Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame] | 1 | /* |
| 2 | * scsi_error.c Copyright (C) 1997 Eric Youngdale |
| 3 | * |
| 4 | * SCSI error/timeout handling |
| 5 | * Initial versions: Eric Youngdale. Based upon conversations with |
| 6 | * Leonard Zubkoff and David Miller at Linux Expo, |
| 7 | * ideas originating from all over the place. |
| 8 | * |
| 9 | * Restructured scsi_unjam_host and associated functions. |
| 10 | * September 04, 2002 Mike Anderson (andmike@us.ibm.com) |
| 11 | * |
| 12 | * Forward port of Russell King's (rmk@arm.linux.org.uk) changes and |
| 13 | * minor cleanups. |
| 14 | * September 30, 2002 Mike Anderson (andmike@us.ibm.com) |
| 15 | */ |
| 16 | |
| 17 | #include <linux/module.h> |
| 18 | #include <linux/sched.h> |
| 19 | #include <linux/gfp.h> |
| 20 | #include <linux/timer.h> |
| 21 | #include <linux/string.h> |
| 22 | #include <linux/kernel.h> |
| 23 | #include <linux/freezer.h> |
| 24 | #include <linux/kthread.h> |
| 25 | #include <linux/interrupt.h> |
| 26 | #include <linux/blkdev.h> |
| 27 | #include <linux/delay.h> |
| 28 | #include <linux/jiffies.h> |
| 29 | |
| 30 | #include <scsi/scsi.h> |
| 31 | #include <scsi/scsi_cmnd.h> |
| 32 | #include <scsi/scsi_dbg.h> |
| 33 | #include <scsi/scsi_device.h> |
| 34 | #include <scsi/scsi_driver.h> |
| 35 | #include <scsi/scsi_eh.h> |
| 36 | #include <scsi/scsi_common.h> |
| 37 | #include <scsi/scsi_transport.h> |
| 38 | #include <scsi/scsi_host.h> |
| 39 | #include <scsi/scsi_ioctl.h> |
| 40 | #include <scsi/scsi_dh.h> |
| 41 | #include <scsi/sg.h> |
| 42 | |
| 43 | #include "scsi_priv.h" |
| 44 | #include "scsi_logging.h" |
| 45 | #include "scsi_transport_api.h" |
| 46 | |
| 47 | #include <trace/events/scsi.h> |
| 48 | |
| 49 | static void scsi_eh_done(struct scsi_cmnd *scmd); |
| 50 | |
| 51 | /* |
| 52 | * These should *probably* be handled by the host itself. |
| 53 | * Since it is allowed to sleep, it probably should. |
| 54 | */ |
| 55 | #define BUS_RESET_SETTLE_TIME (10) |
| 56 | #define HOST_RESET_SETTLE_TIME (10) |
| 57 | |
| 58 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd); |
| 59 | static int scsi_try_to_abort_cmd(struct scsi_host_template *, |
| 60 | struct scsi_cmnd *); |
| 61 | |
| 62 | /* called with shost->host_lock held */ |
| 63 | void scsi_eh_wakeup(struct Scsi_Host *shost) |
| 64 | { |
| 65 | if (atomic_read(&shost->host_busy) == shost->host_failed) { |
| 66 | trace_scsi_eh_wakeup(shost); |
| 67 | wake_up_process(shost->ehandler); |
| 68 | SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost, |
| 69 | "Waking error handler thread\n")); |
| 70 | } |
| 71 | } |
| 72 | |
| 73 | /** |
| 74 | * scsi_schedule_eh - schedule EH for SCSI host |
| 75 | * @shost: SCSI host to invoke error handling on. |
| 76 | * |
| 77 | * Schedule SCSI EH without scmd. |
| 78 | */ |
| 79 | void scsi_schedule_eh(struct Scsi_Host *shost) |
| 80 | { |
| 81 | unsigned long flags; |
| 82 | |
| 83 | spin_lock_irqsave(shost->host_lock, flags); |
| 84 | |
| 85 | if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 || |
| 86 | scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) { |
| 87 | shost->host_eh_scheduled++; |
| 88 | scsi_eh_wakeup(shost); |
| 89 | } |
| 90 | |
| 91 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 92 | } |
| 93 | EXPORT_SYMBOL_GPL(scsi_schedule_eh); |
| 94 | |
| 95 | static int scsi_host_eh_past_deadline(struct Scsi_Host *shost) |
| 96 | { |
| 97 | if (!shost->last_reset || shost->eh_deadline == -1) |
| 98 | return 0; |
| 99 | |
| 100 | /* |
| 101 | * 32bit accesses are guaranteed to be atomic |
| 102 | * (on all supported architectures), so instead |
| 103 | * of using a spinlock we can as well double check |
| 104 | * if eh_deadline has been set to 'off' during the |
| 105 | * time_before call. |
| 106 | */ |
| 107 | if (time_before(jiffies, shost->last_reset + shost->eh_deadline) && |
| 108 | shost->eh_deadline > -1) |
| 109 | return 0; |
| 110 | |
| 111 | return 1; |
| 112 | } |
| 113 | |
| 114 | /** |
| 115 | * scmd_eh_abort_handler - Handle command aborts |
| 116 | * @work: command to be aborted. |
| 117 | */ |
| 118 | void |
| 119 | scmd_eh_abort_handler(struct work_struct *work) |
| 120 | { |
| 121 | struct scsi_cmnd *scmd = |
| 122 | container_of(work, struct scsi_cmnd, abort_work.work); |
| 123 | struct scsi_device *sdev = scmd->device; |
| 124 | int rtn; |
| 125 | |
| 126 | if (scsi_host_eh_past_deadline(sdev->host)) { |
| 127 | SCSI_LOG_ERROR_RECOVERY(3, |
| 128 | scmd_printk(KERN_INFO, scmd, |
| 129 | "eh timeout, not aborting\n")); |
| 130 | } else { |
| 131 | SCSI_LOG_ERROR_RECOVERY(3, |
| 132 | scmd_printk(KERN_INFO, scmd, |
| 133 | "aborting command\n")); |
| 134 | rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); |
| 135 | if (rtn == SUCCESS) { |
| 136 | set_host_byte(scmd, DID_TIME_OUT); |
| 137 | if (scsi_host_eh_past_deadline(sdev->host)) { |
| 138 | SCSI_LOG_ERROR_RECOVERY(3, |
| 139 | scmd_printk(KERN_INFO, scmd, |
| 140 | "eh timeout, not retrying " |
| 141 | "aborted command\n")); |
| 142 | } else if (!scsi_noretry_cmd(scmd) && |
| 143 | (++scmd->retries <= scmd->allowed)) { |
| 144 | SCSI_LOG_ERROR_RECOVERY(3, |
| 145 | scmd_printk(KERN_WARNING, scmd, |
| 146 | "retry aborted command\n")); |
| 147 | scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); |
| 148 | return; |
| 149 | } else { |
| 150 | SCSI_LOG_ERROR_RECOVERY(3, |
| 151 | scmd_printk(KERN_WARNING, scmd, |
| 152 | "finish aborted command\n")); |
| 153 | scsi_finish_command(scmd); |
| 154 | return; |
| 155 | } |
| 156 | } else { |
| 157 | SCSI_LOG_ERROR_RECOVERY(3, |
| 158 | scmd_printk(KERN_INFO, scmd, |
| 159 | "cmd abort %s\n", |
| 160 | (rtn == FAST_IO_FAIL) ? |
| 161 | "not send" : "failed")); |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | if (!scsi_eh_scmd_add(scmd, 0)) { |
| 166 | SCSI_LOG_ERROR_RECOVERY(3, |
| 167 | scmd_printk(KERN_WARNING, scmd, |
| 168 | "terminate aborted command\n")); |
| 169 | set_host_byte(scmd, DID_TIME_OUT); |
| 170 | scsi_finish_command(scmd); |
| 171 | } |
| 172 | } |
| 173 | |
| 174 | /** |
| 175 | * scsi_abort_command - schedule a command abort |
| 176 | * @scmd: scmd to abort. |
| 177 | * |
| 178 | * We only need to abort commands after a command timeout |
| 179 | */ |
| 180 | static int |
| 181 | scsi_abort_command(struct scsi_cmnd *scmd) |
| 182 | { |
| 183 | struct scsi_device *sdev = scmd->device; |
| 184 | struct Scsi_Host *shost = sdev->host; |
| 185 | unsigned long flags; |
| 186 | |
| 187 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { |
| 188 | /* |
| 189 | * Retry after abort failed, escalate to next level. |
| 190 | */ |
| 191 | scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED; |
| 192 | SCSI_LOG_ERROR_RECOVERY(3, |
| 193 | scmd_printk(KERN_INFO, scmd, |
| 194 | "previous abort failed\n")); |
| 195 | BUG_ON(delayed_work_pending(&scmd->abort_work)); |
| 196 | return FAILED; |
| 197 | } |
| 198 | |
| 199 | /* |
| 200 | * Do not try a command abort if |
| 201 | * SCSI EH has already started. |
| 202 | */ |
| 203 | spin_lock_irqsave(shost->host_lock, flags); |
| 204 | if (scsi_host_in_recovery(shost)) { |
| 205 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 206 | SCSI_LOG_ERROR_RECOVERY(3, |
| 207 | scmd_printk(KERN_INFO, scmd, |
| 208 | "not aborting, host in recovery\n")); |
| 209 | return FAILED; |
| 210 | } |
| 211 | |
| 212 | if (shost->eh_deadline != -1 && !shost->last_reset) |
| 213 | shost->last_reset = jiffies; |
| 214 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 215 | |
| 216 | scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED; |
| 217 | SCSI_LOG_ERROR_RECOVERY(3, |
| 218 | scmd_printk(KERN_INFO, scmd, "abort scheduled\n")); |
| 219 | queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100); |
| 220 | return SUCCESS; |
| 221 | } |
| 222 | |
| 223 | /** |
| 224 | * scsi_eh_scmd_add - add scsi cmd to error handling. |
| 225 | * @scmd: scmd to run eh on. |
| 226 | * @eh_flag: optional SCSI_EH flag. |
| 227 | * |
| 228 | * Return value: |
| 229 | * 0 on failure. |
| 230 | */ |
| 231 | int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag) |
| 232 | { |
| 233 | struct Scsi_Host *shost = scmd->device->host; |
| 234 | unsigned long flags; |
| 235 | int ret = 0; |
| 236 | |
| 237 | if (!shost->ehandler) |
| 238 | return 0; |
| 239 | |
| 240 | spin_lock_irqsave(shost->host_lock, flags); |
| 241 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) |
| 242 | if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)) |
| 243 | goto out_unlock; |
| 244 | |
| 245 | if (shost->eh_deadline != -1 && !shost->last_reset) |
| 246 | shost->last_reset = jiffies; |
| 247 | |
| 248 | ret = 1; |
| 249 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) |
| 250 | eh_flag &= ~SCSI_EH_CANCEL_CMD; |
| 251 | scmd->eh_eflags |= eh_flag; |
| 252 | list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q); |
| 253 | shost->host_failed++; |
| 254 | scsi_eh_wakeup(shost); |
| 255 | out_unlock: |
| 256 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 257 | return ret; |
| 258 | } |
| 259 | |
| 260 | /** |
| 261 | * scsi_times_out - Timeout function for normal scsi commands. |
| 262 | * @req: request that is timing out. |
| 263 | * |
| 264 | * Notes: |
| 265 | * We do not need to lock this. There is the potential for a race |
| 266 | * only in that the normal completion handling might run, but if the |
| 267 | * normal completion function determines that the timer has already |
| 268 | * fired, then it mustn't do anything. |
| 269 | */ |
| 270 | enum blk_eh_timer_return scsi_times_out(struct request *req) |
| 271 | { |
| 272 | struct scsi_cmnd *scmd = req->special; |
| 273 | enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED; |
| 274 | struct Scsi_Host *host = scmd->device->host; |
| 275 | |
| 276 | trace_scsi_dispatch_cmd_timeout(scmd); |
| 277 | scsi_log_completion(scmd, TIMEOUT_ERROR); |
| 278 | |
| 279 | if (host->eh_deadline != -1 && !host->last_reset) |
| 280 | host->last_reset = jiffies; |
| 281 | |
| 282 | if (host->transportt->eh_timed_out) |
| 283 | rtn = host->transportt->eh_timed_out(scmd); |
| 284 | else if (host->hostt->eh_timed_out) |
| 285 | rtn = host->hostt->eh_timed_out(scmd); |
| 286 | |
| 287 | if (rtn == BLK_EH_NOT_HANDLED) { |
| 288 | if (!host->hostt->no_async_abort && |
| 289 | scsi_abort_command(scmd) == SUCCESS) |
| 290 | return BLK_EH_NOT_HANDLED; |
| 291 | |
| 292 | set_host_byte(scmd, DID_TIME_OUT); |
| 293 | if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)) |
| 294 | rtn = BLK_EH_HANDLED; |
| 295 | } |
| 296 | |
| 297 | return rtn; |
| 298 | } |
| 299 | |
| 300 | /** |
| 301 | * scsi_block_when_processing_errors - Prevent cmds from being queued. |
| 302 | * @sdev: Device on which we are performing recovery. |
| 303 | * |
| 304 | * Description: |
| 305 | * We block until the host is out of error recovery, and then check to |
| 306 | * see whether the host or the device is offline. |
| 307 | * |
| 308 | * Return value: |
| 309 | * 0 when dev was taken offline by error recovery. 1 OK to proceed. |
| 310 | */ |
| 311 | int scsi_block_when_processing_errors(struct scsi_device *sdev) |
| 312 | { |
| 313 | int online; |
| 314 | |
| 315 | wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host)); |
| 316 | |
| 317 | online = scsi_device_online(sdev); |
| 318 | |
| 319 | SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev, |
| 320 | "%s: rtn: %d\n", __func__, online)); |
| 321 | |
| 322 | return online; |
| 323 | } |
| 324 | EXPORT_SYMBOL(scsi_block_when_processing_errors); |
| 325 | |
| 326 | #ifdef CONFIG_SCSI_LOGGING |
| 327 | /** |
| 328 | * scsi_eh_prt_fail_stats - Log info on failures. |
| 329 | * @shost: scsi host being recovered. |
| 330 | * @work_q: Queue of scsi cmds to process. |
| 331 | */ |
| 332 | static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost, |
| 333 | struct list_head *work_q) |
| 334 | { |
| 335 | struct scsi_cmnd *scmd; |
| 336 | struct scsi_device *sdev; |
| 337 | int total_failures = 0; |
| 338 | int cmd_failed = 0; |
| 339 | int cmd_cancel = 0; |
| 340 | int devices_failed = 0; |
| 341 | |
| 342 | shost_for_each_device(sdev, shost) { |
| 343 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 344 | if (scmd->device == sdev) { |
| 345 | ++total_failures; |
| 346 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) |
| 347 | ++cmd_cancel; |
| 348 | else |
| 349 | ++cmd_failed; |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | if (cmd_cancel || cmd_failed) { |
| 354 | SCSI_LOG_ERROR_RECOVERY(3, |
| 355 | shost_printk(KERN_INFO, shost, |
| 356 | "%s: cmds failed: %d, cancel: %d\n", |
| 357 | __func__, cmd_failed, |
| 358 | cmd_cancel)); |
| 359 | cmd_cancel = 0; |
| 360 | cmd_failed = 0; |
| 361 | ++devices_failed; |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost, |
| 366 | "Total of %d commands on %d" |
| 367 | " devices require eh work\n", |
| 368 | total_failures, devices_failed)); |
| 369 | } |
| 370 | #endif |
| 371 | |
| 372 | /** |
| 373 | * scsi_report_lun_change - Set flag on all *other* devices on the same target |
| 374 | * to indicate that a UNIT ATTENTION is expected. |
| 375 | * @sdev: Device reporting the UNIT ATTENTION |
| 376 | */ |
| 377 | static void scsi_report_lun_change(struct scsi_device *sdev) |
| 378 | { |
| 379 | sdev->sdev_target->expecting_lun_change = 1; |
| 380 | } |
| 381 | |
| 382 | /** |
| 383 | * scsi_report_sense - Examine scsi sense information and log messages for |
| 384 | * certain conditions, also issue uevents for some of them. |
| 385 | * @sdev: Device reporting the sense code |
| 386 | * @sshdr: sshdr to be examined |
| 387 | */ |
| 388 | static void scsi_report_sense(struct scsi_device *sdev, |
| 389 | struct scsi_sense_hdr *sshdr) |
| 390 | { |
| 391 | enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */ |
| 392 | |
| 393 | if (sshdr->sense_key == UNIT_ATTENTION) { |
| 394 | if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) { |
| 395 | evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED; |
| 396 | sdev_printk(KERN_WARNING, sdev, |
| 397 | "Inquiry data has changed"); |
| 398 | } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) { |
| 399 | evt_type = SDEV_EVT_LUN_CHANGE_REPORTED; |
| 400 | scsi_report_lun_change(sdev); |
| 401 | sdev_printk(KERN_WARNING, sdev, |
| 402 | "Warning! Received an indication that the " |
| 403 | "LUN assignments on this target have " |
| 404 | "changed. The Linux SCSI layer does not " |
| 405 | "automatically remap LUN assignments.\n"); |
| 406 | } else if (sshdr->asc == 0x3f) |
| 407 | sdev_printk(KERN_WARNING, sdev, |
| 408 | "Warning! Received an indication that the " |
| 409 | "operating parameters on this target have " |
| 410 | "changed. The Linux SCSI layer does not " |
| 411 | "automatically adjust these parameters.\n"); |
| 412 | |
| 413 | if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) { |
| 414 | evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED; |
| 415 | sdev_printk(KERN_WARNING, sdev, |
| 416 | "Warning! Received an indication that the " |
| 417 | "LUN reached a thin provisioning soft " |
| 418 | "threshold.\n"); |
| 419 | } |
| 420 | |
| 421 | if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) { |
| 422 | evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED; |
| 423 | sdev_printk(KERN_WARNING, sdev, |
| 424 | "Mode parameters changed"); |
| 425 | } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) { |
| 426 | evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED; |
| 427 | sdev_printk(KERN_WARNING, sdev, |
| 428 | "Asymmetric access state changed"); |
| 429 | } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) { |
| 430 | evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED; |
| 431 | sdev_printk(KERN_WARNING, sdev, |
| 432 | "Capacity data has changed"); |
| 433 | } else if (sshdr->asc == 0x2a) |
| 434 | sdev_printk(KERN_WARNING, sdev, |
| 435 | "Parameters changed"); |
| 436 | } |
| 437 | |
| 438 | if (evt_type != SDEV_EVT_MAXBITS) { |
| 439 | set_bit(evt_type, sdev->pending_events); |
| 440 | schedule_work(&sdev->event_work); |
| 441 | } |
| 442 | } |
| 443 | |
| 444 | /** |
| 445 | * scsi_check_sense - Examine scsi cmd sense |
| 446 | * @scmd: Cmd to have sense checked. |
| 447 | * |
| 448 | * Return value: |
| 449 | * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE |
| 450 | * |
| 451 | * Notes: |
| 452 | * When a deferred error is detected the current command has |
| 453 | * not been executed and needs retrying. |
| 454 | */ |
| 455 | static int scsi_check_sense(struct scsi_cmnd *scmd) |
| 456 | { |
| 457 | struct scsi_device *sdev = scmd->device; |
| 458 | struct scsi_sense_hdr sshdr; |
| 459 | |
| 460 | if (! scsi_command_normalize_sense(scmd, &sshdr)) |
| 461 | return FAILED; /* no valid sense data */ |
| 462 | |
| 463 | scsi_report_sense(sdev, &sshdr); |
| 464 | |
| 465 | if (scsi_sense_is_deferred(&sshdr)) |
| 466 | return NEEDS_RETRY; |
| 467 | |
| 468 | if (sdev->handler && sdev->handler->check_sense) { |
| 469 | int rc; |
| 470 | |
| 471 | rc = sdev->handler->check_sense(sdev, &sshdr); |
| 472 | if (rc != SCSI_RETURN_NOT_HANDLED) |
| 473 | return rc; |
| 474 | /* handler does not care. Drop down to default handling */ |
| 475 | } |
| 476 | |
| 477 | if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done) |
| 478 | /* |
| 479 | * nasty: for mid-layer issued TURs, we need to return the |
| 480 | * actual sense data without any recovery attempt. For eh |
| 481 | * issued ones, we need to try to recover and interpret |
| 482 | */ |
| 483 | return SUCCESS; |
| 484 | |
| 485 | /* |
| 486 | * Previous logic looked for FILEMARK, EOM or ILI which are |
| 487 | * mainly associated with tapes and returned SUCCESS. |
| 488 | */ |
| 489 | if (sshdr.response_code == 0x70) { |
| 490 | /* fixed format */ |
| 491 | if (scmd->sense_buffer[2] & 0xe0) |
| 492 | return SUCCESS; |
| 493 | } else { |
| 494 | /* |
| 495 | * descriptor format: look for "stream commands sense data |
| 496 | * descriptor" (see SSC-3). Assume single sense data |
| 497 | * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG. |
| 498 | */ |
| 499 | if ((sshdr.additional_length > 3) && |
| 500 | (scmd->sense_buffer[8] == 0x4) && |
| 501 | (scmd->sense_buffer[11] & 0xe0)) |
| 502 | return SUCCESS; |
| 503 | } |
| 504 | |
| 505 | switch (sshdr.sense_key) { |
| 506 | case NO_SENSE: |
| 507 | return SUCCESS; |
| 508 | case RECOVERED_ERROR: |
| 509 | return /* soft_error */ SUCCESS; |
| 510 | |
| 511 | case ABORTED_COMMAND: |
| 512 | if (sshdr.asc == 0x10) /* DIF */ |
| 513 | return SUCCESS; |
| 514 | |
| 515 | return NEEDS_RETRY; |
| 516 | case NOT_READY: |
| 517 | case UNIT_ATTENTION: |
| 518 | /* |
| 519 | * if we are expecting a cc/ua because of a bus reset that we |
| 520 | * performed, treat this just as a retry. otherwise this is |
| 521 | * information that we should pass up to the upper-level driver |
| 522 | * so that we can deal with it there. |
| 523 | */ |
| 524 | if (scmd->device->expecting_cc_ua) { |
| 525 | /* |
| 526 | * Because some device does not queue unit |
| 527 | * attentions correctly, we carefully check |
| 528 | * additional sense code and qualifier so as |
| 529 | * not to squash media change unit attention. |
| 530 | */ |
| 531 | if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) { |
| 532 | scmd->device->expecting_cc_ua = 0; |
| 533 | return NEEDS_RETRY; |
| 534 | } |
| 535 | } |
| 536 | /* |
| 537 | * we might also expect a cc/ua if another LUN on the target |
| 538 | * reported a UA with an ASC/ASCQ of 3F 0E - |
| 539 | * REPORTED LUNS DATA HAS CHANGED. |
| 540 | */ |
| 541 | if (scmd->device->sdev_target->expecting_lun_change && |
| 542 | sshdr.asc == 0x3f && sshdr.ascq == 0x0e) |
| 543 | return NEEDS_RETRY; |
| 544 | /* |
| 545 | * if the device is in the process of becoming ready, we |
| 546 | * should retry. |
| 547 | */ |
| 548 | if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01)) |
| 549 | return NEEDS_RETRY; |
| 550 | /* |
| 551 | * if the device is not started, we need to wake |
| 552 | * the error handler to start the motor |
| 553 | */ |
| 554 | if (scmd->device->allow_restart && |
| 555 | (sshdr.asc == 0x04) && (sshdr.ascq == 0x02)) |
| 556 | return FAILED; |
| 557 | /* |
| 558 | * Pass the UA upwards for a determination in the completion |
| 559 | * functions. |
| 560 | */ |
| 561 | return SUCCESS; |
| 562 | |
| 563 | /* these are not supported */ |
| 564 | case DATA_PROTECT: |
| 565 | if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) { |
| 566 | /* Thin provisioning hard threshold reached */ |
| 567 | set_host_byte(scmd, DID_ALLOC_FAILURE); |
| 568 | return SUCCESS; |
| 569 | } |
| 570 | case COPY_ABORTED: |
| 571 | case VOLUME_OVERFLOW: |
| 572 | case MISCOMPARE: |
| 573 | case BLANK_CHECK: |
| 574 | set_host_byte(scmd, DID_TARGET_FAILURE); |
| 575 | return SUCCESS; |
| 576 | |
| 577 | case MEDIUM_ERROR: |
| 578 | if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */ |
| 579 | sshdr.asc == 0x13 || /* AMNF DATA FIELD */ |
| 580 | sshdr.asc == 0x14) { /* RECORD NOT FOUND */ |
| 581 | set_host_byte(scmd, DID_MEDIUM_ERROR); |
| 582 | return SUCCESS; |
| 583 | } |
| 584 | return NEEDS_RETRY; |
| 585 | |
| 586 | case HARDWARE_ERROR: |
| 587 | if (scmd->device->retry_hwerror) |
| 588 | return ADD_TO_MLQUEUE; |
| 589 | else |
| 590 | set_host_byte(scmd, DID_TARGET_FAILURE); |
| 591 | |
| 592 | case ILLEGAL_REQUEST: |
| 593 | if (sshdr.asc == 0x20 || /* Invalid command operation code */ |
| 594 | sshdr.asc == 0x21 || /* Logical block address out of range */ |
| 595 | sshdr.asc == 0x24 || /* Invalid field in cdb */ |
| 596 | sshdr.asc == 0x26) { /* Parameter value invalid */ |
| 597 | set_host_byte(scmd, DID_TARGET_FAILURE); |
| 598 | } |
| 599 | return SUCCESS; |
| 600 | |
| 601 | default: |
| 602 | return SUCCESS; |
| 603 | } |
| 604 | } |
| 605 | |
| 606 | static void scsi_handle_queue_ramp_up(struct scsi_device *sdev) |
| 607 | { |
| 608 | struct scsi_host_template *sht = sdev->host->hostt; |
| 609 | struct scsi_device *tmp_sdev; |
| 610 | |
| 611 | if (!sht->track_queue_depth || |
| 612 | sdev->queue_depth >= sdev->max_queue_depth) |
| 613 | return; |
| 614 | |
| 615 | if (time_before(jiffies, |
| 616 | sdev->last_queue_ramp_up + sdev->queue_ramp_up_period)) |
| 617 | return; |
| 618 | |
| 619 | if (time_before(jiffies, |
| 620 | sdev->last_queue_full_time + sdev->queue_ramp_up_period)) |
| 621 | return; |
| 622 | |
| 623 | /* |
| 624 | * Walk all devices of a target and do |
| 625 | * ramp up on them. |
| 626 | */ |
| 627 | shost_for_each_device(tmp_sdev, sdev->host) { |
| 628 | if (tmp_sdev->channel != sdev->channel || |
| 629 | tmp_sdev->id != sdev->id || |
| 630 | tmp_sdev->queue_depth == sdev->max_queue_depth) |
| 631 | continue; |
| 632 | |
| 633 | scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1); |
| 634 | sdev->last_queue_ramp_up = jiffies; |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | static void scsi_handle_queue_full(struct scsi_device *sdev) |
| 639 | { |
| 640 | struct scsi_host_template *sht = sdev->host->hostt; |
| 641 | struct scsi_device *tmp_sdev; |
| 642 | |
| 643 | if (!sht->track_queue_depth) |
| 644 | return; |
| 645 | |
| 646 | shost_for_each_device(tmp_sdev, sdev->host) { |
| 647 | if (tmp_sdev->channel != sdev->channel || |
| 648 | tmp_sdev->id != sdev->id) |
| 649 | continue; |
| 650 | /* |
| 651 | * We do not know the number of commands that were at |
| 652 | * the device when we got the queue full so we start |
| 653 | * from the highest possible value and work our way down. |
| 654 | */ |
| 655 | scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1); |
| 656 | } |
| 657 | } |
| 658 | |
| 659 | /** |
| 660 | * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD. |
| 661 | * @scmd: SCSI cmd to examine. |
| 662 | * |
| 663 | * Notes: |
| 664 | * This is *only* called when we are examining the status of commands |
| 665 | * queued during error recovery. the main difference here is that we |
| 666 | * don't allow for the possibility of retries here, and we are a lot |
| 667 | * more restrictive about what we consider acceptable. |
| 668 | */ |
| 669 | static int scsi_eh_completed_normally(struct scsi_cmnd *scmd) |
| 670 | { |
| 671 | /* |
| 672 | * first check the host byte, to see if there is anything in there |
| 673 | * that would indicate what we need to do. |
| 674 | */ |
| 675 | if (host_byte(scmd->result) == DID_RESET) { |
| 676 | /* |
| 677 | * rats. we are already in the error handler, so we now |
| 678 | * get to try and figure out what to do next. if the sense |
| 679 | * is valid, we have a pretty good idea of what to do. |
| 680 | * if not, we mark it as FAILED. |
| 681 | */ |
| 682 | return scsi_check_sense(scmd); |
| 683 | } |
| 684 | if (host_byte(scmd->result) != DID_OK) |
| 685 | return FAILED; |
| 686 | |
| 687 | /* |
| 688 | * next, check the message byte. |
| 689 | */ |
| 690 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) |
| 691 | return FAILED; |
| 692 | |
| 693 | /* |
| 694 | * now, check the status byte to see if this indicates |
| 695 | * anything special. |
| 696 | */ |
| 697 | switch (status_byte(scmd->result)) { |
| 698 | case GOOD: |
| 699 | scsi_handle_queue_ramp_up(scmd->device); |
| 700 | case COMMAND_TERMINATED: |
| 701 | return SUCCESS; |
| 702 | case CHECK_CONDITION: |
| 703 | return scsi_check_sense(scmd); |
| 704 | case CONDITION_GOOD: |
| 705 | case INTERMEDIATE_GOOD: |
| 706 | case INTERMEDIATE_C_GOOD: |
| 707 | /* |
| 708 | * who knows? FIXME(eric) |
| 709 | */ |
| 710 | return SUCCESS; |
| 711 | case RESERVATION_CONFLICT: |
| 712 | if (scmd->cmnd[0] == TEST_UNIT_READY) |
| 713 | /* it is a success, we probed the device and |
| 714 | * found it */ |
| 715 | return SUCCESS; |
| 716 | /* otherwise, we failed to send the command */ |
| 717 | return FAILED; |
| 718 | case QUEUE_FULL: |
| 719 | scsi_handle_queue_full(scmd->device); |
| 720 | /* fall through */ |
| 721 | case BUSY: |
| 722 | return NEEDS_RETRY; |
| 723 | default: |
| 724 | return FAILED; |
| 725 | } |
| 726 | return FAILED; |
| 727 | } |
| 728 | |
| 729 | /** |
| 730 | * scsi_eh_done - Completion function for error handling. |
| 731 | * @scmd: Cmd that is done. |
| 732 | */ |
| 733 | static void scsi_eh_done(struct scsi_cmnd *scmd) |
| 734 | { |
| 735 | struct completion *eh_action; |
| 736 | |
| 737 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 738 | "%s result: %x\n", __func__, scmd->result)); |
| 739 | |
| 740 | eh_action = scmd->device->host->eh_action; |
| 741 | if (eh_action) |
| 742 | complete(eh_action); |
| 743 | } |
| 744 | |
| 745 | /** |
| 746 | * scsi_try_host_reset - ask host adapter to reset itself |
| 747 | * @scmd: SCSI cmd to send host reset. |
| 748 | */ |
| 749 | static int scsi_try_host_reset(struct scsi_cmnd *scmd) |
| 750 | { |
| 751 | unsigned long flags; |
| 752 | int rtn; |
| 753 | struct Scsi_Host *host = scmd->device->host; |
| 754 | struct scsi_host_template *hostt = host->hostt; |
| 755 | |
| 756 | SCSI_LOG_ERROR_RECOVERY(3, |
| 757 | shost_printk(KERN_INFO, host, "Snd Host RST\n")); |
| 758 | |
| 759 | if (!hostt->eh_host_reset_handler) |
| 760 | return FAILED; |
| 761 | |
| 762 | rtn = hostt->eh_host_reset_handler(scmd); |
| 763 | |
| 764 | if (rtn == SUCCESS) { |
| 765 | if (!hostt->skip_settle_delay) |
| 766 | ssleep(HOST_RESET_SETTLE_TIME); |
| 767 | spin_lock_irqsave(host->host_lock, flags); |
| 768 | scsi_report_bus_reset(host, scmd_channel(scmd)); |
| 769 | spin_unlock_irqrestore(host->host_lock, flags); |
| 770 | } |
| 771 | |
| 772 | return rtn; |
| 773 | } |
| 774 | |
| 775 | /** |
| 776 | * scsi_try_bus_reset - ask host to perform a bus reset |
| 777 | * @scmd: SCSI cmd to send bus reset. |
| 778 | */ |
| 779 | static int scsi_try_bus_reset(struct scsi_cmnd *scmd) |
| 780 | { |
| 781 | unsigned long flags; |
| 782 | int rtn; |
| 783 | struct Scsi_Host *host = scmd->device->host; |
| 784 | struct scsi_host_template *hostt = host->hostt; |
| 785 | |
| 786 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 787 | "%s: Snd Bus RST\n", __func__)); |
| 788 | |
| 789 | if (!hostt->eh_bus_reset_handler) |
| 790 | return FAILED; |
| 791 | |
| 792 | rtn = hostt->eh_bus_reset_handler(scmd); |
| 793 | |
| 794 | if (rtn == SUCCESS) { |
| 795 | if (!hostt->skip_settle_delay) |
| 796 | ssleep(BUS_RESET_SETTLE_TIME); |
| 797 | spin_lock_irqsave(host->host_lock, flags); |
| 798 | scsi_report_bus_reset(host, scmd_channel(scmd)); |
| 799 | spin_unlock_irqrestore(host->host_lock, flags); |
| 800 | } |
| 801 | |
| 802 | return rtn; |
| 803 | } |
| 804 | |
| 805 | static void __scsi_report_device_reset(struct scsi_device *sdev, void *data) |
| 806 | { |
| 807 | sdev->was_reset = 1; |
| 808 | sdev->expecting_cc_ua = 1; |
| 809 | } |
| 810 | |
| 811 | /** |
| 812 | * scsi_try_target_reset - Ask host to perform a target reset |
| 813 | * @scmd: SCSI cmd used to send a target reset |
| 814 | * |
| 815 | * Notes: |
| 816 | * There is no timeout for this operation. if this operation is |
| 817 | * unreliable for a given host, then the host itself needs to put a |
| 818 | * timer on it, and set the host back to a consistent state prior to |
| 819 | * returning. |
| 820 | */ |
| 821 | static int scsi_try_target_reset(struct scsi_cmnd *scmd) |
| 822 | { |
| 823 | unsigned long flags; |
| 824 | int rtn; |
| 825 | struct Scsi_Host *host = scmd->device->host; |
| 826 | struct scsi_host_template *hostt = host->hostt; |
| 827 | |
| 828 | if (!hostt->eh_target_reset_handler) |
| 829 | return FAILED; |
| 830 | |
| 831 | rtn = hostt->eh_target_reset_handler(scmd); |
| 832 | if (rtn == SUCCESS) { |
| 833 | spin_lock_irqsave(host->host_lock, flags); |
| 834 | __starget_for_each_device(scsi_target(scmd->device), NULL, |
| 835 | __scsi_report_device_reset); |
| 836 | spin_unlock_irqrestore(host->host_lock, flags); |
| 837 | } |
| 838 | |
| 839 | return rtn; |
| 840 | } |
| 841 | |
| 842 | /** |
| 843 | * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev |
| 844 | * @scmd: SCSI cmd used to send BDR |
| 845 | * |
| 846 | * Notes: |
| 847 | * There is no timeout for this operation. if this operation is |
| 848 | * unreliable for a given host, then the host itself needs to put a |
| 849 | * timer on it, and set the host back to a consistent state prior to |
| 850 | * returning. |
| 851 | */ |
| 852 | static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd) |
| 853 | { |
| 854 | int rtn; |
| 855 | struct scsi_host_template *hostt = scmd->device->host->hostt; |
| 856 | |
| 857 | if (!hostt->eh_device_reset_handler) |
| 858 | return FAILED; |
| 859 | |
| 860 | rtn = hostt->eh_device_reset_handler(scmd); |
| 861 | if (rtn == SUCCESS) |
| 862 | __scsi_report_device_reset(scmd->device, NULL); |
| 863 | return rtn; |
| 864 | } |
| 865 | |
| 866 | /** |
| 867 | * scsi_try_to_abort_cmd - Ask host to abort a SCSI command |
| 868 | * @hostt: SCSI driver host template |
| 869 | * @scmd: SCSI cmd used to send a target reset |
| 870 | * |
| 871 | * Return value: |
| 872 | * SUCCESS, FAILED, or FAST_IO_FAIL |
| 873 | * |
| 874 | * Notes: |
| 875 | * SUCCESS does not necessarily indicate that the command |
| 876 | * has been aborted; it only indicates that the LLDDs |
| 877 | * has cleared all references to that command. |
| 878 | * LLDDs should return FAILED only if an abort was required |
| 879 | * but could not be executed. LLDDs should return FAST_IO_FAIL |
| 880 | * if the device is temporarily unavailable (eg due to a |
| 881 | * link down on FibreChannel) |
| 882 | */ |
| 883 | static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, |
| 884 | struct scsi_cmnd *scmd) |
| 885 | { |
| 886 | if (!hostt->eh_abort_handler) |
| 887 | return FAILED; |
| 888 | |
| 889 | return hostt->eh_abort_handler(scmd); |
| 890 | } |
| 891 | |
| 892 | static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd) |
| 893 | { |
| 894 | if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS) |
| 895 | if (scsi_try_bus_device_reset(scmd) != SUCCESS) |
| 896 | if (scsi_try_target_reset(scmd) != SUCCESS) |
| 897 | if (scsi_try_bus_reset(scmd) != SUCCESS) |
| 898 | scsi_try_host_reset(scmd); |
| 899 | } |
| 900 | |
| 901 | /** |
| 902 | * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery |
| 903 | * @scmd: SCSI command structure to hijack |
| 904 | * @ses: structure to save restore information |
| 905 | * @cmnd: CDB to send. Can be NULL if no new cmnd is needed |
| 906 | * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB) |
| 907 | * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored) |
| 908 | * |
| 909 | * This function is used to save a scsi command information before re-execution |
| 910 | * as part of the error recovery process. If @sense_bytes is 0 the command |
| 911 | * sent must be one that does not transfer any data. If @sense_bytes != 0 |
| 912 | * @cmnd is ignored and this functions sets up a REQUEST_SENSE command |
| 913 | * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer. |
| 914 | */ |
| 915 | void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses, |
| 916 | unsigned char *cmnd, int cmnd_size, unsigned sense_bytes) |
| 917 | { |
| 918 | struct scsi_device *sdev = scmd->device; |
| 919 | |
| 920 | /* |
| 921 | * We need saved copies of a number of fields - this is because |
| 922 | * error handling may need to overwrite these with different values |
| 923 | * to run different commands, and once error handling is complete, |
| 924 | * we will need to restore these values prior to running the actual |
| 925 | * command. |
| 926 | */ |
| 927 | ses->cmd_len = scmd->cmd_len; |
| 928 | ses->cmnd = scmd->cmnd; |
| 929 | ses->data_direction = scmd->sc_data_direction; |
| 930 | ses->sdb = scmd->sdb; |
| 931 | ses->next_rq = scmd->request->next_rq; |
| 932 | ses->result = scmd->result; |
| 933 | ses->underflow = scmd->underflow; |
| 934 | ses->prot_op = scmd->prot_op; |
| 935 | |
| 936 | scmd->prot_op = SCSI_PROT_NORMAL; |
| 937 | scmd->eh_eflags = 0; |
| 938 | scmd->cmnd = ses->eh_cmnd; |
| 939 | memset(scmd->cmnd, 0, BLK_MAX_CDB); |
| 940 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
| 941 | scmd->request->next_rq = NULL; |
| 942 | scmd->result = 0; |
| 943 | |
| 944 | if (sense_bytes) { |
| 945 | scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE, |
| 946 | sense_bytes); |
| 947 | sg_init_one(&ses->sense_sgl, scmd->sense_buffer, |
| 948 | scmd->sdb.length); |
| 949 | scmd->sdb.table.sgl = &ses->sense_sgl; |
| 950 | scmd->sc_data_direction = DMA_FROM_DEVICE; |
| 951 | scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1; |
| 952 | scmd->cmnd[0] = REQUEST_SENSE; |
| 953 | scmd->cmnd[4] = scmd->sdb.length; |
| 954 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
| 955 | } else { |
| 956 | scmd->sc_data_direction = DMA_NONE; |
| 957 | if (cmnd) { |
| 958 | BUG_ON(cmnd_size > BLK_MAX_CDB); |
| 959 | memcpy(scmd->cmnd, cmnd, cmnd_size); |
| 960 | scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]); |
| 961 | } |
| 962 | } |
| 963 | |
| 964 | scmd->underflow = 0; |
| 965 | |
| 966 | if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN) |
| 967 | scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) | |
| 968 | (sdev->lun << 5 & 0xe0); |
| 969 | |
| 970 | /* |
| 971 | * Zero the sense buffer. The scsi spec mandates that any |
| 972 | * untransferred sense data should be interpreted as being zero. |
| 973 | */ |
| 974 | memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); |
| 975 | } |
| 976 | EXPORT_SYMBOL(scsi_eh_prep_cmnd); |
| 977 | |
| 978 | /** |
| 979 | * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery |
| 980 | * @scmd: SCSI command structure to restore |
| 981 | * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd |
| 982 | * |
| 983 | * Undo any damage done by above scsi_eh_prep_cmnd(). |
| 984 | */ |
| 985 | void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses) |
| 986 | { |
| 987 | /* |
| 988 | * Restore original data |
| 989 | */ |
| 990 | scmd->cmd_len = ses->cmd_len; |
| 991 | scmd->cmnd = ses->cmnd; |
| 992 | scmd->sc_data_direction = ses->data_direction; |
| 993 | scmd->sdb = ses->sdb; |
| 994 | scmd->request->next_rq = ses->next_rq; |
| 995 | scmd->result = ses->result; |
| 996 | scmd->underflow = ses->underflow; |
| 997 | scmd->prot_op = ses->prot_op; |
| 998 | } |
| 999 | EXPORT_SYMBOL(scsi_eh_restore_cmnd); |
| 1000 | |
| 1001 | /** |
| 1002 | * scsi_send_eh_cmnd - submit a scsi command as part of error recovery |
| 1003 | * @scmd: SCSI command structure to hijack |
| 1004 | * @cmnd: CDB to send |
| 1005 | * @cmnd_size: size in bytes of @cmnd |
| 1006 | * @timeout: timeout for this request |
| 1007 | * @sense_bytes: size of sense data to copy or 0 |
| 1008 | * |
| 1009 | * This function is used to send a scsi command down to a target device |
| 1010 | * as part of the error recovery process. See also scsi_eh_prep_cmnd() above. |
| 1011 | * |
| 1012 | * Return value: |
| 1013 | * SUCCESS or FAILED or NEEDS_RETRY |
| 1014 | */ |
| 1015 | static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd, |
| 1016 | int cmnd_size, int timeout, unsigned sense_bytes) |
| 1017 | { |
| 1018 | struct scsi_device *sdev = scmd->device; |
| 1019 | struct Scsi_Host *shost = sdev->host; |
| 1020 | DECLARE_COMPLETION_ONSTACK(done); |
| 1021 | unsigned long timeleft = timeout; |
| 1022 | struct scsi_eh_save ses; |
| 1023 | const unsigned long stall_for = msecs_to_jiffies(100); |
| 1024 | int rtn; |
| 1025 | |
| 1026 | retry: |
| 1027 | scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes); |
| 1028 | shost->eh_action = &done; |
| 1029 | |
| 1030 | scsi_log_send(scmd); |
| 1031 | scmd->scsi_done = scsi_eh_done; |
| 1032 | rtn = shost->hostt->queuecommand(shost, scmd); |
| 1033 | if (rtn) { |
| 1034 | if (timeleft > stall_for) { |
| 1035 | scsi_eh_restore_cmnd(scmd, &ses); |
| 1036 | timeleft -= stall_for; |
| 1037 | msleep(jiffies_to_msecs(stall_for)); |
| 1038 | goto retry; |
| 1039 | } |
| 1040 | /* signal not to enter either branch of the if () below */ |
| 1041 | timeleft = 0; |
| 1042 | rtn = FAILED; |
| 1043 | } else { |
| 1044 | timeleft = wait_for_completion_timeout(&done, timeout); |
| 1045 | rtn = SUCCESS; |
| 1046 | } |
| 1047 | |
| 1048 | shost->eh_action = NULL; |
| 1049 | |
| 1050 | scsi_log_completion(scmd, rtn); |
| 1051 | |
| 1052 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1053 | "%s timeleft: %ld\n", |
| 1054 | __func__, timeleft)); |
| 1055 | |
| 1056 | /* |
| 1057 | * If there is time left scsi_eh_done got called, and we will examine |
| 1058 | * the actual status codes to see whether the command actually did |
| 1059 | * complete normally, else if we have a zero return and no time left, |
| 1060 | * the command must still be pending, so abort it and return FAILED. |
| 1061 | * If we never actually managed to issue the command, because |
| 1062 | * ->queuecommand() kept returning non zero, use the rtn = FAILED |
| 1063 | * value above (so don't execute either branch of the if) |
| 1064 | */ |
| 1065 | if (timeleft) { |
| 1066 | rtn = scsi_eh_completed_normally(scmd); |
| 1067 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1068 | "%s: scsi_eh_completed_normally %x\n", __func__, rtn)); |
| 1069 | |
| 1070 | switch (rtn) { |
| 1071 | case SUCCESS: |
| 1072 | case NEEDS_RETRY: |
| 1073 | case FAILED: |
| 1074 | break; |
| 1075 | case ADD_TO_MLQUEUE: |
| 1076 | rtn = NEEDS_RETRY; |
| 1077 | break; |
| 1078 | default: |
| 1079 | rtn = FAILED; |
| 1080 | break; |
| 1081 | } |
| 1082 | } else if (rtn != FAILED) { |
| 1083 | scsi_abort_eh_cmnd(scmd); |
| 1084 | rtn = FAILED; |
| 1085 | } |
| 1086 | |
| 1087 | scsi_eh_restore_cmnd(scmd, &ses); |
| 1088 | |
| 1089 | return rtn; |
| 1090 | } |
| 1091 | |
| 1092 | /** |
| 1093 | * scsi_request_sense - Request sense data from a particular target. |
| 1094 | * @scmd: SCSI cmd for request sense. |
| 1095 | * |
| 1096 | * Notes: |
| 1097 | * Some hosts automatically obtain this information, others require |
| 1098 | * that we obtain it on our own. This function will *not* return until |
| 1099 | * the command either times out, or it completes. |
| 1100 | */ |
| 1101 | static int scsi_request_sense(struct scsi_cmnd *scmd) |
| 1102 | { |
| 1103 | return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0); |
| 1104 | } |
| 1105 | |
| 1106 | static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn) |
| 1107 | { |
| 1108 | if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) { |
| 1109 | struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd); |
| 1110 | if (sdrv->eh_action) |
| 1111 | rtn = sdrv->eh_action(scmd, rtn); |
| 1112 | } |
| 1113 | return rtn; |
| 1114 | } |
| 1115 | |
| 1116 | /** |
| 1117 | * scsi_eh_finish_cmd - Handle a cmd that eh is finished with. |
| 1118 | * @scmd: Original SCSI cmd that eh has finished. |
| 1119 | * @done_q: Queue for processed commands. |
| 1120 | * |
| 1121 | * Notes: |
| 1122 | * We don't want to use the normal command completion while we are are |
| 1123 | * still handling errors - it may cause other commands to be queued, |
| 1124 | * and that would disturb what we are doing. Thus we really want to |
| 1125 | * keep a list of pending commands for final completion, and once we |
| 1126 | * are ready to leave error handling we handle completion for real. |
| 1127 | */ |
| 1128 | void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q) |
| 1129 | { |
| 1130 | scmd->eh_eflags = 0; |
| 1131 | list_move_tail(&scmd->eh_entry, done_q); |
| 1132 | } |
| 1133 | EXPORT_SYMBOL(scsi_eh_finish_cmd); |
| 1134 | |
| 1135 | /** |
| 1136 | * scsi_eh_get_sense - Get device sense data. |
| 1137 | * @work_q: Queue of commands to process. |
| 1138 | * @done_q: Queue of processed commands. |
| 1139 | * |
| 1140 | * Description: |
| 1141 | * See if we need to request sense information. if so, then get it |
| 1142 | * now, so we have a better idea of what to do. |
| 1143 | * |
| 1144 | * Notes: |
| 1145 | * This has the unfortunate side effect that if a shost adapter does |
| 1146 | * not automatically request sense information, we end up shutting |
| 1147 | * it down before we request it. |
| 1148 | * |
| 1149 | * All drivers should request sense information internally these days, |
| 1150 | * so for now all I have to say is tough noogies if you end up in here. |
| 1151 | * |
| 1152 | * XXX: Long term this code should go away, but that needs an audit of |
| 1153 | * all LLDDs first. |
| 1154 | */ |
| 1155 | int scsi_eh_get_sense(struct list_head *work_q, |
| 1156 | struct list_head *done_q) |
| 1157 | { |
| 1158 | struct scsi_cmnd *scmd, *next; |
| 1159 | struct Scsi_Host *shost; |
| 1160 | int rtn; |
| 1161 | |
| 1162 | /* |
| 1163 | * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO, |
| 1164 | * should not get sense. |
| 1165 | */ |
| 1166 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1167 | if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) || |
| 1168 | (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) || |
| 1169 | SCSI_SENSE_VALID(scmd)) |
| 1170 | continue; |
| 1171 | |
| 1172 | shost = scmd->device->host; |
| 1173 | if (scsi_host_eh_past_deadline(shost)) { |
| 1174 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1175 | scmd_printk(KERN_INFO, scmd, |
| 1176 | "%s: skip request sense, past eh deadline\n", |
| 1177 | current->comm)); |
| 1178 | break; |
| 1179 | } |
| 1180 | if (status_byte(scmd->result) != CHECK_CONDITION) |
| 1181 | /* |
| 1182 | * don't request sense if there's no check condition |
| 1183 | * status because the error we're processing isn't one |
| 1184 | * that has a sense code (and some devices get |
| 1185 | * confused by sense requests out of the blue) |
| 1186 | */ |
| 1187 | continue; |
| 1188 | |
| 1189 | SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd, |
| 1190 | "%s: requesting sense\n", |
| 1191 | current->comm)); |
| 1192 | rtn = scsi_request_sense(scmd); |
| 1193 | if (rtn != SUCCESS) |
| 1194 | continue; |
| 1195 | |
| 1196 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1197 | "sense requested, result %x\n", scmd->result)); |
| 1198 | SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd)); |
| 1199 | |
| 1200 | rtn = scsi_decide_disposition(scmd); |
| 1201 | |
| 1202 | /* |
| 1203 | * if the result was normal, then just pass it along to the |
| 1204 | * upper level. |
| 1205 | */ |
| 1206 | if (rtn == SUCCESS) |
| 1207 | /* we don't want this command reissued, just |
| 1208 | * finished with the sense data, so set |
| 1209 | * retries to the max allowed to ensure it |
| 1210 | * won't get reissued */ |
| 1211 | scmd->retries = scmd->allowed; |
| 1212 | else if (rtn != NEEDS_RETRY) |
| 1213 | continue; |
| 1214 | |
| 1215 | scsi_eh_finish_cmd(scmd, done_q); |
| 1216 | } |
| 1217 | |
| 1218 | return list_empty(work_q); |
| 1219 | } |
| 1220 | EXPORT_SYMBOL_GPL(scsi_eh_get_sense); |
| 1221 | |
| 1222 | /** |
| 1223 | * scsi_eh_tur - Send TUR to device. |
| 1224 | * @scmd: &scsi_cmnd to send TUR |
| 1225 | * |
| 1226 | * Return value: |
| 1227 | * 0 - Device is ready. 1 - Device NOT ready. |
| 1228 | */ |
| 1229 | static int scsi_eh_tur(struct scsi_cmnd *scmd) |
| 1230 | { |
| 1231 | static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0}; |
| 1232 | int retry_cnt = 1, rtn; |
| 1233 | |
| 1234 | retry_tur: |
| 1235 | rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, |
| 1236 | scmd->device->eh_timeout, 0); |
| 1237 | |
| 1238 | SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd, |
| 1239 | "%s return: %x\n", __func__, rtn)); |
| 1240 | |
| 1241 | switch (rtn) { |
| 1242 | case NEEDS_RETRY: |
| 1243 | if (retry_cnt--) |
| 1244 | goto retry_tur; |
| 1245 | /*FALLTHRU*/ |
| 1246 | case SUCCESS: |
| 1247 | return 0; |
| 1248 | default: |
| 1249 | return 1; |
| 1250 | } |
| 1251 | } |
| 1252 | |
| 1253 | /** |
| 1254 | * scsi_eh_test_devices - check if devices are responding from error recovery. |
| 1255 | * @cmd_list: scsi commands in error recovery. |
| 1256 | * @work_q: queue for commands which still need more error recovery |
| 1257 | * @done_q: queue for commands which are finished |
| 1258 | * @try_stu: boolean on if a STU command should be tried in addition to TUR. |
| 1259 | * |
| 1260 | * Decription: |
| 1261 | * Tests if devices are in a working state. Commands to devices now in |
| 1262 | * a working state are sent to the done_q while commands to devices which |
| 1263 | * are still failing to respond are returned to the work_q for more |
| 1264 | * processing. |
| 1265 | **/ |
| 1266 | static int scsi_eh_test_devices(struct list_head *cmd_list, |
| 1267 | struct list_head *work_q, |
| 1268 | struct list_head *done_q, int try_stu) |
| 1269 | { |
| 1270 | struct scsi_cmnd *scmd, *next; |
| 1271 | struct scsi_device *sdev; |
| 1272 | int finish_cmds; |
| 1273 | |
| 1274 | while (!list_empty(cmd_list)) { |
| 1275 | scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry); |
| 1276 | sdev = scmd->device; |
| 1277 | |
| 1278 | if (!try_stu) { |
| 1279 | if (scsi_host_eh_past_deadline(sdev->host)) { |
| 1280 | /* Push items back onto work_q */ |
| 1281 | list_splice_init(cmd_list, work_q); |
| 1282 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1283 | sdev_printk(KERN_INFO, sdev, |
| 1284 | "%s: skip test device, past eh deadline", |
| 1285 | current->comm)); |
| 1286 | break; |
| 1287 | } |
| 1288 | } |
| 1289 | |
| 1290 | finish_cmds = !scsi_device_online(scmd->device) || |
| 1291 | (try_stu && !scsi_eh_try_stu(scmd) && |
| 1292 | !scsi_eh_tur(scmd)) || |
| 1293 | !scsi_eh_tur(scmd); |
| 1294 | |
| 1295 | list_for_each_entry_safe(scmd, next, cmd_list, eh_entry) |
| 1296 | if (scmd->device == sdev) { |
| 1297 | if (finish_cmds && |
| 1298 | (try_stu || |
| 1299 | scsi_eh_action(scmd, SUCCESS) == SUCCESS)) |
| 1300 | scsi_eh_finish_cmd(scmd, done_q); |
| 1301 | else |
| 1302 | list_move_tail(&scmd->eh_entry, work_q); |
| 1303 | } |
| 1304 | } |
| 1305 | return list_empty(work_q); |
| 1306 | } |
| 1307 | |
| 1308 | |
| 1309 | /** |
| 1310 | * scsi_eh_abort_cmds - abort pending commands. |
| 1311 | * @work_q: &list_head for pending commands. |
| 1312 | * @done_q: &list_head for processed commands. |
| 1313 | * |
| 1314 | * Decription: |
| 1315 | * Try and see whether or not it makes sense to try and abort the |
| 1316 | * running command. This only works out to be the case if we have one |
| 1317 | * command that has timed out. If the command simply failed, it makes |
| 1318 | * no sense to try and abort the command, since as far as the shost |
| 1319 | * adapter is concerned, it isn't running. |
| 1320 | */ |
| 1321 | static int scsi_eh_abort_cmds(struct list_head *work_q, |
| 1322 | struct list_head *done_q) |
| 1323 | { |
| 1324 | struct scsi_cmnd *scmd, *next; |
| 1325 | LIST_HEAD(check_list); |
| 1326 | int rtn; |
| 1327 | struct Scsi_Host *shost; |
| 1328 | |
| 1329 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1330 | if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD)) |
| 1331 | continue; |
| 1332 | shost = scmd->device->host; |
| 1333 | if (scsi_host_eh_past_deadline(shost)) { |
| 1334 | list_splice_init(&check_list, work_q); |
| 1335 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1336 | scmd_printk(KERN_INFO, scmd, |
| 1337 | "%s: skip aborting cmd, past eh deadline\n", |
| 1338 | current->comm)); |
| 1339 | return list_empty(work_q); |
| 1340 | } |
| 1341 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1342 | scmd_printk(KERN_INFO, scmd, |
| 1343 | "%s: aborting cmd\n", current->comm)); |
| 1344 | rtn = scsi_try_to_abort_cmd(shost->hostt, scmd); |
| 1345 | if (rtn == FAILED) { |
| 1346 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1347 | scmd_printk(KERN_INFO, scmd, |
| 1348 | "%s: aborting cmd failed\n", |
| 1349 | current->comm)); |
| 1350 | list_splice_init(&check_list, work_q); |
| 1351 | return list_empty(work_q); |
| 1352 | } |
| 1353 | scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD; |
| 1354 | if (rtn == FAST_IO_FAIL) |
| 1355 | scsi_eh_finish_cmd(scmd, done_q); |
| 1356 | else |
| 1357 | list_move_tail(&scmd->eh_entry, &check_list); |
| 1358 | } |
| 1359 | |
| 1360 | return scsi_eh_test_devices(&check_list, work_q, done_q, 0); |
| 1361 | } |
| 1362 | |
| 1363 | /** |
| 1364 | * scsi_eh_try_stu - Send START_UNIT to device. |
| 1365 | * @scmd: &scsi_cmnd to send START_UNIT |
| 1366 | * |
| 1367 | * Return value: |
| 1368 | * 0 - Device is ready. 1 - Device NOT ready. |
| 1369 | */ |
| 1370 | static int scsi_eh_try_stu(struct scsi_cmnd *scmd) |
| 1371 | { |
| 1372 | static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0}; |
| 1373 | |
| 1374 | if (scmd->device->allow_restart) { |
| 1375 | int i, rtn = NEEDS_RETRY; |
| 1376 | |
| 1377 | for (i = 0; rtn == NEEDS_RETRY && i < 2; i++) |
| 1378 | rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0); |
| 1379 | |
| 1380 | if (rtn == SUCCESS) |
| 1381 | return 0; |
| 1382 | } |
| 1383 | |
| 1384 | return 1; |
| 1385 | } |
| 1386 | |
| 1387 | /** |
| 1388 | * scsi_eh_stu - send START_UNIT if needed |
| 1389 | * @shost: &scsi host being recovered. |
| 1390 | * @work_q: &list_head for pending commands. |
| 1391 | * @done_q: &list_head for processed commands. |
| 1392 | * |
| 1393 | * Notes: |
| 1394 | * If commands are failing due to not ready, initializing command required, |
| 1395 | * try revalidating the device, which will end up sending a start unit. |
| 1396 | */ |
| 1397 | static int scsi_eh_stu(struct Scsi_Host *shost, |
| 1398 | struct list_head *work_q, |
| 1399 | struct list_head *done_q) |
| 1400 | { |
| 1401 | struct scsi_cmnd *scmd, *stu_scmd, *next; |
| 1402 | struct scsi_device *sdev; |
| 1403 | |
| 1404 | shost_for_each_device(sdev, shost) { |
| 1405 | if (scsi_host_eh_past_deadline(shost)) { |
| 1406 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1407 | sdev_printk(KERN_INFO, sdev, |
| 1408 | "%s: skip START_UNIT, past eh deadline\n", |
| 1409 | current->comm)); |
| 1410 | break; |
| 1411 | } |
| 1412 | stu_scmd = NULL; |
| 1413 | list_for_each_entry(scmd, work_q, eh_entry) |
| 1414 | if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) && |
| 1415 | scsi_check_sense(scmd) == FAILED ) { |
| 1416 | stu_scmd = scmd; |
| 1417 | break; |
| 1418 | } |
| 1419 | |
| 1420 | if (!stu_scmd) |
| 1421 | continue; |
| 1422 | |
| 1423 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1424 | sdev_printk(KERN_INFO, sdev, |
| 1425 | "%s: Sending START_UNIT\n", |
| 1426 | current->comm)); |
| 1427 | |
| 1428 | if (!scsi_eh_try_stu(stu_scmd)) { |
| 1429 | if (!scsi_device_online(sdev) || |
| 1430 | !scsi_eh_tur(stu_scmd)) { |
| 1431 | list_for_each_entry_safe(scmd, next, |
| 1432 | work_q, eh_entry) { |
| 1433 | if (scmd->device == sdev && |
| 1434 | scsi_eh_action(scmd, SUCCESS) == SUCCESS) |
| 1435 | scsi_eh_finish_cmd(scmd, done_q); |
| 1436 | } |
| 1437 | } |
| 1438 | } else { |
| 1439 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1440 | sdev_printk(KERN_INFO, sdev, |
| 1441 | "%s: START_UNIT failed\n", |
| 1442 | current->comm)); |
| 1443 | } |
| 1444 | } |
| 1445 | |
| 1446 | return list_empty(work_q); |
| 1447 | } |
| 1448 | |
| 1449 | |
| 1450 | /** |
| 1451 | * scsi_eh_bus_device_reset - send bdr if needed |
| 1452 | * @shost: scsi host being recovered. |
| 1453 | * @work_q: &list_head for pending commands. |
| 1454 | * @done_q: &list_head for processed commands. |
| 1455 | * |
| 1456 | * Notes: |
| 1457 | * Try a bus device reset. Still, look to see whether we have multiple |
| 1458 | * devices that are jammed or not - if we have multiple devices, it |
| 1459 | * makes no sense to try bus_device_reset - we really would need to try |
| 1460 | * a bus_reset instead. |
| 1461 | */ |
| 1462 | static int scsi_eh_bus_device_reset(struct Scsi_Host *shost, |
| 1463 | struct list_head *work_q, |
| 1464 | struct list_head *done_q) |
| 1465 | { |
| 1466 | struct scsi_cmnd *scmd, *bdr_scmd, *next; |
| 1467 | struct scsi_device *sdev; |
| 1468 | int rtn; |
| 1469 | |
| 1470 | shost_for_each_device(sdev, shost) { |
| 1471 | if (scsi_host_eh_past_deadline(shost)) { |
| 1472 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1473 | sdev_printk(KERN_INFO, sdev, |
| 1474 | "%s: skip BDR, past eh deadline\n", |
| 1475 | current->comm)); |
| 1476 | break; |
| 1477 | } |
| 1478 | bdr_scmd = NULL; |
| 1479 | list_for_each_entry(scmd, work_q, eh_entry) |
| 1480 | if (scmd->device == sdev) { |
| 1481 | bdr_scmd = scmd; |
| 1482 | break; |
| 1483 | } |
| 1484 | |
| 1485 | if (!bdr_scmd) |
| 1486 | continue; |
| 1487 | |
| 1488 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1489 | sdev_printk(KERN_INFO, sdev, |
| 1490 | "%s: Sending BDR\n", current->comm)); |
| 1491 | rtn = scsi_try_bus_device_reset(bdr_scmd); |
| 1492 | if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { |
| 1493 | if (!scsi_device_online(sdev) || |
| 1494 | rtn == FAST_IO_FAIL || |
| 1495 | !scsi_eh_tur(bdr_scmd)) { |
| 1496 | list_for_each_entry_safe(scmd, next, |
| 1497 | work_q, eh_entry) { |
| 1498 | if (scmd->device == sdev && |
| 1499 | scsi_eh_action(scmd, rtn) != FAILED) |
| 1500 | scsi_eh_finish_cmd(scmd, |
| 1501 | done_q); |
| 1502 | } |
| 1503 | } |
| 1504 | } else { |
| 1505 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1506 | sdev_printk(KERN_INFO, sdev, |
| 1507 | "%s: BDR failed\n", current->comm)); |
| 1508 | } |
| 1509 | } |
| 1510 | |
| 1511 | return list_empty(work_q); |
| 1512 | } |
| 1513 | |
| 1514 | /** |
| 1515 | * scsi_eh_target_reset - send target reset if needed |
| 1516 | * @shost: scsi host being recovered. |
| 1517 | * @work_q: &list_head for pending commands. |
| 1518 | * @done_q: &list_head for processed commands. |
| 1519 | * |
| 1520 | * Notes: |
| 1521 | * Try a target reset. |
| 1522 | */ |
| 1523 | static int scsi_eh_target_reset(struct Scsi_Host *shost, |
| 1524 | struct list_head *work_q, |
| 1525 | struct list_head *done_q) |
| 1526 | { |
| 1527 | LIST_HEAD(tmp_list); |
| 1528 | LIST_HEAD(check_list); |
| 1529 | |
| 1530 | list_splice_init(work_q, &tmp_list); |
| 1531 | |
| 1532 | while (!list_empty(&tmp_list)) { |
| 1533 | struct scsi_cmnd *next, *scmd; |
| 1534 | int rtn; |
| 1535 | unsigned int id; |
| 1536 | |
| 1537 | if (scsi_host_eh_past_deadline(shost)) { |
| 1538 | /* push back on work queue for further processing */ |
| 1539 | list_splice_init(&check_list, work_q); |
| 1540 | list_splice_init(&tmp_list, work_q); |
| 1541 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1542 | shost_printk(KERN_INFO, shost, |
| 1543 | "%s: Skip target reset, past eh deadline\n", |
| 1544 | current->comm)); |
| 1545 | return list_empty(work_q); |
| 1546 | } |
| 1547 | |
| 1548 | scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry); |
| 1549 | id = scmd_id(scmd); |
| 1550 | |
| 1551 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1552 | shost_printk(KERN_INFO, shost, |
| 1553 | "%s: Sending target reset to target %d\n", |
| 1554 | current->comm, id)); |
| 1555 | rtn = scsi_try_target_reset(scmd); |
| 1556 | if (rtn != SUCCESS && rtn != FAST_IO_FAIL) |
| 1557 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1558 | shost_printk(KERN_INFO, shost, |
| 1559 | "%s: Target reset failed" |
| 1560 | " target: %d\n", |
| 1561 | current->comm, id)); |
| 1562 | list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) { |
| 1563 | if (scmd_id(scmd) != id) |
| 1564 | continue; |
| 1565 | |
| 1566 | if (rtn == SUCCESS) |
| 1567 | list_move_tail(&scmd->eh_entry, &check_list); |
| 1568 | else if (rtn == FAST_IO_FAIL) |
| 1569 | scsi_eh_finish_cmd(scmd, done_q); |
| 1570 | else |
| 1571 | /* push back on work queue for further processing */ |
| 1572 | list_move(&scmd->eh_entry, work_q); |
| 1573 | } |
| 1574 | } |
| 1575 | |
| 1576 | return scsi_eh_test_devices(&check_list, work_q, done_q, 0); |
| 1577 | } |
| 1578 | |
| 1579 | /** |
| 1580 | * scsi_eh_bus_reset - send a bus reset |
| 1581 | * @shost: &scsi host being recovered. |
| 1582 | * @work_q: &list_head for pending commands. |
| 1583 | * @done_q: &list_head for processed commands. |
| 1584 | */ |
| 1585 | static int scsi_eh_bus_reset(struct Scsi_Host *shost, |
| 1586 | struct list_head *work_q, |
| 1587 | struct list_head *done_q) |
| 1588 | { |
| 1589 | struct scsi_cmnd *scmd, *chan_scmd, *next; |
| 1590 | LIST_HEAD(check_list); |
| 1591 | unsigned int channel; |
| 1592 | int rtn; |
| 1593 | |
| 1594 | /* |
| 1595 | * we really want to loop over the various channels, and do this on |
| 1596 | * a channel by channel basis. we should also check to see if any |
| 1597 | * of the failed commands are on soft_reset devices, and if so, skip |
| 1598 | * the reset. |
| 1599 | */ |
| 1600 | |
| 1601 | for (channel = 0; channel <= shost->max_channel; channel++) { |
| 1602 | if (scsi_host_eh_past_deadline(shost)) { |
| 1603 | list_splice_init(&check_list, work_q); |
| 1604 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1605 | shost_printk(KERN_INFO, shost, |
| 1606 | "%s: skip BRST, past eh deadline\n", |
| 1607 | current->comm)); |
| 1608 | return list_empty(work_q); |
| 1609 | } |
| 1610 | |
| 1611 | chan_scmd = NULL; |
| 1612 | list_for_each_entry(scmd, work_q, eh_entry) { |
| 1613 | if (channel == scmd_channel(scmd)) { |
| 1614 | chan_scmd = scmd; |
| 1615 | break; |
| 1616 | /* |
| 1617 | * FIXME add back in some support for |
| 1618 | * soft_reset devices. |
| 1619 | */ |
| 1620 | } |
| 1621 | } |
| 1622 | |
| 1623 | if (!chan_scmd) |
| 1624 | continue; |
| 1625 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1626 | shost_printk(KERN_INFO, shost, |
| 1627 | "%s: Sending BRST chan: %d\n", |
| 1628 | current->comm, channel)); |
| 1629 | rtn = scsi_try_bus_reset(chan_scmd); |
| 1630 | if (rtn == SUCCESS || rtn == FAST_IO_FAIL) { |
| 1631 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1632 | if (channel == scmd_channel(scmd)) { |
| 1633 | if (rtn == FAST_IO_FAIL) |
| 1634 | scsi_eh_finish_cmd(scmd, |
| 1635 | done_q); |
| 1636 | else |
| 1637 | list_move_tail(&scmd->eh_entry, |
| 1638 | &check_list); |
| 1639 | } |
| 1640 | } |
| 1641 | } else { |
| 1642 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1643 | shost_printk(KERN_INFO, shost, |
| 1644 | "%s: BRST failed chan: %d\n", |
| 1645 | current->comm, channel)); |
| 1646 | } |
| 1647 | } |
| 1648 | return scsi_eh_test_devices(&check_list, work_q, done_q, 0); |
| 1649 | } |
| 1650 | |
| 1651 | /** |
| 1652 | * scsi_eh_host_reset - send a host reset |
| 1653 | * @shost: host to be reset. |
| 1654 | * @work_q: &list_head for pending commands. |
| 1655 | * @done_q: &list_head for processed commands. |
| 1656 | */ |
| 1657 | static int scsi_eh_host_reset(struct Scsi_Host *shost, |
| 1658 | struct list_head *work_q, |
| 1659 | struct list_head *done_q) |
| 1660 | { |
| 1661 | struct scsi_cmnd *scmd, *next; |
| 1662 | LIST_HEAD(check_list); |
| 1663 | int rtn; |
| 1664 | |
| 1665 | if (!list_empty(work_q)) { |
| 1666 | scmd = list_entry(work_q->next, |
| 1667 | struct scsi_cmnd, eh_entry); |
| 1668 | |
| 1669 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1670 | shost_printk(KERN_INFO, shost, |
| 1671 | "%s: Sending HRST\n", |
| 1672 | current->comm)); |
| 1673 | |
| 1674 | rtn = scsi_try_host_reset(scmd); |
| 1675 | if (rtn == SUCCESS) { |
| 1676 | list_splice_init(work_q, &check_list); |
| 1677 | } else if (rtn == FAST_IO_FAIL) { |
| 1678 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1679 | scsi_eh_finish_cmd(scmd, done_q); |
| 1680 | } |
| 1681 | } else { |
| 1682 | SCSI_LOG_ERROR_RECOVERY(3, |
| 1683 | shost_printk(KERN_INFO, shost, |
| 1684 | "%s: HRST failed\n", |
| 1685 | current->comm)); |
| 1686 | } |
| 1687 | } |
| 1688 | return scsi_eh_test_devices(&check_list, work_q, done_q, 1); |
| 1689 | } |
| 1690 | |
| 1691 | /** |
| 1692 | * scsi_eh_offline_sdevs - offline scsi devices that fail to recover |
| 1693 | * @work_q: &list_head for pending commands. |
| 1694 | * @done_q: &list_head for processed commands. |
| 1695 | */ |
| 1696 | static void scsi_eh_offline_sdevs(struct list_head *work_q, |
| 1697 | struct list_head *done_q) |
| 1698 | { |
| 1699 | struct scsi_cmnd *scmd, *next; |
| 1700 | |
| 1701 | list_for_each_entry_safe(scmd, next, work_q, eh_entry) { |
| 1702 | sdev_printk(KERN_INFO, scmd->device, "Device offlined - " |
| 1703 | "not ready after error recovery\n"); |
| 1704 | scsi_device_set_state(scmd->device, SDEV_OFFLINE); |
| 1705 | if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) { |
| 1706 | /* |
| 1707 | * FIXME: Handle lost cmds. |
| 1708 | */ |
| 1709 | } |
| 1710 | scsi_eh_finish_cmd(scmd, done_q); |
| 1711 | } |
| 1712 | return; |
| 1713 | } |
| 1714 | |
| 1715 | /** |
| 1716 | * scsi_noretry_cmd - determine if command should be failed fast |
| 1717 | * @scmd: SCSI cmd to examine. |
| 1718 | */ |
| 1719 | int scsi_noretry_cmd(struct scsi_cmnd *scmd) |
| 1720 | { |
| 1721 | switch (host_byte(scmd->result)) { |
| 1722 | case DID_OK: |
| 1723 | break; |
| 1724 | case DID_TIME_OUT: |
| 1725 | goto check_type; |
| 1726 | case DID_BUS_BUSY: |
| 1727 | return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT); |
| 1728 | case DID_PARITY: |
| 1729 | return (scmd->request->cmd_flags & REQ_FAILFAST_DEV); |
| 1730 | case DID_ERROR: |
| 1731 | if (msg_byte(scmd->result) == COMMAND_COMPLETE && |
| 1732 | status_byte(scmd->result) == RESERVATION_CONFLICT) |
| 1733 | return 0; |
| 1734 | /* fall through */ |
| 1735 | case DID_SOFT_ERROR: |
| 1736 | return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER); |
| 1737 | } |
| 1738 | |
| 1739 | if (status_byte(scmd->result) != CHECK_CONDITION) |
| 1740 | return 0; |
| 1741 | |
| 1742 | check_type: |
| 1743 | /* |
| 1744 | * assume caller has checked sense and determined |
| 1745 | * the check condition was retryable. |
| 1746 | */ |
| 1747 | if (scmd->request->cmd_flags & REQ_FAILFAST_DEV || |
| 1748 | scmd->request->cmd_type == REQ_TYPE_BLOCK_PC) |
| 1749 | return 1; |
| 1750 | else |
| 1751 | return 0; |
| 1752 | } |
| 1753 | |
| 1754 | /** |
| 1755 | * scsi_decide_disposition - Disposition a cmd on return from LLD. |
| 1756 | * @scmd: SCSI cmd to examine. |
| 1757 | * |
| 1758 | * Notes: |
| 1759 | * This is *only* called when we are examining the status after sending |
| 1760 | * out the actual data command. any commands that are queued for error |
| 1761 | * recovery (e.g. test_unit_ready) do *not* come through here. |
| 1762 | * |
| 1763 | * When this routine returns failed, it means the error handler thread |
| 1764 | * is woken. In cases where the error code indicates an error that |
| 1765 | * doesn't require the error handler read (i.e. we don't need to |
| 1766 | * abort/reset), this function should return SUCCESS. |
| 1767 | */ |
| 1768 | int scsi_decide_disposition(struct scsi_cmnd *scmd) |
| 1769 | { |
| 1770 | int rtn; |
| 1771 | |
| 1772 | /* |
| 1773 | * if the device is offline, then we clearly just pass the result back |
| 1774 | * up to the top level. |
| 1775 | */ |
| 1776 | if (!scsi_device_online(scmd->device)) { |
| 1777 | SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd, |
| 1778 | "%s: device offline - report as SUCCESS\n", __func__)); |
| 1779 | return SUCCESS; |
| 1780 | } |
| 1781 | |
| 1782 | /* |
| 1783 | * first check the host byte, to see if there is anything in there |
| 1784 | * that would indicate what we need to do. |
| 1785 | */ |
| 1786 | switch (host_byte(scmd->result)) { |
| 1787 | case DID_PASSTHROUGH: |
| 1788 | /* |
| 1789 | * no matter what, pass this through to the upper layer. |
| 1790 | * nuke this special code so that it looks like we are saying |
| 1791 | * did_ok. |
| 1792 | */ |
| 1793 | scmd->result &= 0xff00ffff; |
| 1794 | return SUCCESS; |
| 1795 | case DID_OK: |
| 1796 | /* |
| 1797 | * looks good. drop through, and check the next byte. |
| 1798 | */ |
| 1799 | break; |
| 1800 | case DID_ABORT: |
| 1801 | if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) { |
| 1802 | set_host_byte(scmd, DID_TIME_OUT); |
| 1803 | return SUCCESS; |
| 1804 | } |
| 1805 | case DID_NO_CONNECT: |
| 1806 | case DID_BAD_TARGET: |
| 1807 | /* |
| 1808 | * note - this means that we just report the status back |
| 1809 | * to the top level driver, not that we actually think |
| 1810 | * that it indicates SUCCESS. |
| 1811 | */ |
| 1812 | return SUCCESS; |
| 1813 | /* |
| 1814 | * when the low level driver returns did_soft_error, |
| 1815 | * it is responsible for keeping an internal retry counter |
| 1816 | * in order to avoid endless loops (db) |
| 1817 | * |
| 1818 | * actually this is a bug in this function here. we should |
| 1819 | * be mindful of the maximum number of retries specified |
| 1820 | * and not get stuck in a loop. |
| 1821 | */ |
| 1822 | case DID_SOFT_ERROR: |
| 1823 | goto maybe_retry; |
| 1824 | case DID_IMM_RETRY: |
| 1825 | return NEEDS_RETRY; |
| 1826 | |
| 1827 | case DID_REQUEUE: |
| 1828 | return ADD_TO_MLQUEUE; |
| 1829 | case DID_TRANSPORT_DISRUPTED: |
| 1830 | /* |
| 1831 | * LLD/transport was disrupted during processing of the IO. |
| 1832 | * The transport class is now blocked/blocking, |
| 1833 | * and the transport will decide what to do with the IO |
| 1834 | * based on its timers and recovery capablilities if |
| 1835 | * there are enough retries. |
| 1836 | */ |
| 1837 | goto maybe_retry; |
| 1838 | case DID_TRANSPORT_FAILFAST: |
| 1839 | /* |
| 1840 | * The transport decided to failfast the IO (most likely |
| 1841 | * the fast io fail tmo fired), so send IO directly upwards. |
| 1842 | */ |
| 1843 | return SUCCESS; |
| 1844 | case DID_ERROR: |
| 1845 | if (msg_byte(scmd->result) == COMMAND_COMPLETE && |
| 1846 | status_byte(scmd->result) == RESERVATION_CONFLICT) |
| 1847 | /* |
| 1848 | * execute reservation conflict processing code |
| 1849 | * lower down |
| 1850 | */ |
| 1851 | break; |
| 1852 | /* fallthrough */ |
| 1853 | case DID_BUS_BUSY: |
| 1854 | case DID_PARITY: |
| 1855 | goto maybe_retry; |
| 1856 | case DID_TIME_OUT: |
| 1857 | /* |
| 1858 | * when we scan the bus, we get timeout messages for |
| 1859 | * these commands if there is no device available. |
| 1860 | * other hosts report did_no_connect for the same thing. |
| 1861 | */ |
| 1862 | if ((scmd->cmnd[0] == TEST_UNIT_READY || |
| 1863 | scmd->cmnd[0] == INQUIRY)) { |
| 1864 | return SUCCESS; |
| 1865 | } else { |
| 1866 | return FAILED; |
| 1867 | } |
| 1868 | case DID_RESET: |
| 1869 | return SUCCESS; |
| 1870 | default: |
| 1871 | return FAILED; |
| 1872 | } |
| 1873 | |
| 1874 | /* |
| 1875 | * next, check the message byte. |
| 1876 | */ |
| 1877 | if (msg_byte(scmd->result) != COMMAND_COMPLETE) |
| 1878 | return FAILED; |
| 1879 | |
| 1880 | /* |
| 1881 | * check the status byte to see if this indicates anything special. |
| 1882 | */ |
| 1883 | switch (status_byte(scmd->result)) { |
| 1884 | case QUEUE_FULL: |
| 1885 | scsi_handle_queue_full(scmd->device); |
| 1886 | /* |
| 1887 | * the case of trying to send too many commands to a |
| 1888 | * tagged queueing device. |
| 1889 | */ |
| 1890 | case BUSY: |
| 1891 | /* |
| 1892 | * device can't talk to us at the moment. Should only |
| 1893 | * occur (SAM-3) when the task queue is empty, so will cause |
| 1894 | * the empty queue handling to trigger a stall in the |
| 1895 | * device. |
| 1896 | */ |
| 1897 | return ADD_TO_MLQUEUE; |
| 1898 | case GOOD: |
| 1899 | if (scmd->cmnd[0] == REPORT_LUNS) |
| 1900 | scmd->device->sdev_target->expecting_lun_change = 0; |
| 1901 | scsi_handle_queue_ramp_up(scmd->device); |
| 1902 | case COMMAND_TERMINATED: |
| 1903 | return SUCCESS; |
| 1904 | case TASK_ABORTED: |
| 1905 | goto maybe_retry; |
| 1906 | case CHECK_CONDITION: |
| 1907 | rtn = scsi_check_sense(scmd); |
| 1908 | if (rtn == NEEDS_RETRY) |
| 1909 | goto maybe_retry; |
| 1910 | /* if rtn == FAILED, we have no sense information; |
| 1911 | * returning FAILED will wake the error handler thread |
| 1912 | * to collect the sense and redo the decide |
| 1913 | * disposition */ |
| 1914 | return rtn; |
| 1915 | case CONDITION_GOOD: |
| 1916 | case INTERMEDIATE_GOOD: |
| 1917 | case INTERMEDIATE_C_GOOD: |
| 1918 | case ACA_ACTIVE: |
| 1919 | /* |
| 1920 | * who knows? FIXME(eric) |
| 1921 | */ |
| 1922 | return SUCCESS; |
| 1923 | |
| 1924 | case RESERVATION_CONFLICT: |
| 1925 | sdev_printk(KERN_INFO, scmd->device, |
| 1926 | "reservation conflict\n"); |
| 1927 | set_host_byte(scmd, DID_NEXUS_FAILURE); |
| 1928 | return SUCCESS; /* causes immediate i/o error */ |
| 1929 | default: |
| 1930 | return FAILED; |
| 1931 | } |
| 1932 | return FAILED; |
| 1933 | |
| 1934 | maybe_retry: |
| 1935 | |
| 1936 | /* we requeue for retry because the error was retryable, and |
| 1937 | * the request was not marked fast fail. Note that above, |
| 1938 | * even if the request is marked fast fail, we still requeue |
| 1939 | * for queue congestion conditions (QUEUE_FULL or BUSY) */ |
| 1940 | if ((++scmd->retries) <= scmd->allowed |
| 1941 | && !scsi_noretry_cmd(scmd)) { |
| 1942 | return NEEDS_RETRY; |
| 1943 | } else { |
| 1944 | /* |
| 1945 | * no more retries - report this one back to upper level. |
| 1946 | */ |
| 1947 | return SUCCESS; |
| 1948 | } |
| 1949 | } |
| 1950 | |
| 1951 | static void eh_lock_door_done(struct request *req, int uptodate) |
| 1952 | { |
| 1953 | __blk_put_request(req->q, req); |
| 1954 | } |
| 1955 | |
| 1956 | /** |
| 1957 | * scsi_eh_lock_door - Prevent medium removal for the specified device |
| 1958 | * @sdev: SCSI device to prevent medium removal |
| 1959 | * |
| 1960 | * Locking: |
| 1961 | * We must be called from process context. |
| 1962 | * |
| 1963 | * Notes: |
| 1964 | * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the |
| 1965 | * head of the devices request queue, and continue. |
| 1966 | */ |
| 1967 | static void scsi_eh_lock_door(struct scsi_device *sdev) |
| 1968 | { |
| 1969 | struct request *req; |
| 1970 | |
| 1971 | /* |
| 1972 | * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a |
| 1973 | * request becomes available |
| 1974 | */ |
| 1975 | req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL); |
| 1976 | if (IS_ERR(req)) |
| 1977 | return; |
| 1978 | |
| 1979 | blk_rq_set_block_pc(req); |
| 1980 | |
| 1981 | req->cmd[0] = ALLOW_MEDIUM_REMOVAL; |
| 1982 | req->cmd[1] = 0; |
| 1983 | req->cmd[2] = 0; |
| 1984 | req->cmd[3] = 0; |
| 1985 | req->cmd[4] = SCSI_REMOVAL_PREVENT; |
| 1986 | req->cmd[5] = 0; |
| 1987 | |
| 1988 | req->cmd_len = COMMAND_SIZE(req->cmd[0]); |
| 1989 | |
| 1990 | req->cmd_flags |= REQ_QUIET; |
| 1991 | req->timeout = 10 * HZ; |
| 1992 | req->retries = 5; |
| 1993 | |
| 1994 | blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done); |
| 1995 | } |
| 1996 | |
| 1997 | /** |
| 1998 | * scsi_restart_operations - restart io operations to the specified host. |
| 1999 | * @shost: Host we are restarting. |
| 2000 | * |
| 2001 | * Notes: |
| 2002 | * When we entered the error handler, we blocked all further i/o to |
| 2003 | * this device. we need to 'reverse' this process. |
| 2004 | */ |
| 2005 | static void scsi_restart_operations(struct Scsi_Host *shost) |
| 2006 | { |
| 2007 | struct scsi_device *sdev; |
| 2008 | unsigned long flags; |
| 2009 | |
| 2010 | /* |
| 2011 | * If the door was locked, we need to insert a door lock request |
| 2012 | * onto the head of the SCSI request queue for the device. There |
| 2013 | * is no point trying to lock the door of an off-line device. |
| 2014 | */ |
| 2015 | shost_for_each_device(sdev, shost) { |
| 2016 | if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) { |
| 2017 | scsi_eh_lock_door(sdev); |
| 2018 | sdev->was_reset = 0; |
| 2019 | } |
| 2020 | } |
| 2021 | |
| 2022 | /* |
| 2023 | * next free up anything directly waiting upon the host. this |
| 2024 | * will be requests for character device operations, and also for |
| 2025 | * ioctls to queued block devices. |
| 2026 | */ |
| 2027 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2028 | shost_printk(KERN_INFO, shost, "waking up host to restart\n")); |
| 2029 | |
| 2030 | spin_lock_irqsave(shost->host_lock, flags); |
| 2031 | if (scsi_host_set_state(shost, SHOST_RUNNING)) |
| 2032 | if (scsi_host_set_state(shost, SHOST_CANCEL)) |
| 2033 | BUG_ON(scsi_host_set_state(shost, SHOST_DEL)); |
| 2034 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 2035 | |
| 2036 | wake_up(&shost->host_wait); |
| 2037 | |
| 2038 | /* |
| 2039 | * finally we need to re-initiate requests that may be pending. we will |
| 2040 | * have had everything blocked while error handling is taking place, and |
| 2041 | * now that error recovery is done, we will need to ensure that these |
| 2042 | * requests are started. |
| 2043 | */ |
| 2044 | scsi_run_host_queues(shost); |
| 2045 | |
| 2046 | /* |
| 2047 | * if eh is active and host_eh_scheduled is pending we need to re-run |
| 2048 | * recovery. we do this check after scsi_run_host_queues() to allow |
| 2049 | * everything pent up since the last eh run a chance to make forward |
| 2050 | * progress before we sync again. Either we'll immediately re-run |
| 2051 | * recovery or scsi_device_unbusy() will wake us again when these |
| 2052 | * pending commands complete. |
| 2053 | */ |
| 2054 | spin_lock_irqsave(shost->host_lock, flags); |
| 2055 | if (shost->host_eh_scheduled) |
| 2056 | if (scsi_host_set_state(shost, SHOST_RECOVERY)) |
| 2057 | WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY)); |
| 2058 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 2059 | } |
| 2060 | |
| 2061 | /** |
| 2062 | * scsi_eh_ready_devs - check device ready state and recover if not. |
| 2063 | * @shost: host to be recovered. |
| 2064 | * @work_q: &list_head for pending commands. |
| 2065 | * @done_q: &list_head for processed commands. |
| 2066 | */ |
| 2067 | void scsi_eh_ready_devs(struct Scsi_Host *shost, |
| 2068 | struct list_head *work_q, |
| 2069 | struct list_head *done_q) |
| 2070 | { |
| 2071 | if (!scsi_eh_stu(shost, work_q, done_q)) |
| 2072 | if (!scsi_eh_bus_device_reset(shost, work_q, done_q)) |
| 2073 | if (!scsi_eh_target_reset(shost, work_q, done_q)) |
| 2074 | if (!scsi_eh_bus_reset(shost, work_q, done_q)) |
| 2075 | if (!scsi_eh_host_reset(shost, work_q, done_q)) |
| 2076 | scsi_eh_offline_sdevs(work_q, |
| 2077 | done_q); |
| 2078 | } |
| 2079 | EXPORT_SYMBOL_GPL(scsi_eh_ready_devs); |
| 2080 | |
| 2081 | /** |
| 2082 | * scsi_eh_flush_done_q - finish processed commands or retry them. |
| 2083 | * @done_q: list_head of processed commands. |
| 2084 | */ |
| 2085 | void scsi_eh_flush_done_q(struct list_head *done_q) |
| 2086 | { |
| 2087 | struct scsi_cmnd *scmd, *next; |
| 2088 | |
| 2089 | list_for_each_entry_safe(scmd, next, done_q, eh_entry) { |
| 2090 | list_del_init(&scmd->eh_entry); |
| 2091 | if (scsi_device_online(scmd->device) && |
| 2092 | !scsi_noretry_cmd(scmd) && |
| 2093 | (++scmd->retries <= scmd->allowed)) { |
| 2094 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2095 | scmd_printk(KERN_INFO, scmd, |
| 2096 | "%s: flush retry cmd\n", |
| 2097 | current->comm)); |
| 2098 | scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY); |
| 2099 | } else { |
| 2100 | /* |
| 2101 | * If just we got sense for the device (called |
| 2102 | * scsi_eh_get_sense), scmd->result is already |
| 2103 | * set, do not set DRIVER_TIMEOUT. |
| 2104 | */ |
| 2105 | if (!scmd->result) |
| 2106 | scmd->result |= (DRIVER_TIMEOUT << 24); |
| 2107 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2108 | scmd_printk(KERN_INFO, scmd, |
| 2109 | "%s: flush finish cmd\n", |
| 2110 | current->comm)); |
| 2111 | scsi_finish_command(scmd); |
| 2112 | } |
| 2113 | } |
| 2114 | } |
| 2115 | EXPORT_SYMBOL(scsi_eh_flush_done_q); |
| 2116 | |
| 2117 | /** |
| 2118 | * scsi_unjam_host - Attempt to fix a host which has a cmd that failed. |
| 2119 | * @shost: Host to unjam. |
| 2120 | * |
| 2121 | * Notes: |
| 2122 | * When we come in here, we *know* that all commands on the bus have |
| 2123 | * either completed, failed or timed out. we also know that no further |
| 2124 | * commands are being sent to the host, so things are relatively quiet |
| 2125 | * and we have freedom to fiddle with things as we wish. |
| 2126 | * |
| 2127 | * This is only the *default* implementation. it is possible for |
| 2128 | * individual drivers to supply their own version of this function, and |
| 2129 | * if the maintainer wishes to do this, it is strongly suggested that |
| 2130 | * this function be taken as a template and modified. this function |
| 2131 | * was designed to correctly handle problems for about 95% of the |
| 2132 | * different cases out there, and it should always provide at least a |
| 2133 | * reasonable amount of error recovery. |
| 2134 | * |
| 2135 | * Any command marked 'failed' or 'timeout' must eventually have |
| 2136 | * scsi_finish_cmd() called for it. we do all of the retry stuff |
| 2137 | * here, so when we restart the host after we return it should have an |
| 2138 | * empty queue. |
| 2139 | */ |
| 2140 | static void scsi_unjam_host(struct Scsi_Host *shost) |
| 2141 | { |
| 2142 | unsigned long flags; |
| 2143 | LIST_HEAD(eh_work_q); |
| 2144 | LIST_HEAD(eh_done_q); |
| 2145 | |
| 2146 | spin_lock_irqsave(shost->host_lock, flags); |
| 2147 | list_splice_init(&shost->eh_cmd_q, &eh_work_q); |
| 2148 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 2149 | |
| 2150 | SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q)); |
| 2151 | |
| 2152 | if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q)) |
| 2153 | if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q)) |
| 2154 | scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q); |
| 2155 | |
| 2156 | spin_lock_irqsave(shost->host_lock, flags); |
| 2157 | if (shost->eh_deadline != -1) |
| 2158 | shost->last_reset = 0; |
| 2159 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 2160 | scsi_eh_flush_done_q(&eh_done_q); |
| 2161 | } |
| 2162 | |
| 2163 | /** |
| 2164 | * scsi_error_handler - SCSI error handler thread |
| 2165 | * @data: Host for which we are running. |
| 2166 | * |
| 2167 | * Notes: |
| 2168 | * This is the main error handling loop. This is run as a kernel thread |
| 2169 | * for every SCSI host and handles all error handling activity. |
| 2170 | */ |
| 2171 | int scsi_error_handler(void *data) |
| 2172 | { |
| 2173 | struct Scsi_Host *shost = data; |
| 2174 | |
| 2175 | /* |
| 2176 | * We use TASK_INTERRUPTIBLE so that the thread is not |
| 2177 | * counted against the load average as a running process. |
| 2178 | * We never actually get interrupted because kthread_run |
| 2179 | * disables signal delivery for the created thread. |
| 2180 | */ |
| 2181 | while (true) { |
| 2182 | /* |
| 2183 | * The sequence in kthread_stop() sets the stop flag first |
| 2184 | * then wakes the process. To avoid missed wakeups, the task |
| 2185 | * should always be in a non running state before the stop |
| 2186 | * flag is checked |
| 2187 | */ |
| 2188 | set_current_state(TASK_INTERRUPTIBLE); |
| 2189 | if (kthread_should_stop()) |
| 2190 | break; |
| 2191 | |
| 2192 | if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) || |
| 2193 | shost->host_failed != atomic_read(&shost->host_busy)) { |
| 2194 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2195 | shost_printk(KERN_INFO, shost, |
| 2196 | "scsi_eh_%d: sleeping\n", |
| 2197 | shost->host_no)); |
| 2198 | schedule(); |
| 2199 | continue; |
| 2200 | } |
| 2201 | |
| 2202 | __set_current_state(TASK_RUNNING); |
| 2203 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2204 | shost_printk(KERN_INFO, shost, |
| 2205 | "scsi_eh_%d: waking up %d/%d/%d\n", |
| 2206 | shost->host_no, shost->host_eh_scheduled, |
| 2207 | shost->host_failed, |
| 2208 | atomic_read(&shost->host_busy))); |
| 2209 | |
| 2210 | /* |
| 2211 | * We have a host that is failing for some reason. Figure out |
| 2212 | * what we need to do to get it up and online again (if we can). |
| 2213 | * If we fail, we end up taking the thing offline. |
| 2214 | */ |
| 2215 | if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) { |
| 2216 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2217 | shost_printk(KERN_ERR, shost, |
| 2218 | "scsi_eh_%d: unable to autoresume\n", |
| 2219 | shost->host_no)); |
| 2220 | continue; |
| 2221 | } |
| 2222 | |
| 2223 | if (shost->transportt->eh_strategy_handler) |
| 2224 | shost->transportt->eh_strategy_handler(shost); |
| 2225 | else |
| 2226 | scsi_unjam_host(shost); |
| 2227 | |
| 2228 | /* All scmds have been handled */ |
| 2229 | shost->host_failed = 0; |
| 2230 | |
| 2231 | /* |
| 2232 | * Note - if the above fails completely, the action is to take |
| 2233 | * individual devices offline and flush the queue of any |
| 2234 | * outstanding requests that may have been pending. When we |
| 2235 | * restart, we restart any I/O to any other devices on the bus |
| 2236 | * which are still online. |
| 2237 | */ |
| 2238 | scsi_restart_operations(shost); |
| 2239 | if (!shost->eh_noresume) |
| 2240 | scsi_autopm_put_host(shost); |
| 2241 | } |
| 2242 | __set_current_state(TASK_RUNNING); |
| 2243 | |
| 2244 | SCSI_LOG_ERROR_RECOVERY(1, |
| 2245 | shost_printk(KERN_INFO, shost, |
| 2246 | "Error handler scsi_eh_%d exiting\n", |
| 2247 | shost->host_no)); |
| 2248 | shost->ehandler = NULL; |
| 2249 | return 0; |
| 2250 | } |
| 2251 | |
| 2252 | /* |
| 2253 | * Function: scsi_report_bus_reset() |
| 2254 | * |
| 2255 | * Purpose: Utility function used by low-level drivers to report that |
| 2256 | * they have observed a bus reset on the bus being handled. |
| 2257 | * |
| 2258 | * Arguments: shost - Host in question |
| 2259 | * channel - channel on which reset was observed. |
| 2260 | * |
| 2261 | * Returns: Nothing |
| 2262 | * |
| 2263 | * Lock status: Host lock must be held. |
| 2264 | * |
| 2265 | * Notes: This only needs to be called if the reset is one which |
| 2266 | * originates from an unknown location. Resets originated |
| 2267 | * by the mid-level itself don't need to call this, but there |
| 2268 | * should be no harm. |
| 2269 | * |
| 2270 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 2271 | * is properly treated. |
| 2272 | */ |
| 2273 | void scsi_report_bus_reset(struct Scsi_Host *shost, int channel) |
| 2274 | { |
| 2275 | struct scsi_device *sdev; |
| 2276 | |
| 2277 | __shost_for_each_device(sdev, shost) { |
| 2278 | if (channel == sdev_channel(sdev)) |
| 2279 | __scsi_report_device_reset(sdev, NULL); |
| 2280 | } |
| 2281 | } |
| 2282 | EXPORT_SYMBOL(scsi_report_bus_reset); |
| 2283 | |
| 2284 | /* |
| 2285 | * Function: scsi_report_device_reset() |
| 2286 | * |
| 2287 | * Purpose: Utility function used by low-level drivers to report that |
| 2288 | * they have observed a device reset on the device being handled. |
| 2289 | * |
| 2290 | * Arguments: shost - Host in question |
| 2291 | * channel - channel on which reset was observed |
| 2292 | * target - target on which reset was observed |
| 2293 | * |
| 2294 | * Returns: Nothing |
| 2295 | * |
| 2296 | * Lock status: Host lock must be held |
| 2297 | * |
| 2298 | * Notes: This only needs to be called if the reset is one which |
| 2299 | * originates from an unknown location. Resets originated |
| 2300 | * by the mid-level itself don't need to call this, but there |
| 2301 | * should be no harm. |
| 2302 | * |
| 2303 | * The main purpose of this is to make sure that a CHECK_CONDITION |
| 2304 | * is properly treated. |
| 2305 | */ |
| 2306 | void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target) |
| 2307 | { |
| 2308 | struct scsi_device *sdev; |
| 2309 | |
| 2310 | __shost_for_each_device(sdev, shost) { |
| 2311 | if (channel == sdev_channel(sdev) && |
| 2312 | target == sdev_id(sdev)) |
| 2313 | __scsi_report_device_reset(sdev, NULL); |
| 2314 | } |
| 2315 | } |
| 2316 | EXPORT_SYMBOL(scsi_report_device_reset); |
| 2317 | |
| 2318 | static void |
| 2319 | scsi_reset_provider_done_command(struct scsi_cmnd *scmd) |
| 2320 | { |
| 2321 | } |
| 2322 | |
| 2323 | /** |
| 2324 | * scsi_ioctl_reset: explicitly reset a host/bus/target/device |
| 2325 | * @dev: scsi_device to operate on |
| 2326 | * @arg: reset type (see sg.h) |
| 2327 | */ |
| 2328 | int |
| 2329 | scsi_ioctl_reset(struct scsi_device *dev, int __user *arg) |
| 2330 | { |
| 2331 | struct scsi_cmnd *scmd; |
| 2332 | struct Scsi_Host *shost = dev->host; |
| 2333 | struct request req; |
| 2334 | unsigned long flags; |
| 2335 | int error = 0, rtn, val; |
| 2336 | |
| 2337 | if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO)) |
| 2338 | return -EACCES; |
| 2339 | |
| 2340 | error = get_user(val, arg); |
| 2341 | if (error) |
| 2342 | return error; |
| 2343 | |
| 2344 | if (scsi_autopm_get_host(shost) < 0) |
| 2345 | return -EIO; |
| 2346 | |
| 2347 | error = -EIO; |
| 2348 | scmd = scsi_get_command(dev, GFP_KERNEL); |
| 2349 | if (!scmd) |
| 2350 | goto out_put_autopm_host; |
| 2351 | |
| 2352 | blk_rq_init(NULL, &req); |
| 2353 | scmd->request = &req; |
| 2354 | |
| 2355 | scmd->cmnd = req.cmd; |
| 2356 | |
| 2357 | scmd->scsi_done = scsi_reset_provider_done_command; |
| 2358 | memset(&scmd->sdb, 0, sizeof(scmd->sdb)); |
| 2359 | |
| 2360 | scmd->cmd_len = 0; |
| 2361 | |
| 2362 | scmd->sc_data_direction = DMA_BIDIRECTIONAL; |
| 2363 | |
| 2364 | spin_lock_irqsave(shost->host_lock, flags); |
| 2365 | shost->tmf_in_progress = 1; |
| 2366 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 2367 | |
| 2368 | switch (val & ~SG_SCSI_RESET_NO_ESCALATE) { |
| 2369 | case SG_SCSI_RESET_NOTHING: |
| 2370 | rtn = SUCCESS; |
| 2371 | break; |
| 2372 | case SG_SCSI_RESET_DEVICE: |
| 2373 | rtn = scsi_try_bus_device_reset(scmd); |
| 2374 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
| 2375 | break; |
| 2376 | /* FALLTHROUGH */ |
| 2377 | case SG_SCSI_RESET_TARGET: |
| 2378 | rtn = scsi_try_target_reset(scmd); |
| 2379 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
| 2380 | break; |
| 2381 | /* FALLTHROUGH */ |
| 2382 | case SG_SCSI_RESET_BUS: |
| 2383 | rtn = scsi_try_bus_reset(scmd); |
| 2384 | if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE)) |
| 2385 | break; |
| 2386 | /* FALLTHROUGH */ |
| 2387 | case SG_SCSI_RESET_HOST: |
| 2388 | rtn = scsi_try_host_reset(scmd); |
| 2389 | if (rtn == SUCCESS) |
| 2390 | break; |
| 2391 | default: |
| 2392 | /* FALLTHROUGH */ |
| 2393 | rtn = FAILED; |
| 2394 | break; |
| 2395 | } |
| 2396 | |
| 2397 | error = (rtn == SUCCESS) ? 0 : -EIO; |
| 2398 | |
| 2399 | spin_lock_irqsave(shost->host_lock, flags); |
| 2400 | shost->tmf_in_progress = 0; |
| 2401 | spin_unlock_irqrestore(shost->host_lock, flags); |
| 2402 | |
| 2403 | /* |
| 2404 | * be sure to wake up anyone who was sleeping or had their queue |
| 2405 | * suspended while we performed the TMF. |
| 2406 | */ |
| 2407 | SCSI_LOG_ERROR_RECOVERY(3, |
| 2408 | shost_printk(KERN_INFO, shost, |
| 2409 | "waking up host to restart after TMF\n")); |
| 2410 | |
| 2411 | wake_up(&shost->host_wait); |
| 2412 | scsi_run_host_queues(shost); |
| 2413 | |
| 2414 | scsi_put_command(scmd); |
| 2415 | |
| 2416 | out_put_autopm_host: |
| 2417 | scsi_autopm_put_host(shost); |
| 2418 | return error; |
| 2419 | } |
| 2420 | EXPORT_SYMBOL(scsi_ioctl_reset); |
| 2421 | |
| 2422 | bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd, |
| 2423 | struct scsi_sense_hdr *sshdr) |
| 2424 | { |
| 2425 | return scsi_normalize_sense(cmd->sense_buffer, |
| 2426 | SCSI_SENSE_BUFFERSIZE, sshdr); |
| 2427 | } |
| 2428 | EXPORT_SYMBOL(scsi_command_normalize_sense); |
| 2429 | |
| 2430 | /** |
| 2431 | * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format) |
| 2432 | * @sense_buffer: byte array of sense data |
| 2433 | * @sb_len: number of valid bytes in sense_buffer |
| 2434 | * @info_out: pointer to 64 integer where 8 or 4 byte information |
| 2435 | * field will be placed if found. |
| 2436 | * |
| 2437 | * Return value: |
| 2438 | * 1 if information field found, 0 if not found. |
| 2439 | */ |
| 2440 | int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len, |
| 2441 | u64 * info_out) |
| 2442 | { |
| 2443 | int j; |
| 2444 | const u8 * ucp; |
| 2445 | u64 ull; |
| 2446 | |
| 2447 | if (sb_len < 7) |
| 2448 | return 0; |
| 2449 | switch (sense_buffer[0] & 0x7f) { |
| 2450 | case 0x70: |
| 2451 | case 0x71: |
| 2452 | if (sense_buffer[0] & 0x80) { |
| 2453 | *info_out = (sense_buffer[3] << 24) + |
| 2454 | (sense_buffer[4] << 16) + |
| 2455 | (sense_buffer[5] << 8) + sense_buffer[6]; |
| 2456 | return 1; |
| 2457 | } else |
| 2458 | return 0; |
| 2459 | case 0x72: |
| 2460 | case 0x73: |
| 2461 | ucp = scsi_sense_desc_find(sense_buffer, sb_len, |
| 2462 | 0 /* info desc */); |
| 2463 | if (ucp && (0xa == ucp[1])) { |
| 2464 | ull = 0; |
| 2465 | for (j = 0; j < 8; ++j) { |
| 2466 | if (j > 0) |
| 2467 | ull <<= 8; |
| 2468 | ull |= ucp[4 + j]; |
| 2469 | } |
| 2470 | *info_out = ull; |
| 2471 | return 1; |
| 2472 | } else |
| 2473 | return 0; |
| 2474 | default: |
| 2475 | return 0; |
| 2476 | } |
| 2477 | } |
| 2478 | EXPORT_SYMBOL(scsi_get_sense_info_fld); |