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
82 unsigned long runtime_flags;
83 u64 dead_conn_timeout;
85 struct nbd_sock **socks;
87 atomic_t live_connections;
88 wait_queue_head_t conn_wait;
90 atomic_t recv_threads;
91 wait_queue_head_t recv_wq;
94 #if IS_ENABLED(CONFIG_DEBUG_FS)
95 struct dentry *dbg_dir;
100 struct blk_mq_tag_set tag_set;
103 refcount_t config_refs;
105 struct nbd_config *config;
106 struct mutex config_lock;
107 struct gendisk *disk;
109 struct list_head list;
110 struct task_struct *task_recv;
111 struct task_struct *task_setup;
115 struct nbd_device *nbd;
118 struct completion send_complete;
122 #if IS_ENABLED(CONFIG_DEBUG_FS)
123 static struct dentry *nbd_dbg_dir;
126 #define nbd_name(nbd) ((nbd)->disk->disk_name)
128 #define NBD_MAGIC 0x68797548
130 static unsigned int nbds_max = 16;
131 static int max_part = 16;
132 static struct workqueue_struct *recv_workqueue;
133 static int part_shift;
135 static int nbd_dev_dbg_init(struct nbd_device *nbd);
136 static void nbd_dev_dbg_close(struct nbd_device *nbd);
137 static void nbd_config_put(struct nbd_device *nbd);
138 static void nbd_connect_reply(struct genl_info *info, int index);
139 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
140 static void nbd_dead_link_work(struct work_struct *work);
142 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
144 return disk_to_dev(nbd->disk);
147 static const char *nbdcmd_to_ascii(int cmd)
150 case NBD_CMD_READ: return "read";
151 case NBD_CMD_WRITE: return "write";
152 case NBD_CMD_DISC: return "disconnect";
153 case NBD_CMD_FLUSH: return "flush";
154 case NBD_CMD_TRIM: return "trim/discard";
159 static ssize_t pid_show(struct device *dev,
160 struct device_attribute *attr, char *buf)
162 struct gendisk *disk = dev_to_disk(dev);
163 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
165 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
168 static const struct device_attribute pid_attr = {
169 .attr = { .name = "pid", .mode = S_IRUGO},
173 static void nbd_dev_remove(struct nbd_device *nbd)
175 struct gendisk *disk = nbd->disk;
178 blk_cleanup_queue(disk->queue);
179 blk_mq_free_tag_set(&nbd->tag_set);
180 disk->private_data = NULL;
186 static void nbd_put(struct nbd_device *nbd)
188 if (refcount_dec_and_mutex_lock(&nbd->refs,
190 idr_remove(&nbd_index_idr, nbd->index);
191 mutex_unlock(&nbd_index_mutex);
196 static int nbd_disconnected(struct nbd_config *config)
198 return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
199 test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
202 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
205 if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
206 struct link_dead_args *args;
207 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
209 INIT_WORK(&args->work, nbd_dead_link_work);
210 args->index = nbd->index;
211 queue_work(system_wq, &args->work);
215 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
216 atomic_dec(&nbd->config->live_connections);
219 nsock->pending = NULL;
223 static void nbd_size_clear(struct nbd_device *nbd)
225 if (nbd->config->bytesize) {
226 set_capacity(nbd->disk, 0);
227 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
231 static void nbd_size_update(struct nbd_device *nbd)
233 struct nbd_config *config = nbd->config;
234 blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
235 blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
236 set_capacity(nbd->disk, config->bytesize >> 9);
237 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
240 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
243 struct nbd_config *config = nbd->config;
244 config->blksize = blocksize;
245 config->bytesize = blocksize * nr_blocks;
246 nbd_size_update(nbd);
249 static void nbd_complete_rq(struct request *req)
251 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
253 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd,
254 cmd->status ? "failed" : "done");
256 blk_mq_end_request(req, cmd->status);
260 * Forcibly shutdown the socket causing all listeners to error
262 static void sock_shutdown(struct nbd_device *nbd)
264 struct nbd_config *config = nbd->config;
267 if (config->num_connections == 0)
269 if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
272 for (i = 0; i < config->num_connections; i++) {
273 struct nbd_sock *nsock = config->socks[i];
274 mutex_lock(&nsock->tx_lock);
275 nbd_mark_nsock_dead(nbd, nsock, 0);
276 mutex_unlock(&nsock->tx_lock);
278 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
281 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
284 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
285 struct nbd_device *nbd = cmd->nbd;
286 struct nbd_config *config;
288 if (!refcount_inc_not_zero(&nbd->config_refs)) {
289 cmd->status = BLK_STS_TIMEOUT;
290 return BLK_EH_HANDLED;
293 /* If we are waiting on our dead timer then we could get timeout
294 * callbacks for our request. For this we just want to reset the timer
295 * and let the queue side take care of everything.
297 if (!completion_done(&cmd->send_complete)) {
299 return BLK_EH_RESET_TIMER;
301 config = nbd->config;
303 if (config->num_connections > 1) {
304 dev_err_ratelimited(nbd_to_dev(nbd),
305 "Connection timed out, retrying\n");
307 * Hooray we have more connections, requeue this IO, the submit
308 * path will put it on a real connection.
310 if (config->socks && config->num_connections > 1) {
311 if (cmd->index < config->num_connections) {
312 struct nbd_sock *nsock =
313 config->socks[cmd->index];
314 mutex_lock(&nsock->tx_lock);
315 /* We can have multiple outstanding requests, so
316 * we don't want to mark the nsock dead if we've
317 * already reconnected with a new socket, so
318 * only mark it dead if its the same socket we
321 if (cmd->cookie == nsock->cookie)
322 nbd_mark_nsock_dead(nbd, nsock, 1);
323 mutex_unlock(&nsock->tx_lock);
325 blk_mq_requeue_request(req, true);
327 return BLK_EH_NOT_HANDLED;
330 dev_err_ratelimited(nbd_to_dev(nbd),
331 "Connection timed out\n");
333 set_bit(NBD_TIMEDOUT, &config->runtime_flags);
334 cmd->status = BLK_STS_IOERR;
338 return BLK_EH_HANDLED;
342 * Send or receive packet.
344 static int sock_xmit(struct nbd_device *nbd, int index, int send,
345 struct iov_iter *iter, int msg_flags, int *sent)
347 struct nbd_config *config = nbd->config;
348 struct socket *sock = config->socks[index]->sock;
351 unsigned int noreclaim_flag;
353 if (unlikely(!sock)) {
354 dev_err_ratelimited(disk_to_dev(nbd->disk),
355 "Attempted %s on closed socket in sock_xmit\n",
356 (send ? "send" : "recv"));
360 msg.msg_iter = *iter;
362 noreclaim_flag = memalloc_noreclaim_save();
364 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
367 msg.msg_control = NULL;
368 msg.msg_controllen = 0;
369 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
372 result = sock_sendmsg(sock, &msg);
374 result = sock_recvmsg(sock, &msg, msg.msg_flags);
378 result = -EPIPE; /* short read */
383 } while (msg_data_left(&msg));
385 memalloc_noreclaim_restore(noreclaim_flag);
390 /* always call with the tx_lock held */
391 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
393 struct request *req = blk_mq_rq_from_pdu(cmd);
394 struct nbd_config *config = nbd->config;
395 struct nbd_sock *nsock = config->socks[index];
397 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
398 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
399 struct iov_iter from;
400 unsigned long size = blk_rq_bytes(req);
403 u32 nbd_cmd_flags = 0;
404 u32 tag = blk_mq_unique_tag(req);
405 int sent = nsock->sent, skip = 0;
407 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
409 switch (req_op(req)) {
414 type = NBD_CMD_FLUSH;
417 type = NBD_CMD_WRITE;
426 if (rq_data_dir(req) == WRITE &&
427 (config->flags & NBD_FLAG_READ_ONLY)) {
428 dev_err_ratelimited(disk_to_dev(nbd->disk),
429 "Write on read-only\n");
433 if (req->cmd_flags & REQ_FUA)
434 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
436 /* We did a partial send previously, and we at least sent the whole
437 * request struct, so just go and send the rest of the pages in the
441 if (sent >= sizeof(request)) {
442 skip = sent - sizeof(request);
445 iov_iter_advance(&from, sent);
448 cmd->cookie = nsock->cookie;
449 request.type = htonl(type | nbd_cmd_flags);
450 if (type != NBD_CMD_FLUSH) {
451 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
452 request.len = htonl(size);
454 memcpy(request.handle, &tag, sizeof(tag));
456 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
457 cmd, nbdcmd_to_ascii(type),
458 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
459 result = sock_xmit(nbd, index, 1, &from,
460 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
462 if (result == -ERESTARTSYS) {
463 /* If we havne't sent anything we can just return BUSY,
464 * however if we have sent something we need to make
465 * sure we only allow this req to be sent until we are
469 nsock->pending = req;
472 return BLK_STS_RESOURCE;
474 dev_err_ratelimited(disk_to_dev(nbd->disk),
475 "Send control failed (result %d)\n", result);
479 if (type != NBD_CMD_WRITE)
484 struct bio *next = bio->bi_next;
485 struct bvec_iter iter;
488 bio_for_each_segment(bvec, bio, iter) {
489 bool is_last = !next && bio_iter_last(bvec, iter);
490 int flags = is_last ? 0 : MSG_MORE;
492 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
494 iov_iter_bvec(&from, ITER_BVEC | WRITE,
495 &bvec, 1, bvec.bv_len);
497 if (skip >= iov_iter_count(&from)) {
498 skip -= iov_iter_count(&from);
501 iov_iter_advance(&from, skip);
504 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
506 if (result == -ERESTARTSYS) {
507 /* We've already sent the header, we
508 * have no choice but to set pending and
511 nsock->pending = req;
513 return BLK_STS_RESOURCE;
515 dev_err(disk_to_dev(nbd->disk),
516 "Send data failed (result %d)\n",
521 * The completion might already have come in,
522 * so break for the last one instead of letting
523 * the iterator do it. This prevents use-after-free
532 nsock->pending = NULL;
537 /* NULL returned = something went wrong, inform userspace */
538 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
540 struct nbd_config *config = nbd->config;
542 struct nbd_reply reply;
544 struct request *req = NULL;
547 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
551 iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
552 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
554 if (!nbd_disconnected(config))
555 dev_err(disk_to_dev(nbd->disk),
556 "Receive control failed (result %d)\n", result);
557 return ERR_PTR(result);
560 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
561 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
562 (unsigned long)ntohl(reply.magic));
563 return ERR_PTR(-EPROTO);
566 memcpy(&tag, reply.handle, sizeof(u32));
568 hwq = blk_mq_unique_tag_to_hwq(tag);
569 if (hwq < nbd->tag_set.nr_hw_queues)
570 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
571 blk_mq_unique_tag_to_tag(tag));
572 if (!req || !blk_mq_request_started(req)) {
573 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
575 return ERR_PTR(-ENOENT);
577 cmd = blk_mq_rq_to_pdu(req);
578 if (ntohl(reply.error)) {
579 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
581 cmd->status = BLK_STS_IOERR;
585 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
586 if (rq_data_dir(req) != WRITE) {
587 struct req_iterator iter;
590 rq_for_each_segment(bvec, req, iter) {
591 iov_iter_bvec(&to, ITER_BVEC | READ,
592 &bvec, 1, bvec.bv_len);
593 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
595 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
598 * If we've disconnected or we only have 1
599 * connection then we need to make sure we
600 * complete this request, otherwise error out
601 * and let the timeout stuff handle resubmitting
602 * this request onto another connection.
604 if (nbd_disconnected(config) ||
605 config->num_connections <= 1) {
606 cmd->status = BLK_STS_IOERR;
609 return ERR_PTR(-EIO);
611 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
615 /* See the comment in nbd_queue_rq. */
616 wait_for_completion(&cmd->send_complete);
621 static void recv_work(struct work_struct *work)
623 struct recv_thread_args *args = container_of(work,
624 struct recv_thread_args,
626 struct nbd_device *nbd = args->nbd;
627 struct nbd_config *config = nbd->config;
631 cmd = nbd_read_stat(nbd, args->index);
633 struct nbd_sock *nsock = config->socks[args->index];
635 mutex_lock(&nsock->tx_lock);
636 nbd_mark_nsock_dead(nbd, nsock, 1);
637 mutex_unlock(&nsock->tx_lock);
641 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
643 atomic_dec(&config->recv_threads);
644 wake_up(&config->recv_wq);
649 static void nbd_clear_req(struct request *req, void *data, bool reserved)
653 if (!blk_mq_request_started(req))
655 cmd = blk_mq_rq_to_pdu(req);
656 cmd->status = BLK_STS_IOERR;
657 blk_mq_complete_request(req);
660 static void nbd_clear_que(struct nbd_device *nbd)
662 blk_mq_quiesce_queue(nbd->disk->queue);
663 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
664 blk_mq_unquiesce_queue(nbd->disk->queue);
665 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
668 static int find_fallback(struct nbd_device *nbd, int index)
670 struct nbd_config *config = nbd->config;
672 struct nbd_sock *nsock = config->socks[index];
673 int fallback = nsock->fallback_index;
675 if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
678 if (config->num_connections <= 1) {
679 dev_err_ratelimited(disk_to_dev(nbd->disk),
680 "Attempted send on invalid socket\n");
684 if (fallback >= 0 && fallback < config->num_connections &&
685 !config->socks[fallback]->dead)
688 if (nsock->fallback_index < 0 ||
689 nsock->fallback_index >= config->num_connections ||
690 config->socks[nsock->fallback_index]->dead) {
692 for (i = 0; i < config->num_connections; i++) {
695 if (!config->socks[i]->dead) {
700 nsock->fallback_index = new_index;
702 dev_err_ratelimited(disk_to_dev(nbd->disk),
703 "Dead connection, failed to find a fallback\n");
707 new_index = nsock->fallback_index;
711 static int wait_for_reconnect(struct nbd_device *nbd)
713 struct nbd_config *config = nbd->config;
714 if (!config->dead_conn_timeout)
716 if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
718 wait_event_interruptible_timeout(config->conn_wait,
719 atomic_read(&config->live_connections),
720 config->dead_conn_timeout);
721 return atomic_read(&config->live_connections);
724 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
726 struct request *req = blk_mq_rq_from_pdu(cmd);
727 struct nbd_device *nbd = cmd->nbd;
728 struct nbd_config *config;
729 struct nbd_sock *nsock;
732 if (!refcount_inc_not_zero(&nbd->config_refs)) {
733 dev_err_ratelimited(disk_to_dev(nbd->disk),
734 "Socks array is empty\n");
737 config = nbd->config;
739 if (index >= config->num_connections) {
740 dev_err_ratelimited(disk_to_dev(nbd->disk),
741 "Attempted send on invalid socket\n");
745 cmd->status = BLK_STS_OK;
747 nsock = config->socks[index];
748 mutex_lock(&nsock->tx_lock);
750 int old_index = index;
751 index = find_fallback(nbd, index);
752 mutex_unlock(&nsock->tx_lock);
754 if (wait_for_reconnect(nbd)) {
758 /* All the sockets should already be down at this point,
759 * we just want to make sure that DISCONNECTED is set so
760 * any requests that come in that were queue'ed waiting
761 * for the reconnect timer don't trigger the timer again
762 * and instead just error out.
771 /* Handle the case that we have a pending request that was partially
772 * transmitted that _has_ to be serviced first. We need to call requeue
773 * here so that it gets put _after_ the request that is already on the
776 if (unlikely(nsock->pending && nsock->pending != req)) {
777 blk_mq_requeue_request(req, true);
782 * Some failures are related to the link going down, so anything that
783 * returns EAGAIN can be retried on a different socket.
785 ret = nbd_send_cmd(nbd, cmd, index);
786 if (ret == -EAGAIN) {
787 dev_err_ratelimited(disk_to_dev(nbd->disk),
788 "Request send failed trying another connection\n");
789 nbd_mark_nsock_dead(nbd, nsock, 1);
790 mutex_unlock(&nsock->tx_lock);
794 mutex_unlock(&nsock->tx_lock);
799 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
800 const struct blk_mq_queue_data *bd)
802 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
806 * Since we look at the bio's to send the request over the network we
807 * need to make sure the completion work doesn't mark this request done
808 * before we are done doing our send. This keeps us from dereferencing
809 * freed data if we have particularly fast completions (ie we get the
810 * completion before we exit sock_xmit on the last bvec) or in the case
811 * that the server is misbehaving (or there was an error) before we're
812 * done sending everything over the wire.
814 init_completion(&cmd->send_complete);
815 blk_mq_start_request(bd->rq);
817 /* We can be called directly from the user space process, which means we
818 * could possibly have signals pending so our sendmsg will fail. In
819 * this case we need to return that we are busy, otherwise error out as
822 ret = nbd_handle_cmd(cmd, hctx->queue_num);
827 complete(&cmd->send_complete);
832 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
835 struct nbd_config *config = nbd->config;
837 struct nbd_sock **socks;
838 struct nbd_sock *nsock;
841 sock = sockfd_lookup(arg, &err);
845 if (!netlink && !nbd->task_setup &&
846 !test_bit(NBD_BOUND, &config->runtime_flags))
847 nbd->task_setup = current;
850 (nbd->task_setup != current ||
851 test_bit(NBD_BOUND, &config->runtime_flags))) {
852 dev_err(disk_to_dev(nbd->disk),
853 "Device being setup by another task");
858 socks = krealloc(config->socks, (config->num_connections + 1) *
859 sizeof(struct nbd_sock *), GFP_KERNEL);
864 nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
870 config->socks = socks;
872 nsock->fallback_index = -1;
874 mutex_init(&nsock->tx_lock);
876 nsock->pending = NULL;
879 socks[config->num_connections++] = nsock;
880 atomic_inc(&config->live_connections);
885 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
887 struct nbd_config *config = nbd->config;
888 struct socket *sock, *old;
889 struct recv_thread_args *args;
893 sock = sockfd_lookup(arg, &err);
897 args = kzalloc(sizeof(*args), GFP_KERNEL);
903 for (i = 0; i < config->num_connections; i++) {
904 struct nbd_sock *nsock = config->socks[i];
909 mutex_lock(&nsock->tx_lock);
911 mutex_unlock(&nsock->tx_lock);
914 sk_set_memalloc(sock->sk);
915 if (nbd->tag_set.timeout)
916 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
917 atomic_inc(&config->recv_threads);
918 refcount_inc(&nbd->config_refs);
920 nsock->fallback_index = -1;
923 INIT_WORK(&args->work, recv_work);
927 mutex_unlock(&nsock->tx_lock);
930 clear_bit(NBD_DISCONNECTED, &config->runtime_flags);
932 /* We take the tx_mutex in an error path in the recv_work, so we
933 * need to queue_work outside of the tx_mutex.
935 queue_work(recv_workqueue, &args->work);
937 atomic_inc(&config->live_connections);
938 wake_up(&config->conn_wait);
946 static void nbd_bdev_reset(struct block_device *bdev)
948 if (bdev->bd_openers > 1)
950 bd_set_size(bdev, 0);
952 blkdev_reread_part(bdev);
953 bdev->bd_invalidated = 1;
957 static void nbd_parse_flags(struct nbd_device *nbd)
959 struct nbd_config *config = nbd->config;
960 if (config->flags & NBD_FLAG_READ_ONLY)
961 set_disk_ro(nbd->disk, true);
963 set_disk_ro(nbd->disk, false);
964 if (config->flags & NBD_FLAG_SEND_TRIM)
965 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
966 if (config->flags & NBD_FLAG_SEND_FLUSH) {
967 if (config->flags & NBD_FLAG_SEND_FUA)
968 blk_queue_write_cache(nbd->disk->queue, true, true);
970 blk_queue_write_cache(nbd->disk->queue, true, false);
973 blk_queue_write_cache(nbd->disk->queue, false, false);
976 static void send_disconnects(struct nbd_device *nbd)
978 struct nbd_config *config = nbd->config;
979 struct nbd_request request = {
980 .magic = htonl(NBD_REQUEST_MAGIC),
981 .type = htonl(NBD_CMD_DISC),
983 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
984 struct iov_iter from;
987 for (i = 0; i < config->num_connections; i++) {
988 struct nbd_sock *nsock = config->socks[i];
990 iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
991 mutex_lock(&nsock->tx_lock);
992 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
994 dev_err(disk_to_dev(nbd->disk),
995 "Send disconnect failed %d\n", ret);
996 mutex_unlock(&nsock->tx_lock);
1000 static int nbd_disconnect(struct nbd_device *nbd)
1002 struct nbd_config *config = nbd->config;
1004 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1005 set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
1006 send_disconnects(nbd);
1010 static void nbd_clear_sock(struct nbd_device *nbd)
1014 nbd->task_setup = NULL;
1017 static void nbd_config_put(struct nbd_device *nbd)
1019 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1020 &nbd->config_lock)) {
1021 struct nbd_config *config = nbd->config;
1022 nbd_dev_dbg_close(nbd);
1023 nbd_size_clear(nbd);
1024 if (test_and_clear_bit(NBD_HAS_PID_FILE,
1025 &config->runtime_flags))
1026 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1027 nbd->task_recv = NULL;
1028 nbd_clear_sock(nbd);
1029 if (config->num_connections) {
1031 for (i = 0; i < config->num_connections; i++) {
1032 sockfd_put(config->socks[i]->sock);
1033 kfree(config->socks[i]);
1035 kfree(config->socks);
1040 nbd->tag_set.timeout = 0;
1041 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1043 mutex_unlock(&nbd->config_lock);
1045 module_put(THIS_MODULE);
1049 static int nbd_start_device(struct nbd_device *nbd)
1051 struct nbd_config *config = nbd->config;
1052 int num_connections = config->num_connections;
1059 if (num_connections > 1 &&
1060 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1061 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1065 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1066 nbd->task_recv = current;
1068 nbd_parse_flags(nbd);
1070 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1072 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1075 set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1077 nbd_dev_dbg_init(nbd);
1078 for (i = 0; i < num_connections; i++) {
1079 struct recv_thread_args *args;
1081 args = kzalloc(sizeof(*args), GFP_KERNEL);
1086 sk_set_memalloc(config->socks[i]->sock->sk);
1087 if (nbd->tag_set.timeout)
1088 config->socks[i]->sock->sk->sk_sndtimeo =
1089 nbd->tag_set.timeout;
1090 atomic_inc(&config->recv_threads);
1091 refcount_inc(&nbd->config_refs);
1092 INIT_WORK(&args->work, recv_work);
1095 queue_work(recv_workqueue, &args->work);
1100 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1102 struct nbd_config *config = nbd->config;
1105 ret = nbd_start_device(nbd);
1109 bd_set_size(bdev, config->bytesize);
1111 bdev->bd_invalidated = 1;
1112 mutex_unlock(&nbd->config_lock);
1113 ret = wait_event_interruptible(config->recv_wq,
1114 atomic_read(&config->recv_threads) == 0);
1117 mutex_lock(&nbd->config_lock);
1118 bd_set_size(bdev, 0);
1119 /* user requested, ignore socket errors */
1120 if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1122 if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1127 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1128 struct block_device *bdev)
1132 nbd_bdev_reset(bdev);
1133 if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1134 &nbd->config->runtime_flags))
1135 nbd_config_put(nbd);
1138 /* Must be called with config_lock held */
1139 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1140 unsigned int cmd, unsigned long arg)
1142 struct nbd_config *config = nbd->config;
1145 case NBD_DISCONNECT:
1146 return nbd_disconnect(nbd);
1147 case NBD_CLEAR_SOCK:
1148 nbd_clear_sock_ioctl(nbd, bdev);
1151 return nbd_add_socket(nbd, arg, false);
1152 case NBD_SET_BLKSIZE:
1153 nbd_size_set(nbd, arg,
1154 div_s64(config->bytesize, arg));
1157 nbd_size_set(nbd, config->blksize,
1158 div_s64(arg, config->blksize));
1160 case NBD_SET_SIZE_BLOCKS:
1161 nbd_size_set(nbd, config->blksize, arg);
1163 case NBD_SET_TIMEOUT:
1165 nbd->tag_set.timeout = arg * HZ;
1166 blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1171 config->flags = arg;
1174 return nbd_start_device_ioctl(nbd, bdev);
1177 * This is for compatibility only. The queue is always cleared
1178 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1181 case NBD_PRINT_DEBUG:
1183 * For compatibility only, we no longer keep a list of
1184 * outstanding requests.
1191 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1192 unsigned int cmd, unsigned long arg)
1194 struct nbd_device *nbd = bdev->bd_disk->private_data;
1195 struct nbd_config *config = nbd->config;
1196 int error = -EINVAL;
1198 if (!capable(CAP_SYS_ADMIN))
1201 /* The block layer will pass back some non-nbd ioctls in case we have
1202 * special handling for them, but we don't so just return an error.
1204 if (_IOC_TYPE(cmd) != 0xab)
1207 mutex_lock(&nbd->config_lock);
1209 /* Don't allow ioctl operations on a nbd device that was created with
1210 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1212 if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1213 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1214 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1216 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1217 mutex_unlock(&nbd->config_lock);
1221 static struct nbd_config *nbd_alloc_config(void)
1223 struct nbd_config *config;
1225 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1228 atomic_set(&config->recv_threads, 0);
1229 init_waitqueue_head(&config->recv_wq);
1230 init_waitqueue_head(&config->conn_wait);
1231 config->blksize = 1024;
1232 atomic_set(&config->live_connections, 0);
1233 try_module_get(THIS_MODULE);
1237 static int nbd_open(struct block_device *bdev, fmode_t mode)
1239 struct nbd_device *nbd;
1242 mutex_lock(&nbd_index_mutex);
1243 nbd = bdev->bd_disk->private_data;
1248 if (!refcount_inc_not_zero(&nbd->refs)) {
1252 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1253 struct nbd_config *config;
1255 mutex_lock(&nbd->config_lock);
1256 if (refcount_inc_not_zero(&nbd->config_refs)) {
1257 mutex_unlock(&nbd->config_lock);
1260 config = nbd->config = nbd_alloc_config();
1263 mutex_unlock(&nbd->config_lock);
1266 refcount_set(&nbd->config_refs, 1);
1267 refcount_inc(&nbd->refs);
1268 mutex_unlock(&nbd->config_lock);
1271 mutex_unlock(&nbd_index_mutex);
1275 static void nbd_release(struct gendisk *disk, fmode_t mode)
1277 struct nbd_device *nbd = disk->private_data;
1278 nbd_config_put(nbd);
1282 static const struct block_device_operations nbd_fops =
1284 .owner = THIS_MODULE,
1286 .release = nbd_release,
1288 .compat_ioctl = nbd_ioctl,
1291 #if IS_ENABLED(CONFIG_DEBUG_FS)
1293 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1295 struct nbd_device *nbd = s->private;
1298 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1303 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1305 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1308 static const struct file_operations nbd_dbg_tasks_ops = {
1309 .open = nbd_dbg_tasks_open,
1311 .llseek = seq_lseek,
1312 .release = single_release,
1315 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1317 struct nbd_device *nbd = s->private;
1318 u32 flags = nbd->config->flags;
1320 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1322 seq_puts(s, "Known flags:\n");
1324 if (flags & NBD_FLAG_HAS_FLAGS)
1325 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1326 if (flags & NBD_FLAG_READ_ONLY)
1327 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1328 if (flags & NBD_FLAG_SEND_FLUSH)
1329 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1330 if (flags & NBD_FLAG_SEND_FUA)
1331 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1332 if (flags & NBD_FLAG_SEND_TRIM)
1333 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1338 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1340 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1343 static const struct file_operations nbd_dbg_flags_ops = {
1344 .open = nbd_dbg_flags_open,
1346 .llseek = seq_lseek,
1347 .release = single_release,
1350 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1353 struct nbd_config *config = nbd->config;
1358 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1360 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1364 config->dbg_dir = dir;
1366 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1367 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1368 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1369 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1370 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1375 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1377 debugfs_remove_recursive(nbd->config->dbg_dir);
1380 static int nbd_dbg_init(void)
1382 struct dentry *dbg_dir;
1384 dbg_dir = debugfs_create_dir("nbd", NULL);
1388 nbd_dbg_dir = dbg_dir;
1393 static void nbd_dbg_close(void)
1395 debugfs_remove_recursive(nbd_dbg_dir);
1398 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1400 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1405 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1409 static int nbd_dbg_init(void)
1414 static void nbd_dbg_close(void)
1420 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1421 unsigned int hctx_idx, unsigned int numa_node)
1423 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1424 cmd->nbd = set->driver_data;
1428 static const struct blk_mq_ops nbd_mq_ops = {
1429 .queue_rq = nbd_queue_rq,
1430 .complete = nbd_complete_rq,
1431 .init_request = nbd_init_request,
1432 .timeout = nbd_xmit_timeout,
1435 static int nbd_dev_add(int index)
1437 struct nbd_device *nbd;
1438 struct gendisk *disk;
1439 struct request_queue *q;
1442 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1446 disk = alloc_disk(1 << part_shift);
1451 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1456 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1465 nbd->tag_set.ops = &nbd_mq_ops;
1466 nbd->tag_set.nr_hw_queues = 1;
1467 nbd->tag_set.queue_depth = 128;
1468 nbd->tag_set.numa_node = NUMA_NO_NODE;
1469 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1470 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1471 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1472 nbd->tag_set.driver_data = nbd;
1474 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1478 q = blk_mq_init_queue(&nbd->tag_set);
1486 * Tell the block layer that we are not a rotational device
1488 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1489 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1490 disk->queue->limits.discard_granularity = 512;
1491 blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1492 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1493 blk_queue_max_segments(disk->queue, USHRT_MAX);
1494 blk_queue_max_hw_sectors(disk->queue, 65536);
1495 disk->queue->limits.max_sectors = 256;
1497 mutex_init(&nbd->config_lock);
1498 refcount_set(&nbd->config_refs, 0);
1499 refcount_set(&nbd->refs, 1);
1500 INIT_LIST_HEAD(&nbd->list);
1501 disk->major = NBD_MAJOR;
1502 disk->first_minor = index << part_shift;
1503 disk->fops = &nbd_fops;
1504 disk->private_data = nbd;
1505 sprintf(disk->disk_name, "nbd%d", index);
1507 nbd_total_devices++;
1511 blk_mq_free_tag_set(&nbd->tag_set);
1513 idr_remove(&nbd_index_idr, index);
1522 static int find_free_cb(int id, void *ptr, void *data)
1524 struct nbd_device *nbd = ptr;
1525 struct nbd_device **found = data;
1527 if (!refcount_read(&nbd->config_refs)) {
1534 /* Netlink interface. */
1535 static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1536 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1537 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1538 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1539 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1540 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1541 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1542 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1543 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1544 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1547 static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1548 [NBD_SOCK_FD] = { .type = NLA_U32 },
1551 /* We don't use this right now since we don't parse the incoming list, but we
1552 * still want it here so userspace knows what to expect.
1554 static struct nla_policy __attribute__((unused))
1555 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1556 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1557 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1560 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1562 struct nbd_device *nbd = NULL;
1563 struct nbd_config *config;
1566 bool put_dev = false;
1568 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1571 if (info->attrs[NBD_ATTR_INDEX])
1572 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1573 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1574 printk(KERN_ERR "nbd: must specify at least one socket\n");
1577 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1578 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1582 mutex_lock(&nbd_index_mutex);
1584 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1587 new_index = nbd_dev_add(-1);
1588 if (new_index < 0) {
1589 mutex_unlock(&nbd_index_mutex);
1590 printk(KERN_ERR "nbd: failed to add new device\n");
1593 nbd = idr_find(&nbd_index_idr, new_index);
1596 nbd = idr_find(&nbd_index_idr, index);
1598 ret = nbd_dev_add(index);
1600 mutex_unlock(&nbd_index_mutex);
1601 printk(KERN_ERR "nbd: failed to add new device\n");
1604 nbd = idr_find(&nbd_index_idr, index);
1608 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1610 mutex_unlock(&nbd_index_mutex);
1613 if (!refcount_inc_not_zero(&nbd->refs)) {
1614 mutex_unlock(&nbd_index_mutex);
1617 printk(KERN_ERR "nbd: device at index %d is going down\n",
1621 mutex_unlock(&nbd_index_mutex);
1623 mutex_lock(&nbd->config_lock);
1624 if (refcount_read(&nbd->config_refs)) {
1625 mutex_unlock(&nbd->config_lock);
1629 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1632 if (WARN_ON(nbd->config)) {
1633 mutex_unlock(&nbd->config_lock);
1637 config = nbd->config = nbd_alloc_config();
1639 mutex_unlock(&nbd->config_lock);
1641 printk(KERN_ERR "nbd: couldn't allocate config\n");
1644 refcount_set(&nbd->config_refs, 1);
1645 set_bit(NBD_BOUND, &config->runtime_flags);
1647 if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1648 u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1649 nbd_size_set(nbd, config->blksize,
1650 div64_u64(bytes, config->blksize));
1652 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1654 nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1655 nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1657 if (info->attrs[NBD_ATTR_TIMEOUT]) {
1658 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1659 nbd->tag_set.timeout = timeout * HZ;
1660 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1662 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1663 config->dead_conn_timeout =
1664 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1665 config->dead_conn_timeout *= HZ;
1667 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1669 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1670 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1671 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1672 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1673 set_bit(NBD_DESTROY_ON_DISCONNECT,
1674 &config->runtime_flags);
1679 if (info->attrs[NBD_ATTR_SOCKETS]) {
1680 struct nlattr *attr;
1683 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1685 struct nlattr *socks[NBD_SOCK_MAX+1];
1687 if (nla_type(attr) != NBD_SOCK_ITEM) {
1688 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1692 ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1693 nbd_sock_policy, info->extack);
1695 printk(KERN_ERR "nbd: error processing sock list\n");
1699 if (!socks[NBD_SOCK_FD])
1701 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1702 ret = nbd_add_socket(nbd, fd, true);
1707 ret = nbd_start_device(nbd);
1709 mutex_unlock(&nbd->config_lock);
1711 set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1712 refcount_inc(&nbd->config_refs);
1713 nbd_connect_reply(info, nbd->index);
1715 nbd_config_put(nbd);
1721 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1723 struct nbd_device *nbd;
1726 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1729 if (!info->attrs[NBD_ATTR_INDEX]) {
1730 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1733 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1734 mutex_lock(&nbd_index_mutex);
1735 nbd = idr_find(&nbd_index_idr, index);
1737 mutex_unlock(&nbd_index_mutex);
1738 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1742 if (!refcount_inc_not_zero(&nbd->refs)) {
1743 mutex_unlock(&nbd_index_mutex);
1744 printk(KERN_ERR "nbd: device at index %d is going down\n",
1748 mutex_unlock(&nbd_index_mutex);
1749 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1753 mutex_lock(&nbd->config_lock);
1754 nbd_disconnect(nbd);
1755 mutex_unlock(&nbd->config_lock);
1756 if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1757 &nbd->config->runtime_flags))
1758 nbd_config_put(nbd);
1759 nbd_config_put(nbd);
1764 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1766 struct nbd_device *nbd = NULL;
1767 struct nbd_config *config;
1770 bool put_dev = false;
1772 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1775 if (!info->attrs[NBD_ATTR_INDEX]) {
1776 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1779 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1780 mutex_lock(&nbd_index_mutex);
1781 nbd = idr_find(&nbd_index_idr, index);
1783 mutex_unlock(&nbd_index_mutex);
1784 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1788 if (!refcount_inc_not_zero(&nbd->refs)) {
1789 mutex_unlock(&nbd_index_mutex);
1790 printk(KERN_ERR "nbd: device at index %d is going down\n",
1794 mutex_unlock(&nbd_index_mutex);
1796 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1797 dev_err(nbd_to_dev(nbd),
1798 "not configured, cannot reconfigure\n");
1803 mutex_lock(&nbd->config_lock);
1804 config = nbd->config;
1805 if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1807 dev_err(nbd_to_dev(nbd),
1808 "not configured, cannot reconfigure\n");
1812 if (info->attrs[NBD_ATTR_TIMEOUT]) {
1813 u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1814 nbd->tag_set.timeout = timeout * HZ;
1815 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1817 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1818 config->dead_conn_timeout =
1819 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1820 config->dead_conn_timeout *= HZ;
1822 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1823 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1824 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1825 if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1826 &config->runtime_flags))
1829 if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1830 &config->runtime_flags))
1831 refcount_inc(&nbd->refs);
1835 if (info->attrs[NBD_ATTR_SOCKETS]) {
1836 struct nlattr *attr;
1839 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1841 struct nlattr *socks[NBD_SOCK_MAX+1];
1843 if (nla_type(attr) != NBD_SOCK_ITEM) {
1844 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1848 ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1849 nbd_sock_policy, info->extack);
1851 printk(KERN_ERR "nbd: error processing sock list\n");
1855 if (!socks[NBD_SOCK_FD])
1857 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1858 ret = nbd_reconnect_socket(nbd, fd);
1864 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
1868 mutex_unlock(&nbd->config_lock);
1869 nbd_config_put(nbd);
1876 static const struct genl_ops nbd_connect_genl_ops[] = {
1878 .cmd = NBD_CMD_CONNECT,
1879 .policy = nbd_attr_policy,
1880 .doit = nbd_genl_connect,
1883 .cmd = NBD_CMD_DISCONNECT,
1884 .policy = nbd_attr_policy,
1885 .doit = nbd_genl_disconnect,
1888 .cmd = NBD_CMD_RECONFIGURE,
1889 .policy = nbd_attr_policy,
1890 .doit = nbd_genl_reconfigure,
1893 .cmd = NBD_CMD_STATUS,
1894 .policy = nbd_attr_policy,
1895 .doit = nbd_genl_status,
1899 static const struct genl_multicast_group nbd_mcast_grps[] = {
1900 { .name = NBD_GENL_MCAST_GROUP_NAME, },
1903 static struct genl_family nbd_genl_family __ro_after_init = {
1905 .name = NBD_GENL_FAMILY_NAME,
1906 .version = NBD_GENL_VERSION,
1907 .module = THIS_MODULE,
1908 .ops = nbd_connect_genl_ops,
1909 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
1910 .maxattr = NBD_ATTR_MAX,
1911 .mcgrps = nbd_mcast_grps,
1912 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
1915 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
1917 struct nlattr *dev_opt;
1921 /* This is a little racey, but for status it's ok. The
1922 * reason we don't take a ref here is because we can't
1923 * take a ref in the index == -1 case as we would need
1924 * to put under the nbd_index_mutex, which could
1925 * deadlock if we are configured to remove ourselves
1926 * once we're disconnected.
1928 if (refcount_read(&nbd->config_refs))
1930 dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM);
1933 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
1936 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
1940 nla_nest_end(reply, dev_opt);
1944 static int status_cb(int id, void *ptr, void *data)
1946 struct nbd_device *nbd = ptr;
1947 return populate_nbd_status(nbd, (struct sk_buff *)data);
1950 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
1952 struct nlattr *dev_list;
1953 struct sk_buff *reply;
1959 if (info->attrs[NBD_ATTR_INDEX])
1960 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1962 mutex_lock(&nbd_index_mutex);
1964 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
1965 nla_attr_size(sizeof(u8)));
1966 msg_size *= (index == -1) ? nbd_total_devices : 1;
1968 reply = genlmsg_new(msg_size, GFP_KERNEL);
1971 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
1978 dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST);
1980 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
1986 struct nbd_device *nbd;
1987 nbd = idr_find(&nbd_index_idr, index);
1989 ret = populate_nbd_status(nbd, reply);
1996 nla_nest_end(reply, dev_list);
1997 genlmsg_end(reply, reply_head);
1998 genlmsg_reply(reply, info);
2001 mutex_unlock(&nbd_index_mutex);
2005 static void nbd_connect_reply(struct genl_info *info, int index)
2007 struct sk_buff *skb;
2011 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2014 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2020 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2025 genlmsg_end(skb, msg_head);
2026 genlmsg_reply(skb, info);
2029 static void nbd_mcast_index(int index)
2031 struct sk_buff *skb;
2035 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2038 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2044 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2049 genlmsg_end(skb, msg_head);
2050 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2053 static void nbd_dead_link_work(struct work_struct *work)
2055 struct link_dead_args *args = container_of(work, struct link_dead_args,
2057 nbd_mcast_index(args->index);
2061 static int __init nbd_init(void)
2065 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2068 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2074 part_shift = fls(max_part);
2077 * Adjust max_part according to part_shift as it is exported
2078 * to user space so that user can know the max number of
2079 * partition kernel should be able to manage.
2081 * Note that -1 is required because partition 0 is reserved
2082 * for the whole disk.
2084 max_part = (1UL << part_shift) - 1;
2087 if ((1UL << part_shift) > DISK_MAX_PARTS)
2090 if (nbds_max > 1UL << (MINORBITS - part_shift))
2092 recv_workqueue = alloc_workqueue("knbd-recv",
2093 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2094 if (!recv_workqueue)
2097 if (register_blkdev(NBD_MAJOR, "nbd")) {
2098 destroy_workqueue(recv_workqueue);
2102 if (genl_register_family(&nbd_genl_family)) {
2103 unregister_blkdev(NBD_MAJOR, "nbd");
2104 destroy_workqueue(recv_workqueue);
2109 mutex_lock(&nbd_index_mutex);
2110 for (i = 0; i < nbds_max; i++)
2112 mutex_unlock(&nbd_index_mutex);
2116 static int nbd_exit_cb(int id, void *ptr, void *data)
2118 struct list_head *list = (struct list_head *)data;
2119 struct nbd_device *nbd = ptr;
2121 list_add_tail(&nbd->list, list);
2125 static void __exit nbd_cleanup(void)
2127 struct nbd_device *nbd;
2128 LIST_HEAD(del_list);
2132 mutex_lock(&nbd_index_mutex);
2133 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2134 mutex_unlock(&nbd_index_mutex);
2136 while (!list_empty(&del_list)) {
2137 nbd = list_first_entry(&del_list, struct nbd_device, list);
2138 list_del_init(&nbd->list);
2139 if (refcount_read(&nbd->refs) != 1)
2140 printk(KERN_ERR "nbd: possibly leaking a device\n");
2144 idr_destroy(&nbd_index_idr);
2145 genl_unregister_family(&nbd_genl_family);
2146 destroy_workqueue(recv_workqueue);
2147 unregister_blkdev(NBD_MAJOR, "nbd");
2150 module_init(nbd_init);
2151 module_exit(nbd_cleanup);
2153 MODULE_DESCRIPTION("Network Block Device");
2154 MODULE_LICENSE("GPL");
2156 module_param(nbds_max, int, 0444);
2157 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2158 module_param(max_part, int, 0444);
2159 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");