Index: head/sys/cam/ctl/ctl.c =================================================================== --- head/sys/cam/ctl/ctl.c (revision 275057) +++ head/sys/cam/ctl/ctl.c (revision 275058) @@ -1,14231 +1,14177 @@ /*- * Copyright (c) 2003-2009 Silicon Graphics International Corp. * Copyright (c) 2012 The FreeBSD Foundation * 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.c#8 $ */ /* * CAM Target Layer, a SCSI device emulation subsystem. * * Author: Ken Merry */ #define _CTL_C #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 struct ctl_softc *control_softc = NULL; /* * Size and alignment macros needed for Copan-specific HA hardware. These * can go away when the HA code is re-written, and uses busdma for any * hardware. */ #define CTL_ALIGN_8B(target, source, type) \ if (((uint32_t)source & 0x7) != 0) \ target = (type)(source + (0x8 - ((uint32_t)source & 0x7)));\ else \ target = (type)source; #define CTL_SIZE_8B(target, size) \ if ((size & 0x7) != 0) \ target = size + (0x8 - (size & 0x7)); \ else \ target = size; #define CTL_ALIGN_8B_MARGIN 16 /* * Template mode pages. */ /* * Note that these are default values only. The actual values will be * filled in when the user does a mode sense. */ static struct copan_debugconf_subpage debugconf_page_default = { DBGCNF_PAGE_CODE | SMPH_SPF, /* page_code */ DBGCNF_SUBPAGE_CODE, /* subpage */ {(sizeof(struct copan_debugconf_subpage) - 4) >> 8, (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */ DBGCNF_VERSION, /* page_version */ {CTL_TIME_IO_DEFAULT_SECS>>8, CTL_TIME_IO_DEFAULT_SECS>>0}, /* ctl_time_io_secs */ }; static struct copan_debugconf_subpage debugconf_page_changeable = { DBGCNF_PAGE_CODE | SMPH_SPF, /* page_code */ DBGCNF_SUBPAGE_CODE, /* subpage */ {(sizeof(struct copan_debugconf_subpage) - 4) >> 8, (sizeof(struct copan_debugconf_subpage) - 4) >> 0}, /* page_length */ 0, /* page_version */ {0xff,0xff}, /* ctl_time_io_secs */ }; 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}, }; 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*/0, /*read_retry_count*/0, /*correction_span*/0, /*head_offset_count*/0, /*data_strobe_offset_cnt*/0, /*byte8*/0, /*write_retry_count*/0, /*reserved2*/0, /*recovery_time_limit*/{0, 0}, }; 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} }; 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} }; 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} }; 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} }; 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} }; 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} }; 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} }; 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, /*eca_and_aen*/SCP_SWP, /*flags4*/0, /*aen_holdoff_period*/{0, 0}, /*busy_timeout_period*/{0, 0}, /*extended_selftest_completion_time*/{0, 0} }; 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_DEXCPT, /*mrie*/0, /*interval_timer*/{0, 0, 0, 0}, /*report_count*/{0, 0, 0, 0} }; 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*/0, /*mrie*/0, /*interval_timer*/{0, 0, 0, 0}, /*report_count*/{0, 0, 0, 0} }; #define CTL_LBPM_LEN (sizeof(struct ctl_logical_block_provisioning_page) - 4) 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}} } }; 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*/0, /*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}} } }; /* * XXX KDM move these into the softc. */ static int rcv_sync_msg; static int persis_offset; static uint8_t ctl_pause_rtr; 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"); /* * 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 int ctl_init(void); void 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_ioctl_online(void *arg); static void ctl_ioctl_offline(void *arg); static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id); static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id); static int ctl_ioctl_do_datamove(struct ctl_scsiio *ctsio); static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio); static int ctl_ioctl_submit_wait(union ctl_io *io); static void ctl_ioctl_datamove(union ctl_io *io); static void ctl_ioctl_done(union ctl_io *io); static void ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask); static void ctl_ioctl_bbrread_callback(void *arg,struct cfi_metatask *metatask); static int 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 uint32_t ctl_map_lun(int port_num, uint32_t lun); static uint32_t ctl_map_lun_back(int port_num, uint32_t lun); #ifdef unused static union ctl_io *ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target, uint32_t targ_lun, int can_wait); static void ctl_kfree_io(union ctl_io *io); #endif /* unused */ static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun, struct ctl_be_lun *be_lun, struct ctl_id target_id); static int ctl_free_lun(struct ctl_lun *lun); static void ctl_create_lun(struct ctl_be_lun *be_lun); /** static void ctl_failover_change_pages(struct ctl_softc *softc, struct ctl_scsiio *ctsio, int master); **/ 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_ha_msg *msg); 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); 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_softc *ctl_softc, struct ctl_lun *lun, const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio); //static int ctl_check_rtr(union ctl_io *pending_io, struct ctl_softc *softc); static void ctl_failover(void); 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_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io); static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io, ctl_ua_type ua_type); static int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type); static int ctl_abort_task(union ctl_io *io); static int ctl_abort_task_set(union ctl_io *io); static int ctl_i_t_nexus_reset(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 int 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); /* * 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"); MALLOC_DEFINE(M_CTLIO, "ctlio", "Memory used for CTL requests"); 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 ioctl_frontend = { .name = "ioctl", }; 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; ctsio->sense_residual = msg_info->scsi.sense_residual; ctsio->residual = msg_info->scsi.residual; memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data, sizeof(ctsio->sense_data)); memcpy(&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, &msg_info->scsi.lbalen, sizeof(msg_info->scsi.lbalen)); 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; #if 0 /* * Attempt to catch the situation where an I/O has * been freed, and we're using it again. */ if (ctsio->io_hdr.io_type == 0xff) { union ctl_io *tmp_io; tmp_io = (union ctl_io *)ctsio; printf("%s: %p use after free!\n", __func__, ctsio); printf("%s: type %d msg %d cdb %x iptl: " "%d:%d:%d:%d tag 0x%04x " "flag %#x status %x\n", __func__, tmp_io->io_hdr.io_type, tmp_io->io_hdr.msg_type, tmp_io->scsiio.cdb[0], tmp_io->io_hdr.nexus.initid.id, tmp_io->io_hdr.nexus.targ_port, tmp_io->io_hdr.nexus.targ_target.id, tmp_io->io_hdr.nexus.targ_lun, (tmp_io->io_hdr.io_type == CTL_IO_TASK) ? tmp_io->taskio.tag_num : tmp_io->scsiio.tag_num, tmp_io->io_hdr.flags, tmp_io->io_hdr.status); } #endif ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO; ctl_enqueue_isc((union ctl_io *)ctsio); } /* * 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 *ctl_softc; union ctl_io *io; struct ctl_prio *presio; ctl_ha_status isc_status; ctl_softc = control_softc; io = NULL; #if 0 printf("CTL: Isc Msg event %d\n", event); #endif if (event == CTL_HA_EVT_MSG_RECV) { union ctl_ha_msg msg_info; isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), /*wait*/ 0); #if 0 printf("CTL: msg_type %d\n", msg_info.msg_type); #endif if (isc_status != 0) { printf("Error receiving message, status = %d\n", isc_status); return; } switch (msg_info.hdr.msg_type) { case CTL_MSG_SERIALIZE: #if 0 printf("Serialize\n"); #endif io = ctl_alloc_io_nowait(ctl_softc->othersc_pool); if (io == NULL) { printf("ctl_isc_event_handler: can't allocate " "ctl_io!\n"); /* Bad Juju */ /* Need to set busy and send msg back */ msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU; msg_info.hdr.status = CTL_SCSI_ERROR; msg_info.scsi.scsi_status = SCSI_STATUS_BUSY; msg_info.scsi.sense_len = 0; if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 0) > CTL_HA_STATUS_SUCCESS){ } goto bailout; } ctl_zero_io(io); // populate ctsio from msg_info io->io_hdr.io_type = CTL_IO_SCSI; io->io_hdr.msg_type = CTL_MSG_SERIALIZE; io->io_hdr.original_sc = msg_info.hdr.original_sc; #if 0 printf("pOrig %x\n", (int)msg_info.original_sc); #endif 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 (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY) io->io_hdr.flags |= CTL_FLAG_INT_COPY; io->io_hdr.nexus = msg_info.hdr.nexus; #if 0 printf("targ %d, port %d, iid %d, lun %d\n", io->io_hdr.nexus.targ_target.id, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.initid.id, io->io_hdr.nexus.targ_lun); #endif io->scsiio.tag_num = msg_info.scsi.tag_num; io->scsiio.tag_type = msg_info.scsi.tag_type; memcpy(io->scsiio.cdb, msg_info.scsi.cdb, CTL_MAX_CDBLEN); if (ctl_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_info.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_info.hdr.serializing_sc; if (msg_info.dt.sg_sequence == 0) { /* * XXX KDM we use the preallocated S/G list * here, but we'll need to change this to * dynamic allocation if we need larger S/G * lists. */ if (msg_info.dt.kern_sg_entries > sizeof(io->io_hdr.remote_sglist) / sizeof(io->io_hdr.remote_sglist[0])) { printf("%s: number of S/G entries " "needed %u > allocated num %zd\n", __func__, msg_info.dt.kern_sg_entries, sizeof(io->io_hdr.remote_sglist)/ sizeof(io->io_hdr.remote_sglist[0])); /* * XXX KDM send a message back to * the other side to shut down the * DMA. The error will come back * through via the normal channel. */ break; } sgl = io->io_hdr.remote_sglist; memset(sgl, 0, sizeof(io->io_hdr.remote_sglist)); io->scsiio.kern_data_ptr = (uint8_t *)sgl; io->scsiio.kern_sg_entries = msg_info.dt.kern_sg_entries; io->scsiio.rem_sg_entries = msg_info.dt.kern_sg_entries; io->scsiio.kern_data_len = msg_info.dt.kern_data_len; io->scsiio.kern_total_len = msg_info.dt.kern_total_len; io->scsiio.kern_data_resid = msg_info.dt.kern_data_resid; io->scsiio.kern_rel_offset = msg_info.dt.kern_rel_offset; /* * Clear out per-DMA flags. */ io->io_hdr.flags &= ~CTL_FLAG_RDMA_MASK; /* * Add per-DMA flags that are set for this * particular DMA request. */ io->io_hdr.flags |= msg_info.dt.flags & CTL_FLAG_RDMA_MASK; } else sgl = (struct ctl_sg_entry *) io->scsiio.kern_data_ptr; for (i = msg_info.dt.sent_sg_entries, j = 0; i < (msg_info.dt.sent_sg_entries + msg_info.dt.cur_sg_entries); i++, j++) { sgl[i].addr = msg_info.dt.sg_list[j].addr; sgl[i].len = msg_info.dt.sg_list[j].len; #if 0 printf("%s: L: %p,%d -> %p,%d j=%d, i=%d\n", __func__, msg_info.dt.sg_list[j].addr, msg_info.dt.sg_list[j].len, sgl[i].addr, sgl[i].len, j, i); #endif } #if 0 memcpy(&sgl[msg_info.dt.sent_sg_entries], msg_info.dt.sg_list, sizeof(*sgl) * msg_info.dt.cur_sg_entries); #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_info.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_info.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_info.hdr.serializing_sc; io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE; io->io_hdr.status = msg_info.hdr.status; io->scsiio.scsi_status = msg_info.scsi.scsi_status; io->scsiio.sense_len = msg_info.scsi.sense_len; io->scsiio.sense_residual =msg_info.scsi.sense_residual; io->io_hdr.port_status = msg_info.scsi.fetd_status; io->scsiio.residual = msg_info.scsi.residual; memcpy(&io->scsiio.sense_data,&msg_info.scsi.sense_data, sizeof(io->scsiio.sense_data)); ctl_enqueue_isc(io); break; } /* Preformed on Originating SC, SER_ONLY mode */ case CTL_MSG_R2R: io = msg_info.hdr.original_sc; if (io == NULL) { printf("%s: Major Bummer\n", __func__); return; } else { #if 0 printf("pOrig %x\n",(int) ctsio); #endif } io->io_hdr.msg_type = CTL_MSG_R2R; io->io_hdr.serializing_sc = msg_info.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 (ctl_softc->ha_mode == CTL_HA_MODE_XFER) ctl_isc_handler_finish_xfer(ctl_softc, &msg_info); else ctl_isc_handler_finish_ser_only(ctl_softc, &msg_info); break; /* Preformed on Originating SC */ case CTL_MSG_BAD_JUJU: io = msg_info.hdr.original_sc; if (io == NULL) { printf("%s: Bad JUJU!, original_sc is NULL!\n", __func__); break; } ctl_copy_sense_data(&msg_info, 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_info.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_nowait( ctl_softc->othersc_pool); if (taskio == NULL) { printf("ctl_isc_event_handler: can't allocate " "ctl_io!\n"); /* Bad Juju */ /* should I just call the proper reset func here??? */ goto bailout; } 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_info.hdr.nexus; taskio->task_action = msg_info.task.task_action; taskio->tag_num = msg_info.task.tag_num; taskio->tag_type = msg_info.task.tag_type; #ifdef CTL_TIME_IO taskio->io_hdr.start_time = time_uptime; getbintime(&taskio->io_hdr.start_bt); #if 0 cs_prof_gettime(&taskio->io_hdr.start_ticks); #endif #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_nowait( ctl_softc->othersc_pool); if (presio == NULL) { printf("ctl_isc_event_handler: can't allocate " "ctl_io!\n"); /* Bad Juju */ /* Need to set busy and send msg back */ goto bailout; } ctl_zero_io((union ctl_io *)presio); presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION; presio->pr_msg = msg_info.pr; ctl_enqueue_isc((union ctl_io *)presio); break; case CTL_MSG_SYNC_FE: rcv_sync_msg = 1; break; default: printf("How did I get here?\n"); } } else if (event == CTL_HA_EVT_MSG_SENT) { if (param != CTL_HA_STATUS_SUCCESS) { printf("Bad status from ctl_ha_msg_send status %d\n", param); } return; } else if (event == CTL_HA_EVT_DISCONNECT) { printf("CTL: Got a disconnect from Isc\n"); return; } else { printf("ctl_isc_event_handler: Unknown event %d\n", event); return; } bailout: return; } static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest) { struct scsi_sense_data *sense; sense = &dest->scsiio.sense_data; bcopy(&src->scsi.sense_data, sense, sizeof(*sense)); dest->scsiio.scsi_status = src->scsi.scsi_status; dest->scsiio.sense_len = src->scsi.sense_len; dest->io_hdr.status = src->hdr.status; } static int ctl_ha_state_sysctl(SYSCTL_HANDLER_ARGS) { struct ctl_softc *softc = (struct ctl_softc *)arg1; struct ctl_lun *lun; int error, value, i; if (softc->flags & CTL_FLAG_ACTIVE_SHELF) value = 0; else value = 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_lock(&lun->lun_lock); for (i = 0; i < CTL_MAX_INITIATORS; i++) lun->pending_ua[i] |= CTL_UA_ASYM_ACC_CHANGE; mtx_unlock(&lun->lun_lock); } mtx_unlock(&softc->ctl_lock); return (0); } static int ctl_init(void) { struct ctl_softc *softc; void *other_pool; struct ctl_port *port; int i, error, retval; //int isc_retval; retval = 0; ctl_pause_rtr = 0; rcv_sync_msg = 0; control_softc = malloc(sizeof(*control_softc), M_DEVBUF, M_WAITOK | M_ZERO); softc = control_softc; softc->dev = make_dev(&ctl_cdevsw, 0, UID_ROOT, GID_OPERATOR, 0600, "cam/ctl"); softc->dev->si_drv1 = softc; /* * By default, return a "bad LUN" peripheral qualifier for unknown * LUNs. The user can override this default using the tunable or * sysctl. See the comment in ctl_inquiry_std() for more details. */ softc->inquiry_pq_no_lun = 1; TUNABLE_INT_FETCH("kern.cam.ctl.inquiry_pq_no_lun", &softc->inquiry_pq_no_lun); 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(control_softc, M_DEVBUF); control_softc = NULL; return (ENOMEM); } SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "inquiry_pq_no_lun", CTLFLAG_RW, &softc->inquiry_pq_no_lun, 0, "Report no lun possible for invalid LUNs"); 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->open_count = 0; /* * Default to actually sending a SYNCHRONIZE CACHE command down to * the drive. */ softc->flags = CTL_FLAG_REAL_SYNC; /* * 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->flags |= CTL_FLAG_ACTIVE_SHELF; softc->is_single = 1; softc->port_offset = 0; } else softc->port_offset = (softc->ha_id - 1) * CTL_MAX_PORTS; persis_offset = softc->port_offset * CTL_MAX_INIT_PER_PORT; /* * XXX KDM need to figure out where we want to get our target ID * and WWID. Is it different on each port? */ softc->target.id = 0; softc->target.wwid[0] = 0x12345678; softc->target.wwid[1] = 0x87654321; 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 (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC, &other_pool) != 0) { printf("ctl: can't allocate %d entry other SC pool, " "exiting\n", CTL_POOL_ENTRIES_OTHER_SC); return (ENOMEM); } softc->othersc_pool = other_pool; 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"); ctl_pool_free(other_pool); return (error); } } error = kproc_kthread_add(ctl_lun_thread, softc, &softc->ctl_proc, NULL, 0, 0, "ctl", "lun"); if (error != 0) { printf("error creating CTL lun thread!\n"); ctl_pool_free(other_pool); return (error); } error = kproc_kthread_add(ctl_thresh_thread, softc, &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh"); if (error != 0) { printf("error creating CTL threshold thread!\n"); ctl_pool_free(other_pool); return (error); } if (bootverbose) printf("ctl: CAM Target Layer loaded\n"); /* * Initialize the ioctl front end. */ ctl_frontend_register(&ioctl_frontend); port = &softc->ioctl_info.port; port->frontend = &ioctl_frontend; sprintf(softc->ioctl_info.port_name, "ioctl"); port->port_type = CTL_PORT_IOCTL; port->num_requested_ctl_io = 100; port->port_name = softc->ioctl_info.port_name; port->port_online = ctl_ioctl_online; port->port_offline = ctl_ioctl_offline; port->onoff_arg = &softc->ioctl_info; port->lun_enable = ctl_ioctl_lun_enable; port->lun_disable = ctl_ioctl_lun_disable; port->targ_lun_arg = &softc->ioctl_info; port->fe_datamove = ctl_ioctl_datamove; port->fe_done = ctl_ioctl_done; port->max_targets = 15; port->max_target_id = 15; if (ctl_port_register(&softc->ioctl_info.port) != 0) { printf("ctl: ioctl front end registration failed, will " "continue anyway\n"); } SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree), OID_AUTO, "ha_state", CTLTYPE_INT | CTLFLAG_RWTUN, softc, 0, ctl_ha_state_sysctl, "I", "HA state for this head"); #ifdef CTL_IO_DELAY if (sizeof(struct callout) > CTL_TIMER_BYTES) { printf("sizeof(struct callout) %zd > CTL_TIMER_BYTES %zd\n", sizeof(struct callout), CTL_TIMER_BYTES); return (EINVAL); } #endif /* CTL_IO_DELAY */ return (0); } void ctl_shutdown(void) { struct ctl_softc *softc; struct ctl_lun *lun, *next_lun; softc = (struct ctl_softc *)control_softc; if (ctl_port_deregister(&softc->ioctl_info.port) != 0) printf("ctl: ioctl front end deregistration failed\n"); mtx_lock(&softc->ctl_lock); /* * Free up each LUN. */ for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){ next_lun = STAILQ_NEXT(lun, links); ctl_free_lun(lun); } mtx_unlock(&softc->ctl_lock); ctl_frontend_deregister(&ioctl_frontend); #if 0 ctl_shutdown_thread(softc->work_thread); mtx_destroy(&softc->queue_lock); #endif ctl_tpc_shutdown(softc); uma_zdestroy(softc->io_zone); mtx_destroy(&softc->ctl_lock); destroy_dev(softc->dev); sysctl_ctx_free(&softc->sysctl_ctx); free(control_softc, M_DEVBUF); control_softc = NULL; if (bootverbose) printf("ctl: CAM Target Layer unloaded\n"); } 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 (EBUSY); 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); } int ctl_port_enable(ctl_port_type port_type) { struct ctl_softc *softc = control_softc; struct ctl_port *port; if (softc->is_single == 0) { union ctl_ha_msg msg_info; int isc_retval; #if 0 printf("%s: HA mode, synchronizing frontend enable\n", __func__); #endif msg_info.hdr.msg_type = CTL_MSG_SYNC_FE; if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 1 )) > CTL_HA_STATUS_SUCCESS) { printf("Sync msg send error retval %d\n", isc_retval); } if (!rcv_sync_msg) { isc_retval=ctl_ha_msg_recv(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 1); } #if 0 printf("CTL:Frontend Enable\n"); } else { printf("%s: single mode, skipping frontend synchronization\n", __func__); #endif } STAILQ_FOREACH(port, &softc->port_list, links) { if (port_type & port->port_type) { #if 0 printf("port %d\n", port->targ_port); #endif ctl_port_online(port); } } return (0); } int ctl_port_disable(ctl_port_type port_type) { struct ctl_softc *softc; struct ctl_port *port; softc = control_softc; STAILQ_FOREACH(port, &softc->port_list, links) { if (port_type & port->port_type) ctl_port_offline(port); } return (0); } /* * Returns 0 for success, 1 for failure. * Currently the only failure mode is if there aren't enough entries * allocated. So, in case of a failure, look at num_entries_dropped, * reallocate and try again. */ int ctl_port_list(struct ctl_port_entry *entries, int num_entries_alloced, int *num_entries_filled, int *num_entries_dropped, ctl_port_type port_type, int no_virtual) { struct ctl_softc *softc; struct ctl_port *port; int entries_dropped, entries_filled; int retval; int i; softc = control_softc; retval = 0; entries_filled = 0; entries_dropped = 0; i = 0; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { struct ctl_port_entry *entry; if ((port->port_type & port_type) == 0) continue; if ((no_virtual != 0) && (port->virtual_port != 0)) continue; if (entries_filled >= num_entries_alloced) { entries_dropped++; continue; } entry = &entries[i]; entry->port_type = port->port_type; strlcpy(entry->port_name, port->port_name, sizeof(entry->port_name)); entry->physical_port = port->physical_port; entry->virtual_port = port->virtual_port; entry->wwnn = port->wwnn; entry->wwpn = port->wwpn; i++; entries_filled++; } mtx_unlock(&softc->ctl_lock); if (entries_dropped > 0) retval = 1; *num_entries_dropped = entries_dropped; *num_entries_filled = entries_filled; return (retval); } static void ctl_ioctl_online(void *arg) { struct ctl_ioctl_info *ioctl_info; ioctl_info = (struct ctl_ioctl_info *)arg; ioctl_info->flags |= CTL_IOCTL_FLAG_ENABLED; } static void ctl_ioctl_offline(void *arg) { struct ctl_ioctl_info *ioctl_info; ioctl_info = (struct ctl_ioctl_info *)arg; ioctl_info->flags &= ~CTL_IOCTL_FLAG_ENABLED; } /* * 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 = control_softc; 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); port->wwpn_iid[iid].in_use--; port->wwpn_iid[iid].last_use = time_uptime; mtx_unlock(&softc->ctl_lock); 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 = control_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); /* * XXX KDM clear have_ca and ua_pending on each LUN for * this initiator. */ } 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); 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)); } } } static int ctl_ioctl_lun_enable(void *arg, struct ctl_id targ_id, int lun_id) { return (0); } static int ctl_ioctl_lun_disable(void *arg, struct ctl_id targ_id, int lun_id) { return (0); } /* * 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, len_copied; int kern_watermark, ext_watermark; int ext_sglist_malloced; int i, j; ext_sglist_malloced = 0; ext_sg_start = 0; ext_offset = 0; 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) { ctsio->ext_data_filled = ctsio->ext_data_len; 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->io_hdr.flags & CTL_FLAG_EDPTR_SGLIST) { 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) { ctl_set_internal_failure(ctsio, /*sks_valid*/ 0, /*retry_count*/ 0); goto bailout; } ext_sg_entries = ctsio->ext_sg_entries; 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->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; len_copied = 0; for (i = ext_sg_start, j = 0; i < ext_sg_entries && j < kern_sg_entries;) { uint8_t *ext_ptr, *kern_ptr; len_to_copy = ctl_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; kern_watermark += len_to_copy; ext_watermark += len_to_copy; 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) { ctl_set_internal_failure(ctsio, /*sks_valid*/ 0, /*retry_count*/ 0); 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){ ctl_set_internal_failure(ctsio, /*sks_valid*/ 0, /*retry_count*/0); goto bailout; } } len_copied += len_to_copy; if (ext_sglist[i].len == ext_watermark) { i++; ext_watermark = 0; } if (kern_sglist[j].len == kern_watermark) { j++; kern_watermark = 0; } } ctsio->ext_data_filled += len_copied; 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)); /* XXX KDM set residual?? */ bailout: if (ext_sglist_malloced != 0) free(ext_sglist, M_CTL); return (CTL_RETVAL_COMPLETE); } /* * 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 int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio) { struct ctl_softc *ctl_softc; union ctl_ha_msg msg_info; struct ctl_lun *lun; int retval = 0; uint32_t targ_lun; ctl_softc = control_softc; targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun; lun = ctl_softc->ctl_luns[targ_lun]; if (lun==NULL) { /* * Why isn't LUN defined? The other side wouldn't * send a cmd if the LUN is undefined. */ printf("%s: Bad JUJU!, LUN is NULL!\n", __func__); /* "Logical unit not supported" */ ctl_set_sense_data(&msg_info.scsi.sense_data, lun, /*sense_format*/SSD_TYPE_NONE, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x25, /*ascq*/ 0x00, SSD_ELEM_NONE); msg_info.scsi.sense_len = SSD_FULL_SIZE; msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND; msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; 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; if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) { } return(1); } mtx_lock(&lun->lun_lock); 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); break; case CTL_ACTION_PASS: case CTL_ACTION_SKIP: if (ctl_softc->ha_mode == CTL_HA_MODE_XFER) { ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; ctl_enqueue_rtr((union ctl_io *)ctsio); } else { /* 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; #if 0 printf("2. pOrig %x\n", (int)msg_info.hdr.original_sc); #endif if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) { } } break; case CTL_ACTION_OVERLAP: /* OVERLAPPED COMMANDS ATTEMPTED */ ctl_set_sense_data(&msg_info.scsi.sense_data, lun, /*sense_format*/SSD_TYPE_NONE, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x4E, /*ascq*/ 0x00, SSD_ELEM_NONE); msg_info.scsi.sense_len = SSD_FULL_SIZE; msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND; msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; 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; #if 0 printf("BAD JUJU:Major Bummer Overlap\n"); #endif TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); retval = 1; if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) { } break; case CTL_ACTION_OVERLAP_TAG: /* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */ ctl_set_sense_data(&msg_info.scsi.sense_data, lun, /*sense_format*/SSD_TYPE_NONE, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x4D, /*ascq*/ ctsio->tag_num & 0xff, SSD_ELEM_NONE); msg_info.scsi.sense_len = SSD_FULL_SIZE; msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND; msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; 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; #if 0 printf("BAD JUJU:Major Bummer Overlap Tag\n"); #endif TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); retval = 1; if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) { } break; case CTL_ACTION_ERROR: default: /* "Internal target failure" */ ctl_set_sense_data(&msg_info.scsi.sense_data, lun, /*sense_format*/SSD_TYPE_NONE, /*current_error*/ 1, /*sense_key*/ SSD_KEY_HARDWARE_ERROR, /*asc*/ 0x44, /*ascq*/ 0x00, SSD_ELEM_NONE); msg_info.scsi.sense_len = SSD_FULL_SIZE; msg_info.scsi.scsi_status = SCSI_STATUS_CHECK_COND; msg_info.hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; 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; #if 0 printf("BAD JUJU:Major Bummer HW Error\n"); #endif TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); retval = 1; if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 0 ) > CTL_HA_STATUS_SUCCESS) { } break; } mtx_unlock(&lun->lun_lock); return (retval); } static int ctl_ioctl_submit_wait(union ctl_io *io) { struct ctl_fe_ioctl_params params; ctl_fe_ioctl_state last_state; int done, retval; retval = 0; 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(("ctl_ioctl_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 occured, 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); } static void ctl_ioctl_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 ctl_ioctl_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 void ctl_ioctl_hard_startstop_callback(void *arg, struct cfi_metatask *metatask) { struct ctl_fe_ioctl_startstop_info *sd_info; sd_info = (struct ctl_fe_ioctl_startstop_info *)arg; sd_info->hs_info.status = metatask->status; sd_info->hs_info.total_luns = metatask->taskinfo.startstop.total_luns; sd_info->hs_info.luns_complete = metatask->taskinfo.startstop.luns_complete; sd_info->hs_info.luns_failed = metatask->taskinfo.startstop.luns_failed; cv_broadcast(&sd_info->sem); } static void ctl_ioctl_bbrread_callback(void *arg, struct cfi_metatask *metatask) { struct ctl_fe_ioctl_bbrread_info *fe_bbr_info; fe_bbr_info = (struct ctl_fe_ioctl_bbrread_info *)arg; mtx_lock(fe_bbr_info->lock); fe_bbr_info->bbr_info->status = metatask->status; fe_bbr_info->bbr_info->bbr_status = metatask->taskinfo.bbrread.status; fe_bbr_info->wakeup_done = 1; mtx_unlock(fe_bbr_info->lock); cv_broadcast(&fe_bbr_info->sem); } /* * Returns 0 for success, errno for failure. */ static int 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; int retval; retval = 0; 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); return (retval); } static void * ctl_copyin_alloc(void *user_addr, 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; 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) { 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); 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; int retval; softc = control_softc; retval = 0; switch (cmd) { case CTL_IO: { union ctl_io *io; void *pool_tmp; /* * If we haven't been "enabled", don't allow any SCSI I/O * to this FETD. */ if ((softc->ioctl_info.flags & CTL_IOCTL_FLAG_ENABLED) == 0) { retval = EPERM; break; } io = ctl_alloc_io(softc->ioctl_info.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; memcpy(io, (void *)addr, sizeof(*io)); io->io_hdr.pool = pool_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 = softc->ioctl_info.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 = softc->ioctl_info.cur_tag_num++; retval = ctl_ioctl_submit_wait(io); if (retval != 0) { ctl_free_io(io); break; } memcpy((void *)addr, io, sizeof(*io)); /* return this to our pool */ ctl_free_io(io); 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; 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) { /* * 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, */ mtx_unlock(&softc->ctl_lock); if (cmd == CTL_ENABLE_PORT) { struct ctl_lun *lun; STAILQ_FOREACH(lun, &softc->lun_list, links) { port->lun_enable(port->targ_lun_arg, lun->target, lun->lun); } ctl_port_online(port); } else if (cmd == CTL_DISABLE_PORT) { struct ctl_lun *lun; ctl_port_offline(port); STAILQ_FOREACH(lun, &softc->lun_list, links) { port->lun_disable( port->targ_lun_arg, lun->target, lun->lun); } } mtx_lock(&softc->ctl_lock); 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_PORT_LIST: { struct ctl_port *port; struct ctl_port_list *list; int i; list = (struct ctl_port_list *)addr; if (list->alloc_len != (list->alloc_num * sizeof(struct ctl_port_entry))) { printf("%s: CTL_GET_PORT_LIST: alloc_len %u != " "alloc_num %u * sizeof(struct ctl_port_entry) " "%zu\n", __func__, list->alloc_len, list->alloc_num, sizeof(struct ctl_port_entry)); retval = EINVAL; break; } list->fill_len = 0; list->fill_num = 0; list->dropped_num = 0; i = 0; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { struct ctl_port_entry entry, *list_entry; if (list->fill_num >= list->alloc_num) { list->dropped_num++; continue; } entry.port_type = port->port_type; strlcpy(entry.port_name, port->port_name, sizeof(entry.port_name)); entry.targ_port = port->targ_port; entry.physical_port = port->physical_port; entry.virtual_port = port->virtual_port; entry.wwnn = port->wwnn; entry.wwpn = port->wwpn; if (port->status & CTL_PORT_STATUS_ONLINE) entry.online = 1; else entry.online = 0; list_entry = &list->entries[i]; retval = copyout(&entry, list_entry, sizeof(entry)); if (retval != 0) { printf("%s: CTL_GET_PORT_LIST: copyout " "returned %d\n", __func__, retval); break; } i++; list->fill_num++; list->fill_len += sizeof(entry); } mtx_unlock(&softc->ctl_lock); /* * If this is non-zero, we had a copyout fault, so there's * probably no point in attempting to set the status inside * the structure. */ if (retval != 0) break; if (list->dropped_num > 0) list->status = CTL_PORT_LIST_NEED_MORE_SPACE; else list->status = CTL_PORT_LIST_OK; break; } case CTL_DUMP_OOA: { struct ctl_lun *lun; union ctl_io *io; char printbuf[128]; struct sbuf sb; mtx_lock(&softc->ctl_lock); printf("Dumping OOA queues:\n"); STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_lock(&lun->lun_lock); for (io = (union ctl_io *)TAILQ_FIRST( &lun->ooa_queue); io != NULL; io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr, ooa_links)) { sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN); sbuf_printf(&sb, "LUN %jd tag 0x%04x%s%s%s%s: ", (intmax_t)lun->lun, io->scsiio.tag_num, (io->io_hdr.flags & CTL_FLAG_BLOCKED) ? "" : " BLOCKED", (io->io_hdr.flags & CTL_FLAG_DMA_INPROG) ? " DMA" : "", (io->io_hdr.flags & CTL_FLAG_ABORT) ? " ABORT" : "", (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR) ? " RTR" : ""); ctl_scsi_command_string(&io->scsiio, NULL, &sb); sbuf_finish(&sb); printf("%s\n", sbuf_data(&sb)); } mtx_unlock(&lun->lun_lock); } printf("OOA queues dump done\n"); mtx_unlock(&softc->ctl_lock); break; } case CTL_GET_OOA: { struct ctl_lun *lun; 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) { retval = ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr, entries); if (retval != 0) break; } if (retval != 0) { mtx_unlock(&softc->ctl_lock); free(entries, M_CTL); break; } } else { lun = softc->ctl_luns[ooa_hdr->lun_num]; retval = 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); } getbintime(&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_CHECK_OOA: { union ctl_io *io; struct ctl_lun *lun; struct ctl_ooa_info *ooa_info; ooa_info = (struct ctl_ooa_info *)addr; if (ooa_info->lun_id >= CTL_MAX_LUNS) { ooa_info->status = CTL_OOA_INVALID_LUN; break; } mtx_lock(&softc->ctl_lock); lun = softc->ctl_luns[ooa_info->lun_id]; if (lun == NULL) { mtx_unlock(&softc->ctl_lock); ooa_info->status = CTL_OOA_INVALID_LUN; break; } mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); ooa_info->num_entries = 0; for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL; io = (union ctl_io *)TAILQ_NEXT( &io->io_hdr, ooa_links)) { ooa_info->num_entries++; } mtx_unlock(&lun->lun_lock); ooa_info->status = CTL_OOA_SUCCESS; break; } case CTL_HARD_START: case CTL_HARD_STOP: { struct ctl_fe_ioctl_startstop_info ss_info; struct cfi_metatask *metatask; struct mtx hs_mtx; mtx_init(&hs_mtx, "HS Mutex", NULL, MTX_DEF); cv_init(&ss_info.sem, "hard start/stop cv" ); metatask = cfi_alloc_metatask(/*can_wait*/ 1); if (metatask == NULL) { retval = ENOMEM; mtx_destroy(&hs_mtx); break; } if (cmd == CTL_HARD_START) metatask->tasktype = CFI_TASK_STARTUP; else metatask->tasktype = CFI_TASK_SHUTDOWN; metatask->callback = ctl_ioctl_hard_startstop_callback; metatask->callback_arg = &ss_info; cfi_action(metatask); /* Wait for the callback */ mtx_lock(&hs_mtx); cv_wait_sig(&ss_info.sem, &hs_mtx); mtx_unlock(&hs_mtx); /* * All information has been copied from the metatask by the * time cv_broadcast() is called, so we free the metatask here. */ cfi_free_metatask(metatask); memcpy((void *)addr, &ss_info.hs_info, sizeof(ss_info.hs_info)); mtx_destroy(&hs_mtx); break; } case CTL_BBRREAD: { struct ctl_bbrread_info *bbr_info; struct ctl_fe_ioctl_bbrread_info fe_bbr_info; struct mtx bbr_mtx; struct cfi_metatask *metatask; bbr_info = (struct ctl_bbrread_info *)addr; bzero(&fe_bbr_info, sizeof(fe_bbr_info)); bzero(&bbr_mtx, sizeof(bbr_mtx)); mtx_init(&bbr_mtx, "BBR Mutex", NULL, MTX_DEF); fe_bbr_info.bbr_info = bbr_info; fe_bbr_info.lock = &bbr_mtx; cv_init(&fe_bbr_info.sem, "BBR read cv"); metatask = cfi_alloc_metatask(/*can_wait*/ 1); if (metatask == NULL) { mtx_destroy(&bbr_mtx); cv_destroy(&fe_bbr_info.sem); retval = ENOMEM; break; } metatask->tasktype = CFI_TASK_BBRREAD; metatask->callback = ctl_ioctl_bbrread_callback; metatask->callback_arg = &fe_bbr_info; metatask->taskinfo.bbrread.lun_num = bbr_info->lun_num; metatask->taskinfo.bbrread.lba = bbr_info->lba; metatask->taskinfo.bbrread.len = bbr_info->len; cfi_action(metatask); mtx_lock(&bbr_mtx); while (fe_bbr_info.wakeup_done == 0) cv_wait_sig(&fe_bbr_info.sem, &bbr_mtx); mtx_unlock(&bbr_mtx); bbr_info->status = metatask->status; bbr_info->bbr_status = metatask->taskinfo.bbrread.status; bbr_info->scsi_status = metatask->taskinfo.bbrread.scsi_status; memcpy(&bbr_info->sense_data, &metatask->taskinfo.bbrread.sense_data, ctl_min(sizeof(bbr_info->sense_data), sizeof(metatask->taskinfo.bbrread.sense_data))); cfi_free_metatask(metatask); mtx_destroy(&bbr_mtx); cv_destroy(&fe_bbr_info.sem); break; } case CTL_DELAY_IO: { struct ctl_io_delay_info *delay_info; #ifdef CTL_IO_DELAY struct ctl_lun *lun; #endif /* CTL_IO_DELAY */ 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) || (softc->ctl_luns[delay_info->lun_id] == NULL)) { delay_info->status = CTL_DELAY_STATUS_INVALID_LUN; } else { lun = softc->ctl_luns[delay_info->lun_id]; mtx_lock(&lun->lun_lock); delay_info->status = CTL_DELAY_STATUS_OK; switch (delay_info->delay_type) { case CTL_DELAY_TYPE_CONT: break; 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); } mtx_unlock(&softc->ctl_lock); #else delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED; #endif /* CTL_IO_DELAY */ break; } case CTL_REALSYNC_SET: { int *syncstate; syncstate = (int *)addr; mtx_lock(&softc->ctl_lock); switch (*syncstate) { case 0: softc->flags &= ~CTL_FLAG_REAL_SYNC; break; case 1: softc->flags |= CTL_FLAG_REAL_SYNC; break; default: retval = EINVAL; break; } mtx_unlock(&softc->ctl_lock); break; } case CTL_REALSYNC_GET: { int *syncstate; syncstate = (int*)addr; mtx_lock(&softc->ctl_lock); if (softc->flags & CTL_FLAG_REAL_SYNC) *syncstate = 1; else *syncstate = 0; mtx_unlock(&softc->ctl_lock); break; } case CTL_SETSYNC: case CTL_GETSYNC: { struct ctl_sync_info *sync_info; struct ctl_lun *lun; sync_info = (struct ctl_sync_info *)addr; mtx_lock(&softc->ctl_lock); lun = softc->ctl_luns[sync_info->lun_id]; if (lun == NULL) { mtx_unlock(&softc->ctl_lock); sync_info->status = CTL_GS_SYNC_NO_LUN; } /* * Get or set the sync interval. We're not bounds checking * in the set case, hopefully the user won't do something * silly. */ mtx_lock(&lun->lun_lock); mtx_unlock(&softc->ctl_lock); if (cmd == CTL_GETSYNC) sync_info->sync_interval = lun->sync_interval; else lun->sync_interval = sync_info->sync_interval; mtx_unlock(&lun->lun_lock); sync_info->status = CTL_GS_SYNC_OK; break; } case CTL_GETSTATS: { struct ctl_stats *stats; struct ctl_lun *lun; int i; stats = (struct ctl_stats *)addr; if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) > stats->alloc_len) { stats->status = CTL_SS_NEED_MORE_SPACE; stats->num_luns = softc->num_luns; break; } /* * XXX KDM no locking here. If the LUN list changes, * things can blow up. */ for (i = 0, lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; i++, lun = STAILQ_NEXT(lun, links)) { retval = copyout(&lun->stats, &stats->lun_stats[i], sizeof(lun->stats)); if (retval != 0) break; } stats->num_luns = softc->num_luns; stats->fill_len = sizeof(struct ctl_lun_io_stats) * softc->num_luns; stats->status = CTL_SS_OK; #ifdef CTL_TIME_IO stats->flags = CTL_STATS_FLAG_TIME_VALID; #else stats->flags = CTL_STATS_FLAG_NONE; #endif getnanouptime(&stats->timestamp); break; } case CTL_ERROR_INJECT: { struct ctl_error_desc *err_desc, *new_err_desc; struct ctl_lun *lun; 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); lun = softc->ctl_luns[err_desc->lun_id]; if (lun == 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; struct ctl_lun *lun; int delete_done; delete_desc = (struct ctl_error_desc *)addr; delete_done = 0; mtx_lock(&softc->ctl_lock); lun = softc->ctl_luns[delete_desc->lun_id]; if (lun == 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 i, j, k, idx; struct ctl_port *port; struct ctl_frontend *fe; mtx_lock(&softc->ctl_lock); printf("CTL Persistent Reservation information start:\n"); for (i = 0; i < CTL_MAX_LUNS; i++) { struct ctl_lun *lun; lun = softc->ctl_luns[i]; if ((lun == NULL) || ((lun->flags & CTL_LUN_DISABLED) != 0)) continue; for (j = 0; j < (CTL_MAX_PORTS * 2); j++) { for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){ idx = j * CTL_MAX_INIT_PER_PORT + k; if (lun->pr_keys[idx] == 0) continue; printf(" LUN %d port %d iid %d key " "%#jx\n", i, j, k, (uintmax_t)lun->pr_keys[idx]); } } } 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; lun_req = (struct ctl_lun_req *)addr; 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) { lun_req->status = CTL_LUN_ERROR; break; } } retval = backend->ioctl(dev, cmd, addr, flag, td); 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); } break; } case CTL_LUN_LIST: { struct sbuf *sb; struct ctl_lun *lun; struct ctl_lun_list *list; struct ctl_option *opt; 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; } 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; req = (struct ctl_req *)addr; 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) { req->status = CTL_LUN_ERROR; break; } } retval = fe->ioctl(dev, cmd, addr, flag, td); if (req->num_args > 0) { ctl_copyout_args(req->num_args, req->kern_args); ctl_free_args(req->num_args, req->kern_args); } break; } case CTL_PORT_LIST: { struct sbuf *sb; struct ctl_port *port; struct ctl_lun_list *list; struct ctl_option *opt; int j; 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; } 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%u %s\n", j, port->wwpn_iid[j].name); else retval = sbuf_printf(sb, "\t%u naa.%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; } 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) { if (nexus->targ_port < CTL_MAX_PORTS) return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT)); else return (nexus->initid.id + ((nexus->targ_port - CTL_MAX_PORTS) * CTL_MAX_INIT_PER_PORT)); } uint32_t ctl_get_resindex(struct ctl_nexus *nexus) { return (nexus->initid.id + (nexus->targ_port * CTL_MAX_INIT_PER_PORT)); } uint32_t ctl_port_idx(int port_num) { if (port_num < CTL_MAX_PORTS) return(port_num); else return(port_num - CTL_MAX_PORTS); } static uint32_t ctl_map_lun(int port_num, uint32_t lun_id) { struct ctl_port *port; port = control_softc->ctl_ports[ctl_port_idx(port_num)]; if (port == NULL) return (UINT32_MAX); if (port->lun_map == NULL) return (lun_id); return (port->lun_map(port->targ_lun_arg, lun_id)); } static uint32_t ctl_map_lun_back(int port_num, uint32_t lun_id) { struct ctl_port *port; uint32_t i; port = control_softc->ctl_ports[ctl_port_idx(port_num)]; if (port->lun_map == NULL) return (lun_id); for (i = 0; i < CTL_MAX_LUNS; i++) { if (port->lun_map(port->targ_lun_arg, i) == lun_id) return (i); } return (UINT32_MAX); } /* * Note: This only works for bitmask sizes that are at least 32 bits, and * that are a power of 2. */ int ctl_ffz(uint32_t *mask, uint32_t size) { uint32_t num_chunks, num_pieces; int i, j; num_chunks = (size >> 5); if (num_chunks == 0) num_chunks++; num_pieces = ctl_min((sizeof(uint32_t) * 8), size); for (i = 0; i < num_chunks; i++) { for (j = 0; j < num_pieces; j++) { if ((mask[i] & (1 << j)) == 0) return ((i << 5) + j); } } 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); } #ifdef unused /* * The bus, target and lun are optional, they can be filled in later. * can_wait is used to determine whether we can wait on the malloc or not. */ union ctl_io* ctl_malloc_io(ctl_io_type io_type, uint32_t targ_port, uint32_t targ_target, uint32_t targ_lun, int can_wait) { union ctl_io *io; if (can_wait) io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_WAITOK); else io = (union ctl_io *)malloc(sizeof(*io), M_CTL, M_NOWAIT); if (io != NULL) { io->io_hdr.io_type = io_type; io->io_hdr.targ_port = targ_port; /* * XXX KDM this needs to change/go away. We need to move * to a preallocated pool of ctl_scsiio structures. */ io->io_hdr.nexus.targ_target.id = targ_target; io->io_hdr.nexus.targ_lun = targ_lun; } return (io); } void ctl_kfree_io(union ctl_io *io) { free(io, M_CTL); } #endif /* unused */ /* * 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) { #ifdef IO_POOLS 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; pool->zone = uma_zsecond_create(pool->name, NULL, NULL, NULL, NULL, ctl_softc->io_zone); /* uma_prealloc(pool->zone, total_ctl_io); */ *npool = pool; #else *npool = ctl_softc->io_zone; #endif return (0); } void ctl_pool_free(struct ctl_io_pool *pool) { if (pool == NULL) return; #ifdef IO_POOLS uma_zdestroy(pool->zone); free(pool, M_CTL); #endif } union ctl_io * ctl_alloc_io(void *pool_ref) { union ctl_io *io; #ifdef IO_POOLS struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref; io = uma_zalloc(pool->zone, M_WAITOK); #else io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK); #endif if (io != NULL) io->io_hdr.pool = pool_ref; return (io); } union ctl_io * ctl_alloc_io_nowait(void *pool_ref) { union ctl_io *io; #ifdef IO_POOLS struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref; io = uma_zalloc(pool->zone, M_NOWAIT); #else io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT); #endif if (io != NULL) io->io_hdr.pool = pool_ref; return (io); } void ctl_free_io(union ctl_io *io) { #ifdef IO_POOLS struct ctl_io_pool *pool; #endif if (io == NULL) return; #ifdef IO_POOLS pool = (struct ctl_io_pool *)io->io_hdr.pool; uma_zfree(pool->zone, io); #else uma_zfree((uma_zone_t)io->io_hdr.pool, io); #endif } void ctl_zero_io(union ctl_io *io) { void *pool_ref; if (io == NULL) return; /* * May need to preserve linked list pointers at some point too. */ pool_ref = io->io_hdr.pool; memset(io, 0, sizeof(*io)); io->io_hdr.pool = pool_ref; } /* * This routine is currently used for internal copies of ctl_ios that need * to persist for some reason after we've already returned status to the * FETD. (Thus the flag set.) * * XXX XXX * Note that this makes a blind copy of all fields in the ctl_io, except * for the pool reference. This includes any memory that has been * allocated! That memory will no longer be valid after done has been * called, so this would be VERY DANGEROUS for command that actually does * any reads or writes. Right now (11/7/2005), this is only used for immediate * start and stop commands, which don't transfer any data, so this is not a * problem. If it is used for anything else, the caller would also need to * allocate data buffer space and this routine would need to be modified to * copy the data buffer(s) as well. */ void ctl_copy_io(union ctl_io *src, union ctl_io *dest) { void *pool_ref; if ((src == NULL) || (dest == NULL)) return; /* * May need to preserve linked list pointers at some point too. */ pool_ref = dest->io_hdr.pool; memcpy(dest, src, ctl_min(sizeof(*src), sizeof(*dest))); dest->io_hdr.pool = pool_ref; /* * We need to know that this is an internal copy, and doesn't need * to get passed back to the FETD that allocated it. */ dest->io_hdr.flags |= CTL_FLAG_INT_COPY; } static 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; 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 this is a disk-only mode page, there's no point in * setting it up. For some pages, we have to have some * basic information about the disk in order to calculate the * mode page data. */ if ((lun->be_lun->lun_type != T_DIRECT) && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY)) continue; switch (page_index->page_code & SMPH_PC_MASK) { case SMS_RW_ERROR_RECOVERY_PAGE: { if (page_index->subpage != SMS_SUBPAGE_PAGE_0) panic("subpage is incorrect!"); 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; if (page_index->subpage != SMS_SUBPAGE_PAGE_0) panic("subpage is incorrect!"); /* * 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__ */ if (page_index->subpage != SMS_SUBPAGE_PAGE_0) panic("invalid subpage value %d", page_index->subpage); /* * 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) { 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_CACHING_PAGE: { struct scsi_caching_page *caching_page; if (page_index->subpage != SMS_SUBPAGE_PAGE_0) panic("invalid subpage value %d", page_index->subpage); 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"); if (value != NULL && strcmp(value, "off") == 0) caching_page->flags1 &= ~SCP_WCE; value = ctl_get_opt(&lun->be_lun->options, "readcache"); 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: { struct scsi_control_page *control_page; if (page_index->subpage != SMS_SUBPAGE_PAGE_0) panic("invalid subpage value %d", page_index->subpage); 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"); 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 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"); 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"); 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"); 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"); 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; } case SMS_VENDOR_SPECIFIC_PAGE:{ switch (page_index->subpage) { case DBGCNF_SUBPAGE_CODE: { struct copan_debugconf_subpage *current_page, *saved_page; memcpy(&lun->mode_pages.debugconf_subpage[ CTL_PAGE_CURRENT], &debugconf_page_default, sizeof(debugconf_page_default)); memcpy(&lun->mode_pages.debugconf_subpage[ CTL_PAGE_CHANGEABLE], &debugconf_page_changeable, sizeof(debugconf_page_changeable)); memcpy(&lun->mode_pages.debugconf_subpage[ CTL_PAGE_DEFAULT], &debugconf_page_default, sizeof(debugconf_page_default)); memcpy(&lun->mode_pages.debugconf_subpage[ CTL_PAGE_SAVED], &debugconf_page_default, sizeof(debugconf_page_default)); page_index->page_data = (uint8_t *)lun->mode_pages.debugconf_subpage; current_page = (struct copan_debugconf_subpage *) (page_index->page_data + (page_index->page_len * CTL_PAGE_CURRENT)); saved_page = (struct copan_debugconf_subpage *) (page_index->page_data + (page_index->page_len * CTL_PAGE_SAVED)); break; } default: panic("invalid subpage value %d", page_index->subpage); break; } break; } default: panic("invalid page value %d", page_index->page_code & SMPH_PC_MASK); break; } } 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 this is a disk-only mode page, there's no point in * setting it up. For some pages, we have to have some * basic information about the disk in order to calculate the * mode page data. */ if ((lun->be_lun->lun_type != T_DIRECT) && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY)) continue; if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING && ((lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) == 0 || 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; 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++) { 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_id target_id) { struct ctl_lun *nlun, *lun; struct ctl_port *port; struct scsi_vpd_id_descriptor *desc; struct scsi_vpd_id_t10 *t10id; const char *eui, *naa, *scsiname, *vendor, *value; int lun_number, i, 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: break; case T_PROCESSOR: 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; /* 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"); 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"); if (eui != NULL) { len += sizeof(struct scsi_vpd_id_descriptor) + 16; } naa = ctl_get_opt(&be_lun->options, "naa"); if (naa != NULL) { len += sizeof(struct scsi_vpd_id_descriptor) + 16; } 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) { 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; } 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); } 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, CTL_MAX_LUNS); if (lun_number == -1) { mtx_unlock(&ctl_softc->ctl_lock); printf("ctl: can't allocate LUN on target %ju, out of " "LUNs\n", (uintmax_t)target_id.id); 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); } } ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number); mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF); lun->target = target_id; 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_OFFLINE) lun->flags |= CTL_LUN_OFFLINE; if (be_lun->flags & CTL_LUN_FLAG_POWERED_OFF) lun->flags |= CTL_LUN_STOPPED; if (be_lun->flags & CTL_LUN_FLAG_INOPERABLE) lun->flags |= CTL_LUN_INOPERABLE; if (be_lun->flags & CTL_LUN_FLAG_PRIMARY) lun->flags |= CTL_LUN_PRIMARY_SC; value = ctl_get_opt(&be_lun->options, "readonly"); if (value != NULL && strcmp(value, "on") == 0) lun->flags |= CTL_LUN_READONLY; lun->ctl_softc = ctl_softc; TAILQ_INIT(&lun->ooa_queue); TAILQ_INIT(&lun->blocked_queue); STAILQ_INIT(&lun->error_list); ctl_tpc_lun_init(lun); /* * Initialize the mode and log page index. */ ctl_init_page_index(lun); ctl_init_log_page_index(lun); /* * Set the poweron UA for all initiators on this LUN only. */ for (i = 0; i < CTL_MAX_INITIATORS; i++) lun->pending_ua[i] = CTL_UA_POWERON; /* * Now, before we insert this lun on the lun list, set the lun * inventory changed UA for all other luns. */ STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) { for (i = 0; i < CTL_MAX_INITIATORS; i++) { nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE; } } STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links); ctl_softc->ctl_luns[lun_number] = lun; ctl_softc->num_luns++; /* Setup statistics gathering */ lun->stats.device_type = be_lun->lun_type; lun->stats.lun_number = lun_number; if (lun->stats.device_type == T_DIRECT) lun->stats.blocksize = be_lun->blocksize; else lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE; for (i = 0;i < CTL_MAX_PORTS;i++) lun->stats.ports[i].targ_port = i; mtx_unlock(&ctl_softc->ctl_lock); lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK); /* * Run through each registered FETD and bring it online if it isn't * already. Enable the target ID if it hasn't been enabled, and * enable this particular LUN. */ STAILQ_FOREACH(port, &ctl_softc->port_list, links) { int retval; retval = port->lun_enable(port->targ_lun_arg, target_id,lun_number); if (retval != 0) { printf("ctl_alloc_lun: FETD %s port %d returned error " "%d for lun_enable on target %ju lun %d\n", port->port_name, port->targ_port, retval, (uintmax_t)target_id.id, lun_number); } else port->status |= CTL_PORT_STATUS_LUN_ONLINE; } 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; #if 0 struct ctl_port *port; #endif struct ctl_lun *nlun; int i; softc = lun->ctl_softc; mtx_assert(&softc->ctl_lock, MA_OWNED); STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links); ctl_clear_mask(softc->ctl_lun_mask, lun->lun); softc->ctl_luns[lun->lun] = NULL; if (!TAILQ_EMPTY(&lun->ooa_queue)) panic("Freeing a LUN %p with outstanding I/O!!\n", lun); softc->num_luns--; /* * XXX KDM this scheme only works for a single target/multiple LUN * setup. It needs to be revamped for a multiple target scheme. * * XXX KDM this results in port->lun_disable() getting called twice, * once when ctl_disable_lun() is called, and a second time here. * We really need to re-think the LUN disable semantics. There * should probably be several steps/levels to LUN removal: * - disable * - invalidate * - free * * Right now we only have a disable method when communicating to * the front end ports, at least for individual LUNs. */ #if 0 STAILQ_FOREACH(port, &softc->port_list, links) { int retval; retval = port->lun_disable(port->targ_lun_arg, lun->target, lun->lun); if (retval != 0) { printf("ctl_free_lun: FETD %s port %d returned error " "%d for lun_disable on target %ju lun %jd\n", port->port_name, port->targ_port, retval, (uintmax_t)lun->target.id, (intmax_t)lun->lun); } if (STAILQ_FIRST(&softc->lun_list) == NULL) { port->status &= ~CTL_PORT_STATUS_LUN_ONLINE; retval = port->targ_disable(port->targ_lun_arg,lun->target); if (retval != 0) { printf("ctl_free_lun: FETD %s port %d " "returned error %d for targ_disable on " "target %ju\n", port->port_name, port->targ_port, retval, (uintmax_t)lun->target.id); } else port->status &= ~CTL_PORT_STATUS_TARG_ONLINE; if ((port->status & CTL_PORT_STATUS_TARG_ONLINE) != 0) continue; #if 0 port->port_offline(port->onoff_arg); port->status &= ~CTL_PORT_STATUS_ONLINE; #endif } } #endif /* * 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); ctl_tpc_lun_shutdown(lun); mtx_destroy(&lun->lun_lock); free(lun->lun_devid, M_CTL); free(lun->write_buffer, M_CTL); if (lun->flags & CTL_LUN_MALLOCED) free(lun, M_CTL); STAILQ_FOREACH(nlun, &softc->lun_list, links) { for (i = 0; i < CTL_MAX_INITIATORS; i++) { nlun->pending_ua[i] |= CTL_UA_LUN_CHANGE; } } return (0); } static void ctl_create_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; ctl_softc = control_softc; /* * ctl_alloc_lun() should handle all potential failure cases. */ ctl_alloc_lun(ctl_softc, NULL, be_lun, ctl_softc->target); } int ctl_add_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc = control_softc; mtx_lock(&ctl_softc->ctl_lock); STAILQ_INSERT_TAIL(&ctl_softc->pending_lun_queue, be_lun, links); mtx_unlock(&ctl_softc->ctl_lock); wakeup(&ctl_softc->pending_lun_queue); return (0); } int ctl_enable_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_port *port, *nport; struct ctl_lun *lun; int retval; ctl_softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&ctl_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(&ctl_softc->ctl_lock); return (0); } lun->flags &= ~CTL_LUN_DISABLED; mtx_unlock(&lun->lun_lock); for (port = STAILQ_FIRST(&ctl_softc->port_list); port != NULL; port = nport) { nport = STAILQ_NEXT(port, links); /* * 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(&ctl_softc->ctl_lock); retval = port->lun_enable(port->targ_lun_arg, lun->target,lun->lun); mtx_lock(&ctl_softc->ctl_lock); if (retval != 0) { printf("%s: FETD %s port %d returned error " "%d for lun_enable on target %ju lun %jd\n", __func__, port->port_name, port->targ_port, retval, (uintmax_t)lun->target.id, (intmax_t)lun->lun); } #if 0 else { /* NOTE: TODO: why does lun enable affect port status? */ port->status |= CTL_PORT_STATUS_LUN_ONLINE; } #endif } mtx_unlock(&ctl_softc->ctl_lock); return (0); } int ctl_disable_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_port *port; struct ctl_lun *lun; int retval; ctl_softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&ctl_softc->ctl_lock); mtx_lock(&lun->lun_lock); if (lun->flags & CTL_LUN_DISABLED) { mtx_unlock(&lun->lun_lock); mtx_unlock(&ctl_softc->ctl_lock); return (0); } lun->flags |= CTL_LUN_DISABLED; mtx_unlock(&lun->lun_lock); STAILQ_FOREACH(port, &ctl_softc->port_list, links) { mtx_unlock(&ctl_softc->ctl_lock); /* * 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. */ retval = port->lun_disable(port->targ_lun_arg, lun->target, lun->lun); mtx_lock(&ctl_softc->ctl_lock); if (retval != 0) { printf("ctl_alloc_lun: FETD %s port %d returned error " "%d for lun_disable on target %ju lun %jd\n", port->port_name, port->targ_port, retval, (uintmax_t)lun->target.id, (intmax_t)lun->lun); } } mtx_unlock(&ctl_softc->ctl_lock); return (0); } int ctl_start_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_lun *lun; ctl_softc = control_softc; 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_softc *ctl_softc; struct ctl_lun *lun; ctl_softc = control_softc; 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_offline(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_lun *lun; ctl_softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags |= CTL_LUN_OFFLINE; mtx_unlock(&lun->lun_lock); return (0); } int ctl_lun_online(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_lun *lun; ctl_softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags &= ~CTL_LUN_OFFLINE; mtx_unlock(&lun->lun_lock); return (0); } int ctl_invalidate_lun(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_lun *lun; ctl_softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; 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); mtx_lock(&ctl_softc->ctl_lock); ctl_free_lun(lun); mtx_unlock(&ctl_softc->ctl_lock); } else mtx_unlock(&lun->lun_lock); return (0); } int ctl_lun_inoperable(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_lun *lun; ctl_softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags |= CTL_LUN_INOPERABLE; mtx_unlock(&lun->lun_lock); return (0); } int ctl_lun_operable(struct ctl_be_lun *be_lun) { struct ctl_softc *ctl_softc; struct ctl_lun *lun; ctl_softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); lun->flags &= ~CTL_LUN_INOPERABLE; mtx_unlock(&lun->lun_lock); return (0); } void ctl_lun_capacity_changed(struct ctl_be_lun *be_lun) { struct ctl_lun *lun; struct ctl_softc *softc; int i; softc = control_softc; lun = (struct ctl_lun *)be_lun->ctl_lun; mtx_lock(&lun->lun_lock); for (i = 0; i < CTL_MAX_INITIATORS; i++) lun->pending_ua[i] |= CTL_UA_CAPACITY_CHANGED; mtx_unlock(&lun->lun_lock); } /* * 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; - retval = CTL_RETVAL_COMPLETE; - - CTL_DEBUG_PRINT(("ctl_config_move_done\n")); - /* - * XXX KDM this shouldn't happen, but what if it does? - */ - if (io->io_hdr.io_type != CTL_IO_SCSI) - panic("I/O type isn't CTL_IO_SCSI!"); + 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.flags & CTL_FLAG_ABORT) == 0) - && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) - io->io_hdr.status = CTL_SUCCESS; - else if ((io->io_hdr.port_status != 0) - && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) - && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){ + 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)) { /* * For hardware error sense keys, the sense key * specific value is defined to be a retry count, * but we use it to pass back an internal FETD * error code. XXX KDM Hopefully the FETD is only * using 16 bits for an error code, since that's * all the space we have in the sks field. */ ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ io->io_hdr.port_status); - if (io->io_hdr.flags & CTL_FLAG_ALLOCATED) - free(io->scsiio.kern_data_ptr, M_CTL); - ctl_done(io); - goto bailout; } - if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) - || ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) - || ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) { + 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); - /* Hopefully the user has already set the status... */ 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? */ if (ctl_debug & CTL_DEBUG_CDB_DATA) ctl_data_print(io); /* * 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); } -bailout: 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); } /* * SCSI release command. */ int ctl_scsi_release(struct ctl_scsiio *ctsio) { int length, longid, thirdparty_id, resv_id; struct ctl_softc *ctl_softc; struct ctl_lun *lun; uint32_t residx; length = 0; resv_id = 0; CTL_DEBUG_PRINT(("ctl_scsi_release\n")); residx = ctl_get_resindex(&ctsio->io_hdr.nexus); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; ctl_softc = control_softc; switch (ctsio->cdb[0]) { case RELEASE_10: { struct scsi_release_10 *cdb; cdb = (struct scsi_release_10 *)ctsio->cdb; if (cdb->byte2 & SR10_LONGID) longid = 1; else thirdparty_id = cdb->thirdparty_id; resv_id = cdb->resv_id; length = scsi_2btoul(cdb->length); break; } } /* * 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. */ length = 0; 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_data_resid = 0; 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); } if (length > 0) thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr); 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); - if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) { free(ctsio->kern_data_ptr, M_CTL); ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED; } + ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_scsi_reserve(struct ctl_scsiio *ctsio) { int extent, thirdparty, longid; int resv_id, length; uint64_t thirdparty_id; struct ctl_softc *ctl_softc; struct ctl_lun *lun; uint32_t residx; extent = 0; thirdparty = 0; longid = 0; resv_id = 0; length = 0; thirdparty_id = 0; CTL_DEBUG_PRINT(("ctl_reserve\n")); residx = ctl_get_resindex(&ctsio->io_hdr.nexus); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; ctl_softc = control_softc; switch (ctsio->cdb[0]) { case RESERVE_10: { struct scsi_reserve_10 *cdb; cdb = (struct scsi_reserve_10 *)ctsio->cdb; if (cdb->byte2 & SR10_LONGID) longid = 1; else thirdparty_id = cdb->thirdparty_id; resv_id = cdb->resv_id; length = scsi_2btoul(cdb->length); break; } } /* * 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. */ length = 0; 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_data_resid = 0; 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); } if (length > 0) thirdparty_id = scsi_8btou64(ctsio->kern_data_ptr); mtx_lock(&lun->lun_lock); if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) { ctl_set_reservation_conflict(ctsio); goto bailout; } lun->flags |= CTL_LUN_RESERVED; lun->res_idx = residx; ctl_set_success(ctsio); bailout: mtx_unlock(&lun->lun_lock); 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_start_stop(struct ctl_scsiio *ctsio) { struct scsi_start_stop_unit *cdb; struct ctl_lun *lun; struct ctl_softc *ctl_softc; int retval; CTL_DEBUG_PRINT(("ctl_start_stop\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; ctl_softc = control_softc; retval = 0; cdb = (struct scsi_start_stop_unit *)ctsio->cdb; /* * XXX KDM * We don't support the immediate bit on a stop unit. In order to * do that, we would need to code up a way to know that a stop is * pending, and hold off any new commands until it completes, one * way or another. Then we could accept or reject those commands * depending on its status. We would almost need to do the reverse * of what we do below for an immediate start -- return the copy of * the ctl_io to the FETD with status to send to the host (and to * free the copy!) and then free the original I/O once the stop * actually completes. That way, the OOA queue mechanism can work * to block commands that shouldn't proceed. Another alternative * would be to put the copy in the queue in place of the original, * and return the original back to the caller. That could be * slightly safer.. */ if ((cdb->byte2 & SSS_IMMED) && ((cdb->how & SSS_START) == 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); } if ((lun->flags & CTL_LUN_PR_RESERVED) && ((cdb->how & SSS_START)==0)) { uint32_t residx; residx = ctl_get_resindex(&ctsio->io_hdr.nexus); if (lun->pr_keys[residx] == 0 || (lun->pr_res_idx!=residx && lun->res_type < 4)) { ctl_set_reservation_conflict(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } } /* * If there is no backend on this device, we can't start or stop * it. In theory we shouldn't get any start/stop commands in the * first place at this level if the LUN doesn't have a backend. * That should get stopped by the command decode code. */ if (lun->backend == NULL) { ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } /* * XXX KDM Copan-specific offline behavior. * Figure out a reasonable way to port this? */ #ifdef NEEDTOPORT mtx_lock(&lun->lun_lock); if (((cdb->byte2 & SSS_ONOFFLINE) == 0) && (lun->flags & CTL_LUN_OFFLINE)) { /* * If the LUN is offline, and the on/offline bit isn't set, * reject the start or stop. Otherwise, let it through. */ mtx_unlock(&lun->lun_lock); ctl_set_lun_not_ready(ctsio); ctl_done((union ctl_io *)ctsio); } else { mtx_unlock(&lun->lun_lock); #endif /* NEEDTOPORT */ /* * This could be a start or a stop when we're online, * or a stop/offline or start/online. A start or stop when * we're offline is covered in the case above. */ /* * In the non-immediate case, we send the request to * the backend and return status to the user when * it is done. * * In the immediate case, we allocate a new ctl_io * to hold a copy of the request, and send that to * the backend. We then set good status on the * user's request and return it immediately. */ if (cdb->byte2 & SSS_IMMED) { union ctl_io *new_io; new_io = ctl_alloc_io(ctsio->io_hdr.pool); ctl_copy_io((union ctl_io *)ctsio, new_io); retval = lun->backend->config_write(new_io); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); } else { retval = lun->backend->config_write( (union ctl_io *)ctsio); } #ifdef NEEDTOPORT } #endif 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; struct ctl_softc *ctl_softc; uint64_t starting_lba; uint32_t block_count; int retval; CTL_DEBUG_PRINT(("ctl_sync_cache\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; ctl_softc = control_softc; 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); 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); 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); ctl_done((union ctl_io *)ctsio); goto bailout; } /* * If this LUN has no backend, we can't flush the cache anyway. */ if (lun->backend == NULL) { ctl_set_invalid_opcode(ctsio); ctl_done((union ctl_io *)ctsio); goto bailout; } /* * Check to see whether we're configured to send the SYNCHRONIZE * CACHE command directly to the back end. */ mtx_lock(&lun->lun_lock); if ((ctl_softc->flags & CTL_FLAG_REAL_SYNC) && (++(lun->sync_count) >= lun->sync_interval)) { lun->sync_count = 0; mtx_unlock(&lun->lun_lock); retval = lun->backend->config_write((union ctl_io *)ctsio); } else { mtx_unlock(&lun->lun_lock); ctl_set_success(ctsio); ctl_done((union ctl_io *)ctsio); } bailout: return (retval); } int ctl_format(struct ctl_scsiio *ctsio) { struct scsi_format *cdb; struct ctl_lun *lun; struct ctl_softc *ctl_softc; int length, defect_list_len; CTL_DEBUG_PRINT(("ctl_format\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; ctl_softc = control_softc; 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_data_resid = 0; 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; } } } /* * The format command will clear out the "Medium format corrupted" * status if set by the configuration code. That status is really * just a way to notify the host that we have lost the media, and * get them to issue a command that will basically make them think * they're blowing away the media. */ mtx_lock(&lun->lun_lock); lun->flags &= ~CTL_LUN_INOPERABLE; mtx_unlock(&lun->lun_lock); 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 scsi_read_buffer *cdb; struct ctl_lun *lun; int buffer_offset, len; static uint8_t descr[4]; static uint8_t echo_descr[4] = { 0 }; CTL_DEBUG_PRINT(("ctl_read_buffer\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; cdb = (struct scsi_read_buffer *)ctsio->cdb; if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA && (cdb->byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR && (cdb->byte2 & RWB_MODE) != RWB_MODE_DESCR) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 4); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } 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 ((cdb->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 ((cdb->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_data_resid = 0; 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 scsi_write_buffer *cdb; struct ctl_lun *lun; int buffer_offset, len; CTL_DEBUG_PRINT(("ctl_write_buffer\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; cdb = (struct scsi_write_buffer *)ctsio->cdb; if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 4); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } 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_data_resid = 0; 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; struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int len, retval; uint8_t byte2; retval = CTL_RETVAL_COMPLETE; CTL_DEBUG_PRINT(("ctl_write_same\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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 */ } /* NDOB and ANCHOR flags can be used only together with UNMAP */ if ((byte2 & SWS_UNMAP) == 0 && (byte2 & (SWS_NDOB | 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); 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_data_resid = 0; 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; 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; retval = CTL_RETVAL_COMPLETE; CTL_DEBUG_PRINT(("ctl_unmap\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_data_resid = 0; 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); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } 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); 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); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } 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); } /* * Note that this function currently doesn't actually do anything inside * CTL to enforce things if the DQue bit is turned on. * * Also note that this function can't be used in the default case, because * the DQue bit isn't set in the changeable mask for the control mode page * anyway. This is just here as an example for how to implement a page * handler, and a placeholder in case we want to allow the user to turn * tagged queueing on and off. * * The D_SENSE bit handling is functional, however, and will turn * descriptor sense on and off for a given LUN. */ int ctl_control_page_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr) { struct scsi_control_page *current_cp, *saved_cp, *user_cp; struct ctl_lun *lun; struct ctl_softc *softc; int set_ua; uint32_t initidx; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); set_ua = 0; user_cp = (struct scsi_control_page *)page_ptr; current_cp = (struct scsi_control_page *) (page_index->page_data + (page_index->page_len * CTL_PAGE_CURRENT)); saved_cp = (struct scsi_control_page *) (page_index->page_data + (page_index->page_len * CTL_PAGE_SAVED)); softc = control_softc; mtx_lock(&lun->lun_lock); if (((current_cp->rlec & SCP_DSENSE) == 0) && ((user_cp->rlec & SCP_DSENSE) != 0)) { /* * Descriptor sense is currently turned off and the user * wants to turn it on. */ current_cp->rlec |= SCP_DSENSE; saved_cp->rlec |= SCP_DSENSE; lun->flags |= CTL_LUN_SENSE_DESC; set_ua = 1; } else if (((current_cp->rlec & SCP_DSENSE) != 0) && ((user_cp->rlec & SCP_DSENSE) == 0)) { /* * Descriptor sense is currently turned on, and the user * wants to turn it off. */ current_cp->rlec &= ~SCP_DSENSE; saved_cp->rlec &= ~SCP_DSENSE; lun->flags &= ~CTL_LUN_SENSE_DESC; set_ua = 1; } if ((current_cp->queue_flags & SCP_QUEUE_ALG_MASK) != (user_cp->queue_flags & SCP_QUEUE_ALG_MASK)) { current_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK; current_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK; saved_cp->queue_flags &= ~SCP_QUEUE_ALG_MASK; saved_cp->queue_flags |= user_cp->queue_flags & SCP_QUEUE_ALG_MASK; set_ua = 1; } if ((current_cp->eca_and_aen & SCP_SWP) != (user_cp->eca_and_aen & SCP_SWP)) { current_cp->eca_and_aen &= ~SCP_SWP; current_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP; saved_cp->eca_and_aen &= ~SCP_SWP; saved_cp->eca_and_aen |= user_cp->eca_and_aen & SCP_SWP; set_ua = 1; } if (set_ua != 0) { int i; /* * Let other initiators know that the mode * parameters for this LUN have changed. */ for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (i == initidx) continue; lun->pending_ua[i] |= CTL_UA_MODE_CHANGE; } } mtx_unlock(&lun->lun_lock); return (0); } int ctl_caching_sp_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr) { struct scsi_caching_page *current_cp, *saved_cp, *user_cp; struct ctl_lun *lun; int set_ua; uint32_t initidx; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; initidx = ctl_get_initindex(&ctsio->io_hdr.nexus); set_ua = 0; user_cp = (struct scsi_caching_page *)page_ptr; current_cp = (struct scsi_caching_page *) (page_index->page_data + (page_index->page_len * CTL_PAGE_CURRENT)); saved_cp = (struct scsi_caching_page *) (page_index->page_data + (page_index->page_len * CTL_PAGE_SAVED)); mtx_lock(&lun->lun_lock); if ((current_cp->flags1 & (SCP_WCE | SCP_RCD)) != (user_cp->flags1 & (SCP_WCE | SCP_RCD))) { current_cp->flags1 &= ~(SCP_WCE | SCP_RCD); current_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD); saved_cp->flags1 &= ~(SCP_WCE | SCP_RCD); saved_cp->flags1 |= user_cp->flags1 & (SCP_WCE | SCP_RCD); set_ua = 1; } if (set_ua != 0) { int i; /* * Let other initiators know that the mode * parameters for this LUN have changed. */ for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (i == initidx) continue; lun->pending_ua[i] |= CTL_UA_MODE_CHANGE; } } mtx_unlock(&lun->lun_lock); return (0); } int ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr) { uint8_t *c; int i; c = ((struct copan_debugconf_subpage *)page_ptr)->ctl_time_io_secs; ctl_time_io_secs = (c[0] << 8) | (c[1] << 0) | 0; CTL_DEBUG_PRINT(("set ctl_time_io_secs to %d\n", ctl_time_io_secs)); printf("set ctl_time_io_secs to %d\n", ctl_time_io_secs); printf("page data:"); for (i=0; i<8; i++) printf(" %.2x",page_ptr[i]); printf("\n"); return (0); } int ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc) { struct copan_debugconf_subpage *page; page = (struct copan_debugconf_subpage *)page_index->page_data + (page_index->page_len * pc); switch (pc) { case SMS_PAGE_CTRL_CHANGEABLE >> 6: case SMS_PAGE_CTRL_DEFAULT >> 6: case SMS_PAGE_CTRL_SAVED >> 6: /* * We don't update the changable or default bits for this page. */ break; case SMS_PAGE_CTRL_CURRENT >> 6: page->ctl_time_io_secs[0] = ctl_time_io_secs >> 8; page->ctl_time_io_secs[1] = ctl_time_io_secs >> 0; break; default: #ifdef NEEDTOPORT EPRINT(0, "Invalid PC %d!!", pc); #endif /* NEEDTOPORT */ break; } return (0); } static int ctl_do_mode_select(union ctl_io *io) { struct scsi_mode_page_header *page_header; struct ctl_page_index *page_index; struct ctl_scsiio *ctsio; int control_dev, page_len; int page_len_offset, page_len_size; union ctl_modepage_info *modepage_info; struct ctl_lun *lun; int *len_left, *len_used; int retval, i; ctsio = &io->scsiio; page_index = NULL; page_len = 0; retval = CTL_RETVAL_COMPLETE; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; if (lun->be_lun->lun_type != T_DIRECT) control_dev = 1; else control_dev = 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++) { if ((control_dev != 0) && (lun->mode_pages.index[i].page_flags & CTL_PAGE_FLAG_DISK_ONLY)) continue; if ((lun->mode_pages.index[i].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 (((lun->mode_pages.index[i].page_code & SMPH_SPF) == 0) && ((page_header->page_code & SMPH_SPF) == 0)) { page_index = &lun->mode_pages.index[i]; page_len = page_header->page_length; break; } /* * If both pages have subpages, then the subpage numbers * have to match. */ if ((lun->mode_pages.index[i].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 (lun->mode_pages.index[i].subpage == sph->subpage) { page_index = &lun->mode_pages.index[i]; 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 ((page_index == NULL) || (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)) || (*len_left < page_index->page_len)) { 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); } /* * 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) { int param_len, pf, sp; int header_size, bd_len; int len_left, len_used; struct ctl_page_index *page_index; struct ctl_lun *lun; int control_dev, page_len; union ctl_modepage_info *modepage_info; int retval; pf = 0; sp = 0; page_len = 0; len_used = 0; len_left = 0; retval = 0; bd_len = 0; page_index = NULL; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; if (lun->be_lun->lun_type != T_DIRECT) control_dev = 1; else control_dev = 0; 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; 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; 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); break; /* NOTREACHED */ } /* * 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_data_resid = 0; 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("Invalid CDB type %#x", ctsio->cdb[0]); break; } 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)); len_left = param_len - header_size - bd_len; len_used = header_size + bd_len; modepage_info->header.len_left = len_left; modepage_info->header.len_used = len_used; return (ctl_do_mode_select((union ctl_io *)ctsio)); } int ctl_mode_sense(struct ctl_scsiio *ctsio) { struct ctl_lun *lun; 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; int control_dev; dbd = 0; llba = 0; block_desc = NULL; page_index = NULL; CTL_DEBUG_PRINT(("ctl_mode_sense\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; if (lun->be_lun->lun_type != T_DIRECT) control_dev = 1; else control_dev = 0; 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: { 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++) { if ((control_dev != 0) && (lun->mode_pages.index[i].page_flags & CTL_PAGE_FLAG_DISK_ONLY)) continue; /* * We don't use this subpage if the user didn't * request all subpages. */ if ((lun->mode_pages.index[i].subpage != 0) && (subpage == SMS_SUBPAGE_PAGE_0)) continue; #if 0 printf("found page %#x len %d\n", lun->mode_pages.index[i].page_code & SMPH_PC_MASK, lun->mode_pages.index[i].page_len); #endif page_len += lun->mode_pages.index[i].page_len; } break; } default: { int i; page_len = 0; for (i = 0; i < CTL_NUM_MODE_PAGES; i++) { /* Look for the right page code */ if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) != page_code) continue; /* Look for the right subpage or the subpage wildcard*/ if ((lun->mode_pages.index[i].subpage != subpage) && (subpage != SMS_SUBPAGE_ALL)) continue; /* Make sure the page is supported for this dev type */ if ((control_dev != 0) && (lun->mode_pages.index[i].page_flags & CTL_PAGE_FLAG_DISK_ONLY)) continue; #if 0 printf("found page %#x len %d\n", lun->mode_pages.index[i].page_code & SMPH_PC_MASK, lun->mode_pages.index[i].page_len); #endif page_len += lun->mode_pages.index[i].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_data_resid = 0; ctsio->kern_rel_offset = 0; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_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 = ctl_min(total_len - 1, 254); if (control_dev == 0) { header->dev_specific = 0x10; /* DPOFUA */ if ((lun->flags & CTL_LUN_READONLY) || (lun->mode_pages.control_page[CTL_PAGE_CURRENT] .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 = ctl_min(total_len - 2, 65533); scsi_ulto2b(datalen, header->datalen); if (control_dev == 0) { header->dev_specific = 0x10; /* DPOFUA */ if ((lun->flags & CTL_LUN_READONLY) || (lun->mode_pages.control_page[CTL_PAGE_CURRENT] .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("invalid CDB type %#x", ctsio->cdb[0]); break; /* NOTREACHED */ } /* * If we've got a disk, use its blocksize in the block * descriptor. Otherwise, just set it to 0. */ if (dbd == 0) { if (control_dev == 0) 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 ((control_dev != 0) && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY)) 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 ((control_dev != 0) && (page_index->page_flags & CTL_PAGE_FLAG_DISK_ONLY)) 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; } } - ctsio->scsi_status = SCSI_STATUS_OK; - + 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; struct scsi_log_param_header *phdr; uint8_t *data; uint64_t val; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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] = 0x01; /* per-LUN */ 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] = 0x02; /* per-pool */ 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_log_sense(struct ctl_scsiio *ctsio) { struct ctl_lun *lun; 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")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_data_resid = 0; ctsio->kern_rel_offset = 0; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } header = (struct scsi_log_header *)ctsio->kern_data_ptr; header->page = page_index->page_code; 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); - ctsio->scsi_status = SCSI_STATUS_OK; + 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 scsi_read_capacity *cdb; struct scsi_read_capacity_data *data; struct ctl_lun *lun; 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); } lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO); data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr; ctsio->residual = 0; ctsio->kern_data_len = sizeof(*data); ctsio->kern_total_len = sizeof(*data); ctsio->kern_data_resid = 0; 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); - ctsio->scsi_status = SCSI_STATUS_OK; - + 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 scsi_read_capacity_16 *cdb; struct scsi_read_capacity_data_long *data; struct ctl_lun *lun; 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); } lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO); data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr; if (sizeof(*data) < alloc_len) { ctsio->residual = alloc_len - sizeof(*data); ctsio->kern_data_len = sizeof(*data); ctsio->kern_total_len = sizeof(*data); } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; 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; - ctsio->scsi_status = SCSI_STATUS_OK; - + 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_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); if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; 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); } - ctsio->scsi_status = SCSI_STATUS_OK; + 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 scsi_maintenance_in *cdb; int retval; int alloc_len, ext, total_len = 0, g, p, pc, pg, gs, os; int num_target_port_groups, num_target_ports; struct ctl_lun *lun; struct ctl_softc *softc; 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; softc = control_softc; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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); } if (softc->is_single) num_target_port_groups = 1; else num_target_port_groups = NUM_TARGET_PORT_GROUPS; num_target_ports = 0; mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(port, &softc->port_list, links) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0) continue; if (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS) continue; num_target_ports++; } mtx_unlock(&softc->ctl_lock); 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) * num_target_port_groups + sizeof(struct scsi_target_port_descriptor) * num_target_ports * num_target_port_groups; alloc_len = scsi_4btoul(cdb->length); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); ctsio->kern_sg_entries = 0; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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_offset / CTL_MAX_PORTS; if (softc->flags & CTL_FLAG_ACTIVE_SHELF) { if (softc->ha_mode == CTL_HA_MODE_ACT_STBY) { gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; os = TPG_ASYMMETRIC_ACCESS_STANDBY; } else if (lun->flags & CTL_LUN_PRIMARY_SC) { gs = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED; } else { gs = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED; os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; } } else { gs = TPG_ASYMMETRIC_ACCESS_STANDBY; os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED; } for (g = 0; g < num_target_port_groups; g++) { tpg_desc->pref_state = (g == pg) ? gs : os; tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP; scsi_ulto2b(g + 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 (ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS) continue; p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS; scsi_ulto2b(p, 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_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n", - ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1], - ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3], - ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5], - ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7])); - ctl_datamove((union ctl_io *)ctsio); return(retval); } int ctl_report_supported_opcodes(struct ctl_scsiio *ctsio) { struct ctl_lun *lun; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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); } 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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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; } + 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_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; 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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr; data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_TRS; data->byte2 |= RST_ITNRS; + 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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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 scsi_per_res_in *cdb; int alloc_len, total_len = 0; /* struct scsi_per_res_in_rsrv in_data; */ struct ctl_lun *lun; struct ctl_softc *softc; CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n")); softc = control_softc; cdb = (struct scsi_per_res_in *)ctsio->cdb; alloc_len = scsi_2btoul(cdb->length); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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("Invalid PR type %x", cdb->action); } mtx_unlock(&lun->lun_lock); ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO); if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; 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->PRGeneration, 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 < 2*CTL_MAX_INITIATORS; i++) { if (lun->pr_keys[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) { #ifdef NEEDTOPORT csevent_log(CSC_CTL | CSC_SHELF_SW | CTL_PR_ERROR, csevent_LogType_Fault, csevent_AlertLevel_Yellow, csevent_FRU_ShelfController, csevent_FRU_Firmware, csevent_FRU_Unknown, "registered keys %d >= key " "count %d", key_count, lun->pr_key_count); #endif key_count++; continue; } scsi_u64to8b(lun->pr_keys[i], 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->PRGeneration, 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(lun->pr_keys[lun->pr_res_idx], res->data.reservation); } res->data.scopetype = lun->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->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->PRGeneration, res_status->header.generation); res_desc = &res_status->desc[0]; for (i = 0; i < 2*CTL_MAX_INITIATORS; i++) { if (lun->pr_keys[i] == 0) continue; scsi_u64to8b(lun->pr_keys[i], 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->res_type; } scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT, res_desc->rel_trgt_port_id); len = 0; port = softc->ctl_ports[ ctl_port_idx(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: /* * This is a bug, because we just checked for this above, * and should have returned an error. */ panic("Invalid PR type %x", cdb->action); break; /* NOTREACHED */ } 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_DEBUG_PRINT(("buf = %x %x %x %x %x %x %x %x\n", - ctsio->kern_data_ptr[0], ctsio->kern_data_ptr[1], - ctsio->kern_data_ptr[2], ctsio->kern_data_ptr[3], - ctsio->kern_data_ptr[4], ctsio->kern_data_ptr[5], - ctsio->kern_data_ptr[6], ctsio->kern_data_ptr[7])); - 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 retval, i; int isc_retval; retval = 0; 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 < 2*CTL_MAX_INITIATORS; i++) { if (i == residx || lun->pr_keys[i] == 0) continue; if (!persis_offset && i pending_ua[i] |= CTL_UA_REG_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i-persis_offset] |= CTL_UA_REG_PREEMPT; lun->pr_keys[i] = 0; } lun->pr_key_count = 1; lun->res_type = type; if (lun->res_type != SPR_TYPE_WR_EX_AR && lun->res_type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = residx; /* 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)); if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned " "from ctl_ha_msg_send %d\n", isc_retval); } } 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 < 2*CTL_MAX_INITIATORS; i++) { if (lun->pr_keys[i] != sa_res_key) continue; found = 1; lun->pr_keys[i] = 0; lun->pr_key_count--; if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_REG_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i-persis_offset] |= 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); } /* 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)); if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned from " "ctl_ha_msg_send %d\n", isc_retval); } } else { /* Reserved but not all registrants */ /* sa_res_key is res holder */ if (sa_res_key == lun->pr_keys[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 < 2*CTL_MAX_INITIATORS; i++) { if (i == residx || lun->pr_keys[i] == 0) continue; if (sa_res_key == lun->pr_keys[i]) { lun->pr_keys[i] = 0; lun->pr_key_count--; if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_REG_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i-persis_offset] |= CTL_UA_REG_PREEMPT; } else if (type != lun->res_type && (lun->res_type == SPR_TYPE_WR_EX_RO || lun->res_type ==SPR_TYPE_EX_AC_RO)){ if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_RES_RELEASE; else if (persis_offset && i >= persis_offset) lun->pending_ua[ i-persis_offset] |= CTL_UA_RES_RELEASE; } } lun->res_type = type; if (lun->res_type != SPR_TYPE_WR_EX_AR && lun->res_type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = residx; else lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS; 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)); if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned " "from ctl_ha_msg_send %d\n", isc_retval); } } else { /* * sa_res_key is not the res holder just * remove registrants */ int found=0; for (i=0; i < 2*CTL_MAX_INITIATORS; i++) { if (sa_res_key != lun->pr_keys[i]) continue; found = 1; lun->pr_keys[i] = 0; lun->pr_key_count--; if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_REG_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i-persis_offset] |= 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); } 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)); if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned " "from ctl_ha_msg_send %d\n", isc_retval); } } } lun->PRGeneration++; mtx_unlock(&lun->lun_lock); return (retval); } 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 != lun->pr_keys[lun->pr_res_idx]) { if (sa_res_key == 0) { /* * Unregister everybody else and build UA for * them */ for(i=0; i < 2*CTL_MAX_INITIATORS; i++) { if (i == msg->pr.pr_info.residx || lun->pr_keys[i] == 0) continue; if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_REG_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i - persis_offset] |= CTL_UA_REG_PREEMPT; lun->pr_keys[i] = 0; } lun->pr_key_count = 1; lun->res_type = msg->pr.pr_info.res_type; if (lun->res_type != SPR_TYPE_WR_EX_AR && lun->res_type != SPR_TYPE_EX_AC_AR) lun->pr_res_idx = msg->pr.pr_info.residx; } else { for (i=0; i < 2*CTL_MAX_INITIATORS; i++) { if (sa_res_key == lun->pr_keys[i]) continue; lun->pr_keys[i] = 0; lun->pr_key_count--; if (!persis_offset && i < persis_offset) lun->pending_ua[i] |= CTL_UA_REG_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i - persis_offset] |= CTL_UA_REG_PREEMPT; } } } else { for (i=0; i < 2*CTL_MAX_INITIATORS; i++) { if (i == msg->pr.pr_info.residx || lun->pr_keys[i] == 0) continue; if (sa_res_key == lun->pr_keys[i]) { lun->pr_keys[i] = 0; lun->pr_key_count--; if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_REG_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i - persis_offset] |= CTL_UA_REG_PREEMPT; } else if (msg->pr.pr_info.res_type != lun->res_type && (lun->res_type == SPR_TYPE_WR_EX_RO || lun->res_type == SPR_TYPE_EX_AC_RO)) { if (!persis_offset && i < persis_offset) lun->pending_ua[i] |= CTL_UA_RES_RELEASE; else if (persis_offset && i >= persis_offset) lun->pending_ua[i - persis_offset] |= CTL_UA_RES_RELEASE; } } lun->res_type = msg->pr.pr_info.res_type; if (lun->res_type != SPR_TYPE_WR_EX_AR && lun->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->PRGeneration++; } int ctl_persistent_reserve_out(struct ctl_scsiio *ctsio) { int retval; int isc_retval; u_int32_t param_len; struct scsi_per_res_out *cdb; struct ctl_lun *lun; struct scsi_per_res_out_parms* param; struct ctl_softc *softc; uint32_t residx; uint64_t res_key, sa_res_key; uint8_t type; union ctl_ha_msg persis_io; int i; CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n")); retval = CTL_RETVAL_COMPLETE; softc = control_softc; cdb = (struct scsi_per_res_out *)ctsio->cdb; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; /* * 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_data_resid = 0; 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_resindex(&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 (lun->pr_keys[residx] != 0) { if (res_key != lun->pr_keys[residx]) { /* * 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 && lun->pr_keys[residx] == 0)) { mtx_unlock(&lun->lun_lock); goto done; } lun->pr_keys[residx] = 0; 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->res_type == SPR_TYPE_WR_EX_RO || lun->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 = 0; i < CTL_MAX_INITIATORS;i++){ if (lun->pr_keys[ i + persis_offset] == 0) continue; lun->pending_ua[i] |= CTL_UA_RES_RELEASE; } } lun->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->res_type = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; } } 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; if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0 )) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned from " "ctl_ha_msg_send %d\n", isc_retval); } } else /* sa_res_key != 0 */ { /* * If we aren't registered currently then increment * the key count and set the registered flag. */ if (lun->pr_keys[residx] == 0) lun->pr_key_count++; lun->pr_keys[residx] = sa_res_key; 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)); if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned from " "ctl_ha_msg_send %d\n", isc_retval); } } lun->PRGeneration++; mtx_unlock(&lun->lun_lock); 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->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->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; if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned from " "ctl_ha_msg_send %d\n", isc_retval); } } 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->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->res_type = 0; /* * if this isn't an exclusive access * res generate UA for all other * registrants. */ if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX) { for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (i == residx || lun->pr_keys[i + persis_offset] == 0) continue; lun->pending_ua[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; if ((isc_retval=ctl_ha_msg_send( CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned from " "ctl_ha_msg_send %d\n", isc_retval); } break; case SPRO_CLEAR: /* send msg to other side */ mtx_lock(&lun->lun_lock); lun->flags &= ~CTL_LUN_PR_RESERVED; lun->res_type = 0; lun->pr_key_count = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; lun->pr_keys[residx] = 0; for (i=0; i < 2*CTL_MAX_INITIATORS; i++) if (lun->pr_keys[i] != 0) { if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_RES_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i-persis_offset] |= CTL_UA_RES_PREEMPT; lun->pr_keys[i] = 0; } lun->PRGeneration++; 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; if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io, sizeof(persis_io), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Persis Out error returned from " "ctl_ha_msg_send %d\n", isc_retval); } 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("Invalid PR type %x", 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_ha_msg *msg) { struct ctl_lun *lun; struct ctl_softc *softc; int i; uint32_t targ_lun; softc = control_softc; targ_lun = msg->hdr.nexus.targ_mapped_lun; lun = softc->ctl_luns[targ_lun]; mtx_lock(&lun->lun_lock); switch(msg->pr.pr_info.action) { case CTL_PR_REG_KEY: if (lun->pr_keys[msg->pr.pr_info.residx] == 0) lun->pr_key_count++; lun->pr_keys[msg->pr.pr_info.residx] = scsi_8btou64(msg->pr.pr_info.sa_res_key); lun->PRGeneration++; break; case CTL_PR_UNREG_KEY: lun->pr_keys[msg->pr.pr_info.residx] = 0; 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->res_type == SPR_TYPE_WR_EX_RO || lun->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 = 0; i < CTL_MAX_INITIATORS; i++) { if (lun->pr_keys[i+ persis_offset] == 0) continue; lun->pending_ua[i] |= CTL_UA_RES_RELEASE; } } lun->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->res_type = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; } } lun->PRGeneration++; break; case CTL_PR_RESERVE: lun->flags |= CTL_LUN_PR_RESERVED; lun->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 res generate UA for all * other registrants. */ if (lun->res_type != SPR_TYPE_EX_AC && lun->res_type != SPR_TYPE_WR_EX) { for (i = 0; i < CTL_MAX_INITIATORS; i++) if (lun->pr_keys[i+persis_offset] != 0) lun->pending_ua[i] |= CTL_UA_RES_RELEASE; } lun->flags &= ~CTL_LUN_PR_RESERVED; lun->pr_res_idx = CTL_PR_NO_RESERVATION; lun->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->res_type = 0; lun->pr_key_count = 0; lun->pr_res_idx = CTL_PR_NO_RESERVATION; for (i=0; i < 2*CTL_MAX_INITIATORS; i++) { if (lun->pr_keys[i] == 0) continue; if (!persis_offset && i < CTL_MAX_INITIATORS) lun->pending_ua[i] |= CTL_UA_RES_PREEMPT; else if (persis_offset && i >= persis_offset) lun->pending_ua[i-persis_offset] |= CTL_UA_RES_PREEMPT; lun->pr_keys[i] = 0; } lun->PRGeneration++; break; } mtx_unlock(&lun->lun_lock); } int ctl_read_write(struct ctl_scsiio *ctsio) { struct ctl_lun *lun; struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int flags, retval; int isread; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0])); flags = 0; retval = CTL_RETVAL_COMPLETE; 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_rw_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_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); 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); 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_pages.caching_page[CTL_PAGE_CURRENT].flags1 & SCP_RCD) != 0) flags |= CTL_LLF_FUA | CTL_LLF_DPO; } else { if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].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_scsiio *ctsio; struct ctl_lun *lun; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int flags, retval; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0])); flags = 0; retval = CTL_RETVAL_COMPLETE; 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); 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_pages.caching_page[CTL_PAGE_CURRENT].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; struct ctl_lba_len_flags *lbalen; uint64_t lba; uint32_t num_blocks; int bytchk, flags; int retval; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0])); bytchk = 0; flags = CTL_LLF_FUA; retval = CTL_RETVAL_COMPLETE; 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); 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 scsi_report_luns *cdb; struct scsi_report_luns_data *lun_data; struct ctl_lun *lun, *request_lun; int num_luns, retval; uint32_t alloc_len, lun_datalen; int num_filled, well_known; uint32_t initidx, targ_lun_id, lun_id; retval = CTL_RETVAL_COMPLETE; well_known = 0; cdb = (struct scsi_report_luns *)ctsio->cdb; CTL_DEBUG_PRINT(("ctl_report_luns\n")); mtx_lock(&control_softc->ctl_lock); num_luns = control_softc->num_luns; mtx_unlock(&control_softc->ctl_lock); switch (cdb->select_report) { case RPL_REPORT_DEFAULT: case RPL_REPORT_ALL: break; case RPL_REPORT_WELLKNOWN: well_known = 1; 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); } request_lun = (struct ctl_lun *) ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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(&control_softc->ctl_lock); for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) { lun_id = ctl_map_lun(ctsio->io_hdr.nexus.targ_port, targ_lun_id); if (lun_id >= CTL_MAX_LUNS) continue; lun = control_softc->ctl_luns[lun_id]; if (lun == NULL) continue; if (targ_lun_id <= 0xff) { /* * Peripheral addressing method, bus number 0. */ lun_data->luns[num_filled].lundata[0] = RPL_LUNDATA_ATYP_PERIPH; lun_data->luns[num_filled].lundata[1] = targ_lun_id; num_filled++; } else if (targ_lun_id <= 0x3fff) { /* * Flat addressing method. */ lun_data->luns[num_filled].lundata[0] = RPL_LUNDATA_ATYP_FLAT | (targ_lun_id >> 8); lun_data->luns[num_filled].lundata[1] = (targ_lun_id & 0xff); num_filled++; } else if (targ_lun_id <= 0xffffff) { /* * Extended flat addressing method. */ lun_data->luns[num_filled].lundata[0] = RPL_LUNDATA_ATYP_EXTLUN | 0x12; scsi_ulto3b(targ_lun_id, &lun_data->luns[num_filled].lundata[1]); num_filled++; } else { printf("ctl_report_luns: bogus LUN number %jd, " "skipping\n", (intmax_t)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); lun->pending_ua[initidx] &= ~CTL_UA_LUN_CHANGE; mtx_unlock(&lun->lun_lock); } } mtx_unlock(&control_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)); if (lun_datalen < alloc_len) { ctsio->residual = alloc_len - lun_datalen; ctsio->kern_data_len = lun_datalen; ctsio->kern_total_len = lun_datalen; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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. */ - ctsio->scsi_status = SCSI_STATUS_OK; - + 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 scsi_request_sense *cdb; struct scsi_sense_data *sense_ptr; struct ctl_lun *lun; uint32_t initidx; int have_error; scsi_sense_data_type sense_format; cdb = (struct scsi_request_sense *)ctsio->cdb; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; /* * 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->residual = 0; ctsio->kern_data_len = cdb->length; ctsio->kern_total_len = cdb->length; ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * If we don't have a LUN, we don't have any pending sense. */ if (lun == NULL) goto no_sense; 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); #ifdef CTL_WITH_CA if (ctl_is_set(lun->have_ca, initidx)) { scsi_sense_data_type stored_format; /* * Check to see which sense format was used for the stored * sense data. */ stored_format = scsi_sense_type(&lun->pending_sense[initidx]); /* * 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 *) &lun->pending_sense[initidx], (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 *) &lun->pending_sense[initidx], (struct scsi_sense_data_fixed *)sense_ptr); else memcpy(sense_ptr, &lun->pending_sense[initidx], ctl_min(sizeof(*sense_ptr), sizeof(lun->pending_sense[initidx]))); ctl_clear_mask(lun->have_ca, initidx); have_error = 1; } else #endif if (lun->pending_ua[initidx] != CTL_UA_NONE) { ctl_ua_type ua_type; ua_type = ctl_build_ua(&lun->pending_ua[initidx], sense_ptr, sense_format); if (ua_type != CTL_UA_NONE) have_error = 1; } mtx_unlock(&lun->lun_lock); /* * We already have a pending error, return it. */ if (have_error != 0) { /* * We report the SCSI status as OK, since the status of the * request sense command itself is OK. - */ - ctsio->scsi_status = SCSI_STATUS_OK; - - /* * We report 0 for the sense length, because we aren't doing * autosense in this case. We're reporting sense as * parameter data. */ - ctsio->sense_len = 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); } no_sense: /* * No sense information to report, so we report that everything is * okay. */ ctl_set_sense_data(sense_ptr, lun, sense_format, /*current_error*/ 1, /*sense_key*/ SSD_KEY_NO_SENSE, /*asc*/ 0x00, /*ascq*/ 0x00, SSD_ELEM_NONE); - ctsio->scsi_status = SCSI_STATUS_OK; - /* * We report 0 for the sense length, because we aren't doing * autosense in this case. We're reporting sense as parameter data. */ - ctsio->sense_len = 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_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); } #ifdef notyet static int ctl_cmddt_inquiry(struct ctl_scsiio *ctsio) { } #endif static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len) { struct scsi_vpd_supported_pages *pages; int sup_page_size; struct ctl_lun *lun; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_sg_entries = 0; if (sup_page_size < alloc_len) { ctsio->residual = alloc_len - sup_page_size; ctsio->kern_data_len = sup_page_size; ctsio->kern_total_len = sup_page_size; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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; pages->length = SCSI_EVPD_NUM_SUPPORTED_PAGES; /* Supported VPD pages */ pages->page_list[0] = SVPD_SUPPORTED_PAGES; /* Serial Number */ pages->page_list[1] = SVPD_UNIT_SERIAL_NUMBER; /* Device Identification */ pages->page_list[2] = SVPD_DEVICE_ID; /* Extended INQUIRY Data */ pages->page_list[3] = SVPD_EXTENDED_INQUIRY_DATA; /* Mode Page Policy */ pages->page_list[4] = SVPD_MODE_PAGE_POLICY; /* SCSI Ports */ pages->page_list[5] = SVPD_SCSI_PORTS; /* Third-party Copy */ pages->page_list[6] = SVPD_SCSI_TPC; /* Block limits */ pages->page_list[7] = SVPD_BLOCK_LIMITS; /* Block Device Characteristics */ pages->page_list[8] = SVPD_BDC; /* Logical Block Provisioning */ pages->page_list[9] = SVPD_LBP; - ctsio->scsi_status = SCSI_STATUS_OK; - + 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_serial(struct ctl_scsiio *ctsio, int alloc_len) { struct scsi_vpd_unit_serial_number *sn_ptr; struct ctl_lun *lun; int data_len; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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); - ctsio->scsi_status = SCSI_STATUS_OK; + 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_eid(struct ctl_scsiio *ctsio, int alloc_len) { struct scsi_vpd_extended_inquiry_data *eid_ptr; struct ctl_lun *lun; int data_len; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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; eid_ptr->page_length = data_len - 4; eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP; eid_ptr->flags3 = SVPD_EID_V_SUP; - ctsio->scsi_status = SCSI_STATUS_OK; + 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 scsi_vpd_mode_page_policy *mpp_ptr; struct ctl_lun *lun; int data_len; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_sg_entries = 0; if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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; - ctsio->scsi_status = SCSI_STATUS_OK; + 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_devid(struct ctl_scsiio *ctsio, int alloc_len) { struct scsi_vpd_device_id *devid_ptr; struct scsi_vpd_id_descriptor *desc; struct ctl_softc *ctl_softc; struct ctl_lun *lun; struct ctl_port *port; int data_len; uint8_t proto; ctl_softc = control_softc; port = ctl_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_devid) data_len += port->port_devid->len; if (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; if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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_type == CTL_PORT_FC) proto = SCSI_PROTO_FC << 4; else if (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_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; scsi_ulto2b(ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS + 1, &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->target_devid) { memcpy(desc, port->target_devid->data, port->target_devid->len); } - ctsio->scsi_status = SCSI_STATUS_OK; + 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 = control_softc; struct scsi_vpd_scsi_ports *sp; struct scsi_vpd_port_designation *pd; struct scsi_vpd_port_designation_cont *pdc; struct ctl_lun *lun; struct ctl_port *port; int data_len, num_target_ports, iid_len, id_len, g, pg, p; int num_target_port_groups; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; if (softc->is_single) num_target_port_groups = 1; else num_target_port_groups = NUM_TARGET_PORT_GROUPS; 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_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS) 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_port_groups * 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; if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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); pg = softc->port_offset / CTL_MAX_PORTS; for (g = 0; g < num_target_port_groups; g++) { STAILQ_FOREACH(port, &softc->port_list, links) { if ((port->status & CTL_PORT_STATUS_ONLINE) == 0) continue; if (lun != NULL && ctl_map_lun_back(port->targ_port, lun->lun) >= CTL_MAX_LUNS) continue; p = port->targ_port % CTL_MAX_PORTS + g * CTL_MAX_PORTS; scsi_ulto2b(p, pd->relative_port_id); if (port->init_devid && g == pg) { 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 && g == pg) { 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); - ctsio->scsi_status = SCSI_STATUS_OK; + 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 scsi_vpd_block_limits *bl_ptr; struct ctl_lun *lun; int bs; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; if (sizeof(*bl_ptr) < alloc_len) { ctsio->residual = alloc_len - sizeof(*bl_ptr); ctsio->kern_data_len = sizeof(*bl_ptr); ctsio->kern_total_len = sizeof(*bl_ptr); } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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) { bs = lun->be_lun->blocksize; scsi_ulto4b(MAXPHYS / bs, bl_ptr->opt_txfer_len); if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) { scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_lba_cnt); scsi_ulto4b(0xffffffff, bl_ptr->max_unmap_blk_cnt); if (lun->be_lun->pblockexp != 0) { scsi_ulto4b((1 << lun->be_lun->pblockexp), bl_ptr->opt_unmap_grain); scsi_ulto4b(0x80000000 | lun->be_lun->pblockoff, 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_u64to8b(UINT64_MAX, bl_ptr->max_write_same_length); - ctsio->scsi_status = SCSI_STATUS_OK; + 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 scsi_vpd_block_device_characteristics *bdc_ptr; struct ctl_lun *lun; const char *value; u_int i; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; if (sizeof(*bdc_ptr) < alloc_len) { ctsio->residual = alloc_len - sizeof(*bdc_ptr); ctsio->kern_data_len = sizeof(*bdc_ptr); ctsio->kern_total_len = sizeof(*bdc_ptr); } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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) 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) i = strtol(value, NULL, 0); else i = 0; bdc_ptr->wab_wac_ff = (i & 0x0f); bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS; - ctsio->scsi_status = SCSI_STATUS_OK; + 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 scsi_vpd_logical_block_prov *lbp_ptr; struct ctl_lun *lun; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; if (sizeof(*lbp_ptr) < alloc_len) { ctsio->residual = alloc_len - sizeof(*lbp_ptr); ctsio->kern_data_len = sizeof(*lbp_ptr); ctsio->kern_total_len = sizeof(*lbp_ptr); } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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); if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) { lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT; lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 | SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP; lbp_ptr->prov_type = SVPD_LBP_THIN; } - ctsio->scsi_status = SCSI_STATUS_OK; + 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(struct ctl_scsiio *ctsio) { struct scsi_inquiry *cdb; int alloc_len, retval; cdb = (struct scsi_inquiry *)ctsio->cdb; retval = CTL_RETVAL_COMPLETE; 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: retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len); break; case SVPD_BDC: retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len); break; case SVPD_LBP: retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len); 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); retval = CTL_RETVAL_COMPLETE; break; } return (retval); } static int ctl_inquiry_std(struct ctl_scsiio *ctsio) { struct scsi_inquiry_data *inq_ptr; struct scsi_inquiry *cdb; struct ctl_softc *ctl_softc; struct ctl_lun *lun; char *val; uint32_t alloc_len, data_len; ctl_port_type port_type; ctl_softc = control_softc; /* * Figure out whether we're talking to a Fibre Channel port or not. * We treat the ioctl front end, and any SCSI adapters, as packetized * SCSI front ends. */ port_type = ctl_softc->ctl_ports[ ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]->port_type; if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL) port_type = CTL_PORT_SCSI; lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_data_resid = 0; ctsio->kern_rel_offset = 0; if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } /* * If we have a LUN configured, report it as connected. Otherwise, * report that it is offline or no device is supported, depending * on the value of inquiry_pq_no_lun. * * According to the spec (SPC-4 r34), the peripheral qualifier * SID_QUAL_LU_OFFLINE (001b) is used in the following scenario: * * "A peripheral device having the specified peripheral device type * is not connected to this logical unit. However, the device * server is capable of supporting the specified peripheral device * type on this logical unit." * * According to the same spec, the peripheral qualifier * SID_QUAL_BAD_LU (011b) is used in this scenario: * * "The device server is not capable of supporting a peripheral * device on this logical unit. For this peripheral qualifier the * peripheral device type shall be set to 1Fh. All other peripheral * device type values are reserved for this peripheral qualifier." * * Given the text, it would seem that we probably want to report that * the LUN is offline here. There is no LUN connected, but we can * support a LUN at the given LUN number. * * In the real world, though, it sounds like things are a little * different: * * - Linux, when presented with a LUN with the offline peripheral * qualifier, will create an sg driver instance for it. So when * you attach it to CTL, you wind up with a ton of sg driver * instances. (One for every LUN that Linux bothered to probe.) * Linux does this despite the fact that it issues a REPORT LUNs * to LUN 0 to get the inventory of supported LUNs. * * - There is other anecdotal evidence (from Emulex folks) about * arrays that use the offline peripheral qualifier for LUNs that * are on the "passive" path in an active/passive array. * * So the solution is provide a hopefully reasonable default * (return bad/no LUN) and allow the user to change the behavior * with a tunable/sysctl variable. */ if (lun != NULL) inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) | lun->be_lun->lun_type; else if (ctl_softc->inquiry_pq_no_lun == 0) inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT; else inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE; /* RMB in byte 2 is 0 */ inq_ptr->version = SCSI_REV_SPC4; /* * 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 reponse 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; /* 16 bit addressing */ if (port_type == CTL_PORT_SCSI) inq_ptr->spc2_flags = SPC2_SID_ADDR16; /* XXX set the SID_MultiP bit here if we're actually going to respond on multiple ports */ inq_ptr->spc2_flags |= SPC2_SID_MultiP; /* 16 bit data bus, synchronous transfers */ if (port_type == CTL_PORT_SCSI) inq_ptr->flags = SID_WBus16 | SID_Sync; /* * XXX KDM do we want to support tagged queueing on the control * device at all? */ if ((lun == NULL) || (lun->be_lun->lun_type != T_PROCESSOR)) inq_ptr->flags |= SID_CmdQue; /* * 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) { 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) { 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; 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) { 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-5 (no version claimed) */ scsi_ulto2b(0x00A0, inq_ptr->version1); /* SPC-4 (no version claimed) */ scsi_ulto2b(0x0460, 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); } 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: default: break; } } - ctsio->scsi_status = SCSI_STATUS_OK; + 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); } /* * 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: case WRITE_ATOMIC_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_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; } 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) { uint64_t endlba1, endlba2; endlba1 = lba1 + len1 - 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) { uint64_t lba1, lba2; uint64_t len1, len2; int retval; if (ctl_get_lba_len(io1, &lba1, &len1) != 0) return (CTL_ACTION_ERROR); retval = ctl_extent_check_unmap(io2, lba1, len1); if (retval != CTL_ACTION_ERROR) return (retval); if (ctl_get_lba_len(io2, &lba2, &len2) != 0) return (CTL_ACTION_ERROR); return (ctl_extent_check_lba(lba1, len1, lba2, len2)); } 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; 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.id == ooa_io->io_hdr.nexus.initid.id)) && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 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.id == ooa_io->io_hdr.nexus.initid.id)) && ((ooa_io->io_hdr.flags & CTL_FLAG_ABORT) == 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); ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL); 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(pending_io, ooa_io)); case CTL_SER_EXTENTOPT: if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED) return (ctl_extent_check(pending_io, ooa_io)); /* FALLTHROUGH */ case CTL_SER_PASS: return (CTL_ACTION_PASS); case CTL_SER_BLOCKOPT: if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].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("invalid serialization value %d", serialize_row[pending_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("invalid action %d", action); break; /* NOTREACHED */ } } 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) { 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: { struct ctl_softc *softc; const struct ctl_cmd_entry *entry; uint32_t initidx; int isc_retval; /* * 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 (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; 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; if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info, sizeof(msg_info), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:Check Blocked error from " "ctl_ha_msg_send %d\n", isc_retval); } break; } entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL); softc = control_softc; initidx = ctl_get_initindex(&cur_blocked->io_hdr.nexus); /* * 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(softc, 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_softc *ctl_softc, struct ctl_lun *lun, const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio) { int retval; uint32_t residx; retval = 0; mtx_assert(&lun->lun_lock, MA_OWNED); /* * If this shelf is a secondary shelf controller, we have to reject * any media access commands. */ if ((ctl_softc->flags & CTL_FLAG_ACTIVE_SHELF) == 0 && (entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0) { ctl_set_lun_standby(ctsio); retval = 1; goto bailout; } if (entry->pattern & CTL_LUN_PAT_WRITE) { if (lun->flags & CTL_LUN_READONLY) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_DATA_PROTECT, /*asc*/ 0x27, /*ascq*/ 0x01, SSD_ELEM_NONE); retval = 1; goto bailout; } if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT] .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_resindex(&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->res_type == SPR_TYPE_WR_EX || lun->res_type == SPR_TYPE_WR_EX_RO || lun->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 (lun->pr_keys[residx] == 0 || (residx != lun->pr_res_idx && lun->res_type < 4)) { ctl_set_reservation_conflict(ctsio); retval = 1; goto bailout; } } if ((lun->flags & CTL_LUN_OFFLINE) && ((entry->flags & CTL_CMD_FLAG_OK_ON_OFFLINE) == 0)) { ctl_set_lun_not_ready(ctsio); retval = 1; goto bailout; } /* * If the LUN is stopped, see if this particular command is allowed * for a stopped lun. Otherwise, reject it with 0x04,0x02. */ if ((lun->flags & CTL_LUN_STOPPED) && ((entry->flags & CTL_CMD_FLAG_OK_ON_STOPPED) == 0)) { /* "Logical unit not ready, initializing cmd. required" */ ctl_set_lun_stopped(ctsio); retval = 1; goto bailout; } if ((lun->flags & CTL_LUN_INOPERABLE) && ((entry->flags & CTL_CMD_FLAG_OK_ON_INOPERABLE) == 0)) { /* "Medium format corrupted" */ ctl_set_medium_format_corrupted(ctsio); 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(void) { struct ctl_lun *lun; struct ctl_softc *ctl_softc; union ctl_io *next_io, *pending_io; union ctl_io *io; int lun_idx; int i; ctl_softc = control_softc; mtx_lock(&ctl_softc->ctl_lock); /* * Remove any cmds from the other SC from the rtr queue. These * will obviously only be for LUNs for which we're the primary. * We can't send status or get/send data for these commands. * Since they haven't been executed yet, we can just remove them. * We'll either abort them or delete them below, depending on * which HA mode we're in. */ #ifdef notyet mtx_lock(&ctl_softc->queue_lock); for (io = (union ctl_io *)STAILQ_FIRST(&ctl_softc->rtr_queue); io != NULL; io = next_io) { next_io = (union ctl_io *)STAILQ_NEXT(&io->io_hdr, links); if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) STAILQ_REMOVE(&ctl_softc->rtr_queue, &io->io_hdr, ctl_io_hdr, links); } mtx_unlock(&ctl_softc->queue_lock); #endif for (lun_idx=0; lun_idx < ctl_softc->num_luns; lun_idx++) { lun = ctl_softc->ctl_luns[lun_idx]; if (lun==NULL) continue; /* * Processor LUNs are primary on both sides. * XXX will this always be true? */ if (lun->be_lun->lun_type == T_PROCESSOR) continue; if ((lun->flags & CTL_LUN_PRIMARY_SC) && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) { printf("FAILOVER: primary lun %d\n", lun_idx); /* * Remove all commands from the other SC. First from the * blocked queue then from the ooa queue. Once we have * removed them. Call ctl_check_blocked to see if there * is anything that can run. */ for (io = (union ctl_io *)TAILQ_FIRST( &lun->blocked_queue); io != NULL; io = next_io) { next_io = (union ctl_io *)TAILQ_NEXT( &io->io_hdr, blocked_links); if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) { TAILQ_REMOVE(&lun->blocked_queue, &io->io_hdr,blocked_links); io->io_hdr.flags &= ~CTL_FLAG_BLOCKED; TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links); ctl_free_io(io); } } for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL; io = next_io) { next_io = (union ctl_io *)TAILQ_NEXT( &io->io_hdr, ooa_links); if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) { TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links); ctl_free_io(io); } } ctl_check_blocked(lun); } else if ((lun->flags & CTL_LUN_PRIMARY_SC) && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) { printf("FAILOVER: primary lun %d\n", lun_idx); /* * Abort all commands from the other SC. We can't * send status back for them now. These should get * cleaned up when they are completed or come out * for a datamove operation. */ for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); io != NULL; io = next_io) { next_io = (union ctl_io *)TAILQ_NEXT( &io->io_hdr, ooa_links); if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) io->io_hdr.flags |= CTL_FLAG_ABORT; } } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0) && (ctl_softc->ha_mode == CTL_HA_MODE_XFER)) { printf("FAILOVER: secondary lun %d\n", lun_idx); lun->flags |= CTL_LUN_PRIMARY_SC; /* * We send all I/O that was sent to this controller * and redirected to the other side back with * busy status, and have the initiator retry it. * Figuring out how much data has been transferred, * etc. and picking up where we left off would be * very tricky. * * XXX KDM need to remove I/O from the blocked * queue as well! */ for (pending_io = (union ctl_io *)TAILQ_FIRST( &lun->ooa_queue); pending_io != NULL; pending_io = next_io) { next_io = (union ctl_io *)TAILQ_NEXT( &pending_io->io_hdr, ooa_links); pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC; if (pending_io->io_hdr.flags & CTL_FLAG_IO_ACTIVE) { pending_io->io_hdr.flags |= CTL_FLAG_FAILOVER; } else { ctl_set_busy(&pending_io->scsiio); ctl_done(pending_io); } } /* * Build Unit Attention */ for (i = 0; i < CTL_MAX_INITIATORS; i++) { lun->pending_ua[i] |= CTL_UA_ASYM_ACC_CHANGE; } } else if (((lun->flags & CTL_LUN_PRIMARY_SC) == 0) && (ctl_softc->ha_mode == CTL_HA_MODE_SER_ONLY)) { printf("FAILOVER: secondary lun %d\n", lun_idx); /* * if the first io on the OOA is not on the RtR queue * add it. */ lun->flags |= CTL_LUN_PRIMARY_SC; pending_io = (union ctl_io *)TAILQ_FIRST( &lun->ooa_queue); if (pending_io==NULL) { printf("Nothing on OOA queue\n"); continue; } pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC; if ((pending_io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR) == 0) { pending_io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; ctl_enqueue_rtr(pending_io); } #if 0 else { printf("Tag 0x%04x is running\n", pending_io->scsiio.tag_num); } #endif next_io = (union ctl_io *)TAILQ_NEXT( &pending_io->io_hdr, ooa_links); for (pending_io=next_io; pending_io != NULL; pending_io = next_io) { pending_io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC; next_io = (union ctl_io *)TAILQ_NEXT( &pending_io->io_hdr, ooa_links); if (pending_io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR) { #if 0 printf("Tag 0x%04x is running\n", pending_io->scsiio.tag_num); #endif continue; } switch (ctl_check_ooa(lun, pending_io, (union ctl_io *)TAILQ_PREV( &pending_io->io_hdr, ctl_ooaq, ooa_links))) { case CTL_ACTION_BLOCK: TAILQ_INSERT_TAIL(&lun->blocked_queue, &pending_io->io_hdr, blocked_links); pending_io->io_hdr.flags |= CTL_FLAG_BLOCKED; break; case CTL_ACTION_PASS: case CTL_ACTION_SKIP: pending_io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR; ctl_enqueue_rtr(pending_io); break; case CTL_ACTION_OVERLAP: ctl_set_overlapped_cmd( (struct ctl_scsiio *)pending_io); ctl_done(pending_io); break; case CTL_ACTION_OVERLAP_TAG: ctl_set_overlapped_tag( (struct ctl_scsiio *)pending_io, pending_io->scsiio.tag_num & 0xff); ctl_done(pending_io); break; case CTL_ACTION_ERROR: default: ctl_set_internal_failure( (struct ctl_scsiio *)pending_io, 0, // sks_valid 0); //retry count ctl_done(pending_io); break; } } /* * Build Unit Attention */ for (i = 0; i < CTL_MAX_INITIATORS; i++) { lun->pending_ua[i] |= CTL_UA_ASYM_ACC_CHANGE; } } else { panic("Unhandled HA mode failover, LUN flags = %#x, " "ha_mode = #%x", lun->flags, ctl_softc->ha_mode); } } ctl_pause_rtr = 0; mtx_unlock(&ctl_softc->ctl_lock); } static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc, struct ctl_scsiio *ctsio) { struct ctl_lun *lun; const struct ctl_cmd_entry *entry; uint32_t initidx, targ_lun; int retval; retval = 0; lun = NULL; targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun; if ((targ_lun < CTL_MAX_LUNS) && ((lun = ctl_softc->ctl_luns[targ_lun]) != NULL)) { /* * 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; ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL; ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL; } else { ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun; ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun; if (lun->be_lun->lun_type == T_PROCESSOR) { ctsio->io_hdr.flags |= CTL_FLAG_CONTROL_DEV; } /* * Every I/O goes into the OOA queue for a * particular LUN, and stays there until completion. */ TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links); } } else { ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL; ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL; } /* 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_ALL_LUNS) { 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); #ifdef CTL_WITH_CA /* * 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) && (ctl_is_set(lun->have_ca, initidx))) ctl_clear_mask(lun->have_ca, initidx); #endif /* * 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; if (lun->pending_ua[initidx] != CTL_UA_NONE) { scsi_sense_data_type sense_format; if (lun != NULL) sense_format = (lun->flags & CTL_LUN_SENSE_DESC) ? SSD_TYPE_DESC : SSD_TYPE_FIXED; else sense_format = SSD_TYPE_FIXED; ua_type = ctl_build_ua(&lun->pending_ua[initidx], &ctsio->sense_data, sense_format); if (ua_type != CTL_UA_NONE) { ctsio->scsi_status = SCSI_STATUS_CHECK_COND; ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; ctsio->sense_len = SSD_FULL_SIZE; mtx_unlock(&lun->lun_lock); ctl_done((union ctl_io *)ctsio); return (retval); } } } if (ctl_scsiio_lun_check(ctl_softc, 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) { union ctl_ha_msg msg_info; int isc_retval; ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC; msg_info.hdr.msg_type = CTL_MSG_SERIALIZE; msg_info.hdr.original_sc = (union ctl_io *)ctsio; #if 0 printf("1. ctsio %p\n", ctsio); #endif 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; memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN); ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; if ((isc_retval=ctl_ha_msg_send(CTL_HA_CHAN_CTL, (void *)&msg_info, sizeof(msg_info), 0)) > CTL_HA_STATUS_SUCCESS) { printf("CTL:precheck, ctl_ha_msg_send returned %d\n", isc_retval); printf("CTL:opcode is %x\n", ctsio->cdb[0]); } else { #if 0 printf("CTL:Precheck sent msg, opcode is %x\n",opcode); #endif } /* * XXX KDM this I/O is off the incoming queue, but hasn't * been inserted on any other queue. We may need to come * up with a holding queue while we wait for serialization * so that we have an idea of what we're waiting for from * the other side. */ mtx_unlock(&lun->lun_lock); 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_PROCESSOR: if (((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0) && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 0)) return (0); break; case T_DIRECT: if (((entry->flags & CTL_CMD_FLAG_OK_ON_SLUN) == 0) && ((entry->flags & CTL_CMD_FLAG_OK_ON_ALL_LUNS) == 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); } /* * Since we only implement one target right now, a bus reset simply resets * our single target. */ static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io) { return(ctl_target_reset(ctl_softc, io, CTL_UA_BUS_RESET)); } static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io, ctl_ua_type ua_type) { struct ctl_lun *lun; int retval; if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) { union ctl_ha_msg msg_info; io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC; msg_info.hdr.nexus = io->io_hdr.nexus; if (ua_type==CTL_UA_TARG_RESET) msg_info.task.task_action = CTL_TASK_TARGET_RESET; else msg_info.task.task_action = CTL_TASK_BUS_RESET; msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS; msg_info.hdr.original_sc = NULL; msg_info.hdr.serializing_sc = NULL; if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL, (void *)&msg_info, sizeof(msg_info), 0)) { } } retval = 0; mtx_lock(&ctl_softc->ctl_lock); STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) retval += ctl_lun_reset(lun, io, ua_type); mtx_unlock(&ctl_softc->ctl_lock); return (retval); } /* * 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 int ctl_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type) { union ctl_io *xio; #if 0 uint32_t initindex; #endif int i; mtx_lock(&lun->lun_lock); /* * Run through the OOA queue and abort each I/O. */ #if 0 TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) { #endif 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; } /* * This version sets unit attention for every */ #if 0 initindex = ctl_get_initindex(&io->io_hdr.nexus); for (i = 0; i < CTL_MAX_INITIATORS; i++) { if (initindex == i) continue; lun->pending_ua[i] |= ua_type; } #endif /* * A reset (any kind, really) clears reservations established with * RESERVE/RELEASE. It does not clear reservations established * with PERSISTENT RESERVE OUT, but we don't support that at the * moment anyway. See SPC-2, section 5.6. SPC-3 doesn't address * reservations made with the RESERVE/RELEASE commands, because * those commands are obsolete in SPC-3. */ lun->flags &= ~CTL_LUN_RESERVED; for (i = 0; i < CTL_MAX_INITIATORS; i++) { #ifdef CTL_WITH_CA ctl_clear_mask(lun->have_ca, i); #endif lun->pending_ua[i] |= ua_type; } mtx_unlock(&lun->lun_lock); 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.id)) { if (targ_port != xio->io_hdr.nexus.targ_port || init_id != xio->io_hdr.nexus.initid.id) 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, (void *)&msg_info, sizeof(msg_info), 0); } } } } static int ctl_abort_task_set(union ctl_io *io) { struct ctl_softc *softc = control_softc; 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) && (softc->ctl_luns[targ_lun] != NULL)) lun = softc->ctl_luns[targ_lun]; else { mtx_unlock(&softc->ctl_lock); 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.id, (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); return (0); } static int ctl_i_t_nexus_reset(union ctl_io *io) { struct ctl_softc *softc = control_softc; struct ctl_lun *lun; uint32_t initindex, residx; initindex = ctl_get_initindex(&io->io_hdr.nexus); residx = ctl_get_resindex(&io->io_hdr.nexus); mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { mtx_lock(&lun->lun_lock); ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.initid.id, (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0); #ifdef CTL_WITH_CA ctl_clear_mask(lun->have_ca, initindex); #endif if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx)) lun->flags &= ~CTL_LUN_RESERVED; lun->pending_ua[initindex] |= CTL_UA_I_T_NEXUS_LOSS; mtx_unlock(&lun->lun_lock); } mtx_unlock(&softc->ctl_lock); return (0); } static int ctl_abort_task(union ctl_io *io) { union ctl_io *xio; struct ctl_lun *lun; struct ctl_softc *ctl_softc; #if 0 struct sbuf sb; char printbuf[128]; #endif int found; uint32_t targ_lun; ctl_softc = control_softc; found = 0; /* * Look up the LUN. */ targ_lun = io->io_hdr.nexus.targ_mapped_lun; mtx_lock(&ctl_softc->ctl_lock); if ((targ_lun < CTL_MAX_LUNS) && (ctl_softc->ctl_luns[targ_lun] != NULL)) lun = ctl_softc->ctl_luns[targ_lun]; else { mtx_unlock(&ctl_softc->ctl_lock); 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(&ctl_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. */ #if 0 TAILQ_FOREACH((struct ctl_io_hdr *)xio, &lun->ooa_queue, ooa_links) { #endif 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.id == io->io_hdr.nexus.initid.id)) { /* * 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; io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC; 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 if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL, (void *)&msg_info, sizeof(msg_info), 0)) { } } #if 0 printf("ctl_abort_task: found I/O to abort\n"); #endif break; } } } 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: " "%d:%d:%d:%d tag %d type %d\n", io->io_hdr.nexus.initid.id, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_target.id, io->io_hdr.nexus.targ_lun, io->taskio.tag_num, io->taskio.tag_type); #endif } return (0); } static void ctl_run_task(union ctl_io *io) { struct ctl_softc *ctl_softc = control_softc; int retval = 1; const char *task_desc; 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)); task_desc = ctl_scsi_task_string(&io->taskio); if (task_desc != NULL) { #ifdef NEEDTOPORT csevent_log(CSC_CTL | CSC_SHELF_SW | CTL_TASK_REPORT, csevent_LogType_Trace, csevent_Severity_Information, csevent_AlertLevel_Green, csevent_FRU_Firmware, csevent_FRU_Unknown, "CTL: received task: %s",task_desc); #endif } else { #ifdef NEEDTOPORT csevent_log(CSC_CTL | CSC_SHELF_SW | CTL_TASK_REPORT, csevent_LogType_Trace, csevent_Severity_Information, csevent_AlertLevel_Green, csevent_FRU_Firmware, csevent_FRU_Unknown, "CTL: received unknown task " "type: %d (%#x)", io->taskio.task_action, io->taskio.task_action); #endif } 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: { struct ctl_lun *lun; uint32_t targ_lun; targ_lun = io->io_hdr.nexus.targ_mapped_lun; mtx_lock(&ctl_softc->ctl_lock); if ((targ_lun < CTL_MAX_LUNS) && (ctl_softc->ctl_luns[targ_lun] != NULL)) lun = ctl_softc->ctl_luns[targ_lun]; else { mtx_unlock(&ctl_softc->ctl_lock); retval = 1; break; } if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) { union ctl_ha_msg msg_info; io->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC; 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; if (CTL_HA_STATUS_SUCCESS != ctl_ha_msg_send(CTL_HA_CHAN_CTL, (void *)&msg_info, sizeof(msg_info), 0)) { } } retval = ctl_lun_reset(lun, io, CTL_UA_LUN_RESET); mtx_unlock(&ctl_softc->ctl_lock); break; } case CTL_TASK_TARGET_RESET: retval = ctl_target_reset(ctl_softc, io, CTL_UA_TARG_RESET); break; case CTL_TASK_BUS_RESET: retval = ctl_bus_reset(ctl_softc, io); break; case CTL_TASK_PORT_LOGIN: break; case CTL_TASK_PORT_LOGOUT: break; default: printf("ctl_run_task: got unknown task management event %d\n", 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) { int free_io; struct ctl_lun *lun; struct ctl_softc *ctl_softc; uint32_t targ_lun; ctl_softc = control_softc; targ_lun = io->io_hdr.nexus.targ_mapped_lun; lun = ctl_softc->ctl_luns[targ_lun]; switch (io->io_hdr.msg_type) { case CTL_MSG_SERIALIZE: free_io = ctl_serialize_other_sc_cmd(&io->scsiio); break; case CTL_MSG_R2R: { const struct ctl_cmd_entry *entry; /* * This is only used in SER_ONLY mode. */ free_io = 0; entry = ctl_get_cmd_entry(&io->scsiio, NULL); mtx_lock(&lun->lun_lock); if (ctl_scsiio_lun_check(ctl_softc, lun, entry, (struct ctl_scsiio *)io) != 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 (ctl_softc->ha_mode == CTL_HA_MODE_XFER) { free_io = 0; ctl_done(io); } else { free_io = 1; mtx_lock(&lun->lun_lock); TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links); ctl_check_blocked(lun); mtx_unlock(&lun->lun_lock); } break; case CTL_MSG_PERS_ACTION: ctl_hndl_per_res_out_on_other_sc( (union ctl_ha_msg *)&io->presio.pr_msg); free_io = 1; break; case CTL_MSG_BAD_JUJU: free_io = 0; ctl_done(io); break; case CTL_MSG_DATAMOVE: /* Only used in XFER mode */ free_io = 0; ctl_datamove_remote(io); break; case CTL_MSG_DATAMOVE_DONE: /* Only used in XFER mode */ free_io = 0; io->scsiio.be_move_done(io); break; default: free_io = 1; printf("%s: Invalid message type %d\n", __func__, io->io_hdr.msg_type); break; } if (free_io) ctl_free_io(io); } /* * 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); /* * 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); 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, ctl_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(&control_softc->ctl_lock, MA_NOTOWNED); CTL_DEBUG_PRINT(("ctl_datamove\n")); #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: printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type); panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type); break; } 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) { struct ctl_lun *lun; lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE; } else { struct ctl_lun *lun; lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; if ((lun != NULL) && (lun->delay_info.datamove_delay > 0)) { struct callout *callout; callout = (struct callout *)&io->io_hdr.timer_bytes; callout_init(callout, /*mpsafe*/ 1); io->io_hdr.flags |= CTL_FLAG_DELAY_DONE; callout_reset(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 (%ju:%d:%ju:%d) aborted\n", io->scsiio.tag_num,(uintmax_t)io->io_hdr.nexus.initid.id, io->io_hdr.nexus.targ_port, (uintmax_t)io->io_hdr.nexus.targ_target.id, 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; } /* * If we're in XFER mode and this I/O is from the other shelf * controller, we need to send the DMA to the other side to * actually transfer the data to/from the host. In serialize only * mode the transfer happens below CTL and ctl_datamove() is only * called on the machine that originally received the I/O. */ if ((control_softc->ha_mode == CTL_HA_MODE_XFER) && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) { union ctl_ha_msg msg; uint32_t sg_entries_sent; int do_sg_copy; int i; memset(&msg, 0, sizeof(msg)); 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.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 this is in cached memory, flush the cache * before we send the DMA request to the other * controller. We want to do this in either the * read or the write case. The read case is * straightforward. In the write case, we want to * make sure nothing is in the local cache that * could overwrite the DMAed data. */ if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) { /* * XXX KDM use bus_dmamap_sync() here. */ } /* * Convert to a physical address if this is a * virtual address. */ 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! */ #if 0 msg.dt.sg_list[0].addr = (void *) vtophys(io->scsiio.kern_data_ptr); #endif } msg.dt.sg_list[0].len = io->scsiio.kern_data_len; do_sg_copy = 0; } else { struct ctl_sg_entry *sgl; do_sg_copy = 1; msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries; sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) { /* * XXX KDM use bus_dmamap_sync() here. */ } } 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 passing it to the */ for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries; msg.dt.sg_sequence++) { msg.dt.cur_sg_entries = ctl_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) { struct ctl_sg_entry *sgl; int j; sgl = (struct ctl_sg_entry *) io->scsiio.kern_data_ptr; /* * If this is in cached memory, flush the cache * before we send the DMA request to the other * controller. We want to do this in either * the * read or the write case. The read * case is straightforward. In the write * case, we want to make sure nothing is * in the local cache that could overwrite * the DMAed data. */ for (i = sg_entries_sent, j = 0; i < msg.dt.cur_sg_entries; i++, j++) { if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) { /* * XXX KDM use bus_dmamap_sync() */ } if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) { /* * XXX KDM use busdma. */ #if 0 msg.dt.sg_list[j].addr =(void *) vtophys(sgl[i].addr); #endif } else { msg.dt.sg_list[j].addr = sgl[i].addr; } msg.dt.sg_list[j].len = sgl[i].len; } } sg_entries_sent += msg.dt.cur_sg_entries; if (sg_entries_sent >= msg.dt.kern_sg_entries) msg.dt.sg_last = 1; else msg.dt.sg_last = 0; /* * XXX KDM drop and reacquire the lock here? */ if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) { /* * XXX do something here. */ } msg.dt.sent_sg_entries = sg_entries_sent; } io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; if (io->io_hdr.flags & CTL_FLAG_FAILOVER) ctl_failover_io(io, /*have_lock*/ 0); } else { /* * Lookup the fe_datamove() function for this particular * front end. */ fe_datamove = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove; fe_datamove(io); } } static void ctl_send_datamove_done(union ctl_io *io, int have_lock) { union ctl_ha_msg msg; int isc_status; 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.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, sizeof(io->scsiio.sense_data)); msg.scsi.sense_len = io->scsiio.sense_len; msg.scsi.sense_residual = io->scsiio.sense_residual; msg.scsi.fetd_status = io->io_hdr.port_status; msg.scsi.residual = io->scsiio.residual; 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; } isc_status = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0); if (isc_status > CTL_HA_STATUS_SUCCESS) { /* XXX do something if this fails */ } } /* * 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; 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); /* * In this case, we had to malloc the memory locally. Free it. */ if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) { int i; for (i = 0; i < io->scsiio.kern_sg_entries; i++) free(io->io_hdr.local_sglist[i].addr, M_CTL); } /* * 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 = 0; 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 = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove; fe_datamove(io); return; } 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 /* * In this case, we had to malloc the memory locally. Free it. */ if ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 0) { int i; for (i = 0; i < io->scsiio.kern_sg_entries; i++) free(io->io_hdr.local_sglist[i].addr, M_CTL); } #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", __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 = control_softc->ctl_ports[ctl_port_idx(io->io_hdr.nexus.targ_port)]->fe_datamove; fe_datamove(io); } static int ctl_datamove_remote_sgl_setup(union ctl_io *io) { struct ctl_sg_entry *local_sglist, *remote_sglist; struct ctl_sg_entry *local_dma_sglist, *remote_dma_sglist; struct ctl_softc *softc; int retval; int i; retval = 0; softc = control_softc; local_sglist = io->io_hdr.local_sglist; local_dma_sglist = io->io_hdr.local_dma_sglist; remote_sglist = io->io_hdr.remote_sglist; remote_dma_sglist = io->io_hdr.remote_dma_sglist; if (io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) { for (i = 0; i < io->scsiio.kern_sg_entries; i++) { local_sglist[i].len = remote_sglist[i].len; /* * XXX Detect the situation where the RS-level I/O * redirector on the other side has already read the * data off of the AOR RS on this side, and * transferred it to remote (mirror) memory on the * other side. Since we already have the data in * memory here, we just need to use it. * * XXX KDM this can probably be removed once we * get the cache device code in and take the * current AOR implementation out. */ #ifdef NEEDTOPORT if ((remote_sglist[i].addr >= (void *)vtophys(softc->mirr->addr)) && (remote_sglist[i].addr < ((void *)vtophys(softc->mirr->addr) + CacheMirrorOffset))) { local_sglist[i].addr = remote_sglist[i].addr - CacheMirrorOffset; if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) io->io_hdr.flags |= CTL_FLAG_REDIR_DONE; } else { local_sglist[i].addr = remote_sglist[i].addr + CacheMirrorOffset; } #endif #if 0 printf("%s: local %p, remote %p, len %d\n", __func__, local_sglist[i].addr, remote_sglist[i].addr, local_sglist[i].len); #endif } } else { uint32_t len_to_go; /* * In this case, we don't have automatically allocated * memory for this I/O on this controller. This typically * happens with internal CTL I/O -- e.g. inquiry, mode * sense, etc. Anything coming from RAIDCore will have * a mirror area available. */ len_to_go = io->scsiio.kern_data_len; /* * Clear the no datasync flag, we have to use malloced * buffers. */ io->io_hdr.flags &= ~CTL_FLAG_NO_DATASYNC; /* * 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; (i < sizeof(io->io_hdr.remote_sglist) / sizeof(io->io_hdr.remote_sglist[0])) && (len_to_go > 0); i++) { local_sglist[i].len = ctl_min(len_to_go, 131072); CTL_SIZE_8B(local_dma_sglist[i].len, local_sglist[i].len); local_sglist[i].addr = malloc(local_dma_sglist[i].len, M_CTL,M_WAITOK); local_dma_sglist[i].addr = local_sglist[i].addr; if (local_sglist[i].addr == NULL) { int j; printf("malloc failed for %zd bytes!", local_dma_sglist[i].len); for (j = 0; j < i; j++) { free(local_sglist[j].addr, M_CTL); } ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ 4857); retval = 1; goto bailout_error; } /* XXX KDM do we need a sync here? */ 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, %d (DMA: %d)\n", __func__, i, local_sglist[i].addr, local_sglist[i].len, local_dma_sglist[i].len); #endif } return (retval); bailout_error: ctl_send_datamove_done(io, /*have_lock*/ 0); 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; struct ctl_sg_entry *remote_dma_sglist, *local_dma_sglist; uint32_t local_used, remote_used, total_used; int retval; int i, j; retval = 0; 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)) ctl_set_busy(&io->scsiio); if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) { 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); retval = 1; goto bailout; } local_sglist = io->io_hdr.local_sglist; local_dma_sglist = io->io_hdr.local_dma_sglist; remote_sglist = io->io_hdr.remote_sglist; remote_dma_sglist = io->io_hdr.remote_dma_sglist; local_used = 0; remote_used = 0; total_used = 0; if (io->io_hdr.flags & CTL_FLAG_REDIR_DONE) { rq->ret = CTL_HA_STATUS_SUCCESS; rq->context = io; callback(rq); goto bailout; } /* * 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. */ for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) { int isc_ret; uint32_t cur_len, dma_length; uint8_t *tmp_ptr; rq->id = CTL_HA_DATA_CTL; 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 = ctl_min(local_sglist[i].len - local_used, remote_sglist[j].len - remote_used); /* * In this case, we have a size issue and need to decrease * the size, except in the case where we actually have less * than 8 bytes left. In that case, we need to increase * the DMA length to get the last bit. */ if ((cur_len & 0x7) != 0) { if (cur_len > 0x7) { cur_len = cur_len - (cur_len & 0x7); dma_length = cur_len; } else { CTL_SIZE_8B(dma_length, cur_len); } } else dma_length = cur_len; /* * If we had to allocate memory for this I/O, instead of using * the non-cached mirror memory, we'll need to flush the cache * before trying to DMA to the other controller. * * We could end up doing this multiple times for the same * segment if we have a larger local segment than remote * segment. That shouldn't be an issue. */ if ((io->io_hdr.flags & CTL_FLAG_NO_DATASYNC) == 0) { /* * XXX KDM use bus_dmamap_sync() here. */ } rq->size = dma_length; tmp_ptr = (uint8_t *)local_sglist[i].addr; tmp_ptr += local_used; /* Use physical addresses when talking to ISC hardware */ if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) { /* XXX KDM use busdma */ #if 0 rq->local = vtophys(tmp_ptr); #endif } else rq->local = tmp_ptr; 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 ((rq->size & 0x7) != 0) { printf("%s: warning: size %d is not on 8b boundary\n", __func__, rq->size); } if (((uintptr_t)rq->local & 0x7) != 0) { printf("%s: warning: local %p not on 8b boundary\n", __func__, rq->local); } if (((uintptr_t)rq->remote & 0x7) != 0) { printf("%s: warning: remote %p not on 8b boundary\n", __func__, rq->local); } #if 0 printf("%s: %s: local %#x remote %#x 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_WAIT) continue; if (isc_ret == CTL_HA_STATUS_DISCONNECT) { rq->ret = CTL_HA_STATUS_SUCCESS; } else { rq->ret = isc_ret; } callback(rq); goto bailout; } bailout: return (retval); } static void ctl_datamove_remote_read(union ctl_io *io) { int retval; int 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) && ((io->io_hdr.flags & CTL_FLAG_AUTO_MIRROR) == 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); } return; } /* * 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) { struct ctl_softc *softc; softc = control_softc; mtx_assert(&softc->ctl_lock, MA_NOTOWNED); /* * 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 (%d:%d:%d:%d) aborted\n", __func__, io->scsiio.tag_num, io->io_hdr.nexus.initid.id, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_target.id, 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 { union ctl_ha_msg msg; struct scsi_sense_data *sense; uint8_t sks[3]; int retry_count; memset(&msg, 0, sizeof(msg)); msg.hdr.msg_type = CTL_MSG_BAD_JUJU; msg.hdr.status = CTL_SCSI_ERROR; msg.scsi.scsi_status = SCSI_STATUS_CHECK_COND; retry_count = 4243; sense = &msg.scsi.sense_data; sks[0] = SSD_SCS_VALID; sks[1] = (retry_count >> 8) & 0xff; sks[2] = retry_count & 0xff; /* "Internal target failure" */ scsi_set_sense_data(sense, /*sense_format*/ SSD_TYPE_NONE, /*current_error*/ 1, /*sense_key*/ SSD_KEY_HARDWARE_ERROR, /*asc*/ 0x44, /*ascq*/ 0x00, /*type*/ SSD_ELEM_SKS, /*size*/ sizeof(sks), /*data*/ sks, SSD_ELEM_NONE); io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE; if (io->io_hdr.flags & CTL_FLAG_FAILOVER) { ctl_failover_io(io, /*have_lock*/ 1); return; } if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) { /* XXX KDM what to do if this fails? */ } return; } } static int ctl_process_done(union ctl_io *io) { struct ctl_lun *lun; struct ctl_softc *ctl_softc = control_softc; void (*fe_done)(union ctl_io *io); uint32_t targ_port = ctl_port_idx(io->io_hdr.nexus.targ_port); CTL_DEBUG_PRINT(("ctl_process_done\n")); fe_done = control_softc->ctl_ports[targ_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: printf("Invalid CTL I/O type %d\n", io->io_hdr.io_type); panic("Invalid CTL I/O type %d\n", io->io_hdr.io_type); break; } 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 (bootverbose || (ctl_debug & CTL_DEBUG_INFO)) ctl_io_error_print(io, NULL); if (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) ctl_free_io(io); else fe_done(io); return (CTL_RETVAL_COMPLETE); default: panic("ctl_process_done: invalid io type %d\n", io->io_hdr.io_type); break; /* NOTREACHED */ } lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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 errors to inject here. We only * inject errors for commands that don't already have errors set. */ - if ((STAILQ_FIRST(&lun->error_list) != NULL) - && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) + if ((STAILQ_FIRST(&lun->error_list) != NULL) && + ((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) { #ifdef CTL_TIME_IO struct bintime cur_bt; #endif int type; 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; lun->stats.ports[targ_port].bytes[type] += io->scsiio.kern_total_len; lun->stats.ports[targ_port].operations[type]++; #ifdef CTL_TIME_IO bintime_add(&lun->stats.ports[targ_port].dma_time[type], &io->io_hdr.dma_bt); lun->stats.ports[targ_port].num_dmas[type] += io->io_hdr.num_dmas; getbintime(&cur_bt); bintime_sub(&cur_bt, &io->io_hdr.start_bt); bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt); #endif } /* * Remove this from the OOA queue. */ TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links); /* * 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); mtx_lock(&ctl_softc->ctl_lock); ctl_free_lun(lun); mtx_unlock(&ctl_softc->ctl_lock); } 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. * We don't print UAs unless debugging was enabled explicitly. */ do { if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) break; if (!bootverbose && (ctl_debug & CTL_DEBUG_INFO) == 0) break; if ((ctl_debug & CTL_DEBUG_INFO) == 0 && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SCSI_ERROR) && (io->scsiio.scsi_status == SCSI_STATUS_CHECK_COND)) { int error_code, sense_key, asc, ascq; scsi_extract_sense_len(&io->scsiio.sense_data, io->scsiio.sense_len, &error_code, &sense_key, &asc, &ascq, /*show_errors*/ 0); if (sense_key == SSD_KEY_UNIT_ATTENTION) break; } ctl_io_error_print(io, NULL); } while (0); /* * 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. * * We only send status to the other controller if we're in XFER * mode. In SER_ONLY mode, the I/O is done on the controller that * received the I/O (from CTL's perspective), and so the status is * generated there. * * XXX KDM if we hold the lock here, we could cause a deadlock * if the frontend comes back in in this context to queue * something. */ if ((ctl_softc->ha_mode == CTL_HA_MODE_XFER) && (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) { union ctl_ha_msg msg; 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; msg.scsi.sense_residual = io->scsiio.sense_residual; msg.scsi.residual = io->scsiio.residual; memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data, sizeof(io->scsiio.sense_data)); /* * We copy this whether or not this is an I/O-related * command. Otherwise, we'd have to go and check to see * whether it's a read/write command, and it really isn't * worth it. */ memcpy(&msg.scsi.lbalen, &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN].bytes, sizeof(msg.scsi.lbalen)); if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg), 0) > CTL_HA_STATUS_SUCCESS) { /* XXX do something here */ } ctl_free_io(io); } else fe_done(io); return (CTL_RETVAL_COMPLETE); } #ifdef CTL_WITH_CA /* * 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_lun *lun; struct ctl_softc *ctl_softc; uint32_t initidx, targ_lun; ctl_softc = control_softc; CTL_DEBUG_PRINT(("ctl_queue_sense\n")); /* * 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. */ mtx_lock(&ctl_softc->ctl_lock); /* * If we don't have a LUN for this, just toss the sense * information. */ targ_lun = io->io_hdr.nexus.targ_lun; targ_lun = ctl_map_lun(io->io_hdr.nexus.targ_port, targ_lun); if ((targ_lun < CTL_MAX_LUNS) && (ctl_softc->ctl_luns[targ_lun] != NULL)) lun = ctl_softc->ctl_luns[targ_lun]; else goto bailout; initidx = ctl_get_initindex(&io->io_hdr.nexus); mtx_lock(&lun->lun_lock); /* * Already have CA set for this LUN...toss the sense information. */ if (ctl_is_set(lun->have_ca, initidx)) { mtx_unlock(&lun->lun_lock); goto bailout; } memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data, ctl_min(sizeof(lun->pending_sense[initidx]), sizeof(io->scsiio.sense_data))); ctl_set_mask(lun->have_ca, initidx); mtx_unlock(&lun->lun_lock); bailout: mtx_unlock(&ctl_softc->ctl_lock); ctl_free_io(io); return (CTL_RETVAL_COMPLETE); } #endif /* * 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_softc *ctl_softc; CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0])); ctl_softc = control_softc; #ifdef CTL_TIME_IO io->io_hdr.start_time = time_uptime; getbintime(&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_map_lun(io->io_hdr.nexus.targ_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_done(union ctl_io *io) { struct ctl_softc *ctl_softc; ctl_softc = control_softc; /* * 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: " "%d:%d:%d:%d 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.id, io->io_hdr.nexus.targ_port, io->io_hdr.nexus.targ_target.id, 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; /* * We need to send a msg to the serializing shelf to finish the IO * as well. We don't send a finish message to the other shelf if * this is a task management command. Task management commands * aren't serialized in the OOA queue, but rather just executed on * both shelf controllers for commands that originated on that * controller. */ if ((io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC) && (io->io_hdr.io_type != CTL_IO_TASK)) { union ctl_ha_msg msg_io; msg_io.hdr.msg_type = CTL_MSG_FINISH_IO; msg_io.hdr.serializing_sc = io->io_hdr.serializing_sc; if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_io, sizeof(msg_io), 0 ) != CTL_HA_STATUS_SUCCESS) { } /* continue on to finish IO */ } #ifdef CTL_IO_DELAY if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) { struct ctl_lun *lun; lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE; } else { struct ctl_lun *lun; lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; if ((lun != NULL) && (lun->delay_info.done_delay > 0)) { struct callout *callout; callout = (struct callout *)&io->io_hdr.timer_bytes; callout_init(callout, /*mpsafe*/ 1); io->io_hdr.flags |= CTL_FLAG_DELAY_DONE; callout_reset(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); } int ctl_isc(struct ctl_scsiio *ctsio) { struct ctl_lun *lun; int retval; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; CTL_DEBUG_PRINT(("ctl_isc: command: %02x\n", ctsio->cdb[0])); CTL_DEBUG_PRINT(("ctl_isc: calling data_submit()\n")); retval = lun->backend->data_submit((union ctl_io *)ctsio); return (retval); } 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")); for (;;) { retval = 0; /* * We handle the queues in this order: * - ISC * - done queue (to free up resources, unblock other commands) * - RtR queue * - incoming 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); retval = 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; } if (!ctl_pause_rtr) { 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); } } static void ctl_lun_thread(void *arg) { struct ctl_softc *softc = (struct ctl_softc *)arg; struct ctl_be_lun *be_lun; int retval; CTL_DEBUG_PRINT(("ctl_lun_thread starting\n")); for (;;) { retval = 0; 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); } } static void ctl_thresh_thread(void *arg) { struct ctl_softc *softc = (struct ctl_softc *)arg; struct ctl_lun *lun; struct ctl_be_lun *be_lun; struct scsi_da_rw_recovery_page *rwpage; struct ctl_logical_block_provisioning_page *page; const char *attr; uint64_t thres, val; int i, e; CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n")); for (;;) { mtx_lock(&softc->ctl_lock); STAILQ_FOREACH(lun, &softc->lun_list, links) { be_lun = lun->be_lun; if ((lun->flags & CTL_LUN_DISABLED) || (lun->flags & CTL_LUN_OFFLINE) || (be_lun->flags & CTL_LUN_FLAG_UNMAP) == 0 || lun->backend->lun_attr == NULL) continue; rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT]; if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0) continue; e = 0; page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT]; 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); } mtx_lock(&lun->lun_lock); if (e) { if (lun->lasttpt == 0 || time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) { lun->lasttpt = time_uptime; for (i = 0; i < CTL_MAX_INITIATORS; i++) lun->pending_ua[i] |= CTL_UA_THIN_PROV_THRES; } } else { lun->lasttpt = 0; for (i = 0; i < CTL_MAX_INITIATORS; i++) lun->pending_ua[i] &= ~CTL_UA_THIN_PROV_THRES; } mtx_unlock(&lun->lun_lock); } mtx_unlock(&softc->ctl_lock); pause("-", CTL_LBP_PERIOD * hz); } } static void ctl_enqueue_incoming(union ctl_io *io) { struct ctl_softc *softc = control_softc; struct ctl_thread *thr; u_int idx; idx = (io->io_hdr.nexus.targ_port * 127 + io->io_hdr.nexus.initid.id) % 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 = control_softc; 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 = control_softc; 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 = control_softc; 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); } /* Initialization and failover */ void ctl_init_isc_msg(void) { printf("CTL: Still calling this thing\n"); } /* * Init component * Initializes component into configuration defined by bootMode * (see hasc-sv.c) * returns hasc_Status: * OK * ERROR - fatal error */ static ctl_ha_comp_status ctl_isc_init(struct ctl_ha_component *c) { ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK; c->status = ret; return ret; } /* Start component * Starts component in state requested. If component starts successfully, * it must set its own state to the requestrd state * When requested state is HASC_STATE_HA, the component may refine it * by adding _SLAVE or _MASTER flags. * Currently allowed state transitions are: * UNKNOWN->HA - initial startup * UNKNOWN->SINGLE - initial startup when no parter detected * HA->SINGLE - failover * returns ctl_ha_comp_status: * OK - component successfully started in requested state * FAILED - could not start the requested state, failover may * be possible * ERROR - fatal error detected, no future startup possible */ static ctl_ha_comp_status ctl_isc_start(struct ctl_ha_component *c, ctl_ha_state state) { ctl_ha_comp_status ret = CTL_HA_COMP_STATUS_OK; printf("%s: go\n", __func__); // UNKNOWN->HA or UNKNOWN->SINGLE (bootstrap) if (c->state == CTL_HA_STATE_UNKNOWN ) { control_softc->is_single = 0; if (ctl_ha_msg_create(CTL_HA_CHAN_CTL, ctl_isc_event_handler) != CTL_HA_STATUS_SUCCESS) { printf("ctl_isc_start: ctl_ha_msg_create failed.\n"); ret = CTL_HA_COMP_STATUS_ERROR; } } else if (CTL_HA_STATE_IS_HA(c->state) && CTL_HA_STATE_IS_SINGLE(state)){ // HA->SINGLE transition ctl_failover(); control_softc->is_single = 1; } else { printf("ctl_isc_start:Invalid state transition %X->%X\n", c->state, state); ret = CTL_HA_COMP_STATUS_ERROR; } if (CTL_HA_STATE_IS_SINGLE(state)) control_softc->is_single = 1; c->state = state; c->status = ret; return ret; } /* * Quiesce component * The component must clear any error conditions (set status to OK) and * prepare itself to another Start call * returns ctl_ha_comp_status: * OK * ERROR */ static ctl_ha_comp_status ctl_isc_quiesce(struct ctl_ha_component *c) { int ret = CTL_HA_COMP_STATUS_OK; ctl_pause_rtr = 1; c->status = ret; return ret; } struct ctl_ha_component ctl_ha_component_ctlisc = { .name = "CTL ISC", .state = CTL_HA_STATE_UNKNOWN, .init = ctl_isc_init, .start = ctl_isc_start, .quiesce = ctl_isc_quiesce }; /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_backend_block.c =================================================================== --- head/sys/cam/ctl/ctl_backend_block.c (revision 275057) +++ head/sys/cam/ctl/ctl_backend_block.c (revision 275058) @@ -1,2637 +1,2646 @@ /*- * Copyright (c) 2003 Silicon Graphics International Corp. * Copyright (c) 2009-2011 Spectra Logic Corporation * Copyright (c) 2012 The FreeBSD Foundation * 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 /* * 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_MULTI_THREAD = 0x08 } 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_devdata { struct cdev *cdev; struct cdevsw *csw; int dev_ref; }; struct ctl_be_block_filedata { struct ucred *cred; }; union ctl_be_block_bedata { struct ctl_be_block_devdata dev; 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; struct ctl_block_disk *disk; 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_getattr_t getattr; uma_zone_t lun_zone; uint64_t size_blocks; uint64_t size_bytes; uint32_t blocksize; int blocksize_shift; uint16_t pblockexp; uint16_t pblockoff; 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 ctl_be_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_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; int num_disks; STAILQ_HEAD(, ctl_block_disk) disk_list; 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 num_errors; struct bintime ds_t0; devstat_tag_type ds_tag_type; devstat_trans_flags ds_trans_type; uint64_t io_len; uint64_t io_offset; struct ctl_be_block_softc *softc; struct ctl_be_block_lun *lun; void (*beio_cont)(struct ctl_be_block_io *beio); /* to continue processing */ }; 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_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_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_softc *softc, 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_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req); static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, 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); int ctl_be_block_init(void); static struct ctl_backend_driver ctl_be_block_driver = { .name = "block", .flags = CTL_BE_FLAG_HAS_CONFIG, .init = ctl_be_block_init, .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 uma_zone_t beio_zone; static struct ctl_be_block_io * ctl_alloc_beio(struct ctl_be_block_softc *softc) { struct ctl_be_block_io *beio; beio = uma_zalloc(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_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 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 int i; beio = (struct ctl_be_block_io *)PRIV(io)->ptr; be_lun = beio->lun; DPRINTF("entered\n"); #ifdef CTL_TIME_IO getbintime(&cur_bt); bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); bintime_add(&io->io_hdr.dma_bt, &cur_bt); io->io_hdr.num_dmas++; #endif 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.port_status == 0) && - ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) && + 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)) { 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. */ for (i = 0; i < beio->num_segs; i++) { if (memcmp(beio->sg_segs[i].addr, beio->sg_segs[i + CTLBLK_HALF_SEGS].addr, beio->sg_segs[i].len) != 0) break; } if (i < beio->num_segs) ctl_set_sense(&io->scsiio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_MISCOMPARE, /*asc*/ 0x1D, /*ascq*/ 0x00, SSD_ELEM_NONE); else ctl_set_success(&io->scsiio); } - } - else if ((io->io_hdr.port_status != 0) - && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) - && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { + } 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)) { /* * For hardware error sense keys, the sense key * specific value is defined to be a retry count, * but we use it to pass back an internal FETD * error code. XXX KDM Hopefully the FETD is only * using 16 bits for an error code, since that's * all the space we have in the sks field. */ ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ io->io_hdr.port_status); } /* * 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); /* * XXX KDM make sure that links is okay to use at this point. * Otherwise, we either need to add another field to ctl_io_hdr, * or deal with resource allocation here. */ 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->num_errors++; 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. */ if (beio->num_errors > 0) { if (error == EOPNOTSUPP) { ctl_set_invalid_opcode(&io->scsiio); } else if (error == ENOSPC) { ctl_set_space_alloc_fail(&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); 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); #ifdef CTL_TIME_IO getbintime(&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, 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, kernel, read, file_start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, kernel, write, file_start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, kernel, read, file_done,"uint64_t"); SDT_PROBE_DEFINE1(cbb, kernel, 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; int flags; int error, 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_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); xuio.uio_rw = UIO_READ; } else { SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 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. * * So, to attempt to provide some barrier semantics in the * BIO_ORDERED case, set both IO_DIRECT and IO_SYNC. */ error = VOP_READ(be_lun->vn, &xuio, flags, file_data->cred); VOP_UNLOCK(be_lun->vn, 0); SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 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. * * So if we've got the BIO_ORDERED flag set, we want * IO_SYNC in either the UFS or ZFS case. */ error = VOP_WRITE(be_lun->vn, &xuio, flags, file_data->cred); VOP_UNLOCK(be_lun->vn, 0); vn_finished_write(mountpoint); SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); } 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) { char path_str[32]; ctl_scsi_path_string(io, path_str, sizeof(path_str)); printf("%s%s command returned errno %d\n", path_str, (beio->bio_cmd == BIO_READ) ? "READ" : "WRITE", error); if (error == ENOSPC) { ctl_set_space_alloc_fail(&io->scsiio); } else ctl_set_medium_error(&io->scsiio); 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); #ifdef CTL_TIME_IO getbintime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } } static void ctl_be_block_dispatch_zvol(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { struct ctl_be_block_devdata *dev_data; union ctl_io *io; struct uio xuio; struct iovec *xiovec; int flags; int error, i; DPRINTF("entered\n"); dev_data = &be_lun->backend.dev; 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_PROBE(cbb, kernel, read, file_start, 0, 0, 0, 0, 0); xuio.uio_rw = UIO_READ; } else { SDT_PROBE(cbb, kernel, write, file_start, 0, 0, 0, 0, 0); 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) { error = (*dev_data->csw->d_read)(dev_data->cdev, &xuio, flags); SDT_PROBE(cbb, kernel, read, file_done, 0, 0, 0, 0, 0); } else { error = (*dev_data->csw->d_write)(dev_data->cdev, &xuio, flags); SDT_PROBE(cbb, kernel, write, file_done, 0, 0, 0, 0, 0); } 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) { ctl_set_space_alloc_fail(&io->scsiio); } else ctl_set_medium_error(&io->scsiio); 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); #ifdef CTL_TIME_IO getbintime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } } static void ctl_be_block_flush_dev(struct ctl_be_block_lun *be_lun, struct ctl_be_block_io *beio) { struct bio *bio; union ctl_io *io; struct ctl_be_block_devdata *dev_data; dev_data = &be_lun->backend.dev; io = beio->io; DPRINTF("entered\n"); /* This can't fail, it's a blocking allocation. */ bio = g_alloc_bio(); bio->bio_cmd = BIO_FLUSH; bio->bio_flags |= BIO_ORDERED; bio->bio_dev = dev_data->cdev; 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); (*dev_data->csw->d_strategy)(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; struct ctl_be_block_devdata *dev_data; uint64_t maxlen; dev_data = &be_lun->backend.dev; maxlen = LONG_MAX - (LONG_MAX % be_lun->blocksize); while (len > 0) { bio = g_alloc_bio(); bio->bio_cmd = BIO_DELETE; bio->bio_dev = dev_data->cdev; 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->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); (*dev_data->csw->d_strategy)(bio); } } 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_be_block_devdata *dev_data; struct ctl_ptr_len_flags *ptrlen; struct scsi_unmap_desc *buf, *end; uint64_t len; dev_data = &be_lun->backend.dev; 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->blocksize; beio->io_len += len; ctl_be_block_unmap_dev_range(be_lun, beio, scsi_8btou64(buf->lba) * be_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); int i; struct bio *bio; struct ctl_be_block_devdata *dev_data; off_t cur_offset; int max_iosize; DPRINTF("entered\n"); dev_data = &be_lun->backend.dev; /* * 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. */ max_iosize = dev_data->cdev->si_iosize_max; if (max_iosize < PAGE_SIZE) 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_data->cdev; 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->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); (*dev_data->csw->d_strategy)(bio); } } static uint64_t ctl_be_block_getattr_dev(struct ctl_be_block_lun *be_lun, const char *attrname) { struct ctl_be_block_devdata *dev_data = &be_lun->backend.dev; struct diocgattr_arg arg; int error; if (dev_data->csw == NULL || dev_data->csw->d_ioctl == NULL) return (UINT64_MAX); strlcpy(arg.name, attrname, sizeof(arg.name)); arg.len = sizeof(arg.value.off); error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGATTR, (caddr_t)&arg, FREAD, curthread); if (error != 0) return (UINT64_MAX); return (arg.value.off); } 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_block_io *beio; struct ctl_be_block_softc *softc; struct ctl_lba_len_flags *lbalen; uint64_t len_left, lba; int i, seglen; uint8_t *buf, *end; DPRINTF("entered\n"); beio = (struct ctl_be_block_io *)PRIV(io)->ptr; softc = be_lun->softc; 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; } 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 & (SWS_UNMAP | SWS_ANCHOR)) { beio->io_offset = lbalen->lba * be_lun->blocksize; beio->io_len = (uint64_t)lbalen->len * be_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); len_left = (uint64_t)lbalen->len * be_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); seglen -= seglen % be_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 += be_lun->blocksize) { memcpy(buf, io->scsiio.kern_data_ptr, be_lun->blocksize); if (lbalen->flags & SWS_LBDATA) scsi_ulto4b(lbalen->lba + lba, buf); lba++; } } beio->io_offset = lbalen->lba * be_lun->blocksize; beio->io_len = lba * be_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_be_block_softc *softc; struct ctl_ptr_len_flags *ptrlen; DPRINTF("entered\n"); beio = (struct ctl_be_block_io *)PRIV(io)->ptr; softc = be_lun->softc; 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; } 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; } 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_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; PRIV(io)->ptr = (void *)beio; switch (io->scsiio.cdb[0]) { case SYNCHRONIZE_CACHE: case SYNCHRONIZE_CACHE_16: beio->bio_cmd = BIO_FLUSH; beio->ds_trans_type = DEVSTAT_NO_DATA; beio->ds_tag_type = DEVSTAT_TAG_ORDERED; beio->io_len = 0; be_lun->lun_flush(be_lun, beio); 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, kernel, read, start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, kernel, write, start, "uint64_t"); SDT_PROBE_DEFINE1(cbb, kernel, read, alloc_done, "uint64_t"); SDT_PROBE_DEFINE1(cbb, kernel, 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); /* * XXX KDM make sure that links is okay to use at this point. * Otherwise, we either need to add another field to ctl_io_hdr, * or deal with resource allocation here. */ 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_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_PROBE(cbb, kernel, write, start, 0, 0, 0, 0, 0); } else { SDT_PROBE(cbb, kernel, read, start, 0, 0, 0, 0, 0); } 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 / be_lun->blocksize); beio->io_offset = (lbalen->lba + bptrlen->len) * be_lun->blocksize; beio->io_len = lbas * be_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_data_resid = 0; io->scsiio.kern_sg_entries = beio->num_segs; io->io_hdr.flags |= CTL_FLAG_ALLOCATED | CTL_FLAG_KDPTR_SGLIST; /* * 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_PROBE(cbb, kernel, read, alloc_done, 0, 0, 0, 0, 0); be_lun->dispatch(be_lun, beio); } else { SDT_PROBE(cbb, kernel, write, alloc_done, 0, 0, 0, 0, 0); #ifdef CTL_TIME_IO getbintime(&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_softc *softc; union ctl_io *io; be_lun = (struct ctl_be_block_lun *)context; softc = be_lun->softc; DPRINTF("entered\n"); mtx_lock(&be_lun->queue_lock); for (;;) { io = (union ctl_io *)STAILQ_FIRST(&be_lun->datamove_queue); if (io != NULL) { struct ctl_be_block_io *beio; 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; be_lun->dispatch(be_lun, beio); mtx_lock(&be_lun->queue_lock); 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); ctl_be_block_cw_dispatch(be_lun, io); mtx_lock(&be_lun->queue_lock); 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); /* * We must drop the lock, since this routine and * its children may sleep. */ ctl_be_block_dispatch(be_lun, 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); } /* * 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 *ctl_be_lun; DPRINTF("entered\n"); ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ CTL_PRIV_BACKEND_LUN].ptr; be_lun = (struct ctl_be_block_lun *)ctl_be_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); /* * XXX KDM make sure that links is okay to use at this point. * Otherwise, we either need to add another field to ctl_io_hdr, * or deal with resource allocation here. */ 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_block_filedata *file_data; struct ctl_lun_create_params *params; struct vattr vattr; off_t pss; int error; error = 0; 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; 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); } /* * Verify that we have the ability to upgrade to exclusive * access on this file so we can trap errors at open instead * of reporting them during first access. */ if (VOP_ISLOCKED(be_lun->vn) != LK_EXCLUSIVE) { vn_lock(be_lun->vn, LK_UPGRADE | LK_RETRY); if (be_lun->vn->v_iflag & VI_DOOMED) { error = EBADF; snprintf(req->error_str, sizeof(req->error_str), "error locking 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; /* * We set the multi thread flag for file operations because all * filesystems (in theory) are capable of allowing multiple readers * of a file at once. So we want to get the maximum possible * concurrency. */ be_lun->flags |= CTL_BE_BLOCK_LUN_MULTI_THREAD; /* * 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) be_lun->blocksize = params->blocksize_bytes; else be_lun->blocksize = 512; pss = vattr.va_blocksize / be_lun->blocksize; if ((pss > 0) && (pss * be_lun->blocksize == vattr.va_blocksize) && ((pss & (pss - 1)) == 0)) { be_lun->pblockexp = fls(pss) - 1; be_lun->pblockoff = 0; } /* * Sanity check. The media size has to be at least one * sector long. */ if (be_lun->size_bytes < be_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, be_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_lun_create_params *params; struct vattr vattr; struct cdev *dev; struct cdevsw *devsw; int error; off_t ps, pss, po, pos; params = &be_lun->params; be_lun->dev_type = CTL_BE_BLOCK_DEV; be_lun->backend.dev.cdev = be_lun->vn->v_rdev; be_lun->backend.dev.csw = dev_refthread(be_lun->backend.dev.cdev, &be_lun->backend.dev.dev_ref); if (be_lun->backend.dev.csw == NULL) panic("Unable to retrieve device switch"); if (strcmp(be_lun->backend.dev.csw->d_name, "zvol") == 0) be_lun->dispatch = ctl_be_block_dispatch_zvol; else be_lun->dispatch = ctl_be_block_dispatch_dev; be_lun->lun_flush = ctl_be_block_flush_dev; be_lun->unmap = ctl_be_block_unmap_dev; be_lun->getattr = ctl_be_block_getattr_dev; error = VOP_GETATTR(be_lun->vn, &vattr, NOCRED); if (error) { snprintf(req->error_str, sizeof(req->error_str), "error getting vnode attributes for device %s", be_lun->dev_path); return (error); } dev = be_lun->vn->v_rdev; devsw = dev->si_devsw; if (!devsw->d_ioctl) { snprintf(req->error_str, sizeof(req->error_str), "no d_ioctl for device %s!", be_lun->dev_path); return (ENODEV); } error = devsw->d_ioctl(dev, DIOCGSECTORSIZE, (caddr_t)&be_lun->blocksize, FREAD, curthread); if (error) { 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 > be_lun->blocksize)) { uint32_t bs_multiple, tmp_blocksize; bs_multiple = params->blocksize_bytes / be_lun->blocksize; tmp_blocksize = bs_multiple * be_lun->blocksize; if (tmp_blocksize == params->blocksize_bytes) { be_lun->blocksize = params->blocksize_bytes; } else { snprintf(req->error_str, sizeof(req->error_str), "requested blocksize %u is not an even " "multiple of backing device blocksize %u", params->blocksize_bytes, be_lun->blocksize); return (EINVAL); } } else if ((params->blocksize_bytes != 0) && (params->blocksize_bytes != be_lun->blocksize)) { snprintf(req->error_str, sizeof(req->error_str), "requested blocksize %u < backing device " "blocksize %u", params->blocksize_bytes, be_lun->blocksize); return (EINVAL); } error = devsw->d_ioctl(dev, DIOCGMEDIASIZE, (caddr_t)&be_lun->size_bytes, FREAD, curthread); if (error) { 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 > be_lun->size_bytes) { snprintf(req->error_str, sizeof(req->error_str), "requested LUN size %ju > backing device " "size %ju", (uintmax_t)params->lun_size_bytes, (uintmax_t)be_lun->size_bytes); return (EINVAL); } be_lun->size_bytes = params->lun_size_bytes; } error = devsw->d_ioctl(dev, DIOCGSTRIPESIZE, (caddr_t)&ps, FREAD, curthread); if (error) ps = po = 0; else { error = devsw->d_ioctl(dev, DIOCGSTRIPEOFFSET, (caddr_t)&po, FREAD, curthread); if (error) po = 0; } pss = ps / be_lun->blocksize; pos = po / be_lun->blocksize; if ((pss > 0) && (pss * be_lun->blocksize == ps) && (pss >= pos) && ((pss & (pss - 1)) == 0) && (pos * be_lun->blocksize == po)) { be_lun->pblockexp = fls(pss) - 1; be_lun->pblockoff = (pss - pos) % pss; } return (0); } static int ctl_be_block_close(struct ctl_be_block_lun *be_lun) { DROP_GIANT(); if (be_lun->vn) { int flags = FREAD | FWRITE; switch (be_lun->dev_type) { case CTL_BE_BLOCK_DEV: if (be_lun->backend.dev.csw) { dev_relthread(be_lun->backend.dev.cdev, be_lun->backend.dev.dev_ref); be_lun->backend.dev.csw = NULL; be_lun->backend.dev.cdev = NULL; } break; case CTL_BE_BLOCK_FILE: break; case CTL_BE_BLOCK_NONE: break; default: panic("Unexpected backend type."); break; } (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."); break; } be_lun->dev_type = CTL_BE_BLOCK_NONE; } PICKUP_GIANT(); return (0); } static int ctl_be_block_open(struct ctl_be_block_softc *softc, struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) { struct nameidata nd; int flags; int error; /* * XXX KDM allow a read-only option? */ flags = FREAD | FWRITE; error = 0; if (rootvnode == NULL) { snprintf(req->error_str, sizeof(req->error_str), "Root filesystem is not mounted"); return (1); } if (!curthread->td_proc->p_fd->fd_cdir) { curthread->td_proc->p_fd->fd_cdir = rootvnode; VREF(rootvnode); } if (!curthread->td_proc->p_fd->fd_rdir) { curthread->td_proc->p_fd->fd_rdir = rootvnode; VREF(rootvnode); } if (!curthread->td_proc->p_fd->fd_jdir) { curthread->td_proc->p_fd->fd_jdir = rootvnode; VREF(rootvnode); } again: NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, be_lun->dev_path, curthread); error = vn_open(&nd, &flags, 0, NULL); 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_path = "/dev/"; char *dev_name; /* Try adding device path at beginning of name */ dev_name = malloc(strlen(be_lun->dev_path) + strlen(dev_path) + 1, M_CTLBLK, M_WAITOK); if (dev_name) { sprintf(dev_name, "%s%s", dev_path, 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); } 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); return (error); } be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; return (0); } static int ctl_be_block_create(struct ctl_be_block_softc *softc, struct ctl_lun_req *req) { struct ctl_be_block_lun *be_lun; struct ctl_lun_create_params *params; char num_thread_str[16]; char tmpstr[32]; char *value; int retval, num_threads, unmap; int tmp_num_threads; params = &req->reqdata.create; retval = 0; req->status = CTL_LUN_OK; num_threads = cbb_num_threads; be_lun = malloc(sizeof(*be_lun), M_CTLBLK, M_ZERO | M_WAITOK); be_lun->params = req->reqdata.create; be_lun->softc = softc; STAILQ_INIT(&be_lun->input_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(&be_lun->ctl_be_lun.options, req->num_be_args, req->kern_be_args); 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) be_lun->ctl_be_lun.lun_type = params->device_type; else be_lun->ctl_be_lun.lun_type = T_DIRECT; if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { value = ctl_get_opt(&be_lun->ctl_be_lun.options, "file"); if (value == NULL) { snprintf(req->error_str, sizeof(req->error_str), "no file argument specified"); goto bailout_error; } be_lun->dev_path = strdup(value, M_CTLBLK); be_lun->blocksize = 512; be_lun->blocksize_shift = fls(be_lun->blocksize) - 1; retval = ctl_be_block_open(softc, be_lun, req); if (retval != 0) { retval = 0; req->status = CTL_LUN_WARNING; } } else { /* * For processor devices, we don't have any size. */ be_lun->blocksize = 0; be_lun->pblockexp = 0; be_lun->pblockoff = 0; be_lun->size_blocks = 0; be_lun->size_bytes = 0; be_lun->ctl_be_lun.maxlba = 0; /* * Default to just 1 thread for processor devices. */ num_threads = 1; } /* * XXX This searching loop might be refactored to be combined with * the loop above, */ value = ctl_get_opt(&be_lun->ctl_be_lun.options, "num_threads"); 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; } unmap = (be_lun->dispatch == ctl_be_block_dispatch_zvol); value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); if (value != NULL) unmap = (strcmp(value, "on") == 0); be_lun->flags = CTL_BE_BLOCK_LUN_UNCONFIGURED; be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; if (be_lun->vn == NULL) be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_OFFLINE; if (unmap) be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; be_lun->ctl_be_lun.be_lun = be_lun; be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 0 : (be_lun->size_blocks - 1); be_lun->ctl_be_lun.blocksize = be_lun->blocksize; be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; if (be_lun->dispatch == ctl_be_block_dispatch_zvol && be_lun->blocksize != 0) be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / be_lun->blocksize; /* Tell the user the blocksize we ended up using */ params->lun_size_bytes = be_lun->size_bytes; params->blocksize_bytes = be_lun->blocksize; if (params->flags & CTL_LUN_FLAG_ID_REQ) { be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; } else be_lun->ctl_be_lun.req_lun_id = 0; be_lun->ctl_be_lun.lun_shutdown = ctl_be_block_lun_shutdown; be_lun->ctl_be_lun.lun_config_status = ctl_be_block_lun_config_status; be_lun->ctl_be_lun.be = &ctl_be_block_driver; if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", softc->num_luns); strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), sizeof(tmpstr))); /* Tell the user what we used for a serial number */ strncpy((char *)params->serial_num, tmpstr, ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); } else { strncpy((char *)be_lun->ctl_be_lun.serial_num, params->serial_num, ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), sizeof(params->serial_num))); } if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, ctl_min(sizeof(be_lun->ctl_be_lun.device_id), sizeof(tmpstr))); /* Tell the user what we used for a device ID */ strncpy((char *)params->device_id, tmpstr, ctl_min(sizeof(params->device_id), sizeof(tmpstr))); } else { strncpy((char *)be_lun->ctl_be_lun.device_id, params->device_id, ctl_min(sizeof(be_lun->ctl_be_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->ctl_be_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 = be_lun->ctl_be_lun.lun_id; } mtx_unlock(&softc->lock); be_lun->disk_stats = devstat_new_entry("cbb", params->req_lun_id, be_lun->blocksize, DEVSTAT_ALL_SUPPORTED, be_lun->ctl_be_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(&be_lun->ctl_be_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; int retval; params = &req->reqdata.rm; mtx_lock(&softc->lock); be_lun = NULL; STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->ctl_be_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; } retval = ctl_disable_lun(&be_lun->ctl_be_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; } retval = ctl_invalidate_lun(&be_lun->ctl_be_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(be_lun->io_taskqueue, &be_lun->io_task); taskqueue_free(be_lun->io_taskqueue); ctl_be_block_close(be_lun); if (be_lun->disk_stats != NULL) devstat_remove_entry(be_lun->disk_stats); uma_zdestroy(be_lun->lun_zone); ctl_free_opts(&be_lun->ctl_be_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_file(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) { struct vattr vattr; int error; struct ctl_lun_create_params *params = &be_lun->params; if (params->lun_size_bytes != 0) { be_lun->size_bytes = params->lun_size_bytes; } else { vn_lock(be_lun->vn, LK_SHARED | LK_RETRY); error = VOP_GETATTR(be_lun->vn, &vattr, curthread->td_ucred); VOP_UNLOCK(be_lun->vn, 0); if (error != 0) { snprintf(req->error_str, sizeof(req->error_str), "error calling VOP_GETATTR() for file %s", be_lun->dev_path); return (error); } be_lun->size_bytes = vattr.va_size; } return (0); } static int ctl_be_block_modify_dev(struct ctl_be_block_lun *be_lun, struct ctl_lun_req *req) { struct ctl_be_block_devdata *dev_data; int error; struct ctl_lun_create_params *params = &be_lun->params; uint64_t size_bytes; dev_data = &be_lun->backend.dev; if (!dev_data->csw->d_ioctl) { snprintf(req->error_str, sizeof(req->error_str), "no d_ioctl for device %s!", be_lun->dev_path); return (ENODEV); } error = dev_data->csw->d_ioctl(dev_data->cdev, DIOCGMEDIASIZE, (caddr_t)&size_bytes, FREAD, curthread); if (error) { 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 > size_bytes) { snprintf(req->error_str, sizeof(req->error_str), "requested LUN size %ju > backing device " "size %ju", (uintmax_t)params->lun_size_bytes, (uintmax_t)size_bytes); return (EINVAL); } be_lun->size_bytes = params->lun_size_bytes; } else { be_lun->size_bytes = size_bytes; } 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; uint64_t oldsize; int error; params = &req->reqdata.modify; mtx_lock(&softc->lock); be_lun = NULL; STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->ctl_be_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; } be_lun->params.lun_size_bytes = params->lun_size_bytes; oldsize = be_lun->size_bytes; if (be_lun->vn == NULL) error = ctl_be_block_open(softc, be_lun, req); else if (be_lun->vn->v_type == VREG) error = ctl_be_block_modify_file(be_lun, req); else error = ctl_be_block_modify_dev(be_lun, req); if (error == 0 && be_lun->size_bytes != oldsize) { be_lun->size_blocks = be_lun->size_bytes >> be_lun->blocksize_shift; /* * The maximum LBA is the size - 1. * * XXX: Note that this field is being updated without locking, * which might cause problems on 32-bit architectures. */ be_lun->ctl_be_lun.maxlba = (be_lun->size_blocks == 0) ? 0 : (be_lun->size_blocks - 1); be_lun->ctl_be_lun.blocksize = be_lun->blocksize; be_lun->ctl_be_lun.pblockexp = be_lun->pblockexp; be_lun->ctl_be_lun.pblockoff = be_lun->pblockoff; if (be_lun->dispatch == ctl_be_block_dispatch_zvol && be_lun->blocksize != 0) be_lun->ctl_be_lun.atomicblock = CTLBLK_MAX_IO_SIZE / be_lun->blocksize; ctl_lun_capacity_changed(&be_lun->ctl_be_lun); if (oldsize == 0 && be_lun->size_blocks != 0) ctl_lun_online(&be_lun->ctl_be_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; struct ctl_be_block_softc *softc; lun = (struct ctl_be_block_lun *)be_lun; 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->ctl_be_lun) != 0) { printf("%s: ctl_enable_lun() failed!\n", __func__); if (ctl_invalidate_lun(&lun->ctl_be_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 *ctl_be_lun; int retval; retval = 0; DPRINTF("entered\n"); ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ CTL_PRIV_BACKEND_LUN].ptr; be_lun = (struct ctl_be_block_lun *)ctl_be_lun->be_lun; 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; cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; if (cdb->how & SSS_START) retval = ctl_start_lun(ctl_be_lun); else { retval = ctl_stop_lun(ctl_be_lun); /* * XXX KDM Copan-specific offline behavior. * Figure out a reasonable way to port this? */ #ifdef NEEDTOPORT if ((retval == 0) && (cdb->byte2 & SSS_ONOFFLINE)) retval = ctl_lun_offline(ctl_be_lun); #endif } /* * In general, the above routines should not fail. They * just set state for the LUN. So we've got something * pretty wrong here if we can't start or stop the LUN. */ if (retval != 0) { ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ 0xf051); retval = CTL_RETVAL_COMPLETE; } else { 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) { return (0); } 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 = 0; 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)); } int ctl_be_block_init(void) { struct ctl_be_block_softc *softc; int retval; softc = &backend_block_softc; retval = 0; mtx_init(&softc->lock, "ctlblock", NULL, MTX_DEF); beio_zone = uma_zcreate("beio", sizeof(struct ctl_be_block_io), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); STAILQ_INIT(&softc->disk_list); STAILQ_INIT(&softc->lun_list); return (retval); } Index: head/sys/cam/ctl/ctl_backend_ramdisk.c =================================================================== --- head/sys/cam/ctl/ctl_backend_ramdisk.c (revision 275057) +++ head/sys/cam/ctl/ctl_backend_ramdisk.c (revision 275058) @@ -1,973 +1,974 @@ /*- * Copyright (c) 2003, 2008 Silicon Graphics International Corp. * Copyright (c) 2012 The FreeBSD Foundation * 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 backend for a "fake" ramdisk. * * 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 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 { char lunname[32]; uint64_t size_bytes; uint64_t size_blocks; struct ctl_be_ramdisk_softc *softc; ctl_be_ramdisk_lun_flags flags; STAILQ_ENTRY(ctl_be_ramdisk_lun) links; struct ctl_be_lun ctl_be_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 rd_size; #ifdef CTL_RAMDISK_PAGES uint8_t **ramdisk_pages; int num_pages; #else uint8_t *ramdisk_buffer; #endif int num_luns; STAILQ_HEAD(, ctl_be_ramdisk_lun) lun_list; }; static struct ctl_be_ramdisk_softc rd_softc; int ctl_backend_ramdisk_init(void); void ctl_backend_ramdisk_shutdown(void); static int ctl_backend_ramdisk_move_done(union ctl_io *io); static int ctl_backend_ramdisk_submit(union ctl_io *io); static void ctl_backend_ramdisk_continue(union ctl_io *io); 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, int do_wait); static int ctl_backend_ramdisk_modify(struct ctl_be_ramdisk_softc *softc, struct ctl_lun_req *req); static void ctl_backend_ramdisk_worker(void *context, int pending); 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 int ctl_backend_ramdisk_config_write(union ctl_io *io); static int ctl_backend_ramdisk_config_read(union ctl_io *io); static struct ctl_backend_driver ctl_be_ramdisk_driver = { .name = "ramdisk", .flags = CTL_BE_FLAG_HAS_CONFIG, .init = ctl_backend_ramdisk_init, .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 }; MALLOC_DEFINE(M_RAMDISK, "ramdisk", "Memory used for CTL RAMdisk"); CTL_BACKEND_DECLARE(cbr, ctl_be_ramdisk_driver); int ctl_backend_ramdisk_init(void) { struct ctl_be_ramdisk_softc *softc; #ifdef CTL_RAMDISK_PAGES int i; #endif softc = &rd_softc; memset(softc, 0, sizeof(*softc)); mtx_init(&softc->lock, "ctlramdisk", NULL, MTX_DEF); STAILQ_INIT(&softc->lun_list); softc->rd_size = 1024 * 1024; #ifdef CTL_RAMDISK_PAGES softc->num_pages = softc->rd_size / PAGE_SIZE; softc->ramdisk_pages = (uint8_t **)malloc(sizeof(uint8_t *) * softc->num_pages, M_RAMDISK, M_WAITOK); for (i = 0; i < softc->num_pages; i++) softc->ramdisk_pages[i] = malloc(PAGE_SIZE, M_RAMDISK,M_WAITOK); #else softc->ramdisk_buffer = (uint8_t *)malloc(softc->rd_size, M_RAMDISK, M_WAITOK); #endif return (0); } void ctl_backend_ramdisk_shutdown(void) { struct ctl_be_ramdisk_softc *softc; struct ctl_be_ramdisk_lun *lun, *next_lun; #ifdef CTL_RAMDISK_PAGES int i; #endif softc = &rd_softc; mtx_lock(&softc->lock); for (lun = STAILQ_FIRST(&softc->lun_list); lun != NULL; lun = next_lun){ /* * Grab the next LUN. The current LUN may get removed by * ctl_invalidate_lun(), which will call our LUN shutdown * routine, if there is no outstanding I/O for this LUN. */ next_lun = STAILQ_NEXT(lun, links); /* * 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->ctl_be_lun); ctl_invalidate_lun(&lun->ctl_be_lun); mtx_lock(&softc->lock); } mtx_unlock(&softc->lock); #ifdef CTL_RAMDISK_PAGES for (i = 0; i < softc->num_pages; i++) free(softc->ramdisk_pages[i], M_RAMDISK); free(softc->ramdisk_pages, M_RAMDISK); #else free(softc->ramdisk_buffer, M_RAMDISK); #endif if (ctl_backend_deregister(&ctl_be_ramdisk_driver) != 0) { printf("ctl_backend_ramdisk_shutdown: " "ctl_backend_deregister() failed!\n"); } } static int ctl_backend_ramdisk_move_done(union ctl_io *io) { struct ctl_be_lun *ctl_be_lun; struct ctl_be_ramdisk_lun *be_lun; #ifdef CTL_TIME_IO struct bintime cur_bt; #endif CTL_DEBUG_PRINT(("ctl_backend_ramdisk_move_done\n")); ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ CTL_PRIV_BACKEND_LUN].ptr; be_lun = (struct ctl_be_ramdisk_lun *)ctl_be_lun->be_lun; #ifdef CTL_TIME_IO getbintime(&cur_bt); bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt); bintime_add(&io->io_hdr.dma_bt, &cur_bt); io->io_hdr.num_dmas++; #endif 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.port_status == 0) - && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) - && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)) { + 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)) { if (io->io_hdr.ctl_private[CTL_PRIV_BACKEND].integer > 0) { 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); - } else if ((io->io_hdr.port_status != 0) - && ((io->io_hdr.flags & CTL_FLAG_ABORT) == 0) - && ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE)){ + } 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)) { /* * For hardware error sense keys, the sense key * specific value is defined to be a retry count, * but we use it to pass back an internal FETD * error code. XXX KDM Hopefully the FETD is only * using 16 bits for an error code, since that's * all the space we have in the sks field. */ ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ io->io_hdr.port_status); } ctl_data_submit_done(io); return(0); } static int ctl_backend_ramdisk_submit(union ctl_io *io) { struct ctl_be_lun *ctl_be_lun; struct ctl_lba_len_flags *lbalen; ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ CTL_PRIV_BACKEND_LUN].ptr; lbalen = (struct ctl_lba_len_flags *)&io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN]; if (lbalen->flags & CTL_LLF_VERIFY) { ctl_set_success(&io->scsiio); ctl_data_submit_done(io); return (CTL_RETVAL_COMPLETE); } io->io_hdr.ctl_private[CTL_PRIV_BACKEND].integer = lbalen->len * ctl_be_lun->blocksize; ctl_backend_ramdisk_continue(io); return (CTL_RETVAL_COMPLETE); } static void ctl_backend_ramdisk_continue(union ctl_io *io) { struct ctl_be_ramdisk_softc *softc; int len, len_filled, sg_filled; #ifdef CTL_RAMDISK_PAGES struct ctl_sg_entry *sg_entries; int i; #endif softc = &rd_softc; len = io->io_hdr.ctl_private[CTL_PRIV_BACKEND].integer; #ifdef CTL_RAMDISK_PAGES sg_filled = min(btoc(len), softc->num_pages); if (sg_filled > 1) { io->scsiio.kern_data_ptr = malloc(sizeof(struct ctl_sg_entry) * sg_filled, M_RAMDISK, M_WAITOK); sg_entries = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; for (i = 0, len_filled = 0; i < sg_filled; i++) { sg_entries[i].addr = softc->ramdisk_pages[i]; sg_entries[i].len = ctl_min(PAGE_SIZE, len - len_filled); len_filled += sg_entries[i].len; } io->io_hdr.flags |= CTL_FLAG_KDPTR_SGLIST; } else { sg_filled = 0; len_filled = len; io->scsiio.kern_data_ptr = softc->ramdisk_pages[0]; } #else sg_filled = 0; len_filled = min(len, softc->rd_size); io->scsiio.kern_data_ptr = softc->ramdisk_buffer; #endif /* CTL_RAMDISK_PAGES */ io->scsiio.be_move_done = ctl_backend_ramdisk_move_done; io->scsiio.kern_data_resid = 0; io->scsiio.kern_data_len = len_filled; io->scsiio.kern_sg_entries = sg_filled; io->io_hdr.flags |= CTL_FLAG_ALLOCATED; io->io_hdr.ctl_private[CTL_PRIV_BACKEND].integer -= len_filled; #ifdef CTL_TIME_IO getbintime(&io->io_hdr.dma_start_bt); #endif ctl_datamove(io); } static void ctl_backend_ramdisk_worker(void *context, int pending) { struct ctl_be_ramdisk_softc *softc; struct ctl_be_ramdisk_lun *be_lun; union ctl_io *io; be_lun = (struct ctl_be_ramdisk_lun *)context; softc = be_lun->softc; 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); ctl_backend_ramdisk_continue(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_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td) { struct ctl_be_ramdisk_softc *softc; int retval; retval = 0; softc = &rd_softc; 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: retval = ctl_backend_ramdisk_create(softc, lun_req, /*do_wait*/ 1); 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; retval = 0; params = &req->reqdata.rm; be_lun = NULL; mtx_lock(&softc->lock); STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->ctl_be_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->ctl_be_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->ctl_be_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(be_lun->io_taskqueue, &be_lun->io_task); taskqueue_free(be_lun->io_taskqueue); ctl_free_opts(&be_lun->ctl_be_lun.options); 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, int do_wait) { struct ctl_be_ramdisk_lun *be_lun; struct ctl_lun_create_params *params; uint32_t blocksize; char *value; char tmpstr[32]; int retval, unmap; retval = 0; params = &req->reqdata.create; if (params->blocksize_bytes != 0) blocksize = params->blocksize_bytes; else blocksize = 512; be_lun = malloc(sizeof(*be_lun), M_RAMDISK, M_ZERO | (do_wait ? M_WAITOK : M_NOWAIT)); if (be_lun == NULL) { snprintf(req->error_str, sizeof(req->error_str), "%s: error allocating %zd bytes", __func__, sizeof(*be_lun)); goto bailout_error; } sprintf(be_lun->lunname, "cram%d", softc->num_luns); ctl_init_opts(&be_lun->ctl_be_lun.options, req->num_be_args, req->kern_be_args); if (params->flags & CTL_LUN_FLAG_DEV_TYPE) be_lun->ctl_be_lun.lun_type = params->device_type; else be_lun->ctl_be_lun.lun_type = T_DIRECT; if (be_lun->ctl_be_lun.lun_type == T_DIRECT) { if (params->lun_size_bytes < blocksize) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN size %ju < blocksize %u", __func__, params->lun_size_bytes, blocksize); goto bailout_error; } be_lun->size_blocks = params->lun_size_bytes / blocksize; be_lun->size_bytes = be_lun->size_blocks * blocksize; be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; } else { be_lun->ctl_be_lun.maxlba = 0; blocksize = 0; be_lun->size_bytes = 0; be_lun->size_blocks = 0; } be_lun->ctl_be_lun.blocksize = blocksize; /* Tell the user the blocksize we ended up using */ params->blocksize_bytes = blocksize; /* Tell the user the exact size we ended up using */ params->lun_size_bytes = be_lun->size_bytes; be_lun->softc = softc; unmap = 0; value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); if (value != NULL && strcmp(value, "on") == 0) unmap = 1; be_lun->flags = CTL_BE_RAMDISK_LUN_UNCONFIGURED; be_lun->ctl_be_lun.flags = CTL_LUN_FLAG_PRIMARY; if (unmap) be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_UNMAP; be_lun->ctl_be_lun.atomicblock = UINT32_MAX; be_lun->ctl_be_lun.be_lun = be_lun; if (params->flags & CTL_LUN_FLAG_ID_REQ) { be_lun->ctl_be_lun.req_lun_id = params->req_lun_id; be_lun->ctl_be_lun.flags |= CTL_LUN_FLAG_ID_REQ; } else be_lun->ctl_be_lun.req_lun_id = 0; be_lun->ctl_be_lun.lun_shutdown = ctl_backend_ramdisk_lun_shutdown; be_lun->ctl_be_lun.lun_config_status = ctl_backend_ramdisk_lun_config_status; be_lun->ctl_be_lun.be = &ctl_be_ramdisk_driver; if ((params->flags & CTL_LUN_FLAG_SERIAL_NUM) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYSERIAL%4d", softc->num_luns); strncpy((char *)be_lun->ctl_be_lun.serial_num, tmpstr, ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), sizeof(tmpstr))); /* Tell the user what we used for a serial number */ strncpy((char *)params->serial_num, tmpstr, ctl_min(sizeof(params->serial_num), sizeof(tmpstr))); } else { strncpy((char *)be_lun->ctl_be_lun.serial_num, params->serial_num, ctl_min(sizeof(be_lun->ctl_be_lun.serial_num), sizeof(params->serial_num))); } if ((params->flags & CTL_LUN_FLAG_DEVID) == 0) { snprintf(tmpstr, sizeof(tmpstr), "MYDEVID%4d", softc->num_luns); strncpy((char *)be_lun->ctl_be_lun.device_id, tmpstr, ctl_min(sizeof(be_lun->ctl_be_lun.device_id), sizeof(tmpstr))); /* Tell the user what we used for a device ID */ strncpy((char *)params->device_id, tmpstr, ctl_min(sizeof(params->device_id), sizeof(tmpstr))); } else { strncpy((char *)be_lun->ctl_be_lun.device_id, params->device_id, ctl_min(sizeof(be_lun->ctl_be_lun.device_id), sizeof(params->device_id))); } STAILQ_INIT(&be_lun->cont_queue); 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->ctl_be_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; } if (do_wait == 0) return (retval); 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 = be_lun->ctl_be_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(&be_lun->ctl_be_lun.options); 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_lun_modify_params *params; uint32_t blocksize; params = &req->reqdata.modify; be_lun = NULL; mtx_lock(&softc->lock); STAILQ_FOREACH(be_lun, &softc->lun_list, links) { if (be_lun->ctl_be_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; } if (params->lun_size_bytes == 0) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN size \"auto\" not supported " "by the ramdisk backend", __func__); goto bailout_error; } blocksize = be_lun->ctl_be_lun.blocksize; if (params->lun_size_bytes < blocksize) { snprintf(req->error_str, sizeof(req->error_str), "%s: LUN size %ju < blocksize %u", __func__, params->lun_size_bytes, blocksize); goto bailout_error; } be_lun->size_blocks = params->lun_size_bytes / blocksize; be_lun->size_bytes = be_lun->size_blocks * blocksize; /* * The maximum LBA is the size - 1. * * XXX: Note that this field is being updated without locking, * which might cause problems on 32-bit architectures. */ be_lun->ctl_be_lun.maxlba = be_lun->size_blocks - 1; ctl_lun_capacity_changed(&be_lun->ctl_be_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; struct ctl_be_ramdisk_softc *softc; int do_free; lun = (struct ctl_be_ramdisk_lun *)be_lun; softc = lun->softc; do_free = 0; 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--; do_free = 1; } mtx_unlock(&softc->lock); if (do_free != 0) free(be_lun, M_RAMDISK); } 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->ctl_be_lun) != 0) { printf("%s: ctl_enable_lun() failed!\n", __func__); if (ctl_invalidate_lun(&lun->ctl_be_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); } static int ctl_backend_ramdisk_config_write(union ctl_io *io) { struct ctl_be_ramdisk_softc *softc; int retval; retval = 0; softc = &rd_softc; switch (io->scsiio.cdb[0]) { case SYNCHRONIZE_CACHE: case SYNCHRONIZE_CACHE_16: /* * The upper level CTL code will filter out any CDBs with * the immediate bit set and return the proper error. It * will also not allow a sync cache command to go to a LUN * that is powered down. * * 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. * * This is obviously just a stubbed out implementation. * The real implementation will be in the RAIDCore/CTL * interface, and can only really happen when RAIDCore * implements a per-array cache sync. */ ctl_set_success(&io->scsiio); ctl_config_write_done(io); break; case START_STOP_UNIT: { struct scsi_start_stop_unit *cdb; struct ctl_be_lun *ctl_be_lun; struct ctl_be_ramdisk_lun *be_lun; cdb = (struct scsi_start_stop_unit *)io->scsiio.cdb; ctl_be_lun = (struct ctl_be_lun *)io->io_hdr.ctl_private[ CTL_PRIV_BACKEND_LUN].ptr; be_lun = (struct ctl_be_ramdisk_lun *)ctl_be_lun->be_lun; if (cdb->how & SSS_START) retval = ctl_start_lun(ctl_be_lun); else { retval = ctl_stop_lun(ctl_be_lun); #ifdef NEEDTOPORT if ((retval == 0) && (cdb->byte2 & SSS_ONOFFLINE)) retval = ctl_lun_offline(ctl_be_lun); #endif } /* * In general, the above routines should not fail. They * just set state for the LUN. So we've got something * pretty wrong here if we can't start or stop the LUN. */ if (retval != 0) { ctl_set_internal_failure(&io->scsiio, /*sks_valid*/ 1, /*retry_count*/ 0xf051); retval = CTL_RETVAL_COMPLETE; } else { ctl_set_success(&io->scsiio); } ctl_config_write_done(io); break; } case WRITE_SAME_10: case WRITE_SAME_16: case UNMAP: 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_backend_ramdisk_config_read(union ctl_io *io) { /* * XXX KDM need to implement!! */ return (0); } Index: head/sys/cam/ctl/ctl_frontend_iscsi.c =================================================================== --- head/sys/cam/ctl/ctl_frontend_iscsi.c (revision 275057) +++ head/sys/cam/ctl/ctl_frontend_iscsi.c (revision 275058) @@ -1,2929 +1,2953 @@ /*- * 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 #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 maxcmdsn_delta = 256; SYSCTL_INT(_kern_cam_ctl_iscsi, OID_AUTO, maxcmdsn_delta, CTLFLAG_RWTUN, &maxcmdsn_delta, 256, "Number of commands the initiator can send " "without confirmation"); #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 int cfiscsi_init(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_lun_enable(void *arg, struct ctl_id target_id, int lun_id); static int cfiscsi_lun_disable(void *arg, struct ctl_id target_id, int lun_id); static uint32_t cfiscsi_lun_map(void *arg, uint32_t lun); 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_target *cfiscsi_target_find(struct cfiscsi_softc *softc, const char *name); static struct cfiscsi_target *cfiscsi_target_find_or_create( struct cfiscsi_softc *softc, const char *name, const char *alias); static void cfiscsi_target_release(struct cfiscsi_target *ct); static void cfiscsi_session_delete(struct cfiscsi_session *cs); static struct cfiscsi_softc cfiscsi_softc; extern struct ctl_softc *control_softc; static struct ctl_frontend cfiscsi_frontend = { .name = "iscsi", .init = cfiscsi_init, .ioctl = cfiscsi_ioctl, }; CTL_FRONTEND_DECLARE(ctlcfiscsi, cfiscsi_frontend); MODULE_DEPEND(ctlcfiscsi, 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 /* * The target MUST silently ignore any non-immediate command outside * of this range. */ if (cmdsn < cs->cs_cmdsn || cmdsn > cs->cs_cmdsn + maxcmdsn_delta) { CFISCSI_SESSION_UNLOCK(cs); CFISCSI_SESSION_WARN(cs, "received PDU with CmdSN %d, " "while expected CmdSN was %d", cmdsn, cs->cs_cmdsn); return (true); } if ((request->ip_bhs->bhs_opcode & ISCSI_BHS_OPCODE_IMMEDIATE) == 0) 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) { struct cfiscsi_session *cs; cs = PDU_SESSION(request); #ifdef ICL_KERNEL_PROXY 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) + 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) + 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 + maxcmdsn_delta); 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 uint32_t cfiscsi_decode_lun(uint64_t encoded) { uint8_t lun[8]; uint32_t result; /* * The LUN field in iSCSI PDUs may look like an ordinary 64 bit number, * but is in fact an evil, multidimensional structure defined * in SCSI Architecture Model 5 (SAM-5), section 4.6. */ memcpy(lun, &encoded, sizeof(lun)); switch (lun[0] & 0xC0) { case 0x00: if ((lun[0] & 0x3f) != 0 || lun[2] != 0 || lun[3] != 0 || lun[4] != 0 || lun[5] != 0 || lun[6] != 0 || lun[7] != 0) { CFISCSI_WARN("malformed LUN " "(peripheral device addressing method): 0x%jx", (uintmax_t)encoded); result = 0xffffffff; break; } result = lun[1]; break; case 0x40: if (lun[2] != 0 || lun[3] != 0 || lun[4] != 0 || lun[5] != 0 || lun[6] != 0 || lun[7] != 0) { CFISCSI_WARN("malformed LUN " "(flat address space addressing method): 0x%jx", (uintmax_t)encoded); result = 0xffffffff; break; } result = ((lun[0] & 0x3f) << 8) + lun[1]; break; case 0xC0: if (lun[0] != 0xD2 || lun[4] != 0 || lun[5] != 0 || lun[6] != 0 || lun[7] != 0) { CFISCSI_WARN("malformed LUN (extended flat " "address space addressing method): 0x%jx", (uintmax_t)encoded); result = 0xffffffff; break; } result = (lun[1] << 16) + (lun[2] << 8) + lun[3]; default: CFISCSI_WARN("unsupported LUN format 0x%jx", (uintmax_t)encoded); result = 0xffffffff; break; } return (result); } 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.id = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_target.id = 0; io->io_hdr.nexus.targ_lun = cfiscsi_decode_lun(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.id = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_target.id = 0; io->io_hdr.nexus.targ_lun = cfiscsi_decode_lun(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_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; 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; 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; } 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); uma_zfree(cfiscsi_data_wait_zone, cdw); 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 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.id = cs->cs_ctl_initid; io->io_hdr.nexus.targ_port = cs->cs_target->ct_port.targ_port; io->io_hdr.nexus.targ_target.id = 0; 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->scsiio.io_hdr.port_status = 42; cdw->cdw_ctl_io->scsiio.be_move_done(cdw->cdw_ctl_io); uma_zfree(cfiscsi_data_wait_zone, 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) cv_wait(&cs->cs_maintenance_cv, &cs->cs_lock); CFISCSI_SESSION_UNLOCK(cs); if (cs->cs_terminating) { /* * 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) { if (cs->cs_terminating) return; 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) { 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 cs->cs_conn = icl_conn_new("cfiscsi", &cs->cs_lock); 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")); 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); mtx_lock(&softc->lock); TAILQ_REMOVE(&softc->sessions, cs, cs_next); cv_signal(&softc->sessions_cv); mtx_unlock(&softc->lock); free(cs, M_CFISCSI); } int cfiscsi_init(void) { struct cfiscsi_softc *softc; int retval; softc = &cfiscsi_softc; retval = 0; 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); } #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); 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_waiting_for_ctld = true; cv_signal(&cfiscsi_softc.accept_cv); } #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); 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); 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_portal_group_tag = cihp->portal_group_tag; cs->cs_cmdsn = cihp->cmdsn; cs->cs_statsn = cihp->statsn; cs->cs_max_data_segment_length = cihp->max_recv_data_segment_length; cs->cs_max_burst_length = cihp->max_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); 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); refcount_acquire(&cs->cs_outstanding_ctl_pdus); 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 && cs->cs_portal_group_tag == cs2->cs_portal_group_tag && strcmp(cs->cs_initiator_id, cs2->cs_initiator_id) == 0) { 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) { cfiscsi_session_terminate(cs); refcount_release(&cs->cs_outstanding_ctl_pdus); 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 refcount_release(&cs->cs_outstanding_ctl_pdus); 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" "%s" "%s" "%zd" "%d" "%d" "\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_conn->ic_header_crc32c ? "CRC32C" : "None", cs->cs_conn->ic_data_crc32c ? "CRC32C" : "None", cs->cs_max_data_segment_length, cs->cs_immediate_data, cs->cs_conn->ic_iser); 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); cilp->fill_len = sbuf_len(sb) + 1; ci->status = CTL_ISCSI_OK; sbuf_delete(sb); } 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_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; } #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, *tag; struct scsi_vpd_id_descriptor *desc; ctl_options_t opts; int retval, len, idlen; 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"); tag = ctl_get_opt(&opts, "cfiscsi_portal_group_tag"); if (target == NULL || tag == NULL) { req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Missing required argument"); ctl_free_opts(&opts); return; } ct = cfiscsi_target_find_or_create(&cfiscsi_softc, target, alias); 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\" already exists", target); cfiscsi_target_release(ct); ctl_free_opts(&opts); return; } port = &ct->ct_port; 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 = strtoul(tag, NULL, 0); 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->lun_enable = cfiscsi_lun_enable; port->lun_disable = cfiscsi_lun_disable; port->lun_map = cfiscsi_lun_map; port->targ_lun_arg = ct; port->fe_datamove = cfiscsi_datamove; port->fe_done = cfiscsi_done; /* XXX KDM what should we report here? */ /* XXX These should probably be fetched from CTL. */ port->max_targets = 1; port->max_target_id = 15; 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, port->physical_port); /* 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); cfiscsi_target_release(ct); free(port->port_devid, M_CFISCSI); free(port->target_devid, M_CFISCSI); req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "ctl_frontend_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 } static void cfiscsi_ioctl_port_remove(struct ctl_req *req) { struct cfiscsi_target *ct; const char *target; ctl_options_t opts; ctl_init_opts(&opts, req->num_args, req->kern_args); target = ctl_get_opt(&opts, "cfiscsi_target"); if (target == NULL) { ctl_free_opts(&opts); req->status = CTL_LUN_ERROR; snprintf(req->error_str, sizeof(req->error_str), "Missing required argument"); return; } ct = cfiscsi_target_find(&cfiscsi_softc, target); 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); } 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_TERMINATE: cfiscsi_ioctl_terminate(ci); break; case CTL_ISCSI_LOGOUT: cfiscsi_ioctl_logout(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) { struct cfiscsi_target *ct; mtx_lock(&softc->lock); TAILQ_FOREACH(ct, &softc->targets, ct_next) { if (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) { struct cfiscsi_target *ct, *newct; int i; 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 (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); } for (i = 0; i < CTL_MAX_LUNS; i++) newct->ct_luns[i] = UINT32_MAX; strlcpy(newct->ct_name, name, sizeof(newct->ct_name)); if (alias != NULL) strlcpy(newct->ct_alias, alias, sizeof(newct->ct_alias)); 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); } /* * Takes LUN from the target space and returns LUN from the CTL space. */ static uint32_t cfiscsi_lun_map(void *arg, uint32_t lun) { struct cfiscsi_target *ct = arg; if (lun >= CTL_MAX_LUNS) { CFISCSI_DEBUG("requested lun number %d is higher " "than maximum %d", lun, CTL_MAX_LUNS - 1); return (UINT32_MAX); } return (ct->ct_luns[lun]); } static int cfiscsi_target_set_lun(struct cfiscsi_target *ct, unsigned long lun_id, unsigned long ctl_lun_id) { if (lun_id >= CTL_MAX_LUNS) { CFISCSI_WARN("requested lun number %ld is higher " "than maximum %d", lun_id, CTL_MAX_LUNS - 1); return (-1); } if (ct->ct_luns[lun_id] < CTL_MAX_LUNS) { /* * CTL calls cfiscsi_lun_enable() twice for each LUN - once * when the LUN is created, and a second time just before * the port is brought online; don't emit warnings * for that case. */ if (ct->ct_luns[lun_id] == ctl_lun_id) return (0); CFISCSI_WARN("lun %ld already allocated", lun_id); return (-1); } #if 0 CFISCSI_DEBUG("adding mapping for lun %ld, target %s " "to ctl lun %ld", lun_id, ct->ct_name, ctl_lun_id); #endif ct->ct_luns[lun_id] = ctl_lun_id; return (0); } static int cfiscsi_lun_enable(void *arg, struct ctl_id target_id, int lun_id) { struct cfiscsi_softc *softc; struct cfiscsi_target *ct; const char *target = NULL; const char *lun = NULL; unsigned long tmp; ct = (struct cfiscsi_target *)arg; softc = ct->ct_softc; target = ctl_get_opt(&control_softc->ctl_luns[lun_id]->be_lun->options, "cfiscsi_target"); lun = ctl_get_opt(&control_softc->ctl_luns[lun_id]->be_lun->options, "cfiscsi_lun"); if (target == NULL && lun == NULL) return (0); if (target == NULL || lun == NULL) { CFISCSI_WARN("lun added with cfiscsi_target, but without " "cfiscsi_lun, or the other way around; ignoring"); return (0); } if (strcmp(target, ct->ct_name) != 0) return (0); tmp = strtoul(lun, NULL, 10); cfiscsi_target_set_lun(ct, tmp, lun_id); return (0); } static int cfiscsi_lun_disable(void *arg, struct ctl_id target_id, int lun_id) { struct cfiscsi_softc *softc; struct cfiscsi_target *ct; int i; ct = (struct cfiscsi_target *)arg; softc = ct->ct_softc; mtx_lock(&softc->lock); for (i = 0; i < CTL_MAX_LUNS; i++) { if (ct->ct_luns[i] != lun_id) continue; ct->ct_luns[i] = UINT32_MAX; break; } mtx_unlock(&softc->lock); return (0); } 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_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_data_segment_length, ("ip_data_len %zd >= max_data_segment_length %zd", response->ip_data_len, cs->cs_max_data_segment_length)); if (response->ip_data_len + len > cs->cs_max_data_segment_length) { len = cs->cs_max_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; 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_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) - bhsdi->bhsdi_flags |= BHSDI_FLAGS_F; + 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) + 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, 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; /* * Report write underflow as error since CTL and backends don't * really support it, and SCSI does not tell how to do it right. */ expected_len = ntohl(bhssc->bhssc_expected_data_transfer_length); if (io->scsiio.kern_rel_offset + io->scsiio.kern_data_len > expected_len) { io->scsiio.io_hdr.port_status = 43; io->scsiio.be_move_done(io); return; } target_transfer_tag = atomic_fetchadd_32(&cs->cs_target_transfer_tag, 1); #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 = uma_zalloc(cfiscsi_data_wait_zone, M_NOWAIT | M_ZERO); 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; } 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 = io->scsiio.kern_data_len; /* 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 = io->scsiio.kern_data_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) { uma_zfree(cfiscsi_data_wait_zone, cdw); io->scsiio.be_move_done(io); return; } } r2t_off = io->scsiio.kern_rel_offset + io->scsiio.ext_data_filled; r2t_len = MIN(io->scsiio.kern_data_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; } 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) { + 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; 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); cdw->cdw_ctl_io->scsiio.be_move_done(cdw->cdw_ctl_io); uma_zfree(cfiscsi_data_wait_zone, cdw); } CFISCSI_SESSION_UNLOCK(cs); } 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; if (io->io_hdr.status == CTL_SUCCESS) { bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_COMPLETE; } else { /* * XXX: How to figure out what exactly went wrong? iSCSI spec * expects us to provide detailed error, e.g. "Task does * not exist" or "LUN does not exist". */ CFISCSI_SESSION_DEBUG(cs, "BHSTMR_RESPONSE_FUNCTION_NOT_SUPPORTED"); bhstmr2->bhstmr_response = BHSTMR_RESPONSE_FUNCTION_NOT_SUPPORTED; } bhstmr2->bhstmr_initiator_task_tag = bhstmr->bhstmr_initiator_task_tag; ctl_free_io(io); icl_pdu_free(request); cfiscsi_pdu_queue(response); } 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); refcount_release(&cs->cs_outstanding_ctl_pdus); 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); } } Index: head/sys/cam/ctl/ctl_io.h =================================================================== --- head/sys/cam/ctl/ctl_io.h (revision 275057) +++ head/sys/cam/ctl/ctl_io.h (revision 275058) @@ -1,525 +1,526 @@ /*- * 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_io.h#5 $ * $FreeBSD$ */ /* * CAM Target Layer data movement structures/interface. * * Author: Ken Merry */ #ifndef _CTL_IO_H_ #define _CTL_IO_H_ #ifdef _CTL_C #define EXTERN(__var,__val) __var = __val #else #define EXTERN(__var,__val) extern __var #endif #define CTL_MAX_CDBLEN 32 /* * Uncomment this next line to enable printing out times for I/Os * that take longer than CTL_TIME_IO_SECS seconds to get to the datamove * and/or done stage. */ #define CTL_TIME_IO #ifdef CTL_TIME_IO #define CTL_TIME_IO_DEFAULT_SECS 90 EXTERN(int ctl_time_io_secs, CTL_TIME_IO_DEFAULT_SECS); #endif /* * Uncomment these next two lines to enable the CTL I/O delay feature. You * can delay I/O at two different points -- datamove and done. This is * useful for diagnosing abort conditions (for hosts that send an abort on a * timeout), and for determining how long a host's timeout is. */ #define CTL_IO_DELAY #define CTL_TIMER_BYTES sizeof(struct callout) typedef enum { CTL_STATUS_NONE, /* No status */ CTL_SUCCESS, /* Transaction completed successfully */ CTL_CMD_TIMEOUT, /* Command timed out, shouldn't happen here */ CTL_SEL_TIMEOUT, /* Selection timeout, shouldn't happen here */ CTL_ERROR, /* General CTL error XXX expand on this? */ CTL_SCSI_ERROR, /* SCSI error, look at status byte/sense data */ CTL_CMD_ABORTED, /* Command aborted, don't return status */ CTL_STATUS_MASK = 0xfff,/* Mask off any status flags */ CTL_AUTOSENSE = 0x1000 /* Autosense performed */ } ctl_io_status; /* * WARNING: Keep the data in/out/none flags where they are. They're used * in conjuction with ctl_cmd_flags. See comment above ctl_cmd_flags * definition in ctl_private.h. */ typedef enum { CTL_FLAG_NONE = 0x00000000, /* no flags */ CTL_FLAG_DATA_IN = 0x00000001, /* DATA IN */ CTL_FLAG_DATA_OUT = 0x00000002, /* DATA OUT */ CTL_FLAG_DATA_NONE = 0x00000003, /* no data */ CTL_FLAG_DATA_MASK = 0x00000003, CTL_FLAG_KDPTR_SGLIST = 0x00000008, /* kern_data_ptr is S/G list*/ CTL_FLAG_EDPTR_SGLIST = 0x00000010, /* ext_data_ptr is S/G list */ CTL_FLAG_DO_AUTOSENSE = 0x00000020, /* grab sense info */ CTL_FLAG_USER_REQ = 0x00000040, /* request came from userland */ CTL_FLAG_CONTROL_DEV = 0x00000080, /* processor device */ CTL_FLAG_ALLOCATED = 0x00000100, /* data space allocated */ CTL_FLAG_BLOCKED = 0x00000200, /* on the blocked queue */ CTL_FLAG_ABORT_STATUS = 0x00000400, /* return TASK ABORTED status */ CTL_FLAG_ABORT = 0x00000800, /* this I/O should be aborted */ CTL_FLAG_DMA_INPROG = 0x00001000, /* DMA in progress */ CTL_FLAG_NO_DATASYNC = 0x00002000, /* don't cache flush data */ CTL_FLAG_DELAY_DONE = 0x00004000, /* delay injection done */ CTL_FLAG_INT_COPY = 0x00008000, /* internal copy, no done call*/ CTL_FLAG_SENT_2OTHER_SC = 0x00010000, CTL_FLAG_FROM_OTHER_SC = 0x00020000, CTL_FLAG_IS_WAS_ON_RTR = 0x00040000, /* Don't rerun cmd on failover*/ CTL_FLAG_BUS_ADDR = 0x00080000, /* ctl_sglist contains BUS addresses, not virtual ones*/ CTL_FLAG_IO_CONT = 0x00100000, /* Continue I/O instead of completing */ CTL_FLAG_AUTO_MIRROR = 0x00200000, /* Automatically use memory from the RC cache mirrored address area. */ #if 0 CTL_FLAG_ALREADY_DONE = 0x00200000 /* I/O already completed */ #endif CTL_FLAG_NO_DATAMOVE = 0x00400000, CTL_FLAG_DMA_QUEUED = 0x00800000, /* DMA queued but not started*/ CTL_FLAG_STATUS_QUEUED = 0x01000000, /* Status queued but not sent*/ CTL_FLAG_REDIR_DONE = 0x02000000, /* Redirection has already been done. */ CTL_FLAG_FAILOVER = 0x04000000, /* Killed by a failover */ CTL_FLAG_IO_ACTIVE = 0x08000000, /* I/O active on this SC */ CTL_FLAG_RDMA_MASK = CTL_FLAG_NO_DATASYNC | CTL_FLAG_BUS_ADDR | - CTL_FLAG_AUTO_MIRROR | CTL_FLAG_REDIR_DONE + CTL_FLAG_AUTO_MIRROR | CTL_FLAG_REDIR_DONE, /* Flags we care about for remote DMA */ + CTL_FLAG_STATUS_SENT = 0x10000000 /* Status sent by datamove */ } ctl_io_flags; struct ctl_lba_len { uint64_t lba; uint32_t len; }; struct ctl_lba_len_flags { uint64_t lba; uint32_t len; uint32_t flags; #define CTL_LLF_FUA 0x04000000 #define CTL_LLF_DPO 0x08000000 #define CTL_LLF_READ 0x10000000 #define CTL_LLF_WRITE 0x20000000 #define CTL_LLF_VERIFY 0x40000000 #define CTL_LLF_COMPARE 0x80000000 }; struct ctl_ptr_len_flags { uint8_t *ptr; uint32_t len; uint32_t flags; }; union ctl_priv { uint8_t bytes[sizeof(uint64_t) * 2]; uint64_t integer; void *ptr; }; /* * Number of CTL private areas. */ #define CTL_NUM_PRIV 6 /* * Which private area are we using for a particular piece of data? */ #define CTL_PRIV_LUN 0 /* CTL LUN pointer goes here */ #define CTL_PRIV_LBA_LEN 1 /* Decoded LBA/len for read/write*/ #define CTL_PRIV_MODEPAGE 1 /* Modepage info for config write */ #define CTL_PRIV_BACKEND 2 /* Reserved for block, RAIDCore */ #define CTL_PRIV_BACKEND_LUN 3 /* Backend LUN pointer */ #define CTL_PRIV_FRONTEND 4 /* LSI driver, ioctl front end */ #define CTL_PRIV_USER 5 /* Userland use */ #define CTL_INVALID_PORTNAME 0xFF #define CTL_UNMAPPED_IID 0xFF /* * XXX KDM this size is for the port_priv variable in struct ctl_io_hdr * below. This should be defined in terms of the size of struct * ctlfe_lun_cmd_info at the moment: * struct ctlfe_lun_cmd_info { * int cur_transfer_index; * ctlfe_cmd_flags flags; * bus_dma_segment_t cam_sglist[32]; * }; * * This isn't really the way I'd prefer to do it, but it does make some * sense, AS LONG AS we can guarantee that there will always only be one * outstanding DMA request per ctl_io. If that assumption isn't valid, * then we've got problems. * * At some point it may be nice switch CTL over to using CCBs for * everything. At that point we can probably use the ATIO/CTIO model, so * that multiple simultaneous DMAs per command will just work. * * Also note that the current size, 600, is appropriate for 64-bit * architectures, but is overkill for 32-bit architectures. Need a way to * figure out the size at compile time, or just get rid of this altogether. */ #define CTL_PORT_PRIV_SIZE 600 struct ctl_sg_entry { void *addr; size_t len; }; struct ctl_id { uint32_t id; uint64_t wwid[2]; }; typedef enum { CTL_IO_NONE, CTL_IO_SCSI, CTL_IO_TASK, } ctl_io_type; struct ctl_nexus { struct ctl_id initid; /* Initiator ID */ uint32_t targ_port; /* Target port, filled in by PORT */ struct ctl_id targ_target; /* Destination target */ uint32_t targ_lun; /* Destination lun */ uint32_t targ_mapped_lun; /* Destination lun CTL-wide */ }; typedef enum { CTL_MSG_SERIALIZE, CTL_MSG_R2R, CTL_MSG_FINISH_IO, CTL_MSG_BAD_JUJU, CTL_MSG_MANAGE_TASKS, CTL_MSG_PERS_ACTION, CTL_MSG_SYNC_FE, CTL_MSG_DATAMOVE, CTL_MSG_DATAMOVE_DONE } ctl_msg_type; struct ctl_scsiio; #define CTL_NUM_SG_ENTRIES 9 struct ctl_io_hdr { uint32_t version; /* interface version XXX */ ctl_io_type io_type; /* task I/O, SCSI I/O, etc. */ ctl_msg_type msg_type; struct ctl_nexus nexus; /* Initiator, port, target, lun */ uint32_t iid_indx; /* the index into the iid mapping */ uint32_t flags; /* transaction flags */ uint32_t status; /* transaction status */ uint32_t port_status; /* trans status, set by PORT, 0 = good*/ uint32_t timeout; /* timeout in ms */ uint32_t retries; /* retry count */ #ifdef CTL_IO_DELAY uint8_t timer_bytes[CTL_TIMER_BYTES]; /* timer kludge */ #endif /* CTL_IO_DELAY */ #ifdef CTL_TIME_IO time_t start_time; /* I/O start time */ struct bintime start_bt; /* Timer start ticks */ struct bintime dma_start_bt; /* DMA start ticks */ struct bintime dma_bt; /* DMA total ticks */ uint32_t num_dmas; /* Number of DMAs */ #endif /* CTL_TIME_IO */ union ctl_io *original_sc; union ctl_io *serializing_sc; void *pool; /* I/O pool */ union ctl_priv ctl_private[CTL_NUM_PRIV];/* CTL private area */ uint8_t port_priv[CTL_PORT_PRIV_SIZE];/* PORT private area*/ struct ctl_sg_entry remote_sglist[CTL_NUM_SG_ENTRIES]; struct ctl_sg_entry remote_dma_sglist[CTL_NUM_SG_ENTRIES]; struct ctl_sg_entry local_sglist[CTL_NUM_SG_ENTRIES]; struct ctl_sg_entry local_dma_sglist[CTL_NUM_SG_ENTRIES]; STAILQ_ENTRY(ctl_io_hdr) links; /* linked list pointer */ TAILQ_ENTRY(ctl_io_hdr) ooa_links; TAILQ_ENTRY(ctl_io_hdr) blocked_links; }; typedef enum { CTL_TAG_UNTAGGED, CTL_TAG_SIMPLE, CTL_TAG_ORDERED, CTL_TAG_HEAD_OF_QUEUE, CTL_TAG_ACA } ctl_tag_type; union ctl_io; /* * SCSI passthrough I/O structure for the CAM Target Layer. Note * that some of these fields are here for completeness, but they aren't * used in the CTL implementation. e.g., timeout and retries won't be * used. * * Note: Make sure the io_hdr is *always* the first element in this * structure. */ struct ctl_scsiio { struct ctl_io_hdr io_hdr; /* common to all I/O types */ /* * The ext_* fields are generally intended for frontend use; CTL itself * doesn't modify or use them. */ uint32_t ext_sg_entries; /* 0 = no S/G list, > 0 = num entries */ uint8_t *ext_data_ptr; /* data buffer or S/G list */ uint32_t ext_data_len; /* Data transfer length */ uint32_t ext_data_filled; /* Amount of data filled so far */ /* * The number of scatter/gather entries in the list pointed to * by kern_data_ptr. 0 means there is no list, just a data pointer. */ uint32_t kern_sg_entries; uint32_t rem_sg_entries; /* Unused. */ /* * The data pointer or a pointer to the scatter/gather list. */ uint8_t *kern_data_ptr; /* * Length of the data buffer or scatter/gather list. It's also * the length of this particular piece of the data transfer, * ie. number of bytes expected to be transferred by the current * invocation of frontend's datamove() callback. It's always * less than or equal to kern_total_len. */ uint32_t kern_data_len; /* * Total length of data to be transferred during this particular * SCSI command, as decoded from SCSI CDB. */ uint32_t kern_total_len; /* * Amount of data left after the current data transfer. */ uint32_t kern_data_resid; /* * Byte offset of this transfer, equal to the amount of data * already transferred for this SCSI command during previous * datamove() invocations. */ uint32_t kern_rel_offset; struct scsi_sense_data sense_data; /* sense data */ uint8_t sense_len; /* Returned sense length */ uint8_t scsi_status; /* SCSI status byte */ uint8_t sense_residual; /* Unused. */ uint32_t residual; /* data residual length */ uint32_t tag_num; /* tag number */ ctl_tag_type tag_type; /* simple, ordered, head of queue,etc.*/ uint8_t cdb_len; /* CDB length */ uint8_t cdb[CTL_MAX_CDBLEN]; /* CDB */ int (*be_move_done)(union ctl_io *io); /* called by fe */ int (*io_cont)(union ctl_io *io); /* to continue processing */ }; typedef enum { CTL_TASK_ABORT_TASK, CTL_TASK_ABORT_TASK_SET, CTL_TASK_CLEAR_ACA, CTL_TASK_CLEAR_TASK_SET, CTL_TASK_I_T_NEXUS_RESET, CTL_TASK_LUN_RESET, CTL_TASK_TARGET_RESET, CTL_TASK_BUS_RESET, CTL_TASK_PORT_LOGIN, CTL_TASK_PORT_LOGOUT } ctl_task_type; /* * Task management I/O structure. Aborts, bus resets, etc., are sent using * this structure. * * Note: Make sure the io_hdr is *always* the first element in this * structure. */ struct ctl_taskio { struct ctl_io_hdr io_hdr; /* common to all I/O types */ ctl_task_type task_action; /* Target Reset, Abort, etc. */ uint32_t tag_num; /* tag number */ ctl_tag_type tag_type; /* simple, ordered, etc. */ }; typedef enum { CTL_PR_REG_KEY, CTL_PR_UNREG_KEY, CTL_PR_PREEMPT, CTL_PR_CLEAR, CTL_PR_RESERVE, CTL_PR_RELEASE } ctl_pr_action; /* * The PR info is specifically for sending Persistent Reserve actions * to the other SC which it must also act on. * * Note: Make sure the io_hdr is *always* the first element in this * structure. */ struct ctl_pr_info { ctl_pr_action action; uint8_t sa_res_key[8]; uint8_t res_type; uint32_t residx; }; struct ctl_ha_msg_hdr { ctl_msg_type msg_type; union ctl_io *original_sc; union ctl_io *serializing_sc; struct ctl_nexus nexus; /* Initiator, port, target, lun */ uint32_t status; /* transaction status */ TAILQ_ENTRY(ctl_ha_msg_hdr) links; }; #define CTL_HA_MAX_SG_ENTRIES 16 /* * Used for CTL_MSG_PERS_ACTION. */ struct ctl_ha_msg_pr { struct ctl_ha_msg_hdr hdr; struct ctl_pr_info pr_info; }; /* * The S/G handling here is a little different than the standard ctl_scsiio * structure, because we can't pass data by reference in between controllers. * The S/G list in the ctl_scsiio struct is normally passed in the * kern_data_ptr field. So kern_sg_entries here will always be non-zero, * even if there is only one entry. * * Used for CTL_MSG_DATAMOVE. */ struct ctl_ha_msg_dt { struct ctl_ha_msg_hdr hdr; ctl_io_flags flags; /* Only I/O flags are used here */ uint32_t sg_sequence; /* S/G portion number */ uint8_t sg_last; /* last S/G batch = 1 */ uint32_t sent_sg_entries; /* previous S/G count */ uint32_t cur_sg_entries; /* current S/G entries */ uint32_t kern_sg_entries; /* total S/G entries */ uint32_t kern_data_len; /* Length of this S/G list */ uint32_t kern_total_len; /* Total length of this transaction */ uint32_t kern_data_resid; /* Length left to transfer after this*/ uint32_t kern_rel_offset; /* Byte Offset of this transfer */ struct ctl_sg_entry sg_list[CTL_HA_MAX_SG_ENTRIES]; }; /* * Used for CTL_MSG_SERIALIZE, CTL_MSG_FINISH_IO, CTL_MSG_BAD_JUJU. */ struct ctl_ha_msg_scsi { struct ctl_ha_msg_hdr hdr; uint8_t cdb[CTL_MAX_CDBLEN]; /* CDB */ uint32_t tag_num; /* tag number */ ctl_tag_type tag_type; /* simple, ordered, etc. */ uint8_t scsi_status; /* SCSI status byte */ struct scsi_sense_data sense_data; /* sense data */ uint8_t sense_len; /* Returned sense length */ uint8_t sense_residual; /* sense residual length */ uint32_t residual; /* data residual length */ uint32_t fetd_status; /* trans status, set by FETD, 0 = good*/ struct ctl_lba_len lbalen; /* used for stats */ }; /* * Used for CTL_MSG_MANAGE_TASKS. */ struct ctl_ha_msg_task { struct ctl_ha_msg_hdr hdr; ctl_task_type task_action; /* Target Reset, Abort, etc. */ uint32_t tag_num; /* tag number */ ctl_tag_type tag_type; /* simple, ordered, etc. */ }; union ctl_ha_msg { struct ctl_ha_msg_hdr hdr; struct ctl_ha_msg_task task; struct ctl_ha_msg_scsi scsi; struct ctl_ha_msg_dt dt; struct ctl_ha_msg_pr pr; }; struct ctl_prio { struct ctl_io_hdr io_hdr; struct ctl_ha_msg_pr pr_msg; }; union ctl_io { struct ctl_io_hdr io_hdr; /* common to all I/O types */ struct ctl_scsiio scsiio; /* Normal SCSI commands */ struct ctl_taskio taskio; /* SCSI task management/reset */ struct ctl_prio presio; /* update per. res info on other SC */ }; #ifdef _KERNEL union ctl_io *ctl_alloc_io(void *pool_ref); union ctl_io *ctl_alloc_io_nowait(void *pool_ref); void ctl_free_io(union ctl_io *io); void ctl_zero_io(union ctl_io *io); void ctl_copy_io(union ctl_io *src, union ctl_io *dest); #endif /* _KERNEL */ #endif /* _CTL_IO_H_ */ /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_tpc.c =================================================================== --- head/sys/cam/ctl/ctl_tpc.c (revision 275057) +++ head/sys/cam/ctl/ctl_tpc.c (revision 275058) @@ -1,2243 +1,2241 @@ /*- * 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 #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; uint64_t lun; 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; uint8_t *buf; 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 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; }; extern struct ctl_softc *control_softc; 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. */ 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); } callout_schedule(&softc->tpc_timeout, hz); } void ctl_tpc_init(struct ctl_softc *softc) { 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->ctl_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->ctl_lock); } void ctl_tpc_lun_init(struct ctl_lun *lun) { TAILQ_INIT(&lun->tpc_lists); } void ctl_tpc_lun_shutdown(struct ctl_lun *lun) { 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_assert(&control_softc->ctl_lock, MA_OWNED); TAILQ_FOREACH_SAFE(token, &control_softc->tpc_tokens, links, ttoken) { if (token->lun != lun->lun || token->active) continue; TAILQ_REMOVE(&control_softc->tpc_tokens, token, links); free(token->params, M_CTL); free(token, M_CTL); } } int ctl_inquiry_evpd_tpc(struct ctl_scsiio *ctsio, int alloc_len) { 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; struct ctl_lun *lun; int data_len; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_sg_entries = 0; if (data_len < alloc_len) { ctsio->residual = alloc_len - data_len; ctsio->kern_data_len = data_len; ctsio->kern_total_len = data_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; ctsio->kern_sg_entries = 0; /* * 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(TPC_MAX_IOCHUNK_SIZE, 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; - ctsio->scsi_status = SCSI_STATUS_OK; + 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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; retval = CTL_RETVAL_COMPLETE; list_id = cdb->list_identifier; mtx_lock(&lun->lun_lock); list = tpc_find_list(lun, list_id, ctl_get_resindex(&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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; retval = CTL_RETVAL_COMPLETE; list_id = cdb->list_identifier; mtx_lock(&lun->lun_lock); list = tpc_find_list(lun, list_id, ctl_get_resindex(&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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_resindex(&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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_resindex(&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) { if (idx == 0xffff) { if (ss && list->lun->be_lun) *ss = list->lun->be_lun->blocksize; return (list->lun->lun); } if (idx >= list->ncscd) return (UINT64_MAX); return (tpcl_resolve(list->init_port, &list->cscd[idx], ss)); } 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, *prun; uint64_t sl, dl; off_t srclba, dstlba, numbytes, donebytes, roundbytes; int numlba; uint32_t srcblock, dstblock; if (list->stage == 1) { while ((tior = TAILQ_FIRST(&list->allio)) != NULL) { TAILQ_REMOVE(&list->allio, tior, links); ctl_free_io(tior->io); free(tior, M_CTL); } free(list->buf, M_CTL); if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x0d, /*ascq*/ 0x01, SSD_ELEM_NONE); 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]; sl = tpc_resolve(list, scsi_2btoul(seg->src_cscd), &srcblock); dl = tpc_resolve(list, scsi_2btoul(seg->dst_cscd), &dstblock); if (sl >= CTL_MAX_LUNS || dl >= CTL_MAX_LUNS) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x08, /*ascq*/ 0x04, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } sdstp = &list->cscd[scsi_2btoul(seg->src_cscd)].dtsp; if (scsi_3btoul(sdstp->block_length) != 0) srcblock = scsi_3btoul(sdstp->block_length); ddstp = &list->cscd[scsi_2btoul(seg->dst_cscd)].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_NONE); return (CTL_RETVAL_ERROR); } list->buf = malloc(numbytes, M_CTL, M_WAITOK); list->segbytes = numbytes; list->segsectors = numbytes / dstblock; donebytes = 0; TAILQ_INIT(&run); prun = &run; list->tbdio = 1; while (donebytes < numbytes) { roundbytes = MIN(numbytes - donebytes, TPC_MAX_IO_SIZE); tior = malloc(sizeof(*tior), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tior->run); tior->list = list; TAILQ_INSERT_TAIL(&list->allio, tior, links); tior->io = tpcl_alloc_io(); ctl_scsi_read_write(tior->io, /*data_ptr*/ &list->buf[donebytes], /*data_len*/ roundbytes, /*read_op*/ 1, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ srclba + donebytes / srcblock, /*num_blocks*/ roundbytes / srcblock, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tior->io->io_hdr.retries = 3; 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*/ &list->buf[donebytes], /*data_len*/ roundbytes, /*read_op*/ 0, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ dstlba + donebytes / dstblock, /*num_blocks*/ roundbytes / dstblock, /*tag_type*/ CTL_TAG_SIMPLE, /*control*/ 0); tiow->io->io_hdr.retries = 3; tiow->lun = dl; tiow->io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = tior; TAILQ_INSERT_TAIL(&tior->run, tiow, rlinks); TAILQ_INSERT_TAIL(prun, tior, rlinks); prun = &tior->run; donebytes += roundbytes; } 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; 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) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x0d, /*ascq*/ 0x01, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } else return (CTL_RETVAL_COMPLETE); } TAILQ_INIT(&list->allio); seg = (struct scsi_ec_segment_verify *)list->seg[list->curseg]; sl = tpc_resolve(list, scsi_2btoul(seg->src_cscd), NULL); if (sl >= CTL_MAX_LUNS) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x08, /*ascq*/ 0x04, 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->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; 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); } free(list->buf, M_CTL); if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x0d, /*ascq*/ 0x01, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } else return (CTL_RETVAL_COMPLETE); } TAILQ_INIT(&list->allio); seg = (struct scsi_ec_segment_register_key *)list->seg[list->curseg]; dl = tpc_resolve(list, scsi_2btoul(seg->dst_cscd), NULL); if (dl >= CTL_MAX_LUNS) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x08, /*ascq*/ 0x04, 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); list->buf = malloc(datalen, M_CTL, M_WAITOK); ctl_scsi_persistent_res_out(tio->io, list->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->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_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, *prun; int drange, srange; off_t doffset, soffset; off_t srclba, dstlba, numbytes, donebytes, roundbytes; uint32_t srcblock, dstblock; 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, M_CTL); } free(list->buf, M_CTL); if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x0d, /*ascq*/ 0x01, SSD_ELEM_NONE); 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; /* 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_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x0d, /*ascq*/ 0x04, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } srclba = scsi_8btou64(list->token->range[srange].lba) + soffset; numbytes = srcblock * omin(TPC_MAX_IOCHUNK_SIZE / srcblock, (scsi_4btoul(list->token->range[srange].length) - soffset)); dstlba = scsi_8btou64(list->range[drange].lba) + doffset; numbytes = omin(numbytes, dstblock * omin(TPC_MAX_IOCHUNK_SIZE / dstblock, (scsi_4btoul(list->range[drange].length) - doffset))); 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_NONE); return (CTL_RETVAL_ERROR); } list->buf = malloc(numbytes, M_CTL, M_WAITOK | (list->token == NULL ? M_ZERO : 0)); 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); prun = &run; list->tbdio = 1; TAILQ_INIT(&list->allio); while (donebytes < numbytes) { roundbytes = MIN(numbytes - donebytes, TPC_MAX_IO_SIZE); tior = malloc(sizeof(*tior), M_CTL, M_WAITOK | M_ZERO); TAILQ_INIT(&tior->run); tior->list = list; TAILQ_INSERT_TAIL(&list->allio, tior, links); tior->io = tpcl_alloc_io(); ctl_scsi_read_write(tior->io, /*data_ptr*/ &list->buf[donebytes], /*data_len*/ roundbytes, /*read_op*/ 1, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ srclba + donebytes / srcblock, /*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*/ &list->buf[donebytes], /*data_len*/ roundbytes, /*read_op*/ 0, /*byte2*/ 0, /*minimum_cdb_size*/ 0, /*lba*/ dstlba + donebytes / dstblock, /*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(prun, tior, rlinks); prun = &tior->run; donebytes += roundbytes; } 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); } free(list->buf, M_CTL); if (list->abort) { ctl_set_task_aborted(list->ctsio); return (CTL_RETVAL_ERROR); } else if (list->error) { ctl_set_sense(list->ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x0d, /*ascq*/ 0x01, SSD_ELEM_NONE); return (CTL_RETVAL_ERROR); } list->cursectors += list->segsectors; list->curbytes += list->segbytes; return (CTL_RETVAL_COMPLETE); } dstblock = list->lun->be_lun->blocksize; list->buf = malloc(dstblock, M_CTL, M_WAITOK | M_ZERO); 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*/ list->buf, /*data_len*/ dstblock, /*byte2*/ 0, /*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 scsi_ec_segment *seg; struct ctl_scsiio *ctsio = list->ctsio; int retval = CTL_RETVAL_COMPLETE; 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: ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_COPY_ABORTED, /*asc*/ 0x26, /*ascq*/ 0x09, 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(&control_softc->ctl_lock); if (--list->token->active == 0) list->token->last_active = time_uptime; mtx_unlock(&control_softc->ctl_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. */ -// if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) -// ctl_io_error_print(io, NULL); 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; 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 scsi_extended_copy *cdb; struct scsi_extended_copy_lid1_data *data; struct ctl_lun *lun; struct tpc_list *list, *tlist; uint8_t *ptr; char *value; int len, off, lencscd, lenseg, leninl, nseg; CTL_DEBUG_PRINT(("ctl_extended_copy_lid1\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; cdb = (struct scsi_extended_copy *)ctsio->cdb; len = scsi_4btoul(cdb->length); 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_data_resid = 0; 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 (len < sizeof(struct scsi_extended_copy_lid1_data) + lencscd + lenseg + leninl || leninl > TPC_MAX_INLINE) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); goto done; } 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 (lencscd + lenseg > TPC_MAX_LIST) { 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"); 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_resindex(&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[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; } list->seg[nseg] = (struct scsi_ec_segment *)(ptr + off); off += sizeof(struct scsi_ec_segment) + scsi_2btoul(list->seg[nseg]->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: ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_extended_copy_lid4(struct ctl_scsiio *ctsio) { struct scsi_extended_copy *cdb; struct scsi_extended_copy_lid4_data *data; struct ctl_lun *lun; struct tpc_list *list, *tlist; uint8_t *ptr; char *value; int len, off, lencscd, lenseg, leninl, nseg; CTL_DEBUG_PRINT(("ctl_extended_copy_lid4\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; cdb = (struct scsi_extended_copy *)ctsio->cdb; len = scsi_4btoul(cdb->length); 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_data_resid = 0; 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 (len < sizeof(struct scsi_extended_copy_lid4_data) + lencscd + lenseg + leninl || leninl > TPC_MAX_INLINE) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 2, /*bit_valid*/ 0, /*bit*/ 0); goto done; } 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 (lencscd + lenseg > TPC_MAX_LIST) { 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"); 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_resindex(&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[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; } list->seg[nseg] = (struct scsi_ec_segment *)(ptr + off); off += sizeof(struct scsi_ec_segment) + scsi_2btoul(list->seg[nseg]->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: 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 scsi_populate_token *cdb; struct scsi_populate_token_data *data; struct ctl_lun *lun; struct ctl_port *port; struct tpc_list *list, *tlist; struct tpc_token *token; int len, lendesc; CTL_DEBUG_PRINT(("ctl_populate_token\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; port = control_softc->ctl_ports[ctl_port_idx(ctsio->io_hdr.nexus.targ_port)]; 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_data_resid = 0; 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; lendesc = scsi_2btoul(data->range_descriptor_length); if (len < sizeof(struct scsi_populate_token_data) + lendesc) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 2, /*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)); */ 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; } 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_resindex(&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; else if (token->timeout > TPC_MAX_TOKEN_TIMEOUT) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 4, /*bit_valid*/ 0, /*bit*/ 0); } 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(&control_softc->ctl_lock); TAILQ_INSERT_TAIL(&control_softc->tpc_tokens, token, links); mtx_unlock(&control_softc->ctl_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); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_write_using_token(struct ctl_scsiio *ctsio) { struct scsi_write_using_token *cdb; struct scsi_write_using_token_data *data; struct ctl_lun *lun; struct tpc_list *list, *tlist; struct tpc_token *token; int len, lendesc; CTL_DEBUG_PRINT(("ctl_write_using_token\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; cdb = (struct scsi_write_using_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_data_resid = 0; 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; lendesc = scsi_2btoul(data->range_descriptor_length); if (len < sizeof(struct scsi_populate_token_data) + lendesc) { ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 0, /*field*/ 2, /*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)); */ 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_resindex(&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(&control_softc->ctl_lock); TAILQ_FOREACH(token, &control_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(&control_softc->ctl_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); ctl_done((union ctl_io *)ctsio); return (CTL_RETVAL_COMPLETE); } int ctl_receive_rod_token_information(struct ctl_scsiio *ctsio) { struct ctl_lun *lun; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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_resindex(&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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; 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_lun *lun; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; retval = CTL_RETVAL_COMPLETE; tokens = 0; mtx_lock(&control_softc->ctl_lock); TAILQ_FOREACH(token, &control_softc->tpc_tokens, links) tokens++; mtx_unlock(&control_softc->ctl_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; if (total_len < alloc_len) { ctsio->residual = alloc_len - total_len; ctsio->kern_data_len = total_len; ctsio->kern_total_len = total_len; } else { ctsio->residual = 0; ctsio->kern_data_len = alloc_len; ctsio->kern_total_len = alloc_len; } ctsio->kern_data_resid = 0; ctsio->kern_rel_offset = 0; data = (struct scsi_report_all_rod_tokens_data *)ctsio->kern_data_ptr; i = 0; mtx_lock(&control_softc->ctl_lock); TAILQ_FOREACH(token, &control_softc->tpc_tokens, links) { if (i >= tokens) break; memcpy(&data->rod_management_token_list[i * 96], token->token, 96); i++; } mtx_unlock(&control_softc->ctl_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/cam/ctl/scsi_ctl.c =================================================================== --- head/sys/cam/ctl/scsi_ctl.c (revision 275057) +++ head/sys/cam/ctl/scsi_ctl.c (revision 275058) @@ -1,2066 +1,2086 @@ /*- * Copyright (c) 2008, 2009 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/scsi_ctl.c#4 $ */ /* * Peripheral driver interface between CAM and CTL (CAM Target Layer). * * 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 struct ctlfe_softc { struct ctl_port port; path_id_t path_id; u_int maxio; struct cam_sim *sim; char port_name[DEV_IDLEN]; struct mtx lun_softc_mtx; STAILQ_HEAD(, ctlfe_lun_softc) lun_softc_list; STAILQ_ENTRY(ctlfe_softc) links; }; STAILQ_HEAD(, ctlfe_softc) ctlfe_softc_list; struct mtx ctlfe_list_mtx; static char ctlfe_mtx_desc[] = "ctlfelist"; #ifdef CTLFE_INIT_ENABLE static int ctlfe_max_targets = 1; static int ctlfe_num_targets = 0; #endif typedef enum { CTLFE_LUN_NONE = 0x00, CTLFE_LUN_WILDCARD = 0x01 } ctlfe_lun_flags; struct ctlfe_lun_softc { struct ctlfe_softc *parent_softc; struct cam_periph *periph; ctlfe_lun_flags flags; uint64_t ccbs_alloced; uint64_t ccbs_freed; uint64_t ctios_sent; uint64_t ctios_returned; uint64_t atios_sent; uint64_t atios_returned; uint64_t inots_sent; uint64_t inots_returned; /* bus_dma_tag_t dma_tag; */ TAILQ_HEAD(, ccb_hdr) work_queue; STAILQ_ENTRY(ctlfe_lun_softc) links; }; typedef enum { CTLFE_CMD_NONE = 0x00, CTLFE_CMD_PIECEWISE = 0x01 } ctlfe_cmd_flags; /* * The size limit of this structure is CTL_PORT_PRIV_SIZE, from ctl_io.h. * Currently that is 600 bytes. */ struct ctlfe_lun_cmd_info { int cur_transfer_index; size_t cur_transfer_off; ctlfe_cmd_flags flags; /* * XXX KDM struct bus_dma_segment is 8 bytes on i386, and 16 * bytes on amd64. So with 32 elements, this is 256 bytes on * i386 and 512 bytes on amd64. */ #define CTLFE_MAX_SEGS 32 bus_dma_segment_t cam_sglist[CTLFE_MAX_SEGS]; }; CTASSERT(sizeof(struct ctlfe_lun_cmd_info) <= CTL_PORT_PRIV_SIZE); /* * When we register the adapter/bus, request that this many ctl_ios be * allocated. This should be the maximum supported by the adapter, but we * currently don't have a way to get that back from the path inquiry. * XXX KDM add that to the path inquiry. */ #define CTLFE_REQ_CTL_IO 4096 /* * Number of Accept Target I/O CCBs to allocate and queue down to the * adapter per LUN. * XXX KDM should this be controlled by CTL? */ #define CTLFE_ATIO_PER_LUN 1024 /* * Number of Immediate Notify CCBs (used for aborts, resets, etc.) to * allocate and queue down to the adapter per LUN. * XXX KDM should this be controlled by CTL? */ #define CTLFE_IN_PER_LUN 1024 /* * Timeout (in seconds) on CTIO CCB allocation for doing a DMA or sending * status to the initiator. The SIM is expected to have its own timeouts, * so we're not putting this timeout around the CCB execution time. The * SIM should timeout and let us know if it has an issue. */ #define CTLFE_DMA_TIMEOUT 60 /* * Turn this on to enable extra debugging prints. */ #if 0 #define CTLFE_DEBUG #endif /* * Use randomly assigned WWNN/WWPN values. This is to work around an issue * in the FreeBSD initiator that makes it unable to rescan the target if * the target gets rebooted and the WWNN/WWPN stay the same. */ #if 0 #define RANDOM_WWNN #endif MALLOC_DEFINE(M_CTLFE, "CAM CTL FE", "CAM CTL FE interface"); #define io_ptr ppriv_ptr0 /* This is only used in the CTIO */ #define ccb_atio ppriv_ptr1 int ctlfeinitialize(void); void ctlfeshutdown(void); static periph_init_t ctlfeperiphinit; static void ctlfeasync(void *callback_arg, uint32_t code, struct cam_path *path, void *arg); static periph_ctor_t ctlferegister; static periph_oninv_t ctlfeoninvalidate; static periph_dtor_t ctlfecleanup; static periph_start_t ctlfestart; static void ctlfedone(struct cam_periph *periph, union ccb *done_ccb); static void ctlfe_onoffline(void *arg, int online); static void ctlfe_online(void *arg); static void ctlfe_offline(void *arg); static int ctlfe_lun_enable(void *arg, struct ctl_id targ_id, int lun_id); static int ctlfe_lun_disable(void *arg, struct ctl_id targ_id, int lun_id); static void ctlfe_dump_sim(struct cam_sim *sim); static void ctlfe_dump_queue(struct ctlfe_lun_softc *softc); static void ctlfe_datamove(union ctl_io *io); static void ctlfe_done(union ctl_io *io); static void ctlfe_dump(void); static struct periph_driver ctlfe_driver = { ctlfeperiphinit, "ctl", TAILQ_HEAD_INITIALIZER(ctlfe_driver.units), /*generation*/ 0, CAM_PERIPH_DRV_EARLY }; static struct ctl_frontend ctlfe_frontend = { .name = "camtgt", .init = ctlfeinitialize, .fe_dump = ctlfe_dump, .shutdown = ctlfeshutdown, }; CTL_FRONTEND_DECLARE(ctlfe, ctlfe_frontend); extern struct ctl_softc *control_softc; void ctlfeshutdown(void) { return; } int ctlfeinitialize(void) { STAILQ_INIT(&ctlfe_softc_list); mtx_init(&ctlfe_list_mtx, ctlfe_mtx_desc, NULL, MTX_DEF); periphdriver_register(&ctlfe_driver); return (0); } void ctlfeperiphinit(void) { cam_status status; status = xpt_register_async(AC_PATH_REGISTERED | AC_PATH_DEREGISTERED | AC_CONTRACT, ctlfeasync, NULL, NULL); if (status != CAM_REQ_CMP) { printf("ctl: Failed to attach async callback due to CAM " "status 0x%x!\n", status); } } static void ctlfeasync(void *callback_arg, uint32_t code, struct cam_path *path, void *arg) { struct ctlfe_softc *softc; #ifdef CTLFEDEBUG printf("%s: entered\n", __func__); #endif mtx_lock(&ctlfe_list_mtx); STAILQ_FOREACH(softc, &ctlfe_softc_list, links) { if (softc->path_id == xpt_path_path_id(path)) break; } mtx_unlock(&ctlfe_list_mtx); /* * When a new path gets registered, and it is capable of target * mode, go ahead and attach. Later on, we may need to be more * selective, but for now this will be sufficient. */ switch (code) { case AC_PATH_REGISTERED: { struct ctl_port *port; struct ccb_pathinq *cpi; int retval; cpi = (struct ccb_pathinq *)arg; /* Don't attach if it doesn't support target mode */ if ((cpi->target_sprt & PIT_PROCESSOR) == 0) { #ifdef CTLFEDEBUG printf("%s: SIM %s%d doesn't support target mode\n", __func__, cpi->dev_name, cpi->unit_number); #endif break; } if (softc != NULL) { #ifdef CTLFEDEBUG printf("%s: CTL port for CAM path %u already exists\n", __func__, xpt_path_path_id(path)); #endif break; } #ifdef CTLFE_INIT_ENABLE if (ctlfe_num_targets >= ctlfe_max_targets) { union ccb *ccb; ccb = (union ccb *)malloc(sizeof(*ccb), M_TEMP, M_NOWAIT | M_ZERO); if (ccb == NULL) { printf("%s: unable to malloc CCB!\n", __func__); return; } xpt_setup_ccb(&ccb->ccb_h, path, CAM_PRIORITY_NONE); ccb->ccb_h.func_code = XPT_SET_SIM_KNOB; ccb->knob.xport_specific.valid = KNOB_VALID_ROLE; ccb->knob.xport_specific.fc.role = KNOB_ROLE_INITIATOR; xpt_action(ccb); if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) { printf("%s: SIM %s%d (path id %d) initiator " "enable failed with status %#x\n", __func__, cpi->dev_name, cpi->unit_number, cpi->ccb_h.path_id, ccb->ccb_h.status); } else { printf("%s: SIM %s%d (path id %d) initiator " "enable succeeded\n", __func__, cpi->dev_name, cpi->unit_number, cpi->ccb_h.path_id); } free(ccb, M_TEMP); break; } else { ctlfe_num_targets++; } printf("%s: ctlfe_num_targets = %d\n", __func__, ctlfe_num_targets); #endif /* CTLFE_INIT_ENABLE */ /* * We're in an interrupt context here, so we have to * use M_NOWAIT. Of course this means trouble if we * can't allocate memory. */ softc = malloc(sizeof(*softc), M_CTLFE, M_NOWAIT | M_ZERO); if (softc == NULL) { printf("%s: unable to malloc %zd bytes for softc\n", __func__, sizeof(*softc)); return; } softc->path_id = cpi->ccb_h.path_id; softc->sim = xpt_path_sim(path); if (cpi->maxio != 0) softc->maxio = cpi->maxio; else softc->maxio = DFLTPHYS; mtx_init(&softc->lun_softc_mtx, "LUN softc mtx", NULL, MTX_DEF); STAILQ_INIT(&softc->lun_softc_list); port = &softc->port; port->frontend = &ctlfe_frontend; /* * XXX KDM should we be more accurate here ? */ if (cpi->transport == XPORT_FC) port->port_type = CTL_PORT_FC; else if (cpi->transport == XPORT_SAS) port->port_type = CTL_PORT_SAS; else port->port_type = CTL_PORT_SCSI; /* XXX KDM what should the real number be here? */ port->num_requested_ctl_io = 4096; snprintf(softc->port_name, sizeof(softc->port_name), "%s%d", cpi->dev_name, cpi->unit_number); /* * XXX KDM it would be nice to allocate storage in the * frontend structure itself. */ port->port_name = softc->port_name; port->physical_port = cpi->bus_id; port->virtual_port = 0; port->port_online = ctlfe_online; port->port_offline = ctlfe_offline; port->onoff_arg = softc; port->lun_enable = ctlfe_lun_enable; port->lun_disable = ctlfe_lun_disable; port->targ_lun_arg = softc; port->fe_datamove = ctlfe_datamove; port->fe_done = ctlfe_done; /* * XXX KDM the path inquiry doesn't give us the maximum * number of targets supported. */ port->max_targets = cpi->max_target; port->max_target_id = cpi->max_target; /* * XXX KDM need to figure out whether we're the master or * slave. */ #ifdef CTLFEDEBUG printf("%s: calling ctl_port_register() for %s%d\n", __func__, cpi->dev_name, cpi->unit_number); #endif retval = ctl_port_register(port); if (retval != 0) { printf("%s: ctl_port_register() failed with " "error %d!\n", __func__, retval); mtx_destroy(&softc->lun_softc_mtx); free(softc, M_CTLFE); break; } else { mtx_lock(&ctlfe_list_mtx); STAILQ_INSERT_TAIL(&ctlfe_softc_list, softc, links); mtx_unlock(&ctlfe_list_mtx); } break; } case AC_PATH_DEREGISTERED: { if (softc != NULL) { /* * XXX KDM are we certain at this point that there * are no outstanding commands for this frontend? */ mtx_lock(&ctlfe_list_mtx); STAILQ_REMOVE(&ctlfe_softc_list, softc, ctlfe_softc, links); mtx_unlock(&ctlfe_list_mtx); ctl_port_deregister(&softc->port); mtx_destroy(&softc->lun_softc_mtx); free(softc, M_CTLFE); } break; } case AC_CONTRACT: { struct ac_contract *ac; ac = (struct ac_contract *)arg; switch (ac->contract_number) { case AC_CONTRACT_DEV_CHG: { struct ac_device_changed *dev_chg; int retval; dev_chg = (struct ac_device_changed *)ac->contract_data; printf("%s: WWPN %#jx port 0x%06x path %u target %u %s\n", __func__, dev_chg->wwpn, dev_chg->port, xpt_path_path_id(path), dev_chg->target, (dev_chg->arrived == 0) ? "left" : "arrived"); if (softc == NULL) { printf("%s: CTL port for CAM path %u not " "found!\n", __func__, xpt_path_path_id(path)); break; } if (dev_chg->arrived != 0) { retval = ctl_add_initiator(&softc->port, dev_chg->target, dev_chg->wwpn, NULL); } else { retval = ctl_remove_initiator(&softc->port, dev_chg->target); } if (retval < 0) { printf("%s: could not %s port %d iid %u " "WWPN %#jx!\n", __func__, (dev_chg->arrived != 0) ? "add" : "remove", softc->port.targ_port, dev_chg->target, (uintmax_t)dev_chg->wwpn); } break; } default: printf("%s: unsupported contract number %ju\n", __func__, (uintmax_t)ac->contract_number); break; } break; } default: break; } } static cam_status ctlferegister(struct cam_periph *periph, void *arg) { struct ctlfe_softc *bus_softc; struct ctlfe_lun_softc *softc; union ccb en_lun_ccb; cam_status status; int i; softc = (struct ctlfe_lun_softc *)arg; bus_softc = softc->parent_softc; TAILQ_INIT(&softc->work_queue); softc->periph = periph; periph->softc = softc; xpt_setup_ccb(&en_lun_ccb.ccb_h, periph->path, CAM_PRIORITY_NONE); en_lun_ccb.ccb_h.func_code = XPT_EN_LUN; en_lun_ccb.cel.grp6_len = 0; en_lun_ccb.cel.grp7_len = 0; en_lun_ccb.cel.enable = 1; xpt_action(&en_lun_ccb); status = (en_lun_ccb.ccb_h.status & CAM_STATUS_MASK); if (status != CAM_REQ_CMP) { xpt_print(periph->path, "%s: Enable LUN failed, status 0x%x\n", __func__, en_lun_ccb.ccb_h.status); return (status); } status = CAM_REQ_CMP; for (i = 0; i < CTLFE_ATIO_PER_LUN; i++) { union ccb *new_ccb; union ctl_io *new_io; new_ccb = (union ccb *)malloc(sizeof(*new_ccb), M_CTLFE, M_ZERO|M_NOWAIT); if (new_ccb == NULL) { status = CAM_RESRC_UNAVAIL; break; } new_io = ctl_alloc_io_nowait(bus_softc->port.ctl_pool_ref); if (new_io == NULL) { free(new_ccb, M_CTLFE); status = CAM_RESRC_UNAVAIL; break; } new_ccb->ccb_h.io_ptr = new_io; xpt_setup_ccb(&new_ccb->ccb_h, periph->path, /*priority*/ 1); new_ccb->ccb_h.func_code = XPT_ACCEPT_TARGET_IO; new_ccb->ccb_h.cbfcnp = ctlfedone; new_ccb->ccb_h.flags |= CAM_UNLOCKED; xpt_action(new_ccb); softc->atios_sent++; status = new_ccb->ccb_h.status; if ((status & CAM_STATUS_MASK) != CAM_REQ_INPROG) { ctl_free_io(new_io); free(new_ccb, M_CTLFE); break; } } status = cam_periph_acquire(periph); if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) { xpt_print(periph->path, "%s: could not acquire reference " "count, status = %#x\n", __func__, status); return (status); } if (i == 0) { xpt_print(periph->path, "%s: could not allocate ATIO CCBs, " "status 0x%x\n", __func__, status); return (CAM_REQ_CMP_ERR); } for (i = 0; i < CTLFE_IN_PER_LUN; i++) { union ccb *new_ccb; union ctl_io *new_io; new_ccb = (union ccb *)malloc(sizeof(*new_ccb), M_CTLFE, M_ZERO|M_NOWAIT); if (new_ccb == NULL) { status = CAM_RESRC_UNAVAIL; break; } new_io = ctl_alloc_io_nowait(bus_softc->port.ctl_pool_ref); if (new_io == NULL) { free(new_ccb, M_CTLFE); status = CAM_RESRC_UNAVAIL; break; } new_ccb->ccb_h.io_ptr = new_io; xpt_setup_ccb(&new_ccb->ccb_h, periph->path, /*priority*/ 1); new_ccb->ccb_h.func_code = XPT_IMMEDIATE_NOTIFY; new_ccb->ccb_h.cbfcnp = ctlfedone; new_ccb->ccb_h.flags |= CAM_UNLOCKED; xpt_action(new_ccb); softc->inots_sent++; status = new_ccb->ccb_h.status; if ((status & CAM_STATUS_MASK) != CAM_REQ_INPROG) { /* * Note that we don't free the CCB here. If the * status is not CAM_REQ_INPROG, then we're * probably talking to a SIM that says it is * target-capable but doesn't support the * XPT_IMMEDIATE_NOTIFY CCB. i.e. it supports the * older API. In that case, it'll call xpt_done() * on the CCB, and we need to free it in our done * routine as a result. */ break; } } if ((i == 0) || (status != CAM_REQ_INPROG)) { xpt_print(periph->path, "%s: could not allocate immediate " "notify CCBs, status 0x%x\n", __func__, status); return (CAM_REQ_CMP_ERR); } return (CAM_REQ_CMP); } static void ctlfeoninvalidate(struct cam_periph *periph) { union ccb en_lun_ccb; cam_status status; struct ctlfe_softc *bus_softc; struct ctlfe_lun_softc *softc; softc = (struct ctlfe_lun_softc *)periph->softc; xpt_setup_ccb(&en_lun_ccb.ccb_h, periph->path, CAM_PRIORITY_NONE); en_lun_ccb.ccb_h.func_code = XPT_EN_LUN; en_lun_ccb.cel.grp6_len = 0; en_lun_ccb.cel.grp7_len = 0; en_lun_ccb.cel.enable = 0; xpt_action(&en_lun_ccb); status = (en_lun_ccb.ccb_h.status & CAM_STATUS_MASK); if (status != CAM_REQ_CMP) { xpt_print(periph->path, "%s: Disable LUN failed, status 0x%x\n", __func__, en_lun_ccb.ccb_h.status); /* * XXX KDM what do we do now? */ } xpt_print(periph->path, "LUN removed, %ju ATIOs outstanding, %ju " "INOTs outstanding, %d refs\n", softc->atios_sent - softc->atios_returned, softc->inots_sent - softc->inots_returned, periph->refcount); bus_softc = softc->parent_softc; mtx_lock(&bus_softc->lun_softc_mtx); STAILQ_REMOVE(&bus_softc->lun_softc_list, softc, ctlfe_lun_softc, links); mtx_unlock(&bus_softc->lun_softc_mtx); } static void ctlfecleanup(struct cam_periph *periph) { struct ctlfe_lun_softc *softc; xpt_print(periph->path, "%s: Called\n", __func__); softc = (struct ctlfe_lun_softc *)periph->softc; /* * XXX KDM is there anything else that needs to be done here? */ free(softc, M_CTLFE); } static void ctlfedata(struct ctlfe_lun_softc *softc, union ctl_io *io, ccb_flags *flags, uint8_t **data_ptr, uint32_t *dxfer_len, u_int16_t *sglist_cnt) { struct ctlfe_softc *bus_softc; struct ctlfe_lun_cmd_info *cmd_info; struct ctl_sg_entry *ctl_sglist; bus_dma_segment_t *cam_sglist; size_t off; int i, idx; cmd_info = (struct ctlfe_lun_cmd_info *)io->io_hdr.port_priv; bus_softc = softc->parent_softc; /* * Set the direction, relative to the initiator. */ *flags &= ~CAM_DIR_MASK; if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) *flags |= CAM_DIR_IN; else *flags |= CAM_DIR_OUT; *flags &= ~CAM_DATA_MASK; idx = cmd_info->cur_transfer_index; off = cmd_info->cur_transfer_off; cmd_info->flags &= ~CTLFE_CMD_PIECEWISE; if (io->scsiio.kern_sg_entries == 0) { /* No S/G list. */ *data_ptr = io->scsiio.kern_data_ptr + off; if (io->scsiio.kern_data_len - off <= bus_softc->maxio) { *dxfer_len = io->scsiio.kern_data_len - off; } else { *dxfer_len = bus_softc->maxio; cmd_info->cur_transfer_index = -1; cmd_info->cur_transfer_off = bus_softc->maxio; cmd_info->flags |= CTLFE_CMD_PIECEWISE; } *sglist_cnt = 0; if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) *flags |= CAM_DATA_PADDR; else *flags |= CAM_DATA_VADDR; } else { /* S/G list with physical or virtual pointers. */ ctl_sglist = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr; cam_sglist = cmd_info->cam_sglist; *dxfer_len = 0; for (i = 0; i < io->scsiio.kern_sg_entries - idx; i++) { cam_sglist[i].ds_addr = (bus_addr_t)ctl_sglist[i + idx].addr + off; if (ctl_sglist[i + idx].len - off <= bus_softc->maxio - *dxfer_len) { cam_sglist[i].ds_len = ctl_sglist[idx + i].len - off; *dxfer_len += cam_sglist[i].ds_len; } else { cam_sglist[i].ds_len = bus_softc->maxio - *dxfer_len; cmd_info->cur_transfer_index = idx + i; cmd_info->cur_transfer_off = cam_sglist[i].ds_len + off; cmd_info->flags |= CTLFE_CMD_PIECEWISE; *dxfer_len += cam_sglist[i].ds_len; if (ctl_sglist[i].len != 0) i++; break; } if (i == (CTLFE_MAX_SEGS - 1) && idx + i < (io->scsiio.kern_sg_entries - 1)) { cmd_info->cur_transfer_index = idx + i + 1; cmd_info->cur_transfer_off = 0; cmd_info->flags |= CTLFE_CMD_PIECEWISE; i++; break; } off = 0; } *sglist_cnt = i; if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) *flags |= CAM_DATA_SG_PADDR; else *flags |= CAM_DATA_SG; *data_ptr = (uint8_t *)cam_sglist; } } static void ctlfestart(struct cam_periph *periph, union ccb *start_ccb) { struct ctlfe_lun_softc *softc; struct ctlfe_lun_cmd_info *cmd_info; struct ccb_hdr *ccb_h; struct ccb_accept_tio *atio; struct ccb_scsiio *csio; uint8_t *data_ptr; uint32_t dxfer_len; ccb_flags flags; union ctl_io *io; uint8_t scsi_status; softc = (struct ctlfe_lun_softc *)periph->softc; softc->ccbs_alloced++; ccb_h = TAILQ_FIRST(&softc->work_queue); if (ccb_h == NULL) { softc->ccbs_freed++; xpt_release_ccb(start_ccb); return; } /* Take the ATIO off the work queue */ TAILQ_REMOVE(&softc->work_queue, ccb_h, periph_links.tqe); atio = (struct ccb_accept_tio *)ccb_h; io = (union ctl_io *)ccb_h->io_ptr; csio = &start_ccb->csio; flags = atio->ccb_h.flags & (CAM_DIS_DISCONNECT|CAM_TAG_ACTION_VALID|CAM_DIR_MASK); + cmd_info = (struct ctlfe_lun_cmd_info *)io->io_hdr.port_priv; + cmd_info->cur_transfer_index = 0; + cmd_info->cur_transfer_off = 0; + cmd_info->flags = 0; if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED) { /* * Datamove call, we need to setup the S/G list. */ - cmd_info = (struct ctlfe_lun_cmd_info *) - io->io_hdr.port_priv; - bzero(cmd_info, sizeof(*cmd_info)); scsi_status = 0; csio->cdb_len = atio->cdb_len; ctlfedata(softc, io, &flags, &data_ptr, &dxfer_len, &csio->sglist_cnt); io->scsiio.ext_data_filled += dxfer_len; if (io->scsiio.ext_data_filled > io->scsiio.kern_total_len) { xpt_print(periph->path, "%s: tag 0x%04x " "fill len %u > total %u\n", __func__, io->scsiio.tag_num, io->scsiio.ext_data_filled, io->scsiio.kern_total_len); } } else { /* * We're done, send status back. */ if ((io->io_hdr.flags & CTL_FLAG_ABORT) && (io->io_hdr.flags & CTL_FLAG_ABORT_STATUS) == 0) { io->io_hdr.flags &= ~CTL_FLAG_STATUS_QUEUED; /* * If this command was aborted, we don't * need to send status back to the SIM. * Just free the CTIO and ctl_io, and * recycle the ATIO back to the SIM. */ xpt_print(periph->path, "%s: aborted " "command 0x%04x discarded\n", __func__, io->scsiio.tag_num); /* * For a wildcard attachment, commands can * come in with a specific target/lun. Reset * the target and LUN fields back to the * wildcard values before we send them back * down to the SIM. The SIM has a wildcard * LUN enabled, not whatever target/lun * these happened to be. */ if (softc->flags & CTLFE_LUN_WILDCARD) { atio->ccb_h.target_id = CAM_TARGET_WILDCARD; atio->ccb_h.target_lun = CAM_LUN_WILDCARD; } if ((atio->ccb_h.status & CAM_DEV_QFRZN) != 0) { cam_release_devq(periph->path, /*relsim_flags*/0, /*reduction*/0, /*timeout*/0, /*getcount_only*/0); atio->ccb_h.status &= ~CAM_DEV_QFRZN; } if (atio->ccb_h.func_code != XPT_ACCEPT_TARGET_IO) { xpt_print(periph->path, "%s: func_code " "is %#x\n", __func__, atio->ccb_h.func_code); } start_ccb->ccb_h.func_code = XPT_ABORT; start_ccb->cab.abort_ccb = (union ccb *)atio; /* Tell the SIM that we've aborted this ATIO */ xpt_action(start_ccb); softc->ccbs_freed++; xpt_release_ccb(start_ccb); /* * Send the ATIO back down to the SIM. */ xpt_action((union ccb *)atio); softc->atios_sent++; /* * If we still have work to do, ask for * another CCB. Otherwise, deactivate our * callout. */ if (!TAILQ_EMPTY(&softc->work_queue)) xpt_schedule(periph, /*priority*/ 1); return; } - + data_ptr = NULL; + dxfer_len = 0; + csio->sglist_cnt = 0; + scsi_status = 0; + } + if ((io->io_hdr.flags & CTL_FLAG_STATUS_QUEUED) && + (cmd_info->flags & CTLFE_CMD_PIECEWISE) == 0 && + ((io->io_hdr.flags & CTL_FLAG_DMA_QUEUED) == 0 || + io->io_hdr.status == CTL_SUCCESS)) { + io->io_hdr.flags |= CTL_FLAG_STATUS_SENT; flags |= CAM_SEND_STATUS; scsi_status = io->scsiio.scsi_status; csio->sense_len = io->scsiio.sense_len; - data_ptr = NULL; - dxfer_len = 0; #ifdef CTLFEDEBUG printf("%s: tag %04x status %x\n", __func__, atio->tag_id, io->io_hdr.status); #endif - csio->sglist_cnt = 0; if (csio->sense_len != 0) { csio->sense_data = io->scsiio.sense_data; flags |= CAM_SEND_SENSE; } else if (scsi_status == SCSI_STATUS_CHECK_COND) { xpt_print(periph->path, "%s: check condition " "with no sense\n", __func__); } } #ifdef CTLFEDEBUG printf("%s: %s: tag %04x flags %x ptr %p len %u\n", __func__, (flags & CAM_SEND_STATUS) ? "done" : "datamove", atio->tag_id, flags, data_ptr, dxfer_len); #endif /* * Valid combinations: * - CAM_SEND_STATUS, CAM_DATA_SG = 0, dxfer_len = 0, * sglist_cnt = 0 * - CAM_SEND_STATUS = 0, CAM_DATA_SG = 0, dxfer_len != 0, * sglist_cnt = 0 * - CAM_SEND_STATUS = 0, CAM_DATA_SG, dxfer_len != 0, * sglist_cnt != 0 */ #ifdef CTLFEDEBUG if (((flags & CAM_SEND_STATUS) && (((flags & CAM_DATA_SG) != 0) || (dxfer_len != 0) || (csio->sglist_cnt != 0))) || (((flags & CAM_SEND_STATUS) == 0) && (dxfer_len == 0)) || ((flags & CAM_DATA_SG) && (csio->sglist_cnt == 0)) || (((flags & CAM_DATA_SG) == 0) && (csio->sglist_cnt != 0))) { printf("%s: tag %04x cdb %02x flags %#x dxfer_len " "%d sg %u\n", __func__, atio->tag_id, atio->cdb_io.cdb_bytes[0], flags, dxfer_len, csio->sglist_cnt); printf("%s: tag %04x io status %#x\n", __func__, atio->tag_id, io->io_hdr.status); } #endif cam_fill_ctio(csio, /*retries*/ 2, ctlfedone, flags, (flags & CAM_TAG_ACTION_VALID) ? MSG_SIMPLE_Q_TAG : 0, atio->tag_id, atio->init_id, scsi_status, /*data_ptr*/ data_ptr, /*dxfer_len*/ dxfer_len, /*timeout*/ 5 * 1000); start_ccb->ccb_h.flags |= CAM_UNLOCKED; start_ccb->ccb_h.ccb_atio = atio; if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED) io->io_hdr.flags |= CTL_FLAG_DMA_INPROG; io->io_hdr.flags &= ~(CTL_FLAG_DMA_QUEUED | CTL_FLAG_STATUS_QUEUED); softc->ctios_sent++; cam_periph_unlock(periph); xpt_action(start_ccb); cam_periph_lock(periph); if ((atio->ccb_h.status & CAM_DEV_QFRZN) != 0) { cam_release_devq(periph->path, /*relsim_flags*/0, /*reduction*/0, /*timeout*/0, /*getcount_only*/0); atio->ccb_h.status &= ~CAM_DEV_QFRZN; } /* * If we still have work to do, ask for another CCB. */ if (!TAILQ_EMPTY(&softc->work_queue)) xpt_schedule(periph, /*priority*/ 1); } static void ctlfe_free_ccb(struct cam_periph *periph, union ccb *ccb) { struct ctlfe_lun_softc *softc; softc = (struct ctlfe_lun_softc *)periph->softc; switch (ccb->ccb_h.func_code) { case XPT_ACCEPT_TARGET_IO: softc->atios_returned++; break; case XPT_IMMEDIATE_NOTIFY: case XPT_NOTIFY_ACKNOWLEDGE: softc->inots_returned++; break; default: break; } ctl_free_io(ccb->ccb_h.io_ptr); free(ccb, M_CTLFE); KASSERT(softc->atios_returned <= softc->atios_sent, ("%s: " "atios_returned %ju > atios_sent %ju", __func__, softc->atios_returned, softc->atios_sent)); KASSERT(softc->inots_returned <= softc->inots_sent, ("%s: " "inots_returned %ju > inots_sent %ju", __func__, softc->inots_returned, softc->inots_sent)); /* * If we have received all of our CCBs, we can release our * reference on the peripheral driver. It will probably go away * now. */ if ((softc->atios_returned == softc->atios_sent) && (softc->inots_returned == softc->inots_sent)) { cam_periph_release_locked(periph); } } static int ctlfe_adjust_cdb(struct ccb_accept_tio *atio, uint32_t offset) { uint64_t lba; uint32_t num_blocks, nbc; uint8_t *cmdbyt = (atio->ccb_h.flags & CAM_CDB_POINTER)? atio->cdb_io.cdb_ptr : atio->cdb_io.cdb_bytes; nbc = offset >> 9; /* ASSUMING 512 BYTE BLOCKS */ switch (cmdbyt[0]) { case READ_6: case WRITE_6: { struct scsi_rw_6 *cdb = (struct scsi_rw_6 *)cmdbyt; lba = scsi_3btoul(cdb->addr); lba &= 0x1fffff; num_blocks = cdb->length; if (num_blocks == 0) num_blocks = 256; lba += nbc; num_blocks -= nbc; scsi_ulto3b(lba, cdb->addr); cdb->length = num_blocks; break; } case READ_10: case WRITE_10: { struct scsi_rw_10 *cdb = (struct scsi_rw_10 *)cmdbyt; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_2btoul(cdb->length); lba += nbc; num_blocks -= nbc; scsi_ulto4b(lba, cdb->addr); scsi_ulto2b(num_blocks, cdb->length); break; } case READ_12: case WRITE_12: { struct scsi_rw_12 *cdb = (struct scsi_rw_12 *)cmdbyt; lba = scsi_4btoul(cdb->addr); num_blocks = scsi_4btoul(cdb->length); lba += nbc; num_blocks -= nbc; scsi_ulto4b(lba, cdb->addr); scsi_ulto4b(num_blocks, cdb->length); break; } case READ_16: case WRITE_16: case WRITE_ATOMIC_16: { struct scsi_rw_16 *cdb = (struct scsi_rw_16 *)cmdbyt; lba = scsi_8btou64(cdb->addr); num_blocks = scsi_4btoul(cdb->length); lba += nbc; num_blocks -= nbc; scsi_u64to8b(lba, cdb->addr); scsi_ulto4b(num_blocks, cdb->length); break; } default: return -1; } return (0); } static void ctlfedone(struct cam_periph *periph, union ccb *done_ccb) { struct ctlfe_lun_softc *softc; struct ctlfe_softc *bus_softc; struct ccb_accept_tio *atio = NULL; union ctl_io *io = NULL; struct mtx *mtx; KASSERT((done_ccb->ccb_h.flags & CAM_UNLOCKED) != 0, ("CCB in ctlfedone() without CAM_UNLOCKED flag")); #ifdef CTLFE_DEBUG printf("%s: entered, func_code = %#x\n", __func__, done_ccb->ccb_h.func_code); #endif softc = (struct ctlfe_lun_softc *)periph->softc; bus_softc = softc->parent_softc; mtx = cam_periph_mtx(periph); mtx_lock(mtx); /* * If the peripheral is invalid, ATIOs and immediate notify CCBs * need to be freed. Most of the ATIOs and INOTs that come back * will be CCBs that are being returned from the SIM as a result of * our disabling the LUN. * * Other CCB types are handled in their respective cases below. */ if (periph->flags & CAM_PERIPH_INVALID) { switch (done_ccb->ccb_h.func_code) { case XPT_ACCEPT_TARGET_IO: case XPT_IMMEDIATE_NOTIFY: case XPT_NOTIFY_ACKNOWLEDGE: ctlfe_free_ccb(periph, done_ccb); goto out; default: break; } } switch (done_ccb->ccb_h.func_code) { case XPT_ACCEPT_TARGET_IO: { atio = &done_ccb->atio; softc->atios_returned++; resubmit: /* * Allocate a ctl_io, pass it to CTL, and wait for the * datamove or done. */ mtx_unlock(mtx); io = done_ccb->ccb_h.io_ptr; ctl_zero_io(io); /* Save pointers on both sides */ io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr = done_ccb; done_ccb->ccb_h.io_ptr = io; /* * Only SCSI I/O comes down this path, resets, etc. come * down the immediate notify path below. */ io->io_hdr.io_type = CTL_IO_SCSI; io->io_hdr.nexus.initid.id = atio->init_id; io->io_hdr.nexus.targ_port = bus_softc->port.targ_port; io->io_hdr.nexus.targ_target.id = atio->ccb_h.target_id; io->io_hdr.nexus.targ_lun = atio->ccb_h.target_lun; io->scsiio.tag_num = atio->tag_id; switch (atio->tag_action) { case CAM_TAG_ACTION_NONE: io->scsiio.tag_type = CTL_TAG_UNTAGGED; break; case MSG_SIMPLE_TASK: io->scsiio.tag_type = CTL_TAG_SIMPLE; break; case MSG_HEAD_OF_QUEUE_TASK: io->scsiio.tag_type = CTL_TAG_HEAD_OF_QUEUE; break; case MSG_ORDERED_TASK: io->scsiio.tag_type = CTL_TAG_ORDERED; break; case MSG_ACA_TASK: io->scsiio.tag_type = CTL_TAG_ACA; break; default: io->scsiio.tag_type = CTL_TAG_UNTAGGED; printf("%s: unhandled tag type %#x!!\n", __func__, atio->tag_action); break; } if (atio->cdb_len > sizeof(io->scsiio.cdb)) { printf("%s: WARNING: CDB len %d > ctl_io space %zd\n", __func__, atio->cdb_len, sizeof(io->scsiio.cdb)); } io->scsiio.cdb_len = min(atio->cdb_len, sizeof(io->scsiio.cdb)); bcopy(atio->cdb_io.cdb_bytes, io->scsiio.cdb, io->scsiio.cdb_len); #ifdef CTLFEDEBUG printf("%s: %ju:%d:%ju:%d: tag %04x CDB %02x\n", __func__, (uintmax_t)io->io_hdr.nexus.initid.id, io->io_hdr.nexus.targ_port, (uintmax_t)io->io_hdr.nexus.targ_target.id, io->io_hdr.nexus.targ_lun, io->scsiio.tag_num, io->scsiio.cdb[0]); #endif ctl_queue(io); return; } case XPT_CONT_TARGET_IO: { int srr = 0; uint32_t srr_off = 0; atio = (struct ccb_accept_tio *)done_ccb->ccb_h.ccb_atio; io = (union ctl_io *)atio->ccb_h.io_ptr; softc->ctios_returned++; #ifdef CTLFEDEBUG printf("%s: got XPT_CONT_TARGET_IO tag %#x flags %#x\n", __func__, atio->tag_id, done_ccb->ccb_h.flags); #endif /* * Handle SRR case were the data pointer is pushed back hack */ if ((done_ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_MESSAGE_RECV && done_ccb->csio.msg_ptr != NULL && done_ccb->csio.msg_ptr[0] == MSG_EXTENDED && done_ccb->csio.msg_ptr[1] == 5 && done_ccb->csio.msg_ptr[2] == 0) { srr = 1; srr_off = (done_ccb->csio.msg_ptr[3] << 24) | (done_ccb->csio.msg_ptr[4] << 16) | (done_ccb->csio.msg_ptr[5] << 8) | (done_ccb->csio.msg_ptr[6]); } if (srr && (io->io_hdr.flags & CTL_FLAG_DMA_INPROG) == 0) { /* * If status was being sent, the back end data is now * history. Hack it up and resubmit a new command with * the CDB adjusted. If the SIM does the right thing, * all of the resid math should work. */ softc->ccbs_freed++; xpt_release_ccb(done_ccb); if (ctlfe_adjust_cdb(atio, srr_off) == 0) { done_ccb = (union ccb *)atio; goto resubmit; } /* * Fall through to doom.... */ } else if (srr) { /* * If we have an srr and we're still sending data, we * should be able to adjust offsets and cycle again. */ io->scsiio.kern_rel_offset = io->scsiio.ext_data_filled = srr_off; io->scsiio.ext_data_len = io->scsiio.kern_total_len - io->scsiio.kern_rel_offset; softc->ccbs_freed++; io->scsiio.io_hdr.status = CTL_STATUS_NONE; xpt_release_ccb(done_ccb); TAILQ_INSERT_HEAD(&softc->work_queue, &atio->ccb_h, periph_links.tqe); xpt_schedule(periph, /*priority*/ 1); break; } /* * If we were sending status back to the initiator, free up * resources. If we were doing a datamove, call the * datamove done routine. */ if ((io->io_hdr.flags & CTL_FLAG_DMA_INPROG) == 0) { softc->ccbs_freed++; xpt_release_ccb(done_ccb); /* * For a wildcard attachment, commands can come in * with a specific target/lun. Reset the target * and LUN fields back to the wildcard values before * we send them back down to the SIM. The SIM has * a wildcard LUN enabled, not whatever target/lun * these happened to be. */ if (softc->flags & CTLFE_LUN_WILDCARD) { atio->ccb_h.target_id = CAM_TARGET_WILDCARD; atio->ccb_h.target_lun = CAM_LUN_WILDCARD; } if (periph->flags & CAM_PERIPH_INVALID) { ctlfe_free_ccb(periph, (union ccb *)atio); } else { softc->atios_sent++; mtx_unlock(mtx); xpt_action((union ccb *)atio); return; } } else { struct ctlfe_lun_cmd_info *cmd_info; struct ccb_scsiio *csio; csio = &done_ccb->csio; cmd_info = (struct ctlfe_lun_cmd_info *) io->io_hdr.port_priv; io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG; io->scsiio.ext_data_len += csio->dxfer_len; if (io->scsiio.ext_data_len > io->scsiio.kern_total_len) { xpt_print(periph->path, "%s: tag 0x%04x " "done len %u > total %u sent %u\n", __func__, io->scsiio.tag_num, io->scsiio.ext_data_len, io->scsiio.kern_total_len, io->scsiio.ext_data_filled); } /* * Translate CAM status to CTL status. Success * does not change the overall, ctl_io status. In * that case we just set port_status to 0. If we * have a failure, though, set a data phase error * for the overall ctl_io. */ switch (done_ccb->ccb_h.status & CAM_STATUS_MASK) { case CAM_REQ_CMP: io->io_hdr.port_status = 0; break; default: /* * XXX KDM we probably need to figure out a * standard set of errors that the SIM * drivers should return in the event of a * data transfer failure. A data phase * error will at least point the user to a * data transfer error of some sort. * Hopefully the SIM printed out some * additional information to give the user * a clue what happened. */ io->io_hdr.port_status = 0xbad1; ctl_set_data_phase_error(&io->scsiio); /* * XXX KDM figure out residual. */ break; } /* * If we had to break this S/G list into multiple * pieces, figure out where we are in the list, and * continue sending pieces if necessary. */ if ((cmd_info->flags & CTLFE_CMD_PIECEWISE) && (io->io_hdr.port_status == 0)) { ccb_flags flags; uint8_t scsi_status; uint8_t *data_ptr; uint32_t dxfer_len; flags = atio->ccb_h.flags & (CAM_DIS_DISCONNECT| CAM_TAG_ACTION_VALID); ctlfedata(softc, io, &flags, &data_ptr, &dxfer_len, &csio->sglist_cnt); scsi_status = 0; if (((flags & CAM_SEND_STATUS) == 0) && (dxfer_len == 0)) { printf("%s: tag %04x no status or " "len cdb = %02x\n", __func__, atio->tag_id, atio->cdb_io.cdb_bytes[0]); printf("%s: tag %04x io status %#x\n", __func__, atio->tag_id, io->io_hdr.status); } cam_fill_ctio(csio, /*retries*/ 2, ctlfedone, flags, (flags & CAM_TAG_ACTION_VALID) ? MSG_SIMPLE_Q_TAG : 0, atio->tag_id, atio->init_id, scsi_status, /*data_ptr*/ data_ptr, /*dxfer_len*/ dxfer_len, /*timeout*/ 5 * 1000); csio->ccb_h.flags |= CAM_UNLOCKED; csio->resid = 0; csio->ccb_h.ccb_atio = atio; io->io_hdr.flags |= CTL_FLAG_DMA_INPROG; softc->ctios_sent++; mtx_unlock(mtx); xpt_action((union ccb *)csio); } else { /* * Release the CTIO. The ATIO will be sent back * down to the SIM once we send status. */ softc->ccbs_freed++; xpt_release_ccb(done_ccb); mtx_unlock(mtx); /* Call the backend move done callback */ io->scsiio.be_move_done(io); } return; } break; } case XPT_IMMEDIATE_NOTIFY: { union ctl_io *io; struct ccb_immediate_notify *inot; cam_status status; int frozen, send_ctl_io; inot = &done_ccb->cin1; softc->inots_returned++; frozen = (done_ccb->ccb_h.status & CAM_DEV_QFRZN) != 0; printf("%s: got XPT_IMMEDIATE_NOTIFY status %#x tag %#x " "seq %#x\n", __func__, inot->ccb_h.status, inot->tag_id, inot->seq_id); io = done_ccb->ccb_h.io_ptr; ctl_zero_io(io); send_ctl_io = 1; io->io_hdr.io_type = CTL_IO_TASK; io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr =done_ccb; inot->ccb_h.io_ptr = io; io->io_hdr.nexus.initid.id = inot->initiator_id; io->io_hdr.nexus.targ_port = bus_softc->port.targ_port; io->io_hdr.nexus.targ_target.id = inot->ccb_h.target_id; io->io_hdr.nexus.targ_lun = inot->ccb_h.target_lun; /* XXX KDM should this be the tag_id? */ io->taskio.tag_num = inot->seq_id; status = inot->ccb_h.status & CAM_STATUS_MASK; switch (status) { case CAM_SCSI_BUS_RESET: io->taskio.task_action = CTL_TASK_BUS_RESET; break; case CAM_BDR_SENT: io->taskio.task_action = CTL_TASK_TARGET_RESET; break; case CAM_MESSAGE_RECV: switch (inot->arg) { case MSG_ABORT_TASK_SET: io->taskio.task_action = CTL_TASK_ABORT_TASK_SET; break; case MSG_TARGET_RESET: io->taskio.task_action = CTL_TASK_TARGET_RESET; break; case MSG_ABORT_TASK: io->taskio.task_action = CTL_TASK_ABORT_TASK; break; case MSG_LOGICAL_UNIT_RESET: io->taskio.task_action = CTL_TASK_LUN_RESET; break; case MSG_CLEAR_TASK_SET: io->taskio.task_action = CTL_TASK_CLEAR_TASK_SET; break; case MSG_CLEAR_ACA: io->taskio.task_action = CTL_TASK_CLEAR_ACA; break; case MSG_NOOP: send_ctl_io = 0; break; default: xpt_print(periph->path, "%s: unsupported message 0x%x\n", __func__, inot->arg); send_ctl_io = 0; break; } break; case CAM_REQ_ABORTED: /* * This request was sent back by the driver. * XXX KDM what do we do here? */ send_ctl_io = 0; break; case CAM_REQ_INVALID: case CAM_PROVIDE_FAIL: default: /* * We should only get here if we're talking * to a talking to a SIM that is target * capable but supports the old API. In * that case, we need to just free the CCB. * If we actually send a notify acknowledge, * it will send that back with an error as * well. */ if ((status != CAM_REQ_INVALID) && (status != CAM_PROVIDE_FAIL)) xpt_print(periph->path, "%s: unsupported CAM status 0x%x\n", __func__, status); ctlfe_free_ccb(periph, done_ccb); goto out; } if (send_ctl_io != 0) { ctl_queue(io); } else { done_ccb->ccb_h.status = CAM_REQ_INPROG; done_ccb->ccb_h.func_code = XPT_NOTIFY_ACKNOWLEDGE; xpt_action(done_ccb); } if (frozen != 0) { cam_release_devq(periph->path, /*relsim_flags*/ 0, /*opening reduction*/ 0, /*timeout*/ 0, /*getcount_only*/ 0); } break; } case XPT_NOTIFY_ACKNOWLEDGE: /* * Queue this back down to the SIM as an immediate notify. */ done_ccb->ccb_h.func_code = XPT_IMMEDIATE_NOTIFY; xpt_action(done_ccb); softc->inots_sent++; break; case XPT_SET_SIM_KNOB: case XPT_GET_SIM_KNOB: break; default: panic("%s: unexpected CCB type %#x", __func__, done_ccb->ccb_h.func_code); break; } out: mtx_unlock(mtx); } static void ctlfe_onoffline(void *arg, int online) { struct ctlfe_softc *bus_softc; union ccb *ccb; cam_status status; struct cam_path *path; int set_wwnn; bus_softc = (struct ctlfe_softc *)arg; set_wwnn = 0; status = xpt_create_path(&path, /*periph*/ NULL, bus_softc->path_id, CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); if (status != CAM_REQ_CMP) { printf("%s: unable to create path!\n", __func__); return; } ccb = (union ccb *)malloc(sizeof(*ccb), M_TEMP, M_NOWAIT | M_ZERO); if (ccb == NULL) { printf("%s: unable to malloc CCB!\n", __func__); xpt_free_path(path); return; } xpt_setup_ccb(&ccb->ccb_h, path, CAM_PRIORITY_NONE); /* * Copan WWN format: * * Bits 63-60: 0x5 NAA, IEEE registered name * Bits 59-36: 0x000ED5 IEEE Company name assigned to Copan * Bits 35-12: Copan SSN (Sequential Serial Number) * Bits 11-8: Type of port: * 1 == N-Port * 2 == F-Port * 3 == NL-Port * Bits 7-0: 0 == Node Name, >0 == Port Number */ if (online != 0) { ccb->ccb_h.func_code = XPT_GET_SIM_KNOB; xpt_action(ccb); if ((ccb->knob.xport_specific.valid & KNOB_VALID_ADDRESS) != 0){ #ifdef RANDOM_WWNN uint64_t random_bits; #endif printf("%s: %s current WWNN %#jx\n", __func__, bus_softc->port_name, ccb->knob.xport_specific.fc.wwnn); printf("%s: %s current WWPN %#jx\n", __func__, bus_softc->port_name, ccb->knob.xport_specific.fc.wwpn); #ifdef RANDOM_WWNN arc4rand(&random_bits, sizeof(random_bits), 0); #endif /* * XXX KDM this is a bit of a kludge for now. We * take the current WWNN/WWPN from the card, and * replace the company identifier and the NL-Port * indicator and the port number (for the WWPN). * This should be replaced later with ddb_GetWWNN, * or possibly a more centralized scheme. (It * would be nice to have the WWNN/WWPN for each * port stored in the ctl_port structure.) */ #ifdef RANDOM_WWNN ccb->knob.xport_specific.fc.wwnn = (random_bits & 0x0000000fffffff00ULL) | /* Company ID */ 0x5000ED5000000000ULL | /* NL-Port */ 0x0300; ccb->knob.xport_specific.fc.wwpn = (random_bits & 0x0000000fffffff00ULL) | /* Company ID */ 0x5000ED5000000000ULL | /* NL-Port */ 0x3000 | /* Port Num */ (bus_softc->port.targ_port & 0xff); /* * This is a bit of an API break/reversal, but if * we're doing the random WWNN that's a little * different anyway. So record what we're actually * using with the frontend code so it's reported * accurately. */ ctl_port_set_wwns(&bus_softc->port, true, ccb->knob.xport_specific.fc.wwnn, true, ccb->knob.xport_specific.fc.wwpn); set_wwnn = 1; #else /* RANDOM_WWNN */ /* * If the user has specified a WWNN/WWPN, send them * down to the SIM. Otherwise, record what the SIM * has reported. */ if ((bus_softc->port.wwnn != 0) && (bus_softc->port.wwpn != 0)) { ccb->knob.xport_specific.fc.wwnn = bus_softc->port.wwnn; ccb->knob.xport_specific.fc.wwpn = bus_softc->port.wwpn; set_wwnn = 1; } else { ctl_port_set_wwns(&bus_softc->port, true, ccb->knob.xport_specific.fc.wwnn, true, ccb->knob.xport_specific.fc.wwpn); } #endif /* RANDOM_WWNN */ if (set_wwnn != 0) { printf("%s: %s new WWNN %#jx\n", __func__, bus_softc->port_name, ccb->knob.xport_specific.fc.wwnn); printf("%s: %s new WWPN %#jx\n", __func__, bus_softc->port_name, ccb->knob.xport_specific.fc.wwpn); } } else { printf("%s: %s has no valid WWNN/WWPN\n", __func__, bus_softc->port_name); } } ccb->ccb_h.func_code = XPT_SET_SIM_KNOB; ccb->knob.xport_specific.valid = KNOB_VALID_ROLE; if (set_wwnn != 0) ccb->knob.xport_specific.valid |= KNOB_VALID_ADDRESS; if (online != 0) ccb->knob.xport_specific.fc.role = KNOB_ROLE_TARGET; else ccb->knob.xport_specific.fc.role = KNOB_ROLE_NONE; xpt_action(ccb); if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) { printf("%s: SIM %s (path id %d) target %s failed with " "status %#x\n", __func__, bus_softc->port_name, bus_softc->path_id, (online != 0) ? "enable" : "disable", ccb->ccb_h.status); } else { printf("%s: SIM %s (path id %d) target %s succeeded\n", __func__, bus_softc->port_name, bus_softc->path_id, (online != 0) ? "enable" : "disable"); } xpt_free_path(path); free(ccb, M_TEMP); return; } static void ctlfe_online(void *arg) { struct ctlfe_softc *bus_softc; struct cam_path *path; cam_status status; struct ctlfe_lun_softc *lun_softc; struct cam_periph *periph; bus_softc = (struct ctlfe_softc *)arg; /* * Create the wildcard LUN before bringing the port online. */ status = xpt_create_path(&path, /*periph*/ NULL, bus_softc->path_id, CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); if (status != CAM_REQ_CMP) { printf("%s: unable to create path for wildcard periph\n", __func__); return; } lun_softc = malloc(sizeof(*lun_softc), M_CTLFE, M_NOWAIT | M_ZERO); if (lun_softc == NULL) { xpt_print(path, "%s: unable to allocate softc for " "wildcard periph\n", __func__); xpt_free_path(path); return; } xpt_path_lock(path); periph = cam_periph_find(path, "ctl"); if (periph != NULL) { /* We've already got a periph, no need to alloc a new one. */ xpt_path_unlock(path); xpt_free_path(path); free(lun_softc, M_CTLFE); return; } lun_softc->parent_softc = bus_softc; lun_softc->flags |= CTLFE_LUN_WILDCARD; status = cam_periph_alloc(ctlferegister, ctlfeoninvalidate, ctlfecleanup, ctlfestart, "ctl", CAM_PERIPH_BIO, path, ctlfeasync, 0, lun_softc); if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) { const struct cam_status_entry *entry; entry = cam_fetch_status_entry(status); printf("%s: CAM error %s (%#x) returned from " "cam_periph_alloc()\n", __func__, (entry != NULL) ? entry->status_text : "Unknown", status); free(lun_softc, M_CTLFE); } else { mtx_lock(&bus_softc->lun_softc_mtx); STAILQ_INSERT_TAIL(&bus_softc->lun_softc_list, lun_softc, links); mtx_unlock(&bus_softc->lun_softc_mtx); ctlfe_onoffline(arg, /*online*/ 1); } xpt_path_unlock(path); xpt_free_path(path); } static void ctlfe_offline(void *arg) { struct ctlfe_softc *bus_softc; struct cam_path *path; cam_status status; struct cam_periph *periph; bus_softc = (struct ctlfe_softc *)arg; /* * Disable the wildcard LUN for this port now that we have taken * the port offline. */ status = xpt_create_path(&path, /*periph*/ NULL, bus_softc->path_id, CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); if (status != CAM_REQ_CMP) { printf("%s: unable to create path for wildcard periph\n", __func__); return; } xpt_path_lock(path); ctlfe_onoffline(arg, /*online*/ 0); if ((periph = cam_periph_find(path, "ctl")) != NULL) cam_periph_invalidate(periph); xpt_path_unlock(path); xpt_free_path(path); } /* * This will get called to enable a LUN on every bus that is attached to * CTL. So we only need to create a path/periph for this particular bus. */ static int ctlfe_lun_enable(void *arg, struct ctl_id targ_id, int lun_id) { struct ctlfe_softc *bus_softc; struct ctlfe_lun_softc *softc; struct cam_path *path; struct cam_periph *periph; cam_status status; bus_softc = (struct ctlfe_softc *)arg; status = xpt_create_path(&path, /*periph*/ NULL, bus_softc->path_id, targ_id.id, lun_id); /* XXX KDM need some way to return status to CTL here? */ if (status != CAM_REQ_CMP) { printf("%s: could not create path, status %#x\n", __func__, status); return (1); } softc = malloc(sizeof(*softc), M_CTLFE, M_WAITOK | M_ZERO); xpt_path_lock(path); periph = cam_periph_find(path, "ctl"); if (periph != NULL) { /* We've already got a periph, no need to alloc a new one. */ xpt_path_unlock(path); xpt_free_path(path); free(softc, M_CTLFE); return (0); } softc->parent_softc = bus_softc; status = cam_periph_alloc(ctlferegister, ctlfeoninvalidate, ctlfecleanup, ctlfestart, "ctl", CAM_PERIPH_BIO, path, ctlfeasync, 0, softc); if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) { const struct cam_status_entry *entry; entry = cam_fetch_status_entry(status); printf("%s: CAM error %s (%#x) returned from " "cam_periph_alloc()\n", __func__, (entry != NULL) ? entry->status_text : "Unknown", status); free(softc, M_CTLFE); } else { mtx_lock(&bus_softc->lun_softc_mtx); STAILQ_INSERT_TAIL(&bus_softc->lun_softc_list, softc, links); mtx_unlock(&bus_softc->lun_softc_mtx); } xpt_path_unlock(path); xpt_free_path(path); return (0); } /* * This will get called when the user removes a LUN to disable that LUN * on every bus that is attached to CTL. */ static int ctlfe_lun_disable(void *arg, struct ctl_id targ_id, int lun_id) { struct ctlfe_softc *softc; struct ctlfe_lun_softc *lun_softc; softc = (struct ctlfe_softc *)arg; mtx_lock(&softc->lun_softc_mtx); STAILQ_FOREACH(lun_softc, &softc->lun_softc_list, links) { struct cam_path *path; path = lun_softc->periph->path; if ((xpt_path_target_id(path) == targ_id.id) && (xpt_path_lun_id(path) == lun_id)) { break; } } if (lun_softc == NULL) { mtx_unlock(&softc->lun_softc_mtx); printf("%s: can't find target %d lun %d\n", __func__, targ_id.id, lun_id); return (1); } cam_periph_acquire(lun_softc->periph); mtx_unlock(&softc->lun_softc_mtx); cam_periph_lock(lun_softc->periph); cam_periph_invalidate(lun_softc->periph); cam_periph_unlock(lun_softc->periph); cam_periph_release(lun_softc->periph); return (0); } static void ctlfe_dump_sim(struct cam_sim *sim) { printf("%s%d: max tagged openings: %d, max dev openings: %d\n", sim->sim_name, sim->unit_number, sim->max_tagged_dev_openings, sim->max_dev_openings); } /* * Assumes that the SIM lock is held. */ static void ctlfe_dump_queue(struct ctlfe_lun_softc *softc) { struct ccb_hdr *hdr; struct cam_periph *periph; int num_items; periph = softc->periph; num_items = 0; TAILQ_FOREACH(hdr, &softc->work_queue, periph_links.tqe) { union ctl_io *io = hdr->io_ptr; num_items++; /* * Only regular SCSI I/O is put on the work * queue, so we can print sense here. There may be no * sense if it's no the queue for a DMA, but this serves to * print out the CCB as well. * * XXX KDM switch this over to scsi_sense_print() when * CTL is merged in with CAM. */ ctl_io_error_print(io, NULL); /* * Print DMA status if we are DMA_QUEUED. */ if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED) { xpt_print(periph->path, "Total %u, Current %u, Resid %u\n", io->scsiio.kern_total_len, io->scsiio.kern_data_len, io->scsiio.kern_data_resid); } } xpt_print(periph->path, "%d requests total waiting for CCBs\n", num_items); xpt_print(periph->path, "%ju CCBs outstanding (%ju allocated, %ju " "freed)\n", (uintmax_t)(softc->ccbs_alloced - softc->ccbs_freed), (uintmax_t)softc->ccbs_alloced, (uintmax_t)softc->ccbs_freed); xpt_print(periph->path, "%ju CTIOs outstanding (%ju sent, %ju " "returned\n", (uintmax_t)(softc->ctios_sent - softc->ctios_returned), softc->ctios_sent, softc->ctios_returned); } /* * Datamove/done routine called by CTL. Put ourselves on the queue to * receive a CCB from CAM so we can queue the continue I/O request down * to the adapter. */ static void ctlfe_datamove(union ctl_io *io) { union ccb *ccb; struct cam_periph *periph; struct ctlfe_lun_softc *softc; KASSERT(io->io_hdr.io_type == CTL_IO_SCSI, ("Unexpected io_type (%d) in ctlfe_datamove", io->io_hdr.io_type)); ccb = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; periph = xpt_path_periph(ccb->ccb_h.path); cam_periph_lock(periph); softc = (struct ctlfe_lun_softc *)periph->softc; io->io_hdr.flags |= CTL_FLAG_DMA_QUEUED; if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE) io->io_hdr.flags |= CTL_FLAG_STATUS_QUEUED; TAILQ_INSERT_TAIL(&softc->work_queue, &ccb->ccb_h, periph_links.tqe); xpt_schedule(periph, /*priority*/ 1); cam_periph_unlock(periph); } static void ctlfe_done(union ctl_io *io) { union ccb *ccb; struct cam_periph *periph; struct ctlfe_lun_softc *softc; ccb = io->io_hdr.ctl_private[CTL_PRIV_FRONTEND].ptr; periph = xpt_path_periph(ccb->ccb_h.path); cam_periph_lock(periph); softc = (struct ctlfe_lun_softc *)periph->softc; if (io->io_hdr.io_type == CTL_IO_TASK) { /* * Task management commands don't require any further * communication back to the adapter. Requeue the CCB * to the adapter, and free the CTL I/O. */ xpt_print(ccb->ccb_h.path, "%s: returning task I/O " "tag %#x seq %#x\n", __func__, ccb->cin1.tag_id, ccb->cin1.seq_id); /* * Send the notify acknowledge down to the SIM, to let it * know we processed the task management command. */ ccb->ccb_h.status = CAM_REQ_INPROG; ccb->ccb_h.func_code = XPT_NOTIFY_ACKNOWLEDGE; xpt_action(ccb); + } else if (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) { + if (softc->flags & CTLFE_LUN_WILDCARD) { + ccb->ccb_h.target_id = CAM_TARGET_WILDCARD; + ccb->ccb_h.target_lun = CAM_LUN_WILDCARD; + } + if (periph->flags & CAM_PERIPH_INVALID) { + ctlfe_free_ccb(periph, ccb); + } else { + softc->atios_sent++; + cam_periph_unlock(periph); + xpt_action(ccb); + return; + } } else { io->io_hdr.flags |= CTL_FLAG_STATUS_QUEUED; TAILQ_INSERT_TAIL(&softc->work_queue, &ccb->ccb_h, periph_links.tqe); xpt_schedule(periph, /*priority*/ 1); } cam_periph_unlock(periph); } static void ctlfe_dump(void) { struct ctlfe_softc *bus_softc; struct ctlfe_lun_softc *lun_softc; STAILQ_FOREACH(bus_softc, &ctlfe_softc_list, links) { ctlfe_dump_sim(bus_softc->sim); STAILQ_FOREACH(lun_softc, &bus_softc->lun_softc_list, links) ctlfe_dump_queue(lun_softc); } }