Loading...
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
36#include <asm/uaccess.h>
37#include <asm/system.h>
38#include <asm/types.h>
39
40#include <linux/nbd.h>
41
42#define LO_MAGIC 0x68797548
43
44#ifdef NDEBUG
45#define dprintk(flags, fmt...)
46#else /* NDEBUG */
47#define dprintk(flags, fmt...) do { \
48 if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
49} while (0)
50#define DBG_IOCTL 0x0004
51#define DBG_INIT 0x0010
52#define DBG_EXIT 0x0020
53#define DBG_BLKDEV 0x0100
54#define DBG_RX 0x0200
55#define DBG_TX 0x0400
56static unsigned int debugflags;
57#endif /* NDEBUG */
58
59static unsigned int nbds_max = 16;
60static struct nbd_device *nbd_dev;
61static int max_part;
62
63/*
64 * Use just one lock (or at most 1 per NIC). Two arguments for this:
65 * 1. Each NIC is essentially a synchronization point for all servers
66 * accessed through that NIC so there's no need to have more locks
67 * than NICs anyway.
68 * 2. More locks lead to more "Dirty cache line bouncing" which will slow
69 * down each lock to the point where they're actually slower than just
70 * a single lock.
71 * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
72 */
73static DEFINE_SPINLOCK(nbd_lock);
74
75#ifndef NDEBUG
76static const char *ioctl_cmd_to_ascii(int cmd)
77{
78 switch (cmd) {
79 case NBD_SET_SOCK: return "set-sock";
80 case NBD_SET_BLKSIZE: return "set-blksize";
81 case NBD_SET_SIZE: return "set-size";
82 case NBD_DO_IT: return "do-it";
83 case NBD_CLEAR_SOCK: return "clear-sock";
84 case NBD_CLEAR_QUE: return "clear-que";
85 case NBD_PRINT_DEBUG: return "print-debug";
86 case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
87 case NBD_DISCONNECT: return "disconnect";
88 case BLKROSET: return "set-read-only";
89 case BLKFLSBUF: return "flush-buffer-cache";
90 }
91 return "unknown";
92}
93
94static const char *nbdcmd_to_ascii(int cmd)
95{
96 switch (cmd) {
97 case NBD_CMD_READ: return "read";
98 case NBD_CMD_WRITE: return "write";
99 case NBD_CMD_DISC: return "disconnect";
100 }
101 return "invalid";
102}
103#endif /* NDEBUG */
104
105static void nbd_end_request(struct request *req)
106{
107 int error = req->errors ? -EIO : 0;
108 struct request_queue *q = req->q;
109 unsigned long flags;
110
111 dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
112 req, error ? "failed" : "done");
113
114 spin_lock_irqsave(q->queue_lock, flags);
115 __blk_end_request_all(req, error);
116 spin_unlock_irqrestore(q->queue_lock, flags);
117}
118
119static void sock_shutdown(struct nbd_device *lo, int lock)
120{
121 /* Forcibly shutdown the socket causing all listeners
122 * to error
123 *
124 * FIXME: This code is duplicated from sys_shutdown, but
125 * there should be a more generic interface rather than
126 * calling socket ops directly here */
127 if (lock)
128 mutex_lock(&lo->tx_lock);
129 if (lo->sock) {
130 printk(KERN_WARNING "%s: shutting down socket\n",
131 lo->disk->disk_name);
132 kernel_sock_shutdown(lo->sock, SHUT_RDWR);
133 lo->sock = NULL;
134 }
135 if (lock)
136 mutex_unlock(&lo->tx_lock);
137}
138
139static void nbd_xmit_timeout(unsigned long arg)
140{
141 struct task_struct *task = (struct task_struct *)arg;
142
143 printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
144 task->comm, task->pid);
145 force_sig(SIGKILL, task);
146}
147
148/*
149 * Send or receive packet.
150 */
151static int sock_xmit(struct nbd_device *lo, int send, void *buf, int size,
152 int msg_flags)
153{
154 struct socket *sock = lo->sock;
155 int result;
156 struct msghdr msg;
157 struct kvec iov;
158 sigset_t blocked, oldset;
159
160 if (unlikely(!sock)) {
161 printk(KERN_ERR "%s: Attempted %s on closed socket in sock_xmit\n",
162 lo->disk->disk_name, (send ? "send" : "recv"));
163 return -EINVAL;
164 }
165
166 /* Allow interception of SIGKILL only
167 * Don't allow other signals to interrupt the transmission */
168 siginitsetinv(&blocked, sigmask(SIGKILL));
169 sigprocmask(SIG_SETMASK, &blocked, &oldset);
170
171 do {
172 sock->sk->sk_allocation = GFP_NOIO;
173 iov.iov_base = buf;
174 iov.iov_len = size;
175 msg.msg_name = NULL;
176 msg.msg_namelen = 0;
177 msg.msg_control = NULL;
178 msg.msg_controllen = 0;
179 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
180
181 if (send) {
182 struct timer_list ti;
183
184 if (lo->xmit_timeout) {
185 init_timer(&ti);
186 ti.function = nbd_xmit_timeout;
187 ti.data = (unsigned long)current;
188 ti.expires = jiffies + lo->xmit_timeout;
189 add_timer(&ti);
190 }
191 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
192 if (lo->xmit_timeout)
193 del_timer_sync(&ti);
194 } else
195 result = kernel_recvmsg(sock, &msg, &iov, 1, size,
196 msg.msg_flags);
197
198 if (signal_pending(current)) {
199 siginfo_t info;
200 printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
201 task_pid_nr(current), current->comm,
202 dequeue_signal_lock(current, ¤t->blocked, &info));
203 result = -EINTR;
204 sock_shutdown(lo, !send);
205 break;
206 }
207
208 if (result <= 0) {
209 if (result == 0)
210 result = -EPIPE; /* short read */
211 break;
212 }
213 size -= result;
214 buf += result;
215 } while (size > 0);
216
217 sigprocmask(SIG_SETMASK, &oldset, NULL);
218
219 return result;
220}
221
222static inline int sock_send_bvec(struct nbd_device *lo, struct bio_vec *bvec,
223 int flags)
224{
225 int result;
226 void *kaddr = kmap(bvec->bv_page);
227 result = sock_xmit(lo, 1, kaddr + bvec->bv_offset, bvec->bv_len, flags);
228 kunmap(bvec->bv_page);
229 return result;
230}
231
232/* always call with the tx_lock held */
233static int nbd_send_req(struct nbd_device *lo, struct request *req)
234{
235 int result, flags;
236 struct nbd_request request;
237 unsigned long size = blk_rq_bytes(req);
238
239 request.magic = htonl(NBD_REQUEST_MAGIC);
240 request.type = htonl(nbd_cmd(req));
241 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
242 request.len = htonl(size);
243 memcpy(request.handle, &req, sizeof(req));
244
245 dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n",
246 lo->disk->disk_name, req,
247 nbdcmd_to_ascii(nbd_cmd(req)),
248 (unsigned long long)blk_rq_pos(req) << 9,
249 blk_rq_bytes(req));
250 result = sock_xmit(lo, 1, &request, sizeof(request),
251 (nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
252 if (result <= 0) {
253 printk(KERN_ERR "%s: Send control failed (result %d)\n",
254 lo->disk->disk_name, result);
255 goto error_out;
256 }
257
258 if (nbd_cmd(req) == NBD_CMD_WRITE) {
259 struct req_iterator iter;
260 struct bio_vec *bvec;
261 /*
262 * we are really probing at internals to determine
263 * whether to set MSG_MORE or not...
264 */
265 rq_for_each_segment(bvec, req, iter) {
266 flags = 0;
267 if (!rq_iter_last(req, iter))
268 flags = MSG_MORE;
269 dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
270 lo->disk->disk_name, req, bvec->bv_len);
271 result = sock_send_bvec(lo, bvec, flags);
272 if (result <= 0) {
273 printk(KERN_ERR "%s: Send data failed (result %d)\n",
274 lo->disk->disk_name, result);
275 goto error_out;
276 }
277 }
278 }
279 return 0;
280
281error_out:
282 return -EIO;
283}
284
285static struct request *nbd_find_request(struct nbd_device *lo,
286 struct request *xreq)
287{
288 struct request *req, *tmp;
289 int err;
290
291 err = wait_event_interruptible(lo->active_wq, lo->active_req != xreq);
292 if (unlikely(err))
293 goto out;
294
295 spin_lock(&lo->queue_lock);
296 list_for_each_entry_safe(req, tmp, &lo->queue_head, queuelist) {
297 if (req != xreq)
298 continue;
299 list_del_init(&req->queuelist);
300 spin_unlock(&lo->queue_lock);
301 return req;
302 }
303 spin_unlock(&lo->queue_lock);
304
305 err = -ENOENT;
306
307out:
308 return ERR_PTR(err);
309}
310
311static inline int sock_recv_bvec(struct nbd_device *lo, struct bio_vec *bvec)
312{
313 int result;
314 void *kaddr = kmap(bvec->bv_page);
315 result = sock_xmit(lo, 0, kaddr + bvec->bv_offset, bvec->bv_len,
316 MSG_WAITALL);
317 kunmap(bvec->bv_page);
318 return result;
319}
320
321/* NULL returned = something went wrong, inform userspace */
322static struct request *nbd_read_stat(struct nbd_device *lo)
323{
324 int result;
325 struct nbd_reply reply;
326 struct request *req;
327
328 reply.magic = 0;
329 result = sock_xmit(lo, 0, &reply, sizeof(reply), MSG_WAITALL);
330 if (result <= 0) {
331 printk(KERN_ERR "%s: Receive control failed (result %d)\n",
332 lo->disk->disk_name, result);
333 goto harderror;
334 }
335
336 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
337 printk(KERN_ERR "%s: Wrong magic (0x%lx)\n",
338 lo->disk->disk_name,
339 (unsigned long)ntohl(reply.magic));
340 result = -EPROTO;
341 goto harderror;
342 }
343
344 req = nbd_find_request(lo, *(struct request **)reply.handle);
345 if (IS_ERR(req)) {
346 result = PTR_ERR(req);
347 if (result != -ENOENT)
348 goto harderror;
349
350 printk(KERN_ERR "%s: Unexpected reply (%p)\n",
351 lo->disk->disk_name, reply.handle);
352 result = -EBADR;
353 goto harderror;
354 }
355
356 if (ntohl(reply.error)) {
357 printk(KERN_ERR "%s: Other side returned error (%d)\n",
358 lo->disk->disk_name, ntohl(reply.error));
359 req->errors++;
360 return req;
361 }
362
363 dprintk(DBG_RX, "%s: request %p: got reply\n",
364 lo->disk->disk_name, req);
365 if (nbd_cmd(req) == NBD_CMD_READ) {
366 struct req_iterator iter;
367 struct bio_vec *bvec;
368
369 rq_for_each_segment(bvec, req, iter) {
370 result = sock_recv_bvec(lo, bvec);
371 if (result <= 0) {
372 printk(KERN_ERR "%s: Receive data failed (result %d)\n",
373 lo->disk->disk_name, result);
374 req->errors++;
375 return req;
376 }
377 dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
378 lo->disk->disk_name, req, bvec->bv_len);
379 }
380 }
381 return req;
382harderror:
383 lo->harderror = result;
384 return NULL;
385}
386
387static ssize_t pid_show(struct device *dev,
388 struct device_attribute *attr, char *buf)
389{
390 struct gendisk *disk = dev_to_disk(dev);
391
392 return sprintf(buf, "%ld\n",
393 (long) ((struct nbd_device *)disk->private_data)->pid);
394}
395
396static struct device_attribute pid_attr = {
397 .attr = { .name = "pid", .mode = S_IRUGO},
398 .show = pid_show,
399};
400
401static int nbd_do_it(struct nbd_device *lo)
402{
403 struct request *req;
404 int ret;
405
406 BUG_ON(lo->magic != LO_MAGIC);
407
408 lo->pid = current->pid;
409 ret = sysfs_create_file(&disk_to_dev(lo->disk)->kobj, &pid_attr.attr);
410 if (ret) {
411 printk(KERN_ERR "nbd: sysfs_create_file failed!");
412 lo->pid = 0;
413 return ret;
414 }
415
416 while ((req = nbd_read_stat(lo)) != NULL)
417 nbd_end_request(req);
418
419 sysfs_remove_file(&disk_to_dev(lo->disk)->kobj, &pid_attr.attr);
420 lo->pid = 0;
421 return 0;
422}
423
424static void nbd_clear_que(struct nbd_device *lo)
425{
426 struct request *req;
427
428 BUG_ON(lo->magic != LO_MAGIC);
429
430 /*
431 * Because we have set lo->sock to NULL under the tx_lock, all
432 * modifications to the list must have completed by now. For
433 * the same reason, the active_req must be NULL.
434 *
435 * As a consequence, we don't need to take the spin lock while
436 * purging the list here.
437 */
438 BUG_ON(lo->sock);
439 BUG_ON(lo->active_req);
440
441 while (!list_empty(&lo->queue_head)) {
442 req = list_entry(lo->queue_head.next, struct request,
443 queuelist);
444 list_del_init(&req->queuelist);
445 req->errors++;
446 nbd_end_request(req);
447 }
448}
449
450
451static void nbd_handle_req(struct nbd_device *lo, struct request *req)
452{
453 if (req->cmd_type != REQ_TYPE_FS)
454 goto error_out;
455
456 nbd_cmd(req) = NBD_CMD_READ;
457 if (rq_data_dir(req) == WRITE) {
458 nbd_cmd(req) = NBD_CMD_WRITE;
459 if (lo->flags & NBD_READ_ONLY) {
460 printk(KERN_ERR "%s: Write on read-only\n",
461 lo->disk->disk_name);
462 goto error_out;
463 }
464 }
465
466 req->errors = 0;
467
468 mutex_lock(&lo->tx_lock);
469 if (unlikely(!lo->sock)) {
470 mutex_unlock(&lo->tx_lock);
471 printk(KERN_ERR "%s: Attempted send on closed socket\n",
472 lo->disk->disk_name);
473 goto error_out;
474 }
475
476 lo->active_req = req;
477
478 if (nbd_send_req(lo, req) != 0) {
479 printk(KERN_ERR "%s: Request send failed\n",
480 lo->disk->disk_name);
481 req->errors++;
482 nbd_end_request(req);
483 } else {
484 spin_lock(&lo->queue_lock);
485 list_add(&req->queuelist, &lo->queue_head);
486 spin_unlock(&lo->queue_lock);
487 }
488
489 lo->active_req = NULL;
490 mutex_unlock(&lo->tx_lock);
491 wake_up_all(&lo->active_wq);
492
493 return;
494
495error_out:
496 req->errors++;
497 nbd_end_request(req);
498}
499
500static int nbd_thread(void *data)
501{
502 struct nbd_device *lo = data;
503 struct request *req;
504
505 set_user_nice(current, -20);
506 while (!kthread_should_stop() || !list_empty(&lo->waiting_queue)) {
507 /* wait for something to do */
508 wait_event_interruptible(lo->waiting_wq,
509 kthread_should_stop() ||
510 !list_empty(&lo->waiting_queue));
511
512 /* extract request */
513 if (list_empty(&lo->waiting_queue))
514 continue;
515
516 spin_lock_irq(&lo->queue_lock);
517 req = list_entry(lo->waiting_queue.next, struct request,
518 queuelist);
519 list_del_init(&req->queuelist);
520 spin_unlock_irq(&lo->queue_lock);
521
522 /* handle request */
523 nbd_handle_req(lo, req);
524 }
525 return 0;
526}
527
528/*
529 * We always wait for result of write, for now. It would be nice to make it optional
530 * in future
531 * if ((rq_data_dir(req) == WRITE) && (lo->flags & NBD_WRITE_NOCHK))
532 * { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
533 */
534
535static void do_nbd_request(struct request_queue *q)
536{
537 struct request *req;
538
539 while ((req = blk_fetch_request(q)) != NULL) {
540 struct nbd_device *lo;
541
542 spin_unlock_irq(q->queue_lock);
543
544 dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
545 req->rq_disk->disk_name, req, req->cmd_type);
546
547 lo = req->rq_disk->private_data;
548
549 BUG_ON(lo->magic != LO_MAGIC);
550
551 if (unlikely(!lo->sock)) {
552 printk(KERN_ERR "%s: Attempted send on closed socket\n",
553 lo->disk->disk_name);
554 req->errors++;
555 nbd_end_request(req);
556 spin_lock_irq(q->queue_lock);
557 continue;
558 }
559
560 spin_lock_irq(&lo->queue_lock);
561 list_add_tail(&req->queuelist, &lo->waiting_queue);
562 spin_unlock_irq(&lo->queue_lock);
563
564 wake_up(&lo->waiting_wq);
565
566 spin_lock_irq(q->queue_lock);
567 }
568}
569
570/* Must be called with tx_lock held */
571
572static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *lo,
573 unsigned int cmd, unsigned long arg)
574{
575 switch (cmd) {
576 case NBD_DISCONNECT: {
577 struct request sreq;
578
579 printk(KERN_INFO "%s: NBD_DISCONNECT\n", lo->disk->disk_name);
580
581 blk_rq_init(NULL, &sreq);
582 sreq.cmd_type = REQ_TYPE_SPECIAL;
583 nbd_cmd(&sreq) = NBD_CMD_DISC;
584 if (!lo->sock)
585 return -EINVAL;
586 nbd_send_req(lo, &sreq);
587 return 0;
588 }
589
590 case NBD_CLEAR_SOCK: {
591 struct file *file;
592
593 lo->sock = NULL;
594 file = lo->file;
595 lo->file = NULL;
596 nbd_clear_que(lo);
597 BUG_ON(!list_empty(&lo->queue_head));
598 if (file)
599 fput(file);
600 return 0;
601 }
602
603 case NBD_SET_SOCK: {
604 struct file *file;
605 if (lo->file)
606 return -EBUSY;
607 file = fget(arg);
608 if (file) {
609 struct inode *inode = file->f_path.dentry->d_inode;
610 if (S_ISSOCK(inode->i_mode)) {
611 lo->file = file;
612 lo->sock = SOCKET_I(inode);
613 if (max_part > 0)
614 bdev->bd_invalidated = 1;
615 return 0;
616 } else {
617 fput(file);
618 }
619 }
620 return -EINVAL;
621 }
622
623 case NBD_SET_BLKSIZE:
624 lo->blksize = arg;
625 lo->bytesize &= ~(lo->blksize-1);
626 bdev->bd_inode->i_size = lo->bytesize;
627 set_blocksize(bdev, lo->blksize);
628 set_capacity(lo->disk, lo->bytesize >> 9);
629 return 0;
630
631 case NBD_SET_SIZE:
632 lo->bytesize = arg & ~(lo->blksize-1);
633 bdev->bd_inode->i_size = lo->bytesize;
634 set_blocksize(bdev, lo->blksize);
635 set_capacity(lo->disk, lo->bytesize >> 9);
636 return 0;
637
638 case NBD_SET_TIMEOUT:
639 lo->xmit_timeout = arg * HZ;
640 return 0;
641
642 case NBD_SET_SIZE_BLOCKS:
643 lo->bytesize = ((u64) arg) * lo->blksize;
644 bdev->bd_inode->i_size = lo->bytesize;
645 set_blocksize(bdev, lo->blksize);
646 set_capacity(lo->disk, lo->bytesize >> 9);
647 return 0;
648
649 case NBD_DO_IT: {
650 struct task_struct *thread;
651 struct file *file;
652 int error;
653
654 if (lo->pid)
655 return -EBUSY;
656 if (!lo->file)
657 return -EINVAL;
658
659 mutex_unlock(&lo->tx_lock);
660
661 thread = kthread_create(nbd_thread, lo, lo->disk->disk_name);
662 if (IS_ERR(thread)) {
663 mutex_lock(&lo->tx_lock);
664 return PTR_ERR(thread);
665 }
666 wake_up_process(thread);
667 error = nbd_do_it(lo);
668 kthread_stop(thread);
669
670 mutex_lock(&lo->tx_lock);
671 if (error)
672 return error;
673 sock_shutdown(lo, 0);
674 file = lo->file;
675 lo->file = NULL;
676 nbd_clear_que(lo);
677 printk(KERN_WARNING "%s: queue cleared\n", lo->disk->disk_name);
678 if (file)
679 fput(file);
680 lo->bytesize = 0;
681 bdev->bd_inode->i_size = 0;
682 set_capacity(lo->disk, 0);
683 if (max_part > 0)
684 ioctl_by_bdev(bdev, BLKRRPART, 0);
685 return lo->harderror;
686 }
687
688 case NBD_CLEAR_QUE:
689 /*
690 * This is for compatibility only. The queue is always cleared
691 * by NBD_DO_IT or NBD_CLEAR_SOCK.
692 */
693 BUG_ON(!lo->sock && !list_empty(&lo->queue_head));
694 return 0;
695
696 case NBD_PRINT_DEBUG:
697 printk(KERN_INFO "%s: next = %p, prev = %p, head = %p\n",
698 bdev->bd_disk->disk_name,
699 lo->queue_head.next, lo->queue_head.prev,
700 &lo->queue_head);
701 return 0;
702 }
703 return -ENOTTY;
704}
705
706static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
707 unsigned int cmd, unsigned long arg)
708{
709 struct nbd_device *lo = bdev->bd_disk->private_data;
710 int error;
711
712 if (!capable(CAP_SYS_ADMIN))
713 return -EPERM;
714
715 BUG_ON(lo->magic != LO_MAGIC);
716
717 /* Anyone capable of this syscall can do *real bad* things */
718 dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
719 lo->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
720
721 mutex_lock(&lo->tx_lock);
722 error = __nbd_ioctl(bdev, lo, cmd, arg);
723 mutex_unlock(&lo->tx_lock);
724
725 return error;
726}
727
728static const struct block_device_operations nbd_fops =
729{
730 .owner = THIS_MODULE,
731 .ioctl = nbd_ioctl,
732};
733
734/*
735 * And here should be modules and kernel interface
736 * (Just smiley confuses emacs :-)
737 */
738
739static int __init nbd_init(void)
740{
741 int err = -ENOMEM;
742 int i;
743 int part_shift;
744
745 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
746
747 if (max_part < 0) {
748 printk(KERN_CRIT "nbd: max_part must be >= 0\n");
749 return -EINVAL;
750 }
751
752 nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
753 if (!nbd_dev)
754 return -ENOMEM;
755
756 part_shift = 0;
757 if (max_part > 0) {
758 part_shift = fls(max_part);
759
760 /*
761 * Adjust max_part according to part_shift as it is exported
762 * to user space so that user can know the max number of
763 * partition kernel should be able to manage.
764 *
765 * Note that -1 is required because partition 0 is reserved
766 * for the whole disk.
767 */
768 max_part = (1UL << part_shift) - 1;
769 }
770
771 if ((1UL << part_shift) > DISK_MAX_PARTS)
772 return -EINVAL;
773
774 if (nbds_max > 1UL << (MINORBITS - part_shift))
775 return -EINVAL;
776
777 for (i = 0; i < nbds_max; i++) {
778 struct gendisk *disk = alloc_disk(1 << part_shift);
779 if (!disk)
780 goto out;
781 nbd_dev[i].disk = disk;
782 /*
783 * The new linux 2.5 block layer implementation requires
784 * every gendisk to have its very own request_queue struct.
785 * These structs are big so we dynamically allocate them.
786 */
787 disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
788 if (!disk->queue) {
789 put_disk(disk);
790 goto out;
791 }
792 /*
793 * Tell the block layer that we are not a rotational device
794 */
795 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
796 }
797
798 if (register_blkdev(NBD_MAJOR, "nbd")) {
799 err = -EIO;
800 goto out;
801 }
802
803 printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
804 dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);
805
806 for (i = 0; i < nbds_max; i++) {
807 struct gendisk *disk = nbd_dev[i].disk;
808 nbd_dev[i].file = NULL;
809 nbd_dev[i].magic = LO_MAGIC;
810 nbd_dev[i].flags = 0;
811 INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
812 spin_lock_init(&nbd_dev[i].queue_lock);
813 INIT_LIST_HEAD(&nbd_dev[i].queue_head);
814 mutex_init(&nbd_dev[i].tx_lock);
815 init_waitqueue_head(&nbd_dev[i].active_wq);
816 init_waitqueue_head(&nbd_dev[i].waiting_wq);
817 nbd_dev[i].blksize = 1024;
818 nbd_dev[i].bytesize = 0;
819 disk->major = NBD_MAJOR;
820 disk->first_minor = i << part_shift;
821 disk->fops = &nbd_fops;
822 disk->private_data = &nbd_dev[i];
823 sprintf(disk->disk_name, "nbd%d", i);
824 set_capacity(disk, 0);
825 add_disk(disk);
826 }
827
828 return 0;
829out:
830 while (i--) {
831 blk_cleanup_queue(nbd_dev[i].disk->queue);
832 put_disk(nbd_dev[i].disk);
833 }
834 kfree(nbd_dev);
835 return err;
836}
837
838static void __exit nbd_cleanup(void)
839{
840 int i;
841 for (i = 0; i < nbds_max; i++) {
842 struct gendisk *disk = nbd_dev[i].disk;
843 nbd_dev[i].magic = 0;
844 if (disk) {
845 del_gendisk(disk);
846 blk_cleanup_queue(disk->queue);
847 put_disk(disk);
848 }
849 }
850 unregister_blkdev(NBD_MAJOR, "nbd");
851 kfree(nbd_dev);
852 printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
853}
854
855module_init(nbd_init);
856module_exit(nbd_cleanup);
857
858MODULE_DESCRIPTION("Network Block Device");
859MODULE_LICENSE("GPL");
860
861module_param(nbds_max, int, 0444);
862MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
863module_param(max_part, int, 0444);
864MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
865#ifndef NDEBUG
866module_param(debugflags, int, 0644);
867MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
868#endif
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Network block device - make block devices work over TCP
4 *
5 * Note that you can not swap over this thing, yet. Seems to work but
6 * deadlocks sometimes - you can not swap over TCP in general.
7 *
8 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
9 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
10 *
11 * (part of code stolen from loop.c)
12 */
13
14#include <linux/major.h>
15
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/init.h>
19#include <linux/sched.h>
20#include <linux/sched/mm.h>
21#include <linux/fs.h>
22#include <linux/bio.h>
23#include <linux/stat.h>
24#include <linux/errno.h>
25#include <linux/file.h>
26#include <linux/ioctl.h>
27#include <linux/mutex.h>
28#include <linux/compiler.h>
29#include <linux/completion.h>
30#include <linux/err.h>
31#include <linux/kernel.h>
32#include <linux/slab.h>
33#include <net/sock.h>
34#include <linux/net.h>
35#include <linux/kthread.h>
36#include <linux/types.h>
37#include <linux/debugfs.h>
38#include <linux/blk-mq.h>
39
40#include <linux/uaccess.h>
41#include <asm/types.h>
42
43#include <linux/nbd.h>
44#include <linux/nbd-netlink.h>
45#include <net/genetlink.h>
46
47#define CREATE_TRACE_POINTS
48#include <trace/events/nbd.h>
49
50static DEFINE_IDR(nbd_index_idr);
51static DEFINE_MUTEX(nbd_index_mutex);
52static int nbd_total_devices = 0;
53
54struct nbd_sock {
55 struct socket *sock;
56 struct mutex tx_lock;
57 struct request *pending;
58 int sent;
59 bool dead;
60 int fallback_index;
61 int cookie;
62};
63
64struct recv_thread_args {
65 struct work_struct work;
66 struct nbd_device *nbd;
67 int index;
68};
69
70struct link_dead_args {
71 struct work_struct work;
72 int index;
73};
74
75#define NBD_RT_TIMEDOUT 0
76#define NBD_RT_DISCONNECT_REQUESTED 1
77#define NBD_RT_DISCONNECTED 2
78#define NBD_RT_HAS_PID_FILE 3
79#define NBD_RT_HAS_CONFIG_REF 4
80#define NBD_RT_BOUND 5
81#define NBD_RT_DESTROY_ON_DISCONNECT 6
82#define NBD_RT_DISCONNECT_ON_CLOSE 7
83
84#define NBD_DESTROY_ON_DISCONNECT 0
85#define NBD_DISCONNECT_REQUESTED 1
86
87struct nbd_config {
88 u32 flags;
89 unsigned long runtime_flags;
90 u64 dead_conn_timeout;
91
92 struct nbd_sock **socks;
93 int num_connections;
94 atomic_t live_connections;
95 wait_queue_head_t conn_wait;
96
97 atomic_t recv_threads;
98 wait_queue_head_t recv_wq;
99 loff_t blksize;
100 loff_t bytesize;
101#if IS_ENABLED(CONFIG_DEBUG_FS)
102 struct dentry *dbg_dir;
103#endif
104};
105
106struct nbd_device {
107 struct blk_mq_tag_set tag_set;
108
109 int index;
110 refcount_t config_refs;
111 refcount_t refs;
112 struct nbd_config *config;
113 struct mutex config_lock;
114 struct gendisk *disk;
115 struct workqueue_struct *recv_workq;
116
117 struct list_head list;
118 struct task_struct *task_recv;
119 struct task_struct *task_setup;
120
121 struct completion *destroy_complete;
122 unsigned long flags;
123};
124
125#define NBD_CMD_REQUEUED 1
126
127struct nbd_cmd {
128 struct nbd_device *nbd;
129 struct mutex lock;
130 int index;
131 int cookie;
132 int retries;
133 blk_status_t status;
134 unsigned long flags;
135 u32 cmd_cookie;
136};
137
138#if IS_ENABLED(CONFIG_DEBUG_FS)
139static struct dentry *nbd_dbg_dir;
140#endif
141
142#define nbd_name(nbd) ((nbd)->disk->disk_name)
143
144#define NBD_MAGIC 0x68797548
145
146#define NBD_DEF_BLKSIZE 1024
147
148static unsigned int nbds_max = 16;
149static int max_part = 16;
150static int part_shift;
151
152static int nbd_dev_dbg_init(struct nbd_device *nbd);
153static void nbd_dev_dbg_close(struct nbd_device *nbd);
154static void nbd_config_put(struct nbd_device *nbd);
155static void nbd_connect_reply(struct genl_info *info, int index);
156static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info);
157static void nbd_dead_link_work(struct work_struct *work);
158static void nbd_disconnect_and_put(struct nbd_device *nbd);
159
160static inline struct device *nbd_to_dev(struct nbd_device *nbd)
161{
162 return disk_to_dev(nbd->disk);
163}
164
165static void nbd_requeue_cmd(struct nbd_cmd *cmd)
166{
167 struct request *req = blk_mq_rq_from_pdu(cmd);
168
169 if (!test_and_set_bit(NBD_CMD_REQUEUED, &cmd->flags))
170 blk_mq_requeue_request(req, true);
171}
172
173#define NBD_COOKIE_BITS 32
174
175static u64 nbd_cmd_handle(struct nbd_cmd *cmd)
176{
177 struct request *req = blk_mq_rq_from_pdu(cmd);
178 u32 tag = blk_mq_unique_tag(req);
179 u64 cookie = cmd->cmd_cookie;
180
181 return (cookie << NBD_COOKIE_BITS) | tag;
182}
183
184static u32 nbd_handle_to_tag(u64 handle)
185{
186 return (u32)handle;
187}
188
189static u32 nbd_handle_to_cookie(u64 handle)
190{
191 return (u32)(handle >> NBD_COOKIE_BITS);
192}
193
194static const char *nbdcmd_to_ascii(int cmd)
195{
196 switch (cmd) {
197 case NBD_CMD_READ: return "read";
198 case NBD_CMD_WRITE: return "write";
199 case NBD_CMD_DISC: return "disconnect";
200 case NBD_CMD_FLUSH: return "flush";
201 case NBD_CMD_TRIM: return "trim/discard";
202 }
203 return "invalid";
204}
205
206static ssize_t pid_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
208{
209 struct gendisk *disk = dev_to_disk(dev);
210 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
211
212 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
213}
214
215static const struct device_attribute pid_attr = {
216 .attr = { .name = "pid", .mode = 0444},
217 .show = pid_show,
218};
219
220static void nbd_dev_remove(struct nbd_device *nbd)
221{
222 struct gendisk *disk = nbd->disk;
223 struct request_queue *q;
224
225 if (disk) {
226 q = disk->queue;
227 del_gendisk(disk);
228 blk_cleanup_queue(q);
229 blk_mq_free_tag_set(&nbd->tag_set);
230 disk->private_data = NULL;
231 put_disk(disk);
232 }
233
234 /*
235 * Place this in the last just before the nbd is freed to
236 * make sure that the disk and the related kobject are also
237 * totally removed to avoid duplicate creation of the same
238 * one.
239 */
240 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) && nbd->destroy_complete)
241 complete(nbd->destroy_complete);
242
243 kfree(nbd);
244}
245
246static void nbd_put(struct nbd_device *nbd)
247{
248 if (refcount_dec_and_mutex_lock(&nbd->refs,
249 &nbd_index_mutex)) {
250 idr_remove(&nbd_index_idr, nbd->index);
251 nbd_dev_remove(nbd);
252 mutex_unlock(&nbd_index_mutex);
253 }
254}
255
256static int nbd_disconnected(struct nbd_config *config)
257{
258 return test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags) ||
259 test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
260}
261
262static void nbd_mark_nsock_dead(struct nbd_device *nbd, struct nbd_sock *nsock,
263 int notify)
264{
265 if (!nsock->dead && notify && !nbd_disconnected(nbd->config)) {
266 struct link_dead_args *args;
267 args = kmalloc(sizeof(struct link_dead_args), GFP_NOIO);
268 if (args) {
269 INIT_WORK(&args->work, nbd_dead_link_work);
270 args->index = nbd->index;
271 queue_work(system_wq, &args->work);
272 }
273 }
274 if (!nsock->dead) {
275 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
276 if (atomic_dec_return(&nbd->config->live_connections) == 0) {
277 if (test_and_clear_bit(NBD_RT_DISCONNECT_REQUESTED,
278 &nbd->config->runtime_flags)) {
279 set_bit(NBD_RT_DISCONNECTED,
280 &nbd->config->runtime_flags);
281 dev_info(nbd_to_dev(nbd),
282 "Disconnected due to user request.\n");
283 }
284 }
285 }
286 nsock->dead = true;
287 nsock->pending = NULL;
288 nsock->sent = 0;
289}
290
291static void nbd_size_clear(struct nbd_device *nbd)
292{
293 if (nbd->config->bytesize) {
294 set_capacity(nbd->disk, 0);
295 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
296 }
297}
298
299static void nbd_size_update(struct nbd_device *nbd)
300{
301 struct nbd_config *config = nbd->config;
302 struct block_device *bdev = bdget_disk(nbd->disk, 0);
303
304 if (config->flags & NBD_FLAG_SEND_TRIM) {
305 nbd->disk->queue->limits.discard_granularity = config->blksize;
306 nbd->disk->queue->limits.discard_alignment = config->blksize;
307 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
308 }
309 blk_queue_logical_block_size(nbd->disk->queue, config->blksize);
310 blk_queue_physical_block_size(nbd->disk->queue, config->blksize);
311 set_capacity(nbd->disk, config->bytesize >> 9);
312 if (bdev) {
313 if (bdev->bd_disk) {
314 bd_set_size(bdev, config->bytesize);
315 set_blocksize(bdev, config->blksize);
316 } else
317 bdev->bd_invalidated = 1;
318 bdput(bdev);
319 }
320 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
321}
322
323static void nbd_size_set(struct nbd_device *nbd, loff_t blocksize,
324 loff_t nr_blocks)
325{
326 struct nbd_config *config = nbd->config;
327 config->blksize = blocksize;
328 config->bytesize = blocksize * nr_blocks;
329 if (nbd->task_recv != NULL)
330 nbd_size_update(nbd);
331}
332
333static void nbd_complete_rq(struct request *req)
334{
335 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
336
337 dev_dbg(nbd_to_dev(cmd->nbd), "request %p: %s\n", req,
338 cmd->status ? "failed" : "done");
339
340 blk_mq_end_request(req, cmd->status);
341}
342
343/*
344 * Forcibly shutdown the socket causing all listeners to error
345 */
346static void sock_shutdown(struct nbd_device *nbd)
347{
348 struct nbd_config *config = nbd->config;
349 int i;
350
351 if (config->num_connections == 0)
352 return;
353 if (test_and_set_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
354 return;
355
356 for (i = 0; i < config->num_connections; i++) {
357 struct nbd_sock *nsock = config->socks[i];
358 mutex_lock(&nsock->tx_lock);
359 nbd_mark_nsock_dead(nbd, nsock, 0);
360 mutex_unlock(&nsock->tx_lock);
361 }
362 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
363}
364
365static u32 req_to_nbd_cmd_type(struct request *req)
366{
367 switch (req_op(req)) {
368 case REQ_OP_DISCARD:
369 return NBD_CMD_TRIM;
370 case REQ_OP_FLUSH:
371 return NBD_CMD_FLUSH;
372 case REQ_OP_WRITE:
373 return NBD_CMD_WRITE;
374 case REQ_OP_READ:
375 return NBD_CMD_READ;
376 default:
377 return U32_MAX;
378 }
379}
380
381static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
382 bool reserved)
383{
384 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
385 struct nbd_device *nbd = cmd->nbd;
386 struct nbd_config *config;
387
388 if (!mutex_trylock(&cmd->lock))
389 return BLK_EH_RESET_TIMER;
390
391 if (!refcount_inc_not_zero(&nbd->config_refs)) {
392 cmd->status = BLK_STS_TIMEOUT;
393 mutex_unlock(&cmd->lock);
394 goto done;
395 }
396 config = nbd->config;
397
398 if (config->num_connections > 1 ||
399 (config->num_connections == 1 && nbd->tag_set.timeout)) {
400 dev_err_ratelimited(nbd_to_dev(nbd),
401 "Connection timed out, retrying (%d/%d alive)\n",
402 atomic_read(&config->live_connections),
403 config->num_connections);
404 /*
405 * Hooray we have more connections, requeue this IO, the submit
406 * path will put it on a real connection. Or if only one
407 * connection is configured, the submit path will wait util
408 * a new connection is reconfigured or util dead timeout.
409 */
410 if (config->socks) {
411 if (cmd->index < config->num_connections) {
412 struct nbd_sock *nsock =
413 config->socks[cmd->index];
414 mutex_lock(&nsock->tx_lock);
415 /* We can have multiple outstanding requests, so
416 * we don't want to mark the nsock dead if we've
417 * already reconnected with a new socket, so
418 * only mark it dead if its the same socket we
419 * were sent out on.
420 */
421 if (cmd->cookie == nsock->cookie)
422 nbd_mark_nsock_dead(nbd, nsock, 1);
423 mutex_unlock(&nsock->tx_lock);
424 }
425 mutex_unlock(&cmd->lock);
426 nbd_requeue_cmd(cmd);
427 nbd_config_put(nbd);
428 return BLK_EH_DONE;
429 }
430 }
431
432 if (!nbd->tag_set.timeout) {
433 /*
434 * Userspace sets timeout=0 to disable socket disconnection,
435 * so just warn and reset the timer.
436 */
437 struct nbd_sock *nsock = config->socks[cmd->index];
438 cmd->retries++;
439 dev_info(nbd_to_dev(nbd), "Possible stuck request %p: control (%s@%llu,%uB). Runtime %u seconds\n",
440 req, nbdcmd_to_ascii(req_to_nbd_cmd_type(req)),
441 (unsigned long long)blk_rq_pos(req) << 9,
442 blk_rq_bytes(req), (req->timeout / HZ) * cmd->retries);
443
444 mutex_lock(&nsock->tx_lock);
445 if (cmd->cookie != nsock->cookie) {
446 nbd_requeue_cmd(cmd);
447 mutex_unlock(&nsock->tx_lock);
448 mutex_unlock(&cmd->lock);
449 nbd_config_put(nbd);
450 return BLK_EH_DONE;
451 }
452 mutex_unlock(&nsock->tx_lock);
453 mutex_unlock(&cmd->lock);
454 nbd_config_put(nbd);
455 return BLK_EH_RESET_TIMER;
456 }
457
458 dev_err_ratelimited(nbd_to_dev(nbd), "Connection timed out\n");
459 set_bit(NBD_RT_TIMEDOUT, &config->runtime_flags);
460 cmd->status = BLK_STS_IOERR;
461 mutex_unlock(&cmd->lock);
462 sock_shutdown(nbd);
463 nbd_config_put(nbd);
464done:
465 blk_mq_complete_request(req);
466 return BLK_EH_DONE;
467}
468
469/*
470 * Send or receive packet.
471 */
472static int sock_xmit(struct nbd_device *nbd, int index, int send,
473 struct iov_iter *iter, int msg_flags, int *sent)
474{
475 struct nbd_config *config = nbd->config;
476 struct socket *sock = config->socks[index]->sock;
477 int result;
478 struct msghdr msg;
479 unsigned int noreclaim_flag;
480
481 if (unlikely(!sock)) {
482 dev_err_ratelimited(disk_to_dev(nbd->disk),
483 "Attempted %s on closed socket in sock_xmit\n",
484 (send ? "send" : "recv"));
485 return -EINVAL;
486 }
487
488 msg.msg_iter = *iter;
489
490 noreclaim_flag = memalloc_noreclaim_save();
491 do {
492 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
493 msg.msg_name = NULL;
494 msg.msg_namelen = 0;
495 msg.msg_control = NULL;
496 msg.msg_controllen = 0;
497 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
498
499 if (send)
500 result = sock_sendmsg(sock, &msg);
501 else
502 result = sock_recvmsg(sock, &msg, msg.msg_flags);
503
504 if (result <= 0) {
505 if (result == 0)
506 result = -EPIPE; /* short read */
507 break;
508 }
509 if (sent)
510 *sent += result;
511 } while (msg_data_left(&msg));
512
513 memalloc_noreclaim_restore(noreclaim_flag);
514
515 return result;
516}
517
518/*
519 * Different settings for sk->sk_sndtimeo can result in different return values
520 * if there is a signal pending when we enter sendmsg, because reasons?
521 */
522static inline int was_interrupted(int result)
523{
524 return result == -ERESTARTSYS || result == -EINTR;
525}
526
527/* always call with the tx_lock held */
528static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
529{
530 struct request *req = blk_mq_rq_from_pdu(cmd);
531 struct nbd_config *config = nbd->config;
532 struct nbd_sock *nsock = config->socks[index];
533 int result;
534 struct nbd_request request = {.magic = htonl(NBD_REQUEST_MAGIC)};
535 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
536 struct iov_iter from;
537 unsigned long size = blk_rq_bytes(req);
538 struct bio *bio;
539 u64 handle;
540 u32 type;
541 u32 nbd_cmd_flags = 0;
542 int sent = nsock->sent, skip = 0;
543
544 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
545
546 type = req_to_nbd_cmd_type(req);
547 if (type == U32_MAX)
548 return -EIO;
549
550 if (rq_data_dir(req) == WRITE &&
551 (config->flags & NBD_FLAG_READ_ONLY)) {
552 dev_err_ratelimited(disk_to_dev(nbd->disk),
553 "Write on read-only\n");
554 return -EIO;
555 }
556
557 if (req->cmd_flags & REQ_FUA)
558 nbd_cmd_flags |= NBD_CMD_FLAG_FUA;
559
560 /* We did a partial send previously, and we at least sent the whole
561 * request struct, so just go and send the rest of the pages in the
562 * request.
563 */
564 if (sent) {
565 if (sent >= sizeof(request)) {
566 skip = sent - sizeof(request);
567
568 /* initialize handle for tracing purposes */
569 handle = nbd_cmd_handle(cmd);
570
571 goto send_pages;
572 }
573 iov_iter_advance(&from, sent);
574 } else {
575 cmd->cmd_cookie++;
576 }
577 cmd->index = index;
578 cmd->cookie = nsock->cookie;
579 cmd->retries = 0;
580 request.type = htonl(type | nbd_cmd_flags);
581 if (type != NBD_CMD_FLUSH) {
582 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
583 request.len = htonl(size);
584 }
585 handle = nbd_cmd_handle(cmd);
586 memcpy(request.handle, &handle, sizeof(handle));
587
588 trace_nbd_send_request(&request, nbd->index, blk_mq_rq_from_pdu(cmd));
589
590 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
591 req, nbdcmd_to_ascii(type),
592 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
593 result = sock_xmit(nbd, index, 1, &from,
594 (type == NBD_CMD_WRITE) ? MSG_MORE : 0, &sent);
595 trace_nbd_header_sent(req, handle);
596 if (result <= 0) {
597 if (was_interrupted(result)) {
598 /* If we havne't sent anything we can just return BUSY,
599 * however if we have sent something we need to make
600 * sure we only allow this req to be sent until we are
601 * completely done.
602 */
603 if (sent) {
604 nsock->pending = req;
605 nsock->sent = sent;
606 }
607 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
608 return BLK_STS_RESOURCE;
609 }
610 dev_err_ratelimited(disk_to_dev(nbd->disk),
611 "Send control failed (result %d)\n", result);
612 return -EAGAIN;
613 }
614send_pages:
615 if (type != NBD_CMD_WRITE)
616 goto out;
617
618 bio = req->bio;
619 while (bio) {
620 struct bio *next = bio->bi_next;
621 struct bvec_iter iter;
622 struct bio_vec bvec;
623
624 bio_for_each_segment(bvec, bio, iter) {
625 bool is_last = !next && bio_iter_last(bvec, iter);
626 int flags = is_last ? 0 : MSG_MORE;
627
628 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
629 req, bvec.bv_len);
630 iov_iter_bvec(&from, WRITE, &bvec, 1, bvec.bv_len);
631 if (skip) {
632 if (skip >= iov_iter_count(&from)) {
633 skip -= iov_iter_count(&from);
634 continue;
635 }
636 iov_iter_advance(&from, skip);
637 skip = 0;
638 }
639 result = sock_xmit(nbd, index, 1, &from, flags, &sent);
640 if (result <= 0) {
641 if (was_interrupted(result)) {
642 /* We've already sent the header, we
643 * have no choice but to set pending and
644 * return BUSY.
645 */
646 nsock->pending = req;
647 nsock->sent = sent;
648 set_bit(NBD_CMD_REQUEUED, &cmd->flags);
649 return BLK_STS_RESOURCE;
650 }
651 dev_err(disk_to_dev(nbd->disk),
652 "Send data failed (result %d)\n",
653 result);
654 return -EAGAIN;
655 }
656 /*
657 * The completion might already have come in,
658 * so break for the last one instead of letting
659 * the iterator do it. This prevents use-after-free
660 * of the bio.
661 */
662 if (is_last)
663 break;
664 }
665 bio = next;
666 }
667out:
668 trace_nbd_payload_sent(req, handle);
669 nsock->pending = NULL;
670 nsock->sent = 0;
671 return 0;
672}
673
674/* NULL returned = something went wrong, inform userspace */
675static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
676{
677 struct nbd_config *config = nbd->config;
678 int result;
679 struct nbd_reply reply;
680 struct nbd_cmd *cmd;
681 struct request *req = NULL;
682 u64 handle;
683 u16 hwq;
684 u32 tag;
685 struct kvec iov = {.iov_base = &reply, .iov_len = sizeof(reply)};
686 struct iov_iter to;
687 int ret = 0;
688
689 reply.magic = 0;
690 iov_iter_kvec(&to, READ, &iov, 1, sizeof(reply));
691 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
692 if (result <= 0) {
693 if (!nbd_disconnected(config))
694 dev_err(disk_to_dev(nbd->disk),
695 "Receive control failed (result %d)\n", result);
696 return ERR_PTR(result);
697 }
698
699 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
700 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
701 (unsigned long)ntohl(reply.magic));
702 return ERR_PTR(-EPROTO);
703 }
704
705 memcpy(&handle, reply.handle, sizeof(handle));
706 tag = nbd_handle_to_tag(handle);
707 hwq = blk_mq_unique_tag_to_hwq(tag);
708 if (hwq < nbd->tag_set.nr_hw_queues)
709 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
710 blk_mq_unique_tag_to_tag(tag));
711 if (!req || !blk_mq_request_started(req)) {
712 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
713 tag, req);
714 return ERR_PTR(-ENOENT);
715 }
716 trace_nbd_header_received(req, handle);
717 cmd = blk_mq_rq_to_pdu(req);
718
719 mutex_lock(&cmd->lock);
720 if (cmd->cmd_cookie != nbd_handle_to_cookie(handle)) {
721 dev_err(disk_to_dev(nbd->disk), "Double reply on req %p, cmd_cookie %u, handle cookie %u\n",
722 req, cmd->cmd_cookie, nbd_handle_to_cookie(handle));
723 ret = -ENOENT;
724 goto out;
725 }
726 if (cmd->status != BLK_STS_OK) {
727 dev_err(disk_to_dev(nbd->disk), "Command already handled %p\n",
728 req);
729 ret = -ENOENT;
730 goto out;
731 }
732 if (test_bit(NBD_CMD_REQUEUED, &cmd->flags)) {
733 dev_err(disk_to_dev(nbd->disk), "Raced with timeout on req %p\n",
734 req);
735 ret = -ENOENT;
736 goto out;
737 }
738 if (ntohl(reply.error)) {
739 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
740 ntohl(reply.error));
741 cmd->status = BLK_STS_IOERR;
742 goto out;
743 }
744
745 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", req);
746 if (rq_data_dir(req) != WRITE) {
747 struct req_iterator iter;
748 struct bio_vec bvec;
749
750 rq_for_each_segment(bvec, req, iter) {
751 iov_iter_bvec(&to, READ, &bvec, 1, bvec.bv_len);
752 result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
753 if (result <= 0) {
754 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
755 result);
756 /*
757 * If we've disconnected, we need to make sure we
758 * complete this request, otherwise error out
759 * and let the timeout stuff handle resubmitting
760 * this request onto another connection.
761 */
762 if (nbd_disconnected(config)) {
763 cmd->status = BLK_STS_IOERR;
764 goto out;
765 }
766 ret = -EIO;
767 goto out;
768 }
769 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
770 req, bvec.bv_len);
771 }
772 }
773out:
774 trace_nbd_payload_received(req, handle);
775 mutex_unlock(&cmd->lock);
776 return ret ? ERR_PTR(ret) : cmd;
777}
778
779static void recv_work(struct work_struct *work)
780{
781 struct recv_thread_args *args = container_of(work,
782 struct recv_thread_args,
783 work);
784 struct nbd_device *nbd = args->nbd;
785 struct nbd_config *config = nbd->config;
786 struct nbd_cmd *cmd;
787 struct request *rq;
788
789 while (1) {
790 cmd = nbd_read_stat(nbd, args->index);
791 if (IS_ERR(cmd)) {
792 struct nbd_sock *nsock = config->socks[args->index];
793
794 mutex_lock(&nsock->tx_lock);
795 nbd_mark_nsock_dead(nbd, nsock, 1);
796 mutex_unlock(&nsock->tx_lock);
797 break;
798 }
799
800 rq = blk_mq_rq_from_pdu(cmd);
801 if (likely(!blk_should_fake_timeout(rq->q)))
802 blk_mq_complete_request(rq);
803 }
804 atomic_dec(&config->recv_threads);
805 wake_up(&config->recv_wq);
806 nbd_config_put(nbd);
807 kfree(args);
808}
809
810static bool nbd_clear_req(struct request *req, void *data, bool reserved)
811{
812 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
813
814 mutex_lock(&cmd->lock);
815 cmd->status = BLK_STS_IOERR;
816 mutex_unlock(&cmd->lock);
817
818 blk_mq_complete_request(req);
819 return true;
820}
821
822static void nbd_clear_que(struct nbd_device *nbd)
823{
824 blk_mq_quiesce_queue(nbd->disk->queue);
825 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
826 blk_mq_unquiesce_queue(nbd->disk->queue);
827 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
828}
829
830static int find_fallback(struct nbd_device *nbd, int index)
831{
832 struct nbd_config *config = nbd->config;
833 int new_index = -1;
834 struct nbd_sock *nsock = config->socks[index];
835 int fallback = nsock->fallback_index;
836
837 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
838 return new_index;
839
840 if (config->num_connections <= 1) {
841 dev_err_ratelimited(disk_to_dev(nbd->disk),
842 "Dead connection, failed to find a fallback\n");
843 return new_index;
844 }
845
846 if (fallback >= 0 && fallback < config->num_connections &&
847 !config->socks[fallback]->dead)
848 return fallback;
849
850 if (nsock->fallback_index < 0 ||
851 nsock->fallback_index >= config->num_connections ||
852 config->socks[nsock->fallback_index]->dead) {
853 int i;
854 for (i = 0; i < config->num_connections; i++) {
855 if (i == index)
856 continue;
857 if (!config->socks[i]->dead) {
858 new_index = i;
859 break;
860 }
861 }
862 nsock->fallback_index = new_index;
863 if (new_index < 0) {
864 dev_err_ratelimited(disk_to_dev(nbd->disk),
865 "Dead connection, failed to find a fallback\n");
866 return new_index;
867 }
868 }
869 new_index = nsock->fallback_index;
870 return new_index;
871}
872
873static int wait_for_reconnect(struct nbd_device *nbd)
874{
875 struct nbd_config *config = nbd->config;
876 if (!config->dead_conn_timeout)
877 return 0;
878 if (test_bit(NBD_RT_DISCONNECTED, &config->runtime_flags))
879 return 0;
880 return wait_event_timeout(config->conn_wait,
881 atomic_read(&config->live_connections) > 0,
882 config->dead_conn_timeout) > 0;
883}
884
885static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
886{
887 struct request *req = blk_mq_rq_from_pdu(cmd);
888 struct nbd_device *nbd = cmd->nbd;
889 struct nbd_config *config;
890 struct nbd_sock *nsock;
891 int ret;
892
893 if (!refcount_inc_not_zero(&nbd->config_refs)) {
894 dev_err_ratelimited(disk_to_dev(nbd->disk),
895 "Socks array is empty\n");
896 blk_mq_start_request(req);
897 return -EINVAL;
898 }
899 config = nbd->config;
900
901 if (index >= config->num_connections) {
902 dev_err_ratelimited(disk_to_dev(nbd->disk),
903 "Attempted send on invalid socket\n");
904 nbd_config_put(nbd);
905 blk_mq_start_request(req);
906 return -EINVAL;
907 }
908 cmd->status = BLK_STS_OK;
909again:
910 nsock = config->socks[index];
911 mutex_lock(&nsock->tx_lock);
912 if (nsock->dead) {
913 int old_index = index;
914 index = find_fallback(nbd, index);
915 mutex_unlock(&nsock->tx_lock);
916 if (index < 0) {
917 if (wait_for_reconnect(nbd)) {
918 index = old_index;
919 goto again;
920 }
921 /* All the sockets should already be down at this point,
922 * we just want to make sure that DISCONNECTED is set so
923 * any requests that come in that were queue'ed waiting
924 * for the reconnect timer don't trigger the timer again
925 * and instead just error out.
926 */
927 sock_shutdown(nbd);
928 nbd_config_put(nbd);
929 blk_mq_start_request(req);
930 return -EIO;
931 }
932 goto again;
933 }
934
935 /* Handle the case that we have a pending request that was partially
936 * transmitted that _has_ to be serviced first. We need to call requeue
937 * here so that it gets put _after_ the request that is already on the
938 * dispatch list.
939 */
940 blk_mq_start_request(req);
941 if (unlikely(nsock->pending && nsock->pending != req)) {
942 nbd_requeue_cmd(cmd);
943 ret = 0;
944 goto out;
945 }
946 /*
947 * Some failures are related to the link going down, so anything that
948 * returns EAGAIN can be retried on a different socket.
949 */
950 ret = nbd_send_cmd(nbd, cmd, index);
951 if (ret == -EAGAIN) {
952 dev_err_ratelimited(disk_to_dev(nbd->disk),
953 "Request send failed, requeueing\n");
954 nbd_mark_nsock_dead(nbd, nsock, 1);
955 nbd_requeue_cmd(cmd);
956 ret = 0;
957 }
958out:
959 mutex_unlock(&nsock->tx_lock);
960 nbd_config_put(nbd);
961 return ret;
962}
963
964static blk_status_t nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
965 const struct blk_mq_queue_data *bd)
966{
967 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
968 int ret;
969
970 /*
971 * Since we look at the bio's to send the request over the network we
972 * need to make sure the completion work doesn't mark this request done
973 * before we are done doing our send. This keeps us from dereferencing
974 * freed data if we have particularly fast completions (ie we get the
975 * completion before we exit sock_xmit on the last bvec) or in the case
976 * that the server is misbehaving (or there was an error) before we're
977 * done sending everything over the wire.
978 */
979 mutex_lock(&cmd->lock);
980 clear_bit(NBD_CMD_REQUEUED, &cmd->flags);
981
982 /* We can be called directly from the user space process, which means we
983 * could possibly have signals pending so our sendmsg will fail. In
984 * this case we need to return that we are busy, otherwise error out as
985 * appropriate.
986 */
987 ret = nbd_handle_cmd(cmd, hctx->queue_num);
988 if (ret < 0)
989 ret = BLK_STS_IOERR;
990 else if (!ret)
991 ret = BLK_STS_OK;
992 mutex_unlock(&cmd->lock);
993
994 return ret;
995}
996
997static struct socket *nbd_get_socket(struct nbd_device *nbd, unsigned long fd,
998 int *err)
999{
1000 struct socket *sock;
1001
1002 *err = 0;
1003 sock = sockfd_lookup(fd, err);
1004 if (!sock)
1005 return NULL;
1006
1007 if (sock->ops->shutdown == sock_no_shutdown) {
1008 dev_err(disk_to_dev(nbd->disk), "Unsupported socket: shutdown callout must be supported.\n");
1009 *err = -EINVAL;
1010 sockfd_put(sock);
1011 return NULL;
1012 }
1013
1014 return sock;
1015}
1016
1017static int nbd_add_socket(struct nbd_device *nbd, unsigned long arg,
1018 bool netlink)
1019{
1020 struct nbd_config *config = nbd->config;
1021 struct socket *sock;
1022 struct nbd_sock **socks;
1023 struct nbd_sock *nsock;
1024 int err;
1025
1026 sock = nbd_get_socket(nbd, arg, &err);
1027 if (!sock)
1028 return err;
1029
1030 if (!netlink && !nbd->task_setup &&
1031 !test_bit(NBD_RT_BOUND, &config->runtime_flags))
1032 nbd->task_setup = current;
1033
1034 if (!netlink &&
1035 (nbd->task_setup != current ||
1036 test_bit(NBD_RT_BOUND, &config->runtime_flags))) {
1037 dev_err(disk_to_dev(nbd->disk),
1038 "Device being setup by another task");
1039 err = -EBUSY;
1040 goto put_socket;
1041 }
1042
1043 nsock = kzalloc(sizeof(*nsock), GFP_KERNEL);
1044 if (!nsock) {
1045 err = -ENOMEM;
1046 goto put_socket;
1047 }
1048
1049 socks = krealloc(config->socks, (config->num_connections + 1) *
1050 sizeof(struct nbd_sock *), GFP_KERNEL);
1051 if (!socks) {
1052 kfree(nsock);
1053 err = -ENOMEM;
1054 goto put_socket;
1055 }
1056
1057 config->socks = socks;
1058
1059 nsock->fallback_index = -1;
1060 nsock->dead = false;
1061 mutex_init(&nsock->tx_lock);
1062 nsock->sock = sock;
1063 nsock->pending = NULL;
1064 nsock->sent = 0;
1065 nsock->cookie = 0;
1066 socks[config->num_connections++] = nsock;
1067 atomic_inc(&config->live_connections);
1068
1069 return 0;
1070
1071put_socket:
1072 sockfd_put(sock);
1073 return err;
1074}
1075
1076static int nbd_reconnect_socket(struct nbd_device *nbd, unsigned long arg)
1077{
1078 struct nbd_config *config = nbd->config;
1079 struct socket *sock, *old;
1080 struct recv_thread_args *args;
1081 int i;
1082 int err;
1083
1084 sock = nbd_get_socket(nbd, arg, &err);
1085 if (!sock)
1086 return err;
1087
1088 args = kzalloc(sizeof(*args), GFP_KERNEL);
1089 if (!args) {
1090 sockfd_put(sock);
1091 return -ENOMEM;
1092 }
1093
1094 for (i = 0; i < config->num_connections; i++) {
1095 struct nbd_sock *nsock = config->socks[i];
1096
1097 if (!nsock->dead)
1098 continue;
1099
1100 mutex_lock(&nsock->tx_lock);
1101 if (!nsock->dead) {
1102 mutex_unlock(&nsock->tx_lock);
1103 continue;
1104 }
1105 sk_set_memalloc(sock->sk);
1106 if (nbd->tag_set.timeout)
1107 sock->sk->sk_sndtimeo = nbd->tag_set.timeout;
1108 atomic_inc(&config->recv_threads);
1109 refcount_inc(&nbd->config_refs);
1110 old = nsock->sock;
1111 nsock->fallback_index = -1;
1112 nsock->sock = sock;
1113 nsock->dead = false;
1114 INIT_WORK(&args->work, recv_work);
1115 args->index = i;
1116 args->nbd = nbd;
1117 nsock->cookie++;
1118 mutex_unlock(&nsock->tx_lock);
1119 sockfd_put(old);
1120
1121 clear_bit(NBD_RT_DISCONNECTED, &config->runtime_flags);
1122
1123 /* We take the tx_mutex in an error path in the recv_work, so we
1124 * need to queue_work outside of the tx_mutex.
1125 */
1126 queue_work(nbd->recv_workq, &args->work);
1127
1128 atomic_inc(&config->live_connections);
1129 wake_up(&config->conn_wait);
1130 return 0;
1131 }
1132 sockfd_put(sock);
1133 kfree(args);
1134 return -ENOSPC;
1135}
1136
1137static void nbd_bdev_reset(struct block_device *bdev)
1138{
1139 if (bdev->bd_openers > 1)
1140 return;
1141 bd_set_size(bdev, 0);
1142}
1143
1144static void nbd_parse_flags(struct nbd_device *nbd)
1145{
1146 struct nbd_config *config = nbd->config;
1147 if (config->flags & NBD_FLAG_READ_ONLY)
1148 set_disk_ro(nbd->disk, true);
1149 else
1150 set_disk_ro(nbd->disk, false);
1151 if (config->flags & NBD_FLAG_SEND_TRIM)
1152 blk_queue_flag_set(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1153 if (config->flags & NBD_FLAG_SEND_FLUSH) {
1154 if (config->flags & NBD_FLAG_SEND_FUA)
1155 blk_queue_write_cache(nbd->disk->queue, true, true);
1156 else
1157 blk_queue_write_cache(nbd->disk->queue, true, false);
1158 }
1159 else
1160 blk_queue_write_cache(nbd->disk->queue, false, false);
1161}
1162
1163static void send_disconnects(struct nbd_device *nbd)
1164{
1165 struct nbd_config *config = nbd->config;
1166 struct nbd_request request = {
1167 .magic = htonl(NBD_REQUEST_MAGIC),
1168 .type = htonl(NBD_CMD_DISC),
1169 };
1170 struct kvec iov = {.iov_base = &request, .iov_len = sizeof(request)};
1171 struct iov_iter from;
1172 int i, ret;
1173
1174 for (i = 0; i < config->num_connections; i++) {
1175 struct nbd_sock *nsock = config->socks[i];
1176
1177 iov_iter_kvec(&from, WRITE, &iov, 1, sizeof(request));
1178 mutex_lock(&nsock->tx_lock);
1179 ret = sock_xmit(nbd, i, 1, &from, 0, NULL);
1180 if (ret <= 0)
1181 dev_err(disk_to_dev(nbd->disk),
1182 "Send disconnect failed %d\n", ret);
1183 mutex_unlock(&nsock->tx_lock);
1184 }
1185}
1186
1187static int nbd_disconnect(struct nbd_device *nbd)
1188{
1189 struct nbd_config *config = nbd->config;
1190
1191 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
1192 set_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags);
1193 set_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags);
1194 send_disconnects(nbd);
1195 return 0;
1196}
1197
1198static void nbd_clear_sock(struct nbd_device *nbd)
1199{
1200 sock_shutdown(nbd);
1201 nbd_clear_que(nbd);
1202 nbd->task_setup = NULL;
1203}
1204
1205static void nbd_config_put(struct nbd_device *nbd)
1206{
1207 if (refcount_dec_and_mutex_lock(&nbd->config_refs,
1208 &nbd->config_lock)) {
1209 struct nbd_config *config = nbd->config;
1210 nbd_dev_dbg_close(nbd);
1211 nbd_size_clear(nbd);
1212 if (test_and_clear_bit(NBD_RT_HAS_PID_FILE,
1213 &config->runtime_flags))
1214 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
1215 nbd->task_recv = NULL;
1216 nbd_clear_sock(nbd);
1217 if (config->num_connections) {
1218 int i;
1219 for (i = 0; i < config->num_connections; i++) {
1220 sockfd_put(config->socks[i]->sock);
1221 kfree(config->socks[i]);
1222 }
1223 kfree(config->socks);
1224 }
1225 kfree(nbd->config);
1226 nbd->config = NULL;
1227
1228 if (nbd->recv_workq)
1229 destroy_workqueue(nbd->recv_workq);
1230 nbd->recv_workq = NULL;
1231
1232 nbd->tag_set.timeout = 0;
1233 nbd->disk->queue->limits.discard_granularity = 0;
1234 nbd->disk->queue->limits.discard_alignment = 0;
1235 blk_queue_max_discard_sectors(nbd->disk->queue, UINT_MAX);
1236 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, nbd->disk->queue);
1237
1238 mutex_unlock(&nbd->config_lock);
1239 nbd_put(nbd);
1240 module_put(THIS_MODULE);
1241 }
1242}
1243
1244static int nbd_start_device(struct nbd_device *nbd)
1245{
1246 struct nbd_config *config = nbd->config;
1247 int num_connections = config->num_connections;
1248 int error = 0, i;
1249
1250 if (nbd->task_recv)
1251 return -EBUSY;
1252 if (!config->socks)
1253 return -EINVAL;
1254 if (num_connections > 1 &&
1255 !(config->flags & NBD_FLAG_CAN_MULTI_CONN)) {
1256 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
1257 return -EINVAL;
1258 }
1259
1260 nbd->recv_workq = alloc_workqueue("knbd%d-recv",
1261 WQ_MEM_RECLAIM | WQ_HIGHPRI |
1262 WQ_UNBOUND, 0, nbd->index);
1263 if (!nbd->recv_workq) {
1264 dev_err(disk_to_dev(nbd->disk), "Could not allocate knbd recv work queue.\n");
1265 return -ENOMEM;
1266 }
1267
1268 blk_mq_update_nr_hw_queues(&nbd->tag_set, config->num_connections);
1269 nbd->task_recv = current;
1270
1271 nbd_parse_flags(nbd);
1272
1273 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
1274 if (error) {
1275 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
1276 return error;
1277 }
1278 set_bit(NBD_RT_HAS_PID_FILE, &config->runtime_flags);
1279
1280 nbd_dev_dbg_init(nbd);
1281 for (i = 0; i < num_connections; i++) {
1282 struct recv_thread_args *args;
1283
1284 args = kzalloc(sizeof(*args), GFP_KERNEL);
1285 if (!args) {
1286 sock_shutdown(nbd);
1287 /*
1288 * If num_connections is m (2 < m),
1289 * and NO.1 ~ NO.n(1 < n < m) kzallocs are successful.
1290 * But NO.(n + 1) failed. We still have n recv threads.
1291 * So, add flush_workqueue here to prevent recv threads
1292 * dropping the last config_refs and trying to destroy
1293 * the workqueue from inside the workqueue.
1294 */
1295 if (i)
1296 flush_workqueue(nbd->recv_workq);
1297 return -ENOMEM;
1298 }
1299 sk_set_memalloc(config->socks[i]->sock->sk);
1300 if (nbd->tag_set.timeout)
1301 config->socks[i]->sock->sk->sk_sndtimeo =
1302 nbd->tag_set.timeout;
1303 atomic_inc(&config->recv_threads);
1304 refcount_inc(&nbd->config_refs);
1305 INIT_WORK(&args->work, recv_work);
1306 args->nbd = nbd;
1307 args->index = i;
1308 queue_work(nbd->recv_workq, &args->work);
1309 }
1310 nbd_size_update(nbd);
1311 return error;
1312}
1313
1314static int nbd_start_device_ioctl(struct nbd_device *nbd, struct block_device *bdev)
1315{
1316 struct nbd_config *config = nbd->config;
1317 int ret;
1318
1319 ret = nbd_start_device(nbd);
1320 if (ret)
1321 return ret;
1322
1323 if (max_part)
1324 bdev->bd_invalidated = 1;
1325 mutex_unlock(&nbd->config_lock);
1326 ret = wait_event_interruptible(config->recv_wq,
1327 atomic_read(&config->recv_threads) == 0);
1328 if (ret)
1329 sock_shutdown(nbd);
1330 flush_workqueue(nbd->recv_workq);
1331
1332 mutex_lock(&nbd->config_lock);
1333 nbd_bdev_reset(bdev);
1334 /* user requested, ignore socket errors */
1335 if (test_bit(NBD_RT_DISCONNECT_REQUESTED, &config->runtime_flags))
1336 ret = 0;
1337 if (test_bit(NBD_RT_TIMEDOUT, &config->runtime_flags))
1338 ret = -ETIMEDOUT;
1339 return ret;
1340}
1341
1342static void nbd_clear_sock_ioctl(struct nbd_device *nbd,
1343 struct block_device *bdev)
1344{
1345 sock_shutdown(nbd);
1346 __invalidate_device(bdev, true);
1347 nbd_bdev_reset(bdev);
1348 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
1349 &nbd->config->runtime_flags))
1350 nbd_config_put(nbd);
1351}
1352
1353static bool nbd_is_valid_blksize(unsigned long blksize)
1354{
1355 if (!blksize || !is_power_of_2(blksize) || blksize < 512 ||
1356 blksize > PAGE_SIZE)
1357 return false;
1358 return true;
1359}
1360
1361static void nbd_set_cmd_timeout(struct nbd_device *nbd, u64 timeout)
1362{
1363 nbd->tag_set.timeout = timeout * HZ;
1364 if (timeout)
1365 blk_queue_rq_timeout(nbd->disk->queue, timeout * HZ);
1366 else
1367 blk_queue_rq_timeout(nbd->disk->queue, 30 * HZ);
1368}
1369
1370/* Must be called with config_lock held */
1371static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
1372 unsigned int cmd, unsigned long arg)
1373{
1374 struct nbd_config *config = nbd->config;
1375
1376 switch (cmd) {
1377 case NBD_DISCONNECT:
1378 return nbd_disconnect(nbd);
1379 case NBD_CLEAR_SOCK:
1380 nbd_clear_sock_ioctl(nbd, bdev);
1381 return 0;
1382 case NBD_SET_SOCK:
1383 return nbd_add_socket(nbd, arg, false);
1384 case NBD_SET_BLKSIZE:
1385 if (!arg)
1386 arg = NBD_DEF_BLKSIZE;
1387 if (!nbd_is_valid_blksize(arg))
1388 return -EINVAL;
1389 nbd_size_set(nbd, arg,
1390 div_s64(config->bytesize, arg));
1391 return 0;
1392 case NBD_SET_SIZE:
1393 nbd_size_set(nbd, config->blksize,
1394 div_s64(arg, config->blksize));
1395 return 0;
1396 case NBD_SET_SIZE_BLOCKS:
1397 nbd_size_set(nbd, config->blksize, arg);
1398 return 0;
1399 case NBD_SET_TIMEOUT:
1400 nbd_set_cmd_timeout(nbd, arg);
1401 return 0;
1402
1403 case NBD_SET_FLAGS:
1404 config->flags = arg;
1405 return 0;
1406 case NBD_DO_IT:
1407 return nbd_start_device_ioctl(nbd, bdev);
1408 case NBD_CLEAR_QUE:
1409 /*
1410 * This is for compatibility only. The queue is always cleared
1411 * by NBD_DO_IT or NBD_CLEAR_SOCK.
1412 */
1413 return 0;
1414 case NBD_PRINT_DEBUG:
1415 /*
1416 * For compatibility only, we no longer keep a list of
1417 * outstanding requests.
1418 */
1419 return 0;
1420 }
1421 return -ENOTTY;
1422}
1423
1424static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
1425 unsigned int cmd, unsigned long arg)
1426{
1427 struct nbd_device *nbd = bdev->bd_disk->private_data;
1428 struct nbd_config *config = nbd->config;
1429 int error = -EINVAL;
1430
1431 if (!capable(CAP_SYS_ADMIN))
1432 return -EPERM;
1433
1434 /* The block layer will pass back some non-nbd ioctls in case we have
1435 * special handling for them, but we don't so just return an error.
1436 */
1437 if (_IOC_TYPE(cmd) != 0xab)
1438 return -EINVAL;
1439
1440 mutex_lock(&nbd->config_lock);
1441
1442 /* Don't allow ioctl operations on a nbd device that was created with
1443 * netlink, unless it's DISCONNECT or CLEAR_SOCK, which are fine.
1444 */
1445 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
1446 (cmd == NBD_DISCONNECT || cmd == NBD_CLEAR_SOCK))
1447 error = __nbd_ioctl(bdev, nbd, cmd, arg);
1448 else
1449 dev_err(nbd_to_dev(nbd), "Cannot use ioctl interface on a netlink controlled device.\n");
1450 mutex_unlock(&nbd->config_lock);
1451 return error;
1452}
1453
1454static struct nbd_config *nbd_alloc_config(void)
1455{
1456 struct nbd_config *config;
1457
1458 config = kzalloc(sizeof(struct nbd_config), GFP_NOFS);
1459 if (!config)
1460 return NULL;
1461 atomic_set(&config->recv_threads, 0);
1462 init_waitqueue_head(&config->recv_wq);
1463 init_waitqueue_head(&config->conn_wait);
1464 config->blksize = NBD_DEF_BLKSIZE;
1465 atomic_set(&config->live_connections, 0);
1466 try_module_get(THIS_MODULE);
1467 return config;
1468}
1469
1470static int nbd_open(struct block_device *bdev, fmode_t mode)
1471{
1472 struct nbd_device *nbd;
1473 int ret = 0;
1474
1475 mutex_lock(&nbd_index_mutex);
1476 nbd = bdev->bd_disk->private_data;
1477 if (!nbd) {
1478 ret = -ENXIO;
1479 goto out;
1480 }
1481 if (!refcount_inc_not_zero(&nbd->refs)) {
1482 ret = -ENXIO;
1483 goto out;
1484 }
1485 if (!refcount_inc_not_zero(&nbd->config_refs)) {
1486 struct nbd_config *config;
1487
1488 mutex_lock(&nbd->config_lock);
1489 if (refcount_inc_not_zero(&nbd->config_refs)) {
1490 mutex_unlock(&nbd->config_lock);
1491 goto out;
1492 }
1493 config = nbd->config = nbd_alloc_config();
1494 if (!config) {
1495 ret = -ENOMEM;
1496 mutex_unlock(&nbd->config_lock);
1497 goto out;
1498 }
1499 refcount_set(&nbd->config_refs, 1);
1500 refcount_inc(&nbd->refs);
1501 mutex_unlock(&nbd->config_lock);
1502 bdev->bd_invalidated = 1;
1503 } else if (nbd_disconnected(nbd->config)) {
1504 bdev->bd_invalidated = 1;
1505 }
1506out:
1507 mutex_unlock(&nbd_index_mutex);
1508 return ret;
1509}
1510
1511static void nbd_release(struct gendisk *disk, fmode_t mode)
1512{
1513 struct nbd_device *nbd = disk->private_data;
1514 struct block_device *bdev = bdget_disk(disk, 0);
1515
1516 if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
1517 bdev->bd_openers == 0)
1518 nbd_disconnect_and_put(nbd);
1519
1520 nbd_config_put(nbd);
1521 nbd_put(nbd);
1522}
1523
1524static const struct block_device_operations nbd_fops =
1525{
1526 .owner = THIS_MODULE,
1527 .open = nbd_open,
1528 .release = nbd_release,
1529 .ioctl = nbd_ioctl,
1530 .compat_ioctl = nbd_ioctl,
1531};
1532
1533#if IS_ENABLED(CONFIG_DEBUG_FS)
1534
1535static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
1536{
1537 struct nbd_device *nbd = s->private;
1538
1539 if (nbd->task_recv)
1540 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
1541
1542 return 0;
1543}
1544
1545static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
1546{
1547 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
1548}
1549
1550static const struct file_operations nbd_dbg_tasks_ops = {
1551 .open = nbd_dbg_tasks_open,
1552 .read = seq_read,
1553 .llseek = seq_lseek,
1554 .release = single_release,
1555};
1556
1557static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
1558{
1559 struct nbd_device *nbd = s->private;
1560 u32 flags = nbd->config->flags;
1561
1562 seq_printf(s, "Hex: 0x%08x\n\n", flags);
1563
1564 seq_puts(s, "Known flags:\n");
1565
1566 if (flags & NBD_FLAG_HAS_FLAGS)
1567 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
1568 if (flags & NBD_FLAG_READ_ONLY)
1569 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
1570 if (flags & NBD_FLAG_SEND_FLUSH)
1571 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
1572 if (flags & NBD_FLAG_SEND_FUA)
1573 seq_puts(s, "NBD_FLAG_SEND_FUA\n");
1574 if (flags & NBD_FLAG_SEND_TRIM)
1575 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
1576
1577 return 0;
1578}
1579
1580static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
1581{
1582 return single_open(file, nbd_dbg_flags_show, inode->i_private);
1583}
1584
1585static const struct file_operations nbd_dbg_flags_ops = {
1586 .open = nbd_dbg_flags_open,
1587 .read = seq_read,
1588 .llseek = seq_lseek,
1589 .release = single_release,
1590};
1591
1592static int nbd_dev_dbg_init(struct nbd_device *nbd)
1593{
1594 struct dentry *dir;
1595 struct nbd_config *config = nbd->config;
1596
1597 if (!nbd_dbg_dir)
1598 return -EIO;
1599
1600 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
1601 if (!dir) {
1602 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
1603 nbd_name(nbd));
1604 return -EIO;
1605 }
1606 config->dbg_dir = dir;
1607
1608 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
1609 debugfs_create_u64("size_bytes", 0444, dir, &config->bytesize);
1610 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
1611 debugfs_create_u64("blocksize", 0444, dir, &config->blksize);
1612 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
1613
1614 return 0;
1615}
1616
1617static void nbd_dev_dbg_close(struct nbd_device *nbd)
1618{
1619 debugfs_remove_recursive(nbd->config->dbg_dir);
1620}
1621
1622static int nbd_dbg_init(void)
1623{
1624 struct dentry *dbg_dir;
1625
1626 dbg_dir = debugfs_create_dir("nbd", NULL);
1627 if (!dbg_dir)
1628 return -EIO;
1629
1630 nbd_dbg_dir = dbg_dir;
1631
1632 return 0;
1633}
1634
1635static void nbd_dbg_close(void)
1636{
1637 debugfs_remove_recursive(nbd_dbg_dir);
1638}
1639
1640#else /* IS_ENABLED(CONFIG_DEBUG_FS) */
1641
1642static int nbd_dev_dbg_init(struct nbd_device *nbd)
1643{
1644 return 0;
1645}
1646
1647static void nbd_dev_dbg_close(struct nbd_device *nbd)
1648{
1649}
1650
1651static int nbd_dbg_init(void)
1652{
1653 return 0;
1654}
1655
1656static void nbd_dbg_close(void)
1657{
1658}
1659
1660#endif
1661
1662static int nbd_init_request(struct blk_mq_tag_set *set, struct request *rq,
1663 unsigned int hctx_idx, unsigned int numa_node)
1664{
1665 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
1666 cmd->nbd = set->driver_data;
1667 cmd->flags = 0;
1668 mutex_init(&cmd->lock);
1669 return 0;
1670}
1671
1672static const struct blk_mq_ops nbd_mq_ops = {
1673 .queue_rq = nbd_queue_rq,
1674 .complete = nbd_complete_rq,
1675 .init_request = nbd_init_request,
1676 .timeout = nbd_xmit_timeout,
1677};
1678
1679static int nbd_dev_add(int index)
1680{
1681 struct nbd_device *nbd;
1682 struct gendisk *disk;
1683 struct request_queue *q;
1684 int err = -ENOMEM;
1685
1686 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1687 if (!nbd)
1688 goto out;
1689
1690 disk = alloc_disk(1 << part_shift);
1691 if (!disk)
1692 goto out_free_nbd;
1693
1694 if (index >= 0) {
1695 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1696 GFP_KERNEL);
1697 if (err == -ENOSPC)
1698 err = -EEXIST;
1699 } else {
1700 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1701 if (err >= 0)
1702 index = err;
1703 }
1704 if (err < 0)
1705 goto out_free_disk;
1706
1707 nbd->index = index;
1708 nbd->disk = disk;
1709 nbd->tag_set.ops = &nbd_mq_ops;
1710 nbd->tag_set.nr_hw_queues = 1;
1711 nbd->tag_set.queue_depth = 128;
1712 nbd->tag_set.numa_node = NUMA_NO_NODE;
1713 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1714 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1715 BLK_MQ_F_BLOCKING;
1716 nbd->tag_set.driver_data = nbd;
1717 nbd->destroy_complete = NULL;
1718
1719 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1720 if (err)
1721 goto out_free_idr;
1722
1723 q = blk_mq_init_queue(&nbd->tag_set);
1724 if (IS_ERR(q)) {
1725 err = PTR_ERR(q);
1726 goto out_free_tags;
1727 }
1728 disk->queue = q;
1729
1730 /*
1731 * Tell the block layer that we are not a rotational device
1732 */
1733 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
1734 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1735 disk->queue->limits.discard_granularity = 0;
1736 disk->queue->limits.discard_alignment = 0;
1737 blk_queue_max_discard_sectors(disk->queue, 0);
1738 blk_queue_max_segment_size(disk->queue, UINT_MAX);
1739 blk_queue_max_segments(disk->queue, USHRT_MAX);
1740 blk_queue_max_hw_sectors(disk->queue, 65536);
1741 disk->queue->limits.max_sectors = 256;
1742
1743 mutex_init(&nbd->config_lock);
1744 refcount_set(&nbd->config_refs, 0);
1745 refcount_set(&nbd->refs, 1);
1746 INIT_LIST_HEAD(&nbd->list);
1747 disk->major = NBD_MAJOR;
1748 disk->first_minor = index << part_shift;
1749 disk->fops = &nbd_fops;
1750 disk->private_data = nbd;
1751 sprintf(disk->disk_name, "nbd%d", index);
1752 add_disk(disk);
1753 nbd_total_devices++;
1754 return index;
1755
1756out_free_tags:
1757 blk_mq_free_tag_set(&nbd->tag_set);
1758out_free_idr:
1759 idr_remove(&nbd_index_idr, index);
1760out_free_disk:
1761 put_disk(disk);
1762out_free_nbd:
1763 kfree(nbd);
1764out:
1765 return err;
1766}
1767
1768static int find_free_cb(int id, void *ptr, void *data)
1769{
1770 struct nbd_device *nbd = ptr;
1771 struct nbd_device **found = data;
1772
1773 if (!refcount_read(&nbd->config_refs)) {
1774 *found = nbd;
1775 return 1;
1776 }
1777 return 0;
1778}
1779
1780/* Netlink interface. */
1781static const struct nla_policy nbd_attr_policy[NBD_ATTR_MAX + 1] = {
1782 [NBD_ATTR_INDEX] = { .type = NLA_U32 },
1783 [NBD_ATTR_SIZE_BYTES] = { .type = NLA_U64 },
1784 [NBD_ATTR_BLOCK_SIZE_BYTES] = { .type = NLA_U64 },
1785 [NBD_ATTR_TIMEOUT] = { .type = NLA_U64 },
1786 [NBD_ATTR_SERVER_FLAGS] = { .type = NLA_U64 },
1787 [NBD_ATTR_CLIENT_FLAGS] = { .type = NLA_U64 },
1788 [NBD_ATTR_SOCKETS] = { .type = NLA_NESTED},
1789 [NBD_ATTR_DEAD_CONN_TIMEOUT] = { .type = NLA_U64 },
1790 [NBD_ATTR_DEVICE_LIST] = { .type = NLA_NESTED},
1791};
1792
1793static const struct nla_policy nbd_sock_policy[NBD_SOCK_MAX + 1] = {
1794 [NBD_SOCK_FD] = { .type = NLA_U32 },
1795};
1796
1797/* We don't use this right now since we don't parse the incoming list, but we
1798 * still want it here so userspace knows what to expect.
1799 */
1800static const struct nla_policy __attribute__((unused))
1801nbd_device_policy[NBD_DEVICE_ATTR_MAX + 1] = {
1802 [NBD_DEVICE_INDEX] = { .type = NLA_U32 },
1803 [NBD_DEVICE_CONNECTED] = { .type = NLA_U8 },
1804};
1805
1806static int nbd_genl_size_set(struct genl_info *info, struct nbd_device *nbd)
1807{
1808 struct nbd_config *config = nbd->config;
1809 u64 bsize = config->blksize;
1810 u64 bytes = config->bytesize;
1811
1812 if (info->attrs[NBD_ATTR_SIZE_BYTES])
1813 bytes = nla_get_u64(info->attrs[NBD_ATTR_SIZE_BYTES]);
1814
1815 if (info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]) {
1816 bsize = nla_get_u64(info->attrs[NBD_ATTR_BLOCK_SIZE_BYTES]);
1817 if (!bsize)
1818 bsize = NBD_DEF_BLKSIZE;
1819 if (!nbd_is_valid_blksize(bsize)) {
1820 printk(KERN_ERR "Invalid block size %llu\n", bsize);
1821 return -EINVAL;
1822 }
1823 }
1824
1825 if (bytes != config->bytesize || bsize != config->blksize)
1826 nbd_size_set(nbd, bsize, div64_u64(bytes, bsize));
1827 return 0;
1828}
1829
1830static int nbd_genl_connect(struct sk_buff *skb, struct genl_info *info)
1831{
1832 DECLARE_COMPLETION_ONSTACK(destroy_complete);
1833 struct nbd_device *nbd = NULL;
1834 struct nbd_config *config;
1835 int index = -1;
1836 int ret;
1837 bool put_dev = false;
1838
1839 if (!netlink_capable(skb, CAP_SYS_ADMIN))
1840 return -EPERM;
1841
1842 if (info->attrs[NBD_ATTR_INDEX])
1843 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
1844 if (!info->attrs[NBD_ATTR_SOCKETS]) {
1845 printk(KERN_ERR "nbd: must specify at least one socket\n");
1846 return -EINVAL;
1847 }
1848 if (!info->attrs[NBD_ATTR_SIZE_BYTES]) {
1849 printk(KERN_ERR "nbd: must specify a size in bytes for the device\n");
1850 return -EINVAL;
1851 }
1852again:
1853 mutex_lock(&nbd_index_mutex);
1854 if (index == -1) {
1855 ret = idr_for_each(&nbd_index_idr, &find_free_cb, &nbd);
1856 if (ret == 0) {
1857 int new_index;
1858 new_index = nbd_dev_add(-1);
1859 if (new_index < 0) {
1860 mutex_unlock(&nbd_index_mutex);
1861 printk(KERN_ERR "nbd: failed to add new device\n");
1862 return new_index;
1863 }
1864 nbd = idr_find(&nbd_index_idr, new_index);
1865 }
1866 } else {
1867 nbd = idr_find(&nbd_index_idr, index);
1868 if (!nbd) {
1869 ret = nbd_dev_add(index);
1870 if (ret < 0) {
1871 mutex_unlock(&nbd_index_mutex);
1872 printk(KERN_ERR "nbd: failed to add new device\n");
1873 return ret;
1874 }
1875 nbd = idr_find(&nbd_index_idr, index);
1876 }
1877 }
1878 if (!nbd) {
1879 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
1880 index);
1881 mutex_unlock(&nbd_index_mutex);
1882 return -EINVAL;
1883 }
1884
1885 if (test_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags) &&
1886 test_bit(NBD_DISCONNECT_REQUESTED, &nbd->flags)) {
1887 nbd->destroy_complete = &destroy_complete;
1888 mutex_unlock(&nbd_index_mutex);
1889
1890 /* Wait untill the the nbd stuff is totally destroyed */
1891 wait_for_completion(&destroy_complete);
1892 goto again;
1893 }
1894
1895 if (!refcount_inc_not_zero(&nbd->refs)) {
1896 mutex_unlock(&nbd_index_mutex);
1897 if (index == -1)
1898 goto again;
1899 printk(KERN_ERR "nbd: device at index %d is going down\n",
1900 index);
1901 return -EINVAL;
1902 }
1903 mutex_unlock(&nbd_index_mutex);
1904
1905 mutex_lock(&nbd->config_lock);
1906 if (refcount_read(&nbd->config_refs)) {
1907 mutex_unlock(&nbd->config_lock);
1908 nbd_put(nbd);
1909 if (index == -1)
1910 goto again;
1911 printk(KERN_ERR "nbd: nbd%d already in use\n", index);
1912 return -EBUSY;
1913 }
1914 if (WARN_ON(nbd->config)) {
1915 mutex_unlock(&nbd->config_lock);
1916 nbd_put(nbd);
1917 return -EINVAL;
1918 }
1919 config = nbd->config = nbd_alloc_config();
1920 if (!nbd->config) {
1921 mutex_unlock(&nbd->config_lock);
1922 nbd_put(nbd);
1923 printk(KERN_ERR "nbd: couldn't allocate config\n");
1924 return -ENOMEM;
1925 }
1926 refcount_set(&nbd->config_refs, 1);
1927 set_bit(NBD_RT_BOUND, &config->runtime_flags);
1928
1929 ret = nbd_genl_size_set(info, nbd);
1930 if (ret)
1931 goto out;
1932
1933 if (info->attrs[NBD_ATTR_TIMEOUT])
1934 nbd_set_cmd_timeout(nbd,
1935 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
1936 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
1937 config->dead_conn_timeout =
1938 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
1939 config->dead_conn_timeout *= HZ;
1940 }
1941 if (info->attrs[NBD_ATTR_SERVER_FLAGS])
1942 config->flags =
1943 nla_get_u64(info->attrs[NBD_ATTR_SERVER_FLAGS]);
1944 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
1945 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
1946 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
1947 set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
1948 &config->runtime_flags);
1949 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1950 put_dev = true;
1951 } else {
1952 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
1953 }
1954 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
1955 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
1956 &config->runtime_flags);
1957 }
1958 }
1959
1960 if (info->attrs[NBD_ATTR_SOCKETS]) {
1961 struct nlattr *attr;
1962 int rem, fd;
1963
1964 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
1965 rem) {
1966 struct nlattr *socks[NBD_SOCK_MAX+1];
1967
1968 if (nla_type(attr) != NBD_SOCK_ITEM) {
1969 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
1970 ret = -EINVAL;
1971 goto out;
1972 }
1973 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
1974 attr,
1975 nbd_sock_policy,
1976 info->extack);
1977 if (ret != 0) {
1978 printk(KERN_ERR "nbd: error processing sock list\n");
1979 ret = -EINVAL;
1980 goto out;
1981 }
1982 if (!socks[NBD_SOCK_FD])
1983 continue;
1984 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
1985 ret = nbd_add_socket(nbd, fd, true);
1986 if (ret)
1987 goto out;
1988 }
1989 }
1990 ret = nbd_start_device(nbd);
1991out:
1992 mutex_unlock(&nbd->config_lock);
1993 if (!ret) {
1994 set_bit(NBD_RT_HAS_CONFIG_REF, &config->runtime_flags);
1995 refcount_inc(&nbd->config_refs);
1996 nbd_connect_reply(info, nbd->index);
1997 }
1998 nbd_config_put(nbd);
1999 if (put_dev)
2000 nbd_put(nbd);
2001 return ret;
2002}
2003
2004static void nbd_disconnect_and_put(struct nbd_device *nbd)
2005{
2006 mutex_lock(&nbd->config_lock);
2007 nbd_disconnect(nbd);
2008 nbd_clear_sock(nbd);
2009 mutex_unlock(&nbd->config_lock);
2010 /*
2011 * Make sure recv thread has finished, so it does not drop the last
2012 * config ref and try to destroy the workqueue from inside the work
2013 * queue.
2014 */
2015 flush_workqueue(nbd->recv_workq);
2016 if (test_and_clear_bit(NBD_RT_HAS_CONFIG_REF,
2017 &nbd->config->runtime_flags))
2018 nbd_config_put(nbd);
2019}
2020
2021static int nbd_genl_disconnect(struct sk_buff *skb, struct genl_info *info)
2022{
2023 struct nbd_device *nbd;
2024 int index;
2025
2026 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2027 return -EPERM;
2028
2029 if (!info->attrs[NBD_ATTR_INDEX]) {
2030 printk(KERN_ERR "nbd: must specify an index to disconnect\n");
2031 return -EINVAL;
2032 }
2033 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2034 mutex_lock(&nbd_index_mutex);
2035 nbd = idr_find(&nbd_index_idr, index);
2036 if (!nbd) {
2037 mutex_unlock(&nbd_index_mutex);
2038 printk(KERN_ERR "nbd: couldn't find device at index %d\n",
2039 index);
2040 return -EINVAL;
2041 }
2042 if (!refcount_inc_not_zero(&nbd->refs)) {
2043 mutex_unlock(&nbd_index_mutex);
2044 printk(KERN_ERR "nbd: device at index %d is going down\n",
2045 index);
2046 return -EINVAL;
2047 }
2048 mutex_unlock(&nbd_index_mutex);
2049 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2050 nbd_put(nbd);
2051 return 0;
2052 }
2053 nbd_disconnect_and_put(nbd);
2054 nbd_config_put(nbd);
2055 nbd_put(nbd);
2056 return 0;
2057}
2058
2059static int nbd_genl_reconfigure(struct sk_buff *skb, struct genl_info *info)
2060{
2061 struct nbd_device *nbd = NULL;
2062 struct nbd_config *config;
2063 int index;
2064 int ret = 0;
2065 bool put_dev = false;
2066
2067 if (!netlink_capable(skb, CAP_SYS_ADMIN))
2068 return -EPERM;
2069
2070 if (!info->attrs[NBD_ATTR_INDEX]) {
2071 printk(KERN_ERR "nbd: must specify a device to reconfigure\n");
2072 return -EINVAL;
2073 }
2074 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2075 mutex_lock(&nbd_index_mutex);
2076 nbd = idr_find(&nbd_index_idr, index);
2077 if (!nbd) {
2078 mutex_unlock(&nbd_index_mutex);
2079 printk(KERN_ERR "nbd: couldn't find a device at index %d\n",
2080 index);
2081 return -EINVAL;
2082 }
2083 if (!refcount_inc_not_zero(&nbd->refs)) {
2084 mutex_unlock(&nbd_index_mutex);
2085 printk(KERN_ERR "nbd: device at index %d is going down\n",
2086 index);
2087 return -EINVAL;
2088 }
2089 mutex_unlock(&nbd_index_mutex);
2090
2091 if (!refcount_inc_not_zero(&nbd->config_refs)) {
2092 dev_err(nbd_to_dev(nbd),
2093 "not configured, cannot reconfigure\n");
2094 nbd_put(nbd);
2095 return -EINVAL;
2096 }
2097
2098 mutex_lock(&nbd->config_lock);
2099 config = nbd->config;
2100 if (!test_bit(NBD_RT_BOUND, &config->runtime_flags) ||
2101 !nbd->task_recv) {
2102 dev_err(nbd_to_dev(nbd),
2103 "not configured, cannot reconfigure\n");
2104 ret = -EINVAL;
2105 goto out;
2106 }
2107
2108 ret = nbd_genl_size_set(info, nbd);
2109 if (ret)
2110 goto out;
2111
2112 if (info->attrs[NBD_ATTR_TIMEOUT])
2113 nbd_set_cmd_timeout(nbd,
2114 nla_get_u64(info->attrs[NBD_ATTR_TIMEOUT]));
2115 if (info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]) {
2116 config->dead_conn_timeout =
2117 nla_get_u64(info->attrs[NBD_ATTR_DEAD_CONN_TIMEOUT]);
2118 config->dead_conn_timeout *= HZ;
2119 }
2120 if (info->attrs[NBD_ATTR_CLIENT_FLAGS]) {
2121 u64 flags = nla_get_u64(info->attrs[NBD_ATTR_CLIENT_FLAGS]);
2122 if (flags & NBD_CFLAG_DESTROY_ON_DISCONNECT) {
2123 if (!test_and_set_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2124 &config->runtime_flags))
2125 put_dev = true;
2126 set_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2127 } else {
2128 if (test_and_clear_bit(NBD_RT_DESTROY_ON_DISCONNECT,
2129 &config->runtime_flags))
2130 refcount_inc(&nbd->refs);
2131 clear_bit(NBD_DESTROY_ON_DISCONNECT, &nbd->flags);
2132 }
2133
2134 if (flags & NBD_CFLAG_DISCONNECT_ON_CLOSE) {
2135 set_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2136 &config->runtime_flags);
2137 } else {
2138 clear_bit(NBD_RT_DISCONNECT_ON_CLOSE,
2139 &config->runtime_flags);
2140 }
2141 }
2142
2143 if (info->attrs[NBD_ATTR_SOCKETS]) {
2144 struct nlattr *attr;
2145 int rem, fd;
2146
2147 nla_for_each_nested(attr, info->attrs[NBD_ATTR_SOCKETS],
2148 rem) {
2149 struct nlattr *socks[NBD_SOCK_MAX+1];
2150
2151 if (nla_type(attr) != NBD_SOCK_ITEM) {
2152 printk(KERN_ERR "nbd: socks must be embedded in a SOCK_ITEM attr\n");
2153 ret = -EINVAL;
2154 goto out;
2155 }
2156 ret = nla_parse_nested_deprecated(socks, NBD_SOCK_MAX,
2157 attr,
2158 nbd_sock_policy,
2159 info->extack);
2160 if (ret != 0) {
2161 printk(KERN_ERR "nbd: error processing sock list\n");
2162 ret = -EINVAL;
2163 goto out;
2164 }
2165 if (!socks[NBD_SOCK_FD])
2166 continue;
2167 fd = (int)nla_get_u32(socks[NBD_SOCK_FD]);
2168 ret = nbd_reconnect_socket(nbd, fd);
2169 if (ret) {
2170 if (ret == -ENOSPC)
2171 ret = 0;
2172 goto out;
2173 }
2174 dev_info(nbd_to_dev(nbd), "reconnected socket\n");
2175 }
2176 }
2177out:
2178 mutex_unlock(&nbd->config_lock);
2179 nbd_config_put(nbd);
2180 nbd_put(nbd);
2181 if (put_dev)
2182 nbd_put(nbd);
2183 return ret;
2184}
2185
2186static const struct genl_ops nbd_connect_genl_ops[] = {
2187 {
2188 .cmd = NBD_CMD_CONNECT,
2189 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2190 .doit = nbd_genl_connect,
2191 },
2192 {
2193 .cmd = NBD_CMD_DISCONNECT,
2194 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2195 .doit = nbd_genl_disconnect,
2196 },
2197 {
2198 .cmd = NBD_CMD_RECONFIGURE,
2199 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2200 .doit = nbd_genl_reconfigure,
2201 },
2202 {
2203 .cmd = NBD_CMD_STATUS,
2204 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2205 .doit = nbd_genl_status,
2206 },
2207};
2208
2209static const struct genl_multicast_group nbd_mcast_grps[] = {
2210 { .name = NBD_GENL_MCAST_GROUP_NAME, },
2211};
2212
2213static struct genl_family nbd_genl_family __ro_after_init = {
2214 .hdrsize = 0,
2215 .name = NBD_GENL_FAMILY_NAME,
2216 .version = NBD_GENL_VERSION,
2217 .module = THIS_MODULE,
2218 .ops = nbd_connect_genl_ops,
2219 .n_ops = ARRAY_SIZE(nbd_connect_genl_ops),
2220 .maxattr = NBD_ATTR_MAX,
2221 .policy = nbd_attr_policy,
2222 .mcgrps = nbd_mcast_grps,
2223 .n_mcgrps = ARRAY_SIZE(nbd_mcast_grps),
2224};
2225
2226static int populate_nbd_status(struct nbd_device *nbd, struct sk_buff *reply)
2227{
2228 struct nlattr *dev_opt;
2229 u8 connected = 0;
2230 int ret;
2231
2232 /* This is a little racey, but for status it's ok. The
2233 * reason we don't take a ref here is because we can't
2234 * take a ref in the index == -1 case as we would need
2235 * to put under the nbd_index_mutex, which could
2236 * deadlock if we are configured to remove ourselves
2237 * once we're disconnected.
2238 */
2239 if (refcount_read(&nbd->config_refs))
2240 connected = 1;
2241 dev_opt = nla_nest_start_noflag(reply, NBD_DEVICE_ITEM);
2242 if (!dev_opt)
2243 return -EMSGSIZE;
2244 ret = nla_put_u32(reply, NBD_DEVICE_INDEX, nbd->index);
2245 if (ret)
2246 return -EMSGSIZE;
2247 ret = nla_put_u8(reply, NBD_DEVICE_CONNECTED,
2248 connected);
2249 if (ret)
2250 return -EMSGSIZE;
2251 nla_nest_end(reply, dev_opt);
2252 return 0;
2253}
2254
2255static int status_cb(int id, void *ptr, void *data)
2256{
2257 struct nbd_device *nbd = ptr;
2258 return populate_nbd_status(nbd, (struct sk_buff *)data);
2259}
2260
2261static int nbd_genl_status(struct sk_buff *skb, struct genl_info *info)
2262{
2263 struct nlattr *dev_list;
2264 struct sk_buff *reply;
2265 void *reply_head;
2266 size_t msg_size;
2267 int index = -1;
2268 int ret = -ENOMEM;
2269
2270 if (info->attrs[NBD_ATTR_INDEX])
2271 index = nla_get_u32(info->attrs[NBD_ATTR_INDEX]);
2272
2273 mutex_lock(&nbd_index_mutex);
2274
2275 msg_size = nla_total_size(nla_attr_size(sizeof(u32)) +
2276 nla_attr_size(sizeof(u8)));
2277 msg_size *= (index == -1) ? nbd_total_devices : 1;
2278
2279 reply = genlmsg_new(msg_size, GFP_KERNEL);
2280 if (!reply)
2281 goto out;
2282 reply_head = genlmsg_put_reply(reply, info, &nbd_genl_family, 0,
2283 NBD_CMD_STATUS);
2284 if (!reply_head) {
2285 nlmsg_free(reply);
2286 goto out;
2287 }
2288
2289 dev_list = nla_nest_start_noflag(reply, NBD_ATTR_DEVICE_LIST);
2290 if (index == -1) {
2291 ret = idr_for_each(&nbd_index_idr, &status_cb, reply);
2292 if (ret) {
2293 nlmsg_free(reply);
2294 goto out;
2295 }
2296 } else {
2297 struct nbd_device *nbd;
2298 nbd = idr_find(&nbd_index_idr, index);
2299 if (nbd) {
2300 ret = populate_nbd_status(nbd, reply);
2301 if (ret) {
2302 nlmsg_free(reply);
2303 goto out;
2304 }
2305 }
2306 }
2307 nla_nest_end(reply, dev_list);
2308 genlmsg_end(reply, reply_head);
2309 ret = genlmsg_reply(reply, info);
2310out:
2311 mutex_unlock(&nbd_index_mutex);
2312 return ret;
2313}
2314
2315static void nbd_connect_reply(struct genl_info *info, int index)
2316{
2317 struct sk_buff *skb;
2318 void *msg_head;
2319 int ret;
2320
2321 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2322 if (!skb)
2323 return;
2324 msg_head = genlmsg_put_reply(skb, info, &nbd_genl_family, 0,
2325 NBD_CMD_CONNECT);
2326 if (!msg_head) {
2327 nlmsg_free(skb);
2328 return;
2329 }
2330 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2331 if (ret) {
2332 nlmsg_free(skb);
2333 return;
2334 }
2335 genlmsg_end(skb, msg_head);
2336 genlmsg_reply(skb, info);
2337}
2338
2339static void nbd_mcast_index(int index)
2340{
2341 struct sk_buff *skb;
2342 void *msg_head;
2343 int ret;
2344
2345 skb = genlmsg_new(nla_total_size(sizeof(u32)), GFP_KERNEL);
2346 if (!skb)
2347 return;
2348 msg_head = genlmsg_put(skb, 0, 0, &nbd_genl_family, 0,
2349 NBD_CMD_LINK_DEAD);
2350 if (!msg_head) {
2351 nlmsg_free(skb);
2352 return;
2353 }
2354 ret = nla_put_u32(skb, NBD_ATTR_INDEX, index);
2355 if (ret) {
2356 nlmsg_free(skb);
2357 return;
2358 }
2359 genlmsg_end(skb, msg_head);
2360 genlmsg_multicast(&nbd_genl_family, skb, 0, 0, GFP_KERNEL);
2361}
2362
2363static void nbd_dead_link_work(struct work_struct *work)
2364{
2365 struct link_dead_args *args = container_of(work, struct link_dead_args,
2366 work);
2367 nbd_mcast_index(args->index);
2368 kfree(args);
2369}
2370
2371static int __init nbd_init(void)
2372{
2373 int i;
2374
2375 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
2376
2377 if (max_part < 0) {
2378 printk(KERN_ERR "nbd: max_part must be >= 0\n");
2379 return -EINVAL;
2380 }
2381
2382 part_shift = 0;
2383 if (max_part > 0) {
2384 part_shift = fls(max_part);
2385
2386 /*
2387 * Adjust max_part according to part_shift as it is exported
2388 * to user space so that user can know the max number of
2389 * partition kernel should be able to manage.
2390 *
2391 * Note that -1 is required because partition 0 is reserved
2392 * for the whole disk.
2393 */
2394 max_part = (1UL << part_shift) - 1;
2395 }
2396
2397 if ((1UL << part_shift) > DISK_MAX_PARTS)
2398 return -EINVAL;
2399
2400 if (nbds_max > 1UL << (MINORBITS - part_shift))
2401 return -EINVAL;
2402
2403 if (register_blkdev(NBD_MAJOR, "nbd"))
2404 return -EIO;
2405
2406 if (genl_register_family(&nbd_genl_family)) {
2407 unregister_blkdev(NBD_MAJOR, "nbd");
2408 return -EINVAL;
2409 }
2410 nbd_dbg_init();
2411
2412 mutex_lock(&nbd_index_mutex);
2413 for (i = 0; i < nbds_max; i++)
2414 nbd_dev_add(i);
2415 mutex_unlock(&nbd_index_mutex);
2416 return 0;
2417}
2418
2419static int nbd_exit_cb(int id, void *ptr, void *data)
2420{
2421 struct list_head *list = (struct list_head *)data;
2422 struct nbd_device *nbd = ptr;
2423
2424 list_add_tail(&nbd->list, list);
2425 return 0;
2426}
2427
2428static void __exit nbd_cleanup(void)
2429{
2430 struct nbd_device *nbd;
2431 LIST_HEAD(del_list);
2432
2433 nbd_dbg_close();
2434
2435 mutex_lock(&nbd_index_mutex);
2436 idr_for_each(&nbd_index_idr, &nbd_exit_cb, &del_list);
2437 mutex_unlock(&nbd_index_mutex);
2438
2439 while (!list_empty(&del_list)) {
2440 nbd = list_first_entry(&del_list, struct nbd_device, list);
2441 list_del_init(&nbd->list);
2442 if (refcount_read(&nbd->refs) != 1)
2443 printk(KERN_ERR "nbd: possibly leaking a device\n");
2444 nbd_put(nbd);
2445 }
2446
2447 idr_destroy(&nbd_index_idr);
2448 genl_unregister_family(&nbd_genl_family);
2449 unregister_blkdev(NBD_MAJOR, "nbd");
2450}
2451
2452module_init(nbd_init);
2453module_exit(nbd_cleanup);
2454
2455MODULE_DESCRIPTION("Network Block Device");
2456MODULE_LICENSE("GPL");
2457
2458module_param(nbds_max, int, 0444);
2459MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
2460module_param(max_part, int, 0444);
2461MODULE_PARM_DESC(max_part, "number of partitions per device (default: 16)");