xref: /linux/drivers/block/nbd.c (revision 8c749ce93ee69e789e46b3be98de9e0cbfcf8ed8)
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/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/err.h>
30 #include <linux/kernel.h>
31 #include <linux/slab.h>
32 #include <net/sock.h>
33 #include <linux/net.h>
34 #include <linux/kthread.h>
35 #include <linux/types.h>
36 #include <linux/debugfs.h>
37 
38 #include <asm/uaccess.h>
39 #include <asm/types.h>
40 
41 #include <linux/nbd.h>
42 
43 struct nbd_device {
44 	u32 flags;
45 	struct socket * sock;	/* If == NULL, device is not ready, yet	*/
46 	int magic;
47 
48 	spinlock_t queue_lock;
49 	struct list_head queue_head;	/* Requests waiting result */
50 	struct request *active_req;
51 	wait_queue_head_t active_wq;
52 	struct list_head waiting_queue;	/* Requests to be sent */
53 	wait_queue_head_t waiting_wq;
54 
55 	struct mutex tx_lock;
56 	struct gendisk *disk;
57 	int blksize;
58 	loff_t bytesize;
59 	int xmit_timeout;
60 	bool disconnect; /* a disconnect has been requested by user */
61 
62 	struct timer_list timeout_timer;
63 	spinlock_t tasks_lock;
64 	struct task_struct *task_recv;
65 	struct task_struct *task_send;
66 
67 #if IS_ENABLED(CONFIG_DEBUG_FS)
68 	struct dentry *dbg_dir;
69 #endif
70 };
71 
72 #if IS_ENABLED(CONFIG_DEBUG_FS)
73 static struct dentry *nbd_dbg_dir;
74 #endif
75 
76 #define nbd_name(nbd) ((nbd)->disk->disk_name)
77 
78 #define NBD_MAGIC 0x68797548
79 
80 static unsigned int nbds_max = 16;
81 static struct nbd_device *nbd_dev;
82 static int max_part;
83 
84 /*
85  * Use just one lock (or at most 1 per NIC). Two arguments for this:
86  * 1. Each NIC is essentially a synchronization point for all servers
87  *    accessed through that NIC so there's no need to have more locks
88  *    than NICs anyway.
89  * 2. More locks lead to more "Dirty cache line bouncing" which will slow
90  *    down each lock to the point where they're actually slower than just
91  *    a single lock.
92  * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
93  */
94 static DEFINE_SPINLOCK(nbd_lock);
95 
96 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
97 {
98 	return disk_to_dev(nbd->disk);
99 }
100 
101 static const char *nbdcmd_to_ascii(int cmd)
102 {
103 	switch (cmd) {
104 	case  NBD_CMD_READ: return "read";
105 	case NBD_CMD_WRITE: return "write";
106 	case  NBD_CMD_DISC: return "disconnect";
107 	case NBD_CMD_FLUSH: return "flush";
108 	case  NBD_CMD_TRIM: return "trim/discard";
109 	}
110 	return "invalid";
111 }
112 
113 static void nbd_end_request(struct nbd_device *nbd, struct request *req)
114 {
115 	int error = req->errors ? -EIO : 0;
116 	struct request_queue *q = req->q;
117 	unsigned long flags;
118 
119 	dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", req,
120 		error ? "failed" : "done");
121 
122 	spin_lock_irqsave(q->queue_lock, flags);
123 	__blk_end_request_all(req, error);
124 	spin_unlock_irqrestore(q->queue_lock, flags);
125 }
126 
127 /*
128  * Forcibly shutdown the socket causing all listeners to error
129  */
130 static void sock_shutdown(struct nbd_device *nbd)
131 {
132 	if (!nbd->sock)
133 		return;
134 
135 	dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n");
136 	kernel_sock_shutdown(nbd->sock, SHUT_RDWR);
137 	nbd->sock = NULL;
138 	del_timer_sync(&nbd->timeout_timer);
139 }
140 
141 static void nbd_xmit_timeout(unsigned long arg)
142 {
143 	struct nbd_device *nbd = (struct nbd_device *)arg;
144 	unsigned long flags;
145 
146 	if (list_empty(&nbd->queue_head))
147 		return;
148 
149 	nbd->disconnect = true;
150 
151 	spin_lock_irqsave(&nbd->tasks_lock, flags);
152 
153 	if (nbd->task_recv)
154 		force_sig(SIGKILL, nbd->task_recv);
155 
156 	if (nbd->task_send)
157 		force_sig(SIGKILL, nbd->task_send);
158 
159 	spin_unlock_irqrestore(&nbd->tasks_lock, flags);
160 
161 	dev_err(nbd_to_dev(nbd), "Connection timed out, killed receiver and sender, shutting down connection\n");
162 }
163 
164 /*
165  *  Send or receive packet.
166  */
167 static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size,
168 		int msg_flags)
169 {
170 	struct socket *sock = nbd->sock;
171 	int result;
172 	struct msghdr msg;
173 	struct kvec iov;
174 	sigset_t blocked, oldset;
175 	unsigned long pflags = current->flags;
176 
177 	if (unlikely(!sock)) {
178 		dev_err(disk_to_dev(nbd->disk),
179 			"Attempted %s on closed socket in sock_xmit\n",
180 			(send ? "send" : "recv"));
181 		return -EINVAL;
182 	}
183 
184 	/* Allow interception of SIGKILL only
185 	 * Don't allow other signals to interrupt the transmission */
186 	siginitsetinv(&blocked, sigmask(SIGKILL));
187 	sigprocmask(SIG_SETMASK, &blocked, &oldset);
188 
189 	current->flags |= PF_MEMALLOC;
190 	do {
191 		sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
192 		iov.iov_base = buf;
193 		iov.iov_len = size;
194 		msg.msg_name = NULL;
195 		msg.msg_namelen = 0;
196 		msg.msg_control = NULL;
197 		msg.msg_controllen = 0;
198 		msg.msg_flags = msg_flags | MSG_NOSIGNAL;
199 
200 		if (send)
201 			result = kernel_sendmsg(sock, &msg, &iov, 1, size);
202 		else
203 			result = kernel_recvmsg(sock, &msg, &iov, 1, size,
204 						msg.msg_flags);
205 
206 		if (result <= 0) {
207 			if (result == 0)
208 				result = -EPIPE; /* short read */
209 			break;
210 		}
211 		size -= result;
212 		buf += result;
213 	} while (size > 0);
214 
215 	sigprocmask(SIG_SETMASK, &oldset, NULL);
216 	tsk_restore_flags(current, pflags, PF_MEMALLOC);
217 
218 	if (!send && nbd->xmit_timeout)
219 		mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout);
220 
221 	return result;
222 }
223 
224 static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec,
225 		int flags)
226 {
227 	int result;
228 	void *kaddr = kmap(bvec->bv_page);
229 	result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset,
230 			   bvec->bv_len, flags);
231 	kunmap(bvec->bv_page);
232 	return result;
233 }
234 
235 /* always call with the tx_lock held */
236 static int nbd_send_req(struct nbd_device *nbd, struct request *req)
237 {
238 	int result, flags;
239 	struct nbd_request request;
240 	unsigned long size = blk_rq_bytes(req);
241 	u32 type;
242 
243 	if (req->cmd_type == REQ_TYPE_DRV_PRIV)
244 		type = NBD_CMD_DISC;
245 	else if (req->cmd_flags & REQ_DISCARD)
246 		type = NBD_CMD_TRIM;
247 	else if (req->cmd_flags & REQ_FLUSH)
248 		type = NBD_CMD_FLUSH;
249 	else if (rq_data_dir(req) == WRITE)
250 		type = NBD_CMD_WRITE;
251 	else
252 		type = NBD_CMD_READ;
253 
254 	memset(&request, 0, sizeof(request));
255 	request.magic = htonl(NBD_REQUEST_MAGIC);
256 	request.type = htonl(type);
257 	if (type != NBD_CMD_FLUSH && type != NBD_CMD_DISC) {
258 		request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
259 		request.len = htonl(size);
260 	}
261 	memcpy(request.handle, &req, sizeof(req));
262 
263 	dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
264 		req, nbdcmd_to_ascii(type),
265 		(unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
266 	result = sock_xmit(nbd, 1, &request, sizeof(request),
267 			(type == NBD_CMD_WRITE) ? MSG_MORE : 0);
268 	if (result <= 0) {
269 		dev_err(disk_to_dev(nbd->disk),
270 			"Send control failed (result %d)\n", result);
271 		return -EIO;
272 	}
273 
274 	if (type == NBD_CMD_WRITE) {
275 		struct req_iterator iter;
276 		struct bio_vec bvec;
277 		/*
278 		 * we are really probing at internals to determine
279 		 * whether to set MSG_MORE or not...
280 		 */
281 		rq_for_each_segment(bvec, req, iter) {
282 			flags = 0;
283 			if (!rq_iter_last(bvec, iter))
284 				flags = MSG_MORE;
285 			dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
286 				req, bvec.bv_len);
287 			result = sock_send_bvec(nbd, &bvec, flags);
288 			if (result <= 0) {
289 				dev_err(disk_to_dev(nbd->disk),
290 					"Send data failed (result %d)\n",
291 					result);
292 				return -EIO;
293 			}
294 		}
295 	}
296 	return 0;
297 }
298 
299 static struct request *nbd_find_request(struct nbd_device *nbd,
300 					struct request *xreq)
301 {
302 	struct request *req, *tmp;
303 	int err;
304 
305 	err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq);
306 	if (unlikely(err))
307 		return ERR_PTR(err);
308 
309 	spin_lock(&nbd->queue_lock);
310 	list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) {
311 		if (req != xreq)
312 			continue;
313 		list_del_init(&req->queuelist);
314 		spin_unlock(&nbd->queue_lock);
315 		return req;
316 	}
317 	spin_unlock(&nbd->queue_lock);
318 
319 	return ERR_PTR(-ENOENT);
320 }
321 
322 static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec)
323 {
324 	int result;
325 	void *kaddr = kmap(bvec->bv_page);
326 	result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len,
327 			MSG_WAITALL);
328 	kunmap(bvec->bv_page);
329 	return result;
330 }
331 
332 /* NULL returned = something went wrong, inform userspace */
333 static struct request *nbd_read_stat(struct nbd_device *nbd)
334 {
335 	int result;
336 	struct nbd_reply reply;
337 	struct request *req;
338 
339 	reply.magic = 0;
340 	result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL);
341 	if (result <= 0) {
342 		dev_err(disk_to_dev(nbd->disk),
343 			"Receive control failed (result %d)\n", result);
344 		return ERR_PTR(result);
345 	}
346 
347 	if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
348 		dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
349 				(unsigned long)ntohl(reply.magic));
350 		return ERR_PTR(-EPROTO);
351 	}
352 
353 	req = nbd_find_request(nbd, *(struct request **)reply.handle);
354 	if (IS_ERR(req)) {
355 		result = PTR_ERR(req);
356 		if (result != -ENOENT)
357 			return ERR_PTR(result);
358 
359 		dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n",
360 			reply.handle);
361 		return ERR_PTR(-EBADR);
362 	}
363 
364 	if (ntohl(reply.error)) {
365 		dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
366 			ntohl(reply.error));
367 		req->errors++;
368 		return req;
369 	}
370 
371 	dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
372 	if (rq_data_dir(req) != WRITE) {
373 		struct req_iterator iter;
374 		struct bio_vec bvec;
375 
376 		rq_for_each_segment(bvec, req, iter) {
377 			result = sock_recv_bvec(nbd, &bvec);
378 			if (result <= 0) {
379 				dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
380 					result);
381 				req->errors++;
382 				return req;
383 			}
384 			dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
385 				req, bvec.bv_len);
386 		}
387 	}
388 	return req;
389 }
390 
391 static ssize_t pid_show(struct device *dev,
392 			struct device_attribute *attr, char *buf)
393 {
394 	struct gendisk *disk = dev_to_disk(dev);
395 	struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
396 
397 	return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
398 }
399 
400 static struct device_attribute pid_attr = {
401 	.attr = { .name = "pid", .mode = S_IRUGO},
402 	.show = pid_show,
403 };
404 
405 static int nbd_thread_recv(struct nbd_device *nbd)
406 {
407 	struct request *req;
408 	int ret;
409 	unsigned long flags;
410 
411 	BUG_ON(nbd->magic != NBD_MAGIC);
412 
413 	sk_set_memalloc(nbd->sock->sk);
414 
415 	spin_lock_irqsave(&nbd->tasks_lock, flags);
416 	nbd->task_recv = current;
417 	spin_unlock_irqrestore(&nbd->tasks_lock, flags);
418 
419 	ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
420 	if (ret) {
421 		dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
422 
423 		spin_lock_irqsave(&nbd->tasks_lock, flags);
424 		nbd->task_recv = NULL;
425 		spin_unlock_irqrestore(&nbd->tasks_lock, flags);
426 
427 		return ret;
428 	}
429 
430 	while (1) {
431 		req = nbd_read_stat(nbd);
432 		if (IS_ERR(req)) {
433 			ret = PTR_ERR(req);
434 			break;
435 		}
436 
437 		nbd_end_request(nbd, req);
438 	}
439 
440 	device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
441 
442 	spin_lock_irqsave(&nbd->tasks_lock, flags);
443 	nbd->task_recv = NULL;
444 	spin_unlock_irqrestore(&nbd->tasks_lock, flags);
445 
446 	if (signal_pending(current)) {
447 		ret = kernel_dequeue_signal(NULL);
448 		dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n",
449 			 task_pid_nr(current), current->comm, ret);
450 		mutex_lock(&nbd->tx_lock);
451 		sock_shutdown(nbd);
452 		mutex_unlock(&nbd->tx_lock);
453 		ret = -ETIMEDOUT;
454 	}
455 
456 	return ret;
457 }
458 
459 static void nbd_clear_que(struct nbd_device *nbd)
460 {
461 	struct request *req;
462 
463 	BUG_ON(nbd->magic != NBD_MAGIC);
464 
465 	/*
466 	 * Because we have set nbd->sock to NULL under the tx_lock, all
467 	 * modifications to the list must have completed by now.  For
468 	 * the same reason, the active_req must be NULL.
469 	 *
470 	 * As a consequence, we don't need to take the spin lock while
471 	 * purging the list here.
472 	 */
473 	BUG_ON(nbd->sock);
474 	BUG_ON(nbd->active_req);
475 
476 	while (!list_empty(&nbd->queue_head)) {
477 		req = list_entry(nbd->queue_head.next, struct request,
478 				 queuelist);
479 		list_del_init(&req->queuelist);
480 		req->errors++;
481 		nbd_end_request(nbd, req);
482 	}
483 
484 	while (!list_empty(&nbd->waiting_queue)) {
485 		req = list_entry(nbd->waiting_queue.next, struct request,
486 				 queuelist);
487 		list_del_init(&req->queuelist);
488 		req->errors++;
489 		nbd_end_request(nbd, req);
490 	}
491 	dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
492 }
493 
494 
495 static void nbd_handle_req(struct nbd_device *nbd, struct request *req)
496 {
497 	if (req->cmd_type != REQ_TYPE_FS)
498 		goto error_out;
499 
500 	if (rq_data_dir(req) == WRITE &&
501 	    (nbd->flags & NBD_FLAG_READ_ONLY)) {
502 		dev_err(disk_to_dev(nbd->disk),
503 			"Write on read-only\n");
504 		goto error_out;
505 	}
506 
507 	req->errors = 0;
508 
509 	mutex_lock(&nbd->tx_lock);
510 	if (unlikely(!nbd->sock)) {
511 		mutex_unlock(&nbd->tx_lock);
512 		dev_err(disk_to_dev(nbd->disk),
513 			"Attempted send on closed socket\n");
514 		goto error_out;
515 	}
516 
517 	nbd->active_req = req;
518 
519 	if (nbd->xmit_timeout && list_empty_careful(&nbd->queue_head))
520 		mod_timer(&nbd->timeout_timer, jiffies + nbd->xmit_timeout);
521 
522 	if (nbd_send_req(nbd, req) != 0) {
523 		dev_err(disk_to_dev(nbd->disk), "Request send failed\n");
524 		req->errors++;
525 		nbd_end_request(nbd, req);
526 	} else {
527 		spin_lock(&nbd->queue_lock);
528 		list_add_tail(&req->queuelist, &nbd->queue_head);
529 		spin_unlock(&nbd->queue_lock);
530 	}
531 
532 	nbd->active_req = NULL;
533 	mutex_unlock(&nbd->tx_lock);
534 	wake_up_all(&nbd->active_wq);
535 
536 	return;
537 
538 error_out:
539 	req->errors++;
540 	nbd_end_request(nbd, req);
541 }
542 
543 static int nbd_thread_send(void *data)
544 {
545 	struct nbd_device *nbd = data;
546 	struct request *req;
547 	unsigned long flags;
548 
549 	spin_lock_irqsave(&nbd->tasks_lock, flags);
550 	nbd->task_send = current;
551 	spin_unlock_irqrestore(&nbd->tasks_lock, flags);
552 
553 	set_user_nice(current, MIN_NICE);
554 	while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) {
555 		/* wait for something to do */
556 		wait_event_interruptible(nbd->waiting_wq,
557 					 kthread_should_stop() ||
558 					 !list_empty(&nbd->waiting_queue));
559 
560 		if (signal_pending(current)) {
561 			int ret = kernel_dequeue_signal(NULL);
562 
563 			dev_warn(nbd_to_dev(nbd), "pid %d, %s, got signal %d\n",
564 				 task_pid_nr(current), current->comm, ret);
565 			mutex_lock(&nbd->tx_lock);
566 			sock_shutdown(nbd);
567 			mutex_unlock(&nbd->tx_lock);
568 			break;
569 		}
570 
571 		/* extract request */
572 		if (list_empty(&nbd->waiting_queue))
573 			continue;
574 
575 		spin_lock_irq(&nbd->queue_lock);
576 		req = list_entry(nbd->waiting_queue.next, struct request,
577 				 queuelist);
578 		list_del_init(&req->queuelist);
579 		spin_unlock_irq(&nbd->queue_lock);
580 
581 		/* handle request */
582 		nbd_handle_req(nbd, req);
583 	}
584 
585 	spin_lock_irqsave(&nbd->tasks_lock, flags);
586 	nbd->task_send = NULL;
587 	spin_unlock_irqrestore(&nbd->tasks_lock, flags);
588 
589 	/* Clear maybe pending signals */
590 	if (signal_pending(current))
591 		kernel_dequeue_signal(NULL);
592 
593 	return 0;
594 }
595 
596 /*
597  * We always wait for result of write, for now. It would be nice to make it optional
598  * in future
599  * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK))
600  *   { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
601  */
602 
603 static void nbd_request_handler(struct request_queue *q)
604 		__releases(q->queue_lock) __acquires(q->queue_lock)
605 {
606 	struct request *req;
607 
608 	while ((req = blk_fetch_request(q)) != NULL) {
609 		struct nbd_device *nbd;
610 
611 		spin_unlock_irq(q->queue_lock);
612 
613 		nbd = req->rq_disk->private_data;
614 
615 		BUG_ON(nbd->magic != NBD_MAGIC);
616 
617 		dev_dbg(nbd_to_dev(nbd), "request %p: dequeued (flags=%x)\n",
618 			req, req->cmd_type);
619 
620 		if (unlikely(!nbd->sock)) {
621 			dev_err(disk_to_dev(nbd->disk),
622 				"Attempted send on closed socket\n");
623 			req->errors++;
624 			nbd_end_request(nbd, req);
625 			spin_lock_irq(q->queue_lock);
626 			continue;
627 		}
628 
629 		spin_lock_irq(&nbd->queue_lock);
630 		list_add_tail(&req->queuelist, &nbd->waiting_queue);
631 		spin_unlock_irq(&nbd->queue_lock);
632 
633 		wake_up(&nbd->waiting_wq);
634 
635 		spin_lock_irq(q->queue_lock);
636 	}
637 }
638 
639 static int nbd_dev_dbg_init(struct nbd_device *nbd);
640 static void nbd_dev_dbg_close(struct nbd_device *nbd);
641 
642 /* Must be called with tx_lock held */
643 
644 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
645 		       unsigned int cmd, unsigned long arg)
646 {
647 	switch (cmd) {
648 	case NBD_DISCONNECT: {
649 		struct request sreq;
650 
651 		dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
652 		if (!nbd->sock)
653 			return -EINVAL;
654 
655 		mutex_unlock(&nbd->tx_lock);
656 		fsync_bdev(bdev);
657 		mutex_lock(&nbd->tx_lock);
658 		blk_rq_init(NULL, &sreq);
659 		sreq.cmd_type = REQ_TYPE_DRV_PRIV;
660 
661 		/* Check again after getting mutex back.  */
662 		if (!nbd->sock)
663 			return -EINVAL;
664 
665 		nbd->disconnect = true;
666 
667 		nbd_send_req(nbd, &sreq);
668 		return 0;
669 	}
670 
671 	case NBD_CLEAR_SOCK: {
672 		struct socket *sock = nbd->sock;
673 		nbd->sock = NULL;
674 		nbd_clear_que(nbd);
675 		BUG_ON(!list_empty(&nbd->queue_head));
676 		BUG_ON(!list_empty(&nbd->waiting_queue));
677 		kill_bdev(bdev);
678 		if (sock)
679 			sockfd_put(sock);
680 		return 0;
681 	}
682 
683 	case NBD_SET_SOCK: {
684 		struct socket *sock;
685 		int err;
686 		if (nbd->sock)
687 			return -EBUSY;
688 		sock = sockfd_lookup(arg, &err);
689 		if (sock) {
690 			nbd->sock = sock;
691 			if (max_part > 0)
692 				bdev->bd_invalidated = 1;
693 			nbd->disconnect = false; /* we're connected now */
694 			return 0;
695 		}
696 		return -EINVAL;
697 	}
698 
699 	case NBD_SET_BLKSIZE:
700 		nbd->blksize = arg;
701 		nbd->bytesize &= ~(nbd->blksize-1);
702 		bdev->bd_inode->i_size = nbd->bytesize;
703 		set_blocksize(bdev, nbd->blksize);
704 		set_capacity(nbd->disk, nbd->bytesize >> 9);
705 		return 0;
706 
707 	case NBD_SET_SIZE:
708 		nbd->bytesize = arg & ~(nbd->blksize-1);
709 		bdev->bd_inode->i_size = nbd->bytesize;
710 		set_blocksize(bdev, nbd->blksize);
711 		set_capacity(nbd->disk, nbd->bytesize >> 9);
712 		return 0;
713 
714 	case NBD_SET_TIMEOUT:
715 		nbd->xmit_timeout = arg * HZ;
716 		if (arg)
717 			mod_timer(&nbd->timeout_timer,
718 				  jiffies + nbd->xmit_timeout);
719 		else
720 			del_timer_sync(&nbd->timeout_timer);
721 
722 		return 0;
723 
724 	case NBD_SET_FLAGS:
725 		nbd->flags = arg;
726 		return 0;
727 
728 	case NBD_SET_SIZE_BLOCKS:
729 		nbd->bytesize = ((u64) arg) * nbd->blksize;
730 		bdev->bd_inode->i_size = nbd->bytesize;
731 		set_blocksize(bdev, nbd->blksize);
732 		set_capacity(nbd->disk, nbd->bytesize >> 9);
733 		return 0;
734 
735 	case NBD_DO_IT: {
736 		struct task_struct *thread;
737 		struct socket *sock;
738 		int error;
739 
740 		if (nbd->task_recv)
741 			return -EBUSY;
742 		if (!nbd->sock)
743 			return -EINVAL;
744 
745 		mutex_unlock(&nbd->tx_lock);
746 
747 		if (nbd->flags & NBD_FLAG_READ_ONLY)
748 			set_device_ro(bdev, true);
749 		if (nbd->flags & NBD_FLAG_SEND_TRIM)
750 			queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
751 				nbd->disk->queue);
752 		if (nbd->flags & NBD_FLAG_SEND_FLUSH)
753 			blk_queue_flush(nbd->disk->queue, REQ_FLUSH);
754 		else
755 			blk_queue_flush(nbd->disk->queue, 0);
756 
757 		thread = kthread_run(nbd_thread_send, nbd, "%s",
758 				     nbd_name(nbd));
759 		if (IS_ERR(thread)) {
760 			mutex_lock(&nbd->tx_lock);
761 			return PTR_ERR(thread);
762 		}
763 
764 		nbd_dev_dbg_init(nbd);
765 		error = nbd_thread_recv(nbd);
766 		nbd_dev_dbg_close(nbd);
767 		kthread_stop(thread);
768 
769 		mutex_lock(&nbd->tx_lock);
770 
771 		sock_shutdown(nbd);
772 		sock = nbd->sock;
773 		nbd->sock = NULL;
774 		nbd_clear_que(nbd);
775 		kill_bdev(bdev);
776 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
777 		set_device_ro(bdev, false);
778 		if (sock)
779 			sockfd_put(sock);
780 		nbd->flags = 0;
781 		nbd->bytesize = 0;
782 		bdev->bd_inode->i_size = 0;
783 		set_capacity(nbd->disk, 0);
784 		if (max_part > 0)
785 			blkdev_reread_part(bdev);
786 		if (nbd->disconnect) /* user requested, ignore socket errors */
787 			return 0;
788 		return error;
789 	}
790 
791 	case NBD_CLEAR_QUE:
792 		/*
793 		 * This is for compatibility only.  The queue is always cleared
794 		 * by NBD_DO_IT or NBD_CLEAR_SOCK.
795 		 */
796 		return 0;
797 
798 	case NBD_PRINT_DEBUG:
799 		dev_info(disk_to_dev(nbd->disk),
800 			"next = %p, prev = %p, head = %p\n",
801 			nbd->queue_head.next, nbd->queue_head.prev,
802 			&nbd->queue_head);
803 		return 0;
804 	}
805 	return -ENOTTY;
806 }
807 
808 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
809 		     unsigned int cmd, unsigned long arg)
810 {
811 	struct nbd_device *nbd = bdev->bd_disk->private_data;
812 	int error;
813 
814 	if (!capable(CAP_SYS_ADMIN))
815 		return -EPERM;
816 
817 	BUG_ON(nbd->magic != NBD_MAGIC);
818 
819 	mutex_lock(&nbd->tx_lock);
820 	error = __nbd_ioctl(bdev, nbd, cmd, arg);
821 	mutex_unlock(&nbd->tx_lock);
822 
823 	return error;
824 }
825 
826 static const struct block_device_operations nbd_fops =
827 {
828 	.owner =	THIS_MODULE,
829 	.ioctl =	nbd_ioctl,
830 	.compat_ioctl =	nbd_ioctl,
831 };
832 
833 #if IS_ENABLED(CONFIG_DEBUG_FS)
834 
835 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
836 {
837 	struct nbd_device *nbd = s->private;
838 
839 	if (nbd->task_recv)
840 		seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
841 	if (nbd->task_send)
842 		seq_printf(s, "send: %d\n", task_pid_nr(nbd->task_send));
843 
844 	return 0;
845 }
846 
847 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
848 {
849 	return single_open(file, nbd_dbg_tasks_show, inode->i_private);
850 }
851 
852 static const struct file_operations nbd_dbg_tasks_ops = {
853 	.open = nbd_dbg_tasks_open,
854 	.read = seq_read,
855 	.llseek = seq_lseek,
856 	.release = single_release,
857 };
858 
859 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
860 {
861 	struct nbd_device *nbd = s->private;
862 	u32 flags = nbd->flags;
863 
864 	seq_printf(s, "Hex: 0x%08x\n\n", flags);
865 
866 	seq_puts(s, "Known flags:\n");
867 
868 	if (flags & NBD_FLAG_HAS_FLAGS)
869 		seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
870 	if (flags & NBD_FLAG_READ_ONLY)
871 		seq_puts(s, "NBD_FLAG_READ_ONLY\n");
872 	if (flags & NBD_FLAG_SEND_FLUSH)
873 		seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
874 	if (flags & NBD_FLAG_SEND_TRIM)
875 		seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
876 
877 	return 0;
878 }
879 
880 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
881 {
882 	return single_open(file, nbd_dbg_flags_show, inode->i_private);
883 }
884 
885 static const struct file_operations nbd_dbg_flags_ops = {
886 	.open = nbd_dbg_flags_open,
887 	.read = seq_read,
888 	.llseek = seq_lseek,
889 	.release = single_release,
890 };
891 
892 static int nbd_dev_dbg_init(struct nbd_device *nbd)
893 {
894 	struct dentry *dir;
895 	struct dentry *f;
896 
897 	dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
898 	if (IS_ERR_OR_NULL(dir)) {
899 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s' (%ld)\n",
900 			nbd_name(nbd), PTR_ERR(dir));
901 		return PTR_ERR(dir);
902 	}
903 	nbd->dbg_dir = dir;
904 
905 	f = debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
906 	if (IS_ERR_OR_NULL(f)) {
907 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'tasks', %ld\n",
908 			PTR_ERR(f));
909 		return PTR_ERR(f);
910 	}
911 
912 	f = debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize);
913 	if (IS_ERR_OR_NULL(f)) {
914 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'size_bytes', %ld\n",
915 			PTR_ERR(f));
916 		return PTR_ERR(f);
917 	}
918 
919 	f = debugfs_create_u32("timeout", 0444, dir, &nbd->xmit_timeout);
920 	if (IS_ERR_OR_NULL(f)) {
921 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'timeout', %ld\n",
922 			PTR_ERR(f));
923 		return PTR_ERR(f);
924 	}
925 
926 	f = debugfs_create_u32("blocksize", 0444, dir, &nbd->blksize);
927 	if (IS_ERR_OR_NULL(f)) {
928 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'blocksize', %ld\n",
929 			PTR_ERR(f));
930 		return PTR_ERR(f);
931 	}
932 
933 	f = debugfs_create_file("flags", 0444, dir, &nbd, &nbd_dbg_flags_ops);
934 	if (IS_ERR_OR_NULL(f)) {
935 		dev_err(nbd_to_dev(nbd), "Failed to create debugfs file 'flags', %ld\n",
936 			PTR_ERR(f));
937 		return PTR_ERR(f);
938 	}
939 
940 	return 0;
941 }
942 
943 static void nbd_dev_dbg_close(struct nbd_device *nbd)
944 {
945 	debugfs_remove_recursive(nbd->dbg_dir);
946 }
947 
948 static int nbd_dbg_init(void)
949 {
950 	struct dentry *dbg_dir;
951 
952 	dbg_dir = debugfs_create_dir("nbd", NULL);
953 	if (IS_ERR(dbg_dir))
954 		return PTR_ERR(dbg_dir);
955 
956 	nbd_dbg_dir = dbg_dir;
957 
958 	return 0;
959 }
960 
961 static void nbd_dbg_close(void)
962 {
963 	debugfs_remove_recursive(nbd_dbg_dir);
964 }
965 
966 #else  /* IS_ENABLED(CONFIG_DEBUG_FS) */
967 
968 static int nbd_dev_dbg_init(struct nbd_device *nbd)
969 {
970 	return 0;
971 }
972 
973 static void nbd_dev_dbg_close(struct nbd_device *nbd)
974 {
975 }
976 
977 static int nbd_dbg_init(void)
978 {
979 	return 0;
980 }
981 
982 static void nbd_dbg_close(void)
983 {
984 }
985 
986 #endif
987 
988 /*
989  * And here should be modules and kernel interface
990  *  (Just smiley confuses emacs :-)
991  */
992 
993 static int __init nbd_init(void)
994 {
995 	int err = -ENOMEM;
996 	int i;
997 	int part_shift;
998 
999 	BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
1000 
1001 	if (max_part < 0) {
1002 		printk(KERN_ERR "nbd: max_part must be >= 0\n");
1003 		return -EINVAL;
1004 	}
1005 
1006 	part_shift = 0;
1007 	if (max_part > 0) {
1008 		part_shift = fls(max_part);
1009 
1010 		/*
1011 		 * Adjust max_part according to part_shift as it is exported
1012 		 * to user space so that user can know the max number of
1013 		 * partition kernel should be able to manage.
1014 		 *
1015 		 * Note that -1 is required because partition 0 is reserved
1016 		 * for the whole disk.
1017 		 */
1018 		max_part = (1UL << part_shift) - 1;
1019 	}
1020 
1021 	if ((1UL << part_shift) > DISK_MAX_PARTS)
1022 		return -EINVAL;
1023 
1024 	if (nbds_max > 1UL << (MINORBITS - part_shift))
1025 		return -EINVAL;
1026 
1027 	nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
1028 	if (!nbd_dev)
1029 		return -ENOMEM;
1030 
1031 	for (i = 0; i < nbds_max; i++) {
1032 		struct gendisk *disk = alloc_disk(1 << part_shift);
1033 		if (!disk)
1034 			goto out;
1035 		nbd_dev[i].disk = disk;
1036 		/*
1037 		 * The new linux 2.5 block layer implementation requires
1038 		 * every gendisk to have its very own request_queue struct.
1039 		 * These structs are big so we dynamically allocate them.
1040 		 */
1041 		disk->queue = blk_init_queue(nbd_request_handler, &nbd_lock);
1042 		if (!disk->queue) {
1043 			put_disk(disk);
1044 			goto out;
1045 		}
1046 		/*
1047 		 * Tell the block layer that we are not a rotational device
1048 		 */
1049 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1050 		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1051 		disk->queue->limits.discard_granularity = 512;
1052 		blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1053 		disk->queue->limits.discard_zeroes_data = 0;
1054 		blk_queue_max_hw_sectors(disk->queue, 65536);
1055 		disk->queue->limits.max_sectors = 256;
1056 	}
1057 
1058 	if (register_blkdev(NBD_MAJOR, "nbd")) {
1059 		err = -EIO;
1060 		goto out;
1061 	}
1062 
1063 	printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
1064 
1065 	nbd_dbg_init();
1066 
1067 	for (i = 0; i < nbds_max; i++) {
1068 		struct gendisk *disk = nbd_dev[i].disk;
1069 		nbd_dev[i].magic = NBD_MAGIC;
1070 		INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
1071 		spin_lock_init(&nbd_dev[i].queue_lock);
1072 		spin_lock_init(&nbd_dev[i].tasks_lock);
1073 		INIT_LIST_HEAD(&nbd_dev[i].queue_head);
1074 		mutex_init(&nbd_dev[i].tx_lock);
1075 		init_timer(&nbd_dev[i].timeout_timer);
1076 		nbd_dev[i].timeout_timer.function = nbd_xmit_timeout;
1077 		nbd_dev[i].timeout_timer.data = (unsigned long)&nbd_dev[i];
1078 		init_waitqueue_head(&nbd_dev[i].active_wq);
1079 		init_waitqueue_head(&nbd_dev[i].waiting_wq);
1080 		nbd_dev[i].blksize = 1024;
1081 		nbd_dev[i].bytesize = 0;
1082 		disk->major = NBD_MAJOR;
1083 		disk->first_minor = i << part_shift;
1084 		disk->fops = &nbd_fops;
1085 		disk->private_data = &nbd_dev[i];
1086 		sprintf(disk->disk_name, "nbd%d", i);
1087 		set_capacity(disk, 0);
1088 		add_disk(disk);
1089 	}
1090 
1091 	return 0;
1092 out:
1093 	while (i--) {
1094 		blk_cleanup_queue(nbd_dev[i].disk->queue);
1095 		put_disk(nbd_dev[i].disk);
1096 	}
1097 	kfree(nbd_dev);
1098 	return err;
1099 }
1100 
1101 static void __exit nbd_cleanup(void)
1102 {
1103 	int i;
1104 
1105 	nbd_dbg_close();
1106 
1107 	for (i = 0; i < nbds_max; i++) {
1108 		struct gendisk *disk = nbd_dev[i].disk;
1109 		nbd_dev[i].magic = 0;
1110 		if (disk) {
1111 			del_gendisk(disk);
1112 			blk_cleanup_queue(disk->queue);
1113 			put_disk(disk);
1114 		}
1115 	}
1116 	unregister_blkdev(NBD_MAJOR, "nbd");
1117 	kfree(nbd_dev);
1118 	printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
1119 }
1120 
1121 module_init(nbd_init);
1122 module_exit(nbd_cleanup);
1123 
1124 MODULE_DESCRIPTION("Network Block Device");
1125 MODULE_LICENSE("GPL");
1126 
1127 module_param(nbds_max, int, 0444);
1128 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
1129 module_param(max_part, int, 0444);
1130 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
1131