blob: dc88bcf320e8e8a859d27ec5b0b3cd2189547f83 [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;
Jens Axboe165125e2007-07-24 09:28:11 +0200103 struct request_queue *q = req->q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 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;
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700125 sigset_t blocked, oldset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127 /* Allow interception of SIGKILL only
128 * Don't allow other signals to interrupt the transmission */
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700129 siginitsetinv(&blocked, sigmask(SIGKILL));
130 sigprocmask(SIG_SETMASK, &blocked, &oldset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
132 do {
133 sock->sk->sk_allocation = GFP_NOIO;
134 iov.iov_base = buf;
135 iov.iov_len = size;
136 msg.msg_name = NULL;
137 msg.msg_namelen = 0;
138 msg.msg_control = NULL;
139 msg.msg_controllen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
141
142 if (send)
143 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
144 else
145 result = kernel_recvmsg(sock, &msg, &iov, 1, size, 0);
146
147 if (signal_pending(current)) {
148 siginfo_t info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700150 current->pid, current->comm,
151 dequeue_signal_lock(current, &current->blocked, &info));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 result = -EINTR;
153 break;
154 }
155
156 if (result <= 0) {
157 if (result == 0)
158 result = -EPIPE; /* short read */
159 break;
160 }
161 size -= result;
162 buf += result;
163 } while (size > 0);
164
Oleg Nesterovbe0ef952007-07-15 23:41:32 -0700165 sigprocmask(SIG_SETMASK, &oldset, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
167 return result;
168}
169
170static inline int sock_send_bvec(struct socket *sock, struct bio_vec *bvec,
171 int flags)
172{
173 int result;
174 void *kaddr = kmap(bvec->bv_page);
175 result = sock_xmit(sock, 1, kaddr + bvec->bv_offset, bvec->bv_len,
176 flags);
177 kunmap(bvec->bv_page);
178 return result;
179}
180
181static int nbd_send_req(struct nbd_device *lo, struct request *req)
182{
NeilBrown5705f702007-09-25 12:35:59 +0200183 int result, flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 struct nbd_request request;
185 unsigned long size = req->nr_sectors << 9;
186 struct socket *sock = lo->sock;
187
188 request.magic = htonl(NBD_REQUEST_MAGIC);
189 request.type = htonl(nbd_cmd(req));
190 request.from = cpu_to_be64((u64) req->sector << 9);
191 request.len = htonl(size);
192 memcpy(request.handle, &req, sizeof(req));
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%luB)\n",
195 lo->disk->disk_name, req,
196 nbdcmd_to_ascii(nbd_cmd(req)),
197 (unsigned long long)req->sector << 9,
198 req->nr_sectors << 9);
199 result = sock_xmit(sock, 1, &request, sizeof(request),
200 (nbd_cmd(req) == NBD_CMD_WRITE)? MSG_MORE: 0);
201 if (result <= 0) {
202 printk(KERN_ERR "%s: Send control failed (result %d)\n",
203 lo->disk->disk_name, result);
204 goto error_out;
205 }
206
207 if (nbd_cmd(req) == NBD_CMD_WRITE) {
NeilBrown5705f702007-09-25 12:35:59 +0200208 struct req_iterator iter;
209 struct bio_vec *bvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 /*
211 * we are really probing at internals to determine
212 * whether to set MSG_MORE or not...
213 */
NeilBrown5705f702007-09-25 12:35:59 +0200214 rq_for_each_segment(bvec, req, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +0200215 flags = 0;
216 if (!rq_iter_last(req, iter))
217 flags = MSG_MORE;
218 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
219 lo->disk->disk_name, req, bvec->bv_len);
220 result = sock_send_bvec(sock, bvec, flags);
221 if (result <= 0) {
222 printk(KERN_ERR "%s: Send data failed (result %d)\n",
223 lo->disk->disk_name, result);
224 goto error_out;
225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 }
227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 return 0;
229
230error_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 return 1;
232}
233
Denis Cheng0cbc591b2007-10-16 23:26:14 -0700234static struct request *nbd_find_request(struct nbd_device *lo,
235 struct request *xreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
Denis Chengd2c97402007-10-16 23:26:14 -0700237 struct request *req, *tmp;
Herbert Xu4b2f0262006-01-06 00:09:47 -0800238 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Herbert Xu4b2f0262006-01-06 00:09:47 -0800240 err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
241 if (unlikely(err))
242 goto out;
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 spin_lock(&lo->queue_lock);
Denis Chengd2c97402007-10-16 23:26:14 -0700245 list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (req != xreq)
247 continue;
248 list_del_init(&req->queuelist);
249 spin_unlock(&lo->queue_lock);
250 return req;
251 }
252 spin_unlock(&lo->queue_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800253
254 err = -ENOENT;
255
256out:
257 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259
260static inline int sock_recv_bvec(struct socket *sock, struct bio_vec *bvec)
261{
262 int result;
263 void *kaddr = kmap(bvec->bv_page);
264 result = sock_xmit(sock, 0, kaddr + bvec->bv_offset, bvec->bv_len,
265 MSG_WAITALL);
266 kunmap(bvec->bv_page);
267 return result;
268}
269
270/* NULL returned = something went wrong, inform userspace */
271static struct request *nbd_read_stat(struct nbd_device *lo)
272{
273 int result;
274 struct nbd_reply reply;
275 struct request *req;
276 struct socket *sock = lo->sock;
277
278 reply.magic = 0;
279 result = sock_xmit(sock, 0, &reply, sizeof(reply), MSG_WAITALL);
280 if (result <= 0) {
281 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
282 lo->disk->disk_name, result);
283 goto harderror;
284 }
Michal Feixe4b57e02006-07-30 03:03:31 -0700285
286 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
287 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
288 lo->disk->disk_name,
289 (unsigned long)ntohl(reply.magic));
290 result = -EPROTO;
291 goto harderror;
292 }
293
Denis Cheng0cbc591b2007-10-16 23:26:14 -0700294 req = nbd_find_request(lo, *(struct request **)reply.handle);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800295 if (unlikely(IS_ERR(req))) {
296 result = PTR_ERR(req);
297 if (result != -ENOENT)
298 goto harderror;
299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
301 lo->disk->disk_name, reply.handle);
302 result = -EBADR;
303 goto harderror;
304 }
305
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 if (ntohl(reply.error)) {
307 printk(KERN_ERR "%s: Other side returned error (%d)\n",
308 lo->disk->disk_name, ntohl(reply.error));
309 req->errors++;
310 return req;
311 }
312
313 dprintk(DBG_RX, "%s: request %p: got reply\n",
314 lo->disk->disk_name, req);
315 if (nbd_cmd(req) == NBD_CMD_READ) {
NeilBrown5705f702007-09-25 12:35:59 +0200316 struct req_iterator iter;
317 struct bio_vec *bvec;
318
319 rq_for_each_segment(bvec, req, iter) {
Jens Axboe6c92e692007-08-16 13:43:12 +0200320 result = sock_recv_bvec(sock, bvec);
321 if (result <= 0) {
322 printk(KERN_ERR "%s: Receive data failed (result %d)\n",
323 lo->disk->disk_name, result);
324 req->errors++;
325 return req;
326 }
327 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
328 lo->disk->disk_name, req, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 }
330 }
331 return req;
332harderror:
333 lo->harderror = result;
334 return NULL;
335}
336
Paul Clements6b39bb62006-12-06 20:40:53 -0800337static ssize_t pid_show(struct gendisk *disk, char *page)
338{
339 return sprintf(page, "%ld\n",
340 (long) ((struct nbd_device *)disk->private_data)->pid);
341}
342
343static struct disk_attribute pid_attr = {
344 .attr = { .name = "pid", .mode = S_IRUGO },
345 .show = pid_show,
346};
347
WANG Cong84963042007-05-09 02:33:36 -0700348static int nbd_do_it(struct nbd_device *lo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349{
350 struct request *req;
WANG Cong84963042007-05-09 02:33:36 -0700351 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 BUG_ON(lo->magic != LO_MAGIC);
354
Paul Clements6b39bb62006-12-06 20:40:53 -0800355 lo->pid = current->pid;
WANG Cong84963042007-05-09 02:33:36 -0700356 ret = sysfs_create_file(&lo->disk->kobj, &pid_attr.attr);
357 if (ret) {
358 printk(KERN_ERR "nbd: sysfs_create_file failed!");
359 return ret;
360 }
Paul Clements6b39bb62006-12-06 20:40:53 -0800361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 while ((req = nbd_read_stat(lo)) != NULL)
363 nbd_end_request(req);
Paul Clements6b39bb62006-12-06 20:40:53 -0800364
365 sysfs_remove_file(&lo->disk->kobj, &pid_attr.attr);
WANG Cong84963042007-05-09 02:33:36 -0700366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
369static void nbd_clear_que(struct nbd_device *lo)
370{
371 struct request *req;
372
373 BUG_ON(lo->magic != LO_MAGIC);
374
Herbert Xu4b2f0262006-01-06 00:09:47 -0800375 /*
376 * Because we have set lo->sock to NULL under the tx_lock, all
377 * modifications to the list must have completed by now. For
378 * the same reason, the active_req must be NULL.
379 *
380 * As a consequence, we don't need to take the spin lock while
381 * purging the list here.
382 */
383 BUG_ON(lo->sock);
384 BUG_ON(lo->active_req);
385
386 while (!list_empty(&lo->queue_head)) {
387 req = list_entry(lo->queue_head.next, struct request,
388 queuelist);
389 list_del_init(&req->queuelist);
390 req->errors++;
391 nbd_end_request(req);
392 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393}
394
395/*
396 * We always wait for result of write, for now. It would be nice to make it optional
397 * in future
Boaz Harroshe654bc42007-06-20 13:53:23 +0200398 * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
400 */
401
Jens Axboe165125e2007-07-24 09:28:11 +0200402static void do_nbd_request(struct request_queue * q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403{
404 struct request *req;
405
406 while ((req = elv_next_request(q)) != NULL) {
407 struct nbd_device *lo;
408
409 blkdev_dequeue_request(req);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200410 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
411 req->rq_disk->disk_name, req, req->cmd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
Jens Axboe4aff5e22006-08-10 08:44:47 +0200413 if (!blk_fs_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 goto error_out;
415
416 lo = req->rq_disk->private_data;
417
418 BUG_ON(lo->magic != LO_MAGIC);
419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 nbd_cmd(req) = NBD_CMD_READ;
421 if (rq_data_dir(req) == WRITE) {
422 nbd_cmd(req) = NBD_CMD_WRITE;
423 if (lo->flags & NBD_READ_ONLY) {
424 printk(KERN_ERR "%s: Write on read-only\n",
425 lo->disk->disk_name);
426 goto error_out;
427 }
428 }
429
430 req->errors = 0;
431 spin_unlock_irq(q->queue_lock);
432
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800433 mutex_lock(&lo->tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800434 if (unlikely(!lo->sock)) {
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800435 mutex_unlock(&lo->tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800436 printk(KERN_ERR "%s: Attempted send on closed socket\n",
437 lo->disk->disk_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 req->errors++;
439 nbd_end_request(req);
440 spin_lock_irq(q->queue_lock);
441 continue;
442 }
443
Herbert Xu4b2f0262006-01-06 00:09:47 -0800444 lo->active_req = req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 if (nbd_send_req(lo, req) != 0) {
447 printk(KERN_ERR "%s: Request send failed\n",
448 lo->disk->disk_name);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800449 req->errors++;
450 nbd_end_request(req);
451 } else {
452 spin_lock(&lo->queue_lock);
453 list_add(&req->queuelist, &lo->queue_head);
454 spin_unlock(&lo->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
456
Herbert Xu4b2f0262006-01-06 00:09:47 -0800457 lo->active_req = NULL;
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800458 mutex_unlock(&lo->tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800459 wake_up_all(&lo->active_wq);
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 spin_lock_irq(q->queue_lock);
462 continue;
463
464error_out:
465 req->errors++;
466 spin_unlock(q->queue_lock);
467 nbd_end_request(req);
468 spin_lock(q->queue_lock);
469 }
470 return;
471}
472
473static int nbd_ioctl(struct inode *inode, struct file *file,
474 unsigned int cmd, unsigned long arg)
475{
476 struct nbd_device *lo = inode->i_bdev->bd_disk->private_data;
477 int error;
478 struct request sreq ;
479
480 if (!capable(CAP_SYS_ADMIN))
481 return -EPERM;
482
483 BUG_ON(lo->magic != LO_MAGIC);
484
485 /* Anyone capable of this syscall can do *real bad* things */
486 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
487 lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
488
489 switch (cmd) {
490 case NBD_DISCONNECT:
491 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
Jens Axboe4aff5e22006-08-10 08:44:47 +0200492 sreq.cmd_type = REQ_TYPE_SPECIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 nbd_cmd(&sreq) = NBD_CMD_DISC;
494 /*
495 * Set these to sane values in case server implementation
496 * fails to check the request type first and also to keep
497 * debugging output cleaner.
498 */
499 sreq.sector = 0;
500 sreq.nr_sectors = 0;
501 if (!lo->sock)
502 return -EINVAL;
503 nbd_send_req(lo, &sreq);
504 return 0;
505
506 case NBD_CLEAR_SOCK:
507 error = 0;
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800508 mutex_lock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 lo->sock = NULL;
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800510 mutex_unlock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 file = lo->file;
512 lo->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 nbd_clear_que(lo);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800514 BUG_ON(!list_empty(&lo->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 if (file)
516 fput(file);
517 return error;
518 case NBD_SET_SOCK:
519 if (lo->file)
520 return -EBUSY;
521 error = -EINVAL;
522 file = fget(arg);
523 if (file) {
Josef Sipek17506042006-12-08 02:37:22 -0800524 inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (S_ISSOCK(inode->i_mode)) {
526 lo->file = file;
527 lo->sock = SOCKET_I(inode);
528 error = 0;
529 } else {
530 fput(file);
531 }
532 }
533 return error;
534 case NBD_SET_BLKSIZE:
535 lo->blksize = arg;
536 lo->bytesize &= ~(lo->blksize-1);
537 inode->i_bdev->bd_inode->i_size = lo->bytesize;
538 set_blocksize(inode->i_bdev, lo->blksize);
539 set_capacity(lo->disk, lo->bytesize >> 9);
540 return 0;
541 case NBD_SET_SIZE:
542 lo->bytesize = arg & ~(lo->blksize-1);
543 inode->i_bdev->bd_inode->i_size = lo->bytesize;
544 set_blocksize(inode->i_bdev, lo->blksize);
545 set_capacity(lo->disk, lo->bytesize >> 9);
546 return 0;
547 case NBD_SET_SIZE_BLOCKS:
548 lo->bytesize = ((u64) arg) * lo->blksize;
549 inode->i_bdev->bd_inode->i_size = lo->bytesize;
550 set_blocksize(inode->i_bdev, lo->blksize);
551 set_capacity(lo->disk, lo->bytesize >> 9);
552 return 0;
553 case NBD_DO_IT:
554 if (!lo->file)
555 return -EINVAL;
WANG Cong84963042007-05-09 02:33:36 -0700556 error = nbd_do_it(lo);
557 if (error)
558 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 /* on return tidy up in case we have a signal */
560 /* Forcibly shutdown the socket causing all listeners
561 * to error
562 *
563 * FIXME: This code is duplicated from sys_shutdown, but
564 * there should be a more generic interface rather than
565 * calling socket ops directly here */
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800566 mutex_lock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 if (lo->sock) {
568 printk(KERN_WARNING "%s: shutting down socket\n",
569 lo->disk->disk_name);
570 lo->sock->ops->shutdown(lo->sock,
571 SEND_SHUTDOWN|RCV_SHUTDOWN);
572 lo->sock = NULL;
573 }
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800574 mutex_unlock(&lo->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 file = lo->file;
576 lo->file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 nbd_clear_que(lo);
578 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
579 if (file)
580 fput(file);
581 return lo->harderror;
582 case NBD_CLEAR_QUE:
Herbert Xu4b2f0262006-01-06 00:09:47 -0800583 /*
584 * This is for compatibility only. The queue is always cleared
585 * by NBD_DO_IT or NBD_CLEAR_SOCK.
586 */
587 BUG_ON(!lo->sock && !list_empty(&lo->queue_head));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return 0;
589 case NBD_PRINT_DEBUG:
590 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
591 inode->i_bdev->bd_disk->disk_name,
592 lo->queue_head.next, lo->queue_head.prev,
593 &lo->queue_head);
594 return 0;
595 }
596 return -EINVAL;
597}
598
599static struct block_device_operations nbd_fops =
600{
601 .owner = THIS_MODULE,
602 .ioctl = nbd_ioctl,
603};
604
605/*
606 * And here should be modules and kernel interface
607 * (Just smiley confuses emacs :-)
608 */
609
610static int __init nbd_init(void)
611{
612 int err = -ENOMEM;
613 int i;
614
Adrian Bunk5b7b18c2006-03-25 03:07:04 -0800615 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700617 if (nbds_max > MAX_NBD) {
618 printk(KERN_CRIT "nbd: cannot allocate more than %u nbds; %u requested.\n", MAX_NBD,
619 nbds_max);
620 return -EINVAL;
621 }
622
623 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 struct gendisk *disk = alloc_disk(1);
625 if (!disk)
626 goto out;
627 nbd_dev[i].disk = disk;
628 /*
629 * The new linux 2.5 block layer implementation requires
630 * every gendisk to have its very own request_queue struct.
631 * These structs are big so we dynamically allocate them.
632 */
633 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
634 if (!disk->queue) {
635 put_disk(disk);
636 goto out;
637 }
638 }
639
640 if (register_blkdev(NBD_MAJOR, "nbd")) {
641 err = -EIO;
642 goto out;
643 }
644
645 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
646 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
647
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700648 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 struct gendisk *disk = nbd_dev[i].disk;
650 nbd_dev[i].file = NULL;
651 nbd_dev[i].magic = LO_MAGIC;
652 nbd_dev[i].flags = 0;
653 spin_lock_init(&nbd_dev[i].queue_lock);
654 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
Ingo Molnar82d4dc52006-03-23 03:00:38 -0800655 mutex_init(&nbd_dev[i].tx_lock);
Herbert Xu4b2f0262006-01-06 00:09:47 -0800656 init_waitqueue_head(&nbd_dev[i].active_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 nbd_dev[i].blksize = 1024;
658 nbd_dev[i].bytesize = 0x7ffffc00ULL << 10; /* 2TB */
659 disk->major = NBD_MAJOR;
660 disk->first_minor = i;
661 disk->fops = &nbd_fops;
662 disk->private_data = &nbd_dev[i];
663 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
664 sprintf(disk->disk_name, "nbd%d", i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 set_capacity(disk, 0x7ffffc00ULL << 1); /* 2 TB */
666 add_disk(disk);
667 }
668
669 return 0;
670out:
671 while (i--) {
672 blk_cleanup_queue(nbd_dev[i].disk->queue);
673 put_disk(nbd_dev[i].disk);
674 }
675 return err;
676}
677
678static void __exit nbd_cleanup(void)
679{
680 int i;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700681 for (i = 0; i < nbds_max; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 struct gendisk *disk = nbd_dev[i].disk;
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700683 nbd_dev[i].magic = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (disk) {
685 del_gendisk(disk);
686 blk_cleanup_queue(disk->queue);
687 put_disk(disk);
688 }
689 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 unregister_blkdev(NBD_MAJOR, "nbd");
691 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
692}
693
694module_init(nbd_init);
695module_exit(nbd_cleanup);
696
697MODULE_DESCRIPTION("Network Block Device");
698MODULE_LICENSE("GPL");
699
Lars Marowsky-Bree40be0c22005-05-01 08:59:07 -0700700module_param(nbds_max, int, 0444);
701MODULE_PARM_DESC(nbds_max, "How many network block devices to initialize.");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702#ifndef NDEBUG
703module_param(debugflags, int, 0644);
704MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
705#endif