Index: head/sys/cam/ctl/ctl.c =================================================================== --- head/sys/cam/ctl/ctl.c (revision 274153) +++ head/sys/cam/ctl/ctl.c (revision 274154) @@ -1,14153 +1,14461 @@ /*- * 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 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*/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} }; -static struct scsi_logical_block_provisioning_page lbp_page_default = { +#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*/{0, sizeof(struct scsi_logical_block_provisioning_page) - 4}, + /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN}, /*flags*/0, /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, - /*descr*/{} + /*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 scsi_logical_block_provisioning_page lbp_page_changeable = { +static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{ /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF, /*subpage_code*/0x02, - /*page_length*/{0, sizeof(struct scsi_logical_block_provisioning_page) - 4}, + /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN}, /*flags*/0, /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, - /*descr*/{} + /*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; static int ctl_is_single = 1; 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((void *)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( (void *)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( (void *)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_init(void) { struct ctl_softc *softc; struct ctl_io_pool *internal_pool, *emergency_pool, *other_pool; struct ctl_port *port; uint8_t sc_id =0; 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); mtx_init(&softc->pool_lock, "CTL pool mutex", NULL, MTX_DEF); 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. */ #ifdef NEEDTOPORT scmicro_rw(SCMICRO_GET_SHELF_ID, &sc_id); #endif if (sc_id == 0) { softc->flags |= CTL_FLAG_MASTER_SHELF; persis_offset = 0; } else persis_offset = CTL_MAX_INITIATORS; /* * 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); STAILQ_INIT(&softc->io_pools); ctl_tpc_init(softc); if (ctl_pool_create(softc, CTL_POOL_INTERNAL, CTL_POOL_ENTRIES_INTERNAL, &internal_pool)!= 0){ printf("ctl: can't allocate %d entry internal pool, " "exiting\n", CTL_POOL_ENTRIES_INTERNAL); return (ENOMEM); } if (ctl_pool_create(softc, CTL_POOL_EMERGENCY, CTL_POOL_ENTRIES_EMERGENCY, &emergency_pool) != 0) { printf("ctl: can't allocate %d entry emergency pool, " "exiting\n", CTL_POOL_ENTRIES_EMERGENCY); ctl_pool_free(internal_pool); return (ENOMEM); } if (ctl_pool_create(softc, CTL_POOL_4OTHERSC, 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); ctl_pool_free(internal_pool); ctl_pool_free(emergency_pool); return (ENOMEM); } softc->internal_pool = internal_pool; softc->emergency_pool = emergency_pool; 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(internal_pool); ctl_pool_free(emergency_pool); 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(internal_pool); ctl_pool_free(emergency_pool); 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(internal_pool); + ctl_pool_free(emergency_pool); + 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, (softc->flags & CTL_FLAG_MASTER_SHELF)) != 0) { printf("ctl: ioctl front end registration failed, will " "continue anyway\n"); } #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; struct ctl_io_pool *pool; 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); /* * This will rip the rug out from under any FETDs or anyone else * that has a pool allocated. Since we increment our module * refcount any time someone outside the main CTL module allocates * a pool, we shouldn't have any problems here. The user won't be * able to unload the CTL module until client modules have * successfully unloaded. */ while ((pool = STAILQ_FIRST(&softc->io_pools)) != NULL) ctl_pool_free(pool); #if 0 ctl_shutdown_thread(softc->work_thread); mtx_destroy(&softc->queue_lock); #endif ctl_tpc_shutdown(softc); mtx_destroy(&softc->pool_lock); 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; struct ctl_port *port; if (ctl_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 } softc = control_softc; 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); if (io == NULL) { printf("ctl_ioctl: can't allocate ctl_io!\n"); retval = ENOSPC; break; } /* * 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_type, total_ctl_io are passed in. * npool is passed out. */ int ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type, uint32_t total_ctl_io, struct ctl_io_pool **npool) { uint32_t i; union ctl_io *cur_io, *next_io; struct ctl_io_pool *pool; int retval; retval = 0; pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL, M_NOWAIT | M_ZERO); if (pool == NULL) { retval = ENOMEM; goto bailout; } pool->type = pool_type; pool->ctl_softc = ctl_softc; mtx_lock(&ctl_softc->pool_lock); pool->id = ctl_softc->cur_pool_id++; mtx_unlock(&ctl_softc->pool_lock); pool->flags = CTL_POOL_FLAG_NONE; pool->refcount = 1; /* Reference for validity. */ STAILQ_INIT(&pool->free_queue); /* * XXX KDM other options here: * - allocate a page at a time * - allocate one big chunk of memory. * Page allocation might work well, but would take a little more * tracking. */ for (i = 0; i < total_ctl_io; i++) { cur_io = (union ctl_io *)malloc(sizeof(*cur_io), M_CTLIO, M_NOWAIT); if (cur_io == NULL) { retval = ENOMEM; break; } cur_io->io_hdr.pool = pool; STAILQ_INSERT_TAIL(&pool->free_queue, &cur_io->io_hdr, links); pool->total_ctl_io++; pool->free_ctl_io++; } if (retval != 0) { for (cur_io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue); cur_io != NULL; cur_io = next_io) { next_io = (union ctl_io *)STAILQ_NEXT(&cur_io->io_hdr, links); STAILQ_REMOVE(&pool->free_queue, &cur_io->io_hdr, ctl_io_hdr, links); free(cur_io, M_CTLIO); } free(pool, M_CTL); goto bailout; } mtx_lock(&ctl_softc->pool_lock); ctl_softc->num_pools++; STAILQ_INSERT_TAIL(&ctl_softc->io_pools, pool, links); /* * Increment our usage count if this is an external consumer, so we * can't get unloaded until the external consumer (most likely a * FETD) unloads and frees his pool. * * XXX KDM will this increment the caller's module use count, or * mine? */ #if 0 if ((pool_type != CTL_POOL_EMERGENCY) && (pool_type != CTL_POOL_INTERNAL) && (pool_type != CTL_POOL_4OTHERSC)) MOD_INC_USE_COUNT; #endif mtx_unlock(&ctl_softc->pool_lock); *npool = pool; bailout: return (retval); } static int ctl_pool_acquire(struct ctl_io_pool *pool) { mtx_assert(&pool->ctl_softc->pool_lock, MA_OWNED); if (pool->flags & CTL_POOL_FLAG_INVALID) return (EINVAL); pool->refcount++; return (0); } static void ctl_pool_release(struct ctl_io_pool *pool) { struct ctl_softc *ctl_softc = pool->ctl_softc; union ctl_io *io; mtx_assert(&ctl_softc->pool_lock, MA_OWNED); if (--pool->refcount != 0) return; while ((io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue)) != NULL) { STAILQ_REMOVE(&pool->free_queue, &io->io_hdr, ctl_io_hdr, links); free(io, M_CTLIO); } STAILQ_REMOVE(&ctl_softc->io_pools, pool, ctl_io_pool, links); ctl_softc->num_pools--; /* * XXX KDM will this decrement the caller's usage count or mine? */ #if 0 if ((pool->type != CTL_POOL_EMERGENCY) && (pool->type != CTL_POOL_INTERNAL) && (pool->type != CTL_POOL_4OTHERSC)) MOD_DEC_USE_COUNT; #endif free(pool, M_CTL); } void ctl_pool_free(struct ctl_io_pool *pool) { struct ctl_softc *ctl_softc; if (pool == NULL) return; ctl_softc = pool->ctl_softc; mtx_lock(&ctl_softc->pool_lock); pool->flags |= CTL_POOL_FLAG_INVALID; ctl_pool_release(pool); mtx_unlock(&ctl_softc->pool_lock); } /* * This routine does not block (except for spinlocks of course). * It tries to allocate a ctl_io union from the caller's pool as quickly as * possible. */ union ctl_io * ctl_alloc_io(void *pool_ref) { union ctl_io *io; struct ctl_softc *ctl_softc; struct ctl_io_pool *pool, *npool; struct ctl_io_pool *emergency_pool; pool = (struct ctl_io_pool *)pool_ref; if (pool == NULL) { printf("%s: pool is NULL\n", __func__); return (NULL); } emergency_pool = NULL; ctl_softc = pool->ctl_softc; mtx_lock(&ctl_softc->pool_lock); /* * First, try to get the io structure from the user's pool. */ if (ctl_pool_acquire(pool) == 0) { io = (union ctl_io *)STAILQ_FIRST(&pool->free_queue); if (io != NULL) { STAILQ_REMOVE_HEAD(&pool->free_queue, links); pool->total_allocated++; pool->free_ctl_io--; mtx_unlock(&ctl_softc->pool_lock); return (io); } else ctl_pool_release(pool); } /* * If he doesn't have any io structures left, search for an * emergency pool and grab one from there. */ STAILQ_FOREACH(npool, &ctl_softc->io_pools, links) { if (npool->type != CTL_POOL_EMERGENCY) continue; if (ctl_pool_acquire(npool) != 0) continue; emergency_pool = npool; io = (union ctl_io *)STAILQ_FIRST(&npool->free_queue); if (io != NULL) { STAILQ_REMOVE_HEAD(&npool->free_queue, links); npool->total_allocated++; npool->free_ctl_io--; mtx_unlock(&ctl_softc->pool_lock); return (io); } else ctl_pool_release(npool); } /* Drop the spinlock before we malloc */ mtx_unlock(&ctl_softc->pool_lock); /* * The emergency pool (if it exists) didn't have one, so try an * atomic (i.e. nonblocking) malloc and see if we get lucky. */ io = (union ctl_io *)malloc(sizeof(*io), M_CTLIO, M_NOWAIT); if (io != NULL) { /* * If the emergency pool exists but is empty, add this * ctl_io to its list when it gets freed. */ if (emergency_pool != NULL) { mtx_lock(&ctl_softc->pool_lock); if (ctl_pool_acquire(emergency_pool) == 0) { io->io_hdr.pool = emergency_pool; emergency_pool->total_ctl_io++; /* * Need to bump this, otherwise * total_allocated and total_freed won't * match when we no longer have anything * outstanding. */ emergency_pool->total_allocated++; } mtx_unlock(&ctl_softc->pool_lock); } else io->io_hdr.pool = NULL; } return (io); } void ctl_free_io(union ctl_io *io) { if (io == NULL) return; /* * If this ctl_io has a pool, return it to that pool. */ if (io->io_hdr.pool != NULL) { struct ctl_io_pool *pool; pool = (struct ctl_io_pool *)io->io_hdr.pool; mtx_lock(&pool->ctl_softc->pool_lock); io->io_hdr.io_type = 0xff; STAILQ_INSERT_TAIL(&pool->free_queue, &io->io_hdr, links); pool->total_freed++; pool->free_ctl_io++; ctl_pool_release(pool); mtx_unlock(&pool->ctl_softc->pool_lock); } else { /* * Otherwise, just free it. We probably malloced it and * the emergency pool wasn't available. */ free(io, M_CTLIO); } } 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_CURRENT], &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)); 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_SAVED], &rigid_disk_page_default, sizeof(rigid_disk_page_default)); 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_CURRENT]; scsi_ulto3b(cylinders, rigid_disk_page->cylinders); rigid_disk_page = &lun->mode_pages.rigid_disk_page[ CTL_PAGE_DEFAULT]; scsi_ulto3b(cylinders, rigid_disk_page->cylinders); rigid_disk_page = &lun->mode_pages.rigid_disk_page[ CTL_PAGE_SAVED]; scsi_ulto3b(cylinders, rigid_disk_page->cylinders); 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: - memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT], + 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_DEFAULT], - &lbp_page_default, - sizeof(lbp_page_default)); 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, prev; + 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; i < CTL_NUM_LOG_PAGES; i++) { + 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[i*2] = page_index->page_code; - lun->log_pages.subpages_page[i*2+1] = page_index->subpage; + 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 = i * 2; + 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!"); 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)){ /* * 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)) { /* * 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); } 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); ctsio->scsi_status = SCSI_STATUS_OK; ctsio->io_hdr.status = CTL_SUCCESS; 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_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; ctsio->scsi_status = SCSI_STATUS_OK; ctsio->io_hdr.status = CTL_SUCCESS; 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); if (new_io == NULL) { ctl_set_busy(ctsio); ctl_done((union ctl_io *)ctsio); } else { 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); ctsio->scsi_status = SCSI_STATUS_OK; ctsio->io_hdr.status = CTL_SUCCESS; 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; 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_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; 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; 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; 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; 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; struct ctl_lun *lun; uint32_t alloc_len, data_len; uint8_t format; CTL_DEBUG_PRINT(("ctl_read_defect\n")); lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; 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; int num_target_port_groups, num_target_ports, single; 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); } single = ctl_is_single; if (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]; } pg = ctsio->io_hdr.nexus.targ_port / CTL_MAX_PORTS; mtx_lock(&softc->ctl_lock); for (g = 0; g < num_target_port_groups; g++) { if (g == pg) tpg_desc->pref_state = TPG_PRIMARY | TPG_ASYMMETRIC_ACCESS_OPTIMIZED; else tpg_desc->pref_state = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED; tpg_desc->support = TPG_AO_SUP; if (!single) tpg_desc->support |= TPG_AN_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); 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; } 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 ctl_lun *lun; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; 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 ctl_lun *lun; 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; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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]); 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); 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: { 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; 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; 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; 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) { struct ctl_lun *lun; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; CTL_DEBUG_PRINT(("ctl_tur\n")); if (lun == NULL) return (EINVAL); ctsio->scsi_status = SCSI_STATUS_OK; ctsio->io_hdr.status = CTL_SUCCESS; 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; 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; 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; 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; 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; 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, single; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; single = ctl_is_single; if (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); if (softc->flags & CTL_FLAG_MASTER_SHELF) pg = 0; else pg = 1; 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; 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; 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 = SVPD_NON_ROTATING; 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; 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_RESOURCE; + lbp_ptr->prov_type = SVPD_LBP_THIN; } ctsio->scsi_status = SCSI_STATUS_OK; 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; struct ctl_lun *lun; int alloc_len, retval; lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; 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; 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 0 /* No longer needed for HA */ if (((ctl_softc->flags & CTL_FLAG_MASTER_SHELF) == 0) && ((entry->flags & CTL_CMD_FLAG_OK_ON_SECONDARY) == 0)) { ctl_set_lun_standby(ctsio); retval = 1; goto bailout; } #endif 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) && (ctl_softc->ctl_luns[targ_lun] != NULL)) { lun = ctl_softc->ctl_luns[targ_lun]; /* * If the LUN is invalid, pretend that it doesn't exist. * It will go away as soon as all pending I/O has been * completed. */ if (lun->flags & CTL_LUN_DISABLED) { lun = 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. */ mtx_lock(&lun->lun_lock); 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; 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)); fe_done(io); goto bailout; } ctl_softc = lun->ctl_softc; 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)) 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); /* * 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); bailout: 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 ) { ctl_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(); ctl_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)) ctl_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.h =================================================================== --- head/sys/cam/ctl/ctl.h (revision 274153) +++ head/sys/cam/ctl/ctl.h (revision 274154) @@ -1,211 +1,215 @@ /*- * Copyright (c) 2003 Silicon Graphics International Corp. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl.h#5 $ * $FreeBSD$ */ /* * Function definitions used both within CTL and potentially in various CTL * clients. * * Author: Ken Merry */ #ifndef _CTL_H_ #define _CTL_H_ #define ctl_min(x,y) (((x) < (y)) ? (x) : (y)) #define CTL_RETVAL_COMPLETE 0 #define CTL_RETVAL_QUEUED 1 #define CTL_RETVAL_ALLOCATED 2 #define CTL_RETVAL_ERROR 3 typedef enum { CTL_PORT_NONE = 0x00, CTL_PORT_FC = 0x01, CTL_PORT_SCSI = 0x02, CTL_PORT_IOCTL = 0x04, CTL_PORT_INTERNAL = 0x08, CTL_PORT_ISCSI = 0x10, CTL_PORT_SAS = 0x20, CTL_PORT_ALL = 0xff, CTL_PORT_ISC = 0x100 // FC port for inter-shelf communication } ctl_port_type; struct ctl_port_entry { ctl_port_type port_type; char port_name[64]; int32_t targ_port; int physical_port; int virtual_port; u_int flags; #define CTL_PORT_WWNN_VALID 0x01 #define CTL_PORT_WWPN_VALID 0x02 uint64_t wwnn; uint64_t wwpn; int online; }; struct ctl_modepage_header { uint8_t page_code; uint8_t subpage; int32_t len_used; int32_t len_left; }; struct ctl_modepage_aps { struct ctl_modepage_header header; uint8_t lock_active; }; union ctl_modepage_info { struct ctl_modepage_header header; struct ctl_modepage_aps aps; }; /* * Serial number length, for VPD page 0x80. */ #define CTL_SN_LEN 16 /* * Device ID length, for VPD page 0x83. */ #define CTL_DEVID_LEN 64 #define CTL_DEVID_MIN_LEN 16 /* * WWPN length, for VPD page 0x83. */ #define CTL_WWPN_LEN 8 #define CTL_DRIVER_NAME_LEN 32 /* * Unit attention types. ASC/ASCQ values for these should be placed in * ctl_build_ua. These are also listed in order of reporting priority. * i.e. a poweron UA is reported first, bus reset second, etc. */ typedef enum { CTL_UA_NONE = 0x0000, CTL_UA_POWERON = 0x0001, CTL_UA_BUS_RESET = 0x0002, CTL_UA_TARG_RESET = 0x0004, CTL_UA_I_T_NEXUS_LOSS = 0x0008, CTL_UA_LUN_RESET = 0x0010, CTL_UA_LUN_CHANGE = 0x0020, CTL_UA_MODE_CHANGE = 0x0040, CTL_UA_LOG_CHANGE = 0x0080, CTL_UA_LVD = 0x0100, CTL_UA_SE = 0x0200, CTL_UA_RES_PREEMPT = 0x0400, CTL_UA_RES_RELEASE = 0x0800, CTL_UA_REG_PREEMPT = 0x1000, CTL_UA_ASYM_ACC_CHANGE = 0x2000, - CTL_UA_CAPACITY_CHANGED = 0x4000 + CTL_UA_CAPACITY_CHANGED = 0x4000, + CTL_UA_THIN_PROV_THRES = 0x8000 } ctl_ua_type; #ifdef _KERNEL MALLOC_DECLARE(M_CTL); struct ctl_page_index; #ifdef SYSCTL_DECL /* from sysctl.h */ SYSCTL_DECL(_kern_cam_ctl); #endif /* * Call these routines to enable or disable front end ports. */ int ctl_port_enable(ctl_port_type port_type); int ctl_port_disable(ctl_port_type port_type); /* * This routine grabs a list of frontend ports. */ 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); /* * Put a string into an sbuf, escaping characters that are illegal or not * recommended in XML. Note this doesn't escape everything, just > < and &. */ int ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size); int ctl_ffz(uint32_t *mask, uint32_t size); int ctl_set_mask(uint32_t *mask, uint32_t bit); int ctl_clear_mask(uint32_t *mask, uint32_t bit); int ctl_is_set(uint32_t *mask, uint32_t bit); int ctl_caching_sp_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr); int ctl_control_page_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr); /** int ctl_failover_sp_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr); **/ int ctl_debugconf_sp_sense_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc); int ctl_debugconf_sp_select_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr); +int ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio, + struct ctl_page_index *page_index, + int pc); int ctl_config_move_done(union ctl_io *io); void ctl_datamove(union ctl_io *io); void ctl_done(union ctl_io *io); void ctl_data_submit_done(union ctl_io *io); void ctl_config_write_done(union ctl_io *io); void ctl_portDB_changed(int portnum); void ctl_init_isc_msg(void); /* * KPI to manipulate LUN/port options */ struct ctl_option { STAILQ_ENTRY(ctl_option) links; char *name; char *value; }; typedef STAILQ_HEAD(ctl_options, ctl_option) ctl_options_t; struct ctl_be_arg; void ctl_init_opts(ctl_options_t *opts, int num_args, struct ctl_be_arg *args); void ctl_free_opts(ctl_options_t *opts); char * ctl_get_opt(ctl_options_t *opts, const char *name); #endif /* _KERNEL */ #endif /* _CTL_H_ */ /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_backend.h =================================================================== --- head/sys/cam/ctl/ctl_backend.h (revision 274153) +++ head/sys/cam/ctl/ctl_backend.h (revision 274154) @@ -1,301 +1,303 @@ /*- * Copyright (c) 2003 Silicon Graphics International Corp. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce at minimum a disclaimer * substantially similar to the "NO WARRANTY" disclaimer below * ("Disclaimer") and any redistribution must be conditioned upon * including a substantially similar Disclaimer requirement for further * binary redistribution. * * NO WARRANTY * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGES. * * $Id: //depot/users/kenm/FreeBSD-test2/sys/cam/ctl/ctl_backend.h#2 $ * $FreeBSD$ */ /* * CTL backend driver definitions * * Author: Ken Merry */ #ifndef _CTL_BACKEND_H_ #define _CTL_BACKEND_H_ /* * XXX KDM move this to another header file? */ #define CTL_BE_NAME_LEN 32 /* * The ID_REQ flag is used to say that the caller has requested a * particular LUN ID in the req_lun_id field. If we cannot allocate that * LUN ID, the ctl_add_lun() call will fail. * * The POWERED_OFF flag tells us that the LUN should default to the powered * off state. It will return 0x04,0x02 until it is powered up. ("Logical * unit not ready, initializing command required.") * * The INOPERABLE flag tells us that this LUN is not operable for whatever * reason. This means that user data may have been (or has been?) lost. * We will return 0x31,0x00 ("Medium format corrupted") until the host * issues a FORMAT UNIT command to clear the error. * * The PRIMARY flag tells us that this LUN is registered as a Primary LUN * which is accessible via the Master shelf controller in an HA. This flag * being set indicates a Primary LUN. This flag being reset represents a * Secondary LUN controlled by the Secondary controller in an HA * configuration. Flag is applicable at this time to T_DIRECT types. * * The SERIAL_NUM flag tells us that the serial_num field is filled in and * valid for use in SCSI INQUIRY VPD page 0x80. * * The DEVID flag tells us that the device_id field is filled in and * valid for use in SCSI INQUIRY VPD page 0x83. * * The DEV_TYPE flag tells us that the device_type field is filled in. * * The UNMAP flag tells us that this LUN supports UNMAP. * * The OFFLINE flag tells us that this LUN can not access backing store. */ typedef enum { CTL_LUN_FLAG_ID_REQ = 0x01, CTL_LUN_FLAG_POWERED_OFF = 0x02, CTL_LUN_FLAG_INOPERABLE = 0x04, CTL_LUN_FLAG_PRIMARY = 0x08, CTL_LUN_FLAG_SERIAL_NUM = 0x10, CTL_LUN_FLAG_DEVID = 0x20, CTL_LUN_FLAG_DEV_TYPE = 0x40, CTL_LUN_FLAG_UNMAP = 0x80, CTL_LUN_FLAG_OFFLINE = 0x100 } ctl_backend_lun_flags; #ifdef _KERNEL #define CTL_BACKEND_DECLARE(name, driver) \ static int name ## _modevent(module_t mod, int type, void *data) \ { \ switch (type) { \ case MOD_LOAD: \ ctl_backend_register( \ (struct ctl_backend_driver *)data); \ break; \ case MOD_UNLOAD: \ printf(#name " module unload - not possible for this module type\n"); \ return EINVAL; \ default: \ return EOPNOTSUPP; \ } \ return 0; \ } \ static moduledata_t name ## _mod = { \ #name, \ name ## _modevent, \ (void *)&driver \ }; \ DECLARE_MODULE(name, name ## _mod, SI_SUB_CONFIGURE, SI_ORDER_FOURTH); \ MODULE_DEPEND(name, ctl, 1, 1, 1); \ MODULE_DEPEND(name, cam, 1, 1, 1) typedef enum { CTL_LUN_CONFIG_OK, CTL_LUN_CONFIG_FAILURE } ctl_lun_config_status; typedef void (*be_callback_t)(void *be_lun); typedef void (*be_lun_config_t)(void *be_lun, ctl_lun_config_status status); /* * The lun_type field is the SCSI device type of this particular LUN. In * general, this should be T_DIRECT, although backends will want to create * a processor LUN, typically at LUN 0. See scsi_all.h for the defines for * the various SCSI device types. * * The flags are described above. * * The be_lun field is the backend driver's own context that will get * passsed back so that it can tell which LUN CTL is referencing. * * maxlba is the maximum accessible LBA on the LUN. Note that this is * different from the capacity of the array. capacity = maxlba + 1 * * blocksize is the size, in bytes, of each LBA on the LUN. In general * this should be 512. In theory CTL should be able to handle other block * sizes. Host application software may not deal with it very well, though. * * pblockexp is the log2() of number of LBAs on the LUN per physical sector. * * pblockoff is the lowest LBA on the LUN aligned ot physical sector. * * atomicblock is the number of blocks that can be written atomically. * * req_lun_id is the requested LUN ID. CTL only pays attention to this * field if the CTL_LUN_FLAG_ID_REQ flag is set. If the requested LUN ID is * not available, the LUN addition will fail. If a particular LUN ID isn't * requested, the first available LUN ID will be allocated. * * serial_num is the device serial number returned in the SCSI INQUIRY VPD * page 0x80. This should be a unique, per-shelf value. The data inside * this field should be ASCII only, left aligned, and any unused space * should be padded out with ASCII spaces. This field should NOT be NULL * terminated. * * device_id is the T10 device identifier returned in the SCSI INQUIRY VPD * page 0x83. This should be a unique, per-LUN value. The data inside * this field should be ASCII only, left aligned, and any unused space * should be padded with ASCII spaces. This field should NOT be NULL * terminated. * * The lun_shutdown() method is the callback for the ctl_invalidate_lun() * call. It is called when all outstanding I/O for that LUN has been * completed and CTL has deleted the resources for that LUN. When the CTL * backend gets this call, it can safely free its per-LUN resources. * * The lun_config_status() method is the callback for the ctl_add_lun() * call. It is called when the LUN is successfully added, or when LUN * addition fails. If the LUN is successfully added, the backend may call * the ctl_enable_lun() method to enable the LUN. * * The be field is a pointer to the ctl_backend_driver structure, which * contains the backend methods to be called by CTL. * * The ctl_lun field is for CTL internal use only, and should not be used * by the backend. * * The links field is for CTL internal use only, and should not be used by * the backend. */ struct ctl_be_lun { uint8_t lun_type; /* passed to CTL */ ctl_backend_lun_flags flags; /* passed to CTL */ void *be_lun; /* passed to CTL */ uint64_t maxlba; /* passed to CTL */ uint32_t blocksize; /* passed to CTL */ uint16_t pblockexp; /* passed to CTL */ uint16_t pblockoff; /* passed to CTL */ uint32_t atomicblock; /* passed to CTL */ uint32_t req_lun_id; /* passed to CTL */ uint32_t lun_id; /* returned from CTL */ uint8_t serial_num[CTL_SN_LEN]; /* passed to CTL */ uint8_t device_id[CTL_DEVID_LEN];/* passed to CTL */ be_callback_t lun_shutdown; /* passed to CTL */ be_lun_config_t lun_config_status; /* passed to CTL */ struct ctl_backend_driver *be; /* passed to CTL */ void *ctl_lun; /* used by CTL */ ctl_options_t options; /* passed to CTL */ STAILQ_ENTRY(ctl_be_lun) links; /* used by CTL */ }; typedef enum { CTL_BE_FLAG_NONE = 0x00, /* no flags */ CTL_BE_FLAG_HAS_CONFIG = 0x01, /* can do config reads, writes */ CTL_BE_FLAG_INTERNAL = 0x02 /* don't inc mod refcount */ } ctl_backend_flags; typedef int (*be_init_t)(void); typedef int (*be_func_t)(union ctl_io *io); typedef void (*be_vfunc_t)(union ctl_io *io); typedef int (*be_ioctl_t)(struct cdev *dev, u_long cmd, caddr_t addr, int flag, struct thread *td); typedef int (*be_luninfo_t)(void *be_lun, struct sbuf *sb); +typedef uint64_t (*be_lunattr_t)(void *be_lun, const char *attrname); struct ctl_backend_driver { char name[CTL_BE_NAME_LEN]; /* passed to CTL */ ctl_backend_flags flags; /* passed to CTL */ be_init_t init; /* passed to CTL */ be_func_t data_submit; /* passed to CTL */ be_func_t data_move_done; /* passed to CTL */ be_func_t config_read; /* passed to CTL */ be_func_t config_write; /* passed to CTL */ be_ioctl_t ioctl; /* passed to CTL */ be_luninfo_t lun_info; /* passed to CTL */ + be_lunattr_t lun_attr; /* passed to CTL */ #ifdef CS_BE_CONFIG_MOVE_DONE_IS_NOT_USED be_func_t config_move_done; /* passed to backend */ #endif #if 0 be_vfunc_t config_write_done; /* passed to backend */ #endif u_int num_luns; /* used by CTL */ STAILQ_ENTRY(ctl_backend_driver) links; /* used by CTL */ }; int ctl_backend_register(struct ctl_backend_driver *be); int ctl_backend_deregister(struct ctl_backend_driver *be); struct ctl_backend_driver *ctl_backend_find(char *backend_name); /* * To add a LUN, first call ctl_add_lun(). You will get the lun_config_status() * callback when the LUN addition has either succeeded or failed. * * Once you get that callback, you can then call ctl_enable_lun() to enable * the LUN. */ int ctl_add_lun(struct ctl_be_lun *be_lun); int ctl_enable_lun(struct ctl_be_lun *be_lun); /* * To delete a LUN, first call ctl_disable_lun(), then * ctl_invalidate_lun(). You will get the lun_shutdown() callback when all * I/O to the LUN has completed and the LUN has been deleted. */ int ctl_disable_lun(struct ctl_be_lun *be_lun); int ctl_invalidate_lun(struct ctl_be_lun *be_lun); /* * To start a LUN (transition from powered off to powered on state) call * ctl_start_lun(). To stop a LUN (transition from powered on to powered * off state) call ctl_stop_lun(). */ int ctl_start_lun(struct ctl_be_lun *be_lun); int ctl_stop_lun(struct ctl_be_lun *be_lun); /* * If a LUN is inoperable, call ctl_lun_inoperable(). Generally the LUN * will become operable once again when the user issues the SCSI FORMAT UNIT * command. (CTL will automatically clear the inoperable flag.) If we * need to re-enable the LUN, we can call ctl_lun_operable() to enable it * without a SCSI command. */ int ctl_lun_inoperable(struct ctl_be_lun *be_lun); int ctl_lun_operable(struct ctl_be_lun *be_lun); /* * To take a LUN offline, call ctl_lun_offline(). Generally the LUN will * be online again once the user sends a SCSI START STOP UNIT command with * the start and on/offline bits set. The backend can bring the LUN back * online via the ctl_lun_online() function, if necessary. */ int ctl_lun_offline(struct ctl_be_lun *be_lun); int ctl_lun_online(struct ctl_be_lun *be_lun); /* * Let the backend notify the initiator about changed capacity. */ void ctl_lun_capacity_changed(struct ctl_be_lun *be_lun); #endif /* _KERNEL */ #endif /* _CTL_BACKEND_H_ */ /* * vim: ts=8 */ Index: head/sys/cam/ctl/ctl_backend_block.c =================================================================== --- head/sys/cam/ctl/ctl_backend_block.c (revision 274153) +++ head/sys/cam/ctl/ctl_backend_block.c (revision 274154) @@ -1,2601 +1,2637 @@ /*- * 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_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) && ((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)) { /* * 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 { #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 { #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 { #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 = 0; + unmap = (be_lun->dispatch == ctl_be_block_dispatch_zvol); value = ctl_get_opt(&be_lun->ctl_be_lun.options, "unmap"); - if (value != NULL && strcmp(value, "on") == 0) - unmap = 1; + 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_blocks; 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_blocks != 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_error.c =================================================================== --- head/sys/cam/ctl/ctl_error.c (revision 274153) +++ head/sys/cam/ctl/ctl_error.c (revision 274154) @@ -1,830 +1,835 @@ /*- * Copyright (c) 2003-2009 Silicon Graphics International Corp. * Copyright (c) 2011 Spectra Logic Corporation * 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_error.c#2 $ */ /* * CAM Target Layer error reporting routines. * * Author: Ken Merry */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include void ctl_set_sense_data_va(struct scsi_sense_data *sense_data, void *lunptr, scsi_sense_data_type sense_format, int current_error, int sense_key, int asc, int ascq, va_list ap) { struct ctl_lun *lun; lun = (struct ctl_lun *)lunptr; /* * Determine whether to return fixed or descriptor format sense * data. */ if (sense_format == SSD_TYPE_NONE) { /* * If the format isn't specified, we only return descriptor * sense if the LUN exists and descriptor sense is turned * on for that LUN. */ if ((lun != NULL) && (lun->flags & CTL_LUN_SENSE_DESC)) sense_format = SSD_TYPE_DESC; else sense_format = SSD_TYPE_FIXED; } scsi_set_sense_data_va(sense_data, sense_format, current_error, sense_key, asc, ascq, ap); } void ctl_set_sense_data(struct scsi_sense_data *sense_data, void *lunptr, scsi_sense_data_type sense_format, int current_error, int sense_key, int asc, int ascq, ...) { va_list ap; va_start(ap, ascq); ctl_set_sense_data_va(sense_data, lunptr, sense_format, current_error, sense_key, asc, ascq, ap); va_end(ap); } void ctl_set_sense(struct ctl_scsiio *ctsio, int current_error, int sense_key, int asc, int ascq, ...) { va_list ap; struct ctl_lun *lun; /* * The LUN can't go away until all of the commands have been * completed. Therefore we can safely access the LUN structure and * flags without the lock. */ lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr; va_start(ap, ascq); ctl_set_sense_data_va(&ctsio->sense_data, lun, SSD_TYPE_NONE, current_error, sense_key, asc, ascq, ap); va_end(ap); ctsio->scsi_status = SCSI_STATUS_CHECK_COND; ctsio->sense_len = SSD_FULL_SIZE; ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE; } /* * Transform fixed sense data into descriptor sense data. * * For simplicity's sake, we assume that both sense structures are * SSD_FULL_SIZE. Otherwise, the logic gets more complicated. */ void ctl_sense_to_desc(struct scsi_sense_data_fixed *sense_src, struct scsi_sense_data_desc *sense_dest) { struct scsi_sense_stream stream_sense; int current_error; uint8_t stream_bits; bzero(sense_dest, sizeof(*sense_dest)); if ((sense_src->error_code & SSD_ERRCODE) == SSD_DEFERRED_ERROR) current_error = 0; else current_error = 1; bzero(&stream_sense, sizeof(stream_sense)); /* * Check to see whether any of the tape-specific bits are set. If * so, we'll need a stream sense descriptor. */ if (sense_src->flags & (SSD_ILI|SSD_EOM|SSD_FILEMARK)) stream_bits = sense_src->flags & ~SSD_KEY; else stream_bits = 0; /* * Utilize our sense setting routine to do the transform. If a * value is set in the fixed sense data, set it in the descriptor * data. Otherwise, skip it. */ ctl_set_sense_data((struct scsi_sense_data *)sense_dest, /*lun*/ NULL, /*sense_format*/ SSD_TYPE_DESC, current_error, /*sense_key*/ sense_src->flags & SSD_KEY, /*asc*/ sense_src->add_sense_code, /*ascq*/ sense_src->add_sense_code_qual, /* Information Bytes */ (scsi_4btoul(sense_src->info) != 0) ? SSD_ELEM_INFO : SSD_ELEM_SKIP, sizeof(sense_src->info), sense_src->info, /* Command specific bytes */ (scsi_4btoul(sense_src->cmd_spec_info) != 0) ? SSD_ELEM_COMMAND : SSD_ELEM_SKIP, sizeof(sense_src->cmd_spec_info), sense_src->cmd_spec_info, /* FRU */ (sense_src->fru != 0) ? SSD_ELEM_FRU : SSD_ELEM_SKIP, sizeof(sense_src->fru), &sense_src->fru, /* Sense Key Specific */ (sense_src->sense_key_spec[0] & SSD_SCS_VALID) ? SSD_ELEM_SKS : SSD_ELEM_SKIP, sizeof(sense_src->sense_key_spec), sense_src->sense_key_spec, /* Tape bits */ (stream_bits != 0) ? SSD_ELEM_STREAM : SSD_ELEM_SKIP, sizeof(stream_bits), &stream_bits, SSD_ELEM_NONE); } /* * Transform descriptor format sense data into fixed sense data. * * Some data may be lost in translation, because there are descriptors * thant can't be represented as fixed sense data. * * For simplicity's sake, we assume that both sense structures are * SSD_FULL_SIZE. Otherwise, the logic gets more complicated. */ void ctl_sense_to_fixed(struct scsi_sense_data_desc *sense_src, struct scsi_sense_data_fixed *sense_dest) { int current_error; uint8_t *info_ptr = NULL, *cmd_ptr = NULL, *fru_ptr = NULL; uint8_t *sks_ptr = NULL, *stream_ptr = NULL; int info_size = 0, cmd_size = 0, fru_size = 0; int sks_size = 0, stream_size = 0; int pos; if ((sense_src->error_code & SSD_ERRCODE) == SSD_DESC_CURRENT_ERROR) current_error = 1; else current_error = 0; for (pos = 0; pos < (int)(sense_src->extra_len - 1);) { struct scsi_sense_desc_header *header; header = (struct scsi_sense_desc_header *) &sense_src->sense_desc[pos]; /* * See if this record goes past the end of the sense data. * It shouldn't, but check just in case. */ if ((pos + header->length + sizeof(*header)) > sense_src->extra_len) break; switch (sense_src->sense_desc[pos]) { case SSD_DESC_INFO: { struct scsi_sense_info *info; info = (struct scsi_sense_info *)header; info_ptr = info->info; info_size = sizeof(info->info); pos += info->length + sizeof(struct scsi_sense_desc_header); break; } case SSD_DESC_COMMAND: { struct scsi_sense_command *cmd; cmd = (struct scsi_sense_command *)header; cmd_ptr = cmd->command_info; cmd_size = sizeof(cmd->command_info); pos += cmd->length + sizeof(struct scsi_sense_desc_header); break; } case SSD_DESC_FRU: { struct scsi_sense_fru *fru; fru = (struct scsi_sense_fru *)header; fru_ptr = &fru->fru; fru_size = sizeof(fru->fru); pos += fru->length + sizeof(struct scsi_sense_desc_header); break; } case SSD_DESC_SKS: { struct scsi_sense_sks *sks; sks = (struct scsi_sense_sks *)header; sks_ptr = sks->sense_key_spec; sks_size = sizeof(sks->sense_key_spec); pos = sks->length + sizeof(struct scsi_sense_desc_header); break; } case SSD_DESC_STREAM: { struct scsi_sense_stream *stream_sense; stream_sense = (struct scsi_sense_stream *)header; stream_ptr = &stream_sense->byte3; stream_size = sizeof(stream_sense->byte3); pos = stream_sense->length + sizeof(struct scsi_sense_desc_header); break; } default: /* * We don't recognize this particular sense * descriptor type, so just skip it. */ pos += sizeof(*header) + header->length; break; } } ctl_set_sense_data((struct scsi_sense_data *)sense_dest, /*lun*/ NULL, /*sense_format*/ SSD_TYPE_FIXED, current_error, /*sense_key*/ sense_src->sense_key & SSD_KEY, /*asc*/ sense_src->add_sense_code, /*ascq*/ sense_src->add_sense_code_qual, /* Information Bytes */ (info_ptr != NULL) ? SSD_ELEM_INFO : SSD_ELEM_SKIP, info_size, info_ptr, /* Command specific bytes */ (cmd_ptr != NULL) ? SSD_ELEM_COMMAND : SSD_ELEM_SKIP, cmd_size, cmd_ptr, /* FRU */ (fru_ptr != NULL) ? SSD_ELEM_FRU : SSD_ELEM_SKIP, fru_size, fru_ptr, /* Sense Key Specific */ (sks_ptr != NULL) ? SSD_ELEM_SKS : SSD_ELEM_SKIP, sks_size, sks_ptr, /* Tape bits */ (stream_ptr != NULL) ? SSD_ELEM_STREAM : SSD_ELEM_SKIP, stream_size, stream_ptr, SSD_ELEM_NONE); } void ctl_set_ua(struct ctl_scsiio *ctsio, int asc, int ascq) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_UNIT_ATTENTION, asc, ascq, SSD_ELEM_NONE); } ctl_ua_type ctl_build_ua(ctl_ua_type *ua_type, struct scsi_sense_data *sense, scsi_sense_data_type sense_format) { ctl_ua_type ua_to_build, ua_to_clear; int asc, ascq; if (*ua_type == CTL_UA_NONE) return (CTL_UA_NONE); ua_to_build = (1 << (ffs(*ua_type) - 1)); ua_to_clear = ua_to_build; switch (ua_to_build) { case CTL_UA_POWERON: /* 29h/01h POWER ON OCCURRED */ asc = 0x29; ascq = 0x01; ua_to_clear = ~0; break; case CTL_UA_BUS_RESET: /* 29h/02h SCSI BUS RESET OCCURRED */ asc = 0x29; ascq = 0x02; ua_to_clear = ~0; break; case CTL_UA_TARG_RESET: /* 29h/03h BUS DEVICE RESET FUNCTION OCCURRED*/ asc = 0x29; ascq = 0x03; ua_to_clear = ~0; break; case CTL_UA_I_T_NEXUS_LOSS: /* 29h/07h I_T NEXUS LOSS OCCURRED */ asc = 0x29; ascq = 0x07; ua_to_clear = ~0; break; case CTL_UA_LUN_RESET: /* 29h/00h POWER ON, RESET, OR BUS DEVICE RESET OCCURRED */ /* * Since we don't have a specific ASC/ASCQ pair for a LUN * reset, just return the generic reset code. */ asc = 0x29; ascq = 0x00; break; case CTL_UA_LUN_CHANGE: /* 3Fh/0Eh REPORTED LUNS DATA HAS CHANGED */ asc = 0x3F; ascq = 0x0E; break; case CTL_UA_MODE_CHANGE: /* 2Ah/01h MODE PARAMETERS CHANGED */ asc = 0x2A; ascq = 0x01; break; case CTL_UA_LOG_CHANGE: /* 2Ah/02h LOG PARAMETERS CHANGED */ asc = 0x2A; ascq = 0x02; break; case CTL_UA_LVD: /* 29h/06h TRANSCEIVER MODE CHANGED TO LVD */ asc = 0x29; ascq = 0x06; break; case CTL_UA_SE: /* 29h/05h TRANSCEIVER MODE CHANGED TO SINGLE-ENDED */ asc = 0x29; ascq = 0x05; break; case CTL_UA_RES_PREEMPT: /* 2Ah/03h RESERVATIONS PREEMPTED */ asc = 0x2A; ascq = 0x03; break; case CTL_UA_RES_RELEASE: /* 2Ah/04h RESERVATIONS RELEASED */ asc = 0x2A; ascq = 0x04; break; case CTL_UA_REG_PREEMPT: /* 2Ah/05h REGISTRATIONS PREEMPTED */ asc = 0x2A; ascq = 0x05; break; case CTL_UA_ASYM_ACC_CHANGE: /* 2Ah/06n ASYMMETRIC ACCESS STATE CHANGED */ asc = 0x2A; ascq = 0x06; break; case CTL_UA_CAPACITY_CHANGED: /* 2Ah/09n CAPACITY DATA HAS CHANGED */ asc = 0x2A; ascq = 0x09; break; + case CTL_UA_THIN_PROV_THRES: + /* 38h/07n THIN PROVISIONING SOFT THRESHOLD REACHED */ + asc = 0x38; + ascq = 0x07; + break; default: panic("ctl_build_ua: Unknown UA %x", ua_to_build); } ctl_set_sense_data(sense, /*lun*/ NULL, sense_format, /*current_error*/ 1, /*sense_key*/ SSD_KEY_UNIT_ATTENTION, asc, ascq, SSD_ELEM_NONE); /* We're reporting this UA, so clear it */ *ua_type &= ~ua_to_clear; return (ua_to_build); } void ctl_set_overlapped_cmd(struct ctl_scsiio *ctsio) { /* OVERLAPPED COMMANDS ATTEMPTED */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x4E, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_overlapped_tag(struct ctl_scsiio *ctsio, uint8_t tag) { /* TAGGED OVERLAPPED COMMANDS (NN = QUEUE TAG) */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x4D, /*ascq*/ tag, SSD_ELEM_NONE); } /* * Tell the user that there was a problem with the command or data he sent. */ void ctl_set_invalid_field(struct ctl_scsiio *ctsio, int sks_valid, int command, int field, int bit_valid, int bit) { uint8_t sks[3]; int asc; if (command != 0) { /* "Invalid field in CDB" */ asc = 0x24; } else { /* "Invalid field in parameter list" */ asc = 0x26; } if (sks_valid) { sks[0] = SSD_SCS_VALID; if (command) sks[0] |= SSD_FIELDPTR_CMD; scsi_ulto2b(field, &sks[1]); if (bit_valid) sks[0] |= SSD_BITPTR_VALID | bit; } ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, asc, /*ascq*/ 0x00, /*type*/ (sks_valid != 0) ? SSD_ELEM_SKS : SSD_ELEM_SKIP, /*size*/ sizeof(sks), /*data*/ sks, SSD_ELEM_NONE); } void ctl_set_invalid_opcode(struct ctl_scsiio *ctsio) { struct scsi_sense_data *sense; uint8_t sks[3]; sense = &ctsio->sense_data; sks[0] = SSD_SCS_VALID | SSD_FIELDPTR_CMD; scsi_ulto2b(0, &sks[1]); /* "Invalid command operation code" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x20, /*ascq*/ 0x00, /*type*/ SSD_ELEM_SKS, /*size*/ sizeof(sks), /*data*/ sks, SSD_ELEM_NONE); } void ctl_set_param_len_error(struct ctl_scsiio *ctsio) { /* "Parameter list length error" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x1a, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_already_locked(struct ctl_scsiio *ctsio) { /* Vendor unique "Somebody already is locked" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x81, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_unsupported_lun(struct ctl_scsiio *ctsio) { /* "Logical unit not supported" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x25, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_internal_failure(struct ctl_scsiio *ctsio, int sks_valid, uint16_t retry_count) { uint8_t sks[3]; if (sks_valid) { sks[0] = SSD_SCS_VALID; sks[1] = (retry_count >> 8) & 0xff; sks[2] = retry_count & 0xff; } /* "Internal target failure" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_HARDWARE_ERROR, /*asc*/ 0x44, /*ascq*/ 0x00, /*type*/ (sks_valid != 0) ? SSD_ELEM_SKS : SSD_ELEM_SKIP, /*size*/ sizeof(sks), /*data*/ sks, SSD_ELEM_NONE); } void ctl_set_medium_error(struct ctl_scsiio *ctsio) { if ((ctsio->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) { /* "Unrecovered read error" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_MEDIUM_ERROR, /*asc*/ 0x11, /*ascq*/ 0x00, SSD_ELEM_NONE); } else { /* "Write error - auto reallocation failed" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_MEDIUM_ERROR, /*asc*/ 0x0C, /*ascq*/ 0x02, SSD_ELEM_NONE); } } void ctl_set_aborted(struct ctl_scsiio *ctsio) { ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ABORTED_COMMAND, /*asc*/ 0x45, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_lba_out_of_range(struct ctl_scsiio *ctsio) { /* "Logical block address out of range" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x21, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_lun_stopped(struct ctl_scsiio *ctsio) { /* "Logical unit not ready, initializing cmd. required" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_NOT_READY, /*asc*/ 0x04, /*ascq*/ 0x02, SSD_ELEM_NONE); } void ctl_set_lun_not_ready(struct ctl_scsiio *ctsio) { /* "Logical unit not ready, manual intervention required" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_NOT_READY, /*asc*/ 0x04, /*ascq*/ 0x03, SSD_ELEM_NONE); } void ctl_set_illegal_pr_release(struct ctl_scsiio *ctsio) { /* "Invalid release of persistent reservation" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST, /*asc*/ 0x26, /*ascq*/ 0x04, SSD_ELEM_NONE); } void ctl_set_lun_standby(struct ctl_scsiio *ctsio) { /* "Logical unit not ready, target port in standby state" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_NOT_READY, /*asc*/ 0x04, /*ascq*/ 0x0b, SSD_ELEM_NONE); } void ctl_set_medium_format_corrupted(struct ctl_scsiio *ctsio) { /* "Medium format corrupted" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_MEDIUM_ERROR, /*asc*/ 0x31, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_medium_magazine_inaccessible(struct ctl_scsiio *ctsio) { /* "Medium magazine not accessible" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_NOT_READY, /*asc*/ 0x3b, /*ascq*/ 0x11, SSD_ELEM_NONE); } void ctl_set_data_phase_error(struct ctl_scsiio *ctsio) { /* "Data phase error" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_NOT_READY, /*asc*/ 0x4b, /*ascq*/ 0x00, SSD_ELEM_NONE); } void ctl_set_reservation_conflict(struct ctl_scsiio *ctsio) { struct scsi_sense_data *sense; sense = &ctsio->sense_data; memset(sense, 0, sizeof(*sense)); ctsio->scsi_status = SCSI_STATUS_RESERV_CONFLICT; ctsio->sense_len = 0; ctsio->io_hdr.status = CTL_SCSI_ERROR; } void ctl_set_queue_full(struct ctl_scsiio *ctsio) { struct scsi_sense_data *sense; sense = &ctsio->sense_data; memset(sense, 0, sizeof(*sense)); ctsio->scsi_status = SCSI_STATUS_QUEUE_FULL; ctsio->sense_len = 0; ctsio->io_hdr.status = CTL_SCSI_ERROR; } void ctl_set_busy(struct ctl_scsiio *ctsio) { struct scsi_sense_data *sense; sense = &ctsio->sense_data; memset(sense, 0, sizeof(*sense)); ctsio->scsi_status = SCSI_STATUS_BUSY; ctsio->sense_len = 0; ctsio->io_hdr.status = CTL_SCSI_ERROR; } void ctl_set_task_aborted(struct ctl_scsiio *ctsio) { struct scsi_sense_data *sense; sense = &ctsio->sense_data; memset(sense, 0, sizeof(*sense)); ctsio->scsi_status = SCSI_STATUS_TASK_ABORTED; ctsio->sense_len = 0; ctsio->io_hdr.status = CTL_CMD_ABORTED; } void ctl_set_space_alloc_fail(struct ctl_scsiio *ctsio) { /* "Space allocation failed write protect" */ ctl_set_sense(ctsio, /*current_error*/ 1, /*sense_key*/ SSD_KEY_DATA_PROTECT, /*asc*/ 0x27, /*ascq*/ 0x07, SSD_ELEM_NONE); } void ctl_set_success(struct ctl_scsiio *ctsio) { struct scsi_sense_data *sense; sense = &ctsio->sense_data; memset(sense, 0, sizeof(*sense)); ctsio->scsi_status = SCSI_STATUS_OK; ctsio->sense_len = 0; ctsio->io_hdr.status = CTL_SUCCESS; } Index: head/sys/cam/ctl/ctl_private.h =================================================================== --- head/sys/cam/ctl/ctl_private.h (revision 274153) +++ head/sys/cam/ctl/ctl_private.h (revision 274154) @@ -1,548 +1,563 @@ /*- * Copyright (c) 2003, 2004, 2005, 2008 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_private.h#7 $ * $FreeBSD$ */ /* * CAM Target Layer driver private data structures/definitions. * * Author: Ken Merry */ #ifndef _CTL_PRIVATE_H_ #define _CTL_PRIVATE_H_ /* * SCSI vendor and product names. */ #define CTL_VENDOR "FREEBSD " #define CTL_DIRECT_PRODUCT "CTLDISK " #define CTL_PROCESSOR_PRODUCT "CTLPROCESSOR " #define CTL_UNKNOWN_PRODUCT "CTLDEVICE " struct ctl_fe_ioctl_startstop_info { struct cv sem; struct ctl_hard_startstop_info hs_info; }; struct ctl_fe_ioctl_bbrread_info { struct cv sem; struct ctl_bbrread_info *bbr_info; int wakeup_done; struct mtx *lock; }; typedef enum { CTL_IOCTL_INPROG, CTL_IOCTL_DATAMOVE, CTL_IOCTL_DONE } ctl_fe_ioctl_state; struct ctl_fe_ioctl_params { struct cv sem; struct mtx ioctl_mtx; ctl_fe_ioctl_state state; }; #define CTL_POOL_ENTRIES_INTERNAL 200 #define CTL_POOL_ENTRIES_EMERGENCY 300 #define CTL_POOL_ENTRIES_OTHER_SC 200 typedef enum { CTL_POOL_INTERNAL, CTL_POOL_FETD, CTL_POOL_EMERGENCY, CTL_POOL_4OTHERSC } ctl_pool_type; typedef enum { CTL_POOL_FLAG_NONE = 0x00, CTL_POOL_FLAG_INVALID = 0x01 } ctl_pool_flags; struct ctl_io_pool { ctl_pool_type type; ctl_pool_flags flags; uint32_t id; struct ctl_softc *ctl_softc; uint32_t refcount; uint64_t total_allocated; uint64_t total_freed; int32_t total_ctl_io; int32_t free_ctl_io; STAILQ_HEAD(, ctl_io_hdr) free_queue; STAILQ_ENTRY(ctl_io_pool) links; }; typedef enum { CTL_IOCTL_FLAG_NONE = 0x00, CTL_IOCTL_FLAG_ENABLED = 0x01 } ctl_ioctl_flags; struct ctl_ioctl_info { ctl_ioctl_flags flags; uint32_t cur_tag_num; struct ctl_port port; char port_name[24]; }; typedef enum { CTL_SER_BLOCK, CTL_SER_BLOCKOPT, CTL_SER_EXTENT, CTL_SER_EXTENTOPT, CTL_SER_PASS, CTL_SER_SKIP } ctl_serialize_action; typedef enum { CTL_ACTION_BLOCK, CTL_ACTION_OVERLAP, CTL_ACTION_OVERLAP_TAG, CTL_ACTION_PASS, CTL_ACTION_SKIP, CTL_ACTION_ERROR } ctl_action; /* * WARNING: Keep the bottom nibble here free, we OR in the data direction * flags for each command. * * Note: "OK_ON_ALL_LUNS" == we don't have to have a lun configured * "OK_ON_BOTH" == we have to have a lun configured * "SA5" == command has 5-bit service action at byte 1 */ typedef enum { CTL_CMD_FLAG_NONE = 0x0000, CTL_CMD_FLAG_NO_SENSE = 0x0010, CTL_CMD_FLAG_OK_ON_ALL_LUNS = 0x0020, CTL_CMD_FLAG_ALLOW_ON_RESV = 0x0040, CTL_CMD_FLAG_ALLOW_ON_PR_WRESV = 0x0080, CTL_CMD_FLAG_OK_ON_PROC = 0x0100, CTL_CMD_FLAG_OK_ON_SLUN = 0x0200, CTL_CMD_FLAG_OK_ON_BOTH = 0x0300, CTL_CMD_FLAG_OK_ON_STOPPED = 0x0400, CTL_CMD_FLAG_OK_ON_INOPERABLE = 0x0800, CTL_CMD_FLAG_OK_ON_OFFLINE = 0x1000, CTL_CMD_FLAG_OK_ON_SECONDARY = 0x2000, CTL_CMD_FLAG_ALLOW_ON_PR_RESV = 0x4000, CTL_CMD_FLAG_SA5 = 0x8000 } ctl_cmd_flags; typedef enum { CTL_SERIDX_TUR = 0, CTL_SERIDX_READ, CTL_SERIDX_WRITE, CTL_SERIDX_UNMAP, CTL_SERIDX_MD_SNS, CTL_SERIDX_MD_SEL, CTL_SERIDX_RQ_SNS, CTL_SERIDX_INQ, CTL_SERIDX_RD_CAP, CTL_SERIDX_RES, CTL_SERIDX_LOG_SNS, CTL_SERIDX_FORMAT, CTL_SERIDX_START, /* TBD: others to be filled in as needed */ CTL_SERIDX_COUNT, /* LAST, not a normal code, provides # codes */ CTL_SERIDX_INVLD = CTL_SERIDX_COUNT } ctl_seridx; typedef int ctl_opfunc(struct ctl_scsiio *ctsio); struct ctl_cmd_entry { ctl_opfunc *execute; ctl_seridx seridx; ctl_cmd_flags flags; ctl_lun_error_pattern pattern; uint8_t length; /* CDB length */ uint8_t usage[15]; /* Mask of allowed CDB bits * after the opcode byte. */ }; typedef enum { CTL_LUN_NONE = 0x000, CTL_LUN_CONTROL = 0x001, CTL_LUN_RESERVED = 0x002, CTL_LUN_INVALID = 0x004, CTL_LUN_DISABLED = 0x008, CTL_LUN_MALLOCED = 0x010, CTL_LUN_STOPPED = 0x020, CTL_LUN_INOPERABLE = 0x040, CTL_LUN_OFFLINE = 0x080, CTL_LUN_PR_RESERVED = 0x100, CTL_LUN_PRIMARY_SC = 0x200, CTL_LUN_SENSE_DESC = 0x400, CTL_LUN_READONLY = 0x800 } ctl_lun_flags; typedef enum { CTLBLOCK_FLAG_NONE = 0x00, CTLBLOCK_FLAG_INVALID = 0x01 } ctlblock_flags; union ctl_softcs { struct ctl_softc *ctl_softc; struct ctlblock_softc *ctlblock_softc; }; /* * Mode page defaults. */ #if 0 /* * These values make Solaris trim off some of the capacity. */ #define CTL_DEFAULT_SECTORS_PER_TRACK 63 #define CTL_DEFAULT_HEADS 255 /* * These values seem to work okay. */ #define CTL_DEFAULT_SECTORS_PER_TRACK 63 #define CTL_DEFAULT_HEADS 16 /* * These values work reasonably well. */ #define CTL_DEFAULT_SECTORS_PER_TRACK 512 #define CTL_DEFAULT_HEADS 64 #endif /* * Solaris is somewhat picky about how many heads and sectors per track you * have defined in mode pages 3 and 4. These values seem to cause Solaris * to get the capacity more or less right when you run the format tool. * They still have problems when dealing with devices larger than 1TB, * but there isn't anything we can do about that. * * For smaller LUN sizes, this ends up causing the number of cylinders to * work out to 0. Solaris actually recognizes that and comes up with its * own bogus geometry to fit the actual capacity of the drive. They really * should just give up on geometry and stick to the read capacity * information alone for modern disk drives. * * One thing worth mentioning about Solaris' mkfs command is that it * doesn't like sectors per track values larger than 256. 512 seems to * work okay for format, but causes problems when you try to make a * filesystem. * * Another caveat about these values: the product of these two values * really should be a power of 2. This is because of the simplistic * shift-based calculation that we have to use on the i386 platform to * calculate the number of cylinders here. (If you use a divide, you end * up calling __udivdi3(), which is a hardware FP call on the PC. On the * XScale, it is done in software, so you can do that from inside the * kernel.) * * So for the current values (256 S/T, 128 H), we get 32768, which works * very nicely for calculating cylinders. * * If you want to change these values so that their product is no longer a * power of 2, re-visit the calculation in ctl_init_page_index(). You may * need to make it a bit more complicated to get the number of cylinders * right. */ #define CTL_DEFAULT_SECTORS_PER_TRACK 256 #define CTL_DEFAULT_HEADS 128 #define CTL_DEFAULT_ROTATION_RATE 10000 struct ctl_page_index; typedef int ctl_modesen_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, int pc); typedef int ctl_modesel_handler(struct ctl_scsiio *ctsio, struct ctl_page_index *page_index, uint8_t *page_ptr); typedef enum { CTL_PAGE_FLAG_NONE = 0x00, CTL_PAGE_FLAG_DISK_ONLY = 0x01 } ctl_page_flags; struct ctl_page_index { uint8_t page_code; uint8_t subpage; uint16_t page_len; uint8_t *page_data; ctl_page_flags page_flags; ctl_modesen_handler *sense_handler; ctl_modesel_handler *select_handler; }; #define CTL_PAGE_CURRENT 0x00 #define CTL_PAGE_CHANGEABLE 0x01 #define CTL_PAGE_DEFAULT 0x02 #define CTL_PAGE_SAVED 0x03 +#define CTL_NUM_LBP_PARAMS 4 +#define CTL_NUM_LBP_THRESH 4 +#define CTL_LBP_EXPONENT 11 /* 2048 sectors */ +#define CTL_LBP_PERIOD 10 /* 10 seconds */ +#define CTL_LBP_UA_PERIOD 300 /* 5 minutes */ + +struct ctl_logical_block_provisioning_page { + struct scsi_logical_block_provisioning_page main; + struct scsi_logical_block_provisioning_page_descr descr[CTL_NUM_LBP_THRESH]; +}; + static const struct ctl_page_index page_index_template[] = { {SMS_RW_ERROR_RECOVERY_PAGE, 0, sizeof(struct scsi_da_rw_recovery_page), NULL, CTL_PAGE_FLAG_DISK_ONLY, NULL, NULL}, {SMS_FORMAT_DEVICE_PAGE, 0, sizeof(struct scsi_format_page), NULL, CTL_PAGE_FLAG_DISK_ONLY, NULL, NULL}, {SMS_RIGID_DISK_PAGE, 0, sizeof(struct scsi_rigid_disk_page), NULL, CTL_PAGE_FLAG_DISK_ONLY, NULL, NULL}, {SMS_CACHING_PAGE, 0, sizeof(struct scsi_caching_page), NULL, CTL_PAGE_FLAG_DISK_ONLY, NULL, ctl_caching_sp_handler}, {SMS_CONTROL_MODE_PAGE, 0, sizeof(struct scsi_control_page), NULL, CTL_PAGE_FLAG_NONE, NULL, ctl_control_page_handler}, {SMS_INFO_EXCEPTIONS_PAGE, 0, sizeof(struct scsi_info_exceptions_page), NULL, CTL_PAGE_FLAG_NONE, NULL, NULL}, {SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF, 0x02, - sizeof(struct scsi_logical_block_provisioning_page), NULL, + sizeof(struct ctl_logical_block_provisioning_page), NULL, CTL_PAGE_FLAG_DISK_ONLY, NULL, NULL}, {SMS_VENDOR_SPECIFIC_PAGE | SMPH_SPF, DBGCNF_SUBPAGE_CODE, sizeof(struct copan_debugconf_subpage), NULL, CTL_PAGE_FLAG_NONE, ctl_debugconf_sp_sense_handler, ctl_debugconf_sp_select_handler}, }; #define CTL_NUM_MODE_PAGES sizeof(page_index_template)/ \ sizeof(page_index_template[0]) struct ctl_mode_pages { struct scsi_da_rw_recovery_page rw_er_page[4]; struct scsi_format_page format_page[4]; struct scsi_rigid_disk_page rigid_disk_page[4]; struct scsi_caching_page caching_page[4]; struct scsi_control_page control_page[4]; struct scsi_info_exceptions_page ie_page[4]; - struct scsi_logical_block_provisioning_page lbp_page[4]; + struct ctl_logical_block_provisioning_page lbp_page[4]; struct copan_debugconf_subpage debugconf_subpage[4]; struct ctl_page_index index[CTL_NUM_MODE_PAGES]; }; static const struct ctl_page_index log_page_index_template[] = { {SLS_SUPPORTED_PAGES_PAGE, 0, 0, NULL, CTL_PAGE_FLAG_NONE, NULL, NULL}, {SLS_SUPPORTED_PAGES_PAGE, SLS_SUPPORTED_SUBPAGES_SUBPAGE, 0, NULL, CTL_PAGE_FLAG_NONE, NULL, NULL}, + {SLS_LOGICAL_BLOCK_PROVISIONING, 0, 0, NULL, + CTL_PAGE_FLAG_NONE, ctl_lbp_log_sense_handler, NULL}, }; #define CTL_NUM_LOG_PAGES sizeof(log_page_index_template)/ \ sizeof(log_page_index_template[0]) struct ctl_log_pages { uint8_t pages_page[CTL_NUM_LOG_PAGES]; uint8_t subpages_page[CTL_NUM_LOG_PAGES * 2]; + uint8_t lbp_page[12*CTL_NUM_LBP_PARAMS]; struct ctl_page_index index[CTL_NUM_LOG_PAGES]; }; struct ctl_lun_delay_info { ctl_delay_type datamove_type; uint32_t datamove_delay; ctl_delay_type done_type; uint32_t done_delay; }; typedef enum { CTL_ERR_INJ_NONE = 0x00, CTL_ERR_INJ_ABORTED = 0x01 } ctl_err_inject_flags; typedef enum { CTL_PR_FLAG_NONE = 0x00, CTL_PR_FLAG_REGISTERED = 0x01, CTL_PR_FLAG_ACTIVE_RES = 0x02 } ctl_per_res_flags; #define CTL_PR_ALL_REGISTRANTS 0xFFFFFFFF #define CTL_PR_NO_RESERVATION 0xFFFFFFF0 struct ctl_devid { int len; uint8_t data[]; }; /* * For report target port groups. */ #define NUM_TARGET_PORT_GROUPS 2 #define CTL_WRITE_BUFFER_SIZE 262144 struct tpc_list; struct ctl_lun { struct mtx lun_lock; struct ctl_id target; uint64_t lun; ctl_lun_flags flags; STAILQ_HEAD(,ctl_error_desc) error_list; uint64_t error_serial; struct ctl_softc *ctl_softc; struct ctl_be_lun *be_lun; struct ctl_backend_driver *backend; int io_count; struct ctl_lun_delay_info delay_info; int sync_interval; int sync_count; TAILQ_HEAD(ctl_ooaq, ctl_io_hdr) ooa_queue; TAILQ_HEAD(ctl_blockq,ctl_io_hdr) blocked_queue; STAILQ_ENTRY(ctl_lun) links; STAILQ_ENTRY(ctl_lun) run_links; #ifdef CTL_WITH_CA uint32_t have_ca[CTL_MAX_INITIATORS >> 5]; struct scsi_sense_data pending_sense[CTL_MAX_INITIATORS]; #endif ctl_ua_type pending_ua[CTL_MAX_INITIATORS]; + time_t lasttpt; struct ctl_mode_pages mode_pages; struct ctl_log_pages log_pages; struct ctl_lun_io_stats stats; uint32_t res_idx; unsigned int PRGeneration; uint64_t pr_keys[2*CTL_MAX_INITIATORS]; int pr_key_count; uint32_t pr_res_idx; uint8_t res_type; uint8_t *write_buffer; struct ctl_devid *lun_devid; TAILQ_HEAD(tpc_lists, tpc_list) tpc_lists; }; typedef enum { CTL_FLAG_REAL_SYNC = 0x02, CTL_FLAG_MASTER_SHELF = 0x04 } ctl_gen_flags; #define CTL_MAX_THREADS 16 struct ctl_thread { struct mtx_padalign queue_lock; struct ctl_softc *ctl_softc; struct thread *thread; STAILQ_HEAD(, ctl_io_hdr) incoming_queue; STAILQ_HEAD(, ctl_io_hdr) rtr_queue; STAILQ_HEAD(, ctl_io_hdr) done_queue; STAILQ_HEAD(, ctl_io_hdr) isc_queue; }; struct tpc_token; struct ctl_softc { struct mtx ctl_lock; struct cdev *dev; int open_count; struct ctl_id target; int num_disks; int num_luns; ctl_gen_flags flags; ctl_ha_mode ha_mode; int inquiry_pq_no_lun; struct sysctl_ctx_list sysctl_ctx; struct sysctl_oid *sysctl_tree; struct ctl_ioctl_info ioctl_info; struct ctl_io_pool *internal_pool; struct ctl_io_pool *emergency_pool; struct ctl_io_pool *othersc_pool; struct proc *ctl_proc; int targ_online; uint32_t ctl_lun_mask[(CTL_MAX_LUNS + 31) / 32]; struct ctl_lun *ctl_luns[CTL_MAX_LUNS]; uint32_t ctl_port_mask[(CTL_MAX_PORTS + 31) / 32]; STAILQ_HEAD(, ctl_lun) lun_list; STAILQ_HEAD(, ctl_be_lun) pending_lun_queue; uint32_t num_frontends; STAILQ_HEAD(, ctl_frontend) fe_list; uint32_t num_ports; STAILQ_HEAD(, ctl_port) port_list; struct ctl_port *ctl_ports[CTL_MAX_PORTS]; uint32_t num_backends; STAILQ_HEAD(, ctl_backend_driver) be_list; struct mtx pool_lock; uint32_t num_pools; uint32_t cur_pool_id; STAILQ_HEAD(, ctl_io_pool) io_pools; struct ctl_thread threads[CTL_MAX_THREADS]; TAILQ_HEAD(tpc_tokens, tpc_token) tpc_tokens; struct callout tpc_timeout; }; #ifdef _KERNEL extern const struct ctl_cmd_entry ctl_cmd_table[256]; uint32_t ctl_get_initindex(struct ctl_nexus *nexus); uint32_t ctl_get_resindex(struct ctl_nexus *nexus); uint32_t ctl_port_idx(int port_num); int ctl_pool_create(struct ctl_softc *ctl_softc, ctl_pool_type pool_type, uint32_t total_ctl_io, struct ctl_io_pool **npool); void ctl_pool_free(struct ctl_io_pool *pool); int ctl_scsi_release(struct ctl_scsiio *ctsio); int ctl_scsi_reserve(struct ctl_scsiio *ctsio); int ctl_start_stop(struct ctl_scsiio *ctsio); int ctl_sync_cache(struct ctl_scsiio *ctsio); int ctl_format(struct ctl_scsiio *ctsio); int ctl_read_buffer(struct ctl_scsiio *ctsio); int ctl_write_buffer(struct ctl_scsiio *ctsio); int ctl_write_same(struct ctl_scsiio *ctsio); int ctl_unmap(struct ctl_scsiio *ctsio); int ctl_mode_select(struct ctl_scsiio *ctsio); int ctl_mode_sense(struct ctl_scsiio *ctsio); int ctl_log_sense(struct ctl_scsiio *ctsio); int ctl_read_capacity(struct ctl_scsiio *ctsio); int ctl_read_capacity_16(struct ctl_scsiio *ctsio); int ctl_read_defect(struct ctl_scsiio *ctsio); int ctl_read_write(struct ctl_scsiio *ctsio); int ctl_cnw(struct ctl_scsiio *ctsio); int ctl_report_luns(struct ctl_scsiio *ctsio); int ctl_request_sense(struct ctl_scsiio *ctsio); int ctl_tur(struct ctl_scsiio *ctsio); int ctl_verify(struct ctl_scsiio *ctsio); int ctl_inquiry(struct ctl_scsiio *ctsio); int ctl_persistent_reserve_in(struct ctl_scsiio *ctsio); int ctl_persistent_reserve_out(struct ctl_scsiio *ctsio); int ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio); int ctl_report_supported_opcodes(struct ctl_scsiio *ctsio); int ctl_report_supported_tmf(struct ctl_scsiio *ctsio); int ctl_report_timestamp(struct ctl_scsiio *ctsio); int ctl_isc(struct ctl_scsiio *ctsio); void ctl_tpc_init(struct ctl_softc *softc); void ctl_tpc_shutdown(struct ctl_softc *softc); void ctl_tpc_lun_init(struct ctl_lun *lun); void ctl_tpc_lun_shutdown(struct ctl_lun *lun); int ctl_inquiry_evpd_tpc(struct ctl_scsiio *ctsio, int alloc_len); int ctl_receive_copy_status_lid1(struct ctl_scsiio *ctsio); int ctl_receive_copy_failure_details(struct ctl_scsiio *ctsio); int ctl_receive_copy_status_lid4(struct ctl_scsiio *ctsio); int ctl_receive_copy_operating_parameters(struct ctl_scsiio *ctsio); int ctl_extended_copy_lid1(struct ctl_scsiio *ctsio); int ctl_extended_copy_lid4(struct ctl_scsiio *ctsio); int ctl_copy_operation_abort(struct ctl_scsiio *ctsio); int ctl_populate_token(struct ctl_scsiio *ctsio); int ctl_write_using_token(struct ctl_scsiio *ctsio); int ctl_receive_rod_token_information(struct ctl_scsiio *ctsio); int ctl_report_all_rod_tokens(struct ctl_scsiio *ctsio); #endif /* _KERNEL */ #endif /* _CTL_PRIVATE_H_ */ /* * vim: ts=8 */ Index: head/sys/cam/scsi/scsi_all.h =================================================================== --- head/sys/cam/scsi/scsi_all.h (revision 274153) +++ head/sys/cam/scsi/scsi_all.h (revision 274154) @@ -1,3714 +1,3720 @@ /*- * Largely written by Julian Elischer (julian@tfs.com) * for TRW Financial Systems. * * TRW Financial Systems, in accordance with their agreement with Carnegie * Mellon University, makes this software available to CMU to distribute * or use in any manner that they see fit as long as this message is kept with * the software. For this reason TFS also grants any other persons or * organisations permission to use or modify this software. * * TFS supplies this software to be publicly redistributed * on the understanding that TFS is not responsible for the correct * functioning of this software in any circumstances. * * Ported to run under 386BSD by Julian Elischer (julian@tfs.com) Sept 1992 * * $FreeBSD$ */ /* * SCSI general interface description */ #ifndef _SCSI_SCSI_ALL_H #define _SCSI_SCSI_ALL_H 1 #include #include #ifdef _KERNEL /* * This is the number of seconds we wait for devices to settle after a SCSI * bus reset. */ extern int scsi_delay; #endif /* _KERNEL */ /* * SCSI command format */ /* * Define dome bits that are in ALL (or a lot of) scsi commands */ #define SCSI_CTL_LINK 0x01 #define SCSI_CTL_FLAG 0x02 #define SCSI_CTL_VENDOR 0xC0 #define SCSI_CMD_LUN 0xA0 /* these two should not be needed */ #define SCSI_CMD_LUN_SHIFT 5 /* LUN in the cmd is no longer SCSI */ #define SCSI_MAX_CDBLEN 16 /* * 16 byte commands are in the * SCSI-3 spec */ #if defined(CAM_MAX_CDBLEN) && (CAM_MAX_CDBLEN < SCSI_MAX_CDBLEN) #error "CAM_MAX_CDBLEN cannot be less than SCSI_MAX_CDBLEN" #endif /* 6byte CDBs special case 0 length to be 256 */ #define SCSI_CDB6_LEN(len) ((len) == 0 ? 256 : len) /* * This type defines actions to be taken when a particular sense code is * received. Right now, these flags are only defined to take up 16 bits, * but can be expanded in the future if necessary. */ typedef enum { SS_NOP = 0x000000, /* Do nothing */ SS_RETRY = 0x010000, /* Retry the command */ SS_FAIL = 0x020000, /* Bail out */ SS_START = 0x030000, /* Send a Start Unit command to the device, * then retry the original command. */ SS_TUR = 0x040000, /* Send a Test Unit Ready command to the * device, then retry the original command. */ SS_MASK = 0xff0000 } scsi_sense_action; typedef enum { SSQ_NONE = 0x0000, SSQ_DECREMENT_COUNT = 0x0100, /* Decrement the retry count */ SSQ_MANY = 0x0200, /* send lots of recovery commands */ SSQ_RANGE = 0x0400, /* * This table entry represents the * end of a range of ASCQs that * have identical error actions * and text. */ SSQ_PRINT_SENSE = 0x0800, SSQ_UA = 0x1000, /* Broadcast UA. */ SSQ_RESCAN = 0x2000, /* Rescan target for LUNs. */ SSQ_LOST = 0x4000, /* Destroy the LUNs. */ SSQ_MASK = 0xff00 } scsi_sense_action_qualifier; /* Mask for error status values */ #define SS_ERRMASK 0xff /* The default, retyable, error action */ #define SS_RDEF SS_RETRY|SSQ_DECREMENT_COUNT|SSQ_PRINT_SENSE|EIO /* The retyable, error action, with table specified error code */ #define SS_RET SS_RETRY|SSQ_DECREMENT_COUNT|SSQ_PRINT_SENSE /* Fatal error action, with table specified error code */ #define SS_FATAL SS_FAIL|SSQ_PRINT_SENSE struct scsi_generic { u_int8_t opcode; u_int8_t bytes[11]; }; struct scsi_request_sense { u_int8_t opcode; u_int8_t byte2; #define SRS_DESC 0x01 u_int8_t unused[2]; u_int8_t length; u_int8_t control; }; struct scsi_test_unit_ready { u_int8_t opcode; u_int8_t byte2; u_int8_t unused[3]; u_int8_t control; }; struct scsi_receive_diag { uint8_t opcode; uint8_t byte2; #define SRD_PCV 0x01 uint8_t page_code; uint8_t length[2]; uint8_t control; }; struct scsi_send_diag { uint8_t opcode; uint8_t byte2; #define SSD_UNITOFFL 0x01 #define SSD_DEVOFFL 0x02 #define SSD_SELFTEST 0x04 #define SSD_PF 0x10 #define SSD_SELF_TEST_CODE_MASK 0xE0 #define SSD_SELF_TEST_CODE_SHIFT 5 #define SSD_SELF_TEST_CODE_NONE 0x00 #define SSD_SELF_TEST_CODE_BG_SHORT 0x01 #define SSD_SELF_TEST_CODE_BG_EXTENDED 0x02 #define SSD_SELF_TEST_CODE_BG_ABORT 0x04 #define SSD_SELF_TEST_CODE_FG_SHORT 0x05 #define SSD_SELF_TEST_CODE_FG_EXTENDED 0x06 uint8_t reserved; uint8_t length[2]; uint8_t control; }; struct scsi_sense { u_int8_t opcode; u_int8_t byte2; u_int8_t unused[2]; u_int8_t length; u_int8_t control; }; struct scsi_inquiry { u_int8_t opcode; u_int8_t byte2; #define SI_EVPD 0x01 #define SI_CMDDT 0x02 u_int8_t page_code; u_int8_t length[2]; u_int8_t control; }; struct scsi_mode_sense_6 { u_int8_t opcode; u_int8_t byte2; #define SMS_DBD 0x08 u_int8_t page; #define SMS_PAGE_CODE 0x3F #define SMS_VENDOR_SPECIFIC_PAGE 0x00 #define SMS_DISCONNECT_RECONNECT_PAGE 0x02 #define SMS_FORMAT_DEVICE_PAGE 0x03 #define SMS_GEOMETRY_PAGE 0x04 #define SMS_CACHE_PAGE 0x08 #define SMS_PERIPHERAL_DEVICE_PAGE 0x09 #define SMS_CONTROL_MODE_PAGE 0x0A #define SMS_PROTO_SPECIFIC_PAGE 0x19 #define SMS_INFO_EXCEPTIONS_PAGE 0x1C #define SMS_ALL_PAGES_PAGE 0x3F #define SMS_PAGE_CTRL_MASK 0xC0 #define SMS_PAGE_CTRL_CURRENT 0x00 #define SMS_PAGE_CTRL_CHANGEABLE 0x40 #define SMS_PAGE_CTRL_DEFAULT 0x80 #define SMS_PAGE_CTRL_SAVED 0xC0 u_int8_t subpage; #define SMS_SUBPAGE_PAGE_0 0x00 #define SMS_SUBPAGE_ALL 0xff u_int8_t length; u_int8_t control; }; struct scsi_mode_sense_10 { u_int8_t opcode; u_int8_t byte2; /* same bits as small version */ #define SMS10_LLBAA 0x10 u_int8_t page; /* same bits as small version */ u_int8_t subpage; u_int8_t unused[3]; u_int8_t length[2]; u_int8_t control; }; struct scsi_mode_select_6 { u_int8_t opcode; u_int8_t byte2; #define SMS_SP 0x01 #define SMS_PF 0x10 u_int8_t unused[2]; u_int8_t length; u_int8_t control; }; struct scsi_mode_select_10 { u_int8_t opcode; u_int8_t byte2; /* same bits as small version */ u_int8_t unused[5]; u_int8_t length[2]; u_int8_t control; }; /* * When sending a mode select to a tape drive, the medium type must be 0. */ struct scsi_mode_hdr_6 { u_int8_t datalen; u_int8_t medium_type; u_int8_t dev_specific; u_int8_t block_descr_len; }; struct scsi_mode_hdr_10 { u_int8_t datalen[2]; u_int8_t medium_type; u_int8_t dev_specific; u_int8_t reserved[2]; u_int8_t block_descr_len[2]; }; struct scsi_mode_block_descr { u_int8_t density_code; u_int8_t num_blocks[3]; u_int8_t reserved; u_int8_t block_len[3]; }; struct scsi_per_res_in { u_int8_t opcode; u_int8_t action; #define SPRI_RK 0x00 #define SPRI_RR 0x01 #define SPRI_RC 0x02 #define SPRI_RS 0x03 u_int8_t reserved[5]; u_int8_t length[2]; #define SPRI_MAX_LEN 0xffff u_int8_t control; }; struct scsi_per_res_in_header { u_int8_t generation[4]; u_int8_t length[4]; }; struct scsi_per_res_key { u_int8_t key[8]; }; struct scsi_per_res_in_keys { struct scsi_per_res_in_header header; struct scsi_per_res_key keys[0]; }; struct scsi_per_res_cap { uint8_t length[2]; uint8_t flags1; #define SPRI_RLR_C 0x80 #define SPRI_CRH 0x10 #define SPRI_SIP_C 0x08 #define SPRI_ATP_C 0x04 #define SPRI_PTPL_C 0x01 uint8_t flags2; #define SPRI_TMV 0x80 #define SPRI_ALLOW_CMD_MASK 0x70 #define SPRI_ALLOW_CMD_SHIFT 4 #define SPRI_ALLOW_NA 0x00 #define SPRI_ALLOW_1 0x10 #define SPRI_ALLOW_2 0x20 #define SPRI_ALLOW_3 0x30 #define SPRI_ALLOW_4 0x40 #define SPRI_ALLOW_5 0x50 #define SPRI_PTPL_A 0x01 uint8_t type_mask[2]; #define SPRI_TM_WR_EX_AR 0x8000 #define SPRI_TM_EX_AC_RO 0x4000 #define SPRI_TM_WR_EX_RO 0x2000 #define SPRI_TM_EX_AC 0x0800 #define SPRI_TM_WR_EX 0x0200 #define SPRI_TM_EX_AC_AR 0x0001 uint8_t reserved[2]; }; struct scsi_per_res_in_rsrv_data { uint8_t reservation[8]; uint8_t scope_addr[4]; uint8_t reserved; uint8_t scopetype; #define SPRT_WE 0x01 #define SPRT_EA 0x03 #define SPRT_WERO 0x05 #define SPRT_EARO 0x06 #define SPRT_WEAR 0x07 #define SPRT_EAAR 0x08 uint8_t extent_length[2]; }; struct scsi_per_res_in_rsrv { struct scsi_per_res_in_header header; struct scsi_per_res_in_rsrv_data data; }; struct scsi_per_res_in_full_desc { struct scsi_per_res_key res_key; uint8_t reserved1[4]; uint8_t flags; #define SPRI_FULL_ALL_TG_PT 0x02 #define SPRI_FULL_R_HOLDER 0x01 uint8_t scopetype; uint8_t reserved2[4]; uint8_t rel_trgt_port_id[2]; uint8_t additional_length[4]; uint8_t transport_id[]; }; struct scsi_per_res_in_full { struct scsi_per_res_in_header header; struct scsi_per_res_in_full_desc desc[]; }; struct scsi_per_res_out { u_int8_t opcode; u_int8_t action; #define SPRO_REGISTER 0x00 #define SPRO_RESERVE 0x01 #define SPRO_RELEASE 0x02 #define SPRO_CLEAR 0x03 #define SPRO_PREEMPT 0x04 #define SPRO_PRE_ABO 0x05 #define SPRO_REG_IGNO 0x06 #define SPRO_REG_MOVE 0x07 #define SPRO_REPL_LOST_RES 0x08 #define SPRO_ACTION_MASK 0x1f u_int8_t scope_type; #define SPR_SCOPE_MASK 0xf0 #define SPR_SCOPE_SHIFT 4 #define SPR_LU_SCOPE 0x00 #define SPR_EXTENT_SCOPE 0x10 #define SPR_ELEMENT_SCOPE 0x20 #define SPR_TYPE_MASK 0x0f #define SPR_TYPE_RD_SHARED 0x00 #define SPR_TYPE_WR_EX 0x01 #define SPR_TYPE_RD_EX 0x02 #define SPR_TYPE_EX_AC 0x03 #define SPR_TYPE_SHARED 0x04 #define SPR_TYPE_WR_EX_RO 0x05 #define SPR_TYPE_EX_AC_RO 0x06 #define SPR_TYPE_WR_EX_AR 0x07 #define SPR_TYPE_EX_AC_AR 0x08 u_int8_t reserved[2]; u_int8_t length[4]; u_int8_t control; }; struct scsi_per_res_out_parms { struct scsi_per_res_key res_key; u_int8_t serv_act_res_key[8]; u_int8_t scope_spec_address[4]; u_int8_t flags; #define SPR_SPEC_I_PT 0x08 #define SPR_ALL_TG_PT 0x04 #define SPR_APTPL 0x01 u_int8_t reserved1; u_int8_t extent_length[2]; u_int8_t transport_id_list[]; }; struct scsi_per_res_out_trans_ids { u_int8_t additional_length[4]; u_int8_t transport_ids[]; }; /* * Used with REGISTER AND MOVE serivce action of the PERSISTENT RESERVE OUT * command. */ struct scsi_per_res_reg_move { struct scsi_per_res_key res_key; u_int8_t serv_act_res_key[8]; u_int8_t reserved; u_int8_t flags; #define SPR_REG_MOVE_UNREG 0x02 #define SPR_REG_MOVE_APTPL 0x01 u_int8_t rel_trgt_port_id[2]; u_int8_t transport_id_length[4]; u_int8_t transport_id[]; }; struct scsi_transportid_header { uint8_t format_protocol; #define SCSI_TRN_FORMAT_MASK 0xc0 #define SCSI_TRN_FORMAT_SHIFT 6 #define SCSI_TRN_PROTO_MASK 0x0f }; struct scsi_transportid_fcp { uint8_t format_protocol; #define SCSI_TRN_FCP_FORMAT_DEFAULT 0x00 uint8_t reserved1[7]; uint8_t n_port_name[8]; uint8_t reserved2[8]; }; struct scsi_transportid_spi { uint8_t format_protocol; #define SCSI_TRN_SPI_FORMAT_DEFAULT 0x00 uint8_t reserved1; uint8_t scsi_addr[2]; uint8_t obsolete[2]; uint8_t rel_trgt_port_id[2]; uint8_t reserved2[16]; }; struct scsi_transportid_1394 { uint8_t format_protocol; #define SCSI_TRN_1394_FORMAT_DEFAULT 0x00 uint8_t reserved1[7]; uint8_t eui64[8]; uint8_t reserved2[8]; }; struct scsi_transportid_rdma { uint8_t format_protocol; #define SCSI_TRN_RDMA_FORMAT_DEFAULT 0x00 uint8_t reserved[7]; #define SCSI_TRN_RDMA_PORT_LEN 16 uint8_t initiator_port_id[SCSI_TRN_RDMA_PORT_LEN]; }; struct scsi_transportid_iscsi_device { uint8_t format_protocol; #define SCSI_TRN_ISCSI_FORMAT_DEVICE 0x00 uint8_t reserved; uint8_t additional_length[2]; uint8_t iscsi_name[]; }; struct scsi_transportid_iscsi_port { uint8_t format_protocol; #define SCSI_TRN_ISCSI_FORMAT_PORT 0x40 uint8_t reserved; uint8_t additional_length[2]; uint8_t iscsi_name[]; /* * Followed by a separator and iSCSI initiator session ID */ }; struct scsi_transportid_sas { uint8_t format_protocol; #define SCSI_TRN_SAS_FORMAT_DEFAULT 0x00 uint8_t reserved1[3]; uint8_t sas_address[8]; uint8_t reserved2[12]; }; struct scsi_sop_routing_id_norm { uint8_t bus; uint8_t devfunc; #define SCSI_TRN_SOP_BUS_MAX 0xff #define SCSI_TRN_SOP_DEV_MAX 0x1f #define SCSI_TRN_SOP_DEV_MASK 0xf8 #define SCSI_TRN_SOP_DEV_SHIFT 3 #define SCSI_TRN_SOP_FUNC_NORM_MASK 0x07 #define SCSI_TRN_SOP_FUNC_NORM_MAX 0x07 }; struct scsi_sop_routing_id_alt { uint8_t bus; uint8_t function; #define SCSI_TRN_SOP_FUNC_ALT_MAX 0xff }; struct scsi_transportid_sop { uint8_t format_protocol; #define SCSI_TRN_SOP_FORMAT_DEFAULT 0x00 uint8_t reserved1; uint8_t routing_id[2]; uint8_t reserved2[20]; }; struct scsi_log_sense { u_int8_t opcode; u_int8_t byte2; #define SLS_SP 0x01 #define SLS_PPC 0x02 u_int8_t page; #define SLS_PAGE_CODE 0x3F #define SLS_SUPPORTED_PAGES_PAGE 0x00 #define SLS_OVERRUN_PAGE 0x01 #define SLS_ERROR_WRITE_PAGE 0x02 #define SLS_ERROR_READ_PAGE 0x03 #define SLS_ERROR_READREVERSE_PAGE 0x04 #define SLS_ERROR_VERIFY_PAGE 0x05 #define SLS_ERROR_NONMEDIUM_PAGE 0x06 #define SLS_ERROR_LASTN_PAGE 0x07 +#define SLS_LOGICAL_BLOCK_PROVISIONING 0x0c #define SLS_SELF_TEST_PAGE 0x10 #define SLS_IE_PAGE 0x2f #define SLS_PAGE_CTRL_MASK 0xC0 #define SLS_PAGE_CTRL_THRESHOLD 0x00 #define SLS_PAGE_CTRL_CUMULATIVE 0x40 #define SLS_PAGE_CTRL_THRESH_DEFAULT 0x80 #define SLS_PAGE_CTRL_CUMUL_DEFAULT 0xC0 u_int8_t subpage; #define SLS_SUPPORTED_SUBPAGES_SUBPAGE 0xff u_int8_t reserved; u_int8_t paramptr[2]; u_int8_t length[2]; u_int8_t control; }; struct scsi_log_select { u_int8_t opcode; u_int8_t byte2; /* SLS_SP 0x01 */ #define SLS_PCR 0x02 u_int8_t page; /* SLS_PAGE_CTRL_MASK 0xC0 */ /* SLS_PAGE_CTRL_THRESHOLD 0x00 */ /* SLS_PAGE_CTRL_CUMULATIVE 0x40 */ /* SLS_PAGE_CTRL_THRESH_DEFAULT 0x80 */ /* SLS_PAGE_CTRL_CUMUL_DEFAULT 0xC0 */ u_int8_t reserved[4]; u_int8_t length[2]; u_int8_t control; }; struct scsi_log_header { u_int8_t page; #define SL_PAGE_CODE 0x3F #define SL_SPF 0x40 #define SL_DS 0x80 u_int8_t subpage; u_int8_t datalen[2]; }; struct scsi_log_param_header { u_int8_t param_code[2]; u_int8_t param_control; #define SLP_LP 0x01 #define SLP_LBIN 0x02 #define SLP_TMC_MASK 0x0C #define SLP_TMC_ALWAYS 0x00 #define SLP_TMC_EQUAL 0x04 #define SLP_TMC_NOTEQUAL 0x08 #define SLP_TMC_GREATER 0x0C #define SLP_ETC 0x10 #define SLP_TSD 0x20 #define SLP_DS 0x40 #define SLP_DU 0x80 u_int8_t param_len; }; struct scsi_control_page { u_int8_t page_code; u_int8_t page_length; u_int8_t rlec; #define SCP_RLEC 0x01 /*Report Log Exception Cond*/ #define SCP_GLTSD 0x02 /*Global Logging target save disable */ #define SCP_DSENSE 0x04 /*Descriptor Sense */ #define SCP_DPICZ 0x08 /*Disable Prot. Info Check if Prot. Field is Zero */ #define SCP_TMF_ONLY 0x10 /*TM Functions Only*/ #define SCP_TST_MASK 0xE0 /*Task Set Type Mask*/ #define SCP_TST_ONE 0x00 /*One Task Set*/ #define SCP_TST_SEPARATE 0x20 /*Separate Task Sets*/ u_int8_t queue_flags; #define SCP_QUEUE_ALG_MASK 0xF0 #define SCP_QUEUE_ALG_RESTRICTED 0x00 #define SCP_QUEUE_ALG_UNRESTRICTED 0x10 #define SCP_NUAR 0x08 /*No UA on release*/ #define SCP_QUEUE_ERR 0x02 /*Queued I/O aborted for CACs*/ #define SCP_QUEUE_DQUE 0x01 /*Queued I/O disabled*/ u_int8_t eca_and_aen; #define SCP_EECA 0x80 /*Enable Extended CA*/ #define SCP_RAC 0x40 /*Report a check*/ #define SCP_SWP 0x08 /*Software Write Protect*/ #define SCP_RAENP 0x04 /*Ready AEN Permission*/ #define SCP_UAAENP 0x02 /*UA AEN Permission*/ #define SCP_EAENP 0x01 /*Error AEN Permission*/ u_int8_t flags4; #define SCP_ATO 0x80 /*Application tag owner*/ #define SCP_TAS 0x40 /*Task aborted status*/ #define SCP_ATMPE 0x20 /*Application tag mode page*/ #define SCP_RWWP 0x10 /*Reject write without prot*/ u_int8_t aen_holdoff_period[2]; u_int8_t busy_timeout_period[2]; u_int8_t extended_selftest_completion_time[2]; }; struct scsi_cache_page { u_int8_t page_code; #define SCHP_PAGE_SAVABLE 0x80 /* Page is savable */ u_int8_t page_length; u_int8_t cache_flags; #define SCHP_FLAGS_WCE 0x04 /* Write Cache Enable */ #define SCHP_FLAGS_MF 0x02 /* Multiplication factor */ #define SCHP_FLAGS_RCD 0x01 /* Read Cache Disable */ u_int8_t rw_cache_policy; u_int8_t dis_prefetch[2]; u_int8_t min_prefetch[2]; u_int8_t max_prefetch[2]; u_int8_t max_prefetch_ceil[2]; }; /* * XXX KDM * Updated version of the cache page, as of SBC. Update this to SBC-3 and * rationalize the two. */ struct scsi_caching_page { uint8_t page_code; #define SMS_CACHING_PAGE 0x08 uint8_t page_length; uint8_t flags1; #define SCP_IC 0x80 #define SCP_ABPF 0x40 #define SCP_CAP 0x20 #define SCP_DISC 0x10 #define SCP_SIZE 0x08 #define SCP_WCE 0x04 #define SCP_MF 0x02 #define SCP_RCD 0x01 uint8_t ret_priority; uint8_t disable_pf_transfer_len[2]; uint8_t min_prefetch[2]; uint8_t max_prefetch[2]; uint8_t max_pf_ceiling[2]; uint8_t flags2; #define SCP_FSW 0x80 #define SCP_LBCSS 0x40 #define SCP_DRA 0x20 #define SCP_VS1 0x10 #define SCP_VS2 0x08 uint8_t cache_segments; uint8_t cache_seg_size[2]; uint8_t reserved; uint8_t non_cache_seg_size[3]; }; /* * XXX KDM move this off to a vendor shim. */ struct copan_debugconf_subpage { uint8_t page_code; #define DBGCNF_PAGE_CODE 0x00 uint8_t subpage; #define DBGCNF_SUBPAGE_CODE 0xF0 uint8_t page_length[2]; uint8_t page_version; #define DBGCNF_VERSION 0x00 uint8_t ctl_time_io_secs[2]; }; struct scsi_info_exceptions_page { u_int8_t page_code; #define SIEP_PAGE_SAVABLE 0x80 /* Page is savable */ u_int8_t page_length; u_int8_t info_flags; #define SIEP_FLAGS_PERF 0x80 #define SIEP_FLAGS_EBF 0x20 #define SIEP_FLAGS_EWASC 0x10 #define SIEP_FLAGS_DEXCPT 0x08 #define SIEP_FLAGS_TEST 0x04 #define SIEP_FLAGS_EBACKERR 0x02 #define SIEP_FLAGS_LOGERR 0x01 u_int8_t mrie; u_int8_t interval_timer[4]; u_int8_t report_count[4]; }; struct scsi_logical_block_provisioning_page_descr { uint8_t flags; +#define SLBPPD_ENABLED 0x80 +#define SLBPPD_TYPE_MASK 0x38 +#define SLBPPD_ARMING_MASK 0x07 +#define SLBPPD_ARMING_DEC 0x02 +#define SLBPPD_ARMING_INC 0x01 uint8_t resource; uint8_t reserved[2]; uint8_t count[4]; }; struct scsi_logical_block_provisioning_page { uint8_t page_code; uint8_t subpage_code; uint8_t page_length[2]; uint8_t flags; #define SLBPP_SITUA 0x01 uint8_t reserved[11]; struct scsi_logical_block_provisioning_page_descr descr[0]; }; /* * SCSI protocol identifier values, current as of SPC4r36l. */ #define SCSI_PROTO_FC 0x00 /* Fibre Channel */ #define SCSI_PROTO_SPI 0x01 /* Parallel SCSI */ #define SCSI_PROTO_SSA 0x02 /* Serial Storage Arch. */ #define SCSI_PROTO_1394 0x03 /* IEEE 1394 (Firewire) */ #define SCSI_PROTO_RDMA 0x04 /* SCSI RDMA Protocol */ #define SCSI_PROTO_ISCSI 0x05 /* Internet SCSI */ #define SCSI_PROTO_iSCSI 0x05 /* Internet SCSI */ #define SCSI_PROTO_SAS 0x06 /* SAS Serial SCSI Protocol */ #define SCSI_PROTO_ADT 0x07 /* Automation/Drive Int. Trans. Prot.*/ #define SCSI_PROTO_ADITP 0x07 /* Automation/Drive Int. Trans. Prot.*/ #define SCSI_PROTO_ATA 0x08 /* AT Attachment Interface */ #define SCSI_PROTO_UAS 0x09 /* USB Atached SCSI */ #define SCSI_PROTO_SOP 0x0a /* SCSI over PCI Express */ #define SCSI_PROTO_NONE 0x0f /* No specific protocol */ struct scsi_proto_specific_page { u_int8_t page_code; #define SPSP_PAGE_SAVABLE 0x80 /* Page is savable */ u_int8_t page_length; u_int8_t protocol; #define SPSP_PROTO_FC SCSI_PROTO_FC #define SPSP_PROTO_SPI SCSI_PROTO_SPI #define SPSP_PROTO_SSA SCSI_PROTO_SSA #define SPSP_PROTO_1394 SCSI_PROTO_1394 #define SPSP_PROTO_RDMA SCSI_PROTO_RDMA #define SPSP_PROTO_ISCSI SCSI_PROTO_ISCSI #define SPSP_PROTO_SAS SCSI_PROTO_SAS #define SPSP_PROTO_ADT SCSI_PROTO_ADITP #define SPSP_PROTO_ATA SCSI_PROTO_ATA #define SPSP_PROTO_UAS SCSI_PROTO_UAS #define SPSP_PROTO_SOP SCSI_PROTO_SOP #define SPSP_PROTO_NONE SCSI_PROTO_NONE }; struct scsi_reserve { u_int8_t opcode; u_int8_t byte2; #define SR_EXTENT 0x01 #define SR_ID_MASK 0x0e #define SR_3RDPTY 0x10 #define SR_LUN_MASK 0xe0 u_int8_t resv_id; u_int8_t length[2]; u_int8_t control; }; struct scsi_reserve_10 { uint8_t opcode; uint8_t byte2; #define SR10_3RDPTY 0x10 #define SR10_LONGID 0x02 #define SR10_EXTENT 0x01 uint8_t resv_id; uint8_t thirdparty_id; uint8_t reserved[3]; uint8_t length[2]; uint8_t control; }; struct scsi_release { u_int8_t opcode; u_int8_t byte2; u_int8_t resv_id; u_int8_t unused[1]; u_int8_t length; u_int8_t control; }; struct scsi_release_10 { uint8_t opcode; uint8_t byte2; uint8_t resv_id; uint8_t thirdparty_id; uint8_t reserved[3]; uint8_t length[2]; uint8_t control; }; struct scsi_prevent { u_int8_t opcode; u_int8_t byte2; u_int8_t unused[2]; u_int8_t how; u_int8_t control; }; #define PR_PREVENT 0x01 #define PR_ALLOW 0x00 struct scsi_sync_cache { u_int8_t opcode; u_int8_t byte2; #define SSC_IMMED 0x02 #define SSC_RELADR 0x01 u_int8_t begin_lba[4]; u_int8_t reserved; u_int8_t lb_count[2]; u_int8_t control; }; struct scsi_sync_cache_16 { uint8_t opcode; uint8_t byte2; uint8_t begin_lba[8]; uint8_t lb_count[4]; uint8_t reserved; uint8_t control; }; struct scsi_format { uint8_t opcode; uint8_t byte2; #define SF_LONGLIST 0x20 #define SF_FMTDATA 0x10 #define SF_CMPLIST 0x08 #define SF_FORMAT_MASK 0x07 #define SF_FORMAT_BLOCK 0x00 #define SF_FORMAT_LONG_BLOCK 0x03 #define SF_FORMAT_BFI 0x04 #define SF_FORMAT_PHYS 0x05 uint8_t vendor; uint8_t interleave[2]; uint8_t control; }; struct scsi_format_header_short { uint8_t reserved; #define SF_DATA_FOV 0x80 #define SF_DATA_DPRY 0x40 #define SF_DATA_DCRT 0x20 #define SF_DATA_STPF 0x10 #define SF_DATA_IP 0x08 #define SF_DATA_DSP 0x04 #define SF_DATA_IMMED 0x02 #define SF_DATA_VS 0x01 uint8_t byte2; uint8_t defect_list_len[2]; }; struct scsi_format_header_long { uint8_t reserved; uint8_t byte2; uint8_t reserved2[2]; uint8_t defect_list_len[4]; }; struct scsi_changedef { u_int8_t opcode; u_int8_t byte2; u_int8_t unused1; u_int8_t how; u_int8_t unused[4]; u_int8_t datalen; u_int8_t control; }; struct scsi_read_buffer { u_int8_t opcode; u_int8_t byte2; #define RWB_MODE 0x1F #define RWB_MODE_HDR_DATA 0x00 #define RWB_MODE_VENDOR 0x01 #define RWB_MODE_DATA 0x02 #define RWB_MODE_DESCR 0x03 #define RWB_MODE_DOWNLOAD 0x04 #define RWB_MODE_DOWNLOAD_SAVE 0x05 #define RWB_MODE_ECHO 0x0A #define RWB_MODE_ECHO_DESCR 0x0B #define RWB_MODE_ERROR_HISTORY 0x1C u_int8_t buffer_id; u_int8_t offset[3]; u_int8_t length[3]; u_int8_t control; }; struct scsi_write_buffer { u_int8_t opcode; u_int8_t byte2; u_int8_t buffer_id; u_int8_t offset[3]; u_int8_t length[3]; u_int8_t control; }; struct scsi_rw_6 { u_int8_t opcode; u_int8_t addr[3]; /* only 5 bits are valid in the MSB address byte */ #define SRW_TOPADDR 0x1F u_int8_t length; u_int8_t control; }; struct scsi_rw_10 { u_int8_t opcode; #define SRW10_RELADDR 0x01 /* EBP defined for WRITE(10) only */ #define SRW10_EBP 0x04 #define SRW10_FUA 0x08 #define SRW10_DPO 0x10 u_int8_t byte2; u_int8_t addr[4]; u_int8_t reserved; u_int8_t length[2]; u_int8_t control; }; struct scsi_rw_12 { u_int8_t opcode; #define SRW12_RELADDR 0x01 #define SRW12_FUA 0x08 #define SRW12_DPO 0x10 u_int8_t byte2; u_int8_t addr[4]; u_int8_t length[4]; u_int8_t reserved; u_int8_t control; }; struct scsi_rw_16 { u_int8_t opcode; #define SRW16_RELADDR 0x01 #define SRW16_FUA 0x08 #define SRW16_DPO 0x10 u_int8_t byte2; u_int8_t addr[8]; u_int8_t length[4]; u_int8_t reserved; u_int8_t control; }; struct scsi_write_same_10 { uint8_t opcode; uint8_t byte2; #define SWS_LBDATA 0x02 #define SWS_PBDATA 0x04 #define SWS_UNMAP 0x08 #define SWS_ANCHOR 0x10 uint8_t addr[4]; uint8_t group; uint8_t length[2]; uint8_t control; }; struct scsi_write_same_16 { uint8_t opcode; uint8_t byte2; #define SWS_NDOB 0x01 uint8_t addr[8]; uint8_t length[4]; uint8_t group; uint8_t control; }; struct scsi_unmap { uint8_t opcode; uint8_t byte2; #define SU_ANCHOR 0x01 uint8_t reserved[4]; uint8_t group; uint8_t length[2]; uint8_t control; }; struct scsi_unmap_header { uint8_t length[2]; uint8_t desc_length[2]; uint8_t reserved[4]; }; struct scsi_unmap_desc { uint8_t lba[8]; uint8_t length[4]; uint8_t reserved[4]; }; struct scsi_write_verify_10 { uint8_t opcode; uint8_t byte2; #define SWV_BYTCHK 0x02 #define SWV_DPO 0x10 #define SWV_WRPROECT_MASK 0xe0 uint8_t addr[4]; uint8_t group; uint8_t length[2]; uint8_t control; }; struct scsi_write_verify_12 { uint8_t opcode; uint8_t byte2; uint8_t addr[4]; uint8_t length[4]; uint8_t group; uint8_t control; }; struct scsi_write_verify_16 { uint8_t opcode; uint8_t byte2; uint8_t addr[8]; uint8_t length[4]; uint8_t group; uint8_t control; }; struct scsi_start_stop_unit { u_int8_t opcode; u_int8_t byte2; #define SSS_IMMED 0x01 u_int8_t reserved[2]; u_int8_t how; #define SSS_START 0x01 #define SSS_LOEJ 0x02 #define SSS_PC_MASK 0xf0 #define SSS_PC_START_VALID 0x00 #define SSS_PC_ACTIVE 0x10 #define SSS_PC_IDLE 0x20 #define SSS_PC_STANDBY 0x30 #define SSS_PC_LU_CONTROL 0x70 #define SSS_PC_FORCE_IDLE_0 0xa0 #define SSS_PC_FORCE_STANDBY_0 0xb0 u_int8_t control; }; struct ata_pass_12 { u_int8_t opcode; u_int8_t protocol; #define AP_PROTO_HARD_RESET (0x00 << 1) #define AP_PROTO_SRST (0x01 << 1) #define AP_PROTO_NON_DATA (0x03 << 1) #define AP_PROTO_PIO_IN (0x04 << 1) #define AP_PROTO_PIO_OUT (0x05 << 1) #define AP_PROTO_DMA (0x06 << 1) #define AP_PROTO_DMA_QUEUED (0x07 << 1) #define AP_PROTO_DEVICE_DIAG (0x08 << 1) #define AP_PROTO_DEVICE_RESET (0x09 << 1) #define AP_PROTO_UDMA_IN (0x0a << 1) #define AP_PROTO_UDMA_OUT (0x0b << 1) #define AP_PROTO_FPDMA (0x0c << 1) #define AP_PROTO_RESP_INFO (0x0f << 1) #define AP_MULTI 0xe0 u_int8_t flags; #define AP_T_LEN 0x03 #define AP_BB 0x04 #define AP_T_DIR 0x08 #define AP_CK_COND 0x20 #define AP_OFFLINE 0x60 u_int8_t features; u_int8_t sector_count; u_int8_t lba_low; u_int8_t lba_mid; u_int8_t lba_high; u_int8_t device; u_int8_t command; u_int8_t reserved; u_int8_t control; }; struct scsi_maintenance_in { uint8_t opcode; uint8_t byte2; #define SERVICE_ACTION_MASK 0x1f #define SA_RPRT_TRGT_GRP 0x0a uint8_t reserved[4]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_report_supported_opcodes { uint8_t opcode; uint8_t service_action; uint8_t options; #define RSO_RCTD 0x80 #define RSO_OPTIONS_MASK 0x07 #define RSO_OPTIONS_ALL 0x00 #define RSO_OPTIONS_OC 0x01 #define RSO_OPTIONS_OC_SA 0x02 uint8_t requested_opcode; uint8_t requested_service_action[2]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_report_supported_opcodes_timeout { uint8_t length[2]; uint8_t reserved; uint8_t cmd_specific; uint8_t nominal_time[4]; uint8_t recommended_time[4]; }; struct scsi_report_supported_opcodes_descr { uint8_t opcode; uint8_t reserved; uint8_t service_action[2]; uint8_t reserved2; uint8_t flags; #define RSO_SERVACTV 0x01 #define RSO_CTDP 0x02 uint8_t cdb_length[2]; struct scsi_report_supported_opcodes_timeout timeout[0]; }; struct scsi_report_supported_opcodes_all { uint8_t length[4]; struct scsi_report_supported_opcodes_descr descr[0]; }; struct scsi_report_supported_opcodes_one { uint8_t reserved; uint8_t support; #define RSO_ONE_CTDP 0x80 uint8_t cdb_length[2]; uint8_t cdb_usage[]; }; struct scsi_report_supported_tmf { uint8_t opcode; uint8_t service_action; uint8_t reserved[4]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_report_supported_tmf_data { uint8_t byte1; #define RST_WAKES 0x01 #define RST_TRS 0x02 #define RST_QTS 0x04 #define RST_LURS 0x08 #define RST_CTSS 0x10 #define RST_CACAS 0x20 #define RST_ATSS 0x40 #define RST_ATS 0x80 uint8_t byte2; #define RST_ITNRS 0x01 #define RST_QTSS 0x02 #define RST_QAES 0x04 uint8_t reserved[2]; }; struct scsi_report_timestamp { uint8_t opcode; uint8_t service_action; uint8_t reserved[4]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_report_timestamp_data { uint8_t length[2]; uint8_t origin; #define RTS_ORIG_MASK 0x00 #define RTS_ORIG_ZERO 0x00 #define RTS_ORIG_SET 0x02 #define RTS_ORIG_OUTSIDE 0x03 uint8_t reserved; uint8_t timestamp[6]; uint8_t reserve2[2]; }; struct scsi_receive_copy_status_lid1 { uint8_t opcode; uint8_t service_action; #define RCS_RCS_LID1 0x00 uint8_t list_identifier; uint8_t reserved[7]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_receive_copy_status_lid1_data { uint8_t available_data[4]; uint8_t copy_command_status; #define RCS_CCS_INPROG 0x00 #define RCS_CCS_COMPLETED 0x01 #define RCS_CCS_ERROR 0x02 uint8_t segments_processed[2]; uint8_t transfer_count_units; #define RCS_TC_BYTES 0x00 #define RCS_TC_KBYTES 0x01 #define RCS_TC_MBYTES 0x02 #define RCS_TC_GBYTES 0x03 #define RCS_TC_TBYTES 0x04 #define RCS_TC_PBYTES 0x05 #define RCS_TC_EBYTES 0x06 #define RCS_TC_LBAS 0xf1 uint8_t transfer_count[4]; }; struct scsi_receive_copy_failure_details { uint8_t opcode; uint8_t service_action; #define RCS_RCFD 0x04 uint8_t list_identifier; uint8_t reserved[7]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_receive_copy_failure_details_data { uint8_t available_data[4]; uint8_t reserved[52]; uint8_t copy_command_status; uint8_t reserved2; uint8_t sense_data_length[2]; uint8_t sense_data[]; }; struct scsi_receive_copy_status_lid4 { uint8_t opcode; uint8_t service_action; #define RCS_RCS_LID4 0x05 uint8_t list_identifier[4]; uint8_t reserved[4]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_receive_copy_status_lid4_data { uint8_t available_data[4]; uint8_t response_to_service_action; uint8_t copy_command_status; #define RCS_CCS_COMPLETED_PROD 0x03 #define RCS_CCS_COMPLETED_RESID 0x04 #define RCS_CCS_INPROG_FGBG 0x10 #define RCS_CCS_INPROG_FG 0x11 #define RCS_CCS_INPROG_BG 0x12 #define RCS_CCS_ABORTED 0x60 uint8_t operation_counter[2]; uint8_t estimated_status_update_delay[4]; uint8_t extended_copy_completion_status; uint8_t length_of_the_sense_data_field; uint8_t sense_data_length; uint8_t transfer_count_units; uint8_t transfer_count[8]; uint8_t segments_processed[2]; uint8_t reserved[6]; uint8_t sense_data[]; }; struct scsi_receive_copy_operating_parameters { uint8_t opcode; uint8_t service_action; #define RCS_RCOP 0x03 uint8_t reserved[8]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_receive_copy_operating_parameters_data { uint8_t length[4]; uint8_t snlid; #define RCOP_SNLID 0x01 uint8_t reserved[3]; uint8_t maximum_cscd_descriptor_count[2]; uint8_t maximum_segment_descriptor_count[2]; uint8_t maximum_descriptor_list_length[4]; uint8_t maximum_segment_length[4]; uint8_t maximum_inline_data_length[4]; uint8_t held_data_limit[4]; uint8_t maximum_stream_device_transfer_size[4]; uint8_t reserved2[2]; uint8_t total_concurrent_copies[2]; uint8_t maximum_concurrent_copies; uint8_t data_segment_granularity; uint8_t inline_data_granularity; uint8_t held_data_granularity; uint8_t reserved3[3]; uint8_t implemented_descriptor_list_length; uint8_t list_of_implemented_descriptor_type_codes[0]; }; struct scsi_extended_copy { uint8_t opcode; uint8_t service_action; #define EC_EC_LID1 0x00 #define EC_EC_LID4 0x01 uint8_t reserved[8]; uint8_t length[4]; uint8_t reserved1; uint8_t control; }; struct scsi_ec_cscd_dtsp { uint8_t flags; #define EC_CSCD_FIXED 0x01 #define EC_CSCD_PAD 0x04 uint8_t block_length[3]; }; struct scsi_ec_cscd { uint8_t type_code; #define EC_CSCD_EXT 0xff uint8_t luidt_pdt; #define EC_LUIDT_MASK 0xc0 #define EC_LUIDT_LUN 0x00 #define EC_LUIDT_PROXY_TOKEN 0x40 uint8_t relative_initiator_port[2]; uint8_t cscd_params[24]; struct scsi_ec_cscd_dtsp dtsp; }; struct scsi_ec_cscd_id { uint8_t type_code; #define EC_CSCD_ID 0xe4 uint8_t luidt_pdt; uint8_t relative_initiator_port[2]; uint8_t codeset; uint8_t id_type; uint8_t reserved; uint8_t length; uint8_t designator[20]; struct scsi_ec_cscd_dtsp dtsp; }; struct scsi_ec_segment { uint8_t type_code; uint8_t flags; #define EC_SEG_DC 0x02 #define EC_SEG_CAT 0x01 uint8_t descr_length[2]; uint8_t params[]; }; struct scsi_ec_segment_b2b { uint8_t type_code; #define EC_SEG_B2B 0x02 uint8_t flags; uint8_t descr_length[2]; uint8_t src_cscd[2]; uint8_t dst_cscd[2]; uint8_t reserved[2]; uint8_t number_of_blocks[2]; uint8_t src_lba[8]; uint8_t dst_lba[8]; }; struct scsi_ec_segment_verify { uint8_t type_code; #define EC_SEG_VERIFY 0x07 uint8_t reserved; uint8_t descr_length[2]; uint8_t src_cscd[2]; uint8_t reserved2[2]; uint8_t tur; uint8_t reserved3[3]; }; struct scsi_ec_segment_register_key { uint8_t type_code; #define EC_SEG_REGISTER_KEY 0x14 uint8_t reserved; uint8_t descr_length[2]; uint8_t reserved2[2]; uint8_t dst_cscd[2]; uint8_t res_key[8]; uint8_t sa_res_key[8]; uint8_t reserved3[4]; }; struct scsi_extended_copy_lid1_data { uint8_t list_identifier; uint8_t flags; #define EC_PRIORITY 0x07 #define EC_LIST_ID_USAGE_MASK 0x18 #define EC_LIST_ID_USAGE_FULL 0x08 #define EC_LIST_ID_USAGE_NOHOLD 0x10 #define EC_LIST_ID_USAGE_NONE 0x18 #define EC_STR 0x20 uint8_t cscd_list_length[2]; uint8_t reserved[4]; uint8_t segment_list_length[4]; uint8_t inline_data_length[4]; uint8_t data[]; }; struct scsi_extended_copy_lid4_data { uint8_t list_format; #define EC_LIST_FORMAT 0x01 uint8_t flags; uint8_t header_cscd_list_length[2]; uint8_t reserved[11]; uint8_t flags2; #define EC_IMMED 0x01 #define EC_G_SENSE 0x02 uint8_t header_cscd_type_code; uint8_t reserved2[3]; uint8_t list_identifier[4]; uint8_t reserved3[18]; uint8_t cscd_list_length[2]; uint8_t segment_list_length[2]; uint8_t inline_data_length[2]; uint8_t data[]; }; struct scsi_copy_operation_abort { uint8_t opcode; uint8_t service_action; #define EC_COA 0x1c uint8_t list_identifier[4]; uint8_t reserved[9]; uint8_t control; }; struct scsi_populate_token { uint8_t opcode; uint8_t service_action; #define EC_PT 0x10 uint8_t reserved[4]; uint8_t list_identifier[4]; uint8_t length[4]; uint8_t group_number; uint8_t control; }; struct scsi_range_desc { uint8_t lba[8]; uint8_t length[4]; uint8_t reserved[4]; }; struct scsi_populate_token_data { uint8_t length[2]; uint8_t flags; #define EC_PT_IMMED 0x01 #define EC_PT_RTV 0x02 uint8_t reserved; uint8_t inactivity_timeout[4]; uint8_t rod_type[4]; uint8_t reserved2[2]; uint8_t range_descriptor_length[2]; struct scsi_range_desc desc[]; }; struct scsi_write_using_token { uint8_t opcode; uint8_t service_action; #define EC_WUT 0x11 uint8_t reserved[4]; uint8_t list_identifier[4]; uint8_t length[4]; uint8_t group_number; uint8_t control; }; struct scsi_write_using_token_data { uint8_t length[2]; uint8_t flags; #define EC_WUT_IMMED 0x01 #define EC_WUT_DEL_TKN 0x02 uint8_t reserved[5]; uint8_t offset_into_rod[8]; uint8_t rod_token[512]; uint8_t reserved2[6]; uint8_t range_descriptor_length[2]; struct scsi_range_desc desc[]; }; struct scsi_receive_rod_token_information { uint8_t opcode; uint8_t service_action; #define RCS_RRTI 0x07 uint8_t list_identifier[4]; uint8_t reserved[4]; uint8_t length[4]; uint8_t reserved2; uint8_t control; }; struct scsi_token { uint8_t type[4]; #define ROD_TYPE_INTERNAL 0x00000000 #define ROD_TYPE_AUR 0x00010000 #define ROD_TYPE_PIT_DEF 0x00800000 #define ROD_TYPE_PIT_VULN 0x00800001 #define ROD_TYPE_PIT_PERS 0x00800002 #define ROD_TYPE_PIT_ANY 0x0080FFFF #define ROD_TYPE_BLOCK_ZERO 0xFFFF0001 uint8_t reserved[2]; uint8_t length[2]; uint8_t body[0]; }; struct scsi_report_all_rod_tokens { uint8_t opcode; uint8_t service_action; #define RCS_RART 0x08 uint8_t reserved[8]; uint8_t length[4]; uint8_t reserved2; uint8_t control; }; struct scsi_report_all_rod_tokens_data { uint8_t available_data[4]; uint8_t reserved[4]; uint8_t rod_management_token_list[]; }; struct ata_pass_16 { u_int8_t opcode; u_int8_t protocol; #define AP_EXTEND 0x01 u_int8_t flags; #define AP_FLAG_TLEN_NO_DATA (0 << 0) #define AP_FLAG_TLEN_FEAT (1 << 0) #define AP_FLAG_TLEN_SECT_CNT (2 << 0) #define AP_FLAG_TLEN_STPSIU (3 << 0) #define AP_FLAG_BYT_BLOK_BYTES (0 << 2) #define AP_FLAG_BYT_BLOK_BLOCKS (1 << 2) #define AP_FLAG_TDIR_TO_DEV (0 << 3) #define AP_FLAG_TDIR_FROM_DEV (1 << 3) #define AP_FLAG_CHK_COND (1 << 5) u_int8_t features_ext; u_int8_t features; u_int8_t sector_count_ext; u_int8_t sector_count; u_int8_t lba_low_ext; u_int8_t lba_low; u_int8_t lba_mid_ext; u_int8_t lba_mid; u_int8_t lba_high_ext; u_int8_t lba_high; u_int8_t device; u_int8_t command; u_int8_t control; }; #define SC_SCSI_1 0x01 #define SC_SCSI_2 0x03 /* * Opcodes */ #define TEST_UNIT_READY 0x00 #define REQUEST_SENSE 0x03 #define READ_6 0x08 #define WRITE_6 0x0A #define INQUIRY 0x12 #define MODE_SELECT_6 0x15 #define MODE_SENSE_6 0x1A #define START_STOP_UNIT 0x1B #define START_STOP 0x1B #define RESERVE 0x16 #define RELEASE 0x17 #define RECEIVE_DIAGNOSTIC 0x1C #define SEND_DIAGNOSTIC 0x1D #define PREVENT_ALLOW 0x1E #define READ_CAPACITY 0x25 #define READ_10 0x28 #define WRITE_10 0x2A #define POSITION_TO_ELEMENT 0x2B #define WRITE_VERIFY_10 0x2E #define VERIFY_10 0x2F #define SYNCHRONIZE_CACHE 0x35 #define READ_DEFECT_DATA_10 0x37 #define WRITE_BUFFER 0x3B #define READ_BUFFER 0x3C #define CHANGE_DEFINITION 0x40 #define WRITE_SAME_10 0x41 #define UNMAP 0x42 #define LOG_SELECT 0x4C #define LOG_SENSE 0x4D #define MODE_SELECT_10 0x55 #define RESERVE_10 0x56 #define RELEASE_10 0x57 #define MODE_SENSE_10 0x5A #define PERSISTENT_RES_IN 0x5E #define PERSISTENT_RES_OUT 0x5F #define EXTENDED_COPY 0x83 #define RECEIVE_COPY_STATUS 0x84 #define ATA_PASS_16 0x85 #define READ_16 0x88 #define COMPARE_AND_WRITE 0x89 #define WRITE_16 0x8A #define WRITE_VERIFY_16 0x8E #define VERIFY_16 0x8F #define SYNCHRONIZE_CACHE_16 0x91 #define WRITE_SAME_16 0x93 #define WRITE_ATOMIC_16 0x9C #define SERVICE_ACTION_IN 0x9E #define REPORT_LUNS 0xA0 #define ATA_PASS_12 0xA1 #define MAINTENANCE_IN 0xA3 #define MAINTENANCE_OUT 0xA4 #define MOVE_MEDIUM 0xA5 #define READ_12 0xA8 #define WRITE_12 0xAA #define WRITE_VERIFY_12 0xAE #define VERIFY_12 0xAF #define READ_ELEMENT_STATUS 0xB8 #define READ_CD 0xBE /* Maintenance In Service Action Codes */ #define REPORT_IDENTIFYING_INFRMATION 0x05 #define REPORT_TARGET_PORT_GROUPS 0x0A #define REPORT_ALIASES 0x0B #define REPORT_SUPPORTED_OPERATION_CODES 0x0C #define REPORT_SUPPORTED_TASK_MANAGEMENT_FUNCTIONS 0x0D #define REPORT_PRIORITY 0x0E #define REPORT_TIMESTAMP 0x0F #define MANAGEMENT_PROTOCOL_IN 0x10 /* Maintenance Out Service Action Codes */ #define SET_IDENTIFY_INFORMATION 0x06 #define SET_TARGET_PORT_GROUPS 0x0A #define CHANGE_ALIASES 0x0B #define SET_PRIORITY 0x0E #define SET_TIMESTAMP 0x0F #define MANGAEMENT_PROTOCOL_OUT 0x10 /* * Device Types */ #define T_DIRECT 0x00 #define T_SEQUENTIAL 0x01 #define T_PRINTER 0x02 #define T_PROCESSOR 0x03 #define T_WORM 0x04 #define T_CDROM 0x05 #define T_SCANNER 0x06 #define T_OPTICAL 0x07 #define T_CHANGER 0x08 #define T_COMM 0x09 #define T_ASC0 0x0a #define T_ASC1 0x0b #define T_STORARRAY 0x0c #define T_ENCLOSURE 0x0d #define T_RBC 0x0e #define T_OCRW 0x0f #define T_OSD 0x11 #define T_ADC 0x12 #define T_NODEVICE 0x1f #define T_ANY 0xff /* Used in Quirk table matches */ #define T_REMOV 1 #define T_FIXED 0 /* * This length is the initial inquiry length used by the probe code, as * well as the length necessary for scsi_print_inquiry() to function * correctly. If either use requires a different length in the future, * the two values should be de-coupled. */ #define SHORT_INQUIRY_LENGTH 36 struct scsi_inquiry_data { u_int8_t device; #define SID_TYPE(inq_data) ((inq_data)->device & 0x1f) #define SID_QUAL(inq_data) (((inq_data)->device & 0xE0) >> 5) #define SID_QUAL_LU_CONNECTED 0x00 /* * The specified peripheral device * type is currently connected to * logical unit. If the target cannot * determine whether or not a physical * device is currently connected, it * shall also use this peripheral * qualifier when returning the INQUIRY * data. This peripheral qualifier * does not mean that the device is * ready for access by the initiator. */ #define SID_QUAL_LU_OFFLINE 0x01 /* * The target is capable of supporting * the specified peripheral device type * on this logical unit; however, the * physical device is not currently * connected to this logical unit. */ #define SID_QUAL_RSVD 0x02 #define SID_QUAL_BAD_LU 0x03 /* * The target is not capable of * supporting a physical device on * this logical unit. For this * peripheral qualifier the peripheral * device type shall be set to 1Fh to * provide compatibility with previous * versions of SCSI. All other * peripheral device type values are * reserved for this peripheral * qualifier. */ #define SID_QUAL_IS_VENDOR_UNIQUE(inq_data) ((SID_QUAL(inq_data) & 0x08) != 0) u_int8_t dev_qual2; #define SID_QUAL2 0x7F #define SID_LU_CONG 0x40 #define SID_RMB 0x80 #define SID_IS_REMOVABLE(inq_data) (((inq_data)->dev_qual2 & SID_RMB) != 0) u_int8_t version; #define SID_ANSI_REV(inq_data) ((inq_data)->version & 0x07) #define SCSI_REV_0 0 #define SCSI_REV_CCS 1 #define SCSI_REV_2 2 #define SCSI_REV_SPC 3 #define SCSI_REV_SPC2 4 #define SCSI_REV_SPC3 5 #define SCSI_REV_SPC4 6 #define SID_ECMA 0x38 #define SID_ISO 0xC0 u_int8_t response_format; #define SID_AENC 0x80 #define SID_TrmIOP 0x40 #define SID_NormACA 0x20 #define SID_HiSup 0x10 u_int8_t additional_length; #define SID_ADDITIONAL_LENGTH(iqd) \ ((iqd)->additional_length + \ __offsetof(struct scsi_inquiry_data, additional_length) + 1) u_int8_t spc3_flags; #define SPC3_SID_PROTECT 0x01 #define SPC3_SID_3PC 0x08 #define SPC3_SID_TPGS_MASK 0x30 #define SPC3_SID_TPGS_IMPLICIT 0x10 #define SPC3_SID_TPGS_EXPLICIT 0x20 #define SPC3_SID_ACC 0x40 #define SPC3_SID_SCCS 0x80 u_int8_t spc2_flags; #define SPC2_SID_ADDR16 0x01 #define SPC2_SID_MChngr 0x08 #define SPC2_SID_MultiP 0x10 #define SPC2_SID_EncServ 0x40 #define SPC2_SID_BQueue 0x80 #define INQ_DATA_TQ_ENABLED(iqd) \ ((SID_ANSI_REV(iqd) < SCSI_REV_SPC2)? ((iqd)->flags & SID_CmdQue) : \ (((iqd)->flags & SID_CmdQue) && !((iqd)->spc2_flags & SPC2_SID_BQueue)) || \ (!((iqd)->flags & SID_CmdQue) && ((iqd)->spc2_flags & SPC2_SID_BQueue))) u_int8_t flags; #define SID_SftRe 0x01 #define SID_CmdQue 0x02 #define SID_Linked 0x08 #define SID_Sync 0x10 #define SID_WBus16 0x20 #define SID_WBus32 0x40 #define SID_RelAdr 0x80 #define SID_VENDOR_SIZE 8 char vendor[SID_VENDOR_SIZE]; #define SID_PRODUCT_SIZE 16 char product[SID_PRODUCT_SIZE]; #define SID_REVISION_SIZE 4 char revision[SID_REVISION_SIZE]; /* * The following fields were taken from SCSI Primary Commands - 2 * (SPC-2) Revision 14, Dated 11 November 1999 */ #define SID_VENDOR_SPECIFIC_0_SIZE 20 u_int8_t vendor_specific0[SID_VENDOR_SPECIFIC_0_SIZE]; /* * An extension of SCSI Parallel Specific Values */ #define SID_SPI_IUS 0x01 #define SID_SPI_QAS 0x02 #define SID_SPI_CLOCK_ST 0x00 #define SID_SPI_CLOCK_DT 0x04 #define SID_SPI_CLOCK_DT_ST 0x0C #define SID_SPI_MASK 0x0F u_int8_t spi3data; u_int8_t reserved2; /* * Version Descriptors, stored 2 byte values. */ u_int8_t version1[2]; u_int8_t version2[2]; u_int8_t version3[2]; u_int8_t version4[2]; u_int8_t version5[2]; u_int8_t version6[2]; u_int8_t version7[2]; u_int8_t version8[2]; u_int8_t reserved3[22]; #define SID_VENDOR_SPECIFIC_1_SIZE 160 u_int8_t vendor_specific1[SID_VENDOR_SPECIFIC_1_SIZE]; }; /* * This structure is more suited to initiator operation, because the * maximum number of supported pages is already allocated. */ struct scsi_vpd_supported_page_list { u_int8_t device; u_int8_t page_code; #define SVPD_SUPPORTED_PAGE_LIST 0x00 #define SVPD_SUPPORTED_PAGES_HDR_LEN 4 u_int8_t reserved; u_int8_t length; /* number of VPD entries */ #define SVPD_SUPPORTED_PAGES_SIZE 251 u_int8_t list[SVPD_SUPPORTED_PAGES_SIZE]; }; /* * This structure is more suited to target operation, because the * number of supported pages is left to the user to allocate. */ struct scsi_vpd_supported_pages { u_int8_t device; u_int8_t page_code; u_int8_t reserved; #define SVPD_SUPPORTED_PAGES 0x00 u_int8_t length; u_int8_t page_list[0]; }; struct scsi_vpd_unit_serial_number { u_int8_t device; u_int8_t page_code; #define SVPD_UNIT_SERIAL_NUMBER 0x80 u_int8_t reserved; u_int8_t length; /* serial number length */ #define SVPD_SERIAL_NUM_SIZE 251 u_int8_t serial_num[SVPD_SERIAL_NUM_SIZE]; }; struct scsi_vpd_device_id { u_int8_t device; u_int8_t page_code; #define SVPD_DEVICE_ID 0x83 #define SVPD_DEVICE_ID_MAX_SIZE 252 #define SVPD_DEVICE_ID_HDR_LEN \ __offsetof(struct scsi_vpd_device_id, desc_list) u_int8_t length[2]; u_int8_t desc_list[]; }; struct scsi_vpd_id_descriptor { u_int8_t proto_codeset; /* * See the SCSI_PROTO definitions above for the protocols. */ #define SVPD_ID_PROTO_SHIFT 4 #define SVPD_ID_CODESET_BINARY 0x01 #define SVPD_ID_CODESET_ASCII 0x02 #define SVPD_ID_CODESET_UTF8 0x03 #define SVPD_ID_CODESET_MASK 0x0f u_int8_t id_type; #define SVPD_ID_PIV 0x80 #define SVPD_ID_ASSOC_LUN 0x00 #define SVPD_ID_ASSOC_PORT 0x10 #define SVPD_ID_ASSOC_TARGET 0x20 #define SVPD_ID_ASSOC_MASK 0x30 #define SVPD_ID_TYPE_VENDOR 0x00 #define SVPD_ID_TYPE_T10 0x01 #define SVPD_ID_TYPE_EUI64 0x02 #define SVPD_ID_TYPE_NAA 0x03 #define SVPD_ID_TYPE_RELTARG 0x04 #define SVPD_ID_TYPE_TPORTGRP 0x05 #define SVPD_ID_TYPE_LUNGRP 0x06 #define SVPD_ID_TYPE_MD5_LUN_ID 0x07 #define SVPD_ID_TYPE_SCSI_NAME 0x08 #define SVPD_ID_TYPE_MASK 0x0f u_int8_t reserved; u_int8_t length; #define SVPD_DEVICE_ID_DESC_HDR_LEN \ __offsetof(struct scsi_vpd_id_descriptor, identifier) u_int8_t identifier[]; }; struct scsi_vpd_id_t10 { u_int8_t vendor[8]; u_int8_t vendor_spec_id[0]; }; struct scsi_vpd_id_eui64 { u_int8_t ieee_company_id[3]; u_int8_t extension_id[5]; }; struct scsi_vpd_id_naa_basic { uint8_t naa; /* big endian, packed: uint8_t naa : 4; uint8_t naa_desig : 4; */ #define SVPD_ID_NAA_NAA_SHIFT 4 #define SVPD_ID_NAA_IEEE_EXT 0x02 #define SVPD_ID_NAA_LOCAL_REG 0x03 #define SVPD_ID_NAA_IEEE_REG 0x05 #define SVPD_ID_NAA_IEEE_REG_EXT 0x06 uint8_t naa_data[]; }; struct scsi_vpd_id_naa_ieee_extended_id { uint8_t naa; uint8_t vendor_specific_id_a; uint8_t ieee_company_id[3]; uint8_t vendor_specific_id_b[4]; }; struct scsi_vpd_id_naa_local_reg { uint8_t naa; uint8_t local_value[7]; }; struct scsi_vpd_id_naa_ieee_reg { uint8_t naa; uint8_t reg_value[7]; /* big endian, packed: uint8_t naa_basic : 4; uint8_t ieee_company_id_0 : 4; uint8_t ieee_company_id_1[2]; uint8_t ieee_company_id_2 : 4; uint8_t vendor_specific_id_0 : 4; uint8_t vendor_specific_id_1[4]; */ }; struct scsi_vpd_id_naa_ieee_reg_extended { uint8_t naa; uint8_t reg_value[15]; /* big endian, packed: uint8_t naa_basic : 4; uint8_t ieee_company_id_0 : 4; uint8_t ieee_company_id_1[2]; uint8_t ieee_company_id_2 : 4; uint8_t vendor_specific_id_0 : 4; uint8_t vendor_specific_id_1[4]; uint8_t vendor_specific_id_ext[8]; */ }; struct scsi_vpd_id_rel_trgt_port_id { uint8_t obsolete[2]; uint8_t rel_trgt_port_id[2]; }; struct scsi_vpd_id_trgt_port_grp_id { uint8_t reserved[2]; uint8_t trgt_port_grp[2]; }; struct scsi_vpd_id_lun_grp_id { uint8_t reserved[2]; uint8_t log_unit_grp[2]; }; struct scsi_vpd_id_md5_lun_id { uint8_t lun_id[16]; }; struct scsi_vpd_id_scsi_name { uint8_t name_string[256]; }; struct scsi_service_action_in { uint8_t opcode; uint8_t service_action; uint8_t action_dependent[13]; uint8_t control; }; struct scsi_vpd_extended_inquiry_data { uint8_t device; uint8_t page_code; #define SVPD_EXTENDED_INQUIRY_DATA 0x86 uint8_t reserved; uint8_t page_length; uint8_t flags1; #define SVPD_EID_AM 0xC0 #define SVPD_EID_SPT 0x38 #define SVPD_EID_SPT_1 0x00 #define SVPD_EID_SPT_12 0x08 #define SVPD_EID_SPT_2 0x10 #define SVPD_EID_SPT_13 0x18 #define SVPD_EID_SPT_3 0x20 #define SVPD_EID_SPT_23 0x28 #define SVPD_EID_SPT_123 0x38 #define SVPD_EID_GRD_CHK 0x04 #define SVPD_EID_APP_CHK 0x02 #define SVPD_EID_REF_CHK 0x01 uint8_t flags2; #define SVPD_EID_UASK_SUP 0x20 #define SVPD_EID_GROUP_SUP 0x10 #define SVPD_EID_PRIOR_SUP 0x08 #define SVPD_EID_HEADSUP 0x04 #define SVPD_EID_ORDSUP 0x02 #define SVPD_EID_SIMPSUP 0x01 uint8_t flags3; #define SVPD_EID_WU_SUP 0x08 #define SVPD_EID_CRD_SUP 0x04 #define SVPD_EID_NV_SUP 0x02 #define SVPD_EID_V_SUP 0x01 uint8_t flags4; #define SVPD_EID_P_I_I_SUP 0x10 #define SVPD_EID_LUICLT 0x01 uint8_t flags5; #define SVPD_EID_R_SUP 0x10 #define SVPD_EID_CBCS 0x01 uint8_t flags6; #define SVPD_EID_MULTI_I_T_FW 0x0F uint8_t est[2]; uint8_t flags7; #define SVPD_EID_POA_SUP 0x80 #define SVPD_EID_HRA_SUP 0x80 #define SVPD_EID_VSA_SUP 0x80 uint8_t max_sense_length; uint8_t reserved2[50]; }; struct scsi_vpd_mode_page_policy_descr { uint8_t page_code; uint8_t subpage_code; uint8_t policy; #define SVPD_MPP_SHARED 0x00 #define SVPD_MPP_PORT 0x01 #define SVPD_MPP_I_T 0x03 #define SVPD_MPP_MLUS 0x80 uint8_t reserved; }; struct scsi_vpd_mode_page_policy { uint8_t device; uint8_t page_code; #define SVPD_MODE_PAGE_POLICY 0x87 uint8_t page_length[2]; struct scsi_vpd_mode_page_policy_descr descr[0]; }; struct scsi_diag_page { uint8_t page_code; uint8_t page_specific_flags; uint8_t length[2]; uint8_t params[0]; }; struct scsi_vpd_port_designation { uint8_t reserved[2]; uint8_t relative_port_id[2]; uint8_t reserved2[2]; uint8_t initiator_transportid_length[2]; uint8_t initiator_transportid[0]; }; struct scsi_vpd_port_designation_cont { uint8_t reserved[2]; uint8_t target_port_descriptors_length[2]; struct scsi_vpd_id_descriptor target_port_descriptors[0]; }; struct scsi_vpd_scsi_ports { u_int8_t device; u_int8_t page_code; #define SVPD_SCSI_PORTS 0x88 u_int8_t page_length[2]; struct scsi_vpd_port_designation design[]; }; /* * ATA Information VPD Page based on * T10/2126-D Revision 04 */ #define SVPD_ATA_INFORMATION 0x89 struct scsi_vpd_tpc_descriptor { uint8_t desc_type[2]; uint8_t desc_length[2]; uint8_t parameters[]; }; struct scsi_vpd_tpc_descriptor_bdrl { uint8_t desc_type[2]; #define SVPD_TPC_BDRL 0x0000 uint8_t desc_length[2]; uint8_t vendor_specific[6]; uint8_t maximum_ranges[2]; uint8_t maximum_inactivity_timeout[4]; uint8_t default_inactivity_timeout[4]; uint8_t maximum_token_transfer_size[8]; uint8_t optimal_transfer_count[8]; }; struct scsi_vpd_tpc_descriptor_sc_descr { uint8_t opcode; uint8_t sa_length; uint8_t supported_service_actions[0]; }; struct scsi_vpd_tpc_descriptor_sc { uint8_t desc_type[2]; #define SVPD_TPC_SC 0x0001 uint8_t desc_length[2]; uint8_t list_length; struct scsi_vpd_tpc_descriptor_sc_descr descr[]; }; struct scsi_vpd_tpc_descriptor_pd { uint8_t desc_type[2]; #define SVPD_TPC_PD 0x0004 uint8_t desc_length[2]; uint8_t reserved[4]; uint8_t maximum_cscd_descriptor_count[2]; uint8_t maximum_segment_descriptor_count[2]; uint8_t maximum_descriptor_list_length[4]; uint8_t maximum_inline_data_length[4]; uint8_t reserved2[12]; }; struct scsi_vpd_tpc_descriptor_sd { uint8_t desc_type[2]; #define SVPD_TPC_SD 0x0008 uint8_t desc_length[2]; uint8_t list_length; uint8_t supported_descriptor_codes[]; }; struct scsi_vpd_tpc_descriptor_sdid { uint8_t desc_type[2]; #define SVPD_TPC_SDID 0x000C uint8_t desc_length[2]; uint8_t list_length[2]; uint8_t supported_descriptor_ids[]; }; struct scsi_vpd_tpc_descriptor_rtf_block { uint8_t type_format; #define SVPD_TPC_RTF_BLOCK 0x00 uint8_t reserved; uint8_t desc_length[2]; uint8_t reserved2[2]; uint8_t optimal_length_granularity[2]; uint8_t maximum_bytes[8]; uint8_t optimal_bytes[8]; uint8_t optimal_bytes_to_token_per_segment[8]; uint8_t optimal_bytes_from_token_per_segment[8]; uint8_t reserved3[8]; }; struct scsi_vpd_tpc_descriptor_rtf { uint8_t desc_type[2]; #define SVPD_TPC_RTF 0x0106 uint8_t desc_length[2]; uint8_t remote_tokens; uint8_t reserved[11]; uint8_t minimum_token_lifetime[4]; uint8_t maximum_token_lifetime[4]; uint8_t maximum_token_inactivity_timeout[4]; uint8_t reserved2[18]; uint8_t type_specific_features_length[2]; uint8_t type_specific_features[0]; }; struct scsi_vpd_tpc_descriptor_srtd { uint8_t rod_type[4]; uint8_t flags; #define SVPD_TPC_SRTD_TOUT 0x01 #define SVPD_TPC_SRTD_TIN 0x02 #define SVPD_TPC_SRTD_ECPY 0x80 uint8_t reserved; uint8_t preference_indicator[2]; uint8_t reserved2[56]; }; struct scsi_vpd_tpc_descriptor_srt { uint8_t desc_type[2]; #define SVPD_TPC_SRT 0x0108 uint8_t desc_length[2]; uint8_t reserved[2]; uint8_t rod_type_descriptors_length[2]; uint8_t rod_type_descriptors[0]; }; struct scsi_vpd_tpc_descriptor_gco { uint8_t desc_type[2]; #define SVPD_TPC_GCO 0x8001 uint8_t desc_length[2]; uint8_t total_concurrent_copies[4]; uint8_t maximum_identified_concurrent_copies[4]; uint8_t maximum_segment_length[4]; uint8_t data_segment_granularity; uint8_t inline_data_granularity; uint8_t reserved[18]; }; struct scsi_vpd_tpc { uint8_t device; uint8_t page_code; #define SVPD_SCSI_TPC 0x8F uint8_t page_length[2]; struct scsi_vpd_tpc_descriptor descr[]; }; /* * Block Device Characteristics VPD Page based on * T10/1799-D Revision 31 */ struct scsi_vpd_block_characteristics { u_int8_t device; u_int8_t page_code; #define SVPD_BDC 0xB1 u_int8_t page_length[2]; u_int8_t medium_rotation_rate[2]; #define SVPD_BDC_RATE_NOT_REPORTED 0x00 #define SVPD_BDC_RATE_NON_ROTATING 0x01 u_int8_t reserved1; u_int8_t nominal_form_factor; #define SVPD_BDC_FORM_NOT_REPORTED 0x00 #define SVPD_BDC_FORM_5_25INCH 0x01 #define SVPD_BDC_FORM_3_5INCH 0x02 #define SVPD_BDC_FORM_2_5INCH 0x03 #define SVPD_BDC_FORM_1_5INCH 0x04 #define SVPD_BDC_FORM_LESSTHAN_1_5INCH 0x05 u_int8_t reserved2[56]; }; /* * Block Device Characteristics VPD Page */ struct scsi_vpd_block_device_characteristics { uint8_t device; uint8_t page_code; #define SVPD_BDC 0xB1 uint8_t page_length[2]; uint8_t medium_rotation_rate[2]; #define SVPD_NOT_REPORTED 0x0000 #define SVPD_NON_ROTATING 0x0001 uint8_t product_type; uint8_t wab_wac_ff; uint8_t flags; #define SVPD_VBULS 0x01 #define SVPD_FUAB 0x02 #define SVPD_HAW_ZBC 0x10 uint8_t reserved[55]; }; /* * Logical Block Provisioning VPD Page based on * T10/1799-D Revision 31 */ struct scsi_vpd_logical_block_prov { u_int8_t device; u_int8_t page_code; #define SVPD_LBP 0xB2 u_int8_t page_length[2]; #define SVPD_LBP_PL_BASIC 0x04 u_int8_t threshold_exponent; u_int8_t flags; #define SVPD_LBP_UNMAP 0x80 #define SVPD_LBP_WS16 0x40 #define SVPD_LBP_WS10 0x20 #define SVPD_LBP_RZ 0x04 #define SVPD_LBP_ANC_SUP 0x02 #define SVPD_LBP_DP 0x01 u_int8_t prov_type; #define SVPD_LBP_RESOURCE 0x01 #define SVPD_LBP_THIN 0x02 u_int8_t reserved; /* * Provisioning Group Descriptor can be here if SVPD_LBP_DP is set * Its size can be determined from page_length - 4 */ }; /* * Block Limits VDP Page based on SBC-4 Revision 2 */ struct scsi_vpd_block_limits { u_int8_t device; u_int8_t page_code; #define SVPD_BLOCK_LIMITS 0xB0 u_int8_t page_length[2]; #define SVPD_BL_PL_BASIC 0x10 #define SVPD_BL_PL_TP 0x3C u_int8_t reserved1; u_int8_t max_cmp_write_len; u_int8_t opt_txfer_len_grain[2]; u_int8_t max_txfer_len[4]; u_int8_t opt_txfer_len[4]; u_int8_t max_prefetch[4]; u_int8_t max_unmap_lba_cnt[4]; u_int8_t max_unmap_blk_cnt[4]; u_int8_t opt_unmap_grain[4]; u_int8_t unmap_grain_align[4]; u_int8_t max_write_same_length[8]; u_int8_t max_atomic_transfer_length[4]; u_int8_t atomic_alignment[4]; u_int8_t atomic_transfer_length_granularity[4]; u_int8_t reserved2[8]; }; struct scsi_read_capacity { u_int8_t opcode; u_int8_t byte2; #define SRC_RELADR 0x01 u_int8_t addr[4]; u_int8_t unused[2]; u_int8_t pmi; #define SRC_PMI 0x01 u_int8_t control; }; struct scsi_read_capacity_16 { uint8_t opcode; #define SRC16_SERVICE_ACTION 0x10 uint8_t service_action; uint8_t addr[8]; uint8_t alloc_len[4]; #define SRC16_PMI 0x01 #define SRC16_RELADR 0x02 uint8_t reladr; uint8_t control; }; struct scsi_read_capacity_data { u_int8_t addr[4]; u_int8_t length[4]; }; struct scsi_read_capacity_data_long { uint8_t addr[8]; uint8_t length[4]; #define SRC16_PROT_EN 0x01 #define SRC16_P_TYPE 0x0e #define SRC16_PTYPE_1 0x00 #define SRC16_PTYPE_2 0x02 #define SRC16_PTYPE_3 0x04 uint8_t prot; #define SRC16_LBPPBE 0x0f #define SRC16_PI_EXPONENT 0xf0 #define SRC16_PI_EXPONENT_SHIFT 4 uint8_t prot_lbppbe; #define SRC16_LALBA 0x3f #define SRC16_LBPRZ 0x40 #define SRC16_LBPME 0x80 /* * Alternate versions of these macros that are intended for use on a 16-bit * version of the lalba_lbp field instead of the array of 2 8 bit numbers. */ #define SRC16_LALBA_A 0x3fff #define SRC16_LBPRZ_A 0x4000 #define SRC16_LBPME_A 0x8000 uint8_t lalba_lbp[2]; uint8_t reserved[16]; }; struct scsi_report_luns { uint8_t opcode; uint8_t reserved1; #define RPL_REPORT_DEFAULT 0x00 #define RPL_REPORT_WELLKNOWN 0x01 #define RPL_REPORT_ALL 0x02 uint8_t select_report; uint8_t reserved2[3]; uint8_t length[4]; uint8_t reserved3; uint8_t control; }; struct scsi_report_luns_lundata { uint8_t lundata[8]; #define RPL_LUNDATA_PERIPH_BUS_MASK 0x3f #define RPL_LUNDATA_FLAT_LUN_MASK 0x3f #define RPL_LUNDATA_FLAT_LUN_BITS 0x06 #define RPL_LUNDATA_LUN_TARG_MASK 0x3f #define RPL_LUNDATA_LUN_BUS_MASK 0xe0 #define RPL_LUNDATA_LUN_LUN_MASK 0x1f #define RPL_LUNDATA_EXT_LEN_MASK 0x30 #define RPL_LUNDATA_EXT_EAM_MASK 0x0f #define RPL_LUNDATA_EXT_EAM_WK 0x01 #define RPL_LUNDATA_EXT_EAM_NOT_SPEC 0x0f #define RPL_LUNDATA_ATYP_MASK 0xc0 /* MBZ for type 0 lun */ #define RPL_LUNDATA_ATYP_PERIPH 0x00 #define RPL_LUNDATA_ATYP_FLAT 0x40 #define RPL_LUNDATA_ATYP_LUN 0x80 #define RPL_LUNDATA_ATYP_EXTLUN 0xc0 }; struct scsi_report_luns_data { u_int8_t length[4]; /* length of LUN inventory, in bytes */ u_int8_t reserved[4]; /* unused */ /* * LUN inventory- we only support the type zero form for now. */ struct scsi_report_luns_lundata luns[0]; }; struct scsi_target_group { uint8_t opcode; uint8_t service_action; #define STG_PDF_MASK 0xe0 #define STG_PDF_LENGTH 0x00 #define STG_PDF_EXTENDED 0x20 uint8_t reserved1[4]; uint8_t length[4]; uint8_t reserved2; uint8_t control; }; struct scsi_target_port_descriptor { uint8_t reserved[2]; uint8_t relative_target_port_identifier[2]; uint8_t desc_list[]; }; struct scsi_target_port_group_descriptor { uint8_t pref_state; #define TPG_PRIMARY 0x80 #define TPG_ASYMMETRIC_ACCESS_STATE_MASK 0xf #define TPG_ASYMMETRIC_ACCESS_OPTIMIZED 0x0 #define TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED 0x1 #define TPG_ASYMMETRIC_ACCESS_STANDBY 0x2 #define TPG_ASYMMETRIC_ACCESS_UNAVAILABLE 0x3 #define TPG_ASYMMETRIC_ACCESS_LBA_DEPENDENT 0x4 #define TPG_ASYMMETRIC_ACCESS_OFFLINE 0xE #define TPG_ASYMMETRIC_ACCESS_TRANSITIONING 0xF uint8_t support; #define TPG_AO_SUP 0x01 #define TPG_AN_SUP 0x02 #define TPG_S_SUP 0x04 #define TPG_U_SUP 0x08 #define TPG_LBD_SUP 0x10 #define TPG_O_SUP 0x40 #define TPG_T_SUP 0x80 uint8_t target_port_group[2]; uint8_t reserved; uint8_t status; #define TPG_UNAVLBL 0 #define TPG_SET_BY_STPG 0x01 #define TPG_IMPLICIT 0x02 uint8_t vendor_specific; uint8_t target_port_count; struct scsi_target_port_descriptor descriptors[]; }; struct scsi_target_group_data { uint8_t length[4]; /* length of returned data, in bytes */ struct scsi_target_port_group_descriptor groups[]; }; struct scsi_target_group_data_extended { uint8_t length[4]; /* length of returned data, in bytes */ uint8_t format_type; /* STG_PDF_LENGTH or STG_PDF_EXTENDED */ uint8_t implicit_transition_time; uint8_t reserved[2]; struct scsi_target_port_group_descriptor groups[]; }; typedef enum { SSD_TYPE_NONE, SSD_TYPE_FIXED, SSD_TYPE_DESC } scsi_sense_data_type; typedef enum { SSD_ELEM_NONE, SSD_ELEM_SKIP, SSD_ELEM_DESC, SSD_ELEM_SKS, SSD_ELEM_COMMAND, SSD_ELEM_INFO, SSD_ELEM_FRU, SSD_ELEM_STREAM, SSD_ELEM_MAX } scsi_sense_elem_type; struct scsi_sense_data { uint8_t error_code; /* * SPC-4 says that the maximum length of sense data is 252 bytes. * So this structure is exactly 252 bytes log. */ #define SSD_FULL_SIZE 252 uint8_t sense_buf[SSD_FULL_SIZE - 1]; /* * XXX KDM is this still a reasonable minimum size? */ #define SSD_MIN_SIZE 18 /* * Maximum value for the extra_len field in the sense data. */ #define SSD_EXTRA_MAX 244 }; /* * Fixed format sense data. */ struct scsi_sense_data_fixed { u_int8_t error_code; #define SSD_ERRCODE 0x7F #define SSD_CURRENT_ERROR 0x70 #define SSD_DEFERRED_ERROR 0x71 #define SSD_ERRCODE_VALID 0x80 u_int8_t segment; u_int8_t flags; #define SSD_KEY 0x0F #define SSD_KEY_NO_SENSE 0x00 #define SSD_KEY_RECOVERED_ERROR 0x01 #define SSD_KEY_NOT_READY 0x02 #define SSD_KEY_MEDIUM_ERROR 0x03 #define SSD_KEY_HARDWARE_ERROR 0x04 #define SSD_KEY_ILLEGAL_REQUEST 0x05 #define SSD_KEY_UNIT_ATTENTION 0x06 #define SSD_KEY_DATA_PROTECT 0x07 #define SSD_KEY_BLANK_CHECK 0x08 #define SSD_KEY_Vendor_Specific 0x09 #define SSD_KEY_COPY_ABORTED 0x0a #define SSD_KEY_ABORTED_COMMAND 0x0b #define SSD_KEY_EQUAL 0x0c #define SSD_KEY_VOLUME_OVERFLOW 0x0d #define SSD_KEY_MISCOMPARE 0x0e #define SSD_KEY_COMPLETED 0x0f #define SSD_ILI 0x20 #define SSD_EOM 0x40 #define SSD_FILEMARK 0x80 u_int8_t info[4]; u_int8_t extra_len; u_int8_t cmd_spec_info[4]; u_int8_t add_sense_code; u_int8_t add_sense_code_qual; u_int8_t fru; u_int8_t sense_key_spec[3]; #define SSD_SCS_VALID 0x80 #define SSD_FIELDPTR_CMD 0x40 #define SSD_BITPTR_VALID 0x08 #define SSD_BITPTR_VALUE 0x07 u_int8_t extra_bytes[14]; #define SSD_FIXED_IS_PRESENT(sense, length, field) \ ((length >= (offsetof(struct scsi_sense_data_fixed, field) + \ sizeof(sense->field))) ? 1 :0) #define SSD_FIXED_IS_FILLED(sense, field) \ ((((offsetof(struct scsi_sense_data_fixed, field) + \ sizeof(sense->field)) - \ (offsetof(struct scsi_sense_data_fixed, extra_len) + \ sizeof(sense->extra_len))) <= sense->extra_len) ? 1 : 0) }; /* * Descriptor format sense data definitions. * Introduced in SPC-3. */ struct scsi_sense_data_desc { uint8_t error_code; #define SSD_DESC_CURRENT_ERROR 0x72 #define SSD_DESC_DEFERRED_ERROR 0x73 uint8_t sense_key; uint8_t add_sense_code; uint8_t add_sense_code_qual; uint8_t reserved[3]; /* * Note that SPC-4, section 4.5.2.1 says that the extra_len field * must be less than or equal to 244. */ uint8_t extra_len; uint8_t sense_desc[0]; #define SSD_DESC_IS_PRESENT(sense, length, field) \ ((length >= (offsetof(struct scsi_sense_data_desc, field) + \ sizeof(sense->field))) ? 1 :0) }; struct scsi_sense_desc_header { uint8_t desc_type; uint8_t length; }; /* * The information provide in the Information descriptor is device type or * command specific information, and defined in a command standard. * * Note that any changes to the field names or positions in this structure, * even reserved fields, should be accompanied by an examination of the * code in ctl_set_sense() that uses them. * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_info { uint8_t desc_type; #define SSD_DESC_INFO 0x00 uint8_t length; uint8_t byte2; #define SSD_INFO_VALID 0x80 uint8_t reserved; uint8_t info[8]; }; /* * Command-specific information depends on the command for which the * reported condition occured. * * Note that any changes to the field names or positions in this structure, * even reserved fields, should be accompanied by an examination of the * code in ctl_set_sense() that uses them. * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_command { uint8_t desc_type; #define SSD_DESC_COMMAND 0x01 uint8_t length; uint8_t reserved[2]; uint8_t command_info[8]; }; /* * Sense key specific descriptor. The sense key specific data format * depends on the sense key in question. * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_sks { uint8_t desc_type; #define SSD_DESC_SKS 0x02 uint8_t length; uint8_t reserved1[2]; uint8_t sense_key_spec[3]; #define SSD_SKS_VALID 0x80 uint8_t reserved2; }; /* * This is used for the Illegal Request sense key (0x05) only. */ struct scsi_sense_sks_field { uint8_t byte0; #define SSD_SKS_FIELD_VALID 0x80 #define SSD_SKS_FIELD_CMD 0x40 #define SSD_SKS_BPV 0x08 #define SSD_SKS_BIT_VALUE 0x07 uint8_t field[2]; }; /* * This is used for the Hardware Error (0x04), Medium Error (0x03) and * Recovered Error (0x01) sense keys. */ struct scsi_sense_sks_retry { uint8_t byte0; #define SSD_SKS_RETRY_VALID 0x80 uint8_t actual_retry_count[2]; }; /* * Used with the NO Sense (0x00) or Not Ready (0x02) sense keys. */ struct scsi_sense_sks_progress { uint8_t byte0; #define SSD_SKS_PROGRESS_VALID 0x80 uint8_t progress[2]; #define SSD_SKS_PROGRESS_DENOM 0x10000 }; /* * Used with the Copy Aborted (0x0a) sense key. */ struct scsi_sense_sks_segment { uint8_t byte0; #define SSD_SKS_SEGMENT_VALID 0x80 #define SSD_SKS_SEGMENT_SD 0x20 #define SSD_SKS_SEGMENT_BPV 0x08 #define SSD_SKS_SEGMENT_BITPTR 0x07 uint8_t field[2]; }; /* * Used with the Unit Attention (0x06) sense key. * * This is currently used to indicate that the unit attention condition * queue has overflowed (when the overflow bit is set). */ struct scsi_sense_sks_overflow { uint8_t byte0; #define SSD_SKS_OVERFLOW_VALID 0x80 #define SSD_SKS_OVERFLOW_SET 0x01 uint8_t reserved[2]; }; /* * This specifies which component is associated with the sense data. There * is no standard meaning for the fru value. * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_fru { uint8_t desc_type; #define SSD_DESC_FRU 0x03 uint8_t length; uint8_t reserved; uint8_t fru; }; /* * Used for Stream commands, defined in SSC-4. * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_stream { uint8_t desc_type; #define SSD_DESC_STREAM 0x04 uint8_t length; uint8_t reserved; uint8_t byte3; #define SSD_DESC_STREAM_FM 0x80 #define SSD_DESC_STREAM_EOM 0x40 #define SSD_DESC_STREAM_ILI 0x20 }; /* * Used for Block commands, defined in SBC-3. * * This is currently (as of SBC-3) only used for the Incorrect Length * Indication (ILI) bit, which says that the data length requested in the * READ LONG or WRITE LONG command did not match the length of the logical * block. * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_block { uint8_t desc_type; #define SSD_DESC_BLOCK 0x05 uint8_t length; uint8_t reserved; uint8_t byte3; #define SSD_DESC_BLOCK_ILI 0x20 }; /* * Used for Object-Based Storage Devices (OSD-3). * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_osd_objid { uint8_t desc_type; #define SSD_DESC_OSD_OBJID 0x06 uint8_t length; uint8_t reserved[6]; /* * XXX KDM provide the bit definitions here? There are a lot of * them, and we don't have an OSD driver yet. */ uint8_t not_init_cmds[4]; uint8_t completed_cmds[4]; uint8_t partition_id[8]; uint8_t object_id[8]; }; /* * Used for Object-Based Storage Devices (OSD-3). * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_osd_integrity { uint8_t desc_type; #define SSD_DESC_OSD_INTEGRITY 0x07 uint8_t length; uint8_t integ_check_val[32]; }; /* * Used for Object-Based Storage Devices (OSD-3). * * Maximum descriptors allowed: 1 (as of SPC-4) */ struct scsi_sense_osd_attr_id { uint8_t desc_type; #define SSD_DESC_OSD_ATTR_ID 0x08 uint8_t length; uint8_t reserved[2]; uint8_t attr_desc[0]; }; /* * Used with Sense keys No Sense (0x00) and Not Ready (0x02). * * Maximum descriptors allowed: 32 (as of SPC-4) */ struct scsi_sense_progress { uint8_t desc_type; #define SSD_DESC_PROGRESS 0x0a uint8_t length; uint8_t sense_key; uint8_t add_sense_code; uint8_t add_sense_code_qual; uint8_t reserved; uint8_t progress[2]; }; /* * This is typically forwarded as the result of an EXTENDED COPY command. * * Maximum descriptors allowed: 2 (as of SPC-4) */ struct scsi_sense_forwarded { uint8_t desc_type; #define SSD_DESC_FORWARDED 0x0c uint8_t length; uint8_t byte2; #define SSD_FORWARDED_FSDT 0x80 #define SSD_FORWARDED_SDS_MASK 0x0f #define SSD_FORWARDED_SDS_UNK 0x00 #define SSD_FORWARDED_SDS_EXSRC 0x01 #define SSD_FORWARDED_SDS_EXDST 0x02 }; /* * Vendor-specific sense descriptor. The desc_type field will be in the * range bewteen MIN and MAX inclusive. */ struct scsi_sense_vendor { uint8_t desc_type; #define SSD_DESC_VENDOR_MIN 0x80 #define SSD_DESC_VENDOR_MAX 0xff uint8_t length; uint8_t data[0]; }; struct scsi_mode_header_6 { u_int8_t data_length; /* Sense data length */ u_int8_t medium_type; u_int8_t dev_spec; u_int8_t blk_desc_len; }; struct scsi_mode_header_10 { u_int8_t data_length[2];/* Sense data length */ u_int8_t medium_type; u_int8_t dev_spec; u_int8_t unused[2]; u_int8_t blk_desc_len[2]; }; struct scsi_mode_page_header { u_int8_t page_code; #define SMPH_PS 0x80 #define SMPH_SPF 0x40 #define SMPH_PC_MASK 0x3f u_int8_t page_length; }; struct scsi_mode_page_header_sp { uint8_t page_code; uint8_t subpage; uint8_t page_length[2]; }; struct scsi_mode_blk_desc { u_int8_t density; u_int8_t nblocks[3]; u_int8_t reserved; u_int8_t blklen[3]; }; #define SCSI_DEFAULT_DENSITY 0x00 /* use 'default' density */ #define SCSI_SAME_DENSITY 0x7f /* use 'same' density- >= SCSI-2 only */ /* * Status Byte */ #define SCSI_STATUS_OK 0x00 #define SCSI_STATUS_CHECK_COND 0x02 #define SCSI_STATUS_COND_MET 0x04 #define SCSI_STATUS_BUSY 0x08 #define SCSI_STATUS_INTERMED 0x10 #define SCSI_STATUS_INTERMED_COND_MET 0x14 #define SCSI_STATUS_RESERV_CONFLICT 0x18 #define SCSI_STATUS_CMD_TERMINATED 0x22 /* Obsolete in SAM-2 */ #define SCSI_STATUS_QUEUE_FULL 0x28 #define SCSI_STATUS_ACA_ACTIVE 0x30 #define SCSI_STATUS_TASK_ABORTED 0x40 struct scsi_inquiry_pattern { u_int8_t type; u_int8_t media_type; #define SIP_MEDIA_REMOVABLE 0x01 #define SIP_MEDIA_FIXED 0x02 const char *vendor; const char *product; const char *revision; }; struct scsi_static_inquiry_pattern { u_int8_t type; u_int8_t media_type; char vendor[SID_VENDOR_SIZE+1]; char product[SID_PRODUCT_SIZE+1]; char revision[SID_REVISION_SIZE+1]; }; struct scsi_sense_quirk_entry { struct scsi_inquiry_pattern inq_pat; int num_sense_keys; int num_ascs; struct sense_key_table_entry *sense_key_info; struct asc_table_entry *asc_info; }; struct sense_key_table_entry { u_int8_t sense_key; u_int32_t action; const char *desc; }; struct asc_table_entry { u_int8_t asc; u_int8_t ascq; u_int32_t action; const char *desc; }; struct op_table_entry { u_int8_t opcode; u_int32_t opmask; const char *desc; }; struct scsi_op_quirk_entry { struct scsi_inquiry_pattern inq_pat; int num_ops; struct op_table_entry *op_table; }; typedef enum { SSS_FLAG_NONE = 0x00, SSS_FLAG_PRINT_COMMAND = 0x01 } scsi_sense_string_flags; struct scsi_nv { const char *name; uint64_t value; }; typedef enum { SCSI_NV_FOUND, SCSI_NV_AMBIGUOUS, SCSI_NV_NOT_FOUND } scsi_nv_status; typedef enum { SCSI_NV_FLAG_NONE = 0x00, SCSI_NV_FLAG_IG_CASE = 0x01 /* Case insensitive comparison */ } scsi_nv_flags; struct ccb_scsiio; struct cam_periph; union ccb; #ifndef _KERNEL struct cam_device; #endif extern const char *scsi_sense_key_text[]; struct sbuf; __BEGIN_DECLS void scsi_sense_desc(int sense_key, int asc, int ascq, struct scsi_inquiry_data *inq_data, const char **sense_key_desc, const char **asc_desc); scsi_sense_action scsi_error_action(struct ccb_scsiio* csio, struct scsi_inquiry_data *inq_data, u_int32_t sense_flags); const char * scsi_status_string(struct ccb_scsiio *csio); void scsi_desc_iterate(struct scsi_sense_data_desc *sense, u_int sense_len, int (*iter_func)(struct scsi_sense_data_desc *sense, u_int, struct scsi_sense_desc_header *, void *), void *arg); uint8_t *scsi_find_desc(struct scsi_sense_data_desc *sense, u_int sense_len, uint8_t desc_type); void scsi_set_sense_data(struct scsi_sense_data *sense_data, scsi_sense_data_type sense_format, int current_error, int sense_key, int asc, int ascq, ...) ; void scsi_set_sense_data_va(struct scsi_sense_data *sense_data, scsi_sense_data_type sense_format, int current_error, int sense_key, int asc, int ascq, va_list ap); int scsi_get_sense_info(struct scsi_sense_data *sense_data, u_int sense_len, uint8_t info_type, uint64_t *info, int64_t *signed_info); int scsi_get_sks(struct scsi_sense_data *sense_data, u_int sense_len, uint8_t *sks); int scsi_get_block_info(struct scsi_sense_data *sense_data, u_int sense_len, struct scsi_inquiry_data *inq_data, uint8_t *block_bits); int scsi_get_stream_info(struct scsi_sense_data *sense_data, u_int sense_len, struct scsi_inquiry_data *inq_data, uint8_t *stream_bits); void scsi_info_sbuf(struct sbuf *sb, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, uint64_t info); void scsi_command_sbuf(struct sbuf *sb, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, uint64_t csi); void scsi_progress_sbuf(struct sbuf *sb, uint16_t progress); int scsi_sks_sbuf(struct sbuf *sb, int sense_key, uint8_t *sks); void scsi_fru_sbuf(struct sbuf *sb, uint64_t fru); void scsi_stream_sbuf(struct sbuf *sb, uint8_t stream_bits, uint64_t info); void scsi_block_sbuf(struct sbuf *sb, uint8_t block_bits, uint64_t info); void scsi_sense_info_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_command_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_sks_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_fru_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_stream_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_block_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_progress_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_generic_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); void scsi_sense_desc_sbuf(struct sbuf *sb, struct scsi_sense_data *sense, u_int sense_len, uint8_t *cdb, int cdb_len, struct scsi_inquiry_data *inq_data, struct scsi_sense_desc_header *header); scsi_sense_data_type scsi_sense_type(struct scsi_sense_data *sense_data); void scsi_sense_only_sbuf(struct scsi_sense_data *sense, u_int sense_len, struct sbuf *sb, char *path_str, struct scsi_inquiry_data *inq_data, uint8_t *cdb, int cdb_len); #ifdef _KERNEL int scsi_command_string(struct ccb_scsiio *csio, struct sbuf *sb); int scsi_sense_sbuf(struct ccb_scsiio *csio, struct sbuf *sb, scsi_sense_string_flags flags); char * scsi_sense_string(struct ccb_scsiio *csio, char *str, int str_len); void scsi_sense_print(struct ccb_scsiio *csio); int scsi_vpd_supported_page(struct cam_periph *periph, uint8_t page_id); #else /* _KERNEL */ int scsi_command_string(struct cam_device *device, struct ccb_scsiio *csio, struct sbuf *sb); int scsi_sense_sbuf(struct cam_device *device, struct ccb_scsiio *csio, struct sbuf *sb, scsi_sense_string_flags flags); char * scsi_sense_string(struct cam_device *device, struct ccb_scsiio *csio, char *str, int str_len); void scsi_sense_print(struct cam_device *device, struct ccb_scsiio *csio, FILE *ofile); #endif /* _KERNEL */ const char * scsi_op_desc(u_int16_t opcode, struct scsi_inquiry_data *inq_data); char * scsi_cdb_string(u_int8_t *cdb_ptr, char *cdb_string, size_t len); void scsi_print_inquiry(struct scsi_inquiry_data *inq_data); void scsi_print_inquiry_short(struct scsi_inquiry_data *inq_data); u_int scsi_calc_syncsrate(u_int period_factor); u_int scsi_calc_syncparam(u_int period); typedef int (*scsi_devid_checkfn_t)(uint8_t *); int scsi_devid_is_naa_ieee_reg(uint8_t *bufp); int scsi_devid_is_sas_target(uint8_t *bufp); int scsi_devid_is_lun_eui64(uint8_t *bufp); int scsi_devid_is_lun_naa(uint8_t *bufp); int scsi_devid_is_lun_name(uint8_t *bufp); int scsi_devid_is_lun_t10(uint8_t *bufp); struct scsi_vpd_id_descriptor * scsi_get_devid(struct scsi_vpd_device_id *id, uint32_t len, scsi_devid_checkfn_t ck_fn); struct scsi_vpd_id_descriptor * scsi_get_devid_desc(struct scsi_vpd_id_descriptor *desc, uint32_t len, scsi_devid_checkfn_t ck_fn); int scsi_transportid_sbuf(struct sbuf *sb, struct scsi_transportid_header *hdr, uint32_t valid_len); const char * scsi_nv_to_str(struct scsi_nv *table, int num_table_entries, uint64_t value); scsi_nv_status scsi_get_nv(struct scsi_nv *table, int num_table_entries, char *name, int *table_entry, scsi_nv_flags flags); int scsi_parse_transportid_64bit(int proto_id, char *id_str, struct scsi_transportid_header **hdr, unsigned int *alloc_len, #ifdef _KERNEL struct malloc_type *type, int flags, #endif char *error_str, int error_str_len); int scsi_parse_transportid_spi(char *id_str, struct scsi_transportid_header **hdr, unsigned int *alloc_len, #ifdef _KERNEL struct malloc_type *type, int flags, #endif char *error_str, int error_str_len); int scsi_parse_transportid_rdma(char *id_str, struct scsi_transportid_header **hdr, unsigned int *alloc_len, #ifdef _KERNEL struct malloc_type *type, int flags, #endif char *error_str, int error_str_len); int scsi_parse_transportid_iscsi(char *id_str, struct scsi_transportid_header **hdr, unsigned int *alloc_len, #ifdef _KERNEL struct malloc_type *type, int flags, #endif char *error_str,int error_str_len); int scsi_parse_transportid_sop(char *id_str, struct scsi_transportid_header **hdr, unsigned int *alloc_len, #ifdef _KERNEL struct malloc_type *type, int flags, #endif char *error_str,int error_str_len); int scsi_parse_transportid(char *transportid_str, struct scsi_transportid_header **hdr, unsigned int *alloc_len, #ifdef _KERNEL struct malloc_type *type, int flags, #endif char *error_str, int error_str_len); void scsi_test_unit_ready(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t sense_len, u_int32_t timeout); void scsi_request_sense(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), void *data_ptr, u_int8_t dxfer_len, u_int8_t tag_action, u_int8_t sense_len, u_int32_t timeout); void scsi_inquiry(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t *inq_buf, u_int32_t inq_len, int evpd, u_int8_t page_code, u_int8_t sense_len, u_int32_t timeout); void scsi_mode_sense(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, int dbd, u_int8_t page_code, u_int8_t page, u_int8_t *param_buf, u_int32_t param_len, u_int8_t sense_len, u_int32_t timeout); void scsi_mode_sense_len(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, int dbd, u_int8_t page_code, u_int8_t page, u_int8_t *param_buf, u_int32_t param_len, int minimum_cmd_size, u_int8_t sense_len, u_int32_t timeout); void scsi_mode_select(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, int scsi_page_fmt, int save_pages, u_int8_t *param_buf, u_int32_t param_len, u_int8_t sense_len, u_int32_t timeout); void scsi_mode_select_len(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, int scsi_page_fmt, int save_pages, u_int8_t *param_buf, u_int32_t param_len, int minimum_cmd_size, u_int8_t sense_len, u_int32_t timeout); void scsi_log_sense(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t page_code, u_int8_t page, int save_pages, int ppc, u_int32_t paramptr, u_int8_t *param_buf, u_int32_t param_len, u_int8_t sense_len, u_int32_t timeout); void scsi_log_select(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t page_code, int save_pages, int pc_reset, u_int8_t *param_buf, u_int32_t param_len, u_int8_t sense_len, u_int32_t timeout); void scsi_prevent(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t action, u_int8_t sense_len, u_int32_t timeout); void scsi_read_capacity(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, struct scsi_read_capacity_data *, u_int8_t sense_len, u_int32_t timeout); void scsi_read_capacity_16(struct ccb_scsiio *csio, uint32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), uint8_t tag_action, uint64_t lba, int reladr, int pmi, uint8_t *rcap_buf, int rcap_buf_len, uint8_t sense_len, uint32_t timeout); void scsi_report_luns(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t select_report, struct scsi_report_luns_data *rpl_buf, u_int32_t alloc_len, u_int8_t sense_len, u_int32_t timeout); void scsi_report_target_group(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t pdf, void *buf, u_int32_t alloc_len, u_int8_t sense_len, u_int32_t timeout); void scsi_set_target_group(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, void *buf, u_int32_t alloc_len, u_int8_t sense_len, u_int32_t timeout); void scsi_synchronize_cache(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int32_t begin_lba, u_int16_t lb_count, u_int8_t sense_len, u_int32_t timeout); void scsi_receive_diagnostic_results(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb*), uint8_t tag_action, int pcv, uint8_t page_code, uint8_t *data_ptr, uint16_t allocation_length, uint8_t sense_len, uint32_t timeout); void scsi_send_diagnostic(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), uint8_t tag_action, int unit_offline, int device_offline, int self_test, int page_format, int self_test_code, uint8_t *data_ptr, uint16_t param_list_length, uint8_t sense_len, uint32_t timeout); void scsi_read_buffer(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb*), uint8_t tag_action, int mode, uint8_t buffer_id, u_int32_t offset, uint8_t *data_ptr, uint32_t allocation_length, uint8_t sense_len, uint32_t timeout); void scsi_write_buffer(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), uint8_t tag_action, int mode, uint8_t buffer_id, u_int32_t offset, uint8_t *data_ptr, uint32_t param_list_length, uint8_t sense_len, uint32_t timeout); #define SCSI_RW_READ 0x0001 #define SCSI_RW_WRITE 0x0002 #define SCSI_RW_DIRMASK 0x0003 #define SCSI_RW_BIO 0x1000 void scsi_read_write(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, int readop, u_int8_t byte2, int minimum_cmd_size, u_int64_t lba, u_int32_t block_count, u_int8_t *data_ptr, u_int32_t dxfer_len, u_int8_t sense_len, u_int32_t timeout); void scsi_write_same(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t byte2, int minimum_cmd_size, u_int64_t lba, u_int32_t block_count, u_int8_t *data_ptr, u_int32_t dxfer_len, u_int8_t sense_len, u_int32_t timeout); void scsi_ata_identify(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t *data_ptr, u_int16_t dxfer_len, u_int8_t sense_len, u_int32_t timeout); void scsi_ata_trim(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int16_t block_count, u_int8_t *data_ptr, u_int16_t dxfer_len, u_int8_t sense_len, u_int32_t timeout); void scsi_ata_pass_16(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int32_t flags, u_int8_t tag_action, u_int8_t protocol, u_int8_t ata_flags, u_int16_t features, u_int16_t sector_count, uint64_t lba, u_int8_t command, u_int8_t control, u_int8_t *data_ptr, u_int16_t dxfer_len, u_int8_t sense_len, u_int32_t timeout); void scsi_unmap(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, u_int8_t byte2, u_int8_t *data_ptr, u_int16_t dxfer_len, u_int8_t sense_len, u_int32_t timeout); void scsi_start_stop(struct ccb_scsiio *csio, u_int32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), u_int8_t tag_action, int start, int load_eject, int immediate, u_int8_t sense_len, u_int32_t timeout); void scsi_persistent_reserve_in(struct ccb_scsiio *csio, uint32_t retries, void (*cbfcnp)(struct cam_periph *,union ccb *), uint8_t tag_action, int service_action, uint8_t *data_ptr, uint32_t dxfer_len, int sense_len, int timeout); void scsi_persistent_reserve_out(struct ccb_scsiio *csio, uint32_t retries, void (*cbfcnp)(struct cam_periph *, union ccb *), uint8_t tag_action, int service_action, int scope, int res_type, uint8_t *data_ptr, uint32_t dxfer_len, int sense_len, int timeout); int scsi_inquiry_match(caddr_t inqbuffer, caddr_t table_entry); int scsi_static_inquiry_match(caddr_t inqbuffer, caddr_t table_entry); int scsi_devid_match(uint8_t *rhs, size_t rhs_len, uint8_t *lhs, size_t lhs_len); void scsi_extract_sense(struct scsi_sense_data *sense, int *error_code, int *sense_key, int *asc, int *ascq); int scsi_extract_sense_ccb(union ccb *ccb, int *error_code, int *sense_key, int *asc, int *ascq); void scsi_extract_sense_len(struct scsi_sense_data *sense, u_int sense_len, int *error_code, int *sense_key, int *asc, int *ascq, int show_errors); int scsi_get_sense_key(struct scsi_sense_data *sense, u_int sense_len, int show_errors); int scsi_get_asc(struct scsi_sense_data *sense, u_int sense_len, int show_errors); int scsi_get_ascq(struct scsi_sense_data *sense, u_int sense_len, int show_errors); static __inline void scsi_ulto2b(u_int32_t val, u_int8_t *bytes); static __inline void scsi_ulto3b(u_int32_t val, u_int8_t *bytes); static __inline void scsi_ulto4b(u_int32_t val, u_int8_t *bytes); static __inline void scsi_u64to8b(u_int64_t val, u_int8_t *bytes); static __inline uint32_t scsi_2btoul(const uint8_t *bytes); static __inline uint32_t scsi_3btoul(const uint8_t *bytes); static __inline int32_t scsi_3btol(const uint8_t *bytes); static __inline uint32_t scsi_4btoul(const uint8_t *bytes); static __inline uint64_t scsi_8btou64(const uint8_t *bytes); static __inline void *find_mode_page_6(struct scsi_mode_header_6 *mode_header); static __inline void *find_mode_page_10(struct scsi_mode_header_10 *mode_header); static __inline void scsi_ulto2b(u_int32_t val, u_int8_t *bytes) { bytes[0] = (val >> 8) & 0xff; bytes[1] = val & 0xff; } static __inline void scsi_ulto3b(u_int32_t val, u_int8_t *bytes) { bytes[0] = (val >> 16) & 0xff; bytes[1] = (val >> 8) & 0xff; bytes[2] = val & 0xff; } static __inline void scsi_ulto4b(u_int32_t val, u_int8_t *bytes) { bytes[0] = (val >> 24) & 0xff; bytes[1] = (val >> 16) & 0xff; bytes[2] = (val >> 8) & 0xff; bytes[3] = val & 0xff; } static __inline void scsi_u64to8b(u_int64_t val, u_int8_t *bytes) { bytes[0] = (val >> 56) & 0xff; bytes[1] = (val >> 48) & 0xff; bytes[2] = (val >> 40) & 0xff; bytes[3] = (val >> 32) & 0xff; bytes[4] = (val >> 24) & 0xff; bytes[5] = (val >> 16) & 0xff; bytes[6] = (val >> 8) & 0xff; bytes[7] = val & 0xff; } static __inline uint32_t scsi_2btoul(const uint8_t *bytes) { uint32_t rv; rv = (bytes[0] << 8) | bytes[1]; return (rv); } static __inline uint32_t scsi_3btoul(const uint8_t *bytes) { uint32_t rv; rv = (bytes[0] << 16) | (bytes[1] << 8) | bytes[2]; return (rv); } static __inline int32_t scsi_3btol(const uint8_t *bytes) { uint32_t rc = scsi_3btoul(bytes); if (rc & 0x00800000) rc |= 0xff000000; return (int32_t) rc; } static __inline uint32_t scsi_4btoul(const uint8_t *bytes) { uint32_t rv; rv = (bytes[0] << 24) | (bytes[1] << 16) | (bytes[2] << 8) | bytes[3]; return (rv); } static __inline uint64_t scsi_8btou64(const uint8_t *bytes) { uint64_t rv; rv = (((uint64_t)bytes[0]) << 56) | (((uint64_t)bytes[1]) << 48) | (((uint64_t)bytes[2]) << 40) | (((uint64_t)bytes[3]) << 32) | (((uint64_t)bytes[4]) << 24) | (((uint64_t)bytes[5]) << 16) | (((uint64_t)bytes[6]) << 8) | bytes[7]; return (rv); } /* * Given the pointer to a returned mode sense buffer, return a pointer to * the start of the first mode page. */ static __inline void * find_mode_page_6(struct scsi_mode_header_6 *mode_header) { void *page_start; page_start = (void *)((u_int8_t *)&mode_header[1] + mode_header->blk_desc_len); return(page_start); } static __inline void * find_mode_page_10(struct scsi_mode_header_10 *mode_header) { void *page_start; page_start = (void *)((u_int8_t *)&mode_header[1] + scsi_2btoul(mode_header->blk_desc_len)); return(page_start); } __END_DECLS #endif /*_SCSI_SCSI_ALL_H*/ Index: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c =================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c (revision 274153) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c (revision 274154) @@ -1,2869 +1,2921 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * * Copyright (c) 2006-2010 Pawel Jakub Dawidek * All rights reserved. * * Portions Copyright 2010 Robert Milkowski * * Copyright 2011 Nexenta Systems, Inc. All rights reserved. * Copyright (c) 2012, 2014 by Delphix. All rights reserved. * Copyright (c) 2013, Joyent, Inc. All rights reserved. */ /* Portions Copyright 2011 Martin Matuska */ /* * ZFS volume emulation driver. * * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. * Volumes are accessed through the symbolic links named: * * /dev/zvol/dsk// * /dev/zvol/rdsk// * * These links are created by the /dev filesystem (sdev_zvolops.c). * Volumes are persistent through reboot. No user command needs to be * run before opening and using a device. * * FreeBSD notes. * On FreeBSD ZVOLs are simply GEOM providers like any other storage device * in the system. */ #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 "zfs_namecheck.h" struct g_class zfs_zvol_class = { .name = "ZFS::ZVOL", .version = G_VERSION, }; DECLARE_GEOM_CLASS(zfs_zvol_class, zfs_zvol); void *zfsdev_state; static char *zvol_tag = "zvol_tag"; #define ZVOL_DUMPSIZE "dumpsize" /* * The spa_namespace_lock protects the zfsdev_state structure from being * modified while it's being used, e.g. an open that comes in before a * create finishes. It also protects temporary opens of the dataset so that, * e.g., an open doesn't get a spurious EBUSY. */ static uint32_t zvol_minors; SYSCTL_DECL(_vfs_zfs); SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME"); static int volmode = ZFS_VOLMODE_GEOM; SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0, "Expose as GEOM providers (1), device files (2) or neither"); typedef struct zvol_extent { list_node_t ze_node; dva_t ze_dva; /* dva associated with this extent */ uint64_t ze_nblks; /* number of blocks in extent */ } zvol_extent_t; /* * The in-core state of each volume. */ typedef struct zvol_state { LIST_ENTRY(zvol_state) zv_links; char zv_name[MAXPATHLEN]; /* pool/dd name */ uint64_t zv_volsize; /* amount of space we advertise */ uint64_t zv_volblocksize; /* volume block size */ struct cdev *zv_dev; /* non-GEOM device */ struct g_provider *zv_provider; /* GEOM provider */ uint8_t zv_min_bs; /* minimum addressable block shift */ uint8_t zv_flags; /* readonly, dumpified, etc. */ objset_t *zv_objset; /* objset handle */ uint32_t zv_total_opens; /* total open count */ zilog_t *zv_zilog; /* ZIL handle */ list_t zv_extents; /* List of extents for dump */ znode_t zv_znode; /* for range locking */ dmu_buf_t *zv_dbuf; /* bonus handle */ int zv_state; int zv_volmode; /* Provide GEOM or cdev */ struct bio_queue_head zv_queue; struct mtx zv_queue_mtx; /* zv_queue mutex */ } zvol_state_t; static LIST_HEAD(, zvol_state) all_zvols; /* * zvol specific flags */ #define ZVOL_RDONLY 0x1 #define ZVOL_DUMPIFIED 0x2 #define ZVOL_EXCL 0x4 #define ZVOL_WCE 0x8 /* * zvol maximum transfer in one DMU tx. */ int zvol_maxphys = DMU_MAX_ACCESS/2; /* * Toggle unmap functionality. */ boolean_t zvol_unmap_enabled = B_TRUE; SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, unmap_enabled, CTLFLAG_RWTUN, &zvol_unmap_enabled, 0, "Enable UNMAP functionality"); static d_open_t zvol_d_open; static d_close_t zvol_d_close; static d_read_t zvol_read; static d_write_t zvol_write; static d_ioctl_t zvol_d_ioctl; static d_strategy_t zvol_strategy; static struct cdevsw zvol_cdevsw = { .d_version = D_VERSION, .d_open = zvol_d_open, .d_close = zvol_d_close, .d_read = zvol_read, .d_write = zvol_write, .d_ioctl = zvol_d_ioctl, .d_strategy = zvol_strategy, .d_name = "zvol", .d_flags = D_DISK | D_TRACKCLOSE, }; extern int zfs_set_prop_nvlist(const char *, zprop_source_t, nvlist_t *, nvlist_t *); static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, boolean_t sync); static int zvol_remove_zv(zvol_state_t *); static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio); static int zvol_dumpify(zvol_state_t *zv); static int zvol_dump_fini(zvol_state_t *zv); static int zvol_dump_init(zvol_state_t *zv, boolean_t resize); static void zvol_geom_run(zvol_state_t *zv); static void zvol_geom_destroy(zvol_state_t *zv); static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace); static void zvol_geom_start(struct bio *bp); static void zvol_geom_worker(void *arg); static void zvol_size_changed(zvol_state_t *zv) { #ifdef sun dev_t dev = makedevice(maj, min); VERIFY(ddi_prop_update_int64(dev, zfs_dip, "Size", volsize) == DDI_SUCCESS); VERIFY(ddi_prop_update_int64(dev, zfs_dip, "Nblocks", lbtodb(volsize)) == DDI_SUCCESS); /* Notify specfs to invalidate the cached size */ spec_size_invalidate(dev, VBLK); spec_size_invalidate(dev, VCHR); #else /* !sun */ if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { struct g_provider *pp; pp = zv->zv_provider; if (pp == NULL) return; g_topology_lock(); g_resize_provider(pp, zv->zv_volsize); g_topology_unlock(); } #endif /* !sun */ } int zvol_check_volsize(uint64_t volsize, uint64_t blocksize) { if (volsize == 0) return (SET_ERROR(EINVAL)); if (volsize % blocksize != 0) return (SET_ERROR(EINVAL)); #ifdef _ILP32 if (volsize - 1 > SPEC_MAXOFFSET_T) return (SET_ERROR(EOVERFLOW)); #endif return (0); } int zvol_check_volblocksize(uint64_t volblocksize) { if (volblocksize < SPA_MINBLOCKSIZE || volblocksize > SPA_MAXBLOCKSIZE || !ISP2(volblocksize)) return (SET_ERROR(EDOM)); return (0); } int zvol_get_stats(objset_t *os, nvlist_t *nv) { int error; dmu_object_info_t doi; uint64_t val; error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); if (error) return (error); dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); error = dmu_object_info(os, ZVOL_OBJ, &doi); if (error == 0) { dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, doi.doi_data_block_size); } return (error); } static zvol_state_t * zvol_minor_lookup(const char *name) { zvol_state_t *zv; ASSERT(MUTEX_HELD(&spa_namespace_lock)); LIST_FOREACH(zv, &all_zvols, zv_links) { if (strcmp(zv->zv_name, name) == 0) break; } return (zv); } /* extent mapping arg */ struct maparg { zvol_state_t *ma_zv; uint64_t ma_blks; }; /*ARGSUSED*/ static int zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) { struct maparg *ma = arg; zvol_extent_t *ze; int bs = ma->ma_zv->zv_volblocksize; if (BP_IS_HOLE(bp) || zb->zb_object != ZVOL_OBJ || zb->zb_level != 0) return (0); VERIFY(!BP_IS_EMBEDDED(bp)); VERIFY3U(ma->ma_blks, ==, zb->zb_blkid); ma->ma_blks++; /* Abort immediately if we have encountered gang blocks */ if (BP_IS_GANG(bp)) return (SET_ERROR(EFRAGS)); /* * See if the block is at the end of the previous extent. */ ze = list_tail(&ma->ma_zv->zv_extents); if (ze && DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) && DVA_GET_OFFSET(BP_IDENTITY(bp)) == DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) { ze->ze_nblks++; return (0); } dprintf_bp(bp, "%s", "next blkptr:"); /* start a new extent */ ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP); ze->ze_dva = bp->blk_dva[0]; /* structure assignment */ ze->ze_nblks = 1; list_insert_tail(&ma->ma_zv->zv_extents, ze); return (0); } static void zvol_free_extents(zvol_state_t *zv) { zvol_extent_t *ze; while (ze = list_head(&zv->zv_extents)) { list_remove(&zv->zv_extents, ze); kmem_free(ze, sizeof (zvol_extent_t)); } } static int zvol_get_lbas(zvol_state_t *zv) { objset_t *os = zv->zv_objset; struct maparg ma; int err; ma.ma_zv = zv; ma.ma_blks = 0; zvol_free_extents(zv); /* commit any in-flight changes before traversing the dataset */ txg_wait_synced(dmu_objset_pool(os), 0); err = traverse_dataset(dmu_objset_ds(os), 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma); if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) { zvol_free_extents(zv); return (err ? err : EIO); } return (0); } /* ARGSUSED */ void zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) { zfs_creat_t *zct = arg; nvlist_t *nvprops = zct->zct_props; int error; uint64_t volblocksize, volsize; VERIFY(nvlist_lookup_uint64(nvprops, zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); if (nvlist_lookup_uint64(nvprops, zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); /* * These properties must be removed from the list so the generic * property setting step won't apply to them. */ VERIFY(nvlist_remove_all(nvprops, zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); (void) nvlist_remove_all(nvprops, zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, DMU_OT_NONE, 0, tx); ASSERT(error == 0); error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, DMU_OT_NONE, 0, tx); ASSERT(error == 0); error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); ASSERT(error == 0); } /* * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we * implement DKIOCFREE/free-long-range. */ static int zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) { uint64_t offset, length; if (byteswap) byteswap_uint64_array(lr, sizeof (*lr)); offset = lr->lr_offset; length = lr->lr_length; return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); } /* * Replay a TX_WRITE ZIL transaction that didn't get committed * after a system failure */ static int zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) { objset_t *os = zv->zv_objset; char *data = (char *)(lr + 1); /* data follows lr_write_t */ uint64_t offset, length; dmu_tx_t *tx; int error; if (byteswap) byteswap_uint64_array(lr, sizeof (*lr)); offset = lr->lr_offset; length = lr->lr_length; /* If it's a dmu_sync() block, write the whole block */ if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) { uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr); if (length < blocksize) { offset -= offset % blocksize; length = blocksize; } } tx = dmu_tx_create(os); dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); } else { dmu_write(os, ZVOL_OBJ, offset, length, data, tx); dmu_tx_commit(tx); } return (error); } /* ARGSUSED */ static int zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) { return (SET_ERROR(ENOTSUP)); } /* * Callback vectors for replaying records. * Only TX_WRITE and TX_TRUNCATE are needed for zvol. */ zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = { zvol_replay_err, /* 0 no such transaction type */ zvol_replay_err, /* TX_CREATE */ zvol_replay_err, /* TX_MKDIR */ zvol_replay_err, /* TX_MKXATTR */ zvol_replay_err, /* TX_SYMLINK */ zvol_replay_err, /* TX_REMOVE */ zvol_replay_err, /* TX_RMDIR */ zvol_replay_err, /* TX_LINK */ zvol_replay_err, /* TX_RENAME */ zvol_replay_write, /* TX_WRITE */ zvol_replay_truncate, /* TX_TRUNCATE */ zvol_replay_err, /* TX_SETATTR */ zvol_replay_err, /* TX_ACL */ zvol_replay_err, /* TX_CREATE_ACL */ zvol_replay_err, /* TX_CREATE_ATTR */ zvol_replay_err, /* TX_CREATE_ACL_ATTR */ zvol_replay_err, /* TX_MKDIR_ACL */ zvol_replay_err, /* TX_MKDIR_ATTR */ zvol_replay_err, /* TX_MKDIR_ACL_ATTR */ zvol_replay_err, /* TX_WRITE2 */ }; #ifdef sun int zvol_name2minor(const char *name, minor_t *minor) { zvol_state_t *zv; mutex_enter(&spa_namespace_lock); zv = zvol_minor_lookup(name); if (minor && zv) *minor = zv->zv_minor; mutex_exit(&spa_namespace_lock); return (zv ? 0 : -1); } #endif /* sun */ /* * Create a minor node (plus a whole lot more) for the specified volume. */ int zvol_create_minor(const char *name) { zfs_soft_state_t *zs; zvol_state_t *zv; objset_t *os; struct cdev *dev; struct g_provider *pp; struct g_geom *gp; dmu_object_info_t doi; uint64_t volsize, mode; int error; ZFS_LOG(1, "Creating ZVOL %s...", name); mutex_enter(&spa_namespace_lock); if (zvol_minor_lookup(name) != NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(EEXIST)); } /* lie and say we're read-only */ error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os); if (error) { mutex_exit(&spa_namespace_lock); return (error); } #ifdef sun if ((minor = zfsdev_minor_alloc()) == 0) { dmu_objset_disown(os, FTAG); mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENXIO)); } if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) { dmu_objset_disown(os, FTAG); mutex_exit(&spa_namespace_lock); return (SET_ERROR(EAGAIN)); } (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME, (char *)name); (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor); if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR, minor, DDI_PSEUDO, 0) == DDI_FAILURE) { ddi_soft_state_free(zfsdev_state, minor); dmu_objset_disown(os, FTAG); mutex_exit(&spa_namespace_lock); return (SET_ERROR(EAGAIN)); } (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor); if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK, minor, DDI_PSEUDO, 0) == DDI_FAILURE) { ddi_remove_minor_node(zfs_dip, chrbuf); ddi_soft_state_free(zfsdev_state, minor); dmu_objset_disown(os, FTAG); mutex_exit(&spa_namespace_lock); return (SET_ERROR(EAGAIN)); } zs = ddi_get_soft_state(zfsdev_state, minor); zs->zss_type = ZSST_ZVOL; zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); #else /* !sun */ zv = kmem_zalloc(sizeof(*zv), KM_SLEEP); zv->zv_state = 0; error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); if (error) { kmem_free(zv, sizeof(*zv)); dmu_objset_disown(os, zvol_tag); mutex_exit(&spa_namespace_lock); return (error); } error = dsl_prop_get_integer(name, zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL); if (error != 0 || mode == ZFS_VOLMODE_DEFAULT) mode = volmode; DROP_GIANT(); zv->zv_volsize = volsize; zv->zv_volmode = mode; if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { g_topology_lock(); gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name); gp->start = zvol_geom_start; gp->access = zvol_geom_access; pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name); pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; pp->sectorsize = DEV_BSIZE; pp->mediasize = zv->zv_volsize; pp->private = zv; zv->zv_provider = pp; bioq_init(&zv->zv_queue); mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF); } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s/%s", ZVOL_DRIVER, name) != 0) { kmem_free(zv, sizeof(*zv)); dmu_objset_disown(os, FTAG); mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENXIO)); } zv->zv_dev = dev; dev->si_iosize_max = MAXPHYS; dev->si_drv2 = zv; } LIST_INSERT_HEAD(&all_zvols, zv, zv_links); #endif /* !sun */ (void) strlcpy(zv->zv_name, name, MAXPATHLEN); zv->zv_min_bs = DEV_BSHIFT; zv->zv_objset = os; if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) zv->zv_flags |= ZVOL_RDONLY; mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL); avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare, sizeof (rl_t), offsetof(rl_t, r_node)); list_create(&zv->zv_extents, sizeof (zvol_extent_t), offsetof(zvol_extent_t, ze_node)); /* get and cache the blocksize */ error = dmu_object_info(os, ZVOL_OBJ, &doi); ASSERT(error == 0); zv->zv_volblocksize = doi.doi_data_block_size; if (spa_writeable(dmu_objset_spa(os))) { if (zil_replay_disable) zil_destroy(dmu_objset_zil(os), B_FALSE); else zil_replay(os, zv, zvol_replay_vector); } dmu_objset_disown(os, FTAG); zv->zv_objset = NULL; zvol_minors++; mutex_exit(&spa_namespace_lock); #ifndef sun if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { zvol_geom_run(zv); g_topology_unlock(); } PICKUP_GIANT(); #endif ZFS_LOG(1, "ZVOL %s created.", name); return (0); } /* * Remove minor node for the specified volume. */ static int zvol_remove_zv(zvol_state_t *zv) { #ifdef sun minor_t minor = zv->zv_minor; #endif ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (zv->zv_total_opens != 0) return (SET_ERROR(EBUSY)); ZFS_LOG(1, "ZVOL %s destroyed.", zv->zv_name); #ifdef sun (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor); ddi_remove_minor_node(zfs_dip, nmbuf); #else LIST_REMOVE(zv, zv_links); if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { g_topology_lock(); zvol_geom_destroy(zv); g_topology_unlock(); } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) destroy_dev(zv->zv_dev); #endif /* sun */ avl_destroy(&zv->zv_znode.z_range_avl); mutex_destroy(&zv->zv_znode.z_range_lock); kmem_free(zv, sizeof(*zv)); zvol_minors--; return (0); } int zvol_remove_minor(const char *name) { zvol_state_t *zv; int rc; mutex_enter(&spa_namespace_lock); if ((zv = zvol_minor_lookup(name)) == NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENXIO)); } rc = zvol_remove_zv(zv); mutex_exit(&spa_namespace_lock); return (rc); } int zvol_first_open(zvol_state_t *zv) { objset_t *os; uint64_t volsize; int error; uint64_t readonly; /* lie and say we're read-only */ error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os); if (error) return (error); error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); if (error) { ASSERT(error == 0); dmu_objset_disown(os, zvol_tag); return (error); } zv->zv_objset = os; error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); if (error) { dmu_objset_disown(os, zvol_tag); return (error); } zv->zv_volsize = volsize; zv->zv_zilog = zil_open(os, zvol_get_data); zvol_size_changed(zv); VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly, NULL) == 0); if (readonly || dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os))) zv->zv_flags |= ZVOL_RDONLY; else zv->zv_flags &= ~ZVOL_RDONLY; return (error); } void zvol_last_close(zvol_state_t *zv) { zil_close(zv->zv_zilog); zv->zv_zilog = NULL; dmu_buf_rele(zv->zv_dbuf, zvol_tag); zv->zv_dbuf = NULL; /* * Evict cached data */ if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && !(zv->zv_flags & ZVOL_RDONLY)) txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); dmu_objset_evict_dbufs(zv->zv_objset); dmu_objset_disown(zv->zv_objset, zvol_tag); zv->zv_objset = NULL; } #ifdef sun int zvol_prealloc(zvol_state_t *zv) { objset_t *os = zv->zv_objset; dmu_tx_t *tx; uint64_t refd, avail, usedobjs, availobjs; uint64_t resid = zv->zv_volsize; uint64_t off = 0; /* Check the space usage before attempting to allocate the space */ dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs); if (avail < zv->zv_volsize) return (SET_ERROR(ENOSPC)); /* Free old extents if they exist */ zvol_free_extents(zv); while (resid != 0) { int error; uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE); tx = dmu_tx_create(os); dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off); return (error); } dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx); dmu_tx_commit(tx); off += bytes; resid -= bytes; } txg_wait_synced(dmu_objset_pool(os), 0); return (0); } #endif /* sun */ static int zvol_update_volsize(objset_t *os, uint64_t volsize) { dmu_tx_t *tx; int error; ASSERT(MUTEX_HELD(&spa_namespace_lock)); tx = dmu_tx_create(os); dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); dmu_tx_mark_netfree(tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); return (error); } error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); dmu_tx_commit(tx); if (error == 0) error = dmu_free_long_range(os, ZVOL_OBJ, volsize, DMU_OBJECT_END); return (error); } void zvol_remove_minors(const char *name) { zvol_state_t *zv, *tzv; size_t namelen; namelen = strlen(name); DROP_GIANT(); mutex_enter(&spa_namespace_lock); LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) { if (strcmp(zv->zv_name, name) == 0 || (strncmp(zv->zv_name, name, namelen) == 0 && strlen(zv->zv_name) > namelen && (zv->zv_name[namelen] == '/' || zv->zv_name[namelen] == '@'))) { (void) zvol_remove_zv(zv); } } mutex_exit(&spa_namespace_lock); PICKUP_GIANT(); } int zvol_set_volsize(const char *name, major_t maj, uint64_t volsize) { zvol_state_t *zv = NULL; objset_t *os; int error; dmu_object_info_t doi; uint64_t old_volsize = 0ULL; uint64_t readonly; mutex_enter(&spa_namespace_lock); zv = zvol_minor_lookup(name); if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { mutex_exit(&spa_namespace_lock); return (error); } if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 || (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0) goto out; VERIFY(dsl_prop_get_integer(name, "readonly", &readonly, NULL) == 0); if (readonly) { error = EROFS; goto out; } error = zvol_update_volsize(os, volsize); /* * Reinitialize the dump area to the new size. If we * failed to resize the dump area then restore it back to * its original size. */ if (zv && error == 0) { #ifdef ZVOL_DUMP if (zv->zv_flags & ZVOL_DUMPIFIED) { old_volsize = zv->zv_volsize; zv->zv_volsize = volsize; if ((error = zvol_dumpify(zv)) != 0 || (error = dumpvp_resize()) != 0) { (void) zvol_update_volsize(os, old_volsize); zv->zv_volsize = old_volsize; error = zvol_dumpify(zv); } } #endif /* ZVOL_DUMP */ if (error == 0) { zv->zv_volsize = volsize; zvol_size_changed(zv); } } #ifdef sun /* * Generate a LUN expansion event. */ if (zv && error == 0) { sysevent_id_t eid; nvlist_t *attr; char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV, zv->zv_minor); VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, ESC_DEV_DLE, attr, &eid, DDI_SLEEP); nvlist_free(attr); kmem_free(physpath, MAXPATHLEN); } #endif /* sun */ out: dmu_objset_rele(os, FTAG); mutex_exit(&spa_namespace_lock); return (error); } /*ARGSUSED*/ static int zvol_open(struct g_provider *pp, int flag, int count) { zvol_state_t *zv; int err = 0; boolean_t locked = B_FALSE; /* * Protect against recursively entering spa_namespace_lock * when spa_open() is used for a pool on a (local) ZVOL(s). * This is needed since we replaced upstream zfsdev_state_lock * with spa_namespace_lock in the ZVOL code. * We are using the same trick as spa_open(). * Note that calls in zvol_first_open which need to resolve * pool name to a spa object will enter spa_open() * recursively, but that function already has all the * necessary protection. */ if (!MUTEX_HELD(&spa_namespace_lock)) { mutex_enter(&spa_namespace_lock); locked = B_TRUE; } zv = pp->private; if (zv == NULL) { if (locked) mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENXIO)); } if (zv->zv_total_opens == 0) { err = zvol_first_open(zv); if (err) { if (locked) mutex_exit(&spa_namespace_lock); return (err); } pp->mediasize = zv->zv_volsize; pp->stripeoffset = 0; pp->stripesize = zv->zv_volblocksize; } if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { err = SET_ERROR(EROFS); goto out; } if (zv->zv_flags & ZVOL_EXCL) { err = SET_ERROR(EBUSY); goto out; } #ifdef FEXCL if (flag & FEXCL) { if (zv->zv_total_opens != 0) { err = SET_ERROR(EBUSY); goto out; } zv->zv_flags |= ZVOL_EXCL; } #endif zv->zv_total_opens += count; if (locked) mutex_exit(&spa_namespace_lock); return (err); out: if (zv->zv_total_opens == 0) zvol_last_close(zv); if (locked) mutex_exit(&spa_namespace_lock); return (err); } /*ARGSUSED*/ static int zvol_close(struct g_provider *pp, int flag, int count) { zvol_state_t *zv; int error = 0; boolean_t locked = B_FALSE; /* See comment in zvol_open(). */ if (!MUTEX_HELD(&spa_namespace_lock)) { mutex_enter(&spa_namespace_lock); locked = B_TRUE; } zv = pp->private; if (zv == NULL) { if (locked) mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENXIO)); } if (zv->zv_flags & ZVOL_EXCL) { ASSERT(zv->zv_total_opens == 1); zv->zv_flags &= ~ZVOL_EXCL; } /* * If the open count is zero, this is a spurious close. * That indicates a bug in the kernel / DDI framework. */ ASSERT(zv->zv_total_opens != 0); /* * You may get multiple opens, but only one close. */ zv->zv_total_opens -= count; if (zv->zv_total_opens == 0) zvol_last_close(zv); if (locked) mutex_exit(&spa_namespace_lock); return (error); } static void zvol_get_done(zgd_t *zgd, int error) { if (zgd->zgd_db) dmu_buf_rele(zgd->zgd_db, zgd); zfs_range_unlock(zgd->zgd_rl); if (error == 0 && zgd->zgd_bp) zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); kmem_free(zgd, sizeof (zgd_t)); } /* * Get data to generate a TX_WRITE intent log record. */ static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) { zvol_state_t *zv = arg; objset_t *os = zv->zv_objset; uint64_t object = ZVOL_OBJ; uint64_t offset = lr->lr_offset; uint64_t size = lr->lr_length; /* length of user data */ blkptr_t *bp = &lr->lr_blkptr; dmu_buf_t *db; zgd_t *zgd; int error; ASSERT(zio != NULL); ASSERT(size != 0); zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP); zgd->zgd_zilog = zv->zv_zilog; zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER); /* * Write records come in two flavors: immediate and indirect. * For small writes it's cheaper to store the data with the * log record (immediate); for large writes it's cheaper to * sync the data and get a pointer to it (indirect) so that * we don't have to write the data twice. */ if (buf != NULL) { /* immediate write */ error = dmu_read(os, object, offset, size, buf, DMU_READ_NO_PREFETCH); } else { size = zv->zv_volblocksize; offset = P2ALIGN(offset, size); error = dmu_buf_hold(os, object, offset, zgd, &db, DMU_READ_NO_PREFETCH); if (error == 0) { blkptr_t *obp = dmu_buf_get_blkptr(db); if (obp) { ASSERT(BP_IS_HOLE(bp)); *bp = *obp; } zgd->zgd_db = db; zgd->zgd_bp = bp; ASSERT(db->db_offset == offset); ASSERT(db->db_size == size); error = dmu_sync(zio, lr->lr_common.lrc_txg, zvol_get_done, zgd); if (error == 0) return (0); } } zvol_get_done(zgd, error); return (error); } /* * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. * * We store data in the log buffers if it's small enough. * Otherwise we will later flush the data out via dmu_sync(). */ ssize_t zvol_immediate_write_sz = 32768; static void zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid, boolean_t sync) { uint32_t blocksize = zv->zv_volblocksize; zilog_t *zilog = zv->zv_zilog; boolean_t slogging; ssize_t immediate_write_sz; if (zil_replaying(zilog, tx)) return; immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) ? 0 : zvol_immediate_write_sz; slogging = spa_has_slogs(zilog->zl_spa) && (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); while (resid) { itx_t *itx; lr_write_t *lr; ssize_t len; itx_wr_state_t write_state; /* * Unlike zfs_log_write() we can be called with * upto DMU_MAX_ACCESS/2 (5MB) writes. */ if (blocksize > immediate_write_sz && !slogging && resid >= blocksize && off % blocksize == 0) { write_state = WR_INDIRECT; /* uses dmu_sync */ len = blocksize; } else if (sync) { write_state = WR_COPIED; len = MIN(ZIL_MAX_LOG_DATA, resid); } else { write_state = WR_NEED_COPY; len = MIN(ZIL_MAX_LOG_DATA, resid); } itx = zil_itx_create(TX_WRITE, sizeof (*lr) + (write_state == WR_COPIED ? len : 0)); lr = (lr_write_t *)&itx->itx_lr; if (write_state == WR_COPIED && dmu_read(zv->zv_objset, ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) { zil_itx_destroy(itx); itx = zil_itx_create(TX_WRITE, sizeof (*lr)); lr = (lr_write_t *)&itx->itx_lr; write_state = WR_NEED_COPY; } itx->itx_wr_state = write_state; if (write_state == WR_NEED_COPY) itx->itx_sod += len; lr->lr_foid = ZVOL_OBJ; lr->lr_offset = off; lr->lr_length = len; lr->lr_blkoff = 0; BP_ZERO(&lr->lr_blkptr); itx->itx_private = zv; itx->itx_sync = sync; zil_itx_assign(zilog, itx, tx); off += len; resid -= len; } } #ifdef sun static int zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset, uint64_t size, boolean_t doread, boolean_t isdump) { vdev_disk_t *dvd; int c; int numerrors = 0; if (vd->vdev_ops == &vdev_mirror_ops || vd->vdev_ops == &vdev_replacing_ops || vd->vdev_ops == &vdev_spare_ops) { for (c = 0; c < vd->vdev_children; c++) { int err = zvol_dumpio_vdev(vd->vdev_child[c], addr, offset, origoffset, size, doread, isdump); if (err != 0) { numerrors++; } else if (doread) { break; } } } if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops) return (numerrors < vd->vdev_children ? 0 : EIO); if (doread && !vdev_readable(vd)) return (SET_ERROR(EIO)); else if (!doread && !vdev_writeable(vd)) return (SET_ERROR(EIO)); if (vd->vdev_ops == &vdev_raidz_ops) { return (vdev_raidz_physio(vd, addr, size, offset, origoffset, doread, isdump)); } offset += VDEV_LABEL_START_SIZE; if (ddi_in_panic() || isdump) { ASSERT(!doread); if (doread) return (SET_ERROR(EIO)); dvd = vd->vdev_tsd; ASSERT3P(dvd, !=, NULL); return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset), lbtodb(size))); } else { dvd = vd->vdev_tsd; ASSERT3P(dvd, !=, NULL); return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size, offset, doread ? B_READ : B_WRITE)); } } static int zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size, boolean_t doread, boolean_t isdump) { vdev_t *vd; int error; zvol_extent_t *ze; spa_t *spa = dmu_objset_spa(zv->zv_objset); /* Must be sector aligned, and not stradle a block boundary. */ if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) || P2BOUNDARY(offset, size, zv->zv_volblocksize)) { return (SET_ERROR(EINVAL)); } ASSERT(size <= zv->zv_volblocksize); /* Locate the extent this belongs to */ ze = list_head(&zv->zv_extents); while (offset >= ze->ze_nblks * zv->zv_volblocksize) { offset -= ze->ze_nblks * zv->zv_volblocksize; ze = list_next(&zv->zv_extents, ze); } if (ze == NULL) return (SET_ERROR(EINVAL)); if (!ddi_in_panic()) spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva)); offset += DVA_GET_OFFSET(&ze->ze_dva); error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva), size, doread, isdump); if (!ddi_in_panic()) spa_config_exit(spa, SCL_STATE, FTAG); return (error); } #endif /* sun */ void zvol_strategy(struct bio *bp) { zvol_state_t *zv; uint64_t off, volsize; size_t resid; char *addr; objset_t *os; rl_t *rl; int error = 0; boolean_t doread = 0; boolean_t is_dumpified; boolean_t sync; if (bp->bio_to) zv = bp->bio_to->private; else zv = bp->bio_dev->si_drv2; if (zv == NULL) { error = ENXIO; goto out; } if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) { error = EROFS; goto out; } switch (bp->bio_cmd) { case BIO_FLUSH: goto sync; case BIO_READ: doread = 1; case BIO_WRITE: case BIO_DELETE: break; default: error = EOPNOTSUPP; goto out; } off = bp->bio_offset; volsize = zv->zv_volsize; os = zv->zv_objset; ASSERT(os != NULL); addr = bp->bio_data; resid = bp->bio_length; if (resid > 0 && (off < 0 || off >= volsize)) { error = EIO; goto out; } #ifdef illumos is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED; #else is_dumpified = B_FALSE; #endif sync = !doread && !is_dumpified && zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS; /* * There must be no buffer changes when doing a dmu_sync() because * we can't change the data whilst calculating the checksum. */ rl = zfs_range_lock(&zv->zv_znode, off, resid, doread ? RL_READER : RL_WRITER); if (bp->bio_cmd == BIO_DELETE) { dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); error = dmu_tx_assign(tx, TXG_WAIT); if (error != 0) { dmu_tx_abort(tx); } else { zvol_log_truncate(zv, tx, off, resid, B_TRUE); dmu_tx_commit(tx); error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, off, resid); resid = 0; } goto unlock; } while (resid != 0 && off < volsize) { size_t size = MIN(resid, zvol_maxphys); #ifdef illumos if (is_dumpified) { size = MIN(size, P2END(off, zv->zv_volblocksize) - off); error = zvol_dumpio(zv, addr, off, size, doread, B_FALSE); } else if (doread) { #else if (doread) { #endif error = dmu_read(os, ZVOL_OBJ, off, size, addr, DMU_READ_PREFETCH); } else { dmu_tx_t *tx = dmu_tx_create(os); dmu_tx_hold_write(tx, ZVOL_OBJ, off, size); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); } else { dmu_write(os, ZVOL_OBJ, off, size, addr, tx); zvol_log_write(zv, tx, off, size, sync); dmu_tx_commit(tx); } } if (error) { /* convert checksum errors into IO errors */ if (error == ECKSUM) error = SET_ERROR(EIO); break; } off += size; addr += size; resid -= size; } unlock: zfs_range_unlock(rl); bp->bio_completed = bp->bio_length - resid; if (bp->bio_completed < bp->bio_length && off > volsize) error = EINVAL; if (sync) { sync: zil_commit(zv->zv_zilog, ZVOL_OBJ); } out: if (bp->bio_to) g_io_deliver(bp, error); else biofinish(bp, NULL, error); } #ifdef sun /* * Set the buffer count to the zvol maximum transfer. * Using our own routine instead of the default minphys() * means that for larger writes we write bigger buffers on X86 * (128K instead of 56K) and flush the disk write cache less often * (every zvol_maxphys - currently 1MB) instead of minphys (currently * 56K on X86 and 128K on sparc). */ void zvol_minphys(struct buf *bp) { if (bp->b_bcount > zvol_maxphys) bp->b_bcount = zvol_maxphys; } int zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks) { minor_t minor = getminor(dev); zvol_state_t *zv; int error = 0; uint64_t size; uint64_t boff; uint64_t resid; zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); if (zv == NULL) return (SET_ERROR(ENXIO)); if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0) return (SET_ERROR(EINVAL)); boff = ldbtob(blkno); resid = ldbtob(nblocks); VERIFY3U(boff + resid, <=, zv->zv_volsize); while (resid) { size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff); error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE); if (error) break; boff += size; addr += size; resid -= size; } return (error); } /*ARGSUSED*/ int zvol_read(dev_t dev, uio_t *uio, cred_t *cr) { minor_t minor = getminor(dev); #else int zvol_read(struct cdev *dev, struct uio *uio, int ioflag) { #endif zvol_state_t *zv; uint64_t volsize; rl_t *rl; int error = 0; #ifdef sun zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); if (zv == NULL) return (SET_ERROR(ENXIO)); #else zv = dev->si_drv2; #endif volsize = zv->zv_volsize; if (uio->uio_resid > 0 && (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) return (SET_ERROR(EIO)); #ifdef illumos if (zv->zv_flags & ZVOL_DUMPIFIED) { error = physio(zvol_strategy, NULL, dev, B_READ, zvol_minphys, uio); return (error); } #endif rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, RL_READER); while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); /* don't read past the end */ if (bytes > volsize - uio->uio_loffset) bytes = volsize - uio->uio_loffset; error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes); if (error) { /* convert checksum errors into IO errors */ if (error == ECKSUM) error = SET_ERROR(EIO); break; } } zfs_range_unlock(rl); return (error); } #ifdef sun /*ARGSUSED*/ int zvol_write(dev_t dev, uio_t *uio, cred_t *cr) { minor_t minor = getminor(dev); #else int zvol_write(struct cdev *dev, struct uio *uio, int ioflag) { #endif zvol_state_t *zv; uint64_t volsize; rl_t *rl; int error = 0; boolean_t sync; #ifdef sun zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); if (zv == NULL) return (SET_ERROR(ENXIO)); #else zv = dev->si_drv2; #endif volsize = zv->zv_volsize; if (uio->uio_resid > 0 && (uio->uio_loffset < 0 || uio->uio_loffset > volsize)) return (SET_ERROR(EIO)); #ifdef illumos if (zv->zv_flags & ZVOL_DUMPIFIED) { error = physio(zvol_strategy, NULL, dev, B_WRITE, zvol_minphys, uio); return (error); } #endif #ifdef sun sync = !(zv->zv_flags & ZVOL_WCE) || #else sync = (ioflag & IO_SYNC) || #endif (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid, RL_WRITER); while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); uint64_t off = uio->uio_loffset; dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); if (bytes > volsize - off) /* don't write past the end */ bytes = volsize - off; dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); break; } error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); if (error == 0) zvol_log_write(zv, tx, off, bytes, sync); dmu_tx_commit(tx); if (error) break; } zfs_range_unlock(rl); if (sync) zil_commit(zv->zv_zilog, ZVOL_OBJ); return (error); } #ifdef sun int zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs) { struct uuid uuid = EFI_RESERVED; efi_gpe_t gpe = { 0 }; uint32_t crc; dk_efi_t efi; int length; char *ptr; if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag)) return (SET_ERROR(EFAULT)); ptr = (char *)(uintptr_t)efi.dki_data_64; length = efi.dki_length; /* * Some clients may attempt to request a PMBR for the * zvol. Currently this interface will return EINVAL to * such requests. These requests could be supported by * adding a check for lba == 0 and consing up an appropriate * PMBR. */ if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0) return (SET_ERROR(EINVAL)); gpe.efi_gpe_StartingLBA = LE_64(34ULL); gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1); UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid); if (efi.dki_lba == 1) { efi_gpt_t gpt = { 0 }; gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE); gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT); gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt)); gpt.efi_gpt_MyLBA = LE_64(1ULL); gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL); gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1); gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL); gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1); gpt.efi_gpt_SizeOfPartitionEntry = LE_32(sizeof (efi_gpe_t)); CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table); gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc); CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table); gpt.efi_gpt_HeaderCRC32 = LE_32(~crc); if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length), flag)) return (SET_ERROR(EFAULT)); ptr += sizeof (gpt); length -= sizeof (gpt); } if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe), length), flag)) return (SET_ERROR(EFAULT)); return (0); } /* * BEGIN entry points to allow external callers access to the volume. */ /* * Return the volume parameters needed for access from an external caller. * These values are invariant as long as the volume is held open. */ int zvol_get_volume_params(minor_t minor, uint64_t *blksize, uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl, void **rl_hdl, void **bonus_hdl) { zvol_state_t *zv; zv = zfsdev_get_soft_state(minor, ZSST_ZVOL); if (zv == NULL) return (SET_ERROR(ENXIO)); if (zv->zv_flags & ZVOL_DUMPIFIED) return (SET_ERROR(ENXIO)); ASSERT(blksize && max_xfer_len && minor_hdl && objset_hdl && zil_hdl && rl_hdl && bonus_hdl); *blksize = zv->zv_volblocksize; *max_xfer_len = (uint64_t)zvol_maxphys; *minor_hdl = zv; *objset_hdl = zv->zv_objset; *zil_hdl = zv->zv_zilog; *rl_hdl = &zv->zv_znode; *bonus_hdl = zv->zv_dbuf; return (0); } /* * Return the current volume size to an external caller. * The size can change while the volume is open. */ uint64_t zvol_get_volume_size(void *minor_hdl) { zvol_state_t *zv = minor_hdl; return (zv->zv_volsize); } /* * Return the current WCE setting to an external caller. * The WCE setting can change while the volume is open. */ int zvol_get_volume_wce(void *minor_hdl) { zvol_state_t *zv = minor_hdl; return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0); } /* * Entry point for external callers to zvol_log_write */ void zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid, boolean_t sync) { zvol_state_t *zv = minor_hdl; zvol_log_write(zv, tx, off, resid, sync); } /* * END entry points to allow external callers access to the volume. */ #endif /* sun */ /* * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. */ static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, boolean_t sync) { itx_t *itx; lr_truncate_t *lr; zilog_t *zilog = zv->zv_zilog; if (zil_replaying(zilog, tx)) return; itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); lr = (lr_truncate_t *)&itx->itx_lr; lr->lr_foid = ZVOL_OBJ; lr->lr_offset = off; lr->lr_length = len; itx->itx_sync = sync; zil_itx_assign(zilog, itx, tx); } #ifdef sun /* * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I). * Also a dirtbag dkio ioctl for unmap/free-block functionality. */ /*ARGSUSED*/ int zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp) { zvol_state_t *zv; struct dk_callback *dkc; int error = 0; rl_t *rl; mutex_enter(&spa_namespace_lock); zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL); if (zv == NULL) { mutex_exit(&spa_namespace_lock); return (SET_ERROR(ENXIO)); } ASSERT(zv->zv_total_opens > 0); switch (cmd) { case DKIOCINFO: { struct dk_cinfo dki; bzero(&dki, sizeof (dki)); (void) strcpy(dki.dki_cname, "zvol"); (void) strcpy(dki.dki_dname, "zvol"); dki.dki_ctype = DKC_UNKNOWN; dki.dki_unit = getminor(dev); dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs); mutex_exit(&spa_namespace_lock); if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag)) error = SET_ERROR(EFAULT); return (error); } case DKIOCGMEDIAINFO: { struct dk_minfo dkm; bzero(&dkm, sizeof (dkm)); dkm.dki_lbsize = 1U << zv->zv_min_bs; dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; dkm.dki_media_type = DK_UNKNOWN; mutex_exit(&spa_namespace_lock); if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag)) error = SET_ERROR(EFAULT); return (error); } case DKIOCGMEDIAINFOEXT: { struct dk_minfo_ext dkmext; bzero(&dkmext, sizeof (dkmext)); dkmext.dki_lbsize = 1U << zv->zv_min_bs; dkmext.dki_pbsize = zv->zv_volblocksize; dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs; dkmext.dki_media_type = DK_UNKNOWN; mutex_exit(&spa_namespace_lock); if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag)) error = SET_ERROR(EFAULT); return (error); } case DKIOCGETEFI: { uint64_t vs = zv->zv_volsize; uint8_t bs = zv->zv_min_bs; mutex_exit(&spa_namespace_lock); error = zvol_getefi((void *)arg, flag, vs, bs); return (error); } case DKIOCFLUSHWRITECACHE: dkc = (struct dk_callback *)arg; mutex_exit(&spa_namespace_lock); zil_commit(zv->zv_zilog, ZVOL_OBJ); if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) { (*dkc->dkc_callback)(dkc->dkc_cookie, error); error = 0; } return (error); case DKIOCGETWCE: { int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0; if (ddi_copyout(&wce, (void *)arg, sizeof (int), flag)) error = SET_ERROR(EFAULT); break; } case DKIOCSETWCE: { int wce; if (ddi_copyin((void *)arg, &wce, sizeof (int), flag)) { error = SET_ERROR(EFAULT); break; } if (wce) { zv->zv_flags |= ZVOL_WCE; mutex_exit(&spa_namespace_lock); } else { zv->zv_flags &= ~ZVOL_WCE; mutex_exit(&spa_namespace_lock); zil_commit(zv->zv_zilog, ZVOL_OBJ); } return (0); } case DKIOCGGEOM: case DKIOCGVTOC: /* * commands using these (like prtvtoc) expect ENOTSUP * since we're emulating an EFI label */ error = SET_ERROR(ENOTSUP); break; case DKIOCDUMPINIT: rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, RL_WRITER); error = zvol_dumpify(zv); zfs_range_unlock(rl); break; case DKIOCDUMPFINI: if (!(zv->zv_flags & ZVOL_DUMPIFIED)) break; rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize, RL_WRITER); error = zvol_dump_fini(zv); zfs_range_unlock(rl); break; case DKIOCFREE: { dkioc_free_t df; dmu_tx_t *tx; if (!zvol_unmap_enabled) break; if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) { error = SET_ERROR(EFAULT); break; } /* * Apply Postel's Law to length-checking. If they overshoot, * just blank out until the end, if there's a need to blank * out anything. */ if (df.df_start >= zv->zv_volsize) break; /* No need to do anything... */ mutex_exit(&spa_namespace_lock); rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length, RL_WRITER); tx = dmu_tx_create(zv->zv_objset); dmu_tx_mark_netfree(tx); error = dmu_tx_assign(tx, TXG_WAIT); if (error != 0) { dmu_tx_abort(tx); } else { zvol_log_truncate(zv, tx, df.df_start, df.df_length, B_TRUE); dmu_tx_commit(tx); error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, df.df_start, df.df_length); } zfs_range_unlock(rl); if (error == 0) { /* * If the write-cache is disabled or 'sync' property * is set to 'always' then treat this as a synchronous * operation (i.e. commit to zil). */ if (!(zv->zv_flags & ZVOL_WCE) || (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) zil_commit(zv->zv_zilog, ZVOL_OBJ); /* * If the caller really wants synchronous writes, and * can't wait for them, don't return until the write * is done. */ if (df.df_flags & DF_WAIT_SYNC) { txg_wait_synced( dmu_objset_pool(zv->zv_objset), 0); } } return (error); } default: error = SET_ERROR(ENOTTY); break; } mutex_exit(&spa_namespace_lock); return (error); } #endif /* sun */ int zvol_busy(void) { return (zvol_minors != 0); } void zvol_init(void) { VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t), 1) == 0); ZFS_LOG(1, "ZVOL Initialized."); } void zvol_fini(void) { ddi_soft_state_fini(&zfsdev_state); ZFS_LOG(1, "ZVOL Deinitialized."); } #ifdef sun /*ARGSUSED*/ static int zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) return (1); return (0); } /*ARGSUSED*/ static void zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx) { spa_t *spa = dmu_tx_pool(tx)->dp_spa; spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx); } static int zvol_dump_init(zvol_state_t *zv, boolean_t resize) { dmu_tx_t *tx; int error; objset_t *os = zv->zv_objset; spa_t *spa = dmu_objset_spa(os); vdev_t *vd = spa->spa_root_vdev; nvlist_t *nv = NULL; uint64_t version = spa_version(spa); enum zio_checksum checksum; ASSERT(MUTEX_HELD(&spa_namespace_lock)); ASSERT(vd->vdev_ops == &vdev_root_ops); error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0, DMU_OBJECT_END); /* wait for dmu_free_long_range to actually free the blocks */ txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); /* * If the pool on which the dump device is being initialized has more * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is * enabled. If so, bump that feature's counter to indicate that the * feature is active. We also check the vdev type to handle the * following case: * # zpool create test raidz disk1 disk2 disk3 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev), * the raidz vdev itself has 3 children. */ if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) { if (!spa_feature_is_enabled(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP)) return (SET_ERROR(ENOTSUP)); (void) dsl_sync_task(spa_name(spa), zfs_mvdev_dump_feature_check, zfs_mvdev_dump_activate_feature_sync, NULL, 2, ZFS_SPACE_CHECK_RESERVED); } tx = dmu_tx_create(os); dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); dmu_tx_hold_bonus(tx, ZVOL_OBJ); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); return (error); } /* * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum * function. Otherwise, use the old default -- OFF. */ checksum = spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY : ZIO_CHECKSUM_OFF; /* * If we are resizing the dump device then we only need to * update the refreservation to match the newly updated * zvolsize. Otherwise, we save off the original state of the * zvol so that we can restore them if the zvol is ever undumpified. */ if (resize) { error = zap_update(os, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &zv->zv_volsize, tx); } else { uint64_t checksum, compress, refresrv, vbs, dedup; error = dsl_prop_get_integer(zv->zv_name, zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL); error = error ? error : dsl_prop_get_integer(zv->zv_name, zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL); error = error ? error : dsl_prop_get_integer(zv->zv_name, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL); error = error ? error : dsl_prop_get_integer(zv->zv_name, zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL); if (version >= SPA_VERSION_DEDUP) { error = error ? error : dsl_prop_get_integer(zv->zv_name, zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL); } error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress, tx); error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx); error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv, tx); error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs, tx); error = error ? error : dmu_object_set_blocksize( os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx); if (version >= SPA_VERSION_DEDUP) { error = error ? error : zap_update(os, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup, tx); } if (error == 0) zv->zv_volblocksize = SPA_MAXBLOCKSIZE; } dmu_tx_commit(tx); /* * We only need update the zvol's property if we are initializing * the dump area for the first time. */ if (!resize) { VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); VERIFY(nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0); VERIFY(nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_COMPRESSION), ZIO_COMPRESS_OFF) == 0); VERIFY(nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum) == 0); if (version >= SPA_VERSION_DEDUP) { VERIFY(nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_DEDUP), ZIO_CHECKSUM_OFF) == 0); } error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, nv, NULL); nvlist_free(nv); if (error) return (error); } /* Allocate the space for the dump */ error = zvol_prealloc(zv); return (error); } static int zvol_dumpify(zvol_state_t *zv) { int error = 0; uint64_t dumpsize = 0; dmu_tx_t *tx; objset_t *os = zv->zv_objset; if (zv->zv_flags & ZVOL_RDONLY) return (SET_ERROR(EROFS)); if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) { boolean_t resize = (dumpsize > 0); if ((error = zvol_dump_init(zv, resize)) != 0) { (void) zvol_dump_fini(zv); return (error); } } /* * Build up our lba mapping. */ error = zvol_get_lbas(zv); if (error) { (void) zvol_dump_fini(zv); return (error); } tx = dmu_tx_create(os); dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); (void) zvol_dump_fini(zv); return (error); } zv->zv_flags |= ZVOL_DUMPIFIED; error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1, &zv->zv_volsize, tx); dmu_tx_commit(tx); if (error) { (void) zvol_dump_fini(zv); return (error); } txg_wait_synced(dmu_objset_pool(os), 0); return (0); } static int zvol_dump_fini(zvol_state_t *zv) { dmu_tx_t *tx; objset_t *os = zv->zv_objset; nvlist_t *nv; int error = 0; uint64_t checksum, compress, refresrv, vbs, dedup; uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset)); /* * Attempt to restore the zvol back to its pre-dumpified state. * This is a best-effort attempt as it's possible that not all * of these properties were initialized during the dumpify process * (i.e. error during zvol_dump_init). */ tx = dmu_tx_create(os); dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); return (error); } (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx); dmu_tx_commit(tx); (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum); (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress); (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv); (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs); VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); (void) nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum); (void) nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress); (void) nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv); if (version >= SPA_VERSION_DEDUP && zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) { (void) nvlist_add_uint64(nv, zfs_prop_to_name(ZFS_PROP_DEDUP), dedup); } (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL, nv, NULL); nvlist_free(nv); zvol_free_extents(zv); zv->zv_flags &= ~ZVOL_DUMPIFIED; (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END); /* wait for dmu_free_long_range to actually free the blocks */ txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, ZVOL_OBJ); error = dmu_tx_assign(tx, TXG_WAIT); if (error) { dmu_tx_abort(tx); return (error); } if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0) zv->zv_volblocksize = vbs; dmu_tx_commit(tx); return (0); } #endif /* sun */ static void zvol_geom_run(zvol_state_t *zv) { struct g_provider *pp; pp = zv->zv_provider; g_error_provider(pp, 0); kproc_kthread_add(zvol_geom_worker, zv, &zfsproc, NULL, 0, 0, "zfskern", "zvol %s", pp->name + sizeof(ZVOL_DRIVER)); } static void zvol_geom_destroy(zvol_state_t *zv) { struct g_provider *pp; g_topology_assert(); mtx_lock(&zv->zv_queue_mtx); zv->zv_state = 1; wakeup_one(&zv->zv_queue); while (zv->zv_state != 2) msleep(&zv->zv_state, &zv->zv_queue_mtx, 0, "zvol:w", 0); mtx_destroy(&zv->zv_queue_mtx); pp = zv->zv_provider; zv->zv_provider = NULL; pp->private = NULL; g_wither_geom(pp->geom, ENXIO); } static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace) { int count, error, flags; g_topology_assert(); /* * To make it easier we expect either open or close, but not both * at the same time. */ KASSERT((acr >= 0 && acw >= 0 && ace >= 0) || (acr <= 0 && acw <= 0 && ace <= 0), ("Unsupported access request to %s (acr=%d, acw=%d, ace=%d).", pp->name, acr, acw, ace)); if (pp->private == NULL) { if (acr <= 0 && acw <= 0 && ace <= 0) return (0); return (pp->error); } /* * We don't pass FEXCL flag to zvol_open()/zvol_close() if ace != 0, * because GEOM already handles that and handles it a bit differently. * GEOM allows for multiple read/exclusive consumers and ZFS allows * only one exclusive consumer, no matter if it is reader or writer. * I like better the way GEOM works so I'll leave it for GEOM to * decide what to do. */ count = acr + acw + ace; if (count == 0) return (0); flags = 0; if (acr != 0 || ace != 0) flags |= FREAD; if (acw != 0) flags |= FWRITE; g_topology_unlock(); if (count > 0) error = zvol_open(pp, flags, count); else error = zvol_close(pp, flags, -count); g_topology_lock(); return (error); } static void zvol_geom_start(struct bio *bp) { zvol_state_t *zv; boolean_t first; zv = bp->bio_to->private; ASSERT(zv != NULL); switch (bp->bio_cmd) { case BIO_FLUSH: if (!THREAD_CAN_SLEEP()) goto enqueue; zil_commit(zv->zv_zilog, ZVOL_OBJ); g_io_deliver(bp, 0); break; case BIO_READ: case BIO_WRITE: case BIO_DELETE: if (!THREAD_CAN_SLEEP()) goto enqueue; zvol_strategy(bp); break; - case BIO_GETATTR: + case BIO_GETATTR: { + spa_t *spa = dmu_objset_spa(zv->zv_objset); + uint64_t refd, avail, usedobjs, availobjs, val; + if (g_handleattr_int(bp, "GEOM::candelete", 1)) return; + if (strcmp(bp->bio_attribute, "blocksavail") == 0) { + dmu_objset_space(zv->zv_objset, &refd, &avail, + &usedobjs, &availobjs); + if (g_handleattr_off_t(bp, "blocksavail", + avail / DEV_BSIZE)) + return; + } else if (strcmp(bp->bio_attribute, "blocksused") == 0) { + dmu_objset_space(zv->zv_objset, &refd, &avail, + &usedobjs, &availobjs); + if (g_handleattr_off_t(bp, "blocksused", + refd / DEV_BSIZE)) + return; + } else if (strcmp(bp->bio_attribute, "poolblocksavail") == 0) { + avail = metaslab_class_get_space(spa_normal_class(spa)); + avail -= metaslab_class_get_alloc(spa_normal_class(spa)); + if (g_handleattr_off_t(bp, "poolblocksavail", + avail / DEV_BSIZE)) + return; + } else if (strcmp(bp->bio_attribute, "poolblocksused") == 0) { + refd = metaslab_class_get_alloc(spa_normal_class(spa)); + if (g_handleattr_off_t(bp, "poolblocksused", + refd / DEV_BSIZE)) + return; + } /* FALLTHROUGH */ + } default: g_io_deliver(bp, EOPNOTSUPP); break; } return; enqueue: mtx_lock(&zv->zv_queue_mtx); first = (bioq_first(&zv->zv_queue) == NULL); bioq_insert_tail(&zv->zv_queue, bp); mtx_unlock(&zv->zv_queue_mtx); if (first) wakeup_one(&zv->zv_queue); } static void zvol_geom_worker(void *arg) { zvol_state_t *zv; struct bio *bp; thread_lock(curthread); sched_prio(curthread, PRIBIO); thread_unlock(curthread); zv = arg; for (;;) { mtx_lock(&zv->zv_queue_mtx); bp = bioq_takefirst(&zv->zv_queue); if (bp == NULL) { if (zv->zv_state == 1) { zv->zv_state = 2; wakeup(&zv->zv_state); mtx_unlock(&zv->zv_queue_mtx); kthread_exit(); } msleep(&zv->zv_queue, &zv->zv_queue_mtx, PRIBIO | PDROP, "zvol:io", 0); continue; } mtx_unlock(&zv->zv_queue_mtx); switch (bp->bio_cmd) { case BIO_FLUSH: zil_commit(zv->zv_zilog, ZVOL_OBJ); g_io_deliver(bp, 0); break; case BIO_READ: case BIO_WRITE: zvol_strategy(bp); break; } } } extern boolean_t dataset_name_hidden(const char *name); static int zvol_create_snapshots(objset_t *os, const char *name) { uint64_t cookie, obj; char *sname; int error, len; cookie = obj = 0; sname = kmem_alloc(MAXPATHLEN, KM_SLEEP); #if 0 (void) dmu_objset_find(name, dmu_objset_prefetch, NULL, DS_FIND_SNAPSHOTS); #endif for (;;) { len = snprintf(sname, MAXPATHLEN, "%s@", name); if (len >= MAXPATHLEN) { dmu_objset_rele(os, FTAG); error = ENAMETOOLONG; break; } dsl_pool_config_enter(dmu_objset_pool(os), FTAG); error = dmu_snapshot_list_next(os, MAXPATHLEN - len, sname + len, &obj, &cookie, NULL); dsl_pool_config_exit(dmu_objset_pool(os), FTAG); if (error != 0) { if (error == ENOENT) error = 0; break; } if ((error = zvol_create_minor(sname)) != 0) { printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", sname, error); break; } } kmem_free(sname, MAXPATHLEN); return (error); } int zvol_create_minors(const char *name) { uint64_t cookie; objset_t *os; char *osname, *p; int error, len; if (dataset_name_hidden(name)) return (0); if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", name, error); return (error); } if (dmu_objset_type(os) == DMU_OST_ZVOL) { dsl_dataset_long_hold(os->os_dsl_dataset, FTAG); dsl_pool_rele(dmu_objset_pool(os), FTAG); error = zvol_create_minor(name); if (error == 0 || error == EEXIST) { error = zvol_create_snapshots(os, name); } else { printf("ZFS WARNING: Unable to create ZVOL %s (error=%d).\n", name, error); } dsl_dataset_long_rele(os->os_dsl_dataset, FTAG); dsl_dataset_rele(os->os_dsl_dataset, FTAG); return (error); } if (dmu_objset_type(os) != DMU_OST_ZFS) { dmu_objset_rele(os, FTAG); return (0); } osname = kmem_alloc(MAXPATHLEN, KM_SLEEP); if (snprintf(osname, MAXPATHLEN, "%s/", name) >= MAXPATHLEN) { dmu_objset_rele(os, FTAG); kmem_free(osname, MAXPATHLEN); return (ENOENT); } p = osname + strlen(osname); len = MAXPATHLEN - (p - osname); #if 0 /* Prefetch the datasets. */ cookie = 0; while (dmu_dir_list_next(os, len, p, NULL, &cookie) == 0) { if (!dataset_name_hidden(osname)) (void) dmu_objset_prefetch(osname, NULL); } #endif cookie = 0; while (dmu_dir_list_next(os, MAXPATHLEN - (p - osname), p, NULL, &cookie) == 0) { dmu_objset_rele(os, FTAG); (void)zvol_create_minors(osname); if ((error = dmu_objset_hold(name, FTAG, &os)) != 0) { printf("ZFS WARNING: Unable to put hold on %s (error=%d).\n", name, error); return (error); } } dmu_objset_rele(os, FTAG); kmem_free(osname, MAXPATHLEN); return (0); } static void zvol_rename_minor(zvol_state_t *zv, const char *newname) { struct g_geom *gp; struct g_provider *pp; struct cdev *dev; ASSERT(MUTEX_HELD(&spa_namespace_lock)); if (zv->zv_volmode == ZFS_VOLMODE_GEOM) { g_topology_lock(); pp = zv->zv_provider; ASSERT(pp != NULL); gp = pp->geom; ASSERT(gp != NULL); zv->zv_provider = NULL; g_wither_provider(pp, ENXIO); pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname); pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND; pp->sectorsize = DEV_BSIZE; pp->mediasize = zv->zv_volsize; pp->private = zv; zv->zv_provider = pp; g_error_provider(pp, 0); g_topology_unlock(); } else if (zv->zv_volmode == ZFS_VOLMODE_DEV) { dev = zv->zv_dev; ASSERT(dev != NULL); zv->zv_dev = NULL; destroy_dev(dev); if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s/%s", ZVOL_DRIVER, newname) == 0) { zv->zv_dev = dev; dev->si_iosize_max = MAXPHYS; dev->si_drv2 = zv; } } strlcpy(zv->zv_name, newname, sizeof(zv->zv_name)); } void zvol_rename_minors(const char *oldname, const char *newname) { char name[MAXPATHLEN]; struct g_provider *pp; struct g_geom *gp; size_t oldnamelen, newnamelen; zvol_state_t *zv; char *namebuf; boolean_t locked = B_FALSE; oldnamelen = strlen(oldname); newnamelen = strlen(newname); DROP_GIANT(); /* See comment in zvol_open(). */ if (!MUTEX_HELD(&spa_namespace_lock)) { mutex_enter(&spa_namespace_lock); locked = B_TRUE; } LIST_FOREACH(zv, &all_zvols, zv_links) { if (strcmp(zv->zv_name, oldname) == 0) { zvol_rename_minor(zv, newname); } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && (zv->zv_name[oldnamelen] == '/' || zv->zv_name[oldnamelen] == '@')) { snprintf(name, sizeof(name), "%s%c%s", newname, zv->zv_name[oldnamelen], zv->zv_name + oldnamelen + 1); zvol_rename_minor(zv, name); } } if (locked) mutex_exit(&spa_namespace_lock); PICKUP_GIANT(); } static int zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td) { zvol_state_t *zv; int err = 0; mutex_enter(&spa_namespace_lock); zv = dev->si_drv2; if (zv == NULL) { mutex_exit(&spa_namespace_lock); return(ENXIO); /* zvol_create_minor() not done yet */ } if (zv->zv_total_opens == 0) err = zvol_first_open(zv); if (err) { mutex_exit(&spa_namespace_lock); return (err); } if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) { err = SET_ERROR(EROFS); goto out; } if (zv->zv_flags & ZVOL_EXCL) { err = SET_ERROR(EBUSY); goto out; } #ifdef FEXCL if (flags & FEXCL) { if (zv->zv_total_opens != 0) { err = SET_ERROR(EBUSY); goto out; } zv->zv_flags |= ZVOL_EXCL; } #endif zv->zv_total_opens++; mutex_exit(&spa_namespace_lock); return (err); out: if (zv->zv_total_opens == 0) zvol_last_close(zv); mutex_exit(&spa_namespace_lock); return (err); } static int zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td) { zvol_state_t *zv; int err = 0; mutex_enter(&spa_namespace_lock); zv = dev->si_drv2; if (zv == NULL) { mutex_exit(&spa_namespace_lock); return(ENXIO); } if (zv->zv_flags & ZVOL_EXCL) { ASSERT(zv->zv_total_opens == 1); zv->zv_flags &= ~ZVOL_EXCL; } /* * If the open count is zero, this is a spurious close. * That indicates a bug in the kernel / DDI framework. */ ASSERT(zv->zv_total_opens != 0); /* * You may get multiple opens, but only one close. */ zv->zv_total_opens--; if (zv->zv_total_opens == 0) zvol_last_close(zv); mutex_exit(&spa_namespace_lock); return (0); } static int zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) { zvol_state_t *zv; rl_t *rl; off_t offset, length, chunk; int i, error; u_int u; zv = dev->si_drv2; error = 0; KASSERT(zv->zv_total_opens > 0, ("Device with zero access count in zvol_d_ioctl")); i = IOCPARM_LEN(cmd); switch (cmd) { case DIOCGSECTORSIZE: *(u_int *)data = DEV_BSIZE; break; case DIOCGMEDIASIZE: *(off_t *)data = zv->zv_volsize; break; case DIOCGFLUSH: zil_commit(zv->zv_zilog, ZVOL_OBJ); break; case DIOCGDELETE: if (!zvol_unmap_enabled) break; offset = ((off_t *)data)[0]; length = ((off_t *)data)[1]; if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 || offset < 0 || offset >= zv->zv_volsize || length <= 0) { printf("%s: offset=%jd length=%jd\n", __func__, offset, length); error = EINVAL; break; } rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER); dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); error = dmu_tx_assign(tx, TXG_WAIT); if (error != 0) { dmu_tx_abort(tx); } else { zvol_log_truncate(zv, tx, offset, length, B_TRUE); dmu_tx_commit(tx); error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length); } zfs_range_unlock(rl); if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) zil_commit(zv->zv_zilog, ZVOL_OBJ); break; case DIOCGSTRIPESIZE: *(off_t *)data = zv->zv_volblocksize; break; case DIOCGSTRIPEOFFSET: *(off_t *)data = 0; break; + case DIOCGATTR: { + spa_t *spa = dmu_objset_spa(zv->zv_objset); + struct diocgattr_arg *arg = (struct diocgattr_arg *)data; + uint64_t refd, avail, usedobjs, availobjs; + + if (strcmp(arg->name, "blocksavail") == 0) { + dmu_objset_space(zv->zv_objset, &refd, &avail, + &usedobjs, &availobjs); + arg->value.off = avail / DEV_BSIZE; + } else if (strcmp(arg->name, "blocksused") == 0) { + dmu_objset_space(zv->zv_objset, &refd, &avail, + &usedobjs, &availobjs); + arg->value.off = refd / DEV_BSIZE; + } else if (strcmp(arg->name, "poolblocksavail") == 0) { + avail = metaslab_class_get_space(spa_normal_class(spa)); + avail -= metaslab_class_get_alloc(spa_normal_class(spa)); + arg->value.off = avail / DEV_BSIZE; + } else if (strcmp(arg->name, "poolblocksused") == 0) { + refd = metaslab_class_get_alloc(spa_normal_class(spa)); + arg->value.off = refd / DEV_BSIZE; + } else + error = ENOIOCTL; + break; + } default: error = ENOIOCTL; } return (error); } Index: head/sys/geom/geom_dev.c =================================================================== --- head/sys/geom/geom_dev.c (revision 274153) +++ head/sys/geom/geom_dev.c (revision 274154) @@ -1,672 +1,682 @@ /*- * Copyright (c) 2002 Poul-Henning Kamp * Copyright (c) 2002 Networks Associates Technology, Inc. * All rights reserved. * * This software was developed for the FreeBSD Project by Poul-Henning Kamp * and NAI Labs, the Security Research Division of Network Associates, Inc. * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the * DARPA CHATS research program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The names of the authors may not be used to endorse or promote * products derived from this software without specific prior written * permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct g_dev_softc { struct mtx sc_mtx; struct cdev *sc_dev; struct cdev *sc_alias; int sc_open; int sc_active; }; static d_open_t g_dev_open; static d_close_t g_dev_close; static d_strategy_t g_dev_strategy; static d_ioctl_t g_dev_ioctl; static struct cdevsw g_dev_cdevsw = { .d_version = D_VERSION, .d_open = g_dev_open, .d_close = g_dev_close, .d_read = physread, .d_write = physwrite, .d_ioctl = g_dev_ioctl, .d_strategy = g_dev_strategy, .d_name = "g_dev", .d_flags = D_DISK | D_TRACKCLOSE, }; static g_init_t g_dev_init; static g_fini_t g_dev_fini; static g_taste_t g_dev_taste; static g_orphan_t g_dev_orphan; static g_attrchanged_t g_dev_attrchanged; static struct g_class g_dev_class = { .name = "DEV", .version = G_VERSION, .init = g_dev_init, .fini = g_dev_fini, .taste = g_dev_taste, .orphan = g_dev_orphan, .attrchanged = g_dev_attrchanged }; /* * We target 262144 (8 x 32768) sectors by default as this significantly * increases the throughput on commonly used SSD's with a marginal * increase in non-interruptible request latency. */ static uint64_t g_dev_del_max_sectors = 262144; SYSCTL_DECL(_kern_geom); SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff"); SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW, &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single " "delete request sent to the provider. Larger requests are chunked " "so they can be interrupted. (0 = disable chunking)"); static char *dumpdev = NULL; static void g_dev_init(struct g_class *mp) { dumpdev = kern_getenv("dumpdev"); } static void g_dev_fini(struct g_class *mp) { freeenv(dumpdev); } static int g_dev_setdumpdev(struct cdev *dev) { struct g_kerneldump kd; struct g_consumer *cp; int error, len; if (dev == NULL) return (set_dumper(NULL, NULL)); cp = dev->si_drv2; len = sizeof(kd); kd.offset = 0; kd.length = OFF_MAX; error = g_io_getattr("GEOM::kerneldump", cp, &len, &kd); if (error == 0) { error = set_dumper(&kd.di, devtoname(dev)); if (error == 0) dev->si_flags |= SI_DUMPDEV; } return (error); } static void init_dumpdev(struct cdev *dev) { if (dumpdev == NULL) return; if (strcmp(devtoname(dev), dumpdev) != 0) return; if (g_dev_setdumpdev(dev) == 0) { freeenv(dumpdev); dumpdev = NULL; } } static void g_dev_destroy(void *arg, int flags __unused) { struct g_consumer *cp; struct g_geom *gp; struct g_dev_softc *sc; g_topology_assert(); cp = arg; gp = cp->geom; sc = cp->private; g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name); if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0) g_access(cp, -cp->acr, -cp->acw, -cp->ace); g_detach(cp); g_destroy_consumer(cp); g_destroy_geom(gp); mtx_destroy(&sc->sc_mtx); g_free(sc); } void g_dev_print(void) { struct g_geom *gp; char const *p = ""; LIST_FOREACH(gp, &g_dev_class.geom, geom) { printf("%s%s", p, gp->name); p = " "; } printf("\n"); } static void g_dev_attrchanged(struct g_consumer *cp, const char *attr) { struct g_dev_softc *sc; struct cdev *dev; char buf[SPECNAMELEN + 6]; sc = cp->private; if (strcmp(attr, "GEOM::media") == 0) { dev = sc->sc_dev; snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK); dev = sc->sc_alias; if (dev != NULL) { snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name); devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK); } return; } if (strcmp(attr, "GEOM::physpath") != 0) return; if (g_access(cp, 1, 0, 0) == 0) { char *physpath; int error, physpath_len; physpath_len = MAXPATHLEN; physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO); error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath); g_access(cp, -1, 0, 0); if (error == 0 && strlen(physpath) != 0) { struct cdev *old_alias_dev; struct cdev **alias_devp; dev = sc->sc_dev; old_alias_dev = sc->sc_alias; alias_devp = (struct cdev **)&sc->sc_alias; make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp, dev, old_alias_dev, physpath); } else if (sc->sc_alias) { destroy_dev((struct cdev *)sc->sc_alias); sc->sc_alias = NULL; } g_free(physpath); } } struct g_provider * g_dev_getprovider(struct cdev *dev) { struct g_consumer *cp; g_topology_assert(); if (dev == NULL) return (NULL); if (dev->si_devsw != &g_dev_cdevsw) return (NULL); cp = dev->si_drv2; return (cp->provider); } static struct g_geom * g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused) { struct g_geom *gp; struct g_consumer *cp; struct g_dev_softc *sc; int error, len; struct cdev *dev, *adev; char buf[64], *val; g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name); g_topology_assert(); gp = g_new_geomf(mp, "%s", pp->name); sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO); mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF); cp = g_new_consumer(gp); cp->private = sc; cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; error = g_attach(cp, pp); KASSERT(error == 0, ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error)); error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev, &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name); if (error != 0) { printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n", __func__, gp->name, error); g_detach(cp); g_destroy_consumer(cp); g_destroy_geom(gp); mtx_destroy(&sc->sc_mtx); g_free(sc); return (NULL); } dev->si_flags |= SI_UNMAPPED; sc->sc_dev = dev; /* Search for device alias name and create it if found. */ adev = NULL; for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) { snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name); buf[14 + len] = 0; val = kern_getenv(buf); if (val != NULL) { snprintf(buf, sizeof(buf), "%s%s", val, gp->name + len); freeenv(val); if ((make_dev_alias_p(MAKEDEV_CHECKNAME|MAKEDEV_WAITOK, &adev, dev, "%s", buf)) != 0) printf("Warning: unable to create device " "alias %s\n", buf); break; } } dev->si_iosize_max = MAXPHYS; dev->si_drv2 = cp; init_dumpdev(dev); if (adev != NULL) { adev->si_iosize_max = MAXPHYS; adev->si_drv2 = cp; adev->si_flags |= SI_UNMAPPED; init_dumpdev(adev); } g_dev_attrchanged(cp, "GEOM::physpath"); return (gp); } static int g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td) { struct g_consumer *cp; struct g_dev_softc *sc; int error, r, w, e; cp = dev->si_drv2; if (cp == NULL) return (ENXIO); /* g_dev_taste() not done yet */ g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)", cp->geom->name, flags, fmt, td); r = flags & FREAD ? 1 : 0; w = flags & FWRITE ? 1 : 0; #ifdef notyet e = flags & O_EXCL ? 1 : 0; #else e = 0; #endif if (w) { /* * When running in very secure mode, do not allow * opens for writing of any disks. */ error = securelevel_ge(td->td_ucred, 2); if (error) return (error); } g_topology_lock(); error = g_access(cp, r, w, e); g_topology_unlock(); if (error == 0) { sc = cp->private; mtx_lock(&sc->sc_mtx); if (sc->sc_open == 0 && sc->sc_active != 0) wakeup(&sc->sc_active); sc->sc_open += r + w + e; mtx_unlock(&sc->sc_mtx); } return (error); } static int g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td) { struct g_consumer *cp; struct g_dev_softc *sc; int error, r, w, e; cp = dev->si_drv2; if (cp == NULL) return (ENXIO); g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)", cp->geom->name, flags, fmt, td); r = flags & FREAD ? -1 : 0; w = flags & FWRITE ? -1 : 0; #ifdef notyet e = flags & O_EXCL ? -1 : 0; #else e = 0; #endif sc = cp->private; mtx_lock(&sc->sc_mtx); sc->sc_open += r + w + e; while (sc->sc_open == 0 && sc->sc_active != 0) msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0); mtx_unlock(&sc->sc_mtx); g_topology_lock(); error = g_access(cp, r, w, e); g_topology_unlock(); return (error); } /* * XXX: Until we have unmessed the ioctl situation, there is a race against * XXX: a concurrent orphanization. We cannot close it by holding topology * XXX: since that would prevent us from doing our job, and stalling events * XXX: will break (actually: stall) the BSD disklabel hacks. */ static int g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td) { struct g_consumer *cp; struct g_provider *pp; off_t offset, length, chunk; int i, error; cp = dev->si_drv2; pp = cp->provider; error = 0; KASSERT(cp->acr || cp->acw, ("Consumer with zero access count in g_dev_ioctl")); i = IOCPARM_LEN(cmd); switch (cmd) { case DIOCGSECTORSIZE: *(u_int *)data = cp->provider->sectorsize; if (*(u_int *)data == 0) error = ENOENT; break; case DIOCGMEDIASIZE: *(off_t *)data = cp->provider->mediasize; if (*(off_t *)data == 0) error = ENOENT; break; case DIOCGFWSECTORS: error = g_io_getattr("GEOM::fwsectors", cp, &i, data); if (error == 0 && *(u_int *)data == 0) error = ENOENT; break; case DIOCGFWHEADS: error = g_io_getattr("GEOM::fwheads", cp, &i, data); if (error == 0 && *(u_int *)data == 0) error = ENOENT; break; case DIOCGFRONTSTUFF: error = g_io_getattr("GEOM::frontstuff", cp, &i, data); break; case DIOCSKERNELDUMP: if (*(u_int *)data == 0) error = g_dev_setdumpdev(NULL); else error = g_dev_setdumpdev(dev); break; case DIOCGFLUSH: error = g_io_flush(cp); break; case DIOCGDELETE: offset = ((off_t *)data)[0]; length = ((off_t *)data)[1]; if ((offset % cp->provider->sectorsize) != 0 || (length % cp->provider->sectorsize) != 0 || length <= 0) { printf("%s: offset=%jd length=%jd\n", __func__, offset, length); error = EINVAL; break; } while (length > 0) { chunk = length; if (g_dev_del_max_sectors != 0 && chunk > g_dev_del_max_sectors * cp->provider->sectorsize) { chunk = g_dev_del_max_sectors * cp->provider->sectorsize; } error = g_delete_data(cp, offset, chunk); length -= chunk; offset += chunk; if (error) break; /* * Since the request size can be large, the service * time can be is likewise. We make this ioctl * interruptible by checking for signals for each bio. */ if (SIGPENDING(td)) break; } break; case DIOCGIDENT: error = g_io_getattr("GEOM::ident", cp, &i, data); break; case DIOCGPROVIDERNAME: if (pp == NULL) return (ENOENT); strlcpy(data, pp->name, i); break; case DIOCGSTRIPESIZE: *(off_t *)data = cp->provider->stripesize; break; case DIOCGSTRIPEOFFSET: *(off_t *)data = cp->provider->stripeoffset; break; case DIOCGPHYSPATH: error = g_io_getattr("GEOM::physpath", cp, &i, data); if (error == 0 && *(char *)data == '\0') error = ENOENT; break; + case DIOCGATTR: { + struct diocgattr_arg *arg = (struct diocgattr_arg *)data; + + if (arg->len > sizeof(arg->value)) { + error = EINVAL; + break; + } + error = g_io_getattr(arg->name, cp, &arg->len, &arg->value); + break; + } default: if (cp->provider->geom->ioctl != NULL) { error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td); } else { error = ENOIOCTL; } } return (error); } static void g_dev_done(struct bio *bp2) { struct g_consumer *cp; struct g_dev_softc *sc; struct bio *bp; int destroy; cp = bp2->bio_from; sc = cp->private; bp = bp2->bio_parent; bp->bio_error = bp2->bio_error; bp->bio_completed = bp2->bio_completed; bp->bio_resid = bp->bio_length - bp2->bio_completed; if (bp2->bio_error != 0) { g_trace(G_T_BIO, "g_dev_done(%p) had error %d", bp2, bp2->bio_error); bp->bio_flags |= BIO_ERROR; } else { g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd", bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed); } g_destroy_bio(bp2); destroy = 0; mtx_lock(&sc->sc_mtx); if ((--sc->sc_active) == 0) { if (sc->sc_open == 0) wakeup(&sc->sc_active); if (sc->sc_dev == NULL) destroy = 1; } mtx_unlock(&sc->sc_mtx); if (destroy) g_post_event(g_dev_destroy, cp, M_WAITOK, NULL); biodone(bp); } static void g_dev_strategy(struct bio *bp) { struct g_consumer *cp; struct bio *bp2; struct cdev *dev; struct g_dev_softc *sc; KASSERT(bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE || bp->bio_cmd == BIO_FLUSH, ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd)); dev = bp->bio_dev; cp = dev->si_drv2; sc = cp->private; KASSERT(cp->acr || cp->acw, ("Consumer with zero access count in g_dev_strategy")); #ifdef INVARIANTS if ((bp->bio_offset % cp->provider->sectorsize) != 0 || (bp->bio_bcount % cp->provider->sectorsize) != 0) { bp->bio_resid = bp->bio_bcount; biofinish(bp, NULL, EINVAL); return; } #endif mtx_lock(&sc->sc_mtx); KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy")); sc->sc_active++; mtx_unlock(&sc->sc_mtx); for (;;) { /* * XXX: This is not an ideal solution, but I belive it to * XXX: deadlock safe, all things considered. */ bp2 = g_clone_bio(bp); if (bp2 != NULL) break; pause("gdstrat", hz / 10); } KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place")); bp2->bio_done = g_dev_done; g_trace(G_T_BIO, "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d", bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length, bp2->bio_data, bp2->bio_cmd); g_io_request(bp2, cp); KASSERT(cp->acr || cp->acw, ("g_dev_strategy raced with g_dev_close and lost")); } /* * g_dev_callback() * * Called by devfs when asynchronous device destruction is completed. * - Mark that we have no attached device any more. * - If there are no outstanding requests, schedule geom destruction. * Otherwise destruction will be scheduled later by g_dev_done(). */ static void g_dev_callback(void *arg) { struct g_consumer *cp; struct g_dev_softc *sc; int destroy; cp = arg; sc = cp->private; g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name); mtx_lock(&sc->sc_mtx); sc->sc_dev = NULL; sc->sc_alias = NULL; destroy = (sc->sc_active == 0); mtx_unlock(&sc->sc_mtx); if (destroy) g_post_event(g_dev_destroy, cp, M_WAITOK, NULL); } /* * g_dev_orphan() * * Called from below when the provider orphaned us. * - Clear any dump settings. * - Request asynchronous device destruction to prevent any more requests * from coming in. The provider is already marked with an error, so * anything which comes in in the interrim will be returned immediately. */ static void g_dev_orphan(struct g_consumer *cp) { struct cdev *dev; struct g_dev_softc *sc; g_topology_assert(); sc = cp->private; dev = sc->sc_dev; g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name); /* Reset any dump-area set on this device */ if (dev->si_flags & SI_DUMPDEV) (void)set_dumper(NULL, NULL); /* Destroy the struct cdev *so we get no more requests */ destroy_dev_sched_cb(dev, g_dev_callback, cp); } DECLARE_GEOM_CLASS(g_dev_class, g_dev); Index: head/sys/sys/disk.h =================================================================== --- head/sys/sys/disk.h (revision 274153) +++ head/sys/sys/disk.h (revision 274154) @@ -1,127 +1,137 @@ /*- * ---------------------------------------------------------------------------- * "THE BEER-WARE LICENSE" (Revision 42): * wrote this file. As long as you retain this notice you * can do whatever you want with this stuff. If we meet some day, and you think * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp * ---------------------------------------------------------------------------- * * $FreeBSD$ * */ #ifndef _SYS_DISK_H_ #define _SYS_DISK_H_ #include #ifdef _KERNEL #ifndef _SYS_CONF_H_ #include /* XXX: temporary to avoid breakage */ #endif void disk_err(struct bio *bp, const char *what, int blkdone, int nl); #endif #define DIOCGSECTORSIZE _IOR('d', 128, u_int) /* * Get the sector size of the device in bytes. The sector size is the * smallest unit of data which can be transferred from this device. * Usually this is a power of 2 but it might not be (i.e. CDROM audio). */ #define DIOCGMEDIASIZE _IOR('d', 129, off_t) /* Get media size in bytes */ /* * Get the size of the entire device in bytes. This should be a * multiple of the sector size. */ #define DIOCGFWSECTORS _IOR('d', 130, u_int) /* Get firmware's sectorcount */ /* * Get the firmware's notion of number of sectors per track. This * value is mostly used for compatibility with various ill designed * disk label formats. Don't use it unless you have to. */ #define DIOCGFWHEADS _IOR('d', 131, u_int) /* Get firmware's headcount */ /* * Get the firmwares notion of number of heads per cylinder. This * value is mostly used for compatibility with various ill designed * disk label formats. Don't use it unless you have to. */ #define DIOCSKERNELDUMP _IOW('d', 133, u_int) /* Set/Clear kernel dumps */ /* * Enable/Disable (the argument is boolean) the device for kernel * core dumps. */ #define DIOCGFRONTSTUFF _IOR('d', 134, off_t) /* * Many disk formats have some amount of space reserved at the * start of the disk to hold bootblocks, various disklabels and * similar stuff. This ioctl returns the number of such bytes * which may apply to the device. */ #define DIOCGFLUSH _IO('d', 135) /* Flush write cache */ /* * Flush write cache of the device. */ #define DIOCGDELETE _IOW('d', 136, off_t[2]) /* Delete data */ /* * Mark data on the device as unused. */ #define DISK_IDENT_SIZE 256 #define DIOCGIDENT _IOR('d', 137, char[DISK_IDENT_SIZE]) /*- * Get the ident of the given provider. Ident is (most of the time) * a uniqe and fixed provider's identifier. Ident's properties are as * follow: * - ident value is preserved between reboots, * - provider can be detached/attached and ident is preserved, * - provider's name can change - ident can't, * - ident value should not be based on on-disk metadata; in other * words copying whole data from one disk to another should not * yield the same ident for the other disk, * - there could be more than one provider with the same ident, but * only if they point at exactly the same physical storage, this is * the case for multipathing for example, * - GEOM classes that consumes single providers and provide single * providers, like geli, gbde, should just attach class name to the * ident of the underlying provider, * - ident is an ASCII string (is printable), * - ident is optional and applications can't relay on its presence. */ #define DIOCGPROVIDERNAME _IOR('d', 138, char[MAXPATHLEN]) /* * Store the provider name, given a device path, in a buffer. The buffer * must be at least MAXPATHLEN bytes long. */ #define DIOCGSTRIPESIZE _IOR('d', 139, off_t) /* Get stripe size in bytes */ /* * Get the size of the device's optimal access block in bytes. * This should be a multiple of the sector size. */ #define DIOCGSTRIPEOFFSET _IOR('d', 140, off_t) /* Get stripe offset in bytes */ /* * Get the offset of the first device's optimal access block in bytes. * This should be a multiple of the sector size. */ #define DIOCGPHYSPATH _IOR('d', 141, char[MAXPATHLEN]) /* * Get a string defining the physical path for a given provider. * This has similar rules to ident, but is intended to uniquely * identify the physical location of the device, not the current * occupant of that location. */ +struct diocgattr_arg { + char name[64]; + int len; + union { + char str[DISK_IDENT_SIZE]; + off_t off; + } value; +}; +#define DIOCGATTR _IOWR('d', 142, struct diocgattr_arg) + #endif /* _SYS_DISK_H_ */ Index: head/usr.sbin/ctladm/ctladm.8 =================================================================== --- head/usr.sbin/ctladm/ctladm.8 (revision 274153) +++ head/usr.sbin/ctladm/ctladm.8 (revision 274154) @@ -1,1114 +1,1122 @@ .\" .\" 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. .\" .\" ctladm utility man page. .\" .\" Author: Ken Merry .\" .\" $Id: //depot/users/kenm/FreeBSD-test2/usr.sbin/ctladm/ctladm.8#3 $ .\" $FreeBSD$ .\" -.Dd October 26, 2014 +.Dd November 5, 2014 .Dt CTLADM 8 .Os .Sh NAME .Nm ctladm .Nd CAM Target Layer control utility .Sh SYNOPSIS .Nm .Aq Ar command .Op target:lun .Op generic args .Op command args .Nm .Ic tur .Aq target:lun .Op general options .Nm .Ic inquiry .Aq target:lun .Op general options .Nm .Ic reqsense .Aq target:lun .Op general options .Nm .Ic reportluns .Aq target:lun .Op general options .Nm .Ic read .Aq target:lun .Op general options .Aq Fl l Ar lba .Aq Fl d Ar datalen .Aq Fl f Ar file|- .Aq Fl b Ar blocksize_bytes .Op Fl c Ar cdbsize .Op Fl N .Nm .Ic write .Aq target:lun .Op general options .Aq Fl l Ar lba .Aq Fl d Ar datalen .Aq Fl f Ar file|- .Aq Fl b Ar blocksize_bytes .Op Fl c Ar cdbsize .Op Fl N .Nm .Ic bbrread .Aq target:lun .Op general options .Aq Fl -l Ar lba .Aq Fl -d Ar datalen .Nm .Ic readcap .Aq target:lun .Op general options .Op Fl c Ar cdbsize .Nm .Ic modesense .Aq target:lun .Aq Fl m Ar page | Fl l .Op Fl P Ar pc .Op Fl d .Op Fl S Ar subpage .Op Fl c Ar size .Nm .Ic start .Aq target:lun .Op general options .Op Fl i .Op Fl o .Nm .Ic stop .Aq target:lun .Op general options .Op Fl i .Op Fl o .Nm .Ic synccache .Aq target:lun .Op general options .Op Fl l Ar lba .Op Fl b Ar blockcount .Op Fl r .Op Fl i .Op Fl c Ar cdbsize .Nm .Ic shutdown .Op general options .Nm .Ic startup .Op general options .Nm .Ic hardstop .Nm .Ic hardstart .Nm .Ic lunlist .Nm .Ic delay .Aq target:lun .Aq Fl l Ar datamove|done .Aq Fl t Ar secs .Op Fl T Ar oneshot|cont .Nm .Ic realsync Aq on|off|query .Nm .Ic setsync interval .Aq target:lun .Aq Fl i Ar interval .Nm .Ic getsync .Aq target:lun .Nm .Ic inject .Aq Fl i Ar action .Aq Fl p Ar pattern .Op Fl r Ar lba,len .Op Fl s Ar len fmt Op Ar args .Op Fl c .Op Fl d Ar delete_id .Nm .Ic create .Aq Fl b Ar backend .Op Fl B Ar blocksize .Op Fl d Ar device_id .Op Fl l Ar lun_id .Op Fl o Ar name=value .Op Fl s Ar size_bytes .Op Fl S Ar serial_num .Op Fl t Ar device_type .Nm .Ic remove .Aq Fl b Ar backend .Aq Fl l Ar lun_id .Op Fl o Ar name=value .Nm .Ic modify .Aq Fl b Ar backend .Aq Fl l Ar lun_id .Aq Fl s Ar size_bytes .Nm .Ic devlist .Op Fl b Ar backend .Op Fl v .Op Fl x .Nm .Ic port .Op Fl l .Op Fl o Ar on|off .Op Fl w Ar wwpn .Op Fl W Ar wwnn .Op Fl p Ar targ_port .Op Fl t Ar fe_type .Op Fl q .Op Fl x .Nm .Ic portlist .Op Fl f Ar frontend .Op Fl i .Op Fl p Ar targ_port .Op Fl q .Op Fl v .Op Fl x .Nm .Ic dumpooa .Nm .Ic dumpstructs .Nm .Ic islist .Op Fl v .Op Fl x .Nm .Ic islogout .Aq Fl a | Fl c Ar connection-id | Fl i Ar name | Fl p Ar portal .Nm .Ic isterminate .Aq Fl a | Fl c Ar connection-id | Fl i Ar name | Fl p Ar portal .Nm .Ic help .Sh DESCRIPTION The .Nm utility is designed to provide a way to access and control the CAM Target Layer (CTL). It provides a way to send .Tn SCSI commands to the CTL layer, and also provides some meta-commands that utilize .Tn SCSI commands. (For instance, the .Ic lunlist command is implemented using the .Tn SCSI REPORT LUNS and INQUIRY commands.) .Pp The .Nm utility has a number of primary functions, many of which require a device identifier. The device identifier takes the following form: .Bl -tag -width 14n .It target:lun Specify the target (almost always 0) and LUN number to operate on. .El Many of the primary functions of the .Nm utility take the following optional arguments: .Bl -tag -width 10n .It Fl C Ar retries Specify the number of times to retry a command in the event of failure. .It Fl D Ar device Specify the device to open. This allows opening a device other than the default device, .Pa /dev/cam/ctl , to be opened for sending commands. .It Fl I Ar id Specify the initiator number to use. By default, .Nm will use 7 as the initiator number. .El .Pp Primary commands: .Bl -tag -width 11n .It Ic tur Send the .Tn SCSI TEST UNIT READY command to the device and report whether or not it is ready. .It Ic inquiry Send the .Tn SCSI INQUIRY command to the device and display some of the returned inquiry data. .It Ic reqsense Send the .Tn SCSI REQUEST SENSE command to the device and display the returned sense information. .It Ic reportluns Send the .Tn SCSI REPORT LUNS command to the device and display supported LUNs. .It Ic read Send a .Tn SCSI READ command to the device, and write the requested data to a file or stdout. .Bl -tag -width 12n .It Fl l Ar lba Specify the starting Logical Block Address for the READ. This can be specified in decimal, octal (starting with 0), hexadecimal (starting with 0x) or any other base supported by .Xr strtoull 3 . .It Fl d Ar datalen Specify the length, in 512 byte blocks, of the READ request. .It Fl f Ar file Specify the destination for the data read by the READ command. Either a filename or .Sq - for stdout may be specified. .It Fl c Ar cdbsize Specify the minimum .Tn SCSI CDB (Command Data Block) size to be used for the READ request. Allowable values are 6, 10, 12 and 16. Depending upon the LBA and amount of data requested, a larger CDB size may be used to satisfy the request. (e.g., for LBAs above 0xffffffff, READ(16) must be used to satisfy the request.) .It Fl b Ar blocksize Specify the blocksize of the underlying .Tn SCSI device, so the transfer length can be calculated accurately. The blocksize can be obtained via the .Tn SCSI READ CAPACITY command. .It Fl N Do not copy data to .Nm from the kernel when doing a read, just execute the command without copying data. This is to be used for performance testing. .El .It Ic write Read data from a file or stdin, and write the data to the device using the .Tn SCSI WRITE command. .Bl -tag -width 12n .It Fl l Ar lba Specify the starting Logical Block Address for the WRITE. This can be specified in decimal, octal (starting with 0), hexadecimal (starting with 0x) or any other base supported by .Xr strtoull 3 . .It Fl d Ar atalen Specify the length, in 512 byte blocks, of the WRITE request. .It Fl f Ar file Specify the source for the data to be written by the WRITE command. Either a filename or .Sq - for stdin may be specified. .It Fl c Ar cdbsize Specify the minimum .Tn SCSI CDB (Command Data Block) size to be used for the READ request. Allowable values are 6, 10, 12 and 16. Depending upon the LBA and amount of data requested, a larger CDB size may be used to satisfy the request. (e.g., for LBAs above 0xffffffff, READ(16) must be used to satisfy the request.) .It Fl b Ar blocksize Specify the blocksize of the underlying .Tn SCSI device, so the transfer length can be calculated accurately. The blocksize can be obtained via the .Tn SCSI READ CAPACITY command. .It Fl N Do not copy data to .Nm to the kernel when doing a write, just execute the command without copying data. This is to be used for performance testing. .El .It Ic bbrread Issue a SCSI READ command to the logical device to potentially force a bad block on a disk in the RAID set to be reconstructed from the other disks in the array. This command should only be used on an array that is in the normal state. If used on a critical array, it could cause the array to go offline if the bad block to be remapped is on one of the disks that is still active in the array. .Pp The data for this particular command will be discarded, and not returned to the user. .Pp In order to determine which LUN to read from, the user should first determine which LUN the disk with a bad block belongs to. Then he should map the bad disk block back to the logical block address for the array in order to determine which LBA to pass in to the .Ic bbrread command. .Pp This command is primarily intended for testing. In practice, bad block remapping will generally be triggered by the in-kernel Disk Aerobics and Disk Scrubbing code. .Bl -tag -width 10n .It Fl l Ar lba Specify the starting Logical Block Address. .It Fl d Ar datalen Specify the amount of data in bytes to read from the LUN. This must be a multiple of the LUN blocksize. .El .It Ic readcap Send the .Tn SCSI READ CAPACITY command to the device and display the device size and device block size. By default, READ CAPACITY(10) is used. If the device returns a maximum LBA of 0xffffffff, however, .Nm will automatically issue a READ CAPACITY(16), which is implemented as a service action of the SERVICE ACTION IN(16) opcode. The user can specify the minimum CDB size with the .Fl c argument. Valid values for the .Fl c option are 10 and 16. If a 10 byte CDB is specified, the request will be automatically reissued with a 16 byte CDB if the maximum LBA returned is 0xffffffff. .It Ic modesense Send a .Tn SCSI MODE SENSE command to the device, and display the requested mode page(s) or page list. .Bl -tag -width 10n .It Fl m Ar page Specify the mode page to display. This option and the .Fl l option are mutually exclusive. One of the two must be specified, though. Mode page numbers may be specified in decimal or hexadecimal. .It Fl l Request that the list of mode pages supported by the device be returned. This option and the .Fl m option are mutually exclusive. One of the two must be specified, though. .It Fl P Ar pc Specify the mode page control value. Possible values are: .Bl -tag -width 2n -compact .It 0 Current values. .It 1 Changeable value bitmask. .It 2 Default values. .It 3 Saved values. .El .It Fl d Disable block descriptors when sending the mode sense request. .It Fl S Ar subpage Specify the subpage used with the mode sense request. .It Fl c Ar cdbsize Specify the CDB size used for the mode sense request. Supported values are 6 and 10. .El .It Ic start Send the .Tn SCSI START STOP UNIT command to the specified LUN with the start bit set. .Bl -tag -width 4n .It Fl i Set the immediate bit in the CDB. Note that CTL does not support the immediate bit, so this is primarily useful for making sure that CTL returns the proper error. .It Fl o Set the Copan proprietary on/offline bit in the CDB. When this flag is used, the LUN will be marked online again (see the description of the .Ic shutdown and .Ic startup commands). When this flag is used with a start command, the LUN will NOT be spun up. You need to use a start command without the .Fl o flag to spin up the disks in the LUN. .El .It Ic stop Send the .Tn SCSI START STOP UNIT command to the specified LUN with the start bit cleared. We use an ordered tag to stop the LUN, so we can guarantee that all pending I/O executes before it is stopped. (CTL guarantees this anyway, but .Nm sends an ordered tag for completeness.) .Bl -tag -width 4n .It Fl i Set the immediate bit in the CDB. Note that CTL does not support the immediate bit, so this is primarily useful for making sure that CTL returns the proper error. .It Fl o Set the Copan proprietary on/offline bit in the CDB. When this flag is used, the LUN will be spun down and taken offline ("Logical unit not ready, manual intervention required"). See the description of the .Ic shutdown and .Ic startup options. .El .It Ic synccache Send the .Tn SCSI SYNCHRONIZE CACHE command to the device. By default, SYNCHRONIZE CACHE(10) is used. If the specified starting LBA is greater than 0xffffffff or the length is greater than 0xffff, though, SYNCHRONIZE CACHE(16) will be used. The 16 byte command will also be used if the user specifies a 16 byte CDB with the .Fl c argument. .Bl -tag -width 14n .It Fl l Ar lba Specify the starting LBA of the cache region to synchronize. This option is a no-op for CTL. If you send a SYNCHRONIZE CACHE command, it will sync the cache for the entire LUN. .It Fl b Ar blockcount Specify the length of the cache region to synchronize. This option is a no-op for CTL. If you send a SYNCHRONIZE CACHE command, it will sync the cache for the entire LUN. .It Fl r Specify relative addressing for the starting LBA. CTL does not support relative addressing, since it only works for linked commands, and CTL does not support linked commands. .It Fl i Tell the target to return status immediately after issuing the SYNCHRONIZE CACHE command rather than waiting for the cache to finish syncing. CTL does not support this bit. .It Fl c Ar cdbsize Specify the minimum CDB size. Valid values are 10 and 16 bytes. .El .It Ic shutdown Issue a .Tn SCSI START STOP UNIT command with the start bit cleared and the on/offline bit set to all direct access LUNs. This will spin down all direct access LUNs, and mark them offline ("Logical unit not ready, manual intervention required"). Once marked offline, the state can only be cleared by sending a START STOP UNIT command with the start bit set and the on/offline bit set. The .Nm commands .Ic startup and .Ic start will accomplish this. Note that the on/offline bit is a non-standard Copan extension to the .Tn SCSI START STOP UNIT command, so merely sending a normal start command from an initiator will not clear the condition. (This is by design.) .It Ic startup Issue a .Tn SCSI START STOP UNIT command with the start bit set and the on/offline bit set to all direct access LUNs. This will mark all direct access LUNs "online" again. It will not cause any LUNs to start up. A separate start command without the on/offline bit set is necessary for that. .It Ic hardstop Use the kernel facility for stopping all direct access LUNs and setting the offline bit. Unlike the .Ic shutdown command above, this command allows shutting down LUNs with I/O active. It will also issue a LUN reset to any reserved LUNs to break the reservation so that the LUN can be stopped. .Ic shutdown command instead. .It Ic hardstart This command is functionally identical to the .Ic startup command described above. The primary difference is that the LUNs are enumerated and commands sent by the in-kernel Front End Target Driver instead of by .Nm . .It Ic lunlist List all LUNs registered with CTL. Because this command uses the ioctl port, it will only work when the FETDs (Front End Target Drivers) are enabled. This command is the equivalent of doing a REPORT LUNS on one LUN and then an INQUIRY on each LUN in the system. .It Ic delay Delay commands at the given location. There are two places where commands may be delayed currently: before data is transferred .Pq Dq datamove and just prior to sending status to the host .Pq Dq done . One of the two must be supplied as an argument to the .Fl l option. The .Fl t option must also be specified. .Bl -tag -width 12n .It Fl l Ar delayloc Delay command(s) at the specified location. This can either be at the data movement stage (datamove) or prior to command completion (done). .It Fl t Ar delaytime Delay command(s) for the specified number of seconds. This must be specified. If set to 0, it will clear out any previously set delay for this particular location (datamove or done). .It Fl T Ar delaytype Specify the delay type. By default, the .Ic delay option will delay the next command sent to the given LUN. With the .Fl T Ar cont option, every command will be delayed by the specified period of time. With the .Fl T Ar oneshot the next command sent to the given LUN will be delayed and all subsequent commands will be completed normally. This is the default. .El .It Ic realsync Query and control CTL's SYNCHRONIZE CACHE behavior. The .Sq query argument will show whether SYNCHRONIZE CACHE commands are being sent to the backend or not. The default is to send SYNCHRONIZE CACHE commands to the backend. The .Sq on argument will cause all SYNCHRONIZE CACHE commands sent to all LUNs to be sent to the backend. The .Sq off argument will cause all SYNCHRONIZE CACHE commands sent to all LUNs to be immediately returned to the initiator with successful status. .It Ic setsync For a given lun, only actually service every Nth SYNCHRONIZE CACHE command that is sent. This can be used for debugging the optimal time period for sending SYNCHRONIZE cache commands. An interval of 0 means that the cache will be flushed for this LUN every time a SYNCHRONIZE CACHE command is received. .Pp You must specify the target and LUN you want to modify. .It Ic getsync Get the interval at which we actually service the SYNCHRONIZE CACHE command, as set by the .Ic setsync command above. The reported number means that we will actually flush the cache on every Nth SYNCHRONIZE CACHE command. A value of 0 means that we will flush the cache every time. .Pp You must specify the target and LUN you want to query. .It Ic inject Inject the specified type of error for the LUN specified, when a command that matches the given pattern is seen. The sense data returned is in either fixed or descriptor format, depending upon the status of the D_SENSE bit in the control mode page (page 0xa) for the LUN. .Pp Errors are only injected for commands that have not already failed for other reasons. By default, only the first command matching the pattern specified is returned with the supplied error. .Pp If the .Fl c flag is specified, all commands matching the pattern will be returned with the specified error until the error injection command is deleted with .Fl d flag. .Bl -tag -width 17n .It Fl i Ar action Specify the error to return: .Bl -tag -width 10n .It aborted Return the next matching command on the specified LUN with the sense key ABORTED COMMAND (0x0b), and the ASC/ASCQ 0x45,0x00 ("Select or reselect failure"). .It mediumerr Return the next matching command on the specified LUN with the sense key MEDIUM ERROR (0x03) and the ASC/ASCQ 0x11,0x00 ("Unrecovered read error") for reads, or ASC/ASCQ 0x0c,0x02 ("Write error - auto reallocation failed") for write errors. .It ua Return the next matching command on the specified LUN with the sense key UNIT ATTENTION (0x06) and the ASC/ASCQ 0x29,0x00 ("POWER ON, RESET, OR BUS DEVICE RESET OCCURRED"). .It custom Return the next matching command on the specified LUN with the supplied sense data. The .Fl s argument must be specified. .El .It Fl p Ar pattern Specify which commands should be returned with the given error. .Bl -tag -width 10n .It read The error should apply to READ(6), READ(10), READ(12), READ(16), etc. .It write The error should apply to WRITE(6), WRITE(10), WRITE(12), WRITE(16), WRITE AND VERIFY(10), etc. .It rw The error should apply to both read and write type commands. .It readcap The error should apply to READ CAPACITY(10) and READ CAPACITY(16) commands. .It tur The error should apply to TEST UNIT READY commands. .It any The error should apply to any command. .El .It Fl r Ar lba,len Specify the starting lba and length of the range of LBAs which should trigger an error. This option is only applies when read and/or write patterns are specified. If used with other command types, the error will never be triggered. .It Fl s Ar len fmt Op Ar args Specify the sense data that is to be returned for custom actions. If the format is .Sq - , len bytes of sense data will be read from standard input and written to the sense buffer. If len is longer than 252 bytes (the maximum allowable .Tn SCSI sense data length), it will be truncated to that length. The sense data format is described in .Xr cam_cdparse 3 . .It Fl c The error injection should be persistent, instead of happening once. Persistent errors must be deleted with the .Fl d argument. .It Fl d Ar delete_id Delete the specified error injection serial number. The serial number is returned when the error is injected. .El .It Ic port Perform one of several CTL frontend port operations. Either get a list of frontend ports .Pq Fl l , turn one or more frontends on or off .Pq Fl o Ar on|off , or set the World Wide Node Name .Pq Fl w Ar wwnn or World Wide Port Name .Pq Fl W Ar wwpn for a given port. One of .Fl l , .Fl o , or .Fl w or .Fl W must be specified. The WWNN and WWPN may both be specified at the same time, but cannot be combined with enabling/disabling or listing ports. .Bl -tag -width 12n .It Fl l List all CTL frontend ports or a specific port type or number. .It Fl o Ar on|off Turn the specified CTL frontend ports off or on. If no port number or port type is specified, all ports are turned on or off. .It Fl p Ar targ_port Specify the frontend port number. The port numbers can be found in the frontend port list. .It Fl q Omit the header in the port list output. .It Fl t Ar fe_type Specify the frontend type. Currently defined port types are .Dq fc (Fibre Channel), .Dq scsi (Parallel SCSI), .Dq ioctl (CTL ioctl interface), and .Dq internal (CTL CAM SIM). .It Fl w Ar wwnn Set the World Wide Node Name for the given port. The .Fl n argument must be specified, since this is only possible to implement on a single port. As a general rule, the WWNN should be the same across all ports on the system. .It Fl W Ar wwpn Set the World Wide Port Name for the given port. The .Fl n argument must be specified, since this is only possible to implement on a single port. As a general rule, the WWPN must be different for every port in the system. .It Fl x Output the port list in XML format. .El .It Ic portlist List CTL frontend ports. .Bl -tag -width 12n .It Fl f Ar frontend Specify the frontend type. .It Fl i Report target and connected initiators addresses .It Fl p Ar targ_port Specify the frontend port number. .It Fl q Omit the header in the port list output. .It Fl v Enable verbose output (report all port options). .It Fl x Output the port list in XML format. .El .It Ic dumpooa Dump the OOA (Order Of Arrival) queue for each LUN registered with CTL. .It Ic dumpstructs Dump the CTL structures to the console. .It Ic create Create a new LUN. The backend must be specified, and depending upon the backend requested, some of the other options may be required. If the LUN is created successfully, the LUN configuration will be displayed. If LUN creation fails, a message will be displayed describing the failure. .Bl -tag -width 14n .It Fl b Ar backend The .Fl b flag is required. This specifies the name backend to use when creating the LUN. Examples are .Dq ramdisk and .Dq block . .It Fl B Ar blocksize Specify the blocksize of the backend in bytes. .It Fl d Ar device_id Specify the LUN-associated string to use in the .Tn SCSI INQUIRY VPD page 0x83 data. .It Fl l Ar lun_id Request that a particular LUN number be assigned. If the requested LUN number is not available, the request will fail. .It Fl o Ar name=value Specify a backend-specific name/value pair. Multiple .Fl o arguments may be specified. Refer to the backend documentation for arguments that may be used. .It Fl s Ar size_bytes Specify the size of the LUN in bytes. Some backends may allow setting the size (e.g. the ramdisk backend) and for others the size may be implicit (e.g. the block backend). .It Fl S Ar serial_num Specify the serial number to be used in the .Tn SCSI INQUIRY VPD page 0x80 data. .It Fl t Ar device_type Specify the numeric SCSI device type to use when creating the LUN. For example, the Direct Access type is 0. If this flag is not used, the type of LUN created is backend-specific. Not all LUN types are supported. Currently CTL only supports Direct Access (type 0) and Processor (type 3) LUNs. The backend requested may or may not support all of the LUN types that CTL supports. .El .It Ic remove Remove a LUN. The backend must be specified, and the LUN number must also be specified. Backend-specific options may also be specified with the .Fl o flag. .Bl -tag -width 14n .It Fl b Ar backend Specify the backend that owns the LUN to be removed. Examples are .Dq ramdisk and .Dq block . .It Fl l Ar lun_id Specify the LUN number to remove. .It Fl o Ar name=value Specify a backend-specific name/value pair. Multiple .Fl o arguments may be specified. Refer to the backend documentation for arguments that may be used. .El .It Ic modify Modify a LUN size. The backend, the LUN number, and the size must be specified. .Bl -tag -width 14n .It Fl b Ar backend Specify the backend that owns the LUN to be removed. Examples are .Dq ramdisk and .Dq block . .It Fl l Ar lun_id Specify the LUN number to remove. .It Fl s Ar size_bytes Specify the size of the LUN in bytes. For the .Dq block backend, an .Dq auto keyword may be passed instead; this will make CTL use the size of backing file or device. .El .It Ic devlist Get a list of all configured LUNs. This also includes the LUN size and blocksize, serial number and device ID. .Bl -tag -width 11n .It Fl b Ar backend Specify the backend. This restricts the LUN list to the named backend. Examples are .Dq ramdisk and .Dq block . .It Fl v Be verbose. This will also display any backend-specific LUN attributes in addition to the standard per-LUN information. .It Fl x Dump the raw XML. The LUN list information from the kernel comes in XML format, and this option allows the display of the raw XML data. This option and the .Fl v and .Fl b options are mutually exclusive. If you specify .Fl x , the entire LUN database is displayed in XML format. .El .It Ic islist Get a list of currently running iSCSI connections. This includes initiator and target names and the unique connection IDs. .Bl -tag -width 11n .It Fl v Verbose mode. .It Fl x Dump the raw XML. The connections list information from the kernel comes in XML format, and this option allows the display of the raw XML data. .El .It Ic islogout Ask the initiator to log out iSCSI connections matching criteria. .Bl -tag -width 11n .It Fl a Log out all connections. .It Fl c Specify connection ID. .It Fl i Specify initiator name. .It Fl p Specify initiator portal (hostname or IP address). .El .It Ic isterminate Forcibly terminate iSCSI connections matching criteria. .Bl -tag -width 11n .It Fl a Terminate all connections. .It Fl c Specify connection ID. .It Fl i Specify initiator name. .It Fl p Specify initiator portal (hostname or IP address). .El .It Ic help Display .Nm usage information. .El .Sh OPTIONS Number of additional configuration options may be specified for LUNs. Some options are global, others are backend-specific. .Pp Global options: .Bl -tag -width 12n .It Va vendor Specifies LUN vendor string up to 8 chars. .It Va product Specifies LUN product string up to 16 chars. .It Va revision Specifies LUN revision string up to 4 chars. .It Va scsiname Specifies LUN SCSI name string. .It Va eui Specifies LUN EUI-64 identifier. .It Va naa Specifies LUN NAA identifier. Either EUI or NAA identifier should be set to UNIQUE value to allow EXTENDED COPY command access the LUN. Non-unique LUN identifiers may lead to data corruption. .It Va insecure_tpc Setting to "on" allows EXTENDED COPY command sent to this LUN access other LUNs on this host, not accessible otherwise. This allows to offload copying between different iSCSI targets residing on the same host in trusted environments. .It Va readcache Set to "off", disables read caching for the LUN, if supported by the backend. .It Va readonly Set to "on", blocks all media write operations to the LUN, reporting it as write protected. .It Va reordering Set to "unrestricted", allows target to process commands with SIMPLE task attribute in arbitrary order. Any data integrity exposures related to command sequence order shall be explicitly handled by the application client through the selection of appropriate commands and task attributes. The default value is "restricted". It improves data integrity, but may introduce some additional delays. .It Va rpm Specifies medium rotation rate of the device: 0 -- not reported, 1 -- non-rotating (SSD), >1024 -- value in revolutions per minute. .It Va formfactor Specifies nominal form factor of the device: 0 -- not reported, 1 -- 5.25", 2 -- 3.5", 3 -- 2.5", 4 -- 1.8", 5 -- less then 1.8". .It Va unmap Set to "on", enables UNMAP support for the LUN, if supported by the backend. +.It Va avail-threshold +.It Va used-threshold +.It Va pool-avail-threshold +.It Va pool-used-threshold +Set per-LUN/-pool thin provisioning soft thresholds for ZVOL-backed LUNs. +LUN will establish UNIT ATTENTION condition if its or pool available space +get below configured avail values, or its or pool used space get above +configured used values. .It Va writecache Set to "off", disables write caching for the LUN, if supported by the backend. .El .Pp Options specific for block backend: .Bl -tag -width 12n .It Va file Specifies file or device name to use for backing store. .It Va num_threads Specifies number of backend threads to use for this LUN. .El .Sh EXAMPLES .Dl ctladm tur 0:1 .Pp Send a .Tn SCSI TEST UNIT READY command to LUN 1. .Pp .Dl ctladm modesense 0:1 -l .Pp Display the list of mode pages supported by LUN 1. .Pp .Dl ctladm modesense 0:0 -m 10 -P 3 -d -c 10 .Pp Display the saved version of the Control mode page (page 10) on LUN 0. Disable fetching block descriptors, and use a 10 byte MODE SENSE command instead of the default 6 byte command. .Bd -literal ctladm read 0:2 -l 0 -d 1 -b 512 -f - > foo .Ed .Pp Read the first 512 byte block from LUN 2 and dump it to the file .Pa foo . .Bd -literal ctladm write 0:3 -l 0xff432140 -d 20 -b 512 -f /tmp/bar .Ed .Pp Read 10240 bytes from the file .Pa /tmp/bar and write it to target 0, LUN 3. starting at LBA 0xff432140. .Pp .Dl ctladm create -b ramdisk -s 10485760000000000 .Pp Create a LUN with the .Dq fake ramdisk as a backing store. The LUN will claim to have a size of approximately 10 terabytes. .Pp .Dl ctladm create -b block -o file=src/usr.sbin/ctladm/ctladm.8 .Pp Create a LUN using the block backend, and specify the file .Pa src/usr.sbin/ctladm/ctladm.8 as the backing store. The size of the LUN will be derived from the size of the file. .Pp .Dl ctladm create -b block -o file=src/usr.sbin/ctladm/ctladm.8 -S MYSERIAL321 -d MYDEVID123 .Pp Create a LUN using the block backend, specify the file .Pa src/usr.sbin/ctladm/ctladm.8 as the backing store, and specify the .Tn SCSI VPD page 0x80 and 0x83 serial number .Fl ( S ) and device ID .Fl ( d ) . .Pp .Dl ctladm remove -b block -l 12 .Pp Remove LUN 12, which is handled by the block backend, from the system. .Pp .Dl ctladm devlist .Pp List configured LUNs in the system, along with their backend and serial number. This works when the Front End Target Drivers are enabled or disabled. .Pp .Dl ctladm lunlist .Pp List all LUNs in the system, along with their inquiry data and device type. This only works when the FETDs are enabled, since the commands go through the ioctl port. .Pp .Dl ctladm inject 0:6 -i mediumerr -p read -r 0,512 -c .Pp Inject a medium error on LUN 6 for every read that covers the first 512 blocks of the LUN. .Bd -literal -offset indent ctladm inject 0:6 -i custom -p tur -s 18 "f0 0 02 s12 04 02" .Ed .Pp Inject a custom error on LUN 6 for the next TEST UNIT READY command only. This will result in a sense key of NOT READY (0x02), and an ASC/ASCQ of 0x04,0x02 ("Logical unit not ready, initializing command required"). .Sh SEE ALSO .Xr cam 3 , .Xr cam_cdbparse 3 , .Xr cam 4 , .Xr ctl 4 , .Xr xpt 4 , .Xr camcontrol 8 , .Xr ctld 8 .Sh HISTORY The .Nm utility was originally written during the Winter/Spring of 2003 as an interface to CTL. .Sh AUTHORS .An Ken Merry Aq Mt ken@FreeBSD.org