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; | |
d8983910 | 268 | unsigned int loff, nr_pages, req_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 | 283 | loff = *ppos & ~PAGE_CACHE_MASK; |
d8983910 FW |
284 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
285 | nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS); | |
5274f052 | 286 | |
eb20796b JA |
287 | /* |
288 | * Lookup the (hopefully) full range of pages we need. | |
289 | */ | |
290 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
431a4820 | 291 | index += spd.nr_pages; |
82aa5d61 | 292 | |
eb20796b JA |
293 | /* |
294 | * If find_get_pages_contig() returned fewer pages than we needed, | |
431a4820 | 295 | * readahead/allocate the rest and fill in the holes. |
eb20796b | 296 | */ |
431a4820 | 297 | if (spd.nr_pages < nr_pages) |
cf914a7d RR |
298 | page_cache_sync_readahead(mapping, &in->f_ra, in, |
299 | index, req_pages - spd.nr_pages); | |
431a4820 | 300 | |
932cc6d4 | 301 | error = 0; |
eb20796b | 302 | while (spd.nr_pages < nr_pages) { |
82aa5d61 | 303 | /* |
eb20796b JA |
304 | * Page could be there, find_get_pages_contig() breaks on |
305 | * the first hole. | |
5274f052 | 306 | */ |
7480a904 JA |
307 | page = find_get_page(mapping, index); |
308 | if (!page) { | |
7480a904 | 309 | /* |
eb20796b | 310 | * page didn't exist, allocate one. |
7480a904 JA |
311 | */ |
312 | page = page_cache_alloc_cold(mapping); | |
313 | if (!page) | |
314 | break; | |
315 | ||
316 | error = add_to_page_cache_lru(page, mapping, index, | |
2ae88149 | 317 | GFP_KERNEL); |
7480a904 JA |
318 | if (unlikely(error)) { |
319 | page_cache_release(page); | |
a0548871 JA |
320 | if (error == -EEXIST) |
321 | continue; | |
7480a904 JA |
322 | break; |
323 | } | |
eb20796b JA |
324 | /* |
325 | * add_to_page_cache() locks the page, unlock it | |
326 | * to avoid convoluting the logic below even more. | |
327 | */ | |
328 | unlock_page(page); | |
7480a904 JA |
329 | } |
330 | ||
eb20796b JA |
331 | pages[spd.nr_pages++] = page; |
332 | index++; | |
333 | } | |
334 | ||
335 | /* | |
336 | * Now loop over the map and see if we need to start IO on any | |
337 | * pages, fill in the partial map, etc. | |
338 | */ | |
339 | index = *ppos >> PAGE_CACHE_SHIFT; | |
340 | nr_pages = spd.nr_pages; | |
341 | spd.nr_pages = 0; | |
342 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
343 | unsigned int this_len; | |
344 | ||
345 | if (!len) | |
346 | break; | |
347 | ||
348 | /* | |
349 | * this_len is the max we'll use from this page | |
350 | */ | |
351 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
352 | page = pages[page_nr]; | |
353 | ||
a08a166f | 354 | if (PageReadahead(page)) |
cf914a7d | 355 | page_cache_async_readahead(mapping, &in->f_ra, in, |
d8983910 | 356 | page, index, req_pages - page_nr); |
a08a166f | 357 | |
7480a904 JA |
358 | /* |
359 | * If the page isn't uptodate, we may need to start io on it | |
360 | */ | |
361 | if (!PageUptodate(page)) { | |
c4f895cb JA |
362 | /* |
363 | * If in nonblock mode then dont block on waiting | |
364 | * for an in-flight io page | |
365 | */ | |
9ae9d68c FW |
366 | if (flags & SPLICE_F_NONBLOCK) { |
367 | if (TestSetPageLocked(page)) | |
368 | break; | |
369 | } else | |
370 | lock_page(page); | |
7480a904 JA |
371 | |
372 | /* | |
373 | * page was truncated, stop here. if this isn't the | |
374 | * first page, we'll just complete what we already | |
375 | * added | |
376 | */ | |
377 | if (!page->mapping) { | |
378 | unlock_page(page); | |
7480a904 JA |
379 | break; |
380 | } | |
381 | /* | |
382 | * page was already under io and is now done, great | |
383 | */ | |
384 | if (PageUptodate(page)) { | |
385 | unlock_page(page); | |
386 | goto fill_it; | |
387 | } | |
5274f052 | 388 | |
7480a904 JA |
389 | /* |
390 | * need to read in the page | |
391 | */ | |
392 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 393 | if (unlikely(error)) { |
eb20796b JA |
394 | /* |
395 | * We really should re-lookup the page here, | |
396 | * but it complicates things a lot. Instead | |
397 | * lets just do what we already stored, and | |
398 | * we'll get it the next time we are called. | |
399 | */ | |
7480a904 | 400 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
401 | error = 0; |
402 | ||
5274f052 JA |
403 | break; |
404 | } | |
620a324b JA |
405 | } |
406 | fill_it: | |
407 | /* | |
408 | * i_size must be checked after PageUptodate. | |
409 | */ | |
410 | isize = i_size_read(mapping->host); | |
411 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
412 | if (unlikely(!isize || index > end_index)) | |
413 | break; | |
414 | ||
415 | /* | |
416 | * if this is the last page, see if we need to shrink | |
417 | * the length and stop | |
418 | */ | |
419 | if (end_index == index) { | |
420 | unsigned int plen; | |
91ad66ef JA |
421 | |
422 | /* | |
620a324b | 423 | * max good bytes in this page |
91ad66ef | 424 | */ |
620a324b JA |
425 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
426 | if (plen <= loff) | |
91ad66ef | 427 | break; |
91ad66ef JA |
428 | |
429 | /* | |
620a324b | 430 | * force quit after adding this page |
91ad66ef | 431 | */ |
620a324b JA |
432 | this_len = min(this_len, plen - loff); |
433 | len = this_len; | |
5274f052 | 434 | } |
620a324b | 435 | |
eb20796b JA |
436 | partial[page_nr].offset = loff; |
437 | partial[page_nr].len = this_len; | |
82aa5d61 | 438 | len -= this_len; |
91ad66ef | 439 | loff = 0; |
eb20796b JA |
440 | spd.nr_pages++; |
441 | index++; | |
5274f052 JA |
442 | } |
443 | ||
eb20796b | 444 | /* |
475ecade | 445 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
446 | * we got, 'nr_pages' is how many pages are in the map. |
447 | */ | |
448 | while (page_nr < nr_pages) | |
449 | page_cache_release(pages[page_nr++]); | |
a08a166f | 450 | in->f_ra.prev_index = index; |
eb20796b | 451 | |
912d35f8 | 452 | if (spd.nr_pages) |
00522fb4 | 453 | return splice_to_pipe(pipe, &spd); |
5274f052 | 454 | |
7480a904 | 455 | return error; |
5274f052 JA |
456 | } |
457 | ||
83f9135b JA |
458 | /** |
459 | * generic_file_splice_read - splice data from file to a pipe | |
460 | * @in: file to splice from | |
932cc6d4 | 461 | * @ppos: position in @in |
83f9135b JA |
462 | * @pipe: pipe to splice to |
463 | * @len: number of bytes to splice | |
464 | * @flags: splice modifier flags | |
465 | * | |
932cc6d4 JA |
466 | * Description: |
467 | * Will read pages from given file and fill them into a pipe. Can be | |
468 | * used as long as the address_space operations for the source implements | |
469 | * a readpage() hook. | |
470 | * | |
83f9135b | 471 | */ |
cbb7e577 JA |
472 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
473 | struct pipe_inode_info *pipe, size_t len, | |
474 | unsigned int flags) | |
5274f052 JA |
475 | { |
476 | ssize_t spliced; | |
477 | int ret; | |
d366d398 JA |
478 | loff_t isize, left; |
479 | ||
480 | isize = i_size_read(in->f_mapping->host); | |
481 | if (unlikely(*ppos >= isize)) | |
482 | return 0; | |
483 | ||
484 | left = isize - *ppos; | |
485 | if (unlikely(left < len)) | |
486 | len = left; | |
5274f052 JA |
487 | |
488 | ret = 0; | |
489 | spliced = 0; | |
51a92c0f | 490 | while (len && !spliced) { |
cbb7e577 | 491 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 492 | |
c4f895cb | 493 | if (ret < 0) |
5274f052 | 494 | break; |
c4f895cb JA |
495 | else if (!ret) { |
496 | if (spliced) | |
497 | break; | |
498 | if (flags & SPLICE_F_NONBLOCK) { | |
499 | ret = -EAGAIN; | |
500 | break; | |
501 | } | |
502 | } | |
5274f052 | 503 | |
cbb7e577 | 504 | *ppos += ret; |
5274f052 JA |
505 | len -= ret; |
506 | spliced += ret; | |
507 | } | |
508 | ||
509 | if (spliced) | |
510 | return spliced; | |
511 | ||
512 | return ret; | |
513 | } | |
514 | ||
059a8f37 JA |
515 | EXPORT_SYMBOL(generic_file_splice_read); |
516 | ||
5274f052 | 517 | /* |
4f6f0bd2 | 518 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 519 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 520 | */ |
76ad4d11 | 521 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
522 | struct pipe_buffer *buf, struct splice_desc *sd) |
523 | { | |
6a14b90b | 524 | struct file *file = sd->u.file; |
5274f052 | 525 | loff_t pos = sd->pos; |
f84d7519 | 526 | int ret, more; |
5274f052 | 527 | |
cac36bb0 | 528 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
529 | if (!ret) { |
530 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 531 | |
f84d7519 JA |
532 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
533 | sd->len, &pos, more); | |
534 | } | |
5274f052 | 535 | |
016b661e | 536 | return ret; |
5274f052 JA |
537 | } |
538 | ||
539 | /* | |
540 | * This is a little more tricky than the file -> pipe splicing. There are | |
541 | * basically three cases: | |
542 | * | |
543 | * - Destination page already exists in the address space and there | |
544 | * are users of it. For that case we have no other option that | |
545 | * copying the data. Tough luck. | |
546 | * - Destination page already exists in the address space, but there | |
547 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
548 | * through to last case. | |
549 | * - Destination page does not exist, we can add the pipe page to | |
550 | * the page cache and avoid the copy. | |
551 | * | |
83f9135b JA |
552 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
553 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
554 | * file address space page cache. This is possible if no one else has | |
555 | * the pipe page referenced outside of the pipe and page cache. If | |
556 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
557 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 558 | */ |
76ad4d11 | 559 | static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
5274f052 JA |
560 | struct splice_desc *sd) |
561 | { | |
6a14b90b | 562 | struct file *file = sd->u.file; |
5274f052 | 563 | struct address_space *mapping = file->f_mapping; |
016b661e | 564 | unsigned int offset, this_len; |
5274f052 | 565 | struct page *page; |
5274f052 | 566 | pgoff_t index; |
3e7ee3e7 | 567 | int ret; |
5274f052 JA |
568 | |
569 | /* | |
49d0b21b | 570 | * make sure the data in this buffer is uptodate |
5274f052 | 571 | */ |
cac36bb0 | 572 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
573 | if (unlikely(ret)) |
574 | return ret; | |
5274f052 JA |
575 | |
576 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
577 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
578 | ||
016b661e JA |
579 | this_len = sd->len; |
580 | if (this_len + offset > PAGE_CACHE_SIZE) | |
581 | this_len = PAGE_CACHE_SIZE - offset; | |
582 | ||
485ddb4b NP |
583 | find_page: |
584 | page = find_lock_page(mapping, index); | |
585 | if (!page) { | |
586 | ret = -ENOMEM; | |
587 | page = page_cache_alloc_cold(mapping); | |
588 | if (unlikely(!page)) | |
589 | goto out_ret; | |
590 | ||
83f9135b | 591 | /* |
485ddb4b | 592 | * This will also lock the page |
83f9135b | 593 | */ |
485ddb4b NP |
594 | ret = add_to_page_cache_lru(page, mapping, index, |
595 | GFP_KERNEL); | |
596 | if (unlikely(ret)) | |
597 | goto out; | |
598 | } | |
9e0267c2 | 599 | |
016b661e | 600 | ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); |
bfc4ee39 JA |
601 | if (unlikely(ret)) { |
602 | loff_t isize = i_size_read(mapping->host); | |
603 | ||
604 | if (ret != AOP_TRUNCATED_PAGE) | |
605 | unlock_page(page); | |
4f6f0bd2 | 606 | page_cache_release(page); |
bfc4ee39 JA |
607 | if (ret == AOP_TRUNCATED_PAGE) |
608 | goto find_page; | |
609 | ||
610 | /* | |
611 | * prepare_write() may have instantiated a few blocks | |
612 | * outside i_size. Trim these off again. | |
613 | */ | |
614 | if (sd->pos + this_len > isize) | |
615 | vmtruncate(mapping->host, isize); | |
616 | ||
e6e80f29 | 617 | goto out_ret; |
bfc4ee39 | 618 | } |
5274f052 | 619 | |
0568b409 | 620 | if (buf->page != page) { |
f84d7519 JA |
621 | /* |
622 | * Careful, ->map() uses KM_USER0! | |
623 | */ | |
76ad4d11 | 624 | char *src = buf->ops->map(pipe, buf, 1); |
f84d7519 | 625 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 626 | |
016b661e | 627 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 628 | flush_dcache_page(page); |
f84d7519 | 629 | kunmap_atomic(dst, KM_USER1); |
76ad4d11 | 630 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 631 | } |
5274f052 | 632 | |
016b661e | 633 | ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); |
d9993c37 DM |
634 | if (ret) { |
635 | if (ret == AOP_TRUNCATED_PAGE) { | |
636 | page_cache_release(page); | |
637 | goto find_page; | |
638 | } | |
639 | if (ret < 0) | |
640 | goto out; | |
0568b409 | 641 | /* |
d9993c37 DM |
642 | * Partial write has happened, so 'ret' already initialized by |
643 | * number of bytes written, Where is nothing we have to do here. | |
0568b409 | 644 | */ |
d9993c37 | 645 | } else |
0568b409 | 646 | ret = this_len; |
d9993c37 DM |
647 | /* |
648 | * Return the number of bytes written and mark page as | |
649 | * accessed, we are now done! | |
650 | */ | |
651 | mark_page_accessed(page); | |
5274f052 | 652 | out: |
0568b409 | 653 | page_cache_release(page); |
9e0267c2 | 654 | unlock_page(page); |
e6e80f29 | 655 | out_ret: |
5274f052 JA |
656 | return ret; |
657 | } | |
658 | ||
932cc6d4 JA |
659 | /** |
660 | * __splice_from_pipe - splice data from a pipe to given actor | |
661 | * @pipe: pipe to splice from | |
662 | * @sd: information to @actor | |
663 | * @actor: handler that splices the data | |
664 | * | |
665 | * Description: | |
666 | * This function does little more than loop over the pipe and call | |
667 | * @actor to do the actual moving of a single struct pipe_buffer to | |
668 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
669 | * pipe_to_user. | |
670 | * | |
83f9135b | 671 | */ |
c66ab6fa JA |
672 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, |
673 | splice_actor *actor) | |
5274f052 | 674 | { |
5274f052 | 675 | int ret, do_wakeup, err; |
5274f052 JA |
676 | |
677 | ret = 0; | |
678 | do_wakeup = 0; | |
679 | ||
5274f052 | 680 | for (;;) { |
6f767b04 JA |
681 | if (pipe->nrbufs) { |
682 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
d4c3cca9 | 683 | const struct pipe_buf_operations *ops = buf->ops; |
5274f052 | 684 | |
c66ab6fa JA |
685 | sd->len = buf->len; |
686 | if (sd->len > sd->total_len) | |
687 | sd->len = sd->total_len; | |
5274f052 | 688 | |
c66ab6fa | 689 | err = actor(pipe, buf, sd); |
016b661e | 690 | if (err <= 0) { |
5274f052 JA |
691 | if (!ret && err != -ENODATA) |
692 | ret = err; | |
693 | ||
694 | break; | |
695 | } | |
696 | ||
016b661e JA |
697 | ret += err; |
698 | buf->offset += err; | |
699 | buf->len -= err; | |
700 | ||
c66ab6fa JA |
701 | sd->len -= err; |
702 | sd->pos += err; | |
703 | sd->total_len -= err; | |
704 | if (sd->len) | |
016b661e | 705 | continue; |
73d62d83 | 706 | |
5274f052 JA |
707 | if (!buf->len) { |
708 | buf->ops = NULL; | |
3a326a2c | 709 | ops->release(pipe, buf); |
6f767b04 JA |
710 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
711 | pipe->nrbufs--; | |
712 | if (pipe->inode) | |
713 | do_wakeup = 1; | |
5274f052 JA |
714 | } |
715 | ||
c66ab6fa | 716 | if (!sd->total_len) |
5274f052 JA |
717 | break; |
718 | } | |
719 | ||
6f767b04 | 720 | if (pipe->nrbufs) |
5274f052 | 721 | continue; |
3a326a2c | 722 | if (!pipe->writers) |
5274f052 | 723 | break; |
3a326a2c | 724 | if (!pipe->waiting_writers) { |
5274f052 JA |
725 | if (ret) |
726 | break; | |
727 | } | |
728 | ||
c66ab6fa | 729 | if (sd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
730 | if (!ret) |
731 | ret = -EAGAIN; | |
732 | break; | |
733 | } | |
734 | ||
5274f052 JA |
735 | if (signal_pending(current)) { |
736 | if (!ret) | |
737 | ret = -ERESTARTSYS; | |
738 | break; | |
739 | } | |
740 | ||
741 | if (do_wakeup) { | |
c0bd1f65 | 742 | smp_mb(); |
3a326a2c IM |
743 | if (waitqueue_active(&pipe->wait)) |
744 | wake_up_interruptible_sync(&pipe->wait); | |
745 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
746 | do_wakeup = 0; |
747 | } | |
748 | ||
3a326a2c | 749 | pipe_wait(pipe); |
5274f052 JA |
750 | } |
751 | ||
5274f052 | 752 | if (do_wakeup) { |
c0bd1f65 | 753 | smp_mb(); |
3a326a2c IM |
754 | if (waitqueue_active(&pipe->wait)) |
755 | wake_up_interruptible(&pipe->wait); | |
756 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
757 | } |
758 | ||
5274f052 | 759 | return ret; |
5274f052 | 760 | } |
40bee44e | 761 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 762 | |
932cc6d4 JA |
763 | /** |
764 | * splice_from_pipe - splice data from a pipe to a file | |
765 | * @pipe: pipe to splice from | |
766 | * @out: file to splice to | |
767 | * @ppos: position in @out | |
768 | * @len: how many bytes to splice | |
769 | * @flags: splice modifier flags | |
770 | * @actor: handler that splices the data | |
771 | * | |
772 | * Description: | |
773 | * See __splice_from_pipe. This function locks the input and output inodes, | |
774 | * otherwise it's identical to __splice_from_pipe(). | |
775 | * | |
776 | */ | |
6da61809 MF |
777 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
778 | loff_t *ppos, size_t len, unsigned int flags, | |
779 | splice_actor *actor) | |
780 | { | |
781 | ssize_t ret; | |
782 | struct inode *inode = out->f_mapping->host; | |
c66ab6fa JA |
783 | struct splice_desc sd = { |
784 | .total_len = len, | |
785 | .flags = flags, | |
786 | .pos = *ppos, | |
6a14b90b | 787 | .u.file = out, |
c66ab6fa | 788 | }; |
6da61809 MF |
789 | |
790 | /* | |
791 | * The actor worker might be calling ->prepare_write and | |
792 | * ->commit_write. Most of the time, these expect i_mutex to | |
793 | * be held. Since this may result in an ABBA deadlock with | |
794 | * pipe->inode, we have to order lock acquiry here. | |
795 | */ | |
796 | inode_double_lock(inode, pipe->inode); | |
c66ab6fa | 797 | ret = __splice_from_pipe(pipe, &sd, actor); |
6da61809 MF |
798 | inode_double_unlock(inode, pipe->inode); |
799 | ||
800 | return ret; | |
801 | } | |
802 | ||
803 | /** | |
804 | * generic_file_splice_write_nolock - generic_file_splice_write without mutexes | |
805 | * @pipe: pipe info | |
806 | * @out: file to write to | |
932cc6d4 | 807 | * @ppos: position in @out |
6da61809 MF |
808 | * @len: number of bytes to splice |
809 | * @flags: splice modifier flags | |
810 | * | |
932cc6d4 JA |
811 | * Description: |
812 | * Will either move or copy pages (determined by @flags options) from | |
813 | * the given pipe inode to the given file. The caller is responsible | |
814 | * for acquiring i_mutex on both inodes. | |
6da61809 MF |
815 | * |
816 | */ | |
817 | ssize_t | |
818 | generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out, | |
819 | loff_t *ppos, size_t len, unsigned int flags) | |
820 | { | |
821 | struct address_space *mapping = out->f_mapping; | |
822 | struct inode *inode = mapping->host; | |
c66ab6fa JA |
823 | struct splice_desc sd = { |
824 | .total_len = len, | |
825 | .flags = flags, | |
826 | .pos = *ppos, | |
6a14b90b | 827 | .u.file = out, |
c66ab6fa | 828 | }; |
6da61809 MF |
829 | ssize_t ret; |
830 | int err; | |
831 | ||
0f7fc9e4 | 832 | err = remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
833 | if (unlikely(err)) |
834 | return err; | |
835 | ||
c66ab6fa | 836 | ret = __splice_from_pipe(pipe, &sd, pipe_to_file); |
6da61809 | 837 | if (ret > 0) { |
17ee4f49 JA |
838 | unsigned long nr_pages; |
839 | ||
6da61809 | 840 | *ppos += ret; |
17ee4f49 | 841 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
6da61809 MF |
842 | |
843 | /* | |
844 | * If file or inode is SYNC and we actually wrote some data, | |
845 | * sync it. | |
846 | */ | |
847 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
848 | err = generic_osync_inode(inode, mapping, | |
849 | OSYNC_METADATA|OSYNC_DATA); | |
850 | ||
851 | if (err) | |
852 | ret = err; | |
853 | } | |
17ee4f49 | 854 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
6da61809 MF |
855 | } |
856 | ||
857 | return ret; | |
858 | } | |
859 | ||
860 | EXPORT_SYMBOL(generic_file_splice_write_nolock); | |
861 | ||
83f9135b JA |
862 | /** |
863 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 864 | * @pipe: pipe info |
83f9135b | 865 | * @out: file to write to |
932cc6d4 | 866 | * @ppos: position in @out |
83f9135b JA |
867 | * @len: number of bytes to splice |
868 | * @flags: splice modifier flags | |
869 | * | |
932cc6d4 JA |
870 | * Description: |
871 | * Will either move or copy pages (determined by @flags options) from | |
872 | * the given pipe inode to the given file. | |
83f9135b JA |
873 | * |
874 | */ | |
3a326a2c IM |
875 | ssize_t |
876 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 877 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 878 | { |
4f6f0bd2 | 879 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 880 | struct inode *inode = mapping->host; |
3a326a2c | 881 | ssize_t ret; |
8c34e2d6 JA |
882 | int err; |
883 | ||
0f7fc9e4 | 884 | err = should_remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
885 | if (unlikely(err)) { |
886 | mutex_lock(&inode->i_mutex); | |
0f7fc9e4 | 887 | err = __remove_suid(out->f_path.dentry, err); |
8c34e2d6 JA |
888 | mutex_unlock(&inode->i_mutex); |
889 | if (err) | |
890 | return err; | |
891 | } | |
3a326a2c | 892 | |
00522fb4 | 893 | ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
a4514ebd | 894 | if (ret > 0) { |
17ee4f49 JA |
895 | unsigned long nr_pages; |
896 | ||
a4514ebd | 897 | *ppos += ret; |
17ee4f49 | 898 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
a4514ebd JA |
899 | |
900 | /* | |
901 | * If file or inode is SYNC and we actually wrote some data, | |
902 | * sync it. | |
903 | */ | |
904 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
a4514ebd JA |
905 | mutex_lock(&inode->i_mutex); |
906 | err = generic_osync_inode(inode, mapping, | |
907 | OSYNC_METADATA|OSYNC_DATA); | |
908 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 909 | |
a4514ebd JA |
910 | if (err) |
911 | ret = err; | |
912 | } | |
17ee4f49 | 913 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
4f6f0bd2 JA |
914 | } |
915 | ||
916 | return ret; | |
5274f052 JA |
917 | } |
918 | ||
059a8f37 JA |
919 | EXPORT_SYMBOL(generic_file_splice_write); |
920 | ||
83f9135b JA |
921 | /** |
922 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 923 | * @pipe: pipe to splice from |
83f9135b | 924 | * @out: socket to write to |
932cc6d4 | 925 | * @ppos: position in @out |
83f9135b JA |
926 | * @len: number of bytes to splice |
927 | * @flags: splice modifier flags | |
928 | * | |
932cc6d4 JA |
929 | * Description: |
930 | * Will send @len bytes from the pipe to a network socket. No data copying | |
931 | * is involved. | |
83f9135b JA |
932 | * |
933 | */ | |
3a326a2c | 934 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 935 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 936 | { |
00522fb4 | 937 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
938 | } |
939 | ||
059a8f37 | 940 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 941 | |
83f9135b JA |
942 | /* |
943 | * Attempt to initiate a splice from pipe to file. | |
944 | */ | |
3a326a2c | 945 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 946 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 947 | { |
5274f052 JA |
948 | int ret; |
949 | ||
49570e9b | 950 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
951 | return -EINVAL; |
952 | ||
49570e9b | 953 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
954 | return -EBADF; |
955 | ||
cbb7e577 | 956 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
957 | if (unlikely(ret < 0)) |
958 | return ret; | |
959 | ||
29ce2058 JM |
960 | ret = security_file_permission(out, MAY_WRITE); |
961 | if (unlikely(ret < 0)) | |
962 | return ret; | |
963 | ||
cbb7e577 | 964 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
965 | } |
966 | ||
83f9135b JA |
967 | /* |
968 | * Attempt to initiate a splice from a file to a pipe. | |
969 | */ | |
cbb7e577 JA |
970 | static long do_splice_to(struct file *in, loff_t *ppos, |
971 | struct pipe_inode_info *pipe, size_t len, | |
972 | unsigned int flags) | |
5274f052 | 973 | { |
5274f052 JA |
974 | int ret; |
975 | ||
49570e9b | 976 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
977 | return -EINVAL; |
978 | ||
49570e9b | 979 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
980 | return -EBADF; |
981 | ||
cbb7e577 | 982 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
983 | if (unlikely(ret < 0)) |
984 | return ret; | |
985 | ||
29ce2058 JM |
986 | ret = security_file_permission(in, MAY_READ); |
987 | if (unlikely(ret < 0)) | |
988 | return ret; | |
989 | ||
cbb7e577 | 990 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
991 | } |
992 | ||
932cc6d4 JA |
993 | /** |
994 | * splice_direct_to_actor - splices data directly between two non-pipes | |
995 | * @in: file to splice from | |
996 | * @sd: actor information on where to splice to | |
997 | * @actor: handles the data splicing | |
998 | * | |
999 | * Description: | |
1000 | * This is a special case helper to splice directly between two | |
1001 | * points, without requiring an explicit pipe. Internally an allocated | |
1002 | * pipe is cached in the process, and reused during the life time of | |
1003 | * that process. | |
1004 | * | |
c66ab6fa JA |
1005 | */ |
1006 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1007 | splice_direct_actor *actor) | |
b92ce558 JA |
1008 | { |
1009 | struct pipe_inode_info *pipe; | |
1010 | long ret, bytes; | |
1011 | umode_t i_mode; | |
c66ab6fa JA |
1012 | size_t len; |
1013 | int i, flags; | |
b92ce558 JA |
1014 | |
1015 | /* | |
1016 | * We require the input being a regular file, as we don't want to | |
1017 | * randomly drop data for eg socket -> socket splicing. Use the | |
1018 | * piped splicing for that! | |
1019 | */ | |
0f7fc9e4 | 1020 | i_mode = in->f_path.dentry->d_inode->i_mode; |
b92ce558 JA |
1021 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1022 | return -EINVAL; | |
1023 | ||
1024 | /* | |
1025 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1026 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1027 | */ | |
1028 | pipe = current->splice_pipe; | |
49570e9b | 1029 | if (unlikely(!pipe)) { |
b92ce558 JA |
1030 | pipe = alloc_pipe_info(NULL); |
1031 | if (!pipe) | |
1032 | return -ENOMEM; | |
1033 | ||
1034 | /* | |
1035 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1036 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1037 | * PIPE_READERS appropriately. |
1038 | */ | |
1039 | pipe->readers = 1; | |
1040 | ||
1041 | current->splice_pipe = pipe; | |
1042 | } | |
1043 | ||
1044 | /* | |
73d62d83 | 1045 | * Do the splice. |
b92ce558 JA |
1046 | */ |
1047 | ret = 0; | |
1048 | bytes = 0; | |
c66ab6fa JA |
1049 | len = sd->total_len; |
1050 | flags = sd->flags; | |
1051 | ||
1052 | /* | |
1053 | * Don't block on output, we have to drain the direct pipe. | |
1054 | */ | |
1055 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
b92ce558 JA |
1056 | |
1057 | while (len) { | |
51a92c0f | 1058 | size_t read_len; |
bcd4f3ac | 1059 | loff_t pos = sd->pos; |
b92ce558 | 1060 | |
bcd4f3ac | 1061 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1062 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1063 | goto out_release; |
1064 | ||
1065 | read_len = ret; | |
c66ab6fa | 1066 | sd->total_len = read_len; |
b92ce558 JA |
1067 | |
1068 | /* | |
1069 | * NOTE: nonblocking mode only applies to the input. We | |
1070 | * must not do the output in nonblocking mode as then we | |
1071 | * could get stuck data in the internal pipe: | |
1072 | */ | |
c66ab6fa | 1073 | ret = actor(pipe, sd); |
51a92c0f | 1074 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1075 | goto out_release; |
1076 | ||
1077 | bytes += ret; | |
1078 | len -= ret; | |
bcd4f3ac | 1079 | sd->pos = pos; |
b92ce558 | 1080 | |
51a92c0f JA |
1081 | if (ret < read_len) |
1082 | goto out_release; | |
b92ce558 JA |
1083 | } |
1084 | ||
1085 | pipe->nrbufs = pipe->curbuf = 0; | |
b92ce558 JA |
1086 | return bytes; |
1087 | ||
1088 | out_release: | |
1089 | /* | |
1090 | * If we did an incomplete transfer we must release | |
1091 | * the pipe buffers in question: | |
1092 | */ | |
1093 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1094 | struct pipe_buffer *buf = pipe->bufs + i; | |
1095 | ||
1096 | if (buf->ops) { | |
1097 | buf->ops->release(pipe, buf); | |
1098 | buf->ops = NULL; | |
1099 | } | |
1100 | } | |
1101 | pipe->nrbufs = pipe->curbuf = 0; | |
1102 | ||
1103 | /* | |
1104 | * If we transferred some data, return the number of bytes: | |
1105 | */ | |
1106 | if (bytes > 0) | |
1107 | return bytes; | |
1108 | ||
1109 | return ret; | |
c66ab6fa JA |
1110 | |
1111 | } | |
1112 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1113 | ||
1114 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1115 | struct splice_desc *sd) | |
1116 | { | |
6a14b90b | 1117 | struct file *file = sd->u.file; |
c66ab6fa JA |
1118 | |
1119 | return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags); | |
1120 | } | |
1121 | ||
932cc6d4 JA |
1122 | /** |
1123 | * do_splice_direct - splices data directly between two files | |
1124 | * @in: file to splice from | |
1125 | * @ppos: input file offset | |
1126 | * @out: file to splice to | |
1127 | * @len: number of bytes to splice | |
1128 | * @flags: splice modifier flags | |
1129 | * | |
1130 | * Description: | |
1131 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1132 | * doing it in the application would incur an extra system call | |
1133 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1134 | * can splice directly through a process-private pipe. | |
1135 | * | |
1136 | */ | |
c66ab6fa JA |
1137 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
1138 | size_t len, unsigned int flags) | |
1139 | { | |
1140 | struct splice_desc sd = { | |
1141 | .len = len, | |
1142 | .total_len = len, | |
1143 | .flags = flags, | |
1144 | .pos = *ppos, | |
6a14b90b | 1145 | .u.file = out, |
c66ab6fa | 1146 | }; |
51a92c0f | 1147 | long ret; |
c66ab6fa JA |
1148 | |
1149 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); | |
51a92c0f JA |
1150 | if (ret > 0) |
1151 | *ppos += ret; | |
1152 | ||
c66ab6fa | 1153 | return ret; |
b92ce558 JA |
1154 | } |
1155 | ||
ddac0d39 JA |
1156 | /* |
1157 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1158 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1159 | * pipe. | |
1160 | */ | |
1161 | static inline struct pipe_inode_info *pipe_info(struct inode *inode) | |
1162 | { | |
1163 | if (S_ISFIFO(inode->i_mode)) | |
1164 | return inode->i_pipe; | |
1165 | ||
1166 | return NULL; | |
1167 | } | |
1168 | ||
83f9135b JA |
1169 | /* |
1170 | * Determine where to splice to/from. | |
1171 | */ | |
529565dc IM |
1172 | static long do_splice(struct file *in, loff_t __user *off_in, |
1173 | struct file *out, loff_t __user *off_out, | |
1174 | size_t len, unsigned int flags) | |
5274f052 | 1175 | { |
3a326a2c | 1176 | struct pipe_inode_info *pipe; |
cbb7e577 | 1177 | loff_t offset, *off; |
a4514ebd | 1178 | long ret; |
5274f052 | 1179 | |
0f7fc9e4 | 1180 | pipe = pipe_info(in->f_path.dentry->d_inode); |
529565dc IM |
1181 | if (pipe) { |
1182 | if (off_in) | |
1183 | return -ESPIPE; | |
b92ce558 JA |
1184 | if (off_out) { |
1185 | if (out->f_op->llseek == no_llseek) | |
1186 | return -EINVAL; | |
cbb7e577 | 1187 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1188 | return -EFAULT; |
cbb7e577 JA |
1189 | off = &offset; |
1190 | } else | |
1191 | off = &out->f_pos; | |
529565dc | 1192 | |
a4514ebd JA |
1193 | ret = do_splice_from(pipe, out, off, len, flags); |
1194 | ||
1195 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1196 | ret = -EFAULT; | |
1197 | ||
1198 | return ret; | |
529565dc | 1199 | } |
5274f052 | 1200 | |
0f7fc9e4 | 1201 | pipe = pipe_info(out->f_path.dentry->d_inode); |
529565dc IM |
1202 | if (pipe) { |
1203 | if (off_out) | |
1204 | return -ESPIPE; | |
b92ce558 JA |
1205 | if (off_in) { |
1206 | if (in->f_op->llseek == no_llseek) | |
1207 | return -EINVAL; | |
cbb7e577 | 1208 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1209 | return -EFAULT; |
cbb7e577 JA |
1210 | off = &offset; |
1211 | } else | |
1212 | off = &in->f_pos; | |
529565dc | 1213 | |
a4514ebd JA |
1214 | ret = do_splice_to(in, off, pipe, len, flags); |
1215 | ||
1216 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1217 | ret = -EFAULT; | |
1218 | ||
1219 | return ret; | |
529565dc | 1220 | } |
5274f052 JA |
1221 | |
1222 | return -EINVAL; | |
1223 | } | |
1224 | ||
912d35f8 JA |
1225 | /* |
1226 | * Map an iov into an array of pages and offset/length tupples. With the | |
1227 | * partial_page structure, we can map several non-contiguous ranges into | |
1228 | * our ones pages[] map instead of splitting that operation into pieces. | |
1229 | * Could easily be exported as a generic helper for other users, in which | |
1230 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1231 | */ | |
1232 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1233 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1234 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1235 | { |
1236 | int buffers = 0, error = 0; | |
1237 | ||
1238 | /* | |
1239 | * It's ok to take the mmap_sem for reading, even | |
1240 | * across a "get_user()". | |
1241 | */ | |
1242 | down_read(¤t->mm->mmap_sem); | |
1243 | ||
1244 | while (nr_vecs) { | |
1245 | unsigned long off, npages; | |
1246 | void __user *base; | |
1247 | size_t len; | |
1248 | int i; | |
1249 | ||
1250 | /* | |
1251 | * Get user address base and length for this iovec. | |
1252 | */ | |
1253 | error = get_user(base, &iov->iov_base); | |
1254 | if (unlikely(error)) | |
1255 | break; | |
1256 | error = get_user(len, &iov->iov_len); | |
1257 | if (unlikely(error)) | |
1258 | break; | |
1259 | ||
1260 | /* | |
1261 | * Sanity check this iovec. 0 read succeeds. | |
1262 | */ | |
1263 | if (unlikely(!len)) | |
1264 | break; | |
1265 | error = -EFAULT; | |
1266 | if (unlikely(!base)) | |
1267 | break; | |
1268 | ||
1269 | /* | |
1270 | * Get this base offset and number of pages, then map | |
1271 | * in the user pages. | |
1272 | */ | |
1273 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1274 | |
1275 | /* | |
1276 | * If asked for alignment, the offset must be zero and the | |
1277 | * length a multiple of the PAGE_SIZE. | |
1278 | */ | |
1279 | error = -EINVAL; | |
1280 | if (aligned && (off || len & ~PAGE_MASK)) | |
1281 | break; | |
1282 | ||
912d35f8 JA |
1283 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1284 | if (npages > PIPE_BUFFERS - buffers) | |
1285 | npages = PIPE_BUFFERS - buffers; | |
1286 | ||
1287 | error = get_user_pages(current, current->mm, | |
1288 | (unsigned long) base, npages, 0, 0, | |
1289 | &pages[buffers], NULL); | |
1290 | ||
1291 | if (unlikely(error <= 0)) | |
1292 | break; | |
1293 | ||
1294 | /* | |
1295 | * Fill this contiguous range into the partial page map. | |
1296 | */ | |
1297 | for (i = 0; i < error; i++) { | |
7591489a | 1298 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1299 | |
1300 | partial[buffers].offset = off; | |
1301 | partial[buffers].len = plen; | |
1302 | ||
1303 | off = 0; | |
1304 | len -= plen; | |
1305 | buffers++; | |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * We didn't complete this iov, stop here since it probably | |
1310 | * means we have to move some of this into a pipe to | |
1311 | * be able to continue. | |
1312 | */ | |
1313 | if (len) | |
1314 | break; | |
1315 | ||
1316 | /* | |
1317 | * Don't continue if we mapped fewer pages than we asked for, | |
1318 | * or if we mapped the max number of pages that we have | |
1319 | * room for. | |
1320 | */ | |
1321 | if (error < npages || buffers == PIPE_BUFFERS) | |
1322 | break; | |
1323 | ||
1324 | nr_vecs--; | |
1325 | iov++; | |
1326 | } | |
1327 | ||
1328 | up_read(¤t->mm->mmap_sem); | |
1329 | ||
1330 | if (buffers) | |
1331 | return buffers; | |
1332 | ||
1333 | return error; | |
1334 | } | |
1335 | ||
6a14b90b JA |
1336 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1337 | struct splice_desc *sd) | |
1338 | { | |
1339 | char *src; | |
1340 | int ret; | |
1341 | ||
cac36bb0 | 1342 | ret = buf->ops->confirm(pipe, buf); |
6a14b90b JA |
1343 | if (unlikely(ret)) |
1344 | return ret; | |
1345 | ||
1346 | /* | |
1347 | * See if we can use the atomic maps, by prefaulting in the | |
1348 | * pages and doing an atomic copy | |
1349 | */ | |
1350 | if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { | |
1351 | src = buf->ops->map(pipe, buf, 1); | |
1352 | ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, | |
1353 | sd->len); | |
1354 | buf->ops->unmap(pipe, buf, src); | |
1355 | if (!ret) { | |
1356 | ret = sd->len; | |
1357 | goto out; | |
1358 | } | |
1359 | } | |
1360 | ||
1361 | /* | |
1362 | * No dice, use slow non-atomic map and copy | |
1363 | */ | |
1364 | src = buf->ops->map(pipe, buf, 0); | |
1365 | ||
1366 | ret = sd->len; | |
1367 | if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) | |
1368 | ret = -EFAULT; | |
1369 | ||
1370 | out: | |
1371 | if (ret > 0) | |
1372 | sd->u.userptr += ret; | |
1373 | buf->ops->unmap(pipe, buf, src); | |
1374 | return ret; | |
1375 | } | |
1376 | ||
1377 | /* | |
1378 | * For lack of a better implementation, implement vmsplice() to userspace | |
1379 | * as a simple copy of the pipes pages to the user iov. | |
1380 | */ | |
1381 | static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, | |
1382 | unsigned long nr_segs, unsigned int flags) | |
1383 | { | |
1384 | struct pipe_inode_info *pipe; | |
1385 | struct splice_desc sd; | |
1386 | ssize_t size; | |
1387 | int error; | |
1388 | long ret; | |
1389 | ||
1390 | pipe = pipe_info(file->f_path.dentry->d_inode); | |
1391 | if (!pipe) | |
1392 | return -EBADF; | |
1393 | ||
1394 | if (pipe->inode) | |
1395 | mutex_lock(&pipe->inode->i_mutex); | |
1396 | ||
1397 | error = ret = 0; | |
1398 | while (nr_segs) { | |
1399 | void __user *base; | |
1400 | size_t len; | |
1401 | ||
1402 | /* | |
1403 | * Get user address base and length for this iovec. | |
1404 | */ | |
1405 | error = get_user(base, &iov->iov_base); | |
1406 | if (unlikely(error)) | |
1407 | break; | |
1408 | error = get_user(len, &iov->iov_len); | |
1409 | if (unlikely(error)) | |
1410 | break; | |
1411 | ||
1412 | /* | |
1413 | * Sanity check this iovec. 0 read succeeds. | |
1414 | */ | |
1415 | if (unlikely(!len)) | |
1416 | break; | |
1417 | if (unlikely(!base)) { | |
1418 | error = -EFAULT; | |
1419 | break; | |
1420 | } | |
1421 | ||
1422 | sd.len = 0; | |
1423 | sd.total_len = len; | |
1424 | sd.flags = flags; | |
1425 | sd.u.userptr = base; | |
1426 | sd.pos = 0; | |
1427 | ||
1428 | size = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1429 | if (size < 0) { | |
1430 | if (!ret) | |
1431 | ret = size; | |
1432 | ||
1433 | break; | |
1434 | } | |
1435 | ||
1436 | ret += size; | |
1437 | ||
1438 | if (size < len) | |
1439 | break; | |
1440 | ||
1441 | nr_segs--; | |
1442 | iov++; | |
1443 | } | |
1444 | ||
1445 | if (pipe->inode) | |
1446 | mutex_unlock(&pipe->inode->i_mutex); | |
1447 | ||
1448 | if (!ret) | |
1449 | ret = error; | |
1450 | ||
1451 | return ret; | |
1452 | } | |
1453 | ||
912d35f8 JA |
1454 | /* |
1455 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1456 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1457 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1458 | */ |
6a14b90b JA |
1459 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1460 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1461 | { |
ddac0d39 | 1462 | struct pipe_inode_info *pipe; |
912d35f8 JA |
1463 | struct page *pages[PIPE_BUFFERS]; |
1464 | struct partial_page partial[PIPE_BUFFERS]; | |
1465 | struct splice_pipe_desc spd = { | |
1466 | .pages = pages, | |
1467 | .partial = partial, | |
1468 | .flags = flags, | |
1469 | .ops = &user_page_pipe_buf_ops, | |
1470 | }; | |
1471 | ||
0f7fc9e4 | 1472 | pipe = pipe_info(file->f_path.dentry->d_inode); |
ddac0d39 | 1473 | if (!pipe) |
912d35f8 | 1474 | return -EBADF; |
912d35f8 | 1475 | |
7afa6fd0 JA |
1476 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1477 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1478 | if (spd.nr_pages <= 0) |
1479 | return spd.nr_pages; | |
1480 | ||
00522fb4 | 1481 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1482 | } |
1483 | ||
6a14b90b JA |
1484 | /* |
1485 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1486 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1487 | * operation that can be supported without any funky alignment restrictions | |
1488 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1489 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1490 | * solutions for that: | |
1491 | * | |
1492 | * - memcpy() the data internally, at which point we might as well just | |
1493 | * do a regular read() on the buffer anyway. | |
1494 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1495 | * has restriction limitations on both ends of the pipe). | |
1496 | * | |
1497 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1498 | * | |
1499 | */ | |
912d35f8 JA |
1500 | asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, |
1501 | unsigned long nr_segs, unsigned int flags) | |
1502 | { | |
1503 | struct file *file; | |
1504 | long error; | |
1505 | int fput; | |
1506 | ||
6a14b90b JA |
1507 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1508 | return -EINVAL; | |
1509 | else if (unlikely(!nr_segs)) | |
1510 | return 0; | |
1511 | ||
912d35f8 JA |
1512 | error = -EBADF; |
1513 | file = fget_light(fd, &fput); | |
1514 | if (file) { | |
1515 | if (file->f_mode & FMODE_WRITE) | |
6a14b90b JA |
1516 | error = vmsplice_to_pipe(file, iov, nr_segs, flags); |
1517 | else if (file->f_mode & FMODE_READ) | |
1518 | error = vmsplice_to_user(file, iov, nr_segs, flags); | |
912d35f8 JA |
1519 | |
1520 | fput_light(file, fput); | |
1521 | } | |
1522 | ||
1523 | return error; | |
1524 | } | |
1525 | ||
529565dc IM |
1526 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
1527 | int fd_out, loff_t __user *off_out, | |
1528 | size_t len, unsigned int flags) | |
5274f052 JA |
1529 | { |
1530 | long error; | |
1531 | struct file *in, *out; | |
1532 | int fput_in, fput_out; | |
1533 | ||
1534 | if (unlikely(!len)) | |
1535 | return 0; | |
1536 | ||
1537 | error = -EBADF; | |
529565dc | 1538 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1539 | if (in) { |
1540 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1541 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1542 | if (out) { |
1543 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1544 | error = do_splice(in, off_in, |
1545 | out, off_out, | |
1546 | len, flags); | |
5274f052 JA |
1547 | fput_light(out, fput_out); |
1548 | } | |
1549 | } | |
1550 | ||
1551 | fput_light(in, fput_in); | |
1552 | } | |
1553 | ||
1554 | return error; | |
1555 | } | |
70524490 | 1556 | |
aadd06e5 JA |
1557 | /* |
1558 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1559 | * return an appropriate error. | |
1560 | */ | |
1561 | static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1562 | { | |
1563 | int ret; | |
1564 | ||
1565 | /* | |
1566 | * Check ->nrbufs without the inode lock first. This function | |
1567 | * is speculative anyways, so missing one is ok. | |
1568 | */ | |
1569 | if (pipe->nrbufs) | |
1570 | return 0; | |
1571 | ||
1572 | ret = 0; | |
1573 | mutex_lock(&pipe->inode->i_mutex); | |
1574 | ||
1575 | while (!pipe->nrbufs) { | |
1576 | if (signal_pending(current)) { | |
1577 | ret = -ERESTARTSYS; | |
1578 | break; | |
1579 | } | |
1580 | if (!pipe->writers) | |
1581 | break; | |
1582 | if (!pipe->waiting_writers) { | |
1583 | if (flags & SPLICE_F_NONBLOCK) { | |
1584 | ret = -EAGAIN; | |
1585 | break; | |
1586 | } | |
1587 | } | |
1588 | pipe_wait(pipe); | |
1589 | } | |
1590 | ||
1591 | mutex_unlock(&pipe->inode->i_mutex); | |
1592 | return ret; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1597 | * return an appropriate error. | |
1598 | */ | |
1599 | static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1600 | { | |
1601 | int ret; | |
1602 | ||
1603 | /* | |
1604 | * Check ->nrbufs without the inode lock first. This function | |
1605 | * is speculative anyways, so missing one is ok. | |
1606 | */ | |
1607 | if (pipe->nrbufs < PIPE_BUFFERS) | |
1608 | return 0; | |
1609 | ||
1610 | ret = 0; | |
1611 | mutex_lock(&pipe->inode->i_mutex); | |
1612 | ||
1613 | while (pipe->nrbufs >= PIPE_BUFFERS) { | |
1614 | if (!pipe->readers) { | |
1615 | send_sig(SIGPIPE, current, 0); | |
1616 | ret = -EPIPE; | |
1617 | break; | |
1618 | } | |
1619 | if (flags & SPLICE_F_NONBLOCK) { | |
1620 | ret = -EAGAIN; | |
1621 | break; | |
1622 | } | |
1623 | if (signal_pending(current)) { | |
1624 | ret = -ERESTARTSYS; | |
1625 | break; | |
1626 | } | |
1627 | pipe->waiting_writers++; | |
1628 | pipe_wait(pipe); | |
1629 | pipe->waiting_writers--; | |
1630 | } | |
1631 | ||
1632 | mutex_unlock(&pipe->inode->i_mutex); | |
1633 | return ret; | |
1634 | } | |
1635 | ||
70524490 JA |
1636 | /* |
1637 | * Link contents of ipipe to opipe. | |
1638 | */ | |
1639 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1640 | struct pipe_inode_info *opipe, | |
1641 | size_t len, unsigned int flags) | |
1642 | { | |
1643 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1644 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1645 | |
1646 | /* | |
1647 | * Potential ABBA deadlock, work around it by ordering lock | |
1648 | * grabbing by inode address. Otherwise two different processes | |
1649 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1650 | */ | |
62752ee1 | 1651 | inode_double_lock(ipipe->inode, opipe->inode); |
70524490 | 1652 | |
aadd06e5 | 1653 | do { |
70524490 JA |
1654 | if (!opipe->readers) { |
1655 | send_sig(SIGPIPE, current, 0); | |
1656 | if (!ret) | |
1657 | ret = -EPIPE; | |
1658 | break; | |
1659 | } | |
70524490 | 1660 | |
aadd06e5 JA |
1661 | /* |
1662 | * If we have iterated all input buffers or ran out of | |
1663 | * output room, break. | |
1664 | */ | |
1665 | if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) | |
1666 | break; | |
70524490 | 1667 | |
aadd06e5 JA |
1668 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); |
1669 | nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
70524490 JA |
1670 | |
1671 | /* | |
aadd06e5 JA |
1672 | * Get a reference to this pipe buffer, |
1673 | * so we can copy the contents over. | |
70524490 | 1674 | */ |
aadd06e5 JA |
1675 | ibuf->ops->get(ipipe, ibuf); |
1676 | ||
1677 | obuf = opipe->bufs + nbuf; | |
1678 | *obuf = *ibuf; | |
1679 | ||
2a27250e | 1680 | /* |
aadd06e5 JA |
1681 | * Don't inherit the gift flag, we need to |
1682 | * prevent multiple steals of this page. | |
2a27250e | 1683 | */ |
aadd06e5 | 1684 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1685 | |
aadd06e5 JA |
1686 | if (obuf->len > len) |
1687 | obuf->len = len; | |
70524490 | 1688 | |
aadd06e5 JA |
1689 | opipe->nrbufs++; |
1690 | ret += obuf->len; | |
1691 | len -= obuf->len; | |
1692 | i++; | |
1693 | } while (len); | |
70524490 | 1694 | |
62752ee1 | 1695 | inode_double_unlock(ipipe->inode, opipe->inode); |
70524490 | 1696 | |
aadd06e5 JA |
1697 | /* |
1698 | * If we put data in the output pipe, wakeup any potential readers. | |
1699 | */ | |
1700 | if (ret > 0) { | |
70524490 JA |
1701 | smp_mb(); |
1702 | if (waitqueue_active(&opipe->wait)) | |
1703 | wake_up_interruptible(&opipe->wait); | |
1704 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1705 | } | |
1706 | ||
1707 | return ret; | |
1708 | } | |
1709 | ||
1710 | /* | |
1711 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1712 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1713 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1714 | * applicable one is SPLICE_F_NONBLOCK. | |
1715 | */ | |
1716 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1717 | unsigned int flags) | |
1718 | { | |
0f7fc9e4 JJS |
1719 | struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); |
1720 | struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); | |
aadd06e5 | 1721 | int ret = -EINVAL; |
70524490 JA |
1722 | |
1723 | /* | |
aadd06e5 JA |
1724 | * Duplicate the contents of ipipe to opipe without actually |
1725 | * copying the data. | |
70524490 | 1726 | */ |
aadd06e5 JA |
1727 | if (ipipe && opipe && ipipe != opipe) { |
1728 | /* | |
1729 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1730 | * ordering doesn't really matter. | |
1731 | */ | |
1732 | ret = link_ipipe_prep(ipipe, flags); | |
1733 | if (!ret) { | |
1734 | ret = link_opipe_prep(opipe, flags); | |
1735 | if (!ret) { | |
1736 | ret = link_pipe(ipipe, opipe, len, flags); | |
1737 | if (!ret && (flags & SPLICE_F_NONBLOCK)) | |
1738 | ret = -EAGAIN; | |
1739 | } | |
1740 | } | |
1741 | } | |
70524490 | 1742 | |
aadd06e5 | 1743 | return ret; |
70524490 JA |
1744 | } |
1745 | ||
1746 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1747 | { | |
1748 | struct file *in; | |
1749 | int error, fput_in; | |
1750 | ||
1751 | if (unlikely(!len)) | |
1752 | return 0; | |
1753 | ||
1754 | error = -EBADF; | |
1755 | in = fget_light(fdin, &fput_in); | |
1756 | if (in) { | |
1757 | if (in->f_mode & FMODE_READ) { | |
1758 | int fput_out; | |
1759 | struct file *out = fget_light(fdout, &fput_out); | |
1760 | ||
1761 | if (out) { | |
1762 | if (out->f_mode & FMODE_WRITE) | |
1763 | error = do_tee(in, out, len, flags); | |
1764 | fput_light(out, fput_out); | |
1765 | } | |
1766 | } | |
1767 | fput_light(in, fput_in); | |
1768 | } | |
1769 | ||
1770 | return error; | |
1771 | } |