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