1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Network block device - make block devices work over TCP
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.
8 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
9 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
11 * (part of code stolen from loop.c)
14 #include <linux/major.h>
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>
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>
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>
40 #include <linux/uaccess.h>
41 #include <asm/types.h>
43 #include <linux/nbd.h>
44 #include <linux/nbd-netlink.h>
45 #include <net/genetlink.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/nbd.h>
50 static DEFINE_IDR(nbd_index_idr);
51 static DEFINE_MUTEX(nbd_index_mutex);
52 static int nbd_total_devices = 0;
57 struct request *pending;
64 struct recv_thread_args {
65 struct work_struct work;
66 struct nbd_device *nbd;
70 struct link_dead_args {
71 struct work_struct work;
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
84 #define NBD_DESTROY_ON_DISCONNECT 0
85 #define NBD_DISCONNECT_REQUESTED 1
89 unsigned long runtime_flags;
90 u64 dead_conn_timeout;
92 struct nbd_sock **socks;
94 atomic_t live_connections;
95 wait_queue_head_t conn_wait;
97 atomic_t recv_threads;
98 wait_queue_head_t recv_wq;
101 #if IS_ENABLED(CONFIG_DEBUG_FS)
102 struct dentry *dbg_dir;
107 struct blk_mq_tag_set tag_set;
110 refcount_t config_refs;
112 struct nbd_config *config;
113 struct mutex config_lock;
114 struct gendisk *disk;
115 struct workqueue_struct *recv_workq;
117 struct list_head list;
118 struct task_struct *task_recv;
119 struct task_struct *task_setup;
121 struct completion *destroy_complete;
125 #define NBD_CMD_REQUEUED 1
128 struct nbd_device *nbd;
138 #if IS_ENABLED(CONFIG_DEBUG_FS)
139 static struct dentry *nbd_dbg_dir;
142 #define nbd_name(nbd) ((nbd)->disk->disk_name)
144 #define NBD_MAGIC 0x68797548
146 #define NBD_DEF_BLKSIZE 1024
148 static unsigned int nbds_max = 16;
149 static int max_part = 16;
150 static int part_shift;
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);
160 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
162 return disk_to_dev(nbd->disk);
165 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
167 struct request *req = blk_mq_rq_from_pdu(cmd);
169 if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
170 blk_mq_requeue_request(req, true);
173 #define NBD_COOKIE_BITS 32
175 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
177 struct request *req = blk_mq_rq_from_pdu(cmd);
178 u32 tag = blk_mq_unique_tag(req);
179 u64 cookie = cmd->cmd_cookie;
181 return (cookie << NBD_COOKIE_BITS) | tag;
184 static u32 nbd_handle_to_tag(u64 handle)
189 static u32 nbd_handle_to_cookie(u64 handle)
191 return (u32)(handle >> NBD_COOKIE_BITS);
194 static const char *nbdcmd_to_ascii(int 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";
206 static ssize_t pid_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
209 struct gendisk *disk = dev_to_disk(dev);
210 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
212 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
215 static const struct device_attribute pid_attr = {
216 .attr = { .name = "pid", .mode = 0444},
220 static void nbd_dev_remove(struct nbd_device *nbd)
222 struct gendisk *disk = nbd->disk;
223 struct request_queue *q;
228 blk_cleanup_queue(q);
229 blk_mq_free_tag_set(&nbd->tag_set);
230 disk->private_data = NULL;
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
240 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) && nbd->destroy_complete)
241 complete(nbd->destroy_complete);
246 static void nbd_put(struct nbd_device *nbd)
248 if (refcount_dec_and_mutex_lock(&nbd->refs,
250 idr_remove(&nbd_index_idr, nbd->index);
252 mutex_unlock(&nbd_index_mutex);
256 static int nbd_disconnected(struct nbd_config *config)
258 return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
259 test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
262 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
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);
269 INIT_WORK(&args->work, nbd_dead_link_work);
270 args->index = nbd->index;
271 queue_work(system_wq, &args->work);
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");
287 nsock->pending = NULL;
291 static void nbd_size_clear(struct nbd_device *nbd)
293 if (nbd->config->bytesize) {
294 set_capacity(nbd->disk, 0);
295 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
299 static void nbd_size_update(struct nbd_device *nbd)
301 struct nbd_config *config = nbd->config;
302 struct block_device *bdev = bdget_disk(nbd->disk, 0);
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);
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);
314 bd_set_size(bdev, config->bytesize);
315 set_blocksize(bdev, config->blksize);
317 bdev->bd_invalidated = 1;
320 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
323 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
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);
333 static void nbd_complete_rq(struct request *req)
335 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
337 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
338 cmd->status ? "failed" : "done");
340 blk_mq_end_request(req, cmd->status);
344 * Forcibly shutdown the socket causing all listeners to error
346 static void sock_shutdown(struct nbd_device *nbd)
348 struct nbd_config *config = nbd->config;
351 if (config->num_connections == 0)
353 if (test_and_set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
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);
362 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
365 static u32 req_to_nbd_cmd_type(struct request *req)
367 switch (req_op(req)) {
371 return NBD_CMD_FLUSH;
373 return NBD_CMD_WRITE;
381 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
384 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
385 struct nbd_device *nbd = cmd->nbd;
386 struct nbd_config *config;
388 if (!mutex_trylock(&cmd->lock))
389 return BLK_EH_RESET_TIMER;
391 if (!refcount_inc_not_zero(&nbd->config_refs)) {
392 cmd->status = BLK_STS_TIMEOUT;
393 mutex_unlock(&cmd->lock);
396 config = nbd->config;
398 if (config->num_connections > 1 ||
399 (config->num_connections == 1 && nbd->tag_set.timeout)) {
400 dev_err_ratelimited(nbd_to_dev(nbd),
401 "Connection timed out, retrying (%d/%d alive)\n",
402 atomic_read(&config->live_connections),
403 config->num_connections);
405 * Hooray we have more connections, requeue this IO, the submit
406 * path will put it on a real connection. Or if only one
407 * connection is configured, the submit path will wait util
408 * a new connection is reconfigured or util dead timeout.
411 if (cmd->index < config->num_connections) {
412 struct nbd_sock *nsock =
413 config->socks[cmd->index];
414 mutex_lock(&nsock->tx_lock);
415 /* We can have multiple outstanding requests, so
416 * we don't want to mark the nsock dead if we've
417 * already reconnected with a new socket, so
418 * only mark it dead if its the same socket we
421 if (cmd->cookie == nsock->cookie)
422 nbd_mark_nsock_dead(nbd, nsock, 1);
423 mutex_unlock(&nsock->tx_lock);
425 mutex_unlock(&cmd->lock);
426 nbd_requeue_cmd(cmd);
432 if (!nbd->tag_set.timeout) {
434 * Userspace sets timeout=0 to disable socket disconnection,
435 * so just warn and reset the timer.
437 struct nbd_sock *nsock = config->socks[cmd->index];
439 dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
440 req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
441 (unsigned long long)blk_rq_pos(req) << 9,
442 blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
444 mutex_lock(&nsock->tx_lock);
445 if (cmd->cookie != nsock->cookie) {
446 nbd_requeue_cmd(cmd);
447 mutex_unlock(&nsock->tx_lock);
448 mutex_unlock(&cmd->lock);
452 mutex_unlock(&nsock->tx_lock);
453 mutex_unlock(&cmd->lock);
455 return BLK_EH_RESET_TIMER;
458 dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
459 set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
460 cmd->status = BLK_STS_IOERR;
461 mutex_unlock(&cmd->lock);
465 blk_mq_complete_request(req);
470 * Send or receive packet.
472 static int sock_xmit(struct nbd_device *nbd, int index, int send,
473 struct iov_iter *iter, int msg_flags, int *sent)
475 struct nbd_config *config = nbd->config;
476 struct socket *sock = config->socks[index]->sock;
479 unsigned int noreclaim_flag;
481 if (unlikely(!sock)) {
482 dev_err_ratelimited(disk_to_dev(nbd->disk),
483 "Attempted %s on closed socket in sock_xmit\n",
484 (send ? "send" : "recv"));
488 msg.msg_iter = *iter;
490 noreclaim_flag = memalloc_noreclaim_save();
492 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
495 msg.msg_control = NULL;
496 msg.msg_controllen = 0;
497 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
500 result = sock_sendmsg(sock, &msg);
502 result = sock_recvmsg(sock, &msg, msg.msg_flags);
506 result = -EPIPE; /* short read */
511 } while (msg_data_left(&msg));
513 memalloc_noreclaim_restore(noreclaim_flag);
519 * Different settings for sk->sk_sndtimeo can result in different return values
520 * if there is a signal pending when we enter sendmsg, because reasons?
522 static inline int was_interrupted(int result)
524 return result == -ERESTARTSYS || result == -EINTR;
527 /* always call with the tx_lock held */
528 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
530 struct request *req = blk_mq_rq_from_pdu(cmd);
531 struct nbd_config *config = nbd->config;
532 struct nbd_sock *nsock = config->socks[index];
534 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
535 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
536 struct iov_iter from;
537 unsigned long size = blk_rq_bytes(req);
541 u32 nbd_cmd_flags = 0;
542 int sent = nsock->sent, skip = 0;
544 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
546 type = req_to_nbd_cmd_type(req);
550 if (rq_data_dir(req) == WRITE &&
551 (config->flags & NBD_FLAG_READ_ONLY)) {
552 dev_err_ratelimited(disk_to_dev(nbd->disk),
553 "Write on read-only\n");
557 if (req->cmd_flags & REQ_FUA)
558 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
560 /* We did a partial send previously, and we at least sent the whole
561 * request struct, so just go and send the rest of the pages in the
565 if (sent >= sizeof(request)) {
566 skip = sent - sizeof(request);
568 /* initialize handle for tracing purposes */
569 handle = nbd_cmd_handle(cmd);
573 iov_iter_advance(&from, sent);
578 cmd->cookie = nsock->cookie;
580 request.type = htonl(type | nbd_cmd_flags);
581 if (type != NBD_CMD_FLUSH) {
582 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
583 request.len = htonl(size);
585 handle = nbd_cmd_handle(cmd);
586 memcpy(request.handle, &handle, sizeof(handle));
588 trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
590 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
591 req, nbdcmd_to_ascii(type),
592 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
593 result = sock_xmit(nbd, index, 1, &from,
594 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
595 trace_nbd_header_sent(req, handle);
597 if (was_interrupted(result)) {
598 /* If we havne't sent anything we can just return BUSY,
599 * however if we have sent something we need to make
600 * sure we only allow this req to be sent until we are
604 nsock->pending = req;
607 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
608 return BLK_STS_RESOURCE;
610 dev_err_ratelimited(disk_to_dev(nbd->disk),
611 "Send control failed (result %d)\n", result);
615 if (type != NBD_CMD_WRITE)
620 struct bio *next = bio->bi_next;
621 struct bvec_iter iter;
624 bio_for_each_segment(bvec, bio, iter) {
625 bool is_last = !next && bio_iter_last(bvec, iter);
626 int flags = is_last ? 0 : MSG_MORE;
628 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
630 iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
632 if (skip >= iov_iter_count(&from)) {
633 skip -= iov_iter_count(&from);
636 iov_iter_advance(&from, skip);
639 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
641 if (was_interrupted(result)) {
642 /* We've already sent the header, we
643 * have no choice but to set pending and
646 nsock->pending = req;
648 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
649 return BLK_STS_RESOURCE;
651 dev_err(disk_to_dev(nbd->disk),
652 "Send data failed (result %d)\n",
657 * The completion might already have come in,
658 * so break for the last one instead of letting
659 * the iterator do it. This prevents use-after-free
668 trace_nbd_payload_sent(req, handle);
669 nsock->pending = NULL;
674 /* NULL returned = something went wrong, inform userspace */
675 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
677 struct nbd_config *config = nbd->config;
679 struct nbd_reply reply;
681 struct request *req = NULL;
685 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
690 iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
691 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
693 if (!nbd_disconnected(config))
694 dev_err(disk_to_dev(nbd->disk),
695 "Receive control failed (result %d)\n", result);
696 return ERR_PTR(result);
699 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
700 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
701 (unsigned long)ntohl(reply.magic));
702 return ERR_PTR(-EPROTO);
705 memcpy(&handle, reply.handle, sizeof(handle));
706 tag = nbd_handle_to_tag(handle);
707 hwq = blk_mq_unique_tag_to_hwq(tag);
708 if (hwq < nbd->tag_set.nr_hw_queues)
709 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
710 blk_mq_unique_tag_to_tag(tag));
711 if (!req || !blk_mq_request_started(req)) {
712 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
714 return ERR_PTR(-ENOENT);
716 trace_nbd_header_received(req, handle);
717 cmd = blk_mq_rq_to_pdu(req);
719 mutex_lock(&cmd->lock);
720 if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
721 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
722 req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
726 if (cmd->status != BLK_STS_OK) {
727 dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
732 if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
733 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
738 if (ntohl(reply.error)) {
739 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
741 cmd->status = BLK_STS_IOERR;
745 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
746 if (rq_data_dir(req) != WRITE) {
747 struct req_iterator iter;
750 rq_for_each_segment(bvec, req, iter) {
751 iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
752 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
754 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
757 * If we've disconnected, we need to make sure we
758 * complete this request, otherwise error out
759 * and let the timeout stuff handle resubmitting
760 * this request onto another connection.
762 if (nbd_disconnected(config)) {
763 cmd->status = BLK_STS_IOERR;
769 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
774 trace_nbd_payload_received(req, handle);
775 mutex_unlock(&cmd->lock);
776 return ret ? ERR_PTR(ret) : cmd;
779 static void recv_work(struct work_struct *work)
781 struct recv_thread_args *args = container_of(work,
782 struct recv_thread_args,
784 struct nbd_device *nbd = args->nbd;
785 struct nbd_config *config = nbd->config;
790 cmd = nbd_read_stat(nbd, args->index);
792 struct nbd_sock *nsock = config->socks[args->index];
794 mutex_lock(&nsock->tx_lock);
795 nbd_mark_nsock_dead(nbd, nsock, 1);
796 mutex_unlock(&nsock->tx_lock);
800 rq = blk_mq_rq_from_pdu(cmd);
801 if (likely(!blk_should_fake_timeout(rq->q)))
802 blk_mq_complete_request(rq);
804 atomic_dec(&config->recv_threads);
805 wake_up(&config->recv_wq);
810 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
812 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
814 mutex_lock(&cmd->lock);
815 cmd->status = BLK_STS_IOERR;
816 mutex_unlock(&cmd->lock);
818 blk_mq_complete_request(req);
822 static void nbd_clear_que(struct nbd_device *nbd)
824 blk_mq_quiesce_queue(nbd->disk->queue);
825 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
826 blk_mq_unquiesce_queue(nbd->disk->queue);
827 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
830 static int find_fallback(struct nbd_device *nbd, int index)
832 struct nbd_config *config = nbd->config;
834 struct nbd_sock *nsock = config->socks[index];
835 int fallback = nsock->fallback_index;
837 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
840 if (config->num_connections <= 1) {
841 dev_err_ratelimited(disk_to_dev(nbd->disk),
842 "Dead connection, failed to find a fallback\n");
846 if (fallback >= 0 && fallback < config->num_connections &&
847 !config->socks[fallback]->dead)
850 if (nsock->fallback_index < 0 ||
851 nsock->fallback_index >= config->num_connections ||
852 config->socks[nsock->fallback_index]->dead) {
854 for (i = 0; i < config->num_connections; i++) {
857 if (!config->socks[i]->dead) {
862 nsock->fallback_index = new_index;
864 dev_err_ratelimited(disk_to_dev(nbd->disk),
865 "Dead connection, failed to find a fallback\n");
869 new_index = nsock->fallback_index;
873 static int wait_for_reconnect(struct nbd_device *nbd)
875 struct nbd_config *config = nbd->config;
876 if (!config->dead_conn_timeout)
878 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
880 return wait_event_timeout(config->conn_wait,
881 atomic_read(&config->live_connections) > 0,
882 config->dead_conn_timeout) > 0;
885 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
887 struct request *req = blk_mq_rq_from_pdu(cmd);
888 struct nbd_device *nbd = cmd->nbd;
889 struct nbd_config *config;
890 struct nbd_sock *nsock;
893 if (!refcount_inc_not_zero(&nbd->config_refs)) {
894 dev_err_ratelimited(disk_to_dev(nbd->disk),
895 "Socks array is empty\n");
896 blk_mq_start_request(req);
899 config = nbd->config;
901 if (index >= config->num_connections) {
902 dev_err_ratelimited(disk_to_dev(nbd->disk),
903 "Attempted send on invalid socket\n");
905 blk_mq_start_request(req);
908 cmd->status = BLK_STS_OK;
910 nsock = config->socks[index];
911 mutex_lock(&nsock->tx_lock);
913 int old_index = index;
914 index = find_fallback(nbd, index);
915 mutex_unlock(&nsock->tx_lock);
917 if (wait_for_reconnect(nbd)) {
921 /* All the sockets should already be down at this point,
922 * we just want to make sure that DISCONNECTED is set so
923 * any requests that come in that were queue'ed waiting
924 * for the reconnect timer don't trigger the timer again
925 * and instead just error out.
929 blk_mq_start_request(req);
935 /* Handle the case that we have a pending request that was partially
936 * transmitted that _has_ to be serviced first. We need to call requeue
937 * here so that it gets put _after_ the request that is already on the
940 blk_mq_start_request(req);
941 if (unlikely(nsock->pending && nsock->pending != req)) {
942 nbd_requeue_cmd(cmd);
947 * Some failures are related to the link going down, so anything that
948 * returns EAGAIN can be retried on a different socket.
950 ret = nbd_send_cmd(nbd, cmd, index);
951 if (ret == -EAGAIN) {
952 dev_err_ratelimited(disk_to_dev(nbd->disk),
953 "Request send failed, requeueing\n");
954 nbd_mark_nsock_dead(nbd, nsock, 1);
955 nbd_requeue_cmd(cmd);
959 mutex_unlock(&nsock->tx_lock);
964 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
965 const struct blk_mq_queue_data *bd)
967 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
971 * Since we look at the bio's to send the request over the network we
972 * need to make sure the completion work doesn't mark this request done
973 * before we are done doing our send. This keeps us from dereferencing
974 * freed data if we have particularly fast completions (ie we get the
975 * completion before we exit sock_xmit on the last bvec) or in the case
976 * that the server is misbehaving (or there was an error) before we're
977 * done sending everything over the wire.
979 mutex_lock(&cmd->lock);
980 clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
982 /* We can be called directly from the user space process, which means we
983 * could possibly have signals pending so our sendmsg will fail. In
984 * this case we need to return that we are busy, otherwise error out as
987 ret = nbd_handle_cmd(cmd, hctx->queue_num);
992 mutex_unlock(&cmd->lock);
997 static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
1000 struct socket *sock;
1003 sock = sockfd_lookup(fd, err);
1007 if (sock->ops->shutdown == sock_no_shutdown) {
1008 dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
1017 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1020 struct nbd_config *config = nbd->config;
1021 struct socket *sock;
1022 struct nbd_sock **socks;
1023 struct nbd_sock *nsock;
1026 sock = nbd_get_socket(nbd, arg, &err);
1030 if (!netlink && !nbd->task_setup &&
1031 !test_bit(NBD_RT_BOUND, &config->runtime_flags))
1032 nbd->task_setup = current;
1035 (nbd->task_setup != current ||
1036 test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1037 dev_err(disk_to_dev(nbd->disk),
1038 "Device being setup by another task");
1043 nsock = kzalloc(sizeof(*nsock), GFP_KERNEL);
1049 socks = krealloc(config->socks, (config->num_connections + 1) *
1050 sizeof(struct nbd_sock *), GFP_KERNEL);
1057 config->socks = socks;
1059 nsock->fallback_index = -1;
1060 nsock->dead = false;
1061 mutex_init(&nsock->tx_lock);
1063 nsock->pending = NULL;
1066 socks[config->num_connections++] = nsock;
1067 atomic_inc(&config->live_connections);
1076 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1078 struct nbd_config *config = nbd->config;
1079 struct socket *sock, *old;
1080 struct recv_thread_args *args;
1084 sock = nbd_get_socket(nbd, arg, &err);
1088 args = kzalloc(sizeof(*args), GFP_KERNEL);
1094 for (i = 0; i < config->num_connections; i++) {
1095 struct nbd_sock *nsock = config->socks[i];
1100 mutex_lock(&nsock->tx_lock);
1102 mutex_unlock(&nsock->tx_lock);
1105 sk_set_memalloc(sock->sk);
1106 if (nbd->tag_set.timeout)
1107 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1108 atomic_inc(&config->recv_threads);
1109 refcount_inc(&nbd->config_refs);
1111 nsock->fallback_index = -1;
1113 nsock->dead = false;
1114 INIT_WORK(&args->work, recv_work);
1118 mutex_unlock(&nsock->tx_lock);
1121 clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1123 /* We take the tx_mutex in an error path in the recv_work, so we
1124 * need to queue_work outside of the tx_mutex.
1126 queue_work(nbd->recv_workq, &args->work);
1128 atomic_inc(&config->live_connections);
1129 wake_up(&config->conn_wait);
1137 static void nbd_bdev_reset(struct block_device *bdev)
1139 if (bdev->bd_openers > 1)
1141 bd_set_size(bdev, 0);
1144 static void nbd_parse_flags(struct nbd_device *nbd)
1146 struct nbd_config *config = nbd->config;
1147 if (config->flags & NBD_FLAG_READ_ONLY)
1148 set_disk_ro(nbd->disk, true);
1150 set_disk_ro(nbd->disk, false);
1151 if (config->flags & NBD_FLAG_SEND_TRIM)
1152 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1153 if (config->flags & NBD_FLAG_SEND_FLUSH) {
1154 if (config->flags & NBD_FLAG_SEND_FUA)
1155 blk_queue_write_cache(nbd->disk->queue, true, true);
1157 blk_queue_write_cache(nbd->disk->queue, true, false);
1160 blk_queue_write_cache(nbd->disk->queue, false, false);
1163 static void send_disconnects(struct nbd_device *nbd)
1165 struct nbd_config *config = nbd->config;
1166 struct nbd_request request = {
1167 .magic = htonl(NBD_REQUEST_MAGIC),
1168 .type = htonl(NBD_CMD_DISC),
1170 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1171 struct iov_iter from;
1174 for (i = 0; i < config->num_connections; i++) {
1175 struct nbd_sock *nsock = config->socks[i];
1177 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1178 mutex_lock(&nsock->tx_lock);
1179 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1181 dev_err(disk_to_dev(nbd->disk),
1182 "Send disconnect failed %d\n", ret);
1183 mutex_unlock(&nsock->tx_lock);
1187 static int nbd_disconnect(struct nbd_device *nbd)
1189 struct nbd_config *config = nbd->config;
1191 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1192 set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1193 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1194 send_disconnects(nbd);
1198 static void nbd_clear_sock(struct nbd_device *nbd)
1202 nbd->task_setup = NULL;
1205 static void nbd_config_put(struct nbd_device *nbd)
1207 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1208 &nbd->config_lock)) {
1209 struct nbd_config *config = nbd->config;
1210 nbd_dev_dbg_close(nbd);
1211 nbd_size_clear(nbd);
1212 if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1213 &config->runtime_flags))
1214 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1215 nbd->task_recv = NULL;
1216 nbd_clear_sock(nbd);
1217 if (config->num_connections) {
1219 for (i = 0; i < config->num_connections; i++) {
1220 sockfd_put(config->socks[i]->sock);
1221 kfree(config->socks[i]);
1223 kfree(config->socks);
1228 if (nbd->recv_workq)
1229 destroy_workqueue(nbd->recv_workq);
1230 nbd->recv_workq = NULL;
1232 nbd->tag_set.timeout = 0;
1233 nbd->disk->queue->limits.discard_granularity = 0;
1234 nbd->disk->queue->limits.discard_alignment = 0;
1235 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1236 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1238 mutex_unlock(&nbd->config_lock);
1240 module_put(THIS_MODULE);
1244 static int nbd_start_device(struct nbd_device *nbd)
1246 struct nbd_config *config = nbd->config;
1247 int num_connections = config->num_connections;
1254 if (num_connections > 1 &&
1255 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1256 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1260 nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1261 WQ_MEM_RECLAIM | WQ_HIGHPRI |
1262 WQ_UNBOUND, 0, nbd->index);
1263 if (!nbd->recv_workq) {
1264 dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1268 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1269 nbd->task_recv = current;
1271 nbd_parse_flags(nbd);
1273 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1275 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1278 set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1280 nbd_dev_dbg_init(nbd);
1281 for (i = 0; i < num_connections; i++) {
1282 struct recv_thread_args *args;
1284 args = kzalloc(sizeof(*args), GFP_KERNEL);
1288 * If num_connections is m (2 < m),
1289 * and NO.1 ~ NO.n(1 < n < m) kzallocs are successful.
1290 * But NO.(n + 1) failed. We still have n recv threads.
1291 * So, add flush_workqueue here to prevent recv threads
1292 * dropping the last config_refs and trying to destroy
1293 * the workqueue from inside the workqueue.
1296 flush_workqueue(nbd->recv_workq);
1299 sk_set_memalloc(config->socks[i]->sock->sk);
1300 if (nbd->tag_set.timeout)
1301 config->socks[i]->sock->sk->sk_sndtimeo =
1302 nbd->tag_set.timeout;
1303 atomic_inc(&config->recv_threads);
1304 refcount_inc(&nbd->config_refs);
1305 INIT_WORK(&args->work, recv_work);
1308 queue_work(nbd->recv_workq, &args->work);
1310 nbd_size_update(nbd);
1314 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1316 struct nbd_config *config = nbd->config;
1319 ret = nbd_start_device(nbd);
1324 bdev->bd_invalidated = 1;
1325 mutex_unlock(&nbd->config_lock);
1326 ret = wait_event_interruptible(config->recv_wq,
1327 atomic_read(&config->recv_threads) == 0);
1330 flush_workqueue(nbd->recv_workq);
1332 mutex_lock(&nbd->config_lock);
1333 nbd_bdev_reset(bdev);
1334 /* user requested, ignore socket errors */
1335 if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1337 if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1342 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1343 struct block_device *bdev)
1346 __invalidate_device(bdev, true);
1347 nbd_bdev_reset(bdev);
1348 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1349 &nbd->config->runtime_flags))
1350 nbd_config_put(nbd);
1353 static bool nbd_is_valid_blksize(unsigned long blksize)
1355 if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1356 blksize > PAGE_SIZE)
1361 static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1363 nbd->tag_set.timeout = timeout * HZ;
1365 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1367 blk_queue_rq_timeout(nbd->disk->queue, 30 * HZ);
1370 /* Must be called with config_lock held */
1371 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1372 unsigned int cmd, unsigned long arg)
1374 struct nbd_config *config = nbd->config;
1377 case NBD_DISCONNECT:
1378 return nbd_disconnect(nbd);
1379 case NBD_CLEAR_SOCK:
1380 nbd_clear_sock_ioctl(nbd, bdev);
1383 return nbd_add_socket(nbd, arg, false);
1384 case NBD_SET_BLKSIZE:
1386 arg = NBD_DEF_BLKSIZE;
1387 if (!nbd_is_valid_blksize(arg))
1389 nbd_size_set(nbd, arg,
1390 div_s64(config->bytesize, arg));
1393 nbd_size_set(nbd, config->blksize,
1394 div_s64(arg, config->blksize));
1396 case NBD_SET_SIZE_BLOCKS:
1397 nbd_size_set(nbd, config->blksize, arg);
1399 case NBD_SET_TIMEOUT:
1400 nbd_set_cmd_timeout(nbd, arg);
1404 config->flags = arg;
1407 return nbd_start_device_ioctl(nbd, bdev);
1410 * This is for compatibility only. The queue is always cleared
1411 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1414 case NBD_PRINT_DEBUG:
1416 * For compatibility only, we no longer keep a list of
1417 * outstanding requests.
1424 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1425 unsigned int cmd, unsigned long arg)
1427 struct nbd_device *nbd = bdev->bd_disk->private_data;
1428 struct nbd_config *config = nbd->config;
1429 int error = -EINVAL;
1431 if (!capable(CAP_SYS_ADMIN))
1434 /* The block layer will pass back some non-nbd ioctls in case we have
1435 * special handling for them, but we don't so just return an error.
1437 if (_IOC_TYPE(cmd) != 0xab)
1440 mutex_lock(&nbd->config_lock);
1442 /* Don't allow ioctl operations on a nbd device that was created with
1443 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1445 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1446 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1447 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1449 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1450 mutex_unlock(&nbd->config_lock);
1454 static struct nbd_config *nbd_alloc_config(void)
1456 struct nbd_config *config;
1458 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1461 atomic_set(&config->recv_threads, 0);
1462 init_waitqueue_head(&config->recv_wq);
1463 init_waitqueue_head(&config->conn_wait);
1464 config->blksize = NBD_DEF_BLKSIZE;
1465 atomic_set(&config->live_connections, 0);
1466 try_module_get(THIS_MODULE);
1470 static int nbd_open(struct block_device *bdev, fmode_t mode)
1472 struct nbd_device *nbd;
1475 mutex_lock(&nbd_index_mutex);
1476 nbd = bdev->bd_disk->private_data;
1481 if (!refcount_inc_not_zero(&nbd->refs)) {
1485 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1486 struct nbd_config *config;
1488 mutex_lock(&nbd->config_lock);
1489 if (refcount_inc_not_zero(&nbd->config_refs)) {
1490 mutex_unlock(&nbd->config_lock);
1493 config = nbd->config = nbd_alloc_config();
1496 mutex_unlock(&nbd->config_lock);
1499 refcount_set(&nbd->config_refs, 1);
1500 refcount_inc(&nbd->refs);
1501 mutex_unlock(&nbd->config_lock);
1502 bdev->bd_invalidated = 1;
1503 } else if (nbd_disconnected(nbd->config)) {
1504 bdev->bd_invalidated = 1;
1507 mutex_unlock(&nbd_index_mutex);
1511 static void nbd_release(struct gendisk *disk, fmode_t mode)
1513 struct nbd_device *nbd = disk->private_data;
1514 struct block_device *bdev = bdget_disk(disk, 0);
1516 if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1517 bdev->bd_openers == 0)
1518 nbd_disconnect_and_put(nbd);
1520 nbd_config_put(nbd);
1524 static const struct block_device_operations nbd_fops =
1526 .owner = THIS_MODULE,
1528 .release = nbd_release,
1530 .compat_ioctl = nbd_ioctl,
1533 #if IS_ENABLED(CONFIG_DEBUG_FS)
1535 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1537 struct nbd_device *nbd = s->private;
1540 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1545 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1547 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1550 static const struct file_operations nbd_dbg_tasks_ops = {
1551 .open = nbd_dbg_tasks_open,
1553 .llseek = seq_lseek,
1554 .release = single_release,
1557 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1559 struct nbd_device *nbd = s->private;
1560 u32 flags = nbd->config->flags;
1562 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1564 seq_puts(s, "Known flags:\n");
1566 if (flags & NBD_FLAG_HAS_FLAGS)
1567 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1568 if (flags & NBD_FLAG_READ_ONLY)
1569 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1570 if (flags & NBD_FLAG_SEND_FLUSH)
1571 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1572 if (flags & NBD_FLAG_SEND_FUA)
1573 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1574 if (flags & NBD_FLAG_SEND_TRIM)
1575 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1580 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1582 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1585 static const struct file_operations nbd_dbg_flags_ops = {
1586 .open = nbd_dbg_flags_open,
1588 .llseek = seq_lseek,
1589 .release = single_release,
1592 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1595 struct nbd_config *config = nbd->config;
1600 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1602 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1606 config->dbg_dir = dir;
1608 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1609 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1610 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1611 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1612 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1617 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1619 debugfs_remove_recursive(nbd->config->dbg_dir);
1622 static int nbd_dbg_init(void)
1624 struct dentry *dbg_dir;
1626 dbg_dir = debugfs_create_dir("nbd", NULL);
1630 nbd_dbg_dir = dbg_dir;
1635 static void nbd_dbg_close(void)
1637 debugfs_remove_recursive(nbd_dbg_dir);
1640 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1642 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1647 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1651 static int nbd_dbg_init(void)
1656 static void nbd_dbg_close(void)
1662 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1663 unsigned int hctx_idx, unsigned int numa_node)
1665 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1666 cmd->nbd = set->driver_data;
1668 mutex_init(&cmd->lock);
1672 static const struct blk_mq_ops nbd_mq_ops = {
1673 .queue_rq = nbd_queue_rq,
1674 .complete = nbd_complete_rq,
1675 .init_request = nbd_init_request,
1676 .timeout = nbd_xmit_timeout,
1679 static int nbd_dev_add(int index)
1681 struct nbd_device *nbd;
1682 struct gendisk *disk;
1683 struct request_queue *q;
1686 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1690 disk = alloc_disk(1 << part_shift);
1695 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1700 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1709 nbd->tag_set.ops = &nbd_mq_ops;
1710 nbd->tag_set.nr_hw_queues = 1;
1711 nbd->tag_set.queue_depth = 128;
1712 nbd->tag_set.numa_node = NUMA_NO_NODE;
1713 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1714 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1716 nbd->tag_set.driver_data = nbd;
1717 nbd->destroy_complete = NULL;
1719 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1723 q = blk_mq_init_queue(&nbd->tag_set);
1731 * Tell the block layer that we are not a rotational device
1733 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1734 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1735 disk->queue->limits.discard_granularity = 0;
1736 disk->queue->limits.discard_alignment = 0;
1737 blk_queue_max_discard_sectors(disk->queue, 0);
1738 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1739 blk_queue_max_segments(disk->queue, USHRT_MAX);
1740 blk_queue_max_hw_sectors(disk->queue, 65536);
1741 disk->queue->limits.max_sectors = 256;
1743 mutex_init(&nbd->config_lock);
1744 refcount_set(&nbd->config_refs, 0);
1745 refcount_set(&nbd->refs, 1);
1746 INIT_LIST_HEAD(&nbd->list);
1747 disk->major = NBD_MAJOR;
1748 disk->first_minor = index << part_shift;
1749 disk->fops = &nbd_fops;
1750 disk->private_data = nbd;
1751 sprintf(disk->disk_name, "nbd%d", index);
1753 nbd_total_devices++;
1757 blk_mq_free_tag_set(&nbd->tag_set);
1759 idr_remove(&nbd_index_idr, index);
1768 static int find_free_cb(int id, void *ptr, void *data)
1770 struct nbd_device *nbd = ptr;
1771 struct nbd_device **found = data;
1773 if (!refcount_read(&nbd->config_refs)) {
1780 /* Netlink interface. */
1781 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1782 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1783 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1784 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1785 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1786 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1787 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1788 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1789 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1790 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1793 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1794 [NBD_SOCK_FD] = { .type = NLA_U32 },
1797 /* We don't use this right now since we don't parse the incoming list, but we
1798 * still want it here so userspace knows what to expect.
1800 static const struct nla_policy __attribute__((unused))
1801 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1802 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1803 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1806 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1808 struct nbd_config *config = nbd->config;
1809 u64 bsize = config->blksize;
1810 u64 bytes = config->bytesize;
1812 if (info->attrs[NBD_ATTR_SIZE_BYTES])
1813 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1815 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1816 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1818 bsize = NBD_DEF_BLKSIZE;
1819 if (!nbd_is_valid_blksize(bsize)) {
1820 printk(KERN_ERR "Invalid block size %llu\n", bsize);
1825 if (bytes != config->bytesize || bsize != config->blksize)
1826 nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1830 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1832 DECLARE_COMPLETION_ONSTACK(destroy_complete);
1833 struct nbd_device *nbd = NULL;
1834 struct nbd_config *config;
1837 bool put_dev = false;
1839 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1842 if (info->attrs[NBD_ATTR_INDEX])
1843 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1844 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1845 printk(KERN_ERR "nbd: must specify at least one socket\n");
1848 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1849 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1853 mutex_lock(&nbd_index_mutex);
1855 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1858 new_index = nbd_dev_add(-1);
1859 if (new_index < 0) {
1860 mutex_unlock(&nbd_index_mutex);
1861 printk(KERN_ERR "nbd: failed to add new device\n");
1864 nbd = idr_find(&nbd_index_idr, new_index);
1867 nbd = idr_find(&nbd_index_idr, index);
1869 ret = nbd_dev_add(index);
1871 mutex_unlock(&nbd_index_mutex);
1872 printk(KERN_ERR "nbd: failed to add new device\n");
1875 nbd = idr_find(&nbd_index_idr, index);
1879 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1881 mutex_unlock(&nbd_index_mutex);
1885 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1886 test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1887 nbd->destroy_complete = &destroy_complete;
1888 mutex_unlock(&nbd_index_mutex);
1890 /* Wait untill the the nbd stuff is totally destroyed */
1891 wait_for_completion(&destroy_complete);
1895 if (!refcount_inc_not_zero(&nbd->refs)) {
1896 mutex_unlock(&nbd_index_mutex);
1899 printk(KERN_ERR "nbd: device at index %d is going down\n",
1903 mutex_unlock(&nbd_index_mutex);
1905 mutex_lock(&nbd->config_lock);
1906 if (refcount_read(&nbd->config_refs)) {
1907 mutex_unlock(&nbd->config_lock);
1911 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1914 if (WARN_ON(nbd->config)) {
1915 mutex_unlock(&nbd->config_lock);
1919 config = nbd->config = nbd_alloc_config();
1921 mutex_unlock(&nbd->config_lock);
1923 printk(KERN_ERR "nbd: couldn't allocate config\n");
1926 refcount_set(&nbd->config_refs, 1);
1927 set_bit(NBD_RT_BOUND, &config->runtime_flags);
1929 ret = nbd_genl_size_set(info, nbd);
1933 if (info->attrs[NBD_ATTR_TIMEOUT])
1934 nbd_set_cmd_timeout(nbd,
1935 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1936 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1937 config->dead_conn_timeout =
1938 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1939 config->dead_conn_timeout *= HZ;
1941 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1943 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1944 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1945 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1946 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1947 set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1948 &config->runtime_flags);
1949 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1952 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1954 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1955 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1956 &config->runtime_flags);
1960 if (info->attrs[NBD_ATTR_SOCKETS]) {
1961 struct nlattr *attr;
1964 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1966 struct nlattr *socks[NBD_SOCK_MAX+1];
1968 if (nla_type(attr) != NBD_SOCK_ITEM) {
1969 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1973 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1978 printk(KERN_ERR "nbd: error processing sock list\n");
1982 if (!socks[NBD_SOCK_FD])
1984 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1985 ret = nbd_add_socket(nbd, fd, true);
1990 ret = nbd_start_device(nbd);
1992 mutex_unlock(&nbd->config_lock);
1994 set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1995 refcount_inc(&nbd->config_refs);
1996 nbd_connect_reply(info, nbd->index);
1998 nbd_config_put(nbd);
2004 static void nbd_disconnect_and_put(struct nbd_device *nbd)
2006 mutex_lock(&nbd->config_lock);
2007 nbd_disconnect(nbd);
2008 nbd_clear_sock(nbd);
2009 mutex_unlock(&nbd->config_lock);
2011 * Make sure recv thread has finished, so it does not drop the last
2012 * config ref and try to destroy the workqueue from inside the work
2015 flush_workqueue(nbd->recv_workq);
2016 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
2017 &nbd->config->runtime_flags))
2018 nbd_config_put(nbd);
2021 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
2023 struct nbd_device *nbd;
2026 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2029 if (!info->attrs[NBD_ATTR_INDEX]) {
2030 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
2033 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2034 mutex_lock(&nbd_index_mutex);
2035 nbd = idr_find(&nbd_index_idr, index);
2037 mutex_unlock(&nbd_index_mutex);
2038 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2042 if (!refcount_inc_not_zero(&nbd->refs)) {
2043 mutex_unlock(&nbd_index_mutex);
2044 printk(KERN_ERR "nbd: device at index %d is going down\n",
2048 mutex_unlock(&nbd_index_mutex);
2049 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2053 nbd_disconnect_and_put(nbd);
2054 nbd_config_put(nbd);
2059 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2061 struct nbd_device *nbd = NULL;
2062 struct nbd_config *config;
2065 bool put_dev = false;
2067 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2070 if (!info->attrs[NBD_ATTR_INDEX]) {
2071 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2074 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2075 mutex_lock(&nbd_index_mutex);
2076 nbd = idr_find(&nbd_index_idr, index);
2078 mutex_unlock(&nbd_index_mutex);
2079 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2083 if (!refcount_inc_not_zero(&nbd->refs)) {
2084 mutex_unlock(&nbd_index_mutex);
2085 printk(KERN_ERR "nbd: device at index %d is going down\n",
2089 mutex_unlock(&nbd_index_mutex);
2091 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2092 dev_err(nbd_to_dev(nbd),
2093 "not configured, cannot reconfigure\n");
2098 mutex_lock(&nbd->config_lock);
2099 config = nbd->config;
2100 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2102 dev_err(nbd_to_dev(nbd),
2103 "not configured, cannot reconfigure\n");
2108 ret = nbd_genl_size_set(info, nbd);
2112 if (info->attrs[NBD_ATTR_TIMEOUT])
2113 nbd_set_cmd_timeout(nbd,
2114 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2115 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2116 config->dead_conn_timeout =
2117 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2118 config->dead_conn_timeout *= HZ;
2120 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2121 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2122 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2123 if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2124 &config->runtime_flags))
2126 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2128 if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2129 &config->runtime_flags))
2130 refcount_inc(&nbd->refs);
2131 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2134 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2135 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2136 &config->runtime_flags);
2138 clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2139 &config->runtime_flags);
2143 if (info->attrs[NBD_ATTR_SOCKETS]) {
2144 struct nlattr *attr;
2147 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2149 struct nlattr *socks[NBD_SOCK_MAX+1];
2151 if (nla_type(attr) != NBD_SOCK_ITEM) {
2152 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2156 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2161 printk(KERN_ERR "nbd: error processing sock list\n");
2165 if (!socks[NBD_SOCK_FD])
2167 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2168 ret = nbd_reconnect_socket(nbd, fd);
2174 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2178 mutex_unlock(&nbd->config_lock);
2179 nbd_config_put(nbd);
2186 static const struct genl_ops nbd_connect_genl_ops[] = {
2188 .cmd = NBD_CMD_CONNECT,
2189 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2190 .doit = nbd_genl_connect,
2193 .cmd = NBD_CMD_DISCONNECT,
2194 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2195 .doit = nbd_genl_disconnect,
2198 .cmd = NBD_CMD_RECONFIGURE,
2199 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2200 .doit = nbd_genl_reconfigure,
2203 .cmd = NBD_CMD_STATUS,
2204 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2205 .doit = nbd_genl_status,
2209 static const struct genl_multicast_group nbd_mcast_grps[] = {
2210 { .name = NBD_GENL_MCAST_GROUP_NAME, },
2213 static struct genl_family nbd_genl_family __ro_after_init = {
2215 .name = NBD_GENL_FAMILY_NAME,
2216 .version = NBD_GENL_VERSION,
2217 .module = THIS_MODULE,
2218 .ops = nbd_connect_genl_ops,
2219 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2220 .maxattr = NBD_ATTR_MAX,
2221 .policy = nbd_attr_policy,
2222 .mcgrps = nbd_mcast_grps,
2223 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2226 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2228 struct nlattr *dev_opt;
2232 /* This is a little racey, but for status it's ok. The
2233 * reason we don't take a ref here is because we can't
2234 * take a ref in the index == -1 case as we would need
2235 * to put under the nbd_index_mutex, which could
2236 * deadlock if we are configured to remove ourselves
2237 * once we're disconnected.
2239 if (refcount_read(&nbd->config_refs))
2241 dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2244 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2247 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2251 nla_nest_end(reply, dev_opt);
2255 static int status_cb(int id, void *ptr, void *data)
2257 struct nbd_device *nbd = ptr;
2258 return populate_nbd_status(nbd, (struct sk_buff *)data);
2261 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2263 struct nlattr *dev_list;
2264 struct sk_buff *reply;
2270 if (info->attrs[NBD_ATTR_INDEX])
2271 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2273 mutex_lock(&nbd_index_mutex);
2275 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2276 nla_attr_size(sizeof(u8)));
2277 msg_size *= (index == -1) ? nbd_total_devices : 1;
2279 reply = genlmsg_new(msg_size, GFP_KERNEL);
2282 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2289 dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2291 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2297 struct nbd_device *nbd;
2298 nbd = idr_find(&nbd_index_idr, index);
2300 ret = populate_nbd_status(nbd, reply);
2307 nla_nest_end(reply, dev_list);
2308 genlmsg_end(reply, reply_head);
2309 ret = genlmsg_reply(reply, info);
2311 mutex_unlock(&nbd_index_mutex);
2315 static void nbd_connect_reply(struct genl_info *info, int index)
2317 struct sk_buff *skb;
2321 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2324 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2330 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2335 genlmsg_end(skb, msg_head);
2336 genlmsg_reply(skb, info);
2339 static void nbd_mcast_index(int index)
2341 struct sk_buff *skb;
2345 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2348 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2354 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2359 genlmsg_end(skb, msg_head);
2360 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2363 static void nbd_dead_link_work(struct work_struct *work)
2365 struct link_dead_args *args = container_of(work, struct link_dead_args,
2367 nbd_mcast_index(args->index);
2371 static int __init nbd_init(void)
2375 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2378 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2384 part_shift = fls(max_part);
2387 * Adjust max_part according to part_shift as it is exported
2388 * to user space so that user can know the max number of
2389 * partition kernel should be able to manage.
2391 * Note that -1 is required because partition 0 is reserved
2392 * for the whole disk.
2394 max_part = (1UL << part_shift) - 1;
2397 if ((1UL << part_shift) > DISK_MAX_PARTS)
2400 if (nbds_max > 1UL << (MINORBITS - part_shift))
2403 if (register_blkdev(NBD_MAJOR, "nbd"))
2406 if (genl_register_family(&nbd_genl_family)) {
2407 unregister_blkdev(NBD_MAJOR, "nbd");
2412 mutex_lock(&nbd_index_mutex);
2413 for (i = 0; i < nbds_max; i++)
2415 mutex_unlock(&nbd_index_mutex);
2419 static int nbd_exit_cb(int id, void *ptr, void *data)
2421 struct list_head *list = (struct list_head *)data;
2422 struct nbd_device *nbd = ptr;
2424 list_add_tail(&nbd->list, list);
2428 static void __exit nbd_cleanup(void)
2430 struct nbd_device *nbd;
2431 LIST_HEAD(del_list);
2435 mutex_lock(&nbd_index_mutex);
2436 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2437 mutex_unlock(&nbd_index_mutex);
2439 while (!list_empty(&del_list)) {
2440 nbd = list_first_entry(&del_list, struct nbd_device, list);
2441 list_del_init(&nbd->list);
2442 if (refcount_read(&nbd->refs) != 1)
2443 printk(KERN_ERR "nbd: possibly leaking a device\n");
2447 idr_destroy(&nbd_index_idr);
2448 genl_unregister_family(&nbd_genl_family);
2449 unregister_blkdev(NBD_MAJOR, "nbd");
2452 module_init(nbd_init);
2453 module_exit(nbd_cleanup);
2455 MODULE_DESCRIPTION("Network Block Device");
2456 MODULE_LICENSE("GPL");
2458 module_param(nbds_max, int, 0444);
2459 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2460 module_param(max_part, int, 0444);
2461 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");