]>
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"; | |
101 | } | |
102 | return "invalid"; | |
103 | } | |
104 | #endif /* NDEBUG */ | |
105 | ||
106 | static void nbd_end_request(struct request *req) | |
107 | { | |
097c94a4 | 108 | int error = req->errors ? -EIO : 0; |
165125e1 | 109 | struct request_queue *q = req->q; |
1da177e4 LT |
110 | unsigned long flags; |
111 | ||
112 | dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name, | |
097c94a4 | 113 | req, error ? "failed" : "done"); |
1da177e4 LT |
114 | |
115 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 116 | __blk_end_request_all(req, error); |
1da177e4 LT |
117 | spin_unlock_irqrestore(q->queue_lock, flags); |
118 | } | |
119 | ||
f4507164 | 120 | static void sock_shutdown(struct nbd_device *nbd, int lock) |
7fdfd406 PC |
121 | { |
122 | /* Forcibly shutdown the socket causing all listeners | |
123 | * to error | |
124 | * | |
125 | * FIXME: This code is duplicated from sys_shutdown, but | |
126 | * there should be a more generic interface rather than | |
127 | * calling socket ops directly here */ | |
128 | if (lock) | |
f4507164 WG |
129 | mutex_lock(&nbd->tx_lock); |
130 | if (nbd->sock) { | |
131 | dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n"); | |
132 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
133 | nbd->sock = NULL; | |
7fdfd406 PC |
134 | } |
135 | if (lock) | |
f4507164 | 136 | mutex_unlock(&nbd->tx_lock); |
7fdfd406 PC |
137 | } |
138 | ||
139 | static void nbd_xmit_timeout(unsigned long arg) | |
140 | { | |
141 | struct task_struct *task = (struct task_struct *)arg; | |
142 | ||
143 | printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n", | |
144 | task->comm, task->pid); | |
145 | force_sig(SIGKILL, task); | |
146 | } | |
147 | ||
1da177e4 LT |
148 | /* |
149 | * Send or receive packet. | |
150 | */ | |
f4507164 | 151 | static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size, |
1da177e4 LT |
152 | int msg_flags) |
153 | { | |
f4507164 | 154 | struct socket *sock = nbd->sock; |
1da177e4 LT |
155 | int result; |
156 | struct msghdr msg; | |
157 | struct kvec iov; | |
be0ef957 | 158 | sigset_t blocked, oldset; |
7f338fe4 | 159 | unsigned long pflags = current->flags; |
1da177e4 | 160 | |
ffc41cf8 | 161 | if (unlikely(!sock)) { |
f4507164 | 162 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
163 | "Attempted %s on closed socket in sock_xmit\n", |
164 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
165 | return -EINVAL; |
166 | } | |
167 | ||
1da177e4 LT |
168 | /* Allow interception of SIGKILL only |
169 | * Don't allow other signals to interrupt the transmission */ | |
be0ef957 ON |
170 | siginitsetinv(&blocked, sigmask(SIGKILL)); |
171 | sigprocmask(SIG_SETMASK, &blocked, &oldset); | |
1da177e4 | 172 | |
7f338fe4 | 173 | current->flags |= PF_MEMALLOC; |
1da177e4 | 174 | do { |
7f338fe4 | 175 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
176 | iov.iov_base = buf; |
177 | iov.iov_len = size; | |
178 | msg.msg_name = NULL; | |
179 | msg.msg_namelen = 0; | |
180 | msg.msg_control = NULL; | |
181 | msg.msg_controllen = 0; | |
1da177e4 LT |
182 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
183 | ||
7fdfd406 PC |
184 | if (send) { |
185 | struct timer_list ti; | |
186 | ||
f4507164 | 187 | if (nbd->xmit_timeout) { |
7fdfd406 PC |
188 | init_timer(&ti); |
189 | ti.function = nbd_xmit_timeout; | |
190 | ti.data = (unsigned long)current; | |
f4507164 | 191 | ti.expires = jiffies + nbd->xmit_timeout; |
7fdfd406 PC |
192 | add_timer(&ti); |
193 | } | |
1da177e4 | 194 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
f4507164 | 195 | if (nbd->xmit_timeout) |
7fdfd406 PC |
196 | del_timer_sync(&ti); |
197 | } else | |
35fbf5bc NK |
198 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
199 | msg.msg_flags); | |
1da177e4 LT |
200 | |
201 | if (signal_pending(current)) { | |
202 | siginfo_t info; | |
1da177e4 | 203 | printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n", |
ba25f9dc | 204 | task_pid_nr(current), current->comm, |
be0ef957 | 205 | dequeue_signal_lock(current, ¤t->blocked, &info)); |
1da177e4 | 206 | result = -EINTR; |
f4507164 | 207 | sock_shutdown(nbd, !send); |
1da177e4 LT |
208 | break; |
209 | } | |
210 | ||
211 | if (result <= 0) { | |
212 | if (result == 0) | |
213 | result = -EPIPE; /* short read */ | |
214 | break; | |
215 | } | |
216 | size -= result; | |
217 | buf += result; | |
218 | } while (size > 0); | |
219 | ||
be0ef957 | 220 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
7f338fe4 | 221 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 LT |
222 | |
223 | return result; | |
224 | } | |
225 | ||
f4507164 | 226 | static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec, |
1da177e4 LT |
227 | int flags) |
228 | { | |
229 | int result; | |
230 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 WG |
231 | result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset, |
232 | bvec->bv_len, flags); | |
1da177e4 LT |
233 | kunmap(bvec->bv_page); |
234 | return result; | |
235 | } | |
236 | ||
7fdfd406 | 237 | /* always call with the tx_lock held */ |
f4507164 | 238 | static int nbd_send_req(struct nbd_device *nbd, struct request *req) |
1da177e4 | 239 | { |
5705f702 | 240 | int result, flags; |
1da177e4 | 241 | struct nbd_request request; |
1011c1b9 | 242 | unsigned long size = blk_rq_bytes(req); |
1da177e4 LT |
243 | |
244 | request.magic = htonl(NBD_REQUEST_MAGIC); | |
245 | request.type = htonl(nbd_cmd(req)); | |
83096ebf | 246 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
1da177e4 LT |
247 | request.len = htonl(size); |
248 | memcpy(request.handle, &req, sizeof(req)); | |
249 | ||
83096ebf | 250 | dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n", |
f4507164 | 251 | nbd->disk->disk_name, req, |
1da177e4 | 252 | nbdcmd_to_ascii(nbd_cmd(req)), |
83096ebf | 253 | (unsigned long long)blk_rq_pos(req) << 9, |
1011c1b9 | 254 | blk_rq_bytes(req)); |
f4507164 | 255 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
7fdfd406 | 256 | (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 257 | if (result <= 0) { |
f4507164 | 258 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 259 | "Send control failed (result %d)\n", result); |
1da177e4 LT |
260 | goto error_out; |
261 | } | |
262 | ||
263 | if (nbd_cmd(req) == NBD_CMD_WRITE) { | |
5705f702 N |
264 | struct req_iterator iter; |
265 | struct bio_vec *bvec; | |
1da177e4 LT |
266 | /* |
267 | * we are really probing at internals to determine | |
268 | * whether to set MSG_MORE or not... | |
269 | */ | |
5705f702 | 270 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 JA |
271 | flags = 0; |
272 | if (!rq_iter_last(req, iter)) | |
273 | flags = MSG_MORE; | |
274 | dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n", | |
f4507164 WG |
275 | nbd->disk->disk_name, req, bvec->bv_len); |
276 | result = sock_send_bvec(nbd, bvec, flags); | |
6c92e699 | 277 | if (result <= 0) { |
f4507164 | 278 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
279 | "Send data failed (result %d)\n", |
280 | result); | |
6c92e699 JA |
281 | goto error_out; |
282 | } | |
1da177e4 LT |
283 | } |
284 | } | |
1da177e4 LT |
285 | return 0; |
286 | ||
287 | error_out: | |
15746fca | 288 | return -EIO; |
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 HX |
298 | if (unlikely(err)) |
299 | goto out; | |
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 HX |
310 | |
311 | err = -ENOENT; | |
312 | ||
313 | out: | |
314 | return ERR_PTR(err); | |
1da177e4 LT |
315 | } |
316 | ||
f4507164 | 317 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
318 | { |
319 | int result; | |
320 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 321 | result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
322 | MSG_WAITALL); |
323 | kunmap(bvec->bv_page); | |
324 | return result; | |
325 | } | |
326 | ||
327 | /* NULL returned = something went wrong, inform userspace */ | |
f4507164 | 328 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
329 | { |
330 | int result; | |
331 | struct nbd_reply reply; | |
332 | struct request *req; | |
1da177e4 LT |
333 | |
334 | reply.magic = 0; | |
f4507164 | 335 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 336 | if (result <= 0) { |
f4507164 | 337 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 338 | "Receive control failed (result %d)\n", result); |
1da177e4 LT |
339 | goto harderror; |
340 | } | |
e4b57e08 MF |
341 | |
342 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 343 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 MF |
344 | (unsigned long)ntohl(reply.magic)); |
345 | result = -EPROTO; | |
346 | goto harderror; | |
347 | } | |
348 | ||
f4507164 | 349 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 350 | if (IS_ERR(req)) { |
4b2f0260 HX |
351 | result = PTR_ERR(req); |
352 | if (result != -ENOENT) | |
353 | goto harderror; | |
354 | ||
f4507164 | 355 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 356 | reply.handle); |
1da177e4 LT |
357 | result = -EBADR; |
358 | goto harderror; | |
359 | } | |
360 | ||
1da177e4 | 361 | if (ntohl(reply.error)) { |
f4507164 | 362 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 363 | ntohl(reply.error)); |
1da177e4 LT |
364 | req->errors++; |
365 | return req; | |
366 | } | |
367 | ||
368 | dprintk(DBG_RX, "%s: request %p: got reply\n", | |
f4507164 | 369 | nbd->disk->disk_name, req); |
1da177e4 | 370 | if (nbd_cmd(req) == NBD_CMD_READ) { |
5705f702 N |
371 | struct req_iterator iter; |
372 | struct bio_vec *bvec; | |
373 | ||
374 | rq_for_each_segment(bvec, req, iter) { | |
f4507164 | 375 | result = sock_recv_bvec(nbd, bvec); |
6c92e699 | 376 | if (result <= 0) { |
f4507164 | 377 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 378 | result); |
6c92e699 JA |
379 | req->errors++; |
380 | return req; | |
381 | } | |
382 | dprintk(DBG_RX, "%s: request %p: got %d bytes data\n", | |
f4507164 | 383 | nbd->disk->disk_name, req, bvec->bv_len); |
1da177e4 LT |
384 | } |
385 | } | |
386 | return req; | |
387 | harderror: | |
f4507164 | 388 | nbd->harderror = result; |
1da177e4 LT |
389 | return NULL; |
390 | } | |
391 | ||
edfaa7c3 KS |
392 | static ssize_t pid_show(struct device *dev, |
393 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 394 | { |
edfaa7c3 KS |
395 | struct gendisk *disk = dev_to_disk(dev); |
396 | ||
397 | return sprintf(buf, "%ld\n", | |
6b39bb65 PC |
398 | (long) ((struct nbd_device *)disk->private_data)->pid); |
399 | } | |
400 | ||
edfaa7c3 | 401 | static struct device_attribute pid_attr = { |
01e8ef11 | 402 | .attr = { .name = "pid", .mode = S_IRUGO}, |
6b39bb65 PC |
403 | .show = pid_show, |
404 | }; | |
405 | ||
f4507164 | 406 | static int nbd_do_it(struct nbd_device *nbd) |
1da177e4 LT |
407 | { |
408 | struct request *req; | |
84963048 | 409 | int ret; |
1da177e4 | 410 | |
f4507164 | 411 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 412 | |
7f338fe4 | 413 | sk_set_memalloc(nbd->sock->sk); |
f4507164 WG |
414 | nbd->pid = task_pid_nr(current); |
415 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); | |
84963048 | 416 | if (ret) { |
f4507164 WG |
417 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
418 | nbd->pid = 0; | |
84963048 WC |
419 | return ret; |
420 | } | |
6b39bb65 | 421 | |
f4507164 | 422 | while ((req = nbd_read_stat(nbd)) != NULL) |
1da177e4 | 423 | nbd_end_request(req); |
6b39bb65 | 424 | |
f4507164 WG |
425 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
426 | nbd->pid = 0; | |
84963048 | 427 | return 0; |
1da177e4 LT |
428 | } |
429 | ||
f4507164 | 430 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
431 | { |
432 | struct request *req; | |
433 | ||
f4507164 | 434 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 435 | |
4b2f0260 | 436 | /* |
f4507164 | 437 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
438 | * modifications to the list must have completed by now. For |
439 | * the same reason, the active_req must be NULL. | |
440 | * | |
441 | * As a consequence, we don't need to take the spin lock while | |
442 | * purging the list here. | |
443 | */ | |
f4507164 WG |
444 | BUG_ON(nbd->sock); |
445 | BUG_ON(nbd->active_req); | |
4b2f0260 | 446 | |
f4507164 WG |
447 | while (!list_empty(&nbd->queue_head)) { |
448 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
449 | queuelist); |
450 | list_del_init(&req->queuelist); | |
451 | req->errors++; | |
452 | nbd_end_request(req); | |
453 | } | |
fded4e09 PC |
454 | |
455 | while (!list_empty(&nbd->waiting_queue)) { | |
456 | req = list_entry(nbd->waiting_queue.next, struct request, | |
457 | queuelist); | |
458 | list_del_init(&req->queuelist); | |
459 | req->errors++; | |
460 | nbd_end_request(req); | |
461 | } | |
1da177e4 LT |
462 | } |
463 | ||
7fdfd406 | 464 | |
f4507164 | 465 | static void nbd_handle_req(struct nbd_device *nbd, struct request *req) |
48cf6061 | 466 | { |
33659ebb | 467 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
468 | goto error_out; |
469 | ||
470 | nbd_cmd(req) = NBD_CMD_READ; | |
471 | if (rq_data_dir(req) == WRITE) { | |
472 | nbd_cmd(req) = NBD_CMD_WRITE; | |
2f012508 | 473 | if (nbd->flags & NBD_FLAG_READ_ONLY) { |
f4507164 | 474 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 475 | "Write on read-only\n"); |
48cf6061 LV |
476 | goto error_out; |
477 | } | |
478 | } | |
479 | ||
480 | req->errors = 0; | |
481 | ||
f4507164 WG |
482 | mutex_lock(&nbd->tx_lock); |
483 | if (unlikely(!nbd->sock)) { | |
484 | mutex_unlock(&nbd->tx_lock); | |
485 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 486 | "Attempted send on closed socket\n"); |
15746fca | 487 | goto error_out; |
48cf6061 LV |
488 | } |
489 | ||
f4507164 | 490 | nbd->active_req = req; |
48cf6061 | 491 | |
f4507164 WG |
492 | if (nbd_send_req(nbd, req) != 0) { |
493 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 LV |
494 | req->errors++; |
495 | nbd_end_request(req); | |
496 | } else { | |
f4507164 | 497 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 498 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 499 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
500 | } |
501 | ||
f4507164 WG |
502 | nbd->active_req = NULL; |
503 | mutex_unlock(&nbd->tx_lock); | |
504 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
505 | |
506 | return; | |
507 | ||
508 | error_out: | |
509 | req->errors++; | |
510 | nbd_end_request(req); | |
511 | } | |
512 | ||
513 | static int nbd_thread(void *data) | |
514 | { | |
f4507164 | 515 | struct nbd_device *nbd = data; |
48cf6061 LV |
516 | struct request *req; |
517 | ||
518 | set_user_nice(current, -20); | |
f4507164 | 519 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 520 | /* wait for something to do */ |
f4507164 | 521 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 522 | kthread_should_stop() || |
f4507164 | 523 | !list_empty(&nbd->waiting_queue)); |
48cf6061 LV |
524 | |
525 | /* extract request */ | |
f4507164 | 526 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
527 | continue; |
528 | ||
f4507164 WG |
529 | spin_lock_irq(&nbd->queue_lock); |
530 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
531 | queuelist); |
532 | list_del_init(&req->queuelist); | |
f4507164 | 533 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
534 | |
535 | /* handle request */ | |
f4507164 | 536 | nbd_handle_req(nbd, req); |
48cf6061 LV |
537 | } |
538 | return 0; | |
539 | } | |
540 | ||
1da177e4 LT |
541 | /* |
542 | * We always wait for result of write, for now. It would be nice to make it optional | |
543 | * in future | |
f4507164 | 544 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
545 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
546 | */ | |
547 | ||
15746fca | 548 | static void do_nbd_request(struct request_queue *q) |
1da177e4 LT |
549 | { |
550 | struct request *req; | |
551 | ||
9934c8c0 | 552 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 553 | struct nbd_device *nbd; |
1da177e4 | 554 | |
48cf6061 LV |
555 | spin_unlock_irq(q->queue_lock); |
556 | ||
4aff5e23 JA |
557 | dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n", |
558 | req->rq_disk->disk_name, req, req->cmd_type); | |
1da177e4 | 559 | |
f4507164 | 560 | nbd = req->rq_disk->private_data; |
1da177e4 | 561 | |
f4507164 | 562 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 563 | |
f4507164 WG |
564 | if (unlikely(!nbd->sock)) { |
565 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 566 | "Attempted send on closed socket\n"); |
4d48a542 PC |
567 | req->errors++; |
568 | nbd_end_request(req); | |
569 | spin_lock_irq(q->queue_lock); | |
570 | continue; | |
571 | } | |
572 | ||
f4507164 WG |
573 | spin_lock_irq(&nbd->queue_lock); |
574 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
575 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 576 | |
f4507164 | 577 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 578 | |
1da177e4 | 579 | spin_lock_irq(q->queue_lock); |
1da177e4 | 580 | } |
1da177e4 LT |
581 | } |
582 | ||
1a2ad211 | 583 | /* Must be called with tx_lock held */ |
1da177e4 | 584 | |
f4507164 | 585 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
586 | unsigned int cmd, unsigned long arg) |
587 | { | |
1da177e4 | 588 | switch (cmd) { |
1a2ad211 PM |
589 | case NBD_DISCONNECT: { |
590 | struct request sreq; | |
591 | ||
f4507164 | 592 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
1a2ad211 | 593 | |
4f54eec8 | 594 | blk_rq_init(NULL, &sreq); |
4aff5e23 | 595 | sreq.cmd_type = REQ_TYPE_SPECIAL; |
1da177e4 | 596 | nbd_cmd(&sreq) = NBD_CMD_DISC; |
f4507164 | 597 | if (!nbd->sock) |
1da177e4 | 598 | return -EINVAL; |
f4507164 | 599 | nbd_send_req(nbd, &sreq); |
1da177e4 | 600 | return 0; |
1a2ad211 | 601 | } |
1da177e4 | 602 | |
1a2ad211 PM |
603 | case NBD_CLEAR_SOCK: { |
604 | struct file *file; | |
605 | ||
f4507164 WG |
606 | nbd->sock = NULL; |
607 | file = nbd->file; | |
608 | nbd->file = NULL; | |
609 | nbd_clear_que(nbd); | |
610 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 611 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
1da177e4 LT |
612 | if (file) |
613 | fput(file); | |
1a2ad211 PM |
614 | return 0; |
615 | } | |
616 | ||
617 | case NBD_SET_SOCK: { | |
618 | struct file *file; | |
f4507164 | 619 | if (nbd->file) |
1da177e4 | 620 | return -EBUSY; |
1da177e4 LT |
621 | file = fget(arg); |
622 | if (file) { | |
a8cdc308 | 623 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 | 624 | if (S_ISSOCK(inode->i_mode)) { |
f4507164 WG |
625 | nbd->file = file; |
626 | nbd->sock = SOCKET_I(inode); | |
d71a6d73 LV |
627 | if (max_part > 0) |
628 | bdev->bd_invalidated = 1; | |
1a2ad211 | 629 | return 0; |
1da177e4 LT |
630 | } else { |
631 | fput(file); | |
632 | } | |
633 | } | |
1a2ad211 PM |
634 | return -EINVAL; |
635 | } | |
636 | ||
1da177e4 | 637 | case NBD_SET_BLKSIZE: |
f4507164 WG |
638 | nbd->blksize = arg; |
639 | nbd->bytesize &= ~(nbd->blksize-1); | |
640 | bdev->bd_inode->i_size = nbd->bytesize; | |
641 | set_blocksize(bdev, nbd->blksize); | |
642 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 643 | return 0; |
1a2ad211 | 644 | |
1da177e4 | 645 | case NBD_SET_SIZE: |
f4507164 WG |
646 | nbd->bytesize = arg & ~(nbd->blksize-1); |
647 | bdev->bd_inode->i_size = nbd->bytesize; | |
648 | set_blocksize(bdev, nbd->blksize); | |
649 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 650 | return 0; |
1a2ad211 | 651 | |
7fdfd406 | 652 | case NBD_SET_TIMEOUT: |
f4507164 | 653 | nbd->xmit_timeout = arg * HZ; |
7fdfd406 | 654 | return 0; |
1a2ad211 | 655 | |
2f012508 PC |
656 | case NBD_SET_FLAGS: |
657 | nbd->flags = arg; | |
658 | return 0; | |
659 | ||
1da177e4 | 660 | case NBD_SET_SIZE_BLOCKS: |
f4507164 WG |
661 | nbd->bytesize = ((u64) arg) * nbd->blksize; |
662 | bdev->bd_inode->i_size = nbd->bytesize; | |
663 | set_blocksize(bdev, nbd->blksize); | |
664 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 665 | return 0; |
1a2ad211 PM |
666 | |
667 | case NBD_DO_IT: { | |
668 | struct task_struct *thread; | |
669 | struct file *file; | |
670 | int error; | |
671 | ||
f4507164 | 672 | if (nbd->pid) |
c91192d6 | 673 | return -EBUSY; |
f4507164 | 674 | if (!nbd->file) |
1da177e4 | 675 | return -EINVAL; |
1a2ad211 | 676 | |
f4507164 | 677 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 678 | |
f4507164 | 679 | thread = kthread_create(nbd_thread, nbd, nbd->disk->disk_name); |
1a2ad211 | 680 | if (IS_ERR(thread)) { |
f4507164 | 681 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 682 | return PTR_ERR(thread); |
1a2ad211 | 683 | } |
48cf6061 | 684 | wake_up_process(thread); |
f4507164 | 685 | error = nbd_do_it(nbd); |
48cf6061 | 686 | kthread_stop(thread); |
1a2ad211 | 687 | |
f4507164 | 688 | mutex_lock(&nbd->tx_lock); |
84963048 WC |
689 | if (error) |
690 | return error; | |
f4507164 WG |
691 | sock_shutdown(nbd, 0); |
692 | file = nbd->file; | |
693 | nbd->file = NULL; | |
694 | nbd_clear_que(nbd); | |
695 | dev_warn(disk_to_dev(nbd->disk), "queue cleared\n"); | |
1da177e4 LT |
696 | if (file) |
697 | fput(file); | |
f4507164 | 698 | nbd->bytesize = 0; |
a8cdc308 | 699 | bdev->bd_inode->i_size = 0; |
f4507164 | 700 | set_capacity(nbd->disk, 0); |
d71a6d73 | 701 | if (max_part > 0) |
a8cdc308 | 702 | ioctl_by_bdev(bdev, BLKRRPART, 0); |
f4507164 | 703 | return nbd->harderror; |
1a2ad211 PM |
704 | } |
705 | ||
1da177e4 | 706 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
707 | /* |
708 | * This is for compatibility only. The queue is always cleared | |
709 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
710 | */ | |
f4507164 | 711 | BUG_ON(!nbd->sock && !list_empty(&nbd->queue_head)); |
1da177e4 | 712 | return 0; |
1a2ad211 | 713 | |
1da177e4 | 714 | case NBD_PRINT_DEBUG: |
f4507164 | 715 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 716 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
717 | nbd->queue_head.next, nbd->queue_head.prev, |
718 | &nbd->queue_head); | |
1da177e4 LT |
719 | return 0; |
720 | } | |
1a2ad211 PM |
721 | return -ENOTTY; |
722 | } | |
723 | ||
724 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
725 | unsigned int cmd, unsigned long arg) | |
726 | { | |
f4507164 | 727 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
728 | int error; |
729 | ||
730 | if (!capable(CAP_SYS_ADMIN)) | |
731 | return -EPERM; | |
732 | ||
f4507164 | 733 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 PM |
734 | |
735 | /* Anyone capable of this syscall can do *real bad* things */ | |
736 | dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n", | |
f4507164 | 737 | nbd->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg); |
1a2ad211 | 738 | |
f4507164 WG |
739 | mutex_lock(&nbd->tx_lock); |
740 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
741 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
742 | |
743 | return error; | |
1da177e4 LT |
744 | } |
745 | ||
83d5cde4 | 746 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
747 | { |
748 | .owner = THIS_MODULE, | |
8a6cfeb6 | 749 | .ioctl = nbd_ioctl, |
1da177e4 LT |
750 | }; |
751 | ||
752 | /* | |
753 | * And here should be modules and kernel interface | |
754 | * (Just smiley confuses emacs :-) | |
755 | */ | |
756 | ||
757 | static int __init nbd_init(void) | |
758 | { | |
759 | int err = -ENOMEM; | |
760 | int i; | |
d71a6d73 | 761 | int part_shift; |
1da177e4 | 762 | |
5b7b18cc | 763 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 764 | |
d71a6d73 | 765 | if (max_part < 0) { |
7742ce4a | 766 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
767 | return -EINVAL; |
768 | } | |
769 | ||
f3944d61 SW |
770 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
771 | if (!nbd_dev) | |
772 | return -ENOMEM; | |
773 | ||
d71a6d73 | 774 | part_shift = 0; |
5988ce23 | 775 | if (max_part > 0) { |
d71a6d73 LV |
776 | part_shift = fls(max_part); |
777 | ||
5988ce23 NK |
778 | /* |
779 | * Adjust max_part according to part_shift as it is exported | |
780 | * to user space so that user can know the max number of | |
781 | * partition kernel should be able to manage. | |
782 | * | |
783 | * Note that -1 is required because partition 0 is reserved | |
784 | * for the whole disk. | |
785 | */ | |
786 | max_part = (1UL << part_shift) - 1; | |
787 | } | |
788 | ||
3b271082 NK |
789 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
790 | return -EINVAL; | |
791 | ||
792 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
793 | return -EINVAL; | |
794 | ||
40be0c28 | 795 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 796 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
797 | if (!disk) |
798 | goto out; | |
799 | nbd_dev[i].disk = disk; | |
800 | /* | |
801 | * The new linux 2.5 block layer implementation requires | |
802 | * every gendisk to have its very own request_queue struct. | |
803 | * These structs are big so we dynamically allocate them. | |
804 | */ | |
805 | disk->queue = blk_init_queue(do_nbd_request, &nbd_lock); | |
806 | if (!disk->queue) { | |
807 | put_disk(disk); | |
808 | goto out; | |
809 | } | |
31dcfab0 JA |
810 | /* |
811 | * Tell the block layer that we are not a rotational device | |
812 | */ | |
813 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
1da177e4 LT |
814 | } |
815 | ||
816 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
817 | err = -EIO; | |
818 | goto out; | |
819 | } | |
820 | ||
821 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
822 | dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags); | |
823 | ||
40be0c28 | 824 | for (i = 0; i < nbds_max; i++) { |
1da177e4 LT |
825 | struct gendisk *disk = nbd_dev[i].disk; |
826 | nbd_dev[i].file = NULL; | |
f4507164 | 827 | nbd_dev[i].magic = NBD_MAGIC; |
1da177e4 | 828 | nbd_dev[i].flags = 0; |
48cf6061 | 829 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 LT |
830 | spin_lock_init(&nbd_dev[i].queue_lock); |
831 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 832 | mutex_init(&nbd_dev[i].tx_lock); |
4b2f0260 | 833 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 834 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 835 | nbd_dev[i].blksize = 1024; |
4b86a872 | 836 | nbd_dev[i].bytesize = 0; |
1da177e4 | 837 | disk->major = NBD_MAJOR; |
d71a6d73 | 838 | disk->first_minor = i << part_shift; |
1da177e4 LT |
839 | disk->fops = &nbd_fops; |
840 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 841 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 842 | set_capacity(disk, 0); |
1da177e4 LT |
843 | add_disk(disk); |
844 | } | |
845 | ||
846 | return 0; | |
847 | out: | |
848 | while (i--) { | |
849 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
850 | put_disk(nbd_dev[i].disk); | |
851 | } | |
f3944d61 | 852 | kfree(nbd_dev); |
1da177e4 LT |
853 | return err; |
854 | } | |
855 | ||
856 | static void __exit nbd_cleanup(void) | |
857 | { | |
858 | int i; | |
40be0c28 | 859 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 860 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 861 | nbd_dev[i].magic = 0; |
1da177e4 LT |
862 | if (disk) { |
863 | del_gendisk(disk); | |
864 | blk_cleanup_queue(disk->queue); | |
865 | put_disk(disk); | |
866 | } | |
867 | } | |
1da177e4 | 868 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 869 | kfree(nbd_dev); |
1da177e4 LT |
870 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
871 | } | |
872 | ||
873 | module_init(nbd_init); | |
874 | module_exit(nbd_cleanup); | |
875 | ||
876 | MODULE_DESCRIPTION("Network Block Device"); | |
877 | MODULE_LICENSE("GPL"); | |
878 | ||
40be0c28 | 879 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
880 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
881 | module_param(max_part, int, 0444); | |
882 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); | |
1da177e4 LT |
883 | #ifndef NDEBUG |
884 | module_param(debugflags, int, 0644); | |
885 | MODULE_PARM_DESC(debugflags, "flags for controlling debug output"); | |
886 | #endif |