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> |
1da177e4 | 35 | |
1da177e4 LT |
36 | #include <asm/uaccess.h> |
37 | #include <asm/types.h> | |
38 | ||
39 | #include <linux/nbd.h> | |
40 | ||
f4507164 | 41 | #define NBD_MAGIC 0x68797548 |
1da177e4 LT |
42 | |
43 | #ifdef NDEBUG | |
44 | #define dprintk(flags, fmt...) | |
45 | #else /* NDEBUG */ | |
46 | #define dprintk(flags, fmt...) do { \ | |
47 | if (debugflags & (flags)) printk(KERN_DEBUG fmt); \ | |
48 | } while (0) | |
49 | #define DBG_IOCTL 0x0004 | |
50 | #define DBG_INIT 0x0010 | |
51 | #define DBG_EXIT 0x0020 | |
52 | #define DBG_BLKDEV 0x0100 | |
53 | #define DBG_RX 0x0200 | |
54 | #define DBG_TX 0x0400 | |
55 | static unsigned int debugflags; | |
56 | #endif /* NDEBUG */ | |
57 | ||
9c7a4169 | 58 | static unsigned int nbds_max = 16; |
20a8143e | 59 | static struct nbd_device *nbd_dev; |
d71a6d73 | 60 | static int max_part; |
1da177e4 LT |
61 | |
62 | /* | |
63 | * Use just one lock (or at most 1 per NIC). Two arguments for this: | |
64 | * 1. Each NIC is essentially a synchronization point for all servers | |
65 | * accessed through that NIC so there's no need to have more locks | |
66 | * than NICs anyway. | |
67 | * 2. More locks lead to more "Dirty cache line bouncing" which will slow | |
68 | * down each lock to the point where they're actually slower than just | |
69 | * a single lock. | |
70 | * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! | |
71 | */ | |
72 | static DEFINE_SPINLOCK(nbd_lock); | |
73 | ||
74 | #ifndef NDEBUG | |
75 | static const char *ioctl_cmd_to_ascii(int cmd) | |
76 | { | |
77 | switch (cmd) { | |
78 | case NBD_SET_SOCK: return "set-sock"; | |
79 | case NBD_SET_BLKSIZE: return "set-blksize"; | |
80 | case NBD_SET_SIZE: return "set-size"; | |
2f012508 PC |
81 | case NBD_SET_TIMEOUT: return "set-timeout"; |
82 | case NBD_SET_FLAGS: return "set-flags"; | |
1da177e4 LT |
83 | case NBD_DO_IT: return "do-it"; |
84 | case NBD_CLEAR_SOCK: return "clear-sock"; | |
85 | case NBD_CLEAR_QUE: return "clear-que"; | |
86 | case NBD_PRINT_DEBUG: return "print-debug"; | |
87 | case NBD_SET_SIZE_BLOCKS: return "set-size-blocks"; | |
88 | case NBD_DISCONNECT: return "disconnect"; | |
89 | case BLKROSET: return "set-read-only"; | |
90 | case BLKFLSBUF: return "flush-buffer-cache"; | |
91 | } | |
92 | return "unknown"; | |
93 | } | |
94 | ||
95 | static const char *nbdcmd_to_ascii(int cmd) | |
96 | { | |
97 | switch (cmd) { | |
98 | case NBD_CMD_READ: return "read"; | |
99 | case NBD_CMD_WRITE: return "write"; | |
100 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 101 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 102 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
103 | } |
104 | return "invalid"; | |
105 | } | |
106 | #endif /* NDEBUG */ | |
107 | ||
108 | static void nbd_end_request(struct request *req) | |
109 | { | |
097c94a4 | 110 | int error = req->errors ? -EIO : 0; |
165125e1 | 111 | struct request_queue *q = req->q; |
1da177e4 LT |
112 | unsigned long flags; |
113 | ||
114 | dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name, | |
097c94a4 | 115 | req, error ? "failed" : "done"); |
1da177e4 LT |
116 | |
117 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 118 | __blk_end_request_all(req, error); |
1da177e4 LT |
119 | spin_unlock_irqrestore(q->queue_lock, flags); |
120 | } | |
121 | ||
f4507164 | 122 | static void sock_shutdown(struct nbd_device *nbd, int lock) |
7fdfd406 PC |
123 | { |
124 | /* Forcibly shutdown the socket causing all listeners | |
125 | * to error | |
126 | * | |
127 | * FIXME: This code is duplicated from sys_shutdown, but | |
128 | * there should be a more generic interface rather than | |
129 | * calling socket ops directly here */ | |
130 | if (lock) | |
f4507164 WG |
131 | mutex_lock(&nbd->tx_lock); |
132 | if (nbd->sock) { | |
133 | dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n"); | |
134 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
135 | nbd->sock = NULL; | |
7fdfd406 PC |
136 | } |
137 | if (lock) | |
f4507164 | 138 | mutex_unlock(&nbd->tx_lock); |
7fdfd406 PC |
139 | } |
140 | ||
141 | static void nbd_xmit_timeout(unsigned long arg) | |
142 | { | |
143 | struct task_struct *task = (struct task_struct *)arg; | |
144 | ||
145 | printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n", | |
146 | task->comm, task->pid); | |
147 | force_sig(SIGKILL, task); | |
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 | ||
7fdfd406 PC |
186 | if (send) { |
187 | struct timer_list ti; | |
188 | ||
f4507164 | 189 | if (nbd->xmit_timeout) { |
7fdfd406 PC |
190 | init_timer(&ti); |
191 | ti.function = nbd_xmit_timeout; | |
192 | ti.data = (unsigned long)current; | |
f4507164 | 193 | ti.expires = jiffies + nbd->xmit_timeout; |
7fdfd406 PC |
194 | add_timer(&ti); |
195 | } | |
1da177e4 | 196 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
f4507164 | 197 | if (nbd->xmit_timeout) |
7fdfd406 PC |
198 | del_timer_sync(&ti); |
199 | } else | |
35fbf5bc NK |
200 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
201 | msg.msg_flags); | |
1da177e4 LT |
202 | |
203 | if (signal_pending(current)) { | |
204 | siginfo_t info; | |
1da177e4 | 205 | printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n", |
ba25f9dc | 206 | task_pid_nr(current), current->comm, |
be0ef957 | 207 | dequeue_signal_lock(current, ¤t->blocked, &info)); |
1da177e4 | 208 | result = -EINTR; |
f4507164 | 209 | sock_shutdown(nbd, !send); |
1da177e4 LT |
210 | break; |
211 | } | |
212 | ||
213 | if (result <= 0) { | |
214 | if (result == 0) | |
215 | result = -EPIPE; /* short read */ | |
216 | break; | |
217 | } | |
218 | size -= result; | |
219 | buf += result; | |
220 | } while (size > 0); | |
221 | ||
be0ef957 | 222 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
7f338fe4 | 223 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 LT |
224 | |
225 | return result; | |
226 | } | |
227 | ||
f4507164 | 228 | static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec, |
1da177e4 LT |
229 | int flags) |
230 | { | |
231 | int result; | |
232 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 WG |
233 | result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset, |
234 | bvec->bv_len, flags); | |
1da177e4 LT |
235 | kunmap(bvec->bv_page); |
236 | return result; | |
237 | } | |
238 | ||
7fdfd406 | 239 | /* always call with the tx_lock held */ |
f4507164 | 240 | static int nbd_send_req(struct nbd_device *nbd, struct request *req) |
1da177e4 | 241 | { |
5705f702 | 242 | int result, flags; |
1da177e4 | 243 | struct nbd_request request; |
1011c1b9 | 244 | unsigned long size = blk_rq_bytes(req); |
1da177e4 LT |
245 | |
246 | request.magic = htonl(NBD_REQUEST_MAGIC); | |
247 | request.type = htonl(nbd_cmd(req)); | |
75f187ab AB |
248 | |
249 | if (nbd_cmd(req) == NBD_CMD_FLUSH) { | |
250 | /* Other values are reserved for FLUSH requests. */ | |
251 | request.from = 0; | |
252 | request.len = 0; | |
253 | } else { | |
254 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); | |
255 | request.len = htonl(size); | |
256 | } | |
1da177e4 LT |
257 | memcpy(request.handle, &req, sizeof(req)); |
258 | ||
83096ebf | 259 | dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n", |
f4507164 | 260 | nbd->disk->disk_name, req, |
1da177e4 | 261 | nbdcmd_to_ascii(nbd_cmd(req)), |
83096ebf | 262 | (unsigned long long)blk_rq_pos(req) << 9, |
1011c1b9 | 263 | blk_rq_bytes(req)); |
f4507164 | 264 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
7fdfd406 | 265 | (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 266 | if (result <= 0) { |
f4507164 | 267 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 268 | "Send control failed (result %d)\n", result); |
1da177e4 LT |
269 | goto error_out; |
270 | } | |
271 | ||
272 | if (nbd_cmd(req) == NBD_CMD_WRITE) { | |
5705f702 N |
273 | struct req_iterator iter; |
274 | struct bio_vec *bvec; | |
1da177e4 LT |
275 | /* |
276 | * we are really probing at internals to determine | |
277 | * whether to set MSG_MORE or not... | |
278 | */ | |
5705f702 | 279 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 JA |
280 | flags = 0; |
281 | if (!rq_iter_last(req, iter)) | |
282 | flags = MSG_MORE; | |
283 | dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n", | |
f4507164 WG |
284 | nbd->disk->disk_name, req, bvec->bv_len); |
285 | result = sock_send_bvec(nbd, bvec, flags); | |
6c92e699 | 286 | if (result <= 0) { |
f4507164 | 287 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
288 | "Send data failed (result %d)\n", |
289 | result); | |
6c92e699 JA |
290 | goto error_out; |
291 | } | |
1da177e4 LT |
292 | } |
293 | } | |
1da177e4 LT |
294 | return 0; |
295 | ||
296 | error_out: | |
15746fca | 297 | return -EIO; |
1da177e4 LT |
298 | } |
299 | ||
f4507164 | 300 | static struct request *nbd_find_request(struct nbd_device *nbd, |
0cbc591b | 301 | struct request *xreq) |
1da177e4 | 302 | { |
d2c9740b | 303 | struct request *req, *tmp; |
4b2f0260 | 304 | int err; |
1da177e4 | 305 | |
f4507164 | 306 | err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq); |
4b2f0260 HX |
307 | if (unlikely(err)) |
308 | goto out; | |
309 | ||
f4507164 WG |
310 | spin_lock(&nbd->queue_lock); |
311 | list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) { | |
1da177e4 LT |
312 | if (req != xreq) |
313 | continue; | |
314 | list_del_init(&req->queuelist); | |
f4507164 | 315 | spin_unlock(&nbd->queue_lock); |
1da177e4 LT |
316 | return req; |
317 | } | |
f4507164 | 318 | spin_unlock(&nbd->queue_lock); |
4b2f0260 HX |
319 | |
320 | err = -ENOENT; | |
321 | ||
322 | out: | |
323 | return ERR_PTR(err); | |
1da177e4 LT |
324 | } |
325 | ||
f4507164 | 326 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
327 | { |
328 | int result; | |
329 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 330 | result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
331 | MSG_WAITALL); |
332 | kunmap(bvec->bv_page); | |
333 | return result; | |
334 | } | |
335 | ||
336 | /* NULL returned = something went wrong, inform userspace */ | |
f4507164 | 337 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
338 | { |
339 | int result; | |
340 | struct nbd_reply reply; | |
341 | struct request *req; | |
1da177e4 LT |
342 | |
343 | reply.magic = 0; | |
f4507164 | 344 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 345 | if (result <= 0) { |
f4507164 | 346 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 347 | "Receive control failed (result %d)\n", result); |
1da177e4 LT |
348 | goto harderror; |
349 | } | |
e4b57e08 MF |
350 | |
351 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 352 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 MF |
353 | (unsigned long)ntohl(reply.magic)); |
354 | result = -EPROTO; | |
355 | goto harderror; | |
356 | } | |
357 | ||
f4507164 | 358 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 359 | if (IS_ERR(req)) { |
4b2f0260 HX |
360 | result = PTR_ERR(req); |
361 | if (result != -ENOENT) | |
362 | goto harderror; | |
363 | ||
f4507164 | 364 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 365 | reply.handle); |
1da177e4 LT |
366 | result = -EBADR; |
367 | goto harderror; | |
368 | } | |
369 | ||
1da177e4 | 370 | if (ntohl(reply.error)) { |
f4507164 | 371 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 372 | ntohl(reply.error)); |
1da177e4 LT |
373 | req->errors++; |
374 | return req; | |
375 | } | |
376 | ||
377 | dprintk(DBG_RX, "%s: request %p: got reply\n", | |
f4507164 | 378 | nbd->disk->disk_name, req); |
1da177e4 | 379 | if (nbd_cmd(req) == NBD_CMD_READ) { |
5705f702 N |
380 | struct req_iterator iter; |
381 | struct bio_vec *bvec; | |
382 | ||
383 | rq_for_each_segment(bvec, req, iter) { | |
f4507164 | 384 | result = sock_recv_bvec(nbd, bvec); |
6c92e699 | 385 | if (result <= 0) { |
f4507164 | 386 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 387 | result); |
6c92e699 JA |
388 | req->errors++; |
389 | return req; | |
390 | } | |
391 | dprintk(DBG_RX, "%s: request %p: got %d bytes data\n", | |
f4507164 | 392 | nbd->disk->disk_name, req, bvec->bv_len); |
1da177e4 LT |
393 | } |
394 | } | |
395 | return req; | |
396 | harderror: | |
f4507164 | 397 | nbd->harderror = result; |
1da177e4 LT |
398 | return NULL; |
399 | } | |
400 | ||
edfaa7c3 KS |
401 | static ssize_t pid_show(struct device *dev, |
402 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 403 | { |
edfaa7c3 KS |
404 | struct gendisk *disk = dev_to_disk(dev); |
405 | ||
406 | return sprintf(buf, "%ld\n", | |
6b39bb65 PC |
407 | (long) ((struct nbd_device *)disk->private_data)->pid); |
408 | } | |
409 | ||
edfaa7c3 | 410 | static struct device_attribute pid_attr = { |
01e8ef11 | 411 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
412 | .show = pid_show, |
413 | }; | |
414 | ||
f4507164 | 415 | static int nbd_do_it(struct nbd_device *nbd) |
1da177e4 LT |
416 | { |
417 | struct request *req; | |
84963048 | 418 | int ret; |
1da177e4 | 419 | |
f4507164 | 420 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 421 | |
7f338fe4 | 422 | sk_set_memalloc(nbd->sock->sk); |
f4507164 WG |
423 | nbd->pid = task_pid_nr(current); |
424 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); | |
84963048 | 425 | if (ret) { |
f4507164 WG |
426 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
427 | nbd->pid = 0; | |
84963048 WC |
428 | return ret; |
429 | } | |
6b39bb65 | 430 | |
f4507164 | 431 | while ((req = nbd_read_stat(nbd)) != NULL) |
1da177e4 | 432 | nbd_end_request(req); |
6b39bb65 | 433 | |
f4507164 WG |
434 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
435 | nbd->pid = 0; | |
84963048 | 436 | return 0; |
1da177e4 LT |
437 | } |
438 | ||
f4507164 | 439 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
440 | { |
441 | struct request *req; | |
442 | ||
f4507164 | 443 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 444 | |
4b2f0260 | 445 | /* |
f4507164 | 446 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
447 | * modifications to the list must have completed by now. For |
448 | * the same reason, the active_req must be NULL. | |
449 | * | |
450 | * As a consequence, we don't need to take the spin lock while | |
451 | * purging the list here. | |
452 | */ | |
f4507164 WG |
453 | BUG_ON(nbd->sock); |
454 | BUG_ON(nbd->active_req); | |
4b2f0260 | 455 | |
f4507164 WG |
456 | while (!list_empty(&nbd->queue_head)) { |
457 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
458 | queuelist); |
459 | list_del_init(&req->queuelist); | |
460 | req->errors++; | |
461 | nbd_end_request(req); | |
462 | } | |
fded4e09 PC |
463 | |
464 | while (!list_empty(&nbd->waiting_queue)) { | |
465 | req = list_entry(nbd->waiting_queue.next, struct request, | |
466 | queuelist); | |
467 | list_del_init(&req->queuelist); | |
468 | req->errors++; | |
469 | nbd_end_request(req); | |
470 | } | |
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 | ||
479 | nbd_cmd(req) = NBD_CMD_READ; | |
480 | if (rq_data_dir(req) == WRITE) { | |
a336d298 PC |
481 | if ((req->cmd_flags & REQ_DISCARD)) { |
482 | WARN_ON(!(nbd->flags & NBD_FLAG_SEND_TRIM)); | |
483 | nbd_cmd(req) = NBD_CMD_TRIM; | |
484 | } else | |
485 | nbd_cmd(req) = NBD_CMD_WRITE; | |
2f012508 | 486 | if (nbd->flags & NBD_FLAG_READ_ONLY) { |
f4507164 | 487 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 488 | "Write on read-only\n"); |
48cf6061 LV |
489 | goto error_out; |
490 | } | |
491 | } | |
492 | ||
75f187ab AB |
493 | if (req->cmd_flags & REQ_FLUSH) { |
494 | BUG_ON(unlikely(blk_rq_sectors(req))); | |
495 | nbd_cmd(req) = NBD_CMD_FLUSH; | |
496 | } | |
497 | ||
48cf6061 LV |
498 | req->errors = 0; |
499 | ||
f4507164 WG |
500 | mutex_lock(&nbd->tx_lock); |
501 | if (unlikely(!nbd->sock)) { | |
502 | mutex_unlock(&nbd->tx_lock); | |
503 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 504 | "Attempted send on closed socket\n"); |
15746fca | 505 | goto error_out; |
48cf6061 LV |
506 | } |
507 | ||
f4507164 | 508 | nbd->active_req = req; |
48cf6061 | 509 | |
f4507164 WG |
510 | if (nbd_send_req(nbd, req) != 0) { |
511 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 LV |
512 | req->errors++; |
513 | nbd_end_request(req); | |
514 | } else { | |
f4507164 | 515 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 516 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 517 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
518 | } |
519 | ||
f4507164 WG |
520 | nbd->active_req = NULL; |
521 | mutex_unlock(&nbd->tx_lock); | |
522 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
523 | |
524 | return; | |
525 | ||
526 | error_out: | |
527 | req->errors++; | |
528 | nbd_end_request(req); | |
529 | } | |
530 | ||
531 | static int nbd_thread(void *data) | |
532 | { | |
f4507164 | 533 | struct nbd_device *nbd = data; |
48cf6061 LV |
534 | struct request *req; |
535 | ||
536 | set_user_nice(current, -20); | |
f4507164 | 537 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 538 | /* wait for something to do */ |
f4507164 | 539 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 540 | kthread_should_stop() || |
f4507164 | 541 | !list_empty(&nbd->waiting_queue)); |
48cf6061 LV |
542 | |
543 | /* extract request */ | |
f4507164 | 544 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
545 | continue; |
546 | ||
f4507164 WG |
547 | spin_lock_irq(&nbd->queue_lock); |
548 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
549 | queuelist); |
550 | list_del_init(&req->queuelist); | |
f4507164 | 551 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
552 | |
553 | /* handle request */ | |
f4507164 | 554 | nbd_handle_req(nbd, req); |
48cf6061 LV |
555 | } |
556 | return 0; | |
557 | } | |
558 | ||
1da177e4 LT |
559 | /* |
560 | * We always wait for result of write, for now. It would be nice to make it optional | |
561 | * in future | |
f4507164 | 562 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
563 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
564 | */ | |
565 | ||
15746fca | 566 | static void do_nbd_request(struct request_queue *q) |
398eb085 | 567 | __releases(q->queue_lock) __acquires(q->queue_lock) |
1da177e4 LT |
568 | { |
569 | struct request *req; | |
570 | ||
9934c8c0 | 571 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 572 | struct nbd_device *nbd; |
1da177e4 | 573 | |
48cf6061 LV |
574 | spin_unlock_irq(q->queue_lock); |
575 | ||
4aff5e23 JA |
576 | dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n", |
577 | req->rq_disk->disk_name, req, req->cmd_type); | |
1da177e4 | 578 | |
f4507164 | 579 | nbd = req->rq_disk->private_data; |
1da177e4 | 580 | |
f4507164 | 581 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 582 | |
f4507164 WG |
583 | if (unlikely(!nbd->sock)) { |
584 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 585 | "Attempted send on closed socket\n"); |
4d48a542 PC |
586 | req->errors++; |
587 | nbd_end_request(req); | |
588 | spin_lock_irq(q->queue_lock); | |
589 | continue; | |
590 | } | |
591 | ||
f4507164 WG |
592 | spin_lock_irq(&nbd->queue_lock); |
593 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
594 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 595 | |
f4507164 | 596 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 597 | |
1da177e4 | 598 | spin_lock_irq(q->queue_lock); |
1da177e4 | 599 | } |
1da177e4 LT |
600 | } |
601 | ||
1a2ad211 | 602 | /* Must be called with tx_lock held */ |
1da177e4 | 603 | |
f4507164 | 604 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
605 | unsigned int cmd, unsigned long arg) |
606 | { | |
1da177e4 | 607 | switch (cmd) { |
1a2ad211 PM |
608 | case NBD_DISCONNECT: { |
609 | struct request sreq; | |
610 | ||
f4507164 | 611 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
3a2d63f8 PB |
612 | if (!nbd->sock) |
613 | return -EINVAL; | |
1a2ad211 | 614 | |
3a2d63f8 PB |
615 | mutex_unlock(&nbd->tx_lock); |
616 | fsync_bdev(bdev); | |
617 | mutex_lock(&nbd->tx_lock); | |
4f54eec8 | 618 | blk_rq_init(NULL, &sreq); |
4aff5e23 | 619 | sreq.cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 | 620 | nbd_cmd(&sreq) = NBD_CMD_DISC; |
3a2d63f8 PB |
621 | |
622 | /* Check again after getting mutex back. */ | |
f4507164 | 623 | if (!nbd->sock) |
1da177e4 | 624 | return -EINVAL; |
3a2d63f8 | 625 | |
f4507164 | 626 | nbd_send_req(nbd, &sreq); |
1da177e4 | 627 | return 0; |
1a2ad211 | 628 | } |
1da177e4 | 629 | |
1a2ad211 PM |
630 | case NBD_CLEAR_SOCK: { |
631 | struct file *file; | |
632 | ||
f4507164 WG |
633 | nbd->sock = NULL; |
634 | file = nbd->file; | |
635 | nbd->file = NULL; | |
636 | nbd_clear_que(nbd); | |
637 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 638 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
3a2d63f8 | 639 | kill_bdev(bdev); |
1da177e4 LT |
640 | if (file) |
641 | fput(file); | |
1a2ad211 PM |
642 | return 0; |
643 | } | |
644 | ||
645 | case NBD_SET_SOCK: { | |
646 | struct file *file; | |
f4507164 | 647 | if (nbd->file) |
1da177e4 | 648 | return -EBUSY; |
1da177e4 LT |
649 | file = fget(arg); |
650 | if (file) { | |
496ad9aa | 651 | struct inode *inode = file_inode(file); |
1da177e4 | 652 | if (S_ISSOCK(inode->i_mode)) { |
f4507164 WG |
653 | nbd->file = file; |
654 | nbd->sock = SOCKET_I(inode); | |
d71a6d73 LV |
655 | if (max_part > 0) |
656 | bdev->bd_invalidated = 1; | |
1a2ad211 | 657 | return 0; |
1da177e4 LT |
658 | } else { |
659 | fput(file); | |
660 | } | |
661 | } | |
1a2ad211 PM |
662 | return -EINVAL; |
663 | } | |
664 | ||
1da177e4 | 665 | case NBD_SET_BLKSIZE: |
f4507164 WG |
666 | nbd->blksize = arg; |
667 | nbd->bytesize &= ~(nbd->blksize-1); | |
668 | bdev->bd_inode->i_size = nbd->bytesize; | |
669 | set_blocksize(bdev, nbd->blksize); | |
670 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 671 | return 0; |
1a2ad211 | 672 | |
1da177e4 | 673 | case NBD_SET_SIZE: |
f4507164 WG |
674 | nbd->bytesize = arg & ~(nbd->blksize-1); |
675 | bdev->bd_inode->i_size = nbd->bytesize; | |
676 | set_blocksize(bdev, nbd->blksize); | |
677 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 678 | return 0; |
1a2ad211 | 679 | |
7fdfd406 | 680 | case NBD_SET_TIMEOUT: |
f4507164 | 681 | nbd->xmit_timeout = arg * HZ; |
7fdfd406 | 682 | return 0; |
1a2ad211 | 683 | |
2f012508 PC |
684 | case NBD_SET_FLAGS: |
685 | nbd->flags = arg; | |
686 | return 0; | |
687 | ||
1da177e4 | 688 | case NBD_SET_SIZE_BLOCKS: |
f4507164 WG |
689 | nbd->bytesize = ((u64) arg) * nbd->blksize; |
690 | bdev->bd_inode->i_size = nbd->bytesize; | |
691 | set_blocksize(bdev, nbd->blksize); | |
692 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 693 | return 0; |
1a2ad211 PM |
694 | |
695 | case NBD_DO_IT: { | |
696 | struct task_struct *thread; | |
697 | struct file *file; | |
698 | int error; | |
699 | ||
f4507164 | 700 | if (nbd->pid) |
c91192d6 | 701 | return -EBUSY; |
f4507164 | 702 | if (!nbd->file) |
1da177e4 | 703 | return -EINVAL; |
1a2ad211 | 704 | |
f4507164 | 705 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 706 | |
a83e814b PB |
707 | if (nbd->flags & NBD_FLAG_READ_ONLY) |
708 | set_device_ro(bdev, true); | |
a336d298 PC |
709 | if (nbd->flags & NBD_FLAG_SEND_TRIM) |
710 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
711 | nbd->disk->queue); | |
75f187ab AB |
712 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) |
713 | blk_queue_flush(nbd->disk->queue, REQ_FLUSH); | |
714 | else | |
715 | blk_queue_flush(nbd->disk->queue, 0); | |
a336d298 | 716 | |
f4507164 | 717 | thread = kthread_create(nbd_thread, nbd, nbd->disk->disk_name); |
1a2ad211 | 718 | if (IS_ERR(thread)) { |
f4507164 | 719 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 720 | return PTR_ERR(thread); |
1a2ad211 | 721 | } |
48cf6061 | 722 | wake_up_process(thread); |
f4507164 | 723 | error = nbd_do_it(nbd); |
48cf6061 | 724 | kthread_stop(thread); |
1a2ad211 | 725 | |
f4507164 | 726 | mutex_lock(&nbd->tx_lock); |
84963048 WC |
727 | if (error) |
728 | return error; | |
f4507164 WG |
729 | sock_shutdown(nbd, 0); |
730 | file = nbd->file; | |
731 | nbd->file = NULL; | |
732 | nbd_clear_que(nbd); | |
733 | dev_warn(disk_to_dev(nbd->disk), "queue cleared\n"); | |
3a2d63f8 | 734 | kill_bdev(bdev); |
a336d298 | 735 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
a83e814b | 736 | set_device_ro(bdev, false); |
1da177e4 LT |
737 | if (file) |
738 | fput(file); | |
75f187ab | 739 | nbd->flags = 0; |
f4507164 | 740 | nbd->bytesize = 0; |
a8cdc308 | 741 | bdev->bd_inode->i_size = 0; |
f4507164 | 742 | set_capacity(nbd->disk, 0); |
d71a6d73 | 743 | if (max_part > 0) |
a8cdc308 | 744 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
f4507164 | 745 | return nbd->harderror; |
1a2ad211 PM |
746 | } |
747 | ||
1da177e4 | 748 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
749 | /* |
750 | * This is for compatibility only. The queue is always cleared | |
751 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
752 | */ | |
f4507164 | 753 | BUG_ON(!nbd->sock && !list_empty(&nbd->queue_head)); |
1da177e4 | 754 | return 0; |
1a2ad211 | 755 | |
1da177e4 | 756 | case NBD_PRINT_DEBUG: |
f4507164 | 757 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 758 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
759 | nbd->queue_head.next, nbd->queue_head.prev, |
760 | &nbd->queue_head); | |
1da177e4 LT |
761 | return 0; |
762 | } | |
1a2ad211 PM |
763 | return -ENOTTY; |
764 | } | |
765 | ||
766 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
767 | unsigned int cmd, unsigned long arg) | |
768 | { | |
f4507164 | 769 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
770 | int error; |
771 | ||
772 | if (!capable(CAP_SYS_ADMIN)) | |
773 | return -EPERM; | |
774 | ||
f4507164 | 775 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 PM |
776 | |
777 | /* Anyone capable of this syscall can do *real bad* things */ | |
778 | dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n", | |
f4507164 | 779 | nbd->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg); |
1a2ad211 | 780 | |
f4507164 WG |
781 | mutex_lock(&nbd->tx_lock); |
782 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
783 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
784 | |
785 | return error; | |
1da177e4 LT |
786 | } |
787 | ||
83d5cde4 | 788 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
789 | { |
790 | .owner = THIS_MODULE, | |
8a6cfeb6 | 791 | .ioctl = nbd_ioctl, |
1da177e4 LT |
792 | }; |
793 | ||
794 | /* | |
795 | * And here should be modules and kernel interface | |
796 | * (Just smiley confuses emacs :-) | |
797 | */ | |
798 | ||
799 | static int __init nbd_init(void) | |
800 | { | |
801 | int err = -ENOMEM; | |
802 | int i; | |
d71a6d73 | 803 | int part_shift; |
1da177e4 | 804 | |
5b7b18cc | 805 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 806 | |
d71a6d73 | 807 | if (max_part < 0) { |
7742ce4a | 808 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
809 | return -EINVAL; |
810 | } | |
811 | ||
f3944d61 SW |
812 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
813 | if (!nbd_dev) | |
814 | return -ENOMEM; | |
815 | ||
d71a6d73 | 816 | part_shift = 0; |
5988ce23 | 817 | if (max_part > 0) { |
d71a6d73 LV |
818 | part_shift = fls(max_part); |
819 | ||
5988ce23 NK |
820 | /* |
821 | * Adjust max_part according to part_shift as it is exported | |
822 | * to user space so that user can know the max number of | |
823 | * partition kernel should be able to manage. | |
824 | * | |
825 | * Note that -1 is required because partition 0 is reserved | |
826 | * for the whole disk. | |
827 | */ | |
828 | max_part = (1UL << part_shift) - 1; | |
829 | } | |
830 | ||
3b271082 NK |
831 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
832 | return -EINVAL; | |
833 | ||
834 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
835 | return -EINVAL; | |
836 | ||
40be0c28 | 837 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 838 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
839 | if (!disk) |
840 | goto out; | |
841 | nbd_dev[i].disk = disk; | |
842 | /* | |
843 | * The new linux 2.5 block layer implementation requires | |
844 | * every gendisk to have its very own request_queue struct. | |
845 | * These structs are big so we dynamically allocate them. | |
846 | */ | |
847 | disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); | |
848 | if (!disk->queue) { | |
849 | put_disk(disk); | |
850 | goto out; | |
851 | } | |
31dcfab0 JA |
852 | /* |
853 | * Tell the block layer that we are not a rotational device | |
854 | */ | |
855 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
a336d298 PC |
856 | disk->queue->limits.discard_granularity = 512; |
857 | disk->queue->limits.max_discard_sectors = UINT_MAX; | |
858 | disk->queue->limits.discard_zeroes_data = 0; | |
1da177e4 LT |
859 | } |
860 | ||
861 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
862 | err = -EIO; | |
863 | goto out; | |
864 | } | |
865 | ||
866 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
867 | dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags); | |
868 | ||
40be0c28 | 869 | for (i = 0; i < nbds_max; i++) { |
1da177e4 LT |
870 | struct gendisk *disk = nbd_dev[i].disk; |
871 | nbd_dev[i].file = NULL; | |
f4507164 | 872 | nbd_dev[i].magic = NBD_MAGIC; |
1da177e4 | 873 | nbd_dev[i].flags = 0; |
48cf6061 | 874 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 LT |
875 | spin_lock_init(&nbd_dev[i].queue_lock); |
876 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 877 | mutex_init(&nbd_dev[i].tx_lock); |
4b2f0260 | 878 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 879 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 880 | nbd_dev[i].blksize = 1024; |
4b86a872 | 881 | nbd_dev[i].bytesize = 0; |
1da177e4 | 882 | disk->major = NBD_MAJOR; |
d71a6d73 | 883 | disk->first_minor = i << part_shift; |
1da177e4 LT |
884 | disk->fops = &nbd_fops; |
885 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 886 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 887 | set_capacity(disk, 0); |
1da177e4 LT |
888 | add_disk(disk); |
889 | } | |
890 | ||
891 | return 0; | |
892 | out: | |
893 | while (i--) { | |
894 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
895 | put_disk(nbd_dev[i].disk); | |
896 | } | |
f3944d61 | 897 | kfree(nbd_dev); |
1da177e4 LT |
898 | return err; |
899 | } | |
900 | ||
901 | static void __exit nbd_cleanup(void) | |
902 | { | |
903 | int i; | |
40be0c28 | 904 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 905 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 906 | nbd_dev[i].magic = 0; |
1da177e4 LT |
907 | if (disk) { |
908 | del_gendisk(disk); | |
909 | blk_cleanup_queue(disk->queue); | |
910 | put_disk(disk); | |
911 | } | |
912 | } | |
1da177e4 | 913 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 914 | kfree(nbd_dev); |
1da177e4 LT |
915 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
916 | } | |
917 | ||
918 | module_init(nbd_init); | |
919 | module_exit(nbd_cleanup); | |
920 | ||
921 | MODULE_DESCRIPTION("Network Block Device"); | |
922 | MODULE_LICENSE("GPL"); | |
923 | ||
40be0c28 | 924 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
925 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
926 | module_param(max_part, int, 0444); | |
927 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); | |
1da177e4 LT |
928 | #ifndef NDEBUG |
929 | module_param(debugflags, int, 0644); | |
930 | MODULE_PARM_DESC(debugflags, "flags for controlling debug output"); | |
931 | #endif |