blob: 090796bef78f16374f8c94f2bf70e23e602deabd [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Pavel Machekdbf492d2006-06-25 05:47:42 -070010 * This file is released under GPLv2 or later.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 *
Pavel Machekdbf492d2006-06-25 05:47:42 -070012 * (part of code stolen from loop.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 */
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>
Herbert Xu4b2f0262006-01-06 00:09:47 -080027#include <linux/compiler.h>
28#include <linux/err.h>
29#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <net/sock.h>
31
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <asm/uaccess.h>
Herbert Xu4b2f0262006-01-06 00:09:47 -080033#include <asm/system.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/types.h>
35
36#include <linux/nbd.h>
37
38#define LO_MAGIC 0x68797548
39
40#ifdef NDEBUG
41#define dprintk(flags, fmt...)
42#else /* NDEBUG */
43#define dprintk(flags, fmt...) do { \
44 if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
45} while (0)
46#define DBG_IOCTL 0x0004
47#define DBG_INIT 0x0010
48#define DBG_EXIT 0x0020
49#define DBG_BLKDEV 0x0100
50#define DBG_RX 0x0200
51#define DBG_TX 0x0400
52static unsigned int debugflags;
53#endif /* NDEBUG */
54
Ingo van Lil9c7a4162006-07-01 04:36:36 -070055static unsigned int nbds_max = 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056static struct nbd_device nbd_dev[MAX_NBD];
57
58/*
59 * Use just one lock (or at most 1 per NIC). Two arguments for this:
60 * 1. Each NIC is essentially a synchronization point for all servers
61 * accessed through that NIC so there's no need to have more locks
62 * than NICs anyway.
63 * 2. More locks lead to more "Dirty cache line bouncing" which will slow
64 * down each lock to the point where they're actually slower than just
65 * a single lock.
66 * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
67 */
68static DEFINE_SPINLOCK(nbd_lock);
69
70#ifndef NDEBUG
71static const char *ioctl_cmd_to_ascii(int cmd)
72{
73 switch (cmd) {
74 case NBD_SET_SOCK: return "set-sock";
75 case NBD_SET_BLKSIZE: return "set-blksize";
76 case NBD_SET_SIZE: return "set-size";
77 case NBD_DO_IT: return "do-it";
78 case NBD_CLEAR_SOCK: return "clear-sock";
79 case NBD_CLEAR_QUE: return "clear-que";
80 case NBD_PRINT_DEBUG: return "print-debug";
81 case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
82 case NBD_DISCONNECT: return "disconnect";
83 case BLKROSET: return "set-read-only";
84 case BLKFLSBUF: return "flush-buffer-cache";
85 }
86 return "unknown";
87}
88
89static const char *nbdcmd_to_ascii(int cmd)
90{
91 switch (cmd) {
92 case NBD_CMD_READ: return "read";
93 case NBD_CMD_WRITE: return "write";
94 case NBD_CMD_DISC: return "disconnect";
95 }
96 return "invalid";
97}
98#endif /* NDEBUG */
99
100static void nbd_end_request(struct request *req)
101{
102 int uptodate = (req->errors == 0) ? 1 : 0;
103 request_queue_t *q = req->q;
104 unsigned long flags;
105
106 dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
107 req, uptodate? "done": "failed");
108
109 spin_lock_irqsave(q->queue_lock, flags);
110 if (!end_that_request_first(req, uptodate, req->nr_sectors)) {
Tejun Heo8ffdc652006-01-06 09:49:03 +0100111 end_that_request_last(req, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 }
113 spin_unlock_irqrestore(q->queue_lock, flags);
114}
115
116/*
117 * Send or receive packet.
118 */
119static int sock_xmit(struct socket *sock, int send, void *buf, int size,
120 int msg_flags)
121{
122 int result;
123 struct msghdr msg;
124 struct kvec iov;
125 unsigned long flags;
126 sigset_t oldset;
127
128 /* Allow interception of SIGKILL only
129 * Don't allow other signals to interrupt the transmission */
130 spin_lock_irqsave(&current->sighand->siglock, flags);
131 oldset = current->blocked;
132 sigfillset(&current->blocked);
133 sigdelsetmask(&current->blocked, sigmask(SIGKILL));
134 recalc_sigpending();
135 spin_unlock_irqrestore(&current->sighand->siglock, flags);
136
137 do {
138 sock->sk->sk_allocation = GFP_NOIO;
139 iov.iov_base = buf;
140 iov.iov_len = size;
141 msg.msg_name = NULL;
142 msg.msg_namelen = 0;
143 msg.msg_control = NULL;
144 msg.msg_controllen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
146
147 if (send)
148 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
149 else
150 result = kernel_recvmsg(sock, &msg, &iov, 1, size, 0);
151
152 if (signal_pending(current)) {
153 siginfo_t info;
154 spin_lock_irqsave(&current->sighand->siglock, flags);
155 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
156 current->pid, current->comm,
157 dequeue_signal(current, &current->blocked, &info));
158 spin_unlock_irqrestore(&current->sighand->siglock, flags);
159 result = -EINTR;
160 break;
161 }
162
163 if (result <= 0) {
164 if (result == 0)
165 result = -EPIPE; /* short read */
166 break;
167 }
168 size -= result;
169 buf += result;
170 } while (size > 0);
171
172 spin_lock_irqsave(&current->sighand->siglock, flags);
173 current->blocked = oldset;
174 recalc_sigpending();
175 spin_unlock_irqrestore(&current->sighand->siglock, flags);
176
177 return result;
178}
179
180static inline int sock_send_bvec(struct socket *sock, struct bio_vec *bvec,
181 int flags)
182{
183 int result;
184 void *kaddr = kmap(bvec->bv_page);
185 result = sock_xmit(sock, 1, kaddr + bvec->bv_offset, bvec->bv_len,
186 flags);
187 kunmap(bvec->bv_page);
188 return result;
189}
190
191static int nbd_send_req(struct nbd_device *lo, struct request *req)
192{
193 int result, i, flags;
194 struct nbd_request request;
195 unsigned long size = req->nr_sectors << 9;
196 struct socket *sock = lo->sock;
197
198 request.magic = htonl(NBD_REQUEST_MAGIC);
199 request.type = htonl(nbd_cmd(req));
200 request.from = cpu_to_be64((u64) req->sector << 9);
201 request.len = htonl(size);
202 memcpy(request.handle, &req, sizeof(req));
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%luB)\n",
205 lo->disk->disk_name, req,
206 nbdcmd_to_ascii(nbd_cmd(req)),
207 (unsigned long long)req->sector << 9,
208 req->nr_sectors << 9);
209 result = sock_xmit(sock, 1, &request, sizeof(request),
210 (nbd_cmd(req) == NBD_CMD_WRITE)? MSG_MORE: 0);
211 if (result <= 0) {
212 printk(KERN_ERR "%s: Send control failed (result %d)\n",
213 lo->disk->disk_name, result);
214 goto error_out;
215 }
216
217 if (nbd_cmd(req) == NBD_CMD_WRITE) {
218 struct bio *bio;
219 /*
220 * we are really probing at internals to determine
221 * whether to set MSG_MORE or not...
222 */
223 rq_for_each_bio(bio, req) {
224 struct bio_vec *bvec;
225 bio_for_each_segment(bvec, bio, i) {
226 flags = 0;
227 if ((i < (bio->bi_vcnt - 1)) || bio->bi_next)
228 flags = MSG_MORE;
229 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
230 lo->disk->disk_name, req,
231 bvec->bv_len);
232 result = sock_send_bvec(sock, bvec, flags);
233 if (result <= 0) {
234 printk(KERN_ERR "%s: Send data failed (result %d)\n",
235 lo->disk->disk_name,
236 result);
237 goto error_out;
238 }
239 }
240 }
241 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 return 0;
243
244error_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 return 1;
246}
247
248static struct request *nbd_find_request(struct nbd_device *lo, char *handle)
249{
250 struct request *req;
251 struct list_head *tmp;
252 struct request *xreq;
Herbert Xu4b2f0262006-01-06 00:09:47 -0800253 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
255 memcpy(&xreq, handle, sizeof(xreq));
256
Herbert Xu4b2f0262006-01-06 00:09:47 -0800257 err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
258 if (unlikely(err))
259 goto out;
260
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 spin_lock(&lo->queue_lock);
262 list_for_each(tmp, &lo->queue_head) {
263 req = list_entry(tmp, struct request, queuelist);
264 if (req != xreq)
265 continue;
266 list_del_init(&req->queuelist);
267 spin_unlock(&lo->queue_lock);
268 return req;
269 }
270 spin_unlock(&lo->queue_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800271
272 err = -ENOENT;
273
274out:
275 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276}
277
278static inline int sock_recv_bvec(struct socket *sock, struct bio_vec *bvec)
279{
280 int result;
281 void *kaddr = kmap(bvec->bv_page);
282 result = sock_xmit(sock, 0, kaddr + bvec->bv_offset, bvec->bv_len,
283 MSG_WAITALL);
284 kunmap(bvec->bv_page);
285 return result;
286}
287
288/* NULL returned = something went wrong, inform userspace */
289static struct request *nbd_read_stat(struct nbd_device *lo)
290{
291 int result;
292 struct nbd_reply reply;
293 struct request *req;
294 struct socket *sock = lo->sock;
295
296 reply.magic = 0;
297 result = sock_xmit(sock, 0, &reply, sizeof(reply), MSG_WAITALL);
298 if (result <= 0) {
299 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
300 lo->disk->disk_name, result);
301 goto harderror;
302 }
Michal Feixe4b57e02006-07-30 03:03:31 -0700303
304 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
305 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
306 lo->disk->disk_name,
307 (unsigned long)ntohl(reply.magic));
308 result = -EPROTO;
309 goto harderror;
310 }
311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 req = nbd_find_request(lo, reply.handle);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800313 if (unlikely(IS_ERR(req))) {
314 result = PTR_ERR(req);
315 if (result != -ENOENT)
316 goto harderror;
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
319 lo->disk->disk_name, reply.handle);
320 result = -EBADR;
321 goto harderror;
322 }
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 if (ntohl(reply.error)) {
325 printk(KERN_ERR "%s: Other side returned error (%d)\n",
326 lo->disk->disk_name, ntohl(reply.error));
327 req->errors++;
328 return req;
329 }
330
331 dprintk(DBG_RX, "%s: request %p: got reply\n",
332 lo->disk->disk_name, req);
333 if (nbd_cmd(req) == NBD_CMD_READ) {
334 int i;
335 struct bio *bio;
336 rq_for_each_bio(bio, req) {
337 struct bio_vec *bvec;
338 bio_for_each_segment(bvec, bio, i) {
339 result = sock_recv_bvec(sock, bvec);
340 if (result <= 0) {
341 printk(KERN_ERR "%s: Receive data failed (result %d)\n",
342 lo->disk->disk_name,
343 result);
Michal Feixf0df33b2006-07-30 03:03:32 -0700344 req->errors++;
345 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 }
347 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
348 lo->disk->disk_name, req, bvec->bv_len);
349 }
350 }
351 }
352 return req;
353harderror:
354 lo->harderror = result;
355 return NULL;
356}
357
Paul Clements6b39bb62006-12-06 20:40:53 -0800358static ssize_t pid_show(struct gendisk *disk, char *page)
359{
360 return sprintf(page, "%ld\n",
361 (long) ((struct nbd_device *)disk->private_data)->pid);
362}
363
364static struct disk_attribute pid_attr = {
365 .attr = { .name = "pid", .mode = S_IRUGO },
366 .show = pid_show,
367};
368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369static void nbd_do_it(struct nbd_device *lo)
370{
371 struct request *req;
372
373 BUG_ON(lo->magic != LO_MAGIC);
374
Paul Clements6b39bb62006-12-06 20:40:53 -0800375 lo->pid = current->pid;
376 sysfs_create_file(&lo->disk->kobj, &pid_attr.attr);
377
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 while ((req = nbd_read_stat(lo)) != NULL)
379 nbd_end_request(req);
Paul Clements6b39bb62006-12-06 20:40:53 -0800380
381 sysfs_remove_file(&lo->disk->kobj, &pid_attr.attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 return;
383}
384
385static void nbd_clear_que(struct nbd_device *lo)
386{
387 struct request *req;
388
389 BUG_ON(lo->magic != LO_MAGIC);
390
Herbert Xu4b2f0262006-01-06 00:09:47 -0800391 /*
392 * Because we have set lo->sock to NULL under the tx_lock, all
393 * modifications to the list must have completed by now. For
394 * the same reason, the active_req must be NULL.
395 *
396 * As a consequence, we don't need to take the spin lock while
397 * purging the list here.
398 */
399 BUG_ON(lo->sock);
400 BUG_ON(lo->active_req);
401
402 while (!list_empty(&lo->queue_head)) {
403 req = list_entry(lo->queue_head.next, struct request,
404 queuelist);
405 list_del_init(&req->queuelist);
406 req->errors++;
407 nbd_end_request(req);
408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409}
410
411/*
412 * We always wait for result of write, for now. It would be nice to make it optional
413 * in future
414 * if ((req->cmd == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
415 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
416 */
417
418static void do_nbd_request(request_queue_t * q)
419{
420 struct request *req;
421
422 while ((req = elv_next_request(q)) != NULL) {
423 struct nbd_device *lo;
424
425 blkdev_dequeue_request(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200426 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
427 req->rq_disk->disk_name, req, req->cmd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Jens Axboe4aff5e22006-08-10 08:44:47 +0200429 if (!blk_fs_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 goto error_out;
431
432 lo = req->rq_disk->private_data;
433
434 BUG_ON(lo->magic != LO_MAGIC);
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 nbd_cmd(req) = NBD_CMD_READ;
437 if (rq_data_dir(req) == WRITE) {
438 nbd_cmd(req) = NBD_CMD_WRITE;
439 if (lo->flags & NBD_READ_ONLY) {
440 printk(KERN_ERR "%s: Write on read-only\n",
441 lo->disk->disk_name);
442 goto error_out;
443 }
444 }
445
446 req->errors = 0;
447 spin_unlock_irq(q->queue_lock);
448
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800449 mutex_lock(&lo->tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800450 if (unlikely(!lo->sock)) {
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800451 mutex_unlock(&lo->tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800452 printk(KERN_ERR "%s: Attempted send on closed socket\n",
453 lo->disk->disk_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 req->errors++;
455 nbd_end_request(req);
456 spin_lock_irq(q->queue_lock);
457 continue;
458 }
459
Herbert Xu4b2f0262006-01-06 00:09:47 -0800460 lo->active_req = req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 if (nbd_send_req(lo, req) != 0) {
463 printk(KERN_ERR "%s: Request send failed\n",
464 lo->disk->disk_name);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800465 req->errors++;
466 nbd_end_request(req);
467 } else {
468 spin_lock(&lo->queue_lock);
469 list_add(&req->queuelist, &lo->queue_head);
470 spin_unlock(&lo->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 }
472
Herbert Xu4b2f0262006-01-06 00:09:47 -0800473 lo->active_req = NULL;
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800474 mutex_unlock(&lo->tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800475 wake_up_all(&lo->active_wq);
476
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 spin_lock_irq(q->queue_lock);
478 continue;
479
480error_out:
481 req->errors++;
482 spin_unlock(q->queue_lock);
483 nbd_end_request(req);
484 spin_lock(q->queue_lock);
485 }
486 return;
487}
488
489static int nbd_ioctl(struct inode *inode, struct file *file,
490 unsigned int cmd, unsigned long arg)
491{
492 struct nbd_device *lo = inode->i_bdev->bd_disk->private_data;
493 int error;
494 struct request sreq ;
495
496 if (!capable(CAP_SYS_ADMIN))
497 return -EPERM;
498
499 BUG_ON(lo->magic != LO_MAGIC);
500
501 /* Anyone capable of this syscall can do *real bad* things */
502 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
503 lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
504
505 switch (cmd) {
506 case NBD_DISCONNECT:
507 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200508 sreq.cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 nbd_cmd(&sreq) = NBD_CMD_DISC;
510 /*
511 * Set these to sane values in case server implementation
512 * fails to check the request type first and also to keep
513 * debugging output cleaner.
514 */
515 sreq.sector = 0;
516 sreq.nr_sectors = 0;
517 if (!lo->sock)
518 return -EINVAL;
519 nbd_send_req(lo, &sreq);
520 return 0;
521
522 case NBD_CLEAR_SOCK:
523 error = 0;
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800524 mutex_lock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 lo->sock = NULL;
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800526 mutex_unlock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 file = lo->file;
528 lo->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 nbd_clear_que(lo);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800530 BUG_ON(!list_empty(&lo->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 if (file)
532 fput(file);
533 return error;
534 case NBD_SET_SOCK:
535 if (lo->file)
536 return -EBUSY;
537 error = -EINVAL;
538 file = fget(arg);
539 if (file) {
Josef Sipek17506042006-12-08 02:37:22 -0800540 inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (S_ISSOCK(inode->i_mode)) {
542 lo->file = file;
543 lo->sock = SOCKET_I(inode);
544 error = 0;
545 } else {
546 fput(file);
547 }
548 }
549 return error;
550 case NBD_SET_BLKSIZE:
551 lo->blksize = arg;
552 lo->bytesize &= ~(lo->blksize-1);
553 inode->i_bdev->bd_inode->i_size = lo->bytesize;
554 set_blocksize(inode->i_bdev, lo->blksize);
555 set_capacity(lo->disk, lo->bytesize >> 9);
556 return 0;
557 case NBD_SET_SIZE:
558 lo->bytesize = arg & ~(lo->blksize-1);
559 inode->i_bdev->bd_inode->i_size = lo->bytesize;
560 set_blocksize(inode->i_bdev, lo->blksize);
561 set_capacity(lo->disk, lo->bytesize >> 9);
562 return 0;
563 case NBD_SET_SIZE_BLOCKS:
564 lo->bytesize = ((u64) arg) * lo->blksize;
565 inode->i_bdev->bd_inode->i_size = lo->bytesize;
566 set_blocksize(inode->i_bdev, lo->blksize);
567 set_capacity(lo->disk, lo->bytesize >> 9);
568 return 0;
569 case NBD_DO_IT:
570 if (!lo->file)
571 return -EINVAL;
572 nbd_do_it(lo);
573 /* on return tidy up in case we have a signal */
574 /* Forcibly shutdown the socket causing all listeners
575 * to error
576 *
577 * FIXME: This code is duplicated from sys_shutdown, but
578 * there should be a more generic interface rather than
579 * calling socket ops directly here */
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800580 mutex_lock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (lo->sock) {
582 printk(KERN_WARNING "%s: shutting down socket\n",
583 lo->disk->disk_name);
584 lo->sock->ops->shutdown(lo->sock,
585 SEND_SHUTDOWN|RCV_SHUTDOWN);
586 lo->sock = NULL;
587 }
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800588 mutex_unlock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 file = lo->file;
590 lo->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 nbd_clear_que(lo);
592 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
593 if (file)
594 fput(file);
595 return lo->harderror;
596 case NBD_CLEAR_QUE:
Herbert Xu4b2f0262006-01-06 00:09:47 -0800597 /*
598 * This is for compatibility only. The queue is always cleared
599 * by NBD_DO_IT or NBD_CLEAR_SOCK.
600 */
601 BUG_ON(!lo->sock && !list_empty(&lo->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return 0;
603 case NBD_PRINT_DEBUG:
604 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
605 inode->i_bdev->bd_disk->disk_name,
606 lo->queue_head.next, lo->queue_head.prev,
607 &lo->queue_head);
608 return 0;
609 }
610 return -EINVAL;
611}
612
613static struct block_device_operations nbd_fops =
614{
615 .owner = THIS_MODULE,
616 .ioctl = nbd_ioctl,
617};
618
619/*
620 * And here should be modules and kernel interface
621 * (Just smiley confuses emacs :-)
622 */
623
624static int __init nbd_init(void)
625{
626 int err = -ENOMEM;
627 int i;
628
Adrian Bunk5b7b18c2006-03-25 03:07:04 -0800629 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700631 if (nbds_max > MAX_NBD) {
632 printk(KERN_CRIT "nbd: cannot allocate more than %u nbds; %u requested.\n", MAX_NBD,
633 nbds_max);
634 return -EINVAL;
635 }
636
637 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 struct gendisk *disk = alloc_disk(1);
639 if (!disk)
640 goto out;
641 nbd_dev[i].disk = disk;
642 /*
643 * The new linux 2.5 block layer implementation requires
644 * every gendisk to have its very own request_queue struct.
645 * These structs are big so we dynamically allocate them.
646 */
647 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
648 if (!disk->queue) {
649 put_disk(disk);
650 goto out;
651 }
652 }
653
654 if (register_blkdev(NBD_MAJOR, "nbd")) {
655 err = -EIO;
656 goto out;
657 }
658
659 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
660 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
661
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700662 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 struct gendisk *disk = nbd_dev[i].disk;
664 nbd_dev[i].file = NULL;
665 nbd_dev[i].magic = LO_MAGIC;
666 nbd_dev[i].flags = 0;
667 spin_lock_init(&nbd_dev[i].queue_lock);
668 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800669 mutex_init(&nbd_dev[i].tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800670 init_waitqueue_head(&nbd_dev[i].active_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 nbd_dev[i].blksize = 1024;
672 nbd_dev[i].bytesize = 0x7ffffc00ULL << 10; /* 2TB */
673 disk->major = NBD_MAJOR;
674 disk->first_minor = i;
675 disk->fops = &nbd_fops;
676 disk->private_data = &nbd_dev[i];
677 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
678 sprintf(disk->disk_name, "nbd%d", i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 set_capacity(disk, 0x7ffffc00ULL << 1); /* 2 TB */
680 add_disk(disk);
681 }
682
683 return 0;
684out:
685 while (i--) {
686 blk_cleanup_queue(nbd_dev[i].disk->queue);
687 put_disk(nbd_dev[i].disk);
688 }
689 return err;
690}
691
692static void __exit nbd_cleanup(void)
693{
694 int i;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700695 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 struct gendisk *disk = nbd_dev[i].disk;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700697 nbd_dev[i].magic = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 if (disk) {
699 del_gendisk(disk);
700 blk_cleanup_queue(disk->queue);
701 put_disk(disk);
702 }
703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 unregister_blkdev(NBD_MAJOR, "nbd");
705 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
706}
707
708module_init(nbd_init);
709module_exit(nbd_cleanup);
710
711MODULE_DESCRIPTION("Network Block Device");
712MODULE_LICENSE("GPL");
713
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700714module_param(nbds_max, int, 0444);
715MODULE_PARM_DESC(nbds_max, "How many network block devices to initialize.");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716#ifndef NDEBUG
717module_param(debugflags, int, 0644);
718MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
719#endif