blob: e8454455894690c9c5d0d4cb67da0d26a3e58dff [file] [log] [blame]
Jens Axboe5274f052006-03-30 15:15:30 +02001/*
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>
Jens Axboe5abc97a2006-03-30 15:16:46 +020024#include <linux/swap.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020025#include <linux/writeback.h>
26#include <linux/buffer_head.h>
Jeff Garzika0f06782006-03-30 23:06:13 -050027#include <linux/module.h>
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020028#include <linux/syscalls.h>
Jens Axboe5274f052006-03-30 15:15:30 +020029
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
Jens Axboe83f91352006-04-02 23:05:09 +020040/*
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 */
Jens Axboe5abc97a2006-03-30 15:16:46 +020046static int page_cache_pipe_buf_steal(struct pipe_inode_info *info,
47 struct pipe_buffer *buf)
48{
49 struct page *page = buf->page;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020050 struct address_space *mapping = page_mapping(page);
Jens Axboe5abc97a2006-03-30 15:16:46 +020051
52 WARN_ON(!PageLocked(page));
53 WARN_ON(!PageUptodate(page));
54
Jens Axboe4f6f0bd2006-04-02 23:04:46 +020055 if (PagePrivate(page))
56 try_to_release_page(page, mapping_gfp_mask(mapping));
57
58 if (!remove_mapping(mapping, page))
Jens Axboe5abc97a2006-03-30 15:16:46 +020059 return 1;
60
61 if (PageLRU(page)) {
62 struct zone *zone = page_zone(page);
63
64 spin_lock_irq(&zone->lru_lock);
65 BUG_ON(!PageLRU(page));
66 __ClearPageLRU(page);
67 del_page_from_lru(zone, page);
68 spin_unlock_irq(&zone->lru_lock);
69 }
70
Jens Axboe5abc97a2006-03-30 15:16:46 +020071 return 0;
72}
73
Jens Axboe5274f052006-03-30 15:15:30 +020074static 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;
79}
80
81static void *page_cache_pipe_buf_map(struct file *file,
82 struct pipe_inode_info *info,
83 struct pipe_buffer *buf)
84{
85 struct page *page = buf->page;
86
87 lock_page(page);
88
89 if (!PageUptodate(page)) {
90 unlock_page(page);
91 return ERR_PTR(-EIO);
92 }
93
94 if (!page->mapping) {
95 unlock_page(page);
96 return ERR_PTR(-ENODATA);
97 }
98
99 return kmap(buf->page);
100}
101
102static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info,
103 struct pipe_buffer *buf)
104{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200105 unlock_page(buf->page);
Jens Axboe5274f052006-03-30 15:15:30 +0200106 kunmap(buf->page);
107}
108
109static struct pipe_buf_operations page_cache_pipe_buf_ops = {
110 .can_merge = 0,
111 .map = page_cache_pipe_buf_map,
112 .unmap = page_cache_pipe_buf_unmap,
113 .release = page_cache_pipe_buf_release,
Jens Axboe5abc97a2006-03-30 15:16:46 +0200114 .steal = page_cache_pipe_buf_steal,
Jens Axboe5274f052006-03-30 15:15:30 +0200115};
116
Jens Axboe83f91352006-04-02 23:05:09 +0200117/*
118 * Pipe output worker. This sets up our pipe format with the page cache
119 * pipe buffer operations. Otherwise very similar to the regular pipe_writev().
120 */
Jens Axboe5274f052006-03-30 15:15:30 +0200121static ssize_t move_to_pipe(struct inode *inode, struct page **pages,
122 int nr_pages, unsigned long offset,
Linus Torvalds29e35092006-04-02 12:46:35 -0700123 unsigned long len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200124{
125 struct pipe_inode_info *info;
126 int ret, do_wakeup, i;
127
128 ret = 0;
129 do_wakeup = 0;
130 i = 0;
131
132 mutex_lock(PIPE_MUTEX(*inode));
133
134 info = inode->i_pipe;
135 for (;;) {
136 int bufs;
137
138 if (!PIPE_READERS(*inode)) {
139 send_sig(SIGPIPE, current, 0);
140 if (!ret)
141 ret = -EPIPE;
142 break;
143 }
144
145 bufs = info->nrbufs;
146 if (bufs < PIPE_BUFFERS) {
147 int newbuf = (info->curbuf + bufs) & (PIPE_BUFFERS - 1);
148 struct pipe_buffer *buf = info->bufs + newbuf;
149 struct page *page = pages[i++];
150 unsigned long this_len;
151
152 this_len = PAGE_CACHE_SIZE - offset;
153 if (this_len > len)
154 this_len = len;
155
156 buf->page = page;
157 buf->offset = offset;
158 buf->len = this_len;
159 buf->ops = &page_cache_pipe_buf_ops;
160 info->nrbufs = ++bufs;
161 do_wakeup = 1;
162
163 ret += this_len;
164 len -= this_len;
165 offset = 0;
166 if (!--nr_pages)
167 break;
168 if (!len)
169 break;
170 if (bufs < PIPE_BUFFERS)
171 continue;
172
173 break;
174 }
175
Linus Torvalds29e35092006-04-02 12:46:35 -0700176 if (flags & SPLICE_F_NONBLOCK) {
177 if (!ret)
178 ret = -EAGAIN;
179 break;
180 }
181
Jens Axboe5274f052006-03-30 15:15:30 +0200182 if (signal_pending(current)) {
183 if (!ret)
184 ret = -ERESTARTSYS;
185 break;
186 }
187
188 if (do_wakeup) {
189 wake_up_interruptible_sync(PIPE_WAIT(*inode));
190 kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO,
191 POLL_IN);
192 do_wakeup = 0;
193 }
194
195 PIPE_WAITING_WRITERS(*inode)++;
196 pipe_wait(inode);
197 PIPE_WAITING_WRITERS(*inode)--;
198 }
199
200 mutex_unlock(PIPE_MUTEX(*inode));
201
202 if (do_wakeup) {
203 wake_up_interruptible(PIPE_WAIT(*inode));
204 kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN);
205 }
206
207 while (i < nr_pages)
208 page_cache_release(pages[i++]);
209
210 return ret;
211}
212
213static int __generic_file_splice_read(struct file *in, struct inode *pipe,
Linus Torvalds29e35092006-04-02 12:46:35 -0700214 size_t len, unsigned int flags)
Jens Axboe5274f052006-03-30 15:15:30 +0200215{
216 struct address_space *mapping = in->f_mapping;
217 unsigned int offset, nr_pages;
218 struct page *pages[PIPE_BUFFERS], *shadow[PIPE_BUFFERS];
219 struct page *page;
220 pgoff_t index, pidx;
221 int i, j;
222
223 index = in->f_pos >> PAGE_CACHE_SHIFT;
224 offset = in->f_pos & ~PAGE_CACHE_MASK;
225 nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
226
227 if (nr_pages > PIPE_BUFFERS)
228 nr_pages = PIPE_BUFFERS;
229
230 /*
231 * initiate read-ahead on this page range
232 */
233 do_page_cache_readahead(mapping, in, index, nr_pages);
234
235 /*
236 * Get as many pages from the page cache as possible..
237 * Start IO on the page cache entries we create (we
238 * can assume that any pre-existing ones we find have
239 * already had IO started on them).
240 */
241 i = find_get_pages(mapping, index, nr_pages, pages);
242
243 /*
244 * common case - we found all pages and they are contiguous,
245 * kick them off
246 */
247 if (i && (pages[i - 1]->index == index + i - 1))
248 goto splice_them;
249
250 /*
251 * fill shadow[] with pages at the right locations, so we only
252 * have to fill holes
253 */
Jens Axboe53cd9ae2006-04-02 23:04:21 +0200254 memset(shadow, 0, nr_pages * sizeof(struct page *));
255 for (j = 0; j < i; j++)
256 shadow[pages[j]->index - index] = pages[j];
Jens Axboe5274f052006-03-30 15:15:30 +0200257
258 /*
259 * now fill in the holes
260 */
261 for (i = 0, pidx = index; i < nr_pages; pidx++, i++) {
262 int error;
263
264 if (shadow[i])
265 continue;
266
267 /*
268 * no page there, look one up / create it
269 */
270 page = find_or_create_page(mapping, pidx,
271 mapping_gfp_mask(mapping));
272 if (!page)
273 break;
274
275 if (PageUptodate(page))
276 unlock_page(page);
277 else {
278 error = mapping->a_ops->readpage(in, page);
279
280 if (unlikely(error)) {
281 page_cache_release(page);
282 break;
283 }
284 }
285 shadow[i] = page;
286 }
287
288 if (!i) {
289 for (i = 0; i < nr_pages; i++) {
290 if (shadow[i])
291 page_cache_release(shadow[i]);
292 }
293 return 0;
294 }
295
296 memcpy(pages, shadow, i * sizeof(struct page *));
297
298 /*
299 * Now we splice them into the pipe..
300 */
301splice_them:
Linus Torvalds29e35092006-04-02 12:46:35 -0700302 return move_to_pipe(pipe, pages, i, offset, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200303}
304
Jens Axboe83f91352006-04-02 23:05:09 +0200305/**
306 * generic_file_splice_read - splice data from file to a pipe
307 * @in: file to splice from
308 * @pipe: pipe to splice to
309 * @len: number of bytes to splice
310 * @flags: splice modifier flags
311 *
312 * Will read pages from given file and fill them into a pipe.
313 *
314 */
Jens Axboe5274f052006-03-30 15:15:30 +0200315ssize_t generic_file_splice_read(struct file *in, struct inode *pipe,
316 size_t len, unsigned int flags)
317{
318 ssize_t spliced;
319 int ret;
320
321 ret = 0;
322 spliced = 0;
323 while (len) {
Linus Torvalds29e35092006-04-02 12:46:35 -0700324 ret = __generic_file_splice_read(in, pipe, len, flags);
Jens Axboe5274f052006-03-30 15:15:30 +0200325
326 if (ret <= 0)
327 break;
328
329 in->f_pos += ret;
330 len -= ret;
331 spliced += ret;
Linus Torvalds29e35092006-04-02 12:46:35 -0700332
333 if (!(flags & SPLICE_F_NONBLOCK))
334 continue;
335 ret = -EAGAIN;
336 break;
Jens Axboe5274f052006-03-30 15:15:30 +0200337 }
338
339 if (spliced)
340 return spliced;
341
342 return ret;
343}
344
345/*
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200346 * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos'
347 * using sendpage().
Jens Axboe5274f052006-03-30 15:15:30 +0200348 */
349static int pipe_to_sendpage(struct pipe_inode_info *info,
350 struct pipe_buffer *buf, struct splice_desc *sd)
351{
352 struct file *file = sd->file;
353 loff_t pos = sd->pos;
354 unsigned int offset;
355 ssize_t ret;
356 void *ptr;
Jens Axboeb2b39fa2006-04-02 23:05:41 +0200357 int more;
Jens Axboe5274f052006-03-30 15:15:30 +0200358
359 /*
360 * sub-optimal, but we are limited by the pipe ->map. we don't
361 * need a kmap'ed buffer here, we just want to make sure we
362 * have the page pinned if the pipe page originates from the
363 * page cache
364 */
365 ptr = buf->ops->map(file, info, buf);
366 if (IS_ERR(ptr))
367 return PTR_ERR(ptr);
368
369 offset = pos & ~PAGE_CACHE_MASK;
Jens Axboeb2b39fa2006-04-02 23:05:41 +0200370 more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len;
Jens Axboe5274f052006-03-30 15:15:30 +0200371
Jens Axboeb2b39fa2006-04-02 23:05:41 +0200372 ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more);
Jens Axboe5274f052006-03-30 15:15:30 +0200373
374 buf->ops->unmap(info, buf);
375 if (ret == sd->len)
376 return 0;
377
378 return -EIO;
379}
380
381/*
382 * This is a little more tricky than the file -> pipe splicing. There are
383 * basically three cases:
384 *
385 * - Destination page already exists in the address space and there
386 * are users of it. For that case we have no other option that
387 * copying the data. Tough luck.
388 * - Destination page already exists in the address space, but there
389 * are no users of it. Make sure it's uptodate, then drop it. Fall
390 * through to last case.
391 * - Destination page does not exist, we can add the pipe page to
392 * the page cache and avoid the copy.
393 *
Jens Axboe83f91352006-04-02 23:05:09 +0200394 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
395 * sd->flags), we attempt to migrate pages from the pipe to the output
396 * file address space page cache. This is possible if no one else has
397 * the pipe page referenced outside of the pipe and page cache. If
398 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
399 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200400 */
401static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf,
402 struct splice_desc *sd)
403{
404 struct file *file = sd->file;
405 struct address_space *mapping = file->f_mapping;
406 unsigned int offset;
407 struct page *page;
Jens Axboe5274f052006-03-30 15:15:30 +0200408 pgoff_t index;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200409 char *src;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200410 int ret, stolen;
Jens Axboe5274f052006-03-30 15:15:30 +0200411
412 /*
413 * after this, page will be locked and unmapped
414 */
415 src = buf->ops->map(file, info, buf);
416 if (IS_ERR(src))
417 return PTR_ERR(src);
418
419 index = sd->pos >> PAGE_CACHE_SHIFT;
420 offset = sd->pos & ~PAGE_CACHE_MASK;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200421 stolen = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200422
Jens Axboe5274f052006-03-30 15:15:30 +0200423 /*
Jens Axboe5abc97a2006-03-30 15:16:46 +0200424 * reuse buf page, if SPLICE_F_MOVE is set
Jens Axboe5274f052006-03-30 15:15:30 +0200425 */
Jens Axboe5abc97a2006-03-30 15:16:46 +0200426 if (sd->flags & SPLICE_F_MOVE) {
Jens Axboe83f91352006-04-02 23:05:09 +0200427 /*
428 * If steal succeeds, buf->page is now pruned from the vm
429 * side (LRU and page cache) and we can reuse it.
430 */
Jens Axboe5abc97a2006-03-30 15:16:46 +0200431 if (buf->ops->steal(info, buf))
432 goto find_page;
Jens Axboe5274f052006-03-30 15:15:30 +0200433
Jens Axboe5abc97a2006-03-30 15:16:46 +0200434 page = buf->page;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200435 stolen = 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200436 if (add_to_page_cache_lru(page, mapping, index,
437 mapping_gfp_mask(mapping)))
438 goto find_page;
439 } else {
440find_page:
441 ret = -ENOMEM;
442 page = find_or_create_page(mapping, index,
443 mapping_gfp_mask(mapping));
444 if (!page)
445 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200446
Jens Axboe5abc97a2006-03-30 15:16:46 +0200447 /*
448 * If the page is uptodate, it is also locked. If it isn't
449 * uptodate, we can mark it uptodate if we are filling the
450 * full page. Otherwise we need to read it in first...
451 */
452 if (!PageUptodate(page)) {
453 if (sd->len < PAGE_CACHE_SIZE) {
454 ret = mapping->a_ops->readpage(file, page);
455 if (unlikely(ret))
456 goto out;
457
458 lock_page(page);
459
460 if (!PageUptodate(page)) {
461 /*
462 * page got invalidated, repeat
463 */
464 if (!page->mapping) {
465 unlock_page(page);
466 page_cache_release(page);
467 goto find_page;
468 }
469 ret = -EIO;
470 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200471 }
Jens Axboe5abc97a2006-03-30 15:16:46 +0200472 } else {
473 WARN_ON(!PageLocked(page));
474 SetPageUptodate(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200475 }
Jens Axboe5274f052006-03-30 15:15:30 +0200476 }
477 }
478
479 ret = mapping->a_ops->prepare_write(file, page, 0, sd->len);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200480 if (ret == AOP_TRUNCATED_PAGE) {
481 page_cache_release(page);
482 goto find_page;
483 } else if (ret)
Jens Axboe5274f052006-03-30 15:15:30 +0200484 goto out;
485
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200486 if (!stolen) {
Jens Axboe5abc97a2006-03-30 15:16:46 +0200487 char *dst = kmap_atomic(page, KM_USER0);
488
489 memcpy(dst + offset, src + buf->offset, sd->len);
490 flush_dcache_page(page);
491 kunmap_atomic(dst, KM_USER0);
492 }
Jens Axboe5274f052006-03-30 15:15:30 +0200493
494 ret = mapping->a_ops->commit_write(file, page, 0, sd->len);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200495 if (ret == AOP_TRUNCATED_PAGE) {
496 page_cache_release(page);
497 goto find_page;
498 } else if (ret)
Jens Axboe5274f052006-03-30 15:15:30 +0200499 goto out;
500
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200501 balance_dirty_pages_ratelimited(mapping);
Jens Axboe5274f052006-03-30 15:15:30 +0200502out:
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200503 if (!stolen) {
Jens Axboe5abc97a2006-03-30 15:16:46 +0200504 page_cache_release(page);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200505 unlock_page(page);
506 }
Jens Axboe5274f052006-03-30 15:15:30 +0200507 buf->ops->unmap(info, buf);
508 return ret;
509}
510
511typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *,
512 struct splice_desc *);
513
Jens Axboe83f91352006-04-02 23:05:09 +0200514/*
515 * Pipe input worker. Most of this logic works like a regular pipe, the
516 * key here is the 'actor' worker passed in that actually moves the data
517 * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
518 */
Jens Axboe5274f052006-03-30 15:15:30 +0200519static ssize_t move_from_pipe(struct inode *inode, struct file *out,
520 size_t len, unsigned int flags,
521 splice_actor *actor)
522{
523 struct pipe_inode_info *info;
524 int ret, do_wakeup, err;
525 struct splice_desc sd;
526
527 ret = 0;
528 do_wakeup = 0;
529
530 sd.total_len = len;
531 sd.flags = flags;
532 sd.file = out;
533 sd.pos = out->f_pos;
534
535 mutex_lock(PIPE_MUTEX(*inode));
536
537 info = inode->i_pipe;
538 for (;;) {
539 int bufs = info->nrbufs;
540
541 if (bufs) {
542 int curbuf = info->curbuf;
543 struct pipe_buffer *buf = info->bufs + curbuf;
544 struct pipe_buf_operations *ops = buf->ops;
545
546 sd.len = buf->len;
547 if (sd.len > sd.total_len)
548 sd.len = sd.total_len;
549
550 err = actor(info, buf, &sd);
551 if (err) {
552 if (!ret && err != -ENODATA)
553 ret = err;
554
555 break;
556 }
557
558 ret += sd.len;
559 buf->offset += sd.len;
560 buf->len -= sd.len;
561 if (!buf->len) {
562 buf->ops = NULL;
563 ops->release(info, buf);
564 curbuf = (curbuf + 1) & (PIPE_BUFFERS - 1);
565 info->curbuf = curbuf;
566 info->nrbufs = --bufs;
567 do_wakeup = 1;
568 }
569
570 sd.pos += sd.len;
571 sd.total_len -= sd.len;
572 if (!sd.total_len)
573 break;
574 }
575
576 if (bufs)
577 continue;
578 if (!PIPE_WRITERS(*inode))
579 break;
580 if (!PIPE_WAITING_WRITERS(*inode)) {
581 if (ret)
582 break;
583 }
584
Linus Torvalds29e35092006-04-02 12:46:35 -0700585 if (flags & SPLICE_F_NONBLOCK) {
586 if (!ret)
587 ret = -EAGAIN;
588 break;
589 }
590
Jens Axboe5274f052006-03-30 15:15:30 +0200591 if (signal_pending(current)) {
592 if (!ret)
593 ret = -ERESTARTSYS;
594 break;
595 }
596
597 if (do_wakeup) {
598 wake_up_interruptible_sync(PIPE_WAIT(*inode));
599 kill_fasync(PIPE_FASYNC_WRITERS(*inode),SIGIO,POLL_OUT);
600 do_wakeup = 0;
601 }
602
603 pipe_wait(inode);
604 }
605
606 mutex_unlock(PIPE_MUTEX(*inode));
607
608 if (do_wakeup) {
609 wake_up_interruptible(PIPE_WAIT(*inode));
610 kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT);
611 }
612
613 mutex_lock(&out->f_mapping->host->i_mutex);
614 out->f_pos = sd.pos;
615 mutex_unlock(&out->f_mapping->host->i_mutex);
616 return ret;
617
618}
619
Jens Axboe83f91352006-04-02 23:05:09 +0200620/**
621 * generic_file_splice_write - splice data from a pipe to a file
622 * @inode: pipe inode
623 * @out: file to write to
624 * @len: number of bytes to splice
625 * @flags: splice modifier flags
626 *
627 * Will either move or copy pages (determined by @flags options) from
628 * the given pipe inode to the given file.
629 *
630 */
Jens Axboe5274f052006-03-30 15:15:30 +0200631ssize_t generic_file_splice_write(struct inode *inode, struct file *out,
632 size_t len, unsigned int flags)
633{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200634 struct address_space *mapping = out->f_mapping;
635 ssize_t ret = move_from_pipe(inode, out, len, flags, pipe_to_file);
636
637 /*
638 * if file or inode is SYNC and we actually wrote some data, sync it
639 */
640 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host))
641 && ret > 0) {
642 struct inode *inode = mapping->host;
643 int err;
644
645 mutex_lock(&inode->i_mutex);
646 err = generic_osync_inode(mapping->host, mapping,
647 OSYNC_METADATA|OSYNC_DATA);
648 mutex_unlock(&inode->i_mutex);
649
650 if (err)
651 ret = err;
652 }
653
654 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200655}
656
Jens Axboe83f91352006-04-02 23:05:09 +0200657/**
658 * generic_splice_sendpage - splice data from a pipe to a socket
659 * @inode: pipe inode
660 * @out: socket to write to
661 * @len: number of bytes to splice
662 * @flags: splice modifier flags
663 *
664 * Will send @len bytes from the pipe to a network socket. No data copying
665 * is involved.
666 *
667 */
Jens Axboe5274f052006-03-30 15:15:30 +0200668ssize_t generic_splice_sendpage(struct inode *inode, struct file *out,
669 size_t len, unsigned int flags)
670{
671 return move_from_pipe(inode, out, len, flags, pipe_to_sendpage);
672}
673
Jeff Garzika0f06782006-03-30 23:06:13 -0500674EXPORT_SYMBOL(generic_file_splice_write);
675EXPORT_SYMBOL(generic_file_splice_read);
676
Jens Axboe83f91352006-04-02 23:05:09 +0200677/*
678 * Attempt to initiate a splice from pipe to file.
679 */
Jens Axboe5274f052006-03-30 15:15:30 +0200680static long do_splice_from(struct inode *pipe, struct file *out, size_t len,
681 unsigned int flags)
682{
683 loff_t pos;
684 int ret;
685
686 if (!out->f_op || !out->f_op->splice_write)
687 return -EINVAL;
688
689 if (!(out->f_mode & FMODE_WRITE))
690 return -EBADF;
691
692 pos = out->f_pos;
693 ret = rw_verify_area(WRITE, out, &pos, len);
694 if (unlikely(ret < 0))
695 return ret;
696
697 return out->f_op->splice_write(pipe, out, len, flags);
698}
699
Jens Axboe83f91352006-04-02 23:05:09 +0200700/*
701 * Attempt to initiate a splice from a file to a pipe.
702 */
Jens Axboe5274f052006-03-30 15:15:30 +0200703static long do_splice_to(struct file *in, struct inode *pipe, size_t len,
704 unsigned int flags)
705{
706 loff_t pos, isize, left;
707 int ret;
708
709 if (!in->f_op || !in->f_op->splice_read)
710 return -EINVAL;
711
712 if (!(in->f_mode & FMODE_READ))
713 return -EBADF;
714
715 pos = in->f_pos;
716 ret = rw_verify_area(READ, in, &pos, len);
717 if (unlikely(ret < 0))
718 return ret;
719
720 isize = i_size_read(in->f_mapping->host);
721 if (unlikely(in->f_pos >= isize))
722 return 0;
723
724 left = isize - in->f_pos;
725 if (left < len)
726 len = left;
727
728 return in->f_op->splice_read(in, pipe, len, flags);
729}
730
Jens Axboe83f91352006-04-02 23:05:09 +0200731/*
732 * Determine where to splice to/from.
733 */
Jens Axboe5274f052006-03-30 15:15:30 +0200734static long do_splice(struct file *in, struct file *out, size_t len,
735 unsigned int flags)
736{
737 struct inode *pipe;
738
739 pipe = in->f_dentry->d_inode;
740 if (pipe->i_pipe)
741 return do_splice_from(pipe, out, len, flags);
742
743 pipe = out->f_dentry->d_inode;
744 if (pipe->i_pipe)
745 return do_splice_to(in, pipe, len, flags);
746
747 return -EINVAL;
748}
749
750asmlinkage long sys_splice(int fdin, int fdout, size_t len, unsigned int flags)
751{
752 long error;
753 struct file *in, *out;
754 int fput_in, fput_out;
755
756 if (unlikely(!len))
757 return 0;
758
759 error = -EBADF;
760 in = fget_light(fdin, &fput_in);
761 if (in) {
762 if (in->f_mode & FMODE_READ) {
763 out = fget_light(fdout, &fput_out);
764 if (out) {
765 if (out->f_mode & FMODE_WRITE)
766 error = do_splice(in, out, len, flags);
767 fput_light(out, fput_out);
768 }
769 }
770
771 fput_light(in, fput_in);
772 }
773
774 return error;
775}