]>
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 | ||
eb20796b JA |
289 | /* |
290 | * Lookup the (hopefully) full range of pages we need. | |
291 | */ | |
292 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
431a4820 | 293 | index += spd.nr_pages; |
82aa5d61 | 294 | |
eb20796b JA |
295 | /* |
296 | * If find_get_pages_contig() returned fewer pages than we needed, | |
431a4820 | 297 | * readahead/allocate the rest and fill in the holes. |
eb20796b | 298 | */ |
431a4820 FW |
299 | if (spd.nr_pages < nr_pages) |
300 | page_cache_readahead_ondemand(mapping, &in->f_ra, in, | |
301 | NULL, index, nr_pages - spd.nr_pages); | |
302 | ||
932cc6d4 | 303 | error = 0; |
eb20796b | 304 | while (spd.nr_pages < nr_pages) { |
82aa5d61 | 305 | /* |
eb20796b JA |
306 | * Page could be there, find_get_pages_contig() breaks on |
307 | * the first hole. | |
5274f052 | 308 | */ |
7480a904 JA |
309 | page = find_get_page(mapping, index); |
310 | if (!page) { | |
7480a904 | 311 | /* |
eb20796b | 312 | * page didn't exist, allocate one. |
7480a904 JA |
313 | */ |
314 | page = page_cache_alloc_cold(mapping); | |
315 | if (!page) | |
316 | break; | |
317 | ||
318 | error = add_to_page_cache_lru(page, mapping, index, | |
2ae88149 | 319 | GFP_KERNEL); |
7480a904 JA |
320 | if (unlikely(error)) { |
321 | page_cache_release(page); | |
a0548871 JA |
322 | if (error == -EEXIST) |
323 | continue; | |
7480a904 JA |
324 | break; |
325 | } | |
eb20796b JA |
326 | /* |
327 | * add_to_page_cache() locks the page, unlock it | |
328 | * to avoid convoluting the logic below even more. | |
329 | */ | |
330 | unlock_page(page); | |
7480a904 JA |
331 | } |
332 | ||
eb20796b JA |
333 | pages[spd.nr_pages++] = page; |
334 | index++; | |
335 | } | |
336 | ||
337 | /* | |
338 | * Now loop over the map and see if we need to start IO on any | |
339 | * pages, fill in the partial map, etc. | |
340 | */ | |
341 | index = *ppos >> PAGE_CACHE_SHIFT; | |
342 | nr_pages = spd.nr_pages; | |
343 | spd.nr_pages = 0; | |
344 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
345 | unsigned int this_len; | |
346 | ||
347 | if (!len) | |
348 | break; | |
349 | ||
350 | /* | |
351 | * this_len is the max we'll use from this page | |
352 | */ | |
353 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
354 | page = pages[page_nr]; | |
355 | ||
a08a166f FW |
356 | if (PageReadahead(page)) |
357 | page_cache_readahead_ondemand(mapping, &in->f_ra, in, | |
358 | page, index, nr_pages - page_nr); | |
359 | ||
7480a904 JA |
360 | /* |
361 | * If the page isn't uptodate, we may need to start io on it | |
362 | */ | |
363 | if (!PageUptodate(page)) { | |
c4f895cb JA |
364 | /* |
365 | * If in nonblock mode then dont block on waiting | |
366 | * for an in-flight io page | |
367 | */ | |
9ae9d68c FW |
368 | if (flags & SPLICE_F_NONBLOCK) { |
369 | if (TestSetPageLocked(page)) | |
370 | break; | |
371 | } else | |
372 | lock_page(page); | |
7480a904 JA |
373 | |
374 | /* | |
375 | * page was truncated, stop here. if this isn't the | |
376 | * first page, we'll just complete what we already | |
377 | * added | |
378 | */ | |
379 | if (!page->mapping) { | |
380 | unlock_page(page); | |
7480a904 JA |
381 | break; |
382 | } | |
383 | /* | |
384 | * page was already under io and is now done, great | |
385 | */ | |
386 | if (PageUptodate(page)) { | |
387 | unlock_page(page); | |
388 | goto fill_it; | |
389 | } | |
5274f052 | 390 | |
7480a904 JA |
391 | /* |
392 | * need to read in the page | |
393 | */ | |
394 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 395 | if (unlikely(error)) { |
eb20796b JA |
396 | /* |
397 | * We really should re-lookup the page here, | |
398 | * but it complicates things a lot. Instead | |
399 | * lets just do what we already stored, and | |
400 | * we'll get it the next time we are called. | |
401 | */ | |
7480a904 | 402 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
403 | error = 0; |
404 | ||
5274f052 JA |
405 | break; |
406 | } | |
620a324b JA |
407 | } |
408 | fill_it: | |
409 | /* | |
410 | * i_size must be checked after PageUptodate. | |
411 | */ | |
412 | isize = i_size_read(mapping->host); | |
413 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
414 | if (unlikely(!isize || index > end_index)) | |
415 | break; | |
416 | ||
417 | /* | |
418 | * if this is the last page, see if we need to shrink | |
419 | * the length and stop | |
420 | */ | |
421 | if (end_index == index) { | |
422 | unsigned int plen; | |
91ad66ef JA |
423 | |
424 | /* | |
620a324b | 425 | * max good bytes in this page |
91ad66ef | 426 | */ |
620a324b JA |
427 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
428 | if (plen <= loff) | |
91ad66ef | 429 | break; |
91ad66ef JA |
430 | |
431 | /* | |
620a324b | 432 | * force quit after adding this page |
91ad66ef | 433 | */ |
620a324b JA |
434 | this_len = min(this_len, plen - loff); |
435 | len = this_len; | |
5274f052 | 436 | } |
620a324b | 437 | |
eb20796b JA |
438 | partial[page_nr].offset = loff; |
439 | partial[page_nr].len = this_len; | |
82aa5d61 | 440 | len -= this_len; |
91ad66ef | 441 | loff = 0; |
eb20796b JA |
442 | spd.nr_pages++; |
443 | index++; | |
5274f052 JA |
444 | } |
445 | ||
eb20796b | 446 | /* |
475ecade | 447 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
448 | * we got, 'nr_pages' is how many pages are in the map. |
449 | */ | |
450 | while (page_nr < nr_pages) | |
451 | page_cache_release(pages[page_nr++]); | |
a08a166f | 452 | in->f_ra.prev_index = index; |
eb20796b | 453 | |
912d35f8 | 454 | if (spd.nr_pages) |
00522fb4 | 455 | return splice_to_pipe(pipe, &spd); |
5274f052 | 456 | |
7480a904 | 457 | return error; |
5274f052 JA |
458 | } |
459 | ||
83f9135b JA |
460 | /** |
461 | * generic_file_splice_read - splice data from file to a pipe | |
462 | * @in: file to splice from | |
932cc6d4 | 463 | * @ppos: position in @in |
83f9135b JA |
464 | * @pipe: pipe to splice to |
465 | * @len: number of bytes to splice | |
466 | * @flags: splice modifier flags | |
467 | * | |
932cc6d4 JA |
468 | * Description: |
469 | * Will read pages from given file and fill them into a pipe. Can be | |
470 | * used as long as the address_space operations for the source implements | |
471 | * a readpage() hook. | |
472 | * | |
83f9135b | 473 | */ |
cbb7e577 JA |
474 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
475 | struct pipe_inode_info *pipe, size_t len, | |
476 | unsigned int flags) | |
5274f052 JA |
477 | { |
478 | ssize_t spliced; | |
479 | int ret; | |
d366d398 JA |
480 | loff_t isize, left; |
481 | ||
482 | isize = i_size_read(in->f_mapping->host); | |
483 | if (unlikely(*ppos >= isize)) | |
484 | return 0; | |
485 | ||
486 | left = isize - *ppos; | |
487 | if (unlikely(left < len)) | |
488 | len = left; | |
5274f052 JA |
489 | |
490 | ret = 0; | |
491 | spliced = 0; | |
51a92c0f | 492 | while (len && !spliced) { |
cbb7e577 | 493 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 494 | |
c4f895cb | 495 | if (ret < 0) |
5274f052 | 496 | break; |
c4f895cb JA |
497 | else if (!ret) { |
498 | if (spliced) | |
499 | break; | |
500 | if (flags & SPLICE_F_NONBLOCK) { | |
501 | ret = -EAGAIN; | |
502 | break; | |
503 | } | |
504 | } | |
5274f052 | 505 | |
cbb7e577 | 506 | *ppos += ret; |
5274f052 JA |
507 | len -= ret; |
508 | spliced += ret; | |
509 | } | |
510 | ||
511 | if (spliced) | |
512 | return spliced; | |
513 | ||
514 | return ret; | |
515 | } | |
516 | ||
059a8f37 JA |
517 | EXPORT_SYMBOL(generic_file_splice_read); |
518 | ||
5274f052 | 519 | /* |
4f6f0bd2 | 520 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 521 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 522 | */ |
76ad4d11 | 523 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
524 | struct pipe_buffer *buf, struct splice_desc *sd) |
525 | { | |
6a14b90b | 526 | struct file *file = sd->u.file; |
5274f052 | 527 | loff_t pos = sd->pos; |
f84d7519 | 528 | int ret, more; |
5274f052 | 529 | |
cac36bb0 | 530 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
531 | if (!ret) { |
532 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 533 | |
f84d7519 JA |
534 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
535 | sd->len, &pos, more); | |
536 | } | |
5274f052 | 537 | |
016b661e | 538 | return ret; |
5274f052 JA |
539 | } |
540 | ||
541 | /* | |
542 | * This is a little more tricky than the file -> pipe splicing. There are | |
543 | * basically three cases: | |
544 | * | |
545 | * - Destination page already exists in the address space and there | |
546 | * are users of it. For that case we have no other option that | |
547 | * copying the data. Tough luck. | |
548 | * - Destination page already exists in the address space, but there | |
549 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
550 | * through to last case. | |
551 | * - Destination page does not exist, we can add the pipe page to | |
552 | * the page cache and avoid the copy. | |
553 | * | |
83f9135b JA |
554 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
555 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
556 | * file address space page cache. This is possible if no one else has | |
557 | * the pipe page referenced outside of the pipe and page cache. If | |
558 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
559 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 560 | */ |
76ad4d11 | 561 | static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
5274f052 JA |
562 | struct splice_desc *sd) |
563 | { | |
6a14b90b | 564 | struct file *file = sd->u.file; |
5274f052 | 565 | struct address_space *mapping = file->f_mapping; |
016b661e | 566 | unsigned int offset, this_len; |
5274f052 | 567 | struct page *page; |
5274f052 | 568 | pgoff_t index; |
3e7ee3e7 | 569 | int ret; |
5274f052 JA |
570 | |
571 | /* | |
49d0b21b | 572 | * make sure the data in this buffer is uptodate |
5274f052 | 573 | */ |
cac36bb0 | 574 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
575 | if (unlikely(ret)) |
576 | return ret; | |
5274f052 JA |
577 | |
578 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
579 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
580 | ||
016b661e JA |
581 | this_len = sd->len; |
582 | if (this_len + offset > PAGE_CACHE_SIZE) | |
583 | this_len = PAGE_CACHE_SIZE - offset; | |
584 | ||
485ddb4b NP |
585 | find_page: |
586 | page = find_lock_page(mapping, index); | |
587 | if (!page) { | |
588 | ret = -ENOMEM; | |
589 | page = page_cache_alloc_cold(mapping); | |
590 | if (unlikely(!page)) | |
591 | goto out_ret; | |
592 | ||
83f9135b | 593 | /* |
485ddb4b | 594 | * This will also lock the page |
83f9135b | 595 | */ |
485ddb4b NP |
596 | ret = add_to_page_cache_lru(page, mapping, index, |
597 | GFP_KERNEL); | |
598 | if (unlikely(ret)) | |
599 | goto out; | |
600 | } | |
9e0267c2 | 601 | |
016b661e | 602 | ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); |
bfc4ee39 JA |
603 | if (unlikely(ret)) { |
604 | loff_t isize = i_size_read(mapping->host); | |
605 | ||
606 | if (ret != AOP_TRUNCATED_PAGE) | |
607 | unlock_page(page); | |
4f6f0bd2 | 608 | page_cache_release(page); |
bfc4ee39 JA |
609 | if (ret == AOP_TRUNCATED_PAGE) |
610 | goto find_page; | |
611 | ||
612 | /* | |
613 | * prepare_write() may have instantiated a few blocks | |
614 | * outside i_size. Trim these off again. | |
615 | */ | |
616 | if (sd->pos + this_len > isize) | |
617 | vmtruncate(mapping->host, isize); | |
618 | ||
e6e80f29 | 619 | goto out_ret; |
bfc4ee39 | 620 | } |
5274f052 | 621 | |
0568b409 | 622 | if (buf->page != page) { |
f84d7519 JA |
623 | /* |
624 | * Careful, ->map() uses KM_USER0! | |
625 | */ | |
76ad4d11 | 626 | char *src = buf->ops->map(pipe, buf, 1); |
f84d7519 | 627 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 628 | |
016b661e | 629 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 630 | flush_dcache_page(page); |
f84d7519 | 631 | kunmap_atomic(dst, KM_USER1); |
76ad4d11 | 632 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 633 | } |
5274f052 | 634 | |
016b661e | 635 | ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); |
d9993c37 DM |
636 | if (ret) { |
637 | if (ret == AOP_TRUNCATED_PAGE) { | |
638 | page_cache_release(page); | |
639 | goto find_page; | |
640 | } | |
641 | if (ret < 0) | |
642 | goto out; | |
0568b409 | 643 | /* |
d9993c37 DM |
644 | * Partial write has happened, so 'ret' already initialized by |
645 | * number of bytes written, Where is nothing we have to do here. | |
0568b409 | 646 | */ |
d9993c37 | 647 | } else |
0568b409 | 648 | ret = this_len; |
d9993c37 DM |
649 | /* |
650 | * Return the number of bytes written and mark page as | |
651 | * accessed, we are now done! | |
652 | */ | |
653 | mark_page_accessed(page); | |
5274f052 | 654 | out: |
0568b409 | 655 | page_cache_release(page); |
9e0267c2 | 656 | unlock_page(page); |
e6e80f29 | 657 | out_ret: |
5274f052 JA |
658 | return ret; |
659 | } | |
660 | ||
932cc6d4 JA |
661 | /** |
662 | * __splice_from_pipe - splice data from a pipe to given actor | |
663 | * @pipe: pipe to splice from | |
664 | * @sd: information to @actor | |
665 | * @actor: handler that splices the data | |
666 | * | |
667 | * Description: | |
668 | * This function does little more than loop over the pipe and call | |
669 | * @actor to do the actual moving of a single struct pipe_buffer to | |
670 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
671 | * pipe_to_user. | |
672 | * | |
83f9135b | 673 | */ |
c66ab6fa JA |
674 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, |
675 | splice_actor *actor) | |
5274f052 | 676 | { |
5274f052 | 677 | int ret, do_wakeup, err; |
5274f052 JA |
678 | |
679 | ret = 0; | |
680 | do_wakeup = 0; | |
681 | ||
5274f052 | 682 | for (;;) { |
6f767b04 JA |
683 | if (pipe->nrbufs) { |
684 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
d4c3cca9 | 685 | const struct pipe_buf_operations *ops = buf->ops; |
5274f052 | 686 | |
c66ab6fa JA |
687 | sd->len = buf->len; |
688 | if (sd->len > sd->total_len) | |
689 | sd->len = sd->total_len; | |
5274f052 | 690 | |
c66ab6fa | 691 | err = actor(pipe, buf, sd); |
016b661e | 692 | if (err <= 0) { |
5274f052 JA |
693 | if (!ret && err != -ENODATA) |
694 | ret = err; | |
695 | ||
696 | break; | |
697 | } | |
698 | ||
016b661e JA |
699 | ret += err; |
700 | buf->offset += err; | |
701 | buf->len -= err; | |
702 | ||
c66ab6fa JA |
703 | sd->len -= err; |
704 | sd->pos += err; | |
705 | sd->total_len -= err; | |
706 | if (sd->len) | |
016b661e | 707 | continue; |
73d62d83 | 708 | |
5274f052 JA |
709 | if (!buf->len) { |
710 | buf->ops = NULL; | |
3a326a2c | 711 | ops->release(pipe, buf); |
6f767b04 JA |
712 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
713 | pipe->nrbufs--; | |
714 | if (pipe->inode) | |
715 | do_wakeup = 1; | |
5274f052 JA |
716 | } |
717 | ||
c66ab6fa | 718 | if (!sd->total_len) |
5274f052 JA |
719 | break; |
720 | } | |
721 | ||
6f767b04 | 722 | if (pipe->nrbufs) |
5274f052 | 723 | continue; |
3a326a2c | 724 | if (!pipe->writers) |
5274f052 | 725 | break; |
3a326a2c | 726 | if (!pipe->waiting_writers) { |
5274f052 JA |
727 | if (ret) |
728 | break; | |
729 | } | |
730 | ||
c66ab6fa | 731 | if (sd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
732 | if (!ret) |
733 | ret = -EAGAIN; | |
734 | break; | |
735 | } | |
736 | ||
5274f052 JA |
737 | if (signal_pending(current)) { |
738 | if (!ret) | |
739 | ret = -ERESTARTSYS; | |
740 | break; | |
741 | } | |
742 | ||
743 | if (do_wakeup) { | |
c0bd1f65 | 744 | smp_mb(); |
3a326a2c IM |
745 | if (waitqueue_active(&pipe->wait)) |
746 | wake_up_interruptible_sync(&pipe->wait); | |
747 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
748 | do_wakeup = 0; |
749 | } | |
750 | ||
3a326a2c | 751 | pipe_wait(pipe); |
5274f052 JA |
752 | } |
753 | ||
5274f052 | 754 | if (do_wakeup) { |
c0bd1f65 | 755 | smp_mb(); |
3a326a2c IM |
756 | if (waitqueue_active(&pipe->wait)) |
757 | wake_up_interruptible(&pipe->wait); | |
758 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
759 | } |
760 | ||
5274f052 | 761 | return ret; |
5274f052 | 762 | } |
40bee44e | 763 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 764 | |
932cc6d4 JA |
765 | /** |
766 | * splice_from_pipe - splice data from a pipe to a file | |
767 | * @pipe: pipe to splice from | |
768 | * @out: file to splice to | |
769 | * @ppos: position in @out | |
770 | * @len: how many bytes to splice | |
771 | * @flags: splice modifier flags | |
772 | * @actor: handler that splices the data | |
773 | * | |
774 | * Description: | |
775 | * See __splice_from_pipe. This function locks the input and output inodes, | |
776 | * otherwise it's identical to __splice_from_pipe(). | |
777 | * | |
778 | */ | |
6da61809 MF |
779 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
780 | loff_t *ppos, size_t len, unsigned int flags, | |
781 | splice_actor *actor) | |
782 | { | |
783 | ssize_t ret; | |
784 | struct inode *inode = out->f_mapping->host; | |
c66ab6fa JA |
785 | struct splice_desc sd = { |
786 | .total_len = len, | |
787 | .flags = flags, | |
788 | .pos = *ppos, | |
6a14b90b | 789 | .u.file = out, |
c66ab6fa | 790 | }; |
6da61809 MF |
791 | |
792 | /* | |
793 | * The actor worker might be calling ->prepare_write and | |
794 | * ->commit_write. Most of the time, these expect i_mutex to | |
795 | * be held. Since this may result in an ABBA deadlock with | |
796 | * pipe->inode, we have to order lock acquiry here. | |
797 | */ | |
798 | inode_double_lock(inode, pipe->inode); | |
c66ab6fa | 799 | ret = __splice_from_pipe(pipe, &sd, actor); |
6da61809 MF |
800 | inode_double_unlock(inode, pipe->inode); |
801 | ||
802 | return ret; | |
803 | } | |
804 | ||
805 | /** | |
806 | * generic_file_splice_write_nolock - generic_file_splice_write without mutexes | |
807 | * @pipe: pipe info | |
808 | * @out: file to write to | |
932cc6d4 | 809 | * @ppos: position in @out |
6da61809 MF |
810 | * @len: number of bytes to splice |
811 | * @flags: splice modifier flags | |
812 | * | |
932cc6d4 JA |
813 | * Description: |
814 | * Will either move or copy pages (determined by @flags options) from | |
815 | * the given pipe inode to the given file. The caller is responsible | |
816 | * for acquiring i_mutex on both inodes. | |
6da61809 MF |
817 | * |
818 | */ | |
819 | ssize_t | |
820 | generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out, | |
821 | loff_t *ppos, size_t len, unsigned int flags) | |
822 | { | |
823 | struct address_space *mapping = out->f_mapping; | |
824 | struct inode *inode = mapping->host; | |
c66ab6fa JA |
825 | struct splice_desc sd = { |
826 | .total_len = len, | |
827 | .flags = flags, | |
828 | .pos = *ppos, | |
6a14b90b | 829 | .u.file = out, |
c66ab6fa | 830 | }; |
6da61809 MF |
831 | ssize_t ret; |
832 | int err; | |
833 | ||
0f7fc9e4 | 834 | err = remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
835 | if (unlikely(err)) |
836 | return err; | |
837 | ||
c66ab6fa | 838 | ret = __splice_from_pipe(pipe, &sd, pipe_to_file); |
6da61809 | 839 | if (ret > 0) { |
17ee4f49 JA |
840 | unsigned long nr_pages; |
841 | ||
6da61809 | 842 | *ppos += ret; |
17ee4f49 | 843 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
6da61809 MF |
844 | |
845 | /* | |
846 | * If file or inode is SYNC and we actually wrote some data, | |
847 | * sync it. | |
848 | */ | |
849 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
850 | err = generic_osync_inode(inode, mapping, | |
851 | OSYNC_METADATA|OSYNC_DATA); | |
852 | ||
853 | if (err) | |
854 | ret = err; | |
855 | } | |
17ee4f49 | 856 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
6da61809 MF |
857 | } |
858 | ||
859 | return ret; | |
860 | } | |
861 | ||
862 | EXPORT_SYMBOL(generic_file_splice_write_nolock); | |
863 | ||
83f9135b JA |
864 | /** |
865 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 866 | * @pipe: pipe info |
83f9135b | 867 | * @out: file to write to |
932cc6d4 | 868 | * @ppos: position in @out |
83f9135b JA |
869 | * @len: number of bytes to splice |
870 | * @flags: splice modifier flags | |
871 | * | |
932cc6d4 JA |
872 | * Description: |
873 | * Will either move or copy pages (determined by @flags options) from | |
874 | * the given pipe inode to the given file. | |
83f9135b JA |
875 | * |
876 | */ | |
3a326a2c IM |
877 | ssize_t |
878 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 879 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 880 | { |
4f6f0bd2 | 881 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 882 | struct inode *inode = mapping->host; |
3a326a2c | 883 | ssize_t ret; |
8c34e2d6 JA |
884 | int err; |
885 | ||
0f7fc9e4 | 886 | err = should_remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
887 | if (unlikely(err)) { |
888 | mutex_lock(&inode->i_mutex); | |
0f7fc9e4 | 889 | err = __remove_suid(out->f_path.dentry, err); |
8c34e2d6 JA |
890 | mutex_unlock(&inode->i_mutex); |
891 | if (err) | |
892 | return err; | |
893 | } | |
3a326a2c | 894 | |
00522fb4 | 895 | ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
a4514ebd | 896 | if (ret > 0) { |
17ee4f49 JA |
897 | unsigned long nr_pages; |
898 | ||
a4514ebd | 899 | *ppos += ret; |
17ee4f49 | 900 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
a4514ebd JA |
901 | |
902 | /* | |
903 | * If file or inode is SYNC and we actually wrote some data, | |
904 | * sync it. | |
905 | */ | |
906 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
a4514ebd JA |
907 | mutex_lock(&inode->i_mutex); |
908 | err = generic_osync_inode(inode, mapping, | |
909 | OSYNC_METADATA|OSYNC_DATA); | |
910 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 911 | |
a4514ebd JA |
912 | if (err) |
913 | ret = err; | |
914 | } | |
17ee4f49 | 915 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
4f6f0bd2 JA |
916 | } |
917 | ||
918 | return ret; | |
5274f052 JA |
919 | } |
920 | ||
059a8f37 JA |
921 | EXPORT_SYMBOL(generic_file_splice_write); |
922 | ||
83f9135b JA |
923 | /** |
924 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 925 | * @pipe: pipe to splice from |
83f9135b | 926 | * @out: socket to write to |
932cc6d4 | 927 | * @ppos: position in @out |
83f9135b JA |
928 | * @len: number of bytes to splice |
929 | * @flags: splice modifier flags | |
930 | * | |
932cc6d4 JA |
931 | * Description: |
932 | * Will send @len bytes from the pipe to a network socket. No data copying | |
933 | * is involved. | |
83f9135b JA |
934 | * |
935 | */ | |
3a326a2c | 936 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 937 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 938 | { |
00522fb4 | 939 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
940 | } |
941 | ||
059a8f37 | 942 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 943 | |
83f9135b JA |
944 | /* |
945 | * Attempt to initiate a splice from pipe to file. | |
946 | */ | |
3a326a2c | 947 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 948 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 949 | { |
5274f052 JA |
950 | int ret; |
951 | ||
49570e9b | 952 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
953 | return -EINVAL; |
954 | ||
49570e9b | 955 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
956 | return -EBADF; |
957 | ||
cbb7e577 | 958 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
959 | if (unlikely(ret < 0)) |
960 | return ret; | |
961 | ||
29ce2058 JM |
962 | ret = security_file_permission(out, MAY_WRITE); |
963 | if (unlikely(ret < 0)) | |
964 | return ret; | |
965 | ||
cbb7e577 | 966 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
967 | } |
968 | ||
83f9135b JA |
969 | /* |
970 | * Attempt to initiate a splice from a file to a pipe. | |
971 | */ | |
cbb7e577 JA |
972 | static long do_splice_to(struct file *in, loff_t *ppos, |
973 | struct pipe_inode_info *pipe, size_t len, | |
974 | unsigned int flags) | |
5274f052 | 975 | { |
5274f052 JA |
976 | int ret; |
977 | ||
49570e9b | 978 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
979 | return -EINVAL; |
980 | ||
49570e9b | 981 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
982 | return -EBADF; |
983 | ||
cbb7e577 | 984 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
985 | if (unlikely(ret < 0)) |
986 | return ret; | |
987 | ||
29ce2058 JM |
988 | ret = security_file_permission(in, MAY_READ); |
989 | if (unlikely(ret < 0)) | |
990 | return ret; | |
991 | ||
cbb7e577 | 992 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
993 | } |
994 | ||
932cc6d4 JA |
995 | /** |
996 | * splice_direct_to_actor - splices data directly between two non-pipes | |
997 | * @in: file to splice from | |
998 | * @sd: actor information on where to splice to | |
999 | * @actor: handles the data splicing | |
1000 | * | |
1001 | * Description: | |
1002 | * This is a special case helper to splice directly between two | |
1003 | * points, without requiring an explicit pipe. Internally an allocated | |
1004 | * pipe is cached in the process, and reused during the life time of | |
1005 | * that process. | |
1006 | * | |
c66ab6fa JA |
1007 | */ |
1008 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1009 | splice_direct_actor *actor) | |
b92ce558 JA |
1010 | { |
1011 | struct pipe_inode_info *pipe; | |
1012 | long ret, bytes; | |
1013 | umode_t i_mode; | |
c66ab6fa JA |
1014 | size_t len; |
1015 | int i, flags; | |
b92ce558 JA |
1016 | |
1017 | /* | |
1018 | * We require the input being a regular file, as we don't want to | |
1019 | * randomly drop data for eg socket -> socket splicing. Use the | |
1020 | * piped splicing for that! | |
1021 | */ | |
0f7fc9e4 | 1022 | i_mode = in->f_path.dentry->d_inode->i_mode; |
b92ce558 JA |
1023 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1024 | return -EINVAL; | |
1025 | ||
1026 | /* | |
1027 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1028 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1029 | */ | |
1030 | pipe = current->splice_pipe; | |
49570e9b | 1031 | if (unlikely(!pipe)) { |
b92ce558 JA |
1032 | pipe = alloc_pipe_info(NULL); |
1033 | if (!pipe) | |
1034 | return -ENOMEM; | |
1035 | ||
1036 | /* | |
1037 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1038 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1039 | * PIPE_READERS appropriately. |
1040 | */ | |
1041 | pipe->readers = 1; | |
1042 | ||
1043 | current->splice_pipe = pipe; | |
1044 | } | |
1045 | ||
1046 | /* | |
73d62d83 | 1047 | * Do the splice. |
b92ce558 JA |
1048 | */ |
1049 | ret = 0; | |
1050 | bytes = 0; | |
c66ab6fa JA |
1051 | len = sd->total_len; |
1052 | flags = sd->flags; | |
1053 | ||
1054 | /* | |
1055 | * Don't block on output, we have to drain the direct pipe. | |
1056 | */ | |
1057 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
b92ce558 JA |
1058 | |
1059 | while (len) { | |
51a92c0f | 1060 | size_t read_len; |
bcd4f3ac | 1061 | loff_t pos = sd->pos; |
b92ce558 | 1062 | |
bcd4f3ac | 1063 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1064 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1065 | goto out_release; |
1066 | ||
1067 | read_len = ret; | |
c66ab6fa | 1068 | sd->total_len = read_len; |
b92ce558 JA |
1069 | |
1070 | /* | |
1071 | * NOTE: nonblocking mode only applies to the input. We | |
1072 | * must not do the output in nonblocking mode as then we | |
1073 | * could get stuck data in the internal pipe: | |
1074 | */ | |
c66ab6fa | 1075 | ret = actor(pipe, sd); |
51a92c0f | 1076 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1077 | goto out_release; |
1078 | ||
1079 | bytes += ret; | |
1080 | len -= ret; | |
bcd4f3ac | 1081 | sd->pos = pos; |
b92ce558 | 1082 | |
51a92c0f JA |
1083 | if (ret < read_len) |
1084 | goto out_release; | |
b92ce558 JA |
1085 | } |
1086 | ||
1087 | pipe->nrbufs = pipe->curbuf = 0; | |
b92ce558 JA |
1088 | return bytes; |
1089 | ||
1090 | out_release: | |
1091 | /* | |
1092 | * If we did an incomplete transfer we must release | |
1093 | * the pipe buffers in question: | |
1094 | */ | |
1095 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1096 | struct pipe_buffer *buf = pipe->bufs + i; | |
1097 | ||
1098 | if (buf->ops) { | |
1099 | buf->ops->release(pipe, buf); | |
1100 | buf->ops = NULL; | |
1101 | } | |
1102 | } | |
1103 | pipe->nrbufs = pipe->curbuf = 0; | |
1104 | ||
1105 | /* | |
1106 | * If we transferred some data, return the number of bytes: | |
1107 | */ | |
1108 | if (bytes > 0) | |
1109 | return bytes; | |
1110 | ||
1111 | return ret; | |
c66ab6fa JA |
1112 | |
1113 | } | |
1114 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1115 | ||
1116 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1117 | struct splice_desc *sd) | |
1118 | { | |
6a14b90b | 1119 | struct file *file = sd->u.file; |
c66ab6fa JA |
1120 | |
1121 | return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags); | |
1122 | } | |
1123 | ||
932cc6d4 JA |
1124 | /** |
1125 | * do_splice_direct - splices data directly between two files | |
1126 | * @in: file to splice from | |
1127 | * @ppos: input file offset | |
1128 | * @out: file to splice to | |
1129 | * @len: number of bytes to splice | |
1130 | * @flags: splice modifier flags | |
1131 | * | |
1132 | * Description: | |
1133 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1134 | * doing it in the application would incur an extra system call | |
1135 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1136 | * can splice directly through a process-private pipe. | |
1137 | * | |
1138 | */ | |
c66ab6fa JA |
1139 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
1140 | size_t len, unsigned int flags) | |
1141 | { | |
1142 | struct splice_desc sd = { | |
1143 | .len = len, | |
1144 | .total_len = len, | |
1145 | .flags = flags, | |
1146 | .pos = *ppos, | |
6a14b90b | 1147 | .u.file = out, |
c66ab6fa | 1148 | }; |
51a92c0f | 1149 | long ret; |
c66ab6fa JA |
1150 | |
1151 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); | |
51a92c0f JA |
1152 | if (ret > 0) |
1153 | *ppos += ret; | |
1154 | ||
c66ab6fa | 1155 | return ret; |
b92ce558 JA |
1156 | } |
1157 | ||
ddac0d39 JA |
1158 | /* |
1159 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1160 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1161 | * pipe. | |
1162 | */ | |
1163 | static inline struct pipe_inode_info *pipe_info(struct inode *inode) | |
1164 | { | |
1165 | if (S_ISFIFO(inode->i_mode)) | |
1166 | return inode->i_pipe; | |
1167 | ||
1168 | return NULL; | |
1169 | } | |
1170 | ||
83f9135b JA |
1171 | /* |
1172 | * Determine where to splice to/from. | |
1173 | */ | |
529565dc IM |
1174 | static long do_splice(struct file *in, loff_t __user *off_in, |
1175 | struct file *out, loff_t __user *off_out, | |
1176 | size_t len, unsigned int flags) | |
5274f052 | 1177 | { |
3a326a2c | 1178 | struct pipe_inode_info *pipe; |
cbb7e577 | 1179 | loff_t offset, *off; |
a4514ebd | 1180 | long ret; |
5274f052 | 1181 | |
0f7fc9e4 | 1182 | pipe = pipe_info(in->f_path.dentry->d_inode); |
529565dc IM |
1183 | if (pipe) { |
1184 | if (off_in) | |
1185 | return -ESPIPE; | |
b92ce558 JA |
1186 | if (off_out) { |
1187 | if (out->f_op->llseek == no_llseek) | |
1188 | return -EINVAL; | |
cbb7e577 | 1189 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1190 | return -EFAULT; |
cbb7e577 JA |
1191 | off = &offset; |
1192 | } else | |
1193 | off = &out->f_pos; | |
529565dc | 1194 | |
a4514ebd JA |
1195 | ret = do_splice_from(pipe, out, off, len, flags); |
1196 | ||
1197 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1198 | ret = -EFAULT; | |
1199 | ||
1200 | return ret; | |
529565dc | 1201 | } |
5274f052 | 1202 | |
0f7fc9e4 | 1203 | pipe = pipe_info(out->f_path.dentry->d_inode); |
529565dc IM |
1204 | if (pipe) { |
1205 | if (off_out) | |
1206 | return -ESPIPE; | |
b92ce558 JA |
1207 | if (off_in) { |
1208 | if (in->f_op->llseek == no_llseek) | |
1209 | return -EINVAL; | |
cbb7e577 | 1210 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1211 | return -EFAULT; |
cbb7e577 JA |
1212 | off = &offset; |
1213 | } else | |
1214 | off = &in->f_pos; | |
529565dc | 1215 | |
a4514ebd JA |
1216 | ret = do_splice_to(in, off, pipe, len, flags); |
1217 | ||
1218 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1219 | ret = -EFAULT; | |
1220 | ||
1221 | return ret; | |
529565dc | 1222 | } |
5274f052 JA |
1223 | |
1224 | return -EINVAL; | |
1225 | } | |
1226 | ||
912d35f8 JA |
1227 | /* |
1228 | * Map an iov into an array of pages and offset/length tupples. With the | |
1229 | * partial_page structure, we can map several non-contiguous ranges into | |
1230 | * our ones pages[] map instead of splitting that operation into pieces. | |
1231 | * Could easily be exported as a generic helper for other users, in which | |
1232 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1233 | */ | |
1234 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1235 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1236 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1237 | { |
1238 | int buffers = 0, error = 0; | |
1239 | ||
1240 | /* | |
1241 | * It's ok to take the mmap_sem for reading, even | |
1242 | * across a "get_user()". | |
1243 | */ | |
1244 | down_read(¤t->mm->mmap_sem); | |
1245 | ||
1246 | while (nr_vecs) { | |
1247 | unsigned long off, npages; | |
1248 | void __user *base; | |
1249 | size_t len; | |
1250 | int i; | |
1251 | ||
1252 | /* | |
1253 | * Get user address base and length for this iovec. | |
1254 | */ | |
1255 | error = get_user(base, &iov->iov_base); | |
1256 | if (unlikely(error)) | |
1257 | break; | |
1258 | error = get_user(len, &iov->iov_len); | |
1259 | if (unlikely(error)) | |
1260 | break; | |
1261 | ||
1262 | /* | |
1263 | * Sanity check this iovec. 0 read succeeds. | |
1264 | */ | |
1265 | if (unlikely(!len)) | |
1266 | break; | |
1267 | error = -EFAULT; | |
1268 | if (unlikely(!base)) | |
1269 | break; | |
1270 | ||
1271 | /* | |
1272 | * Get this base offset and number of pages, then map | |
1273 | * in the user pages. | |
1274 | */ | |
1275 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1276 | |
1277 | /* | |
1278 | * If asked for alignment, the offset must be zero and the | |
1279 | * length a multiple of the PAGE_SIZE. | |
1280 | */ | |
1281 | error = -EINVAL; | |
1282 | if (aligned && (off || len & ~PAGE_MASK)) | |
1283 | break; | |
1284 | ||
912d35f8 JA |
1285 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1286 | if (npages > PIPE_BUFFERS - buffers) | |
1287 | npages = PIPE_BUFFERS - buffers; | |
1288 | ||
1289 | error = get_user_pages(current, current->mm, | |
1290 | (unsigned long) base, npages, 0, 0, | |
1291 | &pages[buffers], NULL); | |
1292 | ||
1293 | if (unlikely(error <= 0)) | |
1294 | break; | |
1295 | ||
1296 | /* | |
1297 | * Fill this contiguous range into the partial page map. | |
1298 | */ | |
1299 | for (i = 0; i < error; i++) { | |
7591489a | 1300 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1301 | |
1302 | partial[buffers].offset = off; | |
1303 | partial[buffers].len = plen; | |
1304 | ||
1305 | off = 0; | |
1306 | len -= plen; | |
1307 | buffers++; | |
1308 | } | |
1309 | ||
1310 | /* | |
1311 | * We didn't complete this iov, stop here since it probably | |
1312 | * means we have to move some of this into a pipe to | |
1313 | * be able to continue. | |
1314 | */ | |
1315 | if (len) | |
1316 | break; | |
1317 | ||
1318 | /* | |
1319 | * Don't continue if we mapped fewer pages than we asked for, | |
1320 | * or if we mapped the max number of pages that we have | |
1321 | * room for. | |
1322 | */ | |
1323 | if (error < npages || buffers == PIPE_BUFFERS) | |
1324 | break; | |
1325 | ||
1326 | nr_vecs--; | |
1327 | iov++; | |
1328 | } | |
1329 | ||
1330 | up_read(¤t->mm->mmap_sem); | |
1331 | ||
1332 | if (buffers) | |
1333 | return buffers; | |
1334 | ||
1335 | return error; | |
1336 | } | |
1337 | ||
6a14b90b JA |
1338 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1339 | struct splice_desc *sd) | |
1340 | { | |
1341 | char *src; | |
1342 | int ret; | |
1343 | ||
cac36bb0 | 1344 | ret = buf->ops->confirm(pipe, buf); |
6a14b90b JA |
1345 | if (unlikely(ret)) |
1346 | return ret; | |
1347 | ||
1348 | /* | |
1349 | * See if we can use the atomic maps, by prefaulting in the | |
1350 | * pages and doing an atomic copy | |
1351 | */ | |
1352 | if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { | |
1353 | src = buf->ops->map(pipe, buf, 1); | |
1354 | ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, | |
1355 | sd->len); | |
1356 | buf->ops->unmap(pipe, buf, src); | |
1357 | if (!ret) { | |
1358 | ret = sd->len; | |
1359 | goto out; | |
1360 | } | |
1361 | } | |
1362 | ||
1363 | /* | |
1364 | * No dice, use slow non-atomic map and copy | |
1365 | */ | |
1366 | src = buf->ops->map(pipe, buf, 0); | |
1367 | ||
1368 | ret = sd->len; | |
1369 | if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) | |
1370 | ret = -EFAULT; | |
1371 | ||
1372 | out: | |
1373 | if (ret > 0) | |
1374 | sd->u.userptr += ret; | |
1375 | buf->ops->unmap(pipe, buf, src); | |
1376 | return ret; | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * For lack of a better implementation, implement vmsplice() to userspace | |
1381 | * as a simple copy of the pipes pages to the user iov. | |
1382 | */ | |
1383 | static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, | |
1384 | unsigned long nr_segs, unsigned int flags) | |
1385 | { | |
1386 | struct pipe_inode_info *pipe; | |
1387 | struct splice_desc sd; | |
1388 | ssize_t size; | |
1389 | int error; | |
1390 | long ret; | |
1391 | ||
1392 | pipe = pipe_info(file->f_path.dentry->d_inode); | |
1393 | if (!pipe) | |
1394 | return -EBADF; | |
1395 | ||
1396 | if (pipe->inode) | |
1397 | mutex_lock(&pipe->inode->i_mutex); | |
1398 | ||
1399 | error = ret = 0; | |
1400 | while (nr_segs) { | |
1401 | void __user *base; | |
1402 | size_t len; | |
1403 | ||
1404 | /* | |
1405 | * Get user address base and length for this iovec. | |
1406 | */ | |
1407 | error = get_user(base, &iov->iov_base); | |
1408 | if (unlikely(error)) | |
1409 | break; | |
1410 | error = get_user(len, &iov->iov_len); | |
1411 | if (unlikely(error)) | |
1412 | break; | |
1413 | ||
1414 | /* | |
1415 | * Sanity check this iovec. 0 read succeeds. | |
1416 | */ | |
1417 | if (unlikely(!len)) | |
1418 | break; | |
1419 | if (unlikely(!base)) { | |
1420 | error = -EFAULT; | |
1421 | break; | |
1422 | } | |
1423 | ||
1424 | sd.len = 0; | |
1425 | sd.total_len = len; | |
1426 | sd.flags = flags; | |
1427 | sd.u.userptr = base; | |
1428 | sd.pos = 0; | |
1429 | ||
1430 | size = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1431 | if (size < 0) { | |
1432 | if (!ret) | |
1433 | ret = size; | |
1434 | ||
1435 | break; | |
1436 | } | |
1437 | ||
1438 | ret += size; | |
1439 | ||
1440 | if (size < len) | |
1441 | break; | |
1442 | ||
1443 | nr_segs--; | |
1444 | iov++; | |
1445 | } | |
1446 | ||
1447 | if (pipe->inode) | |
1448 | mutex_unlock(&pipe->inode->i_mutex); | |
1449 | ||
1450 | if (!ret) | |
1451 | ret = error; | |
1452 | ||
1453 | return ret; | |
1454 | } | |
1455 | ||
912d35f8 JA |
1456 | /* |
1457 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1458 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1459 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1460 | */ |
6a14b90b JA |
1461 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1462 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1463 | { |
ddac0d39 | 1464 | struct pipe_inode_info *pipe; |
912d35f8 JA |
1465 | struct page *pages[PIPE_BUFFERS]; |
1466 | struct partial_page partial[PIPE_BUFFERS]; | |
1467 | struct splice_pipe_desc spd = { | |
1468 | .pages = pages, | |
1469 | .partial = partial, | |
1470 | .flags = flags, | |
1471 | .ops = &user_page_pipe_buf_ops, | |
1472 | }; | |
1473 | ||
0f7fc9e4 | 1474 | pipe = pipe_info(file->f_path.dentry->d_inode); |
ddac0d39 | 1475 | if (!pipe) |
912d35f8 | 1476 | return -EBADF; |
912d35f8 | 1477 | |
7afa6fd0 JA |
1478 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1479 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1480 | if (spd.nr_pages <= 0) |
1481 | return spd.nr_pages; | |
1482 | ||
00522fb4 | 1483 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1484 | } |
1485 | ||
6a14b90b JA |
1486 | /* |
1487 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1488 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1489 | * operation that can be supported without any funky alignment restrictions | |
1490 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1491 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1492 | * solutions for that: | |
1493 | * | |
1494 | * - memcpy() the data internally, at which point we might as well just | |
1495 | * do a regular read() on the buffer anyway. | |
1496 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1497 | * has restriction limitations on both ends of the pipe). | |
1498 | * | |
1499 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1500 | * | |
1501 | */ | |
912d35f8 JA |
1502 | asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, |
1503 | unsigned long nr_segs, unsigned int flags) | |
1504 | { | |
1505 | struct file *file; | |
1506 | long error; | |
1507 | int fput; | |
1508 | ||
6a14b90b JA |
1509 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1510 | return -EINVAL; | |
1511 | else if (unlikely(!nr_segs)) | |
1512 | return 0; | |
1513 | ||
912d35f8 JA |
1514 | error = -EBADF; |
1515 | file = fget_light(fd, &fput); | |
1516 | if (file) { | |
1517 | if (file->f_mode & FMODE_WRITE) | |
6a14b90b JA |
1518 | error = vmsplice_to_pipe(file, iov, nr_segs, flags); |
1519 | else if (file->f_mode & FMODE_READ) | |
1520 | error = vmsplice_to_user(file, iov, nr_segs, flags); | |
912d35f8 JA |
1521 | |
1522 | fput_light(file, fput); | |
1523 | } | |
1524 | ||
1525 | return error; | |
1526 | } | |
1527 | ||
529565dc IM |
1528 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
1529 | int fd_out, loff_t __user *off_out, | |
1530 | size_t len, unsigned int flags) | |
5274f052 JA |
1531 | { |
1532 | long error; | |
1533 | struct file *in, *out; | |
1534 | int fput_in, fput_out; | |
1535 | ||
1536 | if (unlikely(!len)) | |
1537 | return 0; | |
1538 | ||
1539 | error = -EBADF; | |
529565dc | 1540 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1541 | if (in) { |
1542 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1543 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1544 | if (out) { |
1545 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1546 | error = do_splice(in, off_in, |
1547 | out, off_out, | |
1548 | len, flags); | |
5274f052 JA |
1549 | fput_light(out, fput_out); |
1550 | } | |
1551 | } | |
1552 | ||
1553 | fput_light(in, fput_in); | |
1554 | } | |
1555 | ||
1556 | return error; | |
1557 | } | |
70524490 | 1558 | |
aadd06e5 JA |
1559 | /* |
1560 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1561 | * return an appropriate error. | |
1562 | */ | |
1563 | static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1564 | { | |
1565 | int ret; | |
1566 | ||
1567 | /* | |
1568 | * Check ->nrbufs without the inode lock first. This function | |
1569 | * is speculative anyways, so missing one is ok. | |
1570 | */ | |
1571 | if (pipe->nrbufs) | |
1572 | return 0; | |
1573 | ||
1574 | ret = 0; | |
1575 | mutex_lock(&pipe->inode->i_mutex); | |
1576 | ||
1577 | while (!pipe->nrbufs) { | |
1578 | if (signal_pending(current)) { | |
1579 | ret = -ERESTARTSYS; | |
1580 | break; | |
1581 | } | |
1582 | if (!pipe->writers) | |
1583 | break; | |
1584 | if (!pipe->waiting_writers) { | |
1585 | if (flags & SPLICE_F_NONBLOCK) { | |
1586 | ret = -EAGAIN; | |
1587 | break; | |
1588 | } | |
1589 | } | |
1590 | pipe_wait(pipe); | |
1591 | } | |
1592 | ||
1593 | mutex_unlock(&pipe->inode->i_mutex); | |
1594 | return ret; | |
1595 | } | |
1596 | ||
1597 | /* | |
1598 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1599 | * return an appropriate error. | |
1600 | */ | |
1601 | static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1602 | { | |
1603 | int ret; | |
1604 | ||
1605 | /* | |
1606 | * Check ->nrbufs without the inode lock first. This function | |
1607 | * is speculative anyways, so missing one is ok. | |
1608 | */ | |
1609 | if (pipe->nrbufs < PIPE_BUFFERS) | |
1610 | return 0; | |
1611 | ||
1612 | ret = 0; | |
1613 | mutex_lock(&pipe->inode->i_mutex); | |
1614 | ||
1615 | while (pipe->nrbufs >= PIPE_BUFFERS) { | |
1616 | if (!pipe->readers) { | |
1617 | send_sig(SIGPIPE, current, 0); | |
1618 | ret = -EPIPE; | |
1619 | break; | |
1620 | } | |
1621 | if (flags & SPLICE_F_NONBLOCK) { | |
1622 | ret = -EAGAIN; | |
1623 | break; | |
1624 | } | |
1625 | if (signal_pending(current)) { | |
1626 | ret = -ERESTARTSYS; | |
1627 | break; | |
1628 | } | |
1629 | pipe->waiting_writers++; | |
1630 | pipe_wait(pipe); | |
1631 | pipe->waiting_writers--; | |
1632 | } | |
1633 | ||
1634 | mutex_unlock(&pipe->inode->i_mutex); | |
1635 | return ret; | |
1636 | } | |
1637 | ||
70524490 JA |
1638 | /* |
1639 | * Link contents of ipipe to opipe. | |
1640 | */ | |
1641 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1642 | struct pipe_inode_info *opipe, | |
1643 | size_t len, unsigned int flags) | |
1644 | { | |
1645 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1646 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1647 | |
1648 | /* | |
1649 | * Potential ABBA deadlock, work around it by ordering lock | |
1650 | * grabbing by inode address. Otherwise two different processes | |
1651 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1652 | */ | |
62752ee1 | 1653 | inode_double_lock(ipipe->inode, opipe->inode); |
70524490 | 1654 | |
aadd06e5 | 1655 | do { |
70524490 JA |
1656 | if (!opipe->readers) { |
1657 | send_sig(SIGPIPE, current, 0); | |
1658 | if (!ret) | |
1659 | ret = -EPIPE; | |
1660 | break; | |
1661 | } | |
70524490 | 1662 | |
aadd06e5 JA |
1663 | /* |
1664 | * If we have iterated all input buffers or ran out of | |
1665 | * output room, break. | |
1666 | */ | |
1667 | if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) | |
1668 | break; | |
70524490 | 1669 | |
aadd06e5 JA |
1670 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); |
1671 | nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
70524490 JA |
1672 | |
1673 | /* | |
aadd06e5 JA |
1674 | * Get a reference to this pipe buffer, |
1675 | * so we can copy the contents over. | |
70524490 | 1676 | */ |
aadd06e5 JA |
1677 | ibuf->ops->get(ipipe, ibuf); |
1678 | ||
1679 | obuf = opipe->bufs + nbuf; | |
1680 | *obuf = *ibuf; | |
1681 | ||
2a27250e | 1682 | /* |
aadd06e5 JA |
1683 | * Don't inherit the gift flag, we need to |
1684 | * prevent multiple steals of this page. | |
2a27250e | 1685 | */ |
aadd06e5 | 1686 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1687 | |
aadd06e5 JA |
1688 | if (obuf->len > len) |
1689 | obuf->len = len; | |
70524490 | 1690 | |
aadd06e5 JA |
1691 | opipe->nrbufs++; |
1692 | ret += obuf->len; | |
1693 | len -= obuf->len; | |
1694 | i++; | |
1695 | } while (len); | |
70524490 | 1696 | |
62752ee1 | 1697 | inode_double_unlock(ipipe->inode, opipe->inode); |
70524490 | 1698 | |
aadd06e5 JA |
1699 | /* |
1700 | * If we put data in the output pipe, wakeup any potential readers. | |
1701 | */ | |
1702 | if (ret > 0) { | |
70524490 JA |
1703 | smp_mb(); |
1704 | if (waitqueue_active(&opipe->wait)) | |
1705 | wake_up_interruptible(&opipe->wait); | |
1706 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1707 | } | |
1708 | ||
1709 | return ret; | |
1710 | } | |
1711 | ||
1712 | /* | |
1713 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1714 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1715 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1716 | * applicable one is SPLICE_F_NONBLOCK. | |
1717 | */ | |
1718 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1719 | unsigned int flags) | |
1720 | { | |
0f7fc9e4 JJS |
1721 | struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); |
1722 | struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); | |
aadd06e5 | 1723 | int ret = -EINVAL; |
70524490 JA |
1724 | |
1725 | /* | |
aadd06e5 JA |
1726 | * Duplicate the contents of ipipe to opipe without actually |
1727 | * copying the data. | |
70524490 | 1728 | */ |
aadd06e5 JA |
1729 | if (ipipe && opipe && ipipe != opipe) { |
1730 | /* | |
1731 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1732 | * ordering doesn't really matter. | |
1733 | */ | |
1734 | ret = link_ipipe_prep(ipipe, flags); | |
1735 | if (!ret) { | |
1736 | ret = link_opipe_prep(opipe, flags); | |
1737 | if (!ret) { | |
1738 | ret = link_pipe(ipipe, opipe, len, flags); | |
1739 | if (!ret && (flags & SPLICE_F_NONBLOCK)) | |
1740 | ret = -EAGAIN; | |
1741 | } | |
1742 | } | |
1743 | } | |
70524490 | 1744 | |
aadd06e5 | 1745 | return ret; |
70524490 JA |
1746 | } |
1747 | ||
1748 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1749 | { | |
1750 | struct file *in; | |
1751 | int error, fput_in; | |
1752 | ||
1753 | if (unlikely(!len)) | |
1754 | return 0; | |
1755 | ||
1756 | error = -EBADF; | |
1757 | in = fget_light(fdin, &fput_in); | |
1758 | if (in) { | |
1759 | if (in->f_mode & FMODE_READ) { | |
1760 | int fput_out; | |
1761 | struct file *out = fget_light(fdout, &fput_out); | |
1762 | ||
1763 | if (out) { | |
1764 | if (out->f_mode & FMODE_WRITE) | |
1765 | error = do_tee(in, out, len, flags); | |
1766 | fput_light(out, fput_out); | |
1767 | } | |
1768 | } | |
1769 | fput_light(in, fput_in); | |
1770 | } | |
1771 | ||
1772 | return error; | |
1773 | } |