]>
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 | |
c2058e06 JA |
12 | * Jens to support splicing to files, network, direct splicing, etc and |
13 | * fixing lots of bugs. | |
5274f052 | 14 | * |
c2058e06 JA |
15 | * Copyright (C) 2005-2006 Jens Axboe <axboe@suse.de> |
16 | * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> | |
17 | * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> | |
5274f052 JA |
18 | * |
19 | */ | |
20 | #include <linux/fs.h> | |
21 | #include <linux/file.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/pipe_fs_i.h> | |
24 | #include <linux/mm_inline.h> | |
5abc97aa | 25 | #include <linux/swap.h> |
4f6f0bd2 JA |
26 | #include <linux/writeback.h> |
27 | #include <linux/buffer_head.h> | |
a0f06780 | 28 | #include <linux/module.h> |
4f6f0bd2 | 29 | #include <linux/syscalls.h> |
5274f052 JA |
30 | |
31 | /* | |
32 | * Passed to the actors | |
33 | */ | |
34 | struct splice_desc { | |
35 | unsigned int len, total_len; /* current and remaining length */ | |
36 | unsigned int flags; /* splice flags */ | |
37 | struct file *file; /* file to read/write */ | |
38 | loff_t pos; /* file position */ | |
39 | }; | |
40 | ||
83f9135b JA |
41 | /* |
42 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
43 | * a vm helper function, it's already simplified quite a bit by the | |
44 | * addition of remove_mapping(). If success is returned, the caller may | |
45 | * attempt to reuse this page for another destination. | |
46 | */ | |
5abc97aa JA |
47 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *info, |
48 | struct pipe_buffer *buf) | |
49 | { | |
50 | struct page *page = buf->page; | |
4f6f0bd2 | 51 | struct address_space *mapping = page_mapping(page); |
5abc97aa JA |
52 | |
53 | WARN_ON(!PageLocked(page)); | |
54 | WARN_ON(!PageUptodate(page)); | |
55 | ||
ad8d6f0a JA |
56 | /* |
57 | * At least for ext2 with nobh option, we need to wait on writeback | |
58 | * completing on this page, since we'll remove it from the pagecache. | |
59 | * Otherwise truncate wont wait on the page, allowing the disk | |
60 | * blocks to be reused by someone else before we actually wrote our | |
61 | * data to them. fs corruption ensues. | |
62 | */ | |
63 | wait_on_page_writeback(page); | |
64 | ||
4f6f0bd2 JA |
65 | if (PagePrivate(page)) |
66 | try_to_release_page(page, mapping_gfp_mask(mapping)); | |
67 | ||
68 | if (!remove_mapping(mapping, page)) | |
5abc97aa JA |
69 | return 1; |
70 | ||
3e7ee3e7 | 71 | buf->flags |= PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU; |
5abc97aa JA |
72 | return 0; |
73 | } | |
74 | ||
5274f052 JA |
75 | static void page_cache_pipe_buf_release(struct pipe_inode_info *info, |
76 | struct pipe_buffer *buf) | |
77 | { | |
78 | page_cache_release(buf->page); | |
79 | buf->page = NULL; | |
3e7ee3e7 | 80 | buf->flags &= ~(PIPE_BUF_FLAG_STOLEN | PIPE_BUF_FLAG_LRU); |
5274f052 JA |
81 | } |
82 | ||
83 | static void *page_cache_pipe_buf_map(struct file *file, | |
84 | struct pipe_inode_info *info, | |
85 | struct pipe_buffer *buf) | |
86 | { | |
87 | struct page *page = buf->page; | |
49d0b21b | 88 | int err; |
5274f052 JA |
89 | |
90 | if (!PageUptodate(page)) { | |
49d0b21b JA |
91 | lock_page(page); |
92 | ||
93 | /* | |
94 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 95 | * splice, if this is the first page. |
49d0b21b JA |
96 | */ |
97 | if (!page->mapping) { | |
98 | err = -ENODATA; | |
99 | goto error; | |
100 | } | |
5274f052 | 101 | |
49d0b21b | 102 | /* |
73d62d83 | 103 | * Uh oh, read-error from disk. |
49d0b21b JA |
104 | */ |
105 | if (!PageUptodate(page)) { | |
106 | err = -EIO; | |
107 | goto error; | |
108 | } | |
109 | ||
110 | /* | |
73d62d83 | 111 | * Page is ok afterall, fall through to mapping. |
49d0b21b | 112 | */ |
5274f052 | 113 | unlock_page(page); |
5274f052 JA |
114 | } |
115 | ||
49d0b21b JA |
116 | return kmap(page); |
117 | error: | |
118 | unlock_page(page); | |
119 | return ERR_PTR(err); | |
5274f052 JA |
120 | } |
121 | ||
122 | static void page_cache_pipe_buf_unmap(struct pipe_inode_info *info, | |
123 | struct pipe_buffer *buf) | |
124 | { | |
5274f052 JA |
125 | kunmap(buf->page); |
126 | } | |
127 | ||
70524490 JA |
128 | static void page_cache_pipe_buf_get(struct pipe_inode_info *info, |
129 | struct pipe_buffer *buf) | |
130 | { | |
131 | page_cache_get(buf->page); | |
132 | } | |
133 | ||
5274f052 JA |
134 | static struct pipe_buf_operations page_cache_pipe_buf_ops = { |
135 | .can_merge = 0, | |
136 | .map = page_cache_pipe_buf_map, | |
137 | .unmap = page_cache_pipe_buf_unmap, | |
138 | .release = page_cache_pipe_buf_release, | |
5abc97aa | 139 | .steal = page_cache_pipe_buf_steal, |
70524490 | 140 | .get = page_cache_pipe_buf_get, |
5274f052 JA |
141 | }; |
142 | ||
83f9135b JA |
143 | /* |
144 | * Pipe output worker. This sets up our pipe format with the page cache | |
145 | * pipe buffer operations. Otherwise very similar to the regular pipe_writev(). | |
146 | */ | |
3a326a2c | 147 | static ssize_t move_to_pipe(struct pipe_inode_info *pipe, struct page **pages, |
91ad66ef JA |
148 | int nr_pages, unsigned long len, |
149 | unsigned int offset, unsigned int flags) | |
5274f052 | 150 | { |
5274f052 JA |
151 | int ret, do_wakeup, i; |
152 | ||
153 | ret = 0; | |
154 | do_wakeup = 0; | |
155 | i = 0; | |
156 | ||
3a326a2c IM |
157 | if (pipe->inode) |
158 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 159 | |
5274f052 | 160 | for (;;) { |
3a326a2c | 161 | if (!pipe->readers) { |
5274f052 JA |
162 | send_sig(SIGPIPE, current, 0); |
163 | if (!ret) | |
164 | ret = -EPIPE; | |
165 | break; | |
166 | } | |
167 | ||
6f767b04 JA |
168 | if (pipe->nrbufs < PIPE_BUFFERS) { |
169 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 170 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 JA |
171 | struct page *page = pages[i++]; |
172 | unsigned long this_len; | |
173 | ||
174 | this_len = PAGE_CACHE_SIZE - offset; | |
175 | if (this_len > len) | |
176 | this_len = len; | |
177 | ||
178 | buf->page = page; | |
179 | buf->offset = offset; | |
180 | buf->len = this_len; | |
181 | buf->ops = &page_cache_pipe_buf_ops; | |
6f767b04 JA |
182 | pipe->nrbufs++; |
183 | if (pipe->inode) | |
184 | do_wakeup = 1; | |
5274f052 JA |
185 | |
186 | ret += this_len; | |
187 | len -= this_len; | |
188 | offset = 0; | |
189 | if (!--nr_pages) | |
190 | break; | |
191 | if (!len) | |
192 | break; | |
6f767b04 | 193 | if (pipe->nrbufs < PIPE_BUFFERS) |
5274f052 JA |
194 | continue; |
195 | ||
196 | break; | |
197 | } | |
198 | ||
29e35094 LT |
199 | if (flags & SPLICE_F_NONBLOCK) { |
200 | if (!ret) | |
201 | ret = -EAGAIN; | |
202 | break; | |
203 | } | |
204 | ||
5274f052 JA |
205 | if (signal_pending(current)) { |
206 | if (!ret) | |
207 | ret = -ERESTARTSYS; | |
208 | break; | |
209 | } | |
210 | ||
211 | if (do_wakeup) { | |
c0bd1f65 | 212 | smp_mb(); |
3a326a2c IM |
213 | if (waitqueue_active(&pipe->wait)) |
214 | wake_up_interruptible_sync(&pipe->wait); | |
215 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
216 | do_wakeup = 0; |
217 | } | |
218 | ||
3a326a2c IM |
219 | pipe->waiting_writers++; |
220 | pipe_wait(pipe); | |
221 | pipe->waiting_writers--; | |
5274f052 JA |
222 | } |
223 | ||
3a326a2c IM |
224 | if (pipe->inode) |
225 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
226 | |
227 | if (do_wakeup) { | |
c0bd1f65 | 228 | smp_mb(); |
3a326a2c IM |
229 | if (waitqueue_active(&pipe->wait)) |
230 | wake_up_interruptible(&pipe->wait); | |
231 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
232 | } |
233 | ||
234 | while (i < nr_pages) | |
235 | page_cache_release(pages[i++]); | |
236 | ||
237 | return ret; | |
238 | } | |
239 | ||
3a326a2c | 240 | static int |
cbb7e577 JA |
241 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
242 | struct pipe_inode_info *pipe, size_t len, | |
243 | unsigned int flags) | |
5274f052 JA |
244 | { |
245 | struct address_space *mapping = in->f_mapping; | |
91ad66ef | 246 | unsigned int loff, offset, nr_pages; |
16c523dd | 247 | struct page *pages[PIPE_BUFFERS]; |
5274f052 | 248 | struct page *page; |
91ad66ef JA |
249 | pgoff_t index, end_index; |
250 | loff_t isize; | |
251 | size_t bytes; | |
7480a904 | 252 | int i, error; |
5274f052 | 253 | |
cbb7e577 | 254 | index = *ppos >> PAGE_CACHE_SHIFT; |
91ad66ef | 255 | loff = offset = *ppos & ~PAGE_CACHE_MASK; |
5274f052 JA |
256 | nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
257 | ||
258 | if (nr_pages > PIPE_BUFFERS) | |
259 | nr_pages = PIPE_BUFFERS; | |
260 | ||
261 | /* | |
73d62d83 | 262 | * Initiate read-ahead on this page range. however, don't call into |
0b749ce3 JA |
263 | * read-ahead if this is a non-zero offset (we are likely doing small |
264 | * chunk splice and the page is already there) for a single page. | |
5274f052 | 265 | */ |
0b749ce3 JA |
266 | if (!offset || nr_pages > 1) |
267 | do_page_cache_readahead(mapping, in, index, nr_pages); | |
5274f052 | 268 | |
5274f052 | 269 | /* |
73d62d83 | 270 | * Now fill in the holes: |
5274f052 | 271 | */ |
7480a904 | 272 | error = 0; |
91ad66ef | 273 | bytes = 0; |
16c523dd | 274 | for (i = 0; i < nr_pages; i++, index++) { |
7480a904 | 275 | find_page: |
5274f052 | 276 | /* |
7480a904 | 277 | * lookup the page for this index |
5274f052 | 278 | */ |
7480a904 JA |
279 | page = find_get_page(mapping, index); |
280 | if (!page) { | |
7480a904 JA |
281 | /* |
282 | * page didn't exist, allocate one | |
283 | */ | |
284 | page = page_cache_alloc_cold(mapping); | |
285 | if (!page) | |
286 | break; | |
287 | ||
288 | error = add_to_page_cache_lru(page, mapping, index, | |
289 | mapping_gfp_mask(mapping)); | |
290 | if (unlikely(error)) { | |
291 | page_cache_release(page); | |
292 | break; | |
293 | } | |
294 | ||
295 | goto readpage; | |
296 | } | |
297 | ||
298 | /* | |
299 | * If the page isn't uptodate, we may need to start io on it | |
300 | */ | |
301 | if (!PageUptodate(page)) { | |
c4f895cb JA |
302 | /* |
303 | * If in nonblock mode then dont block on waiting | |
304 | * for an in-flight io page | |
305 | */ | |
306 | if (flags & SPLICE_F_NONBLOCK) | |
307 | break; | |
308 | ||
7480a904 JA |
309 | lock_page(page); |
310 | ||
311 | /* | |
312 | * page was truncated, stop here. if this isn't the | |
313 | * first page, we'll just complete what we already | |
314 | * added | |
315 | */ | |
316 | if (!page->mapping) { | |
317 | unlock_page(page); | |
318 | page_cache_release(page); | |
319 | break; | |
320 | } | |
321 | /* | |
322 | * page was already under io and is now done, great | |
323 | */ | |
324 | if (PageUptodate(page)) { | |
325 | unlock_page(page); | |
326 | goto fill_it; | |
327 | } | |
5274f052 | 328 | |
7480a904 JA |
329 | readpage: |
330 | /* | |
331 | * need to read in the page | |
332 | */ | |
333 | error = mapping->a_ops->readpage(in, page); | |
5274f052 JA |
334 | |
335 | if (unlikely(error)) { | |
336 | page_cache_release(page); | |
7480a904 JA |
337 | if (error == AOP_TRUNCATED_PAGE) |
338 | goto find_page; | |
5274f052 JA |
339 | break; |
340 | } | |
91ad66ef JA |
341 | |
342 | /* | |
343 | * i_size must be checked after ->readpage(). | |
344 | */ | |
345 | isize = i_size_read(mapping->host); | |
346 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
347 | if (unlikely(!isize || index > end_index)) { | |
348 | page_cache_release(page); | |
349 | break; | |
350 | } | |
351 | ||
352 | /* | |
353 | * if this is the last page, see if we need to shrink | |
354 | * the length and stop | |
355 | */ | |
356 | if (end_index == index) { | |
357 | loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK); | |
358 | if (bytes + loff > isize) { | |
359 | page_cache_release(page); | |
360 | break; | |
361 | } | |
362 | /* | |
363 | * force quit after adding this page | |
364 | */ | |
365 | nr_pages = i; | |
366 | } | |
5274f052 | 367 | } |
7480a904 | 368 | fill_it: |
16c523dd | 369 | pages[i] = page; |
91ad66ef JA |
370 | bytes += PAGE_CACHE_SIZE - loff; |
371 | loff = 0; | |
5274f052 JA |
372 | } |
373 | ||
16c523dd | 374 | if (i) |
91ad66ef | 375 | return move_to_pipe(pipe, pages, i, bytes, offset, flags); |
5274f052 | 376 | |
7480a904 | 377 | return error; |
5274f052 JA |
378 | } |
379 | ||
83f9135b JA |
380 | /** |
381 | * generic_file_splice_read - splice data from file to a pipe | |
382 | * @in: file to splice from | |
383 | * @pipe: pipe to splice to | |
384 | * @len: number of bytes to splice | |
385 | * @flags: splice modifier flags | |
386 | * | |
387 | * Will read pages from given file and fill them into a pipe. | |
83f9135b | 388 | */ |
cbb7e577 JA |
389 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
390 | struct pipe_inode_info *pipe, size_t len, | |
391 | unsigned int flags) | |
5274f052 JA |
392 | { |
393 | ssize_t spliced; | |
394 | int ret; | |
395 | ||
396 | ret = 0; | |
397 | spliced = 0; | |
3a326a2c | 398 | |
5274f052 | 399 | while (len) { |
cbb7e577 | 400 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 401 | |
c4f895cb | 402 | if (ret < 0) |
5274f052 | 403 | break; |
c4f895cb JA |
404 | else if (!ret) { |
405 | if (spliced) | |
406 | break; | |
407 | if (flags & SPLICE_F_NONBLOCK) { | |
408 | ret = -EAGAIN; | |
409 | break; | |
410 | } | |
411 | } | |
5274f052 | 412 | |
cbb7e577 | 413 | *ppos += ret; |
5274f052 JA |
414 | len -= ret; |
415 | spliced += ret; | |
416 | } | |
417 | ||
418 | if (spliced) | |
419 | return spliced; | |
420 | ||
421 | return ret; | |
422 | } | |
423 | ||
059a8f37 JA |
424 | EXPORT_SYMBOL(generic_file_splice_read); |
425 | ||
5274f052 | 426 | /* |
4f6f0bd2 JA |
427 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
428 | * using sendpage(). | |
5274f052 JA |
429 | */ |
430 | static int pipe_to_sendpage(struct pipe_inode_info *info, | |
431 | struct pipe_buffer *buf, struct splice_desc *sd) | |
432 | { | |
433 | struct file *file = sd->file; | |
434 | loff_t pos = sd->pos; | |
435 | unsigned int offset; | |
436 | ssize_t ret; | |
437 | void *ptr; | |
b2b39fa4 | 438 | int more; |
5274f052 JA |
439 | |
440 | /* | |
73d62d83 | 441 | * Sub-optimal, but we are limited by the pipe ->map. We don't |
5274f052 JA |
442 | * need a kmap'ed buffer here, we just want to make sure we |
443 | * have the page pinned if the pipe page originates from the | |
73d62d83 | 444 | * page cache. |
5274f052 JA |
445 | */ |
446 | ptr = buf->ops->map(file, info, buf); | |
447 | if (IS_ERR(ptr)) | |
448 | return PTR_ERR(ptr); | |
449 | ||
450 | offset = pos & ~PAGE_CACHE_MASK; | |
b2b39fa4 | 451 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; |
5274f052 | 452 | |
b2b39fa4 | 453 | ret = file->f_op->sendpage(file, buf->page, offset, sd->len, &pos,more); |
5274f052 JA |
454 | |
455 | buf->ops->unmap(info, buf); | |
456 | if (ret == sd->len) | |
457 | return 0; | |
458 | ||
459 | return -EIO; | |
460 | } | |
461 | ||
462 | /* | |
463 | * This is a little more tricky than the file -> pipe splicing. There are | |
464 | * basically three cases: | |
465 | * | |
466 | * - Destination page already exists in the address space and there | |
467 | * are users of it. For that case we have no other option that | |
468 | * copying the data. Tough luck. | |
469 | * - Destination page already exists in the address space, but there | |
470 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
471 | * through to last case. | |
472 | * - Destination page does not exist, we can add the pipe page to | |
473 | * the page cache and avoid the copy. | |
474 | * | |
83f9135b JA |
475 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
476 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
477 | * file address space page cache. This is possible if no one else has | |
478 | * the pipe page referenced outside of the pipe and page cache. If | |
479 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
480 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 JA |
481 | */ |
482 | static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf, | |
483 | struct splice_desc *sd) | |
484 | { | |
485 | struct file *file = sd->file; | |
486 | struct address_space *mapping = file->f_mapping; | |
3e7ee3e7 | 487 | gfp_t gfp_mask = mapping_gfp_mask(mapping); |
5274f052 JA |
488 | unsigned int offset; |
489 | struct page *page; | |
5274f052 | 490 | pgoff_t index; |
5abc97aa | 491 | char *src; |
3e7ee3e7 | 492 | int ret; |
5274f052 JA |
493 | |
494 | /* | |
49d0b21b | 495 | * make sure the data in this buffer is uptodate |
5274f052 JA |
496 | */ |
497 | src = buf->ops->map(file, info, buf); | |
498 | if (IS_ERR(src)) | |
499 | return PTR_ERR(src); | |
500 | ||
501 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
502 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
503 | ||
5274f052 | 504 | /* |
73d62d83 | 505 | * Reuse buf page, if SPLICE_F_MOVE is set. |
5274f052 | 506 | */ |
5abc97aa | 507 | if (sd->flags & SPLICE_F_MOVE) { |
83f9135b JA |
508 | /* |
509 | * If steal succeeds, buf->page is now pruned from the vm | |
510 | * side (LRU and page cache) and we can reuse it. | |
511 | */ | |
5abc97aa JA |
512 | if (buf->ops->steal(info, buf)) |
513 | goto find_page; | |
514 | ||
49d0b21b JA |
515 | /* |
516 | * this will also set the page locked | |
517 | */ | |
5abc97aa | 518 | page = buf->page; |
3e7ee3e7 | 519 | if (add_to_page_cache(page, mapping, index, gfp_mask)) |
5abc97aa | 520 | goto find_page; |
3e7ee3e7 JA |
521 | |
522 | if (!(buf->flags & PIPE_BUF_FLAG_LRU)) | |
523 | lru_cache_add(page); | |
5abc97aa JA |
524 | } else { |
525 | find_page: | |
526 | ret = -ENOMEM; | |
3e7ee3e7 | 527 | page = find_or_create_page(mapping, index, gfp_mask); |
5abc97aa | 528 | if (!page) |
9aefe431 | 529 | goto out_nomem; |
5abc97aa JA |
530 | |
531 | /* | |
532 | * If the page is uptodate, it is also locked. If it isn't | |
533 | * uptodate, we can mark it uptodate if we are filling the | |
534 | * full page. Otherwise we need to read it in first... | |
535 | */ | |
536 | if (!PageUptodate(page)) { | |
537 | if (sd->len < PAGE_CACHE_SIZE) { | |
538 | ret = mapping->a_ops->readpage(file, page); | |
539 | if (unlikely(ret)) | |
540 | goto out; | |
541 | ||
542 | lock_page(page); | |
543 | ||
544 | if (!PageUptodate(page)) { | |
545 | /* | |
73d62d83 | 546 | * Page got invalidated, repeat. |
5abc97aa JA |
547 | */ |
548 | if (!page->mapping) { | |
549 | unlock_page(page); | |
550 | page_cache_release(page); | |
551 | goto find_page; | |
552 | } | |
553 | ret = -EIO; | |
554 | goto out; | |
5274f052 | 555 | } |
5abc97aa JA |
556 | } else { |
557 | WARN_ON(!PageLocked(page)); | |
558 | SetPageUptodate(page); | |
5274f052 | 559 | } |
5274f052 JA |
560 | } |
561 | } | |
562 | ||
563 | ret = mapping->a_ops->prepare_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
564 | if (ret == AOP_TRUNCATED_PAGE) { |
565 | page_cache_release(page); | |
566 | goto find_page; | |
567 | } else if (ret) | |
5274f052 JA |
568 | goto out; |
569 | ||
3e7ee3e7 | 570 | if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) { |
5abc97aa JA |
571 | char *dst = kmap_atomic(page, KM_USER0); |
572 | ||
573 | memcpy(dst + offset, src + buf->offset, sd->len); | |
574 | flush_dcache_page(page); | |
575 | kunmap_atomic(dst, KM_USER0); | |
576 | } | |
5274f052 JA |
577 | |
578 | ret = mapping->a_ops->commit_write(file, page, 0, sd->len); | |
4f6f0bd2 JA |
579 | if (ret == AOP_TRUNCATED_PAGE) { |
580 | page_cache_release(page); | |
581 | goto find_page; | |
582 | } else if (ret) | |
5274f052 JA |
583 | goto out; |
584 | ||
c7f21e4f | 585 | mark_page_accessed(page); |
4f6f0bd2 | 586 | balance_dirty_pages_ratelimited(mapping); |
5274f052 | 587 | out: |
3e7ee3e7 | 588 | if (!(buf->flags & PIPE_BUF_FLAG_STOLEN)) { |
5abc97aa | 589 | page_cache_release(page); |
4f6f0bd2 JA |
590 | unlock_page(page); |
591 | } | |
9aefe431 | 592 | out_nomem: |
5274f052 JA |
593 | buf->ops->unmap(info, buf); |
594 | return ret; | |
595 | } | |
596 | ||
597 | typedef int (splice_actor)(struct pipe_inode_info *, struct pipe_buffer *, | |
598 | struct splice_desc *); | |
599 | ||
83f9135b JA |
600 | /* |
601 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
602 | * key here is the 'actor' worker passed in that actually moves the data | |
603 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
604 | */ | |
3a326a2c | 605 | static ssize_t move_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 606 | loff_t *ppos, size_t len, unsigned int flags, |
5274f052 JA |
607 | splice_actor *actor) |
608 | { | |
5274f052 JA |
609 | int ret, do_wakeup, err; |
610 | struct splice_desc sd; | |
611 | ||
612 | ret = 0; | |
613 | do_wakeup = 0; | |
614 | ||
615 | sd.total_len = len; | |
616 | sd.flags = flags; | |
617 | sd.file = out; | |
cbb7e577 | 618 | sd.pos = *ppos; |
5274f052 | 619 | |
3a326a2c IM |
620 | if (pipe->inode) |
621 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 622 | |
5274f052 | 623 | for (;;) { |
6f767b04 JA |
624 | if (pipe->nrbufs) { |
625 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
5274f052 JA |
626 | struct pipe_buf_operations *ops = buf->ops; |
627 | ||
628 | sd.len = buf->len; | |
629 | if (sd.len > sd.total_len) | |
630 | sd.len = sd.total_len; | |
631 | ||
3a326a2c | 632 | err = actor(pipe, buf, &sd); |
5274f052 JA |
633 | if (err) { |
634 | if (!ret && err != -ENODATA) | |
635 | ret = err; | |
636 | ||
637 | break; | |
638 | } | |
639 | ||
640 | ret += sd.len; | |
641 | buf->offset += sd.len; | |
642 | buf->len -= sd.len; | |
73d62d83 | 643 | |
5274f052 JA |
644 | if (!buf->len) { |
645 | buf->ops = NULL; | |
3a326a2c | 646 | ops->release(pipe, buf); |
6f767b04 JA |
647 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
648 | pipe->nrbufs--; | |
649 | if (pipe->inode) | |
650 | do_wakeup = 1; | |
5274f052 JA |
651 | } |
652 | ||
653 | sd.pos += sd.len; | |
654 | sd.total_len -= sd.len; | |
655 | if (!sd.total_len) | |
656 | break; | |
657 | } | |
658 | ||
6f767b04 | 659 | if (pipe->nrbufs) |
5274f052 | 660 | continue; |
3a326a2c | 661 | if (!pipe->writers) |
5274f052 | 662 | break; |
3a326a2c | 663 | if (!pipe->waiting_writers) { |
5274f052 JA |
664 | if (ret) |
665 | break; | |
666 | } | |
667 | ||
29e35094 LT |
668 | if (flags & SPLICE_F_NONBLOCK) { |
669 | if (!ret) | |
670 | ret = -EAGAIN; | |
671 | break; | |
672 | } | |
673 | ||
5274f052 JA |
674 | if (signal_pending(current)) { |
675 | if (!ret) | |
676 | ret = -ERESTARTSYS; | |
677 | break; | |
678 | } | |
679 | ||
680 | if (do_wakeup) { | |
c0bd1f65 | 681 | smp_mb(); |
3a326a2c IM |
682 | if (waitqueue_active(&pipe->wait)) |
683 | wake_up_interruptible_sync(&pipe->wait); | |
684 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
685 | do_wakeup = 0; |
686 | } | |
687 | ||
3a326a2c | 688 | pipe_wait(pipe); |
5274f052 JA |
689 | } |
690 | ||
3a326a2c IM |
691 | if (pipe->inode) |
692 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
693 | |
694 | if (do_wakeup) { | |
c0bd1f65 | 695 | smp_mb(); |
3a326a2c IM |
696 | if (waitqueue_active(&pipe->wait)) |
697 | wake_up_interruptible(&pipe->wait); | |
698 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
699 | } |
700 | ||
5274f052 | 701 | return ret; |
5274f052 JA |
702 | } |
703 | ||
83f9135b JA |
704 | /** |
705 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 706 | * @pipe: pipe info |
83f9135b JA |
707 | * @out: file to write to |
708 | * @len: number of bytes to splice | |
709 | * @flags: splice modifier flags | |
710 | * | |
711 | * Will either move or copy pages (determined by @flags options) from | |
712 | * the given pipe inode to the given file. | |
713 | * | |
714 | */ | |
3a326a2c IM |
715 | ssize_t |
716 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 717 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 718 | { |
4f6f0bd2 | 719 | struct address_space *mapping = out->f_mapping; |
3a326a2c IM |
720 | ssize_t ret; |
721 | ||
cbb7e577 | 722 | ret = move_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
4f6f0bd2 JA |
723 | |
724 | /* | |
73d62d83 | 725 | * If file or inode is SYNC and we actually wrote some data, sync it. |
4f6f0bd2 JA |
726 | */ |
727 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(mapping->host)) | |
728 | && ret > 0) { | |
729 | struct inode *inode = mapping->host; | |
730 | int err; | |
731 | ||
732 | mutex_lock(&inode->i_mutex); | |
733 | err = generic_osync_inode(mapping->host, mapping, | |
49570e9b | 734 | OSYNC_METADATA|OSYNC_DATA); |
4f6f0bd2 JA |
735 | mutex_unlock(&inode->i_mutex); |
736 | ||
737 | if (err) | |
738 | ret = err; | |
739 | } | |
740 | ||
741 | return ret; | |
5274f052 JA |
742 | } |
743 | ||
059a8f37 JA |
744 | EXPORT_SYMBOL(generic_file_splice_write); |
745 | ||
83f9135b JA |
746 | /** |
747 | * generic_splice_sendpage - splice data from a pipe to a socket | |
748 | * @inode: pipe inode | |
749 | * @out: socket to write to | |
750 | * @len: number of bytes to splice | |
751 | * @flags: splice modifier flags | |
752 | * | |
753 | * Will send @len bytes from the pipe to a network socket. No data copying | |
754 | * is involved. | |
755 | * | |
756 | */ | |
3a326a2c | 757 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 758 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 759 | { |
cbb7e577 | 760 | return move_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
761 | } |
762 | ||
059a8f37 | 763 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 764 | |
83f9135b JA |
765 | /* |
766 | * Attempt to initiate a splice from pipe to file. | |
767 | */ | |
3a326a2c | 768 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 769 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 770 | { |
5274f052 JA |
771 | int ret; |
772 | ||
49570e9b | 773 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
774 | return -EINVAL; |
775 | ||
49570e9b | 776 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
777 | return -EBADF; |
778 | ||
cbb7e577 | 779 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
780 | if (unlikely(ret < 0)) |
781 | return ret; | |
782 | ||
cbb7e577 | 783 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
784 | } |
785 | ||
83f9135b JA |
786 | /* |
787 | * Attempt to initiate a splice from a file to a pipe. | |
788 | */ | |
cbb7e577 JA |
789 | static long do_splice_to(struct file *in, loff_t *ppos, |
790 | struct pipe_inode_info *pipe, size_t len, | |
791 | unsigned int flags) | |
5274f052 | 792 | { |
cbb7e577 | 793 | loff_t isize, left; |
5274f052 JA |
794 | int ret; |
795 | ||
49570e9b | 796 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
797 | return -EINVAL; |
798 | ||
49570e9b | 799 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
800 | return -EBADF; |
801 | ||
cbb7e577 | 802 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
803 | if (unlikely(ret < 0)) |
804 | return ret; | |
805 | ||
806 | isize = i_size_read(in->f_mapping->host); | |
cbb7e577 | 807 | if (unlikely(*ppos >= isize)) |
5274f052 JA |
808 | return 0; |
809 | ||
cbb7e577 | 810 | left = isize - *ppos; |
49570e9b | 811 | if (unlikely(left < len)) |
5274f052 JA |
812 | len = left; |
813 | ||
cbb7e577 | 814 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
815 | } |
816 | ||
cbb7e577 JA |
817 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
818 | size_t len, unsigned int flags) | |
b92ce558 JA |
819 | { |
820 | struct pipe_inode_info *pipe; | |
821 | long ret, bytes; | |
cbb7e577 | 822 | loff_t out_off; |
b92ce558 JA |
823 | umode_t i_mode; |
824 | int i; | |
825 | ||
826 | /* | |
827 | * We require the input being a regular file, as we don't want to | |
828 | * randomly drop data for eg socket -> socket splicing. Use the | |
829 | * piped splicing for that! | |
830 | */ | |
831 | i_mode = in->f_dentry->d_inode->i_mode; | |
832 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) | |
833 | return -EINVAL; | |
834 | ||
835 | /* | |
836 | * neither in nor out is a pipe, setup an internal pipe attached to | |
837 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
838 | */ | |
839 | pipe = current->splice_pipe; | |
49570e9b | 840 | if (unlikely(!pipe)) { |
b92ce558 JA |
841 | pipe = alloc_pipe_info(NULL); |
842 | if (!pipe) | |
843 | return -ENOMEM; | |
844 | ||
845 | /* | |
846 | * We don't have an immediate reader, but we'll read the stuff | |
847 | * out of the pipe right after the move_to_pipe(). So set | |
848 | * PIPE_READERS appropriately. | |
849 | */ | |
850 | pipe->readers = 1; | |
851 | ||
852 | current->splice_pipe = pipe; | |
853 | } | |
854 | ||
855 | /* | |
73d62d83 | 856 | * Do the splice. |
b92ce558 JA |
857 | */ |
858 | ret = 0; | |
859 | bytes = 0; | |
cbb7e577 | 860 | out_off = 0; |
b92ce558 JA |
861 | |
862 | while (len) { | |
863 | size_t read_len, max_read_len; | |
864 | ||
865 | /* | |
866 | * Do at most PIPE_BUFFERS pages worth of transfer: | |
867 | */ | |
868 | max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE)); | |
869 | ||
cbb7e577 | 870 | ret = do_splice_to(in, ppos, pipe, max_read_len, flags); |
b92ce558 JA |
871 | if (unlikely(ret < 0)) |
872 | goto out_release; | |
873 | ||
874 | read_len = ret; | |
875 | ||
876 | /* | |
877 | * NOTE: nonblocking mode only applies to the input. We | |
878 | * must not do the output in nonblocking mode as then we | |
879 | * could get stuck data in the internal pipe: | |
880 | */ | |
cbb7e577 | 881 | ret = do_splice_from(pipe, out, &out_off, read_len, |
b92ce558 JA |
882 | flags & ~SPLICE_F_NONBLOCK); |
883 | if (unlikely(ret < 0)) | |
884 | goto out_release; | |
885 | ||
886 | bytes += ret; | |
887 | len -= ret; | |
888 | ||
889 | /* | |
890 | * In nonblocking mode, if we got back a short read then | |
891 | * that was due to either an IO error or due to the | |
892 | * pagecache entry not being there. In the IO error case | |
893 | * the _next_ splice attempt will produce a clean IO error | |
894 | * return value (not a short read), so in both cases it's | |
895 | * correct to break out of the loop here: | |
896 | */ | |
897 | if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len)) | |
898 | break; | |
899 | } | |
900 | ||
901 | pipe->nrbufs = pipe->curbuf = 0; | |
902 | ||
903 | return bytes; | |
904 | ||
905 | out_release: | |
906 | /* | |
907 | * If we did an incomplete transfer we must release | |
908 | * the pipe buffers in question: | |
909 | */ | |
910 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
911 | struct pipe_buffer *buf = pipe->bufs + i; | |
912 | ||
913 | if (buf->ops) { | |
914 | buf->ops->release(pipe, buf); | |
915 | buf->ops = NULL; | |
916 | } | |
917 | } | |
918 | pipe->nrbufs = pipe->curbuf = 0; | |
919 | ||
920 | /* | |
921 | * If we transferred some data, return the number of bytes: | |
922 | */ | |
923 | if (bytes > 0) | |
924 | return bytes; | |
925 | ||
926 | return ret; | |
927 | } | |
928 | ||
929 | EXPORT_SYMBOL(do_splice_direct); | |
930 | ||
83f9135b JA |
931 | /* |
932 | * Determine where to splice to/from. | |
933 | */ | |
529565dc IM |
934 | static long do_splice(struct file *in, loff_t __user *off_in, |
935 | struct file *out, loff_t __user *off_out, | |
936 | size_t len, unsigned int flags) | |
5274f052 | 937 | { |
3a326a2c | 938 | struct pipe_inode_info *pipe; |
cbb7e577 | 939 | loff_t offset, *off; |
5274f052 | 940 | |
3a326a2c | 941 | pipe = in->f_dentry->d_inode->i_pipe; |
529565dc IM |
942 | if (pipe) { |
943 | if (off_in) | |
944 | return -ESPIPE; | |
b92ce558 JA |
945 | if (off_out) { |
946 | if (out->f_op->llseek == no_llseek) | |
947 | return -EINVAL; | |
cbb7e577 | 948 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 949 | return -EFAULT; |
cbb7e577 JA |
950 | off = &offset; |
951 | } else | |
952 | off = &out->f_pos; | |
529565dc | 953 | |
cbb7e577 | 954 | return do_splice_from(pipe, out, off, len, flags); |
529565dc | 955 | } |
5274f052 | 956 | |
3a326a2c | 957 | pipe = out->f_dentry->d_inode->i_pipe; |
529565dc IM |
958 | if (pipe) { |
959 | if (off_out) | |
960 | return -ESPIPE; | |
b92ce558 JA |
961 | if (off_in) { |
962 | if (in->f_op->llseek == no_llseek) | |
963 | return -EINVAL; | |
cbb7e577 | 964 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 965 | return -EFAULT; |
cbb7e577 JA |
966 | off = &offset; |
967 | } else | |
968 | off = &in->f_pos; | |
529565dc | 969 | |
cbb7e577 | 970 | return do_splice_to(in, off, pipe, len, flags); |
529565dc | 971 | } |
5274f052 JA |
972 | |
973 | return -EINVAL; | |
974 | } | |
975 | ||
529565dc IM |
976 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
977 | int fd_out, loff_t __user *off_out, | |
978 | size_t len, unsigned int flags) | |
5274f052 JA |
979 | { |
980 | long error; | |
981 | struct file *in, *out; | |
982 | int fput_in, fput_out; | |
983 | ||
984 | if (unlikely(!len)) | |
985 | return 0; | |
986 | ||
987 | error = -EBADF; | |
529565dc | 988 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
989 | if (in) { |
990 | if (in->f_mode & FMODE_READ) { | |
529565dc | 991 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
992 | if (out) { |
993 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
994 | error = do_splice(in, off_in, |
995 | out, off_out, | |
996 | len, flags); | |
5274f052 JA |
997 | fput_light(out, fput_out); |
998 | } | |
999 | } | |
1000 | ||
1001 | fput_light(in, fput_in); | |
1002 | } | |
1003 | ||
1004 | return error; | |
1005 | } | |
70524490 JA |
1006 | |
1007 | /* | |
1008 | * Link contents of ipipe to opipe. | |
1009 | */ | |
1010 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1011 | struct pipe_inode_info *opipe, | |
1012 | size_t len, unsigned int flags) | |
1013 | { | |
1014 | struct pipe_buffer *ibuf, *obuf; | |
2a27250e JA |
1015 | int ret, do_wakeup, i, ipipe_first; |
1016 | ||
1017 | ret = do_wakeup = ipipe_first = 0; | |
70524490 JA |
1018 | |
1019 | /* | |
1020 | * Potential ABBA deadlock, work around it by ordering lock | |
1021 | * grabbing by inode address. Otherwise two different processes | |
1022 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1023 | */ | |
1024 | if (ipipe->inode < opipe->inode) { | |
2a27250e | 1025 | ipipe_first = 1; |
70524490 JA |
1026 | mutex_lock(&ipipe->inode->i_mutex); |
1027 | mutex_lock(&opipe->inode->i_mutex); | |
1028 | } else { | |
1029 | mutex_lock(&opipe->inode->i_mutex); | |
1030 | mutex_lock(&ipipe->inode->i_mutex); | |
1031 | } | |
1032 | ||
1033 | for (i = 0;; i++) { | |
1034 | if (!opipe->readers) { | |
1035 | send_sig(SIGPIPE, current, 0); | |
1036 | if (!ret) | |
1037 | ret = -EPIPE; | |
1038 | break; | |
1039 | } | |
1040 | if (ipipe->nrbufs - i) { | |
1041 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); | |
1042 | ||
1043 | /* | |
1044 | * If we have room, fill this buffer | |
1045 | */ | |
1046 | if (opipe->nrbufs < PIPE_BUFFERS) { | |
1047 | int nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
1048 | ||
1049 | /* | |
1050 | * Get a reference to this pipe buffer, | |
1051 | * so we can copy the contents over. | |
1052 | */ | |
1053 | ibuf->ops->get(ipipe, ibuf); | |
1054 | ||
1055 | obuf = opipe->bufs + nbuf; | |
1056 | *obuf = *ibuf; | |
1057 | ||
1058 | if (obuf->len > len) | |
1059 | obuf->len = len; | |
1060 | ||
1061 | opipe->nrbufs++; | |
1062 | do_wakeup = 1; | |
1063 | ret += obuf->len; | |
1064 | len -= obuf->len; | |
1065 | ||
1066 | if (!len) | |
1067 | break; | |
1068 | if (opipe->nrbufs < PIPE_BUFFERS) | |
1069 | continue; | |
1070 | } | |
1071 | ||
1072 | /* | |
1073 | * We have input available, but no output room. | |
2a27250e JA |
1074 | * If we already copied data, return that. If we |
1075 | * need to drop the opipe lock, it must be ordered | |
1076 | * last to avoid deadlocks. | |
70524490 | 1077 | */ |
2a27250e | 1078 | if ((flags & SPLICE_F_NONBLOCK) || !ipipe_first) { |
70524490 JA |
1079 | if (!ret) |
1080 | ret = -EAGAIN; | |
1081 | break; | |
1082 | } | |
1083 | if (signal_pending(current)) { | |
1084 | if (!ret) | |
1085 | ret = -ERESTARTSYS; | |
1086 | break; | |
1087 | } | |
1088 | if (do_wakeup) { | |
1089 | smp_mb(); | |
1090 | if (waitqueue_active(&opipe->wait)) | |
1091 | wake_up_interruptible(&opipe->wait); | |
1092 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1093 | do_wakeup = 0; | |
1094 | } | |
1095 | ||
1096 | opipe->waiting_writers++; | |
1097 | pipe_wait(opipe); | |
1098 | opipe->waiting_writers--; | |
1099 | continue; | |
1100 | } | |
1101 | ||
1102 | /* | |
1103 | * No input buffers, do the usual checks for available | |
1104 | * writers and blocking and wait if necessary | |
1105 | */ | |
1106 | if (!ipipe->writers) | |
1107 | break; | |
1108 | if (!ipipe->waiting_writers) { | |
1109 | if (ret) | |
1110 | break; | |
1111 | } | |
2a27250e JA |
1112 | /* |
1113 | * pipe_wait() drops the ipipe mutex. To avoid deadlocks | |
1114 | * with another process, we can only safely do that if | |
1115 | * the ipipe lock is ordered last. | |
1116 | */ | |
1117 | if ((flags & SPLICE_F_NONBLOCK) || ipipe_first) { | |
70524490 JA |
1118 | if (!ret) |
1119 | ret = -EAGAIN; | |
1120 | break; | |
1121 | } | |
1122 | if (signal_pending(current)) { | |
1123 | if (!ret) | |
1124 | ret = -ERESTARTSYS; | |
1125 | break; | |
1126 | } | |
1127 | ||
1128 | if (waitqueue_active(&ipipe->wait)) | |
1129 | wake_up_interruptible_sync(&ipipe->wait); | |
1130 | kill_fasync(&ipipe->fasync_writers, SIGIO, POLL_OUT); | |
1131 | ||
1132 | pipe_wait(ipipe); | |
1133 | } | |
1134 | ||
1135 | mutex_unlock(&ipipe->inode->i_mutex); | |
1136 | mutex_unlock(&opipe->inode->i_mutex); | |
1137 | ||
1138 | if (do_wakeup) { | |
1139 | smp_mb(); | |
1140 | if (waitqueue_active(&opipe->wait)) | |
1141 | wake_up_interruptible(&opipe->wait); | |
1142 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1143 | } | |
1144 | ||
1145 | return ret; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1150 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1151 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1152 | * applicable one is SPLICE_F_NONBLOCK. | |
1153 | */ | |
1154 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1155 | unsigned int flags) | |
1156 | { | |
1157 | struct pipe_inode_info *ipipe = in->f_dentry->d_inode->i_pipe; | |
1158 | struct pipe_inode_info *opipe = out->f_dentry->d_inode->i_pipe; | |
1159 | ||
1160 | /* | |
1161 | * Link ipipe to the two output pipes, consuming as we go along. | |
1162 | */ | |
1163 | if (ipipe && opipe) | |
1164 | return link_pipe(ipipe, opipe, len, flags); | |
1165 | ||
1166 | return -EINVAL; | |
1167 | } | |
1168 | ||
1169 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1170 | { | |
1171 | struct file *in; | |
1172 | int error, fput_in; | |
1173 | ||
1174 | if (unlikely(!len)) | |
1175 | return 0; | |
1176 | ||
1177 | error = -EBADF; | |
1178 | in = fget_light(fdin, &fput_in); | |
1179 | if (in) { | |
1180 | if (in->f_mode & FMODE_READ) { | |
1181 | int fput_out; | |
1182 | struct file *out = fget_light(fdout, &fput_out); | |
1183 | ||
1184 | if (out) { | |
1185 | if (out->f_mode & FMODE_WRITE) | |
1186 | error = do_tee(in, out, len, flags); | |
1187 | fput_light(out, fput_out); | |
1188 | } | |
1189 | } | |
1190 | fput_light(in, fput_in); | |
1191 | } | |
1192 | ||
1193 | return error; | |
1194 | } |