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 #include <linux/smp_lock.h>
  23
  24 #include <scsi/scsi.h>
  25 #include <scsi/scsi_ioctl.h>
  26 #include <scsi/scsi_cmnd.h>
  27 #include <scsi/scsi_device.h>
  28 #include <scsi/scsi_driver.h>
  29 #include <scsi/sg.h>
  30
  31 #define BSG_DESCRIPTION "Block layer SCSI generic (bsg) driver"
  32 #define BSG_VERSION     "0.4"
  33
  34 struct bsg_device {
  35         struct request_queue *queue;
  36         spinlock_t lock;
  37         struct list_head busy_list;
  38         struct list_head done_list;
  39         struct hlist_node dev_list;
  40         atomic_t ref_count;
  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 };
  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, struct bsg_device *bd,
 176                                 int has_write_perm)
 177 {
 178         if (hdr->request_len > BLK_MAX_CDB) {
 179                 rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
 180                 if (!rq->cmd)
 181                         return -ENOMEM;
 182         }
 183
 184         if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
 185                            hdr->request_len))
 186                 return -EFAULT;
 187
 188         if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
 189                 if (blk_verify_command(&q->cmd_filter, rq->cmd, has_write_perm))
 190                         return -EPERM;
 191         } else if (!capable(CAP_SYS_RAWIO))
 192                 return -EPERM;
 193
 194         /*
 195          * fill in request structure
 196          */
 197         rq->cmd_len = hdr->request_len;
 198         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 199
 200         rq->timeout = (hdr->timeout * HZ) / 1000;
 201         if (!rq->timeout)
 202                 rq->timeout = q->sg_timeout;
 203         if (!rq->timeout)
 204                 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
 205
 206         return 0;
 207 }
 208
 209 /*
 210  * Check if sg_io_v4 from user is allowed and valid
 211  */
 212 static int
 213 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
 214 {
 215         int ret = 0;
 216
 217         if (hdr->guard != 'Q')
 218                 return -EINVAL;
 219         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 220             hdr->din_xfer_len > (q->max_sectors << 9))
 221                 return -EIO;
 222
 223         switch (hdr->protocol) {
 224         case BSG_PROTOCOL_SCSI:
 225                 switch (hdr->subprotocol) {
 226                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 227                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 228                         break;
 229                 default:
 230                         ret = -EINVAL;
 231                 }
 232                 break;
 233         default:
 234                 ret = -EINVAL;
 235         }
 236
 237         *rw = hdr->dout_xfer_len ? WRITE : READ;
 238         return ret;
 239 }
 240
 241 /*
 242  * map sg_io_v4 to a request.
 243  */
 244 static struct request *
 245 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, int has_write_perm)
 246 {
 247         struct request_queue *q = bd->queue;
 248         struct request *rq, *next_rq = NULL;
 249         int ret, rw;
 250         unsigned int dxfer_len;
 251         void *dxferp = NULL;
 252
 253         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 254                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 255                 hdr->din_xfer_len);
 256
 257         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 258         if (ret)
 259                 return ERR_PTR(ret);
 260
 261         /*
 262          * map scatter-gather elements seperately and string them to request
 263          */
 264         rq = blk_get_request(q, rw, GFP_KERNEL);
 265         if (!rq)
 266                 return ERR_PTR(-ENOMEM);
 267         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
 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                                        GFP_KERNEL);
 288                 if (ret)
 289                         goto out;
 290         }
 291
 292         if (hdr->dout_xfer_len) {
 293                 dxfer_len = hdr->dout_xfer_len;
 294                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 295         } else if (hdr->din_xfer_len) {
 296                 dxfer_len = hdr->din_xfer_len;
 297                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 298         } else
 299                 dxfer_len = 0;
 300
 301         if (dxfer_len) {
 302                 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len, GFP_KERNEL);
 303                 if (ret)
 304                         goto out;
 305         }
 306         return rq;
 307 out:
 308         if (rq->cmd != rq->__cmd)
 309                 kfree(rq->cmd);
 310         blk_put_request(rq);
 311         if (next_rq) {
 312                 blk_rq_unmap_user(next_rq->bio);
 313                 blk_put_request(next_rq);
 314         }
 315         return ERR_PTR(ret);
 316 }
 317
 318 /*
 319  * async completion call-back from the block layer, when scsi/ide/whatever
 320  * calls end_that_request_last() on a request
 321  */
 322 static void bsg_rq_end_io(struct request *rq, int uptodate)
 323 {
 324         struct bsg_command *bc = rq->end_io_data;
 325         struct bsg_device *bd = bc->bd;
 326         unsigned long flags;
 327
 328         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 329                 bd->name, rq, bc, bc->bio, uptodate);
 330
 331         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 332
 333         spin_lock_irqsave(&bd->lock, flags);
 334         list_move_tail(&bc->list, &bd->done_list);
 335         bd->done_cmds++;
 336         spin_unlock_irqrestore(&bd->lock, flags);
 337
 338         wake_up(&bd->wq_done);
 339 }
 340
 341 /*
 342  * do final setup of a 'bc' and submit the matching 'rq' to the block
 343  * layer for io
 344  */
 345 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
 346                             struct bsg_command *bc, struct request *rq)
 347 {
 348         rq->sense = bc->sense;
 349         rq->sense_len = 0;
 350
 351         /*
 352          * add bc command to busy queue and submit rq for io
 353          */
 354         bc->rq = rq;
 355         bc->bio = rq->bio;
 356         if (rq->next_rq)
 357                 bc->bidi_bio = rq->next_rq->bio;
 358         bc->hdr.duration = jiffies;
 359         spin_lock_irq(&bd->lock);
 360         list_add_tail(&bc->list, &bd->busy_list);
 361         spin_unlock_irq(&bd->lock);
 362
 363         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 364
 365         rq->end_io_data = bc;
 366         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 367 }
 368
 369 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 370 {
 371         struct bsg_command *bc = NULL;
 372
 373         spin_lock_irq(&bd->lock);
 374         if (bd->done_cmds) {
 375                 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
 376                 list_del(&bc->list);
 377                 bd->done_cmds--;
 378         }
 379         spin_unlock_irq(&bd->lock);
 380
 381         return bc;
 382 }
 383
 384 /*
 385  * Get a finished command from the done list
 386  */
 387 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 388 {
 389         struct bsg_command *bc;
 390         int ret;
 391
 392         do {
 393                 bc = bsg_next_done_cmd(bd);
 394                 if (bc)
 395                         break;
 396
 397                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 398                         bc = ERR_PTR(-EAGAIN);
 399                         break;
 400                 }
 401
 402                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 403                 if (ret) {
 404                         bc = ERR_PTR(-ERESTARTSYS);
 405                         break;
 406                 }
 407         } while (1);
 408
 409         dprintk("%s: returning done %p\n", bd->name, bc);
 410
 411         return bc;
 412 }
 413
 414 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 415                                     struct bio *bio, struct bio *bidi_bio)
 416 {
 417         int ret = 0;
 418
 419         dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
 420         /*
 421          * fill in all the output members
 422          */
 423         hdr->device_status = status_byte(rq->errors);
 424         hdr->transport_status = host_byte(rq->errors);
 425         hdr->driver_status = driver_byte(rq->errors);
 426         hdr->info = 0;
 427         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 428                 hdr->info |= SG_INFO_CHECK;
 429         hdr->response_len = 0;
 430
 431         if (rq->sense_len && hdr->response) {
 432                 int len = min_t(unsigned int, hdr->max_response_len,
 433                                         rq->sense_len);
 434
 435                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 436                                    rq->sense, len);
 437                 if (!ret)
 438                         hdr->response_len = len;
 439                 else
 440                         ret = -EFAULT;
 441         }
 442
 443         if (rq->next_rq) {
 444                 hdr->dout_resid = rq->data_len;
 445                 hdr->din_resid = rq->next_rq->data_len;
 446                 blk_rq_unmap_user(bidi_bio);
 447                 blk_put_request(rq->next_rq);
 448         } else if (rq_data_dir(rq) == READ)
 449                 hdr->din_resid = rq->data_len;
 450         else
 451                 hdr->dout_resid = rq->data_len;
 452
 453         /*
 454          * If the request generated a negative error number, return it
 455          * (providing we aren't already returning an error); if it's
 456          * just a protocol response (i.e. non negative), that gets
 457          * processed above.
 458          */
 459         if (!ret && rq->errors < 0)
 460                 ret = rq->errors;
 461
 462         blk_rq_unmap_user(bio);
 463         if (rq->cmd != rq->__cmd)
 464                 kfree(rq->cmd);
 465         blk_put_request(rq);
 466
 467         return ret;
 468 }
 469
 470 static int bsg_complete_all_commands(struct bsg_device *bd)
 471 {
 472         struct bsg_command *bc;
 473         int ret, tret;
 474
 475         dprintk("%s: entered\n", bd->name);
 476
 477         /*
 478          * wait for all commands to complete
 479          */
 480         ret = 0;
 481         do {
 482                 ret = bsg_io_schedule(bd);
 483                 /*
 484                  * look for -ENODATA specifically -- we'll sometimes get
 485                  * -ERESTARTSYS when we've taken a signal, but we can't
 486                  * return until we're done freeing the queue, so ignore
 487                  * it.  The signal will get handled when we're done freeing
 488                  * the bsg_device.
 489                  */
 490         } while (ret != -ENODATA);
 491
 492         /*
 493          * discard done commands
 494          */
 495         ret = 0;
 496         do {
 497                 spin_lock_irq(&bd->lock);
 498                 if (!bd->queued_cmds) {
 499                         spin_unlock_irq(&bd->lock);
 500                         break;
 501                 }
 502                 spin_unlock_irq(&bd->lock);
 503
 504                 bc = bsg_get_done_cmd(bd);
 505                 if (IS_ERR(bc))
 506                         break;
 507
 508                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 509                                                 bc->bidi_bio);
 510                 if (!ret)
 511                         ret = tret;
 512
 513                 bsg_free_command(bc);
 514         } while (1);
 515
 516         return ret;
 517 }
 518
 519 static int
 520 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 521            const struct iovec *iov, ssize_t *bytes_read)
 522 {
 523         struct bsg_command *bc;
 524         int nr_commands, ret;
 525
 526         if (count % sizeof(struct sg_io_v4))
 527                 return -EINVAL;
 528
 529         ret = 0;
 530         nr_commands = count / sizeof(struct sg_io_v4);
 531         while (nr_commands) {
 532                 bc = bsg_get_done_cmd(bd);
 533                 if (IS_ERR(bc)) {
 534                         ret = PTR_ERR(bc);
 535                         break;
 536                 }
 537
 538                 /*
 539                  * this is the only case where we need to copy data back
 540                  * after completing the request. so do that here,
 541                  * bsg_complete_work() cannot do that for us
 542                  */
 543                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 544                                                bc->bidi_bio);
 545
 546                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 547                         ret = -EFAULT;
 548
 549                 bsg_free_command(bc);
 550
 551                 if (ret)
 552                         break;
 553
 554                 buf += sizeof(struct sg_io_v4);
 555                 *bytes_read += sizeof(struct sg_io_v4);
 556                 nr_commands--;
 557         }
 558
 559         return ret;
 560 }
 561
 562 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 563 {
 564         if (file->f_flags & O_NONBLOCK)
 565                 clear_bit(BSG_F_BLOCK, &bd->flags);
 566         else
 567                 set_bit(BSG_F_BLOCK, &bd->flags);
 568 }
 569
 570 /*
 571  * Check if the error is a "real" error that we should return.
 572  */
 573 static inline int err_block_err(int ret)
 574 {
 575         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 576                 return 1;
 577
 578         return 0;
 579 }
 580
 581 static ssize_t
 582 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 583 {
 584         struct bsg_device *bd = file->private_data;
 585         int ret;
 586         ssize_t bytes_read;
 587
 588         dprintk("%s: read %Zd bytes\n", bd->name, count);
 589
 590         bsg_set_block(bd, file);
 591
 592         bytes_read = 0;
 593         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 594         *ppos = bytes_read;
 595
 596         if (!bytes_read || (bytes_read && err_block_err(ret)))
 597                 bytes_read = ret;
 598
 599         return bytes_read;
 600 }
 601
 602 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 603                        size_t count, ssize_t *bytes_written, int has_write_perm)
 604 {
 605         struct bsg_command *bc;
 606         struct request *rq;
 607         int ret, nr_commands;
 608
 609         if (count % sizeof(struct sg_io_v4))
 610                 return -EINVAL;
 611
 612         nr_commands = count / sizeof(struct sg_io_v4);
 613         rq = NULL;
 614         bc = NULL;
 615         ret = 0;
 616         while (nr_commands) {
 617                 struct request_queue *q = bd->queue;
 618
 619                 bc = bsg_alloc_command(bd);
 620                 if (IS_ERR(bc)) {
 621                         ret = PTR_ERR(bc);
 622                         bc = NULL;
 623                         break;
 624                 }
 625
 626                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 627                         ret = -EFAULT;
 628                         break;
 629                 }
 630
 631                 /*
 632                  * get a request, fill in the blanks, and add to request queue
 633                  */
 634                 rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm);
 635                 if (IS_ERR(rq)) {
 636                         ret = PTR_ERR(rq);
 637                         rq = NULL;
 638                         break;
 639                 }
 640
 641                 bsg_add_command(bd, q, bc, rq);
 642                 bc = NULL;
 643                 rq = NULL;
 644                 nr_commands--;
 645                 buf += sizeof(struct sg_io_v4);
 646                 *bytes_written += sizeof(struct sg_io_v4);
 647         }
 648
 649         if (bc)
 650                 bsg_free_command(bc);
 651
 652         return ret;
 653 }
 654
 655 static ssize_t
 656 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 657 {
 658         struct bsg_device *bd = file->private_data;
 659         ssize_t bytes_written;
 660         int ret;
 661
 662         dprintk("%s: write %Zd bytes\n", bd->name, count);
 663
 664         bsg_set_block(bd, file);
 665
 666         bytes_written = 0;
 667         ret = __bsg_write(bd, buf, count, &bytes_written,
 668                           file->f_mode & FMODE_WRITE);
 669
 670         *ppos = bytes_written;
 671
 672         /*
 673          * return bytes written on non-fatal errors
 674          */
 675         if (!bytes_written || (bytes_written && err_block_err(ret)))
 676                 bytes_written = ret;
 677
 678         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 679         return bytes_written;
 680 }
 681
 682 static struct bsg_device *bsg_alloc_device(void)
 683 {
 684         struct bsg_device *bd;
 685
 686         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 687         if (unlikely(!bd))
 688                 return NULL;
 689
 690         spin_lock_init(&bd->lock);
 691
 692         bd->max_queue = BSG_DEFAULT_CMDS;
 693
 694         INIT_LIST_HEAD(&bd->busy_list);
 695         INIT_LIST_HEAD(&bd->done_list);
 696         INIT_HLIST_NODE(&bd->dev_list);
 697
 698         init_waitqueue_head(&bd->wq_free);
 699         init_waitqueue_head(&bd->wq_done);
 700         return bd;
 701 }
 702
 703 static void bsg_kref_release_function(struct kref *kref)
 704 {
 705         struct bsg_class_device *bcd =
 706                 container_of(kref, struct bsg_class_device, ref);
 707         struct device *parent = bcd->parent;
 708
 709         if (bcd->release)
 710                 bcd->release(bcd->parent);
 711
 712         put_device(parent);
 713 }
 714
 715 static int bsg_put_device(struct bsg_device *bd)
 716 {
 717         int ret = 0, do_free;
 718         struct request_queue *q = bd->queue;
 719
 720         mutex_lock(&bsg_mutex);
 721
 722         do_free = atomic_dec_and_test(&bd->ref_count);
 723         if (!do_free) {
 724                 mutex_unlock(&bsg_mutex);
 725                 goto out;
 726         }
 727
 728         hlist_del(&bd->dev_list);
 729         mutex_unlock(&bsg_mutex);
 730
 731         dprintk("%s: tearing down\n", bd->name);
 732
 733         /*
 734          * close can always block
 735          */
 736         set_bit(BSG_F_BLOCK, &bd->flags);
 737
 738         /*
 739          * correct error detection baddies here again. it's the responsibility
 740          * of the app to properly reap commands before close() if it wants
 741          * fool-proof error detection
 742          */
 743         ret = bsg_complete_all_commands(bd);
 744
 745         kfree(bd);
 746 out:
 747         kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
 748         if (do_free)
 749                 blk_put_queue(q);
 750         return ret;
 751 }
 752
 753 static struct bsg_device *bsg_add_device(struct inode *inode,
 754                                          struct request_queue *rq,
 755                                          struct file *file)
 756 {
 757         struct bsg_device *bd;
 758         int ret;
 759 #ifdef BSG_DEBUG
 760         unsigned char buf[32];
 761 #endif
 762         ret = blk_get_queue(rq);
 763         if (ret)
 764                 return ERR_PTR(-ENXIO);
 765
 766         bd = bsg_alloc_device();
 767         if (!bd) {
 768                 blk_put_queue(rq);
 769                 return ERR_PTR(-ENOMEM);
 770         }
 771
 772         bd->queue = rq;
 773
 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;
 838
 839         lock_kernel();
 840         bd = bsg_get_device(inode, file);
 841         unlock_kernel();
 842
 843         if (IS_ERR(bd))
 844                 return PTR_ERR(bd);
 845
 846         file->private_data = bd;
 847         return 0;
 848 }
 849
 850 static int bsg_release(struct inode *inode, struct file *file)
 851 {
 852         struct bsg_device *bd = file->private_data;
 853
 854         file->private_data = NULL;
 855         return bsg_put_device(bd);
 856 }
 857
 858 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 859 {
 860         struct bsg_device *bd = file->private_data;
 861         unsigned int mask = 0;
 862
 863         poll_wait(file, &bd->wq_done, wait);
 864         poll_wait(file, &bd->wq_free, wait);
 865
 866         spin_lock_irq(&bd->lock);
 867         if (!list_empty(&bd->done_list))
 868                 mask |= POLLIN | POLLRDNORM;
 869         if (bd->queued_cmds >= bd->max_queue)
 870                 mask |= POLLOUT;
 871         spin_unlock_irq(&bd->lock);
 872
 873         return mask;
 874 }
 875
 876 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 877 {
 878         struct bsg_device *bd = file->private_data;
 879         int __user *uarg = (int __user *) arg;
 880         int ret;
 881
 882         switch (cmd) {
 883                 /*
 884                  * our own ioctls
 885                  */
 886         case SG_GET_COMMAND_Q:
 887                 return put_user(bd->max_queue, uarg);
 888         case SG_SET_COMMAND_Q: {
 889                 int queue;
 890
 891                 if (get_user(queue, uarg))
 892                         return -EFAULT;
 893                 if (queue < 1)
 894                         return -EINVAL;
 895
 896                 spin_lock_irq(&bd->lock);
 897                 bd->max_queue = queue;
 898                 spin_unlock_irq(&bd->lock);
 899                 return 0;
 900         }
 901
 902         /*
 903          * SCSI/sg ioctls
 904          */
 905         case SG_GET_VERSION_NUM:
 906         case SCSI_IOCTL_GET_IDLUN:
 907         case SCSI_IOCTL_GET_BUS_NUMBER:
 908         case SG_SET_TIMEOUT:
 909         case SG_GET_TIMEOUT:
 910         case SG_GET_RESERVED_SIZE:
 911         case SG_SET_RESERVED_SIZE:
 912         case SG_EMULATED_HOST:
 913         case SCSI_IOCTL_SEND_COMMAND: {
 914                 void __user *uarg = (void __user *) arg;
 915                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 916         }
 917         case SG_IO: {
 918                 struct request *rq;
 919                 struct bio *bio, *bidi_bio = NULL;
 920                 struct sg_io_v4 hdr;
 921
 922                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 923                         return -EFAULT;
 924
 925                 rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE);
 926                 if (IS_ERR(rq))
 927                         return PTR_ERR(rq);
 928
 929                 bio = rq->bio;
 930                 if (rq->next_rq)
 931                         bidi_bio = rq->next_rq->bio;
 932                 blk_execute_rq(bd->queue, NULL, rq, 0);
 933                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 934
 935                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 936                         return -EFAULT;
 937
 938                 return ret;
 939         }
 940         /*
 941          * block device ioctls
 942          */
 943         default:
 944 #if 0
 945                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 946 #else
 947                 return -ENOTTY;
 948 #endif
 949         }
 950 }
 951
 952 static const struct file_operations bsg_fops = {
 953         .read           =       bsg_read,
 954         .write          =       bsg_write,
 955         .poll           =       bsg_poll,
 956         .open           =       bsg_open,
 957         .release        =       bsg_release,
 958         .unlocked_ioctl =       bsg_ioctl,
 959         .owner          =       THIS_MODULE,
 960 };
 961
 962 void bsg_unregister_queue(struct request_queue *q)
 963 {
 964         struct bsg_class_device *bcd = &q->bsg_dev;
 965
 966         if (!bcd->class_dev)
 967                 return;
 968
 969         mutex_lock(&bsg_mutex);
 970         idr_remove(&bsg_minor_idr, bcd->minor);
 971         sysfs_remove_link(&q->kobj, "bsg");
 972         device_unregister(bcd->class_dev);
 973         bcd->class_dev = NULL;
 974         kref_put(&bcd->ref, bsg_kref_release_function);
 975         mutex_unlock(&bsg_mutex);
 976 }
 977 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 978
 979 int bsg_register_queue(struct request_queue *q, struct device *parent,
 980                        const char *name, void (*release)(struct device *))
 981 {
 982         struct bsg_class_device *bcd;
 983         dev_t dev;
 984         int ret, minor;
 985         struct device *class_dev = NULL;
 986         const char *devname;
 987
 988         if (name)
 989                 devname = name;
 990         else
 991                 devname = parent->bus_id;
 992
 993         /*
 994          * we need a proper transport to send commands, not a stacked device
 995          */
 996         if (!q->request_fn)
 997                 return 0;
 998
 999         bcd = &q->bsg_dev;
1000         memset(bcd, 0, sizeof(*bcd));
1001
1002         mutex_lock(&bsg_mutex);
1003
1004         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1005         if (!ret) {
1006                 ret = -ENOMEM;
1007                 goto unlock;
1008         }
1009
1010         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1011         if (ret < 0)
1012                 goto unlock;
1013
1014         if (minor >= BSG_MAX_DEVS) {
1015                 printk(KERN_ERR "bsg: too many bsg devices\n");
1016                 ret = -EINVAL;
1017                 goto remove_idr;
1018         }
1019
1020         bcd->minor = minor;
1021         bcd->queue = q;
1022         bcd->parent = get_device(parent);
1023         bcd->release = release;
1024         kref_init(&bcd->ref);
1025         dev = MKDEV(bsg_major, bcd->minor);
1026         class_dev = device_create_drvdata(bsg_class, parent, dev, NULL,
1027                                           "%s", devname);
1028         if (IS_ERR(class_dev)) {
1029                 ret = PTR_ERR(class_dev);
1030                 goto put_dev;
1031         }
1032         bcd->class_dev = class_dev;
1033
1034         if (q->kobj.sd) {
1035                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1036                 if (ret)
1037                         goto unregister_class_dev;
1038         }
1039
1040         mutex_unlock(&bsg_mutex);
1041         return 0;
1042
1043 unregister_class_dev:
1044         device_unregister(class_dev);
1045 put_dev:
1046         put_device(parent);
1047 remove_idr:
1048         idr_remove(&bsg_minor_idr, minor);
1049 unlock:
1050         mutex_unlock(&bsg_mutex);
1051         return ret;
1052 }
1053 EXPORT_SYMBOL_GPL(bsg_register_queue);
1054
1055 static struct cdev bsg_cdev;
1056
1057 static int __init bsg_init(void)
1058 {
1059         int ret, i;
1060         dev_t devid;
1061
1062         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1063                                 sizeof(struct bsg_command), 0, 0, NULL);
1064         if (!bsg_cmd_cachep) {
1065                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1066                 return -ENOMEM;
1067         }
1068
1069         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1070                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1071
1072         bsg_class = class_create(THIS_MODULE, "bsg");
1073         if (IS_ERR(bsg_class)) {
1074                 ret = PTR_ERR(bsg_class);
1075                 goto destroy_kmemcache;
1076         }
1077
1078         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1079         if (ret)
1080                 goto destroy_bsg_class;
1081
1082         bsg_major = MAJOR(devid);
1083
1084         cdev_init(&bsg_cdev, &bsg_fops);
1085         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1086         if (ret)
1087                 goto unregister_chrdev;
1088
1089         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1090                " loaded (major %d)\n", bsg_major);
1091         return 0;
1092 unregister_chrdev:
1093         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1094 destroy_bsg_class:
1095         class_destroy(bsg_class);
1096 destroy_kmemcache:
1097         kmem_cache_destroy(bsg_cmd_cachep);
1098         return ret;
1099 }
1100
1101 MODULE_AUTHOR("Jens Axboe");
1102 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1103 MODULE_LICENSE("GPL");
1104
1105 device_initcall(bsg_init);