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;
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 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
;
632 cmd
= nbd_read_stat(nbd
, args
->index
);
634 struct nbd_sock
*nsock
= config
->socks
[args
->index
];
636 mutex_lock(&nsock
->tx_lock
);
637 nbd_mark_nsock_dead(nbd
, nsock
, 1);
638 mutex_unlock(&nsock
->tx_lock
);
643 blk_mq_complete_request(blk_mq_rq_from_pdu(cmd
));
645 atomic_dec(&config
->recv_threads
);
646 wake_up(&config
->recv_wq
);
651 static void nbd_clear_req(struct request
*req
, void *data
, bool reserved
)
655 if (!blk_mq_request_started(req
))
657 cmd
= blk_mq_rq_to_pdu(req
);
658 cmd
->status
= BLK_STS_IOERR
;
659 blk_mq_complete_request(req
);
662 static void nbd_clear_que(struct nbd_device
*nbd
)
664 blk_mq_stop_hw_queues(nbd
->disk
->queue
);
665 blk_mq_tagset_busy_iter(&nbd
->tag_set
, nbd_clear_req
, NULL
);
666 blk_mq_start_hw_queues(nbd
->disk
->queue
);
667 dev_dbg(disk_to_dev(nbd
->disk
), "queue cleared\n");
670 static int find_fallback(struct nbd_device
*nbd
, int index
)
672 struct nbd_config
*config
= nbd
->config
;
674 struct nbd_sock
*nsock
= config
->socks
[index
];
675 int fallback
= nsock
->fallback_index
;
677 if (test_bit(NBD_DISCONNECTED
, &config
->runtime_flags
))
680 if (config
->num_connections
<= 1) {
681 dev_err_ratelimited(disk_to_dev(nbd
->disk
),
682 "Attempted send on invalid socket\n");
686 if (fallback
>= 0 && fallback
< config
->num_connections
&&
687 !config
->socks
[fallback
]->dead
)
690 if (nsock
->fallback_index
< 0 ||
691 nsock
->fallback_index
>= config
->num_connections
||
692 config
->socks
[nsock
->fallback_index
]->dead
) {
694 for (i
= 0; i
< config
->num_connections
; i
++) {
697 if (!config
->socks
[i
]->dead
) {
702 nsock
->fallback_index
= new_index
;
704 dev_err_ratelimited(disk_to_dev(nbd
->disk
),
705 "Dead connection, failed to find a fallback\n");
709 new_index
= nsock
->fallback_index
;
713 static int wait_for_reconnect(struct nbd_device
*nbd
)
715 struct nbd_config
*config
= nbd
->config
;
716 if (!config
->dead_conn_timeout
)
718 if (test_bit(NBD_DISCONNECTED
, &config
->runtime_flags
))
720 wait_event_interruptible_timeout(config
->conn_wait
,
721 atomic_read(&config
->live_connections
),
722 config
->dead_conn_timeout
);
723 return atomic_read(&config
->live_connections
);
726 static int nbd_handle_cmd(struct nbd_cmd
*cmd
, int index
)
728 struct request
*req
= blk_mq_rq_from_pdu(cmd
);
729 struct nbd_device
*nbd
= cmd
->nbd
;
730 struct nbd_config
*config
;
731 struct nbd_sock
*nsock
;
734 if (!refcount_inc_not_zero(&nbd
->config_refs
)) {
735 dev_err_ratelimited(disk_to_dev(nbd
->disk
),
736 "Socks array is empty\n");
739 config
= nbd
->config
;
741 if (index
>= config
->num_connections
) {
742 dev_err_ratelimited(disk_to_dev(nbd
->disk
),
743 "Attempted send on invalid socket\n");
747 cmd
->status
= BLK_STS_OK
;
749 nsock
= config
->socks
[index
];
750 mutex_lock(&nsock
->tx_lock
);
752 int old_index
= index
;
753 index
= find_fallback(nbd
, index
);
754 mutex_unlock(&nsock
->tx_lock
);
756 if (wait_for_reconnect(nbd
)) {
760 /* All the sockets should already be down at this point,
761 * we just want to make sure that DISCONNECTED is set so
762 * any requests that come in that were queue'ed waiting
763 * for the reconnect timer don't trigger the timer again
764 * and instead just error out.
773 /* Handle the case that we have a pending request that was partially
774 * transmitted that _has_ to be serviced first. We need to call requeue
775 * here so that it gets put _after_ the request that is already on the
778 if (unlikely(nsock
->pending
&& nsock
->pending
!= req
)) {
779 blk_mq_requeue_request(req
, true);
784 * Some failures are related to the link going down, so anything that
785 * returns EAGAIN can be retried on a different socket.
787 ret
= nbd_send_cmd(nbd
, cmd
, index
);
788 if (ret
== -EAGAIN
) {
789 dev_err_ratelimited(disk_to_dev(nbd
->disk
),
790 "Request send failed trying another connection\n");
791 nbd_mark_nsock_dead(nbd
, nsock
, 1);
792 mutex_unlock(&nsock
->tx_lock
);
796 mutex_unlock(&nsock
->tx_lock
);
801 static blk_status_t
nbd_queue_rq(struct blk_mq_hw_ctx
*hctx
,
802 const struct blk_mq_queue_data
*bd
)
804 struct nbd_cmd
*cmd
= blk_mq_rq_to_pdu(bd
->rq
);
808 * Since we look at the bio's to send the request over the network we
809 * need to make sure the completion work doesn't mark this request done
810 * before we are done doing our send. This keeps us from dereferencing
811 * freed data if we have particularly fast completions (ie we get the
812 * completion before we exit sock_xmit on the last bvec) or in the case
813 * that the server is misbehaving (or there was an error) before we're
814 * done sending everything over the wire.
816 init_completion(&cmd
->send_complete
);
817 blk_mq_start_request(bd
->rq
);
819 /* We can be called directly from the user space process, which means we
820 * could possibly have signals pending so our sendmsg will fail. In
821 * this case we need to return that we are busy, otherwise error out as
824 ret
= nbd_handle_cmd(cmd
, hctx
->queue_num
);
825 complete(&cmd
->send_complete
);
827 return ret
< 0 ? BLK_STS_IOERR
: BLK_STS_OK
;
830 static int nbd_add_socket(struct nbd_device
*nbd
, unsigned long arg
,
833 struct nbd_config
*config
= nbd
->config
;
835 struct nbd_sock
**socks
;
836 struct nbd_sock
*nsock
;
839 sock
= sockfd_lookup(arg
, &err
);
843 if (!netlink
&& !nbd
->task_setup
&&
844 !test_bit(NBD_BOUND
, &config
->runtime_flags
))
845 nbd
->task_setup
= current
;
848 (nbd
->task_setup
!= current
||
849 test_bit(NBD_BOUND
, &config
->runtime_flags
))) {
850 dev_err(disk_to_dev(nbd
->disk
),
851 "Device being setup by another task");
856 socks
= krealloc(config
->socks
, (config
->num_connections
+ 1) *
857 sizeof(struct nbd_sock
*), GFP_KERNEL
);
862 nsock
= kzalloc(sizeof(struct nbd_sock
), GFP_KERNEL
);
868 config
->socks
= socks
;
870 nsock
->fallback_index
= -1;
872 mutex_init(&nsock
->tx_lock
);
874 nsock
->pending
= NULL
;
877 socks
[config
->num_connections
++] = nsock
;
878 atomic_inc(&config
->live_connections
);
883 static int nbd_reconnect_socket(struct nbd_device
*nbd
, unsigned long arg
)
885 struct nbd_config
*config
= nbd
->config
;
886 struct socket
*sock
, *old
;
887 struct recv_thread_args
*args
;
891 sock
= sockfd_lookup(arg
, &err
);
895 args
= kzalloc(sizeof(*args
), GFP_KERNEL
);
901 for (i
= 0; i
< config
->num_connections
; i
++) {
902 struct nbd_sock
*nsock
= config
->socks
[i
];
907 mutex_lock(&nsock
->tx_lock
);
909 mutex_unlock(&nsock
->tx_lock
);
912 sk_set_memalloc(sock
->sk
);
913 sock
->sk
->sk_sndtimeo
= nbd
->tag_set
.timeout
;
914 atomic_inc(&config
->recv_threads
);
915 refcount_inc(&nbd
->config_refs
);
917 nsock
->fallback_index
= -1;
920 INIT_WORK(&args
->work
, recv_work
);
924 mutex_unlock(&nsock
->tx_lock
);
927 /* We take the tx_mutex in an error path in the recv_work, so we
928 * need to queue_work outside of the tx_mutex.
930 queue_work(recv_workqueue
, &args
->work
);
932 atomic_inc(&config
->live_connections
);
933 wake_up(&config
->conn_wait
);
941 static void nbd_bdev_reset(struct block_device
*bdev
)
943 if (bdev
->bd_openers
> 1)
945 bd_set_size(bdev
, 0);
947 blkdev_reread_part(bdev
);
948 bdev
->bd_invalidated
= 1;
952 static void nbd_parse_flags(struct nbd_device
*nbd
)
954 struct nbd_config
*config
= nbd
->config
;
955 if (config
->flags
& NBD_FLAG_READ_ONLY
)
956 set_disk_ro(nbd
->disk
, true);
958 set_disk_ro(nbd
->disk
, false);
959 if (config
->flags
& NBD_FLAG_SEND_TRIM
)
960 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, nbd
->disk
->queue
);
961 if (config
->flags
& NBD_FLAG_SEND_FLUSH
) {
962 if (config
->flags
& NBD_FLAG_SEND_FUA
)
963 blk_queue_write_cache(nbd
->disk
->queue
, true, true);
965 blk_queue_write_cache(nbd
->disk
->queue
, true, false);
968 blk_queue_write_cache(nbd
->disk
->queue
, false, false);
971 static void send_disconnects(struct nbd_device
*nbd
)
973 struct nbd_config
*config
= nbd
->config
;
974 struct nbd_request request
= {
975 .magic
= htonl(NBD_REQUEST_MAGIC
),
976 .type
= htonl(NBD_CMD_DISC
),
978 struct kvec iov
= {.iov_base
= &request
, .iov_len
= sizeof(request
)};
979 struct iov_iter from
;
982 for (i
= 0; i
< config
->num_connections
; i
++) {
983 iov_iter_kvec(&from
, WRITE
| ITER_KVEC
, &iov
, 1, sizeof(request
));
984 ret
= sock_xmit(nbd
, i
, 1, &from
, 0, NULL
);
986 dev_err(disk_to_dev(nbd
->disk
),
987 "Send disconnect failed %d\n", ret
);
991 static int nbd_disconnect(struct nbd_device
*nbd
)
993 struct nbd_config
*config
= nbd
->config
;
995 dev_info(disk_to_dev(nbd
->disk
), "NBD_DISCONNECT\n");
996 if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED
,
997 &config
->runtime_flags
))
998 send_disconnects(nbd
);
1002 static void nbd_clear_sock(struct nbd_device
*nbd
)
1006 nbd
->task_setup
= NULL
;
1009 static void nbd_config_put(struct nbd_device
*nbd
)
1011 if (refcount_dec_and_mutex_lock(&nbd
->config_refs
,
1012 &nbd
->config_lock
)) {
1013 struct nbd_config
*config
= nbd
->config
;
1014 nbd_dev_dbg_close(nbd
);
1015 nbd_size_clear(nbd
);
1016 if (test_and_clear_bit(NBD_HAS_PID_FILE
,
1017 &config
->runtime_flags
))
1018 device_remove_file(disk_to_dev(nbd
->disk
), &pid_attr
);
1019 nbd
->task_recv
= NULL
;
1020 nbd_clear_sock(nbd
);
1021 if (config
->num_connections
) {
1023 for (i
= 0; i
< config
->num_connections
; i
++) {
1024 sockfd_put(config
->socks
[i
]->sock
);
1025 kfree(config
->socks
[i
]);
1027 kfree(config
->socks
);
1032 nbd
->tag_set
.timeout
= 0;
1033 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
, nbd
->disk
->queue
);
1035 mutex_unlock(&nbd
->config_lock
);
1037 module_put(THIS_MODULE
);
1041 static int nbd_start_device(struct nbd_device
*nbd
)
1043 struct nbd_config
*config
= nbd
->config
;
1044 int num_connections
= config
->num_connections
;
1051 if (num_connections
> 1 &&
1052 !(config
->flags
& NBD_FLAG_CAN_MULTI_CONN
)) {
1053 dev_err(disk_to_dev(nbd
->disk
), "server does not support multiple connections per device.\n");
1057 blk_mq_update_nr_hw_queues(&nbd
->tag_set
, config
->num_connections
);
1058 nbd
->task_recv
= current
;
1060 nbd_parse_flags(nbd
);
1062 error
= device_create_file(disk_to_dev(nbd
->disk
), &pid_attr
);
1064 dev_err(disk_to_dev(nbd
->disk
), "device_create_file failed!\n");
1067 set_bit(NBD_HAS_PID_FILE
, &config
->runtime_flags
);
1069 nbd_dev_dbg_init(nbd
);
1070 for (i
= 0; i
< num_connections
; i
++) {
1071 struct recv_thread_args
*args
;
1073 args
= kzalloc(sizeof(*args
), GFP_KERNEL
);
1078 sk_set_memalloc(config
->socks
[i
]->sock
->sk
);
1079 config
->socks
[i
]->sock
->sk
->sk_sndtimeo
= nbd
->tag_set
.timeout
;
1080 atomic_inc(&config
->recv_threads
);
1081 refcount_inc(&nbd
->config_refs
);
1082 INIT_WORK(&args
->work
, recv_work
);
1085 queue_work(recv_workqueue
, &args
->work
);
1090 static int nbd_start_device_ioctl(struct nbd_device
*nbd
, struct block_device
*bdev
)
1092 struct nbd_config
*config
= nbd
->config
;
1095 ret
= nbd_start_device(nbd
);
1099 bd_set_size(bdev
, config
->bytesize
);
1101 bdev
->bd_invalidated
= 1;
1102 mutex_unlock(&nbd
->config_lock
);
1103 ret
= wait_event_interruptible(config
->recv_wq
,
1104 atomic_read(&config
->recv_threads
) == 0);
1107 mutex_lock(&nbd
->config_lock
);
1108 bd_set_size(bdev
, 0);
1109 /* user requested, ignore socket errors */
1110 if (test_bit(NBD_DISCONNECT_REQUESTED
, &config
->runtime_flags
))
1112 if (test_bit(NBD_TIMEDOUT
, &config
->runtime_flags
))
1117 static void nbd_clear_sock_ioctl(struct nbd_device
*nbd
,
1118 struct block_device
*bdev
)
1122 nbd_bdev_reset(bdev
);
1123 if (test_and_clear_bit(NBD_HAS_CONFIG_REF
,
1124 &nbd
->config
->runtime_flags
))
1125 nbd_config_put(nbd
);
1128 /* Must be called with config_lock held */
1129 static int __nbd_ioctl(struct block_device
*bdev
, struct nbd_device
*nbd
,
1130 unsigned int cmd
, unsigned long arg
)
1132 struct nbd_config
*config
= nbd
->config
;
1135 case NBD_DISCONNECT
:
1136 return nbd_disconnect(nbd
);
1137 case NBD_CLEAR_SOCK
:
1138 nbd_clear_sock_ioctl(nbd
, bdev
);
1141 return nbd_add_socket(nbd
, arg
, false);
1142 case NBD_SET_BLKSIZE
:
1143 nbd_size_set(nbd
, arg
,
1144 div_s64(config
->bytesize
, arg
));
1147 nbd_size_set(nbd
, config
->blksize
,
1148 div_s64(arg
, config
->blksize
));
1150 case NBD_SET_SIZE_BLOCKS
:
1151 nbd_size_set(nbd
, config
->blksize
, arg
);
1153 case NBD_SET_TIMEOUT
:
1155 nbd
->tag_set
.timeout
= arg
* HZ
;
1156 blk_queue_rq_timeout(nbd
->disk
->queue
, arg
* HZ
);
1161 config
->flags
= arg
;
1164 return nbd_start_device_ioctl(nbd
, bdev
);
1167 * This is for compatibility only. The queue is always cleared
1168 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1171 case NBD_PRINT_DEBUG
:
1173 * For compatibility only, we no longer keep a list of
1174 * outstanding requests.
1181 static int nbd_ioctl(struct block_device
*bdev
, fmode_t mode
,
1182 unsigned int cmd
, unsigned long arg
)
1184 struct nbd_device
*nbd
= bdev
->bd_disk
->private_data
;
1185 struct nbd_config
*config
= nbd
->config
;
1186 int error
= -EINVAL
;
1188 if (!capable(CAP_SYS_ADMIN
))
1191 mutex_lock(&nbd
->config_lock
);
1193 /* Don't allow ioctl operations on a nbd device that was created with
1194 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1196 if (!test_bit(NBD_BOUND
, &config
->runtime_flags
) ||
1197 (cmd
== NBD_DISCONNECT
|| cmd
== NBD_CLEAR_SOCK
))
1198 error
= __nbd_ioctl(bdev
, nbd
, cmd
, arg
);
1200 dev_err(nbd_to_dev(nbd
), "Cannot use ioctl interface on a netlink controlled device.\n");
1201 mutex_unlock(&nbd
->config_lock
);
1205 static struct nbd_config
*nbd_alloc_config(void)
1207 struct nbd_config
*config
;
1209 config
= kzalloc(sizeof(struct nbd_config
), GFP_NOFS
);
1212 atomic_set(&config
->recv_threads
, 0);
1213 init_waitqueue_head(&config
->recv_wq
);
1214 init_waitqueue_head(&config
->conn_wait
);
1215 config
->blksize
= 1024;
1216 atomic_set(&config
->live_connections
, 0);
1217 try_module_get(THIS_MODULE
);
1221 static int nbd_open(struct block_device
*bdev
, fmode_t mode
)
1223 struct nbd_device
*nbd
;
1226 mutex_lock(&nbd_index_mutex
);
1227 nbd
= bdev
->bd_disk
->private_data
;
1232 if (!refcount_inc_not_zero(&nbd
->refs
)) {
1236 if (!refcount_inc_not_zero(&nbd
->config_refs
)) {
1237 struct nbd_config
*config
;
1239 mutex_lock(&nbd
->config_lock
);
1240 if (refcount_inc_not_zero(&nbd
->config_refs
)) {
1241 mutex_unlock(&nbd
->config_lock
);
1244 config
= nbd
->config
= nbd_alloc_config();
1247 mutex_unlock(&nbd
->config_lock
);
1250 refcount_set(&nbd
->config_refs
, 1);
1251 refcount_inc(&nbd
->refs
);
1252 mutex_unlock(&nbd
->config_lock
);
1255 mutex_unlock(&nbd_index_mutex
);
1259 static void nbd_release(struct gendisk
*disk
, fmode_t mode
)
1261 struct nbd_device
*nbd
= disk
->private_data
;
1262 nbd_config_put(nbd
);
1266 static const struct block_device_operations nbd_fops
=
1268 .owner
= THIS_MODULE
,
1270 .release
= nbd_release
,
1272 .compat_ioctl
= nbd_ioctl
,
1275 #if IS_ENABLED(CONFIG_DEBUG_FS)
1277 static int nbd_dbg_tasks_show(struct seq_file
*s
, void *unused
)
1279 struct nbd_device
*nbd
= s
->private;
1282 seq_printf(s
, "recv: %d\n", task_pid_nr(nbd
->task_recv
));
1287 static int nbd_dbg_tasks_open(struct inode
*inode
, struct file
*file
)
1289 return single_open(file
, nbd_dbg_tasks_show
, inode
->i_private
);
1292 static const struct file_operations nbd_dbg_tasks_ops
= {
1293 .open
= nbd_dbg_tasks_open
,
1295 .llseek
= seq_lseek
,
1296 .release
= single_release
,
1299 static int nbd_dbg_flags_show(struct seq_file
*s
, void *unused
)
1301 struct nbd_device
*nbd
= s
->private;
1302 u32 flags
= nbd
->config
->flags
;
1304 seq_printf(s
, "Hex: 0x%08x\n\n", flags
);
1306 seq_puts(s
, "Known flags:\n");
1308 if (flags
& NBD_FLAG_HAS_FLAGS
)
1309 seq_puts(s
, "NBD_FLAG_HAS_FLAGS\n");
1310 if (flags
& NBD_FLAG_READ_ONLY
)
1311 seq_puts(s
, "NBD_FLAG_READ_ONLY\n");
1312 if (flags
& NBD_FLAG_SEND_FLUSH
)
1313 seq_puts(s
, "NBD_FLAG_SEND_FLUSH\n");
1314 if (flags
& NBD_FLAG_SEND_FUA
)
1315 seq_puts(s
, "NBD_FLAG_SEND_FUA\n");
1316 if (flags
& NBD_FLAG_SEND_TRIM
)
1317 seq_puts(s
, "NBD_FLAG_SEND_TRIM\n");
1322 static int nbd_dbg_flags_open(struct inode
*inode
, struct file
*file
)
1324 return single_open(file
, nbd_dbg_flags_show
, inode
->i_private
);
1327 static const struct file_operations nbd_dbg_flags_ops
= {
1328 .open
= nbd_dbg_flags_open
,
1330 .llseek
= seq_lseek
,
1331 .release
= single_release
,
1334 static int nbd_dev_dbg_init(struct nbd_device
*nbd
)
1337 struct nbd_config
*config
= nbd
->config
;
1342 dir
= debugfs_create_dir(nbd_name(nbd
), nbd_dbg_dir
);
1344 dev_err(nbd_to_dev(nbd
), "Failed to create debugfs dir for '%s'\n",
1348 config
->dbg_dir
= dir
;
1350 debugfs_create_file("tasks", 0444, dir
, nbd
, &nbd_dbg_tasks_ops
);
1351 debugfs_create_u64("size_bytes", 0444, dir
, &config
->bytesize
);
1352 debugfs_create_u32("timeout", 0444, dir
, &nbd
->tag_set
.timeout
);
1353 debugfs_create_u64("blocksize", 0444, dir
, &config
->blksize
);
1354 debugfs_create_file("flags", 0444, dir
, nbd
, &nbd_dbg_flags_ops
);
1359 static void nbd_dev_dbg_close(struct nbd_device
*nbd
)
1361 debugfs_remove_recursive(nbd
->config
->dbg_dir
);
1364 static int nbd_dbg_init(void)
1366 struct dentry
*dbg_dir
;
1368 dbg_dir
= debugfs_create_dir("nbd", NULL
);
1372 nbd_dbg_dir
= dbg_dir
;
1377 static void nbd_dbg_close(void)
1379 debugfs_remove_recursive(nbd_dbg_dir
);
1382 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1384 static int nbd_dev_dbg_init(struct nbd_device
*nbd
)
1389 static void nbd_dev_dbg_close(struct nbd_device
*nbd
)
1393 static int nbd_dbg_init(void)
1398 static void nbd_dbg_close(void)
1404 static int nbd_init_request(struct blk_mq_tag_set
*set
, struct request
*rq
,
1405 unsigned int hctx_idx
, unsigned int numa_node
)
1407 struct nbd_cmd
*cmd
= blk_mq_rq_to_pdu(rq
);
1408 cmd
->nbd
= set
->driver_data
;
1412 static const struct blk_mq_ops nbd_mq_ops
= {
1413 .queue_rq
= nbd_queue_rq
,
1414 .complete
= nbd_complete_rq
,
1415 .init_request
= nbd_init_request
,
1416 .timeout
= nbd_xmit_timeout
,
1419 static int nbd_dev_add(int index
)
1421 struct nbd_device
*nbd
;
1422 struct gendisk
*disk
;
1423 struct request_queue
*q
;
1426 nbd
= kzalloc(sizeof(struct nbd_device
), GFP_KERNEL
);
1430 disk
= alloc_disk(1 << part_shift
);
1435 err
= idr_alloc(&nbd_index_idr
, nbd
, index
, index
+ 1,
1440 err
= idr_alloc(&nbd_index_idr
, nbd
, 0, 0, GFP_KERNEL
);
1449 nbd
->tag_set
.ops
= &nbd_mq_ops
;
1450 nbd
->tag_set
.nr_hw_queues
= 1;
1451 nbd
->tag_set
.queue_depth
= 128;
1452 nbd
->tag_set
.numa_node
= NUMA_NO_NODE
;
1453 nbd
->tag_set
.cmd_size
= sizeof(struct nbd_cmd
);
1454 nbd
->tag_set
.flags
= BLK_MQ_F_SHOULD_MERGE
|
1455 BLK_MQ_F_SG_MERGE
| BLK_MQ_F_BLOCKING
;
1456 nbd
->tag_set
.driver_data
= nbd
;
1458 err
= blk_mq_alloc_tag_set(&nbd
->tag_set
);
1462 q
= blk_mq_init_queue(&nbd
->tag_set
);
1470 * Tell the block layer that we are not a rotational device
1472 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, disk
->queue
);
1473 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, disk
->queue
);
1474 disk
->queue
->limits
.discard_granularity
= 512;
1475 blk_queue_max_discard_sectors(disk
->queue
, UINT_MAX
);
1476 blk_queue_max_segment_size(disk
->queue
, UINT_MAX
);
1477 blk_queue_max_segments(disk
->queue
, USHRT_MAX
);
1478 blk_queue_max_hw_sectors(disk
->queue
, 65536);
1479 disk
->queue
->limits
.max_sectors
= 256;
1481 mutex_init(&nbd
->config_lock
);
1482 refcount_set(&nbd
->config_refs
, 0);
1483 refcount_set(&nbd
->refs
, 1);
1484 INIT_LIST_HEAD(&nbd
->list
);
1485 disk
->major
= NBD_MAJOR
;
1486 disk
->first_minor
= index
<< part_shift
;
1487 disk
->fops
= &nbd_fops
;
1488 disk
->private_data
= nbd
;
1489 sprintf(disk
->disk_name
, "nbd%d", index
);
1491 nbd_total_devices
++;
1495 blk_mq_free_tag_set(&nbd
->tag_set
);
1497 idr_remove(&nbd_index_idr
, index
);
1506 static int find_free_cb(int id
, void *ptr
, void *data
)
1508 struct nbd_device
*nbd
= ptr
;
1509 struct nbd_device
**found
= data
;
1511 if (!refcount_read(&nbd
->config_refs
)) {
1518 /* Netlink interface. */
1519 static struct nla_policy nbd_attr_policy
[NBD_ATTR_MAX
+ 1] = {
1520 [NBD_ATTR_INDEX
] = { .type
= NLA_U32
},
1521 [NBD_ATTR_SIZE_BYTES
] = { .type
= NLA_U64
},
1522 [NBD_ATTR_BLOCK_SIZE_BYTES
] = { .type
= NLA_U64
},
1523 [NBD_ATTR_TIMEOUT
] = { .type
= NLA_U64
},
1524 [NBD_ATTR_SERVER_FLAGS
] = { .type
= NLA_U64
},
1525 [NBD_ATTR_CLIENT_FLAGS
] = { .type
= NLA_U64
},
1526 [NBD_ATTR_SOCKETS
] = { .type
= NLA_NESTED
},
1527 [NBD_ATTR_DEAD_CONN_TIMEOUT
] = { .type
= NLA_U64
},
1528 [NBD_ATTR_DEVICE_LIST
] = { .type
= NLA_NESTED
},
1531 static struct nla_policy nbd_sock_policy
[NBD_SOCK_MAX
+ 1] = {
1532 [NBD_SOCK_FD
] = { .type
= NLA_U32
},
1535 /* We don't use this right now since we don't parse the incoming list, but we
1536 * still want it here so userspace knows what to expect.
1538 static struct nla_policy
__attribute__((unused
))
1539 nbd_device_policy
[NBD_DEVICE_ATTR_MAX
+ 1] = {
1540 [NBD_DEVICE_INDEX
] = { .type
= NLA_U32
},
1541 [NBD_DEVICE_CONNECTED
] = { .type
= NLA_U8
},
1544 static int nbd_genl_connect(struct sk_buff
*skb
, struct genl_info
*info
)
1546 struct nbd_device
*nbd
= NULL
;
1547 struct nbd_config
*config
;
1550 bool put_dev
= false;
1552 if (!netlink_capable(skb
, CAP_SYS_ADMIN
))
1555 if (info
->attrs
[NBD_ATTR_INDEX
])
1556 index
= nla_get_u32(info
->attrs
[NBD_ATTR_INDEX
]);
1557 if (!info
->attrs
[NBD_ATTR_SOCKETS
]) {
1558 printk(KERN_ERR
"nbd: must specify at least one socket\n");
1561 if (!info
->attrs
[NBD_ATTR_SIZE_BYTES
]) {
1562 printk(KERN_ERR
"nbd: must specify a size in bytes for the device\n");
1566 mutex_lock(&nbd_index_mutex
);
1568 ret
= idr_for_each(&nbd_index_idr
, &find_free_cb
, &nbd
);
1571 new_index
= nbd_dev_add(-1);
1572 if (new_index
< 0) {
1573 mutex_unlock(&nbd_index_mutex
);
1574 printk(KERN_ERR
"nbd: failed to add new device\n");
1577 nbd
= idr_find(&nbd_index_idr
, new_index
);
1580 nbd
= idr_find(&nbd_index_idr
, index
);
1583 printk(KERN_ERR
"nbd: couldn't find device at index %d\n",
1585 mutex_unlock(&nbd_index_mutex
);
1588 if (!refcount_inc_not_zero(&nbd
->refs
)) {
1589 mutex_unlock(&nbd_index_mutex
);
1592 printk(KERN_ERR
"nbd: device at index %d is going down\n",
1596 mutex_unlock(&nbd_index_mutex
);
1598 mutex_lock(&nbd
->config_lock
);
1599 if (refcount_read(&nbd
->config_refs
)) {
1600 mutex_unlock(&nbd
->config_lock
);
1604 printk(KERN_ERR
"nbd: nbd%d already in use\n", index
);
1607 if (WARN_ON(nbd
->config
)) {
1608 mutex_unlock(&nbd
->config_lock
);
1612 config
= nbd
->config
= nbd_alloc_config();
1614 mutex_unlock(&nbd
->config_lock
);
1616 printk(KERN_ERR
"nbd: couldn't allocate config\n");
1619 refcount_set(&nbd
->config_refs
, 1);
1620 set_bit(NBD_BOUND
, &config
->runtime_flags
);
1622 if (info
->attrs
[NBD_ATTR_SIZE_BYTES
]) {
1623 u64 bytes
= nla_get_u64(info
->attrs
[NBD_ATTR_SIZE_BYTES
]);
1624 nbd_size_set(nbd
, config
->blksize
,
1625 div64_u64(bytes
, config
->blksize
));
1627 if (info
->attrs
[NBD_ATTR_BLOCK_SIZE_BYTES
]) {
1629 nla_get_u64(info
->attrs
[NBD_ATTR_BLOCK_SIZE_BYTES
]);
1630 nbd_size_set(nbd
, bsize
, div64_u64(config
->bytesize
, bsize
));
1632 if (info
->attrs
[NBD_ATTR_TIMEOUT
]) {
1633 u64 timeout
= nla_get_u64(info
->attrs
[NBD_ATTR_TIMEOUT
]);
1634 nbd
->tag_set
.timeout
= timeout
* HZ
;
1635 blk_queue_rq_timeout(nbd
->disk
->queue
, timeout
* HZ
);
1637 if (info
->attrs
[NBD_ATTR_DEAD_CONN_TIMEOUT
]) {
1638 config
->dead_conn_timeout
=
1639 nla_get_u64(info
->attrs
[NBD_ATTR_DEAD_CONN_TIMEOUT
]);
1640 config
->dead_conn_timeout
*= HZ
;
1642 if (info
->attrs
[NBD_ATTR_SERVER_FLAGS
])
1644 nla_get_u64(info
->attrs
[NBD_ATTR_SERVER_FLAGS
]);
1645 if (info
->attrs
[NBD_ATTR_CLIENT_FLAGS
]) {
1646 u64 flags
= nla_get_u64(info
->attrs
[NBD_ATTR_CLIENT_FLAGS
]);
1647 if (flags
& NBD_CFLAG_DESTROY_ON_DISCONNECT
) {
1648 set_bit(NBD_DESTROY_ON_DISCONNECT
,
1649 &config
->runtime_flags
);
1654 if (info
->attrs
[NBD_ATTR_SOCKETS
]) {
1655 struct nlattr
*attr
;
1658 nla_for_each_nested(attr
, info
->attrs
[NBD_ATTR_SOCKETS
],
1660 struct nlattr
*socks
[NBD_SOCK_MAX
+1];
1662 if (nla_type(attr
) != NBD_SOCK_ITEM
) {
1663 printk(KERN_ERR
"nbd: socks must be embedded in a SOCK_ITEM attr\n");
1667 ret
= nla_parse_nested(socks
, NBD_SOCK_MAX
, attr
,
1668 nbd_sock_policy
, info
->extack
);
1670 printk(KERN_ERR
"nbd: error processing sock list\n");
1674 if (!socks
[NBD_SOCK_FD
])
1676 fd
= (int)nla_get_u32(socks
[NBD_SOCK_FD
]);
1677 ret
= nbd_add_socket(nbd
, fd
, true);
1682 ret
= nbd_start_device(nbd
);
1684 mutex_unlock(&nbd
->config_lock
);
1686 set_bit(NBD_HAS_CONFIG_REF
, &config
->runtime_flags
);
1687 refcount_inc(&nbd
->config_refs
);
1688 nbd_connect_reply(info
, nbd
->index
);
1690 nbd_config_put(nbd
);
1696 static int nbd_genl_disconnect(struct sk_buff
*skb
, struct genl_info
*info
)
1698 struct nbd_device
*nbd
;
1701 if (!netlink_capable(skb
, CAP_SYS_ADMIN
))
1704 if (!info
->attrs
[NBD_ATTR_INDEX
]) {
1705 printk(KERN_ERR
"nbd: must specify an index to disconnect\n");
1708 index
= nla_get_u32(info
->attrs
[NBD_ATTR_INDEX
]);
1709 mutex_lock(&nbd_index_mutex
);
1710 nbd
= idr_find(&nbd_index_idr
, index
);
1712 mutex_unlock(&nbd_index_mutex
);
1713 printk(KERN_ERR
"nbd: couldn't find device at index %d\n",
1717 if (!refcount_inc_not_zero(&nbd
->refs
)) {
1718 mutex_unlock(&nbd_index_mutex
);
1719 printk(KERN_ERR
"nbd: device at index %d is going down\n",
1723 mutex_unlock(&nbd_index_mutex
);
1724 if (!refcount_inc_not_zero(&nbd
->config_refs
)) {
1728 mutex_lock(&nbd
->config_lock
);
1729 nbd_disconnect(nbd
);
1730 mutex_unlock(&nbd
->config_lock
);
1731 if (test_and_clear_bit(NBD_HAS_CONFIG_REF
,
1732 &nbd
->config
->runtime_flags
))
1733 nbd_config_put(nbd
);
1734 nbd_config_put(nbd
);
1739 static int nbd_genl_reconfigure(struct sk_buff
*skb
, struct genl_info
*info
)
1741 struct nbd_device
*nbd
= NULL
;
1742 struct nbd_config
*config
;
1745 bool put_dev
= false;
1747 if (!netlink_capable(skb
, CAP_SYS_ADMIN
))
1750 if (!info
->attrs
[NBD_ATTR_INDEX
]) {
1751 printk(KERN_ERR
"nbd: must specify a device to reconfigure\n");
1754 index
= nla_get_u32(info
->attrs
[NBD_ATTR_INDEX
]);
1755 mutex_lock(&nbd_index_mutex
);
1756 nbd
= idr_find(&nbd_index_idr
, index
);
1758 mutex_unlock(&nbd_index_mutex
);
1759 printk(KERN_ERR
"nbd: couldn't find a device at index %d\n",
1763 if (!refcount_inc_not_zero(&nbd
->refs
)) {
1764 mutex_unlock(&nbd_index_mutex
);
1765 printk(KERN_ERR
"nbd: device at index %d is going down\n",
1769 mutex_unlock(&nbd_index_mutex
);
1771 if (!refcount_inc_not_zero(&nbd
->config_refs
)) {
1772 dev_err(nbd_to_dev(nbd
),
1773 "not configured, cannot reconfigure\n");
1778 mutex_lock(&nbd
->config_lock
);
1779 config
= nbd
->config
;
1780 if (!test_bit(NBD_BOUND
, &config
->runtime_flags
) ||
1782 dev_err(nbd_to_dev(nbd
),
1783 "not configured, cannot reconfigure\n");
1787 if (info
->attrs
[NBD_ATTR_TIMEOUT
]) {
1788 u64 timeout
= nla_get_u64(info
->attrs
[NBD_ATTR_TIMEOUT
]);
1789 nbd
->tag_set
.timeout
= timeout
* HZ
;
1790 blk_queue_rq_timeout(nbd
->disk
->queue
, timeout
* HZ
);
1792 if (info
->attrs
[NBD_ATTR_DEAD_CONN_TIMEOUT
]) {
1793 config
->dead_conn_timeout
=
1794 nla_get_u64(info
->attrs
[NBD_ATTR_DEAD_CONN_TIMEOUT
]);
1795 config
->dead_conn_timeout
*= HZ
;
1797 if (info
->attrs
[NBD_ATTR_CLIENT_FLAGS
]) {
1798 u64 flags
= nla_get_u64(info
->attrs
[NBD_ATTR_CLIENT_FLAGS
]);
1799 if (flags
& NBD_CFLAG_DESTROY_ON_DISCONNECT
) {
1800 if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT
,
1801 &config
->runtime_flags
))
1804 if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT
,
1805 &config
->runtime_flags
))
1806 refcount_inc(&nbd
->refs
);
1810 if (info
->attrs
[NBD_ATTR_SOCKETS
]) {
1811 struct nlattr
*attr
;
1814 nla_for_each_nested(attr
, info
->attrs
[NBD_ATTR_SOCKETS
],
1816 struct nlattr
*socks
[NBD_SOCK_MAX
+1];
1818 if (nla_type(attr
) != NBD_SOCK_ITEM
) {
1819 printk(KERN_ERR
"nbd: socks must be embedded in a SOCK_ITEM attr\n");
1823 ret
= nla_parse_nested(socks
, NBD_SOCK_MAX
, attr
,
1824 nbd_sock_policy
, info
->extack
);
1826 printk(KERN_ERR
"nbd: error processing sock list\n");
1830 if (!socks
[NBD_SOCK_FD
])
1832 fd
= (int)nla_get_u32(socks
[NBD_SOCK_FD
]);
1833 ret
= nbd_reconnect_socket(nbd
, fd
);
1839 dev_info(nbd_to_dev(nbd
), "reconnected socket\n");
1843 mutex_unlock(&nbd
->config_lock
);
1844 nbd_config_put(nbd
);
1851 static const struct genl_ops nbd_connect_genl_ops
[] = {
1853 .cmd
= NBD_CMD_CONNECT
,
1854 .policy
= nbd_attr_policy
,
1855 .doit
= nbd_genl_connect
,
1858 .cmd
= NBD_CMD_DISCONNECT
,
1859 .policy
= nbd_attr_policy
,
1860 .doit
= nbd_genl_disconnect
,
1863 .cmd
= NBD_CMD_RECONFIGURE
,
1864 .policy
= nbd_attr_policy
,
1865 .doit
= nbd_genl_reconfigure
,
1868 .cmd
= NBD_CMD_STATUS
,
1869 .policy
= nbd_attr_policy
,
1870 .doit
= nbd_genl_status
,
1874 static const struct genl_multicast_group nbd_mcast_grps
[] = {
1875 { .name
= NBD_GENL_MCAST_GROUP_NAME
, },
1878 static struct genl_family nbd_genl_family __ro_after_init
= {
1880 .name
= NBD_GENL_FAMILY_NAME
,
1881 .version
= NBD_GENL_VERSION
,
1882 .module
= THIS_MODULE
,
1883 .ops
= nbd_connect_genl_ops
,
1884 .n_ops
= ARRAY_SIZE(nbd_connect_genl_ops
),
1885 .maxattr
= NBD_ATTR_MAX
,
1886 .mcgrps
= nbd_mcast_grps
,
1887 .n_mcgrps
= ARRAY_SIZE(nbd_mcast_grps
),
1890 static int populate_nbd_status(struct nbd_device
*nbd
, struct sk_buff
*reply
)
1892 struct nlattr
*dev_opt
;
1896 /* This is a little racey, but for status it's ok. The
1897 * reason we don't take a ref here is because we can't
1898 * take a ref in the index == -1 case as we would need
1899 * to put under the nbd_index_mutex, which could
1900 * deadlock if we are configured to remove ourselves
1901 * once we're disconnected.
1903 if (refcount_read(&nbd
->config_refs
))
1905 dev_opt
= nla_nest_start(reply
, NBD_DEVICE_ITEM
);
1908 ret
= nla_put_u32(reply
, NBD_DEVICE_INDEX
, nbd
->index
);
1911 ret
= nla_put_u8(reply
, NBD_DEVICE_CONNECTED
,
1915 nla_nest_end(reply
, dev_opt
);
1919 static int status_cb(int id
, void *ptr
, void *data
)
1921 struct nbd_device
*nbd
= ptr
;
1922 return populate_nbd_status(nbd
, (struct sk_buff
*)data
);
1925 static int nbd_genl_status(struct sk_buff
*skb
, struct genl_info
*info
)
1927 struct nlattr
*dev_list
;
1928 struct sk_buff
*reply
;
1934 if (info
->attrs
[NBD_ATTR_INDEX
])
1935 index
= nla_get_u32(info
->attrs
[NBD_ATTR_INDEX
]);
1937 mutex_lock(&nbd_index_mutex
);
1939 msg_size
= nla_total_size(nla_attr_size(sizeof(u32
)) +
1940 nla_attr_size(sizeof(u8
)));
1941 msg_size
*= (index
== -1) ? nbd_total_devices
: 1;
1943 reply
= genlmsg_new(msg_size
, GFP_KERNEL
);
1946 reply_head
= genlmsg_put_reply(reply
, info
, &nbd_genl_family
, 0,
1953 dev_list
= nla_nest_start(reply
, NBD_ATTR_DEVICE_LIST
);
1955 ret
= idr_for_each(&nbd_index_idr
, &status_cb
, reply
);
1961 struct nbd_device
*nbd
;
1962 nbd
= idr_find(&nbd_index_idr
, index
);
1964 ret
= populate_nbd_status(nbd
, reply
);
1971 nla_nest_end(reply
, dev_list
);
1972 genlmsg_end(reply
, reply_head
);
1973 genlmsg_reply(reply
, info
);
1976 mutex_unlock(&nbd_index_mutex
);
1980 static void nbd_connect_reply(struct genl_info
*info
, int index
)
1982 struct sk_buff
*skb
;
1986 skb
= genlmsg_new(nla_total_size(sizeof(u32
)), GFP_KERNEL
);
1989 msg_head
= genlmsg_put_reply(skb
, info
, &nbd_genl_family
, 0,
1995 ret
= nla_put_u32(skb
, NBD_ATTR_INDEX
, index
);
2000 genlmsg_end(skb
, msg_head
);
2001 genlmsg_reply(skb
, info
);
2004 static void nbd_mcast_index(int index
)
2006 struct sk_buff
*skb
;
2010 skb
= genlmsg_new(nla_total_size(sizeof(u32
)), GFP_KERNEL
);
2013 msg_head
= genlmsg_put(skb
, 0, 0, &nbd_genl_family
, 0,
2019 ret
= nla_put_u32(skb
, NBD_ATTR_INDEX
, index
);
2024 genlmsg_end(skb
, msg_head
);
2025 genlmsg_multicast(&nbd_genl_family
, skb
, 0, 0, GFP_KERNEL
);
2028 static void nbd_dead_link_work(struct work_struct
*work
)
2030 struct link_dead_args
*args
= container_of(work
, struct link_dead_args
,
2032 nbd_mcast_index(args
->index
);
2036 static int __init
nbd_init(void)
2040 BUILD_BUG_ON(sizeof(struct nbd_request
) != 28);
2043 printk(KERN_ERR
"nbd: max_part must be >= 0\n");
2049 part_shift
= fls(max_part
);
2052 * Adjust max_part according to part_shift as it is exported
2053 * to user space so that user can know the max number of
2054 * partition kernel should be able to manage.
2056 * Note that -1 is required because partition 0 is reserved
2057 * for the whole disk.
2059 max_part
= (1UL << part_shift
) - 1;
2062 if ((1UL << part_shift
) > DISK_MAX_PARTS
)
2065 if (nbds_max
> 1UL << (MINORBITS
- part_shift
))
2067 recv_workqueue
= alloc_workqueue("knbd-recv",
2068 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
2069 if (!recv_workqueue
)
2072 if (register_blkdev(NBD_MAJOR
, "nbd")) {
2073 destroy_workqueue(recv_workqueue
);
2077 if (genl_register_family(&nbd_genl_family
)) {
2078 unregister_blkdev(NBD_MAJOR
, "nbd");
2079 destroy_workqueue(recv_workqueue
);
2084 mutex_lock(&nbd_index_mutex
);
2085 for (i
= 0; i
< nbds_max
; i
++)
2087 mutex_unlock(&nbd_index_mutex
);
2091 static int nbd_exit_cb(int id
, void *ptr
, void *data
)
2093 struct list_head
*list
= (struct list_head
*)data
;
2094 struct nbd_device
*nbd
= ptr
;
2096 list_add_tail(&nbd
->list
, list
);
2100 static void __exit
nbd_cleanup(void)
2102 struct nbd_device
*nbd
;
2103 LIST_HEAD(del_list
);
2107 mutex_lock(&nbd_index_mutex
);
2108 idr_for_each(&nbd_index_idr
, &nbd_exit_cb
, &del_list
);
2109 mutex_unlock(&nbd_index_mutex
);
2111 while (!list_empty(&del_list
)) {
2112 nbd
= list_first_entry(&del_list
, struct nbd_device
, list
);
2113 list_del_init(&nbd
->list
);
2114 if (refcount_read(&nbd
->refs
) != 1)
2115 printk(KERN_ERR
"nbd: possibly leaking a device\n");
2119 idr_destroy(&nbd_index_idr
);
2120 genl_unregister_family(&nbd_genl_family
);
2121 destroy_workqueue(recv_workqueue
);
2122 unregister_blkdev(NBD_MAJOR
, "nbd");
2125 module_init(nbd_init
);
2126 module_exit(nbd_cleanup
);
2128 MODULE_DESCRIPTION("Network Block Device");
2129 MODULE_LICENSE("GPL");
2131 module_param(nbds_max
, int, 0444);
2132 MODULE_PARM_DESC(nbds_max
, "number of network block devices to initialize (default: 16)");
2133 module_param(max_part
, int, 0444);
2134 MODULE_PARM_DESC(max_part
, "number of partitions per device (default: 0)");