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