xref: /linux/drivers/block/nbd.c (revision b24413180f5600bcb3bb70fbed5cf186b60864bd)
1 /*
2  * Network block device - make block devices work over TCP
3  *
4  * Note that you can not swap over this thing, yet. Seems to work but
5  * deadlocks sometimes - you can not swap over TCP in general.
6  *
7  * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
8  * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
9  *
10  * This file is released under GPLv2 or later.
11  *
12  * (part of code stolen from loop.c)
13  */
14 
15 #include <linux/major.h>
16 
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/sched.h>
21 #include <linux/sched/mm.h>
22 #include <linux/fs.h>
23 #include <linux/bio.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/file.h>
27 #include <linux/ioctl.h>
28 #include <linux/mutex.h>
29 #include <linux/compiler.h>
30 #include <linux/err.h>
31 #include <linux/kernel.h>
32 #include <linux/slab.h>
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 static DEFINE_IDR(nbd_index_idr);
48 static DEFINE_MUTEX(nbd_index_mutex);
49 static int nbd_total_devices = 0;
50 
51 struct nbd_sock {
52 	struct socket *sock;
53 	struct mutex tx_lock;
54 	struct request *pending;
55 	int sent;
56 	bool dead;
57 	int fallback_index;
58 	int cookie;
59 };
60 
61 struct recv_thread_args {
62 	struct work_struct work;
63 	struct nbd_device *nbd;
64 	int index;
65 };
66 
67 struct link_dead_args {
68 	struct work_struct work;
69 	int index;
70 };
71 
72 #define NBD_TIMEDOUT			0
73 #define NBD_DISCONNECT_REQUESTED	1
74 #define NBD_DISCONNECTED		2
75 #define NBD_HAS_PID_FILE		3
76 #define NBD_HAS_CONFIG_REF		4
77 #define NBD_BOUND			5
78 #define NBD_DESTROY_ON_DISCONNECT	6
79 
80 struct nbd_config {
81 	u32 flags;
82 	unsigned long runtime_flags;
83 	u64 dead_conn_timeout;
84 
85 	struct nbd_sock **socks;
86 	int num_connections;
87 	atomic_t live_connections;
88 	wait_queue_head_t conn_wait;
89 
90 	atomic_t recv_threads;
91 	wait_queue_head_t recv_wq;
92 	loff_t blksize;
93 	loff_t bytesize;
94 #if IS_ENABLED(CONFIG_DEBUG_FS)
95 	struct dentry *dbg_dir;
96 #endif
97 };
98 
99 struct nbd_device {
100 	struct blk_mq_tag_set tag_set;
101 
102 	int index;
103 	refcount_t config_refs;
104 	refcount_t refs;
105 	struct nbd_config *config;
106 	struct mutex config_lock;
107 	struct gendisk *disk;
108 
109 	struct list_head list;
110 	struct task_struct *task_recv;
111 	struct task_struct *task_setup;
112 };
113 
114 struct nbd_cmd {
115 	struct nbd_device *nbd;
116 	int index;
117 	int cookie;
118 	struct completion send_complete;
119 	blk_status_t status;
120 };
121 
122 #if IS_ENABLED(CONFIG_DEBUG_FS)
123 static struct dentry *nbd_dbg_dir;
124 #endif
125 
126 #define nbd_name(nbd) ((nbd)->disk->disk_name)
127 
128 #define NBD_MAGIC 0x68797548
129 
130 static unsigned int nbds_max = 16;
131 static int max_part = 16;
132 static struct workqueue_struct *recv_workqueue;
133 static int part_shift;
134 
135 static int nbd_dev_dbg_init(struct nbd_device *nbd);
136 static void nbd_dev_dbg_close(struct nbd_device *nbd);
137 static void nbd_config_put(struct nbd_device *nbd);
138 static void nbd_connect_reply(struct genl_info *info, int index);
139 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
140 static void nbd_dead_link_work(struct work_struct *work);
141 
142 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
143 {
144 	return disk_to_dev(nbd->disk);
145 }
146 
147 static const char *nbdcmd_to_ascii(int cmd)
148 {
149 	switch (cmd) {
150 	case  NBD_CMD_READ: return "read";
151 	case NBD_CMD_WRITE: return "write";
152 	case  NBD_CMD_DISC: return "disconnect";
153 	case NBD_CMD_FLUSH: return "flush";
154 	case  NBD_CMD_TRIM: return "trim/discard";
155 	}
156 	return "invalid";
157 }
158 
159 static ssize_t pid_show(struct device *dev,
160 			struct device_attribute *attr, char *buf)
161 {
162 	struct gendisk *disk = dev_to_disk(dev);
163 	struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
164 
165 	return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
166 }
167 
168 static const struct device_attribute pid_attr = {
169 	.attr = { .name = "pid", .mode = S_IRUGO},
170 	.show = pid_show,
171 };
172 
173 static void nbd_dev_remove(struct nbd_device *nbd)
174 {
175 	struct gendisk *disk = nbd->disk;
176 	if (disk) {
177 		del_gendisk(disk);
178 		blk_cleanup_queue(disk->queue);
179 		blk_mq_free_tag_set(&nbd->tag_set);
180 		disk->private_data = NULL;
181 		put_disk(disk);
182 	}
183 	kfree(nbd);
184 }
185 
186 static void nbd_put(struct nbd_device *nbd)
187 {
188 	if (refcount_dec_and_mutex_lock(&nbd->refs,
189 					&nbd_index_mutex)) {
190 		idr_remove(&nbd_index_idr, nbd->index);
191 		mutex_unlock(&nbd_index_mutex);
192 		nbd_dev_remove(nbd);
193 	}
194 }
195 
196 static int nbd_disconnected(struct nbd_config *config)
197 {
198 	return test_bit(NBD_DISCONNECTED, &config->runtime_flags) ||
199 		test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
200 }
201 
202 static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
203 				int notify)
204 {
205 	if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
206 		struct link_dead_args *args;
207 		args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
208 		if (args) {
209 			INIT_WORK(&args->work, nbd_dead_link_work);
210 			args->index = nbd->index;
211 			queue_work(system_wq, &args->work);
212 		}
213 	}
214 	if (!nsock->dead) {
215 		kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
216 		atomic_dec(&nbd->config->live_connections);
217 	}
218 	nsock->dead = true;
219 	nsock->pending = NULL;
220 	nsock->sent = 0;
221 }
222 
223 static void nbd_size_clear(struct nbd_device *nbd)
224 {
225 	if (nbd->config->bytesize) {
226 		set_capacity(nbd->disk, 0);
227 		kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
228 	}
229 }
230 
231 static void nbd_size_update(struct nbd_device *nbd)
232 {
233 	struct nbd_config *config = nbd->config;
234 	blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
235 	blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
236 	set_capacity(nbd->disk, config->bytesize >> 9);
237 	kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
238 }
239 
240 static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
241 			 loff_t nr_blocks)
242 {
243 	struct nbd_config *config = nbd->config;
244 	config->blksize = blocksize;
245 	config->bytesize = blocksize * nr_blocks;
246 }
247 
248 static void nbd_complete_rq(struct request *req)
249 {
250 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
251 
252 	dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", cmd,
253 		cmd->status ? "failed" : "done");
254 
255 	blk_mq_end_request(req, cmd->status);
256 }
257 
258 /*
259  * Forcibly shutdown the socket causing all listeners to error
260  */
261 static void sock_shutdown(struct nbd_device *nbd)
262 {
263 	struct nbd_config *config = nbd->config;
264 	int i;
265 
266 	if (config->num_connections == 0)
267 		return;
268 	if (test_and_set_bit(NBD_DISCONNECTED, &config->runtime_flags))
269 		return;
270 
271 	for (i = 0; i < config->num_connections; i++) {
272 		struct nbd_sock *nsock = config->socks[i];
273 		mutex_lock(&nsock->tx_lock);
274 		nbd_mark_nsock_dead(nbd, nsock, 0);
275 		mutex_unlock(&nsock->tx_lock);
276 	}
277 	dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
278 }
279 
280 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
281 						 bool reserved)
282 {
283 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
284 	struct nbd_device *nbd = cmd->nbd;
285 	struct nbd_config *config;
286 
287 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
288 		cmd->status = BLK_STS_TIMEOUT;
289 		return BLK_EH_HANDLED;
290 	}
291 
292 	/* If we are waiting on our dead timer then we could get timeout
293 	 * callbacks for our request.  For this we just want to reset the timer
294 	 * and let the queue side take care of everything.
295 	 */
296 	if (!completion_done(&cmd->send_complete)) {
297 		nbd_config_put(nbd);
298 		return BLK_EH_RESET_TIMER;
299 	}
300 	config = nbd->config;
301 
302 	if (config->num_connections > 1) {
303 		dev_err_ratelimited(nbd_to_dev(nbd),
304 				    "Connection timed out, retrying\n");
305 		/*
306 		 * Hooray we have more connections, requeue this IO, the submit
307 		 * path will put it on a real connection.
308 		 */
309 		if (config->socks && config->num_connections > 1) {
310 			if (cmd->index < config->num_connections) {
311 				struct nbd_sock *nsock =
312 					config->socks[cmd->index];
313 				mutex_lock(&nsock->tx_lock);
314 				/* We can have multiple outstanding requests, so
315 				 * we don't want to mark the nsock dead if we've
316 				 * already reconnected with a new socket, so
317 				 * only mark it dead if its the same socket we
318 				 * were sent out on.
319 				 */
320 				if (cmd->cookie == nsock->cookie)
321 					nbd_mark_nsock_dead(nbd, nsock, 1);
322 				mutex_unlock(&nsock->tx_lock);
323 			}
324 			blk_mq_requeue_request(req, true);
325 			nbd_config_put(nbd);
326 			return BLK_EH_NOT_HANDLED;
327 		}
328 	} else {
329 		dev_err_ratelimited(nbd_to_dev(nbd),
330 				    "Connection timed out\n");
331 	}
332 	set_bit(NBD_TIMEDOUT, &config->runtime_flags);
333 	cmd->status = BLK_STS_IOERR;
334 	sock_shutdown(nbd);
335 	nbd_config_put(nbd);
336 
337 	return BLK_EH_HANDLED;
338 }
339 
340 /*
341  *  Send or receive packet.
342  */
343 static int sock_xmit(struct nbd_device *nbd, int index, int send,
344 		     struct iov_iter *iter, int msg_flags, int *sent)
345 {
346 	struct nbd_config *config = nbd->config;
347 	struct socket *sock = config->socks[index]->sock;
348 	int result;
349 	struct msghdr msg;
350 	unsigned int noreclaim_flag;
351 
352 	if (unlikely(!sock)) {
353 		dev_err_ratelimited(disk_to_dev(nbd->disk),
354 			"Attempted %s on closed socket in sock_xmit\n",
355 			(send ? "send" : "recv"));
356 		return -EINVAL;
357 	}
358 
359 	msg.msg_iter = *iter;
360 
361 	noreclaim_flag = memalloc_noreclaim_save();
362 	do {
363 		sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
364 		msg.msg_name = NULL;
365 		msg.msg_namelen = 0;
366 		msg.msg_control = NULL;
367 		msg.msg_controllen = 0;
368 		msg.msg_flags = msg_flags | MSG_NOSIGNAL;
369 
370 		if (send)
371 			result = sock_sendmsg(sock, &msg);
372 		else
373 			result = sock_recvmsg(sock, &msg, msg.msg_flags);
374 
375 		if (result <= 0) {
376 			if (result == 0)
377 				result = -EPIPE; /* short read */
378 			break;
379 		}
380 		if (sent)
381 			*sent += result;
382 	} while (msg_data_left(&msg));
383 
384 	memalloc_noreclaim_restore(noreclaim_flag);
385 
386 	return result;
387 }
388 
389 /* always call with the tx_lock held */
390 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
391 {
392 	struct request *req = blk_mq_rq_from_pdu(cmd);
393 	struct nbd_config *config = nbd->config;
394 	struct nbd_sock *nsock = config->socks[index];
395 	int result;
396 	struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
397 	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
398 	struct iov_iter from;
399 	unsigned long size = blk_rq_bytes(req);
400 	struct bio *bio;
401 	u32 type;
402 	u32 nbd_cmd_flags = 0;
403 	u32 tag = blk_mq_unique_tag(req);
404 	int sent = nsock->sent, skip = 0;
405 
406 	iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
407 
408 	switch (req_op(req)) {
409 	case REQ_OP_DISCARD:
410 		type = NBD_CMD_TRIM;
411 		break;
412 	case REQ_OP_FLUSH:
413 		type = NBD_CMD_FLUSH;
414 		break;
415 	case REQ_OP_WRITE:
416 		type = NBD_CMD_WRITE;
417 		break;
418 	case REQ_OP_READ:
419 		type = NBD_CMD_READ;
420 		break;
421 	default:
422 		return -EIO;
423 	}
424 
425 	if (rq_data_dir(req) == WRITE &&
426 	    (config->flags & NBD_FLAG_READ_ONLY)) {
427 		dev_err_ratelimited(disk_to_dev(nbd->disk),
428 				    "Write on read-only\n");
429 		return -EIO;
430 	}
431 
432 	if (req->cmd_flags & REQ_FUA)
433 		nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
434 
435 	/* We did a partial send previously, and we at least sent the whole
436 	 * request struct, so just go and send the rest of the pages in the
437 	 * request.
438 	 */
439 	if (sent) {
440 		if (sent >= sizeof(request)) {
441 			skip = sent - sizeof(request);
442 			goto send_pages;
443 		}
444 		iov_iter_advance(&from, sent);
445 	}
446 	cmd->index = index;
447 	cmd->cookie = nsock->cookie;
448 	request.type = htonl(type | nbd_cmd_flags);
449 	if (type != NBD_CMD_FLUSH) {
450 		request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
451 		request.len = htonl(size);
452 	}
453 	memcpy(request.handle, &tag, sizeof(tag));
454 
455 	dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
456 		cmd, nbdcmd_to_ascii(type),
457 		(unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
458 	result = sock_xmit(nbd, index, 1, &from,
459 			(type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
460 	if (result <= 0) {
461 		if (result == -ERESTARTSYS) {
462 			/* If we havne't sent anything we can just return BUSY,
463 			 * however if we have sent something we need to make
464 			 * sure we only allow this req to be sent until we are
465 			 * completely done.
466 			 */
467 			if (sent) {
468 				nsock->pending = req;
469 				nsock->sent = sent;
470 			}
471 			return BLK_STS_RESOURCE;
472 		}
473 		dev_err_ratelimited(disk_to_dev(nbd->disk),
474 			"Send control failed (result %d)\n", result);
475 		return -EAGAIN;
476 	}
477 send_pages:
478 	if (type != NBD_CMD_WRITE)
479 		goto out;
480 
481 	bio = req->bio;
482 	while (bio) {
483 		struct bio *next = bio->bi_next;
484 		struct bvec_iter iter;
485 		struct bio_vec bvec;
486 
487 		bio_for_each_segment(bvec, bio, iter) {
488 			bool is_last = !next && bio_iter_last(bvec, iter);
489 			int flags = is_last ? 0 : MSG_MORE;
490 
491 			dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
492 				cmd, bvec.bv_len);
493 			iov_iter_bvec(&from, ITER_BVEC | WRITE,
494 				      &bvec, 1, bvec.bv_len);
495 			if (skip) {
496 				if (skip >= iov_iter_count(&from)) {
497 					skip -= iov_iter_count(&from);
498 					continue;
499 				}
500 				iov_iter_advance(&from, skip);
501 				skip = 0;
502 			}
503 			result = sock_xmit(nbd, index, 1, &from, flags, &sent);
504 			if (result <= 0) {
505 				if (result == -ERESTARTSYS) {
506 					/* We've already sent the header, we
507 					 * have no choice but to set pending and
508 					 * return BUSY.
509 					 */
510 					nsock->pending = req;
511 					nsock->sent = sent;
512 					return BLK_STS_RESOURCE;
513 				}
514 				dev_err(disk_to_dev(nbd->disk),
515 					"Send data failed (result %d)\n",
516 					result);
517 				return -EAGAIN;
518 			}
519 			/*
520 			 * The completion might already have come in,
521 			 * so break for the last one instead of letting
522 			 * the iterator do it. This prevents use-after-free
523 			 * of the bio.
524 			 */
525 			if (is_last)
526 				break;
527 		}
528 		bio = next;
529 	}
530 out:
531 	nsock->pending = NULL;
532 	nsock->sent = 0;
533 	return 0;
534 }
535 
536 /* NULL returned = something went wrong, inform userspace */
537 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
538 {
539 	struct nbd_config *config = nbd->config;
540 	int result;
541 	struct nbd_reply reply;
542 	struct nbd_cmd *cmd;
543 	struct request *req = NULL;
544 	u16 hwq;
545 	u32 tag;
546 	struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
547 	struct iov_iter to;
548 
549 	reply.magic = 0;
550 	iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
551 	result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
552 	if (result <= 0) {
553 		if (!nbd_disconnected(config))
554 			dev_err(disk_to_dev(nbd->disk),
555 				"Receive control failed (result %d)\n", result);
556 		return ERR_PTR(result);
557 	}
558 
559 	if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
560 		dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
561 				(unsigned long)ntohl(reply.magic));
562 		return ERR_PTR(-EPROTO);
563 	}
564 
565 	memcpy(&tag, reply.handle, sizeof(u32));
566 
567 	hwq = blk_mq_unique_tag_to_hwq(tag);
568 	if (hwq < nbd->tag_set.nr_hw_queues)
569 		req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
570 				       blk_mq_unique_tag_to_tag(tag));
571 	if (!req || !blk_mq_request_started(req)) {
572 		dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
573 			tag, req);
574 		return ERR_PTR(-ENOENT);
575 	}
576 	cmd = blk_mq_rq_to_pdu(req);
577 	if (ntohl(reply.error)) {
578 		dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
579 			ntohl(reply.error));
580 		cmd->status = BLK_STS_IOERR;
581 		return cmd;
582 	}
583 
584 	dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
585 	if (rq_data_dir(req) != WRITE) {
586 		struct req_iterator iter;
587 		struct bio_vec bvec;
588 
589 		rq_for_each_segment(bvec, req, iter) {
590 			iov_iter_bvec(&to, ITER_BVEC | READ,
591 				      &bvec, 1, bvec.bv_len);
592 			result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
593 			if (result <= 0) {
594 				dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
595 					result);
596 				/*
597 				 * If we've disconnected or we only have 1
598 				 * connection then we need to make sure we
599 				 * complete this request, otherwise error out
600 				 * and let the timeout stuff handle resubmitting
601 				 * this request onto another connection.
602 				 */
603 				if (nbd_disconnected(config) ||
604 				    config->num_connections <= 1) {
605 					cmd->status = BLK_STS_IOERR;
606 					return cmd;
607 				}
608 				return ERR_PTR(-EIO);
609 			}
610 			dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
611 				cmd, bvec.bv_len);
612 		}
613 	} else {
614 		/* See the comment in nbd_queue_rq. */
615 		wait_for_completion(&cmd->send_complete);
616 	}
617 	return cmd;
618 }
619 
620 static void recv_work(struct work_struct *work)
621 {
622 	struct recv_thread_args *args = container_of(work,
623 						     struct recv_thread_args,
624 						     work);
625 	struct nbd_device *nbd = args->nbd;
626 	struct nbd_config *config = nbd->config;
627 	struct nbd_cmd *cmd;
628 
629 	while (1) {
630 		cmd = nbd_read_stat(nbd, args->index);
631 		if (IS_ERR(cmd)) {
632 			struct nbd_sock *nsock = config->socks[args->index];
633 
634 			mutex_lock(&nsock->tx_lock);
635 			nbd_mark_nsock_dead(nbd, nsock, 1);
636 			mutex_unlock(&nsock->tx_lock);
637 			break;
638 		}
639 
640 		blk_mq_complete_request(blk_mq_rq_from_pdu(cmd));
641 	}
642 	atomic_dec(&config->recv_threads);
643 	wake_up(&config->recv_wq);
644 	nbd_config_put(nbd);
645 	kfree(args);
646 }
647 
648 static void nbd_clear_req(struct request *req, void *data, bool reserved)
649 {
650 	struct nbd_cmd *cmd;
651 
652 	if (!blk_mq_request_started(req))
653 		return;
654 	cmd = blk_mq_rq_to_pdu(req);
655 	cmd->status = BLK_STS_IOERR;
656 	blk_mq_complete_request(req);
657 }
658 
659 static void nbd_clear_que(struct nbd_device *nbd)
660 {
661 	blk_mq_quiesce_queue(nbd->disk->queue);
662 	blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
663 	blk_mq_unquiesce_queue(nbd->disk->queue);
664 	dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
665 }
666 
667 static int find_fallback(struct nbd_device *nbd, int index)
668 {
669 	struct nbd_config *config = nbd->config;
670 	int new_index = -1;
671 	struct nbd_sock *nsock = config->socks[index];
672 	int fallback = nsock->fallback_index;
673 
674 	if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
675 		return new_index;
676 
677 	if (config->num_connections <= 1) {
678 		dev_err_ratelimited(disk_to_dev(nbd->disk),
679 				    "Attempted send on invalid socket\n");
680 		return new_index;
681 	}
682 
683 	if (fallback >= 0 && fallback < config->num_connections &&
684 	    !config->socks[fallback]->dead)
685 		return fallback;
686 
687 	if (nsock->fallback_index < 0 ||
688 	    nsock->fallback_index >= config->num_connections ||
689 	    config->socks[nsock->fallback_index]->dead) {
690 		int i;
691 		for (i = 0; i < config->num_connections; i++) {
692 			if (i == index)
693 				continue;
694 			if (!config->socks[i]->dead) {
695 				new_index = i;
696 				break;
697 			}
698 		}
699 		nsock->fallback_index = new_index;
700 		if (new_index < 0) {
701 			dev_err_ratelimited(disk_to_dev(nbd->disk),
702 					    "Dead connection, failed to find a fallback\n");
703 			return new_index;
704 		}
705 	}
706 	new_index = nsock->fallback_index;
707 	return new_index;
708 }
709 
710 static int wait_for_reconnect(struct nbd_device *nbd)
711 {
712 	struct nbd_config *config = nbd->config;
713 	if (!config->dead_conn_timeout)
714 		return 0;
715 	if (test_bit(NBD_DISCONNECTED, &config->runtime_flags))
716 		return 0;
717 	wait_event_interruptible_timeout(config->conn_wait,
718 					 atomic_read(&config->live_connections),
719 					 config->dead_conn_timeout);
720 	return atomic_read(&config->live_connections);
721 }
722 
723 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
724 {
725 	struct request *req = blk_mq_rq_from_pdu(cmd);
726 	struct nbd_device *nbd = cmd->nbd;
727 	struct nbd_config *config;
728 	struct nbd_sock *nsock;
729 	int ret;
730 
731 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
732 		dev_err_ratelimited(disk_to_dev(nbd->disk),
733 				    "Socks array is empty\n");
734 		return -EINVAL;
735 	}
736 	config = nbd->config;
737 
738 	if (index >= config->num_connections) {
739 		dev_err_ratelimited(disk_to_dev(nbd->disk),
740 				    "Attempted send on invalid socket\n");
741 		nbd_config_put(nbd);
742 		return -EINVAL;
743 	}
744 	cmd->status = BLK_STS_OK;
745 again:
746 	nsock = config->socks[index];
747 	mutex_lock(&nsock->tx_lock);
748 	if (nsock->dead) {
749 		int old_index = index;
750 		index = find_fallback(nbd, index);
751 		mutex_unlock(&nsock->tx_lock);
752 		if (index < 0) {
753 			if (wait_for_reconnect(nbd)) {
754 				index = old_index;
755 				goto again;
756 			}
757 			/* All the sockets should already be down at this point,
758 			 * we just want to make sure that DISCONNECTED is set so
759 			 * any requests that come in that were queue'ed waiting
760 			 * for the reconnect timer don't trigger the timer again
761 			 * and instead just error out.
762 			 */
763 			sock_shutdown(nbd);
764 			nbd_config_put(nbd);
765 			return -EIO;
766 		}
767 		goto again;
768 	}
769 
770 	/* Handle the case that we have a pending request that was partially
771 	 * transmitted that _has_ to be serviced first.  We need to call requeue
772 	 * here so that it gets put _after_ the request that is already on the
773 	 * dispatch list.
774 	 */
775 	if (unlikely(nsock->pending && nsock->pending != req)) {
776 		blk_mq_requeue_request(req, true);
777 		ret = 0;
778 		goto out;
779 	}
780 	/*
781 	 * Some failures are related to the link going down, so anything that
782 	 * returns EAGAIN can be retried on a different socket.
783 	 */
784 	ret = nbd_send_cmd(nbd, cmd, index);
785 	if (ret == -EAGAIN) {
786 		dev_err_ratelimited(disk_to_dev(nbd->disk),
787 				    "Request send failed trying another connection\n");
788 		nbd_mark_nsock_dead(nbd, nsock, 1);
789 		mutex_unlock(&nsock->tx_lock);
790 		goto again;
791 	}
792 out:
793 	mutex_unlock(&nsock->tx_lock);
794 	nbd_config_put(nbd);
795 	return ret;
796 }
797 
798 static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
799 			const struct blk_mq_queue_data *bd)
800 {
801 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
802 	int ret;
803 
804 	/*
805 	 * Since we look at the bio's to send the request over the network we
806 	 * need to make sure the completion work doesn't mark this request done
807 	 * before we are done doing our send.  This keeps us from dereferencing
808 	 * freed data if we have particularly fast completions (ie we get the
809 	 * completion before we exit sock_xmit on the last bvec) or in the case
810 	 * that the server is misbehaving (or there was an error) before we're
811 	 * done sending everything over the wire.
812 	 */
813 	init_completion(&cmd->send_complete);
814 	blk_mq_start_request(bd->rq);
815 
816 	/* We can be called directly from the user space process, which means we
817 	 * could possibly have signals pending so our sendmsg will fail.  In
818 	 * this case we need to return that we are busy, otherwise error out as
819 	 * appropriate.
820 	 */
821 	ret = nbd_handle_cmd(cmd, hctx->queue_num);
822 	if (ret < 0)
823 		ret = BLK_STS_IOERR;
824 	else if (!ret)
825 		ret = BLK_STS_OK;
826 	complete(&cmd->send_complete);
827 
828 	return ret;
829 }
830 
831 static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
832 			  bool netlink)
833 {
834 	struct nbd_config *config = nbd->config;
835 	struct socket *sock;
836 	struct nbd_sock **socks;
837 	struct nbd_sock *nsock;
838 	int err;
839 
840 	sock = sockfd_lookup(arg, &err);
841 	if (!sock)
842 		return err;
843 
844 	if (!netlink && !nbd->task_setup &&
845 	    !test_bit(NBD_BOUND, &config->runtime_flags))
846 		nbd->task_setup = current;
847 
848 	if (!netlink &&
849 	    (nbd->task_setup != current ||
850 	     test_bit(NBD_BOUND, &config->runtime_flags))) {
851 		dev_err(disk_to_dev(nbd->disk),
852 			"Device being setup by another task");
853 		sockfd_put(sock);
854 		return -EBUSY;
855 	}
856 
857 	socks = krealloc(config->socks, (config->num_connections + 1) *
858 			 sizeof(struct nbd_sock *), GFP_KERNEL);
859 	if (!socks) {
860 		sockfd_put(sock);
861 		return -ENOMEM;
862 	}
863 	nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
864 	if (!nsock) {
865 		sockfd_put(sock);
866 		return -ENOMEM;
867 	}
868 
869 	config->socks = socks;
870 
871 	nsock->fallback_index = -1;
872 	nsock->dead = false;
873 	mutex_init(&nsock->tx_lock);
874 	nsock->sock = sock;
875 	nsock->pending = NULL;
876 	nsock->sent = 0;
877 	nsock->cookie = 0;
878 	socks[config->num_connections++] = nsock;
879 	atomic_inc(&config->live_connections);
880 
881 	return 0;
882 }
883 
884 static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
885 {
886 	struct nbd_config *config = nbd->config;
887 	struct socket *sock, *old;
888 	struct recv_thread_args *args;
889 	int i;
890 	int err;
891 
892 	sock = sockfd_lookup(arg, &err);
893 	if (!sock)
894 		return err;
895 
896 	args = kzalloc(sizeof(*args), GFP_KERNEL);
897 	if (!args) {
898 		sockfd_put(sock);
899 		return -ENOMEM;
900 	}
901 
902 	for (i = 0; i < config->num_connections; i++) {
903 		struct nbd_sock *nsock = config->socks[i];
904 
905 		if (!nsock->dead)
906 			continue;
907 
908 		mutex_lock(&nsock->tx_lock);
909 		if (!nsock->dead) {
910 			mutex_unlock(&nsock->tx_lock);
911 			continue;
912 		}
913 		sk_set_memalloc(sock->sk);
914 		if (nbd->tag_set.timeout)
915 			sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
916 		atomic_inc(&config->recv_threads);
917 		refcount_inc(&nbd->config_refs);
918 		old = nsock->sock;
919 		nsock->fallback_index = -1;
920 		nsock->sock = sock;
921 		nsock->dead = false;
922 		INIT_WORK(&args->work, recv_work);
923 		args->index = i;
924 		args->nbd = nbd;
925 		nsock->cookie++;
926 		mutex_unlock(&nsock->tx_lock);
927 		sockfd_put(old);
928 
929 		clear_bit(NBD_DISCONNECTED, &config->runtime_flags);
930 
931 		/* We take the tx_mutex in an error path in the recv_work, so we
932 		 * need to queue_work outside of the tx_mutex.
933 		 */
934 		queue_work(recv_workqueue, &args->work);
935 
936 		atomic_inc(&config->live_connections);
937 		wake_up(&config->conn_wait);
938 		return 0;
939 	}
940 	sockfd_put(sock);
941 	kfree(args);
942 	return -ENOSPC;
943 }
944 
945 static void nbd_bdev_reset(struct block_device *bdev)
946 {
947 	if (bdev->bd_openers > 1)
948 		return;
949 	bd_set_size(bdev, 0);
950 	if (max_part > 0) {
951 		blkdev_reread_part(bdev);
952 		bdev->bd_invalidated = 1;
953 	}
954 }
955 
956 static void nbd_parse_flags(struct nbd_device *nbd)
957 {
958 	struct nbd_config *config = nbd->config;
959 	if (config->flags & NBD_FLAG_READ_ONLY)
960 		set_disk_ro(nbd->disk, true);
961 	else
962 		set_disk_ro(nbd->disk, false);
963 	if (config->flags & NBD_FLAG_SEND_TRIM)
964 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
965 	if (config->flags & NBD_FLAG_SEND_FLUSH) {
966 		if (config->flags & NBD_FLAG_SEND_FUA)
967 			blk_queue_write_cache(nbd->disk->queue, true, true);
968 		else
969 			blk_queue_write_cache(nbd->disk->queue, true, false);
970 	}
971 	else
972 		blk_queue_write_cache(nbd->disk->queue, false, false);
973 }
974 
975 static void send_disconnects(struct nbd_device *nbd)
976 {
977 	struct nbd_config *config = nbd->config;
978 	struct nbd_request request = {
979 		.magic = htonl(NBD_REQUEST_MAGIC),
980 		.type = htonl(NBD_CMD_DISC),
981 	};
982 	struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
983 	struct iov_iter from;
984 	int i, ret;
985 
986 	for (i = 0; i < config->num_connections; i++) {
987 		struct nbd_sock *nsock = config->socks[i];
988 
989 		iov_iter_kvec(&from, WRITE | ITER_KVEC, &iov, 1, sizeof(request));
990 		mutex_lock(&nsock->tx_lock);
991 		ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
992 		if (ret <= 0)
993 			dev_err(disk_to_dev(nbd->disk),
994 				"Send disconnect failed %d\n", ret);
995 		mutex_unlock(&nsock->tx_lock);
996 	}
997 }
998 
999 static int nbd_disconnect(struct nbd_device *nbd)
1000 {
1001 	struct nbd_config *config = nbd->config;
1002 
1003 	dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1004 	set_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags);
1005 	send_disconnects(nbd);
1006 	return 0;
1007 }
1008 
1009 static void nbd_clear_sock(struct nbd_device *nbd)
1010 {
1011 	sock_shutdown(nbd);
1012 	nbd_clear_que(nbd);
1013 	nbd->task_setup = NULL;
1014 }
1015 
1016 static void nbd_config_put(struct nbd_device *nbd)
1017 {
1018 	if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1019 					&nbd->config_lock)) {
1020 		struct nbd_config *config = nbd->config;
1021 		nbd_dev_dbg_close(nbd);
1022 		nbd_size_clear(nbd);
1023 		if (test_and_clear_bit(NBD_HAS_PID_FILE,
1024 				       &config->runtime_flags))
1025 			device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1026 		nbd->task_recv = NULL;
1027 		nbd_clear_sock(nbd);
1028 		if (config->num_connections) {
1029 			int i;
1030 			for (i = 0; i < config->num_connections; i++) {
1031 				sockfd_put(config->socks[i]->sock);
1032 				kfree(config->socks[i]);
1033 			}
1034 			kfree(config->socks);
1035 		}
1036 		kfree(nbd->config);
1037 		nbd->config = NULL;
1038 
1039 		nbd->tag_set.timeout = 0;
1040 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1041 
1042 		mutex_unlock(&nbd->config_lock);
1043 		nbd_put(nbd);
1044 		module_put(THIS_MODULE);
1045 	}
1046 }
1047 
1048 static int nbd_start_device(struct nbd_device *nbd)
1049 {
1050 	struct nbd_config *config = nbd->config;
1051 	int num_connections = config->num_connections;
1052 	int error = 0, i;
1053 
1054 	if (nbd->task_recv)
1055 		return -EBUSY;
1056 	if (!config->socks)
1057 		return -EINVAL;
1058 	if (num_connections > 1 &&
1059 	    !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1060 		dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1061 		return -EINVAL;
1062 	}
1063 
1064 	blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1065 	nbd->task_recv = current;
1066 
1067 	nbd_parse_flags(nbd);
1068 
1069 	error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1070 	if (error) {
1071 		dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1072 		return error;
1073 	}
1074 	set_bit(NBD_HAS_PID_FILE, &config->runtime_flags);
1075 
1076 	nbd_dev_dbg_init(nbd);
1077 	for (i = 0; i < num_connections; i++) {
1078 		struct recv_thread_args *args;
1079 
1080 		args = kzalloc(sizeof(*args), GFP_KERNEL);
1081 		if (!args) {
1082 			sock_shutdown(nbd);
1083 			return -ENOMEM;
1084 		}
1085 		sk_set_memalloc(config->socks[i]->sock->sk);
1086 		if (nbd->tag_set.timeout)
1087 			config->socks[i]->sock->sk->sk_sndtimeo =
1088 				nbd->tag_set.timeout;
1089 		atomic_inc(&config->recv_threads);
1090 		refcount_inc(&nbd->config_refs);
1091 		INIT_WORK(&args->work, recv_work);
1092 		args->nbd = nbd;
1093 		args->index = i;
1094 		queue_work(recv_workqueue, &args->work);
1095 	}
1096 	nbd_size_update(nbd);
1097 	return error;
1098 }
1099 
1100 static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1101 {
1102 	struct nbd_config *config = nbd->config;
1103 	int ret;
1104 
1105 	ret = nbd_start_device(nbd);
1106 	if (ret)
1107 		return ret;
1108 
1109 	bd_set_size(bdev, config->bytesize);
1110 	if (max_part)
1111 		bdev->bd_invalidated = 1;
1112 	mutex_unlock(&nbd->config_lock);
1113 	ret = wait_event_interruptible(config->recv_wq,
1114 					 atomic_read(&config->recv_threads) == 0);
1115 	if (ret)
1116 		sock_shutdown(nbd);
1117 	mutex_lock(&nbd->config_lock);
1118 	bd_set_size(bdev, 0);
1119 	/* user requested, ignore socket errors */
1120 	if (test_bit(NBD_DISCONNECT_REQUESTED, &config->runtime_flags))
1121 		ret = 0;
1122 	if (test_bit(NBD_TIMEDOUT, &config->runtime_flags))
1123 		ret = -ETIMEDOUT;
1124 	return ret;
1125 }
1126 
1127 static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1128 				 struct block_device *bdev)
1129 {
1130 	sock_shutdown(nbd);
1131 	kill_bdev(bdev);
1132 	nbd_bdev_reset(bdev);
1133 	if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1134 			       &nbd->config->runtime_flags))
1135 		nbd_config_put(nbd);
1136 }
1137 
1138 /* Must be called with config_lock held */
1139 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1140 		       unsigned int cmd, unsigned long arg)
1141 {
1142 	struct nbd_config *config = nbd->config;
1143 
1144 	switch (cmd) {
1145 	case NBD_DISCONNECT:
1146 		return nbd_disconnect(nbd);
1147 	case NBD_CLEAR_SOCK:
1148 		nbd_clear_sock_ioctl(nbd, bdev);
1149 		return 0;
1150 	case NBD_SET_SOCK:
1151 		return nbd_add_socket(nbd, arg, false);
1152 	case NBD_SET_BLKSIZE:
1153 		nbd_size_set(nbd, arg,
1154 			     div_s64(config->bytesize, arg));
1155 		return 0;
1156 	case NBD_SET_SIZE:
1157 		nbd_size_set(nbd, config->blksize,
1158 			     div_s64(arg, config->blksize));
1159 		return 0;
1160 	case NBD_SET_SIZE_BLOCKS:
1161 		nbd_size_set(nbd, config->blksize, arg);
1162 		return 0;
1163 	case NBD_SET_TIMEOUT:
1164 		if (arg) {
1165 			nbd->tag_set.timeout = arg * HZ;
1166 			blk_queue_rq_timeout(nbd->disk->queue, arg * HZ);
1167 		}
1168 		return 0;
1169 
1170 	case NBD_SET_FLAGS:
1171 		config->flags = arg;
1172 		return 0;
1173 	case NBD_DO_IT:
1174 		return nbd_start_device_ioctl(nbd, bdev);
1175 	case NBD_CLEAR_QUE:
1176 		/*
1177 		 * This is for compatibility only.  The queue is always cleared
1178 		 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1179 		 */
1180 		return 0;
1181 	case NBD_PRINT_DEBUG:
1182 		/*
1183 		 * For compatibility only, we no longer keep a list of
1184 		 * outstanding requests.
1185 		 */
1186 		return 0;
1187 	}
1188 	return -ENOTTY;
1189 }
1190 
1191 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1192 		     unsigned int cmd, unsigned long arg)
1193 {
1194 	struct nbd_device *nbd = bdev->bd_disk->private_data;
1195 	struct nbd_config *config = nbd->config;
1196 	int error = -EINVAL;
1197 
1198 	if (!capable(CAP_SYS_ADMIN))
1199 		return -EPERM;
1200 
1201 	/* The block layer will pass back some non-nbd ioctls in case we have
1202 	 * special handling for them, but we don't so just return an error.
1203 	 */
1204 	if (_IOC_TYPE(cmd) != 0xab)
1205 		return -EINVAL;
1206 
1207 	mutex_lock(&nbd->config_lock);
1208 
1209 	/* Don't allow ioctl operations on a nbd device that was created with
1210 	 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1211 	 */
1212 	if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1213 	    (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1214 		error = __nbd_ioctl(bdev, nbd, cmd, arg);
1215 	else
1216 		dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1217 	mutex_unlock(&nbd->config_lock);
1218 	return error;
1219 }
1220 
1221 static struct nbd_config *nbd_alloc_config(void)
1222 {
1223 	struct nbd_config *config;
1224 
1225 	config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1226 	if (!config)
1227 		return NULL;
1228 	atomic_set(&config->recv_threads, 0);
1229 	init_waitqueue_head(&config->recv_wq);
1230 	init_waitqueue_head(&config->conn_wait);
1231 	config->blksize = 1024;
1232 	atomic_set(&config->live_connections, 0);
1233 	try_module_get(THIS_MODULE);
1234 	return config;
1235 }
1236 
1237 static int nbd_open(struct block_device *bdev, fmode_t mode)
1238 {
1239 	struct nbd_device *nbd;
1240 	int ret = 0;
1241 
1242 	mutex_lock(&nbd_index_mutex);
1243 	nbd = bdev->bd_disk->private_data;
1244 	if (!nbd) {
1245 		ret = -ENXIO;
1246 		goto out;
1247 	}
1248 	if (!refcount_inc_not_zero(&nbd->refs)) {
1249 		ret = -ENXIO;
1250 		goto out;
1251 	}
1252 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
1253 		struct nbd_config *config;
1254 
1255 		mutex_lock(&nbd->config_lock);
1256 		if (refcount_inc_not_zero(&nbd->config_refs)) {
1257 			mutex_unlock(&nbd->config_lock);
1258 			goto out;
1259 		}
1260 		config = nbd->config = nbd_alloc_config();
1261 		if (!config) {
1262 			ret = -ENOMEM;
1263 			mutex_unlock(&nbd->config_lock);
1264 			goto out;
1265 		}
1266 		refcount_set(&nbd->config_refs, 1);
1267 		refcount_inc(&nbd->refs);
1268 		mutex_unlock(&nbd->config_lock);
1269 	}
1270 out:
1271 	mutex_unlock(&nbd_index_mutex);
1272 	return ret;
1273 }
1274 
1275 static void nbd_release(struct gendisk *disk, fmode_t mode)
1276 {
1277 	struct nbd_device *nbd = disk->private_data;
1278 	nbd_config_put(nbd);
1279 	nbd_put(nbd);
1280 }
1281 
1282 static const struct block_device_operations nbd_fops =
1283 {
1284 	.owner =	THIS_MODULE,
1285 	.open =		nbd_open,
1286 	.release =	nbd_release,
1287 	.ioctl =	nbd_ioctl,
1288 	.compat_ioctl =	nbd_ioctl,
1289 };
1290 
1291 #if IS_ENABLED(CONFIG_DEBUG_FS)
1292 
1293 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1294 {
1295 	struct nbd_device *nbd = s->private;
1296 
1297 	if (nbd->task_recv)
1298 		seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1299 
1300 	return 0;
1301 }
1302 
1303 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1304 {
1305 	return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1306 }
1307 
1308 static const struct file_operations nbd_dbg_tasks_ops = {
1309 	.open = nbd_dbg_tasks_open,
1310 	.read = seq_read,
1311 	.llseek = seq_lseek,
1312 	.release = single_release,
1313 };
1314 
1315 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1316 {
1317 	struct nbd_device *nbd = s->private;
1318 	u32 flags = nbd->config->flags;
1319 
1320 	seq_printf(s, "Hex: 0x%08x\n\n", flags);
1321 
1322 	seq_puts(s, "Known flags:\n");
1323 
1324 	if (flags & NBD_FLAG_HAS_FLAGS)
1325 		seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1326 	if (flags & NBD_FLAG_READ_ONLY)
1327 		seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1328 	if (flags & NBD_FLAG_SEND_FLUSH)
1329 		seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1330 	if (flags & NBD_FLAG_SEND_FUA)
1331 		seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1332 	if (flags & NBD_FLAG_SEND_TRIM)
1333 		seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1334 
1335 	return 0;
1336 }
1337 
1338 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1339 {
1340 	return single_open(file, nbd_dbg_flags_show, inode->i_private);
1341 }
1342 
1343 static const struct file_operations nbd_dbg_flags_ops = {
1344 	.open = nbd_dbg_flags_open,
1345 	.read = seq_read,
1346 	.llseek = seq_lseek,
1347 	.release = single_release,
1348 };
1349 
1350 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1351 {
1352 	struct dentry *dir;
1353 	struct nbd_config *config = nbd->config;
1354 
1355 	if (!nbd_dbg_dir)
1356 		return -EIO;
1357 
1358 	dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1359 	if (!dir) {
1360 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1361 			nbd_name(nbd));
1362 		return -EIO;
1363 	}
1364 	config->dbg_dir = dir;
1365 
1366 	debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1367 	debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1368 	debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1369 	debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1370 	debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1371 
1372 	return 0;
1373 }
1374 
1375 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1376 {
1377 	debugfs_remove_recursive(nbd->config->dbg_dir);
1378 }
1379 
1380 static int nbd_dbg_init(void)
1381 {
1382 	struct dentry *dbg_dir;
1383 
1384 	dbg_dir = debugfs_create_dir("nbd", NULL);
1385 	if (!dbg_dir)
1386 		return -EIO;
1387 
1388 	nbd_dbg_dir = dbg_dir;
1389 
1390 	return 0;
1391 }
1392 
1393 static void nbd_dbg_close(void)
1394 {
1395 	debugfs_remove_recursive(nbd_dbg_dir);
1396 }
1397 
1398 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
1399 
1400 static int nbd_dev_dbg_init(struct nbd_device *nbd)
1401 {
1402 	return 0;
1403 }
1404 
1405 static void nbd_dev_dbg_close(struct nbd_device *nbd)
1406 {
1407 }
1408 
1409 static int nbd_dbg_init(void)
1410 {
1411 	return 0;
1412 }
1413 
1414 static void nbd_dbg_close(void)
1415 {
1416 }
1417 
1418 #endif
1419 
1420 static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1421 			    unsigned int hctx_idx, unsigned int numa_node)
1422 {
1423 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1424 	cmd->nbd = set->driver_data;
1425 	return 0;
1426 }
1427 
1428 static const struct blk_mq_ops nbd_mq_ops = {
1429 	.queue_rq	= nbd_queue_rq,
1430 	.complete	= nbd_complete_rq,
1431 	.init_request	= nbd_init_request,
1432 	.timeout	= nbd_xmit_timeout,
1433 };
1434 
1435 static int nbd_dev_add(int index)
1436 {
1437 	struct nbd_device *nbd;
1438 	struct gendisk *disk;
1439 	struct request_queue *q;
1440 	int err = -ENOMEM;
1441 
1442 	nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1443 	if (!nbd)
1444 		goto out;
1445 
1446 	disk = alloc_disk(1 << part_shift);
1447 	if (!disk)
1448 		goto out_free_nbd;
1449 
1450 	if (index >= 0) {
1451 		err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1452 				GFP_KERNEL);
1453 		if (err == -ENOSPC)
1454 			err = -EEXIST;
1455 	} else {
1456 		err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1457 		if (err >= 0)
1458 			index = err;
1459 	}
1460 	if (err < 0)
1461 		goto out_free_disk;
1462 
1463 	nbd->index = index;
1464 	nbd->disk = disk;
1465 	nbd->tag_set.ops = &nbd_mq_ops;
1466 	nbd->tag_set.nr_hw_queues = 1;
1467 	nbd->tag_set.queue_depth = 128;
1468 	nbd->tag_set.numa_node = NUMA_NO_NODE;
1469 	nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1470 	nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1471 		BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1472 	nbd->tag_set.driver_data = nbd;
1473 
1474 	err = blk_mq_alloc_tag_set(&nbd->tag_set);
1475 	if (err)
1476 		goto out_free_idr;
1477 
1478 	q = blk_mq_init_queue(&nbd->tag_set);
1479 	if (IS_ERR(q)) {
1480 		err = PTR_ERR(q);
1481 		goto out_free_tags;
1482 	}
1483 	disk->queue = q;
1484 
1485 	/*
1486 	 * Tell the block layer that we are not a rotational device
1487 	 */
1488 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1489 	queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1490 	disk->queue->limits.discard_granularity = 512;
1491 	blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1492 	blk_queue_max_segment_size(disk->queue, UINT_MAX);
1493 	blk_queue_max_segments(disk->queue, USHRT_MAX);
1494 	blk_queue_max_hw_sectors(disk->queue, 65536);
1495 	disk->queue->limits.max_sectors = 256;
1496 
1497 	mutex_init(&nbd->config_lock);
1498 	refcount_set(&nbd->config_refs, 0);
1499 	refcount_set(&nbd->refs, 1);
1500 	INIT_LIST_HEAD(&nbd->list);
1501 	disk->major = NBD_MAJOR;
1502 	disk->first_minor = index << part_shift;
1503 	disk->fops = &nbd_fops;
1504 	disk->private_data = nbd;
1505 	sprintf(disk->disk_name, "nbd%d", index);
1506 	add_disk(disk);
1507 	nbd_total_devices++;
1508 	return index;
1509 
1510 out_free_tags:
1511 	blk_mq_free_tag_set(&nbd->tag_set);
1512 out_free_idr:
1513 	idr_remove(&nbd_index_idr, index);
1514 out_free_disk:
1515 	put_disk(disk);
1516 out_free_nbd:
1517 	kfree(nbd);
1518 out:
1519 	return err;
1520 }
1521 
1522 static int find_free_cb(int id, void *ptr, void *data)
1523 {
1524 	struct nbd_device *nbd = ptr;
1525 	struct nbd_device **found = data;
1526 
1527 	if (!refcount_read(&nbd->config_refs)) {
1528 		*found = nbd;
1529 		return 1;
1530 	}
1531 	return 0;
1532 }
1533 
1534 /* Netlink interface. */
1535 static struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1536 	[NBD_ATTR_INDEX]		=	{ .type = NLA_U32 },
1537 	[NBD_ATTR_SIZE_BYTES]		=	{ .type = NLA_U64 },
1538 	[NBD_ATTR_BLOCK_SIZE_BYTES]	=	{ .type = NLA_U64 },
1539 	[NBD_ATTR_TIMEOUT]		=	{ .type = NLA_U64 },
1540 	[NBD_ATTR_SERVER_FLAGS]		=	{ .type = NLA_U64 },
1541 	[NBD_ATTR_CLIENT_FLAGS]		=	{ .type = NLA_U64 },
1542 	[NBD_ATTR_SOCKETS]		=	{ .type = NLA_NESTED},
1543 	[NBD_ATTR_DEAD_CONN_TIMEOUT]	=	{ .type = NLA_U64 },
1544 	[NBD_ATTR_DEVICE_LIST]		=	{ .type = NLA_NESTED},
1545 };
1546 
1547 static struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1548 	[NBD_SOCK_FD]			=	{ .type = NLA_U32 },
1549 };
1550 
1551 /* We don't use this right now since we don't parse the incoming list, but we
1552  * still want it here so userspace knows what to expect.
1553  */
1554 static struct nla_policy __attribute__((unused))
1555 nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1556 	[NBD_DEVICE_INDEX]		=	{ .type = NLA_U32 },
1557 	[NBD_DEVICE_CONNECTED]		=	{ .type = NLA_U8 },
1558 };
1559 
1560 static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1561 {
1562 	struct nbd_device *nbd = NULL;
1563 	struct nbd_config *config;
1564 	int index = -1;
1565 	int ret;
1566 	bool put_dev = false;
1567 
1568 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1569 		return -EPERM;
1570 
1571 	if (info->attrs[NBD_ATTR_INDEX])
1572 		index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1573 	if (!info->attrs[NBD_ATTR_SOCKETS]) {
1574 		printk(KERN_ERR "nbd: must specify at least one socket\n");
1575 		return -EINVAL;
1576 	}
1577 	if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1578 		printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1579 		return -EINVAL;
1580 	}
1581 again:
1582 	mutex_lock(&nbd_index_mutex);
1583 	if (index == -1) {
1584 		ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1585 		if (ret == 0) {
1586 			int new_index;
1587 			new_index = nbd_dev_add(-1);
1588 			if (new_index < 0) {
1589 				mutex_unlock(&nbd_index_mutex);
1590 				printk(KERN_ERR "nbd: failed to add new device\n");
1591 				return ret;
1592 			}
1593 			nbd = idr_find(&nbd_index_idr, new_index);
1594 		}
1595 	} else {
1596 		nbd = idr_find(&nbd_index_idr, index);
1597 		if (!nbd) {
1598 			ret = nbd_dev_add(index);
1599 			if (ret < 0) {
1600 				mutex_unlock(&nbd_index_mutex);
1601 				printk(KERN_ERR "nbd: failed to add new device\n");
1602 				return ret;
1603 			}
1604 			nbd = idr_find(&nbd_index_idr, index);
1605 		}
1606 	}
1607 	if (!nbd) {
1608 		printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1609 		       index);
1610 		mutex_unlock(&nbd_index_mutex);
1611 		return -EINVAL;
1612 	}
1613 	if (!refcount_inc_not_zero(&nbd->refs)) {
1614 		mutex_unlock(&nbd_index_mutex);
1615 		if (index == -1)
1616 			goto again;
1617 		printk(KERN_ERR "nbd: device at index %d is going down\n",
1618 		       index);
1619 		return -EINVAL;
1620 	}
1621 	mutex_unlock(&nbd_index_mutex);
1622 
1623 	mutex_lock(&nbd->config_lock);
1624 	if (refcount_read(&nbd->config_refs)) {
1625 		mutex_unlock(&nbd->config_lock);
1626 		nbd_put(nbd);
1627 		if (index == -1)
1628 			goto again;
1629 		printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1630 		return -EBUSY;
1631 	}
1632 	if (WARN_ON(nbd->config)) {
1633 		mutex_unlock(&nbd->config_lock);
1634 		nbd_put(nbd);
1635 		return -EINVAL;
1636 	}
1637 	config = nbd->config = nbd_alloc_config();
1638 	if (!nbd->config) {
1639 		mutex_unlock(&nbd->config_lock);
1640 		nbd_put(nbd);
1641 		printk(KERN_ERR "nbd: couldn't allocate config\n");
1642 		return -ENOMEM;
1643 	}
1644 	refcount_set(&nbd->config_refs, 1);
1645 	set_bit(NBD_BOUND, &config->runtime_flags);
1646 
1647 	if (info->attrs[NBD_ATTR_SIZE_BYTES]) {
1648 		u64 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1649 		nbd_size_set(nbd, config->blksize,
1650 			     div64_u64(bytes, config->blksize));
1651 	}
1652 	if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1653 		u64 bsize =
1654 			nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1655 		nbd_size_set(nbd, bsize, div64_u64(config->bytesize, bsize));
1656 	}
1657 	if (info->attrs[NBD_ATTR_TIMEOUT]) {
1658 		u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1659 		nbd->tag_set.timeout = timeout * HZ;
1660 		blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1661 	}
1662 	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1663 		config->dead_conn_timeout =
1664 			nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1665 		config->dead_conn_timeout *= HZ;
1666 	}
1667 	if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1668 		config->flags =
1669 			nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1670 	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1671 		u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1672 		if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1673 			set_bit(NBD_DESTROY_ON_DISCONNECT,
1674 				&config->runtime_flags);
1675 			put_dev = true;
1676 		}
1677 	}
1678 
1679 	if (info->attrs[NBD_ATTR_SOCKETS]) {
1680 		struct nlattr *attr;
1681 		int rem, fd;
1682 
1683 		nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1684 				    rem) {
1685 			struct nlattr *socks[NBD_SOCK_MAX+1];
1686 
1687 			if (nla_type(attr) != NBD_SOCK_ITEM) {
1688 				printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1689 				ret = -EINVAL;
1690 				goto out;
1691 			}
1692 			ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1693 					       nbd_sock_policy, info->extack);
1694 			if (ret != 0) {
1695 				printk(KERN_ERR "nbd: error processing sock list\n");
1696 				ret = -EINVAL;
1697 				goto out;
1698 			}
1699 			if (!socks[NBD_SOCK_FD])
1700 				continue;
1701 			fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1702 			ret = nbd_add_socket(nbd, fd, true);
1703 			if (ret)
1704 				goto out;
1705 		}
1706 	}
1707 	ret = nbd_start_device(nbd);
1708 out:
1709 	mutex_unlock(&nbd->config_lock);
1710 	if (!ret) {
1711 		set_bit(NBD_HAS_CONFIG_REF, &config->runtime_flags);
1712 		refcount_inc(&nbd->config_refs);
1713 		nbd_connect_reply(info, nbd->index);
1714 	}
1715 	nbd_config_put(nbd);
1716 	if (put_dev)
1717 		nbd_put(nbd);
1718 	return ret;
1719 }
1720 
1721 static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
1722 {
1723 	struct nbd_device *nbd;
1724 	int index;
1725 
1726 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1727 		return -EPERM;
1728 
1729 	if (!info->attrs[NBD_ATTR_INDEX]) {
1730 		printk(KERN_ERR "nbd: must specify an index to disconnect\n");
1731 		return -EINVAL;
1732 	}
1733 	index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1734 	mutex_lock(&nbd_index_mutex);
1735 	nbd = idr_find(&nbd_index_idr, index);
1736 	if (!nbd) {
1737 		mutex_unlock(&nbd_index_mutex);
1738 		printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1739 		       index);
1740 		return -EINVAL;
1741 	}
1742 	if (!refcount_inc_not_zero(&nbd->refs)) {
1743 		mutex_unlock(&nbd_index_mutex);
1744 		printk(KERN_ERR "nbd: device at index %d is going down\n",
1745 		       index);
1746 		return -EINVAL;
1747 	}
1748 	mutex_unlock(&nbd_index_mutex);
1749 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
1750 		nbd_put(nbd);
1751 		return 0;
1752 	}
1753 	mutex_lock(&nbd->config_lock);
1754 	nbd_disconnect(nbd);
1755 	mutex_unlock(&nbd->config_lock);
1756 	if (test_and_clear_bit(NBD_HAS_CONFIG_REF,
1757 			       &nbd->config->runtime_flags))
1758 		nbd_config_put(nbd);
1759 	nbd_config_put(nbd);
1760 	nbd_put(nbd);
1761 	return 0;
1762 }
1763 
1764 static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
1765 {
1766 	struct nbd_device *nbd = NULL;
1767 	struct nbd_config *config;
1768 	int index;
1769 	int ret = -EINVAL;
1770 	bool put_dev = false;
1771 
1772 	if (!netlink_capable(skb, CAP_SYS_ADMIN))
1773 		return -EPERM;
1774 
1775 	if (!info->attrs[NBD_ATTR_INDEX]) {
1776 		printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
1777 		return -EINVAL;
1778 	}
1779 	index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1780 	mutex_lock(&nbd_index_mutex);
1781 	nbd = idr_find(&nbd_index_idr, index);
1782 	if (!nbd) {
1783 		mutex_unlock(&nbd_index_mutex);
1784 		printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
1785 		       index);
1786 		return -EINVAL;
1787 	}
1788 	if (!refcount_inc_not_zero(&nbd->refs)) {
1789 		mutex_unlock(&nbd_index_mutex);
1790 		printk(KERN_ERR "nbd: device at index %d is going down\n",
1791 		       index);
1792 		return -EINVAL;
1793 	}
1794 	mutex_unlock(&nbd_index_mutex);
1795 
1796 	if (!refcount_inc_not_zero(&nbd->config_refs)) {
1797 		dev_err(nbd_to_dev(nbd),
1798 			"not configured, cannot reconfigure\n");
1799 		nbd_put(nbd);
1800 		return -EINVAL;
1801 	}
1802 
1803 	mutex_lock(&nbd->config_lock);
1804 	config = nbd->config;
1805 	if (!test_bit(NBD_BOUND, &config->runtime_flags) ||
1806 	    !nbd->task_recv) {
1807 		dev_err(nbd_to_dev(nbd),
1808 			"not configured, cannot reconfigure\n");
1809 		goto out;
1810 	}
1811 
1812 	if (info->attrs[NBD_ATTR_TIMEOUT]) {
1813 		u64 timeout = nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]);
1814 		nbd->tag_set.timeout = timeout * HZ;
1815 		blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1816 	}
1817 	if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1818 		config->dead_conn_timeout =
1819 			nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1820 		config->dead_conn_timeout *= HZ;
1821 	}
1822 	if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1823 		u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1824 		if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1825 			if (!test_and_set_bit(NBD_DESTROY_ON_DISCONNECT,
1826 					      &config->runtime_flags))
1827 				put_dev = true;
1828 		} else {
1829 			if (test_and_clear_bit(NBD_DESTROY_ON_DISCONNECT,
1830 					       &config->runtime_flags))
1831 				refcount_inc(&nbd->refs);
1832 		}
1833 	}
1834 
1835 	if (info->attrs[NBD_ATTR_SOCKETS]) {
1836 		struct nlattr *attr;
1837 		int rem, fd;
1838 
1839 		nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1840 				    rem) {
1841 			struct nlattr *socks[NBD_SOCK_MAX+1];
1842 
1843 			if (nla_type(attr) != NBD_SOCK_ITEM) {
1844 				printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1845 				ret = -EINVAL;
1846 				goto out;
1847 			}
1848 			ret = nla_parse_nested(socks, NBD_SOCK_MAX, attr,
1849 					       nbd_sock_policy, info->extack);
1850 			if (ret != 0) {
1851 				printk(KERN_ERR "nbd: error processing sock list\n");
1852 				ret = -EINVAL;
1853 				goto out;
1854 			}
1855 			if (!socks[NBD_SOCK_FD])
1856 				continue;
1857 			fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1858 			ret = nbd_reconnect_socket(nbd, fd);
1859 			if (ret) {
1860 				if (ret == -ENOSPC)
1861 					ret = 0;
1862 				goto out;
1863 			}
1864 			dev_info(nbd_to_dev(nbd), "reconnected socket\n");
1865 		}
1866 	}
1867 out:
1868 	mutex_unlock(&nbd->config_lock);
1869 	nbd_config_put(nbd);
1870 	nbd_put(nbd);
1871 	if (put_dev)
1872 		nbd_put(nbd);
1873 	return ret;
1874 }
1875 
1876 static const struct genl_ops nbd_connect_genl_ops[] = {
1877 	{
1878 		.cmd	= NBD_CMD_CONNECT,
1879 		.policy	= nbd_attr_policy,
1880 		.doit	= nbd_genl_connect,
1881 	},
1882 	{
1883 		.cmd	= NBD_CMD_DISCONNECT,
1884 		.policy	= nbd_attr_policy,
1885 		.doit	= nbd_genl_disconnect,
1886 	},
1887 	{
1888 		.cmd	= NBD_CMD_RECONFIGURE,
1889 		.policy	= nbd_attr_policy,
1890 		.doit	= nbd_genl_reconfigure,
1891 	},
1892 	{
1893 		.cmd	= NBD_CMD_STATUS,
1894 		.policy	= nbd_attr_policy,
1895 		.doit	= nbd_genl_status,
1896 	},
1897 };
1898 
1899 static const struct genl_multicast_group nbd_mcast_grps[] = {
1900 	{ .name = NBD_GENL_MCAST_GROUP_NAME, },
1901 };
1902 
1903 static struct genl_family nbd_genl_family __ro_after_init = {
1904 	.hdrsize	= 0,
1905 	.name		= NBD_GENL_FAMILY_NAME,
1906 	.version	= NBD_GENL_VERSION,
1907 	.module		= THIS_MODULE,
1908 	.ops		= nbd_connect_genl_ops,
1909 	.n_ops		= ARRAY_SIZE(nbd_connect_genl_ops),
1910 	.maxattr	= NBD_ATTR_MAX,
1911 	.mcgrps		= nbd_mcast_grps,
1912 	.n_mcgrps	= ARRAY_SIZE(nbd_mcast_grps),
1913 };
1914 
1915 static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
1916 {
1917 	struct nlattr *dev_opt;
1918 	u8 connected = 0;
1919 	int ret;
1920 
1921 	/* This is a little racey, but for status it's ok.  The
1922 	 * reason we don't take a ref here is because we can't
1923 	 * take a ref in the index == -1 case as we would need
1924 	 * to put under the nbd_index_mutex, which could
1925 	 * deadlock if we are configured to remove ourselves
1926 	 * once we're disconnected.
1927 	 */
1928 	if (refcount_read(&nbd->config_refs))
1929 		connected = 1;
1930 	dev_opt = nla_nest_start(reply, NBD_DEVICE_ITEM);
1931 	if (!dev_opt)
1932 		return -EMSGSIZE;
1933 	ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
1934 	if (ret)
1935 		return -EMSGSIZE;
1936 	ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
1937 			 connected);
1938 	if (ret)
1939 		return -EMSGSIZE;
1940 	nla_nest_end(reply, dev_opt);
1941 	return 0;
1942 }
1943 
1944 static int status_cb(int id, void *ptr, void *data)
1945 {
1946 	struct nbd_device *nbd = ptr;
1947 	return populate_nbd_status(nbd, (struct sk_buff *)data);
1948 }
1949 
1950 static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
1951 {
1952 	struct nlattr *dev_list;
1953 	struct sk_buff *reply;
1954 	void *reply_head;
1955 	size_t msg_size;
1956 	int index = -1;
1957 	int ret = -ENOMEM;
1958 
1959 	if (info->attrs[NBD_ATTR_INDEX])
1960 		index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1961 
1962 	mutex_lock(&nbd_index_mutex);
1963 
1964 	msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
1965 				  nla_attr_size(sizeof(u8)));
1966 	msg_size *= (index == -1) ? nbd_total_devices : 1;
1967 
1968 	reply = genlmsg_new(msg_size, GFP_KERNEL);
1969 	if (!reply)
1970 		goto out;
1971 	reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
1972 				       NBD_CMD_STATUS);
1973 	if (!reply_head) {
1974 		nlmsg_free(reply);
1975 		goto out;
1976 	}
1977 
1978 	dev_list = nla_nest_start(reply, NBD_ATTR_DEVICE_LIST);
1979 	if (index == -1) {
1980 		ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
1981 		if (ret) {
1982 			nlmsg_free(reply);
1983 			goto out;
1984 		}
1985 	} else {
1986 		struct nbd_device *nbd;
1987 		nbd = idr_find(&nbd_index_idr, index);
1988 		if (nbd) {
1989 			ret = populate_nbd_status(nbd, reply);
1990 			if (ret) {
1991 				nlmsg_free(reply);
1992 				goto out;
1993 			}
1994 		}
1995 	}
1996 	nla_nest_end(reply, dev_list);
1997 	genlmsg_end(reply, reply_head);
1998 	genlmsg_reply(reply, info);
1999 	ret = 0;
2000 out:
2001 	mutex_unlock(&nbd_index_mutex);
2002 	return ret;
2003 }
2004 
2005 static void nbd_connect_reply(struct genl_info *info, int index)
2006 {
2007 	struct sk_buff *skb;
2008 	void *msg_head;
2009 	int ret;
2010 
2011 	skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2012 	if (!skb)
2013 		return;
2014 	msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2015 				     NBD_CMD_CONNECT);
2016 	if (!msg_head) {
2017 		nlmsg_free(skb);
2018 		return;
2019 	}
2020 	ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2021 	if (ret) {
2022 		nlmsg_free(skb);
2023 		return;
2024 	}
2025 	genlmsg_end(skb, msg_head);
2026 	genlmsg_reply(skb, info);
2027 }
2028 
2029 static void nbd_mcast_index(int index)
2030 {
2031 	struct sk_buff *skb;
2032 	void *msg_head;
2033 	int ret;
2034 
2035 	skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2036 	if (!skb)
2037 		return;
2038 	msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2039 				     NBD_CMD_LINK_DEAD);
2040 	if (!msg_head) {
2041 		nlmsg_free(skb);
2042 		return;
2043 	}
2044 	ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2045 	if (ret) {
2046 		nlmsg_free(skb);
2047 		return;
2048 	}
2049 	genlmsg_end(skb, msg_head);
2050 	genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2051 }
2052 
2053 static void nbd_dead_link_work(struct work_struct *work)
2054 {
2055 	struct link_dead_args *args = container_of(work, struct link_dead_args,
2056 						   work);
2057 	nbd_mcast_index(args->index);
2058 	kfree(args);
2059 }
2060 
2061 static int __init nbd_init(void)
2062 {
2063 	int i;
2064 
2065 	BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2066 
2067 	if (max_part < 0) {
2068 		printk(KERN_ERR "nbd: max_part must be >= 0\n");
2069 		return -EINVAL;
2070 	}
2071 
2072 	part_shift = 0;
2073 	if (max_part > 0) {
2074 		part_shift = fls(max_part);
2075 
2076 		/*
2077 		 * Adjust max_part according to part_shift as it is exported
2078 		 * to user space so that user can know the max number of
2079 		 * partition kernel should be able to manage.
2080 		 *
2081 		 * Note that -1 is required because partition 0 is reserved
2082 		 * for the whole disk.
2083 		 */
2084 		max_part = (1UL << part_shift) - 1;
2085 	}
2086 
2087 	if ((1UL << part_shift) > DISK_MAX_PARTS)
2088 		return -EINVAL;
2089 
2090 	if (nbds_max > 1UL << (MINORBITS - part_shift))
2091 		return -EINVAL;
2092 	recv_workqueue = alloc_workqueue("knbd-recv",
2093 					 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2094 	if (!recv_workqueue)
2095 		return -ENOMEM;
2096 
2097 	if (register_blkdev(NBD_MAJOR, "nbd")) {
2098 		destroy_workqueue(recv_workqueue);
2099 		return -EIO;
2100 	}
2101 
2102 	if (genl_register_family(&nbd_genl_family)) {
2103 		unregister_blkdev(NBD_MAJOR, "nbd");
2104 		destroy_workqueue(recv_workqueue);
2105 		return -EINVAL;
2106 	}
2107 	nbd_dbg_init();
2108 
2109 	mutex_lock(&nbd_index_mutex);
2110 	for (i = 0; i < nbds_max; i++)
2111 		nbd_dev_add(i);
2112 	mutex_unlock(&nbd_index_mutex);
2113 	return 0;
2114 }
2115 
2116 static int nbd_exit_cb(int id, void *ptr, void *data)
2117 {
2118 	struct list_head *list = (struct list_head *)data;
2119 	struct nbd_device *nbd = ptr;
2120 
2121 	list_add_tail(&nbd->list, list);
2122 	return 0;
2123 }
2124 
2125 static void __exit nbd_cleanup(void)
2126 {
2127 	struct nbd_device *nbd;
2128 	LIST_HEAD(del_list);
2129 
2130 	nbd_dbg_close();
2131 
2132 	mutex_lock(&nbd_index_mutex);
2133 	idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2134 	mutex_unlock(&nbd_index_mutex);
2135 
2136 	while (!list_empty(&del_list)) {
2137 		nbd = list_first_entry(&del_list, struct nbd_device, list);
2138 		list_del_init(&nbd->list);
2139 		if (refcount_read(&nbd->refs) != 1)
2140 			printk(KERN_ERR "nbd: possibly leaking a device\n");
2141 		nbd_put(nbd);
2142 	}
2143 
2144 	idr_destroy(&nbd_index_idr);
2145 	genl_unregister_family(&nbd_genl_family);
2146 	destroy_workqueue(recv_workqueue);
2147 	unregister_blkdev(NBD_MAJOR, "nbd");
2148 }
2149 
2150 module_init(nbd_init);
2151 module_exit(nbd_cleanup);
2152 
2153 MODULE_DESCRIPTION("Network Block Device");
2154 MODULE_LICENSE("GPL");
2155 
2156 module_param(nbds_max, int, 0444);
2157 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2158 module_param(max_part, int, 0444);
2159 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");
2160