]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/block/nbd.c
Remove duplicated unlikely() in IS_ERR()
[net-next-2.6.git] / drivers / block / nbd.c
CommitLineData
1da177e4
LT
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 *
dbf492d6 10 * This file is released under GPLv2 or later.
1da177e4 11 *
dbf492d6 12 * (part of code stolen from loop.c)
1da177e4
LT
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>
4b2f0260
HX
27#include <linux/compiler.h>
28#include <linux/err.h>
29#include <linux/kernel.h>
1da177e4 30#include <net/sock.h>
91cf45f0 31#include <linux/net.h>
48cf6061 32#include <linux/kthread.h>
1da177e4 33
1da177e4 34#include <asm/uaccess.h>
4b2f0260 35#include <asm/system.h>
1da177e4
LT
36#include <asm/types.h>
37
38#include <linux/nbd.h>
39
40#define LO_MAGIC 0x68797548
41
42#ifdef NDEBUG
43#define dprintk(flags, fmt...)
44#else /* NDEBUG */
45#define dprintk(flags, fmt...) do { \
46 if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
47} while (0)
48#define DBG_IOCTL 0x0004
49#define DBG_INIT 0x0010
50#define DBG_EXIT 0x0020
51#define DBG_BLKDEV 0x0100
52#define DBG_RX 0x0200
53#define DBG_TX 0x0400
54static unsigned int debugflags;
55#endif /* NDEBUG */
56
9c7a4169 57static unsigned int nbds_max = 16;
20a8143e 58static struct nbd_device *nbd_dev;
d71a6d73 59static int max_part;
1da177e4
LT
60
61/*
62 * Use just one lock (or at most 1 per NIC). Two arguments for this:
63 * 1. Each NIC is essentially a synchronization point for all servers
64 * accessed through that NIC so there's no need to have more locks
65 * than NICs anyway.
66 * 2. More locks lead to more "Dirty cache line bouncing" which will slow
67 * down each lock to the point where they're actually slower than just
68 * a single lock.
69 * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
70 */
71static DEFINE_SPINLOCK(nbd_lock);
72
73#ifndef NDEBUG
74static const char *ioctl_cmd_to_ascii(int cmd)
75{
76 switch (cmd) {
77 case NBD_SET_SOCK: return "set-sock";
78 case NBD_SET_BLKSIZE: return "set-blksize";
79 case NBD_SET_SIZE: return "set-size";
80 case NBD_DO_IT: return "do-it";
81 case NBD_CLEAR_SOCK: return "clear-sock";
82 case NBD_CLEAR_QUE: return "clear-que";
83 case NBD_PRINT_DEBUG: return "print-debug";
84 case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
85 case NBD_DISCONNECT: return "disconnect";
86 case BLKROSET: return "set-read-only";
87 case BLKFLSBUF: return "flush-buffer-cache";
88 }
89 return "unknown";
90}
91
92static const char *nbdcmd_to_ascii(int cmd)
93{
94 switch (cmd) {
95 case NBD_CMD_READ: return "read";
96 case NBD_CMD_WRITE: return "write";
97 case NBD_CMD_DISC: return "disconnect";
98 }
99 return "invalid";
100}
101#endif /* NDEBUG */
102
103static void nbd_end_request(struct request *req)
104{
097c94a4 105 int error = req->errors ? -EIO : 0;
165125e1 106 struct request_queue *q = req->q;
1da177e4
LT
107 unsigned long flags;
108
109 dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
097c94a4 110 req, error ? "failed" : "done");
1da177e4
LT
111
112 spin_lock_irqsave(q->queue_lock, flags);
097c94a4 113 __blk_end_request(req, error, req->nr_sectors << 9);
1da177e4
LT
114 spin_unlock_irqrestore(q->queue_lock, flags);
115}
116
7fdfd406
PC
117static void sock_shutdown(struct nbd_device *lo, int lock)
118{
119 /* Forcibly shutdown the socket causing all listeners
120 * to error
121 *
122 * FIXME: This code is duplicated from sys_shutdown, but
123 * there should be a more generic interface rather than
124 * calling socket ops directly here */
125 if (lock)
126 mutex_lock(&lo->tx_lock);
127 if (lo->sock) {
128 printk(KERN_WARNING "%s: shutting down socket\n",
129 lo->disk->disk_name);
91cf45f0 130 kernel_sock_shutdown(lo->sock, SHUT_RDWR);
7fdfd406
PC
131 lo->sock = NULL;
132 }
133 if (lock)
134 mutex_unlock(&lo->tx_lock);
135}
136
137static 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
1da177e4
LT
146/*
147 * Send or receive packet.
148 */
7fdfd406 149static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size,
1da177e4
LT
150 int msg_flags)
151{
7fdfd406 152 struct socket *sock = lo->sock;
1da177e4
LT
153 int result;
154 struct msghdr msg;
155 struct kvec iov;
be0ef957 156 sigset_t blocked, oldset;
1da177e4 157
ffc41cf8
MS
158 if (unlikely(!sock)) {
159 printk(KERN_ERR "%s: Attempted %s on closed socket in sock_xmit\n",
160 lo->disk->disk_name, (send ? "send" : "recv"));
161 return -EINVAL;
162 }
163
1da177e4
LT
164 /* Allow interception of SIGKILL only
165 * Don't allow other signals to interrupt the transmission */
be0ef957
ON
166 siginitsetinv(&blocked, sigmask(SIGKILL));
167 sigprocmask(SIG_SETMASK, &blocked, &oldset);
1da177e4
LT
168
169 do {
170 sock->sk->sk_allocation = GFP_NOIO;
171 iov.iov_base = buf;
172 iov.iov_len = size;
173 msg.msg_name = NULL;
174 msg.msg_namelen = 0;
175 msg.msg_control = NULL;
176 msg.msg_controllen = 0;
1da177e4
LT
177 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
178
7fdfd406
PC
179 if (send) {
180 struct timer_list ti;
181
182 if (lo->xmit_timeout) {
183 init_timer(&ti);
184 ti.function = nbd_xmit_timeout;
185 ti.data = (unsigned long)current;
186 ti.expires = jiffies + lo->xmit_timeout;
187 add_timer(&ti);
188 }
1da177e4 189 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
7fdfd406
PC
190 if (lo->xmit_timeout)
191 del_timer_sync(&ti);
192 } else
1da177e4
LT
193 result = kernel_recvmsg(sock, &msg, &iov, 1, size, 0);
194
195 if (signal_pending(current)) {
196 siginfo_t info;
1da177e4 197 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
ba25f9dc 198 task_pid_nr(current), current->comm,
be0ef957 199 dequeue_signal_lock(current, &current->blocked, &info));
1da177e4 200 result = -EINTR;
7fdfd406 201 sock_shutdown(lo, !send);
1da177e4
LT
202 break;
203 }
204
205 if (result <= 0) {
206 if (result == 0)
207 result = -EPIPE; /* short read */
208 break;
209 }
210 size -= result;
211 buf += result;
212 } while (size > 0);
213
be0ef957 214 sigprocmask(SIG_SETMASK, &oldset, NULL);
1da177e4
LT
215
216 return result;
217}
218
7fdfd406 219static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec,
1da177e4
LT
220 int flags)
221{
222 int result;
223 void *kaddr = kmap(bvec->bv_page);
7fdfd406 224 result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, bvec->bv_len, flags);
1da177e4
LT
225 kunmap(bvec->bv_page);
226 return result;
227}
228
7fdfd406 229/* always call with the tx_lock held */
1da177e4
LT
230static int nbd_send_req(struct nbd_device *lo, struct request *req)
231{
5705f702 232 int result, flags;
1da177e4
LT
233 struct nbd_request request;
234 unsigned long size = req->nr_sectors << 9;
1da177e4
LT
235
236 request.magic = htonl(NBD_REQUEST_MAGIC);
237 request.type = htonl(nbd_cmd(req));
238 request.from = cpu_to_be64((u64) req->sector << 9);
239 request.len = htonl(size);
240 memcpy(request.handle, &req, sizeof(req));
241
1da177e4
LT
242 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%luB)\n",
243 lo->disk->disk_name, req,
244 nbdcmd_to_ascii(nbd_cmd(req)),
245 (unsigned long long)req->sector << 9,
246 req->nr_sectors << 9);
7fdfd406
PC
247 result = sock_xmit(lo, 1, &request, sizeof(request),
248 (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
1da177e4
LT
249 if (result <= 0) {
250 printk(KERN_ERR "%s: Send control failed (result %d)\n",
251 lo->disk->disk_name, result);
252 goto error_out;
253 }
254
255 if (nbd_cmd(req) == NBD_CMD_WRITE) {
5705f702
N
256 struct req_iterator iter;
257 struct bio_vec *bvec;
1da177e4
LT
258 /*
259 * we are really probing at internals to determine
260 * whether to set MSG_MORE or not...
261 */
5705f702 262 rq_for_each_segment(bvec, req, iter) {
6c92e699
JA
263 flags = 0;
264 if (!rq_iter_last(req, iter))
265 flags = MSG_MORE;
266 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
267 lo->disk->disk_name, req, bvec->bv_len);
7fdfd406 268 result = sock_send_bvec(lo, bvec, flags);
6c92e699
JA
269 if (result <= 0) {
270 printk(KERN_ERR "%s: Send data failed (result %d)\n",
271 lo->disk->disk_name, result);
272 goto error_out;
273 }
1da177e4
LT
274 }
275 }
1da177e4
LT
276 return 0;
277
278error_out:
1da177e4
LT
279 return 1;
280}
281
0cbc591b
DC
282static struct request *nbd_find_request(struct nbd_device *lo,
283 struct request *xreq)
1da177e4 284{
d2c9740b 285 struct request *req, *tmp;
4b2f0260 286 int err;
1da177e4 287
4b2f0260
HX
288 err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
289 if (unlikely(err))
290 goto out;
291
1da177e4 292 spin_lock(&lo->queue_lock);
d2c9740b 293 list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) {
1da177e4
LT
294 if (req != xreq)
295 continue;
296 list_del_init(&req->queuelist);
297 spin_unlock(&lo->queue_lock);
298 return req;
299 }
300 spin_unlock(&lo->queue_lock);
4b2f0260
HX
301
302 err = -ENOENT;
303
304out:
305 return ERR_PTR(err);
1da177e4
LT
306}
307
7fdfd406 308static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec)
1da177e4
LT
309{
310 int result;
311 void *kaddr = kmap(bvec->bv_page);
7fdfd406 312 result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len,
1da177e4
LT
313 MSG_WAITALL);
314 kunmap(bvec->bv_page);
315 return result;
316}
317
318/* NULL returned = something went wrong, inform userspace */
319static struct request *nbd_read_stat(struct nbd_device *lo)
320{
321 int result;
322 struct nbd_reply reply;
323 struct request *req;
1da177e4
LT
324
325 reply.magic = 0;
7fdfd406 326 result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL);
1da177e4
LT
327 if (result <= 0) {
328 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
329 lo->disk->disk_name, result);
330 goto harderror;
331 }
e4b57e08
MF
332
333 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
334 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
335 lo->disk->disk_name,
336 (unsigned long)ntohl(reply.magic));
337 result = -EPROTO;
338 goto harderror;
339 }
340
0cbc591b 341 req = nbd_find_request(lo, *(struct request **)reply.handle);
801678c5 342 if (IS_ERR(req)) {
4b2f0260
HX
343 result = PTR_ERR(req);
344 if (result != -ENOENT)
345 goto harderror;
346
1da177e4
LT
347 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
348 lo->disk->disk_name, reply.handle);
349 result = -EBADR;
350 goto harderror;
351 }
352
1da177e4
LT
353 if (ntohl(reply.error)) {
354 printk(KERN_ERR "%s: Other side returned error (%d)\n",
355 lo->disk->disk_name, ntohl(reply.error));
356 req->errors++;
357 return req;
358 }
359
360 dprintk(DBG_RX, "%s: request %p: got reply\n",
361 lo->disk->disk_name, req);
362 if (nbd_cmd(req) == NBD_CMD_READ) {
5705f702
N
363 struct req_iterator iter;
364 struct bio_vec *bvec;
365
366 rq_for_each_segment(bvec, req, iter) {
7fdfd406 367 result = sock_recv_bvec(lo, bvec);
6c92e699
JA
368 if (result <= 0) {
369 printk(KERN_ERR "%s: Receive data failed (result %d)\n",
370 lo->disk->disk_name, result);
371 req->errors++;
372 return req;
373 }
374 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
375 lo->disk->disk_name, req, bvec->bv_len);
1da177e4
LT
376 }
377 }
378 return req;
379harderror:
380 lo->harderror = result;
381 return NULL;
382}
383
edfaa7c3
KS
384static ssize_t pid_show(struct device *dev,
385 struct device_attribute *attr, char *buf)
6b39bb65 386{
edfaa7c3
KS
387 struct gendisk *disk = dev_to_disk(dev);
388
389 return sprintf(buf, "%ld\n",
6b39bb65
PC
390 (long) ((struct nbd_device *)disk->private_data)->pid);
391}
392
edfaa7c3
KS
393static struct device_attribute pid_attr = {
394 .attr = { .name = "pid", .mode = S_IRUGO, .owner = THIS_MODULE },
6b39bb65
PC
395 .show = pid_show,
396};
397
84963048 398static int nbd_do_it(struct nbd_device *lo)
1da177e4
LT
399{
400 struct request *req;
84963048 401 int ret;
1da177e4
LT
402
403 BUG_ON(lo->magic != LO_MAGIC);
404
6b39bb65 405 lo->pid = current->pid;
edfaa7c3 406 ret = sysfs_create_file(&lo->disk->dev.kobj, &pid_attr.attr);
84963048
WC
407 if (ret) {
408 printk(KERN_ERR "nbd: sysfs_create_file failed!");
409 return ret;
410 }
6b39bb65 411
1da177e4
LT
412 while ((req = nbd_read_stat(lo)) != NULL)
413 nbd_end_request(req);
6b39bb65 414
edfaa7c3 415 sysfs_remove_file(&lo->disk->dev.kobj, &pid_attr.attr);
84963048 416 return 0;
1da177e4
LT
417}
418
419static void nbd_clear_que(struct nbd_device *lo)
420{
421 struct request *req;
422
423 BUG_ON(lo->magic != LO_MAGIC);
424
4b2f0260
HX
425 /*
426 * Because we have set lo->sock to NULL under the tx_lock, all
427 * modifications to the list must have completed by now. For
428 * the same reason, the active_req must be NULL.
429 *
430 * As a consequence, we don't need to take the spin lock while
431 * purging the list here.
432 */
433 BUG_ON(lo->sock);
434 BUG_ON(lo->active_req);
435
436 while (!list_empty(&lo->queue_head)) {
437 req = list_entry(lo->queue_head.next, struct request,
438 queuelist);
439 list_del_init(&req->queuelist);
440 req->errors++;
441 nbd_end_request(req);
442 }
1da177e4
LT
443}
444
7fdfd406 445
48cf6061
LV
446static void nbd_handle_req(struct nbd_device *lo, struct request *req)
447{
448 if (!blk_fs_request(req))
449 goto error_out;
450
451 nbd_cmd(req) = NBD_CMD_READ;
452 if (rq_data_dir(req) == WRITE) {
453 nbd_cmd(req) = NBD_CMD_WRITE;
454 if (lo->flags & NBD_READ_ONLY) {
455 printk(KERN_ERR "%s: Write on read-only\n",
456 lo->disk->disk_name);
457 goto error_out;
458 }
459 }
460
461 req->errors = 0;
462
463 mutex_lock(&lo->tx_lock);
464 if (unlikely(!lo->sock)) {
465 mutex_unlock(&lo->tx_lock);
466 printk(KERN_ERR "%s: Attempted send on closed socket\n",
467 lo->disk->disk_name);
468 req->errors++;
469 nbd_end_request(req);
470 return;
471 }
472
473 lo->active_req = req;
474
475 if (nbd_send_req(lo, req) != 0) {
476 printk(KERN_ERR "%s: Request send failed\n",
477 lo->disk->disk_name);
478 req->errors++;
479 nbd_end_request(req);
480 } else {
481 spin_lock(&lo->queue_lock);
482 list_add(&req->queuelist, &lo->queue_head);
483 spin_unlock(&lo->queue_lock);
484 }
485
486 lo->active_req = NULL;
487 mutex_unlock(&lo->tx_lock);
488 wake_up_all(&lo->active_wq);
489
490 return;
491
492error_out:
493 req->errors++;
494 nbd_end_request(req);
495}
496
497static int nbd_thread(void *data)
498{
499 struct nbd_device *lo = data;
500 struct request *req;
501
502 set_user_nice(current, -20);
503 while (!kthread_should_stop() || !list_empty(&lo->waiting_queue)) {
504 /* wait for something to do */
505 wait_event_interruptible(lo->waiting_wq,
506 kthread_should_stop() ||
507 !list_empty(&lo->waiting_queue));
508
509 /* extract request */
510 if (list_empty(&lo->waiting_queue))
511 continue;
512
513 spin_lock_irq(&lo->queue_lock);
514 req = list_entry(lo->waiting_queue.next, struct request,
515 queuelist);
516 list_del_init(&req->queuelist);
517 spin_unlock_irq(&lo->queue_lock);
518
519 /* handle request */
520 nbd_handle_req(lo, req);
521 }
522 return 0;
523}
524
1da177e4
LT
525/*
526 * We always wait for result of write, for now. It would be nice to make it optional
527 * in future
e654bc43 528 * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
1da177e4
LT
529 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
530 */
531
165125e1 532static void do_nbd_request(struct request_queue * q)
1da177e4
LT
533{
534 struct request *req;
535
536 while ((req = elv_next_request(q)) != NULL) {
537 struct nbd_device *lo;
538
539 blkdev_dequeue_request(req);
48cf6061
LV
540
541 spin_unlock_irq(q->queue_lock);
542
4aff5e23
JA
543 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
544 req->rq_disk->disk_name, req, req->cmd_type);
1da177e4 545
1da177e4
LT
546 lo = req->rq_disk->private_data;
547
548 BUG_ON(lo->magic != LO_MAGIC);
549
48cf6061
LV
550 spin_lock_irq(&lo->queue_lock);
551 list_add_tail(&req->queuelist, &lo->waiting_queue);
552 spin_unlock_irq(&lo->queue_lock);
1da177e4 553
48cf6061 554 wake_up(&lo->waiting_wq);
4b2f0260 555
1da177e4 556 spin_lock_irq(q->queue_lock);
1da177e4 557 }
1da177e4
LT
558}
559
560static int nbd_ioctl(struct inode *inode, struct file *file,
561 unsigned int cmd, unsigned long arg)
562{
563 struct nbd_device *lo = inode->i_bdev->bd_disk->private_data;
564 int error;
565 struct request sreq ;
48cf6061 566 struct task_struct *thread;
1da177e4
LT
567
568 if (!capable(CAP_SYS_ADMIN))
569 return -EPERM;
570
571 BUG_ON(lo->magic != LO_MAGIC);
572
573 /* Anyone capable of this syscall can do *real bad* things */
574 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
575 lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
576
577 switch (cmd) {
578 case NBD_DISCONNECT:
579 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
4aff5e23 580 sreq.cmd_type = REQ_TYPE_SPECIAL;
1da177e4
LT
581 nbd_cmd(&sreq) = NBD_CMD_DISC;
582 /*
583 * Set these to sane values in case server implementation
584 * fails to check the request type first and also to keep
585 * debugging output cleaner.
586 */
587 sreq.sector = 0;
588 sreq.nr_sectors = 0;
589 if (!lo->sock)
590 return -EINVAL;
7fdfd406 591 mutex_lock(&lo->tx_lock);
1da177e4 592 nbd_send_req(lo, &sreq);
7fdfd406 593 mutex_unlock(&lo->tx_lock);
1da177e4
LT
594 return 0;
595
596 case NBD_CLEAR_SOCK:
597 error = 0;
82d4dc5a 598 mutex_lock(&lo->tx_lock);
1da177e4 599 lo->sock = NULL;
82d4dc5a 600 mutex_unlock(&lo->tx_lock);
1da177e4
LT
601 file = lo->file;
602 lo->file = NULL;
1da177e4 603 nbd_clear_que(lo);
4b2f0260 604 BUG_ON(!list_empty(&lo->queue_head));
1da177e4
LT
605 if (file)
606 fput(file);
607 return error;
608 case NBD_SET_SOCK:
609 if (lo->file)
610 return -EBUSY;
611 error = -EINVAL;
612 file = fget(arg);
613 if (file) {
d71a6d73 614 struct block_device *bdev = inode->i_bdev;
17506041 615 inode = file->f_path.dentry->d_inode;
1da177e4
LT
616 if (S_ISSOCK(inode->i_mode)) {
617 lo->file = file;
618 lo->sock = SOCKET_I(inode);
d71a6d73
LV
619 if (max_part > 0)
620 bdev->bd_invalidated = 1;
1da177e4
LT
621 error = 0;
622 } else {
623 fput(file);
624 }
625 }
626 return error;
627 case NBD_SET_BLKSIZE:
628 lo->blksize = arg;
629 lo->bytesize &= ~(lo->blksize-1);
630 inode->i_bdev->bd_inode->i_size = lo->bytesize;
631 set_blocksize(inode->i_bdev, lo->blksize);
632 set_capacity(lo->disk, lo->bytesize >> 9);
633 return 0;
634 case NBD_SET_SIZE:
635 lo->bytesize = arg & ~(lo->blksize-1);
636 inode->i_bdev->bd_inode->i_size = lo->bytesize;
637 set_blocksize(inode->i_bdev, lo->blksize);
638 set_capacity(lo->disk, lo->bytesize >> 9);
639 return 0;
7fdfd406
PC
640 case NBD_SET_TIMEOUT:
641 lo->xmit_timeout = arg * HZ;
642 return 0;
1da177e4
LT
643 case NBD_SET_SIZE_BLOCKS:
644 lo->bytesize = ((u64) arg) * lo->blksize;
645 inode->i_bdev->bd_inode->i_size = lo->bytesize;
646 set_blocksize(inode->i_bdev, lo->blksize);
647 set_capacity(lo->disk, lo->bytesize >> 9);
648 return 0;
649 case NBD_DO_IT:
650 if (!lo->file)
651 return -EINVAL;
48cf6061
LV
652 thread = kthread_create(nbd_thread, lo, lo->disk->disk_name);
653 if (IS_ERR(thread))
654 return PTR_ERR(thread);
655 wake_up_process(thread);
84963048 656 error = nbd_do_it(lo);
48cf6061 657 kthread_stop(thread);
84963048
WC
658 if (error)
659 return error;
7fdfd406 660 sock_shutdown(lo, 1);
1da177e4
LT
661 file = lo->file;
662 lo->file = NULL;
1da177e4
LT
663 nbd_clear_que(lo);
664 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
665 if (file)
666 fput(file);
4b86a872
PC
667 lo->bytesize = 0;
668 inode->i_bdev->bd_inode->i_size = 0;
669 set_capacity(lo->disk, 0);
d71a6d73
LV
670 if (max_part > 0)
671 ioctl_by_bdev(inode->i_bdev, BLKRRPART, 0);
1da177e4
LT
672 return lo->harderror;
673 case NBD_CLEAR_QUE:
4b2f0260
HX
674 /*
675 * This is for compatibility only. The queue is always cleared
676 * by NBD_DO_IT or NBD_CLEAR_SOCK.
677 */
678 BUG_ON(!lo->sock && !list_empty(&lo->queue_head));
1da177e4
LT
679 return 0;
680 case NBD_PRINT_DEBUG:
681 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
682 inode->i_bdev->bd_disk->disk_name,
683 lo->queue_head.next, lo->queue_head.prev,
684 &lo->queue_head);
685 return 0;
686 }
687 return -EINVAL;
688}
689
690static struct block_device_operations nbd_fops =
691{
692 .owner = THIS_MODULE,
693 .ioctl = nbd_ioctl,
694};
695
696/*
697 * And here should be modules and kernel interface
698 * (Just smiley confuses emacs :-)
699 */
700
701static int __init nbd_init(void)
702{
703 int err = -ENOMEM;
704 int i;
d71a6d73 705 int part_shift;
1da177e4 706
5b7b18cc 707 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
1da177e4 708
20a8143e
PC
709 nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
710 if (!nbd_dev)
711 return -ENOMEM;
40be0c28 712
d71a6d73
LV
713 if (max_part < 0) {
714 printk(KERN_CRIT "nbd: max_part must be >= 0\n");
715 return -EINVAL;
716 }
717
718 part_shift = 0;
719 if (max_part > 0)
720 part_shift = fls(max_part);
721
40be0c28 722 for (i = 0; i < nbds_max; i++) {
d71a6d73 723 struct gendisk *disk = alloc_disk(1 << part_shift);
48f15b93 724 elevator_t *old_e;
1da177e4
LT
725 if (!disk)
726 goto out;
727 nbd_dev[i].disk = disk;
728 /*
729 * The new linux 2.5 block layer implementation requires
730 * every gendisk to have its very own request_queue struct.
731 * These structs are big so we dynamically allocate them.
732 */
733 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
734 if (!disk->queue) {
735 put_disk(disk);
736 goto out;
737 }
48f15b93
PC
738 old_e = disk->queue->elevator;
739 if (elevator_init(disk->queue, "deadline") == 0 ||
740 elevator_init(disk->queue, "noop") == 0) {
741 elevator_exit(old_e);
742 }
1da177e4
LT
743 }
744
745 if (register_blkdev(NBD_MAJOR, "nbd")) {
746 err = -EIO;
747 goto out;
748 }
749
750 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
751 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
752
40be0c28 753 for (i = 0; i < nbds_max; i++) {
1da177e4
LT
754 struct gendisk *disk = nbd_dev[i].disk;
755 nbd_dev[i].file = NULL;
756 nbd_dev[i].magic = LO_MAGIC;
757 nbd_dev[i].flags = 0;
48cf6061 758 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
1da177e4
LT
759 spin_lock_init(&nbd_dev[i].queue_lock);
760 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
82d4dc5a 761 mutex_init(&nbd_dev[i].tx_lock);
4b2f0260 762 init_waitqueue_head(&nbd_dev[i].active_wq);
48cf6061 763 init_waitqueue_head(&nbd_dev[i].waiting_wq);
1da177e4 764 nbd_dev[i].blksize = 1024;
4b86a872 765 nbd_dev[i].bytesize = 0;
1da177e4 766 disk->major = NBD_MAJOR;
d71a6d73 767 disk->first_minor = i << part_shift;
1da177e4
LT
768 disk->fops = &nbd_fops;
769 disk->private_data = &nbd_dev[i];
1da177e4 770 sprintf(disk->disk_name, "nbd%d", i);
4b86a872 771 set_capacity(disk, 0);
1da177e4
LT
772 add_disk(disk);
773 }
774
775 return 0;
776out:
777 while (i--) {
778 blk_cleanup_queue(nbd_dev[i].disk->queue);
779 put_disk(nbd_dev[i].disk);
780 }
781 return err;
782}
783
784static void __exit nbd_cleanup(void)
785{
786 int i;
40be0c28 787 for (i = 0; i < nbds_max; i++) {
1da177e4 788 struct gendisk *disk = nbd_dev[i].disk;
40be0c28 789 nbd_dev[i].magic = 0;
1da177e4
LT
790 if (disk) {
791 del_gendisk(disk);
792 blk_cleanup_queue(disk->queue);
793 put_disk(disk);
794 }
795 }
1da177e4
LT
796 unregister_blkdev(NBD_MAJOR, "nbd");
797 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
798}
799
800module_init(nbd_init);
801module_exit(nbd_cleanup);
802
803MODULE_DESCRIPTION("Network Block Device");
804MODULE_LICENSE("GPL");
805
40be0c28 806module_param(nbds_max, int, 0444);
d71a6d73
LV
807MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
808module_param(max_part, int, 0444);
809MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
1da177e4
LT
810#ifndef NDEBUG
811module_param(debugflags, int, 0644);
812MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
813#endif