]>
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); |
9dc6c806 CH |
233 | u32 type; |
234 | ||
235 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) | |
236 | type = NBD_CMD_DISC; | |
237 | else if (req->cmd_flags & REQ_DISCARD) | |
238 | type = NBD_CMD_TRIM; | |
239 | else if (req->cmd_flags & REQ_FLUSH) | |
240 | type = NBD_CMD_FLUSH; | |
241 | else if (rq_data_dir(req) == WRITE) | |
242 | type = NBD_CMD_WRITE; | |
243 | else | |
244 | type = NBD_CMD_READ; | |
1da177e4 | 245 | |
04cfac4e | 246 | memset(&request, 0, sizeof(request)); |
1da177e4 | 247 | request.magic = htonl(NBD_REQUEST_MAGIC); |
9dc6c806 CH |
248 | request.type = htonl(type); |
249 | if (type != NBD_CMD_FLUSH && type != NBD_CMD_DISC) { | |
75f187ab AB |
250 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
251 | request.len = htonl(size); | |
252 | } | |
1da177e4 LT |
253 | memcpy(request.handle, &req, sizeof(req)); |
254 | ||
d18509f5 | 255 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
9dc6c806 | 256 | req, nbdcmd_to_ascii(type), |
d18509f5 | 257 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
f4507164 | 258 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
9dc6c806 | 259 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 260 | if (result <= 0) { |
f4507164 | 261 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 262 | "Send control failed (result %d)\n", result); |
dab5313a | 263 | return -EIO; |
1da177e4 LT |
264 | } |
265 | ||
9dc6c806 | 266 | if (type == NBD_CMD_WRITE) { |
5705f702 | 267 | struct req_iterator iter; |
7988613b | 268 | struct bio_vec bvec; |
1da177e4 LT |
269 | /* |
270 | * we are really probing at internals to determine | |
271 | * whether to set MSG_MORE or not... | |
272 | */ | |
5705f702 | 273 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 | 274 | flags = 0; |
4550dd6c | 275 | if (!rq_iter_last(bvec, iter)) |
6c92e699 | 276 | flags = MSG_MORE; |
d18509f5 MP |
277 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
278 | req, bvec.bv_len); | |
7988613b | 279 | result = sock_send_bvec(nbd, &bvec, flags); |
6c92e699 | 280 | if (result <= 0) { |
f4507164 | 281 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
282 | "Send data failed (result %d)\n", |
283 | result); | |
dab5313a | 284 | return -EIO; |
6c92e699 | 285 | } |
1da177e4 LT |
286 | } |
287 | } | |
1da177e4 | 288 | return 0; |
1da177e4 LT |
289 | } |
290 | ||
f4507164 | 291 | static struct request *nbd_find_request(struct nbd_device *nbd, |
0cbc591b | 292 | struct request *xreq) |
1da177e4 | 293 | { |
d2c9740b | 294 | struct request *req, *tmp; |
4b2f0260 | 295 | int err; |
1da177e4 | 296 | |
f4507164 | 297 | err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq); |
4b2f0260 | 298 | if (unlikely(err)) |
de9ad6d4 | 299 | return ERR_PTR(err); |
4b2f0260 | 300 | |
f4507164 WG |
301 | spin_lock(&nbd->queue_lock); |
302 | list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) { | |
1da177e4 LT |
303 | if (req != xreq) |
304 | continue; | |
305 | list_del_init(&req->queuelist); | |
f4507164 | 306 | spin_unlock(&nbd->queue_lock); |
1da177e4 LT |
307 | return req; |
308 | } | |
f4507164 | 309 | spin_unlock(&nbd->queue_lock); |
4b2f0260 | 310 | |
de9ad6d4 | 311 | return ERR_PTR(-ENOENT); |
1da177e4 LT |
312 | } |
313 | ||
f4507164 | 314 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
315 | { |
316 | int result; | |
317 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 318 | result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
319 | MSG_WAITALL); |
320 | kunmap(bvec->bv_page); | |
321 | return result; | |
322 | } | |
323 | ||
324 | /* NULL returned = something went wrong, inform userspace */ | |
f4507164 | 325 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
326 | { |
327 | int result; | |
328 | struct nbd_reply reply; | |
329 | struct request *req; | |
1da177e4 LT |
330 | |
331 | reply.magic = 0; | |
f4507164 | 332 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 333 | if (result <= 0) { |
f4507164 | 334 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 335 | "Receive control failed (result %d)\n", result); |
1da177e4 LT |
336 | goto harderror; |
337 | } | |
e4b57e08 MF |
338 | |
339 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 340 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 MF |
341 | (unsigned long)ntohl(reply.magic)); |
342 | result = -EPROTO; | |
343 | goto harderror; | |
344 | } | |
345 | ||
f4507164 | 346 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 347 | if (IS_ERR(req)) { |
4b2f0260 HX |
348 | result = PTR_ERR(req); |
349 | if (result != -ENOENT) | |
350 | goto harderror; | |
351 | ||
f4507164 | 352 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 353 | reply.handle); |
1da177e4 LT |
354 | result = -EBADR; |
355 | goto harderror; | |
356 | } | |
357 | ||
1da177e4 | 358 | if (ntohl(reply.error)) { |
f4507164 | 359 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 360 | ntohl(reply.error)); |
1da177e4 LT |
361 | req->errors++; |
362 | return req; | |
363 | } | |
364 | ||
d18509f5 | 365 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req); |
9dc6c806 | 366 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 367 | struct req_iterator iter; |
7988613b | 368 | struct bio_vec bvec; |
5705f702 N |
369 | |
370 | rq_for_each_segment(bvec, req, iter) { | |
7988613b | 371 | result = sock_recv_bvec(nbd, &bvec); |
6c92e699 | 372 | if (result <= 0) { |
f4507164 | 373 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 374 | result); |
6c92e699 JA |
375 | req->errors++; |
376 | return req; | |
377 | } | |
d18509f5 MP |
378 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
379 | req, bvec.bv_len); | |
1da177e4 LT |
380 | } |
381 | } | |
382 | return req; | |
383 | harderror: | |
f4507164 | 384 | nbd->harderror = result; |
1da177e4 LT |
385 | return NULL; |
386 | } | |
387 | ||
edfaa7c3 KS |
388 | static ssize_t pid_show(struct device *dev, |
389 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 390 | { |
edfaa7c3 KS |
391 | struct gendisk *disk = dev_to_disk(dev); |
392 | ||
393 | return sprintf(buf, "%ld\n", | |
6b39bb65 PC |
394 | (long) ((struct nbd_device *)disk->private_data)->pid); |
395 | } | |
396 | ||
edfaa7c3 | 397 | static struct device_attribute pid_attr = { |
01e8ef11 | 398 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
399 | .show = pid_show, |
400 | }; | |
401 | ||
f4507164 | 402 | static int nbd_do_it(struct nbd_device *nbd) |
1da177e4 LT |
403 | { |
404 | struct request *req; | |
84963048 | 405 | int ret; |
1da177e4 | 406 | |
f4507164 | 407 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 408 | |
7f338fe4 | 409 | sk_set_memalloc(nbd->sock->sk); |
f4507164 WG |
410 | nbd->pid = task_pid_nr(current); |
411 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); | |
84963048 | 412 | if (ret) { |
f4507164 WG |
413 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
414 | nbd->pid = 0; | |
84963048 WC |
415 | return ret; |
416 | } | |
6b39bb65 | 417 | |
f4507164 | 418 | while ((req = nbd_read_stat(nbd)) != NULL) |
d18509f5 | 419 | nbd_end_request(nbd, req); |
6b39bb65 | 420 | |
f4507164 WG |
421 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
422 | nbd->pid = 0; | |
84963048 | 423 | return 0; |
1da177e4 LT |
424 | } |
425 | ||
f4507164 | 426 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
427 | { |
428 | struct request *req; | |
429 | ||
f4507164 | 430 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 431 | |
4b2f0260 | 432 | /* |
f4507164 | 433 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
434 | * modifications to the list must have completed by now. For |
435 | * the same reason, the active_req must be NULL. | |
436 | * | |
437 | * As a consequence, we don't need to take the spin lock while | |
438 | * purging the list here. | |
439 | */ | |
f4507164 WG |
440 | BUG_ON(nbd->sock); |
441 | BUG_ON(nbd->active_req); | |
4b2f0260 | 442 | |
f4507164 WG |
443 | while (!list_empty(&nbd->queue_head)) { |
444 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
445 | queuelist); |
446 | list_del_init(&req->queuelist); | |
447 | req->errors++; | |
d18509f5 | 448 | nbd_end_request(nbd, req); |
4b2f0260 | 449 | } |
fded4e09 PC |
450 | |
451 | while (!list_empty(&nbd->waiting_queue)) { | |
452 | req = list_entry(nbd->waiting_queue.next, struct request, | |
453 | queuelist); | |
454 | list_del_init(&req->queuelist); | |
455 | req->errors++; | |
d18509f5 | 456 | nbd_end_request(nbd, req); |
fded4e09 | 457 | } |
1da177e4 LT |
458 | } |
459 | ||
7fdfd406 | 460 | |
f4507164 | 461 | static void nbd_handle_req(struct nbd_device *nbd, struct request *req) |
48cf6061 | 462 | { |
33659ebb | 463 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
464 | goto error_out; |
465 | ||
9dc6c806 CH |
466 | if (rq_data_dir(req) == WRITE && |
467 | (nbd->flags & NBD_FLAG_READ_ONLY)) { | |
468 | dev_err(disk_to_dev(nbd->disk), | |
469 | "Write on read-only\n"); | |
470 | goto error_out; | |
75f187ab AB |
471 | } |
472 | ||
48cf6061 LV |
473 | req->errors = 0; |
474 | ||
f4507164 WG |
475 | mutex_lock(&nbd->tx_lock); |
476 | if (unlikely(!nbd->sock)) { | |
477 | mutex_unlock(&nbd->tx_lock); | |
478 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 479 | "Attempted send on closed socket\n"); |
15746fca | 480 | goto error_out; |
48cf6061 LV |
481 | } |
482 | ||
f4507164 | 483 | nbd->active_req = req; |
48cf6061 | 484 | |
f4507164 WG |
485 | if (nbd_send_req(nbd, req) != 0) { |
486 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 | 487 | req->errors++; |
d18509f5 | 488 | nbd_end_request(nbd, req); |
48cf6061 | 489 | } else { |
f4507164 | 490 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 491 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 492 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
493 | } |
494 | ||
f4507164 WG |
495 | nbd->active_req = NULL; |
496 | mutex_unlock(&nbd->tx_lock); | |
497 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
498 | |
499 | return; | |
500 | ||
501 | error_out: | |
502 | req->errors++; | |
d18509f5 | 503 | nbd_end_request(nbd, req); |
48cf6061 LV |
504 | } |
505 | ||
506 | static int nbd_thread(void *data) | |
507 | { | |
f4507164 | 508 | struct nbd_device *nbd = data; |
48cf6061 LV |
509 | struct request *req; |
510 | ||
8698a745 | 511 | set_user_nice(current, MIN_NICE); |
f4507164 | 512 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 513 | /* wait for something to do */ |
f4507164 | 514 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 515 | kthread_should_stop() || |
f4507164 | 516 | !list_empty(&nbd->waiting_queue)); |
48cf6061 LV |
517 | |
518 | /* extract request */ | |
f4507164 | 519 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
520 | continue; |
521 | ||
f4507164 WG |
522 | spin_lock_irq(&nbd->queue_lock); |
523 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
524 | queuelist); |
525 | list_del_init(&req->queuelist); | |
f4507164 | 526 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
527 | |
528 | /* handle request */ | |
f4507164 | 529 | nbd_handle_req(nbd, req); |
48cf6061 LV |
530 | } |
531 | return 0; | |
532 | } | |
533 | ||
1da177e4 LT |
534 | /* |
535 | * We always wait for result of write, for now. It would be nice to make it optional | |
536 | * in future | |
f4507164 | 537 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
538 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
539 | */ | |
540 | ||
15746fca | 541 | static void do_nbd_request(struct request_queue *q) |
398eb085 | 542 | __releases(q->queue_lock) __acquires(q->queue_lock) |
1da177e4 LT |
543 | { |
544 | struct request *req; | |
545 | ||
9934c8c0 | 546 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 547 | struct nbd_device *nbd; |
1da177e4 | 548 | |
48cf6061 LV |
549 | spin_unlock_irq(q->queue_lock); |
550 | ||
f4507164 | 551 | nbd = req->rq_disk->private_data; |
1da177e4 | 552 | |
f4507164 | 553 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 554 | |
d18509f5 MP |
555 | dev_dbg(nbd_to_dev(nbd), "request %p: dequeued (flags=%x)\n", |
556 | req, req->cmd_type); | |
557 | ||
f4507164 WG |
558 | if (unlikely(!nbd->sock)) { |
559 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 560 | "Attempted send on closed socket\n"); |
4d48a542 | 561 | req->errors++; |
d18509f5 | 562 | nbd_end_request(nbd, req); |
4d48a542 PC |
563 | spin_lock_irq(q->queue_lock); |
564 | continue; | |
565 | } | |
566 | ||
f4507164 WG |
567 | spin_lock_irq(&nbd->queue_lock); |
568 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
569 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 570 | |
f4507164 | 571 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 572 | |
1da177e4 | 573 | spin_lock_irq(q->queue_lock); |
1da177e4 | 574 | } |
1da177e4 LT |
575 | } |
576 | ||
1a2ad211 | 577 | /* Must be called with tx_lock held */ |
1da177e4 | 578 | |
f4507164 | 579 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
580 | unsigned int cmd, unsigned long arg) |
581 | { | |
1da177e4 | 582 | switch (cmd) { |
1a2ad211 PM |
583 | case NBD_DISCONNECT: { |
584 | struct request sreq; | |
585 | ||
f4507164 | 586 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
3a2d63f8 PB |
587 | if (!nbd->sock) |
588 | return -EINVAL; | |
1a2ad211 | 589 | |
3a2d63f8 PB |
590 | mutex_unlock(&nbd->tx_lock); |
591 | fsync_bdev(bdev); | |
592 | mutex_lock(&nbd->tx_lock); | |
4f54eec8 | 593 | blk_rq_init(NULL, &sreq); |
4f8c9510 | 594 | sreq.cmd_type = REQ_TYPE_DRV_PRIV; |
3a2d63f8 PB |
595 | |
596 | /* Check again after getting mutex back. */ | |
f4507164 | 597 | if (!nbd->sock) |
1da177e4 | 598 | return -EINVAL; |
3a2d63f8 | 599 | |
c378f70a PC |
600 | nbd->disconnect = 1; |
601 | ||
f4507164 | 602 | nbd_send_req(nbd, &sreq); |
c378f70a | 603 | return 0; |
1a2ad211 | 604 | } |
1da177e4 | 605 | |
1a2ad211 | 606 | case NBD_CLEAR_SOCK: { |
e2511578 | 607 | struct socket *sock = nbd->sock; |
f4507164 | 608 | nbd->sock = NULL; |
f4507164 WG |
609 | nbd_clear_que(nbd); |
610 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 611 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
3a2d63f8 | 612 | kill_bdev(bdev); |
e2511578 AV |
613 | if (sock) |
614 | sockfd_put(sock); | |
1a2ad211 PM |
615 | return 0; |
616 | } | |
617 | ||
618 | case NBD_SET_SOCK: { | |
e2511578 AV |
619 | struct socket *sock; |
620 | int err; | |
621 | if (nbd->sock) | |
1da177e4 | 622 | return -EBUSY; |
e2511578 AV |
623 | sock = sockfd_lookup(arg, &err); |
624 | if (sock) { | |
625 | nbd->sock = sock; | |
626 | if (max_part > 0) | |
627 | bdev->bd_invalidated = 1; | |
628 | nbd->disconnect = 0; /* we're connected now */ | |
629 | return 0; | |
1da177e4 | 630 | } |
1a2ad211 PM |
631 | return -EINVAL; |
632 | } | |
633 | ||
1da177e4 | 634 | case NBD_SET_BLKSIZE: |
f4507164 WG |
635 | nbd->blksize = arg; |
636 | nbd->bytesize &= ~(nbd->blksize-1); | |
637 | bdev->bd_inode->i_size = nbd->bytesize; | |
638 | set_blocksize(bdev, nbd->blksize); | |
639 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 640 | return 0; |
1a2ad211 | 641 | |
1da177e4 | 642 | case NBD_SET_SIZE: |
f4507164 WG |
643 | nbd->bytesize = arg & ~(nbd->blksize-1); |
644 | bdev->bd_inode->i_size = nbd->bytesize; | |
645 | set_blocksize(bdev, nbd->blksize); | |
646 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 647 | return 0; |
1a2ad211 | 648 | |
7fdfd406 | 649 | case NBD_SET_TIMEOUT: |
f4507164 | 650 | nbd->xmit_timeout = arg * HZ; |
7fdfd406 | 651 | return 0; |
1a2ad211 | 652 | |
2f012508 PC |
653 | case NBD_SET_FLAGS: |
654 | nbd->flags = arg; | |
655 | return 0; | |
656 | ||
1da177e4 | 657 | case NBD_SET_SIZE_BLOCKS: |
f4507164 WG |
658 | nbd->bytesize = ((u64) arg) * nbd->blksize; |
659 | bdev->bd_inode->i_size = nbd->bytesize; | |
660 | set_blocksize(bdev, nbd->blksize); | |
661 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 662 | return 0; |
1a2ad211 PM |
663 | |
664 | case NBD_DO_IT: { | |
665 | struct task_struct *thread; | |
e2511578 | 666 | struct socket *sock; |
1a2ad211 PM |
667 | int error; |
668 | ||
f4507164 | 669 | if (nbd->pid) |
c91192d6 | 670 | return -EBUSY; |
e2511578 | 671 | if (!nbd->sock) |
1da177e4 | 672 | return -EINVAL; |
1a2ad211 | 673 | |
f4507164 | 674 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 675 | |
a83e814b PB |
676 | if (nbd->flags & NBD_FLAG_READ_ONLY) |
677 | set_device_ro(bdev, true); | |
a336d298 PC |
678 | if (nbd->flags & NBD_FLAG_SEND_TRIM) |
679 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
680 | nbd->disk->queue); | |
75f187ab AB |
681 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) |
682 | blk_queue_flush(nbd->disk->queue, REQ_FLUSH); | |
683 | else | |
684 | blk_queue_flush(nbd->disk->queue, 0); | |
a336d298 | 685 | |
d06df60b MP |
686 | thread = kthread_run(nbd_thread, nbd, "%s", |
687 | nbd->disk->disk_name); | |
1a2ad211 | 688 | if (IS_ERR(thread)) { |
f4507164 | 689 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 690 | return PTR_ERR(thread); |
1a2ad211 | 691 | } |
d06df60b | 692 | |
f4507164 | 693 | error = nbd_do_it(nbd); |
48cf6061 | 694 | kthread_stop(thread); |
1a2ad211 | 695 | |
f4507164 | 696 | mutex_lock(&nbd->tx_lock); |
84963048 WC |
697 | if (error) |
698 | return error; | |
f4507164 | 699 | sock_shutdown(nbd, 0); |
e2511578 AV |
700 | sock = nbd->sock; |
701 | nbd->sock = NULL; | |
f4507164 WG |
702 | nbd_clear_que(nbd); |
703 | dev_warn(disk_to_dev(nbd->disk), "queue cleared\n"); | |
3a2d63f8 | 704 | kill_bdev(bdev); |
a336d298 | 705 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
a83e814b | 706 | set_device_ro(bdev, false); |
e2511578 AV |
707 | if (sock) |
708 | sockfd_put(sock); | |
75f187ab | 709 | nbd->flags = 0; |
f4507164 | 710 | nbd->bytesize = 0; |
a8cdc308 | 711 | bdev->bd_inode->i_size = 0; |
f4507164 | 712 | set_capacity(nbd->disk, 0); |
d71a6d73 | 713 | if (max_part > 0) |
9dcd1379 | 714 | blkdev_reread_part(bdev); |
c378f70a PC |
715 | if (nbd->disconnect) /* user requested, ignore socket errors */ |
716 | return 0; | |
f4507164 | 717 | return nbd->harderror; |
1a2ad211 PM |
718 | } |
719 | ||
1da177e4 | 720 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
721 | /* |
722 | * This is for compatibility only. The queue is always cleared | |
723 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
724 | */ | |
1da177e4 | 725 | return 0; |
1a2ad211 | 726 | |
1da177e4 | 727 | case NBD_PRINT_DEBUG: |
f4507164 | 728 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 729 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
730 | nbd->queue_head.next, nbd->queue_head.prev, |
731 | &nbd->queue_head); | |
1da177e4 LT |
732 | return 0; |
733 | } | |
1a2ad211 PM |
734 | return -ENOTTY; |
735 | } | |
736 | ||
737 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
738 | unsigned int cmd, unsigned long arg) | |
739 | { | |
f4507164 | 740 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
741 | int error; |
742 | ||
743 | if (!capable(CAP_SYS_ADMIN)) | |
744 | return -EPERM; | |
745 | ||
f4507164 | 746 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 | 747 | |
f4507164 WG |
748 | mutex_lock(&nbd->tx_lock); |
749 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
750 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
751 | |
752 | return error; | |
1da177e4 LT |
753 | } |
754 | ||
83d5cde4 | 755 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
756 | { |
757 | .owner = THIS_MODULE, | |
8a6cfeb6 | 758 | .ioctl = nbd_ioctl, |
1da177e4 LT |
759 | }; |
760 | ||
761 | /* | |
762 | * And here should be modules and kernel interface | |
763 | * (Just smiley confuses emacs :-) | |
764 | */ | |
765 | ||
766 | static int __init nbd_init(void) | |
767 | { | |
768 | int err = -ENOMEM; | |
769 | int i; | |
d71a6d73 | 770 | int part_shift; |
1da177e4 | 771 | |
5b7b18cc | 772 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 773 | |
d71a6d73 | 774 | if (max_part < 0) { |
7742ce4a | 775 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
776 | return -EINVAL; |
777 | } | |
778 | ||
779 | part_shift = 0; | |
5988ce23 | 780 | if (max_part > 0) { |
d71a6d73 LV |
781 | part_shift = fls(max_part); |
782 | ||
5988ce23 NK |
783 | /* |
784 | * Adjust max_part according to part_shift as it is exported | |
785 | * to user space so that user can know the max number of | |
786 | * partition kernel should be able to manage. | |
787 | * | |
788 | * Note that -1 is required because partition 0 is reserved | |
789 | * for the whole disk. | |
790 | */ | |
791 | max_part = (1UL << part_shift) - 1; | |
792 | } | |
793 | ||
3b271082 NK |
794 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
795 | return -EINVAL; | |
796 | ||
797 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
798 | return -EINVAL; | |
799 | ||
ff6b8090 SM |
800 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
801 | if (!nbd_dev) | |
802 | return -ENOMEM; | |
803 | ||
40be0c28 | 804 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 805 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
806 | if (!disk) |
807 | goto out; | |
808 | nbd_dev[i].disk = disk; | |
809 | /* | |
810 | * The new linux 2.5 block layer implementation requires | |
811 | * every gendisk to have its very own request_queue struct. | |
812 | * These structs are big so we dynamically allocate them. | |
813 | */ | |
814 | disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); | |
815 | if (!disk->queue) { | |
816 | put_disk(disk); | |
817 | goto out; | |
818 | } | |
31dcfab0 JA |
819 | /* |
820 | * Tell the block layer that we are not a rotational device | |
821 | */ | |
822 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
b277da0a | 823 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); |
a336d298 PC |
824 | disk->queue->limits.discard_granularity = 512; |
825 | disk->queue->limits.max_discard_sectors = UINT_MAX; | |
826 | disk->queue->limits.discard_zeroes_data = 0; | |
078be02b MB |
827 | blk_queue_max_hw_sectors(disk->queue, 65536); |
828 | disk->queue->limits.max_sectors = 256; | |
1da177e4 LT |
829 | } |
830 | ||
831 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
832 | err = -EIO; | |
833 | goto out; | |
834 | } | |
835 | ||
836 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
1da177e4 | 837 | |
40be0c28 | 838 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 839 | struct gendisk *disk = nbd_dev[i].disk; |
f4507164 | 840 | nbd_dev[i].magic = NBD_MAGIC; |
48cf6061 | 841 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 LT |
842 | spin_lock_init(&nbd_dev[i].queue_lock); |
843 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 844 | mutex_init(&nbd_dev[i].tx_lock); |
4b2f0260 | 845 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 846 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 847 | nbd_dev[i].blksize = 1024; |
4b86a872 | 848 | nbd_dev[i].bytesize = 0; |
1da177e4 | 849 | disk->major = NBD_MAJOR; |
d71a6d73 | 850 | disk->first_minor = i << part_shift; |
1da177e4 LT |
851 | disk->fops = &nbd_fops; |
852 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 853 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 854 | set_capacity(disk, 0); |
1da177e4 LT |
855 | add_disk(disk); |
856 | } | |
857 | ||
858 | return 0; | |
859 | out: | |
860 | while (i--) { | |
861 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
862 | put_disk(nbd_dev[i].disk); | |
863 | } | |
f3944d61 | 864 | kfree(nbd_dev); |
1da177e4 LT |
865 | return err; |
866 | } | |
867 | ||
868 | static void __exit nbd_cleanup(void) | |
869 | { | |
870 | int i; | |
40be0c28 | 871 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 872 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 873 | nbd_dev[i].magic = 0; |
1da177e4 LT |
874 | if (disk) { |
875 | del_gendisk(disk); | |
876 | blk_cleanup_queue(disk->queue); | |
877 | put_disk(disk); | |
878 | } | |
879 | } | |
1da177e4 | 880 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 881 | kfree(nbd_dev); |
1da177e4 LT |
882 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
883 | } | |
884 | ||
885 | module_init(nbd_init); | |
886 | module_exit(nbd_cleanup); | |
887 | ||
888 | MODULE_DESCRIPTION("Network Block Device"); | |
889 | MODULE_LICENSE("GPL"); | |
890 | ||
40be0c28 | 891 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
892 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
893 | module_param(max_part, int, 0444); | |
894 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |