1	// SPDX-License-Identifier: GPL-2.0-or-later
2	/*
3	* Network block device - make block devices work over TCP
4	*
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.
7	*
8	* Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
9	* Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
10	*
11	* (part of code stolen from loop.c)
12	*/
13
14	#define pr_fmt(fmt) "nbd: " fmt
15
16	#include <linux/major.h>
17
18	#include <linux/blkdev.h>
19	#include <linux/module.h>
20	#include <linux/init.h>
21	#include <linux/sched.h>
22	#include <linux/sched/mm.h>
23	#include <linux/fs.h>
24	#include <linux/bio.h>
25	#include <linux/stat.h>
26	#include <linux/errno.h>
27	#include <linux/file.h>
28	#include <linux/ioctl.h>
29	#include <linux/mutex.h>
30	#include <linux/compiler.h>
31	#include <linux/completion.h>
32	#include <linux/err.h>
33	#include <linux/kernel.h>
34	#include <linux/slab.h>
35	#include <net/sock.h>
36	#include <linux/net.h>
37	#include <linux/kthread.h>
38	#include <linux/types.h>
39	#include <linux/debugfs.h>
40	#include <linux/blk-mq.h>
41
42	#include <linux/uaccess.h>
43	#include <asm/types.h>
44
45	#include <linux/nbd.h>
46	#include <linux/nbd-netlink.h>
47	#include <net/genetlink.h>
48
49	#define CREATE_TRACE_POINTS
50	#include <trace/events/nbd.h>
51
52	static DEFINE_IDR(nbd_index_idr);
53	static DEFINE_MUTEX(nbd_index_mutex);
54	static struct workqueue_struct *nbd_del_wq;
55	static int nbd_total_devices = 0;
56
57	struct nbd_sock {
58	struct socket *sock;
59	struct mutex tx_lock;
60	struct request *pending;
61	int sent;
62	bool dead;
63	int fallback_index;
64	int cookie;
65	};
66
67	struct recv_thread_args {
68	struct work_struct work;
69	struct nbd_device *nbd;
70	int index;
71	};
72
73	struct link_dead_args {
74	struct work_struct work;
75	int index;
76	};
77
78	#define NBD_RT_TIMEDOUT 0
79	#define NBD_RT_DISCONNECT_REQUESTED 1
80	#define NBD_RT_DISCONNECTED 2
81	#define NBD_RT_HAS_PID_FILE 3
82	#define NBD_RT_HAS_CONFIG_REF 4
83	#define NBD_RT_BOUND 5
84	#define NBD_RT_DISCONNECT_ON_CLOSE 6
85	#define NBD_RT_HAS_BACKEND_FILE 7
86
87	#define NBD_DESTROY_ON_DISCONNECT 0
88	#define NBD_DISCONNECT_REQUESTED 1
89
90	struct nbd_config {
91	u32 flags;
92	unsigned long runtime_flags;
93	u64 dead_conn_timeout;
94
95	struct nbd_sock **socks;
96	int num_connections;
97	atomic_t live_connections;
98	wait_queue_head_t conn_wait;
99
100	atomic_t recv_threads;
101	wait_queue_head_t recv_wq;
102	unsigned int blksize_bits;
103	loff_t bytesize;
104	#if IS_ENABLED(CONFIG_DEBUG_FS)
105	struct dentry *dbg_dir;
106	#endif
107	};
108
109	static inline unsigned int nbd_blksize(struct nbd_config *config)
110	{
111	return 1u << config->blksize_bits;
112	}
113
114	struct nbd_device {
115	struct blk_mq_tag_set tag_set;
116
117	int index;
118	refcount_t config_refs;
119	refcount_t refs;
120	struct nbd_config *config;
121	struct mutex config_lock;
122	struct gendisk *disk;
123	struct workqueue_struct *recv_workq;
124	struct work_struct remove_work;
125
126	struct list_head list;
127	struct task_struct *task_setup;
128
129	unsigned long flags;
130	pid_t pid; /* pid of nbd-client, if attached */
131
132	char *backend;
133	};
134
135	#define NBD_CMD_REQUEUED 1
136	/*
137	* This flag will be set if nbd_queue_rq() succeed, and will be checked and
138	* cleared in completion. Both setting and clearing of the flag are protected
139	* by cmd->lock.
140	*/
141	#define NBD_CMD_INFLIGHT 2
142
143	struct nbd_cmd {
144	struct nbd_device *nbd;
145	struct mutex lock;
146	int index;
147	int cookie;
148	int retries;
149	blk_status_t status;
150	unsigned long flags;
151	u32 cmd_cookie;
152	};
153
154	#if IS_ENABLED(CONFIG_DEBUG_FS)
155	static struct dentry *nbd_dbg_dir;
156	#endif
157
158	#define nbd_name(nbd) ((nbd)->disk->disk_name)
159
160	#define NBD_DEF_BLKSIZE_BITS 10
161
162	static unsigned int nbds_max = 16;
163	static int max_part = 16;
164	static int part_shift;
165
166	static int nbd_dev_dbg_init(struct nbd_device *nbd);
167	static void nbd_dev_dbg_close(struct nbd_device *nbd);
168	static void nbd_config_put(struct nbd_device *nbd);
169	static void nbd_connect_reply(struct genl_info *info, int index);
170	static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
171	static void nbd_dead_link_work(struct work_struct *work);
172	static void nbd_disconnect_and_put(struct nbd_device *nbd);
173
174	static inline struct device *nbd_to_dev(struct nbd_device *nbd)
175	{
176	return disk_to_dev(nbd->disk);
177	}
178
179	static void nbd_requeue_cmd(struct nbd_cmd *cmd)
180	{
181	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
182
183	if (!test_and_set_bit(NBD_CMD_REQUEUED, addr: &cmd->flags))
184	blk_mq_requeue_request(rq: req, kick_requeue_list: true);
185	}
186
187	#define NBD_COOKIE_BITS 32
188
189	static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
190	{
191	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
192	u32 tag = blk_mq_unique_tag(rq: req);
193	u64 cookie = cmd->cmd_cookie;
194
195	return (cookie << NBD_COOKIE_BITS) | tag;
196	}
197
198	static u32 nbd_handle_to_tag(u64 handle)
199	{
200	return (u32)handle;
201	}
202
203	static u32 nbd_handle_to_cookie(u64 handle)
204	{
205	return (u32)(handle >> NBD_COOKIE_BITS);
206	}
207
208	static const char *nbdcmd_to_ascii(int cmd)
209	{
210	switch (cmd) {
211	case NBD_CMD_READ: return "read";
212	case NBD_CMD_WRITE: return "write";
213	case NBD_CMD_DISC: return "disconnect";
214	case NBD_CMD_FLUSH: return "flush";
215	case NBD_CMD_TRIM: return "trim/discard";
216	}
217	return "invalid";
218	}
219
220	static ssize_t pid_show(struct device *dev,
221	struct device_attribute *attr, char *buf)
222	{
223	struct gendisk *disk = dev_to_disk(dev);
224	struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
225
226	return sprintf(buf, fmt: "%d\n", nbd->pid);
227	}
228
229	static const struct device_attribute pid_attr = {
230	.attr = { .name = "pid", .mode = 0444},
231	.show = pid_show,
232	};
233
234	static ssize_t backend_show(struct device *dev,
235	struct device_attribute *attr, char *buf)
236	{
237	struct gendisk *disk = dev_to_disk(dev);
238	struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
239
240	return sprintf(buf, fmt: "%s\n", nbd->backend ?: "");
241	}
242
243	static const struct device_attribute backend_attr = {
244	.attr = { .name = "backend", .mode = 0444},
245	.show = backend_show,
246	};
247
248	static void nbd_dev_remove(struct nbd_device *nbd)
249	{
250	struct gendisk *disk = nbd->disk;
251
252	del_gendisk(gp: disk);
253	put_disk(disk);
254	blk_mq_free_tag_set(set: &nbd->tag_set);
255
256	/*
257	* Remove from idr after del_gendisk() completes, so if the same ID is
258	* reused, the following add_disk() will succeed.
259	*/
260	mutex_lock(&nbd_index_mutex);
261	idr_remove(&nbd_index_idr, id: nbd->index);
262	mutex_unlock(lock: &nbd_index_mutex);
263	destroy_workqueue(wq: nbd->recv_workq);
264	kfree(objp: nbd);
265	}
266
267	static void nbd_dev_remove_work(struct work_struct *work)
268	{
269	nbd_dev_remove(container_of(work, struct nbd_device, remove_work));
270	}
271
272	static void nbd_put(struct nbd_device *nbd)
273	{
274	if (!refcount_dec_and_test(r: &nbd->refs))
275	return;
276
277	/* Call del_gendisk() asynchrounously to prevent deadlock */
278	if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags))
279	queue_work(wq: nbd_del_wq, work: &nbd->remove_work);
280	else
281	nbd_dev_remove(nbd);
282	}
283
284	static int nbd_disconnected(struct nbd_config *config)
285	{
286	return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
287	test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
288	}
289
290	static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
291	int notify)
292	{
293	if (!nsock->dead && notify && !nbd_disconnected(config: nbd->config)) {
294	struct link_dead_args *args;
295	args = kmalloc(size: sizeof(struct link_dead_args), GFP_NOIO);
296	if (args) {
297	INIT_WORK(&args->work, nbd_dead_link_work);
298	args->index = nbd->index;
299	queue_work(wq: system_wq, work: &args->work);
300	}
301	}
302	if (!nsock->dead) {
303	kernel_sock_shutdown(sock: nsock->sock, how: SHUT_RDWR);
304	if (atomic_dec_return(v: &nbd->config->live_connections) == 0) {
305	if (test_and_clear_bit(NBD_RT_DISCONNECT_REQUESTED,
306	addr: &nbd->config->runtime_flags)) {
307	set_bit(NBD_RT_DISCONNECTED,
308	addr: &nbd->config->runtime_flags);
309	dev_info(nbd_to_dev(nbd),
310	"Disconnected due to user request.\n");
311	}
312	}
313	}
314	nsock->dead = true;
315	nsock->pending = NULL;
316	nsock->sent = 0;
317	}
318
319	static int nbd_set_size(struct nbd_device *nbd, loff_t bytesize,
320	loff_t blksize)
321	{
322	if (!blksize)
323	blksize = 1u << NBD_DEF_BLKSIZE_BITS;
324
325	if (blk_validate_block_size(bsize: blksize))
326	return -EINVAL;
327
328	if (bytesize < 0)
329	return -EINVAL;
330
331	nbd->config->bytesize = bytesize;
332	nbd->config->blksize_bits = __ffs(blksize);
333
334	if (!nbd->pid)
335	return 0;
336
337	if (nbd->config->flags & NBD_FLAG_SEND_TRIM) {
338	nbd->disk->queue->limits.discard_granularity = blksize;
339	blk_queue_max_discard_sectors(q: nbd->disk->queue, UINT_MAX);
340	}
341	blk_queue_logical_block_size(nbd->disk->queue, blksize);
342	blk_queue_physical_block_size(nbd->disk->queue, blksize);
343
344	if (max_part)
345	set_bit(GD_NEED_PART_SCAN, addr: &nbd->disk->state);
346	if (!set_capacity_and_notify(disk: nbd->disk, size: bytesize >> 9))
347	kobject_uevent(kobj: &nbd_to_dev(nbd)->kobj, action: KOBJ_CHANGE);
348	return 0;
349	}
350
351	static void nbd_complete_rq(struct request *req)
352	{
353	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: req);
354
355	dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
356	cmd->status ? "failed" : "done");
357
358	blk_mq_end_request(rq: req, error: cmd->status);
359	}
360
361	/*
362	* Forcibly shutdown the socket causing all listeners to error
363	*/
364	static void sock_shutdown(struct nbd_device *nbd)
365	{
366	struct nbd_config *config = nbd->config;
367	int i;
368
369	if (config->num_connections == 0)
370	return;
371	if (test_and_set_bit(NBD_RT_DISCONNECTED, addr: &config->runtime_flags))
372	return;
373
374	for (i = 0; i < config->num_connections; i++) {
375	struct nbd_sock *nsock = config->socks[i];
376	mutex_lock(&nsock->tx_lock);
377	nbd_mark_nsock_dead(nbd, nsock, notify: 0);
378	mutex_unlock(lock: &nsock->tx_lock);
379	}
380	dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
381	}
382
383	static u32 req_to_nbd_cmd_type(struct request *req)
384	{
385	switch (req_op(req)) {
386	case REQ_OP_DISCARD:
387	return NBD_CMD_TRIM;
388	case REQ_OP_FLUSH:
389	return NBD_CMD_FLUSH;
390	case REQ_OP_WRITE:
391	return NBD_CMD_WRITE;
392	case REQ_OP_READ:
393	return NBD_CMD_READ;
394	default:
395	return U32_MAX;
396	}
397	}
398
399	static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req)
400	{
401	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: req);
402	struct nbd_device *nbd = cmd->nbd;
403	struct nbd_config *config;
404
405	if (!mutex_trylock(lock: &cmd->lock))
406	return BLK_EH_RESET_TIMER;
407
408	if (!test_bit(NBD_CMD_INFLIGHT, &cmd->flags)) {
409	mutex_unlock(lock: &cmd->lock);
410	return BLK_EH_DONE;
411	}
412
413	if (!refcount_inc_not_zero(r: &nbd->config_refs)) {
414	cmd->status = BLK_STS_TIMEOUT;
415	__clear_bit(NBD_CMD_INFLIGHT, &cmd->flags);
416	mutex_unlock(lock: &cmd->lock);
417	goto done;
418	}
419	config = nbd->config;
420
421	if (config->num_connections > 1 ||
422	(config->num_connections == 1 && nbd->tag_set.timeout)) {
423	dev_err_ratelimited(nbd_to_dev(nbd),
424	"Connection timed out, retrying (%d/%d alive)\n",
425	atomic_read(&config->live_connections),
426	config->num_connections);
427	/*
428	* Hooray we have more connections, requeue this IO, the submit
429	* path will put it on a real connection. Or if only one
430	* connection is configured, the submit path will wait util
431	* a new connection is reconfigured or util dead timeout.
432	*/
433	if (config->socks) {
434	if (cmd->index < config->num_connections) {
435	struct nbd_sock *nsock =
436	config->socks[cmd->index];
437	mutex_lock(&nsock->tx_lock);
438	/* We can have multiple outstanding requests, so
439	* we don't want to mark the nsock dead if we've
440	* already reconnected with a new socket, so
441	* only mark it dead if its the same socket we
442	* were sent out on.
443	*/
444	if (cmd->cookie == nsock->cookie)
445	nbd_mark_nsock_dead(nbd, nsock, notify: 1);
446	mutex_unlock(lock: &nsock->tx_lock);
447	}
448	mutex_unlock(lock: &cmd->lock);
449	nbd_requeue_cmd(cmd);
450	nbd_config_put(nbd);
451	return BLK_EH_DONE;
452	}
453	}
454
455	if (!nbd->tag_set.timeout) {
456	/*
457	* Userspace sets timeout=0 to disable socket disconnection,
458	* so just warn and reset the timer.
459	*/
460	struct nbd_sock *nsock = config->socks[cmd->index];
461	cmd->retries++;
462	dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
463	req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
464	(unsigned long long)blk_rq_pos(req) << 9,
465	blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
466
467	mutex_lock(&nsock->tx_lock);
468	if (cmd->cookie != nsock->cookie) {
469	nbd_requeue_cmd(cmd);
470	mutex_unlock(lock: &nsock->tx_lock);
471	mutex_unlock(lock: &cmd->lock);
472	nbd_config_put(nbd);
473	return BLK_EH_DONE;
474	}
475	mutex_unlock(lock: &nsock->tx_lock);
476	mutex_unlock(lock: &cmd->lock);
477	nbd_config_put(nbd);
478	return BLK_EH_RESET_TIMER;
479	}
480
481	dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
482	set_bit(NBD_RT_TIMEDOUT, addr: &config->runtime_flags);
483	cmd->status = BLK_STS_IOERR;
484	__clear_bit(NBD_CMD_INFLIGHT, &cmd->flags);
485	mutex_unlock(lock: &cmd->lock);
486	sock_shutdown(nbd);
487	nbd_config_put(nbd);
488	done:
489	blk_mq_complete_request(rq: req);
490	return BLK_EH_DONE;
491	}
492
493	/*
494	* Send or receive packet. Return a positive value on success and
495	* negtive value on failue, and never return 0.
496	*/
497	static int sock_xmit(struct nbd_device *nbd, int index, int send,
498	struct iov_iter *iter, int msg_flags, int *sent)
499	{
500	struct nbd_config *config = nbd->config;
501	struct socket *sock = config->socks[index]->sock;
502	int result;
503	struct msghdr msg;
504	unsigned int noreclaim_flag;
505
506	if (unlikely(!sock)) {
507	dev_err_ratelimited(disk_to_dev(nbd->disk),
508	"Attempted %s on closed socket in sock_xmit\n",
509	(send ? "send" : "recv"));
510	return -EINVAL;
511	}
512
513	msg.msg_iter = *iter;
514
515	noreclaim_flag = memalloc_noreclaim_save();
516	do {
517	sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
518	sock->sk->sk_use_task_frag = false;
519	msg.msg_name = NULL;
520	msg.msg_namelen = 0;
521	msg.msg_control = NULL;
522	msg.msg_controllen = 0;
523	msg.msg_flags = msg_flags | MSG_NOSIGNAL;
524
525	if (send)
526	result = sock_sendmsg(sock, msg: &msg);
527	else
528	result = sock_recvmsg(sock, msg: &msg, flags: msg.msg_flags);
529
530	if (result <= 0) {
531	if (result == 0)
532	result = -EPIPE; /* short read */
533	break;
534	}
535	if (sent)
536	*sent += result;
537	} while (msg_data_left(msg: &msg));
538
539	memalloc_noreclaim_restore(flags: noreclaim_flag);
540
541	return result;
542	}
543
544	/*
545	* Different settings for sk->sk_sndtimeo can result in different return values
546	* if there is a signal pending when we enter sendmsg, because reasons?
547	*/
548	static inline int was_interrupted(int result)
549	{
550	return result == -ERESTARTSYS || result == -EINTR;
551	}
552
553	/* always call with the tx_lock held */
554	static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
555	{
556	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
557	struct nbd_config *config = nbd->config;
558	struct nbd_sock *nsock = config->socks[index];
559	int result;
560	struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
561	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
562	struct iov_iter from;
563	unsigned long size = blk_rq_bytes(rq: req);
564	struct bio *bio;
565	u64 handle;
566	u32 type;
567	u32 nbd_cmd_flags = 0;
568	int sent = nsock->sent, skip = 0;
569
570	iov_iter_kvec(i: &from, ITER_SOURCE, kvec: &iov, nr_segs: 1, count: sizeof(request));
571
572	type = req_to_nbd_cmd_type(req);
573	if (type == U32_MAX)
574	return -EIO;
575
576	if (rq_data_dir(req) == WRITE &&
577	(config->flags & NBD_FLAG_READ_ONLY)) {
578	dev_err_ratelimited(disk_to_dev(nbd->disk),
579	"Write on read-only\n");
580	return -EIO;
581	}
582
583	if (req->cmd_flags & REQ_FUA)
584	nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
585
586	/* We did a partial send previously, and we at least sent the whole
587	* request struct, so just go and send the rest of the pages in the
588	* request.
589	*/
590	if (sent) {
591	if (sent >= sizeof(request)) {
592	skip = sent - sizeof(request);
593
594	/* initialize handle for tracing purposes */
595	handle = nbd_cmd_handle(cmd);
596
597	goto send_pages;
598	}
599	iov_iter_advance(i: &from, bytes: sent);
600	} else {
601	cmd->cmd_cookie++;
602	}
603	cmd->index = index;
604	cmd->cookie = nsock->cookie;
605	cmd->retries = 0;
606	request.type = htonl(type | nbd_cmd_flags);
607	if (type != NBD_CMD_FLUSH) {
608	request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
609	request.len = htonl(size);
610	}
611	handle = nbd_cmd_handle(cmd);
612	request.cookie = cpu_to_be64(handle);
613
614	trace_nbd_send_request(nbd_request: &request, index: nbd->index, rq: blk_mq_rq_from_pdu(pdu: cmd));
615
616	dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
617	req, nbdcmd_to_ascii(type),
618	(unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
619	result = sock_xmit(nbd, index, send: 1, iter: &from,
620	msg_flags: (type == NBD_CMD_WRITE) ? MSG_MORE : 0, sent: &sent);
621	trace_nbd_header_sent(req, handle);
622	if (result < 0) {
623	if (was_interrupted(result)) {
624	/* If we haven't sent anything we can just return BUSY,
625	* however if we have sent something we need to make
626	* sure we only allow this req to be sent until we are
627	* completely done.
628	*/
629	if (sent) {
630	nsock->pending = req;
631	nsock->sent = sent;
632	}
633	set_bit(NBD_CMD_REQUEUED, addr: &cmd->flags);
634	return BLK_STS_RESOURCE;
635	}
636	dev_err_ratelimited(disk_to_dev(nbd->disk),
637	"Send control failed (result %d)\n", result);
638	return -EAGAIN;
639	}
640	send_pages:
641	if (type != NBD_CMD_WRITE)
642	goto out;
643
644	bio = req->bio;
645	while (bio) {
646	struct bio *next = bio->bi_next;
647	struct bvec_iter iter;
648	struct bio_vec bvec;
649
650	bio_for_each_segment(bvec, bio, iter) {
651	bool is_last = !next && bio_iter_last(bvec, iter);
652	int flags = is_last ? 0 : MSG_MORE;
653
654	dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
655	req, bvec.bv_len);
656	iov_iter_bvec(i: &from, ITER_SOURCE, bvec: &bvec, nr_segs: 1, count: bvec.bv_len);
657	if (skip) {
658	if (skip >= iov_iter_count(i: &from)) {
659	skip -= iov_iter_count(i: &from);
660	continue;
661	}
662	iov_iter_advance(i: &from, bytes: skip);
663	skip = 0;
664	}
665	result = sock_xmit(nbd, index, send: 1, iter: &from, msg_flags: flags, sent: &sent);
666	if (result < 0) {
667	if (was_interrupted(result)) {
668	/* We've already sent the header, we
669	* have no choice but to set pending and
670	* return BUSY.
671	*/
672	nsock->pending = req;
673	nsock->sent = sent;
674	set_bit(NBD_CMD_REQUEUED, addr: &cmd->flags);
675	return BLK_STS_RESOURCE;
676	}
677	dev_err(disk_to_dev(nbd->disk),
678	"Send data failed (result %d)\n",
679	result);
680	return -EAGAIN;
681	}
682	/*
683	* The completion might already have come in,
684	* so break for the last one instead of letting
685	* the iterator do it. This prevents use-after-free
686	* of the bio.
687	*/
688	if (is_last)
689	break;
690	}
691	bio = next;
692	}
693	out:
694	trace_nbd_payload_sent(req, handle);
695	nsock->pending = NULL;
696	nsock->sent = 0;
697	return 0;
698	}
699
700	static int nbd_read_reply(struct nbd_device *nbd, int index,
701	struct nbd_reply *reply)
702	{
703	struct kvec iov = {.iov_base = reply, .iov_len = sizeof(*reply)};
704	struct iov_iter to;
705	int result;
706
707	reply->magic = 0;
708	iov_iter_kvec(i: &to, ITER_DEST, kvec: &iov, nr_segs: 1, count: sizeof(*reply));
709	result = sock_xmit(nbd, index, send: 0, iter: &to, MSG_WAITALL, NULL);
710	if (result < 0) {
711	if (!nbd_disconnected(config: nbd->config))
712	dev_err(disk_to_dev(nbd->disk),
713	"Receive control failed (result %d)\n", result);
714	return result;
715	}
716
717	if (ntohl(reply->magic) != NBD_REPLY_MAGIC) {
718	dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
719	(unsigned long)ntohl(reply->magic));
720	return -EPROTO;
721	}
722
723	return 0;
724	}
725
726	/* NULL returned = something went wrong, inform userspace */
727	static struct nbd_cmd *nbd_handle_reply(struct nbd_device *nbd, int index,
728	struct nbd_reply *reply)
729	{
730	int result;
731	struct nbd_cmd *cmd;
732	struct request *req = NULL;
733	u64 handle;
734	u16 hwq;
735	u32 tag;
736	int ret = 0;
737
738	handle = be64_to_cpu(reply->cookie);
739	tag = nbd_handle_to_tag(handle);
740	hwq = blk_mq_unique_tag_to_hwq(unique_tag: tag);
741	if (hwq < nbd->tag_set.nr_hw_queues)
742	req = blk_mq_tag_to_rq(tags: nbd->tag_set.tags[hwq],
743	tag: blk_mq_unique_tag_to_tag(unique_tag: tag));
744	if (!req || !blk_mq_request_started(rq: req)) {
745	dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
746	tag, req);
747	return ERR_PTR(error: -ENOENT);
748	}
749	trace_nbd_header_received(req, handle);
750	cmd = blk_mq_rq_to_pdu(rq: req);
751
752	mutex_lock(&cmd->lock);
753	if (!test_bit(NBD_CMD_INFLIGHT, &cmd->flags)) {
754	dev_err(disk_to_dev(nbd->disk), "Suspicious reply %d (status %u flags %lu)",
755	tag, cmd->status, cmd->flags);
756	ret = -ENOENT;
757	goto out;
758	}
759	if (cmd->index != index) {
760	dev_err(disk_to_dev(nbd->disk), "Unexpected reply %d from different sock %d (expected %d)",
761	tag, index, cmd->index);
762	ret = -ENOENT;
763	goto out;
764	}
765	if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
766	dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
767	req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
768	ret = -ENOENT;
769	goto out;
770	}
771	if (cmd->status != BLK_STS_OK) {
772	dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
773	req);
774	ret = -ENOENT;
775	goto out;
776	}
777	if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
778	dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
779	req);
780	ret = -ENOENT;
781	goto out;
782	}
783	if (ntohl(reply->error)) {
784	dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
785	ntohl(reply->error));
786	cmd->status = BLK_STS_IOERR;
787	goto out;
788	}
789
790	dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
791	if (rq_data_dir(req) != WRITE) {
792	struct req_iterator iter;
793	struct bio_vec bvec;
794	struct iov_iter to;
795
796	rq_for_each_segment(bvec, req, iter) {
797	iov_iter_bvec(i: &to, ITER_DEST, bvec: &bvec, nr_segs: 1, count: bvec.bv_len);
798	result = sock_xmit(nbd, index, send: 0, iter: &to, MSG_WAITALL, NULL);
799	if (result < 0) {
800	dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
801	result);
802	/*
803	* If we've disconnected, we need to make sure we
804	* complete this request, otherwise error out
805	* and let the timeout stuff handle resubmitting
806	* this request onto another connection.
807	*/
808	if (nbd_disconnected(config: nbd->config)) {
809	cmd->status = BLK_STS_IOERR;
810	goto out;
811	}
812	ret = -EIO;
813	goto out;
814	}
815	dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
816	req, bvec.bv_len);
817	}
818	}
819	out:
820	trace_nbd_payload_received(req, handle);
821	mutex_unlock(lock: &cmd->lock);
822	return ret ? ERR_PTR(error: ret) : cmd;
823	}
824
825	static void recv_work(struct work_struct *work)
826	{
827	struct recv_thread_args *args = container_of(work,
828	struct recv_thread_args,
829	work);
830	struct nbd_device *nbd = args->nbd;
831	struct nbd_config *config = nbd->config;
832	struct request_queue *q = nbd->disk->queue;
833	struct nbd_sock *nsock;
834	struct nbd_cmd *cmd;
835	struct request *rq;
836
837	while (1) {
838	struct nbd_reply reply;
839
840	if (nbd_read_reply(nbd, index: args->index, reply: &reply))
841	break;
842
843	/*
844	* Grab .q_usage_counter so request pool won't go away, then no
845	* request use-after-free is possible during nbd_handle_reply().
846	* If queue is frozen, there won't be any inflight requests, we
847	* needn't to handle the incoming garbage message.
848	*/
849	if (!percpu_ref_tryget(ref: &q->q_usage_counter)) {
850	dev_err(disk_to_dev(nbd->disk), "%s: no io inflight\n",
851	__func__);
852	break;
853	}
854
855	cmd = nbd_handle_reply(nbd, index: args->index, reply: &reply);
856	if (IS_ERR(ptr: cmd)) {
857	percpu_ref_put(ref: &q->q_usage_counter);
858	break;
859	}
860
861	rq = blk_mq_rq_from_pdu(pdu: cmd);
862	if (likely(!blk_should_fake_timeout(rq->q))) {
863	bool complete;
864
865	mutex_lock(&cmd->lock);
866	complete = __test_and_clear_bit(NBD_CMD_INFLIGHT,
867	&cmd->flags);
868	mutex_unlock(lock: &cmd->lock);
869	if (complete)
870	blk_mq_complete_request(rq);
871	}
872	percpu_ref_put(ref: &q->q_usage_counter);
873	}
874
875	nsock = config->socks[args->index];
876	mutex_lock(&nsock->tx_lock);
877	nbd_mark_nsock_dead(nbd, nsock, notify: 1);
878	mutex_unlock(lock: &nsock->tx_lock);
879
880	nbd_config_put(nbd);
881	atomic_dec(v: &config->recv_threads);
882	wake_up(&config->recv_wq);
883	kfree(objp: args);
884	}
885
886	static bool nbd_clear_req(struct request *req, void *data)
887	{
888	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: req);
889
890	/* don't abort one completed request */
891	if (blk_mq_request_completed(rq: req))
892	return true;
893
894	mutex_lock(&cmd->lock);
895	if (!__test_and_clear_bit(NBD_CMD_INFLIGHT, &cmd->flags)) {
896	mutex_unlock(lock: &cmd->lock);
897	return true;
898	}
899	cmd->status = BLK_STS_IOERR;
900	mutex_unlock(lock: &cmd->lock);
901
902	blk_mq_complete_request(rq: req);
903	return true;
904	}
905
906	static void nbd_clear_que(struct nbd_device *nbd)
907	{
908	blk_mq_quiesce_queue(q: nbd->disk->queue);
909	blk_mq_tagset_busy_iter(tagset: &nbd->tag_set, fn: nbd_clear_req, NULL);
910	blk_mq_unquiesce_queue(q: nbd->disk->queue);
911	dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
912	}
913
914	static int find_fallback(struct nbd_device *nbd, int index)
915	{
916	struct nbd_config *config = nbd->config;
917	int new_index = -1;
918	struct nbd_sock *nsock = config->socks[index];
919	int fallback = nsock->fallback_index;
920
921	if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
922	return new_index;
923
924	if (config->num_connections <= 1) {
925	dev_err_ratelimited(disk_to_dev(nbd->disk),
926	"Dead connection, failed to find a fallback\n");
927	return new_index;
928	}
929
930	if (fallback >= 0 && fallback < config->num_connections &&
931	!config->socks[fallback]->dead)
932	return fallback;
933
934	if (nsock->fallback_index < 0 ||
935	nsock->fallback_index >= config->num_connections ||
936	config->socks[nsock->fallback_index]->dead) {
937	int i;
938	for (i = 0; i < config->num_connections; i++) {
939	if (i == index)
940	continue;
941	if (!config->socks[i]->dead) {
942	new_index = i;
943	break;
944	}
945	}
946	nsock->fallback_index = new_index;
947	if (new_index < 0) {
948	dev_err_ratelimited(disk_to_dev(nbd->disk),
949	"Dead connection, failed to find a fallback\n");
950	return new_index;
951	}
952	}
953	new_index = nsock->fallback_index;
954	return new_index;
955	}
956
957	static int wait_for_reconnect(struct nbd_device *nbd)
958	{
959	struct nbd_config *config = nbd->config;
960	if (!config->dead_conn_timeout)
961	return 0;
962
963	if (!wait_event_timeout(config->conn_wait,
964	test_bit(NBD_RT_DISCONNECTED,
965	&config->runtime_flags) ||
966	atomic_read(&config->live_connections) > 0,
967	config->dead_conn_timeout))
968	return 0;
969
970	return !test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
971	}
972
973	static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
974	{
975	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
976	struct nbd_device *nbd = cmd->nbd;
977	struct nbd_config *config;
978	struct nbd_sock *nsock;
979	int ret;
980
981	if (!refcount_inc_not_zero(r: &nbd->config_refs)) {
982	dev_err_ratelimited(disk_to_dev(nbd->disk),
983	"Socks array is empty\n");
984	return -EINVAL;
985	}
986	config = nbd->config;
987
988	if (index >= config->num_connections) {
989	dev_err_ratelimited(disk_to_dev(nbd->disk),
990	"Attempted send on invalid socket\n");
991	nbd_config_put(nbd);
992	return -EINVAL;
993	}
994	cmd->status = BLK_STS_OK;
995	again:
996	nsock = config->socks[index];
997	mutex_lock(&nsock->tx_lock);
998	if (nsock->dead) {
999	int old_index = index;
1000	index = find_fallback(nbd, index);
1001	mutex_unlock(lock: &nsock->tx_lock);
1002	if (index < 0) {
1003	if (wait_for_reconnect(nbd)) {
1004	index = old_index;
1005	goto again;
1006	}
1007	/* All the sockets should already be down at this point,
1008	* we just want to make sure that DISCONNECTED is set so
1009	* any requests that come in that were queue'ed waiting
1010	* for the reconnect timer don't trigger the timer again
1011	* and instead just error out.
1012	*/
1013	sock_shutdown(nbd);
1014	nbd_config_put(nbd);
1015	return -EIO;
1016	}
1017	goto again;
1018	}
1019
1020	/* Handle the case that we have a pending request that was partially
1021	* transmitted that _has_ to be serviced first. We need to call requeue
1022	* here so that it gets put _after_ the request that is already on the
1023	* dispatch list.
1024	*/
1025	blk_mq_start_request(rq: req);
1026	if (unlikely(nsock->pending && nsock->pending != req)) {
1027	nbd_requeue_cmd(cmd);
1028	ret = 0;
1029	goto out;
1030	}
1031	/*
1032	* Some failures are related to the link going down, so anything that
1033	* returns EAGAIN can be retried on a different socket.
1034	*/
1035	ret = nbd_send_cmd(nbd, cmd, index);
1036	/*
1037	* Access to this flag is protected by cmd->lock, thus it's safe to set
1038	* the flag after nbd_send_cmd() succeed to send request to server.
1039	*/
1040	if (!ret)
1041	__set_bit(NBD_CMD_INFLIGHT, &cmd->flags);
1042	else if (ret == -EAGAIN) {
1043	dev_err_ratelimited(disk_to_dev(nbd->disk),
1044	"Request send failed, requeueing\n");
1045	nbd_mark_nsock_dead(nbd, nsock, notify: 1);
1046	nbd_requeue_cmd(cmd);
1047	ret = 0;
1048	}
1049	out:
1050	mutex_unlock(lock: &nsock->tx_lock);
1051	nbd_config_put(nbd);
1052	return ret;
1053	}
1054
1055	static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
1056	const struct blk_mq_queue_data *bd)
1057	{
1058	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: bd->rq);
1059	int ret;
1060
1061	/*
1062	* Since we look at the bio's to send the request over the network we
1063	* need to make sure the completion work doesn't mark this request done
1064	* before we are done doing our send. This keeps us from dereferencing
1065	* freed data if we have particularly fast completions (ie we get the
1066	* completion before we exit sock_xmit on the last bvec) or in the case
1067	* that the server is misbehaving (or there was an error) before we're
1068	* done sending everything over the wire.
1069	*/
1070	mutex_lock(&cmd->lock);
1071	clear_bit(NBD_CMD_REQUEUED, addr: &cmd->flags);
1072
1073	/* We can be called directly from the user space process, which means we
1074	* could possibly have signals pending so our sendmsg will fail. In
1075	* this case we need to return that we are busy, otherwise error out as
1076	* appropriate.
1077	*/
1078	ret = nbd_handle_cmd(cmd, index: hctx->queue_num);
1079	if (ret < 0)
1080	ret = BLK_STS_IOERR;
1081	else if (!ret)
1082	ret = BLK_STS_OK;
1083	mutex_unlock(lock: &cmd->lock);
1084
1085	return ret;
1086	}
1087
1088	static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
1089	int *err)
1090	{
1091	struct socket *sock;
1092
1093	*err = 0;
1094	sock = sockfd_lookup(fd, err);
1095	if (!sock)
1096	return NULL;
1097
1098	if (sock->ops->shutdown == sock_no_shutdown) {
1099	dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
1100	*err = -EINVAL;
1101	sockfd_put(sock);
1102	return NULL;
1103	}
1104
1105	return sock;
1106	}
1107
1108	static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1109	bool netlink)
1110	{
1111	struct nbd_config *config = nbd->config;
1112	struct socket *sock;
1113	struct nbd_sock **socks;
1114	struct nbd_sock *nsock;
1115	int err;
1116
1117	/* Arg will be cast to int, check it to avoid overflow */
1118	if (arg > INT_MAX)
1119	return -EINVAL;
1120	sock = nbd_get_socket(nbd, fd: arg, err: &err);
1121	if (!sock)
1122	return err;
1123
1124	/*
1125	* We need to make sure we don't get any errant requests while we're
1126	* reallocating the ->socks array.
1127	*/
1128	blk_mq_freeze_queue(q: nbd->disk->queue);
1129
1130	if (!netlink && !nbd->task_setup &&
1131	!test_bit(NBD_RT_BOUND, &config->runtime_flags))
1132	nbd->task_setup = current;
1133
1134	if (!netlink &&
1135	(nbd->task_setup != current ||
1136	test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1137	dev_err(disk_to_dev(nbd->disk),
1138	"Device being setup by another task");
1139	err = -EBUSY;
1140	goto put_socket;
1141	}
1142
1143	nsock = kzalloc(size: sizeof(*nsock), GFP_KERNEL);
1144	if (!nsock) {
1145	err = -ENOMEM;
1146	goto put_socket;
1147	}
1148
1149	socks = krealloc(objp: config->socks, new_size: (config->num_connections + 1) *
1150	sizeof(struct nbd_sock *), GFP_KERNEL);
1151	if (!socks) {
1152	kfree(objp: nsock);
1153	err = -ENOMEM;
1154	goto put_socket;
1155	}
1156
1157	config->socks = socks;
1158
1159	nsock->fallback_index = -1;
1160	nsock->dead = false;
1161	mutex_init(&nsock->tx_lock);
1162	nsock->sock = sock;
1163	nsock->pending = NULL;
1164	nsock->sent = 0;
1165	nsock->cookie = 0;
1166	socks[config->num_connections++] = nsock;
1167	atomic_inc(v: &config->live_connections);
1168	blk_mq_unfreeze_queue(q: nbd->disk->queue);
1169
1170	return 0;
1171
1172	put_socket:
1173	blk_mq_unfreeze_queue(q: nbd->disk->queue);
1174	sockfd_put(sock);
1175	return err;
1176	}
1177
1178	static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1179	{
1180	struct nbd_config *config = nbd->config;
1181	struct socket *sock, *old;
1182	struct recv_thread_args *args;
1183	int i;
1184	int err;
1185
1186	sock = nbd_get_socket(nbd, fd: arg, err: &err);
1187	if (!sock)
1188	return err;
1189
1190	args = kzalloc(size: sizeof(*args), GFP_KERNEL);
1191	if (!args) {
1192	sockfd_put(sock);
1193	return -ENOMEM;
1194	}
1195
1196	for (i = 0; i < config->num_connections; i++) {
1197	struct nbd_sock *nsock = config->socks[i];
1198
1199	if (!nsock->dead)
1200	continue;
1201
1202	mutex_lock(&nsock->tx_lock);
1203	if (!nsock->dead) {
1204	mutex_unlock(lock: &nsock->tx_lock);
1205	continue;
1206	}
1207	sk_set_memalloc(sk: sock->sk);
1208	if (nbd->tag_set.timeout)
1209	sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1210	atomic_inc(v: &config->recv_threads);
1211	refcount_inc(r: &nbd->config_refs);
1212	old = nsock->sock;
1213	nsock->fallback_index = -1;
1214	nsock->sock = sock;
1215	nsock->dead = false;
1216	INIT_WORK(&args->work, recv_work);
1217	args->index = i;
1218	args->nbd = nbd;
1219	nsock->cookie++;
1220	mutex_unlock(lock: &nsock->tx_lock);
1221	sockfd_put(old);
1222
1223	clear_bit(NBD_RT_DISCONNECTED, addr: &config->runtime_flags);
1224
1225	/* We take the tx_mutex in an error path in the recv_work, so we
1226	* need to queue_work outside of the tx_mutex.
1227	*/
1228	queue_work(wq: nbd->recv_workq, work: &args->work);
1229
1230	atomic_inc(v: &config->live_connections);
1231	wake_up(&config->conn_wait);
1232	return 0;
1233	}
1234	sockfd_put(sock);
1235	kfree(objp: args);
1236	return -ENOSPC;
1237	}
1238
1239	static void nbd_bdev_reset(struct nbd_device *nbd)
1240	{
1241	if (disk_openers(disk: nbd->disk) > 1)
1242	return;
1243	set_capacity(disk: nbd->disk, size: 0);
1244	}
1245
1246	static void nbd_parse_flags(struct nbd_device *nbd)
1247	{
1248	struct nbd_config *config = nbd->config;
1249	if (config->flags & NBD_FLAG_READ_ONLY)
1250	set_disk_ro(disk: nbd->disk, read_only: true);
1251	else
1252	set_disk_ro(disk: nbd->disk, read_only: false);
1253	if (config->flags & NBD_FLAG_SEND_FLUSH) {
1254	if (config->flags & NBD_FLAG_SEND_FUA)
1255	blk_queue_write_cache(q: nbd->disk->queue, enabled: true, fua: true);
1256	else
1257	blk_queue_write_cache(q: nbd->disk->queue, enabled: true, fua: false);
1258	}
1259	else
1260	blk_queue_write_cache(q: nbd->disk->queue, enabled: false, fua: false);
1261	}
1262
1263	static void send_disconnects(struct nbd_device *nbd)
1264	{
1265	struct nbd_config *config = nbd->config;
1266	struct nbd_request request = {
1267	.magic = htonl(NBD_REQUEST_MAGIC),
1268	.type = htonl(NBD_CMD_DISC),
1269	};
1270	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1271	struct iov_iter from;
1272	int i, ret;
1273
1274	for (i = 0; i < config->num_connections; i++) {
1275	struct nbd_sock *nsock = config->socks[i];
1276
1277	iov_iter_kvec(i: &from, ITER_SOURCE, kvec: &iov, nr_segs: 1, count: sizeof(request));
1278	mutex_lock(&nsock->tx_lock);
1279	ret = sock_xmit(nbd, index: i, send: 1, iter: &from, msg_flags: 0, NULL);
1280	if (ret < 0)
1281	dev_err(disk_to_dev(nbd->disk),
1282	"Send disconnect failed %d\n", ret);
1283	mutex_unlock(lock: &nsock->tx_lock);
1284	}
1285	}
1286
1287	static int nbd_disconnect(struct nbd_device *nbd)
1288	{
1289	struct nbd_config *config = nbd->config;
1290
1291	dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1292	set_bit(NBD_RT_DISCONNECT_REQUESTED, addr: &config->runtime_flags);
1293	set_bit(NBD_DISCONNECT_REQUESTED, addr: &nbd->flags);
1294	send_disconnects(nbd);
1295	return 0;
1296	}
1297
1298	static void nbd_clear_sock(struct nbd_device *nbd)
1299	{
1300	sock_shutdown(nbd);
1301	nbd_clear_que(nbd);
1302	nbd->task_setup = NULL;
1303	}
1304
1305	static void nbd_config_put(struct nbd_device *nbd)
1306	{
1307	if (refcount_dec_and_mutex_lock(r: &nbd->config_refs,
1308	lock: &nbd->config_lock)) {
1309	struct nbd_config *config = nbd->config;
1310	nbd_dev_dbg_close(nbd);
1311	invalidate_disk(disk: nbd->disk);
1312	if (nbd->config->bytesize)
1313	kobject_uevent(kobj: &nbd_to_dev(nbd)->kobj, action: KOBJ_CHANGE);
1314	if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1315	addr: &config->runtime_flags))
1316	device_remove_file(disk_to_dev(nbd->disk), attr: &pid_attr);
1317	nbd->pid = 0;
1318	if (test_and_clear_bit(NBD_RT_HAS_BACKEND_FILE,
1319	addr: &config->runtime_flags)) {
1320	device_remove_file(disk_to_dev(nbd->disk), attr: &backend_attr);
1321	kfree(objp: nbd->backend);
1322	nbd->backend = NULL;
1323	}
1324	nbd_clear_sock(nbd);
1325	if (config->num_connections) {
1326	int i;
1327	for (i = 0; i < config->num_connections; i++) {
1328	sockfd_put(config->socks[i]->sock);
1329	kfree(objp: config->socks[i]);
1330	}
1331	kfree(objp: config->socks);
1332	}
1333	kfree(objp: nbd->config);
1334	nbd->config = NULL;
1335
1336	nbd->tag_set.timeout = 0;
1337	nbd->disk->queue->limits.discard_granularity = 0;
1338	blk_queue_max_discard_sectors(q: nbd->disk->queue, max_discard_sectors: 0);
1339
1340	mutex_unlock(lock: &nbd->config_lock);
1341	nbd_put(nbd);
1342	module_put(THIS_MODULE);
1343	}
1344	}
1345
1346	static int nbd_start_device(struct nbd_device *nbd)
1347	{
1348	struct nbd_config *config = nbd->config;
1349	int num_connections = config->num_connections;
1350	int error = 0, i;
1351
1352	if (nbd->pid)
1353	return -EBUSY;
1354	if (!config->socks)
1355	return -EINVAL;
1356	if (num_connections > 1 &&
1357	!(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1358	dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1359	return -EINVAL;
1360	}
1361
1362	blk_mq_update_nr_hw_queues(set: &nbd->tag_set, nr_hw_queues: config->num_connections);
1363	nbd->pid = task_pid_nr(current);
1364
1365	nbd_parse_flags(nbd);
1366
1367	error = device_create_file(disk_to_dev(nbd->disk), entry: &pid_attr);
1368	if (error) {
1369	dev_err(disk_to_dev(nbd->disk), "device_create_file failed for pid!\n");
1370	return error;
1371	}
1372	set_bit(NBD_RT_HAS_PID_FILE, addr: &config->runtime_flags);
1373
1374	nbd_dev_dbg_init(nbd);
1375	for (i = 0; i < num_connections; i++) {
1376	struct recv_thread_args *args;
1377
1378	args = kzalloc(size: sizeof(*args), GFP_KERNEL);
1379	if (!args) {
1380	sock_shutdown(nbd);
1381	/*
1382	* If num_connections is m (2 < m),
1383	* and NO.1 ~ NO.n(1 < n < m) kzallocs are successful.
1384	* But NO.(n + 1) failed. We still have n recv threads.
1385	* So, add flush_workqueue here to prevent recv threads
1386	* dropping the last config_refs and trying to destroy
1387	* the workqueue from inside the workqueue.
1388	*/
1389	if (i)
1390	flush_workqueue(nbd->recv_workq);
1391	return -ENOMEM;
1392	}
1393	sk_set_memalloc(sk: config->socks[i]->sock->sk);
1394	if (nbd->tag_set.timeout)
1395	config->socks[i]->sock->sk->sk_sndtimeo =
1396	nbd->tag_set.timeout;
1397	atomic_inc(v: &config->recv_threads);
1398	refcount_inc(r: &nbd->config_refs);
1399	INIT_WORK(&args->work, recv_work);
1400	args->nbd = nbd;
1401	args->index = i;
1402	queue_work(wq: nbd->recv_workq, work: &args->work);
1403	}
1404	return nbd_set_size(nbd, bytesize: config->bytesize, blksize: nbd_blksize(config));
1405	}
1406
1407	static int nbd_start_device_ioctl(struct nbd_device *nbd)
1408	{
1409	struct nbd_config *config = nbd->config;
1410	int ret;
1411
1412	ret = nbd_start_device(nbd);
1413	if (ret)
1414	return ret;
1415
1416	if (max_part)
1417	set_bit(GD_NEED_PART_SCAN, addr: &nbd->disk->state);
1418	mutex_unlock(lock: &nbd->config_lock);
1419	ret = wait_event_interruptible(config->recv_wq,
1420	atomic_read(&config->recv_threads) == 0);
1421	if (ret) {
1422	sock_shutdown(nbd);
1423	nbd_clear_que(nbd);
1424	}
1425
1426	flush_workqueue(nbd->recv_workq);
1427	mutex_lock(&nbd->config_lock);
1428	nbd_bdev_reset(nbd);
1429	/* user requested, ignore socket errors */
1430	if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1431	ret = 0;
1432	if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1433	ret = -ETIMEDOUT;
1434	return ret;
1435	}
1436
1437	static void nbd_clear_sock_ioctl(struct nbd_device *nbd)
1438	{
1439	nbd_clear_sock(nbd);
1440	disk_force_media_change(disk: nbd->disk);
1441	nbd_bdev_reset(nbd);
1442	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1443	addr: &nbd->config->runtime_flags))
1444	nbd_config_put(nbd);
1445	}
1446
1447	static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1448	{
1449	nbd->tag_set.timeout = timeout * HZ;
1450	if (timeout)
1451	blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1452	else
1453	blk_queue_rq_timeout(nbd->disk->queue, 30 * HZ);
1454	}
1455
1456	/* Must be called with config_lock held */
1457	static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1458	unsigned int cmd, unsigned long arg)
1459	{
1460	struct nbd_config *config = nbd->config;
1461	loff_t bytesize;
1462
1463	switch (cmd) {
1464	case NBD_DISCONNECT:
1465	return nbd_disconnect(nbd);
1466	case NBD_CLEAR_SOCK:
1467	nbd_clear_sock_ioctl(nbd);
1468	return 0;
1469	case NBD_SET_SOCK:
1470	return nbd_add_socket(nbd, arg, netlink: false);
1471	case NBD_SET_BLKSIZE:
1472	return nbd_set_size(nbd, bytesize: config->bytesize, blksize: arg);
1473	case NBD_SET_SIZE:
1474	return nbd_set_size(nbd, bytesize: arg, blksize: nbd_blksize(config));
1475	case NBD_SET_SIZE_BLOCKS:
1476	if (check_shl_overflow(arg, config->blksize_bits, &bytesize))
1477	return -EINVAL;
1478	return nbd_set_size(nbd, bytesize, blksize: nbd_blksize(config));
1479	case NBD_SET_TIMEOUT:
1480	nbd_set_cmd_timeout(nbd, timeout: arg);
1481	return 0;
1482
1483	case NBD_SET_FLAGS:
1484	config->flags = arg;
1485	return 0;
1486	case NBD_DO_IT:
1487	return nbd_start_device_ioctl(nbd);
1488	case NBD_CLEAR_QUE:
1489	/*
1490	* This is for compatibility only. The queue is always cleared
1491	* by NBD_DO_IT or NBD_CLEAR_SOCK.
1492	*/
1493	return 0;
1494	case NBD_PRINT_DEBUG:
1495	/*
1496	* For compatibility only, we no longer keep a list of
1497	* outstanding requests.
1498	*/
1499	return 0;
1500	}
1501	return -ENOTTY;
1502	}
1503
1504	static int nbd_ioctl(struct block_device *bdev, blk_mode_t mode,
1505	unsigned int cmd, unsigned long arg)
1506	{
1507	struct nbd_device *nbd = bdev->bd_disk->private_data;
1508	struct nbd_config *config = nbd->config;
1509	int error = -EINVAL;
1510
1511	if (!capable(CAP_SYS_ADMIN))
1512	return -EPERM;
1513
1514	/* The block layer will pass back some non-nbd ioctls in case we have
1515	* special handling for them, but we don't so just return an error.
1516	*/
1517	if (_IOC_TYPE(cmd) != 0xab)
1518	return -EINVAL;
1519
1520	mutex_lock(&nbd->config_lock);
1521
1522	/* Don't allow ioctl operations on a nbd device that was created with
1523	* netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1524	*/
1525	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1526	(cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1527	error = __nbd_ioctl(bdev, nbd, cmd, arg);
1528	else
1529	dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1530	mutex_unlock(lock: &nbd->config_lock);
1531	return error;
1532	}
1533
1534	static struct nbd_config *nbd_alloc_config(void)
1535	{
1536	struct nbd_config *config;
1537
1538	if (!try_module_get(THIS_MODULE))
1539	return ERR_PTR(error: -ENODEV);
1540
1541	config = kzalloc(size: sizeof(struct nbd_config), GFP_NOFS);
1542	if (!config) {
1543	module_put(THIS_MODULE);
1544	return ERR_PTR(error: -ENOMEM);
1545	}
1546
1547	atomic_set(v: &config->recv_threads, i: 0);
1548	init_waitqueue_head(&config->recv_wq);
1549	init_waitqueue_head(&config->conn_wait);
1550	config->blksize_bits = NBD_DEF_BLKSIZE_BITS;
1551	atomic_set(v: &config->live_connections, i: 0);
1552	return config;
1553	}
1554
1555	static int nbd_open(struct gendisk *disk, blk_mode_t mode)
1556	{
1557	struct nbd_device *nbd;
1558	int ret = 0;
1559
1560	mutex_lock(&nbd_index_mutex);
1561	nbd = disk->private_data;
1562	if (!nbd) {
1563	ret = -ENXIO;
1564	goto out;
1565	}
1566	if (!refcount_inc_not_zero(r: &nbd->refs)) {
1567	ret = -ENXIO;
1568	goto out;
1569	}
1570	if (!refcount_inc_not_zero(r: &nbd->config_refs)) {
1571	struct nbd_config *config;
1572
1573	mutex_lock(&nbd->config_lock);
1574	if (refcount_inc_not_zero(r: &nbd->config_refs)) {
1575	mutex_unlock(lock: &nbd->config_lock);
1576	goto out;
1577	}
1578	config = nbd_alloc_config();
1579	if (IS_ERR(ptr: config)) {
1580	ret = PTR_ERR(ptr: config);
1581	mutex_unlock(lock: &nbd->config_lock);
1582	goto out;
1583	}
1584	nbd->config = config;
1585	refcount_set(r: &nbd->config_refs, n: 1);
1586	refcount_inc(r: &nbd->refs);
1587	mutex_unlock(lock: &nbd->config_lock);
1588	if (max_part)
1589	set_bit(GD_NEED_PART_SCAN, addr: &disk->state);
1590	} else if (nbd_disconnected(config: nbd->config)) {
1591	if (max_part)
1592	set_bit(GD_NEED_PART_SCAN, addr: &disk->state);
1593	}
1594	out:
1595	mutex_unlock(lock: &nbd_index_mutex);
1596	return ret;
1597	}
1598
1599	static void nbd_release(struct gendisk *disk)
1600	{
1601	struct nbd_device *nbd = disk->private_data;
1602
1603	if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1604	disk_openers(disk) == 0)
1605	nbd_disconnect_and_put(nbd);
1606
1607	nbd_config_put(nbd);
1608	nbd_put(nbd);
1609	}
1610
1611	static const struct block_device_operations nbd_fops =
1612	{
1613	.owner = THIS_MODULE,
1614	.open = nbd_open,
1615	.release = nbd_release,
1616	.ioctl = nbd_ioctl,
1617	.compat_ioctl = nbd_ioctl,
1618	};
1619
1620	#if IS_ENABLED(CONFIG_DEBUG_FS)
1621
1622	static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1623	{
1624	struct nbd_device *nbd = s->private;
1625
1626	if (nbd->pid)
1627	seq_printf(m: s, fmt: "recv: %d\n", nbd->pid);
1628
1629	return 0;
1630	}
1631
1632	DEFINE_SHOW_ATTRIBUTE(nbd_dbg_tasks);
1633
1634	static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1635	{
1636	struct nbd_device *nbd = s->private;
1637	u32 flags = nbd->config->flags;
1638
1639	seq_printf(m: s, fmt: "Hex: 0x%08x\n\n", flags);
1640
1641	seq_puts(m: s, s: "Known flags:\n");
1642
1643	if (flags & NBD_FLAG_HAS_FLAGS)
1644	seq_puts(m: s, s: "NBD_FLAG_HAS_FLAGS\n");
1645	if (flags & NBD_FLAG_READ_ONLY)
1646	seq_puts(m: s, s: "NBD_FLAG_READ_ONLY\n");
1647	if (flags & NBD_FLAG_SEND_FLUSH)
1648	seq_puts(m: s, s: "NBD_FLAG_SEND_FLUSH\n");
1649	if (flags & NBD_FLAG_SEND_FUA)
1650	seq_puts(m: s, s: "NBD_FLAG_SEND_FUA\n");
1651	if (flags & NBD_FLAG_SEND_TRIM)
1652	seq_puts(m: s, s: "NBD_FLAG_SEND_TRIM\n");
1653
1654	return 0;
1655	}
1656
1657	DEFINE_SHOW_ATTRIBUTE(nbd_dbg_flags);
1658
1659	static int nbd_dev_dbg_init(struct nbd_device *nbd)
1660	{
1661	struct dentry *dir;
1662	struct nbd_config *config = nbd->config;
1663
1664	if (!nbd_dbg_dir)
1665	return -EIO;
1666
1667	dir = debugfs_create_dir(nbd_name(nbd), parent: nbd_dbg_dir);
1668	if (IS_ERR(ptr: dir)) {
1669	dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1670	nbd_name(nbd));
1671	return -EIO;
1672	}
1673	config->dbg_dir = dir;
1674
1675	debugfs_create_file(name: "tasks", mode: 0444, parent: dir, data: nbd, fops: &nbd_dbg_tasks_fops);
1676	debugfs_create_u64(name: "size_bytes", mode: 0444, parent: dir, value: &config->bytesize);
1677	debugfs_create_u32(name: "timeout", mode: 0444, parent: dir, value: &nbd->tag_set.timeout);
1678	debugfs_create_u32(name: "blocksize_bits", mode: 0444, parent: dir, value: &config->blksize_bits);
1679	debugfs_create_file(name: "flags", mode: 0444, parent: dir, data: nbd, fops: &nbd_dbg_flags_fops);
1680
1681	return 0;
1682	}
1683
1684	static void nbd_dev_dbg_close(struct nbd_device *nbd)
1685	{
1686	debugfs_remove_recursive(dentry: nbd->config->dbg_dir);
1687	}
1688
1689	static int nbd_dbg_init(void)
1690	{
1691	struct dentry *dbg_dir;
1692
1693	dbg_dir = debugfs_create_dir(name: "nbd", NULL);
1694	if (IS_ERR(ptr: dbg_dir))
1695	return -EIO;
1696
1697	nbd_dbg_dir = dbg_dir;
1698
1699	return 0;
1700	}
1701
1702	static void nbd_dbg_close(void)
1703	{
1704	debugfs_remove_recursive(dentry: nbd_dbg_dir);
1705	}
1706
1707	#else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1708
1709	static int nbd_dev_dbg_init(struct nbd_device *nbd)
1710	{
1711	return 0;
1712	}
1713
1714	static void nbd_dev_dbg_close(struct nbd_device *nbd)
1715	{
1716	}
1717
1718	static int nbd_dbg_init(void)
1719	{
1720	return 0;
1721	}
1722
1723	static void nbd_dbg_close(void)
1724	{
1725	}
1726
1727	#endif
1728
1729	static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1730	unsigned int hctx_idx, unsigned int numa_node)
1731	{
1732	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1733	cmd->nbd = set->driver_data;
1734	cmd->flags = 0;
1735	mutex_init(&cmd->lock);
1736	return 0;
1737	}
1738
1739	static const struct blk_mq_ops nbd_mq_ops = {
1740	.queue_rq = nbd_queue_rq,
1741	.complete = nbd_complete_rq,
1742	.init_request = nbd_init_request,
1743	.timeout = nbd_xmit_timeout,
1744	};
1745
1746	static struct nbd_device *nbd_dev_add(int index, unsigned int refs)
1747	{
1748	struct nbd_device *nbd;
1749	struct gendisk *disk;
1750	int err = -ENOMEM;
1751
1752	nbd = kzalloc(size: sizeof(struct nbd_device), GFP_KERNEL);
1753	if (!nbd)
1754	goto out;
1755
1756	nbd->tag_set.ops = &nbd_mq_ops;
1757	nbd->tag_set.nr_hw_queues = 1;
1758	nbd->tag_set.queue_depth = 128;
1759	nbd->tag_set.numa_node = NUMA_NO_NODE;
1760	nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1761	nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1762	BLK_MQ_F_BLOCKING;
1763	nbd->tag_set.driver_data = nbd;
1764	INIT_WORK(&nbd->remove_work, nbd_dev_remove_work);
1765	nbd->backend = NULL;
1766
1767	err = blk_mq_alloc_tag_set(set: &nbd->tag_set);
1768	if (err)
1769	goto out_free_nbd;
1770
1771	mutex_lock(&nbd_index_mutex);
1772	if (index >= 0) {
1773	err = idr_alloc(&nbd_index_idr, ptr: nbd, start: index, end: index + 1,
1774	GFP_KERNEL);
1775	if (err == -ENOSPC)
1776	err = -EEXIST;
1777	} else {
1778	err = idr_alloc(&nbd_index_idr, ptr: nbd, start: 0,
1779	end: (MINORMASK >> part_shift) + 1, GFP_KERNEL);
1780	if (err >= 0)
1781	index = err;
1782	}
1783	nbd->index = index;
1784	mutex_unlock(lock: &nbd_index_mutex);
1785	if (err < 0)
1786	goto out_free_tags;
1787
1788	disk = blk_mq_alloc_disk(&nbd->tag_set, NULL);
1789	if (IS_ERR(ptr: disk)) {
1790	err = PTR_ERR(ptr: disk);
1791	goto out_free_idr;
1792	}
1793	nbd->disk = disk;
1794
1795	nbd->recv_workq = alloc_workqueue(fmt: "nbd%d-recv",
1796	flags: WQ_MEM_RECLAIM | WQ_HIGHPRI |
1797	WQ_UNBOUND, max_active: 0, nbd->index);
1798	if (!nbd->recv_workq) {
1799	dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1800	err = -ENOMEM;
1801	goto out_err_disk;
1802	}
1803
1804	/*
1805	* Tell the block layer that we are not a rotational device
1806	*/
1807	blk_queue_flag_set(QUEUE_FLAG_NONROT, q: disk->queue);
1808	disk->queue->limits.discard_granularity = 0;
1809	blk_queue_max_discard_sectors(q: disk->queue, max_discard_sectors: 0);
1810	blk_queue_max_segment_size(disk->queue, UINT_MAX);
1811	blk_queue_max_segments(disk->queue, USHRT_MAX);
1812	blk_queue_max_hw_sectors(disk->queue, 65536);
1813	disk->queue->limits.max_sectors = 256;
1814
1815	mutex_init(&nbd->config_lock);
1816	refcount_set(r: &nbd->config_refs, n: 0);
1817	/*
1818	* Start out with a zero references to keep other threads from using
1819	* this device until it is fully initialized.
1820	*/
1821	refcount_set(r: &nbd->refs, n: 0);
1822	INIT_LIST_HEAD(list: &nbd->list);
1823	disk->major = NBD_MAJOR;
1824	disk->first_minor = index << part_shift;
1825	disk->minors = 1 << part_shift;
1826	disk->fops = &nbd_fops;
1827	disk->private_data = nbd;
1828	sprintf(buf: disk->disk_name, fmt: "nbd%d", index);
1829	err = add_disk(disk);
1830	if (err)
1831	goto out_free_work;
1832
1833	/*
1834	* Now publish the device.
1835	*/
1836	refcount_set(r: &nbd->refs, n: refs);
1837	nbd_total_devices++;
1838	return nbd;
1839
1840	out_free_work:
1841	destroy_workqueue(wq: nbd->recv_workq);
1842	out_err_disk:
1843	put_disk(disk);
1844	out_free_idr:
1845	mutex_lock(&nbd_index_mutex);
1846	idr_remove(&nbd_index_idr, id: index);
1847	mutex_unlock(lock: &nbd_index_mutex);
1848	out_free_tags:
1849	blk_mq_free_tag_set(set: &nbd->tag_set);
1850	out_free_nbd:
1851	kfree(objp: nbd);
1852	out:
1853	return ERR_PTR(error: err);
1854	}
1855
1856	static struct nbd_device *nbd_find_get_unused(void)
1857	{
1858	struct nbd_device *nbd;
1859	int id;
1860
1861	lockdep_assert_held(&nbd_index_mutex);
1862
1863	idr_for_each_entry(&nbd_index_idr, nbd, id) {
1864	if (refcount_read(r: &nbd->config_refs) ||
1865	test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags))
1866	continue;
1867	if (refcount_inc_not_zero(r: &nbd->refs))
1868	return nbd;
1869	}
1870
1871	return NULL;
1872	}
1873
1874	/* Netlink interface. */
1875	static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1876	[NBD_ATTR_INDEX] = { .type = NLA_U32 },
1877	[NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1878	[NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1879	[NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1880	[NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1881	[NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1882	[NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1883	[NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1884	[NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1885	[NBD_ATTR_BACKEND_IDENTIFIER] = { .type = NLA_STRING},
1886	};
1887
1888	static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1889	[NBD_SOCK_FD] = { .type = NLA_U32 },
1890	};
1891
1892	/* We don't use this right now since we don't parse the incoming list, but we
1893	* still want it here so userspace knows what to expect.
1894	*/
1895	static const struct nla_policy __attribute__((unused))
1896	nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1897	[NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1898	[NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1899	};
1900
1901	static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1902	{
1903	struct nbd_config *config = nbd->config;
1904	u64 bsize = nbd_blksize(config);
1905	u64 bytes = config->bytesize;
1906
1907	if (info->attrs[NBD_ATTR_SIZE_BYTES])
1908	bytes = nla_get_u64(nla: info->attrs[NBD_ATTR_SIZE_BYTES]);
1909
1910	if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES])
1911	bsize = nla_get_u64(nla: info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1912
1913	if (bytes != config->bytesize || bsize != nbd_blksize(config))
1914	return nbd_set_size(nbd, bytesize: bytes, blksize: bsize);
1915	return 0;
1916	}
1917
1918	static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1919	{
1920	struct nbd_device *nbd;
1921	struct nbd_config *config;
1922	int index = -1;
1923	int ret;
1924	bool put_dev = false;
1925
1926	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1927	return -EPERM;
1928
1929	if (info->attrs[NBD_ATTR_INDEX]) {
1930	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
1931
1932	/*
1933	* Too big first_minor can cause duplicate creation of
1934	* sysfs files/links, since index << part_shift might overflow, or
1935	* MKDEV() expect that the max bits of first_minor is 20.
1936	*/
1937	if (index < 0 || index > MINORMASK >> part_shift) {
1938	pr_err("illegal input index %d\n", index);
1939	return -EINVAL;
1940	}
1941	}
1942	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_SOCKETS)) {
1943	pr_err("must specify at least one socket\n");
1944	return -EINVAL;
1945	}
1946	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_SIZE_BYTES)) {
1947	pr_err("must specify a size in bytes for the device\n");
1948	return -EINVAL;
1949	}
1950	again:
1951	mutex_lock(&nbd_index_mutex);
1952	if (index == -1) {
1953	nbd = nbd_find_get_unused();
1954	} else {
1955	nbd = idr_find(&nbd_index_idr, id: index);
1956	if (nbd) {
1957	if ((test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1958	test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) ||
1959	!refcount_inc_not_zero(r: &nbd->refs)) {
1960	mutex_unlock(lock: &nbd_index_mutex);
1961	pr_err("device at index %d is going down\n",
1962	index);
1963	return -EINVAL;
1964	}
1965	}
1966	}
1967	mutex_unlock(lock: &nbd_index_mutex);
1968
1969	if (!nbd) {
1970	nbd = nbd_dev_add(index, refs: 2);
1971	if (IS_ERR(ptr: nbd)) {
1972	pr_err("failed to add new device\n");
1973	return PTR_ERR(ptr: nbd);
1974	}
1975	}
1976
1977	mutex_lock(&nbd->config_lock);
1978	if (refcount_read(r: &nbd->config_refs)) {
1979	mutex_unlock(lock: &nbd->config_lock);
1980	nbd_put(nbd);
1981	if (index == -1)
1982	goto again;
1983	pr_err("nbd%d already in use\n", index);
1984	return -EBUSY;
1985	}
1986	if (WARN_ON(nbd->config)) {
1987	mutex_unlock(lock: &nbd->config_lock);
1988	nbd_put(nbd);
1989	return -EINVAL;
1990	}
1991	config = nbd_alloc_config();
1992	if (IS_ERR(ptr: config)) {
1993	mutex_unlock(lock: &nbd->config_lock);
1994	nbd_put(nbd);
1995	pr_err("couldn't allocate config\n");
1996	return PTR_ERR(ptr: config);
1997	}
1998	nbd->config = config;
1999	refcount_set(r: &nbd->config_refs, n: 1);
2000	set_bit(NBD_RT_BOUND, addr: &config->runtime_flags);
2001
2002	ret = nbd_genl_size_set(info, nbd);
2003	if (ret)
2004	goto out;
2005
2006	if (info->attrs[NBD_ATTR_TIMEOUT])
2007	nbd_set_cmd_timeout(nbd,
2008	timeout: nla_get_u64(nla: info->attrs[NBD_ATTR_TIMEOUT]));
2009	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2010	config->dead_conn_timeout =
2011	nla_get_u64(nla: info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2012	config->dead_conn_timeout *= HZ;
2013	}
2014	if (info->attrs[NBD_ATTR_SERVER_FLAGS])
2015	config->flags =
2016	nla_get_u64(nla: info->attrs[NBD_ATTR_SERVER_FLAGS]);
2017	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2018	u64 flags = nla_get_u64(nla: info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2019	if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2020	/*
2021	* We have 1 ref to keep the device around, and then 1
2022	* ref for our current operation here, which will be
2023	* inherited by the config. If we already have
2024	* DESTROY_ON_DISCONNECT set then we know we don't have
2025	* that extra ref already held so we don't need the
2026	* put_dev.
2027	*/
2028	if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
2029	addr: &nbd->flags))
2030	put_dev = true;
2031	} else {
2032	if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
2033	addr: &nbd->flags))
2034	refcount_inc(r: &nbd->refs);
2035	}
2036	if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2037	set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2038	addr: &config->runtime_flags);
2039	}
2040	}
2041
2042	if (info->attrs[NBD_ATTR_SOCKETS]) {
2043	struct nlattr *attr;
2044	int rem, fd;
2045
2046	nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2047	rem) {
2048	struct nlattr *socks[NBD_SOCK_MAX+1];
2049
2050	if (nla_type(nla: attr) != NBD_SOCK_ITEM) {
2051	pr_err("socks must be embedded in a SOCK_ITEM attr\n");
2052	ret = -EINVAL;
2053	goto out;
2054	}
2055	ret = nla_parse_nested_deprecated(tb: socks, NBD_SOCK_MAX,
2056	nla: attr,
2057	policy: nbd_sock_policy,
2058	extack: info->extack);
2059	if (ret != 0) {
2060	pr_err("error processing sock list\n");
2061	ret = -EINVAL;
2062	goto out;
2063	}
2064	if (!socks[NBD_SOCK_FD])
2065	continue;
2066	fd = (int)nla_get_u32(nla: socks[NBD_SOCK_FD]);
2067	ret = nbd_add_socket(nbd, arg: fd, netlink: true);
2068	if (ret)
2069	goto out;
2070	}
2071	}
2072	ret = nbd_start_device(nbd);
2073	if (ret)
2074	goto out;
2075	if (info->attrs[NBD_ATTR_BACKEND_IDENTIFIER]) {
2076	nbd->backend = nla_strdup(nla: info->attrs[NBD_ATTR_BACKEND_IDENTIFIER],
2077	GFP_KERNEL);
2078	if (!nbd->backend) {
2079	ret = -ENOMEM;
2080	goto out;
2081	}
2082	}
2083	ret = device_create_file(disk_to_dev(nbd->disk), entry: &backend_attr);
2084	if (ret) {
2085	dev_err(disk_to_dev(nbd->disk),
2086	"device_create_file failed for backend!\n");
2087	goto out;
2088	}
2089	set_bit(NBD_RT_HAS_BACKEND_FILE, addr: &config->runtime_flags);
2090	out:
2091	mutex_unlock(lock: &nbd->config_lock);
2092	if (!ret) {
2093	set_bit(NBD_RT_HAS_CONFIG_REF, addr: &config->runtime_flags);
2094	refcount_inc(r: &nbd->config_refs);
2095	nbd_connect_reply(info, index: nbd->index);
2096	}
2097	nbd_config_put(nbd);
2098	if (put_dev)
2099	nbd_put(nbd);
2100	return ret;
2101	}
2102
2103	static void nbd_disconnect_and_put(struct nbd_device *nbd)
2104	{
2105	mutex_lock(&nbd->config_lock);
2106	nbd_disconnect(nbd);
2107	sock_shutdown(nbd);
2108	wake_up(&nbd->config->conn_wait);
2109	/*
2110	* Make sure recv thread has finished, we can safely call nbd_clear_que()
2111	* to cancel the inflight I/Os.
2112	*/
2113	flush_workqueue(nbd->recv_workq);
2114	nbd_clear_que(nbd);
2115	nbd->task_setup = NULL;
2116	mutex_unlock(lock: &nbd->config_lock);
2117
2118	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
2119	addr: &nbd->config->runtime_flags))
2120	nbd_config_put(nbd);
2121	}
2122
2123	static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
2124	{
2125	struct nbd_device *nbd;
2126	int index;
2127
2128	if (!netlink_capable(skb, CAP_SYS_ADMIN))
2129	return -EPERM;
2130
2131	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_INDEX)) {
2132	pr_err("must specify an index to disconnect\n");
2133	return -EINVAL;
2134	}
2135	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
2136	mutex_lock(&nbd_index_mutex);
2137	nbd = idr_find(&nbd_index_idr, id: index);
2138	if (!nbd) {
2139	mutex_unlock(lock: &nbd_index_mutex);
2140	pr_err("couldn't find device at index %d\n", index);
2141	return -EINVAL;
2142	}
2143	if (!refcount_inc_not_zero(r: &nbd->refs)) {
2144	mutex_unlock(lock: &nbd_index_mutex);
2145	pr_err("device at index %d is going down\n", index);
2146	return -EINVAL;
2147	}
2148	mutex_unlock(lock: &nbd_index_mutex);
2149	if (!refcount_inc_not_zero(r: &nbd->config_refs))
2150	goto put_nbd;
2151	nbd_disconnect_and_put(nbd);
2152	nbd_config_put(nbd);
2153	put_nbd:
2154	nbd_put(nbd);
2155	return 0;
2156	}
2157
2158	static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2159	{
2160	struct nbd_device *nbd = NULL;
2161	struct nbd_config *config;
2162	int index;
2163	int ret = 0;
2164	bool put_dev = false;
2165
2166	if (!netlink_capable(skb, CAP_SYS_ADMIN))
2167	return -EPERM;
2168
2169	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_INDEX)) {
2170	pr_err("must specify a device to reconfigure\n");
2171	return -EINVAL;
2172	}
2173	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
2174	mutex_lock(&nbd_index_mutex);
2175	nbd = idr_find(&nbd_index_idr, id: index);
2176	if (!nbd) {
2177	mutex_unlock(lock: &nbd_index_mutex);
2178	pr_err("couldn't find a device at index %d\n", index);
2179	return -EINVAL;
2180	}
2181	if (nbd->backend) {
2182	if (info->attrs[NBD_ATTR_BACKEND_IDENTIFIER]) {
2183	if (nla_strcmp(nla: info->attrs[NBD_ATTR_BACKEND_IDENTIFIER],
2184	str: nbd->backend)) {
2185	mutex_unlock(lock: &nbd_index_mutex);
2186	dev_err(nbd_to_dev(nbd),
2187	"backend image doesn't match with %s\n",
2188	nbd->backend);
2189	return -EINVAL;
2190	}
2191	} else {
2192	mutex_unlock(lock: &nbd_index_mutex);
2193	dev_err(nbd_to_dev(nbd), "must specify backend\n");
2194	return -EINVAL;
2195	}
2196	}
2197	if (!refcount_inc_not_zero(r: &nbd->refs)) {
2198	mutex_unlock(lock: &nbd_index_mutex);
2199	pr_err("device at index %d is going down\n", index);
2200	return -EINVAL;
2201	}
2202	mutex_unlock(lock: &nbd_index_mutex);
2203
2204	if (!refcount_inc_not_zero(r: &nbd->config_refs)) {
2205	dev_err(nbd_to_dev(nbd),
2206	"not configured, cannot reconfigure\n");
2207	nbd_put(nbd);
2208	return -EINVAL;
2209	}
2210
2211	mutex_lock(&nbd->config_lock);
2212	config = nbd->config;
2213	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2214	!nbd->pid) {
2215	dev_err(nbd_to_dev(nbd),
2216	"not configured, cannot reconfigure\n");
2217	ret = -EINVAL;
2218	goto out;
2219	}
2220
2221	ret = nbd_genl_size_set(info, nbd);
2222	if (ret)
2223	goto out;
2224
2225	if (info->attrs[NBD_ATTR_TIMEOUT])
2226	nbd_set_cmd_timeout(nbd,
2227	timeout: nla_get_u64(nla: info->attrs[NBD_ATTR_TIMEOUT]));
2228	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2229	config->dead_conn_timeout =
2230	nla_get_u64(nla: info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2231	config->dead_conn_timeout *= HZ;
2232	}
2233	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2234	u64 flags = nla_get_u64(nla: info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2235	if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2236	if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
2237	addr: &nbd->flags))
2238	put_dev = true;
2239	} else {
2240	if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
2241	addr: &nbd->flags))
2242	refcount_inc(r: &nbd->refs);
2243	}
2244
2245	if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2246	set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2247	addr: &config->runtime_flags);
2248	} else {
2249	clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2250	addr: &config->runtime_flags);
2251	}
2252	}
2253
2254	if (info->attrs[NBD_ATTR_SOCKETS]) {
2255	struct nlattr *attr;
2256	int rem, fd;
2257
2258	nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2259	rem) {
2260	struct nlattr *socks[NBD_SOCK_MAX+1];
2261
2262	if (nla_type(nla: attr) != NBD_SOCK_ITEM) {
2263	pr_err("socks must be embedded in a SOCK_ITEM attr\n");
2264	ret = -EINVAL;
2265	goto out;
2266	}
2267	ret = nla_parse_nested_deprecated(tb: socks, NBD_SOCK_MAX,
2268	nla: attr,
2269	policy: nbd_sock_policy,
2270	extack: info->extack);
2271	if (ret != 0) {
2272	pr_err("error processing sock list\n");
2273	ret = -EINVAL;
2274	goto out;
2275	}
2276	if (!socks[NBD_SOCK_FD])
2277	continue;
2278	fd = (int)nla_get_u32(nla: socks[NBD_SOCK_FD]);
2279	ret = nbd_reconnect_socket(nbd, arg: fd);
2280	if (ret) {
2281	if (ret == -ENOSPC)
2282	ret = 0;
2283	goto out;
2284	}
2285	dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2286	}
2287	}
2288	out:
2289	mutex_unlock(lock: &nbd->config_lock);
2290	nbd_config_put(nbd);
2291	nbd_put(nbd);
2292	if (put_dev)
2293	nbd_put(nbd);
2294	return ret;
2295	}
2296
2297	static const struct genl_small_ops nbd_connect_genl_ops[] = {
2298	{
2299	.cmd = NBD_CMD_CONNECT,
2300	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2301	.doit = nbd_genl_connect,
2302	},
2303	{
2304	.cmd = NBD_CMD_DISCONNECT,
2305	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2306	.doit = nbd_genl_disconnect,
2307	},
2308	{
2309	.cmd = NBD_CMD_RECONFIGURE,
2310	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2311	.doit = nbd_genl_reconfigure,
2312	},
2313	{
2314	.cmd = NBD_CMD_STATUS,
2315	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2316	.doit = nbd_genl_status,
2317	},
2318	};
2319
2320	static const struct genl_multicast_group nbd_mcast_grps[] = {
2321	{ .name = NBD_GENL_MCAST_GROUP_NAME, },
2322	};
2323
2324	static struct genl_family nbd_genl_family __ro_after_init = {
2325	.hdrsize = 0,
2326	.name = NBD_GENL_FAMILY_NAME,
2327	.version = NBD_GENL_VERSION,
2328	.module = THIS_MODULE,
2329	.small_ops = nbd_connect_genl_ops,
2330	.n_small_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2331	.resv_start_op = NBD_CMD_STATUS + 1,
2332	.maxattr = NBD_ATTR_MAX,
2333	.netnsok = 1,
2334	.policy = nbd_attr_policy,
2335	.mcgrps = nbd_mcast_grps,
2336	.n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2337	};
2338	MODULE_ALIAS_GENL_FAMILY(NBD_GENL_FAMILY_NAME);
2339
2340	static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2341	{
2342	struct nlattr *dev_opt;
2343	u8 connected = 0;
2344	int ret;
2345
2346	/* This is a little racey, but for status it's ok. The
2347	* reason we don't take a ref here is because we can't
2348	* take a ref in the index == -1 case as we would need
2349	* to put under the nbd_index_mutex, which could
2350	* deadlock if we are configured to remove ourselves
2351	* once we're disconnected.
2352	*/
2353	if (refcount_read(r: &nbd->config_refs))
2354	connected = 1;
2355	dev_opt = nla_nest_start_noflag(skb: reply, attrtype: NBD_DEVICE_ITEM);
2356	if (!dev_opt)
2357	return -EMSGSIZE;
2358	ret = nla_put_u32(skb: reply, attrtype: NBD_DEVICE_INDEX, value: nbd->index);
2359	if (ret)
2360	return -EMSGSIZE;
2361	ret = nla_put_u8(skb: reply, attrtype: NBD_DEVICE_CONNECTED,
2362	value: connected);
2363	if (ret)
2364	return -EMSGSIZE;
2365	nla_nest_end(skb: reply, start: dev_opt);
2366	return 0;
2367	}
2368
2369	static int status_cb(int id, void *ptr, void *data)
2370	{
2371	struct nbd_device *nbd = ptr;
2372	return populate_nbd_status(nbd, reply: (struct sk_buff *)data);
2373	}
2374
2375	static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2376	{
2377	struct nlattr *dev_list;
2378	struct sk_buff *reply;
2379	void *reply_head;
2380	size_t msg_size;
2381	int index = -1;
2382	int ret = -ENOMEM;
2383
2384	if (info->attrs[NBD_ATTR_INDEX])
2385	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
2386
2387	mutex_lock(&nbd_index_mutex);
2388
2389	msg_size = nla_total_size(payload: nla_attr_size(payload: sizeof(u32)) +
2390	nla_attr_size(payload: sizeof(u8)));
2391	msg_size *= (index == -1) ? nbd_total_devices : 1;
2392
2393	reply = genlmsg_new(payload: msg_size, GFP_KERNEL);
2394	if (!reply)
2395	goto out;
2396	reply_head = genlmsg_put_reply(skb: reply, info, family: &nbd_genl_family, flags: 0,
2397	cmd: NBD_CMD_STATUS);
2398	if (!reply_head) {
2399	nlmsg_free(skb: reply);
2400	goto out;
2401	}
2402
2403	dev_list = nla_nest_start_noflag(skb: reply, attrtype: NBD_ATTR_DEVICE_LIST);
2404	if (index == -1) {
2405	ret = idr_for_each(&nbd_index_idr, fn: &status_cb, data: reply);
2406	if (ret) {
2407	nlmsg_free(skb: reply);
2408	goto out;
2409	}
2410	} else {
2411	struct nbd_device *nbd;
2412	nbd = idr_find(&nbd_index_idr, id: index);
2413	if (nbd) {
2414	ret = populate_nbd_status(nbd, reply);
2415	if (ret) {
2416	nlmsg_free(skb: reply);
2417	goto out;
2418	}
2419	}
2420	}
2421	nla_nest_end(skb: reply, start: dev_list);
2422	genlmsg_end(skb: reply, hdr: reply_head);
2423	ret = genlmsg_reply(skb: reply, info);
2424	out:
2425	mutex_unlock(lock: &nbd_index_mutex);
2426	return ret;
2427	}
2428
2429	static void nbd_connect_reply(struct genl_info *info, int index)
2430	{
2431	struct sk_buff *skb;
2432	void *msg_head;
2433	int ret;
2434
2435	skb = genlmsg_new(payload: nla_total_size(payload: sizeof(u32)), GFP_KERNEL);
2436	if (!skb)
2437	return;
2438	msg_head = genlmsg_put_reply(skb, info, family: &nbd_genl_family, flags: 0,
2439	cmd: NBD_CMD_CONNECT);
2440	if (!msg_head) {
2441	nlmsg_free(skb);
2442	return;
2443	}
2444	ret = nla_put_u32(skb, attrtype: NBD_ATTR_INDEX, value: index);
2445	if (ret) {
2446	nlmsg_free(skb);
2447	return;
2448	}
2449	genlmsg_end(skb, hdr: msg_head);
2450	genlmsg_reply(skb, info);
2451	}
2452
2453	static void nbd_mcast_index(int index)
2454	{
2455	struct sk_buff *skb;
2456	void *msg_head;
2457	int ret;
2458
2459	skb = genlmsg_new(payload: nla_total_size(payload: sizeof(u32)), GFP_KERNEL);
2460	if (!skb)
2461	return;
2462	msg_head = genlmsg_put(skb, portid: 0, seq: 0, family: &nbd_genl_family, flags: 0,
2463	cmd: NBD_CMD_LINK_DEAD);
2464	if (!msg_head) {
2465	nlmsg_free(skb);
2466	return;
2467	}
2468	ret = nla_put_u32(skb, attrtype: NBD_ATTR_INDEX, value: index);
2469	if (ret) {
2470	nlmsg_free(skb);
2471	return;
2472	}
2473	genlmsg_end(skb, hdr: msg_head);
2474	genlmsg_multicast(family: &nbd_genl_family, skb, portid: 0, group: 0, GFP_KERNEL);
2475	}
2476
2477	static void nbd_dead_link_work(struct work_struct *work)
2478	{
2479	struct link_dead_args *args = container_of(work, struct link_dead_args,
2480	work);
2481	nbd_mcast_index(index: args->index);
2482	kfree(objp: args);
2483	}
2484
2485	static int __init nbd_init(void)
2486	{
2487	int i;
2488
2489	BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2490
2491	if (max_part < 0) {
2492	pr_err("max_part must be >= 0\n");
2493	return -EINVAL;
2494	}
2495
2496	part_shift = 0;
2497	if (max_part > 0) {
2498	part_shift = fls(x: max_part);
2499
2500	/*
2501	* Adjust max_part according to part_shift as it is exported
2502	* to user space so that user can know the max number of
2503	* partition kernel should be able to manage.
2504	*
2505	* Note that -1 is required because partition 0 is reserved
2506	* for the whole disk.
2507	*/
2508	max_part = (1UL << part_shift) - 1;
2509	}
2510
2511	if ((1UL << part_shift) > DISK_MAX_PARTS)
2512	return -EINVAL;
2513
2514	if (nbds_max > 1UL << (MINORBITS - part_shift))
2515	return -EINVAL;
2516
2517	if (register_blkdev(NBD_MAJOR, "nbd"))
2518	return -EIO;
2519
2520	nbd_del_wq = alloc_workqueue(fmt: "nbd-del", flags: WQ_UNBOUND, max_active: 0);
2521	if (!nbd_del_wq) {
2522	unregister_blkdev(NBD_MAJOR, name: "nbd");
2523	return -ENOMEM;
2524	}
2525
2526	if (genl_register_family(family: &nbd_genl_family)) {
2527	destroy_workqueue(wq: nbd_del_wq);
2528	unregister_blkdev(NBD_MAJOR, name: "nbd");
2529	return -EINVAL;
2530	}
2531	nbd_dbg_init();
2532
2533	for (i = 0; i < nbds_max; i++)
2534	nbd_dev_add(index: i, refs: 1);
2535	return 0;
2536	}
2537
2538	static int nbd_exit_cb(int id, void *ptr, void *data)
2539	{
2540	struct list_head *list = (struct list_head *)data;
2541	struct nbd_device *nbd = ptr;
2542
2543	/* Skip nbd that is being removed asynchronously */
2544	if (refcount_read(r: &nbd->refs))
2545	list_add_tail(new: &nbd->list, head: list);
2546
2547	return 0;
2548	}
2549
2550	static void __exit nbd_cleanup(void)
2551	{
2552	struct nbd_device *nbd;
2553	LIST_HEAD(del_list);
2554
2555	/*
2556	* Unregister netlink interface prior to waiting
2557	* for the completion of netlink commands.
2558	*/
2559	genl_unregister_family(family: &nbd_genl_family);
2560
2561	nbd_dbg_close();
2562
2563	mutex_lock(&nbd_index_mutex);
2564	idr_for_each(&nbd_index_idr, fn: &nbd_exit_cb, data: &del_list);
2565	mutex_unlock(lock: &nbd_index_mutex);
2566
2567	while (!list_empty(head: &del_list)) {
2568	nbd = list_first_entry(&del_list, struct nbd_device, list);
2569	list_del_init(entry: &nbd->list);
2570	if (refcount_read(r: &nbd->config_refs))
2571	pr_err("possibly leaking nbd_config (ref %d)\n",
2572	refcount_read(&nbd->config_refs));
2573	if (refcount_read(r: &nbd->refs) != 1)
2574	pr_err("possibly leaking a device\n");
2575	nbd_put(nbd);
2576	}
2577
2578	/* Also wait for nbd_dev_remove_work() completes */
2579	destroy_workqueue(wq: nbd_del_wq);
2580
2581	idr_destroy(&nbd_index_idr);
2582	unregister_blkdev(NBD_MAJOR, name: "nbd");
2583	}
2584
2585	module_init(nbd_init);
2586	module_exit(nbd_cleanup);
2587
2588	MODULE_DESCRIPTION("Network Block Device");
2589	MODULE_LICENSE("GPL");
2590
2591	module_param(nbds_max, int, 0444);
2592	MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2593	module_param(max_part, int, 0444);
2594	MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");
2595