Commit | Line | Data |
---|---|---|
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 | * | |
a2531293 | 7 | * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz> |
1da177e4 LT |
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> | |
2a48fc0a | 27 | #include <linux/mutex.h> |
4b2f0260 HX |
28 | #include <linux/compiler.h> |
29 | #include <linux/err.h> | |
30 | #include <linux/kernel.h> | |
5a0e3ad6 | 31 | #include <linux/slab.h> |
1da177e4 | 32 | #include <net/sock.h> |
91cf45f0 | 33 | #include <linux/net.h> |
48cf6061 | 34 | #include <linux/kthread.h> |
b9c495bb | 35 | #include <linux/types.h> |
1da177e4 | 36 | |
1da177e4 LT |
37 | #include <asm/uaccess.h> |
38 | #include <asm/types.h> | |
39 | ||
40 | #include <linux/nbd.h> | |
41 | ||
13e71d69 MP |
42 | struct nbd_device { |
43 | int flags; | |
44 | int harderror; /* Code of hard error */ | |
45 | struct socket * sock; /* If == NULL, device is not ready, yet */ | |
46 | int magic; | |
47 | ||
48 | spinlock_t queue_lock; | |
49 | struct list_head queue_head; /* Requests waiting result */ | |
50 | struct request *active_req; | |
51 | wait_queue_head_t active_wq; | |
52 | struct list_head waiting_queue; /* Requests to be sent */ | |
53 | wait_queue_head_t waiting_wq; | |
54 | ||
55 | struct mutex tx_lock; | |
56 | struct gendisk *disk; | |
57 | int blksize; | |
b9c495bb | 58 | loff_t bytesize; |
13e71d69 MP |
59 | pid_t pid; /* pid of nbd-client, if attached */ |
60 | int xmit_timeout; | |
61 | int disconnect; /* a disconnect has been requested by user */ | |
62 | }; | |
63 | ||
f4507164 | 64 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 65 | |
9c7a4169 | 66 | static unsigned int nbds_max = 16; |
20a8143e | 67 | static struct nbd_device *nbd_dev; |
d71a6d73 | 68 | static int max_part; |
1da177e4 LT |
69 | |
70 | /* | |
71 | * Use just one lock (or at most 1 per NIC). Two arguments for this: | |
72 | * 1. Each NIC is essentially a synchronization point for all servers | |
73 | * accessed through that NIC so there's no need to have more locks | |
74 | * than NICs anyway. | |
75 | * 2. More locks lead to more "Dirty cache line bouncing" which will slow | |
76 | * down each lock to the point where they're actually slower than just | |
77 | * a single lock. | |
78 | * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! | |
79 | */ | |
80 | static DEFINE_SPINLOCK(nbd_lock); | |
81 | ||
d18509f5 | 82 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 83 | { |
d18509f5 | 84 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
85 | } |
86 | ||
87 | static const char *nbdcmd_to_ascii(int cmd) | |
88 | { | |
89 | switch (cmd) { | |
90 | case NBD_CMD_READ: return "read"; | |
91 | case NBD_CMD_WRITE: return "write"; | |
92 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 93 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 94 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
95 | } |
96 | return "invalid"; | |
97 | } | |
1da177e4 | 98 | |
d18509f5 | 99 | static void nbd_end_request(struct nbd_device *nbd, struct request *req) |
1da177e4 | 100 | { |
097c94a4 | 101 | int error = req->errors ? -EIO : 0; |
165125e1 | 102 | struct request_queue *q = req->q; |
1da177e4 LT |
103 | unsigned long flags; |
104 | ||
d18509f5 MP |
105 | dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", req, |
106 | error ? "failed" : "done"); | |
1da177e4 LT |
107 | |
108 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 109 | __blk_end_request_all(req, error); |
1da177e4 LT |
110 | spin_unlock_irqrestore(q->queue_lock, flags); |
111 | } | |
112 | ||
e018e757 MP |
113 | /* |
114 | * Forcibly shutdown the socket causing all listeners to error | |
115 | */ | |
f4507164 | 116 | static void sock_shutdown(struct nbd_device *nbd, int lock) |
7fdfd406 | 117 | { |
7fdfd406 | 118 | if (lock) |
f4507164 WG |
119 | mutex_lock(&nbd->tx_lock); |
120 | if (nbd->sock) { | |
121 | dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n"); | |
122 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
123 | nbd->sock = NULL; | |
7fdfd406 PC |
124 | } |
125 | if (lock) | |
f4507164 | 126 | mutex_unlock(&nbd->tx_lock); |
7fdfd406 PC |
127 | } |
128 | ||
129 | static void nbd_xmit_timeout(unsigned long arg) | |
130 | { | |
131 | struct task_struct *task = (struct task_struct *)arg; | |
132 | ||
133 | printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n", | |
134 | task->comm, task->pid); | |
135 | force_sig(SIGKILL, task); | |
136 | } | |
137 | ||
1da177e4 LT |
138 | /* |
139 | * Send or receive packet. | |
140 | */ | |
f4507164 | 141 | static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size, |
1da177e4 LT |
142 | int msg_flags) |
143 | { | |
f4507164 | 144 | struct socket *sock = nbd->sock; |
1da177e4 LT |
145 | int result; |
146 | struct msghdr msg; | |
147 | struct kvec iov; | |
be0ef957 | 148 | sigset_t blocked, oldset; |
7f338fe4 | 149 | unsigned long pflags = current->flags; |
1da177e4 | 150 | |
ffc41cf8 | 151 | if (unlikely(!sock)) { |
f4507164 | 152 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
153 | "Attempted %s on closed socket in sock_xmit\n", |
154 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
155 | return -EINVAL; |
156 | } | |
157 | ||
1da177e4 LT |
158 | /* Allow interception of SIGKILL only |
159 | * Don't allow other signals to interrupt the transmission */ | |
be0ef957 ON |
160 | siginitsetinv(&blocked, sigmask(SIGKILL)); |
161 | sigprocmask(SIG_SETMASK, &blocked, &oldset); | |
1da177e4 | 162 | |
7f338fe4 | 163 | current->flags |= PF_MEMALLOC; |
1da177e4 | 164 | do { |
7f338fe4 | 165 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
166 | iov.iov_base = buf; |
167 | iov.iov_len = size; | |
168 | msg.msg_name = NULL; | |
169 | msg.msg_namelen = 0; | |
170 | msg.msg_control = NULL; | |
171 | msg.msg_controllen = 0; | |
1da177e4 LT |
172 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
173 | ||
7fdfd406 PC |
174 | if (send) { |
175 | struct timer_list ti; | |
176 | ||
f4507164 | 177 | if (nbd->xmit_timeout) { |
7fdfd406 PC |
178 | init_timer(&ti); |
179 | ti.function = nbd_xmit_timeout; | |
180 | ti.data = (unsigned long)current; | |
f4507164 | 181 | ti.expires = jiffies + nbd->xmit_timeout; |
7fdfd406 PC |
182 | add_timer(&ti); |
183 | } | |
1da177e4 | 184 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
f4507164 | 185 | if (nbd->xmit_timeout) |
7fdfd406 PC |
186 | del_timer_sync(&ti); |
187 | } else | |
35fbf5bc NK |
188 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
189 | msg.msg_flags); | |
1da177e4 LT |
190 | |
191 | if (signal_pending(current)) { | |
192 | siginfo_t info; | |
1da177e4 | 193 | printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n", |
ba25f9dc | 194 | task_pid_nr(current), current->comm, |
be0ef957 | 195 | dequeue_signal_lock(current, ¤t->blocked, &info)); |
1da177e4 | 196 | result = -EINTR; |
f4507164 | 197 | sock_shutdown(nbd, !send); |
1da177e4 LT |
198 | break; |
199 | } | |
200 | ||
201 | if (result <= 0) { | |
202 | if (result == 0) | |
203 | result = -EPIPE; /* short read */ | |
204 | break; | |
205 | } | |
206 | size -= result; | |
207 | buf += result; | |
208 | } while (size > 0); | |
209 | ||
be0ef957 | 210 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
7f338fe4 | 211 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 LT |
212 | |
213 | return result; | |
214 | } | |
215 | ||
f4507164 | 216 | static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec, |
1da177e4 LT |
217 | int flags) |
218 | { | |
219 | int result; | |
220 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 WG |
221 | result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset, |
222 | bvec->bv_len, flags); | |
1da177e4 LT |
223 | kunmap(bvec->bv_page); |
224 | return result; | |
225 | } | |
226 | ||
7fdfd406 | 227 | /* always call with the tx_lock held */ |
f4507164 | 228 | static int nbd_send_req(struct nbd_device *nbd, struct request *req) |
1da177e4 | 229 | { |
5705f702 | 230 | int result, flags; |
1da177e4 | 231 | struct nbd_request request; |
1011c1b9 | 232 | unsigned long size = blk_rq_bytes(req); |
1da177e4 | 233 | |
04cfac4e | 234 | memset(&request, 0, sizeof(request)); |
1da177e4 LT |
235 | request.magic = htonl(NBD_REQUEST_MAGIC); |
236 | request.type = htonl(nbd_cmd(req)); | |
75f187ab | 237 | |
04cfac4e | 238 | if (nbd_cmd(req) != NBD_CMD_FLUSH && nbd_cmd(req) != NBD_CMD_DISC) { |
75f187ab AB |
239 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
240 | request.len = htonl(size); | |
241 | } | |
1da177e4 LT |
242 | memcpy(request.handle, &req, sizeof(req)); |
243 | ||
d18509f5 MP |
244 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
245 | req, nbdcmd_to_ascii(nbd_cmd(req)), | |
246 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); | |
f4507164 | 247 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
7fdfd406 | 248 | (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 249 | if (result <= 0) { |
f4507164 | 250 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 251 | "Send control failed (result %d)\n", result); |
dab5313a | 252 | return -EIO; |
1da177e4 LT |
253 | } |
254 | ||
255 | if (nbd_cmd(req) == NBD_CMD_WRITE) { | |
5705f702 | 256 | struct req_iterator iter; |
7988613b | 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 | 263 | flags = 0; |
4550dd6c | 264 | if (!rq_iter_last(bvec, iter)) |
6c92e699 | 265 | flags = MSG_MORE; |
d18509f5 MP |
266 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
267 | req, bvec.bv_len); | |
7988613b | 268 | result = sock_send_bvec(nbd, &bvec, flags); |
6c92e699 | 269 | if (result <= 0) { |
f4507164 | 270 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
271 | "Send data failed (result %d)\n", |
272 | result); | |
dab5313a | 273 | return -EIO; |
6c92e699 | 274 | } |
1da177e4 LT |
275 | } |
276 | } | |
1da177e4 | 277 | return 0; |
1da177e4 LT |
278 | } |
279 | ||
f4507164 | 280 | static struct request *nbd_find_request(struct nbd_device *nbd, |
0cbc591b | 281 | struct request *xreq) |
1da177e4 | 282 | { |
d2c9740b | 283 | struct request *req, *tmp; |
4b2f0260 | 284 | int err; |
1da177e4 | 285 | |
f4507164 | 286 | err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq); |
4b2f0260 | 287 | if (unlikely(err)) |
de9ad6d4 | 288 | return ERR_PTR(err); |
4b2f0260 | 289 | |
f4507164 WG |
290 | spin_lock(&nbd->queue_lock); |
291 | list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) { | |
1da177e4 LT |
292 | if (req != xreq) |
293 | continue; | |
294 | list_del_init(&req->queuelist); | |
f4507164 | 295 | spin_unlock(&nbd->queue_lock); |
1da177e4 LT |
296 | return req; |
297 | } | |
f4507164 | 298 | spin_unlock(&nbd->queue_lock); |
4b2f0260 | 299 | |
de9ad6d4 | 300 | return ERR_PTR(-ENOENT); |
1da177e4 LT |
301 | } |
302 | ||
f4507164 | 303 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
304 | { |
305 | int result; | |
306 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 307 | result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
308 | MSG_WAITALL); |
309 | kunmap(bvec->bv_page); | |
310 | return result; | |
311 | } | |
312 | ||
313 | /* NULL returned = something went wrong, inform userspace */ | |
f4507164 | 314 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
315 | { |
316 | int result; | |
317 | struct nbd_reply reply; | |
318 | struct request *req; | |
1da177e4 LT |
319 | |
320 | reply.magic = 0; | |
f4507164 | 321 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 322 | if (result <= 0) { |
f4507164 | 323 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 324 | "Receive control failed (result %d)\n", result); |
1da177e4 LT |
325 | goto harderror; |
326 | } | |
e4b57e08 MF |
327 | |
328 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 329 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 MF |
330 | (unsigned long)ntohl(reply.magic)); |
331 | result = -EPROTO; | |
332 | goto harderror; | |
333 | } | |
334 | ||
f4507164 | 335 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 336 | if (IS_ERR(req)) { |
4b2f0260 HX |
337 | result = PTR_ERR(req); |
338 | if (result != -ENOENT) | |
339 | goto harderror; | |
340 | ||
f4507164 | 341 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 342 | reply.handle); |
1da177e4 LT |
343 | result = -EBADR; |
344 | goto harderror; | |
345 | } | |
346 | ||
1da177e4 | 347 | if (ntohl(reply.error)) { |
f4507164 | 348 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 349 | ntohl(reply.error)); |
1da177e4 LT |
350 | req->errors++; |
351 | return req; | |
352 | } | |
353 | ||
d18509f5 | 354 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req); |
1da177e4 | 355 | if (nbd_cmd(req) == NBD_CMD_READ) { |
5705f702 | 356 | struct req_iterator iter; |
7988613b | 357 | struct bio_vec bvec; |
5705f702 N |
358 | |
359 | rq_for_each_segment(bvec, req, iter) { | |
7988613b | 360 | result = sock_recv_bvec(nbd, &bvec); |
6c92e699 | 361 | if (result <= 0) { |
f4507164 | 362 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 363 | result); |
6c92e699 JA |
364 | req->errors++; |
365 | return req; | |
366 | } | |
d18509f5 MP |
367 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
368 | req, bvec.bv_len); | |
1da177e4 LT |
369 | } |
370 | } | |
371 | return req; | |
372 | harderror: | |
f4507164 | 373 | nbd->harderror = result; |
1da177e4 LT |
374 | return NULL; |
375 | } | |
376 | ||
edfaa7c3 KS |
377 | static ssize_t pid_show(struct device *dev, |
378 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 379 | { |
edfaa7c3 KS |
380 | struct gendisk *disk = dev_to_disk(dev); |
381 | ||
382 | return sprintf(buf, "%ld\n", | |
6b39bb65 PC |
383 | (long) ((struct nbd_device *)disk->private_data)->pid); |
384 | } | |
385 | ||
edfaa7c3 | 386 | static struct device_attribute pid_attr = { |
01e8ef11 | 387 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
388 | .show = pid_show, |
389 | }; | |
390 | ||
f4507164 | 391 | static int nbd_do_it(struct nbd_device *nbd) |
1da177e4 LT |
392 | { |
393 | struct request *req; | |
84963048 | 394 | int ret; |
1da177e4 | 395 | |
f4507164 | 396 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 397 | |
7f338fe4 | 398 | sk_set_memalloc(nbd->sock->sk); |
f4507164 WG |
399 | nbd->pid = task_pid_nr(current); |
400 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); | |
84963048 | 401 | if (ret) { |
f4507164 WG |
402 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
403 | nbd->pid = 0; | |
84963048 WC |
404 | return ret; |
405 | } | |
6b39bb65 | 406 | |
f4507164 | 407 | while ((req = nbd_read_stat(nbd)) != NULL) |
d18509f5 | 408 | nbd_end_request(nbd, req); |
6b39bb65 | 409 | |
f4507164 WG |
410 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
411 | nbd->pid = 0; | |
84963048 | 412 | return 0; |
1da177e4 LT |
413 | } |
414 | ||
f4507164 | 415 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
416 | { |
417 | struct request *req; | |
418 | ||
f4507164 | 419 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 420 | |
4b2f0260 | 421 | /* |
f4507164 | 422 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
423 | * modifications to the list must have completed by now. For |
424 | * the same reason, the active_req must be NULL. | |
425 | * | |
426 | * As a consequence, we don't need to take the spin lock while | |
427 | * purging the list here. | |
428 | */ | |
f4507164 WG |
429 | BUG_ON(nbd->sock); |
430 | BUG_ON(nbd->active_req); | |
4b2f0260 | 431 | |
f4507164 WG |
432 | while (!list_empty(&nbd->queue_head)) { |
433 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
434 | queuelist); |
435 | list_del_init(&req->queuelist); | |
436 | req->errors++; | |
d18509f5 | 437 | nbd_end_request(nbd, req); |
4b2f0260 | 438 | } |
fded4e09 PC |
439 | |
440 | while (!list_empty(&nbd->waiting_queue)) { | |
441 | req = list_entry(nbd->waiting_queue.next, struct request, | |
442 | queuelist); | |
443 | list_del_init(&req->queuelist); | |
444 | req->errors++; | |
d18509f5 | 445 | nbd_end_request(nbd, req); |
fded4e09 | 446 | } |
1da177e4 LT |
447 | } |
448 | ||
7fdfd406 | 449 | |
f4507164 | 450 | static void nbd_handle_req(struct nbd_device *nbd, struct request *req) |
48cf6061 | 451 | { |
33659ebb | 452 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
453 | goto error_out; |
454 | ||
455 | nbd_cmd(req) = NBD_CMD_READ; | |
456 | if (rq_data_dir(req) == WRITE) { | |
a336d298 PC |
457 | if ((req->cmd_flags & REQ_DISCARD)) { |
458 | WARN_ON(!(nbd->flags & NBD_FLAG_SEND_TRIM)); | |
459 | nbd_cmd(req) = NBD_CMD_TRIM; | |
460 | } else | |
461 | nbd_cmd(req) = NBD_CMD_WRITE; | |
2f012508 | 462 | if (nbd->flags & NBD_FLAG_READ_ONLY) { |
f4507164 | 463 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 464 | "Write on read-only\n"); |
48cf6061 LV |
465 | goto error_out; |
466 | } | |
467 | } | |
468 | ||
75f187ab AB |
469 | if (req->cmd_flags & REQ_FLUSH) { |
470 | BUG_ON(unlikely(blk_rq_sectors(req))); | |
471 | nbd_cmd(req) = NBD_CMD_FLUSH; | |
472 | } | |
473 | ||
48cf6061 LV |
474 | req->errors = 0; |
475 | ||
f4507164 WG |
476 | mutex_lock(&nbd->tx_lock); |
477 | if (unlikely(!nbd->sock)) { | |
478 | mutex_unlock(&nbd->tx_lock); | |
479 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 480 | "Attempted send on closed socket\n"); |
15746fca | 481 | goto error_out; |
48cf6061 LV |
482 | } |
483 | ||
f4507164 | 484 | nbd->active_req = req; |
48cf6061 | 485 | |
f4507164 WG |
486 | if (nbd_send_req(nbd, req) != 0) { |
487 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 | 488 | req->errors++; |
d18509f5 | 489 | nbd_end_request(nbd, req); |
48cf6061 | 490 | } else { |
f4507164 | 491 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 492 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 493 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
494 | } |
495 | ||
f4507164 WG |
496 | nbd->active_req = NULL; |
497 | mutex_unlock(&nbd->tx_lock); | |
498 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
499 | |
500 | return; | |
501 | ||
502 | error_out: | |
503 | req->errors++; | |
d18509f5 | 504 | nbd_end_request(nbd, req); |
48cf6061 LV |
505 | } |
506 | ||
507 | static int nbd_thread(void *data) | |
508 | { | |
f4507164 | 509 | struct nbd_device *nbd = data; |
48cf6061 LV |
510 | struct request *req; |
511 | ||
8698a745 | 512 | set_user_nice(current, MIN_NICE); |
f4507164 | 513 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 514 | /* wait for something to do */ |
f4507164 | 515 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 516 | kthread_should_stop() || |
f4507164 | 517 | !list_empty(&nbd->waiting_queue)); |
48cf6061 LV |
518 | |
519 | /* extract request */ | |
f4507164 | 520 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
521 | continue; |
522 | ||
f4507164 WG |
523 | spin_lock_irq(&nbd->queue_lock); |
524 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
525 | queuelist); |
526 | list_del_init(&req->queuelist); | |
f4507164 | 527 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
528 | |
529 | /* handle request */ | |
f4507164 | 530 | nbd_handle_req(nbd, req); |
48cf6061 LV |
531 | } |
532 | return 0; | |
533 | } | |
534 | ||
1da177e4 LT |
535 | /* |
536 | * We always wait for result of write, for now. It would be nice to make it optional | |
537 | * in future | |
f4507164 | 538 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
539 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
540 | */ | |
541 | ||
15746fca | 542 | static void do_nbd_request(struct request_queue *q) |
398eb085 | 543 | __releases(q->queue_lock) __acquires(q->queue_lock) |
1da177e4 LT |
544 | { |
545 | struct request *req; | |
546 | ||
9934c8c0 | 547 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 548 | struct nbd_device *nbd; |
1da177e4 | 549 | |
48cf6061 LV |
550 | spin_unlock_irq(q->queue_lock); |
551 | ||
f4507164 | 552 | nbd = req->rq_disk->private_data; |
1da177e4 | 553 | |
f4507164 | 554 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 555 | |
d18509f5 MP |
556 | dev_dbg(nbd_to_dev(nbd), "request %p: dequeued (flags=%x)\n", |
557 | req, req->cmd_type); | |
558 | ||
f4507164 WG |
559 | if (unlikely(!nbd->sock)) { |
560 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 561 | "Attempted send on closed socket\n"); |
4d48a542 | 562 | req->errors++; |
d18509f5 | 563 | nbd_end_request(nbd, req); |
4d48a542 PC |
564 | spin_lock_irq(q->queue_lock); |
565 | continue; | |
566 | } | |
567 | ||
f4507164 WG |
568 | spin_lock_irq(&nbd->queue_lock); |
569 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
570 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 571 | |
f4507164 | 572 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 573 | |
1da177e4 | 574 | spin_lock_irq(q->queue_lock); |
1da177e4 | 575 | } |
1da177e4 LT |
576 | } |
577 | ||
1a2ad211 | 578 | /* Must be called with tx_lock held */ |
1da177e4 | 579 | |
f4507164 | 580 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
581 | unsigned int cmd, unsigned long arg) |
582 | { | |
1da177e4 | 583 | switch (cmd) { |
1a2ad211 PM |
584 | case NBD_DISCONNECT: { |
585 | struct request sreq; | |
586 | ||
f4507164 | 587 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
3a2d63f8 PB |
588 | if (!nbd->sock) |
589 | return -EINVAL; | |
1a2ad211 | 590 | |
3a2d63f8 PB |
591 | mutex_unlock(&nbd->tx_lock); |
592 | fsync_bdev(bdev); | |
593 | mutex_lock(&nbd->tx_lock); | |
4f54eec8 | 594 | blk_rq_init(NULL, &sreq); |
4aff5e23 | 595 | sreq.cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 | 596 | nbd_cmd(&sreq) = NBD_CMD_DISC; |
3a2d63f8 PB |
597 | |
598 | /* Check again after getting mutex back. */ | |
f4507164 | 599 | if (!nbd->sock) |
1da177e4 | 600 | return -EINVAL; |
3a2d63f8 | 601 | |
c378f70a PC |
602 | nbd->disconnect = 1; |
603 | ||
f4507164 | 604 | nbd_send_req(nbd, &sreq); |
c378f70a | 605 | return 0; |
1a2ad211 | 606 | } |
1da177e4 | 607 | |
1a2ad211 | 608 | case NBD_CLEAR_SOCK: { |
e2511578 | 609 | struct socket *sock = nbd->sock; |
f4507164 | 610 | nbd->sock = NULL; |
f4507164 WG |
611 | nbd_clear_que(nbd); |
612 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 613 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
3a2d63f8 | 614 | kill_bdev(bdev); |
e2511578 AV |
615 | if (sock) |
616 | sockfd_put(sock); | |
1a2ad211 PM |
617 | return 0; |
618 | } | |
619 | ||
620 | case NBD_SET_SOCK: { | |
e2511578 AV |
621 | struct socket *sock; |
622 | int err; | |
623 | if (nbd->sock) | |
1da177e4 | 624 | return -EBUSY; |
e2511578 AV |
625 | sock = sockfd_lookup(arg, &err); |
626 | if (sock) { | |
627 | nbd->sock = sock; | |
628 | if (max_part > 0) | |
629 | bdev->bd_invalidated = 1; | |
630 | nbd->disconnect = 0; /* we're connected now */ | |
631 | return 0; | |
1da177e4 | 632 | } |
1a2ad211 PM |
633 | return -EINVAL; |
634 | } | |
635 | ||
1da177e4 | 636 | case NBD_SET_BLKSIZE: |
f4507164 WG |
637 | nbd->blksize = arg; |
638 | nbd->bytesize &= ~(nbd->blksize-1); | |
639 | bdev->bd_inode->i_size = nbd->bytesize; | |
640 | set_blocksize(bdev, nbd->blksize); | |
641 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 642 | return 0; |
1a2ad211 | 643 | |
1da177e4 | 644 | case NBD_SET_SIZE: |
f4507164 WG |
645 | nbd->bytesize = arg & ~(nbd->blksize-1); |
646 | bdev->bd_inode->i_size = nbd->bytesize; | |
647 | set_blocksize(bdev, nbd->blksize); | |
648 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 649 | return 0; |
1a2ad211 | 650 | |
7fdfd406 | 651 | case NBD_SET_TIMEOUT: |
f4507164 | 652 | nbd->xmit_timeout = arg * HZ; |
7fdfd406 | 653 | return 0; |
1a2ad211 | 654 | |
2f012508 PC |
655 | case NBD_SET_FLAGS: |
656 | nbd->flags = arg; | |
657 | return 0; | |
658 | ||
1da177e4 | 659 | case NBD_SET_SIZE_BLOCKS: |
f4507164 WG |
660 | nbd->bytesize = ((u64) arg) * nbd->blksize; |
661 | bdev->bd_inode->i_size = nbd->bytesize; | |
662 | set_blocksize(bdev, nbd->blksize); | |
663 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 664 | return 0; |
1a2ad211 PM |
665 | |
666 | case NBD_DO_IT: { | |
667 | struct task_struct *thread; | |
e2511578 | 668 | struct socket *sock; |
1a2ad211 PM |
669 | int error; |
670 | ||
f4507164 | 671 | if (nbd->pid) |
c91192d6 | 672 | return -EBUSY; |
e2511578 | 673 | if (!nbd->sock) |
1da177e4 | 674 | return -EINVAL; |
1a2ad211 | 675 | |
f4507164 | 676 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 677 | |
a83e814b PB |
678 | if (nbd->flags & NBD_FLAG_READ_ONLY) |
679 | set_device_ro(bdev, true); | |
a336d298 PC |
680 | if (nbd->flags & NBD_FLAG_SEND_TRIM) |
681 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
682 | nbd->disk->queue); | |
75f187ab AB |
683 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) |
684 | blk_queue_flush(nbd->disk->queue, REQ_FLUSH); | |
685 | else | |
686 | blk_queue_flush(nbd->disk->queue, 0); | |
a336d298 | 687 | |
d06df60b MP |
688 | thread = kthread_run(nbd_thread, nbd, "%s", |
689 | nbd->disk->disk_name); | |
1a2ad211 | 690 | if (IS_ERR(thread)) { |
f4507164 | 691 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 692 | return PTR_ERR(thread); |
1a2ad211 | 693 | } |
d06df60b | 694 | |
f4507164 | 695 | error = nbd_do_it(nbd); |
48cf6061 | 696 | kthread_stop(thread); |
1a2ad211 | 697 | |
f4507164 | 698 | mutex_lock(&nbd->tx_lock); |
84963048 WC |
699 | if (error) |
700 | return error; | |
f4507164 | 701 | sock_shutdown(nbd, 0); |
e2511578 AV |
702 | sock = nbd->sock; |
703 | nbd->sock = NULL; | |
f4507164 WG |
704 | nbd_clear_que(nbd); |
705 | dev_warn(disk_to_dev(nbd->disk), "queue cleared\n"); | |
3a2d63f8 | 706 | kill_bdev(bdev); |
a336d298 | 707 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
a83e814b | 708 | set_device_ro(bdev, false); |
e2511578 AV |
709 | if (sock) |
710 | sockfd_put(sock); | |
75f187ab | 711 | nbd->flags = 0; |
f4507164 | 712 | nbd->bytesize = 0; |
a8cdc308 | 713 | bdev->bd_inode->i_size = 0; |
f4507164 | 714 | set_capacity(nbd->disk, 0); |
d71a6d73 | 715 | if (max_part > 0) |
a8cdc308 | 716 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
c378f70a PC |
717 | if (nbd->disconnect) /* user requested, ignore socket errors */ |
718 | return 0; | |
f4507164 | 719 | return nbd->harderror; |
1a2ad211 PM |
720 | } |
721 | ||
1da177e4 | 722 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
723 | /* |
724 | * This is for compatibility only. The queue is always cleared | |
725 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
726 | */ | |
1da177e4 | 727 | return 0; |
1a2ad211 | 728 | |
1da177e4 | 729 | case NBD_PRINT_DEBUG: |
f4507164 | 730 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 731 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
732 | nbd->queue_head.next, nbd->queue_head.prev, |
733 | &nbd->queue_head); | |
1da177e4 LT |
734 | return 0; |
735 | } | |
1a2ad211 PM |
736 | return -ENOTTY; |
737 | } | |
738 | ||
739 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
740 | unsigned int cmd, unsigned long arg) | |
741 | { | |
f4507164 | 742 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
743 | int error; |
744 | ||
745 | if (!capable(CAP_SYS_ADMIN)) | |
746 | return -EPERM; | |
747 | ||
f4507164 | 748 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 | 749 | |
f4507164 WG |
750 | mutex_lock(&nbd->tx_lock); |
751 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
752 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
753 | |
754 | return error; | |
1da177e4 LT |
755 | } |
756 | ||
83d5cde4 | 757 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
758 | { |
759 | .owner = THIS_MODULE, | |
8a6cfeb6 | 760 | .ioctl = nbd_ioctl, |
1da177e4 LT |
761 | }; |
762 | ||
763 | /* | |
764 | * And here should be modules and kernel interface | |
765 | * (Just smiley confuses emacs :-) | |
766 | */ | |
767 | ||
768 | static int __init nbd_init(void) | |
769 | { | |
770 | int err = -ENOMEM; | |
771 | int i; | |
d71a6d73 | 772 | int part_shift; |
1da177e4 | 773 | |
5b7b18cc | 774 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 775 | |
d71a6d73 | 776 | if (max_part < 0) { |
7742ce4a | 777 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
778 | return -EINVAL; |
779 | } | |
780 | ||
781 | part_shift = 0; | |
5988ce23 | 782 | if (max_part > 0) { |
d71a6d73 LV |
783 | part_shift = fls(max_part); |
784 | ||
5988ce23 NK |
785 | /* |
786 | * Adjust max_part according to part_shift as it is exported | |
787 | * to user space so that user can know the max number of | |
788 | * partition kernel should be able to manage. | |
789 | * | |
790 | * Note that -1 is required because partition 0 is reserved | |
791 | * for the whole disk. | |
792 | */ | |
793 | max_part = (1UL << part_shift) - 1; | |
794 | } | |
795 | ||
3b271082 NK |
796 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
797 | return -EINVAL; | |
798 | ||
799 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
800 | return -EINVAL; | |
801 | ||
ff6b8090 SM |
802 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
803 | if (!nbd_dev) | |
804 | return -ENOMEM; | |
805 | ||
40be0c28 | 806 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 807 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
808 | if (!disk) |
809 | goto out; | |
810 | nbd_dev[i].disk = disk; | |
811 | /* | |
812 | * The new linux 2.5 block layer implementation requires | |
813 | * every gendisk to have its very own request_queue struct. | |
814 | * These structs are big so we dynamically allocate them. | |
815 | */ | |
816 | disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); | |
817 | if (!disk->queue) { | |
818 | put_disk(disk); | |
819 | goto out; | |
820 | } | |
31dcfab0 JA |
821 | /* |
822 | * Tell the block layer that we are not a rotational device | |
823 | */ | |
824 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
b277da0a | 825 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); |
a336d298 PC |
826 | disk->queue->limits.discard_granularity = 512; |
827 | disk->queue->limits.max_discard_sectors = UINT_MAX; | |
828 | disk->queue->limits.discard_zeroes_data = 0; | |
078be02b MB |
829 | blk_queue_max_hw_sectors(disk->queue, 65536); |
830 | disk->queue->limits.max_sectors = 256; | |
1da177e4 LT |
831 | } |
832 | ||
833 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
834 | err = -EIO; | |
835 | goto out; | |
836 | } | |
837 | ||
838 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
1da177e4 | 839 | |
40be0c28 | 840 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 841 | struct gendisk *disk = nbd_dev[i].disk; |
f4507164 | 842 | nbd_dev[i].magic = NBD_MAGIC; |
48cf6061 | 843 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 LT |
844 | spin_lock_init(&nbd_dev[i].queue_lock); |
845 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 846 | mutex_init(&nbd_dev[i].tx_lock); |
4b2f0260 | 847 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 848 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 849 | nbd_dev[i].blksize = 1024; |
4b86a872 | 850 | nbd_dev[i].bytesize = 0; |
1da177e4 | 851 | disk->major = NBD_MAJOR; |
d71a6d73 | 852 | disk->first_minor = i << part_shift; |
1da177e4 LT |
853 | disk->fops = &nbd_fops; |
854 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 855 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 856 | set_capacity(disk, 0); |
1da177e4 LT |
857 | add_disk(disk); |
858 | } | |
859 | ||
860 | return 0; | |
861 | out: | |
862 | while (i--) { | |
863 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
864 | put_disk(nbd_dev[i].disk); | |
865 | } | |
f3944d61 | 866 | kfree(nbd_dev); |
1da177e4 LT |
867 | return err; |
868 | } | |
869 | ||
870 | static void __exit nbd_cleanup(void) | |
871 | { | |
872 | int i; | |
40be0c28 | 873 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 874 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 875 | nbd_dev[i].magic = 0; |
1da177e4 LT |
876 | if (disk) { |
877 | del_gendisk(disk); | |
878 | blk_cleanup_queue(disk->queue); | |
879 | put_disk(disk); | |
880 | } | |
881 | } | |
1da177e4 | 882 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 883 | kfree(nbd_dev); |
1da177e4 LT |
884 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
885 | } | |
886 | ||
887 | module_init(nbd_init); | |
888 | module_exit(nbd_cleanup); | |
889 | ||
890 | MODULE_DESCRIPTION("Network Block Device"); | |
891 | MODULE_LICENSE("GPL"); | |
892 | ||
40be0c28 | 893 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
894 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
895 | module_param(max_part, int, 0444); | |
896 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |