?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
net: DCB: Validate DCB_ATTR_DCB_BUFFER argument
[linux/fpc-iii.git] / drivers / block / nbd.c
blob7c577cabb9c3b1faba809a4a59ea5034858eaeae
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 void nbd_size_update(struct nbd_device *nbd)
300 {
301 struct nbd_config *config = nbd->config;
302 struct block_device *bdev = bdget_disk(nbd->disk, 0);
303
304 if (config->flags & NBD_FLAG_SEND_TRIM) {
305 nbd->disk->queue->limits.discard_granularity = config->blksize;
306 nbd->disk->queue->limits.discard_alignment = config->blksize;
307 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
308 }
309 blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
310 blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
311 set_capacity(nbd->disk, config->bytesize >> 9);
312 if (bdev) {
313 if (bdev->bd_disk) {
314 bd_set_size(bdev, config->bytesize);
315 set_blocksize(bdev, config->blksize);
316 } else
317 bdev->bd_invalidated = 1;
318 bdput(bdev);
319 }
320 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
321 }
322
323 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
324 loff_t nr_blocks)
325 {
326 struct nbd_config *config = nbd->config;
327 config->blksize = blocksize;
328 config->bytesize = blocksize * nr_blocks;
329 if (nbd->task_recv != NULL)
330 nbd_size_update(nbd);
331 }
332
333 static void nbd_complete_rq(struct request *req)
334 {
335 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
336
337 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
338 cmd->status ? "failed" : "done");
339
340 blk_mq_end_request(req, cmd->status);
341 }
342
343 /*
344 * Forcibly shutdown the socket causing all listeners to error
345 */
346 static void sock_shutdown(struct nbd_device *nbd)
347 {
348 struct nbd_config *config = nbd->config;
349 int i;
350
351 if (config->num_connections == 0)
352 return;
353 if (test_and_set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
354 return;
355
356 for (i = 0; i < config->num_connections; i++) {
357 struct nbd_sock *nsock = config->socks[i];
358 mutex_lock(&nsock->tx_lock);
359 nbd_mark_nsock_dead(nbd, nsock, 0);
360 mutex_unlock(&nsock->tx_lock);
361 }
362 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
363 }
364
365 static u32 req_to_nbd_cmd_type(struct request *req)
366 {
367 switch (req_op(req)) {
368 case REQ_OP_DISCARD:
369 return NBD_CMD_TRIM;
370 case REQ_OP_FLUSH:
371 return NBD_CMD_FLUSH;
372 case REQ_OP_WRITE:
373 return NBD_CMD_WRITE;
374 case REQ_OP_READ:
375 return NBD_CMD_READ;
376 default:
377 return U32_MAX;
378 }
379 }
380
381 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
382 bool reserved)
383 {
384 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
385 struct nbd_device *nbd = cmd->nbd;
386 struct nbd_config *config;
387
388 if (!mutex_trylock(&cmd->lock))
389 return BLK_EH_RESET_TIMER;
390
391 if (!refcount_inc_not_zero(&nbd->config_refs)) {
392 cmd->status = BLK_STS_TIMEOUT;
393 mutex_unlock(&cmd->lock);
394 goto done;
395 }
396 config = nbd->config;
397
398 if (config->num_connections > 1) {
399 dev_err_ratelimited(nbd_to_dev(nbd),
400 "Connection timed out, retrying (%d/%d alive)\n",
401 atomic_read(&config->live_connections),
402 config->num_connections);
403 /*
404 * Hooray we have more connections, requeue this IO, the submit
405 * path will put it on a real connection.
406 */
407 if (config->socks && config->num_connections > 1) {
408 if (cmd->index < config->num_connections) {
409 struct nbd_sock *nsock =
410 config->socks[cmd->index];
411 mutex_lock(&nsock->tx_lock);
412 /* We can have multiple outstanding requests, so
413 * we don't want to mark the nsock dead if we've
414 * already reconnected with a new socket, so
415 * only mark it dead if its the same socket we
416 * were sent out on.
417 */
418 if (cmd->cookie == nsock->cookie)
419 nbd_mark_nsock_dead(nbd, nsock, 1);
420 mutex_unlock(&nsock->tx_lock);
421 }
422 mutex_unlock(&cmd->lock);
423 nbd_requeue_cmd(cmd);
424 nbd_config_put(nbd);
425 return BLK_EH_DONE;
426 }
427 }
428
429 if (!nbd->tag_set.timeout) {
430 /*
431 * Userspace sets timeout=0 to disable socket disconnection,
432 * so just warn and reset the timer.
433 */
434 cmd->retries++;
435 dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
436 req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
437 (unsigned long long)blk_rq_pos(req) << 9,
438 blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
439
440 mutex_unlock(&cmd->lock);
441 nbd_config_put(nbd);
442 return BLK_EH_RESET_TIMER;
443 }
444
445 dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
446 set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
447 cmd->status = BLK_STS_IOERR;
448 mutex_unlock(&cmd->lock);
449 sock_shutdown(nbd);
450 nbd_config_put(nbd);
451 done:
452 blk_mq_complete_request(req);
453 return BLK_EH_DONE;
454 }
455
456 /*
457 * Send or receive packet.
458 */
459 static int sock_xmit(struct nbd_device *nbd, int index, int send,
460 struct iov_iter *iter, int msg_flags, int *sent)
461 {
462 struct nbd_config *config = nbd->config;
463 struct socket *sock = config->socks[index]->sock;
464 int result;
465 struct msghdr msg;
466 unsigned int noreclaim_flag;
467
468 if (unlikely(!sock)) {
469 dev_err_ratelimited(disk_to_dev(nbd->disk),
470 "Attempted %s on closed socket in sock_xmit\n",
471 (send ? "send" : "recv"));
472 return -EINVAL;
473 }
474
475 msg.msg_iter = *iter;
476
477 noreclaim_flag = memalloc_noreclaim_save();
478 do {
479 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
480 msg.msg_name = NULL;
481 msg.msg_namelen = 0;
482 msg.msg_control = NULL;
483 msg.msg_controllen = 0;
484 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
485
486 if (send)
487 result = sock_sendmsg(sock, &msg);
488 else
489 result = sock_recvmsg(sock, &msg, msg.msg_flags);
490
491 if (result <= 0) {
492 if (result == 0)
493 result = -EPIPE; /* short read */
494 break;
495 }
496 if (sent)
497 *sent += result;
498 } while (msg_data_left(&msg));
499
500 memalloc_noreclaim_restore(noreclaim_flag);
501
502 return result;
503 }
504
505 /*
506 * Different settings for sk->sk_sndtimeo can result in different return values
507 * if there is a signal pending when we enter sendmsg, because reasons?
508 */
509 static inline int was_interrupted(int result)
510 {
511 return result == -ERESTARTSYS || result == -EINTR;
512 }
513
514 /* always call with the tx_lock held */
515 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
516 {
517 struct request *req = blk_mq_rq_from_pdu(cmd);
518 struct nbd_config *config = nbd->config;
519 struct nbd_sock *nsock = config->socks[index];
520 int result;
521 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
522 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
523 struct iov_iter from;
524 unsigned long size = blk_rq_bytes(req);
525 struct bio *bio;
526 u64 handle;
527 u32 type;
528 u32 nbd_cmd_flags = 0;
529 int sent = nsock->sent, skip = 0;
530
531 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
532
533 type = req_to_nbd_cmd_type(req);
534 if (type == U32_MAX)
535 return -EIO;
536
537 if (rq_data_dir(req) == WRITE &&
538 (config->flags & NBD_FLAG_READ_ONLY)) {
539 dev_err_ratelimited(disk_to_dev(nbd->disk),
540 "Write on read-only\n");
541 return -EIO;
542 }
543
544 if (req->cmd_flags & REQ_FUA)
545 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
546
547 /* We did a partial send previously, and we at least sent the whole
548 * request struct, so just go and send the rest of the pages in the
549 * request.
550 */
551 if (sent) {
552 if (sent >= sizeof(request)) {
553 skip = sent - sizeof(request);
554
555 /* initialize handle for tracing purposes */
556 handle = nbd_cmd_handle(cmd);
557
558 goto send_pages;
559 }
560 iov_iter_advance(&from, sent);
561 } else {
562 cmd->cmd_cookie++;
563 }
564 cmd->index = index;
565 cmd->cookie = nsock->cookie;
566 cmd->retries = 0;
567 request.type = htonl(type | nbd_cmd_flags);
568 if (type != NBD_CMD_FLUSH) {
569 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
570 request.len = htonl(size);
571 }
572 handle = nbd_cmd_handle(cmd);
573 memcpy(request.handle, &handle, sizeof(handle));
574
575 trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
576
577 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
578 req, nbdcmd_to_ascii(type),
579 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
580 result = sock_xmit(nbd, index, 1, &from,
581 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
582 trace_nbd_header_sent(req, handle);
583 if (result <= 0) {
584 if (was_interrupted(result)) {
585 /* If we havne't sent anything we can just return BUSY,
586 * however if we have sent something we need to make
587 * sure we only allow this req to be sent until we are
588 * completely done.
589 */
590 if (sent) {
591 nsock->pending = req;
592 nsock->sent = sent;
593 }
594 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
595 return BLK_STS_RESOURCE;
596 }
597 dev_err_ratelimited(disk_to_dev(nbd->disk),
598 "Send control failed (result %d)\n", result);
599 return -EAGAIN;
600 }
601 send_pages:
602 if (type != NBD_CMD_WRITE)
603 goto out;
604
605 bio = req->bio;
606 while (bio) {
607 struct bio *next = bio->bi_next;
608 struct bvec_iter iter;
609 struct bio_vec bvec;
610
611 bio_for_each_segment(bvec, bio, iter) {
612 bool is_last = !next && bio_iter_last(bvec, iter);
613 int flags = is_last ? 0 : MSG_MORE;
614
615 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
616 req, bvec.bv_len);
617 iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
618 if (skip) {
619 if (skip >= iov_iter_count(&from)) {
620 skip -= iov_iter_count(&from);
621 continue;
622 }
623 iov_iter_advance(&from, skip);
624 skip = 0;
625 }
626 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
627 if (result <= 0) {
628 if (was_interrupted(result)) {
629 /* We've already sent the header, we
630 * have no choice but to set pending and
631 * return BUSY.
632 */
633 nsock->pending = req;
634 nsock->sent = sent;
635 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
636 return BLK_STS_RESOURCE;
637 }
638 dev_err(disk_to_dev(nbd->disk),
639 "Send data failed (result %d)\n",
640 result);
641 return -EAGAIN;
642 }
643 /*
644 * The completion might already have come in,
645 * so break for the last one instead of letting
646 * the iterator do it. This prevents use-after-free
647 * of the bio.
648 */
649 if (is_last)
650 break;
651 }
652 bio = next;
653 }
654 out:
655 trace_nbd_payload_sent(req, handle);
656 nsock->pending = NULL;
657 nsock->sent = 0;
658 return 0;
659 }
660
661 /* NULL returned = something went wrong, inform userspace */
662 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
663 {
664 struct nbd_config *config = nbd->config;
665 int result;
666 struct nbd_reply reply;
667 struct nbd_cmd *cmd;
668 struct request *req = NULL;
669 u64 handle;
670 u16 hwq;
671 u32 tag;
672 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
673 struct iov_iter to;
674 int ret = 0;
675
676 reply.magic = 0;
677 iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
678 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
679 if (result <= 0) {
680 if (!nbd_disconnected(config))
681 dev_err(disk_to_dev(nbd->disk),
682 "Receive control failed (result %d)\n", result);
683 return ERR_PTR(result);
684 }
685
686 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
687 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
688 (unsigned long)ntohl(reply.magic));
689 return ERR_PTR(-EPROTO);
690 }
691
692 memcpy(&handle, reply.handle, sizeof(handle));
693 tag = nbd_handle_to_tag(handle);
694 hwq = blk_mq_unique_tag_to_hwq(tag);
695 if (hwq < nbd->tag_set.nr_hw_queues)
696 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
697 blk_mq_unique_tag_to_tag(tag));
698 if (!req || !blk_mq_request_started(req)) {
699 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
700 tag, req);
701 return ERR_PTR(-ENOENT);
702 }
703 trace_nbd_header_received(req, handle);
704 cmd = blk_mq_rq_to_pdu(req);
705
706 mutex_lock(&cmd->lock);
707 if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
708 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
709 req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
710 ret = -ENOENT;
711 goto out;
712 }
713 if (cmd->status != BLK_STS_OK) {
714 dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
715 req);
716 ret = -ENOENT;
717 goto out;
718 }
719 if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
720 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
721 req);
722 ret = -ENOENT;
723 goto out;
724 }
725 if (ntohl(reply.error)) {
726 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
727 ntohl(reply.error));
728 cmd->status = BLK_STS_IOERR;
729 goto out;
730 }
731
732 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
733 if (rq_data_dir(req) != WRITE) {
734 struct req_iterator iter;
735 struct bio_vec bvec;
736
737 rq_for_each_segment(bvec, req, iter) {
738 iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
739 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
740 if (result <= 0) {
741 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
742 result);
743 /*
744 * If we've disconnected or we only have 1
745 * connection then we need to make sure we
746 * complete this request, otherwise error out
747 * and let the timeout stuff handle resubmitting
748 * this request onto another connection.
749 */
750 if (nbd_disconnected(config) ||
751 config->num_connections <= 1) {
752 cmd->status = BLK_STS_IOERR;
753 goto out;
754 }
755 ret = -EIO;
756 goto out;
757 }
758 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
759 req, bvec.bv_len);
760 }
761 }
762 out:
763 trace_nbd_payload_received(req, handle);
764 mutex_unlock(&cmd->lock);
765 return ret ? ERR_PTR(ret) : cmd;
766 }
767
768 static void recv_work(struct work_struct *work)
769 {
770 struct recv_thread_args *args = container_of(work,
771 struct recv_thread_args,
772 work);
773 struct nbd_device *nbd = args->nbd;
774 struct nbd_config *config = nbd->config;
775 struct nbd_cmd *cmd;
776
777 while (1) {
778 cmd = nbd_read_stat(nbd, args->index);
779 if (IS_ERR(cmd)) {
780 struct nbd_sock *nsock = config->socks[args->index];
781
782 mutex_lock(&nsock->tx_lock);
783 nbd_mark_nsock_dead(nbd, nsock, 1);
784 mutex_unlock(&nsock->tx_lock);
785 break;
786 }
787
788 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
789 }
790 atomic_dec(&config->recv_threads);
791 wake_up(&config->recv_wq);
792 nbd_config_put(nbd);
793 kfree(args);
794 }
795
796 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
797 {
798 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
799
800 mutex_lock(&cmd->lock);
801 cmd->status = BLK_STS_IOERR;
802 mutex_unlock(&cmd->lock);
803
804 blk_mq_complete_request(req);
805 return true;
806 }
807
808 static void nbd_clear_que(struct nbd_device *nbd)
809 {
810 blk_mq_quiesce_queue(nbd->disk->queue);
811 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
812 blk_mq_unquiesce_queue(nbd->disk->queue);
813 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
814 }
815
816 static int find_fallback(struct nbd_device *nbd, int index)
817 {
818 struct nbd_config *config = nbd->config;
819 int new_index = -1;
820 struct nbd_sock *nsock = config->socks[index];
821 int fallback = nsock->fallback_index;
822
823 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
824 return new_index;
825
826 if (config->num_connections <= 1) {
827 dev_err_ratelimited(disk_to_dev(nbd->disk),
828 "Attempted send on invalid socket\n");
829 return new_index;
830 }
831
832 if (fallback >= 0 && fallback < config->num_connections &&
833 !config->socks[fallback]->dead)
834 return fallback;
835
836 if (nsock->fallback_index < 0 ||
837 nsock->fallback_index >= config->num_connections ||
838 config->socks[nsock->fallback_index]->dead) {
839 int i;
840 for (i = 0; i < config->num_connections; i++) {
841 if (i == index)
842 continue;
843 if (!config->socks[i]->dead) {
844 new_index = i;
845 break;
846 }
847 }
848 nsock->fallback_index = new_index;
849 if (new_index < 0) {
850 dev_err_ratelimited(disk_to_dev(nbd->disk),
851 "Dead connection, failed to find a fallback\n");
852 return new_index;
853 }
854 }
855 new_index = nsock->fallback_index;
856 return new_index;
857 }
858
859 static int wait_for_reconnect(struct nbd_device *nbd)
860 {
861 struct nbd_config *config = nbd->config;
862 if (!config->dead_conn_timeout)
863 return 0;
864 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
865 return 0;
866 return wait_event_timeout(config->conn_wait,
867 atomic_read(&config->live_connections) > 0,
868 config->dead_conn_timeout) > 0;
869 }
870
871 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
872 {
873 struct request *req = blk_mq_rq_from_pdu(cmd);
874 struct nbd_device *nbd = cmd->nbd;
875 struct nbd_config *config;
876 struct nbd_sock *nsock;
877 int ret;
878
879 if (!refcount_inc_not_zero(&nbd->config_refs)) {
880 dev_err_ratelimited(disk_to_dev(nbd->disk),
881 "Socks array is empty\n");
882 blk_mq_start_request(req);
883 return -EINVAL;
884 }
885 config = nbd->config;
886
887 if (index >= config->num_connections) {
888 dev_err_ratelimited(disk_to_dev(nbd->disk),
889 "Attempted send on invalid socket\n");
890 nbd_config_put(nbd);
891 blk_mq_start_request(req);
892 return -EINVAL;
893 }
894 cmd->status = BLK_STS_OK;
895 again:
896 nsock = config->socks[index];
897 mutex_lock(&nsock->tx_lock);
898 if (nsock->dead) {
899 int old_index = index;
900 index = find_fallback(nbd, index);
901 mutex_unlock(&nsock->tx_lock);
902 if (index < 0) {
903 if (wait_for_reconnect(nbd)) {
904 index = old_index;
905 goto again;
906 }
907 /* All the sockets should already be down at this point,
908 * we just want to make sure that DISCONNECTED is set so
909 * any requests that come in that were queue'ed waiting
910 * for the reconnect timer don't trigger the timer again
911 * and instead just error out.
912 */
913 sock_shutdown(nbd);
914 nbd_config_put(nbd);
915 blk_mq_start_request(req);
916 return -EIO;
917 }
918 goto again;
919 }
920
921 /* Handle the case that we have a pending request that was partially
922 * transmitted that _has_ to be serviced first. We need to call requeue
923 * here so that it gets put _after_ the request that is already on the
924 * dispatch list.
925 */
926 blk_mq_start_request(req);
927 if (unlikely(nsock->pending && nsock->pending != req)) {
928 nbd_requeue_cmd(cmd);
929 ret = 0;
930 goto out;
931 }
932 /*
933 * Some failures are related to the link going down, so anything that
934 * returns EAGAIN can be retried on a different socket.
935 */
936 ret = nbd_send_cmd(nbd, cmd, index);
937 if (ret == -EAGAIN) {
938 dev_err_ratelimited(disk_to_dev(nbd->disk),
939 "Request send failed, requeueing\n");
940 nbd_mark_nsock_dead(nbd, nsock, 1);
941 nbd_requeue_cmd(cmd);
942 ret = 0;
943 }
944 out:
945 mutex_unlock(&nsock->tx_lock);
946 nbd_config_put(nbd);
947 return ret;
948 }
949
950 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
951 const struct blk_mq_queue_data *bd)
952 {
953 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
954 int ret;
955
956 /*
957 * Since we look at the bio's to send the request over the network we
958 * need to make sure the completion work doesn't mark this request done
959 * before we are done doing our send. This keeps us from dereferencing
960 * freed data if we have particularly fast completions (ie we get the
961 * completion before we exit sock_xmit on the last bvec) or in the case
962 * that the server is misbehaving (or there was an error) before we're
963 * done sending everything over the wire.
964 */
965 mutex_lock(&cmd->lock);
966 clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
967
968 /* We can be called directly from the user space process, which means we
969 * could possibly have signals pending so our sendmsg will fail. In
970 * this case we need to return that we are busy, otherwise error out as
971 * appropriate.
972 */
973 ret = nbd_handle_cmd(cmd, hctx->queue_num);
974 if (ret < 0)
975 ret = BLK_STS_IOERR;
976 else if (!ret)
977 ret = BLK_STS_OK;
978 mutex_unlock(&cmd->lock);
979
980 return ret;
981 }
982
983 static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
984 int *err)
985 {
986 struct socket *sock;
987
988 *err = 0;
989 sock = sockfd_lookup(fd, err);
990 if (!sock)
991 return NULL;
992
993 if (sock->ops->shutdown == sock_no_shutdown) {
994 dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
995 *err = -EINVAL;
996 sockfd_put(sock);
997 return NULL;
998 }
999
1000 return sock;
1001 }
1002
1003 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1004 bool netlink)
1005 {
1006 struct nbd_config *config = nbd->config;
1007 struct socket *sock;
1008 struct nbd_sock **socks;
1009 struct nbd_sock *nsock;
1010 int err;
1011
1012 sock = nbd_get_socket(nbd, arg, &err);
1013 if (!sock)
1014 return err;
1015
1016 if (!netlink && !nbd->task_setup &&
1017 !test_bit(NBD_RT_BOUND, &config->runtime_flags))
1018 nbd->task_setup = current;
1019
1020 if (!netlink &&
1021 (nbd->task_setup != current ||
1022 test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1023 dev_err(disk_to_dev(nbd->disk),
1024 "Device being setup by another task");
1025 err = -EBUSY;
1026 goto put_socket;
1027 }
1028
1029 nsock = kzalloc(sizeof(*nsock), GFP_KERNEL);
1030 if (!nsock) {
1031 err = -ENOMEM;
1032 goto put_socket;
1033 }
1034
1035 socks = krealloc(config->socks, (config->num_connections + 1) *
1036 sizeof(struct nbd_sock *), GFP_KERNEL);
1037 if (!socks) {
1038 kfree(nsock);
1039 err = -ENOMEM;
1040 goto put_socket;
1041 }
1042
1043 config->socks = socks;
1044
1045 nsock->fallback_index = -1;
1046 nsock->dead = false;
1047 mutex_init(&nsock->tx_lock);
1048 nsock->sock = sock;
1049 nsock->pending = NULL;
1050 nsock->sent = 0;
1051 nsock->cookie = 0;
1052 socks[config->num_connections++] = nsock;
1053 atomic_inc(&config->live_connections);
1054
1055 return 0;
1056
1057 put_socket:
1058 sockfd_put(sock);
1059 return err;
1060 }
1061
1062 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1063 {
1064 struct nbd_config *config = nbd->config;
1065 struct socket *sock, *old;
1066 struct recv_thread_args *args;
1067 int i;
1068 int err;
1069
1070 sock = nbd_get_socket(nbd, arg, &err);
1071 if (!sock)
1072 return err;
1073
1074 args = kzalloc(sizeof(*args), GFP_KERNEL);
1075 if (!args) {
1076 sockfd_put(sock);
1077 return -ENOMEM;
1078 }
1079
1080 for (i = 0; i < config->num_connections; i++) {
1081 struct nbd_sock *nsock = config->socks[i];
1082
1083 if (!nsock->dead)
1084 continue;
1085
1086 mutex_lock(&nsock->tx_lock);
1087 if (!nsock->dead) {
1088 mutex_unlock(&nsock->tx_lock);
1089 continue;
1090 }
1091 sk_set_memalloc(sock->sk);
1092 if (nbd->tag_set.timeout)
1093 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1094 atomic_inc(&config->recv_threads);
1095 refcount_inc(&nbd->config_refs);
1096 old = nsock->sock;
1097 nsock->fallback_index = -1;
1098 nsock->sock = sock;
1099 nsock->dead = false;
1100 INIT_WORK(&args->work, recv_work);
1101 args->index = i;
1102 args->nbd = nbd;
1103 nsock->cookie++;
1104 mutex_unlock(&nsock->tx_lock);
1105 sockfd_put(old);
1106
1107 clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1108
1109 /* We take the tx_mutex in an error path in the recv_work, so we
1110 * need to queue_work outside of the tx_mutex.
1111 */
1112 queue_work(nbd->recv_workq, &args->work);
1113
1114 atomic_inc(&config->live_connections);
1115 wake_up(&config->conn_wait);
1116 return 0;
1117 }
1118 sockfd_put(sock);
1119 kfree(args);
1120 return -ENOSPC;
1121 }
1122
1123 static void nbd_bdev_reset(struct block_device *bdev)
1124 {
1125 if (bdev->bd_openers > 1)
1126 return;
1127 bd_set_size(bdev, 0);
1128 }
1129
1130 static void nbd_parse_flags(struct nbd_device *nbd)
1131 {
1132 struct nbd_config *config = nbd->config;
1133 if (config->flags & NBD_FLAG_READ_ONLY)
1134 set_disk_ro(nbd->disk, true);
1135 else
1136 set_disk_ro(nbd->disk, false);
1137 if (config->flags & NBD_FLAG_SEND_TRIM)
1138 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1139 if (config->flags & NBD_FLAG_SEND_FLUSH) {
1140 if (config->flags & NBD_FLAG_SEND_FUA)
1141 blk_queue_write_cache(nbd->disk->queue, true, true);
1142 else
1143 blk_queue_write_cache(nbd->disk->queue, true, false);
1144 }
1145 else
1146 blk_queue_write_cache(nbd->disk->queue, false, false);
1147 }
1148
1149 static void send_disconnects(struct nbd_device *nbd)
1150 {
1151 struct nbd_config *config = nbd->config;
1152 struct nbd_request request = {
1153 .magic = htonl(NBD_REQUEST_MAGIC),
1154 .type = htonl(NBD_CMD_DISC),
1155 };
1156 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1157 struct iov_iter from;
1158 int i, ret;
1159
1160 for (i = 0; i < config->num_connections; i++) {
1161 struct nbd_sock *nsock = config->socks[i];
1162
1163 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1164 mutex_lock(&nsock->tx_lock);
1165 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1166 if (ret <= 0)
1167 dev_err(disk_to_dev(nbd->disk),
1168 "Send disconnect failed %d\n", ret);
1169 mutex_unlock(&nsock->tx_lock);
1170 }
1171 }
1172
1173 static int nbd_disconnect(struct nbd_device *nbd)
1174 {
1175 struct nbd_config *config = nbd->config;
1176
1177 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1178 set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1179 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1180 send_disconnects(nbd);
1181 return 0;
1182 }
1183
1184 static void nbd_clear_sock(struct nbd_device *nbd)
1185 {
1186 sock_shutdown(nbd);
1187 nbd_clear_que(nbd);
1188 nbd->task_setup = NULL;
1189 }
1190
1191 static void nbd_config_put(struct nbd_device *nbd)
1192 {
1193 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1194 &nbd->config_lock)) {
1195 struct nbd_config *config = nbd->config;
1196 nbd_dev_dbg_close(nbd);
1197 nbd_size_clear(nbd);
1198 if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1199 &config->runtime_flags))
1200 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1201 nbd->task_recv = NULL;
1202 nbd_clear_sock(nbd);
1203 if (config->num_connections) {
1204 int i;
1205 for (i = 0; i < config->num_connections; i++) {
1206 sockfd_put(config->socks[i]->sock);
1207 kfree(config->socks[i]);
1208 }
1209 kfree(config->socks);
1210 }
1211 kfree(nbd->config);
1212 nbd->config = NULL;
1213
1214 if (nbd->recv_workq)
1215 destroy_workqueue(nbd->recv_workq);
1216 nbd->recv_workq = NULL;
1217
1218 nbd->tag_set.timeout = 0;
1219 nbd->disk->queue->limits.discard_granularity = 0;
1220 nbd->disk->queue->limits.discard_alignment = 0;
1221 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1222 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1223
1224 mutex_unlock(&nbd->config_lock);
1225 nbd_put(nbd);
1226 module_put(THIS_MODULE);
1227 }
1228 }
1229
1230 static int nbd_start_device(struct nbd_device *nbd)
1231 {
1232 struct nbd_config *config = nbd->config;
1233 int num_connections = config->num_connections;
1234 int error = 0, i;
1235
1236 if (nbd->task_recv)
1237 return -EBUSY;
1238 if (!config->socks)
1239 return -EINVAL;
1240 if (num_connections > 1 &&
1241 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1242 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1243 return -EINVAL;
1244 }
1245
1246 nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1247 WQ_MEM_RECLAIM | WQ_HIGHPRI |
1248 WQ_UNBOUND, 0, nbd->index);
1249 if (!nbd->recv_workq) {
1250 dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1251 return -ENOMEM;
1252 }
1253
1254 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1255 nbd->task_recv = current;
1256
1257 nbd_parse_flags(nbd);
1258
1259 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1260 if (error) {
1261 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1262 return error;
1263 }
1264 set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1265
1266 nbd_dev_dbg_init(nbd);
1267 for (i = 0; i < num_connections; i++) {
1268 struct recv_thread_args *args;
1269
1270 args = kzalloc(sizeof(*args), GFP_KERNEL);
1271 if (!args) {
1272 sock_shutdown(nbd);
1273 /*
1274 * If num_connections is m (2 < m),
1275 * and NO.1 ~ NO.n(1 < n < m) kzallocs are successful.
1276 * But NO.(n + 1) failed. We still have n recv threads.
1277 * So, add flush_workqueue here to prevent recv threads
1278 * dropping the last config_refs and trying to destroy
1279 * the workqueue from inside the workqueue.
1280 */
1281 if (i)
1282 flush_workqueue(nbd->recv_workq);
1283 return -ENOMEM;
1284 }
1285 sk_set_memalloc(config->socks[i]->sock->sk);
1286 if (nbd->tag_set.timeout)
1287 config->socks[i]->sock->sk->sk_sndtimeo =
1288 nbd->tag_set.timeout;
1289 atomic_inc(&config->recv_threads);
1290 refcount_inc(&nbd->config_refs);
1291 INIT_WORK(&args->work, recv_work);
1292 args->nbd = nbd;
1293 args->index = i;
1294 queue_work(nbd->recv_workq, &args->work);
1295 }
1296 nbd_size_update(nbd);
1297 return error;
1298 }
1299
1300 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1301 {
1302 struct nbd_config *config = nbd->config;
1303 int ret;
1304
1305 ret = nbd_start_device(nbd);
1306 if (ret)
1307 return ret;
1308
1309 if (max_part)
1310 bdev->bd_invalidated = 1;
1311 mutex_unlock(&nbd->config_lock);
1312 ret = wait_event_interruptible(config->recv_wq,
1313 atomic_read(&config->recv_threads) == 0);
1314 if (ret)
1315 sock_shutdown(nbd);
1316 flush_workqueue(nbd->recv_workq);
1317
1318 mutex_lock(&nbd->config_lock);
1319 nbd_bdev_reset(bdev);
1320 /* user requested, ignore socket errors */
1321 if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1322 ret = 0;
1323 if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1324 ret = -ETIMEDOUT;
1325 return ret;
1326 }
1327
1328 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1329 struct block_device *bdev)
1330 {
1331 sock_shutdown(nbd);
1332 __invalidate_device(bdev, true);
1333 nbd_bdev_reset(bdev);
1334 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1335 &nbd->config->runtime_flags))
1336 nbd_config_put(nbd);
1337 }
1338
1339 static bool nbd_is_valid_blksize(unsigned long blksize)
1340 {
1341 if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1342 blksize > PAGE_SIZE)
1343 return false;
1344 return true;
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 if (!arg)
1372 arg = NBD_DEF_BLKSIZE;
1373 if (!nbd_is_valid_blksize(arg))
1374 return -EINVAL;
1375 nbd_size_set(nbd, arg,
1376 div_s64(config->bytesize, arg));
1377 return 0;
1378 case NBD_SET_SIZE:
1379 nbd_size_set(nbd, config->blksize,
1380 div_s64(arg, config->blksize));
1381 return 0;
1382 case NBD_SET_SIZE_BLOCKS:
1383 nbd_size_set(nbd, config->blksize, arg);
1384 return 0;
1385 case NBD_SET_TIMEOUT:
1386 nbd_set_cmd_timeout(nbd, arg);
1387 return 0;
1388
1389 case NBD_SET_FLAGS:
1390 config->flags = arg;
1391 return 0;
1392 case NBD_DO_IT:
1393 return nbd_start_device_ioctl(nbd, bdev);
1394 case NBD_CLEAR_QUE:
1395 /*
1396 * This is for compatibility only. The queue is always cleared
1397 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1398 */
1399 return 0;
1400 case NBD_PRINT_DEBUG:
1401 /*
1402 * For compatibility only, we no longer keep a list of
1403 * outstanding requests.
1404 */
1405 return 0;
1406 }
1407 return -ENOTTY;
1408 }
1409
1410 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1411 unsigned int cmd, unsigned long arg)
1412 {
1413 struct nbd_device *nbd = bdev->bd_disk->private_data;
1414 struct nbd_config *config = nbd->config;
1415 int error = -EINVAL;
1416
1417 if (!capable(CAP_SYS_ADMIN))
1418 return -EPERM;
1419
1420 /* The block layer will pass back some non-nbd ioctls in case we have
1421 * special handling for them, but we don't so just return an error.
1422 */
1423 if (_IOC_TYPE(cmd) != 0xab)
1424 return -EINVAL;
1425
1426 mutex_lock(&nbd->config_lock);
1427
1428 /* Don't allow ioctl operations on a nbd device that was created with
1429 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1430 */
1431 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1432 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1433 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1434 else
1435 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1436 mutex_unlock(&nbd->config_lock);
1437 return error;
1438 }
1439
1440 static struct nbd_config *nbd_alloc_config(void)
1441 {
1442 struct nbd_config *config;
1443
1444 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1445 if (!config)
1446 return NULL;
1447 atomic_set(&config->recv_threads, 0);
1448 init_waitqueue_head(&config->recv_wq);
1449 init_waitqueue_head(&config->conn_wait);
1450 config->blksize = NBD_DEF_BLKSIZE;
1451 atomic_set(&config->live_connections, 0);
1452 try_module_get(THIS_MODULE);
1453 return config;
1454 }
1455
1456 static int nbd_open(struct block_device *bdev, fmode_t mode)
1457 {
1458 struct nbd_device *nbd;
1459 int ret = 0;
1460
1461 mutex_lock(&nbd_index_mutex);
1462 nbd = bdev->bd_disk->private_data;
1463 if (!nbd) {
1464 ret = -ENXIO;
1465 goto out;
1466 }
1467 if (!refcount_inc_not_zero(&nbd->refs)) {
1468 ret = -ENXIO;
1469 goto out;
1470 }
1471 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1472 struct nbd_config *config;
1473
1474 mutex_lock(&nbd->config_lock);
1475 if (refcount_inc_not_zero(&nbd->config_refs)) {
1476 mutex_unlock(&nbd->config_lock);
1477 goto out;
1478 }
1479 config = nbd->config = nbd_alloc_config();
1480 if (!config) {
1481 ret = -ENOMEM;
1482 mutex_unlock(&nbd->config_lock);
1483 goto out;
1484 }
1485 refcount_set(&nbd->config_refs, 1);
1486 refcount_inc(&nbd->refs);
1487 mutex_unlock(&nbd->config_lock);
1488 bdev->bd_invalidated = 1;
1489 } else if (nbd_disconnected(nbd->config)) {
1490 bdev->bd_invalidated = 1;
1491 }
1492 out:
1493 mutex_unlock(&nbd_index_mutex);
1494 return ret;
1495 }
1496
1497 static void nbd_release(struct gendisk *disk, fmode_t mode)
1498 {
1499 struct nbd_device *nbd = disk->private_data;
1500 struct block_device *bdev = bdget_disk(disk, 0);
1501
1502 if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1503 bdev->bd_openers == 0)
1504 nbd_disconnect_and_put(nbd);
1505
1506 nbd_config_put(nbd);
1507 nbd_put(nbd);
1508 }
1509
1510 static const struct block_device_operations nbd_fops =
1511 {
1512 .owner = THIS_MODULE,
1513 .open = nbd_open,
1514 .release = nbd_release,
1515 .ioctl = nbd_ioctl,
1516 .compat_ioctl = nbd_ioctl,
1517 };
1518
1519 #if IS_ENABLED(CONFIG_DEBUG_FS)
1520
1521 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1522 {
1523 struct nbd_device *nbd = s->private;
1524
1525 if (nbd->task_recv)
1526 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1527
1528 return 0;
1529 }
1530
1531 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1532 {
1533 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1534 }
1535
1536 static const struct file_operations nbd_dbg_tasks_ops = {
1537 .open = nbd_dbg_tasks_open,
1538 .read = seq_read,
1539 .llseek = seq_lseek,
1540 .release = single_release,
1541 };
1542
1543 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1544 {
1545 struct nbd_device *nbd = s->private;
1546 u32 flags = nbd->config->flags;
1547
1548 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1549
1550 seq_puts(s, "Known flags:\n");
1551
1552 if (flags & NBD_FLAG_HAS_FLAGS)
1553 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1554 if (flags & NBD_FLAG_READ_ONLY)
1555 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1556 if (flags & NBD_FLAG_SEND_FLUSH)
1557 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1558 if (flags & NBD_FLAG_SEND_FUA)
1559 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1560 if (flags & NBD_FLAG_SEND_TRIM)
1561 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1562
1563 return 0;
1564 }
1565
1566 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1567 {
1568 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1569 }
1570
1571 static const struct file_operations nbd_dbg_flags_ops = {
1572 .open = nbd_dbg_flags_open,
1573 .read = seq_read,
1574 .llseek = seq_lseek,
1575 .release = single_release,
1576 };
1577
1578 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1579 {
1580 struct dentry *dir;
1581 struct nbd_config *config = nbd->config;
1582
1583 if (!nbd_dbg_dir)
1584 return -EIO;
1585
1586 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1587 if (!dir) {
1588 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1589 nbd_name(nbd));
1590 return -EIO;
1591 }
1592 config->dbg_dir = dir;
1593
1594 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1595 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1596 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1597 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1598 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1599
1600 return 0;
1601 }
1602
1603 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1604 {
1605 debugfs_remove_recursive(nbd->config->dbg_dir);
1606 }
1607
1608 static int nbd_dbg_init(void)
1609 {
1610 struct dentry *dbg_dir;
1611
1612 dbg_dir = debugfs_create_dir("nbd", NULL);
1613 if (!dbg_dir)
1614 return -EIO;
1615
1616 nbd_dbg_dir = dbg_dir;
1617
1618 return 0;
1619 }
1620
1621 static void nbd_dbg_close(void)
1622 {
1623 debugfs_remove_recursive(nbd_dbg_dir);
1624 }
1625
1626 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1627
1628 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1629 {
1630 return 0;
1631 }
1632
1633 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1634 {
1635 }
1636
1637 static int nbd_dbg_init(void)
1638 {
1639 return 0;
1640 }
1641
1642 static void nbd_dbg_close(void)
1643 {
1644 }
1645
1646 #endif
1647
1648 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1649 unsigned int hctx_idx, unsigned int numa_node)
1650 {
1651 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1652 cmd->nbd = set->driver_data;
1653 cmd->flags = 0;
1654 mutex_init(&cmd->lock);
1655 return 0;
1656 }
1657
1658 static const struct blk_mq_ops nbd_mq_ops = {
1659 .queue_rq = nbd_queue_rq,
1660 .complete = nbd_complete_rq,
1661 .init_request = nbd_init_request,
1662 .timeout = nbd_xmit_timeout,
1663 };
1664
1665 static int nbd_dev_add(int index)
1666 {
1667 struct nbd_device *nbd;
1668 struct gendisk *disk;
1669 struct request_queue *q;
1670 int err = -ENOMEM;
1671
1672 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1673 if (!nbd)
1674 goto out;
1675
1676 disk = alloc_disk(1 << part_shift);
1677 if (!disk)
1678 goto out_free_nbd;
1679
1680 if (index >= 0) {
1681 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1682 GFP_KERNEL);
1683 if (err == -ENOSPC)
1684 err = -EEXIST;
1685 } else {
1686 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1687 if (err >= 0)
1688 index = err;
1689 }
1690 if (err < 0)
1691 goto out_free_disk;
1692
1693 nbd->index = index;
1694 nbd->disk = disk;
1695 nbd->tag_set.ops = &nbd_mq_ops;
1696 nbd->tag_set.nr_hw_queues = 1;
1697 nbd->tag_set.queue_depth = 128;
1698 nbd->tag_set.numa_node = NUMA_NO_NODE;
1699 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1700 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1701 BLK_MQ_F_BLOCKING;
1702 nbd->tag_set.driver_data = nbd;
1703 nbd->destroy_complete = NULL;
1704
1705 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1706 if (err)
1707 goto out_free_idr;
1708
1709 q = blk_mq_init_queue(&nbd->tag_set);
1710 if (IS_ERR(q)) {
1711 err = PTR_ERR(q);
1712 goto out_free_tags;
1713 }
1714 disk->queue = q;
1715
1716 /*
1717 * Tell the block layer that we are not a rotational device
1718 */
1719 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1720 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1721 disk->queue->limits.discard_granularity = 0;
1722 disk->queue->limits.discard_alignment = 0;
1723 blk_queue_max_discard_sectors(disk->queue, 0);
1724 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1725 blk_queue_max_segments(disk->queue, USHRT_MAX);
1726 blk_queue_max_hw_sectors(disk->queue, 65536);
1727 disk->queue->limits.max_sectors = 256;
1728
1729 mutex_init(&nbd->config_lock);
1730 refcount_set(&nbd->config_refs, 0);
1731 refcount_set(&nbd->refs, 1);
1732 INIT_LIST_HEAD(&nbd->list);
1733 disk->major = NBD_MAJOR;
1734 disk->first_minor = index << part_shift;
1735 disk->fops = &nbd_fops;
1736 disk->private_data = nbd;
1737 sprintf(disk->disk_name, "nbd%d", index);
1738 add_disk(disk);
1739 nbd_total_devices++;
1740 return index;
1741
1742 out_free_tags:
1743 blk_mq_free_tag_set(&nbd->tag_set);
1744 out_free_idr:
1745 idr_remove(&nbd_index_idr, index);
1746 out_free_disk:
1747 put_disk(disk);
1748 out_free_nbd:
1749 kfree(nbd);
1750 out:
1751 return err;
1752 }
1753
1754 static int find_free_cb(int id, void *ptr, void *data)
1755 {
1756 struct nbd_device *nbd = ptr;
1757 struct nbd_device **found = data;
1758
1759 if (!refcount_read(&nbd->config_refs)) {
1760 *found = nbd;
1761 return 1;
1762 }
1763 return 0;
1764 }
1765
1766 /* Netlink interface. */
1767 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1768 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1769 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1770 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1771 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1772 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1773 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1774 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1775 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1776 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1777 };
1778
1779 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1780 [NBD_SOCK_FD] = { .type = NLA_U32 },
1781 };
1782
1783 /* We don't use this right now since we don't parse the incoming list, but we
1784 * still want it here so userspace knows what to expect.
1785 */
1786 static const struct nla_policy __attribute__((unused))
1787 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1788 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1789 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1790 };
1791
1792 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1793 {
1794 struct nbd_config *config = nbd->config;
1795 u64 bsize = config->blksize;
1796 u64 bytes = config->bytesize;
1797
1798 if (info->attrs[NBD_ATTR_SIZE_BYTES])
1799 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1800
1801 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1802 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1803 if (!bsize)
1804 bsize = NBD_DEF_BLKSIZE;
1805 if (!nbd_is_valid_blksize(bsize)) {
1806 printk(KERN_ERR "Invalid block size %llu\n", bsize);
1807 return -EINVAL;
1808 }
1809 }
1810
1811 if (bytes != config->bytesize || bsize != config->blksize)
1812 nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1813 return 0;
1814 }
1815
1816 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1817 {
1818 DECLARE_COMPLETION_ONSTACK(destroy_complete);
1819 struct nbd_device *nbd = NULL;
1820 struct nbd_config *config;
1821 int index = -1;
1822 int ret;
1823 bool put_dev = false;
1824
1825 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1826 return -EPERM;
1827
1828 if (info->attrs[NBD_ATTR_INDEX])
1829 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1830 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1831 printk(KERN_ERR "nbd: must specify at least one socket\n");
1832 return -EINVAL;
1833 }
1834 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1835 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1836 return -EINVAL;
1837 }
1838 again:
1839 mutex_lock(&nbd_index_mutex);
1840 if (index == -1) {
1841 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1842 if (ret == 0) {
1843 int new_index;
1844 new_index = nbd_dev_add(-1);
1845 if (new_index < 0) {
1846 mutex_unlock(&nbd_index_mutex);
1847 printk(KERN_ERR "nbd: failed to add new device\n");
1848 return new_index;
1849 }
1850 nbd = idr_find(&nbd_index_idr, new_index);
1851 }
1852 } else {
1853 nbd = idr_find(&nbd_index_idr, index);
1854 if (!nbd) {
1855 ret = nbd_dev_add(index);
1856 if (ret < 0) {
1857 mutex_unlock(&nbd_index_mutex);
1858 printk(KERN_ERR "nbd: failed to add new device\n");
1859 return ret;
1860 }
1861 nbd = idr_find(&nbd_index_idr, index);
1862 }
1863 }
1864 if (!nbd) {
1865 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1866 index);
1867 mutex_unlock(&nbd_index_mutex);
1868 return -EINVAL;
1869 }
1870
1871 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1872 test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1873 nbd->destroy_complete = &destroy_complete;
1874 mutex_unlock(&nbd_index_mutex);
1875
1876 /* Wait untill the the nbd stuff is totally destroyed */
1877 wait_for_completion(&destroy_complete);
1878 goto again;
1879 }
1880
1881 if (!refcount_inc_not_zero(&nbd->refs)) {
1882 mutex_unlock(&nbd_index_mutex);
1883 if (index == -1)
1884 goto again;
1885 printk(KERN_ERR "nbd: device at index %d is going down\n",
1886 index);
1887 return -EINVAL;
1888 }
1889 mutex_unlock(&nbd_index_mutex);
1890
1891 mutex_lock(&nbd->config_lock);
1892 if (refcount_read(&nbd->config_refs)) {
1893 mutex_unlock(&nbd->config_lock);
1894 nbd_put(nbd);
1895 if (index == -1)
1896 goto again;
1897 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1898 return -EBUSY;
1899 }
1900 if (WARN_ON(nbd->config)) {
1901 mutex_unlock(&nbd->config_lock);
1902 nbd_put(nbd);
1903 return -EINVAL;
1904 }
1905 config = nbd->config = nbd_alloc_config();
1906 if (!nbd->config) {
1907 mutex_unlock(&nbd->config_lock);
1908 nbd_put(nbd);
1909 printk(KERN_ERR "nbd: couldn't allocate config\n");
1910 return -ENOMEM;
1911 }
1912 refcount_set(&nbd->config_refs, 1);
1913 set_bit(NBD_RT_BOUND, &config->runtime_flags);
1914
1915 ret = nbd_genl_size_set(info, nbd);
1916 if (ret)
1917 goto out;
1918
1919 if (info->attrs[NBD_ATTR_TIMEOUT])
1920 nbd_set_cmd_timeout(nbd,
1921 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1922 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1923 config->dead_conn_timeout =
1924 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1925 config->dead_conn_timeout *= HZ;
1926 }
1927 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1928 config->flags =
1929 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1930 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1931 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1932 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1933 set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1934 &config->runtime_flags);
1935 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1936 put_dev = true;
1937 } else {
1938 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1939 }
1940 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1941 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1942 &config->runtime_flags);
1943 }
1944 }
1945
1946 if (info->attrs[NBD_ATTR_SOCKETS]) {
1947 struct nlattr *attr;
1948 int rem, fd;
1949
1950 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1951 rem) {
1952 struct nlattr *socks[NBD_SOCK_MAX+1];
1953
1954 if (nla_type(attr) != NBD_SOCK_ITEM) {
1955 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1956 ret = -EINVAL;
1957 goto out;
1958 }
1959 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1960 attr,
1961 nbd_sock_policy,
1962 info->extack);
1963 if (ret != 0) {
1964 printk(KERN_ERR "nbd: error processing sock list\n");
1965 ret = -EINVAL;
1966 goto out;
1967 }
1968 if (!socks[NBD_SOCK_FD])
1969 continue;
1970 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1971 ret = nbd_add_socket(nbd, fd, true);
1972 if (ret)
1973 goto out;
1974 }
1975 }
1976 ret = nbd_start_device(nbd);
1977 out:
1978 mutex_unlock(&nbd->config_lock);
1979 if (!ret) {
1980 set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1981 refcount_inc(&nbd->config_refs);
1982 nbd_connect_reply(info, nbd->index);
1983 }
1984 nbd_config_put(nbd);
1985 if (put_dev)
1986 nbd_put(nbd);
1987 return ret;
1988 }
1989
1990 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1991 {
1992 mutex_lock(&nbd->config_lock);
1993 nbd_disconnect(nbd);
1994 nbd_clear_sock(nbd);
1995 mutex_unlock(&nbd->config_lock);
1996 /*
1997 * Make sure recv thread has finished, so it does not drop the last
1998 * config ref and try to destroy the workqueue from inside the work
1999 * queue.
2000 */
2001 flush_workqueue(nbd->recv_workq);
2002 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
2003 &nbd->config->runtime_flags))
2004 nbd_config_put(nbd);
2005 }
2006
2007 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
2008 {
2009 struct nbd_device *nbd;
2010 int index;
2011
2012 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2013 return -EPERM;
2014
2015 if (!info->attrs[NBD_ATTR_INDEX]) {
2016 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
2017 return -EINVAL;
2018 }
2019 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2020 mutex_lock(&nbd_index_mutex);
2021 nbd = idr_find(&nbd_index_idr, index);
2022 if (!nbd) {
2023 mutex_unlock(&nbd_index_mutex);
2024 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2025 index);
2026 return -EINVAL;
2027 }
2028 if (!refcount_inc_not_zero(&nbd->refs)) {
2029 mutex_unlock(&nbd_index_mutex);
2030 printk(KERN_ERR "nbd: device at index %d is going down\n",
2031 index);
2032 return -EINVAL;
2033 }
2034 mutex_unlock(&nbd_index_mutex);
2035 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2036 nbd_put(nbd);
2037 return 0;
2038 }
2039 nbd_disconnect_and_put(nbd);
2040 nbd_config_put(nbd);
2041 nbd_put(nbd);
2042 return 0;
2043 }
2044
2045 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2046 {
2047 struct nbd_device *nbd = NULL;
2048 struct nbd_config *config;
2049 int index;
2050 int ret = 0;
2051 bool put_dev = false;
2052
2053 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2054 return -EPERM;
2055
2056 if (!info->attrs[NBD_ATTR_INDEX]) {
2057 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2058 return -EINVAL;
2059 }
2060 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2061 mutex_lock(&nbd_index_mutex);
2062 nbd = idr_find(&nbd_index_idr, index);
2063 if (!nbd) {
2064 mutex_unlock(&nbd_index_mutex);
2065 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2066 index);
2067 return -EINVAL;
2068 }
2069 if (!refcount_inc_not_zero(&nbd->refs)) {
2070 mutex_unlock(&nbd_index_mutex);
2071 printk(KERN_ERR "nbd: device at index %d is going down\n",
2072 index);
2073 return -EINVAL;
2074 }
2075 mutex_unlock(&nbd_index_mutex);
2076
2077 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2078 dev_err(nbd_to_dev(nbd),
2079 "not configured, cannot reconfigure\n");
2080 nbd_put(nbd);
2081 return -EINVAL;
2082 }
2083
2084 mutex_lock(&nbd->config_lock);
2085 config = nbd->config;
2086 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2087 !nbd->task_recv) {
2088 dev_err(nbd_to_dev(nbd),
2089 "not configured, cannot reconfigure\n");
2090 ret = -EINVAL;
2091 goto out;
2092 }
2093
2094 ret = nbd_genl_size_set(info, nbd);
2095 if (ret)
2096 goto out;
2097
2098 if (info->attrs[NBD_ATTR_TIMEOUT])
2099 nbd_set_cmd_timeout(nbd,
2100 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2101 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2102 config->dead_conn_timeout =
2103 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2104 config->dead_conn_timeout *= HZ;
2105 }
2106 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2107 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2108 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2109 if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2110 &config->runtime_flags))
2111 put_dev = true;
2112 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2113 } else {
2114 if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2115 &config->runtime_flags))
2116 refcount_inc(&nbd->refs);
2117 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2118 }
2119
2120 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2121 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2122 &config->runtime_flags);
2123 } else {
2124 clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2125 &config->runtime_flags);
2126 }
2127 }
2128
2129 if (info->attrs[NBD_ATTR_SOCKETS]) {
2130 struct nlattr *attr;
2131 int rem, fd;
2132
2133 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2134 rem) {
2135 struct nlattr *socks[NBD_SOCK_MAX+1];
2136
2137 if (nla_type(attr) != NBD_SOCK_ITEM) {
2138 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2139 ret = -EINVAL;
2140 goto out;
2141 }
2142 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2143 attr,
2144 nbd_sock_policy,
2145 info->extack);
2146 if (ret != 0) {
2147 printk(KERN_ERR "nbd: error processing sock list\n");
2148 ret = -EINVAL;
2149 goto out;
2150 }
2151 if (!socks[NBD_SOCK_FD])
2152 continue;
2153 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2154 ret = nbd_reconnect_socket(nbd, fd);
2155 if (ret) {
2156 if (ret == -ENOSPC)
2157 ret = 0;
2158 goto out;
2159 }
2160 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2161 }
2162 }
2163 out:
2164 mutex_unlock(&nbd->config_lock);
2165 nbd_config_put(nbd);
2166 nbd_put(nbd);
2167 if (put_dev)
2168 nbd_put(nbd);
2169 return ret;
2170 }
2171
2172 static const struct genl_ops nbd_connect_genl_ops[] = {
2173 {
2174 .cmd = NBD_CMD_CONNECT,
2175 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2176 .doit = nbd_genl_connect,
2177 },
2178 {
2179 .cmd = NBD_CMD_DISCONNECT,
2180 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2181 .doit = nbd_genl_disconnect,
2182 },
2183 {
2184 .cmd = NBD_CMD_RECONFIGURE,
2185 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2186 .doit = nbd_genl_reconfigure,
2187 },
2188 {
2189 .cmd = NBD_CMD_STATUS,
2190 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2191 .doit = nbd_genl_status,
2192 },
2193 };
2194
2195 static const struct genl_multicast_group nbd_mcast_grps[] = {
2196 { .name = NBD_GENL_MCAST_GROUP_NAME, },
2197 };
2198
2199 static struct genl_family nbd_genl_family __ro_after_init = {
2200 .hdrsize = 0,
2201 .name = NBD_GENL_FAMILY_NAME,
2202 .version = NBD_GENL_VERSION,
2203 .module = THIS_MODULE,
2204 .ops = nbd_connect_genl_ops,
2205 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2206 .maxattr = NBD_ATTR_MAX,
2207 .policy = nbd_attr_policy,
2208 .mcgrps = nbd_mcast_grps,
2209 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2210 };
2211
2212 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2213 {
2214 struct nlattr *dev_opt;
2215 u8 connected = 0;
2216 int ret;
2217
2218 /* This is a little racey, but for status it's ok. The
2219 * reason we don't take a ref here is because we can't
2220 * take a ref in the index == -1 case as we would need
2221 * to put under the nbd_index_mutex, which could
2222 * deadlock if we are configured to remove ourselves
2223 * once we're disconnected.
2224 */
2225 if (refcount_read(&nbd->config_refs))
2226 connected = 1;
2227 dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2228 if (!dev_opt)
2229 return -EMSGSIZE;
2230 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2231 if (ret)
2232 return -EMSGSIZE;
2233 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2234 connected);
2235 if (ret)
2236 return -EMSGSIZE;
2237 nla_nest_end(reply, dev_opt);
2238 return 0;
2239 }
2240
2241 static int status_cb(int id, void *ptr, void *data)
2242 {
2243 struct nbd_device *nbd = ptr;
2244 return populate_nbd_status(nbd, (struct sk_buff *)data);
2245 }
2246
2247 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2248 {
2249 struct nlattr *dev_list;
2250 struct sk_buff *reply;
2251 void *reply_head;
2252 size_t msg_size;
2253 int index = -1;
2254 int ret = -ENOMEM;
2255
2256 if (info->attrs[NBD_ATTR_INDEX])
2257 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2258
2259 mutex_lock(&nbd_index_mutex);
2260
2261 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2262 nla_attr_size(sizeof(u8)));
2263 msg_size *= (index == -1) ? nbd_total_devices : 1;
2264
2265 reply = genlmsg_new(msg_size, GFP_KERNEL);
2266 if (!reply)
2267 goto out;
2268 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2269 NBD_CMD_STATUS);
2270 if (!reply_head) {
2271 nlmsg_free(reply);
2272 goto out;
2273 }
2274
2275 dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2276 if (index == -1) {
2277 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2278 if (ret) {
2279 nlmsg_free(reply);
2280 goto out;
2281 }
2282 } else {
2283 struct nbd_device *nbd;
2284 nbd = idr_find(&nbd_index_idr, index);
2285 if (nbd) {
2286 ret = populate_nbd_status(nbd, reply);
2287 if (ret) {
2288 nlmsg_free(reply);
2289 goto out;
2290 }
2291 }
2292 }
2293 nla_nest_end(reply, dev_list);
2294 genlmsg_end(reply, reply_head);
2295 ret = genlmsg_reply(reply, info);
2296 out:
2297 mutex_unlock(&nbd_index_mutex);
2298 return ret;
2299 }
2300
2301 static void nbd_connect_reply(struct genl_info *info, int index)
2302 {
2303 struct sk_buff *skb;
2304 void *msg_head;
2305 int ret;
2306
2307 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2308 if (!skb)
2309 return;
2310 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2311 NBD_CMD_CONNECT);
2312 if (!msg_head) {
2313 nlmsg_free(skb);
2314 return;
2315 }
2316 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2317 if (ret) {
2318 nlmsg_free(skb);
2319 return;
2320 }
2321 genlmsg_end(skb, msg_head);
2322 genlmsg_reply(skb, info);
2323 }
2324
2325 static void nbd_mcast_index(int index)
2326 {
2327 struct sk_buff *skb;
2328 void *msg_head;
2329 int ret;
2330
2331 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2332 if (!skb)
2333 return;
2334 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2335 NBD_CMD_LINK_DEAD);
2336 if (!msg_head) {
2337 nlmsg_free(skb);
2338 return;
2339 }
2340 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2341 if (ret) {
2342 nlmsg_free(skb);
2343 return;
2344 }
2345 genlmsg_end(skb, msg_head);
2346 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2347 }
2348
2349 static void nbd_dead_link_work(struct work_struct *work)
2350 {
2351 struct link_dead_args *args = container_of(work, struct link_dead_args,
2352 work);
2353 nbd_mcast_index(args->index);
2354 kfree(args);
2355 }
2356
2357 static int __init nbd_init(void)
2358 {
2359 int i;
2360
2361 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2362
2363 if (max_part < 0) {
2364 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2365 return -EINVAL;
2366 }
2367
2368 part_shift = 0;
2369 if (max_part > 0) {
2370 part_shift = fls(max_part);
2371
2372 /*
2373 * Adjust max_part according to part_shift as it is exported
2374 * to user space so that user can know the max number of
2375 * partition kernel should be able to manage.
2376 *
2377 * Note that -1 is required because partition 0 is reserved
2378 * for the whole disk.
2379 */
2380 max_part = (1UL << part_shift) - 1;
2381 }
2382
2383 if ((1UL << part_shift) > DISK_MAX_PARTS)
2384 return -EINVAL;
2385
2386 if (nbds_max > 1UL << (MINORBITS - part_shift))
2387 return -EINVAL;
2388
2389 if (register_blkdev(NBD_MAJOR, "nbd"))
2390 return -EIO;
2391
2392 if (genl_register_family(&nbd_genl_family)) {
2393 unregister_blkdev(NBD_MAJOR, "nbd");
2394 return -EINVAL;
2395 }
2396 nbd_dbg_init();
2397
2398 mutex_lock(&nbd_index_mutex);
2399 for (i = 0; i < nbds_max; i++)
2400 nbd_dev_add(i);
2401 mutex_unlock(&nbd_index_mutex);
2402 return 0;
2403 }
2404
2405 static int nbd_exit_cb(int id, void *ptr, void *data)
2406 {
2407 struct list_head *list = (struct list_head *)data;
2408 struct nbd_device *nbd = ptr;
2409
2410 list_add_tail(&nbd->list, list);
2411 return 0;
2412 }
2413
2414 static void __exit nbd_cleanup(void)
2415 {
2416 struct nbd_device *nbd;
2417 LIST_HEAD(del_list);
2418
2419 nbd_dbg_close();
2420
2421 mutex_lock(&nbd_index_mutex);
2422 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2423 mutex_unlock(&nbd_index_mutex);
2424
2425 while (!list_empty(&del_list)) {
2426 nbd = list_first_entry(&del_list, struct nbd_device, list);
2427 list_del_init(&nbd->list);
2428 if (refcount_read(&nbd->refs) != 1)
2429 printk(KERN_ERR "nbd: possibly leaking a device\n");
2430 nbd_put(nbd);
2431 }
2432
2433 idr_destroy(&nbd_index_idr);
2434 genl_unregister_family(&nbd_genl_family);
2435 unregister_blkdev(NBD_MAJOR, "nbd");
2436 }
2437
2438 module_init(nbd_init);
2439 module_exit(nbd_cleanup);
2440
2441 MODULE_DESCRIPTION("Network Block Device");
2442 MODULE_LICENSE("GPL");
2443
2444 module_param(nbds_max, int, 0444);
2445 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2446 module_param(max_part, int, 0444);
2447 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");
Linux with added FPC-III support