]>
Commit | Line | Data |
---|---|---|
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 | * | |
7 | * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz> | |
8 | * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com> | |
9 | * | |
10 | * This file is released under GPLv2 or later. | |
11 | * | |
12 | * (part of code stolen from loop.c) | |
13 | */ | |
14 | ||
15 | #include <linux/major.h> | |
16 | ||
17 | #include <linux/blkdev.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/init.h> | |
20 | #include <linux/sched.h> | |
21 | #include <linux/sched/mm.h> | |
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> | |
28 | #include <linux/mutex.h> | |
29 | #include <linux/compiler.h> | |
30 | #include <linux/err.h> | |
31 | #include <linux/kernel.h> | |
32 | #include <linux/slab.h> | |
33 | #include <net/sock.h> | |
34 | #include <linux/net.h> | |
35 | #include <linux/kthread.h> | |
36 | #include <linux/types.h> | |
37 | #include <linux/debugfs.h> | |
38 | #include <linux/blk-mq.h> | |
39 | ||
40 | #include <linux/uaccess.h> | |
41 | #include <asm/types.h> | |
42 | ||
43 | #include <linux/nbd.h> | |
44 | #include <linux/nbd-netlink.h> | |
45 | #include <net/genetlink.h> | |
46 | ||
47 | static DEFINE_IDR(nbd_index_idr); | |
48 | static DEFINE_MUTEX(nbd_index_mutex); | |
49 | static int nbd_total_devices = 0; | |
50 | ||
51 | struct nbd_sock { | |
52 | struct socket *sock; | |
53 | struct mutex tx_lock; | |
54 | struct request *pending; | |
55 | int sent; | |
56 | bool dead; | |
57 | int fallback_index; | |
58 | int cookie; | |
59 | }; | |
60 | ||
61 | struct recv_thread_args { | |
62 | struct work_struct work; | |
63 | struct nbd_device *nbd; | |
64 | int index; | |
65 | }; | |
66 | ||
67 | struct link_dead_args { | |
68 | struct work_struct work; | |
69 | int index; | |
70 | }; | |
71 | ||
72 | #define NBD_TIMEDOUT 0 | |
73 | #define NBD_DISCONNECT_REQUESTED 1 | |
74 | #define NBD_DISCONNECTED 2 | |
75 | #define NBD_HAS_PID_FILE 3 | |
76 | #define NBD_HAS_CONFIG_REF 4 | |
77 | #define NBD_BOUND 5 | |
78 | #define NBD_DESTROY_ON_DISCONNECT 6 | |
79 | ||
80 | struct nbd_config { | |
81 | u32 flags; | |
82 | unsigned long runtime_flags; | |
83 | u64 dead_conn_timeout; | |
84 | ||
85 | struct nbd_sock **socks; | |
86 | int num_connections; | |
87 | atomic_t live_connections; | |
88 | wait_queue_head_t conn_wait; | |
89 | ||
90 | atomic_t recv_threads; | |
91 | wait_queue_head_t recv_wq; | |
92 | loff_t blksize; | |
93 | loff_t bytesize; | |
94 | #if IS_ENABLED(CONFIG_DEBUG_FS) | |
95 | struct dentry *dbg_dir; | |
96 | #endif | |
97 | }; | |
98 | ||
99 | struct nbd_device { | |
100 | struct blk_mq_tag_set tag_set; | |
101 | ||
102 | int index; | |
103 | refcount_t config_refs; | |
104 | refcount_t refs; | |
105 | struct nbd_config *config; | |
106 | struct mutex config_lock; | |
107 | struct gendisk *disk; | |
108 | ||
109 | struct list_head list; | |
110 | struct task_struct *task_recv; | |
111 | struct task_struct *task_setup; | |
112 | }; | |
113 | ||
114 | struct nbd_cmd { | |
115 | struct nbd_device *nbd; | |
116 | int index; | |
117 | int cookie; | |
118 | struct completion send_complete; | |
119 | blk_status_t status; | |
120 | }; | |
121 | ||
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 | ||
128 | #define NBD_MAGIC 0x68797548 | |
129 | ||
130 | static unsigned int nbds_max = 16; | |
131 | static int max_part = 15; | |
132 | static struct workqueue_struct *recv_workqueue; | |
133 | static int part_shift; | |
134 | ||
135 | static int nbd_dev_dbg_init(struct nbd_device *nbd); | |
136 | static void nbd_dev_dbg_close(struct nbd_device *nbd); | |
137 | static void nbd_config_put(struct nbd_device *nbd); | |
138 | static void nbd_connect_reply(struct genl_info *info, int index); | |
139 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info); | |
140 | static void nbd_dead_link_work(struct work_struct *work); | |
141 | ||
142 | static inline struct device *nbd_to_dev(struct nbd_device *nbd) | |
143 | { | |
144 | return disk_to_dev(nbd->disk); | |
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"; | |
153 | case NBD_CMD_FLUSH: return "flush"; | |
154 | case NBD_CMD_TRIM: return "trim/discard"; | |
155 | } | |
156 | return "invalid"; | |
157 | } | |
158 | ||
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 | ||
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); | |
180 | disk->private_data = NULL; | |
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 | ||
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) | |
204 | { | |
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 | } | |
214 | if (!nsock->dead) { | |
215 | kernel_sock_shutdown(nsock->sock, SHUT_RDWR); | |
216 | atomic_dec(&nbd->config->live_connections); | |
217 | } | |
218 | nsock->dead = true; | |
219 | nsock->pending = NULL; | |
220 | nsock->sent = 0; | |
221 | } | |
222 | ||
223 | static void nbd_size_clear(struct nbd_device *nbd) | |
224 | { | |
225 | if (nbd->config->bytesize) { | |
226 | set_capacity(nbd->disk, 0); | |
227 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
228 | } | |
229 | } | |
230 | ||
231 | static void nbd_size_update(struct nbd_device *nbd) | |
232 | { | |
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); | |
236 | set_capacity(nbd->disk, config->bytesize >> 9); | |
237 | kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE); | |
238 | } | |
239 | ||
240 | static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize, | |
241 | loff_t nr_blocks) | |
242 | { | |
243 | struct nbd_config *config = nbd->config; | |
244 | config->blksize = blocksize; | |
245 | config->bytesize = blocksize * nr_blocks; | |
246 | } | |
247 | ||
248 | static void nbd_complete_rq(struct request *req) | |
249 | { | |
250 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); | |
251 | ||
252 | dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd, | |
253 | cmd->status ? "failed" : "done"); | |
254 | ||
255 | blk_mq_end_request(req, cmd->status); | |
256 | } | |
257 | ||
258 | /* | |
259 | * Forcibly shutdown the socket causing all listeners to error | |
260 | */ | |
261 | static void sock_shutdown(struct nbd_device *nbd) | |
262 | { | |
263 | struct nbd_config *config = nbd->config; | |
264 | int i; | |
265 | ||
266 | if (config->num_connections == 0) | |
267 | return; | |
268 | if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags)) | |
269 | return; | |
270 | ||
271 | for (i = 0; i < config->num_connections; i++) { | |
272 | struct nbd_sock *nsock = config->socks[i]; | |
273 | mutex_lock(&nsock->tx_lock); | |
274 | nbd_mark_nsock_dead(nbd, nsock, 0); | |
275 | mutex_unlock(&nsock->tx_lock); | |
276 | } | |
277 | dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n"); | |
278 | } | |
279 | ||
280 | static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req, | |
281 | bool reserved) | |
282 | { | |
283 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req); | |
284 | struct nbd_device *nbd = cmd->nbd; | |
285 | struct nbd_config *config; | |
286 | ||
287 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
288 | cmd->status = BLK_STS_TIMEOUT; | |
289 | return BLK_EH_HANDLED; | |
290 | } | |
291 | ||
292 | /* If we are waiting on our dead timer then we could get timeout | |
293 | * callbacks for our request. For this we just want to reset the timer | |
294 | * and let the queue side take care of everything. | |
295 | */ | |
296 | if (!completion_done(&cmd->send_complete)) { | |
297 | nbd_config_put(nbd); | |
298 | return BLK_EH_RESET_TIMER; | |
299 | } | |
300 | config = nbd->config; | |
301 | ||
302 | if (config->num_connections > 1) { | |
303 | dev_err_ratelimited(nbd_to_dev(nbd), | |
304 | "Connection timed out, retrying\n"); | |
305 | /* | |
306 | * Hooray we have more connections, requeue this IO, the submit | |
307 | * path will put it on a real connection. | |
308 | */ | |
309 | if (config->socks && config->num_connections > 1) { | |
310 | if (cmd->index < config->num_connections) { | |
311 | struct nbd_sock *nsock = | |
312 | config->socks[cmd->index]; | |
313 | mutex_lock(&nsock->tx_lock); | |
314 | /* We can have multiple outstanding requests, so | |
315 | * we don't want to mark the nsock dead if we've | |
316 | * already reconnected with a new socket, so | |
317 | * only mark it dead if its the same socket we | |
318 | * were sent out on. | |
319 | */ | |
320 | if (cmd->cookie == nsock->cookie) | |
321 | nbd_mark_nsock_dead(nbd, nsock, 1); | |
322 | mutex_unlock(&nsock->tx_lock); | |
323 | } | |
324 | blk_mq_requeue_request(req, true); | |
325 | nbd_config_put(nbd); | |
326 | return BLK_EH_NOT_HANDLED; | |
327 | } | |
328 | } else { | |
329 | dev_err_ratelimited(nbd_to_dev(nbd), | |
330 | "Connection timed out\n"); | |
331 | } | |
332 | set_bit(NBD_TIMEDOUT, &config->runtime_flags); | |
333 | cmd->status = BLK_STS_IOERR; | |
334 | sock_shutdown(nbd); | |
335 | nbd_config_put(nbd); | |
336 | ||
337 | return BLK_EH_HANDLED; | |
338 | } | |
339 | ||
340 | /* | |
341 | * Send or receive packet. | |
342 | */ | |
343 | static int sock_xmit(struct nbd_device *nbd, int index, int send, | |
344 | struct iov_iter *iter, int msg_flags, int *sent) | |
345 | { | |
346 | struct nbd_config *config = nbd->config; | |
347 | struct socket *sock = config->socks[index]->sock; | |
348 | int result; | |
349 | struct msghdr msg; | |
350 | unsigned int noreclaim_flag; | |
351 | ||
352 | if (unlikely(!sock)) { | |
353 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
354 | "Attempted %s on closed socket in sock_xmit\n", | |
355 | (send ? "send" : "recv")); | |
356 | return -EINVAL; | |
357 | } | |
358 | ||
359 | msg.msg_iter = *iter; | |
360 | ||
361 | noreclaim_flag = memalloc_noreclaim_save(); | |
362 | do { | |
363 | sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC; | |
364 | msg.msg_name = NULL; | |
365 | msg.msg_namelen = 0; | |
366 | msg.msg_control = NULL; | |
367 | msg.msg_controllen = 0; | |
368 | msg.msg_flags = msg_flags | MSG_NOSIGNAL; | |
369 | ||
370 | if (send) | |
371 | result = sock_sendmsg(sock, &msg); | |
372 | else | |
373 | result = sock_recvmsg(sock, &msg, msg.msg_flags); | |
374 | ||
375 | if (result <= 0) { | |
376 | if (result == 0) | |
377 | result = -EPIPE; /* short read */ | |
378 | break; | |
379 | } | |
380 | if (sent) | |
381 | *sent += result; | |
382 | } while (msg_data_left(&msg)); | |
383 | ||
384 | memalloc_noreclaim_restore(noreclaim_flag); | |
385 | ||
386 | return result; | |
387 | } | |
388 | ||
389 | /* | |
390 | * Different settings for sk->sk_sndtimeo can result in different return values | |
391 | * if there is a signal pending when we enter sendmsg, because reasons? | |
392 | */ | |
393 | static inline int was_interrupted(int result) | |
394 | { | |
395 | return result == -ERESTARTSYS || result == -EINTR; | |
396 | } | |
397 | ||
398 | /* always call with the tx_lock held */ | |
399 | static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index) | |
400 | { | |
401 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
402 | struct nbd_config *config = nbd->config; | |
403 | struct nbd_sock *nsock = config->socks[index]; | |
404 | int result; | |
405 | struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)}; | |
406 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
407 | struct iov_iter from; | |
408 | unsigned long size = blk_rq_bytes(req); | |
409 | struct bio *bio; | |
410 | u32 type; | |
411 | u32 nbd_cmd_flags = 0; | |
412 | u32 tag = blk_mq_unique_tag(req); | |
413 | int sent = nsock->sent, skip = 0; | |
414 | ||
415 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); | |
416 | ||
417 | switch (req_op(req)) { | |
418 | case REQ_OP_DISCARD: | |
419 | type = NBD_CMD_TRIM; | |
420 | break; | |
421 | case REQ_OP_FLUSH: | |
422 | type = NBD_CMD_FLUSH; | |
423 | break; | |
424 | case REQ_OP_WRITE: | |
425 | type = NBD_CMD_WRITE; | |
426 | break; | |
427 | case REQ_OP_READ: | |
428 | type = NBD_CMD_READ; | |
429 | break; | |
430 | default: | |
431 | return -EIO; | |
432 | } | |
433 | ||
434 | if (rq_data_dir(req) == WRITE && | |
435 | (config->flags & NBD_FLAG_READ_ONLY)) { | |
436 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
437 | "Write on read-only\n"); | |
438 | return -EIO; | |
439 | } | |
440 | ||
441 | if (req->cmd_flags & REQ_FUA) | |
442 | nbd_cmd_flags |= NBD_CMD_FLAG_FUA; | |
443 | ||
444 | /* We did a partial send previously, and we at least sent the whole | |
445 | * request struct, so just go and send the rest of the pages in the | |
446 | * request. | |
447 | */ | |
448 | if (sent) { | |
449 | if (sent >= sizeof(request)) { | |
450 | skip = sent - sizeof(request); | |
451 | goto send_pages; | |
452 | } | |
453 | iov_iter_advance(&from, sent); | |
454 | } | |
455 | cmd->index = index; | |
456 | cmd->cookie = nsock->cookie; | |
457 | request.type = htonl(type | nbd_cmd_flags); | |
458 | if (type != NBD_CMD_FLUSH) { | |
459 | request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9); | |
460 | request.len = htonl(size); | |
461 | } | |
462 | memcpy(request.handle, &tag, sizeof(tag)); | |
463 | ||
464 | dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n", | |
465 | cmd, nbdcmd_to_ascii(type), | |
466 | (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req)); | |
467 | result = sock_xmit(nbd, index, 1, &from, | |
468 | (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent); | |
469 | if (result <= 0) { | |
470 | if (was_interrupted(result)) { | |
471 | /* If we havne't sent anything we can just return BUSY, | |
472 | * however if we have sent something we need to make | |
473 | * sure we only allow this req to be sent until we are | |
474 | * completely done. | |
475 | */ | |
476 | if (sent) { | |
477 | nsock->pending = req; | |
478 | nsock->sent = sent; | |
479 | } | |
480 | return BLK_STS_RESOURCE; | |
481 | } | |
482 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
483 | "Send control failed (result %d)\n", result); | |
484 | return -EAGAIN; | |
485 | } | |
486 | send_pages: | |
487 | if (type != NBD_CMD_WRITE) | |
488 | goto out; | |
489 | ||
490 | bio = req->bio; | |
491 | while (bio) { | |
492 | struct bio *next = bio->bi_next; | |
493 | struct bvec_iter iter; | |
494 | struct bio_vec bvec; | |
495 | ||
496 | bio_for_each_segment(bvec, bio, iter) { | |
497 | bool is_last = !next && bio_iter_last(bvec, iter); | |
498 | int flags = is_last ? 0 : MSG_MORE; | |
499 | ||
500 | dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n", | |
501 | cmd, bvec.bv_len); | |
502 | iov_iter_bvec(&from, ITER_BVEC | WRITE, | |
503 | &bvec, 1, bvec.bv_len); | |
504 | if (skip) { | |
505 | if (skip >= iov_iter_count(&from)) { | |
506 | skip -= iov_iter_count(&from); | |
507 | continue; | |
508 | } | |
509 | iov_iter_advance(&from, skip); | |
510 | skip = 0; | |
511 | } | |
512 | result = sock_xmit(nbd, index, 1, &from, flags, &sent); | |
513 | if (result <= 0) { | |
514 | if (was_interrupted(result)) { | |
515 | /* We've already sent the header, we | |
516 | * have no choice but to set pending and | |
517 | * return BUSY. | |
518 | */ | |
519 | nsock->pending = req; | |
520 | nsock->sent = sent; | |
521 | return BLK_STS_RESOURCE; | |
522 | } | |
523 | dev_err(disk_to_dev(nbd->disk), | |
524 | "Send data failed (result %d)\n", | |
525 | result); | |
526 | return -EAGAIN; | |
527 | } | |
528 | /* | |
529 | * The completion might already have come in, | |
530 | * so break for the last one instead of letting | |
531 | * the iterator do it. This prevents use-after-free | |
532 | * of the bio. | |
533 | */ | |
534 | if (is_last) | |
535 | break; | |
536 | } | |
537 | bio = next; | |
538 | } | |
539 | out: | |
540 | nsock->pending = NULL; | |
541 | nsock->sent = 0; | |
542 | return 0; | |
543 | } | |
544 | ||
545 | /* NULL returned = something went wrong, inform userspace */ | |
546 | static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index) | |
547 | { | |
548 | struct nbd_config *config = nbd->config; | |
549 | int result; | |
550 | struct nbd_reply reply; | |
551 | struct nbd_cmd *cmd; | |
552 | struct request *req = NULL; | |
553 | u16 hwq; | |
554 | u32 tag; | |
555 | struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)}; | |
556 | struct iov_iter to; | |
557 | ||
558 | reply.magic = 0; | |
559 | iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply)); | |
560 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); | |
561 | if (result <= 0) { | |
562 | if (!nbd_disconnected(config)) | |
563 | dev_err(disk_to_dev(nbd->disk), | |
564 | "Receive control failed (result %d)\n", result); | |
565 | return ERR_PTR(result); | |
566 | } | |
567 | ||
568 | if (ntohl(reply.magic) != NBD_REPLY_MAGIC) { | |
569 | dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n", | |
570 | (unsigned long)ntohl(reply.magic)); | |
571 | return ERR_PTR(-EPROTO); | |
572 | } | |
573 | ||
574 | memcpy(&tag, reply.handle, sizeof(u32)); | |
575 | ||
576 | hwq = blk_mq_unique_tag_to_hwq(tag); | |
577 | if (hwq < nbd->tag_set.nr_hw_queues) | |
578 | req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq], | |
579 | blk_mq_unique_tag_to_tag(tag)); | |
580 | if (!req || !blk_mq_request_started(req)) { | |
581 | dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n", | |
582 | tag, req); | |
583 | return ERR_PTR(-ENOENT); | |
584 | } | |
585 | cmd = blk_mq_rq_to_pdu(req); | |
586 | if (ntohl(reply.error)) { | |
587 | dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n", | |
588 | ntohl(reply.error)); | |
589 | cmd->status = BLK_STS_IOERR; | |
590 | return cmd; | |
591 | } | |
592 | ||
593 | dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd); | |
594 | if (rq_data_dir(req) != WRITE) { | |
595 | struct req_iterator iter; | |
596 | struct bio_vec bvec; | |
597 | ||
598 | rq_for_each_segment(bvec, req, iter) { | |
599 | iov_iter_bvec(&to, ITER_BVEC | READ, | |
600 | &bvec, 1, bvec.bv_len); | |
601 | result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL); | |
602 | if (result <= 0) { | |
603 | dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n", | |
604 | result); | |
605 | /* | |
606 | * If we've disconnected or we only have 1 | |
607 | * connection then we need to make sure we | |
608 | * complete this request, otherwise error out | |
609 | * and let the timeout stuff handle resubmitting | |
610 | * this request onto another connection. | |
611 | */ | |
612 | if (nbd_disconnected(config) || | |
613 | config->num_connections <= 1) { | |
614 | cmd->status = BLK_STS_IOERR; | |
615 | return cmd; | |
616 | } | |
617 | return ERR_PTR(-EIO); | |
618 | } | |
619 | dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n", | |
620 | cmd, bvec.bv_len); | |
621 | } | |
622 | } else { | |
623 | /* See the comment in nbd_queue_rq. */ | |
624 | wait_for_completion(&cmd->send_complete); | |
625 | } | |
626 | return cmd; | |
627 | } | |
628 | ||
629 | static void recv_work(struct work_struct *work) | |
630 | { | |
631 | struct recv_thread_args *args = container_of(work, | |
632 | struct recv_thread_args, | |
633 | work); | |
634 | struct nbd_device *nbd = args->nbd; | |
635 | struct nbd_config *config = nbd->config; | |
636 | struct nbd_cmd *cmd; | |
637 | ||
638 | while (1) { | |
639 | cmd = nbd_read_stat(nbd, args->index); | |
640 | if (IS_ERR(cmd)) { | |
641 | struct nbd_sock *nsock = config->socks[args->index]; | |
642 | ||
643 | mutex_lock(&nsock->tx_lock); | |
644 | nbd_mark_nsock_dead(nbd, nsock, 1); | |
645 | mutex_unlock(&nsock->tx_lock); | |
646 | break; | |
647 | } | |
648 | ||
649 | blk_mq_complete_request(blk_mq_rq_from_pdu(cmd)); | |
650 | } | |
651 | atomic_dec(&config->recv_threads); | |
652 | wake_up(&config->recv_wq); | |
653 | nbd_config_put(nbd); | |
654 | kfree(args); | |
655 | } | |
656 | ||
657 | static void nbd_clear_req(struct request *req, void *data, bool reserved) | |
658 | { | |
659 | struct nbd_cmd *cmd; | |
660 | ||
661 | if (!blk_mq_request_started(req)) | |
662 | return; | |
663 | cmd = blk_mq_rq_to_pdu(req); | |
664 | cmd->status = BLK_STS_IOERR; | |
665 | blk_mq_complete_request(req); | |
666 | } | |
667 | ||
668 | static void nbd_clear_que(struct nbd_device *nbd) | |
669 | { | |
670 | blk_mq_quiesce_queue(nbd->disk->queue); | |
671 | blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL); | |
672 | blk_mq_unquiesce_queue(nbd->disk->queue); | |
673 | dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n"); | |
674 | } | |
675 | ||
676 | static int find_fallback(struct nbd_device *nbd, int index) | |
677 | { | |
678 | struct nbd_config *config = nbd->config; | |
679 | int new_index = -1; | |
680 | struct nbd_sock *nsock = config->socks[index]; | |
681 | int fallback = nsock->fallback_index; | |
682 | ||
683 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) | |
684 | return new_index; | |
685 | ||
686 | if (config->num_connections <= 1) { | |
687 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
688 | "Attempted send on invalid socket\n"); | |
689 | return new_index; | |
690 | } | |
691 | ||
692 | if (fallback >= 0 && fallback < config->num_connections && | |
693 | !config->socks[fallback]->dead) | |
694 | return fallback; | |
695 | ||
696 | if (nsock->fallback_index < 0 || | |
697 | nsock->fallback_index >= config->num_connections || | |
698 | config->socks[nsock->fallback_index]->dead) { | |
699 | int i; | |
700 | for (i = 0; i < config->num_connections; i++) { | |
701 | if (i == index) | |
702 | continue; | |
703 | if (!config->socks[i]->dead) { | |
704 | new_index = i; | |
705 | break; | |
706 | } | |
707 | } | |
708 | nsock->fallback_index = new_index; | |
709 | if (new_index < 0) { | |
710 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
711 | "Dead connection, failed to find a fallback\n"); | |
712 | return new_index; | |
713 | } | |
714 | } | |
715 | new_index = nsock->fallback_index; | |
716 | return new_index; | |
717 | } | |
718 | ||
719 | static int wait_for_reconnect(struct nbd_device *nbd) | |
720 | { | |
721 | struct nbd_config *config = nbd->config; | |
722 | if (!config->dead_conn_timeout) | |
723 | return 0; | |
724 | if (test_bit(NBD_DISCONNECTED, &config->runtime_flags)) | |
725 | return 0; | |
726 | wait_event_interruptible_timeout(config->conn_wait, | |
727 | atomic_read(&config->live_connections), | |
728 | config->dead_conn_timeout); | |
729 | return atomic_read(&config->live_connections); | |
730 | } | |
731 | ||
732 | static int nbd_handle_cmd(struct nbd_cmd *cmd, int index) | |
733 | { | |
734 | struct request *req = blk_mq_rq_from_pdu(cmd); | |
735 | struct nbd_device *nbd = cmd->nbd; | |
736 | struct nbd_config *config; | |
737 | struct nbd_sock *nsock; | |
738 | int ret; | |
739 | ||
740 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
741 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
742 | "Socks array is empty\n"); | |
743 | return -EINVAL; | |
744 | } | |
745 | config = nbd->config; | |
746 | ||
747 | if (index >= config->num_connections) { | |
748 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
749 | "Attempted send on invalid socket\n"); | |
750 | nbd_config_put(nbd); | |
751 | return -EINVAL; | |
752 | } | |
753 | cmd->status = BLK_STS_OK; | |
754 | again: | |
755 | nsock = config->socks[index]; | |
756 | mutex_lock(&nsock->tx_lock); | |
757 | if (nsock->dead) { | |
758 | int old_index = index; | |
759 | index = find_fallback(nbd, index); | |
760 | mutex_unlock(&nsock->tx_lock); | |
761 | if (index < 0) { | |
762 | if (wait_for_reconnect(nbd)) { | |
763 | index = old_index; | |
764 | goto again; | |
765 | } | |
766 | /* All the sockets should already be down at this point, | |
767 | * we just want to make sure that DISCONNECTED is set so | |
768 | * any requests that come in that were queue'ed waiting | |
769 | * for the reconnect timer don't trigger the timer again | |
770 | * and instead just error out. | |
771 | */ | |
772 | sock_shutdown(nbd); | |
773 | nbd_config_put(nbd); | |
774 | return -EIO; | |
775 | } | |
776 | goto again; | |
777 | } | |
778 | ||
779 | /* Handle the case that we have a pending request that was partially | |
780 | * transmitted that _has_ to be serviced first. We need to call requeue | |
781 | * here so that it gets put _after_ the request that is already on the | |
782 | * dispatch list. | |
783 | */ | |
784 | if (unlikely(nsock->pending && nsock->pending != req)) { | |
785 | blk_mq_requeue_request(req, true); | |
786 | ret = 0; | |
787 | goto out; | |
788 | } | |
789 | /* | |
790 | * Some failures are related to the link going down, so anything that | |
791 | * returns EAGAIN can be retried on a different socket. | |
792 | */ | |
793 | ret = nbd_send_cmd(nbd, cmd, index); | |
794 | if (ret == -EAGAIN) { | |
795 | dev_err_ratelimited(disk_to_dev(nbd->disk), | |
796 | "Request send failed trying another connection\n"); | |
797 | nbd_mark_nsock_dead(nbd, nsock, 1); | |
798 | mutex_unlock(&nsock->tx_lock); | |
799 | goto again; | |
800 | } | |
801 | out: | |
802 | mutex_unlock(&nsock->tx_lock); | |
803 | nbd_config_put(nbd); | |
804 | return ret; | |
805 | } | |
806 | ||
807 | static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx, | |
808 | const struct blk_mq_queue_data *bd) | |
809 | { | |
810 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq); | |
811 | int ret; | |
812 | ||
813 | /* | |
814 | * Since we look at the bio's to send the request over the network we | |
815 | * need to make sure the completion work doesn't mark this request done | |
816 | * before we are done doing our send. This keeps us from dereferencing | |
817 | * freed data if we have particularly fast completions (ie we get the | |
818 | * completion before we exit sock_xmit on the last bvec) or in the case | |
819 | * that the server is misbehaving (or there was an error) before we're | |
820 | * done sending everything over the wire. | |
821 | */ | |
822 | init_completion(&cmd->send_complete); | |
823 | blk_mq_start_request(bd->rq); | |
824 | ||
825 | /* We can be called directly from the user space process, which means we | |
826 | * could possibly have signals pending so our sendmsg will fail. In | |
827 | * this case we need to return that we are busy, otherwise error out as | |
828 | * appropriate. | |
829 | */ | |
830 | ret = nbd_handle_cmd(cmd, hctx->queue_num); | |
831 | complete(&cmd->send_complete); | |
832 | ||
833 | return ret < 0 ? BLK_STS_IOERR : BLK_STS_OK; | |
834 | } | |
835 | ||
836 | static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg, | |
837 | bool netlink) | |
838 | { | |
839 | struct nbd_config *config = nbd->config; | |
840 | struct socket *sock; | |
841 | struct nbd_sock **socks; | |
842 | struct nbd_sock *nsock; | |
843 | int err; | |
844 | ||
845 | sock = sockfd_lookup(arg, &err); | |
846 | if (!sock) | |
847 | return err; | |
848 | ||
849 | if (!netlink && !nbd->task_setup && | |
850 | !test_bit(NBD_BOUND, &config->runtime_flags)) | |
851 | nbd->task_setup = current; | |
852 | ||
853 | if (!netlink && | |
854 | (nbd->task_setup != current || | |
855 | test_bit(NBD_BOUND, &config->runtime_flags))) { | |
856 | dev_err(disk_to_dev(nbd->disk), | |
857 | "Device being setup by another task"); | |
858 | sockfd_put(sock); | |
859 | return -EBUSY; | |
860 | } | |
861 | ||
862 | socks = krealloc(config->socks, (config->num_connections + 1) * | |
863 | sizeof(struct nbd_sock *), GFP_KERNEL); | |
864 | if (!socks) { | |
865 | sockfd_put(sock); | |
866 | return -ENOMEM; | |
867 | } | |
868 | nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL); | |
869 | if (!nsock) { | |
870 | sockfd_put(sock); | |
871 | return -ENOMEM; | |
872 | } | |
873 | ||
874 | config->socks = socks; | |
875 | ||
876 | nsock->fallback_index = -1; | |
877 | nsock->dead = false; | |
878 | mutex_init(&nsock->tx_lock); | |
879 | nsock->sock = sock; | |
880 | nsock->pending = NULL; | |
881 | nsock->sent = 0; | |
882 | nsock->cookie = 0; | |
883 | socks[config->num_connections++] = nsock; | |
884 | atomic_inc(&config->live_connections); | |
885 | ||
886 | return 0; | |
887 | } | |
888 | ||
889 | static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg) | |
890 | { | |
891 | struct nbd_config *config = nbd->config; | |
892 | struct socket *sock, *old; | |
893 | struct recv_thread_args *args; | |
894 | int i; | |
895 | int err; | |
896 | ||
897 | sock = sockfd_lookup(arg, &err); | |
898 | if (!sock) | |
899 | return err; | |
900 | ||
901 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
902 | if (!args) { | |
903 | sockfd_put(sock); | |
904 | return -ENOMEM; | |
905 | } | |
906 | ||
907 | for (i = 0; i < config->num_connections; i++) { | |
908 | struct nbd_sock *nsock = config->socks[i]; | |
909 | ||
910 | if (!nsock->dead) | |
911 | continue; | |
912 | ||
913 | mutex_lock(&nsock->tx_lock); | |
914 | if (!nsock->dead) { | |
915 | mutex_unlock(&nsock->tx_lock); | |
916 | continue; | |
917 | } | |
918 | sk_set_memalloc(sock->sk); | |
919 | if (nbd->tag_set.timeout) | |
920 | sock->sk->sk_sndtimeo = nbd->tag_set.timeout; | |
921 | atomic_inc(&config->recv_threads); | |
922 | refcount_inc(&nbd->config_refs); | |
923 | old = nsock->sock; | |
924 | nsock->fallback_index = -1; | |
925 | nsock->sock = sock; | |
926 | nsock->dead = false; | |
927 | INIT_WORK(&args->work, recv_work); | |
928 | args->index = i; | |
929 | args->nbd = nbd; | |
930 | nsock->cookie++; | |
931 | mutex_unlock(&nsock->tx_lock); | |
932 | sockfd_put(old); | |
933 | ||
934 | clear_bit(NBD_DISCONNECTED, &config->runtime_flags); | |
935 | ||
936 | /* We take the tx_mutex in an error path in the recv_work, so we | |
937 | * need to queue_work outside of the tx_mutex. | |
938 | */ | |
939 | queue_work(recv_workqueue, &args->work); | |
940 | ||
941 | atomic_inc(&config->live_connections); | |
942 | wake_up(&config->conn_wait); | |
943 | return 0; | |
944 | } | |
945 | sockfd_put(sock); | |
946 | kfree(args); | |
947 | return -ENOSPC; | |
948 | } | |
949 | ||
950 | static void nbd_bdev_reset(struct block_device *bdev) | |
951 | { | |
952 | if (bdev->bd_openers > 1) | |
953 | return; | |
954 | bd_set_size(bdev, 0); | |
955 | if (max_part > 0) { | |
956 | blkdev_reread_part(bdev); | |
957 | bdev->bd_invalidated = 1; | |
958 | } | |
959 | } | |
960 | ||
961 | static void nbd_parse_flags(struct nbd_device *nbd) | |
962 | { | |
963 | struct nbd_config *config = nbd->config; | |
964 | if (config->flags & NBD_FLAG_READ_ONLY) | |
965 | set_disk_ro(nbd->disk, true); | |
966 | else | |
967 | set_disk_ro(nbd->disk, false); | |
968 | if (config->flags & NBD_FLAG_SEND_TRIM) | |
969 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); | |
970 | if (config->flags & NBD_FLAG_SEND_FLUSH) { | |
971 | if (config->flags & NBD_FLAG_SEND_FUA) | |
972 | blk_queue_write_cache(nbd->disk->queue, true, true); | |
973 | else | |
974 | blk_queue_write_cache(nbd->disk->queue, true, false); | |
975 | } | |
976 | else | |
977 | blk_queue_write_cache(nbd->disk->queue, false, false); | |
978 | } | |
979 | ||
980 | static void send_disconnects(struct nbd_device *nbd) | |
981 | { | |
982 | struct nbd_config *config = nbd->config; | |
983 | struct nbd_request request = { | |
984 | .magic = htonl(NBD_REQUEST_MAGIC), | |
985 | .type = htonl(NBD_CMD_DISC), | |
986 | }; | |
987 | struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)}; | |
988 | struct iov_iter from; | |
989 | int i, ret; | |
990 | ||
991 | for (i = 0; i < config->num_connections; i++) { | |
992 | struct nbd_sock *nsock = config->socks[i]; | |
993 | ||
994 | iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request)); | |
995 | mutex_lock(&nsock->tx_lock); | |
996 | ret = sock_xmit(nbd, i, 1, &from, 0, NULL); | |
997 | if (ret <= 0) | |
998 | dev_err(disk_to_dev(nbd->disk), | |
999 | "Send disconnect failed %d\n", ret); | |
1000 | mutex_unlock(&nsock->tx_lock); | |
1001 | } | |
1002 | } | |
1003 | ||
1004 | static int nbd_disconnect(struct nbd_device *nbd) | |
1005 | { | |
1006 | struct nbd_config *config = nbd->config; | |
1007 | ||
1008 | dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n"); | |
1009 | set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags); | |
1010 | send_disconnects(nbd); | |
1011 | return 0; | |
1012 | } | |
1013 | ||
1014 | static void nbd_clear_sock(struct nbd_device *nbd) | |
1015 | { | |
1016 | sock_shutdown(nbd); | |
1017 | nbd_clear_que(nbd); | |
1018 | nbd->task_setup = NULL; | |
1019 | } | |
1020 | ||
1021 | static void nbd_config_put(struct nbd_device *nbd) | |
1022 | { | |
1023 | if (refcount_dec_and_mutex_lock(&nbd->config_refs, | |
1024 | &nbd->config_lock)) { | |
1025 | struct nbd_config *config = nbd->config; | |
1026 | nbd_dev_dbg_close(nbd); | |
1027 | nbd_size_clear(nbd); | |
1028 | if (test_and_clear_bit(NBD_HAS_PID_FILE, | |
1029 | &config->runtime_flags)) | |
1030 | device_remove_file(disk_to_dev(nbd->disk), &pid_attr); | |
1031 | nbd->task_recv = NULL; | |
1032 | nbd_clear_sock(nbd); | |
1033 | if (config->num_connections) { | |
1034 | int i; | |
1035 | for (i = 0; i < config->num_connections; i++) { | |
1036 | sockfd_put(config->socks[i]->sock); | |
1037 | kfree(config->socks[i]); | |
1038 | } | |
1039 | kfree(config->socks); | |
1040 | } | |
1041 | kfree(nbd->config); | |
1042 | nbd->config = NULL; | |
1043 | ||
1044 | nbd->tag_set.timeout = 0; | |
1045 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue); | |
1046 | ||
1047 | mutex_unlock(&nbd->config_lock); | |
1048 | nbd_put(nbd); | |
1049 | module_put(THIS_MODULE); | |
1050 | } | |
1051 | } | |
1052 | ||
1053 | static int nbd_start_device(struct nbd_device *nbd) | |
1054 | { | |
1055 | struct nbd_config *config = nbd->config; | |
1056 | int num_connections = config->num_connections; | |
1057 | int error = 0, i; | |
1058 | ||
1059 | if (nbd->task_recv) | |
1060 | return -EBUSY; | |
1061 | if (!config->socks) | |
1062 | return -EINVAL; | |
1063 | if (num_connections > 1 && | |
1064 | !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) { | |
1065 | dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n"); | |
1066 | return -EINVAL; | |
1067 | } | |
1068 | ||
1069 | blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections); | |
1070 | nbd->task_recv = current; | |
1071 | ||
1072 | nbd_parse_flags(nbd); | |
1073 | ||
1074 | error = device_create_file(disk_to_dev(nbd->disk), &pid_attr); | |
1075 | if (error) { | |
1076 | dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n"); | |
1077 | return error; | |
1078 | } | |
1079 | set_bit(NBD_HAS_PID_FILE, &config->runtime_flags); | |
1080 | ||
1081 | nbd_dev_dbg_init(nbd); | |
1082 | for (i = 0; i < num_connections; i++) { | |
1083 | struct recv_thread_args *args; | |
1084 | ||
1085 | args = kzalloc(sizeof(*args), GFP_KERNEL); | |
1086 | if (!args) { | |
1087 | sock_shutdown(nbd); | |
1088 | return -ENOMEM; | |
1089 | } | |
1090 | sk_set_memalloc(config->socks[i]->sock->sk); | |
1091 | if (nbd->tag_set.timeout) | |
1092 | config->socks[i]->sock->sk->sk_sndtimeo = | |
1093 | nbd->tag_set.timeout; | |
1094 | atomic_inc(&config->recv_threads); | |
1095 | refcount_inc(&nbd->config_refs); | |
1096 | INIT_WORK(&args->work, recv_work); | |
1097 | args->nbd = nbd; | |
1098 | args->index = i; | |
1099 | queue_work(recv_workqueue, &args->work); | |
1100 | } | |
1101 | nbd_size_update(nbd); | |
1102 | return error; | |
1103 | } | |
1104 | ||
1105 | static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev) | |
1106 | { | |
1107 | struct nbd_config *config = nbd->config; | |
1108 | int ret; | |
1109 | ||
1110 | ret = nbd_start_device(nbd); | |
1111 | if (ret) | |
1112 | return ret; | |
1113 | ||
1114 | bd_set_size(bdev, config->bytesize); | |
1115 | if (max_part) | |
1116 | bdev->bd_invalidated = 1; | |
1117 | mutex_unlock(&nbd->config_lock); | |
1118 | ret = wait_event_interruptible(config->recv_wq, | |
1119 | atomic_read(&config->recv_threads) == 0); | |
1120 | if (ret) | |
1121 | sock_shutdown(nbd); | |
1122 | mutex_lock(&nbd->config_lock); | |
1123 | bd_set_size(bdev, 0); | |
1124 | /* user requested, ignore socket errors */ | |
1125 | if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags)) | |
1126 | ret = 0; | |
1127 | if (test_bit(NBD_TIMEDOUT, &config->runtime_flags)) | |
1128 | ret = -ETIMEDOUT; | |
1129 | return ret; | |
1130 | } | |
1131 | ||
1132 | static void nbd_clear_sock_ioctl(struct nbd_device *nbd, | |
1133 | struct block_device *bdev) | |
1134 | { | |
1135 | sock_shutdown(nbd); | |
1136 | kill_bdev(bdev); | |
1137 | nbd_bdev_reset(bdev); | |
1138 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, | |
1139 | &nbd->config->runtime_flags)) | |
1140 | nbd_config_put(nbd); | |
1141 | } | |
1142 | ||
1143 | /* Must be called with config_lock held */ | |
1144 | static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd, | |
1145 | unsigned int cmd, unsigned long arg) | |
1146 | { | |
1147 | struct nbd_config *config = nbd->config; | |
1148 | ||
1149 | switch (cmd) { | |
1150 | case NBD_DISCONNECT: | |
1151 | return nbd_disconnect(nbd); | |
1152 | case NBD_CLEAR_SOCK: | |
1153 | nbd_clear_sock_ioctl(nbd, bdev); | |
1154 | return 0; | |
1155 | case NBD_SET_SOCK: | |
1156 | return nbd_add_socket(nbd, arg, false); | |
1157 | case NBD_SET_BLKSIZE: | |
1158 | nbd_size_set(nbd, arg, | |
1159 | div_s64(config->bytesize, arg)); | |
1160 | return 0; | |
1161 | case NBD_SET_SIZE: | |
1162 | nbd_size_set(nbd, config->blksize, | |
1163 | div_s64(arg, config->blksize)); | |
1164 | return 0; | |
1165 | case NBD_SET_SIZE_BLOCKS: | |
1166 | nbd_size_set(nbd, config->blksize, arg); | |
1167 | return 0; | |
1168 | case NBD_SET_TIMEOUT: | |
1169 | if (arg) { | |
1170 | nbd->tag_set.timeout = arg * HZ; | |
1171 | blk_queue_rq_timeout(nbd->disk->queue, arg * HZ); | |
1172 | } | |
1173 | return 0; | |
1174 | ||
1175 | case NBD_SET_FLAGS: | |
1176 | config->flags = arg; | |
1177 | return 0; | |
1178 | case NBD_DO_IT: | |
1179 | return nbd_start_device_ioctl(nbd, bdev); | |
1180 | case NBD_CLEAR_QUE: | |
1181 | /* | |
1182 | * This is for compatibility only. The queue is always cleared | |
1183 | * by NBD_DO_IT or NBD_CLEAR_SOCK. | |
1184 | */ | |
1185 | return 0; | |
1186 | case NBD_PRINT_DEBUG: | |
1187 | /* | |
1188 | * For compatibility only, we no longer keep a list of | |
1189 | * outstanding requests. | |
1190 | */ | |
1191 | return 0; | |
1192 | } | |
1193 | return -ENOTTY; | |
1194 | } | |
1195 | ||
1196 | static int nbd_ioctl(struct block_device *bdev, fmode_t mode, | |
1197 | unsigned int cmd, unsigned long arg) | |
1198 | { | |
1199 | struct nbd_device *nbd = bdev->bd_disk->private_data; | |
1200 | struct nbd_config *config = nbd->config; | |
1201 | int error = -EINVAL; | |
1202 | ||
1203 | if (!capable(CAP_SYS_ADMIN)) | |
1204 | return -EPERM; | |
1205 | ||
1206 | mutex_lock(&nbd->config_lock); | |
1207 | ||
1208 | /* Don't allow ioctl operations on a nbd device that was created with | |
1209 | * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine. | |
1210 | */ | |
1211 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1212 | (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK)) | |
1213 | error = __nbd_ioctl(bdev, nbd, cmd, arg); | |
1214 | else | |
1215 | dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n"); | |
1216 | mutex_unlock(&nbd->config_lock); | |
1217 | return error; | |
1218 | } | |
1219 | ||
1220 | static struct nbd_config *nbd_alloc_config(void) | |
1221 | { | |
1222 | struct nbd_config *config; | |
1223 | ||
1224 | config = kzalloc(sizeof(struct nbd_config), GFP_NOFS); | |
1225 | if (!config) | |
1226 | return NULL; | |
1227 | atomic_set(&config->recv_threads, 0); | |
1228 | init_waitqueue_head(&config->recv_wq); | |
1229 | init_waitqueue_head(&config->conn_wait); | |
1230 | config->blksize = 1024; | |
1231 | atomic_set(&config->live_connections, 0); | |
1232 | try_module_get(THIS_MODULE); | |
1233 | return config; | |
1234 | } | |
1235 | ||
1236 | static int nbd_open(struct block_device *bdev, fmode_t mode) | |
1237 | { | |
1238 | struct nbd_device *nbd; | |
1239 | int ret = 0; | |
1240 | ||
1241 | mutex_lock(&nbd_index_mutex); | |
1242 | nbd = bdev->bd_disk->private_data; | |
1243 | if (!nbd) { | |
1244 | ret = -ENXIO; | |
1245 | goto out; | |
1246 | } | |
1247 | if (!refcount_inc_not_zero(&nbd->refs)) { | |
1248 | ret = -ENXIO; | |
1249 | goto out; | |
1250 | } | |
1251 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1252 | struct nbd_config *config; | |
1253 | ||
1254 | mutex_lock(&nbd->config_lock); | |
1255 | if (refcount_inc_not_zero(&nbd->config_refs)) { | |
1256 | mutex_unlock(&nbd->config_lock); | |
1257 | goto out; | |
1258 | } | |
1259 | config = nbd->config = nbd_alloc_config(); | |
1260 | if (!config) { | |
1261 | ret = -ENOMEM; | |
1262 | mutex_unlock(&nbd->config_lock); | |
1263 | goto out; | |
1264 | } | |
1265 | refcount_set(&nbd->config_refs, 1); | |
1266 | refcount_inc(&nbd->refs); | |
1267 | mutex_unlock(&nbd->config_lock); | |
1268 | } | |
1269 | out: | |
1270 | mutex_unlock(&nbd_index_mutex); | |
1271 | return ret; | |
1272 | } | |
1273 | ||
1274 | static void nbd_release(struct gendisk *disk, fmode_t mode) | |
1275 | { | |
1276 | struct nbd_device *nbd = disk->private_data; | |
1277 | nbd_config_put(nbd); | |
1278 | nbd_put(nbd); | |
1279 | } | |
1280 | ||
1281 | static const struct block_device_operations nbd_fops = | |
1282 | { | |
1283 | .owner = THIS_MODULE, | |
1284 | .open = nbd_open, | |
1285 | .release = nbd_release, | |
1286 | .ioctl = nbd_ioctl, | |
1287 | .compat_ioctl = nbd_ioctl, | |
1288 | }; | |
1289 | ||
1290 | #if IS_ENABLED(CONFIG_DEBUG_FS) | |
1291 | ||
1292 | static int nbd_dbg_tasks_show(struct seq_file *s, void *unused) | |
1293 | { | |
1294 | struct nbd_device *nbd = s->private; | |
1295 | ||
1296 | if (nbd->task_recv) | |
1297 | seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv)); | |
1298 | ||
1299 | return 0; | |
1300 | } | |
1301 | ||
1302 | static int nbd_dbg_tasks_open(struct inode *inode, struct file *file) | |
1303 | { | |
1304 | return single_open(file, nbd_dbg_tasks_show, inode->i_private); | |
1305 | } | |
1306 | ||
1307 | static const struct file_operations nbd_dbg_tasks_ops = { | |
1308 | .open = nbd_dbg_tasks_open, | |
1309 | .read = seq_read, | |
1310 | .llseek = seq_lseek, | |
1311 | .release = single_release, | |
1312 | }; | |
1313 | ||
1314 | static int nbd_dbg_flags_show(struct seq_file *s, void *unused) | |
1315 | { | |
1316 | struct nbd_device *nbd = s->private; | |
1317 | u32 flags = nbd->config->flags; | |
1318 | ||
1319 | seq_printf(s, "Hex: 0x%08x\n\n", flags); | |
1320 | ||
1321 | seq_puts(s, "Known flags:\n"); | |
1322 | ||
1323 | if (flags & NBD_FLAG_HAS_FLAGS) | |
1324 | seq_puts(s, "NBD_FLAG_HAS_FLAGS\n"); | |
1325 | if (flags & NBD_FLAG_READ_ONLY) | |
1326 | seq_puts(s, "NBD_FLAG_READ_ONLY\n"); | |
1327 | if (flags & NBD_FLAG_SEND_FLUSH) | |
1328 | seq_puts(s, "NBD_FLAG_SEND_FLUSH\n"); | |
1329 | if (flags & NBD_FLAG_SEND_FUA) | |
1330 | seq_puts(s, "NBD_FLAG_SEND_FUA\n"); | |
1331 | if (flags & NBD_FLAG_SEND_TRIM) | |
1332 | seq_puts(s, "NBD_FLAG_SEND_TRIM\n"); | |
1333 | ||
1334 | return 0; | |
1335 | } | |
1336 | ||
1337 | static int nbd_dbg_flags_open(struct inode *inode, struct file *file) | |
1338 | { | |
1339 | return single_open(file, nbd_dbg_flags_show, inode->i_private); | |
1340 | } | |
1341 | ||
1342 | static const struct file_operations nbd_dbg_flags_ops = { | |
1343 | .open = nbd_dbg_flags_open, | |
1344 | .read = seq_read, | |
1345 | .llseek = seq_lseek, | |
1346 | .release = single_release, | |
1347 | }; | |
1348 | ||
1349 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1350 | { | |
1351 | struct dentry *dir; | |
1352 | struct nbd_config *config = nbd->config; | |
1353 | ||
1354 | if (!nbd_dbg_dir) | |
1355 | return -EIO; | |
1356 | ||
1357 | dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir); | |
1358 | if (!dir) { | |
1359 | dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n", | |
1360 | nbd_name(nbd)); | |
1361 | return -EIO; | |
1362 | } | |
1363 | config->dbg_dir = dir; | |
1364 | ||
1365 | debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops); | |
1366 | debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize); | |
1367 | debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout); | |
1368 | debugfs_create_u64("blocksize", 0444, dir, &config->blksize); | |
1369 | debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops); | |
1370 | ||
1371 | return 0; | |
1372 | } | |
1373 | ||
1374 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1375 | { | |
1376 | debugfs_remove_recursive(nbd->config->dbg_dir); | |
1377 | } | |
1378 | ||
1379 | static int nbd_dbg_init(void) | |
1380 | { | |
1381 | struct dentry *dbg_dir; | |
1382 | ||
1383 | dbg_dir = debugfs_create_dir("nbd", NULL); | |
1384 | if (!dbg_dir) | |
1385 | return -EIO; | |
1386 | ||
1387 | nbd_dbg_dir = dbg_dir; | |
1388 | ||
1389 | return 0; | |
1390 | } | |
1391 | ||
1392 | static void nbd_dbg_close(void) | |
1393 | { | |
1394 | debugfs_remove_recursive(nbd_dbg_dir); | |
1395 | } | |
1396 | ||
1397 | #else /* IS_ENABLED(CONFIG_DEBUG_FS) */ | |
1398 | ||
1399 | static int nbd_dev_dbg_init(struct nbd_device *nbd) | |
1400 | { | |
1401 | return 0; | |
1402 | } | |
1403 | ||
1404 | static void nbd_dev_dbg_close(struct nbd_device *nbd) | |
1405 | { | |
1406 | } | |
1407 | ||
1408 | static int nbd_dbg_init(void) | |
1409 | { | |
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | static void nbd_dbg_close(void) | |
1414 | { | |
1415 | } | |
1416 | ||
1417 | #endif | |
1418 | ||
1419 | static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq, | |
1420 | unsigned int hctx_idx, unsigned int numa_node) | |
1421 | { | |
1422 | struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1423 | cmd->nbd = set->driver_data; | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | static const struct blk_mq_ops nbd_mq_ops = { | |
1428 | .queue_rq = nbd_queue_rq, | |
1429 | .complete = nbd_complete_rq, | |
1430 | .init_request = nbd_init_request, | |
1431 | .timeout = nbd_xmit_timeout, | |
1432 | }; | |
1433 | ||
1434 | static int nbd_dev_add(int index) | |
1435 | { | |
1436 | struct nbd_device *nbd; | |
1437 | struct gendisk *disk; | |
1438 | struct request_queue *q; | |
1439 | int err = -ENOMEM; | |
1440 | ||
1441 | nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL); | |
1442 | if (!nbd) | |
1443 | goto out; | |
1444 | ||
1445 | disk = alloc_disk(1 << part_shift); | |
1446 | if (!disk) | |
1447 | goto out_free_nbd; | |
1448 | ||
1449 | if (index >= 0) { | |
1450 | err = idr_alloc(&nbd_index_idr, nbd, index, index + 1, | |
1451 | GFP_KERNEL); | |
1452 | if (err == -ENOSPC) | |
1453 | err = -EEXIST; | |
1454 | } else { | |
1455 | err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL); | |
1456 | if (err >= 0) | |
1457 | index = err; | |
1458 | } | |
1459 | if (err < 0) | |
1460 | goto out_free_disk; | |
1461 | ||
1462 | nbd->index = index; | |
1463 | nbd->disk = disk; | |
1464 | nbd->tag_set.ops = &nbd_mq_ops; | |
1465 | nbd->tag_set.nr_hw_queues = 1; | |
1466 | nbd->tag_set.queue_depth = 128; | |
1467 | nbd->tag_set.numa_node = NUMA_NO_NODE; | |
1468 | nbd->tag_set.cmd_size = sizeof(struct nbd_cmd); | |
1469 | nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | | |
1470 | BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING; | |
1471 | nbd->tag_set.driver_data = nbd; | |
1472 | ||
1473 | err = blk_mq_alloc_tag_set(&nbd->tag_set); | |
1474 | if (err) | |
1475 | goto out_free_idr; | |
1476 | ||
1477 | q = blk_mq_init_queue(&nbd->tag_set); | |
1478 | if (IS_ERR(q)) { | |
1479 | err = PTR_ERR(q); | |
1480 | goto out_free_tags; | |
1481 | } | |
1482 | disk->queue = q; | |
1483 | ||
1484 | /* | |
1485 | * Tell the block layer that we are not a rotational device | |
1486 | */ | |
1487 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue); | |
1488 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue); | |
1489 | disk->queue->limits.discard_granularity = 512; | |
1490 | blk_queue_max_discard_sectors(disk->queue, UINT_MAX); | |
1491 | blk_queue_max_segment_size(disk->queue, UINT_MAX); | |
1492 | blk_queue_max_segments(disk->queue, USHRT_MAX); | |
1493 | blk_queue_max_hw_sectors(disk->queue, 65536); | |
1494 | disk->queue->limits.max_sectors = 256; | |
1495 | ||
1496 | mutex_init(&nbd->config_lock); | |
1497 | refcount_set(&nbd->config_refs, 0); | |
1498 | refcount_set(&nbd->refs, 1); | |
1499 | INIT_LIST_HEAD(&nbd->list); | |
1500 | disk->major = NBD_MAJOR; | |
1501 | disk->first_minor = index << part_shift; | |
1502 | disk->fops = &nbd_fops; | |
1503 | disk->private_data = nbd; | |
1504 | sprintf(disk->disk_name, "nbd%d", index); | |
1505 | add_disk(disk); | |
1506 | nbd_total_devices++; | |
1507 | return index; | |
1508 | ||
1509 | out_free_tags: | |
1510 | blk_mq_free_tag_set(&nbd->tag_set); | |
1511 | out_free_idr: | |
1512 | idr_remove(&nbd_index_idr, index); | |
1513 | out_free_disk: | |
1514 | put_disk(disk); | |
1515 | out_free_nbd: | |
1516 | kfree(nbd); | |
1517 | out: | |
1518 | return err; | |
1519 | } | |
1520 | ||
1521 | static int find_free_cb(int id, void *ptr, void *data) | |
1522 | { | |
1523 | struct nbd_device *nbd = ptr; | |
1524 | struct nbd_device **found = data; | |
1525 | ||
1526 | if (!refcount_read(&nbd->config_refs)) { | |
1527 | *found = nbd; | |
1528 | return 1; | |
1529 | } | |
1530 | return 0; | |
1531 | } | |
1532 | ||
1533 | /* Netlink interface. */ | |
1534 | static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = { | |
1535 | [NBD_ATTR_INDEX] = { .type = NLA_U32 }, | |
1536 | [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 }, | |
1537 | [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 }, | |
1538 | [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 }, | |
1539 | [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 }, | |
1540 | [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 }, | |
1541 | [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED}, | |
1542 | [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 }, | |
1543 | [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED}, | |
1544 | }; | |
1545 | ||
1546 | static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = { | |
1547 | [NBD_SOCK_FD] = { .type = NLA_U32 }, | |
1548 | }; | |
1549 | ||
1550 | /* We don't use this right now since we don't parse the incoming list, but we | |
1551 | * still want it here so userspace knows what to expect. | |
1552 | */ | |
1553 | static struct nla_policy __attribute__((unused)) | |
1554 | nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = { | |
1555 | [NBD_DEVICE_INDEX] = { .type = NLA_U32 }, | |
1556 | [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 }, | |
1557 | }; | |
1558 | ||
1559 | static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info) | |
1560 | { | |
1561 | struct nbd_device *nbd = NULL; | |
1562 | struct nbd_config *config; | |
1563 | int index = -1; | |
1564 | int ret; | |
1565 | bool put_dev = false; | |
1566 | ||
1567 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1568 | return -EPERM; | |
1569 | ||
1570 | if (info->attrs[NBD_ATTR_INDEX]) | |
1571 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1572 | if (!info->attrs[NBD_ATTR_SOCKETS]) { | |
1573 | printk(KERN_ERR "nbd: must specify at least one socket\n"); | |
1574 | return -EINVAL; | |
1575 | } | |
1576 | if (!info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1577 | printk(KERN_ERR "nbd: must specify a size in bytes for the device\n"); | |
1578 | return -EINVAL; | |
1579 | } | |
1580 | again: | |
1581 | mutex_lock(&nbd_index_mutex); | |
1582 | if (index == -1) { | |
1583 | ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd); | |
1584 | if (ret == 0) { | |
1585 | int new_index; | |
1586 | new_index = nbd_dev_add(-1); | |
1587 | if (new_index < 0) { | |
1588 | mutex_unlock(&nbd_index_mutex); | |
1589 | printk(KERN_ERR "nbd: failed to add new device\n"); | |
1590 | return ret; | |
1591 | } | |
1592 | nbd = idr_find(&nbd_index_idr, new_index); | |
1593 | } | |
1594 | } else { | |
1595 | nbd = idr_find(&nbd_index_idr, index); | |
1596 | } | |
1597 | if (!nbd) { | |
1598 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", | |
1599 | index); | |
1600 | mutex_unlock(&nbd_index_mutex); | |
1601 | return -EINVAL; | |
1602 | } | |
1603 | if (!refcount_inc_not_zero(&nbd->refs)) { | |
1604 | mutex_unlock(&nbd_index_mutex); | |
1605 | if (index == -1) | |
1606 | goto again; | |
1607 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1608 | index); | |
1609 | return -EINVAL; | |
1610 | } | |
1611 | mutex_unlock(&nbd_index_mutex); | |
1612 | ||
1613 | mutex_lock(&nbd->config_lock); | |
1614 | if (refcount_read(&nbd->config_refs)) { | |
1615 | mutex_unlock(&nbd->config_lock); | |
1616 | nbd_put(nbd); | |
1617 | if (index == -1) | |
1618 | goto again; | |
1619 | printk(KERN_ERR "nbd: nbd%d already in use\n", index); | |
1620 | return -EBUSY; | |
1621 | } | |
1622 | if (WARN_ON(nbd->config)) { | |
1623 | mutex_unlock(&nbd->config_lock); | |
1624 | nbd_put(nbd); | |
1625 | return -EINVAL; | |
1626 | } | |
1627 | config = nbd->config = nbd_alloc_config(); | |
1628 | if (!nbd->config) { | |
1629 | mutex_unlock(&nbd->config_lock); | |
1630 | nbd_put(nbd); | |
1631 | printk(KERN_ERR "nbd: couldn't allocate config\n"); | |
1632 | return -ENOMEM; | |
1633 | } | |
1634 | refcount_set(&nbd->config_refs, 1); | |
1635 | set_bit(NBD_BOUND, &config->runtime_flags); | |
1636 | ||
1637 | if (info->attrs[NBD_ATTR_SIZE_BYTES]) { | |
1638 | u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]); | |
1639 | nbd_size_set(nbd, config->blksize, | |
1640 | div64_u64(bytes, config->blksize)); | |
1641 | } | |
1642 | if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) { | |
1643 | u64 bsize = | |
1644 | nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]); | |
1645 | nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize)); | |
1646 | } | |
1647 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1648 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1649 | nbd->tag_set.timeout = timeout * HZ; | |
1650 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1651 | } | |
1652 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { | |
1653 | config->dead_conn_timeout = | |
1654 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1655 | config->dead_conn_timeout *= HZ; | |
1656 | } | |
1657 | if (info->attrs[NBD_ATTR_SERVER_FLAGS]) | |
1658 | config->flags = | |
1659 | nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]); | |
1660 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { | |
1661 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1662 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1663 | set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1664 | &config->runtime_flags); | |
1665 | put_dev = true; | |
1666 | } | |
1667 | } | |
1668 | ||
1669 | if (info->attrs[NBD_ATTR_SOCKETS]) { | |
1670 | struct nlattr *attr; | |
1671 | int rem, fd; | |
1672 | ||
1673 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1674 | rem) { | |
1675 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1676 | ||
1677 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1678 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1679 | ret = -EINVAL; | |
1680 | goto out; | |
1681 | } | |
1682 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
1683 | nbd_sock_policy, info->extack); | |
1684 | if (ret != 0) { | |
1685 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1686 | ret = -EINVAL; | |
1687 | goto out; | |
1688 | } | |
1689 | if (!socks[NBD_SOCK_FD]) | |
1690 | continue; | |
1691 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1692 | ret = nbd_add_socket(nbd, fd, true); | |
1693 | if (ret) | |
1694 | goto out; | |
1695 | } | |
1696 | } | |
1697 | ret = nbd_start_device(nbd); | |
1698 | out: | |
1699 | mutex_unlock(&nbd->config_lock); | |
1700 | if (!ret) { | |
1701 | set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags); | |
1702 | refcount_inc(&nbd->config_refs); | |
1703 | nbd_connect_reply(info, nbd->index); | |
1704 | } | |
1705 | nbd_config_put(nbd); | |
1706 | if (put_dev) | |
1707 | nbd_put(nbd); | |
1708 | return ret; | |
1709 | } | |
1710 | ||
1711 | static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info) | |
1712 | { | |
1713 | struct nbd_device *nbd; | |
1714 | int index; | |
1715 | ||
1716 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1717 | return -EPERM; | |
1718 | ||
1719 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1720 | printk(KERN_ERR "nbd: must specify an index to disconnect\n"); | |
1721 | return -EINVAL; | |
1722 | } | |
1723 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1724 | mutex_lock(&nbd_index_mutex); | |
1725 | nbd = idr_find(&nbd_index_idr, index); | |
1726 | if (!nbd) { | |
1727 | mutex_unlock(&nbd_index_mutex); | |
1728 | printk(KERN_ERR "nbd: couldn't find device at index %d\n", | |
1729 | index); | |
1730 | return -EINVAL; | |
1731 | } | |
1732 | if (!refcount_inc_not_zero(&nbd->refs)) { | |
1733 | mutex_unlock(&nbd_index_mutex); | |
1734 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1735 | index); | |
1736 | return -EINVAL; | |
1737 | } | |
1738 | mutex_unlock(&nbd_index_mutex); | |
1739 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1740 | nbd_put(nbd); | |
1741 | return 0; | |
1742 | } | |
1743 | mutex_lock(&nbd->config_lock); | |
1744 | nbd_disconnect(nbd); | |
1745 | mutex_unlock(&nbd->config_lock); | |
1746 | if (test_and_clear_bit(NBD_HAS_CONFIG_REF, | |
1747 | &nbd->config->runtime_flags)) | |
1748 | nbd_config_put(nbd); | |
1749 | nbd_config_put(nbd); | |
1750 | nbd_put(nbd); | |
1751 | return 0; | |
1752 | } | |
1753 | ||
1754 | static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info) | |
1755 | { | |
1756 | struct nbd_device *nbd = NULL; | |
1757 | struct nbd_config *config; | |
1758 | int index; | |
1759 | int ret = -EINVAL; | |
1760 | bool put_dev = false; | |
1761 | ||
1762 | if (!netlink_capable(skb, CAP_SYS_ADMIN)) | |
1763 | return -EPERM; | |
1764 | ||
1765 | if (!info->attrs[NBD_ATTR_INDEX]) { | |
1766 | printk(KERN_ERR "nbd: must specify a device to reconfigure\n"); | |
1767 | return -EINVAL; | |
1768 | } | |
1769 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1770 | mutex_lock(&nbd_index_mutex); | |
1771 | nbd = idr_find(&nbd_index_idr, index); | |
1772 | if (!nbd) { | |
1773 | mutex_unlock(&nbd_index_mutex); | |
1774 | printk(KERN_ERR "nbd: couldn't find a device at index %d\n", | |
1775 | index); | |
1776 | return -EINVAL; | |
1777 | } | |
1778 | if (!refcount_inc_not_zero(&nbd->refs)) { | |
1779 | mutex_unlock(&nbd_index_mutex); | |
1780 | printk(KERN_ERR "nbd: device at index %d is going down\n", | |
1781 | index); | |
1782 | return -EINVAL; | |
1783 | } | |
1784 | mutex_unlock(&nbd_index_mutex); | |
1785 | ||
1786 | if (!refcount_inc_not_zero(&nbd->config_refs)) { | |
1787 | dev_err(nbd_to_dev(nbd), | |
1788 | "not configured, cannot reconfigure\n"); | |
1789 | nbd_put(nbd); | |
1790 | return -EINVAL; | |
1791 | } | |
1792 | ||
1793 | mutex_lock(&nbd->config_lock); | |
1794 | config = nbd->config; | |
1795 | if (!test_bit(NBD_BOUND, &config->runtime_flags) || | |
1796 | !nbd->task_recv) { | |
1797 | dev_err(nbd_to_dev(nbd), | |
1798 | "not configured, cannot reconfigure\n"); | |
1799 | goto out; | |
1800 | } | |
1801 | ||
1802 | if (info->attrs[NBD_ATTR_TIMEOUT]) { | |
1803 | u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]); | |
1804 | nbd->tag_set.timeout = timeout * HZ; | |
1805 | blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ); | |
1806 | } | |
1807 | if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) { | |
1808 | config->dead_conn_timeout = | |
1809 | nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]); | |
1810 | config->dead_conn_timeout *= HZ; | |
1811 | } | |
1812 | if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) { | |
1813 | u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]); | |
1814 | if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) { | |
1815 | if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT, | |
1816 | &config->runtime_flags)) | |
1817 | put_dev = true; | |
1818 | } else { | |
1819 | if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT, | |
1820 | &config->runtime_flags)) | |
1821 | refcount_inc(&nbd->refs); | |
1822 | } | |
1823 | } | |
1824 | ||
1825 | if (info->attrs[NBD_ATTR_SOCKETS]) { | |
1826 | struct nlattr *attr; | |
1827 | int rem, fd; | |
1828 | ||
1829 | nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS], | |
1830 | rem) { | |
1831 | struct nlattr *socks[NBD_SOCK_MAX+1]; | |
1832 | ||
1833 | if (nla_type(attr) != NBD_SOCK_ITEM) { | |
1834 | printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n"); | |
1835 | ret = -EINVAL; | |
1836 | goto out; | |
1837 | } | |
1838 | ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr, | |
1839 | nbd_sock_policy, info->extack); | |
1840 | if (ret != 0) { | |
1841 | printk(KERN_ERR "nbd: error processing sock list\n"); | |
1842 | ret = -EINVAL; | |
1843 | goto out; | |
1844 | } | |
1845 | if (!socks[NBD_SOCK_FD]) | |
1846 | continue; | |
1847 | fd = (int)nla_get_u32(socks[NBD_SOCK_FD]); | |
1848 | ret = nbd_reconnect_socket(nbd, fd); | |
1849 | if (ret) { | |
1850 | if (ret == -ENOSPC) | |
1851 | ret = 0; | |
1852 | goto out; | |
1853 | } | |
1854 | dev_info(nbd_to_dev(nbd), "reconnected socket\n"); | |
1855 | } | |
1856 | } | |
1857 | out: | |
1858 | mutex_unlock(&nbd->config_lock); | |
1859 | nbd_config_put(nbd); | |
1860 | nbd_put(nbd); | |
1861 | if (put_dev) | |
1862 | nbd_put(nbd); | |
1863 | return ret; | |
1864 | } | |
1865 | ||
1866 | static const struct genl_ops nbd_connect_genl_ops[] = { | |
1867 | { | |
1868 | .cmd = NBD_CMD_CONNECT, | |
1869 | .policy = nbd_attr_policy, | |
1870 | .doit = nbd_genl_connect, | |
1871 | }, | |
1872 | { | |
1873 | .cmd = NBD_CMD_DISCONNECT, | |
1874 | .policy = nbd_attr_policy, | |
1875 | .doit = nbd_genl_disconnect, | |
1876 | }, | |
1877 | { | |
1878 | .cmd = NBD_CMD_RECONFIGURE, | |
1879 | .policy = nbd_attr_policy, | |
1880 | .doit = nbd_genl_reconfigure, | |
1881 | }, | |
1882 | { | |
1883 | .cmd = NBD_CMD_STATUS, | |
1884 | .policy = nbd_attr_policy, | |
1885 | .doit = nbd_genl_status, | |
1886 | }, | |
1887 | }; | |
1888 | ||
1889 | static const struct genl_multicast_group nbd_mcast_grps[] = { | |
1890 | { .name = NBD_GENL_MCAST_GROUP_NAME, }, | |
1891 | }; | |
1892 | ||
1893 | static struct genl_family nbd_genl_family __ro_after_init = { | |
1894 | .hdrsize = 0, | |
1895 | .name = NBD_GENL_FAMILY_NAME, | |
1896 | .version = NBD_GENL_VERSION, | |
1897 | .module = THIS_MODULE, | |
1898 | .ops = nbd_connect_genl_ops, | |
1899 | .n_ops = ARRAY_SIZE(nbd_connect_genl_ops), | |
1900 | .maxattr = NBD_ATTR_MAX, | |
1901 | .mcgrps = nbd_mcast_grps, | |
1902 | .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps), | |
1903 | }; | |
1904 | ||
1905 | static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply) | |
1906 | { | |
1907 | struct nlattr *dev_opt; | |
1908 | u8 connected = 0; | |
1909 | int ret; | |
1910 | ||
1911 | /* This is a little racey, but for status it's ok. The | |
1912 | * reason we don't take a ref here is because we can't | |
1913 | * take a ref in the index == -1 case as we would need | |
1914 | * to put under the nbd_index_mutex, which could | |
1915 | * deadlock if we are configured to remove ourselves | |
1916 | * once we're disconnected. | |
1917 | */ | |
1918 | if (refcount_read(&nbd->config_refs)) | |
1919 | connected = 1; | |
1920 | dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM); | |
1921 | if (!dev_opt) | |
1922 | return -EMSGSIZE; | |
1923 | ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index); | |
1924 | if (ret) | |
1925 | return -EMSGSIZE; | |
1926 | ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED, | |
1927 | connected); | |
1928 | if (ret) | |
1929 | return -EMSGSIZE; | |
1930 | nla_nest_end(reply, dev_opt); | |
1931 | return 0; | |
1932 | } | |
1933 | ||
1934 | static int status_cb(int id, void *ptr, void *data) | |
1935 | { | |
1936 | struct nbd_device *nbd = ptr; | |
1937 | return populate_nbd_status(nbd, (struct sk_buff *)data); | |
1938 | } | |
1939 | ||
1940 | static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info) | |
1941 | { | |
1942 | struct nlattr *dev_list; | |
1943 | struct sk_buff *reply; | |
1944 | void *reply_head; | |
1945 | size_t msg_size; | |
1946 | int index = -1; | |
1947 | int ret = -ENOMEM; | |
1948 | ||
1949 | if (info->attrs[NBD_ATTR_INDEX]) | |
1950 | index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]); | |
1951 | ||
1952 | mutex_lock(&nbd_index_mutex); | |
1953 | ||
1954 | msg_size = nla_total_size(nla_attr_size(sizeof(u32)) + | |
1955 | nla_attr_size(sizeof(u8))); | |
1956 | msg_size *= (index == -1) ? nbd_total_devices : 1; | |
1957 | ||
1958 | reply = genlmsg_new(msg_size, GFP_KERNEL); | |
1959 | if (!reply) | |
1960 | goto out; | |
1961 | reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0, | |
1962 | NBD_CMD_STATUS); | |
1963 | if (!reply_head) { | |
1964 | nlmsg_free(reply); | |
1965 | goto out; | |
1966 | } | |
1967 | ||
1968 | dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST); | |
1969 | if (index == -1) { | |
1970 | ret = idr_for_each(&nbd_index_idr, &status_cb, reply); | |
1971 | if (ret) { | |
1972 | nlmsg_free(reply); | |
1973 | goto out; | |
1974 | } | |
1975 | } else { | |
1976 | struct nbd_device *nbd; | |
1977 | nbd = idr_find(&nbd_index_idr, index); | |
1978 | if (nbd) { | |
1979 | ret = populate_nbd_status(nbd, reply); | |
1980 | if (ret) { | |
1981 | nlmsg_free(reply); | |
1982 | goto out; | |
1983 | } | |
1984 | } | |
1985 | } | |
1986 | nla_nest_end(reply, dev_list); | |
1987 | genlmsg_end(reply, reply_head); | |
1988 | genlmsg_reply(reply, info); | |
1989 | ret = 0; | |
1990 | out: | |
1991 | mutex_unlock(&nbd_index_mutex); | |
1992 | return ret; | |
1993 | } | |
1994 | ||
1995 | static void nbd_connect_reply(struct genl_info *info, int index) | |
1996 | { | |
1997 | struct sk_buff *skb; | |
1998 | void *msg_head; | |
1999 | int ret; | |
2000 | ||
2001 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
2002 | if (!skb) | |
2003 | return; | |
2004 | msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0, | |
2005 | NBD_CMD_CONNECT); | |
2006 | if (!msg_head) { | |
2007 | nlmsg_free(skb); | |
2008 | return; | |
2009 | } | |
2010 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
2011 | if (ret) { | |
2012 | nlmsg_free(skb); | |
2013 | return; | |
2014 | } | |
2015 | genlmsg_end(skb, msg_head); | |
2016 | genlmsg_reply(skb, info); | |
2017 | } | |
2018 | ||
2019 | static void nbd_mcast_index(int index) | |
2020 | { | |
2021 | struct sk_buff *skb; | |
2022 | void *msg_head; | |
2023 | int ret; | |
2024 | ||
2025 | skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL); | |
2026 | if (!skb) | |
2027 | return; | |
2028 | msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0, | |
2029 | NBD_CMD_LINK_DEAD); | |
2030 | if (!msg_head) { | |
2031 | nlmsg_free(skb); | |
2032 | return; | |
2033 | } | |
2034 | ret = nla_put_u32(skb, NBD_ATTR_INDEX, index); | |
2035 | if (ret) { | |
2036 | nlmsg_free(skb); | |
2037 | return; | |
2038 | } | |
2039 | genlmsg_end(skb, msg_head); | |
2040 | genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL); | |
2041 | } | |
2042 | ||
2043 | static void nbd_dead_link_work(struct work_struct *work) | |
2044 | { | |
2045 | struct link_dead_args *args = container_of(work, struct link_dead_args, | |
2046 | work); | |
2047 | nbd_mcast_index(args->index); | |
2048 | kfree(args); | |
2049 | } | |
2050 | ||
2051 | static int __init nbd_init(void) | |
2052 | { | |
2053 | int i; | |
2054 | ||
2055 | BUILD_BUG_ON(sizeof(struct nbd_request) != 28); | |
2056 | ||
2057 | if (max_part < 0) { | |
2058 | printk(KERN_ERR "nbd: max_part must be >= 0\n"); | |
2059 | return -EINVAL; | |
2060 | } | |
2061 | ||
2062 | part_shift = 0; | |
2063 | if (max_part > 0) { | |
2064 | part_shift = fls(max_part); | |
2065 | ||
2066 | /* | |
2067 | * Adjust max_part according to part_shift as it is exported | |
2068 | * to user space so that user can know the max number of | |
2069 | * partition kernel should be able to manage. | |
2070 | * | |
2071 | * Note that -1 is required because partition 0 is reserved | |
2072 | * for the whole disk. | |
2073 | */ | |
2074 | max_part = (1UL << part_shift) - 1; | |
2075 | } | |
2076 | ||
2077 | if ((1UL << part_shift) > DISK_MAX_PARTS) | |
2078 | return -EINVAL; | |
2079 | ||
2080 | if (nbds_max > 1UL << (MINORBITS - part_shift)) | |
2081 | return -EINVAL; | |
2082 | recv_workqueue = alloc_workqueue("knbd-recv", | |
2083 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); | |
2084 | if (!recv_workqueue) | |
2085 | return -ENOMEM; | |
2086 | ||
2087 | if (register_blkdev(NBD_MAJOR, "nbd")) { | |
2088 | destroy_workqueue(recv_workqueue); | |
2089 | return -EIO; | |
2090 | } | |
2091 | ||
2092 | if (genl_register_family(&nbd_genl_family)) { | |
2093 | unregister_blkdev(NBD_MAJOR, "nbd"); | |
2094 | destroy_workqueue(recv_workqueue); | |
2095 | return -EINVAL; | |
2096 | } | |
2097 | nbd_dbg_init(); | |
2098 | ||
2099 | mutex_lock(&nbd_index_mutex); | |
2100 | for (i = 0; i < nbds_max; i++) | |
2101 | nbd_dev_add(i); | |
2102 | mutex_unlock(&nbd_index_mutex); | |
2103 | return 0; | |
2104 | } | |
2105 | ||
2106 | static int nbd_exit_cb(int id, void *ptr, void *data) | |
2107 | { | |
2108 | struct list_head *list = (struct list_head *)data; | |
2109 | struct nbd_device *nbd = ptr; | |
2110 | ||
2111 | list_add_tail(&nbd->list, list); | |
2112 | return 0; | |
2113 | } | |
2114 | ||
2115 | static void __exit nbd_cleanup(void) | |
2116 | { | |
2117 | struct nbd_device *nbd; | |
2118 | LIST_HEAD(del_list); | |
2119 | ||
2120 | nbd_dbg_close(); | |
2121 | ||
2122 | mutex_lock(&nbd_index_mutex); | |
2123 | idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list); | |
2124 | mutex_unlock(&nbd_index_mutex); | |
2125 | ||
2126 | while (!list_empty(&del_list)) { | |
2127 | nbd = list_first_entry(&del_list, struct nbd_device, list); | |
2128 | list_del_init(&nbd->list); | |
2129 | if (refcount_read(&nbd->refs) != 1) | |
2130 | printk(KERN_ERR "nbd: possibly leaking a device\n"); | |
2131 | nbd_put(nbd); | |
2132 | } | |
2133 | ||
2134 | idr_destroy(&nbd_index_idr); | |
2135 | genl_unregister_family(&nbd_genl_family); | |
2136 | destroy_workqueue(recv_workqueue); | |
2137 | unregister_blkdev(NBD_MAJOR, "nbd"); | |
2138 | } | |
2139 | ||
2140 | module_init(nbd_init); | |
2141 | module_exit(nbd_cleanup); | |
2142 | ||
2143 | MODULE_DESCRIPTION("Network Block Device"); | |
2144 | MODULE_LICENSE("GPL"); | |
2145 | ||
2146 | module_param(nbds_max, int, 0444); | |
2147 | MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)"); | |
2148 | module_param(max_part, int, 0444); | |
2149 | MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)"); |