block/bsg.c

   1 /*
   2  * bsg.c - block layer implementation of the sg v4 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 #include <linux/module.h>
  13 #include <linux/init.h>
  14 #include <linux/file.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/poll.h>
  17 #include <linux/cdev.h>
  18 #include <linux/percpu.h>
  19 #include <linux/uio.h>
  20 #include <linux/idr.h>
  21 #include <linux/bsg.h>
  22
  23 #include <scsi/scsi.h>
  24 #include <scsi/scsi_ioctl.h>
  25 #include <scsi/scsi_cmnd.h>
  26 #include <scsi/scsi_device.h>
  27 #include <scsi/scsi_driver.h>
  28 #include <scsi/sg.h>
  29
  30 #define BSG_DESCRIPTION "Block layer SCSI generic (bsg) driver"
  31 #define BSG_VERSION     "0.4"
  32
  33 struct bsg_device {
  34         struct request_queue *queue;
  35         spinlock_t lock;
  36         struct list_head busy_list;
  37         struct list_head done_list;
  38         struct hlist_node dev_list;
  39         atomic_t ref_count;
  40         int queued_cmds;
  41         int done_cmds;
  42         wait_queue_head_t wq_done;
  43         wait_queue_head_t wq_free;
  44         char name[BUS_ID_SIZE];
  45         int max_queue;
  46         unsigned long flags;
  47 };
  48
  49 enum {
  50         BSG_F_BLOCK             = 1,
  51         BSG_F_WRITE_PERM        = 2,
  52 };
  53
  54 #define BSG_DEFAULT_CMDS        64
  55 #define BSG_MAX_DEVS            32768
  56
  57 #undef BSG_DEBUG
  58
  59 #ifdef BSG_DEBUG
  60 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args)
  61 #else
  62 #define dprintk(fmt, args...)
  63 #endif
  64
  65 static DEFINE_MUTEX(bsg_mutex);
  66 static DEFINE_IDR(bsg_minor_idr);
  67
  68 #define BSG_LIST_ARRAY_SIZE     8
  69 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
  70
  71 static struct class *bsg_class;
  72 static int bsg_major;
  73
  74 static struct kmem_cache *bsg_cmd_cachep;
  75
  76 /*
  77  * our internal command type
  78  */
  79 struct bsg_command {
  80         struct bsg_device *bd;
  81         struct list_head list;
  82         struct request *rq;
  83         struct bio *bio;
  84         struct bio *bidi_bio;
  85         int err;
  86         struct sg_io_v4 hdr;
  87         char sense[SCSI_SENSE_BUFFERSIZE];
  88 };
  89
  90 static void bsg_free_command(struct bsg_command *bc)
  91 {
  92         struct bsg_device *bd = bc->bd;
  93         unsigned long flags;
  94
  95         kmem_cache_free(bsg_cmd_cachep, bc);
  96
  97         spin_lock_irqsave(&bd->lock, flags);
  98         bd->queued_cmds--;
  99         spin_unlock_irqrestore(&bd->lock, flags);
 100
 101         wake_up(&bd->wq_free);
 102 }
 103
 104 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
 105 {
 106         struct bsg_command *bc = ERR_PTR(-EINVAL);
 107
 108         spin_lock_irq(&bd->lock);
 109
 110         if (bd->queued_cmds >= bd->max_queue)
 111                 goto out;
 112
 113         bd->queued_cmds++;
 114         spin_unlock_irq(&bd->lock);
 115
 116         bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
 117         if (unlikely(!bc)) {
 118                 spin_lock_irq(&bd->lock);
 119                 bd->queued_cmds--;
 120                 bc = ERR_PTR(-ENOMEM);
 121                 goto out;
 122         }
 123
 124         bc->bd = bd;
 125         INIT_LIST_HEAD(&bc->list);
 126         dprintk("%s: returning free cmd %p\n", bd->name, bc);
 127         return bc;
 128 out:
 129         spin_unlock_irq(&bd->lock);
 130         return bc;
 131 }
 132
 133 static inline struct hlist_head *bsg_dev_idx_hash(int index)
 134 {
 135         return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
 136 }
 137
 138 static int bsg_io_schedule(struct bsg_device *bd)
 139 {
 140         DEFINE_WAIT(wait);
 141         int ret = 0;
 142
 143         spin_lock_irq(&bd->lock);
 144
 145         BUG_ON(bd->done_cmds > bd->queued_cmds);
 146
 147         /*
 148          * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
 149          * work to do", even though we return -ENOSPC after this same test
 150          * during bsg_write() -- there, it means our buffer can't have more
 151          * bsg_commands added to it, thus has no space left.
 152          */
 153         if (bd->done_cmds == bd->queued_cmds) {
 154                 ret = -ENODATA;
 155                 goto unlock;
 156         }
 157
 158         if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 159                 ret = -EAGAIN;
 160                 goto unlock;
 161         }
 162
 163         prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
 164         spin_unlock_irq(&bd->lock);
 165         io_schedule();
 166         finish_wait(&bd->wq_done, &wait);
 167
 168         return ret;
 169 unlock:
 170         spin_unlock_irq(&bd->lock);
 171         return ret;
 172 }
 173
 174 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
 175                                 struct sg_io_v4 *hdr, int has_write_perm)
 176 {
 177         memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
 178
 179         if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
 180                            hdr->request_len))
 181                 return -EFAULT;
 182
 183         if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
 184                 if (blk_verify_command(rq->cmd, has_write_perm))
 185                         return -EPERM;
 186         } else if (!capable(CAP_SYS_RAWIO))
 187                 return -EPERM;
 188
 189         /*
 190          * fill in request structure
 191          */
 192         rq->cmd_len = hdr->request_len;
 193         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 194
 195         rq->timeout = (hdr->timeout * HZ) / 1000;
 196         if (!rq->timeout)
 197                 rq->timeout = q->sg_timeout;
 198         if (!rq->timeout)
 199                 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
 200
 201         return 0;
 202 }
 203
 204 /*
 205  * Check if sg_io_v4 from user is allowed and valid
 206  */
 207 static int
 208 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
 209 {
 210         int ret = 0;
 211
 212         if (hdr->guard != 'Q')
 213                 return -EINVAL;
 214         if (hdr->request_len > BLK_MAX_CDB)
 215                 return -EINVAL;
 216         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 217             hdr->din_xfer_len > (q->max_sectors << 9))
 218                 return -EIO;
 219
 220         switch (hdr->protocol) {
 221         case BSG_PROTOCOL_SCSI:
 222                 switch (hdr->subprotocol) {
 223                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 224                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 225                         break;
 226                 default:
 227                         ret = -EINVAL;
 228                 }
 229                 break;
 230         default:
 231                 ret = -EINVAL;
 232         }
 233
 234         *rw = hdr->dout_xfer_len ? WRITE : READ;
 235         return ret;
 236 }
 237
 238 /*
 239  * map sg_io_v4 to a request.
 240  */
 241 static struct request *
 242 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
 243 {
 244         struct request_queue *q = bd->queue;
 245         struct request *rq, *next_rq = NULL;
 246         int ret, rw;
 247         unsigned int dxfer_len;
 248         void *dxferp = NULL;
 249
 250         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 251                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 252                 hdr->din_xfer_len);
 253
 254         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 255         if (ret)
 256                 return ERR_PTR(ret);
 257
 258         /*
 259          * map scatter-gather elements seperately and string them to request
 260          */
 261         rq = blk_get_request(q, rw, GFP_KERNEL);
 262         if (!rq)
 263                 return ERR_PTR(-ENOMEM);
 264         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
 265                                                        &bd->flags));
 266         if (ret)
 267                 goto out;
 268
 269         if (rw == WRITE && hdr->din_xfer_len) {
 270                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 271                         ret = -EOPNOTSUPP;
 272                         goto out;
 273                 }
 274
 275                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 276                 if (!next_rq) {
 277                         ret = -ENOMEM;
 278                         goto out;
 279                 }
 280                 rq->next_rq = next_rq;
 281                 next_rq->cmd_type = rq->cmd_type;
 282
 283                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 284                 ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
 285                 if (ret)
 286                         goto out;
 287         }
 288
 289         if (hdr->dout_xfer_len) {
 290                 dxfer_len = hdr->dout_xfer_len;
 291                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 292         } else if (hdr->din_xfer_len) {
 293                 dxfer_len = hdr->din_xfer_len;
 294                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 295         } else
 296                 dxfer_len = 0;
 297
 298         if (dxfer_len) {
 299                 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
 300                 if (ret)
 301                         goto out;
 302         }
 303         return rq;
 304 out:
 305         blk_put_request(rq);
 306         if (next_rq) {
 307                 blk_rq_unmap_user(next_rq->bio);
 308                 blk_put_request(next_rq);
 309         }
 310         return ERR_PTR(ret);
 311 }
 312
 313 /*
 314  * async completion call-back from the block layer, when scsi/ide/whatever
 315  * calls end_that_request_last() on a request
 316  */
 317 static void bsg_rq_end_io(struct request *rq, int uptodate)
 318 {
 319         struct bsg_command *bc = rq->end_io_data;
 320         struct bsg_device *bd = bc->bd;
 321         unsigned long flags;
 322
 323         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 324                 bd->name, rq, bc, bc->bio, uptodate);
 325
 326         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 327
 328         spin_lock_irqsave(&bd->lock, flags);
 329         list_move_tail(&bc->list, &bd->done_list);
 330         bd->done_cmds++;
 331         spin_unlock_irqrestore(&bd->lock, flags);
 332
 333         wake_up(&bd->wq_done);
 334 }
 335
 336 /*
 337  * do final setup of a 'bc' and submit the matching 'rq' to the block
 338  * layer for io
 339  */
 340 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
 341                             struct bsg_command *bc, struct request *rq)
 342 {
 343         rq->sense = bc->sense;
 344         rq->sense_len = 0;
 345
 346         /*
 347          * add bc command to busy queue and submit rq for io
 348          */
 349         bc->rq = rq;
 350         bc->bio = rq->bio;
 351         if (rq->next_rq)
 352                 bc->bidi_bio = rq->next_rq->bio;
 353         bc->hdr.duration = jiffies;
 354         spin_lock_irq(&bd->lock);
 355         list_add_tail(&bc->list, &bd->busy_list);
 356         spin_unlock_irq(&bd->lock);
 357
 358         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 359
 360         rq->end_io_data = bc;
 361         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 362 }
 363
 364 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 365 {
 366         struct bsg_command *bc = NULL;
 367
 368         spin_lock_irq(&bd->lock);
 369         if (bd->done_cmds) {
 370                 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
 371                 list_del(&bc->list);
 372                 bd->done_cmds--;
 373         }
 374         spin_unlock_irq(&bd->lock);
 375
 376         return bc;
 377 }
 378
 379 /*
 380  * Get a finished command from the done list
 381  */
 382 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 383 {
 384         struct bsg_command *bc;
 385         int ret;
 386
 387         do {
 388                 bc = bsg_next_done_cmd(bd);
 389                 if (bc)
 390                         break;
 391
 392                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 393                         bc = ERR_PTR(-EAGAIN);
 394                         break;
 395                 }
 396
 397                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 398                 if (ret) {
 399                         bc = ERR_PTR(-ERESTARTSYS);
 400                         break;
 401                 }
 402         } while (1);
 403
 404         dprintk("%s: returning done %p\n", bd->name, bc);
 405
 406         return bc;
 407 }
 408
 409 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 410                                     struct bio *bio, struct bio *bidi_bio)
 411 {
 412         int ret = 0;
 413
 414         dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
 415         /*
 416          * fill in all the output members
 417          */
 418         hdr->device_status = status_byte(rq->errors);
 419         hdr->transport_status = host_byte(rq->errors);
 420         hdr->driver_status = driver_byte(rq->errors);
 421         hdr->info = 0;
 422         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 423                 hdr->info |= SG_INFO_CHECK;
 424         hdr->response_len = 0;
 425
 426         if (rq->sense_len && hdr->response) {
 427                 int len = min_t(unsigned int, hdr->max_response_len,
 428                                         rq->sense_len);
 429
 430                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 431                                    rq->sense, len);
 432                 if (!ret)
 433                         hdr->response_len = len;
 434                 else
 435                         ret = -EFAULT;
 436         }
 437
 438         if (rq->next_rq) {
 439                 hdr->dout_resid = rq->data_len;
 440                 hdr->din_resid = rq->next_rq->data_len;
 441                 blk_rq_unmap_user(bidi_bio);
 442                 blk_put_request(rq->next_rq);
 443         } else if (rq_data_dir(rq) == READ)
 444                 hdr->din_resid = rq->data_len;
 445         else
 446                 hdr->dout_resid = rq->data_len;
 447
 448         /*
 449          * If the request generated a negative error number, return it
 450          * (providing we aren't already returning an error); if it's
 451          * just a protocol response (i.e. non negative), that gets
 452          * processed above.
 453          */
 454         if (!ret && rq->errors < 0)
 455                 ret = rq->errors;
 456
 457         blk_rq_unmap_user(bio);
 458         blk_put_request(rq);
 459
 460         return ret;
 461 }
 462
 463 static int bsg_complete_all_commands(struct bsg_device *bd)
 464 {
 465         struct bsg_command *bc;
 466         int ret, tret;
 467
 468         dprintk("%s: entered\n", bd->name);
 469
 470         /*
 471          * wait for all commands to complete
 472          */
 473         ret = 0;
 474         do {
 475                 ret = bsg_io_schedule(bd);
 476                 /*
 477                  * look for -ENODATA specifically -- we'll sometimes get
 478                  * -ERESTARTSYS when we've taken a signal, but we can't
 479                  * return until we're done freeing the queue, so ignore
 480                  * it.  The signal will get handled when we're done freeing
 481                  * the bsg_device.
 482                  */
 483         } while (ret != -ENODATA);
 484
 485         /*
 486          * discard done commands
 487          */
 488         ret = 0;
 489         do {
 490                 spin_lock_irq(&bd->lock);
 491                 if (!bd->queued_cmds) {
 492                         spin_unlock_irq(&bd->lock);
 493                         break;
 494                 }
 495                 spin_unlock_irq(&bd->lock);
 496
 497                 bc = bsg_get_done_cmd(bd);
 498                 if (IS_ERR(bc))
 499                         break;
 500
 501                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 502                                                 bc->bidi_bio);
 503                 if (!ret)
 504                         ret = tret;
 505
 506                 bsg_free_command(bc);
 507         } while (1);
 508
 509         return ret;
 510 }
 511
 512 static int
 513 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 514            const struct iovec *iov, ssize_t *bytes_read)
 515 {
 516         struct bsg_command *bc;
 517         int nr_commands, ret;
 518
 519         if (count % sizeof(struct sg_io_v4))
 520                 return -EINVAL;
 521
 522         ret = 0;
 523         nr_commands = count / sizeof(struct sg_io_v4);
 524         while (nr_commands) {
 525                 bc = bsg_get_done_cmd(bd);
 526                 if (IS_ERR(bc)) {
 527                         ret = PTR_ERR(bc);
 528                         break;
 529                 }
 530
 531                 /*
 532                  * this is the only case where we need to copy data back
 533                  * after completing the request. so do that here,
 534                  * bsg_complete_work() cannot do that for us
 535                  */
 536                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 537                                                bc->bidi_bio);
 538
 539                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 540                         ret = -EFAULT;
 541
 542                 bsg_free_command(bc);
 543
 544                 if (ret)
 545                         break;
 546
 547                 buf += sizeof(struct sg_io_v4);
 548                 *bytes_read += sizeof(struct sg_io_v4);
 549                 nr_commands--;
 550         }
 551
 552         return ret;
 553 }
 554
 555 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 556 {
 557         if (file->f_flags & O_NONBLOCK)
 558                 clear_bit(BSG_F_BLOCK, &bd->flags);
 559         else
 560                 set_bit(BSG_F_BLOCK, &bd->flags);
 561 }
 562
 563 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
 564 {
 565         if (file->f_mode & FMODE_WRITE)
 566                 set_bit(BSG_F_WRITE_PERM, &bd->flags);
 567         else
 568                 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
 569 }
 570
 571 /*
 572  * Check if the error is a "real" error that we should return.
 573  */
 574 static inline int err_block_err(int ret)
 575 {
 576         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 577                 return 1;
 578
 579         return 0;
 580 }
 581
 582 static ssize_t
 583 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 584 {
 585         struct bsg_device *bd = file->private_data;
 586         int ret;
 587         ssize_t bytes_read;
 588
 589         dprintk("%s: read %Zd bytes\n", bd->name, count);
 590
 591         bsg_set_block(bd, file);
 592         bytes_read = 0;
 593         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 594         *ppos = bytes_read;
 595
 596         if (!bytes_read || (bytes_read && err_block_err(ret)))
 597                 bytes_read = ret;
 598
 599         return bytes_read;
 600 }
 601
 602 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 603                        size_t count, ssize_t *bytes_written)
 604 {
 605         struct bsg_command *bc;
 606         struct request *rq;
 607         int ret, nr_commands;
 608
 609         if (count % sizeof(struct sg_io_v4))
 610                 return -EINVAL;
 611
 612         nr_commands = count / sizeof(struct sg_io_v4);
 613         rq = NULL;
 614         bc = NULL;
 615         ret = 0;
 616         while (nr_commands) {
 617                 struct request_queue *q = bd->queue;
 618
 619                 bc = bsg_alloc_command(bd);
 620                 if (IS_ERR(bc)) {
 621                         ret = PTR_ERR(bc);
 622                         bc = NULL;
 623                         break;
 624                 }
 625
 626                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 627                         ret = -EFAULT;
 628                         break;
 629                 }
 630
 631                 /*
 632                  * get a request, fill in the blanks, and add to request queue
 633                  */
 634                 rq = bsg_map_hdr(bd, &bc->hdr);
 635                 if (IS_ERR(rq)) {
 636                         ret = PTR_ERR(rq);
 637                         rq = NULL;
 638                         break;
 639                 }
 640
 641                 bsg_add_command(bd, q, bc, rq);
 642                 bc = NULL;
 643                 rq = NULL;
 644                 nr_commands--;
 645                 buf += sizeof(struct sg_io_v4);
 646                 *bytes_written += sizeof(struct sg_io_v4);
 647         }
 648
 649         if (bc)
 650                 bsg_free_command(bc);
 651
 652         return ret;
 653 }
 654
 655 static ssize_t
 656 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 657 {
 658         struct bsg_device *bd = file->private_data;
 659         ssize_t bytes_written;
 660         int ret;
 661
 662         dprintk("%s: write %Zd bytes\n", bd->name, count);
 663
 664         bsg_set_block(bd, file);
 665         bsg_set_write_perm(bd, file);
 666
 667         bytes_written = 0;
 668         ret = __bsg_write(bd, buf, count, &bytes_written);
 669         *ppos = bytes_written;
 670
 671         /*
 672          * return bytes written on non-fatal errors
 673          */
 674         if (!bytes_written || (bytes_written && err_block_err(ret)))
 675                 bytes_written = ret;
 676
 677         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 678         return bytes_written;
 679 }
 680
 681 static struct bsg_device *bsg_alloc_device(void)
 682 {
 683         struct bsg_device *bd;
 684
 685         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 686         if (unlikely(!bd))
 687                 return NULL;
 688
 689         spin_lock_init(&bd->lock);
 690
 691         bd->max_queue = BSG_DEFAULT_CMDS;
 692
 693         INIT_LIST_HEAD(&bd->busy_list);
 694         INIT_LIST_HEAD(&bd->done_list);
 695         INIT_HLIST_NODE(&bd->dev_list);
 696
 697         init_waitqueue_head(&bd->wq_free);
 698         init_waitqueue_head(&bd->wq_done);
 699         return bd;
 700 }
 701
 702 static void bsg_kref_release_function(struct kref *kref)
 703 {
 704         struct bsg_class_device *bcd =
 705                 container_of(kref, struct bsg_class_device, ref);
 706
 707         if (bcd->release)
 708                 bcd->release(bcd->parent);
 709
 710         put_device(bcd->parent);
 711 }
 712
 713 static int bsg_put_device(struct bsg_device *bd)
 714 {
 715         int ret = 0, do_free;
 716         struct request_queue *q = bd->queue;
 717
 718         mutex_lock(&bsg_mutex);
 719
 720         do_free = atomic_dec_and_test(&bd->ref_count);
 721         if (!do_free)
 722                 goto out;
 723
 724         dprintk("%s: tearing down\n", bd->name);
 725
 726         /*
 727          * close can always block
 728          */
 729         set_bit(BSG_F_BLOCK, &bd->flags);
 730
 731         /*
 732          * correct error detection baddies here again. it's the responsibility
 733          * of the app to properly reap commands before close() if it wants
 734          * fool-proof error detection
 735          */
 736         ret = bsg_complete_all_commands(bd);
 737
 738         hlist_del(&bd->dev_list);
 739         kfree(bd);
 740 out:
 741         mutex_unlock(&bsg_mutex);
 742         kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
 743         if (do_free)
 744                 blk_put_queue(q);
 745         return ret;
 746 }
 747
 748 static struct bsg_device *bsg_add_device(struct inode *inode,
 749                                          struct request_queue *rq,
 750                                          struct file *file)
 751 {
 752         struct bsg_device *bd;
 753         int ret;
 754 #ifdef BSG_DEBUG
 755         unsigned char buf[32];
 756 #endif
 757         ret = blk_get_queue(rq);
 758         if (ret)
 759                 return ERR_PTR(-ENXIO);
 760
 761         bd = bsg_alloc_device();
 762         if (!bd) {
 763                 blk_put_queue(rq);
 764                 return ERR_PTR(-ENOMEM);
 765         }
 766
 767         bd->queue = rq;
 768         bsg_set_block(bd, file);
 769
 770         atomic_set(&bd->ref_count, 1);
 771         mutex_lock(&bsg_mutex);
 772         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
 773
 774         strncpy(bd->name, rq->bsg_dev.class_dev->bus_id, sizeof(bd->name) - 1);
 775         dprintk("bound to <%s>, max queue %d\n",
 776                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 777
 778         mutex_unlock(&bsg_mutex);
 779         return bd;
 780 }
 781
 782 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
 783 {
 784         struct bsg_device *bd;
 785         struct hlist_node *entry;
 786
 787         mutex_lock(&bsg_mutex);
 788
 789         hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
 790                 if (bd->queue == q) {
 791                         atomic_inc(&bd->ref_count);
 792                         goto found;
 793                 }
 794         }
 795         bd = NULL;
 796 found:
 797         mutex_unlock(&bsg_mutex);
 798         return bd;
 799 }
 800
 801 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 802 {
 803         struct bsg_device *bd;
 804         struct bsg_class_device *bcd;
 805
 806         /*
 807          * find the class device
 808          */
 809         mutex_lock(&bsg_mutex);
 810         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 811         if (bcd)
 812                 kref_get(&bcd->ref);
 813         mutex_unlock(&bsg_mutex);
 814
 815         if (!bcd)
 816                 return ERR_PTR(-ENODEV);
 817
 818         bd = __bsg_get_device(iminor(inode), bcd->queue);
 819         if (bd)
 820                 return bd;
 821
 822         bd = bsg_add_device(inode, bcd->queue, file);
 823         if (IS_ERR(bd))
 824                 kref_put(&bcd->ref, bsg_kref_release_function);
 825
 826         return bd;
 827 }
 828
 829 static int bsg_open(struct inode *inode, struct file *file)
 830 {
 831         struct bsg_device *bd = bsg_get_device(inode, file);
 832
 833         if (IS_ERR(bd))
 834                 return PTR_ERR(bd);
 835
 836         file->private_data = bd;
 837         return 0;
 838 }
 839
 840 static int bsg_release(struct inode *inode, struct file *file)
 841 {
 842         struct bsg_device *bd = file->private_data;
 843
 844         file->private_data = NULL;
 845         return bsg_put_device(bd);
 846 }
 847
 848 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 849 {
 850         struct bsg_device *bd = file->private_data;
 851         unsigned int mask = 0;
 852
 853         poll_wait(file, &bd->wq_done, wait);
 854         poll_wait(file, &bd->wq_free, wait);
 855
 856         spin_lock_irq(&bd->lock);
 857         if (!list_empty(&bd->done_list))
 858                 mask |= POLLIN | POLLRDNORM;
 859         if (bd->queued_cmds >= bd->max_queue)
 860                 mask |= POLLOUT;
 861         spin_unlock_irq(&bd->lock);
 862
 863         return mask;
 864 }
 865
 866 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 867 {
 868         struct bsg_device *bd = file->private_data;
 869         int __user *uarg = (int __user *) arg;
 870         int ret;
 871
 872         switch (cmd) {
 873                 /*
 874                  * our own ioctls
 875                  */
 876         case SG_GET_COMMAND_Q:
 877                 return put_user(bd->max_queue, uarg);
 878         case SG_SET_COMMAND_Q: {
 879                 int queue;
 880
 881                 if (get_user(queue, uarg))
 882                         return -EFAULT;
 883                 if (queue < 1)
 884                         return -EINVAL;
 885
 886                 spin_lock_irq(&bd->lock);
 887                 bd->max_queue = queue;
 888                 spin_unlock_irq(&bd->lock);
 889                 return 0;
 890         }
 891
 892         /*
 893          * SCSI/sg ioctls
 894          */
 895         case SG_GET_VERSION_NUM:
 896         case SCSI_IOCTL_GET_IDLUN:
 897         case SCSI_IOCTL_GET_BUS_NUMBER:
 898         case SG_SET_TIMEOUT:
 899         case SG_GET_TIMEOUT:
 900         case SG_GET_RESERVED_SIZE:
 901         case SG_SET_RESERVED_SIZE:
 902         case SG_EMULATED_HOST:
 903         case SCSI_IOCTL_SEND_COMMAND: {
 904                 void __user *uarg = (void __user *) arg;
 905                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 906         }
 907         case SG_IO: {
 908                 struct request *rq;
 909                 struct bio *bio, *bidi_bio = NULL;
 910                 struct sg_io_v4 hdr;
 911
 912                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 913                         return -EFAULT;
 914
 915                 rq = bsg_map_hdr(bd, &hdr);
 916                 if (IS_ERR(rq))
 917                         return PTR_ERR(rq);
 918
 919                 bio = rq->bio;
 920                 if (rq->next_rq)
 921                         bidi_bio = rq->next_rq->bio;
 922                 blk_execute_rq(bd->queue, NULL, rq, 0);
 923                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 924
 925                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 926                         return -EFAULT;
 927
 928                 return ret;
 929         }
 930         /*
 931          * block device ioctls
 932          */
 933         default:
 934 #if 0
 935                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 936 #else
 937                 return -ENOTTY;
 938 #endif
 939         }
 940 }
 941
 942 static const struct file_operations bsg_fops = {
 943         .read           =       bsg_read,
 944         .write          =       bsg_write,
 945         .poll           =       bsg_poll,
 946         .open           =       bsg_open,
 947         .release        =       bsg_release,
 948         .unlocked_ioctl =       bsg_ioctl,
 949         .owner          =       THIS_MODULE,
 950 };
 951
 952 void bsg_unregister_queue(struct request_queue *q)
 953 {
 954         struct bsg_class_device *bcd = &q->bsg_dev;
 955
 956         if (!bcd->class_dev)
 957                 return;
 958
 959         mutex_lock(&bsg_mutex);
 960         idr_remove(&bsg_minor_idr, bcd->minor);
 961         sysfs_remove_link(&q->kobj, "bsg");
 962         device_unregister(bcd->class_dev);
 963         bcd->class_dev = NULL;
 964         kref_put(&bcd->ref, bsg_kref_release_function);
 965         mutex_unlock(&bsg_mutex);
 966 }
 967 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 968
 969 int bsg_register_queue(struct request_queue *q, struct device *parent,
 970                        const char *name, void (*release)(struct device *))
 971 {
 972         struct bsg_class_device *bcd;
 973         dev_t dev;
 974         int ret, minor;
 975         struct device *class_dev = NULL;
 976         const char *devname;
 977
 978         if (name)
 979                 devname = name;
 980         else
 981                 devname = parent->bus_id;
 982
 983         /*
 984          * we need a proper transport to send commands, not a stacked device
 985          */
 986         if (!q->request_fn)
 987                 return 0;
 988
 989         bcd = &q->bsg_dev;
 990         memset(bcd, 0, sizeof(*bcd));
 991
 992         mutex_lock(&bsg_mutex);
 993
 994         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
 995         if (!ret) {
 996                 ret = -ENOMEM;
 997                 goto unlock;
 998         }
 999
1000         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1001         if (ret < 0)
1002                 goto unlock;
1003
1004         if (minor >= BSG_MAX_DEVS) {
1005                 printk(KERN_ERR "bsg: too many bsg devices\n");
1006                 ret = -EINVAL;
1007                 goto remove_idr;
1008         }
1009
1010         bcd->minor = minor;
1011         bcd->queue = q;
1012         bcd->parent = get_device(parent);
1013         bcd->release = release;
1014         kref_init(&bcd->ref);
1015         dev = MKDEV(bsg_major, bcd->minor);
1016         class_dev = device_create(bsg_class, parent, dev, "%s", devname);
1017         if (IS_ERR(class_dev)) {
1018                 ret = PTR_ERR(class_dev);
1019                 goto put_dev;
1020         }
1021         bcd->class_dev = class_dev;
1022
1023         if (q->kobj.sd) {
1024                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1025                 if (ret)
1026                         goto unregister_class_dev;
1027         }
1028
1029         mutex_unlock(&bsg_mutex);
1030         return 0;
1031
1032 unregister_class_dev:
1033         device_unregister(class_dev);
1034 put_dev:
1035         put_device(parent);
1036 remove_idr:
1037         idr_remove(&bsg_minor_idr, minor);
1038 unlock:
1039         mutex_unlock(&bsg_mutex);
1040         return ret;
1041 }
1042 EXPORT_SYMBOL_GPL(bsg_register_queue);
1043
1044 static struct cdev bsg_cdev;
1045
1046 static int __init bsg_init(void)
1047 {
1048         int ret, i;
1049         dev_t devid;
1050
1051         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1052                                 sizeof(struct bsg_command), 0, 0, NULL);
1053         if (!bsg_cmd_cachep) {
1054                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1055                 return -ENOMEM;
1056         }
1057
1058         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1059                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1060
1061         bsg_class = class_create(THIS_MODULE, "bsg");
1062         if (IS_ERR(bsg_class)) {
1063                 ret = PTR_ERR(bsg_class);
1064                 goto destroy_kmemcache;
1065         }
1066
1067         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1068         if (ret)
1069                 goto destroy_bsg_class;
1070
1071         bsg_major = MAJOR(devid);
1072
1073         cdev_init(&bsg_cdev, &bsg_fops);
1074         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1075         if (ret)
1076                 goto unregister_chrdev;
1077
1078         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1079                " loaded (major %d)\n", bsg_major);
1080         return 0;
1081 unregister_chrdev:
1082         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1083 destroy_bsg_class:
1084         class_destroy(bsg_class);
1085 destroy_kmemcache:
1086         kmem_cache_destroy(bsg_cmd_cachep);
1087         return ret;
1088 }
1089
1090 MODULE_AUTHOR("Jens Axboe");
1091 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1092 MODULE_LICENSE("GPL");
1093
1094 device_initcall(bsg_init);