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             u8 *sense)
 249 {
 250         struct request_queue *q = bd->queue;
 251         struct request *rq, *next_rq = NULL;
 252         int ret, rw;
 253         unsigned int dxfer_len;
 254         void *dxferp = NULL;
 255
 256         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 257                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 258                 hdr->din_xfer_len);
 259
 260         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 261         if (ret)
 262                 return ERR_PTR(ret);
 263
 264         /*
 265          * map scatter-gather elements seperately and string them to request
 266          */
 267         rq = blk_get_request(q, rw, GFP_KERNEL);
 268         if (!rq)
 269                 return ERR_PTR(-ENOMEM);
 270         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
 271         if (ret)
 272                 goto out;
 273
 274         if (rw == WRITE && hdr->din_xfer_len) {
 275                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 276                         ret = -EOPNOTSUPP;
 277                         goto out;
 278                 }
 279
 280                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 281                 if (!next_rq) {
 282                         ret = -ENOMEM;
 283                         goto out;
 284                 }
 285                 rq->next_rq = next_rq;
 286                 next_rq->cmd_type = rq->cmd_type;
 287
 288                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 289                 ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
 290                                        hdr->din_xfer_len, GFP_KERNEL);
 291                 if (ret)
 292                         goto out;
 293         }
 294
 295         if (hdr->dout_xfer_len) {
 296                 dxfer_len = hdr->dout_xfer_len;
 297                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 298         } else if (hdr->din_xfer_len) {
 299                 dxfer_len = hdr->din_xfer_len;
 300                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 301         } else
 302                 dxfer_len = 0;
 303
 304         if (dxfer_len) {
 305                 ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
 306                                       GFP_KERNEL);
 307                 if (ret)
 308                         goto out;
 309         }
 310
 311         rq->sense = sense;
 312         rq->sense_len = 0;
 313
 314         return rq;
 315 out:
 316         if (rq->cmd != rq->__cmd)
 317                 kfree(rq->cmd);
 318         blk_put_request(rq);
 319         if (next_rq) {
 320                 blk_rq_unmap_user(next_rq->bio);
 321                 blk_put_request(next_rq);
 322         }
 323         return ERR_PTR(ret);
 324 }
 325
 326 /*
 327  * async completion call-back from the block layer, when scsi/ide/whatever
 328  * calls end_that_request_last() on a request
 329  */
 330 static void bsg_rq_end_io(struct request *rq, int uptodate)
 331 {
 332         struct bsg_command *bc = rq->end_io_data;
 333         struct bsg_device *bd = bc->bd;
 334         unsigned long flags;
 335
 336         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 337                 bd->name, rq, bc, bc->bio, uptodate);
 338
 339         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 340
 341         spin_lock_irqsave(&bd->lock, flags);
 342         list_move_tail(&bc->list, &bd->done_list);
 343         bd->done_cmds++;
 344         spin_unlock_irqrestore(&bd->lock, flags);
 345
 346         wake_up(&bd->wq_done);
 347 }
 348
 349 /*
 350  * do final setup of a 'bc' and submit the matching 'rq' to the block
 351  * layer for io
 352  */
 353 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
 354                             struct bsg_command *bc, struct request *rq)
 355 {
 356         /*
 357          * add bc command to busy queue and submit rq for io
 358          */
 359         bc->rq = rq;
 360         bc->bio = rq->bio;
 361         if (rq->next_rq)
 362                 bc->bidi_bio = rq->next_rq->bio;
 363         bc->hdr.duration = jiffies;
 364         spin_lock_irq(&bd->lock);
 365         list_add_tail(&bc->list, &bd->busy_list);
 366         spin_unlock_irq(&bd->lock);
 367
 368         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 369
 370         rq->end_io_data = bc;
 371         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 372 }
 373
 374 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 375 {
 376         struct bsg_command *bc = NULL;
 377
 378         spin_lock_irq(&bd->lock);
 379         if (bd->done_cmds) {
 380                 bc = list_first_entry(&bd->done_list, struct bsg_command, list);
 381                 list_del(&bc->list);
 382                 bd->done_cmds--;
 383         }
 384         spin_unlock_irq(&bd->lock);
 385
 386         return bc;
 387 }
 388
 389 /*
 390  * Get a finished command from the done list
 391  */
 392 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 393 {
 394         struct bsg_command *bc;
 395         int ret;
 396
 397         do {
 398                 bc = bsg_next_done_cmd(bd);
 399                 if (bc)
 400                         break;
 401
 402                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 403                         bc = ERR_PTR(-EAGAIN);
 404                         break;
 405                 }
 406
 407                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 408                 if (ret) {
 409                         bc = ERR_PTR(-ERESTARTSYS);
 410                         break;
 411                 }
 412         } while (1);
 413
 414         dprintk("%s: returning done %p\n", bd->name, bc);
 415
 416         return bc;
 417 }
 418
 419 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 420                                     struct bio *bio, struct bio *bidi_bio)
 421 {
 422         int ret = 0;
 423
 424         dprintk("rq %p bio %p 0x%x\n", rq, bio, rq->errors);
 425         /*
 426          * fill in all the output members
 427          */
 428         hdr->device_status = status_byte(rq->errors);
 429         hdr->transport_status = host_byte(rq->errors);
 430         hdr->driver_status = driver_byte(rq->errors);
 431         hdr->info = 0;
 432         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 433                 hdr->info |= SG_INFO_CHECK;
 434         hdr->response_len = 0;
 435
 436         if (rq->sense_len && hdr->response) {
 437                 int len = min_t(unsigned int, hdr->max_response_len,
 438                                         rq->sense_len);
 439
 440                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 441                                    rq->sense, len);
 442                 if (!ret)
 443                         hdr->response_len = len;
 444                 else
 445                         ret = -EFAULT;
 446         }
 447
 448         if (rq->next_rq) {
 449                 hdr->dout_resid = rq->data_len;
 450                 hdr->din_resid = rq->next_rq->data_len;
 451                 blk_rq_unmap_user(bidi_bio);
 452                 blk_put_request(rq->next_rq);
 453         } else if (rq_data_dir(rq) == READ)
 454                 hdr->din_resid = rq->data_len;
 455         else
 456                 hdr->dout_resid = rq->data_len;
 457
 458         /*
 459          * If the request generated a negative error number, return it
 460          * (providing we aren't already returning an error); if it's
 461          * just a protocol response (i.e. non negative), that gets
 462          * processed above.
 463          */
 464         if (!ret && rq->errors < 0)
 465                 ret = rq->errors;
 466
 467         blk_rq_unmap_user(bio);
 468         if (rq->cmd != rq->__cmd)
 469                 kfree(rq->cmd);
 470         blk_put_request(rq);
 471
 472         return ret;
 473 }
 474
 475 static int bsg_complete_all_commands(struct bsg_device *bd)
 476 {
 477         struct bsg_command *bc;
 478         int ret, tret;
 479
 480         dprintk("%s: entered\n", bd->name);
 481
 482         /*
 483          * wait for all commands to complete
 484          */
 485         ret = 0;
 486         do {
 487                 ret = bsg_io_schedule(bd);
 488                 /*
 489                  * look for -ENODATA specifically -- we'll sometimes get
 490                  * -ERESTARTSYS when we've taken a signal, but we can't
 491                  * return until we're done freeing the queue, so ignore
 492                  * it.  The signal will get handled when we're done freeing
 493                  * the bsg_device.
 494                  */
 495         } while (ret != -ENODATA);
 496
 497         /*
 498          * discard done commands
 499          */
 500         ret = 0;
 501         do {
 502                 spin_lock_irq(&bd->lock);
 503                 if (!bd->queued_cmds) {
 504                         spin_unlock_irq(&bd->lock);
 505                         break;
 506                 }
 507                 spin_unlock_irq(&bd->lock);
 508
 509                 bc = bsg_get_done_cmd(bd);
 510                 if (IS_ERR(bc))
 511                         break;
 512
 513                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 514                                                 bc->bidi_bio);
 515                 if (!ret)
 516                         ret = tret;
 517
 518                 bsg_free_command(bc);
 519         } while (1);
 520
 521         return ret;
 522 }
 523
 524 static int
 525 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 526            const struct iovec *iov, ssize_t *bytes_read)
 527 {
 528         struct bsg_command *bc;
 529         int nr_commands, ret;
 530
 531         if (count % sizeof(struct sg_io_v4))
 532                 return -EINVAL;
 533
 534         ret = 0;
 535         nr_commands = count / sizeof(struct sg_io_v4);
 536         while (nr_commands) {
 537                 bc = bsg_get_done_cmd(bd);
 538                 if (IS_ERR(bc)) {
 539                         ret = PTR_ERR(bc);
 540                         break;
 541                 }
 542
 543                 /*
 544                  * this is the only case where we need to copy data back
 545                  * after completing the request. so do that here,
 546                  * bsg_complete_work() cannot do that for us
 547                  */
 548                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 549                                                bc->bidi_bio);
 550
 551                 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
 552                         ret = -EFAULT;
 553
 554                 bsg_free_command(bc);
 555
 556                 if (ret)
 557                         break;
 558
 559                 buf += sizeof(struct sg_io_v4);
 560                 *bytes_read += sizeof(struct sg_io_v4);
 561                 nr_commands--;
 562         }
 563
 564         return ret;
 565 }
 566
 567 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 568 {
 569         if (file->f_flags & O_NONBLOCK)
 570                 clear_bit(BSG_F_BLOCK, &bd->flags);
 571         else
 572                 set_bit(BSG_F_BLOCK, &bd->flags);
 573 }
 574
 575 /*
 576  * Check if the error is a "real" error that we should return.
 577  */
 578 static inline int err_block_err(int ret)
 579 {
 580         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 581                 return 1;
 582
 583         return 0;
 584 }
 585
 586 static ssize_t
 587 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 588 {
 589         struct bsg_device *bd = file->private_data;
 590         int ret;
 591         ssize_t bytes_read;
 592
 593         dprintk("%s: read %Zd bytes\n", bd->name, count);
 594
 595         bsg_set_block(bd, file);
 596
 597         bytes_read = 0;
 598         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 599         *ppos = bytes_read;
 600
 601         if (!bytes_read || (bytes_read && err_block_err(ret)))
 602                 bytes_read = ret;
 603
 604         return bytes_read;
 605 }
 606
 607 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
 608                        size_t count, ssize_t *bytes_written,
 609                        fmode_t has_write_perm)
 610 {
 611         struct bsg_command *bc;
 612         struct request *rq;
 613         int ret, nr_commands;
 614
 615         if (count % sizeof(struct sg_io_v4))
 616                 return -EINVAL;
 617
 618         nr_commands = count / sizeof(struct sg_io_v4);
 619         rq = NULL;
 620         bc = NULL;
 621         ret = 0;
 622         while (nr_commands) {
 623                 struct request_queue *q = bd->queue;
 624
 625                 bc = bsg_alloc_command(bd);
 626                 if (IS_ERR(bc)) {
 627                         ret = PTR_ERR(bc);
 628                         bc = NULL;
 629                         break;
 630                 }
 631
 632                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 633                         ret = -EFAULT;
 634                         break;
 635                 }
 636
 637                 /*
 638                  * get a request, fill in the blanks, and add to request queue
 639                  */
 640                 rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm, bc->sense);
 641                 if (IS_ERR(rq)) {
 642                         ret = PTR_ERR(rq);
 643                         rq = NULL;
 644                         break;
 645                 }
 646
 647                 bsg_add_command(bd, q, bc, rq);
 648                 bc = NULL;
 649                 rq = NULL;
 650                 nr_commands--;
 651                 buf += sizeof(struct sg_io_v4);
 652                 *bytes_written += sizeof(struct sg_io_v4);
 653         }
 654
 655         if (bc)
 656                 bsg_free_command(bc);
 657
 658         return ret;
 659 }
 660
 661 static ssize_t
 662 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 663 {
 664         struct bsg_device *bd = file->private_data;
 665         ssize_t bytes_written;
 666         int ret;
 667
 668         dprintk("%s: write %Zd bytes\n", bd->name, count);
 669
 670         bsg_set_block(bd, file);
 671
 672         bytes_written = 0;
 673         ret = __bsg_write(bd, buf, count, &bytes_written,
 674                           file->f_mode & FMODE_WRITE);
 675
 676         *ppos = bytes_written;
 677
 678         /*
 679          * return bytes written on non-fatal errors
 680          */
 681         if (!bytes_written || (bytes_written && err_block_err(ret)))
 682                 bytes_written = ret;
 683
 684         dprintk("%s: returning %Zd\n", bd->name, bytes_written);
 685         return bytes_written;
 686 }
 687
 688 static struct bsg_device *bsg_alloc_device(void)
 689 {
 690         struct bsg_device *bd;
 691
 692         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 693         if (unlikely(!bd))
 694                 return NULL;
 695
 696         spin_lock_init(&bd->lock);
 697
 698         bd->max_queue = BSG_DEFAULT_CMDS;
 699
 700         INIT_LIST_HEAD(&bd->busy_list);
 701         INIT_LIST_HEAD(&bd->done_list);
 702         INIT_HLIST_NODE(&bd->dev_list);
 703
 704         init_waitqueue_head(&bd->wq_free);
 705         init_waitqueue_head(&bd->wq_done);
 706         return bd;
 707 }
 708
 709 static void bsg_kref_release_function(struct kref *kref)
 710 {
 711         struct bsg_class_device *bcd =
 712                 container_of(kref, struct bsg_class_device, ref);
 713         struct device *parent = bcd->parent;
 714
 715         if (bcd->release)
 716                 bcd->release(bcd->parent);
 717
 718         put_device(parent);
 719 }
 720
 721 static int bsg_put_device(struct bsg_device *bd)
 722 {
 723         int ret = 0, do_free;
 724         struct request_queue *q = bd->queue;
 725
 726         mutex_lock(&bsg_mutex);
 727
 728         do_free = atomic_dec_and_test(&bd->ref_count);
 729         if (!do_free) {
 730                 mutex_unlock(&bsg_mutex);
 731                 goto out;
 732         }
 733
 734         hlist_del(&bd->dev_list);
 735         mutex_unlock(&bsg_mutex);
 736
 737         dprintk("%s: tearing down\n", bd->name);
 738
 739         /*
 740          * close can always block
 741          */
 742         set_bit(BSG_F_BLOCK, &bd->flags);
 743
 744         /*
 745          * correct error detection baddies here again. it's the responsibility
 746          * of the app to properly reap commands before close() if it wants
 747          * fool-proof error detection
 748          */
 749         ret = bsg_complete_all_commands(bd);
 750
 751         kfree(bd);
 752 out:
 753         kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
 754         if (do_free)
 755                 blk_put_queue(q);
 756         return ret;
 757 }
 758
 759 static struct bsg_device *bsg_add_device(struct inode *inode,
 760                                          struct request_queue *rq,
 761                                          struct file *file)
 762 {
 763         struct bsg_device *bd;
 764         int ret;
 765 #ifdef BSG_DEBUG
 766         unsigned char buf[32];
 767 #endif
 768         ret = blk_get_queue(rq);
 769         if (ret)
 770                 return ERR_PTR(-ENXIO);
 771
 772         bd = bsg_alloc_device();
 773         if (!bd) {
 774                 blk_put_queue(rq);
 775                 return ERR_PTR(-ENOMEM);
 776         }
 777
 778         bd->queue = rq;
 779
 780         bsg_set_block(bd, file);
 781
 782         atomic_set(&bd->ref_count, 1);
 783         mutex_lock(&bsg_mutex);
 784         hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
 785
 786         strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
 787         dprintk("bound to <%s>, max queue %d\n",
 788                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 789
 790         mutex_unlock(&bsg_mutex);
 791         return bd;
 792 }
 793
 794 static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
 795 {
 796         struct bsg_device *bd;
 797         struct hlist_node *entry;
 798
 799         mutex_lock(&bsg_mutex);
 800
 801         hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
 802                 if (bd->queue == q) {
 803                         atomic_inc(&bd->ref_count);
 804                         goto found;
 805                 }
 806         }
 807         bd = NULL;
 808 found:
 809         mutex_unlock(&bsg_mutex);
 810         return bd;
 811 }
 812
 813 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 814 {
 815         struct bsg_device *bd;
 816         struct bsg_class_device *bcd;
 817
 818         /*
 819          * find the class device
 820          */
 821         mutex_lock(&bsg_mutex);
 822         bcd = idr_find(&bsg_minor_idr, iminor(inode));
 823         if (bcd)
 824                 kref_get(&bcd->ref);
 825         mutex_unlock(&bsg_mutex);
 826
 827         if (!bcd)
 828                 return ERR_PTR(-ENODEV);
 829
 830         bd = __bsg_get_device(iminor(inode), bcd->queue);
 831         if (bd)
 832                 return bd;
 833
 834         bd = bsg_add_device(inode, bcd->queue, file);
 835         if (IS_ERR(bd))
 836                 kref_put(&bcd->ref, bsg_kref_release_function);
 837
 838         return bd;
 839 }
 840
 841 static int bsg_open(struct inode *inode, struct file *file)
 842 {
 843         struct bsg_device *bd;
 844
 845         lock_kernel();
 846         bd = bsg_get_device(inode, file);
 847         unlock_kernel();
 848
 849         if (IS_ERR(bd))
 850                 return PTR_ERR(bd);
 851
 852         file->private_data = bd;
 853         return 0;
 854 }
 855
 856 static int bsg_release(struct inode *inode, struct file *file)
 857 {
 858         struct bsg_device *bd = file->private_data;
 859
 860         file->private_data = NULL;
 861         return bsg_put_device(bd);
 862 }
 863
 864 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 865 {
 866         struct bsg_device *bd = file->private_data;
 867         unsigned int mask = 0;
 868
 869         poll_wait(file, &bd->wq_done, wait);
 870         poll_wait(file, &bd->wq_free, wait);
 871
 872         spin_lock_irq(&bd->lock);
 873         if (!list_empty(&bd->done_list))
 874                 mask |= POLLIN | POLLRDNORM;
 875         if (bd->queued_cmds >= bd->max_queue)
 876                 mask |= POLLOUT;
 877         spin_unlock_irq(&bd->lock);
 878
 879         return mask;
 880 }
 881
 882 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 883 {
 884         struct bsg_device *bd = file->private_data;
 885         int __user *uarg = (int __user *) arg;
 886         int ret;
 887
 888         switch (cmd) {
 889                 /*
 890                  * our own ioctls
 891                  */
 892         case SG_GET_COMMAND_Q:
 893                 return put_user(bd->max_queue, uarg);
 894         case SG_SET_COMMAND_Q: {
 895                 int queue;
 896
 897                 if (get_user(queue, uarg))
 898                         return -EFAULT;
 899                 if (queue < 1)
 900                         return -EINVAL;
 901
 902                 spin_lock_irq(&bd->lock);
 903                 bd->max_queue = queue;
 904                 spin_unlock_irq(&bd->lock);
 905                 return 0;
 906         }
 907
 908         /*
 909          * SCSI/sg ioctls
 910          */
 911         case SG_GET_VERSION_NUM:
 912         case SCSI_IOCTL_GET_IDLUN:
 913         case SCSI_IOCTL_GET_BUS_NUMBER:
 914         case SG_SET_TIMEOUT:
 915         case SG_GET_TIMEOUT:
 916         case SG_GET_RESERVED_SIZE:
 917         case SG_SET_RESERVED_SIZE:
 918         case SG_EMULATED_HOST:
 919         case SCSI_IOCTL_SEND_COMMAND: {
 920                 void __user *uarg = (void __user *) arg;
 921                 return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
 922         }
 923         case SG_IO: {
 924                 struct request *rq;
 925                 struct bio *bio, *bidi_bio = NULL;
 926                 struct sg_io_v4 hdr;
 927                 u8 sense[SCSI_SENSE_BUFFERSIZE];
 928
 929                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 930                         return -EFAULT;
 931
 932                 rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE, sense);
 933                 if (IS_ERR(rq))
 934                         return PTR_ERR(rq);
 935
 936                 bio = rq->bio;
 937                 if (rq->next_rq)
 938                         bidi_bio = rq->next_rq->bio;
 939                 blk_execute_rq(bd->queue, NULL, rq, 0);
 940                 ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 941
 942                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 943                         return -EFAULT;
 944
 945                 return ret;
 946         }
 947         /*
 948          * block device ioctls
 949          */
 950         default:
 951 #if 0
 952                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 953 #else
 954                 return -ENOTTY;
 955 #endif
 956         }
 957 }
 958
 959 static const struct file_operations bsg_fops = {
 960         .read           =       bsg_read,
 961         .write          =       bsg_write,
 962         .poll           =       bsg_poll,
 963         .open           =       bsg_open,
 964         .release        =       bsg_release,
 965         .unlocked_ioctl =       bsg_ioctl,
 966         .owner          =       THIS_MODULE,
 967 };
 968
 969 void bsg_unregister_queue(struct request_queue *q)
 970 {
 971         struct bsg_class_device *bcd = &q->bsg_dev;
 972
 973         if (!bcd->class_dev)
 974                 return;
 975
 976         mutex_lock(&bsg_mutex);
 977         idr_remove(&bsg_minor_idr, bcd->minor);
 978         sysfs_remove_link(&q->kobj, "bsg");
 979         device_unregister(bcd->class_dev);
 980         bcd->class_dev = NULL;
 981         kref_put(&bcd->ref, bsg_kref_release_function);
 982         mutex_unlock(&bsg_mutex);
 983 }
 984 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 985
 986 int bsg_register_queue(struct request_queue *q, struct device *parent,
 987                        const char *name, void (*release)(struct device *))
 988 {
 989         struct bsg_class_device *bcd;
 990         dev_t dev;
 991         int ret, minor;
 992         struct device *class_dev = NULL;
 993         const char *devname;
 994
 995         if (name)
 996                 devname = name;
 997         else
 998                 devname = dev_name(parent);
 999
1000         /*
1001          * we need a proper transport to send commands, not a stacked device
1002          */
1003         if (!q->request_fn)
1004                 return 0;
1005
1006         bcd = &q->bsg_dev;
1007         memset(bcd, 0, sizeof(*bcd));
1008
1009         mutex_lock(&bsg_mutex);
1010
1011         ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
1012         if (!ret) {
1013                 ret = -ENOMEM;
1014                 goto unlock;
1015         }
1016
1017         ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
1018         if (ret < 0)
1019                 goto unlock;
1020
1021         if (minor >= BSG_MAX_DEVS) {
1022                 printk(KERN_ERR "bsg: too many bsg devices\n");
1023                 ret = -EINVAL;
1024                 goto remove_idr;
1025         }
1026
1027         bcd->minor = minor;
1028         bcd->queue = q;
1029         bcd->parent = get_device(parent);
1030         bcd->release = release;
1031         kref_init(&bcd->ref);
1032         dev = MKDEV(bsg_major, bcd->minor);
1033         class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
1034         if (IS_ERR(class_dev)) {
1035                 ret = PTR_ERR(class_dev);
1036                 goto put_dev;
1037         }
1038         bcd->class_dev = class_dev;
1039
1040         if (q->kobj.sd) {
1041                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1042                 if (ret)
1043                         goto unregister_class_dev;
1044         }
1045
1046         mutex_unlock(&bsg_mutex);
1047         return 0;
1048
1049 unregister_class_dev:
1050         device_unregister(class_dev);
1051 put_dev:
1052         put_device(parent);
1053 remove_idr:
1054         idr_remove(&bsg_minor_idr, minor);
1055 unlock:
1056         mutex_unlock(&bsg_mutex);
1057         return ret;
1058 }
1059 EXPORT_SYMBOL_GPL(bsg_register_queue);
1060
1061 static struct cdev bsg_cdev;
1062
1063 static int __init bsg_init(void)
1064 {
1065         int ret, i;
1066         dev_t devid;
1067
1068         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1069                                 sizeof(struct bsg_command), 0, 0, NULL);
1070         if (!bsg_cmd_cachep) {
1071                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1072                 return -ENOMEM;
1073         }
1074
1075         for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1076                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1077
1078         bsg_class = class_create(THIS_MODULE, "bsg");
1079         if (IS_ERR(bsg_class)) {
1080                 ret = PTR_ERR(bsg_class);
1081                 goto destroy_kmemcache;
1082         }
1083
1084         ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1085         if (ret)
1086                 goto destroy_bsg_class;
1087
1088         bsg_major = MAJOR(devid);
1089
1090         cdev_init(&bsg_cdev, &bsg_fops);
1091         ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1092         if (ret)
1093                 goto unregister_chrdev;
1094
1095         printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1096                " loaded (major %d)\n", bsg_major);
1097         return 0;
1098 unregister_chrdev:
1099         unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1100 destroy_bsg_class:
1101         class_destroy(bsg_class);
1102 destroy_kmemcache:
1103         kmem_cache_destroy(bsg_cmd_cachep);
1104         return ret;
1105 }
1106
1107 MODULE_AUTHOR("Jens Axboe");
1108 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1109 MODULE_LICENSE("GPL");
1110
1111 device_initcall(bsg_init);