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/jiffies.h>
  19 #include <linux/percpu.h>
  20 #include <linux/uio.h>
  21 #include <linux/idr.h>
  22 #include <linux/bsg.h>
  23 #include <linux/slab.h>
  24
  25 #include <scsi/scsi.h>
  26 #include <scsi/scsi_ioctl.h>
  27 #include <scsi/scsi_cmnd.h>
  28 #include <scsi/scsi_device.h>
  29 #include <scsi/scsi_driver.h>
  30 #include <scsi/sg.h>
  31
  32 #define BSG_DESCRIPTION "Block layer SCSI generic (bsg) driver"
  33 #define BSG_VERSION     "0.4"
  34
  35 struct bsg_device {
  36         struct request_queue *queue;
  37         spinlock_t lock;
  38         struct list_head busy_list;
  39         struct list_head done_list;
  40         struct hlist_node dev_list;
  41         atomic_t ref_count;
  42         int queued_cmds;
  43         int done_cmds;
  44         wait_queue_head_t wq_done;
  45         wait_queue_head_t wq_free;
  46         char name[20];
  47         int max_queue;
  48         unsigned long flags;
  49 };
  50
  51 enum {
  52         BSG_F_BLOCK             = 1,
  53 };
  54
  55 #define BSG_DEFAULT_CMDS        64
  56 #define BSG_MAX_DEVS            32768
  57
  58 #undef BSG_DEBUG
  59
  60 #ifdef BSG_DEBUG
  61 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __func__, ##args)
  62 #else
  63 #define dprintk(fmt, args...)
  64 #endif
  65
  66 static DEFINE_MUTEX(bsg_mutex);
  67 static DEFINE_IDR(bsg_minor_idr);
  68
  69 #define BSG_LIST_ARRAY_SIZE     8
  70 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
  71
  72 static struct class *bsg_class;
  73 static int bsg_major;
  74
  75 static struct kmem_cache *bsg_cmd_cachep;
  76
  77 /*
  78  * our internal command type
  79  */
  80 struct bsg_command {
  81         struct bsg_device *bd;
  82         struct list_head list;
  83         struct request *rq;
  84         struct bio *bio;
  85         struct bio *bidi_bio;
  86         int err;
  87         struct sg_io_v4 hdr;
  88         char sense[SCSI_SENSE_BUFFERSIZE];
  89 };
  90
  91 static void bsg_free_command(struct bsg_command *bc)
  92 {
  93         struct bsg_device *bd = bc->bd;
  94         unsigned long flags;
  95
  96         kmem_cache_free(bsg_cmd_cachep, bc);
  97
  98         spin_lock_irqsave(&bd->lock, flags);
  99         bd->queued_cmds--;
 100         spin_unlock_irqrestore(&bd->lock, flags);
 101
 102         wake_up(&bd->wq_free);
 103 }
 104
 105 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
 106 {
 107         struct bsg_command *bc = ERR_PTR(-EINVAL);
 108
 109         spin_lock_irq(&bd->lock);
 110
 111         if (bd->queued_cmds >= bd->max_queue)
 112                 goto out;
 113
 114         bd->queued_cmds++;
 115         spin_unlock_irq(&bd->lock);
 116
 117         bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
 118         if (unlikely(!bc)) {
 119                 spin_lock_irq(&bd->lock);
 120                 bd->queued_cmds--;
 121                 bc = ERR_PTR(-ENOMEM);
 122                 goto out;
 123         }
 124
 125         bc->bd = bd;
 126         INIT_LIST_HEAD(&bc->list);
 127         dprintk("%s: returning free cmd %p\n", bd->name, bc);
 128         return bc;
 129 out:
 130         spin_unlock_irq(&bd->lock);
 131         return bc;
 132 }
 133
 134 static inline struct hlist_head *bsg_dev_idx_hash(int index)
 135 {
 136         return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
 137 }
 138
 139 static int bsg_io_schedule(struct bsg_device *bd)
 140 {
 141         DEFINE_WAIT(wait);
 142         int ret = 0;
 143
 144         spin_lock_irq(&bd->lock);
 145
 146         BUG_ON(bd->done_cmds > bd->queued_cmds);
 147
 148         /*
 149          * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
 150          * work to do", even though we return -ENOSPC after this same test
 151          * during bsg_write() -- there, it means our buffer can't have more
 152          * bsg_commands added to it, thus has no space left.
 153          */
 154         if (bd->done_cmds == bd->queued_cmds) {
 155                 ret = -ENODATA;
 156                 goto unlock;
 157         }
 158
 159         if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 160                 ret = -EAGAIN;
 161                 goto unlock;
 162         }
 163
 164         prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
 165         spin_unlock_irq(&bd->lock);
 166         io_schedule();
 167         finish_wait(&bd->wq_done, &wait);
 168
 169         return ret;
 170 unlock:
 171         spin_unlock_irq(&bd->lock);
 172         return ret;
 173 }
 174
 175 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
 176                                 struct sg_io_v4 *hdr, struct bsg_device *bd,
 177                                 fmode_t has_write_perm)
 178 {
 179         if (hdr->request_len > BLK_MAX_CDB) {
 180                 rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
 181                 if (!rq->cmd)
 182                         return -ENOMEM;
 183         }
 184
 185         if (copy_from_user(rq->cmd, (void __user *)(unsigned long)hdr->request,
 186                            hdr->request_len))
 187                 return -EFAULT;
 188
 189         if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
 190                 if (blk_verify_command(rq->cmd, has_write_perm))
 191                         return -EPERM;
 192         } else if (!capable(CAP_SYS_RAWIO))
 193                 return -EPERM;
 194
 195         /*
 196          * fill in request structure
 197          */
 198         rq->cmd_len = hdr->request_len;
 199
 200         rq->timeout = msecs_to_jiffies(hdr->timeout);
 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
 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, fmode_t has_write_perm,
 245             u8 *sense)
 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 __user *dxferp = NULL;
 252         struct bsg_class_device *bcd = &q->bsg_dev;
 253
 254         /* if the LLD has been removed then the bsg_unregister_queue will
 255          * eventually be called and the class_dev was freed, so we can no
 256          * longer use this request_queue. Return no such address.
 257          */
 258         if (!bcd->class_dev)
 259                 return ERR_PTR(-ENXIO);
 260
 261         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 262                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 263                 hdr->din_xfer_len);
 264
 265         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 266         if (ret)
 267                 return ERR_PTR(ret);
 268
 269         /*
 270          * map scatter-gather elements separately and string them to request
 271          */
 272         rq = blk_get_request(q, rw, GFP_KERNEL);
 273         if (!rq)
 274                 return ERR_PTR(-ENOMEM);
 275         blk_rq_set_block_pc(rq);
 276
 277         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
 278         if (ret)
 279                 goto out;
 280
 281         if (rw == WRITE && hdr->din_xfer_len) {
 282                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 283                         ret = -EOPNOTSUPP;
 284                         goto out;
 285                 }
 286
 287                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 288                 if (!next_rq) {
 289                         ret = -ENOMEM;
 290                         goto out;
 291                 }
 292                 rq->next_rq = next_rq;
 293                 next_rq->cmd_type = rq->cmd_type;
 294
 295                 dxferp = (void __user *)(unsigned long)hdr->din_xferp;
 296                 ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
 297                                        hdr->din_xfer_len, GFP_KERNEL);
 298                 if (ret)
 299                         goto out;
 300         }
 301
 302         if (hdr->dout_xfer_len) {
 303                 dxfer_len = hdr->dout_xfer_len;
 304                 dxferp = (void __user *)(unsigned long)hdr->dout_xferp;
 305         } else if (hdr->din_xfer_len) {
 306                 dxfer_len = hdr->din_xfer_len;
 307                 dxferp = (void __user *)(unsigned long)hdr->din_xferp;
 308         } else
 309                 dxfer_len = 0;
 310
 311         if (dxfer_len) {
 312                 ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
 313                                       GFP_KERNEL);
 314                 if (ret)
 315                         goto out;
 316         }
 317
 318         rq->sense = sense;
 319         rq->sense_len = 0;
 320
 321         return rq;
 322 out:
 323         if (rq->cmd != rq->__cmd)
 324                 kfree(rq->cmd);
 325         blk_put_request(rq);
 326         if (next_rq) {
 327                 blk_rq_unmap_user(next_rq->bio);
 328                 blk_put_request(next_rq);
 329         }
 330         return ERR_PTR(ret);
 331 }
 332
 333 /*
 334  * async completion call-back from the block layer, when scsi/ide/whatever
 335  * calls end_that_request_last() on a request
 336  */
 337 static void bsg_rq_end_io(struct request *rq, int uptodate)
 338 {
 339         struct bsg_command *bc = rq->end_io_data;
 340         struct bsg_device *bd = bc->bd;
 341         unsigned long flags;
 342
 343         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 344                 bd->name, rq, bc, bc->bio, uptodate);
 345
 346         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 347
 348         spin_lock_irqsave(&bd->lock, flags);
 349         list_move_tail(&bc->list, &bd->done_list);
 350         bd->done_cmds++;
 351         spin_unlock_irqrestore(&bd->lock, flags);
 352
 353         wake_up(&bd->wq_done);
 354 }
 355
 356 /*
 357  * do final setup of a 'bc' and submit the matching 'rq' to the block
 358  * layer for io
 359  */
 360 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
 361                             struct bsg_command *bc, struct request *rq)
 362 {
 363         int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
 364
 365         /*
 366          * add bc command to busy queue and submit rq for io
 367          */
 368         bc->rq = rq;
 369         bc->bio = rq->bio;
 370         if (rq->next_rq)
 371                 bc->bidi_bio = rq->next_rq->bio;
 372         bc->hdr.duration = jiffies;
 373         spin_lock_irq(&bd->lock);
 374         list_add_tail(&bc->list, &bd->busy_list);
 375         spin_unlock_irq(&bd->lock);
 376
 377         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 378
 379         rq->end_io_data = bc;
 380         blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
 381 }
 382
 383 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 384 {
 385         struct bsg_command *bc = NULL;
 386
 387         spin_lock_irq(&bd->lock);
 388         if (bd->done_cmds) {
 389                 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
 390                 list_del(&bc->list);
 391                 bd->done_cmds--;
 392         }
 393         spin_unlock_irq(&bd->lock);
 394
 395         return bc;
 396 }
 397
 398 /*
 399  * Get a finished command from the done list
 400  */
 401 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 402 {
 403         struct bsg_command *bc;
 404         int ret;
 405
 406         do {
 407                 bc = bsg_next_done_cmd(bd);
 408                 if (bc)
 409                         break;
 410
 411                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 412                         bc = ERR_PTR(-EAGAIN);
 413                         break;
 414                 }
 415
 416                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 417                 if (ret) {
 418                         bc = ERR_PTR(-ERESTARTSYS);
 419                         break;
 420                 }
 421         } while (1);
 422
 423         dprintk("%s: returning done %p\n", bd->name, bc);
 424
 425         return bc;
 426 }
 427
 428 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 429                                     struct bio *bio, struct bio *bidi_bio)
 430 {
 431         int ret = 0;
 432
 433         dprintk("rq %p bio %p 0x%x\n", rq, bio, rq->errors);
 434         /*
 435          * fill in all the output members
 436          */
 437         hdr->device_status = rq->errors & 0xff;
 438         hdr->transport_status = host_byte(rq->errors);
 439         hdr->driver_status = driver_byte(rq->errors);
 440         hdr->info = 0;
 441         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 442                 hdr->info |= SG_INFO_CHECK;
 443         hdr->response_len = 0;
 444
 445         if (rq->sense_len && hdr->response) {
 446                 int len = min_t(unsigned int, hdr->max_response_len,
 447                                         rq->sense_len);
 448
 449                 ret = copy_to_user((void __user *)(unsigned long)hdr->response,
 450                                    rq->sense, len);
 451                 if (!ret)
 452                         hdr->response_len = len;
 453                 else
 454                         ret = -EFAULT;
 455         }
 456
 457         if (rq->next_rq) {
 458                 hdr->dout_resid = rq->resid_len;
 459                 hdr->din_resid = rq->next_rq->resid_len;
 460                 blk_rq_unmap_user(bidi_bio);
 461                 blk_put_request(rq->next_rq);
 462         } else if (rq_data_dir(rq) == READ)
 463                 hdr->din_resid = rq->resid_len;
 464         else
 465                 hdr->dout_resid = rq->resid_len;
 466
 467         /*
 468          * If the request generated a negative error number, return it
 469          * (providing we aren't already returning an error); if it's
 470          * just a protocol response (i.e. non negative), that gets
 471          * processed above.
 472          */
 473         if (!ret && rq->errors < 0)
 474                 ret = rq->errors;
 475
 476         blk_rq_unmap_user(bio);
 477         if (rq->cmd != rq->__cmd)
 478                 kfree(rq->cmd);
 479         blk_put_request(rq);
 480
 481         return ret;
 482 }
 483
 484 static int bsg_complete_all_commands(struct bsg_device *bd)
 485 {
 486         struct bsg_command *bc;
 487         int ret, tret;
 488
 489         dprintk("%s: entered\n", bd->name);
 490
 491         /*
 492          * wait for all commands to complete
 493          */
 494         ret = 0;
 495         do {
 496                 ret = bsg_io_schedule(bd);
 497                 /*
 498                  * look for -ENODATA specifically -- we'll sometimes get
 499                  * -ERESTARTSYS when we've taken a signal, but we can't
 500                  * return until we're done freeing the queue, so ignore
 501                  * it.  The signal will get handled when we're done freeing
 502                  * the bsg_device.
 503                  */
 504         } while (ret != -ENODATA);
 505
 506         /*
 507          * discard done commands
 508          */
 509         ret = 0;
 510         do {
 511                 spin_lock_irq(&bd->lock);
 512                 if (!bd->queued_cmds) {
 513                         spin_unlock_irq(&bd->lock);
 514                         break;
 515                 }
 516                 spin_unlock_irq(&bd->lock);
 517
 518                 bc = bsg_get_done_cmd(bd);
 519                 if (IS_ERR(bc))
 520                         break;
 521
 522                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 523                                                 bc->bidi_bio);
 524                 if (!ret)
 525                         ret = tret;
 526
 527                 bsg_free_command(bc);
 528         } while (1);
 529
 530         return ret;
 531 }
 532
 533 static int
 534 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 535            const struct iovec *iov, ssize_t *bytes_read)
 536 {
 537         struct bsg_command *bc;
 538         int nr_commands, ret;
 539
 540         if (count % sizeof(struct sg_io_v4))
 541                 return -EINVAL;
 542
 543         ret = 0;
 544         nr_commands = count / sizeof(struct sg_io_v4);
 545         while (nr_commands) {
 546                 bc = bsg_get_done_cmd(bd);
 547                 if (IS_ERR(bc)) {
 548                         ret = PTR_ERR(bc);
 549                         break;
 550                 }
 551
 552                 /*
 553                  * this is the only case where we need to copy data back
 554                  * after completing the request. so do that here,
 555                  * bsg_complete_work() cannot do that for us
 556                  */
 557                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 558                                                bc->bidi_bio);
 559
 560                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 561                         ret = -EFAULT;
 562
 563                 bsg_free_command(bc);
 564
 565                 if (ret)
 566                         break;
 567
 568                 buf += sizeof(struct sg_io_v4);
 569                 *bytes_read += sizeof(struct sg_io_v4);
 570                 nr_commands--;
 571         }
 572
 573         return ret;
 574 }
 575
 576 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 577 {
 578         if (file->f_flags & O_NONBLOCK)
 579                 clear_bit(BSG_F_BLOCK, &bd->flags);
 580         else
 581                 set_bit(BSG_F_BLOCK, &bd->flags);
 582 }
 583
 584 /*
 585  * Check if the error is a "real" error that we should return.
 586  */
 587 static inline int err_block_err(int ret)
 588 {
 589         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 590                 return 1;
 591
 592         return 0;
 593 }
 594
 595 static ssize_t
 596 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 597 {
 598         struct bsg_device *bd = file->private_data;
 599         int ret;
 600         ssize_t bytes_read;
 601
 602         dprintk("%s: read %Zd bytes\n", bd->name, count);
 603
 604         bsg_set_block(bd, file);
 605
 606         bytes_read = 0;
 607         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 608         *ppos = bytes_read;
 609
 610         if (!bytes_read || err_block_err(ret))
 611                 bytes_read = ret;
 612
 613         return bytes_read;
 614 }
 615
 616 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 617                        size_t count, ssize_t *bytes_written,
 618                        fmode_t has_write_perm)
 619 {
 620         struct bsg_command *bc;
 621         struct request *rq;
 622         int ret, nr_commands;
 623
 624         if (count % sizeof(struct sg_io_v4))
 625                 return -EINVAL;
 626
 627         nr_commands = count / sizeof(struct sg_io_v4);
 628         rq = NULL;
 629         bc = NULL;
 630         ret = 0;
 631         while (nr_commands) {
 632                 struct request_queue *q = bd->queue;
 633
 634                 bc = bsg_alloc_command(bd);
 635                 if (IS_ERR(bc)) {
 636                         ret = PTR_ERR(bc);
 637                         bc = NULL;
 638                         break;
 639                 }
 640
 641                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 642                         ret = -EFAULT;
 643                         break;
 644                 }
 645
 646                 /*
 647                  * get a request, fill in the blanks, and add to request queue
 648                  */
 649                 rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm, bc->sense);
 650                 if (IS_ERR(rq)) {
 651                         ret = PTR_ERR(rq);
 652                         rq = NULL;
 653                         break;
 654                 }
 655
 656                 bsg_add_command(bd, q, bc, rq);
 657                 bc = NULL;
 658                 rq = NULL;
 659                 nr_commands--;
 660                 buf += sizeof(struct sg_io_v4);
 661                 *bytes_written += sizeof(struct sg_io_v4);
 662         }
 663
 664         if (bc)
 665                 bsg_free_command(bc);
 666
 667         return ret;
 668 }
 669
 670 static ssize_t
 671 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 672 {
 673         struct bsg_device *bd = file->private_data;
 674         ssize_t bytes_written;
 675         int ret;
 676
 677         dprintk("%s: write %Zd bytes\n", bd->name, count);
 678
 679         bsg_set_block(bd, file);
 680
 681         bytes_written = 0;
 682         ret = __bsg_write(bd, buf, count, &bytes_written,
 683                           file->f_mode & FMODE_WRITE);
 684
 685         *ppos = bytes_written;
 686
 687         /*
 688          * return bytes written on non-fatal errors
 689          */
 690         if (!bytes_written || err_block_err(ret))
 691                 bytes_written = ret;
 692
 693         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 694         return bytes_written;
 695 }
 696
 697 static struct bsg_device *bsg_alloc_device(void)
 698 {
 699         struct bsg_device *bd;
 700
 701         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 702         if (unlikely(!bd))
 703                 return NULL;
 704
 705         spin_lock_init(&bd->lock);
 706
 707         bd->max_queue = BSG_DEFAULT_CMDS;
 708
 709         INIT_LIST_HEAD(&bd->busy_list);
 710         INIT_LIST_HEAD(&bd->done_list);
 711         INIT_HLIST_NODE(&bd->dev_list);
 712
 713         init_waitqueue_head(&bd->wq_free);
 714         init_waitqueue_head(&bd->wq_done);
 715         return bd;
 716 }
 717
 718 static void bsg_kref_release_function(struct kref *kref)
 719 {
 720         struct bsg_class_device *bcd =
 721                 container_of(kref, struct bsg_class_device, ref);
 722         struct device *parent = bcd->parent;
 723
 724         if (bcd->release)
 725                 bcd->release(bcd->parent);
 726
 727         put_device(parent);
 728 }
 729
 730 static int bsg_put_device(struct bsg_device *bd)
 731 {
 732         int ret = 0, do_free;
 733         struct request_queue *q = bd->queue;
 734
 735         mutex_lock(&bsg_mutex);
 736
 737         do_free = atomic_dec_and_test(&bd->ref_count);
 738         if (!do_free) {
 739                 mutex_unlock(&bsg_mutex);
 740                 goto out;
 741         }
 742
 743         hlist_del(&bd->dev_list);
 744         mutex_unlock(&bsg_mutex);
 745
 746         dprintk("%s: tearing down\n", bd->name);
 747
 748         /*
 749          * close can always block
 750          */
 751         set_bit(BSG_F_BLOCK, &bd->flags);
 752
 753         /*
 754          * correct error detection baddies here again. it's the responsibility
 755          * of the app to properly reap commands before close() if it wants
 756          * fool-proof error detection
 757          */
 758         ret = bsg_complete_all_commands(bd);
 759
 760         kfree(bd);
 761 out:
 762         kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
 763         if (do_free)
 764                 blk_put_queue(q);
 765         return ret;
 766 }
 767
 768 static struct bsg_device *bsg_add_device(struct inode *inode,
 769                                          struct request_queue *rq,
 770                                          struct file *file)
 771 {
 772         struct bsg_device *bd;
 773 #ifdef BSG_DEBUG
 774         unsigned char buf[32];
 775 #endif
 776         if (!blk_get_queue(rq))
 777                 return ERR_PTR(-ENXIO);
 778
 779         bd = bsg_alloc_device();
 780         if (!bd) {
 781                 blk_put_queue(rq);
 782                 return ERR_PTR(-ENOMEM);
 783         }
 784
 785         bd->queue = rq;
 786
 787         bsg_set_block(bd, file);
 788
 789         atomic_set(&bd->ref_count, 1);
 790         mutex_lock(&bsg_mutex);
 791         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
 792
 793         strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
 794         dprintk("bound to <%s>, max queue %d\n",
 795                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 796
 797         mutex_unlock(&bsg_mutex);
 798         return bd;
 799 }
 800
 801 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
 802 {
 803         struct bsg_device *bd;
 804
 805         mutex_lock(&bsg_mutex);
 806
 807         hlist_for_each_entry(bd, bsg_dev_idx_hash(minor), dev_list) {
 808                 if (bd->queue == q) {
 809                         atomic_inc(&bd->ref_count);
 810                         goto found;
 811                 }
 812         }
 813         bd = NULL;
 814 found:
 815         mutex_unlock(&bsg_mutex);
 816         return bd;
 817 }
 818
 819 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 820 {
 821         struct bsg_device *bd;
 822         struct bsg_class_device *bcd;
 823
 824         /*
 825          * find the class device
 826          */
 827         mutex_lock(&bsg_mutex);
 828         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 829         if (bcd)
 830                 kref_get(&bcd->ref);
 831         mutex_unlock(&bsg_mutex);
 832
 833         if (!bcd)
 834                 return ERR_PTR(-ENODEV);
 835
 836         bd = __bsg_get_device(iminor(inode), bcd->queue);
 837         if (bd)
 838                 return bd;
 839
 840         bd = bsg_add_device(inode, bcd->queue, file);
 841         if (IS_ERR(bd))
 842                 kref_put(&bcd->ref, bsg_kref_release_function);
 843
 844         return bd;
 845 }
 846
 847 static int bsg_open(struct inode *inode, struct file *file)
 848 {
 849         struct bsg_device *bd;
 850
 851         bd = bsg_get_device(inode, file);
 852
 853         if (IS_ERR(bd))
 854                 return PTR_ERR(bd);
 855
 856         file->private_data = bd;
 857         return 0;
 858 }
 859
 860 static int bsg_release(struct inode *inode, struct file *file)
 861 {
 862         struct bsg_device *bd = file->private_data;
 863
 864         file->private_data = NULL;
 865         return bsg_put_device(bd);
 866 }
 867
 868 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 869 {
 870         struct bsg_device *bd = file->private_data;
 871         unsigned int mask = 0;
 872
 873         poll_wait(file, &bd->wq_done, wait);
 874         poll_wait(file, &bd->wq_free, wait);
 875
 876         spin_lock_irq(&bd->lock);
 877         if (!list_empty(&bd->done_list))
 878                 mask |= POLLIN | POLLRDNORM;
 879         if (bd->queued_cmds < bd->max_queue)
 880                 mask |= POLLOUT;
 881         spin_unlock_irq(&bd->lock);
 882
 883         return mask;
 884 }
 885
 886 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 887 {
 888         struct bsg_device *bd = file->private_data;
 889         int __user *uarg = (int __user *) arg;
 890         int ret;
 891
 892         switch (cmd) {
 893                 /*
 894                  * our own ioctls
 895                  */
 896         case SG_GET_COMMAND_Q:
 897                 return put_user(bd->max_queue, uarg);
 898         case SG_SET_COMMAND_Q: {
 899                 int queue;
 900
 901                 if (get_user(queue, uarg))
 902                         return -EFAULT;
 903                 if (queue < 1)
 904                         return -EINVAL;
 905
 906                 spin_lock_irq(&bd->lock);
 907                 bd->max_queue = queue;
 908                 spin_unlock_irq(&bd->lock);
 909                 return 0;
 910         }
 911
 912         /*
 913          * SCSI/sg ioctls
 914          */
 915         case SG_GET_VERSION_NUM:
 916         case SCSI_IOCTL_GET_IDLUN:
 917         case SCSI_IOCTL_GET_BUS_NUMBER:
 918         case SG_SET_TIMEOUT:
 919         case SG_GET_TIMEOUT:
 920         case SG_GET_RESERVED_SIZE:
 921         case SG_SET_RESERVED_SIZE:
 922         case SG_EMULATED_HOST:
 923         case SCSI_IOCTL_SEND_COMMAND: {
 924                 void __user *uarg = (void __user *) arg;
 925                 return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
 926         }
 927         case SG_IO: {
 928                 struct request *rq;
 929                 struct bio *bio, *bidi_bio = NULL;
 930                 struct sg_io_v4 hdr;
 931                 int at_head;
 932                 u8 sense[SCSI_SENSE_BUFFERSIZE];
 933
 934                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 935                         return -EFAULT;
 936
 937                 rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE, sense);
 938                 if (IS_ERR(rq))
 939                         return PTR_ERR(rq);
 940
 941                 bio = rq->bio;
 942                 if (rq->next_rq)
 943                         bidi_bio = rq->next_rq->bio;
 944
 945                 at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
 946                 blk_execute_rq(bd->queue, NULL, rq, at_head);
 947                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 948
 949                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 950                         return -EFAULT;
 951
 952                 return ret;
 953         }
 954         /*
 955          * block device ioctls
 956          */
 957         default:
 958 #if 0
 959                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 960 #else
 961                 return -ENOTTY;
 962 #endif
 963         }
 964 }
 965
 966 static const struct file_operations bsg_fops = {
 967         .read           =       bsg_read,
 968         .write          =       bsg_write,
 969         .poll           =       bsg_poll,
 970         .open           =       bsg_open,
 971         .release        =       bsg_release,
 972         .unlocked_ioctl =       bsg_ioctl,
 973         .owner          =       THIS_MODULE,
 974         .llseek         =       default_llseek,
 975 };
 976
 977 void bsg_unregister_queue(struct request_queue *q)
 978 {
 979         struct bsg_class_device *bcd = &q->bsg_dev;
 980
 981         if (!bcd->class_dev)
 982                 return;
 983
 984         mutex_lock(&bsg_mutex);
 985         idr_remove(&bsg_minor_idr, bcd->minor);
 986         if (q->kobj.sd)
 987                 sysfs_remove_link(&q->kobj, "bsg");
 988         device_unregister(bcd->class_dev);
 989         bcd->class_dev = NULL;
 990         kref_put(&bcd->ref, bsg_kref_release_function);
 991         mutex_unlock(&bsg_mutex);
 992 }
 993 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 994
 995 int bsg_register_queue(struct request_queue *q, struct device *parent,
 996                        const char *name, void (*release)(struct device *))
 997 {
 998         struct bsg_class_device *bcd;
 999         dev_t dev;
1000         int ret;
1001         struct device *class_dev = NULL;
1002         const char *devname;
1003
1004         if (name)
1005                 devname = name;
1006         else
1007                 devname = dev_name(parent);
1008
1009         /*
1010          * we need a proper transport to send commands, not a stacked device
1011          */
1012         if (!queue_is_rq_based(q))
1013                 return 0;
1014
1015         bcd = &q->bsg_dev;
1016         memset(bcd, 0, sizeof(*bcd));
1017
1018         mutex_lock(&bsg_mutex);
1019
1020         ret = idr_alloc(&bsg_minor_idr, bcd, 0, BSG_MAX_DEVS, GFP_KERNEL);
1021         if (ret < 0) {
1022                 if (ret == -ENOSPC) {
1023                         printk(KERN_ERR "bsg: too many bsg devices\n");
1024                         ret = -EINVAL;
1025                 }
1026                 goto unlock;
1027         }
1028
1029         bcd->minor = ret;
1030         bcd->queue = q;
1031         bcd->parent = get_device(parent);
1032         bcd->release = release;
1033         kref_init(&bcd->ref);
1034         dev = MKDEV(bsg_major, bcd->minor);
1035         class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1036         if (IS_ERR(class_dev)) {
1037                 ret = PTR_ERR(class_dev);
1038                 goto put_dev;
1039         }
1040         bcd->class_dev = class_dev;
1041
1042         if (q->kobj.sd) {
1043                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1044                 if (ret)
1045                         goto unregister_class_dev;
1046         }
1047
1048         mutex_unlock(&bsg_mutex);
1049         return 0;
1050
1051 unregister_class_dev:
1052         device_unregister(class_dev);
1053 put_dev:
1054         put_device(parent);
1055         idr_remove(&bsg_minor_idr, bcd->minor);
1056 unlock:
1057         mutex_unlock(&bsg_mutex);
1058         return ret;
1059 }
1060 EXPORT_SYMBOL_GPL(bsg_register_queue);
1061
1062 static struct cdev bsg_cdev;
1063
1064 static char *bsg_devnode(struct device *dev, umode_t *mode)
1065 {
1066         return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
1067 }
1068
1069 static int __init bsg_init(void)
1070 {
1071         int ret, i;
1072         dev_t devid;
1073
1074         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1075                                 sizeof(struct bsg_command), 0, 0, NULL);
1076         if (!bsg_cmd_cachep) {
1077                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1078                 return -ENOMEM;
1079         }
1080
1081         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1082                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1083
1084         bsg_class = class_create(THIS_MODULE, "bsg");
1085         if (IS_ERR(bsg_class)) {
1086                 ret = PTR_ERR(bsg_class);
1087                 goto destroy_kmemcache;
1088         }
1089         bsg_class->devnode = bsg_devnode;
1090
1091         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1092         if (ret)
1093                 goto destroy_bsg_class;
1094
1095         bsg_major = MAJOR(devid);
1096
1097         cdev_init(&bsg_cdev, &bsg_fops);
1098         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1099         if (ret)
1100                 goto unregister_chrdev;
1101
1102         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1103                " loaded (major %d)\n", bsg_major);
1104         return 0;
1105 unregister_chrdev:
1106         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1107 destroy_bsg_class:
1108         class_destroy(bsg_class);
1109 destroy_kmemcache:
1110         kmem_cache_destroy(bsg_cmd_cachep);
1111         return ret;
1112 }
1113
1114 MODULE_AUTHOR("Jens Axboe");
1115 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1116 MODULE_LICENSE("GPL");
1117
1118 device_initcall(bsg_init);