]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/block/nbd.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'drm-misc-fixes-2017-06-02' of git://anongit.freedesktop.org/git/drm-misc...
[mirror_ubuntu-bionic-kernel.git] / drivers / block / nbd.c
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 int 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;
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 nbd_size_update(nbd);
247 }
248
249 static void nbd_complete_rq(struct request *req)
250 {
251 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
252
253 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd,
254 cmd->status ? "failed" : "done");
255
256 blk_mq_end_request(req, cmd->status);
257 }
258
259 /*
260 * Forcibly shutdown the socket causing all listeners to error
261 */
262 static void sock_shutdown(struct nbd_device *nbd)
263 {
264 struct nbd_config *config = nbd->config;
265 int i;
266
267 if (config->num_connections == 0)
268 return;
269 if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
270 return;
271
272 for (i = 0; i < config->num_connections; i++) {
273 struct nbd_sock *nsock = config->socks[i];
274 mutex_lock(&nsock->tx_lock);
275 nbd_mark_nsock_dead(nbd, nsock, 0);
276 mutex_unlock(&nsock->tx_lock);
277 }
278 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
279 }
280
281 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
282 bool reserved)
283 {
284 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
285 struct nbd_device *nbd = cmd->nbd;
286 struct nbd_config *config;
287
288 if (!refcount_inc_not_zero(&nbd->config_refs)) {
289 cmd->status = -EIO;
290 return BLK_EH_HANDLED;
291 }
292
293 /* If we are waiting on our dead timer then we could get timeout
294 * callbacks for our request. For this we just want to reset the timer
295 * and let the queue side take care of everything.
296 */
297 if (!completion_done(&cmd->send_complete)) {
298 nbd_config_put(nbd);
299 return BLK_EH_RESET_TIMER;
300 }
301 config = nbd->config;
302
303 if (config->num_connections > 1) {
304 dev_err_ratelimited(nbd_to_dev(nbd),
305 "Connection timed out, retrying\n");
306 /*
307 * Hooray we have more connections, requeue this IO, the submit
308 * path will put it on a real connection.
309 */
310 if (config->socks && config->num_connections > 1) {
311 if (cmd->index < config->num_connections) {
312 struct nbd_sock *nsock =
313 config->socks[cmd->index];
314 mutex_lock(&nsock->tx_lock);
315 /* We can have multiple outstanding requests, so
316 * we don't want to mark the nsock dead if we've
317 * already reconnected with a new socket, so
318 * only mark it dead if its the same socket we
319 * were sent out on.
320 */
321 if (cmd->cookie == nsock->cookie)
322 nbd_mark_nsock_dead(nbd, nsock, 1);
323 mutex_unlock(&nsock->tx_lock);
324 }
325 blk_mq_requeue_request(req, true);
326 nbd_config_put(nbd);
327 return BLK_EH_NOT_HANDLED;
328 }
329 } else {
330 dev_err_ratelimited(nbd_to_dev(nbd),
331 "Connection timed out\n");
332 }
333 set_bit(NBD_TIMEDOUT, &config->runtime_flags);
334 cmd->status = -EIO;
335 sock_shutdown(nbd);
336 nbd_config_put(nbd);
337
338 return BLK_EH_HANDLED;
339 }
340
341 /*
342 * Send or receive packet.
343 */
344 static int sock_xmit(struct nbd_device *nbd, int index, int send,
345 struct iov_iter *iter, int msg_flags, int *sent)
346 {
347 struct nbd_config *config = nbd->config;
348 struct socket *sock = config->socks[index]->sock;
349 int result;
350 struct msghdr msg;
351 unsigned int noreclaim_flag;
352
353 if (unlikely(!sock)) {
354 dev_err_ratelimited(disk_to_dev(nbd->disk),
355 "Attempted %s on closed socket in sock_xmit\n",
356 (send ? "send" : "recv"));
357 return -EINVAL;
358 }
359
360 msg.msg_iter = *iter;
361
362 noreclaim_flag = memalloc_noreclaim_save();
363 do {
364 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
365 msg.msg_name = NULL;
366 msg.msg_namelen = 0;
367 msg.msg_control = NULL;
368 msg.msg_controllen = 0;
369 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
370
371 if (send)
372 result = sock_sendmsg(sock, &msg);
373 else
374 result = sock_recvmsg(sock, &msg, msg.msg_flags);
375
376 if (result <= 0) {
377 if (result == 0)
378 result = -EPIPE; /* short read */
379 break;
380 }
381 if (sent)
382 *sent += result;
383 } while (msg_data_left(&msg));
384
385 memalloc_noreclaim_restore(noreclaim_flag);
386
387 return result;
388 }
389
390 /* always call with the tx_lock held */
391 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
392 {
393 struct request *req = blk_mq_rq_from_pdu(cmd);
394 struct nbd_config *config = nbd->config;
395 struct nbd_sock *nsock = config->socks[index];
396 int result;
397 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
398 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
399 struct iov_iter from;
400 unsigned long size = blk_rq_bytes(req);
401 struct bio *bio;
402 u32 type;
403 u32 tag = blk_mq_unique_tag(req);
404 int sent = nsock->sent, skip = 0;
405
406 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
407
408 switch (req_op(req)) {
409 case REQ_OP_DISCARD:
410 type = NBD_CMD_TRIM;
411 break;
412 case REQ_OP_FLUSH:
413 type = NBD_CMD_FLUSH;
414 break;
415 case REQ_OP_WRITE:
416 type = NBD_CMD_WRITE;
417 break;
418 case REQ_OP_READ:
419 type = NBD_CMD_READ;
420 break;
421 default:
422 return -EIO;
423 }
424
425 if (rq_data_dir(req) == WRITE &&
426 (config->flags & NBD_FLAG_READ_ONLY)) {
427 dev_err_ratelimited(disk_to_dev(nbd->disk),
428 "Write on read-only\n");
429 return -EIO;
430 }
431
432 /* We did a partial send previously, and we at least sent the whole
433 * request struct, so just go and send the rest of the pages in the
434 * request.
435 */
436 if (sent) {
437 if (sent >= sizeof(request)) {
438 skip = sent - sizeof(request);
439 goto send_pages;
440 }
441 iov_iter_advance(&from, sent);
442 }
443 cmd->index = index;
444 cmd->cookie = nsock->cookie;
445 request.type = htonl(type);
446 if (type != NBD_CMD_FLUSH) {
447 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
448 request.len = htonl(size);
449 }
450 memcpy(request.handle, &tag, sizeof(tag));
451
452 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
453 cmd, nbdcmd_to_ascii(type),
454 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
455 result = sock_xmit(nbd, index, 1, &from,
456 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
457 if (result <= 0) {
458 if (result == -ERESTARTSYS) {
459 /* If we havne't sent anything we can just return BUSY,
460 * however if we have sent something we need to make
461 * sure we only allow this req to be sent until we are
462 * completely done.
463 */
464 if (sent) {
465 nsock->pending = req;
466 nsock->sent = sent;
467 }
468 return BLK_MQ_RQ_QUEUE_BUSY;
469 }
470 dev_err_ratelimited(disk_to_dev(nbd->disk),
471 "Send control failed (result %d)\n", result);
472 return -EAGAIN;
473 }
474 send_pages:
475 if (type != NBD_CMD_WRITE)
476 goto out;
477
478 bio = req->bio;
479 while (bio) {
480 struct bio *next = bio->bi_next;
481 struct bvec_iter iter;
482 struct bio_vec bvec;
483
484 bio_for_each_segment(bvec, bio, iter) {
485 bool is_last = !next && bio_iter_last(bvec, iter);
486 int flags = is_last ? 0 : MSG_MORE;
487
488 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
489 cmd, bvec.bv_len);
490 iov_iter_bvec(&from, ITER_BVEC | WRITE,
491 &bvec, 1, bvec.bv_len);
492 if (skip) {
493 if (skip >= iov_iter_count(&from)) {
494 skip -= iov_iter_count(&from);
495 continue;
496 }
497 iov_iter_advance(&from, skip);
498 skip = 0;
499 }
500 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
501 if (result <= 0) {
502 if (result == -ERESTARTSYS) {
503 /* We've already sent the header, we
504 * have no choice but to set pending and
505 * return BUSY.
506 */
507 nsock->pending = req;
508 nsock->sent = sent;
509 return BLK_MQ_RQ_QUEUE_BUSY;
510 }
511 dev_err(disk_to_dev(nbd->disk),
512 "Send data failed (result %d)\n",
513 result);
514 return -EAGAIN;
515 }
516 /*
517 * The completion might already have come in,
518 * so break for the last one instead of letting
519 * the iterator do it. This prevents use-after-free
520 * of the bio.
521 */
522 if (is_last)
523 break;
524 }
525 bio = next;
526 }
527 out:
528 nsock->pending = NULL;
529 nsock->sent = 0;
530 return 0;
531 }
532
533 /* NULL returned = something went wrong, inform userspace */
534 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
535 {
536 struct nbd_config *config = nbd->config;
537 int result;
538 struct nbd_reply reply;
539 struct nbd_cmd *cmd;
540 struct request *req = NULL;
541 u16 hwq;
542 u32 tag;
543 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
544 struct iov_iter to;
545
546 reply.magic = 0;
547 iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
548 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
549 if (result <= 0) {
550 if (!nbd_disconnected(config))
551 dev_err(disk_to_dev(nbd->disk),
552 "Receive control failed (result %d)\n", result);
553 return ERR_PTR(result);
554 }
555
556 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
557 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
558 (unsigned long)ntohl(reply.magic));
559 return ERR_PTR(-EPROTO);
560 }
561
562 memcpy(&tag, reply.handle, sizeof(u32));
563
564 hwq = blk_mq_unique_tag_to_hwq(tag);
565 if (hwq < nbd->tag_set.nr_hw_queues)
566 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
567 blk_mq_unique_tag_to_tag(tag));
568 if (!req || !blk_mq_request_started(req)) {
569 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
570 tag, req);
571 return ERR_PTR(-ENOENT);
572 }
573 cmd = blk_mq_rq_to_pdu(req);
574 if (ntohl(reply.error)) {
575 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
576 ntohl(reply.error));
577 cmd->status = -EIO;
578 return cmd;
579 }
580
581 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
582 if (rq_data_dir(req) != WRITE) {
583 struct req_iterator iter;
584 struct bio_vec bvec;
585
586 rq_for_each_segment(bvec, req, iter) {
587 iov_iter_bvec(&to, ITER_BVEC | READ,
588 &bvec, 1, bvec.bv_len);
589 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
590 if (result <= 0) {
591 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
592 result);
593 /*
594 * If we've disconnected or we only have 1
595 * connection then we need to make sure we
596 * complete this request, otherwise error out
597 * and let the timeout stuff handle resubmitting
598 * this request onto another connection.
599 */
600 if (nbd_disconnected(config) ||
601 config->num_connections <= 1) {
602 cmd->status = -EIO;
603 return cmd;
604 }
605 return ERR_PTR(-EIO);
606 }
607 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
608 cmd, bvec.bv_len);
609 }
610 } else {
611 /* See the comment in nbd_queue_rq. */
612 wait_for_completion(&cmd->send_complete);
613 }
614 return cmd;
615 }
616
617 static void recv_work(struct work_struct *work)
618 {
619 struct recv_thread_args *args = container_of(work,
620 struct recv_thread_args,
621 work);
622 struct nbd_device *nbd = args->nbd;
623 struct nbd_config *config = nbd->config;
624 struct nbd_cmd *cmd;
625 int ret = 0;
626
627 while (1) {
628 cmd = nbd_read_stat(nbd, args->index);
629 if (IS_ERR(cmd)) {
630 struct nbd_sock *nsock = config->socks[args->index];
631
632 mutex_lock(&nsock->tx_lock);
633 nbd_mark_nsock_dead(nbd, nsock, 1);
634 mutex_unlock(&nsock->tx_lock);
635 ret = PTR_ERR(cmd);
636 break;
637 }
638
639 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
640 }
641 atomic_dec(&config->recv_threads);
642 wake_up(&config->recv_wq);
643 nbd_config_put(nbd);
644 kfree(args);
645 }
646
647 static void nbd_clear_req(struct request *req, void *data, bool reserved)
648 {
649 struct nbd_cmd *cmd;
650
651 if (!blk_mq_request_started(req))
652 return;
653 cmd = blk_mq_rq_to_pdu(req);
654 cmd->status = -EIO;
655 blk_mq_complete_request(req);
656 }
657
658 static void nbd_clear_que(struct nbd_device *nbd)
659 {
660 blk_mq_stop_hw_queues(nbd->disk->queue);
661 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
662 blk_mq_start_hw_queues(nbd->disk->queue);
663 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
664 }
665
666 static int find_fallback(struct nbd_device *nbd, int index)
667 {
668 struct nbd_config *config = nbd->config;
669 int new_index = -1;
670 struct nbd_sock *nsock = config->socks[index];
671 int fallback = nsock->fallback_index;
672
673 if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
674 return new_index;
675
676 if (config->num_connections <= 1) {
677 dev_err_ratelimited(disk_to_dev(nbd->disk),
678 "Attempted send on invalid socket\n");
679 return new_index;
680 }
681
682 if (fallback >= 0 && fallback < config->num_connections &&
683 !config->socks[fallback]->dead)
684 return fallback;
685
686 if (nsock->fallback_index < 0 ||
687 nsock->fallback_index >= config->num_connections ||
688 config->socks[nsock->fallback_index]->dead) {
689 int i;
690 for (i = 0; i < config->num_connections; i++) {
691 if (i == index)
692 continue;
693 if (!config->socks[i]->dead) {
694 new_index = i;
695 break;
696 }
697 }
698 nsock->fallback_index = new_index;
699 if (new_index < 0) {
700 dev_err_ratelimited(disk_to_dev(nbd->disk),
701 "Dead connection, failed to find a fallback\n");
702 return new_index;
703 }
704 }
705 new_index = nsock->fallback_index;
706 return new_index;
707 }
708
709 static int wait_for_reconnect(struct nbd_device *nbd)
710 {
711 struct nbd_config *config = nbd->config;
712 if (!config->dead_conn_timeout)
713 return 0;
714 if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
715 return 0;
716 wait_event_interruptible_timeout(config->conn_wait,
717 atomic_read(&config->live_connections),
718 config->dead_conn_timeout);
719 return atomic_read(&config->live_connections);
720 }
721
722 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
723 {
724 struct request *req = blk_mq_rq_from_pdu(cmd);
725 struct nbd_device *nbd = cmd->nbd;
726 struct nbd_config *config;
727 struct nbd_sock *nsock;
728 int ret;
729
730 if (!refcount_inc_not_zero(&nbd->config_refs)) {
731 dev_err_ratelimited(disk_to_dev(nbd->disk),
732 "Socks array is empty\n");
733 return -EINVAL;
734 }
735 config = nbd->config;
736
737 if (index >= config->num_connections) {
738 dev_err_ratelimited(disk_to_dev(nbd->disk),
739 "Attempted send on invalid socket\n");
740 nbd_config_put(nbd);
741 return -EINVAL;
742 }
743 cmd->status = 0;
744 again:
745 nsock = config->socks[index];
746 mutex_lock(&nsock->tx_lock);
747 if (nsock->dead) {
748 int old_index = index;
749 index = find_fallback(nbd, index);
750 mutex_unlock(&nsock->tx_lock);
751 if (index < 0) {
752 if (wait_for_reconnect(nbd)) {
753 index = old_index;
754 goto again;
755 }
756 /* All the sockets should already be down at this point,
757 * we just want to make sure that DISCONNECTED is set so
758 * any requests that come in that were queue'ed waiting
759 * for the reconnect timer don't trigger the timer again
760 * and instead just error out.
761 */
762 sock_shutdown(nbd);
763 nbd_config_put(nbd);
764 return -EIO;
765 }
766 goto again;
767 }
768
769 /* Handle the case that we have a pending request that was partially
770 * transmitted that _has_ to be serviced first. We need to call requeue
771 * here so that it gets put _after_ the request that is already on the
772 * dispatch list.
773 */
774 if (unlikely(nsock->pending && nsock->pending != req)) {
775 blk_mq_requeue_request(req, true);
776 ret = 0;
777 goto out;
778 }
779 /*
780 * Some failures are related to the link going down, so anything that
781 * returns EAGAIN can be retried on a different socket.
782 */
783 ret = nbd_send_cmd(nbd, cmd, index);
784 if (ret == -EAGAIN) {
785 dev_err_ratelimited(disk_to_dev(nbd->disk),
786 "Request send failed trying another connection\n");
787 nbd_mark_nsock_dead(nbd, nsock, 1);
788 mutex_unlock(&nsock->tx_lock);
789 goto again;
790 }
791 out:
792 mutex_unlock(&nsock->tx_lock);
793 nbd_config_put(nbd);
794 return ret;
795 }
796
797 static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
798 const struct blk_mq_queue_data *bd)
799 {
800 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
801 int ret;
802
803 /*
804 * Since we look at the bio's to send the request over the network we
805 * need to make sure the completion work doesn't mark this request done
806 * before we are done doing our send. This keeps us from dereferencing
807 * freed data if we have particularly fast completions (ie we get the
808 * completion before we exit sock_xmit on the last bvec) or in the case
809 * that the server is misbehaving (or there was an error) before we're
810 * done sending everything over the wire.
811 */
812 init_completion(&cmd->send_complete);
813 blk_mq_start_request(bd->rq);
814
815 /* We can be called directly from the user space process, which means we
816 * could possibly have signals pending so our sendmsg will fail. In
817 * this case we need to return that we are busy, otherwise error out as
818 * appropriate.
819 */
820 ret = nbd_handle_cmd(cmd, hctx->queue_num);
821 if (ret < 0)
822 ret = BLK_MQ_RQ_QUEUE_ERROR;
823 if (!ret)
824 ret = BLK_MQ_RQ_QUEUE_OK;
825 complete(&cmd->send_complete);
826
827 return ret;
828 }
829
830 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
831 bool netlink)
832 {
833 struct nbd_config *config = nbd->config;
834 struct socket *sock;
835 struct nbd_sock **socks;
836 struct nbd_sock *nsock;
837 int err;
838
839 sock = sockfd_lookup(arg, &err);
840 if (!sock)
841 return err;
842
843 if (!netlink && !nbd->task_setup &&
844 !test_bit(NBD_BOUND, &config->runtime_flags))
845 nbd->task_setup = current;
846
847 if (!netlink &&
848 (nbd->task_setup != current ||
849 test_bit(NBD_BOUND, &config->runtime_flags))) {
850 dev_err(disk_to_dev(nbd->disk),
851 "Device being setup by another task");
852 sockfd_put(sock);
853 return -EBUSY;
854 }
855
856 socks = krealloc(config->socks, (config->num_connections + 1) *
857 sizeof(struct nbd_sock *), GFP_KERNEL);
858 if (!socks) {
859 sockfd_put(sock);
860 return -ENOMEM;
861 }
862 nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
863 if (!nsock) {
864 sockfd_put(sock);
865 return -ENOMEM;
866 }
867
868 config->socks = socks;
869
870 nsock->fallback_index = -1;
871 nsock->dead = false;
872 mutex_init(&nsock->tx_lock);
873 nsock->sock = sock;
874 nsock->pending = NULL;
875 nsock->sent = 0;
876 nsock->cookie = 0;
877 socks[config->num_connections++] = nsock;
878 atomic_inc(&config->live_connections);
879
880 return 0;
881 }
882
883 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
884 {
885 struct nbd_config *config = nbd->config;
886 struct socket *sock, *old;
887 struct recv_thread_args *args;
888 int i;
889 int err;
890
891 sock = sockfd_lookup(arg, &err);
892 if (!sock)
893 return err;
894
895 args = kzalloc(sizeof(*args), GFP_KERNEL);
896 if (!args) {
897 sockfd_put(sock);
898 return -ENOMEM;
899 }
900
901 for (i = 0; i < config->num_connections; i++) {
902 struct nbd_sock *nsock = config->socks[i];
903
904 if (!nsock->dead)
905 continue;
906
907 mutex_lock(&nsock->tx_lock);
908 if (!nsock->dead) {
909 mutex_unlock(&nsock->tx_lock);
910 continue;
911 }
912 sk_set_memalloc(sock->sk);
913 atomic_inc(&config->recv_threads);
914 refcount_inc(&nbd->config_refs);
915 old = nsock->sock;
916 nsock->fallback_index = -1;
917 nsock->sock = sock;
918 nsock->dead = false;
919 INIT_WORK(&args->work, recv_work);
920 args->index = i;
921 args->nbd = nbd;
922 nsock->cookie++;
923 mutex_unlock(&nsock->tx_lock);
924 sockfd_put(old);
925
926 /* We take the tx_mutex in an error path in the recv_work, so we
927 * need to queue_work outside of the tx_mutex.
928 */
929 queue_work(recv_workqueue, &args->work);
930
931 atomic_inc(&config->live_connections);
932 wake_up(&config->conn_wait);
933 return 0;
934 }
935 sockfd_put(sock);
936 kfree(args);
937 return -ENOSPC;
938 }
939
940 static void nbd_bdev_reset(struct block_device *bdev)
941 {
942 if (bdev->bd_openers > 1)
943 return;
944 bd_set_size(bdev, 0);
945 if (max_part > 0) {
946 blkdev_reread_part(bdev);
947 bdev->bd_invalidated = 1;
948 }
949 }
950
951 static void nbd_parse_flags(struct nbd_device *nbd)
952 {
953 struct nbd_config *config = nbd->config;
954 if (config->flags & NBD_FLAG_READ_ONLY)
955 set_disk_ro(nbd->disk, true);
956 else
957 set_disk_ro(nbd->disk, false);
958 if (config->flags & NBD_FLAG_SEND_TRIM)
959 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
960 if (config->flags & NBD_FLAG_SEND_FLUSH)
961 blk_queue_write_cache(nbd->disk->queue, true, false);
962 else
963 blk_queue_write_cache(nbd->disk->queue, false, false);
964 }
965
966 static void send_disconnects(struct nbd_device *nbd)
967 {
968 struct nbd_config *config = nbd->config;
969 struct nbd_request request = {
970 .magic = htonl(NBD_REQUEST_MAGIC),
971 .type = htonl(NBD_CMD_DISC),
972 };
973 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
974 struct iov_iter from;
975 int i, ret;
976
977 for (i = 0; i < config->num_connections; i++) {
978 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
979 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
980 if (ret <= 0)
981 dev_err(disk_to_dev(nbd->disk),
982 "Send disconnect failed %d\n", ret);
983 }
984 }
985
986 static int nbd_disconnect(struct nbd_device *nbd)
987 {
988 struct nbd_config *config = nbd->config;
989
990 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
991 if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
992 &config->runtime_flags))
993 send_disconnects(nbd);
994 return 0;
995 }
996
997 static void nbd_clear_sock(struct nbd_device *nbd)
998 {
999 sock_shutdown(nbd);
1000 nbd_clear_que(nbd);
1001 nbd->task_setup = NULL;
1002 }
1003
1004 static void nbd_config_put(struct nbd_device *nbd)
1005 {
1006 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1007 &nbd->config_lock)) {
1008 struct nbd_config *config = nbd->config;
1009 nbd_dev_dbg_close(nbd);
1010 nbd_size_clear(nbd);
1011 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1012 &config->runtime_flags))
1013 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1014 nbd->task_recv = NULL;
1015 nbd_clear_sock(nbd);
1016 if (config->num_connections) {
1017 int i;
1018 for (i = 0; i < config->num_connections; i++) {
1019 sockfd_put(config->socks[i]->sock);
1020 kfree(config->socks[i]);
1021 }
1022 kfree(config->socks);
1023 }
1024 kfree(nbd->config);
1025 nbd->config = NULL;
1026
1027 nbd->tag_set.timeout = 0;
1028 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1029
1030 mutex_unlock(&nbd->config_lock);
1031 nbd_put(nbd);
1032 module_put(THIS_MODULE);
1033 }
1034 }
1035
1036 static int nbd_start_device(struct nbd_device *nbd)
1037 {
1038 struct nbd_config *config = nbd->config;
1039 int num_connections = config->num_connections;
1040 int error = 0, i;
1041
1042 if (nbd->task_recv)
1043 return -EBUSY;
1044 if (!config->socks)
1045 return -EINVAL;
1046 if (num_connections > 1 &&
1047 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1048 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1049 return -EINVAL;
1050 }
1051
1052 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1053 nbd->task_recv = current;
1054
1055 nbd_parse_flags(nbd);
1056
1057 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1058 if (error) {
1059 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1060 return error;
1061 }
1062 set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1063
1064 nbd_dev_dbg_init(nbd);
1065 for (i = 0; i < num_connections; i++) {
1066 struct recv_thread_args *args;
1067
1068 args = kzalloc(sizeof(*args), GFP_KERNEL);
1069 if (!args) {
1070 sock_shutdown(nbd);
1071 return -ENOMEM;
1072 }
1073 sk_set_memalloc(config->socks[i]->sock->sk);
1074 atomic_inc(&config->recv_threads);
1075 refcount_inc(&nbd->config_refs);
1076 INIT_WORK(&args->work, recv_work);
1077 args->nbd = nbd;
1078 args->index = i;
1079 queue_work(recv_workqueue, &args->work);
1080 }
1081 return error;
1082 }
1083
1084 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1085 {
1086 struct nbd_config *config = nbd->config;
1087 int ret;
1088
1089 ret = nbd_start_device(nbd);
1090 if (ret)
1091 return ret;
1092
1093 bd_set_size(bdev, config->bytesize);
1094 if (max_part)
1095 bdev->bd_invalidated = 1;
1096 mutex_unlock(&nbd->config_lock);
1097 ret = wait_event_interruptible(config->recv_wq,
1098 atomic_read(&config->recv_threads) == 0);
1099 if (ret)
1100 sock_shutdown(nbd);
1101 mutex_lock(&nbd->config_lock);
1102 bd_set_size(bdev, 0);
1103 /* user requested, ignore socket errors */
1104 if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1105 ret = 0;
1106 if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1107 ret = -ETIMEDOUT;
1108 return ret;
1109 }
1110
1111 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1112 struct block_device *bdev)
1113 {
1114 sock_shutdown(nbd);
1115 kill_bdev(bdev);
1116 nbd_bdev_reset(bdev);
1117 if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1118 &nbd->config->runtime_flags))
1119 nbd_config_put(nbd);
1120 }
1121
1122 /* Must be called with config_lock held */
1123 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1124 unsigned int cmd, unsigned long arg)
1125 {
1126 struct nbd_config *config = nbd->config;
1127
1128 switch (cmd) {
1129 case NBD_DISCONNECT:
1130 return nbd_disconnect(nbd);
1131 case NBD_CLEAR_SOCK:
1132 nbd_clear_sock_ioctl(nbd, bdev);
1133 return 0;
1134 case NBD_SET_SOCK:
1135 return nbd_add_socket(nbd, arg, false);
1136 case NBD_SET_BLKSIZE:
1137 nbd_size_set(nbd, arg,
1138 div_s64(config->bytesize, arg));
1139 return 0;
1140 case NBD_SET_SIZE:
1141 nbd_size_set(nbd, config->blksize,
1142 div_s64(arg, config->blksize));
1143 return 0;
1144 case NBD_SET_SIZE_BLOCKS:
1145 nbd_size_set(nbd, config->blksize, arg);
1146 return 0;
1147 case NBD_SET_TIMEOUT:
1148 if (arg) {
1149 nbd->tag_set.timeout = arg * HZ;
1150 blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1151 }
1152 return 0;
1153
1154 case NBD_SET_FLAGS:
1155 config->flags = arg;
1156 return 0;
1157 case NBD_DO_IT:
1158 return nbd_start_device_ioctl(nbd, bdev);
1159 case NBD_CLEAR_QUE:
1160 /*
1161 * This is for compatibility only. The queue is always cleared
1162 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1163 */
1164 return 0;
1165 case NBD_PRINT_DEBUG:
1166 /*
1167 * For compatibility only, we no longer keep a list of
1168 * outstanding requests.
1169 */
1170 return 0;
1171 }
1172 return -ENOTTY;
1173 }
1174
1175 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1176 unsigned int cmd, unsigned long arg)
1177 {
1178 struct nbd_device *nbd = bdev->bd_disk->private_data;
1179 struct nbd_config *config = nbd->config;
1180 int error = -EINVAL;
1181
1182 if (!capable(CAP_SYS_ADMIN))
1183 return -EPERM;
1184
1185 mutex_lock(&nbd->config_lock);
1186
1187 /* Don't allow ioctl operations on a nbd device that was created with
1188 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1189 */
1190 if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1191 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1192 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1193 else
1194 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1195 mutex_unlock(&nbd->config_lock);
1196 return error;
1197 }
1198
1199 static struct nbd_config *nbd_alloc_config(void)
1200 {
1201 struct nbd_config *config;
1202
1203 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1204 if (!config)
1205 return NULL;
1206 atomic_set(&config->recv_threads, 0);
1207 init_waitqueue_head(&config->recv_wq);
1208 init_waitqueue_head(&config->conn_wait);
1209 config->blksize = 1024;
1210 atomic_set(&config->live_connections, 0);
1211 try_module_get(THIS_MODULE);
1212 return config;
1213 }
1214
1215 static int nbd_open(struct block_device *bdev, fmode_t mode)
1216 {
1217 struct nbd_device *nbd;
1218 int ret = 0;
1219
1220 mutex_lock(&nbd_index_mutex);
1221 nbd = bdev->bd_disk->private_data;
1222 if (!nbd) {
1223 ret = -ENXIO;
1224 goto out;
1225 }
1226 if (!refcount_inc_not_zero(&nbd->refs)) {
1227 ret = -ENXIO;
1228 goto out;
1229 }
1230 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1231 struct nbd_config *config;
1232
1233 mutex_lock(&nbd->config_lock);
1234 if (refcount_inc_not_zero(&nbd->config_refs)) {
1235 mutex_unlock(&nbd->config_lock);
1236 goto out;
1237 }
1238 config = nbd->config = nbd_alloc_config();
1239 if (!config) {
1240 ret = -ENOMEM;
1241 mutex_unlock(&nbd->config_lock);
1242 goto out;
1243 }
1244 refcount_set(&nbd->config_refs, 1);
1245 refcount_inc(&nbd->refs);
1246 mutex_unlock(&nbd->config_lock);
1247 }
1248 out:
1249 mutex_unlock(&nbd_index_mutex);
1250 return ret;
1251 }
1252
1253 static void nbd_release(struct gendisk *disk, fmode_t mode)
1254 {
1255 struct nbd_device *nbd = disk->private_data;
1256 nbd_config_put(nbd);
1257 nbd_put(nbd);
1258 }
1259
1260 static const struct block_device_operations nbd_fops =
1261 {
1262 .owner = THIS_MODULE,
1263 .open = nbd_open,
1264 .release = nbd_release,
1265 .ioctl = nbd_ioctl,
1266 .compat_ioctl = nbd_ioctl,
1267 };
1268
1269 #if IS_ENABLED(CONFIG_DEBUG_FS)
1270
1271 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1272 {
1273 struct nbd_device *nbd = s->private;
1274
1275 if (nbd->task_recv)
1276 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1277
1278 return 0;
1279 }
1280
1281 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1282 {
1283 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1284 }
1285
1286 static const struct file_operations nbd_dbg_tasks_ops = {
1287 .open = nbd_dbg_tasks_open,
1288 .read = seq_read,
1289 .llseek = seq_lseek,
1290 .release = single_release,
1291 };
1292
1293 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1294 {
1295 struct nbd_device *nbd = s->private;
1296 u32 flags = nbd->config->flags;
1297
1298 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1299
1300 seq_puts(s, "Known flags:\n");
1301
1302 if (flags & NBD_FLAG_HAS_FLAGS)
1303 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1304 if (flags & NBD_FLAG_READ_ONLY)
1305 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1306 if (flags & NBD_FLAG_SEND_FLUSH)
1307 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1308 if (flags & NBD_FLAG_SEND_TRIM)
1309 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1310
1311 return 0;
1312 }
1313
1314 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1315 {
1316 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1317 }
1318
1319 static const struct file_operations nbd_dbg_flags_ops = {
1320 .open = nbd_dbg_flags_open,
1321 .read = seq_read,
1322 .llseek = seq_lseek,
1323 .release = single_release,
1324 };
1325
1326 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1327 {
1328 struct dentry *dir;
1329 struct nbd_config *config = nbd->config;
1330
1331 if (!nbd_dbg_dir)
1332 return -EIO;
1333
1334 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1335 if (!dir) {
1336 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1337 nbd_name(nbd));
1338 return -EIO;
1339 }
1340 config->dbg_dir = dir;
1341
1342 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1343 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1344 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1345 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1346 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1347
1348 return 0;
1349 }
1350
1351 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1352 {
1353 debugfs_remove_recursive(nbd->config->dbg_dir);
1354 }
1355
1356 static int nbd_dbg_init(void)
1357 {
1358 struct dentry *dbg_dir;
1359
1360 dbg_dir = debugfs_create_dir("nbd", NULL);
1361 if (!dbg_dir)
1362 return -EIO;
1363
1364 nbd_dbg_dir = dbg_dir;
1365
1366 return 0;
1367 }
1368
1369 static void nbd_dbg_close(void)
1370 {
1371 debugfs_remove_recursive(nbd_dbg_dir);
1372 }
1373
1374 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1375
1376 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1377 {
1378 return 0;
1379 }
1380
1381 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1382 {
1383 }
1384
1385 static int nbd_dbg_init(void)
1386 {
1387 return 0;
1388 }
1389
1390 static void nbd_dbg_close(void)
1391 {
1392 }
1393
1394 #endif
1395
1396 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1397 unsigned int hctx_idx, unsigned int numa_node)
1398 {
1399 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1400 cmd->nbd = set->driver_data;
1401 return 0;
1402 }
1403
1404 static const struct blk_mq_ops nbd_mq_ops = {
1405 .queue_rq = nbd_queue_rq,
1406 .complete = nbd_complete_rq,
1407 .init_request = nbd_init_request,
1408 .timeout = nbd_xmit_timeout,
1409 };
1410
1411 static int nbd_dev_add(int index)
1412 {
1413 struct nbd_device *nbd;
1414 struct gendisk *disk;
1415 struct request_queue *q;
1416 int err = -ENOMEM;
1417
1418 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1419 if (!nbd)
1420 goto out;
1421
1422 disk = alloc_disk(1 << part_shift);
1423 if (!disk)
1424 goto out_free_nbd;
1425
1426 if (index >= 0) {
1427 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1428 GFP_KERNEL);
1429 if (err == -ENOSPC)
1430 err = -EEXIST;
1431 } else {
1432 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1433 if (err >= 0)
1434 index = err;
1435 }
1436 if (err < 0)
1437 goto out_free_disk;
1438
1439 nbd->index = index;
1440 nbd->disk = disk;
1441 nbd->tag_set.ops = &nbd_mq_ops;
1442 nbd->tag_set.nr_hw_queues = 1;
1443 nbd->tag_set.queue_depth = 128;
1444 nbd->tag_set.numa_node = NUMA_NO_NODE;
1445 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1446 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1447 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1448 nbd->tag_set.driver_data = nbd;
1449
1450 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1451 if (err)
1452 goto out_free_idr;
1453
1454 q = blk_mq_init_queue(&nbd->tag_set);
1455 if (IS_ERR(q)) {
1456 err = PTR_ERR(q);
1457 goto out_free_tags;
1458 }
1459 disk->queue = q;
1460
1461 /*
1462 * Tell the block layer that we are not a rotational device
1463 */
1464 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1465 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1466 disk->queue->limits.discard_granularity = 512;
1467 blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1468 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1469 blk_queue_max_segments(disk->queue, USHRT_MAX);
1470 blk_queue_max_hw_sectors(disk->queue, 65536);
1471 disk->queue->limits.max_sectors = 256;
1472
1473 mutex_init(&nbd->config_lock);
1474 refcount_set(&nbd->config_refs, 0);
1475 refcount_set(&nbd->refs, 1);
1476 INIT_LIST_HEAD(&nbd->list);
1477 disk->major = NBD_MAJOR;
1478 disk->first_minor = index << part_shift;
1479 disk->fops = &nbd_fops;
1480 disk->private_data = nbd;
1481 sprintf(disk->disk_name, "nbd%d", index);
1482 add_disk(disk);
1483 nbd_total_devices++;
1484 return index;
1485
1486 out_free_tags:
1487 blk_mq_free_tag_set(&nbd->tag_set);
1488 out_free_idr:
1489 idr_remove(&nbd_index_idr, index);
1490 out_free_disk:
1491 put_disk(disk);
1492 out_free_nbd:
1493 kfree(nbd);
1494 out:
1495 return err;
1496 }
1497
1498 static int find_free_cb(int id, void *ptr, void *data)
1499 {
1500 struct nbd_device *nbd = ptr;
1501 struct nbd_device **found = data;
1502
1503 if (!refcount_read(&nbd->config_refs)) {
1504 *found = nbd;
1505 return 1;
1506 }
1507 return 0;
1508 }
1509
1510 /* Netlink interface. */
1511 static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1512 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1513 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1514 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1515 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1516 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1517 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1518 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1519 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1520 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1521 };
1522
1523 static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1524 [NBD_SOCK_FD] = { .type = NLA_U32 },
1525 };
1526
1527 /* We don't use this right now since we don't parse the incoming list, but we
1528 * still want it here so userspace knows what to expect.
1529 */
1530 static struct nla_policy __attribute__((unused))
1531 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1532 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1533 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1534 };
1535
1536 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1537 {
1538 struct nbd_device *nbd = NULL;
1539 struct nbd_config *config;
1540 int index = -1;
1541 int ret;
1542 bool put_dev = false;
1543
1544 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1545 return -EPERM;
1546
1547 if (info->attrs[NBD_ATTR_INDEX])
1548 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1549 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1550 printk(KERN_ERR "nbd: must specify at least one socket\n");
1551 return -EINVAL;
1552 }
1553 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1554 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1555 return -EINVAL;
1556 }
1557 again:
1558 mutex_lock(&nbd_index_mutex);
1559 if (index == -1) {
1560 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1561 if (ret == 0) {
1562 int new_index;
1563 new_index = nbd_dev_add(-1);
1564 if (new_index < 0) {
1565 mutex_unlock(&nbd_index_mutex);
1566 printk(KERN_ERR "nbd: failed to add new device\n");
1567 return ret;
1568 }
1569 nbd = idr_find(&nbd_index_idr, new_index);
1570 }
1571 } else {
1572 nbd = idr_find(&nbd_index_idr, index);
1573 }
1574 if (!nbd) {
1575 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1576 index);
1577 mutex_unlock(&nbd_index_mutex);
1578 return -EINVAL;
1579 }
1580 if (!refcount_inc_not_zero(&nbd->refs)) {
1581 mutex_unlock(&nbd_index_mutex);
1582 if (index == -1)
1583 goto again;
1584 printk(KERN_ERR "nbd: device at index %d is going down\n",
1585 index);
1586 return -EINVAL;
1587 }
1588 mutex_unlock(&nbd_index_mutex);
1589
1590 mutex_lock(&nbd->config_lock);
1591 if (refcount_read(&nbd->config_refs)) {
1592 mutex_unlock(&nbd->config_lock);
1593 nbd_put(nbd);
1594 if (index == -1)
1595 goto again;
1596 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1597 return -EBUSY;
1598 }
1599 if (WARN_ON(nbd->config)) {
1600 mutex_unlock(&nbd->config_lock);
1601 nbd_put(nbd);
1602 return -EINVAL;
1603 }
1604 config = nbd->config = nbd_alloc_config();
1605 if (!nbd->config) {
1606 mutex_unlock(&nbd->config_lock);
1607 nbd_put(nbd);
1608 printk(KERN_ERR "nbd: couldn't allocate config\n");
1609 return -ENOMEM;
1610 }
1611 refcount_set(&nbd->config_refs, 1);
1612 set_bit(NBD_BOUND, &config->runtime_flags);
1613
1614 if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1615 u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1616 nbd_size_set(nbd, config->blksize,
1617 div64_u64(bytes, config->blksize));
1618 }
1619 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1620 u64 bsize =
1621 nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1622 nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1623 }
1624 if (info->attrs[NBD_ATTR_TIMEOUT]) {
1625 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1626 nbd->tag_set.timeout = timeout * HZ;
1627 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1628 }
1629 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1630 config->dead_conn_timeout =
1631 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1632 config->dead_conn_timeout *= HZ;
1633 }
1634 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1635 config->flags =
1636 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1637 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1638 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1639 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1640 set_bit(NBD_DESTROY_ON_DISCONNECT,
1641 &config->runtime_flags);
1642 put_dev = true;
1643 }
1644 }
1645
1646 if (info->attrs[NBD_ATTR_SOCKETS]) {
1647 struct nlattr *attr;
1648 int rem, fd;
1649
1650 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1651 rem) {
1652 struct nlattr *socks[NBD_SOCK_MAX+1];
1653
1654 if (nla_type(attr) != NBD_SOCK_ITEM) {
1655 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1656 ret = -EINVAL;
1657 goto out;
1658 }
1659 ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1660 nbd_sock_policy, info->extack);
1661 if (ret != 0) {
1662 printk(KERN_ERR "nbd: error processing sock list\n");
1663 ret = -EINVAL;
1664 goto out;
1665 }
1666 if (!socks[NBD_SOCK_FD])
1667 continue;
1668 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1669 ret = nbd_add_socket(nbd, fd, true);
1670 if (ret)
1671 goto out;
1672 }
1673 }
1674 ret = nbd_start_device(nbd);
1675 out:
1676 mutex_unlock(&nbd->config_lock);
1677 if (!ret) {
1678 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1679 refcount_inc(&nbd->config_refs);
1680 nbd_connect_reply(info, nbd->index);
1681 }
1682 nbd_config_put(nbd);
1683 if (put_dev)
1684 nbd_put(nbd);
1685 return ret;
1686 }
1687
1688 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1689 {
1690 struct nbd_device *nbd;
1691 int index;
1692
1693 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1694 return -EPERM;
1695
1696 if (!info->attrs[NBD_ATTR_INDEX]) {
1697 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1698 return -EINVAL;
1699 }
1700 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1701 mutex_lock(&nbd_index_mutex);
1702 nbd = idr_find(&nbd_index_idr, index);
1703 if (!nbd) {
1704 mutex_unlock(&nbd_index_mutex);
1705 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1706 index);
1707 return -EINVAL;
1708 }
1709 if (!refcount_inc_not_zero(&nbd->refs)) {
1710 mutex_unlock(&nbd_index_mutex);
1711 printk(KERN_ERR "nbd: device at index %d is going down\n",
1712 index);
1713 return -EINVAL;
1714 }
1715 mutex_unlock(&nbd_index_mutex);
1716 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1717 nbd_put(nbd);
1718 return 0;
1719 }
1720 mutex_lock(&nbd->config_lock);
1721 nbd_disconnect(nbd);
1722 mutex_unlock(&nbd->config_lock);
1723 if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1724 &nbd->config->runtime_flags))
1725 nbd_config_put(nbd);
1726 nbd_config_put(nbd);
1727 nbd_put(nbd);
1728 return 0;
1729 }
1730
1731 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1732 {
1733 struct nbd_device *nbd = NULL;
1734 struct nbd_config *config;
1735 int index;
1736 int ret = -EINVAL;
1737 bool put_dev = false;
1738
1739 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1740 return -EPERM;
1741
1742 if (!info->attrs[NBD_ATTR_INDEX]) {
1743 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1744 return -EINVAL;
1745 }
1746 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1747 mutex_lock(&nbd_index_mutex);
1748 nbd = idr_find(&nbd_index_idr, index);
1749 if (!nbd) {
1750 mutex_unlock(&nbd_index_mutex);
1751 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1752 index);
1753 return -EINVAL;
1754 }
1755 if (!refcount_inc_not_zero(&nbd->refs)) {
1756 mutex_unlock(&nbd_index_mutex);
1757 printk(KERN_ERR "nbd: device at index %d is going down\n",
1758 index);
1759 return -EINVAL;
1760 }
1761 mutex_unlock(&nbd_index_mutex);
1762
1763 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1764 dev_err(nbd_to_dev(nbd),
1765 "not configured, cannot reconfigure\n");
1766 nbd_put(nbd);
1767 return -EINVAL;
1768 }
1769
1770 mutex_lock(&nbd->config_lock);
1771 config = nbd->config;
1772 if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1773 !nbd->task_recv) {
1774 dev_err(nbd_to_dev(nbd),
1775 "not configured, cannot reconfigure\n");
1776 goto out;
1777 }
1778
1779 if (info->attrs[NBD_ATTR_TIMEOUT]) {
1780 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1781 nbd->tag_set.timeout = timeout * HZ;
1782 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1783 }
1784 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1785 config->dead_conn_timeout =
1786 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1787 config->dead_conn_timeout *= HZ;
1788 }
1789 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1790 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1791 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1792 if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1793 &config->runtime_flags))
1794 put_dev = true;
1795 } else {
1796 if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1797 &config->runtime_flags))
1798 refcount_inc(&nbd->refs);
1799 }
1800 }
1801
1802 if (info->attrs[NBD_ATTR_SOCKETS]) {
1803 struct nlattr *attr;
1804 int rem, fd;
1805
1806 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1807 rem) {
1808 struct nlattr *socks[NBD_SOCK_MAX+1];
1809
1810 if (nla_type(attr) != NBD_SOCK_ITEM) {
1811 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1812 ret = -EINVAL;
1813 goto out;
1814 }
1815 ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1816 nbd_sock_policy, info->extack);
1817 if (ret != 0) {
1818 printk(KERN_ERR "nbd: error processing sock list\n");
1819 ret = -EINVAL;
1820 goto out;
1821 }
1822 if (!socks[NBD_SOCK_FD])
1823 continue;
1824 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1825 ret = nbd_reconnect_socket(nbd, fd);
1826 if (ret) {
1827 if (ret == -ENOSPC)
1828 ret = 0;
1829 goto out;
1830 }
1831 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
1832 }
1833 }
1834 out:
1835 mutex_unlock(&nbd->config_lock);
1836 nbd_config_put(nbd);
1837 nbd_put(nbd);
1838 if (put_dev)
1839 nbd_put(nbd);
1840 return ret;
1841 }
1842
1843 static const struct genl_ops nbd_connect_genl_ops[] = {
1844 {
1845 .cmd = NBD_CMD_CONNECT,
1846 .policy = nbd_attr_policy,
1847 .doit = nbd_genl_connect,
1848 },
1849 {
1850 .cmd = NBD_CMD_DISCONNECT,
1851 .policy = nbd_attr_policy,
1852 .doit = nbd_genl_disconnect,
1853 },
1854 {
1855 .cmd = NBD_CMD_RECONFIGURE,
1856 .policy = nbd_attr_policy,
1857 .doit = nbd_genl_reconfigure,
1858 },
1859 {
1860 .cmd = NBD_CMD_STATUS,
1861 .policy = nbd_attr_policy,
1862 .doit = nbd_genl_status,
1863 },
1864 };
1865
1866 static const struct genl_multicast_group nbd_mcast_grps[] = {
1867 { .name = NBD_GENL_MCAST_GROUP_NAME, },
1868 };
1869
1870 static struct genl_family nbd_genl_family __ro_after_init = {
1871 .hdrsize = 0,
1872 .name = NBD_GENL_FAMILY_NAME,
1873 .version = NBD_GENL_VERSION,
1874 .module = THIS_MODULE,
1875 .ops = nbd_connect_genl_ops,
1876 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
1877 .maxattr = NBD_ATTR_MAX,
1878 .mcgrps = nbd_mcast_grps,
1879 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
1880 };
1881
1882 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
1883 {
1884 struct nlattr *dev_opt;
1885 u8 connected = 0;
1886 int ret;
1887
1888 /* This is a little racey, but for status it's ok. The
1889 * reason we don't take a ref here is because we can't
1890 * take a ref in the index == -1 case as we would need
1891 * to put under the nbd_index_mutex, which could
1892 * deadlock if we are configured to remove ourselves
1893 * once we're disconnected.
1894 */
1895 if (refcount_read(&nbd->config_refs))
1896 connected = 1;
1897 dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM);
1898 if (!dev_opt)
1899 return -EMSGSIZE;
1900 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
1901 if (ret)
1902 return -EMSGSIZE;
1903 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
1904 connected);
1905 if (ret)
1906 return -EMSGSIZE;
1907 nla_nest_end(reply, dev_opt);
1908 return 0;
1909 }
1910
1911 static int status_cb(int id, void *ptr, void *data)
1912 {
1913 struct nbd_device *nbd = ptr;
1914 return populate_nbd_status(nbd, (struct sk_buff *)data);
1915 }
1916
1917 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
1918 {
1919 struct nlattr *dev_list;
1920 struct sk_buff *reply;
1921 void *reply_head;
1922 size_t msg_size;
1923 int index = -1;
1924 int ret = -ENOMEM;
1925
1926 if (info->attrs[NBD_ATTR_INDEX])
1927 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1928
1929 mutex_lock(&nbd_index_mutex);
1930
1931 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
1932 nla_attr_size(sizeof(u8)));
1933 msg_size *= (index == -1) ? nbd_total_devices : 1;
1934
1935 reply = genlmsg_new(msg_size, GFP_KERNEL);
1936 if (!reply)
1937 goto out;
1938 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
1939 NBD_CMD_STATUS);
1940 if (!reply_head) {
1941 nlmsg_free(reply);
1942 goto out;
1943 }
1944
1945 dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST);
1946 if (index == -1) {
1947 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
1948 if (ret) {
1949 nlmsg_free(reply);
1950 goto out;
1951 }
1952 } else {
1953 struct nbd_device *nbd;
1954 nbd = idr_find(&nbd_index_idr, index);
1955 if (nbd) {
1956 ret = populate_nbd_status(nbd, reply);
1957 if (ret) {
1958 nlmsg_free(reply);
1959 goto out;
1960 }
1961 }
1962 }
1963 nla_nest_end(reply, dev_list);
1964 genlmsg_end(reply, reply_head);
1965 genlmsg_reply(reply, info);
1966 ret = 0;
1967 out:
1968 mutex_unlock(&nbd_index_mutex);
1969 return ret;
1970 }
1971
1972 static void nbd_connect_reply(struct genl_info *info, int index)
1973 {
1974 struct sk_buff *skb;
1975 void *msg_head;
1976 int ret;
1977
1978 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
1979 if (!skb)
1980 return;
1981 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
1982 NBD_CMD_CONNECT);
1983 if (!msg_head) {
1984 nlmsg_free(skb);
1985 return;
1986 }
1987 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
1988 if (ret) {
1989 nlmsg_free(skb);
1990 return;
1991 }
1992 genlmsg_end(skb, msg_head);
1993 genlmsg_reply(skb, info);
1994 }
1995
1996 static void nbd_mcast_index(int index)
1997 {
1998 struct sk_buff *skb;
1999 void *msg_head;
2000 int ret;
2001
2002 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2003 if (!skb)
2004 return;
2005 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2006 NBD_CMD_LINK_DEAD);
2007 if (!msg_head) {
2008 nlmsg_free(skb);
2009 return;
2010 }
2011 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2012 if (ret) {
2013 nlmsg_free(skb);
2014 return;
2015 }
2016 genlmsg_end(skb, msg_head);
2017 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2018 }
2019
2020 static void nbd_dead_link_work(struct work_struct *work)
2021 {
2022 struct link_dead_args *args = container_of(work, struct link_dead_args,
2023 work);
2024 nbd_mcast_index(args->index);
2025 kfree(args);
2026 }
2027
2028 static int __init nbd_init(void)
2029 {
2030 int i;
2031
2032 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2033
2034 if (max_part < 0) {
2035 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2036 return -EINVAL;
2037 }
2038
2039 part_shift = 0;
2040 if (max_part > 0) {
2041 part_shift = fls(max_part);
2042
2043 /*
2044 * Adjust max_part according to part_shift as it is exported
2045 * to user space so that user can know the max number of
2046 * partition kernel should be able to manage.
2047 *
2048 * Note that -1 is required because partition 0 is reserved
2049 * for the whole disk.
2050 */
2051 max_part = (1UL << part_shift) - 1;
2052 }
2053
2054 if ((1UL << part_shift) > DISK_MAX_PARTS)
2055 return -EINVAL;
2056
2057 if (nbds_max > 1UL << (MINORBITS - part_shift))
2058 return -EINVAL;
2059 recv_workqueue = alloc_workqueue("knbd-recv",
2060 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2061 if (!recv_workqueue)
2062 return -ENOMEM;
2063
2064 if (register_blkdev(NBD_MAJOR, "nbd")) {
2065 destroy_workqueue(recv_workqueue);
2066 return -EIO;
2067 }
2068
2069 if (genl_register_family(&nbd_genl_family)) {
2070 unregister_blkdev(NBD_MAJOR, "nbd");
2071 destroy_workqueue(recv_workqueue);
2072 return -EINVAL;
2073 }
2074 nbd_dbg_init();
2075
2076 mutex_lock(&nbd_index_mutex);
2077 for (i = 0; i < nbds_max; i++)
2078 nbd_dev_add(i);
2079 mutex_unlock(&nbd_index_mutex);
2080 return 0;
2081 }
2082
2083 static int nbd_exit_cb(int id, void *ptr, void *data)
2084 {
2085 struct list_head *list = (struct list_head *)data;
2086 struct nbd_device *nbd = ptr;
2087
2088 list_add_tail(&nbd->list, list);
2089 return 0;
2090 }
2091
2092 static void __exit nbd_cleanup(void)
2093 {
2094 struct nbd_device *nbd;
2095 LIST_HEAD(del_list);
2096
2097 nbd_dbg_close();
2098
2099 mutex_lock(&nbd_index_mutex);
2100 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2101 mutex_unlock(&nbd_index_mutex);
2102
2103 while (!list_empty(&del_list)) {
2104 nbd = list_first_entry(&del_list, struct nbd_device, list);
2105 list_del_init(&nbd->list);
2106 if (refcount_read(&nbd->refs) != 1)
2107 printk(KERN_ERR "nbd: possibly leaking a device\n");
2108 nbd_put(nbd);
2109 }
2110
2111 idr_destroy(&nbd_index_idr);
2112 genl_unregister_family(&nbd_genl_family);
2113 destroy_workqueue(recv_workqueue);
2114 unregister_blkdev(NBD_MAJOR, "nbd");
2115 }
2116
2117 module_init(nbd_init);
2118 module_exit(nbd_cleanup);
2119
2120 MODULE_DESCRIPTION("Network Block Device");
2121 MODULE_LICENSE("GPL");
2122
2123 module_param(nbds_max, int, 0444);
2124 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2125 module_param(max_part, int, 0444);
2126 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
ubuntu-bionic-kernel mirror