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, __func__, ##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         if (hdr->request_len > BLK_MAX_CDB) {
 178                 rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
 179                 if (!rq->cmd)
 180                         return -ENOMEM;
 181         }
 182
 183         if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
 184                            hdr->request_len))
 185                 return -EFAULT;
 186
 187         if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
 188                 if (blk_verify_command(rq->cmd, has_write_perm))
 189                         return -EPERM;
 190         } else if (!capable(CAP_SYS_RAWIO))
 191                 return -EPERM;
 192
 193         /*
 194          * fill in request structure
 195          */
 196         rq->cmd_len = hdr->request_len;
 197         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 198
 199         rq->timeout = (hdr->timeout * HZ) / 1000;
 200         if (!rq->timeout)
 201                 rq->timeout = q->sg_timeout;
 202         if (!rq->timeout)
 203                 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
 204
 205         return 0;
 206 }
 207
 208 /*
 209  * Check if sg_io_v4 from user is allowed and valid
 210  */
 211 static int
 212 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
 213 {
 214         int ret = 0;
 215
 216         if (hdr->guard != 'Q')
 217                 return -EINVAL;
 218         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 219             hdr->din_xfer_len > (q->max_sectors << 9))
 220                 return -EIO;
 221
 222         switch (hdr->protocol) {
 223         case BSG_PROTOCOL_SCSI:
 224                 switch (hdr->subprotocol) {
 225                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 226                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 227                         break;
 228                 default:
 229                         ret = -EINVAL;
 230                 }
 231                 break;
 232         default:
 233                 ret = -EINVAL;
 234         }
 235
 236         *rw = hdr->dout_xfer_len ? WRITE : READ;
 237         return ret;
 238 }
 239
 240 /*
 241  * map sg_io_v4 to a request.
 242  */
 243 static struct request *
 244 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
 245 {
 246         struct request_queue *q = bd->queue;
 247         struct request *rq, *next_rq = NULL;
 248         int ret, rw;
 249         unsigned int dxfer_len;
 250         void *dxferp = NULL;
 251
 252         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 253                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 254                 hdr->din_xfer_len);
 255
 256         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 257         if (ret)
 258                 return ERR_PTR(ret);
 259
 260         /*
 261          * map scatter-gather elements seperately and string them to request
 262          */
 263         rq = blk_get_request(q, rw, GFP_KERNEL);
 264         if (!rq)
 265                 return ERR_PTR(-ENOMEM);
 266         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
 267                                                        &bd->flags));
 268         if (ret)
 269                 goto out;
 270
 271         if (rw == WRITE && hdr->din_xfer_len) {
 272                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 273                         ret = -EOPNOTSUPP;
 274                         goto out;
 275                 }
 276
 277                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 278                 if (!next_rq) {
 279                         ret = -ENOMEM;
 280                         goto out;
 281                 }
 282                 rq->next_rq = next_rq;
 283                 next_rq->cmd_type = rq->cmd_type;
 284
 285                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 286                 ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
 287                 if (ret)
 288                         goto out;
 289         }
 290
 291         if (hdr->dout_xfer_len) {
 292                 dxfer_len = hdr->dout_xfer_len;
 293                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 294         } else if (hdr->din_xfer_len) {
 295                 dxfer_len = hdr->din_xfer_len;
 296                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 297         } else
 298                 dxfer_len = 0;
 299
 300         if (dxfer_len) {
 301                 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
 302                 if (ret)
 303                         goto out;
 304         }
 305         return rq;
 306 out:
 307         if (rq->cmd != rq->__cmd)
 308                 kfree(rq->cmd);
 309         blk_put_request(rq);
 310         if (next_rq) {
 311                 blk_rq_unmap_user(next_rq->bio);
 312                 blk_put_request(next_rq);
 313         }
 314         return ERR_PTR(ret);
 315 }
 316
 317 /*
 318  * async completion call-back from the block layer, when scsi/ide/whatever
 319  * calls end_that_request_last() on a request
 320  */
 321 static void bsg_rq_end_io(struct request *rq, int uptodate)
 322 {
 323         struct bsg_command *bc = rq->end_io_data;
 324         struct bsg_device *bd = bc->bd;
 325         unsigned long flags;
 326
 327         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 328                 bd->name, rq, bc, bc->bio, uptodate);
 329
 330         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 331
 332         spin_lock_irqsave(&bd->lock, flags);
 333         list_move_tail(&bc->list, &bd->done_list);
 334         bd->done_cmds++;
 335         spin_unlock_irqrestore(&bd->lock, flags);
 336
 337         wake_up(&bd->wq_done);
 338 }
 339
 340 /*
 341  * do final setup of a 'bc' and submit the matching 'rq' to the block
 342  * layer for io
 343  */
 344 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
 345                             struct bsg_command *bc, struct request *rq)
 346 {
 347         rq->sense = bc->sense;
 348         rq->sense_len = 0;
 349
 350         /*
 351          * add bc command to busy queue and submit rq for io
 352          */
 353         bc->rq = rq;
 354         bc->bio = rq->bio;
 355         if (rq->next_rq)
 356                 bc->bidi_bio = rq->next_rq->bio;
 357         bc->hdr.duration = jiffies;
 358         spin_lock_irq(&bd->lock);
 359         list_add_tail(&bc->list, &bd->busy_list);
 360         spin_unlock_irq(&bd->lock);
 361
 362         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 363
 364         rq->end_io_data = bc;
 365         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 366 }
 367
 368 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 369 {
 370         struct bsg_command *bc = NULL;
 371
 372         spin_lock_irq(&bd->lock);
 373         if (bd->done_cmds) {
 374                 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
 375                 list_del(&bc->list);
 376                 bd->done_cmds--;
 377         }
 378         spin_unlock_irq(&bd->lock);
 379
 380         return bc;
 381 }
 382
 383 /*
 384  * Get a finished command from the done list
 385  */
 386 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 387 {
 388         struct bsg_command *bc;
 389         int ret;
 390
 391         do {
 392                 bc = bsg_next_done_cmd(bd);
 393                 if (bc)
 394                         break;
 395
 396                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 397                         bc = ERR_PTR(-EAGAIN);
 398                         break;
 399                 }
 400
 401                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 402                 if (ret) {
 403                         bc = ERR_PTR(-ERESTARTSYS);
 404                         break;
 405                 }
 406         } while (1);
 407
 408         dprintk("%s: returning done %p\n", bd->name, bc);
 409
 410         return bc;
 411 }
 412
 413 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 414                                     struct bio *bio, struct bio *bidi_bio)
 415 {
 416         int ret = 0;
 417
 418         dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
 419         /*
 420          * fill in all the output members
 421          */
 422         hdr->device_status = status_byte(rq->errors);
 423         hdr->transport_status = host_byte(rq->errors);
 424         hdr->driver_status = driver_byte(rq->errors);
 425         hdr->info = 0;
 426         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 427                 hdr->info |= SG_INFO_CHECK;
 428         hdr->response_len = 0;
 429
 430         if (rq->sense_len && hdr->response) {
 431                 int len = min_t(unsigned int, hdr->max_response_len,
 432                                         rq->sense_len);
 433
 434                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 435                                    rq->sense, len);
 436                 if (!ret)
 437                         hdr->response_len = len;
 438                 else
 439                         ret = -EFAULT;
 440         }
 441
 442         if (rq->next_rq) {
 443                 hdr->dout_resid = rq->data_len;
 444                 hdr->din_resid = rq->next_rq->data_len;
 445                 blk_rq_unmap_user(bidi_bio);
 446                 blk_put_request(rq->next_rq);
 447         } else if (rq_data_dir(rq) == READ)
 448                 hdr->din_resid = rq->data_len;
 449         else
 450                 hdr->dout_resid = rq->data_len;
 451
 452         /*
 453          * If the request generated a negative error number, return it
 454          * (providing we aren't already returning an error); if it's
 455          * just a protocol response (i.e. non negative), that gets
 456          * processed above.
 457          */
 458         if (!ret && rq->errors < 0)
 459                 ret = rq->errors;
 460
 461         blk_rq_unmap_user(bio);
 462         if (rq->cmd != rq->__cmd)
 463                 kfree(rq->cmd);
 464         blk_put_request(rq);
 465
 466         return ret;
 467 }
 468
 469 static int bsg_complete_all_commands(struct bsg_device *bd)
 470 {
 471         struct bsg_command *bc;
 472         int ret, tret;
 473
 474         dprintk("%s: entered\n", bd->name);
 475
 476         /*
 477          * wait for all commands to complete
 478          */
 479         ret = 0;
 480         do {
 481                 ret = bsg_io_schedule(bd);
 482                 /*
 483                  * look for -ENODATA specifically -- we'll sometimes get
 484                  * -ERESTARTSYS when we've taken a signal, but we can't
 485                  * return until we're done freeing the queue, so ignore
 486                  * it.  The signal will get handled when we're done freeing
 487                  * the bsg_device.
 488                  */
 489         } while (ret != -ENODATA);
 490
 491         /*
 492          * discard done commands
 493          */
 494         ret = 0;
 495         do {
 496                 spin_lock_irq(&bd->lock);
 497                 if (!bd->queued_cmds) {
 498                         spin_unlock_irq(&bd->lock);
 499                         break;
 500                 }
 501                 spin_unlock_irq(&bd->lock);
 502
 503                 bc = bsg_get_done_cmd(bd);
 504                 if (IS_ERR(bc))
 505                         break;
 506
 507                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 508                                                 bc->bidi_bio);
 509                 if (!ret)
 510                         ret = tret;
 511
 512                 bsg_free_command(bc);
 513         } while (1);
 514
 515         return ret;
 516 }
 517
 518 static int
 519 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 520            const struct iovec *iov, ssize_t *bytes_read)
 521 {
 522         struct bsg_command *bc;
 523         int nr_commands, ret;
 524
 525         if (count % sizeof(struct sg_io_v4))
 526                 return -EINVAL;
 527
 528         ret = 0;
 529         nr_commands = count / sizeof(struct sg_io_v4);
 530         while (nr_commands) {
 531                 bc = bsg_get_done_cmd(bd);
 532                 if (IS_ERR(bc)) {
 533                         ret = PTR_ERR(bc);
 534                         break;
 535                 }
 536
 537                 /*
 538                  * this is the only case where we need to copy data back
 539                  * after completing the request. so do that here,
 540                  * bsg_complete_work() cannot do that for us
 541                  */
 542                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 543                                                bc->bidi_bio);
 544
 545                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 546                         ret = -EFAULT;
 547
 548                 bsg_free_command(bc);
 549
 550                 if (ret)
 551                         break;
 552
 553                 buf += sizeof(struct sg_io_v4);
 554                 *bytes_read += sizeof(struct sg_io_v4);
 555                 nr_commands--;
 556         }
 557
 558         return ret;
 559 }
 560
 561 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 562 {
 563         if (file->f_flags & O_NONBLOCK)
 564                 clear_bit(BSG_F_BLOCK, &bd->flags);
 565         else
 566                 set_bit(BSG_F_BLOCK, &bd->flags);
 567 }
 568
 569 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
 570 {
 571         if (file->f_mode & FMODE_WRITE)
 572                 set_bit(BSG_F_WRITE_PERM, &bd->flags);
 573         else
 574                 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
 575 }
 576
 577 /*
 578  * Check if the error is a "real" error that we should return.
 579  */
 580 static inline int err_block_err(int ret)
 581 {
 582         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 583                 return 1;
 584
 585         return 0;
 586 }
 587
 588 static ssize_t
 589 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 590 {
 591         struct bsg_device *bd = file->private_data;
 592         int ret;
 593         ssize_t bytes_read;
 594
 595         dprintk("%s: read %Zd bytes\n", bd->name, count);
 596
 597         bsg_set_block(bd, file);
 598         bytes_read = 0;
 599         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 600         *ppos = bytes_read;
 601
 602         if (!bytes_read || (bytes_read && err_block_err(ret)))
 603                 bytes_read = ret;
 604
 605         return bytes_read;
 606 }
 607
 608 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 609                        size_t count, ssize_t *bytes_written)
 610 {
 611         struct bsg_command *bc;
 612         struct request *rq;
 613         int ret, nr_commands;
 614
 615         if (count % sizeof(struct sg_io_v4))
 616                 return -EINVAL;
 617
 618         nr_commands = count / sizeof(struct sg_io_v4);
 619         rq = NULL;
 620         bc = NULL;
 621         ret = 0;
 622         while (nr_commands) {
 623                 struct request_queue *q = bd->queue;
 624
 625                 bc = bsg_alloc_command(bd);
 626                 if (IS_ERR(bc)) {
 627                         ret = PTR_ERR(bc);
 628                         bc = NULL;
 629                         break;
 630                 }
 631
 632                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 633                         ret = -EFAULT;
 634                         break;
 635                 }
 636
 637                 /*
 638                  * get a request, fill in the blanks, and add to request queue
 639                  */
 640                 rq = bsg_map_hdr(bd, &bc->hdr);
 641                 if (IS_ERR(rq)) {
 642                         ret = PTR_ERR(rq);
 643                         rq = NULL;
 644                         break;
 645                 }
 646
 647                 bsg_add_command(bd, q, bc, rq);
 648                 bc = NULL;
 649                 rq = NULL;
 650                 nr_commands--;
 651                 buf += sizeof(struct sg_io_v4);
 652                 *bytes_written += sizeof(struct sg_io_v4);
 653         }
 654
 655         if (bc)
 656                 bsg_free_command(bc);
 657
 658         return ret;
 659 }
 660
 661 static ssize_t
 662 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 663 {
 664         struct bsg_device *bd = file->private_data;
 665         ssize_t bytes_written;
 666         int ret;
 667
 668         dprintk("%s: write %Zd bytes\n", bd->name, count);
 669
 670         bsg_set_block(bd, file);
 671         bsg_set_write_perm(bd, file);
 672
 673         bytes_written = 0;
 674         ret = __bsg_write(bd, buf, count, &bytes_written);
 675         *ppos = bytes_written;
 676
 677         /*
 678          * return bytes written on non-fatal errors
 679          */
 680         if (!bytes_written || (bytes_written && err_block_err(ret)))
 681                 bytes_written = ret;
 682
 683         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 684         return bytes_written;
 685 }
 686
 687 static struct bsg_device *bsg_alloc_device(void)
 688 {
 689         struct bsg_device *bd;
 690
 691         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 692         if (unlikely(!bd))
 693                 return NULL;
 694
 695         spin_lock_init(&bd->lock);
 696
 697         bd->max_queue = BSG_DEFAULT_CMDS;
 698
 699         INIT_LIST_HEAD(&bd->busy_list);
 700         INIT_LIST_HEAD(&bd->done_list);
 701         INIT_HLIST_NODE(&bd->dev_list);
 702
 703         init_waitqueue_head(&bd->wq_free);
 704         init_waitqueue_head(&bd->wq_done);
 705         return bd;
 706 }
 707
 708 static void bsg_kref_release_function(struct kref *kref)
 709 {
 710         struct bsg_class_device *bcd =
 711                 container_of(kref, struct bsg_class_device, ref);
 712
 713         if (bcd->release)
 714                 bcd->release(bcd->parent);
 715
 716         put_device(bcd->parent);
 717 }
 718
 719 static int bsg_put_device(struct bsg_device *bd)
 720 {
 721         int ret = 0, do_free;
 722         struct request_queue *q = bd->queue;
 723
 724         mutex_lock(&bsg_mutex);
 725
 726         do_free = atomic_dec_and_test(&bd->ref_count);
 727         if (!do_free)
 728                 goto out;
 729
 730         dprintk("%s: tearing down\n", bd->name);
 731
 732         /*
 733          * close can always block
 734          */
 735         set_bit(BSG_F_BLOCK, &bd->flags);
 736
 737         /*
 738          * correct error detection baddies here again. it's the responsibility
 739          * of the app to properly reap commands before close() if it wants
 740          * fool-proof error detection
 741          */
 742         ret = bsg_complete_all_commands(bd);
 743
 744         hlist_del(&bd->dev_list);
 745         kfree(bd);
 746 out:
 747         mutex_unlock(&bsg_mutex);
 748         kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
 749         if (do_free)
 750                 blk_put_queue(q);
 751         return ret;
 752 }
 753
 754 static struct bsg_device *bsg_add_device(struct inode *inode,
 755                                          struct request_queue *rq,
 756                                          struct file *file)
 757 {
 758         struct bsg_device *bd;
 759         int ret;
 760 #ifdef BSG_DEBUG
 761         unsigned char buf[32];
 762 #endif
 763         ret = blk_get_queue(rq);
 764         if (ret)
 765                 return ERR_PTR(-ENXIO);
 766
 767         bd = bsg_alloc_device();
 768         if (!bd) {
 769                 blk_put_queue(rq);
 770                 return ERR_PTR(-ENOMEM);
 771         }
 772
 773         bd->queue = rq;
 774         bsg_set_block(bd, file);
 775
 776         atomic_set(&bd->ref_count, 1);
 777         mutex_lock(&bsg_mutex);
 778         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
 779
 780         strncpy(bd->name, rq->bsg_dev.class_dev->bus_id, sizeof(bd->name) - 1);
 781         dprintk("bound to <%s>, max queue %d\n",
 782                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 783
 784         mutex_unlock(&bsg_mutex);
 785         return bd;
 786 }
 787
 788 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
 789 {
 790         struct bsg_device *bd;
 791         struct hlist_node *entry;
 792
 793         mutex_lock(&bsg_mutex);
 794
 795         hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
 796                 if (bd->queue == q) {
 797                         atomic_inc(&bd->ref_count);
 798                         goto found;
 799                 }
 800         }
 801         bd = NULL;
 802 found:
 803         mutex_unlock(&bsg_mutex);
 804         return bd;
 805 }
 806
 807 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 808 {
 809         struct bsg_device *bd;
 810         struct bsg_class_device *bcd;
 811
 812         /*
 813          * find the class device
 814          */
 815         mutex_lock(&bsg_mutex);
 816         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 817         if (bcd)
 818                 kref_get(&bcd->ref);
 819         mutex_unlock(&bsg_mutex);
 820
 821         if (!bcd)
 822                 return ERR_PTR(-ENODEV);
 823
 824         bd = __bsg_get_device(iminor(inode), bcd->queue);
 825         if (bd)
 826                 return bd;
 827
 828         bd = bsg_add_device(inode, bcd->queue, file);
 829         if (IS_ERR(bd))
 830                 kref_put(&bcd->ref, bsg_kref_release_function);
 831
 832         return bd;
 833 }
 834
 835 static int bsg_open(struct inode *inode, struct file *file)
 836 {
 837         struct bsg_device *bd = bsg_get_device(inode, file);
 838
 839         if (IS_ERR(bd))
 840                 return PTR_ERR(bd);
 841
 842         file->private_data = bd;
 843         return 0;
 844 }
 845
 846 static int bsg_release(struct inode *inode, struct file *file)
 847 {
 848         struct bsg_device *bd = file->private_data;
 849
 850         file->private_data = NULL;
 851         return bsg_put_device(bd);
 852 }
 853
 854 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 855 {
 856         struct bsg_device *bd = file->private_data;
 857         unsigned int mask = 0;
 858
 859         poll_wait(file, &bd->wq_done, wait);
 860         poll_wait(file, &bd->wq_free, wait);
 861
 862         spin_lock_irq(&bd->lock);
 863         if (!list_empty(&bd->done_list))
 864                 mask |= POLLIN | POLLRDNORM;
 865         if (bd->queued_cmds >= bd->max_queue)
 866                 mask |= POLLOUT;
 867         spin_unlock_irq(&bd->lock);
 868
 869         return mask;
 870 }
 871
 872 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 873 {
 874         struct bsg_device *bd = file->private_data;
 875         int __user *uarg = (int __user *) arg;
 876         int ret;
 877
 878         switch (cmd) {
 879                 /*
 880                  * our own ioctls
 881                  */
 882         case SG_GET_COMMAND_Q:
 883                 return put_user(bd->max_queue, uarg);
 884         case SG_SET_COMMAND_Q: {
 885                 int queue;
 886
 887                 if (get_user(queue, uarg))
 888                         return -EFAULT;
 889                 if (queue < 1)
 890                         return -EINVAL;
 891
 892                 spin_lock_irq(&bd->lock);
 893                 bd->max_queue = queue;
 894                 spin_unlock_irq(&bd->lock);
 895                 return 0;
 896         }
 897
 898         /*
 899          * SCSI/sg ioctls
 900          */
 901         case SG_GET_VERSION_NUM:
 902         case SCSI_IOCTL_GET_IDLUN:
 903         case SCSI_IOCTL_GET_BUS_NUMBER:
 904         case SG_SET_TIMEOUT:
 905         case SG_GET_TIMEOUT:
 906         case SG_GET_RESERVED_SIZE:
 907         case SG_SET_RESERVED_SIZE:
 908         case SG_EMULATED_HOST:
 909         case SCSI_IOCTL_SEND_COMMAND: {
 910                 void __user *uarg = (void __user *) arg;
 911                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 912         }
 913         case SG_IO: {
 914                 struct request *rq;
 915                 struct bio *bio, *bidi_bio = NULL;
 916                 struct sg_io_v4 hdr;
 917
 918                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 919                         return -EFAULT;
 920
 921                 rq = bsg_map_hdr(bd, &hdr);
 922                 if (IS_ERR(rq))
 923                         return PTR_ERR(rq);
 924
 925                 bio = rq->bio;
 926                 if (rq->next_rq)
 927                         bidi_bio = rq->next_rq->bio;
 928                 blk_execute_rq(bd->queue, NULL, rq, 0);
 929                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 930
 931                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 932                         return -EFAULT;
 933
 934                 return ret;
 935         }
 936         /*
 937          * block device ioctls
 938          */
 939         default:
 940 #if 0
 941                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 942 #else
 943                 return -ENOTTY;
 944 #endif
 945         }
 946 }
 947
 948 static const struct file_operations bsg_fops = {
 949         .read           =       bsg_read,
 950         .write          =       bsg_write,
 951         .poll           =       bsg_poll,
 952         .open           =       bsg_open,
 953         .release        =       bsg_release,
 954         .unlocked_ioctl =       bsg_ioctl,
 955         .owner          =       THIS_MODULE,
 956 };
 957
 958 void bsg_unregister_queue(struct request_queue *q)
 959 {
 960         struct bsg_class_device *bcd = &q->bsg_dev;
 961
 962         if (!bcd->class_dev)
 963                 return;
 964
 965         mutex_lock(&bsg_mutex);
 966         idr_remove(&bsg_minor_idr, bcd->minor);
 967         sysfs_remove_link(&q->kobj, "bsg");
 968         device_unregister(bcd->class_dev);
 969         bcd->class_dev = NULL;
 970         kref_put(&bcd->ref, bsg_kref_release_function);
 971         mutex_unlock(&bsg_mutex);
 972 }
 973 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 974
 975 int bsg_register_queue(struct request_queue *q, struct device *parent,
 976                        const char *name, void (*release)(struct device *))
 977 {
 978         struct bsg_class_device *bcd;
 979         dev_t dev;
 980         int ret, minor;
 981         struct device *class_dev = NULL;
 982         const char *devname;
 983
 984         if (name)
 985                 devname = name;
 986         else
 987                 devname = parent->bus_id;
 988
 989         /*
 990          * we need a proper transport to send commands, not a stacked device
 991          */
 992         if (!q->request_fn)
 993                 return 0;
 994
 995         bcd = &q->bsg_dev;
 996         memset(bcd, 0, sizeof(*bcd));
 997
 998         mutex_lock(&bsg_mutex);
 999
1000         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1001         if (!ret) {
1002                 ret = -ENOMEM;
1003                 goto unlock;
1004         }
1005
1006         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1007         if (ret < 0)
1008                 goto unlock;
1009
1010         if (minor >= BSG_MAX_DEVS) {
1011                 printk(KERN_ERR "bsg: too many bsg devices\n");
1012                 ret = -EINVAL;
1013                 goto remove_idr;
1014         }
1015
1016         bcd->minor = minor;
1017         bcd->queue = q;
1018         bcd->parent = get_device(parent);
1019         bcd->release = release;
1020         kref_init(&bcd->ref);
1021         dev = MKDEV(bsg_major, bcd->minor);
1022         class_dev = device_create(bsg_class, parent, dev, "%s", devname);
1023         if (IS_ERR(class_dev)) {
1024                 ret = PTR_ERR(class_dev);
1025                 goto put_dev;
1026         }
1027         bcd->class_dev = class_dev;
1028
1029         if (q->kobj.sd) {
1030                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1031                 if (ret)
1032                         goto unregister_class_dev;
1033         }
1034
1035         mutex_unlock(&bsg_mutex);
1036         return 0;
1037
1038 unregister_class_dev:
1039         device_unregister(class_dev);
1040 put_dev:
1041         put_device(parent);
1042 remove_idr:
1043         idr_remove(&bsg_minor_idr, minor);
1044 unlock:
1045         mutex_unlock(&bsg_mutex);
1046         return ret;
1047 }
1048 EXPORT_SYMBOL_GPL(bsg_register_queue);
1049
1050 static struct cdev bsg_cdev;
1051
1052 static int __init bsg_init(void)
1053 {
1054         int ret, i;
1055         dev_t devid;
1056
1057         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1058                                 sizeof(struct bsg_command), 0, 0, NULL);
1059         if (!bsg_cmd_cachep) {
1060                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1061                 return -ENOMEM;
1062         }
1063
1064         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1065                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1066
1067         bsg_class = class_create(THIS_MODULE, "bsg");
1068         if (IS_ERR(bsg_class)) {
1069                 ret = PTR_ERR(bsg_class);
1070                 goto destroy_kmemcache;
1071         }
1072
1073         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1074         if (ret)
1075                 goto destroy_bsg_class;
1076
1077         bsg_major = MAJOR(devid);
1078
1079         cdev_init(&bsg_cdev, &bsg_fops);
1080         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1081         if (ret)
1082                 goto unregister_chrdev;
1083
1084         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1085                " loaded (major %d)\n", bsg_major);
1086         return 0;
1087 unregister_chrdev:
1088         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1089 destroy_bsg_class:
1090         class_destroy(bsg_class);
1091 destroy_kmemcache:
1092         kmem_cache_destroy(bsg_cmd_cachep);
1093         return ret;
1094 }
1095
1096 MODULE_AUTHOR("Jens Axboe");
1097 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1098 MODULE_LICENSE("GPL");
1099
1100 device_initcall(bsg_init);