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; | |
13e71d69 MP |
44 | struct socket * sock; /* If == NULL, device is not ready, yet */ |
45 | int magic; | |
46 | ||
47 | spinlock_t queue_lock; | |
48 | struct list_head queue_head; /* Requests waiting result */ | |
49 | struct request *active_req; | |
50 | wait_queue_head_t active_wq; | |
51 | struct list_head waiting_queue; /* Requests to be sent */ | |
52 | wait_queue_head_t waiting_wq; | |
53 | ||
54 | struct mutex tx_lock; | |
55 | struct gendisk *disk; | |
56 | int blksize; | |
b9c495bb | 57 | loff_t bytesize; |
13e71d69 MP |
58 | int xmit_timeout; |
59 | int disconnect; /* a disconnect has been requested by user */ | |
7e2893a1 MP |
60 | |
61 | struct timer_list timeout_timer; | |
62 | struct task_struct *task_recv; | |
63 | struct task_struct *task_send; | |
13e71d69 MP |
64 | }; |
65 | ||
f4507164 | 66 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 67 | |
9c7a4169 | 68 | static unsigned int nbds_max = 16; |
20a8143e | 69 | static struct nbd_device *nbd_dev; |
d71a6d73 | 70 | static int max_part; |
1da177e4 LT |
71 | |
72 | /* | |
73 | * Use just one lock (or at most 1 per NIC). Two arguments for this: | |
74 | * 1. Each NIC is essentially a synchronization point for all servers | |
75 | * accessed through that NIC so there's no need to have more locks | |
76 | * than NICs anyway. | |
77 | * 2. More locks lead to more "Dirty cache line bouncing" which will slow | |
78 | * down each lock to the point where they're actually slower than just | |
79 | * a single lock. | |
80 | * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! | |
81 | */ | |
82 | static DEFINE_SPINLOCK(nbd_lock); | |
83 | ||
d18509f5 | 84 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 85 | { |
d18509f5 | 86 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
87 | } |
88 | ||
89 | static 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"; | |
75f187ab | 95 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 96 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
97 | } |
98 | return "invalid"; | |
99 | } | |
1da177e4 | 100 | |
d18509f5 | 101 | static void nbd_end_request(struct nbd_device *nbd, struct request *req) |
1da177e4 | 102 | { |
097c94a4 | 103 | int error = req->errors ? -EIO : 0; |
165125e1 | 104 | struct request_queue *q = req->q; |
1da177e4 LT |
105 | unsigned long flags; |
106 | ||
d18509f5 MP |
107 | dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", req, |
108 | error ? "failed" : "done"); | |
1da177e4 LT |
109 | |
110 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 111 | __blk_end_request_all(req, error); |
1da177e4 LT |
112 | spin_unlock_irqrestore(q->queue_lock, flags); |
113 | } | |
114 | ||
e018e757 MP |
115 | /* |
116 | * Forcibly shutdown the socket causing all listeners to error | |
117 | */ | |
36e47bee | 118 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 119 | { |
260bbce4 MP |
120 | if (!nbd->sock) |
121 | return; | |
122 | ||
123 | dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n"); | |
124 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
125 | nbd->sock = NULL; | |
126 | del_timer_sync(&nbd->timeout_timer); | |
7fdfd406 PC |
127 | } |
128 | ||
129 | static void nbd_xmit_timeout(unsigned long arg) | |
130 | { | |
7e2893a1 MP |
131 | struct nbd_device *nbd = (struct nbd_device *)arg; |
132 | struct task_struct *task; | |
133 | ||
134 | if (list_empty(&nbd->queue_head)) | |
135 | return; | |
136 | ||
137 | nbd->disconnect = 1; | |
138 | ||
139 | task = READ_ONCE(nbd->task_recv); | |
140 | if (task) | |
141 | force_sig(SIGKILL, task); | |
7fdfd406 | 142 | |
7e2893a1 MP |
143 | task = READ_ONCE(nbd->task_send); |
144 | if (task) | |
145 | force_sig(SIGKILL, nbd->task_send); | |
146 | ||
147 | dev_err(nbd_to_dev(nbd), "Connection timed out, killed receiver and sender, shutting down connection\n"); | |
7fdfd406 PC |
148 | } |
149 | ||
1da177e4 LT |
150 | /* |
151 | * Send or receive packet. | |
152 | */ | |
f4507164 | 153 | static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size, |
1da177e4 LT |
154 | int msg_flags) |
155 | { | |
f4507164 | 156 | struct socket *sock = nbd->sock; |
1da177e4 LT |
157 | int result; |
158 | struct msghdr msg; | |
159 | struct kvec iov; | |
be0ef957 | 160 | sigset_t blocked, oldset; |
7f338fe4 | 161 | unsigned long pflags = current->flags; |
1da177e4 | 162 | |
ffc41cf8 | 163 | if (unlikely(!sock)) { |
f4507164 | 164 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
165 | "Attempted %s on closed socket in sock_xmit\n", |
166 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
167 | return -EINVAL; |
168 | } | |
169 | ||
1da177e4 LT |
170 | /* Allow interception of SIGKILL only |
171 | * Don't allow other signals to interrupt the transmission */ | |
be0ef957 ON |
172 | siginitsetinv(&blocked, sigmask(SIGKILL)); |
173 | sigprocmask(SIG_SETMASK, &blocked, &oldset); | |
1da177e4 | 174 | |
7f338fe4 | 175 | current->flags |= PF_MEMALLOC; |
1da177e4 | 176 | do { |
7f338fe4 | 177 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
178 | iov.iov_base = buf; |
179 | iov.iov_len = size; | |
180 | msg.msg_name = NULL; | |
181 | msg.msg_namelen = 0; | |
182 | msg.msg_control = NULL; | |
183 | msg.msg_controllen = 0; | |
1da177e4 LT |
184 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
185 | ||
7e2893a1 | 186 | if (send) |
1da177e4 | 187 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
7e2893a1 | 188 | else |
35fbf5bc NK |
189 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
190 | msg.msg_flags); | |
1da177e4 | 191 | |
1da177e4 LT |
192 | if (result <= 0) { |
193 | if (result == 0) | |
194 | result = -EPIPE; /* short read */ | |
195 | break; | |
196 | } | |
197 | size -= result; | |
198 | buf += result; | |
199 | } while (size > 0); | |
200 | ||
be0ef957 | 201 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
7f338fe4 | 202 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 | 203 | |
7e2893a1 MP |
204 | if (!send && nbd->xmit_timeout) |
205 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
206 | ||
1da177e4 LT |
207 | return result; |
208 | } | |
209 | ||
f4507164 | 210 | static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec, |
1da177e4 LT |
211 | int flags) |
212 | { | |
213 | int result; | |
214 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 WG |
215 | result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset, |
216 | bvec->bv_len, flags); | |
1da177e4 LT |
217 | kunmap(bvec->bv_page); |
218 | return result; | |
219 | } | |
220 | ||
7fdfd406 | 221 | /* always call with the tx_lock held */ |
f4507164 | 222 | static int nbd_send_req(struct nbd_device *nbd, struct request *req) |
1da177e4 | 223 | { |
5705f702 | 224 | int result, flags; |
1da177e4 | 225 | struct nbd_request request; |
1011c1b9 | 226 | unsigned long size = blk_rq_bytes(req); |
9dc6c806 CH |
227 | u32 type; |
228 | ||
229 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) | |
230 | type = NBD_CMD_DISC; | |
231 | else if (req->cmd_flags & REQ_DISCARD) | |
232 | type = NBD_CMD_TRIM; | |
233 | else if (req->cmd_flags & REQ_FLUSH) | |
234 | type = NBD_CMD_FLUSH; | |
235 | else if (rq_data_dir(req) == WRITE) | |
236 | type = NBD_CMD_WRITE; | |
237 | else | |
238 | type = NBD_CMD_READ; | |
1da177e4 | 239 | |
04cfac4e | 240 | memset(&request, 0, sizeof(request)); |
1da177e4 | 241 | request.magic = htonl(NBD_REQUEST_MAGIC); |
9dc6c806 CH |
242 | request.type = htonl(type); |
243 | if (type != NBD_CMD_FLUSH && type != NBD_CMD_DISC) { | |
75f187ab AB |
244 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
245 | request.len = htonl(size); | |
246 | } | |
1da177e4 LT |
247 | memcpy(request.handle, &req, sizeof(req)); |
248 | ||
d18509f5 | 249 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
9dc6c806 | 250 | req, nbdcmd_to_ascii(type), |
d18509f5 | 251 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
f4507164 | 252 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
9dc6c806 | 253 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 254 | if (result <= 0) { |
f4507164 | 255 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 256 | "Send control failed (result %d)\n", result); |
dab5313a | 257 | return -EIO; |
1da177e4 LT |
258 | } |
259 | ||
9dc6c806 | 260 | if (type == NBD_CMD_WRITE) { |
5705f702 | 261 | struct req_iterator iter; |
7988613b | 262 | struct bio_vec bvec; |
1da177e4 LT |
263 | /* |
264 | * we are really probing at internals to determine | |
265 | * whether to set MSG_MORE or not... | |
266 | */ | |
5705f702 | 267 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 | 268 | flags = 0; |
4550dd6c | 269 | if (!rq_iter_last(bvec, iter)) |
6c92e699 | 270 | flags = MSG_MORE; |
d18509f5 MP |
271 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
272 | req, bvec.bv_len); | |
7988613b | 273 | result = sock_send_bvec(nbd, &bvec, flags); |
6c92e699 | 274 | if (result <= 0) { |
f4507164 | 275 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
276 | "Send data failed (result %d)\n", |
277 | result); | |
dab5313a | 278 | return -EIO; |
6c92e699 | 279 | } |
1da177e4 LT |
280 | } |
281 | } | |
1da177e4 | 282 | return 0; |
1da177e4 LT |
283 | } |
284 | ||
f4507164 | 285 | static struct request *nbd_find_request(struct nbd_device *nbd, |
0cbc591b | 286 | struct request *xreq) |
1da177e4 | 287 | { |
d2c9740b | 288 | struct request *req, *tmp; |
4b2f0260 | 289 | int err; |
1da177e4 | 290 | |
f4507164 | 291 | err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq); |
4b2f0260 | 292 | if (unlikely(err)) |
de9ad6d4 | 293 | return ERR_PTR(err); |
4b2f0260 | 294 | |
f4507164 WG |
295 | spin_lock(&nbd->queue_lock); |
296 | list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) { | |
1da177e4 LT |
297 | if (req != xreq) |
298 | continue; | |
299 | list_del_init(&req->queuelist); | |
f4507164 | 300 | spin_unlock(&nbd->queue_lock); |
1da177e4 LT |
301 | return req; |
302 | } | |
f4507164 | 303 | spin_unlock(&nbd->queue_lock); |
4b2f0260 | 304 | |
de9ad6d4 | 305 | return ERR_PTR(-ENOENT); |
1da177e4 LT |
306 | } |
307 | ||
f4507164 | 308 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
309 | { |
310 | int result; | |
311 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 312 | result = sock_xmit(nbd, 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 */ | |
f4507164 | 319 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
320 | { |
321 | int result; | |
322 | struct nbd_reply reply; | |
323 | struct request *req; | |
1da177e4 LT |
324 | |
325 | reply.magic = 0; | |
f4507164 | 326 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 327 | if (result <= 0) { |
f4507164 | 328 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 329 | "Receive control failed (result %d)\n", result); |
19391830 | 330 | return ERR_PTR(result); |
1da177e4 | 331 | } |
e4b57e08 MF |
332 | |
333 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 334 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 335 | (unsigned long)ntohl(reply.magic)); |
19391830 | 336 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
337 | } |
338 | ||
f4507164 | 339 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 340 | if (IS_ERR(req)) { |
4b2f0260 HX |
341 | result = PTR_ERR(req); |
342 | if (result != -ENOENT) | |
19391830 | 343 | return ERR_PTR(result); |
4b2f0260 | 344 | |
f4507164 | 345 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 346 | reply.handle); |
19391830 | 347 | return ERR_PTR(-EBADR); |
1da177e4 LT |
348 | } |
349 | ||
1da177e4 | 350 | if (ntohl(reply.error)) { |
f4507164 | 351 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 352 | ntohl(reply.error)); |
1da177e4 LT |
353 | req->errors++; |
354 | return req; | |
355 | } | |
356 | ||
d18509f5 | 357 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req); |
9dc6c806 | 358 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 359 | struct req_iterator iter; |
7988613b | 360 | struct bio_vec bvec; |
5705f702 N |
361 | |
362 | rq_for_each_segment(bvec, req, iter) { | |
7988613b | 363 | result = sock_recv_bvec(nbd, &bvec); |
6c92e699 | 364 | if (result <= 0) { |
f4507164 | 365 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 366 | result); |
6c92e699 JA |
367 | req->errors++; |
368 | return req; | |
369 | } | |
d18509f5 MP |
370 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
371 | req, bvec.bv_len); | |
1da177e4 LT |
372 | } |
373 | } | |
374 | return req; | |
1da177e4 LT |
375 | } |
376 | ||
edfaa7c3 KS |
377 | static ssize_t pid_show(struct device *dev, |
378 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 379 | { |
edfaa7c3 | 380 | struct gendisk *disk = dev_to_disk(dev); |
6521d39a | 381 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; |
edfaa7c3 | 382 | |
6521d39a | 383 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); |
6b39bb65 PC |
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); |
6521d39a MP |
399 | |
400 | nbd->task_recv = current; | |
401 | ||
f4507164 | 402 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
84963048 | 403 | if (ret) { |
f4507164 | 404 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
6521d39a | 405 | nbd->task_recv = NULL; |
84963048 WC |
406 | return ret; |
407 | } | |
6b39bb65 | 408 | |
19391830 MP |
409 | while (1) { |
410 | req = nbd_read_stat(nbd); | |
411 | if (IS_ERR(req)) { | |
412 | ret = PTR_ERR(req); | |
413 | break; | |
414 | } | |
415 | ||
d18509f5 | 416 | nbd_end_request(nbd, req); |
19391830 | 417 | } |
6b39bb65 | 418 | |
6521d39a MP |
419 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
420 | ||
7e2893a1 MP |
421 | nbd->task_recv = NULL; |
422 | ||
423 | if (signal_pending(current)) { | |
424 | siginfo_t info; | |
425 | ||
426 | ret = dequeue_signal_lock(current, ¤t->blocked, &info); | |
427 | dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n", | |
428 | task_pid_nr(current), current->comm, ret); | |
36e47bee MP |
429 | mutex_lock(&nbd->tx_lock); |
430 | sock_shutdown(nbd); | |
431 | mutex_unlock(&nbd->tx_lock); | |
7e2893a1 MP |
432 | ret = -ETIMEDOUT; |
433 | } | |
434 | ||
7e2893a1 | 435 | return ret; |
1da177e4 LT |
436 | } |
437 | ||
f4507164 | 438 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
439 | { |
440 | struct request *req; | |
441 | ||
f4507164 | 442 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 443 | |
4b2f0260 | 444 | /* |
f4507164 | 445 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
446 | * modifications to the list must have completed by now. For |
447 | * the same reason, the active_req must be NULL. | |
448 | * | |
449 | * As a consequence, we don't need to take the spin lock while | |
450 | * purging the list here. | |
451 | */ | |
f4507164 WG |
452 | BUG_ON(nbd->sock); |
453 | BUG_ON(nbd->active_req); | |
4b2f0260 | 454 | |
f4507164 WG |
455 | while (!list_empty(&nbd->queue_head)) { |
456 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
457 | queuelist); |
458 | list_del_init(&req->queuelist); | |
459 | req->errors++; | |
d18509f5 | 460 | nbd_end_request(nbd, req); |
4b2f0260 | 461 | } |
fded4e09 PC |
462 | |
463 | while (!list_empty(&nbd->waiting_queue)) { | |
464 | req = list_entry(nbd->waiting_queue.next, struct request, | |
465 | queuelist); | |
466 | list_del_init(&req->queuelist); | |
467 | req->errors++; | |
d18509f5 | 468 | nbd_end_request(nbd, req); |
fded4e09 | 469 | } |
e78273c8 | 470 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
471 | } |
472 | ||
7fdfd406 | 473 | |
f4507164 | 474 | static void nbd_handle_req(struct nbd_device *nbd, struct request *req) |
48cf6061 | 475 | { |
33659ebb | 476 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
477 | goto error_out; |
478 | ||
9dc6c806 CH |
479 | if (rq_data_dir(req) == WRITE && |
480 | (nbd->flags & NBD_FLAG_READ_ONLY)) { | |
481 | dev_err(disk_to_dev(nbd->disk), | |
482 | "Write on read-only\n"); | |
483 | goto error_out; | |
75f187ab AB |
484 | } |
485 | ||
48cf6061 LV |
486 | req->errors = 0; |
487 | ||
f4507164 WG |
488 | mutex_lock(&nbd->tx_lock); |
489 | if (unlikely(!nbd->sock)) { | |
490 | mutex_unlock(&nbd->tx_lock); | |
491 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 492 | "Attempted send on closed socket\n"); |
15746fca | 493 | goto error_out; |
48cf6061 LV |
494 | } |
495 | ||
f4507164 | 496 | nbd->active_req = req; |
48cf6061 | 497 | |
7e2893a1 MP |
498 | if (nbd->xmit_timeout && list_empty_careful(&nbd->queue_head)) |
499 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
500 | ||
f4507164 WG |
501 | if (nbd_send_req(nbd, req) != 0) { |
502 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 | 503 | req->errors++; |
d18509f5 | 504 | nbd_end_request(nbd, req); |
48cf6061 | 505 | } else { |
f4507164 | 506 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 507 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 508 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
509 | } |
510 | ||
f4507164 WG |
511 | nbd->active_req = NULL; |
512 | mutex_unlock(&nbd->tx_lock); | |
513 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
514 | |
515 | return; | |
516 | ||
517 | error_out: | |
518 | req->errors++; | |
d18509f5 | 519 | nbd_end_request(nbd, req); |
48cf6061 LV |
520 | } |
521 | ||
522 | static int nbd_thread(void *data) | |
523 | { | |
f4507164 | 524 | struct nbd_device *nbd = data; |
48cf6061 LV |
525 | struct request *req; |
526 | ||
7e2893a1 MP |
527 | nbd->task_send = current; |
528 | ||
8698a745 | 529 | set_user_nice(current, MIN_NICE); |
f4507164 | 530 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 531 | /* wait for something to do */ |
f4507164 | 532 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 533 | kthread_should_stop() || |
f4507164 | 534 | !list_empty(&nbd->waiting_queue)); |
48cf6061 | 535 | |
7e2893a1 MP |
536 | if (signal_pending(current)) { |
537 | siginfo_t info; | |
538 | int ret; | |
539 | ||
540 | ret = dequeue_signal_lock(current, ¤t->blocked, | |
541 | &info); | |
542 | dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n", | |
543 | task_pid_nr(current), current->comm, ret); | |
36e47bee MP |
544 | mutex_lock(&nbd->tx_lock); |
545 | sock_shutdown(nbd); | |
546 | mutex_unlock(&nbd->tx_lock); | |
7e2893a1 MP |
547 | break; |
548 | } | |
549 | ||
48cf6061 | 550 | /* extract request */ |
f4507164 | 551 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
552 | continue; |
553 | ||
f4507164 WG |
554 | spin_lock_irq(&nbd->queue_lock); |
555 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
556 | queuelist); |
557 | list_del_init(&req->queuelist); | |
f4507164 | 558 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
559 | |
560 | /* handle request */ | |
f4507164 | 561 | nbd_handle_req(nbd, req); |
48cf6061 | 562 | } |
7e2893a1 MP |
563 | |
564 | nbd->task_send = NULL; | |
565 | ||
48cf6061 LV |
566 | return 0; |
567 | } | |
568 | ||
1da177e4 LT |
569 | /* |
570 | * We always wait for result of write, for now. It would be nice to make it optional | |
571 | * in future | |
f4507164 | 572 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
573 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
574 | */ | |
575 | ||
15746fca | 576 | static void do_nbd_request(struct request_queue *q) |
398eb085 | 577 | __releases(q->queue_lock) __acquires(q->queue_lock) |
1da177e4 LT |
578 | { |
579 | struct request *req; | |
580 | ||
9934c8c0 | 581 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 582 | struct nbd_device *nbd; |
1da177e4 | 583 | |
48cf6061 LV |
584 | spin_unlock_irq(q->queue_lock); |
585 | ||
f4507164 | 586 | nbd = req->rq_disk->private_data; |
1da177e4 | 587 | |
f4507164 | 588 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 589 | |
d18509f5 MP |
590 | dev_dbg(nbd_to_dev(nbd), "request %p: dequeued (flags=%x)\n", |
591 | req, req->cmd_type); | |
592 | ||
f4507164 WG |
593 | if (unlikely(!nbd->sock)) { |
594 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 595 | "Attempted send on closed socket\n"); |
4d48a542 | 596 | req->errors++; |
d18509f5 | 597 | nbd_end_request(nbd, req); |
4d48a542 PC |
598 | spin_lock_irq(q->queue_lock); |
599 | continue; | |
600 | } | |
601 | ||
f4507164 WG |
602 | spin_lock_irq(&nbd->queue_lock); |
603 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
604 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 605 | |
f4507164 | 606 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 607 | |
1da177e4 | 608 | spin_lock_irq(q->queue_lock); |
1da177e4 | 609 | } |
1da177e4 LT |
610 | } |
611 | ||
1a2ad211 | 612 | /* Must be called with tx_lock held */ |
1da177e4 | 613 | |
f4507164 | 614 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
615 | unsigned int cmd, unsigned long arg) |
616 | { | |
1da177e4 | 617 | switch (cmd) { |
1a2ad211 PM |
618 | case NBD_DISCONNECT: { |
619 | struct request sreq; | |
620 | ||
f4507164 | 621 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
3a2d63f8 PB |
622 | if (!nbd->sock) |
623 | return -EINVAL; | |
1a2ad211 | 624 | |
3a2d63f8 PB |
625 | mutex_unlock(&nbd->tx_lock); |
626 | fsync_bdev(bdev); | |
627 | mutex_lock(&nbd->tx_lock); | |
4f54eec8 | 628 | blk_rq_init(NULL, &sreq); |
4f8c9510 | 629 | sreq.cmd_type = REQ_TYPE_DRV_PRIV; |
3a2d63f8 PB |
630 | |
631 | /* Check again after getting mutex back. */ | |
f4507164 | 632 | if (!nbd->sock) |
1da177e4 | 633 | return -EINVAL; |
3a2d63f8 | 634 | |
c378f70a PC |
635 | nbd->disconnect = 1; |
636 | ||
f4507164 | 637 | nbd_send_req(nbd, &sreq); |
c378f70a | 638 | return 0; |
1a2ad211 | 639 | } |
1da177e4 | 640 | |
1a2ad211 | 641 | case NBD_CLEAR_SOCK: { |
e2511578 | 642 | struct socket *sock = nbd->sock; |
f4507164 | 643 | nbd->sock = NULL; |
f4507164 WG |
644 | nbd_clear_que(nbd); |
645 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 646 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
3a2d63f8 | 647 | kill_bdev(bdev); |
e2511578 AV |
648 | if (sock) |
649 | sockfd_put(sock); | |
1a2ad211 PM |
650 | return 0; |
651 | } | |
652 | ||
653 | case NBD_SET_SOCK: { | |
e2511578 AV |
654 | struct socket *sock; |
655 | int err; | |
656 | if (nbd->sock) | |
1da177e4 | 657 | return -EBUSY; |
e2511578 AV |
658 | sock = sockfd_lookup(arg, &err); |
659 | if (sock) { | |
660 | nbd->sock = sock; | |
661 | if (max_part > 0) | |
662 | bdev->bd_invalidated = 1; | |
663 | nbd->disconnect = 0; /* we're connected now */ | |
664 | return 0; | |
1da177e4 | 665 | } |
1a2ad211 PM |
666 | return -EINVAL; |
667 | } | |
668 | ||
1da177e4 | 669 | case NBD_SET_BLKSIZE: |
f4507164 WG |
670 | nbd->blksize = arg; |
671 | nbd->bytesize &= ~(nbd->blksize-1); | |
672 | bdev->bd_inode->i_size = nbd->bytesize; | |
673 | set_blocksize(bdev, nbd->blksize); | |
674 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 675 | return 0; |
1a2ad211 | 676 | |
1da177e4 | 677 | case NBD_SET_SIZE: |
f4507164 WG |
678 | nbd->bytesize = arg & ~(nbd->blksize-1); |
679 | bdev->bd_inode->i_size = nbd->bytesize; | |
680 | set_blocksize(bdev, nbd->blksize); | |
681 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 682 | return 0; |
1a2ad211 | 683 | |
7fdfd406 | 684 | case NBD_SET_TIMEOUT: |
f4507164 | 685 | nbd->xmit_timeout = arg * HZ; |
7e2893a1 MP |
686 | if (arg) |
687 | mod_timer(&nbd->timeout_timer, | |
688 | jiffies + nbd->xmit_timeout); | |
689 | else | |
690 | del_timer_sync(&nbd->timeout_timer); | |
691 | ||
7fdfd406 | 692 | return 0; |
1a2ad211 | 693 | |
2f012508 PC |
694 | case NBD_SET_FLAGS: |
695 | nbd->flags = arg; | |
696 | return 0; | |
697 | ||
1da177e4 | 698 | case NBD_SET_SIZE_BLOCKS: |
f4507164 WG |
699 | nbd->bytesize = ((u64) arg) * nbd->blksize; |
700 | bdev->bd_inode->i_size = nbd->bytesize; | |
701 | set_blocksize(bdev, nbd->blksize); | |
702 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 703 | return 0; |
1a2ad211 PM |
704 | |
705 | case NBD_DO_IT: { | |
706 | struct task_struct *thread; | |
e2511578 | 707 | struct socket *sock; |
1a2ad211 PM |
708 | int error; |
709 | ||
6521d39a | 710 | if (nbd->task_recv) |
c91192d6 | 711 | return -EBUSY; |
e2511578 | 712 | if (!nbd->sock) |
1da177e4 | 713 | return -EINVAL; |
1a2ad211 | 714 | |
f4507164 | 715 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 716 | |
a83e814b PB |
717 | if (nbd->flags & NBD_FLAG_READ_ONLY) |
718 | set_device_ro(bdev, true); | |
a336d298 PC |
719 | if (nbd->flags & NBD_FLAG_SEND_TRIM) |
720 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
721 | nbd->disk->queue); | |
75f187ab AB |
722 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) |
723 | blk_queue_flush(nbd->disk->queue, REQ_FLUSH); | |
724 | else | |
725 | blk_queue_flush(nbd->disk->queue, 0); | |
a336d298 | 726 | |
d06df60b MP |
727 | thread = kthread_run(nbd_thread, nbd, "%s", |
728 | nbd->disk->disk_name); | |
1a2ad211 | 729 | if (IS_ERR(thread)) { |
f4507164 | 730 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 731 | return PTR_ERR(thread); |
1a2ad211 | 732 | } |
d06df60b | 733 | |
f4507164 | 734 | error = nbd_do_it(nbd); |
48cf6061 | 735 | kthread_stop(thread); |
1a2ad211 | 736 | |
f4507164 | 737 | mutex_lock(&nbd->tx_lock); |
19391830 | 738 | |
36e47bee | 739 | sock_shutdown(nbd); |
e2511578 AV |
740 | sock = nbd->sock; |
741 | nbd->sock = NULL; | |
f4507164 | 742 | nbd_clear_que(nbd); |
3a2d63f8 | 743 | kill_bdev(bdev); |
a336d298 | 744 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
a83e814b | 745 | set_device_ro(bdev, false); |
e2511578 AV |
746 | if (sock) |
747 | sockfd_put(sock); | |
75f187ab | 748 | nbd->flags = 0; |
f4507164 | 749 | nbd->bytesize = 0; |
a8cdc308 | 750 | bdev->bd_inode->i_size = 0; |
f4507164 | 751 | set_capacity(nbd->disk, 0); |
d71a6d73 | 752 | if (max_part > 0) |
9dcd1379 | 753 | blkdev_reread_part(bdev); |
c378f70a PC |
754 | if (nbd->disconnect) /* user requested, ignore socket errors */ |
755 | return 0; | |
19391830 | 756 | return error; |
1a2ad211 PM |
757 | } |
758 | ||
1da177e4 | 759 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
760 | /* |
761 | * This is for compatibility only. The queue is always cleared | |
762 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
763 | */ | |
1da177e4 | 764 | return 0; |
1a2ad211 | 765 | |
1da177e4 | 766 | case NBD_PRINT_DEBUG: |
f4507164 | 767 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 768 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
769 | nbd->queue_head.next, nbd->queue_head.prev, |
770 | &nbd->queue_head); | |
1da177e4 LT |
771 | return 0; |
772 | } | |
1a2ad211 PM |
773 | return -ENOTTY; |
774 | } | |
775 | ||
776 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
777 | unsigned int cmd, unsigned long arg) | |
778 | { | |
f4507164 | 779 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
780 | int error; |
781 | ||
782 | if (!capable(CAP_SYS_ADMIN)) | |
783 | return -EPERM; | |
784 | ||
f4507164 | 785 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 | 786 | |
f4507164 WG |
787 | mutex_lock(&nbd->tx_lock); |
788 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
789 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
790 | |
791 | return error; | |
1da177e4 LT |
792 | } |
793 | ||
83d5cde4 | 794 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
795 | { |
796 | .owner = THIS_MODULE, | |
8a6cfeb6 | 797 | .ioctl = nbd_ioctl, |
1da177e4 LT |
798 | }; |
799 | ||
800 | /* | |
801 | * And here should be modules and kernel interface | |
802 | * (Just smiley confuses emacs :-) | |
803 | */ | |
804 | ||
805 | static int __init nbd_init(void) | |
806 | { | |
807 | int err = -ENOMEM; | |
808 | int i; | |
d71a6d73 | 809 | int part_shift; |
1da177e4 | 810 | |
5b7b18cc | 811 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 812 | |
d71a6d73 | 813 | if (max_part < 0) { |
7742ce4a | 814 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
815 | return -EINVAL; |
816 | } | |
817 | ||
818 | part_shift = 0; | |
5988ce23 | 819 | if (max_part > 0) { |
d71a6d73 LV |
820 | part_shift = fls(max_part); |
821 | ||
5988ce23 NK |
822 | /* |
823 | * Adjust max_part according to part_shift as it is exported | |
824 | * to user space so that user can know the max number of | |
825 | * partition kernel should be able to manage. | |
826 | * | |
827 | * Note that -1 is required because partition 0 is reserved | |
828 | * for the whole disk. | |
829 | */ | |
830 | max_part = (1UL << part_shift) - 1; | |
831 | } | |
832 | ||
3b271082 NK |
833 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
834 | return -EINVAL; | |
835 | ||
836 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
837 | return -EINVAL; | |
838 | ||
ff6b8090 SM |
839 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
840 | if (!nbd_dev) | |
841 | return -ENOMEM; | |
842 | ||
40be0c28 | 843 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 844 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
845 | if (!disk) |
846 | goto out; | |
847 | nbd_dev[i].disk = disk; | |
848 | /* | |
849 | * The new linux 2.5 block layer implementation requires | |
850 | * every gendisk to have its very own request_queue struct. | |
851 | * These structs are big so we dynamically allocate them. | |
852 | */ | |
853 | disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); | |
854 | if (!disk->queue) { | |
855 | put_disk(disk); | |
856 | goto out; | |
857 | } | |
31dcfab0 JA |
858 | /* |
859 | * Tell the block layer that we are not a rotational device | |
860 | */ | |
861 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
b277da0a | 862 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); |
a336d298 | 863 | disk->queue->limits.discard_granularity = 512; |
2bb4cd5c | 864 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); |
a336d298 | 865 | disk->queue->limits.discard_zeroes_data = 0; |
078be02b MB |
866 | blk_queue_max_hw_sectors(disk->queue, 65536); |
867 | disk->queue->limits.max_sectors = 256; | |
1da177e4 LT |
868 | } |
869 | ||
870 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
871 | err = -EIO; | |
872 | goto out; | |
873 | } | |
874 | ||
875 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
1da177e4 | 876 | |
40be0c28 | 877 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 878 | struct gendisk *disk = nbd_dev[i].disk; |
f4507164 | 879 | nbd_dev[i].magic = NBD_MAGIC; |
48cf6061 | 880 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 LT |
881 | spin_lock_init(&nbd_dev[i].queue_lock); |
882 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 883 | mutex_init(&nbd_dev[i].tx_lock); |
7e2893a1 MP |
884 | init_timer(&nbd_dev[i].timeout_timer); |
885 | nbd_dev[i].timeout_timer.function = nbd_xmit_timeout; | |
886 | nbd_dev[i].timeout_timer.data = (unsigned long)&nbd_dev[i]; | |
4b2f0260 | 887 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 888 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 889 | nbd_dev[i].blksize = 1024; |
4b86a872 | 890 | nbd_dev[i].bytesize = 0; |
1da177e4 | 891 | disk->major = NBD_MAJOR; |
d71a6d73 | 892 | disk->first_minor = i << part_shift; |
1da177e4 LT |
893 | disk->fops = &nbd_fops; |
894 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 895 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 896 | set_capacity(disk, 0); |
1da177e4 LT |
897 | add_disk(disk); |
898 | } | |
899 | ||
900 | return 0; | |
901 | out: | |
902 | while (i--) { | |
903 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
904 | put_disk(nbd_dev[i].disk); | |
905 | } | |
f3944d61 | 906 | kfree(nbd_dev); |
1da177e4 LT |
907 | return err; |
908 | } | |
909 | ||
910 | static void __exit nbd_cleanup(void) | |
911 | { | |
912 | int i; | |
40be0c28 | 913 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 914 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 915 | nbd_dev[i].magic = 0; |
1da177e4 LT |
916 | if (disk) { |
917 | del_gendisk(disk); | |
918 | blk_cleanup_queue(disk->queue); | |
919 | put_disk(disk); | |
920 | } | |
921 | } | |
1da177e4 | 922 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 923 | kfree(nbd_dev); |
1da177e4 LT |
924 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
925 | } | |
926 | ||
927 | module_init(nbd_init); | |
928 | module_exit(nbd_cleanup); | |
929 | ||
930 | MODULE_DESCRIPTION("Network Block Device"); | |
931 | MODULE_LICENSE("GPL"); | |
932 | ||
40be0c28 | 933 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
934 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
935 | module_param(max_part, int, 0444); | |
936 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |