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> | |
f1083048 | 21 | #include <linux/sched/mm.h> |
1da177e4 LT |
22 | #include <linux/fs.h> |
23 | #include <linux/bio.h> | |
24 | #include <linux/stat.h> | |
25 | #include <linux/errno.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/ioctl.h> | |
2a48fc0a | 28 | #include <linux/mutex.h> |
4b2f0260 HX |
29 | #include <linux/compiler.h> |
30 | #include <linux/err.h> | |
31 | #include <linux/kernel.h> | |
5a0e3ad6 | 32 | #include <linux/slab.h> |
1da177e4 | 33 | #include <net/sock.h> |
91cf45f0 | 34 | #include <linux/net.h> |
48cf6061 | 35 | #include <linux/kthread.h> |
b9c495bb | 36 | #include <linux/types.h> |
30d53d9c | 37 | #include <linux/debugfs.h> |
fd8383fd | 38 | #include <linux/blk-mq.h> |
1da177e4 | 39 | |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
1da177e4 LT |
41 | #include <asm/types.h> |
42 | ||
43 | #include <linux/nbd.h> | |
e46c7287 JB |
44 | #include <linux/nbd-netlink.h> |
45 | #include <net/genetlink.h> | |
1da177e4 | 46 | |
b0d9111a JB |
47 | static DEFINE_IDR(nbd_index_idr); |
48 | static DEFINE_MUTEX(nbd_index_mutex); | |
47d902b9 | 49 | static int nbd_total_devices = 0; |
b0d9111a | 50 | |
9561a7ad JB |
51 | struct nbd_sock { |
52 | struct socket *sock; | |
53 | struct mutex tx_lock; | |
9dd5d3ab JB |
54 | struct request *pending; |
55 | int sent; | |
f3733247 JB |
56 | bool dead; |
57 | int fallback_index; | |
799f9a38 | 58 | int cookie; |
9561a7ad JB |
59 | }; |
60 | ||
5ea8d108 JB |
61 | struct recv_thread_args { |
62 | struct work_struct work; | |
63 | struct nbd_device *nbd; | |
64 | int index; | |
65 | }; | |
66 | ||
799f9a38 JB |
67 | struct link_dead_args { |
68 | struct work_struct work; | |
69 | int index; | |
70 | }; | |
71 | ||
9b4a6ba9 JB |
72 | #define NBD_TIMEDOUT 0 |
73 | #define NBD_DISCONNECT_REQUESTED 1 | |
9561a7ad | 74 | #define NBD_DISCONNECTED 2 |
5ea8d108 | 75 | #define NBD_HAS_PID_FILE 3 |
e46c7287 JB |
76 | #define NBD_HAS_CONFIG_REF 4 |
77 | #define NBD_BOUND 5 | |
a2c97909 | 78 | #define NBD_DESTROY_ON_DISCONNECT 6 |
9b0b6258 | 79 | #define NBD_DISCONNECT_ON_CLOSE 7 |
9b4a6ba9 | 80 | |
5ea8d108 | 81 | struct nbd_config { |
22d109c1 | 82 | u32 flags; |
9b4a6ba9 | 83 | unsigned long runtime_flags; |
560bc4b3 | 84 | u64 dead_conn_timeout; |
13e71d69 | 85 | |
5ea8d108 | 86 | struct nbd_sock **socks; |
9561a7ad | 87 | int num_connections; |
560bc4b3 JB |
88 | atomic_t live_connections; |
89 | wait_queue_head_t conn_wait; | |
5ea8d108 | 90 | |
9561a7ad JB |
91 | atomic_t recv_threads; |
92 | wait_queue_head_t recv_wq; | |
ef77b515 | 93 | loff_t blksize; |
b9c495bb | 94 | loff_t bytesize; |
30d53d9c MP |
95 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
96 | struct dentry *dbg_dir; | |
97 | #endif | |
13e71d69 MP |
98 | }; |
99 | ||
5ea8d108 JB |
100 | struct nbd_device { |
101 | struct blk_mq_tag_set tag_set; | |
102 | ||
e46c7287 | 103 | int index; |
5ea8d108 | 104 | refcount_t config_refs; |
c6a4759e | 105 | refcount_t refs; |
5ea8d108 JB |
106 | struct nbd_config *config; |
107 | struct mutex config_lock; | |
108 | struct gendisk *disk; | |
109 | ||
c6a4759e | 110 | struct list_head list; |
5ea8d108 JB |
111 | struct task_struct *task_recv; |
112 | struct task_struct *task_setup; | |
113 | }; | |
114 | ||
ced413c5 JB |
115 | #define NBD_CMD_REQUEUED 1 |
116 | ||
fd8383fd JB |
117 | struct nbd_cmd { |
118 | struct nbd_device *nbd; | |
05ee6166 | 119 | struct mutex lock; |
f3733247 | 120 | int index; |
799f9a38 | 121 | int cookie; |
2a842aca | 122 | blk_status_t status; |
ced413c5 | 123 | unsigned long flags; |
05ee6166 | 124 | u32 cmd_cookie; |
fd8383fd JB |
125 | }; |
126 | ||
30d53d9c MP |
127 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
128 | static struct dentry *nbd_dbg_dir; | |
129 | #endif | |
130 | ||
131 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
132 | ||
f4507164 | 133 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 134 | |
9c7a4169 | 135 | static unsigned int nbds_max = 16; |
7a8362a0 | 136 | static int max_part = 16; |
124d6db0 | 137 | static struct workqueue_struct *recv_workqueue; |
b0d9111a | 138 | static int part_shift; |
1da177e4 | 139 | |
9442b739 JB |
140 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
141 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
5ea8d108 | 142 | static void nbd_config_put(struct nbd_device *nbd); |
e46c7287 | 143 | static void nbd_connect_reply(struct genl_info *info, int index); |
47d902b9 | 144 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info); |
799f9a38 | 145 | static void nbd_dead_link_work(struct work_struct *work); |
9b0b6258 | 146 | static void nbd_disconnect_and_put(struct nbd_device *nbd); |
9442b739 | 147 | |
d18509f5 | 148 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 149 | { |
d18509f5 | 150 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
151 | } |
152 | ||
ced413c5 JB |
153 | static void nbd_requeue_cmd(struct nbd_cmd *cmd) |
154 | { | |
155 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
156 | ||
157 | if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags)) | |
158 | blk_mq_requeue_request(req, true); | |
159 | } | |
160 | ||
05ee6166 JB |
161 | #define NBD_COOKIE_BITS 32 |
162 | ||
163 | static u64 nbd_cmd_handle(struct nbd_cmd *cmd) | |
164 | { | |
165 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
166 | u32 tag = blk_mq_unique_tag(req); | |
167 | u64 cookie = cmd->cmd_cookie; | |
168 | ||
169 | return (cookie << NBD_COOKIE_BITS) | tag; | |
170 | } | |
171 | ||
172 | static u32 nbd_handle_to_tag(u64 handle) | |
173 | { | |
174 | return (u32)handle; | |
175 | } | |
176 | ||
177 | static u32 nbd_handle_to_cookie(u64 handle) | |
178 | { | |
179 | return (u32)(handle >> NBD_COOKIE_BITS); | |
180 | } | |
181 | ||
1da177e4 LT |
182 | static const char *nbdcmd_to_ascii(int cmd) |
183 | { | |
184 | switch (cmd) { | |
185 | case NBD_CMD_READ: return "read"; | |
186 | case NBD_CMD_WRITE: return "write"; | |
187 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 188 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 189 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
190 | } |
191 | return "invalid"; | |
192 | } | |
1da177e4 | 193 | |
5ea8d108 JB |
194 | static ssize_t pid_show(struct device *dev, |
195 | struct device_attribute *attr, char *buf) | |
196 | { | |
197 | struct gendisk *disk = dev_to_disk(dev); | |
198 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; | |
199 | ||
200 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); | |
201 | } | |
202 | ||
dfbde552 | 203 | static const struct device_attribute pid_attr = { |
5ea8d108 JB |
204 | .attr = { .name = "pid", .mode = S_IRUGO}, |
205 | .show = pid_show, | |
206 | }; | |
207 | ||
c6a4759e JB |
208 | static void nbd_dev_remove(struct nbd_device *nbd) |
209 | { | |
210 | struct gendisk *disk = nbd->disk; | |
edee2e82 JB |
211 | struct request_queue *q; |
212 | ||
c6a4759e | 213 | if (disk) { |
edee2e82 | 214 | q = disk->queue; |
c6a4759e | 215 | del_gendisk(disk); |
edee2e82 | 216 | blk_cleanup_queue(q); |
c6a4759e | 217 | blk_mq_free_tag_set(&nbd->tag_set); |
a2c97909 | 218 | disk->private_data = NULL; |
c6a4759e JB |
219 | put_disk(disk); |
220 | } | |
221 | kfree(nbd); | |
222 | } | |
223 | ||
224 | static void nbd_put(struct nbd_device *nbd) | |
225 | { | |
226 | if (refcount_dec_and_mutex_lock(&nbd->refs, | |
227 | &nbd_index_mutex)) { | |
228 | idr_remove(&nbd_index_idr, nbd->index); | |
229 | mutex_unlock(&nbd_index_mutex); | |
230 | nbd_dev_remove(nbd); | |
231 | } | |
232 | } | |
233 | ||
799f9a38 JB |
234 | static int nbd_disconnected(struct nbd_config *config) |
235 | { | |
236 | return test_bit(NBD_DISCONNECTED, &config->runtime_flags) || | |
237 | test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); | |
238 | } | |
239 | ||
240 | static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock, | |
241 | int notify) | |
f3733247 | 242 | { |
799f9a38 JB |
243 | if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) { |
244 | struct link_dead_args *args; | |
245 | args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO); | |
246 | if (args) { | |
247 | INIT_WORK(&args->work, nbd_dead_link_work); | |
248 | args->index = nbd->index; | |
249 | queue_work(system_wq, &args->work); | |
250 | } | |
251 | } | |
560bc4b3 | 252 | if (!nsock->dead) { |
f3733247 | 253 | kernel_sock_shutdown(nsock->sock, SHUT_RDWR); |
560bc4b3 JB |
254 | atomic_dec(&nbd->config->live_connections); |
255 | } | |
f3733247 JB |
256 | nsock->dead = true; |
257 | nsock->pending = NULL; | |
258 | nsock->sent = 0; | |
259 | } | |
260 | ||
29eaadc0 | 261 | static void nbd_size_clear(struct nbd_device *nbd) |
37091fdd | 262 | { |
5ea8d108 | 263 | if (nbd->config->bytesize) { |
5ea8d108 JB |
264 | set_capacity(nbd->disk, 0); |
265 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
266 | } | |
37091fdd MP |
267 | } |
268 | ||
29eaadc0 | 269 | static void nbd_size_update(struct nbd_device *nbd) |
37091fdd | 270 | { |
5ea8d108 | 271 | struct nbd_config *config = nbd->config; |
00946218 JB |
272 | struct block_device *bdev = bdget_disk(nbd->disk, 0); |
273 | ||
5ea8d108 JB |
274 | blk_queue_logical_block_size(nbd->disk->queue, config->blksize); |
275 | blk_queue_physical_block_size(nbd->disk->queue, config->blksize); | |
5ea8d108 | 276 | set_capacity(nbd->disk, config->bytesize >> 9); |
00946218 | 277 | if (bdev) { |
c2912ca3 | 278 | if (bdev->bd_disk) { |
00946218 | 279 | bd_set_size(bdev, config->bytesize); |
c2912ca3 JK |
280 | set_blocksize(bdev, config->blksize); |
281 | } else | |
00946218 JB |
282 | bdev->bd_invalidated = 1; |
283 | bdput(bdev); | |
284 | } | |
37091fdd MP |
285 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); |
286 | } | |
287 | ||
29eaadc0 JB |
288 | static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize, |
289 | loff_t nr_blocks) | |
37091fdd | 290 | { |
5ea8d108 JB |
291 | struct nbd_config *config = nbd->config; |
292 | config->blksize = blocksize; | |
293 | config->bytesize = blocksize * nr_blocks; | |
a477d005 JB |
294 | if (nbd->task_recv != NULL) |
295 | nbd_size_update(nbd); | |
37091fdd MP |
296 | } |
297 | ||
1e388ae0 | 298 | static void nbd_complete_rq(struct request *req) |
1da177e4 | 299 | { |
1e388ae0 | 300 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
1da177e4 | 301 | |
1e388ae0 CH |
302 | dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd, |
303 | cmd->status ? "failed" : "done"); | |
1da177e4 | 304 | |
1e388ae0 | 305 | blk_mq_end_request(req, cmd->status); |
1da177e4 LT |
306 | } |
307 | ||
e018e757 MP |
308 | /* |
309 | * Forcibly shutdown the socket causing all listeners to error | |
310 | */ | |
36e47bee | 311 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 312 | { |
5ea8d108 | 313 | struct nbd_config *config = nbd->config; |
9561a7ad | 314 | int i; |
23272a67 | 315 | |
5ea8d108 | 316 | if (config->num_connections == 0) |
9561a7ad | 317 | return; |
5ea8d108 | 318 | if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
260bbce4 | 319 | return; |
23272a67 | 320 | |
5ea8d108 JB |
321 | for (i = 0; i < config->num_connections; i++) { |
322 | struct nbd_sock *nsock = config->socks[i]; | |
9561a7ad | 323 | mutex_lock(&nsock->tx_lock); |
799f9a38 | 324 | nbd_mark_nsock_dead(nbd, nsock, 0); |
9561a7ad JB |
325 | mutex_unlock(&nsock->tx_lock); |
326 | } | |
327 | dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n"); | |
7fdfd406 PC |
328 | } |
329 | ||
0eadf37a JB |
330 | static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req, |
331 | bool reserved) | |
7fdfd406 | 332 | { |
0eadf37a JB |
333 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
334 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 JB |
335 | struct nbd_config *config; |
336 | ||
337 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
2a842aca | 338 | cmd->status = BLK_STS_TIMEOUT; |
5ea8d108 JB |
339 | return BLK_EH_HANDLED; |
340 | } | |
5ea8d108 | 341 | config = nbd->config; |
dcc909d9 | 342 | |
05ee6166 JB |
343 | if (!mutex_trylock(&cmd->lock)) |
344 | return BLK_EH_RESET_TIMER; | |
345 | ||
5ea8d108 | 346 | if (config->num_connections > 1) { |
f3733247 JB |
347 | dev_err_ratelimited(nbd_to_dev(nbd), |
348 | "Connection timed out, retrying\n"); | |
f3733247 JB |
349 | /* |
350 | * Hooray we have more connections, requeue this IO, the submit | |
351 | * path will put it on a real connection. | |
352 | */ | |
5ea8d108 JB |
353 | if (config->socks && config->num_connections > 1) { |
354 | if (cmd->index < config->num_connections) { | |
f3733247 | 355 | struct nbd_sock *nsock = |
5ea8d108 | 356 | config->socks[cmd->index]; |
f3733247 | 357 | mutex_lock(&nsock->tx_lock); |
799f9a38 JB |
358 | /* We can have multiple outstanding requests, so |
359 | * we don't want to mark the nsock dead if we've | |
360 | * already reconnected with a new socket, so | |
361 | * only mark it dead if its the same socket we | |
362 | * were sent out on. | |
363 | */ | |
364 | if (cmd->cookie == nsock->cookie) | |
365 | nbd_mark_nsock_dead(nbd, nsock, 1); | |
f3733247 JB |
366 | mutex_unlock(&nsock->tx_lock); |
367 | } | |
05ee6166 | 368 | mutex_unlock(&cmd->lock); |
ced413c5 | 369 | nbd_requeue_cmd(cmd); |
5ea8d108 | 370 | nbd_config_put(nbd); |
f3733247 JB |
371 | return BLK_EH_NOT_HANDLED; |
372 | } | |
f3733247 JB |
373 | } else { |
374 | dev_err_ratelimited(nbd_to_dev(nbd), | |
375 | "Connection timed out\n"); | |
376 | } | |
5ea8d108 | 377 | set_bit(NBD_TIMEDOUT, &config->runtime_flags); |
2a842aca | 378 | cmd->status = BLK_STS_IOERR; |
05ee6166 | 379 | mutex_unlock(&cmd->lock); |
9561a7ad | 380 | sock_shutdown(nbd); |
5ea8d108 JB |
381 | nbd_config_put(nbd); |
382 | ||
0eadf37a | 383 | return BLK_EH_HANDLED; |
7fdfd406 PC |
384 | } |
385 | ||
1da177e4 LT |
386 | /* |
387 | * Send or receive packet. | |
388 | */ | |
c9f2b6ae | 389 | static int sock_xmit(struct nbd_device *nbd, int index, int send, |
9dd5d3ab | 390 | struct iov_iter *iter, int msg_flags, int *sent) |
1da177e4 | 391 | { |
5ea8d108 JB |
392 | struct nbd_config *config = nbd->config; |
393 | struct socket *sock = config->socks[index]->sock; | |
1da177e4 LT |
394 | int result; |
395 | struct msghdr msg; | |
f1083048 | 396 | unsigned int noreclaim_flag; |
1da177e4 | 397 | |
ffc41cf8 | 398 | if (unlikely(!sock)) { |
a897b666 | 399 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
400 | "Attempted %s on closed socket in sock_xmit\n", |
401 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
402 | return -EINVAL; |
403 | } | |
404 | ||
c9f2b6ae | 405 | msg.msg_iter = *iter; |
c1696cab | 406 | |
f1083048 | 407 | noreclaim_flag = memalloc_noreclaim_save(); |
1da177e4 | 408 | do { |
7f338fe4 | 409 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
410 | msg.msg_name = NULL; |
411 | msg.msg_namelen = 0; | |
412 | msg.msg_control = NULL; | |
413 | msg.msg_controllen = 0; | |
1da177e4 LT |
414 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
415 | ||
7e2893a1 | 416 | if (send) |
c1696cab | 417 | result = sock_sendmsg(sock, &msg); |
7e2893a1 | 418 | else |
c1696cab | 419 | result = sock_recvmsg(sock, &msg, msg.msg_flags); |
1da177e4 | 420 | |
1da177e4 LT |
421 | if (result <= 0) { |
422 | if (result == 0) | |
423 | result = -EPIPE; /* short read */ | |
424 | break; | |
425 | } | |
9dd5d3ab JB |
426 | if (sent) |
427 | *sent += result; | |
c1696cab | 428 | } while (msg_data_left(&msg)); |
1da177e4 | 429 | |
f1083048 | 430 | memalloc_noreclaim_restore(noreclaim_flag); |
1da177e4 LT |
431 | |
432 | return result; | |
433 | } | |
434 | ||
32e67a3a JB |
435 | /* |
436 | * Different settings for sk->sk_sndtimeo can result in different return values | |
437 | * if there is a signal pending when we enter sendmsg, because reasons? | |
438 | */ | |
439 | static inline int was_interrupted(int result) | |
440 | { | |
441 | return result == -ERESTARTSYS || result == -EINTR; | |
442 | } | |
443 | ||
7fdfd406 | 444 | /* always call with the tx_lock held */ |
9561a7ad | 445 | static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index) |
1da177e4 | 446 | { |
fd8383fd | 447 | struct request *req = blk_mq_rq_from_pdu(cmd); |
5ea8d108 JB |
448 | struct nbd_config *config = nbd->config; |
449 | struct nbd_sock *nsock = config->socks[index]; | |
d61b7f97 | 450 | int result; |
c9f2b6ae AV |
451 | struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)}; |
452 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
453 | struct iov_iter from; | |
1011c1b9 | 454 | unsigned long size = blk_rq_bytes(req); |
429a787b | 455 | struct bio *bio; |
05ee6166 | 456 | u64 handle; |
9dc6c806 | 457 | u32 type; |
685c9b24 | 458 | u32 nbd_cmd_flags = 0; |
9dd5d3ab | 459 | int sent = nsock->sent, skip = 0; |
9dc6c806 | 460 | |
c9f2b6ae AV |
461 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); |
462 | ||
aebf526b CH |
463 | switch (req_op(req)) { |
464 | case REQ_OP_DISCARD: | |
9dc6c806 | 465 | type = NBD_CMD_TRIM; |
aebf526b CH |
466 | break; |
467 | case REQ_OP_FLUSH: | |
9dc6c806 | 468 | type = NBD_CMD_FLUSH; |
aebf526b CH |
469 | break; |
470 | case REQ_OP_WRITE: | |
9dc6c806 | 471 | type = NBD_CMD_WRITE; |
aebf526b CH |
472 | break; |
473 | case REQ_OP_READ: | |
9dc6c806 | 474 | type = NBD_CMD_READ; |
aebf526b CH |
475 | break; |
476 | default: | |
477 | return -EIO; | |
478 | } | |
1da177e4 | 479 | |
09fc54cc | 480 | if (rq_data_dir(req) == WRITE && |
5ea8d108 | 481 | (config->flags & NBD_FLAG_READ_ONLY)) { |
09fc54cc CH |
482 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
483 | "Write on read-only\n"); | |
484 | return -EIO; | |
485 | } | |
486 | ||
685c9b24 SM |
487 | if (req->cmd_flags & REQ_FUA) |
488 | nbd_cmd_flags |= NBD_CMD_FLAG_FUA; | |
489 | ||
9dd5d3ab JB |
490 | /* We did a partial send previously, and we at least sent the whole |
491 | * request struct, so just go and send the rest of the pages in the | |
492 | * request. | |
493 | */ | |
494 | if (sent) { | |
495 | if (sent >= sizeof(request)) { | |
496 | skip = sent - sizeof(request); | |
497 | goto send_pages; | |
498 | } | |
499 | iov_iter_advance(&from, sent); | |
05ee6166 JB |
500 | } else { |
501 | cmd->cmd_cookie++; | |
9dd5d3ab | 502 | } |
f3733247 | 503 | cmd->index = index; |
799f9a38 | 504 | cmd->cookie = nsock->cookie; |
685c9b24 | 505 | request.type = htonl(type | nbd_cmd_flags); |
9561a7ad | 506 | if (type != NBD_CMD_FLUSH) { |
75f187ab AB |
507 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
508 | request.len = htonl(size); | |
509 | } | |
05ee6166 JB |
510 | handle = nbd_cmd_handle(cmd); |
511 | memcpy(request.handle, &handle, sizeof(handle)); | |
1da177e4 | 512 | |
d18509f5 | 513 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
fd8383fd | 514 | cmd, nbdcmd_to_ascii(type), |
d18509f5 | 515 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
c9f2b6ae | 516 | result = sock_xmit(nbd, index, 1, &from, |
9dd5d3ab | 517 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent); |
1da177e4 | 518 | if (result <= 0) { |
32e67a3a | 519 | if (was_interrupted(result)) { |
9dd5d3ab JB |
520 | /* If we havne't sent anything we can just return BUSY, |
521 | * however if we have sent something we need to make | |
522 | * sure we only allow this req to be sent until we are | |
523 | * completely done. | |
524 | */ | |
525 | if (sent) { | |
526 | nsock->pending = req; | |
527 | nsock->sent = sent; | |
528 | } | |
ced413c5 | 529 | set_bit(NBD_CMD_REQUEUED, &cmd->flags); |
fc17b653 | 530 | return BLK_STS_RESOURCE; |
9dd5d3ab | 531 | } |
a897b666 | 532 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 | 533 | "Send control failed (result %d)\n", result); |
f3733247 | 534 | return -EAGAIN; |
1da177e4 | 535 | } |
9dd5d3ab | 536 | send_pages: |
429a787b | 537 | if (type != NBD_CMD_WRITE) |
9dd5d3ab | 538 | goto out; |
429a787b | 539 | |
429a787b JA |
540 | bio = req->bio; |
541 | while (bio) { | |
542 | struct bio *next = bio->bi_next; | |
543 | struct bvec_iter iter; | |
7988613b | 544 | struct bio_vec bvec; |
429a787b JA |
545 | |
546 | bio_for_each_segment(bvec, bio, iter) { | |
547 | bool is_last = !next && bio_iter_last(bvec, iter); | |
d61b7f97 | 548 | int flags = is_last ? 0 : MSG_MORE; |
429a787b | 549 | |
d18509f5 | 550 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
fd8383fd | 551 | cmd, bvec.bv_len); |
c9f2b6ae AV |
552 | iov_iter_bvec(&from, ITER_BVEC | WRITE, |
553 | &bvec, 1, bvec.bv_len); | |
9dd5d3ab JB |
554 | if (skip) { |
555 | if (skip >= iov_iter_count(&from)) { | |
556 | skip -= iov_iter_count(&from); | |
557 | continue; | |
558 | } | |
559 | iov_iter_advance(&from, skip); | |
560 | skip = 0; | |
561 | } | |
562 | result = sock_xmit(nbd, index, 1, &from, flags, &sent); | |
6c92e699 | 563 | if (result <= 0) { |
32e67a3a | 564 | if (was_interrupted(result)) { |
9dd5d3ab JB |
565 | /* We've already sent the header, we |
566 | * have no choice but to set pending and | |
567 | * return BUSY. | |
568 | */ | |
569 | nsock->pending = req; | |
570 | nsock->sent = sent; | |
ced413c5 | 571 | set_bit(NBD_CMD_REQUEUED, &cmd->flags); |
fc17b653 | 572 | return BLK_STS_RESOURCE; |
9dd5d3ab | 573 | } |
f4507164 | 574 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
575 | "Send data failed (result %d)\n", |
576 | result); | |
f3733247 | 577 | return -EAGAIN; |
6c92e699 | 578 | } |
429a787b JA |
579 | /* |
580 | * The completion might already have come in, | |
581 | * so break for the last one instead of letting | |
582 | * the iterator do it. This prevents use-after-free | |
583 | * of the bio. | |
584 | */ | |
585 | if (is_last) | |
586 | break; | |
1da177e4 | 587 | } |
429a787b | 588 | bio = next; |
1da177e4 | 589 | } |
9dd5d3ab JB |
590 | out: |
591 | nsock->pending = NULL; | |
592 | nsock->sent = 0; | |
1da177e4 | 593 | return 0; |
1da177e4 LT |
594 | } |
595 | ||
1da177e4 | 596 | /* NULL returned = something went wrong, inform userspace */ |
9561a7ad | 597 | static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index) |
1da177e4 | 598 | { |
5ea8d108 | 599 | struct nbd_config *config = nbd->config; |
1da177e4 LT |
600 | int result; |
601 | struct nbd_reply reply; | |
fd8383fd JB |
602 | struct nbd_cmd *cmd; |
603 | struct request *req = NULL; | |
05ee6166 | 604 | u64 handle; |
fd8383fd | 605 | u16 hwq; |
9561a7ad | 606 | u32 tag; |
c9f2b6ae AV |
607 | struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)}; |
608 | struct iov_iter to; | |
05ee6166 | 609 | int ret = 0; |
1da177e4 LT |
610 | |
611 | reply.magic = 0; | |
c9f2b6ae | 612 | iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply)); |
9dd5d3ab | 613 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
1da177e4 | 614 | if (result <= 0) { |
5ea8d108 | 615 | if (!nbd_disconnected(config)) |
9561a7ad JB |
616 | dev_err(disk_to_dev(nbd->disk), |
617 | "Receive control failed (result %d)\n", result); | |
19391830 | 618 | return ERR_PTR(result); |
1da177e4 | 619 | } |
e4b57e08 MF |
620 | |
621 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 622 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 623 | (unsigned long)ntohl(reply.magic)); |
19391830 | 624 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
625 | } |
626 | ||
05ee6166 JB |
627 | memcpy(&handle, reply.handle, sizeof(handle)); |
628 | tag = nbd_handle_to_tag(handle); | |
fd8383fd JB |
629 | hwq = blk_mq_unique_tag_to_hwq(tag); |
630 | if (hwq < nbd->tag_set.nr_hw_queues) | |
631 | req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq], | |
632 | blk_mq_unique_tag_to_tag(tag)); | |
633 | if (!req || !blk_mq_request_started(req)) { | |
634 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n", | |
635 | tag, req); | |
636 | return ERR_PTR(-ENOENT); | |
1da177e4 | 637 | } |
fd8383fd | 638 | cmd = blk_mq_rq_to_pdu(req); |
05ee6166 JB |
639 | |
640 | mutex_lock(&cmd->lock); | |
641 | if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) { | |
642 | dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n", | |
643 | req, cmd->cmd_cookie, nbd_handle_to_cookie(handle)); | |
644 | ret = -ENOENT; | |
645 | goto out; | |
646 | } | |
647 | if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) { | |
648 | dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n", | |
649 | req); | |
650 | ret = -ENOENT; | |
651 | goto out; | |
652 | } | |
1da177e4 | 653 | if (ntohl(reply.error)) { |
f4507164 | 654 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 655 | ntohl(reply.error)); |
2a842aca | 656 | cmd->status = BLK_STS_IOERR; |
05ee6166 | 657 | goto out; |
1da177e4 LT |
658 | } |
659 | ||
fd8383fd | 660 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd); |
9dc6c806 | 661 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 662 | struct req_iterator iter; |
7988613b | 663 | struct bio_vec bvec; |
5705f702 N |
664 | |
665 | rq_for_each_segment(bvec, req, iter) { | |
c9f2b6ae AV |
666 | iov_iter_bvec(&to, ITER_BVEC | READ, |
667 | &bvec, 1, bvec.bv_len); | |
9dd5d3ab | 668 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
6c92e699 | 669 | if (result <= 0) { |
f4507164 | 670 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 671 | result); |
f3733247 JB |
672 | /* |
673 | * If we've disconnected or we only have 1 | |
674 | * connection then we need to make sure we | |
675 | * complete this request, otherwise error out | |
676 | * and let the timeout stuff handle resubmitting | |
677 | * this request onto another connection. | |
678 | */ | |
5ea8d108 JB |
679 | if (nbd_disconnected(config) || |
680 | config->num_connections <= 1) { | |
2a842aca | 681 | cmd->status = BLK_STS_IOERR; |
05ee6166 | 682 | goto out; |
f3733247 | 683 | } |
05ee6166 JB |
684 | ret = -EIO; |
685 | goto out; | |
6c92e699 | 686 | } |
d18509f5 | 687 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
fd8383fd | 688 | cmd, bvec.bv_len); |
1da177e4 LT |
689 | } |
690 | } | |
05ee6166 JB |
691 | out: |
692 | mutex_unlock(&cmd->lock); | |
693 | return ret ? ERR_PTR(ret) : cmd; | |
1da177e4 LT |
694 | } |
695 | ||
9561a7ad | 696 | static void recv_work(struct work_struct *work) |
1da177e4 | 697 | { |
9561a7ad JB |
698 | struct recv_thread_args *args = container_of(work, |
699 | struct recv_thread_args, | |
700 | work); | |
701 | struct nbd_device *nbd = args->nbd; | |
5ea8d108 | 702 | struct nbd_config *config = nbd->config; |
fd8383fd | 703 | struct nbd_cmd *cmd; |
1da177e4 | 704 | |
19391830 | 705 | while (1) { |
9561a7ad | 706 | cmd = nbd_read_stat(nbd, args->index); |
fd8383fd | 707 | if (IS_ERR(cmd)) { |
5ea8d108 | 708 | struct nbd_sock *nsock = config->socks[args->index]; |
f3733247 JB |
709 | |
710 | mutex_lock(&nsock->tx_lock); | |
799f9a38 | 711 | nbd_mark_nsock_dead(nbd, nsock, 1); |
f3733247 | 712 | mutex_unlock(&nsock->tx_lock); |
19391830 MP |
713 | break; |
714 | } | |
715 | ||
08e0029a | 716 | blk_mq_complete_request(blk_mq_rq_from_pdu(cmd)); |
19391830 | 717 | } |
5ea8d108 JB |
718 | atomic_dec(&config->recv_threads); |
719 | wake_up(&config->recv_wq); | |
720 | nbd_config_put(nbd); | |
721 | kfree(args); | |
1da177e4 LT |
722 | } |
723 | ||
fd8383fd | 724 | static void nbd_clear_req(struct request *req, void *data, bool reserved) |
1da177e4 | 725 | { |
fd8383fd | 726 | struct nbd_cmd *cmd; |
1da177e4 | 727 | |
fd8383fd JB |
728 | if (!blk_mq_request_started(req)) |
729 | return; | |
730 | cmd = blk_mq_rq_to_pdu(req); | |
2a842aca | 731 | cmd->status = BLK_STS_IOERR; |
08e0029a | 732 | blk_mq_complete_request(req); |
fd8383fd JB |
733 | } |
734 | ||
735 | static void nbd_clear_que(struct nbd_device *nbd) | |
736 | { | |
b52c2e92 | 737 | blk_mq_quiesce_queue(nbd->disk->queue); |
fd8383fd | 738 | blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL); |
b52c2e92 | 739 | blk_mq_unquiesce_queue(nbd->disk->queue); |
e78273c8 | 740 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
741 | } |
742 | ||
f3733247 JB |
743 | static int find_fallback(struct nbd_device *nbd, int index) |
744 | { | |
5ea8d108 | 745 | struct nbd_config *config = nbd->config; |
f3733247 | 746 | int new_index = -1; |
5ea8d108 | 747 | struct nbd_sock *nsock = config->socks[index]; |
f3733247 JB |
748 | int fallback = nsock->fallback_index; |
749 | ||
5ea8d108 | 750 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
f3733247 JB |
751 | return new_index; |
752 | ||
5ea8d108 | 753 | if (config->num_connections <= 1) { |
f3733247 JB |
754 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
755 | "Attempted send on invalid socket\n"); | |
756 | return new_index; | |
757 | } | |
758 | ||
5ea8d108 JB |
759 | if (fallback >= 0 && fallback < config->num_connections && |
760 | !config->socks[fallback]->dead) | |
f3733247 JB |
761 | return fallback; |
762 | ||
763 | if (nsock->fallback_index < 0 || | |
5ea8d108 JB |
764 | nsock->fallback_index >= config->num_connections || |
765 | config->socks[nsock->fallback_index]->dead) { | |
f3733247 | 766 | int i; |
5ea8d108 | 767 | for (i = 0; i < config->num_connections; i++) { |
f3733247 JB |
768 | if (i == index) |
769 | continue; | |
5ea8d108 | 770 | if (!config->socks[i]->dead) { |
f3733247 JB |
771 | new_index = i; |
772 | break; | |
773 | } | |
774 | } | |
775 | nsock->fallback_index = new_index; | |
776 | if (new_index < 0) { | |
777 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
778 | "Dead connection, failed to find a fallback\n"); | |
779 | return new_index; | |
780 | } | |
781 | } | |
782 | new_index = nsock->fallback_index; | |
783 | return new_index; | |
784 | } | |
7fdfd406 | 785 | |
560bc4b3 JB |
786 | static int wait_for_reconnect(struct nbd_device *nbd) |
787 | { | |
788 | struct nbd_config *config = nbd->config; | |
789 | if (!config->dead_conn_timeout) | |
790 | return 0; | |
791 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) | |
792 | return 0; | |
f6b7c54c JB |
793 | wait_event_timeout(config->conn_wait, |
794 | atomic_read(&config->live_connections), | |
795 | config->dead_conn_timeout); | |
560bc4b3 JB |
796 | return atomic_read(&config->live_connections); |
797 | } | |
798 | ||
9dd5d3ab | 799 | static int nbd_handle_cmd(struct nbd_cmd *cmd, int index) |
48cf6061 | 800 | { |
fd8383fd JB |
801 | struct request *req = blk_mq_rq_from_pdu(cmd); |
802 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 | 803 | struct nbd_config *config; |
9561a7ad | 804 | struct nbd_sock *nsock; |
9dd5d3ab | 805 | int ret; |
fd8383fd | 806 | |
5ea8d108 JB |
807 | if (!refcount_inc_not_zero(&nbd->config_refs)) { |
808 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
809 | "Socks array is empty\n"); | |
2031e243 | 810 | blk_mq_start_request(req); |
5ea8d108 JB |
811 | return -EINVAL; |
812 | } | |
813 | config = nbd->config; | |
814 | ||
815 | if (index >= config->num_connections) { | |
a897b666 JB |
816 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
817 | "Attempted send on invalid socket\n"); | |
5ea8d108 | 818 | nbd_config_put(nbd); |
2031e243 | 819 | blk_mq_start_request(req); |
9dd5d3ab | 820 | return -EINVAL; |
9561a7ad | 821 | } |
2a842aca | 822 | cmd->status = BLK_STS_OK; |
f3733247 | 823 | again: |
5ea8d108 | 824 | nsock = config->socks[index]; |
9561a7ad | 825 | mutex_lock(&nsock->tx_lock); |
f3733247 | 826 | if (nsock->dead) { |
560bc4b3 | 827 | int old_index = index; |
f3733247 | 828 | index = find_fallback(nbd, index); |
560bc4b3 | 829 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 830 | if (index < 0) { |
560bc4b3 JB |
831 | if (wait_for_reconnect(nbd)) { |
832 | index = old_index; | |
833 | goto again; | |
834 | } | |
835 | /* All the sockets should already be down at this point, | |
836 | * we just want to make sure that DISCONNECTED is set so | |
837 | * any requests that come in that were queue'ed waiting | |
838 | * for the reconnect timer don't trigger the timer again | |
839 | * and instead just error out. | |
840 | */ | |
841 | sock_shutdown(nbd); | |
842 | nbd_config_put(nbd); | |
2031e243 | 843 | blk_mq_start_request(req); |
560bc4b3 | 844 | return -EIO; |
5ea8d108 | 845 | } |
f3733247 | 846 | goto again; |
48cf6061 LV |
847 | } |
848 | ||
9dd5d3ab JB |
849 | /* Handle the case that we have a pending request that was partially |
850 | * transmitted that _has_ to be serviced first. We need to call requeue | |
851 | * here so that it gets put _after_ the request that is already on the | |
852 | * dispatch list. | |
853 | */ | |
2031e243 | 854 | blk_mq_start_request(req); |
9dd5d3ab | 855 | if (unlikely(nsock->pending && nsock->pending != req)) { |
ced413c5 | 856 | nbd_requeue_cmd(cmd); |
9dd5d3ab JB |
857 | ret = 0; |
858 | goto out; | |
48cf6061 | 859 | } |
f3733247 JB |
860 | /* |
861 | * Some failures are related to the link going down, so anything that | |
862 | * returns EAGAIN can be retried on a different socket. | |
863 | */ | |
9dd5d3ab | 864 | ret = nbd_send_cmd(nbd, cmd, index); |
f3733247 JB |
865 | if (ret == -EAGAIN) { |
866 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
2031e243 | 867 | "Request send failed, requeueing\n"); |
799f9a38 | 868 | nbd_mark_nsock_dead(nbd, nsock, 1); |
ced413c5 | 869 | nbd_requeue_cmd(cmd); |
2031e243 | 870 | ret = 0; |
f3733247 | 871 | } |
9dd5d3ab | 872 | out: |
9561a7ad | 873 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 874 | nbd_config_put(nbd); |
9dd5d3ab | 875 | return ret; |
48cf6061 LV |
876 | } |
877 | ||
fc17b653 | 878 | static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
fd8383fd | 879 | const struct blk_mq_queue_data *bd) |
1da177e4 | 880 | { |
fd8383fd | 881 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); |
9dd5d3ab | 882 | int ret; |
1da177e4 | 883 | |
9561a7ad JB |
884 | /* |
885 | * Since we look at the bio's to send the request over the network we | |
886 | * need to make sure the completion work doesn't mark this request done | |
887 | * before we are done doing our send. This keeps us from dereferencing | |
888 | * freed data if we have particularly fast completions (ie we get the | |
889 | * completion before we exit sock_xmit on the last bvec) or in the case | |
890 | * that the server is misbehaving (or there was an error) before we're | |
891 | * done sending everything over the wire. | |
892 | */ | |
05ee6166 | 893 | mutex_lock(&cmd->lock); |
ced413c5 | 894 | clear_bit(NBD_CMD_REQUEUED, &cmd->flags); |
9dd5d3ab JB |
895 | |
896 | /* We can be called directly from the user space process, which means we | |
897 | * could possibly have signals pending so our sendmsg will fail. In | |
898 | * this case we need to return that we are busy, otherwise error out as | |
899 | * appropriate. | |
900 | */ | |
901 | ret = nbd_handle_cmd(cmd, hctx->queue_num); | |
6e60a3bb JB |
902 | if (ret < 0) |
903 | ret = BLK_STS_IOERR; | |
904 | else if (!ret) | |
905 | ret = BLK_STS_OK; | |
05ee6166 | 906 | mutex_unlock(&cmd->lock); |
9561a7ad | 907 | |
6e60a3bb | 908 | return ret; |
1da177e4 LT |
909 | } |
910 | ||
e46c7287 JB |
911 | static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg, |
912 | bool netlink) | |
23272a67 | 913 | { |
5ea8d108 | 914 | struct nbd_config *config = nbd->config; |
9442b739 | 915 | struct socket *sock; |
9561a7ad JB |
916 | struct nbd_sock **socks; |
917 | struct nbd_sock *nsock; | |
9442b739 JB |
918 | int err; |
919 | ||
920 | sock = sockfd_lookup(arg, &err); | |
921 | if (!sock) | |
922 | return err; | |
23272a67 | 923 | |
e46c7287 JB |
924 | if (!netlink && !nbd->task_setup && |
925 | !test_bit(NBD_BOUND, &config->runtime_flags)) | |
9561a7ad | 926 | nbd->task_setup = current; |
e46c7287 JB |
927 | |
928 | if (!netlink && | |
929 | (nbd->task_setup != current || | |
930 | test_bit(NBD_BOUND, &config->runtime_flags))) { | |
9561a7ad JB |
931 | dev_err(disk_to_dev(nbd->disk), |
932 | "Device being setup by another task"); | |
9b1355d5 | 933 | sockfd_put(sock); |
e46c7287 | 934 | return -EBUSY; |
23272a67 MP |
935 | } |
936 | ||
5ea8d108 | 937 | socks = krealloc(config->socks, (config->num_connections + 1) * |
9561a7ad | 938 | sizeof(struct nbd_sock *), GFP_KERNEL); |
9b1355d5 JB |
939 | if (!socks) { |
940 | sockfd_put(sock); | |
9561a7ad | 941 | return -ENOMEM; |
9b1355d5 | 942 | } |
9561a7ad | 943 | nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL); |
9b1355d5 JB |
944 | if (!nsock) { |
945 | sockfd_put(sock); | |
9561a7ad | 946 | return -ENOMEM; |
9b1355d5 | 947 | } |
9561a7ad | 948 | |
5ea8d108 | 949 | config->socks = socks; |
23272a67 | 950 | |
f3733247 JB |
951 | nsock->fallback_index = -1; |
952 | nsock->dead = false; | |
9561a7ad JB |
953 | mutex_init(&nsock->tx_lock); |
954 | nsock->sock = sock; | |
9dd5d3ab JB |
955 | nsock->pending = NULL; |
956 | nsock->sent = 0; | |
799f9a38 | 957 | nsock->cookie = 0; |
5ea8d108 | 958 | socks[config->num_connections++] = nsock; |
560bc4b3 | 959 | atomic_inc(&config->live_connections); |
23272a67 | 960 | |
9561a7ad | 961 | return 0; |
23272a67 MP |
962 | } |
963 | ||
b7aa3d39 JB |
964 | static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg) |
965 | { | |
966 | struct nbd_config *config = nbd->config; | |
967 | struct socket *sock, *old; | |
968 | struct recv_thread_args *args; | |
969 | int i; | |
970 | int err; | |
971 | ||
972 | sock = sockfd_lookup(arg, &err); | |
973 | if (!sock) | |
974 | return err; | |
975 | ||
976 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
977 | if (!args) { | |
978 | sockfd_put(sock); | |
979 | return -ENOMEM; | |
980 | } | |
981 | ||
982 | for (i = 0; i < config->num_connections; i++) { | |
983 | struct nbd_sock *nsock = config->socks[i]; | |
984 | ||
985 | if (!nsock->dead) | |
986 | continue; | |
987 | ||
988 | mutex_lock(&nsock->tx_lock); | |
989 | if (!nsock->dead) { | |
990 | mutex_unlock(&nsock->tx_lock); | |
991 | continue; | |
992 | } | |
993 | sk_set_memalloc(sock->sk); | |
a7ee8cf1 JB |
994 | if (nbd->tag_set.timeout) |
995 | sock->sk->sk_sndtimeo = nbd->tag_set.timeout; | |
b7aa3d39 JB |
996 | atomic_inc(&config->recv_threads); |
997 | refcount_inc(&nbd->config_refs); | |
998 | old = nsock->sock; | |
999 | nsock->fallback_index = -1; | |
1000 | nsock->sock = sock; | |
1001 | nsock->dead = false; | |
1002 | INIT_WORK(&args->work, recv_work); | |
1003 | args->index = i; | |
1004 | args->nbd = nbd; | |
799f9a38 | 1005 | nsock->cookie++; |
b7aa3d39 JB |
1006 | mutex_unlock(&nsock->tx_lock); |
1007 | sockfd_put(old); | |
1008 | ||
7a362ea9 JB |
1009 | clear_bit(NBD_DISCONNECTED, &config->runtime_flags); |
1010 | ||
b7aa3d39 JB |
1011 | /* We take the tx_mutex in an error path in the recv_work, so we |
1012 | * need to queue_work outside of the tx_mutex. | |
1013 | */ | |
1014 | queue_work(recv_workqueue, &args->work); | |
560bc4b3 JB |
1015 | |
1016 | atomic_inc(&config->live_connections); | |
1017 | wake_up(&config->conn_wait); | |
b7aa3d39 JB |
1018 | return 0; |
1019 | } | |
1020 | sockfd_put(sock); | |
1021 | kfree(args); | |
1022 | return -ENOSPC; | |
1023 | } | |
1024 | ||
0e4f0f6f MP |
1025 | static void nbd_bdev_reset(struct block_device *bdev) |
1026 | { | |
abbbdf12 RMB |
1027 | if (bdev->bd_openers > 1) |
1028 | return; | |
29eaadc0 | 1029 | bd_set_size(bdev, 0); |
0e4f0f6f MP |
1030 | if (max_part > 0) { |
1031 | blkdev_reread_part(bdev); | |
1032 | bdev->bd_invalidated = 1; | |
1033 | } | |
1034 | } | |
1035 | ||
29eaadc0 | 1036 | static void nbd_parse_flags(struct nbd_device *nbd) |
d02cf531 | 1037 | { |
5ea8d108 JB |
1038 | struct nbd_config *config = nbd->config; |
1039 | if (config->flags & NBD_FLAG_READ_ONLY) | |
29eaadc0 JB |
1040 | set_disk_ro(nbd->disk, true); |
1041 | else | |
1042 | set_disk_ro(nbd->disk, false); | |
5ea8d108 | 1043 | if (config->flags & NBD_FLAG_SEND_TRIM) |
d02cf531 | 1044 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
685c9b24 SM |
1045 | if (config->flags & NBD_FLAG_SEND_FLUSH) { |
1046 | if (config->flags & NBD_FLAG_SEND_FUA) | |
1047 | blk_queue_write_cache(nbd->disk->queue, true, true); | |
1048 | else | |
1049 | blk_queue_write_cache(nbd->disk->queue, true, false); | |
1050 | } | |
d02cf531 | 1051 | else |
aafb1eec | 1052 | blk_queue_write_cache(nbd->disk->queue, false, false); |
d02cf531 MP |
1053 | } |
1054 | ||
9561a7ad JB |
1055 | static void send_disconnects(struct nbd_device *nbd) |
1056 | { | |
5ea8d108 | 1057 | struct nbd_config *config = nbd->config; |
c9f2b6ae AV |
1058 | struct nbd_request request = { |
1059 | .magic = htonl(NBD_REQUEST_MAGIC), | |
1060 | .type = htonl(NBD_CMD_DISC), | |
1061 | }; | |
1062 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
1063 | struct iov_iter from; | |
9561a7ad JB |
1064 | int i, ret; |
1065 | ||
5ea8d108 | 1066 | for (i = 0; i < config->num_connections; i++) { |
b4b2aecc JB |
1067 | struct nbd_sock *nsock = config->socks[i]; |
1068 | ||
c9f2b6ae | 1069 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); |
b4b2aecc | 1070 | mutex_lock(&nsock->tx_lock); |
9dd5d3ab | 1071 | ret = sock_xmit(nbd, i, 1, &from, 0, NULL); |
9561a7ad JB |
1072 | if (ret <= 0) |
1073 | dev_err(disk_to_dev(nbd->disk), | |
1074 | "Send disconnect failed %d\n", ret); | |
b4b2aecc | 1075 | mutex_unlock(&nsock->tx_lock); |
9561a7ad JB |
1076 | } |
1077 | } | |
1078 | ||
29eaadc0 | 1079 | static int nbd_disconnect(struct nbd_device *nbd) |
9442b739 | 1080 | { |
5ea8d108 | 1081 | struct nbd_config *config = nbd->config; |
30d53d9c | 1082 | |
5ea8d108 | 1083 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
2e13456f JB |
1084 | set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); |
1085 | send_disconnects(nbd); | |
9442b739 JB |
1086 | return 0; |
1087 | } | |
1088 | ||
29eaadc0 | 1089 | static void nbd_clear_sock(struct nbd_device *nbd) |
1a2ad211 | 1090 | { |
9442b739 JB |
1091 | sock_shutdown(nbd); |
1092 | nbd_clear_que(nbd); | |
5ea8d108 | 1093 | nbd->task_setup = NULL; |
5ea8d108 JB |
1094 | } |
1095 | ||
1096 | static void nbd_config_put(struct nbd_device *nbd) | |
1097 | { | |
1098 | if (refcount_dec_and_mutex_lock(&nbd->config_refs, | |
1099 | &nbd->config_lock)) { | |
5ea8d108 | 1100 | struct nbd_config *config = nbd->config; |
5ea8d108 | 1101 | nbd_dev_dbg_close(nbd); |
29eaadc0 | 1102 | nbd_size_clear(nbd); |
5ea8d108 JB |
1103 | if (test_and_clear_bit(NBD_HAS_PID_FILE, |
1104 | &config->runtime_flags)) | |
1105 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); | |
1106 | nbd->task_recv = NULL; | |
29eaadc0 | 1107 | nbd_clear_sock(nbd); |
5ea8d108 JB |
1108 | if (config->num_connections) { |
1109 | int i; | |
1110 | for (i = 0; i < config->num_connections; i++) { | |
1111 | sockfd_put(config->socks[i]->sock); | |
1112 | kfree(config->socks[i]); | |
1113 | } | |
1114 | kfree(config->socks); | |
1115 | } | |
fa976532 | 1116 | kfree(nbd->config); |
af622b86 ID |
1117 | nbd->config = NULL; |
1118 | ||
1119 | nbd->tag_set.timeout = 0; | |
1120 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); | |
a2c97909 | 1121 | |
5ea8d108 | 1122 | mutex_unlock(&nbd->config_lock); |
c6a4759e | 1123 | nbd_put(nbd); |
5ea8d108 JB |
1124 | module_put(THIS_MODULE); |
1125 | } | |
9442b739 JB |
1126 | } |
1127 | ||
e46c7287 | 1128 | static int nbd_start_device(struct nbd_device *nbd) |
9442b739 | 1129 | { |
5ea8d108 JB |
1130 | struct nbd_config *config = nbd->config; |
1131 | int num_connections = config->num_connections; | |
9442b739 | 1132 | int error = 0, i; |
1a2ad211 | 1133 | |
9442b739 JB |
1134 | if (nbd->task_recv) |
1135 | return -EBUSY; | |
5ea8d108 | 1136 | if (!config->socks) |
9442b739 JB |
1137 | return -EINVAL; |
1138 | if (num_connections > 1 && | |
5ea8d108 | 1139 | !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) { |
9442b739 | 1140 | dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n"); |
5ea8d108 | 1141 | return -EINVAL; |
9442b739 | 1142 | } |
23272a67 | 1143 | |
5ea8d108 | 1144 | blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections); |
9442b739 | 1145 | nbd->task_recv = current; |
23272a67 | 1146 | |
29eaadc0 | 1147 | nbd_parse_flags(nbd); |
23272a67 | 1148 | |
9442b739 JB |
1149 | error = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
1150 | if (error) { | |
1151 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); | |
5ea8d108 | 1152 | return error; |
1a2ad211 | 1153 | } |
29eaadc0 | 1154 | set_bit(NBD_HAS_PID_FILE, &config->runtime_flags); |
37091fdd | 1155 | |
9442b739 JB |
1156 | nbd_dev_dbg_init(nbd); |
1157 | for (i = 0; i < num_connections; i++) { | |
5ea8d108 JB |
1158 | struct recv_thread_args *args; |
1159 | ||
1160 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
1161 | if (!args) { | |
1162 | sock_shutdown(nbd); | |
1163 | return -ENOMEM; | |
1164 | } | |
1165 | sk_set_memalloc(config->socks[i]->sock->sk); | |
a7ee8cf1 JB |
1166 | if (nbd->tag_set.timeout) |
1167 | config->socks[i]->sock->sk->sk_sndtimeo = | |
1168 | nbd->tag_set.timeout; | |
5ea8d108 JB |
1169 | atomic_inc(&config->recv_threads); |
1170 | refcount_inc(&nbd->config_refs); | |
1171 | INIT_WORK(&args->work, recv_work); | |
1172 | args->nbd = nbd; | |
1173 | args->index = i; | |
1174 | queue_work(recv_workqueue, &args->work); | |
37091fdd | 1175 | } |
639812a1 | 1176 | nbd_size_update(nbd); |
e46c7287 JB |
1177 | return error; |
1178 | } | |
1179 | ||
1180 | static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev) | |
1181 | { | |
1182 | struct nbd_config *config = nbd->config; | |
1183 | int ret; | |
1184 | ||
1185 | ret = nbd_start_device(nbd); | |
1186 | if (ret) | |
1187 | return ret; | |
1188 | ||
e46c7287 JB |
1189 | if (max_part) |
1190 | bdev->bd_invalidated = 1; | |
1191 | mutex_unlock(&nbd->config_lock); | |
1192 | ret = wait_event_interruptible(config->recv_wq, | |
5ea8d108 | 1193 | atomic_read(&config->recv_threads) == 0); |
e46c7287 | 1194 | if (ret) |
5ea8d108 | 1195 | sock_shutdown(nbd); |
9442b739 | 1196 | mutex_lock(&nbd->config_lock); |
e46c7287 | 1197 | bd_set_size(bdev, 0); |
9442b739 | 1198 | /* user requested, ignore socket errors */ |
5ea8d108 | 1199 | if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags)) |
e46c7287 | 1200 | ret = 0; |
5ea8d108 | 1201 | if (test_bit(NBD_TIMEDOUT, &config->runtime_flags)) |
e46c7287 JB |
1202 | ret = -ETIMEDOUT; |
1203 | return ret; | |
9442b739 JB |
1204 | } |
1205 | ||
29eaadc0 JB |
1206 | static void nbd_clear_sock_ioctl(struct nbd_device *nbd, |
1207 | struct block_device *bdev) | |
1208 | { | |
2516ab15 | 1209 | sock_shutdown(nbd); |
29eaadc0 JB |
1210 | kill_bdev(bdev); |
1211 | nbd_bdev_reset(bdev); | |
e46c7287 JB |
1212 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, |
1213 | &nbd->config->runtime_flags)) | |
1214 | nbd_config_put(nbd); | |
29eaadc0 JB |
1215 | } |
1216 | ||
9442b739 JB |
1217 | /* Must be called with config_lock held */ |
1218 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, | |
1219 | unsigned int cmd, unsigned long arg) | |
1220 | { | |
5ea8d108 JB |
1221 | struct nbd_config *config = nbd->config; |
1222 | ||
9442b739 JB |
1223 | switch (cmd) { |
1224 | case NBD_DISCONNECT: | |
29eaadc0 | 1225 | return nbd_disconnect(nbd); |
9442b739 | 1226 | case NBD_CLEAR_SOCK: |
29eaadc0 JB |
1227 | nbd_clear_sock_ioctl(nbd, bdev); |
1228 | return 0; | |
9442b739 | 1229 | case NBD_SET_SOCK: |
e46c7287 | 1230 | return nbd_add_socket(nbd, arg, false); |
9442b739 | 1231 | case NBD_SET_BLKSIZE: |
23ecbbad JA |
1232 | if (!arg || !is_power_of_2(arg) || arg < 512 || |
1233 | arg > PAGE_SIZE) | |
1234 | return -EINVAL; | |
29eaadc0 | 1235 | nbd_size_set(nbd, arg, |
5ea8d108 | 1236 | div_s64(config->bytesize, arg)); |
e544541b | 1237 | return 0; |
1da177e4 | 1238 | case NBD_SET_SIZE: |
29eaadc0 | 1239 | nbd_size_set(nbd, config->blksize, |
5ea8d108 | 1240 | div_s64(arg, config->blksize)); |
e544541b | 1241 | return 0; |
37091fdd | 1242 | case NBD_SET_SIZE_BLOCKS: |
29eaadc0 | 1243 | nbd_size_set(nbd, config->blksize, arg); |
e544541b | 1244 | return 0; |
7fdfd406 | 1245 | case NBD_SET_TIMEOUT: |
f8586855 JB |
1246 | if (arg) { |
1247 | nbd->tag_set.timeout = arg * HZ; | |
1248 | blk_queue_rq_timeout(nbd->disk->queue, arg * HZ); | |
1249 | } | |
7fdfd406 | 1250 | return 0; |
1a2ad211 | 1251 | |
2f012508 | 1252 | case NBD_SET_FLAGS: |
5ea8d108 | 1253 | config->flags = arg; |
2f012508 | 1254 | return 0; |
9442b739 | 1255 | case NBD_DO_IT: |
e46c7287 | 1256 | return nbd_start_device_ioctl(nbd, bdev); |
1da177e4 | 1257 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
1258 | /* |
1259 | * This is for compatibility only. The queue is always cleared | |
1260 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
1261 | */ | |
1da177e4 LT |
1262 | return 0; |
1263 | case NBD_PRINT_DEBUG: | |
fd8383fd JB |
1264 | /* |
1265 | * For compatibility only, we no longer keep a list of | |
1266 | * outstanding requests. | |
1267 | */ | |
1da177e4 LT |
1268 | return 0; |
1269 | } | |
1a2ad211 PM |
1270 | return -ENOTTY; |
1271 | } | |
1272 | ||
1273 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
1274 | unsigned int cmd, unsigned long arg) | |
1275 | { | |
f4507164 | 1276 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
e46c7287 JB |
1277 | struct nbd_config *config = nbd->config; |
1278 | int error = -EINVAL; | |
1a2ad211 PM |
1279 | |
1280 | if (!capable(CAP_SYS_ADMIN)) | |
1281 | return -EPERM; | |
1282 | ||
1dae69be JB |
1283 | /* The block layer will pass back some non-nbd ioctls in case we have |
1284 | * special handling for them, but we don't so just return an error. | |
1285 | */ | |
1286 | if (_IOC_TYPE(cmd) != 0xab) | |
1287 | return -EINVAL; | |
1288 | ||
9561a7ad | 1289 | mutex_lock(&nbd->config_lock); |
e46c7287 JB |
1290 | |
1291 | /* Don't allow ioctl operations on a nbd device that was created with | |
1292 | * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine. | |
1293 | */ | |
1294 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1295 | (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK)) | |
1296 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
1297 | else | |
1298 | dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n"); | |
9561a7ad | 1299 | mutex_unlock(&nbd->config_lock); |
1a2ad211 | 1300 | return error; |
1da177e4 LT |
1301 | } |
1302 | ||
5ea8d108 JB |
1303 | static struct nbd_config *nbd_alloc_config(void) |
1304 | { | |
1305 | struct nbd_config *config; | |
1306 | ||
1307 | config = kzalloc(sizeof(struct nbd_config), GFP_NOFS); | |
1308 | if (!config) | |
1309 | return NULL; | |
1310 | atomic_set(&config->recv_threads, 0); | |
1311 | init_waitqueue_head(&config->recv_wq); | |
560bc4b3 | 1312 | init_waitqueue_head(&config->conn_wait); |
5ea8d108 | 1313 | config->blksize = 1024; |
560bc4b3 | 1314 | atomic_set(&config->live_connections, 0); |
5ea8d108 JB |
1315 | try_module_get(THIS_MODULE); |
1316 | return config; | |
1317 | } | |
1318 | ||
1319 | static int nbd_open(struct block_device *bdev, fmode_t mode) | |
1320 | { | |
1321 | struct nbd_device *nbd; | |
1322 | int ret = 0; | |
1323 | ||
1324 | mutex_lock(&nbd_index_mutex); | |
1325 | nbd = bdev->bd_disk->private_data; | |
1326 | if (!nbd) { | |
1327 | ret = -ENXIO; | |
1328 | goto out; | |
1329 | } | |
c6a4759e JB |
1330 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1331 | ret = -ENXIO; | |
1332 | goto out; | |
1333 | } | |
5ea8d108 JB |
1334 | if (!refcount_inc_not_zero(&nbd->config_refs)) { |
1335 | struct nbd_config *config; | |
1336 | ||
1337 | mutex_lock(&nbd->config_lock); | |
1338 | if (refcount_inc_not_zero(&nbd->config_refs)) { | |
1339 | mutex_unlock(&nbd->config_lock); | |
1340 | goto out; | |
1341 | } | |
1342 | config = nbd->config = nbd_alloc_config(); | |
1343 | if (!config) { | |
1344 | ret = -ENOMEM; | |
1345 | mutex_unlock(&nbd->config_lock); | |
1346 | goto out; | |
1347 | } | |
1348 | refcount_set(&nbd->config_refs, 1); | |
c6a4759e | 1349 | refcount_inc(&nbd->refs); |
5ea8d108 JB |
1350 | mutex_unlock(&nbd->config_lock); |
1351 | } | |
1352 | out: | |
1353 | mutex_unlock(&nbd_index_mutex); | |
1354 | return ret; | |
1355 | } | |
1356 | ||
1357 | static void nbd_release(struct gendisk *disk, fmode_t mode) | |
1358 | { | |
1359 | struct nbd_device *nbd = disk->private_data; | |
9b0b6258 DRK |
1360 | struct block_device *bdev = bdget_disk(disk, 0); |
1361 | ||
1362 | if (test_bit(NBD_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) && | |
1363 | bdev->bd_openers == 0) | |
1364 | nbd_disconnect_and_put(nbd); | |
1365 | ||
5ea8d108 | 1366 | nbd_config_put(nbd); |
c6a4759e | 1367 | nbd_put(nbd); |
5ea8d108 JB |
1368 | } |
1369 | ||
83d5cde4 | 1370 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
1371 | { |
1372 | .owner = THIS_MODULE, | |
5ea8d108 JB |
1373 | .open = nbd_open, |
1374 | .release = nbd_release, | |
8a6cfeb6 | 1375 | .ioctl = nbd_ioctl, |
263a3df1 | 1376 | .compat_ioctl = nbd_ioctl, |
1da177e4 LT |
1377 | }; |
1378 | ||
30d53d9c MP |
1379 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
1380 | ||
1381 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
1382 | { | |
1383 | struct nbd_device *nbd = s->private; | |
1384 | ||
1385 | if (nbd->task_recv) | |
1386 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
30d53d9c MP |
1387 | |
1388 | return 0; | |
1389 | } | |
1390 | ||
1391 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
1392 | { | |
1393 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
1394 | } | |
1395 | ||
1396 | static const struct file_operations nbd_dbg_tasks_ops = { | |
1397 | .open = nbd_dbg_tasks_open, | |
1398 | .read = seq_read, | |
1399 | .llseek = seq_lseek, | |
1400 | .release = single_release, | |
1401 | }; | |
1402 | ||
1403 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
1404 | { | |
1405 | struct nbd_device *nbd = s->private; | |
5ea8d108 | 1406 | u32 flags = nbd->config->flags; |
30d53d9c MP |
1407 | |
1408 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
1409 | ||
1410 | seq_puts(s, "Known flags:\n"); | |
1411 | ||
1412 | if (flags & NBD_FLAG_HAS_FLAGS) | |
1413 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
1414 | if (flags & NBD_FLAG_READ_ONLY) | |
1415 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
1416 | if (flags & NBD_FLAG_SEND_FLUSH) | |
1417 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
685c9b24 SM |
1418 | if (flags & NBD_FLAG_SEND_FUA) |
1419 | seq_puts(s, "NBD_FLAG_SEND_FUA\n"); | |
30d53d9c MP |
1420 | if (flags & NBD_FLAG_SEND_TRIM) |
1421 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
1422 | ||
1423 | return 0; | |
1424 | } | |
1425 | ||
1426 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
1427 | { | |
1428 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
1429 | } | |
1430 | ||
1431 | static const struct file_operations nbd_dbg_flags_ops = { | |
1432 | .open = nbd_dbg_flags_open, | |
1433 | .read = seq_read, | |
1434 | .llseek = seq_lseek, | |
1435 | .release = single_release, | |
1436 | }; | |
1437 | ||
1438 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1439 | { | |
1440 | struct dentry *dir; | |
5ea8d108 | 1441 | struct nbd_config *config = nbd->config; |
27ea43fe MP |
1442 | |
1443 | if (!nbd_dbg_dir) | |
1444 | return -EIO; | |
30d53d9c MP |
1445 | |
1446 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
27ea43fe MP |
1447 | if (!dir) { |
1448 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n", | |
1449 | nbd_name(nbd)); | |
1450 | return -EIO; | |
30d53d9c | 1451 | } |
5ea8d108 | 1452 | config->dbg_dir = dir; |
30d53d9c | 1453 | |
27ea43fe | 1454 | debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); |
5ea8d108 | 1455 | debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize); |
0eadf37a | 1456 | debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout); |
5ea8d108 | 1457 | debugfs_create_u64("blocksize", 0444, dir, &config->blksize); |
d366a0ff | 1458 | debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops); |
30d53d9c MP |
1459 | |
1460 | return 0; | |
1461 | } | |
1462 | ||
1463 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1464 | { | |
5ea8d108 | 1465 | debugfs_remove_recursive(nbd->config->dbg_dir); |
30d53d9c MP |
1466 | } |
1467 | ||
1468 | static int nbd_dbg_init(void) | |
1469 | { | |
1470 | struct dentry *dbg_dir; | |
1471 | ||
1472 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
27ea43fe MP |
1473 | if (!dbg_dir) |
1474 | return -EIO; | |
30d53d9c MP |
1475 | |
1476 | nbd_dbg_dir = dbg_dir; | |
1477 | ||
1478 | return 0; | |
1479 | } | |
1480 | ||
1481 | static void nbd_dbg_close(void) | |
1482 | { | |
1483 | debugfs_remove_recursive(nbd_dbg_dir); | |
1484 | } | |
1485 | ||
1486 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
1487 | ||
1488 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1489 | { | |
1490 | return 0; | |
1491 | } | |
1492 | ||
1493 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1494 | { | |
1495 | } | |
1496 | ||
1497 | static int nbd_dbg_init(void) | |
1498 | { | |
1499 | return 0; | |
1500 | } | |
1501 | ||
1502 | static void nbd_dbg_close(void) | |
1503 | { | |
1504 | } | |
1505 | ||
1506 | #endif | |
1507 | ||
d6296d39 CH |
1508 | static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1509 | unsigned int hctx_idx, unsigned int numa_node) | |
fd8383fd JB |
1510 | { |
1511 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
d6296d39 | 1512 | cmd->nbd = set->driver_data; |
ced413c5 | 1513 | cmd->flags = 0; |
05ee6166 | 1514 | mutex_init(&cmd->lock); |
fd8383fd JB |
1515 | return 0; |
1516 | } | |
1517 | ||
f363b089 | 1518 | static const struct blk_mq_ops nbd_mq_ops = { |
fd8383fd | 1519 | .queue_rq = nbd_queue_rq, |
1e388ae0 | 1520 | .complete = nbd_complete_rq, |
fd8383fd | 1521 | .init_request = nbd_init_request, |
0eadf37a | 1522 | .timeout = nbd_xmit_timeout, |
fd8383fd JB |
1523 | }; |
1524 | ||
b0d9111a JB |
1525 | static int nbd_dev_add(int index) |
1526 | { | |
1527 | struct nbd_device *nbd; | |
1528 | struct gendisk *disk; | |
1529 | struct request_queue *q; | |
1530 | int err = -ENOMEM; | |
1531 | ||
1532 | nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL); | |
1533 | if (!nbd) | |
1534 | goto out; | |
1535 | ||
1536 | disk = alloc_disk(1 << part_shift); | |
1537 | if (!disk) | |
1538 | goto out_free_nbd; | |
1539 | ||
1540 | if (index >= 0) { | |
1541 | err = idr_alloc(&nbd_index_idr, nbd, index, index + 1, | |
1542 | GFP_KERNEL); | |
1543 | if (err == -ENOSPC) | |
1544 | err = -EEXIST; | |
1545 | } else { | |
1546 | err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL); | |
1547 | if (err >= 0) | |
1548 | index = err; | |
1549 | } | |
1550 | if (err < 0) | |
1551 | goto out_free_disk; | |
1552 | ||
e46c7287 | 1553 | nbd->index = index; |
b0d9111a JB |
1554 | nbd->disk = disk; |
1555 | nbd->tag_set.ops = &nbd_mq_ops; | |
1556 | nbd->tag_set.nr_hw_queues = 1; | |
1557 | nbd->tag_set.queue_depth = 128; | |
1558 | nbd->tag_set.numa_node = NUMA_NO_NODE; | |
1559 | nbd->tag_set.cmd_size = sizeof(struct nbd_cmd); | |
1560 | nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | | |
1561 | BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING; | |
1562 | nbd->tag_set.driver_data = nbd; | |
1563 | ||
1564 | err = blk_mq_alloc_tag_set(&nbd->tag_set); | |
1565 | if (err) | |
1566 | goto out_free_idr; | |
1567 | ||
1568 | q = blk_mq_init_queue(&nbd->tag_set); | |
1569 | if (IS_ERR(q)) { | |
1570 | err = PTR_ERR(q); | |
1571 | goto out_free_tags; | |
1572 | } | |
1573 | disk->queue = q; | |
1574 | ||
1575 | /* | |
1576 | * Tell the block layer that we are not a rotational device | |
1577 | */ | |
1578 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
1579 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); | |
1580 | disk->queue->limits.discard_granularity = 512; | |
1581 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); | |
ebb16d0d | 1582 | blk_queue_max_segment_size(disk->queue, UINT_MAX); |
1cc1f17a | 1583 | blk_queue_max_segments(disk->queue, USHRT_MAX); |
b0d9111a JB |
1584 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1585 | disk->queue->limits.max_sectors = 256; | |
1586 | ||
b0d9111a | 1587 | mutex_init(&nbd->config_lock); |
5ea8d108 | 1588 | refcount_set(&nbd->config_refs, 0); |
c6a4759e JB |
1589 | refcount_set(&nbd->refs, 1); |
1590 | INIT_LIST_HEAD(&nbd->list); | |
b0d9111a JB |
1591 | disk->major = NBD_MAJOR; |
1592 | disk->first_minor = index << part_shift; | |
1593 | disk->fops = &nbd_fops; | |
1594 | disk->private_data = nbd; | |
1595 | sprintf(disk->disk_name, "nbd%d", index); | |
b0d9111a | 1596 | add_disk(disk); |
47d902b9 | 1597 | nbd_total_devices++; |
b0d9111a JB |
1598 | return index; |
1599 | ||
1600 | out_free_tags: | |
1601 | blk_mq_free_tag_set(&nbd->tag_set); | |
1602 | out_free_idr: | |
1603 | idr_remove(&nbd_index_idr, index); | |
1604 | out_free_disk: | |
1605 | put_disk(disk); | |
1606 | out_free_nbd: | |
1607 | kfree(nbd); | |
1608 | out: | |
1609 | return err; | |
1610 | } | |
1611 | ||
e46c7287 JB |
1612 | static int find_free_cb(int id, void *ptr, void *data) |
1613 | { | |
1614 | struct nbd_device *nbd = ptr; | |
1615 | struct nbd_device **found = data; | |
1616 | ||
1617 | if (!refcount_read(&nbd->config_refs)) { | |
1618 | *found = nbd; | |
1619 | return 1; | |
1620 | } | |
1621 | return 0; | |
1622 | } | |
1623 | ||
1624 | /* Netlink interface. */ | |
1625 | static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = { | |
1626 | [NBD_ATTR_INDEX] = { .type = NLA_U32 }, | |
1627 | [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 }, | |
1628 | [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 }, | |
1629 | [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 }, | |
1630 | [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 }, | |
1631 | [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 }, | |
1632 | [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED}, | |
560bc4b3 | 1633 | [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 }, |
47d902b9 | 1634 | [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED}, |
e46c7287 JB |
1635 | }; |
1636 | ||
1637 | static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = { | |
1638 | [NBD_SOCK_FD] = { .type = NLA_U32 }, | |
1639 | }; | |
1640 | ||
47d902b9 JB |
1641 | /* We don't use this right now since we don't parse the incoming list, but we |
1642 | * still want it here so userspace knows what to expect. | |
1643 | */ | |
1644 | static struct nla_policy __attribute__((unused)) | |
1645 | nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = { | |
1646 | [NBD_DEVICE_INDEX] = { .type = NLA_U32 }, | |
1647 | [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 }, | |
1648 | }; | |
1649 | ||
e46c7287 JB |
1650 | static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info) |
1651 | { | |
1652 | struct nbd_device *nbd = NULL; | |
1653 | struct nbd_config *config; | |
1654 | int index = -1; | |
1655 | int ret; | |
a2c97909 | 1656 | bool put_dev = false; |
e46c7287 JB |
1657 | |
1658 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1659 | return -EPERM; | |
1660 | ||
1661 | if (info->attrs[NBD_ATTR_INDEX]) | |
1662 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1663 | if (!info->attrs[NBD_ATTR_SOCKETS]) { | |
1664 | printk(KERN_ERR "nbd: must specify at least one socket\n"); | |
1665 | return -EINVAL; | |
1666 | } | |
1667 | if (!info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1668 | printk(KERN_ERR "nbd: must specify a size in bytes for the device\n"); | |
1669 | return -EINVAL; | |
1670 | } | |
1671 | again: | |
1672 | mutex_lock(&nbd_index_mutex); | |
1673 | if (index == -1) { | |
1674 | ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd); | |
1675 | if (ret == 0) { | |
1676 | int new_index; | |
1677 | new_index = nbd_dev_add(-1); | |
1678 | if (new_index < 0) { | |
1679 | mutex_unlock(&nbd_index_mutex); | |
1680 | printk(KERN_ERR "nbd: failed to add new device\n"); | |
615bf75c | 1681 | return new_index; |
e46c7287 JB |
1682 | } |
1683 | nbd = idr_find(&nbd_index_idr, new_index); | |
1684 | } | |
1685 | } else { | |
1686 | nbd = idr_find(&nbd_index_idr, index); | |
e6a76272 JB |
1687 | if (!nbd) { |
1688 | ret = nbd_dev_add(index); | |
1689 | if (ret < 0) { | |
1690 | mutex_unlock(&nbd_index_mutex); | |
1691 | printk(KERN_ERR "nbd: failed to add new device\n"); | |
1692 | return ret; | |
1693 | } | |
1694 | nbd = idr_find(&nbd_index_idr, index); | |
1695 | } | |
e46c7287 | 1696 | } |
e46c7287 JB |
1697 | if (!nbd) { |
1698 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", | |
1699 | index); | |
c6a4759e JB |
1700 | mutex_unlock(&nbd_index_mutex); |
1701 | return -EINVAL; | |
1702 | } | |
1703 | if (!refcount_inc_not_zero(&nbd->refs)) { | |
1704 | mutex_unlock(&nbd_index_mutex); | |
1705 | if (index == -1) | |
1706 | goto again; | |
1707 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1708 | index); | |
e46c7287 JB |
1709 | return -EINVAL; |
1710 | } | |
c6a4759e | 1711 | mutex_unlock(&nbd_index_mutex); |
e46c7287 JB |
1712 | |
1713 | mutex_lock(&nbd->config_lock); | |
1714 | if (refcount_read(&nbd->config_refs)) { | |
1715 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1716 | nbd_put(nbd); |
e46c7287 JB |
1717 | if (index == -1) |
1718 | goto again; | |
1719 | printk(KERN_ERR "nbd: nbd%d already in use\n", index); | |
1720 | return -EBUSY; | |
1721 | } | |
1722 | if (WARN_ON(nbd->config)) { | |
1723 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1724 | nbd_put(nbd); |
e46c7287 JB |
1725 | return -EINVAL; |
1726 | } | |
1727 | config = nbd->config = nbd_alloc_config(); | |
1728 | if (!nbd->config) { | |
1729 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1730 | nbd_put(nbd); |
e46c7287 JB |
1731 | printk(KERN_ERR "nbd: couldn't allocate config\n"); |
1732 | return -ENOMEM; | |
1733 | } | |
1734 | refcount_set(&nbd->config_refs, 1); | |
1735 | set_bit(NBD_BOUND, &config->runtime_flags); | |
1736 | ||
1737 | if (info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1738 | u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]); | |
1739 | nbd_size_set(nbd, config->blksize, | |
1740 | div64_u64(bytes, config->blksize)); | |
1741 | } | |
1742 | if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) { | |
1743 | u64 bsize = | |
1744 | nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]); | |
1745 | nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize)); | |
1746 | } | |
1747 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1748 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1749 | nbd->tag_set.timeout = timeout * HZ; | |
1750 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1751 | } | |
560bc4b3 JB |
1752 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1753 | config->dead_conn_timeout = | |
1754 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1755 | config->dead_conn_timeout *= HZ; | |
1756 | } | |
e46c7287 JB |
1757 | if (info->attrs[NBD_ATTR_SERVER_FLAGS]) |
1758 | config->flags = | |
1759 | nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]); | |
a2c97909 JB |
1760 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { |
1761 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1762 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1763 | set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1764 | &config->runtime_flags); | |
1765 | put_dev = true; | |
1766 | } | |
9b0b6258 DRK |
1767 | if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) { |
1768 | set_bit(NBD_DISCONNECT_ON_CLOSE, | |
1769 | &config->runtime_flags); | |
1770 | } | |
a2c97909 JB |
1771 | } |
1772 | ||
e46c7287 JB |
1773 | if (info->attrs[NBD_ATTR_SOCKETS]) { |
1774 | struct nlattr *attr; | |
1775 | int rem, fd; | |
1776 | ||
1777 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1778 | rem) { | |
1779 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1780 | ||
1781 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1782 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1783 | ret = -EINVAL; | |
1784 | goto out; | |
1785 | } | |
1786 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
8d65b08d | 1787 | nbd_sock_policy, info->extack); |
e46c7287 JB |
1788 | if (ret != 0) { |
1789 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1790 | ret = -EINVAL; | |
1791 | goto out; | |
1792 | } | |
1793 | if (!socks[NBD_SOCK_FD]) | |
1794 | continue; | |
1795 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1796 | ret = nbd_add_socket(nbd, fd, true); | |
1797 | if (ret) | |
1798 | goto out; | |
1799 | } | |
1800 | } | |
1801 | ret = nbd_start_device(nbd); | |
1802 | out: | |
1803 | mutex_unlock(&nbd->config_lock); | |
1804 | if (!ret) { | |
1805 | set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags); | |
1806 | refcount_inc(&nbd->config_refs); | |
1807 | nbd_connect_reply(info, nbd->index); | |
1808 | } | |
1809 | nbd_config_put(nbd); | |
a2c97909 JB |
1810 | if (put_dev) |
1811 | nbd_put(nbd); | |
e46c7287 JB |
1812 | return ret; |
1813 | } | |
1814 | ||
9b0b6258 DRK |
1815 | static void nbd_disconnect_and_put(struct nbd_device *nbd) |
1816 | { | |
1817 | mutex_lock(&nbd->config_lock); | |
1818 | nbd_disconnect(nbd); | |
1819 | mutex_unlock(&nbd->config_lock); | |
1820 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, | |
1821 | &nbd->config->runtime_flags)) | |
1822 | nbd_config_put(nbd); | |
1823 | } | |
1824 | ||
e46c7287 JB |
1825 | static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info) |
1826 | { | |
1827 | struct nbd_device *nbd; | |
1828 | int index; | |
1829 | ||
1830 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1831 | return -EPERM; | |
1832 | ||
1833 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1834 | printk(KERN_ERR "nbd: must specify an index to disconnect\n"); | |
1835 | return -EINVAL; | |
1836 | } | |
1837 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1838 | mutex_lock(&nbd_index_mutex); | |
1839 | nbd = idr_find(&nbd_index_idr, index); | |
e46c7287 | 1840 | if (!nbd) { |
c6a4759e | 1841 | mutex_unlock(&nbd_index_mutex); |
e46c7287 JB |
1842 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", |
1843 | index); | |
1844 | return -EINVAL; | |
1845 | } | |
c6a4759e JB |
1846 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1847 | mutex_unlock(&nbd_index_mutex); | |
1848 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1849 | index); | |
1850 | return -EINVAL; | |
1851 | } | |
1852 | mutex_unlock(&nbd_index_mutex); | |
1853 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1854 | nbd_put(nbd); | |
e46c7287 | 1855 | return 0; |
c6a4759e | 1856 | } |
9b0b6258 | 1857 | nbd_disconnect_and_put(nbd); |
e46c7287 | 1858 | nbd_config_put(nbd); |
c6a4759e | 1859 | nbd_put(nbd); |
e46c7287 JB |
1860 | return 0; |
1861 | } | |
1862 | ||
b7aa3d39 JB |
1863 | static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info) |
1864 | { | |
1865 | struct nbd_device *nbd = NULL; | |
1866 | struct nbd_config *config; | |
1867 | int index; | |
9b0b6258 | 1868 | int ret = 0; |
a2c97909 | 1869 | bool put_dev = false; |
b7aa3d39 JB |
1870 | |
1871 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1872 | return -EPERM; | |
1873 | ||
1874 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1875 | printk(KERN_ERR "nbd: must specify a device to reconfigure\n"); | |
1876 | return -EINVAL; | |
1877 | } | |
1878 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1879 | mutex_lock(&nbd_index_mutex); | |
1880 | nbd = idr_find(&nbd_index_idr, index); | |
b7aa3d39 | 1881 | if (!nbd) { |
c6a4759e | 1882 | mutex_unlock(&nbd_index_mutex); |
b7aa3d39 JB |
1883 | printk(KERN_ERR "nbd: couldn't find a device at index %d\n", |
1884 | index); | |
1885 | return -EINVAL; | |
1886 | } | |
c6a4759e JB |
1887 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1888 | mutex_unlock(&nbd_index_mutex); | |
1889 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1890 | index); | |
1891 | return -EINVAL; | |
1892 | } | |
1893 | mutex_unlock(&nbd_index_mutex); | |
b7aa3d39 JB |
1894 | |
1895 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1896 | dev_err(nbd_to_dev(nbd), | |
1897 | "not configured, cannot reconfigure\n"); | |
c6a4759e | 1898 | nbd_put(nbd); |
b7aa3d39 JB |
1899 | return -EINVAL; |
1900 | } | |
1901 | ||
1902 | mutex_lock(&nbd->config_lock); | |
1903 | config = nbd->config; | |
1904 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1905 | !nbd->task_recv) { | |
1906 | dev_err(nbd_to_dev(nbd), | |
1907 | "not configured, cannot reconfigure\n"); | |
9b0b6258 | 1908 | ret = -EINVAL; |
b7aa3d39 JB |
1909 | goto out; |
1910 | } | |
1911 | ||
1912 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1913 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1914 | nbd->tag_set.timeout = timeout * HZ; | |
1915 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1916 | } | |
560bc4b3 JB |
1917 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1918 | config->dead_conn_timeout = | |
1919 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1920 | config->dead_conn_timeout *= HZ; | |
1921 | } | |
a2c97909 JB |
1922 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { |
1923 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1924 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1925 | if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1926 | &config->runtime_flags)) | |
1927 | put_dev = true; | |
1928 | } else { | |
1929 | if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT, | |
1930 | &config->runtime_flags)) | |
1931 | refcount_inc(&nbd->refs); | |
1932 | } | |
9b0b6258 DRK |
1933 | |
1934 | if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) { | |
1935 | set_bit(NBD_DISCONNECT_ON_CLOSE, | |
1936 | &config->runtime_flags); | |
1937 | } else { | |
1938 | clear_bit(NBD_DISCONNECT_ON_CLOSE, | |
1939 | &config->runtime_flags); | |
1940 | } | |
a2c97909 | 1941 | } |
b7aa3d39 JB |
1942 | |
1943 | if (info->attrs[NBD_ATTR_SOCKETS]) { | |
1944 | struct nlattr *attr; | |
1945 | int rem, fd; | |
1946 | ||
1947 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1948 | rem) { | |
1949 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1950 | ||
1951 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1952 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1953 | ret = -EINVAL; | |
1954 | goto out; | |
1955 | } | |
1956 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
8d65b08d | 1957 | nbd_sock_policy, info->extack); |
b7aa3d39 JB |
1958 | if (ret != 0) { |
1959 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1960 | ret = -EINVAL; | |
1961 | goto out; | |
1962 | } | |
1963 | if (!socks[NBD_SOCK_FD]) | |
1964 | continue; | |
1965 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1966 | ret = nbd_reconnect_socket(nbd, fd); | |
1967 | if (ret) { | |
1968 | if (ret == -ENOSPC) | |
1969 | ret = 0; | |
1970 | goto out; | |
1971 | } | |
1972 | dev_info(nbd_to_dev(nbd), "reconnected socket\n"); | |
1973 | } | |
1974 | } | |
1975 | out: | |
1976 | mutex_unlock(&nbd->config_lock); | |
1977 | nbd_config_put(nbd); | |
c6a4759e | 1978 | nbd_put(nbd); |
a2c97909 JB |
1979 | if (put_dev) |
1980 | nbd_put(nbd); | |
b7aa3d39 JB |
1981 | return ret; |
1982 | } | |
1983 | ||
e46c7287 JB |
1984 | static const struct genl_ops nbd_connect_genl_ops[] = { |
1985 | { | |
1986 | .cmd = NBD_CMD_CONNECT, | |
1987 | .policy = nbd_attr_policy, | |
1988 | .doit = nbd_genl_connect, | |
1989 | }, | |
1990 | { | |
1991 | .cmd = NBD_CMD_DISCONNECT, | |
1992 | .policy = nbd_attr_policy, | |
1993 | .doit = nbd_genl_disconnect, | |
1994 | }, | |
b7aa3d39 JB |
1995 | { |
1996 | .cmd = NBD_CMD_RECONFIGURE, | |
1997 | .policy = nbd_attr_policy, | |
1998 | .doit = nbd_genl_reconfigure, | |
1999 | }, | |
47d902b9 JB |
2000 | { |
2001 | .cmd = NBD_CMD_STATUS, | |
2002 | .policy = nbd_attr_policy, | |
2003 | .doit = nbd_genl_status, | |
2004 | }, | |
e46c7287 JB |
2005 | }; |
2006 | ||
799f9a38 JB |
2007 | static const struct genl_multicast_group nbd_mcast_grps[] = { |
2008 | { .name = NBD_GENL_MCAST_GROUP_NAME, }, | |
2009 | }; | |
2010 | ||
e46c7287 JB |
2011 | static struct genl_family nbd_genl_family __ro_after_init = { |
2012 | .hdrsize = 0, | |
2013 | .name = NBD_GENL_FAMILY_NAME, | |
2014 | .version = NBD_GENL_VERSION, | |
2015 | .module = THIS_MODULE, | |
2016 | .ops = nbd_connect_genl_ops, | |
2017 | .n_ops = ARRAY_SIZE(nbd_connect_genl_ops), | |
2018 | .maxattr = NBD_ATTR_MAX, | |
799f9a38 JB |
2019 | .mcgrps = nbd_mcast_grps, |
2020 | .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps), | |
e46c7287 JB |
2021 | }; |
2022 | ||
47d902b9 JB |
2023 | static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply) |
2024 | { | |
2025 | struct nlattr *dev_opt; | |
2026 | u8 connected = 0; | |
2027 | int ret; | |
2028 | ||
2029 | /* This is a little racey, but for status it's ok. The | |
2030 | * reason we don't take a ref here is because we can't | |
2031 | * take a ref in the index == -1 case as we would need | |
2032 | * to put under the nbd_index_mutex, which could | |
2033 | * deadlock if we are configured to remove ourselves | |
2034 | * once we're disconnected. | |
2035 | */ | |
2036 | if (refcount_read(&nbd->config_refs)) | |
2037 | connected = 1; | |
2038 | dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM); | |
2039 | if (!dev_opt) | |
2040 | return -EMSGSIZE; | |
2041 | ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index); | |
2042 | if (ret) | |
2043 | return -EMSGSIZE; | |
2044 | ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED, | |
2045 | connected); | |
2046 | if (ret) | |
2047 | return -EMSGSIZE; | |
2048 | nla_nest_end(reply, dev_opt); | |
2049 | return 0; | |
2050 | } | |
2051 | ||
2052 | static int status_cb(int id, void *ptr, void *data) | |
2053 | { | |
2054 | struct nbd_device *nbd = ptr; | |
2055 | return populate_nbd_status(nbd, (struct sk_buff *)data); | |
2056 | } | |
2057 | ||
2058 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info) | |
2059 | { | |
2060 | struct nlattr *dev_list; | |
2061 | struct sk_buff *reply; | |
2062 | void *reply_head; | |
2063 | size_t msg_size; | |
2064 | int index = -1; | |
2065 | int ret = -ENOMEM; | |
2066 | ||
2067 | if (info->attrs[NBD_ATTR_INDEX]) | |
2068 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
2069 | ||
2070 | mutex_lock(&nbd_index_mutex); | |
2071 | ||
2072 | msg_size = nla_total_size(nla_attr_size(sizeof(u32)) + | |
2073 | nla_attr_size(sizeof(u8))); | |
2074 | msg_size *= (index == -1) ? nbd_total_devices : 1; | |
2075 | ||
2076 | reply = genlmsg_new(msg_size, GFP_KERNEL); | |
2077 | if (!reply) | |
2078 | goto out; | |
2079 | reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0, | |
2080 | NBD_CMD_STATUS); | |
2081 | if (!reply_head) { | |
2082 | nlmsg_free(reply); | |
2083 | goto out; | |
2084 | } | |
2085 | ||
2086 | dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST); | |
2087 | if (index == -1) { | |
2088 | ret = idr_for_each(&nbd_index_idr, &status_cb, reply); | |
2089 | if (ret) { | |
2090 | nlmsg_free(reply); | |
2091 | goto out; | |
2092 | } | |
2093 | } else { | |
2094 | struct nbd_device *nbd; | |
2095 | nbd = idr_find(&nbd_index_idr, index); | |
2096 | if (nbd) { | |
2097 | ret = populate_nbd_status(nbd, reply); | |
2098 | if (ret) { | |
2099 | nlmsg_free(reply); | |
2100 | goto out; | |
2101 | } | |
2102 | } | |
2103 | } | |
2104 | nla_nest_end(reply, dev_list); | |
2105 | genlmsg_end(reply, reply_head); | |
2106 | genlmsg_reply(reply, info); | |
2107 | ret = 0; | |
2108 | out: | |
2109 | mutex_unlock(&nbd_index_mutex); | |
2110 | return ret; | |
2111 | } | |
2112 | ||
e46c7287 JB |
2113 | static void nbd_connect_reply(struct genl_info *info, int index) |
2114 | { | |
2115 | struct sk_buff *skb; | |
2116 | void *msg_head; | |
2117 | int ret; | |
2118 | ||
2119 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
2120 | if (!skb) | |
2121 | return; | |
2122 | msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0, | |
2123 | NBD_CMD_CONNECT); | |
2124 | if (!msg_head) { | |
2125 | nlmsg_free(skb); | |
2126 | return; | |
2127 | } | |
2128 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
2129 | if (ret) { | |
2130 | nlmsg_free(skb); | |
2131 | return; | |
2132 | } | |
2133 | genlmsg_end(skb, msg_head); | |
2134 | genlmsg_reply(skb, info); | |
2135 | } | |
1da177e4 | 2136 | |
799f9a38 JB |
2137 | static void nbd_mcast_index(int index) |
2138 | { | |
2139 | struct sk_buff *skb; | |
2140 | void *msg_head; | |
2141 | int ret; | |
2142 | ||
2143 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
2144 | if (!skb) | |
2145 | return; | |
2146 | msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0, | |
2147 | NBD_CMD_LINK_DEAD); | |
2148 | if (!msg_head) { | |
2149 | nlmsg_free(skb); | |
2150 | return; | |
2151 | } | |
2152 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
2153 | if (ret) { | |
2154 | nlmsg_free(skb); | |
2155 | return; | |
2156 | } | |
2157 | genlmsg_end(skb, msg_head); | |
2158 | genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL); | |
2159 | } | |
2160 | ||
2161 | static void nbd_dead_link_work(struct work_struct *work) | |
2162 | { | |
2163 | struct link_dead_args *args = container_of(work, struct link_dead_args, | |
2164 | work); | |
2165 | nbd_mcast_index(args->index); | |
2166 | kfree(args); | |
2167 | } | |
2168 | ||
1da177e4 LT |
2169 | static int __init nbd_init(void) |
2170 | { | |
1da177e4 LT |
2171 | int i; |
2172 | ||
5b7b18cc | 2173 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 2174 | |
d71a6d73 | 2175 | if (max_part < 0) { |
7742ce4a | 2176 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
2177 | return -EINVAL; |
2178 | } | |
2179 | ||
2180 | part_shift = 0; | |
5988ce23 | 2181 | if (max_part > 0) { |
d71a6d73 LV |
2182 | part_shift = fls(max_part); |
2183 | ||
5988ce23 NK |
2184 | /* |
2185 | * Adjust max_part according to part_shift as it is exported | |
2186 | * to user space so that user can know the max number of | |
2187 | * partition kernel should be able to manage. | |
2188 | * | |
2189 | * Note that -1 is required because partition 0 is reserved | |
2190 | * for the whole disk. | |
2191 | */ | |
2192 | max_part = (1UL << part_shift) - 1; | |
2193 | } | |
2194 | ||
3b271082 NK |
2195 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
2196 | return -EINVAL; | |
2197 | ||
2198 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
2199 | return -EINVAL; | |
124d6db0 JB |
2200 | recv_workqueue = alloc_workqueue("knbd-recv", |
2201 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
2202 | if (!recv_workqueue) | |
2203 | return -ENOMEM; | |
3b271082 | 2204 | |
6330a2d0 JB |
2205 | if (register_blkdev(NBD_MAJOR, "nbd")) { |
2206 | destroy_workqueue(recv_workqueue); | |
b0d9111a | 2207 | return -EIO; |
6330a2d0 | 2208 | } |
1da177e4 | 2209 | |
e46c7287 JB |
2210 | if (genl_register_family(&nbd_genl_family)) { |
2211 | unregister_blkdev(NBD_MAJOR, "nbd"); | |
2212 | destroy_workqueue(recv_workqueue); | |
2213 | return -EINVAL; | |
2214 | } | |
30d53d9c MP |
2215 | nbd_dbg_init(); |
2216 | ||
b0d9111a JB |
2217 | mutex_lock(&nbd_index_mutex); |
2218 | for (i = 0; i < nbds_max; i++) | |
2219 | nbd_dev_add(i); | |
2220 | mutex_unlock(&nbd_index_mutex); | |
2221 | return 0; | |
2222 | } | |
1da177e4 | 2223 | |
b0d9111a JB |
2224 | static int nbd_exit_cb(int id, void *ptr, void *data) |
2225 | { | |
c6a4759e | 2226 | struct list_head *list = (struct list_head *)data; |
b0d9111a | 2227 | struct nbd_device *nbd = ptr; |
c6a4759e | 2228 | |
c6a4759e | 2229 | list_add_tail(&nbd->list, list); |
1da177e4 | 2230 | return 0; |
1da177e4 LT |
2231 | } |
2232 | ||
2233 | static void __exit nbd_cleanup(void) | |
2234 | { | |
c6a4759e JB |
2235 | struct nbd_device *nbd; |
2236 | LIST_HEAD(del_list); | |
2237 | ||
30d53d9c MP |
2238 | nbd_dbg_close(); |
2239 | ||
c6a4759e JB |
2240 | mutex_lock(&nbd_index_mutex); |
2241 | idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list); | |
2242 | mutex_unlock(&nbd_index_mutex); | |
2243 | ||
60ae36ad JB |
2244 | while (!list_empty(&del_list)) { |
2245 | nbd = list_first_entry(&del_list, struct nbd_device, list); | |
2246 | list_del_init(&nbd->list); | |
2247 | if (refcount_read(&nbd->refs) != 1) | |
c6a4759e JB |
2248 | printk(KERN_ERR "nbd: possibly leaking a device\n"); |
2249 | nbd_put(nbd); | |
c6a4759e JB |
2250 | } |
2251 | ||
b0d9111a | 2252 | idr_destroy(&nbd_index_idr); |
e46c7287 | 2253 | genl_unregister_family(&nbd_genl_family); |
124d6db0 | 2254 | destroy_workqueue(recv_workqueue); |
1da177e4 | 2255 | unregister_blkdev(NBD_MAJOR, "nbd"); |
1da177e4 LT |
2256 | } |
2257 | ||
2258 | module_init(nbd_init); | |
2259 | module_exit(nbd_cleanup); | |
2260 | ||
2261 | MODULE_DESCRIPTION("Network Block Device"); | |
2262 | MODULE_LICENSE("GPL"); | |
2263 | ||
40be0c28 | 2264 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
2265 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
2266 | module_param(max_part, int, 0444); | |
7a8362a0 | 2267 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)"); |