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_DISCONNECT_ON_CLOSE 6
83 #define NBD_DESTROY_ON_DISCONNECT 0
84 #define NBD_DISCONNECT_REQUESTED 1
88 unsigned long runtime_flags;
89 u64 dead_conn_timeout;
91 struct nbd_sock **socks;
93 atomic_t live_connections;
94 wait_queue_head_t conn_wait;
96 atomic_t recv_threads;
97 wait_queue_head_t recv_wq;
100 #if IS_ENABLED(CONFIG_DEBUG_FS)
101 struct dentry *dbg_dir;
106 struct blk_mq_tag_set tag_set;
109 refcount_t config_refs;
111 struct nbd_config *config;
112 struct mutex config_lock;
113 struct gendisk *disk;
114 struct workqueue_struct *recv_workq;
116 struct list_head list;
117 struct task_struct *task_recv;
118 struct task_struct *task_setup;
120 struct completion *destroy_complete;
124 #define NBD_CMD_REQUEUED 1
127 struct nbd_device *nbd;
137 #if IS_ENABLED(CONFIG_DEBUG_FS)
138 static struct dentry *nbd_dbg_dir;
141 #define nbd_name(nbd) ((nbd)->disk->disk_name)
143 #define NBD_MAGIC 0x68797548
145 #define NBD_DEF_BLKSIZE 1024
147 static unsigned int nbds_max = 16;
148 static int max_part = 16;
149 static int part_shift;
151 static int nbd_dev_dbg_init(struct nbd_device *nbd);
152 static void nbd_dev_dbg_close(struct nbd_device *nbd);
153 static void nbd_config_put(struct nbd_device *nbd);
154 static void nbd_connect_reply(struct genl_info *info, int index);
155 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
156 static void nbd_dead_link_work(struct work_struct *work);
157 static void nbd_disconnect_and_put(struct nbd_device *nbd);
159 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
161 return disk_to_dev(nbd->disk);
164 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
166 struct request *req = blk_mq_rq_from_pdu(cmd);
168 if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
169 blk_mq_requeue_request(req, true);
172 #define NBD_COOKIE_BITS 32
174 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
176 struct request *req = blk_mq_rq_from_pdu(cmd);
177 u32 tag = blk_mq_unique_tag(req);
178 u64 cookie = cmd->cmd_cookie;
180 return (cookie << NBD_COOKIE_BITS) | tag;
183 static u32 nbd_handle_to_tag(u64 handle)
188 static u32 nbd_handle_to_cookie(u64 handle)
190 return (u32)(handle >> NBD_COOKIE_BITS);
193 static const char *nbdcmd_to_ascii(int cmd)
196 case NBD_CMD_READ: return "read";
197 case NBD_CMD_WRITE: return "write";
198 case NBD_CMD_DISC: return "disconnect";
199 case NBD_CMD_FLUSH: return "flush";
200 case NBD_CMD_TRIM: return "trim/discard";
205 static ssize_t pid_show(struct device *dev,
206 struct device_attribute *attr, char *buf)
208 struct gendisk *disk = dev_to_disk(dev);
209 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
211 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
214 static const struct device_attribute pid_attr = {
215 .attr = { .name = "pid", .mode = 0444},
219 static void nbd_dev_remove(struct nbd_device *nbd)
221 struct gendisk *disk = nbd->disk;
222 struct request_queue *q;
227 blk_cleanup_queue(q);
228 blk_mq_free_tag_set(&nbd->tag_set);
229 disk->private_data = NULL;
234 * Place this in the last just before the nbd is freed to
235 * make sure that the disk and the related kobject are also
236 * totally removed to avoid duplicate creation of the same
239 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) && nbd->destroy_complete)
240 complete(nbd->destroy_complete);
245 static void nbd_put(struct nbd_device *nbd)
247 if (refcount_dec_and_mutex_lock(&nbd->refs,
249 idr_remove(&nbd_index_idr, nbd->index);
251 mutex_unlock(&nbd_index_mutex);
255 static int nbd_disconnected(struct nbd_config *config)
257 return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
258 test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
261 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
264 if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
265 struct link_dead_args *args;
266 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
268 INIT_WORK(&args->work, nbd_dead_link_work);
269 args->index = nbd->index;
270 queue_work(system_wq, &args->work);
274 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
275 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
276 if (test_and_clear_bit(NBD_RT_DISCONNECT_REQUESTED,
277 &nbd->config->runtime_flags)) {
278 set_bit(NBD_RT_DISCONNECTED,
279 &nbd->config->runtime_flags);
280 dev_info(nbd_to_dev(nbd),
281 "Disconnected due to user request.\n");
286 nsock->pending = NULL;
290 static void nbd_size_clear(struct nbd_device *nbd)
292 if (nbd->config->bytesize) {
293 set_capacity(nbd->disk, 0);
294 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
298 static int nbd_set_size(struct nbd_device *nbd, loff_t bytesize,
302 blksize = NBD_DEF_BLKSIZE;
303 if (blksize < 512 || blksize > PAGE_SIZE || !is_power_of_2(blksize))
306 nbd->config->bytesize = bytesize;
307 nbd->config->blksize = blksize;
312 if (nbd->config->flags & NBD_FLAG_SEND_TRIM) {
313 nbd->disk->queue->limits.discard_granularity = blksize;
314 nbd->disk->queue->limits.discard_alignment = blksize;
315 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
317 blk_queue_logical_block_size(nbd->disk->queue, blksize);
318 blk_queue_physical_block_size(nbd->disk->queue, blksize);
321 set_bit(GD_NEED_PART_SCAN, &nbd->disk->state);
322 if (!set_capacity_and_notify(nbd->disk, bytesize >> 9))
323 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
327 static void nbd_complete_rq(struct request *req)
329 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
331 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
332 cmd->status ? "failed" : "done");
334 blk_mq_end_request(req, cmd->status);
338 * Forcibly shutdown the socket causing all listeners to error
340 static void sock_shutdown(struct nbd_device *nbd)
342 struct nbd_config *config = nbd->config;
345 if (config->num_connections == 0)
347 if (test_and_set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
350 for (i = 0; i < config->num_connections; i++) {
351 struct nbd_sock *nsock = config->socks[i];
352 mutex_lock(&nsock->tx_lock);
353 nbd_mark_nsock_dead(nbd, nsock, 0);
354 mutex_unlock(&nsock->tx_lock);
356 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
359 static u32 req_to_nbd_cmd_type(struct request *req)
361 switch (req_op(req)) {
365 return NBD_CMD_FLUSH;
367 return NBD_CMD_WRITE;
375 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
378 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
379 struct nbd_device *nbd = cmd->nbd;
380 struct nbd_config *config;
382 if (!mutex_trylock(&cmd->lock))
383 return BLK_EH_RESET_TIMER;
385 if (!refcount_inc_not_zero(&nbd->config_refs)) {
386 cmd->status = BLK_STS_TIMEOUT;
387 mutex_unlock(&cmd->lock);
390 config = nbd->config;
392 if (config->num_connections > 1 ||
393 (config->num_connections == 1 && nbd->tag_set.timeout)) {
394 dev_err_ratelimited(nbd_to_dev(nbd),
395 "Connection timed out, retrying (%d/%d alive)\n",
396 atomic_read(&config->live_connections),
397 config->num_connections);
399 * Hooray we have more connections, requeue this IO, the submit
400 * path will put it on a real connection. Or if only one
401 * connection is configured, the submit path will wait util
402 * a new connection is reconfigured or util dead timeout.
405 if (cmd->index < config->num_connections) {
406 struct nbd_sock *nsock =
407 config->socks[cmd->index];
408 mutex_lock(&nsock->tx_lock);
409 /* We can have multiple outstanding requests, so
410 * we don't want to mark the nsock dead if we've
411 * already reconnected with a new socket, so
412 * only mark it dead if its the same socket we
415 if (cmd->cookie == nsock->cookie)
416 nbd_mark_nsock_dead(nbd, nsock, 1);
417 mutex_unlock(&nsock->tx_lock);
419 mutex_unlock(&cmd->lock);
420 nbd_requeue_cmd(cmd);
426 if (!nbd->tag_set.timeout) {
428 * Userspace sets timeout=0 to disable socket disconnection,
429 * so just warn and reset the timer.
431 struct nbd_sock *nsock = config->socks[cmd->index];
433 dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
434 req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
435 (unsigned long long)blk_rq_pos(req) << 9,
436 blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
438 mutex_lock(&nsock->tx_lock);
439 if (cmd->cookie != nsock->cookie) {
440 nbd_requeue_cmd(cmd);
441 mutex_unlock(&nsock->tx_lock);
442 mutex_unlock(&cmd->lock);
446 mutex_unlock(&nsock->tx_lock);
447 mutex_unlock(&cmd->lock);
449 return BLK_EH_RESET_TIMER;
452 dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
453 set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
454 cmd->status = BLK_STS_IOERR;
455 mutex_unlock(&cmd->lock);
459 blk_mq_complete_request(req);
464 * Send or receive packet.
466 static int sock_xmit(struct nbd_device *nbd, int index, int send,
467 struct iov_iter *iter, int msg_flags, int *sent)
469 struct nbd_config *config = nbd->config;
470 struct socket *sock = config->socks[index]->sock;
473 unsigned int noreclaim_flag;
475 if (unlikely(!sock)) {
476 dev_err_ratelimited(disk_to_dev(nbd->disk),
477 "Attempted %s on closed socket in sock_xmit\n",
478 (send ? "send" : "recv"));
482 msg.msg_iter = *iter;
484 noreclaim_flag = memalloc_noreclaim_save();
486 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
489 msg.msg_control = NULL;
490 msg.msg_controllen = 0;
491 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
494 result = sock_sendmsg(sock, &msg);
496 result = sock_recvmsg(sock, &msg, msg.msg_flags);
500 result = -EPIPE; /* short read */
505 } while (msg_data_left(&msg));
507 memalloc_noreclaim_restore(noreclaim_flag);
513 * Different settings for sk->sk_sndtimeo can result in different return values
514 * if there is a signal pending when we enter sendmsg, because reasons?
516 static inline int was_interrupted(int result)
518 return result == -ERESTARTSYS || result == -EINTR;
521 /* always call with the tx_lock held */
522 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
524 struct request *req = blk_mq_rq_from_pdu(cmd);
525 struct nbd_config *config = nbd->config;
526 struct nbd_sock *nsock = config->socks[index];
528 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
529 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
530 struct iov_iter from;
531 unsigned long size = blk_rq_bytes(req);
535 u32 nbd_cmd_flags = 0;
536 int sent = nsock->sent, skip = 0;
538 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
540 type = req_to_nbd_cmd_type(req);
544 if (rq_data_dir(req) == WRITE &&
545 (config->flags & NBD_FLAG_READ_ONLY)) {
546 dev_err_ratelimited(disk_to_dev(nbd->disk),
547 "Write on read-only\n");
551 if (req->cmd_flags & REQ_FUA)
552 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
554 /* We did a partial send previously, and we at least sent the whole
555 * request struct, so just go and send the rest of the pages in the
559 if (sent >= sizeof(request)) {
560 skip = sent - sizeof(request);
562 /* initialize handle for tracing purposes */
563 handle = nbd_cmd_handle(cmd);
567 iov_iter_advance(&from, sent);
572 cmd->cookie = nsock->cookie;
574 request.type = htonl(type | nbd_cmd_flags);
575 if (type != NBD_CMD_FLUSH) {
576 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
577 request.len = htonl(size);
579 handle = nbd_cmd_handle(cmd);
580 memcpy(request.handle, &handle, sizeof(handle));
582 trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
584 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
585 req, nbdcmd_to_ascii(type),
586 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
587 result = sock_xmit(nbd, index, 1, &from,
588 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
589 trace_nbd_header_sent(req, handle);
591 if (was_interrupted(result)) {
592 /* If we havne't sent anything we can just return BUSY,
593 * however if we have sent something we need to make
594 * sure we only allow this req to be sent until we are
598 nsock->pending = req;
601 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
602 return BLK_STS_RESOURCE;
604 dev_err_ratelimited(disk_to_dev(nbd->disk),
605 "Send control failed (result %d)\n", result);
609 if (type != NBD_CMD_WRITE)
614 struct bio *next = bio->bi_next;
615 struct bvec_iter iter;
618 bio_for_each_segment(bvec, bio, iter) {
619 bool is_last = !next && bio_iter_last(bvec, iter);
620 int flags = is_last ? 0 : MSG_MORE;
622 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
624 iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
626 if (skip >= iov_iter_count(&from)) {
627 skip -= iov_iter_count(&from);
630 iov_iter_advance(&from, skip);
633 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
635 if (was_interrupted(result)) {
636 /* We've already sent the header, we
637 * have no choice but to set pending and
640 nsock->pending = req;
642 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
643 return BLK_STS_RESOURCE;
645 dev_err(disk_to_dev(nbd->disk),
646 "Send data failed (result %d)\n",
651 * The completion might already have come in,
652 * so break for the last one instead of letting
653 * the iterator do it. This prevents use-after-free
662 trace_nbd_payload_sent(req, handle);
663 nsock->pending = NULL;
668 /* NULL returned = something went wrong, inform userspace */
669 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
671 struct nbd_config *config = nbd->config;
673 struct nbd_reply reply;
675 struct request *req = NULL;
679 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
684 iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
685 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
687 if (!nbd_disconnected(config))
688 dev_err(disk_to_dev(nbd->disk),
689 "Receive control failed (result %d)\n", result);
690 return ERR_PTR(result);
693 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
694 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
695 (unsigned long)ntohl(reply.magic));
696 return ERR_PTR(-EPROTO);
699 memcpy(&handle, reply.handle, sizeof(handle));
700 tag = nbd_handle_to_tag(handle);
701 hwq = blk_mq_unique_tag_to_hwq(tag);
702 if (hwq < nbd->tag_set.nr_hw_queues)
703 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
704 blk_mq_unique_tag_to_tag(tag));
705 if (!req || !blk_mq_request_started(req)) {
706 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
708 return ERR_PTR(-ENOENT);
710 trace_nbd_header_received(req, handle);
711 cmd = blk_mq_rq_to_pdu(req);
713 mutex_lock(&cmd->lock);
714 if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
715 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
716 req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
720 if (cmd->status != BLK_STS_OK) {
721 dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
726 if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
727 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
732 if (ntohl(reply.error)) {
733 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
735 cmd->status = BLK_STS_IOERR;
739 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
740 if (rq_data_dir(req) != WRITE) {
741 struct req_iterator iter;
744 rq_for_each_segment(bvec, req, iter) {
745 iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
746 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
748 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
751 * If we've disconnected, we need to make sure we
752 * complete this request, otherwise error out
753 * and let the timeout stuff handle resubmitting
754 * this request onto another connection.
756 if (nbd_disconnected(config)) {
757 cmd->status = BLK_STS_IOERR;
763 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
768 trace_nbd_payload_received(req, handle);
769 mutex_unlock(&cmd->lock);
770 return ret ? ERR_PTR(ret) : cmd;
773 static void recv_work(struct work_struct *work)
775 struct recv_thread_args *args = container_of(work,
776 struct recv_thread_args,
778 struct nbd_device *nbd = args->nbd;
779 struct nbd_config *config = nbd->config;
784 cmd = nbd_read_stat(nbd, args->index);
786 struct nbd_sock *nsock = config->socks[args->index];
788 mutex_lock(&nsock->tx_lock);
789 nbd_mark_nsock_dead(nbd, nsock, 1);
790 mutex_unlock(&nsock->tx_lock);
794 rq = blk_mq_rq_from_pdu(cmd);
795 if (likely(!blk_should_fake_timeout(rq->q)))
796 blk_mq_complete_request(rq);
799 atomic_dec(&config->recv_threads);
800 wake_up(&config->recv_wq);
804 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
806 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
808 mutex_lock(&cmd->lock);
809 cmd->status = BLK_STS_IOERR;
810 mutex_unlock(&cmd->lock);
812 blk_mq_complete_request(req);
816 static void nbd_clear_que(struct nbd_device *nbd)
818 blk_mq_quiesce_queue(nbd->disk->queue);
819 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
820 blk_mq_unquiesce_queue(nbd->disk->queue);
821 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
824 static int find_fallback(struct nbd_device *nbd, int index)
826 struct nbd_config *config = nbd->config;
828 struct nbd_sock *nsock = config->socks[index];
829 int fallback = nsock->fallback_index;
831 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
834 if (config->num_connections <= 1) {
835 dev_err_ratelimited(disk_to_dev(nbd->disk),
836 "Dead connection, failed to find a fallback\n");
840 if (fallback >= 0 && fallback < config->num_connections &&
841 !config->socks[fallback]->dead)
844 if (nsock->fallback_index < 0 ||
845 nsock->fallback_index >= config->num_connections ||
846 config->socks[nsock->fallback_index]->dead) {
848 for (i = 0; i < config->num_connections; i++) {
851 if (!config->socks[i]->dead) {
856 nsock->fallback_index = new_index;
858 dev_err_ratelimited(disk_to_dev(nbd->disk),
859 "Dead connection, failed to find a fallback\n");
863 new_index = nsock->fallback_index;
867 static int wait_for_reconnect(struct nbd_device *nbd)
869 struct nbd_config *config = nbd->config;
870 if (!config->dead_conn_timeout)
872 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
874 return wait_event_timeout(config->conn_wait,
875 atomic_read(&config->live_connections) > 0,
876 config->dead_conn_timeout) > 0;
879 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
881 struct request *req = blk_mq_rq_from_pdu(cmd);
882 struct nbd_device *nbd = cmd->nbd;
883 struct nbd_config *config;
884 struct nbd_sock *nsock;
887 if (!refcount_inc_not_zero(&nbd->config_refs)) {
888 dev_err_ratelimited(disk_to_dev(nbd->disk),
889 "Socks array is empty\n");
890 blk_mq_start_request(req);
893 config = nbd->config;
895 if (index >= config->num_connections) {
896 dev_err_ratelimited(disk_to_dev(nbd->disk),
897 "Attempted send on invalid socket\n");
899 blk_mq_start_request(req);
902 cmd->status = BLK_STS_OK;
904 nsock = config->socks[index];
905 mutex_lock(&nsock->tx_lock);
907 int old_index = index;
908 index = find_fallback(nbd, index);
909 mutex_unlock(&nsock->tx_lock);
911 if (wait_for_reconnect(nbd)) {
915 /* All the sockets should already be down at this point,
916 * we just want to make sure that DISCONNECTED is set so
917 * any requests that come in that were queue'ed waiting
918 * for the reconnect timer don't trigger the timer again
919 * and instead just error out.
923 blk_mq_start_request(req);
929 /* Handle the case that we have a pending request that was partially
930 * transmitted that _has_ to be serviced first. We need to call requeue
931 * here so that it gets put _after_ the request that is already on the
934 blk_mq_start_request(req);
935 if (unlikely(nsock->pending && nsock->pending != req)) {
936 nbd_requeue_cmd(cmd);
941 * Some failures are related to the link going down, so anything that
942 * returns EAGAIN can be retried on a different socket.
944 ret = nbd_send_cmd(nbd, cmd, index);
945 if (ret == -EAGAIN) {
946 dev_err_ratelimited(disk_to_dev(nbd->disk),
947 "Request send failed, requeueing\n");
948 nbd_mark_nsock_dead(nbd, nsock, 1);
949 nbd_requeue_cmd(cmd);
953 mutex_unlock(&nsock->tx_lock);
958 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
959 const struct blk_mq_queue_data *bd)
961 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
965 * Since we look at the bio's to send the request over the network we
966 * need to make sure the completion work doesn't mark this request done
967 * before we are done doing our send. This keeps us from dereferencing
968 * freed data if we have particularly fast completions (ie we get the
969 * completion before we exit sock_xmit on the last bvec) or in the case
970 * that the server is misbehaving (or there was an error) before we're
971 * done sending everything over the wire.
973 mutex_lock(&cmd->lock);
974 clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
976 /* We can be called directly from the user space process, which means we
977 * could possibly have signals pending so our sendmsg will fail. In
978 * this case we need to return that we are busy, otherwise error out as
981 ret = nbd_handle_cmd(cmd, hctx->queue_num);
986 mutex_unlock(&cmd->lock);
991 static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
997 sock = sockfd_lookup(fd, err);
1001 if (sock->ops->shutdown == sock_no_shutdown) {
1002 dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
1011 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1014 struct nbd_config *config = nbd->config;
1015 struct socket *sock;
1016 struct nbd_sock **socks;
1017 struct nbd_sock *nsock;
1020 sock = nbd_get_socket(nbd, arg, &err);
1025 * We need to make sure we don't get any errant requests while we're
1026 * reallocating the ->socks array.
1028 blk_mq_freeze_queue(nbd->disk->queue);
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);
1068 blk_mq_unfreeze_queue(nbd->disk->queue);
1073 blk_mq_unfreeze_queue(nbd->disk->queue);
1078 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1080 struct nbd_config *config = nbd->config;
1081 struct socket *sock, *old;
1082 struct recv_thread_args *args;
1086 sock = nbd_get_socket(nbd, arg, &err);
1090 args = kzalloc(sizeof(*args), GFP_KERNEL);
1096 for (i = 0; i < config->num_connections; i++) {
1097 struct nbd_sock *nsock = config->socks[i];
1102 mutex_lock(&nsock->tx_lock);
1104 mutex_unlock(&nsock->tx_lock);
1107 sk_set_memalloc(sock->sk);
1108 if (nbd->tag_set.timeout)
1109 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1110 atomic_inc(&config->recv_threads);
1111 refcount_inc(&nbd->config_refs);
1113 nsock->fallback_index = -1;
1115 nsock->dead = false;
1116 INIT_WORK(&args->work, recv_work);
1120 mutex_unlock(&nsock->tx_lock);
1123 clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1125 /* We take the tx_mutex in an error path in the recv_work, so we
1126 * need to queue_work outside of the tx_mutex.
1128 queue_work(nbd->recv_workq, &args->work);
1130 atomic_inc(&config->live_connections);
1131 wake_up(&config->conn_wait);
1139 static void nbd_bdev_reset(struct block_device *bdev)
1141 if (bdev->bd_openers > 1)
1143 set_capacity(bdev->bd_disk, 0);
1146 static void nbd_parse_flags(struct nbd_device *nbd)
1148 struct nbd_config *config = nbd->config;
1149 if (config->flags & NBD_FLAG_READ_ONLY)
1150 set_disk_ro(nbd->disk, true);
1152 set_disk_ro(nbd->disk, false);
1153 if (config->flags & NBD_FLAG_SEND_TRIM)
1154 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1155 if (config->flags & NBD_FLAG_SEND_FLUSH) {
1156 if (config->flags & NBD_FLAG_SEND_FUA)
1157 blk_queue_write_cache(nbd->disk->queue, true, true);
1159 blk_queue_write_cache(nbd->disk->queue, true, false);
1162 blk_queue_write_cache(nbd->disk->queue, false, false);
1165 static void send_disconnects(struct nbd_device *nbd)
1167 struct nbd_config *config = nbd->config;
1168 struct nbd_request request = {
1169 .magic = htonl(NBD_REQUEST_MAGIC),
1170 .type = htonl(NBD_CMD_DISC),
1172 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1173 struct iov_iter from;
1176 for (i = 0; i < config->num_connections; i++) {
1177 struct nbd_sock *nsock = config->socks[i];
1179 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1180 mutex_lock(&nsock->tx_lock);
1181 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1183 dev_err(disk_to_dev(nbd->disk),
1184 "Send disconnect failed %d\n", ret);
1185 mutex_unlock(&nsock->tx_lock);
1189 static int nbd_disconnect(struct nbd_device *nbd)
1191 struct nbd_config *config = nbd->config;
1193 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1194 set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1195 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1196 send_disconnects(nbd);
1200 static void nbd_clear_sock(struct nbd_device *nbd)
1204 nbd->task_setup = NULL;
1207 static void nbd_config_put(struct nbd_device *nbd)
1209 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1210 &nbd->config_lock)) {
1211 struct nbd_config *config = nbd->config;
1212 nbd_dev_dbg_close(nbd);
1213 nbd_size_clear(nbd);
1214 if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1215 &config->runtime_flags))
1216 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1217 nbd->task_recv = NULL;
1218 nbd_clear_sock(nbd);
1219 if (config->num_connections) {
1221 for (i = 0; i < config->num_connections; i++) {
1222 sockfd_put(config->socks[i]->sock);
1223 kfree(config->socks[i]);
1225 kfree(config->socks);
1230 if (nbd->recv_workq)
1231 destroy_workqueue(nbd->recv_workq);
1232 nbd->recv_workq = NULL;
1234 nbd->tag_set.timeout = 0;
1235 nbd->disk->queue->limits.discard_granularity = 0;
1236 nbd->disk->queue->limits.discard_alignment = 0;
1237 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1238 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1240 mutex_unlock(&nbd->config_lock);
1242 module_put(THIS_MODULE);
1246 static int nbd_start_device(struct nbd_device *nbd)
1248 struct nbd_config *config = nbd->config;
1249 int num_connections = config->num_connections;
1256 if (num_connections > 1 &&
1257 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1258 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1262 nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1263 WQ_MEM_RECLAIM | WQ_HIGHPRI |
1264 WQ_UNBOUND, 0, nbd->index);
1265 if (!nbd->recv_workq) {
1266 dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1270 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1271 nbd->task_recv = current;
1273 nbd_parse_flags(nbd);
1275 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1277 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1280 set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1282 nbd_dev_dbg_init(nbd);
1283 for (i = 0; i < num_connections; i++) {
1284 struct recv_thread_args *args;
1286 args = kzalloc(sizeof(*args), GFP_KERNEL);
1290 * If num_connections is m (2 < m),
1291 * and NO.1 ~ NO.n(1 < n < m) kzallocs are successful.
1292 * But NO.(n + 1) failed. We still have n recv threads.
1293 * So, add flush_workqueue here to prevent recv threads
1294 * dropping the last config_refs and trying to destroy
1295 * the workqueue from inside the workqueue.
1298 flush_workqueue(nbd->recv_workq);
1301 sk_set_memalloc(config->socks[i]->sock->sk);
1302 if (nbd->tag_set.timeout)
1303 config->socks[i]->sock->sk->sk_sndtimeo =
1304 nbd->tag_set.timeout;
1305 atomic_inc(&config->recv_threads);
1306 refcount_inc(&nbd->config_refs);
1307 INIT_WORK(&args->work, recv_work);
1310 queue_work(nbd->recv_workq, &args->work);
1312 return nbd_set_size(nbd, config->bytesize, config->blksize);
1315 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1317 struct nbd_config *config = nbd->config;
1320 ret = nbd_start_device(nbd);
1325 set_bit(GD_NEED_PART_SCAN, &nbd->disk->state);
1326 mutex_unlock(&nbd->config_lock);
1327 ret = wait_event_interruptible(config->recv_wq,
1328 atomic_read(&config->recv_threads) == 0);
1331 flush_workqueue(nbd->recv_workq);
1333 mutex_lock(&nbd->config_lock);
1334 nbd_bdev_reset(bdev);
1335 /* user requested, ignore socket errors */
1336 if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1338 if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1343 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1344 struct block_device *bdev)
1347 __invalidate_device(bdev, true);
1348 nbd_bdev_reset(bdev);
1349 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1350 &nbd->config->runtime_flags))
1351 nbd_config_put(nbd);
1354 static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1356 nbd->tag_set.timeout = timeout * HZ;
1358 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1360 blk_queue_rq_timeout(nbd->disk->queue, 30 * HZ);
1363 /* Must be called with config_lock held */
1364 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1365 unsigned int cmd, unsigned long arg)
1367 struct nbd_config *config = nbd->config;
1370 case NBD_DISCONNECT:
1371 return nbd_disconnect(nbd);
1372 case NBD_CLEAR_SOCK:
1373 nbd_clear_sock_ioctl(nbd, bdev);
1376 return nbd_add_socket(nbd, arg, false);
1377 case NBD_SET_BLKSIZE:
1378 return nbd_set_size(nbd, config->bytesize, arg);
1380 return nbd_set_size(nbd, arg, config->blksize);
1381 case NBD_SET_SIZE_BLOCKS:
1382 return nbd_set_size(nbd, arg * config->blksize,
1384 case NBD_SET_TIMEOUT:
1385 nbd_set_cmd_timeout(nbd, arg);
1389 config->flags = arg;
1392 return nbd_start_device_ioctl(nbd, bdev);
1395 * This is for compatibility only. The queue is always cleared
1396 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1399 case NBD_PRINT_DEBUG:
1401 * For compatibility only, we no longer keep a list of
1402 * outstanding requests.
1409 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1410 unsigned int cmd, unsigned long arg)
1412 struct nbd_device *nbd = bdev->bd_disk->private_data;
1413 struct nbd_config *config = nbd->config;
1414 int error = -EINVAL;
1416 if (!capable(CAP_SYS_ADMIN))
1419 /* The block layer will pass back some non-nbd ioctls in case we have
1420 * special handling for them, but we don't so just return an error.
1422 if (_IOC_TYPE(cmd) != 0xab)
1425 mutex_lock(&nbd->config_lock);
1427 /* Don't allow ioctl operations on a nbd device that was created with
1428 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1430 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1431 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1432 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1434 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1435 mutex_unlock(&nbd->config_lock);
1439 static struct nbd_config *nbd_alloc_config(void)
1441 struct nbd_config *config;
1443 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1446 atomic_set(&config->recv_threads, 0);
1447 init_waitqueue_head(&config->recv_wq);
1448 init_waitqueue_head(&config->conn_wait);
1449 config->blksize = NBD_DEF_BLKSIZE;
1450 atomic_set(&config->live_connections, 0);
1451 try_module_get(THIS_MODULE);
1455 static int nbd_open(struct block_device *bdev, fmode_t mode)
1457 struct nbd_device *nbd;
1460 mutex_lock(&nbd_index_mutex);
1461 nbd = bdev->bd_disk->private_data;
1466 if (!refcount_inc_not_zero(&nbd->refs)) {
1470 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1471 struct nbd_config *config;
1473 mutex_lock(&nbd->config_lock);
1474 if (refcount_inc_not_zero(&nbd->config_refs)) {
1475 mutex_unlock(&nbd->config_lock);
1478 config = nbd->config = nbd_alloc_config();
1481 mutex_unlock(&nbd->config_lock);
1484 refcount_set(&nbd->config_refs, 1);
1485 refcount_inc(&nbd->refs);
1486 mutex_unlock(&nbd->config_lock);
1488 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1489 } else if (nbd_disconnected(nbd->config)) {
1491 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1494 mutex_unlock(&nbd_index_mutex);
1498 static void nbd_release(struct gendisk *disk, fmode_t mode)
1500 struct nbd_device *nbd = disk->private_data;
1502 if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1503 disk->part0->bd_openers == 0)
1504 nbd_disconnect_and_put(nbd);
1506 nbd_config_put(nbd);
1510 static const struct block_device_operations nbd_fops =
1512 .owner = THIS_MODULE,
1514 .release = nbd_release,
1516 .compat_ioctl = nbd_ioctl,
1519 #if IS_ENABLED(CONFIG_DEBUG_FS)
1521 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1523 struct nbd_device *nbd = s->private;
1526 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1531 DEFINE_SHOW_ATTRIBUTE(nbd_dbg_tasks);
1533 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1535 struct nbd_device *nbd = s->private;
1536 u32 flags = nbd->config->flags;
1538 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1540 seq_puts(s, "Known flags:\n");
1542 if (flags & NBD_FLAG_HAS_FLAGS)
1543 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1544 if (flags & NBD_FLAG_READ_ONLY)
1545 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1546 if (flags & NBD_FLAG_SEND_FLUSH)
1547 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1548 if (flags & NBD_FLAG_SEND_FUA)
1549 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1550 if (flags & NBD_FLAG_SEND_TRIM)
1551 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1556 DEFINE_SHOW_ATTRIBUTE(nbd_dbg_flags);
1558 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1561 struct nbd_config *config = nbd->config;
1566 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1568 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1572 config->dbg_dir = dir;
1574 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_fops);
1575 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1576 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1577 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1578 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_fops);
1583 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1585 debugfs_remove_recursive(nbd->config->dbg_dir);
1588 static int nbd_dbg_init(void)
1590 struct dentry *dbg_dir;
1592 dbg_dir = debugfs_create_dir("nbd", NULL);
1596 nbd_dbg_dir = dbg_dir;
1601 static void nbd_dbg_close(void)
1603 debugfs_remove_recursive(nbd_dbg_dir);
1606 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1608 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1613 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1617 static int nbd_dbg_init(void)
1622 static void nbd_dbg_close(void)
1628 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1629 unsigned int hctx_idx, unsigned int numa_node)
1631 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1632 cmd->nbd = set->driver_data;
1634 mutex_init(&cmd->lock);
1638 static const struct blk_mq_ops nbd_mq_ops = {
1639 .queue_rq = nbd_queue_rq,
1640 .complete = nbd_complete_rq,
1641 .init_request = nbd_init_request,
1642 .timeout = nbd_xmit_timeout,
1645 static int nbd_dev_add(int index)
1647 struct nbd_device *nbd;
1648 struct gendisk *disk;
1649 struct request_queue *q;
1652 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1656 disk = alloc_disk(1 << part_shift);
1661 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1666 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1675 nbd->tag_set.ops = &nbd_mq_ops;
1676 nbd->tag_set.nr_hw_queues = 1;
1677 nbd->tag_set.queue_depth = 128;
1678 nbd->tag_set.numa_node = NUMA_NO_NODE;
1679 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1680 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1682 nbd->tag_set.driver_data = nbd;
1683 nbd->destroy_complete = NULL;
1685 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1689 q = blk_mq_init_queue(&nbd->tag_set);
1697 * Tell the block layer that we are not a rotational device
1699 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1700 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1701 disk->queue->limits.discard_granularity = 0;
1702 disk->queue->limits.discard_alignment = 0;
1703 blk_queue_max_discard_sectors(disk->queue, 0);
1704 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1705 blk_queue_max_segments(disk->queue, USHRT_MAX);
1706 blk_queue_max_hw_sectors(disk->queue, 65536);
1707 disk->queue->limits.max_sectors = 256;
1709 mutex_init(&nbd->config_lock);
1710 refcount_set(&nbd->config_refs, 0);
1711 refcount_set(&nbd->refs, 1);
1712 INIT_LIST_HEAD(&nbd->list);
1713 disk->major = NBD_MAJOR;
1714 disk->first_minor = index << part_shift;
1715 disk->fops = &nbd_fops;
1716 disk->private_data = nbd;
1717 sprintf(disk->disk_name, "nbd%d", index);
1719 nbd_total_devices++;
1723 blk_mq_free_tag_set(&nbd->tag_set);
1725 idr_remove(&nbd_index_idr, index);
1734 static int find_free_cb(int id, void *ptr, void *data)
1736 struct nbd_device *nbd = ptr;
1737 struct nbd_device **found = data;
1739 if (!refcount_read(&nbd->config_refs)) {
1746 /* Netlink interface. */
1747 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1748 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1749 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1750 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1751 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1752 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1753 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1754 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1755 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1756 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1759 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1760 [NBD_SOCK_FD] = { .type = NLA_U32 },
1763 /* We don't use this right now since we don't parse the incoming list, but we
1764 * still want it here so userspace knows what to expect.
1766 static const struct nla_policy __attribute__((unused))
1767 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1768 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1769 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1772 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1774 struct nbd_config *config = nbd->config;
1775 u64 bsize = config->blksize;
1776 u64 bytes = config->bytesize;
1778 if (info->attrs[NBD_ATTR_SIZE_BYTES])
1779 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1781 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES])
1782 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1784 if (bytes != config->bytesize || bsize != config->blksize)
1785 return nbd_set_size(nbd, bytes, bsize);
1789 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1791 DECLARE_COMPLETION_ONSTACK(destroy_complete);
1792 struct nbd_device *nbd = NULL;
1793 struct nbd_config *config;
1796 bool put_dev = false;
1798 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1801 if (info->attrs[NBD_ATTR_INDEX])
1802 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1803 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1804 printk(KERN_ERR "nbd: must specify at least one socket\n");
1807 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1808 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1812 mutex_lock(&nbd_index_mutex);
1814 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1817 new_index = nbd_dev_add(-1);
1818 if (new_index < 0) {
1819 mutex_unlock(&nbd_index_mutex);
1820 printk(KERN_ERR "nbd: failed to add new device\n");
1823 nbd = idr_find(&nbd_index_idr, new_index);
1826 nbd = idr_find(&nbd_index_idr, index);
1828 ret = nbd_dev_add(index);
1830 mutex_unlock(&nbd_index_mutex);
1831 printk(KERN_ERR "nbd: failed to add new device\n");
1834 nbd = idr_find(&nbd_index_idr, index);
1838 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1840 mutex_unlock(&nbd_index_mutex);
1844 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1845 test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1846 nbd->destroy_complete = &destroy_complete;
1847 mutex_unlock(&nbd_index_mutex);
1849 /* Wait untill the the nbd stuff is totally destroyed */
1850 wait_for_completion(&destroy_complete);
1854 if (!refcount_inc_not_zero(&nbd->refs)) {
1855 mutex_unlock(&nbd_index_mutex);
1858 printk(KERN_ERR "nbd: device at index %d is going down\n",
1862 mutex_unlock(&nbd_index_mutex);
1864 mutex_lock(&nbd->config_lock);
1865 if (refcount_read(&nbd->config_refs)) {
1866 mutex_unlock(&nbd->config_lock);
1870 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1873 if (WARN_ON(nbd->config)) {
1874 mutex_unlock(&nbd->config_lock);
1878 config = nbd->config = nbd_alloc_config();
1880 mutex_unlock(&nbd->config_lock);
1882 printk(KERN_ERR "nbd: couldn't allocate config\n");
1885 refcount_set(&nbd->config_refs, 1);
1886 set_bit(NBD_RT_BOUND, &config->runtime_flags);
1888 ret = nbd_genl_size_set(info, nbd);
1892 if (info->attrs[NBD_ATTR_TIMEOUT])
1893 nbd_set_cmd_timeout(nbd,
1894 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1895 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1896 config->dead_conn_timeout =
1897 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1898 config->dead_conn_timeout *= HZ;
1900 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1902 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1903 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1904 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1905 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1907 * We have 1 ref to keep the device around, and then 1
1908 * ref for our current operation here, which will be
1909 * inherited by the config. If we already have
1910 * DESTROY_ON_DISCONNECT set then we know we don't have
1911 * that extra ref already held so we don't need the
1914 if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1918 if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1920 refcount_inc(&nbd->refs);
1922 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1923 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1924 &config->runtime_flags);
1928 if (info->attrs[NBD_ATTR_SOCKETS]) {
1929 struct nlattr *attr;
1932 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1934 struct nlattr *socks[NBD_SOCK_MAX+1];
1936 if (nla_type(attr) != NBD_SOCK_ITEM) {
1937 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1941 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1946 printk(KERN_ERR "nbd: error processing sock list\n");
1950 if (!socks[NBD_SOCK_FD])
1952 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1953 ret = nbd_add_socket(nbd, fd, true);
1958 ret = nbd_start_device(nbd);
1960 mutex_unlock(&nbd->config_lock);
1962 set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1963 refcount_inc(&nbd->config_refs);
1964 nbd_connect_reply(info, nbd->index);
1966 nbd_config_put(nbd);
1972 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1974 mutex_lock(&nbd->config_lock);
1975 nbd_disconnect(nbd);
1976 nbd_clear_sock(nbd);
1977 mutex_unlock(&nbd->config_lock);
1979 * Make sure recv thread has finished, so it does not drop the last
1980 * config ref and try to destroy the workqueue from inside the work
1983 if (nbd->recv_workq)
1984 flush_workqueue(nbd->recv_workq);
1985 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1986 &nbd->config->runtime_flags))
1987 nbd_config_put(nbd);
1990 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1992 struct nbd_device *nbd;
1995 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1998 if (!info->attrs[NBD_ATTR_INDEX]) {
1999 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
2002 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2003 mutex_lock(&nbd_index_mutex);
2004 nbd = idr_find(&nbd_index_idr, index);
2006 mutex_unlock(&nbd_index_mutex);
2007 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2011 if (!refcount_inc_not_zero(&nbd->refs)) {
2012 mutex_unlock(&nbd_index_mutex);
2013 printk(KERN_ERR "nbd: device at index %d is going down\n",
2017 mutex_unlock(&nbd_index_mutex);
2018 if (!refcount_inc_not_zero(&nbd->config_refs))
2020 nbd_disconnect_and_put(nbd);
2021 nbd_config_put(nbd);
2027 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2029 struct nbd_device *nbd = NULL;
2030 struct nbd_config *config;
2033 bool put_dev = false;
2035 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2038 if (!info->attrs[NBD_ATTR_INDEX]) {
2039 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2042 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2043 mutex_lock(&nbd_index_mutex);
2044 nbd = idr_find(&nbd_index_idr, index);
2046 mutex_unlock(&nbd_index_mutex);
2047 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2051 if (!refcount_inc_not_zero(&nbd->refs)) {
2052 mutex_unlock(&nbd_index_mutex);
2053 printk(KERN_ERR "nbd: device at index %d is going down\n",
2057 mutex_unlock(&nbd_index_mutex);
2059 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2060 dev_err(nbd_to_dev(nbd),
2061 "not configured, cannot reconfigure\n");
2066 mutex_lock(&nbd->config_lock);
2067 config = nbd->config;
2068 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2070 dev_err(nbd_to_dev(nbd),
2071 "not configured, cannot reconfigure\n");
2076 ret = nbd_genl_size_set(info, nbd);
2080 if (info->attrs[NBD_ATTR_TIMEOUT])
2081 nbd_set_cmd_timeout(nbd,
2082 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2083 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2084 config->dead_conn_timeout =
2085 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2086 config->dead_conn_timeout *= HZ;
2088 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2089 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2090 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2091 if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
2095 if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
2097 refcount_inc(&nbd->refs);
2100 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2101 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2102 &config->runtime_flags);
2104 clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2105 &config->runtime_flags);
2109 if (info->attrs[NBD_ATTR_SOCKETS]) {
2110 struct nlattr *attr;
2113 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2115 struct nlattr *socks[NBD_SOCK_MAX+1];
2117 if (nla_type(attr) != NBD_SOCK_ITEM) {
2118 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2122 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2127 printk(KERN_ERR "nbd: error processing sock list\n");
2131 if (!socks[NBD_SOCK_FD])
2133 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2134 ret = nbd_reconnect_socket(nbd, fd);
2140 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2144 mutex_unlock(&nbd->config_lock);
2145 nbd_config_put(nbd);
2152 static const struct genl_small_ops nbd_connect_genl_ops[] = {
2154 .cmd = NBD_CMD_CONNECT,
2155 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2156 .doit = nbd_genl_connect,
2159 .cmd = NBD_CMD_DISCONNECT,
2160 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2161 .doit = nbd_genl_disconnect,
2164 .cmd = NBD_CMD_RECONFIGURE,
2165 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2166 .doit = nbd_genl_reconfigure,
2169 .cmd = NBD_CMD_STATUS,
2170 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2171 .doit = nbd_genl_status,
2175 static const struct genl_multicast_group nbd_mcast_grps[] = {
2176 { .name = NBD_GENL_MCAST_GROUP_NAME, },
2179 static struct genl_family nbd_genl_family __ro_after_init = {
2181 .name = NBD_GENL_FAMILY_NAME,
2182 .version = NBD_GENL_VERSION,
2183 .module = THIS_MODULE,
2184 .small_ops = nbd_connect_genl_ops,
2185 .n_small_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2186 .maxattr = NBD_ATTR_MAX,
2187 .policy = nbd_attr_policy,
2188 .mcgrps = nbd_mcast_grps,
2189 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2192 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2194 struct nlattr *dev_opt;
2198 /* This is a little racey, but for status it's ok. The
2199 * reason we don't take a ref here is because we can't
2200 * take a ref in the index == -1 case as we would need
2201 * to put under the nbd_index_mutex, which could
2202 * deadlock if we are configured to remove ourselves
2203 * once we're disconnected.
2205 if (refcount_read(&nbd->config_refs))
2207 dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2210 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2213 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2217 nla_nest_end(reply, dev_opt);
2221 static int status_cb(int id, void *ptr, void *data)
2223 struct nbd_device *nbd = ptr;
2224 return populate_nbd_status(nbd, (struct sk_buff *)data);
2227 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2229 struct nlattr *dev_list;
2230 struct sk_buff *reply;
2236 if (info->attrs[NBD_ATTR_INDEX])
2237 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2239 mutex_lock(&nbd_index_mutex);
2241 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2242 nla_attr_size(sizeof(u8)));
2243 msg_size *= (index == -1) ? nbd_total_devices : 1;
2245 reply = genlmsg_new(msg_size, GFP_KERNEL);
2248 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2255 dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2257 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2263 struct nbd_device *nbd;
2264 nbd = idr_find(&nbd_index_idr, index);
2266 ret = populate_nbd_status(nbd, reply);
2273 nla_nest_end(reply, dev_list);
2274 genlmsg_end(reply, reply_head);
2275 ret = genlmsg_reply(reply, info);
2277 mutex_unlock(&nbd_index_mutex);
2281 static void nbd_connect_reply(struct genl_info *info, int index)
2283 struct sk_buff *skb;
2287 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2290 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2296 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2301 genlmsg_end(skb, msg_head);
2302 genlmsg_reply(skb, info);
2305 static void nbd_mcast_index(int index)
2307 struct sk_buff *skb;
2311 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2314 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2320 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2325 genlmsg_end(skb, msg_head);
2326 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2329 static void nbd_dead_link_work(struct work_struct *work)
2331 struct link_dead_args *args = container_of(work, struct link_dead_args,
2333 nbd_mcast_index(args->index);
2337 static int __init nbd_init(void)
2341 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2344 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2350 part_shift = fls(max_part);
2353 * Adjust max_part according to part_shift as it is exported
2354 * to user space so that user can know the max number of
2355 * partition kernel should be able to manage.
2357 * Note that -1 is required because partition 0 is reserved
2358 * for the whole disk.
2360 max_part = (1UL << part_shift) - 1;
2363 if ((1UL << part_shift) > DISK_MAX_PARTS)
2366 if (nbds_max > 1UL << (MINORBITS - part_shift))
2369 if (register_blkdev(NBD_MAJOR, "nbd"))
2372 if (genl_register_family(&nbd_genl_family)) {
2373 unregister_blkdev(NBD_MAJOR, "nbd");
2378 mutex_lock(&nbd_index_mutex);
2379 for (i = 0; i < nbds_max; i++)
2381 mutex_unlock(&nbd_index_mutex);
2385 static int nbd_exit_cb(int id, void *ptr, void *data)
2387 struct list_head *list = (struct list_head *)data;
2388 struct nbd_device *nbd = ptr;
2390 list_add_tail(&nbd->list, list);
2394 static void __exit nbd_cleanup(void)
2396 struct nbd_device *nbd;
2397 LIST_HEAD(del_list);
2401 mutex_lock(&nbd_index_mutex);
2402 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2403 mutex_unlock(&nbd_index_mutex);
2405 while (!list_empty(&del_list)) {
2406 nbd = list_first_entry(&del_list, struct nbd_device, list);
2407 list_del_init(&nbd->list);
2408 if (refcount_read(&nbd->refs) != 1)
2409 printk(KERN_ERR "nbd: possibly leaking a device\n");
2413 idr_destroy(&nbd_index_idr);
2414 genl_unregister_family(&nbd_genl_family);
2415 unregister_blkdev(NBD_MAJOR, "nbd");
2418 module_init(nbd_init);
2419 module_exit(nbd_cleanup);
2421 MODULE_DESCRIPTION("Network Block Device");
2422 MODULE_LICENSE("GPL");
2424 module_param(nbds_max, int, 0444);
2425 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2426 module_param(max_part, int, 0444);
2427 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");