2 * Network block device - make block devices work over TCP
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.
7 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
8 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
10 * This file is released under GPLv2 or later.
12 * (part of code stolen from loop.c)
15 #include <linux/major.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/init.h>
20 #include <linux/sched.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>
33 #include <linux/net.h>
34 #include <linux/kthread.h>
35 #include <linux/types.h>
36 #include <linux/debugfs.h>
37 #include <linux/blk-mq.h>
39 #include <linux/uaccess.h>
40 #include <asm/types.h>
42 #include <linux/nbd.h>
44 static DEFINE_IDR(nbd_index_idr);
45 static DEFINE_MUTEX(nbd_index_mutex);
52 #define NBD_TIMEDOUT 0
53 #define NBD_DISCONNECT_REQUESTED 1
54 #define NBD_DISCONNECTED 2
59 unsigned long runtime_flags;
60 struct nbd_sock **socks;
63 struct blk_mq_tag_set tag_set;
65 struct mutex config_lock;
68 atomic_t recv_threads;
69 wait_queue_head_t recv_wq;
73 struct task_struct *task_recv;
74 struct task_struct *task_setup;
76 #if IS_ENABLED(CONFIG_DEBUG_FS)
77 struct dentry *dbg_dir;
82 struct nbd_device *nbd;
83 struct completion send_complete;
86 #if IS_ENABLED(CONFIG_DEBUG_FS)
87 static struct dentry *nbd_dbg_dir;
90 #define nbd_name(nbd) ((nbd)->disk->disk_name)
92 #define NBD_MAGIC 0x68797548
94 static unsigned int nbds_max = 16;
96 static struct workqueue_struct *recv_workqueue;
97 static int part_shift;
99 static int nbd_dev_dbg_init(struct nbd_device *nbd);
100 static void nbd_dev_dbg_close(struct nbd_device *nbd);
103 static inline struct device *nbd_to_dev(struct nbd_device *nbd)
105 return disk_to_dev(nbd->disk);
108 static bool nbd_is_connected(struct nbd_device *nbd)
110 return !!nbd->task_recv;
113 static const char *nbdcmd_to_ascii(int cmd)
116 case NBD_CMD_READ: return "read";
117 case NBD_CMD_WRITE: return "write";
118 case NBD_CMD_DISC: return "disconnect";
119 case NBD_CMD_FLUSH: return "flush";
120 case NBD_CMD_TRIM: return "trim/discard";
125 static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev)
127 bd_set_size(bdev, 0);
128 set_capacity(nbd->disk, 0);
129 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
134 static void nbd_size_update(struct nbd_device *nbd, struct block_device *bdev)
136 blk_queue_logical_block_size(nbd->disk->queue, nbd->blksize);
137 blk_queue_physical_block_size(nbd->disk->queue, nbd->blksize);
138 bd_set_size(bdev, nbd->bytesize);
139 set_capacity(nbd->disk, nbd->bytesize >> 9);
140 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
143 static void nbd_size_set(struct nbd_device *nbd, struct block_device *bdev,
144 loff_t blocksize, loff_t nr_blocks)
146 nbd->blksize = blocksize;
147 nbd->bytesize = blocksize * nr_blocks;
148 if (nbd_is_connected(nbd))
149 nbd_size_update(nbd, bdev);
152 static void nbd_end_request(struct nbd_cmd *cmd)
154 struct nbd_device *nbd = cmd->nbd;
155 struct request *req = blk_mq_rq_from_pdu(cmd);
156 int error = req->errors ? -EIO : 0;
158 dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", cmd,
159 error ? "failed" : "done");
161 blk_mq_complete_request(req, error);
165 * Forcibly shutdown the socket causing all listeners to error
167 static void sock_shutdown(struct nbd_device *nbd)
171 if (nbd->num_connections == 0)
173 if (test_and_set_bit(NBD_DISCONNECTED, &nbd->runtime_flags))
176 for (i = 0; i < nbd->num_connections; i++) {
177 struct nbd_sock *nsock = nbd->socks[i];
178 mutex_lock(&nsock->tx_lock);
179 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
180 mutex_unlock(&nsock->tx_lock);
182 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
185 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
188 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
189 struct nbd_device *nbd = cmd->nbd;
191 dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n");
192 set_bit(NBD_TIMEDOUT, &nbd->runtime_flags);
195 mutex_lock(&nbd->config_lock);
197 mutex_unlock(&nbd->config_lock);
198 return BLK_EH_HANDLED;
202 * Send or receive packet.
204 static int sock_xmit(struct nbd_device *nbd, int index, int send, void *buf,
205 int size, int msg_flags)
207 struct socket *sock = nbd->socks[index]->sock;
211 unsigned long pflags = current->flags;
213 if (unlikely(!sock)) {
214 dev_err_ratelimited(disk_to_dev(nbd->disk),
215 "Attempted %s on closed socket in sock_xmit\n",
216 (send ? "send" : "recv"));
220 current->flags |= PF_MEMALLOC;
222 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
227 msg.msg_control = NULL;
228 msg.msg_controllen = 0;
229 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
232 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
234 result = kernel_recvmsg(sock, &msg, &iov, 1, size,
239 result = -EPIPE; /* short read */
246 tsk_restore_flags(current, pflags, PF_MEMALLOC);
251 static inline int sock_send_bvec(struct nbd_device *nbd, int index,
252 struct bio_vec *bvec, int flags)
255 void *kaddr = kmap(bvec->bv_page);
256 result = sock_xmit(nbd, index, 1, kaddr + bvec->bv_offset,
257 bvec->bv_len, flags);
258 kunmap(bvec->bv_page);
262 /* always call with the tx_lock held */
263 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
265 struct request *req = blk_mq_rq_from_pdu(cmd);
267 struct nbd_request request;
268 unsigned long size = blk_rq_bytes(req);
271 u32 tag = blk_mq_unique_tag(req);
273 switch (req_op(req)) {
278 type = NBD_CMD_FLUSH;
281 type = NBD_CMD_WRITE;
290 if (rq_data_dir(req) == WRITE &&
291 (nbd->flags & NBD_FLAG_READ_ONLY)) {
292 dev_err_ratelimited(disk_to_dev(nbd->disk),
293 "Write on read-only\n");
297 memset(&request, 0, sizeof(request));
298 request.magic = htonl(NBD_REQUEST_MAGIC);
299 request.type = htonl(type);
300 if (type != NBD_CMD_FLUSH) {
301 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
302 request.len = htonl(size);
304 memcpy(request.handle, &tag, sizeof(tag));
306 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
307 cmd, nbdcmd_to_ascii(type),
308 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
309 result = sock_xmit(nbd, index, 1, &request, sizeof(request),
310 (type == NBD_CMD_WRITE) ? MSG_MORE : 0);
312 dev_err_ratelimited(disk_to_dev(nbd->disk),
313 "Send control failed (result %d)\n", result);
317 if (type != NBD_CMD_WRITE)
322 struct bio *next = bio->bi_next;
323 struct bvec_iter iter;
326 bio_for_each_segment(bvec, bio, iter) {
327 bool is_last = !next && bio_iter_last(bvec, iter);
328 int flags = is_last ? 0 : MSG_MORE;
330 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
332 result = sock_send_bvec(nbd, index, &bvec, flags);
334 dev_err(disk_to_dev(nbd->disk),
335 "Send data failed (result %d)\n",
340 * The completion might already have come in,
341 * so break for the last one instead of letting
342 * the iterator do it. This prevents use-after-free
353 static inline int sock_recv_bvec(struct nbd_device *nbd, int index,
354 struct bio_vec *bvec)
357 void *kaddr = kmap(bvec->bv_page);
358 result = sock_xmit(nbd, index, 0, kaddr + bvec->bv_offset,
359 bvec->bv_len, MSG_WAITALL);
360 kunmap(bvec->bv_page);
364 /* NULL returned = something went wrong, inform userspace */
365 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
368 struct nbd_reply reply;
370 struct request *req = NULL;
375 result = sock_xmit(nbd, index, 0, &reply, sizeof(reply), MSG_WAITALL);
377 if (!test_bit(NBD_DISCONNECTED, &nbd->runtime_flags) &&
378 !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
379 dev_err(disk_to_dev(nbd->disk),
380 "Receive control failed (result %d)\n", result);
381 return ERR_PTR(result);
384 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
385 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
386 (unsigned long)ntohl(reply.magic));
387 return ERR_PTR(-EPROTO);
390 memcpy(&tag, reply.handle, sizeof(u32));
392 hwq = blk_mq_unique_tag_to_hwq(tag);
393 if (hwq < nbd->tag_set.nr_hw_queues)
394 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
395 blk_mq_unique_tag_to_tag(tag));
396 if (!req || !blk_mq_request_started(req)) {
397 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
399 return ERR_PTR(-ENOENT);
401 cmd = blk_mq_rq_to_pdu(req);
402 if (ntohl(reply.error)) {
403 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
409 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
410 if (rq_data_dir(req) != WRITE) {
411 struct req_iterator iter;
414 rq_for_each_segment(bvec, req, iter) {
415 result = sock_recv_bvec(nbd, index, &bvec);
417 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
422 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
426 /* See the comment in nbd_queue_rq. */
427 wait_for_completion(&cmd->send_complete);
432 static ssize_t pid_show(struct device *dev,
433 struct device_attribute *attr, char *buf)
435 struct gendisk *disk = dev_to_disk(dev);
436 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
438 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
441 static struct device_attribute pid_attr = {
442 .attr = { .name = "pid", .mode = S_IRUGO},
446 struct recv_thread_args {
447 struct work_struct work;
448 struct nbd_device *nbd;
452 static void recv_work(struct work_struct *work)
454 struct recv_thread_args *args = container_of(work,
455 struct recv_thread_args,
457 struct nbd_device *nbd = args->nbd;
461 BUG_ON(nbd->magic != NBD_MAGIC);
463 cmd = nbd_read_stat(nbd, args->index);
469 nbd_end_request(cmd);
473 * We got an error, shut everybody down if this wasn't the result of a
474 * disconnect request.
476 if (ret && !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
478 atomic_dec(&nbd->recv_threads);
479 wake_up(&nbd->recv_wq);
482 static void nbd_clear_req(struct request *req, void *data, bool reserved)
486 if (!blk_mq_request_started(req))
488 cmd = blk_mq_rq_to_pdu(req);
490 nbd_end_request(cmd);
493 static void nbd_clear_que(struct nbd_device *nbd)
495 BUG_ON(nbd->magic != NBD_MAGIC);
497 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
498 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
502 static void nbd_handle_cmd(struct nbd_cmd *cmd, int index)
504 struct request *req = blk_mq_rq_from_pdu(cmd);
505 struct nbd_device *nbd = cmd->nbd;
506 struct nbd_sock *nsock;
508 if (index >= nbd->num_connections) {
509 dev_err_ratelimited(disk_to_dev(nbd->disk),
510 "Attempted send on invalid socket\n");
514 if (test_bit(NBD_DISCONNECTED, &nbd->runtime_flags)) {
515 dev_err_ratelimited(disk_to_dev(nbd->disk),
516 "Attempted send on closed socket\n");
522 nsock = nbd->socks[index];
523 mutex_lock(&nsock->tx_lock);
524 if (unlikely(!nsock->sock)) {
525 mutex_unlock(&nsock->tx_lock);
526 dev_err_ratelimited(disk_to_dev(nbd->disk),
527 "Attempted send on closed socket\n");
531 if (nbd_send_cmd(nbd, cmd, index) != 0) {
532 dev_err_ratelimited(disk_to_dev(nbd->disk),
533 "Request send failed\n");
535 nbd_end_request(cmd);
538 mutex_unlock(&nsock->tx_lock);
544 nbd_end_request(cmd);
547 static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
548 const struct blk_mq_queue_data *bd)
550 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
553 * Since we look at the bio's to send the request over the network we
554 * need to make sure the completion work doesn't mark this request done
555 * before we are done doing our send. This keeps us from dereferencing
556 * freed data if we have particularly fast completions (ie we get the
557 * completion before we exit sock_xmit on the last bvec) or in the case
558 * that the server is misbehaving (or there was an error) before we're
559 * done sending everything over the wire.
561 init_completion(&cmd->send_complete);
562 blk_mq_start_request(bd->rq);
563 nbd_handle_cmd(cmd, hctx->queue_num);
564 complete(&cmd->send_complete);
566 return BLK_MQ_RQ_QUEUE_OK;
569 static int nbd_add_socket(struct nbd_device *nbd, struct block_device *bdev,
573 struct nbd_sock **socks;
574 struct nbd_sock *nsock;
577 sock = sockfd_lookup(arg, &err);
581 if (!nbd->task_setup)
582 nbd->task_setup = current;
583 if (nbd->task_setup != current) {
584 dev_err(disk_to_dev(nbd->disk),
585 "Device being setup by another task");
589 socks = krealloc(nbd->socks, (nbd->num_connections + 1) *
590 sizeof(struct nbd_sock *), GFP_KERNEL);
593 nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
599 mutex_init(&nsock->tx_lock);
601 socks[nbd->num_connections++] = nsock;
604 bdev->bd_invalidated = 1;
608 /* Reset all properties of an NBD device */
609 static void nbd_reset(struct nbd_device *nbd)
611 nbd->runtime_flags = 0;
614 set_capacity(nbd->disk, 0);
616 nbd->tag_set.timeout = 0;
617 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
620 static void nbd_bdev_reset(struct block_device *bdev)
622 set_device_ro(bdev, false);
623 bdev->bd_inode->i_size = 0;
625 blkdev_reread_part(bdev);
626 bdev->bd_invalidated = 1;
630 static void nbd_parse_flags(struct nbd_device *nbd, struct block_device *bdev)
632 if (nbd->flags & NBD_FLAG_READ_ONLY)
633 set_device_ro(bdev, true);
634 if (nbd->flags & NBD_FLAG_SEND_TRIM)
635 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
636 if (nbd->flags & NBD_FLAG_SEND_FLUSH)
637 blk_queue_write_cache(nbd->disk->queue, true, false);
639 blk_queue_write_cache(nbd->disk->queue, false, false);
642 static void send_disconnects(struct nbd_device *nbd)
644 struct nbd_request request = {};
647 request.magic = htonl(NBD_REQUEST_MAGIC);
648 request.type = htonl(NBD_CMD_DISC);
650 for (i = 0; i < nbd->num_connections; i++) {
651 ret = sock_xmit(nbd, i, 1, &request, sizeof(request), 0);
653 dev_err(disk_to_dev(nbd->disk),
654 "Send disconnect failed %d\n", ret);
658 static int nbd_disconnect(struct nbd_device *nbd, struct block_device *bdev)
660 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
664 mutex_unlock(&nbd->config_lock);
666 mutex_lock(&nbd->config_lock);
668 /* Check again after getting mutex back. */
672 if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
673 &nbd->runtime_flags))
674 send_disconnects(nbd);
678 static int nbd_clear_sock(struct nbd_device *nbd, struct block_device *bdev)
683 nbd_bdev_reset(bdev);
685 * We want to give the run thread a chance to wait for everybody
686 * to clean up and then do it's own cleanup.
688 if (!test_bit(NBD_RUNNING, &nbd->runtime_flags) &&
689 nbd->num_connections) {
692 for (i = 0; i < nbd->num_connections; i++)
693 kfree(nbd->socks[i]);
696 nbd->num_connections = 0;
698 nbd->task_setup = NULL;
703 static int nbd_start_device(struct nbd_device *nbd, struct block_device *bdev)
705 struct recv_thread_args *args;
706 int num_connections = nbd->num_connections;
713 if (num_connections > 1 &&
714 !(nbd->flags & NBD_FLAG_CAN_MULTI_CONN)) {
715 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
720 set_bit(NBD_RUNNING, &nbd->runtime_flags);
721 blk_mq_update_nr_hw_queues(&nbd->tag_set, nbd->num_connections);
722 args = kcalloc(num_connections, sizeof(*args), GFP_KERNEL);
727 nbd->task_recv = current;
728 mutex_unlock(&nbd->config_lock);
730 nbd_parse_flags(nbd, bdev);
732 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
734 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
738 nbd_size_update(nbd, bdev);
740 nbd_dev_dbg_init(nbd);
741 for (i = 0; i < num_connections; i++) {
742 sk_set_memalloc(nbd->socks[i]->sock->sk);
743 atomic_inc(&nbd->recv_threads);
744 INIT_WORK(&args[i].work, recv_work);
747 queue_work(recv_workqueue, &args[i].work);
749 wait_event_interruptible(nbd->recv_wq,
750 atomic_read(&nbd->recv_threads) == 0);
751 for (i = 0; i < num_connections; i++)
752 flush_work(&args[i].work);
753 nbd_dev_dbg_close(nbd);
754 nbd_size_clear(nbd, bdev);
755 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
757 mutex_lock(&nbd->config_lock);
758 nbd->task_recv = NULL;
760 clear_bit(NBD_RUNNING, &nbd->runtime_flags);
761 nbd_clear_sock(nbd, bdev);
763 /* user requested, ignore socket errors */
764 if (test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
766 if (test_bit(NBD_TIMEDOUT, &nbd->runtime_flags))
773 /* Must be called with config_lock held */
774 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
775 unsigned int cmd, unsigned long arg)
779 return nbd_disconnect(nbd, bdev);
781 return nbd_clear_sock(nbd, bdev);
783 return nbd_add_socket(nbd, bdev, arg);
784 case NBD_SET_BLKSIZE:
785 nbd_size_set(nbd, bdev, arg,
786 div_s64(nbd->bytesize, arg));
789 nbd_size_set(nbd, bdev, nbd->blksize,
790 div_s64(arg, nbd->blksize));
792 case NBD_SET_SIZE_BLOCKS:
793 nbd_size_set(nbd, bdev, nbd->blksize, arg);
795 case NBD_SET_TIMEOUT:
796 nbd->tag_set.timeout = arg * HZ;
803 return nbd_start_device(nbd, bdev);
806 * This is for compatibility only. The queue is always cleared
807 * by NBD_DO_IT or NBD_CLEAR_SOCK.
810 case NBD_PRINT_DEBUG:
812 * For compatibility only, we no longer keep a list of
813 * outstanding requests.
820 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
821 unsigned int cmd, unsigned long arg)
823 struct nbd_device *nbd = bdev->bd_disk->private_data;
826 if (!capable(CAP_SYS_ADMIN))
829 BUG_ON(nbd->magic != NBD_MAGIC);
831 mutex_lock(&nbd->config_lock);
832 error = __nbd_ioctl(bdev, nbd, cmd, arg);
833 mutex_unlock(&nbd->config_lock);
838 static const struct block_device_operations nbd_fops =
840 .owner = THIS_MODULE,
842 .compat_ioctl = nbd_ioctl,
845 #if IS_ENABLED(CONFIG_DEBUG_FS)
847 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
849 struct nbd_device *nbd = s->private;
852 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
857 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
859 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
862 static const struct file_operations nbd_dbg_tasks_ops = {
863 .open = nbd_dbg_tasks_open,
866 .release = single_release,
869 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
871 struct nbd_device *nbd = s->private;
872 u32 flags = nbd->flags;
874 seq_printf(s, "Hex: 0x%08x\n\n", flags);
876 seq_puts(s, "Known flags:\n");
878 if (flags & NBD_FLAG_HAS_FLAGS)
879 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
880 if (flags & NBD_FLAG_READ_ONLY)
881 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
882 if (flags & NBD_FLAG_SEND_FLUSH)
883 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
884 if (flags & NBD_FLAG_SEND_TRIM)
885 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
890 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
892 return single_open(file, nbd_dbg_flags_show, inode->i_private);
895 static const struct file_operations nbd_dbg_flags_ops = {
896 .open = nbd_dbg_flags_open,
899 .release = single_release,
902 static int nbd_dev_dbg_init(struct nbd_device *nbd)
909 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
911 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
917 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
918 debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize);
919 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
920 debugfs_create_u64("blocksize", 0444, dir, &nbd->blksize);
921 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
926 static void nbd_dev_dbg_close(struct nbd_device *nbd)
928 debugfs_remove_recursive(nbd->dbg_dir);
931 static int nbd_dbg_init(void)
933 struct dentry *dbg_dir;
935 dbg_dir = debugfs_create_dir("nbd", NULL);
939 nbd_dbg_dir = dbg_dir;
944 static void nbd_dbg_close(void)
946 debugfs_remove_recursive(nbd_dbg_dir);
949 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
951 static int nbd_dev_dbg_init(struct nbd_device *nbd)
956 static void nbd_dev_dbg_close(struct nbd_device *nbd)
960 static int nbd_dbg_init(void)
965 static void nbd_dbg_close(void)
971 static int nbd_init_request(void *data, struct request *rq,
972 unsigned int hctx_idx, unsigned int request_idx,
973 unsigned int numa_node)
975 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
980 static struct blk_mq_ops nbd_mq_ops = {
981 .queue_rq = nbd_queue_rq,
982 .init_request = nbd_init_request,
983 .timeout = nbd_xmit_timeout,
986 static void nbd_dev_remove(struct nbd_device *nbd)
988 struct gendisk *disk = nbd->disk;
992 blk_cleanup_queue(disk->queue);
993 blk_mq_free_tag_set(&nbd->tag_set);
999 static int nbd_dev_add(int index)
1001 struct nbd_device *nbd;
1002 struct gendisk *disk;
1003 struct request_queue *q;
1006 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1010 disk = alloc_disk(1 << part_shift);
1015 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1020 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1028 nbd->tag_set.ops = &nbd_mq_ops;
1029 nbd->tag_set.nr_hw_queues = 1;
1030 nbd->tag_set.queue_depth = 128;
1031 nbd->tag_set.numa_node = NUMA_NO_NODE;
1032 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1033 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1034 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1035 nbd->tag_set.driver_data = nbd;
1037 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1041 q = blk_mq_init_queue(&nbd->tag_set);
1049 * Tell the block layer that we are not a rotational device
1051 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1052 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1053 disk->queue->limits.discard_granularity = 512;
1054 blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1055 disk->queue->limits.discard_zeroes_data = 0;
1056 blk_queue_max_hw_sectors(disk->queue, 65536);
1057 disk->queue->limits.max_sectors = 256;
1059 nbd->magic = NBD_MAGIC;
1060 mutex_init(&nbd->config_lock);
1061 disk->major = NBD_MAJOR;
1062 disk->first_minor = index << part_shift;
1063 disk->fops = &nbd_fops;
1064 disk->private_data = nbd;
1065 sprintf(disk->disk_name, "nbd%d", index);
1066 init_waitqueue_head(&nbd->recv_wq);
1072 blk_mq_free_tag_set(&nbd->tag_set);
1074 idr_remove(&nbd_index_idr, index);
1084 * And here should be modules and kernel interface
1085 * (Just smiley confuses emacs :-)
1088 static int __init nbd_init(void)
1092 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
1095 printk(KERN_ERR "nbd: max_part must be >= 0\n");
1101 part_shift = fls(max_part);
1104 * Adjust max_part according to part_shift as it is exported
1105 * to user space so that user can know the max number of
1106 * partition kernel should be able to manage.
1108 * Note that -1 is required because partition 0 is reserved
1109 * for the whole disk.
1111 max_part = (1UL << part_shift) - 1;
1114 if ((1UL << part_shift) > DISK_MAX_PARTS)
1117 if (nbds_max > 1UL << (MINORBITS - part_shift))
1119 recv_workqueue = alloc_workqueue("knbd-recv",
1120 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1121 if (!recv_workqueue)
1124 if (register_blkdev(NBD_MAJOR, "nbd")) {
1125 destroy_workqueue(recv_workqueue);
1131 mutex_lock(&nbd_index_mutex);
1132 for (i = 0; i < nbds_max; i++)
1134 mutex_unlock(&nbd_index_mutex);
1138 static int nbd_exit_cb(int id, void *ptr, void *data)
1140 struct nbd_device *nbd = ptr;
1141 nbd_dev_remove(nbd);
1145 static void __exit nbd_cleanup(void)
1149 idr_for_each(&nbd_index_idr, &nbd_exit_cb, NULL);
1150 idr_destroy(&nbd_index_idr);
1151 destroy_workqueue(recv_workqueue);
1152 unregister_blkdev(NBD_MAJOR, "nbd");
1155 module_init(nbd_init);
1156 module_exit(nbd_cleanup);
1158 MODULE_DESCRIPTION("Network Block Device");
1159 MODULE_LICENSE("GPL");
1161 module_param(nbds_max, int, 0444);
1162 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
1163 module_param(max_part, int, 0444);
1164 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");