scsi_lib.c revision 79ed24297236b7430d6ce0a1511ff70cf5b6015a
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->data_len > 0 && req->data_len <= bufflen)) 244 memset(buffer + (bufflen - req->data_len), 0, req->data_len); 245 246 if (resid) 247 *resid = req->data_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 280struct scsi_io_context { 281 void *data; 282 void (*done)(void *data, char *sense, int result, int resid); 283 char sense[SCSI_SENSE_BUFFERSIZE]; 284}; 285 286static struct kmem_cache *scsi_io_context_cache; 287 288static void scsi_end_async(struct request *req, int uptodate) 289{ 290 struct scsi_io_context *sioc = req->end_io_data; 291 292 if (sioc->done) 293 sioc->done(sioc->data, sioc->sense, req->errors, req->data_len); 294 295 kmem_cache_free(scsi_io_context_cache, sioc); 296 __blk_put_request(req->q, req); 297} 298 299static int scsi_merge_bio(struct request *rq, struct bio *bio) 300{ 301 struct request_queue *q = rq->q; 302 303 bio->bi_flags &= ~(1 << BIO_SEG_VALID); 304 if (rq_data_dir(rq) == WRITE) 305 bio->bi_rw |= (1 << BIO_RW); 306 blk_queue_bounce(q, &bio); 307 308 return blk_rq_append_bio(q, rq, bio); 309} 310 311static void scsi_bi_endio(struct bio *bio, int error) 312{ 313 bio_put(bio); 314} 315 316/** 317 * scsi_req_map_sg - map a scatterlist into a request 318 * @rq: request to fill 319 * @sgl: scatterlist 320 * @nsegs: number of elements 321 * @bufflen: len of buffer 322 * @gfp: memory allocation flags 323 * 324 * scsi_req_map_sg maps a scatterlist into a request so that the 325 * request can be sent to the block layer. We do not trust the scatterlist 326 * sent to use, as some ULDs use that struct to only organize the pages. 327 */ 328static int scsi_req_map_sg(struct request *rq, struct scatterlist *sgl, 329 int nsegs, unsigned bufflen, gfp_t gfp) 330{ 331 struct request_queue *q = rq->q; 332 int nr_pages = (bufflen + sgl[0].offset + PAGE_SIZE - 1) >> PAGE_SHIFT; 333 unsigned int data_len = bufflen, len, bytes, off; 334 struct scatterlist *sg; 335 struct page *page; 336 struct bio *bio = NULL; 337 int i, err, nr_vecs = 0; 338 339 for_each_sg(sgl, sg, nsegs, i) { 340 page = sg_page(sg); 341 off = sg->offset; 342 len = sg->length; 343 344 while (len > 0 && data_len > 0) { 345 /* 346 * sg sends a scatterlist that is larger than 347 * the data_len it wants transferred for certain 348 * IO sizes 349 */ 350 bytes = min_t(unsigned int, len, PAGE_SIZE - off); 351 bytes = min(bytes, data_len); 352 353 if (!bio) { 354 nr_vecs = min_t(int, BIO_MAX_PAGES, nr_pages); 355 nr_pages -= nr_vecs; 356 357 bio = bio_alloc(gfp, nr_vecs); 358 if (!bio) { 359 err = -ENOMEM; 360 goto free_bios; 361 } 362 bio->bi_end_io = scsi_bi_endio; 363 } 364 365 if (bio_add_pc_page(q, bio, page, bytes, off) != 366 bytes) { 367 bio_put(bio); 368 err = -EINVAL; 369 goto free_bios; 370 } 371 372 if (bio->bi_vcnt >= nr_vecs) { 373 err = scsi_merge_bio(rq, bio); 374 if (err) { 375 bio_endio(bio, 0); 376 goto free_bios; 377 } 378 bio = NULL; 379 } 380 381 page++; 382 len -= bytes; 383 data_len -=bytes; 384 off = 0; 385 } 386 } 387 388 rq->buffer = rq->data = NULL; 389 rq->data_len = bufflen; 390 return 0; 391 392free_bios: 393 while ((bio = rq->bio) != NULL) { 394 rq->bio = bio->bi_next; 395 /* 396 * call endio instead of bio_put incase it was bounced 397 */ 398 bio_endio(bio, 0); 399 } 400 401 return err; 402} 403 404/** 405 * scsi_execute_async - insert request 406 * @sdev: scsi device 407 * @cmd: scsi command 408 * @cmd_len: length of scsi cdb 409 * @data_direction: DMA_TO_DEVICE, DMA_FROM_DEVICE, or DMA_NONE 410 * @buffer: data buffer (this can be a kernel buffer or scatterlist) 411 * @bufflen: len of buffer 412 * @use_sg: if buffer is a scatterlist this is the number of elements 413 * @timeout: request timeout in seconds 414 * @retries: number of times to retry request 415 * @privdata: data passed to done() 416 * @done: callback function when done 417 * @gfp: memory allocation flags 418 */ 419int scsi_execute_async(struct scsi_device *sdev, const unsigned char *cmd, 420 int cmd_len, int data_direction, void *buffer, unsigned bufflen, 421 int use_sg, int timeout, int retries, void *privdata, 422 void (*done)(void *, char *, int, int), gfp_t gfp) 423{ 424 struct request *req; 425 struct scsi_io_context *sioc; 426 int err = 0; 427 int write = (data_direction == DMA_TO_DEVICE); 428 429 sioc = kmem_cache_zalloc(scsi_io_context_cache, gfp); 430 if (!sioc) 431 return DRIVER_ERROR << 24; 432 433 req = blk_get_request(sdev->request_queue, write, gfp); 434 if (!req) 435 goto free_sense; 436 req->cmd_type = REQ_TYPE_BLOCK_PC; 437 req->cmd_flags |= REQ_QUIET; 438 439 if (use_sg) 440 err = scsi_req_map_sg(req, buffer, use_sg, bufflen, gfp); 441 else if (bufflen) 442 err = blk_rq_map_kern(req->q, req, buffer, bufflen, gfp); 443 444 if (err) 445 goto free_req; 446 447 req->cmd_len = cmd_len; 448 memset(req->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */ 449 memcpy(req->cmd, cmd, req->cmd_len); 450 req->sense = sioc->sense; 451 req->sense_len = 0; 452 req->timeout = timeout; 453 req->retries = retries; 454 req->end_io_data = sioc; 455 456 sioc->data = privdata; 457 sioc->done = done; 458 459 blk_execute_rq_nowait(req->q, NULL, req, 1, scsi_end_async); 460 return 0; 461 462free_req: 463 blk_put_request(req); 464free_sense: 465 kmem_cache_free(scsi_io_context_cache, sioc); 466 return DRIVER_ERROR << 24; 467} 468EXPORT_SYMBOL_GPL(scsi_execute_async); 469 470/* 471 * Function: scsi_init_cmd_errh() 472 * 473 * Purpose: Initialize cmd fields related to error handling. 474 * 475 * Arguments: cmd - command that is ready to be queued. 476 * 477 * Notes: This function has the job of initializing a number of 478 * fields related to error handling. Typically this will 479 * be called once for each command, as required. 480 */ 481static void scsi_init_cmd_errh(struct scsi_cmnd *cmd) 482{ 483 cmd->serial_number = 0; 484 scsi_set_resid(cmd, 0); 485 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 486 if (cmd->cmd_len == 0) 487 cmd->cmd_len = scsi_command_size(cmd->cmnd); 488} 489 490void scsi_device_unbusy(struct scsi_device *sdev) 491{ 492 struct Scsi_Host *shost = sdev->host; 493 struct scsi_target *starget = scsi_target(sdev); 494 unsigned long flags; 495 496 spin_lock_irqsave(shost->host_lock, flags); 497 shost->host_busy--; 498 starget->target_busy--; 499 if (unlikely(scsi_host_in_recovery(shost) && 500 (shost->host_failed || shost->host_eh_scheduled))) 501 scsi_eh_wakeup(shost); 502 spin_unlock(shost->host_lock); 503 spin_lock(sdev->request_queue->queue_lock); 504 sdev->device_busy--; 505 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags); 506} 507 508/* 509 * Called for single_lun devices on IO completion. Clear starget_sdev_user, 510 * and call blk_run_queue for all the scsi_devices on the target - 511 * including current_sdev first. 512 * 513 * Called with *no* scsi locks held. 514 */ 515static void scsi_single_lun_run(struct scsi_device *current_sdev) 516{ 517 struct Scsi_Host *shost = current_sdev->host; 518 struct scsi_device *sdev, *tmp; 519 struct scsi_target *starget = scsi_target(current_sdev); 520 unsigned long flags; 521 522 spin_lock_irqsave(shost->host_lock, flags); 523 starget->starget_sdev_user = NULL; 524 spin_unlock_irqrestore(shost->host_lock, flags); 525 526 /* 527 * Call blk_run_queue for all LUNs on the target, starting with 528 * current_sdev. We race with others (to set starget_sdev_user), 529 * but in most cases, we will be first. Ideally, each LU on the 530 * target would get some limited time or requests on the target. 531 */ 532 blk_run_queue(current_sdev->request_queue); 533 534 spin_lock_irqsave(shost->host_lock, flags); 535 if (starget->starget_sdev_user) 536 goto out; 537 list_for_each_entry_safe(sdev, tmp, &starget->devices, 538 same_target_siblings) { 539 if (sdev == current_sdev) 540 continue; 541 if (scsi_device_get(sdev)) 542 continue; 543 544 spin_unlock_irqrestore(shost->host_lock, flags); 545 blk_run_queue(sdev->request_queue); 546 spin_lock_irqsave(shost->host_lock, flags); 547 548 scsi_device_put(sdev); 549 } 550 out: 551 spin_unlock_irqrestore(shost->host_lock, flags); 552} 553 554static inline int scsi_device_is_busy(struct scsi_device *sdev) 555{ 556 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked) 557 return 1; 558 559 return 0; 560} 561 562static inline int scsi_target_is_busy(struct scsi_target *starget) 563{ 564 return ((starget->can_queue > 0 && 565 starget->target_busy >= starget->can_queue) || 566 starget->target_blocked); 567} 568 569static inline int scsi_host_is_busy(struct Scsi_Host *shost) 570{ 571 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) || 572 shost->host_blocked || shost->host_self_blocked) 573 return 1; 574 575 return 0; 576} 577 578/* 579 * Function: scsi_run_queue() 580 * 581 * Purpose: Select a proper request queue to serve next 582 * 583 * Arguments: q - last request's queue 584 * 585 * Returns: Nothing 586 * 587 * Notes: The previous command was completely finished, start 588 * a new one if possible. 589 */ 590static void scsi_run_queue(struct request_queue *q) 591{ 592 struct scsi_device *sdev = q->queuedata; 593 struct Scsi_Host *shost = sdev->host; 594 LIST_HEAD(starved_list); 595 unsigned long flags; 596 597 if (scsi_target(sdev)->single_lun) 598 scsi_single_lun_run(sdev); 599 600 spin_lock_irqsave(shost->host_lock, flags); 601 list_splice_init(&shost->starved_list, &starved_list); 602 603 while (!list_empty(&starved_list)) { 604 int flagset; 605 606 /* 607 * As long as shost is accepting commands and we have 608 * starved queues, call blk_run_queue. scsi_request_fn 609 * drops the queue_lock and can add us back to the 610 * starved_list. 611 * 612 * host_lock protects the starved_list and starved_entry. 613 * scsi_request_fn must get the host_lock before checking 614 * or modifying starved_list or starved_entry. 615 */ 616 if (scsi_host_is_busy(shost)) 617 break; 618 619 sdev = list_entry(starved_list.next, 620 struct scsi_device, starved_entry); 621 list_del_init(&sdev->starved_entry); 622 if (scsi_target_is_busy(scsi_target(sdev))) { 623 list_move_tail(&sdev->starved_entry, 624 &shost->starved_list); 625 continue; 626 } 627 628 spin_unlock(shost->host_lock); 629 630 spin_lock(sdev->request_queue->queue_lock); 631 flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) && 632 !test_bit(QUEUE_FLAG_REENTER, 633 &sdev->request_queue->queue_flags); 634 if (flagset) 635 queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue); 636 __blk_run_queue(sdev->request_queue); 637 if (flagset) 638 queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue); 639 spin_unlock(sdev->request_queue->queue_lock); 640 641 spin_lock(shost->host_lock); 642 } 643 /* put any unprocessed entries back */ 644 list_splice(&starved_list, &shost->starved_list); 645 spin_unlock_irqrestore(shost->host_lock, flags); 646 647 blk_run_queue(q); 648} 649 650/* 651 * Function: scsi_requeue_command() 652 * 653 * Purpose: Handle post-processing of completed commands. 654 * 655 * Arguments: q - queue to operate on 656 * cmd - command that may need to be requeued. 657 * 658 * Returns: Nothing 659 * 660 * Notes: After command completion, there may be blocks left 661 * over which weren't finished by the previous command 662 * this can be for a number of reasons - the main one is 663 * I/O errors in the middle of the request, in which case 664 * we need to request the blocks that come after the bad 665 * sector. 666 * Notes: Upon return, cmd is a stale pointer. 667 */ 668static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd) 669{ 670 struct request *req = cmd->request; 671 unsigned long flags; 672 673 spin_lock_irqsave(q->queue_lock, flags); 674 scsi_unprep_request(req); 675 blk_requeue_request(q, req); 676 spin_unlock_irqrestore(q->queue_lock, flags); 677 678 scsi_run_queue(q); 679} 680 681void scsi_next_command(struct scsi_cmnd *cmd) 682{ 683 struct scsi_device *sdev = cmd->device; 684 struct request_queue *q = sdev->request_queue; 685 686 /* need to hold a reference on the device before we let go of the cmd */ 687 get_device(&sdev->sdev_gendev); 688 689 scsi_put_command(cmd); 690 scsi_run_queue(q); 691 692 /* ok to remove device now */ 693 put_device(&sdev->sdev_gendev); 694} 695 696void scsi_run_host_queues(struct Scsi_Host *shost) 697{ 698 struct scsi_device *sdev; 699 700 shost_for_each_device(sdev, shost) 701 scsi_run_queue(sdev->request_queue); 702} 703 704static void __scsi_release_buffers(struct scsi_cmnd *, int); 705 706/* 707 * Function: scsi_end_request() 708 * 709 * Purpose: Post-processing of completed commands (usually invoked at end 710 * of upper level post-processing and scsi_io_completion). 711 * 712 * Arguments: cmd - command that is complete. 713 * error - 0 if I/O indicates success, < 0 for I/O error. 714 * bytes - number of bytes of completed I/O 715 * requeue - indicates whether we should requeue leftovers. 716 * 717 * Lock status: Assumed that lock is not held upon entry. 718 * 719 * Returns: cmd if requeue required, NULL otherwise. 720 * 721 * Notes: This is called for block device requests in order to 722 * mark some number of sectors as complete. 723 * 724 * We are guaranteeing that the request queue will be goosed 725 * at some point during this call. 726 * Notes: If cmd was requeued, upon return it will be a stale pointer. 727 */ 728static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error, 729 int bytes, int requeue) 730{ 731 struct request_queue *q = cmd->device->request_queue; 732 struct request *req = cmd->request; 733 734 /* 735 * If there are blocks left over at the end, set up the command 736 * to queue the remainder of them. 737 */ 738 if (blk_end_request(req, error, bytes)) { 739 int leftover = (req->hard_nr_sectors << 9); 740 741 if (blk_pc_request(req)) 742 leftover = req->data_len; 743 744 /* kill remainder if no retrys */ 745 if (error && scsi_noretry_cmd(cmd)) 746 blk_end_request(req, error, leftover); 747 else { 748 if (requeue) { 749 /* 750 * Bleah. Leftovers again. Stick the 751 * leftovers in the front of the 752 * queue, and goose the queue again. 753 */ 754 scsi_release_buffers(cmd); 755 scsi_requeue_command(q, cmd); 756 cmd = NULL; 757 } 758 return cmd; 759 } 760 } 761 762 /* 763 * This will goose the queue request function at the end, so we don't 764 * need to worry about launching another command. 765 */ 766 __scsi_release_buffers(cmd, 0); 767 scsi_next_command(cmd); 768 return NULL; 769} 770 771static inline unsigned int scsi_sgtable_index(unsigned short nents) 772{ 773 unsigned int index; 774 775 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS); 776 777 if (nents <= 8) 778 index = 0; 779 else 780 index = get_count_order(nents) - 3; 781 782 return index; 783} 784 785static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents) 786{ 787 struct scsi_host_sg_pool *sgp; 788 789 sgp = scsi_sg_pools + scsi_sgtable_index(nents); 790 mempool_free(sgl, sgp->pool); 791} 792 793static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask) 794{ 795 struct scsi_host_sg_pool *sgp; 796 797 sgp = scsi_sg_pools + scsi_sgtable_index(nents); 798 return mempool_alloc(sgp->pool, gfp_mask); 799} 800 801static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, 802 gfp_t gfp_mask) 803{ 804 int ret; 805 806 BUG_ON(!nents); 807 808 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS, 809 gfp_mask, scsi_sg_alloc); 810 if (unlikely(ret)) 811 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, 812 scsi_sg_free); 813 814 return ret; 815} 816 817static void scsi_free_sgtable(struct scsi_data_buffer *sdb) 818{ 819 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free); 820} 821 822static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check) 823{ 824 825 if (cmd->sdb.table.nents) 826 scsi_free_sgtable(&cmd->sdb); 827 828 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 829 830 if (do_bidi_check && scsi_bidi_cmnd(cmd)) { 831 struct scsi_data_buffer *bidi_sdb = 832 cmd->request->next_rq->special; 833 scsi_free_sgtable(bidi_sdb); 834 kmem_cache_free(scsi_sdb_cache, bidi_sdb); 835 cmd->request->next_rq->special = NULL; 836 } 837 838 if (scsi_prot_sg_count(cmd)) 839 scsi_free_sgtable(cmd->prot_sdb); 840} 841 842/* 843 * Function: scsi_release_buffers() 844 * 845 * Purpose: Completion processing for block device I/O requests. 846 * 847 * Arguments: cmd - command that we are bailing. 848 * 849 * Lock status: Assumed that no lock is held upon entry. 850 * 851 * Returns: Nothing 852 * 853 * Notes: In the event that an upper level driver rejects a 854 * command, we must release resources allocated during 855 * the __init_io() function. Primarily this would involve 856 * the scatter-gather table, and potentially any bounce 857 * buffers. 858 */ 859void scsi_release_buffers(struct scsi_cmnd *cmd) 860{ 861 __scsi_release_buffers(cmd, 1); 862} 863EXPORT_SYMBOL(scsi_release_buffers); 864 865/* 866 * Bidi commands Must be complete as a whole, both sides at once. 867 * If part of the bytes were written and lld returned 868 * scsi_in()->resid and/or scsi_out()->resid this information will be left 869 * in req->data_len and req->next_rq->data_len. The upper-layer driver can 870 * decide what to do with this information. 871 */ 872static void scsi_end_bidi_request(struct scsi_cmnd *cmd) 873{ 874 struct request *req = cmd->request; 875 unsigned int dlen = req->data_len; 876 unsigned int next_dlen = req->next_rq->data_len; 877 878 req->data_len = scsi_out(cmd)->resid; 879 req->next_rq->data_len = scsi_in(cmd)->resid; 880 881 /* The req and req->next_rq have not been completed */ 882 BUG_ON(blk_end_bidi_request(req, 0, dlen, next_dlen)); 883 884 scsi_release_buffers(cmd); 885 886 /* 887 * This will goose the queue request function at the end, so we don't 888 * need to worry about launching another command. 889 */ 890 scsi_next_command(cmd); 891} 892 893/* 894 * Function: scsi_io_completion() 895 * 896 * Purpose: Completion processing for block device I/O requests. 897 * 898 * Arguments: cmd - command that is finished. 899 * 900 * Lock status: Assumed that no lock is held upon entry. 901 * 902 * Returns: Nothing 903 * 904 * Notes: This function is matched in terms of capabilities to 905 * the function that created the scatter-gather list. 906 * In other words, if there are no bounce buffers 907 * (the normal case for most drivers), we don't need 908 * the logic to deal with cleaning up afterwards. 909 * 910 * We must call scsi_end_request(). This will finish off 911 * the specified number of sectors. If we are done, the 912 * command block will be released and the queue function 913 * will be goosed. If we are not done then we have to 914 * figure out what to do next: 915 * 916 * a) We can call scsi_requeue_command(). The request 917 * will be unprepared and put back on the queue. Then 918 * a new command will be created for it. This should 919 * be used if we made forward progress, or if we want 920 * to switch from READ(10) to READ(6) for example. 921 * 922 * b) We can call scsi_queue_insert(). The request will 923 * be put back on the queue and retried using the same 924 * command as before, possibly after a delay. 925 * 926 * c) We can call blk_end_request() with -EIO to fail 927 * the remainder of the request. 928 */ 929void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes) 930{ 931 int result = cmd->result; 932 int this_count; 933 struct request_queue *q = cmd->device->request_queue; 934 struct request *req = cmd->request; 935 int error = 0; 936 struct scsi_sense_hdr sshdr; 937 int sense_valid = 0; 938 int sense_deferred = 0; 939 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY, 940 ACTION_DELAYED_RETRY} action; 941 char *description = NULL; 942 943 if (result) { 944 sense_valid = scsi_command_normalize_sense(cmd, &sshdr); 945 if (sense_valid) 946 sense_deferred = scsi_sense_is_deferred(&sshdr); 947 } 948 949 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */ 950 req->errors = result; 951 if (result) { 952 if (sense_valid && req->sense) { 953 /* 954 * SG_IO wants current and deferred errors 955 */ 956 int len = 8 + cmd->sense_buffer[7]; 957 958 if (len > SCSI_SENSE_BUFFERSIZE) 959 len = SCSI_SENSE_BUFFERSIZE; 960 memcpy(req->sense, cmd->sense_buffer, len); 961 req->sense_len = len; 962 } 963 if (!sense_deferred) 964 error = -EIO; 965 } 966 if (scsi_bidi_cmnd(cmd)) { 967 /* will also release_buffers */ 968 scsi_end_bidi_request(cmd); 969 return; 970 } 971 req->data_len = scsi_get_resid(cmd); 972 } 973 974 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */ 975 976 /* 977 * Next deal with any sectors which we were able to correctly 978 * handle. 979 */ 980 SCSI_LOG_HLCOMPLETE(1, printk("%ld sectors total, " 981 "%d bytes done.\n", 982 req->nr_sectors, good_bytes)); 983 984 /* A number of bytes were successfully read. If there 985 * are leftovers and there is some kind of error 986 * (result != 0), retry the rest. 987 */ 988 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL) 989 return; 990 this_count = blk_rq_bytes(req); 991 992 error = -EIO; 993 994 if (host_byte(result) == DID_RESET) { 995 /* Third party bus reset or reset for error recovery 996 * reasons. Just retry the command and see what 997 * happens. 998 */ 999 action = ACTION_RETRY; 1000 } else if (sense_valid && !sense_deferred) { 1001 switch (sshdr.sense_key) { 1002 case UNIT_ATTENTION: 1003 if (cmd->device->removable) { 1004 /* Detected disc change. Set a bit 1005 * and quietly refuse further access. 1006 */ 1007 cmd->device->changed = 1; 1008 description = "Media Changed"; 1009 action = ACTION_FAIL; 1010 } else { 1011 /* Must have been a power glitch, or a 1012 * bus reset. Could not have been a 1013 * media change, so we just retry the 1014 * command and see what happens. 1015 */ 1016 action = ACTION_RETRY; 1017 } 1018 break; 1019 case ILLEGAL_REQUEST: 1020 /* If we had an ILLEGAL REQUEST returned, then 1021 * we may have performed an unsupported 1022 * command. The only thing this should be 1023 * would be a ten byte read where only a six 1024 * byte read was supported. Also, on a system 1025 * where READ CAPACITY failed, we may have 1026 * read past the end of the disk. 1027 */ 1028 if ((cmd->device->use_10_for_rw && 1029 sshdr.asc == 0x20 && sshdr.ascq == 0x00) && 1030 (cmd->cmnd[0] == READ_10 || 1031 cmd->cmnd[0] == WRITE_10)) { 1032 /* This will issue a new 6-byte command. */ 1033 cmd->device->use_10_for_rw = 0; 1034 action = ACTION_REPREP; 1035 } else if (sshdr.asc == 0x10) /* DIX */ { 1036 description = "Host Data Integrity Failure"; 1037 action = ACTION_FAIL; 1038 error = -EILSEQ; 1039 } else 1040 action = ACTION_FAIL; 1041 break; 1042 case ABORTED_COMMAND: 1043 if (sshdr.asc == 0x10) { /* DIF */ 1044 description = "Target Data Integrity Failure"; 1045 action = ACTION_FAIL; 1046 error = -EILSEQ; 1047 } else 1048 action = ACTION_RETRY; 1049 break; 1050 case NOT_READY: 1051 /* If the device is in the process of becoming 1052 * ready, or has a temporary blockage, retry. 1053 */ 1054 if (sshdr.asc == 0x04) { 1055 switch (sshdr.ascq) { 1056 case 0x01: /* becoming ready */ 1057 case 0x04: /* format in progress */ 1058 case 0x05: /* rebuild in progress */ 1059 case 0x06: /* recalculation in progress */ 1060 case 0x07: /* operation in progress */ 1061 case 0x08: /* Long write in progress */ 1062 case 0x09: /* self test in progress */ 1063 action = ACTION_DELAYED_RETRY; 1064 break; 1065 default: 1066 description = "Device not ready"; 1067 action = ACTION_FAIL; 1068 break; 1069 } 1070 } else { 1071 description = "Device not ready"; 1072 action = ACTION_FAIL; 1073 } 1074 break; 1075 case VOLUME_OVERFLOW: 1076 /* See SSC3rXX or current. */ 1077 action = ACTION_FAIL; 1078 break; 1079 default: 1080 description = "Unhandled sense code"; 1081 action = ACTION_FAIL; 1082 break; 1083 } 1084 } else { 1085 description = "Unhandled error code"; 1086 action = ACTION_FAIL; 1087 } 1088 1089 switch (action) { 1090 case ACTION_FAIL: 1091 /* Give up and fail the remainder of the request */ 1092 scsi_release_buffers(cmd); 1093 if (!(req->cmd_flags & REQ_QUIET)) { 1094 if (description) 1095 scmd_printk(KERN_INFO, cmd, "%s\n", 1096 description); 1097 scsi_print_result(cmd); 1098 if (driver_byte(result) & DRIVER_SENSE) 1099 scsi_print_sense("", cmd); 1100 } 1101 blk_end_request(req, -EIO, blk_rq_bytes(req)); 1102 scsi_next_command(cmd); 1103 break; 1104 case ACTION_REPREP: 1105 /* Unprep the request and put it back at the head of the queue. 1106 * A new command will be prepared and issued. 1107 */ 1108 scsi_release_buffers(cmd); 1109 scsi_requeue_command(q, cmd); 1110 break; 1111 case ACTION_RETRY: 1112 /* Retry the same command immediately */ 1113 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0); 1114 break; 1115 case ACTION_DELAYED_RETRY: 1116 /* Retry the same command after a delay */ 1117 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0); 1118 break; 1119 } 1120} 1121 1122static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb, 1123 gfp_t gfp_mask) 1124{ 1125 int count; 1126 1127 /* 1128 * If sg table allocation fails, requeue request later. 1129 */ 1130 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments, 1131 gfp_mask))) { 1132 return BLKPREP_DEFER; 1133 } 1134 1135 req->buffer = NULL; 1136 1137 /* 1138 * Next, walk the list, and fill in the addresses and sizes of 1139 * each segment. 1140 */ 1141 count = blk_rq_map_sg(req->q, req, sdb->table.sgl); 1142 BUG_ON(count > sdb->table.nents); 1143 sdb->table.nents = count; 1144 if (blk_pc_request(req)) 1145 sdb->length = req->data_len; 1146 else 1147 sdb->length = req->nr_sectors << 9; 1148 return BLKPREP_OK; 1149} 1150 1151/* 1152 * Function: scsi_init_io() 1153 * 1154 * Purpose: SCSI I/O initialize function. 1155 * 1156 * Arguments: cmd - Command descriptor we wish to initialize 1157 * 1158 * Returns: 0 on success 1159 * BLKPREP_DEFER if the failure is retryable 1160 * BLKPREP_KILL if the failure is fatal 1161 */ 1162int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask) 1163{ 1164 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask); 1165 if (error) 1166 goto err_exit; 1167 1168 if (blk_bidi_rq(cmd->request)) { 1169 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc( 1170 scsi_sdb_cache, GFP_ATOMIC); 1171 if (!bidi_sdb) { 1172 error = BLKPREP_DEFER; 1173 goto err_exit; 1174 } 1175 1176 cmd->request->next_rq->special = bidi_sdb; 1177 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb, 1178 GFP_ATOMIC); 1179 if (error) 1180 goto err_exit; 1181 } 1182 1183 if (blk_integrity_rq(cmd->request)) { 1184 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb; 1185 int ivecs, count; 1186 1187 BUG_ON(prot_sdb == NULL); 1188 ivecs = blk_rq_count_integrity_sg(cmd->request); 1189 1190 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) { 1191 error = BLKPREP_DEFER; 1192 goto err_exit; 1193 } 1194 1195 count = blk_rq_map_integrity_sg(cmd->request, 1196 prot_sdb->table.sgl); 1197 BUG_ON(unlikely(count > ivecs)); 1198 1199 cmd->prot_sdb = prot_sdb; 1200 cmd->prot_sdb->table.nents = count; 1201 } 1202 1203 return BLKPREP_OK ; 1204 1205err_exit: 1206 scsi_release_buffers(cmd); 1207 if (error == BLKPREP_KILL) 1208 scsi_put_command(cmd); 1209 else /* BLKPREP_DEFER */ 1210 scsi_unprep_request(cmd->request); 1211 1212 return error; 1213} 1214EXPORT_SYMBOL(scsi_init_io); 1215 1216static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev, 1217 struct request *req) 1218{ 1219 struct scsi_cmnd *cmd; 1220 1221 if (!req->special) { 1222 cmd = scsi_get_command(sdev, GFP_ATOMIC); 1223 if (unlikely(!cmd)) 1224 return NULL; 1225 req->special = cmd; 1226 } else { 1227 cmd = req->special; 1228 } 1229 1230 /* pull a tag out of the request if we have one */ 1231 cmd->tag = req->tag; 1232 cmd->request = req; 1233 1234 cmd->cmnd = req->cmd; 1235 1236 return cmd; 1237} 1238 1239int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req) 1240{ 1241 struct scsi_cmnd *cmd; 1242 int ret = scsi_prep_state_check(sdev, req); 1243 1244 if (ret != BLKPREP_OK) 1245 return ret; 1246 1247 cmd = scsi_get_cmd_from_req(sdev, req); 1248 if (unlikely(!cmd)) 1249 return BLKPREP_DEFER; 1250 1251 /* 1252 * BLOCK_PC requests may transfer data, in which case they must 1253 * a bio attached to them. Or they might contain a SCSI command 1254 * that does not transfer data, in which case they may optionally 1255 * submit a request without an attached bio. 1256 */ 1257 if (req->bio) { 1258 int ret; 1259 1260 BUG_ON(!req->nr_phys_segments); 1261 1262 ret = scsi_init_io(cmd, GFP_ATOMIC); 1263 if (unlikely(ret)) 1264 return ret; 1265 } else { 1266 BUG_ON(req->data_len); 1267 BUG_ON(req->data); 1268 1269 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1270 req->buffer = NULL; 1271 } 1272 1273 cmd->cmd_len = req->cmd_len; 1274 if (!req->data_len) 1275 cmd->sc_data_direction = DMA_NONE; 1276 else if (rq_data_dir(req) == WRITE) 1277 cmd->sc_data_direction = DMA_TO_DEVICE; 1278 else 1279 cmd->sc_data_direction = DMA_FROM_DEVICE; 1280 1281 cmd->transfersize = req->data_len; 1282 cmd->allowed = req->retries; 1283 return BLKPREP_OK; 1284} 1285EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd); 1286 1287/* 1288 * Setup a REQ_TYPE_FS command. These are simple read/write request 1289 * from filesystems that still need to be translated to SCSI CDBs from 1290 * the ULD. 1291 */ 1292int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req) 1293{ 1294 struct scsi_cmnd *cmd; 1295 int ret = scsi_prep_state_check(sdev, req); 1296 1297 if (ret != BLKPREP_OK) 1298 return ret; 1299 1300 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh 1301 && sdev->scsi_dh_data->scsi_dh->prep_fn)) { 1302 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req); 1303 if (ret != BLKPREP_OK) 1304 return ret; 1305 } 1306 1307 /* 1308 * Filesystem requests must transfer data. 1309 */ 1310 BUG_ON(!req->nr_phys_segments); 1311 1312 cmd = scsi_get_cmd_from_req(sdev, req); 1313 if (unlikely(!cmd)) 1314 return BLKPREP_DEFER; 1315 1316 memset(cmd->cmnd, 0, BLK_MAX_CDB); 1317 return scsi_init_io(cmd, GFP_ATOMIC); 1318} 1319EXPORT_SYMBOL(scsi_setup_fs_cmnd); 1320 1321int scsi_prep_state_check(struct scsi_device *sdev, struct request *req) 1322{ 1323 int ret = BLKPREP_OK; 1324 1325 /* 1326 * If the device is not in running state we will reject some 1327 * or all commands. 1328 */ 1329 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) { 1330 switch (sdev->sdev_state) { 1331 case SDEV_OFFLINE: 1332 /* 1333 * If the device is offline we refuse to process any 1334 * commands. The device must be brought online 1335 * before trying any recovery commands. 1336 */ 1337 sdev_printk(KERN_ERR, sdev, 1338 "rejecting I/O to offline device\n"); 1339 ret = BLKPREP_KILL; 1340 break; 1341 case SDEV_DEL: 1342 /* 1343 * If the device is fully deleted, we refuse to 1344 * process any commands as well. 1345 */ 1346 sdev_printk(KERN_ERR, sdev, 1347 "rejecting I/O to dead device\n"); 1348 ret = BLKPREP_KILL; 1349 break; 1350 case SDEV_QUIESCE: 1351 case SDEV_BLOCK: 1352 case SDEV_CREATED_BLOCK: 1353 /* 1354 * If the devices is blocked we defer normal commands. 1355 */ 1356 if (!(req->cmd_flags & REQ_PREEMPT)) 1357 ret = BLKPREP_DEFER; 1358 break; 1359 default: 1360 /* 1361 * For any other not fully online state we only allow 1362 * special commands. In particular any user initiated 1363 * command is not allowed. 1364 */ 1365 if (!(req->cmd_flags & REQ_PREEMPT)) 1366 ret = BLKPREP_KILL; 1367 break; 1368 } 1369 } 1370 return ret; 1371} 1372EXPORT_SYMBOL(scsi_prep_state_check); 1373 1374int scsi_prep_return(struct request_queue *q, struct request *req, int ret) 1375{ 1376 struct scsi_device *sdev = q->queuedata; 1377 1378 switch (ret) { 1379 case BLKPREP_KILL: 1380 req->errors = DID_NO_CONNECT << 16; 1381 /* release the command and kill it */ 1382 if (req->special) { 1383 struct scsi_cmnd *cmd = req->special; 1384 scsi_release_buffers(cmd); 1385 scsi_put_command(cmd); 1386 req->special = NULL; 1387 } 1388 break; 1389 case BLKPREP_DEFER: 1390 /* 1391 * If we defer, the elv_next_request() returns NULL, but the 1392 * queue must be restarted, so we plug here if no returning 1393 * command will automatically do that. 1394 */ 1395 if (sdev->device_busy == 0) 1396 blk_plug_device(q); 1397 break; 1398 default: 1399 req->cmd_flags |= REQ_DONTPREP; 1400 } 1401 1402 return ret; 1403} 1404EXPORT_SYMBOL(scsi_prep_return); 1405 1406int scsi_prep_fn(struct request_queue *q, struct request *req) 1407{ 1408 struct scsi_device *sdev = q->queuedata; 1409 int ret = BLKPREP_KILL; 1410 1411 if (req->cmd_type == REQ_TYPE_BLOCK_PC) 1412 ret = scsi_setup_blk_pc_cmnd(sdev, req); 1413 return scsi_prep_return(q, req, ret); 1414} 1415 1416/* 1417 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else 1418 * return 0. 1419 * 1420 * Called with the queue_lock held. 1421 */ 1422static inline int scsi_dev_queue_ready(struct request_queue *q, 1423 struct scsi_device *sdev) 1424{ 1425 if (sdev->device_busy == 0 && sdev->device_blocked) { 1426 /* 1427 * unblock after device_blocked iterates to zero 1428 */ 1429 if (--sdev->device_blocked == 0) { 1430 SCSI_LOG_MLQUEUE(3, 1431 sdev_printk(KERN_INFO, sdev, 1432 "unblocking device at zero depth\n")); 1433 } else { 1434 blk_plug_device(q); 1435 return 0; 1436 } 1437 } 1438 if (scsi_device_is_busy(sdev)) 1439 return 0; 1440 1441 return 1; 1442} 1443 1444 1445/* 1446 * scsi_target_queue_ready: checks if there we can send commands to target 1447 * @sdev: scsi device on starget to check. 1448 * 1449 * Called with the host lock held. 1450 */ 1451static inline int scsi_target_queue_ready(struct Scsi_Host *shost, 1452 struct scsi_device *sdev) 1453{ 1454 struct scsi_target *starget = scsi_target(sdev); 1455 1456 if (starget->single_lun) { 1457 if (starget->starget_sdev_user && 1458 starget->starget_sdev_user != sdev) 1459 return 0; 1460 starget->starget_sdev_user = sdev; 1461 } 1462 1463 if (starget->target_busy == 0 && starget->target_blocked) { 1464 /* 1465 * unblock after target_blocked iterates to zero 1466 */ 1467 if (--starget->target_blocked == 0) { 1468 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget, 1469 "unblocking target at zero depth\n")); 1470 } else { 1471 blk_plug_device(sdev->request_queue); 1472 return 0; 1473 } 1474 } 1475 1476 if (scsi_target_is_busy(starget)) { 1477 if (list_empty(&sdev->starved_entry)) { 1478 list_add_tail(&sdev->starved_entry, 1479 &shost->starved_list); 1480 return 0; 1481 } 1482 } 1483 1484 /* We're OK to process the command, so we can't be starved */ 1485 if (!list_empty(&sdev->starved_entry)) 1486 list_del_init(&sdev->starved_entry); 1487 return 1; 1488} 1489 1490/* 1491 * scsi_host_queue_ready: if we can send requests to shost, return 1 else 1492 * return 0. We must end up running the queue again whenever 0 is 1493 * returned, else IO can hang. 1494 * 1495 * Called with host_lock held. 1496 */ 1497static inline int scsi_host_queue_ready(struct request_queue *q, 1498 struct Scsi_Host *shost, 1499 struct scsi_device *sdev) 1500{ 1501 if (scsi_host_in_recovery(shost)) 1502 return 0; 1503 if (shost->host_busy == 0 && shost->host_blocked) { 1504 /* 1505 * unblock after host_blocked iterates to zero 1506 */ 1507 if (--shost->host_blocked == 0) { 1508 SCSI_LOG_MLQUEUE(3, 1509 printk("scsi%d unblocking host at zero depth\n", 1510 shost->host_no)); 1511 } else { 1512 return 0; 1513 } 1514 } 1515 if (scsi_host_is_busy(shost)) { 1516 if (list_empty(&sdev->starved_entry)) 1517 list_add_tail(&sdev->starved_entry, &shost->starved_list); 1518 return 0; 1519 } 1520 1521 /* We're OK to process the command, so we can't be starved */ 1522 if (!list_empty(&sdev->starved_entry)) 1523 list_del_init(&sdev->starved_entry); 1524 1525 return 1; 1526} 1527 1528/* 1529 * Busy state exporting function for request stacking drivers. 1530 * 1531 * For efficiency, no lock is taken to check the busy state of 1532 * shost/starget/sdev, since the returned value is not guaranteed and 1533 * may be changed after request stacking drivers call the function, 1534 * regardless of taking lock or not. 1535 * 1536 * When scsi can't dispatch I/Os anymore and needs to kill I/Os 1537 * (e.g. !sdev), scsi needs to return 'not busy'. 1538 * Otherwise, request stacking drivers may hold requests forever. 1539 */ 1540static int scsi_lld_busy(struct request_queue *q) 1541{ 1542 struct scsi_device *sdev = q->queuedata; 1543 struct Scsi_Host *shost; 1544 struct scsi_target *starget; 1545 1546 if (!sdev) 1547 return 0; 1548 1549 shost = sdev->host; 1550 starget = scsi_target(sdev); 1551 1552 if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) || 1553 scsi_target_is_busy(starget) || scsi_device_is_busy(sdev)) 1554 return 1; 1555 1556 return 0; 1557} 1558 1559/* 1560 * Kill a request for a dead device 1561 */ 1562static void scsi_kill_request(struct request *req, struct request_queue *q) 1563{ 1564 struct scsi_cmnd *cmd = req->special; 1565 struct scsi_device *sdev = cmd->device; 1566 struct scsi_target *starget = scsi_target(sdev); 1567 struct Scsi_Host *shost = sdev->host; 1568 1569 blkdev_dequeue_request(req); 1570 1571 if (unlikely(cmd == NULL)) { 1572 printk(KERN_CRIT "impossible request in %s.\n", 1573 __func__); 1574 BUG(); 1575 } 1576 1577 scsi_init_cmd_errh(cmd); 1578 cmd->result = DID_NO_CONNECT << 16; 1579 atomic_inc(&cmd->device->iorequest_cnt); 1580 1581 /* 1582 * SCSI request completion path will do scsi_device_unbusy(), 1583 * bump busy counts. To bump the counters, we need to dance 1584 * with the locks as normal issue path does. 1585 */ 1586 sdev->device_busy++; 1587 spin_unlock(sdev->request_queue->queue_lock); 1588 spin_lock(shost->host_lock); 1589 shost->host_busy++; 1590 starget->target_busy++; 1591 spin_unlock(shost->host_lock); 1592 spin_lock(sdev->request_queue->queue_lock); 1593 1594 blk_complete_request(req); 1595} 1596 1597static void scsi_softirq_done(struct request *rq) 1598{ 1599 struct scsi_cmnd *cmd = rq->special; 1600 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout; 1601 int disposition; 1602 1603 INIT_LIST_HEAD(&cmd->eh_entry); 1604 1605 /* 1606 * Set the serial numbers back to zero 1607 */ 1608 cmd->serial_number = 0; 1609 1610 atomic_inc(&cmd->device->iodone_cnt); 1611 if (cmd->result) 1612 atomic_inc(&cmd->device->ioerr_cnt); 1613 1614 disposition = scsi_decide_disposition(cmd); 1615 if (disposition != SUCCESS && 1616 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) { 1617 sdev_printk(KERN_ERR, cmd->device, 1618 "timing out command, waited %lus\n", 1619 wait_for/HZ); 1620 disposition = SUCCESS; 1621 } 1622 1623 scsi_log_completion(cmd, disposition); 1624 1625 switch (disposition) { 1626 case SUCCESS: 1627 scsi_finish_command(cmd); 1628 break; 1629 case NEEDS_RETRY: 1630 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY); 1631 break; 1632 case ADD_TO_MLQUEUE: 1633 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY); 1634 break; 1635 default: 1636 if (!scsi_eh_scmd_add(cmd, 0)) 1637 scsi_finish_command(cmd); 1638 } 1639} 1640 1641/* 1642 * Function: scsi_request_fn() 1643 * 1644 * Purpose: Main strategy routine for SCSI. 1645 * 1646 * Arguments: q - Pointer to actual queue. 1647 * 1648 * Returns: Nothing 1649 * 1650 * Lock status: IO request lock assumed to be held when called. 1651 */ 1652static void scsi_request_fn(struct request_queue *q) 1653{ 1654 struct scsi_device *sdev = q->queuedata; 1655 struct Scsi_Host *shost; 1656 struct scsi_cmnd *cmd; 1657 struct request *req; 1658 1659 if (!sdev) { 1660 printk("scsi: killing requests for dead queue\n"); 1661 while ((req = elv_next_request(q)) != NULL) 1662 scsi_kill_request(req, q); 1663 return; 1664 } 1665 1666 if(!get_device(&sdev->sdev_gendev)) 1667 /* We must be tearing the block queue down already */ 1668 return; 1669 1670 /* 1671 * To start with, we keep looping until the queue is empty, or until 1672 * the host is no longer able to accept any more requests. 1673 */ 1674 shost = sdev->host; 1675 while (!blk_queue_plugged(q)) { 1676 int rtn; 1677 /* 1678 * get next queueable request. We do this early to make sure 1679 * that the request is fully prepared even if we cannot 1680 * accept it. 1681 */ 1682 req = elv_next_request(q); 1683 if (!req || !scsi_dev_queue_ready(q, sdev)) 1684 break; 1685 1686 if (unlikely(!scsi_device_online(sdev))) { 1687 sdev_printk(KERN_ERR, sdev, 1688 "rejecting I/O to offline device\n"); 1689 scsi_kill_request(req, q); 1690 continue; 1691 } 1692 1693 1694 /* 1695 * Remove the request from the request list. 1696 */ 1697 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req))) 1698 blkdev_dequeue_request(req); 1699 sdev->device_busy++; 1700 1701 spin_unlock(q->queue_lock); 1702 cmd = req->special; 1703 if (unlikely(cmd == NULL)) { 1704 printk(KERN_CRIT "impossible request in %s.\n" 1705 "please mail a stack trace to " 1706 "linux-scsi@vger.kernel.org\n", 1707 __func__); 1708 blk_dump_rq_flags(req, "foo"); 1709 BUG(); 1710 } 1711 spin_lock(shost->host_lock); 1712 1713 /* 1714 * We hit this when the driver is using a host wide 1715 * tag map. For device level tag maps the queue_depth check 1716 * in the device ready fn would prevent us from trying 1717 * to allocate a tag. Since the map is a shared host resource 1718 * we add the dev to the starved list so it eventually gets 1719 * a run when a tag is freed. 1720 */ 1721 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) { 1722 if (list_empty(&sdev->starved_entry)) 1723 list_add_tail(&sdev->starved_entry, 1724 &shost->starved_list); 1725 goto not_ready; 1726 } 1727 1728 if (!scsi_target_queue_ready(shost, sdev)) 1729 goto not_ready; 1730 1731 if (!scsi_host_queue_ready(q, shost, sdev)) 1732 goto not_ready; 1733 1734 scsi_target(sdev)->target_busy++; 1735 shost->host_busy++; 1736 1737 /* 1738 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will 1739 * take the lock again. 1740 */ 1741 spin_unlock_irq(shost->host_lock); 1742 1743 /* 1744 * Finally, initialize any error handling parameters, and set up 1745 * the timers for timeouts. 1746 */ 1747 scsi_init_cmd_errh(cmd); 1748 1749 /* 1750 * Dispatch the command to the low-level driver. 1751 */ 1752 rtn = scsi_dispatch_cmd(cmd); 1753 spin_lock_irq(q->queue_lock); 1754 if(rtn) { 1755 /* we're refusing the command; because of 1756 * the way locks get dropped, we need to 1757 * check here if plugging is required */ 1758 if(sdev->device_busy == 0) 1759 blk_plug_device(q); 1760 1761 break; 1762 } 1763 } 1764 1765 goto out; 1766 1767 not_ready: 1768 spin_unlock_irq(shost->host_lock); 1769 1770 /* 1771 * lock q, handle tag, requeue req, and decrement device_busy. We 1772 * must return with queue_lock held. 1773 * 1774 * Decrementing device_busy without checking it is OK, as all such 1775 * cases (host limits or settings) should run the queue at some 1776 * later time. 1777 */ 1778 spin_lock_irq(q->queue_lock); 1779 blk_requeue_request(q, req); 1780 sdev->device_busy--; 1781 if(sdev->device_busy == 0) 1782 blk_plug_device(q); 1783 out: 1784 /* must be careful here...if we trigger the ->remove() function 1785 * we cannot be holding the q lock */ 1786 spin_unlock_irq(q->queue_lock); 1787 put_device(&sdev->sdev_gendev); 1788 spin_lock_irq(q->queue_lock); 1789} 1790 1791u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost) 1792{ 1793 struct device *host_dev; 1794 u64 bounce_limit = 0xffffffff; 1795 1796 if (shost->unchecked_isa_dma) 1797 return BLK_BOUNCE_ISA; 1798 /* 1799 * Platforms with virtual-DMA translation 1800 * hardware have no practical limit. 1801 */ 1802 if (!PCI_DMA_BUS_IS_PHYS) 1803 return BLK_BOUNCE_ANY; 1804 1805 host_dev = scsi_get_device(shost); 1806 if (host_dev && host_dev->dma_mask) 1807 bounce_limit = *host_dev->dma_mask; 1808 1809 return bounce_limit; 1810} 1811EXPORT_SYMBOL(scsi_calculate_bounce_limit); 1812 1813struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost, 1814 request_fn_proc *request_fn) 1815{ 1816 struct request_queue *q; 1817 struct device *dev = shost->shost_gendev.parent; 1818 1819 q = blk_init_queue(request_fn, NULL); 1820 if (!q) 1821 return NULL; 1822 1823 /* 1824 * this limit is imposed by hardware restrictions 1825 */ 1826 blk_queue_max_hw_segments(q, shost->sg_tablesize); 1827 blk_queue_max_phys_segments(q, SCSI_MAX_SG_CHAIN_SEGMENTS); 1828 1829 blk_queue_max_sectors(q, shost->max_sectors); 1830 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost)); 1831 blk_queue_segment_boundary(q, shost->dma_boundary); 1832 dma_set_seg_boundary(dev, shost->dma_boundary); 1833 1834 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev)); 1835 1836 /* New queue, no concurrency on queue_flags */ 1837 if (!shost->use_clustering) 1838 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q); 1839 1840 /* 1841 * set a reasonable default alignment on word boundaries: the 1842 * host and device may alter it using 1843 * blk_queue_update_dma_alignment() later. 1844 */ 1845 blk_queue_dma_alignment(q, 0x03); 1846 1847 return q; 1848} 1849EXPORT_SYMBOL(__scsi_alloc_queue); 1850 1851struct request_queue *scsi_alloc_queue(struct scsi_device *sdev) 1852{ 1853 struct request_queue *q; 1854 1855 q = __scsi_alloc_queue(sdev->host, scsi_request_fn); 1856 if (!q) 1857 return NULL; 1858 1859 blk_queue_prep_rq(q, scsi_prep_fn); 1860 blk_queue_softirq_done(q, scsi_softirq_done); 1861 blk_queue_rq_timed_out(q, scsi_times_out); 1862 blk_queue_lld_busy(q, scsi_lld_busy); 1863 return q; 1864} 1865 1866void scsi_free_queue(struct request_queue *q) 1867{ 1868 blk_cleanup_queue(q); 1869} 1870 1871/* 1872 * Function: scsi_block_requests() 1873 * 1874 * Purpose: Utility function used by low-level drivers to prevent further 1875 * commands from being queued to the device. 1876 * 1877 * Arguments: shost - Host in question 1878 * 1879 * Returns: Nothing 1880 * 1881 * Lock status: No locks are assumed held. 1882 * 1883 * Notes: There is no timer nor any other means by which the requests 1884 * get unblocked other than the low-level driver calling 1885 * scsi_unblock_requests(). 1886 */ 1887void scsi_block_requests(struct Scsi_Host *shost) 1888{ 1889 shost->host_self_blocked = 1; 1890} 1891EXPORT_SYMBOL(scsi_block_requests); 1892 1893/* 1894 * Function: scsi_unblock_requests() 1895 * 1896 * Purpose: Utility function used by low-level drivers to allow further 1897 * commands from being queued to the device. 1898 * 1899 * Arguments: shost - Host in question 1900 * 1901 * Returns: Nothing 1902 * 1903 * Lock status: No locks are assumed held. 1904 * 1905 * Notes: There is no timer nor any other means by which the requests 1906 * get unblocked other than the low-level driver calling 1907 * scsi_unblock_requests(). 1908 * 1909 * This is done as an API function so that changes to the 1910 * internals of the scsi mid-layer won't require wholesale 1911 * changes to drivers that use this feature. 1912 */ 1913void scsi_unblock_requests(struct Scsi_Host *shost) 1914{ 1915 shost->host_self_blocked = 0; 1916 scsi_run_host_queues(shost); 1917} 1918EXPORT_SYMBOL(scsi_unblock_requests); 1919 1920int __init scsi_init_queue(void) 1921{ 1922 int i; 1923 1924 scsi_io_context_cache = kmem_cache_create("scsi_io_context", 1925 sizeof(struct scsi_io_context), 1926 0, 0, NULL); 1927 if (!scsi_io_context_cache) { 1928 printk(KERN_ERR "SCSI: can't init scsi io context cache\n"); 1929 return -ENOMEM; 1930 } 1931 1932 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer", 1933 sizeof(struct scsi_data_buffer), 1934 0, 0, NULL); 1935 if (!scsi_sdb_cache) { 1936 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n"); 1937 goto cleanup_io_context; 1938 } 1939 1940 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1941 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1942 int size = sgp->size * sizeof(struct scatterlist); 1943 1944 sgp->slab = kmem_cache_create(sgp->name, size, 0, 1945 SLAB_HWCACHE_ALIGN, NULL); 1946 if (!sgp->slab) { 1947 printk(KERN_ERR "SCSI: can't init sg slab %s\n", 1948 sgp->name); 1949 goto cleanup_sdb; 1950 } 1951 1952 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE, 1953 sgp->slab); 1954 if (!sgp->pool) { 1955 printk(KERN_ERR "SCSI: can't init sg mempool %s\n", 1956 sgp->name); 1957 goto cleanup_sdb; 1958 } 1959 } 1960 1961 return 0; 1962 1963cleanup_sdb: 1964 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1965 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1966 if (sgp->pool) 1967 mempool_destroy(sgp->pool); 1968 if (sgp->slab) 1969 kmem_cache_destroy(sgp->slab); 1970 } 1971 kmem_cache_destroy(scsi_sdb_cache); 1972cleanup_io_context: 1973 kmem_cache_destroy(scsi_io_context_cache); 1974 1975 return -ENOMEM; 1976} 1977 1978void scsi_exit_queue(void) 1979{ 1980 int i; 1981 1982 kmem_cache_destroy(scsi_io_context_cache); 1983 kmem_cache_destroy(scsi_sdb_cache); 1984 1985 for (i = 0; i < SG_MEMPOOL_NR; i++) { 1986 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i; 1987 mempool_destroy(sgp->pool); 1988 kmem_cache_destroy(sgp->slab); 1989 } 1990} 1991 1992/** 1993 * scsi_mode_select - issue a mode select 1994 * @sdev: SCSI device to be queried 1995 * @pf: Page format bit (1 == standard, 0 == vendor specific) 1996 * @sp: Save page bit (0 == don't save, 1 == save) 1997 * @modepage: mode page being requested 1998 * @buffer: request buffer (may not be smaller than eight bytes) 1999 * @len: length of request buffer. 2000 * @timeout: command timeout 2001 * @retries: number of retries before failing 2002 * @data: returns a structure abstracting the mode header data 2003 * @sshdr: place to put sense data (or NULL if no sense to be collected). 2004 * must be SCSI_SENSE_BUFFERSIZE big. 2005 * 2006 * Returns zero if successful; negative error number or scsi 2007 * status on error 2008 * 2009 */ 2010int 2011scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage, 2012 unsigned char *buffer, int len, int timeout, int retries, 2013 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 2014{ 2015 unsigned char cmd[10]; 2016 unsigned char *real_buffer; 2017 int ret; 2018 2019 memset(cmd, 0, sizeof(cmd)); 2020 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0); 2021 2022 if (sdev->use_10_for_ms) { 2023 if (len > 65535) 2024 return -EINVAL; 2025 real_buffer = kmalloc(8 + len, GFP_KERNEL); 2026 if (!real_buffer) 2027 return -ENOMEM; 2028 memcpy(real_buffer + 8, buffer, len); 2029 len += 8; 2030 real_buffer[0] = 0; 2031 real_buffer[1] = 0; 2032 real_buffer[2] = data->medium_type; 2033 real_buffer[3] = data->device_specific; 2034 real_buffer[4] = data->longlba ? 0x01 : 0; 2035 real_buffer[5] = 0; 2036 real_buffer[6] = data->block_descriptor_length >> 8; 2037 real_buffer[7] = data->block_descriptor_length; 2038 2039 cmd[0] = MODE_SELECT_10; 2040 cmd[7] = len >> 8; 2041 cmd[8] = len; 2042 } else { 2043 if (len > 255 || data->block_descriptor_length > 255 || 2044 data->longlba) 2045 return -EINVAL; 2046 2047 real_buffer = kmalloc(4 + len, GFP_KERNEL); 2048 if (!real_buffer) 2049 return -ENOMEM; 2050 memcpy(real_buffer + 4, buffer, len); 2051 len += 4; 2052 real_buffer[0] = 0; 2053 real_buffer[1] = data->medium_type; 2054 real_buffer[2] = data->device_specific; 2055 real_buffer[3] = data->block_descriptor_length; 2056 2057 2058 cmd[0] = MODE_SELECT; 2059 cmd[4] = len; 2060 } 2061 2062 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len, 2063 sshdr, timeout, retries, NULL); 2064 kfree(real_buffer); 2065 return ret; 2066} 2067EXPORT_SYMBOL_GPL(scsi_mode_select); 2068 2069/** 2070 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary. 2071 * @sdev: SCSI device to be queried 2072 * @dbd: set if mode sense will allow block descriptors to be returned 2073 * @modepage: mode page being requested 2074 * @buffer: request buffer (may not be smaller than eight bytes) 2075 * @len: length of request buffer. 2076 * @timeout: command timeout 2077 * @retries: number of retries before failing 2078 * @data: returns a structure abstracting the mode header data 2079 * @sshdr: place to put sense data (or NULL if no sense to be collected). 2080 * must be SCSI_SENSE_BUFFERSIZE big. 2081 * 2082 * Returns zero if unsuccessful, or the header offset (either 4 2083 * or 8 depending on whether a six or ten byte command was 2084 * issued) if successful. 2085 */ 2086int 2087scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, 2088 unsigned char *buffer, int len, int timeout, int retries, 2089 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 2090{ 2091 unsigned char cmd[12]; 2092 int use_10_for_ms; 2093 int header_length; 2094 int result; 2095 struct scsi_sense_hdr my_sshdr; 2096 2097 memset(data, 0, sizeof(*data)); 2098 memset(&cmd[0], 0, 12); 2099 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */ 2100 cmd[2] = modepage; 2101 2102 /* caller might not be interested in sense, but we need it */ 2103 if (!sshdr) 2104 sshdr = &my_sshdr; 2105 2106 retry: 2107 use_10_for_ms = sdev->use_10_for_ms; 2108 2109 if (use_10_for_ms) { 2110 if (len < 8) 2111 len = 8; 2112 2113 cmd[0] = MODE_SENSE_10; 2114 cmd[8] = len; 2115 header_length = 8; 2116 } else { 2117 if (len < 4) 2118 len = 4; 2119 2120 cmd[0] = MODE_SENSE; 2121 cmd[4] = len; 2122 header_length = 4; 2123 } 2124 2125 memset(buffer, 0, len); 2126 2127 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len, 2128 sshdr, timeout, retries, NULL); 2129 2130 /* This code looks awful: what it's doing is making sure an 2131 * ILLEGAL REQUEST sense return identifies the actual command 2132 * byte as the problem. MODE_SENSE commands can return 2133 * ILLEGAL REQUEST if the code page isn't supported */ 2134 2135 if (use_10_for_ms && !scsi_status_is_good(result) && 2136 (driver_byte(result) & DRIVER_SENSE)) { 2137 if (scsi_sense_valid(sshdr)) { 2138 if ((sshdr->sense_key == ILLEGAL_REQUEST) && 2139 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) { 2140 /* 2141 * Invalid command operation code 2142 */ 2143 sdev->use_10_for_ms = 0; 2144 goto retry; 2145 } 2146 } 2147 } 2148 2149 if(scsi_status_is_good(result)) { 2150 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b && 2151 (modepage == 6 || modepage == 8))) { 2152 /* Initio breakage? */ 2153 header_length = 0; 2154 data->length = 13; 2155 data->medium_type = 0; 2156 data->device_specific = 0; 2157 data->longlba = 0; 2158 data->block_descriptor_length = 0; 2159 } else if(use_10_for_ms) { 2160 data->length = buffer[0]*256 + buffer[1] + 2; 2161 data->medium_type = buffer[2]; 2162 data->device_specific = buffer[3]; 2163 data->longlba = buffer[4] & 0x01; 2164 data->block_descriptor_length = buffer[6]*256 2165 + buffer[7]; 2166 } else { 2167 data->length = buffer[0] + 1; 2168 data->medium_type = buffer[1]; 2169 data->device_specific = buffer[2]; 2170 data->block_descriptor_length = buffer[3]; 2171 } 2172 data->header_length = header_length; 2173 } 2174 2175 return result; 2176} 2177EXPORT_SYMBOL(scsi_mode_sense); 2178 2179/** 2180 * scsi_test_unit_ready - test if unit is ready 2181 * @sdev: scsi device to change the state of. 2182 * @timeout: command timeout 2183 * @retries: number of retries before failing 2184 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for 2185 * returning sense. Make sure that this is cleared before passing 2186 * in. 2187 * 2188 * Returns zero if unsuccessful or an error if TUR failed. For 2189 * removable media, a return of NOT_READY or UNIT_ATTENTION is 2190 * translated to success, with the ->changed flag updated. 2191 **/ 2192int 2193scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries, 2194 struct scsi_sense_hdr *sshdr_external) 2195{ 2196 char cmd[] = { 2197 TEST_UNIT_READY, 0, 0, 0, 0, 0, 2198 }; 2199 struct scsi_sense_hdr *sshdr; 2200 int result; 2201 2202 if (!sshdr_external) 2203 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL); 2204 else 2205 sshdr = sshdr_external; 2206 2207 /* try to eat the UNIT_ATTENTION if there are enough retries */ 2208 do { 2209 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr, 2210 timeout, retries, NULL); 2211 if (sdev->removable && scsi_sense_valid(sshdr) && 2212 sshdr->sense_key == UNIT_ATTENTION) 2213 sdev->changed = 1; 2214 } while (scsi_sense_valid(sshdr) && 2215 sshdr->sense_key == UNIT_ATTENTION && --retries); 2216 2217 if (!sshdr) 2218 /* could not allocate sense buffer, so can't process it */ 2219 return result; 2220 2221 if (sdev->removable && scsi_sense_valid(sshdr) && 2222 (sshdr->sense_key == UNIT_ATTENTION || 2223 sshdr->sense_key == NOT_READY)) { 2224 sdev->changed = 1; 2225 result = 0; 2226 } 2227 if (!sshdr_external) 2228 kfree(sshdr); 2229 return result; 2230} 2231EXPORT_SYMBOL(scsi_test_unit_ready); 2232 2233/** 2234 * scsi_device_set_state - Take the given device through the device state model. 2235 * @sdev: scsi device to change the state of. 2236 * @state: state to change to. 2237 * 2238 * Returns zero if unsuccessful or an error if the requested 2239 * transition is illegal. 2240 */ 2241int 2242scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state) 2243{ 2244 enum scsi_device_state oldstate = sdev->sdev_state; 2245 2246 if (state == oldstate) 2247 return 0; 2248 2249 switch (state) { 2250 case SDEV_CREATED: 2251 switch (oldstate) { 2252 case SDEV_CREATED_BLOCK: 2253 break; 2254 default: 2255 goto illegal; 2256 } 2257 break; 2258 2259 case SDEV_RUNNING: 2260 switch (oldstate) { 2261 case SDEV_CREATED: 2262 case SDEV_OFFLINE: 2263 case SDEV_QUIESCE: 2264 case SDEV_BLOCK: 2265 break; 2266 default: 2267 goto illegal; 2268 } 2269 break; 2270 2271 case SDEV_QUIESCE: 2272 switch (oldstate) { 2273 case SDEV_RUNNING: 2274 case SDEV_OFFLINE: 2275 break; 2276 default: 2277 goto illegal; 2278 } 2279 break; 2280 2281 case SDEV_OFFLINE: 2282 switch (oldstate) { 2283 case SDEV_CREATED: 2284 case SDEV_RUNNING: 2285 case SDEV_QUIESCE: 2286 case SDEV_BLOCK: 2287 break; 2288 default: 2289 goto illegal; 2290 } 2291 break; 2292 2293 case SDEV_BLOCK: 2294 switch (oldstate) { 2295 case SDEV_RUNNING: 2296 case SDEV_CREATED_BLOCK: 2297 break; 2298 default: 2299 goto illegal; 2300 } 2301 break; 2302 2303 case SDEV_CREATED_BLOCK: 2304 switch (oldstate) { 2305 case SDEV_CREATED: 2306 break; 2307 default: 2308 goto illegal; 2309 } 2310 break; 2311 2312 case SDEV_CANCEL: 2313 switch (oldstate) { 2314 case SDEV_CREATED: 2315 case SDEV_RUNNING: 2316 case SDEV_QUIESCE: 2317 case SDEV_OFFLINE: 2318 case SDEV_BLOCK: 2319 break; 2320 default: 2321 goto illegal; 2322 } 2323 break; 2324 2325 case SDEV_DEL: 2326 switch (oldstate) { 2327 case SDEV_CREATED: 2328 case SDEV_RUNNING: 2329 case SDEV_OFFLINE: 2330 case SDEV_CANCEL: 2331 break; 2332 default: 2333 goto illegal; 2334 } 2335 break; 2336 2337 } 2338 sdev->sdev_state = state; 2339 return 0; 2340 2341 illegal: 2342 SCSI_LOG_ERROR_RECOVERY(1, 2343 sdev_printk(KERN_ERR, sdev, 2344 "Illegal state transition %s->%s\n", 2345 scsi_device_state_name(oldstate), 2346 scsi_device_state_name(state)) 2347 ); 2348 return -EINVAL; 2349} 2350EXPORT_SYMBOL(scsi_device_set_state); 2351 2352/** 2353 * sdev_evt_emit - emit a single SCSI device uevent 2354 * @sdev: associated SCSI device 2355 * @evt: event to emit 2356 * 2357 * Send a single uevent (scsi_event) to the associated scsi_device. 2358 */ 2359static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt) 2360{ 2361 int idx = 0; 2362 char *envp[3]; 2363 2364 switch (evt->evt_type) { 2365 case SDEV_EVT_MEDIA_CHANGE: 2366 envp[idx++] = "SDEV_MEDIA_CHANGE=1"; 2367 break; 2368 2369 default: 2370 /* do nothing */ 2371 break; 2372 } 2373 2374 envp[idx++] = NULL; 2375 2376 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp); 2377} 2378 2379/** 2380 * sdev_evt_thread - send a uevent for each scsi event 2381 * @work: work struct for scsi_device 2382 * 2383 * Dispatch queued events to their associated scsi_device kobjects 2384 * as uevents. 2385 */ 2386void scsi_evt_thread(struct work_struct *work) 2387{ 2388 struct scsi_device *sdev; 2389 LIST_HEAD(event_list); 2390 2391 sdev = container_of(work, struct scsi_device, event_work); 2392 2393 while (1) { 2394 struct scsi_event *evt; 2395 struct list_head *this, *tmp; 2396 unsigned long flags; 2397 2398 spin_lock_irqsave(&sdev->list_lock, flags); 2399 list_splice_init(&sdev->event_list, &event_list); 2400 spin_unlock_irqrestore(&sdev->list_lock, flags); 2401 2402 if (list_empty(&event_list)) 2403 break; 2404 2405 list_for_each_safe(this, tmp, &event_list) { 2406 evt = list_entry(this, struct scsi_event, node); 2407 list_del(&evt->node); 2408 scsi_evt_emit(sdev, evt); 2409 kfree(evt); 2410 } 2411 } 2412} 2413 2414/** 2415 * sdev_evt_send - send asserted event to uevent thread 2416 * @sdev: scsi_device event occurred on 2417 * @evt: event to send 2418 * 2419 * Assert scsi device event asynchronously. 2420 */ 2421void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt) 2422{ 2423 unsigned long flags; 2424 2425#if 0 2426 /* FIXME: currently this check eliminates all media change events 2427 * for polled devices. Need to update to discriminate between AN 2428 * and polled events */ 2429 if (!test_bit(evt->evt_type, sdev->supported_events)) { 2430 kfree(evt); 2431 return; 2432 } 2433#endif 2434 2435 spin_lock_irqsave(&sdev->list_lock, flags); 2436 list_add_tail(&evt->node, &sdev->event_list); 2437 schedule_work(&sdev->event_work); 2438 spin_unlock_irqrestore(&sdev->list_lock, flags); 2439} 2440EXPORT_SYMBOL_GPL(sdev_evt_send); 2441 2442/** 2443 * sdev_evt_alloc - allocate a new scsi event 2444 * @evt_type: type of event to allocate 2445 * @gfpflags: GFP flags for allocation 2446 * 2447 * Allocates and returns a new scsi_event. 2448 */ 2449struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type, 2450 gfp_t gfpflags) 2451{ 2452 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags); 2453 if (!evt) 2454 return NULL; 2455 2456 evt->evt_type = evt_type; 2457 INIT_LIST_HEAD(&evt->node); 2458 2459 /* evt_type-specific initialization, if any */ 2460 switch (evt_type) { 2461 case SDEV_EVT_MEDIA_CHANGE: 2462 default: 2463 /* do nothing */ 2464 break; 2465 } 2466 2467 return evt; 2468} 2469EXPORT_SYMBOL_GPL(sdev_evt_alloc); 2470 2471/** 2472 * sdev_evt_send_simple - send asserted event to uevent thread 2473 * @sdev: scsi_device event occurred on 2474 * @evt_type: type of event to send 2475 * @gfpflags: GFP flags for allocation 2476 * 2477 * Assert scsi device event asynchronously, given an event type. 2478 */ 2479void sdev_evt_send_simple(struct scsi_device *sdev, 2480 enum scsi_device_event evt_type, gfp_t gfpflags) 2481{ 2482 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags); 2483 if (!evt) { 2484 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n", 2485 evt_type); 2486 return; 2487 } 2488 2489 sdev_evt_send(sdev, evt); 2490} 2491EXPORT_SYMBOL_GPL(sdev_evt_send_simple); 2492 2493/** 2494 * scsi_device_quiesce - Block user issued commands. 2495 * @sdev: scsi device to quiesce. 2496 * 2497 * This works by trying to transition to the SDEV_QUIESCE state 2498 * (which must be a legal transition). When the device is in this 2499 * state, only special requests will be accepted, all others will 2500 * be deferred. Since special requests may also be requeued requests, 2501 * a successful return doesn't guarantee the device will be 2502 * totally quiescent. 2503 * 2504 * Must be called with user context, may sleep. 2505 * 2506 * Returns zero if unsuccessful or an error if not. 2507 */ 2508int 2509scsi_device_quiesce(struct scsi_device *sdev) 2510{ 2511 int err = scsi_device_set_state(sdev, SDEV_QUIESCE); 2512 if (err) 2513 return err; 2514 2515 scsi_run_queue(sdev->request_queue); 2516 while (sdev->device_busy) { 2517 msleep_interruptible(200); 2518 scsi_run_queue(sdev->request_queue); 2519 } 2520 return 0; 2521} 2522EXPORT_SYMBOL(scsi_device_quiesce); 2523 2524/** 2525 * scsi_device_resume - Restart user issued commands to a quiesced device. 2526 * @sdev: scsi device to resume. 2527 * 2528 * Moves the device from quiesced back to running and restarts the 2529 * queues. 2530 * 2531 * Must be called with user context, may sleep. 2532 */ 2533void 2534scsi_device_resume(struct scsi_device *sdev) 2535{ 2536 if(scsi_device_set_state(sdev, SDEV_RUNNING)) 2537 return; 2538 scsi_run_queue(sdev->request_queue); 2539} 2540EXPORT_SYMBOL(scsi_device_resume); 2541 2542static void 2543device_quiesce_fn(struct scsi_device *sdev, void *data) 2544{ 2545 scsi_device_quiesce(sdev); 2546} 2547 2548void 2549scsi_target_quiesce(struct scsi_target *starget) 2550{ 2551 starget_for_each_device(starget, NULL, device_quiesce_fn); 2552} 2553EXPORT_SYMBOL(scsi_target_quiesce); 2554 2555static void 2556device_resume_fn(struct scsi_device *sdev, void *data) 2557{ 2558 scsi_device_resume(sdev); 2559} 2560 2561void 2562scsi_target_resume(struct scsi_target *starget) 2563{ 2564 starget_for_each_device(starget, NULL, device_resume_fn); 2565} 2566EXPORT_SYMBOL(scsi_target_resume); 2567 2568/** 2569 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state 2570 * @sdev: device to block 2571 * 2572 * Block request made by scsi lld's to temporarily stop all 2573 * scsi commands on the specified device. Called from interrupt 2574 * or normal process context. 2575 * 2576 * Returns zero if successful or error if not 2577 * 2578 * Notes: 2579 * This routine transitions the device to the SDEV_BLOCK state 2580 * (which must be a legal transition). When the device is in this 2581 * state, all commands are deferred until the scsi lld reenables 2582 * the device with scsi_device_unblock or device_block_tmo fires. 2583 * This routine assumes the host_lock is held on entry. 2584 */ 2585int 2586scsi_internal_device_block(struct scsi_device *sdev) 2587{ 2588 struct request_queue *q = sdev->request_queue; 2589 unsigned long flags; 2590 int err = 0; 2591 2592 err = scsi_device_set_state(sdev, SDEV_BLOCK); 2593 if (err) { 2594 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK); 2595 2596 if (err) 2597 return err; 2598 } 2599 2600 /* 2601 * The device has transitioned to SDEV_BLOCK. Stop the 2602 * block layer from calling the midlayer with this device's 2603 * request queue. 2604 */ 2605 spin_lock_irqsave(q->queue_lock, flags); 2606 blk_stop_queue(q); 2607 spin_unlock_irqrestore(q->queue_lock, flags); 2608 2609 return 0; 2610} 2611EXPORT_SYMBOL_GPL(scsi_internal_device_block); 2612 2613/** 2614 * scsi_internal_device_unblock - resume a device after a block request 2615 * @sdev: device to resume 2616 * 2617 * Called by scsi lld's or the midlayer to restart the device queue 2618 * for the previously suspended scsi device. Called from interrupt or 2619 * normal process context. 2620 * 2621 * Returns zero if successful or error if not. 2622 * 2623 * Notes: 2624 * This routine transitions the device to the SDEV_RUNNING state 2625 * (which must be a legal transition) allowing the midlayer to 2626 * goose the queue for this device. This routine assumes the 2627 * host_lock is held upon entry. 2628 */ 2629int 2630scsi_internal_device_unblock(struct scsi_device *sdev) 2631{ 2632 struct request_queue *q = sdev->request_queue; 2633 int err; 2634 unsigned long flags; 2635 2636 /* 2637 * Try to transition the scsi device to SDEV_RUNNING 2638 * and goose the device queue if successful. 2639 */ 2640 err = scsi_device_set_state(sdev, SDEV_RUNNING); 2641 if (err) { 2642 err = scsi_device_set_state(sdev, SDEV_CREATED); 2643 2644 if (err) 2645 return err; 2646 } 2647 2648 spin_lock_irqsave(q->queue_lock, flags); 2649 blk_start_queue(q); 2650 spin_unlock_irqrestore(q->queue_lock, flags); 2651 2652 return 0; 2653} 2654EXPORT_SYMBOL_GPL(scsi_internal_device_unblock); 2655 2656static void 2657device_block(struct scsi_device *sdev, void *data) 2658{ 2659 scsi_internal_device_block(sdev); 2660} 2661 2662static int 2663target_block(struct device *dev, void *data) 2664{ 2665 if (scsi_is_target_device(dev)) 2666 starget_for_each_device(to_scsi_target(dev), NULL, 2667 device_block); 2668 return 0; 2669} 2670 2671void 2672scsi_target_block(struct device *dev) 2673{ 2674 if (scsi_is_target_device(dev)) 2675 starget_for_each_device(to_scsi_target(dev), NULL, 2676 device_block); 2677 else 2678 device_for_each_child(dev, NULL, target_block); 2679} 2680EXPORT_SYMBOL_GPL(scsi_target_block); 2681 2682static void 2683device_unblock(struct scsi_device *sdev, void *data) 2684{ 2685 scsi_internal_device_unblock(sdev); 2686} 2687 2688static int 2689target_unblock(struct device *dev, void *data) 2690{ 2691 if (scsi_is_target_device(dev)) 2692 starget_for_each_device(to_scsi_target(dev), NULL, 2693 device_unblock); 2694 return 0; 2695} 2696 2697void 2698scsi_target_unblock(struct device *dev) 2699{ 2700 if (scsi_is_target_device(dev)) 2701 starget_for_each_device(to_scsi_target(dev), NULL, 2702 device_unblock); 2703 else 2704 device_for_each_child(dev, NULL, target_unblock); 2705} 2706EXPORT_SYMBOL_GPL(scsi_target_unblock); 2707 2708/** 2709 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt 2710 * @sgl: scatter-gather list 2711 * @sg_count: number of segments in sg 2712 * @offset: offset in bytes into sg, on return offset into the mapped area 2713 * @len: bytes to map, on return number of bytes mapped 2714 * 2715 * Returns virtual address of the start of the mapped page 2716 */ 2717void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count, 2718 size_t *offset, size_t *len) 2719{ 2720 int i; 2721 size_t sg_len = 0, len_complete = 0; 2722 struct scatterlist *sg; 2723 struct page *page; 2724 2725 WARN_ON(!irqs_disabled()); 2726 2727 for_each_sg(sgl, sg, sg_count, i) { 2728 len_complete = sg_len; /* Complete sg-entries */ 2729 sg_len += sg->length; 2730 if (sg_len > *offset) 2731 break; 2732 } 2733 2734 if (unlikely(i == sg_count)) { 2735 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, " 2736 "elements %d\n", 2737 __func__, sg_len, *offset, sg_count); 2738 WARN_ON(1); 2739 return NULL; 2740 } 2741 2742 /* Offset starting from the beginning of first page in this sg-entry */ 2743 *offset = *offset - len_complete + sg->offset; 2744 2745 /* Assumption: contiguous pages can be accessed as "page + i" */ 2746 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT)); 2747 *offset &= ~PAGE_MASK; 2748 2749 /* Bytes in this sg-entry from *offset to the end of the page */ 2750 sg_len = PAGE_SIZE - *offset; 2751 if (*len > sg_len) 2752 *len = sg_len; 2753 2754 return kmap_atomic(page, KM_BIO_SRC_IRQ); 2755} 2756EXPORT_SYMBOL(scsi_kmap_atomic_sg); 2757 2758/** 2759 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg 2760 * @virt: virtual address to be unmapped 2761 */ 2762void scsi_kunmap_atomic_sg(void *virt) 2763{ 2764 kunmap_atomic(virt, KM_BIO_SRC_IRQ); 2765} 2766EXPORT_SYMBOL(scsi_kunmap_atomic_sg); 2767