Commit | Line | Data |
---|---|---|
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 | */ | |
33 | struct 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 |
46 | static 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 | ||
70 | if (PageLRU(page)) { | |
71 | struct zone *zone = page_zone(page); | |
72 | ||
73 | spin_lock_irq(&zone->lru_lock); | |
74 | BUG_ON(!PageLRU(page)); | |
75 | __ClearPageLRU(page); | |
76 | del_page_from_lru(zone, page); | |
77 | spin_unlock_irq(&zone->lru_lock); | |
78 | } | |
79 | ||
5abc97aa JA |
80 | return 0; |
81 | } | |
82 | ||
5274f052 JA |
83 | static void page_cache_pipe_buf_release(struct pipe_inode_info *info, |
84 | struct pipe_buffer *buf) | |
85 | { | |
86 | page_cache_release(buf->page); | |
87 | buf->page = NULL; | |
88 | } | |
89 | ||
90 | static void *page_cache_pipe_buf_map(struct file *file, | |
91 | struct pipe_inode_info *info, | |
92 | struct pipe_buffer *buf) | |
93 | { | |
94 | struct page *page = buf->page; | |
95 | ||
96 | lock_page(page); | |
97 | ||
98 | if (!PageUptodate(page)) { | |
99 | unlock_page(page); | |
100 | return ERR_PTR(-EIO); | |
101 | } | |
102 | ||
103 | if (!page->mapping) { | |
104 | unlock_page(page); | |
105 | return ERR_PTR(-ENODATA); | |
106 | } | |
107 | ||
108 | return kmap(buf->page); | |
109 | } | |
110 | ||
111 | static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info, | |
112 | struct pipe_buffer *buf) | |
113 | { | |
4f6f0bd2 | 114 | unlock_page(buf->page); |
5274f052 JA |
115 | kunmap(buf->page); |
116 | } | |
117 | ||
118 | static struct pipe_buf_operations page_cache_pipe_buf_ops = { | |
119 | .can_merge = 0, | |
120 | .map = page_cache_pipe_buf_map, | |
121 | .unmap = page_cache_pipe_buf_unmap, | |
122 | .release = page_cache_pipe_buf_release, | |
5abc97aa | 123 | .steal = page_cache_pipe_buf_steal, |
5274f052 JA |
124 | }; |
125 | ||
83f9135b JA |
126 | /* |
127 | * Pipe output worker. This sets up our pipe format with the page cache | |
128 | * pipe buffer operations. Otherwise very similar to the regular pipe_writev(). | |
129 | */ | |
5274f052 JA |
130 | static ssize_t move_to_pipe(struct inode *inode, struct page **pages, |
131 | int nr_pages, unsigned long offset, | |
29e35094 | 132 | unsigned long len, unsigned int flags) |
5274f052 JA |
133 | { |
134 | struct pipe_inode_info *info; | |
135 | int ret, do_wakeup, i; | |
136 | ||
137 | ret = 0; | |
138 | do_wakeup = 0; | |
139 | i = 0; | |
140 | ||
141 | mutex_lock(PIPE_MUTEX(*inode)); | |
142 | ||
143 | info = inode->i_pipe; | |
144 | for (;;) { | |
145 | int bufs; | |
146 | ||
147 | if (!PIPE_READERS(*inode)) { | |
148 | send_sig(SIGPIPE, current, 0); | |
149 | if (!ret) | |
150 | ret = -EPIPE; | |
151 | break; | |
152 | } | |
153 | ||
154 | bufs = info->nrbufs; | |
155 | if (bufs < PIPE_BUFFERS) { | |
156 | int newbuf = (info->curbuf + bufs) & (PIPE_BUFFERS - 1); | |
157 | struct pipe_buffer *buf = info->bufs + newbuf; | |
158 | struct page *page = pages[i++]; | |
159 | unsigned long this_len; | |
160 | ||
161 | this_len = PAGE_CACHE_SIZE - offset; | |
162 | if (this_len > len) | |
163 | this_len = len; | |
164 | ||
165 | buf->page = page; | |
166 | buf->offset = offset; | |
167 | buf->len = this_len; | |
168 | buf->ops = &page_cache_pipe_buf_ops; | |
169 | info->nrbufs = ++bufs; | |
170 | do_wakeup = 1; | |
171 | ||
172 | ret += this_len; | |
173 | len -= this_len; | |
174 | offset = 0; | |
175 | if (!--nr_pages) | |
176 | break; | |
177 | if (!len) | |
178 | break; | |
179 | if (bufs < PIPE_BUFFERS) | |
180 | continue; | |
181 | ||
182 | break; | |
183 | } | |
184 | ||
29e35094 LT |
185 | if (flags & SPLICE_F_NONBLOCK) { |
186 | if (!ret) | |
187 | ret = -EAGAIN; | |
188 | break; | |
189 | } | |
190 | ||
5274f052 JA |
191 | if (signal_pending(current)) { |
192 | if (!ret) | |
193 | ret = -ERESTARTSYS; | |
194 | break; | |
195 | } | |
196 | ||
197 | if (do_wakeup) { | |
198 | wake_up_interruptible_sync(PIPE_WAIT(*inode)); | |
199 | kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, | |
200 | POLL_IN); | |
201 | do_wakeup = 0; | |
202 | } | |
203 | ||
204 | PIPE_WAITING_WRITERS(*inode)++; | |
205 | pipe_wait(inode); | |
206 | PIPE_WAITING_WRITERS(*inode)--; | |
207 | } | |
208 | ||
209 | mutex_unlock(PIPE_MUTEX(*inode)); | |
210 | ||
211 | if (do_wakeup) { | |
212 | wake_up_interruptible(PIPE_WAIT(*inode)); | |
213 | kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN); | |
214 | } | |
215 | ||
216 | while (i < nr_pages) | |
217 | page_cache_release(pages[i++]); | |
218 | ||
219 | return ret; | |
220 | } | |
221 | ||
222 | static int __generic_file_splice_read(struct file *in, struct inode *pipe, | |
29e35094 | 223 | size_t len, unsigned int flags) |
5274f052 JA |
224 | { |
225 | struct address_space *mapping = in->f_mapping; | |
226 | unsigned int offset, nr_pages; | |
227 | struct page *pages[PIPE_BUFFERS], *shadow[PIPE_BUFFERS]; | |
228 | struct page *page; | |
229 | pgoff_t index, pidx; | |
230 | int i, j; | |
231 | ||
232 | index = in->f_pos >> PAGE_CACHE_SHIFT; | |
233 | offset = in->f_pos & ~PAGE_CACHE_MASK; | |
234 | nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
235 | ||
236 | if (nr_pages > PIPE_BUFFERS) | |
237 | nr_pages = PIPE_BUFFERS; | |
238 | ||
239 | /* | |
240 | * initiate read-ahead on this page range | |
241 | */ | |
242 | do_page_cache_readahead(mapping, in, index, nr_pages); | |
243 | ||
244 | /* | |
245 | * Get as many pages from the page cache as possible.. | |
246 | * Start IO on the page cache entries we create (we | |
247 | * can assume that any pre-existing ones we find have | |
248 | * already had IO started on them). | |
249 | */ | |
250 | i = find_get_pages(mapping, index, nr_pages, pages); | |
251 | ||
252 | /* | |
253 | * common case - we found all pages and they are contiguous, | |
254 | * kick them off | |
255 | */ | |
256 | if (i && (pages[i - 1]->index == index + i - 1)) | |
257 | goto splice_them; | |
258 | ||
259 | /* | |
260 | * fill shadow[] with pages at the right locations, so we only | |
261 | * have to fill holes | |
262 | */ | |
53cd9ae8 JA |
263 | memset(shadow, 0, nr_pages * sizeof(struct page *)); |
264 | for (j = 0; j < i; j++) | |
265 | shadow[pages[j]->index - index] = pages[j]; | |
5274f052 JA |
266 | |
267 | /* | |
268 | * now fill in the holes | |
269 | */ | |
270 | for (i = 0, pidx = index; i < nr_pages; pidx++, i++) { | |
271 | int error; | |
272 | ||
273 | if (shadow[i]) | |
274 | continue; | |
275 | ||
276 | /* | |
277 | * no page there, look one up / create it | |
278 | */ | |
279 | page = find_or_create_page(mapping, pidx, | |
280 | mapping_gfp_mask(mapping)); | |
281 | if (!page) | |
282 | break; | |
283 | ||
284 | if (PageUptodate(page)) | |
285 | unlock_page(page); | |
286 | else { | |
287 | error = mapping->a_ops->readpage(in, page); | |
288 | ||
289 | if (unlikely(error)) { | |
290 | page_cache_release(page); | |
291 | break; | |
292 | } | |
293 | } | |
294 | shadow[i] = page; | |
295 | } | |
296 | ||
297 | if (!i) { | |
298 | for (i = 0; i < nr_pages; i++) { | |
299 | if (shadow[i]) | |
300 | page_cache_release(shadow[i]); | |
301 | } | |
302 | return 0; | |
303 | } | |
304 | ||
305 | memcpy(pages, shadow, i * sizeof(struct page *)); | |
306 | ||
307 | /* | |
308 | * Now we splice them into the pipe.. | |
309 | */ | |
310 | splice_them: | |
29e35094 | 311 | return move_to_pipe(pipe, pages, i, offset, len, flags); |
5274f052 JA |
312 | } |
313 | ||
83f9135b JA |
314 | /** |
315 | * generic_file_splice_read - splice data from file to a pipe | |
316 | * @in: file to splice from | |
317 | * @pipe: pipe to splice to | |
318 | * @len: number of bytes to splice | |
319 | * @flags: splice modifier flags | |
320 | * | |
321 | * Will read pages from given file and fill them into a pipe. | |
322 | * | |
323 | */ | |
5274f052 JA |
324 | ssize_t generic_file_splice_read(struct file *in, struct inode *pipe, |
325 | size_t len, unsigned int flags) | |
326 | { | |
327 | ssize_t spliced; | |
328 | int ret; | |
329 | ||
330 | ret = 0; | |
331 | spliced = 0; | |
332 | while (len) { | |
29e35094 | 333 | ret = __generic_file_splice_read(in, pipe, len, flags); |
5274f052 JA |
334 | |
335 | if (ret <= 0) | |
336 | break; | |
337 | ||
338 | in->f_pos += ret; | |
339 | len -= ret; | |
340 | spliced += ret; | |
29e35094 LT |
341 | |
342 | if (!(flags & SPLICE_F_NONBLOCK)) | |
343 | continue; | |
344 | ret = -EAGAIN; | |
345 | break; | |
5274f052 JA |
346 | } |
347 | ||
348 | if (spliced) | |
349 | return spliced; | |
350 | ||
351 | return ret; | |
352 | } | |
353 | ||
059a8f37 JA |
354 | EXPORT_SYMBOL(generic_file_splice_read); |
355 | ||
5274f052 | 356 | /* |
4f6f0bd2 JA |
357 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
358 | * using sendpage(). | |
5274f052 JA |
359 | */ |
360 | static int pipe_to_sendpage(struct pipe_inode_info *info, | |
361 | struct pipe_buffer *buf, struct splice_desc *sd) | |
362 | { | |
363 | struct file *file = sd->file; | |
364 | loff_t pos = sd->pos; | |
365 | unsigned int offset; | |
366 | ssize_t ret; | |
367 | void *ptr; | |
b2b39fa4 | 368 | int more; |
5274f052 JA |
369 | |
370 | /* | |
371 | * sub-optimal, but we are limited by the pipe ->map. we don't | |
372 | * need a kmap'ed buffer here, we just want to make sure we | |
373 | * have the page pinned if the pipe page originates from the | |
374 | * page cache | |
375 | */ | |
376 | ptr = buf->ops->map(file, info, buf); | |
377 | if (IS_ERR(ptr)) | |
378 | return PTR_ERR(ptr); | |
379 | ||
380 | offset = pos & ~PAGE_CACHE_MASK; | |
b2b39fa4 | 381 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; |
5274f052 | 382 | |
b2b39fa4 | 383 | ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more); |
5274f052 JA |
384 | |
385 | buf->ops->unmap(info, buf); | |
386 | if (ret == sd->len) | |
387 | return 0; | |
388 | ||
389 | return -EIO; | |
390 | } | |
391 | ||
392 | /* | |
393 | * This is a little more tricky than the file -> pipe splicing. There are | |
394 | * basically three cases: | |
395 | * | |
396 | * - Destination page already exists in the address space and there | |
397 | * are users of it. For that case we have no other option that | |
398 | * copying the data. Tough luck. | |
399 | * - Destination page already exists in the address space, but there | |
400 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
401 | * through to last case. | |
402 | * - Destination page does not exist, we can add the pipe page to | |
403 | * the page cache and avoid the copy. | |
404 | * | |
83f9135b JA |
405 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
406 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
407 | * file address space page cache. This is possible if no one else has | |
408 | * the pipe page referenced outside of the pipe and page cache. If | |
409 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
410 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 JA |
411 | */ |
412 | static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf, | |
413 | struct splice_desc *sd) | |
414 | { | |
415 | struct file *file = sd->file; | |
416 | struct address_space *mapping = file->f_mapping; | |
417 | unsigned int offset; | |
418 | struct page *page; | |
5274f052 | 419 | pgoff_t index; |
5abc97aa | 420 | char *src; |
4f6f0bd2 | 421 | int ret, stolen; |
5274f052 JA |
422 | |
423 | /* | |
424 | * after this, page will be locked and unmapped | |
425 | */ | |
426 | src = buf->ops->map(file, info, buf); | |
427 | if (IS_ERR(src)) | |
428 | return PTR_ERR(src); | |
429 | ||
430 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
431 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
4f6f0bd2 | 432 | stolen = 0; |
5274f052 | 433 | |
5274f052 | 434 | /* |
5abc97aa | 435 | * reuse buf page, if SPLICE_F_MOVE is set |
5274f052 | 436 | */ |
5abc97aa | 437 | if (sd->flags & SPLICE_F_MOVE) { |
83f9135b JA |
438 | /* |
439 | * If steal succeeds, buf->page is now pruned from the vm | |
440 | * side (LRU and page cache) and we can reuse it. | |
441 | */ | |
5abc97aa JA |
442 | if (buf->ops->steal(info, buf)) |
443 | goto find_page; | |
444 | ||
445 | page = buf->page; | |
4f6f0bd2 | 446 | stolen = 1; |
5abc97aa JA |
447 | if (add_to_page_cache_lru(page, mapping, index, |
448 | mapping_gfp_mask(mapping))) | |
449 | goto find_page; | |
450 | } else { | |
451 | find_page: | |
452 | ret = -ENOMEM; | |
453 | page = find_or_create_page(mapping, index, | |
454 | mapping_gfp_mask(mapping)); | |
455 | if (!page) | |
456 | goto out; | |
457 | ||
458 | /* | |
459 | * If the page is uptodate, it is also locked. If it isn't | |
460 | * uptodate, we can mark it uptodate if we are filling the | |
461 | * full page. Otherwise we need to read it in first... | |
462 | */ | |
463 | if (!PageUptodate(page)) { | |
464 | if (sd->len < PAGE_CACHE_SIZE) { | |
465 | ret = mapping->a_ops->readpage(file, page); | |
466 | if (unlikely(ret)) | |
467 | goto out; | |
468 | ||
469 | lock_page(page); | |
470 | ||
471 | if (!PageUptodate(page)) { | |
472 | /* | |
473 | * page got invalidated, repeat | |
474 | */ | |
475 | if (!page->mapping) { | |
476 | unlock_page(page); | |
477 | page_cache_release(page); | |
478 | goto find_page; | |
479 | } | |
480 | ret = -EIO; | |
481 | goto out; | |
5274f052 | 482 | } |
5abc97aa JA |
483 | } else { |
484 | WARN_ON(!PageLocked(page)); | |
485 | SetPageUptodate(page); | |
5274f052 | 486 | } |
5274f052 JA |
487 | } |
488 | } | |
489 | ||
490 | ret = mapping->a_ops->prepare_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
491 | if (ret == AOP_TRUNCATED_PAGE) { |
492 | page_cache_release(page); | |
493 | goto find_page; | |
494 | } else if (ret) | |
5274f052 JA |
495 | goto out; |
496 | ||
4f6f0bd2 | 497 | if (!stolen) { |
5abc97aa JA |
498 | char *dst = kmap_atomic(page, KM_USER0); |
499 | ||
500 | memcpy(dst + offset, src + buf->offset, sd->len); | |
501 | flush_dcache_page(page); | |
502 | kunmap_atomic(dst, KM_USER0); | |
503 | } | |
5274f052 JA |
504 | |
505 | ret = mapping->a_ops->commit_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
506 | if (ret == AOP_TRUNCATED_PAGE) { |
507 | page_cache_release(page); | |
508 | goto find_page; | |
509 | } else if (ret) | |
5274f052 JA |
510 | goto out; |
511 | ||
4f6f0bd2 | 512 | balance_dirty_pages_ratelimited(mapping); |
5274f052 | 513 | out: |
4f6f0bd2 | 514 | if (!stolen) { |
5abc97aa | 515 | page_cache_release(page); |
4f6f0bd2 JA |
516 | unlock_page(page); |
517 | } | |
5274f052 JA |
518 | buf->ops->unmap(info, buf); |
519 | return ret; | |
520 | } | |
521 | ||
522 | typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *, | |
523 | struct splice_desc *); | |
524 | ||
83f9135b JA |
525 | /* |
526 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
527 | * key here is the 'actor' worker passed in that actually moves the data | |
528 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
529 | */ | |
5274f052 JA |
530 | static ssize_t move_from_pipe(struct inode *inode, struct file *out, |
531 | size_t len, unsigned int flags, | |
532 | splice_actor *actor) | |
533 | { | |
534 | struct pipe_inode_info *info; | |
535 | int ret, do_wakeup, err; | |
536 | struct splice_desc sd; | |
537 | ||
538 | ret = 0; | |
539 | do_wakeup = 0; | |
540 | ||
541 | sd.total_len = len; | |
542 | sd.flags = flags; | |
543 | sd.file = out; | |
544 | sd.pos = out->f_pos; | |
545 | ||
546 | mutex_lock(PIPE_MUTEX(*inode)); | |
547 | ||
548 | info = inode->i_pipe; | |
549 | for (;;) { | |
550 | int bufs = info->nrbufs; | |
551 | ||
552 | if (bufs) { | |
553 | int curbuf = info->curbuf; | |
554 | struct pipe_buffer *buf = info->bufs + curbuf; | |
555 | struct pipe_buf_operations *ops = buf->ops; | |
556 | ||
557 | sd.len = buf->len; | |
558 | if (sd.len > sd.total_len) | |
559 | sd.len = sd.total_len; | |
560 | ||
561 | err = actor(info, buf, &sd); | |
562 | if (err) { | |
563 | if (!ret && err != -ENODATA) | |
564 | ret = err; | |
565 | ||
566 | break; | |
567 | } | |
568 | ||
569 | ret += sd.len; | |
570 | buf->offset += sd.len; | |
571 | buf->len -= sd.len; | |
572 | if (!buf->len) { | |
573 | buf->ops = NULL; | |
574 | ops->release(info, buf); | |
575 | curbuf = (curbuf + 1) & (PIPE_BUFFERS - 1); | |
576 | info->curbuf = curbuf; | |
577 | info->nrbufs = --bufs; | |
578 | do_wakeup = 1; | |
579 | } | |
580 | ||
581 | sd.pos += sd.len; | |
582 | sd.total_len -= sd.len; | |
583 | if (!sd.total_len) | |
584 | break; | |
585 | } | |
586 | ||
587 | if (bufs) | |
588 | continue; | |
589 | if (!PIPE_WRITERS(*inode)) | |
590 | break; | |
591 | if (!PIPE_WAITING_WRITERS(*inode)) { | |
592 | if (ret) | |
593 | break; | |
594 | } | |
595 | ||
29e35094 LT |
596 | if (flags & SPLICE_F_NONBLOCK) { |
597 | if (!ret) | |
598 | ret = -EAGAIN; | |
599 | break; | |
600 | } | |
601 | ||
5274f052 JA |
602 | if (signal_pending(current)) { |
603 | if (!ret) | |
604 | ret = -ERESTARTSYS; | |
605 | break; | |
606 | } | |
607 | ||
608 | if (do_wakeup) { | |
609 | wake_up_interruptible_sync(PIPE_WAIT(*inode)); | |
610 | kill_fasync(PIPE_FASYNC_WRITERS(*inode),SIGIO,POLL_OUT); | |
611 | do_wakeup = 0; | |
612 | } | |
613 | ||
614 | pipe_wait(inode); | |
615 | } | |
616 | ||
617 | mutex_unlock(PIPE_MUTEX(*inode)); | |
618 | ||
619 | if (do_wakeup) { | |
620 | wake_up_interruptible(PIPE_WAIT(*inode)); | |
621 | kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT); | |
622 | } | |
623 | ||
624 | mutex_lock(&out->f_mapping->host->i_mutex); | |
625 | out->f_pos = sd.pos; | |
626 | mutex_unlock(&out->f_mapping->host->i_mutex); | |
627 | return ret; | |
628 | ||
629 | } | |
630 | ||
83f9135b JA |
631 | /** |
632 | * generic_file_splice_write - splice data from a pipe to a file | |
633 | * @inode: pipe inode | |
634 | * @out: file to write to | |
635 | * @len: number of bytes to splice | |
636 | * @flags: splice modifier flags | |
637 | * | |
638 | * Will either move or copy pages (determined by @flags options) from | |
639 | * the given pipe inode to the given file. | |
640 | * | |
641 | */ | |
5274f052 JA |
642 | ssize_t generic_file_splice_write(struct inode *inode, struct file *out, |
643 | size_t len, unsigned int flags) | |
644 | { | |
4f6f0bd2 JA |
645 | struct address_space *mapping = out->f_mapping; |
646 | ssize_t ret = move_from_pipe(inode, out, len, flags, pipe_to_file); | |
647 | ||
648 | /* | |
649 | * if file or inode is SYNC and we actually wrote some data, sync it | |
650 | */ | |
651 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host)) | |
652 | && ret > 0) { | |
653 | struct inode *inode = mapping->host; | |
654 | int err; | |
655 | ||
656 | mutex_lock(&inode->i_mutex); | |
657 | err = generic_osync_inode(mapping->host, mapping, | |
658 | OSYNC_METADATA|OSYNC_DATA); | |
659 | mutex_unlock(&inode->i_mutex); | |
660 | ||
661 | if (err) | |
662 | ret = err; | |
663 | } | |
664 | ||
665 | return ret; | |
5274f052 JA |
666 | } |
667 | ||
059a8f37 JA |
668 | EXPORT_SYMBOL(generic_file_splice_write); |
669 | ||
83f9135b JA |
670 | /** |
671 | * generic_splice_sendpage - splice data from a pipe to a socket | |
672 | * @inode: pipe inode | |
673 | * @out: socket to write to | |
674 | * @len: number of bytes to splice | |
675 | * @flags: splice modifier flags | |
676 | * | |
677 | * Will send @len bytes from the pipe to a network socket. No data copying | |
678 | * is involved. | |
679 | * | |
680 | */ | |
5274f052 JA |
681 | ssize_t generic_splice_sendpage(struct inode *inode, struct file *out, |
682 | size_t len, unsigned int flags) | |
683 | { | |
684 | return move_from_pipe(inode, out, len, flags, pipe_to_sendpage); | |
685 | } | |
686 | ||
059a8f37 | 687 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 688 | |
83f9135b JA |
689 | /* |
690 | * Attempt to initiate a splice from pipe to file. | |
691 | */ | |
5274f052 JA |
692 | static long do_splice_from(struct inode *pipe, struct file *out, size_t len, |
693 | unsigned int flags) | |
694 | { | |
695 | loff_t pos; | |
696 | int ret; | |
697 | ||
698 | if (!out->f_op || !out->f_op->splice_write) | |
699 | return -EINVAL; | |
700 | ||
701 | if (!(out->f_mode & FMODE_WRITE)) | |
702 | return -EBADF; | |
703 | ||
704 | pos = out->f_pos; | |
705 | ret = rw_verify_area(WRITE, out, &pos, len); | |
706 | if (unlikely(ret < 0)) | |
707 | return ret; | |
708 | ||
709 | return out->f_op->splice_write(pipe, out, len, flags); | |
710 | } | |
711 | ||
83f9135b JA |
712 | /* |
713 | * Attempt to initiate a splice from a file to a pipe. | |
714 | */ | |
5274f052 JA |
715 | static long do_splice_to(struct file *in, struct inode *pipe, size_t len, |
716 | unsigned int flags) | |
717 | { | |
718 | loff_t pos, isize, left; | |
719 | int ret; | |
720 | ||
721 | if (!in->f_op || !in->f_op->splice_read) | |
722 | return -EINVAL; | |
723 | ||
724 | if (!(in->f_mode & FMODE_READ)) | |
725 | return -EBADF; | |
726 | ||
727 | pos = in->f_pos; | |
728 | ret = rw_verify_area(READ, in, &pos, len); | |
729 | if (unlikely(ret < 0)) | |
730 | return ret; | |
731 | ||
732 | isize = i_size_read(in->f_mapping->host); | |
733 | if (unlikely(in->f_pos >= isize)) | |
734 | return 0; | |
735 | ||
736 | left = isize - in->f_pos; | |
737 | if (left < len) | |
738 | len = left; | |
739 | ||
740 | return in->f_op->splice_read(in, pipe, len, flags); | |
741 | } | |
742 | ||
83f9135b JA |
743 | /* |
744 | * Determine where to splice to/from. | |
745 | */ | |
5274f052 JA |
746 | static long do_splice(struct file *in, struct file *out, size_t len, |
747 | unsigned int flags) | |
748 | { | |
749 | struct inode *pipe; | |
750 | ||
751 | pipe = in->f_dentry->d_inode; | |
752 | if (pipe->i_pipe) | |
753 | return do_splice_from(pipe, out, len, flags); | |
754 | ||
755 | pipe = out->f_dentry->d_inode; | |
756 | if (pipe->i_pipe) | |
757 | return do_splice_to(in, pipe, len, flags); | |
758 | ||
759 | return -EINVAL; | |
760 | } | |
761 | ||
762 | asmlinkage long sys_splice(int fdin, int fdout, size_t len, unsigned int flags) | |
763 | { | |
764 | long error; | |
765 | struct file *in, *out; | |
766 | int fput_in, fput_out; | |
767 | ||
768 | if (unlikely(!len)) | |
769 | return 0; | |
770 | ||
771 | error = -EBADF; | |
772 | in = fget_light(fdin, &fput_in); | |
773 | if (in) { | |
774 | if (in->f_mode & FMODE_READ) { | |
775 | out = fget_light(fdout, &fput_out); | |
776 | if (out) { | |
777 | if (out->f_mode & FMODE_WRITE) | |
778 | error = do_splice(in, out, len, flags); | |
779 | fput_light(out, fput_out); | |
780 | } | |
781 | } | |
782 | ||
783 | fput_light(in, fput_in); | |
784 | } | |
785 | ||
786 | return error; | |
787 | } |