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