scsi_lib.c revision b0790410300abaaf4f25f702803beff701baebf1
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 38struct 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 49static 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 68struct kmem_cache *scsi_sdb_cache; 69 70static 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 */ 84static 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 */ 106static 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 */ 188int 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 */ 208int 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} 254EXPORT_SYMBOL(scsi_execute); 255 256 257int 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} 278EXPORT_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 */ 291static 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 300void 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 */ 325static 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 364static 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 372static 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 379static 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 */ 400static 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 */ 478static 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 491void 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 506void 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 514static 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 */ 538static 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 int leftover = blk_rq_sectors(req) << 9; 550 551 if (blk_pc_request(req)) 552 leftover = req->resid_len; 553 554 /* kill remainder if no retrys */ 555 if (error && scsi_noretry_cmd(cmd)) 556 blk_end_request(req, error, leftover); 557 else { 558 if (requeue) { 559 /* 560 * Bleah. Leftovers again. Stick the 561 * leftovers in the front of the 562 * queue, and goose the queue again. 563 */ 564 scsi_release_buffers(cmd); 565 scsi_requeue_command(q, cmd); 566 cmd = NULL; 567 } 568 return cmd; 569 } 570 } 571 572 /* 573 * This will goose the queue request function at the end, so we don't 574 * need to worry about launching another command. 575 */ 576 __scsi_release_buffers(cmd, 0); 577 scsi_next_command(cmd); 578 return NULL; 579} 580 581static inline unsigned int scsi_sgtable_index(unsigned short nents) 582{ 583 unsigned int index; 584 585 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS); 586 587 if (nents <= 8) 588 index = 0; 589 else 590 index = get_count_order(nents) - 3; 591 592 return index; 593} 594 595static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents) 596{ 597 struct scsi_host_sg_pool *sgp; 598 599 sgp = scsi_sg_pools + scsi_sgtable_index(nents); 600 mempool_free(sgl, sgp->pool); 601} 602 603static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask) 604{ 605 struct scsi_host_sg_pool *sgp; 606 607 sgp = scsi_sg_pools + scsi_sgtable_index(nents); 608 return mempool_alloc(sgp->pool, gfp_mask); 609} 610 611static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, 612 gfp_t gfp_mask) 613{ 614 int ret; 615 616 BUG_ON(!nents); 617 618 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS, 619 gfp_mask, scsi_sg_alloc); 620 if (unlikely(ret)) 621 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, 622 scsi_sg_free); 623 624 return ret; 625} 626 627static void scsi_free_sgtable(struct scsi_data_buffer *sdb) 628{ 629 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free); 630} 631 632static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check) 633{ 634 635 if (cmd->sdb.table.nents) 636 scsi_free_sgtable(&cmd->sdb); 637 638 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 639 640 if (do_bidi_check && scsi_bidi_cmnd(cmd)) { 641 struct scsi_data_buffer *bidi_sdb = 642 cmd->request->next_rq->special; 643 scsi_free_sgtable(bidi_sdb); 644 kmem_cache_free(scsi_sdb_cache, bidi_sdb); 645 cmd->request->next_rq->special = NULL; 646 } 647 648 if (scsi_prot_sg_count(cmd)) 649 scsi_free_sgtable(cmd->prot_sdb); 650} 651 652/* 653 * Function: scsi_release_buffers() 654 * 655 * Purpose: Completion processing for block device I/O requests. 656 * 657 * Arguments: cmd - command that we are bailing. 658 * 659 * Lock status: Assumed that no lock is held upon entry. 660 * 661 * Returns: Nothing 662 * 663 * Notes: In the event that an upper level driver rejects a 664 * command, we must release resources allocated during 665 * the __init_io() function. Primarily this would involve 666 * the scatter-gather table, and potentially any bounce 667 * buffers. 668 */ 669void scsi_release_buffers(struct scsi_cmnd *cmd) 670{ 671 __scsi_release_buffers(cmd, 1); 672} 673EXPORT_SYMBOL(scsi_release_buffers); 674 675/* 676 * Bidi commands Must be complete as a whole, both sides at once. If 677 * part of the bytes were written and lld returned scsi_in()->resid 678 * and/or scsi_out()->resid this information will be left in 679 * req->resid_len and req->next_rq->resid_len. The upper-layer driver 680 * can decide what to do with this information. 681 */ 682static void scsi_end_bidi_request(struct scsi_cmnd *cmd) 683{ 684 struct request *req = cmd->request; 685 686 req->resid_len = scsi_out(cmd)->resid; 687 req->next_rq->resid_len = scsi_in(cmd)->resid; 688 689 /* The req and req->next_rq have not been completed */ 690 BUG_ON(blk_end_bidi_request(req, 0, blk_rq_bytes(req), 691 blk_rq_bytes(req->next_rq))); 692 693 scsi_release_buffers(cmd); 694 695 /* 696 * This will goose the queue request function at the end, so we don't 697 * need to worry about launching another command. 698 */ 699 scsi_next_command(cmd); 700} 701 702/* 703 * Function: scsi_io_completion() 704 * 705 * Purpose: Completion processing for block device I/O requests. 706 * 707 * Arguments: cmd - command that is finished. 708 * 709 * Lock status: Assumed that no lock is held upon entry. 710 * 711 * Returns: Nothing 712 * 713 * Notes: This function is matched in terms of capabilities to 714 * the function that created the scatter-gather list. 715 * In other words, if there are no bounce buffers 716 * (the normal case for most drivers), we don't need 717 * the logic to deal with cleaning up afterwards. 718 * 719 * We must call scsi_end_request(). This will finish off 720 * the specified number of sectors. If we are done, the 721 * command block will be released and the queue function 722 * will be goosed. If we are not done then we have to 723 * figure out what to do next: 724 * 725 * a) We can call scsi_requeue_command(). The request 726 * will be unprepared and put back on the queue. Then 727 * a new command will be created for it. This should 728 * be used if we made forward progress, or if we want 729 * to switch from READ(10) to READ(6) for example. 730 * 731 * b) We can call scsi_queue_insert(). The request will 732 * be put back on the queue and retried using the same 733 * command as before, possibly after a delay. 734 * 735 * c) We can call blk_end_request() with -EIO to fail 736 * the remainder of the request. 737 */ 738void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes) 739{ 740 int result = cmd->result; 741 int this_count; 742 struct request_queue *q = cmd->device->request_queue; 743 struct request *req = cmd->request; 744 int error = 0; 745 struct scsi_sense_hdr sshdr; 746 int sense_valid = 0; 747 int sense_deferred = 0; 748 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY, 749 ACTION_DELAYED_RETRY} action; 750 char *description = NULL; 751 752 if (result) { 753 sense_valid = scsi_command_normalize_sense(cmd, &sshdr); 754 if (sense_valid) 755 sense_deferred = scsi_sense_is_deferred(&sshdr); 756 } 757 758 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */ 759 req->errors = result; 760 if (result) { 761 if (sense_valid && req->sense) { 762 /* 763 * SG_IO wants current and deferred errors 764 */ 765 int len = 8 + cmd->sense_buffer[7]; 766 767 if (len > SCSI_SENSE_BUFFERSIZE) 768 len = SCSI_SENSE_BUFFERSIZE; 769 memcpy(req->sense, cmd->sense_buffer, len); 770 req->sense_len = len; 771 } 772 if (!sense_deferred) 773 error = -EIO; 774 } 775 if (scsi_bidi_cmnd(cmd)) { 776 /* will also release_buffers */ 777 scsi_end_bidi_request(cmd); 778 return; 779 } 780 req->resid_len = scsi_get_resid(cmd); 781 } 782 783 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */ 784 785 /* 786 * Next deal with any sectors which we were able to correctly 787 * handle. 788 */ 789 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, " 790 "%d bytes done.\n", 791 blk_rq_sectors(req), good_bytes)); 792 793 /* 794 * Recovered errors need reporting, but they're always treated 795 * as success, so fiddle the result code here. For BLOCK_PC 796 * we already took a copy of the original into rq->errors which 797 * is what gets returned to the user 798 */ 799 if (sense_valid && sshdr.sense_key == RECOVERED_ERROR) { 800 if (!(req->cmd_flags & REQ_QUIET)) 801 scsi_print_sense("", cmd); 802 result = 0; 803 /* BLOCK_PC may have set error */ 804 error = 0; 805 } 806 807 /* 808 * A number of bytes were successfully read. If there 809 * are leftovers and there is some kind of error 810 * (result != 0), retry the rest. 811 */ 812 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL) 813 return; 814 this_count = blk_rq_bytes(req); 815 816 error = -EIO; 817 818 if (host_byte(result) == DID_RESET) { 819 /* Third party bus reset or reset for error recovery 820 * reasons. Just retry the command and see what 821 * happens. 822 */ 823 action = ACTION_RETRY; 824 } else if (sense_valid && !sense_deferred) { 825 switch (sshdr.sense_key) { 826 case UNIT_ATTENTION: 827 if (cmd->device->removable) { 828 /* Detected disc change. Set a bit 829 * and quietly refuse further access. 830 */ 831 cmd->device->changed = 1; 832 description = "Media Changed"; 833 action = ACTION_FAIL; 834 } else { 835 /* Must have been a power glitch, or a 836 * bus reset. Could not have been a 837 * media change, so we just retry the 838 * command and see what happens. 839 */ 840 action = ACTION_RETRY; 841 } 842 break; 843 case ILLEGAL_REQUEST: 844 /* If we had an ILLEGAL REQUEST returned, then 845 * we may have performed an unsupported 846 * command. The only thing this should be 847 * would be a ten byte read where only a six 848 * byte read was supported. Also, on a system 849 * where READ CAPACITY failed, we may have 850 * read past the end of the disk. 851 */ 852 if ((cmd->device->use_10_for_rw && 853 sshdr.asc == 0x20 && sshdr.ascq == 0x00) && 854 (cmd->cmnd[0] == READ_10 || 855 cmd->cmnd[0] == WRITE_10)) { 856 /* This will issue a new 6-byte command. */ 857 cmd->device->use_10_for_rw = 0; 858 action = ACTION_REPREP; 859 } else if (sshdr.asc == 0x10) /* DIX */ { 860 description = "Host Data Integrity Failure"; 861 action = ACTION_FAIL; 862 error = -EILSEQ; 863 } else 864 action = ACTION_FAIL; 865 break; 866 case ABORTED_COMMAND: 867 action = ACTION_FAIL; 868 if (sshdr.asc == 0x10) { /* DIF */ 869 description = "Target Data Integrity Failure"; 870 error = -EILSEQ; 871 } 872 break; 873 case NOT_READY: 874 /* If the device is in the process of becoming 875 * ready, or has a temporary blockage, retry. 876 */ 877 if (sshdr.asc == 0x04) { 878 switch (sshdr.ascq) { 879 case 0x01: /* becoming ready */ 880 case 0x04: /* format in progress */ 881 case 0x05: /* rebuild in progress */ 882 case 0x06: /* recalculation in progress */ 883 case 0x07: /* operation in progress */ 884 case 0x08: /* Long write in progress */ 885 case 0x09: /* self test in progress */ 886 action = ACTION_DELAYED_RETRY; 887 break; 888 default: 889 description = "Device not ready"; 890 action = ACTION_FAIL; 891 break; 892 } 893 } else { 894 description = "Device not ready"; 895 action = ACTION_FAIL; 896 } 897 break; 898 case VOLUME_OVERFLOW: 899 /* See SSC3rXX or current. */ 900 action = ACTION_FAIL; 901 break; 902 default: 903 description = "Unhandled sense code"; 904 action = ACTION_FAIL; 905 break; 906 } 907 } else { 908 description = "Unhandled error code"; 909 action = ACTION_FAIL; 910 } 911 912 switch (action) { 913 case ACTION_FAIL: 914 /* Give up and fail the remainder of the request */ 915 scsi_release_buffers(cmd); 916 if (!(req->cmd_flags & REQ_QUIET)) { 917 if (description) 918 scmd_printk(KERN_INFO, cmd, "%s\n", 919 description); 920 scsi_print_result(cmd); 921 if (driver_byte(result) & DRIVER_SENSE) 922 scsi_print_sense("", cmd); 923 } 924 blk_end_request_all(req, -EIO); 925 scsi_next_command(cmd); 926 break; 927 case ACTION_REPREP: 928 /* Unprep the request and put it back at the head of the queue. 929 * A new command will be prepared and issued. 930 */ 931 scsi_release_buffers(cmd); 932 scsi_requeue_command(q, cmd); 933 break; 934 case ACTION_RETRY: 935 /* Retry the same command immediately */ 936 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0); 937 break; 938 case ACTION_DELAYED_RETRY: 939 /* Retry the same command after a delay */ 940 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0); 941 break; 942 } 943} 944 945static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb, 946 gfp_t gfp_mask) 947{ 948 int count; 949 950 /* 951 * If sg table allocation fails, requeue request later. 952 */ 953 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments, 954 gfp_mask))) { 955 return BLKPREP_DEFER; 956 } 957 958 req->buffer = NULL; 959 960 /* 961 * Next, walk the list, and fill in the addresses and sizes of 962 * each segment. 963 */ 964 count = blk_rq_map_sg(req->q, req, sdb->table.sgl); 965 BUG_ON(count > sdb->table.nents); 966 sdb->table.nents = count; 967 if (blk_pc_request(req)) 968 sdb->length = blk_rq_bytes(req); 969 else 970 sdb->length = blk_rq_sectors(req) << 9; 971 return BLKPREP_OK; 972} 973 974/* 975 * Function: scsi_init_io() 976 * 977 * Purpose: SCSI I/O initialize function. 978 * 979 * Arguments: cmd - Command descriptor we wish to initialize 980 * 981 * Returns: 0 on success 982 * BLKPREP_DEFER if the failure is retryable 983 * BLKPREP_KILL if the failure is fatal 984 */ 985int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask) 986{ 987 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask); 988 if (error) 989 goto err_exit; 990 991 if (blk_bidi_rq(cmd->request)) { 992 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc( 993 scsi_sdb_cache, GFP_ATOMIC); 994 if (!bidi_sdb) { 995 error = BLKPREP_DEFER; 996 goto err_exit; 997 } 998 999 cmd->request->next_rq->special = bidi_sdb; 1000 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb, 1001 GFP_ATOMIC); 1002 if (error) 1003 goto err_exit; 1004 } 1005 1006 if (blk_integrity_rq(cmd->request)) { 1007 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb; 1008 int ivecs, count; 1009 1010 BUG_ON(prot_sdb == NULL); 1011 ivecs = blk_rq_count_integrity_sg(cmd->request); 1012 1013 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) { 1014 error = BLKPREP_DEFER; 1015 goto err_exit; 1016 } 1017 1018 count = blk_rq_map_integrity_sg(cmd->request, 1019 prot_sdb->table.sgl); 1020 BUG_ON(unlikely(count > ivecs)); 1021 1022 cmd->prot_sdb = prot_sdb; 1023 cmd->prot_sdb->table.nents = count; 1024 } 1025 1026 return BLKPREP_OK ; 1027 1028err_exit: 1029 scsi_release_buffers(cmd); 1030 if (error == BLKPREP_KILL) 1031 scsi_put_command(cmd); 1032 else /* BLKPREP_DEFER */ 1033 scsi_unprep_request(cmd->request); 1034 1035 return error; 1036} 1037EXPORT_SYMBOL(scsi_init_io); 1038 1039static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev, 1040 struct request *req) 1041{ 1042 struct scsi_cmnd *cmd; 1043 1044 if (!req->special) { 1045 cmd = scsi_get_command(sdev, GFP_ATOMIC); 1046 if (unlikely(!cmd)) 1047 return NULL; 1048 req->special = cmd; 1049 } else { 1050 cmd = req->special; 1051 } 1052 1053 /* pull a tag out of the request if we have one */ 1054 cmd->tag = req->tag; 1055 cmd->request = req; 1056 1057 cmd->cmnd = req->cmd; 1058 1059 return cmd; 1060} 1061 1062int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req) 1063{ 1064 struct scsi_cmnd *cmd; 1065 int ret = scsi_prep_state_check(sdev, req); 1066 1067 if (ret != BLKPREP_OK) 1068 return ret; 1069 1070 cmd = scsi_get_cmd_from_req(sdev, req); 1071 if (unlikely(!cmd)) 1072 return BLKPREP_DEFER; 1073 1074 /* 1075 * BLOCK_PC requests may transfer data, in which case they must 1076 * a bio attached to them. Or they might contain a SCSI command 1077 * that does not transfer data, in which case they may optionally 1078 * submit a request without an attached bio. 1079 */ 1080 if (req->bio) { 1081 int ret; 1082 1083 BUG_ON(!req->nr_phys_segments); 1084 1085 ret = scsi_init_io(cmd, GFP_ATOMIC); 1086 if (unlikely(ret)) 1087 return ret; 1088 } else { 1089 BUG_ON(blk_rq_bytes(req)); 1090 1091 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1092 req->buffer = NULL; 1093 } 1094 1095 cmd->cmd_len = req->cmd_len; 1096 if (!blk_rq_bytes(req)) 1097 cmd->sc_data_direction = DMA_NONE; 1098 else if (rq_data_dir(req) == WRITE) 1099 cmd->sc_data_direction = DMA_TO_DEVICE; 1100 else 1101 cmd->sc_data_direction = DMA_FROM_DEVICE; 1102 1103 cmd->transfersize = blk_rq_bytes(req); 1104 cmd->allowed = req->retries; 1105 return BLKPREP_OK; 1106} 1107EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd); 1108 1109/* 1110 * Setup a REQ_TYPE_FS command. These are simple read/write request 1111 * from filesystems that still need to be translated to SCSI CDBs from 1112 * the ULD. 1113 */ 1114int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req) 1115{ 1116 struct scsi_cmnd *cmd; 1117 int ret = scsi_prep_state_check(sdev, req); 1118 1119 if (ret != BLKPREP_OK) 1120 return ret; 1121 1122 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh 1123 && sdev->scsi_dh_data->scsi_dh->prep_fn)) { 1124 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req); 1125 if (ret != BLKPREP_OK) 1126 return ret; 1127 } 1128 1129 /* 1130 * Filesystem requests must transfer data. 1131 */ 1132 BUG_ON(!req->nr_phys_segments); 1133 1134 cmd = scsi_get_cmd_from_req(sdev, req); 1135 if (unlikely(!cmd)) 1136 return BLKPREP_DEFER; 1137 1138 memset(cmd->cmnd, 0, BLK_MAX_CDB); 1139 return scsi_init_io(cmd, GFP_ATOMIC); 1140} 1141EXPORT_SYMBOL(scsi_setup_fs_cmnd); 1142 1143int scsi_prep_state_check(struct scsi_device *sdev, struct request *req) 1144{ 1145 int ret = BLKPREP_OK; 1146 1147 /* 1148 * If the device is not in running state we will reject some 1149 * or all commands. 1150 */ 1151 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) { 1152 switch (sdev->sdev_state) { 1153 case SDEV_OFFLINE: 1154 /* 1155 * If the device is offline we refuse to process any 1156 * commands. The device must be brought online 1157 * before trying any recovery commands. 1158 */ 1159 sdev_printk(KERN_ERR, sdev, 1160 "rejecting I/O to offline device\n"); 1161 ret = BLKPREP_KILL; 1162 break; 1163 case SDEV_DEL: 1164 /* 1165 * If the device is fully deleted, we refuse to 1166 * process any commands as well. 1167 */ 1168 sdev_printk(KERN_ERR, sdev, 1169 "rejecting I/O to dead device\n"); 1170 ret = BLKPREP_KILL; 1171 break; 1172 case SDEV_QUIESCE: 1173 case SDEV_BLOCK: 1174 case SDEV_CREATED_BLOCK: 1175 /* 1176 * If the devices is blocked we defer normal commands. 1177 */ 1178 if (!(req->cmd_flags & REQ_PREEMPT)) 1179 ret = BLKPREP_DEFER; 1180 break; 1181 default: 1182 /* 1183 * For any other not fully online state we only allow 1184 * special commands. In particular any user initiated 1185 * command is not allowed. 1186 */ 1187 if (!(req->cmd_flags & REQ_PREEMPT)) 1188 ret = BLKPREP_KILL; 1189 break; 1190 } 1191 } 1192 return ret; 1193} 1194EXPORT_SYMBOL(scsi_prep_state_check); 1195 1196int scsi_prep_return(struct request_queue *q, struct request *req, int ret) 1197{ 1198 struct scsi_device *sdev = q->queuedata; 1199 1200 switch (ret) { 1201 case BLKPREP_KILL: 1202 req->errors = DID_NO_CONNECT << 16; 1203 /* release the command and kill it */ 1204 if (req->special) { 1205 struct scsi_cmnd *cmd = req->special; 1206 scsi_release_buffers(cmd); 1207 scsi_put_command(cmd); 1208 req->special = NULL; 1209 } 1210 break; 1211 case BLKPREP_DEFER: 1212 /* 1213 * If we defer, the elv_next_request() returns NULL, but the 1214 * queue must be restarted, so we plug here if no returning 1215 * command will automatically do that. 1216 */ 1217 if (sdev->device_busy == 0) 1218 blk_plug_device(q); 1219 break; 1220 default: 1221 req->cmd_flags |= REQ_DONTPREP; 1222 } 1223 1224 return ret; 1225} 1226EXPORT_SYMBOL(scsi_prep_return); 1227 1228int scsi_prep_fn(struct request_queue *q, struct request *req) 1229{ 1230 struct scsi_device *sdev = q->queuedata; 1231 int ret = BLKPREP_KILL; 1232 1233 if (req->cmd_type == REQ_TYPE_BLOCK_PC) 1234 ret = scsi_setup_blk_pc_cmnd(sdev, req); 1235 return scsi_prep_return(q, req, ret); 1236} 1237 1238/* 1239 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else 1240 * return 0. 1241 * 1242 * Called with the queue_lock held. 1243 */ 1244static inline int scsi_dev_queue_ready(struct request_queue *q, 1245 struct scsi_device *sdev) 1246{ 1247 if (sdev->device_busy == 0 && sdev->device_blocked) { 1248 /* 1249 * unblock after device_blocked iterates to zero 1250 */ 1251 if (--sdev->device_blocked == 0) { 1252 SCSI_LOG_MLQUEUE(3, 1253 sdev_printk(KERN_INFO, sdev, 1254 "unblocking device at zero depth\n")); 1255 } else { 1256 blk_plug_device(q); 1257 return 0; 1258 } 1259 } 1260 if (scsi_device_is_busy(sdev)) 1261 return 0; 1262 1263 return 1; 1264} 1265 1266 1267/* 1268 * scsi_target_queue_ready: checks if there we can send commands to target 1269 * @sdev: scsi device on starget to check. 1270 * 1271 * Called with the host lock held. 1272 */ 1273static inline int scsi_target_queue_ready(struct Scsi_Host *shost, 1274 struct scsi_device *sdev) 1275{ 1276 struct scsi_target *starget = scsi_target(sdev); 1277 1278 if (starget->single_lun) { 1279 if (starget->starget_sdev_user && 1280 starget->starget_sdev_user != sdev) 1281 return 0; 1282 starget->starget_sdev_user = sdev; 1283 } 1284 1285 if (starget->target_busy == 0 && starget->target_blocked) { 1286 /* 1287 * unblock after target_blocked iterates to zero 1288 */ 1289 if (--starget->target_blocked == 0) { 1290 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget, 1291 "unblocking target at zero depth\n")); 1292 } else { 1293 blk_plug_device(sdev->request_queue); 1294 return 0; 1295 } 1296 } 1297 1298 if (scsi_target_is_busy(starget)) { 1299 if (list_empty(&sdev->starved_entry)) { 1300 list_add_tail(&sdev->starved_entry, 1301 &shost->starved_list); 1302 return 0; 1303 } 1304 } 1305 1306 /* We're OK to process the command, so we can't be starved */ 1307 if (!list_empty(&sdev->starved_entry)) 1308 list_del_init(&sdev->starved_entry); 1309 return 1; 1310} 1311 1312/* 1313 * scsi_host_queue_ready: if we can send requests to shost, return 1 else 1314 * return 0. We must end up running the queue again whenever 0 is 1315 * returned, else IO can hang. 1316 * 1317 * Called with host_lock held. 1318 */ 1319static inline int scsi_host_queue_ready(struct request_queue *q, 1320 struct Scsi_Host *shost, 1321 struct scsi_device *sdev) 1322{ 1323 if (scsi_host_in_recovery(shost)) 1324 return 0; 1325 if (shost->host_busy == 0 && shost->host_blocked) { 1326 /* 1327 * unblock after host_blocked iterates to zero 1328 */ 1329 if (--shost->host_blocked == 0) { 1330 SCSI_LOG_MLQUEUE(3, 1331 printk("scsi%d unblocking host at zero depth\n", 1332 shost->host_no)); 1333 } else { 1334 return 0; 1335 } 1336 } 1337 if (scsi_host_is_busy(shost)) { 1338 if (list_empty(&sdev->starved_entry)) 1339 list_add_tail(&sdev->starved_entry, &shost->starved_list); 1340 return 0; 1341 } 1342 1343 /* We're OK to process the command, so we can't be starved */ 1344 if (!list_empty(&sdev->starved_entry)) 1345 list_del_init(&sdev->starved_entry); 1346 1347 return 1; 1348} 1349 1350/* 1351 * Busy state exporting function for request stacking drivers. 1352 * 1353 * For efficiency, no lock is taken to check the busy state of 1354 * shost/starget/sdev, since the returned value is not guaranteed and 1355 * may be changed after request stacking drivers call the function, 1356 * regardless of taking lock or not. 1357 * 1358 * When scsi can't dispatch I/Os anymore and needs to kill I/Os 1359 * (e.g. !sdev), scsi needs to return 'not busy'. 1360 * Otherwise, request stacking drivers may hold requests forever. 1361 */ 1362static int scsi_lld_busy(struct request_queue *q) 1363{ 1364 struct scsi_device *sdev = q->queuedata; 1365 struct Scsi_Host *shost; 1366 struct scsi_target *starget; 1367 1368 if (!sdev) 1369 return 0; 1370 1371 shost = sdev->host; 1372 starget = scsi_target(sdev); 1373 1374 if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) || 1375 scsi_target_is_busy(starget) || scsi_device_is_busy(sdev)) 1376 return 1; 1377 1378 return 0; 1379} 1380 1381/* 1382 * Kill a request for a dead device 1383 */ 1384static void scsi_kill_request(struct request *req, struct request_queue *q) 1385{ 1386 struct scsi_cmnd *cmd = req->special; 1387 struct scsi_device *sdev = cmd->device; 1388 struct scsi_target *starget = scsi_target(sdev); 1389 struct Scsi_Host *shost = sdev->host; 1390 1391 blkdev_dequeue_request(req); 1392 1393 if (unlikely(cmd == NULL)) { 1394 printk(KERN_CRIT "impossible request in %s.\n", 1395 __func__); 1396 BUG(); 1397 } 1398 1399 scsi_init_cmd_errh(cmd); 1400 cmd->result = DID_NO_CONNECT << 16; 1401 atomic_inc(&cmd->device->iorequest_cnt); 1402 1403 /* 1404 * SCSI request completion path will do scsi_device_unbusy(), 1405 * bump busy counts. To bump the counters, we need to dance 1406 * with the locks as normal issue path does. 1407 */ 1408 sdev->device_busy++; 1409 spin_unlock(sdev->request_queue->queue_lock); 1410 spin_lock(shost->host_lock); 1411 shost->host_busy++; 1412 starget->target_busy++; 1413 spin_unlock(shost->host_lock); 1414 spin_lock(sdev->request_queue->queue_lock); 1415 1416 blk_complete_request(req); 1417} 1418 1419static void scsi_softirq_done(struct request *rq) 1420{ 1421 struct scsi_cmnd *cmd = rq->special; 1422 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout; 1423 int disposition; 1424 1425 INIT_LIST_HEAD(&cmd->eh_entry); 1426 1427 /* 1428 * Set the serial numbers back to zero 1429 */ 1430 cmd->serial_number = 0; 1431 1432 atomic_inc(&cmd->device->iodone_cnt); 1433 if (cmd->result) 1434 atomic_inc(&cmd->device->ioerr_cnt); 1435 1436 disposition = scsi_decide_disposition(cmd); 1437 if (disposition != SUCCESS && 1438 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) { 1439 sdev_printk(KERN_ERR, cmd->device, 1440 "timing out command, waited %lus\n", 1441 wait_for/HZ); 1442 disposition = SUCCESS; 1443 } 1444 1445 scsi_log_completion(cmd, disposition); 1446 1447 switch (disposition) { 1448 case SUCCESS: 1449 scsi_finish_command(cmd); 1450 break; 1451 case NEEDS_RETRY: 1452 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY); 1453 break; 1454 case ADD_TO_MLQUEUE: 1455 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY); 1456 break; 1457 default: 1458 if (!scsi_eh_scmd_add(cmd, 0)) 1459 scsi_finish_command(cmd); 1460 } 1461} 1462 1463/* 1464 * Function: scsi_request_fn() 1465 * 1466 * Purpose: Main strategy routine for SCSI. 1467 * 1468 * Arguments: q - Pointer to actual queue. 1469 * 1470 * Returns: Nothing 1471 * 1472 * Lock status: IO request lock assumed to be held when called. 1473 */ 1474static void scsi_request_fn(struct request_queue *q) 1475{ 1476 struct scsi_device *sdev = q->queuedata; 1477 struct Scsi_Host *shost; 1478 struct scsi_cmnd *cmd; 1479 struct request *req; 1480 1481 if (!sdev) { 1482 printk("scsi: killing requests for dead queue\n"); 1483 while ((req = elv_next_request(q)) != NULL) 1484 scsi_kill_request(req, q); 1485 return; 1486 } 1487 1488 if(!get_device(&sdev->sdev_gendev)) 1489 /* We must be tearing the block queue down already */ 1490 return; 1491 1492 /* 1493 * To start with, we keep looping until the queue is empty, or until 1494 * the host is no longer able to accept any more requests. 1495 */ 1496 shost = sdev->host; 1497 while (!blk_queue_plugged(q)) { 1498 int rtn; 1499 /* 1500 * get next queueable request. We do this early to make sure 1501 * that the request is fully prepared even if we cannot 1502 * accept it. 1503 */ 1504 req = elv_next_request(q); 1505 if (!req || !scsi_dev_queue_ready(q, sdev)) 1506 break; 1507 1508 if (unlikely(!scsi_device_online(sdev))) { 1509 sdev_printk(KERN_ERR, sdev, 1510 "rejecting I/O to offline device\n"); 1511 scsi_kill_request(req, q); 1512 continue; 1513 } 1514 1515 1516 /* 1517 * Remove the request from the request list. 1518 */ 1519 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req))) 1520 blkdev_dequeue_request(req); 1521 sdev->device_busy++; 1522 1523 spin_unlock(q->queue_lock); 1524 cmd = req->special; 1525 if (unlikely(cmd == NULL)) { 1526 printk(KERN_CRIT "impossible request in %s.\n" 1527 "please mail a stack trace to " 1528 "linux-scsi@vger.kernel.org\n", 1529 __func__); 1530 blk_dump_rq_flags(req, "foo"); 1531 BUG(); 1532 } 1533 spin_lock(shost->host_lock); 1534 1535 /* 1536 * We hit this when the driver is using a host wide 1537 * tag map. For device level tag maps the queue_depth check 1538 * in the device ready fn would prevent us from trying 1539 * to allocate a tag. Since the map is a shared host resource 1540 * we add the dev to the starved list so it eventually gets 1541 * a run when a tag is freed. 1542 */ 1543 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) { 1544 if (list_empty(&sdev->starved_entry)) 1545 list_add_tail(&sdev->starved_entry, 1546 &shost->starved_list); 1547 goto not_ready; 1548 } 1549 1550 if (!scsi_target_queue_ready(shost, sdev)) 1551 goto not_ready; 1552 1553 if (!scsi_host_queue_ready(q, shost, sdev)) 1554 goto not_ready; 1555 1556 scsi_target(sdev)->target_busy++; 1557 shost->host_busy++; 1558 1559 /* 1560 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will 1561 * take the lock again. 1562 */ 1563 spin_unlock_irq(shost->host_lock); 1564 1565 /* 1566 * Finally, initialize any error handling parameters, and set up 1567 * the timers for timeouts. 1568 */ 1569 scsi_init_cmd_errh(cmd); 1570 1571 /* 1572 * Dispatch the command to the low-level driver. 1573 */ 1574 rtn = scsi_dispatch_cmd(cmd); 1575 spin_lock_irq(q->queue_lock); 1576 if(rtn) { 1577 /* we're refusing the command; because of 1578 * the way locks get dropped, we need to 1579 * check here if plugging is required */ 1580 if(sdev->device_busy == 0) 1581 blk_plug_device(q); 1582 1583 break; 1584 } 1585 } 1586 1587 goto out; 1588 1589 not_ready: 1590 spin_unlock_irq(shost->host_lock); 1591 1592 /* 1593 * lock q, handle tag, requeue req, and decrement device_busy. We 1594 * must return with queue_lock held. 1595 * 1596 * Decrementing device_busy without checking it is OK, as all such 1597 * cases (host limits or settings) should run the queue at some 1598 * later time. 1599 */ 1600 spin_lock_irq(q->queue_lock); 1601 blk_requeue_request(q, req); 1602 sdev->device_busy--; 1603 if(sdev->device_busy == 0) 1604 blk_plug_device(q); 1605 out: 1606 /* must be careful here...if we trigger the ->remove() function 1607 * we cannot be holding the q lock */ 1608 spin_unlock_irq(q->queue_lock); 1609 put_device(&sdev->sdev_gendev); 1610 spin_lock_irq(q->queue_lock); 1611} 1612 1613u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost) 1614{ 1615 struct device *host_dev; 1616 u64 bounce_limit = 0xffffffff; 1617 1618 if (shost->unchecked_isa_dma) 1619 return BLK_BOUNCE_ISA; 1620 /* 1621 * Platforms with virtual-DMA translation 1622 * hardware have no practical limit. 1623 */ 1624 if (!PCI_DMA_BUS_IS_PHYS) 1625 return BLK_BOUNCE_ANY; 1626 1627 host_dev = scsi_get_device(shost); 1628 if (host_dev && host_dev->dma_mask) 1629 bounce_limit = *host_dev->dma_mask; 1630 1631 return bounce_limit; 1632} 1633EXPORT_SYMBOL(scsi_calculate_bounce_limit); 1634 1635struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost, 1636 request_fn_proc *request_fn) 1637{ 1638 struct request_queue *q; 1639 struct device *dev = shost->shost_gendev.parent; 1640 1641 q = blk_init_queue(request_fn, NULL); 1642 if (!q) 1643 return NULL; 1644 1645 /* 1646 * this limit is imposed by hardware restrictions 1647 */ 1648 blk_queue_max_hw_segments(q, shost->sg_tablesize); 1649 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS); 1650 1651 blk_queue_max_sectors(q, shost->max_sectors); 1652 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost)); 1653 blk_queue_segment_boundary(q, shost->dma_boundary); 1654 dma_set_seg_boundary(dev, shost->dma_boundary); 1655 1656 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev)); 1657 1658 /* New queue, no concurrency on queue_flags */ 1659 if (!shost->use_clustering) 1660 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q); 1661 1662 /* 1663 * set a reasonable default alignment on word boundaries: the 1664 * host and device may alter it using 1665 * blk_queue_update_dma_alignment() later. 1666 */ 1667 blk_queue_dma_alignment(q, 0x03); 1668 1669 return q; 1670} 1671EXPORT_SYMBOL(__scsi_alloc_queue); 1672 1673struct request_queue *scsi_alloc_queue(struct scsi_device *sdev) 1674{ 1675 struct request_queue *q; 1676 1677 q = __scsi_alloc_queue(sdev->host, scsi_request_fn); 1678 if (!q) 1679 return NULL; 1680 1681 blk_queue_prep_rq(q, scsi_prep_fn); 1682 blk_queue_softirq_done(q, scsi_softirq_done); 1683 blk_queue_rq_timed_out(q, scsi_times_out); 1684 blk_queue_lld_busy(q, scsi_lld_busy); 1685 return q; 1686} 1687 1688void scsi_free_queue(struct request_queue *q) 1689{ 1690 blk_cleanup_queue(q); 1691} 1692 1693/* 1694 * Function: scsi_block_requests() 1695 * 1696 * Purpose: Utility function used by low-level drivers to prevent further 1697 * commands from being queued to the device. 1698 * 1699 * Arguments: shost - Host in question 1700 * 1701 * Returns: Nothing 1702 * 1703 * Lock status: No locks are assumed held. 1704 * 1705 * Notes: There is no timer nor any other means by which the requests 1706 * get unblocked other than the low-level driver calling 1707 * scsi_unblock_requests(). 1708 */ 1709void scsi_block_requests(struct Scsi_Host *shost) 1710{ 1711 shost->host_self_blocked = 1; 1712} 1713EXPORT_SYMBOL(scsi_block_requests); 1714 1715/* 1716 * Function: scsi_unblock_requests() 1717 * 1718 * Purpose: Utility function used by low-level drivers to allow further 1719 * commands from being queued to the device. 1720 * 1721 * Arguments: shost - Host in question 1722 * 1723 * Returns: Nothing 1724 * 1725 * Lock status: No locks are assumed held. 1726 * 1727 * Notes: There is no timer nor any other means by which the requests 1728 * get unblocked other than the low-level driver calling 1729 * scsi_unblock_requests(). 1730 * 1731 * This is done as an API function so that changes to the 1732 * internals of the scsi mid-layer won't require wholesale 1733 * changes to drivers that use this feature. 1734 */ 1735void scsi_unblock_requests(struct Scsi_Host *shost) 1736{ 1737 shost->host_self_blocked = 0; 1738 scsi_run_host_queues(shost); 1739} 1740EXPORT_SYMBOL(scsi_unblock_requests); 1741 1742int __init scsi_init_queue(void) 1743{ 1744 int i; 1745 1746 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer", 1747 sizeof(struct scsi_data_buffer), 1748 0, 0, NULL); 1749 if (!scsi_sdb_cache) { 1750 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n"); 1751 return -ENOMEM; 1752 } 1753 1754 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1755 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1756 int size = sgp->size * sizeof(struct scatterlist); 1757 1758 sgp->slab = kmem_cache_create(sgp->name, size, 0, 1759 SLAB_HWCACHE_ALIGN, NULL); 1760 if (!sgp->slab) { 1761 printk(KERN_ERR "SCSI: can't init sg slab %s\n", 1762 sgp->name); 1763 goto cleanup_sdb; 1764 } 1765 1766 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE, 1767 sgp->slab); 1768 if (!sgp->pool) { 1769 printk(KERN_ERR "SCSI: can't init sg mempool %s\n", 1770 sgp->name); 1771 goto cleanup_sdb; 1772 } 1773 } 1774 1775 return 0; 1776 1777cleanup_sdb: 1778 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1779 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1780 if (sgp->pool) 1781 mempool_destroy(sgp->pool); 1782 if (sgp->slab) 1783 kmem_cache_destroy(sgp->slab); 1784 } 1785 kmem_cache_destroy(scsi_sdb_cache); 1786 1787 return -ENOMEM; 1788} 1789 1790void scsi_exit_queue(void) 1791{ 1792 int i; 1793 1794 kmem_cache_destroy(scsi_sdb_cache); 1795 1796 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1797 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1798 mempool_destroy(sgp->pool); 1799 kmem_cache_destroy(sgp->slab); 1800 } 1801} 1802 1803/** 1804 * scsi_mode_select - issue a mode select 1805 * @sdev: SCSI device to be queried 1806 * @pf: Page format bit (1 == standard, 0 == vendor specific) 1807 * @sp: Save page bit (0 == don't save, 1 == save) 1808 * @modepage: mode page being requested 1809 * @buffer: request buffer (may not be smaller than eight bytes) 1810 * @len: length of request buffer. 1811 * @timeout: command timeout 1812 * @retries: number of retries before failing 1813 * @data: returns a structure abstracting the mode header data 1814 * @sshdr: place to put sense data (or NULL if no sense to be collected). 1815 * must be SCSI_SENSE_BUFFERSIZE big. 1816 * 1817 * Returns zero if successful; negative error number or scsi 1818 * status on error 1819 * 1820 */ 1821int 1822scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage, 1823 unsigned char *buffer, int len, int timeout, int retries, 1824 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 1825{ 1826 unsigned char cmd[10]; 1827 unsigned char *real_buffer; 1828 int ret; 1829 1830 memset(cmd, 0, sizeof(cmd)); 1831 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0); 1832 1833 if (sdev->use_10_for_ms) { 1834 if (len > 65535) 1835 return -EINVAL; 1836 real_buffer = kmalloc(8 + len, GFP_KERNEL); 1837 if (!real_buffer) 1838 return -ENOMEM; 1839 memcpy(real_buffer + 8, buffer, len); 1840 len += 8; 1841 real_buffer[0] = 0; 1842 real_buffer[1] = 0; 1843 real_buffer[2] = data->medium_type; 1844 real_buffer[3] = data->device_specific; 1845 real_buffer[4] = data->longlba ? 0x01 : 0; 1846 real_buffer[5] = 0; 1847 real_buffer[6] = data->block_descriptor_length >> 8; 1848 real_buffer[7] = data->block_descriptor_length; 1849 1850 cmd[0] = MODE_SELECT_10; 1851 cmd[7] = len >> 8; 1852 cmd[8] = len; 1853 } else { 1854 if (len > 255 || data->block_descriptor_length > 255 || 1855 data->longlba) 1856 return -EINVAL; 1857 1858 real_buffer = kmalloc(4 + len, GFP_KERNEL); 1859 if (!real_buffer) 1860 return -ENOMEM; 1861 memcpy(real_buffer + 4, buffer, len); 1862 len += 4; 1863 real_buffer[0] = 0; 1864 real_buffer[1] = data->medium_type; 1865 real_buffer[2] = data->device_specific; 1866 real_buffer[3] = data->block_descriptor_length; 1867 1868 1869 cmd[0] = MODE_SELECT; 1870 cmd[4] = len; 1871 } 1872 1873 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len, 1874 sshdr, timeout, retries, NULL); 1875 kfree(real_buffer); 1876 return ret; 1877} 1878EXPORT_SYMBOL_GPL(scsi_mode_select); 1879 1880/** 1881 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary. 1882 * @sdev: SCSI device to be queried 1883 * @dbd: set if mode sense will allow block descriptors to be returned 1884 * @modepage: mode page being requested 1885 * @buffer: request buffer (may not be smaller than eight bytes) 1886 * @len: length of request buffer. 1887 * @timeout: command timeout 1888 * @retries: number of retries before failing 1889 * @data: returns a structure abstracting the mode header data 1890 * @sshdr: place to put sense data (or NULL if no sense to be collected). 1891 * must be SCSI_SENSE_BUFFERSIZE big. 1892 * 1893 * Returns zero if unsuccessful, or the header offset (either 4 1894 * or 8 depending on whether a six or ten byte command was 1895 * issued) if successful. 1896 */ 1897int 1898scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, 1899 unsigned char *buffer, int len, int timeout, int retries, 1900 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 1901{ 1902 unsigned char cmd[12]; 1903 int use_10_for_ms; 1904 int header_length; 1905 int result; 1906 struct scsi_sense_hdr my_sshdr; 1907 1908 memset(data, 0, sizeof(*data)); 1909 memset(&cmd[0], 0, 12); 1910 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */ 1911 cmd[2] = modepage; 1912 1913 /* caller might not be interested in sense, but we need it */ 1914 if (!sshdr) 1915 sshdr = &my_sshdr; 1916 1917 retry: 1918 use_10_for_ms = sdev->use_10_for_ms; 1919 1920 if (use_10_for_ms) { 1921 if (len < 8) 1922 len = 8; 1923 1924 cmd[0] = MODE_SENSE_10; 1925 cmd[8] = len; 1926 header_length = 8; 1927 } else { 1928 if (len < 4) 1929 len = 4; 1930 1931 cmd[0] = MODE_SENSE; 1932 cmd[4] = len; 1933 header_length = 4; 1934 } 1935 1936 memset(buffer, 0, len); 1937 1938 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len, 1939 sshdr, timeout, retries, NULL); 1940 1941 /* This code looks awful: what it's doing is making sure an 1942 * ILLEGAL REQUEST sense return identifies the actual command 1943 * byte as the problem. MODE_SENSE commands can return 1944 * ILLEGAL REQUEST if the code page isn't supported */ 1945 1946 if (use_10_for_ms && !scsi_status_is_good(result) && 1947 (driver_byte(result) & DRIVER_SENSE)) { 1948 if (scsi_sense_valid(sshdr)) { 1949 if ((sshdr->sense_key == ILLEGAL_REQUEST) && 1950 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) { 1951 /* 1952 * Invalid command operation code 1953 */ 1954 sdev->use_10_for_ms = 0; 1955 goto retry; 1956 } 1957 } 1958 } 1959 1960 if(scsi_status_is_good(result)) { 1961 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b && 1962 (modepage == 6 || modepage == 8))) { 1963 /* Initio breakage? */ 1964 header_length = 0; 1965 data->length = 13; 1966 data->medium_type = 0; 1967 data->device_specific = 0; 1968 data->longlba = 0; 1969 data->block_descriptor_length = 0; 1970 } else if(use_10_for_ms) { 1971 data->length = buffer[0]*256 + buffer[1] + 2; 1972 data->medium_type = buffer[2]; 1973 data->device_specific = buffer[3]; 1974 data->longlba = buffer[4] & 0x01; 1975 data->block_descriptor_length = buffer[6]*256 1976 + buffer[7]; 1977 } else { 1978 data->length = buffer[0] + 1; 1979 data->medium_type = buffer[1]; 1980 data->device_specific = buffer[2]; 1981 data->block_descriptor_length = buffer[3]; 1982 } 1983 data->header_length = header_length; 1984 } 1985 1986 return result; 1987} 1988EXPORT_SYMBOL(scsi_mode_sense); 1989 1990/** 1991 * scsi_test_unit_ready - test if unit is ready 1992 * @sdev: scsi device to change the state of. 1993 * @timeout: command timeout 1994 * @retries: number of retries before failing 1995 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for 1996 * returning sense. Make sure that this is cleared before passing 1997 * in. 1998 * 1999 * Returns zero if unsuccessful or an error if TUR failed. For 2000 * removable media, a return of NOT_READY or UNIT_ATTENTION is 2001 * translated to success, with the ->changed flag updated. 2002 **/ 2003int 2004scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries, 2005 struct scsi_sense_hdr *sshdr_external) 2006{ 2007 char cmd[] = { 2008 TEST_UNIT_READY, 0, 0, 0, 0, 0, 2009 }; 2010 struct scsi_sense_hdr *sshdr; 2011 int result; 2012 2013 if (!sshdr_external) 2014 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL); 2015 else 2016 sshdr = sshdr_external; 2017 2018 /* try to eat the UNIT_ATTENTION if there are enough retries */ 2019 do { 2020 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr, 2021 timeout, retries, NULL); 2022 if (sdev->removable && scsi_sense_valid(sshdr) && 2023 sshdr->sense_key == UNIT_ATTENTION) 2024 sdev->changed = 1; 2025 } while (scsi_sense_valid(sshdr) && 2026 sshdr->sense_key == UNIT_ATTENTION && --retries); 2027 2028 if (!sshdr) 2029 /* could not allocate sense buffer, so can't process it */ 2030 return result; 2031 2032 if (sdev->removable && scsi_sense_valid(sshdr) && 2033 (sshdr->sense_key == UNIT_ATTENTION || 2034 sshdr->sense_key == NOT_READY)) { 2035 sdev->changed = 1; 2036 result = 0; 2037 } 2038 if (!sshdr_external) 2039 kfree(sshdr); 2040 return result; 2041} 2042EXPORT_SYMBOL(scsi_test_unit_ready); 2043 2044/** 2045 * scsi_device_set_state - Take the given device through the device state model. 2046 * @sdev: scsi device to change the state of. 2047 * @state: state to change to. 2048 * 2049 * Returns zero if unsuccessful or an error if the requested 2050 * transition is illegal. 2051 */ 2052int 2053scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state) 2054{ 2055 enum scsi_device_state oldstate = sdev->sdev_state; 2056 2057 if (state == oldstate) 2058 return 0; 2059 2060 switch (state) { 2061 case SDEV_CREATED: 2062 switch (oldstate) { 2063 case SDEV_CREATED_BLOCK: 2064 break; 2065 default: 2066 goto illegal; 2067 } 2068 break; 2069 2070 case SDEV_RUNNING: 2071 switch (oldstate) { 2072 case SDEV_CREATED: 2073 case SDEV_OFFLINE: 2074 case SDEV_QUIESCE: 2075 case SDEV_BLOCK: 2076 break; 2077 default: 2078 goto illegal; 2079 } 2080 break; 2081 2082 case SDEV_QUIESCE: 2083 switch (oldstate) { 2084 case SDEV_RUNNING: 2085 case SDEV_OFFLINE: 2086 break; 2087 default: 2088 goto illegal; 2089 } 2090 break; 2091 2092 case SDEV_OFFLINE: 2093 switch (oldstate) { 2094 case SDEV_CREATED: 2095 case SDEV_RUNNING: 2096 case SDEV_QUIESCE: 2097 case SDEV_BLOCK: 2098 break; 2099 default: 2100 goto illegal; 2101 } 2102 break; 2103 2104 case SDEV_BLOCK: 2105 switch (oldstate) { 2106 case SDEV_RUNNING: 2107 case SDEV_CREATED_BLOCK: 2108 break; 2109 default: 2110 goto illegal; 2111 } 2112 break; 2113 2114 case SDEV_CREATED_BLOCK: 2115 switch (oldstate) { 2116 case SDEV_CREATED: 2117 break; 2118 default: 2119 goto illegal; 2120 } 2121 break; 2122 2123 case SDEV_CANCEL: 2124 switch (oldstate) { 2125 case SDEV_CREATED: 2126 case SDEV_RUNNING: 2127 case SDEV_QUIESCE: 2128 case SDEV_OFFLINE: 2129 case SDEV_BLOCK: 2130 break; 2131 default: 2132 goto illegal; 2133 } 2134 break; 2135 2136 case SDEV_DEL: 2137 switch (oldstate) { 2138 case SDEV_CREATED: 2139 case SDEV_RUNNING: 2140 case SDEV_OFFLINE: 2141 case SDEV_CANCEL: 2142 break; 2143 default: 2144 goto illegal; 2145 } 2146 break; 2147 2148 } 2149 sdev->sdev_state = state; 2150 return 0; 2151 2152 illegal: 2153 SCSI_LOG_ERROR_RECOVERY(1, 2154 sdev_printk(KERN_ERR, sdev, 2155 "Illegal state transition %s->%s\n", 2156 scsi_device_state_name(oldstate), 2157 scsi_device_state_name(state)) 2158 ); 2159 return -EINVAL; 2160} 2161EXPORT_SYMBOL(scsi_device_set_state); 2162 2163/** 2164 * sdev_evt_emit - emit a single SCSI device uevent 2165 * @sdev: associated SCSI device 2166 * @evt: event to emit 2167 * 2168 * Send a single uevent (scsi_event) to the associated scsi_device. 2169 */ 2170static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt) 2171{ 2172 int idx = 0; 2173 char *envp[3]; 2174 2175 switch (evt->evt_type) { 2176 case SDEV_EVT_MEDIA_CHANGE: 2177 envp[idx++] = "SDEV_MEDIA_CHANGE=1"; 2178 break; 2179 2180 default: 2181 /* do nothing */ 2182 break; 2183 } 2184 2185 envp[idx++] = NULL; 2186 2187 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp); 2188} 2189 2190/** 2191 * sdev_evt_thread - send a uevent for each scsi event 2192 * @work: work struct for scsi_device 2193 * 2194 * Dispatch queued events to their associated scsi_device kobjects 2195 * as uevents. 2196 */ 2197void scsi_evt_thread(struct work_struct *work) 2198{ 2199 struct scsi_device *sdev; 2200 LIST_HEAD(event_list); 2201 2202 sdev = container_of(work, struct scsi_device, event_work); 2203 2204 while (1) { 2205 struct scsi_event *evt; 2206 struct list_head *this, *tmp; 2207 unsigned long flags; 2208 2209 spin_lock_irqsave(&sdev->list_lock, flags); 2210 list_splice_init(&sdev->event_list, &event_list); 2211 spin_unlock_irqrestore(&sdev->list_lock, flags); 2212 2213 if (list_empty(&event_list)) 2214 break; 2215 2216 list_for_each_safe(this, tmp, &event_list) { 2217 evt = list_entry(this, struct scsi_event, node); 2218 list_del(&evt->node); 2219 scsi_evt_emit(sdev, evt); 2220 kfree(evt); 2221 } 2222 } 2223} 2224 2225/** 2226 * sdev_evt_send - send asserted event to uevent thread 2227 * @sdev: scsi_device event occurred on 2228 * @evt: event to send 2229 * 2230 * Assert scsi device event asynchronously. 2231 */ 2232void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt) 2233{ 2234 unsigned long flags; 2235 2236#if 0 2237 /* FIXME: currently this check eliminates all media change events 2238 * for polled devices. Need to update to discriminate between AN 2239 * and polled events */ 2240 if (!test_bit(evt->evt_type, sdev->supported_events)) { 2241 kfree(evt); 2242 return; 2243 } 2244#endif 2245 2246 spin_lock_irqsave(&sdev->list_lock, flags); 2247 list_add_tail(&evt->node, &sdev->event_list); 2248 schedule_work(&sdev->event_work); 2249 spin_unlock_irqrestore(&sdev->list_lock, flags); 2250} 2251EXPORT_SYMBOL_GPL(sdev_evt_send); 2252 2253/** 2254 * sdev_evt_alloc - allocate a new scsi event 2255 * @evt_type: type of event to allocate 2256 * @gfpflags: GFP flags for allocation 2257 * 2258 * Allocates and returns a new scsi_event. 2259 */ 2260struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type, 2261 gfp_t gfpflags) 2262{ 2263 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags); 2264 if (!evt) 2265 return NULL; 2266 2267 evt->evt_type = evt_type; 2268 INIT_LIST_HEAD(&evt->node); 2269 2270 /* evt_type-specific initialization, if any */ 2271 switch (evt_type) { 2272 case SDEV_EVT_MEDIA_CHANGE: 2273 default: 2274 /* do nothing */ 2275 break; 2276 } 2277 2278 return evt; 2279} 2280EXPORT_SYMBOL_GPL(sdev_evt_alloc); 2281 2282/** 2283 * sdev_evt_send_simple - send asserted event to uevent thread 2284 * @sdev: scsi_device event occurred on 2285 * @evt_type: type of event to send 2286 * @gfpflags: GFP flags for allocation 2287 * 2288 * Assert scsi device event asynchronously, given an event type. 2289 */ 2290void sdev_evt_send_simple(struct scsi_device *sdev, 2291 enum scsi_device_event evt_type, gfp_t gfpflags) 2292{ 2293 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags); 2294 if (!evt) { 2295 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n", 2296 evt_type); 2297 return; 2298 } 2299 2300 sdev_evt_send(sdev, evt); 2301} 2302EXPORT_SYMBOL_GPL(sdev_evt_send_simple); 2303 2304/** 2305 * scsi_device_quiesce - Block user issued commands. 2306 * @sdev: scsi device to quiesce. 2307 * 2308 * This works by trying to transition to the SDEV_QUIESCE state 2309 * (which must be a legal transition). When the device is in this 2310 * state, only special requests will be accepted, all others will 2311 * be deferred. Since special requests may also be requeued requests, 2312 * a successful return doesn't guarantee the device will be 2313 * totally quiescent. 2314 * 2315 * Must be called with user context, may sleep. 2316 * 2317 * Returns zero if unsuccessful or an error if not. 2318 */ 2319int 2320scsi_device_quiesce(struct scsi_device *sdev) 2321{ 2322 int err = scsi_device_set_state(sdev, SDEV_QUIESCE); 2323 if (err) 2324 return err; 2325 2326 scsi_run_queue(sdev->request_queue); 2327 while (sdev->device_busy) { 2328 msleep_interruptible(200); 2329 scsi_run_queue(sdev->request_queue); 2330 } 2331 return 0; 2332} 2333EXPORT_SYMBOL(scsi_device_quiesce); 2334 2335/** 2336 * scsi_device_resume - Restart user issued commands to a quiesced device. 2337 * @sdev: scsi device to resume. 2338 * 2339 * Moves the device from quiesced back to running and restarts the 2340 * queues. 2341 * 2342 * Must be called with user context, may sleep. 2343 */ 2344void 2345scsi_device_resume(struct scsi_device *sdev) 2346{ 2347 if(scsi_device_set_state(sdev, SDEV_RUNNING)) 2348 return; 2349 scsi_run_queue(sdev->request_queue); 2350} 2351EXPORT_SYMBOL(scsi_device_resume); 2352 2353static void 2354device_quiesce_fn(struct scsi_device *sdev, void *data) 2355{ 2356 scsi_device_quiesce(sdev); 2357} 2358 2359void 2360scsi_target_quiesce(struct scsi_target *starget) 2361{ 2362 starget_for_each_device(starget, NULL, device_quiesce_fn); 2363} 2364EXPORT_SYMBOL(scsi_target_quiesce); 2365 2366static void 2367device_resume_fn(struct scsi_device *sdev, void *data) 2368{ 2369 scsi_device_resume(sdev); 2370} 2371 2372void 2373scsi_target_resume(struct scsi_target *starget) 2374{ 2375 starget_for_each_device(starget, NULL, device_resume_fn); 2376} 2377EXPORT_SYMBOL(scsi_target_resume); 2378 2379/** 2380 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state 2381 * @sdev: device to block 2382 * 2383 * Block request made by scsi lld's to temporarily stop all 2384 * scsi commands on the specified device. Called from interrupt 2385 * or normal process context. 2386 * 2387 * Returns zero if successful or error if not 2388 * 2389 * Notes: 2390 * This routine transitions the device to the SDEV_BLOCK state 2391 * (which must be a legal transition). When the device is in this 2392 * state, all commands are deferred until the scsi lld reenables 2393 * the device with scsi_device_unblock or device_block_tmo fires. 2394 * This routine assumes the host_lock is held on entry. 2395 */ 2396int 2397scsi_internal_device_block(struct scsi_device *sdev) 2398{ 2399 struct request_queue *q = sdev->request_queue; 2400 unsigned long flags; 2401 int err = 0; 2402 2403 err = scsi_device_set_state(sdev, SDEV_BLOCK); 2404 if (err) { 2405 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK); 2406 2407 if (err) 2408 return err; 2409 } 2410 2411 /* 2412 * The device has transitioned to SDEV_BLOCK. Stop the 2413 * block layer from calling the midlayer with this device's 2414 * request queue. 2415 */ 2416 spin_lock_irqsave(q->queue_lock, flags); 2417 blk_stop_queue(q); 2418 spin_unlock_irqrestore(q->queue_lock, flags); 2419 2420 return 0; 2421} 2422EXPORT_SYMBOL_GPL(scsi_internal_device_block); 2423 2424/** 2425 * scsi_internal_device_unblock - resume a device after a block request 2426 * @sdev: device to resume 2427 * 2428 * Called by scsi lld's or the midlayer to restart the device queue 2429 * for the previously suspended scsi device. Called from interrupt or 2430 * normal process context. 2431 * 2432 * Returns zero if successful or error if not. 2433 * 2434 * Notes: 2435 * This routine transitions the device to the SDEV_RUNNING state 2436 * (which must be a legal transition) allowing the midlayer to 2437 * goose the queue for this device. This routine assumes the 2438 * host_lock is held upon entry. 2439 */ 2440int 2441scsi_internal_device_unblock(struct scsi_device *sdev) 2442{ 2443 struct request_queue *q = sdev->request_queue; 2444 int err; 2445 unsigned long flags; 2446 2447 /* 2448 * Try to transition the scsi device to SDEV_RUNNING 2449 * and goose the device queue if successful. 2450 */ 2451 err = scsi_device_set_state(sdev, SDEV_RUNNING); 2452 if (err) { 2453 err = scsi_device_set_state(sdev, SDEV_CREATED); 2454 2455 if (err) 2456 return err; 2457 } 2458 2459 spin_lock_irqsave(q->queue_lock, flags); 2460 blk_start_queue(q); 2461 spin_unlock_irqrestore(q->queue_lock, flags); 2462 2463 return 0; 2464} 2465EXPORT_SYMBOL_GPL(scsi_internal_device_unblock); 2466 2467static void 2468device_block(struct scsi_device *sdev, void *data) 2469{ 2470 scsi_internal_device_block(sdev); 2471} 2472 2473static int 2474target_block(struct device *dev, void *data) 2475{ 2476 if (scsi_is_target_device(dev)) 2477 starget_for_each_device(to_scsi_target(dev), NULL, 2478 device_block); 2479 return 0; 2480} 2481 2482void 2483scsi_target_block(struct device *dev) 2484{ 2485 if (scsi_is_target_device(dev)) 2486 starget_for_each_device(to_scsi_target(dev), NULL, 2487 device_block); 2488 else 2489 device_for_each_child(dev, NULL, target_block); 2490} 2491EXPORT_SYMBOL_GPL(scsi_target_block); 2492 2493static void 2494device_unblock(struct scsi_device *sdev, void *data) 2495{ 2496 scsi_internal_device_unblock(sdev); 2497} 2498 2499static int 2500target_unblock(struct device *dev, void *data) 2501{ 2502 if (scsi_is_target_device(dev)) 2503 starget_for_each_device(to_scsi_target(dev), NULL, 2504 device_unblock); 2505 return 0; 2506} 2507 2508void 2509scsi_target_unblock(struct device *dev) 2510{ 2511 if (scsi_is_target_device(dev)) 2512 starget_for_each_device(to_scsi_target(dev), NULL, 2513 device_unblock); 2514 else 2515 device_for_each_child(dev, NULL, target_unblock); 2516} 2517EXPORT_SYMBOL_GPL(scsi_target_unblock); 2518 2519/** 2520 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt 2521 * @sgl: scatter-gather list 2522 * @sg_count: number of segments in sg 2523 * @offset: offset in bytes into sg, on return offset into the mapped area 2524 * @len: bytes to map, on return number of bytes mapped 2525 * 2526 * Returns virtual address of the start of the mapped page 2527 */ 2528void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count, 2529 size_t *offset, size_t *len) 2530{ 2531 int i; 2532 size_t sg_len = 0, len_complete = 0; 2533 struct scatterlist *sg; 2534 struct page *page; 2535 2536 WARN_ON(!irqs_disabled()); 2537 2538 for_each_sg(sgl, sg, sg_count, i) { 2539 len_complete = sg_len; /* Complete sg-entries */ 2540 sg_len += sg->length; 2541 if (sg_len > *offset) 2542 break; 2543 } 2544 2545 if (unlikely(i == sg_count)) { 2546 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, " 2547 "elements %d\n", 2548 __func__, sg_len, *offset, sg_count); 2549 WARN_ON(1); 2550 return NULL; 2551 } 2552 2553 /* Offset starting from the beginning of first page in this sg-entry */ 2554 *offset = *offset - len_complete + sg->offset; 2555 2556 /* Assumption: contiguous pages can be accessed as "page + i" */ 2557 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT)); 2558 *offset &= ~PAGE_MASK; 2559 2560 /* Bytes in this sg-entry from *offset to the end of the page */ 2561 sg_len = PAGE_SIZE - *offset; 2562 if (*len > sg_len) 2563 *len = sg_len; 2564 2565 return kmap_atomic(page, KM_BIO_SRC_IRQ); 2566} 2567EXPORT_SYMBOL(scsi_kmap_atomic_sg); 2568 2569/** 2570 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg 2571 * @virt: virtual address to be unmapped 2572 */ 2573void scsi_kunmap_atomic_sg(void *virt) 2574{ 2575 kunmap_atomic(virt, KM_BIO_SRC_IRQ); 2576} 2577EXPORT_SYMBOL(scsi_kunmap_atomic_sg); 2578