]>
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> |
fd8383fd | 37 | #include <linux/blk-mq.h> |
1da177e4 | 38 | |
7c0f6ba6 | 39 | #include <linux/uaccess.h> |
1da177e4 LT |
40 | #include <asm/types.h> |
41 | ||
42 | #include <linux/nbd.h> | |
e46c7287 JB |
43 | #include <linux/nbd-netlink.h> |
44 | #include <net/genetlink.h> | |
1da177e4 | 45 | |
b0d9111a JB |
46 | static DEFINE_IDR(nbd_index_idr); |
47 | static DEFINE_MUTEX(nbd_index_mutex); | |
48 | ||
9561a7ad JB |
49 | struct nbd_sock { |
50 | struct socket *sock; | |
51 | struct mutex tx_lock; | |
9dd5d3ab JB |
52 | struct request *pending; |
53 | int sent; | |
f3733247 JB |
54 | bool dead; |
55 | int fallback_index; | |
799f9a38 | 56 | int cookie; |
9561a7ad JB |
57 | }; |
58 | ||
5ea8d108 JB |
59 | struct recv_thread_args { |
60 | struct work_struct work; | |
61 | struct nbd_device *nbd; | |
62 | int index; | |
63 | }; | |
64 | ||
799f9a38 JB |
65 | struct link_dead_args { |
66 | struct work_struct work; | |
67 | int index; | |
68 | }; | |
69 | ||
9b4a6ba9 JB |
70 | #define NBD_TIMEDOUT 0 |
71 | #define NBD_DISCONNECT_REQUESTED 1 | |
9561a7ad | 72 | #define NBD_DISCONNECTED 2 |
5ea8d108 | 73 | #define NBD_HAS_PID_FILE 3 |
e46c7287 JB |
74 | #define NBD_HAS_CONFIG_REF 4 |
75 | #define NBD_BOUND 5 | |
9b4a6ba9 | 76 | |
5ea8d108 | 77 | struct nbd_config { |
22d109c1 | 78 | u32 flags; |
9b4a6ba9 | 79 | unsigned long runtime_flags; |
560bc4b3 | 80 | u64 dead_conn_timeout; |
13e71d69 | 81 | |
5ea8d108 | 82 | struct nbd_sock **socks; |
9561a7ad | 83 | int num_connections; |
560bc4b3 JB |
84 | atomic_t live_connections; |
85 | wait_queue_head_t conn_wait; | |
5ea8d108 | 86 | |
9561a7ad JB |
87 | atomic_t recv_threads; |
88 | wait_queue_head_t recv_wq; | |
ef77b515 | 89 | loff_t blksize; |
b9c495bb | 90 | loff_t bytesize; |
30d53d9c MP |
91 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
92 | struct dentry *dbg_dir; | |
93 | #endif | |
13e71d69 MP |
94 | }; |
95 | ||
5ea8d108 JB |
96 | struct nbd_device { |
97 | struct blk_mq_tag_set tag_set; | |
98 | ||
e46c7287 | 99 | int index; |
5ea8d108 JB |
100 | refcount_t config_refs; |
101 | struct nbd_config *config; | |
102 | struct mutex config_lock; | |
103 | struct gendisk *disk; | |
104 | ||
105 | struct task_struct *task_recv; | |
106 | struct task_struct *task_setup; | |
107 | }; | |
108 | ||
fd8383fd JB |
109 | struct nbd_cmd { |
110 | struct nbd_device *nbd; | |
f3733247 | 111 | int index; |
799f9a38 | 112 | int cookie; |
9561a7ad | 113 | struct completion send_complete; |
fd8383fd JB |
114 | }; |
115 | ||
30d53d9c MP |
116 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
117 | static struct dentry *nbd_dbg_dir; | |
118 | #endif | |
119 | ||
120 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
121 | ||
f4507164 | 122 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 123 | |
9c7a4169 | 124 | static unsigned int nbds_max = 16; |
d71a6d73 | 125 | static int max_part; |
124d6db0 | 126 | static struct workqueue_struct *recv_workqueue; |
b0d9111a | 127 | static int part_shift; |
1da177e4 | 128 | |
9442b739 JB |
129 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
130 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
5ea8d108 | 131 | static void nbd_config_put(struct nbd_device *nbd); |
e46c7287 | 132 | static void nbd_connect_reply(struct genl_info *info, int index); |
799f9a38 | 133 | static void nbd_dead_link_work(struct work_struct *work); |
9442b739 | 134 | |
d18509f5 | 135 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 136 | { |
d18509f5 | 137 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
138 | } |
139 | ||
140 | static const char *nbdcmd_to_ascii(int cmd) | |
141 | { | |
142 | switch (cmd) { | |
143 | case NBD_CMD_READ: return "read"; | |
144 | case NBD_CMD_WRITE: return "write"; | |
145 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 146 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 147 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
148 | } |
149 | return "invalid"; | |
150 | } | |
1da177e4 | 151 | |
5ea8d108 JB |
152 | static ssize_t pid_show(struct device *dev, |
153 | struct device_attribute *attr, char *buf) | |
154 | { | |
155 | struct gendisk *disk = dev_to_disk(dev); | |
156 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; | |
157 | ||
158 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); | |
159 | } | |
160 | ||
161 | static struct device_attribute pid_attr = { | |
162 | .attr = { .name = "pid", .mode = S_IRUGO}, | |
163 | .show = pid_show, | |
164 | }; | |
165 | ||
799f9a38 JB |
166 | static int nbd_disconnected(struct nbd_config *config) |
167 | { | |
168 | return test_bit(NBD_DISCONNECTED, &config->runtime_flags) || | |
169 | test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); | |
170 | } | |
171 | ||
172 | static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock, | |
173 | int notify) | |
f3733247 | 174 | { |
799f9a38 JB |
175 | if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) { |
176 | struct link_dead_args *args; | |
177 | args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO); | |
178 | if (args) { | |
179 | INIT_WORK(&args->work, nbd_dead_link_work); | |
180 | args->index = nbd->index; | |
181 | queue_work(system_wq, &args->work); | |
182 | } | |
183 | } | |
560bc4b3 | 184 | if (!nsock->dead) { |
f3733247 | 185 | kernel_sock_shutdown(nsock->sock, SHUT_RDWR); |
560bc4b3 JB |
186 | atomic_dec(&nbd->config->live_connections); |
187 | } | |
f3733247 JB |
188 | nsock->dead = true; |
189 | nsock->pending = NULL; | |
190 | nsock->sent = 0; | |
191 | } | |
192 | ||
29eaadc0 | 193 | static void nbd_size_clear(struct nbd_device *nbd) |
37091fdd | 194 | { |
5ea8d108 | 195 | if (nbd->config->bytesize) { |
5ea8d108 JB |
196 | set_capacity(nbd->disk, 0); |
197 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
198 | } | |
37091fdd MP |
199 | } |
200 | ||
29eaadc0 | 201 | static void nbd_size_update(struct nbd_device *nbd) |
37091fdd | 202 | { |
5ea8d108 JB |
203 | struct nbd_config *config = nbd->config; |
204 | blk_queue_logical_block_size(nbd->disk->queue, config->blksize); | |
205 | blk_queue_physical_block_size(nbd->disk->queue, config->blksize); | |
5ea8d108 | 206 | set_capacity(nbd->disk, config->bytesize >> 9); |
37091fdd MP |
207 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); |
208 | } | |
209 | ||
29eaadc0 JB |
210 | static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize, |
211 | loff_t nr_blocks) | |
37091fdd | 212 | { |
5ea8d108 JB |
213 | struct nbd_config *config = nbd->config; |
214 | config->blksize = blocksize; | |
215 | config->bytesize = blocksize * nr_blocks; | |
29eaadc0 | 216 | nbd_size_update(nbd); |
37091fdd MP |
217 | } |
218 | ||
fd8383fd | 219 | static void nbd_end_request(struct nbd_cmd *cmd) |
1da177e4 | 220 | { |
fd8383fd JB |
221 | struct nbd_device *nbd = cmd->nbd; |
222 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
097c94a4 | 223 | int error = req->errors ? -EIO : 0; |
1da177e4 | 224 | |
fd8383fd | 225 | dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", cmd, |
d18509f5 | 226 | error ? "failed" : "done"); |
1da177e4 | 227 | |
fd8383fd | 228 | blk_mq_complete_request(req, error); |
1da177e4 LT |
229 | } |
230 | ||
e018e757 MP |
231 | /* |
232 | * Forcibly shutdown the socket causing all listeners to error | |
233 | */ | |
36e47bee | 234 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 235 | { |
5ea8d108 | 236 | struct nbd_config *config = nbd->config; |
9561a7ad | 237 | int i; |
23272a67 | 238 | |
5ea8d108 | 239 | if (config->num_connections == 0) |
9561a7ad | 240 | return; |
5ea8d108 | 241 | if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
260bbce4 | 242 | return; |
23272a67 | 243 | |
5ea8d108 JB |
244 | for (i = 0; i < config->num_connections; i++) { |
245 | struct nbd_sock *nsock = config->socks[i]; | |
9561a7ad | 246 | mutex_lock(&nsock->tx_lock); |
799f9a38 | 247 | nbd_mark_nsock_dead(nbd, nsock, 0); |
9561a7ad JB |
248 | mutex_unlock(&nsock->tx_lock); |
249 | } | |
250 | dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n"); | |
7fdfd406 PC |
251 | } |
252 | ||
0eadf37a JB |
253 | static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req, |
254 | bool reserved) | |
7fdfd406 | 255 | { |
0eadf37a JB |
256 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
257 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 JB |
258 | struct nbd_config *config; |
259 | ||
260 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
261 | req->errors = -EIO; | |
262 | return BLK_EH_HANDLED; | |
263 | } | |
264 | ||
560bc4b3 JB |
265 | /* If we are waiting on our dead timer then we could get timeout |
266 | * callbacks for our request. For this we just want to reset the timer | |
267 | * and let the queue side take care of everything. | |
268 | */ | |
269 | if (!completion_done(&cmd->send_complete)) { | |
270 | nbd_config_put(nbd); | |
271 | return BLK_EH_RESET_TIMER; | |
272 | } | |
5ea8d108 | 273 | config = nbd->config; |
dcc909d9 | 274 | |
5ea8d108 | 275 | if (config->num_connections > 1) { |
f3733247 JB |
276 | dev_err_ratelimited(nbd_to_dev(nbd), |
277 | "Connection timed out, retrying\n"); | |
f3733247 JB |
278 | /* |
279 | * Hooray we have more connections, requeue this IO, the submit | |
280 | * path will put it on a real connection. | |
281 | */ | |
5ea8d108 JB |
282 | if (config->socks && config->num_connections > 1) { |
283 | if (cmd->index < config->num_connections) { | |
f3733247 | 284 | struct nbd_sock *nsock = |
5ea8d108 | 285 | config->socks[cmd->index]; |
f3733247 | 286 | mutex_lock(&nsock->tx_lock); |
799f9a38 JB |
287 | /* We can have multiple outstanding requests, so |
288 | * we don't want to mark the nsock dead if we've | |
289 | * already reconnected with a new socket, so | |
290 | * only mark it dead if its the same socket we | |
291 | * were sent out on. | |
292 | */ | |
293 | if (cmd->cookie == nsock->cookie) | |
294 | nbd_mark_nsock_dead(nbd, nsock, 1); | |
f3733247 JB |
295 | mutex_unlock(&nsock->tx_lock); |
296 | } | |
f3733247 | 297 | blk_mq_requeue_request(req, true); |
5ea8d108 | 298 | nbd_config_put(nbd); |
f3733247 JB |
299 | return BLK_EH_NOT_HANDLED; |
300 | } | |
f3733247 JB |
301 | } else { |
302 | dev_err_ratelimited(nbd_to_dev(nbd), | |
303 | "Connection timed out\n"); | |
304 | } | |
5ea8d108 | 305 | set_bit(NBD_TIMEDOUT, &config->runtime_flags); |
c103b4da | 306 | req->errors = -EIO; |
9561a7ad | 307 | sock_shutdown(nbd); |
5ea8d108 JB |
308 | nbd_config_put(nbd); |
309 | ||
0eadf37a | 310 | return BLK_EH_HANDLED; |
7fdfd406 PC |
311 | } |
312 | ||
1da177e4 LT |
313 | /* |
314 | * Send or receive packet. | |
315 | */ | |
c9f2b6ae | 316 | static int sock_xmit(struct nbd_device *nbd, int index, int send, |
9dd5d3ab | 317 | struct iov_iter *iter, int msg_flags, int *sent) |
1da177e4 | 318 | { |
5ea8d108 JB |
319 | struct nbd_config *config = nbd->config; |
320 | struct socket *sock = config->socks[index]->sock; | |
1da177e4 LT |
321 | int result; |
322 | struct msghdr msg; | |
7f338fe4 | 323 | unsigned long pflags = current->flags; |
1da177e4 | 324 | |
ffc41cf8 | 325 | if (unlikely(!sock)) { |
a897b666 | 326 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
327 | "Attempted %s on closed socket in sock_xmit\n", |
328 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
329 | return -EINVAL; |
330 | } | |
331 | ||
c9f2b6ae | 332 | msg.msg_iter = *iter; |
c1696cab | 333 | |
7f338fe4 | 334 | current->flags |= PF_MEMALLOC; |
1da177e4 | 335 | do { |
7f338fe4 | 336 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
337 | msg.msg_name = NULL; |
338 | msg.msg_namelen = 0; | |
339 | msg.msg_control = NULL; | |
340 | msg.msg_controllen = 0; | |
1da177e4 LT |
341 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
342 | ||
7e2893a1 | 343 | if (send) |
c1696cab | 344 | result = sock_sendmsg(sock, &msg); |
7e2893a1 | 345 | else |
c1696cab | 346 | result = sock_recvmsg(sock, &msg, msg.msg_flags); |
1da177e4 | 347 | |
1da177e4 LT |
348 | if (result <= 0) { |
349 | if (result == 0) | |
350 | result = -EPIPE; /* short read */ | |
351 | break; | |
352 | } | |
9dd5d3ab JB |
353 | if (sent) |
354 | *sent += result; | |
c1696cab | 355 | } while (msg_data_left(&msg)); |
1da177e4 | 356 | |
7f338fe4 | 357 | tsk_restore_flags(current, pflags, PF_MEMALLOC); |
1da177e4 LT |
358 | |
359 | return result; | |
360 | } | |
361 | ||
7fdfd406 | 362 | /* always call with the tx_lock held */ |
9561a7ad | 363 | static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index) |
1da177e4 | 364 | { |
fd8383fd | 365 | struct request *req = blk_mq_rq_from_pdu(cmd); |
5ea8d108 JB |
366 | struct nbd_config *config = nbd->config; |
367 | struct nbd_sock *nsock = config->socks[index]; | |
d61b7f97 | 368 | int result; |
c9f2b6ae AV |
369 | struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)}; |
370 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
371 | struct iov_iter from; | |
1011c1b9 | 372 | unsigned long size = blk_rq_bytes(req); |
429a787b | 373 | struct bio *bio; |
9dc6c806 | 374 | u32 type; |
9561a7ad | 375 | u32 tag = blk_mq_unique_tag(req); |
9dd5d3ab | 376 | int sent = nsock->sent, skip = 0; |
9dc6c806 | 377 | |
c9f2b6ae AV |
378 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); |
379 | ||
aebf526b CH |
380 | switch (req_op(req)) { |
381 | case REQ_OP_DISCARD: | |
9dc6c806 | 382 | type = NBD_CMD_TRIM; |
aebf526b CH |
383 | break; |
384 | case REQ_OP_FLUSH: | |
9dc6c806 | 385 | type = NBD_CMD_FLUSH; |
aebf526b CH |
386 | break; |
387 | case REQ_OP_WRITE: | |
9dc6c806 | 388 | type = NBD_CMD_WRITE; |
aebf526b CH |
389 | break; |
390 | case REQ_OP_READ: | |
9dc6c806 | 391 | type = NBD_CMD_READ; |
aebf526b CH |
392 | break; |
393 | default: | |
394 | return -EIO; | |
395 | } | |
1da177e4 | 396 | |
09fc54cc | 397 | if (rq_data_dir(req) == WRITE && |
5ea8d108 | 398 | (config->flags & NBD_FLAG_READ_ONLY)) { |
09fc54cc CH |
399 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
400 | "Write on read-only\n"); | |
401 | return -EIO; | |
402 | } | |
403 | ||
9dd5d3ab JB |
404 | /* We did a partial send previously, and we at least sent the whole |
405 | * request struct, so just go and send the rest of the pages in the | |
406 | * request. | |
407 | */ | |
408 | if (sent) { | |
409 | if (sent >= sizeof(request)) { | |
410 | skip = sent - sizeof(request); | |
411 | goto send_pages; | |
412 | } | |
413 | iov_iter_advance(&from, sent); | |
414 | } | |
f3733247 | 415 | cmd->index = index; |
799f9a38 | 416 | cmd->cookie = nsock->cookie; |
9dc6c806 | 417 | request.type = htonl(type); |
9561a7ad | 418 | if (type != NBD_CMD_FLUSH) { |
75f187ab AB |
419 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
420 | request.len = htonl(size); | |
421 | } | |
9561a7ad | 422 | memcpy(request.handle, &tag, sizeof(tag)); |
1da177e4 | 423 | |
d18509f5 | 424 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
fd8383fd | 425 | cmd, nbdcmd_to_ascii(type), |
d18509f5 | 426 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
c9f2b6ae | 427 | result = sock_xmit(nbd, index, 1, &from, |
9dd5d3ab | 428 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent); |
1da177e4 | 429 | if (result <= 0) { |
9dd5d3ab JB |
430 | if (result == -ERESTARTSYS) { |
431 | /* If we havne't sent anything we can just return BUSY, | |
432 | * however if we have sent something we need to make | |
433 | * sure we only allow this req to be sent until we are | |
434 | * completely done. | |
435 | */ | |
436 | if (sent) { | |
437 | nsock->pending = req; | |
438 | nsock->sent = sent; | |
439 | } | |
440 | return BLK_MQ_RQ_QUEUE_BUSY; | |
441 | } | |
a897b666 | 442 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 | 443 | "Send control failed (result %d)\n", result); |
f3733247 | 444 | return -EAGAIN; |
1da177e4 | 445 | } |
9dd5d3ab | 446 | send_pages: |
429a787b | 447 | if (type != NBD_CMD_WRITE) |
9dd5d3ab | 448 | goto out; |
429a787b | 449 | |
429a787b JA |
450 | bio = req->bio; |
451 | while (bio) { | |
452 | struct bio *next = bio->bi_next; | |
453 | struct bvec_iter iter; | |
7988613b | 454 | struct bio_vec bvec; |
429a787b JA |
455 | |
456 | bio_for_each_segment(bvec, bio, iter) { | |
457 | bool is_last = !next && bio_iter_last(bvec, iter); | |
d61b7f97 | 458 | int flags = is_last ? 0 : MSG_MORE; |
429a787b | 459 | |
d18509f5 | 460 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
fd8383fd | 461 | cmd, bvec.bv_len); |
c9f2b6ae AV |
462 | iov_iter_bvec(&from, ITER_BVEC | WRITE, |
463 | &bvec, 1, bvec.bv_len); | |
9dd5d3ab JB |
464 | if (skip) { |
465 | if (skip >= iov_iter_count(&from)) { | |
466 | skip -= iov_iter_count(&from); | |
467 | continue; | |
468 | } | |
469 | iov_iter_advance(&from, skip); | |
470 | skip = 0; | |
471 | } | |
472 | result = sock_xmit(nbd, index, 1, &from, flags, &sent); | |
6c92e699 | 473 | if (result <= 0) { |
9dd5d3ab JB |
474 | if (result == -ERESTARTSYS) { |
475 | /* We've already sent the header, we | |
476 | * have no choice but to set pending and | |
477 | * return BUSY. | |
478 | */ | |
479 | nsock->pending = req; | |
480 | nsock->sent = sent; | |
481 | return BLK_MQ_RQ_QUEUE_BUSY; | |
482 | } | |
f4507164 | 483 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
484 | "Send data failed (result %d)\n", |
485 | result); | |
f3733247 | 486 | return -EAGAIN; |
6c92e699 | 487 | } |
429a787b JA |
488 | /* |
489 | * The completion might already have come in, | |
490 | * so break for the last one instead of letting | |
491 | * the iterator do it. This prevents use-after-free | |
492 | * of the bio. | |
493 | */ | |
494 | if (is_last) | |
495 | break; | |
1da177e4 | 496 | } |
429a787b | 497 | bio = next; |
1da177e4 | 498 | } |
9dd5d3ab JB |
499 | out: |
500 | nsock->pending = NULL; | |
501 | nsock->sent = 0; | |
1da177e4 | 502 | return 0; |
1da177e4 LT |
503 | } |
504 | ||
1da177e4 | 505 | /* NULL returned = something went wrong, inform userspace */ |
9561a7ad | 506 | static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index) |
1da177e4 | 507 | { |
5ea8d108 | 508 | struct nbd_config *config = nbd->config; |
1da177e4 LT |
509 | int result; |
510 | struct nbd_reply reply; | |
fd8383fd JB |
511 | struct nbd_cmd *cmd; |
512 | struct request *req = NULL; | |
513 | u16 hwq; | |
9561a7ad | 514 | u32 tag; |
c9f2b6ae AV |
515 | struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)}; |
516 | struct iov_iter to; | |
1da177e4 LT |
517 | |
518 | reply.magic = 0; | |
c9f2b6ae | 519 | iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply)); |
9dd5d3ab | 520 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
1da177e4 | 521 | if (result <= 0) { |
5ea8d108 | 522 | if (!nbd_disconnected(config)) |
9561a7ad JB |
523 | dev_err(disk_to_dev(nbd->disk), |
524 | "Receive control failed (result %d)\n", result); | |
19391830 | 525 | return ERR_PTR(result); |
1da177e4 | 526 | } |
e4b57e08 MF |
527 | |
528 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 529 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 530 | (unsigned long)ntohl(reply.magic)); |
19391830 | 531 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
532 | } |
533 | ||
9561a7ad | 534 | memcpy(&tag, reply.handle, sizeof(u32)); |
4b2f0260 | 535 | |
fd8383fd JB |
536 | hwq = blk_mq_unique_tag_to_hwq(tag); |
537 | if (hwq < nbd->tag_set.nr_hw_queues) | |
538 | req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq], | |
539 | blk_mq_unique_tag_to_tag(tag)); | |
540 | if (!req || !blk_mq_request_started(req)) { | |
541 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n", | |
542 | tag, req); | |
543 | return ERR_PTR(-ENOENT); | |
1da177e4 | 544 | } |
fd8383fd | 545 | cmd = blk_mq_rq_to_pdu(req); |
1da177e4 | 546 | if (ntohl(reply.error)) { |
f4507164 | 547 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 548 | ntohl(reply.error)); |
c103b4da | 549 | req->errors = -EIO; |
fd8383fd | 550 | return cmd; |
1da177e4 LT |
551 | } |
552 | ||
fd8383fd | 553 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd); |
9dc6c806 | 554 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 555 | struct req_iterator iter; |
7988613b | 556 | struct bio_vec bvec; |
5705f702 N |
557 | |
558 | rq_for_each_segment(bvec, req, iter) { | |
c9f2b6ae AV |
559 | iov_iter_bvec(&to, ITER_BVEC | READ, |
560 | &bvec, 1, bvec.bv_len); | |
9dd5d3ab | 561 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
6c92e699 | 562 | if (result <= 0) { |
f4507164 | 563 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 564 | result); |
f3733247 JB |
565 | /* |
566 | * If we've disconnected or we only have 1 | |
567 | * connection then we need to make sure we | |
568 | * complete this request, otherwise error out | |
569 | * and let the timeout stuff handle resubmitting | |
570 | * this request onto another connection. | |
571 | */ | |
5ea8d108 JB |
572 | if (nbd_disconnected(config) || |
573 | config->num_connections <= 1) { | |
f3733247 JB |
574 | req->errors = -EIO; |
575 | return cmd; | |
576 | } | |
577 | return ERR_PTR(-EIO); | |
6c92e699 | 578 | } |
d18509f5 | 579 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
fd8383fd | 580 | cmd, bvec.bv_len); |
1da177e4 | 581 | } |
9561a7ad JB |
582 | } else { |
583 | /* See the comment in nbd_queue_rq. */ | |
584 | wait_for_completion(&cmd->send_complete); | |
1da177e4 | 585 | } |
fd8383fd | 586 | return cmd; |
1da177e4 LT |
587 | } |
588 | ||
9561a7ad | 589 | static void recv_work(struct work_struct *work) |
1da177e4 | 590 | { |
9561a7ad JB |
591 | struct recv_thread_args *args = container_of(work, |
592 | struct recv_thread_args, | |
593 | work); | |
594 | struct nbd_device *nbd = args->nbd; | |
5ea8d108 | 595 | struct nbd_config *config = nbd->config; |
fd8383fd | 596 | struct nbd_cmd *cmd; |
9561a7ad | 597 | int ret = 0; |
1da177e4 | 598 | |
19391830 | 599 | while (1) { |
9561a7ad | 600 | cmd = nbd_read_stat(nbd, args->index); |
fd8383fd | 601 | if (IS_ERR(cmd)) { |
5ea8d108 | 602 | struct nbd_sock *nsock = config->socks[args->index]; |
f3733247 JB |
603 | |
604 | mutex_lock(&nsock->tx_lock); | |
799f9a38 | 605 | nbd_mark_nsock_dead(nbd, nsock, 1); |
f3733247 | 606 | mutex_unlock(&nsock->tx_lock); |
fd8383fd | 607 | ret = PTR_ERR(cmd); |
19391830 MP |
608 | break; |
609 | } | |
610 | ||
fd8383fd | 611 | nbd_end_request(cmd); |
19391830 | 612 | } |
5ea8d108 JB |
613 | atomic_dec(&config->recv_threads); |
614 | wake_up(&config->recv_wq); | |
615 | nbd_config_put(nbd); | |
616 | kfree(args); | |
1da177e4 LT |
617 | } |
618 | ||
fd8383fd | 619 | static void nbd_clear_req(struct request *req, void *data, bool reserved) |
1da177e4 | 620 | { |
fd8383fd | 621 | struct nbd_cmd *cmd; |
1da177e4 | 622 | |
fd8383fd JB |
623 | if (!blk_mq_request_started(req)) |
624 | return; | |
625 | cmd = blk_mq_rq_to_pdu(req); | |
c103b4da | 626 | req->errors = -EIO; |
fd8383fd JB |
627 | nbd_end_request(cmd); |
628 | } | |
629 | ||
630 | static void nbd_clear_que(struct nbd_device *nbd) | |
631 | { | |
2516ab15 | 632 | blk_mq_stop_hw_queues(nbd->disk->queue); |
fd8383fd | 633 | blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL); |
2516ab15 | 634 | blk_mq_start_hw_queues(nbd->disk->queue); |
e78273c8 | 635 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
636 | } |
637 | ||
f3733247 JB |
638 | static int find_fallback(struct nbd_device *nbd, int index) |
639 | { | |
5ea8d108 | 640 | struct nbd_config *config = nbd->config; |
f3733247 | 641 | int new_index = -1; |
5ea8d108 | 642 | struct nbd_sock *nsock = config->socks[index]; |
f3733247 JB |
643 | int fallback = nsock->fallback_index; |
644 | ||
5ea8d108 | 645 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
f3733247 JB |
646 | return new_index; |
647 | ||
5ea8d108 | 648 | if (config->num_connections <= 1) { |
f3733247 JB |
649 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
650 | "Attempted send on invalid socket\n"); | |
651 | return new_index; | |
652 | } | |
653 | ||
5ea8d108 JB |
654 | if (fallback >= 0 && fallback < config->num_connections && |
655 | !config->socks[fallback]->dead) | |
f3733247 JB |
656 | return fallback; |
657 | ||
658 | if (nsock->fallback_index < 0 || | |
5ea8d108 JB |
659 | nsock->fallback_index >= config->num_connections || |
660 | config->socks[nsock->fallback_index]->dead) { | |
f3733247 | 661 | int i; |
5ea8d108 | 662 | for (i = 0; i < config->num_connections; i++) { |
f3733247 JB |
663 | if (i == index) |
664 | continue; | |
5ea8d108 | 665 | if (!config->socks[i]->dead) { |
f3733247 JB |
666 | new_index = i; |
667 | break; | |
668 | } | |
669 | } | |
670 | nsock->fallback_index = new_index; | |
671 | if (new_index < 0) { | |
672 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
673 | "Dead connection, failed to find a fallback\n"); | |
674 | return new_index; | |
675 | } | |
676 | } | |
677 | new_index = nsock->fallback_index; | |
678 | return new_index; | |
679 | } | |
7fdfd406 | 680 | |
560bc4b3 JB |
681 | static int wait_for_reconnect(struct nbd_device *nbd) |
682 | { | |
683 | struct nbd_config *config = nbd->config; | |
684 | if (!config->dead_conn_timeout) | |
685 | return 0; | |
686 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) | |
687 | return 0; | |
688 | wait_event_interruptible_timeout(config->conn_wait, | |
689 | atomic_read(&config->live_connections), | |
690 | config->dead_conn_timeout); | |
691 | return atomic_read(&config->live_connections); | |
692 | } | |
693 | ||
9dd5d3ab | 694 | static int nbd_handle_cmd(struct nbd_cmd *cmd, int index) |
48cf6061 | 695 | { |
fd8383fd JB |
696 | struct request *req = blk_mq_rq_from_pdu(cmd); |
697 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 | 698 | struct nbd_config *config; |
9561a7ad | 699 | struct nbd_sock *nsock; |
9dd5d3ab | 700 | int ret; |
fd8383fd | 701 | |
5ea8d108 JB |
702 | if (!refcount_inc_not_zero(&nbd->config_refs)) { |
703 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
704 | "Socks array is empty\n"); | |
705 | return -EINVAL; | |
706 | } | |
707 | config = nbd->config; | |
708 | ||
709 | if (index >= config->num_connections) { | |
a897b666 JB |
710 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
711 | "Attempted send on invalid socket\n"); | |
5ea8d108 | 712 | nbd_config_put(nbd); |
9dd5d3ab | 713 | return -EINVAL; |
9561a7ad | 714 | } |
48cf6061 | 715 | req->errors = 0; |
f3733247 | 716 | again: |
5ea8d108 | 717 | nsock = config->socks[index]; |
9561a7ad | 718 | mutex_lock(&nsock->tx_lock); |
f3733247 | 719 | if (nsock->dead) { |
560bc4b3 | 720 | int old_index = index; |
f3733247 | 721 | index = find_fallback(nbd, index); |
560bc4b3 | 722 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 723 | if (index < 0) { |
560bc4b3 JB |
724 | if (wait_for_reconnect(nbd)) { |
725 | index = old_index; | |
726 | goto again; | |
727 | } | |
728 | /* All the sockets should already be down at this point, | |
729 | * we just want to make sure that DISCONNECTED is set so | |
730 | * any requests that come in that were queue'ed waiting | |
731 | * for the reconnect timer don't trigger the timer again | |
732 | * and instead just error out. | |
733 | */ | |
734 | sock_shutdown(nbd); | |
735 | nbd_config_put(nbd); | |
736 | return -EIO; | |
5ea8d108 | 737 | } |
f3733247 | 738 | goto again; |
48cf6061 LV |
739 | } |
740 | ||
9dd5d3ab JB |
741 | /* Handle the case that we have a pending request that was partially |
742 | * transmitted that _has_ to be serviced first. We need to call requeue | |
743 | * here so that it gets put _after_ the request that is already on the | |
744 | * dispatch list. | |
745 | */ | |
746 | if (unlikely(nsock->pending && nsock->pending != req)) { | |
747 | blk_mq_requeue_request(req, true); | |
748 | ret = 0; | |
749 | goto out; | |
48cf6061 | 750 | } |
f3733247 JB |
751 | /* |
752 | * Some failures are related to the link going down, so anything that | |
753 | * returns EAGAIN can be retried on a different socket. | |
754 | */ | |
9dd5d3ab | 755 | ret = nbd_send_cmd(nbd, cmd, index); |
f3733247 JB |
756 | if (ret == -EAGAIN) { |
757 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
758 | "Request send failed trying another connection\n"); | |
799f9a38 | 759 | nbd_mark_nsock_dead(nbd, nsock, 1); |
f3733247 JB |
760 | mutex_unlock(&nsock->tx_lock); |
761 | goto again; | |
762 | } | |
9dd5d3ab | 763 | out: |
9561a7ad | 764 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 765 | nbd_config_put(nbd); |
9dd5d3ab | 766 | return ret; |
48cf6061 LV |
767 | } |
768 | ||
fd8383fd JB |
769 | static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
770 | const struct blk_mq_queue_data *bd) | |
1da177e4 | 771 | { |
fd8383fd | 772 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); |
9dd5d3ab | 773 | int ret; |
1da177e4 | 774 | |
9561a7ad JB |
775 | /* |
776 | * Since we look at the bio's to send the request over the network we | |
777 | * need to make sure the completion work doesn't mark this request done | |
778 | * before we are done doing our send. This keeps us from dereferencing | |
779 | * freed data if we have particularly fast completions (ie we get the | |
780 | * completion before we exit sock_xmit on the last bvec) or in the case | |
781 | * that the server is misbehaving (or there was an error) before we're | |
782 | * done sending everything over the wire. | |
783 | */ | |
784 | init_completion(&cmd->send_complete); | |
fd8383fd | 785 | blk_mq_start_request(bd->rq); |
9dd5d3ab JB |
786 | |
787 | /* We can be called directly from the user space process, which means we | |
788 | * could possibly have signals pending so our sendmsg will fail. In | |
789 | * this case we need to return that we are busy, otherwise error out as | |
790 | * appropriate. | |
791 | */ | |
792 | ret = nbd_handle_cmd(cmd, hctx->queue_num); | |
793 | if (ret < 0) | |
794 | ret = BLK_MQ_RQ_QUEUE_ERROR; | |
795 | if (!ret) | |
796 | ret = BLK_MQ_RQ_QUEUE_OK; | |
9561a7ad JB |
797 | complete(&cmd->send_complete); |
798 | ||
9dd5d3ab | 799 | return ret; |
1da177e4 LT |
800 | } |
801 | ||
e46c7287 JB |
802 | static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg, |
803 | bool netlink) | |
23272a67 | 804 | { |
5ea8d108 | 805 | struct nbd_config *config = nbd->config; |
9442b739 | 806 | struct socket *sock; |
9561a7ad JB |
807 | struct nbd_sock **socks; |
808 | struct nbd_sock *nsock; | |
9442b739 JB |
809 | int err; |
810 | ||
811 | sock = sockfd_lookup(arg, &err); | |
812 | if (!sock) | |
813 | return err; | |
23272a67 | 814 | |
e46c7287 JB |
815 | if (!netlink && !nbd->task_setup && |
816 | !test_bit(NBD_BOUND, &config->runtime_flags)) | |
9561a7ad | 817 | nbd->task_setup = current; |
e46c7287 JB |
818 | |
819 | if (!netlink && | |
820 | (nbd->task_setup != current || | |
821 | test_bit(NBD_BOUND, &config->runtime_flags))) { | |
9561a7ad JB |
822 | dev_err(disk_to_dev(nbd->disk), |
823 | "Device being setup by another task"); | |
9b1355d5 | 824 | sockfd_put(sock); |
e46c7287 | 825 | return -EBUSY; |
23272a67 MP |
826 | } |
827 | ||
5ea8d108 | 828 | socks = krealloc(config->socks, (config->num_connections + 1) * |
9561a7ad | 829 | sizeof(struct nbd_sock *), GFP_KERNEL); |
9b1355d5 JB |
830 | if (!socks) { |
831 | sockfd_put(sock); | |
9561a7ad | 832 | return -ENOMEM; |
9b1355d5 | 833 | } |
9561a7ad | 834 | nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL); |
9b1355d5 JB |
835 | if (!nsock) { |
836 | sockfd_put(sock); | |
9561a7ad | 837 | return -ENOMEM; |
9b1355d5 | 838 | } |
9561a7ad | 839 | |
5ea8d108 | 840 | config->socks = socks; |
23272a67 | 841 | |
f3733247 JB |
842 | nsock->fallback_index = -1; |
843 | nsock->dead = false; | |
9561a7ad JB |
844 | mutex_init(&nsock->tx_lock); |
845 | nsock->sock = sock; | |
9dd5d3ab JB |
846 | nsock->pending = NULL; |
847 | nsock->sent = 0; | |
799f9a38 | 848 | nsock->cookie = 0; |
5ea8d108 | 849 | socks[config->num_connections++] = nsock; |
560bc4b3 | 850 | atomic_inc(&config->live_connections); |
23272a67 | 851 | |
9561a7ad | 852 | return 0; |
23272a67 MP |
853 | } |
854 | ||
b7aa3d39 JB |
855 | static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg) |
856 | { | |
857 | struct nbd_config *config = nbd->config; | |
858 | struct socket *sock, *old; | |
859 | struct recv_thread_args *args; | |
860 | int i; | |
861 | int err; | |
862 | ||
863 | sock = sockfd_lookup(arg, &err); | |
864 | if (!sock) | |
865 | return err; | |
866 | ||
867 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
868 | if (!args) { | |
869 | sockfd_put(sock); | |
870 | return -ENOMEM; | |
871 | } | |
872 | ||
873 | for (i = 0; i < config->num_connections; i++) { | |
874 | struct nbd_sock *nsock = config->socks[i]; | |
875 | ||
876 | if (!nsock->dead) | |
877 | continue; | |
878 | ||
879 | mutex_lock(&nsock->tx_lock); | |
880 | if (!nsock->dead) { | |
881 | mutex_unlock(&nsock->tx_lock); | |
882 | continue; | |
883 | } | |
884 | sk_set_memalloc(sock->sk); | |
885 | atomic_inc(&config->recv_threads); | |
886 | refcount_inc(&nbd->config_refs); | |
887 | old = nsock->sock; | |
888 | nsock->fallback_index = -1; | |
889 | nsock->sock = sock; | |
890 | nsock->dead = false; | |
891 | INIT_WORK(&args->work, recv_work); | |
892 | args->index = i; | |
893 | args->nbd = nbd; | |
799f9a38 | 894 | nsock->cookie++; |
b7aa3d39 JB |
895 | mutex_unlock(&nsock->tx_lock); |
896 | sockfd_put(old); | |
897 | ||
898 | /* We take the tx_mutex in an error path in the recv_work, so we | |
899 | * need to queue_work outside of the tx_mutex. | |
900 | */ | |
901 | queue_work(recv_workqueue, &args->work); | |
560bc4b3 JB |
902 | |
903 | atomic_inc(&config->live_connections); | |
904 | wake_up(&config->conn_wait); | |
b7aa3d39 JB |
905 | return 0; |
906 | } | |
907 | sockfd_put(sock); | |
908 | kfree(args); | |
909 | return -ENOSPC; | |
910 | } | |
911 | ||
0e4f0f6f MP |
912 | /* Reset all properties of an NBD device */ |
913 | static void nbd_reset(struct nbd_device *nbd) | |
914 | { | |
5ea8d108 | 915 | nbd->config = NULL; |
0eadf37a | 916 | nbd->tag_set.timeout = 0; |
0e4f0f6f | 917 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
0e4f0f6f MP |
918 | } |
919 | ||
920 | static void nbd_bdev_reset(struct block_device *bdev) | |
921 | { | |
abbbdf12 RMB |
922 | if (bdev->bd_openers > 1) |
923 | return; | |
29eaadc0 | 924 | bd_set_size(bdev, 0); |
0e4f0f6f MP |
925 | if (max_part > 0) { |
926 | blkdev_reread_part(bdev); | |
927 | bdev->bd_invalidated = 1; | |
928 | } | |
929 | } | |
930 | ||
29eaadc0 | 931 | static void nbd_parse_flags(struct nbd_device *nbd) |
d02cf531 | 932 | { |
5ea8d108 JB |
933 | struct nbd_config *config = nbd->config; |
934 | if (config->flags & NBD_FLAG_READ_ONLY) | |
29eaadc0 JB |
935 | set_disk_ro(nbd->disk, true); |
936 | else | |
937 | set_disk_ro(nbd->disk, false); | |
5ea8d108 | 938 | if (config->flags & NBD_FLAG_SEND_TRIM) |
d02cf531 | 939 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
5ea8d108 | 940 | if (config->flags & NBD_FLAG_SEND_FLUSH) |
aafb1eec | 941 | blk_queue_write_cache(nbd->disk->queue, true, false); |
d02cf531 | 942 | else |
aafb1eec | 943 | blk_queue_write_cache(nbd->disk->queue, false, false); |
d02cf531 MP |
944 | } |
945 | ||
9561a7ad JB |
946 | static void send_disconnects(struct nbd_device *nbd) |
947 | { | |
5ea8d108 | 948 | struct nbd_config *config = nbd->config; |
c9f2b6ae AV |
949 | struct nbd_request request = { |
950 | .magic = htonl(NBD_REQUEST_MAGIC), | |
951 | .type = htonl(NBD_CMD_DISC), | |
952 | }; | |
953 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
954 | struct iov_iter from; | |
9561a7ad JB |
955 | int i, ret; |
956 | ||
5ea8d108 | 957 | for (i = 0; i < config->num_connections; i++) { |
c9f2b6ae | 958 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); |
9dd5d3ab | 959 | ret = sock_xmit(nbd, i, 1, &from, 0, NULL); |
9561a7ad JB |
960 | if (ret <= 0) |
961 | dev_err(disk_to_dev(nbd->disk), | |
962 | "Send disconnect failed %d\n", ret); | |
963 | } | |
964 | } | |
965 | ||
29eaadc0 | 966 | static int nbd_disconnect(struct nbd_device *nbd) |
9442b739 | 967 | { |
5ea8d108 | 968 | struct nbd_config *config = nbd->config; |
30d53d9c | 969 | |
5ea8d108 | 970 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
9442b739 | 971 | if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED, |
5ea8d108 | 972 | &config->runtime_flags)) |
9442b739 JB |
973 | send_disconnects(nbd); |
974 | return 0; | |
975 | } | |
976 | ||
29eaadc0 | 977 | static void nbd_clear_sock(struct nbd_device *nbd) |
1a2ad211 | 978 | { |
9442b739 JB |
979 | sock_shutdown(nbd); |
980 | nbd_clear_que(nbd); | |
5ea8d108 | 981 | nbd->task_setup = NULL; |
5ea8d108 JB |
982 | } |
983 | ||
984 | static void nbd_config_put(struct nbd_device *nbd) | |
985 | { | |
986 | if (refcount_dec_and_mutex_lock(&nbd->config_refs, | |
987 | &nbd->config_lock)) { | |
5ea8d108 | 988 | struct nbd_config *config = nbd->config; |
5ea8d108 | 989 | nbd_dev_dbg_close(nbd); |
29eaadc0 | 990 | nbd_size_clear(nbd); |
5ea8d108 JB |
991 | if (test_and_clear_bit(NBD_HAS_PID_FILE, |
992 | &config->runtime_flags)) | |
993 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); | |
994 | nbd->task_recv = NULL; | |
29eaadc0 | 995 | nbd_clear_sock(nbd); |
5ea8d108 JB |
996 | if (config->num_connections) { |
997 | int i; | |
998 | for (i = 0; i < config->num_connections; i++) { | |
999 | sockfd_put(config->socks[i]->sock); | |
1000 | kfree(config->socks[i]); | |
1001 | } | |
1002 | kfree(config->socks); | |
1003 | } | |
1004 | nbd_reset(nbd); | |
1005 | mutex_unlock(&nbd->config_lock); | |
5ea8d108 JB |
1006 | module_put(THIS_MODULE); |
1007 | } | |
9442b739 JB |
1008 | } |
1009 | ||
e46c7287 | 1010 | static int nbd_start_device(struct nbd_device *nbd) |
9442b739 | 1011 | { |
5ea8d108 JB |
1012 | struct nbd_config *config = nbd->config; |
1013 | int num_connections = config->num_connections; | |
9442b739 | 1014 | int error = 0, i; |
1a2ad211 | 1015 | |
9442b739 JB |
1016 | if (nbd->task_recv) |
1017 | return -EBUSY; | |
5ea8d108 | 1018 | if (!config->socks) |
9442b739 JB |
1019 | return -EINVAL; |
1020 | if (num_connections > 1 && | |
5ea8d108 | 1021 | !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) { |
9442b739 | 1022 | dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n"); |
5ea8d108 | 1023 | return -EINVAL; |
9442b739 | 1024 | } |
23272a67 | 1025 | |
5ea8d108 | 1026 | blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections); |
9442b739 | 1027 | nbd->task_recv = current; |
23272a67 | 1028 | |
29eaadc0 | 1029 | nbd_parse_flags(nbd); |
23272a67 | 1030 | |
9442b739 JB |
1031 | error = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
1032 | if (error) { | |
1033 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); | |
5ea8d108 | 1034 | return error; |
1a2ad211 | 1035 | } |
29eaadc0 | 1036 | set_bit(NBD_HAS_PID_FILE, &config->runtime_flags); |
37091fdd | 1037 | |
9442b739 JB |
1038 | nbd_dev_dbg_init(nbd); |
1039 | for (i = 0; i < num_connections; i++) { | |
5ea8d108 JB |
1040 | struct recv_thread_args *args; |
1041 | ||
1042 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
1043 | if (!args) { | |
1044 | sock_shutdown(nbd); | |
1045 | return -ENOMEM; | |
1046 | } | |
1047 | sk_set_memalloc(config->socks[i]->sock->sk); | |
1048 | atomic_inc(&config->recv_threads); | |
1049 | refcount_inc(&nbd->config_refs); | |
1050 | INIT_WORK(&args->work, recv_work); | |
1051 | args->nbd = nbd; | |
1052 | args->index = i; | |
1053 | queue_work(recv_workqueue, &args->work); | |
37091fdd | 1054 | } |
e46c7287 JB |
1055 | return error; |
1056 | } | |
1057 | ||
1058 | static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev) | |
1059 | { | |
1060 | struct nbd_config *config = nbd->config; | |
1061 | int ret; | |
1062 | ||
1063 | ret = nbd_start_device(nbd); | |
1064 | if (ret) | |
1065 | return ret; | |
1066 | ||
1067 | bd_set_size(bdev, config->bytesize); | |
1068 | if (max_part) | |
1069 | bdev->bd_invalidated = 1; | |
1070 | mutex_unlock(&nbd->config_lock); | |
1071 | ret = wait_event_interruptible(config->recv_wq, | |
5ea8d108 | 1072 | atomic_read(&config->recv_threads) == 0); |
e46c7287 | 1073 | if (ret) |
5ea8d108 | 1074 | sock_shutdown(nbd); |
9442b739 | 1075 | mutex_lock(&nbd->config_lock); |
e46c7287 | 1076 | bd_set_size(bdev, 0); |
9442b739 | 1077 | /* user requested, ignore socket errors */ |
5ea8d108 | 1078 | if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags)) |
e46c7287 | 1079 | ret = 0; |
5ea8d108 | 1080 | if (test_bit(NBD_TIMEDOUT, &config->runtime_flags)) |
e46c7287 JB |
1081 | ret = -ETIMEDOUT; |
1082 | return ret; | |
9442b739 JB |
1083 | } |
1084 | ||
29eaadc0 JB |
1085 | static void nbd_clear_sock_ioctl(struct nbd_device *nbd, |
1086 | struct block_device *bdev) | |
1087 | { | |
2516ab15 | 1088 | sock_shutdown(nbd); |
29eaadc0 JB |
1089 | kill_bdev(bdev); |
1090 | nbd_bdev_reset(bdev); | |
e46c7287 JB |
1091 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, |
1092 | &nbd->config->runtime_flags)) | |
1093 | nbd_config_put(nbd); | |
29eaadc0 JB |
1094 | } |
1095 | ||
9442b739 JB |
1096 | /* Must be called with config_lock held */ |
1097 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, | |
1098 | unsigned int cmd, unsigned long arg) | |
1099 | { | |
5ea8d108 JB |
1100 | struct nbd_config *config = nbd->config; |
1101 | ||
9442b739 JB |
1102 | switch (cmd) { |
1103 | case NBD_DISCONNECT: | |
29eaadc0 | 1104 | return nbd_disconnect(nbd); |
9442b739 | 1105 | case NBD_CLEAR_SOCK: |
29eaadc0 JB |
1106 | nbd_clear_sock_ioctl(nbd, bdev); |
1107 | return 0; | |
9442b739 | 1108 | case NBD_SET_SOCK: |
e46c7287 | 1109 | return nbd_add_socket(nbd, arg, false); |
9442b739 | 1110 | case NBD_SET_BLKSIZE: |
29eaadc0 | 1111 | nbd_size_set(nbd, arg, |
5ea8d108 | 1112 | div_s64(config->bytesize, arg)); |
e544541b | 1113 | return 0; |
1da177e4 | 1114 | case NBD_SET_SIZE: |
29eaadc0 | 1115 | nbd_size_set(nbd, config->blksize, |
5ea8d108 | 1116 | div_s64(arg, config->blksize)); |
e544541b | 1117 | return 0; |
37091fdd | 1118 | case NBD_SET_SIZE_BLOCKS: |
29eaadc0 | 1119 | nbd_size_set(nbd, config->blksize, arg); |
e544541b | 1120 | return 0; |
7fdfd406 | 1121 | case NBD_SET_TIMEOUT: |
f8586855 JB |
1122 | if (arg) { |
1123 | nbd->tag_set.timeout = arg * HZ; | |
1124 | blk_queue_rq_timeout(nbd->disk->queue, arg * HZ); | |
1125 | } | |
7fdfd406 | 1126 | return 0; |
1a2ad211 | 1127 | |
2f012508 | 1128 | case NBD_SET_FLAGS: |
5ea8d108 | 1129 | config->flags = arg; |
2f012508 | 1130 | return 0; |
9442b739 | 1131 | case NBD_DO_IT: |
e46c7287 | 1132 | return nbd_start_device_ioctl(nbd, bdev); |
1da177e4 | 1133 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
1134 | /* |
1135 | * This is for compatibility only. The queue is always cleared | |
1136 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
1137 | */ | |
1da177e4 LT |
1138 | return 0; |
1139 | case NBD_PRINT_DEBUG: | |
fd8383fd JB |
1140 | /* |
1141 | * For compatibility only, we no longer keep a list of | |
1142 | * outstanding requests. | |
1143 | */ | |
1da177e4 LT |
1144 | return 0; |
1145 | } | |
1a2ad211 PM |
1146 | return -ENOTTY; |
1147 | } | |
1148 | ||
1149 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
1150 | unsigned int cmd, unsigned long arg) | |
1151 | { | |
f4507164 | 1152 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
e46c7287 JB |
1153 | struct nbd_config *config = nbd->config; |
1154 | int error = -EINVAL; | |
1a2ad211 PM |
1155 | |
1156 | if (!capable(CAP_SYS_ADMIN)) | |
1157 | return -EPERM; | |
1158 | ||
9561a7ad | 1159 | mutex_lock(&nbd->config_lock); |
e46c7287 JB |
1160 | |
1161 | /* Don't allow ioctl operations on a nbd device that was created with | |
1162 | * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine. | |
1163 | */ | |
1164 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1165 | (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK)) | |
1166 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
1167 | else | |
1168 | dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n"); | |
9561a7ad | 1169 | mutex_unlock(&nbd->config_lock); |
1a2ad211 | 1170 | return error; |
1da177e4 LT |
1171 | } |
1172 | ||
5ea8d108 JB |
1173 | static struct nbd_config *nbd_alloc_config(void) |
1174 | { | |
1175 | struct nbd_config *config; | |
1176 | ||
1177 | config = kzalloc(sizeof(struct nbd_config), GFP_NOFS); | |
1178 | if (!config) | |
1179 | return NULL; | |
1180 | atomic_set(&config->recv_threads, 0); | |
1181 | init_waitqueue_head(&config->recv_wq); | |
560bc4b3 | 1182 | init_waitqueue_head(&config->conn_wait); |
5ea8d108 | 1183 | config->blksize = 1024; |
560bc4b3 | 1184 | atomic_set(&config->live_connections, 0); |
5ea8d108 JB |
1185 | try_module_get(THIS_MODULE); |
1186 | return config; | |
1187 | } | |
1188 | ||
1189 | static int nbd_open(struct block_device *bdev, fmode_t mode) | |
1190 | { | |
1191 | struct nbd_device *nbd; | |
1192 | int ret = 0; | |
1193 | ||
1194 | mutex_lock(&nbd_index_mutex); | |
1195 | nbd = bdev->bd_disk->private_data; | |
1196 | if (!nbd) { | |
1197 | ret = -ENXIO; | |
1198 | goto out; | |
1199 | } | |
1200 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1201 | struct nbd_config *config; | |
1202 | ||
1203 | mutex_lock(&nbd->config_lock); | |
1204 | if (refcount_inc_not_zero(&nbd->config_refs)) { | |
1205 | mutex_unlock(&nbd->config_lock); | |
1206 | goto out; | |
1207 | } | |
1208 | config = nbd->config = nbd_alloc_config(); | |
1209 | if (!config) { | |
1210 | ret = -ENOMEM; | |
1211 | mutex_unlock(&nbd->config_lock); | |
1212 | goto out; | |
1213 | } | |
1214 | refcount_set(&nbd->config_refs, 1); | |
1215 | mutex_unlock(&nbd->config_lock); | |
1216 | } | |
1217 | out: | |
1218 | mutex_unlock(&nbd_index_mutex); | |
1219 | return ret; | |
1220 | } | |
1221 | ||
1222 | static void nbd_release(struct gendisk *disk, fmode_t mode) | |
1223 | { | |
1224 | struct nbd_device *nbd = disk->private_data; | |
1225 | nbd_config_put(nbd); | |
1226 | } | |
1227 | ||
83d5cde4 | 1228 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
1229 | { |
1230 | .owner = THIS_MODULE, | |
5ea8d108 JB |
1231 | .open = nbd_open, |
1232 | .release = nbd_release, | |
8a6cfeb6 | 1233 | .ioctl = nbd_ioctl, |
263a3df1 | 1234 | .compat_ioctl = nbd_ioctl, |
1da177e4 LT |
1235 | }; |
1236 | ||
30d53d9c MP |
1237 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
1238 | ||
1239 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
1240 | { | |
1241 | struct nbd_device *nbd = s->private; | |
1242 | ||
1243 | if (nbd->task_recv) | |
1244 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
30d53d9c MP |
1245 | |
1246 | return 0; | |
1247 | } | |
1248 | ||
1249 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
1250 | { | |
1251 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
1252 | } | |
1253 | ||
1254 | static const struct file_operations nbd_dbg_tasks_ops = { | |
1255 | .open = nbd_dbg_tasks_open, | |
1256 | .read = seq_read, | |
1257 | .llseek = seq_lseek, | |
1258 | .release = single_release, | |
1259 | }; | |
1260 | ||
1261 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
1262 | { | |
1263 | struct nbd_device *nbd = s->private; | |
5ea8d108 | 1264 | u32 flags = nbd->config->flags; |
30d53d9c MP |
1265 | |
1266 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
1267 | ||
1268 | seq_puts(s, "Known flags:\n"); | |
1269 | ||
1270 | if (flags & NBD_FLAG_HAS_FLAGS) | |
1271 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
1272 | if (flags & NBD_FLAG_READ_ONLY) | |
1273 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
1274 | if (flags & NBD_FLAG_SEND_FLUSH) | |
1275 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
1276 | if (flags & NBD_FLAG_SEND_TRIM) | |
1277 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
1278 | ||
1279 | return 0; | |
1280 | } | |
1281 | ||
1282 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
1283 | { | |
1284 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
1285 | } | |
1286 | ||
1287 | static const struct file_operations nbd_dbg_flags_ops = { | |
1288 | .open = nbd_dbg_flags_open, | |
1289 | .read = seq_read, | |
1290 | .llseek = seq_lseek, | |
1291 | .release = single_release, | |
1292 | }; | |
1293 | ||
1294 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1295 | { | |
1296 | struct dentry *dir; | |
5ea8d108 | 1297 | struct nbd_config *config = nbd->config; |
27ea43fe MP |
1298 | |
1299 | if (!nbd_dbg_dir) | |
1300 | return -EIO; | |
30d53d9c MP |
1301 | |
1302 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
27ea43fe MP |
1303 | if (!dir) { |
1304 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n", | |
1305 | nbd_name(nbd)); | |
1306 | return -EIO; | |
30d53d9c | 1307 | } |
5ea8d108 | 1308 | config->dbg_dir = dir; |
30d53d9c | 1309 | |
27ea43fe | 1310 | debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); |
5ea8d108 | 1311 | debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize); |
0eadf37a | 1312 | debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout); |
5ea8d108 | 1313 | debugfs_create_u64("blocksize", 0444, dir, &config->blksize); |
d366a0ff | 1314 | debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops); |
30d53d9c MP |
1315 | |
1316 | return 0; | |
1317 | } | |
1318 | ||
1319 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1320 | { | |
5ea8d108 | 1321 | debugfs_remove_recursive(nbd->config->dbg_dir); |
30d53d9c MP |
1322 | } |
1323 | ||
1324 | static int nbd_dbg_init(void) | |
1325 | { | |
1326 | struct dentry *dbg_dir; | |
1327 | ||
1328 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
27ea43fe MP |
1329 | if (!dbg_dir) |
1330 | return -EIO; | |
30d53d9c MP |
1331 | |
1332 | nbd_dbg_dir = dbg_dir; | |
1333 | ||
1334 | return 0; | |
1335 | } | |
1336 | ||
1337 | static void nbd_dbg_close(void) | |
1338 | { | |
1339 | debugfs_remove_recursive(nbd_dbg_dir); | |
1340 | } | |
1341 | ||
1342 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
1343 | ||
1344 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1345 | { | |
1346 | return 0; | |
1347 | } | |
1348 | ||
1349 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1350 | { | |
1351 | } | |
1352 | ||
1353 | static int nbd_dbg_init(void) | |
1354 | { | |
1355 | return 0; | |
1356 | } | |
1357 | ||
1358 | static void nbd_dbg_close(void) | |
1359 | { | |
1360 | } | |
1361 | ||
1362 | #endif | |
1363 | ||
fd8383fd JB |
1364 | static int nbd_init_request(void *data, struct request *rq, |
1365 | unsigned int hctx_idx, unsigned int request_idx, | |
1366 | unsigned int numa_node) | |
1367 | { | |
1368 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
fd8383fd | 1369 | cmd->nbd = data; |
fd8383fd JB |
1370 | return 0; |
1371 | } | |
1372 | ||
f363b089 | 1373 | static const struct blk_mq_ops nbd_mq_ops = { |
fd8383fd | 1374 | .queue_rq = nbd_queue_rq, |
fd8383fd | 1375 | .init_request = nbd_init_request, |
0eadf37a | 1376 | .timeout = nbd_xmit_timeout, |
fd8383fd JB |
1377 | }; |
1378 | ||
b0d9111a JB |
1379 | static void nbd_dev_remove(struct nbd_device *nbd) |
1380 | { | |
1381 | struct gendisk *disk = nbd->disk; | |
b0d9111a JB |
1382 | if (disk) { |
1383 | del_gendisk(disk); | |
1384 | blk_cleanup_queue(disk->queue); | |
1385 | blk_mq_free_tag_set(&nbd->tag_set); | |
1386 | put_disk(disk); | |
1387 | } | |
1388 | kfree(nbd); | |
1389 | } | |
1390 | ||
1391 | static int nbd_dev_add(int index) | |
1392 | { | |
1393 | struct nbd_device *nbd; | |
1394 | struct gendisk *disk; | |
1395 | struct request_queue *q; | |
1396 | int err = -ENOMEM; | |
1397 | ||
1398 | nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL); | |
1399 | if (!nbd) | |
1400 | goto out; | |
1401 | ||
1402 | disk = alloc_disk(1 << part_shift); | |
1403 | if (!disk) | |
1404 | goto out_free_nbd; | |
1405 | ||
1406 | if (index >= 0) { | |
1407 | err = idr_alloc(&nbd_index_idr, nbd, index, index + 1, | |
1408 | GFP_KERNEL); | |
1409 | if (err == -ENOSPC) | |
1410 | err = -EEXIST; | |
1411 | } else { | |
1412 | err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL); | |
1413 | if (err >= 0) | |
1414 | index = err; | |
1415 | } | |
1416 | if (err < 0) | |
1417 | goto out_free_disk; | |
1418 | ||
e46c7287 | 1419 | nbd->index = index; |
b0d9111a JB |
1420 | nbd->disk = disk; |
1421 | nbd->tag_set.ops = &nbd_mq_ops; | |
1422 | nbd->tag_set.nr_hw_queues = 1; | |
1423 | nbd->tag_set.queue_depth = 128; | |
1424 | nbd->tag_set.numa_node = NUMA_NO_NODE; | |
1425 | nbd->tag_set.cmd_size = sizeof(struct nbd_cmd); | |
1426 | nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | | |
1427 | BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING; | |
1428 | nbd->tag_set.driver_data = nbd; | |
1429 | ||
1430 | err = blk_mq_alloc_tag_set(&nbd->tag_set); | |
1431 | if (err) | |
1432 | goto out_free_idr; | |
1433 | ||
1434 | q = blk_mq_init_queue(&nbd->tag_set); | |
1435 | if (IS_ERR(q)) { | |
1436 | err = PTR_ERR(q); | |
1437 | goto out_free_tags; | |
1438 | } | |
1439 | disk->queue = q; | |
1440 | ||
1441 | /* | |
1442 | * Tell the block layer that we are not a rotational device | |
1443 | */ | |
1444 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
1445 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); | |
1446 | disk->queue->limits.discard_granularity = 512; | |
1447 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); | |
b0d9111a JB |
1448 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1449 | disk->queue->limits.max_sectors = 256; | |
1450 | ||
b0d9111a | 1451 | mutex_init(&nbd->config_lock); |
5ea8d108 | 1452 | refcount_set(&nbd->config_refs, 0); |
b0d9111a JB |
1453 | disk->major = NBD_MAJOR; |
1454 | disk->first_minor = index << part_shift; | |
1455 | disk->fops = &nbd_fops; | |
1456 | disk->private_data = nbd; | |
1457 | sprintf(disk->disk_name, "nbd%d", index); | |
b0d9111a JB |
1458 | nbd_reset(nbd); |
1459 | add_disk(disk); | |
1460 | return index; | |
1461 | ||
1462 | out_free_tags: | |
1463 | blk_mq_free_tag_set(&nbd->tag_set); | |
1464 | out_free_idr: | |
1465 | idr_remove(&nbd_index_idr, index); | |
1466 | out_free_disk: | |
1467 | put_disk(disk); | |
1468 | out_free_nbd: | |
1469 | kfree(nbd); | |
1470 | out: | |
1471 | return err; | |
1472 | } | |
1473 | ||
e46c7287 JB |
1474 | static int find_free_cb(int id, void *ptr, void *data) |
1475 | { | |
1476 | struct nbd_device *nbd = ptr; | |
1477 | struct nbd_device **found = data; | |
1478 | ||
1479 | if (!refcount_read(&nbd->config_refs)) { | |
1480 | *found = nbd; | |
1481 | return 1; | |
1482 | } | |
1483 | return 0; | |
1484 | } | |
1485 | ||
1486 | /* Netlink interface. */ | |
1487 | static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = { | |
1488 | [NBD_ATTR_INDEX] = { .type = NLA_U32 }, | |
1489 | [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 }, | |
1490 | [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 }, | |
1491 | [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 }, | |
1492 | [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 }, | |
1493 | [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 }, | |
1494 | [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED}, | |
560bc4b3 | 1495 | [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 }, |
e46c7287 JB |
1496 | }; |
1497 | ||
1498 | static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = { | |
1499 | [NBD_SOCK_FD] = { .type = NLA_U32 }, | |
1500 | }; | |
1501 | ||
1502 | static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info) | |
1503 | { | |
1504 | struct nbd_device *nbd = NULL; | |
1505 | struct nbd_config *config; | |
1506 | int index = -1; | |
1507 | int ret; | |
1508 | ||
1509 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1510 | return -EPERM; | |
1511 | ||
1512 | if (info->attrs[NBD_ATTR_INDEX]) | |
1513 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1514 | if (!info->attrs[NBD_ATTR_SOCKETS]) { | |
1515 | printk(KERN_ERR "nbd: must specify at least one socket\n"); | |
1516 | return -EINVAL; | |
1517 | } | |
1518 | if (!info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1519 | printk(KERN_ERR "nbd: must specify a size in bytes for the device\n"); | |
1520 | return -EINVAL; | |
1521 | } | |
1522 | again: | |
1523 | mutex_lock(&nbd_index_mutex); | |
1524 | if (index == -1) { | |
1525 | ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd); | |
1526 | if (ret == 0) { | |
1527 | int new_index; | |
1528 | new_index = nbd_dev_add(-1); | |
1529 | if (new_index < 0) { | |
1530 | mutex_unlock(&nbd_index_mutex); | |
1531 | printk(KERN_ERR "nbd: failed to add new device\n"); | |
1532 | return ret; | |
1533 | } | |
1534 | nbd = idr_find(&nbd_index_idr, new_index); | |
1535 | } | |
1536 | } else { | |
1537 | nbd = idr_find(&nbd_index_idr, index); | |
1538 | } | |
1539 | mutex_unlock(&nbd_index_mutex); | |
1540 | if (!nbd) { | |
1541 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", | |
1542 | index); | |
1543 | return -EINVAL; | |
1544 | } | |
1545 | ||
1546 | mutex_lock(&nbd->config_lock); | |
1547 | if (refcount_read(&nbd->config_refs)) { | |
1548 | mutex_unlock(&nbd->config_lock); | |
1549 | if (index == -1) | |
1550 | goto again; | |
1551 | printk(KERN_ERR "nbd: nbd%d already in use\n", index); | |
1552 | return -EBUSY; | |
1553 | } | |
1554 | if (WARN_ON(nbd->config)) { | |
1555 | mutex_unlock(&nbd->config_lock); | |
1556 | return -EINVAL; | |
1557 | } | |
1558 | config = nbd->config = nbd_alloc_config(); | |
1559 | if (!nbd->config) { | |
1560 | mutex_unlock(&nbd->config_lock); | |
1561 | printk(KERN_ERR "nbd: couldn't allocate config\n"); | |
1562 | return -ENOMEM; | |
1563 | } | |
1564 | refcount_set(&nbd->config_refs, 1); | |
1565 | set_bit(NBD_BOUND, &config->runtime_flags); | |
1566 | ||
1567 | if (info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1568 | u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]); | |
1569 | nbd_size_set(nbd, config->blksize, | |
1570 | div64_u64(bytes, config->blksize)); | |
1571 | } | |
1572 | if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) { | |
1573 | u64 bsize = | |
1574 | nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]); | |
1575 | nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize)); | |
1576 | } | |
1577 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1578 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1579 | nbd->tag_set.timeout = timeout * HZ; | |
1580 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1581 | } | |
560bc4b3 JB |
1582 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1583 | config->dead_conn_timeout = | |
1584 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1585 | config->dead_conn_timeout *= HZ; | |
1586 | } | |
e46c7287 JB |
1587 | if (info->attrs[NBD_ATTR_SERVER_FLAGS]) |
1588 | config->flags = | |
1589 | nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]); | |
1590 | if (info->attrs[NBD_ATTR_SOCKETS]) { | |
1591 | struct nlattr *attr; | |
1592 | int rem, fd; | |
1593 | ||
1594 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1595 | rem) { | |
1596 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1597 | ||
1598 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1599 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1600 | ret = -EINVAL; | |
1601 | goto out; | |
1602 | } | |
1603 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
1604 | nbd_sock_policy); | |
1605 | if (ret != 0) { | |
1606 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1607 | ret = -EINVAL; | |
1608 | goto out; | |
1609 | } | |
1610 | if (!socks[NBD_SOCK_FD]) | |
1611 | continue; | |
1612 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1613 | ret = nbd_add_socket(nbd, fd, true); | |
1614 | if (ret) | |
1615 | goto out; | |
1616 | } | |
1617 | } | |
1618 | ret = nbd_start_device(nbd); | |
1619 | out: | |
1620 | mutex_unlock(&nbd->config_lock); | |
1621 | if (!ret) { | |
1622 | set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags); | |
1623 | refcount_inc(&nbd->config_refs); | |
1624 | nbd_connect_reply(info, nbd->index); | |
1625 | } | |
1626 | nbd_config_put(nbd); | |
1627 | return ret; | |
1628 | } | |
1629 | ||
1630 | static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info) | |
1631 | { | |
1632 | struct nbd_device *nbd; | |
1633 | int index; | |
1634 | ||
1635 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1636 | return -EPERM; | |
1637 | ||
1638 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1639 | printk(KERN_ERR "nbd: must specify an index to disconnect\n"); | |
1640 | return -EINVAL; | |
1641 | } | |
1642 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1643 | mutex_lock(&nbd_index_mutex); | |
1644 | nbd = idr_find(&nbd_index_idr, index); | |
1645 | mutex_unlock(&nbd_index_mutex); | |
1646 | if (!nbd) { | |
1647 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", | |
1648 | index); | |
1649 | return -EINVAL; | |
1650 | } | |
1651 | if (!refcount_inc_not_zero(&nbd->config_refs)) | |
1652 | return 0; | |
1653 | mutex_lock(&nbd->config_lock); | |
1654 | nbd_disconnect(nbd); | |
1655 | mutex_unlock(&nbd->config_lock); | |
1656 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, | |
1657 | &nbd->config->runtime_flags)) | |
1658 | nbd_config_put(nbd); | |
1659 | nbd_config_put(nbd); | |
1660 | return 0; | |
1661 | } | |
1662 | ||
b7aa3d39 JB |
1663 | static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info) |
1664 | { | |
1665 | struct nbd_device *nbd = NULL; | |
1666 | struct nbd_config *config; | |
1667 | int index; | |
1668 | int ret = -EINVAL; | |
1669 | ||
1670 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1671 | return -EPERM; | |
1672 | ||
1673 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1674 | printk(KERN_ERR "nbd: must specify a device to reconfigure\n"); | |
1675 | return -EINVAL; | |
1676 | } | |
1677 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1678 | mutex_lock(&nbd_index_mutex); | |
1679 | nbd = idr_find(&nbd_index_idr, index); | |
1680 | mutex_unlock(&nbd_index_mutex); | |
1681 | if (!nbd) { | |
1682 | printk(KERN_ERR "nbd: couldn't find a device at index %d\n", | |
1683 | index); | |
1684 | return -EINVAL; | |
1685 | } | |
1686 | ||
1687 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1688 | dev_err(nbd_to_dev(nbd), | |
1689 | "not configured, cannot reconfigure\n"); | |
1690 | return -EINVAL; | |
1691 | } | |
1692 | ||
1693 | mutex_lock(&nbd->config_lock); | |
1694 | config = nbd->config; | |
1695 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1696 | !nbd->task_recv) { | |
1697 | dev_err(nbd_to_dev(nbd), | |
1698 | "not configured, cannot reconfigure\n"); | |
1699 | goto out; | |
1700 | } | |
1701 | ||
1702 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1703 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1704 | nbd->tag_set.timeout = timeout * HZ; | |
1705 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1706 | } | |
560bc4b3 JB |
1707 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1708 | config->dead_conn_timeout = | |
1709 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1710 | config->dead_conn_timeout *= HZ; | |
1711 | } | |
b7aa3d39 JB |
1712 | |
1713 | if (info->attrs[NBD_ATTR_SOCKETS]) { | |
1714 | struct nlattr *attr; | |
1715 | int rem, fd; | |
1716 | ||
1717 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1718 | rem) { | |
1719 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1720 | ||
1721 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1722 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1723 | ret = -EINVAL; | |
1724 | goto out; | |
1725 | } | |
1726 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
1727 | nbd_sock_policy); | |
1728 | if (ret != 0) { | |
1729 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1730 | ret = -EINVAL; | |
1731 | goto out; | |
1732 | } | |
1733 | if (!socks[NBD_SOCK_FD]) | |
1734 | continue; | |
1735 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1736 | ret = nbd_reconnect_socket(nbd, fd); | |
1737 | if (ret) { | |
1738 | if (ret == -ENOSPC) | |
1739 | ret = 0; | |
1740 | goto out; | |
1741 | } | |
1742 | dev_info(nbd_to_dev(nbd), "reconnected socket\n"); | |
1743 | } | |
1744 | } | |
1745 | out: | |
1746 | mutex_unlock(&nbd->config_lock); | |
1747 | nbd_config_put(nbd); | |
1748 | return ret; | |
1749 | } | |
1750 | ||
e46c7287 JB |
1751 | static const struct genl_ops nbd_connect_genl_ops[] = { |
1752 | { | |
1753 | .cmd = NBD_CMD_CONNECT, | |
1754 | .policy = nbd_attr_policy, | |
1755 | .doit = nbd_genl_connect, | |
1756 | }, | |
1757 | { | |
1758 | .cmd = NBD_CMD_DISCONNECT, | |
1759 | .policy = nbd_attr_policy, | |
1760 | .doit = nbd_genl_disconnect, | |
1761 | }, | |
b7aa3d39 JB |
1762 | { |
1763 | .cmd = NBD_CMD_RECONFIGURE, | |
1764 | .policy = nbd_attr_policy, | |
1765 | .doit = nbd_genl_reconfigure, | |
1766 | }, | |
e46c7287 JB |
1767 | }; |
1768 | ||
799f9a38 JB |
1769 | static const struct genl_multicast_group nbd_mcast_grps[] = { |
1770 | { .name = NBD_GENL_MCAST_GROUP_NAME, }, | |
1771 | }; | |
1772 | ||
e46c7287 JB |
1773 | static struct genl_family nbd_genl_family __ro_after_init = { |
1774 | .hdrsize = 0, | |
1775 | .name = NBD_GENL_FAMILY_NAME, | |
1776 | .version = NBD_GENL_VERSION, | |
1777 | .module = THIS_MODULE, | |
1778 | .ops = nbd_connect_genl_ops, | |
1779 | .n_ops = ARRAY_SIZE(nbd_connect_genl_ops), | |
1780 | .maxattr = NBD_ATTR_MAX, | |
799f9a38 JB |
1781 | .mcgrps = nbd_mcast_grps, |
1782 | .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps), | |
e46c7287 JB |
1783 | }; |
1784 | ||
1785 | static void nbd_connect_reply(struct genl_info *info, int index) | |
1786 | { | |
1787 | struct sk_buff *skb; | |
1788 | void *msg_head; | |
1789 | int ret; | |
1790 | ||
1791 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
1792 | if (!skb) | |
1793 | return; | |
1794 | msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0, | |
1795 | NBD_CMD_CONNECT); | |
1796 | if (!msg_head) { | |
1797 | nlmsg_free(skb); | |
1798 | return; | |
1799 | } | |
1800 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
1801 | if (ret) { | |
1802 | nlmsg_free(skb); | |
1803 | return; | |
1804 | } | |
1805 | genlmsg_end(skb, msg_head); | |
1806 | genlmsg_reply(skb, info); | |
1807 | } | |
1da177e4 | 1808 | |
799f9a38 JB |
1809 | static void nbd_mcast_index(int index) |
1810 | { | |
1811 | struct sk_buff *skb; | |
1812 | void *msg_head; | |
1813 | int ret; | |
1814 | ||
1815 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
1816 | if (!skb) | |
1817 | return; | |
1818 | msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0, | |
1819 | NBD_CMD_LINK_DEAD); | |
1820 | if (!msg_head) { | |
1821 | nlmsg_free(skb); | |
1822 | return; | |
1823 | } | |
1824 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
1825 | if (ret) { | |
1826 | nlmsg_free(skb); | |
1827 | return; | |
1828 | } | |
1829 | genlmsg_end(skb, msg_head); | |
1830 | genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL); | |
1831 | } | |
1832 | ||
1833 | static void nbd_dead_link_work(struct work_struct *work) | |
1834 | { | |
1835 | struct link_dead_args *args = container_of(work, struct link_dead_args, | |
1836 | work); | |
1837 | nbd_mcast_index(args->index); | |
1838 | kfree(args); | |
1839 | } | |
1840 | ||
1da177e4 LT |
1841 | static int __init nbd_init(void) |
1842 | { | |
1da177e4 LT |
1843 | int i; |
1844 | ||
5b7b18cc | 1845 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 1846 | |
d71a6d73 | 1847 | if (max_part < 0) { |
7742ce4a | 1848 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
1849 | return -EINVAL; |
1850 | } | |
1851 | ||
1852 | part_shift = 0; | |
5988ce23 | 1853 | if (max_part > 0) { |
d71a6d73 LV |
1854 | part_shift = fls(max_part); |
1855 | ||
5988ce23 NK |
1856 | /* |
1857 | * Adjust max_part according to part_shift as it is exported | |
1858 | * to user space so that user can know the max number of | |
1859 | * partition kernel should be able to manage. | |
1860 | * | |
1861 | * Note that -1 is required because partition 0 is reserved | |
1862 | * for the whole disk. | |
1863 | */ | |
1864 | max_part = (1UL << part_shift) - 1; | |
1865 | } | |
1866 | ||
3b271082 NK |
1867 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
1868 | return -EINVAL; | |
1869 | ||
1870 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
1871 | return -EINVAL; | |
124d6db0 JB |
1872 | recv_workqueue = alloc_workqueue("knbd-recv", |
1873 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
1874 | if (!recv_workqueue) | |
1875 | return -ENOMEM; | |
3b271082 | 1876 | |
6330a2d0 JB |
1877 | if (register_blkdev(NBD_MAJOR, "nbd")) { |
1878 | destroy_workqueue(recv_workqueue); | |
b0d9111a | 1879 | return -EIO; |
6330a2d0 | 1880 | } |
1da177e4 | 1881 | |
e46c7287 JB |
1882 | if (genl_register_family(&nbd_genl_family)) { |
1883 | unregister_blkdev(NBD_MAJOR, "nbd"); | |
1884 | destroy_workqueue(recv_workqueue); | |
1885 | return -EINVAL; | |
1886 | } | |
30d53d9c MP |
1887 | nbd_dbg_init(); |
1888 | ||
b0d9111a JB |
1889 | mutex_lock(&nbd_index_mutex); |
1890 | for (i = 0; i < nbds_max; i++) | |
1891 | nbd_dev_add(i); | |
1892 | mutex_unlock(&nbd_index_mutex); | |
1893 | return 0; | |
1894 | } | |
1da177e4 | 1895 | |
b0d9111a JB |
1896 | static int nbd_exit_cb(int id, void *ptr, void *data) |
1897 | { | |
1898 | struct nbd_device *nbd = ptr; | |
1899 | nbd_dev_remove(nbd); | |
1da177e4 | 1900 | return 0; |
1da177e4 LT |
1901 | } |
1902 | ||
1903 | static void __exit nbd_cleanup(void) | |
1904 | { | |
30d53d9c MP |
1905 | nbd_dbg_close(); |
1906 | ||
b0d9111a JB |
1907 | idr_for_each(&nbd_index_idr, &nbd_exit_cb, NULL); |
1908 | idr_destroy(&nbd_index_idr); | |
e46c7287 | 1909 | genl_unregister_family(&nbd_genl_family); |
124d6db0 | 1910 | destroy_workqueue(recv_workqueue); |
1da177e4 | 1911 | unregister_blkdev(NBD_MAJOR, "nbd"); |
1da177e4 LT |
1912 | } |
1913 | ||
1914 | module_init(nbd_init); | |
1915 | module_exit(nbd_cleanup); | |
1916 | ||
1917 | MODULE_DESCRIPTION("Network Block Device"); | |
1918 | MODULE_LICENSE("GPL"); | |
1919 | ||
40be0c28 | 1920 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
1921 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
1922 | module_param(max_part, int, 0444); | |
1923 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |