2 * Network block device - make block devices work over TCP
4 * Note that you can not swap over this thing, yet. Seems to work but
5 * deadlocks sometimes - you can not swap over TCP in general.
7 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
8 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
10 * This file is released under GPLv2 or later.
12 * (part of code stolen from loop.c)
15 #include <linux/major.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/sched.h>
21 #include <linux/sched/mm.h>
23 #include <linux/bio.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/file.h>
27 #include <linux/ioctl.h>
28 #include <linux/mutex.h>
29 #include <linux/compiler.h>
30 #include <linux/err.h>
31 #include <linux/kernel.h>
32 #include <linux/slab.h>
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 static DEFINE_IDR(nbd_index_idr);
48 static DEFINE_MUTEX(nbd_index_mutex);
49 static int nbd_total_devices = 0;
54 struct request *pending;
61 struct recv_thread_args {
62 struct work_struct work;
63 struct nbd_device *nbd;
67 struct link_dead_args {
68 struct work_struct work;
72 #define NBD_TIMEDOUT 0
73 #define NBD_DISCONNECT_REQUESTED 1
74 #define NBD_DISCONNECTED 2
75 #define NBD_HAS_PID_FILE 3
76 #define NBD_HAS_CONFIG_REF 4
78 #define NBD_DESTROY_ON_DISCONNECT 6
79 #define NBD_DISCONNECT_ON_CLOSE 7
83 unsigned long runtime_flags;
84 u64 dead_conn_timeout;
86 struct nbd_sock **socks;
88 atomic_t live_connections;
89 wait_queue_head_t conn_wait;
91 atomic_t recv_threads;
92 wait_queue_head_t recv_wq;
95 #if IS_ENABLED(CONFIG_DEBUG_FS)
96 struct dentry *dbg_dir;
101 struct blk_mq_tag_set tag_set;
104 refcount_t config_refs;
106 struct nbd_config *config;
107 struct mutex config_lock;
108 struct gendisk *disk;
110 struct list_head list;
111 struct task_struct *task_recv;
112 struct task_struct *task_setup;
116 struct nbd_device *nbd;
119 struct completion send_complete;
123 #if IS_ENABLED(CONFIG_DEBUG_FS)
124 static struct dentry *nbd_dbg_dir;
127 #define nbd_name(nbd) ((nbd)->disk->disk_name)
129 #define NBD_MAGIC 0x68797548
131 static unsigned int nbds_max = 16;
132 static int max_part = 16;
133 static struct workqueue_struct *recv_workqueue;
134 static int part_shift;
136 static int nbd_dev_dbg_init(struct nbd_device *nbd);
137 static void nbd_dev_dbg_close(struct nbd_device *nbd);
138 static void nbd_config_put(struct nbd_device *nbd);
139 static void nbd_connect_reply(struct genl_info *info, int index);
140 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
141 static void nbd_dead_link_work(struct work_struct *work);
142 static void nbd_disconnect_and_put(struct nbd_device *nbd);
144 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
146 return disk_to_dev(nbd->disk);
149 static const char *nbdcmd_to_ascii(int cmd)
152 case NBD_CMD_READ: return "read";
153 case NBD_CMD_WRITE: return "write";
154 case NBD_CMD_DISC: return "disconnect";
155 case NBD_CMD_FLUSH: return "flush";
156 case NBD_CMD_TRIM: return "trim/discard";
161 static ssize_t pid_show(struct device *dev,
162 struct device_attribute *attr, char *buf)
164 struct gendisk *disk = dev_to_disk(dev);
165 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
167 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
170 static const struct device_attribute pid_attr = {
171 .attr = { .name = "pid", .mode = 0444},
175 static void nbd_dev_remove(struct nbd_device *nbd)
177 struct gendisk *disk = nbd->disk;
178 struct request_queue *q;
183 blk_cleanup_queue(q);
184 blk_mq_free_tag_set(&nbd->tag_set);
185 disk->private_data = NULL;
191 static void nbd_put(struct nbd_device *nbd)
193 if (refcount_dec_and_mutex_lock(&nbd->refs,
195 idr_remove(&nbd_index_idr, nbd->index);
196 mutex_unlock(&nbd_index_mutex);
201 static int nbd_disconnected(struct nbd_config *config)
203 return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
204 test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
207 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
210 if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
211 struct link_dead_args *args;
212 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
214 INIT_WORK(&args->work, nbd_dead_link_work);
215 args->index = nbd->index;
216 queue_work(system_wq, &args->work);
220 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
221 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
222 if (test_and_clear_bit(NBD_DISCONNECT_REQUESTED,
223 &nbd->config->runtime_flags)) {
224 set_bit(NBD_DISCONNECTED,
225 &nbd->config->runtime_flags);
226 dev_info(nbd_to_dev(nbd),
227 "Disconnected due to user request.\n");
232 nsock->pending = NULL;
236 static void nbd_size_clear(struct nbd_device *nbd)
238 if (nbd->config->bytesize) {
239 set_capacity(nbd->disk, 0);
240 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
244 static void nbd_size_update(struct nbd_device *nbd)
246 struct nbd_config *config = nbd->config;
247 struct block_device *bdev = bdget_disk(nbd->disk, 0);
249 if (config->flags & NBD_FLAG_SEND_TRIM) {
250 nbd->disk->queue->limits.discard_granularity = config->blksize;
251 nbd->disk->queue->limits.discard_alignment = config->blksize;
252 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
254 blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
255 blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
256 set_capacity(nbd->disk, config->bytesize >> 9);
259 bd_set_size(bdev, config->bytesize);
261 bdev->bd_invalidated = 1;
264 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
267 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
270 struct nbd_config *config = nbd->config;
271 config->blksize = blocksize;
272 config->bytesize = blocksize * nr_blocks;
273 if (nbd->task_recv != NULL)
274 nbd_size_update(nbd);
277 static void nbd_complete_rq(struct request *req)
279 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
281 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
282 cmd->status ? "failed" : "done");
284 blk_mq_end_request(req, cmd->status);
288 * Forcibly shutdown the socket causing all listeners to error
290 static void sock_shutdown(struct nbd_device *nbd)
292 struct nbd_config *config = nbd->config;
295 if (config->num_connections == 0)
297 if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
300 for (i = 0; i < config->num_connections; i++) {
301 struct nbd_sock *nsock = config->socks[i];
302 mutex_lock(&nsock->tx_lock);
303 nbd_mark_nsock_dead(nbd, nsock, 0);
304 mutex_unlock(&nsock->tx_lock);
306 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
309 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
312 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
313 struct nbd_device *nbd = cmd->nbd;
314 struct nbd_config *config;
316 if (!refcount_inc_not_zero(&nbd->config_refs)) {
317 cmd->status = BLK_STS_TIMEOUT;
320 config = nbd->config;
322 if (config->num_connections > 1) {
323 dev_err_ratelimited(nbd_to_dev(nbd),
324 "Connection timed out, retrying (%d/%d alive)\n",
325 atomic_read(&config->live_connections),
326 config->num_connections);
328 * Hooray we have more connections, requeue this IO, the submit
329 * path will put it on a real connection.
331 if (config->socks && config->num_connections > 1) {
332 if (cmd->index < config->num_connections) {
333 struct nbd_sock *nsock =
334 config->socks[cmd->index];
335 mutex_lock(&nsock->tx_lock);
336 /* We can have multiple outstanding requests, so
337 * we don't want to mark the nsock dead if we've
338 * already reconnected with a new socket, so
339 * only mark it dead if its the same socket we
342 if (cmd->cookie == nsock->cookie)
343 nbd_mark_nsock_dead(nbd, nsock, 1);
344 mutex_unlock(&nsock->tx_lock);
346 blk_mq_requeue_request(req, true);
351 dev_err_ratelimited(nbd_to_dev(nbd),
352 "Connection timed out\n");
354 set_bit(NBD_TIMEDOUT, &config->runtime_flags);
355 cmd->status = BLK_STS_IOERR;
359 blk_mq_complete_request(req);
364 * Send or receive packet.
366 static int sock_xmit(struct nbd_device *nbd, int index, int send,
367 struct iov_iter *iter, int msg_flags, int *sent)
369 struct nbd_config *config = nbd->config;
370 struct socket *sock = config->socks[index]->sock;
373 unsigned int noreclaim_flag;
375 if (unlikely(!sock)) {
376 dev_err_ratelimited(disk_to_dev(nbd->disk),
377 "Attempted %s on closed socket in sock_xmit\n",
378 (send ? "send" : "recv"));
382 msg.msg_iter = *iter;
384 noreclaim_flag = memalloc_noreclaim_save();
386 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
389 msg.msg_control = NULL;
390 msg.msg_controllen = 0;
391 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
394 result = sock_sendmsg(sock, &msg);
396 result = sock_recvmsg(sock, &msg, msg.msg_flags);
400 result = -EPIPE; /* short read */
405 } while (msg_data_left(&msg));
407 memalloc_noreclaim_restore(noreclaim_flag);
413 * Different settings for sk->sk_sndtimeo can result in different return values
414 * if there is a signal pending when we enter sendmsg, because reasons?
416 static inline int was_interrupted(int result)
418 return result == -ERESTARTSYS || result == -EINTR;
421 /* always call with the tx_lock held */
422 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
424 struct request *req = blk_mq_rq_from_pdu(cmd);
425 struct nbd_config *config = nbd->config;
426 struct nbd_sock *nsock = config->socks[index];
428 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
429 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
430 struct iov_iter from;
431 unsigned long size = blk_rq_bytes(req);
434 u32 nbd_cmd_flags = 0;
435 u32 tag = blk_mq_unique_tag(req);
436 int sent = nsock->sent, skip = 0;
438 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
440 switch (req_op(req)) {
445 type = NBD_CMD_FLUSH;
448 type = NBD_CMD_WRITE;
457 if (rq_data_dir(req) == WRITE &&
458 (config->flags & NBD_FLAG_READ_ONLY)) {
459 dev_err_ratelimited(disk_to_dev(nbd->disk),
460 "Write on read-only\n");
464 if (req->cmd_flags & REQ_FUA)
465 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
467 /* We did a partial send previously, and we at least sent the whole
468 * request struct, so just go and send the rest of the pages in the
472 if (sent >= sizeof(request)) {
473 skip = sent - sizeof(request);
476 iov_iter_advance(&from, sent);
479 cmd->cookie = nsock->cookie;
480 request.type = htonl(type | nbd_cmd_flags);
481 if (type != NBD_CMD_FLUSH) {
482 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
483 request.len = htonl(size);
485 memcpy(request.handle, &tag, sizeof(tag));
487 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
488 req, nbdcmd_to_ascii(type),
489 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
490 result = sock_xmit(nbd, index, 1, &from,
491 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
493 if (was_interrupted(result)) {
494 /* If we havne't sent anything we can just return BUSY,
495 * however if we have sent something we need to make
496 * sure we only allow this req to be sent until we are
500 nsock->pending = req;
503 return BLK_STS_RESOURCE;
505 dev_err_ratelimited(disk_to_dev(nbd->disk),
506 "Send control failed (result %d)\n", result);
510 if (type != NBD_CMD_WRITE)
515 struct bio *next = bio->bi_next;
516 struct bvec_iter iter;
519 bio_for_each_segment(bvec, bio, iter) {
520 bool is_last = !next && bio_iter_last(bvec, iter);
521 int flags = is_last ? 0 : MSG_MORE;
523 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
525 iov_iter_bvec(&from, ITER_BVEC | WRITE,
526 &bvec, 1, bvec.bv_len);
528 if (skip >= iov_iter_count(&from)) {
529 skip -= iov_iter_count(&from);
532 iov_iter_advance(&from, skip);
535 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
537 if (was_interrupted(result)) {
538 /* We've already sent the header, we
539 * have no choice but to set pending and
542 nsock->pending = req;
544 return BLK_STS_RESOURCE;
546 dev_err(disk_to_dev(nbd->disk),
547 "Send data failed (result %d)\n",
552 * The completion might already have come in,
553 * so break for the last one instead of letting
554 * the iterator do it. This prevents use-after-free
563 nsock->pending = NULL;
568 /* NULL returned = something went wrong, inform userspace */
569 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
571 struct nbd_config *config = nbd->config;
573 struct nbd_reply reply;
575 struct request *req = NULL;
578 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
582 iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
583 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
585 if (!nbd_disconnected(config))
586 dev_err(disk_to_dev(nbd->disk),
587 "Receive control failed (result %d)\n", result);
588 return ERR_PTR(result);
591 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
592 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
593 (unsigned long)ntohl(reply.magic));
594 return ERR_PTR(-EPROTO);
597 memcpy(&tag, reply.handle, sizeof(u32));
599 hwq = blk_mq_unique_tag_to_hwq(tag);
600 if (hwq < nbd->tag_set.nr_hw_queues)
601 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
602 blk_mq_unique_tag_to_tag(tag));
603 if (!req || !blk_mq_request_started(req)) {
604 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
606 return ERR_PTR(-ENOENT);
608 cmd = blk_mq_rq_to_pdu(req);
609 if (ntohl(reply.error)) {
610 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
612 cmd->status = BLK_STS_IOERR;
616 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
617 if (rq_data_dir(req) != WRITE) {
618 struct req_iterator iter;
621 rq_for_each_segment(bvec, req, iter) {
622 iov_iter_bvec(&to, ITER_BVEC | READ,
623 &bvec, 1, bvec.bv_len);
624 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
626 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
629 * If we've disconnected or we only have 1
630 * connection then we need to make sure we
631 * complete this request, otherwise error out
632 * and let the timeout stuff handle resubmitting
633 * this request onto another connection.
635 if (nbd_disconnected(config) ||
636 config->num_connections <= 1) {
637 cmd->status = BLK_STS_IOERR;
640 return ERR_PTR(-EIO);
642 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
646 /* See the comment in nbd_queue_rq. */
647 wait_for_completion(&cmd->send_complete);
652 static void recv_work(struct work_struct *work)
654 struct recv_thread_args *args = container_of(work,
655 struct recv_thread_args,
657 struct nbd_device *nbd = args->nbd;
658 struct nbd_config *config = nbd->config;
662 cmd = nbd_read_stat(nbd, args->index);
664 struct nbd_sock *nsock = config->socks[args->index];
666 mutex_lock(&nsock->tx_lock);
667 nbd_mark_nsock_dead(nbd, nsock, 1);
668 mutex_unlock(&nsock->tx_lock);
672 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
674 atomic_dec(&config->recv_threads);
675 wake_up(&config->recv_wq);
680 static void nbd_clear_req(struct request *req, void *data, bool reserved)
682 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
684 cmd->status = BLK_STS_IOERR;
685 blk_mq_complete_request(req);
688 static void nbd_clear_que(struct nbd_device *nbd)
690 blk_mq_quiesce_queue(nbd->disk->queue);
691 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
692 blk_mq_unquiesce_queue(nbd->disk->queue);
693 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
696 static int find_fallback(struct nbd_device *nbd, int index)
698 struct nbd_config *config = nbd->config;
700 struct nbd_sock *nsock = config->socks[index];
701 int fallback = nsock->fallback_index;
703 if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
706 if (config->num_connections <= 1) {
707 dev_err_ratelimited(disk_to_dev(nbd->disk),
708 "Attempted send on invalid socket\n");
712 if (fallback >= 0 && fallback < config->num_connections &&
713 !config->socks[fallback]->dead)
716 if (nsock->fallback_index < 0 ||
717 nsock->fallback_index >= config->num_connections ||
718 config->socks[nsock->fallback_index]->dead) {
720 for (i = 0; i < config->num_connections; i++) {
723 if (!config->socks[i]->dead) {
728 nsock->fallback_index = new_index;
730 dev_err_ratelimited(disk_to_dev(nbd->disk),
731 "Dead connection, failed to find a fallback\n");
735 new_index = nsock->fallback_index;
739 static int wait_for_reconnect(struct nbd_device *nbd)
741 struct nbd_config *config = nbd->config;
742 if (!config->dead_conn_timeout)
744 if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
746 return wait_event_timeout(config->conn_wait,
747 atomic_read(&config->live_connections) > 0,
748 config->dead_conn_timeout) > 0;
751 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
753 struct request *req = blk_mq_rq_from_pdu(cmd);
754 struct nbd_device *nbd = cmd->nbd;
755 struct nbd_config *config;
756 struct nbd_sock *nsock;
759 if (!refcount_inc_not_zero(&nbd->config_refs)) {
760 dev_err_ratelimited(disk_to_dev(nbd->disk),
761 "Socks array is empty\n");
762 blk_mq_start_request(req);
765 config = nbd->config;
767 if (index >= config->num_connections) {
768 dev_err_ratelimited(disk_to_dev(nbd->disk),
769 "Attempted send on invalid socket\n");
771 blk_mq_start_request(req);
774 cmd->status = BLK_STS_OK;
776 nsock = config->socks[index];
777 mutex_lock(&nsock->tx_lock);
779 int old_index = index;
780 index = find_fallback(nbd, index);
781 mutex_unlock(&nsock->tx_lock);
783 if (wait_for_reconnect(nbd)) {
787 /* All the sockets should already be down at this point,
788 * we just want to make sure that DISCONNECTED is set so
789 * any requests that come in that were queue'ed waiting
790 * for the reconnect timer don't trigger the timer again
791 * and instead just error out.
795 blk_mq_start_request(req);
801 /* Handle the case that we have a pending request that was partially
802 * transmitted that _has_ to be serviced first. We need to call requeue
803 * here so that it gets put _after_ the request that is already on the
806 blk_mq_start_request(req);
807 if (unlikely(nsock->pending && nsock->pending != req)) {
808 blk_mq_requeue_request(req, true);
813 * Some failures are related to the link going down, so anything that
814 * returns EAGAIN can be retried on a different socket.
816 ret = nbd_send_cmd(nbd, cmd, index);
817 if (ret == -EAGAIN) {
818 dev_err_ratelimited(disk_to_dev(nbd->disk),
819 "Request send failed, requeueing\n");
820 nbd_mark_nsock_dead(nbd, nsock, 1);
821 blk_mq_requeue_request(req, true);
825 mutex_unlock(&nsock->tx_lock);
830 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
831 const struct blk_mq_queue_data *bd)
833 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
837 * Since we look at the bio's to send the request over the network we
838 * need to make sure the completion work doesn't mark this request done
839 * before we are done doing our send. This keeps us from dereferencing
840 * freed data if we have particularly fast completions (ie we get the
841 * completion before we exit sock_xmit on the last bvec) or in the case
842 * that the server is misbehaving (or there was an error) before we're
843 * done sending everything over the wire.
845 init_completion(&cmd->send_complete);
847 /* We can be called directly from the user space process, which means we
848 * could possibly have signals pending so our sendmsg will fail. In
849 * this case we need to return that we are busy, otherwise error out as
852 ret = nbd_handle_cmd(cmd, hctx->queue_num);
857 complete(&cmd->send_complete);
862 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
865 struct nbd_config *config = nbd->config;
867 struct nbd_sock **socks;
868 struct nbd_sock *nsock;
871 sock = sockfd_lookup(arg, &err);
875 if (!netlink && !nbd->task_setup &&
876 !test_bit(NBD_BOUND, &config->runtime_flags))
877 nbd->task_setup = current;
880 (nbd->task_setup != current ||
881 test_bit(NBD_BOUND, &config->runtime_flags))) {
882 dev_err(disk_to_dev(nbd->disk),
883 "Device being setup by another task");
888 socks = krealloc(config->socks, (config->num_connections + 1) *
889 sizeof(struct nbd_sock *), GFP_KERNEL);
894 nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
900 config->socks = socks;
902 nsock->fallback_index = -1;
904 mutex_init(&nsock->tx_lock);
906 nsock->pending = NULL;
909 socks[config->num_connections++] = nsock;
910 atomic_inc(&config->live_connections);
915 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
917 struct nbd_config *config = nbd->config;
918 struct socket *sock, *old;
919 struct recv_thread_args *args;
923 sock = sockfd_lookup(arg, &err);
927 args = kzalloc(sizeof(*args), GFP_KERNEL);
933 for (i = 0; i < config->num_connections; i++) {
934 struct nbd_sock *nsock = config->socks[i];
939 mutex_lock(&nsock->tx_lock);
941 mutex_unlock(&nsock->tx_lock);
944 sk_set_memalloc(sock->sk);
945 if (nbd->tag_set.timeout)
946 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
947 atomic_inc(&config->recv_threads);
948 refcount_inc(&nbd->config_refs);
950 nsock->fallback_index = -1;
953 INIT_WORK(&args->work, recv_work);
957 mutex_unlock(&nsock->tx_lock);
960 clear_bit(NBD_DISCONNECTED, &config->runtime_flags);
962 /* We take the tx_mutex in an error path in the recv_work, so we
963 * need to queue_work outside of the tx_mutex.
965 queue_work(recv_workqueue, &args->work);
967 atomic_inc(&config->live_connections);
968 wake_up(&config->conn_wait);
976 static void nbd_bdev_reset(struct block_device *bdev)
978 if (bdev->bd_openers > 1)
980 bd_set_size(bdev, 0);
983 static void nbd_parse_flags(struct nbd_device *nbd)
985 struct nbd_config *config = nbd->config;
986 if (config->flags & NBD_FLAG_READ_ONLY)
987 set_disk_ro(nbd->disk, true);
989 set_disk_ro(nbd->disk, false);
990 if (config->flags & NBD_FLAG_SEND_TRIM)
991 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
992 if (config->flags & NBD_FLAG_SEND_FLUSH) {
993 if (config->flags & NBD_FLAG_SEND_FUA)
994 blk_queue_write_cache(nbd->disk->queue, true, true);
996 blk_queue_write_cache(nbd->disk->queue, true, false);
999 blk_queue_write_cache(nbd->disk->queue, false, false);
1002 static void send_disconnects(struct nbd_device *nbd)
1004 struct nbd_config *config = nbd->config;
1005 struct nbd_request request = {
1006 .magic = htonl(NBD_REQUEST_MAGIC),
1007 .type = htonl(NBD_CMD_DISC),
1009 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1010 struct iov_iter from;
1013 for (i = 0; i < config->num_connections; i++) {
1014 struct nbd_sock *nsock = config->socks[i];
1016 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
1017 mutex_lock(&nsock->tx_lock);
1018 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1020 dev_err(disk_to_dev(nbd->disk),
1021 "Send disconnect failed %d\n", ret);
1022 mutex_unlock(&nsock->tx_lock);
1026 static int nbd_disconnect(struct nbd_device *nbd)
1028 struct nbd_config *config = nbd->config;
1030 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1031 set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
1032 send_disconnects(nbd);
1036 static void nbd_clear_sock(struct nbd_device *nbd)
1040 nbd->task_setup = NULL;
1043 static void nbd_config_put(struct nbd_device *nbd)
1045 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1046 &nbd->config_lock)) {
1047 struct nbd_config *config = nbd->config;
1048 nbd_dev_dbg_close(nbd);
1049 nbd_size_clear(nbd);
1050 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1051 &config->runtime_flags))
1052 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1053 nbd->task_recv = NULL;
1054 nbd_clear_sock(nbd);
1055 if (config->num_connections) {
1057 for (i = 0; i < config->num_connections; i++) {
1058 sockfd_put(config->socks[i]->sock);
1059 kfree(config->socks[i]);
1061 kfree(config->socks);
1066 nbd->tag_set.timeout = 0;
1067 nbd->disk->queue->limits.discard_granularity = 0;
1068 nbd->disk->queue->limits.discard_alignment = 0;
1069 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1070 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1072 mutex_unlock(&nbd->config_lock);
1074 module_put(THIS_MODULE);
1078 static int nbd_start_device(struct nbd_device *nbd)
1080 struct nbd_config *config = nbd->config;
1081 int num_connections = config->num_connections;
1088 if (num_connections > 1 &&
1089 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1090 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1094 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1095 nbd->task_recv = current;
1097 nbd_parse_flags(nbd);
1099 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1101 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1104 set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1106 nbd_dev_dbg_init(nbd);
1107 for (i = 0; i < num_connections; i++) {
1108 struct recv_thread_args *args;
1110 args = kzalloc(sizeof(*args), GFP_KERNEL);
1115 sk_set_memalloc(config->socks[i]->sock->sk);
1116 if (nbd->tag_set.timeout)
1117 config->socks[i]->sock->sk->sk_sndtimeo =
1118 nbd->tag_set.timeout;
1119 atomic_inc(&config->recv_threads);
1120 refcount_inc(&nbd->config_refs);
1121 INIT_WORK(&args->work, recv_work);
1124 queue_work(recv_workqueue, &args->work);
1126 nbd_size_update(nbd);
1130 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1132 struct nbd_config *config = nbd->config;
1135 ret = nbd_start_device(nbd);
1140 bdev->bd_invalidated = 1;
1141 mutex_unlock(&nbd->config_lock);
1142 ret = wait_event_interruptible(config->recv_wq,
1143 atomic_read(&config->recv_threads) == 0);
1146 mutex_lock(&nbd->config_lock);
1147 nbd_bdev_reset(bdev);
1148 /* user requested, ignore socket errors */
1149 if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1151 if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1156 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1157 struct block_device *bdev)
1161 nbd_bdev_reset(bdev);
1162 if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1163 &nbd->config->runtime_flags))
1164 nbd_config_put(nbd);
1167 /* Must be called with config_lock held */
1168 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1169 unsigned int cmd, unsigned long arg)
1171 struct nbd_config *config = nbd->config;
1174 case NBD_DISCONNECT:
1175 return nbd_disconnect(nbd);
1176 case NBD_CLEAR_SOCK:
1177 nbd_clear_sock_ioctl(nbd, bdev);
1180 return nbd_add_socket(nbd, arg, false);
1181 case NBD_SET_BLKSIZE:
1182 nbd_size_set(nbd, arg,
1183 div_s64(config->bytesize, arg));
1186 nbd_size_set(nbd, config->blksize,
1187 div_s64(arg, config->blksize));
1189 case NBD_SET_SIZE_BLOCKS:
1190 nbd_size_set(nbd, config->blksize, arg);
1192 case NBD_SET_TIMEOUT:
1194 nbd->tag_set.timeout = arg * HZ;
1195 blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1200 config->flags = arg;
1203 return nbd_start_device_ioctl(nbd, bdev);
1206 * This is for compatibility only. The queue is always cleared
1207 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1210 case NBD_PRINT_DEBUG:
1212 * For compatibility only, we no longer keep a list of
1213 * outstanding requests.
1220 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1221 unsigned int cmd, unsigned long arg)
1223 struct nbd_device *nbd = bdev->bd_disk->private_data;
1224 struct nbd_config *config = nbd->config;
1225 int error = -EINVAL;
1227 if (!capable(CAP_SYS_ADMIN))
1230 /* The block layer will pass back some non-nbd ioctls in case we have
1231 * special handling for them, but we don't so just return an error.
1233 if (_IOC_TYPE(cmd) != 0xab)
1236 mutex_lock(&nbd->config_lock);
1238 /* Don't allow ioctl operations on a nbd device that was created with
1239 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1241 if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1242 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1243 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1245 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1246 mutex_unlock(&nbd->config_lock);
1250 static struct nbd_config *nbd_alloc_config(void)
1252 struct nbd_config *config;
1254 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1257 atomic_set(&config->recv_threads, 0);
1258 init_waitqueue_head(&config->recv_wq);
1259 init_waitqueue_head(&config->conn_wait);
1260 config->blksize = 1024;
1261 atomic_set(&config->live_connections, 0);
1262 try_module_get(THIS_MODULE);
1266 static int nbd_open(struct block_device *bdev, fmode_t mode)
1268 struct nbd_device *nbd;
1271 mutex_lock(&nbd_index_mutex);
1272 nbd = bdev->bd_disk->private_data;
1277 if (!refcount_inc_not_zero(&nbd->refs)) {
1281 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1282 struct nbd_config *config;
1284 mutex_lock(&nbd->config_lock);
1285 if (refcount_inc_not_zero(&nbd->config_refs)) {
1286 mutex_unlock(&nbd->config_lock);
1289 config = nbd->config = nbd_alloc_config();
1292 mutex_unlock(&nbd->config_lock);
1295 refcount_set(&nbd->config_refs, 1);
1296 refcount_inc(&nbd->refs);
1297 mutex_unlock(&nbd->config_lock);
1298 bdev->bd_invalidated = 1;
1299 } else if (nbd_disconnected(nbd->config)) {
1300 bdev->bd_invalidated = 1;
1303 mutex_unlock(&nbd_index_mutex);
1307 static void nbd_release(struct gendisk *disk, fmode_t mode)
1309 struct nbd_device *nbd = disk->private_data;
1310 struct block_device *bdev = bdget_disk(disk, 0);
1312 if (test_bit(NBD_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1313 bdev->bd_openers == 0)
1314 nbd_disconnect_and_put(nbd);
1316 nbd_config_put(nbd);
1320 static const struct block_device_operations nbd_fops =
1322 .owner = THIS_MODULE,
1324 .release = nbd_release,
1326 .compat_ioctl = nbd_ioctl,
1329 #if IS_ENABLED(CONFIG_DEBUG_FS)
1331 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1333 struct nbd_device *nbd = s->private;
1336 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1341 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1343 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1346 static const struct file_operations nbd_dbg_tasks_ops = {
1347 .open = nbd_dbg_tasks_open,
1349 .llseek = seq_lseek,
1350 .release = single_release,
1353 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1355 struct nbd_device *nbd = s->private;
1356 u32 flags = nbd->config->flags;
1358 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1360 seq_puts(s, "Known flags:\n");
1362 if (flags & NBD_FLAG_HAS_FLAGS)
1363 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1364 if (flags & NBD_FLAG_READ_ONLY)
1365 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1366 if (flags & NBD_FLAG_SEND_FLUSH)
1367 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1368 if (flags & NBD_FLAG_SEND_FUA)
1369 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1370 if (flags & NBD_FLAG_SEND_TRIM)
1371 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1376 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1378 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1381 static const struct file_operations nbd_dbg_flags_ops = {
1382 .open = nbd_dbg_flags_open,
1384 .llseek = seq_lseek,
1385 .release = single_release,
1388 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1391 struct nbd_config *config = nbd->config;
1396 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1398 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1402 config->dbg_dir = dir;
1404 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1405 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1406 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1407 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1408 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1413 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1415 debugfs_remove_recursive(nbd->config->dbg_dir);
1418 static int nbd_dbg_init(void)
1420 struct dentry *dbg_dir;
1422 dbg_dir = debugfs_create_dir("nbd", NULL);
1426 nbd_dbg_dir = dbg_dir;
1431 static void nbd_dbg_close(void)
1433 debugfs_remove_recursive(nbd_dbg_dir);
1436 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1438 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1443 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1447 static int nbd_dbg_init(void)
1452 static void nbd_dbg_close(void)
1458 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1459 unsigned int hctx_idx, unsigned int numa_node)
1461 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1462 cmd->nbd = set->driver_data;
1466 static const struct blk_mq_ops nbd_mq_ops = {
1467 .queue_rq = nbd_queue_rq,
1468 .complete = nbd_complete_rq,
1469 .init_request = nbd_init_request,
1470 .timeout = nbd_xmit_timeout,
1473 static int nbd_dev_add(int index)
1475 struct nbd_device *nbd;
1476 struct gendisk *disk;
1477 struct request_queue *q;
1480 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1484 disk = alloc_disk(1 << part_shift);
1489 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1494 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1503 nbd->tag_set.ops = &nbd_mq_ops;
1504 nbd->tag_set.nr_hw_queues = 1;
1505 nbd->tag_set.queue_depth = 128;
1506 nbd->tag_set.numa_node = NUMA_NO_NODE;
1507 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1508 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1509 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1510 nbd->tag_set.driver_data = nbd;
1512 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1516 q = blk_mq_init_queue(&nbd->tag_set);
1524 * Tell the block layer that we are not a rotational device
1526 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1527 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1528 disk->queue->limits.discard_granularity = 0;
1529 disk->queue->limits.discard_alignment = 0;
1530 blk_queue_max_discard_sectors(disk->queue, 0);
1531 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1532 blk_queue_max_segments(disk->queue, USHRT_MAX);
1533 blk_queue_max_hw_sectors(disk->queue, 65536);
1534 disk->queue->limits.max_sectors = 256;
1536 mutex_init(&nbd->config_lock);
1537 refcount_set(&nbd->config_refs, 0);
1538 refcount_set(&nbd->refs, 1);
1539 INIT_LIST_HEAD(&nbd->list);
1540 disk->major = NBD_MAJOR;
1541 disk->first_minor = index << part_shift;
1542 disk->fops = &nbd_fops;
1543 disk->private_data = nbd;
1544 sprintf(disk->disk_name, "nbd%d", index);
1546 nbd_total_devices++;
1550 blk_mq_free_tag_set(&nbd->tag_set);
1552 idr_remove(&nbd_index_idr, index);
1561 static int find_free_cb(int id, void *ptr, void *data)
1563 struct nbd_device *nbd = ptr;
1564 struct nbd_device **found = data;
1566 if (!refcount_read(&nbd->config_refs)) {
1573 /* Netlink interface. */
1574 static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1575 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1576 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1577 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1578 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1579 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1580 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1581 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1582 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1583 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1586 static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1587 [NBD_SOCK_FD] = { .type = NLA_U32 },
1590 /* We don't use this right now since we don't parse the incoming list, but we
1591 * still want it here so userspace knows what to expect.
1593 static struct nla_policy __attribute__((unused))
1594 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1595 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1596 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1599 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1601 struct nbd_device *nbd = NULL;
1602 struct nbd_config *config;
1605 bool put_dev = false;
1607 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1610 if (info->attrs[NBD_ATTR_INDEX])
1611 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1612 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1613 printk(KERN_ERR "nbd: must specify at least one socket\n");
1616 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1617 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1621 mutex_lock(&nbd_index_mutex);
1623 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1626 new_index = nbd_dev_add(-1);
1627 if (new_index < 0) {
1628 mutex_unlock(&nbd_index_mutex);
1629 printk(KERN_ERR "nbd: failed to add new device\n");
1632 nbd = idr_find(&nbd_index_idr, new_index);
1635 nbd = idr_find(&nbd_index_idr, index);
1637 ret = nbd_dev_add(index);
1639 mutex_unlock(&nbd_index_mutex);
1640 printk(KERN_ERR "nbd: failed to add new device\n");
1643 nbd = idr_find(&nbd_index_idr, index);
1647 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1649 mutex_unlock(&nbd_index_mutex);
1652 if (!refcount_inc_not_zero(&nbd->refs)) {
1653 mutex_unlock(&nbd_index_mutex);
1656 printk(KERN_ERR "nbd: device at index %d is going down\n",
1660 mutex_unlock(&nbd_index_mutex);
1662 mutex_lock(&nbd->config_lock);
1663 if (refcount_read(&nbd->config_refs)) {
1664 mutex_unlock(&nbd->config_lock);
1668 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1671 if (WARN_ON(nbd->config)) {
1672 mutex_unlock(&nbd->config_lock);
1676 config = nbd->config = nbd_alloc_config();
1678 mutex_unlock(&nbd->config_lock);
1680 printk(KERN_ERR "nbd: couldn't allocate config\n");
1683 refcount_set(&nbd->config_refs, 1);
1684 set_bit(NBD_BOUND, &config->runtime_flags);
1686 if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1687 u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1688 nbd_size_set(nbd, config->blksize,
1689 div64_u64(bytes, config->blksize));
1691 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1693 nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1694 nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1696 if (info->attrs[NBD_ATTR_TIMEOUT]) {
1697 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1698 nbd->tag_set.timeout = timeout * HZ;
1699 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1701 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1702 config->dead_conn_timeout =
1703 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1704 config->dead_conn_timeout *= HZ;
1706 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1708 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1709 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1710 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1711 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1712 set_bit(NBD_DESTROY_ON_DISCONNECT,
1713 &config->runtime_flags);
1716 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1717 set_bit(NBD_DISCONNECT_ON_CLOSE,
1718 &config->runtime_flags);
1722 if (info->attrs[NBD_ATTR_SOCKETS]) {
1723 struct nlattr *attr;
1726 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1728 struct nlattr *socks[NBD_SOCK_MAX+1];
1730 if (nla_type(attr) != NBD_SOCK_ITEM) {
1731 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1735 ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1736 nbd_sock_policy, info->extack);
1738 printk(KERN_ERR "nbd: error processing sock list\n");
1742 if (!socks[NBD_SOCK_FD])
1744 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1745 ret = nbd_add_socket(nbd, fd, true);
1750 ret = nbd_start_device(nbd);
1752 mutex_unlock(&nbd->config_lock);
1754 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1755 refcount_inc(&nbd->config_refs);
1756 nbd_connect_reply(info, nbd->index);
1758 nbd_config_put(nbd);
1764 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1766 mutex_lock(&nbd->config_lock);
1767 nbd_disconnect(nbd);
1768 nbd_clear_sock(nbd);
1769 mutex_unlock(&nbd->config_lock);
1770 if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1771 &nbd->config->runtime_flags))
1772 nbd_config_put(nbd);
1775 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1777 struct nbd_device *nbd;
1780 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1783 if (!info->attrs[NBD_ATTR_INDEX]) {
1784 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1787 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1788 mutex_lock(&nbd_index_mutex);
1789 nbd = idr_find(&nbd_index_idr, index);
1791 mutex_unlock(&nbd_index_mutex);
1792 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1796 if (!refcount_inc_not_zero(&nbd->refs)) {
1797 mutex_unlock(&nbd_index_mutex);
1798 printk(KERN_ERR "nbd: device at index %d is going down\n",
1802 mutex_unlock(&nbd_index_mutex);
1803 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1807 nbd_disconnect_and_put(nbd);
1808 nbd_config_put(nbd);
1813 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1815 struct nbd_device *nbd = NULL;
1816 struct nbd_config *config;
1819 bool put_dev = false;
1821 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1824 if (!info->attrs[NBD_ATTR_INDEX]) {
1825 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1828 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1829 mutex_lock(&nbd_index_mutex);
1830 nbd = idr_find(&nbd_index_idr, index);
1832 mutex_unlock(&nbd_index_mutex);
1833 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1837 if (!refcount_inc_not_zero(&nbd->refs)) {
1838 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 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1846 dev_err(nbd_to_dev(nbd),
1847 "not configured, cannot reconfigure\n");
1852 mutex_lock(&nbd->config_lock);
1853 config = nbd->config;
1854 if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1856 dev_err(nbd_to_dev(nbd),
1857 "not configured, cannot reconfigure\n");
1862 if (info->attrs[NBD_ATTR_TIMEOUT]) {
1863 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1864 nbd->tag_set.timeout = timeout * HZ;
1865 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1867 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1868 config->dead_conn_timeout =
1869 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1870 config->dead_conn_timeout *= HZ;
1872 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1873 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1874 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1875 if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1876 &config->runtime_flags))
1879 if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1880 &config->runtime_flags))
1881 refcount_inc(&nbd->refs);
1884 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1885 set_bit(NBD_DISCONNECT_ON_CLOSE,
1886 &config->runtime_flags);
1888 clear_bit(NBD_DISCONNECT_ON_CLOSE,
1889 &config->runtime_flags);
1893 if (info->attrs[NBD_ATTR_SOCKETS]) {
1894 struct nlattr *attr;
1897 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1899 struct nlattr *socks[NBD_SOCK_MAX+1];
1901 if (nla_type(attr) != NBD_SOCK_ITEM) {
1902 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1906 ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1907 nbd_sock_policy, info->extack);
1909 printk(KERN_ERR "nbd: error processing sock list\n");
1913 if (!socks[NBD_SOCK_FD])
1915 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1916 ret = nbd_reconnect_socket(nbd, fd);
1922 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
1926 mutex_unlock(&nbd->config_lock);
1927 nbd_config_put(nbd);
1934 static const struct genl_ops nbd_connect_genl_ops[] = {
1936 .cmd = NBD_CMD_CONNECT,
1937 .policy = nbd_attr_policy,
1938 .doit = nbd_genl_connect,
1941 .cmd = NBD_CMD_DISCONNECT,
1942 .policy = nbd_attr_policy,
1943 .doit = nbd_genl_disconnect,
1946 .cmd = NBD_CMD_RECONFIGURE,
1947 .policy = nbd_attr_policy,
1948 .doit = nbd_genl_reconfigure,
1951 .cmd = NBD_CMD_STATUS,
1952 .policy = nbd_attr_policy,
1953 .doit = nbd_genl_status,
1957 static const struct genl_multicast_group nbd_mcast_grps[] = {
1958 { .name = NBD_GENL_MCAST_GROUP_NAME, },
1961 static struct genl_family nbd_genl_family __ro_after_init = {
1963 .name = NBD_GENL_FAMILY_NAME,
1964 .version = NBD_GENL_VERSION,
1965 .module = THIS_MODULE,
1966 .ops = nbd_connect_genl_ops,
1967 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
1968 .maxattr = NBD_ATTR_MAX,
1969 .mcgrps = nbd_mcast_grps,
1970 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
1973 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
1975 struct nlattr *dev_opt;
1979 /* This is a little racey, but for status it's ok. The
1980 * reason we don't take a ref here is because we can't
1981 * take a ref in the index == -1 case as we would need
1982 * to put under the nbd_index_mutex, which could
1983 * deadlock if we are configured to remove ourselves
1984 * once we're disconnected.
1986 if (refcount_read(&nbd->config_refs))
1988 dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM);
1991 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
1994 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
1998 nla_nest_end(reply, dev_opt);
2002 static int status_cb(int id, void *ptr, void *data)
2004 struct nbd_device *nbd = ptr;
2005 return populate_nbd_status(nbd, (struct sk_buff *)data);
2008 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2010 struct nlattr *dev_list;
2011 struct sk_buff *reply;
2017 if (info->attrs[NBD_ATTR_INDEX])
2018 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2020 mutex_lock(&nbd_index_mutex);
2022 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2023 nla_attr_size(sizeof(u8)));
2024 msg_size *= (index == -1) ? nbd_total_devices : 1;
2026 reply = genlmsg_new(msg_size, GFP_KERNEL);
2029 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2036 dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST);
2038 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2044 struct nbd_device *nbd;
2045 nbd = idr_find(&nbd_index_idr, index);
2047 ret = populate_nbd_status(nbd, reply);
2054 nla_nest_end(reply, dev_list);
2055 genlmsg_end(reply, reply_head);
2056 genlmsg_reply(reply, info);
2059 mutex_unlock(&nbd_index_mutex);
2063 static void nbd_connect_reply(struct genl_info *info, int index)
2065 struct sk_buff *skb;
2069 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2072 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2078 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2083 genlmsg_end(skb, msg_head);
2084 genlmsg_reply(skb, info);
2087 static void nbd_mcast_index(int index)
2089 struct sk_buff *skb;
2093 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2096 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2102 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2107 genlmsg_end(skb, msg_head);
2108 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2111 static void nbd_dead_link_work(struct work_struct *work)
2113 struct link_dead_args *args = container_of(work, struct link_dead_args,
2115 nbd_mcast_index(args->index);
2119 static int __init nbd_init(void)
2123 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2126 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2132 part_shift = fls(max_part);
2135 * Adjust max_part according to part_shift as it is exported
2136 * to user space so that user can know the max number of
2137 * partition kernel should be able to manage.
2139 * Note that -1 is required because partition 0 is reserved
2140 * for the whole disk.
2142 max_part = (1UL << part_shift) - 1;
2145 if ((1UL << part_shift) > DISK_MAX_PARTS)
2148 if (nbds_max > 1UL << (MINORBITS - part_shift))
2150 recv_workqueue = alloc_workqueue("knbd-recv",
2151 WQ_MEM_RECLAIM | WQ_HIGHPRI |
2153 if (!recv_workqueue)
2156 if (register_blkdev(NBD_MAJOR, "nbd")) {
2157 destroy_workqueue(recv_workqueue);
2161 if (genl_register_family(&nbd_genl_family)) {
2162 unregister_blkdev(NBD_MAJOR, "nbd");
2163 destroy_workqueue(recv_workqueue);
2168 mutex_lock(&nbd_index_mutex);
2169 for (i = 0; i < nbds_max; i++)
2171 mutex_unlock(&nbd_index_mutex);
2175 static int nbd_exit_cb(int id, void *ptr, void *data)
2177 struct list_head *list = (struct list_head *)data;
2178 struct nbd_device *nbd = ptr;
2180 list_add_tail(&nbd->list, list);
2184 static void __exit nbd_cleanup(void)
2186 struct nbd_device *nbd;
2187 LIST_HEAD(del_list);
2191 mutex_lock(&nbd_index_mutex);
2192 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2193 mutex_unlock(&nbd_index_mutex);
2195 while (!list_empty(&del_list)) {
2196 nbd = list_first_entry(&del_list, struct nbd_device, list);
2197 list_del_init(&nbd->list);
2198 if (refcount_read(&nbd->refs) != 1)
2199 printk(KERN_ERR "nbd: possibly leaking a device\n");
2203 idr_destroy(&nbd_index_idr);
2204 genl_unregister_family(&nbd_genl_family);
2205 destroy_workqueue(recv_workqueue);
2206 unregister_blkdev(NBD_MAJOR, "nbd");
2209 module_init(nbd_init);
2210 module_exit(nbd_cleanup);
2212 MODULE_DESCRIPTION("Network Block Device");
2213 MODULE_LICENSE("GPL");
2215 module_param(nbds_max, int, 0444);
2216 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2217 module_param(max_part, int, 0444);
2218 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");