nbd.c 22.3 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6
/*
 * Network block device - make block devices work over TCP
 *
 * Note that you can not swap over this thing, yet. Seems to work but
 * deadlocks sometimes - you can not swap over TCP in general.
 * 
P
Pavel Machek 已提交
7
 * Copyright 1997-2000, 2008 Pavel Machek <pavel@ucw.cz>
L
Linus Torvalds 已提交
8 9
 * Parts copyright 2001 Steven Whitehouse <steve@chygwyn.com>
 *
10
 * This file is released under GPLv2 or later.
L
Linus Torvalds 已提交
11
 *
12
 * (part of code stolen from loop.c)
L
Linus Torvalds 已提交
13 14 15 16 17 18 19 20 21 22 23 24 25 26
 */

#include <linux/major.h>

#include <linux/blkdev.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/sched.h>
#include <linux/fs.h>
#include <linux/bio.h>
#include <linux/stat.h>
#include <linux/errno.h>
#include <linux/file.h>
#include <linux/ioctl.h>
27
#include <linux/mutex.h>
28 29 30
#include <linux/compiler.h>
#include <linux/err.h>
#include <linux/kernel.h>
31
#include <linux/slab.h>
L
Linus Torvalds 已提交
32
#include <net/sock.h>
33
#include <linux/net.h>
34
#include <linux/kthread.h>
L
Linus Torvalds 已提交
35 36 37 38 39 40

#include <asm/uaccess.h>
#include <asm/types.h>

#include <linux/nbd.h>

41
#define NBD_MAGIC 0x68797548
L
Linus Torvalds 已提交
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57

#ifdef NDEBUG
#define dprintk(flags, fmt...)
#else /* NDEBUG */
#define dprintk(flags, fmt...) do { \
	if (debugflags & (flags)) printk(KERN_DEBUG fmt); \
} while (0)
#define DBG_IOCTL       0x0004
#define DBG_INIT        0x0010
#define DBG_EXIT        0x0020
#define DBG_BLKDEV      0x0100
#define DBG_RX          0x0200
#define DBG_TX          0x0400
static unsigned int debugflags;
#endif /* NDEBUG */

58
static unsigned int nbds_max = 16;
59
static struct nbd_device *nbd_dev;
L
Laurent Vivier 已提交
60
static int max_part;
L
Linus Torvalds 已提交
61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80

/*
 * Use just one lock (or at most 1 per NIC). Two arguments for this:
 * 1. Each NIC is essentially a synchronization point for all servers
 *    accessed through that NIC so there's no need to have more locks
 *    than NICs anyway.
 * 2. More locks lead to more "Dirty cache line bouncing" which will slow
 *    down each lock to the point where they're actually slower than just
 *    a single lock.
 * Thanks go to Jens Axboe and Al Viro for their LKML emails explaining this!
 */
static DEFINE_SPINLOCK(nbd_lock);

#ifndef NDEBUG
static const char *ioctl_cmd_to_ascii(int cmd)
{
	switch (cmd) {
	case NBD_SET_SOCK: return "set-sock";
	case NBD_SET_BLKSIZE: return "set-blksize";
	case NBD_SET_SIZE: return "set-size";
P
Paul Clements 已提交
81 82
	case NBD_SET_TIMEOUT: return "set-timeout";
	case NBD_SET_FLAGS: return "set-flags";
L
Linus Torvalds 已提交
83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
	case NBD_DO_IT: return "do-it";
	case NBD_CLEAR_SOCK: return "clear-sock";
	case NBD_CLEAR_QUE: return "clear-que";
	case NBD_PRINT_DEBUG: return "print-debug";
	case NBD_SET_SIZE_BLOCKS: return "set-size-blocks";
	case NBD_DISCONNECT: return "disconnect";
	case BLKROSET: return "set-read-only";
	case BLKFLSBUF: return "flush-buffer-cache";
	}
	return "unknown";
}

static const char *nbdcmd_to_ascii(int cmd)
{
	switch (cmd) {
	case  NBD_CMD_READ: return "read";
	case NBD_CMD_WRITE: return "write";
	case  NBD_CMD_DISC: return "disconnect";
A
Alex Bligh 已提交
101
	case NBD_CMD_FLUSH: return "flush";
P
Paul Clements 已提交
102
	case  NBD_CMD_TRIM: return "trim/discard";
L
Linus Torvalds 已提交
103 104 105 106 107 108 109
	}
	return "invalid";
}
#endif /* NDEBUG */

static void nbd_end_request(struct request *req)
{
110
	int error = req->errors ? -EIO : 0;
111
	struct request_queue *q = req->q;
L
Linus Torvalds 已提交
112 113 114
	unsigned long flags;

	dprintk(DBG_BLKDEV, "%s: request %p: %s\n", req->rq_disk->disk_name,
115
			req, error ? "failed" : "done");
L
Linus Torvalds 已提交
116 117

	spin_lock_irqsave(q->queue_lock, flags);
118
	__blk_end_request_all(req, error);
L
Linus Torvalds 已提交
119 120 121
	spin_unlock_irqrestore(q->queue_lock, flags);
}

122
static void sock_shutdown(struct nbd_device *nbd, int lock)
123 124 125 126 127 128 129 130
{
	/* Forcibly shutdown the socket causing all listeners
	 * to error
	 *
	 * FIXME: This code is duplicated from sys_shutdown, but
	 * there should be a more generic interface rather than
	 * calling socket ops directly here */
	if (lock)
131 132 133 134 135
		mutex_lock(&nbd->tx_lock);
	if (nbd->sock) {
		dev_warn(disk_to_dev(nbd->disk), "shutting down socket\n");
		kernel_sock_shutdown(nbd->sock, SHUT_RDWR);
		nbd->sock = NULL;
136 137
	}
	if (lock)
138
		mutex_unlock(&nbd->tx_lock);
139 140 141 142 143 144 145 146 147 148 149
}

static void nbd_xmit_timeout(unsigned long arg)
{
	struct task_struct *task = (struct task_struct *)arg;

	printk(KERN_WARNING "nbd: killing hung xmit (%s, pid: %d)\n",
		task->comm, task->pid);
	force_sig(SIGKILL, task);
}

L
Linus Torvalds 已提交
150 151 152
/*
 *  Send or receive packet.
 */
153
static int sock_xmit(struct nbd_device *nbd, int send, void *buf, int size,
L
Linus Torvalds 已提交
154 155
		int msg_flags)
{
156
	struct socket *sock = nbd->sock;
L
Linus Torvalds 已提交
157 158 159
	int result;
	struct msghdr msg;
	struct kvec iov;
160
	sigset_t blocked, oldset;
161
	unsigned long pflags = current->flags;
L
Linus Torvalds 已提交
162

163
	if (unlikely(!sock)) {
164
		dev_err(disk_to_dev(nbd->disk),
165 166
			"Attempted %s on closed socket in sock_xmit\n",
			(send ? "send" : "recv"));
167 168 169
		return -EINVAL;
	}

L
Linus Torvalds 已提交
170 171
	/* Allow interception of SIGKILL only
	 * Don't allow other signals to interrupt the transmission */
172 173
	siginitsetinv(&blocked, sigmask(SIGKILL));
	sigprocmask(SIG_SETMASK, &blocked, &oldset);
L
Linus Torvalds 已提交
174

175
	current->flags |= PF_MEMALLOC;
L
Linus Torvalds 已提交
176
	do {
177
		sock->sk->sk_allocation = GFP_NOIO | __GFP_MEMALLOC;
L
Linus Torvalds 已提交
178 179 180 181 182 183 184 185
		iov.iov_base = buf;
		iov.iov_len = size;
		msg.msg_name = NULL;
		msg.msg_namelen = 0;
		msg.msg_control = NULL;
		msg.msg_controllen = 0;
		msg.msg_flags = msg_flags | MSG_NOSIGNAL;

186 187 188
		if (send) {
			struct timer_list ti;

189
			if (nbd->xmit_timeout) {
190 191 192
				init_timer(&ti);
				ti.function = nbd_xmit_timeout;
				ti.data = (unsigned long)current;
193
				ti.expires = jiffies + nbd->xmit_timeout;
194 195
				add_timer(&ti);
			}
L
Linus Torvalds 已提交
196
			result = kernel_sendmsg(sock, &msg, &iov, 1, size);
197
			if (nbd->xmit_timeout)
198 199
				del_timer_sync(&ti);
		} else
200 201
			result = kernel_recvmsg(sock, &msg, &iov, 1, size,
						msg.msg_flags);
L
Linus Torvalds 已提交
202 203 204 205

		if (signal_pending(current)) {
			siginfo_t info;
			printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
206
				task_pid_nr(current), current->comm,
207
				dequeue_signal_lock(current, &current->blocked, &info));
L
Linus Torvalds 已提交
208
			result = -EINTR;
209
			sock_shutdown(nbd, !send);
L
Linus Torvalds 已提交
210 211 212 213 214 215 216 217 218 219 220 221
			break;
		}

		if (result <= 0) {
			if (result == 0)
				result = -EPIPE; /* short read */
			break;
		}
		size -= result;
		buf += result;
	} while (size > 0);

222
	sigprocmask(SIG_SETMASK, &oldset, NULL);
223
	tsk_restore_flags(current, pflags, PF_MEMALLOC);
L
Linus Torvalds 已提交
224 225 226 227

	return result;
}

228
static inline int sock_send_bvec(struct nbd_device *nbd, struct bio_vec *bvec,
L
Linus Torvalds 已提交
229 230 231 232
		int flags)
{
	int result;
	void *kaddr = kmap(bvec->bv_page);
233 234
	result = sock_xmit(nbd, 1, kaddr + bvec->bv_offset,
			   bvec->bv_len, flags);
L
Linus Torvalds 已提交
235 236 237 238
	kunmap(bvec->bv_page);
	return result;
}

239
/* always call with the tx_lock held */
240
static int nbd_send_req(struct nbd_device *nbd, struct request *req)
L
Linus Torvalds 已提交
241
{
242
	int result, flags;
L
Linus Torvalds 已提交
243
	struct nbd_request request;
244
	unsigned long size = blk_rq_bytes(req);
L
Linus Torvalds 已提交
245

246
	memset(&request, 0, sizeof(request));
L
Linus Torvalds 已提交
247 248
	request.magic = htonl(NBD_REQUEST_MAGIC);
	request.type = htonl(nbd_cmd(req));
A
Alex Bligh 已提交
249

250
	if (nbd_cmd(req) != NBD_CMD_FLUSH && nbd_cmd(req) != NBD_CMD_DISC) {
A
Alex Bligh 已提交
251 252 253
		request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
		request.len = htonl(size);
	}
L
Linus Torvalds 已提交
254 255
	memcpy(request.handle, &req, sizeof(req));

256
	dprintk(DBG_TX, "%s: request %p: sending control (%s@%llu,%uB)\n",
257
			nbd->disk->disk_name, req,
L
Linus Torvalds 已提交
258
			nbdcmd_to_ascii(nbd_cmd(req)),
259
			(unsigned long long)blk_rq_pos(req) << 9,
260
			blk_rq_bytes(req));
261
	result = sock_xmit(nbd, 1, &request, sizeof(request),
262
			(nbd_cmd(req) == NBD_CMD_WRITE) ? MSG_MORE : 0);
L
Linus Torvalds 已提交
263
	if (result <= 0) {
264
		dev_err(disk_to_dev(nbd->disk),
265
			"Send control failed (result %d)\n", result);
L
Linus Torvalds 已提交
266 267 268 269
		goto error_out;
	}

	if (nbd_cmd(req) == NBD_CMD_WRITE) {
270
		struct req_iterator iter;
271
		struct bio_vec bvec;
L
Linus Torvalds 已提交
272 273 274 275
		/*
		 * we are really probing at internals to determine
		 * whether to set MSG_MORE or not...
		 */
276
		rq_for_each_segment(bvec, req, iter) {
277
			flags = 0;
K
Kent Overstreet 已提交
278
			if (!rq_iter_last(bvec, iter))
279 280
				flags = MSG_MORE;
			dprintk(DBG_TX, "%s: request %p: sending %d bytes data\n",
281 282
					nbd->disk->disk_name, req, bvec.bv_len);
			result = sock_send_bvec(nbd, &bvec, flags);
283
			if (result <= 0) {
284
				dev_err(disk_to_dev(nbd->disk),
285 286
					"Send data failed (result %d)\n",
					result);
287 288
				goto error_out;
			}
L
Linus Torvalds 已提交
289 290 291 292 293
		}
	}
	return 0;

error_out:
P
Pavel Machek 已提交
294
	return -EIO;
L
Linus Torvalds 已提交
295 296
}

297
static struct request *nbd_find_request(struct nbd_device *nbd,
298
					struct request *xreq)
L
Linus Torvalds 已提交
299
{
300
	struct request *req, *tmp;
301
	int err;
L
Linus Torvalds 已提交
302

303
	err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq);
304 305 306
	if (unlikely(err))
		goto out;

307 308
	spin_lock(&nbd->queue_lock);
	list_for_each_entry_safe(req, tmp, &nbd->queue_head, queuelist) {
L
Linus Torvalds 已提交
309 310 311
		if (req != xreq)
			continue;
		list_del_init(&req->queuelist);
312
		spin_unlock(&nbd->queue_lock);
L
Linus Torvalds 已提交
313 314
		return req;
	}
315
	spin_unlock(&nbd->queue_lock);
316 317 318 319 320

	err = -ENOENT;

out:
	return ERR_PTR(err);
L
Linus Torvalds 已提交
321 322
}

323
static inline int sock_recv_bvec(struct nbd_device *nbd, struct bio_vec *bvec)
L
Linus Torvalds 已提交
324 325 326
{
	int result;
	void *kaddr = kmap(bvec->bv_page);
327
	result = sock_xmit(nbd, 0, kaddr + bvec->bv_offset, bvec->bv_len,
L
Linus Torvalds 已提交
328 329 330 331 332 333
			MSG_WAITALL);
	kunmap(bvec->bv_page);
	return result;
}

/* NULL returned = something went wrong, inform userspace */
334
static struct request *nbd_read_stat(struct nbd_device *nbd)
L
Linus Torvalds 已提交
335 336 337 338 339 340
{
	int result;
	struct nbd_reply reply;
	struct request *req;

	reply.magic = 0;
341
	result = sock_xmit(nbd, 0, &reply, sizeof(reply), MSG_WAITALL);
L
Linus Torvalds 已提交
342
	if (result <= 0) {
343
		dev_err(disk_to_dev(nbd->disk),
344
			"Receive control failed (result %d)\n", result);
L
Linus Torvalds 已提交
345 346
		goto harderror;
	}
347 348

	if (ntohl(reply.magic) != NBD_REPLY_MAGIC) {
349
		dev_err(disk_to_dev(nbd->disk), "Wrong magic (0x%lx)\n",
350 351 352 353 354
				(unsigned long)ntohl(reply.magic));
		result = -EPROTO;
		goto harderror;
	}

355
	req = nbd_find_request(nbd, *(struct request **)reply.handle);
356
	if (IS_ERR(req)) {
357 358 359 360
		result = PTR_ERR(req);
		if (result != -ENOENT)
			goto harderror;

361
		dev_err(disk_to_dev(nbd->disk), "Unexpected reply (%p)\n",
362
			reply.handle);
L
Linus Torvalds 已提交
363 364 365 366 367
		result = -EBADR;
		goto harderror;
	}

	if (ntohl(reply.error)) {
368
		dev_err(disk_to_dev(nbd->disk), "Other side returned error (%d)\n",
369
			ntohl(reply.error));
L
Linus Torvalds 已提交
370 371 372 373 374
		req->errors++;
		return req;
	}

	dprintk(DBG_RX, "%s: request %p: got reply\n",
375
			nbd->disk->disk_name, req);
L
Linus Torvalds 已提交
376
	if (nbd_cmd(req) == NBD_CMD_READ) {
377
		struct req_iterator iter;
378
		struct bio_vec bvec;
379 380

		rq_for_each_segment(bvec, req, iter) {
381
			result = sock_recv_bvec(nbd, &bvec);
382
			if (result <= 0) {
383
				dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
384
					result);
385 386 387 388
				req->errors++;
				return req;
			}
			dprintk(DBG_RX, "%s: request %p: got %d bytes data\n",
389
				nbd->disk->disk_name, req, bvec.bv_len);
L
Linus Torvalds 已提交
390 391 392 393
		}
	}
	return req;
harderror:
394
	nbd->harderror = result;
L
Linus Torvalds 已提交
395 396 397
	return NULL;
}

398 399
static ssize_t pid_show(struct device *dev,
			struct device_attribute *attr, char *buf)
400
{
401 402 403
	struct gendisk *disk = dev_to_disk(dev);

	return sprintf(buf, "%ld\n",
404 405 406
		(long) ((struct nbd_device *)disk->private_data)->pid);
}

407
static struct device_attribute pid_attr = {
408
	.attr = { .name = "pid", .mode = S_IRUGO},
409 410 411
	.show = pid_show,
};

412
static int nbd_do_it(struct nbd_device *nbd)
L
Linus Torvalds 已提交
413 414
{
	struct request *req;
415
	int ret;
L
Linus Torvalds 已提交
416

417
	BUG_ON(nbd->magic != NBD_MAGIC);
L
Linus Torvalds 已提交
418

419
	sk_set_memalloc(nbd->sock->sk);
420 421
	nbd->pid = task_pid_nr(current);
	ret = device_create_file(disk_to_dev(nbd->disk), &pid_attr);
422
	if (ret) {
423 424
		dev_err(disk_to_dev(nbd->disk), "device_create_file failed!\n");
		nbd->pid = 0;
425 426
		return ret;
	}
427

428
	while ((req = nbd_read_stat(nbd)) != NULL)
L
Linus Torvalds 已提交
429
		nbd_end_request(req);
430

431 432
	device_remove_file(disk_to_dev(nbd->disk), &pid_attr);
	nbd->pid = 0;
433
	return 0;
L
Linus Torvalds 已提交
434 435
}

436
static void nbd_clear_que(struct nbd_device *nbd)
L
Linus Torvalds 已提交
437 438 439
{
	struct request *req;

440
	BUG_ON(nbd->magic != NBD_MAGIC);
L
Linus Torvalds 已提交
441

442
	/*
443
	 * Because we have set nbd->sock to NULL under the tx_lock, all
444 445 446 447 448 449
	 * modifications to the list must have completed by now.  For
	 * the same reason, the active_req must be NULL.
	 *
	 * As a consequence, we don't need to take the spin lock while
	 * purging the list here.
	 */
450 451
	BUG_ON(nbd->sock);
	BUG_ON(nbd->active_req);
452

453 454
	while (!list_empty(&nbd->queue_head)) {
		req = list_entry(nbd->queue_head.next, struct request,
455 456 457 458 459
				 queuelist);
		list_del_init(&req->queuelist);
		req->errors++;
		nbd_end_request(req);
	}
460 461 462 463 464 465 466 467

	while (!list_empty(&nbd->waiting_queue)) {
		req = list_entry(nbd->waiting_queue.next, struct request,
				 queuelist);
		list_del_init(&req->queuelist);
		req->errors++;
		nbd_end_request(req);
	}
L
Linus Torvalds 已提交
468 469
}

470

471
static void nbd_handle_req(struct nbd_device *nbd, struct request *req)
472
{
473
	if (req->cmd_type != REQ_TYPE_FS)
474 475 476 477
		goto error_out;

	nbd_cmd(req) = NBD_CMD_READ;
	if (rq_data_dir(req) == WRITE) {
P
Paul Clements 已提交
478 479 480 481 482
		if ((req->cmd_flags & REQ_DISCARD)) {
			WARN_ON(!(nbd->flags & NBD_FLAG_SEND_TRIM));
			nbd_cmd(req) = NBD_CMD_TRIM;
		} else
			nbd_cmd(req) = NBD_CMD_WRITE;
P
Paul Clements 已提交
483
		if (nbd->flags & NBD_FLAG_READ_ONLY) {
484
			dev_err(disk_to_dev(nbd->disk),
485
				"Write on read-only\n");
486 487 488 489
			goto error_out;
		}
	}

A
Alex Bligh 已提交
490 491 492 493 494
	if (req->cmd_flags & REQ_FLUSH) {
		BUG_ON(unlikely(blk_rq_sectors(req)));
		nbd_cmd(req) = NBD_CMD_FLUSH;
	}

495 496
	req->errors = 0;

497 498 499 500
	mutex_lock(&nbd->tx_lock);
	if (unlikely(!nbd->sock)) {
		mutex_unlock(&nbd->tx_lock);
		dev_err(disk_to_dev(nbd->disk),
501
			"Attempted send on closed socket\n");
P
Pavel Machek 已提交
502
		goto error_out;
503 504
	}

505
	nbd->active_req = req;
506

507 508
	if (nbd_send_req(nbd, req) != 0) {
		dev_err(disk_to_dev(nbd->disk), "Request send failed\n");
509 510 511
		req->errors++;
		nbd_end_request(req);
	} else {
512
		spin_lock(&nbd->queue_lock);
513
		list_add_tail(&req->queuelist, &nbd->queue_head);
514
		spin_unlock(&nbd->queue_lock);
515 516
	}

517 518 519
	nbd->active_req = NULL;
	mutex_unlock(&nbd->tx_lock);
	wake_up_all(&nbd->active_wq);
520 521 522 523 524 525 526 527 528 529

	return;

error_out:
	req->errors++;
	nbd_end_request(req);
}

static int nbd_thread(void *data)
{
530
	struct nbd_device *nbd = data;
531 532
	struct request *req;

533
	set_user_nice(current, MIN_NICE);
534
	while (!kthread_should_stop() || !list_empty(&nbd->waiting_queue)) {
535
		/* wait for something to do */
536
		wait_event_interruptible(nbd->waiting_wq,
537
					 kthread_should_stop() ||
538
					 !list_empty(&nbd->waiting_queue));
539 540

		/* extract request */
541
		if (list_empty(&nbd->waiting_queue))
542 543
			continue;

544 545
		spin_lock_irq(&nbd->queue_lock);
		req = list_entry(nbd->waiting_queue.next, struct request,
546 547
				 queuelist);
		list_del_init(&req->queuelist);
548
		spin_unlock_irq(&nbd->queue_lock);
549 550

		/* handle request */
551
		nbd_handle_req(nbd, req);
552 553 554 555
	}
	return 0;
}

L
Linus Torvalds 已提交
556 557 558
/*
 * We always wait for result of write, for now. It would be nice to make it optional
 * in future
559
 * if ((rq_data_dir(req) == WRITE) && (nbd->flags & NBD_WRITE_NOCHK))
L
Linus Torvalds 已提交
560 561 562
 *   { printk( "Warning: Ignoring result!\n"); nbd_end_request( req ); }
 */

P
Pavel Machek 已提交
563
static void do_nbd_request(struct request_queue *q)
A
Alex Elder 已提交
564
		__releases(q->queue_lock) __acquires(q->queue_lock)
L
Linus Torvalds 已提交
565 566 567
{
	struct request *req;
	
568
	while ((req = blk_fetch_request(q)) != NULL) {
569
		struct nbd_device *nbd;
L
Linus Torvalds 已提交
570

571 572
		spin_unlock_irq(q->queue_lock);

573 574
		dprintk(DBG_BLKDEV, "%s: request %p: dequeued (flags=%x)\n",
				req->rq_disk->disk_name, req, req->cmd_type);
L
Linus Torvalds 已提交
575

576
		nbd = req->rq_disk->private_data;
L
Linus Torvalds 已提交
577

578
		BUG_ON(nbd->magic != NBD_MAGIC);
L
Linus Torvalds 已提交
579

580 581
		if (unlikely(!nbd->sock)) {
			dev_err(disk_to_dev(nbd->disk),
582
				"Attempted send on closed socket\n");
583 584 585 586 587 588
			req->errors++;
			nbd_end_request(req);
			spin_lock_irq(q->queue_lock);
			continue;
		}

589 590 591
		spin_lock_irq(&nbd->queue_lock);
		list_add_tail(&req->queuelist, &nbd->waiting_queue);
		spin_unlock_irq(&nbd->queue_lock);
L
Linus Torvalds 已提交
592

593
		wake_up(&nbd->waiting_wq);
594

L
Linus Torvalds 已提交
595 596 597 598
		spin_lock_irq(q->queue_lock);
	}
}

P
Pavel Machek 已提交
599
/* Must be called with tx_lock held */
L
Linus Torvalds 已提交
600

601
static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
P
Pavel Machek 已提交
602 603
		       unsigned int cmd, unsigned long arg)
{
L
Linus Torvalds 已提交
604
	switch (cmd) {
P
Pavel Machek 已提交
605 606 607
	case NBD_DISCONNECT: {
		struct request sreq;

608
		dev_info(disk_to_dev(nbd->disk), "NBD_DISCONNECT\n");
609 610
		if (!nbd->sock)
			return -EINVAL;
P
Pavel Machek 已提交
611

612 613 614
		mutex_unlock(&nbd->tx_lock);
		fsync_bdev(bdev);
		mutex_lock(&nbd->tx_lock);
615
		blk_rq_init(NULL, &sreq);
616
		sreq.cmd_type = REQ_TYPE_SPECIAL;
L
Linus Torvalds 已提交
617
		nbd_cmd(&sreq) = NBD_CMD_DISC;
618 619

		/* Check again after getting mutex back.  */
620
		if (!nbd->sock)
L
Linus Torvalds 已提交
621
			return -EINVAL;
622

P
Paul Clements 已提交
623 624
		nbd->disconnect = 1;

625
		nbd_send_req(nbd, &sreq);
P
Paul Clements 已提交
626
		return 0;
P
Pavel Machek 已提交
627
	}
L
Linus Torvalds 已提交
628
 
P
Pavel Machek 已提交
629
	case NBD_CLEAR_SOCK: {
A
Al Viro 已提交
630
		struct socket *sock = nbd->sock;
631 632 633
		nbd->sock = NULL;
		nbd_clear_que(nbd);
		BUG_ON(!list_empty(&nbd->queue_head));
634
		BUG_ON(!list_empty(&nbd->waiting_queue));
635
		kill_bdev(bdev);
A
Al Viro 已提交
636 637
		if (sock)
			sockfd_put(sock);
P
Pavel Machek 已提交
638 639 640 641
		return 0;
	}

	case NBD_SET_SOCK: {
A
Al Viro 已提交
642 643 644
		struct socket *sock;
		int err;
		if (nbd->sock)
L
Linus Torvalds 已提交
645
			return -EBUSY;
A
Al Viro 已提交
646 647 648 649 650 651 652
		sock = sockfd_lookup(arg, &err);
		if (sock) {
			nbd->sock = sock;
			if (max_part > 0)
				bdev->bd_invalidated = 1;
			nbd->disconnect = 0; /* we're connected now */
			return 0;
L
Linus Torvalds 已提交
653
		}
P
Pavel Machek 已提交
654 655 656
		return -EINVAL;
	}

L
Linus Torvalds 已提交
657
	case NBD_SET_BLKSIZE:
658 659 660 661 662
		nbd->blksize = arg;
		nbd->bytesize &= ~(nbd->blksize-1);
		bdev->bd_inode->i_size = nbd->bytesize;
		set_blocksize(bdev, nbd->blksize);
		set_capacity(nbd->disk, nbd->bytesize >> 9);
L
Linus Torvalds 已提交
663
		return 0;
P
Pavel Machek 已提交
664

L
Linus Torvalds 已提交
665
	case NBD_SET_SIZE:
666 667 668 669
		nbd->bytesize = arg & ~(nbd->blksize-1);
		bdev->bd_inode->i_size = nbd->bytesize;
		set_blocksize(bdev, nbd->blksize);
		set_capacity(nbd->disk, nbd->bytesize >> 9);
L
Linus Torvalds 已提交
670
		return 0;
P
Pavel Machek 已提交
671

672
	case NBD_SET_TIMEOUT:
673
		nbd->xmit_timeout = arg * HZ;
674
		return 0;
P
Pavel Machek 已提交
675

P
Paul Clements 已提交
676 677 678 679
	case NBD_SET_FLAGS:
		nbd->flags = arg;
		return 0;

L
Linus Torvalds 已提交
680
	case NBD_SET_SIZE_BLOCKS:
681 682 683 684
		nbd->bytesize = ((u64) arg) * nbd->blksize;
		bdev->bd_inode->i_size = nbd->bytesize;
		set_blocksize(bdev, nbd->blksize);
		set_capacity(nbd->disk, nbd->bytesize >> 9);
L
Linus Torvalds 已提交
685
		return 0;
P
Pavel Machek 已提交
686 687 688

	case NBD_DO_IT: {
		struct task_struct *thread;
A
Al Viro 已提交
689
		struct socket *sock;
P
Pavel Machek 已提交
690 691
		int error;

692
		if (nbd->pid)
693
			return -EBUSY;
A
Al Viro 已提交
694
		if (!nbd->sock)
L
Linus Torvalds 已提交
695
			return -EINVAL;
P
Pavel Machek 已提交
696

697
		mutex_unlock(&nbd->tx_lock);
P
Pavel Machek 已提交
698

699 700
		if (nbd->flags & NBD_FLAG_READ_ONLY)
			set_device_ro(bdev, true);
P
Paul Clements 已提交
701 702 703
		if (nbd->flags & NBD_FLAG_SEND_TRIM)
			queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
				nbd->disk->queue);
A
Alex Bligh 已提交
704 705 706 707
		if (nbd->flags & NBD_FLAG_SEND_FLUSH)
			blk_queue_flush(nbd->disk->queue, REQ_FLUSH);
		else
			blk_queue_flush(nbd->disk->queue, 0);
P
Paul Clements 已提交
708

709 710
		thread = kthread_create(nbd_thread, nbd, "%s",
					nbd->disk->disk_name);
P
Pavel Machek 已提交
711
		if (IS_ERR(thread)) {
712
			mutex_lock(&nbd->tx_lock);
713
			return PTR_ERR(thread);
P
Pavel Machek 已提交
714
		}
715
		wake_up_process(thread);
716
		error = nbd_do_it(nbd);
717
		kthread_stop(thread);
P
Pavel Machek 已提交
718

719
		mutex_lock(&nbd->tx_lock);
720 721
		if (error)
			return error;
722
		sock_shutdown(nbd, 0);
A
Al Viro 已提交
723 724
		sock = nbd->sock;
		nbd->sock = NULL;
725 726
		nbd_clear_que(nbd);
		dev_warn(disk_to_dev(nbd->disk), "queue cleared\n");
727
		kill_bdev(bdev);
P
Paul Clements 已提交
728
		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, nbd->disk->queue);
729
		set_device_ro(bdev, false);
A
Al Viro 已提交
730 731
		if (sock)
			sockfd_put(sock);
A
Alex Bligh 已提交
732
		nbd->flags = 0;
733
		nbd->bytesize = 0;
A
Al Viro 已提交
734
		bdev->bd_inode->i_size = 0;
735
		set_capacity(nbd->disk, 0);
L
Laurent Vivier 已提交
736
		if (max_part > 0)
A
Al Viro 已提交
737
			ioctl_by_bdev(bdev, BLKRRPART, 0);
P
Paul Clements 已提交
738 739
		if (nbd->disconnect) /* user requested, ignore socket errors */
			return 0;
740
		return nbd->harderror;
P
Pavel Machek 已提交
741 742
	}

L
Linus Torvalds 已提交
743
	case NBD_CLEAR_QUE:
744 745 746 747
		/*
		 * This is for compatibility only.  The queue is always cleared
		 * by NBD_DO_IT or NBD_CLEAR_SOCK.
		 */
L
Linus Torvalds 已提交
748
		return 0;
P
Pavel Machek 已提交
749

L
Linus Torvalds 已提交
750
	case NBD_PRINT_DEBUG:
751
		dev_info(disk_to_dev(nbd->disk),
752
			"next = %p, prev = %p, head = %p\n",
753 754
			nbd->queue_head.next, nbd->queue_head.prev,
			&nbd->queue_head);
L
Linus Torvalds 已提交
755 756
		return 0;
	}
P
Pavel Machek 已提交
757 758 759 760 761 762
	return -ENOTTY;
}

static int nbd_ioctl(struct block_device *bdev, fmode_t mode,
		     unsigned int cmd, unsigned long arg)
{
763
	struct nbd_device *nbd = bdev->bd_disk->private_data;
P
Pavel Machek 已提交
764 765 766 767 768
	int error;

	if (!capable(CAP_SYS_ADMIN))
		return -EPERM;

769
	BUG_ON(nbd->magic != NBD_MAGIC);
P
Pavel Machek 已提交
770 771 772

	/* Anyone capable of this syscall can do *real bad* things */
	dprintk(DBG_IOCTL, "%s: nbd_ioctl cmd=%s(0x%x) arg=%lu\n",
773
		nbd->disk->disk_name, ioctl_cmd_to_ascii(cmd), cmd, arg);
P
Pavel Machek 已提交
774

775 776 777
	mutex_lock(&nbd->tx_lock);
	error = __nbd_ioctl(bdev, nbd, cmd, arg);
	mutex_unlock(&nbd->tx_lock);
P
Pavel Machek 已提交
778 779

	return error;
L
Linus Torvalds 已提交
780 781
}

782
static const struct block_device_operations nbd_fops =
L
Linus Torvalds 已提交
783 784
{
	.owner =	THIS_MODULE,
785
	.ioctl =	nbd_ioctl,
L
Linus Torvalds 已提交
786 787 788 789 790 791 792 793 794 795 796
};

/*
 * And here should be modules and kernel interface 
 *  (Just smiley confuses emacs :-)
 */

static int __init nbd_init(void)
{
	int err = -ENOMEM;
	int i;
L
Laurent Vivier 已提交
797
	int part_shift;
L
Linus Torvalds 已提交
798

799
	BUILD_BUG_ON(sizeof(struct nbd_request) != 28);
L
Linus Torvalds 已提交
800

L
Laurent Vivier 已提交
801
	if (max_part < 0) {
802
		printk(KERN_ERR "nbd: max_part must be >= 0\n");
L
Laurent Vivier 已提交
803 804 805
		return -EINVAL;
	}

806 807 808 809
	nbd_dev = kcalloc(nbds_max, sizeof(*nbd_dev), GFP_KERNEL);
	if (!nbd_dev)
		return -ENOMEM;

L
Laurent Vivier 已提交
810
	part_shift = 0;
811
	if (max_part > 0) {
L
Laurent Vivier 已提交
812 813
		part_shift = fls(max_part);

814 815 816 817 818 819 820 821 822 823 824
		/*
		 * Adjust max_part according to part_shift as it is exported
		 * to user space so that user can know the max number of
		 * partition kernel should be able to manage.
		 *
		 * Note that -1 is required because partition 0 is reserved
		 * for the whole disk.
		 */
		max_part = (1UL << part_shift) - 1;
	}

825 826 827 828 829 830
	if ((1UL << part_shift) > DISK_MAX_PARTS)
		return -EINVAL;

	if (nbds_max > 1UL << (MINORBITS - part_shift))
		return -EINVAL;

831
	for (i = 0; i < nbds_max; i++) {
L
Laurent Vivier 已提交
832
		struct gendisk *disk = alloc_disk(1 << part_shift);
L
Linus Torvalds 已提交
833 834 835 836 837 838 839 840 841 842 843 844 845
		if (!disk)
			goto out;
		nbd_dev[i].disk = disk;
		/*
		 * The new linux 2.5 block layer implementation requires
		 * every gendisk to have its very own request_queue struct.
		 * These structs are big so we dynamically allocate them.
		 */
		disk->queue = blk_init_queue(do_nbd_request, &nbd_lock);
		if (!disk->queue) {
			put_disk(disk);
			goto out;
		}
846 847 848 849
		/*
		 * Tell the block layer that we are not a rotational device
		 */
		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, disk->queue);
P
Paul Clements 已提交
850 851 852
		disk->queue->limits.discard_granularity = 512;
		disk->queue->limits.max_discard_sectors = UINT_MAX;
		disk->queue->limits.discard_zeroes_data = 0;
853 854
		blk_queue_max_hw_sectors(disk->queue, 65536);
		disk->queue->limits.max_sectors = 256;
L
Linus Torvalds 已提交
855 856 857 858 859 860 861 862 863 864
	}

	if (register_blkdev(NBD_MAJOR, "nbd")) {
		err = -EIO;
		goto out;
	}

	printk(KERN_INFO "nbd: registered device at major %d\n", NBD_MAJOR);
	dprintk(DBG_INIT, "nbd: debugflags=0x%x\n", debugflags);

865
	for (i = 0; i < nbds_max; i++) {
L
Linus Torvalds 已提交
866
		struct gendisk *disk = nbd_dev[i].disk;
867
		nbd_dev[i].magic = NBD_MAGIC;
868
		INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
L
Linus Torvalds 已提交
869 870
		spin_lock_init(&nbd_dev[i].queue_lock);
		INIT_LIST_HEAD(&nbd_dev[i].queue_head);
871
		mutex_init(&nbd_dev[i].tx_lock);
872
		init_waitqueue_head(&nbd_dev[i].active_wq);
873
		init_waitqueue_head(&nbd_dev[i].waiting_wq);
L
Linus Torvalds 已提交
874
		nbd_dev[i].blksize = 1024;
875
		nbd_dev[i].bytesize = 0;
L
Linus Torvalds 已提交
876
		disk->major = NBD_MAJOR;
L
Laurent Vivier 已提交
877
		disk->first_minor = i << part_shift;
L
Linus Torvalds 已提交
878 879 880
		disk->fops = &nbd_fops;
		disk->private_data = &nbd_dev[i];
		sprintf(disk->disk_name, "nbd%d", i);
881
		set_capacity(disk, 0);
L
Linus Torvalds 已提交
882 883 884 885 886 887 888 889 890
		add_disk(disk);
	}

	return 0;
out:
	while (i--) {
		blk_cleanup_queue(nbd_dev[i].disk->queue);
		put_disk(nbd_dev[i].disk);
	}
891
	kfree(nbd_dev);
L
Linus Torvalds 已提交
892 893 894 895 896 897
	return err;
}

static void __exit nbd_cleanup(void)
{
	int i;
898
	for (i = 0; i < nbds_max; i++) {
L
Linus Torvalds 已提交
899
		struct gendisk *disk = nbd_dev[i].disk;
900
		nbd_dev[i].magic = 0;
L
Linus Torvalds 已提交
901 902 903 904 905 906 907
		if (disk) {
			del_gendisk(disk);
			blk_cleanup_queue(disk->queue);
			put_disk(disk);
		}
	}
	unregister_blkdev(NBD_MAJOR, "nbd");
908
	kfree(nbd_dev);
L
Linus Torvalds 已提交
909 910 911 912 913 914 915 916 917
	printk(KERN_INFO "nbd: unregistered device at major %d\n", NBD_MAJOR);
}

module_init(nbd_init);
module_exit(nbd_cleanup);

MODULE_DESCRIPTION("Network Block Device");
MODULE_LICENSE("GPL");

918
module_param(nbds_max, int, 0444);
L
Laurent Vivier 已提交
919 920 921
MODULE_PARM_DESC(nbds_max, "number of network block devices to initialize (default: 16)");
module_param(max_part, int, 0444);
MODULE_PARM_DESC(max_part, "number of partitions per device (default: 0)");
L
Linus Torvalds 已提交
922 923 924 925
#ifndef NDEBUG
module_param(debugflags, int, 0644);
MODULE_PARM_DESC(debugflags, "flags for controlling debug output");
#endif