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> |
30d53d9c | 36 | #include <linux/debugfs.h> |
1da177e4 | 37 | |
1da177e4 LT |
38 | #include <asm/uaccess.h> |
39 | #include <asm/types.h> | |
40 | ||
41 | #include <linux/nbd.h> | |
42 | ||
13e71d69 | 43 | struct nbd_device { |
22d109c1 | 44 | u32 flags; |
13e71d69 MP |
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 | 59 | int xmit_timeout; |
696697cb | 60 | bool disconnect; /* a disconnect has been requested by user */ |
7e2893a1 MP |
61 | |
62 | struct timer_list timeout_timer; | |
63 | struct task_struct *task_recv; | |
64 | struct task_struct *task_send; | |
30d53d9c MP |
65 | |
66 | #if IS_ENABLED(CONFIG_DEBUG_FS) | |
67 | struct dentry *dbg_dir; | |
68 | #endif | |
13e71d69 MP |
69 | }; |
70 | ||
30d53d9c MP |
71 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
72 | static struct dentry *nbd_dbg_dir; | |
73 | #endif | |
74 | ||
75 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
76 | ||
f4507164 | 77 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 78 | |
9c7a4169 | 79 | static unsigned int nbds_max = 16; |
20a8143e | 80 | static struct nbd_device *nbd_dev; |
d71a6d73 | 81 | static int max_part; |
1da177e4 LT |
82 | |
83 | /* | |
84 | * Use just one lock (or at most 1 per NIC). Two arguments for this: | |
85 | * 1. Each NIC is essentially a synchronization point for all servers | |
86 | * accessed through that NIC so there's no need to have more locks | |
87 | * than NICs anyway. | |
88 | * 2. More locks lead to more "Dirty cache line bouncing" which will slow | |
89 | * down each lock to the point where they're actually slower than just | |
90 | * a single lock. | |
91 | * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this! | |
92 | */ | |
93 | static DEFINE_SPINLOCK(nbd_lock); | |
94 | ||
d18509f5 | 95 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 96 | { |
d18509f5 | 97 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
98 | } |
99 | ||
100 | static const char *nbdcmd_to_ascii(int cmd) | |
101 | { | |
102 | switch (cmd) { | |
103 | case NBD_CMD_READ: return "read"; | |
104 | case NBD_CMD_WRITE: return "write"; | |
105 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 106 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 107 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
108 | } |
109 | return "invalid"; | |
110 | } | |
1da177e4 | 111 | |
d18509f5 | 112 | static void nbd_end_request(struct nbd_device *nbd, struct request *req) |
1da177e4 | 113 | { |
097c94a4 | 114 | int error = req->errors ? -EIO : 0; |
165125e1 | 115 | struct request_queue *q = req->q; |
1da177e4 LT |
116 | unsigned long flags; |
117 | ||
d18509f5 MP |
118 | dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", req, |
119 | error ? "failed" : "done"); | |
1da177e4 LT |
120 | |
121 | spin_lock_irqsave(q->queue_lock, flags); | |
1011c1b9 | 122 | __blk_end_request_all(req, error); |
1da177e4 LT |
123 | spin_unlock_irqrestore(q->queue_lock, flags); |
124 | } | |
125 | ||
e018e757 MP |
126 | /* |
127 | * Forcibly shutdown the socket causing all listeners to error | |
128 | */ | |
36e47bee | 129 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 130 | { |
260bbce4 MP |
131 | if (!nbd->sock) |
132 | return; | |
133 | ||
134 | dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n"); | |
135 | kernel_sock_shutdown(nbd->sock, SHUT_RDWR); | |
136 | nbd->sock = NULL; | |
137 | del_timer_sync(&nbd->timeout_timer); | |
7fdfd406 PC |
138 | } |
139 | ||
140 | static void nbd_xmit_timeout(unsigned long arg) | |
141 | { | |
7e2893a1 MP |
142 | struct nbd_device *nbd = (struct nbd_device *)arg; |
143 | struct task_struct *task; | |
144 | ||
145 | if (list_empty(&nbd->queue_head)) | |
146 | return; | |
147 | ||
696697cb | 148 | nbd->disconnect = true; |
7e2893a1 MP |
149 | |
150 | task = READ_ONCE(nbd->task_recv); | |
151 | if (task) | |
152 | force_sig(SIGKILL, task); | |
7fdfd406 | 153 | |
7e2893a1 MP |
154 | task = READ_ONCE(nbd->task_send); |
155 | if (task) | |
156 | force_sig(SIGKILL, nbd->task_send); | |
157 | ||
158 | dev_err(nbd_to_dev(nbd), "Connection timed out, killed receiver and sender, shutting down connection\n"); | |
7fdfd406 PC |
159 | } |
160 | ||
1da177e4 LT |
161 | /* |
162 | * Send or receive packet. | |
163 | */ | |
f4507164 | 164 | static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size, |
1da177e4 LT |
165 | int msg_flags) |
166 | { | |
f4507164 | 167 | struct socket *sock = nbd->sock; |
1da177e4 LT |
168 | int result; |
169 | struct msghdr msg; | |
170 | struct kvec iov; | |
be0ef957 | 171 | sigset_t blocked, oldset; |
7f338fe4 | 172 | unsigned long pflags = current->flags; |
1da177e4 | 173 | |
ffc41cf8 | 174 | if (unlikely(!sock)) { |
f4507164 | 175 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
176 | "Attempted %s on closed socket in sock_xmit\n", |
177 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
178 | return -EINVAL; |
179 | } | |
180 | ||
1da177e4 LT |
181 | /* Allow interception of SIGKILL only |
182 | * Don't allow other signals to interrupt the transmission */ | |
be0ef957 ON |
183 | siginitsetinv(&blocked, sigmask(SIGKILL)); |
184 | sigprocmask(SIG_SETMASK, &blocked, &oldset); | |
1da177e4 | 185 | |
7f338fe4 | 186 | current->flags |= PF_MEMALLOC; |
1da177e4 | 187 | do { |
7f338fe4 | 188 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
189 | iov.iov_base = buf; |
190 | iov.iov_len = size; | |
191 | msg.msg_name = NULL; | |
192 | msg.msg_namelen = 0; | |
193 | msg.msg_control = NULL; | |
194 | msg.msg_controllen = 0; | |
1da177e4 LT |
195 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
196 | ||
7e2893a1 | 197 | if (send) |
1da177e4 | 198 | result = kernel_sendmsg(sock, &msg, &iov, 1, size); |
7e2893a1 | 199 | else |
35fbf5bc NK |
200 | result = kernel_recvmsg(sock, &msg, &iov, 1, size, |
201 | msg.msg_flags); | |
1da177e4 | 202 | |
1da177e4 LT |
203 | if (result <= 0) { |
204 | if (result == 0) | |
205 | result = -EPIPE; /* short read */ | |
206 | break; | |
207 | } | |
208 | size -= result; | |
209 | buf += result; | |
210 | } while (size > 0); | |
211 | ||
be0ef957 | 212 | sigprocmask(SIG_SETMASK, &oldset, NULL); |
7f338fe4 | 213 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 | 214 | |
7e2893a1 MP |
215 | if (!send && nbd->xmit_timeout) |
216 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
217 | ||
1da177e4 LT |
218 | return result; |
219 | } | |
220 | ||
f4507164 | 221 | static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec, |
1da177e4 LT |
222 | int flags) |
223 | { | |
224 | int result; | |
225 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 WG |
226 | result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset, |
227 | bvec->bv_len, flags); | |
1da177e4 LT |
228 | kunmap(bvec->bv_page); |
229 | return result; | |
230 | } | |
231 | ||
7fdfd406 | 232 | /* always call with the tx_lock held */ |
f4507164 | 233 | static int nbd_send_req(struct nbd_device *nbd, struct request *req) |
1da177e4 | 234 | { |
5705f702 | 235 | int result, flags; |
1da177e4 | 236 | struct nbd_request request; |
1011c1b9 | 237 | unsigned long size = blk_rq_bytes(req); |
9dc6c806 CH |
238 | u32 type; |
239 | ||
240 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) | |
241 | type = NBD_CMD_DISC; | |
242 | else if (req->cmd_flags & REQ_DISCARD) | |
243 | type = NBD_CMD_TRIM; | |
244 | else if (req->cmd_flags & REQ_FLUSH) | |
245 | type = NBD_CMD_FLUSH; | |
246 | else if (rq_data_dir(req) == WRITE) | |
247 | type = NBD_CMD_WRITE; | |
248 | else | |
249 | type = NBD_CMD_READ; | |
1da177e4 | 250 | |
04cfac4e | 251 | memset(&request, 0, sizeof(request)); |
1da177e4 | 252 | request.magic = htonl(NBD_REQUEST_MAGIC); |
9dc6c806 CH |
253 | request.type = htonl(type); |
254 | if (type != NBD_CMD_FLUSH && type != NBD_CMD_DISC) { | |
75f187ab AB |
255 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
256 | request.len = htonl(size); | |
257 | } | |
1da177e4 LT |
258 | memcpy(request.handle, &req, sizeof(req)); |
259 | ||
d18509f5 | 260 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
9dc6c806 | 261 | req, nbdcmd_to_ascii(type), |
d18509f5 | 262 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
f4507164 | 263 | result = sock_xmit(nbd, 1, &request, sizeof(request), |
9dc6c806 | 264 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0); |
1da177e4 | 265 | if (result <= 0) { |
f4507164 | 266 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 267 | "Send control failed (result %d)\n", result); |
dab5313a | 268 | return -EIO; |
1da177e4 LT |
269 | } |
270 | ||
9dc6c806 | 271 | if (type == NBD_CMD_WRITE) { |
5705f702 | 272 | struct req_iterator iter; |
7988613b | 273 | struct bio_vec bvec; |
1da177e4 LT |
274 | /* |
275 | * we are really probing at internals to determine | |
276 | * whether to set MSG_MORE or not... | |
277 | */ | |
5705f702 | 278 | rq_for_each_segment(bvec, req, iter) { |
6c92e699 | 279 | flags = 0; |
4550dd6c | 280 | if (!rq_iter_last(bvec, iter)) |
6c92e699 | 281 | flags = MSG_MORE; |
d18509f5 MP |
282 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
283 | req, bvec.bv_len); | |
7988613b | 284 | result = sock_send_bvec(nbd, &bvec, flags); |
6c92e699 | 285 | if (result <= 0) { |
f4507164 | 286 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
287 | "Send data failed (result %d)\n", |
288 | result); | |
dab5313a | 289 | return -EIO; |
6c92e699 | 290 | } |
1da177e4 LT |
291 | } |
292 | } | |
1da177e4 | 293 | return 0; |
1da177e4 LT |
294 | } |
295 | ||
f4507164 | 296 | static struct request *nbd_find_request(struct nbd_device *nbd, |
0cbc591b | 297 | struct request *xreq) |
1da177e4 | 298 | { |
d2c9740b | 299 | struct request *req, *tmp; |
4b2f0260 | 300 | int err; |
1da177e4 | 301 | |
f4507164 | 302 | err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq); |
4b2f0260 | 303 | if (unlikely(err)) |
de9ad6d4 | 304 | return ERR_PTR(err); |
4b2f0260 | 305 | |
f4507164 WG |
306 | spin_lock(&nbd->queue_lock); |
307 | list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) { | |
1da177e4 LT |
308 | if (req != xreq) |
309 | continue; | |
310 | list_del_init(&req->queuelist); | |
f4507164 | 311 | spin_unlock(&nbd->queue_lock); |
1da177e4 LT |
312 | return req; |
313 | } | |
f4507164 | 314 | spin_unlock(&nbd->queue_lock); |
4b2f0260 | 315 | |
de9ad6d4 | 316 | return ERR_PTR(-ENOENT); |
1da177e4 LT |
317 | } |
318 | ||
f4507164 | 319 | static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec) |
1da177e4 LT |
320 | { |
321 | int result; | |
322 | void *kaddr = kmap(bvec->bv_page); | |
f4507164 | 323 | result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len, |
1da177e4 LT |
324 | MSG_WAITALL); |
325 | kunmap(bvec->bv_page); | |
326 | return result; | |
327 | } | |
328 | ||
329 | /* NULL returned = something went wrong, inform userspace */ | |
f4507164 | 330 | static struct request *nbd_read_stat(struct nbd_device *nbd) |
1da177e4 LT |
331 | { |
332 | int result; | |
333 | struct nbd_reply reply; | |
334 | struct request *req; | |
1da177e4 LT |
335 | |
336 | reply.magic = 0; | |
f4507164 | 337 | result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL); |
1da177e4 | 338 | if (result <= 0) { |
f4507164 | 339 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 | 340 | "Receive control failed (result %d)\n", result); |
19391830 | 341 | return ERR_PTR(result); |
1da177e4 | 342 | } |
e4b57e08 MF |
343 | |
344 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 345 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 346 | (unsigned long)ntohl(reply.magic)); |
19391830 | 347 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
348 | } |
349 | ||
f4507164 | 350 | req = nbd_find_request(nbd, *(struct request **)reply.handle); |
801678c5 | 351 | if (IS_ERR(req)) { |
4b2f0260 HX |
352 | result = PTR_ERR(req); |
353 | if (result != -ENOENT) | |
19391830 | 354 | return ERR_PTR(result); |
4b2f0260 | 355 | |
f4507164 | 356 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n", |
7f1b90f9 | 357 | reply.handle); |
19391830 | 358 | return ERR_PTR(-EBADR); |
1da177e4 LT |
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 | ||
d18509f5 | 368 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req); |
9dc6c806 | 369 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 370 | struct req_iterator iter; |
7988613b | 371 | struct bio_vec bvec; |
5705f702 N |
372 | |
373 | rq_for_each_segment(bvec, req, iter) { | |
7988613b | 374 | result = sock_recv_bvec(nbd, &bvec); |
6c92e699 | 375 | if (result <= 0) { |
f4507164 | 376 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 377 | result); |
6c92e699 JA |
378 | req->errors++; |
379 | return req; | |
380 | } | |
d18509f5 MP |
381 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
382 | req, bvec.bv_len); | |
1da177e4 LT |
383 | } |
384 | } | |
385 | return req; | |
1da177e4 LT |
386 | } |
387 | ||
edfaa7c3 KS |
388 | static ssize_t pid_show(struct device *dev, |
389 | struct device_attribute *attr, char *buf) | |
6b39bb65 | 390 | { |
edfaa7c3 | 391 | struct gendisk *disk = dev_to_disk(dev); |
6521d39a | 392 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; |
edfaa7c3 | 393 | |
6521d39a | 394 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); |
6b39bb65 PC |
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 | ||
cad73b27 | 402 | static int nbd_thread_recv(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); |
6521d39a MP |
410 | |
411 | nbd->task_recv = current; | |
412 | ||
f4507164 | 413 | ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
84963048 | 414 | if (ret) { |
f4507164 | 415 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); |
6521d39a | 416 | nbd->task_recv = NULL; |
84963048 WC |
417 | return ret; |
418 | } | |
6b39bb65 | 419 | |
19391830 MP |
420 | while (1) { |
421 | req = nbd_read_stat(nbd); | |
422 | if (IS_ERR(req)) { | |
423 | ret = PTR_ERR(req); | |
424 | break; | |
425 | } | |
426 | ||
d18509f5 | 427 | nbd_end_request(nbd, req); |
19391830 | 428 | } |
6b39bb65 | 429 | |
6521d39a MP |
430 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); |
431 | ||
7e2893a1 MP |
432 | nbd->task_recv = NULL; |
433 | ||
434 | if (signal_pending(current)) { | |
435 | siginfo_t info; | |
436 | ||
437 | ret = dequeue_signal_lock(current, ¤t->blocked, &info); | |
438 | dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n", | |
439 | task_pid_nr(current), current->comm, ret); | |
36e47bee MP |
440 | mutex_lock(&nbd->tx_lock); |
441 | sock_shutdown(nbd); | |
442 | mutex_unlock(&nbd->tx_lock); | |
7e2893a1 MP |
443 | ret = -ETIMEDOUT; |
444 | } | |
445 | ||
7e2893a1 | 446 | return ret; |
1da177e4 LT |
447 | } |
448 | ||
f4507164 | 449 | static void nbd_clear_que(struct nbd_device *nbd) |
1da177e4 LT |
450 | { |
451 | struct request *req; | |
452 | ||
f4507164 | 453 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 454 | |
4b2f0260 | 455 | /* |
f4507164 | 456 | * Because we have set nbd->sock to NULL under the tx_lock, all |
4b2f0260 HX |
457 | * modifications to the list must have completed by now. For |
458 | * the same reason, the active_req must be NULL. | |
459 | * | |
460 | * As a consequence, we don't need to take the spin lock while | |
461 | * purging the list here. | |
462 | */ | |
f4507164 WG |
463 | BUG_ON(nbd->sock); |
464 | BUG_ON(nbd->active_req); | |
4b2f0260 | 465 | |
f4507164 WG |
466 | while (!list_empty(&nbd->queue_head)) { |
467 | req = list_entry(nbd->queue_head.next, struct request, | |
4b2f0260 HX |
468 | queuelist); |
469 | list_del_init(&req->queuelist); | |
470 | req->errors++; | |
d18509f5 | 471 | nbd_end_request(nbd, req); |
4b2f0260 | 472 | } |
fded4e09 PC |
473 | |
474 | while (!list_empty(&nbd->waiting_queue)) { | |
475 | req = list_entry(nbd->waiting_queue.next, struct request, | |
476 | queuelist); | |
477 | list_del_init(&req->queuelist); | |
478 | req->errors++; | |
d18509f5 | 479 | nbd_end_request(nbd, req); |
fded4e09 | 480 | } |
e78273c8 | 481 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
482 | } |
483 | ||
7fdfd406 | 484 | |
f4507164 | 485 | static void nbd_handle_req(struct nbd_device *nbd, struct request *req) |
48cf6061 | 486 | { |
33659ebb | 487 | if (req->cmd_type != REQ_TYPE_FS) |
48cf6061 LV |
488 | goto error_out; |
489 | ||
9dc6c806 CH |
490 | if (rq_data_dir(req) == WRITE && |
491 | (nbd->flags & NBD_FLAG_READ_ONLY)) { | |
492 | dev_err(disk_to_dev(nbd->disk), | |
493 | "Write on read-only\n"); | |
494 | goto error_out; | |
75f187ab AB |
495 | } |
496 | ||
48cf6061 LV |
497 | req->errors = 0; |
498 | ||
f4507164 WG |
499 | mutex_lock(&nbd->tx_lock); |
500 | if (unlikely(!nbd->sock)) { | |
501 | mutex_unlock(&nbd->tx_lock); | |
502 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 503 | "Attempted send on closed socket\n"); |
15746fca | 504 | goto error_out; |
48cf6061 LV |
505 | } |
506 | ||
f4507164 | 507 | nbd->active_req = req; |
48cf6061 | 508 | |
7e2893a1 MP |
509 | if (nbd->xmit_timeout && list_empty_careful(&nbd->queue_head)) |
510 | mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout); | |
511 | ||
f4507164 WG |
512 | if (nbd_send_req(nbd, req) != 0) { |
513 | dev_err(disk_to_dev(nbd->disk), "Request send failed\n"); | |
48cf6061 | 514 | req->errors++; |
d18509f5 | 515 | nbd_end_request(nbd, req); |
48cf6061 | 516 | } else { |
f4507164 | 517 | spin_lock(&nbd->queue_lock); |
01ff5dbc | 518 | list_add_tail(&req->queuelist, &nbd->queue_head); |
f4507164 | 519 | spin_unlock(&nbd->queue_lock); |
48cf6061 LV |
520 | } |
521 | ||
f4507164 WG |
522 | nbd->active_req = NULL; |
523 | mutex_unlock(&nbd->tx_lock); | |
524 | wake_up_all(&nbd->active_wq); | |
48cf6061 LV |
525 | |
526 | return; | |
527 | ||
528 | error_out: | |
529 | req->errors++; | |
d18509f5 | 530 | nbd_end_request(nbd, req); |
48cf6061 LV |
531 | } |
532 | ||
cad73b27 | 533 | static int nbd_thread_send(void *data) |
48cf6061 | 534 | { |
f4507164 | 535 | struct nbd_device *nbd = data; |
48cf6061 LV |
536 | struct request *req; |
537 | ||
7e2893a1 MP |
538 | nbd->task_send = current; |
539 | ||
8698a745 | 540 | set_user_nice(current, MIN_NICE); |
f4507164 | 541 | while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) { |
48cf6061 | 542 | /* wait for something to do */ |
f4507164 | 543 | wait_event_interruptible(nbd->waiting_wq, |
48cf6061 | 544 | kthread_should_stop() || |
f4507164 | 545 | !list_empty(&nbd->waiting_queue)); |
48cf6061 | 546 | |
7e2893a1 MP |
547 | if (signal_pending(current)) { |
548 | siginfo_t info; | |
549 | int ret; | |
550 | ||
551 | ret = dequeue_signal_lock(current, ¤t->blocked, | |
552 | &info); | |
553 | dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n", | |
554 | task_pid_nr(current), current->comm, ret); | |
36e47bee MP |
555 | mutex_lock(&nbd->tx_lock); |
556 | sock_shutdown(nbd); | |
557 | mutex_unlock(&nbd->tx_lock); | |
7e2893a1 MP |
558 | break; |
559 | } | |
560 | ||
48cf6061 | 561 | /* extract request */ |
f4507164 | 562 | if (list_empty(&nbd->waiting_queue)) |
48cf6061 LV |
563 | continue; |
564 | ||
f4507164 WG |
565 | spin_lock_irq(&nbd->queue_lock); |
566 | req = list_entry(nbd->waiting_queue.next, struct request, | |
48cf6061 LV |
567 | queuelist); |
568 | list_del_init(&req->queuelist); | |
f4507164 | 569 | spin_unlock_irq(&nbd->queue_lock); |
48cf6061 LV |
570 | |
571 | /* handle request */ | |
f4507164 | 572 | nbd_handle_req(nbd, req); |
48cf6061 | 573 | } |
7e2893a1 MP |
574 | |
575 | nbd->task_send = NULL; | |
576 | ||
48cf6061 LV |
577 | return 0; |
578 | } | |
579 | ||
1da177e4 LT |
580 | /* |
581 | * We always wait for result of write, for now. It would be nice to make it optional | |
582 | * in future | |
f4507164 | 583 | * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK)) |
1da177e4 LT |
584 | * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); } |
585 | */ | |
586 | ||
cad73b27 | 587 | static void nbd_request_handler(struct request_queue *q) |
398eb085 | 588 | __releases(q->queue_lock) __acquires(q->queue_lock) |
1da177e4 LT |
589 | { |
590 | struct request *req; | |
591 | ||
9934c8c0 | 592 | while ((req = blk_fetch_request(q)) != NULL) { |
f4507164 | 593 | struct nbd_device *nbd; |
1da177e4 | 594 | |
48cf6061 LV |
595 | spin_unlock_irq(q->queue_lock); |
596 | ||
f4507164 | 597 | nbd = req->rq_disk->private_data; |
1da177e4 | 598 | |
f4507164 | 599 | BUG_ON(nbd->magic != NBD_MAGIC); |
1da177e4 | 600 | |
d18509f5 MP |
601 | dev_dbg(nbd_to_dev(nbd), "request %p: dequeued (flags=%x)\n", |
602 | req, req->cmd_type); | |
603 | ||
f4507164 WG |
604 | if (unlikely(!nbd->sock)) { |
605 | dev_err(disk_to_dev(nbd->disk), | |
7f1b90f9 | 606 | "Attempted send on closed socket\n"); |
4d48a542 | 607 | req->errors++; |
d18509f5 | 608 | nbd_end_request(nbd, req); |
4d48a542 PC |
609 | spin_lock_irq(q->queue_lock); |
610 | continue; | |
611 | } | |
612 | ||
f4507164 WG |
613 | spin_lock_irq(&nbd->queue_lock); |
614 | list_add_tail(&req->queuelist, &nbd->waiting_queue); | |
615 | spin_unlock_irq(&nbd->queue_lock); | |
1da177e4 | 616 | |
f4507164 | 617 | wake_up(&nbd->waiting_wq); |
4b2f0260 | 618 | |
1da177e4 | 619 | spin_lock_irq(q->queue_lock); |
1da177e4 | 620 | } |
1da177e4 LT |
621 | } |
622 | ||
30d53d9c MP |
623 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
624 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
625 | ||
1a2ad211 | 626 | /* Must be called with tx_lock held */ |
1da177e4 | 627 | |
f4507164 | 628 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, |
1a2ad211 PM |
629 | unsigned int cmd, unsigned long arg) |
630 | { | |
1da177e4 | 631 | switch (cmd) { |
1a2ad211 PM |
632 | case NBD_DISCONNECT: { |
633 | struct request sreq; | |
634 | ||
f4507164 | 635 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
3a2d63f8 PB |
636 | if (!nbd->sock) |
637 | return -EINVAL; | |
1a2ad211 | 638 | |
3a2d63f8 PB |
639 | mutex_unlock(&nbd->tx_lock); |
640 | fsync_bdev(bdev); | |
641 | mutex_lock(&nbd->tx_lock); | |
4f54eec8 | 642 | blk_rq_init(NULL, &sreq); |
4f8c9510 | 643 | sreq.cmd_type = REQ_TYPE_DRV_PRIV; |
3a2d63f8 PB |
644 | |
645 | /* Check again after getting mutex back. */ | |
f4507164 | 646 | if (!nbd->sock) |
1da177e4 | 647 | return -EINVAL; |
3a2d63f8 | 648 | |
696697cb | 649 | nbd->disconnect = true; |
c378f70a | 650 | |
f4507164 | 651 | nbd_send_req(nbd, &sreq); |
c378f70a | 652 | return 0; |
1a2ad211 | 653 | } |
1da177e4 | 654 | |
1a2ad211 | 655 | case NBD_CLEAR_SOCK: { |
e2511578 | 656 | struct socket *sock = nbd->sock; |
f4507164 | 657 | nbd->sock = NULL; |
f4507164 WG |
658 | nbd_clear_que(nbd); |
659 | BUG_ON(!list_empty(&nbd->queue_head)); | |
fded4e09 | 660 | BUG_ON(!list_empty(&nbd->waiting_queue)); |
3a2d63f8 | 661 | kill_bdev(bdev); |
e2511578 AV |
662 | if (sock) |
663 | sockfd_put(sock); | |
1a2ad211 PM |
664 | return 0; |
665 | } | |
666 | ||
667 | case NBD_SET_SOCK: { | |
e2511578 AV |
668 | struct socket *sock; |
669 | int err; | |
670 | if (nbd->sock) | |
1da177e4 | 671 | return -EBUSY; |
e2511578 AV |
672 | sock = sockfd_lookup(arg, &err); |
673 | if (sock) { | |
674 | nbd->sock = sock; | |
675 | if (max_part > 0) | |
676 | bdev->bd_invalidated = 1; | |
696697cb | 677 | nbd->disconnect = false; /* we're connected now */ |
e2511578 | 678 | return 0; |
1da177e4 | 679 | } |
1a2ad211 PM |
680 | return -EINVAL; |
681 | } | |
682 | ||
1da177e4 | 683 | case NBD_SET_BLKSIZE: |
f4507164 WG |
684 | nbd->blksize = arg; |
685 | nbd->bytesize &= ~(nbd->blksize-1); | |
686 | bdev->bd_inode->i_size = nbd->bytesize; | |
687 | set_blocksize(bdev, nbd->blksize); | |
688 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 689 | return 0; |
1a2ad211 | 690 | |
1da177e4 | 691 | case NBD_SET_SIZE: |
f4507164 WG |
692 | nbd->bytesize = arg & ~(nbd->blksize-1); |
693 | bdev->bd_inode->i_size = nbd->bytesize; | |
694 | set_blocksize(bdev, nbd->blksize); | |
695 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 696 | return 0; |
1a2ad211 | 697 | |
7fdfd406 | 698 | case NBD_SET_TIMEOUT: |
f4507164 | 699 | nbd->xmit_timeout = arg * HZ; |
7e2893a1 MP |
700 | if (arg) |
701 | mod_timer(&nbd->timeout_timer, | |
702 | jiffies + nbd->xmit_timeout); | |
703 | else | |
704 | del_timer_sync(&nbd->timeout_timer); | |
705 | ||
7fdfd406 | 706 | return 0; |
1a2ad211 | 707 | |
2f012508 PC |
708 | case NBD_SET_FLAGS: |
709 | nbd->flags = arg; | |
710 | return 0; | |
711 | ||
1da177e4 | 712 | case NBD_SET_SIZE_BLOCKS: |
f4507164 WG |
713 | nbd->bytesize = ((u64) arg) * nbd->blksize; |
714 | bdev->bd_inode->i_size = nbd->bytesize; | |
715 | set_blocksize(bdev, nbd->blksize); | |
716 | set_capacity(nbd->disk, nbd->bytesize >> 9); | |
1da177e4 | 717 | return 0; |
1a2ad211 PM |
718 | |
719 | case NBD_DO_IT: { | |
720 | struct task_struct *thread; | |
e2511578 | 721 | struct socket *sock; |
1a2ad211 PM |
722 | int error; |
723 | ||
6521d39a | 724 | if (nbd->task_recv) |
c91192d6 | 725 | return -EBUSY; |
e2511578 | 726 | if (!nbd->sock) |
1da177e4 | 727 | return -EINVAL; |
1a2ad211 | 728 | |
f4507164 | 729 | mutex_unlock(&nbd->tx_lock); |
1a2ad211 | 730 | |
a83e814b PB |
731 | if (nbd->flags & NBD_FLAG_READ_ONLY) |
732 | set_device_ro(bdev, true); | |
a336d298 PC |
733 | if (nbd->flags & NBD_FLAG_SEND_TRIM) |
734 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
735 | nbd->disk->queue); | |
75f187ab AB |
736 | if (nbd->flags & NBD_FLAG_SEND_FLUSH) |
737 | blk_queue_flush(nbd->disk->queue, REQ_FLUSH); | |
738 | else | |
739 | blk_queue_flush(nbd->disk->queue, 0); | |
a336d298 | 740 | |
cad73b27 | 741 | thread = kthread_run(nbd_thread_send, nbd, "%s", |
30d53d9c | 742 | nbd_name(nbd)); |
1a2ad211 | 743 | if (IS_ERR(thread)) { |
f4507164 | 744 | mutex_lock(&nbd->tx_lock); |
48cf6061 | 745 | return PTR_ERR(thread); |
1a2ad211 | 746 | } |
d06df60b | 747 | |
30d53d9c | 748 | nbd_dev_dbg_init(nbd); |
cad73b27 | 749 | error = nbd_thread_recv(nbd); |
30d53d9c | 750 | nbd_dev_dbg_close(nbd); |
48cf6061 | 751 | kthread_stop(thread); |
1a2ad211 | 752 | |
f4507164 | 753 | mutex_lock(&nbd->tx_lock); |
19391830 | 754 | |
36e47bee | 755 | sock_shutdown(nbd); |
e2511578 AV |
756 | sock = nbd->sock; |
757 | nbd->sock = NULL; | |
f4507164 | 758 | nbd_clear_que(nbd); |
3a2d63f8 | 759 | kill_bdev(bdev); |
a336d298 | 760 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
a83e814b | 761 | set_device_ro(bdev, false); |
e2511578 AV |
762 | if (sock) |
763 | sockfd_put(sock); | |
75f187ab | 764 | nbd->flags = 0; |
f4507164 | 765 | nbd->bytesize = 0; |
a8cdc308 | 766 | bdev->bd_inode->i_size = 0; |
f4507164 | 767 | set_capacity(nbd->disk, 0); |
d71a6d73 | 768 | if (max_part > 0) |
9dcd1379 | 769 | blkdev_reread_part(bdev); |
c378f70a PC |
770 | if (nbd->disconnect) /* user requested, ignore socket errors */ |
771 | return 0; | |
19391830 | 772 | return error; |
1a2ad211 PM |
773 | } |
774 | ||
1da177e4 | 775 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
776 | /* |
777 | * This is for compatibility only. The queue is always cleared | |
778 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
779 | */ | |
1da177e4 | 780 | return 0; |
1a2ad211 | 781 | |
1da177e4 | 782 | case NBD_PRINT_DEBUG: |
f4507164 | 783 | dev_info(disk_to_dev(nbd->disk), |
5eedf541 | 784 | "next = %p, prev = %p, head = %p\n", |
f4507164 WG |
785 | nbd->queue_head.next, nbd->queue_head.prev, |
786 | &nbd->queue_head); | |
1da177e4 LT |
787 | return 0; |
788 | } | |
1a2ad211 PM |
789 | return -ENOTTY; |
790 | } | |
791 | ||
792 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
793 | unsigned int cmd, unsigned long arg) | |
794 | { | |
f4507164 | 795 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
1a2ad211 PM |
796 | int error; |
797 | ||
798 | if (!capable(CAP_SYS_ADMIN)) | |
799 | return -EPERM; | |
800 | ||
f4507164 | 801 | BUG_ON(nbd->magic != NBD_MAGIC); |
1a2ad211 | 802 | |
f4507164 WG |
803 | mutex_lock(&nbd->tx_lock); |
804 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
805 | mutex_unlock(&nbd->tx_lock); | |
1a2ad211 PM |
806 | |
807 | return error; | |
1da177e4 LT |
808 | } |
809 | ||
83d5cde4 | 810 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
811 | { |
812 | .owner = THIS_MODULE, | |
8a6cfeb6 | 813 | .ioctl = nbd_ioctl, |
1da177e4 LT |
814 | }; |
815 | ||
30d53d9c MP |
816 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
817 | ||
818 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
819 | { | |
820 | struct nbd_device *nbd = s->private; | |
821 | ||
822 | if (nbd->task_recv) | |
823 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
824 | if (nbd->task_send) | |
825 | seq_printf(s, "send: %d\n", task_pid_nr(nbd->task_send)); | |
826 | ||
827 | return 0; | |
828 | } | |
829 | ||
830 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
831 | { | |
832 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
833 | } | |
834 | ||
835 | static const struct file_operations nbd_dbg_tasks_ops = { | |
836 | .open = nbd_dbg_tasks_open, | |
837 | .read = seq_read, | |
838 | .llseek = seq_lseek, | |
839 | .release = single_release, | |
840 | }; | |
841 | ||
842 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
843 | { | |
844 | struct nbd_device *nbd = s->private; | |
845 | u32 flags = nbd->flags; | |
846 | ||
847 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
848 | ||
849 | seq_puts(s, "Known flags:\n"); | |
850 | ||
851 | if (flags & NBD_FLAG_HAS_FLAGS) | |
852 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
853 | if (flags & NBD_FLAG_READ_ONLY) | |
854 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
855 | if (flags & NBD_FLAG_SEND_FLUSH) | |
856 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
857 | if (flags & NBD_FLAG_SEND_TRIM) | |
858 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
859 | ||
860 | return 0; | |
861 | } | |
862 | ||
863 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
864 | { | |
865 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
866 | } | |
867 | ||
868 | static const struct file_operations nbd_dbg_flags_ops = { | |
869 | .open = nbd_dbg_flags_open, | |
870 | .read = seq_read, | |
871 | .llseek = seq_lseek, | |
872 | .release = single_release, | |
873 | }; | |
874 | ||
875 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
876 | { | |
877 | struct dentry *dir; | |
878 | struct dentry *f; | |
879 | ||
880 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
881 | if (IS_ERR_OR_NULL(dir)) { | |
882 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s' (%ld)\n", | |
883 | nbd_name(nbd), PTR_ERR(dir)); | |
884 | return PTR_ERR(dir); | |
885 | } | |
886 | nbd->dbg_dir = dir; | |
887 | ||
888 | f = debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); | |
889 | if (IS_ERR_OR_NULL(f)) { | |
890 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'tasks', %ld\n", | |
891 | PTR_ERR(f)); | |
892 | return PTR_ERR(f); | |
893 | } | |
894 | ||
895 | f = debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize); | |
896 | if (IS_ERR_OR_NULL(f)) { | |
897 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'size_bytes', %ld\n", | |
898 | PTR_ERR(f)); | |
899 | return PTR_ERR(f); | |
900 | } | |
901 | ||
902 | f = debugfs_create_u32("timeout", 0444, dir, &nbd->xmit_timeout); | |
903 | if (IS_ERR_OR_NULL(f)) { | |
904 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'timeout', %ld\n", | |
905 | PTR_ERR(f)); | |
906 | return PTR_ERR(f); | |
907 | } | |
908 | ||
909 | f = debugfs_create_u32("blocksize", 0444, dir, &nbd->blksize); | |
910 | if (IS_ERR_OR_NULL(f)) { | |
911 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'blocksize', %ld\n", | |
912 | PTR_ERR(f)); | |
913 | return PTR_ERR(f); | |
914 | } | |
915 | ||
916 | f = debugfs_create_file("flags", 0444, dir, &nbd, &nbd_dbg_flags_ops); | |
917 | if (IS_ERR_OR_NULL(f)) { | |
918 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'flags', %ld\n", | |
919 | PTR_ERR(f)); | |
920 | return PTR_ERR(f); | |
921 | } | |
922 | ||
923 | return 0; | |
924 | } | |
925 | ||
926 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
927 | { | |
928 | debugfs_remove_recursive(nbd->dbg_dir); | |
929 | } | |
930 | ||
931 | static int nbd_dbg_init(void) | |
932 | { | |
933 | struct dentry *dbg_dir; | |
934 | ||
935 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
936 | if (IS_ERR(dbg_dir)) | |
937 | return PTR_ERR(dbg_dir); | |
938 | ||
939 | nbd_dbg_dir = dbg_dir; | |
940 | ||
941 | return 0; | |
942 | } | |
943 | ||
944 | static void nbd_dbg_close(void) | |
945 | { | |
946 | debugfs_remove_recursive(nbd_dbg_dir); | |
947 | } | |
948 | ||
949 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
950 | ||
951 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
952 | { | |
953 | return 0; | |
954 | } | |
955 | ||
956 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
957 | { | |
958 | } | |
959 | ||
960 | static int nbd_dbg_init(void) | |
961 | { | |
962 | return 0; | |
963 | } | |
964 | ||
965 | static void nbd_dbg_close(void) | |
966 | { | |
967 | } | |
968 | ||
969 | #endif | |
970 | ||
1da177e4 LT |
971 | /* |
972 | * And here should be modules and kernel interface | |
973 | * (Just smiley confuses emacs :-) | |
974 | */ | |
975 | ||
976 | static int __init nbd_init(void) | |
977 | { | |
978 | int err = -ENOMEM; | |
979 | int i; | |
d71a6d73 | 980 | int part_shift; |
1da177e4 | 981 | |
5b7b18cc | 982 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 983 | |
d71a6d73 | 984 | if (max_part < 0) { |
7742ce4a | 985 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
986 | return -EINVAL; |
987 | } | |
988 | ||
989 | part_shift = 0; | |
5988ce23 | 990 | if (max_part > 0) { |
d71a6d73 LV |
991 | part_shift = fls(max_part); |
992 | ||
5988ce23 NK |
993 | /* |
994 | * Adjust max_part according to part_shift as it is exported | |
995 | * to user space so that user can know the max number of | |
996 | * partition kernel should be able to manage. | |
997 | * | |
998 | * Note that -1 is required because partition 0 is reserved | |
999 | * for the whole disk. | |
1000 | */ | |
1001 | max_part = (1UL << part_shift) - 1; | |
1002 | } | |
1003 | ||
3b271082 NK |
1004 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
1005 | return -EINVAL; | |
1006 | ||
1007 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
1008 | return -EINVAL; | |
1009 | ||
ff6b8090 SM |
1010 | nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL); |
1011 | if (!nbd_dev) | |
1012 | return -ENOMEM; | |
1013 | ||
40be0c28 | 1014 | for (i = 0; i < nbds_max; i++) { |
d71a6d73 | 1015 | struct gendisk *disk = alloc_disk(1 << part_shift); |
1da177e4 LT |
1016 | if (!disk) |
1017 | goto out; | |
1018 | nbd_dev[i].disk = disk; | |
1019 | /* | |
1020 | * The new linux 2.5 block layer implementation requires | |
1021 | * every gendisk to have its very own request_queue struct. | |
1022 | * These structs are big so we dynamically allocate them. | |
1023 | */ | |
cad73b27 | 1024 | disk->queue = blk_init_queue(nbd_request_handler, &nbd_lock); |
1da177e4 LT |
1025 | if (!disk->queue) { |
1026 | put_disk(disk); | |
1027 | goto out; | |
1028 | } | |
31dcfab0 JA |
1029 | /* |
1030 | * Tell the block layer that we are not a rotational device | |
1031 | */ | |
1032 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
b277da0a | 1033 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); |
a336d298 | 1034 | disk->queue->limits.discard_granularity = 512; |
2bb4cd5c | 1035 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); |
a336d298 | 1036 | disk->queue->limits.discard_zeroes_data = 0; |
078be02b MB |
1037 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1038 | disk->queue->limits.max_sectors = 256; | |
1da177e4 LT |
1039 | } |
1040 | ||
1041 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
1042 | err = -EIO; | |
1043 | goto out; | |
1044 | } | |
1045 | ||
1046 | printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR); | |
1da177e4 | 1047 | |
30d53d9c MP |
1048 | nbd_dbg_init(); |
1049 | ||
40be0c28 | 1050 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1051 | struct gendisk *disk = nbd_dev[i].disk; |
f4507164 | 1052 | nbd_dev[i].magic = NBD_MAGIC; |
48cf6061 | 1053 | INIT_LIST_HEAD(&nbd_dev[i].waiting_queue); |
1da177e4 LT |
1054 | spin_lock_init(&nbd_dev[i].queue_lock); |
1055 | INIT_LIST_HEAD(&nbd_dev[i].queue_head); | |
82d4dc5a | 1056 | mutex_init(&nbd_dev[i].tx_lock); |
7e2893a1 MP |
1057 | init_timer(&nbd_dev[i].timeout_timer); |
1058 | nbd_dev[i].timeout_timer.function = nbd_xmit_timeout; | |
1059 | nbd_dev[i].timeout_timer.data = (unsigned long)&nbd_dev[i]; | |
4b2f0260 | 1060 | init_waitqueue_head(&nbd_dev[i].active_wq); |
48cf6061 | 1061 | init_waitqueue_head(&nbd_dev[i].waiting_wq); |
1da177e4 | 1062 | nbd_dev[i].blksize = 1024; |
4b86a872 | 1063 | nbd_dev[i].bytesize = 0; |
1da177e4 | 1064 | disk->major = NBD_MAJOR; |
d71a6d73 | 1065 | disk->first_minor = i << part_shift; |
1da177e4 LT |
1066 | disk->fops = &nbd_fops; |
1067 | disk->private_data = &nbd_dev[i]; | |
1da177e4 | 1068 | sprintf(disk->disk_name, "nbd%d", i); |
4b86a872 | 1069 | set_capacity(disk, 0); |
1da177e4 LT |
1070 | add_disk(disk); |
1071 | } | |
1072 | ||
1073 | return 0; | |
1074 | out: | |
1075 | while (i--) { | |
1076 | blk_cleanup_queue(nbd_dev[i].disk->queue); | |
1077 | put_disk(nbd_dev[i].disk); | |
1078 | } | |
f3944d61 | 1079 | kfree(nbd_dev); |
1da177e4 LT |
1080 | return err; |
1081 | } | |
1082 | ||
1083 | static void __exit nbd_cleanup(void) | |
1084 | { | |
1085 | int i; | |
30d53d9c MP |
1086 | |
1087 | nbd_dbg_close(); | |
1088 | ||
40be0c28 | 1089 | for (i = 0; i < nbds_max; i++) { |
1da177e4 | 1090 | struct gendisk *disk = nbd_dev[i].disk; |
40be0c28 | 1091 | nbd_dev[i].magic = 0; |
1da177e4 LT |
1092 | if (disk) { |
1093 | del_gendisk(disk); | |
1094 | blk_cleanup_queue(disk->queue); | |
1095 | put_disk(disk); | |
1096 | } | |
1097 | } | |
1da177e4 | 1098 | unregister_blkdev(NBD_MAJOR, "nbd"); |
f3944d61 | 1099 | kfree(nbd_dev); |
1da177e4 LT |
1100 | printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR); |
1101 | } | |
1102 | ||
1103 | module_init(nbd_init); | |
1104 | module_exit(nbd_cleanup); | |
1105 | ||
1106 | MODULE_DESCRIPTION("Network Block Device"); | |
1107 | MODULE_LICENSE("GPL"); | |
1108 | ||
40be0c28 | 1109 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
1110 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
1111 | module_param(max_part, int, 0444); | |
1112 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |