1 /* 2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale 3 * 4 * SCSI queueing library. 5 * Initial versions: Eric Youngdale (eric@andante.org). 6 * Based upon conversations with large numbers 7 * of people at Linux Expo. 8 */ 9 10 #include <linux/bio.h> 11 #include <linux/bitops.h> 12 #include <linux/blkdev.h> 13 #include <linux/completion.h> 14 #include <linux/kernel.h> 15 #include <linux/mempool.h> 16 #include <linux/slab.h> 17 #include <linux/init.h> 18 #include <linux/pci.h> 19 #include <linux/delay.h> 20 #include <linux/hardirq.h> 21 #include <linux/scatterlist.h> 22 23 #include <scsi/scsi.h> 24 #include <scsi/scsi_cmnd.h> 25 #include <scsi/scsi_dbg.h> 26 #include <scsi/scsi_device.h> 27 #include <scsi/scsi_driver.h> 28 #include <scsi/scsi_eh.h> 29 #include <scsi/scsi_host.h> 30 31 #include "scsi_priv.h" 32 #include "scsi_logging.h" 33 34 35 #define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools) 36 #define SG_MEMPOOL_SIZE 2 37 38 struct scsi_host_sg_pool { 39 size_t size; 40 char *name; 41 struct kmem_cache *slab; 42 mempool_t *pool; 43 }; 44 45 #define SP(x) { x, "sgpool-" __stringify(x) } 46 #if (SCSI_MAX_SG_SEGMENTS < 32) 47 #error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater) 48 #endif 49 static struct scsi_host_sg_pool scsi_sg_pools[] = { 50 SP(8), 51 SP(16), 52 #if (SCSI_MAX_SG_SEGMENTS > 32) 53 SP(32), 54 #if (SCSI_MAX_SG_SEGMENTS > 64) 55 SP(64), 56 #if (SCSI_MAX_SG_SEGMENTS > 128) 57 SP(128), 58 #if (SCSI_MAX_SG_SEGMENTS > 256) 59 #error SCSI_MAX_SG_SEGMENTS is too large (256 MAX) 60 #endif 61 #endif 62 #endif 63 #endif 64 SP(SCSI_MAX_SG_SEGMENTS) 65 }; 66 #undef SP 67 68 struct kmem_cache *scsi_sdb_cache; 69 70 static void scsi_run_queue(struct request_queue *q); 71 72 /* 73 * Function: scsi_unprep_request() 74 * 75 * Purpose: Remove all preparation done for a request, including its 76 * associated scsi_cmnd, so that it can be requeued. 77 * 78 * Arguments: req - request to unprepare 79 * 80 * Lock status: Assumed that no locks are held upon entry. 81 * 82 * Returns: Nothing. 83 */ 84 static void scsi_unprep_request(struct request *req) 85 { 86 struct scsi_cmnd *cmd = req->special; 87 88 req->cmd_flags &= ~REQ_DONTPREP; 89 req->special = NULL; 90 91 scsi_put_command(cmd); 92 } 93 94 /** 95 * __scsi_queue_insert - private queue insertion 96 * @cmd: The SCSI command being requeued 97 * @reason: The reason for the requeue 98 * @unbusy: Whether the queue should be unbusied 99 * 100 * This is a private queue insertion. The public interface 101 * scsi_queue_insert() always assumes the queue should be unbusied 102 * because it's always called before the completion. This function is 103 * for a requeue after completion, which should only occur in this 104 * file. 105 */ 106 static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy) 107 { 108 struct Scsi_Host *host = cmd->device->host; 109 struct scsi_device *device = cmd->device; 110 struct scsi_target *starget = scsi_target(device); 111 struct request_queue *q = device->request_queue; 112 unsigned long flags; 113 114 SCSI_LOG_MLQUEUE(1, 115 printk("Inserting command %p into mlqueue\n", cmd)); 116 117 /* 118 * Set the appropriate busy bit for the device/host. 119 * 120 * If the host/device isn't busy, assume that something actually 121 * completed, and that we should be able to queue a command now. 122 * 123 * Note that the prior mid-layer assumption that any host could 124 * always queue at least one command is now broken. The mid-layer 125 * will implement a user specifiable stall (see 126 * scsi_host.max_host_blocked and scsi_device.max_device_blocked) 127 * if a command is requeued with no other commands outstanding 128 * either for the device or for the host. 129 */ 130 switch (reason) { 131 case SCSI_MLQUEUE_HOST_BUSY: 132 host->host_blocked = host->max_host_blocked; 133 break; 134 case SCSI_MLQUEUE_DEVICE_BUSY: 135 device->device_blocked = device->max_device_blocked; 136 break; 137 case SCSI_MLQUEUE_TARGET_BUSY: 138 starget->target_blocked = starget->max_target_blocked; 139 break; 140 } 141 142 /* 143 * Decrement the counters, since these commands are no longer 144 * active on the host/device. 145 */ 146 if (unbusy) 147 scsi_device_unbusy(device); 148 149 /* 150 * Requeue this command. It will go before all other commands 151 * that are already in the queue. 152 * 153 * NOTE: there is magic here about the way the queue is plugged if 154 * we have no outstanding commands. 155 * 156 * Although we *don't* plug the queue, we call the request 157 * function. The SCSI request function detects the blocked condition 158 * and plugs the queue appropriately. 159 */ 160 spin_lock_irqsave(q->queue_lock, flags); 161 blk_requeue_request(q, cmd->request); 162 spin_unlock_irqrestore(q->queue_lock, flags); 163 164 scsi_run_queue(q); 165 166 return 0; 167 } 168 169 /* 170 * Function: scsi_queue_insert() 171 * 172 * Purpose: Insert a command in the midlevel queue. 173 * 174 * Arguments: cmd - command that we are adding to queue. 175 * reason - why we are inserting command to queue. 176 * 177 * Lock status: Assumed that lock is not held upon entry. 178 * 179 * Returns: Nothing. 180 * 181 * Notes: We do this for one of two cases. Either the host is busy 182 * and it cannot accept any more commands for the time being, 183 * or the device returned QUEUE_FULL and can accept no more 184 * commands. 185 * Notes: This could be called either from an interrupt context or a 186 * normal process context. 187 */ 188 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason) 189 { 190 return __scsi_queue_insert(cmd, reason, 1); 191 } 192 /** 193 * scsi_execute - insert request and wait for the result 194 * @sdev: scsi device 195 * @cmd: scsi command 196 * @data_direction: data direction 197 * @buffer: data buffer 198 * @bufflen: len of buffer 199 * @sense: optional sense buffer 200 * @timeout: request timeout in seconds 201 * @retries: number of times to retry request 202 * @flags: or into request flags; 203 * @resid: optional residual length 204 * 205 * returns the req->errors value which is the scsi_cmnd result 206 * field. 207 */ 208 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd, 209 int data_direction, void *buffer, unsigned bufflen, 210 unsigned char *sense, int timeout, int retries, int flags, 211 int *resid) 212 { 213 struct request *req; 214 int write = (data_direction == DMA_TO_DEVICE); 215 int ret = DRIVER_ERROR << 24; 216 217 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT); 218 219 if (bufflen && blk_rq_map_kern(sdev->request_queue, req, 220 buffer, bufflen, __GFP_WAIT)) 221 goto out; 222 223 req->cmd_len = COMMAND_SIZE(cmd[0]); 224 memcpy(req->cmd, cmd, req->cmd_len); 225 req->sense = sense; 226 req->sense_len = 0; 227 req->retries = retries; 228 req->timeout = timeout; 229 req->cmd_type = REQ_TYPE_BLOCK_PC; 230 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT; 231 232 /* 233 * head injection *required* here otherwise quiesce won't work 234 */ 235 blk_execute_rq(req->q, NULL, req, 1); 236 237 /* 238 * Some devices (USB mass-storage in particular) may transfer 239 * garbage data together with a residue indicating that the data 240 * is invalid. Prevent the garbage from being misinterpreted 241 * and prevent security leaks by zeroing out the excess data. 242 */ 243 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen)) 244 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len); 245 246 if (resid) 247 *resid = req->resid_len; 248 ret = req->errors; 249 out: 250 blk_put_request(req); 251 252 return ret; 253 } 254 EXPORT_SYMBOL(scsi_execute); 255 256 257 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd, 258 int data_direction, void *buffer, unsigned bufflen, 259 struct scsi_sense_hdr *sshdr, int timeout, int retries, 260 int *resid) 261 { 262 char *sense = NULL; 263 int result; 264 265 if (sshdr) { 266 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO); 267 if (!sense) 268 return DRIVER_ERROR << 24; 269 } 270 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen, 271 sense, timeout, retries, 0, resid); 272 if (sshdr) 273 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr); 274 275 kfree(sense); 276 return result; 277 } 278 EXPORT_SYMBOL(scsi_execute_req); 279 280 /* 281 * Function: scsi_init_cmd_errh() 282 * 283 * Purpose: Initialize cmd fields related to error handling. 284 * 285 * Arguments: cmd - command that is ready to be queued. 286 * 287 * Notes: This function has the job of initializing a number of 288 * fields related to error handling. Typically this will 289 * be called once for each command, as required. 290 */ 291 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd) 292 { 293 cmd->serial_number = 0; 294 scsi_set_resid(cmd, 0); 295 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 296 if (cmd->cmd_len == 0) 297 cmd->cmd_len = scsi_command_size(cmd->cmnd); 298 } 299 300 void scsi_device_unbusy(struct scsi_device *sdev) 301 { 302 struct Scsi_Host *shost = sdev->host; 303 struct scsi_target *starget = scsi_target(sdev); 304 unsigned long flags; 305 306 spin_lock_irqsave(shost->host_lock, flags); 307 shost->host_busy--; 308 starget->target_busy--; 309 if (unlikely(scsi_host_in_recovery(shost) && 310 (shost->host_failed || shost->host_eh_scheduled))) 311 scsi_eh_wakeup(shost); 312 spin_unlock(shost->host_lock); 313 spin_lock(sdev->request_queue->queue_lock); 314 sdev->device_busy--; 315 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags); 316 } 317 318 /* 319 * Called for single_lun devices on IO completion. Clear starget_sdev_user, 320 * and call blk_run_queue for all the scsi_devices on the target - 321 * including current_sdev first. 322 * 323 * Called with *no* scsi locks held. 324 */ 325 static void scsi_single_lun_run(struct scsi_device *current_sdev) 326 { 327 struct Scsi_Host *shost = current_sdev->host; 328 struct scsi_device *sdev, *tmp; 329 struct scsi_target *starget = scsi_target(current_sdev); 330 unsigned long flags; 331 332 spin_lock_irqsave(shost->host_lock, flags); 333 starget->starget_sdev_user = NULL; 334 spin_unlock_irqrestore(shost->host_lock, flags); 335 336 /* 337 * Call blk_run_queue for all LUNs on the target, starting with 338 * current_sdev. We race with others (to set starget_sdev_user), 339 * but in most cases, we will be first. Ideally, each LU on the 340 * target would get some limited time or requests on the target. 341 */ 342 blk_run_queue(current_sdev->request_queue); 343 344 spin_lock_irqsave(shost->host_lock, flags); 345 if (starget->starget_sdev_user) 346 goto out; 347 list_for_each_entry_safe(sdev, tmp, &starget->devices, 348 same_target_siblings) { 349 if (sdev == current_sdev) 350 continue; 351 if (scsi_device_get(sdev)) 352 continue; 353 354 spin_unlock_irqrestore(shost->host_lock, flags); 355 blk_run_queue(sdev->request_queue); 356 spin_lock_irqsave(shost->host_lock, flags); 357 358 scsi_device_put(sdev); 359 } 360 out: 361 spin_unlock_irqrestore(shost->host_lock, flags); 362 } 363 364 static inline int scsi_device_is_busy(struct scsi_device *sdev) 365 { 366 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked) 367 return 1; 368 369 return 0; 370 } 371 372 static inline int scsi_target_is_busy(struct scsi_target *starget) 373 { 374 return ((starget->can_queue > 0 && 375 starget->target_busy >= starget->can_queue) || 376 starget->target_blocked); 377 } 378 379 static inline int scsi_host_is_busy(struct Scsi_Host *shost) 380 { 381 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) || 382 shost->host_blocked || shost->host_self_blocked) 383 return 1; 384 385 return 0; 386 } 387 388 /* 389 * Function: scsi_run_queue() 390 * 391 * Purpose: Select a proper request queue to serve next 392 * 393 * Arguments: q - last request's queue 394 * 395 * Returns: Nothing 396 * 397 * Notes: The previous command was completely finished, start 398 * a new one if possible. 399 */ 400 static void scsi_run_queue(struct request_queue *q) 401 { 402 struct scsi_device *sdev = q->queuedata; 403 struct Scsi_Host *shost = sdev->host; 404 LIST_HEAD(starved_list); 405 unsigned long flags; 406 407 if (scsi_target(sdev)->single_lun) 408 scsi_single_lun_run(sdev); 409 410 spin_lock_irqsave(shost->host_lock, flags); 411 list_splice_init(&shost->starved_list, &starved_list); 412 413 while (!list_empty(&starved_list)) { 414 int flagset; 415 416 /* 417 * As long as shost is accepting commands and we have 418 * starved queues, call blk_run_queue. scsi_request_fn 419 * drops the queue_lock and can add us back to the 420 * starved_list. 421 * 422 * host_lock protects the starved_list and starved_entry. 423 * scsi_request_fn must get the host_lock before checking 424 * or modifying starved_list or starved_entry. 425 */ 426 if (scsi_host_is_busy(shost)) 427 break; 428 429 sdev = list_entry(starved_list.next, 430 struct scsi_device, starved_entry); 431 list_del_init(&sdev->starved_entry); 432 if (scsi_target_is_busy(scsi_target(sdev))) { 433 list_move_tail(&sdev->starved_entry, 434 &shost->starved_list); 435 continue; 436 } 437 438 spin_unlock(shost->host_lock); 439 440 spin_lock(sdev->request_queue->queue_lock); 441 flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) && 442 !test_bit(QUEUE_FLAG_REENTER, 443 &sdev->request_queue->queue_flags); 444 if (flagset) 445 queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue); 446 __blk_run_queue(sdev->request_queue); 447 if (flagset) 448 queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue); 449 spin_unlock(sdev->request_queue->queue_lock); 450 451 spin_lock(shost->host_lock); 452 } 453 /* put any unprocessed entries back */ 454 list_splice(&starved_list, &shost->starved_list); 455 spin_unlock_irqrestore(shost->host_lock, flags); 456 457 blk_run_queue(q); 458 } 459 460 /* 461 * Function: scsi_requeue_command() 462 * 463 * Purpose: Handle post-processing of completed commands. 464 * 465 * Arguments: q - queue to operate on 466 * cmd - command that may need to be requeued. 467 * 468 * Returns: Nothing 469 * 470 * Notes: After command completion, there may be blocks left 471 * over which weren't finished by the previous command 472 * this can be for a number of reasons - the main one is 473 * I/O errors in the middle of the request, in which case 474 * we need to request the blocks that come after the bad 475 * sector. 476 * Notes: Upon return, cmd is a stale pointer. 477 */ 478 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd) 479 { 480 struct request *req = cmd->request; 481 unsigned long flags; 482 483 spin_lock_irqsave(q->queue_lock, flags); 484 scsi_unprep_request(req); 485 blk_requeue_request(q, req); 486 spin_unlock_irqrestore(q->queue_lock, flags); 487 488 scsi_run_queue(q); 489 } 490 491 void scsi_next_command(struct scsi_cmnd *cmd) 492 { 493 struct scsi_device *sdev = cmd->device; 494 struct request_queue *q = sdev->request_queue; 495 496 /* need to hold a reference on the device before we let go of the cmd */ 497 get_device(&sdev->sdev_gendev); 498 499 scsi_put_command(cmd); 500 scsi_run_queue(q); 501 502 /* ok to remove device now */ 503 put_device(&sdev->sdev_gendev); 504 } 505 506 void scsi_run_host_queues(struct Scsi_Host *shost) 507 { 508 struct scsi_device *sdev; 509 510 shost_for_each_device(sdev, shost) 511 scsi_run_queue(sdev->request_queue); 512 } 513 514 static void __scsi_release_buffers(struct scsi_cmnd *, int); 515 516 /* 517 * Function: scsi_end_request() 518 * 519 * Purpose: Post-processing of completed commands (usually invoked at end 520 * of upper level post-processing and scsi_io_completion). 521 * 522 * Arguments: cmd - command that is complete. 523 * error - 0 if I/O indicates success, < 0 for I/O error. 524 * bytes - number of bytes of completed I/O 525 * requeue - indicates whether we should requeue leftovers. 526 * 527 * Lock status: Assumed that lock is not held upon entry. 528 * 529 * Returns: cmd if requeue required, NULL otherwise. 530 * 531 * Notes: This is called for block device requests in order to 532 * mark some number of sectors as complete. 533 * 534 * We are guaranteeing that the request queue will be goosed 535 * at some point during this call. 536 * Notes: If cmd was requeued, upon return it will be a stale pointer. 537 */ 538 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error, 539 int bytes, int requeue) 540 { 541 struct request_queue *q = cmd->device->request_queue; 542 struct request *req = cmd->request; 543 544 /* 545 * If there are blocks left over at the end, set up the command 546 * to queue the remainder of them. 547 */ 548 if (blk_end_request(req, error, bytes)) { 549 /* kill remainder if no retrys */ 550 if (error && scsi_noretry_cmd(cmd)) 551 blk_end_request_all(req, error); 552 else { 553 if (requeue) { 554 /* 555 * Bleah. Leftovers again. Stick the 556 * leftovers in the front of the 557 * queue, and goose the queue again. 558 */ 559 scsi_release_buffers(cmd); 560 scsi_requeue_command(q, cmd); 561 cmd = NULL; 562 } 563 return cmd; 564 } 565 } 566 567 /* 568 * This will goose the queue request function at the end, so we don't 569 * need to worry about launching another command. 570 */ 571 __scsi_release_buffers(cmd, 0); 572 scsi_next_command(cmd); 573 return NULL; 574 } 575 576 static inline unsigned int scsi_sgtable_index(unsigned short nents) 577 { 578 unsigned int index; 579 580 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS); 581 582 if (nents <= 8) 583 index = 0; 584 else 585 index = get_count_order(nents) - 3; 586 587 return index; 588 } 589 590 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents) 591 { 592 struct scsi_host_sg_pool *sgp; 593 594 sgp = scsi_sg_pools + scsi_sgtable_index(nents); 595 mempool_free(sgl, sgp->pool); 596 } 597 598 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask) 599 { 600 struct scsi_host_sg_pool *sgp; 601 602 sgp = scsi_sg_pools + scsi_sgtable_index(nents); 603 return mempool_alloc(sgp->pool, gfp_mask); 604 } 605 606 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, 607 gfp_t gfp_mask) 608 { 609 int ret; 610 611 BUG_ON(!nents); 612 613 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS, 614 gfp_mask, scsi_sg_alloc); 615 if (unlikely(ret)) 616 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, 617 scsi_sg_free); 618 619 return ret; 620 } 621 622 static void scsi_free_sgtable(struct scsi_data_buffer *sdb) 623 { 624 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free); 625 } 626 627 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check) 628 { 629 630 if (cmd->sdb.table.nents) 631 scsi_free_sgtable(&cmd->sdb); 632 633 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 634 635 if (do_bidi_check && scsi_bidi_cmnd(cmd)) { 636 struct scsi_data_buffer *bidi_sdb = 637 cmd->request->next_rq->special; 638 scsi_free_sgtable(bidi_sdb); 639 kmem_cache_free(scsi_sdb_cache, bidi_sdb); 640 cmd->request->next_rq->special = NULL; 641 } 642 643 if (scsi_prot_sg_count(cmd)) 644 scsi_free_sgtable(cmd->prot_sdb); 645 } 646 647 /* 648 * Function: scsi_release_buffers() 649 * 650 * Purpose: Completion processing for block device I/O requests. 651 * 652 * Arguments: cmd - command that we are bailing. 653 * 654 * Lock status: Assumed that no lock is held upon entry. 655 * 656 * Returns: Nothing 657 * 658 * Notes: In the event that an upper level driver rejects a 659 * command, we must release resources allocated during 660 * the __init_io() function. Primarily this would involve 661 * the scatter-gather table, and potentially any bounce 662 * buffers. 663 */ 664 void scsi_release_buffers(struct scsi_cmnd *cmd) 665 { 666 __scsi_release_buffers(cmd, 1); 667 } 668 EXPORT_SYMBOL(scsi_release_buffers); 669 670 /* 671 * Function: scsi_io_completion() 672 * 673 * Purpose: Completion processing for block device I/O requests. 674 * 675 * Arguments: cmd - command that is finished. 676 * 677 * Lock status: Assumed that no lock is held upon entry. 678 * 679 * Returns: Nothing 680 * 681 * Notes: This function is matched in terms of capabilities to 682 * the function that created the scatter-gather list. 683 * In other words, if there are no bounce buffers 684 * (the normal case for most drivers), we don't need 685 * the logic to deal with cleaning up afterwards. 686 * 687 * We must call scsi_end_request(). This will finish off 688 * the specified number of sectors. If we are done, the 689 * command block will be released and the queue function 690 * will be goosed. If we are not done then we have to 691 * figure out what to do next: 692 * 693 * a) We can call scsi_requeue_command(). The request 694 * will be unprepared and put back on the queue. Then 695 * a new command will be created for it. This should 696 * be used if we made forward progress, or if we want 697 * to switch from READ(10) to READ(6) for example. 698 * 699 * b) We can call scsi_queue_insert(). The request will 700 * be put back on the queue and retried using the same 701 * command as before, possibly after a delay. 702 * 703 * c) We can call blk_end_request() with -EIO to fail 704 * the remainder of the request. 705 */ 706 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes) 707 { 708 int result = cmd->result; 709 struct request_queue *q = cmd->device->request_queue; 710 struct request *req = cmd->request; 711 int error = 0; 712 struct scsi_sense_hdr sshdr; 713 int sense_valid = 0; 714 int sense_deferred = 0; 715 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY, 716 ACTION_DELAYED_RETRY} action; 717 char *description = NULL; 718 719 if (result) { 720 sense_valid = scsi_command_normalize_sense(cmd, &sshdr); 721 if (sense_valid) 722 sense_deferred = scsi_sense_is_deferred(&sshdr); 723 } 724 725 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */ 726 req->errors = result; 727 if (result) { 728 if (sense_valid && req->sense) { 729 /* 730 * SG_IO wants current and deferred errors 731 */ 732 int len = 8 + cmd->sense_buffer[7]; 733 734 if (len > SCSI_SENSE_BUFFERSIZE) 735 len = SCSI_SENSE_BUFFERSIZE; 736 memcpy(req->sense, cmd->sense_buffer, len); 737 req->sense_len = len; 738 } 739 if (!sense_deferred) 740 error = -EIO; 741 } 742 743 req->resid_len = scsi_get_resid(cmd); 744 745 if (scsi_bidi_cmnd(cmd)) { 746 /* 747 * Bidi commands Must be complete as a whole, 748 * both sides at once. 749 */ 750 req->next_rq->resid_len = scsi_in(cmd)->resid; 751 752 scsi_release_buffers(cmd); 753 blk_end_request_all(req, 0); 754 755 scsi_next_command(cmd); 756 return; 757 } 758 } 759 760 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */ 761 762 /* 763 * Next deal with any sectors which we were able to correctly 764 * handle. 765 */ 766 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, " 767 "%d bytes done.\n", 768 blk_rq_sectors(req), good_bytes)); 769 770 /* 771 * Recovered errors need reporting, but they're always treated 772 * as success, so fiddle the result code here. For BLOCK_PC 773 * we already took a copy of the original into rq->errors which 774 * is what gets returned to the user 775 */ 776 if (sense_valid && sshdr.sense_key == RECOVERED_ERROR) { 777 if (!(req->cmd_flags & REQ_QUIET)) 778 scsi_print_sense("", cmd); 779 result = 0; 780 /* BLOCK_PC may have set error */ 781 error = 0; 782 } 783 784 /* 785 * A number of bytes were successfully read. If there 786 * are leftovers and there is some kind of error 787 * (result != 0), retry the rest. 788 */ 789 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL) 790 return; 791 792 error = -EIO; 793 794 if (host_byte(result) == DID_RESET) { 795 /* Third party bus reset or reset for error recovery 796 * reasons. Just retry the command and see what 797 * happens. 798 */ 799 action = ACTION_RETRY; 800 } else if (sense_valid && !sense_deferred) { 801 switch (sshdr.sense_key) { 802 case UNIT_ATTENTION: 803 if (cmd->device->removable) { 804 /* Detected disc change. Set a bit 805 * and quietly refuse further access. 806 */ 807 cmd->device->changed = 1; 808 description = "Media Changed"; 809 action = ACTION_FAIL; 810 } else { 811 /* Must have been a power glitch, or a 812 * bus reset. Could not have been a 813 * media change, so we just retry the 814 * command and see what happens. 815 */ 816 action = ACTION_RETRY; 817 } 818 break; 819 case ILLEGAL_REQUEST: 820 /* If we had an ILLEGAL REQUEST returned, then 821 * we may have performed an unsupported 822 * command. The only thing this should be 823 * would be a ten byte read where only a six 824 * byte read was supported. Also, on a system 825 * where READ CAPACITY failed, we may have 826 * read past the end of the disk. 827 */ 828 if ((cmd->device->use_10_for_rw && 829 sshdr.asc == 0x20 && sshdr.ascq == 0x00) && 830 (cmd->cmnd[0] == READ_10 || 831 cmd->cmnd[0] == WRITE_10)) { 832 /* This will issue a new 6-byte command. */ 833 cmd->device->use_10_for_rw = 0; 834 action = ACTION_REPREP; 835 } else if (sshdr.asc == 0x10) /* DIX */ { 836 description = "Host Data Integrity Failure"; 837 action = ACTION_FAIL; 838 error = -EILSEQ; 839 } else 840 action = ACTION_FAIL; 841 break; 842 case ABORTED_COMMAND: 843 action = ACTION_FAIL; 844 if (sshdr.asc == 0x10) { /* DIF */ 845 description = "Target Data Integrity Failure"; 846 error = -EILSEQ; 847 } 848 break; 849 case NOT_READY: 850 /* If the device is in the process of becoming 851 * ready, or has a temporary blockage, retry. 852 */ 853 if (sshdr.asc == 0x04) { 854 switch (sshdr.ascq) { 855 case 0x01: /* becoming ready */ 856 case 0x04: /* format in progress */ 857 case 0x05: /* rebuild in progress */ 858 case 0x06: /* recalculation in progress */ 859 case 0x07: /* operation in progress */ 860 case 0x08: /* Long write in progress */ 861 case 0x09: /* self test in progress */ 862 case 0x14: /* space allocation in progress */ 863 action = ACTION_DELAYED_RETRY; 864 break; 865 default: 866 description = "Device not ready"; 867 action = ACTION_FAIL; 868 break; 869 } 870 } else { 871 description = "Device not ready"; 872 action = ACTION_FAIL; 873 } 874 break; 875 case VOLUME_OVERFLOW: 876 /* See SSC3rXX or current. */ 877 action = ACTION_FAIL; 878 break; 879 default: 880 description = "Unhandled sense code"; 881 action = ACTION_FAIL; 882 break; 883 } 884 } else { 885 description = "Unhandled error code"; 886 action = ACTION_FAIL; 887 } 888 889 switch (action) { 890 case ACTION_FAIL: 891 /* Give up and fail the remainder of the request */ 892 scsi_release_buffers(cmd); 893 if (!(req->cmd_flags & REQ_QUIET)) { 894 if (description) 895 scmd_printk(KERN_INFO, cmd, "%s\n", 896 description); 897 scsi_print_result(cmd); 898 if (driver_byte(result) & DRIVER_SENSE) 899 scsi_print_sense("", cmd); 900 scsi_print_command(cmd); 901 } 902 if (blk_end_request_err(req, error)) 903 scsi_requeue_command(q, cmd); 904 else 905 scsi_next_command(cmd); 906 break; 907 case ACTION_REPREP: 908 /* Unprep the request and put it back at the head of the queue. 909 * A new command will be prepared and issued. 910 */ 911 scsi_release_buffers(cmd); 912 scsi_requeue_command(q, cmd); 913 break; 914 case ACTION_RETRY: 915 /* Retry the same command immediately */ 916 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0); 917 break; 918 case ACTION_DELAYED_RETRY: 919 /* Retry the same command after a delay */ 920 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0); 921 break; 922 } 923 } 924 925 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb, 926 gfp_t gfp_mask) 927 { 928 int count; 929 930 /* 931 * If sg table allocation fails, requeue request later. 932 */ 933 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments, 934 gfp_mask))) { 935 return BLKPREP_DEFER; 936 } 937 938 req->buffer = NULL; 939 940 /* 941 * Next, walk the list, and fill in the addresses and sizes of 942 * each segment. 943 */ 944 count = blk_rq_map_sg(req->q, req, sdb->table.sgl); 945 BUG_ON(count > sdb->table.nents); 946 sdb->table.nents = count; 947 sdb->length = blk_rq_bytes(req); 948 return BLKPREP_OK; 949 } 950 951 /* 952 * Function: scsi_init_io() 953 * 954 * Purpose: SCSI I/O initialize function. 955 * 956 * Arguments: cmd - Command descriptor we wish to initialize 957 * 958 * Returns: 0 on success 959 * BLKPREP_DEFER if the failure is retryable 960 * BLKPREP_KILL if the failure is fatal 961 */ 962 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask) 963 { 964 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask); 965 if (error) 966 goto err_exit; 967 968 if (blk_bidi_rq(cmd->request)) { 969 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc( 970 scsi_sdb_cache, GFP_ATOMIC); 971 if (!bidi_sdb) { 972 error = BLKPREP_DEFER; 973 goto err_exit; 974 } 975 976 cmd->request->next_rq->special = bidi_sdb; 977 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb, 978 GFP_ATOMIC); 979 if (error) 980 goto err_exit; 981 } 982 983 if (blk_integrity_rq(cmd->request)) { 984 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb; 985 int ivecs, count; 986 987 BUG_ON(prot_sdb == NULL); 988 ivecs = blk_rq_count_integrity_sg(cmd->request); 989 990 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) { 991 error = BLKPREP_DEFER; 992 goto err_exit; 993 } 994 995 count = blk_rq_map_integrity_sg(cmd->request, 996 prot_sdb->table.sgl); 997 BUG_ON(unlikely(count > ivecs)); 998 999 cmd->prot_sdb = prot_sdb; 1000 cmd->prot_sdb->table.nents = count; 1001 } 1002 1003 return BLKPREP_OK ; 1004 1005 err_exit: 1006 scsi_release_buffers(cmd); 1007 if (error == BLKPREP_KILL) 1008 scsi_put_command(cmd); 1009 else /* BLKPREP_DEFER */ 1010 scsi_unprep_request(cmd->request); 1011 1012 return error; 1013 } 1014 EXPORT_SYMBOL(scsi_init_io); 1015 1016 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev, 1017 struct request *req) 1018 { 1019 struct scsi_cmnd *cmd; 1020 1021 if (!req->special) { 1022 cmd = scsi_get_command(sdev, GFP_ATOMIC); 1023 if (unlikely(!cmd)) 1024 return NULL; 1025 req->special = cmd; 1026 } else { 1027 cmd = req->special; 1028 } 1029 1030 /* pull a tag out of the request if we have one */ 1031 cmd->tag = req->tag; 1032 cmd->request = req; 1033 1034 cmd->cmnd = req->cmd; 1035 1036 return cmd; 1037 } 1038 1039 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req) 1040 { 1041 struct scsi_cmnd *cmd; 1042 int ret = scsi_prep_state_check(sdev, req); 1043 1044 if (ret != BLKPREP_OK) 1045 return ret; 1046 1047 cmd = scsi_get_cmd_from_req(sdev, req); 1048 if (unlikely(!cmd)) 1049 return BLKPREP_DEFER; 1050 1051 /* 1052 * BLOCK_PC requests may transfer data, in which case they must 1053 * a bio attached to them. Or they might contain a SCSI command 1054 * that does not transfer data, in which case they may optionally 1055 * submit a request without an attached bio. 1056 */ 1057 if (req->bio) { 1058 int ret; 1059 1060 BUG_ON(!req->nr_phys_segments); 1061 1062 ret = scsi_init_io(cmd, GFP_ATOMIC); 1063 if (unlikely(ret)) 1064 return ret; 1065 } else { 1066 BUG_ON(blk_rq_bytes(req)); 1067 1068 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1069 req->buffer = NULL; 1070 } 1071 1072 cmd->cmd_len = req->cmd_len; 1073 if (!blk_rq_bytes(req)) 1074 cmd->sc_data_direction = DMA_NONE; 1075 else if (rq_data_dir(req) == WRITE) 1076 cmd->sc_data_direction = DMA_TO_DEVICE; 1077 else 1078 cmd->sc_data_direction = DMA_FROM_DEVICE; 1079 1080 cmd->transfersize = blk_rq_bytes(req); 1081 cmd->allowed = req->retries; 1082 return BLKPREP_OK; 1083 } 1084 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd); 1085 1086 /* 1087 * Setup a REQ_TYPE_FS command. These are simple read/write request 1088 * from filesystems that still need to be translated to SCSI CDBs from 1089 * the ULD. 1090 */ 1091 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req) 1092 { 1093 struct scsi_cmnd *cmd; 1094 int ret = scsi_prep_state_check(sdev, req); 1095 1096 if (ret != BLKPREP_OK) 1097 return ret; 1098 1099 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh 1100 && sdev->scsi_dh_data->scsi_dh->prep_fn)) { 1101 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req); 1102 if (ret != BLKPREP_OK) 1103 return ret; 1104 } 1105 1106 /* 1107 * Filesystem requests must transfer data. 1108 */ 1109 BUG_ON(!req->nr_phys_segments); 1110 1111 cmd = scsi_get_cmd_from_req(sdev, req); 1112 if (unlikely(!cmd)) 1113 return BLKPREP_DEFER; 1114 1115 memset(cmd->cmnd, 0, BLK_MAX_CDB); 1116 return scsi_init_io(cmd, GFP_ATOMIC); 1117 } 1118 EXPORT_SYMBOL(scsi_setup_fs_cmnd); 1119 1120 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req) 1121 { 1122 int ret = BLKPREP_OK; 1123 1124 /* 1125 * If the device is not in running state we will reject some 1126 * or all commands. 1127 */ 1128 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) { 1129 switch (sdev->sdev_state) { 1130 case SDEV_OFFLINE: 1131 /* 1132 * If the device is offline we refuse to process any 1133 * commands. The device must be brought online 1134 * before trying any recovery commands. 1135 */ 1136 sdev_printk(KERN_ERR, sdev, 1137 "rejecting I/O to offline device\n"); 1138 ret = BLKPREP_KILL; 1139 break; 1140 case SDEV_DEL: 1141 /* 1142 * If the device is fully deleted, we refuse to 1143 * process any commands as well. 1144 */ 1145 sdev_printk(KERN_ERR, sdev, 1146 "rejecting I/O to dead device\n"); 1147 ret = BLKPREP_KILL; 1148 break; 1149 case SDEV_QUIESCE: 1150 case SDEV_BLOCK: 1151 case SDEV_CREATED_BLOCK: 1152 /* 1153 * If the devices is blocked we defer normal commands. 1154 */ 1155 if (!(req->cmd_flags & REQ_PREEMPT)) 1156 ret = BLKPREP_DEFER; 1157 break; 1158 default: 1159 /* 1160 * For any other not fully online state we only allow 1161 * special commands. In particular any user initiated 1162 * command is not allowed. 1163 */ 1164 if (!(req->cmd_flags & REQ_PREEMPT)) 1165 ret = BLKPREP_KILL; 1166 break; 1167 } 1168 } 1169 return ret; 1170 } 1171 EXPORT_SYMBOL(scsi_prep_state_check); 1172 1173 int scsi_prep_return(struct request_queue *q, struct request *req, int ret) 1174 { 1175 struct scsi_device *sdev = q->queuedata; 1176 1177 switch (ret) { 1178 case BLKPREP_KILL: 1179 req->errors = DID_NO_CONNECT << 16; 1180 /* release the command and kill it */ 1181 if (req->special) { 1182 struct scsi_cmnd *cmd = req->special; 1183 scsi_release_buffers(cmd); 1184 scsi_put_command(cmd); 1185 req->special = NULL; 1186 } 1187 break; 1188 case BLKPREP_DEFER: 1189 /* 1190 * If we defer, the blk_peek_request() returns NULL, but the 1191 * queue must be restarted, so we plug here if no returning 1192 * command will automatically do that. 1193 */ 1194 if (sdev->device_busy == 0) 1195 blk_plug_device(q); 1196 break; 1197 default: 1198 req->cmd_flags |= REQ_DONTPREP; 1199 } 1200 1201 return ret; 1202 } 1203 EXPORT_SYMBOL(scsi_prep_return); 1204 1205 int scsi_prep_fn(struct request_queue *q, struct request *req) 1206 { 1207 struct scsi_device *sdev = q->queuedata; 1208 int ret = BLKPREP_KILL; 1209 1210 if (req->cmd_type == REQ_TYPE_BLOCK_PC) 1211 ret = scsi_setup_blk_pc_cmnd(sdev, req); 1212 return scsi_prep_return(q, req, ret); 1213 } 1214 EXPORT_SYMBOL(scsi_prep_fn); 1215 1216 /* 1217 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else 1218 * return 0. 1219 * 1220 * Called with the queue_lock held. 1221 */ 1222 static inline int scsi_dev_queue_ready(struct request_queue *q, 1223 struct scsi_device *sdev) 1224 { 1225 if (sdev->device_busy == 0 && sdev->device_blocked) { 1226 /* 1227 * unblock after device_blocked iterates to zero 1228 */ 1229 if (--sdev->device_blocked == 0) { 1230 SCSI_LOG_MLQUEUE(3, 1231 sdev_printk(KERN_INFO, sdev, 1232 "unblocking device at zero depth\n")); 1233 } else { 1234 blk_plug_device(q); 1235 return 0; 1236 } 1237 } 1238 if (scsi_device_is_busy(sdev)) 1239 return 0; 1240 1241 return 1; 1242 } 1243 1244 1245 /* 1246 * scsi_target_queue_ready: checks if there we can send commands to target 1247 * @sdev: scsi device on starget to check. 1248 * 1249 * Called with the host lock held. 1250 */ 1251 static inline int scsi_target_queue_ready(struct Scsi_Host *shost, 1252 struct scsi_device *sdev) 1253 { 1254 struct scsi_target *starget = scsi_target(sdev); 1255 1256 if (starget->single_lun) { 1257 if (starget->starget_sdev_user && 1258 starget->starget_sdev_user != sdev) 1259 return 0; 1260 starget->starget_sdev_user = sdev; 1261 } 1262 1263 if (starget->target_busy == 0 && starget->target_blocked) { 1264 /* 1265 * unblock after target_blocked iterates to zero 1266 */ 1267 if (--starget->target_blocked == 0) { 1268 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget, 1269 "unblocking target at zero depth\n")); 1270 } else 1271 return 0; 1272 } 1273 1274 if (scsi_target_is_busy(starget)) { 1275 if (list_empty(&sdev->starved_entry)) { 1276 list_add_tail(&sdev->starved_entry, 1277 &shost->starved_list); 1278 return 0; 1279 } 1280 } 1281 1282 /* We're OK to process the command, so we can't be starved */ 1283 if (!list_empty(&sdev->starved_entry)) 1284 list_del_init(&sdev->starved_entry); 1285 return 1; 1286 } 1287 1288 /* 1289 * scsi_host_queue_ready: if we can send requests to shost, return 1 else 1290 * return 0. We must end up running the queue again whenever 0 is 1291 * returned, else IO can hang. 1292 * 1293 * Called with host_lock held. 1294 */ 1295 static inline int scsi_host_queue_ready(struct request_queue *q, 1296 struct Scsi_Host *shost, 1297 struct scsi_device *sdev) 1298 { 1299 if (scsi_host_in_recovery(shost)) 1300 return 0; 1301 if (shost->host_busy == 0 && shost->host_blocked) { 1302 /* 1303 * unblock after host_blocked iterates to zero 1304 */ 1305 if (--shost->host_blocked == 0) { 1306 SCSI_LOG_MLQUEUE(3, 1307 printk("scsi%d unblocking host at zero depth\n", 1308 shost->host_no)); 1309 } else { 1310 return 0; 1311 } 1312 } 1313 if (scsi_host_is_busy(shost)) { 1314 if (list_empty(&sdev->starved_entry)) 1315 list_add_tail(&sdev->starved_entry, &shost->starved_list); 1316 return 0; 1317 } 1318 1319 /* We're OK to process the command, so we can't be starved */ 1320 if (!list_empty(&sdev->starved_entry)) 1321 list_del_init(&sdev->starved_entry); 1322 1323 return 1; 1324 } 1325 1326 /* 1327 * Busy state exporting function for request stacking drivers. 1328 * 1329 * For efficiency, no lock is taken to check the busy state of 1330 * shost/starget/sdev, since the returned value is not guaranteed and 1331 * may be changed after request stacking drivers call the function, 1332 * regardless of taking lock or not. 1333 * 1334 * When scsi can't dispatch I/Os anymore and needs to kill I/Os 1335 * (e.g. !sdev), scsi needs to return 'not busy'. 1336 * Otherwise, request stacking drivers may hold requests forever. 1337 */ 1338 static int scsi_lld_busy(struct request_queue *q) 1339 { 1340 struct scsi_device *sdev = q->queuedata; 1341 struct Scsi_Host *shost; 1342 struct scsi_target *starget; 1343 1344 if (!sdev) 1345 return 0; 1346 1347 shost = sdev->host; 1348 starget = scsi_target(sdev); 1349 1350 if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) || 1351 scsi_target_is_busy(starget) || scsi_device_is_busy(sdev)) 1352 return 1; 1353 1354 return 0; 1355 } 1356 1357 /* 1358 * Kill a request for a dead device 1359 */ 1360 static void scsi_kill_request(struct request *req, struct request_queue *q) 1361 { 1362 struct scsi_cmnd *cmd = req->special; 1363 struct scsi_device *sdev; 1364 struct scsi_target *starget; 1365 struct Scsi_Host *shost; 1366 1367 blk_start_request(req); 1368 1369 if (unlikely(cmd == NULL)) { 1370 printk(KERN_CRIT "impossible request in %s.\n", 1371 __func__); 1372 BUG(); 1373 } 1374 1375 sdev = cmd->device; 1376 starget = scsi_target(sdev); 1377 shost = sdev->host; 1378 scsi_init_cmd_errh(cmd); 1379 cmd->result = DID_NO_CONNECT << 16; 1380 atomic_inc(&cmd->device->iorequest_cnt); 1381 1382 /* 1383 * SCSI request completion path will do scsi_device_unbusy(), 1384 * bump busy counts. To bump the counters, we need to dance 1385 * with the locks as normal issue path does. 1386 */ 1387 sdev->device_busy++; 1388 spin_unlock(sdev->request_queue->queue_lock); 1389 spin_lock(shost->host_lock); 1390 shost->host_busy++; 1391 starget->target_busy++; 1392 spin_unlock(shost->host_lock); 1393 spin_lock(sdev->request_queue->queue_lock); 1394 1395 blk_complete_request(req); 1396 } 1397 1398 static void scsi_softirq_done(struct request *rq) 1399 { 1400 struct scsi_cmnd *cmd = rq->special; 1401 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout; 1402 int disposition; 1403 1404 INIT_LIST_HEAD(&cmd->eh_entry); 1405 1406 /* 1407 * Set the serial numbers back to zero 1408 */ 1409 cmd->serial_number = 0; 1410 1411 atomic_inc(&cmd->device->iodone_cnt); 1412 if (cmd->result) 1413 atomic_inc(&cmd->device->ioerr_cnt); 1414 1415 disposition = scsi_decide_disposition(cmd); 1416 if (disposition != SUCCESS && 1417 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) { 1418 sdev_printk(KERN_ERR, cmd->device, 1419 "timing out command, waited %lus\n", 1420 wait_for/HZ); 1421 disposition = SUCCESS; 1422 } 1423 1424 scsi_log_completion(cmd, disposition); 1425 1426 switch (disposition) { 1427 case SUCCESS: 1428 scsi_finish_command(cmd); 1429 break; 1430 case NEEDS_RETRY: 1431 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY); 1432 break; 1433 case ADD_TO_MLQUEUE: 1434 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY); 1435 break; 1436 default: 1437 if (!scsi_eh_scmd_add(cmd, 0)) 1438 scsi_finish_command(cmd); 1439 } 1440 } 1441 1442 /* 1443 * Function: scsi_request_fn() 1444 * 1445 * Purpose: Main strategy routine for SCSI. 1446 * 1447 * Arguments: q - Pointer to actual queue. 1448 * 1449 * Returns: Nothing 1450 * 1451 * Lock status: IO request lock assumed to be held when called. 1452 */ 1453 static void scsi_request_fn(struct request_queue *q) 1454 { 1455 struct scsi_device *sdev = q->queuedata; 1456 struct Scsi_Host *shost; 1457 struct scsi_cmnd *cmd; 1458 struct request *req; 1459 1460 if (!sdev) { 1461 printk("scsi: killing requests for dead queue\n"); 1462 while ((req = blk_peek_request(q)) != NULL) 1463 scsi_kill_request(req, q); 1464 return; 1465 } 1466 1467 if(!get_device(&sdev->sdev_gendev)) 1468 /* We must be tearing the block queue down already */ 1469 return; 1470 1471 /* 1472 * To start with, we keep looping until the queue is empty, or until 1473 * the host is no longer able to accept any more requests. 1474 */ 1475 shost = sdev->host; 1476 while (!blk_queue_plugged(q)) { 1477 int rtn; 1478 /* 1479 * get next queueable request. We do this early to make sure 1480 * that the request is fully prepared even if we cannot 1481 * accept it. 1482 */ 1483 req = blk_peek_request(q); 1484 if (!req || !scsi_dev_queue_ready(q, sdev)) 1485 break; 1486 1487 if (unlikely(!scsi_device_online(sdev))) { 1488 sdev_printk(KERN_ERR, sdev, 1489 "rejecting I/O to offline device\n"); 1490 scsi_kill_request(req, q); 1491 continue; 1492 } 1493 1494 1495 /* 1496 * Remove the request from the request list. 1497 */ 1498 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req))) 1499 blk_start_request(req); 1500 sdev->device_busy++; 1501 1502 spin_unlock(q->queue_lock); 1503 cmd = req->special; 1504 if (unlikely(cmd == NULL)) { 1505 printk(KERN_CRIT "impossible request in %s.\n" 1506 "please mail a stack trace to " 1507 "linux-scsi@vger.kernel.org\n", 1508 __func__); 1509 blk_dump_rq_flags(req, "foo"); 1510 BUG(); 1511 } 1512 spin_lock(shost->host_lock); 1513 1514 /* 1515 * We hit this when the driver is using a host wide 1516 * tag map. For device level tag maps the queue_depth check 1517 * in the device ready fn would prevent us from trying 1518 * to allocate a tag. Since the map is a shared host resource 1519 * we add the dev to the starved list so it eventually gets 1520 * a run when a tag is freed. 1521 */ 1522 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) { 1523 if (list_empty(&sdev->starved_entry)) 1524 list_add_tail(&sdev->starved_entry, 1525 &shost->starved_list); 1526 goto not_ready; 1527 } 1528 1529 if (!scsi_target_queue_ready(shost, sdev)) 1530 goto not_ready; 1531 1532 if (!scsi_host_queue_ready(q, shost, sdev)) 1533 goto not_ready; 1534 1535 scsi_target(sdev)->target_busy++; 1536 shost->host_busy++; 1537 1538 /* 1539 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will 1540 * take the lock again. 1541 */ 1542 spin_unlock_irq(shost->host_lock); 1543 1544 /* 1545 * Finally, initialize any error handling parameters, and set up 1546 * the timers for timeouts. 1547 */ 1548 scsi_init_cmd_errh(cmd); 1549 1550 /* 1551 * Dispatch the command to the low-level driver. 1552 */ 1553 rtn = scsi_dispatch_cmd(cmd); 1554 spin_lock_irq(q->queue_lock); 1555 if(rtn) { 1556 /* we're refusing the command; because of 1557 * the way locks get dropped, we need to 1558 * check here if plugging is required */ 1559 if(sdev->device_busy == 0) 1560 blk_plug_device(q); 1561 1562 break; 1563 } 1564 } 1565 1566 goto out; 1567 1568 not_ready: 1569 spin_unlock_irq(shost->host_lock); 1570 1571 /* 1572 * lock q, handle tag, requeue req, and decrement device_busy. We 1573 * must return with queue_lock held. 1574 * 1575 * Decrementing device_busy without checking it is OK, as all such 1576 * cases (host limits or settings) should run the queue at some 1577 * later time. 1578 */ 1579 spin_lock_irq(q->queue_lock); 1580 blk_requeue_request(q, req); 1581 sdev->device_busy--; 1582 if(sdev->device_busy == 0) 1583 blk_plug_device(q); 1584 out: 1585 /* must be careful here...if we trigger the ->remove() function 1586 * we cannot be holding the q lock */ 1587 spin_unlock_irq(q->queue_lock); 1588 put_device(&sdev->sdev_gendev); 1589 spin_lock_irq(q->queue_lock); 1590 } 1591 1592 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost) 1593 { 1594 struct device *host_dev; 1595 u64 bounce_limit = 0xffffffff; 1596 1597 if (shost->unchecked_isa_dma) 1598 return BLK_BOUNCE_ISA; 1599 /* 1600 * Platforms with virtual-DMA translation 1601 * hardware have no practical limit. 1602 */ 1603 if (!PCI_DMA_BUS_IS_PHYS) 1604 return BLK_BOUNCE_ANY; 1605 1606 host_dev = scsi_get_device(shost); 1607 if (host_dev && host_dev->dma_mask) 1608 bounce_limit = *host_dev->dma_mask; 1609 1610 return bounce_limit; 1611 } 1612 EXPORT_SYMBOL(scsi_calculate_bounce_limit); 1613 1614 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost, 1615 request_fn_proc *request_fn) 1616 { 1617 struct request_queue *q; 1618 struct device *dev = shost->shost_gendev.parent; 1619 1620 q = blk_init_queue(request_fn, NULL); 1621 if (!q) 1622 return NULL; 1623 1624 /* 1625 * this limit is imposed by hardware restrictions 1626 */ 1627 blk_queue_max_hw_segments(q, shost->sg_tablesize); 1628 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS); 1629 1630 blk_queue_max_sectors(q, shost->max_sectors); 1631 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost)); 1632 blk_queue_segment_boundary(q, shost->dma_boundary); 1633 dma_set_seg_boundary(dev, shost->dma_boundary); 1634 1635 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev)); 1636 1637 /* New queue, no concurrency on queue_flags */ 1638 if (!shost->use_clustering) 1639 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q); 1640 1641 /* 1642 * set a reasonable default alignment on word boundaries: the 1643 * host and device may alter it using 1644 * blk_queue_update_dma_alignment() later. 1645 */ 1646 blk_queue_dma_alignment(q, 0x03); 1647 1648 return q; 1649 } 1650 EXPORT_SYMBOL(__scsi_alloc_queue); 1651 1652 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev) 1653 { 1654 struct request_queue *q; 1655 1656 q = __scsi_alloc_queue(sdev->host, scsi_request_fn); 1657 if (!q) 1658 return NULL; 1659 1660 blk_queue_prep_rq(q, scsi_prep_fn); 1661 blk_queue_softirq_done(q, scsi_softirq_done); 1662 blk_queue_rq_timed_out(q, scsi_times_out); 1663 blk_queue_lld_busy(q, scsi_lld_busy); 1664 return q; 1665 } 1666 1667 void scsi_free_queue(struct request_queue *q) 1668 { 1669 blk_cleanup_queue(q); 1670 } 1671 1672 /* 1673 * Function: scsi_block_requests() 1674 * 1675 * Purpose: Utility function used by low-level drivers to prevent further 1676 * commands from being queued to the device. 1677 * 1678 * Arguments: shost - Host in question 1679 * 1680 * Returns: Nothing 1681 * 1682 * Lock status: No locks are assumed held. 1683 * 1684 * Notes: There is no timer nor any other means by which the requests 1685 * get unblocked other than the low-level driver calling 1686 * scsi_unblock_requests(). 1687 */ 1688 void scsi_block_requests(struct Scsi_Host *shost) 1689 { 1690 shost->host_self_blocked = 1; 1691 } 1692 EXPORT_SYMBOL(scsi_block_requests); 1693 1694 /* 1695 * Function: scsi_unblock_requests() 1696 * 1697 * Purpose: Utility function used by low-level drivers to allow further 1698 * commands from being queued to the device. 1699 * 1700 * Arguments: shost - Host in question 1701 * 1702 * Returns: Nothing 1703 * 1704 * Lock status: No locks are assumed held. 1705 * 1706 * Notes: There is no timer nor any other means by which the requests 1707 * get unblocked other than the low-level driver calling 1708 * scsi_unblock_requests(). 1709 * 1710 * This is done as an API function so that changes to the 1711 * internals of the scsi mid-layer won't require wholesale 1712 * changes to drivers that use this feature. 1713 */ 1714 void scsi_unblock_requests(struct Scsi_Host *shost) 1715 { 1716 shost->host_self_blocked = 0; 1717 scsi_run_host_queues(shost); 1718 } 1719 EXPORT_SYMBOL(scsi_unblock_requests); 1720 1721 int __init scsi_init_queue(void) 1722 { 1723 int i; 1724 1725 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer", 1726 sizeof(struct scsi_data_buffer), 1727 0, 0, NULL); 1728 if (!scsi_sdb_cache) { 1729 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n"); 1730 return -ENOMEM; 1731 } 1732 1733 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1734 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1735 int size = sgp->size * sizeof(struct scatterlist); 1736 1737 sgp->slab = kmem_cache_create(sgp->name, size, 0, 1738 SLAB_HWCACHE_ALIGN, NULL); 1739 if (!sgp->slab) { 1740 printk(KERN_ERR "SCSI: can't init sg slab %s\n", 1741 sgp->name); 1742 goto cleanup_sdb; 1743 } 1744 1745 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE, 1746 sgp->slab); 1747 if (!sgp->pool) { 1748 printk(KERN_ERR "SCSI: can't init sg mempool %s\n", 1749 sgp->name); 1750 goto cleanup_sdb; 1751 } 1752 } 1753 1754 return 0; 1755 1756 cleanup_sdb: 1757 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1758 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1759 if (sgp->pool) 1760 mempool_destroy(sgp->pool); 1761 if (sgp->slab) 1762 kmem_cache_destroy(sgp->slab); 1763 } 1764 kmem_cache_destroy(scsi_sdb_cache); 1765 1766 return -ENOMEM; 1767 } 1768 1769 void scsi_exit_queue(void) 1770 { 1771 int i; 1772 1773 kmem_cache_destroy(scsi_sdb_cache); 1774 1775 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1776 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1777 mempool_destroy(sgp->pool); 1778 kmem_cache_destroy(sgp->slab); 1779 } 1780 } 1781 1782 /** 1783 * scsi_mode_select - issue a mode select 1784 * @sdev: SCSI device to be queried 1785 * @pf: Page format bit (1 == standard, 0 == vendor specific) 1786 * @sp: Save page bit (0 == don't save, 1 == save) 1787 * @modepage: mode page being requested 1788 * @buffer: request buffer (may not be smaller than eight bytes) 1789 * @len: length of request buffer. 1790 * @timeout: command timeout 1791 * @retries: number of retries before failing 1792 * @data: returns a structure abstracting the mode header data 1793 * @sshdr: place to put sense data (or NULL if no sense to be collected). 1794 * must be SCSI_SENSE_BUFFERSIZE big. 1795 * 1796 * Returns zero if successful; negative error number or scsi 1797 * status on error 1798 * 1799 */ 1800 int 1801 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage, 1802 unsigned char *buffer, int len, int timeout, int retries, 1803 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 1804 { 1805 unsigned char cmd[10]; 1806 unsigned char *real_buffer; 1807 int ret; 1808 1809 memset(cmd, 0, sizeof(cmd)); 1810 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0); 1811 1812 if (sdev->use_10_for_ms) { 1813 if (len > 65535) 1814 return -EINVAL; 1815 real_buffer = kmalloc(8 + len, GFP_KERNEL); 1816 if (!real_buffer) 1817 return -ENOMEM; 1818 memcpy(real_buffer + 8, buffer, len); 1819 len += 8; 1820 real_buffer[0] = 0; 1821 real_buffer[1] = 0; 1822 real_buffer[2] = data->medium_type; 1823 real_buffer[3] = data->device_specific; 1824 real_buffer[4] = data->longlba ? 0x01 : 0; 1825 real_buffer[5] = 0; 1826 real_buffer[6] = data->block_descriptor_length >> 8; 1827 real_buffer[7] = data->block_descriptor_length; 1828 1829 cmd[0] = MODE_SELECT_10; 1830 cmd[7] = len >> 8; 1831 cmd[8] = len; 1832 } else { 1833 if (len > 255 || data->block_descriptor_length > 255 || 1834 data->longlba) 1835 return -EINVAL; 1836 1837 real_buffer = kmalloc(4 + len, GFP_KERNEL); 1838 if (!real_buffer) 1839 return -ENOMEM; 1840 memcpy(real_buffer + 4, buffer, len); 1841 len += 4; 1842 real_buffer[0] = 0; 1843 real_buffer[1] = data->medium_type; 1844 real_buffer[2] = data->device_specific; 1845 real_buffer[3] = data->block_descriptor_length; 1846 1847 1848 cmd[0] = MODE_SELECT; 1849 cmd[4] = len; 1850 } 1851 1852 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len, 1853 sshdr, timeout, retries, NULL); 1854 kfree(real_buffer); 1855 return ret; 1856 } 1857 EXPORT_SYMBOL_GPL(scsi_mode_select); 1858 1859 /** 1860 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary. 1861 * @sdev: SCSI device to be queried 1862 * @dbd: set if mode sense will allow block descriptors to be returned 1863 * @modepage: mode page being requested 1864 * @buffer: request buffer (may not be smaller than eight bytes) 1865 * @len: length of request buffer. 1866 * @timeout: command timeout 1867 * @retries: number of retries before failing 1868 * @data: returns a structure abstracting the mode header data 1869 * @sshdr: place to put sense data (or NULL if no sense to be collected). 1870 * must be SCSI_SENSE_BUFFERSIZE big. 1871 * 1872 * Returns zero if unsuccessful, or the header offset (either 4 1873 * or 8 depending on whether a six or ten byte command was 1874 * issued) if successful. 1875 */ 1876 int 1877 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, 1878 unsigned char *buffer, int len, int timeout, int retries, 1879 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 1880 { 1881 unsigned char cmd[12]; 1882 int use_10_for_ms; 1883 int header_length; 1884 int result; 1885 struct scsi_sense_hdr my_sshdr; 1886 1887 memset(data, 0, sizeof(*data)); 1888 memset(&cmd[0], 0, 12); 1889 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */ 1890 cmd[2] = modepage; 1891 1892 /* caller might not be interested in sense, but we need it */ 1893 if (!sshdr) 1894 sshdr = &my_sshdr; 1895 1896 retry: 1897 use_10_for_ms = sdev->use_10_for_ms; 1898 1899 if (use_10_for_ms) { 1900 if (len < 8) 1901 len = 8; 1902 1903 cmd[0] = MODE_SENSE_10; 1904 cmd[8] = len; 1905 header_length = 8; 1906 } else { 1907 if (len < 4) 1908 len = 4; 1909 1910 cmd[0] = MODE_SENSE; 1911 cmd[4] = len; 1912 header_length = 4; 1913 } 1914 1915 memset(buffer, 0, len); 1916 1917 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len, 1918 sshdr, timeout, retries, NULL); 1919 1920 /* This code looks awful: what it's doing is making sure an 1921 * ILLEGAL REQUEST sense return identifies the actual command 1922 * byte as the problem. MODE_SENSE commands can return 1923 * ILLEGAL REQUEST if the code page isn't supported */ 1924 1925 if (use_10_for_ms && !scsi_status_is_good(result) && 1926 (driver_byte(result) & DRIVER_SENSE)) { 1927 if (scsi_sense_valid(sshdr)) { 1928 if ((sshdr->sense_key == ILLEGAL_REQUEST) && 1929 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) { 1930 /* 1931 * Invalid command operation code 1932 */ 1933 sdev->use_10_for_ms = 0; 1934 goto retry; 1935 } 1936 } 1937 } 1938 1939 if(scsi_status_is_good(result)) { 1940 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b && 1941 (modepage == 6 || modepage == 8))) { 1942 /* Initio breakage? */ 1943 header_length = 0; 1944 data->length = 13; 1945 data->medium_type = 0; 1946 data->device_specific = 0; 1947 data->longlba = 0; 1948 data->block_descriptor_length = 0; 1949 } else if(use_10_for_ms) { 1950 data->length = buffer[0]*256 + buffer[1] + 2; 1951 data->medium_type = buffer[2]; 1952 data->device_specific = buffer[3]; 1953 data->longlba = buffer[4] & 0x01; 1954 data->block_descriptor_length = buffer[6]*256 1955 + buffer[7]; 1956 } else { 1957 data->length = buffer[0] + 1; 1958 data->medium_type = buffer[1]; 1959 data->device_specific = buffer[2]; 1960 data->block_descriptor_length = buffer[3]; 1961 } 1962 data->header_length = header_length; 1963 } 1964 1965 return result; 1966 } 1967 EXPORT_SYMBOL(scsi_mode_sense); 1968 1969 /** 1970 * scsi_test_unit_ready - test if unit is ready 1971 * @sdev: scsi device to change the state of. 1972 * @timeout: command timeout 1973 * @retries: number of retries before failing 1974 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for 1975 * returning sense. Make sure that this is cleared before passing 1976 * in. 1977 * 1978 * Returns zero if unsuccessful or an error if TUR failed. For 1979 * removable media, a return of NOT_READY or UNIT_ATTENTION is 1980 * translated to success, with the ->changed flag updated. 1981 **/ 1982 int 1983 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries, 1984 struct scsi_sense_hdr *sshdr_external) 1985 { 1986 char cmd[] = { 1987 TEST_UNIT_READY, 0, 0, 0, 0, 0, 1988 }; 1989 struct scsi_sense_hdr *sshdr; 1990 int result; 1991 1992 if (!sshdr_external) 1993 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL); 1994 else 1995 sshdr = sshdr_external; 1996 1997 /* try to eat the UNIT_ATTENTION if there are enough retries */ 1998 do { 1999 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr, 2000 timeout, retries, NULL); 2001 if (sdev->removable && scsi_sense_valid(sshdr) && 2002 sshdr->sense_key == UNIT_ATTENTION) 2003 sdev->changed = 1; 2004 } while (scsi_sense_valid(sshdr) && 2005 sshdr->sense_key == UNIT_ATTENTION && --retries); 2006 2007 if (!sshdr) 2008 /* could not allocate sense buffer, so can't process it */ 2009 return result; 2010 2011 if (sdev->removable && scsi_sense_valid(sshdr) && 2012 (sshdr->sense_key == UNIT_ATTENTION || 2013 sshdr->sense_key == NOT_READY)) { 2014 sdev->changed = 1; 2015 result = 0; 2016 } 2017 if (!sshdr_external) 2018 kfree(sshdr); 2019 return result; 2020 } 2021 EXPORT_SYMBOL(scsi_test_unit_ready); 2022 2023 /** 2024 * scsi_device_set_state - Take the given device through the device state model. 2025 * @sdev: scsi device to change the state of. 2026 * @state: state to change to. 2027 * 2028 * Returns zero if unsuccessful or an error if the requested 2029 * transition is illegal. 2030 */ 2031 int 2032 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state) 2033 { 2034 enum scsi_device_state oldstate = sdev->sdev_state; 2035 2036 if (state == oldstate) 2037 return 0; 2038 2039 switch (state) { 2040 case SDEV_CREATED: 2041 switch (oldstate) { 2042 case SDEV_CREATED_BLOCK: 2043 break; 2044 default: 2045 goto illegal; 2046 } 2047 break; 2048 2049 case SDEV_RUNNING: 2050 switch (oldstate) { 2051 case SDEV_CREATED: 2052 case SDEV_OFFLINE: 2053 case SDEV_QUIESCE: 2054 case SDEV_BLOCK: 2055 break; 2056 default: 2057 goto illegal; 2058 } 2059 break; 2060 2061 case SDEV_QUIESCE: 2062 switch (oldstate) { 2063 case SDEV_RUNNING: 2064 case SDEV_OFFLINE: 2065 break; 2066 default: 2067 goto illegal; 2068 } 2069 break; 2070 2071 case SDEV_OFFLINE: 2072 switch (oldstate) { 2073 case SDEV_CREATED: 2074 case SDEV_RUNNING: 2075 case SDEV_QUIESCE: 2076 case SDEV_BLOCK: 2077 break; 2078 default: 2079 goto illegal; 2080 } 2081 break; 2082 2083 case SDEV_BLOCK: 2084 switch (oldstate) { 2085 case SDEV_RUNNING: 2086 case SDEV_CREATED_BLOCK: 2087 break; 2088 default: 2089 goto illegal; 2090 } 2091 break; 2092 2093 case SDEV_CREATED_BLOCK: 2094 switch (oldstate) { 2095 case SDEV_CREATED: 2096 break; 2097 default: 2098 goto illegal; 2099 } 2100 break; 2101 2102 case SDEV_CANCEL: 2103 switch (oldstate) { 2104 case SDEV_CREATED: 2105 case SDEV_RUNNING: 2106 case SDEV_QUIESCE: 2107 case SDEV_OFFLINE: 2108 case SDEV_BLOCK: 2109 break; 2110 default: 2111 goto illegal; 2112 } 2113 break; 2114 2115 case SDEV_DEL: 2116 switch (oldstate) { 2117 case SDEV_CREATED: 2118 case SDEV_RUNNING: 2119 case SDEV_OFFLINE: 2120 case SDEV_CANCEL: 2121 break; 2122 default: 2123 goto illegal; 2124 } 2125 break; 2126 2127 } 2128 sdev->sdev_state = state; 2129 return 0; 2130 2131 illegal: 2132 SCSI_LOG_ERROR_RECOVERY(1, 2133 sdev_printk(KERN_ERR, sdev, 2134 "Illegal state transition %s->%s\n", 2135 scsi_device_state_name(oldstate), 2136 scsi_device_state_name(state)) 2137 ); 2138 return -EINVAL; 2139 } 2140 EXPORT_SYMBOL(scsi_device_set_state); 2141 2142 /** 2143 * sdev_evt_emit - emit a single SCSI device uevent 2144 * @sdev: associated SCSI device 2145 * @evt: event to emit 2146 * 2147 * Send a single uevent (scsi_event) to the associated scsi_device. 2148 */ 2149 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt) 2150 { 2151 int idx = 0; 2152 char *envp[3]; 2153 2154 switch (evt->evt_type) { 2155 case SDEV_EVT_MEDIA_CHANGE: 2156 envp[idx++] = "SDEV_MEDIA_CHANGE=1"; 2157 break; 2158 2159 default: 2160 /* do nothing */ 2161 break; 2162 } 2163 2164 envp[idx++] = NULL; 2165 2166 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp); 2167 } 2168 2169 /** 2170 * sdev_evt_thread - send a uevent for each scsi event 2171 * @work: work struct for scsi_device 2172 * 2173 * Dispatch queued events to their associated scsi_device kobjects 2174 * as uevents. 2175 */ 2176 void scsi_evt_thread(struct work_struct *work) 2177 { 2178 struct scsi_device *sdev; 2179 LIST_HEAD(event_list); 2180 2181 sdev = container_of(work, struct scsi_device, event_work); 2182 2183 while (1) { 2184 struct scsi_event *evt; 2185 struct list_head *this, *tmp; 2186 unsigned long flags; 2187 2188 spin_lock_irqsave(&sdev->list_lock, flags); 2189 list_splice_init(&sdev->event_list, &event_list); 2190 spin_unlock_irqrestore(&sdev->list_lock, flags); 2191 2192 if (list_empty(&event_list)) 2193 break; 2194 2195 list_for_each_safe(this, tmp, &event_list) { 2196 evt = list_entry(this, struct scsi_event, node); 2197 list_del(&evt->node); 2198 scsi_evt_emit(sdev, evt); 2199 kfree(evt); 2200 } 2201 } 2202 } 2203 2204 /** 2205 * sdev_evt_send - send asserted event to uevent thread 2206 * @sdev: scsi_device event occurred on 2207 * @evt: event to send 2208 * 2209 * Assert scsi device event asynchronously. 2210 */ 2211 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt) 2212 { 2213 unsigned long flags; 2214 2215 #if 0 2216 /* FIXME: currently this check eliminates all media change events 2217 * for polled devices. Need to update to discriminate between AN 2218 * and polled events */ 2219 if (!test_bit(evt->evt_type, sdev->supported_events)) { 2220 kfree(evt); 2221 return; 2222 } 2223 #endif 2224 2225 spin_lock_irqsave(&sdev->list_lock, flags); 2226 list_add_tail(&evt->node, &sdev->event_list); 2227 schedule_work(&sdev->event_work); 2228 spin_unlock_irqrestore(&sdev->list_lock, flags); 2229 } 2230 EXPORT_SYMBOL_GPL(sdev_evt_send); 2231 2232 /** 2233 * sdev_evt_alloc - allocate a new scsi event 2234 * @evt_type: type of event to allocate 2235 * @gfpflags: GFP flags for allocation 2236 * 2237 * Allocates and returns a new scsi_event. 2238 */ 2239 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type, 2240 gfp_t gfpflags) 2241 { 2242 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags); 2243 if (!evt) 2244 return NULL; 2245 2246 evt->evt_type = evt_type; 2247 INIT_LIST_HEAD(&evt->node); 2248 2249 /* evt_type-specific initialization, if any */ 2250 switch (evt_type) { 2251 case SDEV_EVT_MEDIA_CHANGE: 2252 default: 2253 /* do nothing */ 2254 break; 2255 } 2256 2257 return evt; 2258 } 2259 EXPORT_SYMBOL_GPL(sdev_evt_alloc); 2260 2261 /** 2262 * sdev_evt_send_simple - send asserted event to uevent thread 2263 * @sdev: scsi_device event occurred on 2264 * @evt_type: type of event to send 2265 * @gfpflags: GFP flags for allocation 2266 * 2267 * Assert scsi device event asynchronously, given an event type. 2268 */ 2269 void sdev_evt_send_simple(struct scsi_device *sdev, 2270 enum scsi_device_event evt_type, gfp_t gfpflags) 2271 { 2272 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags); 2273 if (!evt) { 2274 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n", 2275 evt_type); 2276 return; 2277 } 2278 2279 sdev_evt_send(sdev, evt); 2280 } 2281 EXPORT_SYMBOL_GPL(sdev_evt_send_simple); 2282 2283 /** 2284 * scsi_device_quiesce - Block user issued commands. 2285 * @sdev: scsi device to quiesce. 2286 * 2287 * This works by trying to transition to the SDEV_QUIESCE state 2288 * (which must be a legal transition). When the device is in this 2289 * state, only special requests will be accepted, all others will 2290 * be deferred. Since special requests may also be requeued requests, 2291 * a successful return doesn't guarantee the device will be 2292 * totally quiescent. 2293 * 2294 * Must be called with user context, may sleep. 2295 * 2296 * Returns zero if unsuccessful or an error if not. 2297 */ 2298 int 2299 scsi_device_quiesce(struct scsi_device *sdev) 2300 { 2301 int err = scsi_device_set_state(sdev, SDEV_QUIESCE); 2302 if (err) 2303 return err; 2304 2305 scsi_run_queue(sdev->request_queue); 2306 while (sdev->device_busy) { 2307 msleep_interruptible(200); 2308 scsi_run_queue(sdev->request_queue); 2309 } 2310 return 0; 2311 } 2312 EXPORT_SYMBOL(scsi_device_quiesce); 2313 2314 /** 2315 * scsi_device_resume - Restart user issued commands to a quiesced device. 2316 * @sdev: scsi device to resume. 2317 * 2318 * Moves the device from quiesced back to running and restarts the 2319 * queues. 2320 * 2321 * Must be called with user context, may sleep. 2322 */ 2323 void 2324 scsi_device_resume(struct scsi_device *sdev) 2325 { 2326 if(scsi_device_set_state(sdev, SDEV_RUNNING)) 2327 return; 2328 scsi_run_queue(sdev->request_queue); 2329 } 2330 EXPORT_SYMBOL(scsi_device_resume); 2331 2332 static void 2333 device_quiesce_fn(struct scsi_device *sdev, void *data) 2334 { 2335 scsi_device_quiesce(sdev); 2336 } 2337 2338 void 2339 scsi_target_quiesce(struct scsi_target *starget) 2340 { 2341 starget_for_each_device(starget, NULL, device_quiesce_fn); 2342 } 2343 EXPORT_SYMBOL(scsi_target_quiesce); 2344 2345 static void 2346 device_resume_fn(struct scsi_device *sdev, void *data) 2347 { 2348 scsi_device_resume(sdev); 2349 } 2350 2351 void 2352 scsi_target_resume(struct scsi_target *starget) 2353 { 2354 starget_for_each_device(starget, NULL, device_resume_fn); 2355 } 2356 EXPORT_SYMBOL(scsi_target_resume); 2357 2358 /** 2359 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state 2360 * @sdev: device to block 2361 * 2362 * Block request made by scsi lld's to temporarily stop all 2363 * scsi commands on the specified device. Called from interrupt 2364 * or normal process context. 2365 * 2366 * Returns zero if successful or error if not 2367 * 2368 * Notes: 2369 * This routine transitions the device to the SDEV_BLOCK state 2370 * (which must be a legal transition). When the device is in this 2371 * state, all commands are deferred until the scsi lld reenables 2372 * the device with scsi_device_unblock or device_block_tmo fires. 2373 * This routine assumes the host_lock is held on entry. 2374 */ 2375 int 2376 scsi_internal_device_block(struct scsi_device *sdev) 2377 { 2378 struct request_queue *q = sdev->request_queue; 2379 unsigned long flags; 2380 int err = 0; 2381 2382 err = scsi_device_set_state(sdev, SDEV_BLOCK); 2383 if (err) { 2384 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK); 2385 2386 if (err) 2387 return err; 2388 } 2389 2390 /* 2391 * The device has transitioned to SDEV_BLOCK. Stop the 2392 * block layer from calling the midlayer with this device's 2393 * request queue. 2394 */ 2395 spin_lock_irqsave(q->queue_lock, flags); 2396 blk_stop_queue(q); 2397 spin_unlock_irqrestore(q->queue_lock, flags); 2398 2399 return 0; 2400 } 2401 EXPORT_SYMBOL_GPL(scsi_internal_device_block); 2402 2403 /** 2404 * scsi_internal_device_unblock - resume a device after a block request 2405 * @sdev: device to resume 2406 * 2407 * Called by scsi lld's or the midlayer to restart the device queue 2408 * for the previously suspended scsi device. Called from interrupt or 2409 * normal process context. 2410 * 2411 * Returns zero if successful or error if not. 2412 * 2413 * Notes: 2414 * This routine transitions the device to the SDEV_RUNNING state 2415 * (which must be a legal transition) allowing the midlayer to 2416 * goose the queue for this device. This routine assumes the 2417 * host_lock is held upon entry. 2418 */ 2419 int 2420 scsi_internal_device_unblock(struct scsi_device *sdev) 2421 { 2422 struct request_queue *q = sdev->request_queue; 2423 unsigned long flags; 2424 2425 /* 2426 * Try to transition the scsi device to SDEV_RUNNING 2427 * and goose the device queue if successful. 2428 */ 2429 if (sdev->sdev_state == SDEV_BLOCK) 2430 sdev->sdev_state = SDEV_RUNNING; 2431 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) 2432 sdev->sdev_state = SDEV_CREATED; 2433 else 2434 return -EINVAL; 2435 2436 spin_lock_irqsave(q->queue_lock, flags); 2437 blk_start_queue(q); 2438 spin_unlock_irqrestore(q->queue_lock, flags); 2439 2440 return 0; 2441 } 2442 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock); 2443 2444 static void 2445 device_block(struct scsi_device *sdev, void *data) 2446 { 2447 scsi_internal_device_block(sdev); 2448 } 2449 2450 static int 2451 target_block(struct device *dev, void *data) 2452 { 2453 if (scsi_is_target_device(dev)) 2454 starget_for_each_device(to_scsi_target(dev), NULL, 2455 device_block); 2456 return 0; 2457 } 2458 2459 void 2460 scsi_target_block(struct device *dev) 2461 { 2462 if (scsi_is_target_device(dev)) 2463 starget_for_each_device(to_scsi_target(dev), NULL, 2464 device_block); 2465 else 2466 device_for_each_child(dev, NULL, target_block); 2467 } 2468 EXPORT_SYMBOL_GPL(scsi_target_block); 2469 2470 static void 2471 device_unblock(struct scsi_device *sdev, void *data) 2472 { 2473 scsi_internal_device_unblock(sdev); 2474 } 2475 2476 static int 2477 target_unblock(struct device *dev, void *data) 2478 { 2479 if (scsi_is_target_device(dev)) 2480 starget_for_each_device(to_scsi_target(dev), NULL, 2481 device_unblock); 2482 return 0; 2483 } 2484 2485 void 2486 scsi_target_unblock(struct device *dev) 2487 { 2488 if (scsi_is_target_device(dev)) 2489 starget_for_each_device(to_scsi_target(dev), NULL, 2490 device_unblock); 2491 else 2492 device_for_each_child(dev, NULL, target_unblock); 2493 } 2494 EXPORT_SYMBOL_GPL(scsi_target_unblock); 2495 2496 /** 2497 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt 2498 * @sgl: scatter-gather list 2499 * @sg_count: number of segments in sg 2500 * @offset: offset in bytes into sg, on return offset into the mapped area 2501 * @len: bytes to map, on return number of bytes mapped 2502 * 2503 * Returns virtual address of the start of the mapped page 2504 */ 2505 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count, 2506 size_t *offset, size_t *len) 2507 { 2508 int i; 2509 size_t sg_len = 0, len_complete = 0; 2510 struct scatterlist *sg; 2511 struct page *page; 2512 2513 WARN_ON(!irqs_disabled()); 2514 2515 for_each_sg(sgl, sg, sg_count, i) { 2516 len_complete = sg_len; /* Complete sg-entries */ 2517 sg_len += sg->length; 2518 if (sg_len > *offset) 2519 break; 2520 } 2521 2522 if (unlikely(i == sg_count)) { 2523 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, " 2524 "elements %d\n", 2525 __func__, sg_len, *offset, sg_count); 2526 WARN_ON(1); 2527 return NULL; 2528 } 2529 2530 /* Offset starting from the beginning of first page in this sg-entry */ 2531 *offset = *offset - len_complete + sg->offset; 2532 2533 /* Assumption: contiguous pages can be accessed as "page + i" */ 2534 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT)); 2535 *offset &= ~PAGE_MASK; 2536 2537 /* Bytes in this sg-entry from *offset to the end of the page */ 2538 sg_len = PAGE_SIZE - *offset; 2539 if (*len > sg_len) 2540 *len = sg_len; 2541 2542 return kmap_atomic(page, KM_BIO_SRC_IRQ); 2543 } 2544 EXPORT_SYMBOL(scsi_kmap_atomic_sg); 2545 2546 /** 2547 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg 2548 * @virt: virtual address to be unmapped 2549 */ 2550 void scsi_kunmap_atomic_sg(void *virt) 2551 { 2552 kunmap_atomic(virt, KM_BIO_SRC_IRQ); 2553 } 2554 EXPORT_SYMBOL(scsi_kunmap_atomic_sg); 2555