drivers/block/nbd.c

   1 /*
   2  * Network block device - make block devices work over TCP
   3  *
   4  * Note that you can not swap over this thing, yet. Seems to work but
   5  * deadlocks sometimes - you can not swap over TCP in general.
   6  *
   7  * Copyright 1997-2000 Pavel Machek <pavel@ucw.cz>
   8  * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
   9  *
  10  * (part of code stolen from loop.c)
  11  *
  12  * 97-3-25 compiled 0-th version, not yet tested it
  13  *   (it did not work, BTW) (later that day) HEY! it works!
  14  *   (bit later) hmm, not that much... 2:00am next day:
  15  *   yes, it works, but it gives something like 50kB/sec
  16  * 97-4-01 complete rewrite to make it possible for many requests at
  17  *   once to be processed
  18  * 97-4-11 Making protocol independent of endianity etc.
  19  * 97-9-13 Cosmetic changes
  20  * 98-5-13 Attempt to make 64-bit-clean on 64-bit machines
  21  * 99-1-11 Attempt to make 64-bit-clean on 32-bit machines <ankry@mif.pg.gda.pl>
  22  * 01-2-27 Fix to store proper blockcount for kernel (calculated using
  23  *   BLOCK_SIZE_BITS, not device blocksize) <aga@permonline.ru>
  24  * 01-3-11 Make nbd work with new Linux block layer code. It now supports
  25  *   plugging like all the other block devices. Also added in MSG_MORE to
  26  *   reduce number of partial TCP segments sent. <steve@chygwyn.com>
  27  * 01-12-6 Fix deadlock condition by making queue locks independent of
  28  *   the transmit lock. <steve@chygwyn.com>
  29  * 02-10-11 Allow hung xmit to be aborted via SIGKILL & various fixes.
  30  *   <Paul.Clements@SteelEye.com> <James.Bottomley@SteelEye.com>
  31  * 03-06-22 Make nbd work with new linux 2.5 block layer design. This fixes
  32  *   memory corruption from module removal and possible memory corruption
  33  *   from sending/receiving disk data. <ldl@aros.net>
  34  * 03-06-23 Cosmetic changes. <ldl@aros.net>
  35  * 03-06-23 Enhance diagnostics support. <ldl@aros.net>
  36  * 03-06-24 Remove unneeded blksize_bits field from nbd_device struct.
  37  *   <ldl@aros.net>
  38  * 03-06-24 Cleanup PARANOIA usage & code. <ldl@aros.net>
  39  * 04-02-19 Remove PARANOIA, plus various cleanups (Paul Clements)
  40  * possible FIXME: make set_sock / set_blksize / set_size / do_it one syscall
  41  * why not: would need access_ok and friends, would share yet another
  42  *          structure with userland
  43  */
  44
  45 #include <linux/major.h>
  46
  47 #include <linux/blkdev.h>
  48 #include <linux/module.h>
  49 #include <linux/init.h>
  50 #include <linux/sched.h>
  51 #include <linux/fs.h>
  52 #include <linux/bio.h>
  53 #include <linux/stat.h>
  54 #include <linux/errno.h>
  55 #include <linux/file.h>
  56 #include <linux/ioctl.h>
  57 #include <net/sock.h>
  58
  59 #include <linux/devfs_fs_kernel.h>
  60
  61 #include <asm/uaccess.h>
  62 #include <asm/types.h>
  63
  64 #include <linux/nbd.h>
  65
  66 #define LO_MAGIC 0x68797548
  67
  68 #ifdef NDEBUG
  69 #define dprintk(flags, fmt...)
  70 #else /* NDEBUG */
  71 #define dprintk(flags, fmt...) do { \
  72         if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
  73 } while (0)
  74 #define DBG_IOCTL       0x0004
  75 #define DBG_INIT        0x0010
  76 #define DBG_EXIT        0x0020
  77 #define DBG_BLKDEV      0x0100
  78 #define DBG_RX          0x0200
  79 #define DBG_TX          0x0400
  80 static unsigned int debugflags;
  81 #endif /* NDEBUG */
  82
  83 static struct nbd_device nbd_dev[MAX_NBD];
  84
  85 /*
  86  * Use just one lock (or at most 1 per NIC). Two arguments for this:
  87  * 1. Each NIC is essentially a synchronization point for all servers
  88  *    accessed through that NIC so there's no need to have more locks
  89  *    than NICs anyway.
  90  * 2. More locks lead to more "Dirty cache line bouncing" which will slow
  91  *    down each lock to the point where they're actually slower than just
  92  *    a single lock.
  93  * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
  94  */
  95 static DEFINE_SPINLOCK(nbd_lock);
  96
  97 #ifndef NDEBUG
  98 static const char *ioctl_cmd_to_ascii(int cmd)
  99 {
 100         switch (cmd) {
 101         case NBD_SET_SOCK: return "set-sock";
 102         case NBD_SET_BLKSIZE: return "set-blksize";
 103         case NBD_SET_SIZE: return "set-size";
 104         case NBD_DO_IT: return "do-it";
 105         case NBD_CLEAR_SOCK: return "clear-sock";
 106         case NBD_CLEAR_QUE: return "clear-que";
 107         case NBD_PRINT_DEBUG: return "print-debug";
 108         case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
 109         case NBD_DISCONNECT: return "disconnect";
 110         case BLKROSET: return "set-read-only";
 111         case BLKFLSBUF: return "flush-buffer-cache";
 112         }
 113         return "unknown";
 114 }
 115
 116 static const char *nbdcmd_to_ascii(int cmd)
 117 {
 118         switch (cmd) {
 119         case  NBD_CMD_READ: return "read";
 120         case NBD_CMD_WRITE: return "write";
 121         case  NBD_CMD_DISC: return "disconnect";
 122         }
 123         return "invalid";
 124 }
 125 #endif /* NDEBUG */
 126
 127 static void nbd_end_request(struct request *req)
 128 {
 129         int uptodate = (req->errors == 0) ? 1 : 0;
 130         request_queue_t *q = req->q;
 131         unsigned long flags;
 132
 133         dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
 134                         req, uptodate? "done": "failed");
 135
 136         spin_lock_irqsave(q->queue_lock, flags);
 137         if (!end_that_request_first(req, uptodate, req->nr_sectors)) {
 138                 end_that_request_last(req);
 139         }
 140         spin_unlock_irqrestore(q->queue_lock, flags);
 141 }
 142
 143 /*
 144  *  Send or receive packet.
 145  */
 146 static int sock_xmit(struct socket *sock, int send, void *buf, int size,
 147                 int msg_flags)
 148 {
 149         int result;
 150         struct msghdr msg;
 151         struct kvec iov;
 152         unsigned long flags;
 153         sigset_t oldset;
 154
 155         /* Allow interception of SIGKILL only
 156          * Don't allow other signals to interrupt the transmission */
 157         spin_lock_irqsave(&current->sighand->siglock, flags);
 158         oldset = current->blocked;
 159         sigfillset(&current->blocked);
 160         sigdelsetmask(&current->blocked, sigmask(SIGKILL));
 161         recalc_sigpending();
 162         spin_unlock_irqrestore(&current->sighand->siglock, flags);
 163
 164         do {
 165                 sock->sk->sk_allocation = GFP_NOIO;
 166                 iov.iov_base = buf;
 167                 iov.iov_len = size;
 168                 msg.msg_name = NULL;
 169                 msg.msg_namelen = 0;
 170                 msg.msg_control = NULL;
 171                 msg.msg_controllen = 0;
 172                 msg.msg_namelen = 0;
 173                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 174
 175                 if (send)
 176                         result = kernel_sendmsg(sock, &msg, &iov, 1, size);
 177                 else
 178                         result = kernel_recvmsg(sock, &msg, &iov, 1, size, 0);
 179
 180                 if (signal_pending(current)) {
 181                         siginfo_t info;
 182                         spin_lock_irqsave(&current->sighand->siglock, flags);
 183                         printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
 184                                 current->pid, current->comm,
 185                                 dequeue_signal(current, &current->blocked, &info));
 186                         spin_unlock_irqrestore(&current->sighand->siglock, flags);
 187                         result = -EINTR;
 188                         break;
 189                 }
 190
 191                 if (result <= 0) {
 192                         if (result == 0)
 193                                 result = -EPIPE; /* short read */
 194                         break;
 195                 }
 196                 size -= result;
 197                 buf += result;
 198         } while (size > 0);
 199
 200         spin_lock_irqsave(&current->sighand->siglock, flags);
 201         current->blocked = oldset;
 202         recalc_sigpending();
 203         spin_unlock_irqrestore(&current->sighand->siglock, flags);
 204
 205         return result;
 206 }
 207
 208 static inline int sock_send_bvec(struct socket *sock, struct bio_vec *bvec,
 209                 int flags)
 210 {
 211         int result;
 212         void *kaddr = kmap(bvec->bv_page);
 213         result = sock_xmit(sock, 1, kaddr + bvec->bv_offset, bvec->bv_len,
 214                         flags);
 215         kunmap(bvec->bv_page);
 216         return result;
 217 }
 218
 219 static int nbd_send_req(struct nbd_device *lo, struct request *req)
 220 {
 221         int result, i, flags;
 222         struct nbd_request request;
 223         unsigned long size = req->nr_sectors << 9;
 224         struct socket *sock = lo->sock;
 225
 226         request.magic = htonl(NBD_REQUEST_MAGIC);
 227         request.type = htonl(nbd_cmd(req));
 228         request.from = cpu_to_be64((u64) req->sector << 9);
 229         request.len = htonl(size);
 230         memcpy(request.handle, &req, sizeof(req));
 231
 232         down(&lo->tx_lock);
 233
 234         if (!sock || !lo->sock) {
 235                 printk(KERN_ERR "%s: Attempted send on closed socket\n",
 236                                 lo->disk->disk_name);
 237                 goto error_out;
 238         }
 239
 240         dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%luB)\n",
 241                         lo->disk->disk_name, req,
 242                         nbdcmd_to_ascii(nbd_cmd(req)),
 243                         (unsigned long long)req->sector << 9,
 244                         req->nr_sectors << 9);
 245         result = sock_xmit(sock, 1, &request, sizeof(request),
 246                         (nbd_cmd(req) == NBD_CMD_WRITE)? MSG_MORE: 0);
 247         if (result <= 0) {
 248                 printk(KERN_ERR "%s: Send control failed (result %d)\n",
 249                                 lo->disk->disk_name, result);
 250                 goto error_out;
 251         }
 252
 253         if (nbd_cmd(req) == NBD_CMD_WRITE) {
 254                 struct bio *bio;
 255                 /*
 256                  * we are really probing at internals to determine
 257                  * whether to set MSG_MORE or not...
 258                  */
 259                 rq_for_each_bio(bio, req) {
 260                         struct bio_vec *bvec;
 261                         bio_for_each_segment(bvec, bio, i) {
 262                                 flags = 0;
 263                                 if ((i < (bio->bi_vcnt - 1)) || bio->bi_next)
 264                                         flags = MSG_MORE;
 265                                 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
 266                                                 lo->disk->disk_name, req,
 267                                                 bvec->bv_len);
 268                                 result = sock_send_bvec(sock, bvec, flags);
 269                                 if (result <= 0) {
 270                                         printk(KERN_ERR "%s: Send data failed (result %d)\n",
 271                                                         lo->disk->disk_name,
 272                                                         result);
 273                                         goto error_out;
 274                                 }
 275                         }
 276                 }
 277         }
 278         up(&lo->tx_lock);
 279         return 0;
 280
 281 error_out:
 282         up(&lo->tx_lock);
 283         return 1;
 284 }
 285
 286 static struct request *nbd_find_request(struct nbd_device *lo, char *handle)
 287 {
 288         struct request *req;
 289         struct list_head *tmp;
 290         struct request *xreq;
 291
 292         memcpy(&xreq, handle, sizeof(xreq));
 293
 294         spin_lock(&lo->queue_lock);
 295         list_for_each(tmp, &lo->queue_head) {
 296                 req = list_entry(tmp, struct request, queuelist);
 297                 if (req != xreq)
 298                         continue;
 299                 list_del_init(&req->queuelist);
 300                 spin_unlock(&lo->queue_lock);
 301                 return req;
 302         }
 303         spin_unlock(&lo->queue_lock);
 304         return NULL;
 305 }
 306
 307 static inline int sock_recv_bvec(struct socket *sock, struct bio_vec *bvec)
 308 {
 309         int result;
 310         void *kaddr = kmap(bvec->bv_page);
 311         result = sock_xmit(sock, 0, kaddr + bvec->bv_offset, bvec->bv_len,
 312                         MSG_WAITALL);
 313         kunmap(bvec->bv_page);
 314         return result;
 315 }
 316
 317 /* NULL returned = something went wrong, inform userspace */
 318 static struct request *nbd_read_stat(struct nbd_device *lo)
 319 {
 320         int result;
 321         struct nbd_reply reply;
 322         struct request *req;
 323         struct socket *sock = lo->sock;
 324
 325         reply.magic = 0;
 326         result = sock_xmit(sock, 0, &reply, sizeof(reply), MSG_WAITALL);
 327         if (result <= 0) {
 328                 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
 329                                 lo->disk->disk_name, result);
 330                 goto harderror;
 331         }
 332         req = nbd_find_request(lo, reply.handle);
 333         if (req == NULL) {
 334                 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
 335                                 lo->disk->disk_name, reply.handle);
 336                 result = -EBADR;
 337                 goto harderror;
 338         }
 339
 340         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 341                 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
 342                                 lo->disk->disk_name,
 343                                 (unsigned long)ntohl(reply.magic));
 344                 result = -EPROTO;
 345                 goto harderror;
 346         }
 347         if (ntohl(reply.error)) {
 348                 printk(KERN_ERR "%s: Other side returned error (%d)\n",
 349                                 lo->disk->disk_name, ntohl(reply.error));
 350                 req->errors++;
 351                 return req;
 352         }
 353
 354         dprintk(DBG_RX, "%s: request %p: got reply\n",
 355                         lo->disk->disk_name, req);
 356         if (nbd_cmd(req) == NBD_CMD_READ) {
 357                 int i;
 358                 struct bio *bio;
 359                 rq_for_each_bio(bio, req) {
 360                         struct bio_vec *bvec;
 361                         bio_for_each_segment(bvec, bio, i) {
 362                                 result = sock_recv_bvec(sock, bvec);
 363                                 if (result <= 0) {
 364                                         printk(KERN_ERR "%s: Receive data failed (result %d)\n",
 365                                                         lo->disk->disk_name,
 366                                                         result);
 367                                         goto harderror;
 368                                 }
 369                                 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
 370                                         lo->disk->disk_name, req, bvec->bv_len);
 371                         }
 372                 }
 373         }
 374         return req;
 375 harderror:
 376         lo->harderror = result;
 377         return NULL;
 378 }
 379
 380 static void nbd_do_it(struct nbd_device *lo)
 381 {
 382         struct request *req;
 383
 384         BUG_ON(lo->magic != LO_MAGIC);
 385
 386         while ((req = nbd_read_stat(lo)) != NULL)
 387                 nbd_end_request(req);
 388         return;
 389 }
 390
 391 static void nbd_clear_que(struct nbd_device *lo)
 392 {
 393         struct request *req;
 394
 395         BUG_ON(lo->magic != LO_MAGIC);
 396
 397         do {
 398                 req = NULL;
 399                 spin_lock(&lo->queue_lock);
 400                 if (!list_empty(&lo->queue_head)) {
 401                         req = list_entry(lo->queue_head.next, struct request, queuelist);
 402                         list_del_init(&req->queuelist);
 403                 }
 404                 spin_unlock(&lo->queue_lock);
 405                 if (req) {
 406                         req->errors++;
 407                         nbd_end_request(req);
 408                 }
 409         } while (req);
 410 }
 411
 412 /*
 413  * We always wait for result of write, for now. It would be nice to make it optional
 414  * in future
 415  * if ((req->cmd == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
 416  *   { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
 417  */
 418
 419 static void do_nbd_request(request_queue_t * q)
 420 {
 421         struct request *req;
 422
 423         while ((req = elv_next_request(q)) != NULL) {
 424                 struct nbd_device *lo;
 425
 426                 blkdev_dequeue_request(req);
 427                 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%lx)\n",
 428                                 req->rq_disk->disk_name, req, req->flags);
 429
 430                 if (!(req->flags & REQ_CMD))
 431                         goto error_out;
 432
 433                 lo = req->rq_disk->private_data;
 434
 435                 BUG_ON(lo->magic != LO_MAGIC);
 436
 437                 if (!lo->file) {
 438                         printk(KERN_ERR "%s: Request when not-ready\n",
 439                                         lo->disk->disk_name);
 440                         goto error_out;
 441                 }
 442                 nbd_cmd(req) = NBD_CMD_READ;
 443                 if (rq_data_dir(req) == WRITE) {
 444                         nbd_cmd(req) = NBD_CMD_WRITE;
 445                         if (lo->flags & NBD_READ_ONLY) {
 446                                 printk(KERN_ERR "%s: Write on read-only\n",
 447                                                 lo->disk->disk_name);
 448                                 goto error_out;
 449                         }
 450                 }
 451
 452                 req->errors = 0;
 453                 spin_unlock_irq(q->queue_lock);
 454
 455                 spin_lock(&lo->queue_lock);
 456
 457                 if (!lo->file) {
 458                         spin_unlock(&lo->queue_lock);
 459                         printk(KERN_ERR "%s: failed between accept and semaphore, file lost\n",
 460                                         lo->disk->disk_name);
 461                         req->errors++;
 462                         nbd_end_request(req);
 463                         spin_lock_irq(q->queue_lock);
 464                         continue;
 465                 }
 466
 467                 list_add(&req->queuelist, &lo->queue_head);
 468                 spin_unlock(&lo->queue_lock);
 469
 470                 if (nbd_send_req(lo, req) != 0) {
 471                         printk(KERN_ERR "%s: Request send failed\n",
 472                                         lo->disk->disk_name);
 473                         if (nbd_find_request(lo, (char *)&req) != NULL) {
 474                                 /* we still own req */
 475                                 req->errors++;
 476                                 nbd_end_request(req);
 477                         } else /* we're racing with nbd_clear_que */
 478                                 printk(KERN_DEBUG "nbd: can't find req\n");
 479                 }
 480
 481                 spin_lock_irq(q->queue_lock);
 482                 continue;
 483
 484 error_out:
 485                 req->errors++;
 486                 spin_unlock(q->queue_lock);
 487                 nbd_end_request(req);
 488                 spin_lock(q->queue_lock);
 489         }
 490         return;
 491 }
 492
 493 static int nbd_ioctl(struct inode *inode, struct file *file,
 494                      unsigned int cmd, unsigned long arg)
 495 {
 496         struct nbd_device *lo = inode->i_bdev->bd_disk->private_data;
 497         int error;
 498         struct request sreq ;
 499
 500         if (!capable(CAP_SYS_ADMIN))
 501                 return -EPERM;
 502
 503         BUG_ON(lo->magic != LO_MAGIC);
 504
 505         /* Anyone capable of this syscall can do *real bad* things */
 506         dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
 507                         lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
 508
 509         switch (cmd) {
 510         case NBD_DISCONNECT:
 511                 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
 512                 sreq.flags = REQ_SPECIAL;
 513                 nbd_cmd(&sreq) = NBD_CMD_DISC;
 514                 /*
 515                  * Set these to sane values in case server implementation
 516                  * fails to check the request type first and also to keep
 517                  * debugging output cleaner.
 518                  */
 519                 sreq.sector = 0;
 520                 sreq.nr_sectors = 0;
 521                 if (!lo->sock)
 522                         return -EINVAL;
 523                 nbd_send_req(lo, &sreq);
 524                 return 0;
 525
 526         case NBD_CLEAR_SOCK:
 527                 error = 0;
 528                 down(&lo->tx_lock);
 529                 lo->sock = NULL;
 530                 up(&lo->tx_lock);
 531                 spin_lock(&lo->queue_lock);
 532                 file = lo->file;
 533                 lo->file = NULL;
 534                 spin_unlock(&lo->queue_lock);
 535                 nbd_clear_que(lo);
 536                 spin_lock(&lo->queue_lock);
 537                 if (!list_empty(&lo->queue_head)) {
 538                         printk(KERN_ERR "nbd: disconnect: some requests are in progress -> please try again.\n");
 539                         error = -EBUSY;
 540                 }
 541                 spin_unlock(&lo->queue_lock);
 542                 if (file)
 543                         fput(file);
 544                 return error;
 545         case NBD_SET_SOCK:
 546                 if (lo->file)
 547                         return -EBUSY;
 548                 error = -EINVAL;
 549                 file = fget(arg);
 550                 if (file) {
 551                         inode = file->f_dentry->d_inode;
 552                         if (S_ISSOCK(inode->i_mode)) {
 553                                 lo->file = file;
 554                                 lo->sock = SOCKET_I(inode);
 555                                 error = 0;
 556                         } else {
 557                                 fput(file);
 558                         }
 559                 }
 560                 return error;
 561         case NBD_SET_BLKSIZE:
 562                 lo->blksize = arg;
 563                 lo->bytesize &= ~(lo->blksize-1);
 564                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 565                 set_blocksize(inode->i_bdev, lo->blksize);
 566                 set_capacity(lo->disk, lo->bytesize >> 9);
 567                 return 0;
 568         case NBD_SET_SIZE:
 569                 lo->bytesize = arg & ~(lo->blksize-1);
 570                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 571                 set_blocksize(inode->i_bdev, lo->blksize);
 572                 set_capacity(lo->disk, lo->bytesize >> 9);
 573                 return 0;
 574         case NBD_SET_SIZE_BLOCKS:
 575                 lo->bytesize = ((u64) arg) * lo->blksize;
 576                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 577                 set_blocksize(inode->i_bdev, lo->blksize);
 578                 set_capacity(lo->disk, lo->bytesize >> 9);
 579                 return 0;
 580         case NBD_DO_IT:
 581                 if (!lo->file)
 582                         return -EINVAL;
 583                 nbd_do_it(lo);
 584                 /* on return tidy up in case we have a signal */
 585                 /* Forcibly shutdown the socket causing all listeners
 586                  * to error
 587                  *
 588                  * FIXME: This code is duplicated from sys_shutdown, but
 589                  * there should be a more generic interface rather than
 590                  * calling socket ops directly here */
 591                 down(&lo->tx_lock);
 592                 if (lo->sock) {
 593                         printk(KERN_WARNING "%s: shutting down socket\n",
 594                                 lo->disk->disk_name);
 595                         lo->sock->ops->shutdown(lo->sock,
 596                                 SEND_SHUTDOWN|RCV_SHUTDOWN);
 597                         lo->sock = NULL;
 598                 }
 599                 up(&lo->tx_lock);
 600                 spin_lock(&lo->queue_lock);
 601                 file = lo->file;
 602                 lo->file = NULL;
 603                 spin_unlock(&lo->queue_lock);
 604                 nbd_clear_que(lo);
 605                 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
 606                 if (file)
 607                         fput(file);
 608                 return lo->harderror;
 609         case NBD_CLEAR_QUE:
 610                 down(&lo->tx_lock);
 611                 if (lo->sock) {
 612                         up(&lo->tx_lock);
 613                         return 0; /* probably should be error, but that would
 614                                    * break "nbd-client -d", so just return 0 */
 615                 }
 616                 up(&lo->tx_lock);
 617                 nbd_clear_que(lo);
 618                 return 0;
 619         case NBD_PRINT_DEBUG:
 620                 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
 621                         inode->i_bdev->bd_disk->disk_name,
 622                         lo->queue_head.next, lo->queue_head.prev,
 623                         &lo->queue_head);
 624                 return 0;
 625         }
 626         return -EINVAL;
 627 }
 628
 629 static struct block_device_operations nbd_fops =
 630 {
 631         .owner =        THIS_MODULE,
 632         .ioctl =        nbd_ioctl,
 633 };
 634
 635 /*
 636  * And here should be modules and kernel interface
 637  *  (Just smiley confuses emacs :-)
 638  */
 639
 640 static int __init nbd_init(void)
 641 {
 642         int err = -ENOMEM;
 643         int i;
 644
 645         if (sizeof(struct nbd_request) != 28) {
 646                 printk(KERN_CRIT "nbd: sizeof nbd_request needs to be 28 in order to work!\n" );
 647                 return -EIO;
 648         }
 649
 650         for (i = 0; i < MAX_NBD; i++) {
 651                 struct gendisk *disk = alloc_disk(1);
 652                 if (!disk)
 653                         goto out;
 654                 nbd_dev[i].disk = disk;
 655                 /*
 656                  * The new linux 2.5 block layer implementation requires
 657                  * every gendisk to have its very own request_queue struct.
 658                  * These structs are big so we dynamically allocate them.
 659                  */
 660                 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
 661                 if (!disk->queue) {
 662                         put_disk(disk);
 663                         goto out;
 664                 }
 665         }
 666
 667         if (register_blkdev(NBD_MAJOR, "nbd")) {
 668                 err = -EIO;
 669                 goto out;
 670         }
 671
 672         printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
 673         dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
 674
 675         devfs_mk_dir("nbd");
 676         for (i = 0; i < MAX_NBD; i++) {
 677                 struct gendisk *disk = nbd_dev[i].disk;
 678                 nbd_dev[i].file = NULL;
 679                 nbd_dev[i].magic = LO_MAGIC;
 680                 nbd_dev[i].flags = 0;
 681                 spin_lock_init(&nbd_dev[i].queue_lock);
 682                 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
 683                 init_MUTEX(&nbd_dev[i].tx_lock);
 684                 nbd_dev[i].blksize = 1024;
 685                 nbd_dev[i].bytesize = 0x7ffffc00ULL << 10; /* 2TB */
 686                 disk->major = NBD_MAJOR;
 687                 disk->first_minor = i;
 688                 disk->fops = &nbd_fops;
 689                 disk->private_data = &nbd_dev[i];
 690                 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 691                 sprintf(disk->disk_name, "nbd%d", i);
 692                 sprintf(disk->devfs_name, "nbd/%d", i);
 693                 set_capacity(disk, 0x7ffffc00ULL << 1); /* 2 TB */
 694                 add_disk(disk);
 695         }
 696
 697         return 0;
 698 out:
 699         while (i--) {
 700                 blk_cleanup_queue(nbd_dev[i].disk->queue);
 701                 put_disk(nbd_dev[i].disk);
 702         }
 703         return err;
 704 }
 705
 706 static void __exit nbd_cleanup(void)
 707 {
 708         int i;
 709         for (i = 0; i < MAX_NBD; i++) {
 710                 struct gendisk *disk = nbd_dev[i].disk;
 711                 if (disk) {
 712                         del_gendisk(disk);
 713                         blk_cleanup_queue(disk->queue);
 714                         put_disk(disk);
 715                 }
 716         }
 717         devfs_remove("nbd");
 718         unregister_blkdev(NBD_MAJOR, "nbd");
 719         printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
 720 }
 721
 722 module_init(nbd_init);
 723 module_exit(nbd_cleanup);
 724
 725 MODULE_DESCRIPTION("Network Block Device");
 726 MODULE_LICENSE("GPL");
 727
 728 #ifndef NDEBUG
 729 module_param(debugflags, int, 0644);
 730 MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
 731 #endif