linux/drivers/scsi/scsi_lib.c
<<
>>
Prefs
   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/export.h>
  16#include <linux/mempool.h>
  17#include <linux/slab.h>
  18#include <linux/init.h>
  19#include <linux/pci.h>
  20#include <linux/delay.h>
  21#include <linux/hardirq.h>
  22#include <linux/scatterlist.h>
  23
  24#include <scsi/scsi.h>
  25#include <scsi/scsi_cmnd.h>
  26#include <scsi/scsi_dbg.h>
  27#include <scsi/scsi_device.h>
  28#include <scsi/scsi_driver.h>
  29#include <scsi/scsi_eh.h>
  30#include <scsi/scsi_host.h>
  31
  32#include "scsi_priv.h"
  33#include "scsi_logging.h"
  34
  35
  36#define SG_MEMPOOL_NR           ARRAY_SIZE(scsi_sg_pools)
  37#define SG_MEMPOOL_SIZE         2
  38
  39struct scsi_host_sg_pool {
  40        size_t          size;
  41        char            *name;
  42        struct kmem_cache       *slab;
  43        mempool_t       *pool;
  44};
  45
  46#define SP(x) { x, "sgpool-" __stringify(x) }
  47#if (SCSI_MAX_SG_SEGMENTS < 32)
  48#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
  49#endif
  50static struct scsi_host_sg_pool scsi_sg_pools[] = {
  51        SP(8),
  52        SP(16),
  53#if (SCSI_MAX_SG_SEGMENTS > 32)
  54        SP(32),
  55#if (SCSI_MAX_SG_SEGMENTS > 64)
  56        SP(64),
  57#if (SCSI_MAX_SG_SEGMENTS > 128)
  58        SP(128),
  59#if (SCSI_MAX_SG_SEGMENTS > 256)
  60#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
  61#endif
  62#endif
  63#endif
  64#endif
  65        SP(SCSI_MAX_SG_SEGMENTS)
  66};
  67#undef SP
  68
  69struct kmem_cache *scsi_sdb_cache;
  70
  71#ifdef CONFIG_ACPI
  72#include <acpi/acpi_bus.h>
  73
  74static bool acpi_scsi_bus_match(struct device *dev)
  75{
  76        return dev->bus == &scsi_bus_type;
  77}
  78
  79int scsi_register_acpi_bus_type(struct acpi_bus_type *bus)
  80{
  81        bus->match = acpi_scsi_bus_match;
  82        return register_acpi_bus_type(bus);
  83}
  84EXPORT_SYMBOL_GPL(scsi_register_acpi_bus_type);
  85
  86void scsi_unregister_acpi_bus_type(struct acpi_bus_type *bus)
  87{
  88        unregister_acpi_bus_type(bus);
  89}
  90EXPORT_SYMBOL_GPL(scsi_unregister_acpi_bus_type);
  91#endif
  92
  93/*
  94 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
  95 * not change behaviour from the previous unplug mechanism, experimentation
  96 * may prove this needs changing.
  97 */
  98#define SCSI_QUEUE_DELAY        3
  99
 100/*
 101 * Function:    scsi_unprep_request()
 102 *
 103 * Purpose:     Remove all preparation done for a request, including its
 104 *              associated scsi_cmnd, so that it can be requeued.
 105 *
 106 * Arguments:   req     - request to unprepare
 107 *
 108 * Lock status: Assumed that no locks are held upon entry.
 109 *
 110 * Returns:     Nothing.
 111 */
 112static void scsi_unprep_request(struct request *req)
 113{
 114        struct scsi_cmnd *cmd = req->special;
 115
 116        blk_unprep_request(req);
 117        req->special = NULL;
 118
 119        scsi_put_command(cmd);
 120}
 121
 122/**
 123 * __scsi_queue_insert - private queue insertion
 124 * @cmd: The SCSI command being requeued
 125 * @reason:  The reason for the requeue
 126 * @unbusy: Whether the queue should be unbusied
 127 *
 128 * This is a private queue insertion.  The public interface
 129 * scsi_queue_insert() always assumes the queue should be unbusied
 130 * because it's always called before the completion.  This function is
 131 * for a requeue after completion, which should only occur in this
 132 * file.
 133 */
 134static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
 135{
 136        struct Scsi_Host *host = cmd->device->host;
 137        struct scsi_device *device = cmd->device;
 138        struct scsi_target *starget = scsi_target(device);
 139        struct request_queue *q = device->request_queue;
 140        unsigned long flags;
 141
 142        SCSI_LOG_MLQUEUE(1,
 143                 printk("Inserting command %p into mlqueue\n", cmd));
 144
 145        /*
 146         * Set the appropriate busy bit for the device/host.
 147         *
 148         * If the host/device isn't busy, assume that something actually
 149         * completed, and that we should be able to queue a command now.
 150         *
 151         * Note that the prior mid-layer assumption that any host could
 152         * always queue at least one command is now broken.  The mid-layer
 153         * will implement a user specifiable stall (see
 154         * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
 155         * if a command is requeued with no other commands outstanding
 156         * either for the device or for the host.
 157         */
 158        switch (reason) {
 159        case SCSI_MLQUEUE_HOST_BUSY:
 160                host->host_blocked = host->max_host_blocked;
 161                break;
 162        case SCSI_MLQUEUE_DEVICE_BUSY:
 163        case SCSI_MLQUEUE_EH_RETRY:
 164                device->device_blocked = device->max_device_blocked;
 165                break;
 166        case SCSI_MLQUEUE_TARGET_BUSY:
 167                starget->target_blocked = starget->max_target_blocked;
 168                break;
 169        }
 170
 171        /*
 172         * Decrement the counters, since these commands are no longer
 173         * active on the host/device.
 174         */
 175        if (unbusy)
 176                scsi_device_unbusy(device);
 177
 178        /*
 179         * Requeue this command.  It will go before all other commands
 180         * that are already in the queue. Schedule requeue work under
 181         * lock such that the kblockd_schedule_work() call happens
 182         * before blk_cleanup_queue() finishes.
 183         */
 184        spin_lock_irqsave(q->queue_lock, flags);
 185        blk_requeue_request(q, cmd->request);
 186        kblockd_schedule_work(q, &device->requeue_work);
 187        spin_unlock_irqrestore(q->queue_lock, flags);
 188}
 189
 190/*
 191 * Function:    scsi_queue_insert()
 192 *
 193 * Purpose:     Insert a command in the midlevel queue.
 194 *
 195 * Arguments:   cmd    - command that we are adding to queue.
 196 *              reason - why we are inserting command to queue.
 197 *
 198 * Lock status: Assumed that lock is not held upon entry.
 199 *
 200 * Returns:     Nothing.
 201 *
 202 * Notes:       We do this for one of two cases.  Either the host is busy
 203 *              and it cannot accept any more commands for the time being,
 204 *              or the device returned QUEUE_FULL and can accept no more
 205 *              commands.
 206 * Notes:       This could be called either from an interrupt context or a
 207 *              normal process context.
 208 */
 209void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
 210{
 211        __scsi_queue_insert(cmd, reason, 1);
 212}
 213/**
 214 * scsi_execute - insert request and wait for the result
 215 * @sdev:       scsi device
 216 * @cmd:        scsi command
 217 * @data_direction: data direction
 218 * @buffer:     data buffer
 219 * @bufflen:    len of buffer
 220 * @sense:      optional sense buffer
 221 * @timeout:    request timeout in seconds
 222 * @retries:    number of times to retry request
 223 * @flags:      or into request flags;
 224 * @resid:      optional residual length
 225 *
 226 * returns the req->errors value which is the scsi_cmnd result
 227 * field.
 228 */
 229int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
 230                 int data_direction, void *buffer, unsigned bufflen,
 231                 unsigned char *sense, int timeout, int retries, int flags,
 232                 int *resid)
 233{
 234        struct request *req;
 235        int write = (data_direction == DMA_TO_DEVICE);
 236        int ret = DRIVER_ERROR << 24;
 237
 238        req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
 239        if (!req)
 240                return ret;
 241
 242        if (bufflen &&  blk_rq_map_kern(sdev->request_queue, req,
 243                                        buffer, bufflen, __GFP_WAIT))
 244                goto out;
 245
 246        req->cmd_len = COMMAND_SIZE(cmd[0]);
 247        memcpy(req->cmd, cmd, req->cmd_len);
 248        req->sense = sense;
 249        req->sense_len = 0;
 250        req->retries = retries;
 251        req->timeout = timeout;
 252        req->cmd_type = REQ_TYPE_BLOCK_PC;
 253        req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
 254
 255        /*
 256         * head injection *required* here otherwise quiesce won't work
 257         */
 258        blk_execute_rq(req->q, NULL, req, 1);
 259
 260        /*
 261         * Some devices (USB mass-storage in particular) may transfer
 262         * garbage data together with a residue indicating that the data
 263         * is invalid.  Prevent the garbage from being misinterpreted
 264         * and prevent security leaks by zeroing out the excess data.
 265         */
 266        if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
 267                memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
 268
 269        if (resid)
 270                *resid = req->resid_len;
 271        ret = req->errors;
 272 out:
 273        blk_put_request(req);
 274
 275        return ret;
 276}
 277EXPORT_SYMBOL(scsi_execute);
 278
 279
 280int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
 281                     int data_direction, void *buffer, unsigned bufflen,
 282                     struct scsi_sense_hdr *sshdr, int timeout, int retries,
 283                     int *resid)
 284{
 285        char *sense = NULL;
 286        int result;
 287        
 288        if (sshdr) {
 289                sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
 290                if (!sense)
 291                        return DRIVER_ERROR << 24;
 292        }
 293        result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
 294                              sense, timeout, retries, 0, resid);
 295        if (sshdr)
 296                scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
 297
 298        kfree(sense);
 299        return result;
 300}
 301EXPORT_SYMBOL(scsi_execute_req);
 302
 303/*
 304 * Function:    scsi_init_cmd_errh()
 305 *
 306 * Purpose:     Initialize cmd fields related to error handling.
 307 *
 308 * Arguments:   cmd     - command that is ready to be queued.
 309 *
 310 * Notes:       This function has the job of initializing a number of
 311 *              fields related to error handling.   Typically this will
 312 *              be called once for each command, as required.
 313 */
 314static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
 315{
 316        cmd->serial_number = 0;
 317        scsi_set_resid(cmd, 0);
 318        memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 319        if (cmd->cmd_len == 0)
 320                cmd->cmd_len = scsi_command_size(cmd->cmnd);
 321}
 322
 323void scsi_device_unbusy(struct scsi_device *sdev)
 324{
 325        struct Scsi_Host *shost = sdev->host;
 326        struct scsi_target *starget = scsi_target(sdev);
 327        unsigned long flags;
 328
 329        spin_lock_irqsave(shost->host_lock, flags);
 330        shost->host_busy--;
 331        starget->target_busy--;
 332        if (unlikely(scsi_host_in_recovery(shost) &&
 333                     (shost->host_failed || shost->host_eh_scheduled)))
 334                scsi_eh_wakeup(shost);
 335        spin_unlock(shost->host_lock);
 336        spin_lock(sdev->request_queue->queue_lock);
 337        sdev->device_busy--;
 338        spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
 339}
 340
 341/*
 342 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
 343 * and call blk_run_queue for all the scsi_devices on the target -
 344 * including current_sdev first.
 345 *
 346 * Called with *no* scsi locks held.
 347 */
 348static void scsi_single_lun_run(struct scsi_device *current_sdev)
 349{
 350        struct Scsi_Host *shost = current_sdev->host;
 351        struct scsi_device *sdev, *tmp;
 352        struct scsi_target *starget = scsi_target(current_sdev);
 353        unsigned long flags;
 354
 355        spin_lock_irqsave(shost->host_lock, flags);
 356        starget->starget_sdev_user = NULL;
 357        spin_unlock_irqrestore(shost->host_lock, flags);
 358
 359        /*
 360         * Call blk_run_queue for all LUNs on the target, starting with
 361         * current_sdev. We race with others (to set starget_sdev_user),
 362         * but in most cases, we will be first. Ideally, each LU on the
 363         * target would get some limited time or requests on the target.
 364         */
 365        blk_run_queue(current_sdev->request_queue);
 366
 367        spin_lock_irqsave(shost->host_lock, flags);
 368        if (starget->starget_sdev_user)
 369                goto out;
 370        list_for_each_entry_safe(sdev, tmp, &starget->devices,
 371                        same_target_siblings) {
 372                if (sdev == current_sdev)
 373                        continue;
 374                if (scsi_device_get(sdev))
 375                        continue;
 376
 377                spin_unlock_irqrestore(shost->host_lock, flags);
 378                blk_run_queue(sdev->request_queue);
 379                spin_lock_irqsave(shost->host_lock, flags);
 380        
 381                scsi_device_put(sdev);
 382        }
 383 out:
 384        spin_unlock_irqrestore(shost->host_lock, flags);
 385}
 386
 387static inline int scsi_device_is_busy(struct scsi_device *sdev)
 388{
 389        if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
 390                return 1;
 391
 392        return 0;
 393}
 394
 395static inline int scsi_target_is_busy(struct scsi_target *starget)
 396{
 397        return ((starget->can_queue > 0 &&
 398                 starget->target_busy >= starget->can_queue) ||
 399                 starget->target_blocked);
 400}
 401
 402static inline int scsi_host_is_busy(struct Scsi_Host *shost)
 403{
 404        if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
 405            shost->host_blocked || shost->host_self_blocked)
 406                return 1;
 407
 408        return 0;
 409}
 410
 411/*
 412 * Function:    scsi_run_queue()
 413 *
 414 * Purpose:     Select a proper request queue to serve next
 415 *
 416 * Arguments:   q       - last request's queue
 417 *
 418 * Returns:     Nothing
 419 *
 420 * Notes:       The previous command was completely finished, start
 421 *              a new one if possible.
 422 */
 423static void scsi_run_queue(struct request_queue *q)
 424{
 425        struct scsi_device *sdev = q->queuedata;
 426        struct Scsi_Host *shost;
 427        LIST_HEAD(starved_list);
 428        unsigned long flags;
 429
 430        shost = sdev->host;
 431        if (scsi_target(sdev)->single_lun)
 432                scsi_single_lun_run(sdev);
 433
 434        spin_lock_irqsave(shost->host_lock, flags);
 435        list_splice_init(&shost->starved_list, &starved_list);
 436
 437        while (!list_empty(&starved_list)) {
 438                /*
 439                 * As long as shost is accepting commands and we have
 440                 * starved queues, call blk_run_queue. scsi_request_fn
 441                 * drops the queue_lock and can add us back to the
 442                 * starved_list.
 443                 *
 444                 * host_lock protects the starved_list and starved_entry.
 445                 * scsi_request_fn must get the host_lock before checking
 446                 * or modifying starved_list or starved_entry.
 447                 */
 448                if (scsi_host_is_busy(shost))
 449                        break;
 450
 451                sdev = list_entry(starved_list.next,
 452                                  struct scsi_device, starved_entry);
 453                list_del_init(&sdev->starved_entry);
 454                if (scsi_target_is_busy(scsi_target(sdev))) {
 455                        list_move_tail(&sdev->starved_entry,
 456                                       &shost->starved_list);
 457                        continue;
 458                }
 459
 460                spin_unlock(shost->host_lock);
 461                spin_lock(sdev->request_queue->queue_lock);
 462                __blk_run_queue(sdev->request_queue);
 463                spin_unlock(sdev->request_queue->queue_lock);
 464                spin_lock(shost->host_lock);
 465        }
 466        /* put any unprocessed entries back */
 467        list_splice(&starved_list, &shost->starved_list);
 468        spin_unlock_irqrestore(shost->host_lock, flags);
 469
 470        blk_run_queue(q);
 471}
 472
 473void scsi_requeue_run_queue(struct work_struct *work)
 474{
 475        struct scsi_device *sdev;
 476        struct request_queue *q;
 477
 478        sdev = container_of(work, struct scsi_device, requeue_work);
 479        q = sdev->request_queue;
 480        scsi_run_queue(q);
 481}
 482
 483/*
 484 * Function:    scsi_requeue_command()
 485 *
 486 * Purpose:     Handle post-processing of completed commands.
 487 *
 488 * Arguments:   q       - queue to operate on
 489 *              cmd     - command that may need to be requeued.
 490 *
 491 * Returns:     Nothing
 492 *
 493 * Notes:       After command completion, there may be blocks left
 494 *              over which weren't finished by the previous command
 495 *              this can be for a number of reasons - the main one is
 496 *              I/O errors in the middle of the request, in which case
 497 *              we need to request the blocks that come after the bad
 498 *              sector.
 499 * Notes:       Upon return, cmd is a stale pointer.
 500 */
 501static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
 502{
 503        struct scsi_device *sdev = cmd->device;
 504        struct request *req = cmd->request;
 505        unsigned long flags;
 506
 507        /*
 508         * We need to hold a reference on the device to avoid the queue being
 509         * killed after the unlock and before scsi_run_queue is invoked which
 510         * may happen because scsi_unprep_request() puts the command which
 511         * releases its reference on the device.
 512         */
 513        get_device(&sdev->sdev_gendev);
 514
 515        spin_lock_irqsave(q->queue_lock, flags);
 516        scsi_unprep_request(req);
 517        blk_requeue_request(q, req);
 518        spin_unlock_irqrestore(q->queue_lock, flags);
 519
 520        scsi_run_queue(q);
 521
 522        put_device(&sdev->sdev_gendev);
 523}
 524
 525void scsi_next_command(struct scsi_cmnd *cmd)
 526{
 527        struct scsi_device *sdev = cmd->device;
 528        struct request_queue *q = sdev->request_queue;
 529
 530        /* need to hold a reference on the device before we let go of the cmd */
 531        get_device(&sdev->sdev_gendev);
 532
 533        scsi_put_command(cmd);
 534        scsi_run_queue(q);
 535
 536        /* ok to remove device now */
 537        put_device(&sdev->sdev_gendev);
 538}
 539
 540void scsi_run_host_queues(struct Scsi_Host *shost)
 541{
 542        struct scsi_device *sdev;
 543
 544        shost_for_each_device(sdev, shost)
 545                scsi_run_queue(sdev->request_queue);
 546}
 547
 548static void __scsi_release_buffers(struct scsi_cmnd *, int);
 549
 550/*
 551 * Function:    scsi_end_request()
 552 *
 553 * Purpose:     Post-processing of completed commands (usually invoked at end
 554 *              of upper level post-processing and scsi_io_completion).
 555 *
 556 * Arguments:   cmd      - command that is complete.
 557 *              error    - 0 if I/O indicates success, < 0 for I/O error.
 558 *              bytes    - number of bytes of completed I/O
 559 *              requeue  - indicates whether we should requeue leftovers.
 560 *
 561 * Lock status: Assumed that lock is not held upon entry.
 562 *
 563 * Returns:     cmd if requeue required, NULL otherwise.
 564 *
 565 * Notes:       This is called for block device requests in order to
 566 *              mark some number of sectors as complete.
 567 * 
 568 *              We are guaranteeing that the request queue will be goosed
 569 *              at some point during this call.
 570 * Notes:       If cmd was requeued, upon return it will be a stale pointer.
 571 */
 572static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
 573                                          int bytes, int requeue)
 574{
 575        struct request_queue *q = cmd->device->request_queue;
 576        struct request *req = cmd->request;
 577
 578        /*
 579         * If there are blocks left over at the end, set up the command
 580         * to queue the remainder of them.
 581         */
 582        if (blk_end_request(req, error, bytes)) {
 583                /* kill remainder if no retrys */
 584                if (error && scsi_noretry_cmd(cmd))
 585                        blk_end_request_all(req, error);
 586                else {
 587                        if (requeue) {
 588                                /*
 589                                 * Bleah.  Leftovers again.  Stick the
 590                                 * leftovers in the front of the
 591                                 * queue, and goose the queue again.
 592                                 */
 593                                scsi_release_buffers(cmd);
 594                                scsi_requeue_command(q, cmd);
 595                                cmd = NULL;
 596                        }
 597                        return cmd;
 598                }
 599        }
 600
 601        /*
 602         * This will goose the queue request function at the end, so we don't
 603         * need to worry about launching another command.
 604         */
 605        __scsi_release_buffers(cmd, 0);
 606        scsi_next_command(cmd);
 607        return NULL;
 608}
 609
 610static inline unsigned int scsi_sgtable_index(unsigned short nents)
 611{
 612        unsigned int index;
 613
 614        BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
 615
 616        if (nents <= 8)
 617                index = 0;
 618        else
 619                index = get_count_order(nents) - 3;
 620
 621        return index;
 622}
 623
 624static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
 625{
 626        struct scsi_host_sg_pool *sgp;
 627
 628        sgp = scsi_sg_pools + scsi_sgtable_index(nents);
 629        mempool_free(sgl, sgp->pool);
 630}
 631
 632static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
 633{
 634        struct scsi_host_sg_pool *sgp;
 635
 636        sgp = scsi_sg_pools + scsi_sgtable_index(nents);
 637        return mempool_alloc(sgp->pool, gfp_mask);
 638}
 639
 640static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
 641                              gfp_t gfp_mask)
 642{
 643        int ret;
 644
 645        BUG_ON(!nents);
 646
 647        ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
 648                               gfp_mask, scsi_sg_alloc);
 649        if (unlikely(ret))
 650                __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
 651                                scsi_sg_free);
 652
 653        return ret;
 654}
 655
 656static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
 657{
 658        __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
 659}
 660
 661static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
 662{
 663
 664        if (cmd->sdb.table.nents)
 665                scsi_free_sgtable(&cmd->sdb);
 666
 667        memset(&cmd->sdb, 0, sizeof(cmd->sdb));
 668
 669        if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
 670                struct scsi_data_buffer *bidi_sdb =
 671                        cmd->request->next_rq->special;
 672                scsi_free_sgtable(bidi_sdb);
 673                kmem_cache_free(scsi_sdb_cache, bidi_sdb);
 674                cmd->request->next_rq->special = NULL;
 675        }
 676
 677        if (scsi_prot_sg_count(cmd))
 678                scsi_free_sgtable(cmd->prot_sdb);
 679}
 680
 681/*
 682 * Function:    scsi_release_buffers()
 683 *
 684 * Purpose:     Completion processing for block device I/O requests.
 685 *
 686 * Arguments:   cmd     - command that we are bailing.
 687 *
 688 * Lock status: Assumed that no lock is held upon entry.
 689 *
 690 * Returns:     Nothing
 691 *
 692 * Notes:       In the event that an upper level driver rejects a
 693 *              command, we must release resources allocated during
 694 *              the __init_io() function.  Primarily this would involve
 695 *              the scatter-gather table, and potentially any bounce
 696 *              buffers.
 697 */
 698void scsi_release_buffers(struct scsi_cmnd *cmd)
 699{
 700        __scsi_release_buffers(cmd, 1);
 701}
 702EXPORT_SYMBOL(scsi_release_buffers);
 703
 704static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
 705{
 706        int error = 0;
 707
 708        switch(host_byte(result)) {
 709        case DID_TRANSPORT_FAILFAST:
 710                error = -ENOLINK;
 711                break;
 712        case DID_TARGET_FAILURE:
 713                set_host_byte(cmd, DID_OK);
 714                error = -EREMOTEIO;
 715                break;
 716        case DID_NEXUS_FAILURE:
 717                set_host_byte(cmd, DID_OK);
 718                error = -EBADE;
 719                break;
 720        default:
 721                error = -EIO;
 722                break;
 723        }
 724
 725        return error;
 726}
 727
 728/*
 729 * Function:    scsi_io_completion()
 730 *
 731 * Purpose:     Completion processing for block device I/O requests.
 732 *
 733 * Arguments:   cmd   - command that is finished.
 734 *
 735 * Lock status: Assumed that no lock is held upon entry.
 736 *
 737 * Returns:     Nothing
 738 *
 739 * Notes:       This function is matched in terms of capabilities to
 740 *              the function that created the scatter-gather list.
 741 *              In other words, if there are no bounce buffers
 742 *              (the normal case for most drivers), we don't need
 743 *              the logic to deal with cleaning up afterwards.
 744 *
 745 *              We must call scsi_end_request().  This will finish off
 746 *              the specified number of sectors.  If we are done, the
 747 *              command block will be released and the queue function
 748 *              will be goosed.  If we are not done then we have to
 749 *              figure out what to do next:
 750 *
 751 *              a) We can call scsi_requeue_command().  The request
 752 *                 will be unprepared and put back on the queue.  Then
 753 *                 a new command will be created for it.  This should
 754 *                 be used if we made forward progress, or if we want
 755 *                 to switch from READ(10) to READ(6) for example.
 756 *
 757 *              b) We can call scsi_queue_insert().  The request will
 758 *                 be put back on the queue and retried using the same
 759 *                 command as before, possibly after a delay.
 760 *
 761 *              c) We can call blk_end_request() with -EIO to fail
 762 *                 the remainder of the request.
 763 */
 764void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 765{
 766        int result = cmd->result;
 767        struct request_queue *q = cmd->device->request_queue;
 768        struct request *req = cmd->request;
 769        int error = 0;
 770        struct scsi_sense_hdr sshdr;
 771        int sense_valid = 0;
 772        int sense_deferred = 0;
 773        enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
 774              ACTION_DELAYED_RETRY} action;
 775        char *description = NULL;
 776
 777        if (result) {
 778                sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
 779                if (sense_valid)
 780                        sense_deferred = scsi_sense_is_deferred(&sshdr);
 781        }
 782
 783        if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
 784                if (result) {
 785                        if (sense_valid && req->sense) {
 786                                /*
 787                                 * SG_IO wants current and deferred errors
 788                                 */
 789                                int len = 8 + cmd->sense_buffer[7];
 790
 791                                if (len > SCSI_SENSE_BUFFERSIZE)
 792                                        len = SCSI_SENSE_BUFFERSIZE;
 793                                memcpy(req->sense, cmd->sense_buffer,  len);
 794                                req->sense_len = len;
 795                        }
 796                        if (!sense_deferred)
 797                                error = __scsi_error_from_host_byte(cmd, result);
 798                }
 799                /*
 800                 * __scsi_error_from_host_byte may have reset the host_byte
 801                 */
 802                req->errors = cmd->result;
 803
 804                req->resid_len = scsi_get_resid(cmd);
 805
 806                if (scsi_bidi_cmnd(cmd)) {
 807                        /*
 808                         * Bidi commands Must be complete as a whole,
 809                         * both sides at once.
 810                         */
 811                        req->next_rq->resid_len = scsi_in(cmd)->resid;
 812
 813                        scsi_release_buffers(cmd);
 814                        blk_end_request_all(req, 0);
 815
 816                        scsi_next_command(cmd);
 817                        return;
 818                }
 819        }
 820
 821        /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
 822        BUG_ON(blk_bidi_rq(req));
 823
 824        /*
 825         * Next deal with any sectors which we were able to correctly
 826         * handle.
 827         */
 828        SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
 829                                      "%d bytes done.\n",
 830                                      blk_rq_sectors(req), good_bytes));
 831
 832        /*
 833         * Recovered errors need reporting, but they're always treated
 834         * as success, so fiddle the result code here.  For BLOCK_PC
 835         * we already took a copy of the original into rq->errors which
 836         * is what gets returned to the user
 837         */
 838        if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
 839                /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
 840                 * print since caller wants ATA registers. Only occurs on
 841                 * SCSI ATA PASS_THROUGH commands when CK_COND=1
 842                 */
 843                if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
 844                        ;
 845                else if (!(req->cmd_flags & REQ_QUIET))
 846                        scsi_print_sense("", cmd);
 847                result = 0;
 848                /* BLOCK_PC may have set error */
 849                error = 0;
 850        }
 851
 852        /*
 853         * A number of bytes were successfully read.  If there
 854         * are leftovers and there is some kind of error
 855         * (result != 0), retry the rest.
 856         */
 857        if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
 858                return;
 859
 860        error = __scsi_error_from_host_byte(cmd, result);
 861
 862        if (host_byte(result) == DID_RESET) {
 863                /* Third party bus reset or reset for error recovery
 864                 * reasons.  Just retry the command and see what
 865                 * happens.
 866                 */
 867                action = ACTION_RETRY;
 868        } else if (sense_valid && !sense_deferred) {
 869                switch (sshdr.sense_key) {
 870                case UNIT_ATTENTION:
 871                        if (cmd->device->removable) {
 872                                /* Detected disc change.  Set a bit
 873                                 * and quietly refuse further access.
 874                                 */
 875                                cmd->device->changed = 1;
 876                                description = "Media Changed";
 877                                action = ACTION_FAIL;
 878                        } else {
 879                                /* Must have been a power glitch, or a
 880                                 * bus reset.  Could not have been a
 881                                 * media change, so we just retry the
 882                                 * command and see what happens.
 883                                 */
 884                                action = ACTION_RETRY;
 885                        }
 886                        break;
 887                case ILLEGAL_REQUEST:
 888                        /* If we had an ILLEGAL REQUEST returned, then
 889                         * we may have performed an unsupported
 890                         * command.  The only thing this should be
 891                         * would be a ten byte read where only a six
 892                         * byte read was supported.  Also, on a system
 893                         * where READ CAPACITY failed, we may have
 894                         * read past the end of the disk.
 895                         */
 896                        if ((cmd->device->use_10_for_rw &&
 897                            sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
 898                            (cmd->cmnd[0] == READ_10 ||
 899                             cmd->cmnd[0] == WRITE_10)) {
 900                                /* This will issue a new 6-byte command. */
 901                                cmd->device->use_10_for_rw = 0;
 902                                action = ACTION_REPREP;
 903                        } else if (sshdr.asc == 0x10) /* DIX */ {
 904                                description = "Host Data Integrity Failure";
 905                                action = ACTION_FAIL;
 906                                error = -EILSEQ;
 907                        /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
 908                        } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
 909                                switch (cmd->cmnd[0]) {
 910                                case UNMAP:
 911                                        description = "Discard failure";
 912                                        break;
 913                                case WRITE_SAME:
 914                                case WRITE_SAME_16:
 915                                        if (cmd->cmnd[1] & 0x8)
 916                                                description = "Discard failure";
 917                                        else
 918                                                description =
 919                                                        "Write same failure";
 920                                        break;
 921                                default:
 922                                        description = "Invalid command failure";
 923                                        break;
 924                                }
 925                                action = ACTION_FAIL;
 926                                error = -EREMOTEIO;
 927                        } else
 928                                action = ACTION_FAIL;
 929                        break;
 930                case ABORTED_COMMAND:
 931                        action = ACTION_FAIL;
 932                        if (sshdr.asc == 0x10) { /* DIF */
 933                                description = "Target Data Integrity Failure";
 934                                error = -EILSEQ;
 935                        }
 936                        break;
 937                case NOT_READY:
 938                        /* If the device is in the process of becoming
 939                         * ready, or has a temporary blockage, retry.
 940                         */
 941                        if (sshdr.asc == 0x04) {
 942                                switch (sshdr.ascq) {
 943                                case 0x01: /* becoming ready */
 944                                case 0x04: /* format in progress */
 945                                case 0x05: /* rebuild in progress */
 946                                case 0x06: /* recalculation in progress */
 947                                case 0x07: /* operation in progress */
 948                                case 0x08: /* Long write in progress */
 949                                case 0x09: /* self test in progress */
 950                                case 0x14: /* space allocation in progress */
 951                                        action = ACTION_DELAYED_RETRY;
 952                                        break;
 953                                default:
 954                                        description = "Device not ready";
 955                                        action = ACTION_FAIL;
 956                                        break;
 957                                }
 958                        } else {
 959                                description = "Device not ready";
 960                                action = ACTION_FAIL;
 961                        }
 962                        break;
 963                case VOLUME_OVERFLOW:
 964                        /* See SSC3rXX or current. */
 965                        action = ACTION_FAIL;
 966                        break;
 967                default:
 968                        description = "Unhandled sense code";
 969                        action = ACTION_FAIL;
 970                        break;
 971                }
 972        } else {
 973                description = "Unhandled error code";
 974                action = ACTION_FAIL;
 975        }
 976
 977        switch (action) {
 978        case ACTION_FAIL:
 979                /* Give up and fail the remainder of the request */
 980                scsi_release_buffers(cmd);
 981                if (!(req->cmd_flags & REQ_QUIET)) {
 982                        if (description)
 983                                scmd_printk(KERN_INFO, cmd, "%s\n",
 984                                            description);
 985                        scsi_print_result(cmd);
 986                        if (driver_byte(result) & DRIVER_SENSE)
 987                                scsi_print_sense("", cmd);
 988                        scsi_print_command(cmd);
 989                }
 990                if (blk_end_request_err(req, error))
 991                        scsi_requeue_command(q, cmd);
 992                else
 993                        scsi_next_command(cmd);
 994                break;
 995        case ACTION_REPREP:
 996                /* Unprep the request and put it back at the head of the queue.
 997                 * A new command will be prepared and issued.
 998                 */
 999                scsi_release_buffers(cmd);
1000                scsi_requeue_command(q, cmd);
1001                break;
1002        case ACTION_RETRY:
1003                /* Retry the same command immediately */
1004                __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1005                break;
1006        case ACTION_DELAYED_RETRY:
1007                /* Retry the same command after a delay */
1008                __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1009                break;
1010        }
1011}
1012
1013static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1014                             gfp_t gfp_mask)
1015{
1016        int count;
1017
1018        /*
1019         * If sg table allocation fails, requeue request later.
1020         */
1021        if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1022                                        gfp_mask))) {
1023                return BLKPREP_DEFER;
1024        }
1025
1026        req->buffer = NULL;
1027
1028        /* 
1029         * Next, walk the list, and fill in the addresses and sizes of
1030         * each segment.
1031         */
1032        count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1033        BUG_ON(count > sdb->table.nents);
1034        sdb->table.nents = count;
1035        sdb->length = blk_rq_bytes(req);
1036        return BLKPREP_OK;
1037}
1038
1039/*
1040 * Function:    scsi_init_io()
1041 *
1042 * Purpose:     SCSI I/O initialize function.
1043 *
1044 * Arguments:   cmd   - Command descriptor we wish to initialize
1045 *
1046 * Returns:     0 on success
1047 *              BLKPREP_DEFER if the failure is retryable
1048 *              BLKPREP_KILL if the failure is fatal
1049 */
1050int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1051{
1052        struct request *rq = cmd->request;
1053
1054        int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1055        if (error)
1056                goto err_exit;
1057
1058        if (blk_bidi_rq(rq)) {
1059                struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1060                        scsi_sdb_cache, GFP_ATOMIC);
1061                if (!bidi_sdb) {
1062                        error = BLKPREP_DEFER;
1063                        goto err_exit;
1064                }
1065
1066                rq->next_rq->special = bidi_sdb;
1067                error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1068                if (error)
1069                        goto err_exit;
1070        }
1071
1072        if (blk_integrity_rq(rq)) {
1073                struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1074                int ivecs, count;
1075
1076                BUG_ON(prot_sdb == NULL);
1077                ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1078
1079                if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1080                        error = BLKPREP_DEFER;
1081                        goto err_exit;
1082                }
1083
1084                count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1085                                                prot_sdb->table.sgl);
1086                BUG_ON(unlikely(count > ivecs));
1087                BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1088
1089                cmd->prot_sdb = prot_sdb;
1090                cmd->prot_sdb->table.nents = count;
1091        }
1092
1093        return BLKPREP_OK ;
1094
1095err_exit:
1096        scsi_release_buffers(cmd);
1097        cmd->request->special = NULL;
1098        scsi_put_command(cmd);
1099        return error;
1100}
1101EXPORT_SYMBOL(scsi_init_io);
1102
1103static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1104                struct request *req)
1105{
1106        struct scsi_cmnd *cmd;
1107
1108        if (!req->special) {
1109                cmd = scsi_get_command(sdev, GFP_ATOMIC);
1110                if (unlikely(!cmd))
1111                        return NULL;
1112                req->special = cmd;
1113        } else {
1114                cmd = req->special;
1115        }
1116
1117        /* pull a tag out of the request if we have one */
1118        cmd->tag = req->tag;
1119        cmd->request = req;
1120
1121        cmd->cmnd = req->cmd;
1122        cmd->prot_op = SCSI_PROT_NORMAL;
1123
1124        return cmd;
1125}
1126
1127int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1128{
1129        struct scsi_cmnd *cmd;
1130        int ret = scsi_prep_state_check(sdev, req);
1131
1132        if (ret != BLKPREP_OK)
1133                return ret;
1134
1135        cmd = scsi_get_cmd_from_req(sdev, req);
1136        if (unlikely(!cmd))
1137                return BLKPREP_DEFER;
1138
1139        /*
1140         * BLOCK_PC requests may transfer data, in which case they must
1141         * a bio attached to them.  Or they might contain a SCSI command
1142         * that does not transfer data, in which case they may optionally
1143         * submit a request without an attached bio.
1144         */
1145        if (req->bio) {
1146                int ret;
1147
1148                BUG_ON(!req->nr_phys_segments);
1149
1150                ret = scsi_init_io(cmd, GFP_ATOMIC);
1151                if (unlikely(ret))
1152                        return ret;
1153        } else {
1154                BUG_ON(blk_rq_bytes(req));
1155
1156                memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1157                req->buffer = NULL;
1158        }
1159
1160        cmd->cmd_len = req->cmd_len;
1161        if (!blk_rq_bytes(req))
1162                cmd->sc_data_direction = DMA_NONE;
1163        else if (rq_data_dir(req) == WRITE)
1164                cmd->sc_data_direction = DMA_TO_DEVICE;
1165        else
1166                cmd->sc_data_direction = DMA_FROM_DEVICE;
1167        
1168        cmd->transfersize = blk_rq_bytes(req);
1169        cmd->allowed = req->retries;
1170        return BLKPREP_OK;
1171}
1172EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1173
1174/*
1175 * Setup a REQ_TYPE_FS command.  These are simple read/write request
1176 * from filesystems that still need to be translated to SCSI CDBs from
1177 * the ULD.
1178 */
1179int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1180{
1181        struct scsi_cmnd *cmd;
1182        int ret = scsi_prep_state_check(sdev, req);
1183
1184        if (ret != BLKPREP_OK)
1185                return ret;
1186
1187        if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1188                         && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1189                ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1190                if (ret != BLKPREP_OK)
1191                        return ret;
1192        }
1193
1194        /*
1195         * Filesystem requests must transfer data.
1196         */
1197        BUG_ON(!req->nr_phys_segments);
1198
1199        cmd = scsi_get_cmd_from_req(sdev, req);
1200        if (unlikely(!cmd))
1201                return BLKPREP_DEFER;
1202
1203        memset(cmd->cmnd, 0, BLK_MAX_CDB);
1204        return scsi_init_io(cmd, GFP_ATOMIC);
1205}
1206EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1207
1208int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1209{
1210        int ret = BLKPREP_OK;
1211
1212        /*
1213         * If the device is not in running state we will reject some
1214         * or all commands.
1215         */
1216        if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1217                switch (sdev->sdev_state) {
1218                case SDEV_OFFLINE:
1219                case SDEV_TRANSPORT_OFFLINE:
1220                        /*
1221                         * If the device is offline we refuse to process any
1222                         * commands.  The device must be brought online
1223                         * before trying any recovery commands.
1224                         */
1225                        sdev_printk(KERN_ERR, sdev,
1226                                    "rejecting I/O to offline device\n");
1227                        ret = BLKPREP_KILL;
1228                        break;
1229                case SDEV_DEL:
1230                        /*
1231                         * If the device is fully deleted, we refuse to
1232                         * process any commands as well.
1233                         */
1234                        sdev_printk(KERN_ERR, sdev,
1235                                    "rejecting I/O to dead device\n");
1236                        ret = BLKPREP_KILL;
1237                        break;
1238                case SDEV_QUIESCE:
1239                case SDEV_BLOCK:
1240                case SDEV_CREATED_BLOCK:
1241                        /*
1242                         * If the devices is blocked we defer normal commands.
1243                         */
1244                        if (!(req->cmd_flags & REQ_PREEMPT))
1245                                ret = BLKPREP_DEFER;
1246                        break;
1247                default:
1248                        /*
1249                         * For any other not fully online state we only allow
1250                         * special commands.  In particular any user initiated
1251                         * command is not allowed.
1252                         */
1253                        if (!(req->cmd_flags & REQ_PREEMPT))
1254                                ret = BLKPREP_KILL;
1255                        break;
1256                }
1257        }
1258        return ret;
1259}
1260EXPORT_SYMBOL(scsi_prep_state_check);
1261
1262int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1263{
1264        struct scsi_device *sdev = q->queuedata;
1265
1266        switch (ret) {
1267        case BLKPREP_KILL:
1268                req->errors = DID_NO_CONNECT << 16;
1269                /* release the command and kill it */
1270                if (req->special) {
1271                        struct scsi_cmnd *cmd = req->special;
1272                        scsi_release_buffers(cmd);
1273                        scsi_put_command(cmd);
1274                        req->special = NULL;
1275                }
1276                break;
1277        case BLKPREP_DEFER:
1278                /*
1279                 * If we defer, the blk_peek_request() returns NULL, but the
1280                 * queue must be restarted, so we schedule a callback to happen
1281                 * shortly.
1282                 */
1283                if (sdev->device_busy == 0)
1284                        blk_delay_queue(q, SCSI_QUEUE_DELAY);
1285                break;
1286        default:
1287                req->cmd_flags |= REQ_DONTPREP;
1288        }
1289
1290        return ret;
1291}
1292EXPORT_SYMBOL(scsi_prep_return);
1293
1294int scsi_prep_fn(struct request_queue *q, struct request *req)
1295{
1296        struct scsi_device *sdev = q->queuedata;
1297        int ret = BLKPREP_KILL;
1298
1299        if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1300                ret = scsi_setup_blk_pc_cmnd(sdev, req);
1301        return scsi_prep_return(q, req, ret);
1302}
1303EXPORT_SYMBOL(scsi_prep_fn);
1304
1305/*
1306 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1307 * return 0.
1308 *
1309 * Called with the queue_lock held.
1310 */
1311static inline int scsi_dev_queue_ready(struct request_queue *q,
1312                                  struct scsi_device *sdev)
1313{
1314        if (sdev->device_busy == 0 && sdev->device_blocked) {
1315                /*
1316                 * unblock after device_blocked iterates to zero
1317                 */
1318                if (--sdev->device_blocked == 0) {
1319                        SCSI_LOG_MLQUEUE(3,
1320                                   sdev_printk(KERN_INFO, sdev,
1321                                   "unblocking device at zero depth\n"));
1322                } else {
1323                        blk_delay_queue(q, SCSI_QUEUE_DELAY);
1324                        return 0;
1325                }
1326        }
1327        if (scsi_device_is_busy(sdev))
1328                return 0;
1329
1330        return 1;
1331}
1332
1333
1334/*
1335 * scsi_target_queue_ready: checks if there we can send commands to target
1336 * @sdev: scsi device on starget to check.
1337 *
1338 * Called with the host lock held.
1339 */
1340static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1341                                           struct scsi_device *sdev)
1342{
1343        struct scsi_target *starget = scsi_target(sdev);
1344
1345        if (starget->single_lun) {
1346                if (starget->starget_sdev_user &&
1347                    starget->starget_sdev_user != sdev)
1348                        return 0;
1349                starget->starget_sdev_user = sdev;
1350        }
1351
1352        if (starget->target_busy == 0 && starget->target_blocked) {
1353                /*
1354                 * unblock after target_blocked iterates to zero
1355                 */
1356                if (--starget->target_blocked == 0) {
1357                        SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1358                                         "unblocking target at zero depth\n"));
1359                } else
1360                        return 0;
1361        }
1362
1363        if (scsi_target_is_busy(starget)) {
1364                list_move_tail(&sdev->starved_entry, &shost->starved_list);
1365                return 0;
1366        }
1367
1368        return 1;
1369}
1370
1371/*
1372 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1373 * return 0. We must end up running the queue again whenever 0 is
1374 * returned, else IO can hang.
1375 *
1376 * Called with host_lock held.
1377 */
1378static inline int scsi_host_queue_ready(struct request_queue *q,
1379                                   struct Scsi_Host *shost,
1380                                   struct scsi_device *sdev)
1381{
1382        if (scsi_host_in_recovery(shost))
1383                return 0;
1384        if (shost->host_busy == 0 && shost->host_blocked) {
1385                /*
1386                 * unblock after host_blocked iterates to zero
1387                 */
1388                if (--shost->host_blocked == 0) {
1389                        SCSI_LOG_MLQUEUE(3,
1390                                printk("scsi%d unblocking host at zero depth\n",
1391                                        shost->host_no));
1392                } else {
1393                        return 0;
1394                }
1395        }
1396        if (scsi_host_is_busy(shost)) {
1397                if (list_empty(&sdev->starved_entry))
1398                        list_add_tail(&sdev->starved_entry, &shost->starved_list);
1399                return 0;
1400        }
1401
1402        /* We're OK to process the command, so we can't be starved */
1403        if (!list_empty(&sdev->starved_entry))
1404                list_del_init(&sdev->starved_entry);
1405
1406        return 1;
1407}
1408
1409/*
1410 * Busy state exporting function for request stacking drivers.
1411 *
1412 * For efficiency, no lock is taken to check the busy state of
1413 * shost/starget/sdev, since the returned value is not guaranteed and
1414 * may be changed after request stacking drivers call the function,
1415 * regardless of taking lock or not.
1416 *
1417 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1418 * needs to return 'not busy'. Otherwise, request stacking drivers
1419 * may hold requests forever.
1420 */
1421static int scsi_lld_busy(struct request_queue *q)
1422{
1423        struct scsi_device *sdev = q->queuedata;
1424        struct Scsi_Host *shost;
1425
1426        if (blk_queue_dying(q))
1427                return 0;
1428
1429        shost = sdev->host;
1430
1431        /*
1432         * Ignore host/starget busy state.
1433         * Since block layer does not have a concept of fairness across
1434         * multiple queues, congestion of host/starget needs to be handled
1435         * in SCSI layer.
1436         */
1437        if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1438                return 1;
1439
1440        return 0;
1441}
1442
1443/*
1444 * Kill a request for a dead device
1445 */
1446static void scsi_kill_request(struct request *req, struct request_queue *q)
1447{
1448        struct scsi_cmnd *cmd = req->special;
1449        struct scsi_device *sdev;
1450        struct scsi_target *starget;
1451        struct Scsi_Host *shost;
1452
1453        blk_start_request(req);
1454
1455        scmd_printk(KERN_INFO, cmd, "killing request\n");
1456
1457        sdev = cmd->device;
1458        starget = scsi_target(sdev);
1459        shost = sdev->host;
1460        scsi_init_cmd_errh(cmd);
1461        cmd->result = DID_NO_CONNECT << 16;
1462        atomic_inc(&cmd->device->iorequest_cnt);
1463
1464        /*
1465         * SCSI request completion path will do scsi_device_unbusy(),
1466         * bump busy counts.  To bump the counters, we need to dance
1467         * with the locks as normal issue path does.
1468         */
1469        sdev->device_busy++;
1470        spin_unlock(sdev->request_queue->queue_lock);
1471        spin_lock(shost->host_lock);
1472        shost->host_busy++;
1473        starget->target_busy++;
1474        spin_unlock(shost->host_lock);
1475        spin_lock(sdev->request_queue->queue_lock);
1476
1477        blk_complete_request(req);
1478}
1479
1480static void scsi_softirq_done(struct request *rq)
1481{
1482        struct scsi_cmnd *cmd = rq->special;
1483        unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1484        int disposition;
1485
1486        INIT_LIST_HEAD(&cmd->eh_entry);
1487
1488        atomic_inc(&cmd->device->iodone_cnt);
1489        if (cmd->result)
1490                atomic_inc(&cmd->device->ioerr_cnt);
1491
1492        disposition = scsi_decide_disposition(cmd);
1493        if (disposition != SUCCESS &&
1494            time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1495                sdev_printk(KERN_ERR, cmd->device,
1496                            "timing out command, waited %lus\n",
1497                            wait_for/HZ);
1498                disposition = SUCCESS;
1499        }
1500                        
1501        scsi_log_completion(cmd, disposition);
1502
1503        switch (disposition) {
1504                case SUCCESS:
1505                        scsi_finish_command(cmd);
1506                        break;
1507                case NEEDS_RETRY:
1508                        scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1509                        break;
1510                case ADD_TO_MLQUEUE:
1511                        scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1512                        break;
1513                default:
1514                        if (!scsi_eh_scmd_add(cmd, 0))
1515                                scsi_finish_command(cmd);
1516        }
1517}
1518
1519/*
1520 * Function:    scsi_request_fn()
1521 *
1522 * Purpose:     Main strategy routine for SCSI.
1523 *
1524 * Arguments:   q       - Pointer to actual queue.
1525 *
1526 * Returns:     Nothing
1527 *
1528 * Lock status: IO request lock assumed to be held when called.
1529 */
1530static void scsi_request_fn(struct request_queue *q)
1531{
1532        struct scsi_device *sdev = q->queuedata;
1533        struct Scsi_Host *shost;
1534        struct scsi_cmnd *cmd;
1535        struct request *req;
1536
1537        if(!get_device(&sdev->sdev_gendev))
1538                /* We must be tearing the block queue down already */
1539                return;
1540
1541        /*
1542         * To start with, we keep looping until the queue is empty, or until
1543         * the host is no longer able to accept any more requests.
1544         */
1545        shost = sdev->host;
1546        for (;;) {
1547                int rtn;
1548                /*
1549                 * get next queueable request.  We do this early to make sure
1550                 * that the request is fully prepared even if we cannot 
1551                 * accept it.
1552                 */
1553                req = blk_peek_request(q);
1554                if (!req || !scsi_dev_queue_ready(q, sdev))
1555                        break;
1556
1557                if (unlikely(!scsi_device_online(sdev))) {
1558                        sdev_printk(KERN_ERR, sdev,
1559                                    "rejecting I/O to offline device\n");
1560                        scsi_kill_request(req, q);
1561                        continue;
1562                }
1563
1564
1565                /*
1566                 * Remove the request from the request list.
1567                 */
1568                if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1569                        blk_start_request(req);
1570                sdev->device_busy++;
1571
1572                spin_unlock(q->queue_lock);
1573                cmd = req->special;
1574                if (unlikely(cmd == NULL)) {
1575                        printk(KERN_CRIT "impossible request in %s.\n"
1576                                         "please mail a stack trace to "
1577                                         "linux-scsi@vger.kernel.org\n",
1578                                         __func__);
1579                        blk_dump_rq_flags(req, "foo");
1580                        BUG();
1581                }
1582                spin_lock(shost->host_lock);
1583
1584                /*
1585                 * We hit this when the driver is using a host wide
1586                 * tag map. For device level tag maps the queue_depth check
1587                 * in the device ready fn would prevent us from trying
1588                 * to allocate a tag. Since the map is a shared host resource
1589                 * we add the dev to the starved list so it eventually gets
1590                 * a run when a tag is freed.
1591                 */
1592                if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1593                        if (list_empty(&sdev->starved_entry))
1594                                list_add_tail(&sdev->starved_entry,
1595                                              &shost->starved_list);
1596                        goto not_ready;
1597                }
1598
1599                if (!scsi_target_queue_ready(shost, sdev))
1600                        goto not_ready;
1601
1602                if (!scsi_host_queue_ready(q, shost, sdev))
1603                        goto not_ready;
1604
1605                scsi_target(sdev)->target_busy++;
1606                shost->host_busy++;
1607
1608                /*
1609                 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1610                 *              take the lock again.
1611                 */
1612                spin_unlock_irq(shost->host_lock);
1613
1614                /*
1615                 * Finally, initialize any error handling parameters, and set up
1616                 * the timers for timeouts.
1617                 */
1618                scsi_init_cmd_errh(cmd);
1619
1620                /*
1621                 * Dispatch the command to the low-level driver.
1622                 */
1623                rtn = scsi_dispatch_cmd(cmd);
1624                spin_lock_irq(q->queue_lock);
1625                if (rtn)
1626                        goto out_delay;
1627        }
1628
1629        goto out;
1630
1631 not_ready:
1632        spin_unlock_irq(shost->host_lock);
1633
1634        /*
1635         * lock q, handle tag, requeue req, and decrement device_busy. We
1636         * must return with queue_lock held.
1637         *
1638         * Decrementing device_busy without checking it is OK, as all such
1639         * cases (host limits or settings) should run the queue at some
1640         * later time.
1641         */
1642        spin_lock_irq(q->queue_lock);
1643        blk_requeue_request(q, req);
1644        sdev->device_busy--;
1645out_delay:
1646        if (sdev->device_busy == 0)
1647                blk_delay_queue(q, SCSI_QUEUE_DELAY);
1648out:
1649        /* must be careful here...if we trigger the ->remove() function
1650         * we cannot be holding the q lock */
1651        spin_unlock_irq(q->queue_lock);
1652        put_device(&sdev->sdev_gendev);
1653        spin_lock_irq(q->queue_lock);
1654}
1655
1656u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1657{
1658        struct device *host_dev;
1659        u64 bounce_limit = 0xffffffff;
1660
1661        if (shost->unchecked_isa_dma)
1662                return BLK_BOUNCE_ISA;
1663        /*
1664         * Platforms with virtual-DMA translation
1665         * hardware have no practical limit.
1666         */
1667        if (!PCI_DMA_BUS_IS_PHYS)
1668                return BLK_BOUNCE_ANY;
1669
1670        host_dev = scsi_get_device(shost);
1671        if (host_dev && host_dev->dma_mask)
1672                bounce_limit = *host_dev->dma_mask;
1673
1674        return bounce_limit;
1675}
1676EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1677
1678struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1679                                         request_fn_proc *request_fn)
1680{
1681        struct request_queue *q;
1682        struct device *dev = shost->dma_dev;
1683
1684        q = blk_init_queue(request_fn, NULL);
1685        if (!q)
1686                return NULL;
1687
1688        /*
1689         * this limit is imposed by hardware restrictions
1690         */
1691        blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1692                                        SCSI_MAX_SG_CHAIN_SEGMENTS));
1693
1694        if (scsi_host_prot_dma(shost)) {
1695                shost->sg_prot_tablesize =
1696                        min_not_zero(shost->sg_prot_tablesize,
1697                                     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1698                BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1699                blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1700        }
1701
1702        blk_queue_max_hw_sectors(q, shost->max_sectors);
1703        blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1704        blk_queue_segment_boundary(q, shost->dma_boundary);
1705        dma_set_seg_boundary(dev, shost->dma_boundary);
1706
1707        blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1708
1709        if (!shost->use_clustering)
1710                q->limits.cluster = 0;
1711
1712        /*
1713         * set a reasonable default alignment on word boundaries: the
1714         * host and device may alter it using
1715         * blk_queue_update_dma_alignment() later.
1716         */
1717        blk_queue_dma_alignment(q, 0x03);
1718
1719        return q;
1720}
1721EXPORT_SYMBOL(__scsi_alloc_queue);
1722
1723struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1724{
1725        struct request_queue *q;
1726
1727        q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1728        if (!q)
1729                return NULL;
1730
1731        blk_queue_prep_rq(q, scsi_prep_fn);
1732        blk_queue_softirq_done(q, scsi_softirq_done);
1733        blk_queue_rq_timed_out(q, scsi_times_out);
1734        blk_queue_lld_busy(q, scsi_lld_busy);
1735        return q;
1736}
1737
1738/*
1739 * Function:    scsi_block_requests()
1740 *
1741 * Purpose:     Utility function used by low-level drivers to prevent further
1742 *              commands from being queued to the device.
1743 *
1744 * Arguments:   shost       - Host in question
1745 *
1746 * Returns:     Nothing
1747 *
1748 * Lock status: No locks are assumed held.
1749 *
1750 * Notes:       There is no timer nor any other means by which the requests
1751 *              get unblocked other than the low-level driver calling
1752 *              scsi_unblock_requests().
1753 */
1754void scsi_block_requests(struct Scsi_Host *shost)
1755{
1756        shost->host_self_blocked = 1;
1757}
1758EXPORT_SYMBOL(scsi_block_requests);
1759
1760/*
1761 * Function:    scsi_unblock_requests()
1762 *
1763 * Purpose:     Utility function used by low-level drivers to allow further
1764 *              commands from being queued to the device.
1765 *
1766 * Arguments:   shost       - Host in question
1767 *
1768 * Returns:     Nothing
1769 *
1770 * Lock status: No locks are assumed held.
1771 *
1772 * Notes:       There is no timer nor any other means by which the requests
1773 *              get unblocked other than the low-level driver calling
1774 *              scsi_unblock_requests().
1775 *
1776 *              This is done as an API function so that changes to the
1777 *              internals of the scsi mid-layer won't require wholesale
1778 *              changes to drivers that use this feature.
1779 */
1780void scsi_unblock_requests(struct Scsi_Host *shost)
1781{
1782        shost->host_self_blocked = 0;
1783        scsi_run_host_queues(shost);
1784}
1785EXPORT_SYMBOL(scsi_unblock_requests);
1786
1787int __init scsi_init_queue(void)
1788{
1789        int i;
1790
1791        scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1792                                           sizeof(struct scsi_data_buffer),
1793                                           0, 0, NULL);
1794        if (!scsi_sdb_cache) {
1795                printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1796                return -ENOMEM;
1797        }
1798
1799        for (i = 0; i < SG_MEMPOOL_NR; i++) {
1800                struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1801                int size = sgp->size * sizeof(struct scatterlist);
1802
1803                sgp->slab = kmem_cache_create(sgp->name, size, 0,
1804                                SLAB_HWCACHE_ALIGN, NULL);
1805                if (!sgp->slab) {
1806                        printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1807                                        sgp->name);
1808                        goto cleanup_sdb;
1809                }
1810
1811                sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1812                                                     sgp->slab);
1813                if (!sgp->pool) {
1814                        printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1815                                        sgp->name);
1816                        goto cleanup_sdb;
1817                }
1818        }
1819
1820        return 0;
1821
1822cleanup_sdb:
1823        for (i = 0; i < SG_MEMPOOL_NR; i++) {
1824                struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1825                if (sgp->pool)
1826                        mempool_destroy(sgp->pool);
1827                if (sgp->slab)
1828                        kmem_cache_destroy(sgp->slab);
1829        }
1830        kmem_cache_destroy(scsi_sdb_cache);
1831
1832        return -ENOMEM;
1833}
1834
1835void scsi_exit_queue(void)
1836{
1837        int i;
1838
1839        kmem_cache_destroy(scsi_sdb_cache);
1840
1841        for (i = 0; i < SG_MEMPOOL_NR; i++) {
1842                struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1843                mempool_destroy(sgp->pool);
1844                kmem_cache_destroy(sgp->slab);
1845        }
1846}
1847
1848/**
1849 *      scsi_mode_select - issue a mode select
1850 *      @sdev:  SCSI device to be queried
1851 *      @pf:    Page format bit (1 == standard, 0 == vendor specific)
1852 *      @sp:    Save page bit (0 == don't save, 1 == save)
1853 *      @modepage: mode page being requested
1854 *      @buffer: request buffer (may not be smaller than eight bytes)
1855 *      @len:   length of request buffer.
1856 *      @timeout: command timeout
1857 *      @retries: number of retries before failing
1858 *      @data: returns a structure abstracting the mode header data
1859 *      @sshdr: place to put sense data (or NULL if no sense to be collected).
1860 *              must be SCSI_SENSE_BUFFERSIZE big.
1861 *
1862 *      Returns zero if successful; negative error number or scsi
1863 *      status on error
1864 *
1865 */
1866int
1867scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1868                 unsigned char *buffer, int len, int timeout, int retries,
1869                 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1870{
1871        unsigned char cmd[10];
1872        unsigned char *real_buffer;
1873        int ret;
1874
1875        memset(cmd, 0, sizeof(cmd));
1876        cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1877
1878        if (sdev->use_10_for_ms) {
1879                if (len > 65535)
1880                        return -EINVAL;
1881                real_buffer = kmalloc(8 + len, GFP_KERNEL);
1882                if (!real_buffer)
1883                        return -ENOMEM;
1884                memcpy(real_buffer + 8, buffer, len);
1885                len += 8;
1886                real_buffer[0] = 0;
1887                real_buffer[1] = 0;
1888                real_buffer[2] = data->medium_type;
1889                real_buffer[3] = data->device_specific;
1890                real_buffer[4] = data->longlba ? 0x01 : 0;
1891                real_buffer[5] = 0;
1892                real_buffer[6] = data->block_descriptor_length >> 8;
1893                real_buffer[7] = data->block_descriptor_length;
1894
1895                cmd[0] = MODE_SELECT_10;
1896                cmd[7] = len >> 8;
1897                cmd[8] = len;
1898        } else {
1899                if (len > 255 || data->block_descriptor_length > 255 ||
1900                    data->longlba)
1901                        return -EINVAL;
1902
1903                real_buffer = kmalloc(4 + len, GFP_KERNEL);
1904                if (!real_buffer)
1905                        return -ENOMEM;
1906                memcpy(real_buffer + 4, buffer, len);
1907                len += 4;
1908                real_buffer[0] = 0;
1909                real_buffer[1] = data->medium_type;
1910                real_buffer[2] = data->device_specific;
1911                real_buffer[3] = data->block_descriptor_length;
1912                
1913
1914                cmd[0] = MODE_SELECT;
1915                cmd[4] = len;
1916        }
1917
1918        ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1919                               sshdr, timeout, retries, NULL);
1920        kfree(real_buffer);
1921        return ret;
1922}
1923EXPORT_SYMBOL_GPL(scsi_mode_select);
1924
1925/**
1926 *      scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1927 *      @sdev:  SCSI device to be queried
1928 *      @dbd:   set if mode sense will allow block descriptors to be returned
1929 *      @modepage: mode page being requested
1930 *      @buffer: request buffer (may not be smaller than eight bytes)
1931 *      @len:   length of request buffer.
1932 *      @timeout: command timeout
1933 *      @retries: number of retries before failing
1934 *      @data: returns a structure abstracting the mode header data
1935 *      @sshdr: place to put sense data (or NULL if no sense to be collected).
1936 *              must be SCSI_SENSE_BUFFERSIZE big.
1937 *
1938 *      Returns zero if unsuccessful, or the header offset (either 4
1939 *      or 8 depending on whether a six or ten byte command was
1940 *      issued) if successful.
1941 */
1942int
1943scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1944                  unsigned char *buffer, int len, int timeout, int retries,
1945                  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1946{
1947        unsigned char cmd[12];
1948        int use_10_for_ms;
1949        int header_length;
1950        int result;
1951        struct scsi_sense_hdr my_sshdr;
1952
1953        memset(data, 0, sizeof(*data));
1954        memset(&cmd[0], 0, 12);
1955        cmd[1] = dbd & 0x18;    /* allows DBD and LLBA bits */
1956        cmd[2] = modepage;
1957
1958        /* caller might not be interested in sense, but we need it */
1959        if (!sshdr)
1960                sshdr = &my_sshdr;
1961
1962 retry:
1963        use_10_for_ms = sdev->use_10_for_ms;
1964
1965        if (use_10_for_ms) {
1966                if (len < 8)
1967                        len = 8;
1968
1969                cmd[0] = MODE_SENSE_10;
1970                cmd[8] = len;
1971                header_length = 8;
1972        } else {
1973                if (len < 4)
1974                        len = 4;
1975
1976                cmd[0] = MODE_SENSE;
1977                cmd[4] = len;
1978                header_length = 4;
1979        }
1980
1981        memset(buffer, 0, len);
1982
1983        result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1984                                  sshdr, timeout, retries, NULL);
1985
1986        /* This code looks awful: what it's doing is making sure an
1987         * ILLEGAL REQUEST sense return identifies the actual command
1988         * byte as the problem.  MODE_SENSE commands can return
1989         * ILLEGAL REQUEST if the code page isn't supported */
1990
1991        if (use_10_for_ms && !scsi_status_is_good(result) &&
1992            (driver_byte(result) & DRIVER_SENSE)) {
1993                if (scsi_sense_valid(sshdr)) {
1994                        if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1995                            (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1996                                /* 
1997                                 * Invalid command operation code
1998                                 */
1999                                sdev->use_10_for_ms = 0;
2000                                goto retry;
2001                        }
2002                }
2003        }
2004
2005        if(scsi_status_is_good(result)) {
2006                if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2007                             (modepage == 6 || modepage == 8))) {
2008                        /* Initio breakage? */
2009                        header_length = 0;
2010                        data->length = 13;
2011                        data->medium_type = 0;
2012                        data->device_specific = 0;
2013                        data->longlba = 0;
2014                        data->block_descriptor_length = 0;
2015                } else if(use_10_for_ms) {
2016                        data->length = buffer[0]*256 + buffer[1] + 2;
2017                        data->medium_type = buffer[2];
2018                        data->device_specific = buffer[3];
2019                        data->longlba = buffer[4] & 0x01;
2020                        data->block_descriptor_length = buffer[6]*256
2021                                + buffer[7];
2022                } else {
2023                        data->length = buffer[0] + 1;
2024                        data->medium_type = buffer[1];
2025                        data->device_specific = buffer[2];
2026                        data->block_descriptor_length = buffer[3];
2027                }
2028                data->header_length = header_length;
2029        }
2030
2031        return result;
2032}
2033EXPORT_SYMBOL(scsi_mode_sense);
2034
2035/**
2036 *      scsi_test_unit_ready - test if unit is ready
2037 *      @sdev:  scsi device to change the state of.
2038 *      @timeout: command timeout
2039 *      @retries: number of retries before failing
2040 *      @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2041 *              returning sense. Make sure that this is cleared before passing
2042 *              in.
2043 *
2044 *      Returns zero if unsuccessful or an error if TUR failed.  For
2045 *      removable media, UNIT_ATTENTION sets ->changed flag.
2046 **/
2047int
2048scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2049                     struct scsi_sense_hdr *sshdr_external)
2050{
2051        char cmd[] = {
2052                TEST_UNIT_READY, 0, 0, 0, 0, 0,
2053        };
2054        struct scsi_sense_hdr *sshdr;
2055        int result;
2056
2057        if (!sshdr_external)
2058                sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2059        else
2060                sshdr = sshdr_external;
2061
2062        /* try to eat the UNIT_ATTENTION if there are enough retries */
2063        do {
2064                result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2065                                          timeout, retries, NULL);
2066                if (sdev->removable && scsi_sense_valid(sshdr) &&
2067                    sshdr->sense_key == UNIT_ATTENTION)
2068                        sdev->changed = 1;
2069        } while (scsi_sense_valid(sshdr) &&
2070                 sshdr->sense_key == UNIT_ATTENTION && --retries);
2071
2072        if (!sshdr_external)
2073                kfree(sshdr);
2074        return result;
2075}
2076EXPORT_SYMBOL(scsi_test_unit_ready);
2077
2078/**
2079 *      scsi_device_set_state - Take the given device through the device state model.
2080 *      @sdev:  scsi device to change the state of.
2081 *      @state: state to change to.
2082 *
2083 *      Returns zero if unsuccessful or an error if the requested 
2084 *      transition is illegal.
2085 */
2086int
2087scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2088{
2089        enum scsi_device_state oldstate = sdev->sdev_state;
2090
2091        if (state == oldstate)
2092                return 0;
2093
2094        switch (state) {
2095        case SDEV_CREATED:
2096                switch (oldstate) {
2097                case SDEV_CREATED_BLOCK:
2098                        break;
2099                default:
2100                        goto illegal;
2101                }
2102                break;
2103                        
2104        case SDEV_RUNNING:
2105                switch (oldstate) {
2106                case SDEV_CREATED:
2107                case SDEV_OFFLINE:
2108                case SDEV_TRANSPORT_OFFLINE:
2109                case SDEV_QUIESCE:
2110                case SDEV_BLOCK:
2111                        break;
2112                default:
2113                        goto illegal;
2114                }
2115                break;
2116
2117        case SDEV_QUIESCE:
2118                switch (oldstate) {
2119                case SDEV_RUNNING:
2120                case SDEV_OFFLINE:
2121                case SDEV_TRANSPORT_OFFLINE:
2122                        break;
2123                default:
2124                        goto illegal;
2125                }
2126                break;
2127
2128        case SDEV_OFFLINE:
2129        case SDEV_TRANSPORT_OFFLINE:
2130                switch (oldstate) {
2131                case SDEV_CREATED:
2132                case SDEV_RUNNING:
2133                case SDEV_QUIESCE:
2134                case SDEV_BLOCK:
2135                        break;
2136                default:
2137                        goto illegal;
2138                }
2139                break;
2140
2141        case SDEV_BLOCK:
2142                switch (oldstate) {
2143                case SDEV_RUNNING:
2144                case SDEV_CREATED_BLOCK:
2145                        break;
2146                default:
2147                        goto illegal;
2148                }
2149                break;
2150
2151        case SDEV_CREATED_BLOCK:
2152                switch (oldstate) {
2153                case SDEV_CREATED:
2154                        break;
2155                default:
2156                        goto illegal;
2157                }
2158                break;
2159
2160        case SDEV_CANCEL:
2161                switch (oldstate) {
2162                case SDEV_CREATED:
2163                case SDEV_RUNNING:
2164                case SDEV_QUIESCE:
2165                case SDEV_OFFLINE:
2166                case SDEV_TRANSPORT_OFFLINE:
2167                case SDEV_BLOCK:
2168                        break;
2169                default:
2170                        goto illegal;
2171                }
2172                break;
2173
2174        case SDEV_DEL:
2175                switch (oldstate) {
2176                case SDEV_CREATED:
2177                case SDEV_RUNNING:
2178                case SDEV_OFFLINE:
2179                case SDEV_TRANSPORT_OFFLINE:
2180                case SDEV_CANCEL:
2181                        break;
2182                default:
2183                        goto illegal;
2184                }
2185                break;
2186
2187        }
2188        sdev->sdev_state = state;
2189        return 0;
2190
2191 illegal:
2192        SCSI_LOG_ERROR_RECOVERY(1, 
2193                                sdev_printk(KERN_ERR, sdev,
2194                                            "Illegal state transition %s->%s\n",
2195                                            scsi_device_state_name(oldstate),
2196                                            scsi_device_state_name(state))
2197                                );
2198        return -EINVAL;
2199}
2200EXPORT_SYMBOL(scsi_device_set_state);
2201
2202/**
2203 *      sdev_evt_emit - emit a single SCSI device uevent
2204 *      @sdev: associated SCSI device
2205 *      @evt: event to emit
2206 *
2207 *      Send a single uevent (scsi_event) to the associated scsi_device.
2208 */
2209static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2210{
2211        int idx = 0;
2212        char *envp[3];
2213
2214        switch (evt->evt_type) {
2215        case SDEV_EVT_MEDIA_CHANGE:
2216                envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2217                break;
2218
2219        default:
2220                /* do nothing */
2221                break;
2222        }
2223
2224        envp[idx++] = NULL;
2225
2226        kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2227}
2228
2229/**
2230 *      sdev_evt_thread - send a uevent for each scsi event
2231 *      @work: work struct for scsi_device
2232 *
2233 *      Dispatch queued events to their associated scsi_device kobjects
2234 *      as uevents.
2235 */
2236void scsi_evt_thread(struct work_struct *work)
2237{
2238        struct scsi_device *sdev;
2239        LIST_HEAD(event_list);
2240
2241        sdev = container_of(work, struct scsi_device, event_work);
2242
2243        while (1) {
2244                struct scsi_event *evt;
2245                struct list_head *this, *tmp;
2246                unsigned long flags;
2247
2248                spin_lock_irqsave(&sdev->list_lock, flags);
2249                list_splice_init(&sdev->event_list, &event_list);
2250                spin_unlock_irqrestore(&sdev->list_lock, flags);
2251
2252                if (list_empty(&event_list))
2253                        break;
2254
2255                list_for_each_safe(this, tmp, &event_list) {
2256                        evt = list_entry(this, struct scsi_event, node);
2257                        list_del(&evt->node);
2258                        scsi_evt_emit(sdev, evt);
2259                        kfree(evt);
2260                }
2261        }
2262}
2263
2264/**
2265 *      sdev_evt_send - send asserted event to uevent thread
2266 *      @sdev: scsi_device event occurred on
2267 *      @evt: event to send
2268 *
2269 *      Assert scsi device event asynchronously.
2270 */
2271void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2272{
2273        unsigned long flags;
2274
2275#if 0
2276        /* FIXME: currently this check eliminates all media change events
2277         * for polled devices.  Need to update to discriminate between AN
2278         * and polled events */
2279        if (!test_bit(evt->evt_type, sdev->supported_events)) {
2280                kfree(evt);
2281                return;
2282        }
2283#endif
2284
2285        spin_lock_irqsave(&sdev->list_lock, flags);
2286        list_add_tail(&evt->node, &sdev->event_list);
2287        schedule_work(&sdev->event_work);
2288        spin_unlock_irqrestore(&sdev->list_lock, flags);
2289}
2290EXPORT_SYMBOL_GPL(sdev_evt_send);
2291
2292/**
2293 *      sdev_evt_alloc - allocate a new scsi event
2294 *      @evt_type: type of event to allocate
2295 *      @gfpflags: GFP flags for allocation
2296 *
2297 *      Allocates and returns a new scsi_event.
2298 */
2299struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2300                                  gfp_t gfpflags)
2301{
2302        struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2303        if (!evt)
2304                return NULL;
2305
2306        evt->evt_type = evt_type;
2307        INIT_LIST_HEAD(&evt->node);
2308
2309        /* evt_type-specific initialization, if any */
2310        switch (evt_type) {
2311        case SDEV_EVT_MEDIA_CHANGE:
2312        default:
2313                /* do nothing */
2314                break;
2315        }
2316
2317        return evt;
2318}
2319EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2320
2321/**
2322 *      sdev_evt_send_simple - send asserted event to uevent thread
2323 *      @sdev: scsi_device event occurred on
2324 *      @evt_type: type of event to send
2325 *      @gfpflags: GFP flags for allocation
2326 *
2327 *      Assert scsi device event asynchronously, given an event type.
2328 */
2329void sdev_evt_send_simple(struct scsi_device *sdev,
2330                          enum scsi_device_event evt_type, gfp_t gfpflags)
2331{
2332        struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2333        if (!evt) {
2334                sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2335                            evt_type);
2336                return;
2337        }
2338
2339        sdev_evt_send(sdev, evt);
2340}
2341EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2342
2343/**
2344 *      scsi_device_quiesce - Block user issued commands.
2345 *      @sdev:  scsi device to quiesce.
2346 *
2347 *      This works by trying to transition to the SDEV_QUIESCE state
2348 *      (which must be a legal transition).  When the device is in this
2349 *      state, only special requests will be accepted, all others will
2350 *      be deferred.  Since special requests may also be requeued requests,
2351 *      a successful return doesn't guarantee the device will be 
2352 *      totally quiescent.
2353 *
2354 *      Must be called with user context, may sleep.
2355 *
2356 *      Returns zero if unsuccessful or an error if not.
2357 */
2358int
2359scsi_device_quiesce(struct scsi_device *sdev)
2360{
2361        int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2362        if (err)
2363                return err;
2364
2365        scsi_run_queue(sdev->request_queue);
2366        while (sdev->device_busy) {
2367                msleep_interruptible(200);
2368                scsi_run_queue(sdev->request_queue);
2369        }
2370        return 0;
2371}
2372EXPORT_SYMBOL(scsi_device_quiesce);
2373
2374/**
2375 *      scsi_device_resume - Restart user issued commands to a quiesced device.
2376 *      @sdev:  scsi device to resume.
2377 *
2378 *      Moves the device from quiesced back to running and restarts the
2379 *      queues.
2380 *
2381 *      Must be called with user context, may sleep.
2382 */
2383void scsi_device_resume(struct scsi_device *sdev)
2384{
2385        /* check if the device state was mutated prior to resume, and if
2386         * so assume the state is being managed elsewhere (for example
2387         * device deleted during suspend)
2388         */
2389        if (sdev->sdev_state != SDEV_QUIESCE ||
2390            scsi_device_set_state(sdev, SDEV_RUNNING))
2391                return;
2392        scsi_run_queue(sdev->request_queue);
2393}
2394EXPORT_SYMBOL(scsi_device_resume);
2395
2396static void
2397device_quiesce_fn(struct scsi_device *sdev, void *data)
2398{
2399        scsi_device_quiesce(sdev);
2400}
2401
2402void
2403scsi_target_quiesce(struct scsi_target *starget)
2404{
2405        starget_for_each_device(starget, NULL, device_quiesce_fn);
2406}
2407EXPORT_SYMBOL(scsi_target_quiesce);
2408
2409static void
2410device_resume_fn(struct scsi_device *sdev, void *data)
2411{
2412        scsi_device_resume(sdev);
2413}
2414
2415void
2416scsi_target_resume(struct scsi_target *starget)
2417{
2418        starget_for_each_device(starget, NULL, device_resume_fn);
2419}
2420EXPORT_SYMBOL(scsi_target_resume);
2421
2422/**
2423 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2424 * @sdev:       device to block
2425 *
2426 * Block request made by scsi lld's to temporarily stop all
2427 * scsi commands on the specified device.  Called from interrupt
2428 * or normal process context.
2429 *
2430 * Returns zero if successful or error if not
2431 *
2432 * Notes:       
2433 *      This routine transitions the device to the SDEV_BLOCK state
2434 *      (which must be a legal transition).  When the device is in this
2435 *      state, all commands are deferred until the scsi lld reenables
2436 *      the device with scsi_device_unblock or device_block_tmo fires.
2437 */
2438int
2439scsi_internal_device_block(struct scsi_device *sdev)
2440{
2441        struct request_queue *q = sdev->request_queue;
2442        unsigned long flags;
2443        int err = 0;
2444
2445        err = scsi_device_set_state(sdev, SDEV_BLOCK);
2446        if (err) {
2447                err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2448
2449                if (err)
2450                        return err;
2451        }
2452
2453        /* 
2454         * The device has transitioned to SDEV_BLOCK.  Stop the
2455         * block layer from calling the midlayer with this device's
2456         * request queue. 
2457         */
2458        spin_lock_irqsave(q->queue_lock, flags);
2459        blk_stop_queue(q);
2460        spin_unlock_irqrestore(q->queue_lock, flags);
2461
2462        return 0;
2463}
2464EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2465 
2466/**
2467 * scsi_internal_device_unblock - resume a device after a block request
2468 * @sdev:       device to resume
2469 * @new_state:  state to set devices to after unblocking
2470 *
2471 * Called by scsi lld's or the midlayer to restart the device queue
2472 * for the previously suspended scsi device.  Called from interrupt or
2473 * normal process context.
2474 *
2475 * Returns zero if successful or error if not.
2476 *
2477 * Notes:       
2478 *      This routine transitions the device to the SDEV_RUNNING state
2479 *      or to one of the offline states (which must be a legal transition)
2480 *      allowing the midlayer to goose the queue for this device.
2481 */
2482int
2483scsi_internal_device_unblock(struct scsi_device *sdev,
2484                             enum scsi_device_state new_state)
2485{
2486        struct request_queue *q = sdev->request_queue; 
2487        unsigned long flags;
2488
2489        /*
2490         * Try to transition the scsi device to SDEV_RUNNING or one of the
2491         * offlined states and goose the device queue if successful.
2492         */
2493        if ((sdev->sdev_state == SDEV_BLOCK) ||
2494            (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2495                sdev->sdev_state = new_state;
2496        else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2497                if (new_state == SDEV_TRANSPORT_OFFLINE ||
2498                    new_state == SDEV_OFFLINE)
2499                        sdev->sdev_state = new_state;
2500                else
2501                        sdev->sdev_state = SDEV_CREATED;
2502        } else if (sdev->sdev_state != SDEV_CANCEL &&
2503                 sdev->sdev_state != SDEV_OFFLINE)
2504                return -EINVAL;
2505
2506        spin_lock_irqsave(q->queue_lock, flags);
2507        blk_start_queue(q);
2508        spin_unlock_irqrestore(q->queue_lock, flags);
2509
2510        return 0;
2511}
2512EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2513
2514static void
2515device_block(struct scsi_device *sdev, void *data)
2516{
2517        scsi_internal_device_block(sdev);
2518}
2519
2520static int
2521target_block(struct device *dev, void *data)
2522{
2523        if (scsi_is_target_device(dev))
2524                starget_for_each_device(to_scsi_target(dev), NULL,
2525                                        device_block);
2526        return 0;
2527}
2528
2529void
2530scsi_target_block(struct device *dev)
2531{
2532        if (scsi_is_target_device(dev))
2533                starget_for_each_device(to_scsi_target(dev), NULL,
2534                                        device_block);
2535        else
2536                device_for_each_child(dev, NULL, target_block);
2537}
2538EXPORT_SYMBOL_GPL(scsi_target_block);
2539
2540static void
2541device_unblock(struct scsi_device *sdev, void *data)
2542{
2543        scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2544}
2545
2546static int
2547target_unblock(struct device *dev, void *data)
2548{
2549        if (scsi_is_target_device(dev))
2550                starget_for_each_device(to_scsi_target(dev), data,
2551                                        device_unblock);
2552        return 0;
2553}
2554
2555void
2556scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2557{
2558        if (scsi_is_target_device(dev))
2559                starget_for_each_device(to_scsi_target(dev), &new_state,
2560                                        device_unblock);
2561        else
2562                device_for_each_child(dev, &new_state, target_unblock);
2563}
2564EXPORT_SYMBOL_GPL(scsi_target_unblock);
2565
2566/**
2567 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2568 * @sgl:        scatter-gather list
2569 * @sg_count:   number of segments in sg
2570 * @offset:     offset in bytes into sg, on return offset into the mapped area
2571 * @len:        bytes to map, on return number of bytes mapped
2572 *
2573 * Returns virtual address of the start of the mapped page
2574 */
2575void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2576                          size_t *offset, size_t *len)
2577{
2578        int i;
2579        size_t sg_len = 0, len_complete = 0;
2580        struct scatterlist *sg;
2581        struct page *page;
2582
2583        WARN_ON(!irqs_disabled());
2584
2585        for_each_sg(sgl, sg, sg_count, i) {
2586                len_complete = sg_len; /* Complete sg-entries */
2587                sg_len += sg->length;
2588                if (sg_len > *offset)
2589                        break;
2590        }
2591
2592        if (unlikely(i == sg_count)) {
2593                printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2594                        "elements %d\n",
2595                       __func__, sg_len, *offset, sg_count);
2596                WARN_ON(1);
2597                return NULL;
2598        }
2599
2600        /* Offset starting from the beginning of first page in this sg-entry */
2601        *offset = *offset - len_complete + sg->offset;
2602
2603        /* Assumption: contiguous pages can be accessed as "page + i" */
2604        page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2605        *offset &= ~PAGE_MASK;
2606
2607        /* Bytes in this sg-entry from *offset to the end of the page */
2608        sg_len = PAGE_SIZE - *offset;
2609        if (*len > sg_len)
2610                *len = sg_len;
2611
2612        return kmap_atomic(page);
2613}
2614EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2615
2616/**
2617 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2618 * @virt:       virtual address to be unmapped
2619 */
2620void scsi_kunmap_atomic_sg(void *virt)
2621{
2622        kunmap_atomic(virt);
2623}
2624EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2625
2626void sdev_disable_disk_events(struct scsi_device *sdev)
2627{
2628        atomic_inc(&sdev->disk_events_disable_depth);
2629}
2630EXPORT_SYMBOL(sdev_disable_disk_events);
2631
2632void sdev_enable_disk_events(struct scsi_device *sdev)
2633{
2634        if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2635                return;
2636        atomic_dec(&sdev->disk_events_disable_depth);
2637}
2638EXPORT_SYMBOL(sdev_enable_disk_events);
2639
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.