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 verify_area 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 spinlock_t nbd_lock = SPIN_LOCK_UNLOCKED;
  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         struct nbd_device *lo = req->rq_disk->private_data;
 132         unsigned long flags;
 133
 134         dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
 135                         req, uptodate? "done": "failed");
 136
 137         spin_lock(&lo->queue_lock);
 138         while (req->ref_count > 1) { /* still in send */
 139                 spin_unlock(&lo->queue_lock);
 140                 printk(KERN_DEBUG "%s: request %p still in use (%d), waiting\n",
 141                     lo->disk->disk_name, req, req->ref_count);
 142                 set_current_state(TASK_UNINTERRUPTIBLE);
 143                 schedule_timeout(HZ); /* wait a second */
 144                 spin_lock(&lo->queue_lock);
 145         }
 146         spin_unlock(&lo->queue_lock);
 147
 148         spin_lock_irqsave(q->queue_lock, flags);
 149         if (!end_that_request_first(req, uptodate, req->nr_sectors)) {
 150                 end_that_request_last(req);
 151         }
 152         spin_unlock_irqrestore(q->queue_lock, flags);
 153 }
 154
 155 /*
 156  *  Send or receive packet.
 157  */
 158 static int sock_xmit(struct socket *sock, int send, void *buf, int size,
 159                 int msg_flags)
 160 {
 161         mm_segment_t oldfs;
 162         int result;
 163         struct msghdr msg;
 164         struct iovec iov;
 165         unsigned long flags;
 166         sigset_t oldset;
 167
 168         oldfs = get_fs();
 169         set_fs(get_ds());
 170         /* Allow interception of SIGKILL only
 171          * Don't allow other signals to interrupt the transmission */
 172         spin_lock_irqsave(&current->sighand->siglock, flags);
 173         oldset = current->blocked;
 174         sigfillset(&current->blocked);
 175         sigdelsetmask(&current->blocked, sigmask(SIGKILL));
 176         recalc_sigpending();
 177         spin_unlock_irqrestore(&current->sighand->siglock, flags);
 178
 179         do {
 180                 sock->sk->sk_allocation = GFP_NOIO;
 181                 iov.iov_base = buf;
 182                 iov.iov_len = size;
 183                 msg.msg_name = NULL;
 184                 msg.msg_namelen = 0;
 185                 msg.msg_iov = &iov;
 186                 msg.msg_iovlen = 1;
 187                 msg.msg_control = NULL;
 188                 msg.msg_controllen = 0;
 189                 msg.msg_namelen = 0;
 190                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 191
 192                 if (send)
 193                         result = sock_sendmsg(sock, &msg, size);
 194                 else
 195                         result = sock_recvmsg(sock, &msg, size, 0);
 196
 197                 if (signal_pending(current)) {
 198                         siginfo_t info;
 199                         spin_lock_irqsave(&current->sighand->siglock, flags);
 200                         printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
 201                                 current->pid, current->comm,
 202                                 dequeue_signal(current, &current->blocked, &info));
 203                         spin_unlock_irqrestore(&current->sighand->siglock, flags);
 204                         result = -EINTR;
 205                         break;
 206                 }
 207
 208                 if (result <= 0) {
 209                         if (result == 0)
 210                                 result = -EPIPE; /* short read */
 211                         break;
 212                 }
 213                 size -= result;
 214                 buf += result;
 215         } while (size > 0);
 216
 217         spin_lock_irqsave(&current->sighand->siglock, flags);
 218         current->blocked = oldset;
 219         recalc_sigpending();
 220         spin_unlock_irqrestore(&current->sighand->siglock, flags);
 221
 222         set_fs(oldfs);
 223         return result;
 224 }
 225
 226 static inline int sock_send_bvec(struct socket *sock, struct bio_vec *bvec,
 227                 int flags)
 228 {
 229         int result;
 230         void *kaddr = kmap(bvec->bv_page);
 231         result = sock_xmit(sock, 1, kaddr + bvec->bv_offset, bvec->bv_len,
 232                         flags);
 233         kunmap(bvec->bv_page);
 234         return result;
 235 }
 236
 237 void nbd_send_req(struct nbd_device *lo, struct request *req)
 238 {
 239         int result, i, flags;
 240         struct nbd_request request;
 241         unsigned long size = req->nr_sectors << 9;
 242         struct socket *sock = lo->sock;
 243
 244         request.magic = htonl(NBD_REQUEST_MAGIC);
 245         request.type = htonl(nbd_cmd(req));
 246         request.from = cpu_to_be64((u64) req->sector << 9);
 247         request.len = htonl(size);
 248         memcpy(request.handle, &req, sizeof(req));
 249
 250         down(&lo->tx_lock);
 251
 252         if (!sock || !lo->sock) {
 253                 printk(KERN_ERR "%s: Attempted send on closed socket\n",
 254                                 lo->disk->disk_name);
 255                 goto error_out;
 256         }
 257
 258         dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%luB)\n",
 259                         lo->disk->disk_name, req,
 260                         nbdcmd_to_ascii(nbd_cmd(req)),
 261                         (unsigned long long)req->sector << 9,
 262                         req->nr_sectors << 9);
 263         result = sock_xmit(sock, 1, &request, sizeof(request),
 264                         (nbd_cmd(req) == NBD_CMD_WRITE)? MSG_MORE: 0);
 265         if (result <= 0) {
 266                 printk(KERN_ERR "%s: Send control failed (result %d)\n",
 267                                 lo->disk->disk_name, result);
 268                 goto error_out;
 269         }
 270
 271         if (nbd_cmd(req) == NBD_CMD_WRITE) {
 272                 struct bio *bio;
 273                 /*
 274                  * we are really probing at internals to determine
 275                  * whether to set MSG_MORE or not...
 276                  */
 277                 rq_for_each_bio(bio, req) {
 278                         struct bio_vec *bvec;
 279                         bio_for_each_segment(bvec, bio, i) {
 280                                 flags = 0;
 281                                 if ((i < (bio->bi_vcnt - 1)) || bio->bi_next)
 282                                         flags = MSG_MORE;
 283                                 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
 284                                                 lo->disk->disk_name, req,
 285                                                 bvec->bv_len);
 286                                 result = sock_send_bvec(sock, bvec, flags);
 287                                 if (result <= 0) {
 288                                         printk(KERN_ERR "%s: Send data failed (result %d)\n",
 289                                                         lo->disk->disk_name,
 290                                                         result);
 291                                         goto error_out;
 292                                 }
 293                         }
 294                 }
 295         }
 296         up(&lo->tx_lock);
 297         return;
 298
 299 error_out:
 300         up(&lo->tx_lock);
 301         req->errors++;
 302 }
 303
 304 static struct request *nbd_find_request(struct nbd_device *lo, char *handle)
 305 {
 306         struct request *req;
 307         struct list_head *tmp;
 308         struct request *xreq;
 309
 310         memcpy(&xreq, handle, sizeof(xreq));
 311
 312         spin_lock(&lo->queue_lock);
 313         list_for_each(tmp, &lo->queue_head) {
 314                 req = list_entry(tmp, struct request, queuelist);
 315                 if (req != xreq)
 316                         continue;
 317                 list_del_init(&req->queuelist);
 318                 spin_unlock(&lo->queue_lock);
 319                 return req;
 320         }
 321         spin_unlock(&lo->queue_lock);
 322         return NULL;
 323 }
 324
 325 static inline int sock_recv_bvec(struct socket *sock, struct bio_vec *bvec)
 326 {
 327         int result;
 328         void *kaddr = kmap(bvec->bv_page);
 329         result = sock_xmit(sock, 0, kaddr + bvec->bv_offset, bvec->bv_len,
 330                         MSG_WAITALL);
 331         kunmap(bvec->bv_page);
 332         return result;
 333 }
 334
 335 /* NULL returned = something went wrong, inform userspace */
 336 struct request *nbd_read_stat(struct nbd_device *lo)
 337 {
 338         int result;
 339         struct nbd_reply reply;
 340         struct request *req;
 341         struct socket *sock = lo->sock;
 342
 343         reply.magic = 0;
 344         result = sock_xmit(sock, 0, &reply, sizeof(reply), MSG_WAITALL);
 345         if (result <= 0) {
 346                 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
 347                                 lo->disk->disk_name, result);
 348                 goto harderror;
 349         }
 350         req = nbd_find_request(lo, reply.handle);
 351         if (req == NULL) {
 352                 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
 353                                 lo->disk->disk_name, reply.handle);
 354                 result = -EBADR;
 355                 goto harderror;
 356         }
 357
 358         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 359                 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
 360                                 lo->disk->disk_name,
 361                                 (unsigned long)ntohl(reply.magic));
 362                 result = -EPROTO;
 363                 goto harderror;
 364         }
 365         if (ntohl(reply.error)) {
 366                 printk(KERN_ERR "%s: Other side returned error (%d)\n",
 367                                 lo->disk->disk_name, ntohl(reply.error));
 368                 req->errors++;
 369                 return req;
 370         }
 371
 372         dprintk(DBG_RX, "%s: request %p: got reply\n",
 373                         lo->disk->disk_name, req);
 374         if (nbd_cmd(req) == NBD_CMD_READ) {
 375                 int i;
 376                 struct bio *bio;
 377                 rq_for_each_bio(bio, req) {
 378                         struct bio_vec *bvec;
 379                         bio_for_each_segment(bvec, bio, i) {
 380                                 result = sock_recv_bvec(sock, bvec);
 381                                 if (result <= 0) {
 382                                         printk(KERN_ERR "%s: Receive data failed (result %d)\n",
 383                                                         lo->disk->disk_name,
 384                                                         result);
 385                                         goto harderror;
 386                                 }
 387                                 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
 388                                         lo->disk->disk_name, req, bvec->bv_len);
 389                         }
 390                 }
 391         }
 392         return req;
 393 harderror:
 394         lo->harderror = result;
 395         return NULL;
 396 }
 397
 398 void nbd_do_it(struct nbd_device *lo)
 399 {
 400         struct request *req;
 401
 402         BUG_ON(lo->magic != LO_MAGIC);
 403
 404         while ((req = nbd_read_stat(lo)) != NULL)
 405                 nbd_end_request(req);
 406         return;
 407 }
 408
 409 void nbd_clear_que(struct nbd_device *lo)
 410 {
 411         struct request *req;
 412
 413         BUG_ON(lo->magic != LO_MAGIC);
 414
 415         do {
 416                 req = NULL;
 417                 spin_lock(&lo->queue_lock);
 418                 if (!list_empty(&lo->queue_head)) {
 419                         req = list_entry(lo->queue_head.next, struct request, queuelist);
 420                         list_del_init(&req->queuelist);
 421                 }
 422                 spin_unlock(&lo->queue_lock);
 423                 if (req) {
 424                         req->errors++;
 425                         nbd_end_request(req);
 426                 }
 427         } while (req);
 428 }
 429
 430 /*
 431  * We always wait for result of write, for now. It would be nice to make it optional
 432  * in future
 433  * if ((req->cmd == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
 434  *   { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
 435  */
 436
 437 static void do_nbd_request(request_queue_t * q)
 438 {
 439         struct request *req;
 440
 441         while ((req = elv_next_request(q)) != NULL) {
 442                 struct nbd_device *lo;
 443
 444                 blkdev_dequeue_request(req);
 445                 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%lx)\n",
 446                                 req->rq_disk->disk_name, req, req->flags);
 447
 448                 if (!(req->flags & REQ_CMD))
 449                         goto error_out;
 450
 451                 lo = req->rq_disk->private_data;
 452
 453                 BUG_ON(lo->magic != LO_MAGIC);
 454
 455                 if (!lo->file) {
 456                         printk(KERN_ERR "%s: Request when not-ready\n",
 457                                         lo->disk->disk_name);
 458                         goto error_out;
 459                 }
 460                 nbd_cmd(req) = NBD_CMD_READ;
 461                 if (rq_data_dir(req) == WRITE) {
 462                         nbd_cmd(req) = NBD_CMD_WRITE;
 463                         if (lo->flags & NBD_READ_ONLY) {
 464                                 printk(KERN_ERR "%s: Write on read-only\n",
 465                                                 lo->disk->disk_name);
 466                                 goto error_out;
 467                         }
 468                 }
 469
 470                 req->errors = 0;
 471                 spin_unlock_irq(q->queue_lock);
 472
 473                 spin_lock(&lo->queue_lock);
 474
 475                 if (!lo->file) {
 476                         spin_unlock(&lo->queue_lock);
 477                         printk(KERN_ERR "%s: failed between accept and semaphore, file lost\n",
 478                                         lo->disk->disk_name);
 479                         req->errors++;
 480                         nbd_end_request(req);
 481                         spin_lock_irq(q->queue_lock);
 482                         continue;
 483                 }
 484
 485                 list_add(&req->queuelist, &lo->queue_head);
 486                 req->ref_count++; /* make sure req does not get freed */
 487                 spin_unlock(&lo->queue_lock);
 488
 489                 nbd_send_req(lo, req);
 490
 491                 if (req->errors) {
 492                         printk(KERN_ERR "%s: Request send failed\n",
 493                                         lo->disk->disk_name);
 494                         spin_lock(&lo->queue_lock);
 495                         list_del_init(&req->queuelist);
 496                         req->ref_count--;
 497                         spin_unlock(&lo->queue_lock);
 498                         nbd_end_request(req);
 499                         spin_lock_irq(q->queue_lock);
 500                         continue;
 501                 }
 502
 503                 spin_lock(&lo->queue_lock);
 504                 req->ref_count--;
 505                 spin_unlock(&lo->queue_lock);
 506                 spin_lock_irq(q->queue_lock);
 507                 continue;
 508
 509 error_out:
 510                 req->errors++;
 511                 spin_unlock(q->queue_lock);
 512                 nbd_end_request(req);
 513                 spin_lock(q->queue_lock);
 514         }
 515         return;
 516 }
 517
 518 static int nbd_ioctl(struct inode *inode, struct file *file,
 519                      unsigned int cmd, unsigned long arg)
 520 {
 521         struct nbd_device *lo = inode->i_bdev->bd_disk->private_data;
 522         int error;
 523         struct request sreq ;
 524
 525         if (!capable(CAP_SYS_ADMIN))
 526                 return -EPERM;
 527
 528         BUG_ON(lo->magic != LO_MAGIC);
 529
 530         /* Anyone capable of this syscall can do *real bad* things */
 531         dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
 532                         lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
 533
 534         switch (cmd) {
 535         case NBD_DISCONNECT:
 536                 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
 537                 sreq.flags = REQ_SPECIAL;
 538                 nbd_cmd(&sreq) = NBD_CMD_DISC;
 539                 /*
 540                  * Set these to sane values in case server implementation
 541                  * fails to check the request type first and also to keep
 542                  * debugging output cleaner.
 543                  */
 544                 sreq.sector = 0;
 545                 sreq.nr_sectors = 0;
 546                 if (!lo->sock)
 547                         return -EINVAL;
 548                 nbd_send_req(lo, &sreq);
 549                 return 0;
 550
 551         case NBD_CLEAR_SOCK:
 552                 error = 0;
 553                 down(&lo->tx_lock);
 554                 lo->sock = NULL;
 555                 up(&lo->tx_lock);
 556                 spin_lock(&lo->queue_lock);
 557                 file = lo->file;
 558                 lo->file = NULL;
 559                 spin_unlock(&lo->queue_lock);
 560                 nbd_clear_que(lo);
 561                 spin_lock(&lo->queue_lock);
 562                 if (!list_empty(&lo->queue_head)) {
 563                         printk(KERN_ERR "nbd: disconnect: some requests are in progress -> please try again.\n");
 564                         error = -EBUSY;
 565                 }
 566                 spin_unlock(&lo->queue_lock);
 567                 if (file)
 568                         fput(file);
 569                 return error;
 570         case NBD_SET_SOCK:
 571                 if (lo->file)
 572                         return -EBUSY;
 573                 error = -EINVAL;
 574                 file = fget(arg);
 575                 if (file) {
 576                         inode = file->f_dentry->d_inode;
 577                         if (inode->i_sock) {
 578                                 lo->file = file;
 579                                 lo->sock = SOCKET_I(inode);
 580                                 error = 0;
 581                         } else {
 582                                 fput(file);
 583                         }
 584                 }
 585                 return error;
 586         case NBD_SET_BLKSIZE:
 587                 lo->blksize = arg;
 588                 lo->bytesize &= ~(lo->blksize-1);
 589                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 590                 set_blocksize(inode->i_bdev, lo->blksize);
 591                 set_capacity(lo->disk, lo->bytesize >> 9);
 592                 return 0;
 593         case NBD_SET_SIZE:
 594                 lo->bytesize = arg & ~(lo->blksize-1);
 595                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 596                 set_blocksize(inode->i_bdev, lo->blksize);
 597                 set_capacity(lo->disk, lo->bytesize >> 9);
 598                 return 0;
 599         case NBD_SET_SIZE_BLOCKS:
 600                 lo->bytesize = ((u64) arg) * lo->blksize;
 601                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 602                 set_blocksize(inode->i_bdev, lo->blksize);
 603                 set_capacity(lo->disk, lo->bytesize >> 9);
 604                 return 0;
 605         case NBD_DO_IT:
 606                 if (!lo->file)
 607                         return -EINVAL;
 608                 nbd_do_it(lo);
 609                 /* on return tidy up in case we have a signal */
 610                 /* Forcibly shutdown the socket causing all listeners
 611                  * to error
 612                  *
 613                  * FIXME: This code is duplicated from sys_shutdown, but
 614                  * there should be a more generic interface rather than
 615                  * calling socket ops directly here */
 616                 down(&lo->tx_lock);
 617                 if (lo->sock) {
 618                         printk(KERN_WARNING "%s: shutting down socket\n",
 619                                 lo->disk->disk_name);
 620                         lo->sock->ops->shutdown(lo->sock,
 621                                 SEND_SHUTDOWN|RCV_SHUTDOWN);
 622                         lo->sock = NULL;
 623                 }
 624                 up(&lo->tx_lock);
 625                 spin_lock(&lo->queue_lock);
 626                 file = lo->file;
 627                 lo->file = NULL;
 628                 spin_unlock(&lo->queue_lock);
 629                 nbd_clear_que(lo);
 630                 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
 631                 if (file)
 632                         fput(file);
 633                 return lo->harderror;
 634         case NBD_CLEAR_QUE:
 635                 down(&lo->tx_lock);
 636                 if (lo->sock) {
 637                         up(&lo->tx_lock);
 638                         return 0; /* probably should be error, but that would
 639                                    * break "nbd-client -d", so just return 0 */
 640                 }
 641                 up(&lo->tx_lock);
 642                 nbd_clear_que(lo);
 643                 return 0;
 644         case NBD_PRINT_DEBUG:
 645                 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
 646                         inode->i_bdev->bd_disk->disk_name,
 647                         lo->queue_head.next, lo->queue_head.prev,
 648                         &lo->queue_head);
 649                 return 0;
 650         }
 651         return -EINVAL;
 652 }
 653
 654 static struct block_device_operations nbd_fops =
 655 {
 656         .owner =        THIS_MODULE,
 657         .ioctl =        nbd_ioctl,
 658 };
 659
 660 /*
 661  * And here should be modules and kernel interface
 662  *  (Just smiley confuses emacs :-)
 663  */
 664
 665 static int __init nbd_init(void)
 666 {
 667         int err = -ENOMEM;
 668         int i;
 669
 670         if (sizeof(struct nbd_request) != 28) {
 671                 printk(KERN_CRIT "nbd: sizeof nbd_request needs to be 28 in order to work!\n" );
 672                 return -EIO;
 673         }
 674
 675         for (i = 0; i < MAX_NBD; i++) {
 676                 struct gendisk *disk = alloc_disk(1);
 677                 if (!disk)
 678                         goto out;
 679                 nbd_dev[i].disk = disk;
 680                 /*
 681                  * The new linux 2.5 block layer implementation requires
 682                  * every gendisk to have its very own request_queue struct.
 683                  * These structs are big so we dynamically allocate them.
 684                  */
 685                 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
 686                 if (!disk->queue) {
 687                         put_disk(disk);
 688                         goto out;
 689                 }
 690         }
 691
 692         if (register_blkdev(NBD_MAJOR, "nbd")) {
 693                 err = -EIO;
 694                 goto out;
 695         }
 696
 697         printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
 698         dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
 699
 700         devfs_mk_dir("nbd");
 701         for (i = 0; i < MAX_NBD; i++) {
 702                 struct gendisk *disk = nbd_dev[i].disk;
 703                 nbd_dev[i].file = NULL;
 704                 nbd_dev[i].magic = LO_MAGIC;
 705                 nbd_dev[i].flags = 0;
 706                 spin_lock_init(&nbd_dev[i].queue_lock);
 707                 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
 708                 init_MUTEX(&nbd_dev[i].tx_lock);
 709                 nbd_dev[i].blksize = 1024;
 710                 nbd_dev[i].bytesize = 0x7ffffc00ULL << 10; /* 2TB */
 711                 disk->major = NBD_MAJOR;
 712                 disk->first_minor = i;
 713                 disk->fops = &nbd_fops;
 714                 disk->private_data = &nbd_dev[i];
 715                 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 716                 sprintf(disk->disk_name, "nbd%d", i);
 717                 sprintf(disk->devfs_name, "nbd/%d", i);
 718                 set_capacity(disk, 0x7ffffc00ULL << 1); /* 2 TB */
 719                 add_disk(disk);
 720         }
 721
 722         return 0;
 723 out:
 724         while (i--) {
 725                 blk_cleanup_queue(nbd_dev[i].disk->queue);
 726                 put_disk(nbd_dev[i].disk);
 727         }
 728         return err;
 729 }
 730
 731 static void __exit nbd_cleanup(void)
 732 {
 733         int i;
 734         for (i = 0; i < MAX_NBD; i++) {
 735                 struct gendisk *disk = nbd_dev[i].disk;
 736                 if (disk) {
 737                         del_gendisk(disk);
 738                         blk_cleanup_queue(disk->queue);
 739                         put_disk(disk);
 740                 }
 741         }
 742         devfs_remove("nbd");
 743         unregister_blkdev(NBD_MAJOR, "nbd");
 744         printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
 745 }
 746
 747 module_init(nbd_init);
 748 module_exit(nbd_cleanup);
 749
 750 MODULE_DESCRIPTION("Network Block Device");
 751 MODULE_LICENSE("GPL");
 752
 753 #ifndef NDEBUG
 754 MODULE_PARM(debugflags, "i");
 755 MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
 756 #endif