block/bsg.c

   1 /*
   2  * bsg.c - block layer implementation of the sg v3 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 /*
  13  * TODO
  14  *      - Should this get merged, block/scsi_ioctl.c will be migrated into
  15  *        this file. To keep maintenance down, it's easier to have them
  16  *        seperated right now.
  17  *
  18  */
  19 #include <linux/module.h>
  20 #include <linux/init.h>
  21 #include <linux/file.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/poll.h>
  24 #include <linux/cdev.h>
  25 #include <linux/percpu.h>
  26 #include <linux/uio.h>
  27 #include <linux/bsg.h>
  28
  29 #include <scsi/scsi.h>
  30 #include <scsi/scsi_ioctl.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_device.h>
  33 #include <scsi/scsi_driver.h>
  34 #include <scsi/sg.h>
  35
  36 static char bsg_version[] = "block layer sg (bsg) 0.4";
  37
  38 struct bsg_device {
  39         request_queue_t *queue;
  40         spinlock_t lock;
  41         struct list_head busy_list;
  42         struct list_head done_list;
  43         struct hlist_node dev_list;
  44         atomic_t ref_count;
  45         int minor;
  46         int queued_cmds;
  47         int done_cmds;
  48         wait_queue_head_t wq_done;
  49         wait_queue_head_t wq_free;
  50         char name[BUS_ID_SIZE];
  51         int max_queue;
  52         unsigned long flags;
  53 };
  54
  55 enum {
  56         BSG_F_BLOCK             = 1,
  57         BSG_F_WRITE_PERM        = 2,
  58 };
  59
  60 #define BSG_DEFAULT_CMDS        64
  61 #define BSG_MAX_DEVS            32768
  62
  63 #undef BSG_DEBUG
  64
  65 #ifdef BSG_DEBUG
  66 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args)
  67 #else
  68 #define dprintk(fmt, args...)
  69 #endif
  70
  71 #define list_entry_bc(entry)    list_entry((entry), struct bsg_command, list)
  72
  73 /*
  74  * just for testing
  75  */
  76 #define BSG_MAJOR       (240)
  77
  78 static DEFINE_MUTEX(bsg_mutex);
  79 static int bsg_device_nr, bsg_minor_idx;
  80
  81 #define BSG_LIST_SIZE   (8)
  82 #define bsg_list_idx(minor)     ((minor) & (BSG_LIST_SIZE - 1))
  83 static struct hlist_head bsg_device_list[BSG_LIST_SIZE];
  84
  85 static struct class *bsg_class;
  86 static LIST_HEAD(bsg_class_list);
  87
  88 static struct kmem_cache *bsg_cmd_cachep;
  89
  90 /*
  91  * our internal command type
  92  */
  93 struct bsg_command {
  94         struct bsg_device *bd;
  95         struct list_head list;
  96         struct request *rq;
  97         struct bio *bio;
  98         struct bio *bidi_bio;
  99         int err;
 100         struct sg_io_v4 hdr;
 101         struct sg_io_v4 __user *uhdr;
 102         char sense[SCSI_SENSE_BUFFERSIZE];
 103 };
 104
 105 static void bsg_free_command(struct bsg_command *bc)
 106 {
 107         struct bsg_device *bd = bc->bd;
 108         unsigned long flags;
 109
 110         kmem_cache_free(bsg_cmd_cachep, bc);
 111
 112         spin_lock_irqsave(&bd->lock, flags);
 113         bd->queued_cmds--;
 114         spin_unlock_irqrestore(&bd->lock, flags);
 115
 116         wake_up(&bd->wq_free);
 117 }
 118
 119 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
 120 {
 121         struct bsg_command *bc = ERR_PTR(-EINVAL);
 122
 123         spin_lock_irq(&bd->lock);
 124
 125         if (bd->queued_cmds >= bd->max_queue)
 126                 goto out;
 127
 128         bd->queued_cmds++;
 129         spin_unlock_irq(&bd->lock);
 130
 131         bc = kmem_cache_alloc(bsg_cmd_cachep, GFP_USER);
 132         if (unlikely(!bc)) {
 133                 spin_lock_irq(&bd->lock);
 134                 bd->queued_cmds--;
 135                 bc = ERR_PTR(-ENOMEM);
 136                 goto out;
 137         }
 138
 139         memset(bc, 0, sizeof(*bc));
 140         bc->bd = bd;
 141         INIT_LIST_HEAD(&bc->list);
 142         dprintk("%s: returning free cmd %p\n", bd->name, bc);
 143         return bc;
 144 out:
 145         spin_unlock_irq(&bd->lock);
 146         return bc;
 147 }
 148
 149 static inline void
 150 bsg_del_done_cmd(struct bsg_device *bd, struct bsg_command *bc)
 151 {
 152         bd->done_cmds--;
 153         list_del(&bc->list);
 154 }
 155
 156 static inline void
 157 bsg_add_done_cmd(struct bsg_device *bd, struct bsg_command *bc)
 158 {
 159         bd->done_cmds++;
 160         list_add_tail(&bc->list, &bd->done_list);
 161         wake_up(&bd->wq_done);
 162 }
 163
 164 static inline int bsg_io_schedule(struct bsg_device *bd, int state)
 165 {
 166         DEFINE_WAIT(wait);
 167         int ret = 0;
 168
 169         spin_lock_irq(&bd->lock);
 170
 171         BUG_ON(bd->done_cmds > bd->queued_cmds);
 172
 173         /*
 174          * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
 175          * work to do", even though we return -ENOSPC after this same test
 176          * during bsg_write() -- there, it means our buffer can't have more
 177          * bsg_commands added to it, thus has no space left.
 178          */
 179         if (bd->done_cmds == bd->queued_cmds) {
 180                 ret = -ENODATA;
 181                 goto unlock;
 182         }
 183
 184         if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 185                 ret = -EAGAIN;
 186                 goto unlock;
 187         }
 188
 189         prepare_to_wait(&bd->wq_done, &wait, state);
 190         spin_unlock_irq(&bd->lock);
 191         io_schedule();
 192         finish_wait(&bd->wq_done, &wait);
 193
 194         if ((state == TASK_INTERRUPTIBLE) && signal_pending(current))
 195                 ret = -ERESTARTSYS;
 196
 197         return ret;
 198 unlock:
 199         spin_unlock_irq(&bd->lock);
 200         return ret;
 201 }
 202
 203 static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq,
 204                                 struct sg_io_v4 *hdr, int has_write_perm)
 205 {
 206         memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
 207
 208         if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
 209                            hdr->request_len))
 210                 return -EFAULT;
 211
 212         if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
 213                 if (blk_verify_command(rq->cmd, has_write_perm))
 214                         return -EPERM;
 215         } else if (!capable(CAP_SYS_RAWIO))
 216                 return -EPERM;
 217
 218         /*
 219          * fill in request structure
 220          */
 221         rq->cmd_len = hdr->request_len;
 222         rq->cmd_type = REQ_TYPE_BLOCK_PC;
 223
 224         rq->timeout = (hdr->timeout * HZ) / 1000;
 225         if (!rq->timeout)
 226                 rq->timeout = q->sg_timeout;
 227         if (!rq->timeout)
 228                 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
 229
 230         return 0;
 231 }
 232
 233 /*
 234  * Check if sg_io_v4 from user is allowed and valid
 235  */
 236 static int
 237 bsg_validate_sgv4_hdr(request_queue_t *q, struct sg_io_v4 *hdr, int *rw)
 238 {
 239         int ret = 0;
 240
 241         if (hdr->guard != 'Q')
 242                 return -EINVAL;
 243         if (hdr->request_len > BLK_MAX_CDB)
 244                 return -EINVAL;
 245         if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
 246             hdr->din_xfer_len > (q->max_sectors << 9))
 247                 return -EIO;
 248
 249         switch (hdr->protocol) {
 250         case BSG_PROTOCOL_SCSI:
 251                 switch (hdr->subprotocol) {
 252                 case BSG_SUB_PROTOCOL_SCSI_CMD:
 253                 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
 254                         break;
 255                 default:
 256                         ret = -EINVAL;
 257                 }
 258                 break;
 259         default:
 260                 ret = -EINVAL;
 261         }
 262
 263         *rw = hdr->dout_xfer_len ? WRITE : READ;
 264         return ret;
 265 }
 266
 267 /*
 268  * map sg_io_v4 to a request.
 269  */
 270 static struct request *
 271 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
 272 {
 273         request_queue_t *q = bd->queue;
 274         struct request *rq, *next_rq = NULL;
 275         int ret, rw = 0; /* shut up gcc */
 276         unsigned int dxfer_len;
 277         void *dxferp = NULL;
 278
 279         dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
 280                 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
 281                 hdr->din_xfer_len);
 282
 283         ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
 284         if (ret)
 285                 return ERR_PTR(ret);
 286
 287         /*
 288          * map scatter-gather elements seperately and string them to request
 289          */
 290         rq = blk_get_request(q, rw, GFP_KERNEL);
 291         if (!rq)
 292                 return ERR_PTR(-ENOMEM);
 293         ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
 294                                                        &bd->flags));
 295         if (ret)
 296                 goto out;
 297
 298         if (rw == WRITE && hdr->din_xfer_len) {
 299                 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
 300                         ret = -EOPNOTSUPP;
 301                         goto out;
 302                 }
 303
 304                 next_rq = blk_get_request(q, READ, GFP_KERNEL);
 305                 if (!next_rq) {
 306                         ret = -ENOMEM;
 307                         goto out;
 308                 }
 309                 rq->next_rq = next_rq;
 310
 311                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 312                 ret =  blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
 313                 if (ret)
 314                         goto out;
 315         }
 316
 317         if (hdr->dout_xfer_len) {
 318                 dxfer_len = hdr->dout_xfer_len;
 319                 dxferp = (void*)(unsigned long)hdr->dout_xferp;
 320         } else if (hdr->din_xfer_len) {
 321                 dxfer_len = hdr->din_xfer_len;
 322                 dxferp = (void*)(unsigned long)hdr->din_xferp;
 323         } else
 324                 dxfer_len = 0;
 325
 326         if (dxfer_len) {
 327                 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
 328                 if (ret)
 329                         goto out;
 330         }
 331         return rq;
 332 out:
 333         blk_put_request(rq);
 334         if (next_rq) {
 335                 blk_rq_unmap_user(next_rq->bio);
 336                 blk_put_request(next_rq);
 337         }
 338         return ERR_PTR(ret);
 339 }
 340
 341 /*
 342  * async completion call-back from the block layer, when scsi/ide/whatever
 343  * calls end_that_request_last() on a request
 344  */
 345 static void bsg_rq_end_io(struct request *rq, int uptodate)
 346 {
 347         struct bsg_command *bc = rq->end_io_data;
 348         struct bsg_device *bd = bc->bd;
 349         unsigned long flags;
 350
 351         dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
 352                 bd->name, rq, bc, bc->bio, uptodate);
 353
 354         bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 355
 356         spin_lock_irqsave(&bd->lock, flags);
 357         list_del(&bc->list);
 358         bsg_add_done_cmd(bd, bc);
 359         spin_unlock_irqrestore(&bd->lock, flags);
 360 }
 361
 362 /*
 363  * do final setup of a 'bc' and submit the matching 'rq' to the block
 364  * layer for io
 365  */
 366 static void bsg_add_command(struct bsg_device *bd, request_queue_t *q,
 367                             struct bsg_command *bc, struct request *rq)
 368 {
 369         rq->sense = bc->sense;
 370         rq->sense_len = 0;
 371
 372         /*
 373          * add bc command to busy queue and submit rq for io
 374          */
 375         bc->rq = rq;
 376         bc->bio = rq->bio;
 377         if (rq->next_rq)
 378                 bc->bidi_bio = rq->next_rq->bio;
 379         bc->hdr.duration = jiffies;
 380         spin_lock_irq(&bd->lock);
 381         list_add_tail(&bc->list, &bd->busy_list);
 382         spin_unlock_irq(&bd->lock);
 383
 384         dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 385
 386         rq->end_io_data = bc;
 387         blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
 388 }
 389
 390 static inline struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 391 {
 392         struct bsg_command *bc = NULL;
 393
 394         spin_lock_irq(&bd->lock);
 395         if (bd->done_cmds) {
 396                 bc = list_entry_bc(bd->done_list.next);
 397                 bsg_del_done_cmd(bd, bc);
 398         }
 399         spin_unlock_irq(&bd->lock);
 400
 401         return bc;
 402 }
 403
 404 /*
 405  * Get a finished command from the done list
 406  */
 407 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 408 {
 409         struct bsg_command *bc;
 410         int ret;
 411
 412         do {
 413                 bc = bsg_next_done_cmd(bd);
 414                 if (bc)
 415                         break;
 416
 417                 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
 418                         bc = ERR_PTR(-EAGAIN);
 419                         break;
 420                 }
 421
 422                 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
 423                 if (ret) {
 424                         bc = ERR_PTR(-ERESTARTSYS);
 425                         break;
 426                 }
 427         } while (1);
 428
 429         dprintk("%s: returning done %p\n", bd->name, bc);
 430
 431         return bc;
 432 }
 433
 434 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
 435                                     struct bio *bio, struct bio *bidi_bio)
 436 {
 437         int ret = 0;
 438
 439         dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
 440         /*
 441          * fill in all the output members
 442          */
 443         hdr->device_status = status_byte(rq->errors);
 444         hdr->transport_status = host_byte(rq->errors);
 445         hdr->driver_status = driver_byte(rq->errors);
 446         hdr->info = 0;
 447         if (hdr->device_status || hdr->transport_status || hdr->driver_status)
 448                 hdr->info |= SG_INFO_CHECK;
 449         hdr->din_resid = rq->data_len;
 450         hdr->response_len = 0;
 451
 452         if (rq->sense_len && hdr->response) {
 453                 int len = min((unsigned int) hdr->max_response_len,
 454                               rq->sense_len);
 455
 456                 ret = copy_to_user((void*)(unsigned long)hdr->response,
 457                                    rq->sense, len);
 458                 if (!ret)
 459                         hdr->response_len = len;
 460                 else
 461                         ret = -EFAULT;
 462         }
 463
 464         if (rq->next_rq) {
 465                 blk_rq_unmap_user(bidi_bio);
 466                 blk_put_request(rq->next_rq);
 467         }
 468
 469         blk_rq_unmap_user(bio);
 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         set_bit(BSG_F_BLOCK, &bd->flags);
 483
 484         /*
 485          * wait for all commands to complete
 486          */
 487         ret = 0;
 488         do {
 489                 ret = bsg_io_schedule(bd, TASK_UNINTERRUPTIBLE);
 490                 /*
 491                  * look for -ENODATA specifically -- we'll sometimes get
 492                  * -ERESTARTSYS when we've taken a signal, but we can't
 493                  * return until we're done freeing the queue, so ignore
 494                  * it.  The signal will get handled when we're done freeing
 495                  * the bsg_device.
 496                  */
 497         } while (ret != -ENODATA);
 498
 499         /*
 500          * discard done commands
 501          */
 502         ret = 0;
 503         do {
 504                 spin_lock_irq(&bd->lock);
 505                 if (!bd->queued_cmds) {
 506                         spin_unlock_irq(&bd->lock);
 507                         break;
 508                 }
 509                 spin_unlock_irq(&bd->lock);
 510
 511                 bc = bsg_get_done_cmd(bd);
 512                 if (IS_ERR(bc))
 513                         break;
 514
 515                 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 516                                                 bc->bidi_bio);
 517                 if (!ret)
 518                         ret = tret;
 519
 520                 bsg_free_command(bc);
 521         } while (1);
 522
 523         return ret;
 524 }
 525
 526 static ssize_t
 527 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
 528            const struct iovec *iov, ssize_t *bytes_read)
 529 {
 530         struct bsg_command *bc;
 531         int nr_commands, ret;
 532
 533         if (count % sizeof(struct sg_io_v4))
 534                 return -EINVAL;
 535
 536         ret = 0;
 537         nr_commands = count / sizeof(struct sg_io_v4);
 538         while (nr_commands) {
 539                 bc = bsg_get_done_cmd(bd);
 540                 if (IS_ERR(bc)) {
 541                         ret = PTR_ERR(bc);
 542                         break;
 543                 }
 544
 545                 /*
 546                  * this is the only case where we need to copy data back
 547                  * after completing the request. so do that here,
 548                  * bsg_complete_work() cannot do that for us
 549                  */
 550                 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
 551                                                bc->bidi_bio);
 552
 553                 if (copy_to_user(buf, (char *) &bc->hdr, sizeof(bc->hdr)))
 554                         ret = -EFAULT;
 555
 556                 bsg_free_command(bc);
 557
 558                 if (ret)
 559                         break;
 560
 561                 buf += sizeof(struct sg_io_v4);
 562                 *bytes_read += sizeof(struct sg_io_v4);
 563                 nr_commands--;
 564         }
 565
 566         return ret;
 567 }
 568
 569 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
 570 {
 571         if (file->f_flags & O_NONBLOCK)
 572                 clear_bit(BSG_F_BLOCK, &bd->flags);
 573         else
 574                 set_bit(BSG_F_BLOCK, &bd->flags);
 575 }
 576
 577 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
 578 {
 579         if (file->f_mode & FMODE_WRITE)
 580                 set_bit(BSG_F_WRITE_PERM, &bd->flags);
 581         else
 582                 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
 583 }
 584
 585 static inline int err_block_err(int ret)
 586 {
 587         if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
 588                 return 1;
 589
 590         return 0;
 591 }
 592
 593 static ssize_t
 594 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
 595 {
 596         struct bsg_device *bd = file->private_data;
 597         int ret;
 598         ssize_t bytes_read;
 599
 600         dprintk("%s: read %Zd bytes\n", bd->name, count);
 601
 602         bsg_set_block(bd, file);
 603         bytes_read = 0;
 604         ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
 605         *ppos = bytes_read;
 606
 607         if (!bytes_read || (bytes_read && err_block_err(ret)))
 608                 bytes_read = ret;
 609
 610         return bytes_read;
 611 }
 612
 613 static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf,
 614                            size_t count, ssize_t *bytes_read)
 615 {
 616         struct bsg_command *bc;
 617         struct request *rq;
 618         int ret, nr_commands;
 619
 620         if (count % sizeof(struct sg_io_v4))
 621                 return -EINVAL;
 622
 623         nr_commands = count / sizeof(struct sg_io_v4);
 624         rq = NULL;
 625         bc = NULL;
 626         ret = 0;
 627         while (nr_commands) {
 628                 request_queue_t *q = bd->queue;
 629
 630                 bc = bsg_alloc_command(bd);
 631                 if (IS_ERR(bc)) {
 632                         ret = PTR_ERR(bc);
 633                         bc = NULL;
 634                         break;
 635                 }
 636
 637                 bc->uhdr = (struct sg_io_v4 __user *) buf;
 638                 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
 639                         ret = -EFAULT;
 640                         break;
 641                 }
 642
 643                 /*
 644                  * get a request, fill in the blanks, and add to request queue
 645                  */
 646                 rq = bsg_map_hdr(bd, &bc->hdr);
 647                 if (IS_ERR(rq)) {
 648                         ret = PTR_ERR(rq);
 649                         rq = NULL;
 650                         break;
 651                 }
 652
 653                 bsg_add_command(bd, q, bc, rq);
 654                 bc = NULL;
 655                 rq = NULL;
 656                 nr_commands--;
 657                 buf += sizeof(struct sg_io_v4);
 658                 *bytes_read += sizeof(struct sg_io_v4);
 659         }
 660
 661         if (bc)
 662                 bsg_free_command(bc);
 663
 664         return ret;
 665 }
 666
 667 static ssize_t
 668 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 669 {
 670         struct bsg_device *bd = file->private_data;
 671         ssize_t bytes_read;
 672         int ret;
 673
 674         dprintk("%s: write %Zd bytes\n", bd->name, count);
 675
 676         bsg_set_block(bd, file);
 677         bsg_set_write_perm(bd, file);
 678
 679         bytes_read = 0;
 680         ret = __bsg_write(bd, buf, count, &bytes_read);
 681         *ppos = bytes_read;
 682
 683         /*
 684          * return bytes written on non-fatal errors
 685          */
 686         if (!bytes_read || (bytes_read && err_block_err(ret)))
 687                 bytes_read = ret;
 688
 689         dprintk("%s: returning %Zd\n", bd->name, bytes_read);
 690         return bytes_read;
 691 }
 692
 693 static struct bsg_device *bsg_alloc_device(void)
 694 {
 695         struct bsg_device *bd;
 696
 697         bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
 698         if (unlikely(!bd))
 699                 return NULL;
 700
 701         spin_lock_init(&bd->lock);
 702
 703         bd->max_queue = BSG_DEFAULT_CMDS;
 704
 705         INIT_LIST_HEAD(&bd->busy_list);
 706         INIT_LIST_HEAD(&bd->done_list);
 707         INIT_HLIST_NODE(&bd->dev_list);
 708
 709         init_waitqueue_head(&bd->wq_free);
 710         init_waitqueue_head(&bd->wq_done);
 711         return bd;
 712 }
 713
 714 static int bsg_put_device(struct bsg_device *bd)
 715 {
 716         int ret = 0;
 717
 718         mutex_lock(&bsg_mutex);
 719
 720         if (!atomic_dec_and_test(&bd->ref_count))
 721                 goto out;
 722
 723         dprintk("%s: tearing down\n", bd->name);
 724
 725         /*
 726          * close can always block
 727          */
 728         set_bit(BSG_F_BLOCK, &bd->flags);
 729
 730         /*
 731          * correct error detection baddies here again. it's the responsibility
 732          * of the app to properly reap commands before close() if it wants
 733          * fool-proof error detection
 734          */
 735         ret = bsg_complete_all_commands(bd);
 736
 737         blk_put_queue(bd->queue);
 738         hlist_del(&bd->dev_list);
 739         kfree(bd);
 740 out:
 741         mutex_unlock(&bsg_mutex);
 742         return ret;
 743 }
 744
 745 static struct bsg_device *bsg_add_device(struct inode *inode,
 746                                          struct request_queue *rq,
 747                                          struct file *file)
 748 {
 749         struct bsg_device *bd = NULL;
 750 #ifdef BSG_DEBUG
 751         unsigned char buf[32];
 752 #endif
 753
 754         bd = bsg_alloc_device();
 755         if (!bd)
 756                 return ERR_PTR(-ENOMEM);
 757
 758         bd->queue = rq;
 759         kobject_get(&rq->kobj);
 760         bsg_set_block(bd, file);
 761
 762         atomic_set(&bd->ref_count, 1);
 763         bd->minor = iminor(inode);
 764         mutex_lock(&bsg_mutex);
 765         hlist_add_head(&bd->dev_list, &bsg_device_list[bsg_list_idx(bd->minor)]);
 766
 767         strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1);
 768         dprintk("bound to <%s>, max queue %d\n",
 769                 format_dev_t(buf, inode->i_rdev), bd->max_queue);
 770
 771         mutex_unlock(&bsg_mutex);
 772         return bd;
 773 }
 774
 775 static struct bsg_device *__bsg_get_device(int minor)
 776 {
 777         struct hlist_head *list = &bsg_device_list[bsg_list_idx(minor)];
 778         struct bsg_device *bd = NULL;
 779         struct hlist_node *entry;
 780
 781         mutex_lock(&bsg_mutex);
 782
 783         hlist_for_each(entry, list) {
 784                 bd = hlist_entry(entry, struct bsg_device, dev_list);
 785                 if (bd->minor == minor) {
 786                         atomic_inc(&bd->ref_count);
 787                         break;
 788                 }
 789
 790                 bd = NULL;
 791         }
 792
 793         mutex_unlock(&bsg_mutex);
 794         return bd;
 795 }
 796
 797 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 798 {
 799         struct bsg_device *bd = __bsg_get_device(iminor(inode));
 800         struct bsg_class_device *bcd, *__bcd;
 801
 802         if (bd)
 803                 return bd;
 804
 805         /*
 806          * find the class device
 807          */
 808         bcd = NULL;
 809         mutex_lock(&bsg_mutex);
 810         list_for_each_entry(__bcd, &bsg_class_list, list) {
 811                 if (__bcd->minor == iminor(inode)) {
 812                         bcd = __bcd;
 813                         break;
 814                 }
 815         }
 816         mutex_unlock(&bsg_mutex);
 817
 818         if (!bcd)
 819                 return ERR_PTR(-ENODEV);
 820
 821         return bsg_add_device(inode, bcd->queue, file);
 822 }
 823
 824 static int bsg_open(struct inode *inode, struct file *file)
 825 {
 826         struct bsg_device *bd = bsg_get_device(inode, file);
 827
 828         if (IS_ERR(bd))
 829                 return PTR_ERR(bd);
 830
 831         file->private_data = bd;
 832         return 0;
 833 }
 834
 835 static int bsg_release(struct inode *inode, struct file *file)
 836 {
 837         struct bsg_device *bd = file->private_data;
 838
 839         file->private_data = NULL;
 840         return bsg_put_device(bd);
 841 }
 842
 843 static unsigned int bsg_poll(struct file *file, poll_table *wait)
 844 {
 845         struct bsg_device *bd = file->private_data;
 846         unsigned int mask = 0;
 847
 848         poll_wait(file, &bd->wq_done, wait);
 849         poll_wait(file, &bd->wq_free, wait);
 850
 851         spin_lock_irq(&bd->lock);
 852         if (!list_empty(&bd->done_list))
 853                 mask |= POLLIN | POLLRDNORM;
 854         if (bd->queued_cmds >= bd->max_queue)
 855                 mask |= POLLOUT;
 856         spin_unlock_irq(&bd->lock);
 857
 858         return mask;
 859 }
 860
 861 static int
 862 bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
 863           unsigned long arg)
 864 {
 865         struct bsg_device *bd = file->private_data;
 866         int __user *uarg = (int __user *) arg;
 867
 868         if (!bd)
 869                 return -ENXIO;
 870
 871         switch (cmd) {
 872                 /*
 873                  * our own ioctls
 874                  */
 875         case SG_GET_COMMAND_Q:
 876                 return put_user(bd->max_queue, uarg);
 877         case SG_SET_COMMAND_Q: {
 878                 int queue;
 879
 880                 if (get_user(queue, uarg))
 881                         return -EFAULT;
 882                 if (queue < 1)
 883                         return -EINVAL;
 884
 885                 spin_lock_irq(&bd->lock);
 886                 bd->max_queue = queue;
 887                 spin_unlock_irq(&bd->lock);
 888                 return 0;
 889         }
 890
 891         /*
 892          * SCSI/sg ioctls
 893          */
 894         case SG_GET_VERSION_NUM:
 895         case SCSI_IOCTL_GET_IDLUN:
 896         case SCSI_IOCTL_GET_BUS_NUMBER:
 897         case SG_SET_TIMEOUT:
 898         case SG_GET_TIMEOUT:
 899         case SG_GET_RESERVED_SIZE:
 900         case SG_SET_RESERVED_SIZE:
 901         case SG_EMULATED_HOST:
 902         case SCSI_IOCTL_SEND_COMMAND: {
 903                 void __user *uarg = (void __user *) arg;
 904                 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
 905         }
 906         case SG_IO: {
 907                 struct request *rq;
 908                 struct bio *bio, *bidi_bio = NULL;
 909                 struct sg_io_v4 hdr;
 910
 911                 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
 912                         return -EFAULT;
 913
 914                 rq = bsg_map_hdr(bd, &hdr);
 915                 if (IS_ERR(rq))
 916                         return PTR_ERR(rq);
 917
 918                 bio = rq->bio;
 919                 if (rq->next_rq)
 920                         bidi_bio = rq->next_rq->bio;
 921                 blk_execute_rq(bd->queue, NULL, rq, 0);
 922                 blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 923
 924                 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
 925                         return -EFAULT;
 926
 927                 return 0;
 928         }
 929         /*
 930          * block device ioctls
 931          */
 932         default:
 933 #if 0
 934                 return ioctl_by_bdev(bd->bdev, cmd, arg);
 935 #else
 936                 return -ENOTTY;
 937 #endif
 938         }
 939 }
 940
 941 static struct file_operations bsg_fops = {
 942         .read           =       bsg_read,
 943         .write          =       bsg_write,
 944         .poll           =       bsg_poll,
 945         .open           =       bsg_open,
 946         .release        =       bsg_release,
 947         .ioctl          =       bsg_ioctl,
 948         .owner          =       THIS_MODULE,
 949 };
 950
 951 void bsg_unregister_queue(struct request_queue *q)
 952 {
 953         struct bsg_class_device *bcd = &q->bsg_dev;
 954
 955         if (!bcd->class_dev)
 956                 return;
 957
 958         mutex_lock(&bsg_mutex);
 959         sysfs_remove_link(&q->kobj, "bsg");
 960         class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor));
 961         bcd->class_dev = NULL;
 962         list_del_init(&bcd->list);
 963         bsg_device_nr--;
 964         mutex_unlock(&bsg_mutex);
 965 }
 966 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 967
 968 int bsg_register_queue(struct request_queue *q, const char *name)
 969 {
 970         struct bsg_class_device *bcd, *__bcd;
 971         dev_t dev;
 972         int ret = -EMFILE;
 973         struct class_device *class_dev = NULL;
 974
 975         /*
 976          * we need a proper transport to send commands, not a stacked device
 977          */
 978         if (!q->request_fn)
 979                 return 0;
 980
 981         bcd = &q->bsg_dev;
 982         memset(bcd, 0, sizeof(*bcd));
 983         INIT_LIST_HEAD(&bcd->list);
 984
 985         mutex_lock(&bsg_mutex);
 986         if (bsg_device_nr == BSG_MAX_DEVS) {
 987                 printk(KERN_ERR "bsg: too many bsg devices\n");
 988                 goto err;
 989         }
 990
 991 retry:
 992         list_for_each_entry(__bcd, &bsg_class_list, list) {
 993                 if (__bcd->minor == bsg_minor_idx) {
 994                         bsg_minor_idx++;
 995                         if (bsg_minor_idx == BSG_MAX_DEVS)
 996                                 bsg_minor_idx = 0;
 997                         goto retry;
 998                 }
 999         }
1000
1001         bcd->minor = bsg_minor_idx++;
1002         if (bsg_minor_idx == BSG_MAX_DEVS)
1003                 bsg_minor_idx = 0;
1004
1005         bcd->queue = q;
1006         dev = MKDEV(BSG_MAJOR, bcd->minor);
1007         class_dev = class_device_create(bsg_class, NULL, dev, bcd->dev, "%s", name);
1008         if (IS_ERR(class_dev)) {
1009                 ret = PTR_ERR(class_dev);
1010                 goto err;
1011         }
1012         bcd->class_dev = class_dev;
1013
1014         if (q->kobj.sd) {
1015                 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
1016                 if (ret)
1017                         goto err;
1018         }
1019
1020         list_add_tail(&bcd->list, &bsg_class_list);
1021         bsg_device_nr++;
1022
1023         mutex_unlock(&bsg_mutex);
1024         return 0;
1025 err:
1026         if (class_dev)
1027                 class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor));
1028         mutex_unlock(&bsg_mutex);
1029         return ret;
1030 }
1031 EXPORT_SYMBOL_GPL(bsg_register_queue);
1032
1033 static int bsg_add(struct class_device *cl_dev, struct class_interface *cl_intf)
1034 {
1035         int ret;
1036         struct scsi_device *sdp = to_scsi_device(cl_dev->dev);
1037         struct request_queue *rq = sdp->request_queue;
1038
1039         if (rq->kobj.parent)
1040                 ret = bsg_register_queue(rq, kobject_name(rq->kobj.parent));
1041         else
1042                 ret = bsg_register_queue(rq, kobject_name(&sdp->sdev_gendev.kobj));
1043         return ret;
1044 }
1045
1046 static void bsg_remove(struct class_device *cl_dev, struct class_interface *cl_intf)
1047 {
1048         bsg_unregister_queue(to_scsi_device(cl_dev->dev)->request_queue);
1049 }
1050
1051 static struct class_interface bsg_intf = {
1052         .add    = bsg_add,
1053         .remove = bsg_remove,
1054 };
1055
1056 static struct cdev bsg_cdev = {
1057         .kobj   = {.name = "bsg", },
1058         .owner  = THIS_MODULE,
1059 };
1060
1061 static int __init bsg_init(void)
1062 {
1063         int ret, i;
1064
1065         bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1066                                 sizeof(struct bsg_command), 0, 0, NULL, NULL);
1067         if (!bsg_cmd_cachep) {
1068                 printk(KERN_ERR "bsg: failed creating slab cache\n");
1069                 return -ENOMEM;
1070         }
1071
1072         for (i = 0; i < BSG_LIST_SIZE; i++)
1073                 INIT_HLIST_HEAD(&bsg_device_list[i]);
1074
1075         bsg_class = class_create(THIS_MODULE, "bsg");
1076         if (IS_ERR(bsg_class)) {
1077                 kmem_cache_destroy(bsg_cmd_cachep);
1078                 return PTR_ERR(bsg_class);
1079         }
1080
1081         ret = register_chrdev_region(MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS, "bsg");
1082         if (ret) {
1083                 kmem_cache_destroy(bsg_cmd_cachep);
1084                 class_destroy(bsg_class);
1085                 return ret;
1086         }
1087
1088         cdev_init(&bsg_cdev, &bsg_fops);
1089         ret = cdev_add(&bsg_cdev, MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS);
1090         if (ret) {
1091                 kmem_cache_destroy(bsg_cmd_cachep);
1092                 class_destroy(bsg_class);
1093                 unregister_chrdev_region(MKDEV(BSG_MAJOR, 0), BSG_MAX_DEVS);
1094                 return ret;
1095         }
1096
1097         ret = scsi_register_interface(&bsg_intf);
1098         if (ret) {
1099                 printk(KERN_ERR "bsg: failed register scsi interface %d\n", ret);
1100                 kmem_cache_destroy(bsg_cmd_cachep);
1101                 class_destroy(bsg_class);
1102                 unregister_chrdev(BSG_MAJOR, "bsg");
1103                 return ret;
1104         }
1105
1106         printk(KERN_INFO "%s loaded\n", bsg_version);
1107         return 0;
1108 }
1109
1110 MODULE_AUTHOR("Jens Axboe");
1111 MODULE_DESCRIPTION("Block layer SGSI generic (sg) driver");
1112 MODULE_LICENSE("GPL");
1113
1114 device_initcall(bsg_init);