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	struct work_struct work;
66	};
67
68	struct recv_thread_args {
69	struct work_struct work;
70	struct nbd_device *nbd;
71	struct nbd_sock *nsock;
72	int index;
73	};
74
75	struct link_dead_args {
76	struct work_struct work;
77	int index;
78	};
79
80	#define NBD_RT_TIMEDOUT 0
81	#define NBD_RT_DISCONNECT_REQUESTED 1
82	#define NBD_RT_DISCONNECTED 2
83	#define NBD_RT_HAS_PID_FILE 3
84	#define NBD_RT_HAS_CONFIG_REF 4
85	#define NBD_RT_BOUND 5
86	#define NBD_RT_DISCONNECT_ON_CLOSE 6
87	#define NBD_RT_HAS_BACKEND_FILE 7
88
89	#define NBD_DESTROY_ON_DISCONNECT 0
90	#define NBD_DISCONNECT_REQUESTED 1
91
92	struct nbd_config {
93	u32 flags;
94	unsigned long runtime_flags;
95	u64 dead_conn_timeout;
96
97	struct nbd_sock **socks;
98	int num_connections;
99	atomic_t live_connections;
100	wait_queue_head_t conn_wait;
101
102	atomic_t recv_threads;
103	wait_queue_head_t recv_wq;
104	unsigned int blksize_bits;
105	loff_t bytesize;
106	#if IS_ENABLED(CONFIG_DEBUG_FS)
107	struct dentry *dbg_dir;
108	#endif
109	};
110
111	static inline unsigned int nbd_blksize(struct nbd_config *config)
112	{
113	return 1u << config->blksize_bits;
114	}
115
116	struct nbd_device {
117	struct blk_mq_tag_set tag_set;
118
119	int index;
120	refcount_t config_refs;
121	refcount_t refs;
122	struct nbd_config *config;
123	struct mutex config_lock;
124	struct gendisk *disk;
125	struct workqueue_struct *recv_workq;
126	struct work_struct remove_work;
127
128	struct list_head list;
129	struct task_struct *task_setup;
130
131	unsigned long flags;
132	pid_t pid; /* pid of nbd-client, if attached */
133
134	char *backend;
135	};
136
137	#define NBD_CMD_REQUEUED 1
138	/*
139	* This flag will be set if nbd_queue_rq() succeed, and will be checked and
140	* cleared in completion. Both setting and clearing of the flag are protected
141	* by cmd->lock.
142	*/
143	#define NBD_CMD_INFLIGHT 2
144
145	/* Just part of request header or data payload is sent successfully */
146	#define NBD_CMD_PARTIAL_SEND 3
147
148	struct nbd_cmd {
149	struct nbd_device *nbd;
150	struct mutex lock;
151	int index;
152	int cookie;
153	int retries;
154	blk_status_t status;
155	unsigned long flags;
156	u32 cmd_cookie;
157	};
158
159	#if IS_ENABLED(CONFIG_DEBUG_FS)
160	static struct dentry *nbd_dbg_dir;
161	#endif
162
163	#define nbd_name(nbd) ((nbd)->disk->disk_name)
164
165	#define NBD_DEF_BLKSIZE_BITS 10
166
167	static unsigned int nbds_max = 16;
168	static int max_part = 16;
169	static int part_shift;
170
171	static int nbd_dev_dbg_init(struct nbd_device *nbd);
172	static void nbd_dev_dbg_close(struct nbd_device *nbd);
173	static void nbd_config_put(struct nbd_device *nbd);
174	static void nbd_connect_reply(struct genl_info *info, int index);
175	static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
176	static void nbd_dead_link_work(struct work_struct *work);
177	static void nbd_disconnect_and_put(struct nbd_device *nbd);
178
179	static inline struct device *nbd_to_dev(struct nbd_device *nbd)
180	{
181	return disk_to_dev(nbd->disk);
182	}
183
184	static void nbd_requeue_cmd(struct nbd_cmd *cmd)
185	{
186	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
187
188	lockdep_assert_held(&cmd->lock);
189
190	/*
191	* Clear INFLIGHT flag so that this cmd won't be completed in
192	* normal completion path
193	*
194	* INFLIGHT flag will be set when the cmd is queued to nbd next
195	* time.
196	*/
197	__clear_bit(NBD_CMD_INFLIGHT, &cmd->flags);
198
199	if (!test_and_set_bit(NBD_CMD_REQUEUED, addr: &cmd->flags))
200	blk_mq_requeue_request(rq: req, kick_requeue_list: true);
201	}
202
203	#define NBD_COOKIE_BITS 32
204
205	static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
206	{
207	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
208	u32 tag = blk_mq_unique_tag(rq: req);
209	u64 cookie = cmd->cmd_cookie;
210
211	return (cookie << NBD_COOKIE_BITS) | tag;
212	}
213
214	static u32 nbd_handle_to_tag(u64 handle)
215	{
216	return (u32)handle;
217	}
218
219	static u32 nbd_handle_to_cookie(u64 handle)
220	{
221	return (u32)(handle >> NBD_COOKIE_BITS);
222	}
223
224	static const char *nbdcmd_to_ascii(int cmd)
225	{
226	switch (cmd) {
227	case NBD_CMD_READ: return "read";
228	case NBD_CMD_WRITE: return "write";
229	case NBD_CMD_DISC: return "disconnect";
230	case NBD_CMD_FLUSH: return "flush";
231	case NBD_CMD_TRIM: return "trim/discard";
232	}
233	return "invalid";
234	}
235
236	static ssize_t pid_show(struct device *dev,
237	struct device_attribute *attr, char *buf)
238	{
239	struct gendisk *disk = dev_to_disk(dev);
240	struct nbd_device *nbd = disk->private_data;
241
242	return sprintf(buf, fmt: "%d\n", nbd->pid);
243	}
244
245	static const struct device_attribute pid_attr = {
246	.attr = { .name = "pid", .mode = 0444},
247	.show = pid_show,
248	};
249
250	static ssize_t backend_show(struct device *dev,
251	struct device_attribute *attr, char *buf)
252	{
253	struct gendisk *disk = dev_to_disk(dev);
254	struct nbd_device *nbd = disk->private_data;
255
256	return sprintf(buf, fmt: "%s\n", nbd->backend ?: "");
257	}
258
259	static const struct device_attribute backend_attr = {
260	.attr = { .name = "backend", .mode = 0444},
261	.show = backend_show,
262	};
263
264	static void nbd_dev_remove(struct nbd_device *nbd)
265	{
266	struct gendisk *disk = nbd->disk;
267
268	del_gendisk(gp: disk);
269	blk_mq_free_tag_set(set: &nbd->tag_set);
270
271	/*
272	* Remove from idr after del_gendisk() completes, so if the same ID is
273	* reused, the following add_disk() will succeed.
274	*/
275	mutex_lock(&nbd_index_mutex);
276	idr_remove(&nbd_index_idr, id: nbd->index);
277	mutex_unlock(lock: &nbd_index_mutex);
278	destroy_workqueue(wq: nbd->recv_workq);
279	put_disk(disk);
280	}
281
282	static void nbd_dev_remove_work(struct work_struct *work)
283	{
284	nbd_dev_remove(container_of(work, struct nbd_device, remove_work));
285	}
286
287	static void nbd_put(struct nbd_device *nbd)
288	{
289	if (!refcount_dec_and_test(r: &nbd->refs))
290	return;
291
292	/* Call del_gendisk() asynchrounously to prevent deadlock */
293	if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags))
294	queue_work(wq: nbd_del_wq, work: &nbd->remove_work);
295	else
296	nbd_dev_remove(nbd);
297	}
298
299	static int nbd_disconnected(struct nbd_config *config)
300	{
301	return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
302	test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
303	}
304
305	static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
306	int notify)
307	{
308	if (!nsock->dead && notify && !nbd_disconnected(config: nbd->config)) {
309	struct link_dead_args *args;
310	args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
311	if (args) {
312	INIT_WORK(&args->work, nbd_dead_link_work);
313	args->index = nbd->index;
314	queue_work(wq: system_wq, work: &args->work);
315	}
316	}
317	if (!nsock->dead) {
318	kernel_sock_shutdown(sock: nsock->sock, how: SHUT_RDWR);
319	if (atomic_dec_return(v: &nbd->config->live_connections) == 0) {
320	if (test_and_clear_bit(NBD_RT_DISCONNECT_REQUESTED,
321	addr: &nbd->config->runtime_flags)) {
322	set_bit(NBD_RT_DISCONNECTED,
323	addr: &nbd->config->runtime_flags);
324	dev_info(nbd_to_dev(nbd),
325	"Disconnected due to user request.\n");
326	}
327	}
328	}
329	nsock->dead = true;
330	nsock->pending = NULL;
331	nsock->sent = 0;
332	}
333
334	static int nbd_set_size(struct nbd_device *nbd, loff_t bytesize, loff_t blksize)
335	{
336	struct queue_limits lim;
337	int error;
338
339	if (!blksize)
340	blksize = 1u << NBD_DEF_BLKSIZE_BITS;
341
342	if (blk_validate_block_size(bsize: blksize))
343	return -EINVAL;
344
345	if (bytesize < 0)
346	return -EINVAL;
347
348	nbd->config->bytesize = bytesize;
349	nbd->config->blksize_bits = __ffs(blksize);
350
351	if (!nbd->pid)
352	return 0;
353
354	lim = queue_limits_start_update(q: nbd->disk->queue);
355	if (nbd->config->flags & NBD_FLAG_SEND_TRIM)
356	lim.max_hw_discard_sectors = UINT_MAX >> SECTOR_SHIFT;
357	else
358	lim.max_hw_discard_sectors = 0;
359	if (!(nbd->config->flags & NBD_FLAG_SEND_FLUSH)) {
360	lim.features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA);
361	} else if (nbd->config->flags & NBD_FLAG_SEND_FUA) {
362	lim.features |= BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA;
363	} else {
364	lim.features |= BLK_FEAT_WRITE_CACHE;
365	lim.features &= ~BLK_FEAT_FUA;
366	}
367	if (nbd->config->flags & NBD_FLAG_ROTATIONAL)
368	lim.features |= BLK_FEAT_ROTATIONAL;
369	if (nbd->config->flags & NBD_FLAG_SEND_WRITE_ZEROES)
370	lim.max_write_zeroes_sectors = UINT_MAX >> SECTOR_SHIFT;
371
372	lim.logical_block_size = blksize;
373	lim.physical_block_size = blksize;
374	error = queue_limits_commit_update_frozen(q: nbd->disk->queue, lim: &lim);
375	if (error)
376	return error;
377
378	if (max_part)
379	set_bit(GD_NEED_PART_SCAN, addr: &nbd->disk->state);
380	if (!set_capacity_and_notify(disk: nbd->disk, size: bytesize >> 9))
381	kobject_uevent(kobj: &nbd_to_dev(nbd)->kobj, action: KOBJ_CHANGE);
382	return 0;
383	}
384
385	static void nbd_complete_rq(struct request *req)
386	{
387	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: req);
388
389	dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
390	cmd->status ? "failed" : "done");
391
392	blk_mq_end_request(rq: req, error: cmd->status);
393	}
394
395	/*
396	* Forcibly shutdown the socket causing all listeners to error
397	*/
398	static void sock_shutdown(struct nbd_device *nbd)
399	{
400	struct nbd_config *config = nbd->config;
401	int i;
402
403	if (config->num_connections == 0)
404	return;
405	if (test_and_set_bit(NBD_RT_DISCONNECTED, addr: &config->runtime_flags))
406	return;
407
408	for (i = 0; i < config->num_connections; i++) {
409	struct nbd_sock *nsock = config->socks[i];
410	mutex_lock(&nsock->tx_lock);
411	nbd_mark_nsock_dead(nbd, nsock, notify: 0);
412	mutex_unlock(lock: &nsock->tx_lock);
413	}
414	dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
415	}
416
417	static u32 req_to_nbd_cmd_type(struct request *req)
418	{
419	switch (req_op(req)) {
420	case REQ_OP_DISCARD:
421	return NBD_CMD_TRIM;
422	case REQ_OP_FLUSH:
423	return NBD_CMD_FLUSH;
424	case REQ_OP_WRITE:
425	return NBD_CMD_WRITE;
426	case REQ_OP_READ:
427	return NBD_CMD_READ;
428	case REQ_OP_WRITE_ZEROES:
429	return NBD_CMD_WRITE_ZEROES;
430	default:
431	return U32_MAX;
432	}
433	}
434
435	static struct nbd_config *nbd_get_config_unlocked(struct nbd_device *nbd)
436	{
437	if (refcount_inc_not_zero(r: &nbd->config_refs)) {
438	/*
439	* Add smp_mb__after_atomic to ensure that reading nbd->config_refs
440	* and reading nbd->config is ordered. The pair is the barrier in
441	* nbd_alloc_and_init_config(), avoid nbd->config_refs is set
442	* before nbd->config.
443	*/
444	smp_mb__after_atomic();
445	return nbd->config;
446	}
447
448	return NULL;
449	}
450
451	static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req)
452	{
453	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: req);
454	struct nbd_device *nbd = cmd->nbd;
455	struct nbd_config *config;
456
457	if (!mutex_trylock(&cmd->lock))
458	return BLK_EH_RESET_TIMER;
459
460	/* partial send is handled in nbd_sock's work function */
461	if (test_bit(NBD_CMD_PARTIAL_SEND, &cmd->flags)) {
462	mutex_unlock(lock: &cmd->lock);
463	return BLK_EH_RESET_TIMER;
464	}
465
466	if (!test_bit(NBD_CMD_INFLIGHT, &cmd->flags)) {
467	mutex_unlock(lock: &cmd->lock);
468	return BLK_EH_DONE;
469	}
470
471	config = nbd_get_config_unlocked(nbd);
472	if (!config) {
473	cmd->status = BLK_STS_TIMEOUT;
474	__clear_bit(NBD_CMD_INFLIGHT, &cmd->flags);
475	mutex_unlock(lock: &cmd->lock);
476	goto done;
477	}
478
479	if (config->num_connections > 1 ||
480	(config->num_connections == 1 && nbd->tag_set.timeout)) {
481	dev_err_ratelimited(nbd_to_dev(nbd),
482	"Connection timed out, retrying (%d/%d alive)\n",
483	atomic_read(&config->live_connections),
484	config->num_connections);
485	/*
486	* Hooray we have more connections, requeue this IO, the submit
487	* path will put it on a real connection. Or if only one
488	* connection is configured, the submit path will wait util
489	* a new connection is reconfigured or util dead timeout.
490	*/
491	if (config->socks) {
492	if (cmd->index < config->num_connections) {
493	struct nbd_sock *nsock =
494	config->socks[cmd->index];
495	mutex_lock(&nsock->tx_lock);
496	/* We can have multiple outstanding requests, so
497	* we don't want to mark the nsock dead if we've
498	* already reconnected with a new socket, so
499	* only mark it dead if its the same socket we
500	* were sent out on.
501	*/
502	if (cmd->cookie == nsock->cookie)
503	nbd_mark_nsock_dead(nbd, nsock, notify: 1);
504	mutex_unlock(lock: &nsock->tx_lock);
505	}
506	nbd_requeue_cmd(cmd);
507	mutex_unlock(lock: &cmd->lock);
508	nbd_config_put(nbd);
509	return BLK_EH_DONE;
510	}
511	}
512
513	if (!nbd->tag_set.timeout) {
514	/*
515	* Userspace sets timeout=0 to disable socket disconnection,
516	* so just warn and reset the timer.
517	*/
518	struct nbd_sock *nsock = config->socks[cmd->index];
519	cmd->retries++;
520	dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
521	req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
522	(unsigned long long)blk_rq_pos(req) << 9,
523	blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
524
525	mutex_lock(&nsock->tx_lock);
526	if (cmd->cookie != nsock->cookie) {
527	nbd_requeue_cmd(cmd);
528	mutex_unlock(lock: &nsock->tx_lock);
529	mutex_unlock(lock: &cmd->lock);
530	nbd_config_put(nbd);
531	return BLK_EH_DONE;
532	}
533	mutex_unlock(lock: &nsock->tx_lock);
534	mutex_unlock(lock: &cmd->lock);
535	nbd_config_put(nbd);
536	return BLK_EH_RESET_TIMER;
537	}
538
539	dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
540	set_bit(NBD_RT_TIMEDOUT, addr: &config->runtime_flags);
541	cmd->status = BLK_STS_IOERR;
542	__clear_bit(NBD_CMD_INFLIGHT, &cmd->flags);
543	mutex_unlock(lock: &cmd->lock);
544	sock_shutdown(nbd);
545	nbd_config_put(nbd);
546	done:
547	blk_mq_complete_request(rq: req);
548	return BLK_EH_DONE;
549	}
550
551	static int __sock_xmit(struct nbd_device *nbd, struct socket *sock, int send,
552	struct iov_iter *iter, int msg_flags, int *sent)
553	{
554	int result;
555	struct msghdr msg = {} ;
556	unsigned int noreclaim_flag;
557
558	if (unlikely(!sock)) {
559	dev_err_ratelimited(disk_to_dev(nbd->disk),
560	"Attempted %s on closed socket in sock_xmit\n",
561	(send ? "send" : "recv"));
562	return -EINVAL;
563	}
564
565	msg.msg_iter = *iter;
566
567	noreclaim_flag = memalloc_noreclaim_save();
568	do {
569	sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
570	sock->sk->sk_use_task_frag = false;
571	msg.msg_flags = msg_flags | MSG_NOSIGNAL;
572
573	if (send)
574	result = sock_sendmsg(sock, msg: &msg);
575	else
576	result = sock_recvmsg(sock, msg: &msg, flags: msg.msg_flags);
577
578	if (result <= 0) {
579	if (result == 0)
580	result = -EPIPE; /* short read */
581	break;
582	}
583	if (sent)
584	*sent += result;
585	} while (msg_data_left(msg: &msg));
586
587	memalloc_noreclaim_restore(flags: noreclaim_flag);
588
589	return result;
590	}
591
592	/*
593	* Send or receive packet. Return a positive value on success and
594	* negtive value on failure, and never return 0.
595	*/
596	static int sock_xmit(struct nbd_device *nbd, int index, int send,
597	struct iov_iter *iter, int msg_flags, int *sent)
598	{
599	struct nbd_config *config = nbd->config;
600	struct socket *sock = config->socks[index]->sock;
601
602	return __sock_xmit(nbd, sock, send, iter, msg_flags, sent);
603	}
604
605	/*
606	* Different settings for sk->sk_sndtimeo can result in different return values
607	* if there is a signal pending when we enter sendmsg, because reasons?
608	*/
609	static inline int was_interrupted(int result)
610	{
611	return result == -ERESTARTSYS || result == -EINTR;
612	}
613
614	/*
615	* We've already sent header or part of data payload, have no choice but
616	* to set pending and schedule it in work.
617	*
618	* And we have to return BLK_STS_OK to block core, otherwise this same
619	* request may be re-dispatched with different tag, but our header has
620	* been sent out with old tag, and this way does confuse reply handling.
621	*/
622	static void nbd_sched_pending_work(struct nbd_device *nbd,
623	struct nbd_sock *nsock,
624	struct nbd_cmd *cmd, int sent)
625	{
626	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
627
628	/* pending work should be scheduled only once */
629	WARN_ON_ONCE(test_bit(NBD_CMD_PARTIAL_SEND, &cmd->flags));
630
631	nsock->pending = req;
632	nsock->sent = sent;
633	set_bit(NBD_CMD_PARTIAL_SEND, addr: &cmd->flags);
634	refcount_inc(r: &nbd->config_refs);
635	schedule_work(work: &nsock->work);
636	}
637
638	/*
639	* Returns BLK_STS_RESOURCE if the caller should retry after a delay.
640	* Returns BLK_STS_IOERR if sending failed.
641	*/
642	static blk_status_t nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd,
643	int index)
644	{
645	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
646	struct nbd_config *config = nbd->config;
647	struct nbd_sock *nsock = config->socks[index];
648	int result;
649	struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
650	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
651	struct iov_iter from;
652	struct bio *bio;
653	u64 handle;
654	u32 type;
655	u32 nbd_cmd_flags = 0;
656	int sent = nsock->sent, skip = 0;
657
658	lockdep_assert_held(&cmd->lock);
659	lockdep_assert_held(&nsock->tx_lock);
660
661	iov_iter_kvec(i: &from, ITER_SOURCE, kvec: &iov, nr_segs: 1, count: sizeof(request));
662
663	type = req_to_nbd_cmd_type(req);
664	if (type == U32_MAX)
665	return BLK_STS_IOERR;
666
667	if (rq_data_dir(req) == WRITE &&
668	(config->flags & NBD_FLAG_READ_ONLY)) {
669	dev_err_ratelimited(disk_to_dev(nbd->disk),
670	"Write on read-only\n");
671	return BLK_STS_IOERR;
672	}
673
674	if (req->cmd_flags & REQ_FUA)
675	nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
676	if ((req->cmd_flags & REQ_NOUNMAP) && (type == NBD_CMD_WRITE_ZEROES))
677	nbd_cmd_flags |= NBD_CMD_FLAG_NO_HOLE;
678
679	/* We did a partial send previously, and we at least sent the whole
680	* request struct, so just go and send the rest of the pages in the
681	* request.
682	*/
683	if (sent) {
684	if (sent >= sizeof(request)) {
685	skip = sent - sizeof(request);
686
687	/* initialize handle for tracing purposes */
688	handle = nbd_cmd_handle(cmd);
689
690	goto send_pages;
691	}
692	iov_iter_advance(i: &from, bytes: sent);
693	} else {
694	cmd->cmd_cookie++;
695	}
696	cmd->index = index;
697	cmd->cookie = nsock->cookie;
698	cmd->retries = 0;
699	request.type = htonl(type | nbd_cmd_flags);
700	if (type != NBD_CMD_FLUSH) {
701	request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
702	request.len = htonl(blk_rq_bytes(req));
703	}
704	handle = nbd_cmd_handle(cmd);
705	request.cookie = cpu_to_be64(handle);
706
707	trace_nbd_send_request(nbd_request: &request, index: nbd->index, rq: blk_mq_rq_from_pdu(pdu: cmd));
708
709	dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
710	req, nbdcmd_to_ascii(type),
711	(unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
712	result = sock_xmit(nbd, index, send: 1, iter: &from,
713	msg_flags: (type == NBD_CMD_WRITE) ? MSG_MORE : 0, sent: &sent);
714	trace_nbd_header_sent(req, handle);
715	if (result < 0) {
716	if (was_interrupted(result)) {
717	/* If we haven't sent anything we can just return BUSY,
718	* however if we have sent something we need to make
719	* sure we only allow this req to be sent until we are
720	* completely done.
721	*/
722	if (sent) {
723	nbd_sched_pending_work(nbd, nsock, cmd, sent);
724	return BLK_STS_OK;
725	}
726	set_bit(NBD_CMD_REQUEUED, addr: &cmd->flags);
727	return BLK_STS_RESOURCE;
728	}
729	dev_err_ratelimited(disk_to_dev(nbd->disk),
730	"Send control failed (result %d)\n", result);
731	goto requeue;
732	}
733	send_pages:
734	if (type != NBD_CMD_WRITE)
735	goto out;
736
737	bio = req->bio;
738	while (bio) {
739	struct bio *next = bio->bi_next;
740	struct bvec_iter iter;
741	struct bio_vec bvec;
742
743	bio_for_each_segment(bvec, bio, iter) {
744	bool is_last = !next && bio_iter_last(bvec, iter);
745	int flags = is_last ? 0 : MSG_MORE;
746
747	dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
748	req, bvec.bv_len);
749	iov_iter_bvec(i: &from, ITER_SOURCE, bvec: &bvec, nr_segs: 1, count: bvec.bv_len);
750	if (skip) {
751	if (skip >= iov_iter_count(i: &from)) {
752	skip -= iov_iter_count(i: &from);
753	continue;
754	}
755	iov_iter_advance(i: &from, bytes: skip);
756	skip = 0;
757	}
758	result = sock_xmit(nbd, index, send: 1, iter: &from, msg_flags: flags, sent: &sent);
759	if (result < 0) {
760	if (was_interrupted(result)) {
761	nbd_sched_pending_work(nbd, nsock, cmd, sent);
762	return BLK_STS_OK;
763	}
764	dev_err(disk_to_dev(nbd->disk),
765	"Send data failed (result %d)\n",
766	result);
767	goto requeue;
768	}
769	/*
770	* The completion might already have come in,
771	* so break for the last one instead of letting
772	* the iterator do it. This prevents use-after-free
773	* of the bio.
774	*/
775	if (is_last)
776	break;
777	}
778	bio = next;
779	}
780	out:
781	trace_nbd_payload_sent(req, handle);
782	nsock->pending = NULL;
783	nsock->sent = 0;
784	__set_bit(NBD_CMD_INFLIGHT, &cmd->flags);
785	return BLK_STS_OK;
786
787	requeue:
788	/*
789	* Can't requeue in case we are dealing with partial send
790	*
791	* We must run from pending work function.
792	* */
793	if (test_bit(NBD_CMD_PARTIAL_SEND, &cmd->flags))
794	return BLK_STS_OK;
795
796	/* retry on a different socket */
797	dev_err_ratelimited(disk_to_dev(nbd->disk),
798	"Request send failed, requeueing\n");
799	nbd_mark_nsock_dead(nbd, nsock, notify: 1);
800	nbd_requeue_cmd(cmd);
801	return BLK_STS_OK;
802	}
803
804	/* handle partial sending */
805	static void nbd_pending_cmd_work(struct work_struct *work)
806	{
807	struct nbd_sock *nsock = container_of(work, struct nbd_sock, work);
808	struct request *req = nsock->pending;
809	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: req);
810	struct nbd_device *nbd = cmd->nbd;
811	unsigned long deadline = READ_ONCE(req->deadline);
812	unsigned int wait_ms = 2;
813
814	mutex_lock(&cmd->lock);
815
816	WARN_ON_ONCE(test_bit(NBD_CMD_REQUEUED, &cmd->flags));
817	if (WARN_ON_ONCE(!test_bit(NBD_CMD_PARTIAL_SEND, &cmd->flags)))
818	goto out;
819
820	mutex_lock(&nsock->tx_lock);
821	while (true) {
822	nbd_send_cmd(nbd, cmd, index: cmd->index);
823	if (!nsock->pending)
824	break;
825
826	/* don't bother timeout handler for partial sending */
827	if (READ_ONCE(jiffies) + msecs_to_jiffies(m: wait_ms) >= deadline) {
828	cmd->status = BLK_STS_IOERR;
829	blk_mq_complete_request(rq: req);
830	break;
831	}
832	msleep(msecs: wait_ms);
833	wait_ms *= 2;
834	}
835	mutex_unlock(lock: &nsock->tx_lock);
836	clear_bit(NBD_CMD_PARTIAL_SEND, addr: &cmd->flags);
837	out:
838	mutex_unlock(lock: &cmd->lock);
839	nbd_config_put(nbd);
840	}
841
842	static int nbd_read_reply(struct nbd_device *nbd, struct socket *sock,
843	struct nbd_reply *reply)
844	{
845	struct kvec iov = {.iov_base = reply, .iov_len = sizeof(*reply)};
846	struct iov_iter to;
847	int result;
848
849	reply->magic = 0;
850	iov_iter_kvec(i: &to, ITER_DEST, kvec: &iov, nr_segs: 1, count: sizeof(*reply));
851	result = __sock_xmit(nbd, sock, send: 0, iter: &to, MSG_WAITALL, NULL);
852	if (result < 0) {
853	if (!nbd_disconnected(config: nbd->config))
854	dev_err(disk_to_dev(nbd->disk),
855	"Receive control failed (result %d)\n", result);
856	return result;
857	}
858
859	if (ntohl(reply->magic) != NBD_REPLY_MAGIC) {
860	dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
861	(unsigned long)ntohl(reply->magic));
862	return -EPROTO;
863	}
864
865	return 0;
866	}
867
868	/* NULL returned = something went wrong, inform userspace */
869	static struct nbd_cmd *nbd_handle_reply(struct nbd_device *nbd, int index,
870	struct nbd_reply *reply)
871	{
872	int result;
873	struct nbd_cmd *cmd;
874	struct request *req = NULL;
875	u64 handle;
876	u16 hwq;
877	u32 tag;
878	int ret = 0;
879
880	handle = be64_to_cpu(reply->cookie);
881	tag = nbd_handle_to_tag(handle);
882	hwq = blk_mq_unique_tag_to_hwq(unique_tag: tag);
883	if (hwq < nbd->tag_set.nr_hw_queues)
884	req = blk_mq_tag_to_rq(tags: nbd->tag_set.tags[hwq],
885	tag: blk_mq_unique_tag_to_tag(unique_tag: tag));
886	if (!req || !blk_mq_request_started(rq: req)) {
887	dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
888	tag, req);
889	return ERR_PTR(error: -ENOENT);
890	}
891	trace_nbd_header_received(req, handle);
892	cmd = blk_mq_rq_to_pdu(rq: req);
893
894	mutex_lock(&cmd->lock);
895	if (!test_bit(NBD_CMD_INFLIGHT, &cmd->flags)) {
896	dev_err(disk_to_dev(nbd->disk), "Suspicious reply %d (status %u flags %lu)",
897	tag, cmd->status, cmd->flags);
898	ret = -ENOENT;
899	goto out;
900	}
901	if (cmd->index != index) {
902	dev_err(disk_to_dev(nbd->disk), "Unexpected reply %d from different sock %d (expected %d)",
903	tag, index, cmd->index);
904	ret = -ENOENT;
905	goto out;
906	}
907	if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
908	dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
909	req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
910	ret = -ENOENT;
911	goto out;
912	}
913	if (cmd->status != BLK_STS_OK) {
914	dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
915	req);
916	ret = -ENOENT;
917	goto out;
918	}
919	if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
920	dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
921	req);
922	ret = -ENOENT;
923	goto out;
924	}
925	if (ntohl(reply->error)) {
926	dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
927	ntohl(reply->error));
928	cmd->status = BLK_STS_IOERR;
929	goto out;
930	}
931
932	dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
933	if (rq_data_dir(req) != WRITE) {
934	struct req_iterator iter;
935	struct bio_vec bvec;
936	struct iov_iter to;
937
938	rq_for_each_segment(bvec, req, iter) {
939	iov_iter_bvec(i: &to, ITER_DEST, bvec: &bvec, nr_segs: 1, count: bvec.bv_len);
940	result = sock_xmit(nbd, index, send: 0, iter: &to, MSG_WAITALL, NULL);
941	if (result < 0) {
942	dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
943	result);
944	/*
945	* If we've disconnected, we need to make sure we
946	* complete this request, otherwise error out
947	* and let the timeout stuff handle resubmitting
948	* this request onto another connection.
949	*/
950	if (nbd_disconnected(config: nbd->config)) {
951	cmd->status = BLK_STS_IOERR;
952	goto out;
953	}
954	ret = -EIO;
955	goto out;
956	}
957	dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
958	req, bvec.bv_len);
959	}
960	}
961	out:
962	trace_nbd_payload_received(req, handle);
963	mutex_unlock(lock: &cmd->lock);
964	return ret ? ERR_PTR(error: ret) : cmd;
965	}
966
967	static void recv_work(struct work_struct *work)
968	{
969	struct recv_thread_args *args = container_of(work,
970	struct recv_thread_args,
971	work);
972	struct nbd_device *nbd = args->nbd;
973	struct nbd_config *config = nbd->config;
974	struct request_queue *q = nbd->disk->queue;
975	struct nbd_sock *nsock = args->nsock;
976	struct nbd_cmd *cmd;
977	struct request *rq;
978
979	while (1) {
980	struct nbd_reply reply;
981
982	if (nbd_read_reply(nbd, sock: nsock->sock, reply: &reply))
983	break;
984
985	/*
986	* Grab .q_usage_counter so request pool won't go away, then no
987	* request use-after-free is possible during nbd_handle_reply().
988	* If queue is frozen, there won't be any inflight requests, we
989	* needn't to handle the incoming garbage message.
990	*/
991	if (!percpu_ref_tryget(ref: &q->q_usage_counter)) {
992	dev_err(disk_to_dev(nbd->disk), "%s: no io inflight\n",
993	__func__);
994	break;
995	}
996
997	cmd = nbd_handle_reply(nbd, index: args->index, reply: &reply);
998	if (IS_ERR(ptr: cmd)) {
999	percpu_ref_put(ref: &q->q_usage_counter);
1000	break;
1001	}
1002
1003	rq = blk_mq_rq_from_pdu(pdu: cmd);
1004	if (likely(!blk_should_fake_timeout(rq->q))) {
1005	bool complete;
1006
1007	mutex_lock(&cmd->lock);
1008	complete = __test_and_clear_bit(NBD_CMD_INFLIGHT,
1009	&cmd->flags);
1010	mutex_unlock(lock: &cmd->lock);
1011	if (complete)
1012	blk_mq_complete_request(rq);
1013	}
1014	percpu_ref_put(ref: &q->q_usage_counter);
1015	}
1016
1017	mutex_lock(&nsock->tx_lock);
1018	nbd_mark_nsock_dead(nbd, nsock, notify: 1);
1019	mutex_unlock(lock: &nsock->tx_lock);
1020
1021	nbd_config_put(nbd);
1022	atomic_dec(v: &config->recv_threads);
1023	wake_up(&config->recv_wq);
1024	kfree(objp: args);
1025	}
1026
1027	static bool nbd_clear_req(struct request *req, void *data)
1028	{
1029	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: req);
1030
1031	/* don't abort one completed request */
1032	if (blk_mq_request_completed(rq: req))
1033	return true;
1034
1035	mutex_lock(&cmd->lock);
1036	if (!__test_and_clear_bit(NBD_CMD_INFLIGHT, &cmd->flags)) {
1037	mutex_unlock(lock: &cmd->lock);
1038	return true;
1039	}
1040	cmd->status = BLK_STS_IOERR;
1041	mutex_unlock(lock: &cmd->lock);
1042
1043	blk_mq_complete_request(rq: req);
1044	return true;
1045	}
1046
1047	static void nbd_clear_que(struct nbd_device *nbd)
1048	{
1049	blk_mq_quiesce_queue(q: nbd->disk->queue);
1050	blk_mq_tagset_busy_iter(tagset: &nbd->tag_set, fn: nbd_clear_req, NULL);
1051	blk_mq_unquiesce_queue(q: nbd->disk->queue);
1052	dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
1053	}
1054
1055	static int find_fallback(struct nbd_device *nbd, int index)
1056	{
1057	struct nbd_config *config = nbd->config;
1058	int new_index = -1;
1059	struct nbd_sock *nsock = config->socks[index];
1060	int fallback = nsock->fallback_index;
1061
1062	if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
1063	return new_index;
1064
1065	if (config->num_connections <= 1) {
1066	dev_err_ratelimited(disk_to_dev(nbd->disk),
1067	"Dead connection, failed to find a fallback\n");
1068	return new_index;
1069	}
1070
1071	if (fallback >= 0 && fallback < config->num_connections &&
1072	!config->socks[fallback]->dead)
1073	return fallback;
1074
1075	if (nsock->fallback_index < 0 ||
1076	nsock->fallback_index >= config->num_connections ||
1077	config->socks[nsock->fallback_index]->dead) {
1078	int i;
1079	for (i = 0; i < config->num_connections; i++) {
1080	if (i == index)
1081	continue;
1082	if (!config->socks[i]->dead) {
1083	new_index = i;
1084	break;
1085	}
1086	}
1087	nsock->fallback_index = new_index;
1088	if (new_index < 0) {
1089	dev_err_ratelimited(disk_to_dev(nbd->disk),
1090	"Dead connection, failed to find a fallback\n");
1091	return new_index;
1092	}
1093	}
1094	new_index = nsock->fallback_index;
1095	return new_index;
1096	}
1097
1098	static int wait_for_reconnect(struct nbd_device *nbd)
1099	{
1100	struct nbd_config *config = nbd->config;
1101	if (!config->dead_conn_timeout)
1102	return 0;
1103
1104	if (!wait_event_timeout(config->conn_wait,
1105	test_bit(NBD_RT_DISCONNECTED,
1106	&config->runtime_flags) ||
1107	atomic_read(&config->live_connections) > 0,
1108	config->dead_conn_timeout))
1109	return 0;
1110
1111	return !test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1112	}
1113
1114	static blk_status_t nbd_handle_cmd(struct nbd_cmd *cmd, int index)
1115	{
1116	struct request *req = blk_mq_rq_from_pdu(pdu: cmd);
1117	struct nbd_device *nbd = cmd->nbd;
1118	struct nbd_config *config;
1119	struct nbd_sock *nsock;
1120	blk_status_t ret;
1121
1122	lockdep_assert_held(&cmd->lock);
1123
1124	config = nbd_get_config_unlocked(nbd);
1125	if (!config) {
1126	dev_err_ratelimited(disk_to_dev(nbd->disk),
1127	"Socks array is empty\n");
1128	return BLK_STS_IOERR;
1129	}
1130
1131	if (index >= config->num_connections) {
1132	dev_err_ratelimited(disk_to_dev(nbd->disk),
1133	"Attempted send on invalid socket\n");
1134	nbd_config_put(nbd);
1135	return BLK_STS_IOERR;
1136	}
1137	cmd->status = BLK_STS_OK;
1138	again:
1139	nsock = config->socks[index];
1140	mutex_lock(&nsock->tx_lock);
1141	if (nsock->dead) {
1142	int old_index = index;
1143	index = find_fallback(nbd, index);
1144	mutex_unlock(lock: &nsock->tx_lock);
1145	if (index < 0) {
1146	if (wait_for_reconnect(nbd)) {
1147	index = old_index;
1148	goto again;
1149	}
1150	/* All the sockets should already be down at this point,
1151	* we just want to make sure that DISCONNECTED is set so
1152	* any requests that come in that were queue'ed waiting
1153	* for the reconnect timer don't trigger the timer again
1154	* and instead just error out.
1155	*/
1156	sock_shutdown(nbd);
1157	nbd_config_put(nbd);
1158	return BLK_STS_IOERR;
1159	}
1160	goto again;
1161	}
1162
1163	/* Handle the case that we have a pending request that was partially
1164	* transmitted that _has_ to be serviced first. We need to call requeue
1165	* here so that it gets put _after_ the request that is already on the
1166	* dispatch list.
1167	*/
1168	blk_mq_start_request(rq: req);
1169	if (unlikely(nsock->pending && nsock->pending != req)) {
1170	nbd_requeue_cmd(cmd);
1171	ret = BLK_STS_OK;
1172	goto out;
1173	}
1174	ret = nbd_send_cmd(nbd, cmd, index);
1175	out:
1176	mutex_unlock(lock: &nsock->tx_lock);
1177	nbd_config_put(nbd);
1178	return ret;
1179	}
1180
1181	static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
1182	const struct blk_mq_queue_data *bd)
1183	{
1184	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq: bd->rq);
1185	blk_status_t ret;
1186
1187	/*
1188	* Since we look at the bio's to send the request over the network we
1189	* need to make sure the completion work doesn't mark this request done
1190	* before we are done doing our send. This keeps us from dereferencing
1191	* freed data if we have particularly fast completions (ie we get the
1192	* completion before we exit sock_xmit on the last bvec) or in the case
1193	* that the server is misbehaving (or there was an error) before we're
1194	* done sending everything over the wire.
1195	*/
1196	mutex_lock(&cmd->lock);
1197	clear_bit(NBD_CMD_REQUEUED, addr: &cmd->flags);
1198
1199	/* We can be called directly from the user space process, which means we
1200	* could possibly have signals pending so our sendmsg will fail. In
1201	* this case we need to return that we are busy, otherwise error out as
1202	* appropriate.
1203	*/
1204	ret = nbd_handle_cmd(cmd, index: hctx->queue_num);
1205	mutex_unlock(lock: &cmd->lock);
1206
1207	return ret;
1208	}
1209
1210	static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
1211	int *err)
1212	{
1213	struct socket *sock;
1214
1215	*err = 0;
1216	sock = sockfd_lookup(fd, err);
1217	if (!sock)
1218	return NULL;
1219
1220	if (sock->ops->shutdown == sock_no_shutdown) {
1221	dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
1222	*err = -EINVAL;
1223	sockfd_put(sock);
1224	return NULL;
1225	}
1226
1227	return sock;
1228	}
1229
1230	static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1231	bool netlink)
1232	{
1233	struct nbd_config *config = nbd->config;
1234	struct socket *sock;
1235	struct nbd_sock **socks;
1236	struct nbd_sock *nsock;
1237	unsigned int memflags;
1238	int err;
1239
1240	/* Arg will be cast to int, check it to avoid overflow */
1241	if (arg > INT_MAX)
1242	return -EINVAL;
1243	sock = nbd_get_socket(nbd, fd: arg, err: &err);
1244	if (!sock)
1245	return err;
1246
1247	/*
1248	* We need to make sure we don't get any errant requests while we're
1249	* reallocating the ->socks array.
1250	*/
1251	memflags = blk_mq_freeze_queue(q: nbd->disk->queue);
1252
1253	if (!netlink && !nbd->task_setup &&
1254	!test_bit(NBD_RT_BOUND, &config->runtime_flags))
1255	nbd->task_setup = current;
1256
1257	if (!netlink &&
1258	(nbd->task_setup != current ||
1259	test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1260	dev_err(disk_to_dev(nbd->disk),
1261	"Device being setup by another task");
1262	err = -EBUSY;
1263	goto put_socket;
1264	}
1265
1266	nsock = kzalloc(sizeof(*nsock), GFP_KERNEL);
1267	if (!nsock) {
1268	err = -ENOMEM;
1269	goto put_socket;
1270	}
1271
1272	socks = krealloc(config->socks, (config->num_connections + 1) *
1273	sizeof(struct nbd_sock *), GFP_KERNEL);
1274	if (!socks) {
1275	kfree(objp: nsock);
1276	err = -ENOMEM;
1277	goto put_socket;
1278	}
1279
1280	config->socks = socks;
1281
1282	nsock->fallback_index = -1;
1283	nsock->dead = false;
1284	mutex_init(&nsock->tx_lock);
1285	nsock->sock = sock;
1286	nsock->pending = NULL;
1287	nsock->sent = 0;
1288	nsock->cookie = 0;
1289	INIT_WORK(&nsock->work, nbd_pending_cmd_work);
1290	socks[config->num_connections++] = nsock;
1291	atomic_inc(v: &config->live_connections);
1292	blk_mq_unfreeze_queue(q: nbd->disk->queue, memflags);
1293
1294	return 0;
1295
1296	put_socket:
1297	blk_mq_unfreeze_queue(q: nbd->disk->queue, memflags);
1298	sockfd_put(sock);
1299	return err;
1300	}
1301
1302	static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1303	{
1304	struct nbd_config *config = nbd->config;
1305	struct socket *sock, *old;
1306	struct recv_thread_args *args;
1307	int i;
1308	int err;
1309
1310	sock = nbd_get_socket(nbd, fd: arg, err: &err);
1311	if (!sock)
1312	return err;
1313
1314	args = kzalloc(sizeof(*args), GFP_KERNEL);
1315	if (!args) {
1316	sockfd_put(sock);
1317	return -ENOMEM;
1318	}
1319
1320	for (i = 0; i < config->num_connections; i++) {
1321	struct nbd_sock *nsock = config->socks[i];
1322
1323	if (!nsock->dead)
1324	continue;
1325
1326	mutex_lock(&nsock->tx_lock);
1327	if (!nsock->dead) {
1328	mutex_unlock(lock: &nsock->tx_lock);
1329	continue;
1330	}
1331	sk_set_memalloc(sk: sock->sk);
1332	if (nbd->tag_set.timeout)
1333	sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1334	atomic_inc(v: &config->recv_threads);
1335	refcount_inc(r: &nbd->config_refs);
1336	old = nsock->sock;
1337	nsock->fallback_index = -1;
1338	nsock->sock = sock;
1339	nsock->dead = false;
1340	INIT_WORK(&args->work, recv_work);
1341	args->index = i;
1342	args->nbd = nbd;
1343	args->nsock = nsock;
1344	nsock->cookie++;
1345	mutex_unlock(lock: &nsock->tx_lock);
1346	sockfd_put(old);
1347
1348	clear_bit(NBD_RT_DISCONNECTED, addr: &config->runtime_flags);
1349
1350	/* We take the tx_mutex in an error path in the recv_work, so we
1351	* need to queue_work outside of the tx_mutex.
1352	*/
1353	queue_work(wq: nbd->recv_workq, work: &args->work);
1354
1355	atomic_inc(v: &config->live_connections);
1356	wake_up(&config->conn_wait);
1357	return 0;
1358	}
1359	sockfd_put(sock);
1360	kfree(objp: args);
1361	return -ENOSPC;
1362	}
1363
1364	static void nbd_bdev_reset(struct nbd_device *nbd)
1365	{
1366	if (disk_openers(disk: nbd->disk) > 1)
1367	return;
1368	set_capacity(disk: nbd->disk, size: 0);
1369	}
1370
1371	static void nbd_parse_flags(struct nbd_device *nbd)
1372	{
1373	if (nbd->config->flags & NBD_FLAG_READ_ONLY)
1374	set_disk_ro(disk: nbd->disk, read_only: true);
1375	else
1376	set_disk_ro(disk: nbd->disk, read_only: false);
1377	}
1378
1379	static void send_disconnects(struct nbd_device *nbd)
1380	{
1381	struct nbd_config *config = nbd->config;
1382	struct nbd_request request = {
1383	.magic = htonl(NBD_REQUEST_MAGIC),
1384	.type = htonl(NBD_CMD_DISC),
1385	};
1386	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1387	struct iov_iter from;
1388	int i, ret;
1389
1390	for (i = 0; i < config->num_connections; i++) {
1391	struct nbd_sock *nsock = config->socks[i];
1392
1393	iov_iter_kvec(i: &from, ITER_SOURCE, kvec: &iov, nr_segs: 1, count: sizeof(request));
1394	mutex_lock(&nsock->tx_lock);
1395	ret = sock_xmit(nbd, index: i, send: 1, iter: &from, msg_flags: 0, NULL);
1396	if (ret < 0)
1397	dev_err(disk_to_dev(nbd->disk),
1398	"Send disconnect failed %d\n", ret);
1399	mutex_unlock(lock: &nsock->tx_lock);
1400	}
1401	}
1402
1403	static int nbd_disconnect(struct nbd_device *nbd)
1404	{
1405	struct nbd_config *config = nbd->config;
1406
1407	dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1408	set_bit(NBD_RT_DISCONNECT_REQUESTED, addr: &config->runtime_flags);
1409	set_bit(NBD_DISCONNECT_REQUESTED, addr: &nbd->flags);
1410	send_disconnects(nbd);
1411	return 0;
1412	}
1413
1414	static void nbd_clear_sock(struct nbd_device *nbd)
1415	{
1416	sock_shutdown(nbd);
1417	nbd_clear_que(nbd);
1418	nbd->task_setup = NULL;
1419	}
1420
1421	static void nbd_config_put(struct nbd_device *nbd)
1422	{
1423	if (refcount_dec_and_mutex_lock(r: &nbd->config_refs,
1424	lock: &nbd->config_lock)) {
1425	struct nbd_config *config = nbd->config;
1426	nbd_dev_dbg_close(nbd);
1427	invalidate_disk(disk: nbd->disk);
1428	if (nbd->config->bytesize)
1429	kobject_uevent(kobj: &nbd_to_dev(nbd)->kobj, action: KOBJ_CHANGE);
1430	if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1431	addr: &config->runtime_flags))
1432	device_remove_file(disk_to_dev(nbd->disk), attr: &pid_attr);
1433	nbd->pid = 0;
1434	if (test_and_clear_bit(NBD_RT_HAS_BACKEND_FILE,
1435	addr: &config->runtime_flags)) {
1436	device_remove_file(disk_to_dev(nbd->disk), attr: &backend_attr);
1437	kfree(objp: nbd->backend);
1438	nbd->backend = NULL;
1439	}
1440	nbd_clear_sock(nbd);
1441	if (config->num_connections) {
1442	int i;
1443	for (i = 0; i < config->num_connections; i++) {
1444	sockfd_put(config->socks[i]->sock);
1445	kfree(objp: config->socks[i]);
1446	}
1447	kfree(objp: config->socks);
1448	}
1449	kfree(objp: nbd->config);
1450	nbd->config = NULL;
1451
1452	nbd->tag_set.timeout = 0;
1453
1454	mutex_unlock(lock: &nbd->config_lock);
1455	nbd_put(nbd);
1456	module_put(THIS_MODULE);
1457	}
1458	}
1459
1460	static int nbd_start_device(struct nbd_device *nbd)
1461	{
1462	struct nbd_config *config = nbd->config;
1463	int num_connections = config->num_connections;
1464	int error = 0, i;
1465
1466	if (nbd->pid)
1467	return -EBUSY;
1468	if (!config->socks)
1469	return -EINVAL;
1470	if (num_connections > 1 &&
1471	!(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1472	dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1473	return -EINVAL;
1474	}
1475
1476	blk_mq_update_nr_hw_queues(set: &nbd->tag_set, nr_hw_queues: config->num_connections);
1477	nbd->pid = task_pid_nr(current);
1478
1479	nbd_parse_flags(nbd);
1480
1481	error = device_create_file(disk_to_dev(nbd->disk), entry: &pid_attr);
1482	if (error) {
1483	dev_err(disk_to_dev(nbd->disk), "device_create_file failed for pid!\n");
1484	return error;
1485	}
1486	set_bit(NBD_RT_HAS_PID_FILE, addr: &config->runtime_flags);
1487
1488	nbd_dev_dbg_init(nbd);
1489	for (i = 0; i < num_connections; i++) {
1490	struct recv_thread_args *args;
1491
1492	args = kzalloc(sizeof(*args), GFP_KERNEL);
1493	if (!args) {
1494	sock_shutdown(nbd);
1495	/*
1496	* If num_connections is m (2 < m),
1497	* and NO.1 ~ NO.n(1 < n < m) kzallocs are successful.
1498	* But NO.(n + 1) failed. We still have n recv threads.
1499	* So, add flush_workqueue here to prevent recv threads
1500	* dropping the last config_refs and trying to destroy
1501	* the workqueue from inside the workqueue.
1502	*/
1503	if (i)
1504	flush_workqueue(nbd->recv_workq);
1505	return -ENOMEM;
1506	}
1507	sk_set_memalloc(sk: config->socks[i]->sock->sk);
1508	if (nbd->tag_set.timeout)
1509	config->socks[i]->sock->sk->sk_sndtimeo =
1510	nbd->tag_set.timeout;
1511	atomic_inc(v: &config->recv_threads);
1512	refcount_inc(r: &nbd->config_refs);
1513	INIT_WORK(&args->work, recv_work);
1514	args->nbd = nbd;
1515	args->nsock = config->socks[i];
1516	args->index = i;
1517	queue_work(wq: nbd->recv_workq, work: &args->work);
1518	}
1519	return nbd_set_size(nbd, bytesize: config->bytesize, blksize: nbd_blksize(config));
1520	}
1521
1522	static int nbd_start_device_ioctl(struct nbd_device *nbd)
1523	{
1524	struct nbd_config *config = nbd->config;
1525	int ret;
1526
1527	ret = nbd_start_device(nbd);
1528	if (ret)
1529	return ret;
1530
1531	if (max_part)
1532	set_bit(GD_NEED_PART_SCAN, addr: &nbd->disk->state);
1533	mutex_unlock(lock: &nbd->config_lock);
1534	ret = wait_event_interruptible(config->recv_wq,
1535	atomic_read(&config->recv_threads) == 0);
1536	if (ret) {
1537	sock_shutdown(nbd);
1538	nbd_clear_que(nbd);
1539	}
1540
1541	flush_workqueue(nbd->recv_workq);
1542	mutex_lock(&nbd->config_lock);
1543	nbd_bdev_reset(nbd);
1544	/* user requested, ignore socket errors */
1545	if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1546	ret = 0;
1547	if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1548	ret = -ETIMEDOUT;
1549	return ret;
1550	}
1551
1552	static void nbd_clear_sock_ioctl(struct nbd_device *nbd)
1553	{
1554	nbd_clear_sock(nbd);
1555	disk_force_media_change(disk: nbd->disk);
1556	nbd_bdev_reset(nbd);
1557	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1558	addr: &nbd->config->runtime_flags))
1559	nbd_config_put(nbd);
1560	}
1561
1562	static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1563	{
1564	nbd->tag_set.timeout = timeout * HZ;
1565	if (timeout)
1566	blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1567	else
1568	blk_queue_rq_timeout(nbd->disk->queue, 30 * HZ);
1569	}
1570
1571	/* Must be called with config_lock held */
1572	static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1573	unsigned int cmd, unsigned long arg)
1574	{
1575	struct nbd_config *config = nbd->config;
1576	loff_t bytesize;
1577
1578	switch (cmd) {
1579	case NBD_DISCONNECT:
1580	return nbd_disconnect(nbd);
1581	case NBD_CLEAR_SOCK:
1582	nbd_clear_sock_ioctl(nbd);
1583	return 0;
1584	case NBD_SET_SOCK:
1585	return nbd_add_socket(nbd, arg, netlink: false);
1586	case NBD_SET_BLKSIZE:
1587	return nbd_set_size(nbd, bytesize: config->bytesize, blksize: arg);
1588	case NBD_SET_SIZE:
1589	return nbd_set_size(nbd, bytesize: arg, blksize: nbd_blksize(config));
1590	case NBD_SET_SIZE_BLOCKS:
1591	if (check_shl_overflow(arg, config->blksize_bits, &bytesize))
1592	return -EINVAL;
1593	return nbd_set_size(nbd, bytesize, blksize: nbd_blksize(config));
1594	case NBD_SET_TIMEOUT:
1595	nbd_set_cmd_timeout(nbd, timeout: arg);
1596	return 0;
1597
1598	case NBD_SET_FLAGS:
1599	config->flags = arg;
1600	return 0;
1601	case NBD_DO_IT:
1602	return nbd_start_device_ioctl(nbd);
1603	case NBD_CLEAR_QUE:
1604	/*
1605	* This is for compatibility only. The queue is always cleared
1606	* by NBD_DO_IT or NBD_CLEAR_SOCK.
1607	*/
1608	return 0;
1609	case NBD_PRINT_DEBUG:
1610	/*
1611	* For compatibility only, we no longer keep a list of
1612	* outstanding requests.
1613	*/
1614	return 0;
1615	}
1616	return -ENOTTY;
1617	}
1618
1619	static int nbd_ioctl(struct block_device *bdev, blk_mode_t mode,
1620	unsigned int cmd, unsigned long arg)
1621	{
1622	struct nbd_device *nbd = bdev->bd_disk->private_data;
1623	struct nbd_config *config = nbd->config;
1624	int error = -EINVAL;
1625
1626	if (!capable(CAP_SYS_ADMIN))
1627	return -EPERM;
1628
1629	/* The block layer will pass back some non-nbd ioctls in case we have
1630	* special handling for them, but we don't so just return an error.
1631	*/
1632	if (_IOC_TYPE(cmd) != 0xab)
1633	return -EINVAL;
1634
1635	mutex_lock(&nbd->config_lock);
1636
1637	/* Don't allow ioctl operations on a nbd device that was created with
1638	* netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1639	*/
1640	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1641	(cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1642	error = __nbd_ioctl(bdev, nbd, cmd, arg);
1643	else
1644	dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1645	mutex_unlock(lock: &nbd->config_lock);
1646	return error;
1647	}
1648
1649	static int nbd_alloc_and_init_config(struct nbd_device *nbd)
1650	{
1651	struct nbd_config *config;
1652
1653	if (WARN_ON(nbd->config))
1654	return -EINVAL;
1655
1656	if (!try_module_get(THIS_MODULE))
1657	return -ENODEV;
1658
1659	config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1660	if (!config) {
1661	module_put(THIS_MODULE);
1662	return -ENOMEM;
1663	}
1664
1665	atomic_set(v: &config->recv_threads, i: 0);
1666	init_waitqueue_head(&config->recv_wq);
1667	init_waitqueue_head(&config->conn_wait);
1668	config->blksize_bits = NBD_DEF_BLKSIZE_BITS;
1669	atomic_set(v: &config->live_connections, i: 0);
1670
1671	nbd->config = config;
1672	/*
1673	* Order refcount_set(&nbd->config_refs, 1) and nbd->config assignment,
1674	* its pair is the barrier in nbd_get_config_unlocked().
1675	* So nbd_get_config_unlocked() won't see nbd->config as null after
1676	* refcount_inc_not_zero() succeed.
1677	*/
1678	smp_mb__before_atomic();
1679	refcount_set(r: &nbd->config_refs, n: 1);
1680
1681	return 0;
1682	}
1683
1684	static int nbd_open(struct gendisk *disk, blk_mode_t mode)
1685	{
1686	struct nbd_device *nbd;
1687	struct nbd_config *config;
1688	int ret = 0;
1689
1690	mutex_lock(&nbd_index_mutex);
1691	nbd = disk->private_data;
1692	if (!nbd) {
1693	ret = -ENXIO;
1694	goto out;
1695	}
1696	if (!refcount_inc_not_zero(r: &nbd->refs)) {
1697	ret = -ENXIO;
1698	goto out;
1699	}
1700
1701	config = nbd_get_config_unlocked(nbd);
1702	if (!config) {
1703	mutex_lock(&nbd->config_lock);
1704	if (refcount_inc_not_zero(r: &nbd->config_refs)) {
1705	mutex_unlock(lock: &nbd->config_lock);
1706	goto out;
1707	}
1708	ret = nbd_alloc_and_init_config(nbd);
1709	if (ret) {
1710	mutex_unlock(lock: &nbd->config_lock);
1711	goto out;
1712	}
1713
1714	refcount_inc(r: &nbd->refs);
1715	mutex_unlock(lock: &nbd->config_lock);
1716	if (max_part)
1717	set_bit(GD_NEED_PART_SCAN, addr: &disk->state);
1718	} else if (nbd_disconnected(config)) {
1719	if (max_part)
1720	set_bit(GD_NEED_PART_SCAN, addr: &disk->state);
1721	}
1722	out:
1723	mutex_unlock(lock: &nbd_index_mutex);
1724	return ret;
1725	}
1726
1727	static void nbd_release(struct gendisk *disk)
1728	{
1729	struct nbd_device *nbd = disk->private_data;
1730
1731	if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1732	disk_openers(disk) == 0)
1733	nbd_disconnect_and_put(nbd);
1734
1735	nbd_config_put(nbd);
1736	nbd_put(nbd);
1737	}
1738
1739	static void nbd_free_disk(struct gendisk *disk)
1740	{
1741	struct nbd_device *nbd = disk->private_data;
1742
1743	kfree(objp: nbd);
1744	}
1745
1746	static const struct block_device_operations nbd_fops =
1747	{
1748	.owner = THIS_MODULE,
1749	.open = nbd_open,
1750	.release = nbd_release,
1751	.ioctl = nbd_ioctl,
1752	.compat_ioctl = nbd_ioctl,
1753	.free_disk = nbd_free_disk,
1754	};
1755
1756	#if IS_ENABLED(CONFIG_DEBUG_FS)
1757
1758	static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1759	{
1760	struct nbd_device *nbd = s->private;
1761
1762	if (nbd->pid)
1763	seq_printf(m: s, fmt: "recv: %d\n", nbd->pid);
1764
1765	return 0;
1766	}
1767
1768	DEFINE_SHOW_ATTRIBUTE(nbd_dbg_tasks);
1769
1770	static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1771	{
1772	struct nbd_device *nbd = s->private;
1773	u32 flags = nbd->config->flags;
1774
1775	seq_printf(m: s, fmt: "Hex: 0x%08x\n\n", flags);
1776
1777	seq_puts(m: s, s: "Known flags:\n");
1778
1779	if (flags & NBD_FLAG_HAS_FLAGS)
1780	seq_puts(m: s, s: "NBD_FLAG_HAS_FLAGS\n");
1781	if (flags & NBD_FLAG_READ_ONLY)
1782	seq_puts(m: s, s: "NBD_FLAG_READ_ONLY\n");
1783	if (flags & NBD_FLAG_SEND_FLUSH)
1784	seq_puts(m: s, s: "NBD_FLAG_SEND_FLUSH\n");
1785	if (flags & NBD_FLAG_SEND_FUA)
1786	seq_puts(m: s, s: "NBD_FLAG_SEND_FUA\n");
1787	if (flags & NBD_FLAG_SEND_TRIM)
1788	seq_puts(m: s, s: "NBD_FLAG_SEND_TRIM\n");
1789	if (flags & NBD_FLAG_SEND_WRITE_ZEROES)
1790	seq_puts(m: s, s: "NBD_FLAG_SEND_WRITE_ZEROES\n");
1791	if (flags & NBD_FLAG_ROTATIONAL)
1792	seq_puts(m: s, s: "NBD_FLAG_ROTATIONAL\n");
1793
1794	return 0;
1795	}
1796
1797	DEFINE_SHOW_ATTRIBUTE(nbd_dbg_flags);
1798
1799	static int nbd_dev_dbg_init(struct nbd_device *nbd)
1800	{
1801	struct dentry *dir;
1802	struct nbd_config *config = nbd->config;
1803
1804	if (!nbd_dbg_dir)
1805	return -EIO;
1806
1807	dir = debugfs_create_dir(nbd_name(nbd), parent: nbd_dbg_dir);
1808	if (IS_ERR(ptr: dir)) {
1809	dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1810	nbd_name(nbd));
1811	return -EIO;
1812	}
1813	config->dbg_dir = dir;
1814
1815	debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_fops);
1816	debugfs_create_u64(name: "size_bytes", mode: 0444, parent: dir, value: &config->bytesize);
1817	debugfs_create_u32(name: "timeout", mode: 0444, parent: dir, value: &nbd->tag_set.timeout);
1818	debugfs_create_u32(name: "blocksize_bits", mode: 0444, parent: dir, value: &config->blksize_bits);
1819	debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_fops);
1820
1821	return 0;
1822	}
1823
1824	static void nbd_dev_dbg_close(struct nbd_device *nbd)
1825	{
1826	debugfs_remove_recursive(dentry: nbd->config->dbg_dir);
1827	}
1828
1829	static int nbd_dbg_init(void)
1830	{
1831	struct dentry *dbg_dir;
1832
1833	dbg_dir = debugfs_create_dir(name: "nbd", NULL);
1834	if (IS_ERR(ptr: dbg_dir))
1835	return -EIO;
1836
1837	nbd_dbg_dir = dbg_dir;
1838
1839	return 0;
1840	}
1841
1842	static void nbd_dbg_close(void)
1843	{
1844	debugfs_remove_recursive(dentry: nbd_dbg_dir);
1845	}
1846
1847	#else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1848
1849	static int nbd_dev_dbg_init(struct nbd_device *nbd)
1850	{
1851	return 0;
1852	}
1853
1854	static void nbd_dev_dbg_close(struct nbd_device *nbd)
1855	{
1856	}
1857
1858	static int nbd_dbg_init(void)
1859	{
1860	return 0;
1861	}
1862
1863	static void nbd_dbg_close(void)
1864	{
1865	}
1866
1867	#endif
1868
1869	static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1870	unsigned int hctx_idx, unsigned int numa_node)
1871	{
1872	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1873	cmd->nbd = set->driver_data;
1874	cmd->flags = 0;
1875	mutex_init(&cmd->lock);
1876	return 0;
1877	}
1878
1879	static const struct blk_mq_ops nbd_mq_ops = {
1880	.queue_rq = nbd_queue_rq,
1881	.complete = nbd_complete_rq,
1882	.init_request = nbd_init_request,
1883	.timeout = nbd_xmit_timeout,
1884	};
1885
1886	static struct nbd_device *nbd_dev_add(int index, unsigned int refs)
1887	{
1888	struct queue_limits lim = {
1889	.max_hw_sectors = 65536,
1890	.io_opt = 256 << SECTOR_SHIFT,
1891	.max_segments = USHRT_MAX,
1892	.max_segment_size = UINT_MAX,
1893	};
1894	struct nbd_device *nbd;
1895	struct gendisk *disk;
1896	int err = -ENOMEM;
1897
1898	nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1899	if (!nbd)
1900	goto out;
1901
1902	nbd->tag_set.ops = &nbd_mq_ops;
1903	nbd->tag_set.nr_hw_queues = 1;
1904	nbd->tag_set.queue_depth = 128;
1905	nbd->tag_set.numa_node = NUMA_NO_NODE;
1906	nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1907	nbd->tag_set.flags = BLK_MQ_F_BLOCKING;
1908	nbd->tag_set.driver_data = nbd;
1909	INIT_WORK(&nbd->remove_work, nbd_dev_remove_work);
1910	nbd->backend = NULL;
1911
1912	err = blk_mq_alloc_tag_set(set: &nbd->tag_set);
1913	if (err)
1914	goto out_free_nbd;
1915
1916	mutex_lock(&nbd_index_mutex);
1917	if (index >= 0) {
1918	err = idr_alloc(&nbd_index_idr, ptr: nbd, start: index, end: index + 1,
1919	GFP_KERNEL);
1920	if (err == -ENOSPC)
1921	err = -EEXIST;
1922	} else {
1923	err = idr_alloc(&nbd_index_idr, ptr: nbd, start: 0,
1924	end: (MINORMASK >> part_shift) + 1, GFP_KERNEL);
1925	if (err >= 0)
1926	index = err;
1927	}
1928	nbd->index = index;
1929	mutex_unlock(lock: &nbd_index_mutex);
1930	if (err < 0)
1931	goto out_free_tags;
1932
1933	disk = blk_mq_alloc_disk(&nbd->tag_set, &lim, NULL);
1934	if (IS_ERR(ptr: disk)) {
1935	err = PTR_ERR(ptr: disk);
1936	goto out_free_idr;
1937	}
1938	nbd->disk = disk;
1939
1940	nbd->recv_workq = alloc_workqueue(fmt: "nbd%d-recv",
1941	flags: WQ_MEM_RECLAIM | WQ_HIGHPRI |
1942	WQ_UNBOUND, max_active: 0, nbd->index);
1943	if (!nbd->recv_workq) {
1944	dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1945	err = -ENOMEM;
1946	goto out_err_disk;
1947	}
1948
1949	mutex_init(&nbd->config_lock);
1950	refcount_set(r: &nbd->config_refs, n: 0);
1951	/*
1952	* Start out with a zero references to keep other threads from using
1953	* this device until it is fully initialized.
1954	*/
1955	refcount_set(r: &nbd->refs, n: 0);
1956	INIT_LIST_HEAD(list: &nbd->list);
1957	disk->major = NBD_MAJOR;
1958	disk->first_minor = index << part_shift;
1959	disk->minors = 1 << part_shift;
1960	disk->fops = &nbd_fops;
1961	disk->private_data = nbd;
1962	sprintf(buf: disk->disk_name, fmt: "nbd%d", index);
1963	err = add_disk(disk);
1964	if (err)
1965	goto out_free_work;
1966
1967	/*
1968	* Now publish the device.
1969	*/
1970	refcount_set(r: &nbd->refs, n: refs);
1971	nbd_total_devices++;
1972	return nbd;
1973
1974	out_free_work:
1975	destroy_workqueue(wq: nbd->recv_workq);
1976	out_err_disk:
1977	put_disk(disk);
1978	out_free_idr:
1979	mutex_lock(&nbd_index_mutex);
1980	idr_remove(&nbd_index_idr, id: index);
1981	mutex_unlock(lock: &nbd_index_mutex);
1982	out_free_tags:
1983	blk_mq_free_tag_set(set: &nbd->tag_set);
1984	out_free_nbd:
1985	kfree(objp: nbd);
1986	out:
1987	return ERR_PTR(error: err);
1988	}
1989
1990	static struct nbd_device *nbd_find_get_unused(void)
1991	{
1992	struct nbd_device *nbd;
1993	int id;
1994
1995	lockdep_assert_held(&nbd_index_mutex);
1996
1997	idr_for_each_entry(&nbd_index_idr, nbd, id) {
1998	if (refcount_read(r: &nbd->config_refs) ||
1999	test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags))
2000	continue;
2001	if (refcount_inc_not_zero(r: &nbd->refs))
2002	return nbd;
2003	}
2004
2005	return NULL;
2006	}
2007
2008	/* Netlink interface. */
2009	static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
2010	[NBD_ATTR_INDEX] = { .type = NLA_U32 },
2011	[NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
2012	[NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
2013	[NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
2014	[NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
2015	[NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
2016	[NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
2017	[NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
2018	[NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
2019	[NBD_ATTR_BACKEND_IDENTIFIER] = { .type = NLA_STRING},
2020	};
2021
2022	static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
2023	[NBD_SOCK_FD] = { .type = NLA_U32 },
2024	};
2025
2026	/* We don't use this right now since we don't parse the incoming list, but we
2027	* still want it here so userspace knows what to expect.
2028	*/
2029	static const struct nla_policy __attribute__((unused))
2030	nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
2031	[NBD_DEVICE_INDEX] = { .type = NLA_U32 },
2032	[NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
2033	};
2034
2035	static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
2036	{
2037	struct nbd_config *config = nbd->config;
2038	u64 bsize = nbd_blksize(config);
2039	u64 bytes = config->bytesize;
2040
2041	if (info->attrs[NBD_ATTR_SIZE_BYTES])
2042	bytes = nla_get_u64(nla: info->attrs[NBD_ATTR_SIZE_BYTES]);
2043
2044	if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES])
2045	bsize = nla_get_u64(nla: info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
2046
2047	if (bytes != config->bytesize || bsize != nbd_blksize(config))
2048	return nbd_set_size(nbd, bytesize: bytes, blksize: bsize);
2049	return 0;
2050	}
2051
2052	static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
2053	{
2054	struct nbd_device *nbd;
2055	struct nbd_config *config;
2056	int index = -1;
2057	int ret;
2058	bool put_dev = false;
2059
2060	if (!netlink_capable(skb, CAP_SYS_ADMIN))
2061	return -EPERM;
2062
2063	if (info->attrs[NBD_ATTR_INDEX]) {
2064	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
2065
2066	/*
2067	* Too big first_minor can cause duplicate creation of
2068	* sysfs files/links, since index << part_shift might overflow, or
2069	* MKDEV() expect that the max bits of first_minor is 20.
2070	*/
2071	if (index < 0 || index > MINORMASK >> part_shift) {
2072	pr_err("illegal input index %d\n", index);
2073	return -EINVAL;
2074	}
2075	}
2076	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_SOCKETS)) {
2077	pr_err("must specify at least one socket\n");
2078	return -EINVAL;
2079	}
2080	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_SIZE_BYTES)) {
2081	pr_err("must specify a size in bytes for the device\n");
2082	return -EINVAL;
2083	}
2084	again:
2085	mutex_lock(&nbd_index_mutex);
2086	if (index == -1) {
2087	nbd = nbd_find_get_unused();
2088	} else {
2089	nbd = idr_find(&nbd_index_idr, id: index);
2090	if (nbd) {
2091	if ((test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
2092	test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) ||
2093	!refcount_inc_not_zero(r: &nbd->refs)) {
2094	mutex_unlock(lock: &nbd_index_mutex);
2095	pr_err("device at index %d is going down\n",
2096	index);
2097	return -EINVAL;
2098	}
2099	}
2100	}
2101	mutex_unlock(lock: &nbd_index_mutex);
2102
2103	if (!nbd) {
2104	nbd = nbd_dev_add(index, refs: 2);
2105	if (IS_ERR(ptr: nbd)) {
2106	pr_err("failed to add new device\n");
2107	return PTR_ERR(ptr: nbd);
2108	}
2109	}
2110
2111	mutex_lock(&nbd->config_lock);
2112	if (refcount_read(r: &nbd->config_refs)) {
2113	mutex_unlock(lock: &nbd->config_lock);
2114	nbd_put(nbd);
2115	if (index == -1)
2116	goto again;
2117	pr_err("nbd%d already in use\n", index);
2118	return -EBUSY;
2119	}
2120
2121	ret = nbd_alloc_and_init_config(nbd);
2122	if (ret) {
2123	mutex_unlock(lock: &nbd->config_lock);
2124	nbd_put(nbd);
2125	pr_err("couldn't allocate config\n");
2126	return ret;
2127	}
2128
2129	config = nbd->config;
2130	set_bit(NBD_RT_BOUND, addr: &config->runtime_flags);
2131	ret = nbd_genl_size_set(info, nbd);
2132	if (ret)
2133	goto out;
2134
2135	if (info->attrs[NBD_ATTR_TIMEOUT])
2136	nbd_set_cmd_timeout(nbd,
2137	timeout: nla_get_u64(nla: info->attrs[NBD_ATTR_TIMEOUT]));
2138	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2139	config->dead_conn_timeout =
2140	nla_get_u64(nla: info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2141	config->dead_conn_timeout *= HZ;
2142	}
2143	if (info->attrs[NBD_ATTR_SERVER_FLAGS])
2144	config->flags =
2145	nla_get_u64(nla: info->attrs[NBD_ATTR_SERVER_FLAGS]);
2146	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2147	u64 flags = nla_get_u64(nla: info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2148	if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2149	/*
2150	* We have 1 ref to keep the device around, and then 1
2151	* ref for our current operation here, which will be
2152	* inherited by the config. If we already have
2153	* DESTROY_ON_DISCONNECT set then we know we don't have
2154	* that extra ref already held so we don't need the
2155	* put_dev.
2156	*/
2157	if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
2158	addr: &nbd->flags))
2159	put_dev = true;
2160	} else {
2161	if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
2162	addr: &nbd->flags))
2163	refcount_inc(r: &nbd->refs);
2164	}
2165	if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2166	set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2167	addr: &config->runtime_flags);
2168	}
2169	}
2170
2171	if (info->attrs[NBD_ATTR_SOCKETS]) {
2172	struct nlattr *attr;
2173	int rem, fd;
2174
2175	nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2176	rem) {
2177	struct nlattr *socks[NBD_SOCK_MAX+1];
2178
2179	if (nla_type(nla: attr) != NBD_SOCK_ITEM) {
2180	pr_err("socks must be embedded in a SOCK_ITEM attr\n");
2181	ret = -EINVAL;
2182	goto out;
2183	}
2184	ret = nla_parse_nested_deprecated(tb: socks, NBD_SOCK_MAX,
2185	nla: attr,
2186	policy: nbd_sock_policy,
2187	extack: info->extack);
2188	if (ret != 0) {
2189	pr_err("error processing sock list\n");
2190	ret = -EINVAL;
2191	goto out;
2192	}
2193	if (!socks[NBD_SOCK_FD])
2194	continue;
2195	fd = (int)nla_get_u32(nla: socks[NBD_SOCK_FD]);
2196	ret = nbd_add_socket(nbd, arg: fd, netlink: true);
2197	if (ret)
2198	goto out;
2199	}
2200	}
2201	ret = nbd_start_device(nbd);
2202	if (ret)
2203	goto out;
2204	if (info->attrs[NBD_ATTR_BACKEND_IDENTIFIER]) {
2205	nbd->backend = nla_strdup(nla: info->attrs[NBD_ATTR_BACKEND_IDENTIFIER],
2206	GFP_KERNEL);
2207	if (!nbd->backend) {
2208	ret = -ENOMEM;
2209	goto out;
2210	}
2211	}
2212	ret = device_create_file(disk_to_dev(nbd->disk), entry: &backend_attr);
2213	if (ret) {
2214	dev_err(disk_to_dev(nbd->disk),
2215	"device_create_file failed for backend!\n");
2216	goto out;
2217	}
2218	set_bit(NBD_RT_HAS_BACKEND_FILE, addr: &config->runtime_flags);
2219	out:
2220	mutex_unlock(lock: &nbd->config_lock);
2221	if (!ret) {
2222	set_bit(NBD_RT_HAS_CONFIG_REF, addr: &config->runtime_flags);
2223	refcount_inc(r: &nbd->config_refs);
2224	nbd_connect_reply(info, index: nbd->index);
2225	}
2226	nbd_config_put(nbd);
2227	if (put_dev)
2228	nbd_put(nbd);
2229	return ret;
2230	}
2231
2232	static void nbd_disconnect_and_put(struct nbd_device *nbd)
2233	{
2234	mutex_lock(&nbd->config_lock);
2235	nbd_disconnect(nbd);
2236	sock_shutdown(nbd);
2237	wake_up(&nbd->config->conn_wait);
2238	/*
2239	* Make sure recv thread has finished, we can safely call nbd_clear_que()
2240	* to cancel the inflight I/Os.
2241	*/
2242	flush_workqueue(nbd->recv_workq);
2243	nbd_clear_que(nbd);
2244	nbd->task_setup = NULL;
2245	clear_bit(NBD_RT_BOUND, addr: &nbd->config->runtime_flags);
2246	mutex_unlock(lock: &nbd->config_lock);
2247
2248	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
2249	addr: &nbd->config->runtime_flags))
2250	nbd_config_put(nbd);
2251	}
2252
2253	static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
2254	{
2255	struct nbd_device *nbd;
2256	int index;
2257
2258	if (!netlink_capable(skb, CAP_SYS_ADMIN))
2259	return -EPERM;
2260
2261	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_INDEX)) {
2262	pr_err("must specify an index to disconnect\n");
2263	return -EINVAL;
2264	}
2265	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
2266	mutex_lock(&nbd_index_mutex);
2267	nbd = idr_find(&nbd_index_idr, id: index);
2268	if (!nbd) {
2269	mutex_unlock(lock: &nbd_index_mutex);
2270	pr_err("couldn't find device at index %d\n", index);
2271	return -EINVAL;
2272	}
2273	if (!refcount_inc_not_zero(r: &nbd->refs)) {
2274	mutex_unlock(lock: &nbd_index_mutex);
2275	pr_err("device at index %d is going down\n", index);
2276	return -EINVAL;
2277	}
2278	mutex_unlock(lock: &nbd_index_mutex);
2279	if (!refcount_inc_not_zero(r: &nbd->config_refs))
2280	goto put_nbd;
2281	nbd_disconnect_and_put(nbd);
2282	nbd_config_put(nbd);
2283	put_nbd:
2284	nbd_put(nbd);
2285	return 0;
2286	}
2287
2288	static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2289	{
2290	struct nbd_device *nbd = NULL;
2291	struct nbd_config *config;
2292	int index;
2293	int ret = 0;
2294	bool put_dev = false;
2295
2296	if (!netlink_capable(skb, CAP_SYS_ADMIN))
2297	return -EPERM;
2298
2299	if (GENL_REQ_ATTR_CHECK(info, NBD_ATTR_INDEX)) {
2300	pr_err("must specify a device to reconfigure\n");
2301	return -EINVAL;
2302	}
2303	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
2304	mutex_lock(&nbd_index_mutex);
2305	nbd = idr_find(&nbd_index_idr, id: index);
2306	if (!nbd) {
2307	mutex_unlock(lock: &nbd_index_mutex);
2308	pr_err("couldn't find a device at index %d\n", index);
2309	return -EINVAL;
2310	}
2311	if (nbd->backend) {
2312	if (info->attrs[NBD_ATTR_BACKEND_IDENTIFIER]) {
2313	if (nla_strcmp(nla: info->attrs[NBD_ATTR_BACKEND_IDENTIFIER],
2314	str: nbd->backend)) {
2315	mutex_unlock(lock: &nbd_index_mutex);
2316	dev_err(nbd_to_dev(nbd),
2317	"backend image doesn't match with %s\n",
2318	nbd->backend);
2319	return -EINVAL;
2320	}
2321	} else {
2322	mutex_unlock(lock: &nbd_index_mutex);
2323	dev_err(nbd_to_dev(nbd), "must specify backend\n");
2324	return -EINVAL;
2325	}
2326	}
2327	if (!refcount_inc_not_zero(r: &nbd->refs)) {
2328	mutex_unlock(lock: &nbd_index_mutex);
2329	pr_err("device at index %d is going down\n", index);
2330	return -EINVAL;
2331	}
2332	mutex_unlock(lock: &nbd_index_mutex);
2333
2334	config = nbd_get_config_unlocked(nbd);
2335	if (!config) {
2336	dev_err(nbd_to_dev(nbd),
2337	"not configured, cannot reconfigure\n");
2338	nbd_put(nbd);
2339	return -EINVAL;
2340	}
2341
2342	mutex_lock(&nbd->config_lock);
2343	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2344	!nbd->pid) {
2345	dev_err(nbd_to_dev(nbd),
2346	"not configured, cannot reconfigure\n");
2347	ret = -EINVAL;
2348	goto out;
2349	}
2350
2351	ret = nbd_genl_size_set(info, nbd);
2352	if (ret)
2353	goto out;
2354
2355	if (info->attrs[NBD_ATTR_TIMEOUT])
2356	nbd_set_cmd_timeout(nbd,
2357	timeout: nla_get_u64(nla: info->attrs[NBD_ATTR_TIMEOUT]));
2358	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2359	config->dead_conn_timeout =
2360	nla_get_u64(nla: info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2361	config->dead_conn_timeout *= HZ;
2362	}
2363	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2364	u64 flags = nla_get_u64(nla: info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2365	if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2366	if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
2367	addr: &nbd->flags))
2368	put_dev = true;
2369	} else {
2370	if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
2371	addr: &nbd->flags))
2372	refcount_inc(r: &nbd->refs);
2373	}
2374
2375	if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2376	set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2377	addr: &config->runtime_flags);
2378	} else {
2379	clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2380	addr: &config->runtime_flags);
2381	}
2382	}
2383
2384	if (info->attrs[NBD_ATTR_SOCKETS]) {
2385	struct nlattr *attr;
2386	int rem, fd;
2387
2388	nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2389	rem) {
2390	struct nlattr *socks[NBD_SOCK_MAX+1];
2391
2392	if (nla_type(nla: attr) != NBD_SOCK_ITEM) {
2393	pr_err("socks must be embedded in a SOCK_ITEM attr\n");
2394	ret = -EINVAL;
2395	goto out;
2396	}
2397	ret = nla_parse_nested_deprecated(tb: socks, NBD_SOCK_MAX,
2398	nla: attr,
2399	policy: nbd_sock_policy,
2400	extack: info->extack);
2401	if (ret != 0) {
2402	pr_err("error processing sock list\n");
2403	ret = -EINVAL;
2404	goto out;
2405	}
2406	if (!socks[NBD_SOCK_FD])
2407	continue;
2408	fd = (int)nla_get_u32(nla: socks[NBD_SOCK_FD]);
2409	ret = nbd_reconnect_socket(nbd, arg: fd);
2410	if (ret) {
2411	if (ret == -ENOSPC)
2412	ret = 0;
2413	goto out;
2414	}
2415	dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2416	}
2417	}
2418	out:
2419	mutex_unlock(lock: &nbd->config_lock);
2420	nbd_config_put(nbd);
2421	nbd_put(nbd);
2422	if (put_dev)
2423	nbd_put(nbd);
2424	return ret;
2425	}
2426
2427	static const struct genl_small_ops nbd_connect_genl_ops[] = {
2428	{
2429	.cmd = NBD_CMD_CONNECT,
2430	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2431	.doit = nbd_genl_connect,
2432	},
2433	{
2434	.cmd = NBD_CMD_DISCONNECT,
2435	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2436	.doit = nbd_genl_disconnect,
2437	},
2438	{
2439	.cmd = NBD_CMD_RECONFIGURE,
2440	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2441	.doit = nbd_genl_reconfigure,
2442	},
2443	{
2444	.cmd = NBD_CMD_STATUS,
2445	.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2446	.doit = nbd_genl_status,
2447	},
2448	};
2449
2450	static const struct genl_multicast_group nbd_mcast_grps[] = {
2451	{ .name = NBD_GENL_MCAST_GROUP_NAME, },
2452	};
2453
2454	static struct genl_family nbd_genl_family __ro_after_init = {
2455	.hdrsize = 0,
2456	.name = NBD_GENL_FAMILY_NAME,
2457	.version = NBD_GENL_VERSION,
2458	.module = THIS_MODULE,
2459	.small_ops = nbd_connect_genl_ops,
2460	.n_small_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2461	.resv_start_op = NBD_CMD_STATUS + 1,
2462	.maxattr = NBD_ATTR_MAX,
2463	.netnsok = 1,
2464	.policy = nbd_attr_policy,
2465	.mcgrps = nbd_mcast_grps,
2466	.n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2467	};
2468	MODULE_ALIAS_GENL_FAMILY(NBD_GENL_FAMILY_NAME);
2469
2470	static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2471	{
2472	struct nlattr *dev_opt;
2473	u8 connected = 0;
2474	int ret;
2475
2476	/* This is a little racey, but for status it's ok. The
2477	* reason we don't take a ref here is because we can't
2478	* take a ref in the index == -1 case as we would need
2479	* to put under the nbd_index_mutex, which could
2480	* deadlock if we are configured to remove ourselves
2481	* once we're disconnected.
2482	*/
2483	if (refcount_read(r: &nbd->config_refs))
2484	connected = 1;
2485	dev_opt = nla_nest_start_noflag(skb: reply, attrtype: NBD_DEVICE_ITEM);
2486	if (!dev_opt)
2487	return -EMSGSIZE;
2488	ret = nla_put_u32(skb: reply, attrtype: NBD_DEVICE_INDEX, value: nbd->index);
2489	if (ret)
2490	return -EMSGSIZE;
2491	ret = nla_put_u8(skb: reply, attrtype: NBD_DEVICE_CONNECTED,
2492	value: connected);
2493	if (ret)
2494	return -EMSGSIZE;
2495	nla_nest_end(skb: reply, start: dev_opt);
2496	return 0;
2497	}
2498
2499	static int status_cb(int id, void *ptr, void *data)
2500	{
2501	struct nbd_device *nbd = ptr;
2502	return populate_nbd_status(nbd, reply: (struct sk_buff *)data);
2503	}
2504
2505	static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2506	{
2507	struct nlattr *dev_list;
2508	struct sk_buff *reply;
2509	void *reply_head;
2510	size_t msg_size;
2511	int index = -1;
2512	int ret = -ENOMEM;
2513
2514	if (info->attrs[NBD_ATTR_INDEX])
2515	index = nla_get_u32(nla: info->attrs[NBD_ATTR_INDEX]);
2516
2517	mutex_lock(&nbd_index_mutex);
2518
2519	msg_size = nla_total_size(payload: nla_attr_size(payload: sizeof(u32)) +
2520	nla_attr_size(payload: sizeof(u8)));
2521	msg_size *= (index == -1) ? nbd_total_devices : 1;
2522
2523	reply = genlmsg_new(payload: msg_size, GFP_KERNEL);
2524	if (!reply)
2525	goto out;
2526	reply_head = genlmsg_put_reply(skb: reply, info, family: &nbd_genl_family, flags: 0,
2527	cmd: NBD_CMD_STATUS);
2528	if (!reply_head) {
2529	nlmsg_free(skb: reply);
2530	goto out;
2531	}
2532
2533	dev_list = nla_nest_start_noflag(skb: reply, attrtype: NBD_ATTR_DEVICE_LIST);
2534	if (!dev_list) {
2535	nlmsg_free(skb: reply);
2536	ret = -EMSGSIZE;
2537	goto out;
2538	}
2539
2540	if (index == -1) {
2541	ret = idr_for_each(&nbd_index_idr, fn: &status_cb, data: reply);
2542	if (ret) {
2543	nlmsg_free(skb: reply);
2544	goto out;
2545	}
2546	} else {
2547	struct nbd_device *nbd;
2548	nbd = idr_find(&nbd_index_idr, id: index);
2549	if (nbd) {
2550	ret = populate_nbd_status(nbd, reply);
2551	if (ret) {
2552	nlmsg_free(skb: reply);
2553	goto out;
2554	}
2555	}
2556	}
2557	nla_nest_end(skb: reply, start: dev_list);
2558	genlmsg_end(skb: reply, hdr: reply_head);
2559	ret = genlmsg_reply(skb: reply, info);
2560	out:
2561	mutex_unlock(lock: &nbd_index_mutex);
2562	return ret;
2563	}
2564
2565	static void nbd_connect_reply(struct genl_info *info, int index)
2566	{
2567	struct sk_buff *skb;
2568	void *msg_head;
2569	int ret;
2570
2571	skb = genlmsg_new(payload: nla_total_size(payload: sizeof(u32)), GFP_KERNEL);
2572	if (!skb)
2573	return;
2574	msg_head = genlmsg_put_reply(skb, info, family: &nbd_genl_family, flags: 0,
2575	cmd: NBD_CMD_CONNECT);
2576	if (!msg_head) {
2577	nlmsg_free(skb);
2578	return;
2579	}
2580	ret = nla_put_u32(skb, attrtype: NBD_ATTR_INDEX, value: index);
2581	if (ret) {
2582	nlmsg_free(skb);
2583	return;
2584	}
2585	genlmsg_end(skb, hdr: msg_head);
2586	genlmsg_reply(skb, info);
2587	}
2588
2589	static void nbd_mcast_index(int index)
2590	{
2591	struct sk_buff *skb;
2592	void *msg_head;
2593	int ret;
2594
2595	skb = genlmsg_new(payload: nla_total_size(payload: sizeof(u32)), GFP_KERNEL);
2596	if (!skb)
2597	return;
2598	msg_head = genlmsg_put(skb, portid: 0, seq: 0, family: &nbd_genl_family, flags: 0,
2599	cmd: NBD_CMD_LINK_DEAD);
2600	if (!msg_head) {
2601	nlmsg_free(skb);
2602	return;
2603	}
2604	ret = nla_put_u32(skb, attrtype: NBD_ATTR_INDEX, value: index);
2605	if (ret) {
2606	nlmsg_free(skb);
2607	return;
2608	}
2609	genlmsg_end(skb, hdr: msg_head);
2610	genlmsg_multicast(family: &nbd_genl_family, skb, portid: 0, group: 0, GFP_KERNEL);
2611	}
2612
2613	static void nbd_dead_link_work(struct work_struct *work)
2614	{
2615	struct link_dead_args *args = container_of(work, struct link_dead_args,
2616	work);
2617	nbd_mcast_index(index: args->index);
2618	kfree(objp: args);
2619	}
2620
2621	static int __init nbd_init(void)
2622	{
2623	int i;
2624
2625	BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2626
2627	if (max_part < 0) {
2628	pr_err("max_part must be >= 0\n");
2629	return -EINVAL;
2630	}
2631
2632	part_shift = 0;
2633	if (max_part > 0) {
2634	part_shift = fls(x: max_part);
2635
2636	/*
2637	* Adjust max_part according to part_shift as it is exported
2638	* to user space so that user can know the max number of
2639	* partition kernel should be able to manage.
2640	*
2641	* Note that -1 is required because partition 0 is reserved
2642	* for the whole disk.
2643	*/
2644	max_part = (1UL << part_shift) - 1;
2645	}
2646
2647	if ((1UL << part_shift) > DISK_MAX_PARTS)
2648	return -EINVAL;
2649
2650	if (nbds_max > 1UL << (MINORBITS - part_shift))
2651	return -EINVAL;
2652
2653	if (register_blkdev(NBD_MAJOR, "nbd"))
2654	return -EIO;
2655
2656	nbd_del_wq = alloc_workqueue(fmt: "nbd-del", flags: WQ_UNBOUND, max_active: 0);
2657	if (!nbd_del_wq) {
2658	unregister_blkdev(NBD_MAJOR, name: "nbd");
2659	return -ENOMEM;
2660	}
2661
2662	if (genl_register_family(family: &nbd_genl_family)) {
2663	destroy_workqueue(wq: nbd_del_wq);
2664	unregister_blkdev(NBD_MAJOR, name: "nbd");
2665	return -EINVAL;
2666	}
2667	nbd_dbg_init();
2668
2669	for (i = 0; i < nbds_max; i++)
2670	nbd_dev_add(index: i, refs: 1);
2671	return 0;
2672	}
2673
2674	static int nbd_exit_cb(int id, void *ptr, void *data)
2675	{
2676	struct list_head *list = (struct list_head *)data;
2677	struct nbd_device *nbd = ptr;
2678
2679	/* Skip nbd that is being removed asynchronously */
2680	if (refcount_read(r: &nbd->refs))
2681	list_add_tail(new: &nbd->list, head: list);
2682
2683	return 0;
2684	}
2685
2686	static void __exit nbd_cleanup(void)
2687	{
2688	struct nbd_device *nbd;
2689	LIST_HEAD(del_list);
2690
2691	/*
2692	* Unregister netlink interface prior to waiting
2693	* for the completion of netlink commands.
2694	*/
2695	genl_unregister_family(family: &nbd_genl_family);
2696
2697	nbd_dbg_close();
2698
2699	mutex_lock(&nbd_index_mutex);
2700	idr_for_each(&nbd_index_idr, fn: &nbd_exit_cb, data: &del_list);
2701	mutex_unlock(lock: &nbd_index_mutex);
2702
2703	while (!list_empty(head: &del_list)) {
2704	nbd = list_first_entry(&del_list, struct nbd_device, list);
2705	list_del_init(entry: &nbd->list);
2706	if (refcount_read(r: &nbd->config_refs))
2707	pr_err("possibly leaking nbd_config (ref %d)\n",
2708	refcount_read(&nbd->config_refs));
2709	if (refcount_read(r: &nbd->refs) != 1)
2710	pr_err("possibly leaking a device\n");
2711	nbd_put(nbd);
2712	}
2713
2714	/* Also wait for nbd_dev_remove_work() completes */
2715	destroy_workqueue(wq: nbd_del_wq);
2716
2717	idr_destroy(&nbd_index_idr);
2718	unregister_blkdev(NBD_MAJOR, name: "nbd");
2719	}
2720
2721	module_init(nbd_init);
2722	module_exit(nbd_cleanup);
2723
2724	MODULE_DESCRIPTION("Network Block Device");
2725	MODULE_LICENSE("GPL");
2726
2727	module_param(nbds_max, int, 0444);
2728	MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2729	module_param(max_part, int, 0444);
2730	MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");
2731