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 inline struct device *nbd_to_dev(struct nbd_device *nbd)
101 return disk_to_dev(nbd->disk);
104 static bool nbd_is_connected(struct nbd_device *nbd)
106 return !!nbd->task_recv;
109 static const char *nbdcmd_to_ascii(int cmd)
112 case NBD_CMD_READ: return "read";
113 case NBD_CMD_WRITE: return "write";
114 case NBD_CMD_DISC: return "disconnect";
115 case NBD_CMD_FLUSH: return "flush";
116 case NBD_CMD_TRIM: return "trim/discard";
121 static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev)
123 bdev->bd_inode->i_size = 0;
124 set_capacity(nbd->disk, 0);
125 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
130 static void nbd_size_update(struct nbd_device *nbd, struct block_device *bdev)
132 if (!nbd_is_connected(nbd))
135 bdev->bd_inode->i_size = nbd->bytesize;
136 set_capacity(nbd->disk, nbd->bytesize >> 9);
137 kobject_uevent(&nbd_to_dev(nbd)->kobj, KOBJ_CHANGE);
140 static int nbd_size_set(struct nbd_device *nbd, struct block_device *bdev,
141 loff_t blocksize, loff_t nr_blocks)
145 ret = set_blocksize(bdev, blocksize);
149 nbd->blksize = blocksize;
150 nbd->bytesize = blocksize * nr_blocks;
152 nbd_size_update(nbd, bdev);
157 static void nbd_end_request(struct nbd_cmd *cmd)
159 struct nbd_device *nbd = cmd->nbd;
160 struct request *req = blk_mq_rq_from_pdu(cmd);
161 int error = req->errors ? -EIO : 0;
163 dev_dbg(nbd_to_dev(nbd), "request %p: %s\n", cmd,
164 error ? "failed" : "done");
166 blk_mq_complete_request(req, error);
170 * Forcibly shutdown the socket causing all listeners to error
172 static void sock_shutdown(struct nbd_device *nbd)
176 if (nbd->num_connections == 0)
178 if (test_and_set_bit(NBD_DISCONNECTED, &nbd->runtime_flags))
181 for (i = 0; i < nbd->num_connections; i++) {
182 struct nbd_sock *nsock = nbd->socks[i];
183 mutex_lock(&nsock->tx_lock);
184 kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
185 mutex_unlock(&nsock->tx_lock);
187 dev_warn(disk_to_dev(nbd->disk), "shutting down sockets\n");
190 static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
193 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
194 struct nbd_device *nbd = cmd->nbd;
196 dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n");
197 set_bit(NBD_TIMEDOUT, &nbd->runtime_flags);
200 mutex_lock(&nbd->config_lock);
202 mutex_unlock(&nbd->config_lock);
203 return BLK_EH_HANDLED;
207 * Send or receive packet.
209 static int sock_xmit(struct nbd_device *nbd, int index, int send, void *buf,
210 int size, int msg_flags)
212 struct socket *sock = nbd->socks[index]->sock;
216 unsigned long pflags = current->flags;
218 if (unlikely(!sock)) {
219 dev_err_ratelimited(disk_to_dev(nbd->disk),
220 "Attempted %s on closed socket in sock_xmit\n",
221 (send ? "send" : "recv"));
225 current->flags |= PF_MEMALLOC;
227 sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
232 msg.msg_control = NULL;
233 msg.msg_controllen = 0;
234 msg.msg_flags = msg_flags | MSG_NOSIGNAL;
237 result = kernel_sendmsg(sock, &msg, &iov, 1, size);
239 result = kernel_recvmsg(sock, &msg, &iov, 1, size,
244 result = -EPIPE; /* short read */
251 tsk_restore_flags(current, pflags, PF_MEMALLOC);
256 static inline int sock_send_bvec(struct nbd_device *nbd, int index,
257 struct bio_vec *bvec, int flags)
260 void *kaddr = kmap(bvec->bv_page);
261 result = sock_xmit(nbd, index, 1, kaddr + bvec->bv_offset,
262 bvec->bv_len, flags);
263 kunmap(bvec->bv_page);
267 /* always call with the tx_lock held */
268 static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
270 struct request *req = blk_mq_rq_from_pdu(cmd);
272 struct nbd_request request;
273 unsigned long size = blk_rq_bytes(req);
276 u32 tag = blk_mq_unique_tag(req);
278 switch (req_op(req)) {
283 type = NBD_CMD_FLUSH;
286 type = NBD_CMD_WRITE;
295 if (rq_data_dir(req) == WRITE &&
296 (nbd->flags & NBD_FLAG_READ_ONLY)) {
297 dev_err_ratelimited(disk_to_dev(nbd->disk),
298 "Write on read-only\n");
302 memset(&request, 0, sizeof(request));
303 request.magic = htonl(NBD_REQUEST_MAGIC);
304 request.type = htonl(type);
305 if (type != NBD_CMD_FLUSH) {
306 request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
307 request.len = htonl(size);
309 memcpy(request.handle, &tag, sizeof(tag));
311 dev_dbg(nbd_to_dev(nbd), "request %p: sending control (%s@%llu,%uB)\n",
312 cmd, nbdcmd_to_ascii(type),
313 (unsigned long long)blk_rq_pos(req) << 9, blk_rq_bytes(req));
314 result = sock_xmit(nbd, index, 1, &request, sizeof(request),
315 (type == NBD_CMD_WRITE) ? MSG_MORE : 0);
317 dev_err_ratelimited(disk_to_dev(nbd->disk),
318 "Send control failed (result %d)\n", result);
322 if (type != NBD_CMD_WRITE)
327 struct bio *next = bio->bi_next;
328 struct bvec_iter iter;
331 bio_for_each_segment(bvec, bio, iter) {
332 bool is_last = !next && bio_iter_last(bvec, iter);
333 int flags = is_last ? 0 : MSG_MORE;
335 dev_dbg(nbd_to_dev(nbd), "request %p: sending %d bytes data\n",
337 result = sock_send_bvec(nbd, index, &bvec, flags);
339 dev_err(disk_to_dev(nbd->disk),
340 "Send data failed (result %d)\n",
345 * The completion might already have come in,
346 * so break for the last one instead of letting
347 * the iterator do it. This prevents use-after-free
358 static inline int sock_recv_bvec(struct nbd_device *nbd, int index,
359 struct bio_vec *bvec)
362 void *kaddr = kmap(bvec->bv_page);
363 result = sock_xmit(nbd, index, 0, kaddr + bvec->bv_offset,
364 bvec->bv_len, MSG_WAITALL);
365 kunmap(bvec->bv_page);
369 /* NULL returned = something went wrong, inform userspace */
370 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
373 struct nbd_reply reply;
375 struct request *req = NULL;
380 result = sock_xmit(nbd, index, 0, &reply, sizeof(reply), MSG_WAITALL);
382 if (!test_bit(NBD_DISCONNECTED, &nbd->runtime_flags) &&
383 !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
384 dev_err(disk_to_dev(nbd->disk),
385 "Receive control failed (result %d)\n", result);
386 return ERR_PTR(result);
389 if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
390 dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
391 (unsigned long)ntohl(reply.magic));
392 return ERR_PTR(-EPROTO);
395 memcpy(&tag, reply.handle, sizeof(u32));
397 hwq = blk_mq_unique_tag_to_hwq(tag);
398 if (hwq < nbd->tag_set.nr_hw_queues)
399 req = blk_mq_tag_to_rq(nbd->tag_set.tags[hwq],
400 blk_mq_unique_tag_to_tag(tag));
401 if (!req || !blk_mq_request_started(req)) {
402 dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%d) %p\n",
404 return ERR_PTR(-ENOENT);
406 cmd = blk_mq_rq_to_pdu(req);
407 if (ntohl(reply.error)) {
408 dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
414 dev_dbg(nbd_to_dev(nbd), "request %p: got reply\n", cmd);
415 if (rq_data_dir(req) != WRITE) {
416 struct req_iterator iter;
419 rq_for_each_segment(bvec, req, iter) {
420 result = sock_recv_bvec(nbd, index, &bvec);
422 dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
427 dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
431 /* See the comment in nbd_queue_rq. */
432 wait_for_completion(&cmd->send_complete);
437 static ssize_t pid_show(struct device *dev,
438 struct device_attribute *attr, char *buf)
440 struct gendisk *disk = dev_to_disk(dev);
441 struct nbd_device *nbd = (struct nbd_device *)disk->private_data;
443 return sprintf(buf, "%d\n", task_pid_nr(nbd->task_recv));
446 static struct device_attribute pid_attr = {
447 .attr = { .name = "pid", .mode = S_IRUGO},
451 struct recv_thread_args {
452 struct work_struct work;
453 struct nbd_device *nbd;
457 static void recv_work(struct work_struct *work)
459 struct recv_thread_args *args = container_of(work,
460 struct recv_thread_args,
462 struct nbd_device *nbd = args->nbd;
466 BUG_ON(nbd->magic != NBD_MAGIC);
468 cmd = nbd_read_stat(nbd, args->index);
474 nbd_end_request(cmd);
478 * We got an error, shut everybody down if this wasn't the result of a
479 * disconnect request.
481 if (ret && !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
483 atomic_dec(&nbd->recv_threads);
484 wake_up(&nbd->recv_wq);
487 static void nbd_clear_req(struct request *req, void *data, bool reserved)
491 if (!blk_mq_request_started(req))
493 cmd = blk_mq_rq_to_pdu(req);
495 nbd_end_request(cmd);
498 static void nbd_clear_que(struct nbd_device *nbd)
500 BUG_ON(nbd->magic != NBD_MAGIC);
502 blk_mq_tagset_busy_iter(&nbd->tag_set, nbd_clear_req, NULL);
503 dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
507 static void nbd_handle_cmd(struct nbd_cmd *cmd, int index)
509 struct request *req = blk_mq_rq_from_pdu(cmd);
510 struct nbd_device *nbd = cmd->nbd;
511 struct nbd_sock *nsock;
513 if (index >= nbd->num_connections) {
514 dev_err_ratelimited(disk_to_dev(nbd->disk),
515 "Attempted send on invalid socket\n");
519 if (test_bit(NBD_DISCONNECTED, &nbd->runtime_flags)) {
520 dev_err_ratelimited(disk_to_dev(nbd->disk),
521 "Attempted send on closed socket\n");
527 nsock = nbd->socks[index];
528 mutex_lock(&nsock->tx_lock);
529 if (unlikely(!nsock->sock)) {
530 mutex_unlock(&nsock->tx_lock);
531 dev_err_ratelimited(disk_to_dev(nbd->disk),
532 "Attempted send on closed socket\n");
536 if (nbd_send_cmd(nbd, cmd, index) != 0) {
537 dev_err_ratelimited(disk_to_dev(nbd->disk),
538 "Request send failed\n");
540 nbd_end_request(cmd);
543 mutex_unlock(&nsock->tx_lock);
549 nbd_end_request(cmd);
552 static int nbd_queue_rq(struct blk_mq_hw_ctx *hctx,
553 const struct blk_mq_queue_data *bd)
555 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(bd->rq);
558 * Since we look at the bio's to send the request over the network we
559 * need to make sure the completion work doesn't mark this request done
560 * before we are done doing our send. This keeps us from dereferencing
561 * freed data if we have particularly fast completions (ie we get the
562 * completion before we exit sock_xmit on the last bvec) or in the case
563 * that the server is misbehaving (or there was an error) before we're
564 * done sending everything over the wire.
566 init_completion(&cmd->send_complete);
567 blk_mq_start_request(bd->rq);
568 nbd_handle_cmd(cmd, hctx->queue_num);
569 complete(&cmd->send_complete);
571 return BLK_MQ_RQ_QUEUE_OK;
574 static int nbd_add_socket(struct nbd_device *nbd, struct socket *sock)
576 struct nbd_sock **socks;
577 struct nbd_sock *nsock;
579 if (!nbd->task_setup)
580 nbd->task_setup = current;
581 if (nbd->task_setup != current) {
582 dev_err(disk_to_dev(nbd->disk),
583 "Device being setup by another task");
587 socks = krealloc(nbd->socks, (nbd->num_connections + 1) *
588 sizeof(struct nbd_sock *), GFP_KERNEL);
591 nsock = kzalloc(sizeof(struct nbd_sock), GFP_KERNEL);
597 mutex_init(&nsock->tx_lock);
599 socks[nbd->num_connections++] = nsock;
604 /* Reset all properties of an NBD device */
605 static void nbd_reset(struct nbd_device *nbd)
609 for (i = 0; i < nbd->num_connections; i++)
610 kfree(nbd->socks[i]);
613 nbd->runtime_flags = 0;
616 set_capacity(nbd->disk, 0);
618 nbd->tag_set.timeout = 0;
619 nbd->num_connections = 0;
620 nbd->task_setup = NULL;
621 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
624 static void nbd_bdev_reset(struct block_device *bdev)
626 set_device_ro(bdev, false);
627 bdev->bd_inode->i_size = 0;
629 blkdev_reread_part(bdev);
630 bdev->bd_invalidated = 1;
634 static void nbd_parse_flags(struct nbd_device *nbd, struct block_device *bdev)
636 if (nbd->flags & NBD_FLAG_READ_ONLY)
637 set_device_ro(bdev, true);
638 if (nbd->flags & NBD_FLAG_SEND_TRIM)
639 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
640 if (nbd->flags & NBD_FLAG_SEND_FLUSH)
641 blk_queue_write_cache(nbd->disk->queue, true, false);
643 blk_queue_write_cache(nbd->disk->queue, false, false);
646 static void send_disconnects(struct nbd_device *nbd)
648 struct nbd_request request = {};
651 request.magic = htonl(NBD_REQUEST_MAGIC);
652 request.type = htonl(NBD_CMD_DISC);
654 for (i = 0; i < nbd->num_connections; i++) {
655 ret = sock_xmit(nbd, i, 1, &request, sizeof(request), 0);
657 dev_err(disk_to_dev(nbd->disk),
658 "Send disconnect failed %d\n", ret);
662 static int nbd_dev_dbg_init(struct nbd_device *nbd);
663 static void nbd_dev_dbg_close(struct nbd_device *nbd);
665 /* Must be called with config_lock held */
666 static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
667 unsigned int cmd, unsigned long arg)
670 case NBD_DISCONNECT: {
671 dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
675 mutex_unlock(&nbd->config_lock);
677 mutex_lock(&nbd->config_lock);
679 /* Check again after getting mutex back. */
683 if (!test_and_set_bit(NBD_DISCONNECT_REQUESTED,
684 &nbd->runtime_flags))
685 send_disconnects(nbd);
693 nbd_bdev_reset(bdev);
695 * We want to give the run thread a chance to wait for everybody
696 * to clean up and then do it's own cleanup.
698 if (!test_bit(NBD_RUNNING, &nbd->runtime_flags)) {
701 for (i = 0; i < nbd->num_connections; i++)
702 kfree(nbd->socks[i]);
705 nbd->num_connections = 0;
706 nbd->task_setup = NULL;
712 struct socket *sock = sockfd_lookup(arg, &err);
717 err = nbd_add_socket(nbd, sock);
718 if (!err && max_part)
719 bdev->bd_invalidated = 1;
724 case NBD_SET_BLKSIZE: {
725 loff_t bsize = div_s64(nbd->bytesize, arg);
727 return nbd_size_set(nbd, bdev, arg, bsize);
731 return nbd_size_set(nbd, bdev, nbd->blksize,
732 div_s64(arg, nbd->blksize));
734 case NBD_SET_SIZE_BLOCKS:
735 return nbd_size_set(nbd, bdev, nbd->blksize, arg);
737 case NBD_SET_TIMEOUT:
738 nbd->tag_set.timeout = arg * HZ;
746 struct recv_thread_args *args;
747 int num_connections = nbd->num_connections;
754 if (num_connections > 1 &&
755 !(nbd->flags & NBD_FLAG_CAN_MULTI_CONN)) {
756 dev_err(disk_to_dev(nbd->disk), "server does not support multiple connections per device.\n");
761 set_bit(NBD_RUNNING, &nbd->runtime_flags);
762 blk_mq_update_nr_hw_queues(&nbd->tag_set, nbd->num_connections);
763 args = kcalloc(num_connections, sizeof(*args), GFP_KERNEL);
768 nbd->task_recv = current;
769 mutex_unlock(&nbd->config_lock);
771 nbd_parse_flags(nbd, bdev);
773 error = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
775 dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
779 nbd_size_update(nbd, bdev);
781 nbd_dev_dbg_init(nbd);
782 for (i = 0; i < num_connections; i++) {
783 sk_set_memalloc(nbd->socks[i]->sock->sk);
784 atomic_inc(&nbd->recv_threads);
785 INIT_WORK(&args[i].work, recv_work);
788 queue_work(recv_workqueue, &args[i].work);
790 wait_event_interruptible(nbd->recv_wq,
791 atomic_read(&nbd->recv_threads) == 0);
792 for (i = 0; i < num_connections; i++)
793 flush_work(&args[i].work);
794 nbd_dev_dbg_close(nbd);
795 nbd_size_clear(nbd, bdev);
796 device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
798 mutex_lock(&nbd->config_lock);
799 nbd->task_recv = NULL;
804 nbd_bdev_reset(bdev);
806 /* user requested, ignore socket errors */
807 if (test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
809 if (test_bit(NBD_TIMEDOUT, &nbd->runtime_flags))
818 * This is for compatibility only. The queue is always cleared
819 * by NBD_DO_IT or NBD_CLEAR_SOCK.
823 case NBD_PRINT_DEBUG:
825 * For compatibility only, we no longer keep a list of
826 * outstanding requests.
833 static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
834 unsigned int cmd, unsigned long arg)
836 struct nbd_device *nbd = bdev->bd_disk->private_data;
839 if (!capable(CAP_SYS_ADMIN))
842 BUG_ON(nbd->magic != NBD_MAGIC);
844 mutex_lock(&nbd->config_lock);
845 error = __nbd_ioctl(bdev, nbd, cmd, arg);
846 mutex_unlock(&nbd->config_lock);
851 static const struct block_device_operations nbd_fops =
853 .owner = THIS_MODULE,
855 .compat_ioctl = nbd_ioctl,
858 #if IS_ENABLED(CONFIG_DEBUG_FS)
860 static int nbd_dbg_tasks_show(struct seq_file *s, void *unused)
862 struct nbd_device *nbd = s->private;
865 seq_printf(s, "recv: %d\n", task_pid_nr(nbd->task_recv));
870 static int nbd_dbg_tasks_open(struct inode *inode, struct file *file)
872 return single_open(file, nbd_dbg_tasks_show, inode->i_private);
875 static const struct file_operations nbd_dbg_tasks_ops = {
876 .open = nbd_dbg_tasks_open,
879 .release = single_release,
882 static int nbd_dbg_flags_show(struct seq_file *s, void *unused)
884 struct nbd_device *nbd = s->private;
885 u32 flags = nbd->flags;
887 seq_printf(s, "Hex: 0x%08x\n\n", flags);
889 seq_puts(s, "Known flags:\n");
891 if (flags & NBD_FLAG_HAS_FLAGS)
892 seq_puts(s, "NBD_FLAG_HAS_FLAGS\n");
893 if (flags & NBD_FLAG_READ_ONLY)
894 seq_puts(s, "NBD_FLAG_READ_ONLY\n");
895 if (flags & NBD_FLAG_SEND_FLUSH)
896 seq_puts(s, "NBD_FLAG_SEND_FLUSH\n");
897 if (flags & NBD_FLAG_SEND_TRIM)
898 seq_puts(s, "NBD_FLAG_SEND_TRIM\n");
903 static int nbd_dbg_flags_open(struct inode *inode, struct file *file)
905 return single_open(file, nbd_dbg_flags_show, inode->i_private);
908 static const struct file_operations nbd_dbg_flags_ops = {
909 .open = nbd_dbg_flags_open,
912 .release = single_release,
915 static int nbd_dev_dbg_init(struct nbd_device *nbd)
922 dir = debugfs_create_dir(nbd_name(nbd), nbd_dbg_dir);
924 dev_err(nbd_to_dev(nbd), "Failed to create debugfs dir for '%s'\n",
930 debugfs_create_file("tasks", 0444, dir, nbd, &nbd_dbg_tasks_ops);
931 debugfs_create_u64("size_bytes", 0444, dir, &nbd->bytesize);
932 debugfs_create_u32("timeout", 0444, dir, &nbd->tag_set.timeout);
933 debugfs_create_u64("blocksize", 0444, dir, &nbd->blksize);
934 debugfs_create_file("flags", 0444, dir, nbd, &nbd_dbg_flags_ops);
939 static void nbd_dev_dbg_close(struct nbd_device *nbd)
941 debugfs_remove_recursive(nbd->dbg_dir);
944 static int nbd_dbg_init(void)
946 struct dentry *dbg_dir;
948 dbg_dir = debugfs_create_dir("nbd", NULL);
952 nbd_dbg_dir = dbg_dir;
957 static void nbd_dbg_close(void)
959 debugfs_remove_recursive(nbd_dbg_dir);
962 #else /* IS_ENABLED(CONFIG_DEBUG_FS) */
964 static int nbd_dev_dbg_init(struct nbd_device *nbd)
969 static void nbd_dev_dbg_close(struct nbd_device *nbd)
973 static int nbd_dbg_init(void)
978 static void nbd_dbg_close(void)
984 static int nbd_init_request(void *data, struct request *rq,
985 unsigned int hctx_idx, unsigned int request_idx,
986 unsigned int numa_node)
988 struct nbd_cmd *cmd = blk_mq_rq_to_pdu(rq);
993 static struct blk_mq_ops nbd_mq_ops = {
994 .queue_rq = nbd_queue_rq,
995 .init_request = nbd_init_request,
996 .timeout = nbd_xmit_timeout,
999 static void nbd_dev_remove(struct nbd_device *nbd)
1001 struct gendisk *disk = nbd->disk;
1005 blk_cleanup_queue(disk->queue);
1006 blk_mq_free_tag_set(&nbd->tag_set);
1012 static int nbd_dev_add(int index)
1014 struct nbd_device *nbd;
1015 struct gendisk *disk;
1016 struct request_queue *q;
1019 nbd = kzalloc(sizeof(struct nbd_device), GFP_KERNEL);
1023 disk = alloc_disk(1 << part_shift);
1028 err = idr_alloc(&nbd_index_idr, nbd, index, index + 1,
1033 err = idr_alloc(&nbd_index_idr, nbd, 0, 0, GFP_KERNEL);
1041 nbd->tag_set.ops = &nbd_mq_ops;
1042 nbd->tag_set.nr_hw_queues = 1;
1043 nbd->tag_set.queue_depth = 128;
1044 nbd->tag_set.numa_node = NUMA_NO_NODE;
1045 nbd->tag_set.cmd_size = sizeof(struct nbd_cmd);
1046 nbd->tag_set.flags = BLK_MQ_F_SHOULD_MERGE |
1047 BLK_MQ_F_SG_MERGE | BLK_MQ_F_BLOCKING;
1048 nbd->tag_set.driver_data = nbd;
1050 err = blk_mq_alloc_tag_set(&nbd->tag_set);
1054 q = blk_mq_init_queue(&nbd->tag_set);
1062 * Tell the block layer that we are not a rotational device
1064 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
1065 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, disk->queue);
1066 disk->queue->limits.discard_granularity = 512;
1067 blk_queue_max_discard_sectors(disk->queue, UINT_MAX);
1068 disk->queue->limits.discard_zeroes_data = 0;
1069 blk_queue_max_hw_sectors(disk->queue, 65536);
1070 disk->queue->limits.max_sectors = 256;
1072 nbd->magic = NBD_MAGIC;
1073 mutex_init(&nbd->config_lock);
1074 disk->major = NBD_MAJOR;
1075 disk->first_minor = index << part_shift;
1076 disk->fops = &nbd_fops;
1077 disk->private_data = nbd;
1078 sprintf(disk->disk_name, "nbd%d", index);
1079 init_waitqueue_head(&nbd->recv_wq);
1085 blk_mq_free_tag_set(&nbd->tag_set);
1087 idr_remove(&nbd_index_idr, index);
1097 * And here should be modules and kernel interface
1098 * (Just smiley confuses emacs :-)
1101 static int __init nbd_init(void)
1105 BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
1108 printk(KERN_ERR "nbd: max_part must be >= 0\n");
1114 part_shift = fls(max_part);
1117 * Adjust max_part according to part_shift as it is exported
1118 * to user space so that user can know the max number of
1119 * partition kernel should be able to manage.
1121 * Note that -1 is required because partition 0 is reserved
1122 * for the whole disk.
1124 max_part = (1UL << part_shift) - 1;
1127 if ((1UL << part_shift) > DISK_MAX_PARTS)
1130 if (nbds_max > 1UL << (MINORBITS - part_shift))
1132 recv_workqueue = alloc_workqueue("knbd-recv",
1133 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1134 if (!recv_workqueue)
1137 if (register_blkdev(NBD_MAJOR, "nbd"))
1142 mutex_lock(&nbd_index_mutex);
1143 for (i = 0; i < nbds_max; i++)
1145 mutex_unlock(&nbd_index_mutex);
1149 static int nbd_exit_cb(int id, void *ptr, void *data)
1151 struct nbd_device *nbd = ptr;
1152 nbd_dev_remove(nbd);
1156 static void __exit nbd_cleanup(void)
1160 idr_for_each(&nbd_index_idr, &nbd_exit_cb, NULL);
1161 idr_destroy(&nbd_index_idr);
1162 destroy_workqueue(recv_workqueue);
1163 unregister_blkdev(NBD_MAJOR, "nbd");
1166 module_init(nbd_init);
1167 module_exit(nbd_cleanup);
1169 MODULE_DESCRIPTION("Network Block Device");
1170 MODULE_LICENSE("GPL");
1172 module_param(nbds_max, int, 0444);
1173 MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
1174 module_param(max_part, int, 0444);
1175 MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");