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[20];
  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                                 fmode_t 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         if (rq->timeout < BLK_MIN_SG_TIMEOUT)
 206                 rq->timeout = BLK_MIN_SG_TIMEOUT;
 207
 208         return 0;
 209 }
 210
 211 /*
 212  * Check if sg_io_v4 from user is allowed and valid
 213  */
 214 static int
 215 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
 216 {
 217         int ret = 0;
 218
 219         if (hdr->guard != 'Q')
 220                 return -EINVAL;
 221         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 222             hdr->din_xfer_len > (q->max_sectors << 9))
 223                 return -EIO;
 224
 225         switch (hdr->protocol) {
 226         case BSG_PROTOCOL_SCSI:
 227                 switch (hdr->subprotocol) {
 228                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 229                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 230                         break;
 231                 default:
 232                         ret = -EINVAL;
 233                 }
 234                 break;
 235         default:
 236                 ret = -EINVAL;
 237         }
 238
 239         *rw = hdr->dout_xfer_len ? WRITE : READ;
 240         return ret;
 241 }
 242
 243 /*
 244  * map sg_io_v4 to a request.
 245  */
 246 static struct request *
 247 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm)
 248 {
 249         struct request_queue *q = bd->queue;
 250         struct request *rq, *next_rq = NULL;
 251         int ret, rw;
 252         unsigned int dxfer_len;
 253         void *dxferp = NULL;
 254
 255         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 256                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 257                 hdr->din_xfer_len);
 258
 259         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 260         if (ret)
 261                 return ERR_PTR(ret);
 262
 263         /*
 264          * map scatter-gather elements seperately and string them to request
 265          */
 266         rq = blk_get_request(q, rw, GFP_KERNEL);
 267         if (!rq)
 268                 return ERR_PTR(-ENOMEM);
 269         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
 270         if (ret)
 271                 goto out;
 272
 273         if (rw == WRITE && hdr->din_xfer_len) {
 274                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 275                         ret = -EOPNOTSUPP;
 276                         goto out;
 277                 }
 278
 279                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 280                 if (!next_rq) {
 281                         ret = -ENOMEM;
 282                         goto out;
 283                 }
 284                 rq->next_rq = next_rq;
 285                 next_rq->cmd_type = rq->cmd_type;
 286
 287                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 288                 ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
 289                                        hdr->din_xfer_len, GFP_KERNEL);
 290                 if (ret)
 291                         goto out;
 292         }
 293
 294         if (hdr->dout_xfer_len) {
 295                 dxfer_len = hdr->dout_xfer_len;
 296                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 297         } else if (hdr->din_xfer_len) {
 298                 dxfer_len = hdr->din_xfer_len;
 299                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 300         } else
 301                 dxfer_len = 0;
 302
 303         if (dxfer_len) {
 304                 ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
 305                                       GFP_KERNEL);
 306                 if (ret)
 307                         goto out;
 308         }
 309         return rq;
 310 out:
 311         if (rq->cmd != rq->__cmd)
 312                 kfree(rq->cmd);
 313         blk_put_request(rq);
 314         if (next_rq) {
 315                 blk_rq_unmap_user(next_rq->bio);
 316                 blk_put_request(next_rq);
 317         }
 318         return ERR_PTR(ret);
 319 }
 320
 321 /*
 322  * async completion call-back from the block layer, when scsi/ide/whatever
 323  * calls end_that_request_last() on a request
 324  */
 325 static void bsg_rq_end_io(struct request *rq, int uptodate)
 326 {
 327         struct bsg_command *bc = rq->end_io_data;
 328         struct bsg_device *bd = bc->bd;
 329         unsigned long flags;
 330
 331         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 332                 bd->name, rq, bc, bc->bio, uptodate);
 333
 334         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 335
 336         spin_lock_irqsave(&bd->lock, flags);
 337         list_move_tail(&bc->list, &bd->done_list);
 338         bd->done_cmds++;
 339         spin_unlock_irqrestore(&bd->lock, flags);
 340
 341         wake_up(&bd->wq_done);
 342 }
 343
 344 /*
 345  * do final setup of a 'bc' and submit the matching 'rq' to the block
 346  * layer for io
 347  */
 348 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
 349                             struct bsg_command *bc, struct request *rq)
 350 {
 351         rq->sense = bc->sense;
 352         rq->sense_len = 0;
 353
 354         /*
 355          * add bc command to busy queue and submit rq for io
 356          */
 357         bc->rq = rq;
 358         bc->bio = rq->bio;
 359         if (rq->next_rq)
 360                 bc->bidi_bio = rq->next_rq->bio;
 361         bc->hdr.duration = jiffies;
 362         spin_lock_irq(&bd->lock);
 363         list_add_tail(&bc->list, &bd->busy_list);
 364         spin_unlock_irq(&bd->lock);
 365
 366         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 367
 368         rq->end_io_data = bc;
 369         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 370 }
 371
 372 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 373 {
 374         struct bsg_command *bc = NULL;
 375
 376         spin_lock_irq(&bd->lock);
 377         if (bd->done_cmds) {
 378                 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
 379                 list_del(&bc->list);
 380                 bd->done_cmds--;
 381         }
 382         spin_unlock_irq(&bd->lock);
 383
 384         return bc;
 385 }
 386
 387 /*
 388  * Get a finished command from the done list
 389  */
 390 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 391 {
 392         struct bsg_command *bc;
 393         int ret;
 394
 395         do {
 396                 bc = bsg_next_done_cmd(bd);
 397                 if (bc)
 398                         break;
 399
 400                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 401                         bc = ERR_PTR(-EAGAIN);
 402                         break;
 403                 }
 404
 405                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 406                 if (ret) {
 407                         bc = ERR_PTR(-ERESTARTSYS);
 408                         break;
 409                 }
 410         } while (1);
 411
 412         dprintk("%s: returning done %p\n", bd->name, bc);
 413
 414         return bc;
 415 }
 416
 417 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 418                                     struct bio *bio, struct bio *bidi_bio)
 419 {
 420         int ret = 0;
 421
 422         dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
 423         /*
 424          * fill in all the output members
 425          */
 426         hdr->device_status = status_byte(rq->errors);
 427         hdr->transport_status = host_byte(rq->errors);
 428         hdr->driver_status = driver_byte(rq->errors);
 429         hdr->info = 0;
 430         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 431                 hdr->info |= SG_INFO_CHECK;
 432         hdr->response_len = 0;
 433
 434         if (rq->sense_len && hdr->response) {
 435                 int len = min_t(unsigned int, hdr->max_response_len,
 436                                         rq->sense_len);
 437
 438                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 439                                    rq->sense, len);
 440                 if (!ret)
 441                         hdr->response_len = len;
 442                 else
 443                         ret = -EFAULT;
 444         }
 445
 446         if (rq->next_rq) {
 447                 hdr->dout_resid = rq->data_len;
 448                 hdr->din_resid = rq->next_rq->data_len;
 449                 blk_rq_unmap_user(bidi_bio);
 450                 blk_put_request(rq->next_rq);
 451         } else if (rq_data_dir(rq) == READ)
 452                 hdr->din_resid = rq->data_len;
 453         else
 454                 hdr->dout_resid = rq->data_len;
 455
 456         /*
 457          * If the request generated a negative error number, return it
 458          * (providing we aren't already returning an error); if it's
 459          * just a protocol response (i.e. non negative), that gets
 460          * processed above.
 461          */
 462         if (!ret && rq->errors < 0)
 463                 ret = rq->errors;
 464
 465         blk_rq_unmap_user(bio);
 466         if (rq->cmd != rq->__cmd)
 467                 kfree(rq->cmd);
 468         blk_put_request(rq);
 469
 470         return ret;
 471 }
 472
 473 static int bsg_complete_all_commands(struct bsg_device *bd)
 474 {
 475         struct bsg_command *bc;
 476         int ret, tret;
 477
 478         dprintk("%s: entered\n", bd->name);
 479
 480         /*
 481          * wait for all commands to complete
 482          */
 483         ret = 0;
 484         do {
 485                 ret = bsg_io_schedule(bd);
 486                 /*
 487                  * look for -ENODATA specifically -- we'll sometimes get
 488                  * -ERESTARTSYS when we've taken a signal, but we can't
 489                  * return until we're done freeing the queue, so ignore
 490                  * it.  The signal will get handled when we're done freeing
 491                  * the bsg_device.
 492                  */
 493         } while (ret != -ENODATA);
 494
 495         /*
 496          * discard done commands
 497          */
 498         ret = 0;
 499         do {
 500                 spin_lock_irq(&bd->lock);
 501                 if (!bd->queued_cmds) {
 502                         spin_unlock_irq(&bd->lock);
 503                         break;
 504                 }
 505                 spin_unlock_irq(&bd->lock);
 506
 507                 bc = bsg_get_done_cmd(bd);
 508                 if (IS_ERR(bc))
 509                         break;
 510
 511                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 512                                                 bc->bidi_bio);
 513                 if (!ret)
 514                         ret = tret;
 515
 516                 bsg_free_command(bc);
 517         } while (1);
 518
 519         return ret;
 520 }
 521
 522 static int
 523 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 524            const struct iovec *iov, ssize_t *bytes_read)
 525 {
 526         struct bsg_command *bc;
 527         int nr_commands, ret;
 528
 529         if (count % sizeof(struct sg_io_v4))
 530                 return -EINVAL;
 531
 532         ret = 0;
 533         nr_commands = count / sizeof(struct sg_io_v4);
 534         while (nr_commands) {
 535                 bc = bsg_get_done_cmd(bd);
 536                 if (IS_ERR(bc)) {
 537                         ret = PTR_ERR(bc);
 538                         break;
 539                 }
 540
 541                 /*
 542                  * this is the only case where we need to copy data back
 543                  * after completing the request. so do that here,
 544                  * bsg_complete_work() cannot do that for us
 545                  */
 546                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 547                                                bc->bidi_bio);
 548
 549                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 550                         ret = -EFAULT;
 551
 552                 bsg_free_command(bc);
 553
 554                 if (ret)
 555                         break;
 556
 557                 buf += sizeof(struct sg_io_v4);
 558                 *bytes_read += sizeof(struct sg_io_v4);
 559                 nr_commands--;
 560         }
 561
 562         return ret;
 563 }
 564
 565 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 566 {
 567         if (file->f_flags & O_NONBLOCK)
 568                 clear_bit(BSG_F_BLOCK, &bd->flags);
 569         else
 570                 set_bit(BSG_F_BLOCK, &bd->flags);
 571 }
 572
 573 /*
 574  * Check if the error is a "real" error that we should return.
 575  */
 576 static inline int err_block_err(int ret)
 577 {
 578         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 579                 return 1;
 580
 581         return 0;
 582 }
 583
 584 static ssize_t
 585 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 586 {
 587         struct bsg_device *bd = file->private_data;
 588         int ret;
 589         ssize_t bytes_read;
 590
 591         dprintk("%s: read %Zd bytes\n", bd->name, count);
 592
 593         bsg_set_block(bd, file);
 594
 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                        fmode_t has_write_perm)
 608 {
 609         struct bsg_command *bc;
 610         struct request *rq;
 611         int ret, nr_commands;
 612
 613         if (count % sizeof(struct sg_io_v4))
 614                 return -EINVAL;
 615
 616         nr_commands = count / sizeof(struct sg_io_v4);
 617         rq = NULL;
 618         bc = NULL;
 619         ret = 0;
 620         while (nr_commands) {
 621                 struct request_queue *q = bd->queue;
 622
 623                 bc = bsg_alloc_command(bd);
 624                 if (IS_ERR(bc)) {
 625                         ret = PTR_ERR(bc);
 626                         bc = NULL;
 627                         break;
 628                 }
 629
 630                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 631                         ret = -EFAULT;
 632                         break;
 633                 }
 634
 635                 /*
 636                  * get a request, fill in the blanks, and add to request queue
 637                  */
 638                 rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm);
 639                 if (IS_ERR(rq)) {
 640                         ret = PTR_ERR(rq);
 641                         rq = NULL;
 642                         break;
 643                 }
 644
 645                 bsg_add_command(bd, q, bc, rq);
 646                 bc = NULL;
 647                 rq = NULL;
 648                 nr_commands--;
 649                 buf += sizeof(struct sg_io_v4);
 650                 *bytes_written += sizeof(struct sg_io_v4);
 651         }
 652
 653         if (bc)
 654                 bsg_free_command(bc);
 655
 656         return ret;
 657 }
 658
 659 static ssize_t
 660 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 661 {
 662         struct bsg_device *bd = file->private_data;
 663         ssize_t bytes_written;
 664         int ret;
 665
 666         dprintk("%s: write %Zd bytes\n", bd->name, count);
 667
 668         bsg_set_block(bd, file);
 669
 670         bytes_written = 0;
 671         ret = __bsg_write(bd, buf, count, &bytes_written,
 672                           file->f_mode & FMODE_WRITE);
 673
 674         *ppos = bytes_written;
 675
 676         /*
 677          * return bytes written on non-fatal errors
 678          */
 679         if (!bytes_written || (bytes_written && err_block_err(ret)))
 680                 bytes_written = ret;
 681
 682         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 683         return bytes_written;
 684 }
 685
 686 static struct bsg_device *bsg_alloc_device(void)
 687 {
 688         struct bsg_device *bd;
 689
 690         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 691         if (unlikely(!bd))
 692                 return NULL;
 693
 694         spin_lock_init(&bd->lock);
 695
 696         bd->max_queue = BSG_DEFAULT_CMDS;
 697
 698         INIT_LIST_HEAD(&bd->busy_list);
 699         INIT_LIST_HEAD(&bd->done_list);
 700         INIT_HLIST_NODE(&bd->dev_list);
 701
 702         init_waitqueue_head(&bd->wq_free);
 703         init_waitqueue_head(&bd->wq_done);
 704         return bd;
 705 }
 706
 707 static void bsg_kref_release_function(struct kref *kref)
 708 {
 709         struct bsg_class_device *bcd =
 710                 container_of(kref, struct bsg_class_device, ref);
 711         struct device *parent = bcd->parent;
 712
 713         if (bcd->release)
 714                 bcd->release(bcd->parent);
 715
 716         put_device(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                 mutex_unlock(&bsg_mutex);
 729                 goto out;
 730         }
 731
 732         hlist_del(&bd->dev_list);
 733         mutex_unlock(&bsg_mutex);
 734
 735         dprintk("%s: tearing down\n", bd->name);
 736
 737         /*
 738          * close can always block
 739          */
 740         set_bit(BSG_F_BLOCK, &bd->flags);
 741
 742         /*
 743          * correct error detection baddies here again. it's the responsibility
 744          * of the app to properly reap commands before close() if it wants
 745          * fool-proof error detection
 746          */
 747         ret = bsg_complete_all_commands(bd);
 748
 749         kfree(bd);
 750 out:
 751         kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
 752         if (do_free)
 753                 blk_put_queue(q);
 754         return ret;
 755 }
 756
 757 static struct bsg_device *bsg_add_device(struct inode *inode,
 758                                          struct request_queue *rq,
 759                                          struct file *file)
 760 {
 761         struct bsg_device *bd;
 762         int ret;
 763 #ifdef BSG_DEBUG
 764         unsigned char buf[32];
 765 #endif
 766         ret = blk_get_queue(rq);
 767         if (ret)
 768                 return ERR_PTR(-ENXIO);
 769
 770         bd = bsg_alloc_device();
 771         if (!bd) {
 772                 blk_put_queue(rq);
 773                 return ERR_PTR(-ENOMEM);
 774         }
 775
 776         bd->queue = rq;
 777
 778         bsg_set_block(bd, file);
 779
 780         atomic_set(&bd->ref_count, 1);
 781         mutex_lock(&bsg_mutex);
 782         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
 783
 784         strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
 785         dprintk("bound to <%s>, max queue %d\n",
 786                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 787
 788         mutex_unlock(&bsg_mutex);
 789         return bd;
 790 }
 791
 792 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
 793 {
 794         struct bsg_device *bd;
 795         struct hlist_node *entry;
 796
 797         mutex_lock(&bsg_mutex);
 798
 799         hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
 800                 if (bd->queue == q) {
 801                         atomic_inc(&bd->ref_count);
 802                         goto found;
 803                 }
 804         }
 805         bd = NULL;
 806 found:
 807         mutex_unlock(&bsg_mutex);
 808         return bd;
 809 }
 810
 811 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 812 {
 813         struct bsg_device *bd;
 814         struct bsg_class_device *bcd;
 815
 816         /*
 817          * find the class device
 818          */
 819         mutex_lock(&bsg_mutex);
 820         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 821         if (bcd)
 822                 kref_get(&bcd->ref);
 823         mutex_unlock(&bsg_mutex);
 824
 825         if (!bcd)
 826                 return ERR_PTR(-ENODEV);
 827
 828         bd = __bsg_get_device(iminor(inode), bcd->queue);
 829         if (bd)
 830                 return bd;
 831
 832         bd = bsg_add_device(inode, bcd->queue, file);
 833         if (IS_ERR(bd))
 834                 kref_put(&bcd->ref, bsg_kref_release_function);
 835
 836         return bd;
 837 }
 838
 839 static int bsg_open(struct inode *inode, struct file *file)
 840 {
 841         struct bsg_device *bd;
 842
 843         lock_kernel();
 844         bd = bsg_get_device(inode, file);
 845         unlock_kernel();
 846
 847         if (IS_ERR(bd))
 848                 return PTR_ERR(bd);
 849
 850         file->private_data = bd;
 851         return 0;
 852 }
 853
 854 static int bsg_release(struct inode *inode, struct file *file)
 855 {
 856         struct bsg_device *bd = file->private_data;
 857
 858         file->private_data = NULL;
 859         return bsg_put_device(bd);
 860 }
 861
 862 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 863 {
 864         struct bsg_device *bd = file->private_data;
 865         unsigned int mask = 0;
 866
 867         poll_wait(file, &bd->wq_done, wait);
 868         poll_wait(file, &bd->wq_free, wait);
 869
 870         spin_lock_irq(&bd->lock);
 871         if (!list_empty(&bd->done_list))
 872                 mask |= POLLIN | POLLRDNORM;
 873         if (bd->queued_cmds >= bd->max_queue)
 874                 mask |= POLLOUT;
 875         spin_unlock_irq(&bd->lock);
 876
 877         return mask;
 878 }
 879
 880 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 881 {
 882         struct bsg_device *bd = file->private_data;
 883         int __user *uarg = (int __user *) arg;
 884         int ret;
 885
 886         switch (cmd) {
 887                 /*
 888                  * our own ioctls
 889                  */
 890         case SG_GET_COMMAND_Q:
 891                 return put_user(bd->max_queue, uarg);
 892         case SG_SET_COMMAND_Q: {
 893                 int queue;
 894
 895                 if (get_user(queue, uarg))
 896                         return -EFAULT;
 897                 if (queue < 1)
 898                         return -EINVAL;
 899
 900                 spin_lock_irq(&bd->lock);
 901                 bd->max_queue = queue;
 902                 spin_unlock_irq(&bd->lock);
 903                 return 0;
 904         }
 905
 906         /*
 907          * SCSI/sg ioctls
 908          */
 909         case SG_GET_VERSION_NUM:
 910         case SCSI_IOCTL_GET_IDLUN:
 911         case SCSI_IOCTL_GET_BUS_NUMBER:
 912         case SG_SET_TIMEOUT:
 913         case SG_GET_TIMEOUT:
 914         case SG_GET_RESERVED_SIZE:
 915         case SG_SET_RESERVED_SIZE:
 916         case SG_EMULATED_HOST:
 917         case SCSI_IOCTL_SEND_COMMAND: {
 918                 void __user *uarg = (void __user *) arg;
 919                 return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
 920         }
 921         case SG_IO: {
 922                 struct request *rq;
 923                 struct bio *bio, *bidi_bio = NULL;
 924                 struct sg_io_v4 hdr;
 925
 926                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 927                         return -EFAULT;
 928
 929                 rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE);
 930                 if (IS_ERR(rq))
 931                         return PTR_ERR(rq);
 932
 933                 bio = rq->bio;
 934                 if (rq->next_rq)
 935                         bidi_bio = rq->next_rq->bio;
 936                 blk_execute_rq(bd->queue, NULL, rq, 0);
 937                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 938
 939                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 940                         return -EFAULT;
 941
 942                 return ret;
 943         }
 944         /*
 945          * block device ioctls
 946          */
 947         default:
 948 #if 0
 949                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 950 #else
 951                 return -ENOTTY;
 952 #endif
 953         }
 954 }
 955
 956 static const struct file_operations bsg_fops = {
 957         .read           =       bsg_read,
 958         .write          =       bsg_write,
 959         .poll           =       bsg_poll,
 960         .open           =       bsg_open,
 961         .release        =       bsg_release,
 962         .unlocked_ioctl =       bsg_ioctl,
 963         .owner          =       THIS_MODULE,
 964 };
 965
 966 void bsg_unregister_queue(struct request_queue *q)
 967 {
 968         struct bsg_class_device *bcd = &q->bsg_dev;
 969
 970         if (!bcd->class_dev)
 971                 return;
 972
 973         mutex_lock(&bsg_mutex);
 974         idr_remove(&bsg_minor_idr, bcd->minor);
 975         sysfs_remove_link(&q->kobj, "bsg");
 976         device_unregister(bcd->class_dev);
 977         bcd->class_dev = NULL;
 978         kref_put(&bcd->ref, bsg_kref_release_function);
 979         mutex_unlock(&bsg_mutex);
 980 }
 981 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 982
 983 int bsg_register_queue(struct request_queue *q, struct device *parent,
 984                        const char *name, void (*release)(struct device *))
 985 {
 986         struct bsg_class_device *bcd;
 987         dev_t dev;
 988         int ret, minor;
 989         struct device *class_dev = NULL;
 990         const char *devname;
 991
 992         if (name)
 993                 devname = name;
 994         else
 995                 devname = dev_name(parent);
 996
 997         /*
 998          * we need a proper transport to send commands, not a stacked device
 999          */
1000         if (!q->request_fn)
1001                 return 0;
1002
1003         bcd = &q->bsg_dev;
1004         memset(bcd, 0, sizeof(*bcd));
1005
1006         mutex_lock(&bsg_mutex);
1007
1008         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1009         if (!ret) {
1010                 ret = -ENOMEM;
1011                 goto unlock;
1012         }
1013
1014         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1015         if (ret < 0)
1016                 goto unlock;
1017
1018         if (minor >= BSG_MAX_DEVS) {
1019                 printk(KERN_ERR "bsg: too many bsg devices\n");
1020                 ret = -EINVAL;
1021                 goto remove_idr;
1022         }
1023
1024         bcd->minor = minor;
1025         bcd->queue = q;
1026         bcd->parent = get_device(parent);
1027         bcd->release = release;
1028         kref_init(&bcd->ref);
1029         dev = MKDEV(bsg_major, bcd->minor);
1030         class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1031         if (IS_ERR(class_dev)) {
1032                 ret = PTR_ERR(class_dev);
1033                 goto put_dev;
1034         }
1035         bcd->class_dev = class_dev;
1036
1037         if (q->kobj.sd) {
1038                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1039                 if (ret)
1040                         goto unregister_class_dev;
1041         }
1042
1043         mutex_unlock(&bsg_mutex);
1044         return 0;
1045
1046 unregister_class_dev:
1047         device_unregister(class_dev);
1048 put_dev:
1049         put_device(parent);
1050 remove_idr:
1051         idr_remove(&bsg_minor_idr, minor);
1052 unlock:
1053         mutex_unlock(&bsg_mutex);
1054         return ret;
1055 }
1056 EXPORT_SYMBOL_GPL(bsg_register_queue);
1057
1058 static struct cdev bsg_cdev;
1059
1060 static int __init bsg_init(void)
1061 {
1062         int ret, i;
1063         dev_t devid;
1064
1065         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1066                                 sizeof(struct bsg_command), 0, 0, NULL);
1067         if (!bsg_cmd_cachep) {
1068                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1069                 return -ENOMEM;
1070         }
1071
1072         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1073                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1074
1075         bsg_class = class_create(THIS_MODULE, "bsg");
1076         if (IS_ERR(bsg_class)) {
1077                 ret = PTR_ERR(bsg_class);
1078                 goto destroy_kmemcache;
1079         }
1080
1081         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1082         if (ret)
1083                 goto destroy_bsg_class;
1084
1085         bsg_major = MAJOR(devid);
1086
1087         cdev_init(&bsg_cdev, &bsg_fops);
1088         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1089         if (ret)
1090                 goto unregister_chrdev;
1091
1092         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1093                " loaded (major %d)\n", bsg_major);
1094         return 0;
1095 unregister_chrdev:
1096         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1097 destroy_bsg_class:
1098         class_destroy(bsg_class);
1099 destroy_kmemcache:
1100         kmem_cache_destroy(bsg_cmd_cachep);
1101         return ret;
1102 }
1103
1104 MODULE_AUTHOR("Jens Axboe");
1105 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1106 MODULE_LICENSE("GPL");
1107
1108 device_initcall(bsg_init);