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  * This file is released under GPLv2 or later.
  11  *
  12  * (part of code stolen from loop.c)
  13  */
  14
  15 #include <linux/major.h>
  16
  17 #include <linux/blkdev.h>
  18 #include <linux/module.h>
  19 #include <linux/init.h>
  20 #include <linux/sched.h>
  21 #include <linux/fs.h>
  22 #include <linux/bio.h>
  23 #include <linux/stat.h>
  24 #include <linux/errno.h>
  25 #include <linux/file.h>
  26 #include <linux/ioctl.h>
  27 #include <linux/compiler.h>
  28 #include <linux/err.h>
  29 #include <linux/kernel.h>
  30 #include <net/sock.h>
  31 #include <linux/net.h>
  32
  33 #include <asm/uaccess.h>
  34 #include <asm/system.h>
  35 #include <asm/types.h>
  36
  37 #include <linux/nbd.h>
  38
  39 #define LO_MAGIC 0x68797548
  40
  41 #ifdef NDEBUG
  42 #define dprintk(flags, fmt...)
  43 #else /* NDEBUG */
  44 #define dprintk(flags, fmt...) do { \
  45         if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
  46 } while (0)
  47 #define DBG_IOCTL       0x0004
  48 #define DBG_INIT        0x0010
  49 #define DBG_EXIT        0x0020
  50 #define DBG_BLKDEV      0x0100
  51 #define DBG_RX          0x0200
  52 #define DBG_TX          0x0400
  53 static unsigned int debugflags;
  54 #endif /* NDEBUG */
  55
  56 static unsigned int nbds_max = 16;
  57 static struct nbd_device *nbd_dev;
  58
  59 /*
  60  * Use just one lock (or at most 1 per NIC). Two arguments for this:
  61  * 1. Each NIC is essentially a synchronization point for all servers
  62  *    accessed through that NIC so there's no need to have more locks
  63  *    than NICs anyway.
  64  * 2. More locks lead to more "Dirty cache line bouncing" which will slow
  65  *    down each lock to the point where they're actually slower than just
  66  *    a single lock.
  67  * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
  68  */
  69 static DEFINE_SPINLOCK(nbd_lock);
  70
  71 #ifndef NDEBUG
  72 static const char *ioctl_cmd_to_ascii(int cmd)
  73 {
  74         switch (cmd) {
  75         case NBD_SET_SOCK: return "set-sock";
  76         case NBD_SET_BLKSIZE: return "set-blksize";
  77         case NBD_SET_SIZE: return "set-size";
  78         case NBD_DO_IT: return "do-it";
  79         case NBD_CLEAR_SOCK: return "clear-sock";
  80         case NBD_CLEAR_QUE: return "clear-que";
  81         case NBD_PRINT_DEBUG: return "print-debug";
  82         case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
  83         case NBD_DISCONNECT: return "disconnect";
  84         case BLKROSET: return "set-read-only";
  85         case BLKFLSBUF: return "flush-buffer-cache";
  86         }
  87         return "unknown";
  88 }
  89
  90 static const char *nbdcmd_to_ascii(int cmd)
  91 {
  92         switch (cmd) {
  93         case  NBD_CMD_READ: return "read";
  94         case NBD_CMD_WRITE: return "write";
  95         case  NBD_CMD_DISC: return "disconnect";
  96         }
  97         return "invalid";
  98 }
  99 #endif /* NDEBUG */
 100
 101 static void nbd_end_request(struct request *req)
 102 {
 103         int error = req->errors ? -EIO : 0;
 104         struct request_queue *q = req->q;
 105         unsigned long flags;
 106
 107         dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
 108                         req, error ? "failed" : "done");
 109
 110         spin_lock_irqsave(q->queue_lock, flags);
 111         __blk_end_request(req, error, req->nr_sectors << 9);
 112         spin_unlock_irqrestore(q->queue_lock, flags);
 113 }
 114
 115 static void sock_shutdown(struct nbd_device *lo, int lock)
 116 {
 117         /* Forcibly shutdown the socket causing all listeners
 118          * to error
 119          *
 120          * FIXME: This code is duplicated from sys_shutdown, but
 121          * there should be a more generic interface rather than
 122          * calling socket ops directly here */
 123         if (lock)
 124                 mutex_lock(&lo->tx_lock);
 125         if (lo->sock) {
 126                 printk(KERN_WARNING "%s: shutting down socket\n",
 127                         lo->disk->disk_name);
 128                 kernel_sock_shutdown(lo->sock, SHUT_RDWR);
 129                 lo->sock = NULL;
 130         }
 131         if (lock)
 132                 mutex_unlock(&lo->tx_lock);
 133 }
 134
 135 static void nbd_xmit_timeout(unsigned long arg)
 136 {
 137         struct task_struct *task = (struct task_struct *)arg;
 138
 139         printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
 140                 task->comm, task->pid);
 141         force_sig(SIGKILL, task);
 142 }
 143
 144 /*
 145  *  Send or receive packet.
 146  */
 147 static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size,
 148                 int msg_flags)
 149 {
 150         struct socket *sock = lo->sock;
 151         int result;
 152         struct msghdr msg;
 153         struct kvec iov;
 154         sigset_t blocked, oldset;
 155
 156         if (unlikely(!sock)) {
 157                 printk(KERN_ERR "%s: Attempted %s on closed socket in sock_xmit\n",
 158                        lo->disk->disk_name, (send ? "send" : "recv"));
 159                 return -EINVAL;
 160         }
 161
 162         /* Allow interception of SIGKILL only
 163          * Don't allow other signals to interrupt the transmission */
 164         siginitsetinv(&blocked, sigmask(SIGKILL));
 165         sigprocmask(SIG_SETMASK, &blocked, &oldset);
 166
 167         do {
 168                 sock->sk->sk_allocation = GFP_NOIO;
 169                 iov.iov_base = buf;
 170                 iov.iov_len = size;
 171                 msg.msg_name = NULL;
 172                 msg.msg_namelen = 0;
 173                 msg.msg_control = NULL;
 174                 msg.msg_controllen = 0;
 175                 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
 176
 177                 if (send) {
 178                         struct timer_list ti;
 179
 180                         if (lo->xmit_timeout) {
 181                                 init_timer(&ti);
 182                                 ti.function = nbd_xmit_timeout;
 183                                 ti.data = (unsigned long)current;
 184                                 ti.expires = jiffies + lo->xmit_timeout;
 185                                 add_timer(&ti);
 186                         }
 187                         result = kernel_sendmsg(sock, &msg, &iov, 1, size);
 188                         if (lo->xmit_timeout)
 189                                 del_timer_sync(&ti);
 190                 } else
 191                         result = kernel_recvmsg(sock, &msg, &iov, 1, size, 0);
 192
 193                 if (signal_pending(current)) {
 194                         siginfo_t info;
 195                         printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
 196                                 task_pid_nr(current), current->comm,
 197                                 dequeue_signal_lock(current, &current->blocked, &info));
 198                         result = -EINTR;
 199                         sock_shutdown(lo, !send);
 200                         break;
 201                 }
 202
 203                 if (result <= 0) {
 204                         if (result == 0)
 205                                 result = -EPIPE; /* short read */
 206                         break;
 207                 }
 208                 size -= result;
 209                 buf += result;
 210         } while (size > 0);
 211
 212         sigprocmask(SIG_SETMASK, &oldset, NULL);
 213
 214         return result;
 215 }
 216
 217 static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec,
 218                 int flags)
 219 {
 220         int result;
 221         void *kaddr = kmap(bvec->bv_page);
 222         result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, bvec->bv_len, flags);
 223         kunmap(bvec->bv_page);
 224         return result;
 225 }
 226
 227 /* always call with the tx_lock held */
 228 static int nbd_send_req(struct nbd_device *lo, struct request *req)
 229 {
 230         int result, flags;
 231         struct nbd_request request;
 232         unsigned long size = req->nr_sectors << 9;
 233
 234         request.magic = htonl(NBD_REQUEST_MAGIC);
 235         request.type = htonl(nbd_cmd(req));
 236         request.from = cpu_to_be64((u64) req->sector << 9);
 237         request.len = htonl(size);
 238         memcpy(request.handle, &req, sizeof(req));
 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(lo, 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 req_iterator iter;
 255                 struct bio_vec *bvec;
 256                 /*
 257                  * we are really probing at internals to determine
 258                  * whether to set MSG_MORE or not...
 259                  */
 260                 rq_for_each_segment(bvec, req, iter) {
 261                         flags = 0;
 262                         if (!rq_iter_last(req, iter))
 263                                 flags = MSG_MORE;
 264                         dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
 265                                         lo->disk->disk_name, req, bvec->bv_len);
 266                         result = sock_send_bvec(lo, bvec, flags);
 267                         if (result <= 0) {
 268                                 printk(KERN_ERR "%s: Send data failed (result %d)\n",
 269                                                 lo->disk->disk_name, result);
 270                                 goto error_out;
 271                         }
 272                 }
 273         }
 274         return 0;
 275
 276 error_out:
 277         return 1;
 278 }
 279
 280 static struct request *nbd_find_request(struct nbd_device *lo,
 281                                         struct request *xreq)
 282 {
 283         struct request *req, *tmp;
 284         int err;
 285
 286         err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
 287         if (unlikely(err))
 288                 goto out;
 289
 290         spin_lock(&lo->queue_lock);
 291         list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) {
 292                 if (req != xreq)
 293                         continue;
 294                 list_del_init(&req->queuelist);
 295                 spin_unlock(&lo->queue_lock);
 296                 return req;
 297         }
 298         spin_unlock(&lo->queue_lock);
 299
 300         err = -ENOENT;
 301
 302 out:
 303         return ERR_PTR(err);
 304 }
 305
 306 static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec)
 307 {
 308         int result;
 309         void *kaddr = kmap(bvec->bv_page);
 310         result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len,
 311                         MSG_WAITALL);
 312         kunmap(bvec->bv_page);
 313         return result;
 314 }
 315
 316 /* NULL returned = something went wrong, inform userspace */
 317 static struct request *nbd_read_stat(struct nbd_device *lo)
 318 {
 319         int result;
 320         struct nbd_reply reply;
 321         struct request *req;
 322
 323         reply.magic = 0;
 324         result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL);
 325         if (result <= 0) {
 326                 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
 327                                 lo->disk->disk_name, result);
 328                 goto harderror;
 329         }
 330
 331         if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
 332                 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
 333                                 lo->disk->disk_name,
 334                                 (unsigned long)ntohl(reply.magic));
 335                 result = -EPROTO;
 336                 goto harderror;
 337         }
 338
 339         req = nbd_find_request(lo, *(struct request **)reply.handle);
 340         if (unlikely(IS_ERR(req))) {
 341                 result = PTR_ERR(req);
 342                 if (result != -ENOENT)
 343                         goto harderror;
 344
 345                 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
 346                                 lo->disk->disk_name, reply.handle);
 347                 result = -EBADR;
 348                 goto harderror;
 349         }
 350
 351         if (ntohl(reply.error)) {
 352                 printk(KERN_ERR "%s: Other side returned error (%d)\n",
 353                                 lo->disk->disk_name, ntohl(reply.error));
 354                 req->errors++;
 355                 return req;
 356         }
 357
 358         dprintk(DBG_RX, "%s: request %p: got reply\n",
 359                         lo->disk->disk_name, req);
 360         if (nbd_cmd(req) == NBD_CMD_READ) {
 361                 struct req_iterator iter;
 362                 struct bio_vec *bvec;
 363
 364                 rq_for_each_segment(bvec, req, iter) {
 365                         result = sock_recv_bvec(lo, bvec);
 366                         if (result <= 0) {
 367                                 printk(KERN_ERR "%s: Receive data failed (result %d)\n",
 368                                                 lo->disk->disk_name, result);
 369                                 req->errors++;
 370                                 return req;
 371                         }
 372                         dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
 373                                 lo->disk->disk_name, req, bvec->bv_len);
 374                 }
 375         }
 376         return req;
 377 harderror:
 378         lo->harderror = result;
 379         return NULL;
 380 }
 381
 382 static ssize_t pid_show(struct device *dev,
 383                         struct device_attribute *attr, char *buf)
 384 {
 385         struct gendisk *disk = dev_to_disk(dev);
 386
 387         return sprintf(buf, "%ld\n",
 388                 (long) ((struct nbd_device *)disk->private_data)->pid);
 389 }
 390
 391 static struct device_attribute pid_attr = {
 392         .attr = { .name = "pid", .mode = S_IRUGO, .owner = THIS_MODULE },
 393         .show = pid_show,
 394 };
 395
 396 static int nbd_do_it(struct nbd_device *lo)
 397 {
 398         struct request *req;
 399         int ret;
 400
 401         BUG_ON(lo->magic != LO_MAGIC);
 402
 403         lo->pid = current->pid;
 404         ret = sysfs_create_file(&lo->disk->dev.kobj, &pid_attr.attr);
 405         if (ret) {
 406                 printk(KERN_ERR "nbd: sysfs_create_file failed!");
 407                 return ret;
 408         }
 409
 410         while ((req = nbd_read_stat(lo)) != NULL)
 411                 nbd_end_request(req);
 412
 413         sysfs_remove_file(&lo->disk->dev.kobj, &pid_attr.attr);
 414         return 0;
 415 }
 416
 417 static void nbd_clear_que(struct nbd_device *lo)
 418 {
 419         struct request *req;
 420
 421         BUG_ON(lo->magic != LO_MAGIC);
 422
 423         /*
 424          * Because we have set lo->sock to NULL under the tx_lock, all
 425          * modifications to the list must have completed by now.  For
 426          * the same reason, the active_req must be NULL.
 427          *
 428          * As a consequence, we don't need to take the spin lock while
 429          * purging the list here.
 430          */
 431         BUG_ON(lo->sock);
 432         BUG_ON(lo->active_req);
 433
 434         while (!list_empty(&lo->queue_head)) {
 435                 req = list_entry(lo->queue_head.next, struct request,
 436                                  queuelist);
 437                 list_del_init(&req->queuelist);
 438                 req->errors++;
 439                 nbd_end_request(req);
 440         }
 441 }
 442
 443
 444 /*
 445  * We always wait for result of write, for now. It would be nice to make it optional
 446  * in future
 447  * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
 448  *   { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
 449  */
 450
 451 static void do_nbd_request(struct request_queue * q)
 452 {
 453         struct request *req;
 454
 455         while ((req = elv_next_request(q)) != NULL) {
 456                 struct nbd_device *lo;
 457
 458                 blkdev_dequeue_request(req);
 459                 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
 460                                 req->rq_disk->disk_name, req, req->cmd_type);
 461
 462                 if (!blk_fs_request(req))
 463                         goto error_out;
 464
 465                 lo = req->rq_disk->private_data;
 466
 467                 BUG_ON(lo->magic != LO_MAGIC);
 468
 469                 nbd_cmd(req) = NBD_CMD_READ;
 470                 if (rq_data_dir(req) == WRITE) {
 471                         nbd_cmd(req) = NBD_CMD_WRITE;
 472                         if (lo->flags & NBD_READ_ONLY) {
 473                                 printk(KERN_ERR "%s: Write on read-only\n",
 474                                                 lo->disk->disk_name);
 475                                 goto error_out;
 476                         }
 477                 }
 478
 479                 req->errors = 0;
 480                 spin_unlock_irq(q->queue_lock);
 481
 482                 mutex_lock(&lo->tx_lock);
 483                 if (unlikely(!lo->sock)) {
 484                         mutex_unlock(&lo->tx_lock);
 485                         printk(KERN_ERR "%s: Attempted send on closed socket\n",
 486                                lo->disk->disk_name);
 487                         req->errors++;
 488                         nbd_end_request(req);
 489                         spin_lock_irq(q->queue_lock);
 490                         continue;
 491                 }
 492
 493                 lo->active_req = req;
 494
 495                 if (nbd_send_req(lo, req) != 0) {
 496                         printk(KERN_ERR "%s: Request send failed\n",
 497                                         lo->disk->disk_name);
 498                         req->errors++;
 499                         nbd_end_request(req);
 500                 } else {
 501                         spin_lock(&lo->queue_lock);
 502                         list_add(&req->queuelist, &lo->queue_head);
 503                         spin_unlock(&lo->queue_lock);
 504                 }
 505
 506                 lo->active_req = NULL;
 507                 mutex_unlock(&lo->tx_lock);
 508                 wake_up_all(&lo->active_wq);
 509
 510                 spin_lock_irq(q->queue_lock);
 511                 continue;
 512
 513 error_out:
 514                 req->errors++;
 515                 spin_unlock(q->queue_lock);
 516                 nbd_end_request(req);
 517                 spin_lock(q->queue_lock);
 518         }
 519 }
 520
 521 static int nbd_ioctl(struct inode *inode, struct file *file,
 522                      unsigned int cmd, unsigned long arg)
 523 {
 524         struct nbd_device *lo = inode->i_bdev->bd_disk->private_data;
 525         int error;
 526         struct request sreq ;
 527
 528         if (!capable(CAP_SYS_ADMIN))
 529                 return -EPERM;
 530
 531         BUG_ON(lo->magic != LO_MAGIC);
 532
 533         /* Anyone capable of this syscall can do *real bad* things */
 534         dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
 535                         lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
 536
 537         switch (cmd) {
 538         case NBD_DISCONNECT:
 539                 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
 540                 sreq.cmd_type = REQ_TYPE_SPECIAL;
 541                 nbd_cmd(&sreq) = NBD_CMD_DISC;
 542                 /*
 543                  * Set these to sane values in case server implementation
 544                  * fails to check the request type first and also to keep
 545                  * debugging output cleaner.
 546                  */
 547                 sreq.sector = 0;
 548                 sreq.nr_sectors = 0;
 549                 if (!lo->sock)
 550                         return -EINVAL;
 551                 mutex_lock(&lo->tx_lock);
 552                 nbd_send_req(lo, &sreq);
 553                 mutex_unlock(&lo->tx_lock);
 554                 return 0;
 555
 556         case NBD_CLEAR_SOCK:
 557                 error = 0;
 558                 mutex_lock(&lo->tx_lock);
 559                 lo->sock = NULL;
 560                 mutex_unlock(&lo->tx_lock);
 561                 file = lo->file;
 562                 lo->file = NULL;
 563                 nbd_clear_que(lo);
 564                 BUG_ON(!list_empty(&lo->queue_head));
 565                 if (file)
 566                         fput(file);
 567                 return error;
 568         case NBD_SET_SOCK:
 569                 if (lo->file)
 570                         return -EBUSY;
 571                 error = -EINVAL;
 572                 file = fget(arg);
 573                 if (file) {
 574                         inode = file->f_path.dentry->d_inode;
 575                         if (S_ISSOCK(inode->i_mode)) {
 576                                 lo->file = file;
 577                                 lo->sock = SOCKET_I(inode);
 578                                 error = 0;
 579                         } else {
 580                                 fput(file);
 581                         }
 582                 }
 583                 return error;
 584         case NBD_SET_BLKSIZE:
 585                 lo->blksize = arg;
 586                 lo->bytesize &= ~(lo->blksize-1);
 587                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 588                 set_blocksize(inode->i_bdev, lo->blksize);
 589                 set_capacity(lo->disk, lo->bytesize >> 9);
 590                 return 0;
 591         case NBD_SET_SIZE:
 592                 lo->bytesize = arg & ~(lo->blksize-1);
 593                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 594                 set_blocksize(inode->i_bdev, lo->blksize);
 595                 set_capacity(lo->disk, lo->bytesize >> 9);
 596                 return 0;
 597         case NBD_SET_TIMEOUT:
 598                 lo->xmit_timeout = arg * HZ;
 599                 return 0;
 600         case NBD_SET_SIZE_BLOCKS:
 601                 lo->bytesize = ((u64) arg) * lo->blksize;
 602                 inode->i_bdev->bd_inode->i_size = lo->bytesize;
 603                 set_blocksize(inode->i_bdev, lo->blksize);
 604                 set_capacity(lo->disk, lo->bytesize >> 9);
 605                 return 0;
 606         case NBD_DO_IT:
 607                 if (!lo->file)
 608                         return -EINVAL;
 609                 error = nbd_do_it(lo);
 610                 if (error)
 611                         return error;
 612                 sock_shutdown(lo, 1);
 613                 file = lo->file;
 614                 lo->file = NULL;
 615                 nbd_clear_que(lo);
 616                 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
 617                 if (file)
 618                         fput(file);
 619                 lo->bytesize = 0;
 620                 inode->i_bdev->bd_inode->i_size = 0;
 621                 set_capacity(lo->disk, 0);
 622                 return lo->harderror;
 623         case NBD_CLEAR_QUE:
 624                 /*
 625                  * This is for compatibility only.  The queue is always cleared
 626                  * by NBD_DO_IT or NBD_CLEAR_SOCK.
 627                  */
 628                 BUG_ON(!lo->sock && !list_empty(&lo->queue_head));
 629                 return 0;
 630         case NBD_PRINT_DEBUG:
 631                 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
 632                         inode->i_bdev->bd_disk->disk_name,
 633                         lo->queue_head.next, lo->queue_head.prev,
 634                         &lo->queue_head);
 635                 return 0;
 636         }
 637         return -EINVAL;
 638 }
 639
 640 static struct block_device_operations nbd_fops =
 641 {
 642         .owner =        THIS_MODULE,
 643         .ioctl =        nbd_ioctl,
 644 };
 645
 646 /*
 647  * And here should be modules and kernel interface
 648  *  (Just smiley confuses emacs :-)
 649  */
 650
 651 static int __init nbd_init(void)
 652 {
 653         int err = -ENOMEM;
 654         int i;
 655
 656         BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
 657
 658         nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
 659         if (!nbd_dev)
 660                 return -ENOMEM;
 661
 662         for (i = 0; i < nbds_max; i++) {
 663                 struct gendisk *disk = alloc_disk(1);
 664                 elevator_t *old_e;
 665                 if (!disk)
 666                         goto out;
 667                 nbd_dev[i].disk = disk;
 668                 /*
 669                  * The new linux 2.5 block layer implementation requires
 670                  * every gendisk to have its very own request_queue struct.
 671                  * These structs are big so we dynamically allocate them.
 672                  */
 673                 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
 674                 if (!disk->queue) {
 675                         put_disk(disk);
 676                         goto out;
 677                 }
 678                 old_e = disk->queue->elevator;
 679                 if (elevator_init(disk->queue, "deadline") == 0 ||
 680                         elevator_init(disk->queue, "noop") == 0) {
 681                                 elevator_exit(old_e);
 682                 }
 683         }
 684
 685         if (register_blkdev(NBD_MAJOR, "nbd")) {
 686                 err = -EIO;
 687                 goto out;
 688         }
 689
 690         printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
 691         dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
 692
 693         for (i = 0; i < nbds_max; i++) {
 694                 struct gendisk *disk = nbd_dev[i].disk;
 695                 nbd_dev[i].file = NULL;
 696                 nbd_dev[i].magic = LO_MAGIC;
 697                 nbd_dev[i].flags = 0;
 698                 spin_lock_init(&nbd_dev[i].queue_lock);
 699                 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
 700                 mutex_init(&nbd_dev[i].tx_lock);
 701                 init_waitqueue_head(&nbd_dev[i].active_wq);
 702                 nbd_dev[i].blksize = 1024;
 703                 nbd_dev[i].bytesize = 0;
 704                 disk->major = NBD_MAJOR;
 705                 disk->first_minor = i;
 706                 disk->fops = &nbd_fops;
 707                 disk->private_data = &nbd_dev[i];
 708                 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 709                 sprintf(disk->disk_name, "nbd%d", i);
 710                 set_capacity(disk, 0);
 711                 add_disk(disk);
 712         }
 713
 714         return 0;
 715 out:
 716         while (i--) {
 717                 blk_cleanup_queue(nbd_dev[i].disk->queue);
 718                 put_disk(nbd_dev[i].disk);
 719         }
 720         return err;
 721 }
 722
 723 static void __exit nbd_cleanup(void)
 724 {
 725         int i;
 726         for (i = 0; i < nbds_max; i++) {
 727                 struct gendisk *disk = nbd_dev[i].disk;
 728                 nbd_dev[i].magic = 0;
 729                 if (disk) {
 730                         del_gendisk(disk);
 731                         blk_cleanup_queue(disk->queue);
 732                         put_disk(disk);
 733                 }
 734         }
 735         unregister_blkdev(NBD_MAJOR, "nbd");
 736         printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
 737 }
 738
 739 module_init(nbd_init);
 740 module_exit(nbd_cleanup);
 741
 742 MODULE_DESCRIPTION("Network Block Device");
 743 MODULE_LICENSE("GPL");
 744
 745 module_param(nbds_max, int, 0444);
 746 MODULE_PARM_DESC(nbds_max, "How many network block devices to initialize.");
 747 #ifndef NDEBUG
 748 module_param(debugflags, int, 0644);
 749 MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
 750 #endif