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                 next_rq->cmd_type = rq->cmd_type;
 283
 284                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 285                 ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
 286                 if (ret)
 287                         goto out;
 288         }
 289
 290         if (hdr->dout_xfer_len) {
 291                 dxfer_len = hdr->dout_xfer_len;
 292                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 293         } else if (hdr->din_xfer_len) {
 294                 dxfer_len = hdr->din_xfer_len;
 295                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 296         } else
 297                 dxfer_len = 0;
 298
 299         if (dxfer_len) {
 300                 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
 301                 if (ret)
 302                         goto out;
 303         }
 304         return rq;
 305 out:
 306         blk_put_request(rq);
 307         if (next_rq) {
 308                 blk_rq_unmap_user(next_rq->bio);
 309                 blk_put_request(next_rq);
 310         }
 311         return ERR_PTR(ret);
 312 }
 313
 314 /*
 315  * async completion call-back from the block layer, when scsi/ide/whatever
 316  * calls end_that_request_last() on a request
 317  */
 318 static void bsg_rq_end_io(struct request *rq, int uptodate)
 319 {
 320         struct bsg_command *bc = rq->end_io_data;
 321         struct bsg_device *bd = bc->bd;
 322         unsigned long flags;
 323
 324         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 325                 bd->name, rq, bc, bc->bio, uptodate);
 326
 327         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 328
 329         spin_lock_irqsave(&bd->lock, flags);
 330         list_move_tail(&bc->list, &bd->done_list);
 331         bd->done_cmds++;
 332         spin_unlock_irqrestore(&bd->lock, flags);
 333
 334         wake_up(&bd->wq_done);
 335 }
 336
 337 /*
 338  * do final setup of a 'bc' and submit the matching 'rq' to the block
 339  * layer for io
 340  */
 341 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
 342                             struct bsg_command *bc, struct request *rq)
 343 {
 344         rq->sense = bc->sense;
 345         rq->sense_len = 0;
 346
 347         /*
 348          * add bc command to busy queue and submit rq for io
 349          */
 350         bc->rq = rq;
 351         bc->bio = rq->bio;
 352         if (rq->next_rq)
 353                 bc->bidi_bio = rq->next_rq->bio;
 354         bc->hdr.duration = jiffies;
 355         spin_lock_irq(&bd->lock);
 356         list_add_tail(&bc->list, &bd->busy_list);
 357         spin_unlock_irq(&bd->lock);
 358
 359         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 360
 361         rq->end_io_data = bc;
 362         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 363 }
 364
 365 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 366 {
 367         struct bsg_command *bc = NULL;
 368
 369         spin_lock_irq(&bd->lock);
 370         if (bd->done_cmds) {
 371                 bc = list_entry(bd->done_list.next, struct bsg_command, list);
 372                 list_del(&bc->list);
 373                 bd->done_cmds--;
 374         }
 375         spin_unlock_irq(&bd->lock);
 376
 377         return bc;
 378 }
 379
 380 /*
 381  * Get a finished command from the done list
 382  */
 383 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 384 {
 385         struct bsg_command *bc;
 386         int ret;
 387
 388         do {
 389                 bc = bsg_next_done_cmd(bd);
 390                 if (bc)
 391                         break;
 392
 393                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 394                         bc = ERR_PTR(-EAGAIN);
 395                         break;
 396                 }
 397
 398                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 399                 if (ret) {
 400                         bc = ERR_PTR(-ERESTARTSYS);
 401                         break;
 402                 }
 403         } while (1);
 404
 405         dprintk("%s: returning done %p\n", bd->name, bc);
 406
 407         return bc;
 408 }
 409
 410 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 411                                     struct bio *bio, struct bio *bidi_bio)
 412 {
 413         int ret = 0;
 414
 415         dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
 416         /*
 417          * fill in all the output members
 418          */
 419         hdr->device_status = status_byte(rq->errors);
 420         hdr->transport_status = host_byte(rq->errors);
 421         hdr->driver_status = driver_byte(rq->errors);
 422         hdr->info = 0;
 423         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 424                 hdr->info |= SG_INFO_CHECK;
 425         hdr->response_len = 0;
 426
 427         if (rq->sense_len && hdr->response) {
 428                 int len = min_t(unsigned int, hdr->max_response_len,
 429                                         rq->sense_len);
 430
 431                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 432                                    rq->sense, len);
 433                 if (!ret)
 434                         hdr->response_len = len;
 435                 else
 436                         ret = -EFAULT;
 437         }
 438
 439         if (rq->next_rq) {
 440                 hdr->dout_resid = rq->data_len;
 441                 hdr->din_resid = rq->next_rq->data_len;
 442                 blk_rq_unmap_user(bidi_bio);
 443                 blk_put_request(rq->next_rq);
 444         } else if (rq_data_dir(rq) == READ)
 445                 hdr->din_resid = rq->data_len;
 446         else
 447                 hdr->dout_resid = rq->data_len;
 448
 449         /*
 450          * If the request generated a negative error number, return it
 451          * (providing we aren't already returning an error); if it's
 452          * just a protocol response (i.e. non negative), that gets
 453          * processed above.
 454          */
 455         if (!ret && rq->errors < 0)
 456                 ret = rq->errors;
 457
 458         blk_rq_unmap_user(bio);
 459         blk_put_request(rq);
 460
 461         return ret;
 462 }
 463
 464 static int bsg_complete_all_commands(struct bsg_device *bd)
 465 {
 466         struct bsg_command *bc;
 467         int ret, tret;
 468
 469         dprintk("%s: entered\n", bd->name);
 470
 471         set_bit(BSG_F_BLOCK, &bd->flags);
 472
 473         /*
 474          * wait for all commands to complete
 475          */
 476         ret = 0;
 477         do {
 478                 ret = bsg_io_schedule(bd);
 479                 /*
 480                  * look for -ENODATA specifically -- we'll sometimes get
 481                  * -ERESTARTSYS when we've taken a signal, but we can't
 482                  * return until we're done freeing the queue, so ignore
 483                  * it.  The signal will get handled when we're done freeing
 484                  * the bsg_device.
 485                  */
 486         } while (ret != -ENODATA);
 487
 488         /*
 489          * discard done commands
 490          */
 491         ret = 0;
 492         do {
 493                 spin_lock_irq(&bd->lock);
 494                 if (!bd->queued_cmds) {
 495                         spin_unlock_irq(&bd->lock);
 496                         break;
 497                 }
 498                 spin_unlock_irq(&bd->lock);
 499
 500                 bc = bsg_get_done_cmd(bd);
 501                 if (IS_ERR(bc))
 502                         break;
 503
 504                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 505                                                 bc->bidi_bio);
 506                 if (!ret)
 507                         ret = tret;
 508
 509                 bsg_free_command(bc);
 510         } while (1);
 511
 512         return ret;
 513 }
 514
 515 static int
 516 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 517            const struct iovec *iov, ssize_t *bytes_read)
 518 {
 519         struct bsg_command *bc;
 520         int nr_commands, ret;
 521
 522         if (count % sizeof(struct sg_io_v4))
 523                 return -EINVAL;
 524
 525         ret = 0;
 526         nr_commands = count / sizeof(struct sg_io_v4);
 527         while (nr_commands) {
 528                 bc = bsg_get_done_cmd(bd);
 529                 if (IS_ERR(bc)) {
 530                         ret = PTR_ERR(bc);
 531                         break;
 532                 }
 533
 534                 /*
 535                  * this is the only case where we need to copy data back
 536                  * after completing the request. so do that here,
 537                  * bsg_complete_work() cannot do that for us
 538                  */
 539                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 540                                                bc->bidi_bio);
 541
 542                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 543                         ret = -EFAULT;
 544
 545                 bsg_free_command(bc);
 546
 547                 if (ret)
 548                         break;
 549
 550                 buf += sizeof(struct sg_io_v4);
 551                 *bytes_read += sizeof(struct sg_io_v4);
 552                 nr_commands--;
 553         }
 554
 555         return ret;
 556 }
 557
 558 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 559 {
 560         if (file->f_flags & O_NONBLOCK)
 561                 clear_bit(BSG_F_BLOCK, &bd->flags);
 562         else
 563                 set_bit(BSG_F_BLOCK, &bd->flags);
 564 }
 565
 566 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
 567 {
 568         if (file->f_mode & FMODE_WRITE)
 569                 set_bit(BSG_F_WRITE_PERM, &bd->flags);
 570         else
 571                 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
 572 }
 573
 574 /*
 575  * Check if the error is a "real" error that we should return.
 576  */
 577 static inline int err_block_err(int ret)
 578 {
 579         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 580                 return 1;
 581
 582         return 0;
 583 }
 584
 585 static ssize_t
 586 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 587 {
 588         struct bsg_device *bd = file->private_data;
 589         int ret;
 590         ssize_t bytes_read;
 591
 592         dprintk("%s: read %Zd bytes\n", bd->name, count);
 593
 594         bsg_set_block(bd, file);
 595         bytes_read = 0;
 596         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 597         *ppos = bytes_read;
 598
 599         if (!bytes_read || (bytes_read && err_block_err(ret)))
 600                 bytes_read = ret;
 601
 602         return bytes_read;
 603 }
 604
 605 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 606                        size_t count, ssize_t *bytes_written)
 607 {
 608         struct bsg_command *bc;
 609         struct request *rq;
 610         int ret, nr_commands;
 611
 612         if (count % sizeof(struct sg_io_v4))
 613                 return -EINVAL;
 614
 615         nr_commands = count / sizeof(struct sg_io_v4);
 616         rq = NULL;
 617         bc = NULL;
 618         ret = 0;
 619         while (nr_commands) {
 620                 struct request_queue *q = bd->queue;
 621
 622                 bc = bsg_alloc_command(bd);
 623                 if (IS_ERR(bc)) {
 624                         ret = PTR_ERR(bc);
 625                         bc = NULL;
 626                         break;
 627                 }
 628
 629                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 630                         ret = -EFAULT;
 631                         break;
 632                 }
 633
 634                 /*
 635                  * get a request, fill in the blanks, and add to request queue
 636                  */
 637                 rq = bsg_map_hdr(bd, &bc->hdr);
 638                 if (IS_ERR(rq)) {
 639                         ret = PTR_ERR(rq);
 640                         rq = NULL;
 641                         break;
 642                 }
 643
 644                 bsg_add_command(bd, q, bc, rq);
 645                 bc = NULL;
 646                 rq = NULL;
 647                 nr_commands--;
 648                 buf += sizeof(struct sg_io_v4);
 649                 *bytes_written += sizeof(struct sg_io_v4);
 650         }
 651
 652         if (bc)
 653                 bsg_free_command(bc);
 654
 655         return ret;
 656 }
 657
 658 static ssize_t
 659 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 660 {
 661         struct bsg_device *bd = file->private_data;
 662         ssize_t bytes_written;
 663         int ret;
 664
 665         dprintk("%s: write %Zd bytes\n", bd->name, count);
 666
 667         bsg_set_block(bd, file);
 668         bsg_set_write_perm(bd, file);
 669
 670         bytes_written = 0;
 671         ret = __bsg_write(bd, buf, count, &bytes_written);
 672         *ppos = bytes_written;
 673
 674         /*
 675          * return bytes written on non-fatal errors
 676          */
 677         if (!bytes_written || (bytes_written && err_block_err(ret)))
 678                 bytes_written = ret;
 679
 680         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 681         return bytes_written;
 682 }
 683
 684 static struct bsg_device *bsg_alloc_device(void)
 685 {
 686         struct bsg_device *bd;
 687
 688         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 689         if (unlikely(!bd))
 690                 return NULL;
 691
 692         spin_lock_init(&bd->lock);
 693
 694         bd->max_queue = BSG_DEFAULT_CMDS;
 695
 696         INIT_LIST_HEAD(&bd->busy_list);
 697         INIT_LIST_HEAD(&bd->done_list);
 698         INIT_HLIST_NODE(&bd->dev_list);
 699
 700         init_waitqueue_head(&bd->wq_free);
 701         init_waitqueue_head(&bd->wq_done);
 702         return bd;
 703 }
 704
 705 static int bsg_put_device(struct bsg_device *bd)
 706 {
 707         int ret = 0;
 708
 709         mutex_lock(&bsg_mutex);
 710
 711         if (!atomic_dec_and_test(&bd->ref_count))
 712                 goto out;
 713
 714         dprintk("%s: tearing down\n", bd->name);
 715
 716         /*
 717          * close can always block
 718          */
 719         set_bit(BSG_F_BLOCK, &bd->flags);
 720
 721         /*
 722          * correct error detection baddies here again. it's the responsibility
 723          * of the app to properly reap commands before close() if it wants
 724          * fool-proof error detection
 725          */
 726         ret = bsg_complete_all_commands(bd);
 727
 728         blk_put_queue(bd->queue);
 729         hlist_del(&bd->dev_list);
 730         kfree(bd);
 731 out:
 732         mutex_unlock(&bsg_mutex);
 733         return ret;
 734 }
 735
 736 static struct bsg_device *bsg_add_device(struct inode *inode,
 737                                          struct request_queue *rq,
 738                                          struct file *file)
 739 {
 740         struct bsg_device *bd;
 741 #ifdef BSG_DEBUG
 742         unsigned char buf[32];
 743 #endif
 744
 745         bd = bsg_alloc_device();
 746         if (!bd)
 747                 return ERR_PTR(-ENOMEM);
 748
 749         bd->queue = rq;
 750         kobject_get(&rq->kobj);
 751         bsg_set_block(bd, file);
 752
 753         atomic_set(&bd->ref_count, 1);
 754         bd->minor = iminor(inode);
 755         mutex_lock(&bsg_mutex);
 756         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(bd->minor));
 757
 758         strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1);
 759         dprintk("bound to <%s>, max queue %d\n",
 760                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 761
 762         mutex_unlock(&bsg_mutex);
 763         return bd;
 764 }
 765
 766 static struct bsg_device *__bsg_get_device(int minor)
 767 {
 768         struct bsg_device *bd = NULL;
 769         struct hlist_node *entry;
 770
 771         mutex_lock(&bsg_mutex);
 772
 773         hlist_for_each(entry, bsg_dev_idx_hash(minor)) {
 774                 bd = hlist_entry(entry, struct bsg_device, dev_list);
 775                 if (bd->minor == minor) {
 776                         atomic_inc(&bd->ref_count);
 777                         break;
 778                 }
 779
 780                 bd = NULL;
 781         }
 782
 783         mutex_unlock(&bsg_mutex);
 784         return bd;
 785 }
 786
 787 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 788 {
 789         struct bsg_device *bd;
 790         struct bsg_class_device *bcd;
 791
 792         bd = __bsg_get_device(iminor(inode));
 793         if (bd)
 794                 return bd;
 795
 796         /*
 797          * find the class device
 798          */
 799         mutex_lock(&bsg_mutex);
 800         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 801         mutex_unlock(&bsg_mutex);
 802
 803         if (!bcd)
 804                 return ERR_PTR(-ENODEV);
 805
 806         return bsg_add_device(inode, bcd->queue, file);
 807 }
 808
 809 static int bsg_open(struct inode *inode, struct file *file)
 810 {
 811         struct bsg_device *bd = bsg_get_device(inode, file);
 812
 813         if (IS_ERR(bd))
 814                 return PTR_ERR(bd);
 815
 816         file->private_data = bd;
 817         return 0;
 818 }
 819
 820 static int bsg_release(struct inode *inode, struct file *file)
 821 {
 822         struct bsg_device *bd = file->private_data;
 823
 824         file->private_data = NULL;
 825         return bsg_put_device(bd);
 826 }
 827
 828 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 829 {
 830         struct bsg_device *bd = file->private_data;
 831         unsigned int mask = 0;
 832
 833         poll_wait(file, &bd->wq_done, wait);
 834         poll_wait(file, &bd->wq_free, wait);
 835
 836         spin_lock_irq(&bd->lock);
 837         if (!list_empty(&bd->done_list))
 838                 mask |= POLLIN | POLLRDNORM;
 839         if (bd->queued_cmds >= bd->max_queue)
 840                 mask |= POLLOUT;
 841         spin_unlock_irq(&bd->lock);
 842
 843         return mask;
 844 }
 845
 846 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 847 {
 848         struct bsg_device *bd = file->private_data;
 849         int __user *uarg = (int __user *) arg;
 850         int ret;
 851
 852         switch (cmd) {
 853                 /*
 854                  * our own ioctls
 855                  */
 856         case SG_GET_COMMAND_Q:
 857                 return put_user(bd->max_queue, uarg);
 858         case SG_SET_COMMAND_Q: {
 859                 int queue;
 860
 861                 if (get_user(queue, uarg))
 862                         return -EFAULT;
 863                 if (queue < 1)
 864                         return -EINVAL;
 865
 866                 spin_lock_irq(&bd->lock);
 867                 bd->max_queue = queue;
 868                 spin_unlock_irq(&bd->lock);
 869                 return 0;
 870         }
 871
 872         /*
 873          * SCSI/sg ioctls
 874          */
 875         case SG_GET_VERSION_NUM:
 876         case SCSI_IOCTL_GET_IDLUN:
 877         case SCSI_IOCTL_GET_BUS_NUMBER:
 878         case SG_SET_TIMEOUT:
 879         case SG_GET_TIMEOUT:
 880         case SG_GET_RESERVED_SIZE:
 881         case SG_SET_RESERVED_SIZE:
 882         case SG_EMULATED_HOST:
 883         case SCSI_IOCTL_SEND_COMMAND: {
 884                 void __user *uarg = (void __user *) arg;
 885                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 886         }
 887         case SG_IO: {
 888                 struct request *rq;
 889                 struct bio *bio, *bidi_bio = NULL;
 890                 struct sg_io_v4 hdr;
 891
 892                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 893                         return -EFAULT;
 894
 895                 rq = bsg_map_hdr(bd, &hdr);
 896                 if (IS_ERR(rq))
 897                         return PTR_ERR(rq);
 898
 899                 bio = rq->bio;
 900                 if (rq->next_rq)
 901                         bidi_bio = rq->next_rq->bio;
 902                 blk_execute_rq(bd->queue, NULL, rq, 0);
 903                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 904
 905                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 906                         return -EFAULT;
 907
 908                 return ret;
 909         }
 910         /*
 911          * block device ioctls
 912          */
 913         default:
 914 #if 0
 915                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 916 #else
 917                 return -ENOTTY;
 918 #endif
 919         }
 920 }
 921
 922 static const struct file_operations bsg_fops = {
 923         .read           =       bsg_read,
 924         .write          =       bsg_write,
 925         .poll           =       bsg_poll,
 926         .open           =       bsg_open,
 927         .release        =       bsg_release,
 928         .unlocked_ioctl =       bsg_ioctl,
 929         .owner          =       THIS_MODULE,
 930 };
 931
 932 void bsg_unregister_queue(struct request_queue *q)
 933 {
 934         struct bsg_class_device *bcd = &q->bsg_dev;
 935
 936         if (!bcd->class_dev)
 937                 return;
 938
 939         mutex_lock(&bsg_mutex);
 940         idr_remove(&bsg_minor_idr, bcd->minor);
 941         sysfs_remove_link(&q->kobj, "bsg");
 942         class_device_unregister(bcd->class_dev);
 943         put_device(bcd->dev);
 944         bcd->class_dev = NULL;
 945         bcd->dev = NULL;
 946         mutex_unlock(&bsg_mutex);
 947 }
 948 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 949
 950 int bsg_register_queue(struct request_queue *q, struct device *gdev,
 951                        const char *name)
 952 {
 953         struct bsg_class_device *bcd;
 954         dev_t dev;
 955         int ret, minor;
 956         struct class_device *class_dev = NULL;
 957         const char *devname;
 958
 959         if (name)
 960                 devname = name;
 961         else
 962                 devname = gdev->bus_id;
 963
 964         /*
 965          * we need a proper transport to send commands, not a stacked device
 966          */
 967         if (!q->request_fn)
 968                 return 0;
 969
 970         bcd = &q->bsg_dev;
 971         memset(bcd, 0, sizeof(*bcd));
 972
 973         mutex_lock(&bsg_mutex);
 974
 975         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
 976         if (!ret) {
 977                 ret = -ENOMEM;
 978                 goto unlock;
 979         }
 980
 981         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
 982         if (ret < 0)
 983                 goto unlock;
 984
 985         if (minor >= BSG_MAX_DEVS) {
 986                 printk(KERN_ERR "bsg: too many bsg devices\n");
 987                 ret = -EINVAL;
 988                 goto remove_idr;
 989         }
 990
 991         bcd->minor = minor;
 992         bcd->queue = q;
 993         bcd->dev = get_device(gdev);
 994         dev = MKDEV(bsg_major, bcd->minor);
 995         class_dev = class_device_create(bsg_class, NULL, dev, gdev, "%s",
 996                                         devname);
 997         if (IS_ERR(class_dev)) {
 998                 ret = PTR_ERR(class_dev);
 999                 goto put_dev;
1000         }
1001         bcd->class_dev = class_dev;
1002
1003         if (q->kobj.sd) {
1004                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1005                 if (ret)
1006                         goto unregister_class_dev;
1007         }
1008
1009         mutex_unlock(&bsg_mutex);
1010         return 0;
1011
1012 unregister_class_dev:
1013         class_device_unregister(class_dev);
1014 put_dev:
1015         put_device(gdev);
1016 remove_idr:
1017         idr_remove(&bsg_minor_idr, minor);
1018 unlock:
1019         mutex_unlock(&bsg_mutex);
1020         return ret;
1021 }
1022 EXPORT_SYMBOL_GPL(bsg_register_queue);
1023
1024 static struct cdev bsg_cdev;
1025
1026 static int __init bsg_init(void)
1027 {
1028         int ret, i;
1029         dev_t devid;
1030
1031         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1032                                 sizeof(struct bsg_command), 0, 0, NULL);
1033         if (!bsg_cmd_cachep) {
1034                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1035                 return -ENOMEM;
1036         }
1037
1038         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1039                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1040
1041         bsg_class = class_create(THIS_MODULE, "bsg");
1042         if (IS_ERR(bsg_class)) {
1043                 ret = PTR_ERR(bsg_class);
1044                 goto destroy_kmemcache;
1045         }
1046
1047         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1048         if (ret)
1049                 goto destroy_bsg_class;
1050
1051         bsg_major = MAJOR(devid);
1052
1053         cdev_init(&bsg_cdev, &bsg_fops);
1054         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1055         if (ret)
1056                 goto unregister_chrdev;
1057
1058         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1059                " loaded (major %d)\n", bsg_major);
1060         return 0;
1061 unregister_chrdev:
1062         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1063 destroy_bsg_class:
1064         class_destroy(bsg_class);
1065 destroy_kmemcache:
1066         kmem_cache_destroy(bsg_cmd_cachep);
1067         return ret;
1068 }
1069
1070 MODULE_AUTHOR("Jens Axboe");
1071 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1072 MODULE_LICENSE("GPL");
1073
1074 device_initcall(bsg_init);