blob: 34591924c783538552641324ba69a87a714c938b [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;
357
358 /*
359 * sub-optimal, but we are limited by the pipe ->map. we don't
360 * need a kmap'ed buffer here, we just want to make sure we
361 * have the page pinned if the pipe page originates from the
362 * page cache
363 */
364 ptr = buf->ops->map(file, info, buf);
365 if (IS_ERR(ptr))
366 return PTR_ERR(ptr);
367
368 offset = pos & ~PAGE_CACHE_MASK;
369
370 ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,
371 sd->len < sd->total_len);
372
373 buf->ops->unmap(info, buf);
374 if (ret == sd->len)
375 return 0;
376
377 return -EIO;
378}
379
380/*
381 * This is a little more tricky than the file -> pipe splicing. There are
382 * basically three cases:
383 *
384 * - Destination page already exists in the address space and there
385 * are users of it. For that case we have no other option that
386 * copying the data. Tough luck.
387 * - Destination page already exists in the address space, but there
388 * are no users of it. Make sure it's uptodate, then drop it. Fall
389 * through to last case.
390 * - Destination page does not exist, we can add the pipe page to
391 * the page cache and avoid the copy.
392 *
Jens Axboe83f91352006-04-02 23:05:09 +0200393 * If asked to move pages to the output file (SPLICE_F_MOVE is set in
394 * sd->flags), we attempt to migrate pages from the pipe to the output
395 * file address space page cache. This is possible if no one else has
396 * the pipe page referenced outside of the pipe and page cache. If
397 * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create
398 * a new page in the output file page cache and fill/dirty that.
Jens Axboe5274f052006-03-30 15:15:30 +0200399 */
400static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf,
401 struct splice_desc *sd)
402{
403 struct file *file = sd->file;
404 struct address_space *mapping = file->f_mapping;
405 unsigned int offset;
406 struct page *page;
Jens Axboe5274f052006-03-30 15:15:30 +0200407 pgoff_t index;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200408 char *src;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200409 int ret, stolen;
Jens Axboe5274f052006-03-30 15:15:30 +0200410
411 /*
412 * after this, page will be locked and unmapped
413 */
414 src = buf->ops->map(file, info, buf);
415 if (IS_ERR(src))
416 return PTR_ERR(src);
417
418 index = sd->pos >> PAGE_CACHE_SHIFT;
419 offset = sd->pos & ~PAGE_CACHE_MASK;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200420 stolen = 0;
Jens Axboe5274f052006-03-30 15:15:30 +0200421
Jens Axboe5274f052006-03-30 15:15:30 +0200422 /*
Jens Axboe5abc97a2006-03-30 15:16:46 +0200423 * reuse buf page, if SPLICE_F_MOVE is set
Jens Axboe5274f052006-03-30 15:15:30 +0200424 */
Jens Axboe5abc97a2006-03-30 15:16:46 +0200425 if (sd->flags & SPLICE_F_MOVE) {
Jens Axboe83f91352006-04-02 23:05:09 +0200426 /*
427 * If steal succeeds, buf->page is now pruned from the vm
428 * side (LRU and page cache) and we can reuse it.
429 */
Jens Axboe5abc97a2006-03-30 15:16:46 +0200430 if (buf->ops->steal(info, buf))
431 goto find_page;
Jens Axboe5274f052006-03-30 15:15:30 +0200432
Jens Axboe5abc97a2006-03-30 15:16:46 +0200433 page = buf->page;
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200434 stolen = 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200435 if (add_to_page_cache_lru(page, mapping, index,
436 mapping_gfp_mask(mapping)))
437 goto find_page;
438 } else {
439find_page:
440 ret = -ENOMEM;
441 page = find_or_create_page(mapping, index,
442 mapping_gfp_mask(mapping));
443 if (!page)
444 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200445
Jens Axboe5abc97a2006-03-30 15:16:46 +0200446 /*
447 * If the page is uptodate, it is also locked. If it isn't
448 * uptodate, we can mark it uptodate if we are filling the
449 * full page. Otherwise we need to read it in first...
450 */
451 if (!PageUptodate(page)) {
452 if (sd->len < PAGE_CACHE_SIZE) {
453 ret = mapping->a_ops->readpage(file, page);
454 if (unlikely(ret))
455 goto out;
456
457 lock_page(page);
458
459 if (!PageUptodate(page)) {
460 /*
461 * page got invalidated, repeat
462 */
463 if (!page->mapping) {
464 unlock_page(page);
465 page_cache_release(page);
466 goto find_page;
467 }
468 ret = -EIO;
469 goto out;
Jens Axboe5274f052006-03-30 15:15:30 +0200470 }
Jens Axboe5abc97a2006-03-30 15:16:46 +0200471 } else {
472 WARN_ON(!PageLocked(page));
473 SetPageUptodate(page);
Jens Axboe5274f052006-03-30 15:15:30 +0200474 }
Jens Axboe5274f052006-03-30 15:15:30 +0200475 }
476 }
477
478 ret = mapping->a_ops->prepare_write(file, page, 0, sd->len);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200479 if (ret == AOP_TRUNCATED_PAGE) {
480 page_cache_release(page);
481 goto find_page;
482 } else if (ret)
Jens Axboe5274f052006-03-30 15:15:30 +0200483 goto out;
484
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200485 if (!stolen) {
Jens Axboe5abc97a2006-03-30 15:16:46 +0200486 char *dst = kmap_atomic(page, KM_USER0);
487
488 memcpy(dst + offset, src + buf->offset, sd->len);
489 flush_dcache_page(page);
490 kunmap_atomic(dst, KM_USER0);
491 }
Jens Axboe5274f052006-03-30 15:15:30 +0200492
493 ret = mapping->a_ops->commit_write(file, page, 0, sd->len);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200494 if (ret == AOP_TRUNCATED_PAGE) {
495 page_cache_release(page);
496 goto find_page;
497 } else if (ret)
Jens Axboe5274f052006-03-30 15:15:30 +0200498 goto out;
499
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200500 balance_dirty_pages_ratelimited(mapping);
Jens Axboe5274f052006-03-30 15:15:30 +0200501out:
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200502 if (!stolen) {
Jens Axboe5abc97a2006-03-30 15:16:46 +0200503 page_cache_release(page);
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200504 unlock_page(page);
505 }
Jens Axboe5274f052006-03-30 15:15:30 +0200506 buf->ops->unmap(info, buf);
507 return ret;
508}
509
510typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *,
511 struct splice_desc *);
512
Jens Axboe83f91352006-04-02 23:05:09 +0200513/*
514 * Pipe input worker. Most of this logic works like a regular pipe, the
515 * key here is the 'actor' worker passed in that actually moves the data
516 * to the wanted destination. See pipe_to_file/pipe_to_sendpage above.
517 */
Jens Axboe5274f052006-03-30 15:15:30 +0200518static ssize_t move_from_pipe(struct inode *inode, struct file *out,
519 size_t len, unsigned int flags,
520 splice_actor *actor)
521{
522 struct pipe_inode_info *info;
523 int ret, do_wakeup, err;
524 struct splice_desc sd;
525
526 ret = 0;
527 do_wakeup = 0;
528
529 sd.total_len = len;
530 sd.flags = flags;
531 sd.file = out;
532 sd.pos = out->f_pos;
533
534 mutex_lock(PIPE_MUTEX(*inode));
535
536 info = inode->i_pipe;
537 for (;;) {
538 int bufs = info->nrbufs;
539
540 if (bufs) {
541 int curbuf = info->curbuf;
542 struct pipe_buffer *buf = info->bufs + curbuf;
543 struct pipe_buf_operations *ops = buf->ops;
544
545 sd.len = buf->len;
546 if (sd.len > sd.total_len)
547 sd.len = sd.total_len;
548
549 err = actor(info, buf, &sd);
550 if (err) {
551 if (!ret && err != -ENODATA)
552 ret = err;
553
554 break;
555 }
556
557 ret += sd.len;
558 buf->offset += sd.len;
559 buf->len -= sd.len;
560 if (!buf->len) {
561 buf->ops = NULL;
562 ops->release(info, buf);
563 curbuf = (curbuf + 1) & (PIPE_BUFFERS - 1);
564 info->curbuf = curbuf;
565 info->nrbufs = --bufs;
566 do_wakeup = 1;
567 }
568
569 sd.pos += sd.len;
570 sd.total_len -= sd.len;
571 if (!sd.total_len)
572 break;
573 }
574
575 if (bufs)
576 continue;
577 if (!PIPE_WRITERS(*inode))
578 break;
579 if (!PIPE_WAITING_WRITERS(*inode)) {
580 if (ret)
581 break;
582 }
583
Linus Torvalds29e35092006-04-02 12:46:35 -0700584 if (flags & SPLICE_F_NONBLOCK) {
585 if (!ret)
586 ret = -EAGAIN;
587 break;
588 }
589
Jens Axboe5274f052006-03-30 15:15:30 +0200590 if (signal_pending(current)) {
591 if (!ret)
592 ret = -ERESTARTSYS;
593 break;
594 }
595
596 if (do_wakeup) {
597 wake_up_interruptible_sync(PIPE_WAIT(*inode));
598 kill_fasync(PIPE_FASYNC_WRITERS(*inode),SIGIO,POLL_OUT);
599 do_wakeup = 0;
600 }
601
602 pipe_wait(inode);
603 }
604
605 mutex_unlock(PIPE_MUTEX(*inode));
606
607 if (do_wakeup) {
608 wake_up_interruptible(PIPE_WAIT(*inode));
609 kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT);
610 }
611
612 mutex_lock(&out->f_mapping->host->i_mutex);
613 out->f_pos = sd.pos;
614 mutex_unlock(&out->f_mapping->host->i_mutex);
615 return ret;
616
617}
618
Jens Axboe83f91352006-04-02 23:05:09 +0200619/**
620 * generic_file_splice_write - splice data from a pipe to a file
621 * @inode: pipe inode
622 * @out: file to write to
623 * @len: number of bytes to splice
624 * @flags: splice modifier flags
625 *
626 * Will either move or copy pages (determined by @flags options) from
627 * the given pipe inode to the given file.
628 *
629 */
Jens Axboe5274f052006-03-30 15:15:30 +0200630ssize_t generic_file_splice_write(struct inode *inode, struct file *out,
631 size_t len, unsigned int flags)
632{
Jens Axboe4f6f0bd2006-04-02 23:04:46 +0200633 struct address_space *mapping = out->f_mapping;
634 ssize_t ret = move_from_pipe(inode, out, len, flags, pipe_to_file);
635
636 /*
637 * if file or inode is SYNC and we actually wrote some data, sync it
638 */
639 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host))
640 && ret > 0) {
641 struct inode *inode = mapping->host;
642 int err;
643
644 mutex_lock(&inode->i_mutex);
645 err = generic_osync_inode(mapping->host, mapping,
646 OSYNC_METADATA|OSYNC_DATA);
647 mutex_unlock(&inode->i_mutex);
648
649 if (err)
650 ret = err;
651 }
652
653 return ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200654}
655
Jens Axboe83f91352006-04-02 23:05:09 +0200656/**
657 * generic_splice_sendpage - splice data from a pipe to a socket
658 * @inode: pipe inode
659 * @out: socket to write to
660 * @len: number of bytes to splice
661 * @flags: splice modifier flags
662 *
663 * Will send @len bytes from the pipe to a network socket. No data copying
664 * is involved.
665 *
666 */
Jens Axboe5274f052006-03-30 15:15:30 +0200667ssize_t generic_splice_sendpage(struct inode *inode, struct file *out,
668 size_t len, unsigned int flags)
669{
670 return move_from_pipe(inode, out, len, flags, pipe_to_sendpage);
671}
672
Jeff Garzika0f06782006-03-30 23:06:13 -0500673EXPORT_SYMBOL(generic_file_splice_write);
674EXPORT_SYMBOL(generic_file_splice_read);
675
Jens Axboe83f91352006-04-02 23:05:09 +0200676/*
677 * Attempt to initiate a splice from pipe to file.
678 */
Jens Axboe5274f052006-03-30 15:15:30 +0200679static long do_splice_from(struct inode *pipe, struct file *out, size_t len,
680 unsigned int flags)
681{
682 loff_t pos;
683 int ret;
684
685 if (!out->f_op || !out->f_op->splice_write)
686 return -EINVAL;
687
688 if (!(out->f_mode & FMODE_WRITE))
689 return -EBADF;
690
691 pos = out->f_pos;
692 ret = rw_verify_area(WRITE, out, &pos, len);
693 if (unlikely(ret < 0))
694 return ret;
695
696 return out->f_op->splice_write(pipe, out, len, flags);
697}
698
Jens Axboe83f91352006-04-02 23:05:09 +0200699/*
700 * Attempt to initiate a splice from a file to a pipe.
701 */
Jens Axboe5274f052006-03-30 15:15:30 +0200702static long do_splice_to(struct file *in, struct inode *pipe, size_t len,
703 unsigned int flags)
704{
705 loff_t pos, isize, left;
706 int ret;
707
708 if (!in->f_op || !in->f_op->splice_read)
709 return -EINVAL;
710
711 if (!(in->f_mode & FMODE_READ))
712 return -EBADF;
713
714 pos = in->f_pos;
715 ret = rw_verify_area(READ, in, &pos, len);
716 if (unlikely(ret < 0))
717 return ret;
718
719 isize = i_size_read(in->f_mapping->host);
720 if (unlikely(in->f_pos >= isize))
721 return 0;
722
723 left = isize - in->f_pos;
724 if (left < len)
725 len = left;
726
727 return in->f_op->splice_read(in, pipe, len, flags);
728}
729
Jens Axboe83f91352006-04-02 23:05:09 +0200730/*
731 * Determine where to splice to/from.
732 */
Jens Axboe5274f052006-03-30 15:15:30 +0200733static long do_splice(struct file *in, struct file *out, size_t len,
734 unsigned int flags)
735{
736 struct inode *pipe;
737
738 pipe = in->f_dentry->d_inode;
739 if (pipe->i_pipe)
740 return do_splice_from(pipe, out, len, flags);
741
742 pipe = out->f_dentry->d_inode;
743 if (pipe->i_pipe)
744 return do_splice_to(in, pipe, len, flags);
745
746 return -EINVAL;
747}
748
749asmlinkage long sys_splice(int fdin, int fdout, size_t len, unsigned int flags)
750{
751 long error;
752 struct file *in, *out;
753 int fput_in, fput_out;
754
755 if (unlikely(!len))
756 return 0;
757
758 error = -EBADF;
759 in = fget_light(fdin, &fput_in);
760 if (in) {
761 if (in->f_mode & FMODE_READ) {
762 out = fget_light(fdout, &fput_out);
763 if (out) {
764 if (out->f_mode & FMODE_WRITE)
765 error = do_splice(in, out, len, flags);
766 fput_light(out, fput_out);
767 }
768 }
769
770 fput_light(in, fput_in);
771 }
772
773 return error;
774}