?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'block-5.11-2021-01-10' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / drivers / block / nbd.c
blob6727358e147dd8f8d3fb738951276f5881c20a66
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Network block device - make block devices work over TCP
4 *
5 * Note that you can not swap over this thing, yet. Seems to work but
6 * deadlocks sometimes - you can not swap over TCP in general.
7 *
8 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
9 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
10 *
11 * (part of code stolen from loop.c)
12 */
13
14 #include <linux/major.h>
15
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched.h>
20 #include <linux/sched/mm.h>
21 #include <linux/fs.h>
22 #include <linux/bio.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/file.h>
26 #include <linux/ioctl.h>
27 #include <linux/mutex.h>
28 #include <linux/compiler.h>
29 #include <linux/completion.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 #define CREATE_TRACE_POINTS
48 #include <trace/events/nbd.h>
49
50 static DEFINE_IDR(nbd_index_idr);
51 static DEFINE_MUTEX(nbd_index_mutex);
52 static int nbd_total_devices = 0;
53
54 struct nbd_sock {
55 struct socket *sock;
56 struct mutex tx_lock;
57 struct request *pending;
58 int sent;
59 bool dead;
60 int fallback_index;
61 int cookie;
62 };
63
64 struct recv_thread_args {
65 struct work_struct work;
66 struct nbd_device *nbd;
67 int index;
68 };
69
70 struct link_dead_args {
71 struct work_struct work;
72 int index;
73 };
74
75 #define NBD_RT_TIMEDOUT 0
76 #define NBD_RT_DISCONNECT_REQUESTED 1
77 #define NBD_RT_DISCONNECTED 2
78 #define NBD_RT_HAS_PID_FILE 3
79 #define NBD_RT_HAS_CONFIG_REF 4
80 #define NBD_RT_BOUND 5
81 #define NBD_RT_DESTROY_ON_DISCONNECT 6
82 #define NBD_RT_DISCONNECT_ON_CLOSE 7
83
84 #define NBD_DESTROY_ON_DISCONNECT 0
85 #define NBD_DISCONNECT_REQUESTED 1
86
87 struct nbd_config {
88 u32 flags;
89 unsigned long runtime_flags;
90 u64 dead_conn_timeout;
91
92 struct nbd_sock **socks;
93 int num_connections;
94 atomic_t live_connections;
95 wait_queue_head_t conn_wait;
96
97 atomic_t recv_threads;
98 wait_queue_head_t recv_wq;
99 loff_t blksize;
100 loff_t bytesize;
101 #if IS_ENABLED(CONFIG_DEBUG_FS)
102 struct dentry *dbg_dir;
103 #endif
104 };
105
106 struct nbd_device {
107 struct blk_mq_tag_set tag_set;
108
109 int index;
110 refcount_t config_refs;
111 refcount_t refs;
112 struct nbd_config *config;
113 struct mutex config_lock;
114 struct gendisk *disk;
115 struct workqueue_struct *recv_workq;
116
117 struct list_head list;
118 struct task_struct *task_recv;
119 struct task_struct *task_setup;
120
121 struct completion *destroy_complete;
122 unsigned long flags;
123 };
124
125 #define NBD_CMD_REQUEUED 1
126
127 struct nbd_cmd {
128 struct nbd_device *nbd;
129 struct mutex lock;
130 int index;
131 int cookie;
132 int retries;
133 blk_status_t status;
134 unsigned long flags;
135 u32 cmd_cookie;
136 };
137
138 #if IS_ENABLED(CONFIG_DEBUG_FS)
139 static struct dentry *nbd_dbg_dir;
140 #endif
141
142 #define nbd_name(nbd) ((nbd)->disk->disk_name)
143
144 #define NBD_MAGIC 0x68797548
145
146 #define NBD_DEF_BLKSIZE 1024
147
148 static unsigned int nbds_max = 16;
149 static int max_part = 16;
150 static int part_shift;
151
152 static int nbd_dev_dbg_init(struct nbd_device *nbd);
153 static void nbd_dev_dbg_close(struct nbd_device *nbd);
154 static void nbd_config_put(struct nbd_device *nbd);
155 static void nbd_connect_reply(struct genl_info *info, int index);
156 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
157 static void nbd_dead_link_work(struct work_struct *work);
158 static void nbd_disconnect_and_put(struct nbd_device *nbd);
159
160 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
161 {
162 return disk_to_dev(nbd->disk);
163 }
164
165 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
166 {
167 struct request *req = blk_mq_rq_from_pdu(cmd);
168
169 if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
170 blk_mq_requeue_request(req, true);
171 }
172
173 #define NBD_COOKIE_BITS 32
174
175 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
176 {
177 struct request *req = blk_mq_rq_from_pdu(cmd);
178 u32 tag = blk_mq_unique_tag(req);
179 u64 cookie = cmd->cmd_cookie;
180
181 return (cookie << NBD_COOKIE_BITS) | tag;
182 }
183
184 static u32 nbd_handle_to_tag(u64 handle)
185 {
186 return (u32)handle;
187 }
188
189 static u32 nbd_handle_to_cookie(u64 handle)
190 {
191 return (u32)(handle >> NBD_COOKIE_BITS);
192 }
193
194 static const char *nbdcmd_to_ascii(int cmd)
195 {
196 switch (cmd) {
197 case NBD_CMD_READ: return "read";
198 case NBD_CMD_WRITE: return "write";
199 case NBD_CMD_DISC: return "disconnect";
200 case NBD_CMD_FLUSH: return "flush";
201 case NBD_CMD_TRIM: return "trim/discard";
202 }
203 return "invalid";
204 }
205
206 static ssize_t pid_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
208 {
209 struct gendisk *disk = dev_to_disk(dev);
210 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
211
212 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
213 }
214
215 static const struct device_attribute pid_attr = {
216 .attr = { .name = "pid", .mode = 0444},
217 .show = pid_show,
218 };
219
220 static void nbd_dev_remove(struct nbd_device *nbd)
221 {
222 struct gendisk *disk = nbd->disk;
223 struct request_queue *q;
224
225 if (disk) {
226 q = disk->queue;
227 del_gendisk(disk);
228 blk_cleanup_queue(q);
229 blk_mq_free_tag_set(&nbd->tag_set);
230 disk->private_data = NULL;
231 put_disk(disk);
232 }
233
234 /*
235 * Place this in the last just before the nbd is freed to
236 * make sure that the disk and the related kobject are also
237 * totally removed to avoid duplicate creation of the same
238 * one.
239 */
240 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) && nbd->destroy_complete)
241 complete(nbd->destroy_complete);
242
243 kfree(nbd);
244 }
245
246 static void nbd_put(struct nbd_device *nbd)
247 {
248 if (refcount_dec_and_mutex_lock(&nbd->refs,
249 &nbd_index_mutex)) {
250 idr_remove(&nbd_index_idr, nbd->index);
251 nbd_dev_remove(nbd);
252 mutex_unlock(&nbd_index_mutex);
253 }
254 }
255
256 static int nbd_disconnected(struct nbd_config *config)
257 {
258 return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
259 test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
260 }
261
262 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
263 int notify)
264 {
265 if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
266 struct link_dead_args *args;
267 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
268 if (args) {
269 INIT_WORK(&args->work, nbd_dead_link_work);
270 args->index = nbd->index;
271 queue_work(system_wq, &args->work);
272 }
273 }
274 if (!nsock->dead) {
275 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
276 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
277 if (test_and_clear_bit(NBD_RT_DISCONNECT_REQUESTED,
278 &nbd->config->runtime_flags)) {
279 set_bit(NBD_RT_DISCONNECTED,
280 &nbd->config->runtime_flags);
281 dev_info(nbd_to_dev(nbd),
282 "Disconnected due to user request.\n");
283 }
284 }
285 }
286 nsock->dead = true;
287 nsock->pending = NULL;
288 nsock->sent = 0;
289 }
290
291 static void nbd_size_clear(struct nbd_device *nbd)
292 {
293 if (nbd->config->bytesize) {
294 set_capacity(nbd->disk, 0);
295 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
296 }
297 }
298
299 static int nbd_set_size(struct nbd_device *nbd, loff_t bytesize,
300 loff_t blksize)
301 {
302 if (!blksize)
303 blksize = NBD_DEF_BLKSIZE;
304 if (blksize < 512 || blksize > PAGE_SIZE || !is_power_of_2(blksize))
305 return -EINVAL;
306
307 nbd->config->bytesize = bytesize;
308 nbd->config->blksize = blksize;
309
310 if (!nbd->task_recv)
311 return 0;
312
313 if (nbd->config->flags & NBD_FLAG_SEND_TRIM) {
314 nbd->disk->queue->limits.discard_granularity = blksize;
315 nbd->disk->queue->limits.discard_alignment = blksize;
316 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
317 }
318 blk_queue_logical_block_size(nbd->disk->queue, blksize);
319 blk_queue_physical_block_size(nbd->disk->queue, blksize);
320
321 if (max_part)
322 set_bit(GD_NEED_PART_SCAN, &nbd->disk->state);
323 if (!set_capacity_and_notify(nbd->disk, bytesize >> 9))
324 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
325 return 0;
326 }
327
328 static void nbd_complete_rq(struct request *req)
329 {
330 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
331
332 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
333 cmd->status ? "failed" : "done");
334
335 blk_mq_end_request(req, cmd->status);
336 }
337
338 /*
339 * Forcibly shutdown the socket causing all listeners to error
340 */
341 static void sock_shutdown(struct nbd_device *nbd)
342 {
343 struct nbd_config *config = nbd->config;
344 int i;
345
346 if (config->num_connections == 0)
347 return;
348 if (test_and_set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
349 return;
350
351 for (i = 0; i < config->num_connections; i++) {
352 struct nbd_sock *nsock = config->socks[i];
353 mutex_lock(&nsock->tx_lock);
354 nbd_mark_nsock_dead(nbd, nsock, 0);
355 mutex_unlock(&nsock->tx_lock);
356 }
357 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
358 }
359
360 static u32 req_to_nbd_cmd_type(struct request *req)
361 {
362 switch (req_op(req)) {
363 case REQ_OP_DISCARD:
364 return NBD_CMD_TRIM;
365 case REQ_OP_FLUSH:
366 return NBD_CMD_FLUSH;
367 case REQ_OP_WRITE:
368 return NBD_CMD_WRITE;
369 case REQ_OP_READ:
370 return NBD_CMD_READ;
371 default:
372 return U32_MAX;
373 }
374 }
375
376 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
377 bool reserved)
378 {
379 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
380 struct nbd_device *nbd = cmd->nbd;
381 struct nbd_config *config;
382
383 if (!mutex_trylock(&cmd->lock))
384 return BLK_EH_RESET_TIMER;
385
386 if (!refcount_inc_not_zero(&nbd->config_refs)) {
387 cmd->status = BLK_STS_TIMEOUT;
388 mutex_unlock(&cmd->lock);
389 goto done;
390 }
391 config = nbd->config;
392
393 if (config->num_connections > 1 ||
394 (config->num_connections == 1 && nbd->tag_set.timeout)) {
395 dev_err_ratelimited(nbd_to_dev(nbd),
396 "Connection timed out, retrying (%d/%d alive)\n",
397 atomic_read(&config->live_connections),
398 config->num_connections);
399 /*
400 * Hooray we have more connections, requeue this IO, the submit
401 * path will put it on a real connection. Or if only one
402 * connection is configured, the submit path will wait util
403 * a new connection is reconfigured or util dead timeout.
404 */
405 if (config->socks) {
406 if (cmd->index < config->num_connections) {
407 struct nbd_sock *nsock =
408 config->socks[cmd->index];
409 mutex_lock(&nsock->tx_lock);
410 /* We can have multiple outstanding requests, so
411 * we don't want to mark the nsock dead if we've
412 * already reconnected with a new socket, so
413 * only mark it dead if its the same socket we
414 * were sent out on.
415 */
416 if (cmd->cookie == nsock->cookie)
417 nbd_mark_nsock_dead(nbd, nsock, 1);
418 mutex_unlock(&nsock->tx_lock);
419 }
420 mutex_unlock(&cmd->lock);
421 nbd_requeue_cmd(cmd);
422 nbd_config_put(nbd);
423 return BLK_EH_DONE;
424 }
425 }
426
427 if (!nbd->tag_set.timeout) {
428 /*
429 * Userspace sets timeout=0 to disable socket disconnection,
430 * so just warn and reset the timer.
431 */
432 struct nbd_sock *nsock = config->socks[cmd->index];
433 cmd->retries++;
434 dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
435 req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
436 (unsigned long long)blk_rq_pos(req) << 9,
437 blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
438
439 mutex_lock(&nsock->tx_lock);
440 if (cmd->cookie != nsock->cookie) {
441 nbd_requeue_cmd(cmd);
442 mutex_unlock(&nsock->tx_lock);
443 mutex_unlock(&cmd->lock);
444 nbd_config_put(nbd);
445 return BLK_EH_DONE;
446 }
447 mutex_unlock(&nsock->tx_lock);
448 mutex_unlock(&cmd->lock);
449 nbd_config_put(nbd);
450 return BLK_EH_RESET_TIMER;
451 }
452
453 dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
454 set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
455 cmd->status = BLK_STS_IOERR;
456 mutex_unlock(&cmd->lock);
457 sock_shutdown(nbd);
458 nbd_config_put(nbd);
459 done:
460 blk_mq_complete_request(req);
461 return BLK_EH_DONE;
462 }
463
464 /*
465 * Send or receive packet.
466 */
467 static int sock_xmit(struct nbd_device *nbd, int index, int send,
468 struct iov_iter *iter, int msg_flags, int *sent)
469 {
470 struct nbd_config *config = nbd->config;
471 struct socket *sock = config->socks[index]->sock;
472 int result;
473 struct msghdr msg;
474 unsigned int noreclaim_flag;
475
476 if (unlikely(!sock)) {
477 dev_err_ratelimited(disk_to_dev(nbd->disk),
478 "Attempted %s on closed socket in sock_xmit\n",
479 (send ? "send" : "recv"));
480 return -EINVAL;
481 }
482
483 msg.msg_iter = *iter;
484
485 noreclaim_flag = memalloc_noreclaim_save();
486 do {
487 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
488 msg.msg_name = NULL;
489 msg.msg_namelen = 0;
490 msg.msg_control = NULL;
491 msg.msg_controllen = 0;
492 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
493
494 if (send)
495 result = sock_sendmsg(sock, &msg);
496 else
497 result = sock_recvmsg(sock, &msg, msg.msg_flags);
498
499 if (result <= 0) {
500 if (result == 0)
501 result = -EPIPE; /* short read */
502 break;
503 }
504 if (sent)
505 *sent += result;
506 } while (msg_data_left(&msg));
507
508 memalloc_noreclaim_restore(noreclaim_flag);
509
510 return result;
511 }
512
513 /*
514 * Different settings for sk->sk_sndtimeo can result in different return values
515 * if there is a signal pending when we enter sendmsg, because reasons?
516 */
517 static inline int was_interrupted(int result)
518 {
519 return result == -ERESTARTSYS || result == -EINTR;
520 }
521
522 /* always call with the tx_lock held */
523 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
524 {
525 struct request *req = blk_mq_rq_from_pdu(cmd);
526 struct nbd_config *config = nbd->config;
527 struct nbd_sock *nsock = config->socks[index];
528 int result;
529 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
530 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
531 struct iov_iter from;
532 unsigned long size = blk_rq_bytes(req);
533 struct bio *bio;
534 u64 handle;
535 u32 type;
536 u32 nbd_cmd_flags = 0;
537 int sent = nsock->sent, skip = 0;
538
539 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
540
541 type = req_to_nbd_cmd_type(req);
542 if (type == U32_MAX)
543 return -EIO;
544
545 if (rq_data_dir(req) == WRITE &&
546 (config->flags & NBD_FLAG_READ_ONLY)) {
547 dev_err_ratelimited(disk_to_dev(nbd->disk),
548 "Write on read-only\n");
549 return -EIO;
550 }
551
552 if (req->cmd_flags & REQ_FUA)
553 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
554
555 /* We did a partial send previously, and we at least sent the whole
556 * request struct, so just go and send the rest of the pages in the
557 * request.
558 */
559 if (sent) {
560 if (sent >= sizeof(request)) {
561 skip = sent - sizeof(request);
562
563 /* initialize handle for tracing purposes */
564 handle = nbd_cmd_handle(cmd);
565
566 goto send_pages;
567 }
568 iov_iter_advance(&from, sent);
569 } else {
570 cmd->cmd_cookie++;
571 }
572 cmd->index = index;
573 cmd->cookie = nsock->cookie;
574 cmd->retries = 0;
575 request.type = htonl(type | nbd_cmd_flags);
576 if (type != NBD_CMD_FLUSH) {
577 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
578 request.len = htonl(size);
579 }
580 handle = nbd_cmd_handle(cmd);
581 memcpy(request.handle, &handle, sizeof(handle));
582
583 trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
584
585 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
586 req, nbdcmd_to_ascii(type),
587 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
588 result = sock_xmit(nbd, index, 1, &from,
589 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
590 trace_nbd_header_sent(req, handle);
591 if (result <= 0) {
592 if (was_interrupted(result)) {
593 /* If we havne't sent anything we can just return BUSY,
594 * however if we have sent something we need to make
595 * sure we only allow this req to be sent until we are
596 * completely done.
597 */
598 if (sent) {
599 nsock->pending = req;
600 nsock->sent = sent;
601 }
602 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
603 return BLK_STS_RESOURCE;
604 }
605 dev_err_ratelimited(disk_to_dev(nbd->disk),
606 "Send control failed (result %d)\n", result);
607 return -EAGAIN;
608 }
609 send_pages:
610 if (type != NBD_CMD_WRITE)
611 goto out;
612
613 bio = req->bio;
614 while (bio) {
615 struct bio *next = bio->bi_next;
616 struct bvec_iter iter;
617 struct bio_vec bvec;
618
619 bio_for_each_segment(bvec, bio, iter) {
620 bool is_last = !next && bio_iter_last(bvec, iter);
621 int flags = is_last ? 0 : MSG_MORE;
622
623 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
624 req, bvec.bv_len);
625 iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
626 if (skip) {
627 if (skip >= iov_iter_count(&from)) {
628 skip -= iov_iter_count(&from);
629 continue;
630 }
631 iov_iter_advance(&from, skip);
632 skip = 0;
633 }
634 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
635 if (result <= 0) {
636 if (was_interrupted(result)) {
637 /* We've already sent the header, we
638 * have no choice but to set pending and
639 * return BUSY.
640 */
641 nsock->pending = req;
642 nsock->sent = sent;
643 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
644 return BLK_STS_RESOURCE;
645 }
646 dev_err(disk_to_dev(nbd->disk),
647 "Send data failed (result %d)\n",
648 result);
649 return -EAGAIN;
650 }
651 /*
652 * The completion might already have come in,
653 * so break for the last one instead of letting
654 * the iterator do it. This prevents use-after-free
655 * of the bio.
656 */
657 if (is_last)
658 break;
659 }
660 bio = next;
661 }
662 out:
663 trace_nbd_payload_sent(req, handle);
664 nsock->pending = NULL;
665 nsock->sent = 0;
666 return 0;
667 }
668
669 /* NULL returned = something went wrong, inform userspace */
670 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
671 {
672 struct nbd_config *config = nbd->config;
673 int result;
674 struct nbd_reply reply;
675 struct nbd_cmd *cmd;
676 struct request *req = NULL;
677 u64 handle;
678 u16 hwq;
679 u32 tag;
680 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
681 struct iov_iter to;
682 int ret = 0;
683
684 reply.magic = 0;
685 iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
686 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
687 if (result <= 0) {
688 if (!nbd_disconnected(config))
689 dev_err(disk_to_dev(nbd->disk),
690 "Receive control failed (result %d)\n", result);
691 return ERR_PTR(result);
692 }
693
694 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
695 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
696 (unsigned long)ntohl(reply.magic));
697 return ERR_PTR(-EPROTO);
698 }
699
700 memcpy(&handle, reply.handle, sizeof(handle));
701 tag = nbd_handle_to_tag(handle);
702 hwq = blk_mq_unique_tag_to_hwq(tag);
703 if (hwq < nbd->tag_set.nr_hw_queues)
704 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
705 blk_mq_unique_tag_to_tag(tag));
706 if (!req || !blk_mq_request_started(req)) {
707 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
708 tag, req);
709 return ERR_PTR(-ENOENT);
710 }
711 trace_nbd_header_received(req, handle);
712 cmd = blk_mq_rq_to_pdu(req);
713
714 mutex_lock(&cmd->lock);
715 if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
716 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
717 req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
718 ret = -ENOENT;
719 goto out;
720 }
721 if (cmd->status != BLK_STS_OK) {
722 dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
723 req);
724 ret = -ENOENT;
725 goto out;
726 }
727 if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
728 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
729 req);
730 ret = -ENOENT;
731 goto out;
732 }
733 if (ntohl(reply.error)) {
734 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
735 ntohl(reply.error));
736 cmd->status = BLK_STS_IOERR;
737 goto out;
738 }
739
740 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
741 if (rq_data_dir(req) != WRITE) {
742 struct req_iterator iter;
743 struct bio_vec bvec;
744
745 rq_for_each_segment(bvec, req, iter) {
746 iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
747 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
748 if (result <= 0) {
749 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
750 result);
751 /*
752 * If we've disconnected, we need to make sure we
753 * complete this request, otherwise error out
754 * and let the timeout stuff handle resubmitting
755 * this request onto another connection.
756 */
757 if (nbd_disconnected(config)) {
758 cmd->status = BLK_STS_IOERR;
759 goto out;
760 }
761 ret = -EIO;
762 goto out;
763 }
764 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
765 req, bvec.bv_len);
766 }
767 }
768 out:
769 trace_nbd_payload_received(req, handle);
770 mutex_unlock(&cmd->lock);
771 return ret ? ERR_PTR(ret) : cmd;
772 }
773
774 static void recv_work(struct work_struct *work)
775 {
776 struct recv_thread_args *args = container_of(work,
777 struct recv_thread_args,
778 work);
779 struct nbd_device *nbd = args->nbd;
780 struct nbd_config *config = nbd->config;
781 struct nbd_cmd *cmd;
782 struct request *rq;
783
784 while (1) {
785 cmd = nbd_read_stat(nbd, args->index);
786 if (IS_ERR(cmd)) {
787 struct nbd_sock *nsock = config->socks[args->index];
788
789 mutex_lock(&nsock->tx_lock);
790 nbd_mark_nsock_dead(nbd, nsock, 1);
791 mutex_unlock(&nsock->tx_lock);
792 break;
793 }
794
795 rq = blk_mq_rq_from_pdu(cmd);
796 if (likely(!blk_should_fake_timeout(rq->q)))
797 blk_mq_complete_request(rq);
798 }
799 nbd_config_put(nbd);
800 atomic_dec(&config->recv_threads);
801 wake_up(&config->recv_wq);
802 kfree(args);
803 }
804
805 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
806 {
807 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
808
809 mutex_lock(&cmd->lock);
810 cmd->status = BLK_STS_IOERR;
811 mutex_unlock(&cmd->lock);
812
813 blk_mq_complete_request(req);
814 return true;
815 }
816
817 static void nbd_clear_que(struct nbd_device *nbd)
818 {
819 blk_mq_quiesce_queue(nbd->disk->queue);
820 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
821 blk_mq_unquiesce_queue(nbd->disk->queue);
822 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
823 }
824
825 static int find_fallback(struct nbd_device *nbd, int index)
826 {
827 struct nbd_config *config = nbd->config;
828 int new_index = -1;
829 struct nbd_sock *nsock = config->socks[index];
830 int fallback = nsock->fallback_index;
831
832 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
833 return new_index;
834
835 if (config->num_connections <= 1) {
836 dev_err_ratelimited(disk_to_dev(nbd->disk),
837 "Dead connection, failed to find a fallback\n");
838 return new_index;
839 }
840
841 if (fallback >= 0 && fallback < config->num_connections &&
842 !config->socks[fallback]->dead)
843 return fallback;
844
845 if (nsock->fallback_index < 0 ||
846 nsock->fallback_index >= config->num_connections ||
847 config->socks[nsock->fallback_index]->dead) {
848 int i;
849 for (i = 0; i < config->num_connections; i++) {
850 if (i == index)
851 continue;
852 if (!config->socks[i]->dead) {
853 new_index = i;
854 break;
855 }
856 }
857 nsock->fallback_index = new_index;
858 if (new_index < 0) {
859 dev_err_ratelimited(disk_to_dev(nbd->disk),
860 "Dead connection, failed to find a fallback\n");
861 return new_index;
862 }
863 }
864 new_index = nsock->fallback_index;
865 return new_index;
866 }
867
868 static int wait_for_reconnect(struct nbd_device *nbd)
869 {
870 struct nbd_config *config = nbd->config;
871 if (!config->dead_conn_timeout)
872 return 0;
873 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
874 return 0;
875 return wait_event_timeout(config->conn_wait,
876 atomic_read(&config->live_connections) > 0,
877 config->dead_conn_timeout) > 0;
878 }
879
880 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
881 {
882 struct request *req = blk_mq_rq_from_pdu(cmd);
883 struct nbd_device *nbd = cmd->nbd;
884 struct nbd_config *config;
885 struct nbd_sock *nsock;
886 int ret;
887
888 if (!refcount_inc_not_zero(&nbd->config_refs)) {
889 dev_err_ratelimited(disk_to_dev(nbd->disk),
890 "Socks array is empty\n");
891 blk_mq_start_request(req);
892 return -EINVAL;
893 }
894 config = nbd->config;
895
896 if (index >= config->num_connections) {
897 dev_err_ratelimited(disk_to_dev(nbd->disk),
898 "Attempted send on invalid socket\n");
899 nbd_config_put(nbd);
900 blk_mq_start_request(req);
901 return -EINVAL;
902 }
903 cmd->status = BLK_STS_OK;
904 again:
905 nsock = config->socks[index];
906 mutex_lock(&nsock->tx_lock);
907 if (nsock->dead) {
908 int old_index = index;
909 index = find_fallback(nbd, index);
910 mutex_unlock(&nsock->tx_lock);
911 if (index < 0) {
912 if (wait_for_reconnect(nbd)) {
913 index = old_index;
914 goto again;
915 }
916 /* All the sockets should already be down at this point,
917 * we just want to make sure that DISCONNECTED is set so
918 * any requests that come in that were queue'ed waiting
919 * for the reconnect timer don't trigger the timer again
920 * and instead just error out.
921 */
922 sock_shutdown(nbd);
923 nbd_config_put(nbd);
924 blk_mq_start_request(req);
925 return -EIO;
926 }
927 goto again;
928 }
929
930 /* Handle the case that we have a pending request that was partially
931 * transmitted that _has_ to be serviced first. We need to call requeue
932 * here so that it gets put _after_ the request that is already on the
933 * dispatch list.
934 */
935 blk_mq_start_request(req);
936 if (unlikely(nsock->pending && nsock->pending != req)) {
937 nbd_requeue_cmd(cmd);
938 ret = 0;
939 goto out;
940 }
941 /*
942 * Some failures are related to the link going down, so anything that
943 * returns EAGAIN can be retried on a different socket.
944 */
945 ret = nbd_send_cmd(nbd, cmd, index);
946 if (ret == -EAGAIN) {
947 dev_err_ratelimited(disk_to_dev(nbd->disk),
948 "Request send failed, requeueing\n");
949 nbd_mark_nsock_dead(nbd, nsock, 1);
950 nbd_requeue_cmd(cmd);
951 ret = 0;
952 }
953 out:
954 mutex_unlock(&nsock->tx_lock);
955 nbd_config_put(nbd);
956 return ret;
957 }
958
959 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
960 const struct blk_mq_queue_data *bd)
961 {
962 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
963 int ret;
964
965 /*
966 * Since we look at the bio's to send the request over the network we
967 * need to make sure the completion work doesn't mark this request done
968 * before we are done doing our send. This keeps us from dereferencing
969 * freed data if we have particularly fast completions (ie we get the
970 * completion before we exit sock_xmit on the last bvec) or in the case
971 * that the server is misbehaving (or there was an error) before we're
972 * done sending everything over the wire.
973 */
974 mutex_lock(&cmd->lock);
975 clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
976
977 /* We can be called directly from the user space process, which means we
978 * could possibly have signals pending so our sendmsg will fail. In
979 * this case we need to return that we are busy, otherwise error out as
980 * appropriate.
981 */
982 ret = nbd_handle_cmd(cmd, hctx->queue_num);
983 if (ret < 0)
984 ret = BLK_STS_IOERR;
985 else if (!ret)
986 ret = BLK_STS_OK;
987 mutex_unlock(&cmd->lock);
988
989 return ret;
990 }
991
992 static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
993 int *err)
994 {
995 struct socket *sock;
996
997 *err = 0;
998 sock = sockfd_lookup(fd, err);
999 if (!sock)
1000 return NULL;
1001
1002 if (sock->ops->shutdown == sock_no_shutdown) {
1003 dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
1004 *err = -EINVAL;
1005 sockfd_put(sock);
1006 return NULL;
1007 }
1008
1009 return sock;
1010 }
1011
1012 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1013 bool netlink)
1014 {
1015 struct nbd_config *config = nbd->config;
1016 struct socket *sock;
1017 struct nbd_sock **socks;
1018 struct nbd_sock *nsock;
1019 int err;
1020
1021 sock = nbd_get_socket(nbd, arg, &err);
1022 if (!sock)
1023 return err;
1024
1025 if (!netlink && !nbd->task_setup &&
1026 !test_bit(NBD_RT_BOUND, &config->runtime_flags))
1027 nbd->task_setup = current;
1028
1029 if (!netlink &&
1030 (nbd->task_setup != current ||
1031 test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1032 dev_err(disk_to_dev(nbd->disk),
1033 "Device being setup by another task");
1034 err = -EBUSY;
1035 goto put_socket;
1036 }
1037
1038 nsock = kzalloc(sizeof(*nsock), GFP_KERNEL);
1039 if (!nsock) {
1040 err = -ENOMEM;
1041 goto put_socket;
1042 }
1043
1044 socks = krealloc(config->socks, (config->num_connections + 1) *
1045 sizeof(struct nbd_sock *), GFP_KERNEL);
1046 if (!socks) {
1047 kfree(nsock);
1048 err = -ENOMEM;
1049 goto put_socket;
1050 }
1051
1052 config->socks = socks;
1053
1054 nsock->fallback_index = -1;
1055 nsock->dead = false;
1056 mutex_init(&nsock->tx_lock);
1057 nsock->sock = sock;
1058 nsock->pending = NULL;
1059 nsock->sent = 0;
1060 nsock->cookie = 0;
1061 socks[config->num_connections++] = nsock;
1062 atomic_inc(&config->live_connections);
1063
1064 return 0;
1065
1066 put_socket:
1067 sockfd_put(sock);
1068 return err;
1069 }
1070
1071 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1072 {
1073 struct nbd_config *config = nbd->config;
1074 struct socket *sock, *old;
1075 struct recv_thread_args *args;
1076 int i;
1077 int err;
1078
1079 sock = nbd_get_socket(nbd, arg, &err);
1080 if (!sock)
1081 return err;
1082
1083 args = kzalloc(sizeof(*args), GFP_KERNEL);
1084 if (!args) {
1085 sockfd_put(sock);
1086 return -ENOMEM;
1087 }
1088
1089 for (i = 0; i < config->num_connections; i++) {
1090 struct nbd_sock *nsock = config->socks[i];
1091
1092 if (!nsock->dead)
1093 continue;
1094
1095 mutex_lock(&nsock->tx_lock);
1096 if (!nsock->dead) {
1097 mutex_unlock(&nsock->tx_lock);
1098 continue;
1099 }
1100 sk_set_memalloc(sock->sk);
1101 if (nbd->tag_set.timeout)
1102 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1103 atomic_inc(&config->recv_threads);
1104 refcount_inc(&nbd->config_refs);
1105 old = nsock->sock;
1106 nsock->fallback_index = -1;
1107 nsock->sock = sock;
1108 nsock->dead = false;
1109 INIT_WORK(&args->work, recv_work);
1110 args->index = i;
1111 args->nbd = nbd;
1112 nsock->cookie++;
1113 mutex_unlock(&nsock->tx_lock);
1114 sockfd_put(old);
1115
1116 clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1117
1118 /* We take the tx_mutex in an error path in the recv_work, so we
1119 * need to queue_work outside of the tx_mutex.
1120 */
1121 queue_work(nbd->recv_workq, &args->work);
1122
1123 atomic_inc(&config->live_connections);
1124 wake_up(&config->conn_wait);
1125 return 0;
1126 }
1127 sockfd_put(sock);
1128 kfree(args);
1129 return -ENOSPC;
1130 }
1131
1132 static void nbd_bdev_reset(struct block_device *bdev)
1133 {
1134 if (bdev->bd_openers > 1)
1135 return;
1136 set_capacity(bdev->bd_disk, 0);
1137 }
1138
1139 static void nbd_parse_flags(struct nbd_device *nbd)
1140 {
1141 struct nbd_config *config = nbd->config;
1142 if (config->flags & NBD_FLAG_READ_ONLY)
1143 set_disk_ro(nbd->disk, true);
1144 else
1145 set_disk_ro(nbd->disk, false);
1146 if (config->flags & NBD_FLAG_SEND_TRIM)
1147 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1148 if (config->flags & NBD_FLAG_SEND_FLUSH) {
1149 if (config->flags & NBD_FLAG_SEND_FUA)
1150 blk_queue_write_cache(nbd->disk->queue, true, true);
1151 else
1152 blk_queue_write_cache(nbd->disk->queue, true, false);
1153 }
1154 else
1155 blk_queue_write_cache(nbd->disk->queue, false, false);
1156 }
1157
1158 static void send_disconnects(struct nbd_device *nbd)
1159 {
1160 struct nbd_config *config = nbd->config;
1161 struct nbd_request request = {
1162 .magic = htonl(NBD_REQUEST_MAGIC),
1163 .type = htonl(NBD_CMD_DISC),
1164 };
1165 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1166 struct iov_iter from;
1167 int i, ret;
1168
1169 for (i = 0; i < config->num_connections; i++) {
1170 struct nbd_sock *nsock = config->socks[i];
1171
1172 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1173 mutex_lock(&nsock->tx_lock);
1174 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1175 if (ret <= 0)
1176 dev_err(disk_to_dev(nbd->disk),
1177 "Send disconnect failed %d\n", ret);
1178 mutex_unlock(&nsock->tx_lock);
1179 }
1180 }
1181
1182 static int nbd_disconnect(struct nbd_device *nbd)
1183 {
1184 struct nbd_config *config = nbd->config;
1185
1186 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1187 set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1188 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1189 send_disconnects(nbd);
1190 return 0;
1191 }
1192
1193 static void nbd_clear_sock(struct nbd_device *nbd)
1194 {
1195 sock_shutdown(nbd);
1196 nbd_clear_que(nbd);
1197 nbd->task_setup = NULL;
1198 }
1199
1200 static void nbd_config_put(struct nbd_device *nbd)
1201 {
1202 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1203 &nbd->config_lock)) {
1204 struct nbd_config *config = nbd->config;
1205 nbd_dev_dbg_close(nbd);
1206 nbd_size_clear(nbd);
1207 if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1208 &config->runtime_flags))
1209 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1210 nbd->task_recv = NULL;
1211 nbd_clear_sock(nbd);
1212 if (config->num_connections) {
1213 int i;
1214 for (i = 0; i < config->num_connections; i++) {
1215 sockfd_put(config->socks[i]->sock);
1216 kfree(config->socks[i]);
1217 }
1218 kfree(config->socks);
1219 }
1220 kfree(nbd->config);
1221 nbd->config = NULL;
1222
1223 if (nbd->recv_workq)
1224 destroy_workqueue(nbd->recv_workq);
1225 nbd->recv_workq = NULL;
1226
1227 nbd->tag_set.timeout = 0;
1228 nbd->disk->queue->limits.discard_granularity = 0;
1229 nbd->disk->queue->limits.discard_alignment = 0;
1230 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1231 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1232
1233 mutex_unlock(&nbd->config_lock);
1234 nbd_put(nbd);
1235 module_put(THIS_MODULE);
1236 }
1237 }
1238
1239 static int nbd_start_device(struct nbd_device *nbd)
1240 {
1241 struct nbd_config *config = nbd->config;
1242 int num_connections = config->num_connections;
1243 int error = 0, i;
1244
1245 if (nbd->task_recv)
1246 return -EBUSY;
1247 if (!config->socks)
1248 return -EINVAL;
1249 if (num_connections > 1 &&
1250 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1251 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1252 return -EINVAL;
1253 }
1254
1255 nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1256 WQ_MEM_RECLAIM | WQ_HIGHPRI |
1257 WQ_UNBOUND, 0, nbd->index);
1258 if (!nbd->recv_workq) {
1259 dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1260 return -ENOMEM;
1261 }
1262
1263 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1264 nbd->task_recv = current;
1265
1266 nbd_parse_flags(nbd);
1267
1268 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1269 if (error) {
1270 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1271 return error;
1272 }
1273 set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1274
1275 nbd_dev_dbg_init(nbd);
1276 for (i = 0; i < num_connections; i++) {
1277 struct recv_thread_args *args;
1278
1279 args = kzalloc(sizeof(*args), GFP_KERNEL);
1280 if (!args) {
1281 sock_shutdown(nbd);
1282 /*
1283 * If num_connections is m (2 < m),
1284 * and NO.1 ~ NO.n(1 < n < m) kzallocs are successful.
1285 * But NO.(n + 1) failed. We still have n recv threads.
1286 * So, add flush_workqueue here to prevent recv threads
1287 * dropping the last config_refs and trying to destroy
1288 * the workqueue from inside the workqueue.
1289 */
1290 if (i)
1291 flush_workqueue(nbd->recv_workq);
1292 return -ENOMEM;
1293 }
1294 sk_set_memalloc(config->socks[i]->sock->sk);
1295 if (nbd->tag_set.timeout)
1296 config->socks[i]->sock->sk->sk_sndtimeo =
1297 nbd->tag_set.timeout;
1298 atomic_inc(&config->recv_threads);
1299 refcount_inc(&nbd->config_refs);
1300 INIT_WORK(&args->work, recv_work);
1301 args->nbd = nbd;
1302 args->index = i;
1303 queue_work(nbd->recv_workq, &args->work);
1304 }
1305 return nbd_set_size(nbd, config->bytesize, config->blksize);
1306 }
1307
1308 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1309 {
1310 struct nbd_config *config = nbd->config;
1311 int ret;
1312
1313 ret = nbd_start_device(nbd);
1314 if (ret)
1315 return ret;
1316
1317 if (max_part)
1318 set_bit(GD_NEED_PART_SCAN, &nbd->disk->state);
1319 mutex_unlock(&nbd->config_lock);
1320 ret = wait_event_interruptible(config->recv_wq,
1321 atomic_read(&config->recv_threads) == 0);
1322 if (ret)
1323 sock_shutdown(nbd);
1324 flush_workqueue(nbd->recv_workq);
1325
1326 mutex_lock(&nbd->config_lock);
1327 nbd_bdev_reset(bdev);
1328 /* user requested, ignore socket errors */
1329 if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1330 ret = 0;
1331 if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1332 ret = -ETIMEDOUT;
1333 return ret;
1334 }
1335
1336 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1337 struct block_device *bdev)
1338 {
1339 sock_shutdown(nbd);
1340 __invalidate_device(bdev, true);
1341 nbd_bdev_reset(bdev);
1342 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1343 &nbd->config->runtime_flags))
1344 nbd_config_put(nbd);
1345 }
1346
1347 static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1348 {
1349 nbd->tag_set.timeout = timeout * HZ;
1350 if (timeout)
1351 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1352 else
1353 blk_queue_rq_timeout(nbd->disk->queue, 30 * HZ);
1354 }
1355
1356 /* Must be called with config_lock held */
1357 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1358 unsigned int cmd, unsigned long arg)
1359 {
1360 struct nbd_config *config = nbd->config;
1361
1362 switch (cmd) {
1363 case NBD_DISCONNECT:
1364 return nbd_disconnect(nbd);
1365 case NBD_CLEAR_SOCK:
1366 nbd_clear_sock_ioctl(nbd, bdev);
1367 return 0;
1368 case NBD_SET_SOCK:
1369 return nbd_add_socket(nbd, arg, false);
1370 case NBD_SET_BLKSIZE:
1371 return nbd_set_size(nbd, config->bytesize, arg);
1372 case NBD_SET_SIZE:
1373 return nbd_set_size(nbd, arg, config->blksize);
1374 case NBD_SET_SIZE_BLOCKS:
1375 return nbd_set_size(nbd, arg * config->blksize,
1376 config->blksize);
1377 case NBD_SET_TIMEOUT:
1378 nbd_set_cmd_timeout(nbd, arg);
1379 return 0;
1380
1381 case NBD_SET_FLAGS:
1382 config->flags = arg;
1383 return 0;
1384 case NBD_DO_IT:
1385 return nbd_start_device_ioctl(nbd, bdev);
1386 case NBD_CLEAR_QUE:
1387 /*
1388 * This is for compatibility only. The queue is always cleared
1389 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1390 */
1391 return 0;
1392 case NBD_PRINT_DEBUG:
1393 /*
1394 * For compatibility only, we no longer keep a list of
1395 * outstanding requests.
1396 */
1397 return 0;
1398 }
1399 return -ENOTTY;
1400 }
1401
1402 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1403 unsigned int cmd, unsigned long arg)
1404 {
1405 struct nbd_device *nbd = bdev->bd_disk->private_data;
1406 struct nbd_config *config = nbd->config;
1407 int error = -EINVAL;
1408
1409 if (!capable(CAP_SYS_ADMIN))
1410 return -EPERM;
1411
1412 /* The block layer will pass back some non-nbd ioctls in case we have
1413 * special handling for them, but we don't so just return an error.
1414 */
1415 if (_IOC_TYPE(cmd) != 0xab)
1416 return -EINVAL;
1417
1418 mutex_lock(&nbd->config_lock);
1419
1420 /* Don't allow ioctl operations on a nbd device that was created with
1421 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1422 */
1423 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1424 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1425 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1426 else
1427 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1428 mutex_unlock(&nbd->config_lock);
1429 return error;
1430 }
1431
1432 static struct nbd_config *nbd_alloc_config(void)
1433 {
1434 struct nbd_config *config;
1435
1436 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1437 if (!config)
1438 return NULL;
1439 atomic_set(&config->recv_threads, 0);
1440 init_waitqueue_head(&config->recv_wq);
1441 init_waitqueue_head(&config->conn_wait);
1442 config->blksize = NBD_DEF_BLKSIZE;
1443 atomic_set(&config->live_connections, 0);
1444 try_module_get(THIS_MODULE);
1445 return config;
1446 }
1447
1448 static int nbd_open(struct block_device *bdev, fmode_t mode)
1449 {
1450 struct nbd_device *nbd;
1451 int ret = 0;
1452
1453 mutex_lock(&nbd_index_mutex);
1454 nbd = bdev->bd_disk->private_data;
1455 if (!nbd) {
1456 ret = -ENXIO;
1457 goto out;
1458 }
1459 if (!refcount_inc_not_zero(&nbd->refs)) {
1460 ret = -ENXIO;
1461 goto out;
1462 }
1463 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1464 struct nbd_config *config;
1465
1466 mutex_lock(&nbd->config_lock);
1467 if (refcount_inc_not_zero(&nbd->config_refs)) {
1468 mutex_unlock(&nbd->config_lock);
1469 goto out;
1470 }
1471 config = nbd->config = nbd_alloc_config();
1472 if (!config) {
1473 ret = -ENOMEM;
1474 mutex_unlock(&nbd->config_lock);
1475 goto out;
1476 }
1477 refcount_set(&nbd->config_refs, 1);
1478 refcount_inc(&nbd->refs);
1479 mutex_unlock(&nbd->config_lock);
1480 if (max_part)
1481 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1482 } else if (nbd_disconnected(nbd->config)) {
1483 if (max_part)
1484 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1485 }
1486 out:
1487 mutex_unlock(&nbd_index_mutex);
1488 return ret;
1489 }
1490
1491 static void nbd_release(struct gendisk *disk, fmode_t mode)
1492 {
1493 struct nbd_device *nbd = disk->private_data;
1494
1495 if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1496 disk->part0->bd_openers == 0)
1497 nbd_disconnect_and_put(nbd);
1498
1499 nbd_config_put(nbd);
1500 nbd_put(nbd);
1501 }
1502
1503 static const struct block_device_operations nbd_fops =
1504 {
1505 .owner = THIS_MODULE,
1506 .open = nbd_open,
1507 .release = nbd_release,
1508 .ioctl = nbd_ioctl,
1509 .compat_ioctl = nbd_ioctl,
1510 };
1511
1512 #if IS_ENABLED(CONFIG_DEBUG_FS)
1513
1514 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1515 {
1516 struct nbd_device *nbd = s->private;
1517
1518 if (nbd->task_recv)
1519 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1520
1521 return 0;
1522 }
1523
1524 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1525 {
1526 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1527 }
1528
1529 static const struct file_operations nbd_dbg_tasks_ops = {
1530 .open = nbd_dbg_tasks_open,
1531 .read = seq_read,
1532 .llseek = seq_lseek,
1533 .release = single_release,
1534 };
1535
1536 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1537 {
1538 struct nbd_device *nbd = s->private;
1539 u32 flags = nbd->config->flags;
1540
1541 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1542
1543 seq_puts(s, "Known flags:\n");
1544
1545 if (flags & NBD_FLAG_HAS_FLAGS)
1546 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1547 if (flags & NBD_FLAG_READ_ONLY)
1548 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1549 if (flags & NBD_FLAG_SEND_FLUSH)
1550 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1551 if (flags & NBD_FLAG_SEND_FUA)
1552 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1553 if (flags & NBD_FLAG_SEND_TRIM)
1554 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1555
1556 return 0;
1557 }
1558
1559 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1560 {
1561 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1562 }
1563
1564 static const struct file_operations nbd_dbg_flags_ops = {
1565 .open = nbd_dbg_flags_open,
1566 .read = seq_read,
1567 .llseek = seq_lseek,
1568 .release = single_release,
1569 };
1570
1571 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1572 {
1573 struct dentry *dir;
1574 struct nbd_config *config = nbd->config;
1575
1576 if (!nbd_dbg_dir)
1577 return -EIO;
1578
1579 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1580 if (!dir) {
1581 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1582 nbd_name(nbd));
1583 return -EIO;
1584 }
1585 config->dbg_dir = dir;
1586
1587 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1588 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1589 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1590 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1591 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1592
1593 return 0;
1594 }
1595
1596 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1597 {
1598 debugfs_remove_recursive(nbd->config->dbg_dir);
1599 }
1600
1601 static int nbd_dbg_init(void)
1602 {
1603 struct dentry *dbg_dir;
1604
1605 dbg_dir = debugfs_create_dir("nbd", NULL);
1606 if (!dbg_dir)
1607 return -EIO;
1608
1609 nbd_dbg_dir = dbg_dir;
1610
1611 return 0;
1612 }
1613
1614 static void nbd_dbg_close(void)
1615 {
1616 debugfs_remove_recursive(nbd_dbg_dir);
1617 }
1618
1619 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1620
1621 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1622 {
1623 return 0;
1624 }
1625
1626 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1627 {
1628 }
1629
1630 static int nbd_dbg_init(void)
1631 {
1632 return 0;
1633 }
1634
1635 static void nbd_dbg_close(void)
1636 {
1637 }
1638
1639 #endif
1640
1641 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1642 unsigned int hctx_idx, unsigned int numa_node)
1643 {
1644 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1645 cmd->nbd = set->driver_data;
1646 cmd->flags = 0;
1647 mutex_init(&cmd->lock);
1648 return 0;
1649 }
1650
1651 static const struct blk_mq_ops nbd_mq_ops = {
1652 .queue_rq = nbd_queue_rq,
1653 .complete = nbd_complete_rq,
1654 .init_request = nbd_init_request,
1655 .timeout = nbd_xmit_timeout,
1656 };
1657
1658 static int nbd_dev_add(int index)
1659 {
1660 struct nbd_device *nbd;
1661 struct gendisk *disk;
1662 struct request_queue *q;
1663 int err = -ENOMEM;
1664
1665 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1666 if (!nbd)
1667 goto out;
1668
1669 disk = alloc_disk(1 << part_shift);
1670 if (!disk)
1671 goto out_free_nbd;
1672
1673 if (index >= 0) {
1674 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1675 GFP_KERNEL);
1676 if (err == -ENOSPC)
1677 err = -EEXIST;
1678 } else {
1679 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1680 if (err >= 0)
1681 index = err;
1682 }
1683 if (err < 0)
1684 goto out_free_disk;
1685
1686 nbd->index = index;
1687 nbd->disk = disk;
1688 nbd->tag_set.ops = &nbd_mq_ops;
1689 nbd->tag_set.nr_hw_queues = 1;
1690 nbd->tag_set.queue_depth = 128;
1691 nbd->tag_set.numa_node = NUMA_NO_NODE;
1692 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1693 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1694 BLK_MQ_F_BLOCKING;
1695 nbd->tag_set.driver_data = nbd;
1696 nbd->destroy_complete = NULL;
1697
1698 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1699 if (err)
1700 goto out_free_idr;
1701
1702 q = blk_mq_init_queue(&nbd->tag_set);
1703 if (IS_ERR(q)) {
1704 err = PTR_ERR(q);
1705 goto out_free_tags;
1706 }
1707 disk->queue = q;
1708
1709 /*
1710 * Tell the block layer that we are not a rotational device
1711 */
1712 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1713 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1714 disk->queue->limits.discard_granularity = 0;
1715 disk->queue->limits.discard_alignment = 0;
1716 blk_queue_max_discard_sectors(disk->queue, 0);
1717 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1718 blk_queue_max_segments(disk->queue, USHRT_MAX);
1719 blk_queue_max_hw_sectors(disk->queue, 65536);
1720 disk->queue->limits.max_sectors = 256;
1721
1722 mutex_init(&nbd->config_lock);
1723 refcount_set(&nbd->config_refs, 0);
1724 refcount_set(&nbd->refs, 1);
1725 INIT_LIST_HEAD(&nbd->list);
1726 disk->major = NBD_MAJOR;
1727 disk->first_minor = index << part_shift;
1728 disk->fops = &nbd_fops;
1729 disk->private_data = nbd;
1730 sprintf(disk->disk_name, "nbd%d", index);
1731 add_disk(disk);
1732 nbd_total_devices++;
1733 return index;
1734
1735 out_free_tags:
1736 blk_mq_free_tag_set(&nbd->tag_set);
1737 out_free_idr:
1738 idr_remove(&nbd_index_idr, index);
1739 out_free_disk:
1740 put_disk(disk);
1741 out_free_nbd:
1742 kfree(nbd);
1743 out:
1744 return err;
1745 }
1746
1747 static int find_free_cb(int id, void *ptr, void *data)
1748 {
1749 struct nbd_device *nbd = ptr;
1750 struct nbd_device **found = data;
1751
1752 if (!refcount_read(&nbd->config_refs)) {
1753 *found = nbd;
1754 return 1;
1755 }
1756 return 0;
1757 }
1758
1759 /* Netlink interface. */
1760 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1761 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1762 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1763 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1764 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1765 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1766 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1767 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1768 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1769 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1770 };
1771
1772 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1773 [NBD_SOCK_FD] = { .type = NLA_U32 },
1774 };
1775
1776 /* We don't use this right now since we don't parse the incoming list, but we
1777 * still want it here so userspace knows what to expect.
1778 */
1779 static const struct nla_policy __attribute__((unused))
1780 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1781 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1782 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1783 };
1784
1785 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1786 {
1787 struct nbd_config *config = nbd->config;
1788 u64 bsize = config->blksize;
1789 u64 bytes = config->bytesize;
1790
1791 if (info->attrs[NBD_ATTR_SIZE_BYTES])
1792 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1793
1794 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES])
1795 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1796
1797 if (bytes != config->bytesize || bsize != config->blksize)
1798 return nbd_set_size(nbd, bytes, bsize);
1799 return 0;
1800 }
1801
1802 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1803 {
1804 DECLARE_COMPLETION_ONSTACK(destroy_complete);
1805 struct nbd_device *nbd = NULL;
1806 struct nbd_config *config;
1807 int index = -1;
1808 int ret;
1809 bool put_dev = false;
1810
1811 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1812 return -EPERM;
1813
1814 if (info->attrs[NBD_ATTR_INDEX])
1815 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1816 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1817 printk(KERN_ERR "nbd: must specify at least one socket\n");
1818 return -EINVAL;
1819 }
1820 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1821 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1822 return -EINVAL;
1823 }
1824 again:
1825 mutex_lock(&nbd_index_mutex);
1826 if (index == -1) {
1827 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1828 if (ret == 0) {
1829 int new_index;
1830 new_index = nbd_dev_add(-1);
1831 if (new_index < 0) {
1832 mutex_unlock(&nbd_index_mutex);
1833 printk(KERN_ERR "nbd: failed to add new device\n");
1834 return new_index;
1835 }
1836 nbd = idr_find(&nbd_index_idr, new_index);
1837 }
1838 } else {
1839 nbd = idr_find(&nbd_index_idr, index);
1840 if (!nbd) {
1841 ret = nbd_dev_add(index);
1842 if (ret < 0) {
1843 mutex_unlock(&nbd_index_mutex);
1844 printk(KERN_ERR "nbd: failed to add new device\n");
1845 return ret;
1846 }
1847 nbd = idr_find(&nbd_index_idr, index);
1848 }
1849 }
1850 if (!nbd) {
1851 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1852 index);
1853 mutex_unlock(&nbd_index_mutex);
1854 return -EINVAL;
1855 }
1856
1857 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1858 test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1859 nbd->destroy_complete = &destroy_complete;
1860 mutex_unlock(&nbd_index_mutex);
1861
1862 /* Wait untill the the nbd stuff is totally destroyed */
1863 wait_for_completion(&destroy_complete);
1864 goto again;
1865 }
1866
1867 if (!refcount_inc_not_zero(&nbd->refs)) {
1868 mutex_unlock(&nbd_index_mutex);
1869 if (index == -1)
1870 goto again;
1871 printk(KERN_ERR "nbd: device at index %d is going down\n",
1872 index);
1873 return -EINVAL;
1874 }
1875 mutex_unlock(&nbd_index_mutex);
1876
1877 mutex_lock(&nbd->config_lock);
1878 if (refcount_read(&nbd->config_refs)) {
1879 mutex_unlock(&nbd->config_lock);
1880 nbd_put(nbd);
1881 if (index == -1)
1882 goto again;
1883 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1884 return -EBUSY;
1885 }
1886 if (WARN_ON(nbd->config)) {
1887 mutex_unlock(&nbd->config_lock);
1888 nbd_put(nbd);
1889 return -EINVAL;
1890 }
1891 config = nbd->config = nbd_alloc_config();
1892 if (!nbd->config) {
1893 mutex_unlock(&nbd->config_lock);
1894 nbd_put(nbd);
1895 printk(KERN_ERR "nbd: couldn't allocate config\n");
1896 return -ENOMEM;
1897 }
1898 refcount_set(&nbd->config_refs, 1);
1899 set_bit(NBD_RT_BOUND, &config->runtime_flags);
1900
1901 ret = nbd_genl_size_set(info, nbd);
1902 if (ret)
1903 goto out;
1904
1905 if (info->attrs[NBD_ATTR_TIMEOUT])
1906 nbd_set_cmd_timeout(nbd,
1907 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1908 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1909 config->dead_conn_timeout =
1910 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1911 config->dead_conn_timeout *= HZ;
1912 }
1913 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1914 config->flags =
1915 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1916 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1917 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1918 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1919 set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1920 &config->runtime_flags);
1921 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1922 put_dev = true;
1923 } else {
1924 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1925 }
1926 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1927 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1928 &config->runtime_flags);
1929 }
1930 }
1931
1932 if (info->attrs[NBD_ATTR_SOCKETS]) {
1933 struct nlattr *attr;
1934 int rem, fd;
1935
1936 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1937 rem) {
1938 struct nlattr *socks[NBD_SOCK_MAX+1];
1939
1940 if (nla_type(attr) != NBD_SOCK_ITEM) {
1941 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1942 ret = -EINVAL;
1943 goto out;
1944 }
1945 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1946 attr,
1947 nbd_sock_policy,
1948 info->extack);
1949 if (ret != 0) {
1950 printk(KERN_ERR "nbd: error processing sock list\n");
1951 ret = -EINVAL;
1952 goto out;
1953 }
1954 if (!socks[NBD_SOCK_FD])
1955 continue;
1956 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1957 ret = nbd_add_socket(nbd, fd, true);
1958 if (ret)
1959 goto out;
1960 }
1961 }
1962 ret = nbd_start_device(nbd);
1963 out:
1964 mutex_unlock(&nbd->config_lock);
1965 if (!ret) {
1966 set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1967 refcount_inc(&nbd->config_refs);
1968 nbd_connect_reply(info, nbd->index);
1969 }
1970 nbd_config_put(nbd);
1971 if (put_dev)
1972 nbd_put(nbd);
1973 return ret;
1974 }
1975
1976 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1977 {
1978 mutex_lock(&nbd->config_lock);
1979 nbd_disconnect(nbd);
1980 nbd_clear_sock(nbd);
1981 mutex_unlock(&nbd->config_lock);
1982 /*
1983 * Make sure recv thread has finished, so it does not drop the last
1984 * config ref and try to destroy the workqueue from inside the work
1985 * queue.
1986 */
1987 flush_workqueue(nbd->recv_workq);
1988 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1989 &nbd->config->runtime_flags))
1990 nbd_config_put(nbd);
1991 }
1992
1993 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1994 {
1995 struct nbd_device *nbd;
1996 int index;
1997
1998 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1999 return -EPERM;
2000
2001 if (!info->attrs[NBD_ATTR_INDEX]) {
2002 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
2003 return -EINVAL;
2004 }
2005 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2006 mutex_lock(&nbd_index_mutex);
2007 nbd = idr_find(&nbd_index_idr, index);
2008 if (!nbd) {
2009 mutex_unlock(&nbd_index_mutex);
2010 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2011 index);
2012 return -EINVAL;
2013 }
2014 if (!refcount_inc_not_zero(&nbd->refs)) {
2015 mutex_unlock(&nbd_index_mutex);
2016 printk(KERN_ERR "nbd: device at index %d is going down\n",
2017 index);
2018 return -EINVAL;
2019 }
2020 mutex_unlock(&nbd_index_mutex);
2021 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2022 nbd_put(nbd);
2023 return 0;
2024 }
2025 nbd_disconnect_and_put(nbd);
2026 nbd_config_put(nbd);
2027 nbd_put(nbd);
2028 return 0;
2029 }
2030
2031 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2032 {
2033 struct nbd_device *nbd = NULL;
2034 struct nbd_config *config;
2035 int index;
2036 int ret = 0;
2037 bool put_dev = false;
2038
2039 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2040 return -EPERM;
2041
2042 if (!info->attrs[NBD_ATTR_INDEX]) {
2043 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2044 return -EINVAL;
2045 }
2046 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2047 mutex_lock(&nbd_index_mutex);
2048 nbd = idr_find(&nbd_index_idr, index);
2049 if (!nbd) {
2050 mutex_unlock(&nbd_index_mutex);
2051 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2052 index);
2053 return -EINVAL;
2054 }
2055 if (!refcount_inc_not_zero(&nbd->refs)) {
2056 mutex_unlock(&nbd_index_mutex);
2057 printk(KERN_ERR "nbd: device at index %d is going down\n",
2058 index);
2059 return -EINVAL;
2060 }
2061 mutex_unlock(&nbd_index_mutex);
2062
2063 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2064 dev_err(nbd_to_dev(nbd),
2065 "not configured, cannot reconfigure\n");
2066 nbd_put(nbd);
2067 return -EINVAL;
2068 }
2069
2070 mutex_lock(&nbd->config_lock);
2071 config = nbd->config;
2072 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2073 !nbd->task_recv) {
2074 dev_err(nbd_to_dev(nbd),
2075 "not configured, cannot reconfigure\n");
2076 ret = -EINVAL;
2077 goto out;
2078 }
2079
2080 ret = nbd_genl_size_set(info, nbd);
2081 if (ret)
2082 goto out;
2083
2084 if (info->attrs[NBD_ATTR_TIMEOUT])
2085 nbd_set_cmd_timeout(nbd,
2086 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2087 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2088 config->dead_conn_timeout =
2089 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2090 config->dead_conn_timeout *= HZ;
2091 }
2092 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2093 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2094 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2095 if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2096 &config->runtime_flags))
2097 put_dev = true;
2098 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2099 } else {
2100 if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2101 &config->runtime_flags))
2102 refcount_inc(&nbd->refs);
2103 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2104 }
2105
2106 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2107 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2108 &config->runtime_flags);
2109 } else {
2110 clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2111 &config->runtime_flags);
2112 }
2113 }
2114
2115 if (info->attrs[NBD_ATTR_SOCKETS]) {
2116 struct nlattr *attr;
2117 int rem, fd;
2118
2119 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2120 rem) {
2121 struct nlattr *socks[NBD_SOCK_MAX+1];
2122
2123 if (nla_type(attr) != NBD_SOCK_ITEM) {
2124 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2125 ret = -EINVAL;
2126 goto out;
2127 }
2128 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2129 attr,
2130 nbd_sock_policy,
2131 info->extack);
2132 if (ret != 0) {
2133 printk(KERN_ERR "nbd: error processing sock list\n");
2134 ret = -EINVAL;
2135 goto out;
2136 }
2137 if (!socks[NBD_SOCK_FD])
2138 continue;
2139 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2140 ret = nbd_reconnect_socket(nbd, fd);
2141 if (ret) {
2142 if (ret == -ENOSPC)
2143 ret = 0;
2144 goto out;
2145 }
2146 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2147 }
2148 }
2149 out:
2150 mutex_unlock(&nbd->config_lock);
2151 nbd_config_put(nbd);
2152 nbd_put(nbd);
2153 if (put_dev)
2154 nbd_put(nbd);
2155 return ret;
2156 }
2157
2158 static const struct genl_small_ops nbd_connect_genl_ops[] = {
2159 {
2160 .cmd = NBD_CMD_CONNECT,
2161 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2162 .doit = nbd_genl_connect,
2163 },
2164 {
2165 .cmd = NBD_CMD_DISCONNECT,
2166 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2167 .doit = nbd_genl_disconnect,
2168 },
2169 {
2170 .cmd = NBD_CMD_RECONFIGURE,
2171 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2172 .doit = nbd_genl_reconfigure,
2173 },
2174 {
2175 .cmd = NBD_CMD_STATUS,
2176 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2177 .doit = nbd_genl_status,
2178 },
2179 };
2180
2181 static const struct genl_multicast_group nbd_mcast_grps[] = {
2182 { .name = NBD_GENL_MCAST_GROUP_NAME, },
2183 };
2184
2185 static struct genl_family nbd_genl_family __ro_after_init = {
2186 .hdrsize = 0,
2187 .name = NBD_GENL_FAMILY_NAME,
2188 .version = NBD_GENL_VERSION,
2189 .module = THIS_MODULE,
2190 .small_ops = nbd_connect_genl_ops,
2191 .n_small_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2192 .maxattr = NBD_ATTR_MAX,
2193 .policy = nbd_attr_policy,
2194 .mcgrps = nbd_mcast_grps,
2195 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2196 };
2197
2198 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2199 {
2200 struct nlattr *dev_opt;
2201 u8 connected = 0;
2202 int ret;
2203
2204 /* This is a little racey, but for status it's ok. The
2205 * reason we don't take a ref here is because we can't
2206 * take a ref in the index == -1 case as we would need
2207 * to put under the nbd_index_mutex, which could
2208 * deadlock if we are configured to remove ourselves
2209 * once we're disconnected.
2210 */
2211 if (refcount_read(&nbd->config_refs))
2212 connected = 1;
2213 dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2214 if (!dev_opt)
2215 return -EMSGSIZE;
2216 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2217 if (ret)
2218 return -EMSGSIZE;
2219 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2220 connected);
2221 if (ret)
2222 return -EMSGSIZE;
2223 nla_nest_end(reply, dev_opt);
2224 return 0;
2225 }
2226
2227 static int status_cb(int id, void *ptr, void *data)
2228 {
2229 struct nbd_device *nbd = ptr;
2230 return populate_nbd_status(nbd, (struct sk_buff *)data);
2231 }
2232
2233 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2234 {
2235 struct nlattr *dev_list;
2236 struct sk_buff *reply;
2237 void *reply_head;
2238 size_t msg_size;
2239 int index = -1;
2240 int ret = -ENOMEM;
2241
2242 if (info->attrs[NBD_ATTR_INDEX])
2243 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2244
2245 mutex_lock(&nbd_index_mutex);
2246
2247 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2248 nla_attr_size(sizeof(u8)));
2249 msg_size *= (index == -1) ? nbd_total_devices : 1;
2250
2251 reply = genlmsg_new(msg_size, GFP_KERNEL);
2252 if (!reply)
2253 goto out;
2254 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2255 NBD_CMD_STATUS);
2256 if (!reply_head) {
2257 nlmsg_free(reply);
2258 goto out;
2259 }
2260
2261 dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2262 if (index == -1) {
2263 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2264 if (ret) {
2265 nlmsg_free(reply);
2266 goto out;
2267 }
2268 } else {
2269 struct nbd_device *nbd;
2270 nbd = idr_find(&nbd_index_idr, index);
2271 if (nbd) {
2272 ret = populate_nbd_status(nbd, reply);
2273 if (ret) {
2274 nlmsg_free(reply);
2275 goto out;
2276 }
2277 }
2278 }
2279 nla_nest_end(reply, dev_list);
2280 genlmsg_end(reply, reply_head);
2281 ret = genlmsg_reply(reply, info);
2282 out:
2283 mutex_unlock(&nbd_index_mutex);
2284 return ret;
2285 }
2286
2287 static void nbd_connect_reply(struct genl_info *info, int index)
2288 {
2289 struct sk_buff *skb;
2290 void *msg_head;
2291 int ret;
2292
2293 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2294 if (!skb)
2295 return;
2296 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2297 NBD_CMD_CONNECT);
2298 if (!msg_head) {
2299 nlmsg_free(skb);
2300 return;
2301 }
2302 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2303 if (ret) {
2304 nlmsg_free(skb);
2305 return;
2306 }
2307 genlmsg_end(skb, msg_head);
2308 genlmsg_reply(skb, info);
2309 }
2310
2311 static void nbd_mcast_index(int index)
2312 {
2313 struct sk_buff *skb;
2314 void *msg_head;
2315 int ret;
2316
2317 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2318 if (!skb)
2319 return;
2320 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2321 NBD_CMD_LINK_DEAD);
2322 if (!msg_head) {
2323 nlmsg_free(skb);
2324 return;
2325 }
2326 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2327 if (ret) {
2328 nlmsg_free(skb);
2329 return;
2330 }
2331 genlmsg_end(skb, msg_head);
2332 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2333 }
2334
2335 static void nbd_dead_link_work(struct work_struct *work)
2336 {
2337 struct link_dead_args *args = container_of(work, struct link_dead_args,
2338 work);
2339 nbd_mcast_index(args->index);
2340 kfree(args);
2341 }
2342
2343 static int __init nbd_init(void)
2344 {
2345 int i;
2346
2347 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2348
2349 if (max_part < 0) {
2350 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2351 return -EINVAL;
2352 }
2353
2354 part_shift = 0;
2355 if (max_part > 0) {
2356 part_shift = fls(max_part);
2357
2358 /*
2359 * Adjust max_part according to part_shift as it is exported
2360 * to user space so that user can know the max number of
2361 * partition kernel should be able to manage.
2362 *
2363 * Note that -1 is required because partition 0 is reserved
2364 * for the whole disk.
2365 */
2366 max_part = (1UL << part_shift) - 1;
2367 }
2368
2369 if ((1UL << part_shift) > DISK_MAX_PARTS)
2370 return -EINVAL;
2371
2372 if (nbds_max > 1UL << (MINORBITS - part_shift))
2373 return -EINVAL;
2374
2375 if (register_blkdev(NBD_MAJOR, "nbd"))
2376 return -EIO;
2377
2378 if (genl_register_family(&nbd_genl_family)) {
2379 unregister_blkdev(NBD_MAJOR, "nbd");
2380 return -EINVAL;
2381 }
2382 nbd_dbg_init();
2383
2384 mutex_lock(&nbd_index_mutex);
2385 for (i = 0; i < nbds_max; i++)
2386 nbd_dev_add(i);
2387 mutex_unlock(&nbd_index_mutex);
2388 return 0;
2389 }
2390
2391 static int nbd_exit_cb(int id, void *ptr, void *data)
2392 {
2393 struct list_head *list = (struct list_head *)data;
2394 struct nbd_device *nbd = ptr;
2395
2396 list_add_tail(&nbd->list, list);
2397 return 0;
2398 }
2399
2400 static void __exit nbd_cleanup(void)
2401 {
2402 struct nbd_device *nbd;
2403 LIST_HEAD(del_list);
2404
2405 nbd_dbg_close();
2406
2407 mutex_lock(&nbd_index_mutex);
2408 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2409 mutex_unlock(&nbd_index_mutex);
2410
2411 while (!list_empty(&del_list)) {
2412 nbd = list_first_entry(&del_list, struct nbd_device, list);
2413 list_del_init(&nbd->list);
2414 if (refcount_read(&nbd->refs) != 1)
2415 printk(KERN_ERR "nbd: possibly leaking a device\n");
2416 nbd_put(nbd);
2417 }
2418
2419 idr_destroy(&nbd_index_idr);
2420 genl_unregister_family(&nbd_genl_family);
2421 unregister_blkdev(NBD_MAJOR, "nbd");
2422 }
2423
2424 module_init(nbd_init);
2425 module_exit(nbd_cleanup);
2426
2427 MODULE_DESCRIPTION("Network Block Device");
2428 MODULE_LICENSE("GPL");
2429
2430 module_param(nbds_max, int, 0444);
2431 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2432 module_param(max_part, int, 0444);
2433 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");
Linux with added FPC-III support