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 | ||
3e7ee3e7 | 70 | buf->flags |= PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU; |
5abc97aa JA |
71 | return 0; |
72 | } | |
73 | ||
5274f052 JA |
74 | static 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 | ||
82 | static 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); |
116 | error: | |
117 | unlock_page(page); | |
118 | return ERR_PTR(err); | |
5274f052 JA |
119 | } |
120 | ||
121 | static 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 | ||
127 | static 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 | */ | |
5274f052 JA |
139 | static ssize_t move_to_pipe(struct inode *inode, struct page **pages, |
140 | int nr_pages, unsigned long offset, | |
29e35094 | 141 | unsigned long len, unsigned int flags) |
5274f052 JA |
142 | { |
143 | struct pipe_inode_info *info; | |
144 | int ret, do_wakeup, i; | |
145 | ||
146 | ret = 0; | |
147 | do_wakeup = 0; | |
148 | i = 0; | |
149 | ||
150 | mutex_lock(PIPE_MUTEX(*inode)); | |
151 | ||
152 | info = inode->i_pipe; | |
153 | for (;;) { | |
154 | int bufs; | |
155 | ||
156 | if (!PIPE_READERS(*inode)) { | |
157 | send_sig(SIGPIPE, current, 0); | |
158 | if (!ret) | |
159 | ret = -EPIPE; | |
160 | break; | |
161 | } | |
162 | ||
163 | bufs = info->nrbufs; | |
164 | if (bufs < PIPE_BUFFERS) { | |
165 | int newbuf = (info->curbuf + bufs) & (PIPE_BUFFERS - 1); | |
166 | struct pipe_buffer *buf = info->bufs + newbuf; | |
167 | struct page *page = pages[i++]; | |
168 | unsigned long this_len; | |
169 | ||
170 | this_len = PAGE_CACHE_SIZE - offset; | |
171 | if (this_len > len) | |
172 | this_len = len; | |
173 | ||
174 | buf->page = page; | |
175 | buf->offset = offset; | |
176 | buf->len = this_len; | |
177 | buf->ops = &page_cache_pipe_buf_ops; | |
178 | info->nrbufs = ++bufs; | |
179 | do_wakeup = 1; | |
180 | ||
181 | ret += this_len; | |
182 | len -= this_len; | |
183 | offset = 0; | |
184 | if (!--nr_pages) | |
185 | break; | |
186 | if (!len) | |
187 | break; | |
188 | if (bufs < PIPE_BUFFERS) | |
189 | continue; | |
190 | ||
191 | break; | |
192 | } | |
193 | ||
29e35094 LT |
194 | if (flags & SPLICE_F_NONBLOCK) { |
195 | if (!ret) | |
196 | ret = -EAGAIN; | |
197 | break; | |
198 | } | |
199 | ||
5274f052 JA |
200 | if (signal_pending(current)) { |
201 | if (!ret) | |
202 | ret = -ERESTARTSYS; | |
203 | break; | |
204 | } | |
205 | ||
206 | if (do_wakeup) { | |
c0bd1f65 JA |
207 | smp_mb(); |
208 | if (waitqueue_active(PIPE_WAIT(*inode))) | |
209 | wake_up_interruptible_sync(PIPE_WAIT(*inode)); | |
5274f052 JA |
210 | kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, |
211 | POLL_IN); | |
212 | do_wakeup = 0; | |
213 | } | |
214 | ||
215 | PIPE_WAITING_WRITERS(*inode)++; | |
216 | pipe_wait(inode); | |
217 | PIPE_WAITING_WRITERS(*inode)--; | |
218 | } | |
219 | ||
220 | mutex_unlock(PIPE_MUTEX(*inode)); | |
221 | ||
222 | if (do_wakeup) { | |
c0bd1f65 JA |
223 | smp_mb(); |
224 | if (waitqueue_active(PIPE_WAIT(*inode))) | |
225 | wake_up_interruptible(PIPE_WAIT(*inode)); | |
5274f052 JA |
226 | kill_fasync(PIPE_FASYNC_READERS(*inode), SIGIO, POLL_IN); |
227 | } | |
228 | ||
229 | while (i < nr_pages) | |
230 | page_cache_release(pages[i++]); | |
231 | ||
232 | return ret; | |
233 | } | |
234 | ||
235 | static int __generic_file_splice_read(struct file *in, struct inode *pipe, | |
29e35094 | 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 | /* | |
253 | * initiate read-ahead on this page range | |
254 | */ | |
255 | do_page_cache_readahead(mapping, in, index, nr_pages); | |
256 | ||
5274f052 JA |
257 | /* |
258 | * now fill in the holes | |
259 | */ | |
16c523dd | 260 | for (i = 0; i < nr_pages; i++, index++) { |
5274f052 JA |
261 | /* |
262 | * no page there, look one up / create it | |
263 | */ | |
16c523dd | 264 | page = find_or_create_page(mapping, index, |
5274f052 JA |
265 | mapping_gfp_mask(mapping)); |
266 | if (!page) | |
267 | break; | |
268 | ||
269 | if (PageUptodate(page)) | |
270 | unlock_page(page); | |
271 | else { | |
16c523dd | 272 | int error = mapping->a_ops->readpage(in, page); |
5274f052 JA |
273 | |
274 | if (unlikely(error)) { | |
275 | page_cache_release(page); | |
276 | break; | |
277 | } | |
278 | } | |
16c523dd | 279 | pages[i] = page; |
5274f052 JA |
280 | } |
281 | ||
16c523dd JA |
282 | if (i) |
283 | return move_to_pipe(pipe, pages, i, offset, len, flags); | |
5274f052 | 284 | |
16c523dd | 285 | return 0; |
5274f052 JA |
286 | } |
287 | ||
83f9135b JA |
288 | /** |
289 | * generic_file_splice_read - splice data from file to a pipe | |
290 | * @in: file to splice from | |
291 | * @pipe: pipe to splice to | |
292 | * @len: number of bytes to splice | |
293 | * @flags: splice modifier flags | |
294 | * | |
295 | * Will read pages from given file and fill them into a pipe. | |
296 | * | |
297 | */ | |
5274f052 JA |
298 | ssize_t generic_file_splice_read(struct file *in, struct inode *pipe, |
299 | size_t len, unsigned int flags) | |
300 | { | |
301 | ssize_t spliced; | |
302 | int ret; | |
303 | ||
304 | ret = 0; | |
305 | spliced = 0; | |
306 | while (len) { | |
29e35094 | 307 | ret = __generic_file_splice_read(in, pipe, len, flags); |
5274f052 JA |
308 | |
309 | if (ret <= 0) | |
310 | break; | |
311 | ||
312 | in->f_pos += ret; | |
313 | len -= ret; | |
314 | spliced += ret; | |
29e35094 LT |
315 | |
316 | if (!(flags & SPLICE_F_NONBLOCK)) | |
317 | continue; | |
318 | ret = -EAGAIN; | |
319 | break; | |
5274f052 JA |
320 | } |
321 | ||
322 | if (spliced) | |
323 | return spliced; | |
324 | ||
325 | return ret; | |
326 | } | |
327 | ||
059a8f37 JA |
328 | EXPORT_SYMBOL(generic_file_splice_read); |
329 | ||
5274f052 | 330 | /* |
4f6f0bd2 JA |
331 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
332 | * using sendpage(). | |
5274f052 JA |
333 | */ |
334 | static int pipe_to_sendpage(struct pipe_inode_info *info, | |
335 | struct pipe_buffer *buf, struct splice_desc *sd) | |
336 | { | |
337 | struct file *file = sd->file; | |
338 | loff_t pos = sd->pos; | |
339 | unsigned int offset; | |
340 | ssize_t ret; | |
341 | void *ptr; | |
b2b39fa4 | 342 | int more; |
5274f052 JA |
343 | |
344 | /* | |
345 | * sub-optimal, but we are limited by the pipe ->map. we don't | |
346 | * need a kmap'ed buffer here, we just want to make sure we | |
347 | * have the page pinned if the pipe page originates from the | |
348 | * page cache | |
349 | */ | |
350 | ptr = buf->ops->map(file, info, buf); | |
351 | if (IS_ERR(ptr)) | |
352 | return PTR_ERR(ptr); | |
353 | ||
354 | offset = pos & ~PAGE_CACHE_MASK; | |
b2b39fa4 | 355 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; |
5274f052 | 356 | |
b2b39fa4 | 357 | ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more); |
5274f052 JA |
358 | |
359 | buf->ops->unmap(info, buf); | |
360 | if (ret == sd->len) | |
361 | return 0; | |
362 | ||
363 | return -EIO; | |
364 | } | |
365 | ||
366 | /* | |
367 | * This is a little more tricky than the file -> pipe splicing. There are | |
368 | * basically three cases: | |
369 | * | |
370 | * - Destination page already exists in the address space and there | |
371 | * are users of it. For that case we have no other option that | |
372 | * copying the data. Tough luck. | |
373 | * - Destination page already exists in the address space, but there | |
374 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
375 | * through to last case. | |
376 | * - Destination page does not exist, we can add the pipe page to | |
377 | * the page cache and avoid the copy. | |
378 | * | |
83f9135b JA |
379 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
380 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
381 | * file address space page cache. This is possible if no one else has | |
382 | * the pipe page referenced outside of the pipe and page cache. If | |
383 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
384 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 JA |
385 | */ |
386 | static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf, | |
387 | struct splice_desc *sd) | |
388 | { | |
389 | struct file *file = sd->file; | |
390 | struct address_space *mapping = file->f_mapping; | |
3e7ee3e7 | 391 | gfp_t gfp_mask = mapping_gfp_mask(mapping); |
5274f052 JA |
392 | unsigned int offset; |
393 | struct page *page; | |
5274f052 | 394 | pgoff_t index; |
5abc97aa | 395 | char *src; |
3e7ee3e7 | 396 | int ret; |
5274f052 JA |
397 | |
398 | /* | |
49d0b21b | 399 | * make sure the data in this buffer is uptodate |
5274f052 JA |
400 | */ |
401 | src = buf->ops->map(file, info, buf); | |
402 | if (IS_ERR(src)) | |
403 | return PTR_ERR(src); | |
404 | ||
405 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
406 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
407 | ||
5274f052 | 408 | /* |
5abc97aa | 409 | * reuse buf page, if SPLICE_F_MOVE is set |
5274f052 | 410 | */ |
5abc97aa | 411 | if (sd->flags & SPLICE_F_MOVE) { |
83f9135b JA |
412 | /* |
413 | * If steal succeeds, buf->page is now pruned from the vm | |
414 | * side (LRU and page cache) and we can reuse it. | |
415 | */ | |
5abc97aa JA |
416 | if (buf->ops->steal(info, buf)) |
417 | goto find_page; | |
418 | ||
49d0b21b JA |
419 | /* |
420 | * this will also set the page locked | |
421 | */ | |
5abc97aa | 422 | page = buf->page; |
3e7ee3e7 | 423 | if (add_to_page_cache(page, mapping, index, gfp_mask)) |
5abc97aa | 424 | goto find_page; |
3e7ee3e7 JA |
425 | |
426 | if (!(buf->flags & PIPE_BUF_FLAG_LRU)) | |
427 | lru_cache_add(page); | |
5abc97aa JA |
428 | } else { |
429 | find_page: | |
430 | ret = -ENOMEM; | |
3e7ee3e7 | 431 | page = find_or_create_page(mapping, index, gfp_mask); |
5abc97aa | 432 | if (!page) |
9aefe431 | 433 | goto out_nomem; |
5abc97aa JA |
434 | |
435 | /* | |
436 | * If the page is uptodate, it is also locked. If it isn't | |
437 | * uptodate, we can mark it uptodate if we are filling the | |
438 | * full page. Otherwise we need to read it in first... | |
439 | */ | |
440 | if (!PageUptodate(page)) { | |
441 | if (sd->len < PAGE_CACHE_SIZE) { | |
442 | ret = mapping->a_ops->readpage(file, page); | |
443 | if (unlikely(ret)) | |
444 | goto out; | |
445 | ||
446 | lock_page(page); | |
447 | ||
448 | if (!PageUptodate(page)) { | |
449 | /* | |
450 | * page got invalidated, repeat | |
451 | */ | |
452 | if (!page->mapping) { | |
453 | unlock_page(page); | |
454 | page_cache_release(page); | |
455 | goto find_page; | |
456 | } | |
457 | ret = -EIO; | |
458 | goto out; | |
5274f052 | 459 | } |
5abc97aa JA |
460 | } else { |
461 | WARN_ON(!PageLocked(page)); | |
462 | SetPageUptodate(page); | |
5274f052 | 463 | } |
5274f052 JA |
464 | } |
465 | } | |
466 | ||
467 | ret = mapping->a_ops->prepare_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
468 | if (ret == AOP_TRUNCATED_PAGE) { |
469 | page_cache_release(page); | |
470 | goto find_page; | |
471 | } else if (ret) | |
5274f052 JA |
472 | goto out; |
473 | ||
3e7ee3e7 | 474 | if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) { |
5abc97aa JA |
475 | char *dst = kmap_atomic(page, KM_USER0); |
476 | ||
477 | memcpy(dst + offset, src + buf->offset, sd->len); | |
478 | flush_dcache_page(page); | |
479 | kunmap_atomic(dst, KM_USER0); | |
480 | } | |
5274f052 JA |
481 | |
482 | ret = mapping->a_ops->commit_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
483 | if (ret == AOP_TRUNCATED_PAGE) { |
484 | page_cache_release(page); | |
485 | goto find_page; | |
486 | } else if (ret) | |
5274f052 JA |
487 | goto out; |
488 | ||
c7f21e4f | 489 | mark_page_accessed(page); |
4f6f0bd2 | 490 | balance_dirty_pages_ratelimited(mapping); |
5274f052 | 491 | out: |
3e7ee3e7 | 492 | if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) { |
5abc97aa | 493 | page_cache_release(page); |
4f6f0bd2 JA |
494 | unlock_page(page); |
495 | } | |
9aefe431 | 496 | out_nomem: |
5274f052 JA |
497 | buf->ops->unmap(info, buf); |
498 | return ret; | |
499 | } | |
500 | ||
501 | typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *, | |
502 | struct splice_desc *); | |
503 | ||
83f9135b JA |
504 | /* |
505 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
506 | * key here is the 'actor' worker passed in that actually moves the data | |
507 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
508 | */ | |
5274f052 JA |
509 | static ssize_t move_from_pipe(struct inode *inode, struct file *out, |
510 | size_t len, unsigned int flags, | |
511 | splice_actor *actor) | |
512 | { | |
513 | struct pipe_inode_info *info; | |
514 | int ret, do_wakeup, err; | |
515 | struct splice_desc sd; | |
516 | ||
517 | ret = 0; | |
518 | do_wakeup = 0; | |
519 | ||
520 | sd.total_len = len; | |
521 | sd.flags = flags; | |
522 | sd.file = out; | |
523 | sd.pos = out->f_pos; | |
524 | ||
525 | mutex_lock(PIPE_MUTEX(*inode)); | |
526 | ||
527 | info = inode->i_pipe; | |
528 | for (;;) { | |
529 | int bufs = info->nrbufs; | |
530 | ||
531 | if (bufs) { | |
532 | int curbuf = info->curbuf; | |
533 | struct pipe_buffer *buf = info->bufs + curbuf; | |
534 | struct pipe_buf_operations *ops = buf->ops; | |
535 | ||
536 | sd.len = buf->len; | |
537 | if (sd.len > sd.total_len) | |
538 | sd.len = sd.total_len; | |
539 | ||
540 | err = actor(info, buf, &sd); | |
541 | if (err) { | |
542 | if (!ret && err != -ENODATA) | |
543 | ret = err; | |
544 | ||
545 | break; | |
546 | } | |
547 | ||
548 | ret += sd.len; | |
549 | buf->offset += sd.len; | |
550 | buf->len -= sd.len; | |
551 | if (!buf->len) { | |
552 | buf->ops = NULL; | |
553 | ops->release(info, buf); | |
554 | curbuf = (curbuf + 1) & (PIPE_BUFFERS - 1); | |
555 | info->curbuf = curbuf; | |
556 | info->nrbufs = --bufs; | |
557 | do_wakeup = 1; | |
558 | } | |
559 | ||
560 | sd.pos += sd.len; | |
561 | sd.total_len -= sd.len; | |
562 | if (!sd.total_len) | |
563 | break; | |
564 | } | |
565 | ||
566 | if (bufs) | |
567 | continue; | |
568 | if (!PIPE_WRITERS(*inode)) | |
569 | break; | |
570 | if (!PIPE_WAITING_WRITERS(*inode)) { | |
571 | if (ret) | |
572 | break; | |
573 | } | |
574 | ||
29e35094 LT |
575 | if (flags & SPLICE_F_NONBLOCK) { |
576 | if (!ret) | |
577 | ret = -EAGAIN; | |
578 | break; | |
579 | } | |
580 | ||
5274f052 JA |
581 | if (signal_pending(current)) { |
582 | if (!ret) | |
583 | ret = -ERESTARTSYS; | |
584 | break; | |
585 | } | |
586 | ||
587 | if (do_wakeup) { | |
c0bd1f65 JA |
588 | smp_mb(); |
589 | if (waitqueue_active(PIPE_WAIT(*inode))) | |
590 | wake_up_interruptible_sync(PIPE_WAIT(*inode)); | |
5274f052 JA |
591 | kill_fasync(PIPE_FASYNC_WRITERS(*inode),SIGIO,POLL_OUT); |
592 | do_wakeup = 0; | |
593 | } | |
594 | ||
595 | pipe_wait(inode); | |
596 | } | |
597 | ||
598 | mutex_unlock(PIPE_MUTEX(*inode)); | |
599 | ||
600 | if (do_wakeup) { | |
c0bd1f65 JA |
601 | smp_mb(); |
602 | if (waitqueue_active(PIPE_WAIT(*inode))) | |
603 | wake_up_interruptible(PIPE_WAIT(*inode)); | |
5274f052 JA |
604 | kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT); |
605 | } | |
606 | ||
607 | mutex_lock(&out->f_mapping->host->i_mutex); | |
608 | out->f_pos = sd.pos; | |
609 | mutex_unlock(&out->f_mapping->host->i_mutex); | |
610 | return ret; | |
611 | ||
612 | } | |
613 | ||
83f9135b JA |
614 | /** |
615 | * generic_file_splice_write - splice data from a pipe to a file | |
616 | * @inode: pipe inode | |
617 | * @out: file to write to | |
618 | * @len: number of bytes to splice | |
619 | * @flags: splice modifier flags | |
620 | * | |
621 | * Will either move or copy pages (determined by @flags options) from | |
622 | * the given pipe inode to the given file. | |
623 | * | |
624 | */ | |
5274f052 JA |
625 | ssize_t generic_file_splice_write(struct inode *inode, struct file *out, |
626 | size_t len, unsigned int flags) | |
627 | { | |
4f6f0bd2 JA |
628 | struct address_space *mapping = out->f_mapping; |
629 | ssize_t ret = move_from_pipe(inode, out, len, flags, pipe_to_file); | |
630 | ||
631 | /* | |
632 | * if file or inode is SYNC and we actually wrote some data, sync it | |
633 | */ | |
634 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host)) | |
635 | && ret > 0) { | |
636 | struct inode *inode = mapping->host; | |
637 | int err; | |
638 | ||
639 | mutex_lock(&inode->i_mutex); | |
640 | err = generic_osync_inode(mapping->host, mapping, | |
641 | OSYNC_METADATA|OSYNC_DATA); | |
642 | mutex_unlock(&inode->i_mutex); | |
643 | ||
644 | if (err) | |
645 | ret = err; | |
646 | } | |
647 | ||
648 | return ret; | |
5274f052 JA |
649 | } |
650 | ||
059a8f37 JA |
651 | EXPORT_SYMBOL(generic_file_splice_write); |
652 | ||
83f9135b JA |
653 | /** |
654 | * generic_splice_sendpage - splice data from a pipe to a socket | |
655 | * @inode: pipe inode | |
656 | * @out: socket to write to | |
657 | * @len: number of bytes to splice | |
658 | * @flags: splice modifier flags | |
659 | * | |
660 | * Will send @len bytes from the pipe to a network socket. No data copying | |
661 | * is involved. | |
662 | * | |
663 | */ | |
5274f052 JA |
664 | ssize_t generic_splice_sendpage(struct inode *inode, struct file *out, |
665 | size_t len, unsigned int flags) | |
666 | { | |
667 | return move_from_pipe(inode, out, len, flags, pipe_to_sendpage); | |
668 | } | |
669 | ||
059a8f37 | 670 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 671 | |
83f9135b JA |
672 | /* |
673 | * Attempt to initiate a splice from pipe to file. | |
674 | */ | |
5274f052 JA |
675 | static long do_splice_from(struct inode *pipe, struct file *out, size_t len, |
676 | unsigned int flags) | |
677 | { | |
678 | loff_t pos; | |
679 | int ret; | |
680 | ||
681 | if (!out->f_op || !out->f_op->splice_write) | |
682 | return -EINVAL; | |
683 | ||
684 | if (!(out->f_mode & FMODE_WRITE)) | |
685 | return -EBADF; | |
686 | ||
687 | pos = out->f_pos; | |
688 | ret = rw_verify_area(WRITE, out, &pos, len); | |
689 | if (unlikely(ret < 0)) | |
690 | return ret; | |
691 | ||
692 | return out->f_op->splice_write(pipe, out, len, flags); | |
693 | } | |
694 | ||
83f9135b JA |
695 | /* |
696 | * Attempt to initiate a splice from a file to a pipe. | |
697 | */ | |
5274f052 JA |
698 | static long do_splice_to(struct file *in, struct inode *pipe, size_t len, |
699 | unsigned int flags) | |
700 | { | |
701 | loff_t pos, isize, left; | |
702 | int ret; | |
703 | ||
704 | if (!in->f_op || !in->f_op->splice_read) | |
705 | return -EINVAL; | |
706 | ||
707 | if (!(in->f_mode & FMODE_READ)) | |
708 | return -EBADF; | |
709 | ||
710 | pos = in->f_pos; | |
711 | ret = rw_verify_area(READ, in, &pos, len); | |
712 | if (unlikely(ret < 0)) | |
713 | return ret; | |
714 | ||
715 | isize = i_size_read(in->f_mapping->host); | |
716 | if (unlikely(in->f_pos >= isize)) | |
717 | return 0; | |
718 | ||
719 | left = isize - in->f_pos; | |
720 | if (left < len) | |
721 | len = left; | |
722 | ||
723 | return in->f_op->splice_read(in, pipe, len, flags); | |
724 | } | |
725 | ||
83f9135b JA |
726 | /* |
727 | * Determine where to splice to/from. | |
728 | */ | |
5274f052 JA |
729 | static long do_splice(struct file *in, struct file *out, size_t len, |
730 | unsigned int flags) | |
731 | { | |
732 | struct inode *pipe; | |
733 | ||
734 | pipe = in->f_dentry->d_inode; | |
735 | if (pipe->i_pipe) | |
736 | return do_splice_from(pipe, out, len, flags); | |
737 | ||
738 | pipe = out->f_dentry->d_inode; | |
739 | if (pipe->i_pipe) | |
740 | return do_splice_to(in, pipe, len, flags); | |
741 | ||
742 | return -EINVAL; | |
743 | } | |
744 | ||
745 | asmlinkage long sys_splice(int fdin, int fdout, size_t len, unsigned int flags) | |
746 | { | |
747 | long error; | |
748 | struct file *in, *out; | |
749 | int fput_in, fput_out; | |
750 | ||
751 | if (unlikely(!len)) | |
752 | return 0; | |
753 | ||
754 | error = -EBADF; | |
755 | in = fget_light(fdin, &fput_in); | |
756 | if (in) { | |
757 | if (in->f_mode & FMODE_READ) { | |
758 | out = fget_light(fdout, &fput_out); | |
759 | if (out) { | |
760 | if (out->f_mode & FMODE_WRITE) | |
761 | error = do_splice(in, out, len, flags); | |
762 | fput_light(out, fput_out); | |
763 | } | |
764 | } | |
765 | ||
766 | fput_light(in, fput_in); | |
767 | } | |
768 | ||
769 | return error; | |
770 | } |