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 minor;
  41         int queued_cmds;
  42         int done_cmds;
  43         wait_queue_head_t wq_done;
  44         wait_queue_head_t wq_free;
  45         char name[BUS_ID_SIZE];
  46         int max_queue;
  47         unsigned long flags;
  48 };
  49
  50 enum {
  51         BSG_F_BLOCK             = 1,
  52         BSG_F_WRITE_PERM        = 2,
  53 };
  54
  55 #define BSG_DEFAULT_CMDS        64
  56 #define BSG_MAX_DEVS            32768
  57
  58 #undef BSG_DEBUG
  59
  60 #ifdef BSG_DEBUG
  61 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args)
  62 #else
  63 #define dprintk(fmt, args...)
  64 #endif
  65
  66 static DEFINE_MUTEX(bsg_mutex);
  67 static DEFINE_IDR(bsg_minor_idr);
  68
  69 #define BSG_LIST_ARRAY_SIZE     8
  70 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
  71
  72 static struct class *bsg_class;
  73 static int bsg_major;
  74
  75 static struct kmem_cache *bsg_cmd_cachep;
  76
  77 /*
  78  * our internal command type
  79  */
  80 struct bsg_command {
  81         struct bsg_device *bd;
  82         struct list_head list;
  83         struct request *rq;
  84         struct bio *bio;
  85         struct bio *bidi_bio;
  86         int err;
  87         struct sg_io_v4 hdr;
  88         char sense[SCSI_SENSE_BUFFERSIZE];
  89 };
  90
  91 static void bsg_free_command(struct bsg_command *bc)
  92 {
  93         struct bsg_device *bd = bc->bd;
  94         unsigned long flags;
  95
  96         kmem_cache_free(bsg_cmd_cachep, bc);
  97
  98         spin_lock_irqsave(&bd->lock, flags);
  99         bd->queued_cmds--;
 100         spin_unlock_irqrestore(&bd->lock, flags);
 101
 102         wake_up(&bd->wq_free);
 103 }
 104
 105 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
 106 {
 107         struct bsg_command *bc = ERR_PTR(-EINVAL);
 108
 109         spin_lock_irq(&bd->lock);
 110
 111         if (bd->queued_cmds >= bd->max_queue)
 112                 goto out;
 113
 114         bd->queued_cmds++;
 115         spin_unlock_irq(&bd->lock);
 116
 117         bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
 118         if (unlikely(!bc)) {
 119                 spin_lock_irq(&bd->lock);
 120                 bd->queued_cmds--;
 121                 bc = ERR_PTR(-ENOMEM);
 122                 goto out;
 123         }
 124
 125         bc->bd = bd;
 126         INIT_LIST_HEAD(&bc->list);
 127         dprintk("%s: returning free cmd %p\n", bd->name, bc);
 128         return bc;
 129 out:
 130         spin_unlock_irq(&bd->lock);
 131         return bc;
 132 }
 133
 134 static inline struct hlist_head *bsg_dev_idx_hash(int index)
 135 {
 136         return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
 137 }
 138
 139 static int bsg_io_schedule(struct bsg_device *bd)
 140 {
 141         DEFINE_WAIT(wait);
 142         int ret = 0;
 143
 144         spin_lock_irq(&bd->lock);
 145
 146         BUG_ON(bd->done_cmds > bd->queued_cmds);
 147
 148         /*
 149          * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
 150          * work to do", even though we return -ENOSPC after this same test
 151          * during bsg_write() -- there, it means our buffer can't have more
 152          * bsg_commands added to it, thus has no space left.
 153          */
 154         if (bd->done_cmds == bd->queued_cmds) {
 155                 ret = -ENODATA;
 156                 goto unlock;
 157         }
 158
 159         if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 160                 ret = -EAGAIN;
 161                 goto unlock;
 162         }
 163
 164         prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
 165         spin_unlock_irq(&bd->lock);
 166         io_schedule();
 167         finish_wait(&bd->wq_done, &wait);
 168
 169         return ret;
 170 unlock:
 171         spin_unlock_irq(&bd->lock);
 172         return ret;
 173 }
 174
 175 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
 176                                 struct sg_io_v4 *hdr, int has_write_perm)
 177 {
 178         memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
 179
 180         if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
 181                            hdr->request_len))
 182                 return -EFAULT;
 183
 184         if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
 185                 if (blk_verify_command(rq->cmd, has_write_perm))
 186                         return -EPERM;
 187         } else if (!capable(CAP_SYS_RAWIO))
 188                 return -EPERM;
 189
 190         /*
 191          * fill in request structure
 192          */
 193         rq->cmd_len = hdr->request_len;
 194         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 195
 196         rq->timeout = (hdr->timeout * HZ) / 1000;
 197         if (!rq->timeout)
 198                 rq->timeout = q->sg_timeout;
 199         if (!rq->timeout)
 200                 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
 201
 202         return 0;
 203 }
 204
 205 /*
 206  * Check if sg_io_v4 from user is allowed and valid
 207  */
 208 static int
 209 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
 210 {
 211         int ret = 0;
 212
 213         if (hdr->guard != 'Q')
 214                 return -EINVAL;
 215         if (hdr->request_len > BLK_MAX_CDB)
 216                 return -EINVAL;
 217         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 218             hdr->din_xfer_len > (q->max_sectors << 9))
 219                 return -EIO;
 220
 221         switch (hdr->protocol) {
 222         case BSG_PROTOCOL_SCSI:
 223                 switch (hdr->subprotocol) {
 224                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 225                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 226                         break;
 227                 default:
 228                         ret = -EINVAL;
 229                 }
 230                 break;
 231         default:
 232                 ret = -EINVAL;
 233         }
 234
 235         *rw = hdr->dout_xfer_len ? WRITE : READ;
 236         return ret;
 237 }
 238
 239 /*
 240  * map sg_io_v4 to a request.
 241  */
 242 static struct request *
 243 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
 244 {
 245         struct request_queue *q = bd->queue;
 246         struct request *rq, *next_rq = NULL;
 247         int ret, rw;
 248         unsigned int dxfer_len;
 249         void *dxferp = NULL;
 250
 251         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 252                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 253                 hdr->din_xfer_len);
 254
 255         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 256         if (ret)
 257                 return ERR_PTR(ret);
 258
 259         /*
 260          * map scatter-gather elements seperately and string them to request
 261          */
 262         rq = blk_get_request(q, rw, GFP_KERNEL);
 263         if (!rq)
 264                 return ERR_PTR(-ENOMEM);
 265         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
 266                                                        &bd->flags));
 267         if (ret)
 268                 goto out;
 269
 270         if (rw == WRITE && hdr->din_xfer_len) {
 271                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 272                         ret = -EOPNOTSUPP;
 273                         goto out;
 274                 }
 275
 276                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 277                 if (!next_rq) {
 278                         ret = -ENOMEM;
 279                         goto out;
 280                 }
 281                 rq->next_rq = next_rq;
 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_entry(bd->done_list.next, 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         set_bit(BSG_F_BLOCK, &bd->flags);
 471
 472         /*
 473          * wait for all commands to complete
 474          */
 475         ret = 0;
 476         do {
 477                 ret = bsg_io_schedule(bd);
 478                 /*
 479                  * look for -ENODATA specifically -- we'll sometimes get
 480                  * -ERESTARTSYS when we've taken a signal, but we can't
 481                  * return until we're done freeing the queue, so ignore
 482                  * it.  The signal will get handled when we're done freeing
 483                  * the bsg_device.
 484                  */
 485         } while (ret != -ENODATA);
 486
 487         /*
 488          * discard done commands
 489          */
 490         ret = 0;
 491         do {
 492                 spin_lock_irq(&bd->lock);
 493                 if (!bd->queued_cmds) {
 494                         spin_unlock_irq(&bd->lock);
 495                         break;
 496                 }
 497                 spin_unlock_irq(&bd->lock);
 498
 499                 bc = bsg_get_done_cmd(bd);
 500                 if (IS_ERR(bc))
 501                         break;
 502
 503                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 504                                                 bc->bidi_bio);
 505                 if (!ret)
 506                         ret = tret;
 507
 508                 bsg_free_command(bc);
 509         } while (1);
 510
 511         return ret;
 512 }
 513
 514 static int
 515 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 516            const struct iovec *iov, ssize_t *bytes_read)
 517 {
 518         struct bsg_command *bc;
 519         int nr_commands, ret;
 520
 521         if (count % sizeof(struct sg_io_v4))
 522                 return -EINVAL;
 523
 524         ret = 0;
 525         nr_commands = count / sizeof(struct sg_io_v4);
 526         while (nr_commands) {
 527                 bc = bsg_get_done_cmd(bd);
 528                 if (IS_ERR(bc)) {
 529                         ret = PTR_ERR(bc);
 530                         break;
 531                 }
 532
 533                 /*
 534                  * this is the only case where we need to copy data back
 535                  * after completing the request. so do that here,
 536                  * bsg_complete_work() cannot do that for us
 537                  */
 538                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 539                                                bc->bidi_bio);
 540
 541                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 542                         ret = -EFAULT;
 543
 544                 bsg_free_command(bc);
 545
 546                 if (ret)
 547                         break;
 548
 549                 buf += sizeof(struct sg_io_v4);
 550                 *bytes_read += sizeof(struct sg_io_v4);
 551                 nr_commands--;
 552         }
 553
 554         return ret;
 555 }
 556
 557 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 558 {
 559         if (file->f_flags & O_NONBLOCK)
 560                 clear_bit(BSG_F_BLOCK, &bd->flags);
 561         else
 562                 set_bit(BSG_F_BLOCK, &bd->flags);
 563 }
 564
 565 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
 566 {
 567         if (file->f_mode & FMODE_WRITE)
 568                 set_bit(BSG_F_WRITE_PERM, &bd->flags);
 569         else
 570                 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
 571 }
 572
 573 /*
 574  * Check if the error is a "real" error that we should return.
 575  */
 576 static inline int err_block_err(int ret)
 577 {
 578         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 579                 return 1;
 580
 581         return 0;
 582 }
 583
 584 static ssize_t
 585 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 586 {
 587         struct bsg_device *bd = file->private_data;
 588         int ret;
 589         ssize_t bytes_read;
 590
 591         dprintk("%s: read %Zd bytes\n", bd->name, count);
 592
 593         bsg_set_block(bd, file);
 594         bytes_read = 0;
 595         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 596         *ppos = bytes_read;
 597
 598         if (!bytes_read || (bytes_read && err_block_err(ret)))
 599                 bytes_read = ret;
 600
 601         return bytes_read;
 602 }
 603
 604 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 605                        size_t count, ssize_t *bytes_written)
 606 {
 607         struct bsg_command *bc;
 608         struct request *rq;
 609         int ret, nr_commands;
 610
 611         if (count % sizeof(struct sg_io_v4))
 612                 return -EINVAL;
 613
 614         nr_commands = count / sizeof(struct sg_io_v4);
 615         rq = NULL;
 616         bc = NULL;
 617         ret = 0;
 618         while (nr_commands) {
 619                 struct request_queue *q = bd->queue;
 620
 621                 bc = bsg_alloc_command(bd);
 622                 if (IS_ERR(bc)) {
 623                         ret = PTR_ERR(bc);
 624                         bc = NULL;
 625                         break;
 626                 }
 627
 628                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 629                         ret = -EFAULT;
 630                         break;
 631                 }
 632
 633                 /*
 634                  * get a request, fill in the blanks, and add to request queue
 635                  */
 636                 rq = bsg_map_hdr(bd, &bc->hdr);
 637                 if (IS_ERR(rq)) {
 638                         ret = PTR_ERR(rq);
 639                         rq = NULL;
 640                         break;
 641                 }
 642
 643                 bsg_add_command(bd, q, bc, rq);
 644                 bc = NULL;
 645                 rq = NULL;
 646                 nr_commands--;
 647                 buf += sizeof(struct sg_io_v4);
 648                 *bytes_written += sizeof(struct sg_io_v4);
 649         }
 650
 651         if (bc)
 652                 bsg_free_command(bc);
 653
 654         return ret;
 655 }
 656
 657 static ssize_t
 658 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 659 {
 660         struct bsg_device *bd = file->private_data;
 661         ssize_t bytes_written;
 662         int ret;
 663
 664         dprintk("%s: write %Zd bytes\n", bd->name, count);
 665
 666         bsg_set_block(bd, file);
 667         bsg_set_write_perm(bd, file);
 668
 669         bytes_written = 0;
 670         ret = __bsg_write(bd, buf, count, &bytes_written);
 671         *ppos = bytes_written;
 672
 673         /*
 674          * return bytes written on non-fatal errors
 675          */
 676         if (!bytes_written || (bytes_written && err_block_err(ret)))
 677                 bytes_written = ret;
 678
 679         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 680         return bytes_written;
 681 }
 682
 683 static struct bsg_device *bsg_alloc_device(void)
 684 {
 685         struct bsg_device *bd;
 686
 687         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 688         if (unlikely(!bd))
 689                 return NULL;
 690
 691         spin_lock_init(&bd->lock);
 692
 693         bd->max_queue = BSG_DEFAULT_CMDS;
 694
 695         INIT_LIST_HEAD(&bd->busy_list);
 696         INIT_LIST_HEAD(&bd->done_list);
 697         INIT_HLIST_NODE(&bd->dev_list);
 698
 699         init_waitqueue_head(&bd->wq_free);
 700         init_waitqueue_head(&bd->wq_done);
 701         return bd;
 702 }
 703
 704 static int bsg_put_device(struct bsg_device *bd)
 705 {
 706         int ret = 0;
 707
 708         mutex_lock(&bsg_mutex);
 709
 710         if (!atomic_dec_and_test(&bd->ref_count))
 711                 goto out;
 712
 713         dprintk("%s: tearing down\n", bd->name);
 714
 715         /*
 716          * close can always block
 717          */
 718         set_bit(BSG_F_BLOCK, &bd->flags);
 719
 720         /*
 721          * correct error detection baddies here again. it's the responsibility
 722          * of the app to properly reap commands before close() if it wants
 723          * fool-proof error detection
 724          */
 725         ret = bsg_complete_all_commands(bd);
 726
 727         blk_put_queue(bd->queue);
 728         hlist_del(&bd->dev_list);
 729         kfree(bd);
 730 out:
 731         mutex_unlock(&bsg_mutex);
 732         return ret;
 733 }
 734
 735 static struct bsg_device *bsg_add_device(struct inode *inode,
 736                                          struct request_queue *rq,
 737                                          struct file *file)
 738 {
 739         struct bsg_device *bd;
 740 #ifdef BSG_DEBUG
 741         unsigned char buf[32];
 742 #endif
 743
 744         bd = bsg_alloc_device();
 745         if (!bd)
 746                 return ERR_PTR(-ENOMEM);
 747
 748         bd->queue = rq;
 749         kobject_get(&rq->kobj);
 750         bsg_set_block(bd, file);
 751
 752         atomic_set(&bd->ref_count, 1);
 753         bd->minor = iminor(inode);
 754         mutex_lock(&bsg_mutex);
 755         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(bd->minor));
 756
 757         strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1);
 758         dprintk("bound to <%s>, max queue %d\n",
 759                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 760
 761         mutex_unlock(&bsg_mutex);
 762         return bd;
 763 }
 764
 765 static struct bsg_device *__bsg_get_device(int minor)
 766 {
 767         struct bsg_device *bd = NULL;
 768         struct hlist_node *entry;
 769
 770         mutex_lock(&bsg_mutex);
 771
 772         hlist_for_each(entry, bsg_dev_idx_hash(minor)) {
 773                 bd = hlist_entry(entry, struct bsg_device, dev_list);
 774                 if (bd->minor == minor) {
 775                         atomic_inc(&bd->ref_count);
 776                         break;
 777                 }
 778
 779                 bd = NULL;
 780         }
 781
 782         mutex_unlock(&bsg_mutex);
 783         return bd;
 784 }
 785
 786 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 787 {
 788         struct bsg_device *bd;
 789         struct bsg_class_device *bcd;
 790
 791         bd = __bsg_get_device(iminor(inode));
 792         if (bd)
 793                 return bd;
 794
 795         /*
 796          * find the class device
 797          */
 798         mutex_lock(&bsg_mutex);
 799         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 800         mutex_unlock(&bsg_mutex);
 801
 802         if (!bcd)
 803                 return ERR_PTR(-ENODEV);
 804
 805         return bsg_add_device(inode, bcd->queue, file);
 806 }
 807
 808 static int bsg_open(struct inode *inode, struct file *file)
 809 {
 810         struct bsg_device *bd = bsg_get_device(inode, file);
 811
 812         if (IS_ERR(bd))
 813                 return PTR_ERR(bd);
 814
 815         file->private_data = bd;
 816         return 0;
 817 }
 818
 819 static int bsg_release(struct inode *inode, struct file *file)
 820 {
 821         struct bsg_device *bd = file->private_data;
 822
 823         file->private_data = NULL;
 824         return bsg_put_device(bd);
 825 }
 826
 827 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 828 {
 829         struct bsg_device *bd = file->private_data;
 830         unsigned int mask = 0;
 831
 832         poll_wait(file, &bd->wq_done, wait);
 833         poll_wait(file, &bd->wq_free, wait);
 834
 835         spin_lock_irq(&bd->lock);
 836         if (!list_empty(&bd->done_list))
 837                 mask |= POLLIN | POLLRDNORM;
 838         if (bd->queued_cmds >= bd->max_queue)
 839                 mask |= POLLOUT;
 840         spin_unlock_irq(&bd->lock);
 841
 842         return mask;
 843 }
 844
 845 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 846 {
 847         struct bsg_device *bd = file->private_data;
 848         int __user *uarg = (int __user *) arg;
 849         int ret;
 850
 851         switch (cmd) {
 852                 /*
 853                  * our own ioctls
 854                  */
 855         case SG_GET_COMMAND_Q:
 856                 return put_user(bd->max_queue, uarg);
 857         case SG_SET_COMMAND_Q: {
 858                 int queue;
 859
 860                 if (get_user(queue, uarg))
 861                         return -EFAULT;
 862                 if (queue < 1)
 863                         return -EINVAL;
 864
 865                 spin_lock_irq(&bd->lock);
 866                 bd->max_queue = queue;
 867                 spin_unlock_irq(&bd->lock);
 868                 return 0;
 869         }
 870
 871         /*
 872          * SCSI/sg ioctls
 873          */
 874         case SG_GET_VERSION_NUM:
 875         case SCSI_IOCTL_GET_IDLUN:
 876         case SCSI_IOCTL_GET_BUS_NUMBER:
 877         case SG_SET_TIMEOUT:
 878         case SG_GET_TIMEOUT:
 879         case SG_GET_RESERVED_SIZE:
 880         case SG_SET_RESERVED_SIZE:
 881         case SG_EMULATED_HOST:
 882         case SCSI_IOCTL_SEND_COMMAND: {
 883                 void __user *uarg = (void __user *) arg;
 884                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 885         }
 886         case SG_IO: {
 887                 struct request *rq;
 888                 struct bio *bio, *bidi_bio = NULL;
 889                 struct sg_io_v4 hdr;
 890
 891                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 892                         return -EFAULT;
 893
 894                 rq = bsg_map_hdr(bd, &hdr);
 895                 if (IS_ERR(rq))
 896                         return PTR_ERR(rq);
 897
 898                 bio = rq->bio;
 899                 if (rq->next_rq)
 900                         bidi_bio = rq->next_rq->bio;
 901                 blk_execute_rq(bd->queue, NULL, rq, 0);
 902                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 903
 904                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 905                         return -EFAULT;
 906
 907                 return ret;
 908         }
 909         /*
 910          * block device ioctls
 911          */
 912         default:
 913 #if 0
 914                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 915 #else
 916                 return -ENOTTY;
 917 #endif
 918         }
 919 }
 920
 921 static const struct file_operations bsg_fops = {
 922         .read           =       bsg_read,
 923         .write          =       bsg_write,
 924         .poll           =       bsg_poll,
 925         .open           =       bsg_open,
 926         .release        =       bsg_release,
 927         .unlocked_ioctl =       bsg_ioctl,
 928         .owner          =       THIS_MODULE,
 929 };
 930
 931 void bsg_unregister_queue(struct request_queue *q)
 932 {
 933         struct bsg_class_device *bcd = &q->bsg_dev;
 934
 935         if (!bcd->class_dev)
 936                 return;
 937
 938         mutex_lock(&bsg_mutex);
 939         idr_remove(&bsg_minor_idr, bcd->minor);
 940         sysfs_remove_link(&q->kobj, "bsg");
 941         class_device_unregister(bcd->class_dev);
 942         put_device(bcd->dev);
 943         bcd->class_dev = NULL;
 944         bcd->dev = NULL;
 945         mutex_unlock(&bsg_mutex);
 946 }
 947 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 948
 949 int bsg_register_queue(struct request_queue *q, struct device *gdev,
 950                        const char *name)
 951 {
 952         struct bsg_class_device *bcd;
 953         dev_t dev;
 954         int ret, minor;
 955         struct class_device *class_dev = NULL;
 956         const char *devname;
 957
 958         if (name)
 959                 devname = name;
 960         else
 961                 devname = gdev->bus_id;
 962
 963         /*
 964          * we need a proper transport to send commands, not a stacked device
 965          */
 966         if (!q->request_fn)
 967                 return 0;
 968
 969         bcd = &q->bsg_dev;
 970         memset(bcd, 0, sizeof(*bcd));
 971
 972         mutex_lock(&bsg_mutex);
 973
 974         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
 975         if (!ret) {
 976                 ret = -ENOMEM;
 977                 goto unlock;
 978         }
 979
 980         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
 981         if (ret < 0)
 982                 goto unlock;
 983
 984         if (minor >= BSG_MAX_DEVS) {
 985                 printk(KERN_ERR "bsg: too many bsg devices\n");
 986                 ret = -EINVAL;
 987                 goto remove_idr;
 988         }
 989
 990         bcd->minor = minor;
 991         bcd->queue = q;
 992         bcd->dev = get_device(gdev);
 993         dev = MKDEV(bsg_major, bcd->minor);
 994         class_dev = class_device_create(bsg_class, NULL, dev, gdev, "%s",
 995                                         devname);
 996         if (IS_ERR(class_dev)) {
 997                 ret = PTR_ERR(class_dev);
 998                 goto put_dev;
 999         }
1000         bcd->class_dev = class_dev;
1001
1002         if (q->kobj.sd) {
1003                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1004                 if (ret)
1005                         goto unregister_class_dev;
1006         }
1007
1008         mutex_unlock(&bsg_mutex);
1009         return 0;
1010
1011 unregister_class_dev:
1012         class_device_unregister(class_dev);
1013 put_dev:
1014         put_device(gdev);
1015 remove_idr:
1016         idr_remove(&bsg_minor_idr, minor);
1017 unlock:
1018         mutex_unlock(&bsg_mutex);
1019         return ret;
1020 }
1021 EXPORT_SYMBOL_GPL(bsg_register_queue);
1022
1023 static struct cdev bsg_cdev;
1024
1025 static int __init bsg_init(void)
1026 {
1027         int ret, i;
1028         dev_t devid;
1029
1030         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1031                                 sizeof(struct bsg_command), 0, 0, NULL);
1032         if (!bsg_cmd_cachep) {
1033                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1034                 return -ENOMEM;
1035         }
1036
1037         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1038                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1039
1040         bsg_class = class_create(THIS_MODULE, "bsg");
1041         if (IS_ERR(bsg_class)) {
1042                 ret = PTR_ERR(bsg_class);
1043                 goto destroy_kmemcache;
1044         }
1045
1046         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1047         if (ret)
1048                 goto destroy_bsg_class;
1049
1050         bsg_major = MAJOR(devid);
1051
1052         cdev_init(&bsg_cdev, &bsg_fops);
1053         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1054         if (ret)
1055                 goto unregister_chrdev;
1056
1057         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1058                " loaded (major %d)\n", bsg_major);
1059         return 0;
1060 unregister_chrdev:
1061         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1062 destroy_bsg_class:
1063         class_destroy(bsg_class);
1064 destroy_kmemcache:
1065         kmem_cache_destroy(bsg_cmd_cachep);
1066         return ret;
1067 }
1068
1069 MODULE_AUTHOR("Jens Axboe");
1070 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1071 MODULE_LICENSE("GPL");
1072
1073 device_initcall(bsg_init);