splice.c 20.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
/*
 * "splice": joining two ropes together by interweaving their strands.
 *
 * This is the "extended pipe" functionality, where a pipe is used as
 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
 * buffer that you can use to transfer data from one end to the other.
 *
 * The traditional unix read/write is extended with a "splice()" operation
 * that transfers data buffers to or from a pipe buffer.
 *
 * Named by Larry McVoy, original implementation from Linus, extended by
 * Jens to support splicing to files and fixing the initial implementation
 * bugs.
 *
 * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
 * Copyright (C) 2005 Linus Torvalds <torvalds@osdl.org>
 *
 */
#include <linux/fs.h>
#include <linux/file.h>
#include <linux/pagemap.h>
#include <linux/pipe_fs_i.h>
#include <linux/mm_inline.h>
24
#include <linux/swap.h>
25 26
#include <linux/writeback.h>
#include <linux/buffer_head.h>
J
Jeff Garzik 已提交
27
#include <linux/module.h>
28
#include <linux/syscalls.h>
29 30 31 32 33 34 35 36 37 38 39

/*
 * Passed to the actors
 */
struct splice_desc {
	unsigned int len, total_len;	/* current and remaining length */
	unsigned int flags;		/* splice flags */
	struct file *file;		/* file to read/write */
	loff_t pos;			/* file position */
};

40 41 42 43 44 45
/*
 * Attempt to steal a page from a pipe buffer. This should perhaps go into
 * a vm helper function, it's already simplified quite a bit by the
 * addition of remove_mapping(). If success is returned, the caller may
 * attempt to reuse this page for another destination.
 */
46 47 48 49
static int page_cache_pipe_buf_steal(struct pipe_inode_info *info,
				     struct pipe_buffer *buf)
{
	struct page *page = buf->page;
50
	struct address_space *mapping = page_mapping(page);
51 52 53 54

	WARN_ON(!PageLocked(page));
	WARN_ON(!PageUptodate(page));

55 56 57 58 59 60 61 62 63
	/*
	 * At least for ext2 with nobh option, we need to wait on writeback
	 * completing on this page, since we'll remove it from the pagecache.
	 * Otherwise truncate wont wait on the page, allowing the disk
	 * blocks to be reused by someone else before we actually wrote our
	 * data to them. fs corruption ensues.
	 */
	wait_on_page_writeback(page);

64 65 66 67
	if (PagePrivate(page))
		try_to_release_page(page, mapping_gfp_mask(mapping));

	if (!remove_mapping(mapping, page))
68 69
		return 1;

70
	buf->flags |= PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU;
71 72 73
	return 0;
}

74 75 76 77 78
static void page_cache_pipe_buf_release(struct pipe_inode_info *info,
					struct pipe_buffer *buf)
{
	page_cache_release(buf->page);
	buf->page = NULL;
79
	buf->flags &= ~(PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU);
80 81 82 83 84 85 86
}

static void *page_cache_pipe_buf_map(struct file *file,
				     struct pipe_inode_info *info,
				     struct pipe_buffer *buf)
{
	struct page *page = buf->page;
87
	int err;
88 89

	if (!PageUptodate(page)) {
90 91 92 93 94 95 96 97 98 99
		lock_page(page);

		/*
		 * Page got truncated/unhashed. This will cause a 0-byte
		 * splice, if this is the first page
		 */
		if (!page->mapping) {
			err = -ENODATA;
			goto error;
		}
100

101 102 103 104 105 106 107 108 109 110 111
		/*
		 * uh oh, read-error from disk
		 */
		if (!PageUptodate(page)) {
			err = -EIO;
			goto error;
		}

		/*
		 * page is ok afterall, fall through to mapping
		 */
112 113 114
		unlock_page(page);
	}

115 116 117 118
	return kmap(page);
error:
	unlock_page(page);
	return ERR_PTR(err);
119 120 121 122 123 124 125 126 127 128 129 130 131
}

static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info,
				      struct pipe_buffer *buf)
{
	kunmap(buf->page);
}

static struct pipe_buf_operations page_cache_pipe_buf_ops = {
	.can_merge = 0,
	.map = page_cache_pipe_buf_map,
	.unmap = page_cache_pipe_buf_unmap,
	.release = page_cache_pipe_buf_release,
132
	.steal = page_cache_pipe_buf_steal,
133 134
};

135 136 137 138
/*
 * Pipe output worker. This sets up our pipe format with the page cache
 * pipe buffer operations. Otherwise very similar to the regular pipe_writev().
 */
139
static ssize_t move_to_pipe(struct pipe_inode_info *pipe, struct page **pages,
140
			    int nr_pages, unsigned long offset,
141
			    unsigned long len, unsigned int flags)
142 143 144 145 146 147 148
{
	int ret, do_wakeup, i;

	ret = 0;
	do_wakeup = 0;
	i = 0;

149 150
	if (pipe->inode)
		mutex_lock(&pipe->inode->i_mutex);
151 152 153 154

	for (;;) {
		int bufs;

155
		if (!pipe->readers) {
156 157 158 159 160 161
			send_sig(SIGPIPE, current, 0);
			if (!ret)
				ret = -EPIPE;
			break;
		}

162
		bufs = pipe->nrbufs;
163
		if (bufs < PIPE_BUFFERS) {
164 165
			int newbuf = (pipe->curbuf + bufs) & (PIPE_BUFFERS - 1);
			struct pipe_buffer *buf = pipe->bufs + newbuf;
166 167 168 169 170 171 172 173 174 175 176
			struct page *page = pages[i++];
			unsigned long this_len;

			this_len = PAGE_CACHE_SIZE - offset;
			if (this_len > len)
				this_len = len;

			buf->page = page;
			buf->offset = offset;
			buf->len = this_len;
			buf->ops = &page_cache_pipe_buf_ops;
177
			pipe->nrbufs = ++bufs;
178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
			do_wakeup = 1;

			ret += this_len;
			len -= this_len;
			offset = 0;
			if (!--nr_pages)
				break;
			if (!len)
				break;
			if (bufs < PIPE_BUFFERS)
				continue;

			break;
		}

193 194 195 196 197 198
		if (flags & SPLICE_F_NONBLOCK) {
			if (!ret)
				ret = -EAGAIN;
			break;
		}

199 200 201 202 203 204 205
		if (signal_pending(current)) {
			if (!ret)
				ret = -ERESTARTSYS;
			break;
		}

		if (do_wakeup) {
206
			smp_mb();
207 208 209
			if (waitqueue_active(&pipe->wait))
				wake_up_interruptible_sync(&pipe->wait);
			kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
210 211 212
			do_wakeup = 0;
		}

213 214 215
		pipe->waiting_writers++;
		pipe_wait(pipe);
		pipe->waiting_writers--;
216 217
	}

218 219
	if (pipe->inode)
		mutex_unlock(&pipe->inode->i_mutex);
220 221

	if (do_wakeup) {
222
		smp_mb();
223 224 225
		if (waitqueue_active(&pipe->wait))
			wake_up_interruptible(&pipe->wait);
		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
226 227 228 229 230 231 232 233
	}

	while (i < nr_pages)
		page_cache_release(pages[i++]);

	return ret;
}

234 235 236
static int
__generic_file_splice_read(struct file *in, struct pipe_inode_info *pipe,
			   size_t len, unsigned int flags)
237 238 239
{
	struct address_space *mapping = in->f_mapping;
	unsigned int offset, nr_pages;
240
	struct page *pages[PIPE_BUFFERS];
241
	struct page *page;
242 243
	pgoff_t index;
	int i;
244 245 246 247 248 249 250 251 252

	index = in->f_pos >> PAGE_CACHE_SHIFT;
	offset = in->f_pos & ~PAGE_CACHE_MASK;
	nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;

	if (nr_pages > PIPE_BUFFERS)
		nr_pages = PIPE_BUFFERS;

	/*
253 254 255
	 * initiate read-ahead on this page range. however, don't call into
	 * read-ahead if this is a non-zero offset (we are likely doing small
	 * chunk splice and the page is already there) for a single page.
256
	 */
257 258
	if (!offset || nr_pages > 1)
		do_page_cache_readahead(mapping, in, index, nr_pages);
259 260 261 262

	/*
	 * now fill in the holes
	 */
263
	for (i = 0; i < nr_pages; i++, index++) {
264 265 266
		/*
		 * no page there, look one up / create it
		 */
267
		page = find_or_create_page(mapping, index,
268 269 270 271 272 273 274
						   mapping_gfp_mask(mapping));
		if (!page)
			break;

		if (PageUptodate(page))
			unlock_page(page);
		else {
275
			int error = mapping->a_ops->readpage(in, page);
276 277 278 279 280 281

			if (unlikely(error)) {
				page_cache_release(page);
				break;
			}
		}
282
		pages[i] = page;
283 284
	}

285 286
	if (i)
		return move_to_pipe(pipe, pages, i, offset, len, flags);
287

288
	return 0;
289 290
}

291 292 293 294 295 296 297 298 299 300
/**
 * generic_file_splice_read - splice data from file to a pipe
 * @in:		file to splice from
 * @pipe:	pipe to splice to
 * @len:	number of bytes to splice
 * @flags:	splice modifier flags
 *
 * Will read pages from given file and fill them into a pipe.
 *
 */
301
ssize_t generic_file_splice_read(struct file *in, struct pipe_inode_info *pipe,
302 303 304 305 306 307 308
				 size_t len, unsigned int flags)
{
	ssize_t spliced;
	int ret;

	ret = 0;
	spliced = 0;
309

310
	while (len) {
311
		ret = __generic_file_splice_read(in, pipe, len, flags);
312 313 314 315 316 317 318

		if (ret <= 0)
			break;

		in->f_pos += ret;
		len -= ret;
		spliced += ret;
319 320 321 322 323

		if (!(flags & SPLICE_F_NONBLOCK))
			continue;
		ret = -EAGAIN;
		break;
324 325 326 327 328 329 330 331
	}

	if (spliced)
		return spliced;

	return ret;
}

332 333
EXPORT_SYMBOL(generic_file_splice_read);

334
/*
335 336
 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
 * using sendpage().
337 338 339 340 341 342 343 344 345
 */
static int pipe_to_sendpage(struct pipe_inode_info *info,
			    struct pipe_buffer *buf, struct splice_desc *sd)
{
	struct file *file = sd->file;
	loff_t pos = sd->pos;
	unsigned int offset;
	ssize_t ret;
	void *ptr;
346
	int more;
347 348 349 350 351 352 353 354 355 356 357 358

	/*
	 * sub-optimal, but we are limited by the pipe ->map. we don't
	 * need a kmap'ed buffer here, we just want to make sure we
	 * have the page pinned if the pipe page originates from the
	 * page cache
	 */
	ptr = buf->ops->map(file, info, buf);
	if (IS_ERR(ptr))
		return PTR_ERR(ptr);

	offset = pos & ~PAGE_CACHE_MASK;
359
	more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
360

361
	ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more);
362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382

	buf->ops->unmap(info, buf);
	if (ret == sd->len)
		return 0;

	return -EIO;
}

/*
 * This is a little more tricky than the file -> pipe splicing. There are
 * basically three cases:
 *
 *	- Destination page already exists in the address space and there
 *	  are users of it. For that case we have no other option that
 *	  copying the data. Tough luck.
 *	- Destination page already exists in the address space, but there
 *	  are no users of it. Make sure it's uptodate, then drop it. Fall
 *	  through to last case.
 *	- Destination page does not exist, we can add the pipe page to
 *	  the page cache and avoid the copy.
 *
383 384 385 386 387 388
 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
 * sd->flags), we attempt to migrate pages from the pipe to the output
 * file address space page cache. This is possible if no one else has
 * the pipe page referenced outside of the pipe and page cache. If
 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
 * a new page in the output file page cache and fill/dirty that.
389 390 391 392 393 394
 */
static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf,
			struct splice_desc *sd)
{
	struct file *file = sd->file;
	struct address_space *mapping = file->f_mapping;
395
	gfp_t gfp_mask = mapping_gfp_mask(mapping);
396 397 398
	unsigned int offset;
	struct page *page;
	pgoff_t index;
399
	char *src;
400
	int ret;
401 402

	/*
403
	 * make sure the data in this buffer is uptodate
404 405 406 407 408 409 410 411 412
	 */
	src = buf->ops->map(file, info, buf);
	if (IS_ERR(src))
		return PTR_ERR(src);

	index = sd->pos >> PAGE_CACHE_SHIFT;
	offset = sd->pos & ~PAGE_CACHE_MASK;

	/*
413
	 * reuse buf page, if SPLICE_F_MOVE is set
414
	 */
415
	if (sd->flags & SPLICE_F_MOVE) {
416 417 418 419
		/*
		 * If steal succeeds, buf->page is now pruned from the vm
		 * side (LRU and page cache) and we can reuse it.
		 */
420 421 422
		if (buf->ops->steal(info, buf))
			goto find_page;

423 424 425
		/*
		 * this will also set the page locked
		 */
426
		page = buf->page;
427
		if (add_to_page_cache(page, mapping, index, gfp_mask))
428
			goto find_page;
429 430 431

		if (!(buf->flags & PIPE_BUF_FLAG_LRU))
			lru_cache_add(page);
432 433 434
	} else {
find_page:
		ret = -ENOMEM;
435
		page = find_or_create_page(mapping, index, gfp_mask);
436
		if (!page)
437
			goto out_nomem;
438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462

		/*
		 * If the page is uptodate, it is also locked. If it isn't
		 * uptodate, we can mark it uptodate if we are filling the
		 * full page. Otherwise we need to read it in first...
		 */
		if (!PageUptodate(page)) {
			if (sd->len < PAGE_CACHE_SIZE) {
				ret = mapping->a_ops->readpage(file, page);
				if (unlikely(ret))
					goto out;

				lock_page(page);

				if (!PageUptodate(page)) {
					/*
					 * page got invalidated, repeat
					 */
					if (!page->mapping) {
						unlock_page(page);
						page_cache_release(page);
						goto find_page;
					}
					ret = -EIO;
					goto out;
463
				}
464 465 466
			} else {
				WARN_ON(!PageLocked(page));
				SetPageUptodate(page);
467 468 469 470 471
			}
		}
	}

	ret = mapping->a_ops->prepare_write(file, page, 0, sd->len);
472 473 474 475
	if (ret == AOP_TRUNCATED_PAGE) {
		page_cache_release(page);
		goto find_page;
	} else if (ret)
476 477
		goto out;

478
	if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) {
479 480 481 482 483 484
		char *dst = kmap_atomic(page, KM_USER0);

		memcpy(dst + offset, src + buf->offset, sd->len);
		flush_dcache_page(page);
		kunmap_atomic(dst, KM_USER0);
	}
485 486

	ret = mapping->a_ops->commit_write(file, page, 0, sd->len);
487 488 489 490
	if (ret == AOP_TRUNCATED_PAGE) {
		page_cache_release(page);
		goto find_page;
	} else if (ret)
491 492
		goto out;

493
	mark_page_accessed(page);
494
	balance_dirty_pages_ratelimited(mapping);
495
out:
496
	if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) {
497
		page_cache_release(page);
498 499
		unlock_page(page);
	}
500
out_nomem:
501 502 503 504 505 506 507
	buf->ops->unmap(info, buf);
	return ret;
}

typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *,
			   struct splice_desc *);

508 509 510 511 512
/*
 * Pipe input worker. Most of this logic works like a regular pipe, the
 * key here is the 'actor' worker passed in that actually moves the data
 * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
 */
513
static ssize_t move_from_pipe(struct pipe_inode_info *pipe, struct file *out,
514 515 516 517 518 519 520 521 522 523 524 525 526 527
			      size_t len, unsigned int flags,
			      splice_actor *actor)
{
	int ret, do_wakeup, err;
	struct splice_desc sd;

	ret = 0;
	do_wakeup = 0;

	sd.total_len = len;
	sd.flags = flags;
	sd.file = out;
	sd.pos = out->f_pos;

528 529
	if (pipe->inode)
		mutex_lock(&pipe->inode->i_mutex);
530 531

	for (;;) {
532
		int bufs = pipe->nrbufs;
533 534

		if (bufs) {
535 536
			int curbuf = pipe->curbuf;
			struct pipe_buffer *buf = pipe->bufs + curbuf;
537 538 539 540 541 542
			struct pipe_buf_operations *ops = buf->ops;

			sd.len = buf->len;
			if (sd.len > sd.total_len)
				sd.len = sd.total_len;

543
			err = actor(pipe, buf, &sd);
544 545 546 547 548 549 550 551 552 553 554 555
			if (err) {
				if (!ret && err != -ENODATA)
					ret = err;

				break;
			}

			ret += sd.len;
			buf->offset += sd.len;
			buf->len -= sd.len;
			if (!buf->len) {
				buf->ops = NULL;
556
				ops->release(pipe, buf);
557
				curbuf = (curbuf + 1) & (PIPE_BUFFERS - 1);
558 559
				pipe->curbuf = curbuf;
				pipe->nrbufs = --bufs;
560 561 562 563 564 565 566 567 568 569 570
				do_wakeup = 1;
			}

			sd.pos += sd.len;
			sd.total_len -= sd.len;
			if (!sd.total_len)
				break;
		}

		if (bufs)
			continue;
571
		if (!pipe->writers)
572
			break;
573
		if (!pipe->waiting_writers) {
574 575 576 577
			if (ret)
				break;
		}

578 579 580 581 582 583
		if (flags & SPLICE_F_NONBLOCK) {
			if (!ret)
				ret = -EAGAIN;
			break;
		}

584 585 586 587 588 589 590
		if (signal_pending(current)) {
			if (!ret)
				ret = -ERESTARTSYS;
			break;
		}

		if (do_wakeup) {
591
			smp_mb();
592 593 594
			if (waitqueue_active(&pipe->wait))
				wake_up_interruptible_sync(&pipe->wait);
			kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
595 596 597
			do_wakeup = 0;
		}

598
		pipe_wait(pipe);
599 600
	}

601 602
	if (pipe->inode)
		mutex_unlock(&pipe->inode->i_mutex);
603 604

	if (do_wakeup) {
605
		smp_mb();
606 607 608
		if (waitqueue_active(&pipe->wait))
			wake_up_interruptible(&pipe->wait);
		kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
609 610 611 612 613 614 615 616 617
	}

	mutex_lock(&out->f_mapping->host->i_mutex);
	out->f_pos = sd.pos;
	mutex_unlock(&out->f_mapping->host->i_mutex);
	return ret;

}

618 619
/**
 * generic_file_splice_write - splice data from a pipe to a file
620
 * @pipe:	pipe info
621 622 623 624 625 626 627 628
 * @out:	file to write to
 * @len:	number of bytes to splice
 * @flags:	splice modifier flags
 *
 * Will either move or copy pages (determined by @flags options) from
 * the given pipe inode to the given file.
 *
 */
629 630 631
ssize_t
generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
			  size_t len, unsigned int flags)
632
{
633
	struct address_space *mapping = out->f_mapping;
634 635 636
	ssize_t ret;

	ret = move_from_pipe(pipe, out, len, flags, pipe_to_file);
637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655

	/*
	 * if file or inode is SYNC and we actually wrote some data, sync it
	 */
	if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host))
	    && ret > 0) {
		struct inode *inode = mapping->host;
		int err;

		mutex_lock(&inode->i_mutex);
		err = generic_osync_inode(mapping->host, mapping,
						OSYNC_METADATA|OSYNC_DATA);
		mutex_unlock(&inode->i_mutex);

		if (err)
			ret = err;
	}

	return ret;
656 657
}

658 659
EXPORT_SYMBOL(generic_file_splice_write);

660 661 662 663 664 665 666 667 668 669 670
/**
 * generic_splice_sendpage - splice data from a pipe to a socket
 * @inode:	pipe inode
 * @out:	socket to write to
 * @len:	number of bytes to splice
 * @flags:	splice modifier flags
 *
 * Will send @len bytes from the pipe to a network socket. No data copying
 * is involved.
 *
 */
671
ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
672 673
				size_t len, unsigned int flags)
{
674
	return move_from_pipe(pipe, out, len, flags, pipe_to_sendpage);
675 676
}

677
EXPORT_SYMBOL(generic_splice_sendpage);
J
Jeff Garzik 已提交
678

679 680 681
/*
 * Attempt to initiate a splice from pipe to file.
 */
682
static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
683
			   size_t len, unsigned int flags)
684 685 686 687 688 689 690 691 692 693 694
{
	loff_t pos;
	int ret;

	if (!out->f_op || !out->f_op->splice_write)
		return -EINVAL;

	if (!(out->f_mode & FMODE_WRITE))
		return -EBADF;

	pos = out->f_pos;
695

696 697 698 699 700 701 702
	ret = rw_verify_area(WRITE, out, &pos, len);
	if (unlikely(ret < 0))
		return ret;

	return out->f_op->splice_write(pipe, out, len, flags);
}

703 704 705
/*
 * Attempt to initiate a splice from a file to a pipe.
 */
706 707
static long do_splice_to(struct file *in, struct pipe_inode_info *pipe,
			 size_t len, unsigned int flags)
708 709 710 711 712 713 714 715 716 717 718
{
	loff_t pos, isize, left;
	int ret;

	if (!in->f_op || !in->f_op->splice_read)
		return -EINVAL;

	if (!(in->f_mode & FMODE_READ))
		return -EBADF;

	pos = in->f_pos;
719

720 721 722 723 724 725 726 727 728 729 730 731 732 733 734
	ret = rw_verify_area(READ, in, &pos, len);
	if (unlikely(ret < 0))
		return ret;

	isize = i_size_read(in->f_mapping->host);
	if (unlikely(in->f_pos >= isize))
		return 0;
	
	left = isize - in->f_pos;
	if (left < len)
		len = left;

	return in->f_op->splice_read(in, pipe, len, flags);
}

735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846
long do_splice_direct(struct file *in, struct file *out, size_t len,
		      unsigned int flags)
{
	struct pipe_inode_info *pipe;
	long ret, bytes;
	umode_t i_mode;
	int i;

	/*
	 * We require the input being a regular file, as we don't want to
	 * randomly drop data for eg socket -> socket splicing. Use the
	 * piped splicing for that!
	 */
	i_mode = in->f_dentry->d_inode->i_mode;
	if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
		return -EINVAL;

	/*
	 * neither in nor out is a pipe, setup an internal pipe attached to
	 * 'out' and transfer the wanted data from 'in' to 'out' through that
	 */
	pipe = current->splice_pipe;
	if (!pipe) {
		pipe = alloc_pipe_info(NULL);
		if (!pipe)
			return -ENOMEM;

		/*
		 * We don't have an immediate reader, but we'll read the stuff
		 * out of the pipe right after the move_to_pipe(). So set
		 * PIPE_READERS appropriately.
		 */
		pipe->readers = 1;

		current->splice_pipe = pipe;
	}

	/*
	 * do the splice
	 */
	ret = 0;
	bytes = 0;

	while (len) {
		size_t read_len, max_read_len;

		/*
		 * Do at most PIPE_BUFFERS pages worth of transfer:
		 */
		max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE));

		ret = do_splice_to(in, pipe, max_read_len, flags);
		if (unlikely(ret < 0))
			goto out_release;

		read_len = ret;

		/*
		 * NOTE: nonblocking mode only applies to the input. We
		 * must not do the output in nonblocking mode as then we
		 * could get stuck data in the internal pipe:
		 */
		ret = do_splice_from(pipe, out, read_len,
				     flags & ~SPLICE_F_NONBLOCK);
		if (unlikely(ret < 0))
			goto out_release;

		bytes += ret;
		len -= ret;

		/*
		 * In nonblocking mode, if we got back a short read then
		 * that was due to either an IO error or due to the
		 * pagecache entry not being there. In the IO error case
		 * the _next_ splice attempt will produce a clean IO error
		 * return value (not a short read), so in both cases it's
		 * correct to break out of the loop here:
		 */
		if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len))
			break;
	}

	pipe->nrbufs = pipe->curbuf = 0;

	return bytes;

out_release:
	/*
	 * If we did an incomplete transfer we must release
	 * the pipe buffers in question:
	 */
	for (i = 0; i < PIPE_BUFFERS; i++) {
		struct pipe_buffer *buf = pipe->bufs + i;

		if (buf->ops) {
			buf->ops->release(pipe, buf);
			buf->ops = NULL;
		}
	}
	pipe->nrbufs = pipe->curbuf = 0;

	/*
	 * If we transferred some data, return the number of bytes:
	 */
	if (bytes > 0)
		return bytes;

	return ret;
}

EXPORT_SYMBOL(do_splice_direct);

847 848 849
/*
 * Determine where to splice to/from.
 */
850 851 852
static long do_splice(struct file *in, loff_t __user *off_in,
		      struct file *out, loff_t __user *off_out,
		      size_t len, unsigned int flags)
853
{
854
	struct pipe_inode_info *pipe;
855

856
	pipe = in->f_dentry->d_inode->i_pipe;
857 858 859
	if (pipe) {
		if (off_in)
			return -ESPIPE;
860 861 862 863 864 865 866
		if (off_out) {
			if (out->f_op->llseek == no_llseek)
				return -EINVAL;
			if (copy_from_user(&out->f_pos, off_out,
					   sizeof(loff_t)))
				return -EFAULT;
		}
867

868
		return do_splice_from(pipe, out, len, flags);
869
	}
870

871
	pipe = out->f_dentry->d_inode->i_pipe;
872 873 874
	if (pipe) {
		if (off_out)
			return -ESPIPE;
875 876 877 878 879 880
		if (off_in) {
			if (in->f_op->llseek == no_llseek)
				return -EINVAL;
			if (copy_from_user(&in->f_pos, off_in, sizeof(loff_t)))
				return -EFAULT;
		}
881

882
		return do_splice_to(in, pipe, len, flags);
883
	}
884 885 886 887

	return -EINVAL;
}

888 889 890
asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
			   int fd_out, loff_t __user *off_out,
			   size_t len, unsigned int flags)
891 892 893 894 895 896 897 898 899
{
	long error;
	struct file *in, *out;
	int fput_in, fput_out;

	if (unlikely(!len))
		return 0;

	error = -EBADF;
900
	in = fget_light(fd_in, &fput_in);
901 902
	if (in) {
		if (in->f_mode & FMODE_READ) {
903
			out = fget_light(fd_out, &fput_out);
904 905
			if (out) {
				if (out->f_mode & FMODE_WRITE)
906 907 908
					error = do_splice(in, off_in,
							  out, off_out,
							  len, flags);
909 910 911 912 913 914 915 916 917
				fput_light(out, fput_out);
			}
		}

		fput_light(in, fput_in);
	}

	return error;
}