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, 2008 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/mutex.h> 28 #include <linux/compiler.h> 29 #include <linux/err.h> 30 #include <linux/kernel.h> 31 #include <linux/slab.h> 32 #include <net/sock.h> 33 #include <linux/net.h> 34 #include <linux/kthread.h> 35 36 #include <asm/uaccess.h> 37 #include <asm/types.h> 38 39 #include <linux/nbd.h> 40 41 #define LO_MAGIC 0x68797548 42 43 #ifdef NDEBUG 44 #define dprintk(flags, fmt...) 45 #else /* NDEBUG */ 46 #define dprintk(flags, fmt...) do { \ 47 if (debugflags & (flags)) printk(KERN_DEBUG fmt); \ 48 } while (0) 49 #define DBG_IOCTL 0x0004 50 #define DBG_INIT 0x0010 51 #define DBG_EXIT 0x0020 52 #define DBG_BLKDEV 0x0100 53 #define DBG_RX 0x0200 54 #define DBG_TX 0x0400 55 static unsigned int debugflags; 56 #endif /* NDEBUG */ 57 58 static unsigned int nbds_max = 16; 59 static struct nbd_device *nbd_dev; 60 static int max_part; 61 62 /* 63 * Use just one lock (or at most 1 per NIC). Two arguments for this: 64 * 1. Each NIC is essentially a synchronization point for all servers 65 * accessed through that NIC so there's no need to have more locks 66 * than NICs anyway. 67 * 2. More locks lead to more "Dirty cache line bouncing" which will slow 68 * down each lock to the point where they're actually slower than just 69 * a single lock. 70 * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! 71 */ 72 static DEFINE_SPINLOCK(nbd_lock); 73 74 #ifndef NDEBUG 75 static const char *ioctl_cmd_to_ascii(int cmd) 76 { 77 switch (cmd) { 78 case NBD_SET_SOCK: return "set-sock"; 79 case NBD_SET_BLKSIZE: return "set-blksize"; 80 case NBD_SET_SIZE: return "set-size"; 81 case NBD_DO_IT: return "do-it"; 82 case NBD_CLEAR_SOCK: return "clear-sock"; 83 case NBD_CLEAR_QUE: return "clear-que"; 84 case NBD_PRINT_DEBUG: return "print-debug"; 85 case NBD_SET_SIZE_BLOCKS: return "set-size-blocks"; 86 case NBD_DISCONNECT: return "disconnect"; 87 case BLKROSET: return "set-read-only"; 88 case BLKFLSBUF: return "flush-buffer-cache"; 89 } 90 return "unknown"; 91 } 92 93 static const char *nbdcmd_to_ascii(int cmd) 94 { 95 switch (cmd) { 96 case NBD_CMD_READ: return "read"; 97 case NBD_CMD_WRITE: return "write"; 98 case NBD_CMD_DISC: return "disconnect"; 99 } 100 return "invalid"; 101 } 102 #endif /* NDEBUG */ 103 104 static void nbd_end_request(struct request *req) 105 { 106 int error = req->errors ? -EIO : 0; 107 struct request_queue *q = req->q; 108 unsigned long flags; 109 110 dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name, 111 req, error ? "failed" : "done"); 112 113 spin_lock_irqsave(q->queue_lock, flags); 114 __blk_end_request_all(req, error); 115 spin_unlock_irqrestore(q->queue_lock, flags); 116 } 117 118 static void sock_shutdown(struct nbd_device *lo, int lock) 119 { 120 /* Forcibly shutdown the socket causing all listeners 121 * to error 122 * 123 * FIXME: This code is duplicated from sys_shutdown, but 124 * there should be a more generic interface rather than 125 * calling socket ops directly here */ 126 if (lock) 127 mutex_lock(&lo->tx_lock); 128 if (lo->sock) { 129 dev_warn(disk_to_dev(lo->disk), "shutting down socket\n"); 130 kernel_sock_shutdown(lo->sock, SHUT_RDWR); 131 lo->sock = NULL; 132 } 133 if (lock) 134 mutex_unlock(&lo->tx_lock); 135 } 136 137 static void nbd_xmit_timeout(unsigned long arg) 138 { 139 struct task_struct *task = (struct task_struct *)arg; 140 141 printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n", 142 task->comm, task->pid); 143 force_sig(SIGKILL, task); 144 } 145 146 /* 147 * Send or receive packet. 148 */ 149 static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size, 150 int msg_flags) 151 { 152 struct socket *sock = lo->sock; 153 int result; 154 struct msghdr msg; 155 struct kvec iov; 156 sigset_t blocked, oldset; 157 158 if (unlikely(!sock)) { 159 dev_err(disk_to_dev(lo->disk), 160 "Attempted %s on closed socket in sock_xmit\n", 161 (send ? "send" : "recv")); 162 return -EINVAL; 163 } 164 165 /* Allow interception of SIGKILL only 166 * Don't allow other signals to interrupt the transmission */ 167 siginitsetinv(&blocked, sigmask(SIGKILL)); 168 sigprocmask(SIG_SETMASK, &blocked, &oldset); 169 170 do { 171 sock->sk->sk_allocation = GFP_NOIO; 172 iov.iov_base = buf; 173 iov.iov_len = size; 174 msg.msg_name = NULL; 175 msg.msg_namelen = 0; 176 msg.msg_control = NULL; 177 msg.msg_controllen = 0; 178 msg.msg_flags = msg_flags | MSG_NOSIGNAL; 179 180 if (send) { 181 struct timer_list ti; 182 183 if (lo->xmit_timeout) { 184 init_timer(&ti); 185 ti.function = nbd_xmit_timeout; 186 ti.data = (unsigned long)current; 187 ti.expires = jiffies + lo->xmit_timeout; 188 add_timer(&ti); 189 } 190 result = kernel_sendmsg(sock, &msg, &iov, 1, size); 191 if (lo->xmit_timeout) 192 del_timer_sync(&ti); 193 } else 194 result = kernel_recvmsg(sock, &msg, &iov, 1, size, 195 msg.msg_flags); 196 197 if (signal_pending(current)) { 198 siginfo_t info; 199 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n", 200 task_pid_nr(current), current->comm, 201 dequeue_signal_lock(current, ¤t->blocked, &info)); 202 result = -EINTR; 203 sock_shutdown(lo, !send); 204 break; 205 } 206 207 if (result <= 0) { 208 if (result == 0) 209 result = -EPIPE; /* short read */ 210 break; 211 } 212 size -= result; 213 buf += result; 214 } while (size > 0); 215 216 sigprocmask(SIG_SETMASK, &oldset, NULL); 217 218 return result; 219 } 220 221 static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec, 222 int flags) 223 { 224 int result; 225 void *kaddr = kmap(bvec->bv_page); 226 result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, bvec->bv_len, flags); 227 kunmap(bvec->bv_page); 228 return result; 229 } 230 231 /* always call with the tx_lock held */ 232 static int nbd_send_req(struct nbd_device *lo, struct request *req) 233 { 234 int result, flags; 235 struct nbd_request request; 236 unsigned long size = blk_rq_bytes(req); 237 238 request.magic = htonl(NBD_REQUEST_MAGIC); 239 request.type = htonl(nbd_cmd(req)); 240 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); 241 request.len = htonl(size); 242 memcpy(request.handle, &req, sizeof(req)); 243 244 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n", 245 lo->disk->disk_name, req, 246 nbdcmd_to_ascii(nbd_cmd(req)), 247 (unsigned long long)blk_rq_pos(req) << 9, 248 blk_rq_bytes(req)); 249 result = sock_xmit(lo, 1, &request, sizeof(request), 250 (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0); 251 if (result <= 0) { 252 dev_err(disk_to_dev(lo->disk), 253 "Send control failed (result %d)\n", result); 254 goto error_out; 255 } 256 257 if (nbd_cmd(req) == NBD_CMD_WRITE) { 258 struct req_iterator iter; 259 struct bio_vec *bvec; 260 /* 261 * we are really probing at internals to determine 262 * whether to set MSG_MORE or not... 263 */ 264 rq_for_each_segment(bvec, req, iter) { 265 flags = 0; 266 if (!rq_iter_last(req, iter)) 267 flags = MSG_MORE; 268 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n", 269 lo->disk->disk_name, req, bvec->bv_len); 270 result = sock_send_bvec(lo, bvec, flags); 271 if (result <= 0) { 272 dev_err(disk_to_dev(lo->disk), 273 "Send data failed (result %d)\n", 274 result); 275 goto error_out; 276 } 277 } 278 } 279 return 0; 280 281 error_out: 282 return -EIO; 283 } 284 285 static struct request *nbd_find_request(struct nbd_device *lo, 286 struct request *xreq) 287 { 288 struct request *req, *tmp; 289 int err; 290 291 err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq); 292 if (unlikely(err)) 293 goto out; 294 295 spin_lock(&lo->queue_lock); 296 list_for_each_entry_safe(req, tmp, &lo->queue_head, 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); 304 305 err = -ENOENT; 306 307 out: 308 return ERR_PTR(err); 309 } 310 311 static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec) 312 { 313 int result; 314 void *kaddr = kmap(bvec->bv_page); 315 result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len, 316 MSG_WAITALL); 317 kunmap(bvec->bv_page); 318 return result; 319 } 320 321 /* NULL returned = something went wrong, inform userspace */ 322 static struct request *nbd_read_stat(struct nbd_device *lo) 323 { 324 int result; 325 struct nbd_reply reply; 326 struct request *req; 327 328 reply.magic = 0; 329 result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL); 330 if (result <= 0) { 331 dev_err(disk_to_dev(lo->disk), 332 "Receive control failed (result %d)\n", result); 333 goto harderror; 334 } 335 336 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { 337 dev_err(disk_to_dev(lo->disk), "Wrong magic (0x%lx)\n", 338 (unsigned long)ntohl(reply.magic)); 339 result = -EPROTO; 340 goto harderror; 341 } 342 343 req = nbd_find_request(lo, *(struct request **)reply.handle); 344 if (IS_ERR(req)) { 345 result = PTR_ERR(req); 346 if (result != -ENOENT) 347 goto harderror; 348 349 dev_err(disk_to_dev(lo->disk), "Unexpected reply (%p)\n", 350 reply.handle); 351 result = -EBADR; 352 goto harderror; 353 } 354 355 if (ntohl(reply.error)) { 356 dev_err(disk_to_dev(lo->disk), "Other side returned error (%d)\n", 357 ntohl(reply.error)); 358 req->errors++; 359 return req; 360 } 361 362 dprintk(DBG_RX, "%s: request %p: got reply\n", 363 lo->disk->disk_name, req); 364 if (nbd_cmd(req) == NBD_CMD_READ) { 365 struct req_iterator iter; 366 struct bio_vec *bvec; 367 368 rq_for_each_segment(bvec, req, iter) { 369 result = sock_recv_bvec(lo, bvec); 370 if (result <= 0) { 371 dev_err(disk_to_dev(lo->disk), "Receive data failed (result %d)\n", 372 result); 373 req->errors++; 374 return req; 375 } 376 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n", 377 lo->disk->disk_name, req, bvec->bv_len); 378 } 379 } 380 return req; 381 harderror: 382 lo->harderror = result; 383 return NULL; 384 } 385 386 static ssize_t pid_show(struct device *dev, 387 struct device_attribute *attr, char *buf) 388 { 389 struct gendisk *disk = dev_to_disk(dev); 390 391 return sprintf(buf, "%ld\n", 392 (long) ((struct nbd_device *)disk->private_data)->pid); 393 } 394 395 static struct device_attribute pid_attr = { 396 .attr = { .name = "pid", .mode = S_IRUGO}, 397 .show = pid_show, 398 }; 399 400 static int nbd_do_it(struct nbd_device *lo) 401 { 402 struct request *req; 403 int ret; 404 405 BUG_ON(lo->magic != LO_MAGIC); 406 407 lo->pid = task_pid_nr(current); 408 ret = device_create_file(disk_to_dev(lo->disk), &pid_attr); 409 if (ret) { 410 dev_err(disk_to_dev(lo->disk), "device_create_file failed!\n"); 411 lo->pid = 0; 412 return ret; 413 } 414 415 while ((req = nbd_read_stat(lo)) != NULL) 416 nbd_end_request(req); 417 418 device_remove_file(disk_to_dev(lo->disk), &pid_attr); 419 lo->pid = 0; 420 return 0; 421 } 422 423 static void nbd_clear_que(struct nbd_device *lo) 424 { 425 struct request *req; 426 427 BUG_ON(lo->magic != LO_MAGIC); 428 429 /* 430 * Because we have set lo->sock to NULL under the tx_lock, all 431 * modifications to the list must have completed by now. For 432 * the same reason, the active_req must be NULL. 433 * 434 * As a consequence, we don't need to take the spin lock while 435 * purging the list here. 436 */ 437 BUG_ON(lo->sock); 438 BUG_ON(lo->active_req); 439 440 while (!list_empty(&lo->queue_head)) { 441 req = list_entry(lo->queue_head.next, struct request, 442 queuelist); 443 list_del_init(&req->queuelist); 444 req->errors++; 445 nbd_end_request(req); 446 } 447 } 448 449 450 static void nbd_handle_req(struct nbd_device *lo, struct request *req) 451 { 452 if (req->cmd_type != REQ_TYPE_FS) 453 goto error_out; 454 455 nbd_cmd(req) = NBD_CMD_READ; 456 if (rq_data_dir(req) == WRITE) { 457 nbd_cmd(req) = NBD_CMD_WRITE; 458 if (lo->flags & NBD_READ_ONLY) { 459 dev_err(disk_to_dev(lo->disk), 460 "Write on read-only\n"); 461 goto error_out; 462 } 463 } 464 465 req->errors = 0; 466 467 mutex_lock(&lo->tx_lock); 468 if (unlikely(!lo->sock)) { 469 mutex_unlock(&lo->tx_lock); 470 dev_err(disk_to_dev(lo->disk), 471 "Attempted send on closed socket\n"); 472 goto error_out; 473 } 474 475 lo->active_req = req; 476 477 if (nbd_send_req(lo, req) != 0) { 478 dev_err(disk_to_dev(lo->disk), "Request send failed\n"); 479 req->errors++; 480 nbd_end_request(req); 481 } else { 482 spin_lock(&lo->queue_lock); 483 list_add(&req->queuelist, &lo->queue_head); 484 spin_unlock(&lo->queue_lock); 485 } 486 487 lo->active_req = NULL; 488 mutex_unlock(&lo->tx_lock); 489 wake_up_all(&lo->active_wq); 490 491 return; 492 493 error_out: 494 req->errors++; 495 nbd_end_request(req); 496 } 497 498 static int nbd_thread(void *data) 499 { 500 struct nbd_device *lo = data; 501 struct request *req; 502 503 set_user_nice(current, -20); 504 while (!kthread_should_stop() || !list_empty(&lo->waiting_queue)) { 505 /* wait for something to do */ 506 wait_event_interruptible(lo->waiting_wq, 507 kthread_should_stop() || 508 !list_empty(&lo->waiting_queue)); 509 510 /* extract request */ 511 if (list_empty(&lo->waiting_queue)) 512 continue; 513 514 spin_lock_irq(&lo->queue_lock); 515 req = list_entry(lo->waiting_queue.next, struct request, 516 queuelist); 517 list_del_init(&req->queuelist); 518 spin_unlock_irq(&lo->queue_lock); 519 520 /* handle request */ 521 nbd_handle_req(lo, req); 522 } 523 return 0; 524 } 525 526 /* 527 * We always wait for result of write, for now. It would be nice to make it optional 528 * in future 529 * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK)) 530 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } 531 */ 532 533 static void do_nbd_request(struct request_queue *q) 534 { 535 struct request *req; 536 537 while ((req = blk_fetch_request(q)) != NULL) { 538 struct nbd_device *lo; 539 540 spin_unlock_irq(q->queue_lock); 541 542 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n", 543 req->rq_disk->disk_name, req, req->cmd_type); 544 545 lo = req->rq_disk->private_data; 546 547 BUG_ON(lo->magic != LO_MAGIC); 548 549 if (unlikely(!lo->sock)) { 550 dev_err(disk_to_dev(lo->disk), 551 "Attempted send on closed socket\n"); 552 req->errors++; 553 nbd_end_request(req); 554 spin_lock_irq(q->queue_lock); 555 continue; 556 } 557 558 spin_lock_irq(&lo->queue_lock); 559 list_add_tail(&req->queuelist, &lo->waiting_queue); 560 spin_unlock_irq(&lo->queue_lock); 561 562 wake_up(&lo->waiting_wq); 563 564 spin_lock_irq(q->queue_lock); 565 } 566 } 567 568 /* Must be called with tx_lock held */ 569 570 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *lo, 571 unsigned int cmd, unsigned long arg) 572 { 573 switch (cmd) { 574 case NBD_DISCONNECT: { 575 struct request sreq; 576 577 dev_info(disk_to_dev(lo->disk), "NBD_DISCONNECT\n"); 578 579 blk_rq_init(NULL, &sreq); 580 sreq.cmd_type = REQ_TYPE_SPECIAL; 581 nbd_cmd(&sreq) = NBD_CMD_DISC; 582 if (!lo->sock) 583 return -EINVAL; 584 nbd_send_req(lo, &sreq); 585 return 0; 586 } 587 588 case NBD_CLEAR_SOCK: { 589 struct file *file; 590 591 lo->sock = NULL; 592 file = lo->file; 593 lo->file = NULL; 594 nbd_clear_que(lo); 595 BUG_ON(!list_empty(&lo->queue_head)); 596 if (file) 597 fput(file); 598 return 0; 599 } 600 601 case NBD_SET_SOCK: { 602 struct file *file; 603 if (lo->file) 604 return -EBUSY; 605 file = fget(arg); 606 if (file) { 607 struct inode *inode = file->f_path.dentry->d_inode; 608 if (S_ISSOCK(inode->i_mode)) { 609 lo->file = file; 610 lo->sock = SOCKET_I(inode); 611 if (max_part > 0) 612 bdev->bd_invalidated = 1; 613 return 0; 614 } else { 615 fput(file); 616 } 617 } 618 return -EINVAL; 619 } 620 621 case NBD_SET_BLKSIZE: 622 lo->blksize = arg; 623 lo->bytesize &= ~(lo->blksize-1); 624 bdev->bd_inode->i_size = lo->bytesize; 625 set_blocksize(bdev, lo->blksize); 626 set_capacity(lo->disk, lo->bytesize >> 9); 627 return 0; 628 629 case NBD_SET_SIZE: 630 lo->bytesize = arg & ~(lo->blksize-1); 631 bdev->bd_inode->i_size = lo->bytesize; 632 set_blocksize(bdev, lo->blksize); 633 set_capacity(lo->disk, lo->bytesize >> 9); 634 return 0; 635 636 case NBD_SET_TIMEOUT: 637 lo->xmit_timeout = arg * HZ; 638 return 0; 639 640 case NBD_SET_SIZE_BLOCKS: 641 lo->bytesize = ((u64) arg) * lo->blksize; 642 bdev->bd_inode->i_size = lo->bytesize; 643 set_blocksize(bdev, lo->blksize); 644 set_capacity(lo->disk, lo->bytesize >> 9); 645 return 0; 646 647 case NBD_DO_IT: { 648 struct task_struct *thread; 649 struct file *file; 650 int error; 651 652 if (lo->pid) 653 return -EBUSY; 654 if (!lo->file) 655 return -EINVAL; 656 657 mutex_unlock(&lo->tx_lock); 658 659 thread = kthread_create(nbd_thread, lo, lo->disk->disk_name); 660 if (IS_ERR(thread)) { 661 mutex_lock(&lo->tx_lock); 662 return PTR_ERR(thread); 663 } 664 wake_up_process(thread); 665 error = nbd_do_it(lo); 666 kthread_stop(thread); 667 668 mutex_lock(&lo->tx_lock); 669 if (error) 670 return error; 671 sock_shutdown(lo, 0); 672 file = lo->file; 673 lo->file = NULL; 674 nbd_clear_que(lo); 675 dev_warn(disk_to_dev(lo->disk), "queue cleared\n"); 676 if (file) 677 fput(file); 678 lo->bytesize = 0; 679 bdev->bd_inode->i_size = 0; 680 set_capacity(lo->disk, 0); 681 if (max_part > 0) 682 ioctl_by_bdev(bdev, BLKRRPART, 0); 683 return lo->harderror; 684 } 685 686 case NBD_CLEAR_QUE: 687 /* 688 * This is for compatibility only. The queue is always cleared 689 * by NBD_DO_IT or NBD_CLEAR_SOCK. 690 */ 691 BUG_ON(!lo->sock && !list_empty(&lo->queue_head)); 692 return 0; 693 694 case NBD_PRINT_DEBUG: 695 dev_info(disk_to_dev(lo->disk), 696 "next = %p, prev = %p, head = %p\n", 697 lo->queue_head.next, lo->queue_head.prev, 698 &lo->queue_head); 699 return 0; 700 } 701 return -ENOTTY; 702 } 703 704 static int nbd_ioctl(struct block_device *bdev, fmode_t mode, 705 unsigned int cmd, unsigned long arg) 706 { 707 struct nbd_device *lo = bdev->bd_disk->private_data; 708 int error; 709 710 if (!capable(CAP_SYS_ADMIN)) 711 return -EPERM; 712 713 BUG_ON(lo->magic != LO_MAGIC); 714 715 /* Anyone capable of this syscall can do *real bad* things */ 716 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n", 717 lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg); 718 719 mutex_lock(&lo->tx_lock); 720 error = __nbd_ioctl(bdev, lo, cmd, arg); 721 mutex_unlock(&lo->tx_lock); 722 723 return error; 724 } 725 726 static const struct block_device_operations nbd_fops = 727 { 728 .owner = THIS_MODULE, 729 .ioctl = nbd_ioctl, 730 }; 731 732 /* 733 * And here should be modules and kernel interface 734 * (Just smiley confuses emacs :-) 735 */ 736 737 static int __init nbd_init(void) 738 { 739 int err = -ENOMEM; 740 int i; 741 int part_shift; 742 743 BUILD_BUG_ON(sizeof(struct nbd_request) != 28); 744 745 if (max_part < 0) { 746 printk(KERN_ERR "nbd: max_part must be >= 0\n"); 747 return -EINVAL; 748 } 749 750 nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); 751 if (!nbd_dev) 752 return -ENOMEM; 753 754 part_shift = 0; 755 if (max_part > 0) { 756 part_shift = fls(max_part); 757 758 /* 759 * Adjust max_part according to part_shift as it is exported 760 * to user space so that user can know the max number of 761 * partition kernel should be able to manage. 762 * 763 * Note that -1 is required because partition 0 is reserved 764 * for the whole disk. 765 */ 766 max_part = (1UL << part_shift) - 1; 767 } 768 769 if ((1UL << part_shift) > DISK_MAX_PARTS) 770 return -EINVAL; 771 772 if (nbds_max > 1UL << (MINORBITS - part_shift)) 773 return -EINVAL; 774 775 for (i = 0; i < nbds_max; i++) { 776 struct gendisk *disk = alloc_disk(1 << part_shift); 777 if (!disk) 778 goto out; 779 nbd_dev[i].disk = disk; 780 /* 781 * The new linux 2.5 block layer implementation requires 782 * every gendisk to have its very own request_queue struct. 783 * These structs are big so we dynamically allocate them. 784 */ 785 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); 786 if (!disk->queue) { 787 put_disk(disk); 788 goto out; 789 } 790 /* 791 * Tell the block layer that we are not a rotational device 792 */ 793 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); 794 } 795 796 if (register_blkdev(NBD_MAJOR, "nbd")) { 797 err = -EIO; 798 goto out; 799 } 800 801 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); 802 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags); 803 804 for (i = 0; i < nbds_max; i++) { 805 struct gendisk *disk = nbd_dev[i].disk; 806 nbd_dev[i].file = NULL; 807 nbd_dev[i].magic = LO_MAGIC; 808 nbd_dev[i].flags = 0; 809 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); 810 spin_lock_init(&nbd_dev[i].queue_lock); 811 INIT_LIST_HEAD(&nbd_dev[i].queue_head); 812 mutex_init(&nbd_dev[i].tx_lock); 813 init_waitqueue_head(&nbd_dev[i].active_wq); 814 init_waitqueue_head(&nbd_dev[i].waiting_wq); 815 nbd_dev[i].blksize = 1024; 816 nbd_dev[i].bytesize = 0; 817 disk->major = NBD_MAJOR; 818 disk->first_minor = i << part_shift; 819 disk->fops = &nbd_fops; 820 disk->private_data = &nbd_dev[i]; 821 sprintf(disk->disk_name, "nbd%d", i); 822 set_capacity(disk, 0); 823 add_disk(disk); 824 } 825 826 return 0; 827 out: 828 while (i--) { 829 blk_cleanup_queue(nbd_dev[i].disk->queue); 830 put_disk(nbd_dev[i].disk); 831 } 832 kfree(nbd_dev); 833 return err; 834 } 835 836 static void __exit nbd_cleanup(void) 837 { 838 int i; 839 for (i = 0; i < nbds_max; i++) { 840 struct gendisk *disk = nbd_dev[i].disk; 841 nbd_dev[i].magic = 0; 842 if (disk) { 843 del_gendisk(disk); 844 blk_cleanup_queue(disk->queue); 845 put_disk(disk); 846 } 847 } 848 unregister_blkdev(NBD_MAJOR, "nbd"); 849 kfree(nbd_dev); 850 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); 851 } 852 853 module_init(nbd_init); 854 module_exit(nbd_cleanup); 855 856 MODULE_DESCRIPTION("Network Block Device"); 857 MODULE_LICENSE("GPL"); 858 859 module_param(nbds_max, int, 0444); 860 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); 861 module_param(max_part, int, 0444); 862 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); 863 #ifndef NDEBUG 864 module_param(debugflags, int, 0644); 865 MODULE_PARM_DESC(debugflags, "flags for controlling debug output"); 866 #endif 867