Commit | Line | Data |
---|---|---|
eb1fe3bf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * Network block device - make block devices work over TCP | |
4 | * | |
5 | * Note that you can not swap over this thing, yet. Seems to work but | |
6 | * deadlocks sometimes - you can not swap over TCP in general. | |
7 | * | |
a2531293 | 8 | * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz> |
1da177e4 LT |
9 | * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com> |
10 | * | |
dbf492d6 | 11 | * (part of code stolen from loop.c) |
1da177e4 LT |
12 | */ |
13 | ||
14 | #include <linux/major.h> | |
15 | ||
16 | #include <linux/blkdev.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/sched.h> | |
f1083048 | 20 | #include <linux/sched/mm.h> |
1da177e4 LT |
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 | |
ea106722 MM |
46 | #define CREATE_TRACE_POINTS |
47 | #include <trace/events/nbd.h> | |
48 | ||
b0d9111a JB |
49 | static DEFINE_IDR(nbd_index_idr); |
50 | static DEFINE_MUTEX(nbd_index_mutex); | |
47d902b9 | 51 | static int nbd_total_devices = 0; |
b0d9111a | 52 | |
9561a7ad JB |
53 | struct nbd_sock { |
54 | struct socket *sock; | |
55 | struct mutex tx_lock; | |
9dd5d3ab JB |
56 | struct request *pending; |
57 | int sent; | |
f3733247 JB |
58 | bool dead; |
59 | int fallback_index; | |
799f9a38 | 60 | int cookie; |
9561a7ad JB |
61 | }; |
62 | ||
5ea8d108 JB |
63 | struct recv_thread_args { |
64 | struct work_struct work; | |
65 | struct nbd_device *nbd; | |
66 | int index; | |
67 | }; | |
68 | ||
799f9a38 JB |
69 | struct link_dead_args { |
70 | struct work_struct work; | |
71 | int index; | |
72 | }; | |
73 | ||
9b4a6ba9 JB |
74 | #define NBD_TIMEDOUT 0 |
75 | #define NBD_DISCONNECT_REQUESTED 1 | |
9561a7ad | 76 | #define NBD_DISCONNECTED 2 |
5ea8d108 | 77 | #define NBD_HAS_PID_FILE 3 |
e46c7287 JB |
78 | #define NBD_HAS_CONFIG_REF 4 |
79 | #define NBD_BOUND 5 | |
a2c97909 | 80 | #define NBD_DESTROY_ON_DISCONNECT 6 |
08ba91ee | 81 | #define NBD_DISCONNECT_ON_CLOSE 7 |
9b4a6ba9 | 82 | |
5ea8d108 | 83 | struct nbd_config { |
22d109c1 | 84 | u32 flags; |
9b4a6ba9 | 85 | unsigned long runtime_flags; |
560bc4b3 | 86 | u64 dead_conn_timeout; |
13e71d69 | 87 | |
5ea8d108 | 88 | struct nbd_sock **socks; |
9561a7ad | 89 | int num_connections; |
560bc4b3 JB |
90 | atomic_t live_connections; |
91 | wait_queue_head_t conn_wait; | |
5ea8d108 | 92 | |
9561a7ad JB |
93 | atomic_t recv_threads; |
94 | wait_queue_head_t recv_wq; | |
ef77b515 | 95 | loff_t blksize; |
b9c495bb | 96 | loff_t bytesize; |
30d53d9c MP |
97 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
98 | struct dentry *dbg_dir; | |
99 | #endif | |
13e71d69 MP |
100 | }; |
101 | ||
5ea8d108 JB |
102 | struct nbd_device { |
103 | struct blk_mq_tag_set tag_set; | |
104 | ||
e46c7287 | 105 | int index; |
5ea8d108 | 106 | refcount_t config_refs; |
c6a4759e | 107 | refcount_t refs; |
5ea8d108 JB |
108 | struct nbd_config *config; |
109 | struct mutex config_lock; | |
110 | struct gendisk *disk; | |
111 | ||
c6a4759e | 112 | struct list_head list; |
5ea8d108 JB |
113 | struct task_struct *task_recv; |
114 | struct task_struct *task_setup; | |
115 | }; | |
116 | ||
d7d94d48 JB |
117 | #define NBD_CMD_REQUEUED 1 |
118 | ||
fd8383fd JB |
119 | struct nbd_cmd { |
120 | struct nbd_device *nbd; | |
8f3ea359 | 121 | struct mutex lock; |
f3733247 | 122 | int index; |
799f9a38 | 123 | int cookie; |
2a842aca | 124 | blk_status_t status; |
d7d94d48 | 125 | unsigned long flags; |
8f3ea359 | 126 | u32 cmd_cookie; |
fd8383fd JB |
127 | }; |
128 | ||
30d53d9c MP |
129 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
130 | static struct dentry *nbd_dbg_dir; | |
131 | #endif | |
132 | ||
133 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
134 | ||
f4507164 | 135 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 136 | |
9c7a4169 | 137 | static unsigned int nbds_max = 16; |
7a8362a0 | 138 | static int max_part = 16; |
124d6db0 | 139 | static struct workqueue_struct *recv_workqueue; |
b0d9111a | 140 | static int part_shift; |
1da177e4 | 141 | |
9442b739 JB |
142 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
143 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
5ea8d108 | 144 | static void nbd_config_put(struct nbd_device *nbd); |
e46c7287 | 145 | static void nbd_connect_reply(struct genl_info *info, int index); |
47d902b9 | 146 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info); |
799f9a38 | 147 | static void nbd_dead_link_work(struct work_struct *work); |
08ba91ee | 148 | static void nbd_disconnect_and_put(struct nbd_device *nbd); |
9442b739 | 149 | |
d18509f5 | 150 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 151 | { |
d18509f5 | 152 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
153 | } |
154 | ||
d7d94d48 JB |
155 | static void nbd_requeue_cmd(struct nbd_cmd *cmd) |
156 | { | |
157 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
158 | ||
159 | if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags)) | |
160 | blk_mq_requeue_request(req, true); | |
161 | } | |
162 | ||
8f3ea359 JB |
163 | #define NBD_COOKIE_BITS 32 |
164 | ||
165 | static u64 nbd_cmd_handle(struct nbd_cmd *cmd) | |
166 | { | |
167 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
168 | u32 tag = blk_mq_unique_tag(req); | |
169 | u64 cookie = cmd->cmd_cookie; | |
170 | ||
171 | return (cookie << NBD_COOKIE_BITS) | tag; | |
172 | } | |
173 | ||
174 | static u32 nbd_handle_to_tag(u64 handle) | |
175 | { | |
176 | return (u32)handle; | |
177 | } | |
178 | ||
179 | static u32 nbd_handle_to_cookie(u64 handle) | |
180 | { | |
181 | return (u32)(handle >> NBD_COOKIE_BITS); | |
182 | } | |
183 | ||
1da177e4 LT |
184 | static const char *nbdcmd_to_ascii(int cmd) |
185 | { | |
186 | switch (cmd) { | |
187 | case NBD_CMD_READ: return "read"; | |
188 | case NBD_CMD_WRITE: return "write"; | |
189 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 190 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 191 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
192 | } |
193 | return "invalid"; | |
194 | } | |
1da177e4 | 195 | |
5ea8d108 JB |
196 | static ssize_t pid_show(struct device *dev, |
197 | struct device_attribute *attr, char *buf) | |
198 | { | |
199 | struct gendisk *disk = dev_to_disk(dev); | |
200 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; | |
201 | ||
202 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); | |
203 | } | |
204 | ||
dfbde552 | 205 | static const struct device_attribute pid_attr = { |
5657a819 | 206 | .attr = { .name = "pid", .mode = 0444}, |
5ea8d108 JB |
207 | .show = pid_show, |
208 | }; | |
209 | ||
c6a4759e JB |
210 | static void nbd_dev_remove(struct nbd_device *nbd) |
211 | { | |
212 | struct gendisk *disk = nbd->disk; | |
8364da47 JB |
213 | struct request_queue *q; |
214 | ||
c6a4759e | 215 | if (disk) { |
8364da47 | 216 | q = disk->queue; |
c6a4759e | 217 | del_gendisk(disk); |
8364da47 | 218 | blk_cleanup_queue(q); |
c6a4759e | 219 | blk_mq_free_tag_set(&nbd->tag_set); |
a2c97909 | 220 | disk->private_data = NULL; |
c6a4759e JB |
221 | put_disk(disk); |
222 | } | |
223 | kfree(nbd); | |
224 | } | |
225 | ||
226 | static void nbd_put(struct nbd_device *nbd) | |
227 | { | |
228 | if (refcount_dec_and_mutex_lock(&nbd->refs, | |
229 | &nbd_index_mutex)) { | |
230 | idr_remove(&nbd_index_idr, nbd->index); | |
231 | mutex_unlock(&nbd_index_mutex); | |
232 | nbd_dev_remove(nbd); | |
233 | } | |
234 | } | |
235 | ||
799f9a38 JB |
236 | static int nbd_disconnected(struct nbd_config *config) |
237 | { | |
238 | return test_bit(NBD_DISCONNECTED, &config->runtime_flags) || | |
239 | test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); | |
240 | } | |
241 | ||
242 | static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock, | |
243 | int notify) | |
f3733247 | 244 | { |
799f9a38 JB |
245 | if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) { |
246 | struct link_dead_args *args; | |
247 | args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO); | |
248 | if (args) { | |
249 | INIT_WORK(&args->work, nbd_dead_link_work); | |
250 | args->index = nbd->index; | |
251 | queue_work(system_wq, &args->work); | |
252 | } | |
253 | } | |
560bc4b3 | 254 | if (!nsock->dead) { |
f3733247 | 255 | kernel_sock_shutdown(nsock->sock, SHUT_RDWR); |
5e3c3a7e KV |
256 | if (atomic_dec_return(&nbd->config->live_connections) == 0) { |
257 | if (test_and_clear_bit(NBD_DISCONNECT_REQUESTED, | |
258 | &nbd->config->runtime_flags)) { | |
259 | set_bit(NBD_DISCONNECTED, | |
260 | &nbd->config->runtime_flags); | |
261 | dev_info(nbd_to_dev(nbd), | |
262 | "Disconnected due to user request.\n"); | |
263 | } | |
264 | } | |
560bc4b3 | 265 | } |
f3733247 JB |
266 | nsock->dead = true; |
267 | nsock->pending = NULL; | |
268 | nsock->sent = 0; | |
269 | } | |
270 | ||
29eaadc0 | 271 | static void nbd_size_clear(struct nbd_device *nbd) |
37091fdd | 272 | { |
5ea8d108 | 273 | if (nbd->config->bytesize) { |
5ea8d108 JB |
274 | set_capacity(nbd->disk, 0); |
275 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
276 | } | |
37091fdd MP |
277 | } |
278 | ||
29eaadc0 | 279 | static void nbd_size_update(struct nbd_device *nbd) |
37091fdd | 280 | { |
5ea8d108 | 281 | struct nbd_config *config = nbd->config; |
9e2b1967 JB |
282 | struct block_device *bdev = bdget_disk(nbd->disk, 0); |
283 | ||
6df133a1 JB |
284 | if (config->flags & NBD_FLAG_SEND_TRIM) { |
285 | nbd->disk->queue->limits.discard_granularity = config->blksize; | |
07ce213f | 286 | nbd->disk->queue->limits.discard_alignment = config->blksize; |
6df133a1 JB |
287 | blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX); |
288 | } | |
5ea8d108 JB |
289 | blk_queue_logical_block_size(nbd->disk->queue, config->blksize); |
290 | blk_queue_physical_block_size(nbd->disk->queue, config->blksize); | |
5ea8d108 | 291 | set_capacity(nbd->disk, config->bytesize >> 9); |
9e2b1967 | 292 | if (bdev) { |
c8a83a6b | 293 | if (bdev->bd_disk) { |
9e2b1967 | 294 | bd_set_size(bdev, config->bytesize); |
c8a83a6b JK |
295 | set_blocksize(bdev, config->blksize); |
296 | } else | |
9e2b1967 JB |
297 | bdev->bd_invalidated = 1; |
298 | bdput(bdev); | |
299 | } | |
37091fdd MP |
300 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); |
301 | } | |
302 | ||
29eaadc0 JB |
303 | static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize, |
304 | loff_t nr_blocks) | |
37091fdd | 305 | { |
5ea8d108 JB |
306 | struct nbd_config *config = nbd->config; |
307 | config->blksize = blocksize; | |
308 | config->bytesize = blocksize * nr_blocks; | |
c3f7c939 JB |
309 | if (nbd->task_recv != NULL) |
310 | nbd_size_update(nbd); | |
37091fdd MP |
311 | } |
312 | ||
1e388ae0 | 313 | static void nbd_complete_rq(struct request *req) |
1da177e4 | 314 | { |
1e388ae0 | 315 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
1da177e4 | 316 | |
ee57a05c | 317 | dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req, |
1e388ae0 | 318 | cmd->status ? "failed" : "done"); |
1da177e4 | 319 | |
1e388ae0 | 320 | blk_mq_end_request(req, cmd->status); |
1da177e4 LT |
321 | } |
322 | ||
e018e757 MP |
323 | /* |
324 | * Forcibly shutdown the socket causing all listeners to error | |
325 | */ | |
36e47bee | 326 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 327 | { |
5ea8d108 | 328 | struct nbd_config *config = nbd->config; |
9561a7ad | 329 | int i; |
23272a67 | 330 | |
5ea8d108 | 331 | if (config->num_connections == 0) |
9561a7ad | 332 | return; |
5ea8d108 | 333 | if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
260bbce4 | 334 | return; |
23272a67 | 335 | |
5ea8d108 JB |
336 | for (i = 0; i < config->num_connections; i++) { |
337 | struct nbd_sock *nsock = config->socks[i]; | |
9561a7ad | 338 | mutex_lock(&nsock->tx_lock); |
799f9a38 | 339 | nbd_mark_nsock_dead(nbd, nsock, 0); |
9561a7ad JB |
340 | mutex_unlock(&nsock->tx_lock); |
341 | } | |
342 | dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n"); | |
7fdfd406 PC |
343 | } |
344 | ||
0eadf37a JB |
345 | static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req, |
346 | bool reserved) | |
7fdfd406 | 347 | { |
0eadf37a JB |
348 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
349 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 JB |
350 | struct nbd_config *config; |
351 | ||
352 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
2a842aca | 353 | cmd->status = BLK_STS_TIMEOUT; |
e5eab017 | 354 | goto done; |
5ea8d108 | 355 | } |
5ea8d108 | 356 | config = nbd->config; |
dcc909d9 | 357 | |
8f3ea359 JB |
358 | if (!mutex_trylock(&cmd->lock)) |
359 | return BLK_EH_RESET_TIMER; | |
360 | ||
5ea8d108 | 361 | if (config->num_connections > 1) { |
f3733247 | 362 | dev_err_ratelimited(nbd_to_dev(nbd), |
5e3c3a7e KV |
363 | "Connection timed out, retrying (%d/%d alive)\n", |
364 | atomic_read(&config->live_connections), | |
365 | config->num_connections); | |
f3733247 JB |
366 | /* |
367 | * Hooray we have more connections, requeue this IO, the submit | |
368 | * path will put it on a real connection. | |
369 | */ | |
5ea8d108 JB |
370 | if (config->socks && config->num_connections > 1) { |
371 | if (cmd->index < config->num_connections) { | |
f3733247 | 372 | struct nbd_sock *nsock = |
5ea8d108 | 373 | config->socks[cmd->index]; |
f3733247 | 374 | mutex_lock(&nsock->tx_lock); |
799f9a38 JB |
375 | /* We can have multiple outstanding requests, so |
376 | * we don't want to mark the nsock dead if we've | |
377 | * already reconnected with a new socket, so | |
378 | * only mark it dead if its the same socket we | |
379 | * were sent out on. | |
380 | */ | |
381 | if (cmd->cookie == nsock->cookie) | |
382 | nbd_mark_nsock_dead(nbd, nsock, 1); | |
f3733247 JB |
383 | mutex_unlock(&nsock->tx_lock); |
384 | } | |
8f3ea359 | 385 | mutex_unlock(&cmd->lock); |
d7d94d48 | 386 | nbd_requeue_cmd(cmd); |
5ea8d108 | 387 | nbd_config_put(nbd); |
6600593c | 388 | return BLK_EH_DONE; |
f3733247 | 389 | } |
f3733247 JB |
390 | } else { |
391 | dev_err_ratelimited(nbd_to_dev(nbd), | |
392 | "Connection timed out\n"); | |
393 | } | |
5ea8d108 | 394 | set_bit(NBD_TIMEDOUT, &config->runtime_flags); |
2a842aca | 395 | cmd->status = BLK_STS_IOERR; |
8f3ea359 | 396 | mutex_unlock(&cmd->lock); |
9561a7ad | 397 | sock_shutdown(nbd); |
5ea8d108 | 398 | nbd_config_put(nbd); |
e5eab017 CH |
399 | done: |
400 | blk_mq_complete_request(req); | |
401 | return BLK_EH_DONE; | |
7fdfd406 PC |
402 | } |
403 | ||
1da177e4 LT |
404 | /* |
405 | * Send or receive packet. | |
406 | */ | |
c9f2b6ae | 407 | static int sock_xmit(struct nbd_device *nbd, int index, int send, |
9dd5d3ab | 408 | struct iov_iter *iter, int msg_flags, int *sent) |
1da177e4 | 409 | { |
5ea8d108 JB |
410 | struct nbd_config *config = nbd->config; |
411 | struct socket *sock = config->socks[index]->sock; | |
1da177e4 LT |
412 | int result; |
413 | struct msghdr msg; | |
f1083048 | 414 | unsigned int noreclaim_flag; |
1da177e4 | 415 | |
ffc41cf8 | 416 | if (unlikely(!sock)) { |
a897b666 | 417 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
418 | "Attempted %s on closed socket in sock_xmit\n", |
419 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
420 | return -EINVAL; |
421 | } | |
422 | ||
c9f2b6ae | 423 | msg.msg_iter = *iter; |
c1696cab | 424 | |
f1083048 | 425 | noreclaim_flag = memalloc_noreclaim_save(); |
1da177e4 | 426 | do { |
7f338fe4 | 427 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
428 | msg.msg_name = NULL; |
429 | msg.msg_namelen = 0; | |
430 | msg.msg_control = NULL; | |
431 | msg.msg_controllen = 0; | |
1da177e4 LT |
432 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
433 | ||
7e2893a1 | 434 | if (send) |
c1696cab | 435 | result = sock_sendmsg(sock, &msg); |
7e2893a1 | 436 | else |
c1696cab | 437 | result = sock_recvmsg(sock, &msg, msg.msg_flags); |
1da177e4 | 438 | |
1da177e4 LT |
439 | if (result <= 0) { |
440 | if (result == 0) | |
441 | result = -EPIPE; /* short read */ | |
442 | break; | |
443 | } | |
9dd5d3ab JB |
444 | if (sent) |
445 | *sent += result; | |
c1696cab | 446 | } while (msg_data_left(&msg)); |
1da177e4 | 447 | |
f1083048 | 448 | memalloc_noreclaim_restore(noreclaim_flag); |
1da177e4 LT |
449 | |
450 | return result; | |
451 | } | |
452 | ||
32e67a3a JB |
453 | /* |
454 | * Different settings for sk->sk_sndtimeo can result in different return values | |
455 | * if there is a signal pending when we enter sendmsg, because reasons? | |
456 | */ | |
457 | static inline int was_interrupted(int result) | |
458 | { | |
459 | return result == -ERESTARTSYS || result == -EINTR; | |
460 | } | |
461 | ||
7fdfd406 | 462 | /* always call with the tx_lock held */ |
9561a7ad | 463 | static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index) |
1da177e4 | 464 | { |
fd8383fd | 465 | struct request *req = blk_mq_rq_from_pdu(cmd); |
5ea8d108 JB |
466 | struct nbd_config *config = nbd->config; |
467 | struct nbd_sock *nsock = config->socks[index]; | |
d61b7f97 | 468 | int result; |
c9f2b6ae AV |
469 | struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)}; |
470 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
471 | struct iov_iter from; | |
1011c1b9 | 472 | unsigned long size = blk_rq_bytes(req); |
429a787b | 473 | struct bio *bio; |
8f3ea359 | 474 | u64 handle; |
9dc6c806 | 475 | u32 type; |
685c9b24 | 476 | u32 nbd_cmd_flags = 0; |
9dd5d3ab | 477 | int sent = nsock->sent, skip = 0; |
9dc6c806 | 478 | |
aa563d7b | 479 | iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request)); |
c9f2b6ae | 480 | |
aebf526b CH |
481 | switch (req_op(req)) { |
482 | case REQ_OP_DISCARD: | |
9dc6c806 | 483 | type = NBD_CMD_TRIM; |
aebf526b CH |
484 | break; |
485 | case REQ_OP_FLUSH: | |
9dc6c806 | 486 | type = NBD_CMD_FLUSH; |
aebf526b CH |
487 | break; |
488 | case REQ_OP_WRITE: | |
9dc6c806 | 489 | type = NBD_CMD_WRITE; |
aebf526b CH |
490 | break; |
491 | case REQ_OP_READ: | |
9dc6c806 | 492 | type = NBD_CMD_READ; |
aebf526b CH |
493 | break; |
494 | default: | |
495 | return -EIO; | |
496 | } | |
1da177e4 | 497 | |
09fc54cc | 498 | if (rq_data_dir(req) == WRITE && |
5ea8d108 | 499 | (config->flags & NBD_FLAG_READ_ONLY)) { |
09fc54cc CH |
500 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
501 | "Write on read-only\n"); | |
502 | return -EIO; | |
503 | } | |
504 | ||
685c9b24 SM |
505 | if (req->cmd_flags & REQ_FUA) |
506 | nbd_cmd_flags |= NBD_CMD_FLAG_FUA; | |
507 | ||
9dd5d3ab JB |
508 | /* We did a partial send previously, and we at least sent the whole |
509 | * request struct, so just go and send the rest of the pages in the | |
510 | * request. | |
511 | */ | |
512 | if (sent) { | |
513 | if (sent >= sizeof(request)) { | |
514 | skip = sent - sizeof(request); | |
2abd2de7 AH |
515 | |
516 | /* initialize handle for tracing purposes */ | |
517 | handle = nbd_cmd_handle(cmd); | |
518 | ||
9dd5d3ab JB |
519 | goto send_pages; |
520 | } | |
521 | iov_iter_advance(&from, sent); | |
8f3ea359 JB |
522 | } else { |
523 | cmd->cmd_cookie++; | |
9dd5d3ab | 524 | } |
f3733247 | 525 | cmd->index = index; |
799f9a38 | 526 | cmd->cookie = nsock->cookie; |
685c9b24 | 527 | request.type = htonl(type | nbd_cmd_flags); |
9561a7ad | 528 | if (type != NBD_CMD_FLUSH) { |
75f187ab AB |
529 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
530 | request.len = htonl(size); | |
531 | } | |
8f3ea359 JB |
532 | handle = nbd_cmd_handle(cmd); |
533 | memcpy(request.handle, &handle, sizeof(handle)); | |
1da177e4 | 534 | |
ea106722 MM |
535 | trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd)); |
536 | ||
d18509f5 | 537 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
ee57a05c | 538 | req, nbdcmd_to_ascii(type), |
d18509f5 | 539 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
c9f2b6ae | 540 | result = sock_xmit(nbd, index, 1, &from, |
9dd5d3ab | 541 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent); |
2abd2de7 | 542 | trace_nbd_header_sent(req, handle); |
1da177e4 | 543 | if (result <= 0) { |
32e67a3a | 544 | if (was_interrupted(result)) { |
9dd5d3ab JB |
545 | /* If we havne't sent anything we can just return BUSY, |
546 | * however if we have sent something we need to make | |
547 | * sure we only allow this req to be sent until we are | |
548 | * completely done. | |
549 | */ | |
550 | if (sent) { | |
551 | nsock->pending = req; | |
552 | nsock->sent = sent; | |
553 | } | |
d7d94d48 | 554 | set_bit(NBD_CMD_REQUEUED, &cmd->flags); |
fc17b653 | 555 | return BLK_STS_RESOURCE; |
9dd5d3ab | 556 | } |
a897b666 | 557 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 | 558 | "Send control failed (result %d)\n", result); |
f3733247 | 559 | return -EAGAIN; |
1da177e4 | 560 | } |
9dd5d3ab | 561 | send_pages: |
429a787b | 562 | if (type != NBD_CMD_WRITE) |
9dd5d3ab | 563 | goto out; |
429a787b | 564 | |
429a787b JA |
565 | bio = req->bio; |
566 | while (bio) { | |
567 | struct bio *next = bio->bi_next; | |
568 | struct bvec_iter iter; | |
7988613b | 569 | struct bio_vec bvec; |
429a787b JA |
570 | |
571 | bio_for_each_segment(bvec, bio, iter) { | |
572 | bool is_last = !next && bio_iter_last(bvec, iter); | |
d61b7f97 | 573 | int flags = is_last ? 0 : MSG_MORE; |
429a787b | 574 | |
d18509f5 | 575 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
ee57a05c | 576 | req, bvec.bv_len); |
aa563d7b | 577 | iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len); |
9dd5d3ab JB |
578 | if (skip) { |
579 | if (skip >= iov_iter_count(&from)) { | |
580 | skip -= iov_iter_count(&from); | |
581 | continue; | |
582 | } | |
583 | iov_iter_advance(&from, skip); | |
584 | skip = 0; | |
585 | } | |
586 | result = sock_xmit(nbd, index, 1, &from, flags, &sent); | |
6c92e699 | 587 | if (result <= 0) { |
32e67a3a | 588 | if (was_interrupted(result)) { |
9dd5d3ab JB |
589 | /* We've already sent the header, we |
590 | * have no choice but to set pending and | |
591 | * return BUSY. | |
592 | */ | |
593 | nsock->pending = req; | |
594 | nsock->sent = sent; | |
d7d94d48 | 595 | set_bit(NBD_CMD_REQUEUED, &cmd->flags); |
fc17b653 | 596 | return BLK_STS_RESOURCE; |
9dd5d3ab | 597 | } |
f4507164 | 598 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
599 | "Send data failed (result %d)\n", |
600 | result); | |
f3733247 | 601 | return -EAGAIN; |
6c92e699 | 602 | } |
429a787b JA |
603 | /* |
604 | * The completion might already have come in, | |
605 | * so break for the last one instead of letting | |
606 | * the iterator do it. This prevents use-after-free | |
607 | * of the bio. | |
608 | */ | |
609 | if (is_last) | |
610 | break; | |
1da177e4 | 611 | } |
429a787b | 612 | bio = next; |
1da177e4 | 613 | } |
9dd5d3ab | 614 | out: |
2abd2de7 | 615 | trace_nbd_payload_sent(req, handle); |
9dd5d3ab JB |
616 | nsock->pending = NULL; |
617 | nsock->sent = 0; | |
1da177e4 | 618 | return 0; |
1da177e4 LT |
619 | } |
620 | ||
1da177e4 | 621 | /* NULL returned = something went wrong, inform userspace */ |
9561a7ad | 622 | static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index) |
1da177e4 | 623 | { |
5ea8d108 | 624 | struct nbd_config *config = nbd->config; |
1da177e4 LT |
625 | int result; |
626 | struct nbd_reply reply; | |
fd8383fd JB |
627 | struct nbd_cmd *cmd; |
628 | struct request *req = NULL; | |
8f3ea359 | 629 | u64 handle; |
fd8383fd | 630 | u16 hwq; |
9561a7ad | 631 | u32 tag; |
c9f2b6ae AV |
632 | struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)}; |
633 | struct iov_iter to; | |
8f3ea359 | 634 | int ret = 0; |
1da177e4 LT |
635 | |
636 | reply.magic = 0; | |
aa563d7b | 637 | iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply)); |
9dd5d3ab | 638 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
1da177e4 | 639 | if (result <= 0) { |
5ea8d108 | 640 | if (!nbd_disconnected(config)) |
9561a7ad JB |
641 | dev_err(disk_to_dev(nbd->disk), |
642 | "Receive control failed (result %d)\n", result); | |
19391830 | 643 | return ERR_PTR(result); |
1da177e4 | 644 | } |
e4b57e08 MF |
645 | |
646 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 647 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 648 | (unsigned long)ntohl(reply.magic)); |
19391830 | 649 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
650 | } |
651 | ||
8f3ea359 JB |
652 | memcpy(&handle, reply.handle, sizeof(handle)); |
653 | tag = nbd_handle_to_tag(handle); | |
fd8383fd JB |
654 | hwq = blk_mq_unique_tag_to_hwq(tag); |
655 | if (hwq < nbd->tag_set.nr_hw_queues) | |
656 | req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq], | |
657 | blk_mq_unique_tag_to_tag(tag)); | |
658 | if (!req || !blk_mq_request_started(req)) { | |
659 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n", | |
660 | tag, req); | |
661 | return ERR_PTR(-ENOENT); | |
1da177e4 | 662 | } |
2abd2de7 | 663 | trace_nbd_header_received(req, handle); |
fd8383fd | 664 | cmd = blk_mq_rq_to_pdu(req); |
8f3ea359 JB |
665 | |
666 | mutex_lock(&cmd->lock); | |
667 | if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) { | |
668 | dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n", | |
669 | req, cmd->cmd_cookie, nbd_handle_to_cookie(handle)); | |
670 | ret = -ENOENT; | |
671 | goto out; | |
672 | } | |
673 | if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) { | |
674 | dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n", | |
675 | req); | |
676 | ret = -ENOENT; | |
677 | goto out; | |
678 | } | |
1da177e4 | 679 | if (ntohl(reply.error)) { |
f4507164 | 680 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 681 | ntohl(reply.error)); |
2a842aca | 682 | cmd->status = BLK_STS_IOERR; |
8f3ea359 | 683 | goto out; |
1da177e4 LT |
684 | } |
685 | ||
ee57a05c | 686 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req); |
9dc6c806 | 687 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 688 | struct req_iterator iter; |
7988613b | 689 | struct bio_vec bvec; |
5705f702 N |
690 | |
691 | rq_for_each_segment(bvec, req, iter) { | |
aa563d7b | 692 | iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len); |
9dd5d3ab | 693 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
6c92e699 | 694 | if (result <= 0) { |
f4507164 | 695 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 696 | result); |
f3733247 JB |
697 | /* |
698 | * If we've disconnected or we only have 1 | |
699 | * connection then we need to make sure we | |
700 | * complete this request, otherwise error out | |
701 | * and let the timeout stuff handle resubmitting | |
702 | * this request onto another connection. | |
703 | */ | |
5ea8d108 JB |
704 | if (nbd_disconnected(config) || |
705 | config->num_connections <= 1) { | |
2a842aca | 706 | cmd->status = BLK_STS_IOERR; |
8f3ea359 | 707 | goto out; |
f3733247 | 708 | } |
8f3ea359 JB |
709 | ret = -EIO; |
710 | goto out; | |
6c92e699 | 711 | } |
d18509f5 | 712 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
ee57a05c | 713 | req, bvec.bv_len); |
1da177e4 LT |
714 | } |
715 | } | |
8f3ea359 | 716 | out: |
2abd2de7 | 717 | trace_nbd_payload_received(req, handle); |
8f3ea359 JB |
718 | mutex_unlock(&cmd->lock); |
719 | return ret ? ERR_PTR(ret) : cmd; | |
1da177e4 LT |
720 | } |
721 | ||
9561a7ad | 722 | static void recv_work(struct work_struct *work) |
1da177e4 | 723 | { |
9561a7ad JB |
724 | struct recv_thread_args *args = container_of(work, |
725 | struct recv_thread_args, | |
726 | work); | |
727 | struct nbd_device *nbd = args->nbd; | |
5ea8d108 | 728 | struct nbd_config *config = nbd->config; |
fd8383fd | 729 | struct nbd_cmd *cmd; |
1da177e4 | 730 | |
19391830 | 731 | while (1) { |
9561a7ad | 732 | cmd = nbd_read_stat(nbd, args->index); |
fd8383fd | 733 | if (IS_ERR(cmd)) { |
5ea8d108 | 734 | struct nbd_sock *nsock = config->socks[args->index]; |
f3733247 JB |
735 | |
736 | mutex_lock(&nsock->tx_lock); | |
799f9a38 | 737 | nbd_mark_nsock_dead(nbd, nsock, 1); |
f3733247 | 738 | mutex_unlock(&nsock->tx_lock); |
19391830 MP |
739 | break; |
740 | } | |
741 | ||
08e0029a | 742 | blk_mq_complete_request(blk_mq_rq_from_pdu(cmd)); |
19391830 | 743 | } |
5ea8d108 JB |
744 | atomic_dec(&config->recv_threads); |
745 | wake_up(&config->recv_wq); | |
746 | nbd_config_put(nbd); | |
747 | kfree(args); | |
1da177e4 LT |
748 | } |
749 | ||
7baa8572 | 750 | static bool nbd_clear_req(struct request *req, void *data, bool reserved) |
1da177e4 | 751 | { |
d250bf4e | 752 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
1da177e4 | 753 | |
2a842aca | 754 | cmd->status = BLK_STS_IOERR; |
08e0029a | 755 | blk_mq_complete_request(req); |
7baa8572 | 756 | return true; |
fd8383fd JB |
757 | } |
758 | ||
759 | static void nbd_clear_que(struct nbd_device *nbd) | |
760 | { | |
b52c2e92 | 761 | blk_mq_quiesce_queue(nbd->disk->queue); |
fd8383fd | 762 | blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL); |
b52c2e92 | 763 | blk_mq_unquiesce_queue(nbd->disk->queue); |
e78273c8 | 764 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
765 | } |
766 | ||
f3733247 JB |
767 | static int find_fallback(struct nbd_device *nbd, int index) |
768 | { | |
5ea8d108 | 769 | struct nbd_config *config = nbd->config; |
f3733247 | 770 | int new_index = -1; |
5ea8d108 | 771 | struct nbd_sock *nsock = config->socks[index]; |
f3733247 JB |
772 | int fallback = nsock->fallback_index; |
773 | ||
5ea8d108 | 774 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
f3733247 JB |
775 | return new_index; |
776 | ||
5ea8d108 | 777 | if (config->num_connections <= 1) { |
f3733247 JB |
778 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
779 | "Attempted send on invalid socket\n"); | |
780 | return new_index; | |
781 | } | |
782 | ||
5ea8d108 JB |
783 | if (fallback >= 0 && fallback < config->num_connections && |
784 | !config->socks[fallback]->dead) | |
f3733247 JB |
785 | return fallback; |
786 | ||
787 | if (nsock->fallback_index < 0 || | |
5ea8d108 JB |
788 | nsock->fallback_index >= config->num_connections || |
789 | config->socks[nsock->fallback_index]->dead) { | |
f3733247 | 790 | int i; |
5ea8d108 | 791 | for (i = 0; i < config->num_connections; i++) { |
f3733247 JB |
792 | if (i == index) |
793 | continue; | |
5ea8d108 | 794 | if (!config->socks[i]->dead) { |
f3733247 JB |
795 | new_index = i; |
796 | break; | |
797 | } | |
798 | } | |
799 | nsock->fallback_index = new_index; | |
800 | if (new_index < 0) { | |
801 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
802 | "Dead connection, failed to find a fallback\n"); | |
803 | return new_index; | |
804 | } | |
805 | } | |
806 | new_index = nsock->fallback_index; | |
807 | return new_index; | |
808 | } | |
7fdfd406 | 809 | |
560bc4b3 JB |
810 | static int wait_for_reconnect(struct nbd_device *nbd) |
811 | { | |
812 | struct nbd_config *config = nbd->config; | |
813 | if (!config->dead_conn_timeout) | |
814 | return 0; | |
815 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) | |
816 | return 0; | |
5e3c3a7e KV |
817 | return wait_event_timeout(config->conn_wait, |
818 | atomic_read(&config->live_connections) > 0, | |
819 | config->dead_conn_timeout) > 0; | |
560bc4b3 JB |
820 | } |
821 | ||
9dd5d3ab | 822 | static int nbd_handle_cmd(struct nbd_cmd *cmd, int index) |
48cf6061 | 823 | { |
fd8383fd JB |
824 | struct request *req = blk_mq_rq_from_pdu(cmd); |
825 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 | 826 | struct nbd_config *config; |
9561a7ad | 827 | struct nbd_sock *nsock; |
9dd5d3ab | 828 | int ret; |
fd8383fd | 829 | |
5ea8d108 JB |
830 | if (!refcount_inc_not_zero(&nbd->config_refs)) { |
831 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
832 | "Socks array is empty\n"); | |
6a468d59 | 833 | blk_mq_start_request(req); |
5ea8d108 JB |
834 | return -EINVAL; |
835 | } | |
836 | config = nbd->config; | |
837 | ||
838 | if (index >= config->num_connections) { | |
a897b666 JB |
839 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
840 | "Attempted send on invalid socket\n"); | |
5ea8d108 | 841 | nbd_config_put(nbd); |
6a468d59 | 842 | blk_mq_start_request(req); |
9dd5d3ab | 843 | return -EINVAL; |
9561a7ad | 844 | } |
2a842aca | 845 | cmd->status = BLK_STS_OK; |
f3733247 | 846 | again: |
5ea8d108 | 847 | nsock = config->socks[index]; |
9561a7ad | 848 | mutex_lock(&nsock->tx_lock); |
f3733247 | 849 | if (nsock->dead) { |
560bc4b3 | 850 | int old_index = index; |
f3733247 | 851 | index = find_fallback(nbd, index); |
560bc4b3 | 852 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 853 | if (index < 0) { |
560bc4b3 JB |
854 | if (wait_for_reconnect(nbd)) { |
855 | index = old_index; | |
856 | goto again; | |
857 | } | |
858 | /* All the sockets should already be down at this point, | |
859 | * we just want to make sure that DISCONNECTED is set so | |
860 | * any requests that come in that were queue'ed waiting | |
861 | * for the reconnect timer don't trigger the timer again | |
862 | * and instead just error out. | |
863 | */ | |
864 | sock_shutdown(nbd); | |
865 | nbd_config_put(nbd); | |
6a468d59 | 866 | blk_mq_start_request(req); |
560bc4b3 | 867 | return -EIO; |
5ea8d108 | 868 | } |
f3733247 | 869 | goto again; |
48cf6061 LV |
870 | } |
871 | ||
9dd5d3ab JB |
872 | /* Handle the case that we have a pending request that was partially |
873 | * transmitted that _has_ to be serviced first. We need to call requeue | |
874 | * here so that it gets put _after_ the request that is already on the | |
875 | * dispatch list. | |
876 | */ | |
6a468d59 | 877 | blk_mq_start_request(req); |
9dd5d3ab | 878 | if (unlikely(nsock->pending && nsock->pending != req)) { |
d7d94d48 | 879 | nbd_requeue_cmd(cmd); |
9dd5d3ab JB |
880 | ret = 0; |
881 | goto out; | |
48cf6061 | 882 | } |
f3733247 JB |
883 | /* |
884 | * Some failures are related to the link going down, so anything that | |
885 | * returns EAGAIN can be retried on a different socket. | |
886 | */ | |
9dd5d3ab | 887 | ret = nbd_send_cmd(nbd, cmd, index); |
f3733247 JB |
888 | if (ret == -EAGAIN) { |
889 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
6a468d59 | 890 | "Request send failed, requeueing\n"); |
799f9a38 | 891 | nbd_mark_nsock_dead(nbd, nsock, 1); |
d7d94d48 | 892 | nbd_requeue_cmd(cmd); |
6a468d59 | 893 | ret = 0; |
f3733247 | 894 | } |
9dd5d3ab | 895 | out: |
9561a7ad | 896 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 897 | nbd_config_put(nbd); |
9dd5d3ab | 898 | return ret; |
48cf6061 LV |
899 | } |
900 | ||
fc17b653 | 901 | static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
fd8383fd | 902 | const struct blk_mq_queue_data *bd) |
1da177e4 | 903 | { |
fd8383fd | 904 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); |
9dd5d3ab | 905 | int ret; |
1da177e4 | 906 | |
9561a7ad JB |
907 | /* |
908 | * Since we look at the bio's to send the request over the network we | |
909 | * need to make sure the completion work doesn't mark this request done | |
910 | * before we are done doing our send. This keeps us from dereferencing | |
911 | * freed data if we have particularly fast completions (ie we get the | |
912 | * completion before we exit sock_xmit on the last bvec) or in the case | |
913 | * that the server is misbehaving (or there was an error) before we're | |
914 | * done sending everything over the wire. | |
915 | */ | |
8f3ea359 | 916 | mutex_lock(&cmd->lock); |
d7d94d48 | 917 | clear_bit(NBD_CMD_REQUEUED, &cmd->flags); |
9dd5d3ab JB |
918 | |
919 | /* We can be called directly from the user space process, which means we | |
920 | * could possibly have signals pending so our sendmsg will fail. In | |
921 | * this case we need to return that we are busy, otherwise error out as | |
922 | * appropriate. | |
923 | */ | |
924 | ret = nbd_handle_cmd(cmd, hctx->queue_num); | |
6e60a3bb JB |
925 | if (ret < 0) |
926 | ret = BLK_STS_IOERR; | |
927 | else if (!ret) | |
928 | ret = BLK_STS_OK; | |
8f3ea359 | 929 | mutex_unlock(&cmd->lock); |
9561a7ad | 930 | |
6e60a3bb | 931 | return ret; |
1da177e4 LT |
932 | } |
933 | ||
e46c7287 JB |
934 | static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg, |
935 | bool netlink) | |
23272a67 | 936 | { |
5ea8d108 | 937 | struct nbd_config *config = nbd->config; |
9442b739 | 938 | struct socket *sock; |
9561a7ad JB |
939 | struct nbd_sock **socks; |
940 | struct nbd_sock *nsock; | |
9442b739 JB |
941 | int err; |
942 | ||
943 | sock = sockfd_lookup(arg, &err); | |
944 | if (!sock) | |
945 | return err; | |
23272a67 | 946 | |
e46c7287 JB |
947 | if (!netlink && !nbd->task_setup && |
948 | !test_bit(NBD_BOUND, &config->runtime_flags)) | |
9561a7ad | 949 | nbd->task_setup = current; |
e46c7287 JB |
950 | |
951 | if (!netlink && | |
952 | (nbd->task_setup != current || | |
953 | test_bit(NBD_BOUND, &config->runtime_flags))) { | |
9561a7ad JB |
954 | dev_err(disk_to_dev(nbd->disk), |
955 | "Device being setup by another task"); | |
9b1355d5 | 956 | sockfd_put(sock); |
e46c7287 | 957 | return -EBUSY; |
23272a67 MP |
958 | } |
959 | ||
5ea8d108 | 960 | socks = krealloc(config->socks, (config->num_connections + 1) * |
9561a7ad | 961 | sizeof(struct nbd_sock *), GFP_KERNEL); |
9b1355d5 JB |
962 | if (!socks) { |
963 | sockfd_put(sock); | |
9561a7ad | 964 | return -ENOMEM; |
9b1355d5 | 965 | } |
9561a7ad | 966 | nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL); |
9b1355d5 JB |
967 | if (!nsock) { |
968 | sockfd_put(sock); | |
9561a7ad | 969 | return -ENOMEM; |
9b1355d5 | 970 | } |
9561a7ad | 971 | |
5ea8d108 | 972 | config->socks = socks; |
23272a67 | 973 | |
f3733247 JB |
974 | nsock->fallback_index = -1; |
975 | nsock->dead = false; | |
9561a7ad JB |
976 | mutex_init(&nsock->tx_lock); |
977 | nsock->sock = sock; | |
9dd5d3ab JB |
978 | nsock->pending = NULL; |
979 | nsock->sent = 0; | |
799f9a38 | 980 | nsock->cookie = 0; |
5ea8d108 | 981 | socks[config->num_connections++] = nsock; |
560bc4b3 | 982 | atomic_inc(&config->live_connections); |
23272a67 | 983 | |
9561a7ad | 984 | return 0; |
23272a67 MP |
985 | } |
986 | ||
b7aa3d39 JB |
987 | static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg) |
988 | { | |
989 | struct nbd_config *config = nbd->config; | |
990 | struct socket *sock, *old; | |
991 | struct recv_thread_args *args; | |
992 | int i; | |
993 | int err; | |
994 | ||
995 | sock = sockfd_lookup(arg, &err); | |
996 | if (!sock) | |
997 | return err; | |
998 | ||
999 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
1000 | if (!args) { | |
1001 | sockfd_put(sock); | |
1002 | return -ENOMEM; | |
1003 | } | |
1004 | ||
1005 | for (i = 0; i < config->num_connections; i++) { | |
1006 | struct nbd_sock *nsock = config->socks[i]; | |
1007 | ||
1008 | if (!nsock->dead) | |
1009 | continue; | |
1010 | ||
1011 | mutex_lock(&nsock->tx_lock); | |
1012 | if (!nsock->dead) { | |
1013 | mutex_unlock(&nsock->tx_lock); | |
1014 | continue; | |
1015 | } | |
1016 | sk_set_memalloc(sock->sk); | |
a7ee8cf1 JB |
1017 | if (nbd->tag_set.timeout) |
1018 | sock->sk->sk_sndtimeo = nbd->tag_set.timeout; | |
b7aa3d39 JB |
1019 | atomic_inc(&config->recv_threads); |
1020 | refcount_inc(&nbd->config_refs); | |
1021 | old = nsock->sock; | |
1022 | nsock->fallback_index = -1; | |
1023 | nsock->sock = sock; | |
1024 | nsock->dead = false; | |
1025 | INIT_WORK(&args->work, recv_work); | |
1026 | args->index = i; | |
1027 | args->nbd = nbd; | |
799f9a38 | 1028 | nsock->cookie++; |
b7aa3d39 JB |
1029 | mutex_unlock(&nsock->tx_lock); |
1030 | sockfd_put(old); | |
1031 | ||
7a362ea9 JB |
1032 | clear_bit(NBD_DISCONNECTED, &config->runtime_flags); |
1033 | ||
b7aa3d39 JB |
1034 | /* We take the tx_mutex in an error path in the recv_work, so we |
1035 | * need to queue_work outside of the tx_mutex. | |
1036 | */ | |
1037 | queue_work(recv_workqueue, &args->work); | |
560bc4b3 JB |
1038 | |
1039 | atomic_inc(&config->live_connections); | |
1040 | wake_up(&config->conn_wait); | |
b7aa3d39 JB |
1041 | return 0; |
1042 | } | |
1043 | sockfd_put(sock); | |
1044 | kfree(args); | |
1045 | return -ENOSPC; | |
1046 | } | |
1047 | ||
0e4f0f6f MP |
1048 | static void nbd_bdev_reset(struct block_device *bdev) |
1049 | { | |
abbbdf12 RMB |
1050 | if (bdev->bd_openers > 1) |
1051 | return; | |
29eaadc0 | 1052 | bd_set_size(bdev, 0); |
0e4f0f6f MP |
1053 | } |
1054 | ||
29eaadc0 | 1055 | static void nbd_parse_flags(struct nbd_device *nbd) |
d02cf531 | 1056 | { |
5ea8d108 JB |
1057 | struct nbd_config *config = nbd->config; |
1058 | if (config->flags & NBD_FLAG_READ_ONLY) | |
29eaadc0 JB |
1059 | set_disk_ro(nbd->disk, true); |
1060 | else | |
1061 | set_disk_ro(nbd->disk, false); | |
5ea8d108 | 1062 | if (config->flags & NBD_FLAG_SEND_TRIM) |
8b904b5b | 1063 | blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
685c9b24 SM |
1064 | if (config->flags & NBD_FLAG_SEND_FLUSH) { |
1065 | if (config->flags & NBD_FLAG_SEND_FUA) | |
1066 | blk_queue_write_cache(nbd->disk->queue, true, true); | |
1067 | else | |
1068 | blk_queue_write_cache(nbd->disk->queue, true, false); | |
1069 | } | |
d02cf531 | 1070 | else |
aafb1eec | 1071 | blk_queue_write_cache(nbd->disk->queue, false, false); |
d02cf531 MP |
1072 | } |
1073 | ||
9561a7ad JB |
1074 | static void send_disconnects(struct nbd_device *nbd) |
1075 | { | |
5ea8d108 | 1076 | struct nbd_config *config = nbd->config; |
c9f2b6ae AV |
1077 | struct nbd_request request = { |
1078 | .magic = htonl(NBD_REQUEST_MAGIC), | |
1079 | .type = htonl(NBD_CMD_DISC), | |
1080 | }; | |
1081 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
1082 | struct iov_iter from; | |
9561a7ad JB |
1083 | int i, ret; |
1084 | ||
5ea8d108 | 1085 | for (i = 0; i < config->num_connections; i++) { |
b4b2aecc JB |
1086 | struct nbd_sock *nsock = config->socks[i]; |
1087 | ||
aa563d7b | 1088 | iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request)); |
b4b2aecc | 1089 | mutex_lock(&nsock->tx_lock); |
9dd5d3ab | 1090 | ret = sock_xmit(nbd, i, 1, &from, 0, NULL); |
9561a7ad JB |
1091 | if (ret <= 0) |
1092 | dev_err(disk_to_dev(nbd->disk), | |
1093 | "Send disconnect failed %d\n", ret); | |
b4b2aecc | 1094 | mutex_unlock(&nsock->tx_lock); |
9561a7ad JB |
1095 | } |
1096 | } | |
1097 | ||
29eaadc0 | 1098 | static int nbd_disconnect(struct nbd_device *nbd) |
9442b739 | 1099 | { |
5ea8d108 | 1100 | struct nbd_config *config = nbd->config; |
30d53d9c | 1101 | |
5ea8d108 | 1102 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
2e13456f JB |
1103 | set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); |
1104 | send_disconnects(nbd); | |
9442b739 JB |
1105 | return 0; |
1106 | } | |
1107 | ||
29eaadc0 | 1108 | static void nbd_clear_sock(struct nbd_device *nbd) |
1a2ad211 | 1109 | { |
9442b739 JB |
1110 | sock_shutdown(nbd); |
1111 | nbd_clear_que(nbd); | |
5ea8d108 | 1112 | nbd->task_setup = NULL; |
5ea8d108 JB |
1113 | } |
1114 | ||
1115 | static void nbd_config_put(struct nbd_device *nbd) | |
1116 | { | |
1117 | if (refcount_dec_and_mutex_lock(&nbd->config_refs, | |
1118 | &nbd->config_lock)) { | |
5ea8d108 | 1119 | struct nbd_config *config = nbd->config; |
5ea8d108 | 1120 | nbd_dev_dbg_close(nbd); |
29eaadc0 | 1121 | nbd_size_clear(nbd); |
5ea8d108 JB |
1122 | if (test_and_clear_bit(NBD_HAS_PID_FILE, |
1123 | &config->runtime_flags)) | |
1124 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); | |
1125 | nbd->task_recv = NULL; | |
29eaadc0 | 1126 | nbd_clear_sock(nbd); |
5ea8d108 JB |
1127 | if (config->num_connections) { |
1128 | int i; | |
1129 | for (i = 0; i < config->num_connections; i++) { | |
1130 | sockfd_put(config->socks[i]->sock); | |
1131 | kfree(config->socks[i]); | |
1132 | } | |
1133 | kfree(config->socks); | |
1134 | } | |
fa976532 | 1135 | kfree(nbd->config); |
af622b86 ID |
1136 | nbd->config = NULL; |
1137 | ||
1138 | nbd->tag_set.timeout = 0; | |
6df133a1 | 1139 | nbd->disk->queue->limits.discard_granularity = 0; |
07ce213f | 1140 | nbd->disk->queue->limits.discard_alignment = 0; |
6df133a1 | 1141 | blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX); |
8b904b5b | 1142 | blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
a2c97909 | 1143 | |
5ea8d108 | 1144 | mutex_unlock(&nbd->config_lock); |
c6a4759e | 1145 | nbd_put(nbd); |
5ea8d108 JB |
1146 | module_put(THIS_MODULE); |
1147 | } | |
9442b739 JB |
1148 | } |
1149 | ||
e46c7287 | 1150 | static int nbd_start_device(struct nbd_device *nbd) |
9442b739 | 1151 | { |
5ea8d108 JB |
1152 | struct nbd_config *config = nbd->config; |
1153 | int num_connections = config->num_connections; | |
9442b739 | 1154 | int error = 0, i; |
1a2ad211 | 1155 | |
9442b739 JB |
1156 | if (nbd->task_recv) |
1157 | return -EBUSY; | |
5ea8d108 | 1158 | if (!config->socks) |
9442b739 JB |
1159 | return -EINVAL; |
1160 | if (num_connections > 1 && | |
5ea8d108 | 1161 | !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) { |
9442b739 | 1162 | dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n"); |
5ea8d108 | 1163 | return -EINVAL; |
9442b739 | 1164 | } |
23272a67 | 1165 | |
5ea8d108 | 1166 | blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections); |
9442b739 | 1167 | nbd->task_recv = current; |
23272a67 | 1168 | |
29eaadc0 | 1169 | nbd_parse_flags(nbd); |
23272a67 | 1170 | |
9442b739 JB |
1171 | error = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
1172 | if (error) { | |
1173 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); | |
5ea8d108 | 1174 | return error; |
1a2ad211 | 1175 | } |
29eaadc0 | 1176 | set_bit(NBD_HAS_PID_FILE, &config->runtime_flags); |
37091fdd | 1177 | |
9442b739 JB |
1178 | nbd_dev_dbg_init(nbd); |
1179 | for (i = 0; i < num_connections; i++) { | |
5ea8d108 JB |
1180 | struct recv_thread_args *args; |
1181 | ||
1182 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
1183 | if (!args) { | |
1184 | sock_shutdown(nbd); | |
1185 | return -ENOMEM; | |
1186 | } | |
1187 | sk_set_memalloc(config->socks[i]->sock->sk); | |
a7ee8cf1 JB |
1188 | if (nbd->tag_set.timeout) |
1189 | config->socks[i]->sock->sk->sk_sndtimeo = | |
1190 | nbd->tag_set.timeout; | |
5ea8d108 JB |
1191 | atomic_inc(&config->recv_threads); |
1192 | refcount_inc(&nbd->config_refs); | |
1193 | INIT_WORK(&args->work, recv_work); | |
1194 | args->nbd = nbd; | |
1195 | args->index = i; | |
1196 | queue_work(recv_workqueue, &args->work); | |
37091fdd | 1197 | } |
639812a1 | 1198 | nbd_size_update(nbd); |
e46c7287 JB |
1199 | return error; |
1200 | } | |
1201 | ||
1202 | static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev) | |
1203 | { | |
1204 | struct nbd_config *config = nbd->config; | |
1205 | int ret; | |
1206 | ||
1207 | ret = nbd_start_device(nbd); | |
1208 | if (ret) | |
1209 | return ret; | |
1210 | ||
e46c7287 JB |
1211 | if (max_part) |
1212 | bdev->bd_invalidated = 1; | |
1213 | mutex_unlock(&nbd->config_lock); | |
1214 | ret = wait_event_interruptible(config->recv_wq, | |
5ea8d108 | 1215 | atomic_read(&config->recv_threads) == 0); |
e46c7287 | 1216 | if (ret) |
5ea8d108 | 1217 | sock_shutdown(nbd); |
9442b739 | 1218 | mutex_lock(&nbd->config_lock); |
76aa1d34 | 1219 | nbd_bdev_reset(bdev); |
9442b739 | 1220 | /* user requested, ignore socket errors */ |
5ea8d108 | 1221 | if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags)) |
e46c7287 | 1222 | ret = 0; |
5ea8d108 | 1223 | if (test_bit(NBD_TIMEDOUT, &config->runtime_flags)) |
e46c7287 JB |
1224 | ret = -ETIMEDOUT; |
1225 | return ret; | |
9442b739 JB |
1226 | } |
1227 | ||
29eaadc0 JB |
1228 | static void nbd_clear_sock_ioctl(struct nbd_device *nbd, |
1229 | struct block_device *bdev) | |
1230 | { | |
2516ab15 | 1231 | sock_shutdown(nbd); |
29eaadc0 JB |
1232 | kill_bdev(bdev); |
1233 | nbd_bdev_reset(bdev); | |
e46c7287 JB |
1234 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, |
1235 | &nbd->config->runtime_flags)) | |
1236 | nbd_config_put(nbd); | |
29eaadc0 JB |
1237 | } |
1238 | ||
9442b739 JB |
1239 | /* Must be called with config_lock held */ |
1240 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, | |
1241 | unsigned int cmd, unsigned long arg) | |
1242 | { | |
5ea8d108 JB |
1243 | struct nbd_config *config = nbd->config; |
1244 | ||
9442b739 JB |
1245 | switch (cmd) { |
1246 | case NBD_DISCONNECT: | |
29eaadc0 | 1247 | return nbd_disconnect(nbd); |
9442b739 | 1248 | case NBD_CLEAR_SOCK: |
29eaadc0 JB |
1249 | nbd_clear_sock_ioctl(nbd, bdev); |
1250 | return 0; | |
9442b739 | 1251 | case NBD_SET_SOCK: |
e46c7287 | 1252 | return nbd_add_socket(nbd, arg, false); |
9442b739 | 1253 | case NBD_SET_BLKSIZE: |
bc811f05 JA |
1254 | if (!arg || !is_power_of_2(arg) || arg < 512 || |
1255 | arg > PAGE_SIZE) | |
1256 | return -EINVAL; | |
29eaadc0 | 1257 | nbd_size_set(nbd, arg, |
5ea8d108 | 1258 | div_s64(config->bytesize, arg)); |
e544541b | 1259 | return 0; |
1da177e4 | 1260 | case NBD_SET_SIZE: |
29eaadc0 | 1261 | nbd_size_set(nbd, config->blksize, |
5ea8d108 | 1262 | div_s64(arg, config->blksize)); |
e544541b | 1263 | return 0; |
37091fdd | 1264 | case NBD_SET_SIZE_BLOCKS: |
29eaadc0 | 1265 | nbd_size_set(nbd, config->blksize, arg); |
e544541b | 1266 | return 0; |
7fdfd406 | 1267 | case NBD_SET_TIMEOUT: |
f8586855 JB |
1268 | if (arg) { |
1269 | nbd->tag_set.timeout = arg * HZ; | |
1270 | blk_queue_rq_timeout(nbd->disk->queue, arg * HZ); | |
1271 | } | |
7fdfd406 | 1272 | return 0; |
1a2ad211 | 1273 | |
2f012508 | 1274 | case NBD_SET_FLAGS: |
5ea8d108 | 1275 | config->flags = arg; |
2f012508 | 1276 | return 0; |
9442b739 | 1277 | case NBD_DO_IT: |
e46c7287 | 1278 | return nbd_start_device_ioctl(nbd, bdev); |
1da177e4 | 1279 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
1280 | /* |
1281 | * This is for compatibility only. The queue is always cleared | |
1282 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
1283 | */ | |
1da177e4 LT |
1284 | return 0; |
1285 | case NBD_PRINT_DEBUG: | |
fd8383fd JB |
1286 | /* |
1287 | * For compatibility only, we no longer keep a list of | |
1288 | * outstanding requests. | |
1289 | */ | |
1da177e4 LT |
1290 | return 0; |
1291 | } | |
1a2ad211 PM |
1292 | return -ENOTTY; |
1293 | } | |
1294 | ||
1295 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
1296 | unsigned int cmd, unsigned long arg) | |
1297 | { | |
f4507164 | 1298 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
e46c7287 JB |
1299 | struct nbd_config *config = nbd->config; |
1300 | int error = -EINVAL; | |
1a2ad211 PM |
1301 | |
1302 | if (!capable(CAP_SYS_ADMIN)) | |
1303 | return -EPERM; | |
1304 | ||
1dae69be JB |
1305 | /* The block layer will pass back some non-nbd ioctls in case we have |
1306 | * special handling for them, but we don't so just return an error. | |
1307 | */ | |
1308 | if (_IOC_TYPE(cmd) != 0xab) | |
1309 | return -EINVAL; | |
1310 | ||
9561a7ad | 1311 | mutex_lock(&nbd->config_lock); |
e46c7287 JB |
1312 | |
1313 | /* Don't allow ioctl operations on a nbd device that was created with | |
1314 | * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine. | |
1315 | */ | |
1316 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1317 | (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK)) | |
1318 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
1319 | else | |
1320 | dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n"); | |
9561a7ad | 1321 | mutex_unlock(&nbd->config_lock); |
1a2ad211 | 1322 | return error; |
1da177e4 LT |
1323 | } |
1324 | ||
5ea8d108 JB |
1325 | static struct nbd_config *nbd_alloc_config(void) |
1326 | { | |
1327 | struct nbd_config *config; | |
1328 | ||
1329 | config = kzalloc(sizeof(struct nbd_config), GFP_NOFS); | |
1330 | if (!config) | |
1331 | return NULL; | |
1332 | atomic_set(&config->recv_threads, 0); | |
1333 | init_waitqueue_head(&config->recv_wq); | |
560bc4b3 | 1334 | init_waitqueue_head(&config->conn_wait); |
5ea8d108 | 1335 | config->blksize = 1024; |
560bc4b3 | 1336 | atomic_set(&config->live_connections, 0); |
5ea8d108 JB |
1337 | try_module_get(THIS_MODULE); |
1338 | return config; | |
1339 | } | |
1340 | ||
1341 | static int nbd_open(struct block_device *bdev, fmode_t mode) | |
1342 | { | |
1343 | struct nbd_device *nbd; | |
1344 | int ret = 0; | |
1345 | ||
1346 | mutex_lock(&nbd_index_mutex); | |
1347 | nbd = bdev->bd_disk->private_data; | |
1348 | if (!nbd) { | |
1349 | ret = -ENXIO; | |
1350 | goto out; | |
1351 | } | |
c6a4759e JB |
1352 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1353 | ret = -ENXIO; | |
1354 | goto out; | |
1355 | } | |
5ea8d108 JB |
1356 | if (!refcount_inc_not_zero(&nbd->config_refs)) { |
1357 | struct nbd_config *config; | |
1358 | ||
1359 | mutex_lock(&nbd->config_lock); | |
1360 | if (refcount_inc_not_zero(&nbd->config_refs)) { | |
1361 | mutex_unlock(&nbd->config_lock); | |
1362 | goto out; | |
1363 | } | |
1364 | config = nbd->config = nbd_alloc_config(); | |
1365 | if (!config) { | |
1366 | ret = -ENOMEM; | |
1367 | mutex_unlock(&nbd->config_lock); | |
1368 | goto out; | |
1369 | } | |
1370 | refcount_set(&nbd->config_refs, 1); | |
c6a4759e | 1371 | refcount_inc(&nbd->refs); |
5ea8d108 | 1372 | mutex_unlock(&nbd->config_lock); |
fe1f9e66 JB |
1373 | bdev->bd_invalidated = 1; |
1374 | } else if (nbd_disconnected(nbd->config)) { | |
1375 | bdev->bd_invalidated = 1; | |
5ea8d108 JB |
1376 | } |
1377 | out: | |
1378 | mutex_unlock(&nbd_index_mutex); | |
1379 | return ret; | |
1380 | } | |
1381 | ||
1382 | static void nbd_release(struct gendisk *disk, fmode_t mode) | |
1383 | { | |
1384 | struct nbd_device *nbd = disk->private_data; | |
08ba91ee DRK |
1385 | struct block_device *bdev = bdget_disk(disk, 0); |
1386 | ||
1387 | if (test_bit(NBD_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) && | |
1388 | bdev->bd_openers == 0) | |
1389 | nbd_disconnect_and_put(nbd); | |
1390 | ||
5ea8d108 | 1391 | nbd_config_put(nbd); |
c6a4759e | 1392 | nbd_put(nbd); |
5ea8d108 JB |
1393 | } |
1394 | ||
83d5cde4 | 1395 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
1396 | { |
1397 | .owner = THIS_MODULE, | |
5ea8d108 JB |
1398 | .open = nbd_open, |
1399 | .release = nbd_release, | |
8a6cfeb6 | 1400 | .ioctl = nbd_ioctl, |
263a3df1 | 1401 | .compat_ioctl = nbd_ioctl, |
1da177e4 LT |
1402 | }; |
1403 | ||
30d53d9c MP |
1404 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
1405 | ||
1406 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
1407 | { | |
1408 | struct nbd_device *nbd = s->private; | |
1409 | ||
1410 | if (nbd->task_recv) | |
1411 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
30d53d9c MP |
1412 | |
1413 | return 0; | |
1414 | } | |
1415 | ||
1416 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
1417 | { | |
1418 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
1419 | } | |
1420 | ||
1421 | static const struct file_operations nbd_dbg_tasks_ops = { | |
1422 | .open = nbd_dbg_tasks_open, | |
1423 | .read = seq_read, | |
1424 | .llseek = seq_lseek, | |
1425 | .release = single_release, | |
1426 | }; | |
1427 | ||
1428 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
1429 | { | |
1430 | struct nbd_device *nbd = s->private; | |
5ea8d108 | 1431 | u32 flags = nbd->config->flags; |
30d53d9c MP |
1432 | |
1433 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
1434 | ||
1435 | seq_puts(s, "Known flags:\n"); | |
1436 | ||
1437 | if (flags & NBD_FLAG_HAS_FLAGS) | |
1438 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
1439 | if (flags & NBD_FLAG_READ_ONLY) | |
1440 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
1441 | if (flags & NBD_FLAG_SEND_FLUSH) | |
1442 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
685c9b24 SM |
1443 | if (flags & NBD_FLAG_SEND_FUA) |
1444 | seq_puts(s, "NBD_FLAG_SEND_FUA\n"); | |
30d53d9c MP |
1445 | if (flags & NBD_FLAG_SEND_TRIM) |
1446 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
1447 | ||
1448 | return 0; | |
1449 | } | |
1450 | ||
1451 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
1452 | { | |
1453 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
1454 | } | |
1455 | ||
1456 | static const struct file_operations nbd_dbg_flags_ops = { | |
1457 | .open = nbd_dbg_flags_open, | |
1458 | .read = seq_read, | |
1459 | .llseek = seq_lseek, | |
1460 | .release = single_release, | |
1461 | }; | |
1462 | ||
1463 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1464 | { | |
1465 | struct dentry *dir; | |
5ea8d108 | 1466 | struct nbd_config *config = nbd->config; |
27ea43fe MP |
1467 | |
1468 | if (!nbd_dbg_dir) | |
1469 | return -EIO; | |
30d53d9c MP |
1470 | |
1471 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
27ea43fe MP |
1472 | if (!dir) { |
1473 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n", | |
1474 | nbd_name(nbd)); | |
1475 | return -EIO; | |
30d53d9c | 1476 | } |
5ea8d108 | 1477 | config->dbg_dir = dir; |
30d53d9c | 1478 | |
27ea43fe | 1479 | debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); |
5ea8d108 | 1480 | debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize); |
0eadf37a | 1481 | debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout); |
5ea8d108 | 1482 | debugfs_create_u64("blocksize", 0444, dir, &config->blksize); |
d366a0ff | 1483 | debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops); |
30d53d9c MP |
1484 | |
1485 | return 0; | |
1486 | } | |
1487 | ||
1488 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1489 | { | |
5ea8d108 | 1490 | debugfs_remove_recursive(nbd->config->dbg_dir); |
30d53d9c MP |
1491 | } |
1492 | ||
1493 | static int nbd_dbg_init(void) | |
1494 | { | |
1495 | struct dentry *dbg_dir; | |
1496 | ||
1497 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
27ea43fe MP |
1498 | if (!dbg_dir) |
1499 | return -EIO; | |
30d53d9c MP |
1500 | |
1501 | nbd_dbg_dir = dbg_dir; | |
1502 | ||
1503 | return 0; | |
1504 | } | |
1505 | ||
1506 | static void nbd_dbg_close(void) | |
1507 | { | |
1508 | debugfs_remove_recursive(nbd_dbg_dir); | |
1509 | } | |
1510 | ||
1511 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
1512 | ||
1513 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1514 | { | |
1515 | return 0; | |
1516 | } | |
1517 | ||
1518 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1519 | { | |
1520 | } | |
1521 | ||
1522 | static int nbd_dbg_init(void) | |
1523 | { | |
1524 | return 0; | |
1525 | } | |
1526 | ||
1527 | static void nbd_dbg_close(void) | |
1528 | { | |
1529 | } | |
1530 | ||
1531 | #endif | |
1532 | ||
d6296d39 CH |
1533 | static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1534 | unsigned int hctx_idx, unsigned int numa_node) | |
fd8383fd JB |
1535 | { |
1536 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
d6296d39 | 1537 | cmd->nbd = set->driver_data; |
d7d94d48 | 1538 | cmd->flags = 0; |
8f3ea359 | 1539 | mutex_init(&cmd->lock); |
fd8383fd JB |
1540 | return 0; |
1541 | } | |
1542 | ||
f363b089 | 1543 | static const struct blk_mq_ops nbd_mq_ops = { |
fd8383fd | 1544 | .queue_rq = nbd_queue_rq, |
1e388ae0 | 1545 | .complete = nbd_complete_rq, |
fd8383fd | 1546 | .init_request = nbd_init_request, |
0eadf37a | 1547 | .timeout = nbd_xmit_timeout, |
fd8383fd JB |
1548 | }; |
1549 | ||
b0d9111a JB |
1550 | static int nbd_dev_add(int index) |
1551 | { | |
1552 | struct nbd_device *nbd; | |
1553 | struct gendisk *disk; | |
1554 | struct request_queue *q; | |
1555 | int err = -ENOMEM; | |
1556 | ||
1557 | nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL); | |
1558 | if (!nbd) | |
1559 | goto out; | |
1560 | ||
1561 | disk = alloc_disk(1 << part_shift); | |
1562 | if (!disk) | |
1563 | goto out_free_nbd; | |
1564 | ||
1565 | if (index >= 0) { | |
1566 | err = idr_alloc(&nbd_index_idr, nbd, index, index + 1, | |
1567 | GFP_KERNEL); | |
1568 | if (err == -ENOSPC) | |
1569 | err = -EEXIST; | |
1570 | } else { | |
1571 | err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL); | |
1572 | if (err >= 0) | |
1573 | index = err; | |
1574 | } | |
1575 | if (err < 0) | |
1576 | goto out_free_disk; | |
1577 | ||
e46c7287 | 1578 | nbd->index = index; |
b0d9111a JB |
1579 | nbd->disk = disk; |
1580 | nbd->tag_set.ops = &nbd_mq_ops; | |
1581 | nbd->tag_set.nr_hw_queues = 1; | |
1582 | nbd->tag_set.queue_depth = 128; | |
1583 | nbd->tag_set.numa_node = NUMA_NO_NODE; | |
1584 | nbd->tag_set.cmd_size = sizeof(struct nbd_cmd); | |
1585 | nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | | |
56d18f62 | 1586 | BLK_MQ_F_BLOCKING; |
b0d9111a JB |
1587 | nbd->tag_set.driver_data = nbd; |
1588 | ||
1589 | err = blk_mq_alloc_tag_set(&nbd->tag_set); | |
1590 | if (err) | |
1591 | goto out_free_idr; | |
1592 | ||
1593 | q = blk_mq_init_queue(&nbd->tag_set); | |
1594 | if (IS_ERR(q)) { | |
1595 | err = PTR_ERR(q); | |
1596 | goto out_free_tags; | |
1597 | } | |
1598 | disk->queue = q; | |
1599 | ||
1600 | /* | |
1601 | * Tell the block layer that we are not a rotational device | |
1602 | */ | |
8b904b5b BVA |
1603 | blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue); |
1604 | blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue); | |
6df133a1 | 1605 | disk->queue->limits.discard_granularity = 0; |
07ce213f | 1606 | disk->queue->limits.discard_alignment = 0; |
6df133a1 | 1607 | blk_queue_max_discard_sectors(disk->queue, 0); |
ebb16d0d | 1608 | blk_queue_max_segment_size(disk->queue, UINT_MAX); |
1cc1f17a | 1609 | blk_queue_max_segments(disk->queue, USHRT_MAX); |
b0d9111a JB |
1610 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1611 | disk->queue->limits.max_sectors = 256; | |
1612 | ||
b0d9111a | 1613 | mutex_init(&nbd->config_lock); |
5ea8d108 | 1614 | refcount_set(&nbd->config_refs, 0); |
c6a4759e JB |
1615 | refcount_set(&nbd->refs, 1); |
1616 | INIT_LIST_HEAD(&nbd->list); | |
b0d9111a JB |
1617 | disk->major = NBD_MAJOR; |
1618 | disk->first_minor = index << part_shift; | |
1619 | disk->fops = &nbd_fops; | |
1620 | disk->private_data = nbd; | |
1621 | sprintf(disk->disk_name, "nbd%d", index); | |
b0d9111a | 1622 | add_disk(disk); |
47d902b9 | 1623 | nbd_total_devices++; |
b0d9111a JB |
1624 | return index; |
1625 | ||
1626 | out_free_tags: | |
1627 | blk_mq_free_tag_set(&nbd->tag_set); | |
1628 | out_free_idr: | |
1629 | idr_remove(&nbd_index_idr, index); | |
1630 | out_free_disk: | |
1631 | put_disk(disk); | |
1632 | out_free_nbd: | |
1633 | kfree(nbd); | |
1634 | out: | |
1635 | return err; | |
1636 | } | |
1637 | ||
e46c7287 JB |
1638 | static int find_free_cb(int id, void *ptr, void *data) |
1639 | { | |
1640 | struct nbd_device *nbd = ptr; | |
1641 | struct nbd_device **found = data; | |
1642 | ||
1643 | if (!refcount_read(&nbd->config_refs)) { | |
1644 | *found = nbd; | |
1645 | return 1; | |
1646 | } | |
1647 | return 0; | |
1648 | } | |
1649 | ||
1650 | /* Netlink interface. */ | |
a86c4120 | 1651 | static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = { |
e46c7287 JB |
1652 | [NBD_ATTR_INDEX] = { .type = NLA_U32 }, |
1653 | [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 }, | |
1654 | [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 }, | |
1655 | [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 }, | |
1656 | [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 }, | |
1657 | [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 }, | |
1658 | [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED}, | |
560bc4b3 | 1659 | [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 }, |
47d902b9 | 1660 | [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED}, |
e46c7287 JB |
1661 | }; |
1662 | ||
a86c4120 | 1663 | static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = { |
e46c7287 JB |
1664 | [NBD_SOCK_FD] = { .type = NLA_U32 }, |
1665 | }; | |
1666 | ||
47d902b9 JB |
1667 | /* We don't use this right now since we don't parse the incoming list, but we |
1668 | * still want it here so userspace knows what to expect. | |
1669 | */ | |
a86c4120 | 1670 | static const struct nla_policy __attribute__((unused)) |
47d902b9 JB |
1671 | nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = { |
1672 | [NBD_DEVICE_INDEX] = { .type = NLA_U32 }, | |
1673 | [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 }, | |
1674 | }; | |
1675 | ||
e46c7287 JB |
1676 | static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info) |
1677 | { | |
1678 | struct nbd_device *nbd = NULL; | |
1679 | struct nbd_config *config; | |
1680 | int index = -1; | |
1681 | int ret; | |
a2c97909 | 1682 | bool put_dev = false; |
e46c7287 JB |
1683 | |
1684 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1685 | return -EPERM; | |
1686 | ||
1687 | if (info->attrs[NBD_ATTR_INDEX]) | |
1688 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1689 | if (!info->attrs[NBD_ATTR_SOCKETS]) { | |
1690 | printk(KERN_ERR "nbd: must specify at least one socket\n"); | |
1691 | return -EINVAL; | |
1692 | } | |
1693 | if (!info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1694 | printk(KERN_ERR "nbd: must specify a size in bytes for the device\n"); | |
1695 | return -EINVAL; | |
1696 | } | |
1697 | again: | |
1698 | mutex_lock(&nbd_index_mutex); | |
1699 | if (index == -1) { | |
1700 | ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd); | |
1701 | if (ret == 0) { | |
1702 | int new_index; | |
1703 | new_index = nbd_dev_add(-1); | |
1704 | if (new_index < 0) { | |
1705 | mutex_unlock(&nbd_index_mutex); | |
1706 | printk(KERN_ERR "nbd: failed to add new device\n"); | |
0979962f | 1707 | return new_index; |
e46c7287 JB |
1708 | } |
1709 | nbd = idr_find(&nbd_index_idr, new_index); | |
1710 | } | |
1711 | } else { | |
1712 | nbd = idr_find(&nbd_index_idr, index); | |
e6a76272 JB |
1713 | if (!nbd) { |
1714 | ret = nbd_dev_add(index); | |
1715 | if (ret < 0) { | |
1716 | mutex_unlock(&nbd_index_mutex); | |
1717 | printk(KERN_ERR "nbd: failed to add new device\n"); | |
1718 | return ret; | |
1719 | } | |
1720 | nbd = idr_find(&nbd_index_idr, index); | |
1721 | } | |
e46c7287 | 1722 | } |
e46c7287 JB |
1723 | if (!nbd) { |
1724 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", | |
1725 | index); | |
c6a4759e JB |
1726 | mutex_unlock(&nbd_index_mutex); |
1727 | return -EINVAL; | |
1728 | } | |
1729 | if (!refcount_inc_not_zero(&nbd->refs)) { | |
1730 | mutex_unlock(&nbd_index_mutex); | |
1731 | if (index == -1) | |
1732 | goto again; | |
1733 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1734 | index); | |
e46c7287 JB |
1735 | return -EINVAL; |
1736 | } | |
c6a4759e | 1737 | mutex_unlock(&nbd_index_mutex); |
e46c7287 JB |
1738 | |
1739 | mutex_lock(&nbd->config_lock); | |
1740 | if (refcount_read(&nbd->config_refs)) { | |
1741 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1742 | nbd_put(nbd); |
e46c7287 JB |
1743 | if (index == -1) |
1744 | goto again; | |
1745 | printk(KERN_ERR "nbd: nbd%d already in use\n", index); | |
1746 | return -EBUSY; | |
1747 | } | |
1748 | if (WARN_ON(nbd->config)) { | |
1749 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1750 | nbd_put(nbd); |
e46c7287 JB |
1751 | return -EINVAL; |
1752 | } | |
1753 | config = nbd->config = nbd_alloc_config(); | |
1754 | if (!nbd->config) { | |
1755 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1756 | nbd_put(nbd); |
e46c7287 JB |
1757 | printk(KERN_ERR "nbd: couldn't allocate config\n"); |
1758 | return -ENOMEM; | |
1759 | } | |
1760 | refcount_set(&nbd->config_refs, 1); | |
1761 | set_bit(NBD_BOUND, &config->runtime_flags); | |
1762 | ||
1763 | if (info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1764 | u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]); | |
1765 | nbd_size_set(nbd, config->blksize, | |
1766 | div64_u64(bytes, config->blksize)); | |
1767 | } | |
1768 | if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) { | |
1769 | u64 bsize = | |
1770 | nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]); | |
1771 | nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize)); | |
1772 | } | |
1773 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1774 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1775 | nbd->tag_set.timeout = timeout * HZ; | |
1776 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1777 | } | |
560bc4b3 JB |
1778 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1779 | config->dead_conn_timeout = | |
1780 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1781 | config->dead_conn_timeout *= HZ; | |
1782 | } | |
e46c7287 JB |
1783 | if (info->attrs[NBD_ATTR_SERVER_FLAGS]) |
1784 | config->flags = | |
1785 | nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]); | |
a2c97909 JB |
1786 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { |
1787 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1788 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1789 | set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1790 | &config->runtime_flags); | |
1791 | put_dev = true; | |
1792 | } | |
08ba91ee DRK |
1793 | if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) { |
1794 | set_bit(NBD_DISCONNECT_ON_CLOSE, | |
1795 | &config->runtime_flags); | |
1796 | } | |
a2c97909 JB |
1797 | } |
1798 | ||
e46c7287 JB |
1799 | if (info->attrs[NBD_ATTR_SOCKETS]) { |
1800 | struct nlattr *attr; | |
1801 | int rem, fd; | |
1802 | ||
1803 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1804 | rem) { | |
1805 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1806 | ||
1807 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1808 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1809 | ret = -EINVAL; | |
1810 | goto out; | |
1811 | } | |
8cb08174 JB |
1812 | ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX, |
1813 | attr, | |
1814 | nbd_sock_policy, | |
1815 | info->extack); | |
e46c7287 JB |
1816 | if (ret != 0) { |
1817 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1818 | ret = -EINVAL; | |
1819 | goto out; | |
1820 | } | |
1821 | if (!socks[NBD_SOCK_FD]) | |
1822 | continue; | |
1823 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1824 | ret = nbd_add_socket(nbd, fd, true); | |
1825 | if (ret) | |
1826 | goto out; | |
1827 | } | |
1828 | } | |
1829 | ret = nbd_start_device(nbd); | |
1830 | out: | |
1831 | mutex_unlock(&nbd->config_lock); | |
1832 | if (!ret) { | |
1833 | set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags); | |
1834 | refcount_inc(&nbd->config_refs); | |
1835 | nbd_connect_reply(info, nbd->index); | |
1836 | } | |
1837 | nbd_config_put(nbd); | |
a2c97909 JB |
1838 | if (put_dev) |
1839 | nbd_put(nbd); | |
e46c7287 JB |
1840 | return ret; |
1841 | } | |
1842 | ||
08ba91ee DRK |
1843 | static void nbd_disconnect_and_put(struct nbd_device *nbd) |
1844 | { | |
1845 | mutex_lock(&nbd->config_lock); | |
1846 | nbd_disconnect(nbd); | |
1847 | nbd_clear_sock(nbd); | |
1848 | mutex_unlock(&nbd->config_lock); | |
1849 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, | |
1850 | &nbd->config->runtime_flags)) | |
1851 | nbd_config_put(nbd); | |
1852 | } | |
1853 | ||
e46c7287 JB |
1854 | static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info) |
1855 | { | |
1856 | struct nbd_device *nbd; | |
1857 | int index; | |
1858 | ||
1859 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1860 | return -EPERM; | |
1861 | ||
1862 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1863 | printk(KERN_ERR "nbd: must specify an index to disconnect\n"); | |
1864 | return -EINVAL; | |
1865 | } | |
1866 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1867 | mutex_lock(&nbd_index_mutex); | |
1868 | nbd = idr_find(&nbd_index_idr, index); | |
e46c7287 | 1869 | if (!nbd) { |
c6a4759e | 1870 | mutex_unlock(&nbd_index_mutex); |
e46c7287 JB |
1871 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", |
1872 | index); | |
1873 | return -EINVAL; | |
1874 | } | |
c6a4759e JB |
1875 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1876 | mutex_unlock(&nbd_index_mutex); | |
1877 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1878 | index); | |
1879 | return -EINVAL; | |
1880 | } | |
1881 | mutex_unlock(&nbd_index_mutex); | |
1882 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1883 | nbd_put(nbd); | |
e46c7287 | 1884 | return 0; |
c6a4759e | 1885 | } |
08ba91ee | 1886 | nbd_disconnect_and_put(nbd); |
e46c7287 | 1887 | nbd_config_put(nbd); |
c6a4759e | 1888 | nbd_put(nbd); |
e46c7287 JB |
1889 | return 0; |
1890 | } | |
1891 | ||
b7aa3d39 JB |
1892 | static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info) |
1893 | { | |
1894 | struct nbd_device *nbd = NULL; | |
1895 | struct nbd_config *config; | |
1896 | int index; | |
08ba91ee | 1897 | int ret = 0; |
a2c97909 | 1898 | bool put_dev = false; |
b7aa3d39 JB |
1899 | |
1900 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1901 | return -EPERM; | |
1902 | ||
1903 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1904 | printk(KERN_ERR "nbd: must specify a device to reconfigure\n"); | |
1905 | return -EINVAL; | |
1906 | } | |
1907 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1908 | mutex_lock(&nbd_index_mutex); | |
1909 | nbd = idr_find(&nbd_index_idr, index); | |
b7aa3d39 | 1910 | if (!nbd) { |
c6a4759e | 1911 | mutex_unlock(&nbd_index_mutex); |
b7aa3d39 JB |
1912 | printk(KERN_ERR "nbd: couldn't find a device at index %d\n", |
1913 | index); | |
1914 | return -EINVAL; | |
1915 | } | |
c6a4759e JB |
1916 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1917 | mutex_unlock(&nbd_index_mutex); | |
1918 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1919 | index); | |
1920 | return -EINVAL; | |
1921 | } | |
1922 | mutex_unlock(&nbd_index_mutex); | |
b7aa3d39 JB |
1923 | |
1924 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1925 | dev_err(nbd_to_dev(nbd), | |
1926 | "not configured, cannot reconfigure\n"); | |
c6a4759e | 1927 | nbd_put(nbd); |
b7aa3d39 JB |
1928 | return -EINVAL; |
1929 | } | |
1930 | ||
1931 | mutex_lock(&nbd->config_lock); | |
1932 | config = nbd->config; | |
1933 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1934 | !nbd->task_recv) { | |
1935 | dev_err(nbd_to_dev(nbd), | |
1936 | "not configured, cannot reconfigure\n"); | |
08ba91ee | 1937 | ret = -EINVAL; |
b7aa3d39 JB |
1938 | goto out; |
1939 | } | |
1940 | ||
1941 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1942 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1943 | nbd->tag_set.timeout = timeout * HZ; | |
1944 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1945 | } | |
560bc4b3 JB |
1946 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1947 | config->dead_conn_timeout = | |
1948 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1949 | config->dead_conn_timeout *= HZ; | |
1950 | } | |
a2c97909 JB |
1951 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { |
1952 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1953 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1954 | if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1955 | &config->runtime_flags)) | |
1956 | put_dev = true; | |
1957 | } else { | |
1958 | if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT, | |
1959 | &config->runtime_flags)) | |
1960 | refcount_inc(&nbd->refs); | |
1961 | } | |
08ba91ee DRK |
1962 | |
1963 | if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) { | |
1964 | set_bit(NBD_DISCONNECT_ON_CLOSE, | |
1965 | &config->runtime_flags); | |
1966 | } else { | |
1967 | clear_bit(NBD_DISCONNECT_ON_CLOSE, | |
1968 | &config->runtime_flags); | |
1969 | } | |
a2c97909 | 1970 | } |
b7aa3d39 JB |
1971 | |
1972 | if (info->attrs[NBD_ATTR_SOCKETS]) { | |
1973 | struct nlattr *attr; | |
1974 | int rem, fd; | |
1975 | ||
1976 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1977 | rem) { | |
1978 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1979 | ||
1980 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1981 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1982 | ret = -EINVAL; | |
1983 | goto out; | |
1984 | } | |
8cb08174 JB |
1985 | ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX, |
1986 | attr, | |
1987 | nbd_sock_policy, | |
1988 | info->extack); | |
b7aa3d39 JB |
1989 | if (ret != 0) { |
1990 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1991 | ret = -EINVAL; | |
1992 | goto out; | |
1993 | } | |
1994 | if (!socks[NBD_SOCK_FD]) | |
1995 | continue; | |
1996 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1997 | ret = nbd_reconnect_socket(nbd, fd); | |
1998 | if (ret) { | |
1999 | if (ret == -ENOSPC) | |
2000 | ret = 0; | |
2001 | goto out; | |
2002 | } | |
2003 | dev_info(nbd_to_dev(nbd), "reconnected socket\n"); | |
2004 | } | |
2005 | } | |
2006 | out: | |
2007 | mutex_unlock(&nbd->config_lock); | |
2008 | nbd_config_put(nbd); | |
c6a4759e | 2009 | nbd_put(nbd); |
a2c97909 JB |
2010 | if (put_dev) |
2011 | nbd_put(nbd); | |
b7aa3d39 JB |
2012 | return ret; |
2013 | } | |
2014 | ||
e46c7287 JB |
2015 | static const struct genl_ops nbd_connect_genl_ops[] = { |
2016 | { | |
2017 | .cmd = NBD_CMD_CONNECT, | |
ef6243ac | 2018 | .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, |
e46c7287 JB |
2019 | .doit = nbd_genl_connect, |
2020 | }, | |
2021 | { | |
2022 | .cmd = NBD_CMD_DISCONNECT, | |
ef6243ac | 2023 | .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, |
e46c7287 JB |
2024 | .doit = nbd_genl_disconnect, |
2025 | }, | |
b7aa3d39 JB |
2026 | { |
2027 | .cmd = NBD_CMD_RECONFIGURE, | |
ef6243ac | 2028 | .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, |
b7aa3d39 JB |
2029 | .doit = nbd_genl_reconfigure, |
2030 | }, | |
47d902b9 JB |
2031 | { |
2032 | .cmd = NBD_CMD_STATUS, | |
ef6243ac | 2033 | .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP, |
47d902b9 JB |
2034 | .doit = nbd_genl_status, |
2035 | }, | |
e46c7287 JB |
2036 | }; |
2037 | ||
799f9a38 JB |
2038 | static const struct genl_multicast_group nbd_mcast_grps[] = { |
2039 | { .name = NBD_GENL_MCAST_GROUP_NAME, }, | |
2040 | }; | |
2041 | ||
e46c7287 JB |
2042 | static struct genl_family nbd_genl_family __ro_after_init = { |
2043 | .hdrsize = 0, | |
2044 | .name = NBD_GENL_FAMILY_NAME, | |
2045 | .version = NBD_GENL_VERSION, | |
2046 | .module = THIS_MODULE, | |
2047 | .ops = nbd_connect_genl_ops, | |
2048 | .n_ops = ARRAY_SIZE(nbd_connect_genl_ops), | |
2049 | .maxattr = NBD_ATTR_MAX, | |
3b0f31f2 | 2050 | .policy = nbd_attr_policy, |
799f9a38 JB |
2051 | .mcgrps = nbd_mcast_grps, |
2052 | .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps), | |
e46c7287 JB |
2053 | }; |
2054 | ||
47d902b9 JB |
2055 | static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply) |
2056 | { | |
2057 | struct nlattr *dev_opt; | |
2058 | u8 connected = 0; | |
2059 | int ret; | |
2060 | ||
2061 | /* This is a little racey, but for status it's ok. The | |
2062 | * reason we don't take a ref here is because we can't | |
2063 | * take a ref in the index == -1 case as we would need | |
2064 | * to put under the nbd_index_mutex, which could | |
2065 | * deadlock if we are configured to remove ourselves | |
2066 | * once we're disconnected. | |
2067 | */ | |
2068 | if (refcount_read(&nbd->config_refs)) | |
2069 | connected = 1; | |
ae0be8de | 2070 | dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM); |
47d902b9 JB |
2071 | if (!dev_opt) |
2072 | return -EMSGSIZE; | |
2073 | ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index); | |
2074 | if (ret) | |
2075 | return -EMSGSIZE; | |
2076 | ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED, | |
2077 | connected); | |
2078 | if (ret) | |
2079 | return -EMSGSIZE; | |
2080 | nla_nest_end(reply, dev_opt); | |
2081 | return 0; | |
2082 | } | |
2083 | ||
2084 | static int status_cb(int id, void *ptr, void *data) | |
2085 | { | |
2086 | struct nbd_device *nbd = ptr; | |
2087 | return populate_nbd_status(nbd, (struct sk_buff *)data); | |
2088 | } | |
2089 | ||
2090 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info) | |
2091 | { | |
2092 | struct nlattr *dev_list; | |
2093 | struct sk_buff *reply; | |
2094 | void *reply_head; | |
2095 | size_t msg_size; | |
2096 | int index = -1; | |
2097 | int ret = -ENOMEM; | |
2098 | ||
2099 | if (info->attrs[NBD_ATTR_INDEX]) | |
2100 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
2101 | ||
2102 | mutex_lock(&nbd_index_mutex); | |
2103 | ||
2104 | msg_size = nla_total_size(nla_attr_size(sizeof(u32)) + | |
2105 | nla_attr_size(sizeof(u8))); | |
2106 | msg_size *= (index == -1) ? nbd_total_devices : 1; | |
2107 | ||
2108 | reply = genlmsg_new(msg_size, GFP_KERNEL); | |
2109 | if (!reply) | |
2110 | goto out; | |
2111 | reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0, | |
2112 | NBD_CMD_STATUS); | |
2113 | if (!reply_head) { | |
2114 | nlmsg_free(reply); | |
2115 | goto out; | |
2116 | } | |
2117 | ||
ae0be8de | 2118 | dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST); |
47d902b9 JB |
2119 | if (index == -1) { |
2120 | ret = idr_for_each(&nbd_index_idr, &status_cb, reply); | |
2121 | if (ret) { | |
2122 | nlmsg_free(reply); | |
2123 | goto out; | |
2124 | } | |
2125 | } else { | |
2126 | struct nbd_device *nbd; | |
2127 | nbd = idr_find(&nbd_index_idr, index); | |
2128 | if (nbd) { | |
2129 | ret = populate_nbd_status(nbd, reply); | |
2130 | if (ret) { | |
2131 | nlmsg_free(reply); | |
2132 | goto out; | |
2133 | } | |
2134 | } | |
2135 | } | |
2136 | nla_nest_end(reply, dev_list); | |
2137 | genlmsg_end(reply, reply_head); | |
cd46eb89 | 2138 | ret = genlmsg_reply(reply, info); |
47d902b9 JB |
2139 | out: |
2140 | mutex_unlock(&nbd_index_mutex); | |
2141 | return ret; | |
2142 | } | |
2143 | ||
e46c7287 JB |
2144 | static void nbd_connect_reply(struct genl_info *info, int index) |
2145 | { | |
2146 | struct sk_buff *skb; | |
2147 | void *msg_head; | |
2148 | int ret; | |
2149 | ||
2150 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
2151 | if (!skb) | |
2152 | return; | |
2153 | msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0, | |
2154 | NBD_CMD_CONNECT); | |
2155 | if (!msg_head) { | |
2156 | nlmsg_free(skb); | |
2157 | return; | |
2158 | } | |
2159 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
2160 | if (ret) { | |
2161 | nlmsg_free(skb); | |
2162 | return; | |
2163 | } | |
2164 | genlmsg_end(skb, msg_head); | |
2165 | genlmsg_reply(skb, info); | |
2166 | } | |
1da177e4 | 2167 | |
799f9a38 JB |
2168 | static void nbd_mcast_index(int index) |
2169 | { | |
2170 | struct sk_buff *skb; | |
2171 | void *msg_head; | |
2172 | int ret; | |
2173 | ||
2174 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
2175 | if (!skb) | |
2176 | return; | |
2177 | msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0, | |
2178 | NBD_CMD_LINK_DEAD); | |
2179 | if (!msg_head) { | |
2180 | nlmsg_free(skb); | |
2181 | return; | |
2182 | } | |
2183 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
2184 | if (ret) { | |
2185 | nlmsg_free(skb); | |
2186 | return; | |
2187 | } | |
2188 | genlmsg_end(skb, msg_head); | |
2189 | genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL); | |
2190 | } | |
2191 | ||
2192 | static void nbd_dead_link_work(struct work_struct *work) | |
2193 | { | |
2194 | struct link_dead_args *args = container_of(work, struct link_dead_args, | |
2195 | work); | |
2196 | nbd_mcast_index(args->index); | |
2197 | kfree(args); | |
2198 | } | |
2199 | ||
1da177e4 LT |
2200 | static int __init nbd_init(void) |
2201 | { | |
1da177e4 LT |
2202 | int i; |
2203 | ||
5b7b18cc | 2204 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 2205 | |
d71a6d73 | 2206 | if (max_part < 0) { |
7742ce4a | 2207 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
2208 | return -EINVAL; |
2209 | } | |
2210 | ||
2211 | part_shift = 0; | |
5988ce23 | 2212 | if (max_part > 0) { |
d71a6d73 LV |
2213 | part_shift = fls(max_part); |
2214 | ||
5988ce23 NK |
2215 | /* |
2216 | * Adjust max_part according to part_shift as it is exported | |
2217 | * to user space so that user can know the max number of | |
2218 | * partition kernel should be able to manage. | |
2219 | * | |
2220 | * Note that -1 is required because partition 0 is reserved | |
2221 | * for the whole disk. | |
2222 | */ | |
2223 | max_part = (1UL << part_shift) - 1; | |
2224 | } | |
2225 | ||
3b271082 NK |
2226 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
2227 | return -EINVAL; | |
2228 | ||
2229 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
2230 | return -EINVAL; | |
124d6db0 | 2231 | recv_workqueue = alloc_workqueue("knbd-recv", |
2189c97c DM |
2232 | WQ_MEM_RECLAIM | WQ_HIGHPRI | |
2233 | WQ_UNBOUND, 0); | |
124d6db0 JB |
2234 | if (!recv_workqueue) |
2235 | return -ENOMEM; | |
3b271082 | 2236 | |
6330a2d0 JB |
2237 | if (register_blkdev(NBD_MAJOR, "nbd")) { |
2238 | destroy_workqueue(recv_workqueue); | |
b0d9111a | 2239 | return -EIO; |
6330a2d0 | 2240 | } |
1da177e4 | 2241 | |
e46c7287 JB |
2242 | if (genl_register_family(&nbd_genl_family)) { |
2243 | unregister_blkdev(NBD_MAJOR, "nbd"); | |
2244 | destroy_workqueue(recv_workqueue); | |
2245 | return -EINVAL; | |
2246 | } | |
30d53d9c MP |
2247 | nbd_dbg_init(); |
2248 | ||
b0d9111a JB |
2249 | mutex_lock(&nbd_index_mutex); |
2250 | for (i = 0; i < nbds_max; i++) | |
2251 | nbd_dev_add(i); | |
2252 | mutex_unlock(&nbd_index_mutex); | |
2253 | return 0; | |
2254 | } | |
1da177e4 | 2255 | |
b0d9111a JB |
2256 | static int nbd_exit_cb(int id, void *ptr, void *data) |
2257 | { | |
c6a4759e | 2258 | struct list_head *list = (struct list_head *)data; |
b0d9111a | 2259 | struct nbd_device *nbd = ptr; |
c6a4759e | 2260 | |
c6a4759e | 2261 | list_add_tail(&nbd->list, list); |
1da177e4 | 2262 | return 0; |
1da177e4 LT |
2263 | } |
2264 | ||
2265 | static void __exit nbd_cleanup(void) | |
2266 | { | |
c6a4759e JB |
2267 | struct nbd_device *nbd; |
2268 | LIST_HEAD(del_list); | |
2269 | ||
30d53d9c MP |
2270 | nbd_dbg_close(); |
2271 | ||
c6a4759e JB |
2272 | mutex_lock(&nbd_index_mutex); |
2273 | idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list); | |
2274 | mutex_unlock(&nbd_index_mutex); | |
2275 | ||
60ae36ad JB |
2276 | while (!list_empty(&del_list)) { |
2277 | nbd = list_first_entry(&del_list, struct nbd_device, list); | |
2278 | list_del_init(&nbd->list); | |
2279 | if (refcount_read(&nbd->refs) != 1) | |
c6a4759e JB |
2280 | printk(KERN_ERR "nbd: possibly leaking a device\n"); |
2281 | nbd_put(nbd); | |
c6a4759e JB |
2282 | } |
2283 | ||
b0d9111a | 2284 | idr_destroy(&nbd_index_idr); |
e46c7287 | 2285 | genl_unregister_family(&nbd_genl_family); |
124d6db0 | 2286 | destroy_workqueue(recv_workqueue); |
1da177e4 | 2287 | unregister_blkdev(NBD_MAJOR, "nbd"); |
1da177e4 LT |
2288 | } |
2289 | ||
2290 | module_init(nbd_init); | |
2291 | module_exit(nbd_cleanup); | |
2292 | ||
2293 | MODULE_DESCRIPTION("Network Block Device"); | |
2294 | MODULE_LICENSE("GPL"); | |
2295 | ||
40be0c28 | 2296 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
2297 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
2298 | module_param(max_part, int, 0444); | |
7a8362a0 | 2299 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)"); |