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> | |
23 | #include <linux/pipe_fs_i.h> | |
24 | #include <linux/mm_inline.h> | |
5abc97aa | 25 | #include <linux/swap.h> |
4f6f0bd2 JA |
26 | #include <linux/writeback.h> |
27 | #include <linux/buffer_head.h> | |
a0f06780 | 28 | #include <linux/module.h> |
4f6f0bd2 | 29 | #include <linux/syscalls.h> |
912d35f8 | 30 | #include <linux/uio.h> |
5274f052 | 31 | |
912d35f8 JA |
32 | struct partial_page { |
33 | unsigned int offset; | |
34 | unsigned int len; | |
35 | }; | |
36 | ||
37 | /* | |
00522fb4 | 38 | * Passed to splice_to_pipe |
912d35f8 JA |
39 | */ |
40 | struct splice_pipe_desc { | |
41 | struct page **pages; /* page map */ | |
42 | struct partial_page *partial; /* pages[] may not be contig */ | |
43 | int nr_pages; /* number of pages in map */ | |
44 | unsigned int flags; /* splice flags */ | |
d4c3cca9 | 45 | const struct pipe_buf_operations *ops;/* ops associated with output pipe */ |
912d35f8 JA |
46 | }; |
47 | ||
83f9135b JA |
48 | /* |
49 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
50 | * a vm helper function, it's already simplified quite a bit by the | |
51 | * addition of remove_mapping(). If success is returned, the caller may | |
52 | * attempt to reuse this page for another destination. | |
53 | */ | |
76ad4d11 | 54 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, |
5abc97aa JA |
55 | struct pipe_buffer *buf) |
56 | { | |
57 | struct page *page = buf->page; | |
9e94cd4f | 58 | struct address_space *mapping; |
5abc97aa | 59 | |
9e0267c2 JA |
60 | lock_page(page); |
61 | ||
9e94cd4f JA |
62 | mapping = page_mapping(page); |
63 | if (mapping) { | |
64 | WARN_ON(!PageUptodate(page)); | |
5abc97aa | 65 | |
9e94cd4f JA |
66 | /* |
67 | * At least for ext2 with nobh option, we need to wait on | |
68 | * writeback completing on this page, since we'll remove it | |
69 | * from the pagecache. Otherwise truncate wont wait on the | |
70 | * page, allowing the disk blocks to be reused by someone else | |
71 | * before we actually wrote our data to them. fs corruption | |
72 | * ensues. | |
73 | */ | |
74 | wait_on_page_writeback(page); | |
ad8d6f0a | 75 | |
9e94cd4f | 76 | if (PagePrivate(page)) |
2ae88149 | 77 | try_to_release_page(page, GFP_KERNEL); |
4f6f0bd2 | 78 | |
9e94cd4f JA |
79 | /* |
80 | * If we succeeded in removing the mapping, set LRU flag | |
81 | * and return good. | |
82 | */ | |
83 | if (remove_mapping(mapping, page)) { | |
84 | buf->flags |= PIPE_BUF_FLAG_LRU; | |
85 | return 0; | |
86 | } | |
9e0267c2 | 87 | } |
5abc97aa | 88 | |
9e94cd4f JA |
89 | /* |
90 | * Raced with truncate or failed to remove page from current | |
91 | * address space, unlock and return failure. | |
92 | */ | |
93 | unlock_page(page); | |
94 | return 1; | |
5abc97aa JA |
95 | } |
96 | ||
76ad4d11 | 97 | static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, |
5274f052 JA |
98 | struct pipe_buffer *buf) |
99 | { | |
100 | page_cache_release(buf->page); | |
1432873a | 101 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
5274f052 JA |
102 | } |
103 | ||
76ad4d11 | 104 | static int page_cache_pipe_buf_pin(struct pipe_inode_info *pipe, |
f84d7519 | 105 | struct pipe_buffer *buf) |
5274f052 JA |
106 | { |
107 | struct page *page = buf->page; | |
49d0b21b | 108 | int err; |
5274f052 JA |
109 | |
110 | if (!PageUptodate(page)) { | |
49d0b21b JA |
111 | lock_page(page); |
112 | ||
113 | /* | |
114 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 115 | * splice, if this is the first page. |
49d0b21b JA |
116 | */ |
117 | if (!page->mapping) { | |
118 | err = -ENODATA; | |
119 | goto error; | |
120 | } | |
5274f052 | 121 | |
49d0b21b | 122 | /* |
73d62d83 | 123 | * Uh oh, read-error from disk. |
49d0b21b JA |
124 | */ |
125 | if (!PageUptodate(page)) { | |
126 | err = -EIO; | |
127 | goto error; | |
128 | } | |
129 | ||
130 | /* | |
f84d7519 | 131 | * Page is ok afterall, we are done. |
49d0b21b | 132 | */ |
5274f052 | 133 | unlock_page(page); |
5274f052 JA |
134 | } |
135 | ||
f84d7519 | 136 | return 0; |
49d0b21b JA |
137 | error: |
138 | unlock_page(page); | |
f84d7519 | 139 | return err; |
70524490 JA |
140 | } |
141 | ||
d4c3cca9 | 142 | static const struct pipe_buf_operations page_cache_pipe_buf_ops = { |
5274f052 | 143 | .can_merge = 0, |
f84d7519 JA |
144 | .map = generic_pipe_buf_map, |
145 | .unmap = generic_pipe_buf_unmap, | |
146 | .pin = page_cache_pipe_buf_pin, | |
5274f052 | 147 | .release = page_cache_pipe_buf_release, |
5abc97aa | 148 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 149 | .get = generic_pipe_buf_get, |
5274f052 JA |
150 | }; |
151 | ||
912d35f8 JA |
152 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
153 | struct pipe_buffer *buf) | |
154 | { | |
7afa6fd0 JA |
155 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
156 | return 1; | |
157 | ||
1432873a | 158 | buf->flags |= PIPE_BUF_FLAG_LRU; |
330ab716 | 159 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
160 | } |
161 | ||
d4c3cca9 | 162 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
912d35f8 | 163 | .can_merge = 0, |
f84d7519 JA |
164 | .map = generic_pipe_buf_map, |
165 | .unmap = generic_pipe_buf_unmap, | |
166 | .pin = generic_pipe_buf_pin, | |
912d35f8 JA |
167 | .release = page_cache_pipe_buf_release, |
168 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 169 | .get = generic_pipe_buf_get, |
912d35f8 JA |
170 | }; |
171 | ||
83f9135b JA |
172 | /* |
173 | * Pipe output worker. This sets up our pipe format with the page cache | |
174 | * pipe buffer operations. Otherwise very similar to the regular pipe_writev(). | |
175 | */ | |
00522fb4 JA |
176 | static ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
177 | struct splice_pipe_desc *spd) | |
5274f052 | 178 | { |
00de00bd | 179 | unsigned int spd_pages = spd->nr_pages; |
912d35f8 | 180 | int ret, do_wakeup, page_nr; |
5274f052 JA |
181 | |
182 | ret = 0; | |
183 | do_wakeup = 0; | |
912d35f8 | 184 | page_nr = 0; |
5274f052 | 185 | |
3a326a2c IM |
186 | if (pipe->inode) |
187 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 188 | |
5274f052 | 189 | for (;;) { |
3a326a2c | 190 | if (!pipe->readers) { |
5274f052 JA |
191 | send_sig(SIGPIPE, current, 0); |
192 | if (!ret) | |
193 | ret = -EPIPE; | |
194 | break; | |
195 | } | |
196 | ||
6f767b04 JA |
197 | if (pipe->nrbufs < PIPE_BUFFERS) { |
198 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 199 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 200 | |
912d35f8 JA |
201 | buf->page = spd->pages[page_nr]; |
202 | buf->offset = spd->partial[page_nr].offset; | |
203 | buf->len = spd->partial[page_nr].len; | |
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 | ||
3a326a2c IM |
248 | if (pipe->inode) |
249 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
250 | |
251 | if (do_wakeup) { | |
c0bd1f65 | 252 | smp_mb(); |
3a326a2c IM |
253 | if (waitqueue_active(&pipe->wait)) |
254 | wake_up_interruptible(&pipe->wait); | |
255 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
256 | } |
257 | ||
00de00bd | 258 | while (page_nr < spd_pages) |
912d35f8 | 259 | page_cache_release(spd->pages[page_nr++]); |
5274f052 JA |
260 | |
261 | return ret; | |
262 | } | |
263 | ||
3a326a2c | 264 | static int |
cbb7e577 JA |
265 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
266 | struct pipe_inode_info *pipe, size_t len, | |
267 | unsigned int flags) | |
5274f052 JA |
268 | { |
269 | struct address_space *mapping = in->f_mapping; | |
912d35f8 | 270 | unsigned int loff, nr_pages; |
16c523dd | 271 | struct page *pages[PIPE_BUFFERS]; |
912d35f8 | 272 | struct partial_page partial[PIPE_BUFFERS]; |
5274f052 | 273 | struct page *page; |
91ad66ef JA |
274 | pgoff_t index, end_index; |
275 | loff_t isize; | |
eb20796b | 276 | int error, page_nr; |
912d35f8 JA |
277 | struct splice_pipe_desc spd = { |
278 | .pages = pages, | |
279 | .partial = partial, | |
280 | .flags = flags, | |
281 | .ops = &page_cache_pipe_buf_ops, | |
282 | }; | |
5274f052 | 283 | |
cbb7e577 | 284 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 JA |
285 | loff = *ppos & ~PAGE_CACHE_MASK; |
286 | nr_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
5274f052 JA |
287 | |
288 | if (nr_pages > PIPE_BUFFERS) | |
289 | nr_pages = PIPE_BUFFERS; | |
290 | ||
291 | /* | |
86aa5ac5 JA |
292 | * Don't try to 2nd guess the read-ahead logic, call into |
293 | * page_cache_readahead() like the page cache reads would do. | |
5274f052 | 294 | */ |
86aa5ac5 | 295 | page_cache_readahead(mapping, &in->f_ra, in, index, nr_pages); |
5274f052 | 296 | |
5274f052 | 297 | /* |
73d62d83 | 298 | * Now fill in the holes: |
5274f052 | 299 | */ |
7480a904 | 300 | error = 0; |
82aa5d61 | 301 | |
eb20796b JA |
302 | /* |
303 | * Lookup the (hopefully) full range of pages we need. | |
304 | */ | |
305 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
82aa5d61 | 306 | |
eb20796b JA |
307 | /* |
308 | * If find_get_pages_contig() returned fewer pages than we needed, | |
309 | * allocate the rest. | |
310 | */ | |
311 | index += spd.nr_pages; | |
312 | while (spd.nr_pages < nr_pages) { | |
82aa5d61 | 313 | /* |
eb20796b JA |
314 | * Page could be there, find_get_pages_contig() breaks on |
315 | * the first hole. | |
5274f052 | 316 | */ |
7480a904 JA |
317 | page = find_get_page(mapping, index); |
318 | if (!page) { | |
e27dedd8 JA |
319 | /* |
320 | * Make sure the read-ahead engine is notified | |
321 | * about this failure. | |
322 | */ | |
323 | handle_ra_miss(mapping, &in->f_ra, index); | |
324 | ||
7480a904 | 325 | /* |
eb20796b | 326 | * page didn't exist, allocate one. |
7480a904 JA |
327 | */ |
328 | page = page_cache_alloc_cold(mapping); | |
329 | if (!page) | |
330 | break; | |
331 | ||
332 | error = add_to_page_cache_lru(page, mapping, index, | |
2ae88149 | 333 | GFP_KERNEL); |
7480a904 JA |
334 | if (unlikely(error)) { |
335 | page_cache_release(page); | |
a0548871 JA |
336 | if (error == -EEXIST) |
337 | continue; | |
7480a904 JA |
338 | break; |
339 | } | |
eb20796b JA |
340 | /* |
341 | * add_to_page_cache() locks the page, unlock it | |
342 | * to avoid convoluting the logic below even more. | |
343 | */ | |
344 | unlock_page(page); | |
7480a904 JA |
345 | } |
346 | ||
eb20796b JA |
347 | pages[spd.nr_pages++] = page; |
348 | index++; | |
349 | } | |
350 | ||
351 | /* | |
352 | * Now loop over the map and see if we need to start IO on any | |
353 | * pages, fill in the partial map, etc. | |
354 | */ | |
355 | index = *ppos >> PAGE_CACHE_SHIFT; | |
356 | nr_pages = spd.nr_pages; | |
357 | spd.nr_pages = 0; | |
358 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
359 | unsigned int this_len; | |
360 | ||
361 | if (!len) | |
362 | break; | |
363 | ||
364 | /* | |
365 | * this_len is the max we'll use from this page | |
366 | */ | |
367 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
368 | page = pages[page_nr]; | |
369 | ||
7480a904 JA |
370 | /* |
371 | * If the page isn't uptodate, we may need to start io on it | |
372 | */ | |
373 | if (!PageUptodate(page)) { | |
c4f895cb JA |
374 | /* |
375 | * If in nonblock mode then dont block on waiting | |
376 | * for an in-flight io page | |
377 | */ | |
9ae9d68c FW |
378 | if (flags & SPLICE_F_NONBLOCK) { |
379 | if (TestSetPageLocked(page)) | |
380 | break; | |
381 | } else | |
382 | lock_page(page); | |
7480a904 JA |
383 | |
384 | /* | |
385 | * page was truncated, stop here. if this isn't the | |
386 | * first page, we'll just complete what we already | |
387 | * added | |
388 | */ | |
389 | if (!page->mapping) { | |
390 | unlock_page(page); | |
7480a904 JA |
391 | break; |
392 | } | |
393 | /* | |
394 | * page was already under io and is now done, great | |
395 | */ | |
396 | if (PageUptodate(page)) { | |
397 | unlock_page(page); | |
398 | goto fill_it; | |
399 | } | |
5274f052 | 400 | |
7480a904 JA |
401 | /* |
402 | * need to read in the page | |
403 | */ | |
404 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 405 | if (unlikely(error)) { |
eb20796b JA |
406 | /* |
407 | * We really should re-lookup the page here, | |
408 | * but it complicates things a lot. Instead | |
409 | * lets just do what we already stored, and | |
410 | * we'll get it the next time we are called. | |
411 | */ | |
7480a904 | 412 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
413 | error = 0; |
414 | ||
5274f052 JA |
415 | break; |
416 | } | |
620a324b JA |
417 | } |
418 | fill_it: | |
419 | /* | |
420 | * i_size must be checked after PageUptodate. | |
421 | */ | |
422 | isize = i_size_read(mapping->host); | |
423 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
424 | if (unlikely(!isize || index > end_index)) | |
425 | break; | |
426 | ||
427 | /* | |
428 | * if this is the last page, see if we need to shrink | |
429 | * the length and stop | |
430 | */ | |
431 | if (end_index == index) { | |
432 | unsigned int plen; | |
91ad66ef JA |
433 | |
434 | /* | |
620a324b | 435 | * max good bytes in this page |
91ad66ef | 436 | */ |
620a324b JA |
437 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
438 | if (plen <= loff) | |
91ad66ef | 439 | break; |
91ad66ef JA |
440 | |
441 | /* | |
620a324b | 442 | * force quit after adding this page |
91ad66ef | 443 | */ |
620a324b JA |
444 | this_len = min(this_len, plen - loff); |
445 | len = this_len; | |
5274f052 | 446 | } |
620a324b | 447 | |
eb20796b JA |
448 | partial[page_nr].offset = loff; |
449 | partial[page_nr].len = this_len; | |
82aa5d61 | 450 | len -= this_len; |
91ad66ef | 451 | loff = 0; |
eb20796b JA |
452 | spd.nr_pages++; |
453 | index++; | |
5274f052 JA |
454 | } |
455 | ||
eb20796b | 456 | /* |
475ecade | 457 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
458 | * we got, 'nr_pages' is how many pages are in the map. |
459 | */ | |
460 | while (page_nr < nr_pages) | |
461 | page_cache_release(pages[page_nr++]); | |
462 | ||
912d35f8 | 463 | if (spd.nr_pages) |
00522fb4 | 464 | return splice_to_pipe(pipe, &spd); |
5274f052 | 465 | |
7480a904 | 466 | return error; |
5274f052 JA |
467 | } |
468 | ||
83f9135b JA |
469 | /** |
470 | * generic_file_splice_read - splice data from file to a pipe | |
471 | * @in: file to splice from | |
472 | * @pipe: pipe to splice to | |
473 | * @len: number of bytes to splice | |
474 | * @flags: splice modifier flags | |
475 | * | |
476 | * Will read pages from given file and fill them into a pipe. | |
83f9135b | 477 | */ |
cbb7e577 JA |
478 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
479 | struct pipe_inode_info *pipe, size_t len, | |
480 | unsigned int flags) | |
5274f052 JA |
481 | { |
482 | ssize_t spliced; | |
483 | int ret; | |
d366d398 JA |
484 | loff_t isize, left; |
485 | ||
486 | isize = i_size_read(in->f_mapping->host); | |
487 | if (unlikely(*ppos >= isize)) | |
488 | return 0; | |
489 | ||
490 | left = isize - *ppos; | |
491 | if (unlikely(left < len)) | |
492 | len = left; | |
5274f052 JA |
493 | |
494 | ret = 0; | |
495 | spliced = 0; | |
496 | while (len) { | |
cbb7e577 | 497 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 498 | |
c4f895cb | 499 | if (ret < 0) |
5274f052 | 500 | break; |
c4f895cb JA |
501 | else if (!ret) { |
502 | if (spliced) | |
503 | break; | |
504 | if (flags & SPLICE_F_NONBLOCK) { | |
505 | ret = -EAGAIN; | |
506 | break; | |
507 | } | |
508 | } | |
5274f052 | 509 | |
cbb7e577 | 510 | *ppos += ret; |
5274f052 JA |
511 | len -= ret; |
512 | spliced += ret; | |
513 | } | |
514 | ||
515 | if (spliced) | |
516 | return spliced; | |
517 | ||
518 | return ret; | |
519 | } | |
520 | ||
059a8f37 JA |
521 | EXPORT_SYMBOL(generic_file_splice_read); |
522 | ||
5274f052 | 523 | /* |
4f6f0bd2 | 524 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 525 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 526 | */ |
76ad4d11 | 527 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
528 | struct pipe_buffer *buf, struct splice_desc *sd) |
529 | { | |
530 | struct file *file = sd->file; | |
531 | loff_t pos = sd->pos; | |
f84d7519 | 532 | int ret, more; |
5274f052 | 533 | |
76ad4d11 | 534 | ret = buf->ops->pin(pipe, buf); |
f84d7519 JA |
535 | if (!ret) { |
536 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 537 | |
f84d7519 JA |
538 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
539 | sd->len, &pos, more); | |
540 | } | |
5274f052 | 541 | |
016b661e | 542 | return ret; |
5274f052 JA |
543 | } |
544 | ||
545 | /* | |
546 | * This is a little more tricky than the file -> pipe splicing. There are | |
547 | * basically three cases: | |
548 | * | |
549 | * - Destination page already exists in the address space and there | |
550 | * are users of it. For that case we have no other option that | |
551 | * copying the data. Tough luck. | |
552 | * - Destination page already exists in the address space, but there | |
553 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
554 | * through to last case. | |
555 | * - Destination page does not exist, we can add the pipe page to | |
556 | * the page cache and avoid the copy. | |
557 | * | |
83f9135b JA |
558 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
559 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
560 | * file address space page cache. This is possible if no one else has | |
561 | * the pipe page referenced outside of the pipe and page cache. If | |
562 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
563 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 564 | */ |
76ad4d11 | 565 | static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
5274f052 JA |
566 | struct splice_desc *sd) |
567 | { | |
568 | struct file *file = sd->file; | |
569 | struct address_space *mapping = file->f_mapping; | |
016b661e | 570 | unsigned int offset, this_len; |
5274f052 | 571 | struct page *page; |
5274f052 | 572 | pgoff_t index; |
3e7ee3e7 | 573 | int ret; |
5274f052 JA |
574 | |
575 | /* | |
49d0b21b | 576 | * make sure the data in this buffer is uptodate |
5274f052 | 577 | */ |
76ad4d11 | 578 | ret = buf->ops->pin(pipe, buf); |
f84d7519 JA |
579 | if (unlikely(ret)) |
580 | return ret; | |
5274f052 JA |
581 | |
582 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
583 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
584 | ||
016b661e JA |
585 | this_len = sd->len; |
586 | if (this_len + offset > PAGE_CACHE_SIZE) | |
587 | this_len = PAGE_CACHE_SIZE - offset; | |
588 | ||
485ddb4b NP |
589 | find_page: |
590 | page = find_lock_page(mapping, index); | |
591 | if (!page) { | |
592 | ret = -ENOMEM; | |
593 | page = page_cache_alloc_cold(mapping); | |
594 | if (unlikely(!page)) | |
595 | goto out_ret; | |
596 | ||
83f9135b | 597 | /* |
485ddb4b | 598 | * This will also lock the page |
83f9135b | 599 | */ |
485ddb4b NP |
600 | ret = add_to_page_cache_lru(page, mapping, index, |
601 | GFP_KERNEL); | |
602 | if (unlikely(ret)) | |
603 | goto out; | |
604 | } | |
9e0267c2 | 605 | |
016b661e | 606 | ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); |
bfc4ee39 JA |
607 | if (unlikely(ret)) { |
608 | loff_t isize = i_size_read(mapping->host); | |
609 | ||
610 | if (ret != AOP_TRUNCATED_PAGE) | |
611 | unlock_page(page); | |
4f6f0bd2 | 612 | page_cache_release(page); |
bfc4ee39 JA |
613 | if (ret == AOP_TRUNCATED_PAGE) |
614 | goto find_page; | |
615 | ||
616 | /* | |
617 | * prepare_write() may have instantiated a few blocks | |
618 | * outside i_size. Trim these off again. | |
619 | */ | |
620 | if (sd->pos + this_len > isize) | |
621 | vmtruncate(mapping->host, isize); | |
622 | ||
e6e80f29 | 623 | goto out_ret; |
bfc4ee39 | 624 | } |
5274f052 | 625 | |
0568b409 | 626 | if (buf->page != page) { |
f84d7519 JA |
627 | /* |
628 | * Careful, ->map() uses KM_USER0! | |
629 | */ | |
76ad4d11 | 630 | char *src = buf->ops->map(pipe, buf, 1); |
f84d7519 | 631 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 632 | |
016b661e | 633 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 634 | flush_dcache_page(page); |
f84d7519 | 635 | kunmap_atomic(dst, KM_USER1); |
76ad4d11 | 636 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 637 | } |
5274f052 | 638 | |
016b661e | 639 | ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); |
d9993c37 DM |
640 | if (ret) { |
641 | if (ret == AOP_TRUNCATED_PAGE) { | |
642 | page_cache_release(page); | |
643 | goto find_page; | |
644 | } | |
645 | if (ret < 0) | |
646 | goto out; | |
0568b409 | 647 | /* |
d9993c37 DM |
648 | * Partial write has happened, so 'ret' already initialized by |
649 | * number of bytes written, Where is nothing we have to do here. | |
0568b409 | 650 | */ |
d9993c37 | 651 | } else |
0568b409 | 652 | ret = this_len; |
d9993c37 DM |
653 | /* |
654 | * Return the number of bytes written and mark page as | |
655 | * accessed, we are now done! | |
656 | */ | |
657 | mark_page_accessed(page); | |
5274f052 | 658 | out: |
0568b409 | 659 | page_cache_release(page); |
9e0267c2 | 660 | unlock_page(page); |
e6e80f29 | 661 | out_ret: |
5274f052 JA |
662 | return ret; |
663 | } | |
664 | ||
83f9135b JA |
665 | /* |
666 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
667 | * key here is the 'actor' worker passed in that actually moves the data | |
668 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
669 | */ | |
40bee44e MF |
670 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, |
671 | struct file *out, loff_t *ppos, size_t len, | |
672 | unsigned int flags, splice_actor *actor) | |
5274f052 | 673 | { |
5274f052 JA |
674 | int ret, do_wakeup, err; |
675 | struct splice_desc sd; | |
676 | ||
677 | ret = 0; | |
678 | do_wakeup = 0; | |
679 | ||
680 | sd.total_len = len; | |
681 | sd.flags = flags; | |
682 | sd.file = out; | |
cbb7e577 | 683 | sd.pos = *ppos; |
5274f052 | 684 | |
5274f052 | 685 | for (;;) { |
6f767b04 JA |
686 | if (pipe->nrbufs) { |
687 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
d4c3cca9 | 688 | const struct pipe_buf_operations *ops = buf->ops; |
5274f052 JA |
689 | |
690 | sd.len = buf->len; | |
691 | if (sd.len > sd.total_len) | |
692 | sd.len = sd.total_len; | |
693 | ||
3a326a2c | 694 | err = actor(pipe, buf, &sd); |
016b661e | 695 | if (err <= 0) { |
5274f052 JA |
696 | if (!ret && err != -ENODATA) |
697 | ret = err; | |
698 | ||
699 | break; | |
700 | } | |
701 | ||
016b661e JA |
702 | ret += err; |
703 | buf->offset += err; | |
704 | buf->len -= err; | |
705 | ||
706 | sd.len -= err; | |
707 | sd.pos += err; | |
708 | sd.total_len -= err; | |
709 | if (sd.len) | |
710 | continue; | |
73d62d83 | 711 | |
5274f052 JA |
712 | if (!buf->len) { |
713 | buf->ops = NULL; | |
3a326a2c | 714 | ops->release(pipe, buf); |
6f767b04 JA |
715 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
716 | pipe->nrbufs--; | |
717 | if (pipe->inode) | |
718 | do_wakeup = 1; | |
5274f052 JA |
719 | } |
720 | ||
5274f052 JA |
721 | if (!sd.total_len) |
722 | break; | |
723 | } | |
724 | ||
6f767b04 | 725 | if (pipe->nrbufs) |
5274f052 | 726 | continue; |
3a326a2c | 727 | if (!pipe->writers) |
5274f052 | 728 | break; |
3a326a2c | 729 | if (!pipe->waiting_writers) { |
5274f052 JA |
730 | if (ret) |
731 | break; | |
732 | } | |
733 | ||
29e35094 LT |
734 | if (flags & SPLICE_F_NONBLOCK) { |
735 | if (!ret) | |
736 | ret = -EAGAIN; | |
737 | break; | |
738 | } | |
739 | ||
5274f052 JA |
740 | if (signal_pending(current)) { |
741 | if (!ret) | |
742 | ret = -ERESTARTSYS; | |
743 | break; | |
744 | } | |
745 | ||
746 | if (do_wakeup) { | |
c0bd1f65 | 747 | smp_mb(); |
3a326a2c IM |
748 | if (waitqueue_active(&pipe->wait)) |
749 | wake_up_interruptible_sync(&pipe->wait); | |
750 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
751 | do_wakeup = 0; |
752 | } | |
753 | ||
3a326a2c | 754 | pipe_wait(pipe); |
5274f052 JA |
755 | } |
756 | ||
5274f052 | 757 | if (do_wakeup) { |
c0bd1f65 | 758 | smp_mb(); |
3a326a2c IM |
759 | if (waitqueue_active(&pipe->wait)) |
760 | wake_up_interruptible(&pipe->wait); | |
761 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
762 | } |
763 | ||
5274f052 | 764 | return ret; |
5274f052 | 765 | } |
40bee44e | 766 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 767 | |
6da61809 MF |
768 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
769 | loff_t *ppos, size_t len, unsigned int flags, | |
770 | splice_actor *actor) | |
771 | { | |
772 | ssize_t ret; | |
773 | struct inode *inode = out->f_mapping->host; | |
774 | ||
775 | /* | |
776 | * The actor worker might be calling ->prepare_write and | |
777 | * ->commit_write. Most of the time, these expect i_mutex to | |
778 | * be held. Since this may result in an ABBA deadlock with | |
779 | * pipe->inode, we have to order lock acquiry here. | |
780 | */ | |
781 | inode_double_lock(inode, pipe->inode); | |
782 | ret = __splice_from_pipe(pipe, out, ppos, len, flags, actor); | |
783 | inode_double_unlock(inode, pipe->inode); | |
784 | ||
785 | return ret; | |
786 | } | |
787 | ||
788 | /** | |
789 | * generic_file_splice_write_nolock - generic_file_splice_write without mutexes | |
790 | * @pipe: pipe info | |
791 | * @out: file to write to | |
792 | * @len: number of bytes to splice | |
793 | * @flags: splice modifier flags | |
794 | * | |
795 | * Will either move or copy pages (determined by @flags options) from | |
796 | * the given pipe inode to the given file. The caller is responsible | |
797 | * for acquiring i_mutex on both inodes. | |
798 | * | |
799 | */ | |
800 | ssize_t | |
801 | generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out, | |
802 | loff_t *ppos, size_t len, unsigned int flags) | |
803 | { | |
804 | struct address_space *mapping = out->f_mapping; | |
805 | struct inode *inode = mapping->host; | |
806 | ssize_t ret; | |
807 | int err; | |
808 | ||
0f7fc9e4 | 809 | err = remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
810 | if (unlikely(err)) |
811 | return err; | |
812 | ||
6da61809 MF |
813 | ret = __splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
814 | if (ret > 0) { | |
17ee4f49 JA |
815 | unsigned long nr_pages; |
816 | ||
6da61809 | 817 | *ppos += ret; |
17ee4f49 | 818 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
6da61809 MF |
819 | |
820 | /* | |
821 | * If file or inode is SYNC and we actually wrote some data, | |
822 | * sync it. | |
823 | */ | |
824 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
825 | err = generic_osync_inode(inode, mapping, | |
826 | OSYNC_METADATA|OSYNC_DATA); | |
827 | ||
828 | if (err) | |
829 | ret = err; | |
830 | } | |
17ee4f49 | 831 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
6da61809 MF |
832 | } |
833 | ||
834 | return ret; | |
835 | } | |
836 | ||
837 | EXPORT_SYMBOL(generic_file_splice_write_nolock); | |
838 | ||
83f9135b JA |
839 | /** |
840 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 841 | * @pipe: pipe info |
83f9135b JA |
842 | * @out: file to write to |
843 | * @len: number of bytes to splice | |
844 | * @flags: splice modifier flags | |
845 | * | |
846 | * Will either move or copy pages (determined by @flags options) from | |
847 | * the given pipe inode to the given file. | |
848 | * | |
849 | */ | |
3a326a2c IM |
850 | ssize_t |
851 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 852 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 853 | { |
4f6f0bd2 | 854 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 855 | struct inode *inode = mapping->host; |
3a326a2c | 856 | ssize_t ret; |
8c34e2d6 JA |
857 | int err; |
858 | ||
0f7fc9e4 | 859 | err = should_remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
860 | if (unlikely(err)) { |
861 | mutex_lock(&inode->i_mutex); | |
0f7fc9e4 | 862 | err = __remove_suid(out->f_path.dentry, err); |
8c34e2d6 JA |
863 | mutex_unlock(&inode->i_mutex); |
864 | if (err) | |
865 | return err; | |
866 | } | |
3a326a2c | 867 | |
00522fb4 | 868 | ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
a4514ebd | 869 | if (ret > 0) { |
17ee4f49 JA |
870 | unsigned long nr_pages; |
871 | ||
a4514ebd | 872 | *ppos += ret; |
17ee4f49 | 873 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
a4514ebd JA |
874 | |
875 | /* | |
876 | * If file or inode is SYNC and we actually wrote some data, | |
877 | * sync it. | |
878 | */ | |
879 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
a4514ebd JA |
880 | mutex_lock(&inode->i_mutex); |
881 | err = generic_osync_inode(inode, mapping, | |
882 | OSYNC_METADATA|OSYNC_DATA); | |
883 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 884 | |
a4514ebd JA |
885 | if (err) |
886 | ret = err; | |
887 | } | |
17ee4f49 | 888 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
4f6f0bd2 JA |
889 | } |
890 | ||
891 | return ret; | |
5274f052 JA |
892 | } |
893 | ||
059a8f37 JA |
894 | EXPORT_SYMBOL(generic_file_splice_write); |
895 | ||
83f9135b JA |
896 | /** |
897 | * generic_splice_sendpage - splice data from a pipe to a socket | |
898 | * @inode: pipe inode | |
899 | * @out: socket to write to | |
900 | * @len: number of bytes to splice | |
901 | * @flags: splice modifier flags | |
902 | * | |
903 | * Will send @len bytes from the pipe to a network socket. No data copying | |
904 | * is involved. | |
905 | * | |
906 | */ | |
3a326a2c | 907 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 908 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 909 | { |
00522fb4 | 910 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
911 | } |
912 | ||
059a8f37 | 913 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 914 | |
83f9135b JA |
915 | /* |
916 | * Attempt to initiate a splice from pipe to file. | |
917 | */ | |
3a326a2c | 918 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 919 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 920 | { |
5274f052 JA |
921 | int ret; |
922 | ||
49570e9b | 923 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
924 | return -EINVAL; |
925 | ||
49570e9b | 926 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
927 | return -EBADF; |
928 | ||
cbb7e577 | 929 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
930 | if (unlikely(ret < 0)) |
931 | return ret; | |
932 | ||
cbb7e577 | 933 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
934 | } |
935 | ||
83f9135b JA |
936 | /* |
937 | * Attempt to initiate a splice from a file to a pipe. | |
938 | */ | |
cbb7e577 JA |
939 | static long do_splice_to(struct file *in, loff_t *ppos, |
940 | struct pipe_inode_info *pipe, size_t len, | |
941 | unsigned int flags) | |
5274f052 | 942 | { |
5274f052 JA |
943 | int ret; |
944 | ||
49570e9b | 945 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
946 | return -EINVAL; |
947 | ||
49570e9b | 948 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
949 | return -EBADF; |
950 | ||
cbb7e577 | 951 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
952 | if (unlikely(ret < 0)) |
953 | return ret; | |
954 | ||
cbb7e577 | 955 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
956 | } |
957 | ||
cbb7e577 JA |
958 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
959 | size_t len, unsigned int flags) | |
b92ce558 JA |
960 | { |
961 | struct pipe_inode_info *pipe; | |
962 | long ret, bytes; | |
cbb7e577 | 963 | loff_t out_off; |
b92ce558 JA |
964 | umode_t i_mode; |
965 | int i; | |
966 | ||
967 | /* | |
968 | * We require the input being a regular file, as we don't want to | |
969 | * randomly drop data for eg socket -> socket splicing. Use the | |
970 | * piped splicing for that! | |
971 | */ | |
0f7fc9e4 | 972 | i_mode = in->f_path.dentry->d_inode->i_mode; |
b92ce558 JA |
973 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
974 | return -EINVAL; | |
975 | ||
976 | /* | |
977 | * neither in nor out is a pipe, setup an internal pipe attached to | |
978 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
979 | */ | |
980 | pipe = current->splice_pipe; | |
49570e9b | 981 | if (unlikely(!pipe)) { |
b92ce558 JA |
982 | pipe = alloc_pipe_info(NULL); |
983 | if (!pipe) | |
984 | return -ENOMEM; | |
985 | ||
986 | /* | |
987 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 988 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
989 | * PIPE_READERS appropriately. |
990 | */ | |
991 | pipe->readers = 1; | |
992 | ||
993 | current->splice_pipe = pipe; | |
994 | } | |
995 | ||
996 | /* | |
73d62d83 | 997 | * Do the splice. |
b92ce558 JA |
998 | */ |
999 | ret = 0; | |
1000 | bytes = 0; | |
cbb7e577 | 1001 | out_off = 0; |
b92ce558 JA |
1002 | |
1003 | while (len) { | |
1004 | size_t read_len, max_read_len; | |
1005 | ||
1006 | /* | |
1007 | * Do at most PIPE_BUFFERS pages worth of transfer: | |
1008 | */ | |
1009 | max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE)); | |
1010 | ||
cbb7e577 | 1011 | ret = do_splice_to(in, ppos, pipe, max_read_len, flags); |
b92ce558 JA |
1012 | if (unlikely(ret < 0)) |
1013 | goto out_release; | |
1014 | ||
1015 | read_len = ret; | |
1016 | ||
1017 | /* | |
1018 | * NOTE: nonblocking mode only applies to the input. We | |
1019 | * must not do the output in nonblocking mode as then we | |
1020 | * could get stuck data in the internal pipe: | |
1021 | */ | |
cbb7e577 | 1022 | ret = do_splice_from(pipe, out, &out_off, read_len, |
b92ce558 JA |
1023 | flags & ~SPLICE_F_NONBLOCK); |
1024 | if (unlikely(ret < 0)) | |
1025 | goto out_release; | |
1026 | ||
1027 | bytes += ret; | |
1028 | len -= ret; | |
1029 | ||
1030 | /* | |
1031 | * In nonblocking mode, if we got back a short read then | |
1032 | * that was due to either an IO error or due to the | |
1033 | * pagecache entry not being there. In the IO error case | |
1034 | * the _next_ splice attempt will produce a clean IO error | |
1035 | * return value (not a short read), so in both cases it's | |
1036 | * correct to break out of the loop here: | |
1037 | */ | |
1038 | if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len)) | |
1039 | break; | |
1040 | } | |
1041 | ||
1042 | pipe->nrbufs = pipe->curbuf = 0; | |
1043 | ||
1044 | return bytes; | |
1045 | ||
1046 | out_release: | |
1047 | /* | |
1048 | * If we did an incomplete transfer we must release | |
1049 | * the pipe buffers in question: | |
1050 | */ | |
1051 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1052 | struct pipe_buffer *buf = pipe->bufs + i; | |
1053 | ||
1054 | if (buf->ops) { | |
1055 | buf->ops->release(pipe, buf); | |
1056 | buf->ops = NULL; | |
1057 | } | |
1058 | } | |
1059 | pipe->nrbufs = pipe->curbuf = 0; | |
1060 | ||
1061 | /* | |
1062 | * If we transferred some data, return the number of bytes: | |
1063 | */ | |
1064 | if (bytes > 0) | |
1065 | return bytes; | |
1066 | ||
1067 | return ret; | |
1068 | } | |
1069 | ||
ddac0d39 JA |
1070 | /* |
1071 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1072 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1073 | * pipe. | |
1074 | */ | |
1075 | static inline struct pipe_inode_info *pipe_info(struct inode *inode) | |
1076 | { | |
1077 | if (S_ISFIFO(inode->i_mode)) | |
1078 | return inode->i_pipe; | |
1079 | ||
1080 | return NULL; | |
1081 | } | |
1082 | ||
83f9135b JA |
1083 | /* |
1084 | * Determine where to splice to/from. | |
1085 | */ | |
529565dc IM |
1086 | static long do_splice(struct file *in, loff_t __user *off_in, |
1087 | struct file *out, loff_t __user *off_out, | |
1088 | size_t len, unsigned int flags) | |
5274f052 | 1089 | { |
3a326a2c | 1090 | struct pipe_inode_info *pipe; |
cbb7e577 | 1091 | loff_t offset, *off; |
a4514ebd | 1092 | long ret; |
5274f052 | 1093 | |
0f7fc9e4 | 1094 | pipe = pipe_info(in->f_path.dentry->d_inode); |
529565dc IM |
1095 | if (pipe) { |
1096 | if (off_in) | |
1097 | return -ESPIPE; | |
b92ce558 JA |
1098 | if (off_out) { |
1099 | if (out->f_op->llseek == no_llseek) | |
1100 | return -EINVAL; | |
cbb7e577 | 1101 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1102 | return -EFAULT; |
cbb7e577 JA |
1103 | off = &offset; |
1104 | } else | |
1105 | off = &out->f_pos; | |
529565dc | 1106 | |
a4514ebd JA |
1107 | ret = do_splice_from(pipe, out, off, len, flags); |
1108 | ||
1109 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1110 | ret = -EFAULT; | |
1111 | ||
1112 | return ret; | |
529565dc | 1113 | } |
5274f052 | 1114 | |
0f7fc9e4 | 1115 | pipe = pipe_info(out->f_path.dentry->d_inode); |
529565dc IM |
1116 | if (pipe) { |
1117 | if (off_out) | |
1118 | return -ESPIPE; | |
b92ce558 JA |
1119 | if (off_in) { |
1120 | if (in->f_op->llseek == no_llseek) | |
1121 | return -EINVAL; | |
cbb7e577 | 1122 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1123 | return -EFAULT; |
cbb7e577 JA |
1124 | off = &offset; |
1125 | } else | |
1126 | off = &in->f_pos; | |
529565dc | 1127 | |
a4514ebd JA |
1128 | ret = do_splice_to(in, off, pipe, len, flags); |
1129 | ||
1130 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1131 | ret = -EFAULT; | |
1132 | ||
1133 | return ret; | |
529565dc | 1134 | } |
5274f052 JA |
1135 | |
1136 | return -EINVAL; | |
1137 | } | |
1138 | ||
912d35f8 JA |
1139 | /* |
1140 | * Map an iov into an array of pages and offset/length tupples. With the | |
1141 | * partial_page structure, we can map several non-contiguous ranges into | |
1142 | * our ones pages[] map instead of splitting that operation into pieces. | |
1143 | * Could easily be exported as a generic helper for other users, in which | |
1144 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1145 | */ | |
1146 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1147 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1148 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1149 | { |
1150 | int buffers = 0, error = 0; | |
1151 | ||
1152 | /* | |
1153 | * It's ok to take the mmap_sem for reading, even | |
1154 | * across a "get_user()". | |
1155 | */ | |
1156 | down_read(¤t->mm->mmap_sem); | |
1157 | ||
1158 | while (nr_vecs) { | |
1159 | unsigned long off, npages; | |
1160 | void __user *base; | |
1161 | size_t len; | |
1162 | int i; | |
1163 | ||
1164 | /* | |
1165 | * Get user address base and length for this iovec. | |
1166 | */ | |
1167 | error = get_user(base, &iov->iov_base); | |
1168 | if (unlikely(error)) | |
1169 | break; | |
1170 | error = get_user(len, &iov->iov_len); | |
1171 | if (unlikely(error)) | |
1172 | break; | |
1173 | ||
1174 | /* | |
1175 | * Sanity check this iovec. 0 read succeeds. | |
1176 | */ | |
1177 | if (unlikely(!len)) | |
1178 | break; | |
1179 | error = -EFAULT; | |
1180 | if (unlikely(!base)) | |
1181 | break; | |
1182 | ||
1183 | /* | |
1184 | * Get this base offset and number of pages, then map | |
1185 | * in the user pages. | |
1186 | */ | |
1187 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1188 | |
1189 | /* | |
1190 | * If asked for alignment, the offset must be zero and the | |
1191 | * length a multiple of the PAGE_SIZE. | |
1192 | */ | |
1193 | error = -EINVAL; | |
1194 | if (aligned && (off || len & ~PAGE_MASK)) | |
1195 | break; | |
1196 | ||
912d35f8 JA |
1197 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1198 | if (npages > PIPE_BUFFERS - buffers) | |
1199 | npages = PIPE_BUFFERS - buffers; | |
1200 | ||
1201 | error = get_user_pages(current, current->mm, | |
1202 | (unsigned long) base, npages, 0, 0, | |
1203 | &pages[buffers], NULL); | |
1204 | ||
1205 | if (unlikely(error <= 0)) | |
1206 | break; | |
1207 | ||
1208 | /* | |
1209 | * Fill this contiguous range into the partial page map. | |
1210 | */ | |
1211 | for (i = 0; i < error; i++) { | |
7591489a | 1212 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1213 | |
1214 | partial[buffers].offset = off; | |
1215 | partial[buffers].len = plen; | |
1216 | ||
1217 | off = 0; | |
1218 | len -= plen; | |
1219 | buffers++; | |
1220 | } | |
1221 | ||
1222 | /* | |
1223 | * We didn't complete this iov, stop here since it probably | |
1224 | * means we have to move some of this into a pipe to | |
1225 | * be able to continue. | |
1226 | */ | |
1227 | if (len) | |
1228 | break; | |
1229 | ||
1230 | /* | |
1231 | * Don't continue if we mapped fewer pages than we asked for, | |
1232 | * or if we mapped the max number of pages that we have | |
1233 | * room for. | |
1234 | */ | |
1235 | if (error < npages || buffers == PIPE_BUFFERS) | |
1236 | break; | |
1237 | ||
1238 | nr_vecs--; | |
1239 | iov++; | |
1240 | } | |
1241 | ||
1242 | up_read(¤t->mm->mmap_sem); | |
1243 | ||
1244 | if (buffers) | |
1245 | return buffers; | |
1246 | ||
1247 | return error; | |
1248 | } | |
1249 | ||
1250 | /* | |
1251 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1252 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1253 | * to file). In both cases the output is a pipe, naturally. | |
1254 | * | |
1255 | * Note that vmsplice only supports splicing _from_ user memory to a pipe, | |
1256 | * not the other way around. Splicing from user memory is a simple operation | |
1257 | * that can be supported without any funky alignment restrictions or nasty | |
1258 | * vm tricks. We simply map in the user memory and fill them into a pipe. | |
1259 | * The reverse isn't quite as easy, though. There are two possible solutions | |
1260 | * for that: | |
1261 | * | |
1262 | * - memcpy() the data internally, at which point we might as well just | |
1263 | * do a regular read() on the buffer anyway. | |
1264 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1265 | * has restriction limitations on both ends of the pipe). | |
1266 | * | |
1267 | * Alas, it isn't here. | |
1268 | * | |
1269 | */ | |
1270 | static long do_vmsplice(struct file *file, const struct iovec __user *iov, | |
1271 | unsigned long nr_segs, unsigned int flags) | |
1272 | { | |
ddac0d39 | 1273 | struct pipe_inode_info *pipe; |
912d35f8 JA |
1274 | struct page *pages[PIPE_BUFFERS]; |
1275 | struct partial_page partial[PIPE_BUFFERS]; | |
1276 | struct splice_pipe_desc spd = { | |
1277 | .pages = pages, | |
1278 | .partial = partial, | |
1279 | .flags = flags, | |
1280 | .ops = &user_page_pipe_buf_ops, | |
1281 | }; | |
1282 | ||
0f7fc9e4 | 1283 | pipe = pipe_info(file->f_path.dentry->d_inode); |
ddac0d39 | 1284 | if (!pipe) |
912d35f8 JA |
1285 | return -EBADF; |
1286 | if (unlikely(nr_segs > UIO_MAXIOV)) | |
1287 | return -EINVAL; | |
1288 | else if (unlikely(!nr_segs)) | |
1289 | return 0; | |
1290 | ||
7afa6fd0 JA |
1291 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1292 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1293 | if (spd.nr_pages <= 0) |
1294 | return spd.nr_pages; | |
1295 | ||
00522fb4 | 1296 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1297 | } |
1298 | ||
1299 | asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, | |
1300 | unsigned long nr_segs, unsigned int flags) | |
1301 | { | |
1302 | struct file *file; | |
1303 | long error; | |
1304 | int fput; | |
1305 | ||
1306 | error = -EBADF; | |
1307 | file = fget_light(fd, &fput); | |
1308 | if (file) { | |
1309 | if (file->f_mode & FMODE_WRITE) | |
1310 | error = do_vmsplice(file, iov, nr_segs, flags); | |
1311 | ||
1312 | fput_light(file, fput); | |
1313 | } | |
1314 | ||
1315 | return error; | |
1316 | } | |
1317 | ||
529565dc IM |
1318 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
1319 | int fd_out, loff_t __user *off_out, | |
1320 | size_t len, unsigned int flags) | |
5274f052 JA |
1321 | { |
1322 | long error; | |
1323 | struct file *in, *out; | |
1324 | int fput_in, fput_out; | |
1325 | ||
1326 | if (unlikely(!len)) | |
1327 | return 0; | |
1328 | ||
1329 | error = -EBADF; | |
529565dc | 1330 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1331 | if (in) { |
1332 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1333 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1334 | if (out) { |
1335 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1336 | error = do_splice(in, off_in, |
1337 | out, off_out, | |
1338 | len, flags); | |
5274f052 JA |
1339 | fput_light(out, fput_out); |
1340 | } | |
1341 | } | |
1342 | ||
1343 | fput_light(in, fput_in); | |
1344 | } | |
1345 | ||
1346 | return error; | |
1347 | } | |
70524490 | 1348 | |
aadd06e5 JA |
1349 | /* |
1350 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1351 | * return an appropriate error. | |
1352 | */ | |
1353 | static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1354 | { | |
1355 | int ret; | |
1356 | ||
1357 | /* | |
1358 | * Check ->nrbufs without the inode lock first. This function | |
1359 | * is speculative anyways, so missing one is ok. | |
1360 | */ | |
1361 | if (pipe->nrbufs) | |
1362 | return 0; | |
1363 | ||
1364 | ret = 0; | |
1365 | mutex_lock(&pipe->inode->i_mutex); | |
1366 | ||
1367 | while (!pipe->nrbufs) { | |
1368 | if (signal_pending(current)) { | |
1369 | ret = -ERESTARTSYS; | |
1370 | break; | |
1371 | } | |
1372 | if (!pipe->writers) | |
1373 | break; | |
1374 | if (!pipe->waiting_writers) { | |
1375 | if (flags & SPLICE_F_NONBLOCK) { | |
1376 | ret = -EAGAIN; | |
1377 | break; | |
1378 | } | |
1379 | } | |
1380 | pipe_wait(pipe); | |
1381 | } | |
1382 | ||
1383 | mutex_unlock(&pipe->inode->i_mutex); | |
1384 | return ret; | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1389 | * return an appropriate error. | |
1390 | */ | |
1391 | static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1392 | { | |
1393 | int ret; | |
1394 | ||
1395 | /* | |
1396 | * Check ->nrbufs without the inode lock first. This function | |
1397 | * is speculative anyways, so missing one is ok. | |
1398 | */ | |
1399 | if (pipe->nrbufs < PIPE_BUFFERS) | |
1400 | return 0; | |
1401 | ||
1402 | ret = 0; | |
1403 | mutex_lock(&pipe->inode->i_mutex); | |
1404 | ||
1405 | while (pipe->nrbufs >= PIPE_BUFFERS) { | |
1406 | if (!pipe->readers) { | |
1407 | send_sig(SIGPIPE, current, 0); | |
1408 | ret = -EPIPE; | |
1409 | break; | |
1410 | } | |
1411 | if (flags & SPLICE_F_NONBLOCK) { | |
1412 | ret = -EAGAIN; | |
1413 | break; | |
1414 | } | |
1415 | if (signal_pending(current)) { | |
1416 | ret = -ERESTARTSYS; | |
1417 | break; | |
1418 | } | |
1419 | pipe->waiting_writers++; | |
1420 | pipe_wait(pipe); | |
1421 | pipe->waiting_writers--; | |
1422 | } | |
1423 | ||
1424 | mutex_unlock(&pipe->inode->i_mutex); | |
1425 | return ret; | |
1426 | } | |
1427 | ||
70524490 JA |
1428 | /* |
1429 | * Link contents of ipipe to opipe. | |
1430 | */ | |
1431 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1432 | struct pipe_inode_info *opipe, | |
1433 | size_t len, unsigned int flags) | |
1434 | { | |
1435 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1436 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1437 | |
1438 | /* | |
1439 | * Potential ABBA deadlock, work around it by ordering lock | |
1440 | * grabbing by inode address. Otherwise two different processes | |
1441 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1442 | */ | |
62752ee1 | 1443 | inode_double_lock(ipipe->inode, opipe->inode); |
70524490 | 1444 | |
aadd06e5 | 1445 | do { |
70524490 JA |
1446 | if (!opipe->readers) { |
1447 | send_sig(SIGPIPE, current, 0); | |
1448 | if (!ret) | |
1449 | ret = -EPIPE; | |
1450 | break; | |
1451 | } | |
70524490 | 1452 | |
aadd06e5 JA |
1453 | /* |
1454 | * If we have iterated all input buffers or ran out of | |
1455 | * output room, break. | |
1456 | */ | |
1457 | if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) | |
1458 | break; | |
70524490 | 1459 | |
aadd06e5 JA |
1460 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); |
1461 | nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
70524490 JA |
1462 | |
1463 | /* | |
aadd06e5 JA |
1464 | * Get a reference to this pipe buffer, |
1465 | * so we can copy the contents over. | |
70524490 | 1466 | */ |
aadd06e5 JA |
1467 | ibuf->ops->get(ipipe, ibuf); |
1468 | ||
1469 | obuf = opipe->bufs + nbuf; | |
1470 | *obuf = *ibuf; | |
1471 | ||
2a27250e | 1472 | /* |
aadd06e5 JA |
1473 | * Don't inherit the gift flag, we need to |
1474 | * prevent multiple steals of this page. | |
2a27250e | 1475 | */ |
aadd06e5 | 1476 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1477 | |
aadd06e5 JA |
1478 | if (obuf->len > len) |
1479 | obuf->len = len; | |
70524490 | 1480 | |
aadd06e5 JA |
1481 | opipe->nrbufs++; |
1482 | ret += obuf->len; | |
1483 | len -= obuf->len; | |
1484 | i++; | |
1485 | } while (len); | |
70524490 | 1486 | |
62752ee1 | 1487 | inode_double_unlock(ipipe->inode, opipe->inode); |
70524490 | 1488 | |
aadd06e5 JA |
1489 | /* |
1490 | * If we put data in the output pipe, wakeup any potential readers. | |
1491 | */ | |
1492 | if (ret > 0) { | |
70524490 JA |
1493 | smp_mb(); |
1494 | if (waitqueue_active(&opipe->wait)) | |
1495 | wake_up_interruptible(&opipe->wait); | |
1496 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1497 | } | |
1498 | ||
1499 | return ret; | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1504 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1505 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1506 | * applicable one is SPLICE_F_NONBLOCK. | |
1507 | */ | |
1508 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1509 | unsigned int flags) | |
1510 | { | |
0f7fc9e4 JJS |
1511 | struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); |
1512 | struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); | |
aadd06e5 | 1513 | int ret = -EINVAL; |
70524490 JA |
1514 | |
1515 | /* | |
aadd06e5 JA |
1516 | * Duplicate the contents of ipipe to opipe without actually |
1517 | * copying the data. | |
70524490 | 1518 | */ |
aadd06e5 JA |
1519 | if (ipipe && opipe && ipipe != opipe) { |
1520 | /* | |
1521 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1522 | * ordering doesn't really matter. | |
1523 | */ | |
1524 | ret = link_ipipe_prep(ipipe, flags); | |
1525 | if (!ret) { | |
1526 | ret = link_opipe_prep(opipe, flags); | |
1527 | if (!ret) { | |
1528 | ret = link_pipe(ipipe, opipe, len, flags); | |
1529 | if (!ret && (flags & SPLICE_F_NONBLOCK)) | |
1530 | ret = -EAGAIN; | |
1531 | } | |
1532 | } | |
1533 | } | |
70524490 | 1534 | |
aadd06e5 | 1535 | return ret; |
70524490 JA |
1536 | } |
1537 | ||
1538 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1539 | { | |
1540 | struct file *in; | |
1541 | int error, fput_in; | |
1542 | ||
1543 | if (unlikely(!len)) | |
1544 | return 0; | |
1545 | ||
1546 | error = -EBADF; | |
1547 | in = fget_light(fdin, &fput_in); | |
1548 | if (in) { | |
1549 | if (in->f_mode & FMODE_READ) { | |
1550 | int fput_out; | |
1551 | struct file *out = fget_light(fdout, &fput_out); | |
1552 | ||
1553 | if (out) { | |
1554 | if (out->f_mode & FMODE_WRITE) | |
1555 | error = do_tee(in, out, len, flags); | |
1556 | fput_light(out, fput_out); | |
1557 | } | |
1558 | } | |
1559 | fput_light(in, fput_in); | |
1560 | } | |
1561 | ||
1562 | return error; | |
1563 | } |