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 | |
61e0d47c | 185 | pipe_lock(pipe); |
5274f052 | 186 | |
5274f052 | 187 | for (;;) { |
3a326a2c | 188 | if (!pipe->readers) { |
5274f052 JA |
189 | send_sig(SIGPIPE, current, 0); |
190 | if (!ret) | |
191 | ret = -EPIPE; | |
192 | break; | |
193 | } | |
194 | ||
6f767b04 JA |
195 | if (pipe->nrbufs < PIPE_BUFFERS) { |
196 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 197 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 198 | |
912d35f8 JA |
199 | buf->page = spd->pages[page_nr]; |
200 | buf->offset = spd->partial[page_nr].offset; | |
201 | buf->len = spd->partial[page_nr].len; | |
497f9625 | 202 | buf->private = spd->partial[page_nr].private; |
912d35f8 | 203 | buf->ops = spd->ops; |
7afa6fd0 JA |
204 | if (spd->flags & SPLICE_F_GIFT) |
205 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
206 | ||
6f767b04 | 207 | pipe->nrbufs++; |
912d35f8 JA |
208 | page_nr++; |
209 | ret += buf->len; | |
210 | ||
6f767b04 JA |
211 | if (pipe->inode) |
212 | do_wakeup = 1; | |
5274f052 | 213 | |
912d35f8 | 214 | if (!--spd->nr_pages) |
5274f052 | 215 | break; |
6f767b04 | 216 | if (pipe->nrbufs < PIPE_BUFFERS) |
5274f052 JA |
217 | continue; |
218 | ||
219 | break; | |
220 | } | |
221 | ||
912d35f8 | 222 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
223 | if (!ret) |
224 | ret = -EAGAIN; | |
225 | break; | |
226 | } | |
227 | ||
5274f052 JA |
228 | if (signal_pending(current)) { |
229 | if (!ret) | |
230 | ret = -ERESTARTSYS; | |
231 | break; | |
232 | } | |
233 | ||
234 | if (do_wakeup) { | |
c0bd1f65 | 235 | smp_mb(); |
3a326a2c IM |
236 | if (waitqueue_active(&pipe->wait)) |
237 | wake_up_interruptible_sync(&pipe->wait); | |
238 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
239 | do_wakeup = 0; |
240 | } | |
241 | ||
3a326a2c IM |
242 | pipe->waiting_writers++; |
243 | pipe_wait(pipe); | |
244 | pipe->waiting_writers--; | |
5274f052 JA |
245 | } |
246 | ||
61e0d47c | 247 | pipe_unlock(pipe); |
5274f052 | 248 | |
61e0d47c MS |
249 | if (do_wakeup) { |
250 | smp_mb(); | |
251 | if (waitqueue_active(&pipe->wait)) | |
252 | wake_up_interruptible(&pipe->wait); | |
253 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
254 | } |
255 | ||
00de00bd | 256 | while (page_nr < spd_pages) |
bbdfc2f7 | 257 | spd->spd_release(spd, page_nr++); |
5274f052 JA |
258 | |
259 | return ret; | |
260 | } | |
261 | ||
bbdfc2f7 JA |
262 | static void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) |
263 | { | |
264 | page_cache_release(spd->pages[i]); | |
265 | } | |
266 | ||
3a326a2c | 267 | static int |
cbb7e577 JA |
268 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
269 | struct pipe_inode_info *pipe, size_t len, | |
270 | unsigned int flags) | |
5274f052 JA |
271 | { |
272 | struct address_space *mapping = in->f_mapping; | |
d8983910 | 273 | unsigned int loff, nr_pages, req_pages; |
16c523dd | 274 | struct page *pages[PIPE_BUFFERS]; |
912d35f8 | 275 | struct partial_page partial[PIPE_BUFFERS]; |
5274f052 | 276 | struct page *page; |
91ad66ef JA |
277 | pgoff_t index, end_index; |
278 | loff_t isize; | |
eb20796b | 279 | int error, page_nr; |
912d35f8 JA |
280 | struct splice_pipe_desc spd = { |
281 | .pages = pages, | |
282 | .partial = partial, | |
283 | .flags = flags, | |
284 | .ops = &page_cache_pipe_buf_ops, | |
bbdfc2f7 | 285 | .spd_release = spd_release_page, |
912d35f8 | 286 | }; |
5274f052 | 287 | |
cbb7e577 | 288 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 | 289 | loff = *ppos & ~PAGE_CACHE_MASK; |
d8983910 FW |
290 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
291 | nr_pages = min(req_pages, (unsigned)PIPE_BUFFERS); | |
5274f052 | 292 | |
eb20796b JA |
293 | /* |
294 | * Lookup the (hopefully) full range of pages we need. | |
295 | */ | |
296 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
431a4820 | 297 | index += spd.nr_pages; |
82aa5d61 | 298 | |
eb20796b JA |
299 | /* |
300 | * If find_get_pages_contig() returned fewer pages than we needed, | |
431a4820 | 301 | * readahead/allocate the rest and fill in the holes. |
eb20796b | 302 | */ |
431a4820 | 303 | if (spd.nr_pages < nr_pages) |
cf914a7d RR |
304 | page_cache_sync_readahead(mapping, &in->f_ra, in, |
305 | index, req_pages - spd.nr_pages); | |
431a4820 | 306 | |
932cc6d4 | 307 | error = 0; |
eb20796b | 308 | while (spd.nr_pages < nr_pages) { |
82aa5d61 | 309 | /* |
eb20796b JA |
310 | * Page could be there, find_get_pages_contig() breaks on |
311 | * the first hole. | |
5274f052 | 312 | */ |
7480a904 JA |
313 | page = find_get_page(mapping, index); |
314 | if (!page) { | |
7480a904 | 315 | /* |
eb20796b | 316 | * page didn't exist, allocate one. |
7480a904 JA |
317 | */ |
318 | page = page_cache_alloc_cold(mapping); | |
319 | if (!page) | |
320 | break; | |
321 | ||
322 | error = add_to_page_cache_lru(page, mapping, index, | |
4cd13504 | 323 | mapping_gfp_mask(mapping)); |
7480a904 JA |
324 | if (unlikely(error)) { |
325 | page_cache_release(page); | |
a0548871 JA |
326 | if (error == -EEXIST) |
327 | continue; | |
7480a904 JA |
328 | break; |
329 | } | |
eb20796b JA |
330 | /* |
331 | * add_to_page_cache() locks the page, unlock it | |
332 | * to avoid convoluting the logic below even more. | |
333 | */ | |
334 | unlock_page(page); | |
7480a904 JA |
335 | } |
336 | ||
eb20796b JA |
337 | pages[spd.nr_pages++] = page; |
338 | index++; | |
339 | } | |
340 | ||
341 | /* | |
342 | * Now loop over the map and see if we need to start IO on any | |
343 | * pages, fill in the partial map, etc. | |
344 | */ | |
345 | index = *ppos >> PAGE_CACHE_SHIFT; | |
346 | nr_pages = spd.nr_pages; | |
347 | spd.nr_pages = 0; | |
348 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
349 | unsigned int this_len; | |
350 | ||
351 | if (!len) | |
352 | break; | |
353 | ||
354 | /* | |
355 | * this_len is the max we'll use from this page | |
356 | */ | |
357 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
358 | page = pages[page_nr]; | |
359 | ||
a08a166f | 360 | if (PageReadahead(page)) |
cf914a7d | 361 | page_cache_async_readahead(mapping, &in->f_ra, in, |
d8983910 | 362 | page, index, req_pages - page_nr); |
a08a166f | 363 | |
7480a904 JA |
364 | /* |
365 | * If the page isn't uptodate, we may need to start io on it | |
366 | */ | |
367 | if (!PageUptodate(page)) { | |
c4f895cb JA |
368 | /* |
369 | * If in nonblock mode then dont block on waiting | |
370 | * for an in-flight io page | |
371 | */ | |
9ae9d68c | 372 | if (flags & SPLICE_F_NONBLOCK) { |
529ae9aa | 373 | if (!trylock_page(page)) { |
8191ecd1 | 374 | error = -EAGAIN; |
9ae9d68c | 375 | break; |
8191ecd1 | 376 | } |
9ae9d68c FW |
377 | } else |
378 | lock_page(page); | |
7480a904 JA |
379 | |
380 | /* | |
32502b84 MS |
381 | * Page was truncated, or invalidated by the |
382 | * filesystem. Redo the find/create, but this time the | |
383 | * page is kept locked, so there's no chance of another | |
384 | * race with truncate/invalidate. | |
7480a904 JA |
385 | */ |
386 | if (!page->mapping) { | |
387 | unlock_page(page); | |
32502b84 MS |
388 | page = find_or_create_page(mapping, index, |
389 | mapping_gfp_mask(mapping)); | |
390 | ||
391 | if (!page) { | |
392 | error = -ENOMEM; | |
393 | break; | |
394 | } | |
395 | page_cache_release(pages[page_nr]); | |
396 | pages[page_nr] = page; | |
7480a904 JA |
397 | } |
398 | /* | |
399 | * page was already under io and is now done, great | |
400 | */ | |
401 | if (PageUptodate(page)) { | |
402 | unlock_page(page); | |
403 | goto fill_it; | |
404 | } | |
5274f052 | 405 | |
7480a904 JA |
406 | /* |
407 | * need to read in the page | |
408 | */ | |
409 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 410 | if (unlikely(error)) { |
eb20796b JA |
411 | /* |
412 | * We really should re-lookup the page here, | |
413 | * but it complicates things a lot. Instead | |
414 | * lets just do what we already stored, and | |
415 | * we'll get it the next time we are called. | |
416 | */ | |
7480a904 | 417 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
418 | error = 0; |
419 | ||
5274f052 JA |
420 | break; |
421 | } | |
620a324b JA |
422 | } |
423 | fill_it: | |
424 | /* | |
425 | * i_size must be checked after PageUptodate. | |
426 | */ | |
427 | isize = i_size_read(mapping->host); | |
428 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
429 | if (unlikely(!isize || index > end_index)) | |
430 | break; | |
431 | ||
432 | /* | |
433 | * if this is the last page, see if we need to shrink | |
434 | * the length and stop | |
435 | */ | |
436 | if (end_index == index) { | |
437 | unsigned int plen; | |
91ad66ef JA |
438 | |
439 | /* | |
620a324b | 440 | * max good bytes in this page |
91ad66ef | 441 | */ |
620a324b JA |
442 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
443 | if (plen <= loff) | |
91ad66ef | 444 | break; |
91ad66ef JA |
445 | |
446 | /* | |
620a324b | 447 | * force quit after adding this page |
91ad66ef | 448 | */ |
620a324b JA |
449 | this_len = min(this_len, plen - loff); |
450 | len = this_len; | |
5274f052 | 451 | } |
620a324b | 452 | |
eb20796b JA |
453 | partial[page_nr].offset = loff; |
454 | partial[page_nr].len = this_len; | |
82aa5d61 | 455 | len -= this_len; |
91ad66ef | 456 | loff = 0; |
eb20796b JA |
457 | spd.nr_pages++; |
458 | index++; | |
5274f052 JA |
459 | } |
460 | ||
eb20796b | 461 | /* |
475ecade | 462 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
463 | * we got, 'nr_pages' is how many pages are in the map. |
464 | */ | |
465 | while (page_nr < nr_pages) | |
466 | page_cache_release(pages[page_nr++]); | |
f4e6b498 | 467 | in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; |
eb20796b | 468 | |
912d35f8 | 469 | if (spd.nr_pages) |
00522fb4 | 470 | return splice_to_pipe(pipe, &spd); |
5274f052 | 471 | |
7480a904 | 472 | return error; |
5274f052 JA |
473 | } |
474 | ||
83f9135b JA |
475 | /** |
476 | * generic_file_splice_read - splice data from file to a pipe | |
477 | * @in: file to splice from | |
932cc6d4 | 478 | * @ppos: position in @in |
83f9135b JA |
479 | * @pipe: pipe to splice to |
480 | * @len: number of bytes to splice | |
481 | * @flags: splice modifier flags | |
482 | * | |
932cc6d4 JA |
483 | * Description: |
484 | * Will read pages from given file and fill them into a pipe. Can be | |
485 | * used as long as the address_space operations for the source implements | |
486 | * a readpage() hook. | |
487 | * | |
83f9135b | 488 | */ |
cbb7e577 JA |
489 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
490 | struct pipe_inode_info *pipe, size_t len, | |
491 | unsigned int flags) | |
5274f052 | 492 | { |
d366d398 | 493 | loff_t isize, left; |
8191ecd1 | 494 | int ret; |
d366d398 JA |
495 | |
496 | isize = i_size_read(in->f_mapping->host); | |
497 | if (unlikely(*ppos >= isize)) | |
498 | return 0; | |
499 | ||
500 | left = isize - *ppos; | |
501 | if (unlikely(left < len)) | |
502 | len = left; | |
5274f052 | 503 | |
8191ecd1 JA |
504 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
505 | if (ret > 0) | |
cbb7e577 | 506 | *ppos += ret; |
5274f052 JA |
507 | |
508 | return ret; | |
509 | } | |
059a8f37 JA |
510 | EXPORT_SYMBOL(generic_file_splice_read); |
511 | ||
6818173b MS |
512 | static const struct pipe_buf_operations default_pipe_buf_ops = { |
513 | .can_merge = 0, | |
514 | .map = generic_pipe_buf_map, | |
515 | .unmap = generic_pipe_buf_unmap, | |
516 | .confirm = generic_pipe_buf_confirm, | |
517 | .release = generic_pipe_buf_release, | |
518 | .steal = generic_pipe_buf_steal, | |
519 | .get = generic_pipe_buf_get, | |
520 | }; | |
521 | ||
522 | static ssize_t kernel_readv(struct file *file, const struct iovec *vec, | |
523 | unsigned long vlen, loff_t offset) | |
524 | { | |
525 | mm_segment_t old_fs; | |
526 | loff_t pos = offset; | |
527 | ssize_t res; | |
528 | ||
529 | old_fs = get_fs(); | |
530 | set_fs(get_ds()); | |
531 | /* The cast to a user pointer is valid due to the set_fs() */ | |
532 | res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos); | |
533 | set_fs(old_fs); | |
534 | ||
535 | return res; | |
536 | } | |
537 | ||
b2858d7d MS |
538 | static ssize_t kernel_write(struct file *file, const char *buf, size_t count, |
539 | loff_t pos) | |
0b0a47f5 MS |
540 | { |
541 | mm_segment_t old_fs; | |
542 | ssize_t res; | |
543 | ||
544 | old_fs = get_fs(); | |
545 | set_fs(get_ds()); | |
546 | /* The cast to a user pointer is valid due to the set_fs() */ | |
b2858d7d | 547 | res = vfs_write(file, (const char __user *)buf, count, &pos); |
0b0a47f5 MS |
548 | set_fs(old_fs); |
549 | ||
550 | return res; | |
551 | } | |
552 | ||
6818173b MS |
553 | ssize_t default_file_splice_read(struct file *in, loff_t *ppos, |
554 | struct pipe_inode_info *pipe, size_t len, | |
555 | unsigned int flags) | |
556 | { | |
557 | unsigned int nr_pages; | |
558 | unsigned int nr_freed; | |
559 | size_t offset; | |
560 | struct page *pages[PIPE_BUFFERS]; | |
561 | struct partial_page partial[PIPE_BUFFERS]; | |
562 | struct iovec vec[PIPE_BUFFERS]; | |
563 | pgoff_t index; | |
564 | ssize_t res; | |
565 | size_t this_len; | |
566 | int error; | |
567 | int i; | |
568 | struct splice_pipe_desc spd = { | |
569 | .pages = pages, | |
570 | .partial = partial, | |
571 | .flags = flags, | |
572 | .ops = &default_pipe_buf_ops, | |
573 | .spd_release = spd_release_page, | |
574 | }; | |
575 | ||
576 | index = *ppos >> PAGE_CACHE_SHIFT; | |
577 | offset = *ppos & ~PAGE_CACHE_MASK; | |
578 | nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
579 | ||
580 | for (i = 0; i < nr_pages && i < PIPE_BUFFERS && len; i++) { | |
581 | struct page *page; | |
582 | ||
4f231228 | 583 | page = alloc_page(GFP_USER); |
6818173b MS |
584 | error = -ENOMEM; |
585 | if (!page) | |
586 | goto err; | |
587 | ||
588 | this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset); | |
4f231228 | 589 | vec[i].iov_base = (void __user *) page_address(page); |
6818173b MS |
590 | vec[i].iov_len = this_len; |
591 | pages[i] = page; | |
592 | spd.nr_pages++; | |
593 | len -= this_len; | |
594 | offset = 0; | |
595 | } | |
596 | ||
597 | res = kernel_readv(in, vec, spd.nr_pages, *ppos); | |
77f6bf57 AM |
598 | if (res < 0) { |
599 | error = res; | |
6818173b | 600 | goto err; |
77f6bf57 | 601 | } |
6818173b MS |
602 | |
603 | error = 0; | |
604 | if (!res) | |
605 | goto err; | |
606 | ||
607 | nr_freed = 0; | |
608 | for (i = 0; i < spd.nr_pages; i++) { | |
6818173b MS |
609 | this_len = min_t(size_t, vec[i].iov_len, res); |
610 | partial[i].offset = 0; | |
611 | partial[i].len = this_len; | |
612 | if (!this_len) { | |
613 | __free_page(pages[i]); | |
614 | pages[i] = NULL; | |
615 | nr_freed++; | |
616 | } | |
617 | res -= this_len; | |
618 | } | |
619 | spd.nr_pages -= nr_freed; | |
620 | ||
621 | res = splice_to_pipe(pipe, &spd); | |
622 | if (res > 0) | |
623 | *ppos += res; | |
624 | ||
625 | return res; | |
626 | ||
627 | err: | |
4f231228 | 628 | for (i = 0; i < spd.nr_pages; i++) |
6818173b | 629 | __free_page(pages[i]); |
4f231228 | 630 | |
6818173b MS |
631 | return error; |
632 | } | |
633 | EXPORT_SYMBOL(default_file_splice_read); | |
634 | ||
5274f052 | 635 | /* |
4f6f0bd2 | 636 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 637 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 638 | */ |
76ad4d11 | 639 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
640 | struct pipe_buffer *buf, struct splice_desc *sd) |
641 | { | |
6a14b90b | 642 | struct file *file = sd->u.file; |
5274f052 | 643 | loff_t pos = sd->pos; |
f84d7519 | 644 | int ret, more; |
5274f052 | 645 | |
cac36bb0 | 646 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
647 | if (!ret) { |
648 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 649 | |
f84d7519 JA |
650 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
651 | sd->len, &pos, more); | |
652 | } | |
5274f052 | 653 | |
016b661e | 654 | return ret; |
5274f052 JA |
655 | } |
656 | ||
657 | /* | |
658 | * This is a little more tricky than the file -> pipe splicing. There are | |
659 | * basically three cases: | |
660 | * | |
661 | * - Destination page already exists in the address space and there | |
662 | * are users of it. For that case we have no other option that | |
663 | * copying the data. Tough luck. | |
664 | * - Destination page already exists in the address space, but there | |
665 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
666 | * through to last case. | |
667 | * - Destination page does not exist, we can add the pipe page to | |
668 | * the page cache and avoid the copy. | |
669 | * | |
83f9135b JA |
670 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
671 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
672 | * file address space page cache. This is possible if no one else has | |
673 | * the pipe page referenced outside of the pipe and page cache. If | |
674 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
675 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 676 | */ |
328eaaba MS |
677 | int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
678 | struct splice_desc *sd) | |
5274f052 | 679 | { |
6a14b90b | 680 | struct file *file = sd->u.file; |
5274f052 | 681 | struct address_space *mapping = file->f_mapping; |
016b661e | 682 | unsigned int offset, this_len; |
5274f052 | 683 | struct page *page; |
afddba49 | 684 | void *fsdata; |
3e7ee3e7 | 685 | int ret; |
5274f052 JA |
686 | |
687 | /* | |
49d0b21b | 688 | * make sure the data in this buffer is uptodate |
5274f052 | 689 | */ |
cac36bb0 | 690 | ret = buf->ops->confirm(pipe, buf); |
f84d7519 JA |
691 | if (unlikely(ret)) |
692 | return ret; | |
5274f052 | 693 | |
5274f052 JA |
694 | offset = sd->pos & ~PAGE_CACHE_MASK; |
695 | ||
016b661e JA |
696 | this_len = sd->len; |
697 | if (this_len + offset > PAGE_CACHE_SIZE) | |
698 | this_len = PAGE_CACHE_SIZE - offset; | |
699 | ||
afddba49 NP |
700 | ret = pagecache_write_begin(file, mapping, sd->pos, this_len, |
701 | AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); | |
702 | if (unlikely(ret)) | |
703 | goto out; | |
5274f052 | 704 | |
0568b409 | 705 | if (buf->page != page) { |
f84d7519 JA |
706 | /* |
707 | * Careful, ->map() uses KM_USER0! | |
708 | */ | |
76ad4d11 | 709 | char *src = buf->ops->map(pipe, buf, 1); |
f84d7519 | 710 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 711 | |
016b661e | 712 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 713 | flush_dcache_page(page); |
f84d7519 | 714 | kunmap_atomic(dst, KM_USER1); |
76ad4d11 | 715 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 716 | } |
afddba49 NP |
717 | ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len, |
718 | page, fsdata); | |
5274f052 | 719 | out: |
5274f052 JA |
720 | return ret; |
721 | } | |
328eaaba | 722 | EXPORT_SYMBOL(pipe_to_file); |
5274f052 | 723 | |
b3c2d2dd MS |
724 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
725 | { | |
726 | smp_mb(); | |
727 | if (waitqueue_active(&pipe->wait)) | |
728 | wake_up_interruptible(&pipe->wait); | |
729 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
730 | } | |
731 | ||
932cc6d4 | 732 | /** |
b3c2d2dd | 733 | * splice_from_pipe_feed - feed available data from a pipe to a file |
932cc6d4 JA |
734 | * @pipe: pipe to splice from |
735 | * @sd: information to @actor | |
736 | * @actor: handler that splices the data | |
737 | * | |
738 | * Description: | |
b3c2d2dd MS |
739 | * This function loops over the pipe and calls @actor to do the |
740 | * actual moving of a single struct pipe_buffer to the desired | |
741 | * destination. It returns when there's no more buffers left in | |
742 | * the pipe or if the requested number of bytes (@sd->total_len) | |
743 | * have been copied. It returns a positive number (one) if the | |
744 | * pipe needs to be filled with more data, zero if the required | |
745 | * number of bytes have been copied and -errno on error. | |
932cc6d4 | 746 | * |
b3c2d2dd MS |
747 | * This, together with splice_from_pipe_{begin,end,next}, may be |
748 | * used to implement the functionality of __splice_from_pipe() when | |
749 | * locking is required around copying the pipe buffers to the | |
750 | * destination. | |
83f9135b | 751 | */ |
b3c2d2dd MS |
752 | int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
753 | splice_actor *actor) | |
5274f052 | 754 | { |
b3c2d2dd | 755 | int ret; |
5274f052 | 756 | |
b3c2d2dd MS |
757 | while (pipe->nrbufs) { |
758 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
759 | const struct pipe_buf_operations *ops = buf->ops; | |
5274f052 | 760 | |
b3c2d2dd MS |
761 | sd->len = buf->len; |
762 | if (sd->len > sd->total_len) | |
763 | sd->len = sd->total_len; | |
5274f052 | 764 | |
b3c2d2dd MS |
765 | ret = actor(pipe, buf, sd); |
766 | if (ret <= 0) { | |
767 | if (ret == -ENODATA) | |
768 | ret = 0; | |
769 | return ret; | |
770 | } | |
771 | buf->offset += ret; | |
772 | buf->len -= ret; | |
773 | ||
774 | sd->num_spliced += ret; | |
775 | sd->len -= ret; | |
776 | sd->pos += ret; | |
777 | sd->total_len -= ret; | |
778 | ||
779 | if (!buf->len) { | |
780 | buf->ops = NULL; | |
781 | ops->release(pipe, buf); | |
782 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); | |
783 | pipe->nrbufs--; | |
784 | if (pipe->inode) | |
785 | sd->need_wakeup = true; | |
786 | } | |
5274f052 | 787 | |
b3c2d2dd MS |
788 | if (!sd->total_len) |
789 | return 0; | |
790 | } | |
5274f052 | 791 | |
b3c2d2dd MS |
792 | return 1; |
793 | } | |
794 | EXPORT_SYMBOL(splice_from_pipe_feed); | |
5274f052 | 795 | |
b3c2d2dd MS |
796 | /** |
797 | * splice_from_pipe_next - wait for some data to splice from | |
798 | * @pipe: pipe to splice from | |
799 | * @sd: information about the splice operation | |
800 | * | |
801 | * Description: | |
802 | * This function will wait for some data and return a positive | |
803 | * value (one) if pipe buffers are available. It will return zero | |
804 | * or -errno if no more data needs to be spliced. | |
805 | */ | |
806 | int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
807 | { | |
808 | while (!pipe->nrbufs) { | |
809 | if (!pipe->writers) | |
810 | return 0; | |
016b661e | 811 | |
b3c2d2dd MS |
812 | if (!pipe->waiting_writers && sd->num_spliced) |
813 | return 0; | |
73d62d83 | 814 | |
b3c2d2dd MS |
815 | if (sd->flags & SPLICE_F_NONBLOCK) |
816 | return -EAGAIN; | |
5274f052 | 817 | |
b3c2d2dd MS |
818 | if (signal_pending(current)) |
819 | return -ERESTARTSYS; | |
5274f052 | 820 | |
b3c2d2dd MS |
821 | if (sd->need_wakeup) { |
822 | wakeup_pipe_writers(pipe); | |
823 | sd->need_wakeup = false; | |
5274f052 JA |
824 | } |
825 | ||
b3c2d2dd MS |
826 | pipe_wait(pipe); |
827 | } | |
29e35094 | 828 | |
b3c2d2dd MS |
829 | return 1; |
830 | } | |
831 | EXPORT_SYMBOL(splice_from_pipe_next); | |
5274f052 | 832 | |
b3c2d2dd MS |
833 | /** |
834 | * splice_from_pipe_begin - start splicing from pipe | |
b80901bb | 835 | * @sd: information about the splice operation |
b3c2d2dd MS |
836 | * |
837 | * Description: | |
838 | * This function should be called before a loop containing | |
839 | * splice_from_pipe_next() and splice_from_pipe_feed() to | |
840 | * initialize the necessary fields of @sd. | |
841 | */ | |
842 | void splice_from_pipe_begin(struct splice_desc *sd) | |
843 | { | |
844 | sd->num_spliced = 0; | |
845 | sd->need_wakeup = false; | |
846 | } | |
847 | EXPORT_SYMBOL(splice_from_pipe_begin); | |
5274f052 | 848 | |
b3c2d2dd MS |
849 | /** |
850 | * splice_from_pipe_end - finish splicing from pipe | |
851 | * @pipe: pipe to splice from | |
852 | * @sd: information about the splice operation | |
853 | * | |
854 | * Description: | |
855 | * This function will wake up pipe writers if necessary. It should | |
856 | * be called after a loop containing splice_from_pipe_next() and | |
857 | * splice_from_pipe_feed(). | |
858 | */ | |
859 | void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
860 | { | |
861 | if (sd->need_wakeup) | |
862 | wakeup_pipe_writers(pipe); | |
863 | } | |
864 | EXPORT_SYMBOL(splice_from_pipe_end); | |
5274f052 | 865 | |
b3c2d2dd MS |
866 | /** |
867 | * __splice_from_pipe - splice data from a pipe to given actor | |
868 | * @pipe: pipe to splice from | |
869 | * @sd: information to @actor | |
870 | * @actor: handler that splices the data | |
871 | * | |
872 | * Description: | |
873 | * This function does little more than loop over the pipe and call | |
874 | * @actor to do the actual moving of a single struct pipe_buffer to | |
875 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
876 | * pipe_to_user. | |
877 | * | |
878 | */ | |
879 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, | |
880 | splice_actor *actor) | |
881 | { | |
882 | int ret; | |
5274f052 | 883 | |
b3c2d2dd MS |
884 | splice_from_pipe_begin(sd); |
885 | do { | |
886 | ret = splice_from_pipe_next(pipe, sd); | |
887 | if (ret > 0) | |
888 | ret = splice_from_pipe_feed(pipe, sd, actor); | |
889 | } while (ret > 0); | |
890 | splice_from_pipe_end(pipe, sd); | |
891 | ||
892 | return sd->num_spliced ? sd->num_spliced : ret; | |
5274f052 | 893 | } |
40bee44e | 894 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 895 | |
932cc6d4 JA |
896 | /** |
897 | * splice_from_pipe - splice data from a pipe to a file | |
898 | * @pipe: pipe to splice from | |
899 | * @out: file to splice to | |
900 | * @ppos: position in @out | |
901 | * @len: how many bytes to splice | |
902 | * @flags: splice modifier flags | |
903 | * @actor: handler that splices the data | |
904 | * | |
905 | * Description: | |
2933970b | 906 | * See __splice_from_pipe. This function locks the pipe inode, |
932cc6d4 JA |
907 | * otherwise it's identical to __splice_from_pipe(). |
908 | * | |
909 | */ | |
6da61809 MF |
910 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
911 | loff_t *ppos, size_t len, unsigned int flags, | |
912 | splice_actor *actor) | |
913 | { | |
914 | ssize_t ret; | |
c66ab6fa JA |
915 | struct splice_desc sd = { |
916 | .total_len = len, | |
917 | .flags = flags, | |
918 | .pos = *ppos, | |
6a14b90b | 919 | .u.file = out, |
c66ab6fa | 920 | }; |
6da61809 | 921 | |
61e0d47c | 922 | pipe_lock(pipe); |
c66ab6fa | 923 | ret = __splice_from_pipe(pipe, &sd, actor); |
61e0d47c | 924 | pipe_unlock(pipe); |
6da61809 MF |
925 | |
926 | return ret; | |
927 | } | |
928 | ||
83f9135b JA |
929 | /** |
930 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 931 | * @pipe: pipe info |
83f9135b | 932 | * @out: file to write to |
932cc6d4 | 933 | * @ppos: position in @out |
83f9135b JA |
934 | * @len: number of bytes to splice |
935 | * @flags: splice modifier flags | |
936 | * | |
932cc6d4 JA |
937 | * Description: |
938 | * Will either move or copy pages (determined by @flags options) from | |
939 | * the given pipe inode to the given file. | |
83f9135b JA |
940 | * |
941 | */ | |
3a326a2c IM |
942 | ssize_t |
943 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 944 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 945 | { |
4f6f0bd2 | 946 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 947 | struct inode *inode = mapping->host; |
7f3d4ee1 MS |
948 | struct splice_desc sd = { |
949 | .total_len = len, | |
950 | .flags = flags, | |
951 | .pos = *ppos, | |
952 | .u.file = out, | |
953 | }; | |
3a326a2c IM |
954 | ssize_t ret; |
955 | ||
61e0d47c | 956 | pipe_lock(pipe); |
eb443e5a MS |
957 | |
958 | splice_from_pipe_begin(&sd); | |
959 | do { | |
960 | ret = splice_from_pipe_next(pipe, &sd); | |
961 | if (ret <= 0) | |
962 | break; | |
963 | ||
964 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); | |
965 | ret = file_remove_suid(out); | |
966 | if (!ret) | |
967 | ret = splice_from_pipe_feed(pipe, &sd, pipe_to_file); | |
968 | mutex_unlock(&inode->i_mutex); | |
969 | } while (ret > 0); | |
970 | splice_from_pipe_end(pipe, &sd); | |
971 | ||
61e0d47c | 972 | pipe_unlock(pipe); |
eb443e5a MS |
973 | |
974 | if (sd.num_spliced) | |
975 | ret = sd.num_spliced; | |
976 | ||
a4514ebd | 977 | if (ret > 0) { |
17ee4f49 | 978 | unsigned long nr_pages; |
148f948b | 979 | int err; |
17ee4f49 | 980 | |
17ee4f49 | 981 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
a4514ebd | 982 | |
148f948b JK |
983 | err = generic_write_sync(out, *ppos, ret); |
984 | if (err) | |
985 | ret = err; | |
986 | else | |
987 | *ppos += ret; | |
17ee4f49 | 988 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
4f6f0bd2 JA |
989 | } |
990 | ||
991 | return ret; | |
5274f052 JA |
992 | } |
993 | ||
059a8f37 JA |
994 | EXPORT_SYMBOL(generic_file_splice_write); |
995 | ||
b2858d7d MS |
996 | static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
997 | struct splice_desc *sd) | |
0b0a47f5 | 998 | { |
b2858d7d MS |
999 | int ret; |
1000 | void *data; | |
1001 | ||
1002 | ret = buf->ops->confirm(pipe, buf); | |
1003 | if (ret) | |
1004 | return ret; | |
1005 | ||
1006 | data = buf->ops->map(pipe, buf, 0); | |
1007 | ret = kernel_write(sd->u.file, data + buf->offset, sd->len, sd->pos); | |
1008 | buf->ops->unmap(pipe, buf, data); | |
1009 | ||
1010 | return ret; | |
0b0a47f5 MS |
1011 | } |
1012 | ||
1013 | static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, | |
1014 | struct file *out, loff_t *ppos, | |
1015 | size_t len, unsigned int flags) | |
1016 | { | |
b2858d7d | 1017 | ssize_t ret; |
0b0a47f5 | 1018 | |
b2858d7d MS |
1019 | ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); |
1020 | if (ret > 0) | |
1021 | *ppos += ret; | |
0b0a47f5 | 1022 | |
b2858d7d | 1023 | return ret; |
0b0a47f5 MS |
1024 | } |
1025 | ||
83f9135b JA |
1026 | /** |
1027 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 1028 | * @pipe: pipe to splice from |
83f9135b | 1029 | * @out: socket to write to |
932cc6d4 | 1030 | * @ppos: position in @out |
83f9135b JA |
1031 | * @len: number of bytes to splice |
1032 | * @flags: splice modifier flags | |
1033 | * | |
932cc6d4 JA |
1034 | * Description: |
1035 | * Will send @len bytes from the pipe to a network socket. No data copying | |
1036 | * is involved. | |
83f9135b JA |
1037 | * |
1038 | */ | |
3a326a2c | 1039 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1040 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1041 | { |
00522fb4 | 1042 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
1043 | } |
1044 | ||
059a8f37 | 1045 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 1046 | |
83f9135b JA |
1047 | /* |
1048 | * Attempt to initiate a splice from pipe to file. | |
1049 | */ | |
3a326a2c | 1050 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1051 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1052 | { |
0b0a47f5 MS |
1053 | ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, |
1054 | loff_t *, size_t, unsigned int); | |
5274f052 JA |
1055 | int ret; |
1056 | ||
49570e9b | 1057 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
1058 | return -EBADF; |
1059 | ||
efc968d4 LT |
1060 | if (unlikely(out->f_flags & O_APPEND)) |
1061 | return -EINVAL; | |
1062 | ||
cbb7e577 | 1063 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
1064 | if (unlikely(ret < 0)) |
1065 | return ret; | |
1066 | ||
0b0a47f5 MS |
1067 | splice_write = out->f_op->splice_write; |
1068 | if (!splice_write) | |
1069 | splice_write = default_file_splice_write; | |
1070 | ||
1071 | return splice_write(pipe, out, ppos, len, flags); | |
5274f052 JA |
1072 | } |
1073 | ||
83f9135b JA |
1074 | /* |
1075 | * Attempt to initiate a splice from a file to a pipe. | |
1076 | */ | |
cbb7e577 JA |
1077 | static long do_splice_to(struct file *in, loff_t *ppos, |
1078 | struct pipe_inode_info *pipe, size_t len, | |
1079 | unsigned int flags) | |
5274f052 | 1080 | { |
6818173b MS |
1081 | ssize_t (*splice_read)(struct file *, loff_t *, |
1082 | struct pipe_inode_info *, size_t, unsigned int); | |
5274f052 JA |
1083 | int ret; |
1084 | ||
49570e9b | 1085 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
1086 | return -EBADF; |
1087 | ||
cbb7e577 | 1088 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
1089 | if (unlikely(ret < 0)) |
1090 | return ret; | |
1091 | ||
6818173b MS |
1092 | splice_read = in->f_op->splice_read; |
1093 | if (!splice_read) | |
1094 | splice_read = default_file_splice_read; | |
1095 | ||
1096 | return splice_read(in, ppos, pipe, len, flags); | |
5274f052 JA |
1097 | } |
1098 | ||
932cc6d4 JA |
1099 | /** |
1100 | * splice_direct_to_actor - splices data directly between two non-pipes | |
1101 | * @in: file to splice from | |
1102 | * @sd: actor information on where to splice to | |
1103 | * @actor: handles the data splicing | |
1104 | * | |
1105 | * Description: | |
1106 | * This is a special case helper to splice directly between two | |
1107 | * points, without requiring an explicit pipe. Internally an allocated | |
79685b8d | 1108 | * pipe is cached in the process, and reused during the lifetime of |
932cc6d4 JA |
1109 | * that process. |
1110 | * | |
c66ab6fa JA |
1111 | */ |
1112 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1113 | splice_direct_actor *actor) | |
b92ce558 JA |
1114 | { |
1115 | struct pipe_inode_info *pipe; | |
1116 | long ret, bytes; | |
1117 | umode_t i_mode; | |
c66ab6fa JA |
1118 | size_t len; |
1119 | int i, flags; | |
b92ce558 JA |
1120 | |
1121 | /* | |
1122 | * We require the input being a regular file, as we don't want to | |
1123 | * randomly drop data for eg socket -> socket splicing. Use the | |
1124 | * piped splicing for that! | |
1125 | */ | |
0f7fc9e4 | 1126 | i_mode = in->f_path.dentry->d_inode->i_mode; |
b92ce558 JA |
1127 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1128 | return -EINVAL; | |
1129 | ||
1130 | /* | |
1131 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1132 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1133 | */ | |
1134 | pipe = current->splice_pipe; | |
49570e9b | 1135 | if (unlikely(!pipe)) { |
b92ce558 JA |
1136 | pipe = alloc_pipe_info(NULL); |
1137 | if (!pipe) | |
1138 | return -ENOMEM; | |
1139 | ||
1140 | /* | |
1141 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1142 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1143 | * PIPE_READERS appropriately. |
1144 | */ | |
1145 | pipe->readers = 1; | |
1146 | ||
1147 | current->splice_pipe = pipe; | |
1148 | } | |
1149 | ||
1150 | /* | |
73d62d83 | 1151 | * Do the splice. |
b92ce558 JA |
1152 | */ |
1153 | ret = 0; | |
1154 | bytes = 0; | |
c66ab6fa JA |
1155 | len = sd->total_len; |
1156 | flags = sd->flags; | |
1157 | ||
1158 | /* | |
1159 | * Don't block on output, we have to drain the direct pipe. | |
1160 | */ | |
1161 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
b92ce558 JA |
1162 | |
1163 | while (len) { | |
51a92c0f | 1164 | size_t read_len; |
a82c53a0 | 1165 | loff_t pos = sd->pos, prev_pos = pos; |
b92ce558 | 1166 | |
bcd4f3ac | 1167 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1168 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1169 | goto out_release; |
1170 | ||
1171 | read_len = ret; | |
c66ab6fa | 1172 | sd->total_len = read_len; |
b92ce558 JA |
1173 | |
1174 | /* | |
1175 | * NOTE: nonblocking mode only applies to the input. We | |
1176 | * must not do the output in nonblocking mode as then we | |
1177 | * could get stuck data in the internal pipe: | |
1178 | */ | |
c66ab6fa | 1179 | ret = actor(pipe, sd); |
a82c53a0 TZ |
1180 | if (unlikely(ret <= 0)) { |
1181 | sd->pos = prev_pos; | |
b92ce558 | 1182 | goto out_release; |
a82c53a0 | 1183 | } |
b92ce558 JA |
1184 | |
1185 | bytes += ret; | |
1186 | len -= ret; | |
bcd4f3ac | 1187 | sd->pos = pos; |
b92ce558 | 1188 | |
a82c53a0 TZ |
1189 | if (ret < read_len) { |
1190 | sd->pos = prev_pos + ret; | |
51a92c0f | 1191 | goto out_release; |
a82c53a0 | 1192 | } |
b92ce558 JA |
1193 | } |
1194 | ||
9e97198d | 1195 | done: |
b92ce558 | 1196 | pipe->nrbufs = pipe->curbuf = 0; |
80848708 | 1197 | file_accessed(in); |
b92ce558 JA |
1198 | return bytes; |
1199 | ||
1200 | out_release: | |
1201 | /* | |
1202 | * If we did an incomplete transfer we must release | |
1203 | * the pipe buffers in question: | |
1204 | */ | |
1205 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1206 | struct pipe_buffer *buf = pipe->bufs + i; | |
1207 | ||
1208 | if (buf->ops) { | |
1209 | buf->ops->release(pipe, buf); | |
1210 | buf->ops = NULL; | |
1211 | } | |
1212 | } | |
b92ce558 | 1213 | |
9e97198d JA |
1214 | if (!bytes) |
1215 | bytes = ret; | |
c66ab6fa | 1216 | |
9e97198d | 1217 | goto done; |
c66ab6fa JA |
1218 | } |
1219 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1220 | ||
1221 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1222 | struct splice_desc *sd) | |
1223 | { | |
6a14b90b | 1224 | struct file *file = sd->u.file; |
c66ab6fa JA |
1225 | |
1226 | return do_splice_from(pipe, file, &sd->pos, sd->total_len, sd->flags); | |
1227 | } | |
1228 | ||
932cc6d4 JA |
1229 | /** |
1230 | * do_splice_direct - splices data directly between two files | |
1231 | * @in: file to splice from | |
1232 | * @ppos: input file offset | |
1233 | * @out: file to splice to | |
1234 | * @len: number of bytes to splice | |
1235 | * @flags: splice modifier flags | |
1236 | * | |
1237 | * Description: | |
1238 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1239 | * doing it in the application would incur an extra system call | |
1240 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1241 | * can splice directly through a process-private pipe. | |
1242 | * | |
1243 | */ | |
c66ab6fa JA |
1244 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
1245 | size_t len, unsigned int flags) | |
1246 | { | |
1247 | struct splice_desc sd = { | |
1248 | .len = len, | |
1249 | .total_len = len, | |
1250 | .flags = flags, | |
1251 | .pos = *ppos, | |
6a14b90b | 1252 | .u.file = out, |
c66ab6fa | 1253 | }; |
51a92c0f | 1254 | long ret; |
c66ab6fa JA |
1255 | |
1256 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); | |
51a92c0f | 1257 | if (ret > 0) |
a82c53a0 | 1258 | *ppos = sd.pos; |
51a92c0f | 1259 | |
c66ab6fa | 1260 | return ret; |
b92ce558 JA |
1261 | } |
1262 | ||
7c77f0b3 MS |
1263 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
1264 | struct pipe_inode_info *opipe, | |
1265 | size_t len, unsigned int flags); | |
ddac0d39 JA |
1266 | /* |
1267 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1268 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1269 | * pipe. | |
1270 | */ | |
1271 | static inline struct pipe_inode_info *pipe_info(struct inode *inode) | |
1272 | { | |
1273 | if (S_ISFIFO(inode->i_mode)) | |
1274 | return inode->i_pipe; | |
1275 | ||
1276 | return NULL; | |
1277 | } | |
1278 | ||
83f9135b JA |
1279 | /* |
1280 | * Determine where to splice to/from. | |
1281 | */ | |
529565dc IM |
1282 | static long do_splice(struct file *in, loff_t __user *off_in, |
1283 | struct file *out, loff_t __user *off_out, | |
1284 | size_t len, unsigned int flags) | |
5274f052 | 1285 | { |
7c77f0b3 MS |
1286 | struct pipe_inode_info *ipipe; |
1287 | struct pipe_inode_info *opipe; | |
cbb7e577 | 1288 | loff_t offset, *off; |
a4514ebd | 1289 | long ret; |
5274f052 | 1290 | |
7c77f0b3 MS |
1291 | ipipe = pipe_info(in->f_path.dentry->d_inode); |
1292 | opipe = pipe_info(out->f_path.dentry->d_inode); | |
1293 | ||
1294 | if (ipipe && opipe) { | |
1295 | if (off_in || off_out) | |
1296 | return -ESPIPE; | |
1297 | ||
1298 | if (!(in->f_mode & FMODE_READ)) | |
1299 | return -EBADF; | |
1300 | ||
1301 | if (!(out->f_mode & FMODE_WRITE)) | |
1302 | return -EBADF; | |
1303 | ||
1304 | /* Splicing to self would be fun, but... */ | |
1305 | if (ipipe == opipe) | |
1306 | return -EINVAL; | |
1307 | ||
1308 | return splice_pipe_to_pipe(ipipe, opipe, len, flags); | |
1309 | } | |
1310 | ||
1311 | if (ipipe) { | |
529565dc IM |
1312 | if (off_in) |
1313 | return -ESPIPE; | |
b92ce558 JA |
1314 | if (off_out) { |
1315 | if (out->f_op->llseek == no_llseek) | |
1316 | return -EINVAL; | |
cbb7e577 | 1317 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1318 | return -EFAULT; |
cbb7e577 JA |
1319 | off = &offset; |
1320 | } else | |
1321 | off = &out->f_pos; | |
529565dc | 1322 | |
7c77f0b3 | 1323 | ret = do_splice_from(ipipe, out, off, len, flags); |
a4514ebd JA |
1324 | |
1325 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1326 | ret = -EFAULT; | |
1327 | ||
1328 | return ret; | |
529565dc | 1329 | } |
5274f052 | 1330 | |
7c77f0b3 | 1331 | if (opipe) { |
529565dc IM |
1332 | if (off_out) |
1333 | return -ESPIPE; | |
b92ce558 JA |
1334 | if (off_in) { |
1335 | if (in->f_op->llseek == no_llseek) | |
1336 | return -EINVAL; | |
cbb7e577 | 1337 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1338 | return -EFAULT; |
cbb7e577 JA |
1339 | off = &offset; |
1340 | } else | |
1341 | off = &in->f_pos; | |
529565dc | 1342 | |
7c77f0b3 | 1343 | ret = do_splice_to(in, off, opipe, len, flags); |
a4514ebd JA |
1344 | |
1345 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1346 | ret = -EFAULT; | |
1347 | ||
1348 | return ret; | |
529565dc | 1349 | } |
5274f052 JA |
1350 | |
1351 | return -EINVAL; | |
1352 | } | |
1353 | ||
912d35f8 JA |
1354 | /* |
1355 | * Map an iov into an array of pages and offset/length tupples. With the | |
1356 | * partial_page structure, we can map several non-contiguous ranges into | |
1357 | * our ones pages[] map instead of splitting that operation into pieces. | |
1358 | * Could easily be exported as a generic helper for other users, in which | |
1359 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1360 | */ | |
1361 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1362 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1363 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1364 | { |
1365 | int buffers = 0, error = 0; | |
1366 | ||
912d35f8 JA |
1367 | while (nr_vecs) { |
1368 | unsigned long off, npages; | |
75723957 | 1369 | struct iovec entry; |
912d35f8 JA |
1370 | void __user *base; |
1371 | size_t len; | |
1372 | int i; | |
1373 | ||
75723957 | 1374 | error = -EFAULT; |
bc40d73c | 1375 | if (copy_from_user(&entry, iov, sizeof(entry))) |
912d35f8 JA |
1376 | break; |
1377 | ||
75723957 LT |
1378 | base = entry.iov_base; |
1379 | len = entry.iov_len; | |
1380 | ||
912d35f8 JA |
1381 | /* |
1382 | * Sanity check this iovec. 0 read succeeds. | |
1383 | */ | |
75723957 | 1384 | error = 0; |
912d35f8 JA |
1385 | if (unlikely(!len)) |
1386 | break; | |
1387 | error = -EFAULT; | |
712a30e6 | 1388 | if (!access_ok(VERIFY_READ, base, len)) |
912d35f8 JA |
1389 | break; |
1390 | ||
1391 | /* | |
1392 | * Get this base offset and number of pages, then map | |
1393 | * in the user pages. | |
1394 | */ | |
1395 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1396 | |
1397 | /* | |
1398 | * If asked for alignment, the offset must be zero and the | |
1399 | * length a multiple of the PAGE_SIZE. | |
1400 | */ | |
1401 | error = -EINVAL; | |
1402 | if (aligned && (off || len & ~PAGE_MASK)) | |
1403 | break; | |
1404 | ||
912d35f8 JA |
1405 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1406 | if (npages > PIPE_BUFFERS - buffers) | |
1407 | npages = PIPE_BUFFERS - buffers; | |
1408 | ||
bc40d73c NP |
1409 | error = get_user_pages_fast((unsigned long)base, npages, |
1410 | 0, &pages[buffers]); | |
912d35f8 JA |
1411 | |
1412 | if (unlikely(error <= 0)) | |
1413 | break; | |
1414 | ||
1415 | /* | |
1416 | * Fill this contiguous range into the partial page map. | |
1417 | */ | |
1418 | for (i = 0; i < error; i++) { | |
7591489a | 1419 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1420 | |
1421 | partial[buffers].offset = off; | |
1422 | partial[buffers].len = plen; | |
1423 | ||
1424 | off = 0; | |
1425 | len -= plen; | |
1426 | buffers++; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * We didn't complete this iov, stop here since it probably | |
1431 | * means we have to move some of this into a pipe to | |
1432 | * be able to continue. | |
1433 | */ | |
1434 | if (len) | |
1435 | break; | |
1436 | ||
1437 | /* | |
1438 | * Don't continue if we mapped fewer pages than we asked for, | |
1439 | * or if we mapped the max number of pages that we have | |
1440 | * room for. | |
1441 | */ | |
1442 | if (error < npages || buffers == PIPE_BUFFERS) | |
1443 | break; | |
1444 | ||
1445 | nr_vecs--; | |
1446 | iov++; | |
1447 | } | |
1448 | ||
912d35f8 JA |
1449 | if (buffers) |
1450 | return buffers; | |
1451 | ||
1452 | return error; | |
1453 | } | |
1454 | ||
6a14b90b JA |
1455 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1456 | struct splice_desc *sd) | |
1457 | { | |
1458 | char *src; | |
1459 | int ret; | |
1460 | ||
cac36bb0 | 1461 | ret = buf->ops->confirm(pipe, buf); |
6a14b90b JA |
1462 | if (unlikely(ret)) |
1463 | return ret; | |
1464 | ||
1465 | /* | |
1466 | * See if we can use the atomic maps, by prefaulting in the | |
1467 | * pages and doing an atomic copy | |
1468 | */ | |
1469 | if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { | |
1470 | src = buf->ops->map(pipe, buf, 1); | |
1471 | ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, | |
1472 | sd->len); | |
1473 | buf->ops->unmap(pipe, buf, src); | |
1474 | if (!ret) { | |
1475 | ret = sd->len; | |
1476 | goto out; | |
1477 | } | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * No dice, use slow non-atomic map and copy | |
1482 | */ | |
1483 | src = buf->ops->map(pipe, buf, 0); | |
1484 | ||
1485 | ret = sd->len; | |
1486 | if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) | |
1487 | ret = -EFAULT; | |
1488 | ||
6866bef4 | 1489 | buf->ops->unmap(pipe, buf, src); |
6a14b90b JA |
1490 | out: |
1491 | if (ret > 0) | |
1492 | sd->u.userptr += ret; | |
6a14b90b JA |
1493 | return ret; |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * For lack of a better implementation, implement vmsplice() to userspace | |
1498 | * as a simple copy of the pipes pages to the user iov. | |
1499 | */ | |
1500 | static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, | |
1501 | unsigned long nr_segs, unsigned int flags) | |
1502 | { | |
1503 | struct pipe_inode_info *pipe; | |
1504 | struct splice_desc sd; | |
1505 | ssize_t size; | |
1506 | int error; | |
1507 | long ret; | |
1508 | ||
1509 | pipe = pipe_info(file->f_path.dentry->d_inode); | |
1510 | if (!pipe) | |
1511 | return -EBADF; | |
1512 | ||
61e0d47c | 1513 | pipe_lock(pipe); |
6a14b90b JA |
1514 | |
1515 | error = ret = 0; | |
1516 | while (nr_segs) { | |
1517 | void __user *base; | |
1518 | size_t len; | |
1519 | ||
1520 | /* | |
1521 | * Get user address base and length for this iovec. | |
1522 | */ | |
1523 | error = get_user(base, &iov->iov_base); | |
1524 | if (unlikely(error)) | |
1525 | break; | |
1526 | error = get_user(len, &iov->iov_len); | |
1527 | if (unlikely(error)) | |
1528 | break; | |
1529 | ||
1530 | /* | |
1531 | * Sanity check this iovec. 0 read succeeds. | |
1532 | */ | |
1533 | if (unlikely(!len)) | |
1534 | break; | |
1535 | if (unlikely(!base)) { | |
1536 | error = -EFAULT; | |
1537 | break; | |
1538 | } | |
1539 | ||
8811930d JA |
1540 | if (unlikely(!access_ok(VERIFY_WRITE, base, len))) { |
1541 | error = -EFAULT; | |
1542 | break; | |
1543 | } | |
1544 | ||
6a14b90b JA |
1545 | sd.len = 0; |
1546 | sd.total_len = len; | |
1547 | sd.flags = flags; | |
1548 | sd.u.userptr = base; | |
1549 | sd.pos = 0; | |
1550 | ||
1551 | size = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1552 | if (size < 0) { | |
1553 | if (!ret) | |
1554 | ret = size; | |
1555 | ||
1556 | break; | |
1557 | } | |
1558 | ||
1559 | ret += size; | |
1560 | ||
1561 | if (size < len) | |
1562 | break; | |
1563 | ||
1564 | nr_segs--; | |
1565 | iov++; | |
1566 | } | |
1567 | ||
61e0d47c | 1568 | pipe_unlock(pipe); |
6a14b90b JA |
1569 | |
1570 | if (!ret) | |
1571 | ret = error; | |
1572 | ||
1573 | return ret; | |
1574 | } | |
1575 | ||
912d35f8 JA |
1576 | /* |
1577 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1578 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1579 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1580 | */ |
6a14b90b JA |
1581 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1582 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1583 | { |
ddac0d39 | 1584 | struct pipe_inode_info *pipe; |
912d35f8 JA |
1585 | struct page *pages[PIPE_BUFFERS]; |
1586 | struct partial_page partial[PIPE_BUFFERS]; | |
1587 | struct splice_pipe_desc spd = { | |
1588 | .pages = pages, | |
1589 | .partial = partial, | |
1590 | .flags = flags, | |
1591 | .ops = &user_page_pipe_buf_ops, | |
bbdfc2f7 | 1592 | .spd_release = spd_release_page, |
912d35f8 JA |
1593 | }; |
1594 | ||
0f7fc9e4 | 1595 | pipe = pipe_info(file->f_path.dentry->d_inode); |
ddac0d39 | 1596 | if (!pipe) |
912d35f8 | 1597 | return -EBADF; |
912d35f8 | 1598 | |
7afa6fd0 JA |
1599 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1600 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1601 | if (spd.nr_pages <= 0) |
1602 | return spd.nr_pages; | |
1603 | ||
00522fb4 | 1604 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1605 | } |
1606 | ||
6a14b90b JA |
1607 | /* |
1608 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1609 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1610 | * operation that can be supported without any funky alignment restrictions | |
1611 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1612 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1613 | * solutions for that: | |
1614 | * | |
1615 | * - memcpy() the data internally, at which point we might as well just | |
1616 | * do a regular read() on the buffer anyway. | |
1617 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1618 | * has restriction limitations on both ends of the pipe). | |
1619 | * | |
1620 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1621 | * | |
1622 | */ | |
836f92ad HC |
1623 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, |
1624 | unsigned long, nr_segs, unsigned int, flags) | |
912d35f8 JA |
1625 | { |
1626 | struct file *file; | |
1627 | long error; | |
1628 | int fput; | |
1629 | ||
6a14b90b JA |
1630 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1631 | return -EINVAL; | |
1632 | else if (unlikely(!nr_segs)) | |
1633 | return 0; | |
1634 | ||
912d35f8 JA |
1635 | error = -EBADF; |
1636 | file = fget_light(fd, &fput); | |
1637 | if (file) { | |
1638 | if (file->f_mode & FMODE_WRITE) | |
6a14b90b JA |
1639 | error = vmsplice_to_pipe(file, iov, nr_segs, flags); |
1640 | else if (file->f_mode & FMODE_READ) | |
1641 | error = vmsplice_to_user(file, iov, nr_segs, flags); | |
912d35f8 JA |
1642 | |
1643 | fput_light(file, fput); | |
1644 | } | |
1645 | ||
1646 | return error; | |
1647 | } | |
1648 | ||
836f92ad HC |
1649 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
1650 | int, fd_out, loff_t __user *, off_out, | |
1651 | size_t, len, unsigned int, flags) | |
5274f052 JA |
1652 | { |
1653 | long error; | |
1654 | struct file *in, *out; | |
1655 | int fput_in, fput_out; | |
1656 | ||
1657 | if (unlikely(!len)) | |
1658 | return 0; | |
1659 | ||
1660 | error = -EBADF; | |
529565dc | 1661 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1662 | if (in) { |
1663 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1664 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1665 | if (out) { |
1666 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1667 | error = do_splice(in, off_in, |
1668 | out, off_out, | |
1669 | len, flags); | |
5274f052 JA |
1670 | fput_light(out, fput_out); |
1671 | } | |
1672 | } | |
1673 | ||
1674 | fput_light(in, fput_in); | |
1675 | } | |
1676 | ||
1677 | return error; | |
1678 | } | |
70524490 | 1679 | |
aadd06e5 JA |
1680 | /* |
1681 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1682 | * return an appropriate error. | |
1683 | */ | |
7c77f0b3 | 1684 | static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1685 | { |
1686 | int ret; | |
1687 | ||
1688 | /* | |
1689 | * Check ->nrbufs without the inode lock first. This function | |
1690 | * is speculative anyways, so missing one is ok. | |
1691 | */ | |
1692 | if (pipe->nrbufs) | |
1693 | return 0; | |
1694 | ||
1695 | ret = 0; | |
61e0d47c | 1696 | pipe_lock(pipe); |
aadd06e5 JA |
1697 | |
1698 | while (!pipe->nrbufs) { | |
1699 | if (signal_pending(current)) { | |
1700 | ret = -ERESTARTSYS; | |
1701 | break; | |
1702 | } | |
1703 | if (!pipe->writers) | |
1704 | break; | |
1705 | if (!pipe->waiting_writers) { | |
1706 | if (flags & SPLICE_F_NONBLOCK) { | |
1707 | ret = -EAGAIN; | |
1708 | break; | |
1709 | } | |
1710 | } | |
1711 | pipe_wait(pipe); | |
1712 | } | |
1713 | ||
61e0d47c | 1714 | pipe_unlock(pipe); |
aadd06e5 JA |
1715 | return ret; |
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1720 | * return an appropriate error. | |
1721 | */ | |
7c77f0b3 | 1722 | static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1723 | { |
1724 | int ret; | |
1725 | ||
1726 | /* | |
1727 | * Check ->nrbufs without the inode lock first. This function | |
1728 | * is speculative anyways, so missing one is ok. | |
1729 | */ | |
1730 | if (pipe->nrbufs < PIPE_BUFFERS) | |
1731 | return 0; | |
1732 | ||
1733 | ret = 0; | |
61e0d47c | 1734 | pipe_lock(pipe); |
aadd06e5 JA |
1735 | |
1736 | while (pipe->nrbufs >= PIPE_BUFFERS) { | |
1737 | if (!pipe->readers) { | |
1738 | send_sig(SIGPIPE, current, 0); | |
1739 | ret = -EPIPE; | |
1740 | break; | |
1741 | } | |
1742 | if (flags & SPLICE_F_NONBLOCK) { | |
1743 | ret = -EAGAIN; | |
1744 | break; | |
1745 | } | |
1746 | if (signal_pending(current)) { | |
1747 | ret = -ERESTARTSYS; | |
1748 | break; | |
1749 | } | |
1750 | pipe->waiting_writers++; | |
1751 | pipe_wait(pipe); | |
1752 | pipe->waiting_writers--; | |
1753 | } | |
1754 | ||
61e0d47c | 1755 | pipe_unlock(pipe); |
aadd06e5 JA |
1756 | return ret; |
1757 | } | |
1758 | ||
7c77f0b3 MS |
1759 | /* |
1760 | * Splice contents of ipipe to opipe. | |
1761 | */ | |
1762 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, | |
1763 | struct pipe_inode_info *opipe, | |
1764 | size_t len, unsigned int flags) | |
1765 | { | |
1766 | struct pipe_buffer *ibuf, *obuf; | |
1767 | int ret = 0, nbuf; | |
1768 | bool input_wakeup = false; | |
1769 | ||
1770 | ||
1771 | retry: | |
1772 | ret = ipipe_prep(ipipe, flags); | |
1773 | if (ret) | |
1774 | return ret; | |
1775 | ||
1776 | ret = opipe_prep(opipe, flags); | |
1777 | if (ret) | |
1778 | return ret; | |
1779 | ||
1780 | /* | |
1781 | * Potential ABBA deadlock, work around it by ordering lock | |
1782 | * grabbing by pipe info address. Otherwise two different processes | |
1783 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1784 | */ | |
1785 | pipe_double_lock(ipipe, opipe); | |
1786 | ||
1787 | do { | |
1788 | if (!opipe->readers) { | |
1789 | send_sig(SIGPIPE, current, 0); | |
1790 | if (!ret) | |
1791 | ret = -EPIPE; | |
1792 | break; | |
1793 | } | |
1794 | ||
1795 | if (!ipipe->nrbufs && !ipipe->writers) | |
1796 | break; | |
1797 | ||
1798 | /* | |
1799 | * Cannot make any progress, because either the input | |
1800 | * pipe is empty or the output pipe is full. | |
1801 | */ | |
1802 | if (!ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) { | |
1803 | /* Already processed some buffers, break */ | |
1804 | if (ret) | |
1805 | break; | |
1806 | ||
1807 | if (flags & SPLICE_F_NONBLOCK) { | |
1808 | ret = -EAGAIN; | |
1809 | break; | |
1810 | } | |
1811 | ||
1812 | /* | |
1813 | * We raced with another reader/writer and haven't | |
1814 | * managed to process any buffers. A zero return | |
1815 | * value means EOF, so retry instead. | |
1816 | */ | |
1817 | pipe_unlock(ipipe); | |
1818 | pipe_unlock(opipe); | |
1819 | goto retry; | |
1820 | } | |
1821 | ||
1822 | ibuf = ipipe->bufs + ipipe->curbuf; | |
1823 | nbuf = (opipe->curbuf + opipe->nrbufs) % PIPE_BUFFERS; | |
1824 | obuf = opipe->bufs + nbuf; | |
1825 | ||
1826 | if (len >= ibuf->len) { | |
1827 | /* | |
1828 | * Simply move the whole buffer from ipipe to opipe | |
1829 | */ | |
1830 | *obuf = *ibuf; | |
1831 | ibuf->ops = NULL; | |
1832 | opipe->nrbufs++; | |
1833 | ipipe->curbuf = (ipipe->curbuf + 1) % PIPE_BUFFERS; | |
1834 | ipipe->nrbufs--; | |
1835 | input_wakeup = true; | |
1836 | } else { | |
1837 | /* | |
1838 | * Get a reference to this pipe buffer, | |
1839 | * so we can copy the contents over. | |
1840 | */ | |
1841 | ibuf->ops->get(ipipe, ibuf); | |
1842 | *obuf = *ibuf; | |
1843 | ||
1844 | /* | |
1845 | * Don't inherit the gift flag, we need to | |
1846 | * prevent multiple steals of this page. | |
1847 | */ | |
1848 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1849 | ||
1850 | obuf->len = len; | |
1851 | opipe->nrbufs++; | |
1852 | ibuf->offset += obuf->len; | |
1853 | ibuf->len -= obuf->len; | |
1854 | } | |
1855 | ret += obuf->len; | |
1856 | len -= obuf->len; | |
1857 | } while (len); | |
1858 | ||
1859 | pipe_unlock(ipipe); | |
1860 | pipe_unlock(opipe); | |
1861 | ||
1862 | /* | |
1863 | * If we put data in the output pipe, wakeup any potential readers. | |
1864 | */ | |
1865 | if (ret > 0) { | |
1866 | smp_mb(); | |
1867 | if (waitqueue_active(&opipe->wait)) | |
1868 | wake_up_interruptible(&opipe->wait); | |
1869 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1870 | } | |
1871 | if (input_wakeup) | |
1872 | wakeup_pipe_writers(ipipe); | |
1873 | ||
1874 | return ret; | |
1875 | } | |
1876 | ||
70524490 JA |
1877 | /* |
1878 | * Link contents of ipipe to opipe. | |
1879 | */ | |
1880 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1881 | struct pipe_inode_info *opipe, | |
1882 | size_t len, unsigned int flags) | |
1883 | { | |
1884 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1885 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1886 | |
1887 | /* | |
1888 | * Potential ABBA deadlock, work around it by ordering lock | |
61e0d47c | 1889 | * grabbing by pipe info address. Otherwise two different processes |
70524490 JA |
1890 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
1891 | */ | |
61e0d47c | 1892 | pipe_double_lock(ipipe, opipe); |
70524490 | 1893 | |
aadd06e5 | 1894 | do { |
70524490 JA |
1895 | if (!opipe->readers) { |
1896 | send_sig(SIGPIPE, current, 0); | |
1897 | if (!ret) | |
1898 | ret = -EPIPE; | |
1899 | break; | |
1900 | } | |
70524490 | 1901 | |
aadd06e5 JA |
1902 | /* |
1903 | * If we have iterated all input buffers or ran out of | |
1904 | * output room, break. | |
1905 | */ | |
1906 | if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) | |
1907 | break; | |
70524490 | 1908 | |
aadd06e5 JA |
1909 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); |
1910 | nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
70524490 JA |
1911 | |
1912 | /* | |
aadd06e5 JA |
1913 | * Get a reference to this pipe buffer, |
1914 | * so we can copy the contents over. | |
70524490 | 1915 | */ |
aadd06e5 JA |
1916 | ibuf->ops->get(ipipe, ibuf); |
1917 | ||
1918 | obuf = opipe->bufs + nbuf; | |
1919 | *obuf = *ibuf; | |
1920 | ||
2a27250e | 1921 | /* |
aadd06e5 JA |
1922 | * Don't inherit the gift flag, we need to |
1923 | * prevent multiple steals of this page. | |
2a27250e | 1924 | */ |
aadd06e5 | 1925 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1926 | |
aadd06e5 JA |
1927 | if (obuf->len > len) |
1928 | obuf->len = len; | |
70524490 | 1929 | |
aadd06e5 JA |
1930 | opipe->nrbufs++; |
1931 | ret += obuf->len; | |
1932 | len -= obuf->len; | |
1933 | i++; | |
1934 | } while (len); | |
70524490 | 1935 | |
02cf01ae JA |
1936 | /* |
1937 | * return EAGAIN if we have the potential of some data in the | |
1938 | * future, otherwise just return 0 | |
1939 | */ | |
1940 | if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) | |
1941 | ret = -EAGAIN; | |
1942 | ||
61e0d47c MS |
1943 | pipe_unlock(ipipe); |
1944 | pipe_unlock(opipe); | |
70524490 | 1945 | |
aadd06e5 JA |
1946 | /* |
1947 | * If we put data in the output pipe, wakeup any potential readers. | |
1948 | */ | |
1949 | if (ret > 0) { | |
70524490 JA |
1950 | smp_mb(); |
1951 | if (waitqueue_active(&opipe->wait)) | |
1952 | wake_up_interruptible(&opipe->wait); | |
1953 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1954 | } | |
1955 | ||
1956 | return ret; | |
1957 | } | |
1958 | ||
1959 | /* | |
1960 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1961 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1962 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1963 | * applicable one is SPLICE_F_NONBLOCK. | |
1964 | */ | |
1965 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1966 | unsigned int flags) | |
1967 | { | |
0f7fc9e4 JJS |
1968 | struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); |
1969 | struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); | |
aadd06e5 | 1970 | int ret = -EINVAL; |
70524490 JA |
1971 | |
1972 | /* | |
aadd06e5 JA |
1973 | * Duplicate the contents of ipipe to opipe without actually |
1974 | * copying the data. | |
70524490 | 1975 | */ |
aadd06e5 JA |
1976 | if (ipipe && opipe && ipipe != opipe) { |
1977 | /* | |
1978 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1979 | * ordering doesn't really matter. | |
1980 | */ | |
7c77f0b3 | 1981 | ret = ipipe_prep(ipipe, flags); |
aadd06e5 | 1982 | if (!ret) { |
7c77f0b3 | 1983 | ret = opipe_prep(opipe, flags); |
02cf01ae | 1984 | if (!ret) |
aadd06e5 | 1985 | ret = link_pipe(ipipe, opipe, len, flags); |
aadd06e5 JA |
1986 | } |
1987 | } | |
70524490 | 1988 | |
aadd06e5 | 1989 | return ret; |
70524490 JA |
1990 | } |
1991 | ||
836f92ad | 1992 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
70524490 JA |
1993 | { |
1994 | struct file *in; | |
1995 | int error, fput_in; | |
1996 | ||
1997 | if (unlikely(!len)) | |
1998 | return 0; | |
1999 | ||
2000 | error = -EBADF; | |
2001 | in = fget_light(fdin, &fput_in); | |
2002 | if (in) { | |
2003 | if (in->f_mode & FMODE_READ) { | |
2004 | int fput_out; | |
2005 | struct file *out = fget_light(fdout, &fput_out); | |
2006 | ||
2007 | if (out) { | |
2008 | if (out->f_mode & FMODE_WRITE) | |
2009 | error = do_tee(in, out, len, flags); | |
2010 | fput_light(out, fput_out); | |
2011 | } | |
2012 | } | |
2013 | fput_light(in, fput_in); | |
2014 | } | |
2015 | ||
2016 | return error; | |
2017 | } |