]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/splice.c
[PATCH] splice: add direct fd <-> fd splicing support
[net-next-2.6.git] / fs / splice.c
CommitLineData
5274f052
JA
1/*
2 * "splice": joining two ropes together by interweaving their strands.
3 *
4 * This is the "extended pipe" functionality, where a pipe is used as
5 * an arbitrary in-memory buffer. Think of a pipe as a small kernel
6 * buffer that you can use to transfer data from one end to the other.
7 *
8 * The traditional unix read/write is extended with a "splice()" operation
9 * that transfers data buffers to or from a pipe buffer.
10 *
11 * Named by Larry McVoy, original implementation from Linus, extended by
12 * Jens to support splicing to files and fixing the initial implementation
13 * bugs.
14 *
15 * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
16 * Copyright (C) 2005 Linus Torvalds <torvalds@osdl.org>
17 *
18 */
19#include <linux/fs.h>
20#include <linux/file.h>
21#include <linux/pagemap.h>
22#include <linux/pipe_fs_i.h>
23#include <linux/mm_inline.h>
5abc97aa 24#include <linux/swap.h>
4f6f0bd2
JA
25#include <linux/writeback.h>
26#include <linux/buffer_head.h>
a0f06780 27#include <linux/module.h>
4f6f0bd2 28#include <linux/syscalls.h>
5274f052
JA
29
30/*
31 * Passed to the actors
32 */
33struct splice_desc {
34 unsigned int len, total_len; /* current and remaining length */
35 unsigned int flags; /* splice flags */
36 struct file *file; /* file to read/write */
37 loff_t pos; /* file position */
38};
39
83f9135b
JA
40/*
41 * Attempt to steal a page from a pipe buffer. This should perhaps go into
42 * a vm helper function, it's already simplified quite a bit by the
43 * addition of remove_mapping(). If success is returned, the caller may
44 * attempt to reuse this page for another destination.
45 */
5abc97aa
JA
46static int page_cache_pipe_buf_steal(struct pipe_inode_info *info,
47 struct pipe_buffer *buf)
48{
49 struct page *page = buf->page;
4f6f0bd2 50 struct address_space *mapping = page_mapping(page);
5abc97aa
JA
51
52 WARN_ON(!PageLocked(page));
53 WARN_ON(!PageUptodate(page));
54
ad8d6f0a
JA
55 /*
56 * At least for ext2 with nobh option, we need to wait on writeback
57 * completing on this page, since we'll remove it from the pagecache.
58 * Otherwise truncate wont wait on the page, allowing the disk
59 * blocks to be reused by someone else before we actually wrote our
60 * data to them. fs corruption ensues.
61 */
62 wait_on_page_writeback(page);
63
4f6f0bd2
JA
64 if (PagePrivate(page))
65 try_to_release_page(page, mapping_gfp_mask(mapping));
66
67 if (!remove_mapping(mapping, page))
5abc97aa
JA
68 return 1;
69
3e7ee3e7 70 buf->flags |= PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU;
5abc97aa
JA
71 return 0;
72}
73
5274f052
JA
74static void page_cache_pipe_buf_release(struct pipe_inode_info *info,
75 struct pipe_buffer *buf)
76{
77 page_cache_release(buf->page);
78 buf->page = NULL;
3e7ee3e7 79 buf->flags &= ~(PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU);
5274f052
JA
80}
81
82static void *page_cache_pipe_buf_map(struct file *file,
83 struct pipe_inode_info *info,
84 struct pipe_buffer *buf)
85{
86 struct page *page = buf->page;
49d0b21b 87 int err;
5274f052
JA
88
89 if (!PageUptodate(page)) {
49d0b21b
JA
90 lock_page(page);
91
92 /*
93 * Page got truncated/unhashed. This will cause a 0-byte
94 * splice, if this is the first page
95 */
96 if (!page->mapping) {
97 err = -ENODATA;
98 goto error;
99 }
5274f052 100
49d0b21b
JA
101 /*
102 * uh oh, read-error from disk
103 */
104 if (!PageUptodate(page)) {
105 err = -EIO;
106 goto error;
107 }
108
109 /*
110 * page is ok afterall, fall through to mapping
111 */
5274f052 112 unlock_page(page);
5274f052
JA
113 }
114
49d0b21b
JA
115 return kmap(page);
116error:
117 unlock_page(page);
118 return ERR_PTR(err);
5274f052
JA
119}
120
121static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info,
122 struct pipe_buffer *buf)
123{
5274f052
JA
124 kunmap(buf->page);
125}
126
127static struct pipe_buf_operations page_cache_pipe_buf_ops = {
128 .can_merge = 0,
129 .map = page_cache_pipe_buf_map,
130 .unmap = page_cache_pipe_buf_unmap,
131 .release = page_cache_pipe_buf_release,
5abc97aa 132 .steal = page_cache_pipe_buf_steal,
5274f052
JA
133};
134
83f9135b
JA
135/*
136 * Pipe output worker. This sets up our pipe format with the page cache
137 * pipe buffer operations. Otherwise very similar to the regular pipe_writev().
138 */
3a326a2c 139static ssize_t move_to_pipe(struct pipe_inode_info *pipe, struct page **pages,
5274f052 140 int nr_pages, unsigned long offset,
29e35094 141 unsigned long len, unsigned int flags)
5274f052 142{
5274f052
JA
143 int ret, do_wakeup, i;
144
145 ret = 0;
146 do_wakeup = 0;
147 i = 0;
148
3a326a2c
IM
149 if (pipe->inode)
150 mutex_lock(&pipe->inode->i_mutex);
5274f052 151
5274f052
JA
152 for (;;) {
153 int bufs;
154
3a326a2c 155 if (!pipe->readers) {
5274f052
JA
156 send_sig(SIGPIPE, current, 0);
157 if (!ret)
158 ret = -EPIPE;
159 break;
160 }
161
3a326a2c 162 bufs = pipe->nrbufs;
5274f052 163 if (bufs < PIPE_BUFFERS) {
3a326a2c
IM
164 int newbuf = (pipe->curbuf + bufs) & (PIPE_BUFFERS - 1);
165 struct pipe_buffer *buf = pipe->bufs + newbuf;
5274f052
JA
166 struct page *page = pages[i++];
167 unsigned long this_len;
168
169 this_len = PAGE_CACHE_SIZE - offset;
170 if (this_len > len)
171 this_len = len;
172
173 buf->page = page;
174 buf->offset = offset;
175 buf->len = this_len;
176 buf->ops = &page_cache_pipe_buf_ops;
3a326a2c 177 pipe->nrbufs = ++bufs;
5274f052
JA
178 do_wakeup = 1;
179
180 ret += this_len;
181 len -= this_len;
182 offset = 0;
183 if (!--nr_pages)
184 break;
185 if (!len)
186 break;
187 if (bufs < PIPE_BUFFERS)
188 continue;
189
190 break;
191 }
192
29e35094
LT
193 if (flags & SPLICE_F_NONBLOCK) {
194 if (!ret)
195 ret = -EAGAIN;
196 break;
197 }
198
5274f052
JA
199 if (signal_pending(current)) {
200 if (!ret)
201 ret = -ERESTARTSYS;
202 break;
203 }
204
205 if (do_wakeup) {
c0bd1f65 206 smp_mb();
3a326a2c
IM
207 if (waitqueue_active(&pipe->wait))
208 wake_up_interruptible_sync(&pipe->wait);
209 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
5274f052
JA
210 do_wakeup = 0;
211 }
212
3a326a2c
IM
213 pipe->waiting_writers++;
214 pipe_wait(pipe);
215 pipe->waiting_writers--;
5274f052
JA
216 }
217
3a326a2c
IM
218 if (pipe->inode)
219 mutex_unlock(&pipe->inode->i_mutex);
5274f052
JA
220
221 if (do_wakeup) {
c0bd1f65 222 smp_mb();
3a326a2c
IM
223 if (waitqueue_active(&pipe->wait))
224 wake_up_interruptible(&pipe->wait);
225 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
5274f052
JA
226 }
227
228 while (i < nr_pages)
229 page_cache_release(pages[i++]);
230
231 return ret;
232}
233
3a326a2c
IM
234static int
235__generic_file_splice_read(struct file *in, struct pipe_inode_info *pipe,
236 size_t len, unsigned int flags)
5274f052
JA
237{
238 struct address_space *mapping = in->f_mapping;
239 unsigned int offset, nr_pages;
16c523dd 240 struct page *pages[PIPE_BUFFERS];
5274f052 241 struct page *page;
16c523dd
JA
242 pgoff_t index;
243 int i;
5274f052
JA
244
245 index = in->f_pos >> PAGE_CACHE_SHIFT;
246 offset = in->f_pos & ~PAGE_CACHE_MASK;
247 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
248
249 if (nr_pages > PIPE_BUFFERS)
250 nr_pages = PIPE_BUFFERS;
251
252 /*
0b749ce3
JA
253 * initiate read-ahead on this page range. however, don't call into
254 * read-ahead if this is a non-zero offset (we are likely doing small
255 * chunk splice and the page is already there) for a single page.
5274f052 256 */
0b749ce3
JA
257 if (!offset || nr_pages > 1)
258 do_page_cache_readahead(mapping, in, index, nr_pages);
5274f052 259
5274f052
JA
260 /*
261 * now fill in the holes
262 */
16c523dd 263 for (i = 0; i < nr_pages; i++, index++) {
5274f052
JA
264 /*
265 * no page there, look one up / create it
266 */
16c523dd 267 page = find_or_create_page(mapping, index,
5274f052
JA
268 mapping_gfp_mask(mapping));
269 if (!page)
270 break;
271
272 if (PageUptodate(page))
273 unlock_page(page);
274 else {
16c523dd 275 int error = mapping->a_ops->readpage(in, page);
5274f052
JA
276
277 if (unlikely(error)) {
278 page_cache_release(page);
279 break;
280 }
281 }
16c523dd 282 pages[i] = page;
5274f052
JA
283 }
284
16c523dd
JA
285 if (i)
286 return move_to_pipe(pipe, pages, i, offset, len, flags);
5274f052 287
16c523dd 288 return 0;
5274f052
JA
289}
290
83f9135b
JA
291/**
292 * generic_file_splice_read - splice data from file to a pipe
293 * @in: file to splice from
294 * @pipe: pipe to splice to
295 * @len: number of bytes to splice
296 * @flags: splice modifier flags
297 *
298 * Will read pages from given file and fill them into a pipe.
299 *
300 */
3a326a2c 301ssize_t generic_file_splice_read(struct file *in, struct pipe_inode_info *pipe,
5274f052
JA
302 size_t len, unsigned int flags)
303{
304 ssize_t spliced;
305 int ret;
306
307 ret = 0;
308 spliced = 0;
3a326a2c 309
5274f052 310 while (len) {
29e35094 311 ret = __generic_file_splice_read(in, pipe, len, flags);
5274f052
JA
312
313 if (ret <= 0)
314 break;
315
316 in->f_pos += ret;
317 len -= ret;
318 spliced += ret;
29e35094
LT
319
320 if (!(flags & SPLICE_F_NONBLOCK))
321 continue;
322 ret = -EAGAIN;
323 break;
5274f052
JA
324 }
325
326 if (spliced)
327 return spliced;
328
329 return ret;
330}
331
059a8f37
JA
332EXPORT_SYMBOL(generic_file_splice_read);
333
5274f052 334/*
4f6f0bd2
JA
335 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
336 * using sendpage().
5274f052
JA
337 */
338static int pipe_to_sendpage(struct pipe_inode_info *info,
339 struct pipe_buffer *buf, struct splice_desc *sd)
340{
341 struct file *file = sd->file;
342 loff_t pos = sd->pos;
343 unsigned int offset;
344 ssize_t ret;
345 void *ptr;
b2b39fa4 346 int more;
5274f052
JA
347
348 /*
349 * sub-optimal, but we are limited by the pipe ->map. we don't
350 * need a kmap'ed buffer here, we just want to make sure we
351 * have the page pinned if the pipe page originates from the
352 * page cache
353 */
354 ptr = buf->ops->map(file, info, buf);
355 if (IS_ERR(ptr))
356 return PTR_ERR(ptr);
357
358 offset = pos & ~PAGE_CACHE_MASK;
b2b39fa4 359 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
5274f052 360
b2b39fa4 361 ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more);
5274f052
JA
362
363 buf->ops->unmap(info, buf);
364 if (ret == sd->len)
365 return 0;
366
367 return -EIO;
368}
369
370/*
371 * This is a little more tricky than the file -> pipe splicing. There are
372 * basically three cases:
373 *
374 * - Destination page already exists in the address space and there
375 * are users of it. For that case we have no other option that
376 * copying the data. Tough luck.
377 * - Destination page already exists in the address space, but there
378 * are no users of it. Make sure it's uptodate, then drop it. Fall
379 * through to last case.
380 * - Destination page does not exist, we can add the pipe page to
381 * the page cache and avoid the copy.
382 *
83f9135b
JA
383 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
384 * sd->flags), we attempt to migrate pages from the pipe to the output
385 * file address space page cache. This is possible if no one else has
386 * the pipe page referenced outside of the pipe and page cache. If
387 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
388 * a new page in the output file page cache and fill/dirty that.
5274f052
JA
389 */
390static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf,
391 struct splice_desc *sd)
392{
393 struct file *file = sd->file;
394 struct address_space *mapping = file->f_mapping;
3e7ee3e7 395 gfp_t gfp_mask = mapping_gfp_mask(mapping);
5274f052
JA
396 unsigned int offset;
397 struct page *page;
5274f052 398 pgoff_t index;
5abc97aa 399 char *src;
3e7ee3e7 400 int ret;
5274f052
JA
401
402 /*
49d0b21b 403 * make sure the data in this buffer is uptodate
5274f052
JA
404 */
405 src = buf->ops->map(file, info, buf);
406 if (IS_ERR(src))
407 return PTR_ERR(src);
408
409 index = sd->pos >> PAGE_CACHE_SHIFT;
410 offset = sd->pos & ~PAGE_CACHE_MASK;
411
5274f052 412 /*
5abc97aa 413 * reuse buf page, if SPLICE_F_MOVE is set
5274f052 414 */
5abc97aa 415 if (sd->flags & SPLICE_F_MOVE) {
83f9135b
JA
416 /*
417 * If steal succeeds, buf->page is now pruned from the vm
418 * side (LRU and page cache) and we can reuse it.
419 */
5abc97aa
JA
420 if (buf->ops->steal(info, buf))
421 goto find_page;
422
49d0b21b
JA
423 /*
424 * this will also set the page locked
425 */
5abc97aa 426 page = buf->page;
3e7ee3e7 427 if (add_to_page_cache(page, mapping, index, gfp_mask))
5abc97aa 428 goto find_page;
3e7ee3e7
JA
429
430 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
431 lru_cache_add(page);
5abc97aa
JA
432 } else {
433find_page:
434 ret = -ENOMEM;
3e7ee3e7 435 page = find_or_create_page(mapping, index, gfp_mask);
5abc97aa 436 if (!page)
9aefe431 437 goto out_nomem;
5abc97aa
JA
438
439 /*
440 * If the page is uptodate, it is also locked. If it isn't
441 * uptodate, we can mark it uptodate if we are filling the
442 * full page. Otherwise we need to read it in first...
443 */
444 if (!PageUptodate(page)) {
445 if (sd->len < PAGE_CACHE_SIZE) {
446 ret = mapping->a_ops->readpage(file, page);
447 if (unlikely(ret))
448 goto out;
449
450 lock_page(page);
451
452 if (!PageUptodate(page)) {
453 /*
454 * page got invalidated, repeat
455 */
456 if (!page->mapping) {
457 unlock_page(page);
458 page_cache_release(page);
459 goto find_page;
460 }
461 ret = -EIO;
462 goto out;
5274f052 463 }
5abc97aa
JA
464 } else {
465 WARN_ON(!PageLocked(page));
466 SetPageUptodate(page);
5274f052 467 }
5274f052
JA
468 }
469 }
470
471 ret = mapping->a_ops->prepare_write(file, page, 0, sd->len);
4f6f0bd2
JA
472 if (ret == AOP_TRUNCATED_PAGE) {
473 page_cache_release(page);
474 goto find_page;
475 } else if (ret)
5274f052
JA
476 goto out;
477
3e7ee3e7 478 if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) {
5abc97aa
JA
479 char *dst = kmap_atomic(page, KM_USER0);
480
481 memcpy(dst + offset, src + buf->offset, sd->len);
482 flush_dcache_page(page);
483 kunmap_atomic(dst, KM_USER0);
484 }
5274f052
JA
485
486 ret = mapping->a_ops->commit_write(file, page, 0, sd->len);
4f6f0bd2
JA
487 if (ret == AOP_TRUNCATED_PAGE) {
488 page_cache_release(page);
489 goto find_page;
490 } else if (ret)
5274f052
JA
491 goto out;
492
c7f21e4f 493 mark_page_accessed(page);
4f6f0bd2 494 balance_dirty_pages_ratelimited(mapping);
5274f052 495out:
3e7ee3e7 496 if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) {
5abc97aa 497 page_cache_release(page);
4f6f0bd2
JA
498 unlock_page(page);
499 }
9aefe431 500out_nomem:
5274f052
JA
501 buf->ops->unmap(info, buf);
502 return ret;
503}
504
505typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *,
506 struct splice_desc *);
507
83f9135b
JA
508/*
509 * Pipe input worker. Most of this logic works like a regular pipe, the
510 * key here is the 'actor' worker passed in that actually moves the data
511 * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
512 */
3a326a2c 513static ssize_t move_from_pipe(struct pipe_inode_info *pipe, struct file *out,
5274f052
JA
514 size_t len, unsigned int flags,
515 splice_actor *actor)
516{
5274f052
JA
517 int ret, do_wakeup, err;
518 struct splice_desc sd;
519
520 ret = 0;
521 do_wakeup = 0;
522
523 sd.total_len = len;
524 sd.flags = flags;
525 sd.file = out;
526 sd.pos = out->f_pos;
527
3a326a2c
IM
528 if (pipe->inode)
529 mutex_lock(&pipe->inode->i_mutex);
5274f052 530
5274f052 531 for (;;) {
3a326a2c 532 int bufs = pipe->nrbufs;
5274f052
JA
533
534 if (bufs) {
3a326a2c
IM
535 int curbuf = pipe->curbuf;
536 struct pipe_buffer *buf = pipe->bufs + curbuf;
5274f052
JA
537 struct pipe_buf_operations *ops = buf->ops;
538
539 sd.len = buf->len;
540 if (sd.len > sd.total_len)
541 sd.len = sd.total_len;
542
3a326a2c 543 err = actor(pipe, buf, &sd);
5274f052
JA
544 if (err) {
545 if (!ret && err != -ENODATA)
546 ret = err;
547
548 break;
549 }
550
551 ret += sd.len;
552 buf->offset += sd.len;
553 buf->len -= sd.len;
554 if (!buf->len) {
555 buf->ops = NULL;
3a326a2c 556 ops->release(pipe, buf);
5274f052 557 curbuf = (curbuf + 1) & (PIPE_BUFFERS - 1);
3a326a2c
IM
558 pipe->curbuf = curbuf;
559 pipe->nrbufs = --bufs;
5274f052
JA
560 do_wakeup = 1;
561 }
562
563 sd.pos += sd.len;
564 sd.total_len -= sd.len;
565 if (!sd.total_len)
566 break;
567 }
568
569 if (bufs)
570 continue;
3a326a2c 571 if (!pipe->writers)
5274f052 572 break;
3a326a2c 573 if (!pipe->waiting_writers) {
5274f052
JA
574 if (ret)
575 break;
576 }
577
29e35094
LT
578 if (flags & SPLICE_F_NONBLOCK) {
579 if (!ret)
580 ret = -EAGAIN;
581 break;
582 }
583
5274f052
JA
584 if (signal_pending(current)) {
585 if (!ret)
586 ret = -ERESTARTSYS;
587 break;
588 }
589
590 if (do_wakeup) {
c0bd1f65 591 smp_mb();
3a326a2c
IM
592 if (waitqueue_active(&pipe->wait))
593 wake_up_interruptible_sync(&pipe->wait);
594 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
5274f052
JA
595 do_wakeup = 0;
596 }
597
3a326a2c 598 pipe_wait(pipe);
5274f052
JA
599 }
600
3a326a2c
IM
601 if (pipe->inode)
602 mutex_unlock(&pipe->inode->i_mutex);
5274f052
JA
603
604 if (do_wakeup) {
c0bd1f65 605 smp_mb();
3a326a2c
IM
606 if (waitqueue_active(&pipe->wait))
607 wake_up_interruptible(&pipe->wait);
608 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
5274f052
JA
609 }
610
611 mutex_lock(&out->f_mapping->host->i_mutex);
612 out->f_pos = sd.pos;
613 mutex_unlock(&out->f_mapping->host->i_mutex);
614 return ret;
615
616}
617
83f9135b
JA
618/**
619 * generic_file_splice_write - splice data from a pipe to a file
3a326a2c 620 * @pipe: pipe info
83f9135b
JA
621 * @out: file to write to
622 * @len: number of bytes to splice
623 * @flags: splice modifier flags
624 *
625 * Will either move or copy pages (determined by @flags options) from
626 * the given pipe inode to the given file.
627 *
628 */
3a326a2c
IM
629ssize_t
630generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out,
631 size_t len, unsigned int flags)
5274f052 632{
4f6f0bd2 633 struct address_space *mapping = out->f_mapping;
3a326a2c
IM
634 ssize_t ret;
635
636 ret = move_from_pipe(pipe, out, len, flags, pipe_to_file);
4f6f0bd2
JA
637
638 /*
639 * if file or inode is SYNC and we actually wrote some data, sync it
640 */
641 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host))
642 && ret > 0) {
643 struct inode *inode = mapping->host;
644 int err;
645
646 mutex_lock(&inode->i_mutex);
647 err = generic_osync_inode(mapping->host, mapping,
648 OSYNC_METADATA|OSYNC_DATA);
649 mutex_unlock(&inode->i_mutex);
650
651 if (err)
652 ret = err;
653 }
654
655 return ret;
5274f052
JA
656}
657
059a8f37
JA
658EXPORT_SYMBOL(generic_file_splice_write);
659
83f9135b
JA
660/**
661 * generic_splice_sendpage - splice data from a pipe to a socket
662 * @inode: pipe inode
663 * @out: socket to write to
664 * @len: number of bytes to splice
665 * @flags: splice modifier flags
666 *
667 * Will send @len bytes from the pipe to a network socket. No data copying
668 * is involved.
669 *
670 */
3a326a2c 671ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out,
5274f052
JA
672 size_t len, unsigned int flags)
673{
3a326a2c 674 return move_from_pipe(pipe, out, len, flags, pipe_to_sendpage);
5274f052
JA
675}
676
059a8f37 677EXPORT_SYMBOL(generic_splice_sendpage);
a0f06780 678
83f9135b
JA
679/*
680 * Attempt to initiate a splice from pipe to file.
681 */
3a326a2c 682static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
b92ce558 683 size_t len, unsigned int flags)
5274f052
JA
684{
685 loff_t pos;
686 int ret;
687
688 if (!out->f_op || !out->f_op->splice_write)
689 return -EINVAL;
690
691 if (!(out->f_mode & FMODE_WRITE))
692 return -EBADF;
693
694 pos = out->f_pos;
529565dc 695
5274f052
JA
696 ret = rw_verify_area(WRITE, out, &pos, len);
697 if (unlikely(ret < 0))
698 return ret;
699
700 return out->f_op->splice_write(pipe, out, len, flags);
701}
702
83f9135b
JA
703/*
704 * Attempt to initiate a splice from a file to a pipe.
705 */
b92ce558
JA
706static long do_splice_to(struct file *in, struct pipe_inode_info *pipe,
707 size_t len, unsigned int flags)
5274f052
JA
708{
709 loff_t pos, isize, left;
710 int ret;
711
712 if (!in->f_op || !in->f_op->splice_read)
713 return -EINVAL;
714
715 if (!(in->f_mode & FMODE_READ))
716 return -EBADF;
717
718 pos = in->f_pos;
529565dc 719
5274f052
JA
720 ret = rw_verify_area(READ, in, &pos, len);
721 if (unlikely(ret < 0))
722 return ret;
723
724 isize = i_size_read(in->f_mapping->host);
725 if (unlikely(in->f_pos >= isize))
726 return 0;
727
728 left = isize - in->f_pos;
729 if (left < len)
730 len = left;
731
732 return in->f_op->splice_read(in, pipe, len, flags);
733}
734
b92ce558
JA
735long do_splice_direct(struct file *in, struct file *out, size_t len,
736 unsigned int flags)
737{
738 struct pipe_inode_info *pipe;
739 long ret, bytes;
740 umode_t i_mode;
741 int i;
742
743 /*
744 * We require the input being a regular file, as we don't want to
745 * randomly drop data for eg socket -> socket splicing. Use the
746 * piped splicing for that!
747 */
748 i_mode = in->f_dentry->d_inode->i_mode;
749 if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode)))
750 return -EINVAL;
751
752 /*
753 * neither in nor out is a pipe, setup an internal pipe attached to
754 * 'out' and transfer the wanted data from 'in' to 'out' through that
755 */
756 pipe = current->splice_pipe;
757 if (!pipe) {
758 pipe = alloc_pipe_info(NULL);
759 if (!pipe)
760 return -ENOMEM;
761
762 /*
763 * We don't have an immediate reader, but we'll read the stuff
764 * out of the pipe right after the move_to_pipe(). So set
765 * PIPE_READERS appropriately.
766 */
767 pipe->readers = 1;
768
769 current->splice_pipe = pipe;
770 }
771
772 /*
773 * do the splice
774 */
775 ret = 0;
776 bytes = 0;
777
778 while (len) {
779 size_t read_len, max_read_len;
780
781 /*
782 * Do at most PIPE_BUFFERS pages worth of transfer:
783 */
784 max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE));
785
786 ret = do_splice_to(in, pipe, max_read_len, flags);
787 if (unlikely(ret < 0))
788 goto out_release;
789
790 read_len = ret;
791
792 /*
793 * NOTE: nonblocking mode only applies to the input. We
794 * must not do the output in nonblocking mode as then we
795 * could get stuck data in the internal pipe:
796 */
797 ret = do_splice_from(pipe, out, read_len,
798 flags & ~SPLICE_F_NONBLOCK);
799 if (unlikely(ret < 0))
800 goto out_release;
801
802 bytes += ret;
803 len -= ret;
804
805 /*
806 * In nonblocking mode, if we got back a short read then
807 * that was due to either an IO error or due to the
808 * pagecache entry not being there. In the IO error case
809 * the _next_ splice attempt will produce a clean IO error
810 * return value (not a short read), so in both cases it's
811 * correct to break out of the loop here:
812 */
813 if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len))
814 break;
815 }
816
817 pipe->nrbufs = pipe->curbuf = 0;
818
819 return bytes;
820
821out_release:
822 /*
823 * If we did an incomplete transfer we must release
824 * the pipe buffers in question:
825 */
826 for (i = 0; i < PIPE_BUFFERS; i++) {
827 struct pipe_buffer *buf = pipe->bufs + i;
828
829 if (buf->ops) {
830 buf->ops->release(pipe, buf);
831 buf->ops = NULL;
832 }
833 }
834 pipe->nrbufs = pipe->curbuf = 0;
835
836 /*
837 * If we transferred some data, return the number of bytes:
838 */
839 if (bytes > 0)
840 return bytes;
841
842 return ret;
843}
844
845EXPORT_SYMBOL(do_splice_direct);
846
83f9135b
JA
847/*
848 * Determine where to splice to/from.
849 */
529565dc
IM
850static long do_splice(struct file *in, loff_t __user *off_in,
851 struct file *out, loff_t __user *off_out,
852 size_t len, unsigned int flags)
5274f052 853{
3a326a2c 854 struct pipe_inode_info *pipe;
5274f052 855
3a326a2c 856 pipe = in->f_dentry->d_inode->i_pipe;
529565dc
IM
857 if (pipe) {
858 if (off_in)
859 return -ESPIPE;
b92ce558
JA
860 if (off_out) {
861 if (out->f_op->llseek == no_llseek)
862 return -EINVAL;
863 if (copy_from_user(&out->f_pos, off_out,
864 sizeof(loff_t)))
865 return -EFAULT;
866 }
529565dc 867
b92ce558 868 return do_splice_from(pipe, out, len, flags);
529565dc 869 }
5274f052 870
3a326a2c 871 pipe = out->f_dentry->d_inode->i_pipe;
529565dc
IM
872 if (pipe) {
873 if (off_out)
874 return -ESPIPE;
b92ce558
JA
875 if (off_in) {
876 if (in->f_op->llseek == no_llseek)
877 return -EINVAL;
878 if (copy_from_user(&in->f_pos, off_in, sizeof(loff_t)))
879 return -EFAULT;
880 }
529565dc 881
b92ce558 882 return do_splice_to(in, pipe, len, flags);
529565dc 883 }
5274f052
JA
884
885 return -EINVAL;
886}
887
529565dc
IM
888asmlinkage long sys_splice(int fd_in, loff_t __user *off_in,
889 int fd_out, loff_t __user *off_out,
890 size_t len, unsigned int flags)
5274f052
JA
891{
892 long error;
893 struct file *in, *out;
894 int fput_in, fput_out;
895
896 if (unlikely(!len))
897 return 0;
898
899 error = -EBADF;
529565dc 900 in = fget_light(fd_in, &fput_in);
5274f052
JA
901 if (in) {
902 if (in->f_mode & FMODE_READ) {
529565dc 903 out = fget_light(fd_out, &fput_out);
5274f052
JA
904 if (out) {
905 if (out->f_mode & FMODE_WRITE)
529565dc
IM
906 error = do_splice(in, off_in,
907 out, off_out,
908 len, flags);
5274f052
JA
909 fput_light(out, fput_out);
910 }
911 }
912
913 fput_light(in, fput_in);
914 }
915
916 return error;
917}