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 | * |
0fe23479 | 15 | * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk> |
c2058e06 JA |
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> | |
d6b29d7c | 23 | #include <linux/splice.h> |
5274f052 | 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> |
912d35f8 | 30 | #include <linux/uio.h> |
29ce2058 | 31 | #include <linux/security.h> |
5274f052 | 32 | |
83f9135b JA |
33 | /* |
34 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
35 | * a vm helper function, it's already simplified quite a bit by the | |
36 | * addition of remove_mapping(). If success is returned, the caller may | |
37 | * attempt to reuse this page for another destination. | |
38 | */ | |
76ad4d11 | 39 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, |
5abc97aa JA |
40 | struct pipe_buffer *buf) |
41 | { | |
42 | struct page *page = buf->page; | |
9e94cd4f | 43 | struct address_space *mapping; |
5abc97aa | 44 | |
9e0267c2 JA |
45 | lock_page(page); |
46 | ||
9e94cd4f JA |
47 | mapping = page_mapping(page); |
48 | if (mapping) { | |
49 | WARN_ON(!PageUptodate(page)); | |
5abc97aa | 50 | |
9e94cd4f JA |
51 | /* |
52 | * At least for ext2 with nobh option, we need to wait on | |
53 | * writeback completing on this page, since we'll remove it | |
54 | * from the pagecache. Otherwise truncate wont wait on the | |
55 | * page, allowing the disk blocks to be reused by someone else | |
56 | * before we actually wrote our data to them. fs corruption | |
57 | * ensues. | |
58 | */ | |
59 | wait_on_page_writeback(page); | |
ad8d6f0a | 60 | |
9e94cd4f | 61 | if (PagePrivate(page)) |
2ae88149 | 62 | try_to_release_page(page, GFP_KERNEL); |
4f6f0bd2 | 63 | |
9e94cd4f JA |
64 | /* |
65 | * If we succeeded in removing the mapping, set LRU flag | |
66 | * and return good. | |
67 | */ | |
68 | if (remove_mapping(mapping, page)) { | |
69 | buf->flags |= PIPE_BUF_FLAG_LRU; | |
70 | return 0; | |
71 | } | |
9e0267c2 | 72 | } |
5abc97aa | 73 | |
9e94cd4f JA |
74 | /* |
75 | * Raced with truncate or failed to remove page from current | |
76 | * address space, unlock and return failure. | |
77 | */ | |
78 | unlock_page(page); | |
79 | return 1; | |
5abc97aa JA |
80 | } |
81 | ||
76ad4d11 | 82 | static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, |
5274f052 JA |
83 | struct pipe_buffer *buf) |
84 | { | |
85 | page_cache_release(buf->page); | |
1432873a | 86 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
5274f052 JA |
87 | } |
88 | ||
0845718d JA |
89 | /* |
90 | * Check whether the contents of buf is OK to access. Since the content | |
91 | * is a page cache page, IO may be in flight. | |
92 | */ | |
cac36bb0 JA |
93 | static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, |
94 | struct pipe_buffer *buf) | |
5274f052 JA |
95 | { |
96 | struct page *page = buf->page; | |
49d0b21b | 97 | int err; |
5274f052 JA |
98 | |
99 | if (!PageUptodate(page)) { | |
49d0b21b JA |
100 | lock_page(page); |
101 | ||
102 | /* | |
103 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 104 | * splice, if this is the first page. |
49d0b21b JA |
105 | */ |
106 | if (!page->mapping) { | |
107 | err = -ENODATA; | |
108 | goto error; | |
109 | } | |
5274f052 | 110 | |
49d0b21b | 111 | /* |
73d62d83 | 112 | * Uh oh, read-error from disk. |
49d0b21b JA |
113 | */ |
114 | if (!PageUptodate(page)) { | |
115 | err = -EIO; | |
116 | goto error; | |
117 | } | |
118 | ||
119 | /* | |
f84d7519 | 120 | * Page is ok afterall, we are done. |
49d0b21b | 121 | */ |
5274f052 | 122 | unlock_page(page); |
5274f052 JA |
123 | } |
124 | ||
f84d7519 | 125 | return 0; |
49d0b21b JA |
126 | error: |
127 | unlock_page(page); | |
f84d7519 | 128 | return err; |
70524490 JA |
129 | } |
130 | ||
d4c3cca9 | 131 | static const struct pipe_buf_operations page_cache_pipe_buf_ops = { |
5274f052 | 132 | .can_merge = 0, |
f84d7519 JA |
133 | .map = generic_pipe_buf_map, |
134 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 135 | .confirm = page_cache_pipe_buf_confirm, |
5274f052 | 136 | .release = page_cache_pipe_buf_release, |
5abc97aa | 137 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 138 | .get = generic_pipe_buf_get, |
5274f052 JA |
139 | }; |
140 | ||
912d35f8 JA |
141 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
142 | struct pipe_buffer *buf) | |
143 | { | |
7afa6fd0 JA |
144 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
145 | return 1; | |
146 | ||
1432873a | 147 | buf->flags |= PIPE_BUF_FLAG_LRU; |
330ab716 | 148 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
149 | } |
150 | ||
d4c3cca9 | 151 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
912d35f8 | 152 | .can_merge = 0, |
f84d7519 JA |
153 | .map = generic_pipe_buf_map, |
154 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 155 | .confirm = generic_pipe_buf_confirm, |
912d35f8 JA |
156 | .release = page_cache_pipe_buf_release, |
157 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 158 | .get = generic_pipe_buf_get, |
912d35f8 JA |
159 | }; |
160 | ||
932cc6d4 JA |
161 | /** |
162 | * splice_to_pipe - fill passed data into a pipe | |
163 | * @pipe: pipe to fill | |
164 | * @spd: data to fill | |
165 | * | |
166 | * Description: | |
167 | * @spd contains a map of pages and len/offset tupples, a long with | |
168 | * the struct pipe_buf_operations associated with these pages. This | |
169 | * function will link that data to the pipe. | |
170 | * | |
83f9135b | 171 | */ |
d6b29d7c JA |
172 | ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
173 | struct splice_pipe_desc *spd) | |
5274f052 | 174 | { |
00de00bd | 175 | unsigned int spd_pages = spd->nr_pages; |
912d35f8 | 176 | int ret, do_wakeup, page_nr; |
5274f052 JA |
177 | |
178 | ret = 0; | |
179 | do_wakeup = 0; | |
912d35f8 | 180 | page_nr = 0; |
5274f052 | 181 | |
3a326a2c IM |
182 | if (pipe->inode) |
183 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 184 | |
5274f052 | 185 | for (;;) { |
3a326a2c | 186 | if (!pipe->readers) { |
5274f052 JA |
187 | send_sig(SIGPIPE, current, 0); |
188 | if (!ret) | |
189 | ret = -EPIPE; | |
190 | break; | |
191 | } | |
192 | ||
6f767b04 JA |
193 | if (pipe->nrbufs < PIPE_BUFFERS) { |
194 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 195 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 196 | |
912d35f8 JA |
197 | buf->page = spd->pages[page_nr]; |
198 | buf->offset = spd->partial[page_nr].offset; | |
199 | buf->len = spd->partial[page_nr].len; | |
497f9625 | 200 | buf->private = spd->partial[page_nr].private; |
912d35f8 | 201 | buf->ops = spd->ops; |
7afa6fd0 JA |
202 | if (spd->flags & SPLICE_F_GIFT) |
203 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
204 | ||
6f767b04 | 205 | pipe->nrbufs++; |
912d35f8 JA |
206 | page_nr++; |
207 | ret += buf->len; | |
208 | ||
6f767b04 JA |
209 | if (pipe->inode) |
210 | do_wakeup = 1; | |
5274f052 | 211 | |
912d35f8 | 212 | if (!--spd->nr_pages) |
5274f052 | 213 | break; |
6f767b04 | 214 | if (pipe->nrbufs < PIPE_BUFFERS) |
5274f052 JA |
215 | continue; |
216 | ||
217 | break; | |
218 | } | |
219 | ||
912d35f8 | 220 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
221 | if (!ret) |
222 | ret = -EAGAIN; | |
223 | break; | |
224 | } | |
225 | ||
5274f052 JA |
226 | if (signal_pending(current)) { |
227 | if (!ret) | |
228 | ret = -ERESTARTSYS; | |
229 | break; | |
230 | } | |
231 | ||
232 | if (do_wakeup) { | |
c0bd1f65 | 233 | smp_mb(); |
3a326a2c IM |
234 | if (waitqueue_active(&pipe->wait)) |
235 | wake_up_interruptible_sync(&pipe->wait); | |
236 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
237 | do_wakeup = 0; |
238 | } | |
239 | ||
3a326a2c IM |
240 | pipe->waiting_writers++; |
241 | pipe_wait(pipe); | |
242 | pipe->waiting_writers--; | |
5274f052 JA |
243 | } |
244 | ||
02676e5a | 245 | if (pipe->inode) { |
3a326a2c | 246 | mutex_unlock(&pipe->inode->i_mutex); |
5274f052 | 247 | |
02676e5a JA |
248 | if (do_wakeup) { |
249 | smp_mb(); | |
250 | if (waitqueue_active(&pipe->wait)) | |
251 | wake_up_interruptible(&pipe->wait); | |
252 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
253 | } | |
5274f052 JA |
254 | } |
255 | ||
00de00bd | 256 | while (page_nr < spd_pages) |
912d35f8 | 257 | page_cache_release(spd->pages[page_nr++]); |
5274f052 JA |
258 | |
259 | return ret; | |
260 | } | |
261 | ||
3a326a2c | 262 | static int |
cbb7e577 JA |
263 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
264 | struct pipe_inode_info *pipe, size_t len, | |
265 | unsigned int flags) | |
5274f052 JA |
266 | { |
267 | struct address_space *mapping = in->f_mapping; | |
912d35f8 | 268 | unsigned int loff, nr_pages; |
16c523dd | 269 | struct page *pages[PIPE_BUFFERS]; |
912d35f8 | 270 | struct partial_page partial[PIPE_BUFFERS]; |
5274f052 | 271 | struct page *page; |
91ad66ef JA |
272 | pgoff_t index, end_index; |
273 | loff_t isize; | |
eb20796b | 274 | int error, page_nr; |
912d35f8 JA |
275 | struct splice_pipe_desc spd = { |
276 | .pages = pages, | |
277 | .partial = partial, | |
278 | .flags = flags, | |
279 | .ops = &page_cache_pipe_buf_ops, | |
280 | }; | |
5274f052 | 281 | |
cbb7e577 | 282 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 JA |
283 | loff = *ppos & ~PAGE_CACHE_MASK; |
284 | nr_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
5274f052 JA |
285 | |
286 | if (nr_pages > PIPE_BUFFERS) | |
287 | nr_pages = PIPE_BUFFERS; | |
288 | ||
289 | /* | |
86aa5ac5 JA |
290 | * Don't try to 2nd guess the read-ahead logic, call into |
291 | * page_cache_readahead() like the page cache reads would do. | |
5274f052 | 292 | */ |
86aa5ac5 | 293 | page_cache_readahead(mapping, &in->f_ra, in, index, nr_pages); |
5274f052 | 294 | |
eb20796b JA |
295 | /* |
296 | * Lookup the (hopefully) full range of pages we need. | |
297 | */ | |
298 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
82aa5d61 | 299 | |
eb20796b JA |
300 | /* |
301 | * If find_get_pages_contig() returned fewer pages than we needed, | |
932cc6d4 | 302 | * allocate the rest and fill in the holes. |
eb20796b | 303 | */ |
932cc6d4 | 304 | error = 0; |
eb20796b JA |
305 | index += spd.nr_pages; |
306 | while (spd.nr_pages < nr_pages) { | |
82aa5d61 | 307 | /* |
eb20796b JA |
308 | * Page could be there, find_get_pages_contig() breaks on |
309 | * the first hole. | |
5274f052 | 310 | */ |
7480a904 JA |
311 | page = find_get_page(mapping, index); |
312 | if (!page) { | |
e27dedd8 JA |
313 | /* |
314 | * Make sure the read-ahead engine is notified | |
315 | * about this failure. | |
316 | */ | |
317 | handle_ra_miss(mapping, &in->f_ra, index); | |
318 | ||
7480a904 | 319 | /* |
eb20796b | 320 | * page didn't exist, allocate one. |
7480a904 JA |
321 | */ |
322 | page = page_cache_alloc_cold(mapping); | |
323 | if (!page) | |
324 | break; | |
325 | ||
326 | error = add_to_page_cache_lru(page, mapping, index, | |
2ae88149 | 327 | GFP_KERNEL); |
7480a904 JA |
328 | if (unlikely(error)) { |
329 | page_cache_release(page); | |
a0548871 JA |
330 | if (error == -EEXIST) |
331 | continue; | |
7480a904 JA |
332 | break; |
333 | } | |
eb20796b JA |
334 | /* |
335 | * add_to_page_cache() locks the page, unlock it | |
336 | * to avoid convoluting the logic below even more. | |
337 | */ | |
338 | unlock_page(page); | |
7480a904 JA |
339 | } |
340 | ||
eb20796b JA |
341 | pages[spd.nr_pages++] = page; |
342 | index++; | |
343 | } | |
344 | ||
345 | /* | |
346 | * Now loop over the map and see if we need to start IO on any | |
347 | * pages, fill in the partial map, etc. | |
348 | */ | |
349 | index = *ppos >> PAGE_CACHE_SHIFT; | |
350 | nr_pages = spd.nr_pages; | |
351 | spd.nr_pages = 0; | |
352 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
353 | unsigned int this_len; | |
354 | ||
355 | if (!len) | |
356 | break; | |
357 | ||
358 | /* | |
359 | * this_len is the max we'll use from this page | |
360 | */ | |
361 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
362 | page = pages[page_nr]; | |
363 | ||
7480a904 JA |
364 | /* |
365 | * If the page isn't uptodate, we may need to start io on it | |
366 | */ | |
367 | if (!PageUptodate(page)) { | |
c4f895cb JA |
368 | /* |
369 | * If in nonblock mode then dont block on waiting | |
370 | * for an in-flight io page | |
371 | */ | |
9ae9d68c FW |
372 | if (flags & SPLICE_F_NONBLOCK) { |
373 | if (TestSetPageLocked(page)) | |
374 | break; | |
375 | } else | |
376 | lock_page(page); | |
7480a904 JA |
377 | |
378 | /* | |
379 | * page was truncated, stop here. if this isn't the | |
380 | * first page, we'll just complete what we already | |
381 | * added | |
382 | */ | |
383 | if (!page->mapping) { | |
384 | unlock_page(page); | |
7480a904 JA |
385 | break; |
386 | } | |
387 | /* | |
388 | * page was already under io and is now done, great | |
389 | */ | |
390 | if (PageUptodate(page)) { | |
391 | unlock_page(page); | |
392 | goto fill_it; | |
393 | } | |
5274f052 | 394 | |
7480a904 JA |
395 | /* |
396 | * need to read in the page | |
397 | */ | |
398 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 399 | if (unlikely(error)) { |
eb20796b JA |
400 | /* |
401 | * We really should re-lookup the page here, | |
402 | * but it complicates things a lot. Instead | |
403 | * lets just do what we already stored, and | |
404 | * we'll get it the next time we are called. | |
405 | */ | |
7480a904 | 406 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
407 | error = 0; |
408 | ||
5274f052 JA |
409 | break; |
410 | } | |
620a324b JA |
411 | } |
412 | fill_it: | |
413 | /* | |
414 | * i_size must be checked after PageUptodate. | |
415 | */ | |
416 | isize = i_size_read(mapping->host); | |
417 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
418 | if (unlikely(!isize || index > end_index)) | |
419 | break; | |
420 | ||
421 | /* | |
422 | * if this is the last page, see if we need to shrink | |
423 | * the length and stop | |
424 | */ | |
425 | if (end_index == index) { | |
426 | unsigned int plen; | |
91ad66ef JA |
427 | |
428 | /* | |
620a324b | 429 | * max good bytes in this page |
91ad66ef | 430 | */ |
620a324b JA |
431 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
432 | if (plen <= loff) | |
91ad66ef | 433 | break; |
91ad66ef JA |
434 | |
435 | /* | |
620a324b | 436 | * force quit after adding this page |
91ad66ef | 437 | */ |
620a324b JA |
438 | this_len = min(this_len, plen - loff); |
439 | len = this_len; | |
5274f052 | 440 | } |
620a324b | 441 | |
eb20796b JA |
442 | partial[page_nr].offset = loff; |
443 | partial[page_nr].len = this_len; | |
82aa5d61 | 444 | len -= this_len; |
91ad66ef | 445 | loff = 0; |
eb20796b JA |
446 | spd.nr_pages++; |
447 | index++; | |
5274f052 JA |
448 | } |
449 | ||
eb20796b | 450 | /* |
475ecade | 451 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
452 | * we got, 'nr_pages' is how many pages are in the map. |
453 | */ | |
454 | while (page_nr < nr_pages) | |
455 | page_cache_release(pages[page_nr++]); | |
456 | ||
912d35f8 | 457 | if (spd.nr_pages) |
00522fb4 | 458 | return splice_to_pipe(pipe, &spd); |
5274f052 | 459 | |
7480a904 | 460 | return error; |
5274f052 JA |
461 | } |
462 | ||
83f9135b JA |
463 | /** |
464 | * generic_file_splice_read - splice data from file to a pipe | |
465 | * @in: file to splice from | |
932cc6d4 | 466 | * @ppos: position in @in |
83f9135b JA |
467 | * @pipe: pipe to splice to |
468 | * @len: number of bytes to splice | |
469 | * @flags: splice modifier flags | |
470 | * | |
932cc6d4 JA |
471 | * Description: |
472 | * Will read pages from given file and fill them into a pipe. Can be | |
473 | * used as long as the address_space operations for the source implements | |
474 | * a readpage() hook. | |
475 | * | |
83f9135b | 476 | */ |
cbb7e577 JA |
477 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
478 | struct pipe_inode_info *pipe, size_t len, | |
479 | unsigned int flags) | |
5274f052 JA |
480 | { |
481 | ssize_t spliced; | |
482 | int ret; | |
d366d398 JA |
483 | loff_t isize, left; |
484 | ||
485 | isize = i_size_read(in->f_mapping->host); | |
486 | if (unlikely(*ppos >= isize)) | |
487 | return 0; | |
488 | ||
489 | left = isize - *ppos; | |
490 | if (unlikely(left < len)) | |
491 | len = left; | |
5274f052 JA |
492 | |
493 | ret = 0; | |
494 | spliced = 0; | |
51a92c0f | 495 | while (len && !spliced) { |
cbb7e577 | 496 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 497 | |
c4f895cb | 498 | if (ret < 0) |
5274f052 | 499 | break; |
c4f895cb JA |
500 | else if (!ret) { |
501 | if (spliced) | |
502 | break; | |
503 | if (flags & SPLICE_F_NONBLOCK) { | |
504 | ret = -EAGAIN; | |
505 | break; | |
506 | } | |
507 | } | |
5274f052 | 508 | |
cbb7e577 | 509 | *ppos += ret; |
5274f052 JA |
510 | len -= ret; |
511 | spliced += ret; | |
512 | } | |
513 | ||
514 | if (spliced) | |
515 | return spliced; | |
516 | ||
517 | return ret; | |
518 | } | |
519 | ||
059a8f37 JA |
520 | EXPORT_SYMBOL(generic_file_splice_read); |
521 | ||
5274f052 | 522 | /* |
4f6f0bd2 | 523 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 524 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 525 | */ |
76ad4d11 | 526 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
527 | struct pipe_buffer *buf, struct splice_desc *sd) |
528 | { | |
6a14b90b | 529 | struct file *file = sd->u.file; |
5274f052 | 530 | loff_t pos = sd->pos; |
f84d7519 | 531 | int ret, more; |
5274f052 | 532 | |
cac36bb0 | 533 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
534 | if (!ret) { |
535 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 536 | |
f84d7519 JA |
537 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
538 | sd->len, &pos, more); | |
539 | } | |
5274f052 | 540 | |
016b661e | 541 | return ret; |
5274f052 JA |
542 | } |
543 | ||
544 | /* | |
545 | * This is a little more tricky than the file -> pipe splicing. There are | |
546 | * basically three cases: | |
547 | * | |
548 | * - Destination page already exists in the address space and there | |
549 | * are users of it. For that case we have no other option that | |
550 | * copying the data. Tough luck. | |
551 | * - Destination page already exists in the address space, but there | |
552 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
553 | * through to last case. | |
554 | * - Destination page does not exist, we can add the pipe page to | |
555 | * the page cache and avoid the copy. | |
556 | * | |
83f9135b JA |
557 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
558 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
559 | * file address space page cache. This is possible if no one else has | |
560 | * the pipe page referenced outside of the pipe and page cache. If | |
561 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
562 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 563 | */ |
76ad4d11 | 564 | static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
5274f052 JA |
565 | struct splice_desc *sd) |
566 | { | |
6a14b90b | 567 | struct file *file = sd->u.file; |
5274f052 | 568 | struct address_space *mapping = file->f_mapping; |
016b661e | 569 | unsigned int offset, this_len; |
5274f052 | 570 | struct page *page; |
5274f052 | 571 | pgoff_t index; |
3e7ee3e7 | 572 | int ret; |
5274f052 JA |
573 | |
574 | /* | |
49d0b21b | 575 | * make sure the data in this buffer is uptodate |
5274f052 | 576 | */ |
cac36bb0 | 577 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
578 | if (unlikely(ret)) |
579 | return ret; | |
5274f052 JA |
580 | |
581 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
582 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
583 | ||
016b661e JA |
584 | this_len = sd->len; |
585 | if (this_len + offset > PAGE_CACHE_SIZE) | |
586 | this_len = PAGE_CACHE_SIZE - offset; | |
587 | ||
485ddb4b NP |
588 | find_page: |
589 | page = find_lock_page(mapping, index); | |
590 | if (!page) { | |
591 | ret = -ENOMEM; | |
592 | page = page_cache_alloc_cold(mapping); | |
593 | if (unlikely(!page)) | |
594 | goto out_ret; | |
595 | ||
83f9135b | 596 | /* |
485ddb4b | 597 | * This will also lock the page |
83f9135b | 598 | */ |
485ddb4b NP |
599 | ret = add_to_page_cache_lru(page, mapping, index, |
600 | GFP_KERNEL); | |
601 | if (unlikely(ret)) | |
602 | goto out; | |
603 | } | |
9e0267c2 | 604 | |
016b661e | 605 | ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); |
bfc4ee39 JA |
606 | if (unlikely(ret)) { |
607 | loff_t isize = i_size_read(mapping->host); | |
608 | ||
609 | if (ret != AOP_TRUNCATED_PAGE) | |
610 | unlock_page(page); | |
4f6f0bd2 | 611 | page_cache_release(page); |
bfc4ee39 JA |
612 | if (ret == AOP_TRUNCATED_PAGE) |
613 | goto find_page; | |
614 | ||
615 | /* | |
616 | * prepare_write() may have instantiated a few blocks | |
617 | * outside i_size. Trim these off again. | |
618 | */ | |
619 | if (sd->pos + this_len > isize) | |
620 | vmtruncate(mapping->host, isize); | |
621 | ||
e6e80f29 | 622 | goto out_ret; |
bfc4ee39 | 623 | } |
5274f052 | 624 | |
0568b409 | 625 | if (buf->page != page) { |
f84d7519 JA |
626 | /* |
627 | * Careful, ->map() uses KM_USER0! | |
628 | */ | |
76ad4d11 | 629 | char *src = buf->ops->map(pipe, buf, 1); |
f84d7519 | 630 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 631 | |
016b661e | 632 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 633 | flush_dcache_page(page); |
f84d7519 | 634 | kunmap_atomic(dst, KM_USER1); |
76ad4d11 | 635 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 636 | } |
5274f052 | 637 | |
016b661e | 638 | ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); |
d9993c37 DM |
639 | if (ret) { |
640 | if (ret == AOP_TRUNCATED_PAGE) { | |
641 | page_cache_release(page); | |
642 | goto find_page; | |
643 | } | |
644 | if (ret < 0) | |
645 | goto out; | |
0568b409 | 646 | /* |
d9993c37 DM |
647 | * Partial write has happened, so 'ret' already initialized by |
648 | * number of bytes written, Where is nothing we have to do here. | |
0568b409 | 649 | */ |
d9993c37 | 650 | } else |
0568b409 | 651 | ret = this_len; |
d9993c37 DM |
652 | /* |
653 | * Return the number of bytes written and mark page as | |
654 | * accessed, we are now done! | |
655 | */ | |
656 | mark_page_accessed(page); | |
5274f052 | 657 | out: |
0568b409 | 658 | page_cache_release(page); |
9e0267c2 | 659 | unlock_page(page); |
e6e80f29 | 660 | out_ret: |
5274f052 JA |
661 | return ret; |
662 | } | |
663 | ||
932cc6d4 JA |
664 | /** |
665 | * __splice_from_pipe - splice data from a pipe to given actor | |
666 | * @pipe: pipe to splice from | |
667 | * @sd: information to @actor | |
668 | * @actor: handler that splices the data | |
669 | * | |
670 | * Description: | |
671 | * This function does little more than loop over the pipe and call | |
672 | * @actor to do the actual moving of a single struct pipe_buffer to | |
673 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
674 | * pipe_to_user. | |
675 | * | |
83f9135b | 676 | */ |
c66ab6fa JA |
677 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, |
678 | splice_actor *actor) | |
5274f052 | 679 | { |
5274f052 | 680 | int ret, do_wakeup, err; |
5274f052 JA |
681 | |
682 | ret = 0; | |
683 | do_wakeup = 0; | |
684 | ||
5274f052 | 685 | for (;;) { |
6f767b04 JA |
686 | if (pipe->nrbufs) { |
687 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
d4c3cca9 | 688 | const struct pipe_buf_operations *ops = buf->ops; |
5274f052 | 689 | |
c66ab6fa JA |
690 | sd->len = buf->len; |
691 | if (sd->len > sd->total_len) | |
692 | sd->len = sd->total_len; | |
5274f052 | 693 | |
c66ab6fa | 694 | err = actor(pipe, buf, sd); |
016b661e | 695 | if (err <= 0) { |
5274f052 JA |
696 | if (!ret && err != -ENODATA) |
697 | ret = err; | |
698 | ||
699 | break; | |
700 | } | |
701 | ||
016b661e JA |
702 | ret += err; |
703 | buf->offset += err; | |
704 | buf->len -= err; | |
705 | ||
c66ab6fa JA |
706 | sd->len -= err; |
707 | sd->pos += err; | |
708 | sd->total_len -= err; | |
709 | if (sd->len) | |
016b661e | 710 | continue; |
73d62d83 | 711 | |
5274f052 JA |
712 | if (!buf->len) { |
713 | buf->ops = NULL; | |
3a326a2c | 714 | ops->release(pipe, buf); |
6f767b04 JA |
715 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
716 | pipe->nrbufs--; | |
717 | if (pipe->inode) | |
718 | do_wakeup = 1; | |
5274f052 JA |
719 | } |
720 | ||
c66ab6fa | 721 | if (!sd->total_len) |
5274f052 JA |
722 | break; |
723 | } | |
724 | ||
6f767b04 | 725 | if (pipe->nrbufs) |
5274f052 | 726 | continue; |
3a326a2c | 727 | if (!pipe->writers) |
5274f052 | 728 | break; |
3a326a2c | 729 | if (!pipe->waiting_writers) { |
5274f052 JA |
730 | if (ret) |
731 | break; | |
732 | } | |
733 | ||
c66ab6fa | 734 | if (sd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
735 | if (!ret) |
736 | ret = -EAGAIN; | |
737 | break; | |
738 | } | |
739 | ||
5274f052 JA |
740 | if (signal_pending(current)) { |
741 | if (!ret) | |
742 | ret = -ERESTARTSYS; | |
743 | break; | |
744 | } | |
745 | ||
746 | if (do_wakeup) { | |
c0bd1f65 | 747 | smp_mb(); |
3a326a2c IM |
748 | if (waitqueue_active(&pipe->wait)) |
749 | wake_up_interruptible_sync(&pipe->wait); | |
750 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
751 | do_wakeup = 0; |
752 | } | |
753 | ||
3a326a2c | 754 | pipe_wait(pipe); |
5274f052 JA |
755 | } |
756 | ||
5274f052 | 757 | if (do_wakeup) { |
c0bd1f65 | 758 | smp_mb(); |
3a326a2c IM |
759 | if (waitqueue_active(&pipe->wait)) |
760 | wake_up_interruptible(&pipe->wait); | |
761 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
762 | } |
763 | ||
5274f052 | 764 | return ret; |
5274f052 | 765 | } |
40bee44e | 766 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 767 | |
932cc6d4 JA |
768 | /** |
769 | * splice_from_pipe - splice data from a pipe to a file | |
770 | * @pipe: pipe to splice from | |
771 | * @out: file to splice to | |
772 | * @ppos: position in @out | |
773 | * @len: how many bytes to splice | |
774 | * @flags: splice modifier flags | |
775 | * @actor: handler that splices the data | |
776 | * | |
777 | * Description: | |
778 | * See __splice_from_pipe. This function locks the input and output inodes, | |
779 | * otherwise it's identical to __splice_from_pipe(). | |
780 | * | |
781 | */ | |
6da61809 MF |
782 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
783 | loff_t *ppos, size_t len, unsigned int flags, | |
784 | splice_actor *actor) | |
785 | { | |
786 | ssize_t ret; | |
787 | struct inode *inode = out->f_mapping->host; | |
c66ab6fa JA |
788 | struct splice_desc sd = { |
789 | .total_len = len, | |
790 | .flags = flags, | |
791 | .pos = *ppos, | |
6a14b90b | 792 | .u.file = out, |
c66ab6fa | 793 | }; |
6da61809 MF |
794 | |
795 | /* | |
796 | * The actor worker might be calling ->prepare_write and | |
797 | * ->commit_write. Most of the time, these expect i_mutex to | |
798 | * be held. Since this may result in an ABBA deadlock with | |
799 | * pipe->inode, we have to order lock acquiry here. | |
800 | */ | |
801 | inode_double_lock(inode, pipe->inode); | |
c66ab6fa | 802 | ret = __splice_from_pipe(pipe, &sd, actor); |
6da61809 MF |
803 | inode_double_unlock(inode, pipe->inode); |
804 | ||
805 | return ret; | |
806 | } | |
807 | ||
808 | /** | |
809 | * generic_file_splice_write_nolock - generic_file_splice_write without mutexes | |
810 | * @pipe: pipe info | |
811 | * @out: file to write to | |
932cc6d4 | 812 | * @ppos: position in @out |
6da61809 MF |
813 | * @len: number of bytes to splice |
814 | * @flags: splice modifier flags | |
815 | * | |
932cc6d4 JA |
816 | * Description: |
817 | * Will either move or copy pages (determined by @flags options) from | |
818 | * the given pipe inode to the given file. The caller is responsible | |
819 | * for acquiring i_mutex on both inodes. | |
6da61809 MF |
820 | * |
821 | */ | |
822 | ssize_t | |
823 | generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out, | |
824 | loff_t *ppos, size_t len, unsigned int flags) | |
825 | { | |
826 | struct address_space *mapping = out->f_mapping; | |
827 | struct inode *inode = mapping->host; | |
c66ab6fa JA |
828 | struct splice_desc sd = { |
829 | .total_len = len, | |
830 | .flags = flags, | |
831 | .pos = *ppos, | |
6a14b90b | 832 | .u.file = out, |
c66ab6fa | 833 | }; |
6da61809 MF |
834 | ssize_t ret; |
835 | int err; | |
836 | ||
0f7fc9e4 | 837 | err = remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
838 | if (unlikely(err)) |
839 | return err; | |
840 | ||
c66ab6fa | 841 | ret = __splice_from_pipe(pipe, &sd, pipe_to_file); |
6da61809 | 842 | if (ret > 0) { |
17ee4f49 JA |
843 | unsigned long nr_pages; |
844 | ||
6da61809 | 845 | *ppos += ret; |
17ee4f49 | 846 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
6da61809 MF |
847 | |
848 | /* | |
849 | * If file or inode is SYNC and we actually wrote some data, | |
850 | * sync it. | |
851 | */ | |
852 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
853 | err = generic_osync_inode(inode, mapping, | |
854 | OSYNC_METADATA|OSYNC_DATA); | |
855 | ||
856 | if (err) | |
857 | ret = err; | |
858 | } | |
17ee4f49 | 859 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
6da61809 MF |
860 | } |
861 | ||
862 | return ret; | |
863 | } | |
864 | ||
865 | EXPORT_SYMBOL(generic_file_splice_write_nolock); | |
866 | ||
83f9135b JA |
867 | /** |
868 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 869 | * @pipe: pipe info |
83f9135b | 870 | * @out: file to write to |
932cc6d4 | 871 | * @ppos: position in @out |
83f9135b JA |
872 | * @len: number of bytes to splice |
873 | * @flags: splice modifier flags | |
874 | * | |
932cc6d4 JA |
875 | * Description: |
876 | * Will either move or copy pages (determined by @flags options) from | |
877 | * the given pipe inode to the given file. | |
83f9135b JA |
878 | * |
879 | */ | |
3a326a2c IM |
880 | ssize_t |
881 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 882 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 883 | { |
4f6f0bd2 | 884 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 885 | struct inode *inode = mapping->host; |
3a326a2c | 886 | ssize_t ret; |
8c34e2d6 JA |
887 | int err; |
888 | ||
0f7fc9e4 | 889 | err = should_remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
890 | if (unlikely(err)) { |
891 | mutex_lock(&inode->i_mutex); | |
0f7fc9e4 | 892 | err = __remove_suid(out->f_path.dentry, err); |
8c34e2d6 JA |
893 | mutex_unlock(&inode->i_mutex); |
894 | if (err) | |
895 | return err; | |
896 | } | |
3a326a2c | 897 | |
00522fb4 | 898 | ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
a4514ebd | 899 | if (ret > 0) { |
17ee4f49 JA |
900 | unsigned long nr_pages; |
901 | ||
a4514ebd | 902 | *ppos += ret; |
17ee4f49 | 903 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
a4514ebd JA |
904 | |
905 | /* | |
906 | * If file or inode is SYNC and we actually wrote some data, | |
907 | * sync it. | |
908 | */ | |
909 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
a4514ebd JA |
910 | mutex_lock(&inode->i_mutex); |
911 | err = generic_osync_inode(inode, mapping, | |
912 | OSYNC_METADATA|OSYNC_DATA); | |
913 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 914 | |
a4514ebd JA |
915 | if (err) |
916 | ret = err; | |
917 | } | |
17ee4f49 | 918 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
4f6f0bd2 JA |
919 | } |
920 | ||
921 | return ret; | |
5274f052 JA |
922 | } |
923 | ||
059a8f37 JA |
924 | EXPORT_SYMBOL(generic_file_splice_write); |
925 | ||
83f9135b JA |
926 | /** |
927 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 928 | * @pipe: pipe to splice from |
83f9135b | 929 | * @out: socket to write to |
932cc6d4 | 930 | * @ppos: position in @out |
83f9135b JA |
931 | * @len: number of bytes to splice |
932 | * @flags: splice modifier flags | |
933 | * | |
932cc6d4 JA |
934 | * Description: |
935 | * Will send @len bytes from the pipe to a network socket. No data copying | |
936 | * is involved. | |
83f9135b JA |
937 | * |
938 | */ | |
3a326a2c | 939 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 940 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 941 | { |
00522fb4 | 942 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
943 | } |
944 | ||
059a8f37 | 945 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 946 | |
83f9135b JA |
947 | /* |
948 | * Attempt to initiate a splice from pipe to file. | |
949 | */ | |
3a326a2c | 950 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 951 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 952 | { |
5274f052 JA |
953 | int ret; |
954 | ||
49570e9b | 955 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
956 | return -EINVAL; |
957 | ||
49570e9b | 958 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
959 | return -EBADF; |
960 | ||
cbb7e577 | 961 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
962 | if (unlikely(ret < 0)) |
963 | return ret; | |
964 | ||
29ce2058 JM |
965 | ret = security_file_permission(out, MAY_WRITE); |
966 | if (unlikely(ret < 0)) | |
967 | return ret; | |
968 | ||
cbb7e577 | 969 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
970 | } |
971 | ||
83f9135b JA |
972 | /* |
973 | * Attempt to initiate a splice from a file to a pipe. | |
974 | */ | |
cbb7e577 JA |
975 | static long do_splice_to(struct file *in, loff_t *ppos, |
976 | struct pipe_inode_info *pipe, size_t len, | |
977 | unsigned int flags) | |
5274f052 | 978 | { |
5274f052 JA |
979 | int ret; |
980 | ||
49570e9b | 981 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
982 | return -EINVAL; |
983 | ||
49570e9b | 984 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
985 | return -EBADF; |
986 | ||
cbb7e577 | 987 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
988 | if (unlikely(ret < 0)) |
989 | return ret; | |
990 | ||
29ce2058 JM |
991 | ret = security_file_permission(in, MAY_READ); |
992 | if (unlikely(ret < 0)) | |
993 | return ret; | |
994 | ||
cbb7e577 | 995 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
996 | } |
997 | ||
932cc6d4 JA |
998 | /** |
999 | * splice_direct_to_actor - splices data directly between two non-pipes | |
1000 | * @in: file to splice from | |
1001 | * @sd: actor information on where to splice to | |
1002 | * @actor: handles the data splicing | |
1003 | * | |
1004 | * Description: | |
1005 | * This is a special case helper to splice directly between two | |
1006 | * points, without requiring an explicit pipe. Internally an allocated | |
1007 | * pipe is cached in the process, and reused during the life time of | |
1008 | * that process. | |
1009 | * | |
c66ab6fa JA |
1010 | */ |
1011 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1012 | splice_direct_actor *actor) | |
b92ce558 JA |
1013 | { |
1014 | struct pipe_inode_info *pipe; | |
1015 | long ret, bytes; | |
1016 | umode_t i_mode; | |
c66ab6fa JA |
1017 | size_t len; |
1018 | int i, flags; | |
b92ce558 JA |
1019 | |
1020 | /* | |
1021 | * We require the input being a regular file, as we don't want to | |
1022 | * randomly drop data for eg socket -> socket splicing. Use the | |
1023 | * piped splicing for that! | |
1024 | */ | |
0f7fc9e4 | 1025 | i_mode = in->f_path.dentry->d_inode->i_mode; |
b92ce558 JA |
1026 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1027 | return -EINVAL; | |
1028 | ||
1029 | /* | |
1030 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1031 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1032 | */ | |
1033 | pipe = current->splice_pipe; | |
49570e9b | 1034 | if (unlikely(!pipe)) { |
b92ce558 JA |
1035 | pipe = alloc_pipe_info(NULL); |
1036 | if (!pipe) | |
1037 | return -ENOMEM; | |
1038 | ||
1039 | /* | |
1040 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1041 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1042 | * PIPE_READERS appropriately. |
1043 | */ | |
1044 | pipe->readers = 1; | |
1045 | ||
1046 | current->splice_pipe = pipe; | |
1047 | } | |
1048 | ||
1049 | /* | |
73d62d83 | 1050 | * Do the splice. |
b92ce558 JA |
1051 | */ |
1052 | ret = 0; | |
1053 | bytes = 0; | |
c66ab6fa JA |
1054 | len = sd->total_len; |
1055 | flags = sd->flags; | |
1056 | ||
1057 | /* | |
1058 | * Don't block on output, we have to drain the direct pipe. | |
1059 | */ | |
1060 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
b92ce558 JA |
1061 | |
1062 | while (len) { | |
51a92c0f | 1063 | size_t read_len; |
bcd4f3ac | 1064 | loff_t pos = sd->pos; |
b92ce558 | 1065 | |
bcd4f3ac | 1066 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1067 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1068 | goto out_release; |
1069 | ||
1070 | read_len = ret; | |
c66ab6fa | 1071 | sd->total_len = read_len; |
b92ce558 JA |
1072 | |
1073 | /* | |
1074 | * NOTE: nonblocking mode only applies to the input. We | |
1075 | * must not do the output in nonblocking mode as then we | |
1076 | * could get stuck data in the internal pipe: | |
1077 | */ | |
c66ab6fa | 1078 | ret = actor(pipe, sd); |
51a92c0f | 1079 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1080 | goto out_release; |
1081 | ||
1082 | bytes += ret; | |
1083 | len -= ret; | |
bcd4f3ac | 1084 | sd->pos = pos; |
b92ce558 | 1085 | |
51a92c0f JA |
1086 | if (ret < read_len) |
1087 | goto out_release; | |
b92ce558 JA |
1088 | } |
1089 | ||
1090 | pipe->nrbufs = pipe->curbuf = 0; | |
b92ce558 JA |
1091 | return bytes; |
1092 | ||
1093 | out_release: | |
1094 | /* | |
1095 | * If we did an incomplete transfer we must release | |
1096 | * the pipe buffers in question: | |
1097 | */ | |
1098 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1099 | struct pipe_buffer *buf = pipe->bufs + i; | |
1100 | ||
1101 | if (buf->ops) { | |
1102 | buf->ops->release(pipe, buf); | |
1103 | buf->ops = NULL; | |
1104 | } | |
1105 | } | |
1106 | pipe->nrbufs = pipe->curbuf = 0; | |
1107 | ||
1108 | /* | |
1109 | * If we transferred some data, return the number of bytes: | |
1110 | */ | |
1111 | if (bytes > 0) | |
1112 | return bytes; | |
1113 | ||
1114 | return ret; | |
c66ab6fa JA |
1115 | |
1116 | } | |
1117 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1118 | ||
1119 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1120 | struct splice_desc *sd) | |
1121 | { | |
6a14b90b | 1122 | struct file *file = sd->u.file; |
c66ab6fa JA |
1123 | |
1124 | return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags); | |
1125 | } | |
1126 | ||
932cc6d4 JA |
1127 | /** |
1128 | * do_splice_direct - splices data directly between two files | |
1129 | * @in: file to splice from | |
1130 | * @ppos: input file offset | |
1131 | * @out: file to splice to | |
1132 | * @len: number of bytes to splice | |
1133 | * @flags: splice modifier flags | |
1134 | * | |
1135 | * Description: | |
1136 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1137 | * doing it in the application would incur an extra system call | |
1138 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1139 | * can splice directly through a process-private pipe. | |
1140 | * | |
1141 | */ | |
c66ab6fa JA |
1142 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
1143 | size_t len, unsigned int flags) | |
1144 | { | |
1145 | struct splice_desc sd = { | |
1146 | .len = len, | |
1147 | .total_len = len, | |
1148 | .flags = flags, | |
1149 | .pos = *ppos, | |
6a14b90b | 1150 | .u.file = out, |
c66ab6fa | 1151 | }; |
51a92c0f | 1152 | long ret; |
c66ab6fa JA |
1153 | |
1154 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); | |
51a92c0f JA |
1155 | if (ret > 0) |
1156 | *ppos += ret; | |
1157 | ||
c66ab6fa | 1158 | return ret; |
b92ce558 JA |
1159 | } |
1160 | ||
ddac0d39 JA |
1161 | /* |
1162 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1163 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1164 | * pipe. | |
1165 | */ | |
1166 | static inline struct pipe_inode_info *pipe_info(struct inode *inode) | |
1167 | { | |
1168 | if (S_ISFIFO(inode->i_mode)) | |
1169 | return inode->i_pipe; | |
1170 | ||
1171 | return NULL; | |
1172 | } | |
1173 | ||
83f9135b JA |
1174 | /* |
1175 | * Determine where to splice to/from. | |
1176 | */ | |
529565dc IM |
1177 | static long do_splice(struct file *in, loff_t __user *off_in, |
1178 | struct file *out, loff_t __user *off_out, | |
1179 | size_t len, unsigned int flags) | |
5274f052 | 1180 | { |
3a326a2c | 1181 | struct pipe_inode_info *pipe; |
cbb7e577 | 1182 | loff_t offset, *off; |
a4514ebd | 1183 | long ret; |
5274f052 | 1184 | |
0f7fc9e4 | 1185 | pipe = pipe_info(in->f_path.dentry->d_inode); |
529565dc IM |
1186 | if (pipe) { |
1187 | if (off_in) | |
1188 | return -ESPIPE; | |
b92ce558 JA |
1189 | if (off_out) { |
1190 | if (out->f_op->llseek == no_llseek) | |
1191 | return -EINVAL; | |
cbb7e577 | 1192 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1193 | return -EFAULT; |
cbb7e577 JA |
1194 | off = &offset; |
1195 | } else | |
1196 | off = &out->f_pos; | |
529565dc | 1197 | |
a4514ebd JA |
1198 | ret = do_splice_from(pipe, out, off, len, flags); |
1199 | ||
1200 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1201 | ret = -EFAULT; | |
1202 | ||
1203 | return ret; | |
529565dc | 1204 | } |
5274f052 | 1205 | |
0f7fc9e4 | 1206 | pipe = pipe_info(out->f_path.dentry->d_inode); |
529565dc IM |
1207 | if (pipe) { |
1208 | if (off_out) | |
1209 | return -ESPIPE; | |
b92ce558 JA |
1210 | if (off_in) { |
1211 | if (in->f_op->llseek == no_llseek) | |
1212 | return -EINVAL; | |
cbb7e577 | 1213 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1214 | return -EFAULT; |
cbb7e577 JA |
1215 | off = &offset; |
1216 | } else | |
1217 | off = &in->f_pos; | |
529565dc | 1218 | |
a4514ebd JA |
1219 | ret = do_splice_to(in, off, pipe, len, flags); |
1220 | ||
1221 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1222 | ret = -EFAULT; | |
1223 | ||
1224 | return ret; | |
529565dc | 1225 | } |
5274f052 JA |
1226 | |
1227 | return -EINVAL; | |
1228 | } | |
1229 | ||
912d35f8 JA |
1230 | /* |
1231 | * Map an iov into an array of pages and offset/length tupples. With the | |
1232 | * partial_page structure, we can map several non-contiguous ranges into | |
1233 | * our ones pages[] map instead of splitting that operation into pieces. | |
1234 | * Could easily be exported as a generic helper for other users, in which | |
1235 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1236 | */ | |
1237 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1238 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1239 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1240 | { |
1241 | int buffers = 0, error = 0; | |
1242 | ||
1243 | /* | |
1244 | * It's ok to take the mmap_sem for reading, even | |
1245 | * across a "get_user()". | |
1246 | */ | |
1247 | down_read(¤t->mm->mmap_sem); | |
1248 | ||
1249 | while (nr_vecs) { | |
1250 | unsigned long off, npages; | |
1251 | void __user *base; | |
1252 | size_t len; | |
1253 | int i; | |
1254 | ||
1255 | /* | |
1256 | * Get user address base and length for this iovec. | |
1257 | */ | |
1258 | error = get_user(base, &iov->iov_base); | |
1259 | if (unlikely(error)) | |
1260 | break; | |
1261 | error = get_user(len, &iov->iov_len); | |
1262 | if (unlikely(error)) | |
1263 | break; | |
1264 | ||
1265 | /* | |
1266 | * Sanity check this iovec. 0 read succeeds. | |
1267 | */ | |
1268 | if (unlikely(!len)) | |
1269 | break; | |
1270 | error = -EFAULT; | |
1271 | if (unlikely(!base)) | |
1272 | break; | |
1273 | ||
1274 | /* | |
1275 | * Get this base offset and number of pages, then map | |
1276 | * in the user pages. | |
1277 | */ | |
1278 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1279 | |
1280 | /* | |
1281 | * If asked for alignment, the offset must be zero and the | |
1282 | * length a multiple of the PAGE_SIZE. | |
1283 | */ | |
1284 | error = -EINVAL; | |
1285 | if (aligned && (off || len & ~PAGE_MASK)) | |
1286 | break; | |
1287 | ||
912d35f8 JA |
1288 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1289 | if (npages > PIPE_BUFFERS - buffers) | |
1290 | npages = PIPE_BUFFERS - buffers; | |
1291 | ||
1292 | error = get_user_pages(current, current->mm, | |
1293 | (unsigned long) base, npages, 0, 0, | |
1294 | &pages[buffers], NULL); | |
1295 | ||
1296 | if (unlikely(error <= 0)) | |
1297 | break; | |
1298 | ||
1299 | /* | |
1300 | * Fill this contiguous range into the partial page map. | |
1301 | */ | |
1302 | for (i = 0; i < error; i++) { | |
7591489a | 1303 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1304 | |
1305 | partial[buffers].offset = off; | |
1306 | partial[buffers].len = plen; | |
1307 | ||
1308 | off = 0; | |
1309 | len -= plen; | |
1310 | buffers++; | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * We didn't complete this iov, stop here since it probably | |
1315 | * means we have to move some of this into a pipe to | |
1316 | * be able to continue. | |
1317 | */ | |
1318 | if (len) | |
1319 | break; | |
1320 | ||
1321 | /* | |
1322 | * Don't continue if we mapped fewer pages than we asked for, | |
1323 | * or if we mapped the max number of pages that we have | |
1324 | * room for. | |
1325 | */ | |
1326 | if (error < npages || buffers == PIPE_BUFFERS) | |
1327 | break; | |
1328 | ||
1329 | nr_vecs--; | |
1330 | iov++; | |
1331 | } | |
1332 | ||
1333 | up_read(¤t->mm->mmap_sem); | |
1334 | ||
1335 | if (buffers) | |
1336 | return buffers; | |
1337 | ||
1338 | return error; | |
1339 | } | |
1340 | ||
6a14b90b JA |
1341 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1342 | struct splice_desc *sd) | |
1343 | { | |
1344 | char *src; | |
1345 | int ret; | |
1346 | ||
cac36bb0 | 1347 | ret = buf->ops->confirm(pipe, buf); |
6a14b90b JA |
1348 | if (unlikely(ret)) |
1349 | return ret; | |
1350 | ||
1351 | /* | |
1352 | * See if we can use the atomic maps, by prefaulting in the | |
1353 | * pages and doing an atomic copy | |
1354 | */ | |
1355 | if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { | |
1356 | src = buf->ops->map(pipe, buf, 1); | |
1357 | ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, | |
1358 | sd->len); | |
1359 | buf->ops->unmap(pipe, buf, src); | |
1360 | if (!ret) { | |
1361 | ret = sd->len; | |
1362 | goto out; | |
1363 | } | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * No dice, use slow non-atomic map and copy | |
1368 | */ | |
1369 | src = buf->ops->map(pipe, buf, 0); | |
1370 | ||
1371 | ret = sd->len; | |
1372 | if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) | |
1373 | ret = -EFAULT; | |
1374 | ||
1375 | out: | |
1376 | if (ret > 0) | |
1377 | sd->u.userptr += ret; | |
1378 | buf->ops->unmap(pipe, buf, src); | |
1379 | return ret; | |
1380 | } | |
1381 | ||
1382 | /* | |
1383 | * For lack of a better implementation, implement vmsplice() to userspace | |
1384 | * as a simple copy of the pipes pages to the user iov. | |
1385 | */ | |
1386 | static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, | |
1387 | unsigned long nr_segs, unsigned int flags) | |
1388 | { | |
1389 | struct pipe_inode_info *pipe; | |
1390 | struct splice_desc sd; | |
1391 | ssize_t size; | |
1392 | int error; | |
1393 | long ret; | |
1394 | ||
1395 | pipe = pipe_info(file->f_path.dentry->d_inode); | |
1396 | if (!pipe) | |
1397 | return -EBADF; | |
1398 | ||
1399 | if (pipe->inode) | |
1400 | mutex_lock(&pipe->inode->i_mutex); | |
1401 | ||
1402 | error = ret = 0; | |
1403 | while (nr_segs) { | |
1404 | void __user *base; | |
1405 | size_t len; | |
1406 | ||
1407 | /* | |
1408 | * Get user address base and length for this iovec. | |
1409 | */ | |
1410 | error = get_user(base, &iov->iov_base); | |
1411 | if (unlikely(error)) | |
1412 | break; | |
1413 | error = get_user(len, &iov->iov_len); | |
1414 | if (unlikely(error)) | |
1415 | break; | |
1416 | ||
1417 | /* | |
1418 | * Sanity check this iovec. 0 read succeeds. | |
1419 | */ | |
1420 | if (unlikely(!len)) | |
1421 | break; | |
1422 | if (unlikely(!base)) { | |
1423 | error = -EFAULT; | |
1424 | break; | |
1425 | } | |
1426 | ||
1427 | sd.len = 0; | |
1428 | sd.total_len = len; | |
1429 | sd.flags = flags; | |
1430 | sd.u.userptr = base; | |
1431 | sd.pos = 0; | |
1432 | ||
1433 | size = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1434 | if (size < 0) { | |
1435 | if (!ret) | |
1436 | ret = size; | |
1437 | ||
1438 | break; | |
1439 | } | |
1440 | ||
1441 | ret += size; | |
1442 | ||
1443 | if (size < len) | |
1444 | break; | |
1445 | ||
1446 | nr_segs--; | |
1447 | iov++; | |
1448 | } | |
1449 | ||
1450 | if (pipe->inode) | |
1451 | mutex_unlock(&pipe->inode->i_mutex); | |
1452 | ||
1453 | if (!ret) | |
1454 | ret = error; | |
1455 | ||
1456 | return ret; | |
1457 | } | |
1458 | ||
912d35f8 JA |
1459 | /* |
1460 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1461 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1462 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1463 | */ |
6a14b90b JA |
1464 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1465 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1466 | { |
ddac0d39 | 1467 | struct pipe_inode_info *pipe; |
912d35f8 JA |
1468 | struct page *pages[PIPE_BUFFERS]; |
1469 | struct partial_page partial[PIPE_BUFFERS]; | |
1470 | struct splice_pipe_desc spd = { | |
1471 | .pages = pages, | |
1472 | .partial = partial, | |
1473 | .flags = flags, | |
1474 | .ops = &user_page_pipe_buf_ops, | |
1475 | }; | |
1476 | ||
0f7fc9e4 | 1477 | pipe = pipe_info(file->f_path.dentry->d_inode); |
ddac0d39 | 1478 | if (!pipe) |
912d35f8 | 1479 | return -EBADF; |
912d35f8 | 1480 | |
7afa6fd0 JA |
1481 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1482 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1483 | if (spd.nr_pages <= 0) |
1484 | return spd.nr_pages; | |
1485 | ||
00522fb4 | 1486 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1487 | } |
1488 | ||
6a14b90b JA |
1489 | /* |
1490 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1491 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1492 | * operation that can be supported without any funky alignment restrictions | |
1493 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1494 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1495 | * solutions for that: | |
1496 | * | |
1497 | * - memcpy() the data internally, at which point we might as well just | |
1498 | * do a regular read() on the buffer anyway. | |
1499 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1500 | * has restriction limitations on both ends of the pipe). | |
1501 | * | |
1502 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1503 | * | |
1504 | */ | |
912d35f8 JA |
1505 | asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, |
1506 | unsigned long nr_segs, unsigned int flags) | |
1507 | { | |
1508 | struct file *file; | |
1509 | long error; | |
1510 | int fput; | |
1511 | ||
6a14b90b JA |
1512 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1513 | return -EINVAL; | |
1514 | else if (unlikely(!nr_segs)) | |
1515 | return 0; | |
1516 | ||
912d35f8 JA |
1517 | error = -EBADF; |
1518 | file = fget_light(fd, &fput); | |
1519 | if (file) { | |
1520 | if (file->f_mode & FMODE_WRITE) | |
6a14b90b JA |
1521 | error = vmsplice_to_pipe(file, iov, nr_segs, flags); |
1522 | else if (file->f_mode & FMODE_READ) | |
1523 | error = vmsplice_to_user(file, iov, nr_segs, flags); | |
912d35f8 JA |
1524 | |
1525 | fput_light(file, fput); | |
1526 | } | |
1527 | ||
1528 | return error; | |
1529 | } | |
1530 | ||
529565dc IM |
1531 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
1532 | int fd_out, loff_t __user *off_out, | |
1533 | size_t len, unsigned int flags) | |
5274f052 JA |
1534 | { |
1535 | long error; | |
1536 | struct file *in, *out; | |
1537 | int fput_in, fput_out; | |
1538 | ||
1539 | if (unlikely(!len)) | |
1540 | return 0; | |
1541 | ||
1542 | error = -EBADF; | |
529565dc | 1543 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1544 | if (in) { |
1545 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1546 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1547 | if (out) { |
1548 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1549 | error = do_splice(in, off_in, |
1550 | out, off_out, | |
1551 | len, flags); | |
5274f052 JA |
1552 | fput_light(out, fput_out); |
1553 | } | |
1554 | } | |
1555 | ||
1556 | fput_light(in, fput_in); | |
1557 | } | |
1558 | ||
1559 | return error; | |
1560 | } | |
70524490 | 1561 | |
aadd06e5 JA |
1562 | /* |
1563 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1564 | * return an appropriate error. | |
1565 | */ | |
1566 | static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1567 | { | |
1568 | int ret; | |
1569 | ||
1570 | /* | |
1571 | * Check ->nrbufs without the inode lock first. This function | |
1572 | * is speculative anyways, so missing one is ok. | |
1573 | */ | |
1574 | if (pipe->nrbufs) | |
1575 | return 0; | |
1576 | ||
1577 | ret = 0; | |
1578 | mutex_lock(&pipe->inode->i_mutex); | |
1579 | ||
1580 | while (!pipe->nrbufs) { | |
1581 | if (signal_pending(current)) { | |
1582 | ret = -ERESTARTSYS; | |
1583 | break; | |
1584 | } | |
1585 | if (!pipe->writers) | |
1586 | break; | |
1587 | if (!pipe->waiting_writers) { | |
1588 | if (flags & SPLICE_F_NONBLOCK) { | |
1589 | ret = -EAGAIN; | |
1590 | break; | |
1591 | } | |
1592 | } | |
1593 | pipe_wait(pipe); | |
1594 | } | |
1595 | ||
1596 | mutex_unlock(&pipe->inode->i_mutex); | |
1597 | return ret; | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1602 | * return an appropriate error. | |
1603 | */ | |
1604 | static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1605 | { | |
1606 | int ret; | |
1607 | ||
1608 | /* | |
1609 | * Check ->nrbufs without the inode lock first. This function | |
1610 | * is speculative anyways, so missing one is ok. | |
1611 | */ | |
1612 | if (pipe->nrbufs < PIPE_BUFFERS) | |
1613 | return 0; | |
1614 | ||
1615 | ret = 0; | |
1616 | mutex_lock(&pipe->inode->i_mutex); | |
1617 | ||
1618 | while (pipe->nrbufs >= PIPE_BUFFERS) { | |
1619 | if (!pipe->readers) { | |
1620 | send_sig(SIGPIPE, current, 0); | |
1621 | ret = -EPIPE; | |
1622 | break; | |
1623 | } | |
1624 | if (flags & SPLICE_F_NONBLOCK) { | |
1625 | ret = -EAGAIN; | |
1626 | break; | |
1627 | } | |
1628 | if (signal_pending(current)) { | |
1629 | ret = -ERESTARTSYS; | |
1630 | break; | |
1631 | } | |
1632 | pipe->waiting_writers++; | |
1633 | pipe_wait(pipe); | |
1634 | pipe->waiting_writers--; | |
1635 | } | |
1636 | ||
1637 | mutex_unlock(&pipe->inode->i_mutex); | |
1638 | return ret; | |
1639 | } | |
1640 | ||
70524490 JA |
1641 | /* |
1642 | * Link contents of ipipe to opipe. | |
1643 | */ | |
1644 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1645 | struct pipe_inode_info *opipe, | |
1646 | size_t len, unsigned int flags) | |
1647 | { | |
1648 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1649 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1650 | |
1651 | /* | |
1652 | * Potential ABBA deadlock, work around it by ordering lock | |
1653 | * grabbing by inode address. Otherwise two different processes | |
1654 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1655 | */ | |
62752ee1 | 1656 | inode_double_lock(ipipe->inode, opipe->inode); |
70524490 | 1657 | |
aadd06e5 | 1658 | do { |
70524490 JA |
1659 | if (!opipe->readers) { |
1660 | send_sig(SIGPIPE, current, 0); | |
1661 | if (!ret) | |
1662 | ret = -EPIPE; | |
1663 | break; | |
1664 | } | |
70524490 | 1665 | |
aadd06e5 JA |
1666 | /* |
1667 | * If we have iterated all input buffers or ran out of | |
1668 | * output room, break. | |
1669 | */ | |
1670 | if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) | |
1671 | break; | |
70524490 | 1672 | |
aadd06e5 JA |
1673 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); |
1674 | nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
70524490 JA |
1675 | |
1676 | /* | |
aadd06e5 JA |
1677 | * Get a reference to this pipe buffer, |
1678 | * so we can copy the contents over. | |
70524490 | 1679 | */ |
aadd06e5 JA |
1680 | ibuf->ops->get(ipipe, ibuf); |
1681 | ||
1682 | obuf = opipe->bufs + nbuf; | |
1683 | *obuf = *ibuf; | |
1684 | ||
2a27250e | 1685 | /* |
aadd06e5 JA |
1686 | * Don't inherit the gift flag, we need to |
1687 | * prevent multiple steals of this page. | |
2a27250e | 1688 | */ |
aadd06e5 | 1689 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1690 | |
aadd06e5 JA |
1691 | if (obuf->len > len) |
1692 | obuf->len = len; | |
70524490 | 1693 | |
aadd06e5 JA |
1694 | opipe->nrbufs++; |
1695 | ret += obuf->len; | |
1696 | len -= obuf->len; | |
1697 | i++; | |
1698 | } while (len); | |
70524490 | 1699 | |
62752ee1 | 1700 | inode_double_unlock(ipipe->inode, opipe->inode); |
70524490 | 1701 | |
aadd06e5 JA |
1702 | /* |
1703 | * If we put data in the output pipe, wakeup any potential readers. | |
1704 | */ | |
1705 | if (ret > 0) { | |
70524490 JA |
1706 | smp_mb(); |
1707 | if (waitqueue_active(&opipe->wait)) | |
1708 | wake_up_interruptible(&opipe->wait); | |
1709 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1710 | } | |
1711 | ||
1712 | return ret; | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1717 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1718 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1719 | * applicable one is SPLICE_F_NONBLOCK. | |
1720 | */ | |
1721 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1722 | unsigned int flags) | |
1723 | { | |
0f7fc9e4 JJS |
1724 | struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); |
1725 | struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); | |
aadd06e5 | 1726 | int ret = -EINVAL; |
70524490 JA |
1727 | |
1728 | /* | |
aadd06e5 JA |
1729 | * Duplicate the contents of ipipe to opipe without actually |
1730 | * copying the data. | |
70524490 | 1731 | */ |
aadd06e5 JA |
1732 | if (ipipe && opipe && ipipe != opipe) { |
1733 | /* | |
1734 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1735 | * ordering doesn't really matter. | |
1736 | */ | |
1737 | ret = link_ipipe_prep(ipipe, flags); | |
1738 | if (!ret) { | |
1739 | ret = link_opipe_prep(opipe, flags); | |
1740 | if (!ret) { | |
1741 | ret = link_pipe(ipipe, opipe, len, flags); | |
1742 | if (!ret && (flags & SPLICE_F_NONBLOCK)) | |
1743 | ret = -EAGAIN; | |
1744 | } | |
1745 | } | |
1746 | } | |
70524490 | 1747 | |
aadd06e5 | 1748 | return ret; |
70524490 JA |
1749 | } |
1750 | ||
1751 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1752 | { | |
1753 | struct file *in; | |
1754 | int error, fput_in; | |
1755 | ||
1756 | if (unlikely(!len)) | |
1757 | return 0; | |
1758 | ||
1759 | error = -EBADF; | |
1760 | in = fget_light(fdin, &fput_in); | |
1761 | if (in) { | |
1762 | if (in->f_mode & FMODE_READ) { | |
1763 | int fput_out; | |
1764 | struct file *out = fget_light(fdout, &fput_out); | |
1765 | ||
1766 | if (out) { | |
1767 | if (out->f_mode & FMODE_WRITE) | |
1768 | error = do_tee(in, out, len, flags); | |
1769 | fput_light(out, fput_out); | |
1770 | } | |
1771 | } | |
1772 | fput_light(in, fput_in); | |
1773 | } | |
1774 | ||
1775 | return error; | |
1776 | } |