]>
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> |
08e552c6 | 24 | #include <linux/memcontrol.h> |
5274f052 | 25 | #include <linux/mm_inline.h> |
5abc97aa | 26 | #include <linux/swap.h> |
4f6f0bd2 JA |
27 | #include <linux/writeback.h> |
28 | #include <linux/buffer_head.h> | |
a0f06780 | 29 | #include <linux/module.h> |
4f6f0bd2 | 30 | #include <linux/syscalls.h> |
912d35f8 | 31 | #include <linux/uio.h> |
29ce2058 | 32 | #include <linux/security.h> |
5274f052 | 33 | |
83f9135b JA |
34 | /* |
35 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
36 | * a vm helper function, it's already simplified quite a bit by the | |
37 | * addition of remove_mapping(). If success is returned, the caller may | |
38 | * attempt to reuse this page for another destination. | |
39 | */ | |
76ad4d11 | 40 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, |
5abc97aa JA |
41 | struct pipe_buffer *buf) |
42 | { | |
43 | struct page *page = buf->page; | |
9e94cd4f | 44 | struct address_space *mapping; |
5abc97aa | 45 | |
9e0267c2 JA |
46 | lock_page(page); |
47 | ||
9e94cd4f JA |
48 | mapping = page_mapping(page); |
49 | if (mapping) { | |
50 | WARN_ON(!PageUptodate(page)); | |
5abc97aa | 51 | |
9e94cd4f JA |
52 | /* |
53 | * At least for ext2 with nobh option, we need to wait on | |
54 | * writeback completing on this page, since we'll remove it | |
55 | * from the pagecache. Otherwise truncate wont wait on the | |
56 | * page, allowing the disk blocks to be reused by someone else | |
57 | * before we actually wrote our data to them. fs corruption | |
58 | * ensues. | |
59 | */ | |
60 | wait_on_page_writeback(page); | |
ad8d6f0a | 61 | |
266cf658 DH |
62 | if (page_has_private(page) && |
63 | !try_to_release_page(page, GFP_KERNEL)) | |
ca39d651 | 64 | goto out_unlock; |
4f6f0bd2 | 65 | |
9e94cd4f JA |
66 | /* |
67 | * If we succeeded in removing the mapping, set LRU flag | |
68 | * and return good. | |
69 | */ | |
70 | if (remove_mapping(mapping, page)) { | |
71 | buf->flags |= PIPE_BUF_FLAG_LRU; | |
72 | return 0; | |
73 | } | |
9e0267c2 | 74 | } |
5abc97aa | 75 | |
9e94cd4f JA |
76 | /* |
77 | * Raced with truncate or failed to remove page from current | |
78 | * address space, unlock and return failure. | |
79 | */ | |
ca39d651 | 80 | out_unlock: |
9e94cd4f JA |
81 | unlock_page(page); |
82 | return 1; | |
5abc97aa JA |
83 | } |
84 | ||
76ad4d11 | 85 | static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, |
5274f052 JA |
86 | struct pipe_buffer *buf) |
87 | { | |
88 | page_cache_release(buf->page); | |
1432873a | 89 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
5274f052 JA |
90 | } |
91 | ||
0845718d JA |
92 | /* |
93 | * Check whether the contents of buf is OK to access. Since the content | |
94 | * is a page cache page, IO may be in flight. | |
95 | */ | |
cac36bb0 JA |
96 | static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, |
97 | struct pipe_buffer *buf) | |
5274f052 JA |
98 | { |
99 | struct page *page = buf->page; | |
49d0b21b | 100 | int err; |
5274f052 JA |
101 | |
102 | if (!PageUptodate(page)) { | |
49d0b21b JA |
103 | lock_page(page); |
104 | ||
105 | /* | |
106 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 107 | * splice, if this is the first page. |
49d0b21b JA |
108 | */ |
109 | if (!page->mapping) { | |
110 | err = -ENODATA; | |
111 | goto error; | |
112 | } | |
5274f052 | 113 | |
49d0b21b | 114 | /* |
73d62d83 | 115 | * Uh oh, read-error from disk. |
49d0b21b JA |
116 | */ |
117 | if (!PageUptodate(page)) { | |
118 | err = -EIO; | |
119 | goto error; | |
120 | } | |
121 | ||
122 | /* | |
f84d7519 | 123 | * Page is ok afterall, we are done. |
49d0b21b | 124 | */ |
5274f052 | 125 | unlock_page(page); |
5274f052 JA |
126 | } |
127 | ||
f84d7519 | 128 | return 0; |
49d0b21b JA |
129 | error: |
130 | unlock_page(page); | |
f84d7519 | 131 | return err; |
70524490 JA |
132 | } |
133 | ||
d4c3cca9 | 134 | static const struct pipe_buf_operations page_cache_pipe_buf_ops = { |
5274f052 | 135 | .can_merge = 0, |
f84d7519 JA |
136 | .map = generic_pipe_buf_map, |
137 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 138 | .confirm = page_cache_pipe_buf_confirm, |
5274f052 | 139 | .release = page_cache_pipe_buf_release, |
5abc97aa | 140 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 141 | .get = generic_pipe_buf_get, |
5274f052 JA |
142 | }; |
143 | ||
912d35f8 JA |
144 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
145 | struct pipe_buffer *buf) | |
146 | { | |
7afa6fd0 JA |
147 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
148 | return 1; | |
149 | ||
1432873a | 150 | buf->flags |= PIPE_BUF_FLAG_LRU; |
330ab716 | 151 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
152 | } |
153 | ||
d4c3cca9 | 154 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
912d35f8 | 155 | .can_merge = 0, |
f84d7519 JA |
156 | .map = generic_pipe_buf_map, |
157 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 158 | .confirm = generic_pipe_buf_confirm, |
912d35f8 JA |
159 | .release = page_cache_pipe_buf_release, |
160 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 161 | .get = generic_pipe_buf_get, |
912d35f8 JA |
162 | }; |
163 | ||
932cc6d4 JA |
164 | /** |
165 | * splice_to_pipe - fill passed data into a pipe | |
166 | * @pipe: pipe to fill | |
167 | * @spd: data to fill | |
168 | * | |
169 | * Description: | |
79685b8d | 170 | * @spd contains a map of pages and len/offset tuples, along with |
932cc6d4 JA |
171 | * the struct pipe_buf_operations associated with these pages. This |
172 | * function will link that data to the pipe. | |
173 | * | |
83f9135b | 174 | */ |
d6b29d7c JA |
175 | ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
176 | struct splice_pipe_desc *spd) | |
5274f052 | 177 | { |
00de00bd | 178 | unsigned int spd_pages = spd->nr_pages; |
912d35f8 | 179 | int ret, do_wakeup, page_nr; |
5274f052 JA |
180 | |
181 | ret = 0; | |
182 | do_wakeup = 0; | |
912d35f8 | 183 | page_nr = 0; |
5274f052 | 184 | |
3a326a2c IM |
185 | if (pipe->inode) |
186 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 187 | |
5274f052 | 188 | for (;;) { |
3a326a2c | 189 | if (!pipe->readers) { |
5274f052 JA |
190 | send_sig(SIGPIPE, current, 0); |
191 | if (!ret) | |
192 | ret = -EPIPE; | |
193 | break; | |
194 | } | |
195 | ||
6f767b04 JA |
196 | if (pipe->nrbufs < PIPE_BUFFERS) { |
197 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 198 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 199 | |
912d35f8 JA |
200 | buf->page = spd->pages[page_nr]; |
201 | buf->offset = spd->partial[page_nr].offset; | |
202 | buf->len = spd->partial[page_nr].len; | |
497f9625 | 203 | buf->private = spd->partial[page_nr].private; |
912d35f8 | 204 | buf->ops = spd->ops; |
7afa6fd0 JA |
205 | if (spd->flags & SPLICE_F_GIFT) |
206 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
207 | ||
6f767b04 | 208 | pipe->nrbufs++; |
912d35f8 JA |
209 | page_nr++; |
210 | ret += buf->len; | |
211 | ||
6f767b04 JA |
212 | if (pipe->inode) |
213 | do_wakeup = 1; | |
5274f052 | 214 | |
912d35f8 | 215 | if (!--spd->nr_pages) |
5274f052 | 216 | break; |
6f767b04 | 217 | if (pipe->nrbufs < PIPE_BUFFERS) |
5274f052 JA |
218 | continue; |
219 | ||
220 | break; | |
221 | } | |
222 | ||
912d35f8 | 223 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
224 | if (!ret) |
225 | ret = -EAGAIN; | |
226 | break; | |
227 | } | |
228 | ||
5274f052 JA |
229 | if (signal_pending(current)) { |
230 | if (!ret) | |
231 | ret = -ERESTARTSYS; | |
232 | break; | |
233 | } | |
234 | ||
235 | if (do_wakeup) { | |
c0bd1f65 | 236 | smp_mb(); |
3a326a2c IM |
237 | if (waitqueue_active(&pipe->wait)) |
238 | wake_up_interruptible_sync(&pipe->wait); | |
239 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
240 | do_wakeup = 0; |
241 | } | |
242 | ||
3a326a2c IM |
243 | pipe->waiting_writers++; |
244 | pipe_wait(pipe); | |
245 | pipe->waiting_writers--; | |
5274f052 JA |
246 | } |
247 | ||
02676e5a | 248 | if (pipe->inode) { |
3a326a2c | 249 | mutex_unlock(&pipe->inode->i_mutex); |
5274f052 | 250 | |
02676e5a JA |
251 | if (do_wakeup) { |
252 | smp_mb(); | |
253 | if (waitqueue_active(&pipe->wait)) | |
254 | wake_up_interruptible(&pipe->wait); | |
255 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
256 | } | |
5274f052 JA |
257 | } |
258 | ||
00de00bd | 259 | while (page_nr < spd_pages) |
bbdfc2f7 | 260 | spd->spd_release(spd, page_nr++); |
5274f052 JA |
261 | |
262 | return ret; | |
263 | } | |
264 | ||
bbdfc2f7 JA |
265 | static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) |
266 | { | |
267 | page_cache_release(spd->pages[i]); | |
268 | } | |
269 | ||
3a326a2c | 270 | static int |
cbb7e577 JA |
271 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
272 | struct pipe_inode_info *pipe, size_t len, | |
273 | unsigned int flags) | |
5274f052 JA |
274 | { |
275 | struct address_space *mapping = in->f_mapping; | |
d8983910 | 276 | unsigned int loff, nr_pages, req_pages; |
16c523dd | 277 | struct page *pages[PIPE_BUFFERS]; |
912d35f8 | 278 | struct partial_page partial[PIPE_BUFFERS]; |
5274f052 | 279 | struct page *page; |
91ad66ef JA |
280 | pgoff_t index, end_index; |
281 | loff_t isize; | |
eb20796b | 282 | int error, page_nr; |
912d35f8 JA |
283 | struct splice_pipe_desc spd = { |
284 | .pages = pages, | |
285 | .partial = partial, | |
286 | .flags = flags, | |
287 | .ops = &page_cache_pipe_buf_ops, | |
bbdfc2f7 | 288 | .spd_release = spd_release_page, |
912d35f8 | 289 | }; |
5274f052 | 290 | |
cbb7e577 | 291 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 | 292 | loff = *ppos & ~PAGE_CACHE_MASK; |
d8983910 FW |
293 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
294 | nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS); | |
5274f052 | 295 | |
eb20796b JA |
296 | /* |
297 | * Lookup the (hopefully) full range of pages we need. | |
298 | */ | |
299 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
431a4820 | 300 | index += spd.nr_pages; |
82aa5d61 | 301 | |
eb20796b JA |
302 | /* |
303 | * If find_get_pages_contig() returned fewer pages than we needed, | |
431a4820 | 304 | * readahead/allocate the rest and fill in the holes. |
eb20796b | 305 | */ |
431a4820 | 306 | if (spd.nr_pages < nr_pages) |
cf914a7d RR |
307 | page_cache_sync_readahead(mapping, &in->f_ra, in, |
308 | index, req_pages - spd.nr_pages); | |
431a4820 | 309 | |
932cc6d4 | 310 | error = 0; |
eb20796b | 311 | while (spd.nr_pages < nr_pages) { |
82aa5d61 | 312 | /* |
eb20796b JA |
313 | * Page could be there, find_get_pages_contig() breaks on |
314 | * the first hole. | |
5274f052 | 315 | */ |
7480a904 JA |
316 | page = find_get_page(mapping, index); |
317 | if (!page) { | |
7480a904 | 318 | /* |
eb20796b | 319 | * page didn't exist, allocate one. |
7480a904 JA |
320 | */ |
321 | page = page_cache_alloc_cold(mapping); | |
322 | if (!page) | |
323 | break; | |
324 | ||
325 | error = add_to_page_cache_lru(page, mapping, index, | |
4cd13504 | 326 | mapping_gfp_mask(mapping)); |
7480a904 JA |
327 | if (unlikely(error)) { |
328 | page_cache_release(page); | |
a0548871 JA |
329 | if (error == -EEXIST) |
330 | continue; | |
7480a904 JA |
331 | break; |
332 | } | |
eb20796b JA |
333 | /* |
334 | * add_to_page_cache() locks the page, unlock it | |
335 | * to avoid convoluting the logic below even more. | |
336 | */ | |
337 | unlock_page(page); | |
7480a904 JA |
338 | } |
339 | ||
eb20796b JA |
340 | pages[spd.nr_pages++] = page; |
341 | index++; | |
342 | } | |
343 | ||
344 | /* | |
345 | * Now loop over the map and see if we need to start IO on any | |
346 | * pages, fill in the partial map, etc. | |
347 | */ | |
348 | index = *ppos >> PAGE_CACHE_SHIFT; | |
349 | nr_pages = spd.nr_pages; | |
350 | spd.nr_pages = 0; | |
351 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
352 | unsigned int this_len; | |
353 | ||
354 | if (!len) | |
355 | break; | |
356 | ||
357 | /* | |
358 | * this_len is the max we'll use from this page | |
359 | */ | |
360 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
361 | page = pages[page_nr]; | |
362 | ||
a08a166f | 363 | if (PageReadahead(page)) |
cf914a7d | 364 | page_cache_async_readahead(mapping, &in->f_ra, in, |
d8983910 | 365 | page, index, req_pages - page_nr); |
a08a166f | 366 | |
7480a904 JA |
367 | /* |
368 | * If the page isn't uptodate, we may need to start io on it | |
369 | */ | |
370 | if (!PageUptodate(page)) { | |
c4f895cb JA |
371 | /* |
372 | * If in nonblock mode then dont block on waiting | |
373 | * for an in-flight io page | |
374 | */ | |
9ae9d68c | 375 | if (flags & SPLICE_F_NONBLOCK) { |
529ae9aa | 376 | if (!trylock_page(page)) { |
8191ecd1 | 377 | error = -EAGAIN; |
9ae9d68c | 378 | break; |
8191ecd1 | 379 | } |
9ae9d68c FW |
380 | } else |
381 | lock_page(page); | |
7480a904 JA |
382 | |
383 | /* | |
32502b84 MS |
384 | * Page was truncated, or invalidated by the |
385 | * filesystem. Redo the find/create, but this time the | |
386 | * page is kept locked, so there's no chance of another | |
387 | * race with truncate/invalidate. | |
7480a904 JA |
388 | */ |
389 | if (!page->mapping) { | |
390 | unlock_page(page); | |
32502b84 MS |
391 | page = find_or_create_page(mapping, index, |
392 | mapping_gfp_mask(mapping)); | |
393 | ||
394 | if (!page) { | |
395 | error = -ENOMEM; | |
396 | break; | |
397 | } | |
398 | page_cache_release(pages[page_nr]); | |
399 | pages[page_nr] = page; | |
7480a904 JA |
400 | } |
401 | /* | |
402 | * page was already under io and is now done, great | |
403 | */ | |
404 | if (PageUptodate(page)) { | |
405 | unlock_page(page); | |
406 | goto fill_it; | |
407 | } | |
5274f052 | 408 | |
7480a904 JA |
409 | /* |
410 | * need to read in the page | |
411 | */ | |
412 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 413 | if (unlikely(error)) { |
eb20796b JA |
414 | /* |
415 | * We really should re-lookup the page here, | |
416 | * but it complicates things a lot. Instead | |
417 | * lets just do what we already stored, and | |
418 | * we'll get it the next time we are called. | |
419 | */ | |
7480a904 | 420 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
421 | error = 0; |
422 | ||
5274f052 JA |
423 | break; |
424 | } | |
620a324b JA |
425 | } |
426 | fill_it: | |
427 | /* | |
428 | * i_size must be checked after PageUptodate. | |
429 | */ | |
430 | isize = i_size_read(mapping->host); | |
431 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
432 | if (unlikely(!isize || index > end_index)) | |
433 | break; | |
434 | ||
435 | /* | |
436 | * if this is the last page, see if we need to shrink | |
437 | * the length and stop | |
438 | */ | |
439 | if (end_index == index) { | |
440 | unsigned int plen; | |
91ad66ef JA |
441 | |
442 | /* | |
620a324b | 443 | * max good bytes in this page |
91ad66ef | 444 | */ |
620a324b JA |
445 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
446 | if (plen <= loff) | |
91ad66ef | 447 | break; |
91ad66ef JA |
448 | |
449 | /* | |
620a324b | 450 | * force quit after adding this page |
91ad66ef | 451 | */ |
620a324b JA |
452 | this_len = min(this_len, plen - loff); |
453 | len = this_len; | |
5274f052 | 454 | } |
620a324b | 455 | |
eb20796b JA |
456 | partial[page_nr].offset = loff; |
457 | partial[page_nr].len = this_len; | |
82aa5d61 | 458 | len -= this_len; |
91ad66ef | 459 | loff = 0; |
eb20796b JA |
460 | spd.nr_pages++; |
461 | index++; | |
5274f052 JA |
462 | } |
463 | ||
eb20796b | 464 | /* |
475ecade | 465 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
466 | * we got, 'nr_pages' is how many pages are in the map. |
467 | */ | |
468 | while (page_nr < nr_pages) | |
469 | page_cache_release(pages[page_nr++]); | |
f4e6b498 | 470 | in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; |
eb20796b | 471 | |
912d35f8 | 472 | if (spd.nr_pages) |
00522fb4 | 473 | return splice_to_pipe(pipe, &spd); |
5274f052 | 474 | |
7480a904 | 475 | return error; |
5274f052 JA |
476 | } |
477 | ||
83f9135b JA |
478 | /** |
479 | * generic_file_splice_read - splice data from file to a pipe | |
480 | * @in: file to splice from | |
932cc6d4 | 481 | * @ppos: position in @in |
83f9135b JA |
482 | * @pipe: pipe to splice to |
483 | * @len: number of bytes to splice | |
484 | * @flags: splice modifier flags | |
485 | * | |
932cc6d4 JA |
486 | * Description: |
487 | * Will read pages from given file and fill them into a pipe. Can be | |
488 | * used as long as the address_space operations for the source implements | |
489 | * a readpage() hook. | |
490 | * | |
83f9135b | 491 | */ |
cbb7e577 JA |
492 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
493 | struct pipe_inode_info *pipe, size_t len, | |
494 | unsigned int flags) | |
5274f052 | 495 | { |
d366d398 | 496 | loff_t isize, left; |
8191ecd1 | 497 | int ret; |
d366d398 JA |
498 | |
499 | isize = i_size_read(in->f_mapping->host); | |
500 | if (unlikely(*ppos >= isize)) | |
501 | return 0; | |
502 | ||
503 | left = isize - *ppos; | |
504 | if (unlikely(left < len)) | |
505 | len = left; | |
5274f052 | 506 | |
8191ecd1 JA |
507 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
508 | if (ret > 0) | |
cbb7e577 | 509 | *ppos += ret; |
5274f052 JA |
510 | |
511 | return ret; | |
512 | } | |
513 | ||
059a8f37 JA |
514 | EXPORT_SYMBOL(generic_file_splice_read); |
515 | ||
5274f052 | 516 | /* |
4f6f0bd2 | 517 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 518 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 519 | */ |
76ad4d11 | 520 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
521 | struct pipe_buffer *buf, struct splice_desc *sd) |
522 | { | |
6a14b90b | 523 | struct file *file = sd->u.file; |
5274f052 | 524 | loff_t pos = sd->pos; |
f84d7519 | 525 | int ret, more; |
5274f052 | 526 | |
cac36bb0 | 527 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
528 | if (!ret) { |
529 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 530 | |
f84d7519 JA |
531 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
532 | sd->len, &pos, more); | |
533 | } | |
5274f052 | 534 | |
016b661e | 535 | return ret; |
5274f052 JA |
536 | } |
537 | ||
538 | /* | |
539 | * This is a little more tricky than the file -> pipe splicing. There are | |
540 | * basically three cases: | |
541 | * | |
542 | * - Destination page already exists in the address space and there | |
543 | * are users of it. For that case we have no other option that | |
544 | * copying the data. Tough luck. | |
545 | * - Destination page already exists in the address space, but there | |
546 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
547 | * through to last case. | |
548 | * - Destination page does not exist, we can add the pipe page to | |
549 | * the page cache and avoid the copy. | |
550 | * | |
83f9135b JA |
551 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
552 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
553 | * file address space page cache. This is possible if no one else has | |
554 | * the pipe page referenced outside of the pipe and page cache. If | |
555 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
556 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 557 | */ |
76ad4d11 | 558 | static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
5274f052 JA |
559 | struct splice_desc *sd) |
560 | { | |
6a14b90b | 561 | struct file *file = sd->u.file; |
5274f052 | 562 | struct address_space *mapping = file->f_mapping; |
016b661e | 563 | unsigned int offset, this_len; |
5274f052 | 564 | struct page *page; |
afddba49 | 565 | void *fsdata; |
3e7ee3e7 | 566 | int ret; |
5274f052 JA |
567 | |
568 | /* | |
49d0b21b | 569 | * make sure the data in this buffer is uptodate |
5274f052 | 570 | */ |
cac36bb0 | 571 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
572 | if (unlikely(ret)) |
573 | return ret; | |
5274f052 | 574 | |
5274f052 JA |
575 | offset = sd->pos & ~PAGE_CACHE_MASK; |
576 | ||
016b661e JA |
577 | this_len = sd->len; |
578 | if (this_len + offset > PAGE_CACHE_SIZE) | |
579 | this_len = PAGE_CACHE_SIZE - offset; | |
580 | ||
afddba49 NP |
581 | ret = pagecache_write_begin(file, mapping, sd->pos, this_len, |
582 | AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); | |
583 | if (unlikely(ret)) | |
584 | goto out; | |
5274f052 | 585 | |
0568b409 | 586 | if (buf->page != page) { |
f84d7519 JA |
587 | /* |
588 | * Careful, ->map() uses KM_USER0! | |
589 | */ | |
76ad4d11 | 590 | char *src = buf->ops->map(pipe, buf, 1); |
f84d7519 | 591 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 592 | |
016b661e | 593 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 594 | flush_dcache_page(page); |
f84d7519 | 595 | kunmap_atomic(dst, KM_USER1); |
76ad4d11 | 596 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 597 | } |
afddba49 NP |
598 | ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len, |
599 | page, fsdata); | |
5274f052 | 600 | out: |
5274f052 JA |
601 | return ret; |
602 | } | |
603 | ||
b3c2d2dd MS |
604 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
605 | { | |
606 | smp_mb(); | |
607 | if (waitqueue_active(&pipe->wait)) | |
608 | wake_up_interruptible(&pipe->wait); | |
609 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
610 | } | |
611 | ||
932cc6d4 | 612 | /** |
b3c2d2dd | 613 | * splice_from_pipe_feed - feed available data from a pipe to a file |
932cc6d4 JA |
614 | * @pipe: pipe to splice from |
615 | * @sd: information to @actor | |
616 | * @actor: handler that splices the data | |
617 | * | |
618 | * Description: | |
b3c2d2dd MS |
619 | |
620 | * This function loops over the pipe and calls @actor to do the | |
621 | * actual moving of a single struct pipe_buffer to the desired | |
622 | * destination. It returns when there's no more buffers left in | |
623 | * the pipe or if the requested number of bytes (@sd->total_len) | |
624 | * have been copied. It returns a positive number (one) if the | |
625 | * pipe needs to be filled with more data, zero if the required | |
626 | * number of bytes have been copied and -errno on error. | |
932cc6d4 | 627 | * |
b3c2d2dd MS |
628 | * This, together with splice_from_pipe_{begin,end,next}, may be |
629 | * used to implement the functionality of __splice_from_pipe() when | |
630 | * locking is required around copying the pipe buffers to the | |
631 | * destination. | |
83f9135b | 632 | */ |
b3c2d2dd MS |
633 | int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
634 | splice_actor *actor) | |
5274f052 | 635 | { |
b3c2d2dd | 636 | int ret; |
5274f052 | 637 | |
b3c2d2dd MS |
638 | while (pipe->nrbufs) { |
639 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
640 | const struct pipe_buf_operations *ops = buf->ops; | |
5274f052 | 641 | |
b3c2d2dd MS |
642 | sd->len = buf->len; |
643 | if (sd->len > sd->total_len) | |
644 | sd->len = sd->total_len; | |
5274f052 | 645 | |
b3c2d2dd MS |
646 | ret = actor(pipe, buf, sd); |
647 | if (ret <= 0) { | |
648 | if (ret == -ENODATA) | |
649 | ret = 0; | |
650 | return ret; | |
651 | } | |
652 | buf->offset += ret; | |
653 | buf->len -= ret; | |
654 | ||
655 | sd->num_spliced += ret; | |
656 | sd->len -= ret; | |
657 | sd->pos += ret; | |
658 | sd->total_len -= ret; | |
659 | ||
660 | if (!buf->len) { | |
661 | buf->ops = NULL; | |
662 | ops->release(pipe, buf); | |
663 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); | |
664 | pipe->nrbufs--; | |
665 | if (pipe->inode) | |
666 | sd->need_wakeup = true; | |
667 | } | |
5274f052 | 668 | |
b3c2d2dd MS |
669 | if (!sd->total_len) |
670 | return 0; | |
671 | } | |
5274f052 | 672 | |
b3c2d2dd MS |
673 | return 1; |
674 | } | |
675 | EXPORT_SYMBOL(splice_from_pipe_feed); | |
5274f052 | 676 | |
b3c2d2dd MS |
677 | /** |
678 | * splice_from_pipe_next - wait for some data to splice from | |
679 | * @pipe: pipe to splice from | |
680 | * @sd: information about the splice operation | |
681 | * | |
682 | * Description: | |
683 | * This function will wait for some data and return a positive | |
684 | * value (one) if pipe buffers are available. It will return zero | |
685 | * or -errno if no more data needs to be spliced. | |
686 | */ | |
687 | int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
688 | { | |
689 | while (!pipe->nrbufs) { | |
690 | if (!pipe->writers) | |
691 | return 0; | |
016b661e | 692 | |
b3c2d2dd MS |
693 | if (!pipe->waiting_writers && sd->num_spliced) |
694 | return 0; | |
73d62d83 | 695 | |
b3c2d2dd MS |
696 | if (sd->flags & SPLICE_F_NONBLOCK) |
697 | return -EAGAIN; | |
5274f052 | 698 | |
b3c2d2dd MS |
699 | if (signal_pending(current)) |
700 | return -ERESTARTSYS; | |
5274f052 | 701 | |
b3c2d2dd MS |
702 | if (sd->need_wakeup) { |
703 | wakeup_pipe_writers(pipe); | |
704 | sd->need_wakeup = false; | |
5274f052 JA |
705 | } |
706 | ||
b3c2d2dd MS |
707 | pipe_wait(pipe); |
708 | } | |
29e35094 | 709 | |
b3c2d2dd MS |
710 | return 1; |
711 | } | |
712 | EXPORT_SYMBOL(splice_from_pipe_next); | |
5274f052 | 713 | |
b3c2d2dd MS |
714 | /** |
715 | * splice_from_pipe_begin - start splicing from pipe | |
716 | * @pipe: pipe to splice from | |
717 | * | |
718 | * Description: | |
719 | * This function should be called before a loop containing | |
720 | * splice_from_pipe_next() and splice_from_pipe_feed() to | |
721 | * initialize the necessary fields of @sd. | |
722 | */ | |
723 | void splice_from_pipe_begin(struct splice_desc *sd) | |
724 | { | |
725 | sd->num_spliced = 0; | |
726 | sd->need_wakeup = false; | |
727 | } | |
728 | EXPORT_SYMBOL(splice_from_pipe_begin); | |
5274f052 | 729 | |
b3c2d2dd MS |
730 | /** |
731 | * splice_from_pipe_end - finish splicing from pipe | |
732 | * @pipe: pipe to splice from | |
733 | * @sd: information about the splice operation | |
734 | * | |
735 | * Description: | |
736 | * This function will wake up pipe writers if necessary. It should | |
737 | * be called after a loop containing splice_from_pipe_next() and | |
738 | * splice_from_pipe_feed(). | |
739 | */ | |
740 | void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
741 | { | |
742 | if (sd->need_wakeup) | |
743 | wakeup_pipe_writers(pipe); | |
744 | } | |
745 | EXPORT_SYMBOL(splice_from_pipe_end); | |
5274f052 | 746 | |
b3c2d2dd MS |
747 | /** |
748 | * __splice_from_pipe - splice data from a pipe to given actor | |
749 | * @pipe: pipe to splice from | |
750 | * @sd: information to @actor | |
751 | * @actor: handler that splices the data | |
752 | * | |
753 | * Description: | |
754 | * This function does little more than loop over the pipe and call | |
755 | * @actor to do the actual moving of a single struct pipe_buffer to | |
756 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
757 | * pipe_to_user. | |
758 | * | |
759 | */ | |
760 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, | |
761 | splice_actor *actor) | |
762 | { | |
763 | int ret; | |
5274f052 | 764 | |
b3c2d2dd MS |
765 | splice_from_pipe_begin(sd); |
766 | do { | |
767 | ret = splice_from_pipe_next(pipe, sd); | |
768 | if (ret > 0) | |
769 | ret = splice_from_pipe_feed(pipe, sd, actor); | |
770 | } while (ret > 0); | |
771 | splice_from_pipe_end(pipe, sd); | |
772 | ||
773 | return sd->num_spliced ? sd->num_spliced : ret; | |
5274f052 | 774 | } |
40bee44e | 775 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 776 | |
932cc6d4 JA |
777 | /** |
778 | * splice_from_pipe - splice data from a pipe to a file | |
779 | * @pipe: pipe to splice from | |
780 | * @out: file to splice to | |
781 | * @ppos: position in @out | |
782 | * @len: how many bytes to splice | |
783 | * @flags: splice modifier flags | |
784 | * @actor: handler that splices the data | |
785 | * | |
786 | * Description: | |
2933970b | 787 | * See __splice_from_pipe. This function locks the pipe inode, |
932cc6d4 JA |
788 | * otherwise it's identical to __splice_from_pipe(). |
789 | * | |
790 | */ | |
6da61809 MF |
791 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
792 | loff_t *ppos, size_t len, unsigned int flags, | |
793 | splice_actor *actor) | |
794 | { | |
795 | ssize_t ret; | |
c66ab6fa JA |
796 | struct splice_desc sd = { |
797 | .total_len = len, | |
798 | .flags = flags, | |
799 | .pos = *ppos, | |
6a14b90b | 800 | .u.file = out, |
c66ab6fa | 801 | }; |
6da61809 | 802 | |
7bfac9ec | 803 | if (pipe->inode) |
2933970b | 804 | mutex_lock(&pipe->inode->i_mutex); |
c66ab6fa | 805 | ret = __splice_from_pipe(pipe, &sd, actor); |
7bfac9ec MS |
806 | if (pipe->inode) |
807 | mutex_unlock(&pipe->inode->i_mutex); | |
6da61809 MF |
808 | |
809 | return ret; | |
810 | } | |
811 | ||
812 | /** | |
813 | * generic_file_splice_write_nolock - generic_file_splice_write without mutexes | |
814 | * @pipe: pipe info | |
815 | * @out: file to write to | |
932cc6d4 | 816 | * @ppos: position in @out |
6da61809 MF |
817 | * @len: number of bytes to splice |
818 | * @flags: splice modifier flags | |
819 | * | |
932cc6d4 JA |
820 | * Description: |
821 | * Will either move or copy pages (determined by @flags options) from | |
822 | * the given pipe inode to the given file. The caller is responsible | |
823 | * for acquiring i_mutex on both inodes. | |
6da61809 MF |
824 | * |
825 | */ | |
826 | ssize_t | |
827 | generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out, | |
828 | loff_t *ppos, size_t len, unsigned int flags) | |
829 | { | |
830 | struct address_space *mapping = out->f_mapping; | |
831 | struct inode *inode = mapping->host; | |
c66ab6fa JA |
832 | struct splice_desc sd = { |
833 | .total_len = len, | |
834 | .flags = flags, | |
835 | .pos = *ppos, | |
6a14b90b | 836 | .u.file = out, |
c66ab6fa | 837 | }; |
6da61809 MF |
838 | ssize_t ret; |
839 | int err; | |
840 | ||
2f1936b8 | 841 | err = file_remove_suid(out); |
8c34e2d6 JA |
842 | if (unlikely(err)) |
843 | return err; | |
844 | ||
c66ab6fa | 845 | ret = __splice_from_pipe(pipe, &sd, pipe_to_file); |
6da61809 | 846 | if (ret > 0) { |
17ee4f49 JA |
847 | unsigned long nr_pages; |
848 | ||
6da61809 | 849 | *ppos += ret; |
17ee4f49 | 850 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
6da61809 MF |
851 | |
852 | /* | |
853 | * If file or inode is SYNC and we actually wrote some data, | |
854 | * sync it. | |
855 | */ | |
856 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
857 | err = generic_osync_inode(inode, mapping, | |
858 | OSYNC_METADATA|OSYNC_DATA); | |
859 | ||
860 | if (err) | |
861 | ret = err; | |
862 | } | |
17ee4f49 | 863 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
6da61809 MF |
864 | } |
865 | ||
866 | return ret; | |
867 | } | |
868 | ||
869 | EXPORT_SYMBOL(generic_file_splice_write_nolock); | |
870 | ||
83f9135b JA |
871 | /** |
872 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 873 | * @pipe: pipe info |
83f9135b | 874 | * @out: file to write to |
932cc6d4 | 875 | * @ppos: position in @out |
83f9135b JA |
876 | * @len: number of bytes to splice |
877 | * @flags: splice modifier flags | |
878 | * | |
932cc6d4 JA |
879 | * Description: |
880 | * Will either move or copy pages (determined by @flags options) from | |
881 | * the given pipe inode to the given file. | |
83f9135b JA |
882 | * |
883 | */ | |
3a326a2c IM |
884 | ssize_t |
885 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 886 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 887 | { |
4f6f0bd2 | 888 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 889 | struct inode *inode = mapping->host; |
7f3d4ee1 MS |
890 | struct splice_desc sd = { |
891 | .total_len = len, | |
892 | .flags = flags, | |
893 | .pos = *ppos, | |
894 | .u.file = out, | |
895 | }; | |
3a326a2c IM |
896 | ssize_t ret; |
897 | ||
eb443e5a MS |
898 | if (pipe->inode) |
899 | mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT); | |
900 | ||
901 | splice_from_pipe_begin(&sd); | |
902 | do { | |
903 | ret = splice_from_pipe_next(pipe, &sd); | |
904 | if (ret <= 0) | |
905 | break; | |
906 | ||
907 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); | |
908 | ret = file_remove_suid(out); | |
909 | if (!ret) | |
910 | ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file); | |
911 | mutex_unlock(&inode->i_mutex); | |
912 | } while (ret > 0); | |
913 | splice_from_pipe_end(pipe, &sd); | |
914 | ||
915 | if (pipe->inode) | |
916 | mutex_unlock(&pipe->inode->i_mutex); | |
917 | ||
918 | if (sd.num_spliced) | |
919 | ret = sd.num_spliced; | |
920 | ||
a4514ebd | 921 | if (ret > 0) { |
17ee4f49 JA |
922 | unsigned long nr_pages; |
923 | ||
a4514ebd | 924 | *ppos += ret; |
17ee4f49 | 925 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
a4514ebd JA |
926 | |
927 | /* | |
928 | * If file or inode is SYNC and we actually wrote some data, | |
929 | * sync it. | |
930 | */ | |
931 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
7f3d4ee1 MS |
932 | int err; |
933 | ||
a4514ebd JA |
934 | mutex_lock(&inode->i_mutex); |
935 | err = generic_osync_inode(inode, mapping, | |
936 | OSYNC_METADATA|OSYNC_DATA); | |
937 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 938 | |
a4514ebd JA |
939 | if (err) |
940 | ret = err; | |
941 | } | |
17ee4f49 | 942 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
4f6f0bd2 JA |
943 | } |
944 | ||
945 | return ret; | |
5274f052 JA |
946 | } |
947 | ||
059a8f37 JA |
948 | EXPORT_SYMBOL(generic_file_splice_write); |
949 | ||
83f9135b JA |
950 | /** |
951 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 952 | * @pipe: pipe to splice from |
83f9135b | 953 | * @out: socket to write to |
932cc6d4 | 954 | * @ppos: position in @out |
83f9135b JA |
955 | * @len: number of bytes to splice |
956 | * @flags: splice modifier flags | |
957 | * | |
932cc6d4 JA |
958 | * Description: |
959 | * Will send @len bytes from the pipe to a network socket. No data copying | |
960 | * is involved. | |
83f9135b JA |
961 | * |
962 | */ | |
3a326a2c | 963 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 964 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 965 | { |
00522fb4 | 966 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
967 | } |
968 | ||
059a8f37 | 969 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 970 | |
83f9135b JA |
971 | /* |
972 | * Attempt to initiate a splice from pipe to file. | |
973 | */ | |
3a326a2c | 974 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 975 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 976 | { |
5274f052 JA |
977 | int ret; |
978 | ||
49570e9b | 979 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
980 | return -EINVAL; |
981 | ||
49570e9b | 982 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
983 | return -EBADF; |
984 | ||
efc968d4 LT |
985 | if (unlikely(out->f_flags & O_APPEND)) |
986 | return -EINVAL; | |
987 | ||
cbb7e577 | 988 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
989 | if (unlikely(ret < 0)) |
990 | return ret; | |
991 | ||
cbb7e577 | 992 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
993 | } |
994 | ||
83f9135b JA |
995 | /* |
996 | * Attempt to initiate a splice from a file to a pipe. | |
997 | */ | |
cbb7e577 JA |
998 | static long do_splice_to(struct file *in, loff_t *ppos, |
999 | struct pipe_inode_info *pipe, size_t len, | |
1000 | unsigned int flags) | |
5274f052 | 1001 | { |
5274f052 JA |
1002 | int ret; |
1003 | ||
49570e9b | 1004 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
1005 | return -EINVAL; |
1006 | ||
49570e9b | 1007 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
1008 | return -EBADF; |
1009 | ||
cbb7e577 | 1010 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
1011 | if (unlikely(ret < 0)) |
1012 | return ret; | |
1013 | ||
cbb7e577 | 1014 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
1015 | } |
1016 | ||
932cc6d4 JA |
1017 | /** |
1018 | * splice_direct_to_actor - splices data directly between two non-pipes | |
1019 | * @in: file to splice from | |
1020 | * @sd: actor information on where to splice to | |
1021 | * @actor: handles the data splicing | |
1022 | * | |
1023 | * Description: | |
1024 | * This is a special case helper to splice directly between two | |
1025 | * points, without requiring an explicit pipe. Internally an allocated | |
79685b8d | 1026 | * pipe is cached in the process, and reused during the lifetime of |
932cc6d4 JA |
1027 | * that process. |
1028 | * | |
c66ab6fa JA |
1029 | */ |
1030 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1031 | splice_direct_actor *actor) | |
b92ce558 JA |
1032 | { |
1033 | struct pipe_inode_info *pipe; | |
1034 | long ret, bytes; | |
1035 | umode_t i_mode; | |
c66ab6fa JA |
1036 | size_t len; |
1037 | int i, flags; | |
b92ce558 JA |
1038 | |
1039 | /* | |
1040 | * We require the input being a regular file, as we don't want to | |
1041 | * randomly drop data for eg socket -> socket splicing. Use the | |
1042 | * piped splicing for that! | |
1043 | */ | |
0f7fc9e4 | 1044 | i_mode = in->f_path.dentry->d_inode->i_mode; |
b92ce558 JA |
1045 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1046 | return -EINVAL; | |
1047 | ||
1048 | /* | |
1049 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1050 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1051 | */ | |
1052 | pipe = current->splice_pipe; | |
49570e9b | 1053 | if (unlikely(!pipe)) { |
b92ce558 JA |
1054 | pipe = alloc_pipe_info(NULL); |
1055 | if (!pipe) | |
1056 | return -ENOMEM; | |
1057 | ||
1058 | /* | |
1059 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1060 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1061 | * PIPE_READERS appropriately. |
1062 | */ | |
1063 | pipe->readers = 1; | |
1064 | ||
1065 | current->splice_pipe = pipe; | |
1066 | } | |
1067 | ||
1068 | /* | |
73d62d83 | 1069 | * Do the splice. |
b92ce558 JA |
1070 | */ |
1071 | ret = 0; | |
1072 | bytes = 0; | |
c66ab6fa JA |
1073 | len = sd->total_len; |
1074 | flags = sd->flags; | |
1075 | ||
1076 | /* | |
1077 | * Don't block on output, we have to drain the direct pipe. | |
1078 | */ | |
1079 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
b92ce558 JA |
1080 | |
1081 | while (len) { | |
51a92c0f | 1082 | size_t read_len; |
a82c53a0 | 1083 | loff_t pos = sd->pos, prev_pos = pos; |
b92ce558 | 1084 | |
bcd4f3ac | 1085 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1086 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1087 | goto out_release; |
1088 | ||
1089 | read_len = ret; | |
c66ab6fa | 1090 | sd->total_len = read_len; |
b92ce558 JA |
1091 | |
1092 | /* | |
1093 | * NOTE: nonblocking mode only applies to the input. We | |
1094 | * must not do the output in nonblocking mode as then we | |
1095 | * could get stuck data in the internal pipe: | |
1096 | */ | |
c66ab6fa | 1097 | ret = actor(pipe, sd); |
a82c53a0 TZ |
1098 | if (unlikely(ret <= 0)) { |
1099 | sd->pos = prev_pos; | |
b92ce558 | 1100 | goto out_release; |
a82c53a0 | 1101 | } |
b92ce558 JA |
1102 | |
1103 | bytes += ret; | |
1104 | len -= ret; | |
bcd4f3ac | 1105 | sd->pos = pos; |
b92ce558 | 1106 | |
a82c53a0 TZ |
1107 | if (ret < read_len) { |
1108 | sd->pos = prev_pos + ret; | |
51a92c0f | 1109 | goto out_release; |
a82c53a0 | 1110 | } |
b92ce558 JA |
1111 | } |
1112 | ||
9e97198d | 1113 | done: |
b92ce558 | 1114 | pipe->nrbufs = pipe->curbuf = 0; |
80848708 | 1115 | file_accessed(in); |
b92ce558 JA |
1116 | return bytes; |
1117 | ||
1118 | out_release: | |
1119 | /* | |
1120 | * If we did an incomplete transfer we must release | |
1121 | * the pipe buffers in question: | |
1122 | */ | |
1123 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1124 | struct pipe_buffer *buf = pipe->bufs + i; | |
1125 | ||
1126 | if (buf->ops) { | |
1127 | buf->ops->release(pipe, buf); | |
1128 | buf->ops = NULL; | |
1129 | } | |
1130 | } | |
b92ce558 | 1131 | |
9e97198d JA |
1132 | if (!bytes) |
1133 | bytes = ret; | |
c66ab6fa | 1134 | |
9e97198d | 1135 | goto done; |
c66ab6fa JA |
1136 | } |
1137 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1138 | ||
1139 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1140 | struct splice_desc *sd) | |
1141 | { | |
6a14b90b | 1142 | struct file *file = sd->u.file; |
c66ab6fa JA |
1143 | |
1144 | return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags); | |
1145 | } | |
1146 | ||
932cc6d4 JA |
1147 | /** |
1148 | * do_splice_direct - splices data directly between two files | |
1149 | * @in: file to splice from | |
1150 | * @ppos: input file offset | |
1151 | * @out: file to splice to | |
1152 | * @len: number of bytes to splice | |
1153 | * @flags: splice modifier flags | |
1154 | * | |
1155 | * Description: | |
1156 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1157 | * doing it in the application would incur an extra system call | |
1158 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1159 | * can splice directly through a process-private pipe. | |
1160 | * | |
1161 | */ | |
c66ab6fa JA |
1162 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
1163 | size_t len, unsigned int flags) | |
1164 | { | |
1165 | struct splice_desc sd = { | |
1166 | .len = len, | |
1167 | .total_len = len, | |
1168 | .flags = flags, | |
1169 | .pos = *ppos, | |
6a14b90b | 1170 | .u.file = out, |
c66ab6fa | 1171 | }; |
51a92c0f | 1172 | long ret; |
c66ab6fa JA |
1173 | |
1174 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); | |
51a92c0f | 1175 | if (ret > 0) |
a82c53a0 | 1176 | *ppos = sd.pos; |
51a92c0f | 1177 | |
c66ab6fa | 1178 | return ret; |
b92ce558 JA |
1179 | } |
1180 | ||
ddac0d39 JA |
1181 | /* |
1182 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1183 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1184 | * pipe. | |
1185 | */ | |
1186 | static inline struct pipe_inode_info *pipe_info(struct inode *inode) | |
1187 | { | |
1188 | if (S_ISFIFO(inode->i_mode)) | |
1189 | return inode->i_pipe; | |
1190 | ||
1191 | return NULL; | |
1192 | } | |
1193 | ||
83f9135b JA |
1194 | /* |
1195 | * Determine where to splice to/from. | |
1196 | */ | |
529565dc IM |
1197 | static long do_splice(struct file *in, loff_t __user *off_in, |
1198 | struct file *out, loff_t __user *off_out, | |
1199 | size_t len, unsigned int flags) | |
5274f052 | 1200 | { |
3a326a2c | 1201 | struct pipe_inode_info *pipe; |
cbb7e577 | 1202 | loff_t offset, *off; |
a4514ebd | 1203 | long ret; |
5274f052 | 1204 | |
0f7fc9e4 | 1205 | pipe = pipe_info(in->f_path.dentry->d_inode); |
529565dc IM |
1206 | if (pipe) { |
1207 | if (off_in) | |
1208 | return -ESPIPE; | |
b92ce558 JA |
1209 | if (off_out) { |
1210 | if (out->f_op->llseek == no_llseek) | |
1211 | return -EINVAL; | |
cbb7e577 | 1212 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1213 | return -EFAULT; |
cbb7e577 JA |
1214 | off = &offset; |
1215 | } else | |
1216 | off = &out->f_pos; | |
529565dc | 1217 | |
a4514ebd JA |
1218 | ret = do_splice_from(pipe, out, off, len, flags); |
1219 | ||
1220 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1221 | ret = -EFAULT; | |
1222 | ||
1223 | return ret; | |
529565dc | 1224 | } |
5274f052 | 1225 | |
0f7fc9e4 | 1226 | pipe = pipe_info(out->f_path.dentry->d_inode); |
529565dc IM |
1227 | if (pipe) { |
1228 | if (off_out) | |
1229 | return -ESPIPE; | |
b92ce558 JA |
1230 | if (off_in) { |
1231 | if (in->f_op->llseek == no_llseek) | |
1232 | return -EINVAL; | |
cbb7e577 | 1233 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1234 | return -EFAULT; |
cbb7e577 JA |
1235 | off = &offset; |
1236 | } else | |
1237 | off = &in->f_pos; | |
529565dc | 1238 | |
a4514ebd JA |
1239 | ret = do_splice_to(in, off, pipe, len, flags); |
1240 | ||
1241 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1242 | ret = -EFAULT; | |
1243 | ||
1244 | return ret; | |
529565dc | 1245 | } |
5274f052 JA |
1246 | |
1247 | return -EINVAL; | |
1248 | } | |
1249 | ||
912d35f8 JA |
1250 | /* |
1251 | * Map an iov into an array of pages and offset/length tupples. With the | |
1252 | * partial_page structure, we can map several non-contiguous ranges into | |
1253 | * our ones pages[] map instead of splitting that operation into pieces. | |
1254 | * Could easily be exported as a generic helper for other users, in which | |
1255 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1256 | */ | |
1257 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1258 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1259 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1260 | { |
1261 | int buffers = 0, error = 0; | |
1262 | ||
912d35f8 JA |
1263 | while (nr_vecs) { |
1264 | unsigned long off, npages; | |
75723957 | 1265 | struct iovec entry; |
912d35f8 JA |
1266 | void __user *base; |
1267 | size_t len; | |
1268 | int i; | |
1269 | ||
75723957 | 1270 | error = -EFAULT; |
bc40d73c | 1271 | if (copy_from_user(&entry, iov, sizeof(entry))) |
912d35f8 JA |
1272 | break; |
1273 | ||
75723957 LT |
1274 | base = entry.iov_base; |
1275 | len = entry.iov_len; | |
1276 | ||
912d35f8 JA |
1277 | /* |
1278 | * Sanity check this iovec. 0 read succeeds. | |
1279 | */ | |
75723957 | 1280 | error = 0; |
912d35f8 JA |
1281 | if (unlikely(!len)) |
1282 | break; | |
1283 | error = -EFAULT; | |
712a30e6 | 1284 | if (!access_ok(VERIFY_READ, base, len)) |
912d35f8 JA |
1285 | break; |
1286 | ||
1287 | /* | |
1288 | * Get this base offset and number of pages, then map | |
1289 | * in the user pages. | |
1290 | */ | |
1291 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1292 | |
1293 | /* | |
1294 | * If asked for alignment, the offset must be zero and the | |
1295 | * length a multiple of the PAGE_SIZE. | |
1296 | */ | |
1297 | error = -EINVAL; | |
1298 | if (aligned && (off || len & ~PAGE_MASK)) | |
1299 | break; | |
1300 | ||
912d35f8 JA |
1301 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1302 | if (npages > PIPE_BUFFERS - buffers) | |
1303 | npages = PIPE_BUFFERS - buffers; | |
1304 | ||
bc40d73c NP |
1305 | error = get_user_pages_fast((unsigned long)base, npages, |
1306 | 0, &pages[buffers]); | |
912d35f8 JA |
1307 | |
1308 | if (unlikely(error <= 0)) | |
1309 | break; | |
1310 | ||
1311 | /* | |
1312 | * Fill this contiguous range into the partial page map. | |
1313 | */ | |
1314 | for (i = 0; i < error; i++) { | |
7591489a | 1315 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1316 | |
1317 | partial[buffers].offset = off; | |
1318 | partial[buffers].len = plen; | |
1319 | ||
1320 | off = 0; | |
1321 | len -= plen; | |
1322 | buffers++; | |
1323 | } | |
1324 | ||
1325 | /* | |
1326 | * We didn't complete this iov, stop here since it probably | |
1327 | * means we have to move some of this into a pipe to | |
1328 | * be able to continue. | |
1329 | */ | |
1330 | if (len) | |
1331 | break; | |
1332 | ||
1333 | /* | |
1334 | * Don't continue if we mapped fewer pages than we asked for, | |
1335 | * or if we mapped the max number of pages that we have | |
1336 | * room for. | |
1337 | */ | |
1338 | if (error < npages || buffers == PIPE_BUFFERS) | |
1339 | break; | |
1340 | ||
1341 | nr_vecs--; | |
1342 | iov++; | |
1343 | } | |
1344 | ||
912d35f8 JA |
1345 | if (buffers) |
1346 | return buffers; | |
1347 | ||
1348 | return error; | |
1349 | } | |
1350 | ||
6a14b90b JA |
1351 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1352 | struct splice_desc *sd) | |
1353 | { | |
1354 | char *src; | |
1355 | int ret; | |
1356 | ||
cac36bb0 | 1357 | ret = buf->ops->confirm(pipe, buf); |
6a14b90b JA |
1358 | if (unlikely(ret)) |
1359 | return ret; | |
1360 | ||
1361 | /* | |
1362 | * See if we can use the atomic maps, by prefaulting in the | |
1363 | * pages and doing an atomic copy | |
1364 | */ | |
1365 | if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { | |
1366 | src = buf->ops->map(pipe, buf, 1); | |
1367 | ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, | |
1368 | sd->len); | |
1369 | buf->ops->unmap(pipe, buf, src); | |
1370 | if (!ret) { | |
1371 | ret = sd->len; | |
1372 | goto out; | |
1373 | } | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * No dice, use slow non-atomic map and copy | |
1378 | */ | |
1379 | src = buf->ops->map(pipe, buf, 0); | |
1380 | ||
1381 | ret = sd->len; | |
1382 | if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) | |
1383 | ret = -EFAULT; | |
1384 | ||
6866bef4 | 1385 | buf->ops->unmap(pipe, buf, src); |
6a14b90b JA |
1386 | out: |
1387 | if (ret > 0) | |
1388 | sd->u.userptr += ret; | |
6a14b90b JA |
1389 | return ret; |
1390 | } | |
1391 | ||
1392 | /* | |
1393 | * For lack of a better implementation, implement vmsplice() to userspace | |
1394 | * as a simple copy of the pipes pages to the user iov. | |
1395 | */ | |
1396 | static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, | |
1397 | unsigned long nr_segs, unsigned int flags) | |
1398 | { | |
1399 | struct pipe_inode_info *pipe; | |
1400 | struct splice_desc sd; | |
1401 | ssize_t size; | |
1402 | int error; | |
1403 | long ret; | |
1404 | ||
1405 | pipe = pipe_info(file->f_path.dentry->d_inode); | |
1406 | if (!pipe) | |
1407 | return -EBADF; | |
1408 | ||
1409 | if (pipe->inode) | |
1410 | mutex_lock(&pipe->inode->i_mutex); | |
1411 | ||
1412 | error = ret = 0; | |
1413 | while (nr_segs) { | |
1414 | void __user *base; | |
1415 | size_t len; | |
1416 | ||
1417 | /* | |
1418 | * Get user address base and length for this iovec. | |
1419 | */ | |
1420 | error = get_user(base, &iov->iov_base); | |
1421 | if (unlikely(error)) | |
1422 | break; | |
1423 | error = get_user(len, &iov->iov_len); | |
1424 | if (unlikely(error)) | |
1425 | break; | |
1426 | ||
1427 | /* | |
1428 | * Sanity check this iovec. 0 read succeeds. | |
1429 | */ | |
1430 | if (unlikely(!len)) | |
1431 | break; | |
1432 | if (unlikely(!base)) { | |
1433 | error = -EFAULT; | |
1434 | break; | |
1435 | } | |
1436 | ||
8811930d JA |
1437 | if (unlikely(!access_ok(VERIFY_WRITE, base, len))) { |
1438 | error = -EFAULT; | |
1439 | break; | |
1440 | } | |
1441 | ||
6a14b90b JA |
1442 | sd.len = 0; |
1443 | sd.total_len = len; | |
1444 | sd.flags = flags; | |
1445 | sd.u.userptr = base; | |
1446 | sd.pos = 0; | |
1447 | ||
1448 | size = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1449 | if (size < 0) { | |
1450 | if (!ret) | |
1451 | ret = size; | |
1452 | ||
1453 | break; | |
1454 | } | |
1455 | ||
1456 | ret += size; | |
1457 | ||
1458 | if (size < len) | |
1459 | break; | |
1460 | ||
1461 | nr_segs--; | |
1462 | iov++; | |
1463 | } | |
1464 | ||
1465 | if (pipe->inode) | |
1466 | mutex_unlock(&pipe->inode->i_mutex); | |
1467 | ||
1468 | if (!ret) | |
1469 | ret = error; | |
1470 | ||
1471 | return ret; | |
1472 | } | |
1473 | ||
912d35f8 JA |
1474 | /* |
1475 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1476 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1477 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1478 | */ |
6a14b90b JA |
1479 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1480 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1481 | { |
ddac0d39 | 1482 | struct pipe_inode_info *pipe; |
912d35f8 JA |
1483 | struct page *pages[PIPE_BUFFERS]; |
1484 | struct partial_page partial[PIPE_BUFFERS]; | |
1485 | struct splice_pipe_desc spd = { | |
1486 | .pages = pages, | |
1487 | .partial = partial, | |
1488 | .flags = flags, | |
1489 | .ops = &user_page_pipe_buf_ops, | |
bbdfc2f7 | 1490 | .spd_release = spd_release_page, |
912d35f8 JA |
1491 | }; |
1492 | ||
0f7fc9e4 | 1493 | pipe = pipe_info(file->f_path.dentry->d_inode); |
ddac0d39 | 1494 | if (!pipe) |
912d35f8 | 1495 | return -EBADF; |
912d35f8 | 1496 | |
7afa6fd0 JA |
1497 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1498 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1499 | if (spd.nr_pages <= 0) |
1500 | return spd.nr_pages; | |
1501 | ||
00522fb4 | 1502 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1503 | } |
1504 | ||
6a14b90b JA |
1505 | /* |
1506 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1507 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1508 | * operation that can be supported without any funky alignment restrictions | |
1509 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1510 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1511 | * solutions for that: | |
1512 | * | |
1513 | * - memcpy() the data internally, at which point we might as well just | |
1514 | * do a regular read() on the buffer anyway. | |
1515 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1516 | * has restriction limitations on both ends of the pipe). | |
1517 | * | |
1518 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1519 | * | |
1520 | */ | |
836f92ad HC |
1521 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, |
1522 | unsigned long, nr_segs, unsigned int, flags) | |
912d35f8 JA |
1523 | { |
1524 | struct file *file; | |
1525 | long error; | |
1526 | int fput; | |
1527 | ||
6a14b90b JA |
1528 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1529 | return -EINVAL; | |
1530 | else if (unlikely(!nr_segs)) | |
1531 | return 0; | |
1532 | ||
912d35f8 JA |
1533 | error = -EBADF; |
1534 | file = fget_light(fd, &fput); | |
1535 | if (file) { | |
1536 | if (file->f_mode & FMODE_WRITE) | |
6a14b90b JA |
1537 | error = vmsplice_to_pipe(file, iov, nr_segs, flags); |
1538 | else if (file->f_mode & FMODE_READ) | |
1539 | error = vmsplice_to_user(file, iov, nr_segs, flags); | |
912d35f8 JA |
1540 | |
1541 | fput_light(file, fput); | |
1542 | } | |
1543 | ||
1544 | return error; | |
1545 | } | |
1546 | ||
836f92ad HC |
1547 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
1548 | int, fd_out, loff_t __user *, off_out, | |
1549 | size_t, len, unsigned int, flags) | |
5274f052 JA |
1550 | { |
1551 | long error; | |
1552 | struct file *in, *out; | |
1553 | int fput_in, fput_out; | |
1554 | ||
1555 | if (unlikely(!len)) | |
1556 | return 0; | |
1557 | ||
1558 | error = -EBADF; | |
529565dc | 1559 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1560 | if (in) { |
1561 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1562 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1563 | if (out) { |
1564 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1565 | error = do_splice(in, off_in, |
1566 | out, off_out, | |
1567 | len, flags); | |
5274f052 JA |
1568 | fput_light(out, fput_out); |
1569 | } | |
1570 | } | |
1571 | ||
1572 | fput_light(in, fput_in); | |
1573 | } | |
1574 | ||
1575 | return error; | |
1576 | } | |
70524490 | 1577 | |
aadd06e5 JA |
1578 | /* |
1579 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1580 | * return an appropriate error. | |
1581 | */ | |
1582 | static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1583 | { | |
1584 | int ret; | |
1585 | ||
1586 | /* | |
1587 | * Check ->nrbufs without the inode lock first. This function | |
1588 | * is speculative anyways, so missing one is ok. | |
1589 | */ | |
1590 | if (pipe->nrbufs) | |
1591 | return 0; | |
1592 | ||
1593 | ret = 0; | |
1594 | mutex_lock(&pipe->inode->i_mutex); | |
1595 | ||
1596 | while (!pipe->nrbufs) { | |
1597 | if (signal_pending(current)) { | |
1598 | ret = -ERESTARTSYS; | |
1599 | break; | |
1600 | } | |
1601 | if (!pipe->writers) | |
1602 | break; | |
1603 | if (!pipe->waiting_writers) { | |
1604 | if (flags & SPLICE_F_NONBLOCK) { | |
1605 | ret = -EAGAIN; | |
1606 | break; | |
1607 | } | |
1608 | } | |
1609 | pipe_wait(pipe); | |
1610 | } | |
1611 | ||
1612 | mutex_unlock(&pipe->inode->i_mutex); | |
1613 | return ret; | |
1614 | } | |
1615 | ||
1616 | /* | |
1617 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1618 | * return an appropriate error. | |
1619 | */ | |
1620 | static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1621 | { | |
1622 | int ret; | |
1623 | ||
1624 | /* | |
1625 | * Check ->nrbufs without the inode lock first. This function | |
1626 | * is speculative anyways, so missing one is ok. | |
1627 | */ | |
1628 | if (pipe->nrbufs < PIPE_BUFFERS) | |
1629 | return 0; | |
1630 | ||
1631 | ret = 0; | |
1632 | mutex_lock(&pipe->inode->i_mutex); | |
1633 | ||
1634 | while (pipe->nrbufs >= PIPE_BUFFERS) { | |
1635 | if (!pipe->readers) { | |
1636 | send_sig(SIGPIPE, current, 0); | |
1637 | ret = -EPIPE; | |
1638 | break; | |
1639 | } | |
1640 | if (flags & SPLICE_F_NONBLOCK) { | |
1641 | ret = -EAGAIN; | |
1642 | break; | |
1643 | } | |
1644 | if (signal_pending(current)) { | |
1645 | ret = -ERESTARTSYS; | |
1646 | break; | |
1647 | } | |
1648 | pipe->waiting_writers++; | |
1649 | pipe_wait(pipe); | |
1650 | pipe->waiting_writers--; | |
1651 | } | |
1652 | ||
1653 | mutex_unlock(&pipe->inode->i_mutex); | |
1654 | return ret; | |
1655 | } | |
1656 | ||
70524490 JA |
1657 | /* |
1658 | * Link contents of ipipe to opipe. | |
1659 | */ | |
1660 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1661 | struct pipe_inode_info *opipe, | |
1662 | size_t len, unsigned int flags) | |
1663 | { | |
1664 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1665 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1666 | |
1667 | /* | |
1668 | * Potential ABBA deadlock, work around it by ordering lock | |
1669 | * grabbing by inode address. Otherwise two different processes | |
1670 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1671 | */ | |
62752ee1 | 1672 | inode_double_lock(ipipe->inode, opipe->inode); |
70524490 | 1673 | |
aadd06e5 | 1674 | do { |
70524490 JA |
1675 | if (!opipe->readers) { |
1676 | send_sig(SIGPIPE, current, 0); | |
1677 | if (!ret) | |
1678 | ret = -EPIPE; | |
1679 | break; | |
1680 | } | |
70524490 | 1681 | |
aadd06e5 JA |
1682 | /* |
1683 | * If we have iterated all input buffers or ran out of | |
1684 | * output room, break. | |
1685 | */ | |
1686 | if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) | |
1687 | break; | |
70524490 | 1688 | |
aadd06e5 JA |
1689 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); |
1690 | nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
70524490 JA |
1691 | |
1692 | /* | |
aadd06e5 JA |
1693 | * Get a reference to this pipe buffer, |
1694 | * so we can copy the contents over. | |
70524490 | 1695 | */ |
aadd06e5 JA |
1696 | ibuf->ops->get(ipipe, ibuf); |
1697 | ||
1698 | obuf = opipe->bufs + nbuf; | |
1699 | *obuf = *ibuf; | |
1700 | ||
2a27250e | 1701 | /* |
aadd06e5 JA |
1702 | * Don't inherit the gift flag, we need to |
1703 | * prevent multiple steals of this page. | |
2a27250e | 1704 | */ |
aadd06e5 | 1705 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1706 | |
aadd06e5 JA |
1707 | if (obuf->len > len) |
1708 | obuf->len = len; | |
70524490 | 1709 | |
aadd06e5 JA |
1710 | opipe->nrbufs++; |
1711 | ret += obuf->len; | |
1712 | len -= obuf->len; | |
1713 | i++; | |
1714 | } while (len); | |
70524490 | 1715 | |
02cf01ae JA |
1716 | /* |
1717 | * return EAGAIN if we have the potential of some data in the | |
1718 | * future, otherwise just return 0 | |
1719 | */ | |
1720 | if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) | |
1721 | ret = -EAGAIN; | |
1722 | ||
62752ee1 | 1723 | inode_double_unlock(ipipe->inode, opipe->inode); |
70524490 | 1724 | |
aadd06e5 JA |
1725 | /* |
1726 | * If we put data in the output pipe, wakeup any potential readers. | |
1727 | */ | |
1728 | if (ret > 0) { | |
70524490 JA |
1729 | smp_mb(); |
1730 | if (waitqueue_active(&opipe->wait)) | |
1731 | wake_up_interruptible(&opipe->wait); | |
1732 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1733 | } | |
1734 | ||
1735 | return ret; | |
1736 | } | |
1737 | ||
1738 | /* | |
1739 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1740 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1741 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1742 | * applicable one is SPLICE_F_NONBLOCK. | |
1743 | */ | |
1744 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1745 | unsigned int flags) | |
1746 | { | |
0f7fc9e4 JJS |
1747 | struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); |
1748 | struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); | |
aadd06e5 | 1749 | int ret = -EINVAL; |
70524490 JA |
1750 | |
1751 | /* | |
aadd06e5 JA |
1752 | * Duplicate the contents of ipipe to opipe without actually |
1753 | * copying the data. | |
70524490 | 1754 | */ |
aadd06e5 JA |
1755 | if (ipipe && opipe && ipipe != opipe) { |
1756 | /* | |
1757 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1758 | * ordering doesn't really matter. | |
1759 | */ | |
1760 | ret = link_ipipe_prep(ipipe, flags); | |
1761 | if (!ret) { | |
1762 | ret = link_opipe_prep(opipe, flags); | |
02cf01ae | 1763 | if (!ret) |
aadd06e5 | 1764 | ret = link_pipe(ipipe, opipe, len, flags); |
aadd06e5 JA |
1765 | } |
1766 | } | |
70524490 | 1767 | |
aadd06e5 | 1768 | return ret; |
70524490 JA |
1769 | } |
1770 | ||
836f92ad | 1771 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
70524490 JA |
1772 | { |
1773 | struct file *in; | |
1774 | int error, fput_in; | |
1775 | ||
1776 | if (unlikely(!len)) | |
1777 | return 0; | |
1778 | ||
1779 | error = -EBADF; | |
1780 | in = fget_light(fdin, &fput_in); | |
1781 | if (in) { | |
1782 | if (in->f_mode & FMODE_READ) { | |
1783 | int fput_out; | |
1784 | struct file *out = fget_light(fdout, &fput_out); | |
1785 | ||
1786 | if (out) { | |
1787 | if (out->f_mode & FMODE_WRITE) | |
1788 | error = do_tee(in, out, len, flags); | |
1789 | fput_light(out, fput_out); | |
1790 | } | |
1791 | } | |
1792 | fput_light(in, fput_in); | |
1793 | } | |
1794 | ||
1795 | return error; | |
1796 | } |