xref: /openbmc/linux/drivers/block/nbd.c (revision 1372a51b)
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 		sockfd_put(sock);
997 		return NULL;
998 	}
999 
1000 	return sock;
1001 }
1002 
1003 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1004 			  bool netlink)
1005 {
1006 	struct nbd_config *config = nbd->config;
1007 	struct socket *sock;
1008 	struct nbd_sock **socks;
1009 	struct nbd_sock *nsock;
1010 	int err;
1011 
1012 	sock = nbd_get_socket(nbd, arg, &err);
1013 	if (!sock)
1014 		return err;
1015 
1016 	if (!netlink && !nbd->task_setup &&
1017 	    !test_bit(NBD_RT_BOUND, &config->runtime_flags))
1018 		nbd->task_setup = current;
1019 
1020 	if (!netlink &&
1021 	    (nbd->task_setup != current ||
1022 	     test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1023 		dev_err(disk_to_dev(nbd->disk),
1024 			"Device being setup by another task");
1025 		sockfd_put(sock);
1026 		return -EBUSY;
1027 	}
1028 
1029 	socks = krealloc(config->socks, (config->num_connections + 1) *
1030 			 sizeof(struct nbd_sock *), GFP_KERNEL);
1031 	if (!socks) {
1032 		sockfd_put(sock);
1033 		return -ENOMEM;
1034 	}
1035 
1036 	config->socks = socks;
1037 
1038 	nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
1039 	if (!nsock) {
1040 		sockfd_put(sock);
1041 		return -ENOMEM;
1042 	}
1043 
1044 	nsock->fallback_index = -1;
1045 	nsock->dead = false;
1046 	mutex_init(&nsock->tx_lock);
1047 	nsock->sock = sock;
1048 	nsock->pending = NULL;
1049 	nsock->sent = 0;
1050 	nsock->cookie = 0;
1051 	socks[config->num_connections++] = nsock;
1052 	atomic_inc(&config->live_connections);
1053 
1054 	return 0;
1055 }
1056 
1057 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1058 {
1059 	struct nbd_config *config = nbd->config;
1060 	struct socket *sock, *old;
1061 	struct recv_thread_args *args;
1062 	int i;
1063 	int err;
1064 
1065 	sock = nbd_get_socket(nbd, arg, &err);
1066 	if (!sock)
1067 		return err;
1068 
1069 	args = kzalloc(sizeof(*args), GFP_KERNEL);
1070 	if (!args) {
1071 		sockfd_put(sock);
1072 		return -ENOMEM;
1073 	}
1074 
1075 	for (i = 0; i < config->num_connections; i++) {
1076 		struct nbd_sock *nsock = config->socks[i];
1077 
1078 		if (!nsock->dead)
1079 			continue;
1080 
1081 		mutex_lock(&nsock->tx_lock);
1082 		if (!nsock->dead) {
1083 			mutex_unlock(&nsock->tx_lock);
1084 			continue;
1085 		}
1086 		sk_set_memalloc(sock->sk);
1087 		if (nbd->tag_set.timeout)
1088 			sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1089 		atomic_inc(&config->recv_threads);
1090 		refcount_inc(&nbd->config_refs);
1091 		old = nsock->sock;
1092 		nsock->fallback_index = -1;
1093 		nsock->sock = sock;
1094 		nsock->dead = false;
1095 		INIT_WORK(&args->work, recv_work);
1096 		args->index = i;
1097 		args->nbd = nbd;
1098 		nsock->cookie++;
1099 		mutex_unlock(&nsock->tx_lock);
1100 		sockfd_put(old);
1101 
1102 		clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1103 
1104 		/* We take the tx_mutex in an error path in the recv_work, so we
1105 		 * need to queue_work outside of the tx_mutex.
1106 		 */
1107 		queue_work(nbd->recv_workq, &args->work);
1108 
1109 		atomic_inc(&config->live_connections);
1110 		wake_up(&config->conn_wait);
1111 		return 0;
1112 	}
1113 	sockfd_put(sock);
1114 	kfree(args);
1115 	return -ENOSPC;
1116 }
1117 
1118 static void nbd_bdev_reset(struct block_device *bdev)
1119 {
1120 	if (bdev->bd_openers > 1)
1121 		return;
1122 	bd_set_size(bdev, 0);
1123 }
1124 
1125 static void nbd_parse_flags(struct nbd_device *nbd)
1126 {
1127 	struct nbd_config *config = nbd->config;
1128 	if (config->flags & NBD_FLAG_READ_ONLY)
1129 		set_disk_ro(nbd->disk, true);
1130 	else
1131 		set_disk_ro(nbd->disk, false);
1132 	if (config->flags & NBD_FLAG_SEND_TRIM)
1133 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1134 	if (config->flags & NBD_FLAG_SEND_FLUSH) {
1135 		if (config->flags & NBD_FLAG_SEND_FUA)
1136 			blk_queue_write_cache(nbd->disk->queue, true, true);
1137 		else
1138 			blk_queue_write_cache(nbd->disk->queue, true, false);
1139 	}
1140 	else
1141 		blk_queue_write_cache(nbd->disk->queue, false, false);
1142 }
1143 
1144 static void send_disconnects(struct nbd_device *nbd)
1145 {
1146 	struct nbd_config *config = nbd->config;
1147 	struct nbd_request request = {
1148 		.magic = htonl(NBD_REQUEST_MAGIC),
1149 		.type = htonl(NBD_CMD_DISC),
1150 	};
1151 	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1152 	struct iov_iter from;
1153 	int i, ret;
1154 
1155 	for (i = 0; i < config->num_connections; i++) {
1156 		struct nbd_sock *nsock = config->socks[i];
1157 
1158 		iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1159 		mutex_lock(&nsock->tx_lock);
1160 		ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1161 		if (ret <= 0)
1162 			dev_err(disk_to_dev(nbd->disk),
1163 				"Send disconnect failed %d\n", ret);
1164 		mutex_unlock(&nsock->tx_lock);
1165 	}
1166 }
1167 
1168 static int nbd_disconnect(struct nbd_device *nbd)
1169 {
1170 	struct nbd_config *config = nbd->config;
1171 
1172 	dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1173 	set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1174 	set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1175 	send_disconnects(nbd);
1176 	return 0;
1177 }
1178 
1179 static void nbd_clear_sock(struct nbd_device *nbd)
1180 {
1181 	sock_shutdown(nbd);
1182 	nbd_clear_que(nbd);
1183 	nbd->task_setup = NULL;
1184 }
1185 
1186 static void nbd_config_put(struct nbd_device *nbd)
1187 {
1188 	if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1189 					&nbd->config_lock)) {
1190 		struct nbd_config *config = nbd->config;
1191 		nbd_dev_dbg_close(nbd);
1192 		nbd_size_clear(nbd);
1193 		if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1194 				       &config->runtime_flags))
1195 			device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1196 		nbd->task_recv = NULL;
1197 		nbd_clear_sock(nbd);
1198 		if (config->num_connections) {
1199 			int i;
1200 			for (i = 0; i < config->num_connections; i++) {
1201 				sockfd_put(config->socks[i]->sock);
1202 				kfree(config->socks[i]);
1203 			}
1204 			kfree(config->socks);
1205 		}
1206 		kfree(nbd->config);
1207 		nbd->config = NULL;
1208 
1209 		if (nbd->recv_workq)
1210 			destroy_workqueue(nbd->recv_workq);
1211 		nbd->recv_workq = NULL;
1212 
1213 		nbd->tag_set.timeout = 0;
1214 		nbd->disk->queue->limits.discard_granularity = 0;
1215 		nbd->disk->queue->limits.discard_alignment = 0;
1216 		blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1217 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1218 
1219 		mutex_unlock(&nbd->config_lock);
1220 		nbd_put(nbd);
1221 		module_put(THIS_MODULE);
1222 	}
1223 }
1224 
1225 static int nbd_start_device(struct nbd_device *nbd)
1226 {
1227 	struct nbd_config *config = nbd->config;
1228 	int num_connections = config->num_connections;
1229 	int error = 0, i;
1230 
1231 	if (nbd->task_recv)
1232 		return -EBUSY;
1233 	if (!config->socks)
1234 		return -EINVAL;
1235 	if (num_connections > 1 &&
1236 	    !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1237 		dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1238 		return -EINVAL;
1239 	}
1240 
1241 	nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1242 					  WQ_MEM_RECLAIM | WQ_HIGHPRI |
1243 					  WQ_UNBOUND, 0, nbd->index);
1244 	if (!nbd->recv_workq) {
1245 		dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1246 		return -ENOMEM;
1247 	}
1248 
1249 	blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1250 	nbd->task_recv = current;
1251 
1252 	nbd_parse_flags(nbd);
1253 
1254 	error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1255 	if (error) {
1256 		dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1257 		return error;
1258 	}
1259 	set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1260 
1261 	nbd_dev_dbg_init(nbd);
1262 	for (i = 0; i < num_connections; i++) {
1263 		struct recv_thread_args *args;
1264 
1265 		args = kzalloc(sizeof(*args), GFP_KERNEL);
1266 		if (!args) {
1267 			sock_shutdown(nbd);
1268 			return -ENOMEM;
1269 		}
1270 		sk_set_memalloc(config->socks[i]->sock->sk);
1271 		if (nbd->tag_set.timeout)
1272 			config->socks[i]->sock->sk->sk_sndtimeo =
1273 				nbd->tag_set.timeout;
1274 		atomic_inc(&config->recv_threads);
1275 		refcount_inc(&nbd->config_refs);
1276 		INIT_WORK(&args->work, recv_work);
1277 		args->nbd = nbd;
1278 		args->index = i;
1279 		queue_work(nbd->recv_workq, &args->work);
1280 	}
1281 	nbd_size_update(nbd);
1282 	return error;
1283 }
1284 
1285 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1286 {
1287 	struct nbd_config *config = nbd->config;
1288 	int ret;
1289 
1290 	ret = nbd_start_device(nbd);
1291 	if (ret)
1292 		return ret;
1293 
1294 	if (max_part)
1295 		bdev->bd_invalidated = 1;
1296 	mutex_unlock(&nbd->config_lock);
1297 	ret = wait_event_interruptible(config->recv_wq,
1298 					 atomic_read(&config->recv_threads) == 0);
1299 	if (ret) {
1300 		sock_shutdown(nbd);
1301 		flush_workqueue(nbd->recv_workq);
1302 	}
1303 	mutex_lock(&nbd->config_lock);
1304 	nbd_bdev_reset(bdev);
1305 	/* user requested, ignore socket errors */
1306 	if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1307 		ret = 0;
1308 	if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1309 		ret = -ETIMEDOUT;
1310 	return ret;
1311 }
1312 
1313 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1314 				 struct block_device *bdev)
1315 {
1316 	sock_shutdown(nbd);
1317 	__invalidate_device(bdev, true);
1318 	nbd_bdev_reset(bdev);
1319 	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1320 			       &nbd->config->runtime_flags))
1321 		nbd_config_put(nbd);
1322 }
1323 
1324 static bool nbd_is_valid_blksize(unsigned long blksize)
1325 {
1326 	if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1327 	    blksize > PAGE_SIZE)
1328 		return false;
1329 	return true;
1330 }
1331 
1332 static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1333 {
1334 	nbd->tag_set.timeout = timeout * HZ;
1335 	if (timeout)
1336 		blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1337 }
1338 
1339 /* Must be called with config_lock held */
1340 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1341 		       unsigned int cmd, unsigned long arg)
1342 {
1343 	struct nbd_config *config = nbd->config;
1344 
1345 	switch (cmd) {
1346 	case NBD_DISCONNECT:
1347 		return nbd_disconnect(nbd);
1348 	case NBD_CLEAR_SOCK:
1349 		nbd_clear_sock_ioctl(nbd, bdev);
1350 		return 0;
1351 	case NBD_SET_SOCK:
1352 		return nbd_add_socket(nbd, arg, false);
1353 	case NBD_SET_BLKSIZE:
1354 		if (!arg)
1355 			arg = NBD_DEF_BLKSIZE;
1356 		if (!nbd_is_valid_blksize(arg))
1357 			return -EINVAL;
1358 		nbd_size_set(nbd, arg,
1359 			     div_s64(config->bytesize, arg));
1360 		return 0;
1361 	case NBD_SET_SIZE:
1362 		nbd_size_set(nbd, config->blksize,
1363 			     div_s64(arg, config->blksize));
1364 		return 0;
1365 	case NBD_SET_SIZE_BLOCKS:
1366 		nbd_size_set(nbd, config->blksize, arg);
1367 		return 0;
1368 	case NBD_SET_TIMEOUT:
1369 		nbd_set_cmd_timeout(nbd, arg);
1370 		return 0;
1371 
1372 	case NBD_SET_FLAGS:
1373 		config->flags = arg;
1374 		return 0;
1375 	case NBD_DO_IT:
1376 		return nbd_start_device_ioctl(nbd, bdev);
1377 	case NBD_CLEAR_QUE:
1378 		/*
1379 		 * This is for compatibility only.  The queue is always cleared
1380 		 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1381 		 */
1382 		return 0;
1383 	case NBD_PRINT_DEBUG:
1384 		/*
1385 		 * For compatibility only, we no longer keep a list of
1386 		 * outstanding requests.
1387 		 */
1388 		return 0;
1389 	}
1390 	return -ENOTTY;
1391 }
1392 
1393 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1394 		     unsigned int cmd, unsigned long arg)
1395 {
1396 	struct nbd_device *nbd = bdev->bd_disk->private_data;
1397 	struct nbd_config *config = nbd->config;
1398 	int error = -EINVAL;
1399 
1400 	if (!capable(CAP_SYS_ADMIN))
1401 		return -EPERM;
1402 
1403 	/* The block layer will pass back some non-nbd ioctls in case we have
1404 	 * special handling for them, but we don't so just return an error.
1405 	 */
1406 	if (_IOC_TYPE(cmd) != 0xab)
1407 		return -EINVAL;
1408 
1409 	mutex_lock(&nbd->config_lock);
1410 
1411 	/* Don't allow ioctl operations on a nbd device that was created with
1412 	 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1413 	 */
1414 	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1415 	    (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1416 		error = __nbd_ioctl(bdev, nbd, cmd, arg);
1417 	else
1418 		dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1419 	mutex_unlock(&nbd->config_lock);
1420 	return error;
1421 }
1422 
1423 static struct nbd_config *nbd_alloc_config(void)
1424 {
1425 	struct nbd_config *config;
1426 
1427 	config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1428 	if (!config)
1429 		return NULL;
1430 	atomic_set(&config->recv_threads, 0);
1431 	init_waitqueue_head(&config->recv_wq);
1432 	init_waitqueue_head(&config->conn_wait);
1433 	config->blksize = NBD_DEF_BLKSIZE;
1434 	atomic_set(&config->live_connections, 0);
1435 	try_module_get(THIS_MODULE);
1436 	return config;
1437 }
1438 
1439 static int nbd_open(struct block_device *bdev, fmode_t mode)
1440 {
1441 	struct nbd_device *nbd;
1442 	int ret = 0;
1443 
1444 	mutex_lock(&nbd_index_mutex);
1445 	nbd = bdev->bd_disk->private_data;
1446 	if (!nbd) {
1447 		ret = -ENXIO;
1448 		goto out;
1449 	}
1450 	if (!refcount_inc_not_zero(&nbd->refs)) {
1451 		ret = -ENXIO;
1452 		goto out;
1453 	}
1454 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
1455 		struct nbd_config *config;
1456 
1457 		mutex_lock(&nbd->config_lock);
1458 		if (refcount_inc_not_zero(&nbd->config_refs)) {
1459 			mutex_unlock(&nbd->config_lock);
1460 			goto out;
1461 		}
1462 		config = nbd->config = nbd_alloc_config();
1463 		if (!config) {
1464 			ret = -ENOMEM;
1465 			mutex_unlock(&nbd->config_lock);
1466 			goto out;
1467 		}
1468 		refcount_set(&nbd->config_refs, 1);
1469 		refcount_inc(&nbd->refs);
1470 		mutex_unlock(&nbd->config_lock);
1471 		bdev->bd_invalidated = 1;
1472 	} else if (nbd_disconnected(nbd->config)) {
1473 		bdev->bd_invalidated = 1;
1474 	}
1475 out:
1476 	mutex_unlock(&nbd_index_mutex);
1477 	return ret;
1478 }
1479 
1480 static void nbd_release(struct gendisk *disk, fmode_t mode)
1481 {
1482 	struct nbd_device *nbd = disk->private_data;
1483 	struct block_device *bdev = bdget_disk(disk, 0);
1484 
1485 	if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1486 			bdev->bd_openers == 0)
1487 		nbd_disconnect_and_put(nbd);
1488 
1489 	nbd_config_put(nbd);
1490 	nbd_put(nbd);
1491 }
1492 
1493 static const struct block_device_operations nbd_fops =
1494 {
1495 	.owner =	THIS_MODULE,
1496 	.open =		nbd_open,
1497 	.release =	nbd_release,
1498 	.ioctl =	nbd_ioctl,
1499 	.compat_ioctl =	nbd_ioctl,
1500 };
1501 
1502 #if IS_ENABLED(CONFIG_DEBUG_FS)
1503 
1504 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1505 {
1506 	struct nbd_device *nbd = s->private;
1507 
1508 	if (nbd->task_recv)
1509 		seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1510 
1511 	return 0;
1512 }
1513 
1514 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1515 {
1516 	return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1517 }
1518 
1519 static const struct file_operations nbd_dbg_tasks_ops = {
1520 	.open = nbd_dbg_tasks_open,
1521 	.read = seq_read,
1522 	.llseek = seq_lseek,
1523 	.release = single_release,
1524 };
1525 
1526 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1527 {
1528 	struct nbd_device *nbd = s->private;
1529 	u32 flags = nbd->config->flags;
1530 
1531 	seq_printf(s, "Hex: 0x%08x\n\n", flags);
1532 
1533 	seq_puts(s, "Known flags:\n");
1534 
1535 	if (flags & NBD_FLAG_HAS_FLAGS)
1536 		seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1537 	if (flags & NBD_FLAG_READ_ONLY)
1538 		seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1539 	if (flags & NBD_FLAG_SEND_FLUSH)
1540 		seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1541 	if (flags & NBD_FLAG_SEND_FUA)
1542 		seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1543 	if (flags & NBD_FLAG_SEND_TRIM)
1544 		seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1545 
1546 	return 0;
1547 }
1548 
1549 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1550 {
1551 	return single_open(file, nbd_dbg_flags_show, inode->i_private);
1552 }
1553 
1554 static const struct file_operations nbd_dbg_flags_ops = {
1555 	.open = nbd_dbg_flags_open,
1556 	.read = seq_read,
1557 	.llseek = seq_lseek,
1558 	.release = single_release,
1559 };
1560 
1561 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1562 {
1563 	struct dentry *dir;
1564 	struct nbd_config *config = nbd->config;
1565 
1566 	if (!nbd_dbg_dir)
1567 		return -EIO;
1568 
1569 	dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1570 	if (!dir) {
1571 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1572 			nbd_name(nbd));
1573 		return -EIO;
1574 	}
1575 	config->dbg_dir = dir;
1576 
1577 	debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1578 	debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1579 	debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1580 	debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1581 	debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1582 
1583 	return 0;
1584 }
1585 
1586 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1587 {
1588 	debugfs_remove_recursive(nbd->config->dbg_dir);
1589 }
1590 
1591 static int nbd_dbg_init(void)
1592 {
1593 	struct dentry *dbg_dir;
1594 
1595 	dbg_dir = debugfs_create_dir("nbd", NULL);
1596 	if (!dbg_dir)
1597 		return -EIO;
1598 
1599 	nbd_dbg_dir = dbg_dir;
1600 
1601 	return 0;
1602 }
1603 
1604 static void nbd_dbg_close(void)
1605 {
1606 	debugfs_remove_recursive(nbd_dbg_dir);
1607 }
1608 
1609 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1610 
1611 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1612 {
1613 	return 0;
1614 }
1615 
1616 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1617 {
1618 }
1619 
1620 static int nbd_dbg_init(void)
1621 {
1622 	return 0;
1623 }
1624 
1625 static void nbd_dbg_close(void)
1626 {
1627 }
1628 
1629 #endif
1630 
1631 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1632 			    unsigned int hctx_idx, unsigned int numa_node)
1633 {
1634 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1635 	cmd->nbd = set->driver_data;
1636 	cmd->flags = 0;
1637 	mutex_init(&cmd->lock);
1638 	return 0;
1639 }
1640 
1641 static const struct blk_mq_ops nbd_mq_ops = {
1642 	.queue_rq	= nbd_queue_rq,
1643 	.complete	= nbd_complete_rq,
1644 	.init_request	= nbd_init_request,
1645 	.timeout	= nbd_xmit_timeout,
1646 };
1647 
1648 static int nbd_dev_add(int index)
1649 {
1650 	struct nbd_device *nbd;
1651 	struct gendisk *disk;
1652 	struct request_queue *q;
1653 	int err = -ENOMEM;
1654 
1655 	nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1656 	if (!nbd)
1657 		goto out;
1658 
1659 	disk = alloc_disk(1 << part_shift);
1660 	if (!disk)
1661 		goto out_free_nbd;
1662 
1663 	if (index >= 0) {
1664 		err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1665 				GFP_KERNEL);
1666 		if (err == -ENOSPC)
1667 			err = -EEXIST;
1668 	} else {
1669 		err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1670 		if (err >= 0)
1671 			index = err;
1672 	}
1673 	if (err < 0)
1674 		goto out_free_disk;
1675 
1676 	nbd->index = index;
1677 	nbd->disk = disk;
1678 	nbd->tag_set.ops = &nbd_mq_ops;
1679 	nbd->tag_set.nr_hw_queues = 1;
1680 	nbd->tag_set.queue_depth = 128;
1681 	nbd->tag_set.numa_node = NUMA_NO_NODE;
1682 	nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1683 	nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1684 		BLK_MQ_F_BLOCKING;
1685 	nbd->tag_set.driver_data = nbd;
1686 	nbd->destroy_complete = NULL;
1687 
1688 	err = blk_mq_alloc_tag_set(&nbd->tag_set);
1689 	if (err)
1690 		goto out_free_idr;
1691 
1692 	q = blk_mq_init_queue(&nbd->tag_set);
1693 	if (IS_ERR(q)) {
1694 		err = PTR_ERR(q);
1695 		goto out_free_tags;
1696 	}
1697 	disk->queue = q;
1698 
1699 	/*
1700 	 * Tell the block layer that we are not a rotational device
1701 	 */
1702 	blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1703 	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1704 	disk->queue->limits.discard_granularity = 0;
1705 	disk->queue->limits.discard_alignment = 0;
1706 	blk_queue_max_discard_sectors(disk->queue, 0);
1707 	blk_queue_max_segment_size(disk->queue, UINT_MAX);
1708 	blk_queue_max_segments(disk->queue, USHRT_MAX);
1709 	blk_queue_max_hw_sectors(disk->queue, 65536);
1710 	disk->queue->limits.max_sectors = 256;
1711 
1712 	mutex_init(&nbd->config_lock);
1713 	refcount_set(&nbd->config_refs, 0);
1714 	refcount_set(&nbd->refs, 1);
1715 	INIT_LIST_HEAD(&nbd->list);
1716 	disk->major = NBD_MAJOR;
1717 	disk->first_minor = index << part_shift;
1718 	disk->fops = &nbd_fops;
1719 	disk->private_data = nbd;
1720 	sprintf(disk->disk_name, "nbd%d", index);
1721 	add_disk(disk);
1722 	nbd_total_devices++;
1723 	return index;
1724 
1725 out_free_tags:
1726 	blk_mq_free_tag_set(&nbd->tag_set);
1727 out_free_idr:
1728 	idr_remove(&nbd_index_idr, index);
1729 out_free_disk:
1730 	put_disk(disk);
1731 out_free_nbd:
1732 	kfree(nbd);
1733 out:
1734 	return err;
1735 }
1736 
1737 static int find_free_cb(int id, void *ptr, void *data)
1738 {
1739 	struct nbd_device *nbd = ptr;
1740 	struct nbd_device **found = data;
1741 
1742 	if (!refcount_read(&nbd->config_refs)) {
1743 		*found = nbd;
1744 		return 1;
1745 	}
1746 	return 0;
1747 }
1748 
1749 /* Netlink interface. */
1750 static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1751 	[NBD_ATTR_INDEX]		=	{ .type = NLA_U32 },
1752 	[NBD_ATTR_SIZE_BYTES]		=	{ .type = NLA_U64 },
1753 	[NBD_ATTR_BLOCK_SIZE_BYTES]	=	{ .type = NLA_U64 },
1754 	[NBD_ATTR_TIMEOUT]		=	{ .type = NLA_U64 },
1755 	[NBD_ATTR_SERVER_FLAGS]		=	{ .type = NLA_U64 },
1756 	[NBD_ATTR_CLIENT_FLAGS]		=	{ .type = NLA_U64 },
1757 	[NBD_ATTR_SOCKETS]		=	{ .type = NLA_NESTED},
1758 	[NBD_ATTR_DEAD_CONN_TIMEOUT]	=	{ .type = NLA_U64 },
1759 	[NBD_ATTR_DEVICE_LIST]		=	{ .type = NLA_NESTED},
1760 };
1761 
1762 static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1763 	[NBD_SOCK_FD]			=	{ .type = NLA_U32 },
1764 };
1765 
1766 /* We don't use this right now since we don't parse the incoming list, but we
1767  * still want it here so userspace knows what to expect.
1768  */
1769 static const struct nla_policy __attribute__((unused))
1770 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1771 	[NBD_DEVICE_INDEX]		=	{ .type = NLA_U32 },
1772 	[NBD_DEVICE_CONNECTED]		=	{ .type = NLA_U8 },
1773 };
1774 
1775 static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1776 {
1777 	struct nbd_config *config = nbd->config;
1778 	u64 bsize = config->blksize;
1779 	u64 bytes = config->bytesize;
1780 
1781 	if (info->attrs[NBD_ATTR_SIZE_BYTES])
1782 		bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1783 
1784 	if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1785 		bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1786 		if (!bsize)
1787 			bsize = NBD_DEF_BLKSIZE;
1788 		if (!nbd_is_valid_blksize(bsize)) {
1789 			printk(KERN_ERR "Invalid block size %llu\n", bsize);
1790 			return -EINVAL;
1791 		}
1792 	}
1793 
1794 	if (bytes != config->bytesize || bsize != config->blksize)
1795 		nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1796 	return 0;
1797 }
1798 
1799 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1800 {
1801 	DECLARE_COMPLETION_ONSTACK(destroy_complete);
1802 	struct nbd_device *nbd = NULL;
1803 	struct nbd_config *config;
1804 	int index = -1;
1805 	int ret;
1806 	bool put_dev = false;
1807 
1808 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1809 		return -EPERM;
1810 
1811 	if (info->attrs[NBD_ATTR_INDEX])
1812 		index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1813 	if (!info->attrs[NBD_ATTR_SOCKETS]) {
1814 		printk(KERN_ERR "nbd: must specify at least one socket\n");
1815 		return -EINVAL;
1816 	}
1817 	if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1818 		printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1819 		return -EINVAL;
1820 	}
1821 again:
1822 	mutex_lock(&nbd_index_mutex);
1823 	if (index == -1) {
1824 		ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1825 		if (ret == 0) {
1826 			int new_index;
1827 			new_index = nbd_dev_add(-1);
1828 			if (new_index < 0) {
1829 				mutex_unlock(&nbd_index_mutex);
1830 				printk(KERN_ERR "nbd: failed to add new device\n");
1831 				return new_index;
1832 			}
1833 			nbd = idr_find(&nbd_index_idr, new_index);
1834 		}
1835 	} else {
1836 		nbd = idr_find(&nbd_index_idr, index);
1837 		if (!nbd) {
1838 			ret = nbd_dev_add(index);
1839 			if (ret < 0) {
1840 				mutex_unlock(&nbd_index_mutex);
1841 				printk(KERN_ERR "nbd: failed to add new device\n");
1842 				return ret;
1843 			}
1844 			nbd = idr_find(&nbd_index_idr, index);
1845 		}
1846 	}
1847 	if (!nbd) {
1848 		printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1849 		       index);
1850 		mutex_unlock(&nbd_index_mutex);
1851 		return -EINVAL;
1852 	}
1853 
1854 	if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1855 	    test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1856 		nbd->destroy_complete = &destroy_complete;
1857 		mutex_unlock(&nbd_index_mutex);
1858 
1859 		/* Wait untill the the nbd stuff is totally destroyed */
1860 		wait_for_completion(&destroy_complete);
1861 		goto again;
1862 	}
1863 
1864 	if (!refcount_inc_not_zero(&nbd->refs)) {
1865 		mutex_unlock(&nbd_index_mutex);
1866 		if (index == -1)
1867 			goto again;
1868 		printk(KERN_ERR "nbd: device at index %d is going down\n",
1869 		       index);
1870 		return -EINVAL;
1871 	}
1872 	mutex_unlock(&nbd_index_mutex);
1873 
1874 	mutex_lock(&nbd->config_lock);
1875 	if (refcount_read(&nbd->config_refs)) {
1876 		mutex_unlock(&nbd->config_lock);
1877 		nbd_put(nbd);
1878 		if (index == -1)
1879 			goto again;
1880 		printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1881 		return -EBUSY;
1882 	}
1883 	if (WARN_ON(nbd->config)) {
1884 		mutex_unlock(&nbd->config_lock);
1885 		nbd_put(nbd);
1886 		return -EINVAL;
1887 	}
1888 	config = nbd->config = nbd_alloc_config();
1889 	if (!nbd->config) {
1890 		mutex_unlock(&nbd->config_lock);
1891 		nbd_put(nbd);
1892 		printk(KERN_ERR "nbd: couldn't allocate config\n");
1893 		return -ENOMEM;
1894 	}
1895 	refcount_set(&nbd->config_refs, 1);
1896 	set_bit(NBD_RT_BOUND, &config->runtime_flags);
1897 
1898 	ret = nbd_genl_size_set(info, nbd);
1899 	if (ret)
1900 		goto out;
1901 
1902 	if (info->attrs[NBD_ATTR_TIMEOUT])
1903 		nbd_set_cmd_timeout(nbd,
1904 				    nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1905 	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1906 		config->dead_conn_timeout =
1907 			nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1908 		config->dead_conn_timeout *= HZ;
1909 	}
1910 	if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1911 		config->flags =
1912 			nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1913 	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1914 		u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1915 		if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1916 			set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1917 				&config->runtime_flags);
1918 			set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1919 			put_dev = true;
1920 		} else {
1921 			clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1922 		}
1923 		if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1924 			set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1925 				&config->runtime_flags);
1926 		}
1927 	}
1928 
1929 	if (info->attrs[NBD_ATTR_SOCKETS]) {
1930 		struct nlattr *attr;
1931 		int rem, fd;
1932 
1933 		nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1934 				    rem) {
1935 			struct nlattr *socks[NBD_SOCK_MAX+1];
1936 
1937 			if (nla_type(attr) != NBD_SOCK_ITEM) {
1938 				printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1939 				ret = -EINVAL;
1940 				goto out;
1941 			}
1942 			ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1943 							  attr,
1944 							  nbd_sock_policy,
1945 							  info->extack);
1946 			if (ret != 0) {
1947 				printk(KERN_ERR "nbd: error processing sock list\n");
1948 				ret = -EINVAL;
1949 				goto out;
1950 			}
1951 			if (!socks[NBD_SOCK_FD])
1952 				continue;
1953 			fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1954 			ret = nbd_add_socket(nbd, fd, true);
1955 			if (ret)
1956 				goto out;
1957 		}
1958 	}
1959 	ret = nbd_start_device(nbd);
1960 out:
1961 	mutex_unlock(&nbd->config_lock);
1962 	if (!ret) {
1963 		set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1964 		refcount_inc(&nbd->config_refs);
1965 		nbd_connect_reply(info, nbd->index);
1966 	}
1967 	nbd_config_put(nbd);
1968 	if (put_dev)
1969 		nbd_put(nbd);
1970 	return ret;
1971 }
1972 
1973 static void nbd_disconnect_and_put(struct nbd_device *nbd)
1974 {
1975 	mutex_lock(&nbd->config_lock);
1976 	nbd_disconnect(nbd);
1977 	nbd_clear_sock(nbd);
1978 	mutex_unlock(&nbd->config_lock);
1979 	/*
1980 	 * Make sure recv thread has finished, so it does not drop the last
1981 	 * config ref and try to destroy the workqueue from inside the work
1982 	 * queue.
1983 	 */
1984 	flush_workqueue(nbd->recv_workq);
1985 	if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1986 			       &nbd->config->runtime_flags))
1987 		nbd_config_put(nbd);
1988 }
1989 
1990 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1991 {
1992 	struct nbd_device *nbd;
1993 	int index;
1994 
1995 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1996 		return -EPERM;
1997 
1998 	if (!info->attrs[NBD_ATTR_INDEX]) {
1999 		printk(KERN_ERR "nbd: must specify an index to disconnect\n");
2000 		return -EINVAL;
2001 	}
2002 	index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2003 	mutex_lock(&nbd_index_mutex);
2004 	nbd = idr_find(&nbd_index_idr, index);
2005 	if (!nbd) {
2006 		mutex_unlock(&nbd_index_mutex);
2007 		printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2008 		       index);
2009 		return -EINVAL;
2010 	}
2011 	if (!refcount_inc_not_zero(&nbd->refs)) {
2012 		mutex_unlock(&nbd_index_mutex);
2013 		printk(KERN_ERR "nbd: device at index %d is going down\n",
2014 		       index);
2015 		return -EINVAL;
2016 	}
2017 	mutex_unlock(&nbd_index_mutex);
2018 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
2019 		nbd_put(nbd);
2020 		return 0;
2021 	}
2022 	nbd_disconnect_and_put(nbd);
2023 	nbd_config_put(nbd);
2024 	nbd_put(nbd);
2025 	return 0;
2026 }
2027 
2028 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2029 {
2030 	struct nbd_device *nbd = NULL;
2031 	struct nbd_config *config;
2032 	int index;
2033 	int ret = 0;
2034 	bool put_dev = false;
2035 
2036 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
2037 		return -EPERM;
2038 
2039 	if (!info->attrs[NBD_ATTR_INDEX]) {
2040 		printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2041 		return -EINVAL;
2042 	}
2043 	index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2044 	mutex_lock(&nbd_index_mutex);
2045 	nbd = idr_find(&nbd_index_idr, index);
2046 	if (!nbd) {
2047 		mutex_unlock(&nbd_index_mutex);
2048 		printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2049 		       index);
2050 		return -EINVAL;
2051 	}
2052 	if (!refcount_inc_not_zero(&nbd->refs)) {
2053 		mutex_unlock(&nbd_index_mutex);
2054 		printk(KERN_ERR "nbd: device at index %d is going down\n",
2055 		       index);
2056 		return -EINVAL;
2057 	}
2058 	mutex_unlock(&nbd_index_mutex);
2059 
2060 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
2061 		dev_err(nbd_to_dev(nbd),
2062 			"not configured, cannot reconfigure\n");
2063 		nbd_put(nbd);
2064 		return -EINVAL;
2065 	}
2066 
2067 	mutex_lock(&nbd->config_lock);
2068 	config = nbd->config;
2069 	if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2070 	    !nbd->task_recv) {
2071 		dev_err(nbd_to_dev(nbd),
2072 			"not configured, cannot reconfigure\n");
2073 		ret = -EINVAL;
2074 		goto out;
2075 	}
2076 
2077 	ret = nbd_genl_size_set(info, nbd);
2078 	if (ret)
2079 		goto out;
2080 
2081 	if (info->attrs[NBD_ATTR_TIMEOUT])
2082 		nbd_set_cmd_timeout(nbd,
2083 				    nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2084 	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2085 		config->dead_conn_timeout =
2086 			nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2087 		config->dead_conn_timeout *= HZ;
2088 	}
2089 	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2090 		u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2091 		if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2092 			if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2093 					      &config->runtime_flags))
2094 				put_dev = true;
2095 			set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2096 		} else {
2097 			if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2098 					       &config->runtime_flags))
2099 				refcount_inc(&nbd->refs);
2100 			clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2101 		}
2102 
2103 		if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2104 			set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2105 					&config->runtime_flags);
2106 		} else {
2107 			clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2108 					&config->runtime_flags);
2109 		}
2110 	}
2111 
2112 	if (info->attrs[NBD_ATTR_SOCKETS]) {
2113 		struct nlattr *attr;
2114 		int rem, fd;
2115 
2116 		nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2117 				    rem) {
2118 			struct nlattr *socks[NBD_SOCK_MAX+1];
2119 
2120 			if (nla_type(attr) != NBD_SOCK_ITEM) {
2121 				printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2122 				ret = -EINVAL;
2123 				goto out;
2124 			}
2125 			ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2126 							  attr,
2127 							  nbd_sock_policy,
2128 							  info->extack);
2129 			if (ret != 0) {
2130 				printk(KERN_ERR "nbd: error processing sock list\n");
2131 				ret = -EINVAL;
2132 				goto out;
2133 			}
2134 			if (!socks[NBD_SOCK_FD])
2135 				continue;
2136 			fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2137 			ret = nbd_reconnect_socket(nbd, fd);
2138 			if (ret) {
2139 				if (ret == -ENOSPC)
2140 					ret = 0;
2141 				goto out;
2142 			}
2143 			dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2144 		}
2145 	}
2146 out:
2147 	mutex_unlock(&nbd->config_lock);
2148 	nbd_config_put(nbd);
2149 	nbd_put(nbd);
2150 	if (put_dev)
2151 		nbd_put(nbd);
2152 	return ret;
2153 }
2154 
2155 static const struct genl_ops nbd_connect_genl_ops[] = {
2156 	{
2157 		.cmd	= NBD_CMD_CONNECT,
2158 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2159 		.doit	= nbd_genl_connect,
2160 	},
2161 	{
2162 		.cmd	= NBD_CMD_DISCONNECT,
2163 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2164 		.doit	= nbd_genl_disconnect,
2165 	},
2166 	{
2167 		.cmd	= NBD_CMD_RECONFIGURE,
2168 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2169 		.doit	= nbd_genl_reconfigure,
2170 	},
2171 	{
2172 		.cmd	= NBD_CMD_STATUS,
2173 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2174 		.doit	= nbd_genl_status,
2175 	},
2176 };
2177 
2178 static const struct genl_multicast_group nbd_mcast_grps[] = {
2179 	{ .name = NBD_GENL_MCAST_GROUP_NAME, },
2180 };
2181 
2182 static struct genl_family nbd_genl_family __ro_after_init = {
2183 	.hdrsize	= 0,
2184 	.name		= NBD_GENL_FAMILY_NAME,
2185 	.version	= NBD_GENL_VERSION,
2186 	.module		= THIS_MODULE,
2187 	.ops		= nbd_connect_genl_ops,
2188 	.n_ops		= ARRAY_SIZE(nbd_connect_genl_ops),
2189 	.maxattr	= NBD_ATTR_MAX,
2190 	.policy = nbd_attr_policy,
2191 	.mcgrps		= nbd_mcast_grps,
2192 	.n_mcgrps	= ARRAY_SIZE(nbd_mcast_grps),
2193 };
2194 
2195 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2196 {
2197 	struct nlattr *dev_opt;
2198 	u8 connected = 0;
2199 	int ret;
2200 
2201 	/* This is a little racey, but for status it's ok.  The
2202 	 * reason we don't take a ref here is because we can't
2203 	 * take a ref in the index == -1 case as we would need
2204 	 * to put under the nbd_index_mutex, which could
2205 	 * deadlock if we are configured to remove ourselves
2206 	 * once we're disconnected.
2207 	 */
2208 	if (refcount_read(&nbd->config_refs))
2209 		connected = 1;
2210 	dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2211 	if (!dev_opt)
2212 		return -EMSGSIZE;
2213 	ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2214 	if (ret)
2215 		return -EMSGSIZE;
2216 	ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2217 			 connected);
2218 	if (ret)
2219 		return -EMSGSIZE;
2220 	nla_nest_end(reply, dev_opt);
2221 	return 0;
2222 }
2223 
2224 static int status_cb(int id, void *ptr, void *data)
2225 {
2226 	struct nbd_device *nbd = ptr;
2227 	return populate_nbd_status(nbd, (struct sk_buff *)data);
2228 }
2229 
2230 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2231 {
2232 	struct nlattr *dev_list;
2233 	struct sk_buff *reply;
2234 	void *reply_head;
2235 	size_t msg_size;
2236 	int index = -1;
2237 	int ret = -ENOMEM;
2238 
2239 	if (info->attrs[NBD_ATTR_INDEX])
2240 		index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2241 
2242 	mutex_lock(&nbd_index_mutex);
2243 
2244 	msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2245 				  nla_attr_size(sizeof(u8)));
2246 	msg_size *= (index == -1) ? nbd_total_devices : 1;
2247 
2248 	reply = genlmsg_new(msg_size, GFP_KERNEL);
2249 	if (!reply)
2250 		goto out;
2251 	reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2252 				       NBD_CMD_STATUS);
2253 	if (!reply_head) {
2254 		nlmsg_free(reply);
2255 		goto out;
2256 	}
2257 
2258 	dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2259 	if (index == -1) {
2260 		ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2261 		if (ret) {
2262 			nlmsg_free(reply);
2263 			goto out;
2264 		}
2265 	} else {
2266 		struct nbd_device *nbd;
2267 		nbd = idr_find(&nbd_index_idr, index);
2268 		if (nbd) {
2269 			ret = populate_nbd_status(nbd, reply);
2270 			if (ret) {
2271 				nlmsg_free(reply);
2272 				goto out;
2273 			}
2274 		}
2275 	}
2276 	nla_nest_end(reply, dev_list);
2277 	genlmsg_end(reply, reply_head);
2278 	ret = genlmsg_reply(reply, info);
2279 out:
2280 	mutex_unlock(&nbd_index_mutex);
2281 	return ret;
2282 }
2283 
2284 static void nbd_connect_reply(struct genl_info *info, int index)
2285 {
2286 	struct sk_buff *skb;
2287 	void *msg_head;
2288 	int ret;
2289 
2290 	skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2291 	if (!skb)
2292 		return;
2293 	msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2294 				     NBD_CMD_CONNECT);
2295 	if (!msg_head) {
2296 		nlmsg_free(skb);
2297 		return;
2298 	}
2299 	ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2300 	if (ret) {
2301 		nlmsg_free(skb);
2302 		return;
2303 	}
2304 	genlmsg_end(skb, msg_head);
2305 	genlmsg_reply(skb, info);
2306 }
2307 
2308 static void nbd_mcast_index(int index)
2309 {
2310 	struct sk_buff *skb;
2311 	void *msg_head;
2312 	int ret;
2313 
2314 	skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2315 	if (!skb)
2316 		return;
2317 	msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2318 				     NBD_CMD_LINK_DEAD);
2319 	if (!msg_head) {
2320 		nlmsg_free(skb);
2321 		return;
2322 	}
2323 	ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2324 	if (ret) {
2325 		nlmsg_free(skb);
2326 		return;
2327 	}
2328 	genlmsg_end(skb, msg_head);
2329 	genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2330 }
2331 
2332 static void nbd_dead_link_work(struct work_struct *work)
2333 {
2334 	struct link_dead_args *args = container_of(work, struct link_dead_args,
2335 						   work);
2336 	nbd_mcast_index(args->index);
2337 	kfree(args);
2338 }
2339 
2340 static int __init nbd_init(void)
2341 {
2342 	int i;
2343 
2344 	BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2345 
2346 	if (max_part < 0) {
2347 		printk(KERN_ERR "nbd: max_part must be >= 0\n");
2348 		return -EINVAL;
2349 	}
2350 
2351 	part_shift = 0;
2352 	if (max_part > 0) {
2353 		part_shift = fls(max_part);
2354 
2355 		/*
2356 		 * Adjust max_part according to part_shift as it is exported
2357 		 * to user space so that user can know the max number of
2358 		 * partition kernel should be able to manage.
2359 		 *
2360 		 * Note that -1 is required because partition 0 is reserved
2361 		 * for the whole disk.
2362 		 */
2363 		max_part = (1UL << part_shift) - 1;
2364 	}
2365 
2366 	if ((1UL << part_shift) > DISK_MAX_PARTS)
2367 		return -EINVAL;
2368 
2369 	if (nbds_max > 1UL << (MINORBITS - part_shift))
2370 		return -EINVAL;
2371 
2372 	if (register_blkdev(NBD_MAJOR, "nbd"))
2373 		return -EIO;
2374 
2375 	if (genl_register_family(&nbd_genl_family)) {
2376 		unregister_blkdev(NBD_MAJOR, "nbd");
2377 		return -EINVAL;
2378 	}
2379 	nbd_dbg_init();
2380 
2381 	mutex_lock(&nbd_index_mutex);
2382 	for (i = 0; i < nbds_max; i++)
2383 		nbd_dev_add(i);
2384 	mutex_unlock(&nbd_index_mutex);
2385 	return 0;
2386 }
2387 
2388 static int nbd_exit_cb(int id, void *ptr, void *data)
2389 {
2390 	struct list_head *list = (struct list_head *)data;
2391 	struct nbd_device *nbd = ptr;
2392 
2393 	list_add_tail(&nbd->list, list);
2394 	return 0;
2395 }
2396 
2397 static void __exit nbd_cleanup(void)
2398 {
2399 	struct nbd_device *nbd;
2400 	LIST_HEAD(del_list);
2401 
2402 	nbd_dbg_close();
2403 
2404 	mutex_lock(&nbd_index_mutex);
2405 	idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2406 	mutex_unlock(&nbd_index_mutex);
2407 
2408 	while (!list_empty(&del_list)) {
2409 		nbd = list_first_entry(&del_list, struct nbd_device, list);
2410 		list_del_init(&nbd->list);
2411 		if (refcount_read(&nbd->refs) != 1)
2412 			printk(KERN_ERR "nbd: possibly leaking a device\n");
2413 		nbd_put(nbd);
2414 	}
2415 
2416 	idr_destroy(&nbd_index_idr);
2417 	genl_unregister_family(&nbd_genl_family);
2418 	unregister_blkdev(NBD_MAJOR, "nbd");
2419 }
2420 
2421 module_init(nbd_init);
2422 module_exit(nbd_cleanup);
2423 
2424 MODULE_DESCRIPTION("Network Block Device");
2425 MODULE_LICENSE("GPL");
2426 
2427 module_param(nbds_max, int, 0444);
2428 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2429 module_param(max_part, int, 0444);
2430 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");
2431