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);
251 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 (!refcount_inc_not_zero(&nbd->config_refs)) {
389 cmd->status = BLK_STS_TIMEOUT;
392 config = nbd->config;
394 if (!mutex_trylock(&cmd->lock)) {
396 return BLK_EH_RESET_TIMER;
399 if (config->num_connections > 1) {
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.
408 if (config->socks && config->num_connections > 1) {
409 if (cmd->index < config->num_connections) {
410 struct nbd_sock *nsock =
411 config->socks[cmd->index];
412 mutex_lock(&nsock->tx_lock);
413 /* We can have multiple outstanding requests, so
414 * we don't want to mark the nsock dead if we've
415 * already reconnected with a new socket, so
416 * only mark it dead if its the same socket we
419 if (cmd->cookie == nsock->cookie)
420 nbd_mark_nsock_dead(nbd, nsock, 1);
421 mutex_unlock(&nsock->tx_lock);
423 mutex_unlock(&cmd->lock);
424 nbd_requeue_cmd(cmd);
430 if (!nbd->tag_set.timeout) {
432 * Userspace sets timeout=0 to disable socket disconnection,
433 * so just warn and reset the timer.
436 dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
437 req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
438 (unsigned long long)blk_rq_pos(req) << 9,
439 blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
441 mutex_unlock(&cmd->lock);
443 return BLK_EH_RESET_TIMER;
446 dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
447 set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
448 cmd->status = BLK_STS_IOERR;
449 mutex_unlock(&cmd->lock);
453 blk_mq_complete_request(req);
458 * Send or receive packet.
460 static int sock_xmit(struct nbd_device *nbd, int index, int send,
461 struct iov_iter *iter, int msg_flags, int *sent)
463 struct nbd_config *config = nbd->config;
464 struct socket *sock = config->socks[index]->sock;
467 unsigned int noreclaim_flag;
469 if (unlikely(!sock)) {
470 dev_err_ratelimited(disk_to_dev(nbd->disk),
471 "Attempted %s on closed socket in sock_xmit\n",
472 (send ? "send" : "recv"));
476 msg.msg_iter = *iter;
478 noreclaim_flag = memalloc_noreclaim_save();
480 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
483 msg.msg_control = NULL;
484 msg.msg_controllen = 0;
485 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
488 result = sock_sendmsg(sock, &msg);
490 result = sock_recvmsg(sock, &msg, msg.msg_flags);
494 result = -EPIPE; /* short read */
499 } while (msg_data_left(&msg));
501 memalloc_noreclaim_restore(noreclaim_flag);
507 * Different settings for sk->sk_sndtimeo can result in different return values
508 * if there is a signal pending when we enter sendmsg, because reasons?
510 static inline int was_interrupted(int result)
512 return result == -ERESTARTSYS || result == -EINTR;
515 /* always call with the tx_lock held */
516 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
518 struct request *req = blk_mq_rq_from_pdu(cmd);
519 struct nbd_config *config = nbd->config;
520 struct nbd_sock *nsock = config->socks[index];
522 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
523 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
524 struct iov_iter from;
525 unsigned long size = blk_rq_bytes(req);
529 u32 nbd_cmd_flags = 0;
530 int sent = nsock->sent, skip = 0;
532 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
534 type = req_to_nbd_cmd_type(req);
538 if (rq_data_dir(req) == WRITE &&
539 (config->flags & NBD_FLAG_READ_ONLY)) {
540 dev_err_ratelimited(disk_to_dev(nbd->disk),
541 "Write on read-only\n");
545 if (req->cmd_flags & REQ_FUA)
546 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
548 /* We did a partial send previously, and we at least sent the whole
549 * request struct, so just go and send the rest of the pages in the
553 if (sent >= sizeof(request)) {
554 skip = sent - sizeof(request);
556 /* initialize handle for tracing purposes */
557 handle = nbd_cmd_handle(cmd);
561 iov_iter_advance(&from, sent);
566 cmd->cookie = nsock->cookie;
568 request.type = htonl(type | nbd_cmd_flags);
569 if (type != NBD_CMD_FLUSH) {
570 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
571 request.len = htonl(size);
573 handle = nbd_cmd_handle(cmd);
574 memcpy(request.handle, &handle, sizeof(handle));
576 trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
578 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
579 req, nbdcmd_to_ascii(type),
580 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
581 result = sock_xmit(nbd, index, 1, &from,
582 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
583 trace_nbd_header_sent(req, handle);
585 if (was_interrupted(result)) {
586 /* If we havne't sent anything we can just return BUSY,
587 * however if we have sent something we need to make
588 * sure we only allow this req to be sent until we are
592 nsock->pending = req;
595 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
596 return BLK_STS_RESOURCE;
598 dev_err_ratelimited(disk_to_dev(nbd->disk),
599 "Send control failed (result %d)\n", result);
603 if (type != NBD_CMD_WRITE)
608 struct bio *next = bio->bi_next;
609 struct bvec_iter iter;
612 bio_for_each_segment(bvec, bio, iter) {
613 bool is_last = !next && bio_iter_last(bvec, iter);
614 int flags = is_last ? 0 : MSG_MORE;
616 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
618 iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
620 if (skip >= iov_iter_count(&from)) {
621 skip -= iov_iter_count(&from);
624 iov_iter_advance(&from, skip);
627 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
629 if (was_interrupted(result)) {
630 /* We've already sent the header, we
631 * have no choice but to set pending and
634 nsock->pending = req;
636 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
637 return BLK_STS_RESOURCE;
639 dev_err(disk_to_dev(nbd->disk),
640 "Send data failed (result %d)\n",
645 * The completion might already have come in,
646 * so break for the last one instead of letting
647 * the iterator do it. This prevents use-after-free
656 trace_nbd_payload_sent(req, handle);
657 nsock->pending = NULL;
662 /* NULL returned = something went wrong, inform userspace */
663 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
665 struct nbd_config *config = nbd->config;
667 struct nbd_reply reply;
669 struct request *req = NULL;
673 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
678 iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
679 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
681 if (!nbd_disconnected(config))
682 dev_err(disk_to_dev(nbd->disk),
683 "Receive control failed (result %d)\n", result);
684 return ERR_PTR(result);
687 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
688 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
689 (unsigned long)ntohl(reply.magic));
690 return ERR_PTR(-EPROTO);
693 memcpy(&handle, reply.handle, sizeof(handle));
694 tag = nbd_handle_to_tag(handle);
695 hwq = blk_mq_unique_tag_to_hwq(tag);
696 if (hwq < nbd->tag_set.nr_hw_queues)
697 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
698 blk_mq_unique_tag_to_tag(tag));
699 if (!req || !blk_mq_request_started(req)) {
700 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
702 return ERR_PTR(-ENOENT);
704 trace_nbd_header_received(req, handle);
705 cmd = blk_mq_rq_to_pdu(req);
707 mutex_lock(&cmd->lock);
708 if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
709 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
710 req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
714 if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
715 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
720 if (ntohl(reply.error)) {
721 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
723 cmd->status = BLK_STS_IOERR;
727 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
728 if (rq_data_dir(req) != WRITE) {
729 struct req_iterator iter;
732 rq_for_each_segment(bvec, req, iter) {
733 iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
734 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
736 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
739 * If we've disconnected or we only have 1
740 * connection then we need to make sure we
741 * complete this request, otherwise error out
742 * and let the timeout stuff handle resubmitting
743 * this request onto another connection.
745 if (nbd_disconnected(config) ||
746 config->num_connections <= 1) {
747 cmd->status = BLK_STS_IOERR;
753 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
758 trace_nbd_payload_received(req, handle);
759 mutex_unlock(&cmd->lock);
760 return ret ? ERR_PTR(ret) : cmd;
763 static void recv_work(struct work_struct *work)
765 struct recv_thread_args *args = container_of(work,
766 struct recv_thread_args,
768 struct nbd_device *nbd = args->nbd;
769 struct nbd_config *config = nbd->config;
773 cmd = nbd_read_stat(nbd, args->index);
775 struct nbd_sock *nsock = config->socks[args->index];
777 mutex_lock(&nsock->tx_lock);
778 nbd_mark_nsock_dead(nbd, nsock, 1);
779 mutex_unlock(&nsock->tx_lock);
783 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
785 atomic_dec(&config->recv_threads);
786 wake_up(&config->recv_wq);
791 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
793 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
795 cmd->status = BLK_STS_IOERR;
796 blk_mq_complete_request(req);
800 static void nbd_clear_que(struct nbd_device *nbd)
802 blk_mq_quiesce_queue(nbd->disk->queue);
803 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
804 blk_mq_unquiesce_queue(nbd->disk->queue);
805 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
808 static int find_fallback(struct nbd_device *nbd, int index)
810 struct nbd_config *config = nbd->config;
812 struct nbd_sock *nsock = config->socks[index];
813 int fallback = nsock->fallback_index;
815 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
818 if (config->num_connections <= 1) {
819 dev_err_ratelimited(disk_to_dev(nbd->disk),
820 "Attempted send on invalid socket\n");
824 if (fallback >= 0 && fallback < config->num_connections &&
825 !config->socks[fallback]->dead)
828 if (nsock->fallback_index < 0 ||
829 nsock->fallback_index >= config->num_connections ||
830 config->socks[nsock->fallback_index]->dead) {
832 for (i = 0; i < config->num_connections; i++) {
835 if (!config->socks[i]->dead) {
840 nsock->fallback_index = new_index;
842 dev_err_ratelimited(disk_to_dev(nbd->disk),
843 "Dead connection, failed to find a fallback\n");
847 new_index = nsock->fallback_index;
851 static int wait_for_reconnect(struct nbd_device *nbd)
853 struct nbd_config *config = nbd->config;
854 if (!config->dead_conn_timeout)
856 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
858 return wait_event_timeout(config->conn_wait,
859 atomic_read(&config->live_connections) > 0,
860 config->dead_conn_timeout) > 0;
863 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
865 struct request *req = blk_mq_rq_from_pdu(cmd);
866 struct nbd_device *nbd = cmd->nbd;
867 struct nbd_config *config;
868 struct nbd_sock *nsock;
871 if (!refcount_inc_not_zero(&nbd->config_refs)) {
872 dev_err_ratelimited(disk_to_dev(nbd->disk),
873 "Socks array is empty\n");
874 blk_mq_start_request(req);
877 config = nbd->config;
879 if (index >= config->num_connections) {
880 dev_err_ratelimited(disk_to_dev(nbd->disk),
881 "Attempted send on invalid socket\n");
883 blk_mq_start_request(req);
886 cmd->status = BLK_STS_OK;
888 nsock = config->socks[index];
889 mutex_lock(&nsock->tx_lock);
891 int old_index = index;
892 index = find_fallback(nbd, index);
893 mutex_unlock(&nsock->tx_lock);
895 if (wait_for_reconnect(nbd)) {
899 /* All the sockets should already be down at this point,
900 * we just want to make sure that DISCONNECTED is set so
901 * any requests that come in that were queue'ed waiting
902 * for the reconnect timer don't trigger the timer again
903 * and instead just error out.
907 blk_mq_start_request(req);
913 /* Handle the case that we have a pending request that was partially
914 * transmitted that _has_ to be serviced first. We need to call requeue
915 * here so that it gets put _after_ the request that is already on the
918 blk_mq_start_request(req);
919 if (unlikely(nsock->pending && nsock->pending != req)) {
920 nbd_requeue_cmd(cmd);
925 * Some failures are related to the link going down, so anything that
926 * returns EAGAIN can be retried on a different socket.
928 ret = nbd_send_cmd(nbd, cmd, index);
929 if (ret == -EAGAIN) {
930 dev_err_ratelimited(disk_to_dev(nbd->disk),
931 "Request send failed, requeueing\n");
932 nbd_mark_nsock_dead(nbd, nsock, 1);
933 nbd_requeue_cmd(cmd);
937 mutex_unlock(&nsock->tx_lock);
942 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
943 const struct blk_mq_queue_data *bd)
945 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
949 * Since we look at the bio's to send the request over the network we
950 * need to make sure the completion work doesn't mark this request done
951 * before we are done doing our send. This keeps us from dereferencing
952 * freed data if we have particularly fast completions (ie we get the
953 * completion before we exit sock_xmit on the last bvec) or in the case
954 * that the server is misbehaving (or there was an error) before we're
955 * done sending everything over the wire.
957 mutex_lock(&cmd->lock);
958 clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
960 /* We can be called directly from the user space process, which means we
961 * could possibly have signals pending so our sendmsg will fail. In
962 * this case we need to return that we are busy, otherwise error out as
965 ret = nbd_handle_cmd(cmd, hctx->queue_num);
970 mutex_unlock(&cmd->lock);
975 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
978 struct nbd_config *config = nbd->config;
980 struct nbd_sock **socks;
981 struct nbd_sock *nsock;
984 sock = sockfd_lookup(arg, &err);
988 if (!netlink && !nbd->task_setup &&
989 !test_bit(NBD_RT_BOUND, &config->runtime_flags))
990 nbd->task_setup = current;
993 (nbd->task_setup != current ||
994 test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
995 dev_err(disk_to_dev(nbd->disk),
996 "Device being setup by another task");
1001 socks = krealloc(config->socks, (config->num_connections + 1) *
1002 sizeof(struct nbd_sock *), GFP_KERNEL);
1007 nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
1013 config->socks = socks;
1015 nsock->fallback_index = -1;
1016 nsock->dead = false;
1017 mutex_init(&nsock->tx_lock);
1019 nsock->pending = NULL;
1022 socks[config->num_connections++] = nsock;
1023 atomic_inc(&config->live_connections);
1028 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1030 struct nbd_config *config = nbd->config;
1031 struct socket *sock, *old;
1032 struct recv_thread_args *args;
1036 sock = sockfd_lookup(arg, &err);
1040 args = kzalloc(sizeof(*args), GFP_KERNEL);
1046 for (i = 0; i < config->num_connections; i++) {
1047 struct nbd_sock *nsock = config->socks[i];
1052 mutex_lock(&nsock->tx_lock);
1054 mutex_unlock(&nsock->tx_lock);
1057 sk_set_memalloc(sock->sk);
1058 if (nbd->tag_set.timeout)
1059 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1060 atomic_inc(&config->recv_threads);
1061 refcount_inc(&nbd->config_refs);
1063 nsock->fallback_index = -1;
1065 nsock->dead = false;
1066 INIT_WORK(&args->work, recv_work);
1070 mutex_unlock(&nsock->tx_lock);
1073 clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1075 /* We take the tx_mutex in an error path in the recv_work, so we
1076 * need to queue_work outside of the tx_mutex.
1078 queue_work(nbd->recv_workq, &args->work);
1080 atomic_inc(&config->live_connections);
1081 wake_up(&config->conn_wait);
1089 static void nbd_bdev_reset(struct block_device *bdev)
1091 if (bdev->bd_openers > 1)
1093 bd_set_size(bdev, 0);
1096 static void nbd_parse_flags(struct nbd_device *nbd)
1098 struct nbd_config *config = nbd->config;
1099 if (config->flags & NBD_FLAG_READ_ONLY)
1100 set_disk_ro(nbd->disk, true);
1102 set_disk_ro(nbd->disk, false);
1103 if (config->flags & NBD_FLAG_SEND_TRIM)
1104 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1105 if (config->flags & NBD_FLAG_SEND_FLUSH) {
1106 if (config->flags & NBD_FLAG_SEND_FUA)
1107 blk_queue_write_cache(nbd->disk->queue, true, true);
1109 blk_queue_write_cache(nbd->disk->queue, true, false);
1112 blk_queue_write_cache(nbd->disk->queue, false, false);
1115 static void send_disconnects(struct nbd_device *nbd)
1117 struct nbd_config *config = nbd->config;
1118 struct nbd_request request = {
1119 .magic = htonl(NBD_REQUEST_MAGIC),
1120 .type = htonl(NBD_CMD_DISC),
1122 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1123 struct iov_iter from;
1126 for (i = 0; i < config->num_connections; i++) {
1127 struct nbd_sock *nsock = config->socks[i];
1129 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1130 mutex_lock(&nsock->tx_lock);
1131 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1133 dev_err(disk_to_dev(nbd->disk),
1134 "Send disconnect failed %d\n", ret);
1135 mutex_unlock(&nsock->tx_lock);
1139 static int nbd_disconnect(struct nbd_device *nbd)
1141 struct nbd_config *config = nbd->config;
1143 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1144 set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1145 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1146 send_disconnects(nbd);
1150 static void nbd_clear_sock(struct nbd_device *nbd)
1154 nbd->task_setup = NULL;
1157 static void nbd_config_put(struct nbd_device *nbd)
1159 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1160 &nbd->config_lock)) {
1161 struct nbd_config *config = nbd->config;
1162 nbd_dev_dbg_close(nbd);
1163 nbd_size_clear(nbd);
1164 if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1165 &config->runtime_flags))
1166 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1167 nbd->task_recv = NULL;
1168 nbd_clear_sock(nbd);
1169 if (config->num_connections) {
1171 for (i = 0; i < config->num_connections; i++) {
1172 sockfd_put(config->socks[i]->sock);
1173 kfree(config->socks[i]);
1175 kfree(config->socks);
1180 if (nbd->recv_workq)
1181 destroy_workqueue(nbd->recv_workq);
1182 nbd->recv_workq = NULL;
1184 nbd->tag_set.timeout = 0;
1185 nbd->disk->queue->limits.discard_granularity = 0;
1186 nbd->disk->queue->limits.discard_alignment = 0;
1187 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1188 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1190 mutex_unlock(&nbd->config_lock);
1192 module_put(THIS_MODULE);
1196 static int nbd_start_device(struct nbd_device *nbd)
1198 struct nbd_config *config = nbd->config;
1199 int num_connections = config->num_connections;
1206 if (num_connections > 1 &&
1207 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1208 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1212 nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1213 WQ_MEM_RECLAIM | WQ_HIGHPRI |
1214 WQ_UNBOUND, 0, nbd->index);
1215 if (!nbd->recv_workq) {
1216 dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1220 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1221 nbd->task_recv = current;
1223 nbd_parse_flags(nbd);
1225 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1227 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1230 set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1232 nbd_dev_dbg_init(nbd);
1233 for (i = 0; i < num_connections; i++) {
1234 struct recv_thread_args *args;
1236 args = kzalloc(sizeof(*args), GFP_KERNEL);
1241 sk_set_memalloc(config->socks[i]->sock->sk);
1242 if (nbd->tag_set.timeout)
1243 config->socks[i]->sock->sk->sk_sndtimeo =
1244 nbd->tag_set.timeout;
1245 atomic_inc(&config->recv_threads);
1246 refcount_inc(&nbd->config_refs);
1247 INIT_WORK(&args->work, recv_work);
1250 queue_work(nbd->recv_workq, &args->work);
1252 nbd_size_update(nbd);
1256 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1258 struct nbd_config *config = nbd->config;
1261 ret = nbd_start_device(nbd);
1266 bdev->bd_invalidated = 1;
1267 mutex_unlock(&nbd->config_lock);
1268 ret = wait_event_interruptible(config->recv_wq,
1269 atomic_read(&config->recv_threads) == 0);
1272 flush_workqueue(nbd->recv_workq);
1274 mutex_lock(&nbd->config_lock);
1275 nbd_bdev_reset(bdev);
1276 /* user requested, ignore socket errors */
1277 if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1279 if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1284 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1285 struct block_device *bdev)
1288 __invalidate_device(bdev, true);
1289 nbd_bdev_reset(bdev);
1290 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1291 &nbd->config->runtime_flags))
1292 nbd_config_put(nbd);
1295 static bool nbd_is_valid_blksize(unsigned long blksize)
1297 if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1298 blksize > PAGE_SIZE)
1303 static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1305 nbd->tag_set.timeout = timeout * HZ;
1307 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1310 /* Must be called with config_lock held */
1311 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1312 unsigned int cmd, unsigned long arg)
1314 struct nbd_config *config = nbd->config;
1317 case NBD_DISCONNECT:
1318 return nbd_disconnect(nbd);
1319 case NBD_CLEAR_SOCK:
1320 nbd_clear_sock_ioctl(nbd, bdev);
1323 return nbd_add_socket(nbd, arg, false);
1324 case NBD_SET_BLKSIZE:
1326 arg = NBD_DEF_BLKSIZE;
1327 if (!nbd_is_valid_blksize(arg))
1329 nbd_size_set(nbd, arg,
1330 div_s64(config->bytesize, arg));
1333 nbd_size_set(nbd, config->blksize,
1334 div_s64(arg, config->blksize));
1336 case NBD_SET_SIZE_BLOCKS:
1337 nbd_size_set(nbd, config->blksize, arg);
1339 case NBD_SET_TIMEOUT:
1340 nbd_set_cmd_timeout(nbd, arg);
1344 config->flags = arg;
1347 return nbd_start_device_ioctl(nbd, bdev);
1350 * This is for compatibility only. The queue is always cleared
1351 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1354 case NBD_PRINT_DEBUG:
1356 * For compatibility only, we no longer keep a list of
1357 * outstanding requests.
1364 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1365 unsigned int cmd, unsigned long arg)
1367 struct nbd_device *nbd = bdev->bd_disk->private_data;
1368 struct nbd_config *config = nbd->config;
1369 int error = -EINVAL;
1371 if (!capable(CAP_SYS_ADMIN))
1374 /* The block layer will pass back some non-nbd ioctls in case we have
1375 * special handling for them, but we don't so just return an error.
1377 if (_IOC_TYPE(cmd) != 0xab)
1380 mutex_lock(&nbd->config_lock);
1382 /* Don't allow ioctl operations on a nbd device that was created with
1383 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1385 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1386 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1387 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1389 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1390 mutex_unlock(&nbd->config_lock);
1394 static struct nbd_config *nbd_alloc_config(void)
1396 struct nbd_config *config;
1398 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1401 atomic_set(&config->recv_threads, 0);
1402 init_waitqueue_head(&config->recv_wq);
1403 init_waitqueue_head(&config->conn_wait);
1404 config->blksize = NBD_DEF_BLKSIZE;
1405 atomic_set(&config->live_connections, 0);
1406 try_module_get(THIS_MODULE);
1410 static int nbd_open(struct block_device *bdev, fmode_t mode)
1412 struct nbd_device *nbd;
1415 mutex_lock(&nbd_index_mutex);
1416 nbd = bdev->bd_disk->private_data;
1421 if (!refcount_inc_not_zero(&nbd->refs)) {
1425 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1426 struct nbd_config *config;
1428 mutex_lock(&nbd->config_lock);
1429 if (refcount_inc_not_zero(&nbd->config_refs)) {
1430 mutex_unlock(&nbd->config_lock);
1433 config = nbd->config = nbd_alloc_config();
1436 mutex_unlock(&nbd->config_lock);
1439 refcount_set(&nbd->config_refs, 1);
1440 refcount_inc(&nbd->refs);
1441 mutex_unlock(&nbd->config_lock);
1442 bdev->bd_invalidated = 1;
1443 } else if (nbd_disconnected(nbd->config)) {
1444 bdev->bd_invalidated = 1;
1447 mutex_unlock(&nbd_index_mutex);
1451 static void nbd_release(struct gendisk *disk, fmode_t mode)
1453 struct nbd_device *nbd = disk->private_data;
1454 struct block_device *bdev = bdget_disk(disk, 0);
1456 if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1457 bdev->bd_openers == 0)
1458 nbd_disconnect_and_put(nbd);
1460 nbd_config_put(nbd);
1464 static const struct block_device_operations nbd_fops =
1466 .owner = THIS_MODULE,
1468 .release = nbd_release,
1470 .compat_ioctl = nbd_ioctl,
1473 #if IS_ENABLED(CONFIG_DEBUG_FS)
1475 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1477 struct nbd_device *nbd = s->private;
1480 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1485 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1487 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1490 static const struct file_operations nbd_dbg_tasks_ops = {
1491 .open = nbd_dbg_tasks_open,
1493 .llseek = seq_lseek,
1494 .release = single_release,
1497 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1499 struct nbd_device *nbd = s->private;
1500 u32 flags = nbd->config->flags;
1502 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1504 seq_puts(s, "Known flags:\n");
1506 if (flags & NBD_FLAG_HAS_FLAGS)
1507 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1508 if (flags & NBD_FLAG_READ_ONLY)
1509 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1510 if (flags & NBD_FLAG_SEND_FLUSH)
1511 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1512 if (flags & NBD_FLAG_SEND_FUA)
1513 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1514 if (flags & NBD_FLAG_SEND_TRIM)
1515 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1520 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1522 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1525 static const struct file_operations nbd_dbg_flags_ops = {
1526 .open = nbd_dbg_flags_open,
1528 .llseek = seq_lseek,
1529 .release = single_release,
1532 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1535 struct nbd_config *config = nbd->config;
1540 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1542 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1546 config->dbg_dir = dir;
1548 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1549 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1550 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1551 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1552 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1557 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1559 debugfs_remove_recursive(nbd->config->dbg_dir);
1562 static int nbd_dbg_init(void)
1564 struct dentry *dbg_dir;
1566 dbg_dir = debugfs_create_dir("nbd", NULL);
1570 nbd_dbg_dir = dbg_dir;
1575 static void nbd_dbg_close(void)
1577 debugfs_remove_recursive(nbd_dbg_dir);
1580 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1582 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1587 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1591 static int nbd_dbg_init(void)
1596 static void nbd_dbg_close(void)
1602 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1603 unsigned int hctx_idx, unsigned int numa_node)
1605 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1606 cmd->nbd = set->driver_data;
1608 mutex_init(&cmd->lock);
1612 static const struct blk_mq_ops nbd_mq_ops = {
1613 .queue_rq = nbd_queue_rq,
1614 .complete = nbd_complete_rq,
1615 .init_request = nbd_init_request,
1616 .timeout = nbd_xmit_timeout,
1619 static int nbd_dev_add(int index)
1621 struct nbd_device *nbd;
1622 struct gendisk *disk;
1623 struct request_queue *q;
1626 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1630 disk = alloc_disk(1 << part_shift);
1635 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1640 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1649 nbd->tag_set.ops = &nbd_mq_ops;
1650 nbd->tag_set.nr_hw_queues = 1;
1651 nbd->tag_set.queue_depth = 128;
1652 nbd->tag_set.numa_node = NUMA_NO_NODE;
1653 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1654 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1656 nbd->tag_set.driver_data = nbd;
1657 nbd->destroy_complete = NULL;
1659 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1663 q = blk_mq_init_queue(&nbd->tag_set);
1671 * Tell the block layer that we are not a rotational device
1673 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1674 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1675 disk->queue->limits.discard_granularity = 0;
1676 disk->queue->limits.discard_alignment = 0;
1677 blk_queue_max_discard_sectors(disk->queue, 0);
1678 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1679 blk_queue_max_segments(disk->queue, USHRT_MAX);
1680 blk_queue_max_hw_sectors(disk->queue, 65536);
1681 disk->queue->limits.max_sectors = 256;
1683 mutex_init(&nbd->config_lock);
1684 refcount_set(&nbd->config_refs, 0);
1685 refcount_set(&nbd->refs, 1);
1686 INIT_LIST_HEAD(&nbd->list);
1687 disk->major = NBD_MAJOR;
1688 disk->first_minor = index << part_shift;
1689 disk->fops = &nbd_fops;
1690 disk->private_data = nbd;
1691 sprintf(disk->disk_name, "nbd%d", index);
1693 nbd_total_devices++;
1697 blk_mq_free_tag_set(&nbd->tag_set);
1699 idr_remove(&nbd_index_idr, index);
1708 static int find_free_cb(int id, void *ptr, void *data)
1710 struct nbd_device *nbd = ptr;
1711 struct nbd_device **found = data;
1713 if (!refcount_read(&nbd->config_refs)) {
1720 /* Netlink interface. */
1721 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1722 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1723 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1724 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1725 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1726 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1727 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1728 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1729 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1730 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1733 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1734 [NBD_SOCK_FD] = { .type = NLA_U32 },
1737 /* We don't use this right now since we don't parse the incoming list, but we
1738 * still want it here so userspace knows what to expect.
1740 static const struct nla_policy __attribute__((unused))
1741 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1742 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1743 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1746 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1748 struct nbd_config *config = nbd->config;
1749 u64 bsize = config->blksize;
1750 u64 bytes = config->bytesize;
1752 if (info->attrs[NBD_ATTR_SIZE_BYTES])
1753 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1755 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1756 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1758 bsize = NBD_DEF_BLKSIZE;
1759 if (!nbd_is_valid_blksize(bsize)) {
1760 printk(KERN_ERR "Invalid block size %llu\n", bsize);
1765 if (bytes != config->bytesize || bsize != config->blksize)
1766 nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1770 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1772 DECLARE_COMPLETION_ONSTACK(destroy_complete);
1773 struct nbd_device *nbd = NULL;
1774 struct nbd_config *config;
1777 bool put_dev = false;
1779 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1782 if (info->attrs[NBD_ATTR_INDEX])
1783 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1784 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1785 printk(KERN_ERR "nbd: must specify at least one socket\n");
1788 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1789 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1793 mutex_lock(&nbd_index_mutex);
1795 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1798 new_index = nbd_dev_add(-1);
1799 if (new_index < 0) {
1800 mutex_unlock(&nbd_index_mutex);
1801 printk(KERN_ERR "nbd: failed to add new device\n");
1804 nbd = idr_find(&nbd_index_idr, new_index);
1807 nbd = idr_find(&nbd_index_idr, index);
1809 ret = nbd_dev_add(index);
1811 mutex_unlock(&nbd_index_mutex);
1812 printk(KERN_ERR "nbd: failed to add new device\n");
1815 nbd = idr_find(&nbd_index_idr, index);
1819 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1821 mutex_unlock(&nbd_index_mutex);
1825 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1826 test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1827 nbd->destroy_complete = &destroy_complete;
1828 mutex_unlock(&nbd_index_mutex);
1830 /* Wait untill the the nbd stuff is totally destroyed */
1831 wait_for_completion(&destroy_complete);
1835 if (!refcount_inc_not_zero(&nbd->refs)) {
1836 mutex_unlock(&nbd_index_mutex);
1839 printk(KERN_ERR "nbd: device at index %d is going down\n",
1843 mutex_unlock(&nbd_index_mutex);
1845 mutex_lock(&nbd->config_lock);
1846 if (refcount_read(&nbd->config_refs)) {
1847 mutex_unlock(&nbd->config_lock);
1851 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1854 if (WARN_ON(nbd->config)) {
1855 mutex_unlock(&nbd->config_lock);
1859 config = nbd->config = nbd_alloc_config();
1861 mutex_unlock(&nbd->config_lock);
1863 printk(KERN_ERR "nbd: couldn't allocate config\n");
1866 refcount_set(&nbd->config_refs, 1);
1867 set_bit(NBD_RT_BOUND, &config->runtime_flags);
1869 ret = nbd_genl_size_set(info, nbd);
1873 if (info->attrs[NBD_ATTR_TIMEOUT])
1874 nbd_set_cmd_timeout(nbd,
1875 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1876 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1877 config->dead_conn_timeout =
1878 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1879 config->dead_conn_timeout *= HZ;
1881 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1883 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1884 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1885 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1886 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1887 set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1888 &config->runtime_flags);
1889 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1892 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1894 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1895 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1896 &config->runtime_flags);
1900 if (info->attrs[NBD_ATTR_SOCKETS]) {
1901 struct nlattr *attr;
1904 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1906 struct nlattr *socks[NBD_SOCK_MAX+1];
1908 if (nla_type(attr) != NBD_SOCK_ITEM) {
1909 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1913 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1918 printk(KERN_ERR "nbd: error processing sock list\n");
1922 if (!socks[NBD_SOCK_FD])
1924 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1925 ret = nbd_add_socket(nbd, fd, true);
1930 ret = nbd_start_device(nbd);
1932 mutex_unlock(&nbd->config_lock);
1934 set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1935 refcount_inc(&nbd->config_refs);
1936 nbd_connect_reply(info, nbd->index);
1938 nbd_config_put(nbd);
1944 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1946 mutex_lock(&nbd->config_lock);
1947 nbd_disconnect(nbd);
1948 nbd_clear_sock(nbd);
1949 mutex_unlock(&nbd->config_lock);
1951 * Make sure recv thread has finished, so it does not drop the last
1952 * config ref and try to destroy the workqueue from inside the work
1955 flush_workqueue(nbd->recv_workq);
1956 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1957 &nbd->config->runtime_flags))
1958 nbd_config_put(nbd);
1961 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1963 struct nbd_device *nbd;
1966 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1969 if (!info->attrs[NBD_ATTR_INDEX]) {
1970 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1973 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1974 mutex_lock(&nbd_index_mutex);
1975 nbd = idr_find(&nbd_index_idr, index);
1977 mutex_unlock(&nbd_index_mutex);
1978 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1982 if (!refcount_inc_not_zero(&nbd->refs)) {
1983 mutex_unlock(&nbd_index_mutex);
1984 printk(KERN_ERR "nbd: device at index %d is going down\n",
1988 mutex_unlock(&nbd_index_mutex);
1989 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1993 nbd_disconnect_and_put(nbd);
1994 nbd_config_put(nbd);
1999 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2001 struct nbd_device *nbd = NULL;
2002 struct nbd_config *config;
2005 bool put_dev = false;
2007 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2010 if (!info->attrs[NBD_ATTR_INDEX]) {
2011 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2014 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2015 mutex_lock(&nbd_index_mutex);
2016 nbd = idr_find(&nbd_index_idr, index);
2018 mutex_unlock(&nbd_index_mutex);
2019 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2023 if (!refcount_inc_not_zero(&nbd->refs)) {
2024 mutex_unlock(&nbd_index_mutex);
2025 printk(KERN_ERR "nbd: device at index %d is going down\n",
2029 mutex_unlock(&nbd_index_mutex);
2031 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2032 dev_err(nbd_to_dev(nbd),
2033 "not configured, cannot reconfigure\n");
2038 mutex_lock(&nbd->config_lock);
2039 config = nbd->config;
2040 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2042 dev_err(nbd_to_dev(nbd),
2043 "not configured, cannot reconfigure\n");
2048 ret = nbd_genl_size_set(info, nbd);
2052 if (info->attrs[NBD_ATTR_TIMEOUT])
2053 nbd_set_cmd_timeout(nbd,
2054 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2055 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2056 config->dead_conn_timeout =
2057 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2058 config->dead_conn_timeout *= HZ;
2060 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2061 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2062 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2063 if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2064 &config->runtime_flags))
2066 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2068 if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2069 &config->runtime_flags))
2070 refcount_inc(&nbd->refs);
2071 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2074 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2075 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2076 &config->runtime_flags);
2078 clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2079 &config->runtime_flags);
2083 if (info->attrs[NBD_ATTR_SOCKETS]) {
2084 struct nlattr *attr;
2087 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2089 struct nlattr *socks[NBD_SOCK_MAX+1];
2091 if (nla_type(attr) != NBD_SOCK_ITEM) {
2092 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2096 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2101 printk(KERN_ERR "nbd: error processing sock list\n");
2105 if (!socks[NBD_SOCK_FD])
2107 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2108 ret = nbd_reconnect_socket(nbd, fd);
2114 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2118 mutex_unlock(&nbd->config_lock);
2119 nbd_config_put(nbd);
2126 static const struct genl_ops nbd_connect_genl_ops[] = {
2128 .cmd = NBD_CMD_CONNECT,
2129 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2130 .doit = nbd_genl_connect,
2133 .cmd = NBD_CMD_DISCONNECT,
2134 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2135 .doit = nbd_genl_disconnect,
2138 .cmd = NBD_CMD_RECONFIGURE,
2139 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2140 .doit = nbd_genl_reconfigure,
2143 .cmd = NBD_CMD_STATUS,
2144 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2145 .doit = nbd_genl_status,
2149 static const struct genl_multicast_group nbd_mcast_grps[] = {
2150 { .name = NBD_GENL_MCAST_GROUP_NAME, },
2153 static struct genl_family nbd_genl_family __ro_after_init = {
2155 .name = NBD_GENL_FAMILY_NAME,
2156 .version = NBD_GENL_VERSION,
2157 .module = THIS_MODULE,
2158 .ops = nbd_connect_genl_ops,
2159 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2160 .maxattr = NBD_ATTR_MAX,
2161 .policy = nbd_attr_policy,
2162 .mcgrps = nbd_mcast_grps,
2163 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2166 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2168 struct nlattr *dev_opt;
2172 /* This is a little racey, but for status it's ok. The
2173 * reason we don't take a ref here is because we can't
2174 * take a ref in the index == -1 case as we would need
2175 * to put under the nbd_index_mutex, which could
2176 * deadlock if we are configured to remove ourselves
2177 * once we're disconnected.
2179 if (refcount_read(&nbd->config_refs))
2181 dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2184 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2187 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2191 nla_nest_end(reply, dev_opt);
2195 static int status_cb(int id, void *ptr, void *data)
2197 struct nbd_device *nbd = ptr;
2198 return populate_nbd_status(nbd, (struct sk_buff *)data);
2201 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2203 struct nlattr *dev_list;
2204 struct sk_buff *reply;
2210 if (info->attrs[NBD_ATTR_INDEX])
2211 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2213 mutex_lock(&nbd_index_mutex);
2215 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2216 nla_attr_size(sizeof(u8)));
2217 msg_size *= (index == -1) ? nbd_total_devices : 1;
2219 reply = genlmsg_new(msg_size, GFP_KERNEL);
2222 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2229 dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2231 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2237 struct nbd_device *nbd;
2238 nbd = idr_find(&nbd_index_idr, index);
2240 ret = populate_nbd_status(nbd, reply);
2247 nla_nest_end(reply, dev_list);
2248 genlmsg_end(reply, reply_head);
2249 ret = genlmsg_reply(reply, info);
2251 mutex_unlock(&nbd_index_mutex);
2255 static void nbd_connect_reply(struct genl_info *info, int index)
2257 struct sk_buff *skb;
2261 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2264 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2270 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2275 genlmsg_end(skb, msg_head);
2276 genlmsg_reply(skb, info);
2279 static void nbd_mcast_index(int index)
2281 struct sk_buff *skb;
2285 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2288 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2294 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2299 genlmsg_end(skb, msg_head);
2300 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2303 static void nbd_dead_link_work(struct work_struct *work)
2305 struct link_dead_args *args = container_of(work, struct link_dead_args,
2307 nbd_mcast_index(args->index);
2311 static int __init nbd_init(void)
2315 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2318 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2324 part_shift = fls(max_part);
2327 * Adjust max_part according to part_shift as it is exported
2328 * to user space so that user can know the max number of
2329 * partition kernel should be able to manage.
2331 * Note that -1 is required because partition 0 is reserved
2332 * for the whole disk.
2334 max_part = (1UL << part_shift) - 1;
2337 if ((1UL << part_shift) > DISK_MAX_PARTS)
2340 if (nbds_max > 1UL << (MINORBITS - part_shift))
2343 if (register_blkdev(NBD_MAJOR, "nbd"))
2346 if (genl_register_family(&nbd_genl_family)) {
2347 unregister_blkdev(NBD_MAJOR, "nbd");
2352 mutex_lock(&nbd_index_mutex);
2353 for (i = 0; i < nbds_max; i++)
2355 mutex_unlock(&nbd_index_mutex);
2359 static int nbd_exit_cb(int id, void *ptr, void *data)
2361 struct list_head *list = (struct list_head *)data;
2362 struct nbd_device *nbd = ptr;
2364 list_add_tail(&nbd->list, list);
2368 static void __exit nbd_cleanup(void)
2370 struct nbd_device *nbd;
2371 LIST_HEAD(del_list);
2375 mutex_lock(&nbd_index_mutex);
2376 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2377 mutex_unlock(&nbd_index_mutex);
2379 while (!list_empty(&del_list)) {
2380 nbd = list_first_entry(&del_list, struct nbd_device, list);
2381 list_del_init(&nbd->list);
2382 if (refcount_read(&nbd->refs) != 1)
2383 printk(KERN_ERR "nbd: possibly leaking a device\n");
2387 idr_destroy(&nbd_index_idr);
2388 genl_unregister_family(&nbd_genl_family);
2389 unregister_blkdev(NBD_MAJOR, "nbd");
2392 module_init(nbd_init);
2393 module_exit(nbd_cleanup);
2395 MODULE_DESCRIPTION("Network Block Device");
2396 MODULE_LICENSE("GPL");
2398 module_param(nbds_max, int, 0444);
2399 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2400 module_param(max_part, int, 0444);
2401 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");