]>
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 | /* | |
0b749ce3 JA |
253 | * initiate read-ahead on this page range. however, don't call into |
254 | * read-ahead if this is a non-zero offset (we are likely doing small | |
255 | * chunk splice and the page is already there) for a single page. | |
5274f052 | 256 | */ |
0b749ce3 JA |
257 | if (!offset || nr_pages > 1) |
258 | do_page_cache_readahead(mapping, in, index, nr_pages); | |
5274f052 | 259 | |
5274f052 JA |
260 | /* |
261 | * now fill in the holes | |
262 | */ | |
16c523dd | 263 | for (i = 0; i < nr_pages; i++, index++) { |
5274f052 JA |
264 | /* |
265 | * no page there, look one up / create it | |
266 | */ | |
16c523dd | 267 | page = find_or_create_page(mapping, index, |
5274f052 JA |
268 | mapping_gfp_mask(mapping)); |
269 | if (!page) | |
270 | break; | |
271 | ||
272 | if (PageUptodate(page)) | |
273 | unlock_page(page); | |
274 | else { | |
16c523dd | 275 | int error = mapping->a_ops->readpage(in, page); |
5274f052 JA |
276 | |
277 | if (unlikely(error)) { | |
278 | page_cache_release(page); | |
279 | break; | |
280 | } | |
281 | } | |
16c523dd | 282 | pages[i] = page; |
5274f052 JA |
283 | } |
284 | ||
16c523dd JA |
285 | if (i) |
286 | return move_to_pipe(pipe, pages, i, offset, len, flags); | |
5274f052 | 287 | |
16c523dd | 288 | return 0; |
5274f052 JA |
289 | } |
290 | ||
83f9135b JA |
291 | /** |
292 | * generic_file_splice_read - splice data from file to a pipe | |
293 | * @in: file to splice from | |
294 | * @pipe: pipe to splice to | |
295 | * @len: number of bytes to splice | |
296 | * @flags: splice modifier flags | |
297 | * | |
298 | * Will read pages from given file and fill them into a pipe. | |
299 | * | |
300 | */ | |
5274f052 JA |
301 | ssize_t generic_file_splice_read(struct file *in, struct inode *pipe, |
302 | size_t len, unsigned int flags) | |
303 | { | |
304 | ssize_t spliced; | |
305 | int ret; | |
306 | ||
307 | ret = 0; | |
308 | spliced = 0; | |
309 | while (len) { | |
29e35094 | 310 | ret = __generic_file_splice_read(in, pipe, len, flags); |
5274f052 JA |
311 | |
312 | if (ret <= 0) | |
313 | break; | |
314 | ||
315 | in->f_pos += ret; | |
316 | len -= ret; | |
317 | spliced += ret; | |
29e35094 LT |
318 | |
319 | if (!(flags & SPLICE_F_NONBLOCK)) | |
320 | continue; | |
321 | ret = -EAGAIN; | |
322 | break; | |
5274f052 JA |
323 | } |
324 | ||
325 | if (spliced) | |
326 | return spliced; | |
327 | ||
328 | return ret; | |
329 | } | |
330 | ||
059a8f37 JA |
331 | EXPORT_SYMBOL(generic_file_splice_read); |
332 | ||
5274f052 | 333 | /* |
4f6f0bd2 JA |
334 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
335 | * using sendpage(). | |
5274f052 JA |
336 | */ |
337 | static int pipe_to_sendpage(struct pipe_inode_info *info, | |
338 | struct pipe_buffer *buf, struct splice_desc *sd) | |
339 | { | |
340 | struct file *file = sd->file; | |
341 | loff_t pos = sd->pos; | |
342 | unsigned int offset; | |
343 | ssize_t ret; | |
344 | void *ptr; | |
b2b39fa4 | 345 | int more; |
5274f052 JA |
346 | |
347 | /* | |
348 | * sub-optimal, but we are limited by the pipe ->map. we don't | |
349 | * need a kmap'ed buffer here, we just want to make sure we | |
350 | * have the page pinned if the pipe page originates from the | |
351 | * page cache | |
352 | */ | |
353 | ptr = buf->ops->map(file, info, buf); | |
354 | if (IS_ERR(ptr)) | |
355 | return PTR_ERR(ptr); | |
356 | ||
357 | offset = pos & ~PAGE_CACHE_MASK; | |
b2b39fa4 | 358 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; |
5274f052 | 359 | |
b2b39fa4 | 360 | ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more); |
5274f052 JA |
361 | |
362 | buf->ops->unmap(info, buf); | |
363 | if (ret == sd->len) | |
364 | return 0; | |
365 | ||
366 | return -EIO; | |
367 | } | |
368 | ||
369 | /* | |
370 | * This is a little more tricky than the file -> pipe splicing. There are | |
371 | * basically three cases: | |
372 | * | |
373 | * - Destination page already exists in the address space and there | |
374 | * are users of it. For that case we have no other option that | |
375 | * copying the data. Tough luck. | |
376 | * - Destination page already exists in the address space, but there | |
377 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
378 | * through to last case. | |
379 | * - Destination page does not exist, we can add the pipe page to | |
380 | * the page cache and avoid the copy. | |
381 | * | |
83f9135b JA |
382 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
383 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
384 | * file address space page cache. This is possible if no one else has | |
385 | * the pipe page referenced outside of the pipe and page cache. If | |
386 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
387 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 JA |
388 | */ |
389 | static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf, | |
390 | struct splice_desc *sd) | |
391 | { | |
392 | struct file *file = sd->file; | |
393 | struct address_space *mapping = file->f_mapping; | |
3e7ee3e7 | 394 | gfp_t gfp_mask = mapping_gfp_mask(mapping); |
5274f052 JA |
395 | unsigned int offset; |
396 | struct page *page; | |
5274f052 | 397 | pgoff_t index; |
5abc97aa | 398 | char *src; |
3e7ee3e7 | 399 | int ret; |
5274f052 JA |
400 | |
401 | /* | |
49d0b21b | 402 | * make sure the data in this buffer is uptodate |
5274f052 JA |
403 | */ |
404 | src = buf->ops->map(file, info, buf); | |
405 | if (IS_ERR(src)) | |
406 | return PTR_ERR(src); | |
407 | ||
408 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
409 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
410 | ||
5274f052 | 411 | /* |
5abc97aa | 412 | * reuse buf page, if SPLICE_F_MOVE is set |
5274f052 | 413 | */ |
5abc97aa | 414 | if (sd->flags & SPLICE_F_MOVE) { |
83f9135b JA |
415 | /* |
416 | * If steal succeeds, buf->page is now pruned from the vm | |
417 | * side (LRU and page cache) and we can reuse it. | |
418 | */ | |
5abc97aa JA |
419 | if (buf->ops->steal(info, buf)) |
420 | goto find_page; | |
421 | ||
49d0b21b JA |
422 | /* |
423 | * this will also set the page locked | |
424 | */ | |
5abc97aa | 425 | page = buf->page; |
3e7ee3e7 | 426 | if (add_to_page_cache(page, mapping, index, gfp_mask)) |
5abc97aa | 427 | goto find_page; |
3e7ee3e7 JA |
428 | |
429 | if (!(buf->flags & PIPE_BUF_FLAG_LRU)) | |
430 | lru_cache_add(page); | |
5abc97aa JA |
431 | } else { |
432 | find_page: | |
433 | ret = -ENOMEM; | |
3e7ee3e7 | 434 | page = find_or_create_page(mapping, index, gfp_mask); |
5abc97aa | 435 | if (!page) |
9aefe431 | 436 | goto out_nomem; |
5abc97aa JA |
437 | |
438 | /* | |
439 | * If the page is uptodate, it is also locked. If it isn't | |
440 | * uptodate, we can mark it uptodate if we are filling the | |
441 | * full page. Otherwise we need to read it in first... | |
442 | */ | |
443 | if (!PageUptodate(page)) { | |
444 | if (sd->len < PAGE_CACHE_SIZE) { | |
445 | ret = mapping->a_ops->readpage(file, page); | |
446 | if (unlikely(ret)) | |
447 | goto out; | |
448 | ||
449 | lock_page(page); | |
450 | ||
451 | if (!PageUptodate(page)) { | |
452 | /* | |
453 | * page got invalidated, repeat | |
454 | */ | |
455 | if (!page->mapping) { | |
456 | unlock_page(page); | |
457 | page_cache_release(page); | |
458 | goto find_page; | |
459 | } | |
460 | ret = -EIO; | |
461 | goto out; | |
5274f052 | 462 | } |
5abc97aa JA |
463 | } else { |
464 | WARN_ON(!PageLocked(page)); | |
465 | SetPageUptodate(page); | |
5274f052 | 466 | } |
5274f052 JA |
467 | } |
468 | } | |
469 | ||
470 | ret = mapping->a_ops->prepare_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
471 | if (ret == AOP_TRUNCATED_PAGE) { |
472 | page_cache_release(page); | |
473 | goto find_page; | |
474 | } else if (ret) | |
5274f052 JA |
475 | goto out; |
476 | ||
3e7ee3e7 | 477 | if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) { |
5abc97aa JA |
478 | char *dst = kmap_atomic(page, KM_USER0); |
479 | ||
480 | memcpy(dst + offset, src + buf->offset, sd->len); | |
481 | flush_dcache_page(page); | |
482 | kunmap_atomic(dst, KM_USER0); | |
483 | } | |
5274f052 JA |
484 | |
485 | ret = mapping->a_ops->commit_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
486 | if (ret == AOP_TRUNCATED_PAGE) { |
487 | page_cache_release(page); | |
488 | goto find_page; | |
489 | } else if (ret) | |
5274f052 JA |
490 | goto out; |
491 | ||
c7f21e4f | 492 | mark_page_accessed(page); |
4f6f0bd2 | 493 | balance_dirty_pages_ratelimited(mapping); |
5274f052 | 494 | out: |
3e7ee3e7 | 495 | if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) { |
5abc97aa | 496 | page_cache_release(page); |
4f6f0bd2 JA |
497 | unlock_page(page); |
498 | } | |
9aefe431 | 499 | out_nomem: |
5274f052 JA |
500 | buf->ops->unmap(info, buf); |
501 | return ret; | |
502 | } | |
503 | ||
504 | typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *, | |
505 | struct splice_desc *); | |
506 | ||
83f9135b JA |
507 | /* |
508 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
509 | * key here is the 'actor' worker passed in that actually moves the data | |
510 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
511 | */ | |
5274f052 JA |
512 | static ssize_t move_from_pipe(struct inode *inode, struct file *out, |
513 | size_t len, unsigned int flags, | |
514 | splice_actor *actor) | |
515 | { | |
516 | struct pipe_inode_info *info; | |
517 | int ret, do_wakeup, err; | |
518 | struct splice_desc sd; | |
519 | ||
520 | ret = 0; | |
521 | do_wakeup = 0; | |
522 | ||
523 | sd.total_len = len; | |
524 | sd.flags = flags; | |
525 | sd.file = out; | |
526 | sd.pos = out->f_pos; | |
527 | ||
528 | mutex_lock(PIPE_MUTEX(*inode)); | |
529 | ||
530 | info = inode->i_pipe; | |
531 | for (;;) { | |
532 | int bufs = info->nrbufs; | |
533 | ||
534 | if (bufs) { | |
535 | int curbuf = info->curbuf; | |
536 | struct pipe_buffer *buf = info->bufs + curbuf; | |
537 | struct pipe_buf_operations *ops = buf->ops; | |
538 | ||
539 | sd.len = buf->len; | |
540 | if (sd.len > sd.total_len) | |
541 | sd.len = sd.total_len; | |
542 | ||
543 | err = actor(info, buf, &sd); | |
544 | if (err) { | |
545 | if (!ret && err != -ENODATA) | |
546 | ret = err; | |
547 | ||
548 | break; | |
549 | } | |
550 | ||
551 | ret += sd.len; | |
552 | buf->offset += sd.len; | |
553 | buf->len -= sd.len; | |
554 | if (!buf->len) { | |
555 | buf->ops = NULL; | |
556 | ops->release(info, buf); | |
557 | curbuf = (curbuf + 1) & (PIPE_BUFFERS - 1); | |
558 | info->curbuf = curbuf; | |
559 | info->nrbufs = --bufs; | |
560 | do_wakeup = 1; | |
561 | } | |
562 | ||
563 | sd.pos += sd.len; | |
564 | sd.total_len -= sd.len; | |
565 | if (!sd.total_len) | |
566 | break; | |
567 | } | |
568 | ||
569 | if (bufs) | |
570 | continue; | |
571 | if (!PIPE_WRITERS(*inode)) | |
572 | break; | |
573 | if (!PIPE_WAITING_WRITERS(*inode)) { | |
574 | if (ret) | |
575 | break; | |
576 | } | |
577 | ||
29e35094 LT |
578 | if (flags & SPLICE_F_NONBLOCK) { |
579 | if (!ret) | |
580 | ret = -EAGAIN; | |
581 | break; | |
582 | } | |
583 | ||
5274f052 JA |
584 | if (signal_pending(current)) { |
585 | if (!ret) | |
586 | ret = -ERESTARTSYS; | |
587 | break; | |
588 | } | |
589 | ||
590 | if (do_wakeup) { | |
c0bd1f65 JA |
591 | smp_mb(); |
592 | if (waitqueue_active(PIPE_WAIT(*inode))) | |
593 | wake_up_interruptible_sync(PIPE_WAIT(*inode)); | |
5274f052 JA |
594 | kill_fasync(PIPE_FASYNC_WRITERS(*inode),SIGIO,POLL_OUT); |
595 | do_wakeup = 0; | |
596 | } | |
597 | ||
598 | pipe_wait(inode); | |
599 | } | |
600 | ||
601 | mutex_unlock(PIPE_MUTEX(*inode)); | |
602 | ||
603 | if (do_wakeup) { | |
c0bd1f65 JA |
604 | smp_mb(); |
605 | if (waitqueue_active(PIPE_WAIT(*inode))) | |
606 | wake_up_interruptible(PIPE_WAIT(*inode)); | |
5274f052 JA |
607 | kill_fasync(PIPE_FASYNC_WRITERS(*inode), SIGIO, POLL_OUT); |
608 | } | |
609 | ||
610 | mutex_lock(&out->f_mapping->host->i_mutex); | |
611 | out->f_pos = sd.pos; | |
612 | mutex_unlock(&out->f_mapping->host->i_mutex); | |
613 | return ret; | |
614 | ||
615 | } | |
616 | ||
83f9135b JA |
617 | /** |
618 | * generic_file_splice_write - splice data from a pipe to a file | |
619 | * @inode: pipe inode | |
620 | * @out: file to write to | |
621 | * @len: number of bytes to splice | |
622 | * @flags: splice modifier flags | |
623 | * | |
624 | * Will either move or copy pages (determined by @flags options) from | |
625 | * the given pipe inode to the given file. | |
626 | * | |
627 | */ | |
5274f052 JA |
628 | ssize_t generic_file_splice_write(struct inode *inode, struct file *out, |
629 | size_t len, unsigned int flags) | |
630 | { | |
4f6f0bd2 JA |
631 | struct address_space *mapping = out->f_mapping; |
632 | ssize_t ret = move_from_pipe(inode, out, len, flags, pipe_to_file); | |
633 | ||
634 | /* | |
635 | * if file or inode is SYNC and we actually wrote some data, sync it | |
636 | */ | |
637 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host)) | |
638 | && ret > 0) { | |
639 | struct inode *inode = mapping->host; | |
640 | int err; | |
641 | ||
642 | mutex_lock(&inode->i_mutex); | |
643 | err = generic_osync_inode(mapping->host, mapping, | |
644 | OSYNC_METADATA|OSYNC_DATA); | |
645 | mutex_unlock(&inode->i_mutex); | |
646 | ||
647 | if (err) | |
648 | ret = err; | |
649 | } | |
650 | ||
651 | return ret; | |
5274f052 JA |
652 | } |
653 | ||
059a8f37 JA |
654 | EXPORT_SYMBOL(generic_file_splice_write); |
655 | ||
83f9135b JA |
656 | /** |
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 | */ | |
5274f052 JA |
667 | ssize_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 | ||
059a8f37 | 673 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 674 | |
83f9135b JA |
675 | /* |
676 | * Attempt to initiate a splice from pipe to file. | |
677 | */ | |
5274f052 JA |
678 | static long do_splice_from(struct inode *pipe, struct file *out, size_t len, |
679 | unsigned int flags) | |
680 | { | |
681 | loff_t pos; | |
682 | int ret; | |
683 | ||
684 | if (!out->f_op || !out->f_op->splice_write) | |
685 | return -EINVAL; | |
686 | ||
687 | if (!(out->f_mode & FMODE_WRITE)) | |
688 | return -EBADF; | |
689 | ||
690 | pos = out->f_pos; | |
691 | ret = rw_verify_area(WRITE, out, &pos, len); | |
692 | if (unlikely(ret < 0)) | |
693 | return ret; | |
694 | ||
695 | return out->f_op->splice_write(pipe, out, len, flags); | |
696 | } | |
697 | ||
83f9135b JA |
698 | /* |
699 | * Attempt to initiate a splice from a file to a pipe. | |
700 | */ | |
5274f052 JA |
701 | static long do_splice_to(struct file *in, struct inode *pipe, size_t len, |
702 | unsigned int flags) | |
703 | { | |
704 | loff_t pos, isize, left; | |
705 | int ret; | |
706 | ||
707 | if (!in->f_op || !in->f_op->splice_read) | |
708 | return -EINVAL; | |
709 | ||
710 | if (!(in->f_mode & FMODE_READ)) | |
711 | return -EBADF; | |
712 | ||
713 | pos = in->f_pos; | |
714 | ret = rw_verify_area(READ, in, &pos, len); | |
715 | if (unlikely(ret < 0)) | |
716 | return ret; | |
717 | ||
718 | isize = i_size_read(in->f_mapping->host); | |
719 | if (unlikely(in->f_pos >= isize)) | |
720 | return 0; | |
721 | ||
722 | left = isize - in->f_pos; | |
723 | if (left < len) | |
724 | len = left; | |
725 | ||
726 | return in->f_op->splice_read(in, pipe, len, flags); | |
727 | } | |
728 | ||
83f9135b JA |
729 | /* |
730 | * Determine where to splice to/from. | |
731 | */ | |
5274f052 JA |
732 | static long do_splice(struct file *in, struct file *out, size_t len, |
733 | unsigned int flags) | |
734 | { | |
735 | struct inode *pipe; | |
736 | ||
737 | pipe = in->f_dentry->d_inode; | |
738 | if (pipe->i_pipe) | |
739 | return do_splice_from(pipe, out, len, flags); | |
740 | ||
741 | pipe = out->f_dentry->d_inode; | |
742 | if (pipe->i_pipe) | |
743 | return do_splice_to(in, pipe, len, flags); | |
744 | ||
745 | return -EINVAL; | |
746 | } | |
747 | ||
748 | asmlinkage long sys_splice(int fdin, int fdout, size_t len, unsigned int flags) | |
749 | { | |
750 | long error; | |
751 | struct file *in, *out; | |
752 | int fput_in, fput_out; | |
753 | ||
754 | if (unlikely(!len)) | |
755 | return 0; | |
756 | ||
757 | error = -EBADF; | |
758 | in = fget_light(fdin, &fput_in); | |
759 | if (in) { | |
760 | if (in->f_mode & FMODE_READ) { | |
761 | out = fget_light(fdout, &fput_out); | |
762 | if (out) { | |
763 | if (out->f_mode & FMODE_WRITE) | |
764 | error = do_splice(in, out, len, flags); | |
765 | fput_light(out, fput_out); | |
766 | } | |
767 | } | |
768 | ||
769 | fput_light(in, fput_in); | |
770 | } | |
771 | ||
772 | return error; | |
773 | } |