xref: /openbmc/linux/drivers/block/nbd.c (revision d2a99dbd)
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 #include <linux/major.h>
15 
16 #include <linux/blkdev.h>
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched.h>
20 #include <linux/sched/mm.h>
21 #include <linux/fs.h>
22 #include <linux/bio.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/file.h>
26 #include <linux/ioctl.h>
27 #include <linux/mutex.h>
28 #include <linux/compiler.h>
29 #include <linux/completion.h>
30 #include <linux/err.h>
31 #include <linux/kernel.h>
32 #include <linux/slab.h>
33 #include <net/sock.h>
34 #include <linux/net.h>
35 #include <linux/kthread.h>
36 #include <linux/types.h>
37 #include <linux/debugfs.h>
38 #include <linux/blk-mq.h>
39 
40 #include <linux/uaccess.h>
41 #include <asm/types.h>
42 
43 #include <linux/nbd.h>
44 #include <linux/nbd-netlink.h>
45 #include <net/genetlink.h>
46 
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/nbd.h>
49 
50 static DEFINE_IDR(nbd_index_idr);
51 static DEFINE_MUTEX(nbd_index_mutex);
52 static int nbd_total_devices = 0;
53 
54 struct nbd_sock {
55 	struct socket *sock;
56 	struct mutex tx_lock;
57 	struct request *pending;
58 	int sent;
59 	bool dead;
60 	int fallback_index;
61 	int cookie;
62 };
63 
64 struct recv_thread_args {
65 	struct work_struct work;
66 	struct nbd_device *nbd;
67 	int index;
68 };
69 
70 struct link_dead_args {
71 	struct work_struct work;
72 	int index;
73 };
74 
75 #define NBD_RT_TIMEDOUT			0
76 #define NBD_RT_DISCONNECT_REQUESTED	1
77 #define NBD_RT_DISCONNECTED		2
78 #define NBD_RT_HAS_PID_FILE		3
79 #define NBD_RT_HAS_CONFIG_REF		4
80 #define NBD_RT_BOUND			5
81 #define NBD_RT_DESTROY_ON_DISCONNECT	6
82 #define NBD_RT_DISCONNECT_ON_CLOSE	7
83 
84 #define NBD_DESTROY_ON_DISCONNECT	0
85 #define NBD_DISCONNECT_REQUESTED	1
86 
87 struct nbd_config {
88 	u32 flags;
89 	unsigned long runtime_flags;
90 	u64 dead_conn_timeout;
91 
92 	struct nbd_sock **socks;
93 	int num_connections;
94 	atomic_t live_connections;
95 	wait_queue_head_t conn_wait;
96 
97 	atomic_t recv_threads;
98 	wait_queue_head_t recv_wq;
99 	loff_t blksize;
100 	loff_t bytesize;
101 #if IS_ENABLED(CONFIG_DEBUG_FS)
102 	struct dentry *dbg_dir;
103 #endif
104 };
105 
106 struct nbd_device {
107 	struct blk_mq_tag_set tag_set;
108 
109 	int index;
110 	refcount_t config_refs;
111 	refcount_t refs;
112 	struct nbd_config *config;
113 	struct mutex config_lock;
114 	struct gendisk *disk;
115 	struct workqueue_struct *recv_workq;
116 
117 	struct list_head list;
118 	struct task_struct *task_recv;
119 	struct task_struct *task_setup;
120 
121 	struct completion *destroy_complete;
122 	unsigned long flags;
123 };
124 
125 #define NBD_CMD_REQUEUED	1
126 
127 struct nbd_cmd {
128 	struct nbd_device *nbd;
129 	struct mutex lock;
130 	int index;
131 	int cookie;
132 	int retries;
133 	blk_status_t status;
134 	unsigned long flags;
135 	u32 cmd_cookie;
136 };
137 
138 #if IS_ENABLED(CONFIG_DEBUG_FS)
139 static struct dentry *nbd_dbg_dir;
140 #endif
141 
142 #define nbd_name(nbd) ((nbd)->disk->disk_name)
143 
144 #define NBD_MAGIC 0x68797548
145 
146 #define NBD_DEF_BLKSIZE 1024
147 
148 static unsigned int nbds_max = 16;
149 static int max_part = 16;
150 static int part_shift;
151 
152 static int nbd_dev_dbg_init(struct nbd_device *nbd);
153 static void nbd_dev_dbg_close(struct nbd_device *nbd);
154 static void nbd_config_put(struct nbd_device *nbd);
155 static void nbd_connect_reply(struct genl_info *info, int index);
156 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
157 static void nbd_dead_link_work(struct work_struct *work);
158 static void nbd_disconnect_and_put(struct nbd_device *nbd);
159 
160 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
161 {
162 	return disk_to_dev(nbd->disk);
163 }
164 
165 static void nbd_requeue_cmd(struct nbd_cmd *cmd)
166 {
167 	struct request *req = blk_mq_rq_from_pdu(cmd);
168 
169 	if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
170 		blk_mq_requeue_request(req, true);
171 }
172 
173 #define NBD_COOKIE_BITS 32
174 
175 static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
176 {
177 	struct request *req = blk_mq_rq_from_pdu(cmd);
178 	u32 tag = blk_mq_unique_tag(req);
179 	u64 cookie = cmd->cmd_cookie;
180 
181 	return (cookie << NBD_COOKIE_BITS) | tag;
182 }
183 
184 static u32 nbd_handle_to_tag(u64 handle)
185 {
186 	return (u32)handle;
187 }
188 
189 static u32 nbd_handle_to_cookie(u64 handle)
190 {
191 	return (u32)(handle >> NBD_COOKIE_BITS);
192 }
193 
194 static const char *nbdcmd_to_ascii(int cmd)
195 {
196 	switch (cmd) {
197 	case  NBD_CMD_READ: return "read";
198 	case NBD_CMD_WRITE: return "write";
199 	case  NBD_CMD_DISC: return "disconnect";
200 	case NBD_CMD_FLUSH: return "flush";
201 	case  NBD_CMD_TRIM: return "trim/discard";
202 	}
203 	return "invalid";
204 }
205 
206 static ssize_t pid_show(struct device *dev,
207 			struct device_attribute *attr, char *buf)
208 {
209 	struct gendisk *disk = dev_to_disk(dev);
210 	struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
211 
212 	return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
213 }
214 
215 static const struct device_attribute pid_attr = {
216 	.attr = { .name = "pid", .mode = 0444},
217 	.show = pid_show,
218 };
219 
220 static void nbd_dev_remove(struct nbd_device *nbd)
221 {
222 	struct gendisk *disk = nbd->disk;
223 	struct request_queue *q;
224 
225 	if (disk) {
226 		q = disk->queue;
227 		del_gendisk(disk);
228 		blk_cleanup_queue(q);
229 		blk_mq_free_tag_set(&nbd->tag_set);
230 		disk->private_data = NULL;
231 		put_disk(disk);
232 	}
233 
234 	/*
235 	 * Place this in the last just before the nbd is freed to
236 	 * make sure that the disk and the related kobject are also
237 	 * totally removed to avoid duplicate creation of the same
238 	 * one.
239 	 */
240 	if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) && nbd->destroy_complete)
241 		complete(nbd->destroy_complete);
242 
243 	kfree(nbd);
244 }
245 
246 static void nbd_put(struct nbd_device *nbd)
247 {
248 	if (refcount_dec_and_mutex_lock(&nbd->refs,
249 					&nbd_index_mutex)) {
250 		idr_remove(&nbd_index_idr, nbd->index);
251 		nbd_dev_remove(nbd);
252 		mutex_unlock(&nbd_index_mutex);
253 	}
254 }
255 
256 static int nbd_disconnected(struct nbd_config *config)
257 {
258 	return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
259 		test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
260 }
261 
262 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
263 				int notify)
264 {
265 	if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
266 		struct link_dead_args *args;
267 		args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
268 		if (args) {
269 			INIT_WORK(&args->work, nbd_dead_link_work);
270 			args->index = nbd->index;
271 			queue_work(system_wq, &args->work);
272 		}
273 	}
274 	if (!nsock->dead) {
275 		kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
276 		if (atomic_dec_return(&nbd->config->live_connections) == 0) {
277 			if (test_and_clear_bit(NBD_RT_DISCONNECT_REQUESTED,
278 					       &nbd->config->runtime_flags)) {
279 				set_bit(NBD_RT_DISCONNECTED,
280 					&nbd->config->runtime_flags);
281 				dev_info(nbd_to_dev(nbd),
282 					"Disconnected due to user request.\n");
283 			}
284 		}
285 	}
286 	nsock->dead = true;
287 	nsock->pending = NULL;
288 	nsock->sent = 0;
289 }
290 
291 static void nbd_size_clear(struct nbd_device *nbd)
292 {
293 	if (nbd->config->bytesize) {
294 		set_capacity(nbd->disk, 0);
295 		kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
296 	}
297 }
298 
299 static void nbd_size_update(struct nbd_device *nbd)
300 {
301 	struct nbd_config *config = nbd->config;
302 	struct block_device *bdev = bdget_disk(nbd->disk, 0);
303 
304 	if (config->flags & NBD_FLAG_SEND_TRIM) {
305 		nbd->disk->queue->limits.discard_granularity = config->blksize;
306 		nbd->disk->queue->limits.discard_alignment = config->blksize;
307 		blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
308 	}
309 	blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
310 	blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
311 	set_capacity(nbd->disk, config->bytesize >> 9);
312 	if (bdev) {
313 		if (bdev->bd_disk) {
314 			bd_set_size(bdev, config->bytesize);
315 			set_blocksize(bdev, config->blksize);
316 		} else
317 			bdev->bd_invalidated = 1;
318 		bdput(bdev);
319 	}
320 	kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
321 }
322 
323 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
324 			 loff_t nr_blocks)
325 {
326 	struct nbd_config *config = nbd->config;
327 	config->blksize = blocksize;
328 	config->bytesize = blocksize * nr_blocks;
329 	if (nbd->task_recv != NULL)
330 		nbd_size_update(nbd);
331 }
332 
333 static void nbd_complete_rq(struct request *req)
334 {
335 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
336 
337 	dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
338 		cmd->status ? "failed" : "done");
339 
340 	blk_mq_end_request(req, cmd->status);
341 }
342 
343 /*
344  * Forcibly shutdown the socket causing all listeners to error
345  */
346 static void sock_shutdown(struct nbd_device *nbd)
347 {
348 	struct nbd_config *config = nbd->config;
349 	int i;
350 
351 	if (config->num_connections == 0)
352 		return;
353 	if (test_and_set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
354 		return;
355 
356 	for (i = 0; i < config->num_connections; i++) {
357 		struct nbd_sock *nsock = config->socks[i];
358 		mutex_lock(&nsock->tx_lock);
359 		nbd_mark_nsock_dead(nbd, nsock, 0);
360 		mutex_unlock(&nsock->tx_lock);
361 	}
362 	dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
363 }
364 
365 static u32 req_to_nbd_cmd_type(struct request *req)
366 {
367 	switch (req_op(req)) {
368 	case REQ_OP_DISCARD:
369 		return NBD_CMD_TRIM;
370 	case REQ_OP_FLUSH:
371 		return NBD_CMD_FLUSH;
372 	case REQ_OP_WRITE:
373 		return NBD_CMD_WRITE;
374 	case REQ_OP_READ:
375 		return NBD_CMD_READ;
376 	default:
377 		return U32_MAX;
378 	}
379 }
380 
381 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
382 						 bool reserved)
383 {
384 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
385 	struct nbd_device *nbd = cmd->nbd;
386 	struct nbd_config *config;
387 
388 	if (!mutex_trylock(&cmd->lock))
389 		return BLK_EH_RESET_TIMER;
390 
391 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
392 		cmd->status = BLK_STS_TIMEOUT;
393 		mutex_unlock(&cmd->lock);
394 		goto done;
395 	}
396 	config = nbd->config;
397 
398 	if (config->num_connections > 1) {
399 		dev_err_ratelimited(nbd_to_dev(nbd),
400 				    "Connection timed out, retrying (%d/%d alive)\n",
401 				    atomic_read(&config->live_connections),
402 				    config->num_connections);
403 		/*
404 		 * Hooray we have more connections, requeue this IO, the submit
405 		 * path will put it on a real connection.
406 		 */
407 		if (config->socks && config->num_connections > 1) {
408 			if (cmd->index < config->num_connections) {
409 				struct nbd_sock *nsock =
410 					config->socks[cmd->index];
411 				mutex_lock(&nsock->tx_lock);
412 				/* We can have multiple outstanding requests, so
413 				 * we don't want to mark the nsock dead if we've
414 				 * already reconnected with a new socket, so
415 				 * only mark it dead if its the same socket we
416 				 * were sent out on.
417 				 */
418 				if (cmd->cookie == nsock->cookie)
419 					nbd_mark_nsock_dead(nbd, nsock, 1);
420 				mutex_unlock(&nsock->tx_lock);
421 			}
422 			mutex_unlock(&cmd->lock);
423 			nbd_requeue_cmd(cmd);
424 			nbd_config_put(nbd);
425 			return BLK_EH_DONE;
426 		}
427 	}
428 
429 	if (!nbd->tag_set.timeout) {
430 		/*
431 		 * Userspace sets timeout=0 to disable socket disconnection,
432 		 * so just warn and reset the timer.
433 		 */
434 		cmd->retries++;
435 		dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
436 			req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
437 			(unsigned long long)blk_rq_pos(req) << 9,
438 			blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
439 
440 		mutex_unlock(&cmd->lock);
441 		nbd_config_put(nbd);
442 		return BLK_EH_RESET_TIMER;
443 	}
444 
445 	dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
446 	set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
447 	cmd->status = BLK_STS_IOERR;
448 	mutex_unlock(&cmd->lock);
449 	sock_shutdown(nbd);
450 	nbd_config_put(nbd);
451 done:
452 	blk_mq_complete_request(req);
453 	return BLK_EH_DONE;
454 }
455 
456 /*
457  *  Send or receive packet.
458  */
459 static int sock_xmit(struct nbd_device *nbd, int index, int send,
460 		     struct iov_iter *iter, int msg_flags, int *sent)
461 {
462 	struct nbd_config *config = nbd->config;
463 	struct socket *sock = config->socks[index]->sock;
464 	int result;
465 	struct msghdr msg;
466 	unsigned int noreclaim_flag;
467 
468 	if (unlikely(!sock)) {
469 		dev_err_ratelimited(disk_to_dev(nbd->disk),
470 			"Attempted %s on closed socket in sock_xmit\n",
471 			(send ? "send" : "recv"));
472 		return -EINVAL;
473 	}
474 
475 	msg.msg_iter = *iter;
476 
477 	noreclaim_flag = memalloc_noreclaim_save();
478 	do {
479 		sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
480 		msg.msg_name = NULL;
481 		msg.msg_namelen = 0;
482 		msg.msg_control = NULL;
483 		msg.msg_controllen = 0;
484 		msg.msg_flags = msg_flags | MSG_NOSIGNAL;
485 
486 		if (send)
487 			result = sock_sendmsg(sock, &msg);
488 		else
489 			result = sock_recvmsg(sock, &msg, msg.msg_flags);
490 
491 		if (result <= 0) {
492 			if (result == 0)
493 				result = -EPIPE; /* short read */
494 			break;
495 		}
496 		if (sent)
497 			*sent += result;
498 	} while (msg_data_left(&msg));
499 
500 	memalloc_noreclaim_restore(noreclaim_flag);
501 
502 	return result;
503 }
504 
505 /*
506  * Different settings for sk->sk_sndtimeo can result in different return values
507  * if there is a signal pending when we enter sendmsg, because reasons?
508  */
509 static inline int was_interrupted(int result)
510 {
511 	return result == -ERESTARTSYS || result == -EINTR;
512 }
513 
514 /* always call with the tx_lock held */
515 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
516 {
517 	struct request *req = blk_mq_rq_from_pdu(cmd);
518 	struct nbd_config *config = nbd->config;
519 	struct nbd_sock *nsock = config->socks[index];
520 	int result;
521 	struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
522 	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
523 	struct iov_iter from;
524 	unsigned long size = blk_rq_bytes(req);
525 	struct bio *bio;
526 	u64 handle;
527 	u32 type;
528 	u32 nbd_cmd_flags = 0;
529 	int sent = nsock->sent, skip = 0;
530 
531 	iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
532 
533 	type = req_to_nbd_cmd_type(req);
534 	if (type == U32_MAX)
535 		return -EIO;
536 
537 	if (rq_data_dir(req) == WRITE &&
538 	    (config->flags & NBD_FLAG_READ_ONLY)) {
539 		dev_err_ratelimited(disk_to_dev(nbd->disk),
540 				    "Write on read-only\n");
541 		return -EIO;
542 	}
543 
544 	if (req->cmd_flags & REQ_FUA)
545 		nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
546 
547 	/* We did a partial send previously, and we at least sent the whole
548 	 * request struct, so just go and send the rest of the pages in the
549 	 * request.
550 	 */
551 	if (sent) {
552 		if (sent >= sizeof(request)) {
553 			skip = sent - sizeof(request);
554 
555 			/* initialize handle for tracing purposes */
556 			handle = nbd_cmd_handle(cmd);
557 
558 			goto send_pages;
559 		}
560 		iov_iter_advance(&from, sent);
561 	} else {
562 		cmd->cmd_cookie++;
563 	}
564 	cmd->index = index;
565 	cmd->cookie = nsock->cookie;
566 	cmd->retries = 0;
567 	request.type = htonl(type | nbd_cmd_flags);
568 	if (type != NBD_CMD_FLUSH) {
569 		request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
570 		request.len = htonl(size);
571 	}
572 	handle = nbd_cmd_handle(cmd);
573 	memcpy(request.handle, &handle, sizeof(handle));
574 
575 	trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
576 
577 	dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
578 		req, nbdcmd_to_ascii(type),
579 		(unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
580 	result = sock_xmit(nbd, index, 1, &from,
581 			(type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
582 	trace_nbd_header_sent(req, handle);
583 	if (result <= 0) {
584 		if (was_interrupted(result)) {
585 			/* If we havne't sent anything we can just return BUSY,
586 			 * however if we have sent something we need to make
587 			 * sure we only allow this req to be sent until we are
588 			 * completely done.
589 			 */
590 			if (sent) {
591 				nsock->pending = req;
592 				nsock->sent = sent;
593 			}
594 			set_bit(NBD_CMD_REQUEUED, &cmd->flags);
595 			return BLK_STS_RESOURCE;
596 		}
597 		dev_err_ratelimited(disk_to_dev(nbd->disk),
598 			"Send control failed (result %d)\n", result);
599 		return -EAGAIN;
600 	}
601 send_pages:
602 	if (type != NBD_CMD_WRITE)
603 		goto out;
604 
605 	bio = req->bio;
606 	while (bio) {
607 		struct bio *next = bio->bi_next;
608 		struct bvec_iter iter;
609 		struct bio_vec bvec;
610 
611 		bio_for_each_segment(bvec, bio, iter) {
612 			bool is_last = !next && bio_iter_last(bvec, iter);
613 			int flags = is_last ? 0 : MSG_MORE;
614 
615 			dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
616 				req, bvec.bv_len);
617 			iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
618 			if (skip) {
619 				if (skip >= iov_iter_count(&from)) {
620 					skip -= iov_iter_count(&from);
621 					continue;
622 				}
623 				iov_iter_advance(&from, skip);
624 				skip = 0;
625 			}
626 			result = sock_xmit(nbd, index, 1, &from, flags, &sent);
627 			if (result <= 0) {
628 				if (was_interrupted(result)) {
629 					/* We've already sent the header, we
630 					 * have no choice but to set pending and
631 					 * return BUSY.
632 					 */
633 					nsock->pending = req;
634 					nsock->sent = sent;
635 					set_bit(NBD_CMD_REQUEUED, &cmd->flags);
636 					return BLK_STS_RESOURCE;
637 				}
638 				dev_err(disk_to_dev(nbd->disk),
639 					"Send data failed (result %d)\n",
640 					result);
641 				return -EAGAIN;
642 			}
643 			/*
644 			 * The completion might already have come in,
645 			 * so break for the last one instead of letting
646 			 * the iterator do it. This prevents use-after-free
647 			 * of the bio.
648 			 */
649 			if (is_last)
650 				break;
651 		}
652 		bio = next;
653 	}
654 out:
655 	trace_nbd_payload_sent(req, handle);
656 	nsock->pending = NULL;
657 	nsock->sent = 0;
658 	return 0;
659 }
660 
661 /* NULL returned = something went wrong, inform userspace */
662 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
663 {
664 	struct nbd_config *config = nbd->config;
665 	int result;
666 	struct nbd_reply reply;
667 	struct nbd_cmd *cmd;
668 	struct request *req = NULL;
669 	u64 handle;
670 	u16 hwq;
671 	u32 tag;
672 	struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
673 	struct iov_iter to;
674 	int ret = 0;
675 
676 	reply.magic = 0;
677 	iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
678 	result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
679 	if (result <= 0) {
680 		if (!nbd_disconnected(config))
681 			dev_err(disk_to_dev(nbd->disk),
682 				"Receive control failed (result %d)\n", result);
683 		return ERR_PTR(result);
684 	}
685 
686 	if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
687 		dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
688 				(unsigned long)ntohl(reply.magic));
689 		return ERR_PTR(-EPROTO);
690 	}
691 
692 	memcpy(&handle, reply.handle, sizeof(handle));
693 	tag = nbd_handle_to_tag(handle);
694 	hwq = blk_mq_unique_tag_to_hwq(tag);
695 	if (hwq < nbd->tag_set.nr_hw_queues)
696 		req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
697 				       blk_mq_unique_tag_to_tag(tag));
698 	if (!req || !blk_mq_request_started(req)) {
699 		dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
700 			tag, req);
701 		return ERR_PTR(-ENOENT);
702 	}
703 	trace_nbd_header_received(req, handle);
704 	cmd = blk_mq_rq_to_pdu(req);
705 
706 	mutex_lock(&cmd->lock);
707 	if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
708 		dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
709 			req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
710 		ret = -ENOENT;
711 		goto out;
712 	}
713 	if (cmd->status != BLK_STS_OK) {
714 		dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
715 			req);
716 		ret = -ENOENT;
717 		goto out;
718 	}
719 	if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
720 		dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
721 			req);
722 		ret = -ENOENT;
723 		goto out;
724 	}
725 	if (ntohl(reply.error)) {
726 		dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
727 			ntohl(reply.error));
728 		cmd->status = BLK_STS_IOERR;
729 		goto out;
730 	}
731 
732 	dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
733 	if (rq_data_dir(req) != WRITE) {
734 		struct req_iterator iter;
735 		struct bio_vec bvec;
736 
737 		rq_for_each_segment(bvec, req, iter) {
738 			iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
739 			result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
740 			if (result <= 0) {
741 				dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
742 					result);
743 				/*
744 				 * If we've disconnected or we only have 1
745 				 * connection then we need to make sure we
746 				 * complete this request, otherwise error out
747 				 * and let the timeout stuff handle resubmitting
748 				 * this request onto another connection.
749 				 */
750 				if (nbd_disconnected(config) ||
751 				    config->num_connections <= 1) {
752 					cmd->status = BLK_STS_IOERR;
753 					goto out;
754 				}
755 				ret = -EIO;
756 				goto out;
757 			}
758 			dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
759 				req, bvec.bv_len);
760 		}
761 	}
762 out:
763 	trace_nbd_payload_received(req, handle);
764 	mutex_unlock(&cmd->lock);
765 	return ret ? ERR_PTR(ret) : cmd;
766 }
767 
768 static void recv_work(struct work_struct *work)
769 {
770 	struct recv_thread_args *args = container_of(work,
771 						     struct recv_thread_args,
772 						     work);
773 	struct nbd_device *nbd = args->nbd;
774 	struct nbd_config *config = nbd->config;
775 	struct nbd_cmd *cmd;
776 
777 	while (1) {
778 		cmd = nbd_read_stat(nbd, args->index);
779 		if (IS_ERR(cmd)) {
780 			struct nbd_sock *nsock = config->socks[args->index];
781 
782 			mutex_lock(&nsock->tx_lock);
783 			nbd_mark_nsock_dead(nbd, nsock, 1);
784 			mutex_unlock(&nsock->tx_lock);
785 			break;
786 		}
787 
788 		blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
789 	}
790 	atomic_dec(&config->recv_threads);
791 	wake_up(&config->recv_wq);
792 	nbd_config_put(nbd);
793 	kfree(args);
794 }
795 
796 static bool nbd_clear_req(struct request *req, void *data, bool reserved)
797 {
798 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
799 
800 	mutex_lock(&cmd->lock);
801 	cmd->status = BLK_STS_IOERR;
802 	mutex_unlock(&cmd->lock);
803 
804 	blk_mq_complete_request(req);
805 	return true;
806 }
807 
808 static void nbd_clear_que(struct nbd_device *nbd)
809 {
810 	blk_mq_quiesce_queue(nbd->disk->queue);
811 	blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
812 	blk_mq_unquiesce_queue(nbd->disk->queue);
813 	dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
814 }
815 
816 static int find_fallback(struct nbd_device *nbd, int index)
817 {
818 	struct nbd_config *config = nbd->config;
819 	int new_index = -1;
820 	struct nbd_sock *nsock = config->socks[index];
821 	int fallback = nsock->fallback_index;
822 
823 	if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
824 		return new_index;
825 
826 	if (config->num_connections <= 1) {
827 		dev_err_ratelimited(disk_to_dev(nbd->disk),
828 				    "Attempted send on invalid socket\n");
829 		return new_index;
830 	}
831 
832 	if (fallback >= 0 && fallback < config->num_connections &&
833 	    !config->socks[fallback]->dead)
834 		return fallback;
835 
836 	if (nsock->fallback_index < 0 ||
837 	    nsock->fallback_index >= config->num_connections ||
838 	    config->socks[nsock->fallback_index]->dead) {
839 		int i;
840 		for (i = 0; i < config->num_connections; i++) {
841 			if (i == index)
842 				continue;
843 			if (!config->socks[i]->dead) {
844 				new_index = i;
845 				break;
846 			}
847 		}
848 		nsock->fallback_index = new_index;
849 		if (new_index < 0) {
850 			dev_err_ratelimited(disk_to_dev(nbd->disk),
851 					    "Dead connection, failed to find a fallback\n");
852 			return new_index;
853 		}
854 	}
855 	new_index = nsock->fallback_index;
856 	return new_index;
857 }
858 
859 static int wait_for_reconnect(struct nbd_device *nbd)
860 {
861 	struct nbd_config *config = nbd->config;
862 	if (!config->dead_conn_timeout)
863 		return 0;
864 	if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
865 		return 0;
866 	return wait_event_timeout(config->conn_wait,
867 				  atomic_read(&config->live_connections) > 0,
868 				  config->dead_conn_timeout) > 0;
869 }
870 
871 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
872 {
873 	struct request *req = blk_mq_rq_from_pdu(cmd);
874 	struct nbd_device *nbd = cmd->nbd;
875 	struct nbd_config *config;
876 	struct nbd_sock *nsock;
877 	int ret;
878 
879 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
880 		dev_err_ratelimited(disk_to_dev(nbd->disk),
881 				    "Socks array is empty\n");
882 		blk_mq_start_request(req);
883 		return -EINVAL;
884 	}
885 	config = nbd->config;
886 
887 	if (index >= config->num_connections) {
888 		dev_err_ratelimited(disk_to_dev(nbd->disk),
889 				    "Attempted send on invalid socket\n");
890 		nbd_config_put(nbd);
891 		blk_mq_start_request(req);
892 		return -EINVAL;
893 	}
894 	cmd->status = BLK_STS_OK;
895 again:
896 	nsock = config->socks[index];
897 	mutex_lock(&nsock->tx_lock);
898 	if (nsock->dead) {
899 		int old_index = index;
900 		index = find_fallback(nbd, index);
901 		mutex_unlock(&nsock->tx_lock);
902 		if (index < 0) {
903 			if (wait_for_reconnect(nbd)) {
904 				index = old_index;
905 				goto again;
906 			}
907 			/* All the sockets should already be down at this point,
908 			 * we just want to make sure that DISCONNECTED is set so
909 			 * any requests that come in that were queue'ed waiting
910 			 * for the reconnect timer don't trigger the timer again
911 			 * and instead just error out.
912 			 */
913 			sock_shutdown(nbd);
914 			nbd_config_put(nbd);
915 			blk_mq_start_request(req);
916 			return -EIO;
917 		}
918 		goto again;
919 	}
920 
921 	/* Handle the case that we have a pending request that was partially
922 	 * transmitted that _has_ to be serviced first.  We need to call requeue
923 	 * here so that it gets put _after_ the request that is already on the
924 	 * dispatch list.
925 	 */
926 	blk_mq_start_request(req);
927 	if (unlikely(nsock->pending && nsock->pending != req)) {
928 		nbd_requeue_cmd(cmd);
929 		ret = 0;
930 		goto out;
931 	}
932 	/*
933 	 * Some failures are related to the link going down, so anything that
934 	 * returns EAGAIN can be retried on a different socket.
935 	 */
936 	ret = nbd_send_cmd(nbd, cmd, index);
937 	if (ret == -EAGAIN) {
938 		dev_err_ratelimited(disk_to_dev(nbd->disk),
939 				    "Request send failed, requeueing\n");
940 		nbd_mark_nsock_dead(nbd, nsock, 1);
941 		nbd_requeue_cmd(cmd);
942 		ret = 0;
943 	}
944 out:
945 	mutex_unlock(&nsock->tx_lock);
946 	nbd_config_put(nbd);
947 	return ret;
948 }
949 
950 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
951 			const struct blk_mq_queue_data *bd)
952 {
953 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
954 	int ret;
955 
956 	/*
957 	 * Since we look at the bio's to send the request over the network we
958 	 * need to make sure the completion work doesn't mark this request done
959 	 * before we are done doing our send.  This keeps us from dereferencing
960 	 * freed data if we have particularly fast completions (ie we get the
961 	 * completion before we exit sock_xmit on the last bvec) or in the case
962 	 * that the server is misbehaving (or there was an error) before we're
963 	 * done sending everything over the wire.
964 	 */
965 	mutex_lock(&cmd->lock);
966 	clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
967 
968 	/* We can be called directly from the user space process, which means we
969 	 * could possibly have signals pending so our sendmsg will fail.  In
970 	 * this case we need to return that we are busy, otherwise error out as
971 	 * appropriate.
972 	 */
973 	ret = nbd_handle_cmd(cmd, hctx->queue_num);
974 	if (ret < 0)
975 		ret = BLK_STS_IOERR;
976 	else if (!ret)
977 		ret = BLK_STS_OK;
978 	mutex_unlock(&cmd->lock);
979 
980 	return ret;
981 }
982 
983 static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
984 				     int *err)
985 {
986 	struct socket *sock;
987 
988 	*err = 0;
989 	sock = sockfd_lookup(fd, err);
990 	if (!sock)
991 		return NULL;
992 
993 	if (sock->ops->shutdown == sock_no_shutdown) {
994 		dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
995 		*err = -EINVAL;
996 		return NULL;
997 	}
998 
999 	return sock;
1000 }
1001 
1002 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1003 			  bool netlink)
1004 {
1005 	struct nbd_config *config = nbd->config;
1006 	struct socket *sock;
1007 	struct nbd_sock **socks;
1008 	struct nbd_sock *nsock;
1009 	int err;
1010 
1011 	sock = nbd_get_socket(nbd, arg, &err);
1012 	if (!sock)
1013 		return err;
1014 
1015 	if (!netlink && !nbd->task_setup &&
1016 	    !test_bit(NBD_RT_BOUND, &config->runtime_flags))
1017 		nbd->task_setup = current;
1018 
1019 	if (!netlink &&
1020 	    (nbd->task_setup != current ||
1021 	     test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1022 		dev_err(disk_to_dev(nbd->disk),
1023 			"Device being setup by another task");
1024 		sockfd_put(sock);
1025 		return -EBUSY;
1026 	}
1027 
1028 	socks = krealloc(config->socks, (config->num_connections + 1) *
1029 			 sizeof(struct nbd_sock *), GFP_KERNEL);
1030 	if (!socks) {
1031 		sockfd_put(sock);
1032 		return -ENOMEM;
1033 	}
1034 	nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
1035 	if (!nsock) {
1036 		sockfd_put(sock);
1037 		return -ENOMEM;
1038 	}
1039 
1040 	config->socks = socks;
1041 
1042 	nsock->fallback_index = -1;
1043 	nsock->dead = false;
1044 	mutex_init(&nsock->tx_lock);
1045 	nsock->sock = sock;
1046 	nsock->pending = NULL;
1047 	nsock->sent = 0;
1048 	nsock->cookie = 0;
1049 	socks[config->num_connections++] = nsock;
1050 	atomic_inc(&config->live_connections);
1051 
1052 	return 0;
1053 }
1054 
1055 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1056 {
1057 	struct nbd_config *config = nbd->config;
1058 	struct socket *sock, *old;
1059 	struct recv_thread_args *args;
1060 	int i;
1061 	int err;
1062 
1063 	sock = nbd_get_socket(nbd, arg, &err);
1064 	if (!sock)
1065 		return err;
1066 
1067 	args = kzalloc(sizeof(*args), GFP_KERNEL);
1068 	if (!args) {
1069 		sockfd_put(sock);
1070 		return -ENOMEM;
1071 	}
1072 
1073 	for (i = 0; i < config->num_connections; i++) {
1074 		struct nbd_sock *nsock = config->socks[i];
1075 
1076 		if (!nsock->dead)
1077 			continue;
1078 
1079 		mutex_lock(&nsock->tx_lock);
1080 		if (!nsock->dead) {
1081 			mutex_unlock(&nsock->tx_lock);
1082 			continue;
1083 		}
1084 		sk_set_memalloc(sock->sk);
1085 		if (nbd->tag_set.timeout)
1086 			sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1087 		atomic_inc(&config->recv_threads);
1088 		refcount_inc(&nbd->config_refs);
1089 		old = nsock->sock;
1090 		nsock->fallback_index = -1;
1091 		nsock->sock = sock;
1092 		nsock->dead = false;
1093 		INIT_WORK(&args->work, recv_work);
1094 		args->index = i;
1095 		args->nbd = nbd;
1096 		nsock->cookie++;
1097 		mutex_unlock(&nsock->tx_lock);
1098 		sockfd_put(old);
1099 
1100 		clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1101 
1102 		/* We take the tx_mutex in an error path in the recv_work, so we
1103 		 * need to queue_work outside of the tx_mutex.
1104 		 */
1105 		queue_work(nbd->recv_workq, &args->work);
1106 
1107 		atomic_inc(&config->live_connections);
1108 		wake_up(&config->conn_wait);
1109 		return 0;
1110 	}
1111 	sockfd_put(sock);
1112 	kfree(args);
1113 	return -ENOSPC;
1114 }
1115 
1116 static void nbd_bdev_reset(struct block_device *bdev)
1117 {
1118 	if (bdev->bd_openers > 1)
1119 		return;
1120 	bd_set_size(bdev, 0);
1121 }
1122 
1123 static void nbd_parse_flags(struct nbd_device *nbd)
1124 {
1125 	struct nbd_config *config = nbd->config;
1126 	if (config->flags & NBD_FLAG_READ_ONLY)
1127 		set_disk_ro(nbd->disk, true);
1128 	else
1129 		set_disk_ro(nbd->disk, false);
1130 	if (config->flags & NBD_FLAG_SEND_TRIM)
1131 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1132 	if (config->flags & NBD_FLAG_SEND_FLUSH) {
1133 		if (config->flags & NBD_FLAG_SEND_FUA)
1134 			blk_queue_write_cache(nbd->disk->queue, true, true);
1135 		else
1136 			blk_queue_write_cache(nbd->disk->queue, true, false);
1137 	}
1138 	else
1139 		blk_queue_write_cache(nbd->disk->queue, false, false);
1140 }
1141 
1142 static void send_disconnects(struct nbd_device *nbd)
1143 {
1144 	struct nbd_config *config = nbd->config;
1145 	struct nbd_request request = {
1146 		.magic = htonl(NBD_REQUEST_MAGIC),
1147 		.type = htonl(NBD_CMD_DISC),
1148 	};
1149 	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1150 	struct iov_iter from;
1151 	int i, ret;
1152 
1153 	for (i = 0; i < config->num_connections; i++) {
1154 		struct nbd_sock *nsock = config->socks[i];
1155 
1156 		iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1157 		mutex_lock(&nsock->tx_lock);
1158 		ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1159 		if (ret <= 0)
1160 			dev_err(disk_to_dev(nbd->disk),
1161 				"Send disconnect failed %d\n", ret);
1162 		mutex_unlock(&nsock->tx_lock);
1163 	}
1164 }
1165 
1166 static int nbd_disconnect(struct nbd_device *nbd)
1167 {
1168 	struct nbd_config *config = nbd->config;
1169 
1170 	dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1171 	set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1172 	set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1173 	send_disconnects(nbd);
1174 	return 0;
1175 }
1176 
1177 static void nbd_clear_sock(struct nbd_device *nbd)
1178 {
1179 	sock_shutdown(nbd);
1180 	nbd_clear_que(nbd);
1181 	nbd->task_setup = NULL;
1182 }
1183 
1184 static void nbd_config_put(struct nbd_device *nbd)
1185 {
1186 	if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1187 					&nbd->config_lock)) {
1188 		struct nbd_config *config = nbd->config;
1189 		nbd_dev_dbg_close(nbd);
1190 		nbd_size_clear(nbd);
1191 		if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1192 				       &config->runtime_flags))
1193 			device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1194 		nbd->task_recv = NULL;
1195 		nbd_clear_sock(nbd);
1196 		if (config->num_connections) {
1197 			int i;
1198 			for (i = 0; i < config->num_connections; i++) {
1199 				sockfd_put(config->socks[i]->sock);
1200 				kfree(config->socks[i]);
1201 			}
1202 			kfree(config->socks);
1203 		}
1204 		kfree(nbd->config);
1205 		nbd->config = NULL;
1206 
1207 		if (nbd->recv_workq)
1208 			destroy_workqueue(nbd->recv_workq);
1209 		nbd->recv_workq = NULL;
1210 
1211 		nbd->tag_set.timeout = 0;
1212 		nbd->disk->queue->limits.discard_granularity = 0;
1213 		nbd->disk->queue->limits.discard_alignment = 0;
1214 		blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1215 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1216 
1217 		mutex_unlock(&nbd->config_lock);
1218 		nbd_put(nbd);
1219 		module_put(THIS_MODULE);
1220 	}
1221 }
1222 
1223 static int nbd_start_device(struct nbd_device *nbd)
1224 {
1225 	struct nbd_config *config = nbd->config;
1226 	int num_connections = config->num_connections;
1227 	int error = 0, i;
1228 
1229 	if (nbd->task_recv)
1230 		return -EBUSY;
1231 	if (!config->socks)
1232 		return -EINVAL;
1233 	if (num_connections > 1 &&
1234 	    !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1235 		dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1236 		return -EINVAL;
1237 	}
1238 
1239 	nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1240 					  WQ_MEM_RECLAIM | WQ_HIGHPRI |
1241 					  WQ_UNBOUND, 0, nbd->index);
1242 	if (!nbd->recv_workq) {
1243 		dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1244 		return -ENOMEM;
1245 	}
1246 
1247 	blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1248 	nbd->task_recv = current;
1249 
1250 	nbd_parse_flags(nbd);
1251 
1252 	error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1253 	if (error) {
1254 		dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1255 		return error;
1256 	}
1257 	set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1258 
1259 	nbd_dev_dbg_init(nbd);
1260 	for (i = 0; i < num_connections; i++) {
1261 		struct recv_thread_args *args;
1262 
1263 		args = kzalloc(sizeof(*args), GFP_KERNEL);
1264 		if (!args) {
1265 			sock_shutdown(nbd);
1266 			return -ENOMEM;
1267 		}
1268 		sk_set_memalloc(config->socks[i]->sock->sk);
1269 		if (nbd->tag_set.timeout)
1270 			config->socks[i]->sock->sk->sk_sndtimeo =
1271 				nbd->tag_set.timeout;
1272 		atomic_inc(&config->recv_threads);
1273 		refcount_inc(&nbd->config_refs);
1274 		INIT_WORK(&args->work, recv_work);
1275 		args->nbd = nbd;
1276 		args->index = i;
1277 		queue_work(nbd->recv_workq, &args->work);
1278 	}
1279 	nbd_size_update(nbd);
1280 	return error;
1281 }
1282 
1283 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1284 {
1285 	struct nbd_config *config = nbd->config;
1286 	int ret;
1287 
1288 	ret = nbd_start_device(nbd);
1289 	if (ret)
1290 		return ret;
1291 
1292 	if (max_part)
1293 		bdev->bd_invalidated = 1;
1294 	mutex_unlock(&nbd->config_lock);
1295 	ret = wait_event_interruptible(config->recv_wq,
1296 					 atomic_read(&config->recv_threads) == 0);
1297 	if (ret) {
1298 		sock_shutdown(nbd);
1299 		flush_workqueue(nbd->recv_workq);
1300 	}
1301 	mutex_lock(&nbd->config_lock);
1302 	nbd_bdev_reset(bdev);
1303 	/* user requested, ignore socket errors */
1304 	if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1305 		ret = 0;
1306 	if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1307 		ret = -ETIMEDOUT;
1308 	return ret;
1309 }
1310 
1311 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1312 				 struct block_device *bdev)
1313 {
1314 	sock_shutdown(nbd);
1315 	__invalidate_device(bdev, true);
1316 	nbd_bdev_reset(bdev);
1317 	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1318 			       &nbd->config->runtime_flags))
1319 		nbd_config_put(nbd);
1320 }
1321 
1322 static bool nbd_is_valid_blksize(unsigned long blksize)
1323 {
1324 	if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1325 	    blksize > PAGE_SIZE)
1326 		return false;
1327 	return true;
1328 }
1329 
1330 static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1331 {
1332 	nbd->tag_set.timeout = timeout * HZ;
1333 	if (timeout)
1334 		blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1335 }
1336 
1337 /* Must be called with config_lock held */
1338 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1339 		       unsigned int cmd, unsigned long arg)
1340 {
1341 	struct nbd_config *config = nbd->config;
1342 
1343 	switch (cmd) {
1344 	case NBD_DISCONNECT:
1345 		return nbd_disconnect(nbd);
1346 	case NBD_CLEAR_SOCK:
1347 		nbd_clear_sock_ioctl(nbd, bdev);
1348 		return 0;
1349 	case NBD_SET_SOCK:
1350 		return nbd_add_socket(nbd, arg, false);
1351 	case NBD_SET_BLKSIZE:
1352 		if (!arg)
1353 			arg = NBD_DEF_BLKSIZE;
1354 		if (!nbd_is_valid_blksize(arg))
1355 			return -EINVAL;
1356 		nbd_size_set(nbd, arg,
1357 			     div_s64(config->bytesize, arg));
1358 		return 0;
1359 	case NBD_SET_SIZE:
1360 		nbd_size_set(nbd, config->blksize,
1361 			     div_s64(arg, config->blksize));
1362 		return 0;
1363 	case NBD_SET_SIZE_BLOCKS:
1364 		nbd_size_set(nbd, config->blksize, arg);
1365 		return 0;
1366 	case NBD_SET_TIMEOUT:
1367 		nbd_set_cmd_timeout(nbd, arg);
1368 		return 0;
1369 
1370 	case NBD_SET_FLAGS:
1371 		config->flags = arg;
1372 		return 0;
1373 	case NBD_DO_IT:
1374 		return nbd_start_device_ioctl(nbd, bdev);
1375 	case NBD_CLEAR_QUE:
1376 		/*
1377 		 * This is for compatibility only.  The queue is always cleared
1378 		 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1379 		 */
1380 		return 0;
1381 	case NBD_PRINT_DEBUG:
1382 		/*
1383 		 * For compatibility only, we no longer keep a list of
1384 		 * outstanding requests.
1385 		 */
1386 		return 0;
1387 	}
1388 	return -ENOTTY;
1389 }
1390 
1391 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1392 		     unsigned int cmd, unsigned long arg)
1393 {
1394 	struct nbd_device *nbd = bdev->bd_disk->private_data;
1395 	struct nbd_config *config = nbd->config;
1396 	int error = -EINVAL;
1397 
1398 	if (!capable(CAP_SYS_ADMIN))
1399 		return -EPERM;
1400 
1401 	/* The block layer will pass back some non-nbd ioctls in case we have
1402 	 * special handling for them, but we don't so just return an error.
1403 	 */
1404 	if (_IOC_TYPE(cmd) != 0xab)
1405 		return -EINVAL;
1406 
1407 	mutex_lock(&nbd->config_lock);
1408 
1409 	/* Don't allow ioctl operations on a nbd device that was created with
1410 	 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1411 	 */
1412 	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1413 	    (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1414 		error = __nbd_ioctl(bdev, nbd, cmd, arg);
1415 	else
1416 		dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1417 	mutex_unlock(&nbd->config_lock);
1418 	return error;
1419 }
1420 
1421 static struct nbd_config *nbd_alloc_config(void)
1422 {
1423 	struct nbd_config *config;
1424 
1425 	config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1426 	if (!config)
1427 		return NULL;
1428 	atomic_set(&config->recv_threads, 0);
1429 	init_waitqueue_head(&config->recv_wq);
1430 	init_waitqueue_head(&config->conn_wait);
1431 	config->blksize = NBD_DEF_BLKSIZE;
1432 	atomic_set(&config->live_connections, 0);
1433 	try_module_get(THIS_MODULE);
1434 	return config;
1435 }
1436 
1437 static int nbd_open(struct block_device *bdev, fmode_t mode)
1438 {
1439 	struct nbd_device *nbd;
1440 	int ret = 0;
1441 
1442 	mutex_lock(&nbd_index_mutex);
1443 	nbd = bdev->bd_disk->private_data;
1444 	if (!nbd) {
1445 		ret = -ENXIO;
1446 		goto out;
1447 	}
1448 	if (!refcount_inc_not_zero(&nbd->refs)) {
1449 		ret = -ENXIO;
1450 		goto out;
1451 	}
1452 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
1453 		struct nbd_config *config;
1454 
1455 		mutex_lock(&nbd->config_lock);
1456 		if (refcount_inc_not_zero(&nbd->config_refs)) {
1457 			mutex_unlock(&nbd->config_lock);
1458 			goto out;
1459 		}
1460 		config = nbd->config = nbd_alloc_config();
1461 		if (!config) {
1462 			ret = -ENOMEM;
1463 			mutex_unlock(&nbd->config_lock);
1464 			goto out;
1465 		}
1466 		refcount_set(&nbd->config_refs, 1);
1467 		refcount_inc(&nbd->refs);
1468 		mutex_unlock(&nbd->config_lock);
1469 		bdev->bd_invalidated = 1;
1470 	} else if (nbd_disconnected(nbd->config)) {
1471 		bdev->bd_invalidated = 1;
1472 	}
1473 out:
1474 	mutex_unlock(&nbd_index_mutex);
1475 	return ret;
1476 }
1477 
1478 static void nbd_release(struct gendisk *disk, fmode_t mode)
1479 {
1480 	struct nbd_device *nbd = disk->private_data;
1481 	struct block_device *bdev = bdget_disk(disk, 0);
1482 
1483 	if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1484 			bdev->bd_openers == 0)
1485 		nbd_disconnect_and_put(nbd);
1486 
1487 	nbd_config_put(nbd);
1488 	nbd_put(nbd);
1489 }
1490 
1491 static const struct block_device_operations nbd_fops =
1492 {
1493 	.owner =	THIS_MODULE,
1494 	.open =		nbd_open,
1495 	.release =	nbd_release,
1496 	.ioctl =	nbd_ioctl,
1497 	.compat_ioctl =	nbd_ioctl,
1498 };
1499 
1500 #if IS_ENABLED(CONFIG_DEBUG_FS)
1501 
1502 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1503 {
1504 	struct nbd_device *nbd = s->private;
1505 
1506 	if (nbd->task_recv)
1507 		seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1508 
1509 	return 0;
1510 }
1511 
1512 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1513 {
1514 	return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1515 }
1516 
1517 static const struct file_operations nbd_dbg_tasks_ops = {
1518 	.open = nbd_dbg_tasks_open,
1519 	.read = seq_read,
1520 	.llseek = seq_lseek,
1521 	.release = single_release,
1522 };
1523 
1524 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1525 {
1526 	struct nbd_device *nbd = s->private;
1527 	u32 flags = nbd->config->flags;
1528 
1529 	seq_printf(s, "Hex: 0x%08x\n\n", flags);
1530 
1531 	seq_puts(s, "Known flags:\n");
1532 
1533 	if (flags & NBD_FLAG_HAS_FLAGS)
1534 		seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1535 	if (flags & NBD_FLAG_READ_ONLY)
1536 		seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1537 	if (flags & NBD_FLAG_SEND_FLUSH)
1538 		seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1539 	if (flags & NBD_FLAG_SEND_FUA)
1540 		seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1541 	if (flags & NBD_FLAG_SEND_TRIM)
1542 		seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1543 
1544 	return 0;
1545 }
1546 
1547 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1548 {
1549 	return single_open(file, nbd_dbg_flags_show, inode->i_private);
1550 }
1551 
1552 static const struct file_operations nbd_dbg_flags_ops = {
1553 	.open = nbd_dbg_flags_open,
1554 	.read = seq_read,
1555 	.llseek = seq_lseek,
1556 	.release = single_release,
1557 };
1558 
1559 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1560 {
1561 	struct dentry *dir;
1562 	struct nbd_config *config = nbd->config;
1563 
1564 	if (!nbd_dbg_dir)
1565 		return -EIO;
1566 
1567 	dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1568 	if (!dir) {
1569 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1570 			nbd_name(nbd));
1571 		return -EIO;
1572 	}
1573 	config->dbg_dir = dir;
1574 
1575 	debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1576 	debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1577 	debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1578 	debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1579 	debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1580 
1581 	return 0;
1582 }
1583 
1584 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1585 {
1586 	debugfs_remove_recursive(nbd->config->dbg_dir);
1587 }
1588 
1589 static int nbd_dbg_init(void)
1590 {
1591 	struct dentry *dbg_dir;
1592 
1593 	dbg_dir = debugfs_create_dir("nbd", NULL);
1594 	if (!dbg_dir)
1595 		return -EIO;
1596 
1597 	nbd_dbg_dir = dbg_dir;
1598 
1599 	return 0;
1600 }
1601 
1602 static void nbd_dbg_close(void)
1603 {
1604 	debugfs_remove_recursive(nbd_dbg_dir);
1605 }
1606 
1607 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1608 
1609 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1610 {
1611 	return 0;
1612 }
1613 
1614 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1615 {
1616 }
1617 
1618 static int nbd_dbg_init(void)
1619 {
1620 	return 0;
1621 }
1622 
1623 static void nbd_dbg_close(void)
1624 {
1625 }
1626 
1627 #endif
1628 
1629 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1630 			    unsigned int hctx_idx, unsigned int numa_node)
1631 {
1632 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1633 	cmd->nbd = set->driver_data;
1634 	cmd->flags = 0;
1635 	mutex_init(&cmd->lock);
1636 	return 0;
1637 }
1638 
1639 static const struct blk_mq_ops nbd_mq_ops = {
1640 	.queue_rq	= nbd_queue_rq,
1641 	.complete	= nbd_complete_rq,
1642 	.init_request	= nbd_init_request,
1643 	.timeout	= nbd_xmit_timeout,
1644 };
1645 
1646 static int nbd_dev_add(int index)
1647 {
1648 	struct nbd_device *nbd;
1649 	struct gendisk *disk;
1650 	struct request_queue *q;
1651 	int err = -ENOMEM;
1652 
1653 	nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1654 	if (!nbd)
1655 		goto out;
1656 
1657 	disk = alloc_disk(1 << part_shift);
1658 	if (!disk)
1659 		goto out_free_nbd;
1660 
1661 	if (index >= 0) {
1662 		err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1663 				GFP_KERNEL);
1664 		if (err == -ENOSPC)
1665 			err = -EEXIST;
1666 	} else {
1667 		err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1668 		if (err >= 0)
1669 			index = err;
1670 	}
1671 	if (err < 0)
1672 		goto out_free_disk;
1673 
1674 	nbd->index = index;
1675 	nbd->disk = disk;
1676 	nbd->tag_set.ops = &nbd_mq_ops;
1677 	nbd->tag_set.nr_hw_queues = 1;
1678 	nbd->tag_set.queue_depth = 128;
1679 	nbd->tag_set.numa_node = NUMA_NO_NODE;
1680 	nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1681 	nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1682 		BLK_MQ_F_BLOCKING;
1683 	nbd->tag_set.driver_data = nbd;
1684 	nbd->destroy_complete = NULL;
1685 
1686 	err = blk_mq_alloc_tag_set(&nbd->tag_set);
1687 	if (err)
1688 		goto out_free_idr;
1689 
1690 	q = blk_mq_init_queue(&nbd->tag_set);
1691 	if (IS_ERR(q)) {
1692 		err = PTR_ERR(q);
1693 		goto out_free_tags;
1694 	}
1695 	disk->queue = q;
1696 
1697 	/*
1698 	 * Tell the block layer that we are not a rotational device
1699 	 */
1700 	blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1701 	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1702 	disk->queue->limits.discard_granularity = 0;
1703 	disk->queue->limits.discard_alignment = 0;
1704 	blk_queue_max_discard_sectors(disk->queue, 0);
1705 	blk_queue_max_segment_size(disk->queue, UINT_MAX);
1706 	blk_queue_max_segments(disk->queue, USHRT_MAX);
1707 	blk_queue_max_hw_sectors(disk->queue, 65536);
1708 	disk->queue->limits.max_sectors = 256;
1709 
1710 	mutex_init(&nbd->config_lock);
1711 	refcount_set(&nbd->config_refs, 0);
1712 	refcount_set(&nbd->refs, 1);
1713 	INIT_LIST_HEAD(&nbd->list);
1714 	disk->major = NBD_MAJOR;
1715 	disk->first_minor = index << part_shift;
1716 	disk->fops = &nbd_fops;
1717 	disk->private_data = nbd;
1718 	sprintf(disk->disk_name, "nbd%d", index);
1719 	add_disk(disk);
1720 	nbd_total_devices++;
1721 	return index;
1722 
1723 out_free_tags:
1724 	blk_mq_free_tag_set(&nbd->tag_set);
1725 out_free_idr:
1726 	idr_remove(&nbd_index_idr, index);
1727 out_free_disk:
1728 	put_disk(disk);
1729 out_free_nbd:
1730 	kfree(nbd);
1731 out:
1732 	return err;
1733 }
1734 
1735 static int find_free_cb(int id, void *ptr, void *data)
1736 {
1737 	struct nbd_device *nbd = ptr;
1738 	struct nbd_device **found = data;
1739 
1740 	if (!refcount_read(&nbd->config_refs)) {
1741 		*found = nbd;
1742 		return 1;
1743 	}
1744 	return 0;
1745 }
1746 
1747 /* Netlink interface. */
1748 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1749 	[NBD_ATTR_INDEX]		=	{ .type = NLA_U32 },
1750 	[NBD_ATTR_SIZE_BYTES]		=	{ .type = NLA_U64 },
1751 	[NBD_ATTR_BLOCK_SIZE_BYTES]	=	{ .type = NLA_U64 },
1752 	[NBD_ATTR_TIMEOUT]		=	{ .type = NLA_U64 },
1753 	[NBD_ATTR_SERVER_FLAGS]		=	{ .type = NLA_U64 },
1754 	[NBD_ATTR_CLIENT_FLAGS]		=	{ .type = NLA_U64 },
1755 	[NBD_ATTR_SOCKETS]		=	{ .type = NLA_NESTED},
1756 	[NBD_ATTR_DEAD_CONN_TIMEOUT]	=	{ .type = NLA_U64 },
1757 	[NBD_ATTR_DEVICE_LIST]		=	{ .type = NLA_NESTED},
1758 };
1759 
1760 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1761 	[NBD_SOCK_FD]			=	{ .type = NLA_U32 },
1762 };
1763 
1764 /* We don't use this right now since we don't parse the incoming list, but we
1765  * still want it here so userspace knows what to expect.
1766  */
1767 static const struct nla_policy __attribute__((unused))
1768 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1769 	[NBD_DEVICE_INDEX]		=	{ .type = NLA_U32 },
1770 	[NBD_DEVICE_CONNECTED]		=	{ .type = NLA_U8 },
1771 };
1772 
1773 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1774 {
1775 	struct nbd_config *config = nbd->config;
1776 	u64 bsize = config->blksize;
1777 	u64 bytes = config->bytesize;
1778 
1779 	if (info->attrs[NBD_ATTR_SIZE_BYTES])
1780 		bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1781 
1782 	if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1783 		bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1784 		if (!bsize)
1785 			bsize = NBD_DEF_BLKSIZE;
1786 		if (!nbd_is_valid_blksize(bsize)) {
1787 			printk(KERN_ERR "Invalid block size %llu\n", bsize);
1788 			return -EINVAL;
1789 		}
1790 	}
1791 
1792 	if (bytes != config->bytesize || bsize != config->blksize)
1793 		nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1794 	return 0;
1795 }
1796 
1797 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1798 {
1799 	DECLARE_COMPLETION_ONSTACK(destroy_complete);
1800 	struct nbd_device *nbd = NULL;
1801 	struct nbd_config *config;
1802 	int index = -1;
1803 	int ret;
1804 	bool put_dev = false;
1805 
1806 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1807 		return -EPERM;
1808 
1809 	if (info->attrs[NBD_ATTR_INDEX])
1810 		index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1811 	if (!info->attrs[NBD_ATTR_SOCKETS]) {
1812 		printk(KERN_ERR "nbd: must specify at least one socket\n");
1813 		return -EINVAL;
1814 	}
1815 	if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1816 		printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1817 		return -EINVAL;
1818 	}
1819 again:
1820 	mutex_lock(&nbd_index_mutex);
1821 	if (index == -1) {
1822 		ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1823 		if (ret == 0) {
1824 			int new_index;
1825 			new_index = nbd_dev_add(-1);
1826 			if (new_index < 0) {
1827 				mutex_unlock(&nbd_index_mutex);
1828 				printk(KERN_ERR "nbd: failed to add new device\n");
1829 				return new_index;
1830 			}
1831 			nbd = idr_find(&nbd_index_idr, new_index);
1832 		}
1833 	} else {
1834 		nbd = idr_find(&nbd_index_idr, index);
1835 		if (!nbd) {
1836 			ret = nbd_dev_add(index);
1837 			if (ret < 0) {
1838 				mutex_unlock(&nbd_index_mutex);
1839 				printk(KERN_ERR "nbd: failed to add new device\n");
1840 				return ret;
1841 			}
1842 			nbd = idr_find(&nbd_index_idr, index);
1843 		}
1844 	}
1845 	if (!nbd) {
1846 		printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1847 		       index);
1848 		mutex_unlock(&nbd_index_mutex);
1849 		return -EINVAL;
1850 	}
1851 
1852 	if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1853 	    test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1854 		nbd->destroy_complete = &destroy_complete;
1855 		mutex_unlock(&nbd_index_mutex);
1856 
1857 		/* Wait untill the the nbd stuff is totally destroyed */
1858 		wait_for_completion(&destroy_complete);
1859 		goto again;
1860 	}
1861 
1862 	if (!refcount_inc_not_zero(&nbd->refs)) {
1863 		mutex_unlock(&nbd_index_mutex);
1864 		if (index == -1)
1865 			goto again;
1866 		printk(KERN_ERR "nbd: device at index %d is going down\n",
1867 		       index);
1868 		return -EINVAL;
1869 	}
1870 	mutex_unlock(&nbd_index_mutex);
1871 
1872 	mutex_lock(&nbd->config_lock);
1873 	if (refcount_read(&nbd->config_refs)) {
1874 		mutex_unlock(&nbd->config_lock);
1875 		nbd_put(nbd);
1876 		if (index == -1)
1877 			goto again;
1878 		printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1879 		return -EBUSY;
1880 	}
1881 	if (WARN_ON(nbd->config)) {
1882 		mutex_unlock(&nbd->config_lock);
1883 		nbd_put(nbd);
1884 		return -EINVAL;
1885 	}
1886 	config = nbd->config = nbd_alloc_config();
1887 	if (!nbd->config) {
1888 		mutex_unlock(&nbd->config_lock);
1889 		nbd_put(nbd);
1890 		printk(KERN_ERR "nbd: couldn't allocate config\n");
1891 		return -ENOMEM;
1892 	}
1893 	refcount_set(&nbd->config_refs, 1);
1894 	set_bit(NBD_RT_BOUND, &config->runtime_flags);
1895 
1896 	ret = nbd_genl_size_set(info, nbd);
1897 	if (ret)
1898 		goto out;
1899 
1900 	if (info->attrs[NBD_ATTR_TIMEOUT])
1901 		nbd_set_cmd_timeout(nbd,
1902 				    nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1903 	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1904 		config->dead_conn_timeout =
1905 			nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1906 		config->dead_conn_timeout *= HZ;
1907 	}
1908 	if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1909 		config->flags =
1910 			nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1911 	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1912 		u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1913 		if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1914 			set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1915 				&config->runtime_flags);
1916 			set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1917 			put_dev = true;
1918 		} else {
1919 			clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1920 		}
1921 		if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1922 			set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1923 				&config->runtime_flags);
1924 		}
1925 	}
1926 
1927 	if (info->attrs[NBD_ATTR_SOCKETS]) {
1928 		struct nlattr *attr;
1929 		int rem, fd;
1930 
1931 		nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1932 				    rem) {
1933 			struct nlattr *socks[NBD_SOCK_MAX+1];
1934 
1935 			if (nla_type(attr) != NBD_SOCK_ITEM) {
1936 				printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1937 				ret = -EINVAL;
1938 				goto out;
1939 			}
1940 			ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1941 							  attr,
1942 							  nbd_sock_policy,
1943 							  info->extack);
1944 			if (ret != 0) {
1945 				printk(KERN_ERR "nbd: error processing sock list\n");
1946 				ret = -EINVAL;
1947 				goto out;
1948 			}
1949 			if (!socks[NBD_SOCK_FD])
1950 				continue;
1951 			fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1952 			ret = nbd_add_socket(nbd, fd, true);
1953 			if (ret)
1954 				goto out;
1955 		}
1956 	}
1957 	ret = nbd_start_device(nbd);
1958 out:
1959 	mutex_unlock(&nbd->config_lock);
1960 	if (!ret) {
1961 		set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1962 		refcount_inc(&nbd->config_refs);
1963 		nbd_connect_reply(info, nbd->index);
1964 	}
1965 	nbd_config_put(nbd);
1966 	if (put_dev)
1967 		nbd_put(nbd);
1968 	return ret;
1969 }
1970 
1971 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1972 {
1973 	mutex_lock(&nbd->config_lock);
1974 	nbd_disconnect(nbd);
1975 	nbd_clear_sock(nbd);
1976 	mutex_unlock(&nbd->config_lock);
1977 	/*
1978 	 * Make sure recv thread has finished, so it does not drop the last
1979 	 * config ref and try to destroy the workqueue from inside the work
1980 	 * queue.
1981 	 */
1982 	flush_workqueue(nbd->recv_workq);
1983 	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1984 			       &nbd->config->runtime_flags))
1985 		nbd_config_put(nbd);
1986 }
1987 
1988 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1989 {
1990 	struct nbd_device *nbd;
1991 	int index;
1992 
1993 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1994 		return -EPERM;
1995 
1996 	if (!info->attrs[NBD_ATTR_INDEX]) {
1997 		printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1998 		return -EINVAL;
1999 	}
2000 	index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2001 	mutex_lock(&nbd_index_mutex);
2002 	nbd = idr_find(&nbd_index_idr, index);
2003 	if (!nbd) {
2004 		mutex_unlock(&nbd_index_mutex);
2005 		printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2006 		       index);
2007 		return -EINVAL;
2008 	}
2009 	if (!refcount_inc_not_zero(&nbd->refs)) {
2010 		mutex_unlock(&nbd_index_mutex);
2011 		printk(KERN_ERR "nbd: device at index %d is going down\n",
2012 		       index);
2013 		return -EINVAL;
2014 	}
2015 	mutex_unlock(&nbd_index_mutex);
2016 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
2017 		nbd_put(nbd);
2018 		return 0;
2019 	}
2020 	nbd_disconnect_and_put(nbd);
2021 	nbd_config_put(nbd);
2022 	nbd_put(nbd);
2023 	return 0;
2024 }
2025 
2026 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2027 {
2028 	struct nbd_device *nbd = NULL;
2029 	struct nbd_config *config;
2030 	int index;
2031 	int ret = 0;
2032 	bool put_dev = false;
2033 
2034 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
2035 		return -EPERM;
2036 
2037 	if (!info->attrs[NBD_ATTR_INDEX]) {
2038 		printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2039 		return -EINVAL;
2040 	}
2041 	index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2042 	mutex_lock(&nbd_index_mutex);
2043 	nbd = idr_find(&nbd_index_idr, index);
2044 	if (!nbd) {
2045 		mutex_unlock(&nbd_index_mutex);
2046 		printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2047 		       index);
2048 		return -EINVAL;
2049 	}
2050 	if (!refcount_inc_not_zero(&nbd->refs)) {
2051 		mutex_unlock(&nbd_index_mutex);
2052 		printk(KERN_ERR "nbd: device at index %d is going down\n",
2053 		       index);
2054 		return -EINVAL;
2055 	}
2056 	mutex_unlock(&nbd_index_mutex);
2057 
2058 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
2059 		dev_err(nbd_to_dev(nbd),
2060 			"not configured, cannot reconfigure\n");
2061 		nbd_put(nbd);
2062 		return -EINVAL;
2063 	}
2064 
2065 	mutex_lock(&nbd->config_lock);
2066 	config = nbd->config;
2067 	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2068 	    !nbd->task_recv) {
2069 		dev_err(nbd_to_dev(nbd),
2070 			"not configured, cannot reconfigure\n");
2071 		ret = -EINVAL;
2072 		goto out;
2073 	}
2074 
2075 	ret = nbd_genl_size_set(info, nbd);
2076 	if (ret)
2077 		goto out;
2078 
2079 	if (info->attrs[NBD_ATTR_TIMEOUT])
2080 		nbd_set_cmd_timeout(nbd,
2081 				    nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2082 	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2083 		config->dead_conn_timeout =
2084 			nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2085 		config->dead_conn_timeout *= HZ;
2086 	}
2087 	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2088 		u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2089 		if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2090 			if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2091 					      &config->runtime_flags))
2092 				put_dev = true;
2093 			set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2094 		} else {
2095 			if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2096 					       &config->runtime_flags))
2097 				refcount_inc(&nbd->refs);
2098 			clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2099 		}
2100 
2101 		if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2102 			set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2103 					&config->runtime_flags);
2104 		} else {
2105 			clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2106 					&config->runtime_flags);
2107 		}
2108 	}
2109 
2110 	if (info->attrs[NBD_ATTR_SOCKETS]) {
2111 		struct nlattr *attr;
2112 		int rem, fd;
2113 
2114 		nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2115 				    rem) {
2116 			struct nlattr *socks[NBD_SOCK_MAX+1];
2117 
2118 			if (nla_type(attr) != NBD_SOCK_ITEM) {
2119 				printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2120 				ret = -EINVAL;
2121 				goto out;
2122 			}
2123 			ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2124 							  attr,
2125 							  nbd_sock_policy,
2126 							  info->extack);
2127 			if (ret != 0) {
2128 				printk(KERN_ERR "nbd: error processing sock list\n");
2129 				ret = -EINVAL;
2130 				goto out;
2131 			}
2132 			if (!socks[NBD_SOCK_FD])
2133 				continue;
2134 			fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2135 			ret = nbd_reconnect_socket(nbd, fd);
2136 			if (ret) {
2137 				if (ret == -ENOSPC)
2138 					ret = 0;
2139 				goto out;
2140 			}
2141 			dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2142 		}
2143 	}
2144 out:
2145 	mutex_unlock(&nbd->config_lock);
2146 	nbd_config_put(nbd);
2147 	nbd_put(nbd);
2148 	if (put_dev)
2149 		nbd_put(nbd);
2150 	return ret;
2151 }
2152 
2153 static const struct genl_ops nbd_connect_genl_ops[] = {
2154 	{
2155 		.cmd	= NBD_CMD_CONNECT,
2156 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2157 		.doit	= nbd_genl_connect,
2158 	},
2159 	{
2160 		.cmd	= NBD_CMD_DISCONNECT,
2161 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2162 		.doit	= nbd_genl_disconnect,
2163 	},
2164 	{
2165 		.cmd	= NBD_CMD_RECONFIGURE,
2166 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2167 		.doit	= nbd_genl_reconfigure,
2168 	},
2169 	{
2170 		.cmd	= NBD_CMD_STATUS,
2171 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2172 		.doit	= nbd_genl_status,
2173 	},
2174 };
2175 
2176 static const struct genl_multicast_group nbd_mcast_grps[] = {
2177 	{ .name = NBD_GENL_MCAST_GROUP_NAME, },
2178 };
2179 
2180 static struct genl_family nbd_genl_family __ro_after_init = {
2181 	.hdrsize	= 0,
2182 	.name		= NBD_GENL_FAMILY_NAME,
2183 	.version	= NBD_GENL_VERSION,
2184 	.module		= THIS_MODULE,
2185 	.ops		= nbd_connect_genl_ops,
2186 	.n_ops		= ARRAY_SIZE(nbd_connect_genl_ops),
2187 	.maxattr	= NBD_ATTR_MAX,
2188 	.policy = nbd_attr_policy,
2189 	.mcgrps		= nbd_mcast_grps,
2190 	.n_mcgrps	= ARRAY_SIZE(nbd_mcast_grps),
2191 };
2192 
2193 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2194 {
2195 	struct nlattr *dev_opt;
2196 	u8 connected = 0;
2197 	int ret;
2198 
2199 	/* This is a little racey, but for status it's ok.  The
2200 	 * reason we don't take a ref here is because we can't
2201 	 * take a ref in the index == -1 case as we would need
2202 	 * to put under the nbd_index_mutex, which could
2203 	 * deadlock if we are configured to remove ourselves
2204 	 * once we're disconnected.
2205 	 */
2206 	if (refcount_read(&nbd->config_refs))
2207 		connected = 1;
2208 	dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2209 	if (!dev_opt)
2210 		return -EMSGSIZE;
2211 	ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2212 	if (ret)
2213 		return -EMSGSIZE;
2214 	ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2215 			 connected);
2216 	if (ret)
2217 		return -EMSGSIZE;
2218 	nla_nest_end(reply, dev_opt);
2219 	return 0;
2220 }
2221 
2222 static int status_cb(int id, void *ptr, void *data)
2223 {
2224 	struct nbd_device *nbd = ptr;
2225 	return populate_nbd_status(nbd, (struct sk_buff *)data);
2226 }
2227 
2228 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2229 {
2230 	struct nlattr *dev_list;
2231 	struct sk_buff *reply;
2232 	void *reply_head;
2233 	size_t msg_size;
2234 	int index = -1;
2235 	int ret = -ENOMEM;
2236 
2237 	if (info->attrs[NBD_ATTR_INDEX])
2238 		index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2239 
2240 	mutex_lock(&nbd_index_mutex);
2241 
2242 	msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2243 				  nla_attr_size(sizeof(u8)));
2244 	msg_size *= (index == -1) ? nbd_total_devices : 1;
2245 
2246 	reply = genlmsg_new(msg_size, GFP_KERNEL);
2247 	if (!reply)
2248 		goto out;
2249 	reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2250 				       NBD_CMD_STATUS);
2251 	if (!reply_head) {
2252 		nlmsg_free(reply);
2253 		goto out;
2254 	}
2255 
2256 	dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2257 	if (index == -1) {
2258 		ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2259 		if (ret) {
2260 			nlmsg_free(reply);
2261 			goto out;
2262 		}
2263 	} else {
2264 		struct nbd_device *nbd;
2265 		nbd = idr_find(&nbd_index_idr, index);
2266 		if (nbd) {
2267 			ret = populate_nbd_status(nbd, reply);
2268 			if (ret) {
2269 				nlmsg_free(reply);
2270 				goto out;
2271 			}
2272 		}
2273 	}
2274 	nla_nest_end(reply, dev_list);
2275 	genlmsg_end(reply, reply_head);
2276 	ret = genlmsg_reply(reply, info);
2277 out:
2278 	mutex_unlock(&nbd_index_mutex);
2279 	return ret;
2280 }
2281 
2282 static void nbd_connect_reply(struct genl_info *info, int index)
2283 {
2284 	struct sk_buff *skb;
2285 	void *msg_head;
2286 	int ret;
2287 
2288 	skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2289 	if (!skb)
2290 		return;
2291 	msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2292 				     NBD_CMD_CONNECT);
2293 	if (!msg_head) {
2294 		nlmsg_free(skb);
2295 		return;
2296 	}
2297 	ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2298 	if (ret) {
2299 		nlmsg_free(skb);
2300 		return;
2301 	}
2302 	genlmsg_end(skb, msg_head);
2303 	genlmsg_reply(skb, info);
2304 }
2305 
2306 static void nbd_mcast_index(int index)
2307 {
2308 	struct sk_buff *skb;
2309 	void *msg_head;
2310 	int ret;
2311 
2312 	skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2313 	if (!skb)
2314 		return;
2315 	msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2316 				     NBD_CMD_LINK_DEAD);
2317 	if (!msg_head) {
2318 		nlmsg_free(skb);
2319 		return;
2320 	}
2321 	ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2322 	if (ret) {
2323 		nlmsg_free(skb);
2324 		return;
2325 	}
2326 	genlmsg_end(skb, msg_head);
2327 	genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2328 }
2329 
2330 static void nbd_dead_link_work(struct work_struct *work)
2331 {
2332 	struct link_dead_args *args = container_of(work, struct link_dead_args,
2333 						   work);
2334 	nbd_mcast_index(args->index);
2335 	kfree(args);
2336 }
2337 
2338 static int __init nbd_init(void)
2339 {
2340 	int i;
2341 
2342 	BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2343 
2344 	if (max_part < 0) {
2345 		printk(KERN_ERR "nbd: max_part must be >= 0\n");
2346 		return -EINVAL;
2347 	}
2348 
2349 	part_shift = 0;
2350 	if (max_part > 0) {
2351 		part_shift = fls(max_part);
2352 
2353 		/*
2354 		 * Adjust max_part according to part_shift as it is exported
2355 		 * to user space so that user can know the max number of
2356 		 * partition kernel should be able to manage.
2357 		 *
2358 		 * Note that -1 is required because partition 0 is reserved
2359 		 * for the whole disk.
2360 		 */
2361 		max_part = (1UL << part_shift) - 1;
2362 	}
2363 
2364 	if ((1UL << part_shift) > DISK_MAX_PARTS)
2365 		return -EINVAL;
2366 
2367 	if (nbds_max > 1UL << (MINORBITS - part_shift))
2368 		return -EINVAL;
2369 
2370 	if (register_blkdev(NBD_MAJOR, "nbd"))
2371 		return -EIO;
2372 
2373 	if (genl_register_family(&nbd_genl_family)) {
2374 		unregister_blkdev(NBD_MAJOR, "nbd");
2375 		return -EINVAL;
2376 	}
2377 	nbd_dbg_init();
2378 
2379 	mutex_lock(&nbd_index_mutex);
2380 	for (i = 0; i < nbds_max; i++)
2381 		nbd_dev_add(i);
2382 	mutex_unlock(&nbd_index_mutex);
2383 	return 0;
2384 }
2385 
2386 static int nbd_exit_cb(int id, void *ptr, void *data)
2387 {
2388 	struct list_head *list = (struct list_head *)data;
2389 	struct nbd_device *nbd = ptr;
2390 
2391 	list_add_tail(&nbd->list, list);
2392 	return 0;
2393 }
2394 
2395 static void __exit nbd_cleanup(void)
2396 {
2397 	struct nbd_device *nbd;
2398 	LIST_HEAD(del_list);
2399 
2400 	nbd_dbg_close();
2401 
2402 	mutex_lock(&nbd_index_mutex);
2403 	idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2404 	mutex_unlock(&nbd_index_mutex);
2405 
2406 	while (!list_empty(&del_list)) {
2407 		nbd = list_first_entry(&del_list, struct nbd_device, list);
2408 		list_del_init(&nbd->list);
2409 		if (refcount_read(&nbd->refs) != 1)
2410 			printk(KERN_ERR "nbd: possibly leaking a device\n");
2411 		nbd_put(nbd);
2412 	}
2413 
2414 	idr_destroy(&nbd_index_idr);
2415 	genl_unregister_family(&nbd_genl_family);
2416 	unregister_blkdev(NBD_MAJOR, "nbd");
2417 }
2418 
2419 module_init(nbd_init);
2420 module_exit(nbd_cleanup);
2421 
2422 MODULE_DESCRIPTION("Network Block Device");
2423 MODULE_LICENSE("GPL");
2424 
2425 module_param(nbds_max, int, 0444);
2426 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2427 module_param(max_part, int, 0444);
2428 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");
2429