block/bsg.c

   1 /*
   2  * bsg.c - block layer implementation of the sg v4 interface
   3  *
   4  * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
   5  * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
   6  *
   7  *  This file is subject to the terms and conditions of the GNU General Public
   8  *  License version 2.  See the file "COPYING" in the main directory of this
   9  *  archive for more details.
  10  *
  11  */
  12 #include <linux/module.h>
  13 #include <linux/init.h>
  14 #include <linux/file.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/poll.h>
  17 #include <linux/cdev.h>
  18 #include <linux/percpu.h>
  19 #include <linux/uio.h>
  20 #include <linux/idr.h>
  21 #include <linux/bsg.h>
  22
  23 #include <scsi/scsi.h>
  24 #include <scsi/scsi_ioctl.h>
  25 #include <scsi/scsi_cmnd.h>
  26 #include <scsi/scsi_device.h>
  27 #include <scsi/scsi_driver.h>
  28 #include <scsi/sg.h>
  29
  30 #define BSG_DESCRIPTION "Block layer SCSI generic (bsg) driver"
  31 #define BSG_VERSION     "0.4"
  32
  33 struct bsg_device {
  34         struct request_queue *queue;
  35         spinlock_t lock;
  36         struct list_head busy_list;
  37         struct list_head done_list;
  38         struct hlist_node dev_list;
  39         atomic_t ref_count;
  40         int queued_cmds;
  41         int done_cmds;
  42         wait_queue_head_t wq_done;
  43         wait_queue_head_t wq_free;
  44         char name[BUS_ID_SIZE];
  45         int max_queue;
  46         unsigned long flags;
  47         struct blk_scsi_cmd_filter *cmd_filter;
  48         mode_t *f_mode;
  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 {
 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_cmd_filter_verify_command(bd->cmd_filter, rq->cmd,
 190                                                  bd->f_mode))
 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         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 200
 201         rq->timeout = (hdr->timeout * HZ) / 1000;
 202         if (!rq->timeout)
 203                 rq->timeout = q->sg_timeout;
 204         if (!rq->timeout)
 205                 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
 206
 207         return 0;
 208 }
 209
 210 /*
 211  * Check if sg_io_v4 from user is allowed and valid
 212  */
 213 static int
 214 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
 215 {
 216         int ret = 0;
 217
 218         if (hdr->guard != 'Q')
 219                 return -EINVAL;
 220         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 221             hdr->din_xfer_len > (q->max_sectors << 9))
 222                 return -EIO;
 223
 224         switch (hdr->protocol) {
 225         case BSG_PROTOCOL_SCSI:
 226                 switch (hdr->subprotocol) {
 227                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 228                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 229                         break;
 230                 default:
 231                         ret = -EINVAL;
 232                 }
 233                 break;
 234         default:
 235                 ret = -EINVAL;
 236         }
 237
 238         *rw = hdr->dout_xfer_len ? WRITE : READ;
 239         return ret;
 240 }
 241
 242 /*
 243  * map sg_io_v4 to a request.
 244  */
 245 static struct request *
 246 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
 247 {
 248         struct request_queue *q = bd->queue;
 249         struct request *rq, *next_rq = NULL;
 250         int ret, rw;
 251         unsigned int dxfer_len;
 252         void *dxferp = NULL;
 253
 254         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 255                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 256                 hdr->din_xfer_len);
 257
 258         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 259         if (ret)
 260                 return ERR_PTR(ret);
 261
 262         /*
 263          * map scatter-gather elements seperately and string them to request
 264          */
 265         rq = blk_get_request(q, rw, GFP_KERNEL);
 266         if (!rq)
 267                 return ERR_PTR(-ENOMEM);
 268         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd);
 269         if (ret)
 270                 goto out;
 271
 272         if (rw == WRITE && hdr->din_xfer_len) {
 273                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 274                         ret = -EOPNOTSUPP;
 275                         goto out;
 276                 }
 277
 278                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 279                 if (!next_rq) {
 280                         ret = -ENOMEM;
 281                         goto out;
 282                 }
 283                 rq->next_rq = next_rq;
 284                 next_rq->cmd_type = rq->cmd_type;
 285
 286                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 287                 ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
 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);
 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 static void bsg_set_cmd_filter(struct bsg_device *bd,
 571                            struct file *file)
 572 {
 573         struct inode *inode;
 574         struct gendisk *disk;
 575
 576         if (!file)
 577                 return;
 578
 579         inode = file->f_dentry->d_inode;
 580         if (!inode)
 581                 return;
 582
 583         disk = inode->i_bdev->bd_disk;
 584
 585         bd->cmd_filter = &disk->cmd_filter;
 586         bd->f_mode = &file->f_mode;
 587 }
 588
 589 /*
 590  * Check if the error is a "real" error that we should return.
 591  */
 592 static inline int err_block_err(int ret)
 593 {
 594         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 595                 return 1;
 596
 597         return 0;
 598 }
 599
 600 static ssize_t
 601 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 602 {
 603         struct bsg_device *bd = file->private_data;
 604         int ret;
 605         ssize_t bytes_read;
 606
 607         dprintk("%s: read %Zd bytes\n", bd->name, count);
 608
 609         bsg_set_block(bd, file);
 610         bsg_set_cmd_filter(bd, file);
 611
 612         bytes_read = 0;
 613         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 614         *ppos = bytes_read;
 615
 616         if (!bytes_read || (bytes_read && err_block_err(ret)))
 617                 bytes_read = ret;
 618
 619         return bytes_read;
 620 }
 621
 622 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 623                        size_t count, ssize_t *bytes_written)
 624 {
 625         struct bsg_command *bc;
 626         struct request *rq;
 627         int ret, nr_commands;
 628
 629         if (count % sizeof(struct sg_io_v4))
 630                 return -EINVAL;
 631
 632         nr_commands = count / sizeof(struct sg_io_v4);
 633         rq = NULL;
 634         bc = NULL;
 635         ret = 0;
 636         while (nr_commands) {
 637                 struct request_queue *q = bd->queue;
 638
 639                 bc = bsg_alloc_command(bd);
 640                 if (IS_ERR(bc)) {
 641                         ret = PTR_ERR(bc);
 642                         bc = NULL;
 643                         break;
 644                 }
 645
 646                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 647                         ret = -EFAULT;
 648                         break;
 649                 }
 650
 651                 /*
 652                  * get a request, fill in the blanks, and add to request queue
 653                  */
 654                 rq = bsg_map_hdr(bd, &bc->hdr);
 655                 if (IS_ERR(rq)) {
 656                         ret = PTR_ERR(rq);
 657                         rq = NULL;
 658                         break;
 659                 }
 660
 661                 bsg_add_command(bd, q, bc, rq);
 662                 bc = NULL;
 663                 rq = NULL;
 664                 nr_commands--;
 665                 buf += sizeof(struct sg_io_v4);
 666                 *bytes_written += sizeof(struct sg_io_v4);
 667         }
 668
 669         if (bc)
 670                 bsg_free_command(bc);
 671
 672         return ret;
 673 }
 674
 675 static ssize_t
 676 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 677 {
 678         struct bsg_device *bd = file->private_data;
 679         ssize_t bytes_written;
 680         int ret;
 681
 682         dprintk("%s: write %Zd bytes\n", bd->name, count);
 683
 684         bsg_set_block(bd, file);
 685         bsg_set_cmd_filter(bd, file);
 686
 687         bytes_written = 0;
 688         ret = __bsg_write(bd, buf, count, &bytes_written);
 689         *ppos = bytes_written;
 690
 691         /*
 692          * return bytes written on non-fatal errors
 693          */
 694         if (!bytes_written || (bytes_written && err_block_err(ret)))
 695                 bytes_written = ret;
 696
 697         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 698         return bytes_written;
 699 }
 700
 701 static struct bsg_device *bsg_alloc_device(void)
 702 {
 703         struct bsg_device *bd;
 704
 705         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 706         if (unlikely(!bd))
 707                 return NULL;
 708
 709         spin_lock_init(&bd->lock);
 710
 711         bd->max_queue = BSG_DEFAULT_CMDS;
 712
 713         INIT_LIST_HEAD(&bd->busy_list);
 714         INIT_LIST_HEAD(&bd->done_list);
 715         INIT_HLIST_NODE(&bd->dev_list);
 716
 717         init_waitqueue_head(&bd->wq_free);
 718         init_waitqueue_head(&bd->wq_done);
 719         return bd;
 720 }
 721
 722 static void bsg_kref_release_function(struct kref *kref)
 723 {
 724         struct bsg_class_device *bcd =
 725                 container_of(kref, struct bsg_class_device, ref);
 726
 727         if (bcd->release)
 728                 bcd->release(bcd->parent);
 729
 730         put_device(bcd->parent);
 731 }
 732
 733 static int bsg_put_device(struct bsg_device *bd)
 734 {
 735         int ret = 0, do_free;
 736         struct request_queue *q = bd->queue;
 737
 738         mutex_lock(&bsg_mutex);
 739
 740         do_free = atomic_dec_and_test(&bd->ref_count);
 741         if (!do_free)
 742                 goto out;
 743
 744         dprintk("%s: tearing down\n", bd->name);
 745
 746         /*
 747          * close can always block
 748          */
 749         set_bit(BSG_F_BLOCK, &bd->flags);
 750
 751         /*
 752          * correct error detection baddies here again. it's the responsibility
 753          * of the app to properly reap commands before close() if it wants
 754          * fool-proof error detection
 755          */
 756         ret = bsg_complete_all_commands(bd);
 757
 758         hlist_del(&bd->dev_list);
 759         kfree(bd);
 760 out:
 761         mutex_unlock(&bsg_mutex);
 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         int ret;
 774 #ifdef BSG_DEBUG
 775         unsigned char buf[32];
 776 #endif
 777         ret = blk_get_queue(rq);
 778         if (ret)
 779                 return ERR_PTR(-ENXIO);
 780
 781         bd = bsg_alloc_device();
 782         if (!bd) {
 783                 blk_put_queue(rq);
 784                 return ERR_PTR(-ENOMEM);
 785         }
 786
 787         bd->queue = rq;
 788
 789         bsg_set_block(bd, file);
 790         bsg_set_cmd_filter(bd, file);
 791
 792         atomic_set(&bd->ref_count, 1);
 793         mutex_lock(&bsg_mutex);
 794         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
 795
 796         strncpy(bd->name, rq->bsg_dev.class_dev->bus_id, sizeof(bd->name) - 1);
 797         dprintk("bound to <%s>, max queue %d\n",
 798                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 799
 800         mutex_unlock(&bsg_mutex);
 801         return bd;
 802 }
 803
 804 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
 805 {
 806         struct bsg_device *bd;
 807         struct hlist_node *entry;
 808
 809         mutex_lock(&bsg_mutex);
 810
 811         hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
 812                 if (bd->queue == q) {
 813                         atomic_inc(&bd->ref_count);
 814                         goto found;
 815                 }
 816         }
 817         bd = NULL;
 818 found:
 819         mutex_unlock(&bsg_mutex);
 820         return bd;
 821 }
 822
 823 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 824 {
 825         struct bsg_device *bd;
 826         struct bsg_class_device *bcd;
 827
 828         /*
 829          * find the class device
 830          */
 831         mutex_lock(&bsg_mutex);
 832         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 833         if (bcd)
 834                 kref_get(&bcd->ref);
 835         mutex_unlock(&bsg_mutex);
 836
 837         if (!bcd)
 838                 return ERR_PTR(-ENODEV);
 839
 840         bd = __bsg_get_device(iminor(inode), bcd->queue);
 841         if (bd)
 842                 return bd;
 843
 844         bd = bsg_add_device(inode, bcd->queue, file);
 845         if (IS_ERR(bd))
 846                 kref_put(&bcd->ref, bsg_kref_release_function);
 847
 848         return bd;
 849 }
 850
 851 static int bsg_open(struct inode *inode, struct file *file)
 852 {
 853         struct bsg_device *bd = bsg_get_device(inode, file);
 854
 855         if (IS_ERR(bd))
 856                 return PTR_ERR(bd);
 857
 858         file->private_data = bd;
 859         return 0;
 860 }
 861
 862 static int bsg_release(struct inode *inode, struct file *file)
 863 {
 864         struct bsg_device *bd = file->private_data;
 865
 866         file->private_data = NULL;
 867         return bsg_put_device(bd);
 868 }
 869
 870 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 871 {
 872         struct bsg_device *bd = file->private_data;
 873         unsigned int mask = 0;
 874
 875         poll_wait(file, &bd->wq_done, wait);
 876         poll_wait(file, &bd->wq_free, wait);
 877
 878         spin_lock_irq(&bd->lock);
 879         if (!list_empty(&bd->done_list))
 880                 mask |= POLLIN | POLLRDNORM;
 881         if (bd->queued_cmds >= bd->max_queue)
 882                 mask |= POLLOUT;
 883         spin_unlock_irq(&bd->lock);
 884
 885         return mask;
 886 }
 887
 888 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 889 {
 890         struct bsg_device *bd = file->private_data;
 891         int __user *uarg = (int __user *) arg;
 892         int ret;
 893
 894         switch (cmd) {
 895                 /*
 896                  * our own ioctls
 897                  */
 898         case SG_GET_COMMAND_Q:
 899                 return put_user(bd->max_queue, uarg);
 900         case SG_SET_COMMAND_Q: {
 901                 int queue;
 902
 903                 if (get_user(queue, uarg))
 904                         return -EFAULT;
 905                 if (queue < 1)
 906                         return -EINVAL;
 907
 908                 spin_lock_irq(&bd->lock);
 909                 bd->max_queue = queue;
 910                 spin_unlock_irq(&bd->lock);
 911                 return 0;
 912         }
 913
 914         /*
 915          * SCSI/sg ioctls
 916          */
 917         case SG_GET_VERSION_NUM:
 918         case SCSI_IOCTL_GET_IDLUN:
 919         case SCSI_IOCTL_GET_BUS_NUMBER:
 920         case SG_SET_TIMEOUT:
 921         case SG_GET_TIMEOUT:
 922         case SG_GET_RESERVED_SIZE:
 923         case SG_SET_RESERVED_SIZE:
 924         case SG_EMULATED_HOST:
 925         case SCSI_IOCTL_SEND_COMMAND: {
 926                 void __user *uarg = (void __user *) arg;
 927                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 928         }
 929         case SG_IO: {
 930                 struct request *rq;
 931                 struct bio *bio, *bidi_bio = NULL;
 932                 struct sg_io_v4 hdr;
 933
 934                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 935                         return -EFAULT;
 936
 937                 rq = bsg_map_hdr(bd, &hdr);
 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                 blk_execute_rq(bd->queue, NULL, rq, 0);
 945                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 946
 947                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 948                         return -EFAULT;
 949
 950                 return ret;
 951         }
 952         /*
 953          * block device ioctls
 954          */
 955         default:
 956 #if 0
 957                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 958 #else
 959                 return -ENOTTY;
 960 #endif
 961         }
 962 }
 963
 964 static const struct file_operations bsg_fops = {
 965         .read           =       bsg_read,
 966         .write          =       bsg_write,
 967         .poll           =       bsg_poll,
 968         .open           =       bsg_open,
 969         .release        =       bsg_release,
 970         .unlocked_ioctl =       bsg_ioctl,
 971         .owner          =       THIS_MODULE,
 972 };
 973
 974 void bsg_unregister_queue(struct request_queue *q)
 975 {
 976         struct bsg_class_device *bcd = &q->bsg_dev;
 977
 978         if (!bcd->class_dev)
 979                 return;
 980
 981         mutex_lock(&bsg_mutex);
 982         idr_remove(&bsg_minor_idr, bcd->minor);
 983         sysfs_remove_link(&q->kobj, "bsg");
 984         device_unregister(bcd->class_dev);
 985         bcd->class_dev = NULL;
 986         kref_put(&bcd->ref, bsg_kref_release_function);
 987         mutex_unlock(&bsg_mutex);
 988 }
 989 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 990
 991 int bsg_register_queue(struct request_queue *q, struct device *parent,
 992                        const char *name, void (*release)(struct device *))
 993 {
 994         struct bsg_class_device *bcd;
 995         dev_t dev;
 996         int ret, minor;
 997         struct device *class_dev = NULL;
 998         const char *devname;
 999
1000         if (name)
1001                 devname = name;
1002         else
1003                 devname = parent->bus_id;
1004
1005         /*
1006          * we need a proper transport to send commands, not a stacked device
1007          */
1008         if (!q->request_fn)
1009                 return 0;
1010
1011         bcd = &q->bsg_dev;
1012         memset(bcd, 0, sizeof(*bcd));
1013
1014         mutex_lock(&bsg_mutex);
1015
1016         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1017         if (!ret) {
1018                 ret = -ENOMEM;
1019                 goto unlock;
1020         }
1021
1022         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1023         if (ret < 0)
1024                 goto unlock;
1025
1026         if (minor >= BSG_MAX_DEVS) {
1027                 printk(KERN_ERR "bsg: too many bsg devices\n");
1028                 ret = -EINVAL;
1029                 goto remove_idr;
1030         }
1031
1032         bcd->minor = minor;
1033         bcd->queue = q;
1034         bcd->parent = get_device(parent);
1035         bcd->release = release;
1036         kref_init(&bcd->ref);
1037         dev = MKDEV(bsg_major, bcd->minor);
1038         class_dev = device_create(bsg_class, parent, dev, "%s", devname);
1039         if (IS_ERR(class_dev)) {
1040                 ret = PTR_ERR(class_dev);
1041                 goto put_dev;
1042         }
1043         bcd->class_dev = class_dev;
1044
1045         if (q->kobj.sd) {
1046                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1047                 if (ret)
1048                         goto unregister_class_dev;
1049         }
1050
1051         mutex_unlock(&bsg_mutex);
1052         return 0;
1053
1054 unregister_class_dev:
1055         device_unregister(class_dev);
1056 put_dev:
1057         put_device(parent);
1058 remove_idr:
1059         idr_remove(&bsg_minor_idr, minor);
1060 unlock:
1061         mutex_unlock(&bsg_mutex);
1062         return ret;
1063 }
1064 EXPORT_SYMBOL_GPL(bsg_register_queue);
1065
1066 static struct cdev bsg_cdev;
1067
1068 static int __init bsg_init(void)
1069 {
1070         int ret, i;
1071         dev_t devid;
1072
1073         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1074                                 sizeof(struct bsg_command), 0, 0, NULL);
1075         if (!bsg_cmd_cachep) {
1076                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1077                 return -ENOMEM;
1078         }
1079
1080         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1081                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1082
1083         bsg_class = class_create(THIS_MODULE, "bsg");
1084         if (IS_ERR(bsg_class)) {
1085                 ret = PTR_ERR(bsg_class);
1086                 goto destroy_kmemcache;
1087         }
1088
1089         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1090         if (ret)
1091                 goto destroy_bsg_class;
1092
1093         bsg_major = MAJOR(devid);
1094
1095         cdev_init(&bsg_cdev, &bsg_fops);
1096         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1097         if (ret)
1098                 goto unregister_chrdev;
1099
1100         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1101                " loaded (major %d)\n", bsg_major);
1102         return 0;
1103 unregister_chrdev:
1104         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1105 destroy_bsg_class:
1106         class_destroy(bsg_class);
1107 destroy_kmemcache:
1108         kmem_cache_destroy(bsg_cmd_cachep);
1109         return ret;
1110 }
1111
1112 MODULE_AUTHOR("Jens Axboe");
1113 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1114 MODULE_LICENSE("GPL");
1115
1116 device_initcall(bsg_init);