block/bsg.c

   1 /*
   2  * bsg.c - block layer implementation of the sg v3 interface
   3  *
   4  * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
   5  * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
   6  *
   7  *  This file is subject to the terms and conditions of the GNU General Public
   8  *  License version 2.  See the file "COPYING" in the main directory of this
   9  *  archive for more details.
  10  *
  11  */
  12 /*
  13  * TODO
  14  *      - Should this get merged, block/scsi_ioctl.c will be migrated into
  15  *        this file. To keep maintenance down, it's easier to have them
  16  *        seperated right now.
  17  *
  18  */
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/file.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/poll.h>
  24 #include <linux/cdev.h>
  25 #include <linux/percpu.h>
  26 #include <linux/uio.h>
  27 #include <linux/bsg.h>
  28
  29 #include <scsi/scsi.h>
  30 #include <scsi/scsi_ioctl.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_device.h>
  33 #include <scsi/scsi_driver.h>
  34 #include <scsi/sg.h>
  35
  36 #define BSG_DESCRIPTION "Block layer SCSI generic (bsg) driver"
  37 #define BSG_VERSION     "0.4"
  38
  39 struct bsg_device {
  40         request_queue_t *queue;
  41         spinlock_t lock;
  42         struct list_head busy_list;
  43         struct list_head done_list;
  44         struct hlist_node dev_list;
  45         atomic_t ref_count;
  46         int minor;
  47         int queued_cmds;
  48         int done_cmds;
  49         wait_queue_head_t wq_done;
  50         wait_queue_head_t wq_free;
  51         char name[BUS_ID_SIZE];
  52         int max_queue;
  53         unsigned long flags;
  54 };
  55
  56 enum {
  57         BSG_F_BLOCK             = 1,
  58         BSG_F_WRITE_PERM        = 2,
  59 };
  60
  61 #define BSG_DEFAULT_CMDS        64
  62 #define BSG_MAX_DEVS            32768
  63
  64 #undef BSG_DEBUG
  65
  66 #ifdef BSG_DEBUG
  67 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args)
  68 #else
  69 #define dprintk(fmt, args...)
  70 #endif
  71
  72 static DEFINE_MUTEX(bsg_mutex);
  73 static int bsg_device_nr, bsg_minor_idx;
  74
  75 #define BSG_LIST_ARRAY_SIZE     8
  76 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
  77
  78 static struct class *bsg_class;
  79 static LIST_HEAD(bsg_class_list);
  80 static int bsg_major;
  81
  82 static struct kmem_cache *bsg_cmd_cachep;
  83
  84 /*
  85  * our internal command type
  86  */
  87 struct bsg_command {
  88         struct bsg_device *bd;
  89         struct list_head list;
  90         struct request *rq;
  91         struct bio *bio;
  92         struct bio *bidi_bio;
  93         int err;
  94         struct sg_io_v4 hdr;
  95         struct sg_io_v4 __user *uhdr;
  96         char sense[SCSI_SENSE_BUFFERSIZE];
  97 };
  98
  99 static void bsg_free_command(struct bsg_command *bc)
 100 {
 101         struct bsg_device *bd = bc->bd;
 102         unsigned long flags;
 103
 104         kmem_cache_free(bsg_cmd_cachep, bc);
 105
 106         spin_lock_irqsave(&bd->lock, flags);
 107         bd->queued_cmds--;
 108         spin_unlock_irqrestore(&bd->lock, flags);
 109
 110         wake_up(&bd->wq_free);
 111 }
 112
 113 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
 114 {
 115         struct bsg_command *bc = ERR_PTR(-EINVAL);
 116
 117         spin_lock_irq(&bd->lock);
 118
 119         if (bd->queued_cmds >= bd->max_queue)
 120                 goto out;
 121
 122         bd->queued_cmds++;
 123         spin_unlock_irq(&bd->lock);
 124
 125         bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
 126         if (unlikely(!bc)) {
 127                 spin_lock_irq(&bd->lock);
 128                 bd->queued_cmds--;
 129                 bc = ERR_PTR(-ENOMEM);
 130                 goto out;
 131         }
 132
 133         bc->bd = bd;
 134         INIT_LIST_HEAD(&bc->list);
 135         dprintk("%s: returning free cmd %p\n", bd->name, bc);
 136         return bc;
 137 out:
 138         spin_unlock_irq(&bd->lock);
 139         return bc;
 140 }
 141
 142 static inline struct hlist_head *bsg_dev_idx_hash(int index)
 143 {
 144         return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
 145 }
 146
 147 static int bsg_io_schedule(struct bsg_device *bd)
 148 {
 149         DEFINE_WAIT(wait);
 150         int ret = 0;
 151
 152         spin_lock_irq(&bd->lock);
 153
 154         BUG_ON(bd->done_cmds > bd->queued_cmds);
 155
 156         /*
 157          * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
 158          * work to do", even though we return -ENOSPC after this same test
 159          * during bsg_write() -- there, it means our buffer can't have more
 160          * bsg_commands added to it, thus has no space left.
 161          */
 162         if (bd->done_cmds == bd->queued_cmds) {
 163                 ret = -ENODATA;
 164                 goto unlock;
 165         }
 166
 167         if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 168                 ret = -EAGAIN;
 169                 goto unlock;
 170         }
 171
 172         prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
 173         spin_unlock_irq(&bd->lock);
 174         io_schedule();
 175         finish_wait(&bd->wq_done, &wait);
 176
 177         return ret;
 178 unlock:
 179         spin_unlock_irq(&bd->lock);
 180         return ret;
 181 }
 182
 183 static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq,
 184                                 struct sg_io_v4 *hdr, int has_write_perm)
 185 {
 186         memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
 187
 188         if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
 189                            hdr->request_len))
 190                 return -EFAULT;
 191
 192         if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
 193                 if (blk_verify_command(rq->cmd, has_write_perm))
 194                         return -EPERM;
 195         } else if (!capable(CAP_SYS_RAWIO))
 196                 return -EPERM;
 197
 198         /*
 199          * fill in request structure
 200          */
 201         rq->cmd_len = hdr->request_len;
 202         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 203
 204         rq->timeout = (hdr->timeout * HZ) / 1000;
 205         if (!rq->timeout)
 206                 rq->timeout = q->sg_timeout;
 207         if (!rq->timeout)
 208                 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
 209
 210         return 0;
 211 }
 212
 213 /*
 214  * Check if sg_io_v4 from user is allowed and valid
 215  */
 216 static int
 217 bsg_validate_sgv4_hdr(request_queue_t *q, struct sg_io_v4 *hdr, int *rw)
 218 {
 219         int ret = 0;
 220
 221         if (hdr->guard != 'Q')
 222                 return -EINVAL;
 223         if (hdr->request_len > BLK_MAX_CDB)
 224                 return -EINVAL;
 225         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 226             hdr->din_xfer_len > (q->max_sectors << 9))
 227                 return -EIO;
 228
 229         switch (hdr->protocol) {
 230         case BSG_PROTOCOL_SCSI:
 231                 switch (hdr->subprotocol) {
 232                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 233                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 234                         break;
 235                 default:
 236                         ret = -EINVAL;
 237                 }
 238                 break;
 239         default:
 240                 ret = -EINVAL;
 241         }
 242
 243         *rw = hdr->dout_xfer_len ? WRITE : READ;
 244         return ret;
 245 }
 246
 247 /*
 248  * map sg_io_v4 to a request.
 249  */
 250 static struct request *
 251 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
 252 {
 253         request_queue_t *q = bd->queue;
 254         struct request *rq, *next_rq = NULL;
 255         int ret, rw;
 256         unsigned int dxfer_len;
 257         void *dxferp = NULL;
 258
 259         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 260                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 261                 hdr->din_xfer_len);
 262
 263         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 264         if (ret)
 265                 return ERR_PTR(ret);
 266
 267         /*
 268          * map scatter-gather elements seperately and string them to request
 269          */
 270         rq = blk_get_request(q, rw, GFP_KERNEL);
 271         if (!rq)
 272                 return ERR_PTR(-ENOMEM);
 273         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
 274                                                        &bd->flags));
 275         if (ret)
 276                 goto out;
 277
 278         if (rw == WRITE && hdr->din_xfer_len) {
 279                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 280                         ret = -EOPNOTSUPP;
 281                         goto out;
 282                 }
 283
 284                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 285                 if (!next_rq) {
 286                         ret = -ENOMEM;
 287                         goto out;
 288                 }
 289                 rq->next_rq = next_rq;
 290
 291                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 292                 ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
 293                 if (ret)
 294                         goto out;
 295         }
 296
 297         if (hdr->dout_xfer_len) {
 298                 dxfer_len = hdr->dout_xfer_len;
 299                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 300         } else if (hdr->din_xfer_len) {
 301                 dxfer_len = hdr->din_xfer_len;
 302                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 303         } else
 304                 dxfer_len = 0;
 305
 306         if (dxfer_len) {
 307                 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
 308                 if (ret)
 309                         goto out;
 310         }
 311         return rq;
 312 out:
 313         blk_put_request(rq);
 314         if (next_rq) {
 315                 blk_rq_unmap_user(next_rq->bio);
 316                 blk_put_request(next_rq);
 317         }
 318         return ERR_PTR(ret);
 319 }
 320
 321 /*
 322  * async completion call-back from the block layer, when scsi/ide/whatever
 323  * calls end_that_request_last() on a request
 324  */
 325 static void bsg_rq_end_io(struct request *rq, int uptodate)
 326 {
 327         struct bsg_command *bc = rq->end_io_data;
 328         struct bsg_device *bd = bc->bd;
 329         unsigned long flags;
 330
 331         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 332                 bd->name, rq, bc, bc->bio, uptodate);
 333
 334         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 335
 336         spin_lock_irqsave(&bd->lock, flags);
 337         list_move_tail(&bc->list, &bd->done_list);
 338         bd->done_cmds++;
 339         spin_unlock_irqrestore(&bd->lock, flags);
 340
 341         wake_up(&bd->wq_done);
 342 }
 343
 344 /*
 345  * do final setup of a 'bc' and submit the matching 'rq' to the block
 346  * layer for io
 347  */
 348 static void bsg_add_command(struct bsg_device *bd, request_queue_t *q,
 349                             struct bsg_command *bc, struct request *rq)
 350 {
 351         rq->sense = bc->sense;
 352         rq->sense_len = 0;
 353
 354         /*
 355          * add bc command to busy queue and submit rq for io
 356          */
 357         bc->rq = rq;
 358         bc->bio = rq->bio;
 359         if (rq->next_rq)
 360                 bc->bidi_bio = rq->next_rq->bio;
 361         bc->hdr.duration = jiffies;
 362         spin_lock_irq(&bd->lock);
 363         list_add_tail(&bc->list, &bd->busy_list);
 364         spin_unlock_irq(&bd->lock);
 365
 366         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 367
 368         rq->end_io_data = bc;
 369         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 370 }
 371
 372 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 373 {
 374         struct bsg_command *bc = NULL;
 375
 376         spin_lock_irq(&bd->lock);
 377         if (bd->done_cmds) {
 378                 bc = list_entry(bd->done_list.next, struct bsg_command, list);
 379                 list_del(&bc->list);
 380                 bd->done_cmds--;
 381         }
 382         spin_unlock_irq(&bd->lock);
 383
 384         return bc;
 385 }
 386
 387 /*
 388  * Get a finished command from the done list
 389  */
 390 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 391 {
 392         struct bsg_command *bc;
 393         int ret;
 394
 395         do {
 396                 bc = bsg_next_done_cmd(bd);
 397                 if (bc)
 398                         break;
 399
 400                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 401                         bc = ERR_PTR(-EAGAIN);
 402                         break;
 403                 }
 404
 405                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 406                 if (ret) {
 407                         bc = ERR_PTR(-ERESTARTSYS);
 408                         break;
 409                 }
 410         } while (1);
 411
 412         dprintk("%s: returning done %p\n", bd->name, bc);
 413
 414         return bc;
 415 }
 416
 417 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 418                                     struct bio *bio, struct bio *bidi_bio)
 419 {
 420         int ret = 0;
 421
 422         dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
 423         /*
 424          * fill in all the output members
 425          */
 426         hdr->device_status = status_byte(rq->errors);
 427         hdr->transport_status = host_byte(rq->errors);
 428         hdr->driver_status = driver_byte(rq->errors);
 429         hdr->info = 0;
 430         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 431                 hdr->info |= SG_INFO_CHECK;
 432         hdr->din_resid = rq->data_len;
 433         hdr->response_len = 0;
 434
 435         if (rq->sense_len && hdr->response) {
 436                 int len = min_t(unsigned int, hdr->max_response_len,
 437                                         rq->sense_len);
 438
 439                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 440                                    rq->sense, len);
 441                 if (!ret)
 442                         hdr->response_len = len;
 443                 else
 444                         ret = -EFAULT;
 445         }
 446
 447         if (rq->next_rq) {
 448                 blk_rq_unmap_user(bidi_bio);
 449                 blk_put_request(rq->next_rq);
 450         }
 451
 452         blk_rq_unmap_user(bio);
 453         blk_put_request(rq);
 454
 455         return ret;
 456 }
 457
 458 static int bsg_complete_all_commands(struct bsg_device *bd)
 459 {
 460         struct bsg_command *bc;
 461         int ret, tret;
 462
 463         dprintk("%s: entered\n", bd->name);
 464
 465         set_bit(BSG_F_BLOCK, &bd->flags);
 466
 467         /*
 468          * wait for all commands to complete
 469          */
 470         ret = 0;
 471         do {
 472                 ret = bsg_io_schedule(bd);
 473                 /*
 474                  * look for -ENODATA specifically -- we'll sometimes get
 475                  * -ERESTARTSYS when we've taken a signal, but we can't
 476                  * return until we're done freeing the queue, so ignore
 477                  * it.  The signal will get handled when we're done freeing
 478                  * the bsg_device.
 479                  */
 480         } while (ret != -ENODATA);
 481
 482         /*
 483          * discard done commands
 484          */
 485         ret = 0;
 486         do {
 487                 spin_lock_irq(&bd->lock);
 488                 if (!bd->queued_cmds) {
 489                         spin_unlock_irq(&bd->lock);
 490                         break;
 491                 }
 492                 spin_unlock_irq(&bd->lock);
 493
 494                 bc = bsg_get_done_cmd(bd);
 495                 if (IS_ERR(bc))
 496                         break;
 497
 498                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 499                                                 bc->bidi_bio);
 500                 if (!ret)
 501                         ret = tret;
 502
 503                 bsg_free_command(bc);
 504         } while (1);
 505
 506         return ret;
 507 }
 508
 509 static int
 510 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 511            const struct iovec *iov, ssize_t *bytes_read)
 512 {
 513         struct bsg_command *bc;
 514         int nr_commands, ret;
 515
 516         if (count % sizeof(struct sg_io_v4))
 517                 return -EINVAL;
 518
 519         ret = 0;
 520         nr_commands = count / sizeof(struct sg_io_v4);
 521         while (nr_commands) {
 522                 bc = bsg_get_done_cmd(bd);
 523                 if (IS_ERR(bc)) {
 524                         ret = PTR_ERR(bc);
 525                         break;
 526                 }
 527
 528                 /*
 529                  * this is the only case where we need to copy data back
 530                  * after completing the request. so do that here,
 531                  * bsg_complete_work() cannot do that for us
 532                  */
 533                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 534                                                bc->bidi_bio);
 535
 536                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 537                         ret = -EFAULT;
 538
 539                 bsg_free_command(bc);
 540
 541                 if (ret)
 542                         break;
 543
 544                 buf += sizeof(struct sg_io_v4);
 545                 *bytes_read += sizeof(struct sg_io_v4);
 546                 nr_commands--;
 547         }
 548
 549         return ret;
 550 }
 551
 552 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 553 {
 554         if (file->f_flags & O_NONBLOCK)
 555                 clear_bit(BSG_F_BLOCK, &bd->flags);
 556         else
 557                 set_bit(BSG_F_BLOCK, &bd->flags);
 558 }
 559
 560 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
 561 {
 562         if (file->f_mode & FMODE_WRITE)
 563                 set_bit(BSG_F_WRITE_PERM, &bd->flags);
 564         else
 565                 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
 566 }
 567
 568 /*
 569  * Check if the error is a "real" error that we should return.
 570  */
 571 static inline int err_block_err(int ret)
 572 {
 573         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 574                 return 1;
 575
 576         return 0;
 577 }
 578
 579 static ssize_t
 580 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 581 {
 582         struct bsg_device *bd = file->private_data;
 583         int ret;
 584         ssize_t bytes_read;
 585
 586         dprintk("%s: read %Zd bytes\n", bd->name, count);
 587
 588         bsg_set_block(bd, file);
 589         bytes_read = 0;
 590         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 591         *ppos = bytes_read;
 592
 593         if (!bytes_read || (bytes_read && err_block_err(ret)))
 594                 bytes_read = ret;
 595
 596         return bytes_read;
 597 }
 598
 599 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 600                        size_t count, ssize_t *bytes_written)
 601 {
 602         struct bsg_command *bc;
 603         struct request *rq;
 604         int ret, nr_commands;
 605
 606         if (count % sizeof(struct sg_io_v4))
 607                 return -EINVAL;
 608
 609         nr_commands = count / sizeof(struct sg_io_v4);
 610         rq = NULL;
 611         bc = NULL;
 612         ret = 0;
 613         while (nr_commands) {
 614                 request_queue_t *q = bd->queue;
 615
 616                 bc = bsg_alloc_command(bd);
 617                 if (IS_ERR(bc)) {
 618                         ret = PTR_ERR(bc);
 619                         bc = NULL;
 620                         break;
 621                 }
 622
 623                 bc->uhdr = (struct sg_io_v4 __user *) buf;
 624                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 625                         ret = -EFAULT;
 626                         break;
 627                 }
 628
 629                 /*
 630                  * get a request, fill in the blanks, and add to request queue
 631                  */
 632                 rq = bsg_map_hdr(bd, &bc->hdr);
 633                 if (IS_ERR(rq)) {
 634                         ret = PTR_ERR(rq);
 635                         rq = NULL;
 636                         break;
 637                 }
 638
 639                 bsg_add_command(bd, q, bc, rq);
 640                 bc = NULL;
 641                 rq = NULL;
 642                 nr_commands--;
 643                 buf += sizeof(struct sg_io_v4);
 644                 *bytes_written += sizeof(struct sg_io_v4);
 645         }
 646
 647         if (bc)
 648                 bsg_free_command(bc);
 649
 650         return ret;
 651 }
 652
 653 static ssize_t
 654 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 655 {
 656         struct bsg_device *bd = file->private_data;
 657         ssize_t bytes_written;
 658         int ret;
 659
 660         dprintk("%s: write %Zd bytes\n", bd->name, count);
 661
 662         bsg_set_block(bd, file);
 663         bsg_set_write_perm(bd, file);
 664
 665         bytes_written = 0;
 666         ret = __bsg_write(bd, buf, count, &bytes_written);
 667         *ppos = bytes_written;
 668
 669         /*
 670          * return bytes written on non-fatal errors
 671          */
 672         if (!bytes_written || (bytes_written && err_block_err(ret)))
 673                 bytes_written = ret;
 674
 675         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 676         return bytes_written;
 677 }
 678
 679 static struct bsg_device *bsg_alloc_device(void)
 680 {
 681         struct bsg_device *bd;
 682
 683         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 684         if (unlikely(!bd))
 685                 return NULL;
 686
 687         spin_lock_init(&bd->lock);
 688
 689         bd->max_queue = BSG_DEFAULT_CMDS;
 690
 691         INIT_LIST_HEAD(&bd->busy_list);
 692         INIT_LIST_HEAD(&bd->done_list);
 693         INIT_HLIST_NODE(&bd->dev_list);
 694
 695         init_waitqueue_head(&bd->wq_free);
 696         init_waitqueue_head(&bd->wq_done);
 697         return bd;
 698 }
 699
 700 static int bsg_put_device(struct bsg_device *bd)
 701 {
 702         int ret = 0;
 703
 704         mutex_lock(&bsg_mutex);
 705
 706         if (!atomic_dec_and_test(&bd->ref_count))
 707                 goto out;
 708
 709         dprintk("%s: tearing down\n", bd->name);
 710
 711         /*
 712          * close can always block
 713          */
 714         set_bit(BSG_F_BLOCK, &bd->flags);
 715
 716         /*
 717          * correct error detection baddies here again. it's the responsibility
 718          * of the app to properly reap commands before close() if it wants
 719          * fool-proof error detection
 720          */
 721         ret = bsg_complete_all_commands(bd);
 722
 723         blk_put_queue(bd->queue);
 724         hlist_del(&bd->dev_list);
 725         kfree(bd);
 726 out:
 727         mutex_unlock(&bsg_mutex);
 728         return ret;
 729 }
 730
 731 static struct bsg_device *bsg_add_device(struct inode *inode,
 732                                          struct request_queue *rq,
 733                                          struct file *file)
 734 {
 735         struct bsg_device *bd;
 736 #ifdef BSG_DEBUG
 737         unsigned char buf[32];
 738 #endif
 739
 740         bd = bsg_alloc_device();
 741         if (!bd)
 742                 return ERR_PTR(-ENOMEM);
 743
 744         bd->queue = rq;
 745         kobject_get(&rq->kobj);
 746         bsg_set_block(bd, file);
 747
 748         atomic_set(&bd->ref_count, 1);
 749         bd->minor = iminor(inode);
 750         mutex_lock(&bsg_mutex);
 751         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(bd->minor));
 752
 753         strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1);
 754         dprintk("bound to <%s>, max queue %d\n",
 755                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 756
 757         mutex_unlock(&bsg_mutex);
 758         return bd;
 759 }
 760
 761 static struct bsg_device *__bsg_get_device(int minor)
 762 {
 763         struct bsg_device *bd = NULL;
 764         struct hlist_node *entry;
 765
 766         mutex_lock(&bsg_mutex);
 767
 768         hlist_for_each(entry, bsg_dev_idx_hash(minor)) {
 769                 bd = hlist_entry(entry, struct bsg_device, dev_list);
 770                 if (bd->minor == minor) {
 771                         atomic_inc(&bd->ref_count);
 772                         break;
 773                 }
 774
 775                 bd = NULL;
 776         }
 777
 778         mutex_unlock(&bsg_mutex);
 779         return bd;
 780 }
 781
 782 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 783 {
 784         struct bsg_device *bd = __bsg_get_device(iminor(inode));
 785         struct bsg_class_device *bcd, *__bcd;
 786
 787         if (bd)
 788                 return bd;
 789
 790         /*
 791          * find the class device
 792          */
 793         bcd = NULL;
 794         mutex_lock(&bsg_mutex);
 795         list_for_each_entry(__bcd, &bsg_class_list, list) {
 796                 if (__bcd->minor == iminor(inode)) {
 797                         bcd = __bcd;
 798                         break;
 799                 }
 800         }
 801         mutex_unlock(&bsg_mutex);
 802
 803         if (!bcd)
 804                 return ERR_PTR(-ENODEV);
 805
 806         return bsg_add_device(inode, bcd->queue, file);
 807 }
 808
 809 static int bsg_open(struct inode *inode, struct file *file)
 810 {
 811         struct bsg_device *bd = bsg_get_device(inode, file);
 812
 813         if (IS_ERR(bd))
 814                 return PTR_ERR(bd);
 815
 816         file->private_data = bd;
 817         return 0;
 818 }
 819
 820 static int bsg_release(struct inode *inode, struct file *file)
 821 {
 822         struct bsg_device *bd = file->private_data;
 823
 824         file->private_data = NULL;
 825         return bsg_put_device(bd);
 826 }
 827
 828 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 829 {
 830         struct bsg_device *bd = file->private_data;
 831         unsigned int mask = 0;
 832
 833         poll_wait(file, &bd->wq_done, wait);
 834         poll_wait(file, &bd->wq_free, wait);
 835
 836         spin_lock_irq(&bd->lock);
 837         if (!list_empty(&bd->done_list))
 838                 mask |= POLLIN | POLLRDNORM;
 839         if (bd->queued_cmds >= bd->max_queue)
 840                 mask |= POLLOUT;
 841         spin_unlock_irq(&bd->lock);
 842
 843         return mask;
 844 }
 845
 846 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 847 {
 848         struct bsg_device *bd = file->private_data;
 849         int __user *uarg = (int __user *) arg;
 850
 851         switch (cmd) {
 852                 /*
 853                  * our own ioctls
 854                  */
 855         case SG_GET_COMMAND_Q:
 856                 return put_user(bd->max_queue, uarg);
 857         case SG_SET_COMMAND_Q: {
 858                 int queue;
 859
 860                 if (get_user(queue, uarg))
 861                         return -EFAULT;
 862                 if (queue < 1)
 863                         return -EINVAL;
 864
 865                 spin_lock_irq(&bd->lock);
 866                 bd->max_queue = queue;
 867                 spin_unlock_irq(&bd->lock);
 868                 return 0;
 869         }
 870
 871         /*
 872          * SCSI/sg ioctls
 873          */
 874         case SG_GET_VERSION_NUM:
 875         case SCSI_IOCTL_GET_IDLUN:
 876         case SCSI_IOCTL_GET_BUS_NUMBER:
 877         case SG_SET_TIMEOUT:
 878         case SG_GET_TIMEOUT:
 879         case SG_GET_RESERVED_SIZE:
 880         case SG_SET_RESERVED_SIZE:
 881         case SG_EMULATED_HOST:
 882         case SCSI_IOCTL_SEND_COMMAND: {
 883                 void __user *uarg = (void __user *) arg;
 884                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 885         }
 886         case SG_IO: {
 887                 struct request *rq;
 888                 struct bio *bio, *bidi_bio = NULL;
 889                 struct sg_io_v4 hdr;
 890
 891                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 892                         return -EFAULT;
 893
 894                 rq = bsg_map_hdr(bd, &hdr);
 895                 if (IS_ERR(rq))
 896                         return PTR_ERR(rq);
 897
 898                 bio = rq->bio;
 899                 if (rq->next_rq)
 900                         bidi_bio = rq->next_rq->bio;
 901                 blk_execute_rq(bd->queue, NULL, rq, 0);
 902                 blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 903
 904                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 905                         return -EFAULT;
 906
 907                 return 0;
 908         }
 909         /*
 910          * block device ioctls
 911          */
 912         default:
 913 #if 0
 914                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 915 #else
 916                 return -ENOTTY;
 917 #endif
 918         }
 919 }
 920
 921 static struct file_operations bsg_fops = {
 922         .read           =       bsg_read,
 923         .write          =       bsg_write,
 924         .poll           =       bsg_poll,
 925         .open           =       bsg_open,
 926         .release        =       bsg_release,
 927         .unlocked_ioctl =       bsg_ioctl,
 928         .owner          =       THIS_MODULE,
 929 };
 930
 931 void bsg_unregister_queue(struct request_queue *q)
 932 {
 933         struct bsg_class_device *bcd = &q->bsg_dev;
 934
 935         WARN_ON(!bcd->class_dev);
 936
 937         mutex_lock(&bsg_mutex);
 938         sysfs_remove_link(&q->kobj, "bsg");
 939         class_device_destroy(bsg_class, MKDEV(bsg_major, bcd->minor));
 940         bcd->class_dev = NULL;
 941         list_del_init(&bcd->list);
 942         bsg_device_nr--;
 943         mutex_unlock(&bsg_mutex);
 944 }
 945 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 946
 947 int bsg_register_queue(struct request_queue *q, const char *name)
 948 {
 949         struct bsg_class_device *bcd, *__bcd;
 950         dev_t dev;
 951         int ret = -EMFILE;
 952         struct class_device *class_dev = NULL;
 953
 954         /*
 955          * we need a proper transport to send commands, not a stacked device
 956          */
 957         if (!q->request_fn)
 958                 return 0;
 959
 960         bcd = &q->bsg_dev;
 961         memset(bcd, 0, sizeof(*bcd));
 962         INIT_LIST_HEAD(&bcd->list);
 963
 964         mutex_lock(&bsg_mutex);
 965         if (bsg_device_nr == BSG_MAX_DEVS) {
 966                 printk(KERN_ERR "bsg: too many bsg devices\n");
 967                 goto err;
 968         }
 969
 970 retry:
 971         list_for_each_entry(__bcd, &bsg_class_list, list) {
 972                 if (__bcd->minor == bsg_minor_idx) {
 973                         bsg_minor_idx++;
 974                         if (bsg_minor_idx == BSG_MAX_DEVS)
 975                                 bsg_minor_idx = 0;
 976                         goto retry;
 977                 }
 978         }
 979
 980         bcd->minor = bsg_minor_idx++;
 981         if (bsg_minor_idx == BSG_MAX_DEVS)
 982                 bsg_minor_idx = 0;
 983
 984         bcd->queue = q;
 985         dev = MKDEV(bsg_major, bcd->minor);
 986         class_dev = class_device_create(bsg_class, NULL, dev, bcd->dev, "%s", name);
 987         if (IS_ERR(class_dev)) {
 988                 ret = PTR_ERR(class_dev);
 989                 goto err;
 990         }
 991         bcd->class_dev = class_dev;
 992
 993         if (q->kobj.sd) {
 994                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
 995                 if (ret)
 996                         goto err;
 997         }
 998
 999         list_add_tail(&bcd->list, &bsg_class_list);
1000         bsg_device_nr++;
1001
1002         mutex_unlock(&bsg_mutex);
1003         return 0;
1004 err:
1005         if (class_dev)
1006                 class_device_destroy(bsg_class, MKDEV(bsg_major, bcd->minor));
1007         mutex_unlock(&bsg_mutex);
1008         return ret;
1009 }
1010 EXPORT_SYMBOL_GPL(bsg_register_queue);
1011
1012 static int bsg_add(struct class_device *cl_dev, struct class_interface *cl_intf)
1013 {
1014         int ret;
1015         struct scsi_device *sdp = to_scsi_device(cl_dev->dev);
1016         struct request_queue *rq = sdp->request_queue;
1017
1018         if (rq->kobj.parent)
1019                 ret = bsg_register_queue(rq, kobject_name(rq->kobj.parent));
1020         else
1021                 ret = bsg_register_queue(rq, kobject_name(&sdp->sdev_gendev.kobj));
1022         return ret;
1023 }
1024
1025 static void bsg_remove(struct class_device *cl_dev, struct class_interface *cl_intf)
1026 {
1027         bsg_unregister_queue(to_scsi_device(cl_dev->dev)->request_queue);
1028 }
1029
1030 static struct class_interface bsg_intf = {
1031         .add    = bsg_add,
1032         .remove = bsg_remove,
1033 };
1034
1035 static struct cdev bsg_cdev = {
1036         .kobj   = {.name = "bsg", },
1037         .owner  = THIS_MODULE,
1038 };
1039
1040 static int __init bsg_init(void)
1041 {
1042         int ret, i;
1043         dev_t devid;
1044
1045         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1046                                 sizeof(struct bsg_command), 0, 0, NULL, NULL);
1047         if (!bsg_cmd_cachep) {
1048                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1049                 return -ENOMEM;
1050         }
1051
1052         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1053                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1054
1055         bsg_class = class_create(THIS_MODULE, "bsg");
1056         if (IS_ERR(bsg_class)) {
1057                 ret = PTR_ERR(bsg_class);
1058                 goto destroy_kmemcache;
1059         }
1060
1061         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1062         if (ret)
1063                 goto destroy_bsg_class;
1064
1065         bsg_major = MAJOR(devid);
1066
1067         cdev_init(&bsg_cdev, &bsg_fops);
1068         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1069         if (ret)
1070                 goto unregister_chrdev;
1071
1072         ret = scsi_register_interface(&bsg_intf);
1073         if (ret)
1074                 goto remove_cdev;
1075
1076         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1077                " loaded (major %d)\n", bsg_major);
1078         return 0;
1079 remove_cdev:
1080         printk(KERN_ERR "bsg: failed register scsi interface %d\n", ret);
1081         cdev_del(&bsg_cdev);
1082 unregister_chrdev:
1083         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1084 destroy_bsg_class:
1085         class_destroy(bsg_class);
1086 destroy_kmemcache:
1087         kmem_cache_destroy(bsg_cmd_cachep);
1088         return ret;
1089 }
1090
1091 MODULE_AUTHOR("Jens Axboe");
1092 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1093 MODULE_LICENSE("GPL");
1094
1095 device_initcall(bsg_init);