]>
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 |
9b4a6ba9 | 79 | |
5ea8d108 | 80 | struct nbd_config { |
22d109c1 | 81 | u32 flags; |
9b4a6ba9 | 82 | unsigned long runtime_flags; |
560bc4b3 | 83 | u64 dead_conn_timeout; |
13e71d69 | 84 | |
5ea8d108 | 85 | struct nbd_sock **socks; |
9561a7ad | 86 | int num_connections; |
560bc4b3 JB |
87 | atomic_t live_connections; |
88 | wait_queue_head_t conn_wait; | |
5ea8d108 | 89 | |
9561a7ad JB |
90 | atomic_t recv_threads; |
91 | wait_queue_head_t recv_wq; | |
ef77b515 | 92 | loff_t blksize; |
b9c495bb | 93 | loff_t bytesize; |
30d53d9c MP |
94 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
95 | struct dentry *dbg_dir; | |
96 | #endif | |
13e71d69 MP |
97 | }; |
98 | ||
5ea8d108 JB |
99 | struct nbd_device { |
100 | struct blk_mq_tag_set tag_set; | |
101 | ||
e46c7287 | 102 | int index; |
5ea8d108 | 103 | refcount_t config_refs; |
c6a4759e | 104 | refcount_t refs; |
5ea8d108 JB |
105 | struct nbd_config *config; |
106 | struct mutex config_lock; | |
107 | struct gendisk *disk; | |
108 | ||
c6a4759e | 109 | struct list_head list; |
5ea8d108 JB |
110 | struct task_struct *task_recv; |
111 | struct task_struct *task_setup; | |
112 | }; | |
113 | ||
fd8383fd JB |
114 | struct nbd_cmd { |
115 | struct nbd_device *nbd; | |
f3733247 | 116 | int index; |
799f9a38 | 117 | int cookie; |
9561a7ad | 118 | struct completion send_complete; |
2a842aca | 119 | blk_status_t status; |
fd8383fd JB |
120 | }; |
121 | ||
30d53d9c MP |
122 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
123 | static struct dentry *nbd_dbg_dir; | |
124 | #endif | |
125 | ||
126 | #define nbd_name(nbd) ((nbd)->disk->disk_name) | |
127 | ||
f4507164 | 128 | #define NBD_MAGIC 0x68797548 |
1da177e4 | 129 | |
9c7a4169 | 130 | static unsigned int nbds_max = 16; |
d71a6d73 | 131 | static int max_part; |
124d6db0 | 132 | static struct workqueue_struct *recv_workqueue; |
b0d9111a | 133 | static int part_shift; |
1da177e4 | 134 | |
9442b739 JB |
135 | static int nbd_dev_dbg_init(struct nbd_device *nbd); |
136 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
5ea8d108 | 137 | static void nbd_config_put(struct nbd_device *nbd); |
e46c7287 | 138 | static void nbd_connect_reply(struct genl_info *info, int index); |
47d902b9 | 139 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info); |
799f9a38 | 140 | static void nbd_dead_link_work(struct work_struct *work); |
9442b739 | 141 | |
d18509f5 | 142 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) |
1da177e4 | 143 | { |
d18509f5 | 144 | return disk_to_dev(nbd->disk); |
1da177e4 LT |
145 | } |
146 | ||
147 | static const char *nbdcmd_to_ascii(int cmd) | |
148 | { | |
149 | switch (cmd) { | |
150 | case NBD_CMD_READ: return "read"; | |
151 | case NBD_CMD_WRITE: return "write"; | |
152 | case NBD_CMD_DISC: return "disconnect"; | |
75f187ab | 153 | case NBD_CMD_FLUSH: return "flush"; |
a336d298 | 154 | case NBD_CMD_TRIM: return "trim/discard"; |
1da177e4 LT |
155 | } |
156 | return "invalid"; | |
157 | } | |
1da177e4 | 158 | |
5ea8d108 JB |
159 | static ssize_t pid_show(struct device *dev, |
160 | struct device_attribute *attr, char *buf) | |
161 | { | |
162 | struct gendisk *disk = dev_to_disk(dev); | |
163 | struct nbd_device *nbd = (struct nbd_device *)disk->private_data; | |
164 | ||
165 | return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv)); | |
166 | } | |
167 | ||
168 | static struct device_attribute pid_attr = { | |
169 | .attr = { .name = "pid", .mode = S_IRUGO}, | |
170 | .show = pid_show, | |
171 | }; | |
172 | ||
c6a4759e JB |
173 | static void nbd_dev_remove(struct nbd_device *nbd) |
174 | { | |
175 | struct gendisk *disk = nbd->disk; | |
176 | if (disk) { | |
177 | del_gendisk(disk); | |
178 | blk_cleanup_queue(disk->queue); | |
179 | blk_mq_free_tag_set(&nbd->tag_set); | |
a2c97909 | 180 | disk->private_data = NULL; |
c6a4759e JB |
181 | put_disk(disk); |
182 | } | |
183 | kfree(nbd); | |
184 | } | |
185 | ||
186 | static void nbd_put(struct nbd_device *nbd) | |
187 | { | |
188 | if (refcount_dec_and_mutex_lock(&nbd->refs, | |
189 | &nbd_index_mutex)) { | |
190 | idr_remove(&nbd_index_idr, nbd->index); | |
191 | mutex_unlock(&nbd_index_mutex); | |
192 | nbd_dev_remove(nbd); | |
193 | } | |
194 | } | |
195 | ||
799f9a38 JB |
196 | static int nbd_disconnected(struct nbd_config *config) |
197 | { | |
198 | return test_bit(NBD_DISCONNECTED, &config->runtime_flags) || | |
199 | test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); | |
200 | } | |
201 | ||
202 | static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock, | |
203 | int notify) | |
f3733247 | 204 | { |
799f9a38 JB |
205 | if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) { |
206 | struct link_dead_args *args; | |
207 | args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO); | |
208 | if (args) { | |
209 | INIT_WORK(&args->work, nbd_dead_link_work); | |
210 | args->index = nbd->index; | |
211 | queue_work(system_wq, &args->work); | |
212 | } | |
213 | } | |
560bc4b3 | 214 | if (!nsock->dead) { |
f3733247 | 215 | kernel_sock_shutdown(nsock->sock, SHUT_RDWR); |
560bc4b3 JB |
216 | atomic_dec(&nbd->config->live_connections); |
217 | } | |
f3733247 JB |
218 | nsock->dead = true; |
219 | nsock->pending = NULL; | |
220 | nsock->sent = 0; | |
221 | } | |
222 | ||
29eaadc0 | 223 | static void nbd_size_clear(struct nbd_device *nbd) |
37091fdd | 224 | { |
5ea8d108 | 225 | if (nbd->config->bytesize) { |
5ea8d108 JB |
226 | set_capacity(nbd->disk, 0); |
227 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
228 | } | |
37091fdd MP |
229 | } |
230 | ||
29eaadc0 | 231 | static void nbd_size_update(struct nbd_device *nbd) |
37091fdd | 232 | { |
5ea8d108 JB |
233 | struct nbd_config *config = nbd->config; |
234 | blk_queue_logical_block_size(nbd->disk->queue, config->blksize); | |
235 | blk_queue_physical_block_size(nbd->disk->queue, config->blksize); | |
5ea8d108 | 236 | set_capacity(nbd->disk, config->bytesize >> 9); |
37091fdd MP |
237 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); |
238 | } | |
239 | ||
29eaadc0 JB |
240 | static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize, |
241 | loff_t nr_blocks) | |
37091fdd | 242 | { |
5ea8d108 JB |
243 | struct nbd_config *config = nbd->config; |
244 | config->blksize = blocksize; | |
245 | config->bytesize = blocksize * nr_blocks; | |
29eaadc0 | 246 | nbd_size_update(nbd); |
37091fdd MP |
247 | } |
248 | ||
1e388ae0 | 249 | static void nbd_complete_rq(struct request *req) |
1da177e4 | 250 | { |
1e388ae0 | 251 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
1da177e4 | 252 | |
1e388ae0 CH |
253 | dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd, |
254 | cmd->status ? "failed" : "done"); | |
1da177e4 | 255 | |
1e388ae0 | 256 | blk_mq_end_request(req, cmd->status); |
1da177e4 LT |
257 | } |
258 | ||
e018e757 MP |
259 | /* |
260 | * Forcibly shutdown the socket causing all listeners to error | |
261 | */ | |
36e47bee | 262 | static void sock_shutdown(struct nbd_device *nbd) |
7fdfd406 | 263 | { |
5ea8d108 | 264 | struct nbd_config *config = nbd->config; |
9561a7ad | 265 | int i; |
23272a67 | 266 | |
5ea8d108 | 267 | if (config->num_connections == 0) |
9561a7ad | 268 | return; |
5ea8d108 | 269 | if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
260bbce4 | 270 | return; |
23272a67 | 271 | |
5ea8d108 JB |
272 | for (i = 0; i < config->num_connections; i++) { |
273 | struct nbd_sock *nsock = config->socks[i]; | |
9561a7ad | 274 | mutex_lock(&nsock->tx_lock); |
799f9a38 | 275 | nbd_mark_nsock_dead(nbd, nsock, 0); |
9561a7ad JB |
276 | mutex_unlock(&nsock->tx_lock); |
277 | } | |
278 | dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n"); | |
7fdfd406 PC |
279 | } |
280 | ||
0eadf37a JB |
281 | static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req, |
282 | bool reserved) | |
7fdfd406 | 283 | { |
0eadf37a JB |
284 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); |
285 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 JB |
286 | struct nbd_config *config; |
287 | ||
288 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
2a842aca | 289 | cmd->status = BLK_STS_TIMEOUT; |
5ea8d108 JB |
290 | return BLK_EH_HANDLED; |
291 | } | |
292 | ||
560bc4b3 JB |
293 | /* If we are waiting on our dead timer then we could get timeout |
294 | * callbacks for our request. For this we just want to reset the timer | |
295 | * and let the queue side take care of everything. | |
296 | */ | |
297 | if (!completion_done(&cmd->send_complete)) { | |
298 | nbd_config_put(nbd); | |
299 | return BLK_EH_RESET_TIMER; | |
300 | } | |
5ea8d108 | 301 | config = nbd->config; |
dcc909d9 | 302 | |
5ea8d108 | 303 | if (config->num_connections > 1) { |
f3733247 JB |
304 | dev_err_ratelimited(nbd_to_dev(nbd), |
305 | "Connection timed out, retrying\n"); | |
f3733247 JB |
306 | /* |
307 | * Hooray we have more connections, requeue this IO, the submit | |
308 | * path will put it on a real connection. | |
309 | */ | |
5ea8d108 JB |
310 | if (config->socks && config->num_connections > 1) { |
311 | if (cmd->index < config->num_connections) { | |
f3733247 | 312 | struct nbd_sock *nsock = |
5ea8d108 | 313 | config->socks[cmd->index]; |
f3733247 | 314 | mutex_lock(&nsock->tx_lock); |
799f9a38 JB |
315 | /* We can have multiple outstanding requests, so |
316 | * we don't want to mark the nsock dead if we've | |
317 | * already reconnected with a new socket, so | |
318 | * only mark it dead if its the same socket we | |
319 | * were sent out on. | |
320 | */ | |
321 | if (cmd->cookie == nsock->cookie) | |
322 | nbd_mark_nsock_dead(nbd, nsock, 1); | |
f3733247 JB |
323 | mutex_unlock(&nsock->tx_lock); |
324 | } | |
f3733247 | 325 | blk_mq_requeue_request(req, true); |
5ea8d108 | 326 | nbd_config_put(nbd); |
f3733247 JB |
327 | return BLK_EH_NOT_HANDLED; |
328 | } | |
f3733247 JB |
329 | } else { |
330 | dev_err_ratelimited(nbd_to_dev(nbd), | |
331 | "Connection timed out\n"); | |
332 | } | |
5ea8d108 | 333 | set_bit(NBD_TIMEDOUT, &config->runtime_flags); |
2a842aca | 334 | cmd->status = BLK_STS_IOERR; |
9561a7ad | 335 | sock_shutdown(nbd); |
5ea8d108 JB |
336 | nbd_config_put(nbd); |
337 | ||
0eadf37a | 338 | return BLK_EH_HANDLED; |
7fdfd406 PC |
339 | } |
340 | ||
1da177e4 LT |
341 | /* |
342 | * Send or receive packet. | |
343 | */ | |
c9f2b6ae | 344 | static int sock_xmit(struct nbd_device *nbd, int index, int send, |
9dd5d3ab | 345 | struct iov_iter *iter, int msg_flags, int *sent) |
1da177e4 | 346 | { |
5ea8d108 JB |
347 | struct nbd_config *config = nbd->config; |
348 | struct socket *sock = config->socks[index]->sock; | |
1da177e4 LT |
349 | int result; |
350 | struct msghdr msg; | |
f1083048 | 351 | unsigned int noreclaim_flag; |
1da177e4 | 352 | |
ffc41cf8 | 353 | if (unlikely(!sock)) { |
a897b666 | 354 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
355 | "Attempted %s on closed socket in sock_xmit\n", |
356 | (send ? "send" : "recv")); | |
ffc41cf8 MS |
357 | return -EINVAL; |
358 | } | |
359 | ||
c9f2b6ae | 360 | msg.msg_iter = *iter; |
c1696cab | 361 | |
f1083048 | 362 | noreclaim_flag = memalloc_noreclaim_save(); |
1da177e4 | 363 | do { |
7f338fe4 | 364 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; |
1da177e4 LT |
365 | msg.msg_name = NULL; |
366 | msg.msg_namelen = 0; | |
367 | msg.msg_control = NULL; | |
368 | msg.msg_controllen = 0; | |
1da177e4 LT |
369 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; |
370 | ||
7e2893a1 | 371 | if (send) |
c1696cab | 372 | result = sock_sendmsg(sock, &msg); |
7e2893a1 | 373 | else |
c1696cab | 374 | result = sock_recvmsg(sock, &msg, msg.msg_flags); |
1da177e4 | 375 | |
1da177e4 LT |
376 | if (result <= 0) { |
377 | if (result == 0) | |
378 | result = -EPIPE; /* short read */ | |
379 | break; | |
380 | } | |
9dd5d3ab JB |
381 | if (sent) |
382 | *sent += result; | |
c1696cab | 383 | } while (msg_data_left(&msg)); |
1da177e4 | 384 | |
f1083048 | 385 | memalloc_noreclaim_restore(noreclaim_flag); |
1da177e4 LT |
386 | |
387 | return result; | |
388 | } | |
389 | ||
7fdfd406 | 390 | /* always call with the tx_lock held */ |
9561a7ad | 391 | static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index) |
1da177e4 | 392 | { |
fd8383fd | 393 | struct request *req = blk_mq_rq_from_pdu(cmd); |
5ea8d108 JB |
394 | struct nbd_config *config = nbd->config; |
395 | struct nbd_sock *nsock = config->socks[index]; | |
d61b7f97 | 396 | int result; |
c9f2b6ae AV |
397 | struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)}; |
398 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
399 | struct iov_iter from; | |
1011c1b9 | 400 | unsigned long size = blk_rq_bytes(req); |
429a787b | 401 | struct bio *bio; |
9dc6c806 | 402 | u32 type; |
685c9b24 | 403 | u32 nbd_cmd_flags = 0; |
9561a7ad | 404 | u32 tag = blk_mq_unique_tag(req); |
9dd5d3ab | 405 | int sent = nsock->sent, skip = 0; |
9dc6c806 | 406 | |
c9f2b6ae AV |
407 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); |
408 | ||
aebf526b CH |
409 | switch (req_op(req)) { |
410 | case REQ_OP_DISCARD: | |
9dc6c806 | 411 | type = NBD_CMD_TRIM; |
aebf526b CH |
412 | break; |
413 | case REQ_OP_FLUSH: | |
9dc6c806 | 414 | type = NBD_CMD_FLUSH; |
aebf526b CH |
415 | break; |
416 | case REQ_OP_WRITE: | |
9dc6c806 | 417 | type = NBD_CMD_WRITE; |
aebf526b CH |
418 | break; |
419 | case REQ_OP_READ: | |
9dc6c806 | 420 | type = NBD_CMD_READ; |
aebf526b CH |
421 | break; |
422 | default: | |
423 | return -EIO; | |
424 | } | |
1da177e4 | 425 | |
09fc54cc | 426 | if (rq_data_dir(req) == WRITE && |
5ea8d108 | 427 | (config->flags & NBD_FLAG_READ_ONLY)) { |
09fc54cc CH |
428 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
429 | "Write on read-only\n"); | |
430 | return -EIO; | |
431 | } | |
432 | ||
685c9b24 SM |
433 | if (req->cmd_flags & REQ_FUA) |
434 | nbd_cmd_flags |= NBD_CMD_FLAG_FUA; | |
435 | ||
9dd5d3ab JB |
436 | /* We did a partial send previously, and we at least sent the whole |
437 | * request struct, so just go and send the rest of the pages in the | |
438 | * request. | |
439 | */ | |
440 | if (sent) { | |
441 | if (sent >= sizeof(request)) { | |
442 | skip = sent - sizeof(request); | |
443 | goto send_pages; | |
444 | } | |
445 | iov_iter_advance(&from, sent); | |
446 | } | |
f3733247 | 447 | cmd->index = index; |
799f9a38 | 448 | cmd->cookie = nsock->cookie; |
685c9b24 | 449 | request.type = htonl(type | nbd_cmd_flags); |
9561a7ad | 450 | if (type != NBD_CMD_FLUSH) { |
75f187ab AB |
451 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); |
452 | request.len = htonl(size); | |
453 | } | |
9561a7ad | 454 | memcpy(request.handle, &tag, sizeof(tag)); |
1da177e4 | 455 | |
d18509f5 | 456 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", |
fd8383fd | 457 | cmd, nbdcmd_to_ascii(type), |
d18509f5 | 458 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); |
c9f2b6ae | 459 | result = sock_xmit(nbd, index, 1, &from, |
9dd5d3ab | 460 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent); |
1da177e4 | 461 | if (result <= 0) { |
9dd5d3ab JB |
462 | if (result == -ERESTARTSYS) { |
463 | /* If we havne't sent anything we can just return BUSY, | |
464 | * however if we have sent something we need to make | |
465 | * sure we only allow this req to be sent until we are | |
466 | * completely done. | |
467 | */ | |
468 | if (sent) { | |
469 | nsock->pending = req; | |
470 | nsock->sent = sent; | |
471 | } | |
fc17b653 | 472 | return BLK_STS_RESOURCE; |
9dd5d3ab | 473 | } |
a897b666 | 474 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
7f1b90f9 | 475 | "Send control failed (result %d)\n", result); |
f3733247 | 476 | return -EAGAIN; |
1da177e4 | 477 | } |
9dd5d3ab | 478 | send_pages: |
429a787b | 479 | if (type != NBD_CMD_WRITE) |
9dd5d3ab | 480 | goto out; |
429a787b | 481 | |
429a787b JA |
482 | bio = req->bio; |
483 | while (bio) { | |
484 | struct bio *next = bio->bi_next; | |
485 | struct bvec_iter iter; | |
7988613b | 486 | struct bio_vec bvec; |
429a787b JA |
487 | |
488 | bio_for_each_segment(bvec, bio, iter) { | |
489 | bool is_last = !next && bio_iter_last(bvec, iter); | |
d61b7f97 | 490 | int flags = is_last ? 0 : MSG_MORE; |
429a787b | 491 | |
d18509f5 | 492 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", |
fd8383fd | 493 | cmd, bvec.bv_len); |
c9f2b6ae AV |
494 | iov_iter_bvec(&from, ITER_BVEC | WRITE, |
495 | &bvec, 1, bvec.bv_len); | |
9dd5d3ab JB |
496 | if (skip) { |
497 | if (skip >= iov_iter_count(&from)) { | |
498 | skip -= iov_iter_count(&from); | |
499 | continue; | |
500 | } | |
501 | iov_iter_advance(&from, skip); | |
502 | skip = 0; | |
503 | } | |
504 | result = sock_xmit(nbd, index, 1, &from, flags, &sent); | |
6c92e699 | 505 | if (result <= 0) { |
9dd5d3ab JB |
506 | if (result == -ERESTARTSYS) { |
507 | /* We've already sent the header, we | |
508 | * have no choice but to set pending and | |
509 | * return BUSY. | |
510 | */ | |
511 | nsock->pending = req; | |
512 | nsock->sent = sent; | |
fc17b653 | 513 | return BLK_STS_RESOURCE; |
9dd5d3ab | 514 | } |
f4507164 | 515 | dev_err(disk_to_dev(nbd->disk), |
7f1b90f9 WC |
516 | "Send data failed (result %d)\n", |
517 | result); | |
f3733247 | 518 | return -EAGAIN; |
6c92e699 | 519 | } |
429a787b JA |
520 | /* |
521 | * The completion might already have come in, | |
522 | * so break for the last one instead of letting | |
523 | * the iterator do it. This prevents use-after-free | |
524 | * of the bio. | |
525 | */ | |
526 | if (is_last) | |
527 | break; | |
1da177e4 | 528 | } |
429a787b | 529 | bio = next; |
1da177e4 | 530 | } |
9dd5d3ab JB |
531 | out: |
532 | nsock->pending = NULL; | |
533 | nsock->sent = 0; | |
1da177e4 | 534 | return 0; |
1da177e4 LT |
535 | } |
536 | ||
1da177e4 | 537 | /* NULL returned = something went wrong, inform userspace */ |
9561a7ad | 538 | static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index) |
1da177e4 | 539 | { |
5ea8d108 | 540 | struct nbd_config *config = nbd->config; |
1da177e4 LT |
541 | int result; |
542 | struct nbd_reply reply; | |
fd8383fd JB |
543 | struct nbd_cmd *cmd; |
544 | struct request *req = NULL; | |
545 | u16 hwq; | |
9561a7ad | 546 | u32 tag; |
c9f2b6ae AV |
547 | struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)}; |
548 | struct iov_iter to; | |
1da177e4 LT |
549 | |
550 | reply.magic = 0; | |
c9f2b6ae | 551 | iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply)); |
9dd5d3ab | 552 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
1da177e4 | 553 | if (result <= 0) { |
5ea8d108 | 554 | if (!nbd_disconnected(config)) |
9561a7ad JB |
555 | dev_err(disk_to_dev(nbd->disk), |
556 | "Receive control failed (result %d)\n", result); | |
19391830 | 557 | return ERR_PTR(result); |
1da177e4 | 558 | } |
e4b57e08 MF |
559 | |
560 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
f4507164 | 561 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", |
e4b57e08 | 562 | (unsigned long)ntohl(reply.magic)); |
19391830 | 563 | return ERR_PTR(-EPROTO); |
e4b57e08 MF |
564 | } |
565 | ||
9561a7ad | 566 | memcpy(&tag, reply.handle, sizeof(u32)); |
4b2f0260 | 567 | |
fd8383fd JB |
568 | hwq = blk_mq_unique_tag_to_hwq(tag); |
569 | if (hwq < nbd->tag_set.nr_hw_queues) | |
570 | req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq], | |
571 | blk_mq_unique_tag_to_tag(tag)); | |
572 | if (!req || !blk_mq_request_started(req)) { | |
573 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n", | |
574 | tag, req); | |
575 | return ERR_PTR(-ENOENT); | |
1da177e4 | 576 | } |
fd8383fd | 577 | cmd = blk_mq_rq_to_pdu(req); |
1da177e4 | 578 | if (ntohl(reply.error)) { |
f4507164 | 579 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", |
7f1b90f9 | 580 | ntohl(reply.error)); |
2a842aca | 581 | cmd->status = BLK_STS_IOERR; |
fd8383fd | 582 | return cmd; |
1da177e4 LT |
583 | } |
584 | ||
fd8383fd | 585 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd); |
9dc6c806 | 586 | if (rq_data_dir(req) != WRITE) { |
5705f702 | 587 | struct req_iterator iter; |
7988613b | 588 | struct bio_vec bvec; |
5705f702 N |
589 | |
590 | rq_for_each_segment(bvec, req, iter) { | |
c9f2b6ae AV |
591 | iov_iter_bvec(&to, ITER_BVEC | READ, |
592 | &bvec, 1, bvec.bv_len); | |
9dd5d3ab | 593 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); |
6c92e699 | 594 | if (result <= 0) { |
f4507164 | 595 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", |
7f1b90f9 | 596 | result); |
f3733247 JB |
597 | /* |
598 | * If we've disconnected or we only have 1 | |
599 | * connection then we need to make sure we | |
600 | * complete this request, otherwise error out | |
601 | * and let the timeout stuff handle resubmitting | |
602 | * this request onto another connection. | |
603 | */ | |
5ea8d108 JB |
604 | if (nbd_disconnected(config) || |
605 | config->num_connections <= 1) { | |
2a842aca | 606 | cmd->status = BLK_STS_IOERR; |
f3733247 JB |
607 | return cmd; |
608 | } | |
609 | return ERR_PTR(-EIO); | |
6c92e699 | 610 | } |
d18509f5 | 611 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", |
fd8383fd | 612 | cmd, bvec.bv_len); |
1da177e4 | 613 | } |
9561a7ad JB |
614 | } else { |
615 | /* See the comment in nbd_queue_rq. */ | |
616 | wait_for_completion(&cmd->send_complete); | |
1da177e4 | 617 | } |
fd8383fd | 618 | return cmd; |
1da177e4 LT |
619 | } |
620 | ||
9561a7ad | 621 | static void recv_work(struct work_struct *work) |
1da177e4 | 622 | { |
9561a7ad JB |
623 | struct recv_thread_args *args = container_of(work, |
624 | struct recv_thread_args, | |
625 | work); | |
626 | struct nbd_device *nbd = args->nbd; | |
5ea8d108 | 627 | struct nbd_config *config = nbd->config; |
fd8383fd | 628 | struct nbd_cmd *cmd; |
1da177e4 | 629 | |
19391830 | 630 | while (1) { |
9561a7ad | 631 | cmd = nbd_read_stat(nbd, args->index); |
fd8383fd | 632 | if (IS_ERR(cmd)) { |
5ea8d108 | 633 | struct nbd_sock *nsock = config->socks[args->index]; |
f3733247 JB |
634 | |
635 | mutex_lock(&nsock->tx_lock); | |
799f9a38 | 636 | nbd_mark_nsock_dead(nbd, nsock, 1); |
f3733247 | 637 | mutex_unlock(&nsock->tx_lock); |
19391830 MP |
638 | break; |
639 | } | |
640 | ||
08e0029a | 641 | blk_mq_complete_request(blk_mq_rq_from_pdu(cmd)); |
19391830 | 642 | } |
5ea8d108 JB |
643 | atomic_dec(&config->recv_threads); |
644 | wake_up(&config->recv_wq); | |
645 | nbd_config_put(nbd); | |
646 | kfree(args); | |
1da177e4 LT |
647 | } |
648 | ||
fd8383fd | 649 | static void nbd_clear_req(struct request *req, void *data, bool reserved) |
1da177e4 | 650 | { |
fd8383fd | 651 | struct nbd_cmd *cmd; |
1da177e4 | 652 | |
fd8383fd JB |
653 | if (!blk_mq_request_started(req)) |
654 | return; | |
655 | cmd = blk_mq_rq_to_pdu(req); | |
2a842aca | 656 | cmd->status = BLK_STS_IOERR; |
08e0029a | 657 | blk_mq_complete_request(req); |
fd8383fd JB |
658 | } |
659 | ||
660 | static void nbd_clear_que(struct nbd_device *nbd) | |
661 | { | |
b52c2e92 | 662 | blk_mq_quiesce_queue(nbd->disk->queue); |
fd8383fd | 663 | blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL); |
b52c2e92 | 664 | blk_mq_unquiesce_queue(nbd->disk->queue); |
e78273c8 | 665 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); |
1da177e4 LT |
666 | } |
667 | ||
f3733247 JB |
668 | static int find_fallback(struct nbd_device *nbd, int index) |
669 | { | |
5ea8d108 | 670 | struct nbd_config *config = nbd->config; |
f3733247 | 671 | int new_index = -1; |
5ea8d108 | 672 | struct nbd_sock *nsock = config->socks[index]; |
f3733247 JB |
673 | int fallback = nsock->fallback_index; |
674 | ||
5ea8d108 | 675 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) |
f3733247 JB |
676 | return new_index; |
677 | ||
5ea8d108 | 678 | if (config->num_connections <= 1) { |
f3733247 JB |
679 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
680 | "Attempted send on invalid socket\n"); | |
681 | return new_index; | |
682 | } | |
683 | ||
5ea8d108 JB |
684 | if (fallback >= 0 && fallback < config->num_connections && |
685 | !config->socks[fallback]->dead) | |
f3733247 JB |
686 | return fallback; |
687 | ||
688 | if (nsock->fallback_index < 0 || | |
5ea8d108 JB |
689 | nsock->fallback_index >= config->num_connections || |
690 | config->socks[nsock->fallback_index]->dead) { | |
f3733247 | 691 | int i; |
5ea8d108 | 692 | for (i = 0; i < config->num_connections; i++) { |
f3733247 JB |
693 | if (i == index) |
694 | continue; | |
5ea8d108 | 695 | if (!config->socks[i]->dead) { |
f3733247 JB |
696 | new_index = i; |
697 | break; | |
698 | } | |
699 | } | |
700 | nsock->fallback_index = new_index; | |
701 | if (new_index < 0) { | |
702 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
703 | "Dead connection, failed to find a fallback\n"); | |
704 | return new_index; | |
705 | } | |
706 | } | |
707 | new_index = nsock->fallback_index; | |
708 | return new_index; | |
709 | } | |
7fdfd406 | 710 | |
560bc4b3 JB |
711 | static int wait_for_reconnect(struct nbd_device *nbd) |
712 | { | |
713 | struct nbd_config *config = nbd->config; | |
714 | if (!config->dead_conn_timeout) | |
715 | return 0; | |
716 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) | |
717 | return 0; | |
718 | wait_event_interruptible_timeout(config->conn_wait, | |
719 | atomic_read(&config->live_connections), | |
720 | config->dead_conn_timeout); | |
721 | return atomic_read(&config->live_connections); | |
722 | } | |
723 | ||
9dd5d3ab | 724 | static int nbd_handle_cmd(struct nbd_cmd *cmd, int index) |
48cf6061 | 725 | { |
fd8383fd JB |
726 | struct request *req = blk_mq_rq_from_pdu(cmd); |
727 | struct nbd_device *nbd = cmd->nbd; | |
5ea8d108 | 728 | struct nbd_config *config; |
9561a7ad | 729 | struct nbd_sock *nsock; |
9dd5d3ab | 730 | int ret; |
fd8383fd | 731 | |
5ea8d108 JB |
732 | if (!refcount_inc_not_zero(&nbd->config_refs)) { |
733 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
734 | "Socks array is empty\n"); | |
735 | return -EINVAL; | |
736 | } | |
737 | config = nbd->config; | |
738 | ||
739 | if (index >= config->num_connections) { | |
a897b666 JB |
740 | dev_err_ratelimited(disk_to_dev(nbd->disk), |
741 | "Attempted send on invalid socket\n"); | |
5ea8d108 | 742 | nbd_config_put(nbd); |
9dd5d3ab | 743 | return -EINVAL; |
9561a7ad | 744 | } |
2a842aca | 745 | cmd->status = BLK_STS_OK; |
f3733247 | 746 | again: |
5ea8d108 | 747 | nsock = config->socks[index]; |
9561a7ad | 748 | mutex_lock(&nsock->tx_lock); |
f3733247 | 749 | if (nsock->dead) { |
560bc4b3 | 750 | int old_index = index; |
f3733247 | 751 | index = find_fallback(nbd, index); |
560bc4b3 | 752 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 753 | if (index < 0) { |
560bc4b3 JB |
754 | if (wait_for_reconnect(nbd)) { |
755 | index = old_index; | |
756 | goto again; | |
757 | } | |
758 | /* All the sockets should already be down at this point, | |
759 | * we just want to make sure that DISCONNECTED is set so | |
760 | * any requests that come in that were queue'ed waiting | |
761 | * for the reconnect timer don't trigger the timer again | |
762 | * and instead just error out. | |
763 | */ | |
764 | sock_shutdown(nbd); | |
765 | nbd_config_put(nbd); | |
766 | return -EIO; | |
5ea8d108 | 767 | } |
f3733247 | 768 | goto again; |
48cf6061 LV |
769 | } |
770 | ||
9dd5d3ab JB |
771 | /* Handle the case that we have a pending request that was partially |
772 | * transmitted that _has_ to be serviced first. We need to call requeue | |
773 | * here so that it gets put _after_ the request that is already on the | |
774 | * dispatch list. | |
775 | */ | |
776 | if (unlikely(nsock->pending && nsock->pending != req)) { | |
777 | blk_mq_requeue_request(req, true); | |
778 | ret = 0; | |
779 | goto out; | |
48cf6061 | 780 | } |
f3733247 JB |
781 | /* |
782 | * Some failures are related to the link going down, so anything that | |
783 | * returns EAGAIN can be retried on a different socket. | |
784 | */ | |
9dd5d3ab | 785 | ret = nbd_send_cmd(nbd, cmd, index); |
f3733247 JB |
786 | if (ret == -EAGAIN) { |
787 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
788 | "Request send failed trying another connection\n"); | |
799f9a38 | 789 | nbd_mark_nsock_dead(nbd, nsock, 1); |
f3733247 JB |
790 | mutex_unlock(&nsock->tx_lock); |
791 | goto again; | |
792 | } | |
9dd5d3ab | 793 | out: |
9561a7ad | 794 | mutex_unlock(&nsock->tx_lock); |
5ea8d108 | 795 | nbd_config_put(nbd); |
9dd5d3ab | 796 | return ret; |
48cf6061 LV |
797 | } |
798 | ||
fc17b653 | 799 | static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx, |
fd8383fd | 800 | const struct blk_mq_queue_data *bd) |
1da177e4 | 801 | { |
fd8383fd | 802 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); |
9dd5d3ab | 803 | int ret; |
1da177e4 | 804 | |
9561a7ad JB |
805 | /* |
806 | * Since we look at the bio's to send the request over the network we | |
807 | * need to make sure the completion work doesn't mark this request done | |
808 | * before we are done doing our send. This keeps us from dereferencing | |
809 | * freed data if we have particularly fast completions (ie we get the | |
810 | * completion before we exit sock_xmit on the last bvec) or in the case | |
811 | * that the server is misbehaving (or there was an error) before we're | |
812 | * done sending everything over the wire. | |
813 | */ | |
814 | init_completion(&cmd->send_complete); | |
fd8383fd | 815 | blk_mq_start_request(bd->rq); |
9dd5d3ab JB |
816 | |
817 | /* We can be called directly from the user space process, which means we | |
818 | * could possibly have signals pending so our sendmsg will fail. In | |
819 | * this case we need to return that we are busy, otherwise error out as | |
820 | * appropriate. | |
821 | */ | |
822 | ret = nbd_handle_cmd(cmd, hctx->queue_num); | |
9561a7ad JB |
823 | complete(&cmd->send_complete); |
824 | ||
fc17b653 | 825 | return ret < 0 ? BLK_STS_IOERR : BLK_STS_OK; |
1da177e4 LT |
826 | } |
827 | ||
e46c7287 JB |
828 | static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg, |
829 | bool netlink) | |
23272a67 | 830 | { |
5ea8d108 | 831 | struct nbd_config *config = nbd->config; |
9442b739 | 832 | struct socket *sock; |
9561a7ad JB |
833 | struct nbd_sock **socks; |
834 | struct nbd_sock *nsock; | |
9442b739 JB |
835 | int err; |
836 | ||
837 | sock = sockfd_lookup(arg, &err); | |
838 | if (!sock) | |
839 | return err; | |
23272a67 | 840 | |
e46c7287 JB |
841 | if (!netlink && !nbd->task_setup && |
842 | !test_bit(NBD_BOUND, &config->runtime_flags)) | |
9561a7ad | 843 | nbd->task_setup = current; |
e46c7287 JB |
844 | |
845 | if (!netlink && | |
846 | (nbd->task_setup != current || | |
847 | test_bit(NBD_BOUND, &config->runtime_flags))) { | |
9561a7ad JB |
848 | dev_err(disk_to_dev(nbd->disk), |
849 | "Device being setup by another task"); | |
9b1355d5 | 850 | sockfd_put(sock); |
e46c7287 | 851 | return -EBUSY; |
23272a67 MP |
852 | } |
853 | ||
5ea8d108 | 854 | socks = krealloc(config->socks, (config->num_connections + 1) * |
9561a7ad | 855 | sizeof(struct nbd_sock *), GFP_KERNEL); |
9b1355d5 JB |
856 | if (!socks) { |
857 | sockfd_put(sock); | |
9561a7ad | 858 | return -ENOMEM; |
9b1355d5 | 859 | } |
9561a7ad | 860 | nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL); |
9b1355d5 JB |
861 | if (!nsock) { |
862 | sockfd_put(sock); | |
9561a7ad | 863 | return -ENOMEM; |
9b1355d5 | 864 | } |
9561a7ad | 865 | |
5ea8d108 | 866 | config->socks = socks; |
23272a67 | 867 | |
f3733247 JB |
868 | nsock->fallback_index = -1; |
869 | nsock->dead = false; | |
9561a7ad JB |
870 | mutex_init(&nsock->tx_lock); |
871 | nsock->sock = sock; | |
9dd5d3ab JB |
872 | nsock->pending = NULL; |
873 | nsock->sent = 0; | |
799f9a38 | 874 | nsock->cookie = 0; |
5ea8d108 | 875 | socks[config->num_connections++] = nsock; |
560bc4b3 | 876 | atomic_inc(&config->live_connections); |
23272a67 | 877 | |
9561a7ad | 878 | return 0; |
23272a67 MP |
879 | } |
880 | ||
b7aa3d39 JB |
881 | static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg) |
882 | { | |
883 | struct nbd_config *config = nbd->config; | |
884 | struct socket *sock, *old; | |
885 | struct recv_thread_args *args; | |
886 | int i; | |
887 | int err; | |
888 | ||
889 | sock = sockfd_lookup(arg, &err); | |
890 | if (!sock) | |
891 | return err; | |
892 | ||
893 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
894 | if (!args) { | |
895 | sockfd_put(sock); | |
896 | return -ENOMEM; | |
897 | } | |
898 | ||
899 | for (i = 0; i < config->num_connections; i++) { | |
900 | struct nbd_sock *nsock = config->socks[i]; | |
901 | ||
902 | if (!nsock->dead) | |
903 | continue; | |
904 | ||
905 | mutex_lock(&nsock->tx_lock); | |
906 | if (!nsock->dead) { | |
907 | mutex_unlock(&nsock->tx_lock); | |
908 | continue; | |
909 | } | |
910 | sk_set_memalloc(sock->sk); | |
dc88e34d | 911 | sock->sk->sk_sndtimeo = nbd->tag_set.timeout; |
b7aa3d39 JB |
912 | atomic_inc(&config->recv_threads); |
913 | refcount_inc(&nbd->config_refs); | |
914 | old = nsock->sock; | |
915 | nsock->fallback_index = -1; | |
916 | nsock->sock = sock; | |
917 | nsock->dead = false; | |
918 | INIT_WORK(&args->work, recv_work); | |
919 | args->index = i; | |
920 | args->nbd = nbd; | |
799f9a38 | 921 | nsock->cookie++; |
b7aa3d39 JB |
922 | mutex_unlock(&nsock->tx_lock); |
923 | sockfd_put(old); | |
924 | ||
925 | /* We take the tx_mutex in an error path in the recv_work, so we | |
926 | * need to queue_work outside of the tx_mutex. | |
927 | */ | |
928 | queue_work(recv_workqueue, &args->work); | |
560bc4b3 JB |
929 | |
930 | atomic_inc(&config->live_connections); | |
931 | wake_up(&config->conn_wait); | |
b7aa3d39 JB |
932 | return 0; |
933 | } | |
934 | sockfd_put(sock); | |
935 | kfree(args); | |
936 | return -ENOSPC; | |
937 | } | |
938 | ||
0e4f0f6f MP |
939 | static void nbd_bdev_reset(struct block_device *bdev) |
940 | { | |
abbbdf12 RMB |
941 | if (bdev->bd_openers > 1) |
942 | return; | |
29eaadc0 | 943 | bd_set_size(bdev, 0); |
0e4f0f6f MP |
944 | if (max_part > 0) { |
945 | blkdev_reread_part(bdev); | |
946 | bdev->bd_invalidated = 1; | |
947 | } | |
948 | } | |
949 | ||
29eaadc0 | 950 | static void nbd_parse_flags(struct nbd_device *nbd) |
d02cf531 | 951 | { |
5ea8d108 JB |
952 | struct nbd_config *config = nbd->config; |
953 | if (config->flags & NBD_FLAG_READ_ONLY) | |
29eaadc0 JB |
954 | set_disk_ro(nbd->disk, true); |
955 | else | |
956 | set_disk_ro(nbd->disk, false); | |
5ea8d108 | 957 | if (config->flags & NBD_FLAG_SEND_TRIM) |
d02cf531 | 958 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); |
685c9b24 SM |
959 | if (config->flags & NBD_FLAG_SEND_FLUSH) { |
960 | if (config->flags & NBD_FLAG_SEND_FUA) | |
961 | blk_queue_write_cache(nbd->disk->queue, true, true); | |
962 | else | |
963 | blk_queue_write_cache(nbd->disk->queue, true, false); | |
964 | } | |
d02cf531 | 965 | else |
aafb1eec | 966 | blk_queue_write_cache(nbd->disk->queue, false, false); |
d02cf531 MP |
967 | } |
968 | ||
9561a7ad JB |
969 | static void send_disconnects(struct nbd_device *nbd) |
970 | { | |
5ea8d108 | 971 | struct nbd_config *config = nbd->config; |
c9f2b6ae AV |
972 | struct nbd_request request = { |
973 | .magic = htonl(NBD_REQUEST_MAGIC), | |
974 | .type = htonl(NBD_CMD_DISC), | |
975 | }; | |
976 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
977 | struct iov_iter from; | |
9561a7ad JB |
978 | int i, ret; |
979 | ||
5ea8d108 | 980 | for (i = 0; i < config->num_connections; i++) { |
b4b2aecc JB |
981 | struct nbd_sock *nsock = config->socks[i]; |
982 | ||
c9f2b6ae | 983 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); |
b4b2aecc | 984 | mutex_lock(&nsock->tx_lock); |
9dd5d3ab | 985 | ret = sock_xmit(nbd, i, 1, &from, 0, NULL); |
9561a7ad JB |
986 | if (ret <= 0) |
987 | dev_err(disk_to_dev(nbd->disk), | |
988 | "Send disconnect failed %d\n", ret); | |
b4b2aecc | 989 | mutex_unlock(&nsock->tx_lock); |
9561a7ad JB |
990 | } |
991 | } | |
992 | ||
29eaadc0 | 993 | static int nbd_disconnect(struct nbd_device *nbd) |
9442b739 | 994 | { |
5ea8d108 | 995 | struct nbd_config *config = nbd->config; |
30d53d9c | 996 | |
5ea8d108 | 997 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); |
2e13456f JB |
998 | set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); |
999 | send_disconnects(nbd); | |
9442b739 JB |
1000 | return 0; |
1001 | } | |
1002 | ||
29eaadc0 | 1003 | static void nbd_clear_sock(struct nbd_device *nbd) |
1a2ad211 | 1004 | { |
9442b739 JB |
1005 | sock_shutdown(nbd); |
1006 | nbd_clear_que(nbd); | |
5ea8d108 | 1007 | nbd->task_setup = NULL; |
5ea8d108 JB |
1008 | } |
1009 | ||
1010 | static void nbd_config_put(struct nbd_device *nbd) | |
1011 | { | |
1012 | if (refcount_dec_and_mutex_lock(&nbd->config_refs, | |
1013 | &nbd->config_lock)) { | |
5ea8d108 | 1014 | struct nbd_config *config = nbd->config; |
5ea8d108 | 1015 | nbd_dev_dbg_close(nbd); |
29eaadc0 | 1016 | nbd_size_clear(nbd); |
5ea8d108 JB |
1017 | if (test_and_clear_bit(NBD_HAS_PID_FILE, |
1018 | &config->runtime_flags)) | |
1019 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); | |
1020 | nbd->task_recv = NULL; | |
29eaadc0 | 1021 | nbd_clear_sock(nbd); |
5ea8d108 JB |
1022 | if (config->num_connections) { |
1023 | int i; | |
1024 | for (i = 0; i < config->num_connections; i++) { | |
1025 | sockfd_put(config->socks[i]->sock); | |
1026 | kfree(config->socks[i]); | |
1027 | } | |
1028 | kfree(config->socks); | |
1029 | } | |
fa976532 | 1030 | kfree(nbd->config); |
af622b86 ID |
1031 | nbd->config = NULL; |
1032 | ||
1033 | nbd->tag_set.timeout = 0; | |
1034 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); | |
a2c97909 | 1035 | |
5ea8d108 | 1036 | mutex_unlock(&nbd->config_lock); |
c6a4759e | 1037 | nbd_put(nbd); |
5ea8d108 JB |
1038 | module_put(THIS_MODULE); |
1039 | } | |
9442b739 JB |
1040 | } |
1041 | ||
e46c7287 | 1042 | static int nbd_start_device(struct nbd_device *nbd) |
9442b739 | 1043 | { |
5ea8d108 JB |
1044 | struct nbd_config *config = nbd->config; |
1045 | int num_connections = config->num_connections; | |
9442b739 | 1046 | int error = 0, i; |
1a2ad211 | 1047 | |
9442b739 JB |
1048 | if (nbd->task_recv) |
1049 | return -EBUSY; | |
5ea8d108 | 1050 | if (!config->socks) |
9442b739 JB |
1051 | return -EINVAL; |
1052 | if (num_connections > 1 && | |
5ea8d108 | 1053 | !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) { |
9442b739 | 1054 | dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n"); |
5ea8d108 | 1055 | return -EINVAL; |
9442b739 | 1056 | } |
23272a67 | 1057 | |
5ea8d108 | 1058 | blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections); |
9442b739 | 1059 | nbd->task_recv = current; |
23272a67 | 1060 | |
29eaadc0 | 1061 | nbd_parse_flags(nbd); |
23272a67 | 1062 | |
9442b739 JB |
1063 | error = device_create_file(disk_to_dev(nbd->disk), &pid_attr); |
1064 | if (error) { | |
1065 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); | |
5ea8d108 | 1066 | return error; |
1a2ad211 | 1067 | } |
29eaadc0 | 1068 | set_bit(NBD_HAS_PID_FILE, &config->runtime_flags); |
37091fdd | 1069 | |
9442b739 JB |
1070 | nbd_dev_dbg_init(nbd); |
1071 | for (i = 0; i < num_connections; i++) { | |
5ea8d108 JB |
1072 | struct recv_thread_args *args; |
1073 | ||
1074 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
1075 | if (!args) { | |
1076 | sock_shutdown(nbd); | |
1077 | return -ENOMEM; | |
1078 | } | |
1079 | sk_set_memalloc(config->socks[i]->sock->sk); | |
dc88e34d | 1080 | config->socks[i]->sock->sk->sk_sndtimeo = nbd->tag_set.timeout; |
5ea8d108 JB |
1081 | atomic_inc(&config->recv_threads); |
1082 | refcount_inc(&nbd->config_refs); | |
1083 | INIT_WORK(&args->work, recv_work); | |
1084 | args->nbd = nbd; | |
1085 | args->index = i; | |
1086 | queue_work(recv_workqueue, &args->work); | |
37091fdd | 1087 | } |
e46c7287 JB |
1088 | return error; |
1089 | } | |
1090 | ||
1091 | static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev) | |
1092 | { | |
1093 | struct nbd_config *config = nbd->config; | |
1094 | int ret; | |
1095 | ||
1096 | ret = nbd_start_device(nbd); | |
1097 | if (ret) | |
1098 | return ret; | |
1099 | ||
1100 | bd_set_size(bdev, config->bytesize); | |
1101 | if (max_part) | |
1102 | bdev->bd_invalidated = 1; | |
1103 | mutex_unlock(&nbd->config_lock); | |
1104 | ret = wait_event_interruptible(config->recv_wq, | |
5ea8d108 | 1105 | atomic_read(&config->recv_threads) == 0); |
e46c7287 | 1106 | if (ret) |
5ea8d108 | 1107 | sock_shutdown(nbd); |
9442b739 | 1108 | mutex_lock(&nbd->config_lock); |
e46c7287 | 1109 | bd_set_size(bdev, 0); |
9442b739 | 1110 | /* user requested, ignore socket errors */ |
5ea8d108 | 1111 | if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags)) |
e46c7287 | 1112 | ret = 0; |
5ea8d108 | 1113 | if (test_bit(NBD_TIMEDOUT, &config->runtime_flags)) |
e46c7287 JB |
1114 | ret = -ETIMEDOUT; |
1115 | return ret; | |
9442b739 JB |
1116 | } |
1117 | ||
29eaadc0 JB |
1118 | static void nbd_clear_sock_ioctl(struct nbd_device *nbd, |
1119 | struct block_device *bdev) | |
1120 | { | |
2516ab15 | 1121 | sock_shutdown(nbd); |
29eaadc0 JB |
1122 | kill_bdev(bdev); |
1123 | nbd_bdev_reset(bdev); | |
e46c7287 JB |
1124 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, |
1125 | &nbd->config->runtime_flags)) | |
1126 | nbd_config_put(nbd); | |
29eaadc0 JB |
1127 | } |
1128 | ||
9442b739 JB |
1129 | /* Must be called with config_lock held */ |
1130 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, | |
1131 | unsigned int cmd, unsigned long arg) | |
1132 | { | |
5ea8d108 JB |
1133 | struct nbd_config *config = nbd->config; |
1134 | ||
9442b739 JB |
1135 | switch (cmd) { |
1136 | case NBD_DISCONNECT: | |
29eaadc0 | 1137 | return nbd_disconnect(nbd); |
9442b739 | 1138 | case NBD_CLEAR_SOCK: |
29eaadc0 JB |
1139 | nbd_clear_sock_ioctl(nbd, bdev); |
1140 | return 0; | |
9442b739 | 1141 | case NBD_SET_SOCK: |
e46c7287 | 1142 | return nbd_add_socket(nbd, arg, false); |
9442b739 | 1143 | case NBD_SET_BLKSIZE: |
29eaadc0 | 1144 | nbd_size_set(nbd, arg, |
5ea8d108 | 1145 | div_s64(config->bytesize, arg)); |
e544541b | 1146 | return 0; |
1da177e4 | 1147 | case NBD_SET_SIZE: |
29eaadc0 | 1148 | nbd_size_set(nbd, config->blksize, |
5ea8d108 | 1149 | div_s64(arg, config->blksize)); |
e544541b | 1150 | return 0; |
37091fdd | 1151 | case NBD_SET_SIZE_BLOCKS: |
29eaadc0 | 1152 | nbd_size_set(nbd, config->blksize, arg); |
e544541b | 1153 | return 0; |
7fdfd406 | 1154 | case NBD_SET_TIMEOUT: |
f8586855 JB |
1155 | if (arg) { |
1156 | nbd->tag_set.timeout = arg * HZ; | |
1157 | blk_queue_rq_timeout(nbd->disk->queue, arg * HZ); | |
1158 | } | |
7fdfd406 | 1159 | return 0; |
1a2ad211 | 1160 | |
2f012508 | 1161 | case NBD_SET_FLAGS: |
5ea8d108 | 1162 | config->flags = arg; |
2f012508 | 1163 | return 0; |
9442b739 | 1164 | case NBD_DO_IT: |
e46c7287 | 1165 | return nbd_start_device_ioctl(nbd, bdev); |
1da177e4 | 1166 | case NBD_CLEAR_QUE: |
4b2f0260 HX |
1167 | /* |
1168 | * This is for compatibility only. The queue is always cleared | |
1169 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
1170 | */ | |
1da177e4 LT |
1171 | return 0; |
1172 | case NBD_PRINT_DEBUG: | |
fd8383fd JB |
1173 | /* |
1174 | * For compatibility only, we no longer keep a list of | |
1175 | * outstanding requests. | |
1176 | */ | |
1da177e4 LT |
1177 | return 0; |
1178 | } | |
1a2ad211 PM |
1179 | return -ENOTTY; |
1180 | } | |
1181 | ||
1182 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
1183 | unsigned int cmd, unsigned long arg) | |
1184 | { | |
f4507164 | 1185 | struct nbd_device *nbd = bdev->bd_disk->private_data; |
e46c7287 JB |
1186 | struct nbd_config *config = nbd->config; |
1187 | int error = -EINVAL; | |
1a2ad211 PM |
1188 | |
1189 | if (!capable(CAP_SYS_ADMIN)) | |
1190 | return -EPERM; | |
1191 | ||
9561a7ad | 1192 | mutex_lock(&nbd->config_lock); |
e46c7287 JB |
1193 | |
1194 | /* Don't allow ioctl operations on a nbd device that was created with | |
1195 | * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine. | |
1196 | */ | |
1197 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1198 | (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK)) | |
1199 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
1200 | else | |
1201 | dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n"); | |
9561a7ad | 1202 | mutex_unlock(&nbd->config_lock); |
1a2ad211 | 1203 | return error; |
1da177e4 LT |
1204 | } |
1205 | ||
5ea8d108 JB |
1206 | static struct nbd_config *nbd_alloc_config(void) |
1207 | { | |
1208 | struct nbd_config *config; | |
1209 | ||
1210 | config = kzalloc(sizeof(struct nbd_config), GFP_NOFS); | |
1211 | if (!config) | |
1212 | return NULL; | |
1213 | atomic_set(&config->recv_threads, 0); | |
1214 | init_waitqueue_head(&config->recv_wq); | |
560bc4b3 | 1215 | init_waitqueue_head(&config->conn_wait); |
5ea8d108 | 1216 | config->blksize = 1024; |
560bc4b3 | 1217 | atomic_set(&config->live_connections, 0); |
5ea8d108 JB |
1218 | try_module_get(THIS_MODULE); |
1219 | return config; | |
1220 | } | |
1221 | ||
1222 | static int nbd_open(struct block_device *bdev, fmode_t mode) | |
1223 | { | |
1224 | struct nbd_device *nbd; | |
1225 | int ret = 0; | |
1226 | ||
1227 | mutex_lock(&nbd_index_mutex); | |
1228 | nbd = bdev->bd_disk->private_data; | |
1229 | if (!nbd) { | |
1230 | ret = -ENXIO; | |
1231 | goto out; | |
1232 | } | |
c6a4759e JB |
1233 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1234 | ret = -ENXIO; | |
1235 | goto out; | |
1236 | } | |
5ea8d108 JB |
1237 | if (!refcount_inc_not_zero(&nbd->config_refs)) { |
1238 | struct nbd_config *config; | |
1239 | ||
1240 | mutex_lock(&nbd->config_lock); | |
1241 | if (refcount_inc_not_zero(&nbd->config_refs)) { | |
1242 | mutex_unlock(&nbd->config_lock); | |
1243 | goto out; | |
1244 | } | |
1245 | config = nbd->config = nbd_alloc_config(); | |
1246 | if (!config) { | |
1247 | ret = -ENOMEM; | |
1248 | mutex_unlock(&nbd->config_lock); | |
1249 | goto out; | |
1250 | } | |
1251 | refcount_set(&nbd->config_refs, 1); | |
c6a4759e | 1252 | refcount_inc(&nbd->refs); |
5ea8d108 JB |
1253 | mutex_unlock(&nbd->config_lock); |
1254 | } | |
1255 | out: | |
1256 | mutex_unlock(&nbd_index_mutex); | |
1257 | return ret; | |
1258 | } | |
1259 | ||
1260 | static void nbd_release(struct gendisk *disk, fmode_t mode) | |
1261 | { | |
1262 | struct nbd_device *nbd = disk->private_data; | |
1263 | nbd_config_put(nbd); | |
c6a4759e | 1264 | nbd_put(nbd); |
5ea8d108 JB |
1265 | } |
1266 | ||
83d5cde4 | 1267 | static const struct block_device_operations nbd_fops = |
1da177e4 LT |
1268 | { |
1269 | .owner = THIS_MODULE, | |
5ea8d108 JB |
1270 | .open = nbd_open, |
1271 | .release = nbd_release, | |
8a6cfeb6 | 1272 | .ioctl = nbd_ioctl, |
263a3df1 | 1273 | .compat_ioctl = nbd_ioctl, |
1da177e4 LT |
1274 | }; |
1275 | ||
30d53d9c MP |
1276 | #if IS_ENABLED(CONFIG_DEBUG_FS) |
1277 | ||
1278 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
1279 | { | |
1280 | struct nbd_device *nbd = s->private; | |
1281 | ||
1282 | if (nbd->task_recv) | |
1283 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
30d53d9c MP |
1284 | |
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
1289 | { | |
1290 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
1291 | } | |
1292 | ||
1293 | static const struct file_operations nbd_dbg_tasks_ops = { | |
1294 | .open = nbd_dbg_tasks_open, | |
1295 | .read = seq_read, | |
1296 | .llseek = seq_lseek, | |
1297 | .release = single_release, | |
1298 | }; | |
1299 | ||
1300 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
1301 | { | |
1302 | struct nbd_device *nbd = s->private; | |
5ea8d108 | 1303 | u32 flags = nbd->config->flags; |
30d53d9c MP |
1304 | |
1305 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
1306 | ||
1307 | seq_puts(s, "Known flags:\n"); | |
1308 | ||
1309 | if (flags & NBD_FLAG_HAS_FLAGS) | |
1310 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
1311 | if (flags & NBD_FLAG_READ_ONLY) | |
1312 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
1313 | if (flags & NBD_FLAG_SEND_FLUSH) | |
1314 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
685c9b24 SM |
1315 | if (flags & NBD_FLAG_SEND_FUA) |
1316 | seq_puts(s, "NBD_FLAG_SEND_FUA\n"); | |
30d53d9c MP |
1317 | if (flags & NBD_FLAG_SEND_TRIM) |
1318 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
1319 | ||
1320 | return 0; | |
1321 | } | |
1322 | ||
1323 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
1324 | { | |
1325 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
1326 | } | |
1327 | ||
1328 | static const struct file_operations nbd_dbg_flags_ops = { | |
1329 | .open = nbd_dbg_flags_open, | |
1330 | .read = seq_read, | |
1331 | .llseek = seq_lseek, | |
1332 | .release = single_release, | |
1333 | }; | |
1334 | ||
1335 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1336 | { | |
1337 | struct dentry *dir; | |
5ea8d108 | 1338 | struct nbd_config *config = nbd->config; |
27ea43fe MP |
1339 | |
1340 | if (!nbd_dbg_dir) | |
1341 | return -EIO; | |
30d53d9c MP |
1342 | |
1343 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
27ea43fe MP |
1344 | if (!dir) { |
1345 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n", | |
1346 | nbd_name(nbd)); | |
1347 | return -EIO; | |
30d53d9c | 1348 | } |
5ea8d108 | 1349 | config->dbg_dir = dir; |
30d53d9c | 1350 | |
27ea43fe | 1351 | debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); |
5ea8d108 | 1352 | debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize); |
0eadf37a | 1353 | debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout); |
5ea8d108 | 1354 | debugfs_create_u64("blocksize", 0444, dir, &config->blksize); |
d366a0ff | 1355 | debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops); |
30d53d9c MP |
1356 | |
1357 | return 0; | |
1358 | } | |
1359 | ||
1360 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1361 | { | |
5ea8d108 | 1362 | debugfs_remove_recursive(nbd->config->dbg_dir); |
30d53d9c MP |
1363 | } |
1364 | ||
1365 | static int nbd_dbg_init(void) | |
1366 | { | |
1367 | struct dentry *dbg_dir; | |
1368 | ||
1369 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
27ea43fe MP |
1370 | if (!dbg_dir) |
1371 | return -EIO; | |
30d53d9c MP |
1372 | |
1373 | nbd_dbg_dir = dbg_dir; | |
1374 | ||
1375 | return 0; | |
1376 | } | |
1377 | ||
1378 | static void nbd_dbg_close(void) | |
1379 | { | |
1380 | debugfs_remove_recursive(nbd_dbg_dir); | |
1381 | } | |
1382 | ||
1383 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
1384 | ||
1385 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1386 | { | |
1387 | return 0; | |
1388 | } | |
1389 | ||
1390 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1391 | { | |
1392 | } | |
1393 | ||
1394 | static int nbd_dbg_init(void) | |
1395 | { | |
1396 | return 0; | |
1397 | } | |
1398 | ||
1399 | static void nbd_dbg_close(void) | |
1400 | { | |
1401 | } | |
1402 | ||
1403 | #endif | |
1404 | ||
d6296d39 CH |
1405 | static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq, |
1406 | unsigned int hctx_idx, unsigned int numa_node) | |
fd8383fd JB |
1407 | { |
1408 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
d6296d39 | 1409 | cmd->nbd = set->driver_data; |
fd8383fd JB |
1410 | return 0; |
1411 | } | |
1412 | ||
f363b089 | 1413 | static const struct blk_mq_ops nbd_mq_ops = { |
fd8383fd | 1414 | .queue_rq = nbd_queue_rq, |
1e388ae0 | 1415 | .complete = nbd_complete_rq, |
fd8383fd | 1416 | .init_request = nbd_init_request, |
0eadf37a | 1417 | .timeout = nbd_xmit_timeout, |
fd8383fd JB |
1418 | }; |
1419 | ||
b0d9111a JB |
1420 | static int nbd_dev_add(int index) |
1421 | { | |
1422 | struct nbd_device *nbd; | |
1423 | struct gendisk *disk; | |
1424 | struct request_queue *q; | |
1425 | int err = -ENOMEM; | |
1426 | ||
1427 | nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL); | |
1428 | if (!nbd) | |
1429 | goto out; | |
1430 | ||
1431 | disk = alloc_disk(1 << part_shift); | |
1432 | if (!disk) | |
1433 | goto out_free_nbd; | |
1434 | ||
1435 | if (index >= 0) { | |
1436 | err = idr_alloc(&nbd_index_idr, nbd, index, index + 1, | |
1437 | GFP_KERNEL); | |
1438 | if (err == -ENOSPC) | |
1439 | err = -EEXIST; | |
1440 | } else { | |
1441 | err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL); | |
1442 | if (err >= 0) | |
1443 | index = err; | |
1444 | } | |
1445 | if (err < 0) | |
1446 | goto out_free_disk; | |
1447 | ||
e46c7287 | 1448 | nbd->index = index; |
b0d9111a JB |
1449 | nbd->disk = disk; |
1450 | nbd->tag_set.ops = &nbd_mq_ops; | |
1451 | nbd->tag_set.nr_hw_queues = 1; | |
1452 | nbd->tag_set.queue_depth = 128; | |
1453 | nbd->tag_set.numa_node = NUMA_NO_NODE; | |
1454 | nbd->tag_set.cmd_size = sizeof(struct nbd_cmd); | |
1455 | nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | | |
1456 | BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING; | |
1457 | nbd->tag_set.driver_data = nbd; | |
1458 | ||
1459 | err = blk_mq_alloc_tag_set(&nbd->tag_set); | |
1460 | if (err) | |
1461 | goto out_free_idr; | |
1462 | ||
1463 | q = blk_mq_init_queue(&nbd->tag_set); | |
1464 | if (IS_ERR(q)) { | |
1465 | err = PTR_ERR(q); | |
1466 | goto out_free_tags; | |
1467 | } | |
1468 | disk->queue = q; | |
1469 | ||
1470 | /* | |
1471 | * Tell the block layer that we are not a rotational device | |
1472 | */ | |
1473 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
1474 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); | |
1475 | disk->queue->limits.discard_granularity = 512; | |
1476 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); | |
ebb16d0d | 1477 | blk_queue_max_segment_size(disk->queue, UINT_MAX); |
1cc1f17a | 1478 | blk_queue_max_segments(disk->queue, USHRT_MAX); |
b0d9111a JB |
1479 | blk_queue_max_hw_sectors(disk->queue, 65536); |
1480 | disk->queue->limits.max_sectors = 256; | |
1481 | ||
b0d9111a | 1482 | mutex_init(&nbd->config_lock); |
5ea8d108 | 1483 | refcount_set(&nbd->config_refs, 0); |
c6a4759e JB |
1484 | refcount_set(&nbd->refs, 1); |
1485 | INIT_LIST_HEAD(&nbd->list); | |
b0d9111a JB |
1486 | disk->major = NBD_MAJOR; |
1487 | disk->first_minor = index << part_shift; | |
1488 | disk->fops = &nbd_fops; | |
1489 | disk->private_data = nbd; | |
1490 | sprintf(disk->disk_name, "nbd%d", index); | |
b0d9111a | 1491 | add_disk(disk); |
47d902b9 | 1492 | nbd_total_devices++; |
b0d9111a JB |
1493 | return index; |
1494 | ||
1495 | out_free_tags: | |
1496 | blk_mq_free_tag_set(&nbd->tag_set); | |
1497 | out_free_idr: | |
1498 | idr_remove(&nbd_index_idr, index); | |
1499 | out_free_disk: | |
1500 | put_disk(disk); | |
1501 | out_free_nbd: | |
1502 | kfree(nbd); | |
1503 | out: | |
1504 | return err; | |
1505 | } | |
1506 | ||
e46c7287 JB |
1507 | static int find_free_cb(int id, void *ptr, void *data) |
1508 | { | |
1509 | struct nbd_device *nbd = ptr; | |
1510 | struct nbd_device **found = data; | |
1511 | ||
1512 | if (!refcount_read(&nbd->config_refs)) { | |
1513 | *found = nbd; | |
1514 | return 1; | |
1515 | } | |
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | /* Netlink interface. */ | |
1520 | static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = { | |
1521 | [NBD_ATTR_INDEX] = { .type = NLA_U32 }, | |
1522 | [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 }, | |
1523 | [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 }, | |
1524 | [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 }, | |
1525 | [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 }, | |
1526 | [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 }, | |
1527 | [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED}, | |
560bc4b3 | 1528 | [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 }, |
47d902b9 | 1529 | [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED}, |
e46c7287 JB |
1530 | }; |
1531 | ||
1532 | static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = { | |
1533 | [NBD_SOCK_FD] = { .type = NLA_U32 }, | |
1534 | }; | |
1535 | ||
47d902b9 JB |
1536 | /* We don't use this right now since we don't parse the incoming list, but we |
1537 | * still want it here so userspace knows what to expect. | |
1538 | */ | |
1539 | static struct nla_policy __attribute__((unused)) | |
1540 | nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = { | |
1541 | [NBD_DEVICE_INDEX] = { .type = NLA_U32 }, | |
1542 | [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 }, | |
1543 | }; | |
1544 | ||
e46c7287 JB |
1545 | static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info) |
1546 | { | |
1547 | struct nbd_device *nbd = NULL; | |
1548 | struct nbd_config *config; | |
1549 | int index = -1; | |
1550 | int ret; | |
a2c97909 | 1551 | bool put_dev = false; |
e46c7287 JB |
1552 | |
1553 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1554 | return -EPERM; | |
1555 | ||
1556 | if (info->attrs[NBD_ATTR_INDEX]) | |
1557 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1558 | if (!info->attrs[NBD_ATTR_SOCKETS]) { | |
1559 | printk(KERN_ERR "nbd: must specify at least one socket\n"); | |
1560 | return -EINVAL; | |
1561 | } | |
1562 | if (!info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1563 | printk(KERN_ERR "nbd: must specify a size in bytes for the device\n"); | |
1564 | return -EINVAL; | |
1565 | } | |
1566 | again: | |
1567 | mutex_lock(&nbd_index_mutex); | |
1568 | if (index == -1) { | |
1569 | ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd); | |
1570 | if (ret == 0) { | |
1571 | int new_index; | |
1572 | new_index = nbd_dev_add(-1); | |
1573 | if (new_index < 0) { | |
1574 | mutex_unlock(&nbd_index_mutex); | |
1575 | printk(KERN_ERR "nbd: failed to add new device\n"); | |
1576 | return ret; | |
1577 | } | |
1578 | nbd = idr_find(&nbd_index_idr, new_index); | |
1579 | } | |
1580 | } else { | |
1581 | nbd = idr_find(&nbd_index_idr, index); | |
1582 | } | |
e46c7287 JB |
1583 | if (!nbd) { |
1584 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", | |
1585 | index); | |
c6a4759e JB |
1586 | mutex_unlock(&nbd_index_mutex); |
1587 | return -EINVAL; | |
1588 | } | |
1589 | if (!refcount_inc_not_zero(&nbd->refs)) { | |
1590 | mutex_unlock(&nbd_index_mutex); | |
1591 | if (index == -1) | |
1592 | goto again; | |
1593 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1594 | index); | |
e46c7287 JB |
1595 | return -EINVAL; |
1596 | } | |
c6a4759e | 1597 | mutex_unlock(&nbd_index_mutex); |
e46c7287 JB |
1598 | |
1599 | mutex_lock(&nbd->config_lock); | |
1600 | if (refcount_read(&nbd->config_refs)) { | |
1601 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1602 | nbd_put(nbd); |
e46c7287 JB |
1603 | if (index == -1) |
1604 | goto again; | |
1605 | printk(KERN_ERR "nbd: nbd%d already in use\n", index); | |
1606 | return -EBUSY; | |
1607 | } | |
1608 | if (WARN_ON(nbd->config)) { | |
1609 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1610 | nbd_put(nbd); |
e46c7287 JB |
1611 | return -EINVAL; |
1612 | } | |
1613 | config = nbd->config = nbd_alloc_config(); | |
1614 | if (!nbd->config) { | |
1615 | mutex_unlock(&nbd->config_lock); | |
c6a4759e | 1616 | nbd_put(nbd); |
e46c7287 JB |
1617 | printk(KERN_ERR "nbd: couldn't allocate config\n"); |
1618 | return -ENOMEM; | |
1619 | } | |
1620 | refcount_set(&nbd->config_refs, 1); | |
1621 | set_bit(NBD_BOUND, &config->runtime_flags); | |
1622 | ||
1623 | if (info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1624 | u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]); | |
1625 | nbd_size_set(nbd, config->blksize, | |
1626 | div64_u64(bytes, config->blksize)); | |
1627 | } | |
1628 | if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) { | |
1629 | u64 bsize = | |
1630 | nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]); | |
1631 | nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize)); | |
1632 | } | |
1633 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1634 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1635 | nbd->tag_set.timeout = timeout * HZ; | |
1636 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1637 | } | |
560bc4b3 JB |
1638 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1639 | config->dead_conn_timeout = | |
1640 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1641 | config->dead_conn_timeout *= HZ; | |
1642 | } | |
e46c7287 JB |
1643 | if (info->attrs[NBD_ATTR_SERVER_FLAGS]) |
1644 | config->flags = | |
1645 | nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]); | |
a2c97909 JB |
1646 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { |
1647 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1648 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1649 | set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1650 | &config->runtime_flags); | |
1651 | put_dev = true; | |
1652 | } | |
1653 | } | |
1654 | ||
e46c7287 JB |
1655 | if (info->attrs[NBD_ATTR_SOCKETS]) { |
1656 | struct nlattr *attr; | |
1657 | int rem, fd; | |
1658 | ||
1659 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1660 | rem) { | |
1661 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1662 | ||
1663 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1664 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1665 | ret = -EINVAL; | |
1666 | goto out; | |
1667 | } | |
1668 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
8d65b08d | 1669 | nbd_sock_policy, info->extack); |
e46c7287 JB |
1670 | if (ret != 0) { |
1671 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1672 | ret = -EINVAL; | |
1673 | goto out; | |
1674 | } | |
1675 | if (!socks[NBD_SOCK_FD]) | |
1676 | continue; | |
1677 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1678 | ret = nbd_add_socket(nbd, fd, true); | |
1679 | if (ret) | |
1680 | goto out; | |
1681 | } | |
1682 | } | |
1683 | ret = nbd_start_device(nbd); | |
1684 | out: | |
1685 | mutex_unlock(&nbd->config_lock); | |
1686 | if (!ret) { | |
1687 | set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags); | |
1688 | refcount_inc(&nbd->config_refs); | |
1689 | nbd_connect_reply(info, nbd->index); | |
1690 | } | |
1691 | nbd_config_put(nbd); | |
a2c97909 JB |
1692 | if (put_dev) |
1693 | nbd_put(nbd); | |
e46c7287 JB |
1694 | return ret; |
1695 | } | |
1696 | ||
1697 | static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info) | |
1698 | { | |
1699 | struct nbd_device *nbd; | |
1700 | int index; | |
1701 | ||
1702 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1703 | return -EPERM; | |
1704 | ||
1705 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1706 | printk(KERN_ERR "nbd: must specify an index to disconnect\n"); | |
1707 | return -EINVAL; | |
1708 | } | |
1709 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1710 | mutex_lock(&nbd_index_mutex); | |
1711 | nbd = idr_find(&nbd_index_idr, index); | |
e46c7287 | 1712 | if (!nbd) { |
c6a4759e | 1713 | mutex_unlock(&nbd_index_mutex); |
e46c7287 JB |
1714 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", |
1715 | index); | |
1716 | return -EINVAL; | |
1717 | } | |
c6a4759e JB |
1718 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1719 | mutex_unlock(&nbd_index_mutex); | |
1720 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1721 | index); | |
1722 | return -EINVAL; | |
1723 | } | |
1724 | mutex_unlock(&nbd_index_mutex); | |
1725 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1726 | nbd_put(nbd); | |
e46c7287 | 1727 | return 0; |
c6a4759e | 1728 | } |
e46c7287 JB |
1729 | mutex_lock(&nbd->config_lock); |
1730 | nbd_disconnect(nbd); | |
1731 | mutex_unlock(&nbd->config_lock); | |
1732 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, | |
1733 | &nbd->config->runtime_flags)) | |
1734 | nbd_config_put(nbd); | |
1735 | nbd_config_put(nbd); | |
c6a4759e | 1736 | nbd_put(nbd); |
e46c7287 JB |
1737 | return 0; |
1738 | } | |
1739 | ||
b7aa3d39 JB |
1740 | static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info) |
1741 | { | |
1742 | struct nbd_device *nbd = NULL; | |
1743 | struct nbd_config *config; | |
1744 | int index; | |
1745 | int ret = -EINVAL; | |
a2c97909 | 1746 | bool put_dev = false; |
b7aa3d39 JB |
1747 | |
1748 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1749 | return -EPERM; | |
1750 | ||
1751 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1752 | printk(KERN_ERR "nbd: must specify a device to reconfigure\n"); | |
1753 | return -EINVAL; | |
1754 | } | |
1755 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1756 | mutex_lock(&nbd_index_mutex); | |
1757 | nbd = idr_find(&nbd_index_idr, index); | |
b7aa3d39 | 1758 | if (!nbd) { |
c6a4759e | 1759 | mutex_unlock(&nbd_index_mutex); |
b7aa3d39 JB |
1760 | printk(KERN_ERR "nbd: couldn't find a device at index %d\n", |
1761 | index); | |
1762 | return -EINVAL; | |
1763 | } | |
c6a4759e JB |
1764 | if (!refcount_inc_not_zero(&nbd->refs)) { |
1765 | mutex_unlock(&nbd_index_mutex); | |
1766 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1767 | index); | |
1768 | return -EINVAL; | |
1769 | } | |
1770 | mutex_unlock(&nbd_index_mutex); | |
b7aa3d39 JB |
1771 | |
1772 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1773 | dev_err(nbd_to_dev(nbd), | |
1774 | "not configured, cannot reconfigure\n"); | |
c6a4759e | 1775 | nbd_put(nbd); |
b7aa3d39 JB |
1776 | return -EINVAL; |
1777 | } | |
1778 | ||
1779 | mutex_lock(&nbd->config_lock); | |
1780 | config = nbd->config; | |
1781 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1782 | !nbd->task_recv) { | |
1783 | dev_err(nbd_to_dev(nbd), | |
1784 | "not configured, cannot reconfigure\n"); | |
1785 | goto out; | |
1786 | } | |
1787 | ||
1788 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1789 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1790 | nbd->tag_set.timeout = timeout * HZ; | |
1791 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1792 | } | |
560bc4b3 JB |
1793 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { |
1794 | config->dead_conn_timeout = | |
1795 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1796 | config->dead_conn_timeout *= HZ; | |
1797 | } | |
a2c97909 JB |
1798 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { |
1799 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1800 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1801 | if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1802 | &config->runtime_flags)) | |
1803 | put_dev = true; | |
1804 | } else { | |
1805 | if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT, | |
1806 | &config->runtime_flags)) | |
1807 | refcount_inc(&nbd->refs); | |
1808 | } | |
1809 | } | |
b7aa3d39 JB |
1810 | |
1811 | if (info->attrs[NBD_ATTR_SOCKETS]) { | |
1812 | struct nlattr *attr; | |
1813 | int rem, fd; | |
1814 | ||
1815 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1816 | rem) { | |
1817 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1818 | ||
1819 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1820 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1821 | ret = -EINVAL; | |
1822 | goto out; | |
1823 | } | |
1824 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
8d65b08d | 1825 | nbd_sock_policy, info->extack); |
b7aa3d39 JB |
1826 | if (ret != 0) { |
1827 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1828 | ret = -EINVAL; | |
1829 | goto out; | |
1830 | } | |
1831 | if (!socks[NBD_SOCK_FD]) | |
1832 | continue; | |
1833 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1834 | ret = nbd_reconnect_socket(nbd, fd); | |
1835 | if (ret) { | |
1836 | if (ret == -ENOSPC) | |
1837 | ret = 0; | |
1838 | goto out; | |
1839 | } | |
1840 | dev_info(nbd_to_dev(nbd), "reconnected socket\n"); | |
1841 | } | |
1842 | } | |
1843 | out: | |
1844 | mutex_unlock(&nbd->config_lock); | |
1845 | nbd_config_put(nbd); | |
c6a4759e | 1846 | nbd_put(nbd); |
a2c97909 JB |
1847 | if (put_dev) |
1848 | nbd_put(nbd); | |
b7aa3d39 JB |
1849 | return ret; |
1850 | } | |
1851 | ||
e46c7287 JB |
1852 | static const struct genl_ops nbd_connect_genl_ops[] = { |
1853 | { | |
1854 | .cmd = NBD_CMD_CONNECT, | |
1855 | .policy = nbd_attr_policy, | |
1856 | .doit = nbd_genl_connect, | |
1857 | }, | |
1858 | { | |
1859 | .cmd = NBD_CMD_DISCONNECT, | |
1860 | .policy = nbd_attr_policy, | |
1861 | .doit = nbd_genl_disconnect, | |
1862 | }, | |
b7aa3d39 JB |
1863 | { |
1864 | .cmd = NBD_CMD_RECONFIGURE, | |
1865 | .policy = nbd_attr_policy, | |
1866 | .doit = nbd_genl_reconfigure, | |
1867 | }, | |
47d902b9 JB |
1868 | { |
1869 | .cmd = NBD_CMD_STATUS, | |
1870 | .policy = nbd_attr_policy, | |
1871 | .doit = nbd_genl_status, | |
1872 | }, | |
e46c7287 JB |
1873 | }; |
1874 | ||
799f9a38 JB |
1875 | static const struct genl_multicast_group nbd_mcast_grps[] = { |
1876 | { .name = NBD_GENL_MCAST_GROUP_NAME, }, | |
1877 | }; | |
1878 | ||
e46c7287 JB |
1879 | static struct genl_family nbd_genl_family __ro_after_init = { |
1880 | .hdrsize = 0, | |
1881 | .name = NBD_GENL_FAMILY_NAME, | |
1882 | .version = NBD_GENL_VERSION, | |
1883 | .module = THIS_MODULE, | |
1884 | .ops = nbd_connect_genl_ops, | |
1885 | .n_ops = ARRAY_SIZE(nbd_connect_genl_ops), | |
1886 | .maxattr = NBD_ATTR_MAX, | |
799f9a38 JB |
1887 | .mcgrps = nbd_mcast_grps, |
1888 | .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps), | |
e46c7287 JB |
1889 | }; |
1890 | ||
47d902b9 JB |
1891 | static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply) |
1892 | { | |
1893 | struct nlattr *dev_opt; | |
1894 | u8 connected = 0; | |
1895 | int ret; | |
1896 | ||
1897 | /* This is a little racey, but for status it's ok. The | |
1898 | * reason we don't take a ref here is because we can't | |
1899 | * take a ref in the index == -1 case as we would need | |
1900 | * to put under the nbd_index_mutex, which could | |
1901 | * deadlock if we are configured to remove ourselves | |
1902 | * once we're disconnected. | |
1903 | */ | |
1904 | if (refcount_read(&nbd->config_refs)) | |
1905 | connected = 1; | |
1906 | dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM); | |
1907 | if (!dev_opt) | |
1908 | return -EMSGSIZE; | |
1909 | ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index); | |
1910 | if (ret) | |
1911 | return -EMSGSIZE; | |
1912 | ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED, | |
1913 | connected); | |
1914 | if (ret) | |
1915 | return -EMSGSIZE; | |
1916 | nla_nest_end(reply, dev_opt); | |
1917 | return 0; | |
1918 | } | |
1919 | ||
1920 | static int status_cb(int id, void *ptr, void *data) | |
1921 | { | |
1922 | struct nbd_device *nbd = ptr; | |
1923 | return populate_nbd_status(nbd, (struct sk_buff *)data); | |
1924 | } | |
1925 | ||
1926 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info) | |
1927 | { | |
1928 | struct nlattr *dev_list; | |
1929 | struct sk_buff *reply; | |
1930 | void *reply_head; | |
1931 | size_t msg_size; | |
1932 | int index = -1; | |
1933 | int ret = -ENOMEM; | |
1934 | ||
1935 | if (info->attrs[NBD_ATTR_INDEX]) | |
1936 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1937 | ||
1938 | mutex_lock(&nbd_index_mutex); | |
1939 | ||
1940 | msg_size = nla_total_size(nla_attr_size(sizeof(u32)) + | |
1941 | nla_attr_size(sizeof(u8))); | |
1942 | msg_size *= (index == -1) ? nbd_total_devices : 1; | |
1943 | ||
1944 | reply = genlmsg_new(msg_size, GFP_KERNEL); | |
1945 | if (!reply) | |
1946 | goto out; | |
1947 | reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0, | |
1948 | NBD_CMD_STATUS); | |
1949 | if (!reply_head) { | |
1950 | nlmsg_free(reply); | |
1951 | goto out; | |
1952 | } | |
1953 | ||
1954 | dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST); | |
1955 | if (index == -1) { | |
1956 | ret = idr_for_each(&nbd_index_idr, &status_cb, reply); | |
1957 | if (ret) { | |
1958 | nlmsg_free(reply); | |
1959 | goto out; | |
1960 | } | |
1961 | } else { | |
1962 | struct nbd_device *nbd; | |
1963 | nbd = idr_find(&nbd_index_idr, index); | |
1964 | if (nbd) { | |
1965 | ret = populate_nbd_status(nbd, reply); | |
1966 | if (ret) { | |
1967 | nlmsg_free(reply); | |
1968 | goto out; | |
1969 | } | |
1970 | } | |
1971 | } | |
1972 | nla_nest_end(reply, dev_list); | |
1973 | genlmsg_end(reply, reply_head); | |
1974 | genlmsg_reply(reply, info); | |
1975 | ret = 0; | |
1976 | out: | |
1977 | mutex_unlock(&nbd_index_mutex); | |
1978 | return ret; | |
1979 | } | |
1980 | ||
e46c7287 JB |
1981 | static void nbd_connect_reply(struct genl_info *info, int index) |
1982 | { | |
1983 | struct sk_buff *skb; | |
1984 | void *msg_head; | |
1985 | int ret; | |
1986 | ||
1987 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
1988 | if (!skb) | |
1989 | return; | |
1990 | msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0, | |
1991 | NBD_CMD_CONNECT); | |
1992 | if (!msg_head) { | |
1993 | nlmsg_free(skb); | |
1994 | return; | |
1995 | } | |
1996 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
1997 | if (ret) { | |
1998 | nlmsg_free(skb); | |
1999 | return; | |
2000 | } | |
2001 | genlmsg_end(skb, msg_head); | |
2002 | genlmsg_reply(skb, info); | |
2003 | } | |
1da177e4 | 2004 | |
799f9a38 JB |
2005 | static void nbd_mcast_index(int index) |
2006 | { | |
2007 | struct sk_buff *skb; | |
2008 | void *msg_head; | |
2009 | int ret; | |
2010 | ||
2011 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
2012 | if (!skb) | |
2013 | return; | |
2014 | msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0, | |
2015 | NBD_CMD_LINK_DEAD); | |
2016 | if (!msg_head) { | |
2017 | nlmsg_free(skb); | |
2018 | return; | |
2019 | } | |
2020 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
2021 | if (ret) { | |
2022 | nlmsg_free(skb); | |
2023 | return; | |
2024 | } | |
2025 | genlmsg_end(skb, msg_head); | |
2026 | genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL); | |
2027 | } | |
2028 | ||
2029 | static void nbd_dead_link_work(struct work_struct *work) | |
2030 | { | |
2031 | struct link_dead_args *args = container_of(work, struct link_dead_args, | |
2032 | work); | |
2033 | nbd_mcast_index(args->index); | |
2034 | kfree(args); | |
2035 | } | |
2036 | ||
1da177e4 LT |
2037 | static int __init nbd_init(void) |
2038 | { | |
1da177e4 LT |
2039 | int i; |
2040 | ||
5b7b18cc | 2041 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); |
1da177e4 | 2042 | |
d71a6d73 | 2043 | if (max_part < 0) { |
7742ce4a | 2044 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); |
d71a6d73 LV |
2045 | return -EINVAL; |
2046 | } | |
2047 | ||
2048 | part_shift = 0; | |
5988ce23 | 2049 | if (max_part > 0) { |
d71a6d73 LV |
2050 | part_shift = fls(max_part); |
2051 | ||
5988ce23 NK |
2052 | /* |
2053 | * Adjust max_part according to part_shift as it is exported | |
2054 | * to user space so that user can know the max number of | |
2055 | * partition kernel should be able to manage. | |
2056 | * | |
2057 | * Note that -1 is required because partition 0 is reserved | |
2058 | * for the whole disk. | |
2059 | */ | |
2060 | max_part = (1UL << part_shift) - 1; | |
2061 | } | |
2062 | ||
3b271082 NK |
2063 | if ((1UL << part_shift) > DISK_MAX_PARTS) |
2064 | return -EINVAL; | |
2065 | ||
2066 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
2067 | return -EINVAL; | |
124d6db0 JB |
2068 | recv_workqueue = alloc_workqueue("knbd-recv", |
2069 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
2070 | if (!recv_workqueue) | |
2071 | return -ENOMEM; | |
3b271082 | 2072 | |
6330a2d0 JB |
2073 | if (register_blkdev(NBD_MAJOR, "nbd")) { |
2074 | destroy_workqueue(recv_workqueue); | |
b0d9111a | 2075 | return -EIO; |
6330a2d0 | 2076 | } |
1da177e4 | 2077 | |
e46c7287 JB |
2078 | if (genl_register_family(&nbd_genl_family)) { |
2079 | unregister_blkdev(NBD_MAJOR, "nbd"); | |
2080 | destroy_workqueue(recv_workqueue); | |
2081 | return -EINVAL; | |
2082 | } | |
30d53d9c MP |
2083 | nbd_dbg_init(); |
2084 | ||
b0d9111a JB |
2085 | mutex_lock(&nbd_index_mutex); |
2086 | for (i = 0; i < nbds_max; i++) | |
2087 | nbd_dev_add(i); | |
2088 | mutex_unlock(&nbd_index_mutex); | |
2089 | return 0; | |
2090 | } | |
1da177e4 | 2091 | |
b0d9111a JB |
2092 | static int nbd_exit_cb(int id, void *ptr, void *data) |
2093 | { | |
c6a4759e | 2094 | struct list_head *list = (struct list_head *)data; |
b0d9111a | 2095 | struct nbd_device *nbd = ptr; |
c6a4759e | 2096 | |
c6a4759e | 2097 | list_add_tail(&nbd->list, list); |
1da177e4 | 2098 | return 0; |
1da177e4 LT |
2099 | } |
2100 | ||
2101 | static void __exit nbd_cleanup(void) | |
2102 | { | |
c6a4759e JB |
2103 | struct nbd_device *nbd; |
2104 | LIST_HEAD(del_list); | |
2105 | ||
30d53d9c MP |
2106 | nbd_dbg_close(); |
2107 | ||
c6a4759e JB |
2108 | mutex_lock(&nbd_index_mutex); |
2109 | idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list); | |
2110 | mutex_unlock(&nbd_index_mutex); | |
2111 | ||
60ae36ad JB |
2112 | while (!list_empty(&del_list)) { |
2113 | nbd = list_first_entry(&del_list, struct nbd_device, list); | |
2114 | list_del_init(&nbd->list); | |
2115 | if (refcount_read(&nbd->refs) != 1) | |
c6a4759e JB |
2116 | printk(KERN_ERR "nbd: possibly leaking a device\n"); |
2117 | nbd_put(nbd); | |
c6a4759e JB |
2118 | } |
2119 | ||
b0d9111a | 2120 | idr_destroy(&nbd_index_idr); |
e46c7287 | 2121 | genl_unregister_family(&nbd_genl_family); |
124d6db0 | 2122 | destroy_workqueue(recv_workqueue); |
1da177e4 | 2123 | unregister_blkdev(NBD_MAJOR, "nbd"); |
1da177e4 LT |
2124 | } |
2125 | ||
2126 | module_init(nbd_init); | |
2127 | module_exit(nbd_cleanup); | |
2128 | ||
2129 | MODULE_DESCRIPTION("Network Block Device"); | |
2130 | MODULE_LICENSE("GPL"); | |
2131 | ||
40be0c28 | 2132 | module_param(nbds_max, int, 0444); |
d71a6d73 LV |
2133 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); |
2134 | module_param(max_part, int, 0444); | |
2135 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |