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