1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Network block device - make block devices work over TCP
5 * Note that you can not swap over this thing, yet. Seems to work but
6 * deadlocks sometimes - you can not swap over TCP in general.
8 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
9 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
11 * (part of code stolen from loop.c)
14 #include <linux/major.h>
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched.h>
20 #include <linux/sched/mm.h>
22 #include <linux/bio.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/file.h>
26 #include <linux/ioctl.h>
27 #include <linux/mutex.h>
28 #include <linux/compiler.h>
29 #include <linux/completion.h>
30 #include <linux/err.h>
31 #include <linux/kernel.h>
32 #include <linux/slab.h>
34 #include <linux/net.h>
35 #include <linux/kthread.h>
36 #include <linux/types.h>
37 #include <linux/debugfs.h>
38 #include <linux/blk-mq.h>
40 #include <linux/uaccess.h>
41 #include <asm/types.h>
43 #include <linux/nbd.h>
44 #include <linux/nbd-netlink.h>
45 #include <net/genetlink.h>
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/nbd.h>
50 static DEFINE_IDR(nbd_index_idr);
51 static DEFINE_MUTEX(nbd_index_mutex);
52 static int nbd_total_devices = 0;
57 struct request *pending;
64 struct recv_thread_args {
65 struct work_struct work;
66 struct nbd_device *nbd;
70 struct link_dead_args {
71 struct work_struct work;
75 #define NBD_RT_TIMEDOUT 0
76 #define NBD_RT_DISCONNECT_REQUESTED 1
77 #define NBD_RT_DISCONNECTED 2
78 #define NBD_RT_HAS_PID_FILE 3
79 #define NBD_RT_HAS_CONFIG_REF 4
80 #define NBD_RT_BOUND 5
81 #define NBD_RT_DESTROY_ON_DISCONNECT 6
82 #define NBD_RT_DISCONNECT_ON_CLOSE 7
84 #define NBD_DESTROY_ON_DISCONNECT 0
85 #define NBD_DISCONNECT_REQUESTED 1
89 unsigned long runtime_flags;
90 u64 dead_conn_timeout;
92 struct nbd_sock **socks;
94 atomic_t live_connections;
95 wait_queue_head_t conn_wait;
97 atomic_t recv_threads;
98 wait_queue_head_t recv_wq;
101 #if IS_ENABLED(CONFIG_DEBUG_FS)
102 struct dentry *dbg_dir;
107 struct blk_mq_tag_set tag_set;
110 refcount_t config_refs;
112 struct nbd_config *config;
113 struct mutex config_lock;
114 struct gendisk *disk;
115 struct workqueue_struct *recv_workq;
117 struct list_head list;
118 struct task_struct *task_recv;
119 struct task_struct *task_setup;
121 struct completion *destroy_complete;
125 #define NBD_CMD_REQUEUED 1
128 struct nbd_device *nbd;
138 #if IS_ENABLED(CONFIG_DEBUG_FS)
139 static struct dentry *nbd_dbg_dir;
142 #define nbd_name(nbd) ((nbd)->disk->disk_name)
144 #define NBD_MAGIC 0x68797548
146 #define NBD_DEF_BLKSIZE 1024
148 static unsigned int nbds_max = 16;
149 static int max_part = 16;
150 static int part_shift;
152 static int nbd_dev_dbg_init(struct nbd_device *nbd);
153 static void nbd_dev_dbg_close(struct nbd_device *nbd);
154 static void nbd_config_put(struct nbd_device *nbd);
155 static void nbd_connect_reply(struct genl_info *info, int index);
156 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
157 static void nbd_dead_link_work(struct work_struct *work);
158 static void nbd_disconnect_and_put(struct nbd_device *nbd);
160 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
162 return disk_to_dev(nbd->disk);
165 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
167 struct request *req = blk_mq_rq_from_pdu(cmd);
169 if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
170 blk_mq_requeue_request(req, true);
173 #define NBD_COOKIE_BITS 32
175 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
177 struct request *req = blk_mq_rq_from_pdu(cmd);
178 u32 tag = blk_mq_unique_tag(req);
179 u64 cookie = cmd->cmd_cookie;
181 return (cookie << NBD_COOKIE_BITS) | tag;
184 static u32 nbd_handle_to_tag(u64 handle)
189 static u32 nbd_handle_to_cookie(u64 handle)
191 return (u32)(handle >> NBD_COOKIE_BITS);
194 static const char *nbdcmd_to_ascii(int cmd)
197 case NBD_CMD_READ: return "read";
198 case NBD_CMD_WRITE: return "write";
199 case NBD_CMD_DISC: return "disconnect";
200 case NBD_CMD_FLUSH: return "flush";
201 case NBD_CMD_TRIM: return "trim/discard";
206 static ssize_t pid_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
209 struct gendisk *disk = dev_to_disk(dev);
210 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
212 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
215 static const struct device_attribute pid_attr = {
216 .attr = { .name = "pid", .mode = 0444},
220 static void nbd_dev_remove(struct nbd_device *nbd)
222 struct gendisk *disk = nbd->disk;
223 struct request_queue *q;
228 blk_cleanup_queue(q);
229 blk_mq_free_tag_set(&nbd->tag_set);
230 disk->private_data = NULL;
235 * Place this in the last just before the nbd is freed to
236 * make sure that the disk and the related kobject are also
237 * totally removed to avoid duplicate creation of the same
240 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) && nbd->destroy_complete)
241 complete(nbd->destroy_complete);
246 static void nbd_put(struct nbd_device *nbd)
248 if (refcount_dec_and_mutex_lock(&nbd->refs,
250 idr_remove(&nbd_index_idr, nbd->index);
252 mutex_unlock(&nbd_index_mutex);
256 static int nbd_disconnected(struct nbd_config *config)
258 return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
259 test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
262 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
265 if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
266 struct link_dead_args *args;
267 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
269 INIT_WORK(&args->work, nbd_dead_link_work);
270 args->index = nbd->index;
271 queue_work(system_wq, &args->work);
275 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
276 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
277 if (test_and_clear_bit(NBD_RT_DISCONNECT_REQUESTED,
278 &nbd->config->runtime_flags)) {
279 set_bit(NBD_RT_DISCONNECTED,
280 &nbd->config->runtime_flags);
281 dev_info(nbd_to_dev(nbd),
282 "Disconnected due to user request.\n");
287 nsock->pending = NULL;
291 static void nbd_size_clear(struct nbd_device *nbd)
293 if (nbd->config->bytesize) {
294 set_capacity(nbd->disk, 0);
295 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
299 static void nbd_size_update(struct nbd_device *nbd)
301 struct nbd_config *config = nbd->config;
302 struct block_device *bdev = bdget_disk(nbd->disk, 0);
304 if (config->flags & NBD_FLAG_SEND_TRIM) {
305 nbd->disk->queue->limits.discard_granularity = config->blksize;
306 nbd->disk->queue->limits.discard_alignment = config->blksize;
307 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
309 blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
310 blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
311 set_capacity(nbd->disk, config->bytesize >> 9);
314 bd_set_size(bdev, config->bytesize);
315 set_blocksize(bdev, config->blksize);
317 bdev->bd_invalidated = 1;
320 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
323 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
326 struct nbd_config *config = nbd->config;
327 config->blksize = blocksize;
328 config->bytesize = blocksize * nr_blocks;
329 if (nbd->task_recv != NULL)
330 nbd_size_update(nbd);
333 static void nbd_complete_rq(struct request *req)
335 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
337 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
338 cmd->status ? "failed" : "done");
340 blk_mq_end_request(req, cmd->status);
344 * Forcibly shutdown the socket causing all listeners to error
346 static void sock_shutdown(struct nbd_device *nbd)
348 struct nbd_config *config = nbd->config;
351 if (config->num_connections == 0)
353 if (test_and_set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
356 for (i = 0; i < config->num_connections; i++) {
357 struct nbd_sock *nsock = config->socks[i];
358 mutex_lock(&nsock->tx_lock);
359 nbd_mark_nsock_dead(nbd, nsock, 0);
360 mutex_unlock(&nsock->tx_lock);
362 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
365 static u32 req_to_nbd_cmd_type(struct request *req)
367 switch (req_op(req)) {
371 return NBD_CMD_FLUSH;
373 return NBD_CMD_WRITE;
381 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
384 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
385 struct nbd_device *nbd = cmd->nbd;
386 struct nbd_config *config;
388 if (!mutex_trylock(&cmd->lock))
389 return BLK_EH_RESET_TIMER;
391 if (!refcount_inc_not_zero(&nbd->config_refs)) {
392 cmd->status = BLK_STS_TIMEOUT;
393 mutex_unlock(&cmd->lock);
396 config = nbd->config;
398 if (config->num_connections > 1) {
399 dev_err_ratelimited(nbd_to_dev(nbd),
400 "Connection timed out, retrying (%d/%d alive)\n",
401 atomic_read(&config->live_connections),
402 config->num_connections);
404 * Hooray we have more connections, requeue this IO, the submit
405 * path will put it on a real connection.
407 if (config->socks && config->num_connections > 1) {
408 if (cmd->index < config->num_connections) {
409 struct nbd_sock *nsock =
410 config->socks[cmd->index];
411 mutex_lock(&nsock->tx_lock);
412 /* We can have multiple outstanding requests, so
413 * we don't want to mark the nsock dead if we've
414 * already reconnected with a new socket, so
415 * only mark it dead if its the same socket we
418 if (cmd->cookie == nsock->cookie)
419 nbd_mark_nsock_dead(nbd, nsock, 1);
420 mutex_unlock(&nsock->tx_lock);
422 mutex_unlock(&cmd->lock);
423 nbd_requeue_cmd(cmd);
429 if (!nbd->tag_set.timeout) {
431 * Userspace sets timeout=0 to disable socket disconnection,
432 * so just warn and reset the timer.
435 dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
436 req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
437 (unsigned long long)blk_rq_pos(req) << 9,
438 blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
440 mutex_unlock(&cmd->lock);
442 return BLK_EH_RESET_TIMER;
445 dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
446 set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
447 cmd->status = BLK_STS_IOERR;
448 mutex_unlock(&cmd->lock);
452 blk_mq_complete_request(req);
457 * Send or receive packet.
459 static int sock_xmit(struct nbd_device *nbd, int index, int send,
460 struct iov_iter *iter, int msg_flags, int *sent)
462 struct nbd_config *config = nbd->config;
463 struct socket *sock = config->socks[index]->sock;
466 unsigned int noreclaim_flag;
468 if (unlikely(!sock)) {
469 dev_err_ratelimited(disk_to_dev(nbd->disk),
470 "Attempted %s on closed socket in sock_xmit\n",
471 (send ? "send" : "recv"));
475 msg.msg_iter = *iter;
477 noreclaim_flag = memalloc_noreclaim_save();
479 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
482 msg.msg_control = NULL;
483 msg.msg_controllen = 0;
484 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
487 result = sock_sendmsg(sock, &msg);
489 result = sock_recvmsg(sock, &msg, msg.msg_flags);
493 result = -EPIPE; /* short read */
498 } while (msg_data_left(&msg));
500 memalloc_noreclaim_restore(noreclaim_flag);
506 * Different settings for sk->sk_sndtimeo can result in different return values
507 * if there is a signal pending when we enter sendmsg, because reasons?
509 static inline int was_interrupted(int result)
511 return result == -ERESTARTSYS || result == -EINTR;
514 /* always call with the tx_lock held */
515 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
517 struct request *req = blk_mq_rq_from_pdu(cmd);
518 struct nbd_config *config = nbd->config;
519 struct nbd_sock *nsock = config->socks[index];
521 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
522 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
523 struct iov_iter from;
524 unsigned long size = blk_rq_bytes(req);
528 u32 nbd_cmd_flags = 0;
529 int sent = nsock->sent, skip = 0;
531 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
533 type = req_to_nbd_cmd_type(req);
537 if (rq_data_dir(req) == WRITE &&
538 (config->flags & NBD_FLAG_READ_ONLY)) {
539 dev_err_ratelimited(disk_to_dev(nbd->disk),
540 "Write on read-only\n");
544 if (req->cmd_flags & REQ_FUA)
545 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
547 /* We did a partial send previously, and we at least sent the whole
548 * request struct, so just go and send the rest of the pages in the
552 if (sent >= sizeof(request)) {
553 skip = sent - sizeof(request);
555 /* initialize handle for tracing purposes */
556 handle = nbd_cmd_handle(cmd);
560 iov_iter_advance(&from, sent);
565 cmd->cookie = nsock->cookie;
567 request.type = htonl(type | nbd_cmd_flags);
568 if (type != NBD_CMD_FLUSH) {
569 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
570 request.len = htonl(size);
572 handle = nbd_cmd_handle(cmd);
573 memcpy(request.handle, &handle, sizeof(handle));
575 trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
577 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
578 req, nbdcmd_to_ascii(type),
579 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
580 result = sock_xmit(nbd, index, 1, &from,
581 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
582 trace_nbd_header_sent(req, handle);
584 if (was_interrupted(result)) {
585 /* If we havne't sent anything we can just return BUSY,
586 * however if we have sent something we need to make
587 * sure we only allow this req to be sent until we are
591 nsock->pending = req;
594 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
595 return BLK_STS_RESOURCE;
597 dev_err_ratelimited(disk_to_dev(nbd->disk),
598 "Send control failed (result %d)\n", result);
602 if (type != NBD_CMD_WRITE)
607 struct bio *next = bio->bi_next;
608 struct bvec_iter iter;
611 bio_for_each_segment(bvec, bio, iter) {
612 bool is_last = !next && bio_iter_last(bvec, iter);
613 int flags = is_last ? 0 : MSG_MORE;
615 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
617 iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
619 if (skip >= iov_iter_count(&from)) {
620 skip -= iov_iter_count(&from);
623 iov_iter_advance(&from, skip);
626 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
628 if (was_interrupted(result)) {
629 /* We've already sent the header, we
630 * have no choice but to set pending and
633 nsock->pending = req;
635 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
636 return BLK_STS_RESOURCE;
638 dev_err(disk_to_dev(nbd->disk),
639 "Send data failed (result %d)\n",
644 * The completion might already have come in,
645 * so break for the last one instead of letting
646 * the iterator do it. This prevents use-after-free
655 trace_nbd_payload_sent(req, handle);
656 nsock->pending = NULL;
661 /* NULL returned = something went wrong, inform userspace */
662 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
664 struct nbd_config *config = nbd->config;
666 struct nbd_reply reply;
668 struct request *req = NULL;
672 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
677 iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
678 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
680 if (!nbd_disconnected(config))
681 dev_err(disk_to_dev(nbd->disk),
682 "Receive control failed (result %d)\n", result);
683 return ERR_PTR(result);
686 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
687 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
688 (unsigned long)ntohl(reply.magic));
689 return ERR_PTR(-EPROTO);
692 memcpy(&handle, reply.handle, sizeof(handle));
693 tag = nbd_handle_to_tag(handle);
694 hwq = blk_mq_unique_tag_to_hwq(tag);
695 if (hwq < nbd->tag_set.nr_hw_queues)
696 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
697 blk_mq_unique_tag_to_tag(tag));
698 if (!req || !blk_mq_request_started(req)) {
699 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
701 return ERR_PTR(-ENOENT);
703 trace_nbd_header_received(req, handle);
704 cmd = blk_mq_rq_to_pdu(req);
706 mutex_lock(&cmd->lock);
707 if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
708 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
709 req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
713 if (cmd->status != BLK_STS_OK) {
714 dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
719 if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
720 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
725 if (ntohl(reply.error)) {
726 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
728 cmd->status = BLK_STS_IOERR;
732 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
733 if (rq_data_dir(req) != WRITE) {
734 struct req_iterator iter;
737 rq_for_each_segment(bvec, req, iter) {
738 iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
739 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
741 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
744 * If we've disconnected or we only have 1
745 * connection then we need to make sure we
746 * complete this request, otherwise error out
747 * and let the timeout stuff handle resubmitting
748 * this request onto another connection.
750 if (nbd_disconnected(config) ||
751 config->num_connections <= 1) {
752 cmd->status = BLK_STS_IOERR;
758 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
763 trace_nbd_payload_received(req, handle);
764 mutex_unlock(&cmd->lock);
765 return ret ? ERR_PTR(ret) : cmd;
768 static void recv_work(struct work_struct *work)
770 struct recv_thread_args *args = container_of(work,
771 struct recv_thread_args,
773 struct nbd_device *nbd = args->nbd;
774 struct nbd_config *config = nbd->config;
778 cmd = nbd_read_stat(nbd, args->index);
780 struct nbd_sock *nsock = config->socks[args->index];
782 mutex_lock(&nsock->tx_lock);
783 nbd_mark_nsock_dead(nbd, nsock, 1);
784 mutex_unlock(&nsock->tx_lock);
788 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
790 atomic_dec(&config->recv_threads);
791 wake_up(&config->recv_wq);
796 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
798 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
800 mutex_lock(&cmd->lock);
801 cmd->status = BLK_STS_IOERR;
802 mutex_unlock(&cmd->lock);
804 blk_mq_complete_request(req);
808 static void nbd_clear_que(struct nbd_device *nbd)
810 blk_mq_quiesce_queue(nbd->disk->queue);
811 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
812 blk_mq_unquiesce_queue(nbd->disk->queue);
813 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
816 static int find_fallback(struct nbd_device *nbd, int index)
818 struct nbd_config *config = nbd->config;
820 struct nbd_sock *nsock = config->socks[index];
821 int fallback = nsock->fallback_index;
823 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
826 if (config->num_connections <= 1) {
827 dev_err_ratelimited(disk_to_dev(nbd->disk),
828 "Attempted send on invalid socket\n");
832 if (fallback >= 0 && fallback < config->num_connections &&
833 !config->socks[fallback]->dead)
836 if (nsock->fallback_index < 0 ||
837 nsock->fallback_index >= config->num_connections ||
838 config->socks[nsock->fallback_index]->dead) {
840 for (i = 0; i < config->num_connections; i++) {
843 if (!config->socks[i]->dead) {
848 nsock->fallback_index = new_index;
850 dev_err_ratelimited(disk_to_dev(nbd->disk),
851 "Dead connection, failed to find a fallback\n");
855 new_index = nsock->fallback_index;
859 static int wait_for_reconnect(struct nbd_device *nbd)
861 struct nbd_config *config = nbd->config;
862 if (!config->dead_conn_timeout)
864 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
866 return wait_event_timeout(config->conn_wait,
867 atomic_read(&config->live_connections) > 0,
868 config->dead_conn_timeout) > 0;
871 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
873 struct request *req = blk_mq_rq_from_pdu(cmd);
874 struct nbd_device *nbd = cmd->nbd;
875 struct nbd_config *config;
876 struct nbd_sock *nsock;
879 if (!refcount_inc_not_zero(&nbd->config_refs)) {
880 dev_err_ratelimited(disk_to_dev(nbd->disk),
881 "Socks array is empty\n");
882 blk_mq_start_request(req);
885 config = nbd->config;
887 if (index >= config->num_connections) {
888 dev_err_ratelimited(disk_to_dev(nbd->disk),
889 "Attempted send on invalid socket\n");
891 blk_mq_start_request(req);
894 cmd->status = BLK_STS_OK;
896 nsock = config->socks[index];
897 mutex_lock(&nsock->tx_lock);
899 int old_index = index;
900 index = find_fallback(nbd, index);
901 mutex_unlock(&nsock->tx_lock);
903 if (wait_for_reconnect(nbd)) {
907 /* All the sockets should already be down at this point,
908 * we just want to make sure that DISCONNECTED is set so
909 * any requests that come in that were queue'ed waiting
910 * for the reconnect timer don't trigger the timer again
911 * and instead just error out.
915 blk_mq_start_request(req);
921 /* Handle the case that we have a pending request that was partially
922 * transmitted that _has_ to be serviced first. We need to call requeue
923 * here so that it gets put _after_ the request that is already on the
926 blk_mq_start_request(req);
927 if (unlikely(nsock->pending && nsock->pending != req)) {
928 nbd_requeue_cmd(cmd);
933 * Some failures are related to the link going down, so anything that
934 * returns EAGAIN can be retried on a different socket.
936 ret = nbd_send_cmd(nbd, cmd, index);
937 if (ret == -EAGAIN) {
938 dev_err_ratelimited(disk_to_dev(nbd->disk),
939 "Request send failed, requeueing\n");
940 nbd_mark_nsock_dead(nbd, nsock, 1);
941 nbd_requeue_cmd(cmd);
945 mutex_unlock(&nsock->tx_lock);
950 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
951 const struct blk_mq_queue_data *bd)
953 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
957 * Since we look at the bio's to send the request over the network we
958 * need to make sure the completion work doesn't mark this request done
959 * before we are done doing our send. This keeps us from dereferencing
960 * freed data if we have particularly fast completions (ie we get the
961 * completion before we exit sock_xmit on the last bvec) or in the case
962 * that the server is misbehaving (or there was an error) before we're
963 * done sending everything over the wire.
965 mutex_lock(&cmd->lock);
966 clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
968 /* We can be called directly from the user space process, which means we
969 * could possibly have signals pending so our sendmsg will fail. In
970 * this case we need to return that we are busy, otherwise error out as
973 ret = nbd_handle_cmd(cmd, hctx->queue_num);
978 mutex_unlock(&cmd->lock);
983 static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
989 sock = sockfd_lookup(fd, err);
993 if (sock->ops->shutdown == sock_no_shutdown) {
994 dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
1003 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1006 struct nbd_config *config = nbd->config;
1007 struct socket *sock;
1008 struct nbd_sock **socks;
1009 struct nbd_sock *nsock;
1012 sock = nbd_get_socket(nbd, arg, &err);
1016 if (!netlink && !nbd->task_setup &&
1017 !test_bit(NBD_RT_BOUND, &config->runtime_flags))
1018 nbd->task_setup = current;
1021 (nbd->task_setup != current ||
1022 test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1023 dev_err(disk_to_dev(nbd->disk),
1024 "Device being setup by another task");
1029 socks = krealloc(config->socks, (config->num_connections + 1) *
1030 sizeof(struct nbd_sock *), GFP_KERNEL);
1036 config->socks = socks;
1038 nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
1044 nsock->fallback_index = -1;
1045 nsock->dead = false;
1046 mutex_init(&nsock->tx_lock);
1048 nsock->pending = NULL;
1051 socks[config->num_connections++] = nsock;
1052 atomic_inc(&config->live_connections);
1057 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1059 struct nbd_config *config = nbd->config;
1060 struct socket *sock, *old;
1061 struct recv_thread_args *args;
1065 sock = nbd_get_socket(nbd, arg, &err);
1069 args = kzalloc(sizeof(*args), GFP_KERNEL);
1075 for (i = 0; i < config->num_connections; i++) {
1076 struct nbd_sock *nsock = config->socks[i];
1081 mutex_lock(&nsock->tx_lock);
1083 mutex_unlock(&nsock->tx_lock);
1086 sk_set_memalloc(sock->sk);
1087 if (nbd->tag_set.timeout)
1088 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1089 atomic_inc(&config->recv_threads);
1090 refcount_inc(&nbd->config_refs);
1092 nsock->fallback_index = -1;
1094 nsock->dead = false;
1095 INIT_WORK(&args->work, recv_work);
1099 mutex_unlock(&nsock->tx_lock);
1102 clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1104 /* We take the tx_mutex in an error path in the recv_work, so we
1105 * need to queue_work outside of the tx_mutex.
1107 queue_work(nbd->recv_workq, &args->work);
1109 atomic_inc(&config->live_connections);
1110 wake_up(&config->conn_wait);
1118 static void nbd_bdev_reset(struct block_device *bdev)
1120 if (bdev->bd_openers > 1)
1122 bd_set_size(bdev, 0);
1125 static void nbd_parse_flags(struct nbd_device *nbd)
1127 struct nbd_config *config = nbd->config;
1128 if (config->flags & NBD_FLAG_READ_ONLY)
1129 set_disk_ro(nbd->disk, true);
1131 set_disk_ro(nbd->disk, false);
1132 if (config->flags & NBD_FLAG_SEND_TRIM)
1133 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1134 if (config->flags & NBD_FLAG_SEND_FLUSH) {
1135 if (config->flags & NBD_FLAG_SEND_FUA)
1136 blk_queue_write_cache(nbd->disk->queue, true, true);
1138 blk_queue_write_cache(nbd->disk->queue, true, false);
1141 blk_queue_write_cache(nbd->disk->queue, false, false);
1144 static void send_disconnects(struct nbd_device *nbd)
1146 struct nbd_config *config = nbd->config;
1147 struct nbd_request request = {
1148 .magic = htonl(NBD_REQUEST_MAGIC),
1149 .type = htonl(NBD_CMD_DISC),
1151 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1152 struct iov_iter from;
1155 for (i = 0; i < config->num_connections; i++) {
1156 struct nbd_sock *nsock = config->socks[i];
1158 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1159 mutex_lock(&nsock->tx_lock);
1160 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1162 dev_err(disk_to_dev(nbd->disk),
1163 "Send disconnect failed %d\n", ret);
1164 mutex_unlock(&nsock->tx_lock);
1168 static int nbd_disconnect(struct nbd_device *nbd)
1170 struct nbd_config *config = nbd->config;
1172 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1173 set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1174 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1175 send_disconnects(nbd);
1179 static void nbd_clear_sock(struct nbd_device *nbd)
1183 nbd->task_setup = NULL;
1186 static void nbd_config_put(struct nbd_device *nbd)
1188 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1189 &nbd->config_lock)) {
1190 struct nbd_config *config = nbd->config;
1191 nbd_dev_dbg_close(nbd);
1192 nbd_size_clear(nbd);
1193 if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1194 &config->runtime_flags))
1195 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1196 nbd->task_recv = NULL;
1197 nbd_clear_sock(nbd);
1198 if (config->num_connections) {
1200 for (i = 0; i < config->num_connections; i++) {
1201 sockfd_put(config->socks[i]->sock);
1202 kfree(config->socks[i]);
1204 kfree(config->socks);
1209 if (nbd->recv_workq)
1210 destroy_workqueue(nbd->recv_workq);
1211 nbd->recv_workq = NULL;
1213 nbd->tag_set.timeout = 0;
1214 nbd->disk->queue->limits.discard_granularity = 0;
1215 nbd->disk->queue->limits.discard_alignment = 0;
1216 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1217 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1219 mutex_unlock(&nbd->config_lock);
1221 module_put(THIS_MODULE);
1225 static int nbd_start_device(struct nbd_device *nbd)
1227 struct nbd_config *config = nbd->config;
1228 int num_connections = config->num_connections;
1235 if (num_connections > 1 &&
1236 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1237 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1241 nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1242 WQ_MEM_RECLAIM | WQ_HIGHPRI |
1243 WQ_UNBOUND, 0, nbd->index);
1244 if (!nbd->recv_workq) {
1245 dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1249 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1250 nbd->task_recv = current;
1252 nbd_parse_flags(nbd);
1254 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1256 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1259 set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1261 nbd_dev_dbg_init(nbd);
1262 for (i = 0; i < num_connections; i++) {
1263 struct recv_thread_args *args;
1265 args = kzalloc(sizeof(*args), GFP_KERNEL);
1270 sk_set_memalloc(config->socks[i]->sock->sk);
1271 if (nbd->tag_set.timeout)
1272 config->socks[i]->sock->sk->sk_sndtimeo =
1273 nbd->tag_set.timeout;
1274 atomic_inc(&config->recv_threads);
1275 refcount_inc(&nbd->config_refs);
1276 INIT_WORK(&args->work, recv_work);
1279 queue_work(nbd->recv_workq, &args->work);
1281 nbd_size_update(nbd);
1285 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1287 struct nbd_config *config = nbd->config;
1290 ret = nbd_start_device(nbd);
1295 bdev->bd_invalidated = 1;
1296 mutex_unlock(&nbd->config_lock);
1297 ret = wait_event_interruptible(config->recv_wq,
1298 atomic_read(&config->recv_threads) == 0);
1301 flush_workqueue(nbd->recv_workq);
1303 mutex_lock(&nbd->config_lock);
1304 nbd_bdev_reset(bdev);
1305 /* user requested, ignore socket errors */
1306 if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1308 if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1313 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1314 struct block_device *bdev)
1317 __invalidate_device(bdev, true);
1318 nbd_bdev_reset(bdev);
1319 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1320 &nbd->config->runtime_flags))
1321 nbd_config_put(nbd);
1324 static bool nbd_is_valid_blksize(unsigned long blksize)
1326 if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1327 blksize > PAGE_SIZE)
1332 static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1334 nbd->tag_set.timeout = timeout * HZ;
1336 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1339 /* Must be called with config_lock held */
1340 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1341 unsigned int cmd, unsigned long arg)
1343 struct nbd_config *config = nbd->config;
1346 case NBD_DISCONNECT:
1347 return nbd_disconnect(nbd);
1348 case NBD_CLEAR_SOCK:
1349 nbd_clear_sock_ioctl(nbd, bdev);
1352 return nbd_add_socket(nbd, arg, false);
1353 case NBD_SET_BLKSIZE:
1355 arg = NBD_DEF_BLKSIZE;
1356 if (!nbd_is_valid_blksize(arg))
1358 nbd_size_set(nbd, arg,
1359 div_s64(config->bytesize, arg));
1362 nbd_size_set(nbd, config->blksize,
1363 div_s64(arg, config->blksize));
1365 case NBD_SET_SIZE_BLOCKS:
1366 nbd_size_set(nbd, config->blksize, arg);
1368 case NBD_SET_TIMEOUT:
1369 nbd_set_cmd_timeout(nbd, arg);
1373 config->flags = arg;
1376 return nbd_start_device_ioctl(nbd, bdev);
1379 * This is for compatibility only. The queue is always cleared
1380 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1383 case NBD_PRINT_DEBUG:
1385 * For compatibility only, we no longer keep a list of
1386 * outstanding requests.
1393 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1394 unsigned int cmd, unsigned long arg)
1396 struct nbd_device *nbd = bdev->bd_disk->private_data;
1397 struct nbd_config *config = nbd->config;
1398 int error = -EINVAL;
1400 if (!capable(CAP_SYS_ADMIN))
1403 /* The block layer will pass back some non-nbd ioctls in case we have
1404 * special handling for them, but we don't so just return an error.
1406 if (_IOC_TYPE(cmd) != 0xab)
1409 mutex_lock(&nbd->config_lock);
1411 /* Don't allow ioctl operations on a nbd device that was created with
1412 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1414 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1415 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1416 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1418 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1419 mutex_unlock(&nbd->config_lock);
1423 static struct nbd_config *nbd_alloc_config(void)
1425 struct nbd_config *config;
1427 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1430 atomic_set(&config->recv_threads, 0);
1431 init_waitqueue_head(&config->recv_wq);
1432 init_waitqueue_head(&config->conn_wait);
1433 config->blksize = NBD_DEF_BLKSIZE;
1434 atomic_set(&config->live_connections, 0);
1435 try_module_get(THIS_MODULE);
1439 static int nbd_open(struct block_device *bdev, fmode_t mode)
1441 struct nbd_device *nbd;
1444 mutex_lock(&nbd_index_mutex);
1445 nbd = bdev->bd_disk->private_data;
1450 if (!refcount_inc_not_zero(&nbd->refs)) {
1454 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1455 struct nbd_config *config;
1457 mutex_lock(&nbd->config_lock);
1458 if (refcount_inc_not_zero(&nbd->config_refs)) {
1459 mutex_unlock(&nbd->config_lock);
1462 config = nbd->config = nbd_alloc_config();
1465 mutex_unlock(&nbd->config_lock);
1468 refcount_set(&nbd->config_refs, 1);
1469 refcount_inc(&nbd->refs);
1470 mutex_unlock(&nbd->config_lock);
1471 bdev->bd_invalidated = 1;
1472 } else if (nbd_disconnected(nbd->config)) {
1473 bdev->bd_invalidated = 1;
1476 mutex_unlock(&nbd_index_mutex);
1480 static void nbd_release(struct gendisk *disk, fmode_t mode)
1482 struct nbd_device *nbd = disk->private_data;
1483 struct block_device *bdev = bdget_disk(disk, 0);
1485 if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1486 bdev->bd_openers == 0)
1487 nbd_disconnect_and_put(nbd);
1489 nbd_config_put(nbd);
1493 static const struct block_device_operations nbd_fops =
1495 .owner = THIS_MODULE,
1497 .release = nbd_release,
1499 .compat_ioctl = nbd_ioctl,
1502 #if IS_ENABLED(CONFIG_DEBUG_FS)
1504 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1506 struct nbd_device *nbd = s->private;
1509 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1514 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1516 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1519 static const struct file_operations nbd_dbg_tasks_ops = {
1520 .open = nbd_dbg_tasks_open,
1522 .llseek = seq_lseek,
1523 .release = single_release,
1526 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1528 struct nbd_device *nbd = s->private;
1529 u32 flags = nbd->config->flags;
1531 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1533 seq_puts(s, "Known flags:\n");
1535 if (flags & NBD_FLAG_HAS_FLAGS)
1536 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1537 if (flags & NBD_FLAG_READ_ONLY)
1538 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1539 if (flags & NBD_FLAG_SEND_FLUSH)
1540 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1541 if (flags & NBD_FLAG_SEND_FUA)
1542 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1543 if (flags & NBD_FLAG_SEND_TRIM)
1544 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1549 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1551 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1554 static const struct file_operations nbd_dbg_flags_ops = {
1555 .open = nbd_dbg_flags_open,
1557 .llseek = seq_lseek,
1558 .release = single_release,
1561 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1564 struct nbd_config *config = nbd->config;
1569 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1571 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1575 config->dbg_dir = dir;
1577 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1578 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1579 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1580 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1581 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1586 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1588 debugfs_remove_recursive(nbd->config->dbg_dir);
1591 static int nbd_dbg_init(void)
1593 struct dentry *dbg_dir;
1595 dbg_dir = debugfs_create_dir("nbd", NULL);
1599 nbd_dbg_dir = dbg_dir;
1604 static void nbd_dbg_close(void)
1606 debugfs_remove_recursive(nbd_dbg_dir);
1609 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1611 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1616 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1620 static int nbd_dbg_init(void)
1625 static void nbd_dbg_close(void)
1631 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1632 unsigned int hctx_idx, unsigned int numa_node)
1634 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1635 cmd->nbd = set->driver_data;
1637 mutex_init(&cmd->lock);
1641 static const struct blk_mq_ops nbd_mq_ops = {
1642 .queue_rq = nbd_queue_rq,
1643 .complete = nbd_complete_rq,
1644 .init_request = nbd_init_request,
1645 .timeout = nbd_xmit_timeout,
1648 static int nbd_dev_add(int index)
1650 struct nbd_device *nbd;
1651 struct gendisk *disk;
1652 struct request_queue *q;
1655 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1659 disk = alloc_disk(1 << part_shift);
1664 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1669 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1678 nbd->tag_set.ops = &nbd_mq_ops;
1679 nbd->tag_set.nr_hw_queues = 1;
1680 nbd->tag_set.queue_depth = 128;
1681 nbd->tag_set.numa_node = NUMA_NO_NODE;
1682 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1683 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1685 nbd->tag_set.driver_data = nbd;
1686 nbd->destroy_complete = NULL;
1688 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1692 q = blk_mq_init_queue(&nbd->tag_set);
1700 * Tell the block layer that we are not a rotational device
1702 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1703 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1704 disk->queue->limits.discard_granularity = 0;
1705 disk->queue->limits.discard_alignment = 0;
1706 blk_queue_max_discard_sectors(disk->queue, 0);
1707 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1708 blk_queue_max_segments(disk->queue, USHRT_MAX);
1709 blk_queue_max_hw_sectors(disk->queue, 65536);
1710 disk->queue->limits.max_sectors = 256;
1712 mutex_init(&nbd->config_lock);
1713 refcount_set(&nbd->config_refs, 0);
1714 refcount_set(&nbd->refs, 1);
1715 INIT_LIST_HEAD(&nbd->list);
1716 disk->major = NBD_MAJOR;
1717 disk->first_minor = index << part_shift;
1718 disk->fops = &nbd_fops;
1719 disk->private_data = nbd;
1720 sprintf(disk->disk_name, "nbd%d", index);
1722 nbd_total_devices++;
1726 blk_mq_free_tag_set(&nbd->tag_set);
1728 idr_remove(&nbd_index_idr, index);
1737 static int find_free_cb(int id, void *ptr, void *data)
1739 struct nbd_device *nbd = ptr;
1740 struct nbd_device **found = data;
1742 if (!refcount_read(&nbd->config_refs)) {
1749 /* Netlink interface. */
1750 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1751 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1752 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1753 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1754 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1755 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1756 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1757 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1758 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1759 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1762 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1763 [NBD_SOCK_FD] = { .type = NLA_U32 },
1766 /* We don't use this right now since we don't parse the incoming list, but we
1767 * still want it here so userspace knows what to expect.
1769 static const struct nla_policy __attribute__((unused))
1770 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1771 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1772 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1775 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1777 struct nbd_config *config = nbd->config;
1778 u64 bsize = config->blksize;
1779 u64 bytes = config->bytesize;
1781 if (info->attrs[NBD_ATTR_SIZE_BYTES])
1782 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1784 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1785 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1787 bsize = NBD_DEF_BLKSIZE;
1788 if (!nbd_is_valid_blksize(bsize)) {
1789 printk(KERN_ERR "Invalid block size %llu\n", bsize);
1794 if (bytes != config->bytesize || bsize != config->blksize)
1795 nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1799 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1801 DECLARE_COMPLETION_ONSTACK(destroy_complete);
1802 struct nbd_device *nbd = NULL;
1803 struct nbd_config *config;
1806 bool put_dev = false;
1808 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1811 if (info->attrs[NBD_ATTR_INDEX])
1812 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1813 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1814 printk(KERN_ERR "nbd: must specify at least one socket\n");
1817 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1818 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1822 mutex_lock(&nbd_index_mutex);
1824 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1827 new_index = nbd_dev_add(-1);
1828 if (new_index < 0) {
1829 mutex_unlock(&nbd_index_mutex);
1830 printk(KERN_ERR "nbd: failed to add new device\n");
1833 nbd = idr_find(&nbd_index_idr, new_index);
1836 nbd = idr_find(&nbd_index_idr, index);
1838 ret = nbd_dev_add(index);
1840 mutex_unlock(&nbd_index_mutex);
1841 printk(KERN_ERR "nbd: failed to add new device\n");
1844 nbd = idr_find(&nbd_index_idr, index);
1848 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1850 mutex_unlock(&nbd_index_mutex);
1854 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1855 test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1856 nbd->destroy_complete = &destroy_complete;
1857 mutex_unlock(&nbd_index_mutex);
1859 /* Wait untill the the nbd stuff is totally destroyed */
1860 wait_for_completion(&destroy_complete);
1864 if (!refcount_inc_not_zero(&nbd->refs)) {
1865 mutex_unlock(&nbd_index_mutex);
1868 printk(KERN_ERR "nbd: device at index %d is going down\n",
1872 mutex_unlock(&nbd_index_mutex);
1874 mutex_lock(&nbd->config_lock);
1875 if (refcount_read(&nbd->config_refs)) {
1876 mutex_unlock(&nbd->config_lock);
1880 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1883 if (WARN_ON(nbd->config)) {
1884 mutex_unlock(&nbd->config_lock);
1888 config = nbd->config = nbd_alloc_config();
1890 mutex_unlock(&nbd->config_lock);
1892 printk(KERN_ERR "nbd: couldn't allocate config\n");
1895 refcount_set(&nbd->config_refs, 1);
1896 set_bit(NBD_RT_BOUND, &config->runtime_flags);
1898 ret = nbd_genl_size_set(info, nbd);
1902 if (info->attrs[NBD_ATTR_TIMEOUT])
1903 nbd_set_cmd_timeout(nbd,
1904 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1905 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1906 config->dead_conn_timeout =
1907 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1908 config->dead_conn_timeout *= HZ;
1910 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1912 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1913 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1914 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1915 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1916 set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1917 &config->runtime_flags);
1918 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1921 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1923 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1924 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1925 &config->runtime_flags);
1929 if (info->attrs[NBD_ATTR_SOCKETS]) {
1930 struct nlattr *attr;
1933 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1935 struct nlattr *socks[NBD_SOCK_MAX+1];
1937 if (nla_type(attr) != NBD_SOCK_ITEM) {
1938 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1942 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1947 printk(KERN_ERR "nbd: error processing sock list\n");
1951 if (!socks[NBD_SOCK_FD])
1953 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1954 ret = nbd_add_socket(nbd, fd, true);
1959 ret = nbd_start_device(nbd);
1961 mutex_unlock(&nbd->config_lock);
1963 set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1964 refcount_inc(&nbd->config_refs);
1965 nbd_connect_reply(info, nbd->index);
1967 nbd_config_put(nbd);
1973 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1975 mutex_lock(&nbd->config_lock);
1976 nbd_disconnect(nbd);
1977 nbd_clear_sock(nbd);
1978 mutex_unlock(&nbd->config_lock);
1980 * Make sure recv thread has finished, so it does not drop the last
1981 * config ref and try to destroy the workqueue from inside the work
1984 flush_workqueue(nbd->recv_workq);
1985 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1986 &nbd->config->runtime_flags))
1987 nbd_config_put(nbd);
1990 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1992 struct nbd_device *nbd;
1995 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1998 if (!info->attrs[NBD_ATTR_INDEX]) {
1999 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
2002 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2003 mutex_lock(&nbd_index_mutex);
2004 nbd = idr_find(&nbd_index_idr, index);
2006 mutex_unlock(&nbd_index_mutex);
2007 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2011 if (!refcount_inc_not_zero(&nbd->refs)) {
2012 mutex_unlock(&nbd_index_mutex);
2013 printk(KERN_ERR "nbd: device at index %d is going down\n",
2017 mutex_unlock(&nbd_index_mutex);
2018 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2022 nbd_disconnect_and_put(nbd);
2023 nbd_config_put(nbd);
2028 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2030 struct nbd_device *nbd = NULL;
2031 struct nbd_config *config;
2034 bool put_dev = false;
2036 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2039 if (!info->attrs[NBD_ATTR_INDEX]) {
2040 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2043 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2044 mutex_lock(&nbd_index_mutex);
2045 nbd = idr_find(&nbd_index_idr, index);
2047 mutex_unlock(&nbd_index_mutex);
2048 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2052 if (!refcount_inc_not_zero(&nbd->refs)) {
2053 mutex_unlock(&nbd_index_mutex);
2054 printk(KERN_ERR "nbd: device at index %d is going down\n",
2058 mutex_unlock(&nbd_index_mutex);
2060 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2061 dev_err(nbd_to_dev(nbd),
2062 "not configured, cannot reconfigure\n");
2067 mutex_lock(&nbd->config_lock);
2068 config = nbd->config;
2069 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2071 dev_err(nbd_to_dev(nbd),
2072 "not configured, cannot reconfigure\n");
2077 ret = nbd_genl_size_set(info, nbd);
2081 if (info->attrs[NBD_ATTR_TIMEOUT])
2082 nbd_set_cmd_timeout(nbd,
2083 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2084 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2085 config->dead_conn_timeout =
2086 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2087 config->dead_conn_timeout *= HZ;
2089 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2090 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2091 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2092 if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2093 &config->runtime_flags))
2095 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2097 if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2098 &config->runtime_flags))
2099 refcount_inc(&nbd->refs);
2100 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2103 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2104 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2105 &config->runtime_flags);
2107 clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2108 &config->runtime_flags);
2112 if (info->attrs[NBD_ATTR_SOCKETS]) {
2113 struct nlattr *attr;
2116 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2118 struct nlattr *socks[NBD_SOCK_MAX+1];
2120 if (nla_type(attr) != NBD_SOCK_ITEM) {
2121 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2125 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2130 printk(KERN_ERR "nbd: error processing sock list\n");
2134 if (!socks[NBD_SOCK_FD])
2136 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2137 ret = nbd_reconnect_socket(nbd, fd);
2143 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2147 mutex_unlock(&nbd->config_lock);
2148 nbd_config_put(nbd);
2155 static const struct genl_ops nbd_connect_genl_ops[] = {
2157 .cmd = NBD_CMD_CONNECT,
2158 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2159 .doit = nbd_genl_connect,
2162 .cmd = NBD_CMD_DISCONNECT,
2163 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2164 .doit = nbd_genl_disconnect,
2167 .cmd = NBD_CMD_RECONFIGURE,
2168 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2169 .doit = nbd_genl_reconfigure,
2172 .cmd = NBD_CMD_STATUS,
2173 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2174 .doit = nbd_genl_status,
2178 static const struct genl_multicast_group nbd_mcast_grps[] = {
2179 { .name = NBD_GENL_MCAST_GROUP_NAME, },
2182 static struct genl_family nbd_genl_family __ro_after_init = {
2184 .name = NBD_GENL_FAMILY_NAME,
2185 .version = NBD_GENL_VERSION,
2186 .module = THIS_MODULE,
2187 .ops = nbd_connect_genl_ops,
2188 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2189 .maxattr = NBD_ATTR_MAX,
2190 .policy = nbd_attr_policy,
2191 .mcgrps = nbd_mcast_grps,
2192 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2195 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2197 struct nlattr *dev_opt;
2201 /* This is a little racey, but for status it's ok. The
2202 * reason we don't take a ref here is because we can't
2203 * take a ref in the index == -1 case as we would need
2204 * to put under the nbd_index_mutex, which could
2205 * deadlock if we are configured to remove ourselves
2206 * once we're disconnected.
2208 if (refcount_read(&nbd->config_refs))
2210 dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2213 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2216 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2220 nla_nest_end(reply, dev_opt);
2224 static int status_cb(int id, void *ptr, void *data)
2226 struct nbd_device *nbd = ptr;
2227 return populate_nbd_status(nbd, (struct sk_buff *)data);
2230 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2232 struct nlattr *dev_list;
2233 struct sk_buff *reply;
2239 if (info->attrs[NBD_ATTR_INDEX])
2240 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2242 mutex_lock(&nbd_index_mutex);
2244 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2245 nla_attr_size(sizeof(u8)));
2246 msg_size *= (index == -1) ? nbd_total_devices : 1;
2248 reply = genlmsg_new(msg_size, GFP_KERNEL);
2251 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2258 dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2260 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2266 struct nbd_device *nbd;
2267 nbd = idr_find(&nbd_index_idr, index);
2269 ret = populate_nbd_status(nbd, reply);
2276 nla_nest_end(reply, dev_list);
2277 genlmsg_end(reply, reply_head);
2278 ret = genlmsg_reply(reply, info);
2280 mutex_unlock(&nbd_index_mutex);
2284 static void nbd_connect_reply(struct genl_info *info, int index)
2286 struct sk_buff *skb;
2290 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2293 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2299 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2304 genlmsg_end(skb, msg_head);
2305 genlmsg_reply(skb, info);
2308 static void nbd_mcast_index(int index)
2310 struct sk_buff *skb;
2314 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2317 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2323 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2328 genlmsg_end(skb, msg_head);
2329 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2332 static void nbd_dead_link_work(struct work_struct *work)
2334 struct link_dead_args *args = container_of(work, struct link_dead_args,
2336 nbd_mcast_index(args->index);
2340 static int __init nbd_init(void)
2344 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2347 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2353 part_shift = fls(max_part);
2356 * Adjust max_part according to part_shift as it is exported
2357 * to user space so that user can know the max number of
2358 * partition kernel should be able to manage.
2360 * Note that -1 is required because partition 0 is reserved
2361 * for the whole disk.
2363 max_part = (1UL << part_shift) - 1;
2366 if ((1UL << part_shift) > DISK_MAX_PARTS)
2369 if (nbds_max > 1UL << (MINORBITS - part_shift))
2372 if (register_blkdev(NBD_MAJOR, "nbd"))
2375 if (genl_register_family(&nbd_genl_family)) {
2376 unregister_blkdev(NBD_MAJOR, "nbd");
2381 mutex_lock(&nbd_index_mutex);
2382 for (i = 0; i < nbds_max; i++)
2384 mutex_unlock(&nbd_index_mutex);
2388 static int nbd_exit_cb(int id, void *ptr, void *data)
2390 struct list_head *list = (struct list_head *)data;
2391 struct nbd_device *nbd = ptr;
2393 list_add_tail(&nbd->list, list);
2397 static void __exit nbd_cleanup(void)
2399 struct nbd_device *nbd;
2400 LIST_HEAD(del_list);
2404 mutex_lock(&nbd_index_mutex);
2405 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2406 mutex_unlock(&nbd_index_mutex);
2408 while (!list_empty(&del_list)) {
2409 nbd = list_first_entry(&del_list, struct nbd_device, list);
2410 list_del_init(&nbd->list);
2411 if (refcount_read(&nbd->refs) != 1)
2412 printk(KERN_ERR "nbd: possibly leaking a device\n");
2416 idr_destroy(&nbd_index_idr);
2417 genl_unregister_family(&nbd_genl_family);
2418 unregister_blkdev(NBD_MAJOR, "nbd");
2421 module_init(nbd_init);
2422 module_exit(nbd_cleanup);
2424 MODULE_DESCRIPTION("Network Block Device");
2425 MODULE_LICENSE("GPL");
2427 module_param(nbds_max, int, 0444);
2428 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2429 module_param(max_part, int, 0444);
2430 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");