| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * scsi_error.c Copyright (C) 1997 Eric Youngdale |
| 4 | * |
| 5 | * SCSI error/timeout handling |
| 6 | * Initial versions: Eric Youngdale. Based upon conversations with |
| 7 | * Leonard Zubkoff and David Miller at Linux Expo, |
| 8 | * ideas originating from all over the place. |
| 9 | * |
| 10 | * Restructured scsi_unjam_host and associated functions. |
| 11 | * September 04, 2002 Mike Anderson (andmike@us.ibm.com) |
| 12 | * |
| 13 | * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and |
| 14 | * minor cleanups. |
| 15 | * September 30, 2002 Mike Anderson (andmike@us.ibm.com) |
| 16 | */ |
| 17 | |
| 18 | #include <linux/module.h> |
| 19 | #include <linux/sched.h> |
| 20 | #include <linux/gfp.h> |
| 21 | #include <linux/timer.h> |
| 22 | #include <linux/string.h> |
| 23 | #include <linux/kernel.h> |
| 24 | #include <linux/freezer.h> |
| 25 | #include <linux/kthread.h> |
| 26 | #include <linux/interrupt.h> |
| 27 | #include <linux/blkdev.h> |
| 28 | #include <linux/delay.h> |
| 29 | #include <linux/jiffies.h> |
| 30 | |
| 31 | #include <scsi/scsi.h> |
| 32 | #include <scsi/scsi_cmnd.h> |
| 33 | #include <scsi/scsi_dbg.h> |
| 34 | #include <scsi/scsi_device.h> |
| 35 | #include <scsi/scsi_driver.h> |
| 36 | #include <scsi/scsi_eh.h> |
| 37 | #include <scsi/scsi_common.h> |
| 38 | #include <scsi/scsi_transport.h> |
| 39 | #include <scsi/scsi_host.h> |
| 40 | #include <scsi/scsi_ioctl.h> |
| 41 | #include <scsi/scsi_dh.h> |
| 42 | #include <scsi/scsi_devinfo.h> |
| 43 | #include <scsi/sg.h> |
| 44 | |
| 45 | #include "scsi_priv.h" |
| 46 | #include "scsi_logging.h" |
| 47 | #include "scsi_transport_api.h" |
| 48 | |
| 49 | #include <trace/events/scsi.h> |
| 50 | |
| 51 | #include <linux/unaligned.h> |
| 52 | |
| 53 | /* |
| 54 | * These should *probably* be handled by the host itself. |
| 55 | * Since it is allowed to sleep, it probably should. |
| 56 | */ |
| 57 | #define BUS_RESET_SETTLE_TIME (10) |
| 58 | #define HOST_RESET_SETTLE_TIME (10) |
| 59 | |
| 60 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd); |
| 61 | static enum scsi_disposition scsi_try_to_abort_cmd(const struct scsi_host_template *, |
| 62 | struct scsi_cmnd *); |
| 63 | |
| 64 | void scsi_eh_wakeup(struct Scsi_Host *shost, unsigned int busy) |
| 65 | { |
| 66 | lockdep_assert_held(shost->host_lock); |
| 67 | |
| 68 | if (busy == shost->host_failed) { |
| 69 | trace_scsi_eh_wakeup(shost); |
| 70 | wake_up_process(tsk: shost->ehandler); |
| 71 | SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, |
| 72 | "Waking error handler thread\n" )); |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | /** |
| 77 | * scsi_schedule_eh - schedule EH for SCSI host |
| 78 | * @shost: SCSI host to invoke error handling on. |
| 79 | * |
| 80 | * Schedule SCSI EH without scmd. |
| 81 | */ |
| 82 | void scsi_schedule_eh(struct Scsi_Host *shost) |
| 83 | { |
| 84 | unsigned long flags; |
| 85 | |
| 86 | spin_lock_irqsave(shost->host_lock, flags); |
| 87 | |
| 88 | if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || |
| 89 | scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { |
| 90 | shost->host_eh_scheduled++; |
| 91 | scsi_eh_wakeup(shost, busy: scsi_host_busy(shost)); |
| 92 | } |
| 93 | |
| 94 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 95 | } |
| 96 | EXPORT_SYMBOL_GPL(scsi_schedule_eh); |
| 97 | |
| 98 | static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) |
| 99 | { |
| 100 | if (!shost->last_reset || shost->eh_deadline == -1) |
| 101 | return 0; |
| 102 | |
| 103 | /* |
| 104 | * 32bit accesses are guaranteed to be atomic |
| 105 | * (on all supported architectures), so instead |
| 106 | * of using a spinlock we can as well double check |
| 107 | * if eh_deadline has been set to 'off' during the |
| 108 | * time_before call. |
| 109 | */ |
| 110 | if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && |
| 111 | shost->eh_deadline > -1) |
| 112 | return 0; |
| 113 | |
| 114 | return 1; |
| 115 | } |
| 116 | |
| 117 | static bool scsi_cmd_retry_allowed(struct scsi_cmnd *cmd) |
| 118 | { |
| 119 | if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) |
| 120 | return true; |
| 121 | |
| 122 | return ++cmd->retries <= cmd->allowed; |
| 123 | } |
| 124 | |
| 125 | static bool scsi_eh_should_retry_cmd(struct scsi_cmnd *cmd) |
| 126 | { |
| 127 | struct scsi_device *sdev = cmd->device; |
| 128 | struct Scsi_Host *host = sdev->host; |
| 129 | |
| 130 | if (host->hostt->eh_should_retry_cmd) |
| 131 | return host->hostt->eh_should_retry_cmd(cmd); |
| 132 | |
| 133 | return true; |
| 134 | } |
| 135 | |
| 136 | /** |
| 137 | * scmd_eh_abort_handler - Handle command aborts |
| 138 | * @work: command to be aborted. |
| 139 | * |
| 140 | * Note: this function must be called only for a command that has timed out. |
| 141 | * Because the block layer marks a request as complete before it calls |
| 142 | * scsi_timeout(), a .scsi_done() call from the LLD for a command that has |
| 143 | * timed out do not have any effect. Hence it is safe to call |
| 144 | * scsi_finish_command() from this function. |
| 145 | */ |
| 146 | void |
| 147 | scmd_eh_abort_handler(struct work_struct *work) |
| 148 | { |
| 149 | struct scsi_cmnd *scmd = |
| 150 | container_of(work, struct scsi_cmnd, abort_work.work); |
| 151 | struct scsi_device *sdev = scmd->device; |
| 152 | struct Scsi_Host *shost = sdev->host; |
| 153 | enum scsi_disposition rtn; |
| 154 | unsigned long flags; |
| 155 | |
| 156 | if (scsi_host_eh_past_deadline(shost)) { |
| 157 | SCSI_LOG_ERROR_RECOVERY(3, |
| 158 | scmd_printk(KERN_INFO, scmd, |
| 159 | "eh timeout, not aborting\n" )); |
| 160 | goto out; |
| 161 | } |
| 162 | |
| 163 | SCSI_LOG_ERROR_RECOVERY(3, |
| 164 | scmd_printk(KERN_INFO, scmd, |
| 165 | "aborting command\n" )); |
| 166 | rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); |
| 167 | if (rtn != SUCCESS) { |
| 168 | SCSI_LOG_ERROR_RECOVERY(3, |
| 169 | scmd_printk(KERN_INFO, scmd, |
| 170 | "cmd abort %s\n" , |
| 171 | (rtn == FAST_IO_FAIL) ? |
| 172 | "not send" : "failed" )); |
| 173 | goto out; |
| 174 | } |
| 175 | set_host_byte(cmd: scmd, status: DID_TIME_OUT); |
| 176 | if (scsi_host_eh_past_deadline(shost)) { |
| 177 | SCSI_LOG_ERROR_RECOVERY(3, |
| 178 | scmd_printk(KERN_INFO, scmd, |
| 179 | "eh timeout, not retrying " |
| 180 | "aborted command\n" )); |
| 181 | goto out; |
| 182 | } |
| 183 | |
| 184 | spin_lock_irqsave(shost->host_lock, flags); |
| 185 | list_del_init(entry: &scmd->eh_entry); |
| 186 | |
| 187 | /* |
| 188 | * If the abort succeeds, and there is no further |
| 189 | * EH action, clear the ->last_reset time. |
| 190 | */ |
| 191 | if (list_empty(head: &shost->eh_abort_list) && |
| 192 | list_empty(head: &shost->eh_cmd_q)) |
| 193 | if (shost->eh_deadline != -1) |
| 194 | shost->last_reset = 0; |
| 195 | |
| 196 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 197 | |
| 198 | if (!scsi_noretry_cmd(scmd) && |
| 199 | scsi_cmd_retry_allowed(cmd: scmd) && |
| 200 | scsi_eh_should_retry_cmd(cmd: scmd)) { |
| 201 | SCSI_LOG_ERROR_RECOVERY(3, |
| 202 | scmd_printk(KERN_WARNING, scmd, |
| 203 | "retry aborted command\n" )); |
| 204 | scsi_queue_insert(cmd: scmd, SCSI_MLQUEUE_EH_RETRY); |
| 205 | } else { |
| 206 | SCSI_LOG_ERROR_RECOVERY(3, |
| 207 | scmd_printk(KERN_WARNING, scmd, |
| 208 | "finish aborted command\n" )); |
| 209 | scsi_finish_command(cmd: scmd); |
| 210 | } |
| 211 | return; |
| 212 | |
| 213 | out: |
| 214 | spin_lock_irqsave(shost->host_lock, flags); |
| 215 | list_del_init(entry: &scmd->eh_entry); |
| 216 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 217 | |
| 218 | scsi_eh_scmd_add(scmd); |
| 219 | } |
| 220 | |
| 221 | /** |
| 222 | * scsi_abort_command - schedule a command abort |
| 223 | * @scmd: scmd to abort. |
| 224 | * |
| 225 | * We only need to abort commands after a command timeout |
| 226 | */ |
| 227 | static int |
| 228 | scsi_abort_command(struct scsi_cmnd *scmd) |
| 229 | { |
| 230 | struct scsi_device *sdev = scmd->device; |
| 231 | struct Scsi_Host *shost = sdev->host; |
| 232 | unsigned long flags; |
| 233 | |
| 234 | if (!shost->hostt->eh_abort_handler) { |
| 235 | /* No abort handler, fail command directly */ |
| 236 | return FAILED; |
| 237 | } |
| 238 | |
| 239 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { |
| 240 | /* |
| 241 | * Retry after abort failed, escalate to next level. |
| 242 | */ |
| 243 | SCSI_LOG_ERROR_RECOVERY(3, |
| 244 | scmd_printk(KERN_INFO, scmd, |
| 245 | "previous abort failed\n" )); |
| 246 | BUG_ON(delayed_work_pending(&scmd->abort_work)); |
| 247 | return FAILED; |
| 248 | } |
| 249 | |
| 250 | spin_lock_irqsave(shost->host_lock, flags); |
| 251 | if (shost->eh_deadline != -1 && !shost->last_reset) |
| 252 | shost->last_reset = jiffies; |
| 253 | BUG_ON(!list_empty(&scmd->eh_entry)); |
| 254 | list_add_tail(new: &scmd->eh_entry, head: &shost->eh_abort_list); |
| 255 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 256 | |
| 257 | scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; |
| 258 | SCSI_LOG_ERROR_RECOVERY(3, |
| 259 | scmd_printk(KERN_INFO, scmd, "abort scheduled\n" )); |
| 260 | queue_delayed_work(wq: shost->tmf_work_q, dwork: &scmd->abort_work, HZ / 100); |
| 261 | return SUCCESS; |
| 262 | } |
| 263 | |
| 264 | /** |
| 265 | * scsi_eh_reset - call into ->eh_action to reset internal counters |
| 266 | * @scmd: scmd to run eh on. |
| 267 | * |
| 268 | * The scsi driver might be carrying internal state about the |
| 269 | * devices, so we need to call into the driver to reset the |
| 270 | * internal state once the error handler is started. |
| 271 | */ |
| 272 | static void scsi_eh_reset(struct scsi_cmnd *scmd) |
| 273 | { |
| 274 | if (!blk_rq_is_passthrough(rq: scsi_cmd_to_rq(scmd))) { |
| 275 | struct scsi_driver *sdrv = scsi_cmd_to_driver(cmd: scmd); |
| 276 | if (sdrv->eh_reset) |
| 277 | sdrv->eh_reset(scmd); |
| 278 | } |
| 279 | } |
| 280 | |
| 281 | static void scsi_eh_inc_host_failed(struct rcu_head *head) |
| 282 | { |
| 283 | struct scsi_cmnd *scmd = container_of(head, typeof(*scmd), rcu); |
| 284 | struct Scsi_Host *shost = scmd->device->host; |
| 285 | unsigned int busy; |
| 286 | unsigned long flags; |
| 287 | |
| 288 | spin_lock_irqsave(shost->host_lock, flags); |
| 289 | shost->host_failed++; |
| 290 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 291 | /* |
| 292 | * The counting of busy requests needs to occur after adding to |
| 293 | * host_failed or after the lock acquire for adding to host_failed |
| 294 | * to prevent a race with host unbusy and missing an eh wakeup. |
| 295 | */ |
| 296 | busy = scsi_host_busy(shost); |
| 297 | |
| 298 | spin_lock_irqsave(shost->host_lock, flags); |
| 299 | scsi_eh_wakeup(shost, busy); |
| 300 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 301 | } |
| 302 | |
| 303 | /** |
| 304 | * scsi_eh_scmd_add - add scsi cmd to error handling. |
| 305 | * @scmd: scmd to run eh on. |
| 306 | */ |
| 307 | void scsi_eh_scmd_add(struct scsi_cmnd *scmd) |
| 308 | { |
| 309 | struct Scsi_Host *shost = scmd->device->host; |
| 310 | unsigned long flags; |
| 311 | int ret; |
| 312 | |
| 313 | WARN_ON_ONCE(!shost->ehandler); |
| 314 | WARN_ON_ONCE(!test_bit(SCMD_STATE_INFLIGHT, &scmd->state)); |
| 315 | |
| 316 | spin_lock_irqsave(shost->host_lock, flags); |
| 317 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) { |
| 318 | ret = scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY); |
| 319 | WARN_ON_ONCE(ret); |
| 320 | } |
| 321 | if (shost->eh_deadline != -1 && !shost->last_reset) |
| 322 | shost->last_reset = jiffies; |
| 323 | |
| 324 | scsi_eh_reset(scmd); |
| 325 | list_add_tail(new: &scmd->eh_entry, head: &shost->eh_cmd_q); |
| 326 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 327 | /* |
| 328 | * Ensure that all tasks observe the host state change before the |
| 329 | * host_failed change. |
| 330 | */ |
| 331 | call_rcu_hurry(head: &scmd->rcu, func: scsi_eh_inc_host_failed); |
| 332 | } |
| 333 | |
| 334 | /** |
| 335 | * scsi_timeout - Timeout function for normal scsi commands. |
| 336 | * @req: request that is timing out. |
| 337 | * |
| 338 | * Notes: |
| 339 | * We do not need to lock this. There is the potential for a race |
| 340 | * only in that the normal completion handling might run, but if the |
| 341 | * normal completion function determines that the timer has already |
| 342 | * fired, then it mustn't do anything. |
| 343 | */ |
| 344 | enum blk_eh_timer_return scsi_timeout(struct request *req) |
| 345 | { |
| 346 | struct scsi_cmnd *scmd = blk_mq_rq_to_pdu(rq: req); |
| 347 | struct Scsi_Host *host = scmd->device->host; |
| 348 | |
| 349 | trace_scsi_dispatch_cmd_timeout(cmd: scmd); |
| 350 | scsi_log_completion(cmd: scmd, disposition: TIMEOUT_ERROR); |
| 351 | |
| 352 | atomic_inc(v: &scmd->device->iotmo_cnt); |
| 353 | if (host->eh_deadline != -1 && !host->last_reset) |
| 354 | host->last_reset = jiffies; |
| 355 | |
| 356 | if (host->hostt->eh_timed_out) { |
| 357 | switch (host->hostt->eh_timed_out(scmd)) { |
| 358 | case SCSI_EH_DONE: |
| 359 | return BLK_EH_DONE; |
| 360 | case SCSI_EH_RESET_TIMER: |
| 361 | return BLK_EH_RESET_TIMER; |
| 362 | case SCSI_EH_NOT_HANDLED: |
| 363 | break; |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | /* |
| 368 | * If scsi_done() has already set SCMD_STATE_COMPLETE, do not modify |
| 369 | * *scmd. |
| 370 | */ |
| 371 | if (test_and_set_bit(SCMD_STATE_COMPLETE, addr: &scmd->state)) |
| 372 | return BLK_EH_DONE; |
| 373 | atomic_inc(v: &scmd->device->iodone_cnt); |
| 374 | if (scsi_abort_command(scmd) != SUCCESS) { |
| 375 | set_host_byte(cmd: scmd, status: DID_TIME_OUT); |
| 376 | scsi_eh_scmd_add(scmd); |
| 377 | } |
| 378 | |
| 379 | return BLK_EH_DONE; |
| 380 | } |
| 381 | |
| 382 | /** |
| 383 | * scsi_block_when_processing_errors - Prevent cmds from being queued. |
| 384 | * @sdev: Device on which we are performing recovery. |
| 385 | * |
| 386 | * Description: |
| 387 | * We block until the host is out of error recovery, and then check to |
| 388 | * see whether the host or the device is offline. |
| 389 | * |
| 390 | * Return value: |
| 391 | * 0 when dev was taken offline by error recovery. 1 OK to proceed. |
| 392 | */ |
| 393 | int scsi_block_when_processing_errors(struct scsi_device *sdev) |
| 394 | { |
| 395 | int online; |
| 396 | |
| 397 | wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); |
| 398 | |
| 399 | online = scsi_device_online(sdev); |
| 400 | |
| 401 | return online; |
| 402 | } |
| 403 | EXPORT_SYMBOL(scsi_block_when_processing_errors); |
| 404 | |
| 405 | #ifdef CONFIG_SCSI_LOGGING |
| 406 | /** |
| 407 | * scsi_eh_prt_fail_stats - Log info on failures. |
| 408 | * @shost: scsi host being recovered. |
| 409 | * @work_q: Queue of scsi cmds to process. |
| 410 | */ |
| 411 | static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, |
| 412 | struct list_head *work_q) |
| 413 | { |
| 414 | struct scsi_cmnd *scmd; |
| 415 | struct scsi_device *sdev; |
| 416 | int total_failures = 0; |
| 417 | int cmd_failed = 0; |
| 418 | int cmd_cancel = 0; |
| 419 | int devices_failed = 0; |
| 420 | |
| 421 | shost_for_each_device(sdev, shost) { |
| 422 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 423 | if (scmd->device == sdev) { |
| 424 | ++total_failures; |
| 425 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) |
| 426 | ++cmd_cancel; |
| 427 | else |
| 428 | ++cmd_failed; |
| 429 | } |
| 430 | } |
| 431 | |
| 432 | if (cmd_cancel || cmd_failed) { |
| 433 | SCSI_LOG_ERROR_RECOVERY(3, |
| 434 | shost_printk(KERN_INFO, shost, |
| 435 | "%s: cmds failed: %d, cancel: %d\n" , |
| 436 | __func__, cmd_failed, |
| 437 | cmd_cancel)); |
| 438 | cmd_cancel = 0; |
| 439 | cmd_failed = 0; |
| 440 | ++devices_failed; |
| 441 | } |
| 442 | } |
| 443 | |
| 444 | SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, |
| 445 | "Total of %d commands on %d" |
| 446 | " devices require eh work\n" , |
| 447 | total_failures, devices_failed)); |
| 448 | } |
| 449 | #endif |
| 450 | |
| 451 | /** |
| 452 | * scsi_report_lun_change - Set flag on all *other* devices on the same target |
| 453 | * to indicate that a UNIT ATTENTION is expected. |
| 454 | * @sdev: Device reporting the UNIT ATTENTION |
| 455 | */ |
| 456 | static void scsi_report_lun_change(struct scsi_device *sdev) |
| 457 | { |
| 458 | sdev->sdev_target->expecting_lun_change = 1; |
| 459 | } |
| 460 | |
| 461 | /** |
| 462 | * scsi_report_sense - Examine scsi sense information and log messages for |
| 463 | * certain conditions, also issue uevents for some of them. |
| 464 | * @sdev: Device reporting the sense code |
| 465 | * @sshdr: sshdr to be examined |
| 466 | */ |
| 467 | static void scsi_report_sense(struct scsi_device *sdev, |
| 468 | struct scsi_sense_hdr *sshdr) |
| 469 | { |
| 470 | enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ |
| 471 | |
| 472 | if (sshdr->sense_key == UNIT_ATTENTION) { |
| 473 | if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { |
| 474 | evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; |
| 475 | sdev_printk(KERN_WARNING, sdev, |
| 476 | "Inquiry data has changed" ); |
| 477 | } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { |
| 478 | evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; |
| 479 | scsi_report_lun_change(sdev); |
| 480 | sdev_printk(KERN_WARNING, sdev, |
| 481 | "LUN assignments on this target have " |
| 482 | "changed. The Linux SCSI layer does not " |
| 483 | "automatically remap LUN assignments.\n" ); |
| 484 | } else if (sshdr->asc == 0x3f) |
| 485 | sdev_printk(KERN_WARNING, sdev, |
| 486 | "Operating parameters on this target have " |
| 487 | "changed. The Linux SCSI layer does not " |
| 488 | "automatically adjust these parameters.\n" ); |
| 489 | |
| 490 | if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { |
| 491 | evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; |
| 492 | sdev_printk(KERN_WARNING, sdev, |
| 493 | "Warning! Received an indication that the " |
| 494 | "LUN reached a thin provisioning soft " |
| 495 | "threshold.\n" ); |
| 496 | } |
| 497 | |
| 498 | if (sshdr->asc == 0x29) { |
| 499 | evt_type = SDEV_EVT_POWER_ON_RESET_OCCURRED; |
| 500 | /* |
| 501 | * Do not print message if it is an expected side-effect |
| 502 | * of runtime PM. |
| 503 | */ |
| 504 | if (!sdev->silence_suspend) |
| 505 | sdev_printk(KERN_WARNING, sdev, |
| 506 | "Power-on or device reset occurred\n" ); |
| 507 | } |
| 508 | |
| 509 | if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { |
| 510 | evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; |
| 511 | sdev_printk(KERN_WARNING, sdev, |
| 512 | "Mode parameters changed" ); |
| 513 | } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { |
| 514 | evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; |
| 515 | sdev_printk(KERN_WARNING, sdev, |
| 516 | "Asymmetric access state changed" ); |
| 517 | } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { |
| 518 | evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; |
| 519 | sdev_printk(KERN_WARNING, sdev, |
| 520 | "Capacity data has changed" ); |
| 521 | } else if (sshdr->asc == 0x2a) |
| 522 | sdev_printk(KERN_WARNING, sdev, |
| 523 | "Parameters changed" ); |
| 524 | } |
| 525 | |
| 526 | if (evt_type != SDEV_EVT_MAXBITS) { |
| 527 | set_bit(nr: evt_type, addr: sdev->pending_events); |
| 528 | schedule_work(work: &sdev->event_work); |
| 529 | } |
| 530 | } |
| 531 | |
| 532 | static inline void set_scsi_ml_byte(struct scsi_cmnd *cmd, u8 status) |
| 533 | { |
| 534 | cmd->result = (cmd->result & 0xffff00ff) | (status << 8); |
| 535 | } |
| 536 | |
| 537 | /** |
| 538 | * scsi_check_sense - Examine scsi cmd sense |
| 539 | * @scmd: Cmd to have sense checked. |
| 540 | * |
| 541 | * Return value: |
| 542 | * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE |
| 543 | * |
| 544 | * Notes: |
| 545 | * When a deferred error is detected the current command has |
| 546 | * not been executed and needs retrying. |
| 547 | */ |
| 548 | enum scsi_disposition scsi_check_sense(struct scsi_cmnd *scmd) |
| 549 | { |
| 550 | struct request *req = scsi_cmd_to_rq(scmd); |
| 551 | struct scsi_device *sdev = scmd->device; |
| 552 | struct scsi_sense_hdr sshdr; |
| 553 | |
| 554 | if (! scsi_command_normalize_sense(cmd: scmd, sshdr: &sshdr)) |
| 555 | return FAILED; /* no valid sense data */ |
| 556 | |
| 557 | scsi_report_sense(sdev, sshdr: &sshdr); |
| 558 | |
| 559 | if (sshdr.sense_key == UNIT_ATTENTION) { |
| 560 | /* |
| 561 | * Increment the counters for Power on/Reset or New Media so |
| 562 | * that all ULDs interested in these can see that those have |
| 563 | * happened, even if someone else gets the sense data. |
| 564 | */ |
| 565 | if (sshdr.asc == 0x28) |
| 566 | atomic_inc(v: &sdev->ua_new_media_ctr); |
| 567 | else if (sshdr.asc == 0x29) |
| 568 | atomic_inc(v: &sdev->ua_por_ctr); |
| 569 | } |
| 570 | |
| 571 | if (scsi_sense_is_deferred(sshdr: &sshdr)) |
| 572 | return NEEDS_RETRY; |
| 573 | |
| 574 | if (sdev->handler && sdev->handler->check_sense) { |
| 575 | enum scsi_disposition rc; |
| 576 | |
| 577 | rc = sdev->handler->check_sense(sdev, &sshdr); |
| 578 | if (rc != SCSI_RETURN_NOT_HANDLED) |
| 579 | return rc; |
| 580 | /* handler does not care. Drop down to default handling */ |
| 581 | } |
| 582 | |
| 583 | if (scmd->cmnd[0] == TEST_UNIT_READY && |
| 584 | scmd->submitter != SUBMITTED_BY_SCSI_ERROR_HANDLER) |
| 585 | /* |
| 586 | * nasty: for mid-layer issued TURs, we need to return the |
| 587 | * actual sense data without any recovery attempt. For eh |
| 588 | * issued ones, we need to try to recover and interpret |
| 589 | */ |
| 590 | return SUCCESS; |
| 591 | |
| 592 | /* |
| 593 | * Previous logic looked for FILEMARK, EOM or ILI which are |
| 594 | * mainly associated with tapes and returned SUCCESS. |
| 595 | */ |
| 596 | if (sshdr.response_code == 0x70) { |
| 597 | /* fixed format */ |
| 598 | if (scmd->sense_buffer[2] & 0xe0) |
| 599 | return SUCCESS; |
| 600 | } else { |
| 601 | /* |
| 602 | * descriptor format: look for "stream commands sense data |
| 603 | * descriptor" (see SSC-3). Assume single sense data |
| 604 | * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. |
| 605 | */ |
| 606 | if ((sshdr.additional_length > 3) && |
| 607 | (scmd->sense_buffer[8] == 0x4) && |
| 608 | (scmd->sense_buffer[11] & 0xe0)) |
| 609 | return SUCCESS; |
| 610 | } |
| 611 | |
| 612 | switch (sshdr.sense_key) { |
| 613 | case NO_SENSE: |
| 614 | return SUCCESS; |
| 615 | case RECOVERED_ERROR: |
| 616 | return /* soft_error */ SUCCESS; |
| 617 | |
| 618 | case ABORTED_COMMAND: |
| 619 | if (sshdr.asc == 0x10) /* DIF */ |
| 620 | return SUCCESS; |
| 621 | |
| 622 | /* |
| 623 | * Check aborts due to command duration limit policy: |
| 624 | * ABORTED COMMAND additional sense code with the |
| 625 | * COMMAND TIMEOUT BEFORE PROCESSING or |
| 626 | * COMMAND TIMEOUT DURING PROCESSING or |
| 627 | * COMMAND TIMEOUT DURING PROCESSING DUE TO ERROR RECOVERY |
| 628 | * additional sense code qualifiers. |
| 629 | */ |
| 630 | if (sshdr.asc == 0x2e && |
| 631 | sshdr.ascq >= 0x01 && sshdr.ascq <= 0x03) { |
| 632 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_DL_TIMEOUT); |
| 633 | req->cmd_flags |= REQ_FAILFAST_DEV; |
| 634 | req->rq_flags |= RQF_QUIET; |
| 635 | return SUCCESS; |
| 636 | } |
| 637 | |
| 638 | if (sshdr.asc == 0x44 && sdev->sdev_bflags & BLIST_RETRY_ITF) |
| 639 | return ADD_TO_MLQUEUE; |
| 640 | if (sshdr.asc == 0xc1 && sshdr.ascq == 0x01 && |
| 641 | sdev->sdev_bflags & BLIST_RETRY_ASC_C1) |
| 642 | return ADD_TO_MLQUEUE; |
| 643 | |
| 644 | return NEEDS_RETRY; |
| 645 | case NOT_READY: |
| 646 | case UNIT_ATTENTION: |
| 647 | /* |
| 648 | * if we are expecting a cc/ua because of a bus reset that we |
| 649 | * performed, treat this just as a retry. otherwise this is |
| 650 | * information that we should pass up to the upper-level driver |
| 651 | * so that we can deal with it there. |
| 652 | */ |
| 653 | if (scmd->device->expecting_cc_ua) { |
| 654 | /* |
| 655 | * Because some device does not queue unit |
| 656 | * attentions correctly, we carefully check |
| 657 | * additional sense code and qualifier so as |
| 658 | * not to squash media change unit attention. |
| 659 | */ |
| 660 | if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { |
| 661 | scmd->device->expecting_cc_ua = 0; |
| 662 | return NEEDS_RETRY; |
| 663 | } |
| 664 | } |
| 665 | /* |
| 666 | * we might also expect a cc/ua if another LUN on the target |
| 667 | * reported a UA with an ASC/ASCQ of 3F 0E - |
| 668 | * REPORTED LUNS DATA HAS CHANGED. |
| 669 | */ |
| 670 | if (scmd->device->sdev_target->expecting_lun_change && |
| 671 | sshdr.asc == 0x3f && sshdr.ascq == 0x0e) |
| 672 | return NEEDS_RETRY; |
| 673 | /* |
| 674 | * if the device is in the process of becoming ready, we |
| 675 | * should retry. |
| 676 | */ |
| 677 | if ((sshdr.asc == 0x04) && |
| 678 | (sshdr.ascq == 0x01 || sshdr.ascq == 0x0a)) |
| 679 | return NEEDS_RETRY; |
| 680 | /* |
| 681 | * if the device is not started, we need to wake |
| 682 | * the error handler to start the motor |
| 683 | */ |
| 684 | if (scmd->device->allow_restart && |
| 685 | (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) |
| 686 | return FAILED; |
| 687 | /* |
| 688 | * Pass the UA upwards for a determination in the completion |
| 689 | * functions. |
| 690 | */ |
| 691 | return SUCCESS; |
| 692 | |
| 693 | /* these are not supported */ |
| 694 | case DATA_PROTECT: |
| 695 | if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { |
| 696 | /* Thin provisioning hard threshold reached */ |
| 697 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_NOSPC); |
| 698 | return SUCCESS; |
| 699 | } |
| 700 | fallthrough; |
| 701 | case COPY_ABORTED: |
| 702 | case VOLUME_OVERFLOW: |
| 703 | case MISCOMPARE: |
| 704 | case BLANK_CHECK: |
| 705 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_TGT_FAILURE); |
| 706 | return SUCCESS; |
| 707 | |
| 708 | case MEDIUM_ERROR: |
| 709 | if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ |
| 710 | sshdr.asc == 0x13 || /* AMNF DATA FIELD */ |
| 711 | sshdr.asc == 0x14) { /* RECORD NOT FOUND */ |
| 712 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_MED_ERROR); |
| 713 | return SUCCESS; |
| 714 | } |
| 715 | return NEEDS_RETRY; |
| 716 | |
| 717 | case HARDWARE_ERROR: |
| 718 | if (scmd->device->retry_hwerror) |
| 719 | return ADD_TO_MLQUEUE; |
| 720 | else |
| 721 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_TGT_FAILURE); |
| 722 | fallthrough; |
| 723 | |
| 724 | case ILLEGAL_REQUEST: |
| 725 | if (sshdr.asc == 0x20 || /* Invalid command operation code */ |
| 726 | sshdr.asc == 0x21 || /* Logical block address out of range */ |
| 727 | sshdr.asc == 0x22 || /* Invalid function */ |
| 728 | sshdr.asc == 0x24 || /* Invalid field in cdb */ |
| 729 | sshdr.asc == 0x26 || /* Parameter value invalid */ |
| 730 | sshdr.asc == 0x27) { /* Write protected */ |
| 731 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_TGT_FAILURE); |
| 732 | } |
| 733 | return SUCCESS; |
| 734 | |
| 735 | case COMPLETED: |
| 736 | /* |
| 737 | * A command using command duration limits (CDL) with a |
| 738 | * descriptor set with policy 0xD may be completed with success |
| 739 | * and the sense data DATA CURRENTLY UNAVAILABLE, indicating |
| 740 | * that the command was in fact aborted because it exceeded its |
| 741 | * duration limit. Never retry these commands. |
| 742 | */ |
| 743 | if (sshdr.asc == 0x55 && sshdr.ascq == 0x0a) { |
| 744 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_DL_TIMEOUT); |
| 745 | req->cmd_flags |= REQ_FAILFAST_DEV; |
| 746 | req->rq_flags |= RQF_QUIET; |
| 747 | } |
| 748 | return SUCCESS; |
| 749 | |
| 750 | default: |
| 751 | return SUCCESS; |
| 752 | } |
| 753 | } |
| 754 | EXPORT_SYMBOL_GPL(scsi_check_sense); |
| 755 | |
| 756 | static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) |
| 757 | { |
| 758 | const struct scsi_host_template *sht = sdev->host->hostt; |
| 759 | struct scsi_device *tmp_sdev; |
| 760 | |
| 761 | if (!sdev->budget_map.map) |
| 762 | return; |
| 763 | |
| 764 | if (!sht->track_queue_depth || |
| 765 | sdev->queue_depth >= sdev->max_queue_depth) |
| 766 | return; |
| 767 | |
| 768 | if (time_before(jiffies, |
| 769 | sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) |
| 770 | return; |
| 771 | |
| 772 | if (time_before(jiffies, |
| 773 | sdev->last_queue_full_time + sdev->queue_ramp_up_period)) |
| 774 | return; |
| 775 | |
| 776 | /* |
| 777 | * Walk all devices of a target and do |
| 778 | * ramp up on them. |
| 779 | */ |
| 780 | shost_for_each_device(tmp_sdev, sdev->host) { |
| 781 | if (tmp_sdev->channel != sdev->channel || |
| 782 | tmp_sdev->id != sdev->id || |
| 783 | tmp_sdev->queue_depth == sdev->max_queue_depth) |
| 784 | continue; |
| 785 | |
| 786 | scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); |
| 787 | sdev->last_queue_ramp_up = jiffies; |
| 788 | } |
| 789 | } |
| 790 | |
| 791 | static void scsi_handle_queue_full(struct scsi_device *sdev) |
| 792 | { |
| 793 | const struct scsi_host_template *sht = sdev->host->hostt; |
| 794 | struct scsi_device *tmp_sdev; |
| 795 | |
| 796 | if (!sht->track_queue_depth) |
| 797 | return; |
| 798 | |
| 799 | shost_for_each_device(tmp_sdev, sdev->host) { |
| 800 | if (tmp_sdev->channel != sdev->channel || |
| 801 | tmp_sdev->id != sdev->id) |
| 802 | continue; |
| 803 | /* |
| 804 | * We do not know the number of commands that were at |
| 805 | * the device when we got the queue full so we start |
| 806 | * from the highest possible value and work our way down. |
| 807 | */ |
| 808 | scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | /** |
| 813 | * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. |
| 814 | * @scmd: SCSI cmd to examine. |
| 815 | * |
| 816 | * Notes: |
| 817 | * This is *only* called when we are examining the status of commands |
| 818 | * queued during error recovery. the main difference here is that we |
| 819 | * don't allow for the possibility of retries here, and we are a lot |
| 820 | * more restrictive about what we consider acceptable. |
| 821 | */ |
| 822 | static enum scsi_disposition scsi_eh_completed_normally(struct scsi_cmnd *scmd) |
| 823 | { |
| 824 | /* |
| 825 | * first check the host byte, to see if there is anything in there |
| 826 | * that would indicate what we need to do. |
| 827 | */ |
| 828 | if (host_byte(scmd->result) == DID_RESET) { |
| 829 | /* |
| 830 | * rats. we are already in the error handler, so we now |
| 831 | * get to try and figure out what to do next. if the sense |
| 832 | * is valid, we have a pretty good idea of what to do. |
| 833 | * if not, we mark it as FAILED. |
| 834 | */ |
| 835 | return scsi_check_sense(scmd); |
| 836 | } |
| 837 | if (host_byte(scmd->result) != DID_OK) |
| 838 | return FAILED; |
| 839 | |
| 840 | /* |
| 841 | * now, check the status byte to see if this indicates |
| 842 | * anything special. |
| 843 | */ |
| 844 | switch (get_status_byte(cmd: scmd)) { |
| 845 | case SAM_STAT_GOOD: |
| 846 | scsi_handle_queue_ramp_up(sdev: scmd->device); |
| 847 | if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) |
| 848 | /* |
| 849 | * If we have sense data, call scsi_check_sense() in |
| 850 | * order to set the correct SCSI ML byte (if any). |
| 851 | * No point in checking the return value, since the |
| 852 | * command has already completed successfully. |
| 853 | */ |
| 854 | scsi_check_sense(scmd); |
| 855 | fallthrough; |
| 856 | case SAM_STAT_COMMAND_TERMINATED: |
| 857 | return SUCCESS; |
| 858 | case SAM_STAT_CHECK_CONDITION: |
| 859 | return scsi_check_sense(scmd); |
| 860 | case SAM_STAT_CONDITION_MET: |
| 861 | case SAM_STAT_INTERMEDIATE: |
| 862 | case SAM_STAT_INTERMEDIATE_CONDITION_MET: |
| 863 | /* |
| 864 | * who knows? FIXME(eric) |
| 865 | */ |
| 866 | return SUCCESS; |
| 867 | case SAM_STAT_RESERVATION_CONFLICT: |
| 868 | if (scmd->cmnd[0] == TEST_UNIT_READY) |
| 869 | /* it is a success, we probed the device and |
| 870 | * found it */ |
| 871 | return SUCCESS; |
| 872 | /* otherwise, we failed to send the command */ |
| 873 | return FAILED; |
| 874 | case SAM_STAT_TASK_SET_FULL: |
| 875 | scsi_handle_queue_full(sdev: scmd->device); |
| 876 | fallthrough; |
| 877 | case SAM_STAT_BUSY: |
| 878 | return NEEDS_RETRY; |
| 879 | default: |
| 880 | return FAILED; |
| 881 | } |
| 882 | return FAILED; |
| 883 | } |
| 884 | |
| 885 | /** |
| 886 | * scsi_eh_done - Completion function for error handling. |
| 887 | * @scmd: Cmd that is done. |
| 888 | */ |
| 889 | void scsi_eh_done(struct scsi_cmnd *scmd) |
| 890 | { |
| 891 | struct completion *eh_action; |
| 892 | |
| 893 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 894 | "%s result: %x\n" , __func__, scmd->result)); |
| 895 | |
| 896 | eh_action = scmd->device->host->eh_action; |
| 897 | if (eh_action) |
| 898 | complete(eh_action); |
| 899 | } |
| 900 | |
| 901 | /** |
| 902 | * scsi_try_host_reset - ask host adapter to reset itself |
| 903 | * @scmd: SCSI cmd to send host reset. |
| 904 | */ |
| 905 | static enum scsi_disposition scsi_try_host_reset(struct scsi_cmnd *scmd) |
| 906 | { |
| 907 | unsigned long flags; |
| 908 | enum scsi_disposition rtn; |
| 909 | struct Scsi_Host *host = scmd->device->host; |
| 910 | const struct scsi_host_template *hostt = host->hostt; |
| 911 | |
| 912 | SCSI_LOG_ERROR_RECOVERY(3, |
| 913 | shost_printk(KERN_INFO, host, "Snd Host RST\n" )); |
| 914 | |
| 915 | if (!hostt->eh_host_reset_handler) |
| 916 | return FAILED; |
| 917 | |
| 918 | rtn = hostt->eh_host_reset_handler(scmd); |
| 919 | |
| 920 | if (rtn == SUCCESS) { |
| 921 | if (!hostt->skip_settle_delay) |
| 922 | ssleep(HOST_RESET_SETTLE_TIME); |
| 923 | spin_lock_irqsave(host->host_lock, flags); |
| 924 | scsi_report_bus_reset(host, scmd_channel(scmd)); |
| 925 | spin_unlock_irqrestore(lock: host->host_lock, flags); |
| 926 | } |
| 927 | |
| 928 | return rtn; |
| 929 | } |
| 930 | |
| 931 | /** |
| 932 | * scsi_try_bus_reset - ask host to perform a bus reset |
| 933 | * @scmd: SCSI cmd to send bus reset. |
| 934 | */ |
| 935 | static enum scsi_disposition scsi_try_bus_reset(struct scsi_cmnd *scmd) |
| 936 | { |
| 937 | unsigned long flags; |
| 938 | enum scsi_disposition rtn; |
| 939 | struct Scsi_Host *host = scmd->device->host; |
| 940 | const struct scsi_host_template *hostt = host->hostt; |
| 941 | |
| 942 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 943 | "%s: Snd Bus RST\n" , __func__)); |
| 944 | |
| 945 | if (!hostt->eh_bus_reset_handler) |
| 946 | return FAILED; |
| 947 | |
| 948 | rtn = hostt->eh_bus_reset_handler(scmd); |
| 949 | |
| 950 | if (rtn == SUCCESS) { |
| 951 | if (!hostt->skip_settle_delay) |
| 952 | ssleep(BUS_RESET_SETTLE_TIME); |
| 953 | spin_lock_irqsave(host->host_lock, flags); |
| 954 | scsi_report_bus_reset(host, scmd_channel(scmd)); |
| 955 | spin_unlock_irqrestore(lock: host->host_lock, flags); |
| 956 | } |
| 957 | |
| 958 | return rtn; |
| 959 | } |
| 960 | |
| 961 | static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) |
| 962 | { |
| 963 | sdev->was_reset = 1; |
| 964 | sdev->expecting_cc_ua = 1; |
| 965 | } |
| 966 | |
| 967 | /** |
| 968 | * scsi_try_target_reset - Ask host to perform a target reset |
| 969 | * @scmd: SCSI cmd used to send a target reset |
| 970 | * |
| 971 | * Notes: |
| 972 | * There is no timeout for this operation. if this operation is |
| 973 | * unreliable for a given host, then the host itself needs to put a |
| 974 | * timer on it, and set the host back to a consistent state prior to |
| 975 | * returning. |
| 976 | */ |
| 977 | static enum scsi_disposition scsi_try_target_reset(struct scsi_cmnd *scmd) |
| 978 | { |
| 979 | unsigned long flags; |
| 980 | enum scsi_disposition rtn; |
| 981 | struct Scsi_Host *host = scmd->device->host; |
| 982 | const struct scsi_host_template *hostt = host->hostt; |
| 983 | |
| 984 | if (!hostt->eh_target_reset_handler) |
| 985 | return FAILED; |
| 986 | |
| 987 | rtn = hostt->eh_target_reset_handler(scmd); |
| 988 | if (rtn == SUCCESS) { |
| 989 | spin_lock_irqsave(host->host_lock, flags); |
| 990 | __starget_for_each_device(scsi_target(sdev: scmd->device), NULL, |
| 991 | fn: __scsi_report_device_reset); |
| 992 | spin_unlock_irqrestore(lock: host->host_lock, flags); |
| 993 | } |
| 994 | |
| 995 | return rtn; |
| 996 | } |
| 997 | |
| 998 | /** |
| 999 | * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev |
| 1000 | * @scmd: SCSI cmd used to send BDR |
| 1001 | * |
| 1002 | * Notes: |
| 1003 | * There is no timeout for this operation. if this operation is |
| 1004 | * unreliable for a given host, then the host itself needs to put a |
| 1005 | * timer on it, and set the host back to a consistent state prior to |
| 1006 | * returning. |
| 1007 | */ |
| 1008 | static enum scsi_disposition scsi_try_bus_device_reset(struct scsi_cmnd *scmd) |
| 1009 | { |
| 1010 | enum scsi_disposition rtn; |
| 1011 | const struct scsi_host_template *hostt = scmd->device->host->hostt; |
| 1012 | |
| 1013 | if (!hostt->eh_device_reset_handler) |
| 1014 | return FAILED; |
| 1015 | |
| 1016 | rtn = hostt->eh_device_reset_handler(scmd); |
| 1017 | if (rtn == SUCCESS) |
| 1018 | __scsi_report_device_reset(sdev: scmd->device, NULL); |
| 1019 | return rtn; |
| 1020 | } |
| 1021 | |
| 1022 | /** |
| 1023 | * scsi_try_to_abort_cmd - Ask host to abort a SCSI command |
| 1024 | * @hostt: SCSI driver host template |
| 1025 | * @scmd: SCSI cmd used to send a target reset |
| 1026 | * |
| 1027 | * Return value: |
| 1028 | * SUCCESS, FAILED, or FAST_IO_FAIL |
| 1029 | * |
| 1030 | * Notes: |
| 1031 | * SUCCESS does not necessarily indicate that the command |
| 1032 | * has been aborted; it only indicates that the LLDDs |
| 1033 | * has cleared all references to that command. |
| 1034 | * LLDDs should return FAILED only if an abort was required |
| 1035 | * but could not be executed. LLDDs should return FAST_IO_FAIL |
| 1036 | * if the device is temporarily unavailable (eg due to a |
| 1037 | * link down on FibreChannel) |
| 1038 | */ |
| 1039 | static enum scsi_disposition |
| 1040 | scsi_try_to_abort_cmd(const struct scsi_host_template *hostt, struct scsi_cmnd *scmd) |
| 1041 | { |
| 1042 | if (!hostt->eh_abort_handler) |
| 1043 | return FAILED; |
| 1044 | |
| 1045 | return hostt->eh_abort_handler(scmd); |
| 1046 | } |
| 1047 | |
| 1048 | static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) |
| 1049 | { |
| 1050 | if (scsi_try_to_abort_cmd(hostt: scmd->device->host->hostt, scmd) != SUCCESS) |
| 1051 | if (scsi_try_bus_device_reset(scmd) != SUCCESS) |
| 1052 | if (scsi_try_target_reset(scmd) != SUCCESS) |
| 1053 | if (scsi_try_bus_reset(scmd) != SUCCESS) |
| 1054 | scsi_try_host_reset(scmd); |
| 1055 | } |
| 1056 | |
| 1057 | /** |
| 1058 | * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery |
| 1059 | * @scmd: SCSI command structure to hijack |
| 1060 | * @ses: structure to save restore information |
| 1061 | * @cmnd: CDB to send. Can be NULL if no new cmnd is needed |
| 1062 | * @cmnd_size: size in bytes of @cmnd (must be <= MAX_COMMAND_SIZE) |
| 1063 | * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) |
| 1064 | * |
| 1065 | * This function is used to save a scsi command information before re-execution |
| 1066 | * as part of the error recovery process. If @sense_bytes is 0 the command |
| 1067 | * sent must be one that does not transfer any data. If @sense_bytes != 0 |
| 1068 | * @cmnd is ignored and this functions sets up a REQUEST_SENSE command |
| 1069 | * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. |
| 1070 | */ |
| 1071 | void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, |
| 1072 | unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) |
| 1073 | { |
| 1074 | struct scsi_device *sdev = scmd->device; |
| 1075 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION |
| 1076 | struct request *rq = scsi_cmd_to_rq(scmd); |
| 1077 | #endif |
| 1078 | |
| 1079 | /* |
| 1080 | * We need saved copies of a number of fields - this is because |
| 1081 | * error handling may need to overwrite these with different values |
| 1082 | * to run different commands, and once error handling is complete, |
| 1083 | * we will need to restore these values prior to running the actual |
| 1084 | * command. |
| 1085 | */ |
| 1086 | ses->cmd_len = scmd->cmd_len; |
| 1087 | ses->data_direction = scmd->sc_data_direction; |
| 1088 | ses->sdb = scmd->sdb; |
| 1089 | ses->result = scmd->result; |
| 1090 | ses->resid_len = scmd->resid_len; |
| 1091 | ses->underflow = scmd->underflow; |
| 1092 | ses->prot_op = scmd->prot_op; |
| 1093 | ses->eh_eflags = scmd->eh_eflags; |
| 1094 | |
| 1095 | scmd->prot_op = SCSI_PROT_NORMAL; |
| 1096 | scmd->eh_eflags = 0; |
| 1097 | memcpy(ses->cmnd, scmd->cmnd, sizeof(ses->cmnd)); |
| 1098 | memset(scmd->cmnd, 0, sizeof(scmd->cmnd)); |
| 1099 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
| 1100 | scmd->result = 0; |
| 1101 | scmd->resid_len = 0; |
| 1102 | |
| 1103 | if (sense_bytes) { |
| 1104 | scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, |
| 1105 | sense_bytes); |
| 1106 | sg_init_one(&ses->sense_sgl, scmd->sense_buffer, |
| 1107 | scmd->sdb.length); |
| 1108 | scmd->sdb.table.sgl = &ses->sense_sgl; |
| 1109 | scmd->sc_data_direction = DMA_FROM_DEVICE; |
| 1110 | scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; |
| 1111 | scmd->cmnd[0] = REQUEST_SENSE; |
| 1112 | scmd->cmnd[4] = scmd->sdb.length; |
| 1113 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
| 1114 | } else { |
| 1115 | scmd->sc_data_direction = DMA_NONE; |
| 1116 | if (cmnd) { |
| 1117 | BUG_ON(cmnd_size > sizeof(scmd->cmnd)); |
| 1118 | memcpy(scmd->cmnd, cmnd, cmnd_size); |
| 1119 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
| 1120 | } |
| 1121 | } |
| 1122 | |
| 1123 | scmd->underflow = 0; |
| 1124 | |
| 1125 | if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) |
| 1126 | scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | |
| 1127 | (sdev->lun << 5 & 0xe0); |
| 1128 | |
| 1129 | /* |
| 1130 | * Encryption must be disabled for the commands submitted by the error handler. |
| 1131 | * Hence, clear the encryption context information. |
| 1132 | */ |
| 1133 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION |
| 1134 | ses->rq_crypt_keyslot = rq->crypt_keyslot; |
| 1135 | ses->rq_crypt_ctx = rq->crypt_ctx; |
| 1136 | |
| 1137 | rq->crypt_keyslot = NULL; |
| 1138 | rq->crypt_ctx = NULL; |
| 1139 | #endif |
| 1140 | |
| 1141 | /* |
| 1142 | * Zero the sense buffer. The scsi spec mandates that any |
| 1143 | * untransferred sense data should be interpreted as being zero. |
| 1144 | */ |
| 1145 | memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); |
| 1146 | } |
| 1147 | EXPORT_SYMBOL(scsi_eh_prep_cmnd); |
| 1148 | |
| 1149 | /** |
| 1150 | * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery |
| 1151 | * @scmd: SCSI command structure to restore |
| 1152 | * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd |
| 1153 | * |
| 1154 | * Undo any damage done by above scsi_eh_prep_cmnd(). |
| 1155 | */ |
| 1156 | void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) |
| 1157 | { |
| 1158 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION |
| 1159 | struct request *rq = scsi_cmd_to_rq(scmd); |
| 1160 | #endif |
| 1161 | |
| 1162 | /* |
| 1163 | * Restore original data |
| 1164 | */ |
| 1165 | scmd->cmd_len = ses->cmd_len; |
| 1166 | memcpy(scmd->cmnd, ses->cmnd, sizeof(ses->cmnd)); |
| 1167 | scmd->sc_data_direction = ses->data_direction; |
| 1168 | scmd->sdb = ses->sdb; |
| 1169 | scmd->result = ses->result; |
| 1170 | scmd->resid_len = ses->resid_len; |
| 1171 | scmd->underflow = ses->underflow; |
| 1172 | scmd->prot_op = ses->prot_op; |
| 1173 | scmd->eh_eflags = ses->eh_eflags; |
| 1174 | |
| 1175 | #ifdef CONFIG_BLK_INLINE_ENCRYPTION |
| 1176 | rq->crypt_keyslot = ses->rq_crypt_keyslot; |
| 1177 | rq->crypt_ctx = ses->rq_crypt_ctx; |
| 1178 | #endif |
| 1179 | } |
| 1180 | EXPORT_SYMBOL(scsi_eh_restore_cmnd); |
| 1181 | |
| 1182 | /** |
| 1183 | * scsi_send_eh_cmnd - submit a scsi command as part of error recovery |
| 1184 | * @scmd: SCSI command structure to hijack |
| 1185 | * @cmnd: CDB to send |
| 1186 | * @cmnd_size: size in bytes of @cmnd |
| 1187 | * @timeout: timeout for this request |
| 1188 | * @sense_bytes: size of sense data to copy or 0 |
| 1189 | * |
| 1190 | * This function is used to send a scsi command down to a target device |
| 1191 | * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. |
| 1192 | * |
| 1193 | * Return value: |
| 1194 | * SUCCESS or FAILED or NEEDS_RETRY |
| 1195 | */ |
| 1196 | static enum scsi_disposition scsi_send_eh_cmnd(struct scsi_cmnd *scmd, |
| 1197 | unsigned char *cmnd, int cmnd_size, int timeout, unsigned sense_bytes) |
| 1198 | { |
| 1199 | struct scsi_device *sdev = scmd->device; |
| 1200 | struct Scsi_Host *shost = sdev->host; |
| 1201 | DECLARE_COMPLETION_ONSTACK(done); |
| 1202 | unsigned long timeleft = timeout, delay; |
| 1203 | struct scsi_eh_save ses; |
| 1204 | const unsigned long stall_for = msecs_to_jiffies(m: 100); |
| 1205 | int rtn; |
| 1206 | |
| 1207 | retry: |
| 1208 | scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); |
| 1209 | shost->eh_action = &done; |
| 1210 | |
| 1211 | scsi_log_send(cmd: scmd); |
| 1212 | scmd->submitter = SUBMITTED_BY_SCSI_ERROR_HANDLER; |
| 1213 | scmd->flags |= SCMD_LAST; |
| 1214 | |
| 1215 | /* |
| 1216 | * Lock sdev->state_mutex to avoid that scsi_device_quiesce() can |
| 1217 | * change the SCSI device state after we have examined it and before |
| 1218 | * .queuecommand() is called. |
| 1219 | */ |
| 1220 | mutex_lock(&sdev->state_mutex); |
| 1221 | while (sdev->sdev_state == SDEV_BLOCK && timeleft > 0) { |
| 1222 | mutex_unlock(lock: &sdev->state_mutex); |
| 1223 | SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_DEBUG, sdev, |
| 1224 | "%s: state %d <> %d\n" , __func__, sdev->sdev_state, |
| 1225 | SDEV_BLOCK)); |
| 1226 | delay = min(timeleft, stall_for); |
| 1227 | timeleft -= delay; |
| 1228 | msleep(msecs: jiffies_to_msecs(j: delay)); |
| 1229 | mutex_lock(&sdev->state_mutex); |
| 1230 | } |
| 1231 | if (sdev->sdev_state != SDEV_BLOCK) |
| 1232 | rtn = shost->hostt->queuecommand(shost, scmd); |
| 1233 | else |
| 1234 | rtn = FAILED; |
| 1235 | mutex_unlock(lock: &sdev->state_mutex); |
| 1236 | |
| 1237 | if (rtn) { |
| 1238 | if (timeleft > stall_for) { |
| 1239 | scsi_eh_restore_cmnd(scmd, &ses); |
| 1240 | |
| 1241 | timeleft -= stall_for; |
| 1242 | msleep(msecs: jiffies_to_msecs(j: stall_for)); |
| 1243 | goto retry; |
| 1244 | } |
| 1245 | /* signal not to enter either branch of the if () below */ |
| 1246 | timeleft = 0; |
| 1247 | rtn = FAILED; |
| 1248 | } else { |
| 1249 | timeleft = wait_for_completion_timeout(x: &done, timeout); |
| 1250 | rtn = SUCCESS; |
| 1251 | } |
| 1252 | |
| 1253 | shost->eh_action = NULL; |
| 1254 | |
| 1255 | scsi_log_completion(cmd: scmd, disposition: rtn); |
| 1256 | |
| 1257 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1258 | "%s timeleft: %ld\n" , |
| 1259 | __func__, timeleft)); |
| 1260 | |
| 1261 | /* |
| 1262 | * If there is time left scsi_eh_done got called, and we will examine |
| 1263 | * the actual status codes to see whether the command actually did |
| 1264 | * complete normally, else if we have a zero return and no time left, |
| 1265 | * the command must still be pending, so abort it and return FAILED. |
| 1266 | * If we never actually managed to issue the command, because |
| 1267 | * ->queuecommand() kept returning non zero, use the rtn = FAILED |
| 1268 | * value above (so don't execute either branch of the if) |
| 1269 | */ |
| 1270 | if (timeleft) { |
| 1271 | rtn = scsi_eh_completed_normally(scmd); |
| 1272 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1273 | "%s: scsi_eh_completed_normally %x\n" , __func__, rtn)); |
| 1274 | |
| 1275 | switch (rtn) { |
| 1276 | case SUCCESS: |
| 1277 | case NEEDS_RETRY: |
| 1278 | case FAILED: |
| 1279 | break; |
| 1280 | case ADD_TO_MLQUEUE: |
| 1281 | rtn = NEEDS_RETRY; |
| 1282 | break; |
| 1283 | default: |
| 1284 | rtn = FAILED; |
| 1285 | break; |
| 1286 | } |
| 1287 | } else if (rtn != FAILED) { |
| 1288 | scsi_abort_eh_cmnd(scmd); |
| 1289 | rtn = FAILED; |
| 1290 | } |
| 1291 | |
| 1292 | scsi_eh_restore_cmnd(scmd, &ses); |
| 1293 | |
| 1294 | return rtn; |
| 1295 | } |
| 1296 | |
| 1297 | /** |
| 1298 | * scsi_request_sense - Request sense data from a particular target. |
| 1299 | * @scmd: SCSI cmd for request sense. |
| 1300 | * |
| 1301 | * Notes: |
| 1302 | * Some hosts automatically obtain this information, others require |
| 1303 | * that we obtain it on our own. This function will *not* return until |
| 1304 | * the command either times out, or it completes. |
| 1305 | */ |
| 1306 | static enum scsi_disposition scsi_request_sense(struct scsi_cmnd *scmd) |
| 1307 | { |
| 1308 | return scsi_send_eh_cmnd(scmd, NULL, cmnd_size: 0, timeout: scmd->device->eh_timeout, sense_bytes: ~0); |
| 1309 | } |
| 1310 | |
| 1311 | static enum scsi_disposition |
| 1312 | scsi_eh_action(struct scsi_cmnd *scmd, enum scsi_disposition rtn) |
| 1313 | { |
| 1314 | if (!blk_rq_is_passthrough(rq: scsi_cmd_to_rq(scmd))) { |
| 1315 | struct scsi_driver *sdrv = scsi_cmd_to_driver(cmd: scmd); |
| 1316 | if (sdrv->eh_action) |
| 1317 | rtn = sdrv->eh_action(scmd, rtn); |
| 1318 | } |
| 1319 | return rtn; |
| 1320 | } |
| 1321 | |
| 1322 | /** |
| 1323 | * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. |
| 1324 | * @scmd: Original SCSI cmd that eh has finished. |
| 1325 | * @done_q: Queue for processed commands. |
| 1326 | * |
| 1327 | * Notes: |
| 1328 | * We don't want to use the normal command completion while we are are |
| 1329 | * still handling errors - it may cause other commands to be queued, |
| 1330 | * and that would disturb what we are doing. Thus we really want to |
| 1331 | * keep a list of pending commands for final completion, and once we |
| 1332 | * are ready to leave error handling we handle completion for real. |
| 1333 | */ |
| 1334 | void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) |
| 1335 | { |
| 1336 | list_move_tail(list: &scmd->eh_entry, head: done_q); |
| 1337 | } |
| 1338 | EXPORT_SYMBOL(scsi_eh_finish_cmd); |
| 1339 | |
| 1340 | /** |
| 1341 | * scsi_eh_get_sense - Get device sense data. |
| 1342 | * @work_q: Queue of commands to process. |
| 1343 | * @done_q: Queue of processed commands. |
| 1344 | * |
| 1345 | * Description: |
| 1346 | * See if we need to request sense information. if so, then get it |
| 1347 | * now, so we have a better idea of what to do. |
| 1348 | * |
| 1349 | * Notes: |
| 1350 | * This has the unfortunate side effect that if a shost adapter does |
| 1351 | * not automatically request sense information, we end up shutting |
| 1352 | * it down before we request it. |
| 1353 | * |
| 1354 | * All drivers should request sense information internally these days, |
| 1355 | * so for now all I have to say is tough noogies if you end up in here. |
| 1356 | * |
| 1357 | * XXX: Long term this code should go away, but that needs an audit of |
| 1358 | * all LLDDs first. |
| 1359 | */ |
| 1360 | int scsi_eh_get_sense(struct list_head *work_q, |
| 1361 | struct list_head *done_q) |
| 1362 | { |
| 1363 | struct scsi_cmnd *scmd, *next; |
| 1364 | struct Scsi_Host *shost; |
| 1365 | enum scsi_disposition rtn; |
| 1366 | |
| 1367 | /* |
| 1368 | * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, |
| 1369 | * should not get sense. |
| 1370 | */ |
| 1371 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1372 | if ((scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || |
| 1373 | SCSI_SENSE_VALID(scmd)) |
| 1374 | continue; |
| 1375 | |
| 1376 | shost = scmd->device->host; |
| 1377 | if (scsi_host_eh_past_deadline(shost)) { |
| 1378 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1379 | scmd_printk(KERN_INFO, scmd, |
| 1380 | "%s: skip request sense, past eh deadline\n" , |
| 1381 | current->comm)); |
| 1382 | break; |
| 1383 | } |
| 1384 | if (!scsi_status_is_check_condition(status: scmd->result)) |
| 1385 | /* |
| 1386 | * don't request sense if there's no check condition |
| 1387 | * status because the error we're processing isn't one |
| 1388 | * that has a sense code (and some devices get |
| 1389 | * confused by sense requests out of the blue) |
| 1390 | */ |
| 1391 | continue; |
| 1392 | |
| 1393 | SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, |
| 1394 | "%s: requesting sense\n" , |
| 1395 | current->comm)); |
| 1396 | rtn = scsi_request_sense(scmd); |
| 1397 | if (rtn != SUCCESS) |
| 1398 | continue; |
| 1399 | |
| 1400 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1401 | "sense requested, result %x\n" , scmd->result)); |
| 1402 | SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); |
| 1403 | |
| 1404 | rtn = scsi_decide_disposition(cmd: scmd); |
| 1405 | |
| 1406 | /* |
| 1407 | * if the result was normal, then just pass it along to the |
| 1408 | * upper level. |
| 1409 | */ |
| 1410 | if (rtn == SUCCESS) |
| 1411 | /* |
| 1412 | * We don't want this command reissued, just finished |
| 1413 | * with the sense data, so set retries to the max |
| 1414 | * allowed to ensure it won't get reissued. If the user |
| 1415 | * has requested infinite retries, we also want to |
| 1416 | * finish this command, so force completion by setting |
| 1417 | * retries and allowed to the same value. |
| 1418 | */ |
| 1419 | if (scmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) |
| 1420 | scmd->retries = scmd->allowed = 1; |
| 1421 | else |
| 1422 | scmd->retries = scmd->allowed; |
| 1423 | else if (rtn != NEEDS_RETRY) |
| 1424 | continue; |
| 1425 | |
| 1426 | scsi_eh_finish_cmd(scmd, done_q); |
| 1427 | } |
| 1428 | |
| 1429 | return list_empty(head: work_q); |
| 1430 | } |
| 1431 | EXPORT_SYMBOL_GPL(scsi_eh_get_sense); |
| 1432 | |
| 1433 | /** |
| 1434 | * scsi_eh_tur - Send TUR to device. |
| 1435 | * @scmd: &scsi_cmnd to send TUR |
| 1436 | * |
| 1437 | * Return value: |
| 1438 | * 0 - Device is ready. 1 - Device NOT ready. |
| 1439 | */ |
| 1440 | static int scsi_eh_tur(struct scsi_cmnd *scmd) |
| 1441 | { |
| 1442 | static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; |
| 1443 | int retry_cnt = 1; |
| 1444 | enum scsi_disposition rtn; |
| 1445 | |
| 1446 | retry_tur: |
| 1447 | rtn = scsi_send_eh_cmnd(scmd, cmnd: tur_command, cmnd_size: 6, |
| 1448 | timeout: scmd->device->eh_timeout, sense_bytes: 0); |
| 1449 | |
| 1450 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1451 | "%s return: %x\n" , __func__, rtn)); |
| 1452 | |
| 1453 | switch (rtn) { |
| 1454 | case NEEDS_RETRY: |
| 1455 | if (retry_cnt--) |
| 1456 | goto retry_tur; |
| 1457 | fallthrough; |
| 1458 | case SUCCESS: |
| 1459 | return 0; |
| 1460 | default: |
| 1461 | return 1; |
| 1462 | } |
| 1463 | } |
| 1464 | |
| 1465 | /** |
| 1466 | * scsi_eh_test_devices - check if devices are responding from error recovery. |
| 1467 | * @cmd_list: scsi commands in error recovery. |
| 1468 | * @work_q: queue for commands which still need more error recovery |
| 1469 | * @done_q: queue for commands which are finished |
| 1470 | * @try_stu: boolean on if a STU command should be tried in addition to TUR. |
| 1471 | * |
| 1472 | * Decription: |
| 1473 | * Tests if devices are in a working state. Commands to devices now in |
| 1474 | * a working state are sent to the done_q while commands to devices which |
| 1475 | * are still failing to respond are returned to the work_q for more |
| 1476 | * processing. |
| 1477 | **/ |
| 1478 | static int scsi_eh_test_devices(struct list_head *cmd_list, |
| 1479 | struct list_head *work_q, |
| 1480 | struct list_head *done_q, int try_stu) |
| 1481 | { |
| 1482 | struct scsi_cmnd *scmd, *next; |
| 1483 | struct scsi_device *sdev; |
| 1484 | int finish_cmds; |
| 1485 | |
| 1486 | while (!list_empty(head: cmd_list)) { |
| 1487 | scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); |
| 1488 | sdev = scmd->device; |
| 1489 | |
| 1490 | if (!try_stu) { |
| 1491 | if (scsi_host_eh_past_deadline(shost: sdev->host)) { |
| 1492 | /* Push items back onto work_q */ |
| 1493 | list_splice_init(list: cmd_list, head: work_q); |
| 1494 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1495 | sdev_printk(KERN_INFO, sdev, |
| 1496 | "%s: skip test device, past eh deadline" , |
| 1497 | current->comm)); |
| 1498 | break; |
| 1499 | } |
| 1500 | } |
| 1501 | |
| 1502 | finish_cmds = !scsi_device_online(sdev: scmd->device) || |
| 1503 | (try_stu && !scsi_eh_try_stu(scmd) && |
| 1504 | !scsi_eh_tur(scmd)) || |
| 1505 | !scsi_eh_tur(scmd); |
| 1506 | |
| 1507 | list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) |
| 1508 | if (scmd->device == sdev) { |
| 1509 | if (finish_cmds && |
| 1510 | (try_stu || |
| 1511 | scsi_eh_action(scmd, rtn: SUCCESS) == SUCCESS)) |
| 1512 | scsi_eh_finish_cmd(scmd, done_q); |
| 1513 | else |
| 1514 | list_move_tail(list: &scmd->eh_entry, head: work_q); |
| 1515 | } |
| 1516 | } |
| 1517 | return list_empty(head: work_q); |
| 1518 | } |
| 1519 | |
| 1520 | /** |
| 1521 | * scsi_eh_try_stu - Send START_UNIT to device. |
| 1522 | * @scmd: &scsi_cmnd to send START_UNIT |
| 1523 | * |
| 1524 | * Return value: |
| 1525 | * 0 - Device is ready. 1 - Device NOT ready. |
| 1526 | */ |
| 1527 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd) |
| 1528 | { |
| 1529 | static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; |
| 1530 | |
| 1531 | if (scmd->device->allow_restart) { |
| 1532 | int i; |
| 1533 | enum scsi_disposition rtn = NEEDS_RETRY; |
| 1534 | |
| 1535 | for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) |
| 1536 | rtn = scsi_send_eh_cmnd(scmd, cmnd: stu_command, cmnd_size: 6, |
| 1537 | timeout: scmd->device->eh_timeout, sense_bytes: 0); |
| 1538 | |
| 1539 | if (rtn == SUCCESS) |
| 1540 | return 0; |
| 1541 | } |
| 1542 | |
| 1543 | return 1; |
| 1544 | } |
| 1545 | |
| 1546 | /** |
| 1547 | * scsi_eh_stu - send START_UNIT if needed |
| 1548 | * @shost: &scsi host being recovered. |
| 1549 | * @work_q: &list_head for pending commands. |
| 1550 | * @done_q: &list_head for processed commands. |
| 1551 | * |
| 1552 | * Notes: |
| 1553 | * If commands are failing due to not ready, initializing command required, |
| 1554 | * try revalidating the device, which will end up sending a start unit. |
| 1555 | */ |
| 1556 | static int scsi_eh_stu(struct Scsi_Host *shost, |
| 1557 | struct list_head *work_q, |
| 1558 | struct list_head *done_q) |
| 1559 | { |
| 1560 | struct scsi_cmnd *scmd, *stu_scmd, *next; |
| 1561 | struct scsi_device *sdev; |
| 1562 | |
| 1563 | shost_for_each_device(sdev, shost) { |
| 1564 | if (scsi_host_eh_past_deadline(shost)) { |
| 1565 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1566 | sdev_printk(KERN_INFO, sdev, |
| 1567 | "%s: skip START_UNIT, past eh deadline\n" , |
| 1568 | current->comm)); |
| 1569 | scsi_device_put(sdev); |
| 1570 | break; |
| 1571 | } |
| 1572 | stu_scmd = NULL; |
| 1573 | list_for_each_entry(scmd, work_q, eh_entry) |
| 1574 | if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && |
| 1575 | scsi_check_sense(scmd) == FAILED ) { |
| 1576 | stu_scmd = scmd; |
| 1577 | break; |
| 1578 | } |
| 1579 | |
| 1580 | if (!stu_scmd) |
| 1581 | continue; |
| 1582 | |
| 1583 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1584 | sdev_printk(KERN_INFO, sdev, |
| 1585 | "%s: Sending START_UNIT\n" , |
| 1586 | current->comm)); |
| 1587 | |
| 1588 | if (!scsi_eh_try_stu(scmd: stu_scmd)) { |
| 1589 | if (!scsi_device_online(sdev) || |
| 1590 | !scsi_eh_tur(scmd: stu_scmd)) { |
| 1591 | list_for_each_entry_safe(scmd, next, |
| 1592 | work_q, eh_entry) { |
| 1593 | if (scmd->device == sdev && |
| 1594 | scsi_eh_action(scmd, rtn: SUCCESS) == SUCCESS) |
| 1595 | scsi_eh_finish_cmd(scmd, done_q); |
| 1596 | } |
| 1597 | } |
| 1598 | } else { |
| 1599 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1600 | sdev_printk(KERN_INFO, sdev, |
| 1601 | "%s: START_UNIT failed\n" , |
| 1602 | current->comm)); |
| 1603 | } |
| 1604 | } |
| 1605 | |
| 1606 | return list_empty(head: work_q); |
| 1607 | } |
| 1608 | |
| 1609 | |
| 1610 | /** |
| 1611 | * scsi_eh_bus_device_reset - send bdr if needed |
| 1612 | * @shost: scsi host being recovered. |
| 1613 | * @work_q: &list_head for pending commands. |
| 1614 | * @done_q: &list_head for processed commands. |
| 1615 | * |
| 1616 | * Notes: |
| 1617 | * Try a bus device reset. Still, look to see whether we have multiple |
| 1618 | * devices that are jammed or not - if we have multiple devices, it |
| 1619 | * makes no sense to try bus_device_reset - we really would need to try |
| 1620 | * a bus_reset instead. |
| 1621 | */ |
| 1622 | static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, |
| 1623 | struct list_head *work_q, |
| 1624 | struct list_head *done_q) |
| 1625 | { |
| 1626 | struct scsi_cmnd *scmd, *bdr_scmd, *next; |
| 1627 | struct scsi_device *sdev; |
| 1628 | enum scsi_disposition rtn; |
| 1629 | |
| 1630 | shost_for_each_device(sdev, shost) { |
| 1631 | if (scsi_host_eh_past_deadline(shost)) { |
| 1632 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1633 | sdev_printk(KERN_INFO, sdev, |
| 1634 | "%s: skip BDR, past eh deadline\n" , |
| 1635 | current->comm)); |
| 1636 | scsi_device_put(sdev); |
| 1637 | break; |
| 1638 | } |
| 1639 | bdr_scmd = NULL; |
| 1640 | list_for_each_entry(scmd, work_q, eh_entry) |
| 1641 | if (scmd->device == sdev) { |
| 1642 | bdr_scmd = scmd; |
| 1643 | break; |
| 1644 | } |
| 1645 | |
| 1646 | if (!bdr_scmd) |
| 1647 | continue; |
| 1648 | |
| 1649 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1650 | sdev_printk(KERN_INFO, sdev, |
| 1651 | "%s: Sending BDR\n" , current->comm)); |
| 1652 | rtn = scsi_try_bus_device_reset(scmd: bdr_scmd); |
| 1653 | if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { |
| 1654 | if (!scsi_device_online(sdev) || |
| 1655 | rtn == FAST_IO_FAIL || |
| 1656 | !scsi_eh_tur(scmd: bdr_scmd)) { |
| 1657 | list_for_each_entry_safe(scmd, next, |
| 1658 | work_q, eh_entry) { |
| 1659 | if (scmd->device == sdev && |
| 1660 | scsi_eh_action(scmd, rtn) != FAILED) |
| 1661 | scsi_eh_finish_cmd(scmd, |
| 1662 | done_q); |
| 1663 | } |
| 1664 | } |
| 1665 | } else { |
| 1666 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1667 | sdev_printk(KERN_INFO, sdev, |
| 1668 | "%s: BDR failed\n" , current->comm)); |
| 1669 | } |
| 1670 | } |
| 1671 | |
| 1672 | return list_empty(head: work_q); |
| 1673 | } |
| 1674 | |
| 1675 | /** |
| 1676 | * scsi_eh_target_reset - send target reset if needed |
| 1677 | * @shost: scsi host being recovered. |
| 1678 | * @work_q: &list_head for pending commands. |
| 1679 | * @done_q: &list_head for processed commands. |
| 1680 | * |
| 1681 | * Notes: |
| 1682 | * Try a target reset. |
| 1683 | */ |
| 1684 | static int scsi_eh_target_reset(struct Scsi_Host *shost, |
| 1685 | struct list_head *work_q, |
| 1686 | struct list_head *done_q) |
| 1687 | { |
| 1688 | LIST_HEAD(tmp_list); |
| 1689 | LIST_HEAD(check_list); |
| 1690 | |
| 1691 | list_splice_init(list: work_q, head: &tmp_list); |
| 1692 | |
| 1693 | while (!list_empty(head: &tmp_list)) { |
| 1694 | struct scsi_cmnd *next, *scmd; |
| 1695 | enum scsi_disposition rtn; |
| 1696 | unsigned int id; |
| 1697 | |
| 1698 | if (scsi_host_eh_past_deadline(shost)) { |
| 1699 | /* push back on work queue for further processing */ |
| 1700 | list_splice_init(list: &check_list, head: work_q); |
| 1701 | list_splice_init(list: &tmp_list, head: work_q); |
| 1702 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1703 | shost_printk(KERN_INFO, shost, |
| 1704 | "%s: Skip target reset, past eh deadline\n" , |
| 1705 | current->comm)); |
| 1706 | return list_empty(head: work_q); |
| 1707 | } |
| 1708 | |
| 1709 | scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); |
| 1710 | id = scmd_id(scmd); |
| 1711 | |
| 1712 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1713 | shost_printk(KERN_INFO, shost, |
| 1714 | "%s: Sending target reset to target %d\n" , |
| 1715 | current->comm, id)); |
| 1716 | rtn = scsi_try_target_reset(scmd); |
| 1717 | if (rtn != SUCCESS && rtn != FAST_IO_FAIL) |
| 1718 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1719 | shost_printk(KERN_INFO, shost, |
| 1720 | "%s: Target reset failed" |
| 1721 | " target: %d\n" , |
| 1722 | current->comm, id)); |
| 1723 | list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { |
| 1724 | if (scmd_id(scmd) != id) |
| 1725 | continue; |
| 1726 | |
| 1727 | if (rtn == SUCCESS) |
| 1728 | list_move_tail(list: &scmd->eh_entry, head: &check_list); |
| 1729 | else if (rtn == FAST_IO_FAIL) |
| 1730 | scsi_eh_finish_cmd(scmd, done_q); |
| 1731 | else |
| 1732 | /* push back on work queue for further processing */ |
| 1733 | list_move(list: &scmd->eh_entry, head: work_q); |
| 1734 | } |
| 1735 | } |
| 1736 | |
| 1737 | return scsi_eh_test_devices(cmd_list: &check_list, work_q, done_q, try_stu: 0); |
| 1738 | } |
| 1739 | |
| 1740 | /** |
| 1741 | * scsi_eh_bus_reset - send a bus reset |
| 1742 | * @shost: &scsi host being recovered. |
| 1743 | * @work_q: &list_head for pending commands. |
| 1744 | * @done_q: &list_head for processed commands. |
| 1745 | */ |
| 1746 | static int scsi_eh_bus_reset(struct Scsi_Host *shost, |
| 1747 | struct list_head *work_q, |
| 1748 | struct list_head *done_q) |
| 1749 | { |
| 1750 | struct scsi_cmnd *scmd, *chan_scmd, *next; |
| 1751 | LIST_HEAD(check_list); |
| 1752 | unsigned int channel; |
| 1753 | enum scsi_disposition rtn; |
| 1754 | |
| 1755 | /* |
| 1756 | * we really want to loop over the various channels, and do this on |
| 1757 | * a channel by channel basis. we should also check to see if any |
| 1758 | * of the failed commands are on soft_reset devices, and if so, skip |
| 1759 | * the reset. |
| 1760 | */ |
| 1761 | |
| 1762 | for (channel = 0; channel <= shost->max_channel; channel++) { |
| 1763 | if (scsi_host_eh_past_deadline(shost)) { |
| 1764 | list_splice_init(list: &check_list, head: work_q); |
| 1765 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1766 | shost_printk(KERN_INFO, shost, |
| 1767 | "%s: skip BRST, past eh deadline\n" , |
| 1768 | current->comm)); |
| 1769 | return list_empty(head: work_q); |
| 1770 | } |
| 1771 | |
| 1772 | chan_scmd = NULL; |
| 1773 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 1774 | if (channel == scmd_channel(scmd)) { |
| 1775 | chan_scmd = scmd; |
| 1776 | break; |
| 1777 | /* |
| 1778 | * FIXME add back in some support for |
| 1779 | * soft_reset devices. |
| 1780 | */ |
| 1781 | } |
| 1782 | } |
| 1783 | |
| 1784 | if (!chan_scmd) |
| 1785 | continue; |
| 1786 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1787 | shost_printk(KERN_INFO, shost, |
| 1788 | "%s: Sending BRST chan: %d\n" , |
| 1789 | current->comm, channel)); |
| 1790 | rtn = scsi_try_bus_reset(scmd: chan_scmd); |
| 1791 | if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { |
| 1792 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1793 | if (channel == scmd_channel(scmd)) { |
| 1794 | if (rtn == FAST_IO_FAIL) |
| 1795 | scsi_eh_finish_cmd(scmd, |
| 1796 | done_q); |
| 1797 | else |
| 1798 | list_move_tail(list: &scmd->eh_entry, |
| 1799 | head: &check_list); |
| 1800 | } |
| 1801 | } |
| 1802 | } else { |
| 1803 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1804 | shost_printk(KERN_INFO, shost, |
| 1805 | "%s: BRST failed chan: %d\n" , |
| 1806 | current->comm, channel)); |
| 1807 | } |
| 1808 | } |
| 1809 | return scsi_eh_test_devices(cmd_list: &check_list, work_q, done_q, try_stu: 0); |
| 1810 | } |
| 1811 | |
| 1812 | /** |
| 1813 | * scsi_eh_host_reset - send a host reset |
| 1814 | * @shost: host to be reset. |
| 1815 | * @work_q: &list_head for pending commands. |
| 1816 | * @done_q: &list_head for processed commands. |
| 1817 | */ |
| 1818 | static int scsi_eh_host_reset(struct Scsi_Host *shost, |
| 1819 | struct list_head *work_q, |
| 1820 | struct list_head *done_q) |
| 1821 | { |
| 1822 | struct scsi_cmnd *scmd, *next; |
| 1823 | LIST_HEAD(check_list); |
| 1824 | enum scsi_disposition rtn; |
| 1825 | |
| 1826 | if (!list_empty(head: work_q)) { |
| 1827 | scmd = list_entry(work_q->next, |
| 1828 | struct scsi_cmnd, eh_entry); |
| 1829 | |
| 1830 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1831 | shost_printk(KERN_INFO, shost, |
| 1832 | "%s: Sending HRST\n" , |
| 1833 | current->comm)); |
| 1834 | |
| 1835 | rtn = scsi_try_host_reset(scmd); |
| 1836 | if (rtn == SUCCESS) { |
| 1837 | list_splice_init(list: work_q, head: &check_list); |
| 1838 | } else if (rtn == FAST_IO_FAIL) { |
| 1839 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1840 | scsi_eh_finish_cmd(scmd, done_q); |
| 1841 | } |
| 1842 | } else { |
| 1843 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1844 | shost_printk(KERN_INFO, shost, |
| 1845 | "%s: HRST failed\n" , |
| 1846 | current->comm)); |
| 1847 | } |
| 1848 | } |
| 1849 | return scsi_eh_test_devices(cmd_list: &check_list, work_q, done_q, try_stu: 1); |
| 1850 | } |
| 1851 | |
| 1852 | /** |
| 1853 | * scsi_eh_offline_sdevs - offline scsi devices that fail to recover |
| 1854 | * @work_q: &list_head for pending commands. |
| 1855 | * @done_q: &list_head for processed commands. |
| 1856 | */ |
| 1857 | static void scsi_eh_offline_sdevs(struct list_head *work_q, |
| 1858 | struct list_head *done_q) |
| 1859 | { |
| 1860 | struct scsi_cmnd *scmd, *next; |
| 1861 | struct scsi_device *sdev; |
| 1862 | |
| 1863 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1864 | sdev_printk(KERN_INFO, scmd->device, "Device offlined - " |
| 1865 | "not ready after error recovery\n" ); |
| 1866 | sdev = scmd->device; |
| 1867 | |
| 1868 | mutex_lock(&sdev->state_mutex); |
| 1869 | scsi_device_set_state(sdev, state: SDEV_OFFLINE); |
| 1870 | mutex_unlock(lock: &sdev->state_mutex); |
| 1871 | |
| 1872 | scsi_eh_finish_cmd(scmd, done_q); |
| 1873 | } |
| 1874 | return; |
| 1875 | } |
| 1876 | |
| 1877 | /** |
| 1878 | * scsi_noretry_cmd - determine if command should be failed fast |
| 1879 | * @scmd: SCSI cmd to examine. |
| 1880 | */ |
| 1881 | bool scsi_noretry_cmd(struct scsi_cmnd *scmd) |
| 1882 | { |
| 1883 | struct request *req = scsi_cmd_to_rq(scmd); |
| 1884 | |
| 1885 | switch (host_byte(scmd->result)) { |
| 1886 | case DID_OK: |
| 1887 | break; |
| 1888 | case DID_TIME_OUT: |
| 1889 | goto check_type; |
| 1890 | case DID_BUS_BUSY: |
| 1891 | return !!(req->cmd_flags & REQ_FAILFAST_TRANSPORT); |
| 1892 | case DID_PARITY: |
| 1893 | return !!(req->cmd_flags & REQ_FAILFAST_DEV); |
| 1894 | case DID_ERROR: |
| 1895 | if (get_status_byte(cmd: scmd) == SAM_STAT_RESERVATION_CONFLICT) |
| 1896 | return false; |
| 1897 | fallthrough; |
| 1898 | case DID_SOFT_ERROR: |
| 1899 | return !!(req->cmd_flags & REQ_FAILFAST_DRIVER); |
| 1900 | } |
| 1901 | |
| 1902 | /* Never retry commands aborted due to a duration limit timeout */ |
| 1903 | if (scsi_ml_byte(result: scmd->result) == SCSIML_STAT_DL_TIMEOUT) |
| 1904 | return true; |
| 1905 | |
| 1906 | if (!scsi_status_is_check_condition(status: scmd->result)) |
| 1907 | return false; |
| 1908 | |
| 1909 | check_type: |
| 1910 | /* |
| 1911 | * assume caller has checked sense and determined |
| 1912 | * the check condition was retryable. |
| 1913 | */ |
| 1914 | if (req->cmd_flags & REQ_FAILFAST_DEV || blk_rq_is_passthrough(rq: req)) |
| 1915 | return true; |
| 1916 | |
| 1917 | return false; |
| 1918 | } |
| 1919 | |
| 1920 | /** |
| 1921 | * scsi_decide_disposition - Disposition a cmd on return from LLD. |
| 1922 | * @scmd: SCSI cmd to examine. |
| 1923 | * |
| 1924 | * Notes: |
| 1925 | * This is *only* called when we are examining the status after sending |
| 1926 | * out the actual data command. any commands that are queued for error |
| 1927 | * recovery (e.g. test_unit_ready) do *not* come through here. |
| 1928 | * |
| 1929 | * When this routine returns failed, it means the error handler thread |
| 1930 | * is woken. In cases where the error code indicates an error that |
| 1931 | * doesn't require the error handler read (i.e. we don't need to |
| 1932 | * abort/reset), this function should return SUCCESS. |
| 1933 | */ |
| 1934 | enum scsi_disposition scsi_decide_disposition(struct scsi_cmnd *scmd) |
| 1935 | { |
| 1936 | enum scsi_disposition rtn; |
| 1937 | |
| 1938 | /* |
| 1939 | * if the device is offline, then we clearly just pass the result back |
| 1940 | * up to the top level. |
| 1941 | */ |
| 1942 | if (!scsi_device_online(sdev: scmd->device)) { |
| 1943 | SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, |
| 1944 | "%s: device offline - report as SUCCESS\n" , __func__)); |
| 1945 | return SUCCESS; |
| 1946 | } |
| 1947 | |
| 1948 | /* |
| 1949 | * first check the host byte, to see if there is anything in there |
| 1950 | * that would indicate what we need to do. |
| 1951 | */ |
| 1952 | switch (host_byte(scmd->result)) { |
| 1953 | case DID_PASSTHROUGH: |
| 1954 | /* |
| 1955 | * no matter what, pass this through to the upper layer. |
| 1956 | * nuke this special code so that it looks like we are saying |
| 1957 | * did_ok. |
| 1958 | */ |
| 1959 | scmd->result &= 0xff00ffff; |
| 1960 | return SUCCESS; |
| 1961 | case DID_OK: |
| 1962 | /* |
| 1963 | * looks good. drop through, and check the next byte. |
| 1964 | */ |
| 1965 | break; |
| 1966 | case DID_ABORT: |
| 1967 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { |
| 1968 | set_host_byte(cmd: scmd, status: DID_TIME_OUT); |
| 1969 | return SUCCESS; |
| 1970 | } |
| 1971 | fallthrough; |
| 1972 | case DID_NO_CONNECT: |
| 1973 | case DID_BAD_TARGET: |
| 1974 | /* |
| 1975 | * note - this means that we just report the status back |
| 1976 | * to the top level driver, not that we actually think |
| 1977 | * that it indicates SUCCESS. |
| 1978 | */ |
| 1979 | return SUCCESS; |
| 1980 | case DID_SOFT_ERROR: |
| 1981 | /* |
| 1982 | * when the low level driver returns did_soft_error, |
| 1983 | * it is responsible for keeping an internal retry counter |
| 1984 | * in order to avoid endless loops (db) |
| 1985 | */ |
| 1986 | goto maybe_retry; |
| 1987 | case DID_IMM_RETRY: |
| 1988 | return NEEDS_RETRY; |
| 1989 | |
| 1990 | case DID_REQUEUE: |
| 1991 | return ADD_TO_MLQUEUE; |
| 1992 | case DID_TRANSPORT_DISRUPTED: |
| 1993 | /* |
| 1994 | * LLD/transport was disrupted during processing of the IO. |
| 1995 | * The transport class is now blocked/blocking, |
| 1996 | * and the transport will decide what to do with the IO |
| 1997 | * based on its timers and recovery capablilities if |
| 1998 | * there are enough retries. |
| 1999 | */ |
| 2000 | goto maybe_retry; |
| 2001 | case DID_TRANSPORT_FAILFAST: |
| 2002 | /* |
| 2003 | * The transport decided to failfast the IO (most likely |
| 2004 | * the fast io fail tmo fired), so send IO directly upwards. |
| 2005 | */ |
| 2006 | return SUCCESS; |
| 2007 | case DID_TRANSPORT_MARGINAL: |
| 2008 | /* |
| 2009 | * caller has decided not to do retries on |
| 2010 | * abort success, so send IO directly upwards |
| 2011 | */ |
| 2012 | return SUCCESS; |
| 2013 | case DID_ERROR: |
| 2014 | if (get_status_byte(cmd: scmd) == SAM_STAT_RESERVATION_CONFLICT) |
| 2015 | /* |
| 2016 | * execute reservation conflict processing code |
| 2017 | * lower down |
| 2018 | */ |
| 2019 | break; |
| 2020 | fallthrough; |
| 2021 | case DID_BUS_BUSY: |
| 2022 | case DID_PARITY: |
| 2023 | goto maybe_retry; |
| 2024 | case DID_TIME_OUT: |
| 2025 | /* |
| 2026 | * when we scan the bus, we get timeout messages for |
| 2027 | * these commands if there is no device available. |
| 2028 | * other hosts report did_no_connect for the same thing. |
| 2029 | */ |
| 2030 | if ((scmd->cmnd[0] == TEST_UNIT_READY || |
| 2031 | scmd->cmnd[0] == INQUIRY)) { |
| 2032 | return SUCCESS; |
| 2033 | } else { |
| 2034 | return FAILED; |
| 2035 | } |
| 2036 | case DID_RESET: |
| 2037 | return SUCCESS; |
| 2038 | default: |
| 2039 | return FAILED; |
| 2040 | } |
| 2041 | |
| 2042 | /* |
| 2043 | * check the status byte to see if this indicates anything special. |
| 2044 | */ |
| 2045 | switch (get_status_byte(cmd: scmd)) { |
| 2046 | case SAM_STAT_TASK_SET_FULL: |
| 2047 | scsi_handle_queue_full(sdev: scmd->device); |
| 2048 | /* |
| 2049 | * the case of trying to send too many commands to a |
| 2050 | * tagged queueing device. |
| 2051 | */ |
| 2052 | fallthrough; |
| 2053 | case SAM_STAT_BUSY: |
| 2054 | /* |
| 2055 | * device can't talk to us at the moment. Should only |
| 2056 | * occur (SAM-3) when the task queue is empty, so will cause |
| 2057 | * the empty queue handling to trigger a stall in the |
| 2058 | * device. |
| 2059 | */ |
| 2060 | return ADD_TO_MLQUEUE; |
| 2061 | case SAM_STAT_GOOD: |
| 2062 | if (scmd->cmnd[0] == REPORT_LUNS) |
| 2063 | scmd->device->sdev_target->expecting_lun_change = 0; |
| 2064 | scsi_handle_queue_ramp_up(sdev: scmd->device); |
| 2065 | if (scmd->sense_buffer && SCSI_SENSE_VALID(scmd)) |
| 2066 | /* |
| 2067 | * If we have sense data, call scsi_check_sense() in |
| 2068 | * order to set the correct SCSI ML byte (if any). |
| 2069 | * No point in checking the return value, since the |
| 2070 | * command has already completed successfully. |
| 2071 | */ |
| 2072 | scsi_check_sense(scmd); |
| 2073 | fallthrough; |
| 2074 | case SAM_STAT_COMMAND_TERMINATED: |
| 2075 | return SUCCESS; |
| 2076 | case SAM_STAT_TASK_ABORTED: |
| 2077 | goto maybe_retry; |
| 2078 | case SAM_STAT_CHECK_CONDITION: |
| 2079 | rtn = scsi_check_sense(scmd); |
| 2080 | if (rtn == NEEDS_RETRY) |
| 2081 | goto maybe_retry; |
| 2082 | /* if rtn == FAILED, we have no sense information; |
| 2083 | * returning FAILED will wake the error handler thread |
| 2084 | * to collect the sense and redo the decide |
| 2085 | * disposition */ |
| 2086 | return rtn; |
| 2087 | case SAM_STAT_CONDITION_MET: |
| 2088 | case SAM_STAT_INTERMEDIATE: |
| 2089 | case SAM_STAT_INTERMEDIATE_CONDITION_MET: |
| 2090 | case SAM_STAT_ACA_ACTIVE: |
| 2091 | /* |
| 2092 | * who knows? FIXME(eric) |
| 2093 | */ |
| 2094 | return SUCCESS; |
| 2095 | |
| 2096 | case SAM_STAT_RESERVATION_CONFLICT: |
| 2097 | sdev_printk(KERN_INFO, scmd->device, |
| 2098 | "reservation conflict\n" ); |
| 2099 | set_scsi_ml_byte(cmd: scmd, status: SCSIML_STAT_RESV_CONFLICT); |
| 2100 | return SUCCESS; /* causes immediate i/o error */ |
| 2101 | } |
| 2102 | return FAILED; |
| 2103 | |
| 2104 | maybe_retry: |
| 2105 | |
| 2106 | /* we requeue for retry because the error was retryable, and |
| 2107 | * the request was not marked fast fail. Note that above, |
| 2108 | * even if the request is marked fast fail, we still requeue |
| 2109 | * for queue congestion conditions (QUEUE_FULL or BUSY) */ |
| 2110 | if (scsi_cmd_retry_allowed(cmd: scmd) && !scsi_noretry_cmd(scmd)) { |
| 2111 | return NEEDS_RETRY; |
| 2112 | } else { |
| 2113 | /* |
| 2114 | * no more retries - report this one back to upper level. |
| 2115 | */ |
| 2116 | return SUCCESS; |
| 2117 | } |
| 2118 | } |
| 2119 | |
| 2120 | static enum rq_end_io_ret eh_lock_door_done(struct request *req, |
| 2121 | blk_status_t status) |
| 2122 | { |
| 2123 | blk_mq_free_request(rq: req); |
| 2124 | return RQ_END_IO_NONE; |
| 2125 | } |
| 2126 | |
| 2127 | /** |
| 2128 | * scsi_eh_lock_door - Prevent medium removal for the specified device |
| 2129 | * @sdev: SCSI device to prevent medium removal |
| 2130 | * |
| 2131 | * Locking: |
| 2132 | * We must be called from process context. |
| 2133 | * |
| 2134 | * Notes: |
| 2135 | * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the |
| 2136 | * head of the devices request queue, and continue. |
| 2137 | */ |
| 2138 | static void scsi_eh_lock_door(struct scsi_device *sdev) |
| 2139 | { |
| 2140 | struct scsi_cmnd *scmd; |
| 2141 | struct request *req; |
| 2142 | |
| 2143 | req = scsi_alloc_request(q: sdev->request_queue, opf: REQ_OP_DRV_IN, flags: 0); |
| 2144 | if (IS_ERR(ptr: req)) |
| 2145 | return; |
| 2146 | scmd = blk_mq_rq_to_pdu(rq: req); |
| 2147 | |
| 2148 | scmd->cmnd[0] = ALLOW_MEDIUM_REMOVAL; |
| 2149 | scmd->cmnd[1] = 0; |
| 2150 | scmd->cmnd[2] = 0; |
| 2151 | scmd->cmnd[3] = 0; |
| 2152 | scmd->cmnd[4] = SCSI_REMOVAL_PREVENT; |
| 2153 | scmd->cmnd[5] = 0; |
| 2154 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
| 2155 | scmd->allowed = 5; |
| 2156 | |
| 2157 | req->rq_flags |= RQF_QUIET; |
| 2158 | req->timeout = 10 * HZ; |
| 2159 | req->end_io = eh_lock_door_done; |
| 2160 | |
| 2161 | blk_execute_rq_nowait(rq: req, at_head: true); |
| 2162 | } |
| 2163 | |
| 2164 | /** |
| 2165 | * scsi_restart_operations - restart io operations to the specified host. |
| 2166 | * @shost: Host we are restarting. |
| 2167 | * |
| 2168 | * Notes: |
| 2169 | * When we entered the error handler, we blocked all further i/o to |
| 2170 | * this device. we need to 'reverse' this process. |
| 2171 | */ |
| 2172 | static void scsi_restart_operations(struct Scsi_Host *shost) |
| 2173 | { |
| 2174 | struct scsi_device *sdev; |
| 2175 | unsigned long flags; |
| 2176 | |
| 2177 | /* |
| 2178 | * If the door was locked, we need to insert a door lock request |
| 2179 | * onto the head of the SCSI request queue for the device. There |
| 2180 | * is no point trying to lock the door of an off-line device. |
| 2181 | */ |
| 2182 | shost_for_each_device(sdev, shost) { |
| 2183 | if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { |
| 2184 | scsi_eh_lock_door(sdev); |
| 2185 | sdev->was_reset = 0; |
| 2186 | } |
| 2187 | } |
| 2188 | |
| 2189 | /* |
| 2190 | * next free up anything directly waiting upon the host. this |
| 2191 | * will be requests for character device operations, and also for |
| 2192 | * ioctls to queued block devices. |
| 2193 | */ |
| 2194 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2195 | shost_printk(KERN_INFO, shost, "waking up host to restart\n" )); |
| 2196 | |
| 2197 | spin_lock_irqsave(shost->host_lock, flags); |
| 2198 | if (scsi_host_set_state(shost, SHOST_RUNNING)) |
| 2199 | if (scsi_host_set_state(shost, SHOST_CANCEL)) |
| 2200 | BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); |
| 2201 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 2202 | |
| 2203 | wake_up(&shost->host_wait); |
| 2204 | |
| 2205 | /* |
| 2206 | * finally we need to re-initiate requests that may be pending. we will |
| 2207 | * have had everything blocked while error handling is taking place, and |
| 2208 | * now that error recovery is done, we will need to ensure that these |
| 2209 | * requests are started. |
| 2210 | */ |
| 2211 | scsi_run_host_queues(shost); |
| 2212 | |
| 2213 | /* |
| 2214 | * if eh is active and host_eh_scheduled is pending we need to re-run |
| 2215 | * recovery. we do this check after scsi_run_host_queues() to allow |
| 2216 | * everything pent up since the last eh run a chance to make forward |
| 2217 | * progress before we sync again. Either we'll immediately re-run |
| 2218 | * recovery or scsi_device_unbusy() will wake us again when these |
| 2219 | * pending commands complete. |
| 2220 | */ |
| 2221 | spin_lock_irqsave(shost->host_lock, flags); |
| 2222 | if (shost->host_eh_scheduled) |
| 2223 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) |
| 2224 | WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); |
| 2225 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 2226 | } |
| 2227 | |
| 2228 | /** |
| 2229 | * scsi_eh_ready_devs - check device ready state and recover if not. |
| 2230 | * @shost: host to be recovered. |
| 2231 | * @work_q: &list_head for pending commands. |
| 2232 | * @done_q: &list_head for processed commands. |
| 2233 | */ |
| 2234 | void scsi_eh_ready_devs(struct Scsi_Host *shost, |
| 2235 | struct list_head *work_q, |
| 2236 | struct list_head *done_q) |
| 2237 | { |
| 2238 | if (!scsi_eh_stu(shost, work_q, done_q)) |
| 2239 | if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) |
| 2240 | if (!scsi_eh_target_reset(shost, work_q, done_q)) |
| 2241 | if (!scsi_eh_bus_reset(shost, work_q, done_q)) |
| 2242 | if (!scsi_eh_host_reset(shost, work_q, done_q)) |
| 2243 | scsi_eh_offline_sdevs(work_q, |
| 2244 | done_q); |
| 2245 | } |
| 2246 | EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); |
| 2247 | |
| 2248 | /** |
| 2249 | * scsi_eh_flush_done_q - finish processed commands or retry them. |
| 2250 | * @done_q: list_head of processed commands. |
| 2251 | */ |
| 2252 | void scsi_eh_flush_done_q(struct list_head *done_q) |
| 2253 | { |
| 2254 | struct scsi_cmnd *scmd, *next; |
| 2255 | |
| 2256 | list_for_each_entry_safe(scmd, next, done_q, eh_entry) { |
| 2257 | struct scsi_device *sdev = scmd->device; |
| 2258 | |
| 2259 | list_del_init(entry: &scmd->eh_entry); |
| 2260 | if (scsi_device_online(sdev) && !scsi_noretry_cmd(scmd) && |
| 2261 | scsi_cmd_retry_allowed(cmd: scmd) && |
| 2262 | scsi_eh_should_retry_cmd(cmd: scmd)) { |
| 2263 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2264 | scmd_printk(KERN_INFO, scmd, |
| 2265 | "%s: flush retry cmd\n" , |
| 2266 | current->comm)); |
| 2267 | scsi_queue_insert(cmd: scmd, SCSI_MLQUEUE_EH_RETRY); |
| 2268 | blk_mq_kick_requeue_list(q: sdev->request_queue); |
| 2269 | } else { |
| 2270 | /* |
| 2271 | * If just we got sense for the device (called |
| 2272 | * scsi_eh_get_sense), scmd->result is already |
| 2273 | * set, do not set DID_TIME_OUT. |
| 2274 | */ |
| 2275 | if (!scmd->result && |
| 2276 | !(scmd->flags & SCMD_FORCE_EH_SUCCESS)) |
| 2277 | scmd->result |= (DID_TIME_OUT << 16); |
| 2278 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2279 | scmd_printk(KERN_INFO, scmd, |
| 2280 | "%s: flush finish cmd\n" , |
| 2281 | current->comm)); |
| 2282 | scsi_finish_command(cmd: scmd); |
| 2283 | } |
| 2284 | } |
| 2285 | } |
| 2286 | EXPORT_SYMBOL(scsi_eh_flush_done_q); |
| 2287 | |
| 2288 | /** |
| 2289 | * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. |
| 2290 | * @shost: Host to unjam. |
| 2291 | * |
| 2292 | * Notes: |
| 2293 | * When we come in here, we *know* that all commands on the bus have |
| 2294 | * either completed, failed or timed out. we also know that no further |
| 2295 | * commands are being sent to the host, so things are relatively quiet |
| 2296 | * and we have freedom to fiddle with things as we wish. |
| 2297 | * |
| 2298 | * This is only the *default* implementation. it is possible for |
| 2299 | * individual drivers to supply their own version of this function, and |
| 2300 | * if the maintainer wishes to do this, it is strongly suggested that |
| 2301 | * this function be taken as a template and modified. this function |
| 2302 | * was designed to correctly handle problems for about 95% of the |
| 2303 | * different cases out there, and it should always provide at least a |
| 2304 | * reasonable amount of error recovery. |
| 2305 | * |
| 2306 | * Any command marked 'failed' or 'timeout' must eventually have |
| 2307 | * scsi_finish_cmd() called for it. we do all of the retry stuff |
| 2308 | * here, so when we restart the host after we return it should have an |
| 2309 | * empty queue. |
| 2310 | */ |
| 2311 | static void scsi_unjam_host(struct Scsi_Host *shost) |
| 2312 | { |
| 2313 | unsigned long flags; |
| 2314 | LIST_HEAD(eh_work_q); |
| 2315 | LIST_HEAD(eh_done_q); |
| 2316 | |
| 2317 | spin_lock_irqsave(shost->host_lock, flags); |
| 2318 | list_splice_init(list: &shost->eh_cmd_q, head: &eh_work_q); |
| 2319 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 2320 | |
| 2321 | SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); |
| 2322 | |
| 2323 | if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) |
| 2324 | scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); |
| 2325 | |
| 2326 | spin_lock_irqsave(shost->host_lock, flags); |
| 2327 | if (shost->eh_deadline != -1) |
| 2328 | shost->last_reset = 0; |
| 2329 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 2330 | scsi_eh_flush_done_q(&eh_done_q); |
| 2331 | } |
| 2332 | |
| 2333 | /** |
| 2334 | * scsi_error_handler - SCSI error handler thread |
| 2335 | * @data: Host for which we are running. |
| 2336 | * |
| 2337 | * Notes: |
| 2338 | * This is the main error handling loop. This is run as a kernel thread |
| 2339 | * for every SCSI host and handles all error handling activity. |
| 2340 | */ |
| 2341 | int scsi_error_handler(void *data) |
| 2342 | { |
| 2343 | struct Scsi_Host *shost = data; |
| 2344 | |
| 2345 | /* |
| 2346 | * We use TASK_INTERRUPTIBLE so that the thread is not |
| 2347 | * counted against the load average as a running process. |
| 2348 | * We never actually get interrupted because kthread_run |
| 2349 | * disables signal delivery for the created thread. |
| 2350 | */ |
| 2351 | while (true) { |
| 2352 | /* |
| 2353 | * The sequence in kthread_stop() sets the stop flag first |
| 2354 | * then wakes the process. To avoid missed wakeups, the task |
| 2355 | * should always be in a non running state before the stop |
| 2356 | * flag is checked |
| 2357 | */ |
| 2358 | set_current_state(TASK_INTERRUPTIBLE); |
| 2359 | if (kthread_should_stop()) |
| 2360 | break; |
| 2361 | |
| 2362 | if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || |
| 2363 | shost->host_failed != scsi_host_busy(shost)) { |
| 2364 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2365 | shost_printk(KERN_INFO, shost, |
| 2366 | "scsi_eh_%d: sleeping\n" , |
| 2367 | shost->host_no)); |
| 2368 | schedule(); |
| 2369 | continue; |
| 2370 | } |
| 2371 | |
| 2372 | __set_current_state(TASK_RUNNING); |
| 2373 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2374 | shost_printk(KERN_INFO, shost, |
| 2375 | "scsi_eh_%d: waking up %d/%d/%d\n" , |
| 2376 | shost->host_no, shost->host_eh_scheduled, |
| 2377 | shost->host_failed, |
| 2378 | scsi_host_busy(shost))); |
| 2379 | |
| 2380 | /* |
| 2381 | * We have a host that is failing for some reason. Figure out |
| 2382 | * what we need to do to get it up and online again (if we can). |
| 2383 | * If we fail, we end up taking the thing offline. |
| 2384 | */ |
| 2385 | if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { |
| 2386 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2387 | shost_printk(KERN_ERR, shost, |
| 2388 | "scsi_eh_%d: unable to autoresume\n" , |
| 2389 | shost->host_no)); |
| 2390 | continue; |
| 2391 | } |
| 2392 | |
| 2393 | if (shost->transportt->eh_strategy_handler) |
| 2394 | shost->transportt->eh_strategy_handler(shost); |
| 2395 | else |
| 2396 | scsi_unjam_host(shost); |
| 2397 | |
| 2398 | /* All scmds have been handled */ |
| 2399 | shost->host_failed = 0; |
| 2400 | |
| 2401 | /* |
| 2402 | * Note - if the above fails completely, the action is to take |
| 2403 | * individual devices offline and flush the queue of any |
| 2404 | * outstanding requests that may have been pending. When we |
| 2405 | * restart, we restart any I/O to any other devices on the bus |
| 2406 | * which are still online. |
| 2407 | */ |
| 2408 | scsi_restart_operations(shost); |
| 2409 | if (!shost->eh_noresume) |
| 2410 | scsi_autopm_put_host(shost); |
| 2411 | } |
| 2412 | __set_current_state(TASK_RUNNING); |
| 2413 | |
| 2414 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2415 | shost_printk(KERN_INFO, shost, |
| 2416 | "Error handler scsi_eh_%d exiting\n" , |
| 2417 | shost->host_no)); |
| 2418 | shost->ehandler = NULL; |
| 2419 | return 0; |
| 2420 | } |
| 2421 | |
| 2422 | /** |
| 2423 | * scsi_report_bus_reset() - report bus reset observed |
| 2424 | * |
| 2425 | * Utility function used by low-level drivers to report that |
| 2426 | * they have observed a bus reset on the bus being handled. |
| 2427 | * |
| 2428 | * @shost: Host in question |
| 2429 | * @channel: channel on which reset was observed. |
| 2430 | * |
| 2431 | * Returns: Nothing |
| 2432 | * |
| 2433 | * Lock status: Host lock must be held. |
| 2434 | * |
| 2435 | * Notes: This only needs to be called if the reset is one which |
| 2436 | * originates from an unknown location. Resets originated |
| 2437 | * by the mid-level itself don't need to call this, but there |
| 2438 | * should be no harm. |
| 2439 | * |
| 2440 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 2441 | * is properly treated. |
| 2442 | */ |
| 2443 | void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) |
| 2444 | { |
| 2445 | struct scsi_device *sdev; |
| 2446 | |
| 2447 | __shost_for_each_device(sdev, shost) { |
| 2448 | if (channel == sdev_channel(sdev)) |
| 2449 | __scsi_report_device_reset(sdev, NULL); |
| 2450 | } |
| 2451 | } |
| 2452 | EXPORT_SYMBOL(scsi_report_bus_reset); |
| 2453 | |
| 2454 | /** |
| 2455 | * scsi_report_device_reset() - report device reset observed |
| 2456 | * |
| 2457 | * Utility function used by low-level drivers to report that |
| 2458 | * they have observed a device reset on the device being handled. |
| 2459 | * |
| 2460 | * @shost: Host in question |
| 2461 | * @channel: channel on which reset was observed |
| 2462 | * @target: target on which reset was observed |
| 2463 | * |
| 2464 | * Returns: Nothing |
| 2465 | * |
| 2466 | * Lock status: Host lock must be held |
| 2467 | * |
| 2468 | * Notes: This only needs to be called if the reset is one which |
| 2469 | * originates from an unknown location. Resets originated |
| 2470 | * by the mid-level itself don't need to call this, but there |
| 2471 | * should be no harm. |
| 2472 | * |
| 2473 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 2474 | * is properly treated. |
| 2475 | */ |
| 2476 | void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) |
| 2477 | { |
| 2478 | struct scsi_device *sdev; |
| 2479 | |
| 2480 | __shost_for_each_device(sdev, shost) { |
| 2481 | if (channel == sdev_channel(sdev) && |
| 2482 | target == sdev_id(sdev)) |
| 2483 | __scsi_report_device_reset(sdev, NULL); |
| 2484 | } |
| 2485 | } |
| 2486 | EXPORT_SYMBOL(scsi_report_device_reset); |
| 2487 | |
| 2488 | /** |
| 2489 | * scsi_ioctl_reset: explicitly reset a host/bus/target/device |
| 2490 | * @dev: scsi_device to operate on |
| 2491 | * @arg: reset type (see sg.h) |
| 2492 | */ |
| 2493 | int |
| 2494 | scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) |
| 2495 | { |
| 2496 | struct scsi_cmnd *scmd; |
| 2497 | struct Scsi_Host *shost = dev->host; |
| 2498 | struct request *rq; |
| 2499 | unsigned long flags; |
| 2500 | int error = 0, val; |
| 2501 | enum scsi_disposition rtn; |
| 2502 | |
| 2503 | if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) |
| 2504 | return -EACCES; |
| 2505 | |
| 2506 | error = get_user(val, arg); |
| 2507 | if (error) |
| 2508 | return error; |
| 2509 | |
| 2510 | if (scsi_autopm_get_host(shost) < 0) |
| 2511 | return -EIO; |
| 2512 | |
| 2513 | error = -EIO; |
| 2514 | rq = kzalloc(sizeof(struct request) + sizeof(struct scsi_cmnd) + |
| 2515 | shost->hostt->cmd_size, GFP_KERNEL); |
| 2516 | if (!rq) |
| 2517 | goto out_put_autopm_host; |
| 2518 | blk_rq_init(NULL, rq); |
| 2519 | |
| 2520 | scmd = (struct scsi_cmnd *)(rq + 1); |
| 2521 | scsi_init_command(dev, cmd: scmd); |
| 2522 | |
| 2523 | scmd->submitter = SUBMITTED_BY_SCSI_RESET_IOCTL; |
| 2524 | scmd->flags |= SCMD_LAST; |
| 2525 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
| 2526 | |
| 2527 | scmd->cmd_len = 0; |
| 2528 | |
| 2529 | scmd->sc_data_direction = DMA_BIDIRECTIONAL; |
| 2530 | |
| 2531 | spin_lock_irqsave(shost->host_lock, flags); |
| 2532 | shost->tmf_in_progress = 1; |
| 2533 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 2534 | |
| 2535 | switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { |
| 2536 | case SG_SCSI_RESET_NOTHING: |
| 2537 | rtn = SUCCESS; |
| 2538 | break; |
| 2539 | case SG_SCSI_RESET_DEVICE: |
| 2540 | rtn = scsi_try_bus_device_reset(scmd); |
| 2541 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
| 2542 | break; |
| 2543 | fallthrough; |
| 2544 | case SG_SCSI_RESET_TARGET: |
| 2545 | rtn = scsi_try_target_reset(scmd); |
| 2546 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
| 2547 | break; |
| 2548 | fallthrough; |
| 2549 | case SG_SCSI_RESET_BUS: |
| 2550 | rtn = scsi_try_bus_reset(scmd); |
| 2551 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
| 2552 | break; |
| 2553 | fallthrough; |
| 2554 | case SG_SCSI_RESET_HOST: |
| 2555 | rtn = scsi_try_host_reset(scmd); |
| 2556 | if (rtn == SUCCESS) |
| 2557 | break; |
| 2558 | fallthrough; |
| 2559 | default: |
| 2560 | rtn = FAILED; |
| 2561 | break; |
| 2562 | } |
| 2563 | |
| 2564 | error = (rtn == SUCCESS) ? 0 : -EIO; |
| 2565 | |
| 2566 | spin_lock_irqsave(shost->host_lock, flags); |
| 2567 | shost->tmf_in_progress = 0; |
| 2568 | spin_unlock_irqrestore(lock: shost->host_lock, flags); |
| 2569 | |
| 2570 | /* |
| 2571 | * be sure to wake up anyone who was sleeping or had their queue |
| 2572 | * suspended while we performed the TMF. |
| 2573 | */ |
| 2574 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2575 | shost_printk(KERN_INFO, shost, |
| 2576 | "waking up host to restart after TMF\n" )); |
| 2577 | |
| 2578 | wake_up(&shost->host_wait); |
| 2579 | scsi_run_host_queues(shost); |
| 2580 | |
| 2581 | kfree(objp: rq); |
| 2582 | |
| 2583 | out_put_autopm_host: |
| 2584 | scsi_autopm_put_host(shost); |
| 2585 | return error; |
| 2586 | } |
| 2587 | |
| 2588 | bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, |
| 2589 | struct scsi_sense_hdr *sshdr) |
| 2590 | { |
| 2591 | return scsi_normalize_sense(sense_buffer: cmd->sense_buffer, |
| 2592 | SCSI_SENSE_BUFFERSIZE, sshdr); |
| 2593 | } |
| 2594 | EXPORT_SYMBOL(scsi_command_normalize_sense); |
| 2595 | |
| 2596 | /** |
| 2597 | * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) |
| 2598 | * @sense_buffer: byte array of sense data |
| 2599 | * @sb_len: number of valid bytes in sense_buffer |
| 2600 | * @info_out: pointer to 64 integer where 8 or 4 byte information |
| 2601 | * field will be placed if found. |
| 2602 | * |
| 2603 | * Return value: |
| 2604 | * true if information field found, false if not found. |
| 2605 | */ |
| 2606 | bool scsi_get_sense_info_fld(const u8 *sense_buffer, int sb_len, |
| 2607 | u64 *info_out) |
| 2608 | { |
| 2609 | const u8 * ucp; |
| 2610 | |
| 2611 | if (sb_len < 7) |
| 2612 | return false; |
| 2613 | switch (sense_buffer[0] & 0x7f) { |
| 2614 | case 0x70: |
| 2615 | case 0x71: |
| 2616 | if (sense_buffer[0] & 0x80) { |
| 2617 | *info_out = get_unaligned_be32(p: &sense_buffer[3]); |
| 2618 | return true; |
| 2619 | } |
| 2620 | return false; |
| 2621 | case 0x72: |
| 2622 | case 0x73: |
| 2623 | ucp = scsi_sense_desc_find(sense_buffer, sb_len, |
| 2624 | desc_type: 0 /* info desc */); |
| 2625 | if (ucp && (0xa == ucp[1])) { |
| 2626 | *info_out = get_unaligned_be64(p: &ucp[4]); |
| 2627 | return true; |
| 2628 | } |
| 2629 | return false; |
| 2630 | default: |
| 2631 | return false; |
| 2632 | } |
| 2633 | } |
| 2634 | EXPORT_SYMBOL(scsi_get_sense_info_fld); |
| 2635 | |