr->param_len; } if (lun->backend->lun_attr != NULL && (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused")) != UINT64_MAX) { phdr = (struct scsi_log_param_header *)data; scsi_ulto2b(0x00f2, phdr->param_code); phdr->param_control = SLP_LBIN | SLP_LP; phdr->param_len = 8; data = (uint8_t *)(phdr + 1); scsi_ulto4b(val >> CTL_LBP_EXPONENT, data); data[4] = 0x02; /* per-pool */ data += phdr->param_len; } page_index->page_len = data - page_index->page_data; return (0); } int ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc) { struct ctl_lun *lun = CTL_LUN(ctsio); struct stat_page *data; struct bintime *t; data = (struct stat_page *)page_index->page_data; scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code); data->sap.hdr.param_control = SLP_LBIN; data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) - sizeof(struct scsi_log_param_header); scsi_u64to8b(lun->stats.operations[CTL_STATS_READ], data->sap.read_num); scsi_u64to8b(lun->stats.operations[CTL_STATS_WRITE], data->sap.write_num); if (lun->be_lun->blocksize > 0) { scsi_u64to8b(lun->stats.bytes[CTL_STATS_WRITE] / lun->be_lun->blocksize, data->sap.recvieved_lba); scsi_u64to8b(lun->stats.bytes[CTL_STATS_READ] / lun->be_lun->blocksize, data->sap.transmitted_lba); } t = &lun->stats.time[CTL_STATS_READ]; scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000), data->sap.read_int); t = &lun->stats.time[CTL_STATS_WRITE]; scsi_u64to8b((uint64_t)t->sec * 1000 + t->frac / (UINT64_MAX / 1000), data->sap.write_int); scsi_u64to8b(0, data->sap.weighted_num); scsi_u64to8b(0, data->sap.weighted_int); scsi_ulto2b(SLP_IT, data->it.hdr.param_code); data->it.hdr.param_control = SLP_LBIN; data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) - sizeof(struct scsi_log_param_header); #ifdef CTL_TIME_IO scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int); #endif scsi_ulto2b(SLP_TI, data->ti.hdr.param_code); data->it.hdr.param_control = SLP_LBIN; data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) - sizeof(struct scsi_log_param_header); scsi_ulto4b(3, data->ti.exponent); scsi_ulto4b(1, data->ti.integer); return (0); } int ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_log_informational_exceptions *data; data = (struct scsi_log_informational_exceptions *)page_index->page_data; scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code); data->hdr.param_control = SLP_LBIN; data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) - sizeof(struct scsi_log_param_header); data->ie_asc = lun->ie_asc; data->ie_ascq = lun->ie_ascq; data->temperature = 0xff; return (0); } int ctl_log_sense(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); int i, pc, page_code, subpage; int alloc_len, total_len; struct ctl_page_index *page_index; struct scsi_log_sense *cdb; struct scsi_log_header *header; CTL_DEBUG_PRINT(("ctl_log_sense\n")); cdb = (struct scsi_log_sense *)ctsio->cdb; pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6; page_code = cdb->page & SLS_PAGE_CODE; subpage = cdb->subpage; alloc_len = scsi_2btoul(cdb->length); page_index = NULL; for (i = 0; i < CTL_NUM_LOG_PAGES; i++) { page_index = &lun->log_pages.index[i]; /* Look for the right page code */ if ((page_index->page_code & SL_PAGE_CODE) != page_code) continue; /* Look for the right subpage or the subpage wildcard*/ if (page_index->subpage != subpage) continue; break; } if (i >= CTL_NUM_LOG_PAGES) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } total_len = sizeof(struct scsi_log_header) + page_index->page_len; ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; header = (struct scsi_log_header *)ctsio->kern_data_ptr; header->page = page_index->page_code; if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING) header->page |= SL_DS; if (page_index->subpage) { header->page |= SL_SPF; header->subpage = page_index->subpage; } scsi_ulto2b(page_index->page_len, header->datalen); /* * Call the handler, if it exists, to update the * page to the latest values. */ if (page_index->sense_handler != NULL) page_index->sense_handler(ctsio, page_index, pc); memcpy(header + 1, page_index->page_data, page_index->page_len); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_read_capacity(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_read_capacity *cdb; struct scsi_read_capacity_data *data; uint32_t lba; CTL_DEBUG_PRINT(("ctl_read_capacity\n")); cdb = (struct scsi_read_capacity *)ctsio->cdb; lba = scsi_4btoul(cdb->addr); if (((cdb->pmi & SRC_PMI) == 0) && (lba != 0)) { ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO); data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr; ctsio->kern_data_len = sizeof(*data); ctsio->kern_total_len = sizeof(*data); ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * If the maximum LBA is greater than 0xfffffffe, the user must * issue a SERVICE ACTION IN (16) command, with the read capacity * serivce action set. */ if (lun->be_lun->maxlba > 0xfffffffe) scsi_ulto4b(0xffffffff, data->addr); else scsi_ulto4b(lun->be_lun->maxlba, data->addr); /* * XXX KDM this may not be 512 bytes... */ scsi_ulto4b(lun->be_lun->blocksize, data->length); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_read_capacity_16(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_read_capacity_16 *cdb; struct scsi_read_capacity_data_long *data; uint64_t lba; uint32_t alloc_len; CTL_DEBUG_PRINT(("ctl_read_capacity_16\n")); cdb = (struct scsi_read_capacity_16 *)ctsio->cdb; alloc_len = scsi_4btoul(cdb->alloc_len); lba = scsi_8btou64(cdb->addr); if ((cdb->reladr & SRC16_PMI) && (lba != 0)) { ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO); data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(sizeof(*data), alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; scsi_u64to8b(lun->be_lun->maxlba, data->addr); /* XXX KDM this may not be 512 bytes... */ scsi_ulto4b(lun->be_lun->blocksize, data->length); data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE; scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp); if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_get_lba_status(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_get_lba_status *cdb; struct scsi_get_lba_status_data *data; struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t alloc_len, total_len; int retval; CTL_DEBUG_PRINT(("ctl_get_lba_status\n")); cdb = (struct scsi_get_lba_status *)ctsio->cdb; lba = scsi_8btou64(cdb->addr); alloc_len = scsi_4btoul(cdb->alloc_len); if (lba > lun->be_lun->maxlba) { ctl_set_lba_out_of_range(ctsio, lba); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } total_len = sizeof(*data) + sizeof(data->descr[0]); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* Fill dummy data in case backend can't tell anything. */ scsi_ulto4b(4 + sizeof(data->descr[0]), data->length); scsi_u64to8b(lba, data->descr[0].addr); scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba), data->descr[0].length); data->descr[0].status = 0; /* Mapped or unknown. */ ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; lbalen->lba = lba; lbalen->len = total_len; lbalen->flags = 0; retval = lun->backend->config_read((union ctl_io *)ctsio); return (retval); } int ctl_read_defect(struct ctl_scsiio *ctsio) { struct scsi_read_defect_data_10 *ccb10; struct scsi_read_defect_data_12 *ccb12; struct scsi_read_defect_data_hdr_10 *data10; struct scsi_read_defect_data_hdr_12 *data12; uint32_t alloc_len, data_len; uint8_t format; CTL_DEBUG_PRINT(("ctl_read_defect\n")); if (ctsio->cdb[0] == READ_DEFECT_DATA_10) { ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb; format = ccb10->format; alloc_len = scsi_2btoul(ccb10->alloc_length); data_len = sizeof(*data10); } else { ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb; format = ccb12->format; alloc_len = scsi_4btoul(ccb12->alloc_length); data_len = sizeof(*data12); } if (alloc_len == 0) { ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; if (ctsio->cdb[0] == READ_DEFECT_DATA_10) { data10 = (struct scsi_read_defect_data_hdr_10 *) ctsio->kern_data_ptr; data10->format = format; scsi_ulto2b(0, data10->length); } else { data12 = (struct scsi_read_defect_data_hdr_12 *) ctsio->kern_data_ptr; data12->format = format; scsi_ulto2b(0, data12->generation); scsi_ulto4b(0, data12->length); } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_maintenance_in *cdb; int retval; int alloc_len, ext, total_len = 0, g, pc, pg, ts, os; int num_ha_groups, num_target_ports, shared_group; struct ctl_port *port; struct scsi_target_group_data *rtg_ptr; struct scsi_target_group_data_extended *rtg_ext_ptr; struct scsi_target_port_group_descriptor *tpg_desc; CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n")); cdb = (struct scsi_maintenance_in *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; switch (cdb->byte2 & STG_PDF_MASK) { case STG_PDF_LENGTH: ext = 0; break; case STG_PDF_EXTENDED: ext = 1; break; default: ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 5); ctl_done((union ctl_io *)ctsio); return(retval); } num_target_ports = 0; shared_group = (softc->is_single != 0); mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0) continue; if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; num_target_ports++; if (port->status & CTL_PORT_STATUS_HA_SHARED) shared_group = 1; } mtx_unlock(&softc->ctl_lock); num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES; if (ext) total_len = sizeof(struct scsi_target_group_data_extended); else total_len = sizeof(struct scsi_target_group_data); total_len += sizeof(struct scsi_target_port_group_descriptor) * (shared_group + num_ha_groups) + sizeof(struct scsi_target_port_descriptor) * num_target_ports; alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; if (ext) { rtg_ext_ptr = (struct scsi_target_group_data_extended *) ctsio->kern_data_ptr; scsi_ulto4b(total_len - 4, rtg_ext_ptr->length); rtg_ext_ptr->format_type = 0x10; rtg_ext_ptr->implicit_transition_time = 0; tpg_desc = &rtg_ext_ptr->groups[0]; } else { rtg_ptr = (struct scsi_target_group_data *) ctsio->kern_data_ptr; scsi_ulto4b(total_len - 4, rtg_ptr->length); tpg_desc = &rtg_ptr->groups[0]; } mtx_lock(&softc->ctl_lock); pg = softc->port_min / softc->port_cnt; if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) { /* Some shelf is known to be primary. */ if (softc->ha_link == CTL_HA_LINK_OFFLINE) os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE; else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING; else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) os = TPG_ASYMMETRIC_ACCESS_STANDBY; else os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED; if (lun->flags & CTL_LUN_PRIMARY_SC) { ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; } else { ts = os; os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; } } else { /* No known primary shelf. */ if (softc->ha_link == CTL_HA_LINK_OFFLINE) { ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE; os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) { ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING; os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; } else { ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING; } } if (shared_group) { tpg_desc->pref_state = ts; tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP | TPG_U_SUP | TPG_T_SUP; scsi_ulto2b(1, tpg_desc->target_port_group); tpg_desc->status = TPG_IMPLICIT; pc = 0; STAILQ_FOREACH(port, &softc->port_list, links) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0) continue; if (!softc->is_single && (port->status & CTL_PORT_STATUS_HA_SHARED) == 0) continue; if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc]. relative_target_port_identifier); pc++; } tpg_desc->target_port_count = pc; tpg_desc = (struct scsi_target_port_group_descriptor *) &tpg_desc->descriptors[pc]; } for (g = 0; g < num_ha_groups; g++) { tpg_desc->pref_state = (g == pg) ? ts : os; tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP | TPG_U_SUP | TPG_T_SUP; scsi_ulto2b(2 + g, tpg_desc->target_port_group); tpg_desc->status = TPG_IMPLICIT; pc = 0; STAILQ_FOREACH(port, &softc->port_list, links) { if (port->targ_port < g * softc->port_cnt || port->targ_port >= (g + 1) * softc->port_cnt) continue; if ((port->status & CTL_PORT_STATUS_ONLINE) == 0) continue; if (port->status & CTL_PORT_STATUS_HA_SHARED) continue; if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc]. relative_target_port_identifier); pc++; } tpg_desc->target_port_count = pc; tpg_desc = (struct scsi_target_port_group_descriptor *) &tpg_desc->descriptors[pc]; } mtx_unlock(&softc->ctl_lock); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return(retval); } int ctl_report_supported_opcodes(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_report_supported_opcodes *cdb; const struct ctl_cmd_entry *entry, *sentry; struct scsi_report_supported_opcodes_all *all; struct scsi_report_supported_opcodes_descr *descr; struct scsi_report_supported_opcodes_one *one; int retval; int alloc_len, total_len; int opcode, service_action, i, j, num; CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n")); cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; opcode = cdb->requested_opcode; service_action = scsi_2btoul(cdb->requested_service_action); switch (cdb->options & RSO_OPTIONS_MASK) { case RSO_OPTIONS_ALL: num = 0; for (i = 0; i < 256; i++) { entry = &ctl_cmd_table[i]; if (entry->flags & CTL_CMD_FLAG_SA5) { for (j = 0; j < 32; j++) { sentry = &((const struct ctl_cmd_entry *) entry->execute)[j]; if (ctl_cmd_applicable( lun->be_lun->lun_type, sentry)) num++; } } else { if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) num++; } } total_len = sizeof(struct scsi_report_supported_opcodes_all) + num * sizeof(struct scsi_report_supported_opcodes_descr); break; case RSO_OPTIONS_OC: if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) { ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 2); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32; break; case RSO_OPTIONS_OC_SA: if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 || service_action >= 32) { ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 2); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* FALLTHROUGH */ case RSO_OPTIONS_OC_ASA: total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32; break; default: ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 2); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; switch (cdb->options & RSO_OPTIONS_MASK) { case RSO_OPTIONS_ALL: all = (struct scsi_report_supported_opcodes_all *) ctsio->kern_data_ptr; num = 0; for (i = 0; i < 256; i++) { entry = &ctl_cmd_table[i]; if (entry->flags & CTL_CMD_FLAG_SA5) { for (j = 0; j < 32; j++) { sentry = &((const struct ctl_cmd_entry *) entry->execute)[j]; if (!ctl_cmd_applicable( lun->be_lun->lun_type, sentry)) continue; descr = &all->descr[num++]; descr->opcode = i; scsi_ulto2b(j, descr->service_action); descr->flags = RSO_SERVACTV; scsi_ulto2b(sentry->length, descr->cdb_length); } } else { if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) continue; descr = &all->descr[num++]; descr->opcode = i; scsi_ulto2b(0, descr->service_action); descr->flags = 0; scsi_ulto2b(entry->length, descr->cdb_length); } } scsi_ulto4b( num * sizeof(struct scsi_report_supported_opcodes_descr), all->length); break; case RSO_OPTIONS_OC: one = (struct scsi_report_supported_opcodes_one *) ctsio->kern_data_ptr; entry = &ctl_cmd_table[opcode]; goto fill_one; case RSO_OPTIONS_OC_SA: one = (struct scsi_report_supported_opcodes_one *) ctsio->kern_data_ptr; entry = &ctl_cmd_table[opcode]; entry = &((const struct ctl_cmd_entry *) entry->execute)[service_action]; fill_one: if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) { one->support = 3; scsi_ulto2b(entry->length, one->cdb_length); one->cdb_usage[0] = opcode; memcpy(&one->cdb_usage[1], entry->usage, entry->length - 1); } else one->support = 1; break; case RSO_OPTIONS_OC_ASA: one = (struct scsi_report_supported_opcodes_one *) ctsio->kern_data_ptr; entry = &ctl_cmd_table[opcode]; if (entry->flags & CTL_CMD_FLAG_SA5) { entry = &((const struct ctl_cmd_entry *) entry->execute)[service_action]; } else if (service_action != 0) { one->support = 1; break; } goto fill_one; } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return(retval); } int ctl_report_supported_tmf(struct ctl_scsiio *ctsio) { struct scsi_report_supported_tmf *cdb; struct scsi_report_supported_tmf_ext_data *data; int retval; int alloc_len, total_len; CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n")); cdb = (struct scsi_report_supported_tmf *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; if (cdb->options & RST_REPD) total_len = sizeof(struct scsi_report_supported_tmf_ext_data); else total_len = sizeof(struct scsi_report_supported_tmf_data); alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr; data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS | RST_TRS; data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS; data->length = total_len - 4; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } int ctl_report_timestamp(struct ctl_scsiio *ctsio) { struct scsi_report_timestamp *cdb; struct scsi_report_timestamp_data *data; struct timeval tv; int64_t timestamp; int retval; int alloc_len, total_len; CTL_DEBUG_PRINT(("ctl_report_timestamp\n")); cdb = (struct scsi_report_timestamp *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; total_len = sizeof(struct scsi_report_timestamp_data); alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr; scsi_ulto2b(sizeof(*data) - 2, data->length); data->origin = RTS_ORIG_OUTSIDE; getmicrotime(&tv); timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000; scsi_ulto4b(timestamp >> 16, data->timestamp); scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } int ctl_persistent_reserve_in(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_per_res_in *cdb; int alloc_len, total_len = 0; /* struct scsi_per_res_in_rsrv in_data; */ uint64_t key; CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n")); cdb = (struct scsi_per_res_in *)ctsio->cdb; alloc_len = scsi_2btoul(cdb->length); retry: mtx_lock(&lun->lun_lock); switch (cdb->action) { case SPRI_RK: /* read keys */ total_len = sizeof(struct scsi_per_res_in_keys) + lun->pr_key_count * sizeof(struct scsi_per_res_key); break; case SPRI_RR: /* read reservation */ if (lun->flags & CTL_LUN_PR_RESERVED) total_len = sizeof(struct scsi_per_res_in_rsrv); else total_len = sizeof(struct scsi_per_res_in_header); break; case SPRI_RC: /* report capabilities */ total_len = sizeof(struct scsi_per_res_cap); break; case SPRI_RS: /* read full status */ total_len = sizeof(struct scsi_per_res_in_header) + (sizeof(struct scsi_per_res_in_full_desc) + 256) * lun->pr_key_count; break; default: panic("%s: Invalid PR type %#x", __func__, cdb->action); } mtx_unlock(&lun->lun_lock); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; mtx_lock(&lun->lun_lock); switch (cdb->action) { case SPRI_RK: { // read keys struct scsi_per_res_in_keys *res_keys; int i, key_count; res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr; /* * We had to drop the lock to allocate our buffer, which * leaves time for someone to come in with another * persistent reservation. (That is unlikely, though, * since this should be the only persistent reservation * command active right now.) */ if (total_len != (sizeof(struct scsi_per_res_in_keys) + (lun->pr_key_count * sizeof(struct scsi_per_res_key)))){ mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); printf("%s: reservation length changed, retrying\n", __func__); goto retry; } scsi_ulto4b(lun->pr_generation, res_keys->header.generation); scsi_ulto4b(sizeof(struct scsi_per_res_key) * lun->pr_key_count, res_keys->header.length); for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) { if ((key = ctl_get_prkey(lun, i)) == 0) continue; /* * We used lun->pr_key_count to calculate the * size to allocate. If it turns out the number of * initiators with the registered flag set is * larger than that (i.e. they haven't been kept in * sync), we've got a problem. */ if (key_count >= lun->pr_key_count) { key_count++; continue; } scsi_u64to8b(key, res_keys->keys[key_count].key); key_count++; } break; } case SPRI_RR: { // read reservation struct scsi_per_res_in_rsrv *res; int tmp_len, header_only; res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr; scsi_ulto4b(lun->pr_generation, res->header.generation); if (lun->flags & CTL_LUN_PR_RESERVED) { tmp_len = sizeof(struct scsi_per_res_in_rsrv); scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data), res->header.length); header_only = 0; } else { tmp_len = sizeof(struct scsi_per_res_in_header); scsi_ulto4b(0, res->header.length); header_only = 1; } /* * We had to drop the lock to allocate our buffer, which * leaves time for someone to come in with another * persistent reservation. (That is unlikely, though, * since this should be the only persistent reservation * command active right now.) */ if (tmp_len != total_len) { mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); printf("%s: reservation status changed, retrying\n", __func__); goto retry; } /* * No reservation held, so we're done. */ if (header_only != 0) break; /* * If the registration is an All Registrants type, the key * is 0, since it doesn't really matter. */ if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) { scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx), res->data.reservation); } res->data.scopetype = lun->pr_res_type; break; } case SPRI_RC: //report capabilities { struct scsi_per_res_cap *res_cap; uint16_t type_mask; res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr; scsi_ulto2b(sizeof(*res_cap), res_cap->length); res_cap->flags1 = SPRI_CRH; res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5; type_mask = SPRI_TM_WR_EX_AR | SPRI_TM_EX_AC_RO | SPRI_TM_WR_EX_RO | SPRI_TM_EX_AC | SPRI_TM_WR_EX | SPRI_TM_EX_AC_AR; scsi_ulto2b(type_mask, res_cap->type_mask); break; } case SPRI_RS: { // read full status struct scsi_per_res_in_full *res_status; struct scsi_per_res_in_full_desc *res_desc; struct ctl_port *port; int i, len; res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr; /* * We had to drop the lock to allocate our buffer, which * leaves time for someone to come in with another * persistent reservation. (That is unlikely, though, * since this should be the only persistent reservation * command active right now.) */ if (total_len < (sizeof(struct scsi_per_res_in_header) + (sizeof(struct scsi_per_res_in_full_desc) + 256) * lun->pr_key_count)){ mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); printf("%s: reservation length changed, retrying\n", __func__); goto retry; } scsi_ulto4b(lun->pr_generation, res_status->header.generation); res_desc = &res_status->desc[0]; for (i = 0; i < CTL_MAX_INITIATORS; i++) { if ((key = ctl_get_prkey(lun, i)) == 0) continue; scsi_u64to8b(key, res_desc->res_key.key); if ((lun->flags & CTL_LUN_PR_RESERVED) && (lun->pr_res_idx == i || lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) { res_desc->flags = SPRI_FULL_R_HOLDER; res_desc->scopetype = lun->pr_res_type; } scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT, res_desc->rel_trgt_port_id); len = 0; port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT]; if (port != NULL) len = ctl_create_iid(port, i % CTL_MAX_INIT_PER_PORT, res_desc->transport_id); scsi_ulto4b(len, res_desc->additional_length); res_desc = (struct scsi_per_res_in_full_desc *) &res_desc->transport_id[len]; } scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0], res_status->header.length); break; } default: panic("%s: Invalid PR type %#x", __func__, cdb->action); } mtx_unlock(&lun->lun_lock); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if * it should return. */ static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key, uint64_t sa_res_key, uint8_t type, uint32_t residx, struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb, struct scsi_per_res_out_parms* param) { union ctl_ha_msg persis_io; int i; mtx_lock(&lun->lun_lock); if (sa_res_key == 0) { if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) { /* validate scope and type */ if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 4); ctl_done((union ctl_io *)ctsio); return (1); } if (type>8 || type==2 || type==4 || type==0) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (1); } /* * Unregister everybody else and build UA for * them */ for(i = 0; i < CTL_MAX_INITIATORS; i++) { if (i == residx || ctl_get_prkey(lun, i) == 0) continue; ctl_clr_prkey(lun, i); ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } lun->pr_key_count = 1; lun->pr_res_type = type; if (lun->pr_res_type != SPR_TYPE_WR_EX_AR && lun->pr_res_type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = residx; lun->pr_generation++; mtx_unlock(&lun->lun_lock); /* send msg to other side */ persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_PREEMPT; persis_io.pr.pr_info.residx = lun->pr_res_idx; persis_io.pr.pr_info.res_type = type; memcpy(persis_io.pr.pr_info.sa_res_key, param->serv_act_res_key, sizeof(param->serv_act_res_key)); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); } else { /* not all registrants */ mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 8, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (1); } } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS || !(lun->flags & CTL_LUN_PR_RESERVED)) { int found = 0; if (res_key == sa_res_key) { /* special case */ /* * The spec implies this is not good but doesn't * say what to do. There are two choices either * generate a res conflict or check condition * with illegal field in parameter data. Since * that is what is done when the sa_res_key is * zero I'll take that approach since this has * to do with the sa_res_key. */ mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 8, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (1); } for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (ctl_get_prkey(lun, i) != sa_res_key) continue; found = 1; ctl_clr_prkey(lun, i); lun->pr_key_count--; ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } if (!found) { mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } lun->pr_generation++; mtx_unlock(&lun->lun_lock); /* send msg to other side */ persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_PREEMPT; persis_io.pr.pr_info.residx = lun->pr_res_idx; persis_io.pr.pr_info.res_type = type; memcpy(persis_io.pr.pr_info.sa_res_key, param->serv_act_res_key, sizeof(param->serv_act_res_key)); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); } else { /* Reserved but not all registrants */ /* sa_res_key is res holder */ if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) { /* validate scope and type */ if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 4); ctl_done((union ctl_io *)ctsio); return (1); } if (type>8 || type==2 || type==4 || type==0) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (1); } /* * Do the following: * if sa_res_key != res_key remove all * registrants w/sa_res_key and generate UA * for these registrants(Registrations * Preempted) if it wasn't an exclusive * reservation generate UA(Reservations * Preempted) for all other registered nexuses * if the type has changed. Establish the new * reservation and holder. If res_key and * sa_res_key are the same do the above * except don't unregister the res holder. */ for(i = 0; i < CTL_MAX_INITIATORS; i++) { if (i == residx || ctl_get_prkey(lun, i) == 0) continue; if (sa_res_key == ctl_get_prkey(lun, i)) { ctl_clr_prkey(lun, i); lun->pr_key_count--; ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } else if (type != lun->pr_res_type && (lun->pr_res_type == SPR_TYPE_WR_EX_RO || lun->pr_res_type == SPR_TYPE_EX_AC_RO)) { ctl_est_ua(lun, i, CTL_UA_RES_RELEASE); } } lun->pr_res_type = type; if (lun->pr_res_type != SPR_TYPE_WR_EX_AR && lun->pr_res_type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = residx; else lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS; lun->pr_generation++; mtx_unlock(&lun->lun_lock); persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_PREEMPT; persis_io.pr.pr_info.residx = lun->pr_res_idx; persis_io.pr.pr_info.res_type = type; memcpy(persis_io.pr.pr_info.sa_res_key, param->serv_act_res_key, sizeof(param->serv_act_res_key)); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); } else { /* * sa_res_key is not the res holder just * remove registrants */ int found=0; for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (sa_res_key != ctl_get_prkey(lun, i)) continue; found = 1; ctl_clr_prkey(lun, i); lun->pr_key_count--; ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } if (!found) { mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (1); } lun->pr_generation++; mtx_unlock(&lun->lun_lock); persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_PREEMPT; persis_io.pr.pr_info.residx = lun->pr_res_idx; persis_io.pr.pr_info.res_type = type; memcpy(persis_io.pr.pr_info.sa_res_key, param->serv_act_res_key, sizeof(param->serv_act_res_key)); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); } } return (0); } static void ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg) { uint64_t sa_res_key; int i; sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key); if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS || lun->pr_res_idx == CTL_PR_NO_RESERVATION || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) { if (sa_res_key == 0) { /* * Unregister everybody else and build UA for * them */ for(i = 0; i < CTL_MAX_INITIATORS; i++) { if (i == msg->pr.pr_info.residx || ctl_get_prkey(lun, i) == 0) continue; ctl_clr_prkey(lun, i); ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } lun->pr_key_count = 1; lun->pr_res_type = msg->pr.pr_info.res_type; if (lun->pr_res_type != SPR_TYPE_WR_EX_AR && lun->pr_res_type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = msg->pr.pr_info.residx; } else { for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (sa_res_key == ctl_get_prkey(lun, i)) continue; ctl_clr_prkey(lun, i); lun->pr_key_count--; ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } } } else { for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (i == msg->pr.pr_info.residx || ctl_get_prkey(lun, i) == 0) continue; if (sa_res_key == ctl_get_prkey(lun, i)) { ctl_clr_prkey(lun, i); lun->pr_key_count--; ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } else if (msg->pr.pr_info.res_type != lun->pr_res_type && (lun->pr_res_type == SPR_TYPE_WR_EX_RO || lun->pr_res_type == SPR_TYPE_EX_AC_RO)) { ctl_est_ua(lun, i, CTL_UA_RES_RELEASE); } } lun->pr_res_type = msg->pr.pr_info.res_type; if (lun->pr_res_type != SPR_TYPE_WR_EX_AR && lun->pr_res_type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = msg->pr.pr_info.residx; else lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS; } lun->pr_generation++; } int ctl_persistent_reserve_out(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); int retval; u_int32_t param_len; struct scsi_per_res_out *cdb; struct scsi_per_res_out_parms* param; uint32_t residx; uint64_t res_key, sa_res_key, key; uint8_t type; union ctl_ha_msg persis_io; int i; CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n")); cdb = (struct scsi_per_res_out *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; /* * We only support whole-LUN scope. The scope & type are ignored for * register, register and ignore existing key and clear. * We sometimes ignore scope and type on preempts too!! * Verify reservation type here as well. */ type = cdb->scope_type & SPR_TYPE_MASK; if ((cdb->action == SPRO_RESERVE) || (cdb->action == SPRO_RELEASE)) { if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) { ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 4); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } if (type>8 || type==2 || type==4 || type==0) { ctl_set_invalid_field(/*ctsio*/ ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 1, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } } param_len = scsi_4btoul(cdb->length); if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK); ctsio->kern_data_len = param_len; ctsio->kern_total_len = param_len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr; residx = ctl_get_initindex(&ctsio->io_hdr.nexus); res_key = scsi_8btou64(param->res_key.key); sa_res_key = scsi_8btou64(param->serv_act_res_key); /* * Validate the reservation key here except for SPRO_REG_IGNO * This must be done for all other service actions */ if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) { mtx_lock(&lun->lun_lock); if ((key = ctl_get_prkey(lun, residx)) != 0) { if (res_key != key) { /* * The current key passed in doesn't match * the one the initiator previously * registered. */ mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) { /* * We are not registered */ mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } else if (res_key != 0) { /* * We are not registered and trying to register but * the register key isn't zero. */ mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } mtx_unlock(&lun->lun_lock); } switch (cdb->action & SPRO_ACTION_MASK) { case SPRO_REGISTER: case SPRO_REG_IGNO: { #if 0 printf("Registration received\n"); #endif /* * We don't support any of these options, as we report in * the read capabilities request (see * ctl_persistent_reserve_in(), above). */ if ((param->flags & SPR_SPEC_I_PT) || (param->flags & SPR_ALL_TG_PT) || (param->flags & SPR_APTPL)) { int bit_ptr; if (param->flags & SPR_APTPL) bit_ptr = 0; else if (param->flags & SPR_ALL_TG_PT) bit_ptr = 2; else /* SPR_SPEC_I_PT */ bit_ptr = 3; free(ctsio->kern_data_ptr, M_CTL); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 20, /*bit_valid*/ 1, /*bit*/ bit_ptr); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } mtx_lock(&lun->lun_lock); /* * The initiator wants to clear the * key/unregister. */ if (sa_res_key == 0) { if ((res_key == 0 && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER) || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO && ctl_get_prkey(lun, residx) == 0)) { mtx_unlock(&lun->lun_lock); goto done; } ctl_clr_prkey(lun, residx); lun->pr_key_count--; if (residx == lun->pr_res_idx) { lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_idx = CTL_PR_NO_RESERVATION; if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO || lun->pr_res_type == SPR_TYPE_EX_AC_RO) && lun->pr_key_count) { /* * If the reservation is a registrants * only type we need to generate a UA * for other registered inits. The * sense code should be RESERVATIONS * RELEASED */ for (i = softc->init_min; i < softc->init_max; i++){ if (ctl_get_prkey(lun, i) == 0) continue; ctl_est_ua(lun, i, CTL_UA_RES_RELEASE); } } lun->pr_res_type = 0; } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) { if (lun->pr_key_count==0) { lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_type = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; } } lun->pr_generation++; mtx_unlock(&lun->lun_lock); persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY; persis_io.pr.pr_info.residx = residx; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); } else /* sa_res_key != 0 */ { /* * If we aren't registered currently then increment * the key count and set the registered flag. */ ctl_alloc_prkey(lun, residx); if (ctl_get_prkey(lun, residx) == 0) lun->pr_key_count++; ctl_set_prkey(lun, residx, sa_res_key); lun->pr_generation++; mtx_unlock(&lun->lun_lock); persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_REG_KEY; persis_io.pr.pr_info.residx = residx; memcpy(persis_io.pr.pr_info.sa_res_key, param->serv_act_res_key, sizeof(param->serv_act_res_key)); ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); } break; } case SPRO_RESERVE: #if 0 printf("Reserve executed type %d\n", type); #endif mtx_lock(&lun->lun_lock); if (lun->flags & CTL_LUN_PR_RESERVED) { /* * if this isn't the reservation holder and it's * not a "all registrants" type or if the type is * different then we have a conflict */ if ((lun->pr_res_idx != residx && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) || lun->pr_res_type != type) { mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } mtx_unlock(&lun->lun_lock); } else /* create a reservation */ { /* * If it's not an "all registrants" type record * reservation holder */ if (type != SPR_TYPE_WR_EX_AR && type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = residx; /* Res holder */ else lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS; lun->flags |= CTL_LUN_PR_RESERVED; lun->pr_res_type = type; mtx_unlock(&lun->lun_lock); /* send msg to other side */ persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_RESERVE; persis_io.pr.pr_info.residx = lun->pr_res_idx; persis_io.pr.pr_info.res_type = type; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); } break; case SPRO_RELEASE: mtx_lock(&lun->lun_lock); if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) { /* No reservation exists return good status */ mtx_unlock(&lun->lun_lock); goto done; } /* * Is this nexus a reservation holder? */ if (lun->pr_res_idx != residx && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) { /* * not a res holder return good status but * do nothing */ mtx_unlock(&lun->lun_lock); goto done; } if (lun->pr_res_type != type) { mtx_unlock(&lun->lun_lock); free(ctsio->kern_data_ptr, M_CTL); ctl_set_illegal_pr_release(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* okay to release */ lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_idx = CTL_PR_NO_RESERVATION; lun->pr_res_type = 0; /* * If this isn't an exclusive access reservation and NUAR * is not set, generate UA for all other registrants. */ if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX && (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) { for (i = softc->init_min; i < softc->init_max; i++) { if (i == residx || ctl_get_prkey(lun, i) == 0) continue; ctl_est_ua(lun, i, CTL_UA_RES_RELEASE); } } mtx_unlock(&lun->lun_lock); /* Send msg to other side */ persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_RELEASE; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); break; case SPRO_CLEAR: /* send msg to other side */ mtx_lock(&lun->lun_lock); lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_type = 0; lun->pr_key_count = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; ctl_clr_prkey(lun, residx); for (i = 0; i < CTL_MAX_INITIATORS; i++) if (ctl_get_prkey(lun, i) != 0) { ctl_clr_prkey(lun, i); ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } lun->pr_generation++; mtx_unlock(&lun->lun_lock); persis_io.hdr.nexus = ctsio->io_hdr.nexus; persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION; persis_io.pr.pr_info.action = CTL_PR_CLEAR; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io.pr), M_WAITOK); break; case SPRO_PREEMPT: case SPRO_PRE_ABO: { int nretval; nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type, residx, ctsio, cdb, param); if (nretval != 0) return (CTL_RETVAL_COMPLETE); break; } default: panic("%s: Invalid PR type %#x", __func__, cdb->action); } done: free(ctsio->kern_data_ptr, M_CTL); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (retval); } /* * This routine is for handling a message from the other SC pertaining to * persistent reserve out. All the error checking will have been done * so only perorming the action need be done here to keep the two * in sync. */ static void ctl_hndl_per_res_out_on_other_sc(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); union ctl_ha_msg *msg = (union ctl_ha_msg *)&io->presio.pr_msg; struct ctl_lun *lun; int i; uint32_t residx, targ_lun; targ_lun = msg->hdr.nexus.targ_mapped_lun; mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); return; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); return; } residx = ctl_get_initindex(&msg->hdr.nexus); switch(msg->pr.pr_info.action) { case CTL_PR_REG_KEY: ctl_alloc_prkey(lun, msg->pr.pr_info.residx); if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0) lun->pr_key_count++; ctl_set_prkey(lun, msg->pr.pr_info.residx, scsi_8btou64(msg->pr.pr_info.sa_res_key)); lun->pr_generation++; break; case CTL_PR_UNREG_KEY: ctl_clr_prkey(lun, msg->pr.pr_info.residx); lun->pr_key_count--; /* XXX Need to see if the reservation has been released */ /* if so do we need to generate UA? */ if (msg->pr.pr_info.residx == lun->pr_res_idx) { lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_idx = CTL_PR_NO_RESERVATION; if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO || lun->pr_res_type == SPR_TYPE_EX_AC_RO) && lun->pr_key_count) { /* * If the reservation is a registrants * only type we need to generate a UA * for other registered inits. The * sense code should be RESERVATIONS * RELEASED */ for (i = softc->init_min; i < softc->init_max; i++) { if (ctl_get_prkey(lun, i) == 0) continue; ctl_est_ua(lun, i, CTL_UA_RES_RELEASE); } } lun->pr_res_type = 0; } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) { if (lun->pr_key_count==0) { lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_type = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; } } lun->pr_generation++; break; case CTL_PR_RESERVE: lun->flags |= CTL_LUN_PR_RESERVED; lun->pr_res_type = msg->pr.pr_info.res_type; lun->pr_res_idx = msg->pr.pr_info.residx; break; case CTL_PR_RELEASE: /* * If this isn't an exclusive access reservation and NUAR * is not set, generate UA for all other registrants. */ if (lun->pr_res_type != SPR_TYPE_EX_AC && lun->pr_res_type != SPR_TYPE_WR_EX && (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) { for (i = softc->init_min; i < softc->init_max; i++) if (i == residx || ctl_get_prkey(lun, i) == 0) continue; ctl_est_ua(lun, i, CTL_UA_RES_RELEASE); } lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_idx = CTL_PR_NO_RESERVATION; lun->pr_res_type = 0; break; case CTL_PR_PREEMPT: ctl_pro_preempt_other(lun, msg); break; case CTL_PR_CLEAR: lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_type = 0; lun->pr_key_count = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; for (i=0; i < CTL_MAX_INITIATORS; i++) { if (ctl_get_prkey(lun, i) == 0) continue; ctl_clr_prkey(lun, i); ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT); } lun->pr_generation++; break; } mtx_unlock(&lun->lun_lock); } int ctl_read_write(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int flags, retval; int isread; CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0])); flags = 0; isread = ctsio->cdb[0] == READ_6 || ctsio->cdb[0] == READ_10 || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16; switch (ctsio->cdb[0]) { case READ_6: case WRITE_6: { struct scsi_rw_6 *cdb; cdb = (struct scsi_rw_6 *)ctsio->cdb; lba = scsi_3btoul(cdb->addr); /* only 5 bits are valid in the most significant address byte */ lba &= 0x1fffff; num_blocks = cdb->length; /* * This is correct according to SBC-2. */ if (num_blocks == 0) num_blocks = 256; break; } case READ_10: case WRITE_10: { struct scsi_rw_10 *cdb; cdb = (struct scsi_rw_10 *)ctsio->cdb; if (cdb->byte2 & SRW10_FUA) flags |= CTL_LLF_FUA; if (cdb->byte2 & SRW10_DPO) flags |= CTL_LLF_DPO; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_2btoul(cdb->length); break; } case WRITE_VERIFY_10: { struct scsi_write_verify_10 *cdb; cdb = (struct scsi_write_verify_10 *)ctsio->cdb; flags |= CTL_LLF_FUA; if (cdb->byte2 & SWV_DPO) flags |= CTL_LLF_DPO; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_2btoul(cdb->length); break; } case READ_12: case WRITE_12: { struct scsi_rw_12 *cdb; cdb = (struct scsi_rw_12 *)ctsio->cdb; if (cdb->byte2 & SRW12_FUA) flags |= CTL_LLF_FUA; if (cdb->byte2 & SRW12_DPO) flags |= CTL_LLF_DPO; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_4btoul(cdb->length); break; } case WRITE_VERIFY_12: { struct scsi_write_verify_12 *cdb; cdb = (struct scsi_write_verify_12 *)ctsio->cdb; flags |= CTL_LLF_FUA; if (cdb->byte2 & SWV_DPO) flags |= CTL_LLF_DPO; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_4btoul(cdb->length); break; } case READ_16: case WRITE_16: { struct scsi_rw_16 *cdb; cdb = (struct scsi_rw_16 *)ctsio->cdb; if (cdb->byte2 & SRW12_FUA) flags |= CTL_LLF_FUA; if (cdb->byte2 & SRW12_DPO) flags |= CTL_LLF_DPO; lba = scsi_8btou64(cdb->addr); num_blocks = scsi_4btoul(cdb->length); break; } case WRITE_ATOMIC_16: { struct scsi_write_atomic_16 *cdb; if (lun->be_lun->atomicblock == 0) { ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } cdb = (struct scsi_write_atomic_16 *)ctsio->cdb; if (cdb->byte2 & SRW12_FUA) flags |= CTL_LLF_FUA; if (cdb->byte2 & SRW12_DPO) flags |= CTL_LLF_DPO; lba = scsi_8btou64(cdb->addr); num_blocks = scsi_2btoul(cdb->length); if (num_blocks > lun->be_lun->atomicblock) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } break; } case WRITE_VERIFY_16: { struct scsi_write_verify_16 *cdb; cdb = (struct scsi_write_verify_16 *)ctsio->cdb; flags |= CTL_LLF_FUA; if (cdb->byte2 & SWV_DPO) flags |= CTL_LLF_DPO; lba = scsi_8btou64(cdb->addr); num_blocks = scsi_4btoul(cdb->length); break; } default: /* * We got a command we don't support. This shouldn't * happen, commands should be filtered out above us. */ ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); break; /* NOTREACHED */ } /* * The first check is to make sure we're in bounds, the second * check is to catch wrap-around problems. If the lba + num blocks * is less than the lba, then we've wrapped around and the block * range is invalid anyway. */ if (((lba + num_blocks) > (lun->be_lun->maxlba + 1)) || ((lba + num_blocks) < lba)) { ctl_set_lba_out_of_range(ctsio, MAX(lba, lun->be_lun->maxlba + 1)); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * According to SBC-3, a transfer length of 0 is not an error. * Note that this cannot happen with WRITE(6) or READ(6), since 0 * translates to 256 blocks for those commands. */ if (num_blocks == 0) { ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* Set FUA and/or DPO if caches are disabled. */ if (isread) { if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0) flags |= CTL_LLF_FUA | CTL_LLF_DPO; } else { if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0) flags |= CTL_LLF_FUA; } lbalen = (struct ctl_lba_len_flags *) &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; lbalen->lba = lba; lbalen->len = num_blocks; lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags; ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize; ctsio->kern_rel_offset = 0; CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n")); retval = lun->backend->data_submit((union ctl_io *)ctsio); return (retval); } static int ctl_cnw_cont(union ctl_io *io) { struct ctl_lun *lun = CTL_LUN(io); struct ctl_scsiio *ctsio; struct ctl_lba_len_flags *lbalen; int retval; ctsio = &io->scsiio; ctsio->io_hdr.status = CTL_STATUS_NONE; ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT; lbalen = (struct ctl_lba_len_flags *) &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; lbalen->flags &= ~CTL_LLF_COMPARE; lbalen->flags |= CTL_LLF_WRITE; CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n")); retval = lun->backend->data_submit((union ctl_io *)ctsio); return (retval); } int ctl_cnw(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int flags, retval; CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0])); flags = 0; switch (ctsio->cdb[0]) { case COMPARE_AND_WRITE: { struct scsi_compare_and_write *cdb; cdb = (struct scsi_compare_and_write *)ctsio->cdb; if (cdb->byte2 & SRW10_FUA) flags |= CTL_LLF_FUA; if (cdb->byte2 & SRW10_DPO) flags |= CTL_LLF_DPO; lba = scsi_8btou64(cdb->addr); num_blocks = cdb->length; break; } default: /* * We got a command we don't support. This shouldn't * happen, commands should be filtered out above us. */ ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); break; /* NOTREACHED */ } /* * The first check is to make sure we're in bounds, the second * check is to catch wrap-around problems. If the lba + num blocks * is less than the lba, then we've wrapped around and the block * range is invalid anyway. */ if (((lba + num_blocks) > (lun->be_lun->maxlba + 1)) || ((lba + num_blocks) < lba)) { ctl_set_lba_out_of_range(ctsio, MAX(lba, lun->be_lun->maxlba + 1)); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * According to SBC-3, a transfer length of 0 is not an error. */ if (num_blocks == 0) { ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* Set FUA if write cache is disabled. */ if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0) flags |= CTL_LLF_FUA; ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize; ctsio->kern_rel_offset = 0; /* * Set the IO_CONT flag, so that if this I/O gets passed to * ctl_data_submit_done(), it'll get passed back to * ctl_ctl_cnw_cont() for further processing. */ ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT; ctsio->io_cont = ctl_cnw_cont; lbalen = (struct ctl_lba_len_flags *) &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; lbalen->lba = lba; lbalen->len = num_blocks; lbalen->flags = CTL_LLF_COMPARE | flags; CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n")); retval = lun->backend->data_submit((union ctl_io *)ctsio); return (retval); } int ctl_verify(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int bytchk, flags; int retval; CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0])); bytchk = 0; flags = CTL_LLF_FUA; switch (ctsio->cdb[0]) { case VERIFY_10: { struct scsi_verify_10 *cdb; cdb = (struct scsi_verify_10 *)ctsio->cdb; if (cdb->byte2 & SVFY_BYTCHK) bytchk = 1; if (cdb->byte2 & SVFY_DPO) flags |= CTL_LLF_DPO; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_2btoul(cdb->length); break; } case VERIFY_12: { struct scsi_verify_12 *cdb; cdb = (struct scsi_verify_12 *)ctsio->cdb; if (cdb->byte2 & SVFY_BYTCHK) bytchk = 1; if (cdb->byte2 & SVFY_DPO) flags |= CTL_LLF_DPO; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_4btoul(cdb->length); break; } case VERIFY_16: { struct scsi_rw_16 *cdb; cdb = (struct scsi_rw_16 *)ctsio->cdb; if (cdb->byte2 & SVFY_BYTCHK) bytchk = 1; if (cdb->byte2 & SVFY_DPO) flags |= CTL_LLF_DPO; lba = scsi_8btou64(cdb->addr); num_blocks = scsi_4btoul(cdb->length); break; } default: /* * We got a command we don't support. This shouldn't * happen, commands should be filtered out above us. */ ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * The first check is to make sure we're in bounds, the second * check is to catch wrap-around problems. If the lba + num blocks * is less than the lba, then we've wrapped around and the block * range is invalid anyway. */ if (((lba + num_blocks) > (lun->be_lun->maxlba + 1)) || ((lba + num_blocks) < lba)) { ctl_set_lba_out_of_range(ctsio, MAX(lba, lun->be_lun->maxlba + 1)); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * According to SBC-3, a transfer length of 0 is not an error. */ if (num_blocks == 0) { ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } lbalen = (struct ctl_lba_len_flags *) &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; lbalen->lba = lba; lbalen->len = num_blocks; if (bytchk) { lbalen->flags = CTL_LLF_COMPARE | flags; ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize; } else { lbalen->flags = CTL_LLF_VERIFY | flags; ctsio->kern_total_len = 0; } ctsio->kern_rel_offset = 0; CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n")); retval = lun->backend->data_submit((union ctl_io *)ctsio); return (retval); } int ctl_report_luns(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_port *port = CTL_PORT(ctsio); struct ctl_lun *lun, *request_lun = CTL_LUN(ctsio); struct scsi_report_luns *cdb; struct scsi_report_luns_data *lun_data; int num_filled, num_luns, num_port_luns, retval; uint32_t alloc_len, lun_datalen; uint32_t initidx, targ_lun_id, lun_id; retval = CTL_RETVAL_COMPLETE; cdb = (struct scsi_report_luns *)ctsio->cdb; CTL_DEBUG_PRINT(("ctl_report_luns\n")); num_luns = 0; num_port_luns = port->lun_map ? port->lun_map_size : ctl_max_luns; mtx_lock(&softc->ctl_lock); for (targ_lun_id = 0; targ_lun_id < num_port_luns; targ_lun_id++) { if (ctl_lun_map_from_port(port, targ_lun_id) != UINT32_MAX) num_luns++; } mtx_unlock(&softc->ctl_lock); switch (cdb->select_report) { case RPL_REPORT_DEFAULT: case RPL_REPORT_ALL: case RPL_REPORT_NONSUBSID: break; case RPL_REPORT_WELLKNOWN: case RPL_REPORT_ADMIN: case RPL_REPORT_CONGLOM: num_luns = 0; break; default: ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (retval); break; /* NOTREACHED */ } alloc_len = scsi_4btoul(cdb->length); /* * The initiator has to allocate at least 16 bytes for this request, * so he can at least get the header and the first LUN. Otherwise * we reject the request (per SPC-3 rev 14, section 6.21). */ if (alloc_len < (sizeof(struct scsi_report_luns_data) + sizeof(struct scsi_report_luns_lundata))) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 6, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (retval); } lun_datalen = sizeof(*lun_data) + (num_luns * sizeof(struct scsi_report_luns_lundata)); ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO); lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); mtx_lock(&softc->ctl_lock); for (targ_lun_id = 0, num_filled = 0; targ_lun_id < num_port_luns && num_filled < num_luns; targ_lun_id++) { lun_id = ctl_lun_map_from_port(port, targ_lun_id); if (lun_id == UINT32_MAX) continue; lun = softc->ctl_luns[lun_id]; if (lun == NULL) continue; be64enc(lun_data->luns[num_filled++].lundata, ctl_encode_lun(targ_lun_id)); /* * According to SPC-3, rev 14 section 6.21: * * "The execution of a REPORT LUNS command to any valid and * installed logical unit shall clear the REPORTED LUNS DATA * HAS CHANGED unit attention condition for all logical * units of that target with respect to the requesting * initiator. A valid and installed logical unit is one * having a PERIPHERAL QUALIFIER of 000b in the standard * INQUIRY data (see 6.4.2)." * * If request_lun is NULL, the LUN this report luns command * was issued to is either disabled or doesn't exist. In that * case, we shouldn't clear any pending lun change unit * attention. */ if (request_lun != NULL) { mtx_lock(&lun->lun_lock); ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE); mtx_unlock(&lun->lun_lock); } } mtx_unlock(&softc->ctl_lock); /* * It's quite possible that we've returned fewer LUNs than we allocated * space for. Trim it. */ lun_datalen = sizeof(*lun_data) + (num_filled * sizeof(struct scsi_report_luns_lundata)); ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(lun_datalen, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * We set this to the actual data length, regardless of how much * space we actually have to return results. If the user looks at * this value, he'll know whether or not he allocated enough space * and reissue the command if necessary. We don't support well * known logical units, so if the user asks for that, return none. */ scsi_ulto4b(lun_datalen - 8, lun_data->length); /* * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy * this request. */ ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } int ctl_request_sense(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_request_sense *cdb; struct scsi_sense_data *sense_ptr, *ps; uint32_t initidx; int have_error; u_int sense_len = SSD_FULL_SIZE; scsi_sense_data_type sense_format; ctl_ua_type ua_type; uint8_t asc = 0, ascq = 0; cdb = (struct scsi_request_sense *)ctsio->cdb; CTL_DEBUG_PRINT(("ctl_request_sense\n")); /* * Determine which sense format the user wants. */ if (cdb->byte2 & SRS_DESC) sense_format = SSD_TYPE_DESC; else sense_format = SSD_TYPE_FIXED; ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK); sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; /* * struct scsi_sense_data, which is currently set to 256 bytes, is * larger than the largest allowed value for the length field in the * REQUEST SENSE CDB, which is 252 bytes as of SPC-4. */ ctsio->kern_data_len = cdb->length; ctsio->kern_total_len = cdb->length; /* * If we don't have a LUN, we don't have any pending sense. */ if (lun == NULL || ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 && softc->ha_link < CTL_HA_LINK_UNKNOWN)) { /* "Logical unit not supported" */ ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x25, /*ascq*/ 0x00, SSD_ELEM_NONE); goto send; } have_error = 0; initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); /* * Check for pending sense, and then for pending unit attentions. * Pending sense gets returned first, then pending unit attentions. */ mtx_lock(&lun->lun_lock); ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT]; if (ps != NULL) ps += initidx % CTL_MAX_INIT_PER_PORT; if (ps != NULL && ps->error_code != 0) { scsi_sense_data_type stored_format; /* * Check to see which sense format was used for the stored * sense data. */ stored_format = scsi_sense_type(ps); /* * If the user requested a different sense format than the * one we stored, then we need to convert it to the other * format. If we're going from descriptor to fixed format * sense data, we may lose things in translation, depending * on what options were used. * * If the stored format is SSD_TYPE_NONE (i.e. invalid), * for some reason we'll just copy it out as-is. */ if ((stored_format == SSD_TYPE_FIXED) && (sense_format == SSD_TYPE_DESC)) ctl_sense_to_desc((struct scsi_sense_data_fixed *) ps, (struct scsi_sense_data_desc *)sense_ptr); else if ((stored_format == SSD_TYPE_DESC) && (sense_format == SSD_TYPE_FIXED)) ctl_sense_to_fixed((struct scsi_sense_data_desc *) ps, (struct scsi_sense_data_fixed *)sense_ptr); else memcpy(sense_ptr, ps, sizeof(*sense_ptr)); ps->error_code = 0; have_error = 1; } else { ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len, sense_format); if (ua_type != CTL_UA_NONE) have_error = 1; } if (have_error == 0) { /* * Report informational exception if have one and allowed. */ if (lun->MODE_IE.mrie != SIEP_MRIE_NO) { asc = lun->ie_asc; ascq = lun->ie_ascq; } ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format, /*current_error*/ 1, /*sense_key*/ SSD_KEY_NO_SENSE, /*asc*/ asc, /*ascq*/ ascq, SSD_ELEM_NONE); } mtx_unlock(&lun->lun_lock); send: /* * We report the SCSI status as OK, since the status of the command * itself is OK. We're reporting sense as parameter data. */ ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_tur(struct ctl_scsiio *ctsio) { CTL_DEBUG_PRINT(("ctl_tur\n")); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * SCSI VPD page 0x00, the Supported VPD Pages page. */ static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_supported_pages *pages; int sup_page_size; int p; sup_page_size = sizeof(struct scsi_vpd_supported_pages) * SCSI_EVPD_NUM_SUPPORTED_PAGES; ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO); pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(sup_page_size, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. Need to change this * to figure out whether the disk device is actually online or not. */ if (lun != NULL) pages->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; p = 0; /* Supported VPD pages */ pages->page_list[p++] = SVPD_SUPPORTED_PAGES; /* Serial Number */ pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER; /* Device Identification */ pages->page_list[p++] = SVPD_DEVICE_ID; /* Extended INQUIRY Data */ pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA; /* Mode Page Policy */ pages->page_list[p++] = SVPD_MODE_PAGE_POLICY; /* SCSI Ports */ pages->page_list[p++] = SVPD_SCSI_PORTS; /* Third-party Copy */ pages->page_list[p++] = SVPD_SCSI_TPC; if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) { /* Block limits */ pages->page_list[p++] = SVPD_BLOCK_LIMITS; /* Block Device Characteristics */ pages->page_list[p++] = SVPD_BDC; /* Logical Block Provisioning */ pages->page_list[p++] = SVPD_LBP; } pages->length = p; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * SCSI VPD page 0x80, the Unit Serial Number page. */ static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_unit_serial_number *sn_ptr; int data_len; data_len = 4 + CTL_SN_LEN; ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. Need to change this * to figure out whether the disk device is actually online or not. */ if (lun != NULL) sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER; sn_ptr->length = CTL_SN_LEN; /* * If we don't have a LUN, we just leave the serial number as * all spaces. */ if (lun != NULL) { strncpy((char *)sn_ptr->serial_num, (char *)lun->be_lun->serial_num, CTL_SN_LEN); } else memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * SCSI VPD page 0x86, the Extended INQUIRY Data page. */ static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_extended_inquiry_data *eid_ptr; int data_len; data_len = sizeof(struct scsi_vpd_extended_inquiry_data); ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. */ if (lun != NULL) eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA; scsi_ulto2b(data_len - 4, eid_ptr->page_length); /* * We support head of queue, ordered and simple tags. */ eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP; /* * Volatile cache supported. */ eid_ptr->flags3 = SVPD_EID_V_SUP; /* * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit * attention for a particular IT nexus on all LUNs once we report * it to that nexus once. This bit is required as of SPC-4. */ eid_ptr->flags4 = SVPD_EID_LUICLR; /* * We support revert to defaults (RTD) bit in MODE SELECT. */ eid_ptr->flags5 = SVPD_EID_RTD_SUP; /* * XXX KDM in order to correctly answer this, we would need * information from the SIM to determine how much sense data it * can send. So this would really be a path inquiry field, most * likely. This can be set to a maximum of 252 according to SPC-4, * but the hardware may or may not be able to support that much. * 0 just means that the maximum sense data length is not reported. */ eid_ptr->max_sense_length = 0; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_mode_page_policy *mpp_ptr; int data_len; data_len = sizeof(struct scsi_vpd_mode_page_policy) + sizeof(struct scsi_vpd_mode_page_policy_descr); ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. */ if (lun != NULL) mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY; scsi_ulto2b(data_len - 4, mpp_ptr->page_length); mpp_ptr->descr[0].page_code = 0x3f; mpp_ptr->descr[0].subpage_code = 0xff; mpp_ptr->descr[0].policy = SVPD_MPP_SHARED; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * SCSI VPD page 0x83, the Device Identification page. */ static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_port *port = CTL_PORT(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_device_id *devid_ptr; struct scsi_vpd_id_descriptor *desc; int data_len, g; uint8_t proto; data_len = sizeof(struct scsi_vpd_device_id) + sizeof(struct scsi_vpd_id_descriptor) + sizeof(struct scsi_vpd_id_rel_trgt_port_id) + sizeof(struct scsi_vpd_id_descriptor) + sizeof(struct scsi_vpd_id_trgt_port_grp_id); if (lun && lun->lun_devid) data_len += lun->lun_devid->len; if (port && port->port_devid) data_len += port->port_devid->len; if (port && port->target_devid) data_len += port->target_devid->len; ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. */ if (lun != NULL) devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; devid_ptr->page_code = SVPD_DEVICE_ID; scsi_ulto2b(data_len - 4, devid_ptr->length); if (port && port->port_type == CTL_PORT_FC) proto = SCSI_PROTO_FC << 4; else if (port && port->port_type == CTL_PORT_SAS) proto = SCSI_PROTO_SAS << 4; else if (port && port->port_type == CTL_PORT_ISCSI) proto = SCSI_PROTO_ISCSI << 4; else proto = SCSI_PROTO_SPI << 4; desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list; /* * We're using a LUN association here. i.e., this device ID is a * per-LUN identifier. */ if (lun && lun->lun_devid) { memcpy(desc, lun->lun_devid->data, lun->lun_devid->len); desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc + lun->lun_devid->len); } /* * This is for the WWPN which is a port association. */ if (port && port->port_devid) { memcpy(desc, port->port_devid->data, port->port_devid->len); desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc + port->port_devid->len); } /* * This is for the Relative Target Port(type 4h) identifier */ desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_RELTARG; desc->length = 4; scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]); desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] + sizeof(struct scsi_vpd_id_rel_trgt_port_id)); /* * This is for the Target Port Group(type 5h) identifier */ desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_TPORTGRP; desc->length = 4; if (softc->is_single || (port && port->status & CTL_PORT_STATUS_HA_SHARED)) g = 1; else g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt; scsi_ulto2b(g, &desc->identifier[2]); desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] + sizeof(struct scsi_vpd_id_trgt_port_grp_id)); /* * This is for the Target identifier */ if (port && port->target_devid) { memcpy(desc, port->target_devid->data, port->target_devid->len); } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_scsi_ports *sp; struct scsi_vpd_port_designation *pd; struct scsi_vpd_port_designation_cont *pdc; struct ctl_port *port; int data_len, num_target_ports, iid_len, id_len; num_target_ports = 0; iid_len = 0; id_len = 0; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0) continue; if (lun != NULL && ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; num_target_ports++; if (port->init_devid) iid_len += port->init_devid->len; if (port->port_devid) id_len += port->port_devid->len; } mtx_unlock(&softc->ctl_lock); data_len = sizeof(struct scsi_vpd_scsi_ports) + num_target_ports * (sizeof(struct scsi_vpd_port_designation) + sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len; ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. Need to change this * to figure out whether the disk device is actually online or not. */ if (lun != NULL) sp->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; sp->page_code = SVPD_SCSI_PORTS; scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports), sp->page_length); pd = &sp->design[0]; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0) continue; if (lun != NULL && ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; scsi_ulto2b(port->targ_port, pd->relative_port_id); if (port->init_devid) { iid_len = port->init_devid->len; memcpy(pd->initiator_transportid, port->init_devid->data, port->init_devid->len); } else iid_len = 0; scsi_ulto2b(iid_len, pd->initiator_transportid_length); pdc = (struct scsi_vpd_port_designation_cont *) (&pd->initiator_transportid[iid_len]); if (port->port_devid) { id_len = port->port_devid->len; memcpy(pdc->target_port_descriptors, port->port_devid->data, port->port_devid->len); } else id_len = 0; scsi_ulto2b(id_len, pdc->target_port_descriptors_length); pd = (struct scsi_vpd_port_designation *) ((uint8_t *)pdc->target_port_descriptors + id_len); } mtx_unlock(&softc->ctl_lock); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_block_limits *bl_ptr; + const char *val; uint64_t ival; ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO); bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(sizeof(*bl_ptr), alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. Need to change this * to figure out whether the disk device is actually online or not. */ if (lun != NULL) bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; bl_ptr->page_code = SVPD_BLOCK_LIMITS; scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length); bl_ptr->max_cmp_write_len = 0xff; scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len); if (lun != NULL) { scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len); if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) { ival = 0xffffffff; - ctl_get_opt_number(&lun->be_lun->options, - "unmap_max_lba", &ival); + val = dnvlist_get_string(lun->be_lun->options, + "unmap_max_lba", NULL); + if (val != NULL) + ctl_expand_number(val, &ival); scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt); ival = 0xffffffff; - ctl_get_opt_number(&lun->be_lun->options, - "unmap_max_descr", &ival); + val = dnvlist_get_string(lun->be_lun->options, + "unmap_max_descr", NULL); + if (val != NULL) + ctl_expand_number(val, &ival); scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt); if (lun->be_lun->ublockexp != 0) { scsi_ulto4b((1 << lun->be_lun->ublockexp), bl_ptr->opt_unmap_grain); scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff, bl_ptr->unmap_grain_align); } } scsi_ulto4b(lun->be_lun->atomicblock, bl_ptr->max_atomic_transfer_length); scsi_ulto4b(0, bl_ptr->atomic_alignment); scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity); scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary); scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size); ival = UINT64_MAX; - ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival); + val = dnvlist_get_string(lun->be_lun->options, + "write_same_max_lba", NULL); + if (val != NULL) + ctl_expand_number(val, &ival); scsi_u64to8b(ival, bl_ptr->max_write_same_length); } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_block_device_characteristics *bdc_ptr; const char *value; u_int i; ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO); bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(sizeof(*bdc_ptr), alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. Need to change this * to figure out whether the disk device is actually online or not. */ if (lun != NULL) bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; bdc_ptr->page_code = SVPD_BDC; scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length); if (lun != NULL && - (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL) + (value = dnvlist_get_string(lun->be_lun->options, "rpm", NULL)) != NULL) i = strtol(value, NULL, 0); else i = CTL_DEFAULT_ROTATION_RATE; scsi_ulto2b(i, bdc_ptr->medium_rotation_rate); if (lun != NULL && - (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL) + (value = dnvlist_get_string(lun->be_lun->options, "formfactor", NULL)) != NULL) i = strtol(value, NULL, 0); else i = 0; bdc_ptr->wab_wac_ff = (i & 0x0f); bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_logical_block_prov *lbp_ptr; const char *value; ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO); lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(sizeof(*lbp_ptr), alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. Need to change this * to figure out whether the disk device is actually online or not. */ if (lun != NULL) lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; lbp_ptr->page_code = SVPD_LBP; scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length); lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT; if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) { lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP; - value = ctl_get_opt(&lun->be_lun->options, "provisioning_type"); + value = dnvlist_get_string(lun->be_lun->options, + "provisioning_type", NULL); if (value != NULL) { if (strcmp(value, "resource") == 0) lbp_ptr->prov_type = SVPD_LBP_RESOURCE; else if (strcmp(value, "thin") == 0) lbp_ptr->prov_type = SVPD_LBP_THIN; } else lbp_ptr->prov_type = SVPD_LBP_THIN; } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * INQUIRY with the EVPD bit set. */ static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_inquiry *cdb; int alloc_len, retval; cdb = (struct scsi_inquiry *)ctsio->cdb; alloc_len = scsi_2btoul(cdb->length); switch (cdb->page_code) { case SVPD_SUPPORTED_PAGES: retval = ctl_inquiry_evpd_supported(ctsio, alloc_len); break; case SVPD_UNIT_SERIAL_NUMBER: retval = ctl_inquiry_evpd_serial(ctsio, alloc_len); break; case SVPD_DEVICE_ID: retval = ctl_inquiry_evpd_devid(ctsio, alloc_len); break; case SVPD_EXTENDED_INQUIRY_DATA: retval = ctl_inquiry_evpd_eid(ctsio, alloc_len); break; case SVPD_MODE_PAGE_POLICY: retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len); break; case SVPD_SCSI_PORTS: retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len); break; case SVPD_SCSI_TPC: retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len); break; case SVPD_BLOCK_LIMITS: if (lun == NULL || lun->be_lun->lun_type != T_DIRECT) goto err; retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len); break; case SVPD_BDC: if (lun == NULL || lun->be_lun->lun_type != T_DIRECT) goto err; retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len); break; case SVPD_LBP: if (lun == NULL || lun->be_lun->lun_type != T_DIRECT) goto err; retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len); break; default: err: ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); retval = CTL_RETVAL_COMPLETE; break; } return (retval); } /* * Standard INQUIRY data. */ static int ctl_inquiry_std(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_port *port = CTL_PORT(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_inquiry_data *inq_ptr; struct scsi_inquiry *cdb; - char *val; + const char *val; uint32_t alloc_len, data_len; ctl_port_type port_type; port_type = port->port_type; if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL) port_type = CTL_PORT_SCSI; cdb = (struct scsi_inquiry *)ctsio->cdb; alloc_len = scsi_2btoul(cdb->length); /* * We malloc the full inquiry data size here and fill it * in. If the user only asks for less, we'll give him * that much. */ data_len = offsetof(struct scsi_inquiry_data, vendor_specific1); ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr; ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; if (lun != NULL) { if ((lun->flags & CTL_LUN_PRIMARY_SC) || softc->ha_link >= CTL_HA_LINK_UNKNOWN) { inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; } else { inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | lun->be_lun->lun_type; } if (lun->flags & CTL_LUN_REMOVABLE) inq_ptr->dev_qual2 |= SID_RMB; } else inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE; /* RMB in byte 2 is 0 */ inq_ptr->version = SCSI_REV_SPC5; /* * According to SAM-3, even if a device only supports a single * level of LUN addressing, it should still set the HISUP bit: * * 4.9.1 Logical unit numbers overview * * All logical unit number formats described in this standard are * hierarchical in structure even when only a single level in that * hierarchy is used. The HISUP bit shall be set to one in the * standard INQUIRY data (see SPC-2) when any logical unit number * format described in this standard is used. Non-hierarchical * formats are outside the scope of this standard. * * Therefore we set the HiSup bit here. * * The response format is 2, per SPC-3. */ inq_ptr->response_format = SID_HiSup | 2; inq_ptr->additional_length = data_len - (offsetof(struct scsi_inquiry_data, additional_length) + 1); CTL_DEBUG_PRINT(("additional_length = %d\n", inq_ptr->additional_length)); inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT; if (port_type == CTL_PORT_SCSI) inq_ptr->spc2_flags = SPC2_SID_ADDR16; inq_ptr->spc2_flags |= SPC2_SID_MultiP; inq_ptr->flags = SID_CmdQue; if (port_type == CTL_PORT_SCSI) inq_ptr->flags |= SID_WBus16 | SID_Sync; /* * Per SPC-3, unused bytes in ASCII strings are filled with spaces. * We have 8 bytes for the vendor name, and 16 bytes for the device * name and 4 bytes for the revision. */ - if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options, - "vendor")) == NULL) { + if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options, + "vendor", NULL)) == NULL) { strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor)); } else { memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor)); strncpy(inq_ptr->vendor, val, min(sizeof(inq_ptr->vendor), strlen(val))); } if (lun == NULL) { strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT, sizeof(inq_ptr->product)); - } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) { + } else if ((val = dnvlist_get_string(lun->be_lun->options, "product", + NULL)) == NULL) { switch (lun->be_lun->lun_type) { case T_DIRECT: strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT, sizeof(inq_ptr->product)); break; case T_PROCESSOR: strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT, sizeof(inq_ptr->product)); break; case T_CDROM: strncpy(inq_ptr->product, CTL_CDROM_PRODUCT, sizeof(inq_ptr->product)); break; default: strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT, sizeof(inq_ptr->product)); break; } } else { memset(inq_ptr->product, ' ', sizeof(inq_ptr->product)); strncpy(inq_ptr->product, val, min(sizeof(inq_ptr->product), strlen(val))); } /* * XXX make this a macro somewhere so it automatically gets * incremented when we make changes. */ - if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options, - "revision")) == NULL) { + if (lun == NULL || (val = dnvlist_get_string(lun->be_lun->options, + "revision", NULL)) == NULL) { strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision)); } else { memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision)); strncpy(inq_ptr->revision, val, min(sizeof(inq_ptr->revision), strlen(val))); } /* * For parallel SCSI, we support double transition and single * transition clocking. We also support QAS (Quick Arbitration * and Selection) and Information Unit transfers on both the * control and array devices. */ if (port_type == CTL_PORT_SCSI) inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS | SID_SPI_IUS; /* SAM-6 (no version claimed) */ scsi_ulto2b(0x00C0, inq_ptr->version1); /* SPC-5 (no version claimed) */ scsi_ulto2b(0x05C0, inq_ptr->version2); if (port_type == CTL_PORT_FC) { /* FCP-2 ANSI INCITS.350:2003 */ scsi_ulto2b(0x0917, inq_ptr->version3); } else if (port_type == CTL_PORT_SCSI) { /* SPI-4 ANSI INCITS.362:200x */ scsi_ulto2b(0x0B56, inq_ptr->version3); } else if (port_type == CTL_PORT_ISCSI) { /* iSCSI (no version claimed) */ scsi_ulto2b(0x0960, inq_ptr->version3); } else if (port_type == CTL_PORT_SAS) { /* SAS (no version claimed) */ scsi_ulto2b(0x0BE0, inq_ptr->version3); } else if (port_type == CTL_PORT_UMASS) { /* USB Mass Storage Class Bulk-Only Transport, Revision 1.0 */ scsi_ulto2b(0x1730, inq_ptr->version3); } if (lun == NULL) { /* SBC-4 (no version claimed) */ scsi_ulto2b(0x0600, inq_ptr->version4); } else { switch (lun->be_lun->lun_type) { case T_DIRECT: /* SBC-4 (no version claimed) */ scsi_ulto2b(0x0600, inq_ptr->version4); break; case T_PROCESSOR: break; case T_CDROM: /* MMC-6 (no version claimed) */ scsi_ulto2b(0x04E0, inq_ptr->version4); break; default: break; } } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_inquiry(struct ctl_scsiio *ctsio) { struct scsi_inquiry *cdb; int retval; CTL_DEBUG_PRINT(("ctl_inquiry\n")); cdb = (struct scsi_inquiry *)ctsio->cdb; if (cdb->byte2 & SI_EVPD) retval = ctl_inquiry_evpd(ctsio); else if (cdb->page_code == 0) retval = ctl_inquiry_std(ctsio); else { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } return (retval); } int ctl_get_config(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_get_config_header *hdr; struct scsi_get_config_feature *feature; struct scsi_get_config *cdb; uint32_t alloc_len, data_len; int rt, starting; cdb = (struct scsi_get_config *)ctsio->cdb; rt = (cdb->rt & SGC_RT_MASK); starting = scsi_2btoul(cdb->starting_feature); alloc_len = scsi_2btoul(cdb->length); data_len = sizeof(struct scsi_get_config_header) + sizeof(struct scsi_get_config_feature) + 8 + sizeof(struct scsi_get_config_feature) + 8 + sizeof(struct scsi_get_config_feature) + 4 + sizeof(struct scsi_get_config_feature) + 4 + sizeof(struct scsi_get_config_feature) + 8 + sizeof(struct scsi_get_config_feature) + sizeof(struct scsi_get_config_feature) + 4 + sizeof(struct scsi_get_config_feature) + 4 + sizeof(struct scsi_get_config_feature) + 4 + sizeof(struct scsi_get_config_feature) + 4 + sizeof(struct scsi_get_config_feature) + 4 + sizeof(struct scsi_get_config_feature) + 4; ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr; if (lun->flags & CTL_LUN_NO_MEDIA) scsi_ulto2b(0x0000, hdr->current_profile); else scsi_ulto2b(0x0010, hdr->current_profile); feature = (struct scsi_get_config_feature *)(hdr + 1); if (starting > 0x003b) goto done; if (starting > 0x003a) goto f3b; if (starting > 0x002b) goto f3a; if (starting > 0x002a) goto f2b; if (starting > 0x001f) goto f2a; if (starting > 0x001e) goto f1f; if (starting > 0x001d) goto f1e; if (starting > 0x0010) goto f1d; if (starting > 0x0003) goto f10; if (starting > 0x0002) goto f3; if (starting > 0x0001) goto f2; if (starting > 0x0000) goto f1; /* Profile List */ scsi_ulto2b(0x0000, feature->feature_code); feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT; feature->add_length = 8; scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */ feature->feature_data[2] = 0x00; scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */ feature->feature_data[6] = 0x01; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f1: /* Core */ scsi_ulto2b(0x0001, feature->feature_code); feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT; feature->add_length = 8; scsi_ulto4b(0x00000000, &feature->feature_data[0]); feature->feature_data[4] = 0x03; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f2: /* Morphing */ scsi_ulto2b(0x0002, feature->feature_code); feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x02; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f3: /* Removable Medium */ scsi_ulto2b(0x0003, feature->feature_code); feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x39; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA)) goto done; f10: /* Random Read */ scsi_ulto2b(0x0010, feature->feature_code); feature->flags = 0x00; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 8; scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]); scsi_ulto2b(1, &feature->feature_data[4]); feature->feature_data[6] = 0x00; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f1d: /* Multi-Read */ scsi_ulto2b(0x001D, feature->feature_code); feature->flags = 0x00; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 0; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f1e: /* CD Read */ scsi_ulto2b(0x001E, feature->feature_code); feature->flags = 0x00; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x00; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f1f: /* DVD Read */ scsi_ulto2b(0x001F, feature->feature_code); feature->flags = 0x08; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x01; feature->feature_data[2] = 0x03; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f2a: /* DVD+RW */ scsi_ulto2b(0x002A, feature->feature_code); feature->flags = 0x04; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x00; feature->feature_data[1] = 0x00; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f2b: /* DVD+R */ scsi_ulto2b(0x002B, feature->feature_code); feature->flags = 0x00; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x00; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f3a: /* DVD+RW Dual Layer */ scsi_ulto2b(0x003A, feature->feature_code); feature->flags = 0x00; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x00; feature->feature_data[1] = 0x00; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; f3b: /* DVD+R Dual Layer */ scsi_ulto2b(0x003B, feature->feature_code); feature->flags = 0x00; if ((lun->flags & CTL_LUN_NO_MEDIA) == 0) feature->flags |= SGC_F_CURRENT; feature->add_length = 4; feature->feature_data[0] = 0x00; feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; done: data_len = (uint8_t *)feature - (uint8_t *)hdr; if (rt == SGC_RT_SPECIFIC && data_len > 4) { feature = (struct scsi_get_config_feature *)(hdr + 1); if (scsi_2btoul(feature->feature_code) == starting) feature = (struct scsi_get_config_feature *) &feature->feature_data[feature->add_length]; data_len = (uint8_t *)feature - (uint8_t *)hdr; } scsi_ulto4b(data_len - 4, hdr->data_length); ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_get_event_status(struct ctl_scsiio *ctsio) { struct scsi_get_event_status_header *hdr; struct scsi_get_event_status *cdb; uint32_t alloc_len, data_len; int notif_class; cdb = (struct scsi_get_event_status *)ctsio->cdb; if ((cdb->byte2 & SGESN_POLLED) == 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } notif_class = cdb->notif_class; alloc_len = scsi_2btoul(cdb->length); data_len = sizeof(struct scsi_get_event_status_header); ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr; scsi_ulto2b(0, hdr->descr_length); hdr->nea_class = SGESN_NEA; hdr->supported_class = 0; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_mechanism_status(struct ctl_scsiio *ctsio) { struct scsi_mechanism_status_header *hdr; struct scsi_mechanism_status *cdb; uint32_t alloc_len, data_len; cdb = (struct scsi_mechanism_status *)ctsio->cdb; alloc_len = scsi_2btoul(cdb->length); data_len = sizeof(struct scsi_mechanism_status_header); ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr; hdr->state1 = 0x00; hdr->state2 = 0xe0; scsi_ulto3b(0, hdr->lba); hdr->slots_num = 0; scsi_ulto2b(0, hdr->slots_length); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } static void ctl_ultomsf(uint32_t lba, uint8_t *buf) { lba += 150; buf[0] = 0; buf[1] = bin2bcd((lba / 75) / 60); buf[2] = bin2bcd((lba / 75) % 60); buf[3] = bin2bcd(lba % 75); } int ctl_read_toc(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_read_toc_hdr *hdr; struct scsi_read_toc_type01_descr *descr; struct scsi_read_toc *cdb; uint32_t alloc_len, data_len; int format, msf; cdb = (struct scsi_read_toc *)ctsio->cdb; msf = (cdb->byte2 & CD_MSF) != 0; format = cdb->format; alloc_len = scsi_2btoul(cdb->data_len); data_len = sizeof(struct scsi_read_toc_hdr); if (format == 0) data_len += 2 * sizeof(struct scsi_read_toc_type01_descr); else data_len += sizeof(struct scsi_read_toc_type01_descr); ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr; if (format == 0) { scsi_ulto2b(0x12, hdr->data_length); hdr->first = 1; hdr->last = 1; descr = (struct scsi_read_toc_type01_descr *)(hdr + 1); descr->addr_ctl = 0x14; descr->track_number = 1; if (msf) ctl_ultomsf(0, descr->track_start); else scsi_ulto4b(0, descr->track_start); descr++; descr->addr_ctl = 0x14; descr->track_number = 0xaa; if (msf) ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start); else scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start); } else { scsi_ulto2b(0x0a, hdr->data_length); hdr->first = 1; hdr->last = 1; descr = (struct scsi_read_toc_type01_descr *)(hdr + 1); descr->addr_ctl = 0x14; descr->track_number = 1; if (msf) ctl_ultomsf(0, descr->track_start); else scsi_ulto4b(0, descr->track_start); } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * For known CDB types, parse the LBA and length. */ static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len) { if (io->io_hdr.io_type != CTL_IO_SCSI) return (1); switch (io->scsiio.cdb[0]) { case COMPARE_AND_WRITE: { struct scsi_compare_and_write *cdb; cdb = (struct scsi_compare_and_write *)io->scsiio.cdb; *lba = scsi_8btou64(cdb->addr); *len = cdb->length; break; } case READ_6: case WRITE_6: { struct scsi_rw_6 *cdb; cdb = (struct scsi_rw_6 *)io->scsiio.cdb; *lba = scsi_3btoul(cdb->addr); /* only 5 bits are valid in the most significant address byte */ *lba &= 0x1fffff; *len = cdb->length; break; } case READ_10: case WRITE_10: { struct scsi_rw_10 *cdb; cdb = (struct scsi_rw_10 *)io->scsiio.cdb; *lba = scsi_4btoul(cdb->addr); *len = scsi_2btoul(cdb->length); break; } case WRITE_VERIFY_10: { struct scsi_write_verify_10 *cdb; cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb; *lba = scsi_4btoul(cdb->addr); *len = scsi_2btoul(cdb->length); break; } case READ_12: case WRITE_12: { struct scsi_rw_12 *cdb; cdb = (struct scsi_rw_12 *)io->scsiio.cdb; *lba = scsi_4btoul(cdb->addr); *len = scsi_4btoul(cdb->length); break; } case WRITE_VERIFY_12: { struct scsi_write_verify_12 *cdb; cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb; *lba = scsi_4btoul(cdb->addr); *len = scsi_4btoul(cdb->length); break; } case READ_16: case WRITE_16: { struct scsi_rw_16 *cdb; cdb = (struct scsi_rw_16 *)io->scsiio.cdb; *lba = scsi_8btou64(cdb->addr); *len = scsi_4btoul(cdb->length); break; } case WRITE_ATOMIC_16: { struct scsi_write_atomic_16 *cdb; cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb; *lba = scsi_8btou64(cdb->addr); *len = scsi_2btoul(cdb->length); break; } case WRITE_VERIFY_16: { struct scsi_write_verify_16 *cdb; cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb; *lba = scsi_8btou64(cdb->addr); *len = scsi_4btoul(cdb->length); break; } case WRITE_SAME_10: { struct scsi_write_same_10 *cdb; cdb = (struct scsi_write_same_10 *)io->scsiio.cdb; *lba = scsi_4btoul(cdb->addr); *len = scsi_2btoul(cdb->length); break; } case WRITE_SAME_16: { struct scsi_write_same_16 *cdb; cdb = (struct scsi_write_same_16 *)io->scsiio.cdb; *lba = scsi_8btou64(cdb->addr); *len = scsi_4btoul(cdb->length); break; } case VERIFY_10: { struct scsi_verify_10 *cdb; cdb = (struct scsi_verify_10 *)io->scsiio.cdb; *lba = scsi_4btoul(cdb->addr); *len = scsi_2btoul(cdb->length); break; } case VERIFY_12: { struct scsi_verify_12 *cdb; cdb = (struct scsi_verify_12 *)io->scsiio.cdb; *lba = scsi_4btoul(cdb->addr); *len = scsi_4btoul(cdb->length); break; } case VERIFY_16: { struct scsi_verify_16 *cdb; cdb = (struct scsi_verify_16 *)io->scsiio.cdb; *lba = scsi_8btou64(cdb->addr); *len = scsi_4btoul(cdb->length); break; } case UNMAP: { *lba = 0; *len = UINT64_MAX; break; } case SERVICE_ACTION_IN: { /* GET LBA STATUS */ struct scsi_get_lba_status *cdb; cdb = (struct scsi_get_lba_status *)io->scsiio.cdb; *lba = scsi_8btou64(cdb->addr); *len = UINT32_MAX; break; } default: return (1); break; /* NOTREACHED */ } return (0); } static ctl_action ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2, bool seq) { uint64_t endlba1, endlba2; endlba1 = lba1 + len1 - (seq ? 0 : 1); endlba2 = lba2 + len2 - 1; if ((endlba1 < lba2) || (endlba2 < lba1)) return (CTL_ACTION_PASS); else return (CTL_ACTION_BLOCK); } static int ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2) { struct ctl_ptr_len_flags *ptrlen; struct scsi_unmap_desc *buf, *end, *range; uint64_t lba; uint32_t len; /* If not UNMAP -- go other way. */ if (io->io_hdr.io_type != CTL_IO_SCSI || io->scsiio.cdb[0] != UNMAP) return (CTL_ACTION_ERROR); /* If UNMAP without data -- block and wait for data. */ ptrlen = (struct ctl_ptr_len_flags *) &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 || ptrlen->ptr == NULL) return (CTL_ACTION_BLOCK); /* UNMAP with data -- check for collision. */ buf = (struct scsi_unmap_desc *)ptrlen->ptr; end = buf + ptrlen->len / sizeof(*buf); for (range = buf; range < end; range++) { lba = scsi_8btou64(range->lba); len = scsi_4btoul(range->length); if ((lba < lba2 + len2) && (lba + len > lba2)) return (CTL_ACTION_BLOCK); } return (CTL_ACTION_PASS); } static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq) { uint64_t lba1, lba2; uint64_t len1, len2; int retval; if (ctl_get_lba_len(io2, &lba2, &len2) != 0) return (CTL_ACTION_ERROR); retval = ctl_extent_check_unmap(io1, lba2, len2); if (retval != CTL_ACTION_ERROR) return (retval); if (ctl_get_lba_len(io1, &lba1, &len1) != 0) return (CTL_ACTION_ERROR); if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE) seq = FALSE; return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq)); } static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2) { uint64_t lba1, lba2; uint64_t len1, len2; if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE) return (CTL_ACTION_PASS); if (ctl_get_lba_len(io1, &lba1, &len1) != 0) return (CTL_ACTION_ERROR); if (ctl_get_lba_len(io2, &lba2, &len2) != 0) return (CTL_ACTION_ERROR); if (lba1 + len1 == lba2) return (CTL_ACTION_BLOCK); return (CTL_ACTION_PASS); } static ctl_action ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io, union ctl_io *ooa_io) { const struct ctl_cmd_entry *pending_entry, *ooa_entry; const ctl_serialize_action *serialize_row; /* * The initiator attempted multiple untagged commands at the same * time. Can't do that. */ if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED) && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED) && ((pending_io->io_hdr.nexus.targ_port == ooa_io->io_hdr.nexus.targ_port) && (pending_io->io_hdr.nexus.initid == ooa_io->io_hdr.nexus.initid)) && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT | CTL_FLAG_STATUS_SENT)) == 0)) return (CTL_ACTION_OVERLAP); /* * The initiator attempted to send multiple tagged commands with * the same ID. (It's fine if different initiators have the same * tag ID.) * * Even if all of those conditions are true, we don't kill the I/O * if the command ahead of us has been aborted. We won't end up * sending it to the FETD, and it's perfectly legal to resend a * command with the same tag number as long as the previous * instance of this tag number has been aborted somehow. */ if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED) && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED) && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num) && ((pending_io->io_hdr.nexus.targ_port == ooa_io->io_hdr.nexus.targ_port) && (pending_io->io_hdr.nexus.initid == ooa_io->io_hdr.nexus.initid)) && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT | CTL_FLAG_STATUS_SENT)) == 0)) return (CTL_ACTION_OVERLAP_TAG); /* * If we get a head of queue tag, SAM-3 says that we should * immediately execute it. * * What happens if this command would normally block for some other * reason? e.g. a request sense with a head of queue tag * immediately after a write. Normally that would block, but this * will result in its getting executed immediately... * * We currently return "pass" instead of "skip", so we'll end up * going through the rest of the queue to check for overlapped tags. * * XXX KDM check for other types of blockage first?? */ if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE) return (CTL_ACTION_PASS); /* * Ordered tags have to block until all items ahead of them * have completed. If we get called with an ordered tag, we always * block, if something else is ahead of us in the queue. */ if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED) return (CTL_ACTION_BLOCK); /* * Simple tags get blocked until all head of queue and ordered tags * ahead of them have completed. I'm lumping untagged commands in * with simple tags here. XXX KDM is that the right thing to do? */ if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED) || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE)) && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE) || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED))) return (CTL_ACTION_BLOCK); pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL); KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT, ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p", __func__, pending_entry->seridx, pending_io->scsiio.cdb[0], pending_io->scsiio.cdb[1], pending_io)); ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL); if (ooa_entry->seridx == CTL_SERIDX_INVLD) return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */ KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT, ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p", __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0], ooa_io->scsiio.cdb[1], ooa_io)); serialize_row = ctl_serialize_table[ooa_entry->seridx]; switch (serialize_row[pending_entry->seridx]) { case CTL_SER_BLOCK: return (CTL_ACTION_BLOCK); case CTL_SER_EXTENT: return (ctl_extent_check(ooa_io, pending_io, (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON))); case CTL_SER_EXTENTOPT: if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED) return (ctl_extent_check(ooa_io, pending_io, (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON))); return (CTL_ACTION_PASS); case CTL_SER_EXTENTSEQ: if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF) return (ctl_extent_check_seq(ooa_io, pending_io)); return (CTL_ACTION_PASS); case CTL_SER_PASS: return (CTL_ACTION_PASS); case CTL_SER_BLOCKOPT: if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED) return (CTL_ACTION_BLOCK); return (CTL_ACTION_PASS); case CTL_SER_SKIP: return (CTL_ACTION_SKIP); default: panic("%s: Invalid serialization value %d for %d => %d", __func__, serialize_row[pending_entry->seridx], pending_entry->seridx, ooa_entry->seridx); } return (CTL_ACTION_ERROR); } /* * Check for blockage or overlaps against the OOA (Order Of Arrival) queue. * Assumptions: * - pending_io is generally either incoming, or on the blocked queue * - starting I/O is the I/O we want to start the check with. */ static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io, union ctl_io *starting_io) { union ctl_io *ooa_io; ctl_action action; mtx_assert(&lun->lun_lock, MA_OWNED); /* * Run back along the OOA queue, starting with the current * blocked I/O and going through every I/O before it on the * queue. If starting_io is NULL, we'll just end up returning * CTL_ACTION_PASS. */ for (ooa_io = starting_io; ooa_io != NULL; ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq, ooa_links)){ /* * This routine just checks to see whether * cur_blocked is blocked by ooa_io, which is ahead * of it in the queue. It doesn't queue/dequeue * cur_blocked. */ action = ctl_check_for_blockage(lun, pending_io, ooa_io); switch (action) { case CTL_ACTION_BLOCK: case CTL_ACTION_OVERLAP: case CTL_ACTION_OVERLAP_TAG: case CTL_ACTION_SKIP: case CTL_ACTION_ERROR: return (action); break; /* NOTREACHED */ case CTL_ACTION_PASS: break; default: panic("%s: Invalid action %d\n", __func__, action); } } return (CTL_ACTION_PASS); } /* * Assumptions: * - An I/O has just completed, and has been removed from the per-LUN OOA * queue, so some items on the blocked queue may now be unblocked. */ static int ctl_check_blocked(struct ctl_lun *lun) { struct ctl_softc *softc = lun->ctl_softc; union ctl_io *cur_blocked, *next_blocked; mtx_assert(&lun->lun_lock, MA_OWNED); /* * Run forward from the head of the blocked queue, checking each * entry against the I/Os prior to it on the OOA queue to see if * there is still any blockage. * * We cannot use the TAILQ_FOREACH() macro, because it can't deal * with our removing a variable on it while it is traversing the * list. */ for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue); cur_blocked != NULL; cur_blocked = next_blocked) { union ctl_io *prev_ooa; ctl_action action; next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr, blocked_links); prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr, ctl_ooaq, ooa_links); /* * If cur_blocked happens to be the first item in the OOA * queue now, prev_ooa will be NULL, and the action * returned will just be CTL_ACTION_PASS. */ action = ctl_check_ooa(lun, cur_blocked, prev_ooa); switch (action) { case CTL_ACTION_BLOCK: /* Nothing to do here, still blocked */ break; case CTL_ACTION_OVERLAP: case CTL_ACTION_OVERLAP_TAG: /* * This shouldn't happen! In theory we've already * checked this command for overlap... */ break; case CTL_ACTION_PASS: case CTL_ACTION_SKIP: { const struct ctl_cmd_entry *entry; /* * The skip case shouldn't happen, this transaction * should have never made it onto the blocked queue. */ /* * This I/O is no longer blocked, we can remove it * from the blocked queue. Since this is a TAILQ * (doubly linked list), we can do O(1) removals * from any place on the list. */ TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr, blocked_links); cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED; if ((softc->ha_mode != CTL_HA_MODE_XFER) && (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){ /* * Need to send IO back to original side to * run */ union ctl_ha_msg msg_info; cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; msg_info.hdr.original_sc = cur_blocked->io_hdr.original_sc; msg_info.hdr.serializing_sc = cur_blocked; msg_info.hdr.msg_type = CTL_MSG_R2R; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.hdr), M_NOWAIT); break; } entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL); /* * Check this I/O for LUN state changes that may * have happened while this command was blocked. * The LUN state may have been changed by a command * ahead of us in the queue, so we need to re-check * for any states that can be caused by SCSI * commands. */ if (ctl_scsiio_lun_check(lun, entry, &cur_blocked->scsiio) == 0) { cur_blocked->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; ctl_enqueue_rtr(cur_blocked); } else ctl_done(cur_blocked); break; } default: /* * This probably shouldn't happen -- we shouldn't * get CTL_ACTION_ERROR, or anything else. */ break; } } return (CTL_RETVAL_COMPLETE); } /* * This routine (with one exception) checks LUN flags that can be set by * commands ahead of us in the OOA queue. These flags have to be checked * when a command initially comes in, and when we pull a command off the * blocked queue and are preparing to execute it. The reason we have to * check these flags for commands on the blocked queue is that the LUN * state may have been changed by a command ahead of us while we're on the * blocked queue. * * Ordering is somewhat important with these checks, so please pay * careful attention to the placement of any new checks. */ static int ctl_scsiio_lun_check(struct ctl_lun *lun, const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio) { struct ctl_softc *softc = lun->ctl_softc; int retval; uint32_t residx; retval = 0; mtx_assert(&lun->lun_lock, MA_OWNED); /* * If this shelf is a secondary shelf controller, we may have to * reject some commands disallowed by HA mode and link state. */ if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) { if (softc->ha_link == CTL_HA_LINK_OFFLINE && (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) { ctl_set_lun_unavail(ctsio); retval = 1; goto bailout; } if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 && (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) { ctl_set_lun_transit(ctsio); retval = 1; goto bailout; } if (softc->ha_mode == CTL_HA_MODE_ACT_STBY && (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) { ctl_set_lun_standby(ctsio); retval = 1; goto bailout; } /* The rest of checks are only done on executing side */ if (softc->ha_mode == CTL_HA_MODE_XFER) goto bailout; } if (entry->pattern & CTL_LUN_PAT_WRITE) { if (lun->be_lun && lun->be_lun->flags & CTL_LUN_FLAG_READONLY) { ctl_set_hw_write_protected(ctsio); retval = 1; goto bailout; } if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_DATA_PROTECT, /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE); retval = 1; goto bailout; } } /* * Check for a reservation conflict. If this command isn't allowed * even on reserved LUNs, and if this initiator isn't the one who * reserved us, reject the command with a reservation conflict. */ residx = ctl_get_initindex(&ctsio->io_hdr.nexus); if ((lun->flags & CTL_LUN_RESERVED) && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) { if (lun->res_idx != residx) { ctl_set_reservation_conflict(ctsio); retval = 1; goto bailout; } } if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 || (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) { /* No reservation or command is allowed. */; } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) && (lun->pr_res_type == SPR_TYPE_WR_EX || lun->pr_res_type == SPR_TYPE_WR_EX_RO || lun->pr_res_type == SPR_TYPE_WR_EX_AR)) { /* The command is allowed for Write Exclusive resv. */; } else { /* * if we aren't registered or it's a res holder type * reservation and this isn't the res holder then set a * conflict. */ if (ctl_get_prkey(lun, residx) == 0 || (residx != lun->pr_res_idx && lun->pr_res_type < 4)) { ctl_set_reservation_conflict(ctsio); retval = 1; goto bailout; } } if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) { if (lun->flags & CTL_LUN_EJECTED) ctl_set_lun_ejected(ctsio); else if (lun->flags & CTL_LUN_NO_MEDIA) { if (lun->flags & CTL_LUN_REMOVABLE) ctl_set_lun_no_media(ctsio); else ctl_set_lun_int_reqd(ctsio); } else if (lun->flags & CTL_LUN_STOPPED) ctl_set_lun_stopped(ctsio); else goto bailout; retval = 1; goto bailout; } bailout: return (retval); } static void ctl_failover_io(union ctl_io *io, int have_lock) { ctl_set_busy(&io->scsiio); ctl_done(io); } static void ctl_failover_lun(union ctl_io *rio) { struct ctl_softc *softc = CTL_SOFTC(rio); struct ctl_lun *lun; struct ctl_io_hdr *io, *next_io; uint32_t targ_lun; targ_lun = rio->io_hdr.nexus.targ_mapped_lun; CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun)); /* Find and lock the LUN. */ mtx_lock(&softc->ctl_lock); if (targ_lun > ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); return; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); return; } if (softc->ha_mode == CTL_HA_MODE_XFER) { TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) { /* We are master */ if (io->flags & CTL_FLAG_FROM_OTHER_SC) { if (io->flags & CTL_FLAG_IO_ACTIVE) { io->flags |= CTL_FLAG_ABORT; io->flags |= CTL_FLAG_FAILOVER; } else { /* This can be only due to DATAMOVE */ io->msg_type = CTL_MSG_DATAMOVE_DONE; io->flags &= ~CTL_FLAG_DMA_INPROG; io->flags |= CTL_FLAG_IO_ACTIVE; io->port_status = 31340; ctl_enqueue_isc((union ctl_io *)io); } } /* We are slave */ if (io->flags & CTL_FLAG_SENT_2OTHER_SC) { io->flags &= ~CTL_FLAG_SENT_2OTHER_SC; if (io->flags & CTL_FLAG_IO_ACTIVE) { io->flags |= CTL_FLAG_FAILOVER; } else { ctl_set_busy(&((union ctl_io *)io)-> scsiio); ctl_done((union ctl_io *)io); } } } } else { /* SERIALIZE modes */ TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links, next_io) { /* We are master */ if (io->flags & CTL_FLAG_FROM_OTHER_SC) { TAILQ_REMOVE(&lun->blocked_queue, io, blocked_links); io->flags &= ~CTL_FLAG_BLOCKED; TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links); ctl_free_io((union ctl_io *)io); } } TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) { /* We are master */ if (io->flags & CTL_FLAG_FROM_OTHER_SC) { TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links); ctl_free_io((union ctl_io *)io); } /* We are slave */ if (io->flags & CTL_FLAG_SENT_2OTHER_SC) { io->flags &= ~CTL_FLAG_SENT_2OTHER_SC; if (!(io->flags & CTL_FLAG_IO_ACTIVE)) { ctl_set_busy(&((union ctl_io *)io)-> scsiio); ctl_done((union ctl_io *)io); } } } ctl_check_blocked(lun); } mtx_unlock(&lun->lun_lock); } static int ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio) { struct ctl_lun *lun; const struct ctl_cmd_entry *entry; uint32_t initidx, targ_lun; int retval = 0; lun = NULL; targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun; if (targ_lun < ctl_max_luns) lun = softc->ctl_luns[targ_lun]; if (lun) { /* * If the LUN is invalid, pretend that it doesn't exist. * It will go away as soon as all pending I/O has been * completed. */ mtx_lock(&lun->lun_lock); if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); lun = NULL; } } CTL_LUN(ctsio) = lun; if (lun) { CTL_BACKEND_LUN(ctsio) = lun->be_lun; /* * Every I/O goes into the OOA queue for a particular LUN, * and stays there until completion. */ #ifdef CTL_TIME_IO if (TAILQ_EMPTY(&lun->ooa_queue)) lun->idle_time += getsbinuptime() - lun->last_busy; #endif TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); } /* Get command entry and return error if it is unsuppotyed. */ entry = ctl_validate_command(ctsio); if (entry == NULL) { if (lun) mtx_unlock(&lun->lun_lock); return (retval); } ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK; ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK; /* * Check to see whether we can send this command to LUNs that don't * exist. This should pretty much only be the case for inquiry * and request sense. Further checks, below, really require having * a LUN, so we can't really check the command anymore. Just put * it on the rtr queue. */ if (lun == NULL) { if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) { ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; ctl_enqueue_rtr((union ctl_io *)ctsio); return (retval); } ctl_set_unsupported_lun(ctsio); ctl_done((union ctl_io *)ctsio); CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n")); return (retval); } else { /* * Make sure we support this particular command on this LUN. * e.g., we don't support writes to the control LUN. */ if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (retval); } } initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); /* * If we've got a request sense, it'll clear the contingent * allegiance condition. Otherwise, if we have a CA condition for * this initiator, clear it, because it sent down a command other * than request sense. */ if (ctsio->cdb[0] != REQUEST_SENSE) { struct scsi_sense_data *ps; ps = lun->pending_sense[initidx / CTL_MAX_INIT_PER_PORT]; if (ps != NULL) ps[initidx % CTL_MAX_INIT_PER_PORT].error_code = 0; } /* * If the command has this flag set, it handles its own unit * attention reporting, we shouldn't do anything. Otherwise we * check for any pending unit attentions, and send them back to the * initiator. We only do this when a command initially comes in, * not when we pull it off the blocked queue. * * According to SAM-3, section 5.3.2, the order that things get * presented back to the host is basically unit attentions caused * by some sort of reset event, busy status, reservation conflicts * or task set full, and finally any other status. * * One issue here is that some of the unit attentions we report * don't fall into the "reset" category (e.g. "reported luns data * has changed"). So reporting it here, before the reservation * check, may be technically wrong. I guess the only thing to do * would be to check for and report the reset events here, and then * check for the other unit attention types after we check for a * reservation conflict. * * XXX KDM need to fix this */ if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) { ctl_ua_type ua_type; u_int sense_len = 0; ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data, &sense_len, SSD_TYPE_NONE); if (ua_type != CTL_UA_NONE) { mtx_unlock(&lun->lun_lock); ctsio->scsi_status = SCSI_STATUS_CHECK_COND; ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; ctsio->sense_len = sense_len; ctl_done((union ctl_io *)ctsio); return (retval); } } if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) { mtx_unlock(&lun->lun_lock); ctl_done((union ctl_io *)ctsio); return (retval); } /* * XXX CHD this is where we want to send IO to other side if * this LUN is secondary on this SC. We will need to make a copy * of the IO and flag the IO on this side as SENT_2OTHER and the flag * the copy we send as FROM_OTHER. * We also need to stuff the address of the original IO so we can * find it easily. Something similar will need be done on the other * side so when we are done we can find the copy. */ if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 && (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 && (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) { union ctl_ha_msg msg_info; int isc_retval; ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC; ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; mtx_unlock(&lun->lun_lock); msg_info.hdr.msg_type = CTL_MSG_SERIALIZE; msg_info.hdr.original_sc = (union ctl_io *)ctsio; msg_info.hdr.serializing_sc = NULL; msg_info.hdr.nexus = ctsio->io_hdr.nexus; msg_info.scsi.tag_num = ctsio->tag_num; msg_info.scsi.tag_type = ctsio->tag_type; msg_info.scsi.cdb_len = ctsio->cdb_len; memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN); if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data), M_WAITOK)) > CTL_HA_STATUS_SUCCESS) { ctl_set_busy(ctsio); ctl_done((union ctl_io *)ctsio); return (retval); } return (retval); } switch (ctl_check_ooa(lun, (union ctl_io *)ctsio, (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq, ooa_links))) { case CTL_ACTION_BLOCK: ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED; TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr, blocked_links); mtx_unlock(&lun->lun_lock); return (retval); case CTL_ACTION_PASS: case CTL_ACTION_SKIP: ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; mtx_unlock(&lun->lun_lock); ctl_enqueue_rtr((union ctl_io *)ctsio); break; case CTL_ACTION_OVERLAP: mtx_unlock(&lun->lun_lock); ctl_set_overlapped_cmd(ctsio); ctl_done((union ctl_io *)ctsio); break; case CTL_ACTION_OVERLAP_TAG: mtx_unlock(&lun->lun_lock); ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff); ctl_done((union ctl_io *)ctsio); break; case CTL_ACTION_ERROR: default: mtx_unlock(&lun->lun_lock); ctl_set_internal_failure(ctsio, /*sks_valid*/ 0, /*retry_count*/ 0); ctl_done((union ctl_io *)ctsio); break; } return (retval); } const struct ctl_cmd_entry * ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa) { const struct ctl_cmd_entry *entry; int service_action; entry = &ctl_cmd_table[ctsio->cdb[0]]; if (sa) *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0); if (entry->flags & CTL_CMD_FLAG_SA5) { service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK; entry = &((const struct ctl_cmd_entry *) entry->execute)[service_action]; } return (entry); } const struct ctl_cmd_entry * ctl_validate_command(struct ctl_scsiio *ctsio) { const struct ctl_cmd_entry *entry; int i, sa; uint8_t diff; entry = ctl_get_cmd_entry(ctsio, &sa); if (entry->execute == NULL) { if (sa) ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 4); else ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (NULL); } KASSERT(entry->length > 0, ("Not defined length for command 0x%02x/0x%02x", ctsio->cdb[0], ctsio->cdb[1])); for (i = 1; i < entry->length; i++) { diff = ctsio->cdb[i] & ~entry->usage[i - 1]; if (diff == 0) continue; ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ i, /*bit_valid*/ 1, /*bit*/ fls(diff) - 1); ctl_done((union ctl_io *)ctsio); return (NULL); } return (entry); } static int ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry) { switch (lun_type) { case T_DIRECT: if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0) return (0); break; case T_PROCESSOR: if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) return (0); break; case T_CDROM: if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0) return (0); break; default: return (0); } return (1); } static int ctl_scsiio(struct ctl_scsiio *ctsio) { int retval; const struct ctl_cmd_entry *entry; retval = CTL_RETVAL_COMPLETE; CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0])); entry = ctl_get_cmd_entry(ctsio, NULL); /* * If this I/O has been aborted, just send it straight to * ctl_done() without executing it. */ if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) { ctl_done((union ctl_io *)ctsio); goto bailout; } /* * All the checks should have been handled by ctl_scsiio_precheck(). * We should be clear now to just execute the I/O. */ retval = entry->execute(ctsio); bailout: return (retval); } static int ctl_target_reset(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_port *port = CTL_PORT(io); struct ctl_lun *lun; uint32_t initidx; ctl_ua_type ua_type; if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) { union ctl_ha_msg msg_info; msg_info.hdr.nexus = io->io_hdr.nexus; msg_info.task.task_action = io->taskio.task_action; msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS; msg_info.hdr.original_sc = NULL; msg_info.hdr.serializing_sc = NULL; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.task), M_WAITOK); } initidx = ctl_get_initindex(&io->io_hdr.nexus); if (io->taskio.task_action == CTL_TASK_TARGET_RESET) ua_type = CTL_UA_TARG_RESET; else ua_type = CTL_UA_BUS_RESET; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { if (port != NULL && ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; ctl_do_lun_reset(lun, initidx, ua_type); } mtx_unlock(&softc->ctl_lock); io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE; return (0); } /* * The LUN should always be set. The I/O is optional, and is used to * distinguish between I/Os sent by this initiator, and by other * initiators. We set unit attention for initiators other than this one. * SAM-3 is vague on this point. It does say that a unit attention should * be established for other initiators when a LUN is reset (see section * 5.7.3), but it doesn't specifically say that the unit attention should * be established for this particular initiator when a LUN is reset. Here * is the relevant text, from SAM-3 rev 8: * * 5.7.2 When a SCSI initiator port aborts its own tasks * * When a SCSI initiator port causes its own task(s) to be aborted, no * notification that the task(s) have been aborted shall be returned to * the SCSI initiator port other than the completion response for the * command or task management function action that caused the task(s) to * be aborted and notification(s) associated with related effects of the * action (e.g., a reset unit attention condition). * * XXX KDM for now, we're setting unit attention for all initiators. */ static void ctl_do_lun_reset(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua_type) { union ctl_io *xio; int i; mtx_lock(&lun->lun_lock); /* Abort tasks. */ for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL; xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) { xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS; } /* Clear CA. */ for (i = 0; i < ctl_max_ports; i++) { free(lun->pending_sense[i], M_CTL); lun->pending_sense[i] = NULL; } /* Clear reservation. */ lun->flags &= ~CTL_LUN_RESERVED; /* Clear prevent media removal. */ if (lun->prevent) { for (i = 0; i < CTL_MAX_INITIATORS; i++) ctl_clear_mask(lun->prevent, i); lun->prevent_count = 0; } /* Clear TPC status */ ctl_tpc_lun_clear(lun, -1); /* Establish UA. */ #if 0 ctl_est_ua_all(lun, initidx, ua_type); #else ctl_est_ua_all(lun, -1, ua_type); #endif mtx_unlock(&lun->lun_lock); } static int ctl_lun_reset(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_lun *lun; uint32_t targ_lun, initidx; targ_lun = io->io_hdr.nexus.targ_mapped_lun; initidx = ctl_get_initindex(&io->io_hdr.nexus); mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST; return (1); } ctl_do_lun_reset(lun, initidx, CTL_UA_LUN_RESET); mtx_unlock(&softc->ctl_lock); io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE; if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) { union ctl_ha_msg msg_info; msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS; msg_info.hdr.nexus = io->io_hdr.nexus; msg_info.task.task_action = CTL_TASK_LUN_RESET; msg_info.hdr.original_sc = NULL; msg_info.hdr.serializing_sc = NULL; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.task), M_WAITOK); } return (0); } static void ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id, int other_sc) { union ctl_io *xio; mtx_assert(&lun->lun_lock, MA_OWNED); /* * Run through the OOA queue and attempt to find the given I/O. * The target port, initiator ID, tag type and tag number have to * match the values that we got from the initiator. If we have an * untagged command to abort, simply abort the first untagged command * we come to. We only allow one untagged command at a time of course. */ for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL; xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) { if ((targ_port == UINT32_MAX || targ_port == xio->io_hdr.nexus.targ_port) && (init_id == UINT32_MAX || init_id == xio->io_hdr.nexus.initid)) { if (targ_port != xio->io_hdr.nexus.targ_port || init_id != xio->io_hdr.nexus.initid) xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS; xio->io_hdr.flags |= CTL_FLAG_ABORT; if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) { union ctl_ha_msg msg_info; msg_info.hdr.nexus = xio->io_hdr.nexus; msg_info.task.task_action = CTL_TASK_ABORT_TASK; msg_info.task.tag_num = xio->scsiio.tag_num; msg_info.task.tag_type = xio->scsiio.tag_type; msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS; msg_info.hdr.original_sc = NULL; msg_info.hdr.serializing_sc = NULL; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.task), M_NOWAIT); } } } } static int ctl_abort_task_set(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_lun *lun; uint32_t targ_lun; /* * Look up the LUN. */ targ_lun = io->io_hdr.nexus.targ_mapped_lun; mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST; return (1); } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) { ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.initid, (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0); } else { /* CTL_TASK_CLEAR_TASK_SET */ ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX, (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0); } mtx_unlock(&lun->lun_lock); io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE; return (0); } static void ctl_i_t_nexus_loss(struct ctl_softc *softc, uint32_t initidx, ctl_ua_type ua_type) { struct ctl_lun *lun; struct scsi_sense_data *ps; uint32_t p, i; p = initidx / CTL_MAX_INIT_PER_PORT; i = initidx % CTL_MAX_INIT_PER_PORT; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_lock(&lun->lun_lock); /* Abort tasks. */ ctl_abort_tasks_lun(lun, p, i, 1); /* Clear CA. */ ps = lun->pending_sense[p]; if (ps != NULL) ps[i].error_code = 0; /* Clear reservation. */ if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx)) lun->flags &= ~CTL_LUN_RESERVED; /* Clear prevent media removal. */ if (lun->prevent && ctl_is_set(lun->prevent, initidx)) { ctl_clear_mask(lun->prevent, initidx); lun->prevent_count--; } /* Clear TPC status */ ctl_tpc_lun_clear(lun, initidx); /* Establish UA. */ ctl_est_ua(lun, initidx, ua_type); mtx_unlock(&lun->lun_lock); } mtx_unlock(&softc->ctl_lock); } static int ctl_i_t_nexus_reset(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); uint32_t initidx; if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) { union ctl_ha_msg msg_info; msg_info.hdr.nexus = io->io_hdr.nexus; msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET; msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS; msg_info.hdr.original_sc = NULL; msg_info.hdr.serializing_sc = NULL; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.task), M_WAITOK); } initidx = ctl_get_initindex(&io->io_hdr.nexus); ctl_i_t_nexus_loss(softc, initidx, CTL_UA_I_T_NEXUS_LOSS); io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE; return (0); } static int ctl_abort_task(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); union ctl_io *xio; struct ctl_lun *lun; #if 0 struct sbuf sb; char printbuf[128]; #endif int found; uint32_t targ_lun; found = 0; /* * Look up the LUN. */ targ_lun = io->io_hdr.nexus.targ_mapped_lun; mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST; return (1); } #if 0 printf("ctl_abort_task: called for lun %lld, tag %d type %d\n", lun->lun, io->taskio.tag_num, io->taskio.tag_type); #endif mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); /* * Run through the OOA queue and attempt to find the given I/O. * The target port, initiator ID, tag type and tag number have to * match the values that we got from the initiator. If we have an * untagged command to abort, simply abort the first untagged command * we come to. We only allow one untagged command at a time of course. */ for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL; xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) { #if 0 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN); sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ", lun->lun, xio->scsiio.tag_num, xio->scsiio.tag_type, (xio->io_hdr.blocked_links.tqe_prev == NULL) ? "" : " BLOCKED", (xio->io_hdr.flags & CTL_FLAG_DMA_INPROG) ? " DMA" : "", (xio->io_hdr.flags & CTL_FLAG_ABORT) ? " ABORT" : "", (xio->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : "")); ctl_scsi_command_string(&xio->scsiio, NULL, &sb); sbuf_finish(&sb); printf("%s\n", sbuf_data(&sb)); #endif if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port) || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid) || (xio->io_hdr.flags & CTL_FLAG_ABORT)) continue; /* * If the abort says that the task is untagged, the * task in the queue must be untagged. Otherwise, * we just check to see whether the tag numbers * match. This is because the QLogic firmware * doesn't pass back the tag type in an abort * request. */ #if 0 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED) && (io->taskio.tag_type == CTL_TAG_UNTAGGED)) || (xio->scsiio.tag_num == io->taskio.tag_num)) #endif /* * XXX KDM we've got problems with FC, because it * doesn't send down a tag type with aborts. So we * can only really go by the tag number... * This may cause problems with parallel SCSI. * Need to figure that out!! */ if (xio->scsiio.tag_num == io->taskio.tag_num) { xio->io_hdr.flags |= CTL_FLAG_ABORT; found = 1; if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 && !(lun->flags & CTL_LUN_PRIMARY_SC)) { union ctl_ha_msg msg_info; msg_info.hdr.nexus = io->io_hdr.nexus; msg_info.task.task_action = CTL_TASK_ABORT_TASK; msg_info.task.tag_num = io->taskio.tag_num; msg_info.task.tag_type = io->taskio.tag_type; msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS; msg_info.hdr.original_sc = NULL; msg_info.hdr.serializing_sc = NULL; #if 0 printf("Sent Abort to other side\n"); #endif ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info.task), M_NOWAIT); } #if 0 printf("ctl_abort_task: found I/O to abort\n"); #endif } } mtx_unlock(&lun->lun_lock); if (found == 0) { /* * This isn't really an error. It's entirely possible for * the abort and command completion to cross on the wire. * This is more of an informative/diagnostic error. */ #if 0 printf("ctl_abort_task: ABORT sent for nonexistent I/O: " "%u:%u:%u tag %d type %d\n", io->io_hdr.nexus.initid, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun, io->taskio.tag_num, io->taskio.tag_type); #endif } io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE; return (0); } static int ctl_query_task(union ctl_io *io, int task_set) { struct ctl_softc *softc = CTL_SOFTC(io); union ctl_io *xio; struct ctl_lun *lun; int found = 0; uint32_t targ_lun; targ_lun = io->io_hdr.nexus.targ_mapped_lun; mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST; return (1); } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL; xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) { if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port) || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid) || (xio->io_hdr.flags & CTL_FLAG_ABORT)) continue; if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) { found = 1; break; } } mtx_unlock(&lun->lun_lock); if (found) io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED; else io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE; return (0); } static int ctl_query_async_event(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_lun *lun; ctl_ua_type ua; uint32_t targ_lun, initidx; targ_lun = io->io_hdr.nexus.targ_mapped_lun; mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST; return (1); } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); initidx = ctl_get_initindex(&io->io_hdr.nexus); ua = ctl_build_qae(lun, initidx, io->taskio.task_resp); mtx_unlock(&lun->lun_lock); if (ua != CTL_UA_NONE) io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED; else io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE; return (0); } static void ctl_run_task(union ctl_io *io) { int retval = 1; CTL_DEBUG_PRINT(("ctl_run_task\n")); KASSERT(io->io_hdr.io_type == CTL_IO_TASK, ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type)); io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED; bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp)); switch (io->taskio.task_action) { case CTL_TASK_ABORT_TASK: retval = ctl_abort_task(io); break; case CTL_TASK_ABORT_TASK_SET: case CTL_TASK_CLEAR_TASK_SET: retval = ctl_abort_task_set(io); break; case CTL_TASK_CLEAR_ACA: break; case CTL_TASK_I_T_NEXUS_RESET: retval = ctl_i_t_nexus_reset(io); break; case CTL_TASK_LUN_RESET: retval = ctl_lun_reset(io); break; case CTL_TASK_TARGET_RESET: case CTL_TASK_BUS_RESET: retval = ctl_target_reset(io); break; case CTL_TASK_PORT_LOGIN: break; case CTL_TASK_PORT_LOGOUT: break; case CTL_TASK_QUERY_TASK: retval = ctl_query_task(io, 0); break; case CTL_TASK_QUERY_TASK_SET: retval = ctl_query_task(io, 1); break; case CTL_TASK_QUERY_ASYNC_EVENT: retval = ctl_query_async_event(io); break; default: printf("%s: got unknown task management event %d\n", __func__, io->taskio.task_action); break; } if (retval == 0) io->io_hdr.status = CTL_SUCCESS; else io->io_hdr.status = CTL_ERROR; ctl_done(io); } /* * For HA operation. Handle commands that come in from the other * controller. */ static void ctl_handle_isc(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_lun *lun; const struct ctl_cmd_entry *entry; uint32_t targ_lun; targ_lun = io->io_hdr.nexus.targ_mapped_lun; switch (io->io_hdr.msg_type) { case CTL_MSG_SERIALIZE: ctl_serialize_other_sc_cmd(&io->scsiio); break; case CTL_MSG_R2R: /* Only used in SER_ONLY mode. */ entry = ctl_get_cmd_entry(&io->scsiio, NULL); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { ctl_done(io); break; } mtx_lock(&lun->lun_lock); if (ctl_scsiio_lun_check(lun, entry, &io->scsiio) != 0) { mtx_unlock(&lun->lun_lock); ctl_done(io); break; } io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; mtx_unlock(&lun->lun_lock); ctl_enqueue_rtr(io); break; case CTL_MSG_FINISH_IO: if (softc->ha_mode == CTL_HA_MODE_XFER) { ctl_done(io); break; } if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { ctl_free_io(io); break; } mtx_lock(&lun->lun_lock); TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links); ctl_check_blocked(lun); mtx_unlock(&lun->lun_lock); ctl_free_io(io); break; case CTL_MSG_PERS_ACTION: ctl_hndl_per_res_out_on_other_sc(io); ctl_free_io(io); break; case CTL_MSG_BAD_JUJU: ctl_done(io); break; case CTL_MSG_DATAMOVE: /* Only used in XFER mode */ ctl_datamove_remote(io); break; case CTL_MSG_DATAMOVE_DONE: /* Only used in XFER mode */ io->scsiio.be_move_done(io); break; case CTL_MSG_FAILOVER: ctl_failover_lun(io); ctl_free_io(io); break; default: printf("%s: Invalid message type %d\n", __func__, io->io_hdr.msg_type); ctl_free_io(io); break; } } /* * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if * there is no match. */ static ctl_lun_error_pattern ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc) { const struct ctl_cmd_entry *entry; ctl_lun_error_pattern filtered_pattern, pattern; pattern = desc->error_pattern; /* * XXX KDM we need more data passed into this function to match a * custom pattern, and we actually need to implement custom pattern * matching. */ if (pattern & CTL_LUN_PAT_CMD) return (CTL_LUN_PAT_CMD); if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY) return (CTL_LUN_PAT_ANY); entry = ctl_get_cmd_entry(ctsio, NULL); filtered_pattern = entry->pattern & pattern; /* * If the user requested specific flags in the pattern (e.g. * CTL_LUN_PAT_RANGE), make sure the command supports all of those * flags. * * If the user did not specify any flags, it doesn't matter whether * or not the command supports the flags. */ if ((filtered_pattern & ~CTL_LUN_PAT_MASK) != (pattern & ~CTL_LUN_PAT_MASK)) return (CTL_LUN_PAT_NONE); /* * If the user asked for a range check, see if the requested LBA * range overlaps with this command's LBA range. */ if (filtered_pattern & CTL_LUN_PAT_RANGE) { uint64_t lba1; uint64_t len1; ctl_action action; int retval; retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1); if (retval != 0) return (CTL_LUN_PAT_NONE); action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba, desc->lba_range.len, FALSE); /* * A "pass" means that the LBA ranges don't overlap, so * this doesn't match the user's range criteria. */ if (action == CTL_ACTION_PASS) return (CTL_LUN_PAT_NONE); } return (filtered_pattern); } static void ctl_inject_error(struct ctl_lun *lun, union ctl_io *io) { struct ctl_error_desc *desc, *desc2; mtx_assert(&lun->lun_lock, MA_OWNED); STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) { ctl_lun_error_pattern pattern; /* * Check to see whether this particular command matches * the pattern in the descriptor. */ pattern = ctl_cmd_pattern_match(&io->scsiio, desc); if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE) continue; switch (desc->lun_error & CTL_LUN_INJ_TYPE) { case CTL_LUN_INJ_ABORTED: ctl_set_aborted(&io->scsiio); break; case CTL_LUN_INJ_MEDIUM_ERR: ctl_set_medium_error(&io->scsiio, (io->io_hdr.flags & CTL_FLAG_DATA_MASK) != CTL_FLAG_DATA_OUT); break; case CTL_LUN_INJ_UA: /* 29h/00h POWER ON, RESET, OR BUS DEVICE RESET * OCCURRED */ ctl_set_ua(&io->scsiio, 0x29, 0x00); break; case CTL_LUN_INJ_CUSTOM: /* * We're assuming the user knows what he is doing. * Just copy the sense information without doing * checks. */ bcopy(&desc->custom_sense, &io->scsiio.sense_data, MIN(sizeof(desc->custom_sense), sizeof(io->scsiio.sense_data))); io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND; io->scsiio.sense_len = SSD_FULL_SIZE; io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; break; case CTL_LUN_INJ_NONE: default: /* * If this is an error injection type we don't know * about, clear the continuous flag (if it is set) * so it will get deleted below. */ desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS; break; } /* * By default, each error injection action is a one-shot */ if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS) continue; STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links); free(desc, M_CTL); } } #ifdef CTL_IO_DELAY static void ctl_datamove_timer_wakeup(void *arg) { union ctl_io *io; io = (union ctl_io *)arg; ctl_datamove(io); } #endif /* CTL_IO_DELAY */ void ctl_datamove(union ctl_io *io) { void (*fe_datamove)(union ctl_io *io); mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED); CTL_DEBUG_PRINT(("ctl_datamove\n")); /* No data transferred yet. Frontend must update this when done. */ io->scsiio.kern_data_resid = io->scsiio.kern_data_len; #ifdef CTL_TIME_IO if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) { char str[256]; char path_str[64]; struct sbuf sb; ctl_scsi_path_string(io, path_str, sizeof(path_str)); sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN); sbuf_cat(&sb, path_str); switch (io->io_hdr.io_type) { case CTL_IO_SCSI: ctl_scsi_command_string(&io->scsiio, NULL, &sb); sbuf_printf(&sb, "\n"); sbuf_cat(&sb, path_str); sbuf_printf(&sb, "Tag: 0x%04x, type %d\n", io->scsiio.tag_num, io->scsiio.tag_type); break; case CTL_IO_TASK: sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, " "Tag Type: %d\n", io->taskio.task_action, io->taskio.tag_num, io->taskio.tag_type); break; default: panic("%s: Invalid CTL I/O type %d\n", __func__, io->io_hdr.io_type); } sbuf_cat(&sb, path_str); sbuf_printf(&sb, "ctl_datamove: %jd seconds\n", (intmax_t)time_uptime - io->io_hdr.start_time); sbuf_finish(&sb); printf("%s", sbuf_data(&sb)); } #endif /* CTL_TIME_IO */ #ifdef CTL_IO_DELAY if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) { io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE; } else { struct ctl_lun *lun; lun = CTL_LUN(io); if ((lun != NULL) && (lun->delay_info.datamove_delay > 0)) { callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1); io->io_hdr.flags |= CTL_FLAG_DELAY_DONE; callout_reset(&io->io_hdr.delay_callout, lun->delay_info.datamove_delay * hz, ctl_datamove_timer_wakeup, io); if (lun->delay_info.datamove_type == CTL_DELAY_TYPE_ONESHOT) lun->delay_info.datamove_delay = 0; return; } } #endif /* * This command has been aborted. Set the port status, so we fail * the data move. */ if (io->io_hdr.flags & CTL_FLAG_ABORT) { printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n", io->scsiio.tag_num, io->io_hdr.nexus.initid, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun); io->io_hdr.port_status = 31337; /* * Note that the backend, in this case, will get the * callback in its context. In other cases it may get * called in the frontend's interrupt thread context. */ io->scsiio.be_move_done(io); return; } /* Don't confuse frontend with zero length data move. */ if (io->scsiio.kern_data_len == 0) { io->scsiio.be_move_done(io); return; } fe_datamove = CTL_PORT(io)->fe_datamove; fe_datamove(io); } static void ctl_send_datamove_done(union ctl_io *io, int have_lock) { union ctl_ha_msg msg; #ifdef CTL_TIME_IO struct bintime cur_bt; #endif memset(&msg, 0, sizeof(msg)); msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE; msg.hdr.original_sc = io; msg.hdr.serializing_sc = io->io_hdr.serializing_sc; msg.hdr.nexus = io->io_hdr.nexus; msg.hdr.status = io->io_hdr.status; msg.scsi.kern_data_resid = io->scsiio.kern_data_resid; msg.scsi.tag_num = io->scsiio.tag_num; msg.scsi.tag_type = io->scsiio.tag_type; msg.scsi.scsi_status = io->scsiio.scsi_status; memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data, io->scsiio.sense_len); msg.scsi.sense_len = io->scsiio.sense_len; msg.scsi.port_status = io->io_hdr.port_status; io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; if (io->io_hdr.flags & CTL_FLAG_FAILOVER) { ctl_failover_io(io, /*have_lock*/ have_lock); return; } ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) + msg.scsi.sense_len, M_WAITOK); #ifdef CTL_TIME_IO getbinuptime(&cur_bt); bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); bintime_add(&io->io_hdr.dma_bt, &cur_bt); #endif io->io_hdr.num_dmas++; } /* * The DMA to the remote side is done, now we need to tell the other side * we're done so it can continue with its data movement. */ static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq) { union ctl_io *io; uint32_t i; io = rq->context; if (rq->ret != CTL_HA_STATUS_SUCCESS) { printf("%s: ISC DMA write failed with error %d", __func__, rq->ret); ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ rq->ret); } ctl_dt_req_free(rq); for (i = 0; i < io->scsiio.kern_sg_entries; i++) free(io->io_hdr.local_sglist[i].addr, M_CTL); free(io->io_hdr.remote_sglist, M_CTL); io->io_hdr.remote_sglist = NULL; io->io_hdr.local_sglist = NULL; /* * The data is in local and remote memory, so now we need to send * status (good or back) back to the other side. */ ctl_send_datamove_done(io, /*have_lock*/ 0); } /* * We've moved the data from the host/controller into local memory. Now we * need to push it over to the remote controller's memory. */ static int ctl_datamove_remote_dm_write_cb(union ctl_io *io) { int retval; retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE, ctl_datamove_remote_write_cb); return (retval); } static void ctl_datamove_remote_write(union ctl_io *io) { int retval; void (*fe_datamove)(union ctl_io *io); /* * - Get the data from the host/HBA into local memory. * - DMA memory from the local controller to the remote controller. * - Send status back to the remote controller. */ retval = ctl_datamove_remote_sgl_setup(io); if (retval != 0) return; /* Switch the pointer over so the FETD knows what to do */ io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist; /* * Use a custom move done callback, since we need to send completion * back to the other controller, not to the backend on this side. */ io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb; fe_datamove = CTL_PORT(io)->fe_datamove; fe_datamove(io); } static int ctl_datamove_remote_dm_read_cb(union ctl_io *io) { #if 0 char str[256]; char path_str[64]; struct sbuf sb; #endif uint32_t i; for (i = 0; i < io->scsiio.kern_sg_entries; i++) free(io->io_hdr.local_sglist[i].addr, M_CTL); free(io->io_hdr.remote_sglist, M_CTL); io->io_hdr.remote_sglist = NULL; io->io_hdr.local_sglist = NULL; #if 0 scsi_path_string(io, path_str, sizeof(path_str)); sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN); sbuf_cat(&sb, path_str); scsi_command_string(&io->scsiio, NULL, &sb); sbuf_printf(&sb, "\n"); sbuf_cat(&sb, path_str); sbuf_printf(&sb, "Tag: 0x%04x, type %d\n", io->scsiio.tag_num, io->scsiio.tag_type); sbuf_cat(&sb, path_str); sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__, io->io_hdr.flags, io->io_hdr.status); sbuf_finish(&sb); printk("%s", sbuf_data(&sb)); #endif /* * The read is done, now we need to send status (good or bad) back * to the other side. */ ctl_send_datamove_done(io, /*have_lock*/ 0); return (0); } static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq) { union ctl_io *io; void (*fe_datamove)(union ctl_io *io); io = rq->context; if (rq->ret != CTL_HA_STATUS_SUCCESS) { printf("%s: ISC DMA read failed with error %d\n", __func__, rq->ret); ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ rq->ret); } ctl_dt_req_free(rq); /* Switch the pointer over so the FETD knows what to do */ io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist; /* * Use a custom move done callback, since we need to send completion * back to the other controller, not to the backend on this side. */ io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb; /* XXX KDM add checks like the ones in ctl_datamove? */ fe_datamove = CTL_PORT(io)->fe_datamove; fe_datamove(io); } static int ctl_datamove_remote_sgl_setup(union ctl_io *io) { struct ctl_sg_entry *local_sglist; uint32_t len_to_go; int retval; int i; retval = 0; local_sglist = io->io_hdr.local_sglist; len_to_go = io->scsiio.kern_data_len; /* * The difficult thing here is that the size of the various * S/G segments may be different than the size from the * remote controller. That'll make it harder when DMAing * the data back to the other side. */ for (i = 0; len_to_go > 0; i++) { local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT); local_sglist[i].addr = malloc(local_sglist[i].len, M_CTL, M_WAITOK); len_to_go -= local_sglist[i].len; } /* * Reset the number of S/G entries accordingly. The original * number of S/G entries is available in rem_sg_entries. */ io->scsiio.kern_sg_entries = i; #if 0 printf("%s: kern_sg_entries = %d\n", __func__, io->scsiio.kern_sg_entries); for (i = 0; i < io->scsiio.kern_sg_entries; i++) printf("%s: sg[%d] = %p, %lu\n", __func__, i, local_sglist[i].addr, local_sglist[i].len); #endif return (retval); } static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command, ctl_ha_dt_cb callback) { struct ctl_ha_dt_req *rq; struct ctl_sg_entry *remote_sglist, *local_sglist; uint32_t local_used, remote_used, total_used; int i, j, isc_ret; rq = ctl_dt_req_alloc(); /* * If we failed to allocate the request, and if the DMA didn't fail * anyway, set busy status. This is just a resource allocation * failure. */ if ((rq == NULL) && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) ctl_set_busy(&io->scsiio); if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) { if (rq != NULL) ctl_dt_req_free(rq); /* * The data move failed. We need to return status back * to the other controller. No point in trying to DMA * data to the remote controller. */ ctl_send_datamove_done(io, /*have_lock*/ 0); return (1); } local_sglist = io->io_hdr.local_sglist; remote_sglist = io->io_hdr.remote_sglist; local_used = 0; remote_used = 0; total_used = 0; /* * Pull/push the data over the wire from/to the other controller. * This takes into account the possibility that the local and * remote sglists may not be identical in terms of the size of * the elements and the number of elements. * * One fundamental assumption here is that the length allocated for * both the local and remote sglists is identical. Otherwise, we've * essentially got a coding error of some sort. */ isc_ret = CTL_HA_STATUS_SUCCESS; for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) { uint32_t cur_len; uint8_t *tmp_ptr; rq->command = command; rq->context = io; /* * Both pointers should be aligned. But it is possible * that the allocation length is not. They should both * also have enough slack left over at the end, though, * to round up to the next 8 byte boundary. */ cur_len = MIN(local_sglist[i].len - local_used, remote_sglist[j].len - remote_used); rq->size = cur_len; tmp_ptr = (uint8_t *)local_sglist[i].addr; tmp_ptr += local_used; #if 0 /* Use physical addresses when talking to ISC hardware */ if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) { /* XXX KDM use busdma */ rq->local = vtophys(tmp_ptr); } else rq->local = tmp_ptr; #else KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0, ("HA does not support BUS_ADDR")); rq->local = tmp_ptr; #endif tmp_ptr = (uint8_t *)remote_sglist[j].addr; tmp_ptr += remote_used; rq->remote = tmp_ptr; rq->callback = NULL; local_used += cur_len; if (local_used >= local_sglist[i].len) { i++; local_used = 0; } remote_used += cur_len; if (remote_used >= remote_sglist[j].len) { j++; remote_used = 0; } total_used += cur_len; if (total_used >= io->scsiio.kern_data_len) rq->callback = callback; #if 0 printf("%s: %s: local %p remote %p size %d\n", __func__, (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ", rq->local, rq->remote, rq->size); #endif isc_ret = ctl_dt_single(rq); if (isc_ret > CTL_HA_STATUS_SUCCESS) break; } if (isc_ret != CTL_HA_STATUS_WAIT) { rq->ret = isc_ret; callback(rq); } return (0); } static void ctl_datamove_remote_read(union ctl_io *io) { int retval; uint32_t i; /* * This will send an error to the other controller in the case of a * failure. */ retval = ctl_datamove_remote_sgl_setup(io); if (retval != 0) return; retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ, ctl_datamove_remote_read_cb); if (retval != 0) { /* * Make sure we free memory if there was an error.. The * ctl_datamove_remote_xfer() function will send the * datamove done message, or call the callback with an * error if there is a problem. */ for (i = 0; i < io->scsiio.kern_sg_entries; i++) free(io->io_hdr.local_sglist[i].addr, M_CTL); free(io->io_hdr.remote_sglist, M_CTL); io->io_hdr.remote_sglist = NULL; io->io_hdr.local_sglist = NULL; } } /* * Process a datamove request from the other controller. This is used for * XFER mode only, not SER_ONLY mode. For writes, we DMA into local memory * first. Once that is complete, the data gets DMAed into the remote * controller's memory. For reads, we DMA from the remote controller's * memory into our memory first, and then move it out to the FETD. */ static void ctl_datamove_remote(union ctl_io *io) { mtx_assert(&((struct ctl_softc *)CTL_SOFTC(io))->ctl_lock, MA_NOTOWNED); if (io->io_hdr.flags & CTL_FLAG_FAILOVER) { ctl_failover_io(io, /*have_lock*/ 0); return; } /* * Note that we look for an aborted I/O here, but don't do some of * the other checks that ctl_datamove() normally does. * We don't need to run the datamove delay code, since that should * have been done if need be on the other controller. */ if (io->io_hdr.flags & CTL_FLAG_ABORT) { printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__, io->scsiio.tag_num, io->io_hdr.nexus.initid, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun); io->io_hdr.port_status = 31338; ctl_send_datamove_done(io, /*have_lock*/ 0); return; } if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) ctl_datamove_remote_write(io); else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ctl_datamove_remote_read(io); else { io->io_hdr.port_status = 31339; ctl_send_datamove_done(io, /*have_lock*/ 0); } } static void ctl_process_done(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_port *port = CTL_PORT(io); struct ctl_lun *lun = CTL_LUN(io); void (*fe_done)(union ctl_io *io); union ctl_ha_msg msg; CTL_DEBUG_PRINT(("ctl_process_done\n")); fe_done = port->fe_done; #ifdef CTL_TIME_IO if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) { char str[256]; char path_str[64]; struct sbuf sb; ctl_scsi_path_string(io, path_str, sizeof(path_str)); sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN); sbuf_cat(&sb, path_str); switch (io->io_hdr.io_type) { case CTL_IO_SCSI: ctl_scsi_command_string(&io->scsiio, NULL, &sb); sbuf_printf(&sb, "\n"); sbuf_cat(&sb, path_str); sbuf_printf(&sb, "Tag: 0x%04x, type %d\n", io->scsiio.tag_num, io->scsiio.tag_type); break; case CTL_IO_TASK: sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, " "Tag Type: %d\n", io->taskio.task_action, io->taskio.tag_num, io->taskio.tag_type); break; default: panic("%s: Invalid CTL I/O type %d\n", __func__, io->io_hdr.io_type); } sbuf_cat(&sb, path_str); sbuf_printf(&sb, "ctl_process_done: %jd seconds\n", (intmax_t)time_uptime - io->io_hdr.start_time); sbuf_finish(&sb); printf("%s", sbuf_data(&sb)); } #endif /* CTL_TIME_IO */ switch (io->io_hdr.io_type) { case CTL_IO_SCSI: break; case CTL_IO_TASK: if (ctl_debug & CTL_DEBUG_INFO) ctl_io_error_print(io, NULL); fe_done(io); return; default: panic("%s: Invalid CTL I/O type %d\n", __func__, io->io_hdr.io_type); } if (lun == NULL) { CTL_DEBUG_PRINT(("NULL LUN for lun %d\n", io->io_hdr.nexus.targ_mapped_lun)); goto bailout; } mtx_lock(&lun->lun_lock); /* * Check to see if we have any informational exception and status * of this command can be modified to report it in form of either * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field. */ if (lun->ie_reported == 0 && lun->ie_asc != 0 && io->io_hdr.status == CTL_SUCCESS && (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) { uint8_t mrie = lun->MODE_IE.mrie; uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) || (lun->MODE_VER.byte3 & SMS_VER_PER)); if (((mrie == SIEP_MRIE_REC_COND && per) || mrie == SIEP_MRIE_REC_UNCOND || mrie == SIEP_MRIE_NO_SENSE) && (ctl_get_cmd_entry(&io->scsiio, NULL)->flags & CTL_CMD_FLAG_NO_SENSE) == 0) { ctl_set_sense(&io->scsiio, /*current_error*/ 1, /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ? SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR, /*asc*/ lun->ie_asc, /*ascq*/ lun->ie_ascq, SSD_ELEM_NONE); lun->ie_reported = 1; } } else if (lun->ie_reported < 0) lun->ie_reported = 0; /* * Check to see if we have any errors to inject here. We only * inject errors for commands that don't already have errors set. */ if (!STAILQ_EMPTY(&lun->error_list) && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) && ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0)) ctl_inject_error(lun, io); /* * XXX KDM how do we treat commands that aren't completed * successfully? * * XXX KDM should we also track I/O latency? */ if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS && io->io_hdr.io_type == CTL_IO_SCSI) { int type; #ifdef CTL_TIME_IO struct bintime bt; getbinuptime(&bt); bintime_sub(&bt, &io->io_hdr.start_bt); #endif if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) type = CTL_STATS_READ; else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT) type = CTL_STATS_WRITE; else type = CTL_STATS_NO_IO; #ifdef CTL_LEGACY_STATS uint32_t targ_port = port->targ_port; lun->legacy_stats.ports[targ_port].bytes[type] += io->scsiio.kern_total_len; lun->legacy_stats.ports[targ_port].operations[type] ++; lun->legacy_stats.ports[targ_port].num_dmas[type] += io->io_hdr.num_dmas; #ifdef CTL_TIME_IO bintime_add(&lun->legacy_stats.ports[targ_port].dma_time[type], &io->io_hdr.dma_bt); bintime_add(&lun->legacy_stats.ports[targ_port].time[type], &bt); #endif #endif /* CTL_LEGACY_STATS */ lun->stats.bytes[type] += io->scsiio.kern_total_len; lun->stats.operations[type] ++; lun->stats.dmas[type] += io->io_hdr.num_dmas; #ifdef CTL_TIME_IO bintime_add(&lun->stats.dma_time[type], &io->io_hdr.dma_bt); bintime_add(&lun->stats.time[type], &bt); #endif mtx_lock(&port->port_lock); port->stats.bytes[type] += io->scsiio.kern_total_len; port->stats.operations[type] ++; port->stats.dmas[type] += io->io_hdr.num_dmas; #ifdef CTL_TIME_IO bintime_add(&port->stats.dma_time[type], &io->io_hdr.dma_bt); bintime_add(&port->stats.time[type], &bt); #endif mtx_unlock(&port->port_lock); } /* * Remove this from the OOA queue. */ TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links); #ifdef CTL_TIME_IO if (TAILQ_EMPTY(&lun->ooa_queue)) lun->last_busy = getsbinuptime(); #endif /* * Run through the blocked queue on this LUN and see if anything * has become unblocked, now that this transaction is done. */ ctl_check_blocked(lun); /* * If the LUN has been invalidated, free it if there is nothing * left on its OOA queue. */ if ((lun->flags & CTL_LUN_INVALID) && TAILQ_EMPTY(&lun->ooa_queue)) { mtx_unlock(&lun->lun_lock); ctl_free_lun(lun); } else mtx_unlock(&lun->lun_lock); bailout: /* * If this command has been aborted, make sure we set the status * properly. The FETD is responsible for freeing the I/O and doing * whatever it needs to do to clean up its state. */ if (io->io_hdr.flags & CTL_FLAG_ABORT) ctl_set_task_aborted(&io->scsiio); /* * If enabled, print command error status. */ if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS && (ctl_debug & CTL_DEBUG_INFO) != 0) ctl_io_error_print(io, NULL); /* * Tell the FETD or the other shelf controller we're done with this * command. Note that only SCSI commands get to this point. Task * management commands are completed above. */ if ((softc->ha_mode != CTL_HA_MODE_XFER) && (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) { memset(&msg, 0, sizeof(msg)); msg.hdr.msg_type = CTL_MSG_FINISH_IO; msg.hdr.serializing_sc = io->io_hdr.serializing_sc; msg.hdr.nexus = io->io_hdr.nexus; ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.scsi) - sizeof(msg.scsi.sense_data), M_WAITOK); } fe_done(io); } /* * Front end should call this if it doesn't do autosense. When the request * sense comes back in from the initiator, we'll dequeue this and send it. */ int ctl_queue_sense(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_port *port = CTL_PORT(io); struct ctl_lun *lun; struct scsi_sense_data *ps; uint32_t initidx, p, targ_lun; CTL_DEBUG_PRINT(("ctl_queue_sense\n")); targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun); /* * LUN lookup will likely move to the ctl_work_thread() once we * have our new queueing infrastructure (that doesn't put things on * a per-LUN queue initially). That is so that we can handle * things like an INQUIRY to a LUN that we don't have enabled. We * can't deal with that right now. * If we don't have a LUN for this, just toss the sense information. */ mtx_lock(&softc->ctl_lock); if (targ_lun >= ctl_max_luns || (lun = softc->ctl_luns[targ_lun]) == NULL) { mtx_unlock(&softc->ctl_lock); goto bailout; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); initidx = ctl_get_initindex(&io->io_hdr.nexus); p = initidx / CTL_MAX_INIT_PER_PORT; if (lun->pending_sense[p] == NULL) { lun->pending_sense[p] = malloc(sizeof(*ps) * CTL_MAX_INIT_PER_PORT, M_CTL, M_NOWAIT | M_ZERO); } if ((ps = lun->pending_sense[p]) != NULL) { ps += initidx % CTL_MAX_INIT_PER_PORT; memset(ps, 0, sizeof(*ps)); memcpy(ps, &io->scsiio.sense_data, io->scsiio.sense_len); } mtx_unlock(&lun->lun_lock); bailout: ctl_free_io(io); return (CTL_RETVAL_COMPLETE); } /* * Primary command inlet from frontend ports. All SCSI and task I/O * requests must go through this function. */ int ctl_queue(union ctl_io *io) { struct ctl_port *port = CTL_PORT(io); CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0])); #ifdef CTL_TIME_IO io->io_hdr.start_time = time_uptime; getbinuptime(&io->io_hdr.start_bt); #endif /* CTL_TIME_IO */ /* Map FE-specific LUN ID into global one. */ io->io_hdr.nexus.targ_mapped_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun); switch (io->io_hdr.io_type) { case CTL_IO_SCSI: case CTL_IO_TASK: if (ctl_debug & CTL_DEBUG_CDB) ctl_io_print(io); ctl_enqueue_incoming(io); break; default: printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type); return (EINVAL); } return (CTL_RETVAL_COMPLETE); } #ifdef CTL_IO_DELAY static void ctl_done_timer_wakeup(void *arg) { union ctl_io *io; io = (union ctl_io *)arg; ctl_done(io); } #endif /* CTL_IO_DELAY */ void ctl_serseq_done(union ctl_io *io) { struct ctl_lun *lun = CTL_LUN(io);; if (lun->be_lun == NULL || lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF) return; mtx_lock(&lun->lun_lock); io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE; ctl_check_blocked(lun); mtx_unlock(&lun->lun_lock); } void ctl_done(union ctl_io *io) { /* * Enable this to catch duplicate completion issues. */ #if 0 if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) { printf("%s: type %d msg %d cdb %x iptl: " "%u:%u:%u tag 0x%04x " "flag %#x status %x\n", __func__, io->io_hdr.io_type, io->io_hdr.msg_type, io->scsiio.cdb[0], io->io_hdr.nexus.initid, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_lun, (io->io_hdr.io_type == CTL_IO_TASK) ? io->taskio.tag_num : io->scsiio.tag_num, io->io_hdr.flags, io->io_hdr.status); } else io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE; #endif /* * This is an internal copy of an I/O, and should not go through * the normal done processing logic. */ if (io->io_hdr.flags & CTL_FLAG_INT_COPY) return; #ifdef CTL_IO_DELAY if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) { io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE; } else { struct ctl_lun *lun = CTL_LUN(io); if ((lun != NULL) && (lun->delay_info.done_delay > 0)) { callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1); io->io_hdr.flags |= CTL_FLAG_DELAY_DONE; callout_reset(&io->io_hdr.delay_callout, lun->delay_info.done_delay * hz, ctl_done_timer_wakeup, io); if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT) lun->delay_info.done_delay = 0; return; } } #endif /* CTL_IO_DELAY */ ctl_enqueue_done(io); } static void ctl_work_thread(void *arg) { struct ctl_thread *thr = (struct ctl_thread *)arg; struct ctl_softc *softc = thr->ctl_softc; union ctl_io *io; int retval; CTL_DEBUG_PRINT(("ctl_work_thread starting\n")); while (!softc->shutdown) { /* * We handle the queues in this order: * - ISC * - done queue (to free up resources, unblock other commands) * - incoming queue * - RtR queue * * If those queues are empty, we break out of the loop and * go to sleep. */ mtx_lock(&thr->queue_lock); io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue); if (io != NULL) { STAILQ_REMOVE_HEAD(&thr->isc_queue, links); mtx_unlock(&thr->queue_lock); ctl_handle_isc(io); continue; } io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue); if (io != NULL) { STAILQ_REMOVE_HEAD(&thr->done_queue, links); /* clear any blocked commands, call fe_done */ mtx_unlock(&thr->queue_lock); ctl_process_done(io); continue; } io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue); if (io != NULL) { STAILQ_REMOVE_HEAD(&thr->incoming_queue, links); mtx_unlock(&thr->queue_lock); if (io->io_hdr.io_type == CTL_IO_TASK) ctl_run_task(io); else ctl_scsiio_precheck(softc, &io->scsiio); continue; } io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue); if (io != NULL) { STAILQ_REMOVE_HEAD(&thr->rtr_queue, links); mtx_unlock(&thr->queue_lock); retval = ctl_scsiio(&io->scsiio); if (retval != CTL_RETVAL_COMPLETE) CTL_DEBUG_PRINT(("ctl_scsiio failed\n")); continue; } /* Sleep until we have something to do. */ mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0); } thr->thread = NULL; kthread_exit(); } static void ctl_lun_thread(void *arg) { struct ctl_softc *softc = (struct ctl_softc *)arg; struct ctl_be_lun *be_lun; CTL_DEBUG_PRINT(("ctl_lun_thread starting\n")); while (!softc->shutdown) { mtx_lock(&softc->ctl_lock); be_lun = STAILQ_FIRST(&softc->pending_lun_queue); if (be_lun != NULL) { STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links); mtx_unlock(&softc->ctl_lock); ctl_create_lun(be_lun); continue; } /* Sleep until we have something to do. */ mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock, PDROP | PRIBIO, "-", 0); } softc->lun_thread = NULL; kthread_exit(); } static void ctl_thresh_thread(void *arg) { struct ctl_softc *softc = (struct ctl_softc *)arg; struct ctl_lun *lun; struct ctl_logical_block_provisioning_page *page; const char *attr; union ctl_ha_msg msg; uint64_t thres, val; int i, e, set; CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n")); while (!softc->shutdown) { mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { if ((lun->flags & CTL_LUN_DISABLED) || (lun->flags & CTL_LUN_NO_MEDIA) || lun->backend->lun_attr == NULL) continue; if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 && softc->ha_mode == CTL_HA_MODE_XFER) continue; if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0) continue; e = 0; page = &lun->MODE_LBP; for (i = 0; i < CTL_NUM_LBP_THRESH; i++) { if ((page->descr[i].flags & SLBPPD_ENABLED) == 0) continue; thres = scsi_4btoul(page->descr[i].count); thres <<= CTL_LBP_EXPONENT; switch (page->descr[i].resource) { case 0x01: attr = "blocksavail"; break; case 0x02: attr = "blocksused"; break; case 0xf1: attr = "poolblocksavail"; break; case 0xf2: attr = "poolblocksused"; break; default: continue; } mtx_unlock(&softc->ctl_lock); // XXX val = lun->backend->lun_attr( lun->be_lun->be_lun, attr); mtx_lock(&softc->ctl_lock); if (val == UINT64_MAX) continue; if ((page->descr[i].flags & SLBPPD_ARMING_MASK) == SLBPPD_ARMING_INC) e = (val >= thres); else e = (val <= thres); if (e) break; } mtx_lock(&lun->lun_lock); if (e) { scsi_u64to8b((uint8_t *)&page->descr[i] - (uint8_t *)page, lun->ua_tpt_info); if (lun->lasttpt == 0 || time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) { lun->lasttpt = time_uptime; ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES); set = 1; } else set = 0; } else { lun->lasttpt = 0; ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES); set = -1; } mtx_unlock(&lun->lun_lock); if (set != 0 && lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) { /* Send msg to other side. */ bzero(&msg.ua, sizeof(msg.ua)); msg.hdr.msg_type = CTL_MSG_UA; msg.hdr.nexus.initid = -1; msg.hdr.nexus.targ_port = -1; msg.hdr.nexus.targ_lun = lun->lun; msg.hdr.nexus.targ_mapped_lun = lun->lun; msg.ua.ua_all = 1; msg.ua.ua_set = (set > 0); msg.ua.ua_type = CTL_UA_THIN_PROV_THRES; memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8); mtx_unlock(&softc->ctl_lock); // XXX ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua), M_WAITOK); mtx_lock(&softc->ctl_lock); } } mtx_sleep(&softc->thresh_thread, &softc->ctl_lock, PDROP | PRIBIO, "-", CTL_LBP_PERIOD * hz); } softc->thresh_thread = NULL; kthread_exit(); } static void ctl_enqueue_incoming(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_thread *thr; u_int idx; idx = (io->io_hdr.nexus.targ_port * 127 + io->io_hdr.nexus.initid) % worker_threads; thr = &softc->threads[idx]; mtx_lock(&thr->queue_lock); STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links); mtx_unlock(&thr->queue_lock); wakeup(thr); } static void ctl_enqueue_rtr(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_thread *thr; thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads]; mtx_lock(&thr->queue_lock); STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links); mtx_unlock(&thr->queue_lock); wakeup(thr); } static void ctl_enqueue_done(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_thread *thr; thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads]; mtx_lock(&thr->queue_lock); STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links); mtx_unlock(&thr->queue_lock); wakeup(thr); } static void ctl_enqueue_isc(union ctl_io *io) { struct ctl_softc *softc = CTL_SOFTC(io); struct ctl_thread *thr; thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads]; mtx_lock(&thr->queue_lock); STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links); mtx_unlock(&thr->queue_lock); wakeup(thr); } /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl.h =================================================================== --- head/sys/cam/ctl/ctl.h (revision 333445) +++ head/sys/cam/ctl/ctl.h (revision 333446) @@ -1,225 +1,207 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003 Silicon Graphics International Corp. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.h#5 $ * $FreeBSD$ */ /* * Function definitions used both within CTL and potentially in various CTL * clients. * * Author: Ken Merry */ #ifndef _CTL_H_ #define _CTL_H_ #define CTL_RETVAL_COMPLETE 0 #define CTL_RETVAL_QUEUED 1 #define CTL_RETVAL_ALLOCATED 2 #define CTL_RETVAL_ERROR 3 typedef enum { CTL_PORT_NONE = 0x00, CTL_PORT_FC = 0x01, CTL_PORT_SCSI = 0x02, CTL_PORT_IOCTL = 0x04, CTL_PORT_INTERNAL = 0x08, CTL_PORT_ISCSI = 0x10, CTL_PORT_SAS = 0x20, CTL_PORT_UMASS = 0x40, CTL_PORT_ALL = 0xff, CTL_PORT_ISC = 0x100 // FC port for inter-shelf communication } ctl_port_type; struct ctl_port_entry { ctl_port_type port_type; char port_name[64]; int32_t targ_port; int physical_port; int virtual_port; u_int flags; #define CTL_PORT_WWNN_VALID 0x01 #define CTL_PORT_WWPN_VALID 0x02 uint64_t wwnn; uint64_t wwpn; int online; }; struct ctl_modepage_header { uint8_t page_code; uint8_t subpage; uint16_t len_used; uint16_t len_left; }; union ctl_modepage_info { struct ctl_modepage_header header; }; /* * Serial number length, for VPD page 0x80. */ #define CTL_SN_LEN 16 /* * Device ID length, for VPD page 0x83. */ #define CTL_DEVID_LEN 64 #define CTL_DEVID_MIN_LEN 16 /* * WWPN length, for VPD page 0x83. */ #define CTL_WWPN_LEN 8 #define CTL_DRIVER_NAME_LEN 32 /* * Unit attention types. ASC/ASCQ values for these should be placed in * ctl_build_ua. These are also listed in order of reporting priority. * i.e. a poweron UA is reported first, bus reset second, etc. */ typedef enum { CTL_UA_NONE = 0x0000, CTL_UA_POWERON = 0x0001, CTL_UA_BUS_RESET = 0x0002, CTL_UA_TARG_RESET = 0x0004, CTL_UA_I_T_NEXUS_LOSS = 0x0008, CTL_UA_LUN_RESET = 0x0010, CTL_UA_LUN_CHANGE = 0x0020, CTL_UA_MODE_CHANGE = 0x0040, CTL_UA_LOG_CHANGE = 0x0080, CTL_UA_INQ_CHANGE = 0x0100, CTL_UA_RES_PREEMPT = 0x0400, CTL_UA_RES_RELEASE = 0x0800, CTL_UA_REG_PREEMPT = 0x1000, CTL_UA_ASYM_ACC_CHANGE = 0x2000, CTL_UA_CAPACITY_CHANGE = 0x4000, CTL_UA_THIN_PROV_THRES = 0x8000, CTL_UA_MEDIUM_CHANGE = 0x10000, CTL_UA_IE = 0x20000 } ctl_ua_type; #ifdef _KERNEL MALLOC_DECLARE(M_CTL); struct ctl_page_index; #ifdef SYSCTL_DECL /* from sysctl.h */ SYSCTL_DECL(_kern_cam_ctl); #endif struct ctl_lun; struct ctl_port; struct ctl_softc; /* * Put a string into an sbuf, escaping characters that are illegal or not * recommended in XML. Note this doesn't escape everything, just > < and &. */ int ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size); int ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last); int ctl_set_mask(uint32_t *mask, uint32_t bit); int ctl_clear_mask(uint32_t *mask, uint32_t bit); int ctl_is_set(uint32_t *mask, uint32_t bit); int ctl_default_page_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr); int ctl_ie_page_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr); int ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc); int ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc); int ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc); int ctl_config_move_done(union ctl_io *io); void ctl_datamove(union ctl_io *io); void ctl_serseq_done(union ctl_io *io); void ctl_done(union ctl_io *io); void ctl_data_submit_done(union ctl_io *io); void ctl_config_read_done(union ctl_io *io); void ctl_config_write_done(union ctl_io *io); void ctl_portDB_changed(int portnum); int ctl_ioctl_io(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); void ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua); void ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua); void ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua); void ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua); void ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua); void ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx, ctl_ua_type ua_type); uint32_t ctl_decode_lun(uint64_t encoded); uint64_t ctl_encode_lun(uint32_t decoded); void ctl_isc_announce_lun(struct ctl_lun *lun); void ctl_isc_announce_port(struct ctl_port *port); void ctl_isc_announce_iid(struct ctl_port *port, int iid); void ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx, uint8_t page, uint8_t subpage); -/* - * KPI to manipulate LUN/port options - */ - -struct ctl_option { - STAILQ_ENTRY(ctl_option) links; - char *name; - char *value; -}; -typedef STAILQ_HEAD(ctl_options, ctl_option) ctl_options_t; - -struct ctl_be_arg; -void ctl_init_opts(ctl_options_t *opts, int num_args, struct ctl_be_arg *args); -void ctl_update_opts(ctl_options_t *opts, int num_args, - struct ctl_be_arg *args); -void ctl_free_opts(ctl_options_t *opts); -char * ctl_get_opt(ctl_options_t *opts, const char *name); -int ctl_get_opt_number(ctl_options_t *opts, const char *name, uint64_t *num); int ctl_expand_number(const char *buf, uint64_t *num); #endif /* _KERNEL */ #endif /* _CTL_H_ */ /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_backend.c =================================================================== --- head/sys/cam/ctl/ctl_backend.c (revision 333445) +++ head/sys/cam/ctl/ctl_backend.c (revision 333446) @@ -1,233 +1,143 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003 Silicon Graphics International Corp. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend.c#3 $ */ /* * CTL backend driver registration routines * * Author: Ken Merry */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include extern struct ctl_softc *control_softc; int ctl_backend_register(struct ctl_backend_driver *be) { struct ctl_softc *softc = control_softc; struct ctl_backend_driver *be_tmp; int error; /* Sanity check, make sure this isn't a duplicate registration. */ mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(be_tmp, &softc->be_list, links) { if (strcmp(be_tmp->name, be->name) == 0) { mtx_unlock(&softc->ctl_lock); return (-1); } } mtx_unlock(&softc->ctl_lock); #ifdef CS_BE_CONFIG_MOVE_DONE_IS_NOT_USED be->config_move_done = ctl_config_move_done; #endif be->num_luns = 0; /* Call the backend's initialization routine. */ if (be->init != NULL) { if ((error = be->init()) != 0) { printf("%s backend init error: %d\n", be->name, error); return (error); } } mtx_lock(&softc->ctl_lock); STAILQ_INSERT_TAIL(&softc->be_list, be, links); softc->num_backends++; mtx_unlock(&softc->ctl_lock); return (0); } int ctl_backend_deregister(struct ctl_backend_driver *be) { struct ctl_softc *softc = control_softc; int error; /* Call the backend's shutdown routine. */ if (be->shutdown != NULL) { if ((error = be->shutdown()) != 0) { printf("%s backend shutdown error: %d\n", be->name, error); return (error); } } mtx_lock(&softc->ctl_lock); STAILQ_REMOVE(&softc->be_list, be, ctl_backend_driver, links); softc->num_backends--; mtx_unlock(&softc->ctl_lock); return (0); } struct ctl_backend_driver * ctl_backend_find(char *backend_name) { struct ctl_softc *softc = control_softc; struct ctl_backend_driver *be_tmp; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(be_tmp, &softc->be_list, links) { if (strcmp(be_tmp->name, backend_name) == 0) { mtx_unlock(&softc->ctl_lock); return (be_tmp); } } mtx_unlock(&softc->ctl_lock); return (NULL); } -void -ctl_init_opts(ctl_options_t *opts, int num_args, struct ctl_be_arg *args) -{ - struct ctl_option *opt; - int i; - - STAILQ_INIT(opts); - for (i = 0; i < num_args; i++) { - if ((args[i].flags & CTL_BEARG_RD) == 0) - continue; - if ((args[i].flags & CTL_BEARG_ASCII) == 0) - continue; - opt = malloc(sizeof(*opt), M_CTL, M_WAITOK); - opt->name = strdup(args[i].kname, M_CTL); - opt->value = strdup(args[i].kvalue, M_CTL); - STAILQ_INSERT_TAIL(opts, opt, links); - } -} - -void -ctl_update_opts(ctl_options_t *opts, int num_args, struct ctl_be_arg *args) -{ - struct ctl_option *opt; - int i; - - for (i = 0; i < num_args; i++) { - if ((args[i].flags & CTL_BEARG_RD) == 0) - continue; - if ((args[i].flags & CTL_BEARG_ASCII) == 0) - continue; - STAILQ_FOREACH(opt, opts, links) { - if (strcmp(opt->name, args[i].kname) == 0) - break; - } - if (args[i].kvalue != NULL && - ((char *)args[i].kvalue)[0] != 0) { - if (opt) { - free(opt->value, M_CTL); - opt->value = strdup(args[i].kvalue, M_CTL); - } else { - opt = malloc(sizeof(*opt), M_CTL, M_WAITOK); - opt->name = strdup(args[i].kname, M_CTL); - opt->value = strdup(args[i].kvalue, M_CTL); - STAILQ_INSERT_TAIL(opts, opt, links); - } - } else if (opt) { - STAILQ_REMOVE(opts, opt, ctl_option, links); - free(opt->name, M_CTL); - free(opt->value, M_CTL); - free(opt, M_CTL); - } - } -} - -void -ctl_free_opts(ctl_options_t *opts) -{ - struct ctl_option *opt; - - while ((opt = STAILQ_FIRST(opts)) != NULL) { - STAILQ_REMOVE_HEAD(opts, links); - free(opt->name, M_CTL); - free(opt->value, M_CTL); - free(opt, M_CTL); - } -} - -char * -ctl_get_opt(ctl_options_t *opts, const char *name) -{ - struct ctl_option *opt; - - STAILQ_FOREACH(opt, opts, links) { - if (strcmp(opt->name, name) == 0) { - return (opt->value); - } - } - return (NULL); -} - -int -ctl_get_opt_number(ctl_options_t *opts, const char *name, uint64_t *val) -{ - const char *value; - - value = ctl_get_opt(opts, name); - if (value == NULL) - return (-2); - return (ctl_expand_number(value, val)); -} Index: head/sys/cam/ctl/ctl_backend.h =================================================================== --- head/sys/cam/ctl/ctl_backend.h (revision 333445) +++ head/sys/cam/ctl/ctl_backend.h (revision 333446) @@ -1,271 +1,272 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003 Silicon Graphics International Corp. * Copyright (c) 2014-2017 Alexander Motin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend.h#2 $ * $FreeBSD$ */ /* * CTL backend driver definitions * * Author: Ken Merry */ #ifndef _CTL_BACKEND_H_ #define _CTL_BACKEND_H_ #include +#include typedef enum { CTL_LUN_SERSEQ_OFF, CTL_LUN_SERSEQ_READ, CTL_LUN_SERSEQ_ON } ctl_lun_serseq; #ifdef _KERNEL #define CTL_BACKEND_DECLARE(name, driver) \ static int name ## _modevent(module_t mod, int type, void *data) \ { \ switch (type) { \ case MOD_LOAD: \ return (ctl_backend_register( \ (struct ctl_backend_driver *)data)); \ break; \ case MOD_UNLOAD: \ return (ctl_backend_deregister( \ (struct ctl_backend_driver *)data)); \ break; \ default: \ return EOPNOTSUPP; \ } \ return 0; \ } \ static moduledata_t name ## _mod = { \ #name, \ name ## _modevent, \ (void *)&driver \ }; \ DECLARE_MODULE(name, name ## _mod, SI_SUB_CONFIGURE, SI_ORDER_FOURTH); \ MODULE_DEPEND(name, ctl, 1, 1, 1); \ MODULE_DEPEND(name, cam, 1, 1, 1) typedef enum { CTL_LUN_CONFIG_OK, CTL_LUN_CONFIG_FAILURE } ctl_lun_config_status; typedef void (*be_callback_t)(void *be_lun); typedef void (*be_lun_config_t)(void *be_lun, ctl_lun_config_status status); /* * The lun_type field is the SCSI device type of this particular LUN. In * general, this should be T_DIRECT, although backends will want to create * a processor LUN, typically at LUN 0. See scsi_all.h for the defines for * the various SCSI device types. * * The flags are described above. * * The be_lun field is the backend driver's own context that will get * passsed back so that it can tell which LUN CTL is referencing. * * maxlba is the maximum accessible LBA on the LUN. Note that this is * different from the capacity of the array. capacity = maxlba + 1 * * blocksize is the size, in bytes, of each LBA on the LUN. In general * this should be 512. In theory CTL should be able to handle other block * sizes. Host application software may not deal with it very well, though. * * pblockexp is the log2() of number of LBAs on the LUN per physical sector. * * pblockoff is the lowest LBA on the LUN aligned to physical sector. * * ublockexp is the log2() of number of LBAs on the LUN per UNMAP block. * * ublockoff is the lowest LBA on the LUN aligned to UNMAP block. * * atomicblock is the number of blocks that can be written atomically. * * opttxferlen is the number of blocks that can be written in one operation. * * req_lun_id is the requested LUN ID. CTL only pays attention to this * field if the CTL_LUN_FLAG_ID_REQ flag is set. If the requested LUN ID is * not available, the LUN addition will fail. If a particular LUN ID isn't * requested, the first available LUN ID will be allocated. * * serial_num is the device serial number returned in the SCSI INQUIRY VPD * page 0x80. This should be a unique, per-shelf value. The data inside * this field should be ASCII only, left aligned, and any unused space * should be padded out with ASCII spaces. This field should NOT be NULL * terminated. * * device_id is the T10 device identifier returned in the SCSI INQUIRY VPD * page 0x83. This should be a unique, per-LUN value. The data inside * this field should be ASCII only, left aligned, and any unused space * should be padded with ASCII spaces. This field should NOT be NULL * terminated. * * The lun_shutdown() method is the callback for the ctl_invalidate_lun() * call. It is called when all outstanding I/O for that LUN has been * completed and CTL has deleted the resources for that LUN. When the CTL * backend gets this call, it can safely free its per-LUN resources. * * The lun_config_status() method is the callback for the ctl_add_lun() * call. It is called when the LUN is successfully added, or when LUN * addition fails. If the LUN is successfully added, the backend may call * the ctl_enable_lun() method to enable the LUN. * * The be field is a pointer to the ctl_backend_driver structure, which * contains the backend methods to be called by CTL. * * The ctl_lun field is for CTL internal use only, and should not be used * by the backend. * * The links field is for CTL internal use only, and should not be used by * the backend. */ struct ctl_be_lun { uint8_t lun_type; /* passed to CTL */ ctl_backend_lun_flags flags; /* passed to CTL */ ctl_lun_serseq serseq; /* passed to CTL */ void *be_lun; /* passed to CTL */ uint64_t maxlba; /* passed to CTL */ uint32_t blocksize; /* passed to CTL */ uint16_t pblockexp; /* passed to CTL */ uint16_t pblockoff; /* passed to CTL */ uint16_t ublockexp; /* passed to CTL */ uint16_t ublockoff; /* passed to CTL */ uint32_t atomicblock; /* passed to CTL */ uint32_t opttxferlen; /* passed to CTL */ uint32_t req_lun_id; /* passed to CTL */ uint32_t lun_id; /* returned from CTL */ uint8_t serial_num[CTL_SN_LEN]; /* passed to CTL */ uint8_t device_id[CTL_DEVID_LEN];/* passed to CTL */ be_callback_t lun_shutdown; /* passed to CTL */ be_lun_config_t lun_config_status; /* passed to CTL */ struct ctl_backend_driver *be; /* passed to CTL */ void *ctl_lun; /* used by CTL */ - ctl_options_t options; /* passed to CTL */ + nvlist_t *options; /* passed to CTL */ STAILQ_ENTRY(ctl_be_lun) links; /* used by CTL */ }; typedef enum { CTL_BE_FLAG_NONE = 0x00, /* no flags */ CTL_BE_FLAG_HAS_CONFIG = 0x01, /* can do config reads, writes */ } ctl_backend_flags; typedef int (*be_init_t)(void); typedef int (*be_shutdown_t)(void); typedef int (*be_func_t)(union ctl_io *io); typedef void (*be_vfunc_t)(union ctl_io *io); typedef int (*be_ioctl_t)(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); typedef int (*be_luninfo_t)(void *be_lun, struct sbuf *sb); typedef uint64_t (*be_lunattr_t)(void *be_lun, const char *attrname); struct ctl_backend_driver { char name[CTL_BE_NAME_LEN]; /* passed to CTL */ ctl_backend_flags flags; /* passed to CTL */ be_init_t init; /* passed to CTL */ be_shutdown_t shutdown; /* passed to CTL */ be_func_t data_submit; /* passed to CTL */ be_func_t data_move_done; /* passed to CTL */ be_func_t config_read; /* passed to CTL */ be_func_t config_write; /* passed to CTL */ be_ioctl_t ioctl; /* passed to CTL */ be_luninfo_t lun_info; /* passed to CTL */ be_lunattr_t lun_attr; /* passed to CTL */ #ifdef CS_BE_CONFIG_MOVE_DONE_IS_NOT_USED be_func_t config_move_done; /* passed to backend */ #endif #if 0 be_vfunc_t config_write_done; /* passed to backend */ #endif u_int num_luns; /* used by CTL */ STAILQ_ENTRY(ctl_backend_driver) links; /* used by CTL */ }; int ctl_backend_register(struct ctl_backend_driver *be); int ctl_backend_deregister(struct ctl_backend_driver *be); struct ctl_backend_driver *ctl_backend_find(char *backend_name); /* * To add a LUN, first call ctl_add_lun(). You will get the lun_config_status() * callback when the LUN addition has either succeeded or failed. * * Once you get that callback, you can then call ctl_enable_lun() to enable * the LUN. */ int ctl_add_lun(struct ctl_be_lun *be_lun); int ctl_enable_lun(struct ctl_be_lun *be_lun); /* * To delete a LUN, first call ctl_disable_lun(), then * ctl_invalidate_lun(). You will get the lun_shutdown() callback when all * I/O to the LUN has completed and the LUN has been deleted. */ int ctl_disable_lun(struct ctl_be_lun *be_lun); int ctl_invalidate_lun(struct ctl_be_lun *be_lun); /* * To start a LUN (transition from powered off to powered on state) call * ctl_start_lun(). To stop a LUN (transition from powered on to powered * off state) call ctl_stop_lun(). */ int ctl_start_lun(struct ctl_be_lun *be_lun); int ctl_stop_lun(struct ctl_be_lun *be_lun); /* * Methods to notify about media and tray status changes. */ int ctl_lun_no_media(struct ctl_be_lun *be_lun); int ctl_lun_has_media(struct ctl_be_lun *be_lun); int ctl_lun_ejected(struct ctl_be_lun *be_lun); /* * Called on LUN HA role change. */ int ctl_lun_primary(struct ctl_be_lun *be_lun); int ctl_lun_secondary(struct ctl_be_lun *be_lun); /* * Let the backend notify the initiators about changes. */ void ctl_lun_capacity_changed(struct ctl_be_lun *be_lun); #endif /* _KERNEL */ #endif /* _CTL_BACKEND_H_ */ /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_backend_block.c =================================================================== --- head/sys/cam/ctl/ctl_backend_block.c (revision 333445) +++ head/sys/cam/ctl/ctl_backend_block.c (revision 333446) @@ -1,2898 +1,2901 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003 Silicon Graphics International Corp. * Copyright (c) 2009-2011 Spectra Logic Corporation * Copyright (c) 2012 The FreeBSD Foundation * Copyright (c) 2014-2015 Alexander Motin * All rights reserved. * * Portions of this software were developed by Edward Tomasz Napierala * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_block.c#5 $ */ /* * CAM Target Layer driver backend for block devices. * * Author: Ken Merry */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include #include #include #include #include #include #include /* * The idea here is that we'll allocate enough S/G space to hold a 1MB * I/O. If we get an I/O larger than that, we'll split it. */ #define CTLBLK_HALF_IO_SIZE (512 * 1024) #define CTLBLK_MAX_IO_SIZE (CTLBLK_HALF_IO_SIZE * 2) #define CTLBLK_MAX_SEG MAXPHYS #define CTLBLK_HALF_SEGS MAX(CTLBLK_HALF_IO_SIZE / CTLBLK_MAX_SEG, 1) #define CTLBLK_MAX_SEGS (CTLBLK_HALF_SEGS * 2) #ifdef CTLBLK_DEBUG #define DPRINTF(fmt, args...) \ printf("cbb(%s:%d): " fmt, __FUNCTION__, __LINE__, ##args) #else #define DPRINTF(fmt, args...) do {} while(0) #endif #define PRIV(io) \ ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) #define ARGS(io) \ ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) SDT_PROVIDER_DEFINE(cbb); typedef enum { CTL_BE_BLOCK_LUN_UNCONFIGURED = 0x01, CTL_BE_BLOCK_LUN_CONFIG_ERR = 0x02, CTL_BE_BLOCK_LUN_WAITING = 0x04, } ctl_be_block_lun_flags; typedef enum { CTL_BE_BLOCK_NONE, CTL_BE_BLOCK_DEV, CTL_BE_BLOCK_FILE } ctl_be_block_type; struct ctl_be_block_filedata { struct ucred *cred; }; union ctl_be_block_bedata { struct ctl_be_block_filedata file; }; struct ctl_be_block_io; struct ctl_be_block_lun; typedef void (*cbb_dispatch_t)(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio); typedef uint64_t (*cbb_getattr_t)(struct ctl_be_block_lun *be_lun, const char *attrname); /* * Backend LUN structure. There is a 1:1 mapping between a block device * and a backend block LUN, and between a backend block LUN and a CTL LUN. */ struct ctl_be_block_lun { struct ctl_lun_create_params params; char lunname[32]; char *dev_path; ctl_be_block_type dev_type; struct vnode *vn; union ctl_be_block_bedata backend; cbb_dispatch_t dispatch; cbb_dispatch_t lun_flush; cbb_dispatch_t unmap; cbb_dispatch_t get_lba_status; cbb_getattr_t getattr; uma_zone_t lun_zone; uint64_t size_blocks; uint64_t size_bytes; struct ctl_be_block_softc *softc; struct devstat *disk_stats; ctl_be_block_lun_flags flags; STAILQ_ENTRY(ctl_be_block_lun) links; struct ctl_be_lun cbe_lun; struct taskqueue *io_taskqueue; struct task io_task; int num_threads; STAILQ_HEAD(, ctl_io_hdr) input_queue; STAILQ_HEAD(, ctl_io_hdr) config_read_queue; STAILQ_HEAD(, ctl_io_hdr) config_write_queue; STAILQ_HEAD(, ctl_io_hdr) datamove_queue; struct mtx_padalign io_lock; struct mtx_padalign queue_lock; }; /* * Overall softc structure for the block backend module. */ struct ctl_be_block_softc { struct mtx lock; uma_zone_t beio_zone; int num_luns; STAILQ_HEAD(, ctl_be_block_lun) lun_list; }; static struct ctl_be_block_softc backend_block_softc; /* * Per-I/O information. */ struct ctl_be_block_io { union ctl_io *io; struct ctl_sg_entry sg_segs[CTLBLK_MAX_SEGS]; struct iovec xiovecs[CTLBLK_MAX_SEGS]; int bio_cmd; int num_segs; int num_bios_sent; int num_bios_done; int send_complete; int first_error; uint64_t first_error_offset; struct bintime ds_t0; devstat_tag_type ds_tag_type; devstat_trans_flags ds_trans_type; uint64_t io_len; uint64_t io_offset; int io_arg; struct ctl_be_block_softc *softc; struct ctl_be_block_lun *lun; void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ }; extern struct ctl_softc *control_softc; static int cbb_num_threads = 14; SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, block, CTLFLAG_RD, 0, "CAM Target Layer Block Backend"); SYSCTL_INT(_kern_cam_ctl_block, OID_AUTO, num_threads, CTLFLAG_RWTUN, &cbb_num_threads, 0, "Number of threads per backing file"); static struct ctl_be_block_io *ctl_alloc_beio(struct ctl_be_block_softc *softc); static void ctl_free_beio(struct ctl_be_block_io *beio); static void ctl_complete_beio(struct ctl_be_block_io *beio); static int ctl_be_block_move_done(union ctl_io *io); static void ctl_be_block_biodone(struct bio *bio); static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio); static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio); static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio); static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname); static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio); static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio); static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio); static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname); static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, union ctl_io *io); static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, union ctl_io *io); static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, union ctl_io *io); static void ctl_be_block_worker(void *context, int pending); static int ctl_be_block_submit(union ctl_io *io); static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req); static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req); static int ctl_be_block_close(struct ctl_be_block_lun *be_lun); static int ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req); static int ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req); static int ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req); static int ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req); static void ctl_be_block_lun_shutdown(void *be_lun); static void ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status); static int ctl_be_block_config_write(union ctl_io *io); static int ctl_be_block_config_read(union ctl_io *io); static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb); static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname); static int ctl_be_block_init(void); static int ctl_be_block_shutdown(void); static struct ctl_backend_driver ctl_be_block_driver = { .name = "block", .flags = CTL_BE_FLAG_HAS_CONFIG, .init = ctl_be_block_init, .shutdown = ctl_be_block_shutdown, .data_submit = ctl_be_block_submit, .data_move_done = ctl_be_block_move_done, .config_read = ctl_be_block_config_read, .config_write = ctl_be_block_config_write, .ioctl = ctl_be_block_ioctl, .lun_info = ctl_be_block_lun_info, .lun_attr = ctl_be_block_lun_attr }; MALLOC_DEFINE(M_CTLBLK, "ctlblk", "Memory used for CTL block backend"); CTL_BACKEND_DECLARE(cbb, ctl_be_block_driver); static struct ctl_be_block_io * ctl_alloc_beio(struct ctl_be_block_softc *softc) { struct ctl_be_block_io *beio; beio = uma_zalloc(softc->beio_zone, M_WAITOK | M_ZERO); beio->softc = softc; return (beio); } static void ctl_free_beio(struct ctl_be_block_io *beio) { int duplicate_free; int i; duplicate_free = 0; for (i = 0; i < beio->num_segs; i++) { if (beio->sg_segs[i].addr == NULL) duplicate_free++; uma_zfree(beio->lun->lun_zone, beio->sg_segs[i].addr); beio->sg_segs[i].addr = NULL; /* For compare we had two equal S/G lists. */ if (ARGS(beio->io)->flags & CTL_LLF_COMPARE) { uma_zfree(beio->lun->lun_zone, beio->sg_segs[i + CTLBLK_HALF_SEGS].addr); beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = NULL; } } if (duplicate_free > 0) { printf("%s: %d duplicate frees out of %d segments\n", __func__, duplicate_free, beio->num_segs); } uma_zfree(beio->softc->beio_zone, beio); } static void ctl_complete_beio(struct ctl_be_block_io *beio) { union ctl_io *io = beio->io; if (beio->beio_cont != NULL) { beio->beio_cont(beio); } else { ctl_free_beio(beio); ctl_data_submit_done(io); } } static size_t cmp(uint8_t *a, uint8_t *b, size_t size) { size_t i; for (i = 0; i < size; i++) { if (a[i] != b[i]) break; } return (i); } static void ctl_be_block_compare(union ctl_io *io) { struct ctl_be_block_io *beio; uint64_t off, res; int i; uint8_t info[8]; beio = (struct ctl_be_block_io *)PRIV(io)->ptr; off = 0; for (i = 0; i < beio->num_segs; i++) { res = cmp(beio->sg_segs[i].addr, beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, beio->sg_segs[i].len); off += res; if (res < beio->sg_segs[i].len) break; } if (i < beio->num_segs) { scsi_u64to8b(off, info); ctl_set_sense(&io->scsiio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_MISCOMPARE, /*asc*/ 0x1D, /*ascq*/ 0x00, /*type*/ SSD_ELEM_INFO, /*size*/ sizeof(info), /*data*/ &info, /*type*/ SSD_ELEM_NONE); } else ctl_set_success(&io->scsiio); } static int ctl_be_block_move_done(union ctl_io *io) { struct ctl_be_block_io *beio; struct ctl_be_block_lun *be_lun; struct ctl_lba_len_flags *lbalen; #ifdef CTL_TIME_IO struct bintime cur_bt; #endif beio = (struct ctl_be_block_io *)PRIV(io)->ptr; be_lun = beio->lun; DPRINTF("entered\n"); #ifdef CTL_TIME_IO getbinuptime(&cur_bt); bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); bintime_add(&io->io_hdr.dma_bt, &cur_bt); #endif io->io_hdr.num_dmas++; io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; /* * We set status at this point for read commands, and write * commands with errors. */ if (io->io_hdr.flags & CTL_FLAG_ABORT) { ; } else if ((io->io_hdr.port_status != 0) && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ io->io_hdr.port_status); } else if (io->scsiio.kern_data_resid != 0 && (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { ctl_set_invalid_field_ciu(&io->scsiio); } else if ((io->io_hdr.port_status == 0) && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { lbalen = ARGS(beio->io); if (lbalen->flags & CTL_LLF_READ) { ctl_set_success(&io->scsiio); } else if (lbalen->flags & CTL_LLF_COMPARE) { /* We have two data blocks ready for comparison. */ ctl_be_block_compare(io); } } /* * If this is a read, or a write with errors, it is done. */ if ((beio->bio_cmd == BIO_READ) || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0) || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE)) { ctl_complete_beio(beio); return (0); } /* * At this point, we have a write and the DMA completed * successfully. We now have to queue it to the task queue to * execute the backend I/O. That is because we do blocking * memory allocations, and in the file backing case, blocking I/O. * This move done routine is generally called in the SIM's * interrupt context, and therefore we cannot block. */ mtx_lock(&be_lun->queue_lock); STAILQ_INSERT_TAIL(&be_lun->datamove_queue, &io->io_hdr, links); mtx_unlock(&be_lun->queue_lock); taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); return (0); } static void ctl_be_block_biodone(struct bio *bio) { struct ctl_be_block_io *beio; struct ctl_be_block_lun *be_lun; union ctl_io *io; int error; beio = bio->bio_caller1; be_lun = beio->lun; io = beio->io; DPRINTF("entered\n"); error = bio->bio_error; mtx_lock(&be_lun->io_lock); if (error != 0 && (beio->first_error == 0 || bio->bio_offset < beio->first_error_offset)) { beio->first_error = error; beio->first_error_offset = bio->bio_offset; } beio->num_bios_done++; /* * XXX KDM will this cause WITNESS to complain? Holding a lock * during the free might cause it to complain. */ g_destroy_bio(bio); /* * If the send complete bit isn't set, or we aren't the last I/O to * complete, then we're done. */ if ((beio->send_complete == 0) || (beio->num_bios_done < beio->num_bios_sent)) { mtx_unlock(&be_lun->io_lock); return; } /* * At this point, we've verified that we are the last I/O to * complete, so it's safe to drop the lock. */ devstat_end_transaction(beio->lun->disk_stats, beio->io_len, beio->ds_tag_type, beio->ds_trans_type, /*now*/ NULL, /*then*/&beio->ds_t0); mtx_unlock(&be_lun->io_lock); /* * If there are any errors from the backing device, we fail the * entire I/O with a medium error. */ error = beio->first_error; if (error != 0) { if (error == EOPNOTSUPP) { ctl_set_invalid_opcode(&io->scsiio); } else if (error == ENOSPC || error == EDQUOT) { ctl_set_space_alloc_fail(&io->scsiio); } else if (error == EROFS || error == EACCES) { ctl_set_hw_write_protected(&io->scsiio); } else if (beio->bio_cmd == BIO_FLUSH) { /* XXX KDM is there is a better error here? */ ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ 0xbad2); } else { ctl_set_medium_error(&io->scsiio, beio->bio_cmd == BIO_READ); } ctl_complete_beio(beio); return; } /* * If this is a write, a flush, a delete or verify, we're all done. * If this is a read, we can now send the data to the user. */ if ((beio->bio_cmd == BIO_WRITE) || (beio->bio_cmd == BIO_FLUSH) || (beio->bio_cmd == BIO_DELETE) || (ARGS(io)->flags & CTL_LLF_VERIFY)) { ctl_set_success(&io->scsiio); ctl_complete_beio(beio); } else { if ((ARGS(io)->flags & CTL_LLF_READ) && beio->beio_cont == NULL) { ctl_set_success(&io->scsiio); ctl_serseq_done(io); } #ifdef CTL_TIME_IO getbinuptime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } } static void ctl_be_block_flush_file(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { union ctl_io *io = beio->io; struct mount *mountpoint; int error, lock_flags; DPRINTF("entered\n"); binuptime(&beio->ds_t0); mtx_lock(&be_lun->io_lock); devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); mtx_unlock(&be_lun->io_lock); (void) vn_start_write(be_lun->vn, &mountpoint, V_WAIT); if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount))) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_lock(be_lun->vn, lock_flags | LK_RETRY); error = VOP_FSYNC(be_lun->vn, beio->io_arg ? MNT_NOWAIT : MNT_WAIT, curthread); VOP_UNLOCK(be_lun->vn, 0); vn_finished_write(mountpoint); mtx_lock(&be_lun->io_lock); devstat_end_transaction(beio->lun->disk_stats, beio->io_len, beio->ds_tag_type, beio->ds_trans_type, /*now*/ NULL, /*then*/&beio->ds_t0); mtx_unlock(&be_lun->io_lock); if (error == 0) ctl_set_success(&io->scsiio); else { /* XXX KDM is there is a better error here? */ ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ 0xbad1); } ctl_complete_beio(beio); } SDT_PROBE_DEFINE1(cbb, , read, file_start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, , write, file_start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, , read, file_done,"uint64_t"); SDT_PROBE_DEFINE1(cbb, , write, file_done, "uint64_t"); static void ctl_be_block_dispatch_file(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { struct ctl_be_block_filedata *file_data; union ctl_io *io; struct uio xuio; struct iovec *xiovec; size_t s; int error, flags, i; DPRINTF("entered\n"); file_data = &be_lun->backend.file; io = beio->io; flags = 0; if (ARGS(io)->flags & CTL_LLF_DPO) flags |= IO_DIRECT; if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) flags |= IO_SYNC; bzero(&xuio, sizeof(xuio)); if (beio->bio_cmd == BIO_READ) { SDT_PROBE0(cbb, , read, file_start); xuio.uio_rw = UIO_READ; } else { SDT_PROBE0(cbb, , write, file_start); xuio.uio_rw = UIO_WRITE; } xuio.uio_offset = beio->io_offset; xuio.uio_resid = beio->io_len; xuio.uio_segflg = UIO_SYSSPACE; xuio.uio_iov = beio->xiovecs; xuio.uio_iovcnt = beio->num_segs; xuio.uio_td = curthread; for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { xiovec->iov_base = beio->sg_segs[i].addr; xiovec->iov_len = beio->sg_segs[i].len; } binuptime(&beio->ds_t0); mtx_lock(&be_lun->io_lock); devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); mtx_unlock(&be_lun->io_lock); if (beio->bio_cmd == BIO_READ) { vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); /* * UFS pays attention to IO_DIRECT for reads. If the * DIRECTIO option is configured into the kernel, it calls * ffs_rawread(). But that only works for single-segment * uios with user space addresses. In our case, with a * kernel uio, it still reads into the buffer cache, but it * will just try to release the buffer from the cache later * on in ffs_read(). * * ZFS does not pay attention to IO_DIRECT for reads. * * UFS does not pay attention to IO_SYNC for reads. * * ZFS pays attention to IO_SYNC (which translates into the * Solaris define FRSYNC for zfs_read()) for reads. It * attempts to sync the file before reading. */ error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); VOP_UNLOCK(be_lun->vn, 0); SDT_PROBE0(cbb, , read, file_done); if (error == 0 && xuio.uio_resid > 0) { /* * If we red less then requested (EOF), then * we should clean the rest of the buffer. */ s = beio->io_len - xuio.uio_resid; for (i = 0; i < beio->num_segs; i++) { if (s >= beio->sg_segs[i].len) { s -= beio->sg_segs[i].len; continue; } bzero((uint8_t *)beio->sg_segs[i].addr + s, beio->sg_segs[i].len - s); s = 0; } } } else { struct mount *mountpoint; int lock_flags; (void)vn_start_write(be_lun->vn, &mountpoint, V_WAIT); if (MNT_SHARED_WRITES(mountpoint) || ((mountpoint == NULL) && MNT_SHARED_WRITES(be_lun->vn->v_mount))) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_lock(be_lun->vn, lock_flags | LK_RETRY); /* * UFS pays attention to IO_DIRECT for writes. The write * is done asynchronously. (Normally the write would just * get put into cache. * * UFS pays attention to IO_SYNC for writes. It will * attempt to write the buffer out synchronously if that * flag is set. * * ZFS does not pay attention to IO_DIRECT for writes. * * ZFS pays attention to IO_SYNC (a.k.a. FSYNC or FRSYNC) * for writes. It will flush the transaction from the * cache before returning. */ error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); VOP_UNLOCK(be_lun->vn, 0); vn_finished_write(mountpoint); SDT_PROBE0(cbb, , write, file_done); } mtx_lock(&be_lun->io_lock); devstat_end_transaction(beio->lun->disk_stats, beio->io_len, beio->ds_tag_type, beio->ds_trans_type, /*now*/ NULL, /*then*/&beio->ds_t0); mtx_unlock(&be_lun->io_lock); /* * If we got an error, set the sense data to "MEDIUM ERROR" and * return the I/O to the user. */ if (error != 0) { if (error == ENOSPC || error == EDQUOT) { ctl_set_space_alloc_fail(&io->scsiio); } else if (error == EROFS || error == EACCES) { ctl_set_hw_write_protected(&io->scsiio); } else { ctl_set_medium_error(&io->scsiio, beio->bio_cmd == BIO_READ); } ctl_complete_beio(beio); return; } /* * If this is a write or a verify, we're all done. * If this is a read, we can now send the data to the user. */ if ((beio->bio_cmd == BIO_WRITE) || (ARGS(io)->flags & CTL_LLF_VERIFY)) { ctl_set_success(&io->scsiio); ctl_complete_beio(beio); } else { if ((ARGS(io)->flags & CTL_LLF_READ) && beio->beio_cont == NULL) { ctl_set_success(&io->scsiio); ctl_serseq_done(io); } #ifdef CTL_TIME_IO getbinuptime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } } static void ctl_be_block_gls_file(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { union ctl_io *io = beio->io; struct ctl_lba_len_flags *lbalen = ARGS(io); struct scsi_get_lba_status_data *data; off_t roff, off; int error, status; DPRINTF("entered\n"); off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); error = VOP_IOCTL(be_lun->vn, FIOSEEKHOLE, &off, 0, curthread->td_ucred, curthread); if (error == 0 && off > roff) status = 0; /* mapped up to off */ else { error = VOP_IOCTL(be_lun->vn, FIOSEEKDATA, &off, 0, curthread->td_ucred, curthread); if (error == 0 && off > roff) status = 1; /* deallocated up to off */ else { status = 0; /* unknown up to the end */ off = be_lun->size_bytes; } } VOP_UNLOCK(be_lun->vn, 0); data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; scsi_u64to8b(lbalen->lba, data->descr[0].addr); scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - lbalen->lba), data->descr[0].length); data->descr[0].status = status; ctl_complete_beio(beio); } static uint64_t ctl_be_block_getattr_file(struct ctl_be_block_lun *be_lun, const char *attrname) { struct vattr vattr; struct statfs statfs; uint64_t val; int error; val = UINT64_MAX; if (be_lun->vn == NULL) return (val); vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); if (strcmp(attrname, "blocksused") == 0) { error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); if (error == 0) val = vattr.va_bytes / be_lun->cbe_lun.blocksize; } if (strcmp(attrname, "blocksavail") == 0 && (be_lun->vn->v_iflag & VI_DOOMED) == 0) { error = VFS_STATFS(be_lun->vn->v_mount, &statfs); if (error == 0) val = statfs.f_bavail * statfs.f_bsize / be_lun->cbe_lun.blocksize; } VOP_UNLOCK(be_lun->vn, 0); return (val); } static void ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { union ctl_io *io; struct cdevsw *csw; struct cdev *dev; struct uio xuio; struct iovec *xiovec; int error, flags, i, ref; DPRINTF("entered\n"); io = beio->io; flags = 0; if (ARGS(io)->flags & CTL_LLF_DPO) flags |= IO_DIRECT; if (beio->bio_cmd == BIO_WRITE && ARGS(io)->flags & CTL_LLF_FUA) flags |= IO_SYNC; bzero(&xuio, sizeof(xuio)); if (beio->bio_cmd == BIO_READ) { SDT_PROBE0(cbb, , read, file_start); xuio.uio_rw = UIO_READ; } else { SDT_PROBE0(cbb, , write, file_start); xuio.uio_rw = UIO_WRITE; } xuio.uio_offset = beio->io_offset; xuio.uio_resid = beio->io_len; xuio.uio_segflg = UIO_SYSSPACE; xuio.uio_iov = beio->xiovecs; xuio.uio_iovcnt = beio->num_segs; xuio.uio_td = curthread; for (i = 0, xiovec = xuio.uio_iov; i < xuio.uio_iovcnt; i++, xiovec++) { xiovec->iov_base = beio->sg_segs[i].addr; xiovec->iov_len = beio->sg_segs[i].len; } binuptime(&beio->ds_t0); mtx_lock(&be_lun->io_lock); devstat_start_transaction(beio->lun->disk_stats, &beio->ds_t0); mtx_unlock(&be_lun->io_lock); csw = devvn_refthread(be_lun->vn, &dev, &ref); if (csw) { if (beio->bio_cmd == BIO_READ) error = csw->d_read(dev, &xuio, flags); else error = csw->d_write(dev, &xuio, flags); dev_relthread(dev, ref); } else error = ENXIO; if (beio->bio_cmd == BIO_READ) SDT_PROBE0(cbb, , read, file_done); else SDT_PROBE0(cbb, , write, file_done); mtx_lock(&be_lun->io_lock); devstat_end_transaction(beio->lun->disk_stats, beio->io_len, beio->ds_tag_type, beio->ds_trans_type, /*now*/ NULL, /*then*/&beio->ds_t0); mtx_unlock(&be_lun->io_lock); /* * If we got an error, set the sense data to "MEDIUM ERROR" and * return the I/O to the user. */ if (error != 0) { if (error == ENOSPC || error == EDQUOT) { ctl_set_space_alloc_fail(&io->scsiio); } else if (error == EROFS || error == EACCES) { ctl_set_hw_write_protected(&io->scsiio); } else { ctl_set_medium_error(&io->scsiio, beio->bio_cmd == BIO_READ); } ctl_complete_beio(beio); return; } /* * If this is a write or a verify, we're all done. * If this is a read, we can now send the data to the user. */ if ((beio->bio_cmd == BIO_WRITE) || (ARGS(io)->flags & CTL_LLF_VERIFY)) { ctl_set_success(&io->scsiio); ctl_complete_beio(beio); } else { if ((ARGS(io)->flags & CTL_LLF_READ) && beio->beio_cont == NULL) { ctl_set_success(&io->scsiio); ctl_serseq_done(io); } #ifdef CTL_TIME_IO getbinuptime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } } static void ctl_be_block_gls_zvol(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { union ctl_io *io = beio->io; struct cdevsw *csw; struct cdev *dev; struct ctl_lba_len_flags *lbalen = ARGS(io); struct scsi_get_lba_status_data *data; off_t roff, off; int error, ref, status; DPRINTF("entered\n"); csw = devvn_refthread(be_lun->vn, &dev, &ref); if (csw == NULL) { status = 0; /* unknown up to the end */ off = be_lun->size_bytes; goto done; } off = roff = ((off_t)lbalen->lba) * be_lun->cbe_lun.blocksize; error = csw->d_ioctl(dev, FIOSEEKHOLE, (caddr_t)&off, FREAD, curthread); if (error == 0 && off > roff) status = 0; /* mapped up to off */ else { error = csw->d_ioctl(dev, FIOSEEKDATA, (caddr_t)&off, FREAD, curthread); if (error == 0 && off > roff) status = 1; /* deallocated up to off */ else { status = 0; /* unknown up to the end */ off = be_lun->size_bytes; } } dev_relthread(dev, ref); done: data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; scsi_u64to8b(lbalen->lba, data->descr[0].addr); scsi_ulto4b(MIN(UINT32_MAX, off / be_lun->cbe_lun.blocksize - lbalen->lba), data->descr[0].length); data->descr[0].status = status; ctl_complete_beio(beio); } static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { struct bio *bio; struct cdevsw *csw; struct cdev *dev; int ref; DPRINTF("entered\n"); /* This can't fail, it's a blocking allocation. */ bio = g_alloc_bio(); bio->bio_cmd = BIO_FLUSH; bio->bio_offset = 0; bio->bio_data = 0; bio->bio_done = ctl_be_block_biodone; bio->bio_caller1 = beio; bio->bio_pblkno = 0; /* * We don't need to acquire the LUN lock here, because we are only * sending one bio, and so there is no other context to synchronize * with. */ beio->num_bios_sent = 1; beio->send_complete = 1; binuptime(&beio->ds_t0); mtx_lock(&be_lun->io_lock); devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); mtx_unlock(&be_lun->io_lock); csw = devvn_refthread(be_lun->vn, &dev, &ref); if (csw) { bio->bio_dev = dev; csw->d_strategy(bio); dev_relthread(dev, ref); } else { bio->bio_error = ENXIO; ctl_be_block_biodone(bio); } } static void ctl_be_block_unmap_dev_range(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio, uint64_t off, uint64_t len, int last) { struct bio *bio; uint64_t maxlen; struct cdevsw *csw; struct cdev *dev; int ref; csw = devvn_refthread(be_lun->vn, &dev, &ref); maxlen = LONG_MAX - (LONG_MAX % be_lun->cbe_lun.blocksize); while (len > 0) { bio = g_alloc_bio(); bio->bio_cmd = BIO_DELETE; bio->bio_dev = dev; bio->bio_offset = off; bio->bio_length = MIN(len, maxlen); bio->bio_data = 0; bio->bio_done = ctl_be_block_biodone; bio->bio_caller1 = beio; bio->bio_pblkno = off / be_lun->cbe_lun.blocksize; off += bio->bio_length; len -= bio->bio_length; mtx_lock(&be_lun->io_lock); beio->num_bios_sent++; if (last && len == 0) beio->send_complete = 1; mtx_unlock(&be_lun->io_lock); if (csw) { csw->d_strategy(bio); } else { bio->bio_error = ENXIO; ctl_be_block_biodone(bio); } } if (csw) dev_relthread(dev, ref); } static void ctl_be_block_unmap_dev(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { union ctl_io *io; struct ctl_ptr_len_flags *ptrlen; struct scsi_unmap_desc *buf, *end; uint64_t len; io = beio->io; DPRINTF("entered\n"); binuptime(&beio->ds_t0); mtx_lock(&be_lun->io_lock); devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); mtx_unlock(&be_lun->io_lock); if (beio->io_offset == -1) { beio->io_len = 0; ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; buf = (struct scsi_unmap_desc *)ptrlen->ptr; end = buf + ptrlen->len / sizeof(*buf); for (; buf < end; buf++) { len = (uint64_t)scsi_4btoul(buf->length) * be_lun->cbe_lun.blocksize; beio->io_len += len; ctl_be_block_unmap_dev_range(be_lun, beio, scsi_8btou64(buf->lba) * be_lun->cbe_lun.blocksize, len, (end - buf < 2) ? TRUE : FALSE); } } else ctl_be_block_unmap_dev_range(be_lun, beio, beio->io_offset, beio->io_len, TRUE); } static void ctl_be_block_dispatch_dev(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue); struct bio *bio; struct cdevsw *csw; struct cdev *dev; off_t cur_offset; int i, max_iosize, ref; DPRINTF("entered\n"); csw = devvn_refthread(be_lun->vn, &dev, &ref); /* * We have to limit our I/O size to the maximum supported by the * backend device. Hopefully it is MAXPHYS. If the driver doesn't * set it properly, use DFLTPHYS. */ if (csw) { max_iosize = dev->si_iosize_max; if (max_iosize < PAGE_SIZE) max_iosize = DFLTPHYS; } else max_iosize = DFLTPHYS; cur_offset = beio->io_offset; for (i = 0; i < beio->num_segs; i++) { size_t cur_size; uint8_t *cur_ptr; cur_size = beio->sg_segs[i].len; cur_ptr = beio->sg_segs[i].addr; while (cur_size > 0) { /* This can't fail, it's a blocking allocation. */ bio = g_alloc_bio(); KASSERT(bio != NULL, ("g_alloc_bio() failed!\n")); bio->bio_cmd = beio->bio_cmd; bio->bio_dev = dev; bio->bio_caller1 = beio; bio->bio_length = min(cur_size, max_iosize); bio->bio_offset = cur_offset; bio->bio_data = cur_ptr; bio->bio_done = ctl_be_block_biodone; bio->bio_pblkno = cur_offset / be_lun->cbe_lun.blocksize; cur_offset += bio->bio_length; cur_ptr += bio->bio_length; cur_size -= bio->bio_length; TAILQ_INSERT_TAIL(&queue, bio, bio_queue); beio->num_bios_sent++; } } binuptime(&beio->ds_t0); mtx_lock(&be_lun->io_lock); devstat_start_transaction(be_lun->disk_stats, &beio->ds_t0); beio->send_complete = 1; mtx_unlock(&be_lun->io_lock); /* * Fire off all allocated requests! */ while ((bio = TAILQ_FIRST(&queue)) != NULL) { TAILQ_REMOVE(&queue, bio, bio_queue); if (csw) csw->d_strategy(bio); else { bio->bio_error = ENXIO; ctl_be_block_biodone(bio); } } if (csw) dev_relthread(dev, ref); } static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) { struct diocgattr_arg arg; struct cdevsw *csw; struct cdev *dev; int error, ref; csw = devvn_refthread(be_lun->vn, &dev, &ref); if (csw == NULL) return (UINT64_MAX); strlcpy(arg.name, attrname, sizeof(arg.name)); arg.len = sizeof(arg.value.off); if (csw->d_ioctl) { error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, curthread); } else error = ENODEV; dev_relthread(dev, ref); if (error != 0) return (UINT64_MAX); return (arg.value.off); } static void ctl_be_block_cw_dispatch_sync(struct ctl_be_block_lun *be_lun, union ctl_io *io) { struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; struct ctl_be_block_io *beio; struct ctl_lba_len_flags *lbalen; DPRINTF("entered\n"); beio = (struct ctl_be_block_io *)PRIV(io)->ptr; lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; beio->io_len = lbalen->len * cbe_lun->blocksize; beio->io_offset = lbalen->lba * cbe_lun->blocksize; beio->io_arg = (lbalen->flags & SSC_IMMED) != 0; beio->bio_cmd = BIO_FLUSH; beio->ds_trans_type = DEVSTAT_NO_DATA; DPRINTF("SYNC\n"); be_lun->lun_flush(be_lun, beio); } static void ctl_be_block_cw_done_ws(struct ctl_be_block_io *beio) { union ctl_io *io; io = beio->io; ctl_free_beio(beio); if ((io->io_hdr.flags & CTL_FLAG_ABORT) || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { ctl_config_write_done(io); return; } ctl_be_block_config_write(io); } static void ctl_be_block_cw_dispatch_ws(struct ctl_be_block_lun *be_lun, union ctl_io *io) { struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; struct ctl_be_block_io *beio; struct ctl_lba_len_flags *lbalen; uint64_t len_left, lba; uint32_t pb, pbo, adj; int i, seglen; uint8_t *buf, *end; DPRINTF("entered\n"); beio = (struct ctl_be_block_io *)PRIV(io)->ptr; lbalen = ARGS(beio->io); if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB) || (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR) && be_lun->unmap == NULL)) { ctl_free_beio(beio); ctl_set_invalid_field(&io->scsiio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 0, /*bit*/ 0); ctl_config_write_done(io); return; } if (lbalen->flags & (SWS_UNMAP | SWS_ANCHOR)) { beio->io_offset = lbalen->lba * cbe_lun->blocksize; beio->io_len = (uint64_t)lbalen->len * cbe_lun->blocksize; beio->bio_cmd = BIO_DELETE; beio->ds_trans_type = DEVSTAT_FREE; be_lun->unmap(be_lun, beio); return; } beio->bio_cmd = BIO_WRITE; beio->ds_trans_type = DEVSTAT_WRITE; DPRINTF("WRITE SAME at LBA %jx len %u\n", (uintmax_t)lbalen->lba, lbalen->len); pb = cbe_lun->blocksize << be_lun->cbe_lun.pblockexp; if (be_lun->cbe_lun.pblockoff > 0) pbo = pb - cbe_lun->blocksize * be_lun->cbe_lun.pblockoff; else pbo = 0; len_left = (uint64_t)lbalen->len * cbe_lun->blocksize; for (i = 0, lba = 0; i < CTLBLK_MAX_SEGS && len_left > 0; i++) { /* * Setup the S/G entry for this chunk. */ seglen = MIN(CTLBLK_MAX_SEG, len_left); if (pb > cbe_lun->blocksize) { adj = ((lbalen->lba + lba) * cbe_lun->blocksize + seglen - pbo) % pb; if (seglen > adj) seglen -= adj; else seglen -= seglen % cbe_lun->blocksize; } else seglen -= seglen % cbe_lun->blocksize; beio->sg_segs[i].len = seglen; beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); DPRINTF("segment %d addr %p len %zd\n", i, beio->sg_segs[i].addr, beio->sg_segs[i].len); beio->num_segs++; len_left -= seglen; buf = beio->sg_segs[i].addr; end = buf + seglen; for (; buf < end; buf += cbe_lun->blocksize) { if (lbalen->flags & SWS_NDOB) { memset(buf, 0, cbe_lun->blocksize); } else { memcpy(buf, io->scsiio.kern_data_ptr, cbe_lun->blocksize); } if (lbalen->flags & SWS_LBDATA) scsi_ulto4b(lbalen->lba + lba, buf); lba++; } } beio->io_offset = lbalen->lba * cbe_lun->blocksize; beio->io_len = lba * cbe_lun->blocksize; /* We can not do all in one run. Correct and schedule rerun. */ if (len_left > 0) { lbalen->lba += lba; lbalen->len -= lba; beio->beio_cont = ctl_be_block_cw_done_ws; } be_lun->dispatch(be_lun, beio); } static void ctl_be_block_cw_dispatch_unmap(struct ctl_be_block_lun *be_lun, union ctl_io *io) { struct ctl_be_block_io *beio; struct ctl_ptr_len_flags *ptrlen; DPRINTF("entered\n"); beio = (struct ctl_be_block_io *)PRIV(io)->ptr; ptrlen = (struct ctl_ptr_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; if ((ptrlen->flags & ~SU_ANCHOR) != 0 || be_lun->unmap == NULL) { ctl_free_beio(beio); ctl_set_invalid_field(&io->scsiio, /*sks_valid*/ 0, /*command*/ 1, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); ctl_config_write_done(io); return; } beio->io_len = 0; beio->io_offset = -1; beio->bio_cmd = BIO_DELETE; beio->ds_trans_type = DEVSTAT_FREE; DPRINTF("UNMAP\n"); be_lun->unmap(be_lun, beio); } static void ctl_be_block_cr_done(struct ctl_be_block_io *beio) { union ctl_io *io; io = beio->io; ctl_free_beio(beio); ctl_config_read_done(io); } static void ctl_be_block_cr_dispatch(struct ctl_be_block_lun *be_lun, union ctl_io *io) { struct ctl_be_block_io *beio; struct ctl_be_block_softc *softc; DPRINTF("entered\n"); softc = be_lun->softc; beio = ctl_alloc_beio(softc); beio->io = io; beio->lun = be_lun; beio->beio_cont = ctl_be_block_cr_done; PRIV(io)->ptr = (void *)beio; switch (io->scsiio.cdb[0]) { case SERVICE_ACTION_IN: /* GET LBA STATUS */ beio->bio_cmd = -1; beio->ds_trans_type = DEVSTAT_NO_DATA; beio->ds_tag_type = DEVSTAT_TAG_ORDERED; beio->io_len = 0; if (be_lun->get_lba_status) be_lun->get_lba_status(be_lun, beio); else ctl_be_block_cr_done(beio); break; default: panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); break; } } static void ctl_be_block_cw_done(struct ctl_be_block_io *beio) { union ctl_io *io; io = beio->io; ctl_free_beio(beio); ctl_config_write_done(io); } static void ctl_be_block_cw_dispatch(struct ctl_be_block_lun *be_lun, union ctl_io *io) { struct ctl_be_block_io *beio; struct ctl_be_block_softc *softc; DPRINTF("entered\n"); softc = be_lun->softc; beio = ctl_alloc_beio(softc); beio->io = io; beio->lun = be_lun; beio->beio_cont = ctl_be_block_cw_done; switch (io->scsiio.tag_type) { case CTL_TAG_ORDERED: beio->ds_tag_type = DEVSTAT_TAG_ORDERED; break; case CTL_TAG_HEAD_OF_QUEUE: beio->ds_tag_type = DEVSTAT_TAG_HEAD; break; case CTL_TAG_UNTAGGED: case CTL_TAG_SIMPLE: case CTL_TAG_ACA: default: beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; break; } PRIV(io)->ptr = (void *)beio; switch (io->scsiio.cdb[0]) { case SYNCHRONIZE_CACHE: case SYNCHRONIZE_CACHE_16: ctl_be_block_cw_dispatch_sync(be_lun, io); break; case WRITE_SAME_10: case WRITE_SAME_16: ctl_be_block_cw_dispatch_ws(be_lun, io); break; case UNMAP: ctl_be_block_cw_dispatch_unmap(be_lun, io); break; default: panic("Unhandled CDB type %#x", io->scsiio.cdb[0]); break; } } SDT_PROBE_DEFINE1(cbb, , read, start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, , write, start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, , read, alloc_done, "uint64_t"); SDT_PROBE_DEFINE1(cbb, , write, alloc_done, "uint64_t"); static void ctl_be_block_next(struct ctl_be_block_io *beio) { struct ctl_be_block_lun *be_lun; union ctl_io *io; io = beio->io; be_lun = beio->lun; ctl_free_beio(beio); if ((io->io_hdr.flags & CTL_FLAG_ABORT) || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE && (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) { ctl_data_submit_done(io); return; } io->io_hdr.status &= ~CTL_STATUS_MASK; io->io_hdr.status |= CTL_STATUS_NONE; mtx_lock(&be_lun->queue_lock); STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); mtx_unlock(&be_lun->queue_lock); taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); } static void ctl_be_block_dispatch(struct ctl_be_block_lun *be_lun, union ctl_io *io) { struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; struct ctl_be_block_io *beio; struct ctl_be_block_softc *softc; struct ctl_lba_len_flags *lbalen; struct ctl_ptr_len_flags *bptrlen; uint64_t len_left, lbas; int i; softc = be_lun->softc; DPRINTF("entered\n"); lbalen = ARGS(io); if (lbalen->flags & CTL_LLF_WRITE) { SDT_PROBE0(cbb, , write, start); } else { SDT_PROBE0(cbb, , read, start); } beio = ctl_alloc_beio(softc); beio->io = io; beio->lun = be_lun; bptrlen = PRIV(io); bptrlen->ptr = (void *)beio; switch (io->scsiio.tag_type) { case CTL_TAG_ORDERED: beio->ds_tag_type = DEVSTAT_TAG_ORDERED; break; case CTL_TAG_HEAD_OF_QUEUE: beio->ds_tag_type = DEVSTAT_TAG_HEAD; break; case CTL_TAG_UNTAGGED: case CTL_TAG_SIMPLE: case CTL_TAG_ACA: default: beio->ds_tag_type = DEVSTAT_TAG_SIMPLE; break; } if (lbalen->flags & CTL_LLF_WRITE) { beio->bio_cmd = BIO_WRITE; beio->ds_trans_type = DEVSTAT_WRITE; } else { beio->bio_cmd = BIO_READ; beio->ds_trans_type = DEVSTAT_READ; } DPRINTF("%s at LBA %jx len %u @%ju\n", (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", (uintmax_t)lbalen->lba, lbalen->len, bptrlen->len); if (lbalen->flags & CTL_LLF_COMPARE) lbas = CTLBLK_HALF_IO_SIZE; else lbas = CTLBLK_MAX_IO_SIZE; lbas = MIN(lbalen->len - bptrlen->len, lbas / cbe_lun->blocksize); beio->io_offset = (lbalen->lba + bptrlen->len) * cbe_lun->blocksize; beio->io_len = lbas * cbe_lun->blocksize; bptrlen->len += lbas; for (i = 0, len_left = beio->io_len; len_left > 0; i++) { KASSERT(i < CTLBLK_MAX_SEGS, ("Too many segs (%d >= %d)", i, CTLBLK_MAX_SEGS)); /* * Setup the S/G entry for this chunk. */ beio->sg_segs[i].len = min(CTLBLK_MAX_SEG, len_left); beio->sg_segs[i].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); DPRINTF("segment %d addr %p len %zd\n", i, beio->sg_segs[i].addr, beio->sg_segs[i].len); /* Set up second segment for compare operation. */ if (lbalen->flags & CTL_LLF_COMPARE) { beio->sg_segs[i + CTLBLK_HALF_SEGS].len = beio->sg_segs[i].len; beio->sg_segs[i + CTLBLK_HALF_SEGS].addr = uma_zalloc(be_lun->lun_zone, M_WAITOK); } beio->num_segs++; len_left -= beio->sg_segs[i].len; } if (bptrlen->len < lbalen->len) beio->beio_cont = ctl_be_block_next; io->scsiio.be_move_done = ctl_be_block_move_done; /* For compare we have separate S/G lists for read and datamove. */ if (lbalen->flags & CTL_LLF_COMPARE) io->scsiio.kern_data_ptr = (uint8_t *)&beio->sg_segs[CTLBLK_HALF_SEGS]; else io->scsiio.kern_data_ptr = (uint8_t *)beio->sg_segs; io->scsiio.kern_data_len = beio->io_len; io->scsiio.kern_sg_entries = beio->num_segs; io->io_hdr.flags |= CTL_FLAG_ALLOCATED; /* * For the read case, we need to read the data into our buffers and * then we can send it back to the user. For the write case, we * need to get the data from the user first. */ if (beio->bio_cmd == BIO_READ) { SDT_PROBE0(cbb, , read, alloc_done); be_lun->dispatch(be_lun, beio); } else { SDT_PROBE0(cbb, , write, alloc_done); #ifdef CTL_TIME_IO getbinuptime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } } static void ctl_be_block_worker(void *context, int pending) { struct ctl_be_block_lun *be_lun = (struct ctl_be_block_lun *)context; struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; union ctl_io *io; struct ctl_be_block_io *beio; DPRINTF("entered\n"); /* * Fetch and process I/Os from all queues. If we detect LUN * CTL_LUN_FLAG_NO_MEDIA status here -- it is result of a race, * so make response maximally opaque to not confuse initiator. */ for (;;) { mtx_lock(&be_lun->queue_lock); io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); if (io != NULL) { DPRINTF("datamove queue\n"); STAILQ_REMOVE(&be_lun->datamove_queue, &io->io_hdr, ctl_io_hdr, links); mtx_unlock(&be_lun->queue_lock); beio = (struct ctl_be_block_io *)PRIV(io)->ptr; if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { ctl_set_busy(&io->scsiio); ctl_complete_beio(beio); return; } be_lun->dispatch(be_lun, beio); continue; } io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_write_queue); if (io != NULL) { DPRINTF("config write queue\n"); STAILQ_REMOVE(&be_lun->config_write_queue, &io->io_hdr, ctl_io_hdr, links); mtx_unlock(&be_lun->queue_lock); if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { ctl_set_busy(&io->scsiio); ctl_config_write_done(io); return; } ctl_be_block_cw_dispatch(be_lun, io); continue; } io = (union ctl_io *)STAILQ_FIRST(&be_lun->config_read_queue); if (io != NULL) { DPRINTF("config read queue\n"); STAILQ_REMOVE(&be_lun->config_read_queue, &io->io_hdr, ctl_io_hdr, links); mtx_unlock(&be_lun->queue_lock); if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { ctl_set_busy(&io->scsiio); ctl_config_read_done(io); return; } ctl_be_block_cr_dispatch(be_lun, io); continue; } io = (union ctl_io *)STAILQ_FIRST(&be_lun->input_queue); if (io != NULL) { DPRINTF("input queue\n"); STAILQ_REMOVE(&be_lun->input_queue, &io->io_hdr, ctl_io_hdr, links); mtx_unlock(&be_lun->queue_lock); if (cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) { ctl_set_busy(&io->scsiio); ctl_data_submit_done(io); return; } ctl_be_block_dispatch(be_lun, io); continue; } /* * If we get here, there is no work left in the queues, so * just break out and let the task queue go to sleep. */ mtx_unlock(&be_lun->queue_lock); break; } } /* * Entry point from CTL to the backend for I/O. We queue everything to a * work thread, so this just puts the I/O on a queue and wakes up the * thread. */ static int ctl_be_block_submit(union ctl_io *io) { struct ctl_be_block_lun *be_lun; struct ctl_be_lun *cbe_lun; DPRINTF("entered\n"); cbe_lun = CTL_BACKEND_LUN(io); be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; /* * Make sure we only get SCSI I/O. */ KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Non-SCSI I/O (type " "%#x) encountered", io->io_hdr.io_type)); PRIV(io)->len = 0; mtx_lock(&be_lun->queue_lock); STAILQ_INSERT_TAIL(&be_lun->input_queue, &io->io_hdr, links); mtx_unlock(&be_lun->queue_lock); taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); return (CTL_RETVAL_COMPLETE); } static int ctl_be_block_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td) { struct ctl_be_block_softc *softc; int error; softc = &backend_block_softc; error = 0; switch (cmd) { case CTL_LUN_REQ: { struct ctl_lun_req *lun_req; lun_req = (struct ctl_lun_req *)addr; switch (lun_req->reqtype) { case CTL_LUNREQ_CREATE: error = ctl_be_block_create(softc, lun_req); break; case CTL_LUNREQ_RM: error = ctl_be_block_rm(softc, lun_req); break; case CTL_LUNREQ_MODIFY: error = ctl_be_block_modify(softc, lun_req); break; default: lun_req->status = CTL_LUN_ERROR; snprintf(lun_req->error_str, sizeof(lun_req->error_str), "invalid LUN request type %d", lun_req->reqtype); break; } break; } default: error = ENOTTY; break; } return (error); } static int ctl_be_block_open_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) { struct ctl_be_lun *cbe_lun; struct ctl_be_block_filedata *file_data; struct ctl_lun_create_params *params; - char *value; + const char *value; struct vattr vattr; off_t ps, pss, po, pos, us, uss, uo, uos; int error; cbe_lun = &be_lun->cbe_lun; file_data = &be_lun->backend.file; params = &be_lun->params; be_lun->dev_type = CTL_BE_BLOCK_FILE; be_lun->dispatch = ctl_be_block_dispatch_file; be_lun->lun_flush = ctl_be_block_flush_file; be_lun->get_lba_status = ctl_be_block_gls_file; be_lun->getattr = ctl_be_block_getattr_file; be_lun->unmap = NULL; cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); if (error != 0) { snprintf(req->error_str, sizeof(req->error_str), "error calling VOP_GETATTR() for file %s", be_lun->dev_path); return (error); } file_data->cred = crhold(curthread->td_ucred); if (params->lun_size_bytes != 0) be_lun->size_bytes = params->lun_size_bytes; else be_lun->size_bytes = vattr.va_size; /* * For files we can use any logical block size. Prefer 512 bytes * for compatibility reasons. If file's vattr.va_blocksize * (preferred I/O block size) is bigger and multiple to chosen * logical block size -- report it as physical block size. */ if (params->blocksize_bytes != 0) cbe_lun->blocksize = params->blocksize_bytes; else if (cbe_lun->lun_type == T_CDROM) cbe_lun->blocksize = 2048; else cbe_lun->blocksize = 512; be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 0 : (be_lun->size_blocks - 1); us = ps = vattr.va_blocksize; uo = po = 0; - value = ctl_get_opt(&cbe_lun->options, "pblocksize"); + value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL); if (value != NULL) ctl_expand_number(value, &ps); - value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); + value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL); if (value != NULL) ctl_expand_number(value, &po); pss = ps / cbe_lun->blocksize; pos = po / cbe_lun->blocksize; if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { cbe_lun->pblockexp = fls(pss) - 1; cbe_lun->pblockoff = (pss - pos) % pss; } - value = ctl_get_opt(&cbe_lun->options, "ublocksize"); + value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL); if (value != NULL) ctl_expand_number(value, &us); - value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); + value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL); if (value != NULL) ctl_expand_number(value, &uo); uss = us / cbe_lun->blocksize; uos = uo / cbe_lun->blocksize; if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { cbe_lun->ublockexp = fls(uss) - 1; cbe_lun->ublockoff = (uss - uos) % uss; } /* * Sanity check. The media size has to be at least one * sector long. */ if (be_lun->size_bytes < cbe_lun->blocksize) { error = EINVAL; snprintf(req->error_str, sizeof(req->error_str), "file %s size %ju < block size %u", be_lun->dev_path, (uintmax_t)be_lun->size_bytes, cbe_lun->blocksize); } cbe_lun->opttxferlen = CTLBLK_MAX_IO_SIZE / cbe_lun->blocksize; return (error); } static int ctl_be_block_open_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) { struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; struct ctl_lun_create_params *params; struct cdevsw *csw; struct cdev *dev; - char *value; + const char *value; int error, atomic, maxio, ref, unmap, tmp; off_t ps, pss, po, pos, us, uss, uo, uos, otmp; params = &be_lun->params; be_lun->dev_type = CTL_BE_BLOCK_DEV; csw = devvn_refthread(be_lun->vn, &dev, &ref); if (csw == NULL) return (ENXIO); if (strcmp(csw->d_name, "zvol") == 0) { be_lun->dispatch = ctl_be_block_dispatch_zvol; be_lun->get_lba_status = ctl_be_block_gls_zvol; atomic = maxio = CTLBLK_MAX_IO_SIZE; } else { be_lun->dispatch = ctl_be_block_dispatch_dev; be_lun->get_lba_status = NULL; atomic = 0; maxio = dev->si_iosize_max; if (maxio <= 0) maxio = DFLTPHYS; if (maxio > CTLBLK_MAX_IO_SIZE) maxio = CTLBLK_MAX_IO_SIZE; } be_lun->lun_flush = ctl_be_block_flush_dev; be_lun->getattr = ctl_be_block_getattr_dev; be_lun->unmap = ctl_be_block_unmap_dev; if (!csw->d_ioctl) { dev_relthread(dev, ref); snprintf(req->error_str, sizeof(req->error_str), "no d_ioctl for device %s!", be_lun->dev_path); return (ENODEV); } error = csw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&tmp, FREAD, curthread); if (error) { dev_relthread(dev, ref); snprintf(req->error_str, sizeof(req->error_str), "error %d returned for DIOCGSECTORSIZE ioctl " "on %s!", error, be_lun->dev_path); return (error); } /* * If the user has asked for a blocksize that is greater than the * backing device's blocksize, we can do it only if the blocksize * the user is asking for is an even multiple of the underlying * device's blocksize. */ if ((params->blocksize_bytes != 0) && (params->blocksize_bytes >= tmp)) { if (params->blocksize_bytes % tmp == 0) { cbe_lun->blocksize = params->blocksize_bytes; } else { dev_relthread(dev, ref); snprintf(req->error_str, sizeof(req->error_str), "requested blocksize %u is not an even " "multiple of backing device blocksize %u", params->blocksize_bytes, tmp); return (EINVAL); } } else if (params->blocksize_bytes != 0) { dev_relthread(dev, ref); snprintf(req->error_str, sizeof(req->error_str), "requested blocksize %u < backing device " "blocksize %u", params->blocksize_bytes, tmp); return (EINVAL); } else if (cbe_lun->lun_type == T_CDROM) cbe_lun->blocksize = MAX(tmp, 2048); else cbe_lun->blocksize = tmp; error = csw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&otmp, FREAD, curthread); if (error) { dev_relthread(dev, ref); snprintf(req->error_str, sizeof(req->error_str), "error %d returned for DIOCGMEDIASIZE " " ioctl on %s!", error, be_lun->dev_path); return (error); } if (params->lun_size_bytes != 0) { if (params->lun_size_bytes > otmp) { dev_relthread(dev, ref); snprintf(req->error_str, sizeof(req->error_str), "requested LUN size %ju > backing device " "size %ju", (uintmax_t)params->lun_size_bytes, (uintmax_t)otmp); return (EINVAL); } be_lun->size_bytes = params->lun_size_bytes; } else be_lun->size_bytes = otmp; be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 0 : (be_lun->size_blocks - 1); error = csw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, curthread); if (error) ps = po = 0; else { error = csw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, FREAD, curthread); if (error) po = 0; } us = ps; uo = po; - value = ctl_get_opt(&cbe_lun->options, "pblocksize"); + value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL); if (value != NULL) ctl_expand_number(value, &ps); - value = ctl_get_opt(&cbe_lun->options, "pblockoffset"); + value = dnvlist_get_string(cbe_lun->options, "pblockoffset", NULL); if (value != NULL) ctl_expand_number(value, &po); pss = ps / cbe_lun->blocksize; pos = po / cbe_lun->blocksize; if ((pss > 0) && (pss * cbe_lun->blocksize == ps) && (pss >= pos) && ((pss & (pss - 1)) == 0) && (pos * cbe_lun->blocksize == po)) { cbe_lun->pblockexp = fls(pss) - 1; cbe_lun->pblockoff = (pss - pos) % pss; } - value = ctl_get_opt(&cbe_lun->options, "ublocksize"); + value = dnvlist_get_string(cbe_lun->options, "ublocksize", NULL); if (value != NULL) ctl_expand_number(value, &us); - value = ctl_get_opt(&cbe_lun->options, "ublockoffset"); + value = dnvlist_get_string(cbe_lun->options, "ublockoffset", NULL); if (value != NULL) ctl_expand_number(value, &uo); uss = us / cbe_lun->blocksize; uos = uo / cbe_lun->blocksize; if ((uss > 0) && (uss * cbe_lun->blocksize == us) && (uss >= uos) && ((uss & (uss - 1)) == 0) && (uos * cbe_lun->blocksize == uo)) { cbe_lun->ublockexp = fls(uss) - 1; cbe_lun->ublockoff = (uss - uos) % uss; } cbe_lun->atomicblock = atomic / cbe_lun->blocksize; cbe_lun->opttxferlen = maxio / cbe_lun->blocksize; if (be_lun->dispatch == ctl_be_block_dispatch_zvol) { unmap = 1; } else { struct diocgattr_arg arg; strlcpy(arg.name, "GEOM::candelete", sizeof(arg.name)); arg.len = sizeof(arg.value.i); error = csw->d_ioctl(dev, DIOCGATTR, (caddr_t)&arg, FREAD, curthread); unmap = (error == 0) ? arg.value.i : 0; } - value = ctl_get_opt(&cbe_lun->options, "unmap"); + value = dnvlist_get_string(cbe_lun->options, "unmap", NULL); if (value != NULL) unmap = (strcmp(value, "on") == 0); if (unmap) cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; else cbe_lun->flags &= ~CTL_LUN_FLAG_UNMAP; dev_relthread(dev, ref); return (0); } static int ctl_be_block_close(struct ctl_be_block_lun *be_lun) { struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; int flags; if (be_lun->vn) { flags = FREAD; if ((cbe_lun->flags & CTL_LUN_FLAG_READONLY) == 0) flags |= FWRITE; (void)vn_close(be_lun->vn, flags, NOCRED, curthread); be_lun->vn = NULL; switch (be_lun->dev_type) { case CTL_BE_BLOCK_DEV: break; case CTL_BE_BLOCK_FILE: if (be_lun->backend.file.cred != NULL) { crfree(be_lun->backend.file.cred); be_lun->backend.file.cred = NULL; } break; case CTL_BE_BLOCK_NONE: break; default: panic("Unexpected backend type %d", be_lun->dev_type); break; } be_lun->dev_type = CTL_BE_BLOCK_NONE; } return (0); } static int ctl_be_block_open(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) { struct ctl_be_lun *cbe_lun = &be_lun->cbe_lun; struct nameidata nd; - char *value; + const char *value; int error, flags; error = 0; if (rootvnode == NULL) { snprintf(req->error_str, sizeof(req->error_str), "Root filesystem is not mounted"); return (1); } pwd_ensure_dirs(); - value = ctl_get_opt(&cbe_lun->options, "file"); + value = dnvlist_get_string(cbe_lun->options, "file", NULL); if (value == NULL) { snprintf(req->error_str, sizeof(req->error_str), "no file argument specified"); return (1); } free(be_lun->dev_path, M_CTLBLK); be_lun->dev_path = strdup(value, M_CTLBLK); flags = FREAD; - value = ctl_get_opt(&cbe_lun->options, "readonly"); + value = dnvlist_get_string(cbe_lun->options, "readonly", NULL); if (value != NULL) { if (strcmp(value, "on") != 0) flags |= FWRITE; } else if (cbe_lun->lun_type == T_DIRECT) flags |= FWRITE; again: NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); error = vn_open(&nd, &flags, 0, NULL); if ((error == EROFS || error == EACCES) && (flags & FWRITE)) { flags &= ~FWRITE; goto again; } if (error) { /* * This is the only reasonable guess we can make as far as * path if the user doesn't give us a fully qualified path. * If they want to specify a file, they need to specify the * full path. */ if (be_lun->dev_path[0] != '/') { char *dev_name; asprintf(&dev_name, M_CTLBLK, "/dev/%s", be_lun->dev_path); free(be_lun->dev_path, M_CTLBLK); be_lun->dev_path = dev_name; goto again; } snprintf(req->error_str, sizeof(req->error_str), "error opening %s: %d", be_lun->dev_path, error); return (error); } if (flags & FWRITE) cbe_lun->flags &= ~CTL_LUN_FLAG_READONLY; else cbe_lun->flags |= CTL_LUN_FLAG_READONLY; NDFREE(&nd, NDF_ONLY_PNBUF); be_lun->vn = nd.ni_vp; /* We only support disks and files. */ if (vn_isdisk(be_lun->vn, &error)) { error = ctl_be_block_open_dev(be_lun, req); } else if (be_lun->vn->v_type == VREG) { error = ctl_be_block_open_file(be_lun, req); } else { error = EINVAL; snprintf(req->error_str, sizeof(req->error_str), "%s is not a disk or plain file", be_lun->dev_path); } VOP_UNLOCK(be_lun->vn, 0); if (error != 0) ctl_be_block_close(be_lun); cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; if (be_lun->dispatch != ctl_be_block_dispatch_dev) cbe_lun->serseq = CTL_LUN_SERSEQ_READ; - value = ctl_get_opt(&cbe_lun->options, "serseq"); + value = dnvlist_get_string(cbe_lun->options, "serseq", NULL); if (value != NULL && strcmp(value, "on") == 0) cbe_lun->serseq = CTL_LUN_SERSEQ_ON; else if (value != NULL && strcmp(value, "read") == 0) cbe_lun->serseq = CTL_LUN_SERSEQ_READ; else if (value != NULL && strcmp(value, "off") == 0) cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; return (0); } static int ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) { struct ctl_be_lun *cbe_lun; struct ctl_be_block_lun *be_lun; struct ctl_lun_create_params *params; char num_thread_str[16]; char tmpstr[32]; - char *value; + const char *value; int retval, num_threads; int tmp_num_threads; params = &req->reqdata.create; retval = 0; req->status = CTL_LUN_OK; be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); cbe_lun = &be_lun->cbe_lun; cbe_lun->be_lun = be_lun; be_lun->params = req->reqdata.create; be_lun->softc = softc; STAILQ_INIT(&be_lun->input_queue); STAILQ_INIT(&be_lun->config_read_queue); STAILQ_INIT(&be_lun->config_write_queue); STAILQ_INIT(&be_lun->datamove_queue); sprintf(be_lun->lunname, "cblk%d", softc->num_luns); mtx_init(&be_lun->io_lock, "cblk io lock", NULL, MTX_DEF); mtx_init(&be_lun->queue_lock, "cblk queue lock", NULL, MTX_DEF); - ctl_init_opts(&cbe_lun->options, - req->num_be_args, req->kern_be_args); + cbe_lun->options = nvlist_clone(req->args_nvl); be_lun->lun_zone = uma_zcreate(be_lun->lunname, CTLBLK_MAX_SEG, NULL, NULL, NULL, NULL, /*align*/ 0, /*flags*/0); if (be_lun->lun_zone == NULL) { snprintf(req->error_str, sizeof(req->error_str), "error allocating UMA zone"); goto bailout_error; } if (params->flags & CTL_LUN_FLAG_DEV_TYPE) cbe_lun->lun_type = params->device_type; else cbe_lun->lun_type = T_DIRECT; be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; cbe_lun->flags = 0; - value = ctl_get_opt(&cbe_lun->options, "ha_role"); + value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); if (value != NULL) { if (strcmp(value, "primary") == 0) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; if (cbe_lun->lun_type == T_DIRECT || cbe_lun->lun_type == T_CDROM) { be_lun->size_bytes = params->lun_size_bytes; if (params->blocksize_bytes != 0) cbe_lun->blocksize = params->blocksize_bytes; else if (cbe_lun->lun_type == T_CDROM) cbe_lun->blocksize = 2048; else cbe_lun->blocksize = 512; be_lun->size_blocks = be_lun->size_bytes / cbe_lun->blocksize; cbe_lun->maxlba = (be_lun->size_blocks == 0) ? 0 : (be_lun->size_blocks - 1); if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { retval = ctl_be_block_open(be_lun, req); if (retval != 0) { retval = 0; req->status = CTL_LUN_WARNING; } } num_threads = cbb_num_threads; } else { num_threads = 1; } - value = ctl_get_opt(&cbe_lun->options, "num_threads"); + value = dnvlist_get_string(cbe_lun->options, "num_threads", NULL); if (value != NULL) { tmp_num_threads = strtol(value, NULL, 0); /* * We don't let the user specify less than one * thread, but hope he's clueful enough not to * specify 1000 threads. */ if (tmp_num_threads < 1) { snprintf(req->error_str, sizeof(req->error_str), "invalid number of threads %s", num_thread_str); goto bailout_error; } num_threads = tmp_num_threads; } if (be_lun->vn == NULL) cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; /* Tell the user the blocksize we ended up using */ params->lun_size_bytes = be_lun->size_bytes; params->blocksize_bytes = cbe_lun->blocksize; if (params->flags & CTL_LUN_FLAG_ID_REQ) { cbe_lun->req_lun_id = params->req_lun_id; cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; } else cbe_lun->req_lun_id = 0; cbe_lun->lun_shutdown = ctl_be_block_lun_shutdown; cbe_lun->lun_config_status = ctl_be_block_lun_config_status; cbe_lun->be = &ctl_be_block_driver; if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d", softc->num_luns); strncpy((char *)cbe_lun->serial_num, tmpstr, MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); /* Tell the user what we used for a serial number */ strncpy((char *)params->serial_num, tmpstr, MIN(sizeof(params->serial_num), sizeof(tmpstr))); } else { strncpy((char *)cbe_lun->serial_num, params->serial_num, MIN(sizeof(cbe_lun->serial_num), sizeof(params->serial_num))); } if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns); strncpy((char *)cbe_lun->device_id, tmpstr, MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); /* Tell the user what we used for a device ID */ strncpy((char *)params->device_id, tmpstr, MIN(sizeof(params->device_id), sizeof(tmpstr))); } else { strncpy((char *)cbe_lun->device_id, params->device_id, MIN(sizeof(cbe_lun->device_id), sizeof(params->device_id))); } TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_be_block_worker, be_lun); be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); if (be_lun->io_taskqueue == NULL) { snprintf(req->error_str, sizeof(req->error_str), "unable to create taskqueue"); goto bailout_error; } /* * Note that we start the same number of threads by default for * both the file case and the block device case. For the file * case, we need multiple threads to allow concurrency, because the * vnode interface is designed to be a blocking interface. For the * block device case, ZFS zvols at least will block the caller's * context in many instances, and so we need multiple threads to * overcome that problem. Other block devices don't need as many * threads, but they shouldn't cause too many problems. * * If the user wants to just have a single thread for a block * device, he can specify that when the LUN is created, or change * the tunable/sysctl to alter the default number of threads. */ retval = taskqueue_start_threads(&be_lun->io_taskqueue, /*num threads*/num_threads, /*priority*/PWAIT, /*thread name*/ "%s taskq", be_lun->lunname); if (retval != 0) goto bailout_error; be_lun->num_threads = num_threads; mtx_lock(&softc->lock); softc->num_luns++; STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); mtx_unlock(&softc->lock); retval = ctl_add_lun(&be_lun->cbe_lun); if (retval != 0) { mtx_lock(&softc->lock); STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); softc->num_luns--; mtx_unlock(&softc->lock); snprintf(req->error_str, sizeof(req->error_str), "ctl_add_lun() returned error %d, see dmesg for " "details", retval); retval = 0; goto bailout_error; } mtx_lock(&softc->lock); /* * Tell the config_status routine that we're waiting so it won't * clean up the LUN in the event of an error. */ be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; while (be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) { retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); if (retval == EINTR) break; } be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; if (be_lun->flags & CTL_BE_BLOCK_LUN_CONFIG_ERR) { snprintf(req->error_str, sizeof(req->error_str), "LUN configuration error, see dmesg for details"); STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); softc->num_luns--; mtx_unlock(&softc->lock); goto bailout_error; } else { params->req_lun_id = cbe_lun->lun_id; } mtx_unlock(&softc->lock); be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, cbe_lun->blocksize, DEVSTAT_ALL_SUPPORTED, cbe_lun->lun_type | DEVSTAT_TYPE_IF_OTHER, DEVSTAT_PRIORITY_OTHER); return (retval); bailout_error: req->status = CTL_LUN_ERROR; if (be_lun->io_taskqueue != NULL) taskqueue_free(be_lun->io_taskqueue); ctl_be_block_close(be_lun); if (be_lun->dev_path != NULL) free(be_lun->dev_path, M_CTLBLK); if (be_lun->lun_zone != NULL) uma_zdestroy(be_lun->lun_zone); - ctl_free_opts(&cbe_lun->options); + nvlist_destroy(cbe_lun->options); mtx_destroy(&be_lun->queue_lock); mtx_destroy(&be_lun->io_lock); free(be_lun, M_CTLBLK); return (retval); } static int ctl_be_block_rm(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) { struct ctl_lun_rm_params *params; struct ctl_be_block_lun *be_lun; struct ctl_be_lun *cbe_lun; int retval; params = &req->reqdata.rm; mtx_lock(&softc->lock); STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->cbe_lun.lun_id == params->lun_id) break; } mtx_unlock(&softc->lock); if (be_lun == NULL) { snprintf(req->error_str, sizeof(req->error_str), "LUN %u is not managed by the block backend", params->lun_id); goto bailout_error; } cbe_lun = &be_lun->cbe_lun; retval = ctl_disable_lun(cbe_lun); if (retval != 0) { snprintf(req->error_str, sizeof(req->error_str), "error %d returned from ctl_disable_lun() for " "LUN %d", retval, params->lun_id); goto bailout_error; } if (be_lun->vn != NULL) { cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; ctl_lun_no_media(cbe_lun); taskqueue_drain_all(be_lun->io_taskqueue); ctl_be_block_close(be_lun); } retval = ctl_invalidate_lun(cbe_lun); if (retval != 0) { snprintf(req->error_str, sizeof(req->error_str), "error %d returned from ctl_invalidate_lun() for " "LUN %d", retval, params->lun_id); goto bailout_error; } mtx_lock(&softc->lock); be_lun->flags |= CTL_BE_BLOCK_LUN_WAITING; while ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { retval = msleep(be_lun, &softc->lock, PCATCH, "ctlblk", 0); if (retval == EINTR) break; } be_lun->flags &= ~CTL_BE_BLOCK_LUN_WAITING; if ((be_lun->flags & CTL_BE_BLOCK_LUN_UNCONFIGURED) == 0) { snprintf(req->error_str, sizeof(req->error_str), "interrupted waiting for LUN to be freed"); mtx_unlock(&softc->lock); goto bailout_error; } STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_block_lun, links); softc->num_luns--; mtx_unlock(&softc->lock); taskqueue_drain_all(be_lun->io_taskqueue); taskqueue_free(be_lun->io_taskqueue); if (be_lun->disk_stats != NULL) devstat_remove_entry(be_lun->disk_stats); uma_zdestroy(be_lun->lun_zone); - ctl_free_opts(&cbe_lun->options); + nvlist_destroy(cbe_lun->options); free(be_lun->dev_path, M_CTLBLK); mtx_destroy(&be_lun->queue_lock); mtx_destroy(&be_lun->io_lock); free(be_lun, M_CTLBLK); req->status = CTL_LUN_OK; return (0); bailout_error: req->status = CTL_LUN_ERROR; return (0); } static int ctl_be_block_modify(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) { struct ctl_lun_modify_params *params; struct ctl_be_block_lun *be_lun; struct ctl_be_lun *cbe_lun; - char *value; + const char *value; uint64_t oldsize; int error, wasprim; params = &req->reqdata.modify; mtx_lock(&softc->lock); STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->cbe_lun.lun_id == params->lun_id) break; } mtx_unlock(&softc->lock); if (be_lun == NULL) { snprintf(req->error_str, sizeof(req->error_str), "LUN %u is not managed by the block backend", params->lun_id); goto bailout_error; } cbe_lun = &be_lun->cbe_lun; if (params->lun_size_bytes != 0) be_lun->params.lun_size_bytes = params->lun_size_bytes; - ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); + nvlist_destroy(cbe_lun->options); + cbe_lun->options = nvlist_clone(req->args_nvl); + wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); - value = ctl_get_opt(&cbe_lun->options, "ha_role"); + value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); if (value != NULL) { if (strcmp(value, "primary") == 0) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; else cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; else cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ctl_lun_primary(cbe_lun); else ctl_lun_secondary(cbe_lun); } oldsize = be_lun->size_blocks; if ((cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) || control_softc->ha_mode == CTL_HA_MODE_SER_ONLY) { if (be_lun->vn == NULL) error = ctl_be_block_open(be_lun, req); else if (vn_isdisk(be_lun->vn, &error)) error = ctl_be_block_open_dev(be_lun, req); else if (be_lun->vn->v_type == VREG) { vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); error = ctl_be_block_open_file(be_lun, req); VOP_UNLOCK(be_lun->vn, 0); } else error = EINVAL; if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) && be_lun->vn != NULL) { cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; ctl_lun_has_media(cbe_lun); } else if ((cbe_lun->flags & CTL_LUN_FLAG_NO_MEDIA) == 0 && be_lun->vn == NULL) { cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; ctl_lun_no_media(cbe_lun); } cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; } else { if (be_lun->vn != NULL) { cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; ctl_lun_no_media(cbe_lun); taskqueue_drain_all(be_lun->io_taskqueue); error = ctl_be_block_close(be_lun); } else error = 0; } if (be_lun->size_blocks != oldsize) ctl_lun_capacity_changed(cbe_lun); /* Tell the user the exact size we ended up using */ params->lun_size_bytes = be_lun->size_bytes; req->status = error ? CTL_LUN_WARNING : CTL_LUN_OK; return (0); bailout_error: req->status = CTL_LUN_ERROR; return (0); } static void ctl_be_block_lun_shutdown(void *be_lun) { struct ctl_be_block_lun *lun = be_lun; struct ctl_be_block_softc *softc = lun->softc; mtx_lock(&softc->lock); lun->flags |= CTL_BE_BLOCK_LUN_UNCONFIGURED; if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) wakeup(lun); mtx_unlock(&softc->lock); } static void ctl_be_block_lun_config_status(void *be_lun, ctl_lun_config_status status) { struct ctl_be_block_lun *lun; struct ctl_be_block_softc *softc; lun = (struct ctl_be_block_lun *)be_lun; softc = lun->softc; if (status == CTL_LUN_CONFIG_OK) { mtx_lock(&softc->lock); lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; if (lun->flags & CTL_BE_BLOCK_LUN_WAITING) wakeup(lun); mtx_unlock(&softc->lock); /* * We successfully added the LUN, attempt to enable it. */ if (ctl_enable_lun(&lun->cbe_lun) != 0) { printf("%s: ctl_enable_lun() failed!\n", __func__); if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { printf("%s: ctl_invalidate_lun() failed!\n", __func__); } } return; } mtx_lock(&softc->lock); lun->flags &= ~CTL_BE_BLOCK_LUN_UNCONFIGURED; lun->flags |= CTL_BE_BLOCK_LUN_CONFIG_ERR; wakeup(lun); mtx_unlock(&softc->lock); } static int ctl_be_block_config_write(union ctl_io *io) { struct ctl_be_block_lun *be_lun; struct ctl_be_lun *cbe_lun; int retval; DPRINTF("entered\n"); cbe_lun = CTL_BACKEND_LUN(io); be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; retval = 0; switch (io->scsiio.cdb[0]) { case SYNCHRONIZE_CACHE: case SYNCHRONIZE_CACHE_16: case WRITE_SAME_10: case WRITE_SAME_16: case UNMAP: /* * The upper level CTL code will filter out any CDBs with * the immediate bit set and return the proper error. * * We don't really need to worry about what LBA range the * user asked to be synced out. When they issue a sync * cache command, we'll sync out the whole thing. */ mtx_lock(&be_lun->queue_lock); STAILQ_INSERT_TAIL(&be_lun->config_write_queue, &io->io_hdr, links); mtx_unlock(&be_lun->queue_lock); taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); break; case START_STOP_UNIT: { struct scsi_start_stop_unit *cdb; struct ctl_lun_req req; cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; if ((cdb->how & SSS_PC_MASK) != 0) { ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; } if (cdb->how & SSS_START) { if ((cdb->how & SSS_LOEJ) && be_lun->vn == NULL) { retval = ctl_be_block_open(be_lun, &req); cbe_lun->flags &= ~CTL_LUN_FLAG_EJECTED; if (retval == 0) { cbe_lun->flags &= ~CTL_LUN_FLAG_NO_MEDIA; ctl_lun_has_media(cbe_lun); } else { cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; ctl_lun_no_media(cbe_lun); } } ctl_start_lun(cbe_lun); } else { ctl_stop_lun(cbe_lun); if (cdb->how & SSS_LOEJ) { cbe_lun->flags |= CTL_LUN_FLAG_NO_MEDIA; cbe_lun->flags |= CTL_LUN_FLAG_EJECTED; ctl_lun_ejected(cbe_lun); if (be_lun->vn != NULL) ctl_be_block_close(be_lun); } } ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; } case PREVENT_ALLOW: ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; default: ctl_set_invalid_opcode(&io->scsiio); ctl_config_write_done(io); retval = CTL_RETVAL_COMPLETE; break; } return (retval); } static int ctl_be_block_config_read(union ctl_io *io) { struct ctl_be_block_lun *be_lun; struct ctl_be_lun *cbe_lun; int retval = 0; DPRINTF("entered\n"); cbe_lun = CTL_BACKEND_LUN(io); be_lun = (struct ctl_be_block_lun *)cbe_lun->be_lun; switch (io->scsiio.cdb[0]) { case SERVICE_ACTION_IN: if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { mtx_lock(&be_lun->queue_lock); STAILQ_INSERT_TAIL(&be_lun->config_read_queue, &io->io_hdr, links); mtx_unlock(&be_lun->queue_lock); taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); retval = CTL_RETVAL_QUEUED; break; } ctl_set_invalid_field(&io->scsiio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 4); ctl_config_read_done(io); retval = CTL_RETVAL_COMPLETE; break; default: ctl_set_invalid_opcode(&io->scsiio); ctl_config_read_done(io); retval = CTL_RETVAL_COMPLETE; break; } return (retval); } static int ctl_be_block_lun_info(void *be_lun, struct sbuf *sb) { struct ctl_be_block_lun *lun; int retval; lun = (struct ctl_be_block_lun *)be_lun; retval = sbuf_printf(sb, "\t"); if (retval != 0) goto bailout; retval = sbuf_printf(sb, "%d", lun->num_threads); if (retval != 0) goto bailout; retval = sbuf_printf(sb, "\n"); bailout: return (retval); } static uint64_t ctl_be_block_lun_attr(void *be_lun, const char *attrname) { struct ctl_be_block_lun *lun = (struct ctl_be_block_lun *)be_lun; if (lun->getattr == NULL) return (UINT64_MAX); return (lun->getattr(lun, attrname)); } static int ctl_be_block_init(void) { struct ctl_be_block_softc *softc = &backend_block_softc; mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); softc->beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); STAILQ_INIT(&softc->lun_list); return (0); } static int ctl_be_block_shutdown(void) { struct ctl_be_block_softc *softc = &backend_block_softc; struct ctl_be_block_lun *lun, *next_lun; mtx_lock(&softc->lock); STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun) { /* * Drop our lock here. Since ctl_invalidate_lun() can call * back into us, this could potentially lead to a recursive * lock of the same mutex, which would cause a hang. */ mtx_unlock(&softc->lock); ctl_disable_lun(&lun->cbe_lun); ctl_invalidate_lun(&lun->cbe_lun); mtx_lock(&softc->lock); } mtx_unlock(&softc->lock); uma_zdestroy(softc->beio_zone); mtx_destroy(&softc->lock); return (0); } Index: head/sys/cam/ctl/ctl_backend_ramdisk.c =================================================================== --- head/sys/cam/ctl/ctl_backend_ramdisk.c (revision 333445) +++ head/sys/cam/ctl/ctl_backend_ramdisk.c (revision 333446) @@ -1,1362 +1,1366 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003, 2008 Silicon Graphics International Corp. * Copyright (c) 2012 The FreeBSD Foundation * Copyright (c) 2014-2017 Alexander Motin * All rights reserved. * * Portions of this software were developed by Edward Tomasz Napierala * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend_ramdisk.c#3 $ */ /* * CAM Target Layer black hole and RAM disk backend. * * Author: Ken Merry */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include #include #include #include #include #include #define PRIV(io) \ ((struct ctl_ptr_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_BACKEND]) #define ARGS(io) \ ((struct ctl_lba_len_flags *)&(io)->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]) #define PPP (PAGE_SIZE / sizeof(uint8_t **)) #ifdef __LP64__ #define PPPS (PAGE_SHIFT - 3) #else #define PPPS (PAGE_SHIFT - 2) #endif #define SGPP (PAGE_SIZE / sizeof(struct ctl_sg_entry)) #define P_UNMAPPED NULL /* Page is unmapped. */ #define P_ANCHORED ((void *)(uintptr_t)1) /* Page is anchored. */ typedef enum { GP_READ, /* Return data page or zero page. */ GP_WRITE, /* Return data page, try allocate if none. */ GP_ANCHOR, /* Return data page, try anchor if none. */ GP_OTHER, /* Return what present, do not allocate/anchor. */ } getpage_op_t; typedef enum { CTL_BE_RAMDISK_LUN_UNCONFIGURED = 0x01, CTL_BE_RAMDISK_LUN_CONFIG_ERR = 0x02, CTL_BE_RAMDISK_LUN_WAITING = 0x04 } ctl_be_ramdisk_lun_flags; struct ctl_be_ramdisk_lun { struct ctl_lun_create_params params; char lunname[32]; int indir; uint8_t **pages; uint8_t *zero_page; struct sx page_lock; u_int pblocksize; u_int pblockmul; uint64_t size_bytes; uint64_t size_blocks; uint64_t cap_bytes; uint64_t cap_used; struct ctl_be_ramdisk_softc *softc; ctl_be_ramdisk_lun_flags flags; STAILQ_ENTRY(ctl_be_ramdisk_lun) links; struct ctl_be_lun cbe_lun; struct taskqueue *io_taskqueue; struct task io_task; STAILQ_HEAD(, ctl_io_hdr) cont_queue; struct mtx_padalign queue_lock; }; struct ctl_be_ramdisk_softc { struct mtx lock; int num_luns; STAILQ_HEAD(, ctl_be_ramdisk_lun) lun_list; }; static struct ctl_be_ramdisk_softc rd_softc; extern struct ctl_softc *control_softc; static int ctl_backend_ramdisk_init(void); static int ctl_backend_ramdisk_shutdown(void); static int ctl_backend_ramdisk_move_done(union ctl_io *io); static void ctl_backend_ramdisk_compare(union ctl_io *io); static void ctl_backend_ramdisk_rw(union ctl_io *io); static int ctl_backend_ramdisk_submit(union ctl_io *io); static void ctl_backend_ramdisk_worker(void *context, int pending); static int ctl_backend_ramdisk_config_read(union ctl_io *io); static int ctl_backend_ramdisk_config_write(union ctl_io *io); static uint64_t ctl_backend_ramdisk_lun_attr(void *be_lun, const char *attrname); static int ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); static int ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc, struct ctl_lun_req *req); static int ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc, struct ctl_lun_req *req); static int ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc, struct ctl_lun_req *req); static void ctl_backend_ramdisk_lun_shutdown(void *be_lun); static void ctl_backend_ramdisk_lun_config_status(void *be_lun, ctl_lun_config_status status); static struct ctl_backend_driver ctl_be_ramdisk_driver = { .name = "ramdisk", .flags = CTL_BE_FLAG_HAS_CONFIG, .init = ctl_backend_ramdisk_init, .shutdown = ctl_backend_ramdisk_shutdown, .data_submit = ctl_backend_ramdisk_submit, .data_move_done = ctl_backend_ramdisk_move_done, .config_read = ctl_backend_ramdisk_config_read, .config_write = ctl_backend_ramdisk_config_write, .ioctl = ctl_backend_ramdisk_ioctl, .lun_attr = ctl_backend_ramdisk_lun_attr, }; MALLOC_DEFINE(M_RAMDISK, "ramdisk", "Memory used for CTL RAMdisk"); CTL_BACKEND_DECLARE(cbr, ctl_be_ramdisk_driver); static int ctl_backend_ramdisk_init(void) { struct ctl_be_ramdisk_softc *softc = &rd_softc; memset(softc, 0, sizeof(*softc)); mtx_init(&softc->lock, "ctlramdisk", NULL, MTX_DEF); STAILQ_INIT(&softc->lun_list); return (0); } static int ctl_backend_ramdisk_shutdown(void) { struct ctl_be_ramdisk_softc *softc = &rd_softc; struct ctl_be_ramdisk_lun *lun, *next_lun; mtx_lock(&softc->lock); STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun) { /* * Drop our lock here. Since ctl_invalidate_lun() can call * back into us, this could potentially lead to a recursive * lock of the same mutex, which would cause a hang. */ mtx_unlock(&softc->lock); ctl_disable_lun(&lun->cbe_lun); ctl_invalidate_lun(&lun->cbe_lun); mtx_lock(&softc->lock); } mtx_unlock(&softc->lock); mtx_destroy(&softc->lock); return (0); } static uint8_t * ctl_backend_ramdisk_getpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn, getpage_op_t op) { uint8_t **p, ***pp; off_t i; int s; if (be_lun->cap_bytes == 0) { switch (op) { case GP_READ: return (be_lun->zero_page); case GP_WRITE: return ((uint8_t *)be_lun->pages); case GP_ANCHOR: return (P_ANCHORED); default: return (P_UNMAPPED); } } if (op == GP_WRITE || op == GP_ANCHOR) { sx_xlock(&be_lun->page_lock); pp = &be_lun->pages; for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) { if (*pp == NULL) { *pp = malloc(PAGE_SIZE, M_RAMDISK, M_WAITOK|M_ZERO); } i = pn >> s; pp = (uint8_t ***)&(*pp)[i]; pn -= i << s; } if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) { if (op == GP_WRITE) { *pp = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK|M_ZERO); } else *pp = P_ANCHORED; be_lun->cap_used += be_lun->pblocksize; } else if (*pp == P_ANCHORED && op == GP_WRITE) { *pp = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK|M_ZERO); } sx_xunlock(&be_lun->page_lock); return ((uint8_t *)*pp); } else { sx_slock(&be_lun->page_lock); p = be_lun->pages; for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) { if (p == NULL) break; i = pn >> s; p = (uint8_t **)p[i]; pn -= i << s; } sx_sunlock(&be_lun->page_lock); if ((p == P_UNMAPPED || p == P_ANCHORED) && op == GP_READ) return (be_lun->zero_page); return ((uint8_t *)p); } }; static void ctl_backend_ramdisk_unmappage(struct ctl_be_ramdisk_lun *be_lun, off_t pn) { uint8_t ***pp; off_t i; int s; if (be_lun->cap_bytes == 0) return; sx_xlock(&be_lun->page_lock); pp = &be_lun->pages; for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) { if (*pp == NULL) goto noindir; i = pn >> s; pp = (uint8_t ***)&(*pp)[i]; pn -= i << s; } if (*pp == P_ANCHORED) { be_lun->cap_used -= be_lun->pblocksize; *pp = P_UNMAPPED; } else if (*pp != P_UNMAPPED) { free(*pp, M_RAMDISK); be_lun->cap_used -= be_lun->pblocksize; *pp = P_UNMAPPED; } noindir: sx_xunlock(&be_lun->page_lock); }; static void ctl_backend_ramdisk_anchorpage(struct ctl_be_ramdisk_lun *be_lun, off_t pn) { uint8_t ***pp; off_t i; int s; if (be_lun->cap_bytes == 0) return; sx_xlock(&be_lun->page_lock); pp = &be_lun->pages; for (s = (be_lun->indir - 1) * PPPS; s >= 0; s -= PPPS) { if (*pp == NULL) goto noindir; i = pn >> s; pp = (uint8_t ***)&(*pp)[i]; pn -= i << s; } if (*pp == P_UNMAPPED && be_lun->cap_used < be_lun->cap_bytes) { be_lun->cap_used += be_lun->pblocksize; *pp = P_ANCHORED; } else if (*pp != P_ANCHORED) { free(*pp, M_RAMDISK); *pp = P_ANCHORED; } noindir: sx_xunlock(&be_lun->page_lock); }; static void ctl_backend_ramdisk_freeallpages(uint8_t **p, int indir) { int i; if (p == NULL) return; if (indir == 0) { free(p, M_RAMDISK); return; } for (i = 0; i < PPP; i++) { if (p[i] == NULL) continue; ctl_backend_ramdisk_freeallpages((uint8_t **)p[i], indir - 1); } free(p, M_RAMDISK); }; static size_t cmp(uint8_t *a, uint8_t *b, size_t size) { size_t i; for (i = 0; i < size; i++) { if (a[i] != b[i]) break; } return (i); } static int ctl_backend_ramdisk_cmp(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun; uint8_t *page; uint8_t info[8]; uint64_t lba; u_int lbaoff, lbas, res, off; lbas = io->scsiio.kern_data_len / cbe_lun->blocksize; lba = ARGS(io)->lba + PRIV(io)->len - lbas; off = 0; for (; lbas > 0; lbas--, lba++) { page = ctl_backend_ramdisk_getpage(be_lun, lba >> cbe_lun->pblockexp, GP_READ); lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp); page += lbaoff * cbe_lun->blocksize; res = cmp(io->scsiio.kern_data_ptr + off, page, cbe_lun->blocksize); off += res; if (res < cbe_lun->blocksize) break; } if (lbas > 0) { off += io->scsiio.kern_rel_offset - io->scsiio.kern_data_len; scsi_u64to8b(off, info); ctl_set_sense(&io->scsiio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_MISCOMPARE, /*asc*/ 0x1D, /*ascq*/ 0x00, /*type*/ SSD_ELEM_INFO, /*size*/ sizeof(info), /*data*/ &info, /*type*/ SSD_ELEM_NONE); return (1); } return (0); } static int ctl_backend_ramdisk_move_done(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun; #ifdef CTL_TIME_IO struct bintime cur_bt; #endif CTL_DEBUG_PRINT(("ctl_backend_ramdisk_move_done\n")); #ifdef CTL_TIME_IO getbinuptime(&cur_bt); bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); bintime_add(&io->io_hdr.dma_bt, &cur_bt); #endif io->io_hdr.num_dmas++; if (io->scsiio.kern_sg_entries > 0) free(io->scsiio.kern_data_ptr, M_RAMDISK); io->scsiio.kern_rel_offset += io->scsiio.kern_data_len; if (io->io_hdr.flags & CTL_FLAG_ABORT) { ; } else if (io->io_hdr.port_status != 0 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ io->io_hdr.port_status); } else if (io->scsiio.kern_data_resid != 0 && (io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE || (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) { ctl_set_invalid_field_ciu(&io->scsiio); } else if ((io->io_hdr.port_status == 0) && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { if (ARGS(io)->flags & CTL_LLF_COMPARE) { /* We have data block ready for comparison. */ if (ctl_backend_ramdisk_cmp(io)) goto done; } if (ARGS(io)->len > PRIV(io)->len) { mtx_lock(&be_lun->queue_lock); STAILQ_INSERT_TAIL(&be_lun->cont_queue, &io->io_hdr, links); mtx_unlock(&be_lun->queue_lock); taskqueue_enqueue(be_lun->io_taskqueue, &be_lun->io_task); return (0); } ctl_set_success(&io->scsiio); } done: ctl_data_submit_done(io); return(0); } static void ctl_backend_ramdisk_compare(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); u_int lbas, len; lbas = ARGS(io)->len - PRIV(io)->len; lbas = MIN(lbas, 131072 / cbe_lun->blocksize); len = lbas * cbe_lun->blocksize; io->scsiio.be_move_done = ctl_backend_ramdisk_move_done; io->scsiio.kern_data_ptr = malloc(len, M_RAMDISK, M_WAITOK); io->scsiio.kern_data_len = len; io->scsiio.kern_sg_entries = 0; io->io_hdr.flags |= CTL_FLAG_ALLOCATED; PRIV(io)->len += lbas; #ifdef CTL_TIME_IO getbinuptime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } static void ctl_backend_ramdisk_rw(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun; struct ctl_sg_entry *sg_entries; uint8_t *page; uint64_t lba; u_int i, len, lbaoff, lbas, sgs, off; getpage_op_t op; lba = ARGS(io)->lba + PRIV(io)->len; lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp); lbas = ARGS(io)->len - PRIV(io)->len; lbas = MIN(lbas, (SGPP << cbe_lun->pblockexp) - lbaoff); sgs = (lbas + lbaoff + be_lun->pblockmul - 1) >> cbe_lun->pblockexp; off = lbaoff * cbe_lun->blocksize; op = (ARGS(io)->flags & CTL_LLF_WRITE) ? GP_WRITE : GP_READ; if (sgs > 1) { io->scsiio.kern_data_ptr = malloc(sizeof(struct ctl_sg_entry) * sgs, M_RAMDISK, M_WAITOK); sg_entries = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; len = lbas * cbe_lun->blocksize; for (i = 0; i < sgs; i++) { page = ctl_backend_ramdisk_getpage(be_lun, (lba >> cbe_lun->pblockexp) + i, op); if (page == P_UNMAPPED || page == P_ANCHORED) { free(io->scsiio.kern_data_ptr, M_RAMDISK); nospc: ctl_set_space_alloc_fail(&io->scsiio); ctl_data_submit_done(io); return; } sg_entries[i].addr = page + off; sg_entries[i].len = MIN(len, be_lun->pblocksize - off); len -= sg_entries[i].len; off = 0; } } else { page = ctl_backend_ramdisk_getpage(be_lun, lba >> cbe_lun->pblockexp, op); if (page == P_UNMAPPED || page == P_ANCHORED) goto nospc; sgs = 0; io->scsiio.kern_data_ptr = page + off; } io->scsiio.be_move_done = ctl_backend_ramdisk_move_done; io->scsiio.kern_data_len = lbas * cbe_lun->blocksize; io->scsiio.kern_sg_entries = sgs; io->io_hdr.flags |= CTL_FLAG_ALLOCATED; PRIV(io)->len += lbas; if ((ARGS(io)->flags & CTL_LLF_READ) && ARGS(io)->len <= PRIV(io)->len) { ctl_set_success(&io->scsiio); ctl_serseq_done(io); } #ifdef CTL_TIME_IO getbinuptime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } static int ctl_backend_ramdisk_submit(union ctl_io *io) { struct ctl_lba_len_flags *lbalen = ARGS(io); if (lbalen->flags & CTL_LLF_VERIFY) { ctl_set_success(&io->scsiio); ctl_data_submit_done(io); return (CTL_RETVAL_COMPLETE); } PRIV(io)->len = 0; if (lbalen->flags & CTL_LLF_COMPARE) ctl_backend_ramdisk_compare(io); else ctl_backend_ramdisk_rw(io); return (CTL_RETVAL_COMPLETE); } static void ctl_backend_ramdisk_worker(void *context, int pending) { struct ctl_be_ramdisk_lun *be_lun; union ctl_io *io; be_lun = (struct ctl_be_ramdisk_lun *)context; mtx_lock(&be_lun->queue_lock); for (;;) { io = (union ctl_io *)STAILQ_FIRST(&be_lun->cont_queue); if (io != NULL) { STAILQ_REMOVE(&be_lun->cont_queue, &io->io_hdr, ctl_io_hdr, links); mtx_unlock(&be_lun->queue_lock); if (ARGS(io)->flags & CTL_LLF_COMPARE) ctl_backend_ramdisk_compare(io); else ctl_backend_ramdisk_rw(io); mtx_lock(&be_lun->queue_lock); continue; } /* * If we get here, there is no work left in the queues, so * just break out and let the task queue go to sleep. */ break; } mtx_unlock(&be_lun->queue_lock); } static int ctl_backend_ramdisk_gls(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun; struct scsi_get_lba_status_data *data; uint8_t *page; u_int lbaoff; data = (struct scsi_get_lba_status_data *)io->scsiio.kern_data_ptr; scsi_u64to8b(ARGS(io)->lba, data->descr[0].addr); lbaoff = ARGS(io)->lba & ~(UINT_MAX << cbe_lun->pblockexp); scsi_ulto4b(be_lun->pblockmul - lbaoff, data->descr[0].length); page = ctl_backend_ramdisk_getpage(be_lun, ARGS(io)->lba >> cbe_lun->pblockexp, GP_OTHER); if (page == P_UNMAPPED) data->descr[0].status = 1; else if (page == P_ANCHORED) data->descr[0].status = 2; else data->descr[0].status = 0; ctl_config_read_done(io); return (CTL_RETVAL_COMPLETE); } static int ctl_backend_ramdisk_config_read(union ctl_io *io) { int retval = 0; switch (io->scsiio.cdb[0]) { case SERVICE_ACTION_IN: if (io->scsiio.cdb[1] == SGLS_SERVICE_ACTION) { retval = ctl_backend_ramdisk_gls(io); break; } ctl_set_invalid_field(&io->scsiio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 4); ctl_config_read_done(io); retval = CTL_RETVAL_COMPLETE; break; default: ctl_set_invalid_opcode(&io->scsiio); ctl_config_read_done(io); retval = CTL_RETVAL_COMPLETE; break; } return (retval); } static void ctl_backend_ramdisk_delete(struct ctl_be_lun *cbe_lun, off_t lba, off_t len, int anchor) { struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun; uint8_t *page; uint64_t p, lp; u_int lbaoff; getpage_op_t op = anchor ? GP_ANCHOR : GP_OTHER; /* Partially zero first partial page. */ p = lba >> cbe_lun->pblockexp; lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp); if (lbaoff != 0) { page = ctl_backend_ramdisk_getpage(be_lun, p, op); if (page != P_UNMAPPED && page != P_ANCHORED) { memset(page + lbaoff * cbe_lun->blocksize, 0, min(len, be_lun->pblockmul - lbaoff) * cbe_lun->blocksize); } p++; } /* Partially zero last partial page. */ lp = (lba + len) >> cbe_lun->pblockexp; lbaoff = (lba + len) & ~(UINT_MAX << cbe_lun->pblockexp); if (p <= lp && lbaoff != 0) { page = ctl_backend_ramdisk_getpage(be_lun, lp, op); if (page != P_UNMAPPED && page != P_ANCHORED) memset(page, 0, lbaoff * cbe_lun->blocksize); } /* Delete remaining full pages. */ if (anchor) { for (; p < lp; p++) ctl_backend_ramdisk_anchorpage(be_lun, p); } else { for (; p < lp; p++) ctl_backend_ramdisk_unmappage(be_lun, p); } } static void ctl_backend_ramdisk_ws(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); struct ctl_be_ramdisk_lun *be_lun = cbe_lun->be_lun; struct ctl_lba_len_flags *lbalen = ARGS(io); uint8_t *page; uint64_t lba; u_int lbaoff, lbas; if (lbalen->flags & ~(SWS_LBDATA | SWS_UNMAP | SWS_ANCHOR | SWS_NDOB)) { ctl_set_invalid_field(&io->scsiio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 0, /*bit*/ 0); ctl_config_write_done(io); return; } if (lbalen->flags & SWS_UNMAP) { ctl_backend_ramdisk_delete(cbe_lun, lbalen->lba, lbalen->len, (lbalen->flags & SWS_ANCHOR) != 0); ctl_set_success(&io->scsiio); ctl_config_write_done(io); return; } for (lba = lbalen->lba, lbas = lbalen->len; lbas > 0; lba++, lbas--) { page = ctl_backend_ramdisk_getpage(be_lun, lba >> cbe_lun->pblockexp, GP_WRITE); if (page == P_UNMAPPED || page == P_ANCHORED) { ctl_set_space_alloc_fail(&io->scsiio); ctl_data_submit_done(io); return; } lbaoff = lba & ~(UINT_MAX << cbe_lun->pblockexp); page += lbaoff * cbe_lun->blocksize; if (lbalen->flags & SWS_NDOB) { memset(page, 0, cbe_lun->blocksize); } else { memcpy(page, io->scsiio.kern_data_ptr, cbe_lun->blocksize); } if (lbalen->flags & SWS_LBDATA) scsi_ulto4b(lba, page); } ctl_set_success(&io->scsiio); ctl_config_write_done(io); } static void ctl_backend_ramdisk_unmap(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); struct ctl_ptr_len_flags *ptrlen = (struct ctl_ptr_len_flags *)ARGS(io); struct scsi_unmap_desc *buf, *end; if ((ptrlen->flags & ~SU_ANCHOR) != 0) { ctl_set_invalid_field(&io->scsiio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); ctl_config_write_done(io); return; } buf = (struct scsi_unmap_desc *)ptrlen->ptr; end = buf + ptrlen->len / sizeof(*buf); for (; buf < end; buf++) { ctl_backend_ramdisk_delete(cbe_lun, scsi_8btou64(buf->lba), scsi_4btoul(buf->length), (ptrlen->flags & SU_ANCHOR) != 0); } ctl_set_success(&io->scsiio); ctl_config_write_done(io); } static int ctl_backend_ramdisk_config_write(union ctl_io *io) { struct ctl_be_lun *cbe_lun = CTL_BACKEND_LUN(io); int retval = 0; switch (io->scsiio.cdb[0]) { case SYNCHRONIZE_CACHE: case SYNCHRONIZE_CACHE_16: /* We have no cache to flush. */ ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; case START_STOP_UNIT: { struct scsi_start_stop_unit *cdb; cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; if ((cdb->how & SSS_PC_MASK) != 0) { ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; } if (cdb->how & SSS_START) { if (cdb->how & SSS_LOEJ) ctl_lun_has_media(cbe_lun); ctl_start_lun(cbe_lun); } else { ctl_stop_lun(cbe_lun); if (cdb->how & SSS_LOEJ) ctl_lun_ejected(cbe_lun); } ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; } case PREVENT_ALLOW: ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; case WRITE_SAME_10: case WRITE_SAME_16: ctl_backend_ramdisk_ws(io); break; case UNMAP: ctl_backend_ramdisk_unmap(io); break; default: ctl_set_invalid_opcode(&io->scsiio); ctl_config_write_done(io); retval = CTL_RETVAL_COMPLETE; break; } return (retval); } static uint64_t ctl_backend_ramdisk_lun_attr(void *arg, const char *attrname) { struct ctl_be_ramdisk_lun *be_lun = arg; uint64_t val; val = UINT64_MAX; if (be_lun->cap_bytes == 0) return (val); sx_slock(&be_lun->page_lock); if (strcmp(attrname, "blocksused") == 0) { val = be_lun->cap_used / be_lun->cbe_lun.blocksize; } else if (strcmp(attrname, "blocksavail") == 0) { val = (be_lun->cap_bytes - be_lun->cap_used) / be_lun->cbe_lun.blocksize; } sx_sunlock(&be_lun->page_lock); return (val); } static int ctl_backend_ramdisk_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td) { struct ctl_be_ramdisk_softc *softc = &rd_softc; struct ctl_lun_req *lun_req; int retval; retval = 0; switch (cmd) { case CTL_LUN_REQ: lun_req = (struct ctl_lun_req *)addr; switch (lun_req->reqtype) { case CTL_LUNREQ_CREATE: retval = ctl_backend_ramdisk_create(softc, lun_req); break; case CTL_LUNREQ_RM: retval = ctl_backend_ramdisk_rm(softc, lun_req); break; case CTL_LUNREQ_MODIFY: retval = ctl_backend_ramdisk_modify(softc, lun_req); break; default: lun_req->status = CTL_LUN_ERROR; snprintf(lun_req->error_str, sizeof(lun_req->error_str), "%s: invalid LUN request type %d", __func__, lun_req->reqtype); break; } break; default: retval = ENOTTY; break; } return (retval); } static int ctl_backend_ramdisk_rm(struct ctl_be_ramdisk_softc *softc, struct ctl_lun_req *req) { struct ctl_be_ramdisk_lun *be_lun; struct ctl_lun_rm_params *params; int retval; params = &req->reqdata.rm; mtx_lock(&softc->lock); STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->cbe_lun.lun_id == params->lun_id) break; } mtx_unlock(&softc->lock); if (be_lun == NULL) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN %u is not managed by the ramdisk backend", __func__, params->lun_id); goto bailout_error; } retval = ctl_disable_lun(&be_lun->cbe_lun); if (retval != 0) { snprintf(req->error_str, sizeof(req->error_str), "%s: error %d returned from ctl_disable_lun() for " "LUN %d", __func__, retval, params->lun_id); goto bailout_error; } /* * Set the waiting flag before we invalidate the LUN. Our shutdown * routine can be called any time after we invalidate the LUN, * and can be called from our context. * * This tells the shutdown routine that we're waiting, or we're * going to wait for the shutdown to happen. */ mtx_lock(&softc->lock); be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING; mtx_unlock(&softc->lock); retval = ctl_invalidate_lun(&be_lun->cbe_lun); if (retval != 0) { snprintf(req->error_str, sizeof(req->error_str), "%s: error %d returned from ctl_invalidate_lun() for " "LUN %d", __func__, retval, params->lun_id); mtx_lock(&softc->lock); be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING; mtx_unlock(&softc->lock); goto bailout_error; } mtx_lock(&softc->lock); while ((be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) == 0) { retval = msleep(be_lun, &softc->lock, PCATCH, "ctlram", 0); if (retval == EINTR) break; } be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING; /* * We only remove this LUN from the list and free it (below) if * retval == 0. If the user interrupted the wait, we just bail out * without actually freeing the LUN. We let the shutdown routine * free the LUN if that happens. */ if (retval == 0) { STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_ramdisk_lun, links); softc->num_luns--; } mtx_unlock(&softc->lock); if (retval == 0) { taskqueue_drain_all(be_lun->io_taskqueue); taskqueue_free(be_lun->io_taskqueue); - ctl_free_opts(&be_lun->cbe_lun.options); + nvlist_destroy(be_lun->cbe_lun.options); free(be_lun->zero_page, M_RAMDISK); ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir); sx_destroy(&be_lun->page_lock); mtx_destroy(&be_lun->queue_lock); free(be_lun, M_RAMDISK); } req->status = CTL_LUN_OK; return (retval); bailout_error: req->status = CTL_LUN_ERROR; return (0); } static int ctl_backend_ramdisk_create(struct ctl_be_ramdisk_softc *softc, struct ctl_lun_req *req) { struct ctl_be_ramdisk_lun *be_lun; struct ctl_be_lun *cbe_lun; struct ctl_lun_create_params *params; - char *value; + const char *value; char tmpstr[32]; uint64_t t; int retval; retval = 0; params = &req->reqdata.create; be_lun = malloc(sizeof(*be_lun), M_RAMDISK, M_ZERO | M_WAITOK); cbe_lun = &be_lun->cbe_lun; cbe_lun->be_lun = be_lun; + cbe_lun->options = nvlist_clone(req->args_nvl); be_lun->params = req->reqdata.create; be_lun->softc = softc; sprintf(be_lun->lunname, "cram%d", softc->num_luns); - ctl_init_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); if (params->flags & CTL_LUN_FLAG_DEV_TYPE) cbe_lun->lun_type = params->device_type; else cbe_lun->lun_type = T_DIRECT; be_lun->flags = CTL_BE_RAMDISK_LUN_UNCONFIGURED; cbe_lun->flags = 0; - value = ctl_get_opt(&cbe_lun->options, "ha_role"); + value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); if (value != NULL) { if (strcmp(value, "primary") == 0) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; be_lun->pblocksize = PAGE_SIZE; - value = ctl_get_opt(&cbe_lun->options, "pblocksize"); + value = dnvlist_get_string(cbe_lun->options, "pblocksize", NULL); if (value != NULL) { ctl_expand_number(value, &t); be_lun->pblocksize = t; } if (be_lun->pblocksize < 512 || be_lun->pblocksize > 131072) { snprintf(req->error_str, sizeof(req->error_str), "%s: unsupported pblocksize %u", __func__, be_lun->pblocksize); goto bailout_error; } if (cbe_lun->lun_type == T_DIRECT || cbe_lun->lun_type == T_CDROM) { if (params->blocksize_bytes != 0) cbe_lun->blocksize = params->blocksize_bytes; else if (cbe_lun->lun_type == T_CDROM) cbe_lun->blocksize = 2048; else cbe_lun->blocksize = 512; be_lun->pblockmul = be_lun->pblocksize / cbe_lun->blocksize; if (be_lun->pblockmul < 1 || !powerof2(be_lun->pblockmul)) { snprintf(req->error_str, sizeof(req->error_str), "%s: pblocksize %u not exp2 of blocksize %u", __func__, be_lun->pblocksize, cbe_lun->blocksize); goto bailout_error; } if (params->lun_size_bytes < cbe_lun->blocksize) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN size %ju < blocksize %u", __func__, params->lun_size_bytes, cbe_lun->blocksize); goto bailout_error; } be_lun->size_blocks = params->lun_size_bytes / cbe_lun->blocksize; be_lun->size_bytes = be_lun->size_blocks * cbe_lun->blocksize; be_lun->indir = 0; t = be_lun->size_bytes / be_lun->pblocksize; while (t > 1) { t /= PPP; be_lun->indir++; } cbe_lun->maxlba = be_lun->size_blocks - 1; cbe_lun->pblockexp = fls(be_lun->pblockmul) - 1; cbe_lun->pblockoff = 0; cbe_lun->ublockexp = cbe_lun->pblockexp; cbe_lun->ublockoff = 0; cbe_lun->atomicblock = be_lun->pblocksize; cbe_lun->opttxferlen = SGPP * be_lun->pblocksize; - value = ctl_get_opt(&cbe_lun->options, "capacity"); + value = dnvlist_get_string(cbe_lun->options, "capacity", NULL); if (value != NULL) ctl_expand_number(value, &be_lun->cap_bytes); } else { be_lun->pblockmul = 1; cbe_lun->pblockexp = 0; } /* Tell the user the blocksize we ended up using */ params->blocksize_bytes = cbe_lun->blocksize; params->lun_size_bytes = be_lun->size_bytes; - value = ctl_get_opt(&cbe_lun->options, "unmap"); - if (value == NULL || strcmp(value, "off") != 0) + value = dnvlist_get_string(cbe_lun->options, "unmap", NULL); + if (value != NULL && strcmp(value, "off") != 0) cbe_lun->flags |= CTL_LUN_FLAG_UNMAP; - value = ctl_get_opt(&cbe_lun->options, "readonly"); + value = dnvlist_get_string(cbe_lun->options, "readonly", NULL); if (value != NULL) { if (strcmp(value, "on") == 0) cbe_lun->flags |= CTL_LUN_FLAG_READONLY; } else if (cbe_lun->lun_type != T_DIRECT) cbe_lun->flags |= CTL_LUN_FLAG_READONLY; cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; - value = ctl_get_opt(&cbe_lun->options, "serseq"); + value = dnvlist_get_string(cbe_lun->options, "serseq", NULL); if (value != NULL && strcmp(value, "on") == 0) cbe_lun->serseq = CTL_LUN_SERSEQ_ON; else if (value != NULL && strcmp(value, "read") == 0) cbe_lun->serseq = CTL_LUN_SERSEQ_READ; else if (value != NULL && strcmp(value, "off") == 0) cbe_lun->serseq = CTL_LUN_SERSEQ_OFF; if (params->flags & CTL_LUN_FLAG_ID_REQ) { cbe_lun->req_lun_id = params->req_lun_id; cbe_lun->flags |= CTL_LUN_FLAG_ID_REQ; } else cbe_lun->req_lun_id = 0; cbe_lun->lun_shutdown = ctl_backend_ramdisk_lun_shutdown; cbe_lun->lun_config_status = ctl_backend_ramdisk_lun_config_status; cbe_lun->be = &ctl_be_ramdisk_driver; if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%04d", softc->num_luns); strncpy((char *)cbe_lun->serial_num, tmpstr, MIN(sizeof(cbe_lun->serial_num), sizeof(tmpstr))); /* Tell the user what we used for a serial number */ strncpy((char *)params->serial_num, tmpstr, MIN(sizeof(params->serial_num), sizeof(tmpstr))); } else { strncpy((char *)cbe_lun->serial_num, params->serial_num, MIN(sizeof(cbe_lun->serial_num), sizeof(params->serial_num))); } if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%04d", softc->num_luns); strncpy((char *)cbe_lun->device_id, tmpstr, MIN(sizeof(cbe_lun->device_id), sizeof(tmpstr))); /* Tell the user what we used for a device ID */ strncpy((char *)params->device_id, tmpstr, MIN(sizeof(params->device_id), sizeof(tmpstr))); } else { strncpy((char *)cbe_lun->device_id, params->device_id, MIN(sizeof(cbe_lun->device_id), sizeof(params->device_id))); } STAILQ_INIT(&be_lun->cont_queue); sx_init(&be_lun->page_lock, "cram page lock"); if (be_lun->cap_bytes == 0) { be_lun->indir = 0; be_lun->pages = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK); } be_lun->zero_page = malloc(be_lun->pblocksize, M_RAMDISK, M_WAITOK|M_ZERO); mtx_init(&be_lun->queue_lock, "cram queue lock", NULL, MTX_DEF); TASK_INIT(&be_lun->io_task, /*priority*/0, ctl_backend_ramdisk_worker, be_lun); be_lun->io_taskqueue = taskqueue_create(be_lun->lunname, M_WAITOK, taskqueue_thread_enqueue, /*context*/&be_lun->io_taskqueue); if (be_lun->io_taskqueue == NULL) { snprintf(req->error_str, sizeof(req->error_str), "%s: Unable to create taskqueue", __func__); goto bailout_error; } retval = taskqueue_start_threads(&be_lun->io_taskqueue, /*num threads*/1, /*priority*/PWAIT, /*thread name*/ "%s taskq", be_lun->lunname); if (retval != 0) goto bailout_error; mtx_lock(&softc->lock); softc->num_luns++; STAILQ_INSERT_TAIL(&softc->lun_list, be_lun, links); mtx_unlock(&softc->lock); retval = ctl_add_lun(&be_lun->cbe_lun); if (retval != 0) { mtx_lock(&softc->lock); STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_ramdisk_lun, links); softc->num_luns--; mtx_unlock(&softc->lock); snprintf(req->error_str, sizeof(req->error_str), "%s: ctl_add_lun() returned error %d, see dmesg for " "details", __func__, retval); retval = 0; goto bailout_error; } mtx_lock(&softc->lock); /* * Tell the config_status routine that we're waiting so it won't * clean up the LUN in the event of an error. */ be_lun->flags |= CTL_BE_RAMDISK_LUN_WAITING; while (be_lun->flags & CTL_BE_RAMDISK_LUN_UNCONFIGURED) { retval = msleep(be_lun, &softc->lock, PCATCH, "ctlram", 0); if (retval == EINTR) break; } be_lun->flags &= ~CTL_BE_RAMDISK_LUN_WAITING; if (be_lun->flags & CTL_BE_RAMDISK_LUN_CONFIG_ERR) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN configuration error, see dmesg for details", __func__); STAILQ_REMOVE(&softc->lun_list, be_lun, ctl_be_ramdisk_lun, links); softc->num_luns--; mtx_unlock(&softc->lock); goto bailout_error; } else { params->req_lun_id = cbe_lun->lun_id; } mtx_unlock(&softc->lock); req->status = CTL_LUN_OK; return (retval); bailout_error: req->status = CTL_LUN_ERROR; if (be_lun != NULL) { if (be_lun->io_taskqueue != NULL) taskqueue_free(be_lun->io_taskqueue); - ctl_free_opts(&cbe_lun->options); + nvlist_destroy(cbe_lun->options); free(be_lun->zero_page, M_RAMDISK); ctl_backend_ramdisk_freeallpages(be_lun->pages, be_lun->indir); sx_destroy(&be_lun->page_lock); mtx_destroy(&be_lun->queue_lock); free(be_lun, M_RAMDISK); } return (retval); } static int ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc, struct ctl_lun_req *req) { struct ctl_be_ramdisk_lun *be_lun; struct ctl_be_lun *cbe_lun; struct ctl_lun_modify_params *params; - char *value; + const char *value; uint32_t blocksize; int wasprim; params = &req->reqdata.modify; mtx_lock(&softc->lock); STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->cbe_lun.lun_id == params->lun_id) break; } mtx_unlock(&softc->lock); if (be_lun == NULL) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN %u is not managed by the ramdisk backend", __func__, params->lun_id); goto bailout_error; } cbe_lun = &be_lun->cbe_lun; if (params->lun_size_bytes != 0) be_lun->params.lun_size_bytes = params->lun_size_bytes; - ctl_update_opts(&cbe_lun->options, req->num_be_args, req->kern_be_args); + nvlist_destroy(cbe_lun->options); + cbe_lun->options = nvlist_clone(req->args_nvl); + wasprim = (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY); - value = ctl_get_opt(&cbe_lun->options, "ha_role"); + value = dnvlist_get_string(cbe_lun->options, "ha_role", NULL); if (value != NULL) { if (strcmp(value, "primary") == 0) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; else cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; } else if (control_softc->flags & CTL_FLAG_ACTIVE_SHELF) cbe_lun->flags |= CTL_LUN_FLAG_PRIMARY; else cbe_lun->flags &= ~CTL_LUN_FLAG_PRIMARY; if (wasprim != (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY)) { if (cbe_lun->flags & CTL_LUN_FLAG_PRIMARY) ctl_lun_primary(cbe_lun); else ctl_lun_secondary(cbe_lun); } blocksize = be_lun->cbe_lun.blocksize; if (be_lun->params.lun_size_bytes < blocksize) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN size %ju < blocksize %u", __func__, be_lun->params.lun_size_bytes, blocksize); goto bailout_error; } be_lun->size_blocks = be_lun->params.lun_size_bytes / blocksize; be_lun->size_bytes = be_lun->size_blocks * blocksize; be_lun->cbe_lun.maxlba = be_lun->size_blocks - 1; ctl_lun_capacity_changed(&be_lun->cbe_lun); /* Tell the user the exact size we ended up using */ params->lun_size_bytes = be_lun->size_bytes; req->status = CTL_LUN_OK; return (0); bailout_error: req->status = CTL_LUN_ERROR; return (0); } static void ctl_backend_ramdisk_lun_shutdown(void *be_lun) { struct ctl_be_ramdisk_lun *lun = be_lun; struct ctl_be_ramdisk_softc *softc = lun->softc; mtx_lock(&softc->lock); lun->flags |= CTL_BE_RAMDISK_LUN_UNCONFIGURED; if (lun->flags & CTL_BE_RAMDISK_LUN_WAITING) { wakeup(lun); } else { STAILQ_REMOVE(&softc->lun_list, lun, ctl_be_ramdisk_lun, links); softc->num_luns--; free(be_lun, M_RAMDISK); } mtx_unlock(&softc->lock); } static void ctl_backend_ramdisk_lun_config_status(void *be_lun, ctl_lun_config_status status) { struct ctl_be_ramdisk_lun *lun; struct ctl_be_ramdisk_softc *softc; lun = (struct ctl_be_ramdisk_lun *)be_lun; softc = lun->softc; if (status == CTL_LUN_CONFIG_OK) { mtx_lock(&softc->lock); lun->flags &= ~CTL_BE_RAMDISK_LUN_UNCONFIGURED; if (lun->flags & CTL_BE_RAMDISK_LUN_WAITING) wakeup(lun); mtx_unlock(&softc->lock); /* * We successfully added the LUN, attempt to enable it. */ if (ctl_enable_lun(&lun->cbe_lun) != 0) { printf("%s: ctl_enable_lun() failed!\n", __func__); if (ctl_invalidate_lun(&lun->cbe_lun) != 0) { printf("%s: ctl_invalidate_lun() failed!\n", __func__); } } return; } mtx_lock(&softc->lock); lun->flags &= ~CTL_BE_RAMDISK_LUN_UNCONFIGURED; /* * If we have a user waiting, let him handle the cleanup. If not, * clean things up here. */ if (lun->flags & CTL_BE_RAMDISK_LUN_WAITING) { lun->flags |= CTL_BE_RAMDISK_LUN_CONFIG_ERR; wakeup(lun); } else { STAILQ_REMOVE(&softc->lun_list, lun, ctl_be_ramdisk_lun, links); softc->num_luns--; free(lun, M_RAMDISK); } mtx_unlock(&softc->lock); } Index: head/sys/cam/ctl/ctl_frontend.c =================================================================== --- head/sys/cam/ctl/ctl_frontend.c (revision 333445) +++ head/sys/cam/ctl/ctl_frontend.c (revision 333446) @@ -1,391 +1,393 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003 Silicon Graphics International Corp. * Copyright (c) 2014-2017 Alexander Motin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_frontend.c#4 $ */ /* * CAM Target Layer front end interface code * * Author: Ken Merry */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include /* XXX KDM move defines from ctl_ioctl.h to somewhere else */ #include #include #include #include extern struct ctl_softc *control_softc; int ctl_frontend_register(struct ctl_frontend *fe) { struct ctl_softc *softc = control_softc; struct ctl_frontend *fe_tmp; int error; KASSERT(softc != NULL, ("CTL is not initialized")); /* Sanity check, make sure this isn't a duplicate registration. */ mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(fe_tmp, &softc->fe_list, links) { if (strcmp(fe_tmp->name, fe->name) == 0) { mtx_unlock(&softc->ctl_lock); return (-1); } } mtx_unlock(&softc->ctl_lock); STAILQ_INIT(&fe->port_list); /* Call the frontend's initialization routine. */ if (fe->init != NULL) { if ((error = fe->init()) != 0) { printf("%s frontend init error: %d\n", fe->name, error); return (error); } } mtx_lock(&softc->ctl_lock); softc->num_frontends++; STAILQ_INSERT_TAIL(&softc->fe_list, fe, links); mtx_unlock(&softc->ctl_lock); return (0); } int ctl_frontend_deregister(struct ctl_frontend *fe) { struct ctl_softc *softc = control_softc; int error; /* Call the frontend's shutdown routine.*/ if (fe->shutdown != NULL) { if ((error = fe->shutdown()) != 0) { printf("%s frontend shutdown error: %d\n", fe->name, error); return (error); } } mtx_lock(&softc->ctl_lock); STAILQ_REMOVE(&softc->fe_list, fe, ctl_frontend, links); softc->num_frontends--; mtx_unlock(&softc->ctl_lock); return (0); } struct ctl_frontend * ctl_frontend_find(char *frontend_name) { struct ctl_softc *softc = control_softc; struct ctl_frontend *fe; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(fe, &softc->fe_list, links) { if (strcmp(fe->name, frontend_name) == 0) { mtx_unlock(&softc->ctl_lock); return (fe); } } mtx_unlock(&softc->ctl_lock); return (NULL); } int ctl_port_register(struct ctl_port *port) { struct ctl_softc *softc = control_softc; struct ctl_port *tport, *nport; void *pool; int port_num; int retval; KASSERT(softc != NULL, ("CTL is not initialized")); port->ctl_softc = softc; mtx_lock(&softc->ctl_lock); if (port->targ_port >= 0) port_num = port->targ_port; else port_num = ctl_ffz(softc->ctl_port_mask, softc->port_min, softc->port_max); if ((port_num < 0) || (ctl_set_mask(softc->ctl_port_mask, port_num) < 0)) { mtx_unlock(&softc->ctl_lock); return (1); } softc->num_ports++; mtx_unlock(&softc->ctl_lock); /* * Initialize the initiator and portname mappings */ port->max_initiators = CTL_MAX_INIT_PER_PORT; port->wwpn_iid = malloc(sizeof(*port->wwpn_iid) * port->max_initiators, M_CTL, M_NOWAIT | M_ZERO); if (port->wwpn_iid == NULL) { retval = ENOMEM; goto error; } /* * We add 20 to whatever the caller requests, so he doesn't get * burned by queueing things back to the pending sense queue. In * theory, there should probably only be one outstanding item, at * most, on the pending sense queue for a LUN. We'll clear the * pending sense queue on the next command, whether or not it is * a REQUEST SENSE. */ retval = ctl_pool_create(softc, port->port_name, port->num_requested_ctl_io + 20, &pool); if (retval != 0) { free(port->wwpn_iid, M_CTL); error: port->targ_port = -1; mtx_lock(&softc->ctl_lock); ctl_clear_mask(softc->ctl_port_mask, port_num); mtx_unlock(&softc->ctl_lock); return (retval); } port->targ_port = port_num; port->ctl_pool_ref = pool; - if (port->options.stqh_first == NULL) - STAILQ_INIT(&port->options); + if (port->options == NULL) + port->options = nvlist_create(0); port->stats.item = port_num; mtx_init(&port->port_lock, "CTL port", NULL, MTX_DEF); mtx_lock(&softc->ctl_lock); STAILQ_INSERT_TAIL(&port->frontend->port_list, port, fe_links); for (tport = NULL, nport = STAILQ_FIRST(&softc->port_list); nport != NULL && nport->targ_port < port_num; tport = nport, nport = STAILQ_NEXT(tport, links)) { } if (tport) STAILQ_INSERT_AFTER(&softc->port_list, tport, port, links); else STAILQ_INSERT_HEAD(&softc->port_list, port, links); softc->ctl_ports[port->targ_port] = port; mtx_unlock(&softc->ctl_lock); return (retval); } int ctl_port_deregister(struct ctl_port *port) { struct ctl_softc *softc = port->ctl_softc; struct ctl_io_pool *pool = (struct ctl_io_pool *)port->ctl_pool_ref; int i; if (port->targ_port == -1) return (1); mtx_lock(&softc->ctl_lock); STAILQ_REMOVE(&softc->port_list, port, ctl_port, links); STAILQ_REMOVE(&port->frontend->port_list, port, ctl_port, fe_links); softc->num_ports--; ctl_clear_mask(softc->ctl_port_mask, port->targ_port); softc->ctl_ports[port->targ_port] = NULL; mtx_unlock(&softc->ctl_lock); ctl_pool_free(pool); - ctl_free_opts(&port->options); + nvlist_destroy(port->options); ctl_lun_map_deinit(port); free(port->port_devid, M_CTL); port->port_devid = NULL; free(port->target_devid, M_CTL); port->target_devid = NULL; free(port->init_devid, M_CTL); port->init_devid = NULL; for (i = 0; i < port->max_initiators; i++) free(port->wwpn_iid[i].name, M_CTL); free(port->wwpn_iid, M_CTL); mtx_destroy(&port->port_lock); return (0); } void ctl_port_set_wwns(struct ctl_port *port, int wwnn_valid, uint64_t wwnn, int wwpn_valid, uint64_t wwpn) { struct scsi_vpd_id_descriptor *desc; int len, proto; if (port->port_type == CTL_PORT_FC) proto = SCSI_PROTO_FC << 4; else if (port->port_type == CTL_PORT_SAS) proto = SCSI_PROTO_SAS << 4; else if (port->port_type == CTL_PORT_ISCSI) proto = SCSI_PROTO_ISCSI << 4; else proto = SCSI_PROTO_SPI << 4; if (wwnn_valid) { port->wwnn = wwnn; free(port->target_devid, M_CTL); len = sizeof(struct scsi_vpd_device_id) + CTL_WWPN_LEN; port->target_devid = malloc(sizeof(struct ctl_devid) + len, M_CTL, M_WAITOK | M_ZERO); port->target_devid->len = len; desc = (struct scsi_vpd_id_descriptor *)port->target_devid->data; desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_TARGET | SVPD_ID_TYPE_NAA; desc->length = CTL_WWPN_LEN; scsi_u64to8b(port->wwnn, desc->identifier); } if (wwpn_valid) { port->wwpn = wwpn; free(port->port_devid, M_CTL); len = sizeof(struct scsi_vpd_device_id) + CTL_WWPN_LEN; port->port_devid = malloc(sizeof(struct ctl_devid) + len, M_CTL, M_WAITOK | M_ZERO); port->port_devid->len = len; desc = (struct scsi_vpd_id_descriptor *)port->port_devid->data; desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_NAA; desc->length = CTL_WWPN_LEN; scsi_u64to8b(port->wwpn, desc->identifier); } } void ctl_port_online(struct ctl_port *port) { struct ctl_softc *softc = port->ctl_softc; struct ctl_lun *lun; const char *value; uint32_t l; if (port->lun_enable != NULL) { if (port->lun_map) { for (l = 0; l < port->lun_map_size; l++) { if (ctl_lun_map_from_port(port, l) == UINT32_MAX) continue; port->lun_enable(port->targ_lun_arg, l); } } else { STAILQ_FOREACH(lun, &softc->lun_list, links) port->lun_enable(port->targ_lun_arg, lun->lun); } } if (port->port_online != NULL) port->port_online(port->onoff_arg); mtx_lock(&softc->ctl_lock); if (softc->is_single == 0) { - value = ctl_get_opt(&port->options, "ha_shared"); + value = dnvlist_get_string(port->options, "ha_shared", NULL); if (value != NULL && strcmp(value, "on") == 0) port->status |= CTL_PORT_STATUS_HA_SHARED; else port->status &= ~CTL_PORT_STATUS_HA_SHARED; } port->status |= CTL_PORT_STATUS_ONLINE; STAILQ_FOREACH(lun, &softc->lun_list, links) { if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; mtx_lock(&lun->lun_lock); ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE); mtx_unlock(&lun->lun_lock); } mtx_unlock(&softc->ctl_lock); ctl_isc_announce_port(port); } void ctl_port_offline(struct ctl_port *port) { struct ctl_softc *softc = port->ctl_softc; struct ctl_lun *lun; uint32_t l; if (port->port_offline != NULL) port->port_offline(port->onoff_arg); if (port->lun_disable != NULL) { if (port->lun_map) { for (l = 0; l < port->lun_map_size; l++) { if (ctl_lun_map_from_port(port, l) == UINT32_MAX) continue; port->lun_disable(port->targ_lun_arg, l); } } else { STAILQ_FOREACH(lun, &softc->lun_list, links) port->lun_disable(port->targ_lun_arg, lun->lun); } } mtx_lock(&softc->ctl_lock); port->status &= ~CTL_PORT_STATUS_ONLINE; STAILQ_FOREACH(lun, &softc->lun_list, links) { if (ctl_lun_map_to_port(port, lun->lun) == UINT32_MAX) continue; mtx_lock(&lun->lun_lock); ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE); mtx_unlock(&lun->lun_lock); } mtx_unlock(&softc->ctl_lock); ctl_isc_announce_port(port); } /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_frontend.h =================================================================== --- head/sys/cam/ctl/ctl_frontend.h (revision 333445) +++ head/sys/cam/ctl/ctl_frontend.h (revision 333446) @@ -1,338 +1,339 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003 Silicon Graphics International Corp. * Copyright (c) 2014-2017 Alexander Motin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_frontend.h#2 $ * $FreeBSD$ */ /* * CAM Target Layer front end registration hooks * * Author: Ken Merry */ #ifndef _CTL_FRONTEND_H_ #define _CTL_FRONTEND_H_ #include +#include typedef enum { CTL_PORT_STATUS_NONE = 0x00, CTL_PORT_STATUS_ONLINE = 0x01, CTL_PORT_STATUS_HA_SHARED = 0x02 } ctl_port_status; typedef int (*fe_init_t)(void); typedef int (*fe_shutdown_t)(void); typedef void (*port_func_t)(void *onoff_arg); typedef int (*port_info_func_t)(void *onoff_arg, struct sbuf *sb); typedef int (*lun_func_t)(void *arg, int lun_id); typedef int (*fe_ioctl_t)(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); #define CTL_FRONTEND_DECLARE(name, driver) \ static int name ## _modevent(module_t mod, int type, void *data) \ { \ switch (type) { \ case MOD_LOAD: \ return (ctl_frontend_register( \ (struct ctl_frontend *)data)); \ break; \ case MOD_UNLOAD: \ return (ctl_frontend_deregister( \ (struct ctl_frontend *)data)); \ break; \ default: \ return EOPNOTSUPP; \ } \ return 0; \ } \ static moduledata_t name ## _mod = { \ #name, \ name ## _modevent, \ (void *)&driver \ }; \ DECLARE_MODULE(name, name ## _mod, SI_SUB_CONFIGURE, SI_ORDER_FOURTH); \ MODULE_DEPEND(name, ctl, 1, 1, 1); \ MODULE_DEPEND(name, cam, 1, 1, 1) struct ctl_wwpn_iid { int in_use; time_t last_use; uint64_t wwpn; char *name; }; /* * The ctl_frontend structure is the registration mechanism between a FETD * (Front End Target Driver) and the CTL layer. Here is a description of * the fields: * * port_type: This field tells CTL what kind of front end it is * dealing with. This field serves two purposes. * The first is to let CTL know whether the frontend * in question is inside the main CTL module (i.e. * the ioctl front end), and therefore its module * reference count shouldn't be incremented. The * CTL ioctl front end should continue to use the * CTL_PORT_IOCTL argument as long as it is part of * the main CTL module. The second is to let CTL * know what kind of front end it is dealing with, so * it can return the proper inquiry data for that * particular port. * * num_requested_ctl_io: This is the number of ctl_io structures that the * front end needs for its pool. This should * generally be the maximum number of outstanding * transactions that the FETD can handle. The CTL * layer will add a few to this to account for * ctl_io buffers queued for pending sense data. * (Pending sense only gets queued if the FETD * doesn't support autosense. e.g. non-packetized * parallel SCSI doesn't support autosense.) * * port_name: A string describing the FETD. e.g. "LSI 1030T U320" * or whatever you want to use to describe the driver. * * physical_port: This is the physical port number of this * particular port within the driver/hardware. This * number is hardware/driver specific. * virtual_port: This is the virtual port number of this * particular port. This is for things like NP-IV. * * port_online(): This function is called, with onoff_arg as its * argument, by the CTL layer when it wants the FETD * to start responding to selections on the specified * target ID. * * port_offline(): This function is called, with onoff_arg as its * argument, by the CTL layer when it wants the FETD * to stop responding to selection on the specified * target ID. * * onoff_arg: This is supplied as an argument to port_online() * and port_offline(). This is specified by the * FETD. * * lun_enable(): This function is called, with targ_lun_arg, a target * ID and a LUN ID as its arguments, by CTL when it * wants the FETD to enable a particular LUN. If the * FETD doesn't really know about LUNs, it should * just ignore this call and return 0. If the FETD * cannot enable the requested LUN for some reason, the * FETD should return non-zero status. * * lun_disable(): This function is called, with targ_lun_arg, a target * ID and LUN ID as its arguments, by CTL when it * wants the FETD to disable a particular LUN. If the * FETD doesn't really know about LUNs, it should just * ignore this call and return 0. If the FETD cannot * disable the requested LUN for some reason, the * FETD should return non-zero status. * * targ_lun_arg: This is supplied as an argument to the targ/lun * enable/disable() functions. This is specified by * the FETD. * * fe_datamove(): This function is called one or more times per I/O * by the CTL layer to tell the FETD to initiate a * DMA to or from the data buffer(s) specified by * the passed-in ctl_io structure. * * fe_done(): This function is called by the CTL layer when a * particular SCSI I/O or task management command has * completed. For SCSI I/O requests (CTL_IO_SCSI), * sense data is always supplied if the status is * CTL_SCSI_ERROR and the SCSI status byte is * SCSI_STATUS_CHECK_COND. If the FETD doesn't * support autosense, the sense should be queued * back to the CTL layer via ctl_queue_sense(). * * fe_dump(): This function, if it exists, is called by CTL * to request a dump of any debugging information or * state to the console. * * targ_port: The CTL layer assigns a "port number" to every * FETD. This port number should be passed back in * in the header of every ctl_io that is queued to * the CTL layer. This enables us to determine * which bus the command came in on. * * ctl_pool_ref: Memory pool reference used by the FETD in calls to * ctl_alloc_io(). * * max_initiators: Maximum number of initiators that the FETD is * allowed to have. Initiators should be numbered * from 0 to max_initiators - 1. This value will * typically be 16, and thus not a problem for * parallel SCSI. This may present issues for Fibre * Channel. * * wwnn World Wide Node Name to be used by the FETD. * Note that this is set *after* registration. It * will be set prior to the online function getting * called. * * wwpn World Wide Port Name to be used by the FETD. * Note that this is set *after* registration. It * will be set prior to the online function getting * called. * * status: Used by CTL to keep track of per-FETD state. * * links: Linked list pointers, used by CTL. The FETD * shouldn't touch this field. */ struct ctl_port { struct ctl_softc *ctl_softc; struct ctl_frontend *frontend; ctl_port_type port_type; /* passed to CTL */ int num_requested_ctl_io; /* passed to CTL */ char *port_name; /* passed to CTL */ int physical_port; /* passed to CTL */ int virtual_port; /* passed to CTL */ port_func_t port_online; /* passed to CTL */ port_func_t port_offline; /* passed to CTL */ port_info_func_t port_info; /* passed to CTL */ void *onoff_arg; /* passed to CTL */ lun_func_t lun_enable; /* passed to CTL */ lun_func_t lun_disable; /* passed to CTL */ int lun_map_size; /* passed to CTL */ uint32_t *lun_map; /* passed to CTL */ void *targ_lun_arg; /* passed to CTL */ void (*fe_datamove)(union ctl_io *io); /* passed to CTL */ void (*fe_done)(union ctl_io *io); /* passed to CTL */ int32_t targ_port; /* passed back to FETD */ void *ctl_pool_ref; /* passed back to FETD */ uint32_t max_initiators; /* passed back to FETD */ struct ctl_wwpn_iid *wwpn_iid; /* used by CTL */ uint64_t wwnn; /* set by CTL before online */ uint64_t wwpn; /* set by CTL before online */ ctl_port_status status; /* used by CTL */ - ctl_options_t options; /* passed to CTL */ + nvlist_t *options; /* passed to CTL */ struct ctl_devid *port_devid; /* passed to CTL */ struct ctl_devid *target_devid; /* passed to CTL */ struct ctl_devid *init_devid; /* passed to CTL */ struct ctl_io_stats stats; /* used by CTL */ struct mtx port_lock; /* used by CTL */ STAILQ_ENTRY(ctl_port) fe_links; /* used by CTL */ STAILQ_ENTRY(ctl_port) links; /* used by CTL */ }; struct ctl_frontend { char name[CTL_DRIVER_NAME_LEN]; /* passed to CTL */ fe_init_t init; /* passed to CTL */ fe_ioctl_t ioctl; /* passed to CTL */ void (*fe_dump)(void); /* passed to CTL */ fe_shutdown_t shutdown; /* passed to CTL */ STAILQ_HEAD(, ctl_port) port_list; /* used by CTL */ STAILQ_ENTRY(ctl_frontend) links; /* used by CTL */ }; /* * This may block until resources are allocated. Called at FETD module load * time. Returns 0 for success, non-zero for failure. */ int ctl_frontend_register(struct ctl_frontend *fe); /* * Called at FETD module unload time. * Returns 0 for success, non-zero for failure. */ int ctl_frontend_deregister(struct ctl_frontend *fe); /* * Find the frontend by its name. Returns NULL if not found. */ struct ctl_frontend * ctl_frontend_find(char *frontend_name); /* * This may block until resources are allocated. Called at FETD module load * time. Returns 0 for success, non-zero for failure. */ int ctl_port_register(struct ctl_port *port); /* * Called at FETD module unload time. * Returns 0 for success, non-zero for failure. */ int ctl_port_deregister(struct ctl_port *port); /* * Called to set the WWNN and WWPN for a particular frontend. */ void ctl_port_set_wwns(struct ctl_port *port, int wwnn_valid, uint64_t wwnn, int wwpn_valid, uint64_t wwpn); /* * Called to bring a particular frontend online. */ void ctl_port_online(struct ctl_port *fe); /* * Called to take a particular frontend offline. */ void ctl_port_offline(struct ctl_port *fe); /* * This routine queues I/O and task management requests from the FETD to the * CTL layer. Returns immediately. Returns 0 for success, non-zero for * failure. */ int ctl_queue(union ctl_io *io); /* * This routine is used if the front end interface doesn't support * autosense (e.g. non-packetized parallel SCSI). This will queue the * scsiio structure back to a per-lun pending sense queue. This MUST be * called BEFORE any request sense can get queued to the CTL layer -- I * need it in the queue in order to service the request. The scsiio * structure passed in here will be freed by the CTL layer when sense is * retrieved by the initiator. Returns 0 for success, non-zero for failure. */ int ctl_queue_sense(union ctl_io *io); /* * This routine adds an initiator to CTL's port database. * The iid field should be the same as the iid passed in the nexus of each * ctl_io from this initiator. * The WWPN should be the FC WWPN, if available. */ int ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name); /* * This routine will remove an initiator from CTL's port database. * The iid field should be the same as the iid passed in the nexus of each * ctl_io from this initiator. */ int ctl_remove_initiator(struct ctl_port *port, int iid); #endif /* _CTL_FRONTEND_H_ */ Index: head/sys/cam/ctl/ctl_frontend_ioctl.c =================================================================== --- head/sys/cam/ctl/ctl_frontend_ioctl.c (revision 333445) +++ head/sys/cam/ctl/ctl_frontend_ioctl.c (revision 333446) @@ -1,434 +1,644 @@ /*- + * SPDX-License-Identifier: BSD-2-Clause-FreeBSD + * * Copyright (c) 2003-2009 Silicon Graphics International Corp. * Copyright (c) 2012 The FreeBSD Foundation * Copyright (c) 2015 Alexander Motin + * Copyright (c) 2017 Jakub Wojciech Klama * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer, * without modification, immediately at the beginning of the file. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include typedef enum { CTL_IOCTL_INPROG, CTL_IOCTL_DATAMOVE, CTL_IOCTL_DONE } ctl_fe_ioctl_state; struct ctl_fe_ioctl_params { struct cv sem; struct mtx ioctl_mtx; ctl_fe_ioctl_state state; }; -struct cfi_softc { +struct cfi_port { + TAILQ_ENTRY(cfi_port) link; uint32_t cur_tag_num; + struct cdev * dev; struct ctl_port port; }; +struct cfi_softc { + TAILQ_HEAD(, cfi_port) ports; +}; + + static struct cfi_softc cfi_softc; + static int cfi_init(void); static int cfi_shutdown(void); static void cfi_datamove(union ctl_io *io); static void cfi_done(union ctl_io *io); +static int cfi_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, + struct thread *td); +static void cfi_ioctl_port_create(struct ctl_req *req); +static void cfi_ioctl_port_remove(struct ctl_req *req); +static struct cdevsw cfi_cdevsw = { + .d_version = D_VERSION, + .d_flags = 0, + .d_ioctl = ctl_ioctl_io +}; + static struct ctl_frontend cfi_frontend = { .name = "ioctl", .init = cfi_init, + .ioctl = cfi_ioctl, .shutdown = cfi_shutdown, }; CTL_FRONTEND_DECLARE(ctlioctl, cfi_frontend); static int cfi_init(void) { struct cfi_softc *isoftc = &cfi_softc; + struct cfi_port *cfi; struct ctl_port *port; int error = 0; memset(isoftc, 0, sizeof(*isoftc)); + TAILQ_INIT(&isoftc->ports); - port = &isoftc->port; + cfi = malloc(sizeof(*cfi), M_CTL, M_WAITOK | M_ZERO); + port = &cfi->port; port->frontend = &cfi_frontend; port->port_type = CTL_PORT_IOCTL; port->num_requested_ctl_io = 100; port->port_name = "ioctl"; port->fe_datamove = cfi_datamove; port->fe_done = cfi_done; + port->physical_port = 0; port->targ_port = -1; - port->max_initiators = 1; if ((error = ctl_port_register(port)) != 0) { printf("%s: ioctl port registration failed\n", __func__); return (error); } + ctl_port_online(port); + TAILQ_INSERT_TAIL(&isoftc->ports, cfi, link); return (0); } static int cfi_shutdown(void) { struct cfi_softc *isoftc = &cfi_softc; - struct ctl_port *port = &isoftc->port; - int error = 0; + struct cfi_port *cfi, *temp; + struct ctl_port *port; + int error; - ctl_port_offline(port); - if ((error = ctl_port_deregister(port)) != 0) - printf("%s: ioctl port deregistration failed\n", __func__); - return (error); + TAILQ_FOREACH_SAFE(cfi, &isoftc->ports, link, temp) { + port = &cfi->port; + ctl_port_offline(port); + error = ctl_port_deregister(port); + if (error != 0) { + printf("%s: ctl_frontend_deregister() failed\n", + __func__); + return (error); + } + + TAILQ_REMOVE(&isoftc->ports, cfi, link); + free(cfi, M_CTL); + } + + return (0); } +static void +cfi_ioctl_port_create(struct ctl_req *req) +{ + struct cfi_softc *isoftc = &cfi_softc; + struct cfi_port *cfi; + struct ctl_port *port; + struct make_dev_args args; + const char *val; + int retval; + int pp = -1, vp = 0; + + val = dnvlist_get_string(req->args_nvl, "pp", NULL); + if (val != NULL) + pp = strtol(val, NULL, 10); + + val = dnvlist_get_string(req->args_nvl, "vp", NULL); + if (val != NULL) + vp = strtol(val, NULL, 10); + + if (pp != -1) { + /* Check for duplicates */ + TAILQ_FOREACH(cfi, &isoftc->ports, link) { + if (pp == cfi->port.physical_port && + vp == cfi->port.virtual_port) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "port %d already exists", pp); + + return; + } + } + } else { + /* Find free port number */ + TAILQ_FOREACH(cfi, &isoftc->ports, link) { + pp = MAX(pp, cfi->port.physical_port); + } + + pp++; + } + + cfi = malloc(sizeof(*cfi), M_CTL, M_WAITOK | M_ZERO); + port = &cfi->port; + port->frontend = &cfi_frontend; + port->port_type = CTL_PORT_IOCTL; + port->num_requested_ctl_io = 100; + port->port_name = "ioctl"; + port->fe_datamove = cfi_datamove; + port->fe_done = cfi_done; + port->physical_port = pp; + port->virtual_port = vp; + port->targ_port = -1; + + retval = ctl_port_register(port); + if (retval != 0) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "ctl_port_register() failed with error %d", retval); + free(port, M_CTL); + return; + } + + req->result_nvl = nvlist_create(0); + nvlist_add_number(req->result_nvl, "port_id", port->targ_port); + ctl_port_online(port); + + make_dev_args_init(&args); + args.mda_devsw = &cfi_cdevsw; + args.mda_uid = UID_ROOT; + args.mda_gid = GID_OPERATOR; + args.mda_mode = 0600; + args.mda_si_drv1 = NULL; + args.mda_si_drv2 = cfi; + + retval = make_dev_s(&args, &cfi->dev, "cam/ctl%d.%d", pp, vp); + if (retval != 0) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "make_dev_s() failed with error %d", retval); + free(port, M_CTL); + return; + } + + req->status = CTL_LUN_OK; + TAILQ_INSERT_TAIL(&isoftc->ports, cfi, link); +} + +static void +cfi_ioctl_port_remove(struct ctl_req *req) +{ + struct cfi_softc *isoftc = &cfi_softc; + struct cfi_port *cfi = NULL; + const char *val; + int port_id = -1; + + val = dnvlist_get_string(req->args_nvl, "port_id", NULL); + if (val != NULL) + port_id = strtol(val, NULL, 10); + + if (port_id == -1) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "port_id not provided"); + return; + } + + TAILQ_FOREACH(cfi, &isoftc->ports, link) { + if (cfi->port.targ_port == port_id) + break; + } + + if (cfi == NULL) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "cannot find port %d", port_id); + + return; + } + + if (cfi->port.physical_port == 0 && cfi->port.virtual_port == 0) { + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "cannot destroy default ioctl port"); + + return; + } + + ctl_port_offline(&cfi->port); + ctl_port_deregister(&cfi->port); + TAILQ_REMOVE(&isoftc->ports, cfi, link); + destroy_dev(cfi->dev); + free(cfi, M_CTL); + req->status = CTL_LUN_OK; +} + +static int +cfi_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, + struct thread *td) +{ + struct ctl_req *req; + + if (cmd == CTL_PORT_REQ) { + req = (struct ctl_req *)addr; + switch (req->reqtype) { + case CTL_REQ_CREATE: + cfi_ioctl_port_create(req); + break; + case CTL_REQ_REMOVE: + cfi_ioctl_port_remove(req); + break; + default: + req->status = CTL_LUN_ERROR; + snprintf(req->error_str, sizeof(req->error_str), + "Unsupported request type %d", req->reqtype); + } + return (0); + } + + return (ENOTTY); +} + /* * Data movement routine for the CTL ioctl frontend port. */ static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio) { struct ctl_sg_entry *ext_sglist, *kern_sglist; struct ctl_sg_entry ext_entry, kern_entry; int ext_sglen, ext_sg_entries, kern_sg_entries; int ext_sg_start, ext_offset; int len_to_copy; int kern_watermark, ext_watermark; int ext_sglist_malloced; int i, j; CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove\n")); /* * If this flag is set, fake the data transfer. */ if (ctsio->io_hdr.flags & CTL_FLAG_NO_DATAMOVE) { ext_sglist_malloced = 0; ctsio->ext_data_filled += ctsio->kern_data_len; ctsio->kern_data_resid = 0; goto bailout; } /* * To simplify things here, if we have a single buffer, stick it in * a S/G entry and just make it a single entry S/G list. */ if (ctsio->ext_sg_entries > 0) { int len_seen; ext_sglen = ctsio->ext_sg_entries * sizeof(*ext_sglist); ext_sglist = (struct ctl_sg_entry *)malloc(ext_sglen, M_CTL, M_WAITOK); ext_sglist_malloced = 1; if (copyin(ctsio->ext_data_ptr, ext_sglist, ext_sglen) != 0) { ctsio->io_hdr.port_status = 31343; goto bailout; } ext_sg_entries = ctsio->ext_sg_entries; ext_sg_start = ext_sg_entries; ext_offset = 0; len_seen = 0; for (i = 0; i < ext_sg_entries; i++) { if ((len_seen + ext_sglist[i].len) >= ctsio->ext_data_filled) { ext_sg_start = i; ext_offset = ctsio->ext_data_filled - len_seen; break; } len_seen += ext_sglist[i].len; } } else { ext_sglist = &ext_entry; ext_sglist_malloced = 0; ext_sglist->addr = ctsio->ext_data_ptr; ext_sglist->len = ctsio->ext_data_len; ext_sg_entries = 1; ext_sg_start = 0; ext_offset = ctsio->ext_data_filled; } if (ctsio->kern_sg_entries > 0) { kern_sglist = (struct ctl_sg_entry *)ctsio->kern_data_ptr; kern_sg_entries = ctsio->kern_sg_entries; } else { kern_sglist = &kern_entry; kern_sglist->addr = ctsio->kern_data_ptr; kern_sglist->len = ctsio->kern_data_len; kern_sg_entries = 1; } kern_watermark = 0; ext_watermark = ext_offset; for (i = ext_sg_start, j = 0; i < ext_sg_entries && j < kern_sg_entries;) { uint8_t *ext_ptr, *kern_ptr; len_to_copy = MIN(ext_sglist[i].len - ext_watermark, kern_sglist[j].len - kern_watermark); ext_ptr = (uint8_t *)ext_sglist[i].addr; ext_ptr = ext_ptr + ext_watermark; if (ctsio->io_hdr.flags & CTL_FLAG_BUS_ADDR) { /* * XXX KDM fix this! */ panic("need to implement bus address support"); #if 0 kern_ptr = bus_to_virt(kern_sglist[j].addr); #endif } else kern_ptr = (uint8_t *)kern_sglist[j].addr; kern_ptr = kern_ptr + kern_watermark; if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d " "bytes to user\n", len_to_copy)); CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p " "to %p\n", kern_ptr, ext_ptr)); if (copyout(kern_ptr, ext_ptr, len_to_copy) != 0) { ctsio->io_hdr.port_status = 31344; goto bailout; } } else { CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: copying %d " "bytes from user\n", len_to_copy)); CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: from %p " "to %p\n", ext_ptr, kern_ptr)); if (copyin(ext_ptr, kern_ptr, len_to_copy)!= 0){ ctsio->io_hdr.port_status = 31345; goto bailout; } } ctsio->ext_data_filled += len_to_copy; ctsio->kern_data_resid -= len_to_copy; ext_watermark += len_to_copy; if (ext_sglist[i].len == ext_watermark) { i++; ext_watermark = 0; } kern_watermark += len_to_copy; if (kern_sglist[j].len == kern_watermark) { j++; kern_watermark = 0; } } CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_sg_entries: %d, " "kern_sg_entries: %d\n", ext_sg_entries, kern_sg_entries)); CTL_DEBUG_PRINT(("ctl_ioctl_do_datamove: ext_data_len = %d, " "kern_data_len = %d\n", ctsio->ext_data_len, ctsio->kern_data_len)); bailout: if (ext_sglist_malloced != 0) free(ext_sglist, M_CTL); return (CTL_RETVAL_COMPLETE); } static void cfi_datamove(union ctl_io *io) { struct ctl_fe_ioctl_params *params; params = (struct ctl_fe_ioctl_params *) io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; mtx_lock(¶ms->ioctl_mtx); params->state = CTL_IOCTL_DATAMOVE; cv_broadcast(¶ms->sem); mtx_unlock(¶ms->ioctl_mtx); } static void cfi_done(union ctl_io *io) { struct ctl_fe_ioctl_params *params; params = (struct ctl_fe_ioctl_params *) io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; mtx_lock(¶ms->ioctl_mtx); params->state = CTL_IOCTL_DONE; cv_broadcast(¶ms->sem); mtx_unlock(¶ms->ioctl_mtx); } static int cfi_submit_wait(union ctl_io *io) { struct ctl_fe_ioctl_params params; ctl_fe_ioctl_state last_state; int done, retval; bzero(¶ms, sizeof(params)); mtx_init(¶ms.ioctl_mtx, "ctliocmtx", NULL, MTX_DEF); cv_init(¶ms.sem, "ctlioccv"); params.state = CTL_IOCTL_INPROG; last_state = params.state; io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = ¶ms; CTL_DEBUG_PRINT(("cfi_submit_wait\n")); /* This shouldn't happen */ if ((retval = ctl_queue(io)) != CTL_RETVAL_COMPLETE) return (retval); done = 0; do { mtx_lock(¶ms.ioctl_mtx); /* * Check the state here, and don't sleep if the state has * already changed (i.e. wakeup has already occurred, but we * weren't waiting yet). */ if (params.state == last_state) { /* XXX KDM cv_wait_sig instead? */ cv_wait(¶ms.sem, ¶ms.ioctl_mtx); } last_state = params.state; switch (params.state) { case CTL_IOCTL_INPROG: /* Why did we wake up? */ /* XXX KDM error here? */ mtx_unlock(¶ms.ioctl_mtx); break; case CTL_IOCTL_DATAMOVE: CTL_DEBUG_PRINT(("got CTL_IOCTL_DATAMOVE\n")); /* * change last_state back to INPROG to avoid * deadlock on subsequent data moves. */ params.state = last_state = CTL_IOCTL_INPROG; mtx_unlock(¶ms.ioctl_mtx); ctl_ioctl_do_datamove(&io->scsiio); /* * Note that in some cases, most notably writes, * this will queue the I/O and call us back later. * In other cases, generally reads, this routine * will immediately call back and wake us up, * probably using our own context. */ io->scsiio.be_move_done(io); break; case CTL_IOCTL_DONE: mtx_unlock(¶ms.ioctl_mtx); CTL_DEBUG_PRINT(("got CTL_IOCTL_DONE\n")); done = 1; break; default: mtx_unlock(¶ms.ioctl_mtx); /* XXX KDM error here? */ break; } } while (done == 0); mtx_destroy(¶ms.ioctl_mtx); cv_destroy(¶ms.sem); return (CTL_RETVAL_COMPLETE); } int ctl_ioctl_io(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td) { + struct cfi_port *cfi; union ctl_io *io; void *pool_tmp, *sc_tmp; int retval = 0; + if (cmd != CTL_IO) + return (ENOTTY); + + cfi = dev->si_drv2 == NULL + ? TAILQ_FIRST(&cfi_softc.ports) + : dev->si_drv2; + /* * If we haven't been "enabled", don't allow any SCSI I/O * to this FETD. */ - if ((cfi_softc.port.status & CTL_PORT_STATUS_ONLINE) == 0) + if ((cfi->port.status & CTL_PORT_STATUS_ONLINE) == 0) return (EPERM); - io = ctl_alloc_io(cfi_softc.port.ctl_pool_ref); + io = ctl_alloc_io(cfi->port.ctl_pool_ref); /* * Need to save the pool reference so it doesn't get * spammed by the user's ctl_io. */ pool_tmp = io->io_hdr.pool; sc_tmp = CTL_SOFTC(io); memcpy(io, (void *)addr, sizeof(*io)); io->io_hdr.pool = pool_tmp; CTL_SOFTC(io) = sc_tmp; /* * No status yet, so make sure the status is set properly. */ io->io_hdr.status = CTL_STATUS_NONE; /* * The user sets the initiator ID, target and LUN IDs. */ - io->io_hdr.nexus.targ_port = cfi_softc.port.targ_port; + io->io_hdr.nexus.targ_port = cfi->port.targ_port; io->io_hdr.flags |= CTL_FLAG_USER_REQ; if ((io->io_hdr.io_type == CTL_IO_SCSI) && (io->scsiio.tag_type != CTL_TAG_UNTAGGED)) - io->scsiio.tag_num = cfi_softc.cur_tag_num++; + io->scsiio.tag_num = cfi->cur_tag_num++; retval = cfi_submit_wait(io); if (retval == 0) memcpy((void *)addr, io, sizeof(*io)); + ctl_free_io(io); return (retval); } Index: head/sys/cam/ctl/ctl_frontend_iscsi.c =================================================================== --- head/sys/cam/ctl/ctl_frontend_iscsi.c (revision 333445) +++ head/sys/cam/ctl/ctl_frontend_iscsi.c (revision 333446) @@ -1,3027 +1,3022 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 The FreeBSD Foundation * All rights reserved. * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ /* * CTL frontend for the iSCSI protocol. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef ICL_KERNEL_PROXY #include #endif #ifdef ICL_KERNEL_PROXY FEATURE(cfiscsi_kernel_proxy, "iSCSI target built with ICL_KERNEL_PROXY"); #endif static MALLOC_DEFINE(M_CFISCSI, "cfiscsi", "Memory used for CTL iSCSI frontend"); static uma_zone_t cfiscsi_data_wait_zone; SYSCTL_NODE(_kern_cam_ctl, OID_AUTO, iscsi, CTLFLAG_RD, 0, "CAM Target Layer iSCSI Frontend"); static int debug = 1; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, debug, CTLFLAG_RWTUN, &debug, 1, "Enable debug messages"); static int ping_timeout = 5; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, ping_timeout, CTLFLAG_RWTUN, &ping_timeout, 5, "Interval between ping (NOP-Out) requests, in seconds"); static int login_timeout = 60; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, login_timeout, CTLFLAG_RWTUN, &login_timeout, 60, "Time to wait for ctld(8) to finish Login Phase, in seconds"); static int maxtags = 256; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, maxtags, CTLFLAG_RWTUN, &maxtags, 0, "Max number of requests queued by initiator"); #define CFISCSI_DEBUG(X, ...) \ do { \ if (debug > 1) { \ printf("%s: " X "\n", \ __func__, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_WARN(X, ...) \ do { \ if (debug > 0) { \ printf("WARNING: %s: " X "\n", \ __func__, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_SESSION_DEBUG(S, X, ...) \ do { \ if (debug > 1) { \ printf("%s: %s (%s): " X "\n", \ __func__, S->cs_initiator_addr, \ S->cs_initiator_name, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_SESSION_WARN(S, X, ...) \ do { \ if (debug > 0) { \ printf("WARNING: %s (%s): " X "\n", \ S->cs_initiator_addr, \ S->cs_initiator_name, ## __VA_ARGS__); \ } \ } while (0) #define CFISCSI_SESSION_LOCK(X) mtx_lock(&X->cs_lock) #define CFISCSI_SESSION_UNLOCK(X) mtx_unlock(&X->cs_lock) #define CFISCSI_SESSION_LOCK_ASSERT(X) mtx_assert(&X->cs_lock, MA_OWNED) #define CONN_SESSION(X) ((struct cfiscsi_session *)(X)->ic_prv0) #define PDU_SESSION(X) CONN_SESSION((X)->ip_conn) #define PDU_EXPDATASN(X) (X)->ip_prv0 #define PDU_TOTAL_TRANSFER_LEN(X) (X)->ip_prv1 #define PDU_R2TSN(X) (X)->ip_prv2 static int cfiscsi_init(void); static int cfiscsi_shutdown(void); static void cfiscsi_online(void *arg); static void cfiscsi_offline(void *arg); static int cfiscsi_info(void *arg, struct sbuf *sb); static int cfiscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); static void cfiscsi_datamove(union ctl_io *io); static void cfiscsi_datamove_in(union ctl_io *io); static void cfiscsi_datamove_out(union ctl_io *io); static void cfiscsi_done(union ctl_io *io); static bool cfiscsi_pdu_update_cmdsn(const struct icl_pdu *request); static void cfiscsi_pdu_handle_nop_out(struct icl_pdu *request); static void cfiscsi_pdu_handle_scsi_command(struct icl_pdu *request); static void cfiscsi_pdu_handle_task_request(struct icl_pdu *request); static void cfiscsi_pdu_handle_data_out(struct icl_pdu *request); static void cfiscsi_pdu_handle_logout_request(struct icl_pdu *request); static void cfiscsi_session_terminate(struct cfiscsi_session *cs); static struct cfiscsi_data_wait *cfiscsi_data_wait_new( struct cfiscsi_session *cs, union ctl_io *io, uint32_t initiator_task_tag, uint32_t *target_transfer_tagp); static void cfiscsi_data_wait_free(struct cfiscsi_session *cs, struct cfiscsi_data_wait *cdw); static struct cfiscsi_target *cfiscsi_target_find(struct cfiscsi_softc *softc, const char *name, uint16_t tag); static struct cfiscsi_target *cfiscsi_target_find_or_create( struct cfiscsi_softc *softc, const char *name, const char *alias, uint16_t tag); static void cfiscsi_target_release(struct cfiscsi_target *ct); static void cfiscsi_session_delete(struct cfiscsi_session *cs); static struct cfiscsi_softc cfiscsi_softc; static struct ctl_frontend cfiscsi_frontend = { .name = "iscsi", .init = cfiscsi_init, .ioctl = cfiscsi_ioctl, .shutdown = cfiscsi_shutdown, }; CTL_FRONTEND_DECLARE(cfiscsi, cfiscsi_frontend); MODULE_DEPEND(cfiscsi, icl, 1, 1, 1); static struct icl_pdu * cfiscsi_pdu_new_response(struct icl_pdu *request, int flags) { return (icl_pdu_new(request->ip_conn, flags)); } static bool cfiscsi_pdu_update_cmdsn(const struct icl_pdu *request) { const struct iscsi_bhs_scsi_command *bhssc; struct cfiscsi_session *cs; uint32_t cmdsn, expstatsn; cs = PDU_SESSION(request); /* * Every incoming PDU - not just NOP-Out - resets the ping timer. * The purpose of the timeout is to reset the connection when it stalls; * we don't want this to happen when NOP-In or NOP-Out ends up delayed * in some queue. * * XXX: Locking? */ cs->cs_timeout = 0; /* * Data-Out PDUs don't contain CmdSN. */ if ((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_DATA_OUT) return (false); /* * We're only using fields common for all the request * (initiator -> target) PDUs. */ bhssc = (const struct iscsi_bhs_scsi_command *)request->ip_bhs; cmdsn = ntohl(bhssc->bhssc_cmdsn); expstatsn = ntohl(bhssc->bhssc_expstatsn); CFISCSI_SESSION_LOCK(cs); #if 0 if (expstatsn != cs->cs_statsn) { CFISCSI_SESSION_DEBUG(cs, "received PDU with ExpStatSN %d, " "while current StatSN is %d", expstatsn, cs->cs_statsn); } #endif if ((request->ip_bhs->bhs_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) { /* * The target MUST silently ignore any non-immediate command * outside of this range. */ if (ISCSI_SNLT(cmdsn, cs->cs_cmdsn) || ISCSI_SNGT(cmdsn, cs->cs_cmdsn - 1 + maxtags)) { CFISCSI_SESSION_UNLOCK(cs); CFISCSI_SESSION_WARN(cs, "received PDU with CmdSN %u, " "while expected %u", cmdsn, cs->cs_cmdsn); return (true); } /* * We don't support multiple connections now, so any * discontinuity in CmdSN means lost PDUs. Since we don't * support PDU retransmission -- terminate the connection. */ if (cmdsn != cs->cs_cmdsn) { CFISCSI_SESSION_UNLOCK(cs); CFISCSI_SESSION_WARN(cs, "received PDU with CmdSN %u, " "while expected %u; dropping connection", cmdsn, cs->cs_cmdsn); cfiscsi_session_terminate(cs); return (true); } cs->cs_cmdsn++; } CFISCSI_SESSION_UNLOCK(cs); return (false); } static void cfiscsi_pdu_handle(struct icl_pdu *request) { struct cfiscsi_session *cs; bool ignore; cs = PDU_SESSION(request); ignore = cfiscsi_pdu_update_cmdsn(request); if (ignore) { icl_pdu_free(request); return; } /* * Handle the PDU; this includes e.g. receiving the remaining * part of PDU and submitting the SCSI command to CTL * or queueing a reply. The handling routine is responsible * for freeing the PDU when it's no longer needed. */ switch (request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) { case ISCSI_BHS_OPCODE_NOP_OUT: cfiscsi_pdu_handle_nop_out(request); break; case ISCSI_BHS_OPCODE_SCSI_COMMAND: cfiscsi_pdu_handle_scsi_command(request); break; case ISCSI_BHS_OPCODE_TASK_REQUEST: cfiscsi_pdu_handle_task_request(request); break; case ISCSI_BHS_OPCODE_SCSI_DATA_OUT: cfiscsi_pdu_handle_data_out(request); break; case ISCSI_BHS_OPCODE_LOGOUT_REQUEST: cfiscsi_pdu_handle_logout_request(request); break; default: CFISCSI_SESSION_WARN(cs, "received PDU with unsupported " "opcode 0x%x; dropping connection", request->ip_bhs->bhs_opcode); icl_pdu_free(request); cfiscsi_session_terminate(cs); } } static void cfiscsi_receive_callback(struct icl_pdu *request) { #ifdef ICL_KERNEL_PROXY struct cfiscsi_session *cs; cs = PDU_SESSION(request); if (cs->cs_waiting_for_ctld || cs->cs_login_phase) { if (cs->cs_login_pdu == NULL) cs->cs_login_pdu = request; else icl_pdu_free(request); cv_signal(&cs->cs_login_cv); return; } #endif cfiscsi_pdu_handle(request); } static void cfiscsi_error_callback(struct icl_conn *ic) { struct cfiscsi_session *cs; cs = CONN_SESSION(ic); CFISCSI_SESSION_WARN(cs, "connection error; dropping connection"); cfiscsi_session_terminate(cs); } static int cfiscsi_pdu_prepare(struct icl_pdu *response) { struct cfiscsi_session *cs; struct iscsi_bhs_scsi_response *bhssr; bool advance_statsn = true; cs = PDU_SESSION(response); CFISCSI_SESSION_LOCK_ASSERT(cs); /* * We're only using fields common for all the response * (target -> initiator) PDUs. */ bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs; /* * 10.8.3: "The StatSN for this connection is not advanced * after this PDU is sent." */ if (bhssr->bhssr_opcode == ISCSI_BHS_OPCODE_R2T) advance_statsn = false; /* * 10.19.2: "However, when the Initiator Task Tag is set to 0xffffffff, * StatSN for the connection is not advanced after this PDU is sent." */ if (bhssr->bhssr_opcode == ISCSI_BHS_OPCODE_NOP_IN && bhssr->bhssr_initiator_task_tag == 0xffffffff) advance_statsn = false; /* * See the comment below - StatSN is not meaningful and must * not be advanced. */ if (bhssr->bhssr_opcode == ISCSI_BHS_OPCODE_SCSI_DATA_IN && (bhssr->bhssr_flags & BHSDI_FLAGS_S) == 0) advance_statsn = false; /* * 10.7.3: "The fields StatSN, Status, and Residual Count * only have meaningful content if the S bit is set to 1." */ if (bhssr->bhssr_opcode != ISCSI_BHS_OPCODE_SCSI_DATA_IN || (bhssr->bhssr_flags & BHSDI_FLAGS_S)) bhssr->bhssr_statsn = htonl(cs->cs_statsn); bhssr->bhssr_expcmdsn = htonl(cs->cs_cmdsn); bhssr->bhssr_maxcmdsn = htonl(cs->cs_cmdsn - 1 + imax(0, maxtags - cs->cs_outstanding_ctl_pdus)); if (advance_statsn) cs->cs_statsn++; return (0); } static void cfiscsi_pdu_queue(struct icl_pdu *response) { struct cfiscsi_session *cs; cs = PDU_SESSION(response); CFISCSI_SESSION_LOCK(cs); cfiscsi_pdu_prepare(response); icl_pdu_queue(response); CFISCSI_SESSION_UNLOCK(cs); } static void cfiscsi_pdu_handle_nop_out(struct icl_pdu *request) { struct cfiscsi_session *cs; struct iscsi_bhs_nop_out *bhsno; struct iscsi_bhs_nop_in *bhsni; struct icl_pdu *response; void *data = NULL; size_t datasize; int error; cs = PDU_SESSION(request); bhsno = (struct iscsi_bhs_nop_out *)request->ip_bhs; if (bhsno->bhsno_initiator_task_tag == 0xffffffff) { /* * Nothing to do, iscsi_pdu_update_statsn() already * zeroed the timeout. */ icl_pdu_free(request); return; } datasize = icl_pdu_data_segment_length(request); if (datasize > 0) { data = malloc(datasize, M_CFISCSI, M_NOWAIT | M_ZERO); if (data == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } icl_pdu_get_data(request, 0, data, datasize); } response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "droppping connection"); free(data, M_CFISCSI); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhsni = (struct iscsi_bhs_nop_in *)response->ip_bhs; bhsni->bhsni_opcode = ISCSI_BHS_OPCODE_NOP_IN; bhsni->bhsni_flags = 0x80; bhsni->bhsni_initiator_task_tag = bhsno->bhsno_initiator_task_tag; bhsni->bhsni_target_transfer_tag = 0xffffffff; if (datasize > 0) { error = icl_pdu_append_data(response, data, datasize, M_NOWAIT); if (error != 0) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "dropping connection"); free(data, M_CFISCSI); icl_pdu_free(request); icl_pdu_free(response); cfiscsi_session_terminate(cs); return; } free(data, M_CFISCSI); } icl_pdu_free(request); cfiscsi_pdu_queue(response); } static void cfiscsi_pdu_handle_scsi_command(struct icl_pdu *request) { struct iscsi_bhs_scsi_command *bhssc; struct cfiscsi_session *cs; union ctl_io *io; int error; cs = PDU_SESSION(request); bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs; //CFISCSI_SESSION_DEBUG(cs, "initiator task tag 0x%x", // bhssc->bhssc_initiator_task_tag); if (request->ip_data_len > 0 && cs->cs_immediate_data == false) { CFISCSI_SESSION_WARN(cs, "unsolicited data with " "ImmediateData=No; dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } io = ctl_alloc_io(cs->cs_target->ct_port.ctl_pool_ref); ctl_zero_io(io); io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = request; io->io_hdr.io_type = CTL_IO_SCSI; io->io_hdr.nexus.initid = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_lun = ctl_decode_lun(be64toh(bhssc->bhssc_lun)); io->scsiio.tag_num = bhssc->bhssc_initiator_task_tag; switch ((bhssc->bhssc_flags & BHSSC_FLAGS_ATTR)) { case BHSSC_FLAGS_ATTR_UNTAGGED: io->scsiio.tag_type = CTL_TAG_UNTAGGED; break; case BHSSC_FLAGS_ATTR_SIMPLE: io->scsiio.tag_type = CTL_TAG_SIMPLE; break; case BHSSC_FLAGS_ATTR_ORDERED: io->scsiio.tag_type = CTL_TAG_ORDERED; break; case BHSSC_FLAGS_ATTR_HOQ: io->scsiio.tag_type = CTL_TAG_HEAD_OF_QUEUE; break; case BHSSC_FLAGS_ATTR_ACA: io->scsiio.tag_type = CTL_TAG_ACA; break; default: io->scsiio.tag_type = CTL_TAG_UNTAGGED; CFISCSI_SESSION_WARN(cs, "unhandled tag type %d", bhssc->bhssc_flags & BHSSC_FLAGS_ATTR); break; } io->scsiio.cdb_len = sizeof(bhssc->bhssc_cdb); /* Which is 16. */ memcpy(io->scsiio.cdb, bhssc->bhssc_cdb, sizeof(bhssc->bhssc_cdb)); refcount_acquire(&cs->cs_outstanding_ctl_pdus); error = ctl_queue(io); if (error != CTL_RETVAL_COMPLETE) { CFISCSI_SESSION_WARN(cs, "ctl_queue() failed; error %d; " "dropping connection", error); ctl_free_io(io); refcount_release(&cs->cs_outstanding_ctl_pdus); icl_pdu_free(request); cfiscsi_session_terminate(cs); } } static void cfiscsi_pdu_handle_task_request(struct icl_pdu *request) { struct iscsi_bhs_task_management_request *bhstmr; struct iscsi_bhs_task_management_response *bhstmr2; struct icl_pdu *response; struct cfiscsi_session *cs; union ctl_io *io; int error; cs = PDU_SESSION(request); bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs; io = ctl_alloc_io(cs->cs_target->ct_port.ctl_pool_ref); ctl_zero_io(io); io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = request; io->io_hdr.io_type = CTL_IO_TASK; io->io_hdr.nexus.initid = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_lun = ctl_decode_lun(be64toh(bhstmr->bhstmr_lun)); io->taskio.tag_type = CTL_TAG_SIMPLE; /* XXX */ switch (bhstmr->bhstmr_function & ~0x80) { case BHSTMR_FUNCTION_ABORT_TASK: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_ABORT_TASK"); #endif io->taskio.task_action = CTL_TASK_ABORT_TASK; io->taskio.tag_num = bhstmr->bhstmr_referenced_task_tag; break; case BHSTMR_FUNCTION_ABORT_TASK_SET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_ABORT_TASK_SET"); #endif io->taskio.task_action = CTL_TASK_ABORT_TASK_SET; break; case BHSTMR_FUNCTION_CLEAR_TASK_SET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_CLEAR_TASK_SET"); #endif io->taskio.task_action = CTL_TASK_CLEAR_TASK_SET; break; case BHSTMR_FUNCTION_LOGICAL_UNIT_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_LOGICAL_UNIT_RESET"); #endif io->taskio.task_action = CTL_TASK_LUN_RESET; break; case BHSTMR_FUNCTION_TARGET_WARM_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_TARGET_WARM_RESET"); #endif io->taskio.task_action = CTL_TASK_TARGET_RESET; break; case BHSTMR_FUNCTION_TARGET_COLD_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_TARGET_COLD_RESET"); #endif io->taskio.task_action = CTL_TASK_TARGET_RESET; break; case BHSTMR_FUNCTION_QUERY_TASK: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_QUERY_TASK"); #endif io->taskio.task_action = CTL_TASK_QUERY_TASK; io->taskio.tag_num = bhstmr->bhstmr_referenced_task_tag; break; case BHSTMR_FUNCTION_QUERY_TASK_SET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_QUERY_TASK_SET"); #endif io->taskio.task_action = CTL_TASK_QUERY_TASK_SET; break; case BHSTMR_FUNCTION_I_T_NEXUS_RESET: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_I_T_NEXUS_RESET"); #endif io->taskio.task_action = CTL_TASK_I_T_NEXUS_RESET; break; case BHSTMR_FUNCTION_QUERY_ASYNC_EVENT: #if 0 CFISCSI_SESSION_DEBUG(cs, "BHSTMR_FUNCTION_QUERY_ASYNC_EVENT"); #endif io->taskio.task_action = CTL_TASK_QUERY_ASYNC_EVENT; break; default: CFISCSI_SESSION_DEBUG(cs, "unsupported function 0x%x", bhstmr->bhstmr_function & ~0x80); ctl_free_io(io); response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; " "dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhstmr2 = (struct iscsi_bhs_task_management_response *) response->ip_bhs; bhstmr2->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_RESPONSE; bhstmr2->bhstmr_flags = 0x80; bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_NOT_SUPPORTED; bhstmr2->bhstmr_initiator_task_tag = bhstmr->bhstmr_initiator_task_tag; icl_pdu_free(request); cfiscsi_pdu_queue(response); return; } refcount_acquire(&cs->cs_outstanding_ctl_pdus); error = ctl_queue(io); if (error != CTL_RETVAL_COMPLETE) { CFISCSI_SESSION_WARN(cs, "ctl_queue() failed; error %d; " "dropping connection", error); ctl_free_io(io); refcount_release(&cs->cs_outstanding_ctl_pdus); icl_pdu_free(request); cfiscsi_session_terminate(cs); } } static bool cfiscsi_handle_data_segment(struct icl_pdu *request, struct cfiscsi_data_wait *cdw) { struct iscsi_bhs_data_out *bhsdo; struct cfiscsi_session *cs; struct ctl_sg_entry ctl_sg_entry, *ctl_sglist; size_t copy_len, len, off, buffer_offset; int ctl_sg_count; union ctl_io *io; cs = PDU_SESSION(request); KASSERT((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_DATA_OUT || (request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("bad opcode 0x%x", request->ip_bhs->bhs_opcode)); /* * We're only using fields common for Data-Out and SCSI Command PDUs. */ bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs; io = cdw->cdw_ctl_io; KASSERT((io->io_hdr.flags & CTL_FLAG_DATA_MASK) != CTL_FLAG_DATA_IN, ("CTL_FLAG_DATA_IN")); #if 0 CFISCSI_SESSION_DEBUG(cs, "received %zd bytes out of %d", request->ip_data_len, io->scsiio.kern_total_len); #endif if (io->scsiio.kern_sg_entries > 0) { ctl_sglist = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; ctl_sg_count = io->scsiio.kern_sg_entries; } else { ctl_sglist = &ctl_sg_entry; ctl_sglist->addr = io->scsiio.kern_data_ptr; ctl_sglist->len = io->scsiio.kern_data_len; ctl_sg_count = 1; } if ((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_DATA_OUT) buffer_offset = ntohl(bhsdo->bhsdo_buffer_offset); else buffer_offset = 0; len = icl_pdu_data_segment_length(request); /* * Make sure the offset, as sent by the initiator, matches the offset * we're supposed to be at in the scatter-gather list. */ if (buffer_offset > io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled || buffer_offset + len <= io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled) { CFISCSI_SESSION_WARN(cs, "received bad buffer offset %zd, " "expected %zd; dropping connection", buffer_offset, (size_t)io->scsiio.kern_rel_offset + (size_t)io->scsiio.ext_data_filled); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } /* * This is the offset within the PDU data segment, as opposed * to buffer_offset, which is the offset within the task (SCSI * command). */ off = io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled - buffer_offset; /* * Iterate over the scatter/gather segments, filling them with data * from the PDU data segment. Note that this can get called multiple * times for one SCSI command; the cdw structure holds state for the * scatter/gather list. */ for (;;) { KASSERT(cdw->cdw_sg_index < ctl_sg_count, ("cdw->cdw_sg_index >= ctl_sg_count")); if (cdw->cdw_sg_len == 0) { cdw->cdw_sg_addr = ctl_sglist[cdw->cdw_sg_index].addr; cdw->cdw_sg_len = ctl_sglist[cdw->cdw_sg_index].len; } KASSERT(off <= len, ("len > off")); copy_len = len - off; if (copy_len > cdw->cdw_sg_len) copy_len = cdw->cdw_sg_len; icl_pdu_get_data(request, off, cdw->cdw_sg_addr, copy_len); cdw->cdw_sg_addr += copy_len; cdw->cdw_sg_len -= copy_len; off += copy_len; io->scsiio.ext_data_filled += copy_len; io->scsiio.kern_data_resid -= copy_len; if (cdw->cdw_sg_len == 0) { /* * End of current segment. */ if (cdw->cdw_sg_index == ctl_sg_count - 1) { /* * Last segment in scatter/gather list. */ break; } cdw->cdw_sg_index++; } if (off == len) { /* * End of PDU payload. */ break; } } if (len > off) { /* * In case of unsolicited data, it's possible that the buffer * provided by CTL is smaller than negotiated FirstBurstLength. * Just ignore the superfluous data; will ask for them with R2T * on next call to cfiscsi_datamove(). * * This obviously can only happen with SCSI Command PDU. */ if ((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND) return (true); CFISCSI_SESSION_WARN(cs, "received too much data: got %zd bytes, " "expected %zd; dropping connection", icl_pdu_data_segment_length(request), off); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } if (io->scsiio.ext_data_filled == cdw->cdw_r2t_end && (bhsdo->bhsdo_flags & BHSDO_FLAGS_F) == 0) { CFISCSI_SESSION_WARN(cs, "got the final packet without " "the F flag; flags = 0x%x; dropping connection", bhsdo->bhsdo_flags); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } if (io->scsiio.ext_data_filled != cdw->cdw_r2t_end && (bhsdo->bhsdo_flags & BHSDO_FLAGS_F) != 0) { if ((request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_DATA_OUT) { CFISCSI_SESSION_WARN(cs, "got the final packet, but the " "transmitted size was %zd bytes instead of %d; " "dropping connection", (size_t)io->scsiio.ext_data_filled, cdw->cdw_r2t_end); ctl_set_data_phase_error(&io->scsiio); cfiscsi_session_terminate(cs); return (true); } else { /* * For SCSI Command PDU, this just means we need to * solicit more data by sending R2T. */ return (false); } } if (io->scsiio.ext_data_filled == cdw->cdw_r2t_end) { #if 0 CFISCSI_SESSION_DEBUG(cs, "no longer expecting Data-Out with target " "transfer tag 0x%x", cdw->cdw_target_transfer_tag); #endif return (true); } return (false); } static void cfiscsi_pdu_handle_data_out(struct icl_pdu *request) { struct iscsi_bhs_data_out *bhsdo; struct cfiscsi_session *cs; struct cfiscsi_data_wait *cdw = NULL; union ctl_io *io; bool done; cs = PDU_SESSION(request); bhsdo = (struct iscsi_bhs_data_out *)request->ip_bhs; CFISCSI_SESSION_LOCK(cs); TAILQ_FOREACH(cdw, &cs->cs_waiting_for_data_out, cdw_next) { #if 0 CFISCSI_SESSION_DEBUG(cs, "have ttt 0x%x, itt 0x%x; looking for " "ttt 0x%x, itt 0x%x", bhsdo->bhsdo_target_transfer_tag, bhsdo->bhsdo_initiator_task_tag, cdw->cdw_target_transfer_tag, cdw->cdw_initiator_task_tag)); #endif if (bhsdo->bhsdo_target_transfer_tag == cdw->cdw_target_transfer_tag) break; } CFISCSI_SESSION_UNLOCK(cs); if (cdw == NULL) { CFISCSI_SESSION_WARN(cs, "data transfer tag 0x%x, initiator task tag " "0x%x, not found; dropping connection", bhsdo->bhsdo_target_transfer_tag, bhsdo->bhsdo_initiator_task_tag); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } if (cdw->cdw_datasn != ntohl(bhsdo->bhsdo_datasn)) { CFISCSI_SESSION_WARN(cs, "received Data-Out PDU with " "DataSN %u, while expected %u; dropping connection", ntohl(bhsdo->bhsdo_datasn), cdw->cdw_datasn); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } cdw->cdw_datasn++; io = cdw->cdw_ctl_io; KASSERT((io->io_hdr.flags & CTL_FLAG_DATA_MASK) != CTL_FLAG_DATA_IN, ("CTL_FLAG_DATA_IN")); done = cfiscsi_handle_data_segment(request, cdw); if (done) { CFISCSI_SESSION_LOCK(cs); TAILQ_REMOVE(&cs->cs_waiting_for_data_out, cdw, cdw_next); CFISCSI_SESSION_UNLOCK(cs); done = (io->scsiio.ext_data_filled != cdw->cdw_r2t_end || io->scsiio.ext_data_filled == io->scsiio.kern_data_len); cfiscsi_data_wait_free(cs, cdw); io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; if (done) io->scsiio.be_move_done(io); else cfiscsi_datamove_out(io); } icl_pdu_free(request); } static void cfiscsi_pdu_handle_logout_request(struct icl_pdu *request) { struct iscsi_bhs_logout_request *bhslr; struct iscsi_bhs_logout_response *bhslr2; struct icl_pdu *response; struct cfiscsi_session *cs; cs = PDU_SESSION(request); bhslr = (struct iscsi_bhs_logout_request *)request->ip_bhs; switch (bhslr->bhslr_reason & 0x7f) { case BHSLR_REASON_CLOSE_SESSION: case BHSLR_REASON_CLOSE_CONNECTION: response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_DEBUG(cs, "failed to allocate memory"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhslr2 = (struct iscsi_bhs_logout_response *)response->ip_bhs; bhslr2->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_RESPONSE; bhslr2->bhslr_flags = 0x80; bhslr2->bhslr_response = BHSLR_RESPONSE_CLOSED_SUCCESSFULLY; bhslr2->bhslr_initiator_task_tag = bhslr->bhslr_initiator_task_tag; icl_pdu_free(request); cfiscsi_pdu_queue(response); cfiscsi_session_terminate(cs); break; case BHSLR_REASON_REMOVE_FOR_RECOVERY: response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory; dropping connection"); icl_pdu_free(request); cfiscsi_session_terminate(cs); return; } bhslr2 = (struct iscsi_bhs_logout_response *)response->ip_bhs; bhslr2->bhslr_opcode = ISCSI_BHS_OPCODE_LOGOUT_RESPONSE; bhslr2->bhslr_flags = 0x80; bhslr2->bhslr_response = BHSLR_RESPONSE_RECOVERY_NOT_SUPPORTED; bhslr2->bhslr_initiator_task_tag = bhslr->bhslr_initiator_task_tag; icl_pdu_free(request); cfiscsi_pdu_queue(response); break; default: CFISCSI_SESSION_WARN(cs, "invalid reason 0%x; dropping connection", bhslr->bhslr_reason); icl_pdu_free(request); cfiscsi_session_terminate(cs); break; } } static void cfiscsi_callout(void *context) { struct icl_pdu *cp; struct iscsi_bhs_nop_in *bhsni; struct cfiscsi_session *cs; cs = context; if (cs->cs_terminating) return; callout_schedule(&cs->cs_callout, 1 * hz); atomic_add_int(&cs->cs_timeout, 1); #ifdef ICL_KERNEL_PROXY if (cs->cs_waiting_for_ctld || cs->cs_login_phase) { if (login_timeout > 0 && cs->cs_timeout > login_timeout) { CFISCSI_SESSION_WARN(cs, "login timed out after " "%d seconds; dropping connection", cs->cs_timeout); cfiscsi_session_terminate(cs); } return; } #endif if (ping_timeout <= 0) { /* * Pings are disabled. Don't send NOP-In in this case; * user might have disabled pings to work around problems * with certain initiators that can't properly handle * NOP-In, such as iPXE. Reset the timeout, to avoid * triggering reconnection, should the user decide to * reenable them. */ cs->cs_timeout = 0; return; } if (cs->cs_timeout >= ping_timeout) { CFISCSI_SESSION_WARN(cs, "no ping reply (NOP-Out) after %d seconds; " "dropping connection", ping_timeout); cfiscsi_session_terminate(cs); return; } /* * If the ping was reset less than one second ago - which means * that we've received some PDU during the last second - assume * the traffic flows correctly and don't bother sending a NOP-Out. * * (It's 2 - one for one second, and one for incrementing is_timeout * earlier in this routine.) */ if (cs->cs_timeout < 2) return; cp = icl_pdu_new(cs->cs_conn, M_NOWAIT); if (cp == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate memory"); return; } bhsni = (struct iscsi_bhs_nop_in *)cp->ip_bhs; bhsni->bhsni_opcode = ISCSI_BHS_OPCODE_NOP_IN; bhsni->bhsni_flags = 0x80; bhsni->bhsni_initiator_task_tag = 0xffffffff; cfiscsi_pdu_queue(cp); } static struct cfiscsi_data_wait * cfiscsi_data_wait_new(struct cfiscsi_session *cs, union ctl_io *io, uint32_t initiator_task_tag, uint32_t *target_transfer_tagp) { struct cfiscsi_data_wait *cdw; int error; cdw = uma_zalloc(cfiscsi_data_wait_zone, M_NOWAIT | M_ZERO); if (cdw == NULL) { CFISCSI_SESSION_WARN(cs, "failed to allocate %zd bytes", sizeof(*cdw)); return (NULL); } error = icl_conn_transfer_setup(cs->cs_conn, io, target_transfer_tagp, &cdw->cdw_icl_prv); if (error != 0) { CFISCSI_SESSION_WARN(cs, "icl_conn_transfer_setup() failed with error %d", error); uma_zfree(cfiscsi_data_wait_zone, cdw); return (NULL); } cdw->cdw_ctl_io = io; cdw->cdw_target_transfer_tag = *target_transfer_tagp; cdw->cdw_initiator_task_tag = initiator_task_tag; return (cdw); } static void cfiscsi_data_wait_free(struct cfiscsi_session *cs, struct cfiscsi_data_wait *cdw) { icl_conn_transfer_done(cs->cs_conn, cdw->cdw_icl_prv); uma_zfree(cfiscsi_data_wait_zone, cdw); } static void cfiscsi_session_terminate_tasks(struct cfiscsi_session *cs) { struct cfiscsi_data_wait *cdw; union ctl_io *io; int error, last, wait; if (cs->cs_target == NULL) return; /* No target yet, so nothing to do. */ io = ctl_alloc_io(cs->cs_target->ct_port.ctl_pool_ref); ctl_zero_io(io); io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = cs; io->io_hdr.io_type = CTL_IO_TASK; io->io_hdr.nexus.initid = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_lun = 0; io->taskio.tag_type = CTL_TAG_SIMPLE; /* XXX */ io->taskio.task_action = CTL_TASK_I_T_NEXUS_RESET; wait = cs->cs_outstanding_ctl_pdus; refcount_acquire(&cs->cs_outstanding_ctl_pdus); error = ctl_queue(io); if (error != CTL_RETVAL_COMPLETE) { CFISCSI_SESSION_WARN(cs, "ctl_queue() failed; error %d", error); refcount_release(&cs->cs_outstanding_ctl_pdus); ctl_free_io(io); } CFISCSI_SESSION_LOCK(cs); while ((cdw = TAILQ_FIRST(&cs->cs_waiting_for_data_out)) != NULL) { TAILQ_REMOVE(&cs->cs_waiting_for_data_out, cdw, cdw_next); CFISCSI_SESSION_UNLOCK(cs); /* * Set nonzero port status; this prevents backends from * assuming that the data transfer actually succeeded * and writing uninitialized data to disk. */ cdw->cdw_ctl_io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; cdw->cdw_ctl_io->scsiio.io_hdr.port_status = 42; cdw->cdw_ctl_io->scsiio.be_move_done(cdw->cdw_ctl_io); cfiscsi_data_wait_free(cs, cdw); CFISCSI_SESSION_LOCK(cs); } CFISCSI_SESSION_UNLOCK(cs); /* * Wait for CTL to terminate all the tasks. */ if (wait > 0) CFISCSI_SESSION_WARN(cs, "waiting for CTL to terminate %d tasks", wait); for (;;) { refcount_acquire(&cs->cs_outstanding_ctl_pdus); last = refcount_release(&cs->cs_outstanding_ctl_pdus); if (last != 0) break; tsleep(__DEVOLATILE(void *, &cs->cs_outstanding_ctl_pdus), 0, "cfiscsi_terminate", hz / 100); } if (wait > 0) CFISCSI_SESSION_WARN(cs, "tasks terminated"); } static void cfiscsi_maintenance_thread(void *arg) { struct cfiscsi_session *cs; cs = arg; for (;;) { CFISCSI_SESSION_LOCK(cs); if (cs->cs_terminating == false || cs->cs_handoff_in_progress) cv_wait(&cs->cs_maintenance_cv, &cs->cs_lock); CFISCSI_SESSION_UNLOCK(cs); if (cs->cs_terminating && cs->cs_handoff_in_progress == false) { /* * We used to wait up to 30 seconds to deliver queued * PDUs to the initiator. We also tried hard to deliver * SCSI Responses for the aborted PDUs. We don't do * that anymore. We might need to revisit that. */ callout_drain(&cs->cs_callout); icl_conn_close(cs->cs_conn); /* * At this point ICL receive thread is no longer * running; no new tasks can be queued. */ cfiscsi_session_terminate_tasks(cs); cfiscsi_session_delete(cs); kthread_exit(); return; } CFISCSI_SESSION_DEBUG(cs, "nothing to do"); } } static void cfiscsi_session_terminate(struct cfiscsi_session *cs) { cs->cs_terminating = true; cv_signal(&cs->cs_maintenance_cv); #ifdef ICL_KERNEL_PROXY cv_signal(&cs->cs_login_cv); #endif } static int cfiscsi_session_register_initiator(struct cfiscsi_session *cs) { struct cfiscsi_target *ct; char *name; int i; KASSERT(cs->cs_ctl_initid == -1, ("already registered")); ct = cs->cs_target; name = strdup(cs->cs_initiator_id, M_CTL); i = ctl_add_initiator(&ct->ct_port, -1, 0, name); if (i < 0) { CFISCSI_SESSION_WARN(cs, "ctl_add_initiator failed with error %d", i); cs->cs_ctl_initid = -1; return (1); } cs->cs_ctl_initid = i; #if 0 CFISCSI_SESSION_DEBUG(cs, "added initiator id %d", i); #endif return (0); } static void cfiscsi_session_unregister_initiator(struct cfiscsi_session *cs) { int error; if (cs->cs_ctl_initid == -1) return; error = ctl_remove_initiator(&cs->cs_target->ct_port, cs->cs_ctl_initid); if (error != 0) { CFISCSI_SESSION_WARN(cs, "ctl_remove_initiator failed with error %d", error); } cs->cs_ctl_initid = -1; } static struct cfiscsi_session * cfiscsi_session_new(struct cfiscsi_softc *softc, const char *offload) { struct cfiscsi_session *cs; int error; cs = malloc(sizeof(*cs), M_CFISCSI, M_NOWAIT | M_ZERO); if (cs == NULL) { CFISCSI_WARN("malloc failed"); return (NULL); } cs->cs_ctl_initid = -1; refcount_init(&cs->cs_outstanding_ctl_pdus, 0); TAILQ_INIT(&cs->cs_waiting_for_data_out); mtx_init(&cs->cs_lock, "cfiscsi_lock", NULL, MTX_DEF); cv_init(&cs->cs_maintenance_cv, "cfiscsi_mt"); #ifdef ICL_KERNEL_PROXY cv_init(&cs->cs_login_cv, "cfiscsi_login"); #endif /* * The purpose of this is to avoid racing with session shutdown. * Otherwise we could have the maintenance thread call icl_conn_close() * before we call icl_conn_handoff(). */ cs->cs_handoff_in_progress = true; cs->cs_conn = icl_new_conn(offload, false, "cfiscsi", &cs->cs_lock); if (cs->cs_conn == NULL) { free(cs, M_CFISCSI); return (NULL); } cs->cs_conn->ic_receive = cfiscsi_receive_callback; cs->cs_conn->ic_error = cfiscsi_error_callback; cs->cs_conn->ic_prv0 = cs; error = kthread_add(cfiscsi_maintenance_thread, cs, NULL, NULL, 0, 0, "cfiscsimt"); if (error != 0) { CFISCSI_SESSION_WARN(cs, "kthread_add(9) failed with error %d", error); free(cs, M_CFISCSI); return (NULL); } mtx_lock(&softc->lock); cs->cs_id = ++softc->last_session_id; TAILQ_INSERT_TAIL(&softc->sessions, cs, cs_next); mtx_unlock(&softc->lock); /* * Start pinging the initiator. */ callout_init(&cs->cs_callout, 1); callout_reset(&cs->cs_callout, 1 * hz, cfiscsi_callout, cs); return (cs); } static void cfiscsi_session_delete(struct cfiscsi_session *cs) { struct cfiscsi_softc *softc; softc = &cfiscsi_softc; KASSERT(cs->cs_outstanding_ctl_pdus == 0, ("destroying session with outstanding CTL pdus")); KASSERT(TAILQ_EMPTY(&cs->cs_waiting_for_data_out), ("destroying session with non-empty queue")); mtx_lock(&softc->lock); TAILQ_REMOVE(&softc->sessions, cs, cs_next); mtx_unlock(&softc->lock); cfiscsi_session_unregister_initiator(cs); if (cs->cs_target != NULL) cfiscsi_target_release(cs->cs_target); icl_conn_close(cs->cs_conn); icl_conn_free(cs->cs_conn); free(cs, M_CFISCSI); cv_signal(&softc->sessions_cv); } static int cfiscsi_init(void) { struct cfiscsi_softc *softc; softc = &cfiscsi_softc; bzero(softc, sizeof(*softc)); mtx_init(&softc->lock, "cfiscsi", NULL, MTX_DEF); cv_init(&softc->sessions_cv, "cfiscsi_sessions"); #ifdef ICL_KERNEL_PROXY cv_init(&softc->accept_cv, "cfiscsi_accept"); #endif TAILQ_INIT(&softc->sessions); TAILQ_INIT(&softc->targets); cfiscsi_data_wait_zone = uma_zcreate("cfiscsi_data_wait", sizeof(struct cfiscsi_data_wait), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); return (0); } static int cfiscsi_shutdown(void) { struct cfiscsi_softc *softc = &cfiscsi_softc; if (!TAILQ_EMPTY(&softc->sessions) || !TAILQ_EMPTY(&softc->targets)) return (EBUSY); uma_zdestroy(cfiscsi_data_wait_zone); #ifdef ICL_KERNEL_PROXY cv_destroy(&softc->accept_cv); #endif cv_destroy(&softc->sessions_cv); mtx_destroy(&softc->lock); return (0); } #ifdef ICL_KERNEL_PROXY static void cfiscsi_accept(struct socket *so, struct sockaddr *sa, int portal_id) { struct cfiscsi_session *cs; cs = cfiscsi_session_new(&cfiscsi_softc, NULL); if (cs == NULL) { CFISCSI_WARN("failed to create session"); return; } icl_conn_handoff_sock(cs->cs_conn, so); cs->cs_initiator_sa = sa; cs->cs_portal_id = portal_id; cs->cs_handoff_in_progress = false; cs->cs_waiting_for_ctld = true; cv_signal(&cfiscsi_softc.accept_cv); CFISCSI_SESSION_LOCK(cs); /* * Wake up the maintenance thread if we got scheduled for termination * somewhere between cfiscsi_session_new() and icl_conn_handoff_sock(). */ if (cs->cs_terminating) cfiscsi_session_terminate(cs); CFISCSI_SESSION_UNLOCK(cs); } #endif static void cfiscsi_online(void *arg) { struct cfiscsi_softc *softc; struct cfiscsi_target *ct; int online; ct = (struct cfiscsi_target *)arg; softc = ct->ct_softc; mtx_lock(&softc->lock); if (ct->ct_online) { mtx_unlock(&softc->lock); return; } ct->ct_online = 1; online = softc->online++; mtx_unlock(&softc->lock); if (online > 0) return; #ifdef ICL_KERNEL_PROXY if (softc->listener != NULL) icl_listen_free(softc->listener); softc->listener = icl_listen_new(cfiscsi_accept); #endif } static void cfiscsi_offline(void *arg) { struct cfiscsi_softc *softc; struct cfiscsi_target *ct; struct cfiscsi_session *cs; int online; ct = (struct cfiscsi_target *)arg; softc = ct->ct_softc; mtx_lock(&softc->lock); if (!ct->ct_online) { mtx_unlock(&softc->lock); return; } ct->ct_online = 0; online = --softc->online; TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (cs->cs_target == ct) cfiscsi_session_terminate(cs); } do { TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (cs->cs_target == ct) break; } if (cs != NULL) cv_wait(&softc->sessions_cv, &softc->lock); } while (cs != NULL && ct->ct_online == 0); mtx_unlock(&softc->lock); if (online > 0) return; #ifdef ICL_KERNEL_PROXY icl_listen_free(softc->listener); softc->listener = NULL; #endif } static int cfiscsi_info(void *arg, struct sbuf *sb) { struct cfiscsi_target *ct = (struct cfiscsi_target *)arg; int retval; retval = sbuf_printf(sb, "\t%d\n", ct->ct_state); return (retval); } static void cfiscsi_ioctl_handoff(struct ctl_iscsi *ci) { struct cfiscsi_softc *softc; struct cfiscsi_session *cs, *cs2; struct cfiscsi_target *ct; struct ctl_iscsi_handoff_params *cihp; int error; cihp = (struct ctl_iscsi_handoff_params *)&(ci->data); softc = &cfiscsi_softc; CFISCSI_DEBUG("new connection from %s (%s) to %s", cihp->initiator_name, cihp->initiator_addr, cihp->target_name); ct = cfiscsi_target_find(softc, cihp->target_name, cihp->portal_group_tag); if (ct == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: target not found", __func__); return; } #ifdef ICL_KERNEL_PROXY if (cihp->socket > 0 && cihp->connection_id > 0) { snprintf(ci->error_str, sizeof(ci->error_str), "both socket and connection_id set"); ci->status = CTL_ISCSI_ERROR; cfiscsi_target_release(ct); return; } if (cihp->socket == 0) { mtx_lock(&cfiscsi_softc.lock); TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_id == cihp->connection_id) break; } if (cs == NULL) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "connection not found"); ci->status = CTL_ISCSI_ERROR; cfiscsi_target_release(ct); return; } mtx_unlock(&cfiscsi_softc.lock); } else { #endif cs = cfiscsi_session_new(softc, cihp->offload); if (cs == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: cfiscsi_session_new failed", __func__); cfiscsi_target_release(ct); return; } #ifdef ICL_KERNEL_PROXY } #endif /* * First PDU of Full Feature phase has the same CmdSN as the last * PDU from the Login Phase received from the initiator. Thus, * the -1 below. */ cs->cs_cmdsn = cihp->cmdsn; cs->cs_statsn = cihp->statsn; cs->cs_max_recv_data_segment_length = cihp->max_recv_data_segment_length; cs->cs_max_send_data_segment_length = cihp->max_send_data_segment_length; cs->cs_max_burst_length = cihp->max_burst_length; cs->cs_first_burst_length = cihp->first_burst_length; cs->cs_immediate_data = !!cihp->immediate_data; if (cihp->header_digest == CTL_ISCSI_DIGEST_CRC32C) cs->cs_conn->ic_header_crc32c = true; if (cihp->data_digest == CTL_ISCSI_DIGEST_CRC32C) cs->cs_conn->ic_data_crc32c = true; strlcpy(cs->cs_initiator_name, cihp->initiator_name, sizeof(cs->cs_initiator_name)); strlcpy(cs->cs_initiator_addr, cihp->initiator_addr, sizeof(cs->cs_initiator_addr)); strlcpy(cs->cs_initiator_alias, cihp->initiator_alias, sizeof(cs->cs_initiator_alias)); memcpy(cs->cs_initiator_isid, cihp->initiator_isid, sizeof(cs->cs_initiator_isid)); snprintf(cs->cs_initiator_id, sizeof(cs->cs_initiator_id), "%s,i,0x%02x%02x%02x%02x%02x%02x", cs->cs_initiator_name, cihp->initiator_isid[0], cihp->initiator_isid[1], cihp->initiator_isid[2], cihp->initiator_isid[3], cihp->initiator_isid[4], cihp->initiator_isid[5]); mtx_lock(&softc->lock); if (ct->ct_online == 0) { mtx_unlock(&softc->lock); cs->cs_handoff_in_progress = false; cfiscsi_session_terminate(cs); cfiscsi_target_release(ct); ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: port offline", __func__); return; } cs->cs_target = ct; mtx_unlock(&softc->lock); restart: if (!cs->cs_terminating) { mtx_lock(&softc->lock); TAILQ_FOREACH(cs2, &softc->sessions, cs_next) { if (cs2 != cs && cs2->cs_tasks_aborted == false && cs->cs_target == cs2->cs_target && strcmp(cs->cs_initiator_id, cs2->cs_initiator_id) == 0) { if (strcmp(cs->cs_initiator_addr, cs2->cs_initiator_addr) != 0) { CFISCSI_SESSION_WARN(cs2, "session reinstatement from " "different address %s", cs->cs_initiator_addr); } else { CFISCSI_SESSION_DEBUG(cs2, "session reinstatement"); } cfiscsi_session_terminate(cs2); mtx_unlock(&softc->lock); pause("cfiscsi_reinstate", 1); goto restart; } } mtx_unlock(&softc->lock); } /* * Register initiator with CTL. */ cfiscsi_session_register_initiator(cs); #ifdef ICL_KERNEL_PROXY if (cihp->socket > 0) { #endif error = icl_conn_handoff(cs->cs_conn, cihp->socket); if (error != 0) { cs->cs_handoff_in_progress = false; cfiscsi_session_terminate(cs); ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: icl_conn_handoff failed with error %d", __func__, error); return; } #ifdef ICL_KERNEL_PROXY } #endif #ifdef ICL_KERNEL_PROXY cs->cs_login_phase = false; /* * First PDU of the Full Feature phase has likely already arrived. * We have to pick it up and execute properly. */ if (cs->cs_login_pdu != NULL) { CFISCSI_SESSION_DEBUG(cs, "picking up first PDU"); cfiscsi_pdu_handle(cs->cs_login_pdu); cs->cs_login_pdu = NULL; } #endif CFISCSI_SESSION_LOCK(cs); cs->cs_handoff_in_progress = false; /* * Wake up the maintenance thread if we got scheduled for termination. */ if (cs->cs_terminating) cfiscsi_session_terminate(cs); CFISCSI_SESSION_UNLOCK(cs); ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_list(struct ctl_iscsi *ci) { struct ctl_iscsi_list_params *cilp; struct cfiscsi_session *cs; struct cfiscsi_softc *softc; struct sbuf *sb; int error; cilp = (struct ctl_iscsi_list_params *)&(ci->data); softc = &cfiscsi_softc; sb = sbuf_new(NULL, NULL, cilp->alloc_len, SBUF_FIXEDLEN); if (sb == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "Unable to allocate %d bytes for iSCSI session list", cilp->alloc_len); return; } sbuf_printf(sb, "\n"); mtx_lock(&softc->lock); TAILQ_FOREACH(cs, &softc->sessions, cs_next) { #ifdef ICL_KERNEL_PROXY if (cs->cs_target == NULL) continue; #endif error = sbuf_printf(sb, "" "%s" "%s" "%s" "%s" "%s" "%u" "%s" "%s" "%d" "%d" "%d" "%d" "%d" "%d" "%s" "\n", cs->cs_id, cs->cs_initiator_name, cs->cs_initiator_addr, cs->cs_initiator_alias, cs->cs_target->ct_name, cs->cs_target->ct_alias, cs->cs_target->ct_tag, cs->cs_conn->ic_header_crc32c ? "CRC32C" : "None", cs->cs_conn->ic_data_crc32c ? "CRC32C" : "None", cs->cs_max_recv_data_segment_length, cs->cs_max_send_data_segment_length, cs->cs_max_burst_length, cs->cs_first_burst_length, cs->cs_immediate_data, cs->cs_conn->ic_iser, cs->cs_conn->ic_offload); if (error != 0) break; } mtx_unlock(&softc->lock); error = sbuf_printf(sb, "\n"); if (error != 0) { sbuf_delete(sb); ci->status = CTL_ISCSI_LIST_NEED_MORE_SPACE; snprintf(ci->error_str, sizeof(ci->error_str), "Out of space, %d bytes is too small", cilp->alloc_len); return; } sbuf_finish(sb); error = copyout(sbuf_data(sb), cilp->conn_xml, sbuf_len(sb) + 1); if (error != 0) { sbuf_delete(sb); snprintf(ci->error_str, sizeof(ci->error_str), "copyout failed with error %d", error); ci->status = CTL_ISCSI_ERROR; return; } cilp->fill_len = sbuf_len(sb) + 1; ci->status = CTL_ISCSI_OK; sbuf_delete(sb); } static void cfiscsi_ioctl_logout(struct ctl_iscsi *ci) { struct icl_pdu *response; struct iscsi_bhs_asynchronous_message *bhsam; struct ctl_iscsi_logout_params *cilp; struct cfiscsi_session *cs; struct cfiscsi_softc *softc; int found = 0; cilp = (struct ctl_iscsi_logout_params *)&(ci->data); softc = &cfiscsi_softc; mtx_lock(&softc->lock); TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (cilp->all == 0 && cs->cs_id != cilp->connection_id && strcmp(cs->cs_initiator_name, cilp->initiator_name) != 0 && strcmp(cs->cs_initiator_addr, cilp->initiator_addr) != 0) continue; response = icl_pdu_new(cs->cs_conn, M_NOWAIT); if (response == NULL) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "Unable to allocate memory"); mtx_unlock(&softc->lock); return; } bhsam = (struct iscsi_bhs_asynchronous_message *)response->ip_bhs; bhsam->bhsam_opcode = ISCSI_BHS_OPCODE_ASYNC_MESSAGE; bhsam->bhsam_flags = 0x80; bhsam->bhsam_async_event = BHSAM_EVENT_TARGET_REQUESTS_LOGOUT; bhsam->bhsam_parameter3 = htons(10); cfiscsi_pdu_queue(response); found++; } mtx_unlock(&softc->lock); if (found == 0) { ci->status = CTL_ISCSI_SESSION_NOT_FOUND; snprintf(ci->error_str, sizeof(ci->error_str), "No matching connections found"); return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_terminate(struct ctl_iscsi *ci) { struct icl_pdu *response; struct iscsi_bhs_asynchronous_message *bhsam; struct ctl_iscsi_terminate_params *citp; struct cfiscsi_session *cs; struct cfiscsi_softc *softc; int found = 0; citp = (struct ctl_iscsi_terminate_params *)&(ci->data); softc = &cfiscsi_softc; mtx_lock(&softc->lock); TAILQ_FOREACH(cs, &softc->sessions, cs_next) { if (citp->all == 0 && cs->cs_id != citp->connection_id && strcmp(cs->cs_initiator_name, citp->initiator_name) != 0 && strcmp(cs->cs_initiator_addr, citp->initiator_addr) != 0) continue; response = icl_pdu_new(cs->cs_conn, M_NOWAIT); if (response == NULL) { /* * Oh well. Just terminate the connection. */ } else { bhsam = (struct iscsi_bhs_asynchronous_message *) response->ip_bhs; bhsam->bhsam_opcode = ISCSI_BHS_OPCODE_ASYNC_MESSAGE; bhsam->bhsam_flags = 0x80; bhsam->bhsam_0xffffffff = 0xffffffff; bhsam->bhsam_async_event = BHSAM_EVENT_TARGET_TERMINATES_SESSION; cfiscsi_pdu_queue(response); } cfiscsi_session_terminate(cs); found++; } mtx_unlock(&softc->lock); if (found == 0) { ci->status = CTL_ISCSI_SESSION_NOT_FOUND; snprintf(ci->error_str, sizeof(ci->error_str), "No matching connections found"); return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_limits(struct ctl_iscsi *ci) { struct ctl_iscsi_limits_params *cilp; struct icl_drv_limits idl; int error; cilp = (struct ctl_iscsi_limits_params *)&(ci->data); error = icl_limits(cilp->offload, false, &idl); if (error != 0) { ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: icl_limits failed with error %d", __func__, error); return; } cilp->max_recv_data_segment_length = idl.idl_max_recv_data_segment_length; cilp->max_send_data_segment_length = idl.idl_max_send_data_segment_length; cilp->max_burst_length = idl.idl_max_burst_length; cilp->first_burst_length = idl.idl_first_burst_length; ci->status = CTL_ISCSI_OK; } #ifdef ICL_KERNEL_PROXY static void cfiscsi_ioctl_listen(struct ctl_iscsi *ci) { struct ctl_iscsi_listen_params *cilp; struct sockaddr *sa; int error; cilp = (struct ctl_iscsi_listen_params *)&(ci->data); if (cfiscsi_softc.listener == NULL) { CFISCSI_DEBUG("no listener"); snprintf(ci->error_str, sizeof(ci->error_str), "no listener"); ci->status = CTL_ISCSI_ERROR; return; } error = getsockaddr(&sa, (void *)cilp->addr, cilp->addrlen); if (error != 0) { CFISCSI_DEBUG("getsockaddr, error %d", error); snprintf(ci->error_str, sizeof(ci->error_str), "getsockaddr failed"); ci->status = CTL_ISCSI_ERROR; return; } error = icl_listen_add(cfiscsi_softc.listener, cilp->iser, cilp->domain, cilp->socktype, cilp->protocol, sa, cilp->portal_id); if (error != 0) { free(sa, M_SONAME); CFISCSI_DEBUG("icl_listen_add, error %d", error); snprintf(ci->error_str, sizeof(ci->error_str), "icl_listen_add failed, error %d", error); ci->status = CTL_ISCSI_ERROR; return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_accept(struct ctl_iscsi *ci) { struct ctl_iscsi_accept_params *ciap; struct cfiscsi_session *cs; int error; ciap = (struct ctl_iscsi_accept_params *)&(ci->data); mtx_lock(&cfiscsi_softc.lock); for (;;) { TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_waiting_for_ctld) break; } if (cs != NULL) break; error = cv_wait_sig(&cfiscsi_softc.accept_cv, &cfiscsi_softc.lock); if (error != 0) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "interrupted"); ci->status = CTL_ISCSI_ERROR; return; } } mtx_unlock(&cfiscsi_softc.lock); cs->cs_waiting_for_ctld = false; cs->cs_login_phase = true; ciap->connection_id = cs->cs_id; ciap->portal_id = cs->cs_portal_id; ciap->initiator_addrlen = cs->cs_initiator_sa->sa_len; error = copyout(cs->cs_initiator_sa, ciap->initiator_addr, cs->cs_initiator_sa->sa_len); if (error != 0) { snprintf(ci->error_str, sizeof(ci->error_str), "copyout failed with error %d", error); ci->status = CTL_ISCSI_ERROR; return; } ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_send(struct ctl_iscsi *ci) { struct ctl_iscsi_send_params *cisp; struct cfiscsi_session *cs; struct icl_pdu *ip; size_t datalen; void *data; int error; cisp = (struct ctl_iscsi_send_params *)&(ci->data); mtx_lock(&cfiscsi_softc.lock); TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_id == cisp->connection_id) break; } if (cs == NULL) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "connection not found"); ci->status = CTL_ISCSI_ERROR; return; } mtx_unlock(&cfiscsi_softc.lock); #if 0 if (cs->cs_login_phase == false) return (EBUSY); #endif if (cs->cs_terminating) { snprintf(ci->error_str, sizeof(ci->error_str), "connection is terminating"); ci->status = CTL_ISCSI_ERROR; return; } datalen = cisp->data_segment_len; /* * XXX */ //if (datalen > CFISCSI_MAX_DATA_SEGMENT_LENGTH) { if (datalen > 65535) { snprintf(ci->error_str, sizeof(ci->error_str), "data segment too big"); ci->status = CTL_ISCSI_ERROR; return; } if (datalen > 0) { data = malloc(datalen, M_CFISCSI, M_WAITOK); error = copyin(cisp->data_segment, data, datalen); if (error != 0) { free(data, M_CFISCSI); snprintf(ci->error_str, sizeof(ci->error_str), "copyin error %d", error); ci->status = CTL_ISCSI_ERROR; return; } } ip = icl_pdu_new(cs->cs_conn, M_WAITOK); memcpy(ip->ip_bhs, cisp->bhs, sizeof(*ip->ip_bhs)); if (datalen > 0) { icl_pdu_append_data(ip, data, datalen, M_WAITOK); free(data, M_CFISCSI); } CFISCSI_SESSION_LOCK(cs); icl_pdu_queue(ip); CFISCSI_SESSION_UNLOCK(cs); ci->status = CTL_ISCSI_OK; } static void cfiscsi_ioctl_receive(struct ctl_iscsi *ci) { struct ctl_iscsi_receive_params *cirp; struct cfiscsi_session *cs; struct icl_pdu *ip; void *data; int error; cirp = (struct ctl_iscsi_receive_params *)&(ci->data); mtx_lock(&cfiscsi_softc.lock); TAILQ_FOREACH(cs, &cfiscsi_softc.sessions, cs_next) { if (cs->cs_id == cirp->connection_id) break; } if (cs == NULL) { mtx_unlock(&cfiscsi_softc.lock); snprintf(ci->error_str, sizeof(ci->error_str), "connection not found"); ci->status = CTL_ISCSI_ERROR; return; } mtx_unlock(&cfiscsi_softc.lock); #if 0 if (is->is_login_phase == false) return (EBUSY); #endif CFISCSI_SESSION_LOCK(cs); while (cs->cs_login_pdu == NULL && cs->cs_terminating == false) { error = cv_wait_sig(&cs->cs_login_cv, &cs->cs_lock); if (error != 0) { CFISCSI_SESSION_UNLOCK(cs); snprintf(ci->error_str, sizeof(ci->error_str), "interrupted by signal"); ci->status = CTL_ISCSI_ERROR; return; } } if (cs->cs_terminating) { CFISCSI_SESSION_UNLOCK(cs); snprintf(ci->error_str, sizeof(ci->error_str), "connection terminating"); ci->status = CTL_ISCSI_ERROR; return; } ip = cs->cs_login_pdu; cs->cs_login_pdu = NULL; CFISCSI_SESSION_UNLOCK(cs); if (ip->ip_data_len > cirp->data_segment_len) { icl_pdu_free(ip); snprintf(ci->error_str, sizeof(ci->error_str), "data segment too big"); ci->status = CTL_ISCSI_ERROR; return; } copyout(ip->ip_bhs, cirp->bhs, sizeof(*ip->ip_bhs)); if (ip->ip_data_len > 0) { data = malloc(ip->ip_data_len, M_CFISCSI, M_WAITOK); icl_pdu_get_data(ip, 0, data, ip->ip_data_len); copyout(data, cirp->data_segment, ip->ip_data_len); free(data, M_CFISCSI); } icl_pdu_free(ip); ci->status = CTL_ISCSI_OK; } #endif /* !ICL_KERNEL_PROXY */ static void cfiscsi_ioctl_port_create(struct ctl_req *req) { struct cfiscsi_target *ct; struct ctl_port *port; - const char *target, *alias, *tags; + const char *target, *alias, *val; struct scsi_vpd_id_descriptor *desc; - ctl_options_t opts; int retval, len, idlen; uint16_t tag; - ctl_init_opts(&opts, req->num_args, req->kern_args); - target = ctl_get_opt(&opts, "cfiscsi_target"); - alias = ctl_get_opt(&opts, "cfiscsi_target_alias"); - tags = ctl_get_opt(&opts, "cfiscsi_portal_group_tag"); - if (target == NULL || tags == NULL) { + target = dnvlist_get_string(req->args_nvl, "cfiscsi_target", NULL); + alias = dnvlist_get_string(req->args_nvl, "cfiscsi_target_alias", NULL); + val = dnvlist_get_string(req->args_nvl, "cfiscsi_portal_group_tag", + NULL); + + + if (target == NULL || val == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Missing required argument"); - ctl_free_opts(&opts); return; } - tag = strtol(tags, (char **)NULL, 10); + + tag = strtoul(val, NULL, 0); ct = cfiscsi_target_find_or_create(&cfiscsi_softc, target, alias, tag); if (ct == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "failed to create target \"%s\"", target); - ctl_free_opts(&opts); return; } if (ct->ct_state == CFISCSI_TARGET_STATE_ACTIVE) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "target \"%s\" for portal group tag %u already exists", target, tag); cfiscsi_target_release(ct); - ctl_free_opts(&opts); return; } port = &ct->ct_port; // WAT if (ct->ct_state == CFISCSI_TARGET_STATE_DYING) goto done; port->frontend = &cfiscsi_frontend; port->port_type = CTL_PORT_ISCSI; /* XXX KDM what should the real number be here? */ port->num_requested_ctl_io = 4096; port->port_name = "iscsi"; - port->physical_port = tag; + port->physical_port = (int)tag; port->virtual_port = ct->ct_target_id; port->port_online = cfiscsi_online; port->port_offline = cfiscsi_offline; port->port_info = cfiscsi_info; port->onoff_arg = ct; port->fe_datamove = cfiscsi_datamove; port->fe_done = cfiscsi_done; port->targ_port = -1; + port->options = nvlist_clone(req->args_nvl); - port->options = opts; - STAILQ_INIT(&opts); - /* Generate Port ID. */ idlen = strlen(target) + strlen(",t,0x0001") + 1; idlen = roundup2(idlen, 4); len = sizeof(struct scsi_vpd_device_id) + idlen; port->port_devid = malloc(sizeof(struct ctl_devid) + len, M_CTL, M_WAITOK | M_ZERO); port->port_devid->len = len; desc = (struct scsi_vpd_id_descriptor *)port->port_devid->data; desc->proto_codeset = (SCSI_PROTO_ISCSI << 4) | SVPD_ID_CODESET_UTF8; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT | SVPD_ID_TYPE_SCSI_NAME; desc->length = idlen; snprintf(desc->identifier, idlen, "%s,t,0x%4.4x", target, tag); /* Generate Target ID. */ idlen = strlen(target) + 1; idlen = roundup2(idlen, 4); len = sizeof(struct scsi_vpd_device_id) + idlen; port->target_devid = malloc(sizeof(struct ctl_devid) + len, M_CTL, M_WAITOK | M_ZERO); port->target_devid->len = len; desc = (struct scsi_vpd_id_descriptor *)port->target_devid->data; desc->proto_codeset = (SCSI_PROTO_ISCSI << 4) | SVPD_ID_CODESET_UTF8; desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_TARGET | SVPD_ID_TYPE_SCSI_NAME; desc->length = idlen; strlcpy(desc->identifier, target, idlen); retval = ctl_port_register(port); if (retval != 0) { - ctl_free_opts(&port->options); free(port->port_devid, M_CFISCSI); free(port->target_devid, M_CFISCSI); cfiscsi_target_release(ct); req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "ctl_port_register() failed with error %d", retval); return; } done: ct->ct_state = CFISCSI_TARGET_STATE_ACTIVE; req->status = CTL_LUN_OK; - memcpy(req->kern_args[0].kvalue, &port->targ_port, - sizeof(port->targ_port)); //XXX + req->result_nvl = nvlist_create(0); + nvlist_add_number(req->result_nvl, "port_id", port->targ_port); } static void cfiscsi_ioctl_port_remove(struct ctl_req *req) { struct cfiscsi_target *ct; - const char *target, *tags; - ctl_options_t opts; + const char *target, *val; uint16_t tag; - ctl_init_opts(&opts, req->num_args, req->kern_args); - target = ctl_get_opt(&opts, "cfiscsi_target"); - tags = ctl_get_opt(&opts, "cfiscsi_portal_group_tag"); - if (target == NULL || tags == NULL) { - ctl_free_opts(&opts); + target = dnvlist_get_string(req->args_nvl, "cfiscsi_target", NULL); + val = dnvlist_get_string(req->args_nvl, "cfiscsi_portal_group_tag", + NULL); + + if (target == NULL || val == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Missing required argument"); return; } - tag = strtol(tags, (char **)NULL, 10); + + tag = strtoul(val, NULL, 0); ct = cfiscsi_target_find(&cfiscsi_softc, target, tag); if (ct == NULL) { - ctl_free_opts(&opts); req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "can't find target \"%s\"", target); return; } if (ct->ct_state != CFISCSI_TARGET_STATE_ACTIVE) { - ctl_free_opts(&opts); req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "target \"%s\" is already dying", target); return; } - ctl_free_opts(&opts); ct->ct_state = CFISCSI_TARGET_STATE_DYING; ctl_port_offline(&ct->ct_port); cfiscsi_target_release(ct); cfiscsi_target_release(ct); req->status = CTL_LUN_OK; } static int cfiscsi_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td) { struct ctl_iscsi *ci; struct ctl_req *req; if (cmd == CTL_PORT_REQ) { req = (struct ctl_req *)addr; switch (req->reqtype) { case CTL_REQ_CREATE: cfiscsi_ioctl_port_create(req); break; case CTL_REQ_REMOVE: cfiscsi_ioctl_port_remove(req); break; default: req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Unsupported request type %d", req->reqtype); } return (0); } if (cmd != CTL_ISCSI) return (ENOTTY); ci = (struct ctl_iscsi *)addr; switch (ci->type) { case CTL_ISCSI_HANDOFF: cfiscsi_ioctl_handoff(ci); break; case CTL_ISCSI_LIST: cfiscsi_ioctl_list(ci); break; case CTL_ISCSI_LOGOUT: cfiscsi_ioctl_logout(ci); break; case CTL_ISCSI_TERMINATE: cfiscsi_ioctl_terminate(ci); break; case CTL_ISCSI_LIMITS: cfiscsi_ioctl_limits(ci); break; #ifdef ICL_KERNEL_PROXY case CTL_ISCSI_LISTEN: cfiscsi_ioctl_listen(ci); break; case CTL_ISCSI_ACCEPT: cfiscsi_ioctl_accept(ci); break; case CTL_ISCSI_SEND: cfiscsi_ioctl_send(ci); break; case CTL_ISCSI_RECEIVE: cfiscsi_ioctl_receive(ci); break; #else case CTL_ISCSI_LISTEN: case CTL_ISCSI_ACCEPT: case CTL_ISCSI_SEND: case CTL_ISCSI_RECEIVE: ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: CTL compiled without ICL_KERNEL_PROXY", __func__); break; #endif /* !ICL_KERNEL_PROXY */ default: ci->status = CTL_ISCSI_ERROR; snprintf(ci->error_str, sizeof(ci->error_str), "%s: invalid iSCSI request type %d", __func__, ci->type); break; } return (0); } static void cfiscsi_target_hold(struct cfiscsi_target *ct) { refcount_acquire(&ct->ct_refcount); } static void cfiscsi_target_release(struct cfiscsi_target *ct) { struct cfiscsi_softc *softc; softc = ct->ct_softc; mtx_lock(&softc->lock); if (refcount_release(&ct->ct_refcount)) { TAILQ_REMOVE(&softc->targets, ct, ct_next); mtx_unlock(&softc->lock); if (ct->ct_state != CFISCSI_TARGET_STATE_INVALID) { ct->ct_state = CFISCSI_TARGET_STATE_INVALID; if (ctl_port_deregister(&ct->ct_port) != 0) printf("%s: ctl_port_deregister() failed\n", __func__); } free(ct, M_CFISCSI); return; } mtx_unlock(&softc->lock); } static struct cfiscsi_target * cfiscsi_target_find(struct cfiscsi_softc *softc, const char *name, uint16_t tag) { struct cfiscsi_target *ct; mtx_lock(&softc->lock); TAILQ_FOREACH(ct, &softc->targets, ct_next) { if (ct->ct_tag != tag || strcmp(name, ct->ct_name) != 0 || ct->ct_state != CFISCSI_TARGET_STATE_ACTIVE) continue; cfiscsi_target_hold(ct); mtx_unlock(&softc->lock); return (ct); } mtx_unlock(&softc->lock); return (NULL); } static struct cfiscsi_target * cfiscsi_target_find_or_create(struct cfiscsi_softc *softc, const char *name, const char *alias, uint16_t tag) { struct cfiscsi_target *ct, *newct; if (name[0] == '\0' || strlen(name) >= CTL_ISCSI_NAME_LEN) return (NULL); newct = malloc(sizeof(*newct), M_CFISCSI, M_WAITOK | M_ZERO); mtx_lock(&softc->lock); TAILQ_FOREACH(ct, &softc->targets, ct_next) { if (ct->ct_tag != tag || strcmp(name, ct->ct_name) != 0 || ct->ct_state == CFISCSI_TARGET_STATE_INVALID) continue; cfiscsi_target_hold(ct); mtx_unlock(&softc->lock); free(newct, M_CFISCSI); return (ct); } strlcpy(newct->ct_name, name, sizeof(newct->ct_name)); if (alias != NULL) strlcpy(newct->ct_alias, alias, sizeof(newct->ct_alias)); newct->ct_tag = tag; refcount_init(&newct->ct_refcount, 1); newct->ct_softc = softc; if (TAILQ_EMPTY(&softc->targets)) softc->last_target_id = 0; newct->ct_target_id = ++softc->last_target_id; TAILQ_INSERT_TAIL(&softc->targets, newct, ct_next); mtx_unlock(&softc->lock); return (newct); } static void cfiscsi_datamove_in(union ctl_io *io) { struct cfiscsi_session *cs; struct icl_pdu *request, *response; const struct iscsi_bhs_scsi_command *bhssc; struct iscsi_bhs_data_in *bhsdi; struct ctl_sg_entry ctl_sg_entry, *ctl_sglist; size_t len, expected_len, sg_len, buffer_offset; const char *sg_addr; int ctl_sg_count, error, i; request = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; cs = PDU_SESSION(request); bhssc = (const struct iscsi_bhs_scsi_command *)request->ip_bhs; KASSERT((bhssc->bhssc_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_COMMAND")); if (io->scsiio.kern_sg_entries > 0) { ctl_sglist = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; ctl_sg_count = io->scsiio.kern_sg_entries; } else { ctl_sglist = &ctl_sg_entry; ctl_sglist->addr = io->scsiio.kern_data_ptr; ctl_sglist->len = io->scsiio.kern_data_len; ctl_sg_count = 1; } /* * This is the total amount of data to be transferred within the current * SCSI command. We need to record it so that we can properly report * underflow/underflow. */ PDU_TOTAL_TRANSFER_LEN(request) = io->scsiio.kern_total_len; /* * This is the offset within the current SCSI command; for the first * call to cfiscsi_datamove() it will be 0, and for subsequent ones * it will be the sum of lengths of previous ones. */ buffer_offset = io->scsiio.kern_rel_offset; /* * This is the transfer length expected by the initiator. In theory, * it could be different from the correct amount of data from the SCSI * point of view, even if that doesn't make any sense. */ expected_len = ntohl(bhssc->bhssc_expected_data_transfer_length); #if 0 if (expected_len != io->scsiio.kern_total_len) { CFISCSI_SESSION_DEBUG(cs, "expected transfer length %zd, " "actual length %zd", expected_len, (size_t)io->scsiio.kern_total_len); } #endif if (buffer_offset >= expected_len) { #if 0 CFISCSI_SESSION_DEBUG(cs, "buffer_offset = %zd, " "already sent the expected len", buffer_offset); #endif io->scsiio.be_move_done(io); return; } i = 0; sg_addr = NULL; sg_len = 0; response = NULL; bhsdi = NULL; for (;;) { if (response == NULL) { response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); ctl_set_busy(&io->scsiio); io->scsiio.be_move_done(io); cfiscsi_session_terminate(cs); return; } bhsdi = (struct iscsi_bhs_data_in *)response->ip_bhs; bhsdi->bhsdi_opcode = ISCSI_BHS_OPCODE_SCSI_DATA_IN; bhsdi->bhsdi_initiator_task_tag = bhssc->bhssc_initiator_task_tag; bhsdi->bhsdi_target_transfer_tag = 0xffffffff; bhsdi->bhsdi_datasn = htonl(PDU_EXPDATASN(request)); PDU_EXPDATASN(request)++; bhsdi->bhsdi_buffer_offset = htonl(buffer_offset); } KASSERT(i < ctl_sg_count, ("i >= ctl_sg_count")); if (sg_len == 0) { sg_addr = ctl_sglist[i].addr; sg_len = ctl_sglist[i].len; KASSERT(sg_len > 0, ("sg_len <= 0")); } len = sg_len; /* * Truncate to maximum data segment length. */ KASSERT(response->ip_data_len < cs->cs_max_send_data_segment_length, ("ip_data_len %zd >= max_send_data_segment_length %d", response->ip_data_len, cs->cs_max_send_data_segment_length)); if (response->ip_data_len + len > cs->cs_max_send_data_segment_length) { len = cs->cs_max_send_data_segment_length - response->ip_data_len; KASSERT(len <= sg_len, ("len %zd > sg_len %zd", len, sg_len)); } /* * Truncate to expected data transfer length. */ KASSERT(buffer_offset + response->ip_data_len < expected_len, ("buffer_offset %zd + ip_data_len %zd >= expected_len %zd", buffer_offset, response->ip_data_len, expected_len)); if (buffer_offset + response->ip_data_len + len > expected_len) { CFISCSI_SESSION_DEBUG(cs, "truncating from %zd " "to expected data transfer length %zd", buffer_offset + response->ip_data_len + len, expected_len); len = expected_len - (buffer_offset + response->ip_data_len); KASSERT(len <= sg_len, ("len %zd > sg_len %zd", len, sg_len)); } error = icl_pdu_append_data(response, sg_addr, len, M_NOWAIT); if (error != 0) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); icl_pdu_free(response); ctl_set_busy(&io->scsiio); io->scsiio.be_move_done(io); cfiscsi_session_terminate(cs); return; } sg_addr += len; sg_len -= len; io->scsiio.kern_data_resid -= len; KASSERT(buffer_offset + response->ip_data_len <= expected_len, ("buffer_offset %zd + ip_data_len %zd > expected_len %zd", buffer_offset, response->ip_data_len, expected_len)); if (buffer_offset + response->ip_data_len == expected_len) { /* * Already have the amount of data the initiator wanted. */ break; } if (sg_len == 0) { /* * End of scatter-gather segment; * proceed to the next one... */ if (i == ctl_sg_count - 1) { /* * ... unless this was the last one. */ break; } i++; } if (response->ip_data_len == cs->cs_max_send_data_segment_length) { /* * Can't stuff more data into the current PDU; * queue it. Note that's not enough to check * for kern_data_resid == 0 instead; there * may be several Data-In PDUs for the final * call to cfiscsi_datamove(), and we want * to set the F flag only on the last of them. */ buffer_offset += response->ip_data_len; if (buffer_offset == io->scsiio.kern_total_len || buffer_offset == expected_len) { buffer_offset -= response->ip_data_len; break; } cfiscsi_pdu_queue(response); response = NULL; bhsdi = NULL; } } if (response != NULL) { buffer_offset += response->ip_data_len; if (buffer_offset == io->scsiio.kern_total_len || buffer_offset == expected_len) { bhsdi->bhsdi_flags |= BHSDI_FLAGS_F; if (io->io_hdr.status == CTL_SUCCESS) { bhsdi->bhsdi_flags |= BHSDI_FLAGS_S; if (PDU_TOTAL_TRANSFER_LEN(request) < ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhsdi->bhsdi_flags |= BHSSR_FLAGS_RESIDUAL_UNDERFLOW; bhsdi->bhsdi_residual_count = htonl(ntohl(bhssc->bhssc_expected_data_transfer_length) - PDU_TOTAL_TRANSFER_LEN(request)); } else if (PDU_TOTAL_TRANSFER_LEN(request) > ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhsdi->bhsdi_flags |= BHSSR_FLAGS_RESIDUAL_OVERFLOW; bhsdi->bhsdi_residual_count = htonl(PDU_TOTAL_TRANSFER_LEN(request) - ntohl(bhssc->bhssc_expected_data_transfer_length)); } bhsdi->bhsdi_status = io->scsiio.scsi_status; io->io_hdr.flags |= CTL_FLAG_STATUS_SENT; } } KASSERT(response->ip_data_len > 0, ("sending empty Data-In")); cfiscsi_pdu_queue(response); } io->scsiio.be_move_done(io); } static void cfiscsi_datamove_out(union ctl_io *io) { struct cfiscsi_session *cs; struct icl_pdu *request, *response; const struct iscsi_bhs_scsi_command *bhssc; struct iscsi_bhs_r2t *bhsr2t; struct cfiscsi_data_wait *cdw; struct ctl_sg_entry ctl_sg_entry, *ctl_sglist; uint32_t expected_len, datamove_len, r2t_off, r2t_len; uint32_t target_transfer_tag; bool done; request = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; cs = PDU_SESSION(request); bhssc = (const struct iscsi_bhs_scsi_command *)request->ip_bhs; KASSERT((bhssc->bhssc_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("bhssc->bhssc_opcode != ISCSI_BHS_OPCODE_SCSI_COMMAND")); /* * We need to record it so that we can properly report * underflow/underflow. */ PDU_TOTAL_TRANSFER_LEN(request) = io->scsiio.kern_total_len; /* * Complete write underflow. Not a single byte to read. Return. */ expected_len = ntohl(bhssc->bhssc_expected_data_transfer_length); if (io->scsiio.kern_rel_offset >= expected_len) { io->scsiio.be_move_done(io); return; } datamove_len = MIN(io->scsiio.kern_data_len, expected_len - io->scsiio.kern_rel_offset); target_transfer_tag = atomic_fetchadd_32(&cs->cs_target_transfer_tag, 1); cdw = cfiscsi_data_wait_new(cs, io, bhssc->bhssc_initiator_task_tag, &target_transfer_tag); if (cdw == NULL) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); ctl_set_busy(&io->scsiio); io->scsiio.be_move_done(io); cfiscsi_session_terminate(cs); return; } #if 0 CFISCSI_SESSION_DEBUG(cs, "expecting Data-Out with initiator " "task tag 0x%x, target transfer tag 0x%x", bhssc->bhssc_initiator_task_tag, target_transfer_tag); #endif cdw->cdw_ctl_io = io; cdw->cdw_target_transfer_tag = target_transfer_tag; cdw->cdw_initiator_task_tag = bhssc->bhssc_initiator_task_tag; cdw->cdw_r2t_end = datamove_len; cdw->cdw_datasn = 0; /* Set initial data pointer for the CDW respecting ext_data_filled. */ if (io->scsiio.kern_sg_entries > 0) { ctl_sglist = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; } else { ctl_sglist = &ctl_sg_entry; ctl_sglist->addr = io->scsiio.kern_data_ptr; ctl_sglist->len = datamove_len; } cdw->cdw_sg_index = 0; cdw->cdw_sg_addr = ctl_sglist[cdw->cdw_sg_index].addr; cdw->cdw_sg_len = ctl_sglist[cdw->cdw_sg_index].len; r2t_off = io->scsiio.ext_data_filled; while (r2t_off > 0) { if (r2t_off >= cdw->cdw_sg_len) { r2t_off -= cdw->cdw_sg_len; cdw->cdw_sg_index++; cdw->cdw_sg_addr = ctl_sglist[cdw->cdw_sg_index].addr; cdw->cdw_sg_len = ctl_sglist[cdw->cdw_sg_index].len; continue; } cdw->cdw_sg_addr += r2t_off; cdw->cdw_sg_len -= r2t_off; r2t_off = 0; } if (cs->cs_immediate_data && io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled < icl_pdu_data_segment_length(request)) { done = cfiscsi_handle_data_segment(request, cdw); if (done) { cfiscsi_data_wait_free(cs, cdw); io->scsiio.be_move_done(io); return; } } r2t_off = io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled; r2t_len = MIN(datamove_len - io->scsiio.ext_data_filled, cs->cs_max_burst_length); cdw->cdw_r2t_end = io->scsiio.ext_data_filled + r2t_len; CFISCSI_SESSION_LOCK(cs); TAILQ_INSERT_TAIL(&cs->cs_waiting_for_data_out, cdw, cdw_next); CFISCSI_SESSION_UNLOCK(cs); /* * XXX: We should limit the number of outstanding R2T PDUs * per task to MaxOutstandingR2T. */ response = cfiscsi_pdu_new_response(request, M_NOWAIT); if (response == NULL) { CFISCSI_SESSION_WARN(cs, "failed to " "allocate memory; dropping connection"); ctl_set_busy(&io->scsiio); io->scsiio.be_move_done(io); cfiscsi_session_terminate(cs); return; } io->io_hdr.flags |= CTL_FLAG_DMA_INPROG; bhsr2t = (struct iscsi_bhs_r2t *)response->ip_bhs; bhsr2t->bhsr2t_opcode = ISCSI_BHS_OPCODE_R2T; bhsr2t->bhsr2t_flags = 0x80; bhsr2t->bhsr2t_lun = bhssc->bhssc_lun; bhsr2t->bhsr2t_initiator_task_tag = bhssc->bhssc_initiator_task_tag; bhsr2t->bhsr2t_target_transfer_tag = target_transfer_tag; /* * XXX: Here we assume that cfiscsi_datamove() won't ever * be running concurrently on several CPUs for a given * command. */ bhsr2t->bhsr2t_r2tsn = htonl(PDU_R2TSN(request)); PDU_R2TSN(request)++; /* * This is the offset within the current SCSI command; * i.e. for the first call of datamove(), it will be 0, * and for subsequent ones it will be the sum of lengths * of previous ones. * * The ext_data_filled is to account for unsolicited * (immediate) data that might have already arrived. */ bhsr2t->bhsr2t_buffer_offset = htonl(r2t_off); /* * This is the total length (sum of S/G lengths) this call * to cfiscsi_datamove() is supposed to handle, limited by * MaxBurstLength. */ bhsr2t->bhsr2t_desired_data_transfer_length = htonl(r2t_len); cfiscsi_pdu_queue(response); } static void cfiscsi_datamove(union ctl_io *io) { if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) cfiscsi_datamove_in(io); else { /* We hadn't received anything during this datamove yet. */ io->scsiio.ext_data_filled = 0; cfiscsi_datamove_out(io); } } static void cfiscsi_scsi_command_done(union ctl_io *io) { struct icl_pdu *request, *response; struct iscsi_bhs_scsi_command *bhssc; struct iscsi_bhs_scsi_response *bhssr; #ifdef DIAGNOSTIC struct cfiscsi_data_wait *cdw; #endif struct cfiscsi_session *cs; uint16_t sense_length; request = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; cs = PDU_SESSION(request); bhssc = (struct iscsi_bhs_scsi_command *)request->ip_bhs; KASSERT((bhssc->bhssc_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_SCSI_COMMAND, ("replying to wrong opcode 0x%x", bhssc->bhssc_opcode)); //CFISCSI_SESSION_DEBUG(cs, "initiator task tag 0x%x", // bhssc->bhssc_initiator_task_tag); #ifdef DIAGNOSTIC CFISCSI_SESSION_LOCK(cs); TAILQ_FOREACH(cdw, &cs->cs_waiting_for_data_out, cdw_next) KASSERT(bhssc->bhssc_initiator_task_tag != cdw->cdw_initiator_task_tag, ("dangling cdw")); CFISCSI_SESSION_UNLOCK(cs); #endif /* * Do not return status for aborted commands. * There are exceptions, but none supported by CTL yet. */ if (((io->io_hdr.flags & CTL_FLAG_ABORT) && (io->io_hdr.flags & CTL_FLAG_ABORT_STATUS) == 0) || (io->io_hdr.flags & CTL_FLAG_STATUS_SENT)) { ctl_free_io(io); icl_pdu_free(request); return; } response = cfiscsi_pdu_new_response(request, M_WAITOK); bhssr = (struct iscsi_bhs_scsi_response *)response->ip_bhs; bhssr->bhssr_opcode = ISCSI_BHS_OPCODE_SCSI_RESPONSE; bhssr->bhssr_flags = 0x80; /* * XXX: We don't deal with bidirectional under/overflows; * does anything actually support those? */ if (PDU_TOTAL_TRANSFER_LEN(request) < ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhssr->bhssr_flags |= BHSSR_FLAGS_RESIDUAL_UNDERFLOW; bhssr->bhssr_residual_count = htonl(ntohl(bhssc->bhssc_expected_data_transfer_length) - PDU_TOTAL_TRANSFER_LEN(request)); //CFISCSI_SESSION_DEBUG(cs, "underflow; residual count %d", // ntohl(bhssr->bhssr_residual_count)); } else if (PDU_TOTAL_TRANSFER_LEN(request) > ntohl(bhssc->bhssc_expected_data_transfer_length)) { bhssr->bhssr_flags |= BHSSR_FLAGS_RESIDUAL_OVERFLOW; bhssr->bhssr_residual_count = htonl(PDU_TOTAL_TRANSFER_LEN(request) - ntohl(bhssc->bhssc_expected_data_transfer_length)); //CFISCSI_SESSION_DEBUG(cs, "overflow; residual count %d", // ntohl(bhssr->bhssr_residual_count)); } bhssr->bhssr_response = BHSSR_RESPONSE_COMMAND_COMPLETED; bhssr->bhssr_status = io->scsiio.scsi_status; bhssr->bhssr_initiator_task_tag = bhssc->bhssc_initiator_task_tag; bhssr->bhssr_expdatasn = htonl(PDU_EXPDATASN(request)); if (io->scsiio.sense_len > 0) { #if 0 CFISCSI_SESSION_DEBUG(cs, "returning %d bytes of sense data", io->scsiio.sense_len); #endif sense_length = htons(io->scsiio.sense_len); icl_pdu_append_data(response, &sense_length, sizeof(sense_length), M_WAITOK); icl_pdu_append_data(response, &io->scsiio.sense_data, io->scsiio.sense_len, M_WAITOK); } ctl_free_io(io); icl_pdu_free(request); cfiscsi_pdu_queue(response); } static void cfiscsi_task_management_done(union ctl_io *io) { struct icl_pdu *request, *response; struct iscsi_bhs_task_management_request *bhstmr; struct iscsi_bhs_task_management_response *bhstmr2; struct cfiscsi_data_wait *cdw, *tmpcdw; struct cfiscsi_session *cs, *tcs; struct cfiscsi_softc *softc; int cold_reset = 0; request = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; cs = PDU_SESSION(request); bhstmr = (struct iscsi_bhs_task_management_request *)request->ip_bhs; KASSERT((bhstmr->bhstmr_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) == ISCSI_BHS_OPCODE_TASK_REQUEST, ("replying to wrong opcode 0x%x", bhstmr->bhstmr_opcode)); #if 0 CFISCSI_SESSION_DEBUG(cs, "initiator task tag 0x%x; referenced task tag 0x%x", bhstmr->bhstmr_initiator_task_tag, bhstmr->bhstmr_referenced_task_tag); #endif if ((bhstmr->bhstmr_function & ~0x80) == BHSTMR_FUNCTION_ABORT_TASK) { /* * Make sure we no longer wait for Data-Out for this command. */ CFISCSI_SESSION_LOCK(cs); TAILQ_FOREACH_SAFE(cdw, &cs->cs_waiting_for_data_out, cdw_next, tmpcdw) { if (bhstmr->bhstmr_referenced_task_tag != cdw->cdw_initiator_task_tag) continue; #if 0 CFISCSI_SESSION_DEBUG(cs, "removing csw for initiator task " "tag 0x%x", bhstmr->bhstmr_initiator_task_tag); #endif TAILQ_REMOVE(&cs->cs_waiting_for_data_out, cdw, cdw_next); io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; cdw->cdw_ctl_io->scsiio.io_hdr.port_status = 43; cdw->cdw_ctl_io->scsiio.be_move_done(cdw->cdw_ctl_io); cfiscsi_data_wait_free(cs, cdw); } CFISCSI_SESSION_UNLOCK(cs); } if ((bhstmr->bhstmr_function & ~0x80) == BHSTMR_FUNCTION_TARGET_COLD_RESET && io->io_hdr.status == CTL_SUCCESS) cold_reset = 1; response = cfiscsi_pdu_new_response(request, M_WAITOK); bhstmr2 = (struct iscsi_bhs_task_management_response *) response->ip_bhs; bhstmr2->bhstmr_opcode = ISCSI_BHS_OPCODE_TASK_RESPONSE; bhstmr2->bhstmr_flags = 0x80; switch (io->taskio.task_status) { case CTL_TASK_FUNCTION_COMPLETE: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_COMPLETE; break; case CTL_TASK_FUNCTION_SUCCEEDED: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_SUCCEEDED; break; case CTL_TASK_LUN_DOES_NOT_EXIST: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_LUN_DOES_NOT_EXIST; break; case CTL_TASK_FUNCTION_NOT_SUPPORTED: default: bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_NOT_SUPPORTED; break; } memcpy(bhstmr2->bhstmr_additional_reponse_information, io->taskio.task_resp, sizeof(io->taskio.task_resp)); bhstmr2->bhstmr_initiator_task_tag = bhstmr->bhstmr_initiator_task_tag; ctl_free_io(io); icl_pdu_free(request); cfiscsi_pdu_queue(response); if (cold_reset) { softc = cs->cs_target->ct_softc; mtx_lock(&softc->lock); TAILQ_FOREACH(tcs, &softc->sessions, cs_next) { if (tcs->cs_target == cs->cs_target) cfiscsi_session_terminate(tcs); } mtx_unlock(&softc->lock); } } static void cfiscsi_done(union ctl_io *io) { struct icl_pdu *request; struct cfiscsi_session *cs; KASSERT(((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE), ("invalid CTL status %#x", io->io_hdr.status)); if (io->io_hdr.io_type == CTL_IO_TASK && io->taskio.task_action == CTL_TASK_I_T_NEXUS_RESET) { /* * Implicit task termination has just completed; nothing to do. */ cs = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; cs->cs_tasks_aborted = true; refcount_release(&cs->cs_outstanding_ctl_pdus); wakeup(__DEVOLATILE(void *, &cs->cs_outstanding_ctl_pdus)); ctl_free_io(io); return; } request = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; cs = PDU_SESSION(request); switch (request->ip_bhs->bhs_opcode & ~ISCSI_BHS_OPCODE_IMMEDIATE) { case ISCSI_BHS_OPCODE_SCSI_COMMAND: cfiscsi_scsi_command_done(io); break; case ISCSI_BHS_OPCODE_TASK_REQUEST: cfiscsi_task_management_done(io); break; default: panic("cfiscsi_done called with wrong opcode 0x%x", request->ip_bhs->bhs_opcode); } refcount_release(&cs->cs_outstanding_ctl_pdus); } Index: head/sys/cam/ctl/ctl_ioctl.h =================================================================== --- head/sys/cam/ctl/ctl_ioctl.h (revision 333445) +++ head/sys/cam/ctl/ctl_ioctl.h (revision 333446) @@ -1,843 +1,817 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003 Silicon Graphics International Corp. * Copyright (c) 2011 Spectra Logic Corporation * Copyright (c) 2014-2017 Alexander Motin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_ioctl.h#4 $ * $FreeBSD$ */ /* * CAM Target Layer ioctl interface. * * Author: Ken Merry */ #ifndef _CTL_IOCTL_H_ #define _CTL_IOCTL_H_ #ifdef ICL_KERNEL_PROXY #include #endif #include +#include #define CTL_DEFAULT_DEV "/dev/cam/ctl" /* * Maximum number of targets we support. */ #define CTL_MAX_TARGETS 1 /* * Maximum target ID we support. */ #define CTL_MAX_TARGID 15 /* * Maximum number of initiators per port. */ #define CTL_MAX_INIT_PER_PORT 2048 /* Hopefully this won't conflict with new misc devices that pop up */ #define CTL_MINOR 225 /* Legacy statistics accumulated for every port for every LU. */ //#define CTL_LEGACY_STATS 1 typedef enum { CTL_DELAY_TYPE_NONE, CTL_DELAY_TYPE_CONT, CTL_DELAY_TYPE_ONESHOT } ctl_delay_type; typedef enum { CTL_DELAY_LOC_NONE, CTL_DELAY_LOC_DATAMOVE, CTL_DELAY_LOC_DONE, } ctl_delay_location; typedef enum { CTL_DELAY_STATUS_NONE, CTL_DELAY_STATUS_OK, CTL_DELAY_STATUS_INVALID_LUN, CTL_DELAY_STATUS_INVALID_TYPE, CTL_DELAY_STATUS_INVALID_LOC, CTL_DELAY_STATUS_NOT_IMPLEMENTED } ctl_delay_status; struct ctl_io_delay_info { uint32_t lun_id; ctl_delay_type delay_type; ctl_delay_location delay_loc; uint32_t delay_secs; ctl_delay_status status; }; typedef enum { CTL_STATS_NO_IO, CTL_STATS_READ, CTL_STATS_WRITE } ctl_stat_types; #define CTL_STATS_NUM_TYPES 3 typedef enum { CTL_SS_OK, CTL_SS_NEED_MORE_SPACE, CTL_SS_ERROR } ctl_stats_status; typedef enum { CTL_STATS_FLAG_NONE = 0x00, CTL_STATS_FLAG_TIME_VALID = 0x01 } ctl_stats_flags; #ifdef CTL_LEGACY_STATS typedef enum { CTL_LUN_STATS_NO_BLOCKSIZE = 0x01 } ctl_lun_stats_flags; struct ctl_lun_io_port_stats { uint32_t targ_port; uint64_t bytes[CTL_STATS_NUM_TYPES]; uint64_t operations[CTL_STATS_NUM_TYPES]; struct bintime time[CTL_STATS_NUM_TYPES]; uint64_t num_dmas[CTL_STATS_NUM_TYPES]; struct bintime dma_time[CTL_STATS_NUM_TYPES]; }; struct ctl_lun_io_stats { uint8_t device_type; uint64_t lun_number; uint32_t blocksize; ctl_lun_stats_flags flags; struct ctl_lun_io_port_stats *ports; }; struct ctl_stats { int alloc_len; /* passed to kernel */ struct ctl_lun_io_stats *lun_stats; /* passed to/from kernel */ int fill_len; /* passed to userland */ int num_luns; /* passed to userland */ ctl_stats_status status; /* passed to userland */ ctl_stats_flags flags; /* passed to userland */ struct timespec timestamp; /* passed to userland */ }; #endif /* CTL_LEGACY_STATS */ struct ctl_io_stats { uint32_t item; uint64_t bytes[CTL_STATS_NUM_TYPES]; uint64_t operations[CTL_STATS_NUM_TYPES]; uint64_t dmas[CTL_STATS_NUM_TYPES]; struct bintime time[CTL_STATS_NUM_TYPES]; struct bintime dma_time[CTL_STATS_NUM_TYPES]; }; struct ctl_get_io_stats { struct ctl_io_stats *stats; /* passed to/from kernel */ size_t alloc_len; /* passed to kernel */ size_t fill_len; /* passed to userland */ int first_item; /* passed to kernel */ int num_items; /* passed to userland */ ctl_stats_status status; /* passed to userland */ ctl_stats_flags flags; /* passed to userland */ struct timespec timestamp; /* passed to userland */ }; /* * The types of errors that can be injected: * * NONE: No error specified. * ABORTED: SSD_KEY_ABORTED_COMMAND, 0x45, 0x00 * MEDIUM_ERR: Medium error, different asc/ascq depending on read/write. * UA: Unit attention. * CUSTOM: User specifies the sense data. * TYPE: Mask to use with error types. * * Flags that affect injection behavior: * CONTINUOUS: This error will stay around until explicitly cleared. * DESCRIPTOR: Use descriptor sense instead of fixed sense. */ typedef enum { CTL_LUN_INJ_NONE = 0x000, CTL_LUN_INJ_ABORTED = 0x001, CTL_LUN_INJ_MEDIUM_ERR = 0x002, CTL_LUN_INJ_UA = 0x003, CTL_LUN_INJ_CUSTOM = 0x004, CTL_LUN_INJ_TYPE = 0x0ff, CTL_LUN_INJ_CONTINUOUS = 0x100, CTL_LUN_INJ_DESCRIPTOR = 0x200 } ctl_lun_error; /* * Flags to specify what type of command the given error pattern will * execute on. The first group of types can be ORed together. * * READ: Any read command. * WRITE: Any write command. * READWRITE: Any read or write command. * READCAP: Any read capacity command. * TUR: Test Unit Ready. * ANY: Any command. * MASK: Mask for basic command patterns. * * Special types: * * CMD: The CDB to act on is specified in struct ctl_error_desc_cmd. * RANGE: For read/write commands, act when the LBA is in the * specified range. */ typedef enum { CTL_LUN_PAT_NONE = 0x000, CTL_LUN_PAT_READ = 0x001, CTL_LUN_PAT_WRITE = 0x002, CTL_LUN_PAT_READWRITE = CTL_LUN_PAT_READ | CTL_LUN_PAT_WRITE, CTL_LUN_PAT_READCAP = 0x004, CTL_LUN_PAT_TUR = 0x008, CTL_LUN_PAT_ANY = 0x0ff, CTL_LUN_PAT_MASK = 0x0ff, CTL_LUN_PAT_CMD = 0x100, CTL_LUN_PAT_RANGE = 0x200 } ctl_lun_error_pattern; /* * This structure allows the user to specify a particular CDB pattern to * look for. * * cdb_pattern: Fill in the relevant bytes to look for in the CDB. * cdb_valid_bytes: Bitmask specifying valid bytes in the cdb_pattern. * flags: Specify any command flags (see ctl_io_flags) that * should be set. */ struct ctl_error_desc_cmd { uint8_t cdb_pattern[CTL_MAX_CDBLEN]; uint32_t cdb_valid_bytes; uint32_t flags; }; /* * Error injection descriptor. * * lun_id LUN to act on. * lun_error: The type of error to inject. See above for descriptions. * error_pattern: What kind of command to act on. See above. * cmd_desc: For CTL_LUN_PAT_CMD only. * lba_range: For CTL_LUN_PAT_RANGE only. * custom_sense: Specify sense. For CTL_LUN_INJ_CUSTOM only. * serial: Serial number returned by the kernel. Use for deletion. * links: Kernel use only. */ struct ctl_error_desc { uint32_t lun_id; /* To kernel */ ctl_lun_error lun_error; /* To kernel */ ctl_lun_error_pattern error_pattern; /* To kernel */ struct ctl_error_desc_cmd cmd_desc; /* To kernel */ struct ctl_lba_len lba_range; /* To kernel */ struct scsi_sense_data custom_sense; /* To kernel */ uint64_t serial; /* From kernel */ STAILQ_ENTRY(ctl_error_desc) links; /* Kernel use only */ }; typedef enum { CTL_OOA_FLAG_NONE = 0x00, CTL_OOA_FLAG_ALL_LUNS = 0x01 } ctl_ooa_flags; typedef enum { CTL_OOA_OK, CTL_OOA_NEED_MORE_SPACE, CTL_OOA_ERROR } ctl_get_ooa_status; typedef enum { CTL_OOACMD_FLAG_NONE = 0x00, CTL_OOACMD_FLAG_DMA = 0x01, CTL_OOACMD_FLAG_BLOCKED = 0x02, CTL_OOACMD_FLAG_ABORT = 0x04, CTL_OOACMD_FLAG_RTR = 0x08, CTL_OOACMD_FLAG_DMA_QUEUED = 0x10 } ctl_ooa_cmd_flags; struct ctl_ooa_entry { ctl_ooa_cmd_flags cmd_flags; uint8_t cdb[CTL_MAX_CDBLEN]; uint8_t cdb_len; uint32_t tag_num; uint32_t lun_num; struct bintime start_bt; }; struct ctl_ooa { ctl_ooa_flags flags; /* passed to kernel */ uint64_t lun_num; /* passed to kernel */ uint32_t alloc_len; /* passed to kernel */ uint32_t alloc_num; /* passed to kernel */ struct ctl_ooa_entry *entries; /* filled in kernel */ uint32_t fill_len; /* passed to userland */ uint32_t fill_num; /* passed to userland */ uint32_t dropped_num; /* passed to userland */ struct bintime cur_bt; /* passed to userland */ ctl_get_ooa_status status; /* passed to userland */ }; typedef enum { CTL_LUN_NOSTATUS, CTL_LUN_OK, CTL_LUN_ERROR, CTL_LUN_WARNING } ctl_lun_status; #define CTL_ERROR_STR_LEN 160 -#define CTL_BEARG_RD 0x01 -#define CTL_BEARG_WR 0x02 -#define CTL_BEARG_RW (CTL_BEARG_RD|CTL_BEARG_WR) -#define CTL_BEARG_ASCII 0x04 - -/* - * Backend Argument: - * - * namelen: Length of the name field, including the terminating NUL. - * - * name: Name of the parameter. This must be NUL-terminated. - * - * flags: Flags for the parameter, see above for values. - * - * vallen: Length of the value in bytes, including the terminating NUL. - * - * value: Value to be set/fetched. This must be NUL-terminated. - * - * kname: For kernel use only. - * - * kvalue: For kernel use only. - */ -struct ctl_be_arg { - unsigned int namelen; - char *name; - int flags; - unsigned int vallen; - void *value; - - char *kname; - void *kvalue; -}; - typedef enum { CTL_LUNREQ_CREATE, CTL_LUNREQ_RM, CTL_LUNREQ_MODIFY, } ctl_lunreq_type; /* * The ID_REQ flag is used to say that the caller has requested a * particular LUN ID in the req_lun_id field. If we cannot allocate that * LUN ID, the ctl_add_lun() call will fail. * * The STOPPED flag tells us that the LUN should default to the powered * off state. It will return 0x04,0x02 until it is powered up. ("Logical * unit not ready, initializing command required.") * * The NO_MEDIA flag tells us that the LUN has no media inserted. * * The PRIMARY flag tells us that this LUN is registered as a Primary LUN * which is accessible via the Master shelf controller in an HA. This flag * being set indicates a Primary LUN. This flag being reset represents a * Secondary LUN controlled by the Secondary controller in an HA * configuration. Flag is applicable at this time to T_DIRECT types. * * The SERIAL_NUM flag tells us that the serial_num field is filled in and * valid for use in SCSI INQUIRY VPD page 0x80. * * The DEVID flag tells us that the device_id field is filled in and * valid for use in SCSI INQUIRY VPD page 0x83. * * The DEV_TYPE flag tells us that the device_type field is filled in. * * The EJECTED flag tells us that the removable LUN has tray open. * * The UNMAP flag tells us that this LUN supports UNMAP. * * The OFFLINE flag tells us that this LUN can not access backing store. */ typedef enum { CTL_LUN_FLAG_ID_REQ = 0x01, CTL_LUN_FLAG_STOPPED = 0x02, CTL_LUN_FLAG_NO_MEDIA = 0x04, CTL_LUN_FLAG_PRIMARY = 0x08, CTL_LUN_FLAG_SERIAL_NUM = 0x10, CTL_LUN_FLAG_DEVID = 0x20, CTL_LUN_FLAG_DEV_TYPE = 0x40, CTL_LUN_FLAG_UNMAP = 0x80, CTL_LUN_FLAG_EJECTED = 0x100, CTL_LUN_FLAG_READONLY = 0x200 } ctl_backend_lun_flags; /* * LUN creation parameters: * * flags: Various LUN flags, see above. * * device_type: The SCSI device type. e.g. 0 for Direct Access, * 3 for Processor, etc. Only certain backends may * support setting this field. The CTL_LUN_FLAG_DEV_TYPE * flag should be set in the flags field if the device * type is set. * * lun_size_bytes: The size of the LUN in bytes. For some backends * this is relevant (e.g. ramdisk), for others, it may * be ignored in favor of using the properties of the * backing store. If specified, this should be a * multiple of the blocksize. * * The actual size of the LUN is returned in this * field. * * blocksize_bytes: The LUN blocksize in bytes. For some backends this * is relevant, for others it may be ignored in * favor of using the properties of the backing store. * * The actual blocksize of the LUN is returned in this * field. * * req_lun_id: The requested LUN ID. The CTL_LUN_FLAG_ID_REQ flag * should be set if this is set. The request will be * granted if the LUN number is available, otherwise * the LUN addition request will fail. * * The allocated LUN number is returned in this field. * * serial_num: This is the value returned in SCSI INQUIRY VPD page * 0x80. If it is specified, the CTL_LUN_FLAG_SERIAL_NUM * flag should be set. * * The serial number value used is returned in this * field. * * device_id: This is the value returned in the T10 vendor ID * based DESIGNATOR field in the SCSI INQUIRY VPD page * 0x83 data. If it is specified, the CTL_LUN_FLAG_DEVID * flag should be set. * * The device id value used is returned in this field. */ struct ctl_lun_create_params { ctl_backend_lun_flags flags; uint8_t device_type; uint64_t lun_size_bytes; uint32_t blocksize_bytes; uint32_t req_lun_id; uint8_t serial_num[CTL_SN_LEN]; uint8_t device_id[CTL_DEVID_LEN]; }; /* * LUN removal parameters: * * lun_id: The number of the LUN to delete. This must be set. * The LUN must be backed by the given backend. */ struct ctl_lun_rm_params { uint32_t lun_id; }; /* * LUN modification parameters: * * lun_id: The number of the LUN to modify. This must be set. * The LUN must be backed by the given backend. * * lun_size_bytes: The size of the LUN in bytes. If zero, update * the size using the backing file size, if possible. */ struct ctl_lun_modify_params { uint32_t lun_id; uint64_t lun_size_bytes; }; /* * Union of request type data. Fill in the appropriate union member for * the request type. */ union ctl_lunreq_data { struct ctl_lun_create_params create; struct ctl_lun_rm_params rm; struct ctl_lun_modify_params modify; }; /* * LUN request interface: * * backend: This is required, and is NUL-terminated a string * that is the name of the backend, like "ramdisk" or * "block". * * reqtype: The type of request, CTL_LUNREQ_CREATE to create a * LUN, CTL_LUNREQ_RM to delete a LUN. * * reqdata: Request type-specific information. See the * description of individual the union members above * for more information. * * num_be_args: This is the number of backend-specific arguments * in the be_args array. * * be_args: This is an array of backend-specific arguments. * See above for a description of the fields in this * structure. * * status: Status of the LUN request. * * error_str: If the status is CTL_LUN_ERROR, this will * contain a string describing the error. * * kern_be_args: For kernel use only. */ struct ctl_lun_req { #define CTL_BE_NAME_LEN 32 char backend[CTL_BE_NAME_LEN]; ctl_lunreq_type reqtype; union ctl_lunreq_data reqdata; - int num_be_args; - struct ctl_be_arg *be_args; + void * args; + nvlist_t * args_nvl; + size_t args_len; + void * result; + nvlist_t * result_nvl; + size_t result_len; ctl_lun_status status; char error_str[CTL_ERROR_STR_LEN]; - struct ctl_be_arg *kern_be_args; }; /* * LUN list status: * * NONE: No status. * * OK: Request completed successfully. * * NEED_MORE_SPACE: The allocated length of the entries field is too * small for the available data. * * ERROR: An error occurred, look at the error string for a * description of the error. */ typedef enum { CTL_LUN_LIST_NONE, CTL_LUN_LIST_OK, CTL_LUN_LIST_NEED_MORE_SPACE, CTL_LUN_LIST_ERROR } ctl_lun_list_status; /* * LUN list interface * * backend_name: This is a NUL-terminated string. If the string * length is 0, then all LUNs on all backends will * be enumerated. Otherwise this is the name of the * backend to be enumerated, like "ramdisk" or "block". * * alloc_len: The length of the data buffer allocated for entries. * In order to properly size the buffer, make one call * with alloc_len set to 0, and then use the returned * dropped_len as the buffer length to allocate and * pass in on a subsequent call. * * lun_xml: XML-formatted information on the requested LUNs. * * fill_len: The amount of data filled in the storage for entries. * * status: The status of the request. See above for the * description of the values of this field. * * error_str: If the status indicates an error, this string will * be filled in to describe the error. */ struct ctl_lun_list { char backend[CTL_BE_NAME_LEN]; /* passed to kernel*/ uint32_t alloc_len; /* passed to kernel */ char *lun_xml; /* filled in kernel */ uint32_t fill_len; /* passed to userland */ ctl_lun_list_status status; /* passed to userland */ char error_str[CTL_ERROR_STR_LEN]; /* passed to userland */ }; /* * Port request interface: * * driver: This is required, and is NUL-terminated a string * that is the name of the frontend, like "iscsi" . * * reqtype: The type of request, CTL_REQ_CREATE to create a * port, CTL_REQ_REMOVE to delete a port. * * num_be_args: This is the number of frontend-specific arguments * in the be_args array. * * be_args: This is an array of frontend-specific arguments. * See above for a description of the fields in this * structure. * * status: Status of the request. * * error_str: If the status is CTL_LUN_ERROR, this will * contain a string describing the error. * * kern_be_args: For kernel use only. */ typedef enum { CTL_REQ_CREATE, CTL_REQ_REMOVE, CTL_REQ_MODIFY, } ctl_req_type; struct ctl_req { char driver[CTL_DRIVER_NAME_LEN]; ctl_req_type reqtype; - int num_args; - struct ctl_be_arg *args; + void * args; + nvlist_t * args_nvl; + size_t args_len; + void * result; + nvlist_t * result_nvl; + size_t result_len; ctl_lun_status status; char error_str[CTL_ERROR_STR_LEN]; - struct ctl_be_arg *kern_args; }; /* * iSCSI status * * OK: Request completed successfully. * * ERROR: An error occurred, look at the error string for a * description of the error. * * CTL_ISCSI_LIST_NEED_MORE_SPACE: * User has to pass larger buffer for CTL_ISCSI_LIST ioctl. */ typedef enum { CTL_ISCSI_OK, CTL_ISCSI_ERROR, CTL_ISCSI_LIST_NEED_MORE_SPACE, CTL_ISCSI_SESSION_NOT_FOUND } ctl_iscsi_status; typedef enum { CTL_ISCSI_HANDOFF, CTL_ISCSI_LIST, CTL_ISCSI_LOGOUT, CTL_ISCSI_TERMINATE, CTL_ISCSI_LIMITS, #if defined(ICL_KERNEL_PROXY) || 1 /* * We actually need those in all cases, but leave the ICL_KERNEL_PROXY, * to remember to remove them along with rest of proxy code, eventually. */ CTL_ISCSI_LISTEN, CTL_ISCSI_ACCEPT, CTL_ISCSI_SEND, CTL_ISCSI_RECEIVE, #endif } ctl_iscsi_type; typedef enum { CTL_ISCSI_DIGEST_NONE, CTL_ISCSI_DIGEST_CRC32C } ctl_iscsi_digest; #define CTL_ISCSI_NAME_LEN 224 /* 223 bytes, by RFC 3720, + '\0' */ #define CTL_ISCSI_ADDR_LEN 47 /* INET6_ADDRSTRLEN + '\0' */ #define CTL_ISCSI_ALIAS_LEN 128 /* Arbitrary. */ #define CTL_ISCSI_OFFLOAD_LEN 8 /* Arbitrary. */ struct ctl_iscsi_handoff_params { char initiator_name[CTL_ISCSI_NAME_LEN]; char initiator_addr[CTL_ISCSI_ADDR_LEN]; char initiator_alias[CTL_ISCSI_ALIAS_LEN]; uint8_t initiator_isid[6]; char target_name[CTL_ISCSI_NAME_LEN]; int socket; int portal_group_tag; /* * Connection parameters negotiated by ctld(8). */ ctl_iscsi_digest header_digest; ctl_iscsi_digest data_digest; uint32_t cmdsn; uint32_t statsn; int max_recv_data_segment_length; int max_burst_length; int first_burst_length; uint32_t immediate_data; char offload[CTL_ISCSI_OFFLOAD_LEN]; #ifdef ICL_KERNEL_PROXY int connection_id; #else int spare; #endif int max_send_data_segment_length; }; struct ctl_iscsi_list_params { uint32_t alloc_len; /* passed to kernel */ char *conn_xml; /* filled in kernel */ uint32_t fill_len; /* passed to userland */ int spare[4]; }; struct ctl_iscsi_logout_params { int connection_id; /* passed to kernel */ char initiator_name[CTL_ISCSI_NAME_LEN]; /* passed to kernel */ char initiator_addr[CTL_ISCSI_ADDR_LEN]; /* passed to kernel */ int all; /* passed to kernel */ int spare[4]; }; struct ctl_iscsi_terminate_params { int connection_id; /* passed to kernel */ char initiator_name[CTL_ISCSI_NAME_LEN]; /* passed to kernel */ char initiator_addr[CTL_ISCSI_NAME_LEN]; /* passed to kernel */ int all; /* passed to kernel */ int spare[4]; }; struct ctl_iscsi_limits_params { /* passed to kernel */ char offload[CTL_ISCSI_OFFLOAD_LEN]; /* passed to userland */ size_t spare; int max_recv_data_segment_length; int max_send_data_segment_length; int max_burst_length; int first_burst_length; }; #ifdef ICL_KERNEL_PROXY struct ctl_iscsi_listen_params { int iser; int domain; int socktype; int protocol; struct sockaddr *addr; socklen_t addrlen; int portal_id; int spare[4]; }; struct ctl_iscsi_accept_params { int connection_id; int portal_id; struct sockaddr *initiator_addr; socklen_t initiator_addrlen; int spare[4]; }; struct ctl_iscsi_send_params { int connection_id; void *bhs; size_t spare; void *spare2; size_t data_segment_len; void *data_segment; int spare3[4]; }; struct ctl_iscsi_receive_params { int connection_id; void *bhs; size_t spare; void *spare2; size_t data_segment_len; void *data_segment; int spare3[4]; }; #endif /* ICL_KERNEL_PROXY */ union ctl_iscsi_data { struct ctl_iscsi_handoff_params handoff; struct ctl_iscsi_list_params list; struct ctl_iscsi_logout_params logout; struct ctl_iscsi_terminate_params terminate; struct ctl_iscsi_limits_params limits; #ifdef ICL_KERNEL_PROXY struct ctl_iscsi_listen_params listen; struct ctl_iscsi_accept_params accept; struct ctl_iscsi_send_params send; struct ctl_iscsi_receive_params receive; #endif }; /* * iSCSI interface * * status: The status of the request. See above for the * description of the values of this field. * * error_str: If the status indicates an error, this string will * be filled in to describe the error. */ struct ctl_iscsi { ctl_iscsi_type type; /* passed to kernel */ union ctl_iscsi_data data; /* passed to kernel */ ctl_iscsi_status status; /* passed to userland */ char error_str[CTL_ERROR_STR_LEN]; /* passed to userland */ }; struct ctl_lun_map { uint32_t port; uint32_t plun; uint32_t lun; }; #define CTL_IO _IOWR(CTL_MINOR, 0x00, union ctl_io) #define CTL_ENABLE_PORT _IOW(CTL_MINOR, 0x04, struct ctl_port_entry) #define CTL_DISABLE_PORT _IOW(CTL_MINOR, 0x05, struct ctl_port_entry) #define CTL_DELAY_IO _IOWR(CTL_MINOR, 0x10, struct ctl_io_delay_info) #define CTL_GETSTATS _IOWR(CTL_MINOR, 0x15, struct ctl_stats) #define CTL_ERROR_INJECT _IOWR(CTL_MINOR, 0x16, struct ctl_error_desc) #define CTL_GET_OOA _IOWR(CTL_MINOR, 0x18, struct ctl_ooa) #define CTL_DUMP_STRUCTS _IO(CTL_MINOR, 0x19) #define CTL_LUN_REQ _IOWR(CTL_MINOR, 0x21, struct ctl_lun_req) #define CTL_LUN_LIST _IOWR(CTL_MINOR, 0x22, struct ctl_lun_list) #define CTL_ERROR_INJECT_DELETE _IOW(CTL_MINOR, 0x23, struct ctl_error_desc) #define CTL_SET_PORT_WWNS _IOW(CTL_MINOR, 0x24, struct ctl_port_entry) #define CTL_ISCSI _IOWR(CTL_MINOR, 0x25, struct ctl_iscsi) #define CTL_PORT_REQ _IOWR(CTL_MINOR, 0x26, struct ctl_req) #define CTL_PORT_LIST _IOWR(CTL_MINOR, 0x27, struct ctl_lun_list) #define CTL_LUN_MAP _IOW(CTL_MINOR, 0x28, struct ctl_lun_map) #define CTL_GET_LUN_STATS _IOWR(CTL_MINOR, 0x29, struct ctl_get_io_stats) #define CTL_GET_PORT_STATS _IOWR(CTL_MINOR, 0x2a, struct ctl_get_io_stats) #endif /* _CTL_IOCTL_H_ */ /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_tpc.c =================================================================== --- head/sys/cam/ctl/ctl_tpc.c (revision 333445) +++ head/sys/cam/ctl/ctl_tpc.c (revision 333446) @@ -1,2473 +1,2475 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2014 Alexander Motin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer, * without modification, immediately at the beginning of the file. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TPC_MAX_CSCDS 64 #define TPC_MAX_SEGS 64 #define TPC_MAX_SEG 0 #define TPC_MAX_LIST 8192 #define TPC_MAX_INLINE 0 #define TPC_MAX_LISTS 255 #define TPC_MAX_IO_SIZE (1024 * 1024) #define TPC_MAX_IOCHUNK_SIZE (TPC_MAX_IO_SIZE * 16) #define TPC_MIN_TOKEN_TIMEOUT 1 #define TPC_DFL_TOKEN_TIMEOUT 60 #define TPC_MAX_TOKEN_TIMEOUT 600 MALLOC_DEFINE(M_CTL_TPC, "ctltpc", "CTL TPC"); typedef enum { TPC_ERR_RETRY = 0x000, TPC_ERR_FAIL = 0x001, TPC_ERR_MASK = 0x0ff, TPC_ERR_NO_DECREMENT = 0x100 } tpc_error_action; struct tpc_list; TAILQ_HEAD(runl, tpc_io); struct tpc_io { union ctl_io *io; uint8_t target; uint32_t cscd; uint64_t lun; uint8_t *buf; struct tpc_list *list; struct runl run; TAILQ_ENTRY(tpc_io) rlinks; TAILQ_ENTRY(tpc_io) links; }; struct tpc_token { uint8_t token[512]; uint64_t lun; uint32_t blocksize; uint8_t *params; struct scsi_range_desc *range; int nrange; int active; time_t last_active; uint32_t timeout; TAILQ_ENTRY(tpc_token) links; }; struct tpc_list { uint8_t service_action; int init_port; uint32_t init_idx; uint32_t list_id; uint8_t flags; uint8_t *params; struct scsi_ec_cscd *cscd; struct scsi_ec_segment *seg[TPC_MAX_SEGS]; uint8_t *inl; int ncscd; int nseg; int leninl; struct tpc_token *token; struct scsi_range_desc *range; int nrange; off_t offset_into_rod; int curseg; off_t cursectors; off_t curbytes; int curops; int stage; off_t segsectors; off_t segbytes; int tbdio; int error; int abort; int completed; time_t last_active; TAILQ_HEAD(, tpc_io) allio; struct scsi_sense_data fwd_sense_data; uint8_t fwd_sense_len; uint8_t fwd_scsi_status; uint8_t fwd_target; uint16_t fwd_cscd; struct scsi_sense_data sense_data; uint8_t sense_len; uint8_t scsi_status; struct ctl_scsiio *ctsio; struct ctl_lun *lun; int res_token_valid; uint8_t res_token[512]; TAILQ_ENTRY(tpc_list) links; }; static void tpc_timeout(void *arg) { struct ctl_softc *softc = arg; struct ctl_lun *lun; struct tpc_token *token, *ttoken; struct tpc_list *list, *tlist; /* Free completed lists with expired timeout. */ STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_lock(&lun->lun_lock); TAILQ_FOREACH_SAFE(list, &lun->tpc_lists, links, tlist) { if (!list->completed || time_uptime < list->last_active + TPC_DFL_TOKEN_TIMEOUT) continue; TAILQ_REMOVE(&lun->tpc_lists, list, links); free(list, M_CTL); } mtx_unlock(&lun->lun_lock); } /* Free inactive ROD tokens with expired timeout. */ mtx_lock(&softc->tpc_lock); TAILQ_FOREACH_SAFE(token, &softc->tpc_tokens, links, ttoken) { if (token->active || time_uptime < token->last_active + token->timeout + 1) continue; TAILQ_REMOVE(&softc->tpc_tokens, token, links); free(token->params, M_CTL); free(token, M_CTL); } mtx_unlock(&softc->tpc_lock); callout_schedule(&softc->tpc_timeout, hz); } void ctl_tpc_init(struct ctl_softc *softc) { mtx_init(&softc->tpc_lock, "CTL TPC mutex", NULL, MTX_DEF); TAILQ_INIT(&softc->tpc_tokens); callout_init_mtx(&softc->tpc_timeout, &softc->ctl_lock, 0); callout_reset(&softc->tpc_timeout, hz, tpc_timeout, softc); } void ctl_tpc_shutdown(struct ctl_softc *softc) { struct tpc_token *token; callout_drain(&softc->tpc_timeout); /* Free ROD tokens. */ mtx_lock(&softc->tpc_lock); while ((token = TAILQ_FIRST(&softc->tpc_tokens)) != NULL) { TAILQ_REMOVE(&softc->tpc_tokens, token, links); free(token->params, M_CTL); free(token, M_CTL); } mtx_unlock(&softc->tpc_lock); mtx_destroy(&softc->tpc_lock); } void ctl_tpc_lun_init(struct ctl_lun *lun) { TAILQ_INIT(&lun->tpc_lists); } void ctl_tpc_lun_clear(struct ctl_lun *lun, uint32_t initidx) { struct tpc_list *list, *tlist; TAILQ_FOREACH_SAFE(list, &lun->tpc_lists, links, tlist) { if (initidx != -1 && list->init_idx != initidx) continue; if (!list->completed) continue; TAILQ_REMOVE(&lun->tpc_lists, list, links); free(list, M_CTL); } } void ctl_tpc_lun_shutdown(struct ctl_lun *lun) { struct ctl_softc *softc = lun->ctl_softc; struct tpc_list *list; struct tpc_token *token, *ttoken; /* Free lists for this LUN. */ while ((list = TAILQ_FIRST(&lun->tpc_lists)) != NULL) { TAILQ_REMOVE(&lun->tpc_lists, list, links); KASSERT(list->completed, ("Not completed TPC (%p) on shutdown", list)); free(list, M_CTL); } /* Free ROD tokens for this LUN. */ mtx_lock(&softc->tpc_lock); TAILQ_FOREACH_SAFE(token, &softc->tpc_tokens, links, ttoken) { if (token->lun != lun->lun || token->active) continue; TAILQ_REMOVE(&softc->tpc_tokens, token, links); free(token->params, M_CTL); free(token, M_CTL); } mtx_unlock(&softc->tpc_lock); } int ctl_inquiry_evpd_tpc(struct ctl_scsiio *ctsio, int alloc_len) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_vpd_tpc *tpc_ptr; struct scsi_vpd_tpc_descriptor *d_ptr; struct scsi_vpd_tpc_descriptor_bdrl *bdrl_ptr; struct scsi_vpd_tpc_descriptor_sc *sc_ptr; struct scsi_vpd_tpc_descriptor_sc_descr *scd_ptr; struct scsi_vpd_tpc_descriptor_pd *pd_ptr; struct scsi_vpd_tpc_descriptor_sd *sd_ptr; struct scsi_vpd_tpc_descriptor_sdid *sdid_ptr; struct scsi_vpd_tpc_descriptor_rtf *rtf_ptr; struct scsi_vpd_tpc_descriptor_rtf_block *rtfb_ptr; struct scsi_vpd_tpc_descriptor_srt *srt_ptr; struct scsi_vpd_tpc_descriptor_srtd *srtd_ptr; struct scsi_vpd_tpc_descriptor_gco *gco_ptr; int data_len; data_len = sizeof(struct scsi_vpd_tpc) + sizeof(struct scsi_vpd_tpc_descriptor_bdrl) + roundup2(sizeof(struct scsi_vpd_tpc_descriptor_sc) + 2 * sizeof(struct scsi_vpd_tpc_descriptor_sc_descr) + 11, 4) + sizeof(struct scsi_vpd_tpc_descriptor_pd) + roundup2(sizeof(struct scsi_vpd_tpc_descriptor_sd) + 4, 4) + roundup2(sizeof(struct scsi_vpd_tpc_descriptor_sdid) + 2, 4) + sizeof(struct scsi_vpd_tpc_descriptor_rtf) + sizeof(struct scsi_vpd_tpc_descriptor_rtf_block) + sizeof(struct scsi_vpd_tpc_descriptor_srt) + 2*sizeof(struct scsi_vpd_tpc_descriptor_srtd) + sizeof(struct scsi_vpd_tpc_descriptor_gco); ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO); tpc_ptr = (struct scsi_vpd_tpc *)ctsio->kern_data_ptr; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->kern_data_len = min(data_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; /* * The control device is always connected. The disk device, on the * other hand, may not be online all the time. */ if (lun != NULL) tpc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else tpc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; tpc_ptr->page_code = SVPD_SCSI_TPC; scsi_ulto2b(data_len - 4, tpc_ptr->page_length); /* Block Device ROD Limits */ d_ptr = (struct scsi_vpd_tpc_descriptor *)&tpc_ptr->descr[0]; bdrl_ptr = (struct scsi_vpd_tpc_descriptor_bdrl *)d_ptr; scsi_ulto2b(SVPD_TPC_BDRL, bdrl_ptr->desc_type); scsi_ulto2b(sizeof(*bdrl_ptr) - 4, bdrl_ptr->desc_length); scsi_ulto2b(TPC_MAX_SEGS, bdrl_ptr->maximum_ranges); scsi_ulto4b(TPC_MAX_TOKEN_TIMEOUT, bdrl_ptr->maximum_inactivity_timeout); scsi_ulto4b(TPC_DFL_TOKEN_TIMEOUT, bdrl_ptr->default_inactivity_timeout); scsi_u64to8b(0, bdrl_ptr->maximum_token_transfer_size); scsi_u64to8b(0, bdrl_ptr->optimal_transfer_count); /* Supported commands */ d_ptr = (struct scsi_vpd_tpc_descriptor *) (&d_ptr->parameters[0] + scsi_2btoul(d_ptr->desc_length)); sc_ptr = (struct scsi_vpd_tpc_descriptor_sc *)d_ptr; scsi_ulto2b(SVPD_TPC_SC, sc_ptr->desc_type); sc_ptr->list_length = 2 * sizeof(*scd_ptr) + 11; scsi_ulto2b(roundup2(1 + sc_ptr->list_length, 4), sc_ptr->desc_length); scd_ptr = &sc_ptr->descr[0]; scd_ptr->opcode = EXTENDED_COPY; scd_ptr->sa_length = 5; scd_ptr->supported_service_actions[0] = EC_EC_LID1; scd_ptr->supported_service_actions[1] = EC_EC_LID4; scd_ptr->supported_service_actions[2] = EC_PT; scd_ptr->supported_service_actions[3] = EC_WUT; scd_ptr->supported_service_actions[4] = EC_COA; scd_ptr = (struct scsi_vpd_tpc_descriptor_sc_descr *) &scd_ptr->supported_service_actions[scd_ptr->sa_length]; scd_ptr->opcode = RECEIVE_COPY_STATUS; scd_ptr->sa_length = 6; scd_ptr->supported_service_actions[0] = RCS_RCS_LID1; scd_ptr->supported_service_actions[1] = RCS_RCFD; scd_ptr->supported_service_actions[2] = RCS_RCS_LID4; scd_ptr->supported_service_actions[3] = RCS_RCOP; scd_ptr->supported_service_actions[4] = RCS_RRTI; scd_ptr->supported_service_actions[5] = RCS_RART; /* Parameter data. */ d_ptr = (struct scsi_vpd_tpc_descriptor *) (&d_ptr->parameters[0] + scsi_2btoul(d_ptr->desc_length)); pd_ptr = (struct scsi_vpd_tpc_descriptor_pd *)d_ptr; scsi_ulto2b(SVPD_TPC_PD, pd_ptr->desc_type); scsi_ulto2b(sizeof(*pd_ptr) - 4, pd_ptr->desc_length); scsi_ulto2b(TPC_MAX_CSCDS, pd_ptr->maximum_cscd_descriptor_count); scsi_ulto2b(TPC_MAX_SEGS, pd_ptr->maximum_segment_descriptor_count); scsi_ulto4b(TPC_MAX_LIST, pd_ptr->maximum_descriptor_list_length); scsi_ulto4b(TPC_MAX_INLINE, pd_ptr->maximum_inline_data_length); /* Supported Descriptors */ d_ptr = (struct scsi_vpd_tpc_descriptor *) (&d_ptr->parameters[0] + scsi_2btoul(d_ptr->desc_length)); sd_ptr = (struct scsi_vpd_tpc_descriptor_sd *)d_ptr; scsi_ulto2b(SVPD_TPC_SD, sd_ptr->desc_type); scsi_ulto2b(roundup2(sizeof(*sd_ptr) - 4 + 4, 4), sd_ptr->desc_length); sd_ptr->list_length = 4; sd_ptr->supported_descriptor_codes[0] = EC_SEG_B2B; sd_ptr->supported_descriptor_codes[1] = EC_SEG_VERIFY; sd_ptr->supported_descriptor_codes[2] = EC_SEG_REGISTER_KEY; sd_ptr->supported_descriptor_codes[3] = EC_CSCD_ID; /* Supported CSCD Descriptor IDs */ d_ptr = (struct scsi_vpd_tpc_descriptor *) (&d_ptr->parameters[0] + scsi_2btoul(d_ptr->desc_length)); sdid_ptr = (struct scsi_vpd_tpc_descriptor_sdid *)d_ptr; scsi_ulto2b(SVPD_TPC_SDID, sdid_ptr->desc_type); scsi_ulto2b(roundup2(sizeof(*sdid_ptr) - 4 + 2, 4), sdid_ptr->desc_length); scsi_ulto2b(2, sdid_ptr->list_length); scsi_ulto2b(0xffff, &sdid_ptr->supported_descriptor_ids[0]); /* ROD Token Features */ d_ptr = (struct scsi_vpd_tpc_descriptor *) (&d_ptr->parameters[0] + scsi_2btoul(d_ptr->desc_length)); rtf_ptr = (struct scsi_vpd_tpc_descriptor_rtf *)d_ptr; scsi_ulto2b(SVPD_TPC_RTF, rtf_ptr->desc_type); scsi_ulto2b(sizeof(*rtf_ptr) - 4 + sizeof(*rtfb_ptr), rtf_ptr->desc_length); rtf_ptr->remote_tokens = 0; scsi_ulto4b(TPC_MIN_TOKEN_TIMEOUT, rtf_ptr->minimum_token_lifetime); scsi_ulto4b(UINT32_MAX, rtf_ptr->maximum_token_lifetime); scsi_ulto4b(TPC_MAX_TOKEN_TIMEOUT, rtf_ptr->maximum_token_inactivity_timeout); scsi_ulto2b(sizeof(*rtfb_ptr), rtf_ptr->type_specific_features_length); rtfb_ptr = (struct scsi_vpd_tpc_descriptor_rtf_block *) &rtf_ptr->type_specific_features; rtfb_ptr->type_format = SVPD_TPC_RTF_BLOCK; scsi_ulto2b(sizeof(*rtfb_ptr) - 4, rtfb_ptr->desc_length); scsi_ulto2b(0, rtfb_ptr->optimal_length_granularity); scsi_u64to8b(0, rtfb_ptr->maximum_bytes); scsi_u64to8b(0, rtfb_ptr->optimal_bytes); scsi_u64to8b(UINT64_MAX, rtfb_ptr->optimal_bytes_to_token_per_segment); scsi_u64to8b(TPC_MAX_IOCHUNK_SIZE, rtfb_ptr->optimal_bytes_from_token_per_segment); /* Supported ROD Tokens */ d_ptr = (struct scsi_vpd_tpc_descriptor *) (&d_ptr->parameters[0] + scsi_2btoul(d_ptr->desc_length)); srt_ptr = (struct scsi_vpd_tpc_descriptor_srt *)d_ptr; scsi_ulto2b(SVPD_TPC_SRT, srt_ptr->desc_type); scsi_ulto2b(sizeof(*srt_ptr) - 4 + 2*sizeof(*srtd_ptr), srt_ptr->desc_length); scsi_ulto2b(2*sizeof(*srtd_ptr), srt_ptr->rod_type_descriptors_length); srtd_ptr = (struct scsi_vpd_tpc_descriptor_srtd *) &srt_ptr->rod_type_descriptors; scsi_ulto4b(ROD_TYPE_AUR, srtd_ptr->rod_type); srtd_ptr->flags = SVPD_TPC_SRTD_TIN | SVPD_TPC_SRTD_TOUT; scsi_ulto2b(0, srtd_ptr->preference_indicator); srtd_ptr++; scsi_ulto4b(ROD_TYPE_BLOCK_ZERO, srtd_ptr->rod_type); srtd_ptr->flags = SVPD_TPC_SRTD_TIN; scsi_ulto2b(0, srtd_ptr->preference_indicator); /* General Copy Operations */ d_ptr = (struct scsi_vpd_tpc_descriptor *) (&d_ptr->parameters[0] + scsi_2btoul(d_ptr->desc_length)); gco_ptr = (struct scsi_vpd_tpc_descriptor_gco *)d_ptr; scsi_ulto2b(SVPD_TPC_GCO, gco_ptr->desc_type); scsi_ulto2b(sizeof(*gco_ptr) - 4, gco_ptr->desc_length); scsi_ulto4b(TPC_MAX_LISTS, gco_ptr->total_concurrent_copies); scsi_ulto4b(TPC_MAX_LISTS, gco_ptr->maximum_identified_concurrent_copies); scsi_ulto4b(TPC_MAX_SEG, gco_ptr->maximum_segment_length); gco_ptr->data_segment_granularity = 0; gco_ptr->inline_data_granularity = 0; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_receive_copy_operating_parameters(struct ctl_scsiio *ctsio) { struct scsi_receive_copy_operating_parameters *cdb; struct scsi_receive_copy_operating_parameters_data *data; int retval; int alloc_len, total_len; CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n")); cdb = (struct scsi_receive_copy_operating_parameters *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; total_len = sizeof(*data) + 4; alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_receive_copy_operating_parameters_data *)ctsio->kern_data_ptr; scsi_ulto4b(sizeof(*data) - 4 + 4, data->length); data->snlid = RCOP_SNLID; scsi_ulto2b(TPC_MAX_CSCDS, data->maximum_cscd_descriptor_count); scsi_ulto2b(TPC_MAX_SEGS, data->maximum_segment_descriptor_count); scsi_ulto4b(TPC_MAX_LIST, data->maximum_descriptor_list_length); scsi_ulto4b(TPC_MAX_SEG, data->maximum_segment_length); scsi_ulto4b(TPC_MAX_INLINE, data->maximum_inline_data_length); scsi_ulto4b(0, data->held_data_limit); scsi_ulto4b(0, data->maximum_stream_device_transfer_size); scsi_ulto2b(TPC_MAX_LISTS, data->total_concurrent_copies); data->maximum_concurrent_copies = TPC_MAX_LISTS; data->data_segment_granularity = 0; data->inline_data_granularity = 0; data->held_data_granularity = 0; data->implemented_descriptor_list_length = 4; data->list_of_implemented_descriptor_type_codes[0] = EC_SEG_B2B; data->list_of_implemented_descriptor_type_codes[1] = EC_SEG_VERIFY; data->list_of_implemented_descriptor_type_codes[2] = EC_SEG_REGISTER_KEY; data->list_of_implemented_descriptor_type_codes[3] = EC_CSCD_ID; ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } static struct tpc_list * tpc_find_list(struct ctl_lun *lun, uint32_t list_id, uint32_t init_idx) { struct tpc_list *list; mtx_assert(&lun->lun_lock, MA_OWNED); TAILQ_FOREACH(list, &lun->tpc_lists, links) { if ((list->flags & EC_LIST_ID_USAGE_MASK) != EC_LIST_ID_USAGE_NONE && list->list_id == list_id && list->init_idx == init_idx) break; } return (list); } int ctl_receive_copy_status_lid1(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_receive_copy_status_lid1 *cdb; struct scsi_receive_copy_status_lid1_data *data; struct tpc_list *list; struct tpc_list list_copy; int retval; int alloc_len, total_len; uint32_t list_id; CTL_DEBUG_PRINT(("ctl_receive_copy_status_lid1\n")); cdb = (struct scsi_receive_copy_status_lid1 *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; list_id = cdb->list_identifier; mtx_lock(&lun->lun_lock); list = tpc_find_list(lun, list_id, ctl_get_initindex(&ctsio->io_hdr.nexus)); if (list == NULL) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (retval); } list_copy = *list; if (list->completed) { TAILQ_REMOVE(&lun->tpc_lists, list, links); free(list, M_CTL); } mtx_unlock(&lun->lun_lock); total_len = sizeof(*data); alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_receive_copy_status_lid1_data *)ctsio->kern_data_ptr; scsi_ulto4b(sizeof(*data) - 4, data->available_data); if (list_copy.completed) { if (list_copy.error || list_copy.abort) data->copy_command_status = RCS_CCS_ERROR; else data->copy_command_status = RCS_CCS_COMPLETED; } else data->copy_command_status = RCS_CCS_INPROG; scsi_ulto2b(list_copy.curseg, data->segments_processed); if (list_copy.curbytes <= UINT32_MAX) { data->transfer_count_units = RCS_TC_BYTES; scsi_ulto4b(list_copy.curbytes, data->transfer_count); } else { data->transfer_count_units = RCS_TC_MBYTES; scsi_ulto4b(list_copy.curbytes >> 20, data->transfer_count); } ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } int ctl_receive_copy_failure_details(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_receive_copy_failure_details *cdb; struct scsi_receive_copy_failure_details_data *data; struct tpc_list *list; struct tpc_list list_copy; int retval; int alloc_len, total_len; uint32_t list_id; CTL_DEBUG_PRINT(("ctl_receive_copy_failure_details\n")); cdb = (struct scsi_receive_copy_failure_details *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; list_id = cdb->list_identifier; mtx_lock(&lun->lun_lock); list = tpc_find_list(lun, list_id, ctl_get_initindex(&ctsio->io_hdr.nexus)); if (list == NULL || !list->completed) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (retval); } list_copy = *list; TAILQ_REMOVE(&lun->tpc_lists, list, links); free(list, M_CTL); mtx_unlock(&lun->lun_lock); total_len = sizeof(*data) + list_copy.sense_len; alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_receive_copy_failure_details_data *)ctsio->kern_data_ptr; if (list_copy.completed && (list_copy.error || list_copy.abort)) { scsi_ulto4b(sizeof(*data) - 4 + list_copy.sense_len, data->available_data); data->copy_command_status = RCS_CCS_ERROR; } else scsi_ulto4b(0, data->available_data); scsi_ulto2b(list_copy.sense_len, data->sense_data_length); memcpy(data->sense_data, &list_copy.sense_data, list_copy.sense_len); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } int ctl_receive_copy_status_lid4(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_receive_copy_status_lid4 *cdb; struct scsi_receive_copy_status_lid4_data *data; struct tpc_list *list; struct tpc_list list_copy; int retval; int alloc_len, total_len; uint32_t list_id; CTL_DEBUG_PRINT(("ctl_receive_copy_status_lid4\n")); cdb = (struct scsi_receive_copy_status_lid4 *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; list_id = scsi_4btoul(cdb->list_identifier); mtx_lock(&lun->lun_lock); list = tpc_find_list(lun, list_id, ctl_get_initindex(&ctsio->io_hdr.nexus)); if (list == NULL) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (retval); } list_copy = *list; if (list->completed) { TAILQ_REMOVE(&lun->tpc_lists, list, links); free(list, M_CTL); } mtx_unlock(&lun->lun_lock); total_len = sizeof(*data) + list_copy.sense_len; alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_receive_copy_status_lid4_data *)ctsio->kern_data_ptr; scsi_ulto4b(sizeof(*data) - 4 + list_copy.sense_len, data->available_data); data->response_to_service_action = list_copy.service_action; if (list_copy.completed) { if (list_copy.error) data->copy_command_status = RCS_CCS_ERROR; else if (list_copy.abort) data->copy_command_status = RCS_CCS_ABORTED; else data->copy_command_status = RCS_CCS_COMPLETED; } else data->copy_command_status = RCS_CCS_INPROG_FG; scsi_ulto2b(list_copy.curops, data->operation_counter); scsi_ulto4b(UINT32_MAX, data->estimated_status_update_delay); data->transfer_count_units = RCS_TC_BYTES; scsi_u64to8b(list_copy.curbytes, data->transfer_count); scsi_ulto2b(list_copy.curseg, data->segments_processed); data->length_of_the_sense_data_field = list_copy.sense_len; data->sense_data_length = list_copy.sense_len; memcpy(data->sense_data, &list_copy.sense_data, list_copy.sense_len); ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } int ctl_copy_operation_abort(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_copy_operation_abort *cdb; struct tpc_list *list; int retval; uint32_t list_id; CTL_DEBUG_PRINT(("ctl_copy_operation_abort\n")); cdb = (struct scsi_copy_operation_abort *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; list_id = scsi_4btoul(cdb->list_identifier); mtx_lock(&lun->lun_lock); list = tpc_find_list(lun, list_id, ctl_get_initindex(&ctsio->io_hdr.nexus)); if (list == NULL) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (retval); } list->abort = 1; mtx_unlock(&lun->lun_lock); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (retval); } static uint64_t tpc_resolve(struct tpc_list *list, uint16_t idx, uint32_t *ss, uint32_t *pb, uint32_t *pbo) { if (idx == 0xffff) { if (ss && list->lun->be_lun) *ss = list->lun->be_lun->blocksize; if (pb && list->lun->be_lun) *pb = list->lun->be_lun->blocksize << list->lun->be_lun->pblockexp; if (pbo && list->lun->be_lun) *pbo = list->lun->be_lun->blocksize * list->lun->be_lun->pblockoff; return (list->lun->lun); } if (idx >= list->ncscd) return (UINT64_MAX); return (tpcl_resolve(list->lun->ctl_softc, list->init_port, &list->cscd[idx], ss, pb, pbo)); } static void tpc_set_io_error_sense(struct tpc_list *list) { int flen; uint8_t csi[4]; uint8_t sks[3]; uint8_t fbuf[4 + 64]; scsi_ulto4b(list->curseg, csi); if (list->fwd_cscd <= 0x07ff) { sks[0] = SSD_SKS_SEGMENT_VALID; scsi_ulto2b((uint8_t *)&list->cscd[list->fwd_cscd] - list->params, &sks[1]); } else sks[0] = 0; if (list->fwd_scsi_status) { fbuf[0] = 0x0c; fbuf[2] = list->fwd_target; flen = list->fwd_sense_len; if (flen > 64) { flen = 64; fbuf[2] |= SSD_FORWARDED_FSDT; } fbuf[1] = 2 + flen; fbuf[3] = list->fwd_scsi_status; bcopy(&list->fwd_sense_data, &fbuf[4], flen); flen += 4; } else flen = 0; ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x0d, /*ascq*/ 0x01, SSD_ELEM_COMMAND, sizeof(csi), csi, sks[0] ? SSD_ELEM_SKS : SSD_ELEM_SKIP, sizeof(sks), sks, flen ? SSD_ELEM_DESC : SSD_ELEM_SKIP, flen, fbuf, SSD_ELEM_NONE); } static int tpc_process_b2b(struct tpc_list *list) { struct scsi_ec_segment_b2b *seg; struct scsi_ec_cscd_dtsp *sdstp, *ddstp; struct tpc_io *tior, *tiow; struct runl run; uint64_t sl, dl; off_t srclba, dstlba, numbytes, donebytes, roundbytes; int numlba; uint32_t srcblock, dstblock, pb, pbo, adj; uint16_t scscd, dcscd; uint8_t csi[4]; scsi_ulto4b(list->curseg, csi); if (list->stage == 1) { while ((tior = TAILQ_FIRST(&list->allio)) != NULL) { TAILQ_REMOVE(&list->allio, tior, links); ctl_free_io(tior->io); free(tior->buf, M_CTL); free(tior, M_CTL); } if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { tpc_set_io_error_sense(list); return (CTL_RETVAL_ERROR); } list->cursectors += list->segsectors; list->curbytes += list->segbytes; return (CTL_RETVAL_COMPLETE); } TAILQ_INIT(&list->allio); seg = (struct scsi_ec_segment_b2b *)list->seg[list->curseg]; scscd = scsi_2btoul(seg->src_cscd); dcscd = scsi_2btoul(seg->dst_cscd); sl = tpc_resolve(list, scscd, &srcblock, NULL, NULL); dl = tpc_resolve(list, dcscd, &dstblock, &pb, &pbo); if (sl == UINT64_MAX || dl == UINT64_MAX) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x08, /*ascq*/ 0x04, SSD_ELEM_COMMAND, sizeof(csi), csi, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } if (pbo > 0) pbo = pb - pbo; sdstp = &list->cscd[scscd].dtsp; if (scsi_3btoul(sdstp->block_length) != 0) srcblock = scsi_3btoul(sdstp->block_length); ddstp = &list->cscd[dcscd].dtsp; if (scsi_3btoul(ddstp->block_length) != 0) dstblock = scsi_3btoul(ddstp->block_length); numlba = scsi_2btoul(seg->number_of_blocks); if (seg->flags & EC_SEG_DC) numbytes = (off_t)numlba * dstblock; else numbytes = (off_t)numlba * srcblock; srclba = scsi_8btou64(seg->src_lba); dstlba = scsi_8btou64(seg->dst_lba); // printf("Copy %ju bytes from %ju @ %ju to %ju @ %ju\n", // (uintmax_t)numbytes, sl, scsi_8btou64(seg->src_lba), // dl, scsi_8btou64(seg->dst_lba)); if (numbytes == 0) return (CTL_RETVAL_COMPLETE); if (numbytes % srcblock != 0 || numbytes % dstblock != 0) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x26, /*ascq*/ 0x0A, SSD_ELEM_COMMAND, sizeof(csi), csi, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } list->segbytes = numbytes; list->segsectors = numbytes / dstblock; donebytes = 0; TAILQ_INIT(&run); list->tbdio = 0; while (donebytes < numbytes) { roundbytes = numbytes - donebytes; if (roundbytes > TPC_MAX_IO_SIZE) { roundbytes = TPC_MAX_IO_SIZE; roundbytes -= roundbytes % dstblock; if (pb > dstblock) { adj = (dstlba * dstblock + roundbytes - pbo) % pb; if (roundbytes > adj) roundbytes -= adj; } } tior = malloc(sizeof(*tior), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tior->run); tior->buf = malloc(roundbytes, M_CTL, M_WAITOK); tior->list = list; TAILQ_INSERT_TAIL(&list->allio, tior, links); tior->io = tpcl_alloc_io(); ctl_scsi_read_write(tior->io, /*data_ptr*/ tior->buf, /*data_len*/ roundbytes, /*read_op*/ 1, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ srclba, /*num_blocks*/ roundbytes / srcblock, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tior->io->io_hdr.retries = 3; tior->target = SSD_FORWARDED_SDS_EXSRC; tior->cscd = scscd; tior->lun = sl; tior->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tior; tiow = malloc(sizeof(*tior), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tiow->run); tiow->list = list; TAILQ_INSERT_TAIL(&list->allio, tiow, links); tiow->io = tpcl_alloc_io(); ctl_scsi_read_write(tiow->io, /*data_ptr*/ tior->buf, /*data_len*/ roundbytes, /*read_op*/ 0, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ dstlba, /*num_blocks*/ roundbytes / dstblock, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tiow->io->io_hdr.retries = 3; tiow->target = SSD_FORWARDED_SDS_EXDST; tiow->cscd = dcscd; tiow->lun = dl; tiow->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tiow; TAILQ_INSERT_TAIL(&tior->run, tiow, rlinks); TAILQ_INSERT_TAIL(&run, tior, rlinks); list->tbdio++; donebytes += roundbytes; srclba += roundbytes / srcblock; dstlba += roundbytes / dstblock; } while ((tior = TAILQ_FIRST(&run)) != NULL) { TAILQ_REMOVE(&run, tior, rlinks); if (tpcl_queue(tior->io, tior->lun) != CTL_RETVAL_COMPLETE) panic("tpcl_queue() error"); } list->stage++; return (CTL_RETVAL_QUEUED); } static int tpc_process_verify(struct tpc_list *list) { struct scsi_ec_segment_verify *seg; struct tpc_io *tio; uint64_t sl; uint16_t cscd; uint8_t csi[4]; scsi_ulto4b(list->curseg, csi); if (list->stage == 1) { while ((tio = TAILQ_FIRST(&list->allio)) != NULL) { TAILQ_REMOVE(&list->allio, tio, links); ctl_free_io(tio->io); free(tio, M_CTL); } if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { tpc_set_io_error_sense(list); return (CTL_RETVAL_ERROR); } else return (CTL_RETVAL_COMPLETE); } TAILQ_INIT(&list->allio); seg = (struct scsi_ec_segment_verify *)list->seg[list->curseg]; cscd = scsi_2btoul(seg->src_cscd); sl = tpc_resolve(list, cscd, NULL, NULL, NULL); if (sl == UINT64_MAX) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x08, /*ascq*/ 0x04, SSD_ELEM_COMMAND, sizeof(csi), csi, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } // printf("Verify %ju\n", sl); if ((seg->tur & 0x01) == 0) return (CTL_RETVAL_COMPLETE); list->tbdio = 1; tio = malloc(sizeof(*tio), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tio->run); tio->list = list; TAILQ_INSERT_TAIL(&list->allio, tio, links); tio->io = tpcl_alloc_io(); ctl_scsi_tur(tio->io, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tio->io->io_hdr.retries = 3; tio->target = SSD_FORWARDED_SDS_EXSRC; tio->cscd = cscd; tio->lun = sl; tio->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tio; list->stage++; if (tpcl_queue(tio->io, tio->lun) != CTL_RETVAL_COMPLETE) panic("tpcl_queue() error"); return (CTL_RETVAL_QUEUED); } static int tpc_process_register_key(struct tpc_list *list) { struct scsi_ec_segment_register_key *seg; struct tpc_io *tio; uint64_t dl; int datalen; uint16_t cscd; uint8_t csi[4]; scsi_ulto4b(list->curseg, csi); if (list->stage == 1) { while ((tio = TAILQ_FIRST(&list->allio)) != NULL) { TAILQ_REMOVE(&list->allio, tio, links); ctl_free_io(tio->io); free(tio->buf, M_CTL); free(tio, M_CTL); } if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { tpc_set_io_error_sense(list); return (CTL_RETVAL_ERROR); } else return (CTL_RETVAL_COMPLETE); } TAILQ_INIT(&list->allio); seg = (struct scsi_ec_segment_register_key *)list->seg[list->curseg]; cscd = scsi_2btoul(seg->dst_cscd); dl = tpc_resolve(list, cscd, NULL, NULL, NULL); if (dl == UINT64_MAX) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x08, /*ascq*/ 0x04, SSD_ELEM_COMMAND, sizeof(csi), csi, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } // printf("Register Key %ju\n", dl); list->tbdio = 1; tio = malloc(sizeof(*tio), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tio->run); tio->list = list; TAILQ_INSERT_TAIL(&list->allio, tio, links); tio->io = tpcl_alloc_io(); datalen = sizeof(struct scsi_per_res_out_parms); tio->buf = malloc(datalen, M_CTL, M_WAITOK); ctl_scsi_persistent_res_out(tio->io, tio->buf, datalen, SPRO_REGISTER, -1, scsi_8btou64(seg->res_key), scsi_8btou64(seg->sa_res_key), /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tio->io->io_hdr.retries = 3; tio->target = SSD_FORWARDED_SDS_EXDST; tio->cscd = cscd; tio->lun = dl; tio->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tio; list->stage++; if (tpcl_queue(tio->io, tio->lun) != CTL_RETVAL_COMPLETE) panic("tpcl_queue() error"); return (CTL_RETVAL_QUEUED); } static off_t tpc_ranges_length(struct scsi_range_desc *range, int nrange) { off_t length = 0; int r; for (r = 0; r < nrange; r++) length += scsi_4btoul(range[r].length); return (length); } static int tpc_check_ranges_l(struct scsi_range_desc *range, int nrange, uint64_t maxlba, uint64_t *lba) { uint64_t b1; uint32_t l1; int i; for (i = 0; i < nrange; i++) { b1 = scsi_8btou64(range[i].lba); l1 = scsi_4btoul(range[i].length); if (b1 + l1 < b1 || b1 + l1 > maxlba + 1) { *lba = MAX(b1, maxlba + 1); return (-1); } } return (0); } static int tpc_check_ranges_x(struct scsi_range_desc *range, int nrange) { uint64_t b1, b2; uint32_t l1, l2; int i, j; for (i = 0; i < nrange - 1; i++) { b1 = scsi_8btou64(range[i].lba); l1 = scsi_4btoul(range[i].length); for (j = i + 1; j < nrange; j++) { b2 = scsi_8btou64(range[j].lba); l2 = scsi_4btoul(range[j].length); if (b1 + l1 > b2 && b2 + l2 > b1) return (-1); } } return (0); } static int tpc_skip_ranges(struct scsi_range_desc *range, int nrange, off_t skip, int *srange, off_t *soffset) { off_t off; int r; r = 0; off = 0; while (r < nrange) { if (skip - off < scsi_4btoul(range[r].length)) { *srange = r; *soffset = skip - off; return (0); } off += scsi_4btoul(range[r].length); r++; } return (-1); } static int tpc_process_wut(struct tpc_list *list) { struct tpc_io *tio, *tior, *tiow; struct runl run; int drange, srange; off_t doffset, soffset; off_t srclba, dstlba, numbytes, donebytes, roundbytes; uint32_t srcblock, dstblock, pb, pbo, adj; if (list->stage > 0) { /* Cleanup after previous rounds. */ while ((tio = TAILQ_FIRST(&list->allio)) != NULL) { TAILQ_REMOVE(&list->allio, tio, links); ctl_free_io(tio->io); free(tio->buf, M_CTL); free(tio, M_CTL); } if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { if (list->fwd_scsi_status) { list->ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; list->ctsio->scsi_status = list->fwd_scsi_status; list->ctsio->sense_data = list->fwd_sense_data; list->ctsio->sense_len = list->fwd_sense_len; } else { ctl_set_invalid_field(list->ctsio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); } return (CTL_RETVAL_ERROR); } list->cursectors += list->segsectors; list->curbytes += list->segbytes; } /* Check where we are on destination ranges list. */ if (tpc_skip_ranges(list->range, list->nrange, list->cursectors, &drange, &doffset) != 0) return (CTL_RETVAL_COMPLETE); dstblock = list->lun->be_lun->blocksize; pb = dstblock << list->lun->be_lun->pblockexp; if (list->lun->be_lun->pblockoff > 0) pbo = pb - dstblock * list->lun->be_lun->pblockoff; else pbo = 0; /* Check where we are on source ranges list. */ srcblock = list->token->blocksize; if (tpc_skip_ranges(list->token->range, list->token->nrange, list->offset_into_rod + list->cursectors * dstblock / srcblock, &srange, &soffset) != 0) { ctl_set_invalid_field(list->ctsio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); return (CTL_RETVAL_ERROR); } srclba = scsi_8btou64(list->token->range[srange].lba) + soffset; dstlba = scsi_8btou64(list->range[drange].lba) + doffset; numbytes = srcblock * (scsi_4btoul(list->token->range[srange].length) - soffset); numbytes = omin(numbytes, dstblock * (scsi_4btoul(list->range[drange].length) - doffset)); if (numbytes > TPC_MAX_IOCHUNK_SIZE) { numbytes = TPC_MAX_IOCHUNK_SIZE; numbytes -= numbytes % dstblock; if (pb > dstblock) { adj = (dstlba * dstblock + numbytes - pbo) % pb; if (numbytes > adj) numbytes -= adj; } } if (numbytes % srcblock != 0 || numbytes % dstblock != 0) { ctl_set_invalid_field(list->ctsio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); return (CTL_RETVAL_ERROR); } list->segbytes = numbytes; list->segsectors = numbytes / dstblock; //printf("Copy chunk of %ju sectors from %ju to %ju\n", list->segsectors, // srclba, dstlba); donebytes = 0; TAILQ_INIT(&run); list->tbdio = 0; TAILQ_INIT(&list->allio); while (donebytes < numbytes) { roundbytes = numbytes - donebytes; if (roundbytes > TPC_MAX_IO_SIZE) { roundbytes = TPC_MAX_IO_SIZE; roundbytes -= roundbytes % dstblock; if (pb > dstblock) { adj = (dstlba * dstblock + roundbytes - pbo) % pb; if (roundbytes > adj) roundbytes -= adj; } } tior = malloc(sizeof(*tior), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tior->run); tior->buf = malloc(roundbytes, M_CTL, M_WAITOK); tior->list = list; TAILQ_INSERT_TAIL(&list->allio, tior, links); tior->io = tpcl_alloc_io(); ctl_scsi_read_write(tior->io, /*data_ptr*/ tior->buf, /*data_len*/ roundbytes, /*read_op*/ 1, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ srclba, /*num_blocks*/ roundbytes / srcblock, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tior->io->io_hdr.retries = 3; tior->lun = list->token->lun; tior->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tior; tiow = malloc(sizeof(*tiow), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tiow->run); tiow->list = list; TAILQ_INSERT_TAIL(&list->allio, tiow, links); tiow->io = tpcl_alloc_io(); ctl_scsi_read_write(tiow->io, /*data_ptr*/ tior->buf, /*data_len*/ roundbytes, /*read_op*/ 0, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ dstlba, /*num_blocks*/ roundbytes / dstblock, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tiow->io->io_hdr.retries = 3; tiow->lun = list->lun->lun; tiow->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tiow; TAILQ_INSERT_TAIL(&tior->run, tiow, rlinks); TAILQ_INSERT_TAIL(&run, tior, rlinks); list->tbdio++; donebytes += roundbytes; srclba += roundbytes / srcblock; dstlba += roundbytes / dstblock; } while ((tior = TAILQ_FIRST(&run)) != NULL) { TAILQ_REMOVE(&run, tior, rlinks); if (tpcl_queue(tior->io, tior->lun) != CTL_RETVAL_COMPLETE) panic("tpcl_queue() error"); } list->stage++; return (CTL_RETVAL_QUEUED); } static int tpc_process_zero_wut(struct tpc_list *list) { struct tpc_io *tio, *tiow; struct runl run, *prun; int r; uint32_t dstblock, len; if (list->stage > 0) { complete: /* Cleanup after previous rounds. */ while ((tio = TAILQ_FIRST(&list->allio)) != NULL) { TAILQ_REMOVE(&list->allio, tio, links); ctl_free_io(tio->io); free(tio, M_CTL); } if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { if (list->fwd_scsi_status) { list->ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; list->ctsio->scsi_status = list->fwd_scsi_status; list->ctsio->sense_data = list->fwd_sense_data; list->ctsio->sense_len = list->fwd_sense_len; } else { ctl_set_invalid_field(list->ctsio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); } return (CTL_RETVAL_ERROR); } list->cursectors += list->segsectors; list->curbytes += list->segbytes; return (CTL_RETVAL_COMPLETE); } dstblock = list->lun->be_lun->blocksize; TAILQ_INIT(&run); prun = &run; list->tbdio = 1; TAILQ_INIT(&list->allio); list->segsectors = 0; for (r = 0; r < list->nrange; r++) { len = scsi_4btoul(list->range[r].length); if (len == 0) continue; tiow = malloc(sizeof(*tiow), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tiow->run); tiow->list = list; TAILQ_INSERT_TAIL(&list->allio, tiow, links); tiow->io = tpcl_alloc_io(); ctl_scsi_write_same(tiow->io, /*data_ptr*/ NULL, /*data_len*/ 0, /*byte2*/ SWS_NDOB, /*lba*/ scsi_8btou64(list->range[r].lba), /*num_blocks*/ len, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tiow->io->io_hdr.retries = 3; tiow->lun = list->lun->lun; tiow->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tiow; TAILQ_INSERT_TAIL(prun, tiow, rlinks); prun = &tiow->run; list->segsectors += len; } list->segbytes = list->segsectors * dstblock; if (TAILQ_EMPTY(&run)) goto complete; while ((tiow = TAILQ_FIRST(&run)) != NULL) { TAILQ_REMOVE(&run, tiow, rlinks); if (tpcl_queue(tiow->io, tiow->lun) != CTL_RETVAL_COMPLETE) panic("tpcl_queue() error"); } list->stage++; return (CTL_RETVAL_QUEUED); } static void tpc_process(struct tpc_list *list) { struct ctl_lun *lun = list->lun; struct ctl_softc *softc = lun->ctl_softc; struct scsi_ec_segment *seg; struct ctl_scsiio *ctsio = list->ctsio; int retval = CTL_RETVAL_COMPLETE; uint8_t csi[4]; if (list->service_action == EC_WUT) { if (list->token != NULL) retval = tpc_process_wut(list); else retval = tpc_process_zero_wut(list); if (retval == CTL_RETVAL_QUEUED) return; if (retval == CTL_RETVAL_ERROR) { list->error = 1; goto done; } } else { //printf("ZZZ %d cscd, %d segs\n", list->ncscd, list->nseg); while (list->curseg < list->nseg) { seg = list->seg[list->curseg]; switch (seg->type_code) { case EC_SEG_B2B: retval = tpc_process_b2b(list); break; case EC_SEG_VERIFY: retval = tpc_process_verify(list); break; case EC_SEG_REGISTER_KEY: retval = tpc_process_register_key(list); break; default: scsi_ulto4b(list->curseg, csi); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x26, /*ascq*/ 0x09, SSD_ELEM_COMMAND, sizeof(csi), csi, SSD_ELEM_NONE); goto done; } if (retval == CTL_RETVAL_QUEUED) return; if (retval == CTL_RETVAL_ERROR) { list->error = 1; goto done; } list->curseg++; list->stage = 0; } } ctl_set_success(ctsio); done: //printf("ZZZ done\n"); free(list->params, M_CTL); list->params = NULL; if (list->token) { mtx_lock(&softc->tpc_lock); if (--list->token->active == 0) list->token->last_active = time_uptime; mtx_unlock(&softc->tpc_lock); list->token = NULL; } mtx_lock(&lun->lun_lock); if ((list->flags & EC_LIST_ID_USAGE_MASK) == EC_LIST_ID_USAGE_NONE) { TAILQ_REMOVE(&lun->tpc_lists, list, links); free(list, M_CTL); } else { list->completed = 1; list->last_active = time_uptime; list->sense_data = ctsio->sense_data; list->sense_len = ctsio->sense_len; list->scsi_status = ctsio->scsi_status; } mtx_unlock(&lun->lun_lock); ctl_done((union ctl_io *)ctsio); } /* * For any sort of check condition, busy, etc., we just retry. We do not * decrement the retry count for unit attention type errors. These are * normal, and we want to save the retry count for "real" errors. Otherwise, * we could end up with situations where a command will succeed in some * situations and fail in others, depending on whether a unit attention is * pending. Also, some of our error recovery actions, most notably the * LUN reset action, will cause a unit attention. * * We can add more detail here later if necessary. */ static tpc_error_action tpc_checkcond_parse(union ctl_io *io) { tpc_error_action error_action; int error_code, sense_key, asc, ascq; /* * Default to retrying the command. */ error_action = TPC_ERR_RETRY; scsi_extract_sense_len(&io->scsiio.sense_data, io->scsiio.sense_len, &error_code, &sense_key, &asc, &ascq, /*show_errors*/ 1); switch (error_code) { case SSD_DEFERRED_ERROR: case SSD_DESC_DEFERRED_ERROR: error_action |= TPC_ERR_NO_DECREMENT; break; case SSD_CURRENT_ERROR: case SSD_DESC_CURRENT_ERROR: default: switch (sense_key) { case SSD_KEY_UNIT_ATTENTION: error_action |= TPC_ERR_NO_DECREMENT; break; case SSD_KEY_HARDWARE_ERROR: /* * This is our generic "something bad happened" * error code. It often isn't recoverable. */ if ((asc == 0x44) && (ascq == 0x00)) error_action = TPC_ERR_FAIL; break; case SSD_KEY_NOT_READY: /* * If the LUN is powered down, there likely isn't * much point in retrying right now. */ if ((asc == 0x04) && (ascq == 0x02)) error_action = TPC_ERR_FAIL; /* * If the LUN is offline, there probably isn't much * point in retrying, either. */ if ((asc == 0x04) && (ascq == 0x03)) error_action = TPC_ERR_FAIL; break; } } return (error_action); } static tpc_error_action tpc_error_parse(union ctl_io *io) { tpc_error_action error_action = TPC_ERR_RETRY; switch (io->io_hdr.io_type) { case CTL_IO_SCSI: switch (io->io_hdr.status & CTL_STATUS_MASK) { case CTL_SCSI_ERROR: switch (io->scsiio.scsi_status) { case SCSI_STATUS_CHECK_COND: error_action = tpc_checkcond_parse(io); break; default: break; } break; default: break; } break; case CTL_IO_TASK: break; default: panic("%s: invalid ctl_io type %d\n", __func__, io->io_hdr.io_type); break; } return (error_action); } void tpc_done(union ctl_io *io) { struct tpc_io *tio, *tior; /* * Very minimal retry logic. We basically retry if we got an error * back, and the retry count is greater than 0. If we ever want * more sophisticated initiator type behavior, the CAM error * recovery code in ../common might be helpful. */ tio = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; if (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) && (io->io_hdr.retries > 0)) { ctl_io_status old_status; tpc_error_action error_action; error_action = tpc_error_parse(io); switch (error_action & TPC_ERR_MASK) { case TPC_ERR_FAIL: break; case TPC_ERR_RETRY: default: if ((error_action & TPC_ERR_NO_DECREMENT) == 0) io->io_hdr.retries--; old_status = io->io_hdr.status; io->io_hdr.status = CTL_STATUS_NONE; io->io_hdr.flags &= ~CTL_FLAG_ABORT; io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC; if (tpcl_queue(io, tio->lun) != CTL_RETVAL_COMPLETE) { printf("%s: error returned from ctl_queue()!\n", __func__); io->io_hdr.status = old_status; } else return; } } if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) { tio->list->error = 1; if (io->io_hdr.io_type == CTL_IO_SCSI && (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) { tio->list->fwd_scsi_status = io->scsiio.scsi_status; tio->list->fwd_sense_data = io->scsiio.sense_data; tio->list->fwd_sense_len = io->scsiio.sense_len; tio->list->fwd_target = tio->target; tio->list->fwd_cscd = tio->cscd; } } else atomic_add_int(&tio->list->curops, 1); if (!tio->list->error && !tio->list->abort) { while ((tior = TAILQ_FIRST(&tio->run)) != NULL) { TAILQ_REMOVE(&tio->run, tior, rlinks); atomic_add_int(&tio->list->tbdio, 1); if (tpcl_queue(tior->io, tior->lun) != CTL_RETVAL_COMPLETE) panic("tpcl_queue() error"); } } if (atomic_fetchadd_int(&tio->list->tbdio, -1) == 1) tpc_process(tio->list); } int ctl_extended_copy_lid1(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_extended_copy *cdb; struct scsi_extended_copy_lid1_data *data; struct scsi_ec_cscd *cscd; struct scsi_ec_segment *seg; struct tpc_list *list, *tlist; uint8_t *ptr; - char *value; + const char *value; int len, off, lencscd, lenseg, leninl, nseg; CTL_DEBUG_PRINT(("ctl_extended_copy_lid1\n")); cdb = (struct scsi_extended_copy *)ctsio->cdb; len = scsi_4btoul(cdb->length); if (len == 0) { ctl_set_success(ctsio); goto done; } if (len < sizeof(struct scsi_extended_copy_lid1_data) || len > sizeof(struct scsi_extended_copy_lid1_data) + TPC_MAX_LIST + TPC_MAX_INLINE) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 9, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* * If we've got a kernel request that hasn't been malloced yet, * malloc it and tell the caller the data buffer is here. */ if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK); ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } data = (struct scsi_extended_copy_lid1_data *)ctsio->kern_data_ptr; lencscd = scsi_2btoul(data->cscd_list_length); lenseg = scsi_4btoul(data->segment_list_length); leninl = scsi_4btoul(data->inline_data_length); if (lencscd > TPC_MAX_CSCDS * sizeof(struct scsi_ec_cscd)) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x06, SSD_ELEM_NONE); goto done; } if (lenseg > TPC_MAX_SEGS * sizeof(struct scsi_ec_segment)) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x08, SSD_ELEM_NONE); goto done; } if (lencscd + lenseg > TPC_MAX_LIST || leninl > TPC_MAX_INLINE || len < sizeof(struct scsi_extended_copy_lid1_data) + lencscd + lenseg + leninl) { ctl_set_param_len_error(ctsio); goto done; } list = malloc(sizeof(struct tpc_list), M_CTL, M_WAITOK | M_ZERO); list->service_action = cdb->service_action; - value = ctl_get_opt(&lun->be_lun->options, "insecure_tpc"); + value = dnvlist_get_string(lun->be_lun->options, "insecure_tpc", NULL); if (value != NULL && strcmp(value, "on") == 0) list->init_port = -1; else list->init_port = ctsio->io_hdr.nexus.targ_port; list->init_idx = ctl_get_initindex(&ctsio->io_hdr.nexus); list->list_id = data->list_identifier; list->flags = data->flags; list->params = ctsio->kern_data_ptr; list->cscd = (struct scsi_ec_cscd *)&data->data[0]; ptr = &data->data[0]; for (off = 0; off < lencscd; off += sizeof(struct scsi_ec_cscd)) { cscd = (struct scsi_ec_cscd *)(ptr + off); if (cscd->type_code != EC_CSCD_ID) { free(list, M_CTL); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x07, SSD_ELEM_NONE); goto done; } } ptr = &data->data[lencscd]; for (nseg = 0, off = 0; off < lenseg; nseg++) { if (nseg >= TPC_MAX_SEGS) { free(list, M_CTL); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x08, SSD_ELEM_NONE); goto done; } seg = (struct scsi_ec_segment *)(ptr + off); if (seg->type_code != EC_SEG_B2B && seg->type_code != EC_SEG_VERIFY && seg->type_code != EC_SEG_REGISTER_KEY) { free(list, M_CTL); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x09, SSD_ELEM_NONE); goto done; } list->seg[nseg] = seg; off += sizeof(struct scsi_ec_segment) + scsi_2btoul(seg->descr_length); } list->inl = &data->data[lencscd + lenseg]; list->ncscd = lencscd / sizeof(struct scsi_ec_cscd); list->nseg = nseg; list->leninl = leninl; list->ctsio = ctsio; list->lun = lun; mtx_lock(&lun->lun_lock); if ((list->flags & EC_LIST_ID_USAGE_MASK) != EC_LIST_ID_USAGE_NONE) { tlist = tpc_find_list(lun, list->list_id, list->init_idx); if (tlist != NULL && !tlist->completed) { mtx_unlock(&lun->lun_lock); free(list, M_CTL); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } if (tlist != NULL) { TAILQ_REMOVE(&lun->tpc_lists, tlist, links); free(tlist, M_CTL); } } TAILQ_INSERT_TAIL(&lun->tpc_lists, list, links); mtx_unlock(&lun->lun_lock); tpc_process(list); return (CTL_RETVAL_COMPLETE); done: if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) { free(ctsio->kern_data_ptr, M_CTL); ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED; } ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_extended_copy_lid4(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_extended_copy *cdb; struct scsi_extended_copy_lid4_data *data; struct scsi_ec_cscd *cscd; struct scsi_ec_segment *seg; struct tpc_list *list, *tlist; uint8_t *ptr; - char *value; + const char *value; int len, off, lencscd, lenseg, leninl, nseg; CTL_DEBUG_PRINT(("ctl_extended_copy_lid4\n")); cdb = (struct scsi_extended_copy *)ctsio->cdb; len = scsi_4btoul(cdb->length); if (len == 0) { ctl_set_success(ctsio); goto done; } if (len < sizeof(struct scsi_extended_copy_lid4_data) || len > sizeof(struct scsi_extended_copy_lid4_data) + TPC_MAX_LIST + TPC_MAX_INLINE) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 9, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* * If we've got a kernel request that hasn't been malloced yet, * malloc it and tell the caller the data buffer is here. */ if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK); ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } data = (struct scsi_extended_copy_lid4_data *)ctsio->kern_data_ptr; lencscd = scsi_2btoul(data->cscd_list_length); lenseg = scsi_2btoul(data->segment_list_length); leninl = scsi_2btoul(data->inline_data_length); if (lencscd > TPC_MAX_CSCDS * sizeof(struct scsi_ec_cscd)) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x06, SSD_ELEM_NONE); goto done; } if (lenseg > TPC_MAX_SEGS * sizeof(struct scsi_ec_segment)) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x08, SSD_ELEM_NONE); goto done; } if (lencscd + lenseg > TPC_MAX_LIST || leninl > TPC_MAX_INLINE || len < sizeof(struct scsi_extended_copy_lid1_data) + lencscd + lenseg + leninl) { ctl_set_param_len_error(ctsio); goto done; } list = malloc(sizeof(struct tpc_list), M_CTL, M_WAITOK | M_ZERO); list->service_action = cdb->service_action; - value = ctl_get_opt(&lun->be_lun->options, "insecure_tpc"); + value = dnvlist_get_string(lun->be_lun->options, "insecure_tpc", NULL); if (value != NULL && strcmp(value, "on") == 0) list->init_port = -1; else list->init_port = ctsio->io_hdr.nexus.targ_port; list->init_idx = ctl_get_initindex(&ctsio->io_hdr.nexus); list->list_id = scsi_4btoul(data->list_identifier); list->flags = data->flags; list->params = ctsio->kern_data_ptr; list->cscd = (struct scsi_ec_cscd *)&data->data[0]; ptr = &data->data[0]; for (off = 0; off < lencscd; off += sizeof(struct scsi_ec_cscd)) { cscd = (struct scsi_ec_cscd *)(ptr + off); if (cscd->type_code != EC_CSCD_ID) { free(list, M_CTL); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x07, SSD_ELEM_NONE); goto done; } } ptr = &data->data[lencscd]; for (nseg = 0, off = 0; off < lenseg; nseg++) { if (nseg >= TPC_MAX_SEGS) { free(list, M_CTL); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x08, SSD_ELEM_NONE); goto done; } seg = (struct scsi_ec_segment *)(ptr + off); if (seg->type_code != EC_SEG_B2B && seg->type_code != EC_SEG_VERIFY && seg->type_code != EC_SEG_REGISTER_KEY) { free(list, M_CTL); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x09, SSD_ELEM_NONE); goto done; } list->seg[nseg] = seg; off += sizeof(struct scsi_ec_segment) + scsi_2btoul(seg->descr_length); } list->inl = &data->data[lencscd + lenseg]; list->ncscd = lencscd / sizeof(struct scsi_ec_cscd); list->nseg = nseg; list->leninl = leninl; list->ctsio = ctsio; list->lun = lun; mtx_lock(&lun->lun_lock); if ((list->flags & EC_LIST_ID_USAGE_MASK) != EC_LIST_ID_USAGE_NONE) { tlist = tpc_find_list(lun, list->list_id, list->init_idx); if (tlist != NULL && !tlist->completed) { mtx_unlock(&lun->lun_lock); free(list, M_CTL); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } if (tlist != NULL) { TAILQ_REMOVE(&lun->tpc_lists, tlist, links); free(tlist, M_CTL); } } TAILQ_INSERT_TAIL(&lun->tpc_lists, list, links); mtx_unlock(&lun->lun_lock); tpc_process(list); return (CTL_RETVAL_COMPLETE); done: if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) { free(ctsio->kern_data_ptr, M_CTL); ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED; } ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } static void tpc_create_token(struct ctl_lun *lun, struct ctl_port *port, off_t len, struct scsi_token *token) { static int id = 0; struct scsi_vpd_id_descriptor *idd = NULL; struct scsi_ec_cscd_id *cscd; struct scsi_read_capacity_data_long *dtsd; int targid_len; scsi_ulto4b(ROD_TYPE_AUR, token->type); scsi_ulto2b(0x01f8, token->length); scsi_u64to8b(atomic_fetchadd_int(&id, 1), &token->body[0]); if (lun->lun_devid) idd = scsi_get_devid_desc((struct scsi_vpd_id_descriptor *) lun->lun_devid->data, lun->lun_devid->len, scsi_devid_is_lun_naa); if (idd == NULL && lun->lun_devid) idd = scsi_get_devid_desc((struct scsi_vpd_id_descriptor *) lun->lun_devid->data, lun->lun_devid->len, scsi_devid_is_lun_eui64); if (idd != NULL) { cscd = (struct scsi_ec_cscd_id *)&token->body[8]; cscd->type_code = EC_CSCD_ID; cscd->luidt_pdt = T_DIRECT; memcpy(&cscd->codeset, idd, 4 + idd->length); scsi_ulto3b(lun->be_lun->blocksize, cscd->dtsp.block_length); } scsi_u64to8b(0, &token->body[40]); /* XXX: Should be 128bit value. */ scsi_u64to8b(len, &token->body[48]); /* ROD token device type specific data (RC16 without first field) */ dtsd = (struct scsi_read_capacity_data_long *)&token->body[88 - 8]; scsi_ulto4b(lun->be_lun->blocksize, dtsd->length); dtsd->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE; scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, dtsd->lalba_lbp); if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) dtsd->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ; if (port->target_devid) { targid_len = port->target_devid->len; memcpy(&token->body[120], port->target_devid->data, targid_len); } else targid_len = 32; arc4rand(&token->body[120 + targid_len], 384 - targid_len, 0); }; int ctl_populate_token(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_port *port = CTL_PORT(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_populate_token *cdb; struct scsi_populate_token_data *data; struct tpc_list *list, *tlist; struct tpc_token *token; uint64_t lba; int len, lendata, lendesc; CTL_DEBUG_PRINT(("ctl_populate_token\n")); cdb = (struct scsi_populate_token *)ctsio->cdb; len = scsi_4btoul(cdb->length); if (len < sizeof(struct scsi_populate_token_data) || len > sizeof(struct scsi_populate_token_data) + TPC_MAX_SEGS * sizeof(struct scsi_range_desc)) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 9, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* * If we've got a kernel request that hasn't been malloced yet, * malloc it and tell the caller the data buffer is here. */ if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK); ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } data = (struct scsi_populate_token_data *)ctsio->kern_data_ptr; lendata = scsi_2btoul(data->length); if (lendata < sizeof(struct scsi_populate_token_data) - 2 + sizeof(struct scsi_range_desc)) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } lendesc = scsi_2btoul(data->range_descriptor_length); if (lendesc < sizeof(struct scsi_range_desc) || len < sizeof(struct scsi_populate_token_data) + lendesc || lendata < sizeof(struct scsi_populate_token_data) - 2 + lendesc) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 14, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* printf("PT(list=%u) flags=%x to=%d rt=%x len=%x\n", scsi_4btoul(cdb->list_identifier), data->flags, scsi_4btoul(data->inactivity_timeout), scsi_4btoul(data->rod_type), scsi_2btoul(data->range_descriptor_length)); */ /* Validate INACTIVITY TIMEOUT field */ if (scsi_4btoul(data->inactivity_timeout) > TPC_MAX_TOKEN_TIMEOUT) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 4, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* Validate ROD TYPE field */ if ((data->flags & EC_PT_RTV) && scsi_4btoul(data->rod_type) != ROD_TYPE_AUR) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 8, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* Validate list of ranges */ if (tpc_check_ranges_l(&data->desc[0], scsi_2btoul(data->range_descriptor_length) / sizeof(struct scsi_range_desc), lun->be_lun->maxlba, &lba) != 0) { ctl_set_lba_out_of_range(ctsio, lba); goto done; } if (tpc_check_ranges_x(&data->desc[0], scsi_2btoul(data->range_descriptor_length) / sizeof(struct scsi_range_desc)) != 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } list = malloc(sizeof(struct tpc_list), M_CTL, M_WAITOK | M_ZERO); list->service_action = cdb->service_action; list->init_port = ctsio->io_hdr.nexus.targ_port; list->init_idx = ctl_get_initindex(&ctsio->io_hdr.nexus); list->list_id = scsi_4btoul(cdb->list_identifier); list->flags = data->flags; list->ctsio = ctsio; list->lun = lun; mtx_lock(&lun->lun_lock); tlist = tpc_find_list(lun, list->list_id, list->init_idx); if (tlist != NULL && !tlist->completed) { mtx_unlock(&lun->lun_lock); free(list, M_CTL); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } if (tlist != NULL) { TAILQ_REMOVE(&lun->tpc_lists, tlist, links); free(tlist, M_CTL); } TAILQ_INSERT_TAIL(&lun->tpc_lists, list, links); mtx_unlock(&lun->lun_lock); token = malloc(sizeof(*token), M_CTL, M_WAITOK | M_ZERO); token->lun = lun->lun; token->blocksize = lun->be_lun->blocksize; token->params = ctsio->kern_data_ptr; token->range = &data->desc[0]; token->nrange = scsi_2btoul(data->range_descriptor_length) / sizeof(struct scsi_range_desc); list->cursectors = tpc_ranges_length(token->range, token->nrange); list->curbytes = (off_t)list->cursectors * lun->be_lun->blocksize; tpc_create_token(lun, port, list->curbytes, (struct scsi_token *)token->token); token->active = 0; token->last_active = time_uptime; token->timeout = scsi_4btoul(data->inactivity_timeout); if (token->timeout == 0) token->timeout = TPC_DFL_TOKEN_TIMEOUT; else if (token->timeout < TPC_MIN_TOKEN_TIMEOUT) token->timeout = TPC_MIN_TOKEN_TIMEOUT; memcpy(list->res_token, token->token, sizeof(list->res_token)); list->res_token_valid = 1; list->curseg = 0; list->completed = 1; list->last_active = time_uptime; mtx_lock(&softc->tpc_lock); TAILQ_INSERT_TAIL(&softc->tpc_tokens, token, links); mtx_unlock(&softc->tpc_lock); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); done: if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) { free(ctsio->kern_data_ptr, M_CTL); ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED; } ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_write_using_token(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_write_using_token *cdb; struct scsi_write_using_token_data *data; struct tpc_list *list, *tlist; struct tpc_token *token; uint64_t lba; int len, lendata, lendesc; CTL_DEBUG_PRINT(("ctl_write_using_token\n")); cdb = (struct scsi_write_using_token *)ctsio->cdb; len = scsi_4btoul(cdb->length); if (len < sizeof(struct scsi_write_using_token_data) || len > sizeof(struct scsi_write_using_token_data) + TPC_MAX_SEGS * sizeof(struct scsi_range_desc)) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 9, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* * If we've got a kernel request that hasn't been malloced yet, * malloc it and tell the caller the data buffer is here. */ if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) { ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK); ctsio->kern_data_len = len; ctsio->kern_total_len = len; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } data = (struct scsi_write_using_token_data *)ctsio->kern_data_ptr; lendata = scsi_2btoul(data->length); if (lendata < sizeof(struct scsi_write_using_token_data) - 2 + sizeof(struct scsi_range_desc)) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } lendesc = scsi_2btoul(data->range_descriptor_length); if (lendesc < sizeof(struct scsi_range_desc) || len < sizeof(struct scsi_write_using_token_data) + lendesc || lendata < sizeof(struct scsi_write_using_token_data) - 2 + lendesc) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 534, /*bit_valid*/ 0, /*bit*/ 0); goto done; } /* printf("WUT(list=%u) flags=%x off=%ju len=%x\n", scsi_4btoul(cdb->list_identifier), data->flags, scsi_8btou64(data->offset_into_rod), scsi_2btoul(data->range_descriptor_length)); */ /* Validate list of ranges */ if (tpc_check_ranges_l(&data->desc[0], scsi_2btoul(data->range_descriptor_length) / sizeof(struct scsi_range_desc), lun->be_lun->maxlba, &lba) != 0) { ctl_set_lba_out_of_range(ctsio, lba); goto done; } if (tpc_check_ranges_x(&data->desc[0], scsi_2btoul(data->range_descriptor_length) / sizeof(struct scsi_range_desc)) != 0) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 0, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } list = malloc(sizeof(struct tpc_list), M_CTL, M_WAITOK | M_ZERO); list->service_action = cdb->service_action; list->init_port = ctsio->io_hdr.nexus.targ_port; list->init_idx = ctl_get_initindex(&ctsio->io_hdr.nexus); list->list_id = scsi_4btoul(cdb->list_identifier); list->flags = data->flags; list->params = ctsio->kern_data_ptr; list->range = &data->desc[0]; list->nrange = scsi_2btoul(data->range_descriptor_length) / sizeof(struct scsi_range_desc); list->offset_into_rod = scsi_8btou64(data->offset_into_rod); list->ctsio = ctsio; list->lun = lun; mtx_lock(&lun->lun_lock); tlist = tpc_find_list(lun, list->list_id, list->init_idx); if (tlist != NULL && !tlist->completed) { mtx_unlock(&lun->lun_lock); free(list, M_CTL); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 0, /*bit_valid*/ 0, /*bit*/ 0); goto done; } if (tlist != NULL) { TAILQ_REMOVE(&lun->tpc_lists, tlist, links); free(tlist, M_CTL); } TAILQ_INSERT_TAIL(&lun->tpc_lists, list, links); mtx_unlock(&lun->lun_lock); /* Block device zero ROD token -> no token. */ if (scsi_4btoul(data->rod_token) == ROD_TYPE_BLOCK_ZERO) { tpc_process(list); return (CTL_RETVAL_COMPLETE); } mtx_lock(&softc->tpc_lock); TAILQ_FOREACH(token, &softc->tpc_tokens, links) { if (memcmp(token->token, data->rod_token, sizeof(data->rod_token)) == 0) break; } if (token != NULL) { token->active++; list->token = token; if (data->flags & EC_WUT_DEL_TKN) token->timeout = 0; } mtx_unlock(&softc->tpc_lock); if (token == NULL) { mtx_lock(&lun->lun_lock); TAILQ_REMOVE(&lun->tpc_lists, list, links); mtx_unlock(&lun->lun_lock); free(list, M_CTL); ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x23, /*ascq*/ 0x04, SSD_ELEM_NONE); goto done; } tpc_process(list); return (CTL_RETVAL_COMPLETE); done: if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) { free(ctsio->kern_data_ptr, M_CTL); ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED; } ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_receive_rod_token_information(struct ctl_scsiio *ctsio) { struct ctl_lun *lun = CTL_LUN(ctsio); struct scsi_receive_rod_token_information *cdb; struct scsi_receive_copy_status_lid4_data *data; struct tpc_list *list; struct tpc_list list_copy; uint8_t *ptr; int retval; int alloc_len, total_len, token_len; uint32_t list_id; CTL_DEBUG_PRINT(("ctl_receive_rod_token_information\n")); cdb = (struct scsi_receive_rod_token_information *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; list_id = scsi_4btoul(cdb->list_identifier); mtx_lock(&lun->lun_lock); list = tpc_find_list(lun, list_id, ctl_get_initindex(&ctsio->io_hdr.nexus)); if (list == NULL) { mtx_unlock(&lun->lun_lock); ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); ctl_done((union ctl_io *)ctsio); return (retval); } list_copy = *list; if (list->completed) { TAILQ_REMOVE(&lun->tpc_lists, list, links); free(list, M_CTL); } mtx_unlock(&lun->lun_lock); token_len = list_copy.res_token_valid ? 2 + sizeof(list_copy.res_token) : 0; total_len = sizeof(*data) + list_copy.sense_len + 4 + token_len; alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_receive_copy_status_lid4_data *)ctsio->kern_data_ptr; scsi_ulto4b(sizeof(*data) - 4 + list_copy.sense_len + 4 + token_len, data->available_data); data->response_to_service_action = list_copy.service_action; if (list_copy.completed) { if (list_copy.error) data->copy_command_status = RCS_CCS_ERROR; else if (list_copy.abort) data->copy_command_status = RCS_CCS_ABORTED; else data->copy_command_status = RCS_CCS_COMPLETED; } else data->copy_command_status = RCS_CCS_INPROG_FG; scsi_ulto2b(list_copy.curops, data->operation_counter); scsi_ulto4b(UINT32_MAX, data->estimated_status_update_delay); data->transfer_count_units = RCS_TC_LBAS; scsi_u64to8b(list_copy.cursectors, data->transfer_count); scsi_ulto2b(list_copy.curseg, data->segments_processed); data->length_of_the_sense_data_field = list_copy.sense_len; data->sense_data_length = list_copy.sense_len; memcpy(data->sense_data, &list_copy.sense_data, list_copy.sense_len); ptr = &data->sense_data[data->length_of_the_sense_data_field]; scsi_ulto4b(token_len, &ptr[0]); if (list_copy.res_token_valid) { scsi_ulto2b(0, &ptr[4]); memcpy(&ptr[6], list_copy.res_token, sizeof(list_copy.res_token)); } /* printf("RRTI(list=%u) valid=%d\n", scsi_4btoul(cdb->list_identifier), list_copy.res_token_valid); */ ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } int ctl_report_all_rod_tokens(struct ctl_scsiio *ctsio) { struct ctl_softc *softc = CTL_SOFTC(ctsio); struct scsi_report_all_rod_tokens *cdb; struct scsi_report_all_rod_tokens_data *data; struct tpc_token *token; int retval; int alloc_len, total_len, tokens, i; CTL_DEBUG_PRINT(("ctl_receive_rod_token_information\n")); cdb = (struct scsi_report_all_rod_tokens *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; tokens = 0; mtx_lock(&softc->tpc_lock); TAILQ_FOREACH(token, &softc->tpc_tokens, links) tokens++; mtx_unlock(&softc->tpc_lock); if (tokens > 512) tokens = 512; total_len = sizeof(*data) + tokens * 96; alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; ctsio->kern_rel_offset = 0; ctsio->kern_data_len = min(total_len, alloc_len); ctsio->kern_total_len = ctsio->kern_data_len; data = (struct scsi_report_all_rod_tokens_data *)ctsio->kern_data_ptr; i = 0; mtx_lock(&softc->tpc_lock); TAILQ_FOREACH(token, &softc->tpc_tokens, links) { if (i >= tokens) break; memcpy(&data->rod_management_token_list[i * 96], token->token, 96); i++; } mtx_unlock(&softc->tpc_lock); scsi_ulto4b(sizeof(*data) - 4 + i * 96, data->available_data); /* printf("RART tokens=%d\n", i); */ ctl_set_success(ctsio); ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED; ctsio->be_move_done = ctl_config_move_done; ctl_datamove((union ctl_io *)ctsio); return (retval); } Index: head/sys/sys/param.h =================================================================== --- head/sys/sys/param.h (revision 333445) +++ head/sys/sys/param.h (revision 333446) @@ -1,365 +1,365 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)param.h 8.3 (Berkeley) 4/4/95 * $FreeBSD$ */ #ifndef _SYS_PARAM_H_ #define _SYS_PARAM_H_ #include #define BSD 199506 /* System version (year & month). */ #define BSD4_3 1 #define BSD4_4 1 /* * __FreeBSD_version numbers are documented in the Porter's Handbook. * If you bump the version for any reason, you should update the documentation * there. * Currently this lives here in the doc/ repository: * * head/en_US.ISO8859-1/books/porters-handbook/versions/chapter.xml * * scheme is: Rxx * 'R' is in the range 0 to 4 if this is a release branch or * X.0-CURRENT before releng/X.0 is created, otherwise 'R' is * in the range 5 to 9. */ #undef __FreeBSD_version -#define __FreeBSD_version 1200062 /* Master, propagated to newvers */ +#define __FreeBSD_version 1200063 /* Master, propagated to newvers */ /* * __FreeBSD_kernel__ indicates that this system uses the kernel of FreeBSD, * which by definition is always true on FreeBSD. This macro is also defined * on other systems that use the kernel of FreeBSD, such as GNU/kFreeBSD. * * It is tempting to use this macro in userland code when we want to enable * kernel-specific routines, and in fact it's fine to do this in code that * is part of FreeBSD itself. However, be aware that as presence of this * macro is still not widespread (e.g. older FreeBSD versions, 3rd party * compilers, etc), it is STRONGLY DISCOURAGED to check for this macro in * external applications without also checking for __FreeBSD__ as an * alternative. */ #undef __FreeBSD_kernel__ #define __FreeBSD_kernel__ #if defined(_KERNEL) || defined(IN_RTLD) #define P_OSREL_SIGWAIT 700000 #define P_OSREL_SIGSEGV 700004 #define P_OSREL_MAP_ANON 800104 #define P_OSREL_MAP_FSTRICT 1100036 #define P_OSREL_SHUTDOWN_ENOTCONN 1100077 #define P_OSREL_MAP_GUARD 1200035 #define P_OSREL_WRFSBASE 1200041 #define P_OSREL_CK_CYLGRP 1200046 #define P_OSREL_VMTOTAL64 1200054 #define P_OSREL_MAJOR(x) ((x) / 100000) #endif #ifndef LOCORE #include #endif /* * Machine-independent constants (some used in following include files). * Redefined constants are from POSIX 1003.1 limits file. * * MAXCOMLEN should be >= sizeof(ac_comm) (see ) */ #include #define MAXCOMLEN 19 /* max command name remembered */ #define MAXINTERP PATH_MAX /* max interpreter file name length */ #define MAXLOGNAME 33 /* max login name length (incl. NUL) */ #define MAXUPRC CHILD_MAX /* max simultaneous processes */ #define NCARGS ARG_MAX /* max bytes for an exec function */ #define NGROUPS (NGROUPS_MAX+1) /* max number groups */ #define NOFILE OPEN_MAX /* max open files per process */ #define NOGROUP 65535 /* marker for empty group set member */ #define MAXHOSTNAMELEN 256 /* max hostname size */ #define SPECNAMELEN 63 /* max length of devicename */ /* More types and definitions used throughout the kernel. */ #ifdef _KERNEL #include #include #ifndef LOCORE #include #include #endif #ifndef FALSE #define FALSE 0 #endif #ifndef TRUE #define TRUE 1 #endif #endif #ifndef _KERNEL /* Signals. */ #include #endif /* Machine type dependent parameters. */ #include #ifndef _KERNEL #include #endif #ifndef DEV_BSHIFT #define DEV_BSHIFT 9 /* log2(DEV_BSIZE) */ #endif #define DEV_BSIZE (1<>PAGE_SHIFT) #endif /* * btodb() is messy and perhaps slow because `bytes' may be an off_t. We * want to shift an unsigned type to avoid sign extension and we don't * want to widen `bytes' unnecessarily. Assume that the result fits in * a daddr_t. */ #ifndef btodb #define btodb(bytes) /* calculates (bytes / DEV_BSIZE) */ \ (sizeof (bytes) > sizeof(long) \ ? (daddr_t)((unsigned long long)(bytes) >> DEV_BSHIFT) \ : (daddr_t)((unsigned long)(bytes) >> DEV_BSHIFT)) #endif #ifndef dbtob #define dbtob(db) /* calculates (db * DEV_BSIZE) */ \ ((off_t)(db) << DEV_BSHIFT) #endif #define PRIMASK 0x0ff #define PCATCH 0x100 /* OR'd with pri for tsleep to check signals */ #define PDROP 0x200 /* OR'd with pri to stop re-entry of interlock mutex */ #define NZERO 0 /* default "nice" */ #define NBBY 8 /* number of bits in a byte */ #define NBPW sizeof(int) /* number of bytes per word (integer) */ #define CMASK 022 /* default file mask: S_IWGRP|S_IWOTH */ #define NODEV (dev_t)(-1) /* non-existent device */ /* * File system parameters and macros. * * MAXBSIZE - Filesystems are made out of blocks of at most MAXBSIZE bytes * per block. MAXBSIZE may be made larger without effecting * any existing filesystems as long as it does not exceed MAXPHYS, * and may be made smaller at the risk of not being able to use * filesystems which require a block size exceeding MAXBSIZE. * * MAXBCACHEBUF - Maximum size of a buffer in the buffer cache. This must * be >= MAXBSIZE and can be set differently for different * architectures by defining it in . * Making this larger allows NFS to do larger reads/writes. * * BKVASIZE - Nominal buffer space per buffer, in bytes. BKVASIZE is the * minimum KVM memory reservation the kernel is willing to make. * Filesystems can of course request smaller chunks. Actual * backing memory uses a chunk size of a page (PAGE_SIZE). * The default value here can be overridden on a per-architecture * basis by defining it in . * * If you make BKVASIZE too small you risk seriously fragmenting * the buffer KVM map which may slow things down a bit. If you * make it too big the kernel will not be able to optimally use * the KVM memory reserved for the buffer cache and will wind * up with too-few buffers. * * The default is 16384, roughly 2x the block size used by a * normal UFS filesystem. */ #define MAXBSIZE 65536 /* must be power of 2 */ #ifndef MAXBCACHEBUF #define MAXBCACHEBUF MAXBSIZE /* must be a power of 2 >= MAXBSIZE */ #endif #ifndef BKVASIZE #define BKVASIZE 16384 /* must be power of 2 */ #endif #define BKVAMASK (BKVASIZE-1) /* * MAXPATHLEN defines the longest permissible path length after expanding * symbolic links. It is used to allocate a temporary buffer from the buffer * pool in which to do the name expansion, hence should be a power of two, * and must be less than or equal to MAXBSIZE. MAXSYMLINKS defines the * maximum number of symbolic links that may be expanded in a path name. * It should be set high enough to allow all legitimate uses, but halt * infinite loops reasonably quickly. */ #define MAXPATHLEN PATH_MAX #define MAXSYMLINKS 32 /* Bit map related macros. */ #define setbit(a,i) (((unsigned char *)(a))[(i)/NBBY] |= 1<<((i)%NBBY)) #define clrbit(a,i) (((unsigned char *)(a))[(i)/NBBY] &= ~(1<<((i)%NBBY))) #define isset(a,i) \ (((const unsigned char *)(a))[(i)/NBBY] & (1<<((i)%NBBY))) #define isclr(a,i) \ ((((const unsigned char *)(a))[(i)/NBBY] & (1<<((i)%NBBY))) == 0) /* Macros for counting and rounding. */ #ifndef howmany #define howmany(x, y) (((x)+((y)-1))/(y)) #endif #define nitems(x) (sizeof((x)) / sizeof((x)[0])) #define rounddown(x, y) (((x)/(y))*(y)) #define rounddown2(x, y) ((x)&(~((y)-1))) /* if y is power of two */ #define roundup(x, y) ((((x)+((y)-1))/(y))*(y)) /* to any y */ #define roundup2(x, y) (((x)+((y)-1))&(~((y)-1))) /* if y is powers of two */ #define powerof2(x) ((((x)-1)&(x))==0) /* Macros for min/max. */ #define MIN(a,b) (((a)<(b))?(a):(b)) #define MAX(a,b) (((a)>(b))?(a):(b)) #ifdef _KERNEL /* * Basic byte order function prototypes for non-inline functions. */ #ifndef LOCORE #ifndef _BYTEORDER_PROTOTYPED #define _BYTEORDER_PROTOTYPED __BEGIN_DECLS __uint32_t htonl(__uint32_t); __uint16_t htons(__uint16_t); __uint32_t ntohl(__uint32_t); __uint16_t ntohs(__uint16_t); __END_DECLS #endif #endif #ifndef _BYTEORDER_FUNC_DEFINED #define _BYTEORDER_FUNC_DEFINED #define htonl(x) __htonl(x) #define htons(x) __htons(x) #define ntohl(x) __ntohl(x) #define ntohs(x) __ntohs(x) #endif /* !_BYTEORDER_FUNC_DEFINED */ #endif /* _KERNEL */ /* * Scale factor for scaled integers used to count %cpu time and load avgs. * * The number of CPU `tick's that map to a unique `%age' can be expressed * by the formula (1 / (2 ^ (FSHIFT - 11))). The maximum load average that * can be calculated (assuming 32 bits) can be closely approximated using * the formula (2 ^ (2 * (16 - FSHIFT))) for (FSHIFT < 15). * * For the scheduler to maintain a 1:1 mapping of CPU `tick' to `%age', * FSHIFT must be at least 11; this gives us a maximum load avg of ~1024. */ #define FSHIFT 11 /* bits to right of fixed binary point */ #define FSCALE (1<> (PAGE_SHIFT - DEV_BSHIFT)) #define ctodb(db) /* calculates pages to devblks */ \ ((db) << (PAGE_SHIFT - DEV_BSHIFT)) /* * Old spelling of __containerof(). */ #define member2struct(s, m, x) \ ((struct s *)(void *)((char *)(x) - offsetof(struct s, m))) /* * Access a variable length array that has been declared as a fixed * length array. */ #define __PAST_END(array, offset) (((__typeof__(*(array)) *)(array))[offset]) #endif /* _SYS_PARAM_H_ */ Index: head/usr.sbin/ctladm/Makefile =================================================================== --- head/usr.sbin/ctladm/Makefile (revision 333445) +++ head/usr.sbin/ctladm/Makefile (revision 333446) @@ -1,26 +1,26 @@ # $FreeBSD$ .include PROG= ctladm SRCS= ctladm.c util.c ctl_util.c ctl_scsi_all.c .PATH: ${SRCTOP}/sys/cam/ctl SDIR= ${SRCTOP}/sys CFLAGS+= -I${SDIR} # This is necessary because of these warnings: # warning: cast increases required alignment of target type # The solution is to either upgrade the compiler (preferred), or do void # pointer gymnastics to get around the warning. For now, disable the # warning instead of doing the void pointer workaround. .if ${MACHINE_CPUARCH} == "arm" WARNS?= 3 .endif -LIBADD= cam sbuf bsdxml util +LIBADD= cam sbuf bsdxml util nv MAN= ctladm.8 .if ${MK_ISCSI} != "no" CFLAGS+= -DWANT_ISCSI .endif .include Index: head/usr.sbin/ctladm/ctladm.8 =================================================================== --- head/usr.sbin/ctladm/ctladm.8 (revision 333445) +++ head/usr.sbin/ctladm/ctladm.8 (revision 333446) @@ -1,1057 +1,1081 @@ .\" .\" Copyright (c) 2003 Silicon Graphics International Corp. .\" Copyright (c) 2015 Alexander Motin +.\" Copyright (c) 2018 Marcelo Araujo .\" All rights reserved. .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice, this list of conditions, and the following disclaimer, .\" without modification. .\" 2. Redistributions in binary form must reproduce at minimum a disclaimer .\" substantially similar to the "NO WARRANTY" disclaimer below .\" ("Disclaimer") and any redistribution must be conditioned upon .\" including a substantially similar Disclaimer requirement for further .\" binary redistribution. .\" .\" NO WARRANTY .\" THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS .\" "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT .\" LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR .\" A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT .\" HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, .\" STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING .\" IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE .\" POSSIBILITY OF SUCH DAMAGES. .\" .\" ctladm utility man page. .\" .\" Author: Ken Merry .\" .\" $Id: //depot/users/kenm/FreeBSD-test2/usr.sbin/ctladm/ctladm.8#3 $ .\" $FreeBSD$ .\" -.Dd July 15, 2017 +.Dd May 10, 2018 .Dt CTLADM 8 .Os .Sh NAME .Nm ctladm .Nd CAM Target Layer control utility .Sh SYNOPSIS .Nm .Aq Ar command .Op lun .Op generic args .Op command args .Nm .Ic tur .Aq lun .Op general options .Nm .Ic inquiry .Aq lun .Op general options .Nm .Ic reqsense .Aq lun .Op general options .Nm .Ic reportluns .Aq lun .Op general options .Nm .Ic read .Aq lun .Op general options .Aq Fl l Ar lba .Aq Fl d Ar datalen .Aq Fl f Ar file|- .Aq Fl b Ar blocksize_bytes .Op Fl c Ar cdbsize .Op Fl N .Nm .Ic write .Aq lun .Op general options .Aq Fl l Ar lba .Aq Fl d Ar datalen .Aq Fl f Ar file|- .Aq Fl b Ar blocksize_bytes .Op Fl c Ar cdbsize .Op Fl N .Nm .Ic readcap .Aq lun .Op general options .Op Fl c Ar cdbsize .Nm .Ic modesense .Aq lun .Aq Fl m Ar page | Fl l .Op Fl P Ar pc .Op Fl d .Op Fl S Ar subpage .Op Fl c Ar size .Nm .Ic start .Aq lun .Op general options .Op Fl i .Op Fl o .Nm .Ic stop .Aq lun .Op general options .Op Fl i .Op Fl o .Nm .Ic synccache .Aq lun .Op general options .Op Fl l Ar lba .Op Fl b Ar blockcount .Op Fl r .Op Fl i .Op Fl c Ar cdbsize .Nm .Ic lunlist .Nm .Ic delay .Aq lun .Aq Fl l Ar datamove|done .Aq Fl t Ar secs .Op Fl T Ar oneshot|cont .Nm .Ic inject .Aq Fl i Ar action .Aq Fl p Ar pattern .Op Fl r Ar lba,len .Op Fl s Ar len fmt Op Ar args .Op Fl c .Op Fl d Ar delete_id .Nm .Ic create .Aq Fl b Ar backend .Op Fl B Ar blocksize .Op Fl d Ar device_id .Op Fl l Ar lun_id .Op Fl o Ar name=value .Op Fl s Ar size_bytes .Op Fl S Ar serial_num .Op Fl t Ar device_type .Nm .Ic remove .Aq Fl b Ar backend .Aq Fl l Ar lun_id .Op Fl o Ar name=value .Nm .Ic modify .Aq Fl b Ar backend .Aq Fl l Ar lun_id .Op Fl o Ar name=value .Aq Fl s Ar size_bytes .Nm .Ic devlist .Op Fl b Ar backend .Op Fl v .Op Fl x .Nm .Ic port +.Op Fl c .Op Fl o Ar on|off .Op Fl w Ar wwpn .Op Fl W Ar wwnn +.Op Fl O Ar pp|vp .Op Fl p Ar targ_port +.Op Fl r Ar targ_port .Op Fl t Ar fe_type .Nm .Ic portlist .Op Fl f Ar frontend .Op Fl i .Op Fl l .Op Fl p Ar targ_port .Op Fl q .Op Fl v .Op Fl x .Nm .Ic lunmap .Aq Fl p Ar targ_port .Op Fl l Ar pLUN .Op Fl L Ar cLUN .Nm .Ic dumpooa .Nm .Ic dumpstructs .Nm .Ic islist .Op Fl v .Op Fl x .Nm .Ic islogout .Aq Fl a | Fl c Ar connection-id | Fl i Ar name | Fl p Ar portal .Nm .Ic isterminate .Aq Fl a | Fl c Ar connection-id | Fl i Ar name | Fl p Ar portal .Nm .Ic help .Sh DESCRIPTION The .Nm utility is designed to provide a way to access and control the CAM Target Layer (CTL). It provides a way to send .Tn SCSI commands to the CTL layer, and also provides some meta-commands that utilize .Tn SCSI commands. (For instance, the .Ic lunlist command is implemented using the .Tn SCSI REPORT LUNS and INQUIRY commands.) .Pp The .Nm utility has a number of primary functions, many of which require a device identifier. The device identifier takes the following form: .Bl -tag -width 14n .It lun Specify the LUN number to operate on. .El Many of the primary functions of the .Nm utility take the following optional arguments: .Bl -tag -width 10n .It Fl C Ar retries Specify the number of times to retry a command in the event of failure. .It Fl D Ar device Specify the device to open. This allows opening a device other than the default device, .Pa /dev/cam/ctl , to be opened for sending commands. .It Fl I Ar id Specify the initiator number to use. By default, .Nm will use 7 as the initiator number. .El .Pp Primary commands: .Bl -tag -width 11n .It Ic tur Send the .Tn SCSI TEST UNIT READY command to the device and report whether or not it is ready. .It Ic inquiry Send the .Tn SCSI INQUIRY command to the device and display some of the returned inquiry data. .It Ic reqsense Send the .Tn SCSI REQUEST SENSE command to the device and display the returned sense information. .It Ic reportluns Send the .Tn SCSI REPORT LUNS command to the device and display supported LUNs. .It Ic read Send a .Tn SCSI READ command to the device, and write the requested data to a file or stdout. .Bl -tag -width 12n .It Fl l Ar lba Specify the starting Logical Block Address for the READ. This can be specified in decimal, octal (starting with 0), hexadecimal (starting with 0x) or any other base supported by .Xr strtoull 3 . .It Fl d Ar datalen Specify the length, in 512 byte blocks, of the READ request. .It Fl f Ar file Specify the destination for the data read by the READ command. Either a filename or .Sq - for stdout may be specified. .It Fl c Ar cdbsize Specify the minimum .Tn SCSI CDB (Command Data Block) size to be used for the READ request. Allowable values are 6, 10, 12 and 16. Depending upon the LBA and amount of data requested, a larger CDB size may be used to satisfy the request. (e.g., for LBAs above 0xffffffff, READ(16) must be used to satisfy the request.) .It Fl b Ar blocksize Specify the blocksize of the underlying .Tn SCSI device, so the transfer length can be calculated accurately. The blocksize can be obtained via the .Tn SCSI READ CAPACITY command. .It Fl N Do not copy data to .Nm from the kernel when doing a read, just execute the command without copying data. This is to be used for performance testing. .El .It Ic write Read data from a file or stdin, and write the data to the device using the .Tn SCSI WRITE command. .Bl -tag -width 12n .It Fl l Ar lba Specify the starting Logical Block Address for the WRITE. This can be specified in decimal, octal (starting with 0), hexadecimal (starting with 0x) or any other base supported by .Xr strtoull 3 . .It Fl d Ar atalen Specify the length, in 512 byte blocks, of the WRITE request. .It Fl f Ar file Specify the source for the data to be written by the WRITE command. Either a filename or .Sq - for stdin may be specified. .It Fl c Ar cdbsize Specify the minimum .Tn SCSI CDB (Command Data Block) size to be used for the READ request. Allowable values are 6, 10, 12 and 16. Depending upon the LBA and amount of data requested, a larger CDB size may be used to satisfy the request. (e.g., for LBAs above 0xffffffff, READ(16) must be used to satisfy the request.) .It Fl b Ar blocksize Specify the blocksize of the underlying .Tn SCSI device, so the transfer length can be calculated accurately. The blocksize can be obtained via the .Tn SCSI READ CAPACITY command. .It Fl N Do not copy data to .Nm to the kernel when doing a write, just execute the command without copying data. This is to be used for performance testing. .El .It Ic readcap Send the .Tn SCSI READ CAPACITY command to the device and display the device size and device block size. By default, READ CAPACITY(10) is used. If the device returns a maximum LBA of 0xffffffff, however, .Nm will automatically issue a READ CAPACITY(16), which is implemented as a service action of the SERVICE ACTION IN(16) opcode. The user can specify the minimum CDB size with the .Fl c argument. Valid values for the .Fl c option are 10 and 16. If a 10 byte CDB is specified, the request will be automatically reissued with a 16 byte CDB if the maximum LBA returned is 0xffffffff. .It Ic modesense Send a .Tn SCSI MODE SENSE command to the device, and display the requested mode page(s) or page list. .Bl -tag -width 10n .It Fl m Ar page Specify the mode page to display. This option and the .Fl l option are mutually exclusive. One of the two must be specified, though. Mode page numbers may be specified in decimal or hexadecimal. .It Fl l Request that the list of mode pages supported by the device be returned. This option and the .Fl m option are mutually exclusive. One of the two must be specified, though. .It Fl P Ar pc Specify the mode page control value. Possible values are: .Bl -tag -width 2n -compact .It 0 Current values. .It 1 Changeable value bitmask. .It 2 Default values. .It 3 Saved values. .El .It Fl d Disable block descriptors when sending the mode sense request. .It Fl S Ar subpage Specify the subpage used with the mode sense request. .It Fl c Ar cdbsize Specify the CDB size used for the mode sense request. Supported values are 6 and 10. .El .It Ic start Send the .Tn SCSI START STOP UNIT command to the specified LUN with the start bit set. .Bl -tag -width 4n .It Fl i Set the immediate bit in the CDB. Note that CTL does not support the immediate bit, so this is primarily useful for making sure that CTL returns the proper error. .El .It Ic stop Send the .Tn SCSI START STOP UNIT command to the specified LUN with the start bit cleared. We use an ordered tag to stop the LUN, so we can guarantee that all pending I/O executes before it is stopped. (CTL guarantees this anyway, but .Nm sends an ordered tag for completeness.) .Bl -tag -width 4n .It Fl i Set the immediate bit in the CDB. Note that CTL does not support the immediate bit, so this is primarily useful for making sure that CTL returns the proper error. .El .It Ic synccache Send the .Tn SCSI SYNCHRONIZE CACHE command to the device. By default, SYNCHRONIZE CACHE(10) is used. If the specified starting LBA is greater than 0xffffffff or the length is greater than 0xffff, though, SYNCHRONIZE CACHE(16) will be used. The 16 byte command will also be used if the user specifies a 16 byte CDB with the .Fl c argument. .Bl -tag -width 14n .It Fl l Ar lba Specify the starting LBA of the cache region to synchronize. This option is a no-op for CTL. If you send a SYNCHRONIZE CACHE command, it will sync the cache for the entire LUN. .It Fl b Ar blockcount Specify the length of the cache region to synchronize. This option is a no-op for CTL. If you send a SYNCHRONIZE CACHE command, it will sync the cache for the entire LUN. .It Fl r Specify relative addressing for the starting LBA. CTL does not support relative addressing, since it only works for linked commands, and CTL does not support linked commands. .It Fl i Tell the target to return status immediately after issuing the SYNCHRONIZE CACHE command rather than waiting for the cache to finish syncing. CTL does not support this bit. .It Fl c Ar cdbsize Specify the minimum CDB size. Valid values are 10 and 16 bytes. .El .It Ic lunlist List all LUNs registered with CTL. Because this command uses the ioctl port, it will only work when the FETDs (Front End Target Drivers) are enabled. This command is the equivalent of doing a REPORT LUNS on one LUN and then an INQUIRY on each LUN in the system. .It Ic delay Delay commands at the given location. There are two places where commands may be delayed currently: before data is transferred .Pq Dq datamove and just prior to sending status to the host .Pq Dq done . One of the two must be supplied as an argument to the .Fl l option. The .Fl t option must also be specified. .Bl -tag -width 12n .It Fl l Ar delayloc Delay command(s) at the specified location. This can either be at the data movement stage (datamove) or prior to command completion (done). .It Fl t Ar delaytime Delay command(s) for the specified number of seconds. This must be specified. If set to 0, it will clear out any previously set delay for this particular location (datamove or done). .It Fl T Ar delaytype Specify the delay type. By default, the .Ic delay option will delay the next command sent to the given LUN. With the .Fl T Ar cont option, every command will be delayed by the specified period of time. With the .Fl T Ar oneshot the next command sent to the given LUN will be delayed and all subsequent commands will be completed normally. This is the default. .El .It Ic inject Inject the specified type of error for the LUN specified, when a command that matches the given pattern is seen. The sense data returned is in either fixed or descriptor format, depending upon the status of the D_SENSE bit in the control mode page (page 0xa) for the LUN. .Pp Errors are only injected for commands that have not already failed for other reasons. By default, only the first command matching the pattern specified is returned with the supplied error. .Pp If the .Fl c flag is specified, all commands matching the pattern will be returned with the specified error until the error injection command is deleted with .Fl d flag. .Bl -tag -width 17n .It Fl i Ar action Specify the error to return: .Bl -tag -width 10n .It aborted Return the next matching command on the specified LUN with the sense key ABORTED COMMAND (0x0b), and the ASC/ASCQ 0x45,0x00 ("Select or reselect failure"). .It mediumerr Return the next matching command on the specified LUN with the sense key MEDIUM ERROR (0x03) and the ASC/ASCQ 0x11,0x00 ("Unrecovered read error") for reads, or ASC/ASCQ 0x0c,0x02 ("Write error - auto reallocation failed") for write errors. .It ua Return the next matching command on the specified LUN with the sense key UNIT ATTENTION (0x06) and the ASC/ASCQ 0x29,0x00 ("POWER ON, RESET, OR BUS DEVICE RESET OCCURRED"). .It custom Return the next matching command on the specified LUN with the supplied sense data. The .Fl s argument must be specified. .El .It Fl p Ar pattern Specify which commands should be returned with the given error. .Bl -tag -width 10n .It read The error should apply to READ(6), READ(10), READ(12), READ(16), etc. .It write The error should apply to WRITE(6), WRITE(10), WRITE(12), WRITE(16), WRITE AND VERIFY(10), etc. .It rw The error should apply to both read and write type commands. .It readcap The error should apply to READ CAPACITY(10) and READ CAPACITY(16) commands. .It tur The error should apply to TEST UNIT READY commands. .It any The error should apply to any command. .El .It Fl r Ar lba,len Specify the starting lba and length of the range of LBAs which should trigger an error. This option is only applies when read and/or write patterns are specified. If used with other command types, the error will never be triggered. .It Fl s Ar len fmt Op Ar args Specify the sense data that is to be returned for custom actions. If the format is .Sq - , len bytes of sense data will be read from standard input and written to the sense buffer. If len is longer than 252 bytes (the maximum allowable .Tn SCSI sense data length), it will be truncated to that length. The sense data format is described in .Xr cam_cdbparse 3 . .It Fl c The error injection should be persistent, instead of happening once. Persistent errors must be deleted with the .Fl d argument. .It Fl d Ar delete_id Delete the specified error injection serial number. The serial number is returned when the error is injected. .El .It Ic port Perform one of several CTL frontend port operations. Either get a list of frontend ports .Pq Fl l , turn one or more frontends on or off .Pq Fl o Ar on|off , or set the World Wide Node Name .Pq Fl w Ar wwnn or World Wide Port Name .Pq Fl W Ar wwpn for a given port. One of .Fl l , .Fl o , or .Fl w or .Fl W must be specified. The WWNN and WWPN may both be specified at the same time, but cannot be combined with enabling/disabling or listing ports. .Bl -tag -width 12n +.It Fl c +Create new frontend port using free pp and vp=0. .It Fl o Ar on|off -Turn the specified CTL frontend ports off or on. +Turn the specified CTL frontend ports on or off. If no port number or port type is specified, all ports are turned on or off. +.It Fl O Ar pp|vp +Specify generic options on the ioctl frontend port. +At present, only pp and vp port numbers can be set. .It Fl p Ar targ_port Specify the frontend port number. The port numbers can be found in the frontend port list. +.It Fl r +Remove port specified with +.Pq Fl p Ar targ_port . .It Fl t Ar fe_type Specify the frontend type. Currently defined port types are .Dq fc (Fibre Channel), .Dq scsi (Parallel SCSI), .Dq ioctl (CTL ioctl interface), and .Dq internal (CTL CAM SIM). .It Fl w Ar wwnn Set the World Wide Node Name for the given port. The .Fl n argument must be specified, since this is only possible to implement on a single port. As a general rule, the WWNN should be the same across all ports on the system. .It Fl W Ar wwpn Set the World Wide Port Name for the given port. The .Fl n argument must be specified, since this is only possible to implement on a single port. As a general rule, the WWPN must be different for every port in the system. .El .It Ic portlist List CTL frontend ports. .Bl -tag -width 12n .It Fl f Ar frontend Specify the frontend type. .It Fl i Report target and connected initiators addresses. .It Fl l Report LUN mapping. .It Fl p Ar targ_port Specify the frontend port number. .It Fl q Omit the header in the port list output. .It Fl v Enable verbose output (report all port options). .It Fl x Output the port list in XML format. .El .It Ic lunmap Change LUN mapping for specified port. If both .Ar pLUN and .Ar cLUN are specified -- LUN will be mapped. If .Ar pLUN is specified, but .Ar cLUN is not -- LUN will be unmapped. If neither .Ar pLUN nor .Ar cLUN are specified -- LUN mapping will be disabled, exposing all CTL LUNs. .Bl -tag -width 12n .It Fl p Ar targ_port Specify the frontend port number. .It Fl l Ar pLUN LUN number visible by specified port. .It Fl L Ar cLUN CTL LUN number. .El .It Ic dumpooa Dump the OOA (Order Of Arrival) queue for each LUN registered with CTL. .It Ic dumpstructs Dump the CTL structures to the console. .It Ic create Create a new LUN. The backend must be specified, and depending upon the backend requested, some of the other options may be required. If the LUN is created successfully, the LUN configuration will be displayed. If LUN creation fails, a message will be displayed describing the failure. .Bl -tag -width 14n .It Fl b Ar backend The .Fl b flag is required. This specifies the name backend to use when creating the LUN. Examples are .Dq ramdisk and .Dq block . .It Fl B Ar blocksize Specify the blocksize of the backend in bytes. .It Fl d Ar device_id Specify the LUN-associated string to use in the .Tn SCSI INQUIRY VPD page 0x83 data. .It Fl l Ar lun_id Request that a particular LUN number be assigned. If the requested LUN number is not available, the request will fail. .It Fl o Ar name=value Specify a backend-specific name/value pair. Multiple .Fl o arguments may be specified. Refer to the backend documentation for arguments that may be used. .It Fl s Ar size_bytes Specify the size of the LUN in bytes. Some backends may allow setting the size (e.g. the ramdisk backend) and for others the size may be implicit (e.g. the block backend). .It Fl S Ar serial_num Specify the serial number to be used in the .Tn SCSI INQUIRY VPD page 0x80 data. .It Fl t Ar device_type Specify the numeric SCSI device type to use when creating the LUN. If this flag is not used, the type of LUN created is backend-specific. Not all LUN types are supported. Currently CTL supports Direct Access (type 0), Processor (type 3) and CD/DVD (type 5) LUNs. The backend requested may or may not support all of the LUN types that CTL supports. .El .It Ic remove Remove a LUN. The backend must be specified, and the LUN number must also be specified. Backend-specific options may also be specified with the .Fl o flag. .Bl -tag -width 14n .It Fl b Ar backend Specify the backend that owns the LUN to be removed. Examples are .Dq ramdisk and .Dq block . .It Fl l Ar lun_id Specify the LUN number to remove. .It Fl o Ar name=value Specify a backend-specific name/value pair. Multiple .Fl o arguments may be specified. Refer to the backend documentation for arguments that may be used. .El .It Ic modify Modify a LUN size. The backend, the LUN number, and the size must be specified. .Bl -tag -width 14n .It Fl b Ar backend Specify the backend that owns the LUN to be modified. Examples are .Dq ramdisk and .Dq block . .It Fl l Ar lun_id Specify the LUN number to modify. .It Fl o Ar name=value Specify a backend-specific name/value pair. Multiple .Fl o arguments may be specified. Refer to the backend documentation for arguments that may be used. .It Fl s Ar size_bytes Specify the size of the LUN in bytes. For the .Dq block backend, an .Dq auto keyword may be passed instead; this will make CTL use the size of backing file or device. .El .It Ic devlist Get a list of all configured LUNs. This also includes the LUN size and blocksize, serial number and device ID. .Bl -tag -width 11n .It Fl b Ar backend Specify the backend. This restricts the LUN list to the named backend. Examples are .Dq ramdisk and .Dq block . .It Fl v Be verbose. This will also display any backend-specific LUN attributes in addition to the standard per-LUN information. .It Fl x Dump the raw XML. The LUN list information from the kernel comes in XML format, and this option allows the display of the raw XML data. This option and the .Fl v and .Fl b options are mutually exclusive. If you specify .Fl x , the entire LUN database is displayed in XML format. .El .It Ic islist Get a list of currently running iSCSI sessions. This includes initiator and target names and the unique connection IDs. .Bl -tag -width 11n .It Fl v Verbose mode. .It Fl x Dump the raw XML. The sessions list information from the kernel comes in XML format, and this option allows the display of the raw XML data. .El .It Ic islogout Ask the initiator to log out iSCSI sessions matching criteria. .Bl -tag -width 11n .It Fl a Log out all sessions. .It Fl c Specify connection ID. .It Fl i Specify initiator name. .It Fl p Specify initiator portal (hostname or IP address). .El .It Ic isterminate Forcibly terminate iSCSI sessions matching criteria. .Bl -tag -width 11n .It Fl a Terminate all sessions. .It Fl c Specify connection ID. .It Fl i Specify initiator name. .It Fl p Specify initiator portal (hostname or IP address). .El .It Ic help Display .Nm usage information. .El .Sh OPTIONS Number of additional configuration options may be specified for LUNs. Some options are global, others are backend-specific. .Pp Global options: .Bl -tag -width 12n .It Va vendor Specifies LUN vendor string up to 8 chars. .It Va product Specifies LUN product string up to 16 chars. .It Va revision Specifies LUN revision string up to 4 chars. .It Va scsiname Specifies LUN SCSI name string. .It Va eui Specifies LUN EUI-64 identifier. .It Va naa Specifies LUN NAA identifier. .It Va uuid Specifies LUN locally assigned RFC 4122 UUID identifier. EUI, NAA or UUID identifier should be set to UNIQUE value to allow EXTENDED COPY command access the LUN. Non-unique LUN identifiers may lead to data corruption. Some initiators may not support later introduced UUID identifiers. .It Va ha_role Setting to "primary" or "secondary" overrides default role of the node in HA cluster, set by kern.cam.ctl.ha_role sysctl. .It Va insecure_tpc Setting to "on" allows EXTENDED COPY command sent to this LUN access other LUNs on this host, not accessible otherwise. This allows to offload copying between different iSCSI targets residing on the same host in trusted environments. .It Va readcache Set to "off", disables read caching for the LUN, if supported by the backend. .It Va readonly Set to "on", blocks all media write operations to the LUN, reporting it as write protected. .It Va removable Set to "on", makes LUN removable. .It Va reordering Set to "unrestricted", allows target to process commands with SIMPLE task attribute in arbitrary order. Any data integrity exposures related to command sequence order shall be explicitly handled by the application client through the selection of appropriate commands and task attributes. The default value is "restricted". It improves data integrity, but may introduce some additional delays. .It Va serseq Set to "on" to serialize consecutive reads/writes. Set to "read" to serialize consecutive reads. Set to "off" to allow them be issued in parallel. Parallel issue of consecutive operations may confuse logic of the backing file system, hurting performance; but it may improve performance of backing stores without prefetch/write-back. .It Va pblocksize .It Va pblockoffset Specify physical block size and offset of the device. .It Va ublocksize .It Va ublockoffset Specify UNMAP block size and offset of the device. .It Va rpm Specifies medium rotation rate of the device: 0 -- not reported, 1 -- non-rotating (SSD), >1024 -- value in revolutions per minute. .It Va formfactor Specifies nominal form factor of the device: 0 -- not reported, 1 -- 5.25", 2 -- 3.5", 3 -- 2.5", 4 -- 1.8", 5 -- less then 1.8". .It Va provisioning_type When UNMAP support is enabled, this option specifies provisioning type: "resource", "thin" or "unknown". Default value is "thin". Logical units without UNMAP support are reported as fully provisioned. .It Va unmap Setting to "on" or "off" controls UNMAP support for the logical unit. Default value is "on" if supported by the backend. .It Va unmap_max_lba .It Va unmap_max_descr Specify maximum allowed number of LBAs and block descriptors per UNMAP command to report in Block Limits VPD page. .It Va write_same_max_lba Specify maximum allowed number of LBAs per WRITE SAME command to report in Block Limits VPD page. .It Va avail-threshold .It Va used-threshold .It Va pool-avail-threshold .It Va pool-used-threshold Set per-LUN/-pool thin provisioning soft thresholds. LUN will establish UNIT ATTENTION condition if its or pool available space get below configured avail values, or its or pool used space get above configured used values. Pool thresholds are working only for ZVOL-backed LUNs. .It Va writecache Set to "off", disables write caching for the LUN, if supported by the backend. .El .Pp Options specific for block backend: .Bl -tag -width 12n .It Va file Specifies file or device name to use for backing store. .It Va num_threads Specifies number of backend threads to use for this LUN. .El .Pp Options specific for ramdisk backend: .Bl -tag -width 12n .It Va capacity Specifies capacity of backing store (maximum RAM for data). The default value is zero, that disables backing store completely, making all writes go to nowhere, while all reads return zeroes. .El .Sh EXAMPLES -.Dl ctladm tur 1 .Pp Send a .Tn SCSI TEST UNIT READY command to LUN 1. .Pp -.Dl ctladm modesense 1 -l +.Dl ctladm tur 1 .Pp Display the list of mode pages supported by LUN 1. .Pp -.Dl ctladm modesense 0 -m 10 -P 3 -d -c 10 +.Dl ctladm modesense 1 -l .Pp Display the saved version of the Control mode page (page 10) on LUN 0. Disable fetching block descriptors, and use a 10 byte MODE SENSE command instead of the default 6 byte command. -.Bd -literal -ctladm read 2 -l 0 -d 1 -b 512 -f - > foo -.Ed .Pp +.Dl ctladm modesense 0 -m 10 -P 3 -d -c 10 +.Pp Read the first 512 byte block from LUN 2 and dump it to the file -.Pa foo . .Bd -literal -ctladm write 3 -l 0xff432140 -d 20 -b 512 -f /tmp/bar +.Dl ctladm read 2 -l 0 -d 1 -b 512 -f - > foo .Ed .Pp Read 10240 bytes from the file .Pa /tmp/bar and write it to LUN 3. starting at LBA 0xff432140. .Pp -.Dl ctladm create -b ramdisk -s 10485760000000000 +.Bd -literal +.Dl ctladm write 3 -l 0xff432140 -d 20 -b 512 -f /tmp/bar +.Ed .Pp Create a LUN with the .Dq fake ramdisk as a backing store. The LUN will claim to have a size of approximately 10 terabytes, while having no real data store (all written data are lost). .Pp -.Dl ctladm create -b ramdisk -s 10T -o capacity=10G +.Dl ctladm create -b ramdisk -s 10485760000000000 .Pp Create a thin provisioned LUN with a ramdisk as a backing store. The LUN will have maximal backing store capacity of 10 gigabytes, while reporting size of 10 terabytes, .Pp -.Dl ctladm create -b block -o file=src/usr.sbin/ctladm/ctladm.8 +.Dl ctladm create -b ramdisk -s 10T -o capacity=10G .Pp Create a LUN using the block backend, and specify the file .Pa src/usr.sbin/ctladm/ctladm.8 as the backing store. The size of the LUN will be derived from the size of the file. .Pp -.Dl ctladm create -b block -o file=src/usr.sbin/ctladm/ctladm.8 -S MYSERIAL321 -d MYDEVID123 +.Dl ctladm create -b block -o file=src/usr.sbin/ctladm/ctladm.8 .Pp Create a LUN using the block backend, specify the file .Pa src/usr.sbin/ctladm/ctladm.8 as the backing store, and specify the .Tn SCSI VPD page 0x80 and 0x83 serial number .Fl ( S ) and device ID .Fl ( d ) . .Pp -.Dl ctladm remove -b block -l 12 +.Dl ctladm create -b block -o file=src/usr.sbin/ctladm/ctladm.8 -S MYSERIAL321 -d MYDEVID123 .Pp +Use to specify generic options on ioctl frontend port, now it is +only possible to set pp and/or vp port number. +.Pp +.Dl ctladm port -c -O pp=11 -O vp=12 +.Pp +Remove specified targ_port. +.Pp +.Dl ctladm port -r -p 4 +.Pp +.Pp Remove LUN 12, which is handled by the block backend, from the system. .Pp -.Dl ctladm devlist +.Dl ctladm remove -b block -l 12 .Pp List configured LUNs in the system, along with their backend and serial number. This works when the Front End Target Drivers are enabled or disabled. .Pp -.Dl ctladm lunlist +.Dl ctladm devlist .Pp List all LUNs in the system, along with their inquiry data and device type. This only works when the FETDs are enabled, since the commands go through the ioctl port. .Pp -.Dl ctladm inject 6 -i mediumerr -p read -r 0,512 -c +.Dl ctladm lunlist .Pp Inject a medium error on LUN 6 for every read that covers the first 512 blocks of the LUN. -.Bd -literal -offset indent -ctladm inject 6 -i custom -p tur -s 18 "f0 0 02 s12 04 02" -.Ed .Pp +.Dl ctladm inject 6 -i mediumerr -p read -r 0,512 -c +.Pp Inject a custom error on LUN 6 for the next TEST UNIT READY command only. This will result in a sense key of NOT READY (0x02), and an ASC/ASCQ of 0x04,0x02 ("Logical unit not ready, initializing command required"). +.Pp +.Bd -literal -offset indent +ctladm inject 6 -i custom -p tur -s 18 "f0 0 02 s12 04 02" +.Ed .Sh SEE ALSO .Xr cam 3 , .Xr cam_cdbparse 3 , .Xr cam 4 , .Xr ctl 4 , .Xr xpt 4 , .Xr camcontrol 8 , .Xr ctld 8 , .Xr ctlstat 8 .Sh HISTORY The .Nm utility was originally written during the Winter/Spring of 2003 as an interface to CTL. .Sh AUTHORS .An Ken Merry Aq Mt ken@FreeBSD.org Index: head/usr.sbin/ctladm/ctladm.c =================================================================== --- head/usr.sbin/ctladm/ctladm.c (revision 333445) +++ head/usr.sbin/ctladm/ctladm.c (revision 333446) @@ -1,4284 +1,4273 @@ /*- + * SPDX-License-Identifier: BSD-2-Clause-FreeBSD + * * Copyright (c) 2003, 2004 Silicon Graphics International Corp. * Copyright (c) 1997-2007 Kenneth D. Merry * Copyright (c) 2012 The FreeBSD Foundation + * Copyright (c) 2018 Marcelo Araujo * All rights reserved. * * Portions of this software were developed by Edward Tomasz Napierala * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/usr.sbin/ctladm/ctladm.c#4 $ */ /* * CAM Target Layer exercise program. * * Author: Ken Merry */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ctladm.h" #ifdef min #undef min #endif #define min(x,y) (x < y) ? x : y typedef enum { CTLADM_CMD_TUR, CTLADM_CMD_INQUIRY, CTLADM_CMD_REQ_SENSE, CTLADM_CMD_ARRAYLIST, CTLADM_CMD_REPORT_LUNS, CTLADM_CMD_HELP, CTLADM_CMD_DEVLIST, CTLADM_CMD_ADDDEV, CTLADM_CMD_RM, CTLADM_CMD_CREATE, CTLADM_CMD_READ, CTLADM_CMD_WRITE, CTLADM_CMD_PORT, CTLADM_CMD_PORTLIST, CTLADM_CMD_READCAPACITY, CTLADM_CMD_MODESENSE, CTLADM_CMD_DUMPOOA, CTLADM_CMD_DUMPSTRUCTS, CTLADM_CMD_START, CTLADM_CMD_STOP, CTLADM_CMD_SYNC_CACHE, CTLADM_CMD_LUNLIST, CTLADM_CMD_DELAY, CTLADM_CMD_ERR_INJECT, CTLADM_CMD_PRES_IN, CTLADM_CMD_PRES_OUT, CTLADM_CMD_INQ_VPD_DEVID, CTLADM_CMD_RTPG, CTLADM_CMD_MODIFY, CTLADM_CMD_ISLIST, CTLADM_CMD_ISLOGOUT, CTLADM_CMD_ISTERMINATE, CTLADM_CMD_LUNMAP } ctladm_cmdfunction; typedef enum { CTLADM_ARG_NONE = 0x0000000, CTLADM_ARG_AUTOSENSE = 0x0000001, CTLADM_ARG_DEVICE = 0x0000002, CTLADM_ARG_ARRAYSIZE = 0x0000004, CTLADM_ARG_BACKEND = 0x0000008, CTLADM_ARG_CDBSIZE = 0x0000010, CTLADM_ARG_DATALEN = 0x0000020, CTLADM_ARG_FILENAME = 0x0000040, CTLADM_ARG_LBA = 0x0000080, CTLADM_ARG_PC = 0x0000100, CTLADM_ARG_PAGE_CODE = 0x0000200, CTLADM_ARG_PAGE_LIST = 0x0000400, CTLADM_ARG_SUBPAGE = 0x0000800, CTLADM_ARG_PAGELIST = 0x0001000, CTLADM_ARG_DBD = 0x0002000, CTLADM_ARG_TARG_LUN = 0x0004000, CTLADM_ARG_BLOCKSIZE = 0x0008000, CTLADM_ARG_IMMED = 0x0010000, CTLADM_ARG_RELADR = 0x0020000, CTLADM_ARG_RETRIES = 0x0040000, CTLADM_ARG_ONOFFLINE = 0x0080000, CTLADM_ARG_ONESHOT = 0x0100000, CTLADM_ARG_TIMEOUT = 0x0200000, CTLADM_ARG_INITIATOR = 0x0400000, CTLADM_ARG_NOCOPY = 0x0800000, CTLADM_ARG_NEED_TL = 0x1000000 } ctladm_cmdargs; struct ctladm_opts { const char *optname; uint32_t cmdnum; ctladm_cmdargs argnum; const char *subopt; }; typedef enum { CC_OR_NOT_FOUND, CC_OR_AMBIGUOUS, CC_OR_FOUND } ctladm_optret; static const char rw_opts[] = "Nb:c:d:f:l:"; static const char startstop_opts[] = "i"; static struct ctladm_opts option_table[] = { {"adddev", CTLADM_CMD_ADDDEV, CTLADM_ARG_NONE, NULL}, {"create", CTLADM_CMD_CREATE, CTLADM_ARG_NONE, "b:B:d:l:o:s:S:t:"}, {"delay", CTLADM_CMD_DELAY, CTLADM_ARG_NEED_TL, "T:l:t:"}, {"devid", CTLADM_CMD_INQ_VPD_DEVID, CTLADM_ARG_NEED_TL, NULL}, {"devlist", CTLADM_CMD_DEVLIST, CTLADM_ARG_NONE, "b:vx"}, {"dumpooa", CTLADM_CMD_DUMPOOA, CTLADM_ARG_NONE, NULL}, {"dumpstructs", CTLADM_CMD_DUMPSTRUCTS, CTLADM_ARG_NONE, NULL}, {"help", CTLADM_CMD_HELP, CTLADM_ARG_NONE, NULL}, {"inject", CTLADM_CMD_ERR_INJECT, CTLADM_ARG_NEED_TL, "cd:i:p:r:s:"}, {"inquiry", CTLADM_CMD_INQUIRY, CTLADM_ARG_NEED_TL, NULL}, {"islist", CTLADM_CMD_ISLIST, CTLADM_ARG_NONE, "vx"}, {"islogout", CTLADM_CMD_ISLOGOUT, CTLADM_ARG_NONE, "ac:i:p:"}, {"isterminate", CTLADM_CMD_ISTERMINATE, CTLADM_ARG_NONE, "ac:i:p:"}, {"lunlist", CTLADM_CMD_LUNLIST, CTLADM_ARG_NONE, NULL}, {"lunmap", CTLADM_CMD_LUNMAP, CTLADM_ARG_NONE, "p:l:L:"}, {"modesense", CTLADM_CMD_MODESENSE, CTLADM_ARG_NEED_TL, "P:S:dlm:c:"}, {"modify", CTLADM_CMD_MODIFY, CTLADM_ARG_NONE, "b:l:o:s:"}, - {"port", CTLADM_CMD_PORT, CTLADM_ARG_NONE, "lo:p:qt:w:W:x"}, + {"port", CTLADM_CMD_PORT, CTLADM_ARG_NONE, "lo:O:d:crp:qt:w:W:x"}, {"portlist", CTLADM_CMD_PORTLIST, CTLADM_ARG_NONE, "f:ilp:qvx"}, {"prin", CTLADM_CMD_PRES_IN, CTLADM_ARG_NEED_TL, "a:"}, {"prout", CTLADM_CMD_PRES_OUT, CTLADM_ARG_NEED_TL, "a:k:r:s:"}, {"read", CTLADM_CMD_READ, CTLADM_ARG_NEED_TL, rw_opts}, {"readcapacity", CTLADM_CMD_READCAPACITY, CTLADM_ARG_NEED_TL, "c:"}, {"remove", CTLADM_CMD_RM, CTLADM_ARG_NONE, "b:l:o:"}, {"reportluns", CTLADM_CMD_REPORT_LUNS, CTLADM_ARG_NEED_TL, NULL}, {"reqsense", CTLADM_CMD_REQ_SENSE, CTLADM_ARG_NEED_TL, NULL}, {"rtpg", CTLADM_CMD_RTPG, CTLADM_ARG_NEED_TL, NULL}, {"start", CTLADM_CMD_START, CTLADM_ARG_NEED_TL, startstop_opts}, {"stop", CTLADM_CMD_STOP, CTLADM_ARG_NEED_TL, startstop_opts}, {"synccache", CTLADM_CMD_SYNC_CACHE, CTLADM_ARG_NEED_TL, "b:c:il:r"}, {"tur", CTLADM_CMD_TUR, CTLADM_ARG_NEED_TL, NULL}, {"write", CTLADM_CMD_WRITE, CTLADM_ARG_NEED_TL, rw_opts}, {"-?", CTLADM_CMD_HELP, CTLADM_ARG_NONE, NULL}, {"-h", CTLADM_CMD_HELP, CTLADM_ARG_NONE, NULL}, {NULL, 0, 0, NULL} }; ctladm_optret getoption(struct ctladm_opts *table, char *arg, uint32_t *cmdnum, ctladm_cmdargs *argnum, const char **subopt); static int cctl_dump_ooa(int fd, int argc, char **argv); static int cctl_port(int fd, int argc, char **argv, char *combinedopt); static int cctl_do_io(int fd, int retries, union ctl_io *io, const char *func); static int cctl_delay(int fd, int lun, int argc, char **argv, char *combinedopt); static int cctl_lunlist(int fd); static int cctl_sync_cache(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt); static int cctl_start_stop(int fd, int lun, int iid, int retries, int start, int argc, char **argv, char *combinedopt); static int cctl_mode_sense(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt); static int cctl_read_capacity(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt); static int cctl_read_write(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt, ctladm_cmdfunction command); static int cctl_get_luns(int fd, int lun, int iid, int retries, struct scsi_report_luns_data **lun_data, uint32_t *num_luns); static int cctl_report_luns(int fd, int lun, int iid, int retries); static int cctl_tur(int fd, int lun, int iid, int retries); static int cctl_get_inquiry(int fd, int lun, int iid, int retries, char *path_str, int path_len, struct scsi_inquiry_data *inq_data); static int cctl_inquiry(int fd, int lun, int iid, int retries); static int cctl_req_sense(int fd, int lun, int iid, int retries); static int cctl_persistent_reserve_in(int fd, int lun, int initiator, int argc, char **argv, char *combinedopt, int retry_count); static int cctl_persistent_reserve_out(int fd, int lun, int initiator, int argc, char **argv, char *combinedopt, int retry_count); static int cctl_create_lun(int fd, int argc, char **argv, char *combinedopt); static int cctl_inquiry_vpd_devid(int fd, int lun, int initiator); static int cctl_report_target_port_group(int fd, int lun, int initiator); static int cctl_modify_lun(int fd, int argc, char **argv, char *combinedopt); static int cctl_portlist(int fd, int argc, char **argv, char *combinedopt); ctladm_optret getoption(struct ctladm_opts *table, char *arg, uint32_t *cmdnum, ctladm_cmdargs *argnum, const char **subopt) { struct ctladm_opts *opts; int num_matches = 0; for (opts = table; (opts != NULL) && (opts->optname != NULL); opts++) { if (strncmp(opts->optname, arg, strlen(arg)) == 0) { *cmdnum = opts->cmdnum; *argnum = opts->argnum; *subopt = opts->subopt; if (strcmp(opts->optname, arg) == 0) return (CC_OR_FOUND); if (++num_matches > 1) return(CC_OR_AMBIGUOUS); } } if (num_matches > 0) return(CC_OR_FOUND); else return(CC_OR_NOT_FOUND); } static int cctl_dump_ooa(int fd, int argc, char **argv) { struct ctl_ooa ooa; long double cmd_latency; int num_entries, len, lun = -1, retval = 0; unsigned int i; num_entries = 104; if ((argc > 2) && (isdigit(argv[2][0]))) lun = strtol(argv[2], NULL, 0); retry: len = num_entries * sizeof(struct ctl_ooa_entry); bzero(&ooa, sizeof(ooa)); ooa.entries = malloc(len); if (ooa.entries == NULL) { warn("%s: error mallocing %d bytes", __func__, len); return (1); } if (lun >= 0) { ooa.lun_num = lun; } else ooa.flags |= CTL_OOA_FLAG_ALL_LUNS; ooa.alloc_len = len; ooa.alloc_num = num_entries; if (ioctl(fd, CTL_GET_OOA, &ooa) == -1) { warn("%s: CTL_GET_OOA ioctl failed", __func__); retval = 1; goto bailout; } if (ooa.status == CTL_OOA_NEED_MORE_SPACE) { num_entries = num_entries * 2; free(ooa.entries); ooa.entries = NULL; goto retry; } if (ooa.status != CTL_OOA_OK) { warnx("%s: CTL_GET_OOA ioctl returned error %d", __func__, ooa.status); retval = 1; goto bailout; } fprintf(stdout, "Dumping OOA queues\n"); for (i = 0; i < ooa.fill_num; i++) { struct ctl_ooa_entry *entry; char cdb_str[(SCSI_MAX_CDBLEN * 3) +1]; struct bintime delta_bt; struct timespec ts; entry = &ooa.entries[i]; delta_bt = ooa.cur_bt; bintime_sub(&delta_bt, &entry->start_bt); bintime2timespec(&delta_bt, &ts); cmd_latency = ts.tv_sec * 1000; if (ts.tv_nsec > 0) cmd_latency += ts.tv_nsec / 1000000; fprintf(stdout, "LUN %jd tag 0x%04x%s%s%s%s%s: %s. CDB: %s " "(%0.0Lf ms)\n", (intmax_t)entry->lun_num, entry->tag_num, (entry->cmd_flags & CTL_OOACMD_FLAG_BLOCKED) ? " BLOCKED" : "", (entry->cmd_flags & CTL_OOACMD_FLAG_DMA) ? " DMA" : "", (entry->cmd_flags & CTL_OOACMD_FLAG_DMA_QUEUED) ? " DMAQUEUED" : "", (entry->cmd_flags & CTL_OOACMD_FLAG_ABORT) ? " ABORT" : "", (entry->cmd_flags & CTL_OOACMD_FLAG_RTR) ? " RTR" :"", scsi_op_desc(entry->cdb[0], NULL), scsi_cdb_string(entry->cdb, cdb_str, sizeof(cdb_str)), cmd_latency); } fprintf(stdout, "OOA queues dump done\n"); bailout: free(ooa.entries); return (retval); } static int cctl_dump_structs(int fd, ctladm_cmdargs cmdargs __unused) { if (ioctl(fd, CTL_DUMP_STRUCTS) == -1) { warn(__func__); return (1); } return (0); } typedef enum { CCTL_PORT_MODE_NONE, CCTL_PORT_MODE_LIST, CCTL_PORT_MODE_SET, CCTL_PORT_MODE_ON, - CCTL_PORT_MODE_OFF + CCTL_PORT_MODE_OFF, + CCTL_PORT_MODE_CREATE, + CCTL_PORT_MODE_REMOVE } cctl_port_mode; static struct ctladm_opts cctl_fe_table[] = { {"fc", CTL_PORT_FC, CTLADM_ARG_NONE, NULL}, {"scsi", CTL_PORT_SCSI, CTLADM_ARG_NONE, NULL}, {"internal", CTL_PORT_INTERNAL, CTLADM_ARG_NONE, NULL}, {"iscsi", CTL_PORT_ISCSI, CTLADM_ARG_NONE, NULL}, {"sas", CTL_PORT_SAS, CTLADM_ARG_NONE, NULL}, {"all", CTL_PORT_ALL, CTLADM_ARG_NONE, NULL}, {NULL, 0, 0, NULL} }; static int cctl_port(int fd, int argc, char **argv, char *combinedopt) { int c; int32_t targ_port = -1; int retval = 0; int wwnn_set = 0, wwpn_set = 0; uint64_t wwnn = 0, wwpn = 0; cctl_port_mode port_mode = CCTL_PORT_MODE_NONE; struct ctl_port_entry entry; + struct ctl_req req; + char *driver = NULL; + nvlist_t *option_list; ctl_port_type port_type = CTL_PORT_NONE; int quiet = 0, xml = 0; + option_list = nvlist_create(0); + if (option_list == NULL) + err(1, "%s: unable to allocate nvlist", __func__); + while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'l': if (port_mode != CCTL_PORT_MODE_NONE) goto bailout_badarg; port_mode = CCTL_PORT_MODE_LIST; break; + case 'c': + port_mode = CCTL_PORT_MODE_CREATE; + break; + case 'r': + port_mode = CCTL_PORT_MODE_REMOVE; + break; case 'o': if (port_mode != CCTL_PORT_MODE_NONE) goto bailout_badarg; if (strcasecmp(optarg, "on") == 0) port_mode = CCTL_PORT_MODE_ON; else if (strcasecmp(optarg, "off") == 0) port_mode = CCTL_PORT_MODE_OFF; else { warnx("Invalid -o argument %s, \"on\" or " "\"off\" are the only valid args", optarg); retval = 1; goto bailout; } break; + case 'O': { + char *tmpstr; + char *name, *value; + + tmpstr = strdup(optarg); + name = strsep(&tmpstr, "="); + if (name == NULL) { + warnx("%s: option -O takes \"name=value\"" + "argument", __func__); + retval = 1; + goto bailout; + } + value = strsep(&tmpstr, "="); + if (value == NULL) { + warnx("%s: option -O takes \"name=value\"" + "argument", __func__); + retval = 1; + goto bailout; + } + + free(tmpstr); + nvlist_add_string(option_list, name, value); + break; + } + case 'd': + if (driver != NULL) { + warnx("%s: option -d cannot be specified twice", + __func__); + retval = 1; + goto bailout; + } + + driver = strdup(optarg); + break; case 'p': targ_port = strtol(optarg, NULL, 0); break; case 'q': quiet = 1; break; case 't': { ctladm_optret optret; ctladm_cmdargs argnum; const char *subopt; ctl_port_type tmp_port_type; optret = getoption(cctl_fe_table, optarg, &tmp_port_type, &argnum, &subopt); if (optret == CC_OR_AMBIGUOUS) { warnx("%s: ambiguous frontend type %s", __func__, optarg); retval = 1; goto bailout; } else if (optret == CC_OR_NOT_FOUND) { warnx("%s: invalid frontend type %s", __func__, optarg); retval = 1; goto bailout; } port_type |= tmp_port_type; break; } case 'w': if ((port_mode != CCTL_PORT_MODE_NONE) && (port_mode != CCTL_PORT_MODE_SET)) goto bailout_badarg; port_mode = CCTL_PORT_MODE_SET; wwnn = strtoull(optarg, NULL, 0); wwnn_set = 1; break; case 'W': if ((port_mode != CCTL_PORT_MODE_NONE) && (port_mode != CCTL_PORT_MODE_SET)) goto bailout_badarg; port_mode = CCTL_PORT_MODE_SET; wwpn = strtoull(optarg, NULL, 0); wwpn_set = 1; break; case 'x': xml = 1; break; } } + if (driver == NULL) + driver = strdup("ioctl"); + /* * The user can specify either one or more frontend types (-t), or * a specific frontend, but not both. * * If the user didn't specify a frontend type or number, set it to * all. This is primarily needed for the enable/disable ioctls. * This will be a no-op for the listing code. For the set ioctl, * we'll throw an error, since that only works on one port at a time. */ if ((port_type != CTL_PORT_NONE) && (targ_port != -1)) { warnx("%s: can only specify one of -t or -n", __func__); retval = 1; goto bailout; } else if ((targ_port == -1) && (port_type == CTL_PORT_NONE)) port_type = CTL_PORT_ALL; bzero(&entry, sizeof(entry)); /* * These are needed for all but list/dump mode. */ entry.port_type = port_type; entry.targ_port = targ_port; switch (port_mode) { case CCTL_PORT_MODE_LIST: { char opts[] = "xq"; char argx[] = "-x"; char argq[] = "-q"; char *argvx[2]; int argcx = 0; optind = 0; optreset = 1; if (xml) argvx[argcx++] = argx; if (quiet) argvx[argcx++] = argq; cctl_portlist(fd, argcx, argvx, opts); break; } + case CCTL_PORT_MODE_REMOVE: + if (targ_port == -1) { + warnx("%s: -r require -p", __func__); + retval = 1; + goto bailout; + } + case CCTL_PORT_MODE_CREATE: { + bzero(&req, sizeof(req)); + strlcpy(req.driver, driver, sizeof(req.driver)); + + if (port_mode == CCTL_PORT_MODE_REMOVE) { + req.reqtype = CTL_REQ_REMOVE; + nvlist_add_stringf(option_list, "port_id", "%d", + targ_port); + } else + req.reqtype = CTL_REQ_CREATE; + + req.args = nvlist_pack(option_list, &req.args_len); + if (req.args == NULL) { + warn("%s: error packing nvlist", __func__); + retval = 1; + goto bailout; + } + + retval = ioctl(fd, CTL_PORT_REQ, &req); + free(req.args); + if (retval == -1) { + warn("%s: CTL_PORT_REQ ioctl failed", __func__); + retval = 1; + goto bailout; + } + + switch (req.status) { + case CTL_LUN_ERROR: + warnx("error: %s", req.error_str); + retval = 1; + goto bailout; + case CTL_LUN_WARNING: + warnx("warning: %s", req.error_str); + break; + case CTL_LUN_OK: + break; + default: + warnx("unknown status: %d", req.status); + retval = 1; + goto bailout; + } + + break; + } case CCTL_PORT_MODE_SET: if (targ_port == -1) { warnx("%s: -w and -W require -n", __func__); retval = 1; goto bailout; } if (wwnn_set) { entry.flags |= CTL_PORT_WWNN_VALID; entry.wwnn = wwnn; } if (wwpn_set) { entry.flags |= CTL_PORT_WWPN_VALID; entry.wwpn = wwpn; } if (ioctl(fd, CTL_SET_PORT_WWNS, &entry) == -1) { warn("%s: CTL_SET_PORT_WWNS ioctl failed", __func__); retval = 1; goto bailout; } break; case CCTL_PORT_MODE_ON: if (ioctl(fd, CTL_ENABLE_PORT, &entry) == -1) { warn("%s: CTL_ENABLE_PORT ioctl failed", __func__); retval = 1; goto bailout; } fprintf(stdout, "Front End Ports enabled\n"); break; case CCTL_PORT_MODE_OFF: if (ioctl(fd, CTL_DISABLE_PORT, &entry) == -1) { warn("%s: CTL_DISABLE_PORT ioctl failed", __func__); retval = 1; goto bailout; } fprintf(stdout, "Front End Ports disabled\n"); break; default: warnx("%s: one of -l, -o or -w/-W must be specified", __func__); retval = 1; goto bailout; break; } bailout: - + nvlist_destroy(req.args_nvl); + free(driver); return (retval); bailout_badarg: warnx("%s: only one of -l, -o or -w/-W may be specified", __func__); return (1); } static int cctl_do_io(int fd, int retries, union ctl_io *io, const char *func) { do { if (ioctl(fd, CTL_IO, io) == -1) { warn("%s: error sending CTL_IO ioctl", func); return (-1); } } while (((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) && (retries-- > 0)); return (0); } static int cctl_delay(int fd, int lun, int argc, char **argv, char *combinedopt) { struct ctl_io_delay_info delay_info; char *delayloc = NULL; char *delaytype = NULL; int delaytime = -1; int retval; int c; retval = 0; memset(&delay_info, 0, sizeof(delay_info)); while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'T': delaytype = strdup(optarg); break; case 'l': delayloc = strdup(optarg); break; case 't': delaytime = strtoul(optarg, NULL, 0); break; } } if (delaytime == -1) { warnx("%s: you must specify the delaytime with -t", __func__); retval = 1; goto bailout; } if (strcasecmp(delayloc, "datamove") == 0) delay_info.delay_loc = CTL_DELAY_LOC_DATAMOVE; else if (strcasecmp(delayloc, "done") == 0) delay_info.delay_loc = CTL_DELAY_LOC_DONE; else { warnx("%s: invalid delay location %s", __func__, delayloc); retval = 1; goto bailout; } if ((delaytype == NULL) || (strcmp(delaytype, "oneshot") == 0)) delay_info.delay_type = CTL_DELAY_TYPE_ONESHOT; else if (strcmp(delaytype, "cont") == 0) delay_info.delay_type = CTL_DELAY_TYPE_CONT; else { warnx("%s: invalid delay type %s", __func__, delaytype); retval = 1; goto bailout; } delay_info.lun_id = lun; delay_info.delay_secs = delaytime; if (ioctl(fd, CTL_DELAY_IO, &delay_info) == -1) { warn("%s: CTL_DELAY_IO ioctl failed", __func__); retval = 1; goto bailout; } switch (delay_info.status) { case CTL_DELAY_STATUS_NONE: warnx("%s: no delay status??", __func__); retval = 1; break; case CTL_DELAY_STATUS_OK: break; case CTL_DELAY_STATUS_INVALID_LUN: warnx("%s: invalid lun %d", __func__, lun); retval = 1; break; case CTL_DELAY_STATUS_INVALID_TYPE: warnx("%s: invalid delay type %d", __func__, delay_info.delay_type); retval = 1; break; case CTL_DELAY_STATUS_INVALID_LOC: warnx("%s: delay location %s not implemented?", __func__, delayloc); retval = 1; break; case CTL_DELAY_STATUS_NOT_IMPLEMENTED: warnx("%s: delay not implemented in the kernel", __func__); warnx("%s: recompile with the CTL_IO_DELAY flag set", __func__); retval = 1; break; default: warnx("%s: unknown delay return status %d", __func__, delay_info.status); retval = 1; break; } bailout: free(delayloc); free(delaytype); return (retval); } static struct ctladm_opts cctl_err_types[] = { {"aborted", CTL_LUN_INJ_ABORTED, CTLADM_ARG_NONE, NULL}, {"mediumerr", CTL_LUN_INJ_MEDIUM_ERR, CTLADM_ARG_NONE, NULL}, {"ua", CTL_LUN_INJ_UA, CTLADM_ARG_NONE, NULL}, {"custom", CTL_LUN_INJ_CUSTOM, CTLADM_ARG_NONE, NULL}, {NULL, 0, 0, NULL} }; static struct ctladm_opts cctl_err_patterns[] = { {"read", CTL_LUN_PAT_READ, CTLADM_ARG_NONE, NULL}, {"write", CTL_LUN_PAT_WRITE, CTLADM_ARG_NONE, NULL}, {"rw", CTL_LUN_PAT_READWRITE, CTLADM_ARG_NONE, NULL}, {"readwrite", CTL_LUN_PAT_READWRITE, CTLADM_ARG_NONE, NULL}, {"readcap", CTL_LUN_PAT_READCAP, CTLADM_ARG_NONE, NULL}, {"tur", CTL_LUN_PAT_TUR, CTLADM_ARG_NONE, NULL}, {"any", CTL_LUN_PAT_ANY, CTLADM_ARG_NONE, NULL}, #if 0 {"cmd", CTL_LUN_PAT_CMD, CTLADM_ARG_NONE, NULL}, #endif {NULL, 0, 0, NULL} }; static int cctl_error_inject(int fd, uint32_t lun, int argc, char **argv, char *combinedopt) { int retval = 0; struct ctl_error_desc err_desc; uint64_t lba = 0; uint32_t len = 0; uint64_t delete_id = 0; int delete_id_set = 0; int continuous = 0; int sense_len = 0; int fd_sense = 0; int c; bzero(&err_desc, sizeof(err_desc)); err_desc.lun_id = lun; while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'c': continuous = 1; break; case 'd': delete_id = strtoull(optarg, NULL, 0); delete_id_set = 1; break; case 'i': case 'p': { ctladm_optret optret; ctladm_cmdargs argnum; const char *subopt; if (c == 'i') { ctl_lun_error err_type; if (err_desc.lun_error != CTL_LUN_INJ_NONE) { warnx("%s: can't specify multiple -i " "arguments", __func__); retval = 1; goto bailout; } optret = getoption(cctl_err_types, optarg, &err_type, &argnum, &subopt); err_desc.lun_error = err_type; } else { ctl_lun_error_pattern pattern; optret = getoption(cctl_err_patterns, optarg, &pattern, &argnum, &subopt); err_desc.error_pattern |= pattern; } if (optret == CC_OR_AMBIGUOUS) { warnx("%s: ambiguous argument %s", __func__, optarg); retval = 1; goto bailout; } else if (optret == CC_OR_NOT_FOUND) { warnx("%s: argument %s not found", __func__, optarg); retval = 1; goto bailout; } break; } case 'r': { char *tmpstr, *tmpstr2; tmpstr = strdup(optarg); if (tmpstr == NULL) { warn("%s: error duplicating string %s", __func__, optarg); retval = 1; goto bailout; } tmpstr2 = strsep(&tmpstr, ","); if (tmpstr2 == NULL) { warnx("%s: invalid -r argument %s", __func__, optarg); retval = 1; free(tmpstr); goto bailout; } lba = strtoull(tmpstr2, NULL, 0); tmpstr2 = strsep(&tmpstr, ","); if (tmpstr2 == NULL) { warnx("%s: no len argument for -r lba,len, got" " %s", __func__, optarg); retval = 1; free(tmpstr); goto bailout; } len = strtoul(tmpstr2, NULL, 0); free(tmpstr); break; } case 's': { struct get_hook hook; char *sensestr; sense_len = strtol(optarg, NULL, 0); if (sense_len <= 0) { warnx("invalid number of sense bytes %d", sense_len); retval = 1; goto bailout; } sense_len = MIN(sense_len, SSD_FULL_SIZE); hook.argc = argc - optind; hook.argv = argv + optind; hook.got = 0; sensestr = cget(&hook, NULL); if ((sensestr != NULL) && (sensestr[0] == '-')) { fd_sense = 1; } else { buff_encode_visit( (uint8_t *)&err_desc.custom_sense, sense_len, sensestr, iget, &hook); } optind += hook.got; break; } default: break; } } if (delete_id_set != 0) { err_desc.serial = delete_id; if (ioctl(fd, CTL_ERROR_INJECT_DELETE, &err_desc) == -1) { warn("%s: error issuing CTL_ERROR_INJECT_DELETE ioctl", __func__); retval = 1; } goto bailout; } if (err_desc.lun_error == CTL_LUN_INJ_NONE) { warnx("%s: error injection command (-i) needed", __func__); retval = 1; goto bailout; } else if ((err_desc.lun_error == CTL_LUN_INJ_CUSTOM) && (sense_len == 0)) { warnx("%s: custom error requires -s", __func__); retval = 1; goto bailout; } if (continuous != 0) err_desc.lun_error |= CTL_LUN_INJ_CONTINUOUS; /* * If fd_sense is set, we need to read the sense data the user * wants returned from stdin. */ if (fd_sense == 1) { ssize_t amt_read; int amt_to_read = sense_len; u_int8_t *buf_ptr = (uint8_t *)&err_desc.custom_sense; for (amt_read = 0; amt_to_read > 0; amt_read = read(STDIN_FILENO, buf_ptr, amt_to_read)) { if (amt_read == -1) { warn("error reading sense data from stdin"); retval = 1; goto bailout; } amt_to_read -= amt_read; buf_ptr += amt_read; } } if (err_desc.error_pattern == CTL_LUN_PAT_NONE) { warnx("%s: command pattern (-p) needed", __func__); retval = 1; goto bailout; } if (len != 0) { err_desc.error_pattern |= CTL_LUN_PAT_RANGE; /* * We could check here to see whether it's a read/write * command, but that will be pointless once we allow * custom patterns. At that point, the user could specify * a READ(6) CDB type, and we wouldn't have an easy way here * to verify whether range checking is possible there. The * user will just figure it out when his error never gets * executed. */ #if 0 if ((err_desc.pattern & CTL_LUN_PAT_READWRITE) == 0) { warnx("%s: need read and/or write pattern if range " "is specified", __func__); retval = 1; goto bailout; } #endif err_desc.lba_range.lba = lba; err_desc.lba_range.len = len; } if (ioctl(fd, CTL_ERROR_INJECT, &err_desc) == -1) { warn("%s: error issuing CTL_ERROR_INJECT ioctl", __func__); retval = 1; } else { printf("Error injection succeeded, serial number is %ju\n", (uintmax_t)err_desc.serial); } bailout: return (retval); } static int cctl_lunlist(int fd) { struct scsi_report_luns_data *lun_data; struct scsi_inquiry_data *inq_data; uint32_t num_luns; int initid; unsigned int i; int retval; inq_data = NULL; initid = 7; /* * XXX KDM assuming LUN 0 is fine, but we may need to change this * if we ever acquire the ability to have multiple targets. */ if ((retval = cctl_get_luns(fd, /*lun*/ 0, initid, /*retries*/ 2, &lun_data, &num_luns)) != 0) goto bailout; inq_data = malloc(sizeof(*inq_data)); if (inq_data == NULL) { warn("%s: couldn't allocate memory for inquiry data\n", __func__); retval = 1; goto bailout; } for (i = 0; i < num_luns; i++) { char scsi_path[40]; int lun_val; switch (lun_data->luns[i].lundata[0] & RPL_LUNDATA_ATYP_MASK) { case RPL_LUNDATA_ATYP_PERIPH: lun_val = lun_data->luns[i].lundata[1]; break; case RPL_LUNDATA_ATYP_FLAT: lun_val = (lun_data->luns[i].lundata[0] & RPL_LUNDATA_FLAT_LUN_MASK) | (lun_data->luns[i].lundata[1] << RPL_LUNDATA_FLAT_LUN_BITS); break; case RPL_LUNDATA_ATYP_LUN: case RPL_LUNDATA_ATYP_EXTLUN: default: fprintf(stdout, "Unsupported LUN format %d\n", lun_data->luns[i].lundata[0] & RPL_LUNDATA_ATYP_MASK); lun_val = -1; break; } if (lun_val == -1) continue; if ((retval = cctl_get_inquiry(fd, lun_val, initid, /*retries*/ 2, scsi_path, sizeof(scsi_path), inq_data)) != 0) { goto bailout; } printf("%s", scsi_path); scsi_print_inquiry(inq_data); } bailout: if (lun_data != NULL) free(lun_data); if (inq_data != NULL) free(inq_data); return (retval); } static int cctl_sync_cache(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt) { union ctl_io *io; int cdb_size = -1; int retval; uint64_t our_lba = 0; uint32_t our_block_count = 0; int reladr = 0, immed = 0; int c; retval = 0; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warnx("%s: can't allocate memory", __func__); return (1); } while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'b': our_block_count = strtoul(optarg, NULL, 0); break; case 'c': cdb_size = strtol(optarg, NULL, 0); break; case 'i': immed = 1; break; case 'l': our_lba = strtoull(optarg, NULL, 0); break; case 'r': reladr = 1; break; default: break; } } if (cdb_size != -1) { switch (cdb_size) { case 10: case 16: break; default: warnx("%s: invalid cdbsize %d, valid sizes are 10 " "and 16", __func__, cdb_size); retval = 1; goto bailout; break; /* NOTREACHED */ } } else cdb_size = 10; ctl_scsi_sync_cache(/*io*/ io, /*immed*/ immed, /*reladr*/ reladr, /*minimum_cdb_size*/ cdb_size, /*starting_lba*/ our_lba, /*block_count*/ our_block_count, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { fprintf(stdout, "Cache synchronized successfully\n"); } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); return (retval); } static int cctl_start_stop(int fd, int lun, int iid, int retries, int start, int argc, char **argv, char *combinedopt) { union ctl_io *io; char scsi_path[40]; int immed = 0; int retval, c; retval = 0; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warnx("%s: can't allocate memory", __func__); return (1); } while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'i': immed = 1; break; default: break; } } /* * Use an ordered tag for the stop command, to guarantee that any * pending I/O will finish before the stop command executes. This * would normally be the case anyway, since CTL will basically * treat the start/stop command as an ordered command with respect * to any other command except an INQUIRY. (See ctl_ser_table.c.) */ ctl_scsi_start_stop(/*io*/ io, /*start*/ start, /*load_eject*/ 0, /*immediate*/ immed, /*power_conditions*/ SSS_PC_START_VALID, /*ctl_tag_type*/ start ? CTL_TAG_SIMPLE : CTL_TAG_ORDERED, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } ctl_scsi_path_string(io, scsi_path, sizeof(scsi_path)); if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { fprintf(stdout, "%s LUN %s successfully\n", scsi_path, (start) ? "started" : "stopped"); } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); return (retval); } static int cctl_mode_sense(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt) { union ctl_io *io; uint32_t datalen; uint8_t *dataptr; int pc = -1, cdbsize, retval, dbd = 0, subpage = -1; int list = 0; int page_code = -1; int c; cdbsize = 0; retval = 0; dataptr = NULL; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warn("%s: can't allocate memory", __func__); return (1); } while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'P': pc = strtoul(optarg, NULL, 0); break; case 'S': subpage = strtoul(optarg, NULL, 0); break; case 'd': dbd = 1; break; case 'l': list = 1; break; case 'm': page_code = strtoul(optarg, NULL, 0); break; case 'c': cdbsize = strtol(optarg, NULL, 0); break; default: break; } } if (((list == 0) && (page_code == -1)) || ((list != 0) && (page_code != -1))) { warnx("%s: you must specify either a page code (-m) or -l", __func__); retval = 1; goto bailout; } if ((page_code != -1) && ((page_code > SMS_ALL_PAGES_PAGE) || (page_code < 0))) { warnx("%s: page code %d is out of range", __func__, page_code); retval = 1; goto bailout; } if (list == 1) { page_code = SMS_ALL_PAGES_PAGE; if (pc != -1) { warnx("%s: arg -P makes no sense with -l", __func__); retval = 1; goto bailout; } if (subpage != -1) { warnx("%s: arg -S makes no sense with -l", __func__); retval = 1; goto bailout; } } if (pc == -1) pc = SMS_PAGE_CTRL_CURRENT; else { if ((pc > 3) || (pc < 0)) { warnx("%s: page control value %d is out of range: 0-3", __func__, pc); retval = 1; goto bailout; } } if ((subpage != -1) && ((subpage > 255) || (subpage < 0))) { warnx("%s: subpage code %d is out of range: 0-255", __func__, subpage); retval = 1; goto bailout; } if (cdbsize != 0) { switch (cdbsize) { case 6: case 10: break; default: warnx("%s: invalid cdbsize %d, valid sizes are 6 " "and 10", __func__, cdbsize); retval = 1; goto bailout; break; } } else cdbsize = 6; if (subpage == -1) subpage = 0; if (cdbsize == 6) datalen = 255; else datalen = 65535; dataptr = (uint8_t *)malloc(datalen); if (dataptr == NULL) { warn("%s: can't allocate %d bytes", __func__, datalen); retval = 1; goto bailout; } memset(dataptr, 0, datalen); ctl_scsi_mode_sense(io, /*data_ptr*/ dataptr, /*data_len*/ datalen, /*dbd*/ dbd, /*llbaa*/ 0, /*page_code*/ page_code, /*pc*/ pc << 6, /*subpage*/ subpage, /*minimum_cdb_size*/ cdbsize, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { int pages_len, used_len; uint32_t returned_len; uint8_t *ndataptr; if (io->scsiio.cdb[0] == MODE_SENSE_6) { struct scsi_mode_hdr_6 *hdr6; int bdlen; hdr6 = (struct scsi_mode_hdr_6 *)dataptr; returned_len = hdr6->datalen + 1; bdlen = hdr6->block_descr_len; ndataptr = (uint8_t *)((uint8_t *)&hdr6[1] + bdlen); } else { struct scsi_mode_hdr_10 *hdr10; int bdlen; hdr10 = (struct scsi_mode_hdr_10 *)dataptr; returned_len = scsi_2btoul(hdr10->datalen) + 2; bdlen = scsi_2btoul(hdr10->block_descr_len); ndataptr = (uint8_t *)((uint8_t *)&hdr10[1] + bdlen); } /* just in case they can give us more than we allocated for */ returned_len = min(returned_len, datalen); pages_len = returned_len - (ndataptr - dataptr); #if 0 fprintf(stdout, "returned_len = %d, pages_len = %d\n", returned_len, pages_len); #endif if (list == 1) { fprintf(stdout, "Supported mode pages:\n"); for (used_len = 0; used_len < pages_len;) { struct scsi_mode_page_header *header; header = (struct scsi_mode_page_header *) &ndataptr[used_len]; fprintf(stdout, "%d\n", header->page_code); used_len += header->page_length + 2; } } else { for (used_len = 0; used_len < pages_len; used_len++) { fprintf(stdout, "0x%x ", ndataptr[used_len]); if (((used_len+1) % 16) == 0) fprintf(stdout, "\n"); } fprintf(stdout, "\n"); } } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); if (dataptr != NULL) free(dataptr); return (retval); } static int cctl_read_capacity(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt) { union ctl_io *io; struct scsi_read_capacity_data *data; struct scsi_read_capacity_data_long *longdata; int cdbsize = -1, retval; uint8_t *dataptr; int c; cdbsize = 10; dataptr = NULL; retval = 0; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warn("%s: can't allocate memory\n", __func__); return (1); } while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'c': cdbsize = strtol(optarg, NULL, 0); break; default: break; } } if (cdbsize != -1) { switch (cdbsize) { case 10: case 16: break; default: warnx("%s: invalid cdbsize %d, valid sizes are 10 " "and 16", __func__, cdbsize); retval = 1; goto bailout; break; /* NOTREACHED */ } } else cdbsize = 10; dataptr = (uint8_t *)malloc(sizeof(*longdata)); if (dataptr == NULL) { warn("%s: can't allocate %zd bytes\n", __func__, sizeof(*longdata)); retval = 1; goto bailout; } memset(dataptr, 0, sizeof(*longdata)); retry: switch (cdbsize) { case 10: ctl_scsi_read_capacity(io, /*data_ptr*/ dataptr, /*data_len*/ sizeof(*longdata), /*addr*/ 0, /*reladr*/ 0, /*pmi*/ 0, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); break; case 16: ctl_scsi_read_capacity_16(io, /*data_ptr*/ dataptr, /*data_len*/ sizeof(*longdata), /*addr*/ 0, /*reladr*/ 0, /*pmi*/ 0, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); break; } io->io_hdr.nexus.initid = iid; io->io_hdr.nexus.targ_lun = lun; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { uint64_t maxlba; uint32_t blocksize; if (cdbsize == 10) { data = (struct scsi_read_capacity_data *)dataptr; maxlba = scsi_4btoul(data->addr); blocksize = scsi_4btoul(data->length); if (maxlba == 0xffffffff) { cdbsize = 16; goto retry; } } else { longdata=(struct scsi_read_capacity_data_long *)dataptr; maxlba = scsi_8btou64(longdata->addr); blocksize = scsi_4btoul(longdata->length); } fprintf(stdout, "Disk Capacity: %ju, Blocksize: %d\n", (uintmax_t)maxlba, blocksize); } else { ctl_io_error_print(io, NULL, stderr); } bailout: ctl_scsi_free_io(io); if (dataptr != NULL) free(dataptr); return (retval); } static int cctl_read_write(int fd, int lun, int iid, int retries, int argc, char **argv, char *combinedopt, ctladm_cmdfunction command) { union ctl_io *io; int file_fd, do_stdio; int cdbsize = -1, databytes; uint8_t *dataptr; char *filename = NULL; int datalen = -1, blocksize = -1; uint64_t lba = 0; int lba_set = 0; int retval; int c; retval = 0; do_stdio = 0; dataptr = NULL; file_fd = -1; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warn("%s: can't allocate memory\n", __func__); return (1); } while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'N': io->io_hdr.flags |= CTL_FLAG_NO_DATAMOVE; break; case 'b': blocksize = strtoul(optarg, NULL, 0); break; case 'c': cdbsize = strtoul(optarg, NULL, 0); break; case 'd': datalen = strtoul(optarg, NULL, 0); break; case 'f': filename = strdup(optarg); break; case 'l': lba = strtoull(optarg, NULL, 0); lba_set = 1; break; default: break; } } if (filename == NULL) { warnx("%s: you must supply a filename using -f", __func__); retval = 1; goto bailout; } if (datalen == -1) { warnx("%s: you must specify the data length with -d", __func__); retval = 1; goto bailout; } if (lba_set == 0) { warnx("%s: you must specify the LBA with -l", __func__); retval = 1; goto bailout; } if (blocksize == -1) { warnx("%s: you must specify the blocksize with -b", __func__); retval = 1; goto bailout; } if (cdbsize != -1) { switch (cdbsize) { case 6: case 10: case 12: case 16: break; default: warnx("%s: invalid cdbsize %d, valid sizes are 6, " "10, 12 or 16", __func__, cdbsize); retval = 1; goto bailout; break; /* NOTREACHED */ } } else cdbsize = 6; databytes = datalen * blocksize; dataptr = (uint8_t *)malloc(databytes); if (dataptr == NULL) { warn("%s: can't allocate %d bytes\n", __func__, databytes); retval = 1; goto bailout; } if (strcmp(filename, "-") == 0) { if (command == CTLADM_CMD_READ) file_fd = STDOUT_FILENO; else file_fd = STDIN_FILENO; do_stdio = 1; } else { file_fd = open(filename, O_RDWR | O_CREAT, S_IRUSR | S_IWUSR); if (file_fd == -1) { warn("%s: can't open file %s", __func__, filename); retval = 1; goto bailout; } } memset(dataptr, 0, databytes); if (command == CTLADM_CMD_WRITE) { int bytes_read; bytes_read = read(file_fd, dataptr, databytes); if (bytes_read == -1) { warn("%s: error reading file %s", __func__, filename); retval = 1; goto bailout; } if (bytes_read != databytes) { warnx("%s: only read %d bytes from file %s", __func__, bytes_read, filename); retval = 1; goto bailout; } } ctl_scsi_read_write(io, /*data_ptr*/ dataptr, /*data_len*/ databytes, /*read_op*/ (command == CTLADM_CMD_READ) ? 1 : 0, /*byte2*/ 0, /*minimum_cdb_size*/ cdbsize, /*lba*/ lba, /*num_blocks*/ datalen, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } if (((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) && (command == CTLADM_CMD_READ)) { int bytes_written; bytes_written = write(file_fd, dataptr, databytes); if (bytes_written == -1) { warn("%s: can't write to %s", __func__, filename); goto bailout; } } else if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); if (dataptr != NULL) free(dataptr); if ((do_stdio == 0) && (file_fd != -1)) close(file_fd); return (retval); } static int cctl_get_luns(int fd, int lun, int iid, int retries, struct scsi_report_luns_data **lun_data, uint32_t *num_luns) { union ctl_io *io; uint32_t nluns; int lun_datalen; int retval; retval = 0; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warnx("%s: can't allocate memory", __func__); return (1); } /* * lun_data includes space for 1 lun, allocate space for 4 initially. * If that isn't enough, we'll allocate more. */ nluns = 4; retry: lun_datalen = sizeof(*lun_data) + (nluns * sizeof(struct scsi_report_luns_lundata)); *lun_data = malloc(lun_datalen); if (*lun_data == NULL) { warnx("%s: can't allocate memory", __func__); ctl_scsi_free_io(io); return (1); } ctl_scsi_report_luns(io, /*data_ptr*/ (uint8_t *)*lun_data, /*data_len*/ lun_datalen, /*select_report*/ RPL_REPORT_ALL, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.initid = iid; io->io_hdr.nexus.targ_lun = lun; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { uint32_t returned_len, returned_luns; returned_len = scsi_4btoul((*lun_data)->length); returned_luns = returned_len / 8; if (returned_luns > nluns) { nluns = returned_luns; free(*lun_data); goto retry; } /* These should be the same */ *num_luns = MIN(returned_luns, nluns); } else { ctl_io_error_print(io, NULL, stderr); retval = 1; } bailout: ctl_scsi_free_io(io); return (retval); } static int cctl_report_luns(int fd, int lun, int iid, int retries) { struct scsi_report_luns_data *lun_data; uint32_t num_luns, i; int retval; lun_data = NULL; if ((retval = cctl_get_luns(fd, lun, iid, retries, &lun_data, &num_luns)) != 0) goto bailout; fprintf(stdout, "%u LUNs returned\n", num_luns); for (i = 0; i < num_luns; i++) { int lun_val; /* * XXX KDM figure out a way to share this code with * cctl_lunlist()? */ switch (lun_data->luns[i].lundata[0] & RPL_LUNDATA_ATYP_MASK) { case RPL_LUNDATA_ATYP_PERIPH: lun_val = lun_data->luns[i].lundata[1]; break; case RPL_LUNDATA_ATYP_FLAT: lun_val = (lun_data->luns[i].lundata[0] & RPL_LUNDATA_FLAT_LUN_MASK) | (lun_data->luns[i].lundata[1] << RPL_LUNDATA_FLAT_LUN_BITS); break; case RPL_LUNDATA_ATYP_LUN: case RPL_LUNDATA_ATYP_EXTLUN: default: fprintf(stdout, "Unsupported LUN format %d\n", lun_data->luns[i].lundata[0] & RPL_LUNDATA_ATYP_MASK); lun_val = -1; break; } if (lun_val == -1) continue; fprintf(stdout, "%d\n", lun_val); } bailout: if (lun_data != NULL) free(lun_data); return (retval); } static int cctl_tur(int fd, int lun, int iid, int retries) { union ctl_io *io; io = ctl_scsi_alloc_io(iid); if (io == NULL) { fprintf(stderr, "can't allocate memory\n"); return (1); } ctl_scsi_tur(io, /* tag_type */ CTL_TAG_SIMPLE, /* control */ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retries, io, __func__) != 0) { ctl_scsi_free_io(io); return (1); } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) fprintf(stdout, "Unit is ready\n"); else ctl_io_error_print(io, NULL, stderr); return (0); } static int cctl_get_inquiry(int fd, int lun, int iid, int retries, char *path_str, int path_len, struct scsi_inquiry_data *inq_data) { union ctl_io *io; int retval; retval = 0; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warnx("cctl_inquiry: can't allocate memory\n"); return (1); } ctl_scsi_inquiry(/*io*/ io, /*data_ptr*/ (uint8_t *)inq_data, /*data_len*/ sizeof(*inq_data), /*byte2*/ 0, /*page_code*/ 0, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) { retval = 1; ctl_io_error_print(io, NULL, stderr); } else if (path_str != NULL) ctl_scsi_path_string(io, path_str, path_len); bailout: ctl_scsi_free_io(io); return (retval); } static int cctl_inquiry(int fd, int lun, int iid, int retries) { struct scsi_inquiry_data *inq_data; char scsi_path[40]; int retval; inq_data = malloc(sizeof(*inq_data)); if (inq_data == NULL) { warnx("%s: can't allocate inquiry data", __func__); retval = 1; goto bailout; } if ((retval = cctl_get_inquiry(fd, lun, iid, retries, scsi_path, sizeof(scsi_path), inq_data)) != 0) goto bailout; printf("%s", scsi_path); scsi_print_inquiry(inq_data); bailout: if (inq_data != NULL) free(inq_data); return (retval); } static int cctl_req_sense(int fd, int lun, int iid, int retries) { union ctl_io *io; struct scsi_sense_data *sense_data; int retval; retval = 0; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warnx("cctl_req_sense: can't allocate memory\n"); return (1); } sense_data = malloc(sizeof(*sense_data)); memset(sense_data, 0, sizeof(*sense_data)); ctl_scsi_request_sense(/*io*/ io, /*data_ptr*/ (uint8_t *)sense_data, /*data_len*/ sizeof(*sense_data), /*byte2*/ 0, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retries, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { bcopy(sense_data, &io->scsiio.sense_data, sizeof(*sense_data)); io->scsiio.sense_len = sizeof(*sense_data); ctl_scsi_sense_print(&io->scsiio, NULL, stdout); } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); free(sense_data); return (retval); } static int cctl_report_target_port_group(int fd, int lun, int iid) { union ctl_io *io; uint32_t datalen; uint8_t *dataptr; int retval; dataptr = NULL; retval = 0; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warn("%s: can't allocate memory", __func__); return (1); } datalen = 64; dataptr = (uint8_t *)malloc(datalen); if (dataptr == NULL) { warn("%s: can't allocate %d bytes", __func__, datalen); retval = 1; goto bailout; } memset(dataptr, 0, datalen); ctl_scsi_maintenance_in(/*io*/ io, /*data_ptr*/ dataptr, /*data_len*/ datalen, /*action*/ SA_RPRT_TRGT_GRP, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, 0, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { int returned_len, used_len; returned_len = scsi_4btoul(&dataptr[0]) + 4; for (used_len = 0; used_len < returned_len; used_len++) { fprintf(stdout, "0x%02x ", dataptr[used_len]); if (((used_len+1) % 8) == 0) fprintf(stdout, "\n"); } fprintf(stdout, "\n"); } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); if (dataptr != NULL) free(dataptr); return (retval); } static int cctl_inquiry_vpd_devid(int fd, int lun, int iid) { union ctl_io *io; uint32_t datalen; uint8_t *dataptr; int retval; retval = 0; dataptr = NULL; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warn("%s: can't allocate memory", __func__); return (1); } datalen = 256; dataptr = (uint8_t *)malloc(datalen); if (dataptr == NULL) { warn("%s: can't allocate %d bytes", __func__, datalen); retval = 1; goto bailout; } memset(dataptr, 0, datalen); ctl_scsi_inquiry(/*io*/ io, /*data_ptr*/ dataptr, /*data_len*/ datalen, /*byte2*/ SI_EVPD, /*page_code*/ SVPD_DEVICE_ID, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, 0, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { int returned_len, used_len; returned_len = scsi_2btoul(&dataptr[2]) + 4; for (used_len = 0; used_len < returned_len; used_len++) { fprintf(stdout, "0x%02x ", dataptr[used_len]); if (((used_len+1) % 8) == 0) fprintf(stdout, "\n"); } fprintf(stdout, "\n"); } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); if (dataptr != NULL) free(dataptr); return (retval); } static int cctl_persistent_reserve_in(int fd, int lun, int iid, int argc, char **argv, char *combinedopt, int retry_count) { union ctl_io *io; uint32_t datalen; uint8_t *dataptr; int action = -1; int retval; int c; retval = 0; dataptr = NULL; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warn("%s: can't allocate memory", __func__); return (1); } while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'a': action = strtol(optarg, NULL, 0); break; default: break; } } if (action < 0 || action > 2) { warn("action must be specified and in the range: 0-2"); retval = 1; goto bailout; } datalen = 256; dataptr = (uint8_t *)malloc(datalen); if (dataptr == NULL) { warn("%s: can't allocate %d bytes", __func__, datalen); retval = 1; goto bailout; } memset(dataptr, 0, datalen); ctl_scsi_persistent_res_in(io, /*data_ptr*/ dataptr, /*data_len*/ datalen, /*action*/ action, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retry_count, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { int returned_len, used_len; switch (action) { case 0: returned_len = scsi_4btoul(&dataptr[4]) + 8; returned_len = min(returned_len, 256); break; case 1: returned_len = scsi_4btoul(&dataptr[4]) + 8; break; case 2: returned_len = 8; break; default: warnx("%s: invalid action %d", __func__, action); goto bailout; break; /* NOTREACHED */ } for (used_len = 0; used_len < returned_len; used_len++) { fprintf(stdout, "0x%02x ", dataptr[used_len]); if (((used_len+1) % 8) == 0) fprintf(stdout, "\n"); } fprintf(stdout, "\n"); } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); if (dataptr != NULL) free(dataptr); return (retval); } static int cctl_persistent_reserve_out(int fd, int lun, int iid, int argc, char **argv, char *combinedopt, int retry_count) { union ctl_io *io; uint32_t datalen; uint64_t key = 0, sa_key = 0; int action = -1, restype = -1; uint8_t *dataptr; int retval; int c; retval = 0; dataptr = NULL; io = ctl_scsi_alloc_io(iid); if (io == NULL) { warn("%s: can't allocate memory", __func__); return (1); } while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'a': action = strtol(optarg, NULL, 0); break; case 'k': key = strtoull(optarg, NULL, 0); break; case 'r': restype = strtol(optarg, NULL, 0); break; case 's': sa_key = strtoull(optarg, NULL, 0); break; default: break; } } if (action < 0 || action > 5) { warn("action must be specified and in the range: 0-5"); retval = 1; goto bailout; } if (restype < 0 || restype > 5) { if (action != 0 && action != 5 && action != 3) { warn("'restype' must specified and in the range: 0-5"); retval = 1; goto bailout; } } datalen = 24; dataptr = (uint8_t *)malloc(datalen); if (dataptr == NULL) { warn("%s: can't allocate %d bytes", __func__, datalen); retval = 1; goto bailout; } memset(dataptr, 0, datalen); ctl_scsi_persistent_res_out(io, /*data_ptr*/ dataptr, /*data_len*/ datalen, /*action*/ action, /*type*/ restype, /*key*/ key, /*sa key*/ sa_key, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); io->io_hdr.nexus.targ_lun = lun; io->io_hdr.nexus.initid = iid; if (cctl_do_io(fd, retry_count, io, __func__) != 0) { retval = 1; goto bailout; } if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) { char scsi_path[40]; ctl_scsi_path_string(io, scsi_path, sizeof(scsi_path)); fprintf( stdout, "%sPERSISTENT RESERVE OUT executed " "successfully\n", scsi_path); } else ctl_io_error_print(io, NULL, stderr); bailout: ctl_scsi_free_io(io); if (dataptr != NULL) free(dataptr); return (retval); } -struct cctl_req_option { - char *name; - int namelen; - char *value; - int vallen; - STAILQ_ENTRY(cctl_req_option) links; -}; - static int cctl_create_lun(int fd, int argc, char **argv, char *combinedopt) { struct ctl_lun_req req; int device_type = -1; uint64_t lun_size = 0; uint32_t blocksize = 0, req_lun_id = 0; char *serial_num = NULL; char *device_id = NULL; int lun_size_set = 0, blocksize_set = 0, lun_id_set = 0; char *backend_name = NULL; - STAILQ_HEAD(, cctl_req_option) option_list; - int num_options = 0; + nvlist_t *option_list; int retval = 0, c; - STAILQ_INIT(&option_list); + option_list = nvlist_create(0); + if (option_list == NULL) + err(1, "%s: unable to allocate nvlist", __func__); while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'b': backend_name = strdup(optarg); break; case 'B': blocksize = strtoul(optarg, NULL, 0); blocksize_set = 1; break; case 'd': device_id = strdup(optarg); break; case 'l': req_lun_id = strtoul(optarg, NULL, 0); lun_id_set = 1; break; case 'o': { - struct cctl_req_option *option; char *tmpstr; char *name, *value; tmpstr = strdup(optarg); name = strsep(&tmpstr, "="); if (name == NULL) { warnx("%s: option -o takes \"name=value\"" "argument", __func__); retval = 1; goto bailout; } value = strsep(&tmpstr, "="); if (value == NULL) { warnx("%s: option -o takes \"name=value\"" "argument", __func__); retval = 1; goto bailout; } - option = malloc(sizeof(*option)); - if (option == NULL) { - warn("%s: error allocating %zd bytes", - __func__, sizeof(*option)); - retval = 1; - goto bailout; - } - option->name = strdup(name); - option->namelen = strlen(name) + 1; - option->value = strdup(value); - option->vallen = strlen(value) + 1; free(tmpstr); - - STAILQ_INSERT_TAIL(&option_list, option, links); - num_options++; + nvlist_add_string(option_list, name, value); break; } case 's': if (strcasecmp(optarg, "auto") != 0) { retval = expand_number(optarg, &lun_size); if (retval != 0) { warn("%s: invalid -s argument", __func__); retval = 1; goto bailout; } } lun_size_set = 1; break; case 'S': serial_num = strdup(optarg); break; case 't': device_type = strtoul(optarg, NULL, 0); break; default: break; } } if (backend_name == NULL) { warnx("%s: backend name (-b) must be specified", __func__); retval = 1; goto bailout; } bzero(&req, sizeof(req)); strlcpy(req.backend, backend_name, sizeof(req.backend)); req.reqtype = CTL_LUNREQ_CREATE; if (blocksize_set != 0) req.reqdata.create.blocksize_bytes = blocksize; if (lun_size_set != 0) req.reqdata.create.lun_size_bytes = lun_size; if (lun_id_set != 0) { req.reqdata.create.flags |= CTL_LUN_FLAG_ID_REQ; req.reqdata.create.req_lun_id = req_lun_id; } req.reqdata.create.flags |= CTL_LUN_FLAG_DEV_TYPE; if (device_type != -1) req.reqdata.create.device_type = device_type; else req.reqdata.create.device_type = T_DIRECT; if (serial_num != NULL) { strlcpy(req.reqdata.create.serial_num, serial_num, sizeof(req.reqdata.create.serial_num)); req.reqdata.create.flags |= CTL_LUN_FLAG_SERIAL_NUM; } if (device_id != NULL) { strlcpy(req.reqdata.create.device_id, device_id, sizeof(req.reqdata.create.device_id)); req.reqdata.create.flags |= CTL_LUN_FLAG_DEVID; } - req.num_be_args = num_options; - if (num_options > 0) { - struct cctl_req_option *option, *next_option; - int i; - - req.be_args = malloc(num_options * sizeof(*req.be_args)); - if (req.be_args == NULL) { - warn("%s: error allocating %zd bytes", __func__, - num_options * sizeof(*req.be_args)); - retval = 1; - goto bailout; - } - - for (i = 0, option = STAILQ_FIRST(&option_list); - i < num_options; i++, option = next_option) { - next_option = STAILQ_NEXT(option, links); - - req.be_args[i].namelen = option->namelen; - req.be_args[i].name = strdup(option->name); - req.be_args[i].vallen = option->vallen; - req.be_args[i].value = strdup(option->value); - /* - * XXX KDM do we want a way to specify a writeable - * flag of some sort? Do we want a way to specify - * binary data? - */ - req.be_args[i].flags = CTL_BEARG_ASCII | CTL_BEARG_RD; - - STAILQ_REMOVE(&option_list, option, cctl_req_option, - links); - free(option->name); - free(option->value); - free(option); - } + req.args = nvlist_pack(option_list, &req.args_len); + if (req.args == NULL) { + warn("%s: error packing nvlist", __func__); + retval = 1; + goto bailout; } - if (ioctl(fd, CTL_LUN_REQ, &req) == -1) { + retval = ioctl(fd, CTL_LUN_REQ, &req); + free(req.args); + if (retval == -1) { warn("%s: error issuing CTL_LUN_REQ ioctl", __func__); retval = 1; goto bailout; } switch (req.status) { case CTL_LUN_ERROR: warnx("LUN creation error: %s", req.error_str); retval = 1; goto bailout; case CTL_LUN_WARNING: warnx("LUN creation warning: %s", req.error_str); break; case CTL_LUN_OK: break; default: warnx("unknown LUN creation status: %d", req.status); retval = 1; goto bailout; } fprintf(stdout, "LUN created successfully\n"); fprintf(stdout, "backend: %s\n", req.backend); fprintf(stdout, "device type: %d\n",req.reqdata.create.device_type); fprintf(stdout, "LUN size: %ju bytes\n", (uintmax_t)req.reqdata.create.lun_size_bytes); fprintf(stdout, "blocksize %u bytes\n", req.reqdata.create.blocksize_bytes); fprintf(stdout, "LUN ID: %d\n", req.reqdata.create.req_lun_id); fprintf(stdout, "Serial Number: %s\n", req.reqdata.create.serial_num); - fprintf(stdout, "Device ID; %s\n", req.reqdata.create.device_id); + fprintf(stdout, "Device ID: %s\n", req.reqdata.create.device_id); bailout: + nvlist_destroy(req.args_nvl); return (retval); } static int cctl_rm_lun(int fd, int argc, char **argv, char *combinedopt) { struct ctl_lun_req req; uint32_t lun_id = 0; int lun_id_set = 0; char *backend_name = NULL; - STAILQ_HEAD(, cctl_req_option) option_list; - int num_options = 0; + nvlist_t *option_list; int retval = 0, c; - STAILQ_INIT(&option_list); + option_list = nvlist_create(0); + if (option_list == NULL) + err(1, "%s: unable to allocate nvlist", __func__); while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'b': backend_name = strdup(optarg); break; case 'l': lun_id = strtoul(optarg, NULL, 0); lun_id_set = 1; break; case 'o': { - struct cctl_req_option *option; char *tmpstr; char *name, *value; tmpstr = strdup(optarg); name = strsep(&tmpstr, "="); if (name == NULL) { warnx("%s: option -o takes \"name=value\"" "argument", __func__); retval = 1; goto bailout; } value = strsep(&tmpstr, "="); if (value == NULL) { warnx("%s: option -o takes \"name=value\"" "argument", __func__); retval = 1; goto bailout; } - option = malloc(sizeof(*option)); - if (option == NULL) { - warn("%s: error allocating %zd bytes", - __func__, sizeof(*option)); - retval = 1; - goto bailout; - } - option->name = strdup(name); - option->namelen = strlen(name) + 1; - option->value = strdup(value); - option->vallen = strlen(value) + 1; free(tmpstr); - - STAILQ_INSERT_TAIL(&option_list, option, links); - num_options++; + nvlist_add_string(option_list, name, value); break; } default: break; } } if (backend_name == NULL) errx(1, "%s: backend name (-b) must be specified", __func__); if (lun_id_set == 0) errx(1, "%s: LUN id (-l) must be specified", __func__); bzero(&req, sizeof(req)); strlcpy(req.backend, backend_name, sizeof(req.backend)); req.reqtype = CTL_LUNREQ_RM; req.reqdata.rm.lun_id = lun_id; - - req.num_be_args = num_options; - if (num_options > 0) { - struct cctl_req_option *option, *next_option; - int i; - - req.be_args = malloc(num_options * sizeof(*req.be_args)); - if (req.be_args == NULL) { - warn("%s: error allocating %zd bytes", __func__, - num_options * sizeof(*req.be_args)); - retval = 1; - goto bailout; - } - - for (i = 0, option = STAILQ_FIRST(&option_list); - i < num_options; i++, option = next_option) { - next_option = STAILQ_NEXT(option, links); - - req.be_args[i].namelen = option->namelen; - req.be_args[i].name = strdup(option->name); - req.be_args[i].vallen = option->vallen; - req.be_args[i].value = strdup(option->value); - /* - * XXX KDM do we want a way to specify a writeable - * flag of some sort? Do we want a way to specify - * binary data? - */ - req.be_args[i].flags = CTL_BEARG_ASCII | CTL_BEARG_RD; - - STAILQ_REMOVE(&option_list, option, cctl_req_option, - links); - free(option->name); - free(option->value); - free(option); - } + + req.args = nvlist_pack(option_list, &req.args_len); + if (req.args == NULL) { + warn("%s: error packing nvlist", __func__); + retval = 1; + goto bailout; } - if (ioctl(fd, CTL_LUN_REQ, &req) == -1) { + retval = ioctl(fd, CTL_LUN_REQ, &req); + free(req.args); + if (retval == -1) { warn("%s: error issuing CTL_LUN_REQ ioctl", __func__); retval = 1; goto bailout; } switch (req.status) { case CTL_LUN_ERROR: warnx("LUN removal error: %s", req.error_str); retval = 1; goto bailout; case CTL_LUN_WARNING: warnx("LUN removal warning: %s", req.error_str); break; case CTL_LUN_OK: break; default: warnx("unknown LUN removal status: %d", req.status); retval = 1; goto bailout; } printf("LUN %d removed successfully\n", lun_id); bailout: + nvlist_destroy(req.args_nvl); return (retval); } static int cctl_modify_lun(int fd, int argc, char **argv, char *combinedopt) { struct ctl_lun_req req; uint64_t lun_size = 0; uint32_t lun_id = 0; int lun_id_set = 0, lun_size_set = 0; char *backend_name = NULL; - STAILQ_HEAD(, cctl_req_option) option_list; - int num_options = 0; + nvlist_t *option_list; int retval = 0, c; - STAILQ_INIT(&option_list); + option_list = nvlist_create(0); + if (option_list == NULL) + err(1, "%s: unable to allocate nvlist", __func__); + while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'b': backend_name = strdup(optarg); break; case 'l': lun_id = strtoul(optarg, NULL, 0); lun_id_set = 1; break; case 'o': { - struct cctl_req_option *option; char *tmpstr; char *name, *value; tmpstr = strdup(optarg); name = strsep(&tmpstr, "="); if (name == NULL) { warnx("%s: option -o takes \"name=value\"" "argument", __func__); retval = 1; goto bailout; } value = strsep(&tmpstr, "="); if (value == NULL) { warnx("%s: option -o takes \"name=value\"" "argument", __func__); retval = 1; goto bailout; } - option = malloc(sizeof(*option)); - if (option == NULL) { - warn("%s: error allocating %zd bytes", - __func__, sizeof(*option)); - retval = 1; - goto bailout; - } - option->name = strdup(name); - option->namelen = strlen(name) + 1; - option->value = strdup(value); - option->vallen = strlen(value) + 1; free(tmpstr); - - STAILQ_INSERT_TAIL(&option_list, option, links); - num_options++; + nvlist_add_string(option_list, name, value); break; } case 's': if (strcasecmp(optarg, "auto") != 0) { retval = expand_number(optarg, &lun_size); if (retval != 0) { warn("%s: invalid -s argument", __func__); retval = 1; goto bailout; } } lun_size_set = 1; break; default: break; } } if (backend_name == NULL) errx(1, "%s: backend name (-b) must be specified", __func__); if (lun_id_set == 0) errx(1, "%s: LUN id (-l) must be specified", __func__); - if (lun_size_set == 0 && num_options == 0) + if (lun_size_set == 0 && nvlist_empty(option_list)) errx(1, "%s: size (-s) or options (-o) must be specified", __func__); bzero(&req, sizeof(req)); strlcpy(req.backend, backend_name, sizeof(req.backend)); req.reqtype = CTL_LUNREQ_MODIFY; req.reqdata.modify.lun_id = lun_id; req.reqdata.modify.lun_size_bytes = lun_size; - req.num_be_args = num_options; - if (num_options > 0) { - struct cctl_req_option *option, *next_option; - int i; - - req.be_args = malloc(num_options * sizeof(*req.be_args)); - if (req.be_args == NULL) { - warn("%s: error allocating %zd bytes", __func__, - num_options * sizeof(*req.be_args)); - retval = 1; - goto bailout; - } - - for (i = 0, option = STAILQ_FIRST(&option_list); - i < num_options; i++, option = next_option) { - next_option = STAILQ_NEXT(option, links); - - req.be_args[i].namelen = option->namelen; - req.be_args[i].name = strdup(option->name); - req.be_args[i].vallen = option->vallen; - req.be_args[i].value = strdup(option->value); - /* - * XXX KDM do we want a way to specify a writeable - * flag of some sort? Do we want a way to specify - * binary data? - */ - req.be_args[i].flags = CTL_BEARG_ASCII | CTL_BEARG_RD; - - STAILQ_REMOVE(&option_list, option, cctl_req_option, - links); - free(option->name); - free(option->value); - free(option); - } + req.args = nvlist_pack(option_list, &req.args_len); + if (req.args == NULL) { + warn("%s: error packing nvlist", __func__); + retval = 1; + goto bailout; } - if (ioctl(fd, CTL_LUN_REQ, &req) == -1) { + retval = ioctl(fd, CTL_LUN_REQ, &req); + free(req.args); + if (retval == -1) { warn("%s: error issuing CTL_LUN_REQ ioctl", __func__); retval = 1; goto bailout; } switch (req.status) { case CTL_LUN_ERROR: warnx("LUN modification error: %s", req.error_str); retval = 1; goto bailout; case CTL_LUN_WARNING: warnx("LUN modification warning: %s", req.error_str); break; case CTL_LUN_OK: break; default: warnx("unknown LUN modification status: %d", req.status); retval = 1; goto bailout; } printf("LUN %d modified successfully\n", lun_id); bailout: + nvlist_destroy(req.args_nvl); return (retval); } struct cctl_islist_conn { int connection_id; char *initiator; char *initiator_addr; char *initiator_alias; char *target; char *target_alias; char *header_digest; char *data_digest; char *max_recv_data_segment_length; char *max_send_data_segment_length; char *max_burst_length; char *first_burst_length; char *offload; int immediate_data; int iser; STAILQ_ENTRY(cctl_islist_conn) links; }; struct cctl_islist_data { int num_conns; STAILQ_HEAD(,cctl_islist_conn) conn_list; struct cctl_islist_conn *cur_conn; int level; struct sbuf *cur_sb[32]; }; static void cctl_islist_start_element(void *user_data, const char *name, const char **attr) { int i; struct cctl_islist_data *islist; struct cctl_islist_conn *cur_conn; islist = (struct cctl_islist_data *)user_data; cur_conn = islist->cur_conn; islist->level++; if ((u_int)islist->level >= (sizeof(islist->cur_sb) / sizeof(islist->cur_sb[0]))) errx(1, "%s: too many nesting levels, %zd max", __func__, sizeof(islist->cur_sb) / sizeof(islist->cur_sb[0])); islist->cur_sb[islist->level] = sbuf_new_auto(); if (islist->cur_sb[islist->level] == NULL) err(1, "%s: Unable to allocate sbuf", __func__); if (strcmp(name, "connection") == 0) { if (cur_conn != NULL) errx(1, "%s: improper connection element nesting", __func__); cur_conn = calloc(1, sizeof(*cur_conn)); if (cur_conn == NULL) err(1, "%s: cannot allocate %zd bytes", __func__, sizeof(*cur_conn)); islist->num_conns++; islist->cur_conn = cur_conn; STAILQ_INSERT_TAIL(&islist->conn_list, cur_conn, links); for (i = 0; attr[i] != NULL; i += 2) { if (strcmp(attr[i], "id") == 0) { cur_conn->connection_id = strtoull(attr[i+1], NULL, 0); } else { errx(1, "%s: invalid connection attribute %s = %s", __func__, attr[i], attr[i+1]); } } } } static void cctl_islist_end_element(void *user_data, const char *name) { struct cctl_islist_data *islist; struct cctl_islist_conn *cur_conn; char *str; islist = (struct cctl_islist_data *)user_data; cur_conn = islist->cur_conn; if ((cur_conn == NULL) && (strcmp(name, "ctlislist") != 0)) errx(1, "%s: cur_conn == NULL! (name = %s)", __func__, name); if (islist->cur_sb[islist->level] == NULL) errx(1, "%s: no valid sbuf at level %d (name %s)", __func__, islist->level, name); sbuf_finish(islist->cur_sb[islist->level]); str = strdup(sbuf_data(islist->cur_sb[islist->level])); if (str == NULL) err(1, "%s can't allocate %zd bytes for string", __func__, sbuf_len(islist->cur_sb[islist->level])); sbuf_delete(islist->cur_sb[islist->level]); islist->cur_sb[islist->level] = NULL; islist->level--; if (strcmp(name, "initiator") == 0) { cur_conn->initiator = str; str = NULL; } else if (strcmp(name, "initiator_addr") == 0) { cur_conn->initiator_addr = str; str = NULL; } else if (strcmp(name, "initiator_alias") == 0) { cur_conn->initiator_alias = str; str = NULL; } else if (strcmp(name, "target") == 0) { cur_conn->target = str; str = NULL; } else if (strcmp(name, "target_alias") == 0) { cur_conn->target_alias = str; str = NULL; } else if (strcmp(name, "target_portal_group_tag") == 0) { } else if (strcmp(name, "header_digest") == 0) { cur_conn->header_digest = str; str = NULL; } else if (strcmp(name, "data_digest") == 0) { cur_conn->data_digest = str; str = NULL; } else if (strcmp(name, "max_recv_data_segment_length") == 0) { cur_conn->max_recv_data_segment_length = str; str = NULL; } else if (strcmp(name, "max_send_data_segment_length") == 0) { cur_conn->max_send_data_segment_length = str; str = NULL; } else if (strcmp(name, "max_burst_length") == 0) { cur_conn->max_burst_length = str; str = NULL; } else if (strcmp(name, "first_burst_length") == 0) { cur_conn->first_burst_length = str; str = NULL; } else if (strcmp(name, "offload") == 0) { cur_conn->offload = str; str = NULL; } else if (strcmp(name, "immediate_data") == 0) { cur_conn->immediate_data = atoi(str); } else if (strcmp(name, "iser") == 0) { cur_conn->iser = atoi(str); } else if (strcmp(name, "connection") == 0) { islist->cur_conn = NULL; } else if (strcmp(name, "ctlislist") == 0) { /* Nothing. */ } else { /* * Unknown element; ignore it for forward compatibility. */ } free(str); } static void cctl_islist_char_handler(void *user_data, const XML_Char *str, int len) { struct cctl_islist_data *islist; islist = (struct cctl_islist_data *)user_data; sbuf_bcat(islist->cur_sb[islist->level], str, len); } static int cctl_islist(int fd, int argc, char **argv, char *combinedopt) { struct ctl_iscsi req; struct cctl_islist_data islist; struct cctl_islist_conn *conn; XML_Parser parser; char *conn_str; int conn_len; int dump_xml = 0; int c, retval, verbose = 0; retval = 0; conn_len = 4096; bzero(&islist, sizeof(islist)); STAILQ_INIT(&islist.conn_list); while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'v': verbose = 1; break; case 'x': dump_xml = 1; break; default: break; } } retry: conn_str = malloc(conn_len); bzero(&req, sizeof(req)); req.type = CTL_ISCSI_LIST; req.data.list.alloc_len = conn_len; req.data.list.conn_xml = conn_str; if (ioctl(fd, CTL_ISCSI, &req) == -1) { warn("%s: error issuing CTL_ISCSI ioctl", __func__); retval = 1; goto bailout; } if (req.status == CTL_ISCSI_ERROR) { warnx("%s: error returned from CTL_ISCSI ioctl:\n%s", __func__, req.error_str); } else if (req.status == CTL_ISCSI_LIST_NEED_MORE_SPACE) { conn_len = conn_len << 1; goto retry; } if (dump_xml != 0) { printf("%s", conn_str); goto bailout; } parser = XML_ParserCreate(NULL); if (parser == NULL) { warn("%s: Unable to create XML parser", __func__); retval = 1; goto bailout; } XML_SetUserData(parser, &islist); XML_SetElementHandler(parser, cctl_islist_start_element, cctl_islist_end_element); XML_SetCharacterDataHandler(parser, cctl_islist_char_handler); retval = XML_Parse(parser, conn_str, strlen(conn_str), 1); if (retval != 1) { warnx("%s: Unable to parse XML: Error %d", __func__, XML_GetErrorCode(parser)); XML_ParserFree(parser); retval = 1; goto bailout; } retval = 0; XML_ParserFree(parser); if (verbose != 0) { STAILQ_FOREACH(conn, &islist.conn_list, links) { printf("%-25s %d\n", "Session ID:", conn->connection_id); printf("%-25s %s\n", "Initiator name:", conn->initiator); printf("%-25s %s\n", "Initiator portal:", conn->initiator_addr); printf("%-25s %s\n", "Initiator alias:", conn->initiator_alias); printf("%-25s %s\n", "Target name:", conn->target); printf("%-25s %s\n", "Target alias:", conn->target_alias); printf("%-25s %s\n", "Header digest:", conn->header_digest); printf("%-25s %s\n", "Data digest:", conn->data_digest); printf("%-25s %s\n", "MaxRecvDataSegmentLength:", conn->max_recv_data_segment_length); printf("%-25s %s\n", "MaxSendDataSegmentLength:", conn->max_send_data_segment_length); printf("%-25s %s\n", "MaxBurstLen:", conn->max_burst_length); printf("%-25s %s\n", "FirstBurstLen:", conn->first_burst_length); printf("%-25s %s\n", "ImmediateData:", conn->immediate_data ? "Yes" : "No"); printf("%-25s %s\n", "iSER (RDMA):", conn->iser ? "Yes" : "No"); printf("%-25s %s\n", "Offload driver:", conn->offload); printf("\n"); } } else { printf("%4s %-16s %-36s %-36s\n", "ID", "Portal", "Initiator name", "Target name"); STAILQ_FOREACH(conn, &islist.conn_list, links) { printf("%4u %-16s %-36s %-36s\n", conn->connection_id, conn->initiator_addr, conn->initiator, conn->target); } } bailout: free(conn_str); return (retval); } static int cctl_islogout(int fd, int argc, char **argv, char *combinedopt) { struct ctl_iscsi req; int retval = 0, c; int all = 0, connection_id = -1, nargs = 0; char *initiator_name = NULL, *initiator_addr = NULL; while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'a': all = 1; nargs++; break; case 'c': connection_id = strtoul(optarg, NULL, 0); nargs++; break; case 'i': initiator_name = strdup(optarg); if (initiator_name == NULL) err(1, "%s: strdup", __func__); nargs++; break; case 'p': initiator_addr = strdup(optarg); if (initiator_addr == NULL) err(1, "%s: strdup", __func__); nargs++; break; default: break; } } if (nargs == 0) errx(1, "%s: either -a, -c, -i, or -p must be specified", __func__); if (nargs > 1) errx(1, "%s: only one of -a, -c, -i, or -p may be specified", __func__); bzero(&req, sizeof(req)); req.type = CTL_ISCSI_LOGOUT; req.data.logout.connection_id = connection_id; if (initiator_addr != NULL) strlcpy(req.data.logout.initiator_addr, initiator_addr, sizeof(req.data.logout.initiator_addr)); if (initiator_name != NULL) strlcpy(req.data.logout.initiator_name, initiator_name, sizeof(req.data.logout.initiator_name)); if (all != 0) req.data.logout.all = 1; if (ioctl(fd, CTL_ISCSI, &req) == -1) { warn("%s: error issuing CTL_ISCSI ioctl", __func__); retval = 1; goto bailout; } if (req.status != CTL_ISCSI_OK) { warnx("%s: error returned from CTL iSCSI logout request:\n%s", __func__, req.error_str); retval = 1; goto bailout; } printf("iSCSI logout requests submitted\n"); bailout: return (retval); } static int cctl_isterminate(int fd, int argc, char **argv, char *combinedopt) { struct ctl_iscsi req; int retval = 0, c; int all = 0, connection_id = -1, nargs = 0; char *initiator_name = NULL, *initiator_addr = NULL; while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'a': all = 1; nargs++; break; case 'c': connection_id = strtoul(optarg, NULL, 0); nargs++; break; case 'i': initiator_name = strdup(optarg); if (initiator_name == NULL) err(1, "%s: strdup", __func__); nargs++; break; case 'p': initiator_addr = strdup(optarg); if (initiator_addr == NULL) err(1, "%s: strdup", __func__); nargs++; break; default: break; } } if (nargs == 0) errx(1, "%s: either -a, -c, -i, or -p must be specified", __func__); if (nargs > 1) errx(1, "%s: only one of -a, -c, -i, or -p may be specified", __func__); bzero(&req, sizeof(req)); req.type = CTL_ISCSI_TERMINATE; req.data.terminate.connection_id = connection_id; if (initiator_addr != NULL) strlcpy(req.data.terminate.initiator_addr, initiator_addr, sizeof(req.data.terminate.initiator_addr)); if (initiator_name != NULL) strlcpy(req.data.terminate.initiator_name, initiator_name, sizeof(req.data.terminate.initiator_name)); if (all != 0) req.data.terminate.all = 1; if (ioctl(fd, CTL_ISCSI, &req) == -1) { warn("%s: error issuing CTL_ISCSI ioctl", __func__); retval = 1; goto bailout; } if (req.status != CTL_ISCSI_OK) { warnx("%s: error returned from CTL iSCSI connection " "termination request:\n%s", __func__, req.error_str); retval = 1; goto bailout; } printf("iSCSI connections terminated\n"); bailout: return (retval); } /* * Name/value pair used for per-LUN attributes. */ struct cctl_lun_nv { char *name; char *value; STAILQ_ENTRY(cctl_lun_nv) links; }; /* * Backend LUN information. */ struct cctl_lun { uint64_t lun_id; char *backend_type; uint64_t size_blocks; uint32_t blocksize; char *serial_number; char *device_id; STAILQ_HEAD(,cctl_lun_nv) attr_list; STAILQ_ENTRY(cctl_lun) links; }; struct cctl_devlist_data { int num_luns; STAILQ_HEAD(,cctl_lun) lun_list; struct cctl_lun *cur_lun; int level; struct sbuf *cur_sb[32]; }; static void cctl_start_element(void *user_data, const char *name, const char **attr) { int i; struct cctl_devlist_data *devlist; struct cctl_lun *cur_lun; devlist = (struct cctl_devlist_data *)user_data; cur_lun = devlist->cur_lun; devlist->level++; if ((u_int)devlist->level >= (sizeof(devlist->cur_sb) / sizeof(devlist->cur_sb[0]))) errx(1, "%s: too many nesting levels, %zd max", __func__, sizeof(devlist->cur_sb) / sizeof(devlist->cur_sb[0])); devlist->cur_sb[devlist->level] = sbuf_new_auto(); if (devlist->cur_sb[devlist->level] == NULL) err(1, "%s: Unable to allocate sbuf", __func__); if (strcmp(name, "lun") == 0) { if (cur_lun != NULL) errx(1, "%s: improper lun element nesting", __func__); cur_lun = calloc(1, sizeof(*cur_lun)); if (cur_lun == NULL) err(1, "%s: cannot allocate %zd bytes", __func__, sizeof(*cur_lun)); devlist->num_luns++; devlist->cur_lun = cur_lun; STAILQ_INIT(&cur_lun->attr_list); STAILQ_INSERT_TAIL(&devlist->lun_list, cur_lun, links); for (i = 0; attr[i] != NULL; i += 2) { if (strcmp(attr[i], "id") == 0) { cur_lun->lun_id = strtoull(attr[i+1], NULL, 0); } else { errx(1, "%s: invalid LUN attribute %s = %s", __func__, attr[i], attr[i+1]); } } } } static void cctl_end_element(void *user_data, const char *name) { struct cctl_devlist_data *devlist; struct cctl_lun *cur_lun; char *str; devlist = (struct cctl_devlist_data *)user_data; cur_lun = devlist->cur_lun; if ((cur_lun == NULL) && (strcmp(name, "ctllunlist") != 0)) errx(1, "%s: cur_lun == NULL! (name = %s)", __func__, name); if (devlist->cur_sb[devlist->level] == NULL) errx(1, "%s: no valid sbuf at level %d (name %s)", __func__, devlist->level, name); if (sbuf_finish(devlist->cur_sb[devlist->level]) != 0) err(1, "%s: sbuf_finish", __func__); str = strdup(sbuf_data(devlist->cur_sb[devlist->level])); if (str == NULL) err(1, "%s can't allocate %zd bytes for string", __func__, sbuf_len(devlist->cur_sb[devlist->level])); if (strlen(str) == 0) { free(str); str = NULL; } sbuf_delete(devlist->cur_sb[devlist->level]); devlist->cur_sb[devlist->level] = NULL; devlist->level--; if (strcmp(name, "backend_type") == 0) { cur_lun->backend_type = str; str = NULL; } else if (strcmp(name, "size") == 0) { cur_lun->size_blocks = strtoull(str, NULL, 0); } else if (strcmp(name, "blocksize") == 0) { cur_lun->blocksize = strtoul(str, NULL, 0); } else if (strcmp(name, "serial_number") == 0) { cur_lun->serial_number = str; str = NULL; } else if (strcmp(name, "device_id") == 0) { cur_lun->device_id = str; str = NULL; } else if (strcmp(name, "lun") == 0) { devlist->cur_lun = NULL; } else if (strcmp(name, "ctllunlist") == 0) { /* Nothing. */ } else { struct cctl_lun_nv *nv; nv = calloc(1, sizeof(*nv)); if (nv == NULL) err(1, "%s: can't allocate %zd bytes for nv pair", __func__, sizeof(*nv)); nv->name = strdup(name); if (nv->name == NULL) err(1, "%s: can't allocated %zd bytes for string", __func__, strlen(name)); nv->value = str; str = NULL; STAILQ_INSERT_TAIL(&cur_lun->attr_list, nv, links); } free(str); } static void cctl_char_handler(void *user_data, const XML_Char *str, int len) { struct cctl_devlist_data *devlist; devlist = (struct cctl_devlist_data *)user_data; sbuf_bcat(devlist->cur_sb[devlist->level], str, len); } static int cctl_devlist(int fd, int argc, char **argv, char *combinedopt) { struct ctl_lun_list list; struct cctl_devlist_data devlist; struct cctl_lun *lun; XML_Parser parser; char *lun_str; int lun_len; int dump_xml = 0; int retval, c; char *backend = NULL; int verbose = 0; retval = 0; lun_len = 4096; bzero(&devlist, sizeof(devlist)); STAILQ_INIT(&devlist.lun_list); while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'b': backend = strdup(optarg); break; case 'v': verbose++; break; case 'x': dump_xml = 1; break; default: break; } } retry: lun_str = malloc(lun_len); bzero(&list, sizeof(list)); list.alloc_len = lun_len; list.status = CTL_LUN_LIST_NONE; list.lun_xml = lun_str; if (ioctl(fd, CTL_LUN_LIST, &list) == -1) { warn("%s: error issuing CTL_LUN_LIST ioctl", __func__); retval = 1; goto bailout; } if (list.status == CTL_LUN_LIST_ERROR) { warnx("%s: error returned from CTL_LUN_LIST ioctl:\n%s", __func__, list.error_str); } else if (list.status == CTL_LUN_LIST_NEED_MORE_SPACE) { lun_len = lun_len << 1; goto retry; } if (dump_xml != 0) { printf("%s", lun_str); goto bailout; } parser = XML_ParserCreate(NULL); if (parser == NULL) { warn("%s: Unable to create XML parser", __func__); retval = 1; goto bailout; } XML_SetUserData(parser, &devlist); XML_SetElementHandler(parser, cctl_start_element, cctl_end_element); XML_SetCharacterDataHandler(parser, cctl_char_handler); retval = XML_Parse(parser, lun_str, strlen(lun_str), 1); if (retval != 1) { warnx("%s: Unable to parse XML: Error %d", __func__, XML_GetErrorCode(parser)); XML_ParserFree(parser); retval = 1; goto bailout; } retval = 0; XML_ParserFree(parser); printf("LUN Backend %18s %4s %-16s %-16s\n", "Size (Blocks)", "BS", "Serial Number", "Device ID"); STAILQ_FOREACH(lun, &devlist.lun_list, links) { struct cctl_lun_nv *nv; if ((backend != NULL) && (strcmp(lun->backend_type, backend) != 0)) continue; printf("%3ju %-8s %18ju %4u %-16s %-16s\n", (uintmax_t)lun->lun_id, lun->backend_type, (uintmax_t)lun->size_blocks, lun->blocksize, lun->serial_number, lun->device_id); if (verbose == 0) continue; STAILQ_FOREACH(nv, &lun->attr_list, links) { printf(" %s=%s\n", nv->name, nv->value); } } bailout: free(lun_str); return (retval); } /* * Port information. */ struct cctl_port { uint64_t port_id; char *online; char *frontend_type; char *name; int pp, vp; char *target, *port, *lun_map; STAILQ_HEAD(,cctl_lun_nv) init_list; STAILQ_HEAD(,cctl_lun_nv) lun_list; STAILQ_HEAD(,cctl_lun_nv) attr_list; STAILQ_ENTRY(cctl_port) links; }; struct cctl_portlist_data { int num_ports; STAILQ_HEAD(,cctl_port) port_list; struct cctl_port *cur_port; int level; uint64_t cur_id; struct sbuf *cur_sb[32]; }; static void cctl_start_pelement(void *user_data, const char *name, const char **attr) { int i; struct cctl_portlist_data *portlist; struct cctl_port *cur_port; portlist = (struct cctl_portlist_data *)user_data; cur_port = portlist->cur_port; portlist->level++; if ((u_int)portlist->level >= (sizeof(portlist->cur_sb) / sizeof(portlist->cur_sb[0]))) errx(1, "%s: too many nesting levels, %zd max", __func__, sizeof(portlist->cur_sb) / sizeof(portlist->cur_sb[0])); portlist->cur_sb[portlist->level] = sbuf_new_auto(); if (portlist->cur_sb[portlist->level] == NULL) err(1, "%s: Unable to allocate sbuf", __func__); portlist->cur_id = 0; for (i = 0; attr[i] != NULL; i += 2) { if (strcmp(attr[i], "id") == 0) { portlist->cur_id = strtoull(attr[i+1], NULL, 0); break; } } if (strcmp(name, "targ_port") == 0) { if (cur_port != NULL) errx(1, "%s: improper port element nesting", __func__); cur_port = calloc(1, sizeof(*cur_port)); if (cur_port == NULL) err(1, "%s: cannot allocate %zd bytes", __func__, sizeof(*cur_port)); portlist->num_ports++; portlist->cur_port = cur_port; STAILQ_INIT(&cur_port->init_list); STAILQ_INIT(&cur_port->lun_list); STAILQ_INIT(&cur_port->attr_list); cur_port->port_id = portlist->cur_id; STAILQ_INSERT_TAIL(&portlist->port_list, cur_port, links); } } static void cctl_end_pelement(void *user_data, const char *name) { struct cctl_portlist_data *portlist; struct cctl_port *cur_port; char *str; portlist = (struct cctl_portlist_data *)user_data; cur_port = portlist->cur_port; if ((cur_port == NULL) && (strcmp(name, "ctlportlist") != 0)) errx(1, "%s: cur_port == NULL! (name = %s)", __func__, name); if (portlist->cur_sb[portlist->level] == NULL) errx(1, "%s: no valid sbuf at level %d (name %s)", __func__, portlist->level, name); if (sbuf_finish(portlist->cur_sb[portlist->level]) != 0) err(1, "%s: sbuf_finish", __func__); str = strdup(sbuf_data(portlist->cur_sb[portlist->level])); if (str == NULL) err(1, "%s can't allocate %zd bytes for string", __func__, sbuf_len(portlist->cur_sb[portlist->level])); if (strlen(str) == 0) { free(str); str = NULL; } sbuf_delete(portlist->cur_sb[portlist->level]); portlist->cur_sb[portlist->level] = NULL; portlist->level--; if (strcmp(name, "frontend_type") == 0) { cur_port->frontend_type = str; str = NULL; } else if (strcmp(name, "port_name") == 0) { cur_port->name = str; str = NULL; } else if (strcmp(name, "online") == 0) { cur_port->online = str; str = NULL; } else if (strcmp(name, "physical_port") == 0) { cur_port->pp = strtoull(str, NULL, 0); } else if (strcmp(name, "virtual_port") == 0) { cur_port->vp = strtoull(str, NULL, 0); } else if (strcmp(name, "target") == 0) { cur_port->target = str; str = NULL; } else if (strcmp(name, "port") == 0) { cur_port->port = str; str = NULL; } else if (strcmp(name, "lun_map") == 0) { cur_port->lun_map = str; str = NULL; } else if (strcmp(name, "targ_port") == 0) { portlist->cur_port = NULL; } else if (strcmp(name, "ctlportlist") == 0) { /* Nothing. */ } else { struct cctl_lun_nv *nv; nv = calloc(1, sizeof(*nv)); if (nv == NULL) err(1, "%s: can't allocate %zd bytes for nv pair", __func__, sizeof(*nv)); if (strcmp(name, "initiator") == 0 || strcmp(name, "lun") == 0) asprintf(&nv->name, "%ju", portlist->cur_id); else nv->name = strdup(name); if (nv->name == NULL) err(1, "%s: can't allocated %zd bytes for string", __func__, strlen(name)); nv->value = str; str = NULL; if (strcmp(name, "initiator") == 0) STAILQ_INSERT_TAIL(&cur_port->init_list, nv, links); else if (strcmp(name, "lun") == 0) STAILQ_INSERT_TAIL(&cur_port->lun_list, nv, links); else STAILQ_INSERT_TAIL(&cur_port->attr_list, nv, links); } free(str); } static void cctl_char_phandler(void *user_data, const XML_Char *str, int len) { struct cctl_portlist_data *portlist; portlist = (struct cctl_portlist_data *)user_data; sbuf_bcat(portlist->cur_sb[portlist->level], str, len); } static int cctl_portlist(int fd, int argc, char **argv, char *combinedopt) { struct ctl_lun_list list; struct cctl_portlist_data portlist; struct cctl_port *port; XML_Parser parser; - char *port_str; + char *port_str = NULL; int port_len; int dump_xml = 0; int retval, c; char *frontend = NULL; uint64_t portarg = UINT64_MAX; int verbose = 0, init = 0, lun = 0, quiet = 0; retval = 0; port_len = 4096; bzero(&portlist, sizeof(portlist)); STAILQ_INIT(&portlist.port_list); while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'f': frontend = strdup(optarg); break; case 'i': init++; break; case 'l': lun++; break; case 'p': portarg = strtoll(optarg, NULL, 0); break; case 'q': quiet++; break; case 'v': verbose++; break; case 'x': dump_xml = 1; break; default: break; } } retry: - port_str = malloc(port_len); + port_str = (char *)realloc(port_str, port_len); bzero(&list, sizeof(list)); list.alloc_len = port_len; list.status = CTL_LUN_LIST_NONE; list.lun_xml = port_str; if (ioctl(fd, CTL_PORT_LIST, &list) == -1) { warn("%s: error issuing CTL_PORT_LIST ioctl", __func__); retval = 1; goto bailout; } if (list.status == CTL_LUN_LIST_ERROR) { warnx("%s: error returned from CTL_PORT_LIST ioctl:\n%s", __func__, list.error_str); } else if (list.status == CTL_LUN_LIST_NEED_MORE_SPACE) { port_len = port_len << 1; goto retry; } if (dump_xml != 0) { printf("%s", port_str); goto bailout; } parser = XML_ParserCreate(NULL); if (parser == NULL) { warn("%s: Unable to create XML parser", __func__); retval = 1; goto bailout; } XML_SetUserData(parser, &portlist); XML_SetElementHandler(parser, cctl_start_pelement, cctl_end_pelement); XML_SetCharacterDataHandler(parser, cctl_char_phandler); retval = XML_Parse(parser, port_str, strlen(port_str), 1); if (retval != 1) { warnx("%s: Unable to parse XML: Error %d", __func__, XML_GetErrorCode(parser)); XML_ParserFree(parser); retval = 1; goto bailout; } retval = 0; XML_ParserFree(parser); if (quiet == 0) printf("Port Online Frontend Name pp vp\n"); STAILQ_FOREACH(port, &portlist.port_list, links) { struct cctl_lun_nv *nv; if ((frontend != NULL) && (strcmp(port->frontend_type, frontend) != 0)) continue; if ((portarg != UINT64_MAX) && (portarg != port->port_id)) continue; printf("%-4ju %-6s %-8s %-8s %-2d %-2d %s\n", (uintmax_t)port->port_id, port->online, port->frontend_type, port->name, port->pp, port->vp, port->port ? port->port : ""); if (init || verbose) { if (port->target) printf(" Target: %s\n", port->target); STAILQ_FOREACH(nv, &port->init_list, links) { printf(" Initiator %s: %s\n", nv->name, nv->value); } } if (lun || verbose) { if (port->lun_map) { STAILQ_FOREACH(nv, &port->lun_list, links) printf(" LUN %s: %s\n", nv->name, nv->value); if (STAILQ_EMPTY(&port->lun_list)) printf(" No LUNs mapped\n"); } else printf(" All LUNs mapped\n"); } if (verbose) { STAILQ_FOREACH(nv, &port->attr_list, links) { printf(" %s=%s\n", nv->name, nv->value); } } } bailout: free(port_str); return (retval); } static int cctl_lunmap(int fd, int argc, char **argv, char *combinedopt) { struct ctl_lun_map lm; int retval = 0, c; retval = 0; lm.port = UINT32_MAX; lm.plun = UINT32_MAX; lm.lun = UINT32_MAX; while ((c = getopt(argc, argv, combinedopt)) != -1) { switch (c) { case 'p': lm.port = strtoll(optarg, NULL, 0); break; case 'l': lm.plun = strtoll(optarg, NULL, 0); break; case 'L': lm.lun = strtoll(optarg, NULL, 0); break; default: break; } } if (ioctl(fd, CTL_LUN_MAP, &lm) == -1) { warn("%s: error issuing CTL_LUN_MAP ioctl", __func__); retval = 1; } return (retval); } void usage(int error) { fprintf(error ? stderr : stdout, "Usage:\n" "Primary commands:\n" " ctladm tur [dev_id][general options]\n" " ctladm inquiry [dev_id][general options]\n" " ctladm devid [dev_id][general options]\n" " ctladm reqsense [dev_id][general options]\n" " ctladm reportluns [dev_id][general options]\n" " ctladm read [dev_id][general options] <-l lba> <-d len>\n" " <-f file|-> <-b blocksize> [-c cdbsize][-N]\n" " ctladm write [dev_id][general options] <-l lba> <-d len>\n" " <-f file|-> <-b blocksize> [-c cdbsize][-N]\n" " ctladm readcap [dev_id][general options] [-c cdbsize]\n" " ctladm modesense [dev_id][general options] <-m page|-l> [-P pc]\n" " [-d] [-S subpage] [-c cdbsize]\n" " ctladm prin [dev_id][general options] <-a action>\n" " ctladm prout [dev_id][general options] <-a action>\n" " <-r restype] [-k key] [-s sa_key]\n" " ctladm rtpg [dev_id][general options]\n" " ctladm start [dev_id][general options] [-i] [-o]\n" " ctladm stop [dev_id][general options] [-i] [-o]\n" " ctladm synccache [dev_id][general options] [-l lba]\n" " [-b blockcount] [-r] [-i] [-c cdbsize]\n" " ctladm create <-b backend> [-B blocksize] [-d device_id]\n" " [-l lun_id] [-o name=value] [-s size_bytes]\n" " [-S serial_num] [-t dev_type]\n" " ctladm remove <-b backend> <-l lun_id> [-o name=value]\n" " ctladm modify <-b backend> <-l lun_id> <-s size_bytes>\n" " ctladm devlist [-b backend] [-v] [-x]\n" " ctladm lunlist\n" " ctladm lunmap -p targ_port [-l pLUN] [-L cLUN]\n" " ctladm delay [dev_id] <-l datamove|done> [-T oneshot|cont]\n" " [-t secs]\n" " ctladm inject [dev_id] <-i action> <-p pattern> [-r lba,len]\n" " [-s len fmt [args]] [-c] [-d delete_id]\n" " ctladm port <-o | [-w wwnn][-W wwpn]>\n" " [-p targ_port] [-t port_type]\n" +" <-c> [-d driver] [-O name=value]\n" +" <-r> <-p targ_port>\n" " ctladm portlist [-f frontend] [-i] [-p targ_port] [-q] [-v] [-x]\n" " ctladm islist [-v | -x]\n" " ctladm islogout <-a | -c connection-id | -i name | -p portal>\n" " ctladm isterminate <-a | -c connection-id | -i name | -p portal>\n" " ctladm dumpooa\n" " ctladm dumpstructs\n" " ctladm help\n" "General Options:\n" "-I intiator_id : defaults to 7, used to change the initiator id\n" "-C retries : specify the number of times to retry this command\n" "-D devicename : specify the device to operate on\n" " : (default is %s)\n" "read/write options:\n" "-l lba : logical block address\n" "-d len : read/write length, in blocks\n" "-f file|- : write/read data to/from file or stdout/stdin\n" "-b blocksize : block size, in bytes\n" "-c cdbsize : specify minimum cdb size: 6, 10, 12 or 16\n" "-N : do not copy data to/from userland\n" "readcapacity options:\n" "-c cdbsize : specify minimum cdb size: 10 or 16\n" "modesense options:\n" "-m page : specify the mode page to view\n" "-l : request a list of supported pages\n" "-P pc : specify the page control value: 0-3 (current,\n" " changeable, default, saved, respectively)\n" "-d : disable block descriptors for mode sense\n" "-S subpage : specify a subpage\n" "-c cdbsize : specify minimum cdb size: 6 or 10\n" "persistent reserve in options:\n" "-a action : specify the action value: 0-2 (read key, read\n" " reservation, read capabilities, respectively)\n" "persistent reserve out options:\n" "-a action : specify the action value: 0-5 (register, reserve,\n" " release, clear, preempt, register and ignore)\n" "-k key : key value\n" "-s sa_key : service action value\n" "-r restype : specify the reservation type: 0-5(wr ex, ex ac,\n" " wr ex ro, ex ac ro, wr ex ar, ex ac ar)\n" "start/stop options:\n" "-i : set the immediate bit (CTL does not support this)\n" "-o : set the on/offline bit\n" "synccache options:\n" "-l lba : set the starting LBA\n" "-b blockcount : set the length to sync in blocks\n" "-r : set the relative addressing bit\n" "-i : set the immediate bit\n" "-c cdbsize : specify minimum cdb size: 10 or 16\n" "create options:\n" "-b backend : backend name (\"block\", \"ramdisk\", etc.)\n" "-B blocksize : LUN blocksize in bytes (some backends)\n" "-d device_id : SCSI VPD page 0x83 ID\n" "-l lun_id : requested LUN number\n" "-o name=value : backend-specific options, multiple allowed\n" "-s size_bytes : LUN size in bytes (some backends)\n" "-S serial_num : SCSI VPD page 0x80 serial number\n" "-t dev_type : SCSI device type (0=disk, 3=processor)\n" "remove options:\n" "-b backend : backend name (\"block\", \"ramdisk\", etc.)\n" "-l lun_id : LUN number to delete\n" "-o name=value : backend-specific options, multiple allowed\n" "devlist options:\n" "-b backend : list devices from specified backend only\n" "-v : be verbose, show backend attributes\n" "-x : dump raw XML\n" "delay options:\n" "-l datamove|done : delay command at datamove or done phase\n" "-T oneshot : delay one command, then resume normal completion\n" "-T cont : delay all commands\n" "-t secs : number of seconds to delay\n" "inject options:\n" "-i error_action : action to perform\n" "-p pattern : command pattern to look for\n" "-r lba,len : LBA range for pattern\n" "-s len fmt [args] : sense data for custom sense action\n" "-c : continuous operation\n" "-d delete_id : error id to delete\n" "port options:\n" +"-c : create new ioctl or iscsi frontend port\n" +"-d : specify ioctl or iscsi frontend type\n" "-l : list frontend ports\n" "-o on|off : turn frontend ports on or off\n" +"-O pp|vp : create new frontend port using pp and/or vp\n" "-w wwnn : set WWNN for one frontend\n" "-W wwpn : set WWPN for one frontend\n" "-t port_type : specify fc, scsi, ioctl, internal frontend type\n" "-p targ_port : specify target port number\n" +"-r : remove frontend port\n" "-q : omit header in list output\n" "-x : output port list in XML format\n" "portlist options:\n" "-f frontend : specify frontend type\n" "-i : report target and initiators addresses\n" "-l : report LUN mapping\n" "-p targ_port : specify target port number\n" "-q : omit header in list output\n" "-v : verbose output (report all port options)\n" "-x : output port list in XML format\n" "lunmap options:\n" "-p targ_port : specify target port number\n" "-L pLUN : specify port-visible LUN\n" "-L cLUN : specify CTL LUN\n", CTL_DEFAULT_DEV); } int main(int argc, char **argv) { int c; ctladm_cmdfunction command; ctladm_cmdargs cmdargs; ctladm_optret optreturn; char *device; const char *mainopt = "C:D:I:"; const char *subopt = NULL; char combinedopt[256]; int lun; int optstart = 2; int retval, fd; int retries; int initid; int saved_errno; retval = 0; cmdargs = CTLADM_ARG_NONE; command = CTLADM_CMD_HELP; device = NULL; fd = -1; retries = 0; lun = 0; initid = 7; if (argc < 2) { usage(1); retval = 1; goto bailout; } /* * Get the base option. */ optreturn = getoption(option_table,argv[1], &command, &cmdargs,&subopt); if (optreturn == CC_OR_AMBIGUOUS) { warnx("ambiguous option %s", argv[1]); usage(0); exit(1); } else if (optreturn == CC_OR_NOT_FOUND) { warnx("option %s not found", argv[1]); usage(0); exit(1); } if (cmdargs & CTLADM_ARG_NEED_TL) { if ((argc < 3) || (!isdigit(argv[2][0]))) { warnx("option %s requires a lun argument", argv[1]); usage(0); exit(1); } lun = strtol(argv[2], NULL, 0); cmdargs |= CTLADM_ARG_TARG_LUN; optstart++; } /* * Ahh, getopt(3) is a pain. * * This is a gross hack. There really aren't many other good * options (excuse the pun) for parsing options in a situation like * this. getopt is kinda braindead, so you end up having to run * through the options twice, and give each invocation of getopt * the option string for the other invocation. * * You would think that you could just have two groups of options. * The first group would get parsed by the first invocation of * getopt, and the second group would get parsed by the second * invocation of getopt. It doesn't quite work out that way. When * the first invocation of getopt finishes, it leaves optind pointing * to the argument _after_ the first argument in the second group. * So when the second invocation of getopt comes around, it doesn't * recognize the first argument it gets and then bails out. * * A nice alternative would be to have a flag for getopt that says * "just keep parsing arguments even when you encounter an unknown * argument", but there isn't one. So there's no real clean way to * easily parse two sets of arguments without having one invocation * of getopt know about the other. * * Without this hack, the first invocation of getopt would work as * long as the generic arguments are first, but the second invocation * (in the subfunction) would fail in one of two ways. In the case * where you don't set optreset, it would fail because optind may be * pointing to the argument after the one it should be pointing at. * In the case where you do set optreset, and reset optind, it would * fail because getopt would run into the first set of options, which * it doesn't understand. * * All of this would "sort of" work if you could somehow figure out * whether optind had been incremented one option too far. The * mechanics of that, however, are more daunting than just giving * both invocations all of the expect options for either invocation. * * Needless to say, I wouldn't mind if someone invented a better * (non-GPL!) command line parsing interface than getopt. I * wouldn't mind if someone added more knobs to getopt to make it * work better. Who knows, I may talk myself into doing it someday, * if the standards weenies let me. As it is, it just leads to * hackery like this and causes people to avoid it in some cases. * * KDM, September 8th, 1998 */ if (subopt != NULL) sprintf(combinedopt, "%s%s", mainopt, subopt); else sprintf(combinedopt, "%s", mainopt); /* * Start getopt processing at argv[2/3], since we've already * accepted argv[1..2] as the command name, and as a possible * device name. */ optind = optstart; /* * Now we run through the argument list looking for generic * options, and ignoring options that possibly belong to * subfunctions. */ while ((c = getopt(argc, argv, combinedopt))!= -1){ switch (c) { case 'C': cmdargs |= CTLADM_ARG_RETRIES; retries = strtol(optarg, NULL, 0); break; case 'D': device = strdup(optarg); cmdargs |= CTLADM_ARG_DEVICE; break; case 'I': cmdargs |= CTLADM_ARG_INITIATOR; initid = strtol(optarg, NULL, 0); break; default: break; } } if ((cmdargs & CTLADM_ARG_INITIATOR) == 0) initid = 7; optind = optstart; optreset = 1; /* * Default to opening the CTL device for now. */ if (((cmdargs & CTLADM_ARG_DEVICE) == 0) && (command != CTLADM_CMD_HELP)) { device = strdup(CTL_DEFAULT_DEV); cmdargs |= CTLADM_ARG_DEVICE; } if ((cmdargs & CTLADM_ARG_DEVICE) && (command != CTLADM_CMD_HELP)) { fd = open(device, O_RDWR); if (fd == -1 && errno == ENOENT) { saved_errno = errno; retval = kldload("ctl"); if (retval != -1) fd = open(device, O_RDWR); else errno = saved_errno; } if (fd == -1) { fprintf(stderr, "%s: error opening %s: %s\n", argv[0], device, strerror(errno)); retval = 1; goto bailout; } #ifdef WANT_ISCSI else { if (modfind("cfiscsi") == -1 && kldload("cfiscsi") == -1) warn("couldn't load cfiscsi"); } #endif } else if ((command != CTLADM_CMD_HELP) && ((cmdargs & CTLADM_ARG_DEVICE) == 0)) { fprintf(stderr, "%s: you must specify a device with the " "--device argument for this command\n", argv[0]); command = CTLADM_CMD_HELP; retval = 1; } switch (command) { case CTLADM_CMD_TUR: retval = cctl_tur(fd, lun, initid, retries); break; case CTLADM_CMD_INQUIRY: retval = cctl_inquiry(fd, lun, initid, retries); break; case CTLADM_CMD_REQ_SENSE: retval = cctl_req_sense(fd, lun, initid, retries); break; case CTLADM_CMD_REPORT_LUNS: retval = cctl_report_luns(fd, lun, initid, retries); break; case CTLADM_CMD_CREATE: retval = cctl_create_lun(fd, argc, argv, combinedopt); break; case CTLADM_CMD_RM: retval = cctl_rm_lun(fd, argc, argv, combinedopt); break; case CTLADM_CMD_DEVLIST: retval = cctl_devlist(fd, argc, argv, combinedopt); break; case CTLADM_CMD_READ: case CTLADM_CMD_WRITE: retval = cctl_read_write(fd, lun, initid, retries, argc, argv, combinedopt, command); break; case CTLADM_CMD_PORT: retval = cctl_port(fd, argc, argv, combinedopt); break; case CTLADM_CMD_PORTLIST: retval = cctl_portlist(fd, argc, argv, combinedopt); break; case CTLADM_CMD_LUNMAP: retval = cctl_lunmap(fd, argc, argv, combinedopt); break; case CTLADM_CMD_READCAPACITY: retval = cctl_read_capacity(fd, lun, initid, retries, argc, argv, combinedopt); break; case CTLADM_CMD_MODESENSE: retval = cctl_mode_sense(fd, lun, initid, retries, argc, argv, combinedopt); break; case CTLADM_CMD_START: case CTLADM_CMD_STOP: retval = cctl_start_stop(fd, lun, initid, retries, (command == CTLADM_CMD_START) ? 1 : 0, argc, argv, combinedopt); break; case CTLADM_CMD_SYNC_CACHE: retval = cctl_sync_cache(fd, lun, initid, retries, argc, argv, combinedopt); break; case CTLADM_CMD_LUNLIST: retval = cctl_lunlist(fd); break; case CTLADM_CMD_DELAY: retval = cctl_delay(fd, lun, argc, argv, combinedopt); break; case CTLADM_CMD_ERR_INJECT: retval = cctl_error_inject(fd, lun, argc, argv, combinedopt); break; case CTLADM_CMD_DUMPOOA: retval = cctl_dump_ooa(fd, argc, argv); break; case CTLADM_CMD_DUMPSTRUCTS: retval = cctl_dump_structs(fd, cmdargs); break; case CTLADM_CMD_PRES_IN: retval = cctl_persistent_reserve_in(fd, lun, initid, argc, argv, combinedopt, retries); break; case CTLADM_CMD_PRES_OUT: retval = cctl_persistent_reserve_out(fd, lun, initid, argc, argv, combinedopt, retries); break; case CTLADM_CMD_INQ_VPD_DEVID: retval = cctl_inquiry_vpd_devid(fd, lun, initid); break; case CTLADM_CMD_RTPG: retval = cctl_report_target_port_group(fd, lun, initid); break; case CTLADM_CMD_MODIFY: retval = cctl_modify_lun(fd, argc, argv, combinedopt); break; case CTLADM_CMD_ISLIST: retval = cctl_islist(fd, argc, argv, combinedopt); break; case CTLADM_CMD_ISLOGOUT: retval = cctl_islogout(fd, argc, argv, combinedopt); break; case CTLADM_CMD_ISTERMINATE: retval = cctl_isterminate(fd, argc, argv, combinedopt); break; case CTLADM_CMD_HELP: default: usage(retval); break; } bailout: if (fd != -1) close(fd); exit (retval); } /* * vim: ts=8 */ Index: head/usr.sbin/ctld/Makefile =================================================================== --- head/usr.sbin/ctld/Makefile (revision 333445) +++ head/usr.sbin/ctld/Makefile (revision 333446) @@ -1,30 +1,30 @@ # $FreeBSD$ .include CFLAGS+=-I${SRCTOP}/contrib/libucl/include .PATH: ${SRCTOP}/contrib/libucl/include PROG= ctld SRCS= chap.c ctld.c discovery.c isns.c kernel.c keys.c log.c SRCS+= login.c parse.y pdu.c token.l y.tab.h uclparse.c CFLAGS+= -I${.CURDIR} CFLAGS+= -I${SRCTOP}/sys CFLAGS+= -I${SRCTOP}/sys/cam/ctl CFLAGS+= -I${SRCTOP}/sys/dev/iscsi #CFLAGS+= -DICL_KERNEL_PROXY MAN= ctld.8 ctl.conf.5 -LIBADD= bsdxml l md sbuf util ucl m +LIBADD= bsdxml l md sbuf util ucl m nv YFLAGS+= -v CLEANFILES= y.tab.c y.tab.h y.output WARNS?= 6 NO_WMISSING_VARIABLE_DECLARATIONS= .if ${MK_ISCSI} != "no" CFLAGS+= -DWANT_ISCSI .endif .include Index: head/usr.sbin/ctld/ctld.c =================================================================== --- head/usr.sbin/ctld/ctld.c (revision 333445) +++ head/usr.sbin/ctld/ctld.c (revision 333446) @@ -1,2744 +1,2790 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 The FreeBSD Foundation * All rights reserved. * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ctld.h" #include "isns.h" bool proxy_mode = false; static volatile bool sighup_received = false; static volatile bool sigterm_received = false; static volatile bool sigalrm_received = false; static int nchildren = 0; static uint16_t last_portal_group_tag = 0xff; static void usage(void) { fprintf(stderr, "usage: ctld [-d][-u][-f config-file]\n"); exit(1); } char * checked_strdup(const char *s) { char *c; c = strdup(s); if (c == NULL) log_err(1, "strdup"); return (c); } struct conf * conf_new(void) { struct conf *conf; conf = calloc(1, sizeof(*conf)); if (conf == NULL) log_err(1, "calloc"); TAILQ_INIT(&conf->conf_luns); TAILQ_INIT(&conf->conf_targets); TAILQ_INIT(&conf->conf_auth_groups); TAILQ_INIT(&conf->conf_ports); TAILQ_INIT(&conf->conf_portal_groups); TAILQ_INIT(&conf->conf_pports); TAILQ_INIT(&conf->conf_isns); conf->conf_isns_period = 900; conf->conf_isns_timeout = 5; conf->conf_debug = 0; conf->conf_timeout = 60; conf->conf_maxproc = 30; return (conf); } void conf_delete(struct conf *conf) { struct lun *lun, *ltmp; struct target *targ, *tmp; struct auth_group *ag, *cagtmp; struct portal_group *pg, *cpgtmp; struct pport *pp, *pptmp; struct isns *is, *istmp; assert(conf->conf_pidfh == NULL); TAILQ_FOREACH_SAFE(lun, &conf->conf_luns, l_next, ltmp) lun_delete(lun); TAILQ_FOREACH_SAFE(targ, &conf->conf_targets, t_next, tmp) target_delete(targ); TAILQ_FOREACH_SAFE(ag, &conf->conf_auth_groups, ag_next, cagtmp) auth_group_delete(ag); TAILQ_FOREACH_SAFE(pg, &conf->conf_portal_groups, pg_next, cpgtmp) portal_group_delete(pg); TAILQ_FOREACH_SAFE(pp, &conf->conf_pports, pp_next, pptmp) pport_delete(pp); TAILQ_FOREACH_SAFE(is, &conf->conf_isns, i_next, istmp) isns_delete(is); assert(TAILQ_EMPTY(&conf->conf_ports)); free(conf->conf_pidfile_path); free(conf); } static struct auth * auth_new(struct auth_group *ag) { struct auth *auth; auth = calloc(1, sizeof(*auth)); if (auth == NULL) log_err(1, "calloc"); auth->a_auth_group = ag; TAILQ_INSERT_TAIL(&ag->ag_auths, auth, a_next); return (auth); } static void auth_delete(struct auth *auth) { TAILQ_REMOVE(&auth->a_auth_group->ag_auths, auth, a_next); free(auth->a_user); free(auth->a_secret); free(auth->a_mutual_user); free(auth->a_mutual_secret); free(auth); } const struct auth * auth_find(const struct auth_group *ag, const char *user) { const struct auth *auth; TAILQ_FOREACH(auth, &ag->ag_auths, a_next) { if (strcmp(auth->a_user, user) == 0) return (auth); } return (NULL); } static void auth_check_secret_length(struct auth *auth) { size_t len; len = strlen(auth->a_secret); if (len > 16) { if (auth->a_auth_group->ag_name != NULL) log_warnx("secret for user \"%s\", auth-group \"%s\", " "is too long; it should be at most 16 characters " "long", auth->a_user, auth->a_auth_group->ag_name); else log_warnx("secret for user \"%s\", target \"%s\", " "is too long; it should be at most 16 characters " "long", auth->a_user, auth->a_auth_group->ag_target->t_name); } if (len < 12) { if (auth->a_auth_group->ag_name != NULL) log_warnx("secret for user \"%s\", auth-group \"%s\", " "is too short; it should be at least 12 characters " "long", auth->a_user, auth->a_auth_group->ag_name); else log_warnx("secret for user \"%s\", target \"%s\", " "is too short; it should be at least 12 characters " "long", auth->a_user, auth->a_auth_group->ag_target->t_name); } if (auth->a_mutual_secret != NULL) { len = strlen(auth->a_mutual_secret); if (len > 16) { if (auth->a_auth_group->ag_name != NULL) log_warnx("mutual secret for user \"%s\", " "auth-group \"%s\", is too long; it should " "be at most 16 characters long", auth->a_user, auth->a_auth_group->ag_name); else log_warnx("mutual secret for user \"%s\", " "target \"%s\", is too long; it should " "be at most 16 characters long", auth->a_user, auth->a_auth_group->ag_target->t_name); } if (len < 12) { if (auth->a_auth_group->ag_name != NULL) log_warnx("mutual secret for user \"%s\", " "auth-group \"%s\", is too short; it " "should be at least 12 characters long", auth->a_user, auth->a_auth_group->ag_name); else log_warnx("mutual secret for user \"%s\", " "target \"%s\", is too short; it should be " "at least 12 characters long", auth->a_user, auth->a_auth_group->ag_target->t_name); } } } const struct auth * auth_new_chap(struct auth_group *ag, const char *user, const char *secret) { struct auth *auth; if (ag->ag_type == AG_TYPE_UNKNOWN) ag->ag_type = AG_TYPE_CHAP; if (ag->ag_type != AG_TYPE_CHAP) { if (ag->ag_name != NULL) log_warnx("cannot mix \"chap\" authentication with " "other types for auth-group \"%s\"", ag->ag_name); else log_warnx("cannot mix \"chap\" authentication with " "other types for target \"%s\"", ag->ag_target->t_name); return (NULL); } auth = auth_new(ag); auth->a_user = checked_strdup(user); auth->a_secret = checked_strdup(secret); auth_check_secret_length(auth); return (auth); } const struct auth * auth_new_chap_mutual(struct auth_group *ag, const char *user, const char *secret, const char *user2, const char *secret2) { struct auth *auth; if (ag->ag_type == AG_TYPE_UNKNOWN) ag->ag_type = AG_TYPE_CHAP_MUTUAL; if (ag->ag_type != AG_TYPE_CHAP_MUTUAL) { if (ag->ag_name != NULL) log_warnx("cannot mix \"chap-mutual\" authentication " "with other types for auth-group \"%s\"", ag->ag_name); else log_warnx("cannot mix \"chap-mutual\" authentication " "with other types for target \"%s\"", ag->ag_target->t_name); return (NULL); } auth = auth_new(ag); auth->a_user = checked_strdup(user); auth->a_secret = checked_strdup(secret); auth->a_mutual_user = checked_strdup(user2); auth->a_mutual_secret = checked_strdup(secret2); auth_check_secret_length(auth); return (auth); } const struct auth_name * auth_name_new(struct auth_group *ag, const char *name) { struct auth_name *an; an = calloc(1, sizeof(*an)); if (an == NULL) log_err(1, "calloc"); an->an_auth_group = ag; an->an_initator_name = checked_strdup(name); TAILQ_INSERT_TAIL(&ag->ag_names, an, an_next); return (an); } static void auth_name_delete(struct auth_name *an) { TAILQ_REMOVE(&an->an_auth_group->ag_names, an, an_next); free(an->an_initator_name); free(an); } bool auth_name_defined(const struct auth_group *ag) { if (TAILQ_EMPTY(&ag->ag_names)) return (false); return (true); } const struct auth_name * auth_name_find(const struct auth_group *ag, const char *name) { const struct auth_name *auth_name; TAILQ_FOREACH(auth_name, &ag->ag_names, an_next) { if (strcmp(auth_name->an_initator_name, name) == 0) return (auth_name); } return (NULL); } int auth_name_check(const struct auth_group *ag, const char *initiator_name) { if (!auth_name_defined(ag)) return (0); if (auth_name_find(ag, initiator_name) == NULL) return (1); return (0); } const struct auth_portal * auth_portal_new(struct auth_group *ag, const char *portal) { struct auth_portal *ap; char *net, *mask, *str, *tmp; int len, dm, m; ap = calloc(1, sizeof(*ap)); if (ap == NULL) log_err(1, "calloc"); ap->ap_auth_group = ag; ap->ap_initator_portal = checked_strdup(portal); mask = str = checked_strdup(portal); net = strsep(&mask, "/"); if (net[0] == '[') net++; len = strlen(net); if (len == 0) goto error; if (net[len - 1] == ']') net[len - 1] = 0; if (strchr(net, ':') != NULL) { struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&ap->ap_sa; sin6->sin6_len = sizeof(*sin6); sin6->sin6_family = AF_INET6; if (inet_pton(AF_INET6, net, &sin6->sin6_addr) <= 0) goto error; dm = 128; } else { struct sockaddr_in *sin = (struct sockaddr_in *)&ap->ap_sa; sin->sin_len = sizeof(*sin); sin->sin_family = AF_INET; if (inet_pton(AF_INET, net, &sin->sin_addr) <= 0) goto error; dm = 32; } if (mask != NULL) { m = strtol(mask, &tmp, 0); if (m < 0 || m > dm || tmp[0] != 0) goto error; } else m = dm; ap->ap_mask = m; free(str); TAILQ_INSERT_TAIL(&ag->ag_portals, ap, ap_next); return (ap); error: free(str); free(ap); log_warnx("incorrect initiator portal \"%s\"", portal); return (NULL); } static void auth_portal_delete(struct auth_portal *ap) { TAILQ_REMOVE(&ap->ap_auth_group->ag_portals, ap, ap_next); free(ap->ap_initator_portal); free(ap); } bool auth_portal_defined(const struct auth_group *ag) { if (TAILQ_EMPTY(&ag->ag_portals)) return (false); return (true); } const struct auth_portal * auth_portal_find(const struct auth_group *ag, const struct sockaddr_storage *ss) { const struct auth_portal *ap; const uint8_t *a, *b; int i; uint8_t bmask; TAILQ_FOREACH(ap, &ag->ag_portals, ap_next) { if (ap->ap_sa.ss_family != ss->ss_family) continue; if (ss->ss_family == AF_INET) { a = (const uint8_t *) &((const struct sockaddr_in *)ss)->sin_addr; b = (const uint8_t *) &((const struct sockaddr_in *)&ap->ap_sa)->sin_addr; } else { a = (const uint8_t *) &((const struct sockaddr_in6 *)ss)->sin6_addr; b = (const uint8_t *) &((const struct sockaddr_in6 *)&ap->ap_sa)->sin6_addr; } for (i = 0; i < ap->ap_mask / 8; i++) { if (a[i] != b[i]) goto next; } if (ap->ap_mask % 8) { bmask = 0xff << (8 - (ap->ap_mask % 8)); if ((a[i] & bmask) != (b[i] & bmask)) goto next; } return (ap); next: ; } return (NULL); } int auth_portal_check(const struct auth_group *ag, const struct sockaddr_storage *sa) { if (!auth_portal_defined(ag)) return (0); if (auth_portal_find(ag, sa) == NULL) return (1); return (0); } struct auth_group * auth_group_new(struct conf *conf, const char *name) { struct auth_group *ag; if (name != NULL) { ag = auth_group_find(conf, name); if (ag != NULL) { log_warnx("duplicated auth-group \"%s\"", name); return (NULL); } } ag = calloc(1, sizeof(*ag)); if (ag == NULL) log_err(1, "calloc"); if (name != NULL) ag->ag_name = checked_strdup(name); TAILQ_INIT(&ag->ag_auths); TAILQ_INIT(&ag->ag_names); TAILQ_INIT(&ag->ag_portals); ag->ag_conf = conf; TAILQ_INSERT_TAIL(&conf->conf_auth_groups, ag, ag_next); return (ag); } void auth_group_delete(struct auth_group *ag) { struct auth *auth, *auth_tmp; struct auth_name *auth_name, *auth_name_tmp; struct auth_portal *auth_portal, *auth_portal_tmp; TAILQ_REMOVE(&ag->ag_conf->conf_auth_groups, ag, ag_next); TAILQ_FOREACH_SAFE(auth, &ag->ag_auths, a_next, auth_tmp) auth_delete(auth); TAILQ_FOREACH_SAFE(auth_name, &ag->ag_names, an_next, auth_name_tmp) auth_name_delete(auth_name); TAILQ_FOREACH_SAFE(auth_portal, &ag->ag_portals, ap_next, auth_portal_tmp) auth_portal_delete(auth_portal); free(ag->ag_name); free(ag); } struct auth_group * auth_group_find(const struct conf *conf, const char *name) { struct auth_group *ag; TAILQ_FOREACH(ag, &conf->conf_auth_groups, ag_next) { if (ag->ag_name != NULL && strcmp(ag->ag_name, name) == 0) return (ag); } return (NULL); } int auth_group_set_type(struct auth_group *ag, const char *str) { int type; if (strcmp(str, "none") == 0) { type = AG_TYPE_NO_AUTHENTICATION; } else if (strcmp(str, "deny") == 0) { type = AG_TYPE_DENY; } else if (strcmp(str, "chap") == 0) { type = AG_TYPE_CHAP; } else if (strcmp(str, "chap-mutual") == 0) { type = AG_TYPE_CHAP_MUTUAL; } else { if (ag->ag_name != NULL) log_warnx("invalid auth-type \"%s\" for auth-group " "\"%s\"", str, ag->ag_name); else log_warnx("invalid auth-type \"%s\" for target " "\"%s\"", str, ag->ag_target->t_name); return (1); } if (ag->ag_type != AG_TYPE_UNKNOWN && ag->ag_type != type) { if (ag->ag_name != NULL) { log_warnx("cannot set auth-type to \"%s\" for " "auth-group \"%s\"; already has a different " "type", str, ag->ag_name); } else { log_warnx("cannot set auth-type to \"%s\" for target " "\"%s\"; already has a different type", str, ag->ag_target->t_name); } return (1); } ag->ag_type = type; return (0); } static struct portal * portal_new(struct portal_group *pg) { struct portal *portal; portal = calloc(1, sizeof(*portal)); if (portal == NULL) log_err(1, "calloc"); TAILQ_INIT(&portal->p_targets); portal->p_portal_group = pg; TAILQ_INSERT_TAIL(&pg->pg_portals, portal, p_next); return (portal); } static void portal_delete(struct portal *portal) { TAILQ_REMOVE(&portal->p_portal_group->pg_portals, portal, p_next); if (portal->p_ai != NULL) freeaddrinfo(portal->p_ai); free(portal->p_listen); free(portal); } struct portal_group * portal_group_new(struct conf *conf, const char *name) { struct portal_group *pg; pg = portal_group_find(conf, name); if (pg != NULL) { log_warnx("duplicated portal-group \"%s\"", name); return (NULL); } pg = calloc(1, sizeof(*pg)); if (pg == NULL) log_err(1, "calloc"); pg->pg_name = checked_strdup(name); TAILQ_INIT(&pg->pg_options); TAILQ_INIT(&pg->pg_portals); TAILQ_INIT(&pg->pg_ports); pg->pg_conf = conf; pg->pg_tag = 0; /* Assigned later in conf_apply(). */ TAILQ_INSERT_TAIL(&conf->conf_portal_groups, pg, pg_next); return (pg); } void portal_group_delete(struct portal_group *pg) { struct portal *portal, *tmp; struct port *port, *tport; struct option *o, *otmp; TAILQ_FOREACH_SAFE(port, &pg->pg_ports, p_pgs, tport) port_delete(port); TAILQ_REMOVE(&pg->pg_conf->conf_portal_groups, pg, pg_next); TAILQ_FOREACH_SAFE(portal, &pg->pg_portals, p_next, tmp) portal_delete(portal); TAILQ_FOREACH_SAFE(o, &pg->pg_options, o_next, otmp) option_delete(&pg->pg_options, o); free(pg->pg_name); free(pg->pg_offload); free(pg->pg_redirection); free(pg); } struct portal_group * portal_group_find(const struct conf *conf, const char *name) { struct portal_group *pg; TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { if (strcmp(pg->pg_name, name) == 0) return (pg); } return (NULL); } static int parse_addr_port(char *arg, const char *def_port, struct addrinfo **ai) { struct addrinfo hints; char *str, *addr, *ch; const char *port; int error, colons = 0; str = arg = strdup(arg); if (arg[0] == '[') { /* * IPv6 address in square brackets, perhaps with port. */ arg++; addr = strsep(&arg, "]"); if (arg == NULL) { free(str); return (1); } if (arg[0] == '\0') { port = def_port; } else if (arg[0] == ':') { port = arg + 1; } else { free(str); return (1); } } else { /* * Either IPv6 address without brackets - and without * a port - or IPv4 address. Just count the colons. */ for (ch = arg; *ch != '\0'; ch++) { if (*ch == ':') colons++; } if (colons > 1) { addr = arg; port = def_port; } else { addr = strsep(&arg, ":"); if (arg == NULL) port = def_port; else port = arg; } } memset(&hints, 0, sizeof(hints)); hints.ai_family = PF_UNSPEC; hints.ai_socktype = SOCK_STREAM; hints.ai_flags = AI_PASSIVE; error = getaddrinfo(addr, port, &hints, ai); free(str); return ((error != 0) ? 1 : 0); } int portal_group_add_listen(struct portal_group *pg, const char *value, bool iser) { struct portal *portal; portal = portal_new(pg); portal->p_listen = checked_strdup(value); portal->p_iser = iser; if (parse_addr_port(portal->p_listen, "3260", &portal->p_ai)) { log_warnx("invalid listen address %s", portal->p_listen); portal_delete(portal); return (1); } /* * XXX: getaddrinfo(3) may return multiple addresses; we should turn * those into multiple portals. */ return (0); } int isns_new(struct conf *conf, const char *addr) { struct isns *isns; isns = calloc(1, sizeof(*isns)); if (isns == NULL) log_err(1, "calloc"); isns->i_conf = conf; TAILQ_INSERT_TAIL(&conf->conf_isns, isns, i_next); isns->i_addr = checked_strdup(addr); if (parse_addr_port(isns->i_addr, "3205", &isns->i_ai)) { log_warnx("invalid iSNS address %s", isns->i_addr); isns_delete(isns); return (1); } /* * XXX: getaddrinfo(3) may return multiple addresses; we should turn * those into multiple servers. */ return (0); } void isns_delete(struct isns *isns) { TAILQ_REMOVE(&isns->i_conf->conf_isns, isns, i_next); free(isns->i_addr); if (isns->i_ai != NULL) freeaddrinfo(isns->i_ai); free(isns); } static int isns_do_connect(struct isns *isns) { int s; s = socket(isns->i_ai->ai_family, isns->i_ai->ai_socktype, isns->i_ai->ai_protocol); if (s < 0) { log_warn("socket(2) failed for %s", isns->i_addr); return (-1); } if (connect(s, isns->i_ai->ai_addr, isns->i_ai->ai_addrlen)) { log_warn("connect(2) failed for %s", isns->i_addr); close(s); return (-1); } return(s); } static int isns_do_register(struct isns *isns, int s, const char *hostname) { struct conf *conf = isns->i_conf; struct target *target; struct portal *portal; struct portal_group *pg; struct port *port; struct isns_req *req; int res = 0; uint32_t error; req = isns_req_create(ISNS_FUNC_DEVATTRREG, ISNS_FLAG_CLIENT); isns_req_add_str(req, 32, TAILQ_FIRST(&conf->conf_targets)->t_name); isns_req_add_delim(req); isns_req_add_str(req, 1, hostname); isns_req_add_32(req, 2, 2); /* 2 -- iSCSI */ isns_req_add_32(req, 6, conf->conf_isns_period); TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { if (pg->pg_unassigned) continue; TAILQ_FOREACH(portal, &pg->pg_portals, p_next) { isns_req_add_addr(req, 16, portal->p_ai); isns_req_add_port(req, 17, portal->p_ai); } } TAILQ_FOREACH(target, &conf->conf_targets, t_next) { isns_req_add_str(req, 32, target->t_name); isns_req_add_32(req, 33, 1); /* 1 -- Target*/ if (target->t_alias != NULL) isns_req_add_str(req, 34, target->t_alias); TAILQ_FOREACH(port, &target->t_ports, p_ts) { if ((pg = port->p_portal_group) == NULL) continue; isns_req_add_32(req, 51, pg->pg_tag); TAILQ_FOREACH(portal, &pg->pg_portals, p_next) { isns_req_add_addr(req, 49, portal->p_ai); isns_req_add_port(req, 50, portal->p_ai); } } } res = isns_req_send(s, req); if (res < 0) { log_warn("send(2) failed for %s", isns->i_addr); goto quit; } res = isns_req_receive(s, req); if (res < 0) { log_warn("receive(2) failed for %s", isns->i_addr); goto quit; } error = isns_req_get_status(req); if (error != 0) { log_warnx("iSNS register error %d for %s", error, isns->i_addr); res = -1; } quit: isns_req_free(req); return (res); } static int isns_do_check(struct isns *isns, int s, const char *hostname) { struct conf *conf = isns->i_conf; struct isns_req *req; int res = 0; uint32_t error; req = isns_req_create(ISNS_FUNC_DEVATTRQRY, ISNS_FLAG_CLIENT); isns_req_add_str(req, 32, TAILQ_FIRST(&conf->conf_targets)->t_name); isns_req_add_str(req, 1, hostname); isns_req_add_delim(req); isns_req_add(req, 2, 0, NULL); res = isns_req_send(s, req); if (res < 0) { log_warn("send(2) failed for %s", isns->i_addr); goto quit; } res = isns_req_receive(s, req); if (res < 0) { log_warn("receive(2) failed for %s", isns->i_addr); goto quit; } error = isns_req_get_status(req); if (error != 0) { log_warnx("iSNS check error %d for %s", error, isns->i_addr); res = -1; } quit: isns_req_free(req); return (res); } static int isns_do_deregister(struct isns *isns, int s, const char *hostname) { struct conf *conf = isns->i_conf; struct isns_req *req; int res = 0; uint32_t error; req = isns_req_create(ISNS_FUNC_DEVDEREG, ISNS_FLAG_CLIENT); isns_req_add_str(req, 32, TAILQ_FIRST(&conf->conf_targets)->t_name); isns_req_add_delim(req); isns_req_add_str(req, 1, hostname); res = isns_req_send(s, req); if (res < 0) { log_warn("send(2) failed for %s", isns->i_addr); goto quit; } res = isns_req_receive(s, req); if (res < 0) { log_warn("receive(2) failed for %s", isns->i_addr); goto quit; } error = isns_req_get_status(req); if (error != 0) { log_warnx("iSNS deregister error %d for %s", error, isns->i_addr); res = -1; } quit: isns_req_free(req); return (res); } void isns_register(struct isns *isns, struct isns *oldisns) { struct conf *conf = isns->i_conf; int s; char hostname[256]; if (TAILQ_EMPTY(&conf->conf_targets) || TAILQ_EMPTY(&conf->conf_portal_groups)) return; set_timeout(conf->conf_isns_timeout, false); s = isns_do_connect(isns); if (s < 0) { set_timeout(0, false); return; } gethostname(hostname, sizeof(hostname)); if (oldisns == NULL || TAILQ_EMPTY(&oldisns->i_conf->conf_targets)) oldisns = isns; isns_do_deregister(oldisns, s, hostname); isns_do_register(isns, s, hostname); close(s); set_timeout(0, false); } void isns_check(struct isns *isns) { struct conf *conf = isns->i_conf; int s, res; char hostname[256]; if (TAILQ_EMPTY(&conf->conf_targets) || TAILQ_EMPTY(&conf->conf_portal_groups)) return; set_timeout(conf->conf_isns_timeout, false); s = isns_do_connect(isns); if (s < 0) { set_timeout(0, false); return; } gethostname(hostname, sizeof(hostname)); res = isns_do_check(isns, s, hostname); if (res < 0) { isns_do_deregister(isns, s, hostname); isns_do_register(isns, s, hostname); } close(s); set_timeout(0, false); } void isns_deregister(struct isns *isns) { struct conf *conf = isns->i_conf; int s; char hostname[256]; if (TAILQ_EMPTY(&conf->conf_targets) || TAILQ_EMPTY(&conf->conf_portal_groups)) return; set_timeout(conf->conf_isns_timeout, false); s = isns_do_connect(isns); if (s < 0) return; gethostname(hostname, sizeof(hostname)); isns_do_deregister(isns, s, hostname); close(s); set_timeout(0, false); } int portal_group_set_filter(struct portal_group *pg, const char *str) { int filter; if (strcmp(str, "none") == 0) { filter = PG_FILTER_NONE; } else if (strcmp(str, "portal") == 0) { filter = PG_FILTER_PORTAL; } else if (strcmp(str, "portal-name") == 0) { filter = PG_FILTER_PORTAL_NAME; } else if (strcmp(str, "portal-name-auth") == 0) { filter = PG_FILTER_PORTAL_NAME_AUTH; } else { log_warnx("invalid discovery-filter \"%s\" for portal-group " "\"%s\"; valid values are \"none\", \"portal\", " "\"portal-name\", and \"portal-name-auth\"", str, pg->pg_name); return (1); } if (pg->pg_discovery_filter != PG_FILTER_UNKNOWN && pg->pg_discovery_filter != filter) { log_warnx("cannot set discovery-filter to \"%s\" for " "portal-group \"%s\"; already has a different " "value", str, pg->pg_name); return (1); } pg->pg_discovery_filter = filter; return (0); } int portal_group_set_offload(struct portal_group *pg, const char *offload) { if (pg->pg_offload != NULL) { log_warnx("cannot set offload to \"%s\" for " "portal-group \"%s\"; already defined", offload, pg->pg_name); return (1); } pg->pg_offload = checked_strdup(offload); return (0); } int portal_group_set_redirection(struct portal_group *pg, const char *addr) { if (pg->pg_redirection != NULL) { log_warnx("cannot set redirection to \"%s\" for " "portal-group \"%s\"; already defined", addr, pg->pg_name); return (1); } pg->pg_redirection = checked_strdup(addr); return (0); } static bool valid_hex(const char ch) { switch (ch) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': case 'a': case 'A': case 'b': case 'B': case 'c': case 'C': case 'd': case 'D': case 'e': case 'E': case 'f': case 'F': return (true); default: return (false); } } bool valid_iscsi_name(const char *name) { int i; if (strlen(name) >= MAX_NAME_LEN) { log_warnx("overlong name for target \"%s\"; max length allowed " "by iSCSI specification is %d characters", name, MAX_NAME_LEN); return (false); } /* * In the cases below, we don't return an error, just in case the admin * was right, and we're wrong. */ if (strncasecmp(name, "iqn.", strlen("iqn.")) == 0) { for (i = strlen("iqn."); name[i] != '\0'; i++) { /* * XXX: We should verify UTF-8 normalisation, as defined * by 3.2.6.2: iSCSI Name Encoding. */ if (isalnum(name[i])) continue; if (name[i] == '-' || name[i] == '.' || name[i] == ':') continue; log_warnx("invalid character \"%c\" in target name " "\"%s\"; allowed characters are letters, digits, " "'-', '.', and ':'", name[i], name); break; } /* * XXX: Check more stuff: valid date and a valid reversed domain. */ } else if (strncasecmp(name, "eui.", strlen("eui.")) == 0) { if (strlen(name) != strlen("eui.") + 16) log_warnx("invalid target name \"%s\"; the \"eui.\" " "should be followed by exactly 16 hexadecimal " "digits", name); for (i = strlen("eui."); name[i] != '\0'; i++) { if (!valid_hex(name[i])) { log_warnx("invalid character \"%c\" in target " "name \"%s\"; allowed characters are 1-9 " "and A-F", name[i], name); break; } } } else if (strncasecmp(name, "naa.", strlen("naa.")) == 0) { if (strlen(name) > strlen("naa.") + 32) log_warnx("invalid target name \"%s\"; the \"naa.\" " "should be followed by at most 32 hexadecimal " "digits", name); for (i = strlen("naa."); name[i] != '\0'; i++) { if (!valid_hex(name[i])) { log_warnx("invalid character \"%c\" in target " "name \"%s\"; allowed characters are 1-9 " "and A-F", name[i], name); break; } } } else { log_warnx("invalid target name \"%s\"; should start with " "either \"iqn.\", \"eui.\", or \"naa.\"", name); } return (true); } struct pport * pport_new(struct conf *conf, const char *name, uint32_t ctl_port) { struct pport *pp; pp = calloc(1, sizeof(*pp)); if (pp == NULL) log_err(1, "calloc"); pp->pp_conf = conf; pp->pp_name = checked_strdup(name); pp->pp_ctl_port = ctl_port; TAILQ_INIT(&pp->pp_ports); TAILQ_INSERT_TAIL(&conf->conf_pports, pp, pp_next); return (pp); } struct pport * pport_find(const struct conf *conf, const char *name) { struct pport *pp; TAILQ_FOREACH(pp, &conf->conf_pports, pp_next) { if (strcasecmp(pp->pp_name, name) == 0) return (pp); } return (NULL); } struct pport * pport_copy(struct pport *pp, struct conf *conf) { struct pport *ppnew; ppnew = pport_new(conf, pp->pp_name, pp->pp_ctl_port); return (ppnew); } void pport_delete(struct pport *pp) { struct port *port, *tport; TAILQ_FOREACH_SAFE(port, &pp->pp_ports, p_ts, tport) port_delete(port); TAILQ_REMOVE(&pp->pp_conf->conf_pports, pp, pp_next); free(pp->pp_name); free(pp); } struct port * port_new(struct conf *conf, struct target *target, struct portal_group *pg) { struct port *port; char *name; int ret; ret = asprintf(&name, "%s-%s", pg->pg_name, target->t_name); if (ret <= 0) log_err(1, "asprintf"); if (port_find(conf, name) != NULL) { log_warnx("duplicate port \"%s\"", name); free(name); return (NULL); } port = calloc(1, sizeof(*port)); if (port == NULL) log_err(1, "calloc"); port->p_conf = conf; port->p_name = name; + port->p_ioctl_port = 0; TAILQ_INSERT_TAIL(&conf->conf_ports, port, p_next); TAILQ_INSERT_TAIL(&target->t_ports, port, p_ts); port->p_target = target; TAILQ_INSERT_TAIL(&pg->pg_ports, port, p_pgs); port->p_portal_group = pg; return (port); } struct port * +port_new_ioctl(struct conf *conf, struct target *target, int pp, int vp) +{ + struct pport *pport; + struct port *port; + char *pname; + char *name; + int ret; + + ret = asprintf(&pname, "ioctl/%d/%d", pp, vp); + if (ret <= 0) { + log_err(1, "asprintf"); + return (NULL); + } + + pport = pport_find(conf, pname); + if (pport != NULL) { + free(pname); + return (port_new_pp(conf, target, pport)); + } + + ret = asprintf(&name, "%s-%s", pname, target->t_name); + free(pname); + + if (ret <= 0) + log_err(1, "asprintf"); + if (port_find(conf, name) != NULL) { + log_warnx("duplicate port \"%s\"", name); + free(name); + return (NULL); + } + port = calloc(1, sizeof(*port)); + if (port == NULL) + log_err(1, "calloc"); + port->p_conf = conf; + port->p_name = name; + port->p_ioctl_port = 1; + port->p_ioctl_pp = pp; + port->p_ioctl_vp = vp; + TAILQ_INSERT_TAIL(&conf->conf_ports, port, p_next); + TAILQ_INSERT_TAIL(&target->t_ports, port, p_ts); + port->p_target = target; + return (port); +} + +struct port * port_new_pp(struct conf *conf, struct target *target, struct pport *pp) { struct port *port; char *name; int ret; ret = asprintf(&name, "%s-%s", pp->pp_name, target->t_name); if (ret <= 0) log_err(1, "asprintf"); if (port_find(conf, name) != NULL) { log_warnx("duplicate port \"%s\"", name); free(name); return (NULL); } port = calloc(1, sizeof(*port)); if (port == NULL) log_err(1, "calloc"); port->p_conf = conf; port->p_name = name; TAILQ_INSERT_TAIL(&conf->conf_ports, port, p_next); TAILQ_INSERT_TAIL(&target->t_ports, port, p_ts); port->p_target = target; TAILQ_INSERT_TAIL(&pp->pp_ports, port, p_pps); port->p_pport = pp; return (port); } struct port * port_find(const struct conf *conf, const char *name) { struct port *port; TAILQ_FOREACH(port, &conf->conf_ports, p_next) { if (strcasecmp(port->p_name, name) == 0) return (port); } return (NULL); } struct port * port_find_in_pg(const struct portal_group *pg, const char *target) { struct port *port; TAILQ_FOREACH(port, &pg->pg_ports, p_pgs) { if (strcasecmp(port->p_target->t_name, target) == 0) return (port); } return (NULL); } void port_delete(struct port *port) { if (port->p_portal_group) TAILQ_REMOVE(&port->p_portal_group->pg_ports, port, p_pgs); if (port->p_pport) TAILQ_REMOVE(&port->p_pport->pp_ports, port, p_pps); if (port->p_target) TAILQ_REMOVE(&port->p_target->t_ports, port, p_ts); TAILQ_REMOVE(&port->p_conf->conf_ports, port, p_next); free(port->p_name); free(port); } int port_is_dummy(struct port *port) { if (port->p_portal_group) { if (port->p_portal_group->pg_foreign) return (1); if (TAILQ_EMPTY(&port->p_portal_group->pg_portals)) return (1); } return (0); } struct target * target_new(struct conf *conf, const char *name) { struct target *targ; int i, len; targ = target_find(conf, name); if (targ != NULL) { log_warnx("duplicated target \"%s\"", name); return (NULL); } if (valid_iscsi_name(name) == false) { log_warnx("target name \"%s\" is invalid", name); return (NULL); } targ = calloc(1, sizeof(*targ)); if (targ == NULL) log_err(1, "calloc"); targ->t_name = checked_strdup(name); /* * RFC 3722 requires us to normalize the name to lowercase. */ len = strlen(name); for (i = 0; i < len; i++) targ->t_name[i] = tolower(targ->t_name[i]); targ->t_conf = conf; TAILQ_INIT(&targ->t_ports); TAILQ_INSERT_TAIL(&conf->conf_targets, targ, t_next); return (targ); } void target_delete(struct target *targ) { struct port *port, *tport; TAILQ_FOREACH_SAFE(port, &targ->t_ports, p_ts, tport) port_delete(port); TAILQ_REMOVE(&targ->t_conf->conf_targets, targ, t_next); free(targ->t_name); free(targ->t_redirection); free(targ); } struct target * target_find(struct conf *conf, const char *name) { struct target *targ; TAILQ_FOREACH(targ, &conf->conf_targets, t_next) { if (strcasecmp(targ->t_name, name) == 0) return (targ); } return (NULL); } int target_set_redirection(struct target *target, const char *addr) { if (target->t_redirection != NULL) { log_warnx("cannot set redirection to \"%s\" for " "target \"%s\"; already defined", addr, target->t_name); return (1); } target->t_redirection = checked_strdup(addr); return (0); } struct lun * lun_new(struct conf *conf, const char *name) { struct lun *lun; lun = lun_find(conf, name); if (lun != NULL) { log_warnx("duplicated lun \"%s\"", name); return (NULL); } lun = calloc(1, sizeof(*lun)); if (lun == NULL) log_err(1, "calloc"); lun->l_conf = conf; lun->l_name = checked_strdup(name); TAILQ_INIT(&lun->l_options); TAILQ_INSERT_TAIL(&conf->conf_luns, lun, l_next); lun->l_ctl_lun = -1; return (lun); } void lun_delete(struct lun *lun) { struct target *targ; struct option *o, *tmp; int i; TAILQ_FOREACH(targ, &lun->l_conf->conf_targets, t_next) { for (i = 0; i < MAX_LUNS; i++) { if (targ->t_luns[i] == lun) targ->t_luns[i] = NULL; } } TAILQ_REMOVE(&lun->l_conf->conf_luns, lun, l_next); TAILQ_FOREACH_SAFE(o, &lun->l_options, o_next, tmp) option_delete(&lun->l_options, o); free(lun->l_name); free(lun->l_backend); free(lun->l_device_id); free(lun->l_path); free(lun->l_scsiname); free(lun->l_serial); free(lun); } struct lun * lun_find(const struct conf *conf, const char *name) { struct lun *lun; TAILQ_FOREACH(lun, &conf->conf_luns, l_next) { if (strcmp(lun->l_name, name) == 0) return (lun); } return (NULL); } void lun_set_backend(struct lun *lun, const char *value) { free(lun->l_backend); lun->l_backend = checked_strdup(value); } void lun_set_blocksize(struct lun *lun, size_t value) { lun->l_blocksize = value; } void lun_set_device_type(struct lun *lun, uint8_t value) { lun->l_device_type = value; } void lun_set_device_id(struct lun *lun, const char *value) { free(lun->l_device_id); lun->l_device_id = checked_strdup(value); } void lun_set_path(struct lun *lun, const char *value) { free(lun->l_path); lun->l_path = checked_strdup(value); } void lun_set_scsiname(struct lun *lun, const char *value) { free(lun->l_scsiname); lun->l_scsiname = checked_strdup(value); } void lun_set_serial(struct lun *lun, const char *value) { free(lun->l_serial); lun->l_serial = checked_strdup(value); } void lun_set_size(struct lun *lun, size_t value) { lun->l_size = value; } void lun_set_ctl_lun(struct lun *lun, uint32_t value) { lun->l_ctl_lun = value; } struct option * option_new(struct options *options, const char *name, const char *value) { struct option *o; o = option_find(options, name); if (o != NULL) { log_warnx("duplicated option \"%s\"", name); return (NULL); } o = calloc(1, sizeof(*o)); if (o == NULL) log_err(1, "calloc"); o->o_name = checked_strdup(name); o->o_value = checked_strdup(value); TAILQ_INSERT_TAIL(options, o, o_next); return (o); } void option_delete(struct options *options, struct option *o) { TAILQ_REMOVE(options, o, o_next); free(o->o_name); free(o->o_value); free(o); } struct option * option_find(const struct options *options, const char *name) { struct option *o; TAILQ_FOREACH(o, options, o_next) { if (strcmp(o->o_name, name) == 0) return (o); } return (NULL); } void option_set(struct option *o, const char *value) { free(o->o_value); o->o_value = checked_strdup(value); } static struct connection * connection_new(struct portal *portal, int fd, const char *host, const struct sockaddr *client_sa) { struct connection *conn; conn = calloc(1, sizeof(*conn)); if (conn == NULL) log_err(1, "calloc"); conn->conn_portal = portal; conn->conn_socket = fd; conn->conn_initiator_addr = checked_strdup(host); memcpy(&conn->conn_initiator_sa, client_sa, client_sa->sa_len); /* * Default values, from RFC 3720, section 12. */ conn->conn_max_recv_data_segment_length = 8192; conn->conn_max_send_data_segment_length = 8192; conn->conn_max_burst_length = 262144; conn->conn_first_burst_length = 65536; conn->conn_immediate_data = true; return (conn); } #if 0 static void conf_print(struct conf *conf) { struct auth_group *ag; struct auth *auth; struct auth_name *auth_name; struct auth_portal *auth_portal; struct portal_group *pg; struct portal *portal; struct target *targ; struct lun *lun; struct option *o; TAILQ_FOREACH(ag, &conf->conf_auth_groups, ag_next) { fprintf(stderr, "auth-group %s {\n", ag->ag_name); TAILQ_FOREACH(auth, &ag->ag_auths, a_next) fprintf(stderr, "\t chap-mutual %s %s %s %s\n", auth->a_user, auth->a_secret, auth->a_mutual_user, auth->a_mutual_secret); TAILQ_FOREACH(auth_name, &ag->ag_names, an_next) fprintf(stderr, "\t initiator-name %s\n", auth_name->an_initator_name); - TAILQ_FOREACH(auth_portal, &ag->ag_portals, an_next) + TAILQ_FOREACH(auth_portal, &ag->ag_portals, ap_next) fprintf(stderr, "\t initiator-portal %s\n", - auth_portal->an_initator_portal); + auth_portal->ap_initator_portal); fprintf(stderr, "}\n"); } TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { fprintf(stderr, "portal-group %s {\n", pg->pg_name); TAILQ_FOREACH(portal, &pg->pg_portals, p_next) fprintf(stderr, "\t listen %s\n", portal->p_listen); fprintf(stderr, "}\n"); } TAILQ_FOREACH(lun, &conf->conf_luns, l_next) { fprintf(stderr, "\tlun %s {\n", lun->l_name); fprintf(stderr, "\t\tpath %s\n", lun->l_path); TAILQ_FOREACH(o, &lun->l_options, o_next) fprintf(stderr, "\t\toption %s %s\n", - lo->o_name, lo->o_value); + o->o_name, o->o_value); fprintf(stderr, "\t}\n"); } TAILQ_FOREACH(targ, &conf->conf_targets, t_next) { fprintf(stderr, "target %s {\n", targ->t_name); if (targ->t_alias != NULL) fprintf(stderr, "\t alias %s\n", targ->t_alias); fprintf(stderr, "}\n"); } } #endif static int conf_verify_lun(struct lun *lun) { const struct lun *lun2; if (lun->l_backend == NULL) lun_set_backend(lun, "block"); if (strcmp(lun->l_backend, "block") == 0) { if (lun->l_path == NULL) { log_warnx("missing path for lun \"%s\"", lun->l_name); return (1); } } else if (strcmp(lun->l_backend, "ramdisk") == 0) { if (lun->l_size == 0) { log_warnx("missing size for ramdisk-backed lun \"%s\"", lun->l_name); return (1); } if (lun->l_path != NULL) { log_warnx("path must not be specified " "for ramdisk-backed lun \"%s\"", lun->l_name); return (1); } } if (lun->l_blocksize == 0) { if (lun->l_device_type == 5) lun_set_blocksize(lun, DEFAULT_CD_BLOCKSIZE); else lun_set_blocksize(lun, DEFAULT_BLOCKSIZE); } else if (lun->l_blocksize < 0) { log_warnx("invalid blocksize for lun \"%s\"; " "must be larger than 0", lun->l_name); return (1); } if (lun->l_size != 0 && lun->l_size % lun->l_blocksize != 0) { log_warnx("invalid size for lun \"%s\"; " "must be multiple of blocksize", lun->l_name); return (1); } TAILQ_FOREACH(lun2, &lun->l_conf->conf_luns, l_next) { if (lun == lun2) continue; if (lun->l_path != NULL && lun2->l_path != NULL && strcmp(lun->l_path, lun2->l_path) == 0) { log_debugx("WARNING: path \"%s\" duplicated " "between lun \"%s\", and " "lun \"%s\"", lun->l_path, lun->l_name, lun2->l_name); } } return (0); } int conf_verify(struct conf *conf) { struct auth_group *ag; struct portal_group *pg; struct port *port; struct target *targ; struct lun *lun; bool found; int error, i; if (conf->conf_pidfile_path == NULL) conf->conf_pidfile_path = checked_strdup(DEFAULT_PIDFILE); TAILQ_FOREACH(lun, &conf->conf_luns, l_next) { error = conf_verify_lun(lun); if (error != 0) return (error); } TAILQ_FOREACH(targ, &conf->conf_targets, t_next) { if (targ->t_auth_group == NULL) { targ->t_auth_group = auth_group_find(conf, "default"); assert(targ->t_auth_group != NULL); } if (TAILQ_EMPTY(&targ->t_ports)) { pg = portal_group_find(conf, "default"); assert(pg != NULL); port_new(conf, targ, pg); } found = false; for (i = 0; i < MAX_LUNS; i++) { if (targ->t_luns[i] != NULL) found = true; } if (!found && targ->t_redirection == NULL) { log_warnx("no LUNs defined for target \"%s\"", targ->t_name); } if (found && targ->t_redirection != NULL) { log_debugx("target \"%s\" contains luns, " " but configured for redirection", targ->t_name); } } TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { assert(pg->pg_name != NULL); if (pg->pg_discovery_auth_group == NULL) { pg->pg_discovery_auth_group = auth_group_find(conf, "default"); assert(pg->pg_discovery_auth_group != NULL); } if (pg->pg_discovery_filter == PG_FILTER_UNKNOWN) pg->pg_discovery_filter = PG_FILTER_NONE; if (pg->pg_redirection != NULL) { if (!TAILQ_EMPTY(&pg->pg_ports)) { log_debugx("portal-group \"%s\" assigned " "to target, but configured " "for redirection", pg->pg_name); } pg->pg_unassigned = false; } else if (!TAILQ_EMPTY(&pg->pg_ports)) { pg->pg_unassigned = false; } else { if (strcmp(pg->pg_name, "default") != 0) log_warnx("portal-group \"%s\" not assigned " "to any target", pg->pg_name); pg->pg_unassigned = true; } } TAILQ_FOREACH(ag, &conf->conf_auth_groups, ag_next) { if (ag->ag_name == NULL) assert(ag->ag_target != NULL); else assert(ag->ag_target == NULL); found = false; TAILQ_FOREACH(targ, &conf->conf_targets, t_next) { if (targ->t_auth_group == ag) { found = true; break; } } TAILQ_FOREACH(port, &conf->conf_ports, p_next) { if (port->p_auth_group == ag) { found = true; break; } } TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { if (pg->pg_discovery_auth_group == ag) { found = true; break; } } if (!found && ag->ag_name != NULL && strcmp(ag->ag_name, "default") != 0 && strcmp(ag->ag_name, "no-authentication") != 0 && strcmp(ag->ag_name, "no-access") != 0) { log_warnx("auth-group \"%s\" not assigned " "to any target", ag->ag_name); } } return (0); } static int conf_apply(struct conf *oldconf, struct conf *newconf) { struct lun *oldlun, *newlun, *tmplun; struct portal_group *oldpg, *newpg; struct portal *oldp, *newp; struct port *oldport, *newport, *tmpport; struct isns *oldns, *newns; pid_t otherpid; int changed, cumulated_error = 0, error, sockbuf; int one = 1; if (oldconf->conf_debug != newconf->conf_debug) { log_debugx("changing debug level to %d", newconf->conf_debug); log_init(newconf->conf_debug); } if (oldconf->conf_pidfh != NULL) { assert(oldconf->conf_pidfile_path != NULL); if (newconf->conf_pidfile_path != NULL && strcmp(oldconf->conf_pidfile_path, newconf->conf_pidfile_path) == 0) { newconf->conf_pidfh = oldconf->conf_pidfh; oldconf->conf_pidfh = NULL; } else { log_debugx("removing pidfile %s", oldconf->conf_pidfile_path); pidfile_remove(oldconf->conf_pidfh); oldconf->conf_pidfh = NULL; } } if (newconf->conf_pidfh == NULL && newconf->conf_pidfile_path != NULL) { log_debugx("opening pidfile %s", newconf->conf_pidfile_path); newconf->conf_pidfh = pidfile_open(newconf->conf_pidfile_path, 0600, &otherpid); if (newconf->conf_pidfh == NULL) { if (errno == EEXIST) log_errx(1, "daemon already running, pid: %jd.", (intmax_t)otherpid); log_err(1, "cannot open or create pidfile \"%s\"", newconf->conf_pidfile_path); } } /* * Go through the new portal groups, assigning tags or preserving old. */ TAILQ_FOREACH(newpg, &newconf->conf_portal_groups, pg_next) { if (newpg->pg_tag != 0) continue; oldpg = portal_group_find(oldconf, newpg->pg_name); if (oldpg != NULL) newpg->pg_tag = oldpg->pg_tag; else newpg->pg_tag = ++last_portal_group_tag; } /* Deregister on removed iSNS servers. */ TAILQ_FOREACH(oldns, &oldconf->conf_isns, i_next) { TAILQ_FOREACH(newns, &newconf->conf_isns, i_next) { if (strcmp(oldns->i_addr, newns->i_addr) == 0) break; } if (newns == NULL) isns_deregister(oldns); } /* * XXX: If target or lun removal fails, we should somehow "move" * the old lun or target into newconf, so that subsequent * conf_apply() would try to remove them again. That would * be somewhat hairy, though, and lun deletion failures don't * really happen, so leave it as it is for now. */ /* * First, remove any ports present in the old configuration * and missing in the new one. */ TAILQ_FOREACH_SAFE(oldport, &oldconf->conf_ports, p_next, tmpport) { if (port_is_dummy(oldport)) continue; newport = port_find(newconf, oldport->p_name); if (newport != NULL && !port_is_dummy(newport)) continue; log_debugx("removing port \"%s\"", oldport->p_name); error = kernel_port_remove(oldport); if (error != 0) { log_warnx("failed to remove port %s", oldport->p_name); /* * XXX: Uncomment after fixing the root cause. * * cumulated_error++; */ } } /* * Second, remove any LUNs present in the old configuration * and missing in the new one. */ TAILQ_FOREACH_SAFE(oldlun, &oldconf->conf_luns, l_next, tmplun) { newlun = lun_find(newconf, oldlun->l_name); if (newlun == NULL) { log_debugx("lun \"%s\", CTL lun %d " "not found in new configuration; " "removing", oldlun->l_name, oldlun->l_ctl_lun); error = kernel_lun_remove(oldlun); if (error != 0) { log_warnx("failed to remove lun \"%s\", " "CTL lun %d", oldlun->l_name, oldlun->l_ctl_lun); cumulated_error++; } continue; } /* * Also remove the LUNs changed by more than size. */ changed = 0; assert(oldlun->l_backend != NULL); assert(newlun->l_backend != NULL); if (strcmp(newlun->l_backend, oldlun->l_backend) != 0) { log_debugx("backend for lun \"%s\", " "CTL lun %d changed; removing", oldlun->l_name, oldlun->l_ctl_lun); changed = 1; } if (oldlun->l_blocksize != newlun->l_blocksize) { log_debugx("blocksize for lun \"%s\", " "CTL lun %d changed; removing", oldlun->l_name, oldlun->l_ctl_lun); changed = 1; } if (newlun->l_device_id != NULL && (oldlun->l_device_id == NULL || strcmp(oldlun->l_device_id, newlun->l_device_id) != 0)) { log_debugx("device-id for lun \"%s\", " "CTL lun %d changed; removing", oldlun->l_name, oldlun->l_ctl_lun); changed = 1; } if (newlun->l_path != NULL && (oldlun->l_path == NULL || strcmp(oldlun->l_path, newlun->l_path) != 0)) { log_debugx("path for lun \"%s\", " "CTL lun %d, changed; removing", oldlun->l_name, oldlun->l_ctl_lun); changed = 1; } if (newlun->l_serial != NULL && (oldlun->l_serial == NULL || strcmp(oldlun->l_serial, newlun->l_serial) != 0)) { log_debugx("serial for lun \"%s\", " "CTL lun %d changed; removing", oldlun->l_name, oldlun->l_ctl_lun); changed = 1; } if (changed) { error = kernel_lun_remove(oldlun); if (error != 0) { log_warnx("failed to remove lun \"%s\", " "CTL lun %d", oldlun->l_name, oldlun->l_ctl_lun); cumulated_error++; } lun_delete(oldlun); continue; } lun_set_ctl_lun(newlun, oldlun->l_ctl_lun); } TAILQ_FOREACH_SAFE(newlun, &newconf->conf_luns, l_next, tmplun) { oldlun = lun_find(oldconf, newlun->l_name); if (oldlun != NULL) { log_debugx("modifying lun \"%s\", CTL lun %d", newlun->l_name, newlun->l_ctl_lun); error = kernel_lun_modify(newlun); if (error != 0) { log_warnx("failed to " "modify lun \"%s\", CTL lun %d", newlun->l_name, newlun->l_ctl_lun); cumulated_error++; } continue; } log_debugx("adding lun \"%s\"", newlun->l_name); error = kernel_lun_add(newlun); if (error != 0) { log_warnx("failed to add lun \"%s\"", newlun->l_name); lun_delete(newlun); cumulated_error++; } } /* * Now add new ports or modify existing ones. */ TAILQ_FOREACH(newport, &newconf->conf_ports, p_next) { if (port_is_dummy(newport)) continue; oldport = port_find(oldconf, newport->p_name); if (oldport == NULL || port_is_dummy(oldport)) { log_debugx("adding port \"%s\"", newport->p_name); error = kernel_port_add(newport); } else { log_debugx("updating port \"%s\"", newport->p_name); newport->p_ctl_port = oldport->p_ctl_port; error = kernel_port_update(newport, oldport); } if (error != 0) { log_warnx("failed to %s port %s", (oldport == NULL) ? "add" : "update", newport->p_name); /* * XXX: Uncomment after fixing the root cause. * * cumulated_error++; */ } } /* * Go through the new portals, opening the sockets as necessary. */ TAILQ_FOREACH(newpg, &newconf->conf_portal_groups, pg_next) { if (newpg->pg_foreign) continue; if (newpg->pg_unassigned) { log_debugx("not listening on portal-group \"%s\", " "not assigned to any target", newpg->pg_name); continue; } TAILQ_FOREACH(newp, &newpg->pg_portals, p_next) { /* * Try to find already open portal and reuse * the listening socket. We don't care about * what portal or portal group that was, what * matters is the listening address. */ TAILQ_FOREACH(oldpg, &oldconf->conf_portal_groups, pg_next) { TAILQ_FOREACH(oldp, &oldpg->pg_portals, p_next) { if (strcmp(newp->p_listen, oldp->p_listen) == 0 && oldp->p_socket > 0) { newp->p_socket = oldp->p_socket; oldp->p_socket = 0; break; } } } if (newp->p_socket > 0) { /* * We're done with this portal. */ continue; } #ifdef ICL_KERNEL_PROXY if (proxy_mode) { newpg->pg_conf->conf_portal_id++; newp->p_id = newpg->pg_conf->conf_portal_id; log_debugx("listening on %s, portal-group " "\"%s\", portal id %d, using ICL proxy", newp->p_listen, newpg->pg_name, newp->p_id); kernel_listen(newp->p_ai, newp->p_iser, newp->p_id); continue; } #endif assert(proxy_mode == false); assert(newp->p_iser == false); log_debugx("listening on %s, portal-group \"%s\"", newp->p_listen, newpg->pg_name); newp->p_socket = socket(newp->p_ai->ai_family, newp->p_ai->ai_socktype, newp->p_ai->ai_protocol); if (newp->p_socket < 0) { log_warn("socket(2) failed for %s", newp->p_listen); cumulated_error++; continue; } sockbuf = SOCKBUF_SIZE; if (setsockopt(newp->p_socket, SOL_SOCKET, SO_RCVBUF, &sockbuf, sizeof(sockbuf)) == -1) log_warn("setsockopt(SO_RCVBUF) failed " "for %s", newp->p_listen); sockbuf = SOCKBUF_SIZE; if (setsockopt(newp->p_socket, SOL_SOCKET, SO_SNDBUF, &sockbuf, sizeof(sockbuf)) == -1) log_warn("setsockopt(SO_SNDBUF) failed " "for %s", newp->p_listen); error = setsockopt(newp->p_socket, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(one)); if (error != 0) { log_warn("setsockopt(SO_REUSEADDR) failed " "for %s", newp->p_listen); close(newp->p_socket); newp->p_socket = 0; cumulated_error++; continue; } error = bind(newp->p_socket, newp->p_ai->ai_addr, newp->p_ai->ai_addrlen); if (error != 0) { log_warn("bind(2) failed for %s", newp->p_listen); close(newp->p_socket); newp->p_socket = 0; cumulated_error++; continue; } error = listen(newp->p_socket, -1); if (error != 0) { log_warn("listen(2) failed for %s", newp->p_listen); close(newp->p_socket); newp->p_socket = 0; cumulated_error++; continue; } } } /* * Go through the no longer used sockets, closing them. */ TAILQ_FOREACH(oldpg, &oldconf->conf_portal_groups, pg_next) { TAILQ_FOREACH(oldp, &oldpg->pg_portals, p_next) { if (oldp->p_socket <= 0) continue; log_debugx("closing socket for %s, portal-group \"%s\"", oldp->p_listen, oldpg->pg_name); close(oldp->p_socket); oldp->p_socket = 0; } } /* (Re-)Register on remaining/new iSNS servers. */ TAILQ_FOREACH(newns, &newconf->conf_isns, i_next) { TAILQ_FOREACH(oldns, &oldconf->conf_isns, i_next) { if (strcmp(oldns->i_addr, newns->i_addr) == 0) break; } isns_register(newns, oldns); } /* Schedule iSNS update */ if (!TAILQ_EMPTY(&newconf->conf_isns)) set_timeout((newconf->conf_isns_period + 2) / 3, false); return (cumulated_error); } bool timed_out(void) { return (sigalrm_received); } static void sigalrm_handler_fatal(int dummy __unused) { /* * It would be easiest to just log an error and exit. We can't * do this, though, because log_errx() is not signal safe, since * it calls syslog(3). Instead, set a flag checked by pdu_send() * and pdu_receive(), to call log_errx() there. Should they fail * to notice, we'll exit here one second later. */ if (sigalrm_received) { /* * Oh well. Just give up and quit. */ _exit(2); } sigalrm_received = true; } static void sigalrm_handler(int dummy __unused) { sigalrm_received = true; } void set_timeout(int timeout, int fatal) { struct sigaction sa; struct itimerval itv; int error; if (timeout <= 0) { log_debugx("session timeout disabled"); bzero(&itv, sizeof(itv)); error = setitimer(ITIMER_REAL, &itv, NULL); if (error != 0) log_err(1, "setitimer"); sigalrm_received = false; return; } sigalrm_received = false; bzero(&sa, sizeof(sa)); if (fatal) sa.sa_handler = sigalrm_handler_fatal; else sa.sa_handler = sigalrm_handler; sigfillset(&sa.sa_mask); error = sigaction(SIGALRM, &sa, NULL); if (error != 0) log_err(1, "sigaction"); /* * First SIGALRM will arive after conf_timeout seconds. * If we do nothing, another one will arrive a second later. */ log_debugx("setting session timeout to %d seconds", timeout); bzero(&itv, sizeof(itv)); itv.it_interval.tv_sec = 1; itv.it_value.tv_sec = timeout; error = setitimer(ITIMER_REAL, &itv, NULL); if (error != 0) log_err(1, "setitimer"); } static int wait_for_children(bool block) { pid_t pid; int status; int num = 0; for (;;) { /* * If "block" is true, wait for at least one process. */ if (block && num == 0) pid = wait4(-1, &status, 0, NULL); else pid = wait4(-1, &status, WNOHANG, NULL); if (pid <= 0) break; if (WIFSIGNALED(status)) { log_warnx("child process %d terminated with signal %d", pid, WTERMSIG(status)); } else if (WEXITSTATUS(status) != 0) { log_warnx("child process %d terminated with exit status %d", pid, WEXITSTATUS(status)); } else { log_debugx("child process %d terminated gracefully", pid); } num++; } return (num); } static void handle_connection(struct portal *portal, int fd, const struct sockaddr *client_sa, bool dont_fork) { struct connection *conn; int error; pid_t pid; char host[NI_MAXHOST + 1]; struct conf *conf; conf = portal->p_portal_group->pg_conf; if (dont_fork) { log_debugx("incoming connection; not forking due to -d flag"); } else { nchildren -= wait_for_children(false); assert(nchildren >= 0); while (conf->conf_maxproc > 0 && nchildren >= conf->conf_maxproc) { log_debugx("maxproc limit of %d child processes hit; " "waiting for child process to exit", conf->conf_maxproc); nchildren -= wait_for_children(true); assert(nchildren >= 0); } log_debugx("incoming connection; forking child process #%d", nchildren); nchildren++; pid = fork(); if (pid < 0) log_err(1, "fork"); if (pid > 0) { close(fd); return; } } pidfile_close(conf->conf_pidfh); error = getnameinfo(client_sa, client_sa->sa_len, host, sizeof(host), NULL, 0, NI_NUMERICHOST); if (error != 0) log_errx(1, "getnameinfo: %s", gai_strerror(error)); log_debugx("accepted connection from %s; portal group \"%s\"", host, portal->p_portal_group->pg_name); log_set_peer_addr(host); setproctitle("%s", host); conn = connection_new(portal, fd, host, client_sa); set_timeout(conf->conf_timeout, true); kernel_capsicate(); login(conn); if (conn->conn_session_type == CONN_SESSION_TYPE_NORMAL) { kernel_handoff(conn); log_debugx("connection handed off to the kernel"); } else { assert(conn->conn_session_type == CONN_SESSION_TYPE_DISCOVERY); discovery(conn); } log_debugx("nothing more to do; exiting"); exit(0); } static int fd_add(int fd, fd_set *fdset, int nfds) { /* * Skip sockets which we failed to bind. */ if (fd <= 0) return (nfds); FD_SET(fd, fdset); if (fd > nfds) nfds = fd; return (nfds); } static void main_loop(struct conf *conf, bool dont_fork) { struct portal_group *pg; struct portal *portal; struct sockaddr_storage client_sa; socklen_t client_salen; #ifdef ICL_KERNEL_PROXY int connection_id; int portal_id; #endif fd_set fdset; int error, nfds, client_fd; pidfile_write(conf->conf_pidfh); for (;;) { if (sighup_received || sigterm_received || timed_out()) return; #ifdef ICL_KERNEL_PROXY if (proxy_mode) { client_salen = sizeof(client_sa); kernel_accept(&connection_id, &portal_id, (struct sockaddr *)&client_sa, &client_salen); assert(client_salen >= client_sa.ss_len); log_debugx("incoming connection, id %d, portal id %d", connection_id, portal_id); TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { TAILQ_FOREACH(portal, &pg->pg_portals, p_next) { if (portal->p_id == portal_id) { goto found; } } } log_errx(1, "kernel returned invalid portal_id %d", portal_id); found: handle_connection(portal, connection_id, (struct sockaddr *)&client_sa, dont_fork); } else { #endif assert(proxy_mode == false); FD_ZERO(&fdset); nfds = 0; TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { TAILQ_FOREACH(portal, &pg->pg_portals, p_next) nfds = fd_add(portal->p_socket, &fdset, nfds); } error = select(nfds + 1, &fdset, NULL, NULL, NULL); if (error <= 0) { if (errno == EINTR) return; log_err(1, "select"); } TAILQ_FOREACH(pg, &conf->conf_portal_groups, pg_next) { TAILQ_FOREACH(portal, &pg->pg_portals, p_next) { if (!FD_ISSET(portal->p_socket, &fdset)) continue; client_salen = sizeof(client_sa); client_fd = accept(portal->p_socket, (struct sockaddr *)&client_sa, &client_salen); if (client_fd < 0) { if (errno == ECONNABORTED) continue; log_err(1, "accept"); } assert(client_salen >= client_sa.ss_len); handle_connection(portal, client_fd, (struct sockaddr *)&client_sa, dont_fork); break; } } #ifdef ICL_KERNEL_PROXY } #endif } } static void sighup_handler(int dummy __unused) { sighup_received = true; } static void sigterm_handler(int dummy __unused) { sigterm_received = true; } static void sigchld_handler(int dummy __unused) { /* * The only purpose of this handler is to make SIGCHLD * interrupt the ISCSIDWAIT ioctl(2), so we can call * wait_for_children(). */ } static void register_signals(void) { struct sigaction sa; int error; bzero(&sa, sizeof(sa)); sa.sa_handler = sighup_handler; sigfillset(&sa.sa_mask); error = sigaction(SIGHUP, &sa, NULL); if (error != 0) log_err(1, "sigaction"); sa.sa_handler = sigterm_handler; error = sigaction(SIGTERM, &sa, NULL); if (error != 0) log_err(1, "sigaction"); sa.sa_handler = sigterm_handler; error = sigaction(SIGINT, &sa, NULL); if (error != 0) log_err(1, "sigaction"); sa.sa_handler = sigchld_handler; error = sigaction(SIGCHLD, &sa, NULL); if (error != 0) log_err(1, "sigaction"); } static void check_perms(const char *path) { struct stat sb; int error; error = stat(path, &sb); if (error != 0) { log_warn("stat"); return; } if (sb.st_mode & S_IWOTH) { log_warnx("%s is world-writable", path); } else if (sb.st_mode & S_IROTH) { log_warnx("%s is world-readable", path); } else if (sb.st_mode & S_IXOTH) { /* * Ok, this one doesn't matter, but still do it, * just for consistency. */ log_warnx("%s is world-executable", path); } /* * XXX: Should we also check for owner != 0? */ } static struct conf * conf_new_from_file(const char *path, struct conf *oldconf, bool ucl) { struct conf *conf; struct auth_group *ag; struct portal_group *pg; struct pport *pp; int error; log_debugx("obtaining configuration from %s", path); conf = conf_new(); TAILQ_FOREACH(pp, &oldconf->conf_pports, pp_next) pport_copy(pp, conf); ag = auth_group_new(conf, "default"); assert(ag != NULL); ag = auth_group_new(conf, "no-authentication"); assert(ag != NULL); ag->ag_type = AG_TYPE_NO_AUTHENTICATION; ag = auth_group_new(conf, "no-access"); assert(ag != NULL); ag->ag_type = AG_TYPE_DENY; pg = portal_group_new(conf, "default"); assert(pg != NULL); if (ucl) error = uclparse_conf(conf, path); else error = parse_conf(conf, path); if (error != 0) { conf_delete(conf); return (NULL); } check_perms(path); if (conf->conf_default_ag_defined == false) { log_debugx("auth-group \"default\" not defined; " "going with defaults"); ag = auth_group_find(conf, "default"); assert(ag != NULL); ag->ag_type = AG_TYPE_DENY; } if (conf->conf_default_pg_defined == false) { log_debugx("portal-group \"default\" not defined; " "going with defaults"); pg = portal_group_find(conf, "default"); assert(pg != NULL); portal_group_add_listen(pg, "0.0.0.0:3260", false); portal_group_add_listen(pg, "[::]:3260", false); } conf->conf_kernel_port_on = true; error = conf_verify(conf); if (error != 0) { conf_delete(conf); return (NULL); } return (conf); } int main(int argc, char **argv) { struct conf *oldconf, *newconf, *tmpconf; struct isns *newns; const char *config_path = DEFAULT_CONFIG_PATH; int debug = 0, ch, error; bool dont_daemonize = false; bool use_ucl = false; while ((ch = getopt(argc, argv, "duf:R")) != -1) { switch (ch) { case 'd': dont_daemonize = true; debug++; break; case 'u': use_ucl = true; break; case 'f': config_path = optarg; break; case 'R': #ifndef ICL_KERNEL_PROXY log_errx(1, "ctld(8) compiled without ICL_KERNEL_PROXY " "does not support iSER protocol"); #endif proxy_mode = true; break; case '?': default: usage(); } } argc -= optind; if (argc != 0) usage(); log_init(debug); kernel_init(); oldconf = conf_new_from_kernel(); newconf = conf_new_from_file(config_path, oldconf, use_ucl); if (newconf == NULL) log_errx(1, "configuration error; exiting"); if (debug > 0) { oldconf->conf_debug = debug; newconf->conf_debug = debug; } error = conf_apply(oldconf, newconf); if (error != 0) log_errx(1, "failed to apply configuration; exiting"); conf_delete(oldconf); oldconf = NULL; register_signals(); if (dont_daemonize == false) { log_debugx("daemonizing"); if (daemon(0, 0) == -1) { log_warn("cannot daemonize"); pidfile_remove(newconf->conf_pidfh); exit(1); } } /* Schedule iSNS update */ if (!TAILQ_EMPTY(&newconf->conf_isns)) set_timeout((newconf->conf_isns_period + 2) / 3, false); for (;;) { main_loop(newconf, dont_daemonize); if (sighup_received) { sighup_received = false; log_debugx("received SIGHUP, reloading configuration"); tmpconf = conf_new_from_file(config_path, newconf, use_ucl); if (tmpconf == NULL) { log_warnx("configuration error, " "continuing with old configuration"); } else { if (debug > 0) tmpconf->conf_debug = debug; oldconf = newconf; newconf = tmpconf; error = conf_apply(oldconf, newconf); if (error != 0) log_warnx("failed to reload " "configuration"); conf_delete(oldconf); oldconf = NULL; } } else if (sigterm_received) { log_debugx("exiting on signal; " "reloading empty configuration"); log_debugx("removing CTL iSCSI ports " "and terminating all connections"); oldconf = newconf; newconf = conf_new(); if (debug > 0) newconf->conf_debug = debug; error = conf_apply(oldconf, newconf); if (error != 0) log_warnx("failed to apply configuration"); conf_delete(oldconf); oldconf = NULL; log_warnx("exiting on signal"); exit(0); } else { nchildren -= wait_for_children(false); assert(nchildren >= 0); if (timed_out()) { set_timeout(0, false); TAILQ_FOREACH(newns, &newconf->conf_isns, i_next) isns_check(newns); /* Schedule iSNS update */ if (!TAILQ_EMPTY(&newconf->conf_isns)) { set_timeout((newconf->conf_isns_period + 2) / 3, false); } } } } /* NOTREACHED */ } Index: head/usr.sbin/ctld/ctld.h =================================================================== --- head/usr.sbin/ctld/ctld.h (revision 333445) +++ head/usr.sbin/ctld/ctld.h (revision 333446) @@ -1,466 +1,471 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 The FreeBSD Foundation * All rights reserved. * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef CTLD_H #define CTLD_H #include #ifdef ICL_KERNEL_PROXY #include #endif #include #include #include #define DEFAULT_CONFIG_PATH "/etc/ctl.conf" #define DEFAULT_PIDFILE "/var/run/ctld.pid" #define DEFAULT_BLOCKSIZE 512 #define DEFAULT_CD_BLOCKSIZE 2048 #define MAX_LUNS 1024 #define MAX_NAME_LEN 223 #define MAX_DATA_SEGMENT_LENGTH (128 * 1024) #define SOCKBUF_SIZE 1048576 struct auth { TAILQ_ENTRY(auth) a_next; struct auth_group *a_auth_group; char *a_user; char *a_secret; char *a_mutual_user; char *a_mutual_secret; }; struct auth_name { TAILQ_ENTRY(auth_name) an_next; struct auth_group *an_auth_group; char *an_initator_name; }; struct auth_portal { TAILQ_ENTRY(auth_portal) ap_next; struct auth_group *ap_auth_group; char *ap_initator_portal; struct sockaddr_storage ap_sa; int ap_mask; }; #define AG_TYPE_UNKNOWN 0 #define AG_TYPE_DENY 1 #define AG_TYPE_NO_AUTHENTICATION 2 #define AG_TYPE_CHAP 3 #define AG_TYPE_CHAP_MUTUAL 4 struct auth_group { TAILQ_ENTRY(auth_group) ag_next; struct conf *ag_conf; char *ag_name; struct target *ag_target; int ag_type; TAILQ_HEAD(, auth) ag_auths; TAILQ_HEAD(, auth_name) ag_names; TAILQ_HEAD(, auth_portal) ag_portals; }; struct portal { TAILQ_ENTRY(portal) p_next; struct portal_group *p_portal_group; bool p_iser; char *p_listen; struct addrinfo *p_ai; #ifdef ICL_KERNEL_PROXY int p_id; #endif TAILQ_HEAD(, target) p_targets; int p_socket; }; TAILQ_HEAD(options, option); #define PG_FILTER_UNKNOWN 0 #define PG_FILTER_NONE 1 #define PG_FILTER_PORTAL 2 #define PG_FILTER_PORTAL_NAME 3 #define PG_FILTER_PORTAL_NAME_AUTH 4 struct portal_group { TAILQ_ENTRY(portal_group) pg_next; struct conf *pg_conf; struct options pg_options; char *pg_name; struct auth_group *pg_discovery_auth_group; int pg_discovery_filter; int pg_foreign; bool pg_unassigned; TAILQ_HEAD(, portal) pg_portals; TAILQ_HEAD(, port) pg_ports; char *pg_offload; char *pg_redirection; uint16_t pg_tag; }; struct pport { TAILQ_ENTRY(pport) pp_next; TAILQ_HEAD(, port) pp_ports; struct conf *pp_conf; char *pp_name; uint32_t pp_ctl_port; }; struct port { TAILQ_ENTRY(port) p_next; TAILQ_ENTRY(port) p_pgs; TAILQ_ENTRY(port) p_pps; TAILQ_ENTRY(port) p_ts; struct conf *p_conf; char *p_name; struct auth_group *p_auth_group; struct portal_group *p_portal_group; struct pport *p_pport; struct target *p_target; + int p_ioctl_port; + int p_ioctl_pp; + int p_ioctl_vp; uint32_t p_ctl_port; }; struct option { TAILQ_ENTRY(option) o_next; char *o_name; char *o_value; }; struct lun { TAILQ_ENTRY(lun) l_next; struct conf *l_conf; struct options l_options; char *l_name; char *l_backend; uint8_t l_device_type; int l_blocksize; char *l_device_id; char *l_path; char *l_scsiname; char *l_serial; int64_t l_size; int l_ctl_lun; }; struct target { TAILQ_ENTRY(target) t_next; struct conf *t_conf; struct lun *t_luns[MAX_LUNS]; struct auth_group *t_auth_group; TAILQ_HEAD(, port) t_ports; char *t_name; char *t_alias; char *t_redirection; }; struct isns { TAILQ_ENTRY(isns) i_next; struct conf *i_conf; char *i_addr; struct addrinfo *i_ai; }; struct conf { char *conf_pidfile_path; TAILQ_HEAD(, lun) conf_luns; TAILQ_HEAD(, target) conf_targets; TAILQ_HEAD(, auth_group) conf_auth_groups; TAILQ_HEAD(, port) conf_ports; TAILQ_HEAD(, portal_group) conf_portal_groups; TAILQ_HEAD(, pport) conf_pports; TAILQ_HEAD(, isns) conf_isns; int conf_isns_period; int conf_isns_timeout; int conf_debug; int conf_timeout; int conf_maxproc; #ifdef ICL_KERNEL_PROXY int conf_portal_id; #endif struct pidfh *conf_pidfh; bool conf_default_pg_defined; bool conf_default_ag_defined; bool conf_kernel_port_on; }; #define CONN_SESSION_TYPE_NONE 0 #define CONN_SESSION_TYPE_DISCOVERY 1 #define CONN_SESSION_TYPE_NORMAL 2 #define CONN_DIGEST_NONE 0 #define CONN_DIGEST_CRC32C 1 struct connection { struct portal *conn_portal; struct port *conn_port; struct target *conn_target; int conn_socket; int conn_session_type; char *conn_initiator_name; char *conn_initiator_addr; char *conn_initiator_alias; uint8_t conn_initiator_isid[6]; struct sockaddr_storage conn_initiator_sa; uint32_t conn_cmdsn; uint32_t conn_statsn; int conn_max_recv_data_segment_limit; int conn_max_send_data_segment_limit; int conn_max_burst_limit; int conn_first_burst_limit; int conn_max_recv_data_segment_length; int conn_max_send_data_segment_length; int conn_max_burst_length; int conn_first_burst_length; int conn_immediate_data; int conn_header_digest; int conn_data_digest; const char *conn_user; struct chap *conn_chap; }; struct pdu { struct connection *pdu_connection; struct iscsi_bhs *pdu_bhs; char *pdu_data; size_t pdu_data_len; }; #define KEYS_MAX 1024 struct keys { char *keys_names[KEYS_MAX]; char *keys_values[KEYS_MAX]; char *keys_data; size_t keys_data_len; }; #define CHAP_CHALLENGE_LEN 1024 #define CHAP_DIGEST_LEN 16 /* Equal to MD5 digest size. */ struct chap { unsigned char chap_id; char chap_challenge[CHAP_CHALLENGE_LEN]; char chap_response[CHAP_DIGEST_LEN]; }; struct rchap { char *rchap_secret; unsigned char rchap_id; void *rchap_challenge; size_t rchap_challenge_len; }; struct chap *chap_new(void); char *chap_get_id(const struct chap *chap); char *chap_get_challenge(const struct chap *chap); int chap_receive(struct chap *chap, const char *response); int chap_authenticate(struct chap *chap, const char *secret); void chap_delete(struct chap *chap); struct rchap *rchap_new(const char *secret); int rchap_receive(struct rchap *rchap, const char *id, const char *challenge); char *rchap_get_response(struct rchap *rchap); void rchap_delete(struct rchap *rchap); int parse_conf(struct conf *conf, const char *path); int uclparse_conf(struct conf *conf, const char *path); struct conf *conf_new(void); struct conf *conf_new_from_kernel(void); void conf_delete(struct conf *conf); int conf_verify(struct conf *conf); struct auth_group *auth_group_new(struct conf *conf, const char *name); void auth_group_delete(struct auth_group *ag); struct auth_group *auth_group_find(const struct conf *conf, const char *name); int auth_group_set_type(struct auth_group *ag, const char *type); const struct auth *auth_new_chap(struct auth_group *ag, const char *user, const char *secret); const struct auth *auth_new_chap_mutual(struct auth_group *ag, const char *user, const char *secret, const char *user2, const char *secret2); const struct auth *auth_find(const struct auth_group *ag, const char *user); const struct auth_name *auth_name_new(struct auth_group *ag, const char *initiator_name); bool auth_name_defined(const struct auth_group *ag); const struct auth_name *auth_name_find(const struct auth_group *ag, const char *initiator_name); int auth_name_check(const struct auth_group *ag, const char *initiator_name); const struct auth_portal *auth_portal_new(struct auth_group *ag, const char *initiator_portal); bool auth_portal_defined(const struct auth_group *ag); const struct auth_portal *auth_portal_find(const struct auth_group *ag, const struct sockaddr_storage *sa); int auth_portal_check(const struct auth_group *ag, const struct sockaddr_storage *sa); struct portal_group *portal_group_new(struct conf *conf, const char *name); void portal_group_delete(struct portal_group *pg); struct portal_group *portal_group_find(const struct conf *conf, const char *name); int portal_group_add_listen(struct portal_group *pg, const char *listen, bool iser); int portal_group_set_filter(struct portal_group *pg, const char *filter); int portal_group_set_offload(struct portal_group *pg, const char *offload); int portal_group_set_redirection(struct portal_group *pg, const char *addr); int isns_new(struct conf *conf, const char *addr); void isns_delete(struct isns *is); void isns_register(struct isns *isns, struct isns *oldisns); void isns_check(struct isns *isns); void isns_deregister(struct isns *isns); struct pport *pport_new(struct conf *conf, const char *name, uint32_t ctl_port); struct pport *pport_find(const struct conf *conf, const char *name); struct pport *pport_copy(struct pport *pport, struct conf *conf); void pport_delete(struct pport *pport); struct port *port_new(struct conf *conf, struct target *target, struct portal_group *pg); +struct port *port_new_ioctl(struct conf *conf, struct target *target, + int pp, int vp); struct port *port_new_pp(struct conf *conf, struct target *target, struct pport *pp); struct port *port_find(const struct conf *conf, const char *name); struct port *port_find_in_pg(const struct portal_group *pg, const char *target); void port_delete(struct port *port); int port_is_dummy(struct port *port); struct target *target_new(struct conf *conf, const char *name); void target_delete(struct target *target); struct target *target_find(struct conf *conf, const char *name); int target_set_redirection(struct target *target, const char *addr); struct lun *lun_new(struct conf *conf, const char *name); void lun_delete(struct lun *lun); struct lun *lun_find(const struct conf *conf, const char *name); void lun_set_backend(struct lun *lun, const char *value); void lun_set_device_type(struct lun *lun, uint8_t value); void lun_set_blocksize(struct lun *lun, size_t value); void lun_set_device_id(struct lun *lun, const char *value); void lun_set_path(struct lun *lun, const char *value); void lun_set_scsiname(struct lun *lun, const char *value); void lun_set_serial(struct lun *lun, const char *value); void lun_set_size(struct lun *lun, size_t value); void lun_set_ctl_lun(struct lun *lun, uint32_t value); struct option *option_new(struct options *os, const char *name, const char *value); void option_delete(struct options *os, struct option *co); struct option *option_find(const struct options *os, const char *name); void option_set(struct option *o, const char *value); void kernel_init(void); int kernel_lun_add(struct lun *lun); int kernel_lun_modify(struct lun *lun); int kernel_lun_remove(struct lun *lun); void kernel_handoff(struct connection *conn); void kernel_limits(const char *offload, int *max_recv_data_segment_length, int *max_send_data_segment_length, int *max_burst_length, int *first_burst_length); int kernel_port_add(struct port *port); int kernel_port_update(struct port *port, struct port *old); int kernel_port_remove(struct port *port); void kernel_capsicate(void); #ifdef ICL_KERNEL_PROXY void kernel_listen(struct addrinfo *ai, bool iser, int portal_id); void kernel_accept(int *connection_id, int *portal_id, struct sockaddr *client_sa, socklen_t *client_salen); void kernel_send(struct pdu *pdu); void kernel_receive(struct pdu *pdu); #endif struct keys *keys_new(void); void keys_delete(struct keys *keys); void keys_load(struct keys *keys, const struct pdu *pdu); void keys_save(struct keys *keys, struct pdu *pdu); const char *keys_find(struct keys *keys, const char *name); void keys_add(struct keys *keys, const char *name, const char *value); void keys_add_int(struct keys *keys, const char *name, int value); struct pdu *pdu_new(struct connection *conn); struct pdu *pdu_new_response(struct pdu *request); void pdu_delete(struct pdu *pdu); void pdu_receive(struct pdu *request); void pdu_send(struct pdu *response); void login(struct connection *conn); void discovery(struct connection *conn); void log_init(int level); void log_set_peer_name(const char *name); void log_set_peer_addr(const char *addr); void log_err(int, const char *, ...) __dead2 __printflike(2, 3); void log_errx(int, const char *, ...) __dead2 __printflike(2, 3); void log_warn(const char *, ...) __printflike(1, 2); void log_warnx(const char *, ...) __printflike(1, 2); void log_debugx(const char *, ...) __printflike(1, 2); char *checked_strdup(const char *); bool valid_iscsi_name(const char *name); void set_timeout(int timeout, int fatal); bool timed_out(void); #endif /* !CTLD_H */ Index: head/usr.sbin/ctld/kernel.c =================================================================== --- head/usr.sbin/ctld/kernel.c (revision 333445) +++ head/usr.sbin/ctld/kernel.c (revision 333446) @@ -1,1312 +1,1339 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2003, 2004 Silicon Graphics International Corp. * Copyright (c) 1997-2007 Kenneth D. Merry * Copyright (c) 2012 The FreeBSD Foundation + * Copyright (c) 2017 Jakub Wojciech Klama * All rights reserved. * * Portions of this software were developed by Edward Tomasz Napierala * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "ctld.h" #ifdef ICL_KERNEL_PROXY #include #endif +#define NVLIST_BUFSIZE 1024 + extern bool proxy_mode; static int ctl_fd = 0; void kernel_init(void) { int retval, saved_errno; ctl_fd = open(CTL_DEFAULT_DEV, O_RDWR); if (ctl_fd < 0 && errno == ENOENT) { saved_errno = errno; retval = kldload("ctl"); if (retval != -1) ctl_fd = open(CTL_DEFAULT_DEV, O_RDWR); else errno = saved_errno; } if (ctl_fd < 0) log_err(1, "failed to open %s", CTL_DEFAULT_DEV); #ifdef WANT_ISCSI else { saved_errno = errno; if (modfind("cfiscsi") == -1 && kldload("cfiscsi") == -1) log_warn("couldn't load cfiscsi"); errno = saved_errno; } #endif } /* * Name/value pair used for per-LUN attributes. */ struct cctl_lun_nv { char *name; char *value; STAILQ_ENTRY(cctl_lun_nv) links; }; /* * Backend LUN information. */ struct cctl_lun { uint64_t lun_id; char *backend_type; uint8_t device_type; uint64_t size_blocks; uint32_t blocksize; char *serial_number; char *device_id; char *ctld_name; STAILQ_HEAD(,cctl_lun_nv) attr_list; STAILQ_ENTRY(cctl_lun) links; }; struct cctl_port { uint32_t port_id; char *port_frontend; char *port_name; int pp; int vp; int cfiscsi_state; char *cfiscsi_target; uint16_t cfiscsi_portal_group_tag; char *ctld_portal_group_name; STAILQ_HEAD(,cctl_lun_nv) attr_list; STAILQ_ENTRY(cctl_port) links; }; struct cctl_devlist_data { int num_luns; STAILQ_HEAD(,cctl_lun) lun_list; struct cctl_lun *cur_lun; int num_ports; STAILQ_HEAD(,cctl_port) port_list; struct cctl_port *cur_port; int level; struct sbuf *cur_sb[32]; }; static void cctl_start_element(void *user_data, const char *name, const char **attr) { int i; struct cctl_devlist_data *devlist; struct cctl_lun *cur_lun; devlist = (struct cctl_devlist_data *)user_data; cur_lun = devlist->cur_lun; devlist->level++; if ((u_int)devlist->level >= (sizeof(devlist->cur_sb) / sizeof(devlist->cur_sb[0]))) log_errx(1, "%s: too many nesting levels, %zd max", __func__, sizeof(devlist->cur_sb) / sizeof(devlist->cur_sb[0])); devlist->cur_sb[devlist->level] = sbuf_new_auto(); if (devlist->cur_sb[devlist->level] == NULL) log_err(1, "%s: unable to allocate sbuf", __func__); if (strcmp(name, "lun") == 0) { if (cur_lun != NULL) log_errx(1, "%s: improper lun element nesting", __func__); cur_lun = calloc(1, sizeof(*cur_lun)); if (cur_lun == NULL) log_err(1, "%s: cannot allocate %zd bytes", __func__, sizeof(*cur_lun)); devlist->num_luns++; devlist->cur_lun = cur_lun; STAILQ_INIT(&cur_lun->attr_list); STAILQ_INSERT_TAIL(&devlist->lun_list, cur_lun, links); for (i = 0; attr[i] != NULL; i += 2) { if (strcmp(attr[i], "id") == 0) { cur_lun->lun_id = strtoull(attr[i+1], NULL, 0); } else { log_errx(1, "%s: invalid LUN attribute %s = %s", __func__, attr[i], attr[i+1]); } } } } static void cctl_end_element(void *user_data, const char *name) { struct cctl_devlist_data *devlist; struct cctl_lun *cur_lun; char *str; devlist = (struct cctl_devlist_data *)user_data; cur_lun = devlist->cur_lun; if ((cur_lun == NULL) && (strcmp(name, "ctllunlist") != 0)) log_errx(1, "%s: cur_lun == NULL! (name = %s)", __func__, name); if (devlist->cur_sb[devlist->level] == NULL) log_errx(1, "%s: no valid sbuf at level %d (name %s)", __func__, devlist->level, name); sbuf_finish(devlist->cur_sb[devlist->level]); str = checked_strdup(sbuf_data(devlist->cur_sb[devlist->level])); if (strlen(str) == 0) { free(str); str = NULL; } sbuf_delete(devlist->cur_sb[devlist->level]); devlist->cur_sb[devlist->level] = NULL; devlist->level--; if (strcmp(name, "backend_type") == 0) { cur_lun->backend_type = str; str = NULL; } else if (strcmp(name, "lun_type") == 0) { cur_lun->device_type = strtoull(str, NULL, 0); } else if (strcmp(name, "size") == 0) { cur_lun->size_blocks = strtoull(str, NULL, 0); } else if (strcmp(name, "blocksize") == 0) { cur_lun->blocksize = strtoul(str, NULL, 0); } else if (strcmp(name, "serial_number") == 0) { cur_lun->serial_number = str; str = NULL; } else if (strcmp(name, "device_id") == 0) { cur_lun->device_id = str; str = NULL; } else if (strcmp(name, "ctld_name") == 0) { cur_lun->ctld_name = str; str = NULL; } else if (strcmp(name, "lun") == 0) { devlist->cur_lun = NULL; } else if (strcmp(name, "ctllunlist") == 0) { /* Nothing. */ } else { struct cctl_lun_nv *nv; nv = calloc(1, sizeof(*nv)); if (nv == NULL) log_err(1, "%s: can't allocate %zd bytes for nv pair", __func__, sizeof(*nv)); nv->name = checked_strdup(name); nv->value = str; str = NULL; STAILQ_INSERT_TAIL(&cur_lun->attr_list, nv, links); } free(str); } static void cctl_start_pelement(void *user_data, const char *name, const char **attr) { int i; struct cctl_devlist_data *devlist; struct cctl_port *cur_port; devlist = (struct cctl_devlist_data *)user_data; cur_port = devlist->cur_port; devlist->level++; if ((u_int)devlist->level >= (sizeof(devlist->cur_sb) / sizeof(devlist->cur_sb[0]))) log_errx(1, "%s: too many nesting levels, %zd max", __func__, sizeof(devlist->cur_sb) / sizeof(devlist->cur_sb[0])); devlist->cur_sb[devlist->level] = sbuf_new_auto(); if (devlist->cur_sb[devlist->level] == NULL) log_err(1, "%s: unable to allocate sbuf", __func__); if (strcmp(name, "targ_port") == 0) { if (cur_port != NULL) log_errx(1, "%s: improper port element nesting (%s)", __func__, name); cur_port = calloc(1, sizeof(*cur_port)); if (cur_port == NULL) log_err(1, "%s: cannot allocate %zd bytes", __func__, sizeof(*cur_port)); devlist->num_ports++; devlist->cur_port = cur_port; STAILQ_INIT(&cur_port->attr_list); STAILQ_INSERT_TAIL(&devlist->port_list, cur_port, links); for (i = 0; attr[i] != NULL; i += 2) { if (strcmp(attr[i], "id") == 0) { cur_port->port_id = strtoul(attr[i+1], NULL, 0); } else { log_errx(1, "%s: invalid LUN attribute %s = %s", __func__, attr[i], attr[i+1]); } } } } static void cctl_end_pelement(void *user_data, const char *name) { struct cctl_devlist_data *devlist; struct cctl_port *cur_port; char *str; devlist = (struct cctl_devlist_data *)user_data; cur_port = devlist->cur_port; if ((cur_port == NULL) && (strcmp(name, "ctlportlist") != 0)) log_errx(1, "%s: cur_port == NULL! (name = %s)", __func__, name); if (devlist->cur_sb[devlist->level] == NULL) log_errx(1, "%s: no valid sbuf at level %d (name %s)", __func__, devlist->level, name); sbuf_finish(devlist->cur_sb[devlist->level]); str = checked_strdup(sbuf_data(devlist->cur_sb[devlist->level])); if (strlen(str) == 0) { free(str); str = NULL; } sbuf_delete(devlist->cur_sb[devlist->level]); devlist->cur_sb[devlist->level] = NULL; devlist->level--; if (strcmp(name, "frontend_type") == 0) { cur_port->port_frontend = str; str = NULL; } else if (strcmp(name, "port_name") == 0) { cur_port->port_name = str; str = NULL; } else if (strcmp(name, "physical_port") == 0) { cur_port->pp = strtoul(str, NULL, 0); } else if (strcmp(name, "virtual_port") == 0) { cur_port->vp = strtoul(str, NULL, 0); } else if (strcmp(name, "cfiscsi_target") == 0) { cur_port->cfiscsi_target = str; str = NULL; } else if (strcmp(name, "cfiscsi_state") == 0) { cur_port->cfiscsi_state = strtoul(str, NULL, 0); } else if (strcmp(name, "cfiscsi_portal_group_tag") == 0) { cur_port->cfiscsi_portal_group_tag = strtoul(str, NULL, 0); } else if (strcmp(name, "ctld_portal_group_name") == 0) { cur_port->ctld_portal_group_name = str; str = NULL; } else if (strcmp(name, "targ_port") == 0) { devlist->cur_port = NULL; } else if (strcmp(name, "ctlportlist") == 0) { /* Nothing. */ } else { struct cctl_lun_nv *nv; nv = calloc(1, sizeof(*nv)); if (nv == NULL) log_err(1, "%s: can't allocate %zd bytes for nv pair", __func__, sizeof(*nv)); nv->name = checked_strdup(name); nv->value = str; str = NULL; STAILQ_INSERT_TAIL(&cur_port->attr_list, nv, links); } free(str); } static void cctl_char_handler(void *user_data, const XML_Char *str, int len) { struct cctl_devlist_data *devlist; devlist = (struct cctl_devlist_data *)user_data; sbuf_bcat(devlist->cur_sb[devlist->level], str, len); } struct conf * conf_new_from_kernel(void) { struct conf *conf = NULL; struct target *targ; struct portal_group *pg; struct pport *pp; struct port *cp; struct lun *cl; struct option *o; struct ctl_lun_list list; struct cctl_devlist_data devlist; struct cctl_lun *lun; struct cctl_port *port; XML_Parser parser; char *str, *name; int len, retval; bzero(&devlist, sizeof(devlist)); STAILQ_INIT(&devlist.lun_list); STAILQ_INIT(&devlist.port_list); log_debugx("obtaining previously configured CTL luns from the kernel"); str = NULL; len = 4096; retry: str = realloc(str, len); if (str == NULL) log_err(1, "realloc"); bzero(&list, sizeof(list)); list.alloc_len = len; list.status = CTL_LUN_LIST_NONE; list.lun_xml = str; if (ioctl(ctl_fd, CTL_LUN_LIST, &list) == -1) { log_warn("error issuing CTL_LUN_LIST ioctl"); free(str); return (NULL); } if (list.status == CTL_LUN_LIST_ERROR) { log_warnx("error returned from CTL_LUN_LIST ioctl: %s", list.error_str); free(str); return (NULL); } if (list.status == CTL_LUN_LIST_NEED_MORE_SPACE) { len = len << 1; goto retry; } parser = XML_ParserCreate(NULL); if (parser == NULL) { log_warnx("unable to create XML parser"); free(str); return (NULL); } XML_SetUserData(parser, &devlist); XML_SetElementHandler(parser, cctl_start_element, cctl_end_element); XML_SetCharacterDataHandler(parser, cctl_char_handler); retval = XML_Parse(parser, str, strlen(str), 1); XML_ParserFree(parser); free(str); if (retval != 1) { log_warnx("XML_Parse failed"); return (NULL); } str = NULL; len = 4096; retry_port: str = realloc(str, len); if (str == NULL) log_err(1, "realloc"); bzero(&list, sizeof(list)); list.alloc_len = len; list.status = CTL_LUN_LIST_NONE; list.lun_xml = str; if (ioctl(ctl_fd, CTL_PORT_LIST, &list) == -1) { log_warn("error issuing CTL_PORT_LIST ioctl"); free(str); return (NULL); } if (list.status == CTL_LUN_LIST_ERROR) { log_warnx("error returned from CTL_PORT_LIST ioctl: %s", list.error_str); free(str); return (NULL); } if (list.status == CTL_LUN_LIST_NEED_MORE_SPACE) { len = len << 1; goto retry_port; } parser = XML_ParserCreate(NULL); if (parser == NULL) { log_warnx("unable to create XML parser"); free(str); return (NULL); } XML_SetUserData(parser, &devlist); XML_SetElementHandler(parser, cctl_start_pelement, cctl_end_pelement); XML_SetCharacterDataHandler(parser, cctl_char_handler); retval = XML_Parse(parser, str, strlen(str), 1); XML_ParserFree(parser); free(str); if (retval != 1) { log_warnx("XML_Parse failed"); return (NULL); } conf = conf_new(); name = NULL; STAILQ_FOREACH(port, &devlist.port_list, links) { if (strcmp(port->port_frontend, "ha") == 0) continue; free(name); if (port->pp == 0 && port->vp == 0) { name = checked_strdup(port->port_name); } else if (port->vp == 0) { retval = asprintf(&name, "%s/%d", port->port_name, port->pp); if (retval <= 0) log_err(1, "asprintf"); } else { retval = asprintf(&name, "%s/%d/%d", port->port_name, port->pp, port->vp); if (retval <= 0) log_err(1, "asprintf"); } if (port->cfiscsi_target == NULL) { log_debugx("CTL port %u \"%s\" wasn't managed by ctld; ", port->port_id, name); pp = pport_find(conf, name); if (pp == NULL) { #if 0 log_debugx("found new kernel port %u \"%s\"", port->port_id, name); #endif pp = pport_new(conf, name, port->port_id); if (pp == NULL) { log_warnx("pport_new failed"); continue; } } continue; } if (port->cfiscsi_state != 1) { log_debugx("CTL port %ju is not active (%d); ignoring", (uintmax_t)port->port_id, port->cfiscsi_state); continue; } targ = target_find(conf, port->cfiscsi_target); if (targ == NULL) { #if 0 log_debugx("found new kernel target %s for CTL port %ld", port->cfiscsi_target, port->port_id); #endif targ = target_new(conf, port->cfiscsi_target); if (targ == NULL) { log_warnx("target_new failed"); continue; } } if (port->ctld_portal_group_name == NULL) continue; pg = portal_group_find(conf, port->ctld_portal_group_name); if (pg == NULL) { #if 0 log_debugx("found new kernel portal group %s for CTL port %ld", port->ctld_portal_group_name, port->port_id); #endif pg = portal_group_new(conf, port->ctld_portal_group_name); if (pg == NULL) { log_warnx("portal_group_new failed"); continue; } } pg->pg_tag = port->cfiscsi_portal_group_tag; cp = port_new(conf, targ, pg); if (cp == NULL) { log_warnx("port_new failed"); continue; } cp->p_ctl_port = port->port_id; } free(name); STAILQ_FOREACH(lun, &devlist.lun_list, links) { struct cctl_lun_nv *nv; if (lun->ctld_name == NULL) { log_debugx("CTL lun %ju wasn't managed by ctld; " "ignoring", (uintmax_t)lun->lun_id); continue; } cl = lun_find(conf, lun->ctld_name); if (cl != NULL) { log_warnx("found CTL lun %ju \"%s\", " "also backed by CTL lun %d; ignoring", (uintmax_t)lun->lun_id, lun->ctld_name, cl->l_ctl_lun); continue; } log_debugx("found CTL lun %ju \"%s\"", (uintmax_t)lun->lun_id, lun->ctld_name); cl = lun_new(conf, lun->ctld_name); if (cl == NULL) { log_warnx("lun_new failed"); continue; } lun_set_backend(cl, lun->backend_type); lun_set_device_type(cl, lun->device_type); lun_set_blocksize(cl, lun->blocksize); lun_set_device_id(cl, lun->device_id); lun_set_serial(cl, lun->serial_number); lun_set_size(cl, lun->size_blocks * cl->l_blocksize); lun_set_ctl_lun(cl, lun->lun_id); STAILQ_FOREACH(nv, &lun->attr_list, links) { if (strcmp(nv->name, "file") == 0 || strcmp(nv->name, "dev") == 0) { lun_set_path(cl, nv->value); continue; } o = option_new(&cl->l_options, nv->name, nv->value); if (o == NULL) log_warnx("unable to add CTL lun option %s " "for CTL lun %ju \"%s\"", nv->name, (uintmax_t) lun->lun_id, cl->l_name); } } return (conf); } -static void -str_arg(struct ctl_be_arg *arg, const char *name, const char *value) -{ - - arg->namelen = strlen(name) + 1; - arg->name = __DECONST(char *, name); - arg->vallen = strlen(value) + 1; - arg->value = __DECONST(char *, value); - arg->flags = CTL_BEARG_ASCII | CTL_BEARG_RD; -} - int kernel_lun_add(struct lun *lun) { struct option *o; struct ctl_lun_req req; - int error, i, num_options; + int error; bzero(&req, sizeof(req)); strlcpy(req.backend, lun->l_backend, sizeof(req.backend)); req.reqtype = CTL_LUNREQ_CREATE; req.reqdata.create.blocksize_bytes = lun->l_blocksize; if (lun->l_size != 0) req.reqdata.create.lun_size_bytes = lun->l_size; if (lun->l_ctl_lun >= 0) { req.reqdata.create.req_lun_id = lun->l_ctl_lun; req.reqdata.create.flags |= CTL_LUN_FLAG_ID_REQ; } req.reqdata.create.flags |= CTL_LUN_FLAG_DEV_TYPE; req.reqdata.create.device_type = lun->l_device_type; if (lun->l_serial != NULL) { strncpy(req.reqdata.create.serial_num, lun->l_serial, sizeof(req.reqdata.create.serial_num)); req.reqdata.create.flags |= CTL_LUN_FLAG_SERIAL_NUM; } if (lun->l_device_id != NULL) { strncpy(req.reqdata.create.device_id, lun->l_device_id, sizeof(req.reqdata.create.device_id)); req.reqdata.create.flags |= CTL_LUN_FLAG_DEVID; } if (lun->l_path != NULL) { o = option_find(&lun->l_options, "file"); if (o != NULL) { option_set(o, lun->l_path); } else { o = option_new(&lun->l_options, "file", lun->l_path); assert(o != NULL); } } o = option_find(&lun->l_options, "ctld_name"); if (o != NULL) { option_set(o, lun->l_name); } else { o = option_new(&lun->l_options, "ctld_name", lun->l_name); assert(o != NULL); } o = option_find(&lun->l_options, "scsiname"); if (o == NULL && lun->l_scsiname != NULL) { o = option_new(&lun->l_options, "scsiname", lun->l_scsiname); assert(o != NULL); } - num_options = 0; - TAILQ_FOREACH(o, &lun->l_options, o_next) - num_options++; - - req.num_be_args = num_options; - if (num_options > 0) { - req.be_args = malloc(num_options * sizeof(*req.be_args)); - if (req.be_args == NULL) { - log_warn("error allocating %zd bytes", - num_options * sizeof(*req.be_args)); + if (!TAILQ_EMPTY(&lun->l_options)) { + req.args_nvl = nvlist_create(0); + if (req.args_nvl == NULL) { + log_warn("error allocating nvlist"); return (1); } - i = 0; - TAILQ_FOREACH(o, &lun->l_options, o_next) { - str_arg(&req.be_args[i], o->o_name, o->o_value); - i++; + TAILQ_FOREACH(o, &lun->l_options, o_next) + nvlist_add_string(req.args_nvl, o->o_name, o->o_value); + + req.args = nvlist_pack(req.args_nvl, &req.args_len); + if (req.args == NULL) { + log_warn("error packing nvlist"); + return (1); } - assert(i == num_options); } error = ioctl(ctl_fd, CTL_LUN_REQ, &req); - free(req.be_args); + nvlist_destroy(req.args_nvl); + if (error != 0) { log_warn("error issuing CTL_LUN_REQ ioctl"); return (1); } switch (req.status) { case CTL_LUN_ERROR: log_warnx("LUN creation error: %s", req.error_str); return (1); case CTL_LUN_WARNING: log_warnx("LUN creation warning: %s", req.error_str); break; case CTL_LUN_OK: break; default: log_warnx("unknown LUN creation status: %d", req.status); return (1); } lun_set_ctl_lun(lun, req.reqdata.create.req_lun_id); return (0); } int kernel_lun_modify(struct lun *lun) { struct option *o; struct ctl_lun_req req; - int error, i, num_options; + int error; bzero(&req, sizeof(req)); strlcpy(req.backend, lun->l_backend, sizeof(req.backend)); req.reqtype = CTL_LUNREQ_MODIFY; req.reqdata.modify.lun_id = lun->l_ctl_lun; req.reqdata.modify.lun_size_bytes = lun->l_size; - num_options = 0; - TAILQ_FOREACH(o, &lun->l_options, o_next) - num_options++; - - req.num_be_args = num_options; - if (num_options > 0) { - req.be_args = malloc(num_options * sizeof(*req.be_args)); - if (req.be_args == NULL) { - log_warn("error allocating %zd bytes", - num_options * sizeof(*req.be_args)); + if (!TAILQ_EMPTY(&lun->l_options)) { + req.args_nvl = nvlist_create(0); + if (req.args_nvl == NULL) { + log_warn("error allocating nvlist"); return (1); } - i = 0; - TAILQ_FOREACH(o, &lun->l_options, o_next) { - str_arg(&req.be_args[i], o->o_name, o->o_value); - i++; + TAILQ_FOREACH(o, &lun->l_options, o_next) + nvlist_add_string(req.args_nvl, o->o_name, o->o_value); + + req.args = nvlist_pack(req.args_nvl, &req.args_len); + if (req.args == NULL) { + log_warn("error packing nvlist"); + return (1); } - assert(i == num_options); } error = ioctl(ctl_fd, CTL_LUN_REQ, &req); - free(req.be_args); + nvlist_destroy(req.args_nvl); + if (error != 0) { log_warn("error issuing CTL_LUN_REQ ioctl"); return (1); } switch (req.status) { case CTL_LUN_ERROR: log_warnx("LUN modification error: %s", req.error_str); return (1); case CTL_LUN_WARNING: log_warnx("LUN modification warning: %s", req.error_str); break; case CTL_LUN_OK: break; default: log_warnx("unknown LUN modification status: %d", req.status); return (1); } return (0); } int kernel_lun_remove(struct lun *lun) { struct ctl_lun_req req; bzero(&req, sizeof(req)); strlcpy(req.backend, lun->l_backend, sizeof(req.backend)); req.reqtype = CTL_LUNREQ_RM; req.reqdata.rm.lun_id = lun->l_ctl_lun; if (ioctl(ctl_fd, CTL_LUN_REQ, &req) == -1) { log_warn("error issuing CTL_LUN_REQ ioctl"); return (1); } switch (req.status) { case CTL_LUN_ERROR: log_warnx("LUN removal error: %s", req.error_str); return (1); case CTL_LUN_WARNING: log_warnx("LUN removal warning: %s", req.error_str); break; case CTL_LUN_OK: break; default: log_warnx("unknown LUN removal status: %d", req.status); return (1); } return (0); } void kernel_handoff(struct connection *conn) { struct ctl_iscsi req; bzero(&req, sizeof(req)); req.type = CTL_ISCSI_HANDOFF; strlcpy(req.data.handoff.initiator_name, conn->conn_initiator_name, sizeof(req.data.handoff.initiator_name)); strlcpy(req.data.handoff.initiator_addr, conn->conn_initiator_addr, sizeof(req.data.handoff.initiator_addr)); if (conn->conn_initiator_alias != NULL) { strlcpy(req.data.handoff.initiator_alias, conn->conn_initiator_alias, sizeof(req.data.handoff.initiator_alias)); } memcpy(req.data.handoff.initiator_isid, conn->conn_initiator_isid, sizeof(req.data.handoff.initiator_isid)); strlcpy(req.data.handoff.target_name, conn->conn_target->t_name, sizeof(req.data.handoff.target_name)); if (conn->conn_portal->p_portal_group->pg_offload != NULL) { strlcpy(req.data.handoff.offload, conn->conn_portal->p_portal_group->pg_offload, sizeof(req.data.handoff.offload)); } #ifdef ICL_KERNEL_PROXY if (proxy_mode) req.data.handoff.connection_id = conn->conn_socket; else req.data.handoff.socket = conn->conn_socket; #else req.data.handoff.socket = conn->conn_socket; #endif req.data.handoff.portal_group_tag = conn->conn_portal->p_portal_group->pg_tag; if (conn->conn_header_digest == CONN_DIGEST_CRC32C) req.data.handoff.header_digest = CTL_ISCSI_DIGEST_CRC32C; if (conn->conn_data_digest == CONN_DIGEST_CRC32C) req.data.handoff.data_digest = CTL_ISCSI_DIGEST_CRC32C; req.data.handoff.cmdsn = conn->conn_cmdsn; req.data.handoff.statsn = conn->conn_statsn; req.data.handoff.max_recv_data_segment_length = conn->conn_max_recv_data_segment_length; req.data.handoff.max_send_data_segment_length = conn->conn_max_send_data_segment_length; req.data.handoff.max_burst_length = conn->conn_max_burst_length; req.data.handoff.first_burst_length = conn->conn_first_burst_length; req.data.handoff.immediate_data = conn->conn_immediate_data; if (ioctl(ctl_fd, CTL_ISCSI, &req) == -1) { log_err(1, "error issuing CTL_ISCSI ioctl; " "dropping connection"); } if (req.status != CTL_ISCSI_OK) { log_errx(1, "error returned from CTL iSCSI handoff request: " "%s; dropping connection", req.error_str); } } void kernel_limits(const char *offload, int *max_recv_dsl, int *max_send_dsl, int *max_burst_length, int *first_burst_length) { struct ctl_iscsi req; struct ctl_iscsi_limits_params *cilp; bzero(&req, sizeof(req)); req.type = CTL_ISCSI_LIMITS; cilp = (struct ctl_iscsi_limits_params *)&(req.data.limits); if (offload != NULL) { strlcpy(cilp->offload, offload, sizeof(cilp->offload)); } if (ioctl(ctl_fd, CTL_ISCSI, &req) == -1) { log_err(1, "error issuing CTL_ISCSI ioctl; " "dropping connection"); } if (req.status != CTL_ISCSI_OK) { log_errx(1, "error returned from CTL iSCSI limits request: " "%s; dropping connection", req.error_str); } if (cilp->max_recv_data_segment_length != 0) { *max_recv_dsl = cilp->max_recv_data_segment_length; *max_send_dsl = cilp->max_recv_data_segment_length; } if (cilp->max_send_data_segment_length != 0) *max_send_dsl = cilp->max_send_data_segment_length; if (cilp->max_burst_length != 0) *max_burst_length = cilp->max_burst_length; if (cilp->first_burst_length != 0) *first_burst_length = cilp->first_burst_length; if (*max_burst_length < *first_burst_length) *first_burst_length = *max_burst_length; if (offload != NULL) { log_debugx("Kernel limits for offload \"%s\" are " "MaxRecvDataSegment=%d, max_send_dsl=%d, " "MaxBurstLength=%d, FirstBurstLength=%d", offload, *max_recv_dsl, *max_send_dsl, *max_burst_length, *first_burst_length); } else { log_debugx("Kernel limits are " "MaxRecvDataSegment=%d, max_send_dsl=%d, " "MaxBurstLength=%d, FirstBurstLength=%d", *max_recv_dsl, *max_send_dsl, *max_burst_length, *first_burst_length); } } int kernel_port_add(struct port *port) { struct option *o; struct ctl_port_entry entry; struct ctl_req req; struct ctl_lun_map lm; struct target *targ = port->p_target; struct portal_group *pg = port->p_portal_group; - char tagstr[16]; - int error, i, n; + char result_buf[NVLIST_BUFSIZE]; + int error, i; /* Create iSCSI port. */ - if (port->p_portal_group) { + if (port->p_portal_group || port->p_ioctl_port) { bzero(&req, sizeof(req)); - strlcpy(req.driver, "iscsi", sizeof(req.driver)); req.reqtype = CTL_REQ_CREATE; - req.num_args = 5; - TAILQ_FOREACH(o, &pg->pg_options, o_next) - req.num_args++; - req.args = malloc(req.num_args * sizeof(*req.args)); - if (req.args == NULL) - log_err(1, "malloc"); - n = 0; - req.args[n].namelen = sizeof("port_id"); - req.args[n].name = __DECONST(char *, "port_id"); - req.args[n].vallen = sizeof(port->p_ctl_port); - req.args[n].value = &port->p_ctl_port; - req.args[n++].flags = CTL_BEARG_WR; - str_arg(&req.args[n++], "cfiscsi_target", targ->t_name); - snprintf(tagstr, sizeof(tagstr), "%d", pg->pg_tag); - str_arg(&req.args[n++], "cfiscsi_portal_group_tag", tagstr); - if (targ->t_alias) - str_arg(&req.args[n++], "cfiscsi_target_alias", targ->t_alias); - str_arg(&req.args[n++], "ctld_portal_group_name", pg->pg_name); - TAILQ_FOREACH(o, &pg->pg_options, o_next) - str_arg(&req.args[n++], o->o_name, o->o_value); - req.num_args = n; + + if (port->p_portal_group) { + strlcpy(req.driver, "iscsi", sizeof(req.driver)); + req.args_nvl = nvlist_create(0); + nvlist_add_string(req.args_nvl, "cfiscsi_target", + targ->t_name); + nvlist_add_string(req.args_nvl, + "ctld_portal_group_name", pg->pg_name); + nvlist_add_stringf(req.args_nvl, + "cfiscsi_portal_group_tag", "%u", pg->pg_tag); + + if (targ->t_alias) { + nvlist_add_string(req.args_nvl, + "cfiscsi_target_alias", targ->t_alias); + } + + TAILQ_FOREACH(o, &pg->pg_options, o_next) + nvlist_add_string(req.args_nvl, o->o_name, + o->o_value); + } + + if (port->p_ioctl_port) { + strlcpy(req.driver, "ioctl", sizeof(req.driver)); + req.args_nvl = nvlist_create(0); + nvlist_add_stringf(req.args_nvl, "pp", "%d", + port->p_ioctl_pp); + nvlist_add_stringf(req.args_nvl, "vp", "%d", + port->p_ioctl_vp); + } + + req.args = nvlist_pack(req.args_nvl, &req.args_len); + if (req.args == NULL) { + log_warn("error packing nvlist"); + return (1); + } + + req.result = result_buf; + req.result_len = sizeof(result_buf); error = ioctl(ctl_fd, CTL_PORT_REQ, &req); - free(req.args); + nvlist_destroy(req.args_nvl); + if (error != 0) { log_warn("error issuing CTL_PORT_REQ ioctl"); return (1); } if (req.status == CTL_LUN_ERROR) { log_warnx("error returned from port creation request: %s", req.error_str); return (1); } if (req.status != CTL_LUN_OK) { log_warnx("unknown port creation request status %d", req.status); return (1); } + + req.result_nvl = nvlist_unpack(result_buf, req.result_len, 0); + if (req.result_nvl == NULL) { + log_warnx("error unpacking result nvlist"); + return (1); + } + + port->p_ctl_port = nvlist_get_number(req.result_nvl, "port_id"); + nvlist_destroy(req.result_nvl); } else if (port->p_pport) { port->p_ctl_port = port->p_pport->pp_ctl_port; if (strncmp(targ->t_name, "naa.", 4) == 0 && strlen(targ->t_name) == 20) { bzero(&entry, sizeof(entry)); entry.port_type = CTL_PORT_NONE; entry.targ_port = port->p_ctl_port; entry.flags |= CTL_PORT_WWNN_VALID; entry.wwnn = strtoull(targ->t_name + 4, NULL, 16); if (ioctl(ctl_fd, CTL_SET_PORT_WWNS, &entry) == -1) log_warn("CTL_SET_PORT_WWNS ioctl failed"); } } /* Explicitly enable mapping to block any access except allowed. */ lm.port = port->p_ctl_port; lm.plun = UINT32_MAX; lm.lun = 0; error = ioctl(ctl_fd, CTL_LUN_MAP, &lm); if (error != 0) log_warn("CTL_LUN_MAP ioctl failed"); /* Map configured LUNs */ for (i = 0; i < MAX_LUNS; i++) { if (targ->t_luns[i] == NULL) continue; lm.port = port->p_ctl_port; lm.plun = i; lm.lun = targ->t_luns[i]->l_ctl_lun; error = ioctl(ctl_fd, CTL_LUN_MAP, &lm); if (error != 0) log_warn("CTL_LUN_MAP ioctl failed"); } /* Enable port */ bzero(&entry, sizeof(entry)); entry.targ_port = port->p_ctl_port; error = ioctl(ctl_fd, CTL_ENABLE_PORT, &entry); if (error != 0) { log_warn("CTL_ENABLE_PORT ioctl failed"); return (-1); } return (0); } int kernel_port_update(struct port *port, struct port *oport) { struct ctl_lun_map lm; struct target *targ = port->p_target; struct target *otarg = oport->p_target; int error, i; uint32_t olun; /* Map configured LUNs and unmap others */ for (i = 0; i < MAX_LUNS; i++) { lm.port = port->p_ctl_port; lm.plun = i; if (targ->t_luns[i] == NULL) lm.lun = UINT32_MAX; else lm.lun = targ->t_luns[i]->l_ctl_lun; if (otarg->t_luns[i] == NULL) olun = UINT32_MAX; else olun = otarg->t_luns[i]->l_ctl_lun; if (lm.lun == olun) continue; error = ioctl(ctl_fd, CTL_LUN_MAP, &lm); if (error != 0) log_warn("CTL_LUN_MAP ioctl failed"); } return (0); } int kernel_port_remove(struct port *port) { struct ctl_port_entry entry; struct ctl_lun_map lm; struct ctl_req req; - char tagstr[16]; struct target *targ = port->p_target; struct portal_group *pg = port->p_portal_group; int error; /* Disable port */ bzero(&entry, sizeof(entry)); entry.targ_port = port->p_ctl_port; error = ioctl(ctl_fd, CTL_DISABLE_PORT, &entry); if (error != 0) { log_warn("CTL_DISABLE_PORT ioctl failed"); return (-1); } - /* Remove iSCSI port. */ - if (port->p_portal_group) { + /* Remove iSCSI or ioctl port. */ + if (port->p_portal_group || port->p_ioctl_port) { bzero(&req, sizeof(req)); - strlcpy(req.driver, "iscsi", sizeof(req.driver)); + strlcpy(req.driver, port->p_ioctl_port ? "ioctl" : "iscsi", + sizeof(req.driver)); req.reqtype = CTL_REQ_REMOVE; - req.num_args = 2; - req.args = malloc(req.num_args * sizeof(*req.args)); - if (req.args == NULL) - log_err(1, "malloc"); - str_arg(&req.args[0], "cfiscsi_target", targ->t_name); - snprintf(tagstr, sizeof(tagstr), "%d", pg->pg_tag); - str_arg(&req.args[1], "cfiscsi_portal_group_tag", tagstr); + req.args_nvl = nvlist_create(0); + if (req.args_nvl == NULL) + log_err(1, "nvlist_create"); + + if (port->p_ioctl_port) + nvlist_add_stringf(req.args_nvl, "port_id", "%d", + port->p_ctl_port); + else { + nvlist_add_string(req.args_nvl, "cfiscsi_target", + targ->t_name); + nvlist_add_stringf(req.args_nvl, + "cfiscsi_portal_group_tag", "%u", pg->pg_tag); + } + + req.args = nvlist_pack(req.args_nvl, &req.args_len); + if (req.args == NULL) { + log_warn("error packing nvlist"); + return (1); + } + error = ioctl(ctl_fd, CTL_PORT_REQ, &req); - free(req.args); + nvlist_destroy(req.args_nvl); + if (error != 0) { log_warn("error issuing CTL_PORT_REQ ioctl"); return (1); } if (req.status == CTL_LUN_ERROR) { log_warnx("error returned from port removal request: %s", req.error_str); return (1); } if (req.status != CTL_LUN_OK) { log_warnx("unknown port removal request status %d", req.status); return (1); } } else { /* Disable LUN mapping. */ lm.port = port->p_ctl_port; lm.plun = UINT32_MAX; lm.lun = UINT32_MAX; error = ioctl(ctl_fd, CTL_LUN_MAP, &lm); if (error != 0) log_warn("CTL_LUN_MAP ioctl failed"); } return (0); } #ifdef ICL_KERNEL_PROXY void kernel_listen(struct addrinfo *ai, bool iser, int portal_id) { struct ctl_iscsi req; bzero(&req, sizeof(req)); req.type = CTL_ISCSI_LISTEN; req.data.listen.iser = iser; req.data.listen.domain = ai->ai_family; req.data.listen.socktype = ai->ai_socktype; req.data.listen.protocol = ai->ai_protocol; req.data.listen.addr = ai->ai_addr; req.data.listen.addrlen = ai->ai_addrlen; req.data.listen.portal_id = portal_id; if (ioctl(ctl_fd, CTL_ISCSI, &req) == -1) log_err(1, "error issuing CTL_ISCSI ioctl"); if (req.status != CTL_ISCSI_OK) { log_errx(1, "error returned from CTL iSCSI listen: %s", req.error_str); } } void kernel_accept(int *connection_id, int *portal_id, struct sockaddr *client_sa, socklen_t *client_salen) { struct ctl_iscsi req; struct sockaddr_storage ss; bzero(&req, sizeof(req)); req.type = CTL_ISCSI_ACCEPT; req.data.accept.initiator_addr = (struct sockaddr *)&ss; if (ioctl(ctl_fd, CTL_ISCSI, &req) == -1) log_err(1, "error issuing CTL_ISCSI ioctl"); if (req.status != CTL_ISCSI_OK) { log_errx(1, "error returned from CTL iSCSI accept: %s", req.error_str); } *connection_id = req.data.accept.connection_id; *portal_id = req.data.accept.portal_id; *client_salen = req.data.accept.initiator_addrlen; memcpy(client_sa, &ss, *client_salen); } void kernel_send(struct pdu *pdu) { struct ctl_iscsi req; bzero(&req, sizeof(req)); req.type = CTL_ISCSI_SEND; req.data.send.connection_id = pdu->pdu_connection->conn_socket; req.data.send.bhs = pdu->pdu_bhs; req.data.send.data_segment_len = pdu->pdu_data_len; req.data.send.data_segment = pdu->pdu_data; if (ioctl(ctl_fd, CTL_ISCSI, &req) == -1) { log_err(1, "error issuing CTL_ISCSI ioctl; " "dropping connection"); } if (req.status != CTL_ISCSI_OK) { log_errx(1, "error returned from CTL iSCSI send: " "%s; dropping connection", req.error_str); } } void kernel_receive(struct pdu *pdu) { struct connection *conn; struct ctl_iscsi req; conn = pdu->pdu_connection; pdu->pdu_data = malloc(conn->conn_max_recv_data_segment_length); if (pdu->pdu_data == NULL) log_err(1, "malloc"); bzero(&req, sizeof(req)); req.type = CTL_ISCSI_RECEIVE; req.data.receive.connection_id = conn->conn_socket; req.data.receive.bhs = pdu->pdu_bhs; req.data.receive.data_segment_len = conn->conn_max_recv_data_segment_length; req.data.receive.data_segment = pdu->pdu_data; if (ioctl(ctl_fd, CTL_ISCSI, &req) == -1) { log_err(1, "error issuing CTL_ISCSI ioctl; " "dropping connection"); } if (req.status != CTL_ISCSI_OK) { log_errx(1, "error returned from CTL iSCSI receive: " "%s; dropping connection", req.error_str); } } #endif /* ICL_KERNEL_PROXY */ /* * XXX: I CANT INTO LATIN */ void kernel_capsicate(void) { int error; cap_rights_t rights; const unsigned long cmds[] = { CTL_ISCSI }; cap_rights_init(&rights, CAP_IOCTL); error = cap_rights_limit(ctl_fd, &rights); if (error != 0 && errno != ENOSYS) log_err(1, "cap_rights_limit"); error = cap_ioctls_limit(ctl_fd, cmds, nitems(cmds)); if (error != 0 && errno != ENOSYS) log_err(1, "cap_ioctls_limit"); error = cap_enter(); if (error != 0 && errno != ENOSYS) log_err(1, "cap_enter"); if (cap_sandboxed()) log_debugx("Capsicum capability mode enabled"); else log_warnx("Capsicum capability mode not supported"); } Index: head/usr.sbin/ctld/parse.y =================================================================== --- head/usr.sbin/ctld/parse.y (revision 333445) +++ head/usr.sbin/ctld/parse.y (revision 333446) @@ -1,1081 +1,1094 @@ %{ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 The FreeBSD Foundation * All rights reserved. * * This software was developed by Edward Tomasz Napierala under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include #include "ctld.h" extern FILE *yyin; extern char *yytext; extern int lineno; static struct conf *conf = NULL; static struct auth_group *auth_group = NULL; static struct portal_group *portal_group = NULL; static struct target *target = NULL; static struct lun *lun = NULL; extern void yyerror(const char *); extern int yylex(void); extern void yyrestart(FILE *); %} %token ALIAS AUTH_GROUP AUTH_TYPE BACKEND BLOCKSIZE CHAP CHAP_MUTUAL %token CLOSING_BRACKET CTL_LUN DEBUG DEVICE_ID DEVICE_TYPE %token DISCOVERY_AUTH_GROUP DISCOVERY_FILTER FOREIGN %token INITIATOR_NAME INITIATOR_PORTAL ISNS_SERVER ISNS_PERIOD ISNS_TIMEOUT %token LISTEN LISTEN_ISER LUN MAXPROC OFFLOAD OPENING_BRACKET OPTION %token PATH PIDFILE PORT PORTAL_GROUP REDIRECT SEMICOLON SERIAL SIZE STR %token TAG TARGET TIMEOUT %union { char *str; } %token STR %% statements: | statements statement | statements statement SEMICOLON ; statement: debug | timeout | maxproc | pidfile | isns_server | isns_period | isns_timeout | auth_group | portal_group | lun | target ; debug: DEBUG STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } conf->conf_debug = tmp; } ; timeout: TIMEOUT STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } conf->conf_timeout = tmp; } ; maxproc: MAXPROC STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } conf->conf_maxproc = tmp; } ; pidfile: PIDFILE STR { if (conf->conf_pidfile_path != NULL) { log_warnx("pidfile specified more than once"); free($2); return (1); } conf->conf_pidfile_path = $2; } ; isns_server: ISNS_SERVER STR { int error; error = isns_new(conf, $2); free($2); if (error != 0) return (1); } ; isns_period: ISNS_PERIOD STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } conf->conf_isns_period = tmp; } ; isns_timeout: ISNS_TIMEOUT STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } conf->conf_isns_timeout = tmp; } ; auth_group: AUTH_GROUP auth_group_name OPENING_BRACKET auth_group_entries CLOSING_BRACKET { auth_group = NULL; } ; auth_group_name: STR { /* * Make it possible to redefine default * auth-group. but only once. */ if (strcmp($1, "default") == 0 && conf->conf_default_ag_defined == false) { auth_group = auth_group_find(conf, $1); conf->conf_default_ag_defined = true; } else { auth_group = auth_group_new(conf, $1); } free($1); if (auth_group == NULL) return (1); } ; auth_group_entries: | auth_group_entries auth_group_entry | auth_group_entries auth_group_entry SEMICOLON ; auth_group_entry: auth_group_auth_type | auth_group_chap | auth_group_chap_mutual | auth_group_initiator_name | auth_group_initiator_portal ; auth_group_auth_type: AUTH_TYPE STR { int error; error = auth_group_set_type(auth_group, $2); free($2); if (error != 0) return (1); } ; auth_group_chap: CHAP STR STR { const struct auth *ca; ca = auth_new_chap(auth_group, $2, $3); free($2); free($3); if (ca == NULL) return (1); } ; auth_group_chap_mutual: CHAP_MUTUAL STR STR STR STR { const struct auth *ca; ca = auth_new_chap_mutual(auth_group, $2, $3, $4, $5); free($2); free($3); free($4); free($5); if (ca == NULL) return (1); } ; auth_group_initiator_name: INITIATOR_NAME STR { const struct auth_name *an; an = auth_name_new(auth_group, $2); free($2); if (an == NULL) return (1); } ; auth_group_initiator_portal: INITIATOR_PORTAL STR { const struct auth_portal *ap; ap = auth_portal_new(auth_group, $2); free($2); if (ap == NULL) return (1); } ; portal_group: PORTAL_GROUP portal_group_name OPENING_BRACKET portal_group_entries CLOSING_BRACKET { portal_group = NULL; } ; portal_group_name: STR { /* * Make it possible to redefine default * portal-group. but only once. */ if (strcmp($1, "default") == 0 && conf->conf_default_pg_defined == false) { portal_group = portal_group_find(conf, $1); conf->conf_default_pg_defined = true; } else { portal_group = portal_group_new(conf, $1); } free($1); if (portal_group == NULL) return (1); } ; portal_group_entries: | portal_group_entries portal_group_entry | portal_group_entries portal_group_entry SEMICOLON ; portal_group_entry: portal_group_discovery_auth_group | portal_group_discovery_filter | portal_group_foreign | portal_group_listen | portal_group_listen_iser | portal_group_offload | portal_group_option | portal_group_redirect | portal_group_tag ; portal_group_discovery_auth_group: DISCOVERY_AUTH_GROUP STR { if (portal_group->pg_discovery_auth_group != NULL) { log_warnx("discovery-auth-group for portal-group " "\"%s\" specified more than once", portal_group->pg_name); return (1); } portal_group->pg_discovery_auth_group = auth_group_find(conf, $2); if (portal_group->pg_discovery_auth_group == NULL) { log_warnx("unknown discovery-auth-group \"%s\" " "for portal-group \"%s\"", $2, portal_group->pg_name); return (1); } free($2); } ; portal_group_discovery_filter: DISCOVERY_FILTER STR { int error; error = portal_group_set_filter(portal_group, $2); free($2); if (error != 0) return (1); } ; portal_group_foreign: FOREIGN { portal_group->pg_foreign = 1; } ; portal_group_listen: LISTEN STR { int error; error = portal_group_add_listen(portal_group, $2, false); free($2); if (error != 0) return (1); } ; portal_group_listen_iser: LISTEN_ISER STR { int error; error = portal_group_add_listen(portal_group, $2, true); free($2); if (error != 0) return (1); } ; portal_group_offload: OFFLOAD STR { int error; error = portal_group_set_offload(portal_group, $2); free($2); if (error != 0) return (1); } ; portal_group_option: OPTION STR STR { struct option *o; o = option_new(&portal_group->pg_options, $2, $3); free($2); free($3); if (o == NULL) return (1); } ; portal_group_redirect: REDIRECT STR { int error; error = portal_group_set_redirection(portal_group, $2); free($2); if (error != 0) return (1); } ; portal_group_tag: TAG STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } portal_group->pg_tag = tmp; } ; lun: LUN lun_name OPENING_BRACKET lun_entries CLOSING_BRACKET { lun = NULL; } ; lun_name: STR { lun = lun_new(conf, $1); free($1); if (lun == NULL) return (1); } ; target: TARGET target_name OPENING_BRACKET target_entries CLOSING_BRACKET { target = NULL; } ; target_name: STR { target = target_new(conf, $1); free($1); if (target == NULL) return (1); } ; target_entries: | target_entries target_entry | target_entries target_entry SEMICOLON ; target_entry: target_alias | target_auth_group | target_auth_type | target_chap | target_chap_mutual | target_initiator_name | target_initiator_portal | target_portal_group | target_port | target_redirect | target_lun | target_lun_ref ; target_alias: ALIAS STR { if (target->t_alias != NULL) { log_warnx("alias for target \"%s\" " "specified more than once", target->t_name); return (1); } target->t_alias = $2; } ; target_auth_group: AUTH_GROUP STR { if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) log_warnx("auth-group for target \"%s\" " "specified more than once", target->t_name); else log_warnx("cannot use both auth-group and explicit " "authorisations for target \"%s\"", target->t_name); return (1); } target->t_auth_group = auth_group_find(conf, $2); if (target->t_auth_group == NULL) { log_warnx("unknown auth-group \"%s\" for target " "\"%s\"", $2, target->t_name); return (1); } free($2); } ; target_auth_type: AUTH_TYPE STR { int error; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "auth-type for target \"%s\"", target->t_name); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) { free($2); return (1); } target->t_auth_group->ag_target = target; } error = auth_group_set_type(target->t_auth_group, $2); free($2); if (error != 0) return (1); } ; target_chap: CHAP STR STR { const struct auth *ca; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "chap for target \"%s\"", target->t_name); free($2); free($3); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) { free($2); free($3); return (1); } target->t_auth_group->ag_target = target; } ca = auth_new_chap(target->t_auth_group, $2, $3); free($2); free($3); if (ca == NULL) return (1); } ; target_chap_mutual: CHAP_MUTUAL STR STR STR STR { const struct auth *ca; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "chap-mutual for target \"%s\"", target->t_name); free($2); free($3); free($4); free($5); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) { free($2); free($3); free($4); free($5); return (1); } target->t_auth_group->ag_target = target; } ca = auth_new_chap_mutual(target->t_auth_group, $2, $3, $4, $5); free($2); free($3); free($4); free($5); if (ca == NULL) return (1); } ; target_initiator_name: INITIATOR_NAME STR { const struct auth_name *an; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "initiator-name for target \"%s\"", target->t_name); free($2); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) { free($2); return (1); } target->t_auth_group->ag_target = target; } an = auth_name_new(target->t_auth_group, $2); free($2); if (an == NULL) return (1); } ; target_initiator_portal: INITIATOR_PORTAL STR { const struct auth_portal *ap; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "initiator-portal for target \"%s\"", target->t_name); free($2); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) { free($2); return (1); } target->t_auth_group->ag_target = target; } ap = auth_portal_new(target->t_auth_group, $2); free($2); if (ap == NULL) return (1); } ; target_portal_group: PORTAL_GROUP STR STR { struct portal_group *tpg; struct auth_group *tag; struct port *tp; tpg = portal_group_find(conf, $2); if (tpg == NULL) { log_warnx("unknown portal-group \"%s\" for target " "\"%s\"", $2, target->t_name); free($2); free($3); return (1); } tag = auth_group_find(conf, $3); if (tag == NULL) { log_warnx("unknown auth-group \"%s\" for target " "\"%s\"", $3, target->t_name); free($2); free($3); return (1); } tp = port_new(conf, target, tpg); if (tp == NULL) { log_warnx("can't link portal-group \"%s\" to target " "\"%s\"", $2, target->t_name); free($2); return (1); } tp->p_auth_group = tag; free($2); free($3); } | PORTAL_GROUP STR { struct portal_group *tpg; struct port *tp; tpg = portal_group_find(conf, $2); if (tpg == NULL) { log_warnx("unknown portal-group \"%s\" for target " "\"%s\"", $2, target->t_name); free($2); return (1); } tp = port_new(conf, target, tpg); if (tp == NULL) { log_warnx("can't link portal-group \"%s\" to target " "\"%s\"", $2, target->t_name); free($2); return (1); } free($2); } ; target_port: PORT STR { struct pport *pp; struct port *tp; + int ret, i_pp, i_vp = 0; - pp = pport_find(conf, $2); - if (pp == NULL) { - log_warnx("unknown port \"%s\" for target \"%s\"", - $2, target->t_name); - free($2); - return (1); + ret = sscanf($2, "ioctl/%d/%d", &i_pp, &i_vp); + if (ret > 0) { + tp = port_new_ioctl(conf, target, i_pp, i_vp); + if (tp == NULL) { + log_warnx("can't create new ioctl port for " + "target \"%s\"", target->t_name); + free($2); + return (1); + } + } else { + pp = pport_find(conf, $2); + if (pp == NULL) { + log_warnx("unknown port \"%s\" for target \"%s\"", + $2, target->t_name); + free($2); + return (1); + } + if (!TAILQ_EMPTY(&pp->pp_ports)) { + log_warnx("can't link port \"%s\" to target \"%s\", " + "port already linked to some target", + $2, target->t_name); + free($2); + return (1); + } + tp = port_new_pp(conf, target, pp); + if (tp == NULL) { + log_warnx("can't link port \"%s\" to target \"%s\"", + $2, target->t_name); + free($2); + return (1); + } } - if (!TAILQ_EMPTY(&pp->pp_ports)) { - log_warnx("can't link port \"%s\" to target \"%s\", " - "port already linked to some target", - $2, target->t_name); - free($2); - return (1); - } - tp = port_new_pp(conf, target, pp); - if (tp == NULL) { - log_warnx("can't link port \"%s\" to target \"%s\"", - $2, target->t_name); - free($2); - return (1); - } + free($2); } ; target_redirect: REDIRECT STR { int error; error = target_set_redirection(target, $2); free($2); if (error != 0) return (1); } ; target_lun: LUN lun_number OPENING_BRACKET lun_entries CLOSING_BRACKET { lun = NULL; } ; lun_number: STR { uint64_t tmp; int ret; char *name; if (expand_number($1, &tmp) != 0) { yyerror("invalid numeric value"); free($1); return (1); } if (tmp >= MAX_LUNS) { yyerror("LU number is too big"); free($1); return (1); } ret = asprintf(&name, "%s,lun,%ju", target->t_name, tmp); if (ret <= 0) log_err(1, "asprintf"); lun = lun_new(conf, name); if (lun == NULL) return (1); lun_set_scsiname(lun, name); target->t_luns[tmp] = lun; } ; target_lun_ref: LUN STR STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); free($3); return (1); } free($2); if (tmp >= MAX_LUNS) { yyerror("LU number is too big"); free($3); return (1); } lun = lun_find(conf, $3); free($3); if (lun == NULL) return (1); target->t_luns[tmp] = lun; } ; lun_entries: | lun_entries lun_entry | lun_entries lun_entry SEMICOLON ; lun_entry: lun_backend | lun_blocksize | lun_device_id | lun_device_type | lun_ctl_lun | lun_option | lun_path | lun_serial | lun_size ; lun_backend: BACKEND STR { if (lun->l_backend != NULL) { log_warnx("backend for lun \"%s\" " "specified more than once", lun->l_name); free($2); return (1); } lun_set_backend(lun, $2); free($2); } ; lun_blocksize: BLOCKSIZE STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } if (lun->l_blocksize != 0) { log_warnx("blocksize for lun \"%s\" " "specified more than once", lun->l_name); return (1); } lun_set_blocksize(lun, tmp); } ; lun_device_id: DEVICE_ID STR { if (lun->l_device_id != NULL) { log_warnx("device_id for lun \"%s\" " "specified more than once", lun->l_name); free($2); return (1); } lun_set_device_id(lun, $2); free($2); } ; lun_device_type: DEVICE_TYPE STR { uint64_t tmp; if (strcasecmp($2, "disk") == 0 || strcasecmp($2, "direct") == 0) tmp = 0; else if (strcasecmp($2, "processor") == 0) tmp = 3; else if (strcasecmp($2, "cd") == 0 || strcasecmp($2, "cdrom") == 0 || strcasecmp($2, "dvd") == 0 || strcasecmp($2, "dvdrom") == 0) tmp = 5; else if (expand_number($2, &tmp) != 0 || tmp > 15) { yyerror("invalid numeric value"); free($2); return (1); } lun_set_device_type(lun, tmp); } ; lun_ctl_lun: CTL_LUN STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } if (lun->l_ctl_lun >= 0) { log_warnx("ctl_lun for lun \"%s\" " "specified more than once", lun->l_name); return (1); } lun_set_ctl_lun(lun, tmp); } ; lun_option: OPTION STR STR { struct option *o; o = option_new(&lun->l_options, $2, $3); free($2); free($3); if (o == NULL) return (1); } ; lun_path: PATH STR { if (lun->l_path != NULL) { log_warnx("path for lun \"%s\" " "specified more than once", lun->l_name); free($2); return (1); } lun_set_path(lun, $2); free($2); } ; lun_serial: SERIAL STR { if (lun->l_serial != NULL) { log_warnx("serial for lun \"%s\" " "specified more than once", lun->l_name); free($2); return (1); } lun_set_serial(lun, $2); free($2); } ; lun_size: SIZE STR { uint64_t tmp; if (expand_number($2, &tmp) != 0) { yyerror("invalid numeric value"); free($2); return (1); } if (lun->l_size != 0) { log_warnx("size for lun \"%s\" " "specified more than once", lun->l_name); return (1); } lun_set_size(lun, tmp); } ; %% void yyerror(const char *str) { log_warnx("error in configuration file at line %d near '%s': %s", lineno, yytext, str); } int parse_conf(struct conf *newconf, const char *path) { int error; conf = newconf; yyin = fopen(path, "r"); if (yyin == NULL) { log_warn("unable to open configuration file %s", path); return (1); } lineno = 1; yyrestart(yyin); error = yyparse(); auth_group = NULL; portal_group = NULL; target = NULL; lun = NULL; fclose(yyin); return (error); } Index: head/usr.sbin/ctld/uclparse.c =================================================================== --- head/usr.sbin/ctld/uclparse.c (revision 333445) +++ head/usr.sbin/ctld/uclparse.c (revision 333446) @@ -1,918 +1,931 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2015 iXsystems Inc. * All rights reserved. * * This software was developed by Jakub Klama * under sponsorship from iXsystems Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include "ctld.h" static struct conf *conf = NULL; static int uclparse_toplevel(const ucl_object_t *); static int uclparse_chap(struct auth_group *, const ucl_object_t *); static int uclparse_chap_mutual(struct auth_group *, const ucl_object_t *); static int uclparse_lun(const char *, const ucl_object_t *); static int uclparse_auth_group(const char *, const ucl_object_t *); static int uclparse_portal_group(const char *, const ucl_object_t *); static int uclparse_target(const char *, const ucl_object_t *); static int uclparse_target_portal_group(struct target *, const ucl_object_t *); static int uclparse_target_lun(struct target *, const ucl_object_t *); static int uclparse_chap(struct auth_group *auth_group, const ucl_object_t *obj) { const struct auth *ca; const ucl_object_t *user, *secret; user = ucl_object_find_key(obj, "user"); if (!user || user->type != UCL_STRING) { log_warnx("chap section in auth-group \"%s\" is missing " "\"user\" string key", auth_group->ag_name); return (1); } secret = ucl_object_find_key(obj, "secret"); if (!secret || secret->type != UCL_STRING) { log_warnx("chap section in auth-group \"%s\" is missing " "\"secret\" string key", auth_group->ag_name); } ca = auth_new_chap(auth_group, ucl_object_tostring(user), ucl_object_tostring(secret)); if (ca == NULL) return (1); return (0); } static int uclparse_chap_mutual(struct auth_group *auth_group, const ucl_object_t *obj) { const struct auth *ca; const ucl_object_t *user, *secret, *mutual_user; const ucl_object_t *mutual_secret; user = ucl_object_find_key(obj, "user"); if (!user || user->type != UCL_STRING) { log_warnx("chap-mutual section in auth-group \"%s\" is missing " "\"user\" string key", auth_group->ag_name); return (1); } secret = ucl_object_find_key(obj, "secret"); if (!secret || secret->type != UCL_STRING) { log_warnx("chap-mutual section in auth-group \"%s\" is missing " "\"secret\" string key", auth_group->ag_name); return (1); } mutual_user = ucl_object_find_key(obj, "mutual-user"); if (!user || user->type != UCL_STRING) { log_warnx("chap-mutual section in auth-group \"%s\" is missing " "\"mutual-user\" string key", auth_group->ag_name); return (1); } mutual_secret = ucl_object_find_key(obj, "mutual-secret"); if (!secret || secret->type != UCL_STRING) { log_warnx("chap-mutual section in auth-group \"%s\" is missing " "\"mutual-secret\" string key", auth_group->ag_name); return (1); } ca = auth_new_chap_mutual(auth_group, ucl_object_tostring(user), ucl_object_tostring(secret), ucl_object_tostring(mutual_user), ucl_object_tostring(mutual_secret)); if (ca == NULL) return (1); return (0); } static int uclparse_target_portal_group(struct target *target, const ucl_object_t *obj) { struct portal_group *tpg; struct auth_group *tag = NULL; struct port *tp; const ucl_object_t *portal_group, *auth_group; portal_group = ucl_object_find_key(obj, "name"); if (!portal_group || portal_group->type != UCL_STRING) { log_warnx("portal-group section in target \"%s\" is missing " "\"name\" string key", target->t_name); return (1); } auth_group = ucl_object_find_key(obj, "auth-group-name"); if (auth_group && auth_group->type != UCL_STRING) { log_warnx("portal-group section in target \"%s\" is missing " "\"auth-group-name\" string key", target->t_name); return (1); } tpg = portal_group_find(conf, ucl_object_tostring(portal_group)); if (tpg == NULL) { log_warnx("unknown portal-group \"%s\" for target " "\"%s\"", ucl_object_tostring(portal_group), target->t_name); return (1); } if (auth_group) { tag = auth_group_find(conf, ucl_object_tostring(auth_group)); if (tag == NULL) { log_warnx("unknown auth-group \"%s\" for target " "\"%s\"", ucl_object_tostring(auth_group), target->t_name); return (1); } } tp = port_new(conf, target, tpg); if (tp == NULL) { log_warnx("can't link portal-group \"%s\" to target " "\"%s\"", ucl_object_tostring(portal_group), target->t_name); return (1); } tp->p_auth_group = tag; return (0); } static int uclparse_target_lun(struct target *target, const ucl_object_t *obj) { struct lun *lun; uint64_t tmp; if (obj->type == UCL_INT) { char *name; tmp = ucl_object_toint(obj); if (tmp >= MAX_LUNS) { log_warnx("LU number %ju in target \"%s\" is too big", tmp, target->t_name); return (1); } asprintf(&name, "%s,lun,%ju", target->t_name, tmp); lun = lun_new(conf, name); if (lun == NULL) return (1); lun_set_scsiname(lun, name); target->t_luns[tmp] = lun; return (0); } if (obj->type == UCL_OBJECT) { const ucl_object_t *num = ucl_object_find_key(obj, "number"); const ucl_object_t *name = ucl_object_find_key(obj, "name"); if (num == NULL || num->type != UCL_INT) { log_warnx("lun section in target \"%s\" is missing " "\"number\" integer property", target->t_name); return (1); } tmp = ucl_object_toint(num); if (tmp >= MAX_LUNS) { log_warnx("LU number %ju in target \"%s\" is too big", tmp, target->t_name); return (1); } if (name == NULL || name->type != UCL_STRING) { log_warnx("lun section in target \"%s\" is missing " "\"name\" string property", target->t_name); return (1); } lun = lun_find(conf, ucl_object_tostring(name)); if (lun == NULL) return (1); target->t_luns[tmp] = lun; } return (0); } static int uclparse_toplevel(const ucl_object_t *top) { ucl_object_iter_t it = NULL, iter = NULL; const ucl_object_t *obj = NULL, *child = NULL; int err = 0; /* Pass 1 - everything except targets */ while ((obj = ucl_iterate_object(top, &it, true))) { const char *key = ucl_object_key(obj); if (!strcmp(key, "debug")) { if (obj->type == UCL_INT) conf->conf_debug = ucl_object_toint(obj); else { log_warnx("\"debug\" property value is not integer"); return (1); } } if (!strcmp(key, "timeout")) { if (obj->type == UCL_INT) conf->conf_timeout = ucl_object_toint(obj); else { log_warnx("\"timeout\" property value is not integer"); return (1); } } if (!strcmp(key, "maxproc")) { if (obj->type == UCL_INT) conf->conf_maxproc = ucl_object_toint(obj); else { log_warnx("\"maxproc\" property value is not integer"); return (1); } } if (!strcmp(key, "pidfile")) { if (obj->type == UCL_STRING) conf->conf_pidfile_path = strdup( ucl_object_tostring(obj)); else { log_warnx("\"pidfile\" property value is not string"); return (1); } } if (!strcmp(key, "isns-server")) { if (obj->type == UCL_ARRAY) { iter = NULL; while ((child = ucl_iterate_object(obj, &iter, true))) { if (child->type != UCL_STRING) return (1); err = isns_new(conf, ucl_object_tostring(child)); if (err != 0) { return (1); } } } else { log_warnx("\"isns-server\" property value is " "not an array"); return (1); } } if (!strcmp(key, "isns-period")) { if (obj->type == UCL_INT) conf->conf_timeout = ucl_object_toint(obj); else { log_warnx("\"isns-period\" property value is not integer"); return (1); } } if (!strcmp(key, "isns-timeout")) { if (obj->type == UCL_INT) conf->conf_timeout = ucl_object_toint(obj); else { log_warnx("\"isns-timeout\" property value is not integer"); return (1); } } if (!strcmp(key, "auth-group")) { if (obj->type == UCL_OBJECT) { iter = NULL; while ((child = ucl_iterate_object(obj, &iter, true))) { uclparse_auth_group(ucl_object_key(child), child); } } else { log_warnx("\"auth-group\" section is not an object"); return (1); } } if (!strcmp(key, "portal-group")) { if (obj->type == UCL_OBJECT) { iter = NULL; while ((child = ucl_iterate_object(obj, &iter, true))) { uclparse_portal_group(ucl_object_key(child), child); } } else { log_warnx("\"portal-group\" section is not an object"); return (1); } } if (!strcmp(key, "lun")) { if (obj->type == UCL_OBJECT) { iter = NULL; while ((child = ucl_iterate_object(obj, &iter, true))) { uclparse_lun(ucl_object_key(child), child); } } else { log_warnx("\"lun\" section is not an object"); return (1); } } } /* Pass 2 - targets */ it = NULL; while ((obj = ucl_iterate_object(top, &it, true))) { const char *key = ucl_object_key(obj); if (!strcmp(key, "target")) { if (obj->type == UCL_OBJECT) { iter = NULL; while ((child = ucl_iterate_object(obj, &iter, true))) { uclparse_target(ucl_object_key(child), child); } } else { log_warnx("\"target\" section is not an object"); return (1); } } } return (0); } static int uclparse_auth_group(const char *name, const ucl_object_t *top) { struct auth_group *auth_group; const struct auth_name *an; const struct auth_portal *ap; ucl_object_iter_t it = NULL, it2 = NULL; const ucl_object_t *obj = NULL, *tmp = NULL; const char *key; int err; if (!strcmp(name, "default") && conf->conf_default_ag_defined == false) { auth_group = auth_group_find(conf, name); conf->conf_default_ag_defined = true; } else { auth_group = auth_group_new(conf, name); } if (auth_group == NULL) return (1); while ((obj = ucl_iterate_object(top, &it, true))) { key = ucl_object_key(obj); if (!strcmp(key, "auth-type")) { const char *value = ucl_object_tostring(obj); err = auth_group_set_type(auth_group, value); if (err) return (1); } if (!strcmp(key, "chap")) { if (obj->type != UCL_ARRAY) { log_warnx("\"chap\" property of " "auth-group \"%s\" is not an array", name); return (1); } it2 = NULL; while ((tmp = ucl_iterate_object(obj, &it2, true))) { if (uclparse_chap(auth_group, tmp) != 0) return (1); } } if (!strcmp(key, "chap-mutual")) { if (obj->type != UCL_ARRAY) { log_warnx("\"chap-mutual\" property of " "auth-group \"%s\" is not an array", name); return (1); } it2 = NULL; while ((tmp = ucl_iterate_object(obj, &it2, true))) { if (uclparse_chap_mutual(auth_group, tmp) != 0) return (1); } } if (!strcmp(key, "initiator-name")) { if (obj->type != UCL_ARRAY) { log_warnx("\"initiator-name\" property of " "auth-group \"%s\" is not an array", name); return (1); } it2 = NULL; while ((tmp = ucl_iterate_object(obj, &it2, true))) { const char *value = ucl_object_tostring(tmp); an = auth_name_new(auth_group, value); if (an == NULL) return (1); } } if (!strcmp(key, "initiator-portal")) { if (obj->type != UCL_ARRAY) { log_warnx("\"initiator-portal\" property of " "auth-group \"%s\" is not an array", name); return (1); } it2 = NULL; while ((tmp = ucl_iterate_object(obj, &it2, true))) { const char *value = ucl_object_tostring(tmp); ap = auth_portal_new(auth_group, value); if (ap == NULL) return (1); } } } return (0); } static int uclparse_portal_group(const char *name, const ucl_object_t *top) { struct portal_group *portal_group; ucl_object_iter_t it = NULL, it2 = NULL; const ucl_object_t *obj = NULL, *tmp = NULL; const char *key; if (strcmp(name, "default") == 0 && conf->conf_default_pg_defined == false) { portal_group = portal_group_find(conf, name); conf->conf_default_pg_defined = true; } else { portal_group = portal_group_new(conf, name); } if (portal_group == NULL) return (1); while ((obj = ucl_iterate_object(top, &it, true))) { key = ucl_object_key(obj); if (!strcmp(key, "discovery-auth-group")) { portal_group->pg_discovery_auth_group = auth_group_find(conf, ucl_object_tostring(obj)); if (portal_group->pg_discovery_auth_group == NULL) { log_warnx("unknown discovery-auth-group \"%s\" " "for portal-group \"%s\"", ucl_object_tostring(obj), portal_group->pg_name); return (1); } } if (!strcmp(key, "discovery-filter")) { if (obj->type != UCL_STRING) { log_warnx("\"discovery-filter\" property of " "portal-group \"%s\" is not a string", portal_group->pg_name); return (1); } if (portal_group_set_filter(portal_group, ucl_object_tostring(obj)) != 0) return (1); } if (!strcmp(key, "listen")) { if (obj->type == UCL_STRING) { if (portal_group_add_listen(portal_group, ucl_object_tostring(obj), false) != 0) return (1); } else if (obj->type == UCL_ARRAY) { while ((tmp = ucl_iterate_object(obj, &it2, true))) { if (portal_group_add_listen( portal_group, ucl_object_tostring(tmp), false) != 0) return (1); } } else { log_warnx("\"listen\" property of " "portal-group \"%s\" is not a string", portal_group->pg_name); return (1); } } if (!strcmp(key, "listen-iser")) { if (obj->type == UCL_STRING) { if (portal_group_add_listen(portal_group, ucl_object_tostring(obj), true) != 0) return (1); } else if (obj->type == UCL_ARRAY) { while ((tmp = ucl_iterate_object(obj, &it2, true))) { if (portal_group_add_listen( portal_group, ucl_object_tostring(tmp), true) != 0) return (1); } } else { log_warnx("\"listen\" property of " "portal-group \"%s\" is not a string", portal_group->pg_name); return (1); } } if (!strcmp(key, "redirect")) { if (obj->type != UCL_STRING) { log_warnx("\"listen\" property of " "portal-group \"%s\" is not a string", portal_group->pg_name); return (1); } if (portal_group_set_redirection(portal_group, ucl_object_tostring(obj)) != 0) return (1); } if (!strcmp(key, "options")) { if (obj->type != UCL_OBJECT) { log_warnx("\"options\" property of portal group " "\"%s\" is not an object", portal_group->pg_name); return (1); } while ((tmp = ucl_iterate_object(obj, &it2, true))) { option_new(&portal_group->pg_options, ucl_object_key(tmp), ucl_object_tostring_forced(tmp)); } } } return (0); } static int uclparse_target(const char *name, const ucl_object_t *top) { struct target *target; ucl_object_iter_t it = NULL, it2 = NULL; const ucl_object_t *obj = NULL, *tmp = NULL; const char *key; target = target_new(conf, name); if (target == NULL) return (1); while ((obj = ucl_iterate_object(top, &it, true))) { key = ucl_object_key(obj); if (!strcmp(key, "alias")) { if (obj->type != UCL_STRING) { log_warnx("\"alias\" property of target " "\"%s\" is not a string", target->t_name); return (1); } target->t_alias = strdup(ucl_object_tostring(obj)); } if (!strcmp(key, "auth-group")) { if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) log_warnx("auth-group for target \"%s\" " "specified more than once", target->t_name); else log_warnx("cannot use both auth-group " "and explicit authorisations for " "target \"%s\"", target->t_name); return (1); } target->t_auth_group = auth_group_find(conf, ucl_object_tostring(obj)); if (target->t_auth_group == NULL) { log_warnx("unknown auth-group \"%s\" for target " "\"%s\"", ucl_object_tostring(obj), target->t_name); return (1); } } if (!strcmp(key, "auth-type")) { int error; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "auth-type for target \"%s\"", target->t_name); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) return (1); target->t_auth_group->ag_target = target; } error = auth_group_set_type(target->t_auth_group, ucl_object_tostring(obj)); if (error != 0) return (1); } if (!strcmp(key, "chap")) { if (uclparse_chap(target->t_auth_group, obj) != 0) return (1); } if (!strcmp(key, "chap-mutual")) { if (uclparse_chap_mutual(target->t_auth_group, obj) != 0) return (1); } if (!strcmp(key, "initiator-name")) { const struct auth_name *an; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "initiator-name for target \"%s\"", target->t_name); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) return (1); target->t_auth_group->ag_target = target; } an = auth_name_new(target->t_auth_group, ucl_object_tostring(obj)); if (an == NULL) return (1); } if (!strcmp(key, "initiator-portal")) { const struct auth_portal *ap; if (target->t_auth_group != NULL) { if (target->t_auth_group->ag_name != NULL) { log_warnx("cannot use both auth-group and " "initiator-portal for target \"%s\"", target->t_name); return (1); } } else { target->t_auth_group = auth_group_new(conf, NULL); if (target->t_auth_group == NULL) return (1); target->t_auth_group->ag_target = target; } ap = auth_portal_new(target->t_auth_group, ucl_object_tostring(obj)); if (ap == NULL) return (1); } if (!strcmp(key, "portal-group")) { if (obj->type == UCL_OBJECT) { if (uclparse_target_portal_group(target, obj) != 0) return (1); } if (obj->type == UCL_ARRAY) { while ((tmp = ucl_iterate_object(obj, &it2, true))) { if (uclparse_target_portal_group(target, tmp) != 0) return (1); } } } if (!strcmp(key, "port")) { struct pport *pp; struct port *tp; const char *value = ucl_object_tostring(obj); + int ret, i_pp, i_vp = 0; + + ret = sscanf(value, "ioctl/%d/%d", &i_pp, &i_vp); + if (ret > 0) { + tp = port_new_ioctl(conf, target, i_pp, i_vp); + if (tp == NULL) { + log_warnx("can't create new ioctl port " + "for target \"%s\"", target->t_name); + return (1); + } + + return (0); + } pp = pport_find(conf, value); if (pp == NULL) { log_warnx("unknown port \"%s\" for target \"%s\"", value, target->t_name); return (1); } if (!TAILQ_EMPTY(&pp->pp_ports)) { log_warnx("can't link port \"%s\" to target \"%s\", " "port already linked to some target", value, target->t_name); return (1); } tp = port_new_pp(conf, target, pp); if (tp == NULL) { log_warnx("can't link port \"%s\" to target \"%s\"", value, target->t_name); return (1); } } if (!strcmp(key, "redirect")) { if (obj->type != UCL_STRING) { log_warnx("\"redirect\" property of target " "\"%s\" is not a string", target->t_name); return (1); } if (target_set_redirection(target, ucl_object_tostring(obj)) != 0) return (1); } if (!strcmp(key, "lun")) { while ((tmp = ucl_iterate_object(obj, &it2, true))) { if (uclparse_target_lun(target, tmp) != 0) return (1); } } } return (0); } static int uclparse_lun(const char *name, const ucl_object_t *top) { struct lun *lun; ucl_object_iter_t it = NULL, child_it = NULL; const ucl_object_t *obj = NULL, *child = NULL; const char *key; lun = lun_new(conf, name); if (lun == NULL) return (1); while ((obj = ucl_iterate_object(top, &it, true))) { key = ucl_object_key(obj); if (!strcmp(key, "backend")) { if (obj->type != UCL_STRING) { log_warnx("\"backend\" property of lun " "\"%s\" is not a string", lun->l_name); return (1); } lun_set_backend(lun, ucl_object_tostring(obj)); } if (!strcmp(key, "blocksize")) { if (obj->type != UCL_INT) { log_warnx("\"blocksize\" property of lun " "\"%s\" is not an integer", lun->l_name); return (1); } lun_set_blocksize(lun, ucl_object_toint(obj)); } if (!strcmp(key, "device-id")) { if (obj->type != UCL_STRING) { log_warnx("\"device-id\" property of lun " "\"%s\" is not an integer", lun->l_name); return (1); } lun_set_device_id(lun, ucl_object_tostring(obj)); } if (!strcmp(key, "options")) { if (obj->type != UCL_OBJECT) { log_warnx("\"options\" property of lun " "\"%s\" is not an object", lun->l_name); return (1); } while ((child = ucl_iterate_object(obj, &child_it, true))) { option_new(&lun->l_options, ucl_object_key(child), ucl_object_tostring_forced(child)); } } if (!strcmp(key, "path")) { if (obj->type != UCL_STRING) { log_warnx("\"path\" property of lun " "\"%s\" is not a string", lun->l_name); return (1); } lun_set_path(lun, ucl_object_tostring(obj)); } if (!strcmp(key, "serial")) { if (obj->type != UCL_STRING) { log_warnx("\"serial\" property of lun " "\"%s\" is not a string", lun->l_name); return (1); } lun_set_serial(lun, ucl_object_tostring(obj)); } if (!strcmp(key, "size")) { if (obj->type != UCL_INT) { log_warnx("\"size\" property of lun " "\"%s\" is not an integer", lun->l_name); return (1); } lun_set_size(lun, ucl_object_toint(obj)); } } return (0); } int uclparse_conf(struct conf *newconf, const char *path) { struct ucl_parser *parser; int error; conf = newconf; parser = ucl_parser_new(0); if (!ucl_parser_add_file(parser, path)) { log_warn("unable to parse configuration file %s: %s", path, ucl_parser_get_error(parser)); return (1); } error = uclparse_toplevel(ucl_parser_get_object(parser)); return (error); }