]>
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 | { |
912d35f8 | 179 | int ret, do_wakeup, page_nr; |
5274f052 JA |
180 | |
181 | ret = 0; | |
182 | do_wakeup = 0; | |
912d35f8 | 183 | page_nr = 0; |
5274f052 | 184 | |
3a326a2c IM |
185 | if (pipe->inode) |
186 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 187 | |
5274f052 | 188 | for (;;) { |
3a326a2c | 189 | if (!pipe->readers) { |
5274f052 JA |
190 | send_sig(SIGPIPE, current, 0); |
191 | if (!ret) | |
192 | ret = -EPIPE; | |
193 | break; | |
194 | } | |
195 | ||
6f767b04 JA |
196 | if (pipe->nrbufs < PIPE_BUFFERS) { |
197 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 198 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 199 | |
912d35f8 JA |
200 | buf->page = spd->pages[page_nr]; |
201 | buf->offset = spd->partial[page_nr].offset; | |
202 | buf->len = spd->partial[page_nr].len; | |
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 | ||
3a326a2c IM |
247 | if (pipe->inode) |
248 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
249 | |
250 | if (do_wakeup) { | |
c0bd1f65 | 251 | smp_mb(); |
3a326a2c IM |
252 | if (waitqueue_active(&pipe->wait)) |
253 | wake_up_interruptible(&pipe->wait); | |
254 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
255 | } |
256 | ||
912d35f8 JA |
257 | while (page_nr < spd->nr_pages) |
258 | page_cache_release(spd->pages[page_nr++]); | |
5274f052 JA |
259 | |
260 | return ret; | |
261 | } | |
262 | ||
3a326a2c | 263 | static int |
cbb7e577 JA |
264 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
265 | struct pipe_inode_info *pipe, size_t len, | |
266 | unsigned int flags) | |
5274f052 JA |
267 | { |
268 | struct address_space *mapping = in->f_mapping; | |
912d35f8 | 269 | unsigned int loff, nr_pages; |
16c523dd | 270 | struct page *pages[PIPE_BUFFERS]; |
912d35f8 | 271 | struct partial_page partial[PIPE_BUFFERS]; |
5274f052 | 272 | struct page *page; |
91ad66ef JA |
273 | pgoff_t index, end_index; |
274 | loff_t isize; | |
912d35f8 | 275 | size_t total_len; |
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 | /* | |
73d62d83 | 292 | * Initiate read-ahead on this page range. however, don't call into |
0b749ce3 JA |
293 | * read-ahead if this is a non-zero offset (we are likely doing small |
294 | * chunk splice and the page is already there) for a single page. | |
5274f052 | 295 | */ |
eb645a24 JA |
296 | if (!loff || nr_pages > 1) |
297 | page_cache_readahead(mapping, &in->f_ra, in, index, nr_pages); | |
5274f052 | 298 | |
5274f052 | 299 | /* |
73d62d83 | 300 | * Now fill in the holes: |
5274f052 | 301 | */ |
7480a904 | 302 | error = 0; |
912d35f8 | 303 | total_len = 0; |
82aa5d61 | 304 | |
eb20796b JA |
305 | /* |
306 | * Lookup the (hopefully) full range of pages we need. | |
307 | */ | |
308 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
82aa5d61 | 309 | |
eb20796b JA |
310 | /* |
311 | * If find_get_pages_contig() returned fewer pages than we needed, | |
312 | * allocate the rest. | |
313 | */ | |
314 | index += spd.nr_pages; | |
315 | while (spd.nr_pages < nr_pages) { | |
82aa5d61 | 316 | /* |
eb20796b JA |
317 | * Page could be there, find_get_pages_contig() breaks on |
318 | * the first hole. | |
5274f052 | 319 | */ |
7480a904 JA |
320 | page = find_get_page(mapping, index); |
321 | if (!page) { | |
e27dedd8 JA |
322 | /* |
323 | * Make sure the read-ahead engine is notified | |
324 | * about this failure. | |
325 | */ | |
326 | handle_ra_miss(mapping, &in->f_ra, index); | |
327 | ||
7480a904 | 328 | /* |
eb20796b | 329 | * page didn't exist, allocate one. |
7480a904 JA |
330 | */ |
331 | page = page_cache_alloc_cold(mapping); | |
332 | if (!page) | |
333 | break; | |
334 | ||
335 | error = add_to_page_cache_lru(page, mapping, index, | |
2ae88149 | 336 | GFP_KERNEL); |
7480a904 JA |
337 | if (unlikely(error)) { |
338 | page_cache_release(page); | |
a0548871 JA |
339 | if (error == -EEXIST) |
340 | continue; | |
7480a904 JA |
341 | break; |
342 | } | |
eb20796b JA |
343 | /* |
344 | * add_to_page_cache() locks the page, unlock it | |
345 | * to avoid convoluting the logic below even more. | |
346 | */ | |
347 | unlock_page(page); | |
7480a904 JA |
348 | } |
349 | ||
eb20796b JA |
350 | pages[spd.nr_pages++] = page; |
351 | index++; | |
352 | } | |
353 | ||
354 | /* | |
355 | * Now loop over the map and see if we need to start IO on any | |
356 | * pages, fill in the partial map, etc. | |
357 | */ | |
358 | index = *ppos >> PAGE_CACHE_SHIFT; | |
359 | nr_pages = spd.nr_pages; | |
360 | spd.nr_pages = 0; | |
361 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
362 | unsigned int this_len; | |
363 | ||
364 | if (!len) | |
365 | break; | |
366 | ||
367 | /* | |
368 | * this_len is the max we'll use from this page | |
369 | */ | |
370 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
371 | page = pages[page_nr]; | |
372 | ||
7480a904 JA |
373 | /* |
374 | * If the page isn't uptodate, we may need to start io on it | |
375 | */ | |
376 | if (!PageUptodate(page)) { | |
c4f895cb JA |
377 | /* |
378 | * If in nonblock mode then dont block on waiting | |
379 | * for an in-flight io page | |
380 | */ | |
381 | if (flags & SPLICE_F_NONBLOCK) | |
382 | break; | |
383 | ||
7480a904 JA |
384 | lock_page(page); |
385 | ||
386 | /* | |
387 | * page was truncated, stop here. if this isn't the | |
388 | * first page, we'll just complete what we already | |
389 | * added | |
390 | */ | |
391 | if (!page->mapping) { | |
392 | unlock_page(page); | |
7480a904 JA |
393 | break; |
394 | } | |
395 | /* | |
396 | * page was already under io and is now done, great | |
397 | */ | |
398 | if (PageUptodate(page)) { | |
399 | unlock_page(page); | |
400 | goto fill_it; | |
401 | } | |
5274f052 | 402 | |
7480a904 JA |
403 | /* |
404 | * need to read in the page | |
405 | */ | |
406 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 407 | if (unlikely(error)) { |
eb20796b JA |
408 | /* |
409 | * We really should re-lookup the page here, | |
410 | * but it complicates things a lot. Instead | |
411 | * lets just do what we already stored, and | |
412 | * we'll get it the next time we are called. | |
413 | */ | |
7480a904 | 414 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
415 | error = 0; |
416 | ||
5274f052 JA |
417 | break; |
418 | } | |
91ad66ef JA |
419 | |
420 | /* | |
421 | * i_size must be checked after ->readpage(). | |
422 | */ | |
423 | isize = i_size_read(mapping->host); | |
424 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
eb20796b | 425 | if (unlikely(!isize || index > end_index)) |
91ad66ef | 426 | break; |
91ad66ef JA |
427 | |
428 | /* | |
429 | * if this is the last page, see if we need to shrink | |
430 | * the length and stop | |
431 | */ | |
432 | if (end_index == index) { | |
433 | loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK); | |
eb20796b | 434 | if (total_len + loff > isize) |
91ad66ef | 435 | break; |
91ad66ef JA |
436 | /* |
437 | * force quit after adding this page | |
438 | */ | |
eb20796b | 439 | len = this_len; |
82aa5d61 | 440 | this_len = min(this_len, loff); |
912d35f8 | 441 | loff = 0; |
91ad66ef | 442 | } |
5274f052 | 443 | } |
7480a904 | 444 | fill_it: |
eb20796b JA |
445 | partial[page_nr].offset = loff; |
446 | partial[page_nr].len = this_len; | |
82aa5d61 | 447 | len -= this_len; |
912d35f8 | 448 | total_len += this_len; |
91ad66ef | 449 | loff = 0; |
eb20796b JA |
450 | spd.nr_pages++; |
451 | index++; | |
5274f052 JA |
452 | } |
453 | ||
eb20796b JA |
454 | /* |
455 | * Release any pages at the end, if we quit early. 'i' is how far | |
456 | * we got, 'nr_pages' is how many pages are in the map. | |
457 | */ | |
458 | while (page_nr < nr_pages) | |
459 | page_cache_release(pages[page_nr++]); | |
460 | ||
912d35f8 | 461 | if (spd.nr_pages) |
00522fb4 | 462 | return splice_to_pipe(pipe, &spd); |
5274f052 | 463 | |
7480a904 | 464 | return error; |
5274f052 JA |
465 | } |
466 | ||
83f9135b JA |
467 | /** |
468 | * generic_file_splice_read - splice data from file to a pipe | |
469 | * @in: file to splice from | |
470 | * @pipe: pipe to splice to | |
471 | * @len: number of bytes to splice | |
472 | * @flags: splice modifier flags | |
473 | * | |
474 | * Will read pages from given file and fill them into a pipe. | |
83f9135b | 475 | */ |
cbb7e577 JA |
476 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
477 | struct pipe_inode_info *pipe, size_t len, | |
478 | unsigned int flags) | |
5274f052 JA |
479 | { |
480 | ssize_t spliced; | |
481 | int ret; | |
482 | ||
483 | ret = 0; | |
484 | spliced = 0; | |
3a326a2c | 485 | |
5274f052 | 486 | while (len) { |
cbb7e577 | 487 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 488 | |
c4f895cb | 489 | if (ret < 0) |
5274f052 | 490 | break; |
c4f895cb JA |
491 | else if (!ret) { |
492 | if (spliced) | |
493 | break; | |
494 | if (flags & SPLICE_F_NONBLOCK) { | |
495 | ret = -EAGAIN; | |
496 | break; | |
497 | } | |
498 | } | |
5274f052 | 499 | |
cbb7e577 | 500 | *ppos += ret; |
5274f052 JA |
501 | len -= ret; |
502 | spliced += ret; | |
503 | } | |
504 | ||
505 | if (spliced) | |
506 | return spliced; | |
507 | ||
508 | return ret; | |
509 | } | |
510 | ||
059a8f37 JA |
511 | EXPORT_SYMBOL(generic_file_splice_read); |
512 | ||
5274f052 | 513 | /* |
4f6f0bd2 | 514 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 515 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 516 | */ |
76ad4d11 | 517 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
518 | struct pipe_buffer *buf, struct splice_desc *sd) |
519 | { | |
520 | struct file *file = sd->file; | |
521 | loff_t pos = sd->pos; | |
f84d7519 | 522 | int ret, more; |
5274f052 | 523 | |
76ad4d11 | 524 | ret = buf->ops->pin(pipe, buf); |
f84d7519 JA |
525 | if (!ret) { |
526 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 527 | |
f84d7519 JA |
528 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
529 | sd->len, &pos, more); | |
530 | } | |
5274f052 | 531 | |
016b661e | 532 | return ret; |
5274f052 JA |
533 | } |
534 | ||
535 | /* | |
536 | * This is a little more tricky than the file -> pipe splicing. There are | |
537 | * basically three cases: | |
538 | * | |
539 | * - Destination page already exists in the address space and there | |
540 | * are users of it. For that case we have no other option that | |
541 | * copying the data. Tough luck. | |
542 | * - Destination page already exists in the address space, but there | |
543 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
544 | * through to last case. | |
545 | * - Destination page does not exist, we can add the pipe page to | |
546 | * the page cache and avoid the copy. | |
547 | * | |
83f9135b JA |
548 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
549 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
550 | * file address space page cache. This is possible if no one else has | |
551 | * the pipe page referenced outside of the pipe and page cache. If | |
552 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
553 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 554 | */ |
76ad4d11 | 555 | static int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
5274f052 JA |
556 | struct splice_desc *sd) |
557 | { | |
558 | struct file *file = sd->file; | |
559 | struct address_space *mapping = file->f_mapping; | |
016b661e | 560 | unsigned int offset, this_len; |
5274f052 | 561 | struct page *page; |
5274f052 | 562 | pgoff_t index; |
3e7ee3e7 | 563 | int ret; |
5274f052 JA |
564 | |
565 | /* | |
49d0b21b | 566 | * make sure the data in this buffer is uptodate |
5274f052 | 567 | */ |
76ad4d11 | 568 | ret = buf->ops->pin(pipe, buf); |
f84d7519 JA |
569 | if (unlikely(ret)) |
570 | return ret; | |
5274f052 JA |
571 | |
572 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
573 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
574 | ||
016b661e JA |
575 | this_len = sd->len; |
576 | if (this_len + offset > PAGE_CACHE_SIZE) | |
577 | this_len = PAGE_CACHE_SIZE - offset; | |
578 | ||
5274f052 | 579 | /* |
0568b409 JA |
580 | * Reuse buf page, if SPLICE_F_MOVE is set and we are doing a full |
581 | * page. | |
5274f052 | 582 | */ |
0568b409 | 583 | if ((sd->flags & SPLICE_F_MOVE) && this_len == PAGE_CACHE_SIZE) { |
83f9135b | 584 | /* |
1432873a JA |
585 | * If steal succeeds, buf->page is now pruned from the |
586 | * pagecache and we can reuse it. The page will also be | |
587 | * locked on successful return. | |
83f9135b | 588 | */ |
76ad4d11 | 589 | if (buf->ops->steal(pipe, buf)) |
5abc97aa JA |
590 | goto find_page; |
591 | ||
592 | page = buf->page; | |
2ae88149 | 593 | if (add_to_page_cache(page, mapping, index, GFP_KERNEL)) { |
46e678c9 | 594 | unlock_page(page); |
5abc97aa | 595 | goto find_page; |
46e678c9 | 596 | } |
1432873a JA |
597 | |
598 | page_cache_get(page); | |
599 | ||
600 | if (!(buf->flags & PIPE_BUF_FLAG_LRU)) | |
601 | lru_cache_add(page); | |
5abc97aa JA |
602 | } else { |
603 | find_page: | |
9e0267c2 JA |
604 | page = find_lock_page(mapping, index); |
605 | if (!page) { | |
606 | ret = -ENOMEM; | |
607 | page = page_cache_alloc_cold(mapping); | |
608 | if (unlikely(!page)) | |
e6e80f29 | 609 | goto out_ret; |
9e0267c2 JA |
610 | |
611 | /* | |
612 | * This will also lock the page | |
613 | */ | |
614 | ret = add_to_page_cache_lru(page, mapping, index, | |
2ae88149 | 615 | GFP_KERNEL); |
9e0267c2 JA |
616 | if (unlikely(ret)) |
617 | goto out; | |
618 | } | |
5abc97aa JA |
619 | |
620 | /* | |
9e0267c2 JA |
621 | * We get here with the page locked. If the page is also |
622 | * uptodate, we don't need to do more. If it isn't, we | |
623 | * may need to bring it in if we are not going to overwrite | |
624 | * the full page. | |
5abc97aa JA |
625 | */ |
626 | if (!PageUptodate(page)) { | |
016b661e | 627 | if (this_len < PAGE_CACHE_SIZE) { |
5abc97aa JA |
628 | ret = mapping->a_ops->readpage(file, page); |
629 | if (unlikely(ret)) | |
630 | goto out; | |
631 | ||
632 | lock_page(page); | |
633 | ||
634 | if (!PageUptodate(page)) { | |
635 | /* | |
73d62d83 | 636 | * Page got invalidated, repeat. |
5abc97aa JA |
637 | */ |
638 | if (!page->mapping) { | |
639 | unlock_page(page); | |
640 | page_cache_release(page); | |
641 | goto find_page; | |
642 | } | |
643 | ret = -EIO; | |
644 | goto out; | |
5274f052 | 645 | } |
9e0267c2 | 646 | } else |
5abc97aa | 647 | SetPageUptodate(page); |
5274f052 JA |
648 | } |
649 | } | |
650 | ||
016b661e | 651 | ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); |
bfc4ee39 JA |
652 | if (unlikely(ret)) { |
653 | loff_t isize = i_size_read(mapping->host); | |
654 | ||
655 | if (ret != AOP_TRUNCATED_PAGE) | |
656 | unlock_page(page); | |
4f6f0bd2 | 657 | page_cache_release(page); |
bfc4ee39 JA |
658 | if (ret == AOP_TRUNCATED_PAGE) |
659 | goto find_page; | |
660 | ||
661 | /* | |
662 | * prepare_write() may have instantiated a few blocks | |
663 | * outside i_size. Trim these off again. | |
664 | */ | |
665 | if (sd->pos + this_len > isize) | |
666 | vmtruncate(mapping->host, isize); | |
667 | ||
e6e80f29 | 668 | goto out_ret; |
bfc4ee39 | 669 | } |
5274f052 | 670 | |
0568b409 | 671 | if (buf->page != page) { |
f84d7519 JA |
672 | /* |
673 | * Careful, ->map() uses KM_USER0! | |
674 | */ | |
76ad4d11 | 675 | char *src = buf->ops->map(pipe, buf, 1); |
f84d7519 | 676 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 677 | |
016b661e | 678 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 679 | flush_dcache_page(page); |
f84d7519 | 680 | kunmap_atomic(dst, KM_USER1); |
76ad4d11 | 681 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 682 | } |
5274f052 | 683 | |
016b661e | 684 | ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); |
0568b409 JA |
685 | if (!ret) { |
686 | /* | |
687 | * Return the number of bytes written and mark page as | |
688 | * accessed, we are now done! | |
689 | */ | |
690 | ret = this_len; | |
691 | mark_page_accessed(page); | |
692 | balance_dirty_pages_ratelimited(mapping); | |
693 | } else if (ret == AOP_TRUNCATED_PAGE) { | |
4f6f0bd2 JA |
694 | page_cache_release(page); |
695 | goto find_page; | |
0568b409 | 696 | } |
5274f052 | 697 | out: |
0568b409 | 698 | page_cache_release(page); |
9e0267c2 | 699 | unlock_page(page); |
e6e80f29 | 700 | out_ret: |
5274f052 JA |
701 | return ret; |
702 | } | |
703 | ||
83f9135b JA |
704 | /* |
705 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
706 | * key here is the 'actor' worker passed in that actually moves the data | |
707 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
708 | */ | |
6da61809 MF |
709 | static ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, |
710 | struct file *out, loff_t *ppos, size_t len, | |
711 | unsigned int flags, splice_actor *actor) | |
5274f052 | 712 | { |
5274f052 JA |
713 | int ret, do_wakeup, err; |
714 | struct splice_desc sd; | |
715 | ||
716 | ret = 0; | |
717 | do_wakeup = 0; | |
718 | ||
719 | sd.total_len = len; | |
720 | sd.flags = flags; | |
721 | sd.file = out; | |
cbb7e577 | 722 | sd.pos = *ppos; |
5274f052 | 723 | |
5274f052 | 724 | for (;;) { |
6f767b04 JA |
725 | if (pipe->nrbufs) { |
726 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
d4c3cca9 | 727 | const struct pipe_buf_operations *ops = buf->ops; |
5274f052 JA |
728 | |
729 | sd.len = buf->len; | |
730 | if (sd.len > sd.total_len) | |
731 | sd.len = sd.total_len; | |
732 | ||
3a326a2c | 733 | err = actor(pipe, buf, &sd); |
016b661e | 734 | if (err <= 0) { |
5274f052 JA |
735 | if (!ret && err != -ENODATA) |
736 | ret = err; | |
737 | ||
738 | break; | |
739 | } | |
740 | ||
016b661e JA |
741 | ret += err; |
742 | buf->offset += err; | |
743 | buf->len -= err; | |
744 | ||
745 | sd.len -= err; | |
746 | sd.pos += err; | |
747 | sd.total_len -= err; | |
748 | if (sd.len) | |
749 | continue; | |
73d62d83 | 750 | |
5274f052 JA |
751 | if (!buf->len) { |
752 | buf->ops = NULL; | |
3a326a2c | 753 | ops->release(pipe, buf); |
6f767b04 JA |
754 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
755 | pipe->nrbufs--; | |
756 | if (pipe->inode) | |
757 | do_wakeup = 1; | |
5274f052 JA |
758 | } |
759 | ||
5274f052 JA |
760 | if (!sd.total_len) |
761 | break; | |
762 | } | |
763 | ||
6f767b04 | 764 | if (pipe->nrbufs) |
5274f052 | 765 | continue; |
3a326a2c | 766 | if (!pipe->writers) |
5274f052 | 767 | break; |
3a326a2c | 768 | if (!pipe->waiting_writers) { |
5274f052 JA |
769 | if (ret) |
770 | break; | |
771 | } | |
772 | ||
29e35094 LT |
773 | if (flags & SPLICE_F_NONBLOCK) { |
774 | if (!ret) | |
775 | ret = -EAGAIN; | |
776 | break; | |
777 | } | |
778 | ||
5274f052 JA |
779 | if (signal_pending(current)) { |
780 | if (!ret) | |
781 | ret = -ERESTARTSYS; | |
782 | break; | |
783 | } | |
784 | ||
785 | if (do_wakeup) { | |
c0bd1f65 | 786 | smp_mb(); |
3a326a2c IM |
787 | if (waitqueue_active(&pipe->wait)) |
788 | wake_up_interruptible_sync(&pipe->wait); | |
789 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
790 | do_wakeup = 0; |
791 | } | |
792 | ||
3a326a2c | 793 | pipe_wait(pipe); |
5274f052 JA |
794 | } |
795 | ||
5274f052 | 796 | if (do_wakeup) { |
c0bd1f65 | 797 | smp_mb(); |
3a326a2c IM |
798 | if (waitqueue_active(&pipe->wait)) |
799 | wake_up_interruptible(&pipe->wait); | |
800 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
801 | } |
802 | ||
5274f052 | 803 | return ret; |
5274f052 JA |
804 | } |
805 | ||
6da61809 MF |
806 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
807 | loff_t *ppos, size_t len, unsigned int flags, | |
808 | splice_actor *actor) | |
809 | { | |
810 | ssize_t ret; | |
811 | struct inode *inode = out->f_mapping->host; | |
812 | ||
813 | /* | |
814 | * The actor worker might be calling ->prepare_write and | |
815 | * ->commit_write. Most of the time, these expect i_mutex to | |
816 | * be held. Since this may result in an ABBA deadlock with | |
817 | * pipe->inode, we have to order lock acquiry here. | |
818 | */ | |
819 | inode_double_lock(inode, pipe->inode); | |
820 | ret = __splice_from_pipe(pipe, out, ppos, len, flags, actor); | |
821 | inode_double_unlock(inode, pipe->inode); | |
822 | ||
823 | return ret; | |
824 | } | |
825 | ||
826 | /** | |
827 | * generic_file_splice_write_nolock - generic_file_splice_write without mutexes | |
828 | * @pipe: pipe info | |
829 | * @out: file to write to | |
830 | * @len: number of bytes to splice | |
831 | * @flags: splice modifier flags | |
832 | * | |
833 | * Will either move or copy pages (determined by @flags options) from | |
834 | * the given pipe inode to the given file. The caller is responsible | |
835 | * for acquiring i_mutex on both inodes. | |
836 | * | |
837 | */ | |
838 | ssize_t | |
839 | generic_file_splice_write_nolock(struct pipe_inode_info *pipe, struct file *out, | |
840 | loff_t *ppos, size_t len, unsigned int flags) | |
841 | { | |
842 | struct address_space *mapping = out->f_mapping; | |
843 | struct inode *inode = mapping->host; | |
844 | ssize_t ret; | |
845 | int err; | |
846 | ||
0f7fc9e4 | 847 | err = remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
848 | if (unlikely(err)) |
849 | return err; | |
850 | ||
6da61809 MF |
851 | ret = __splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
852 | if (ret > 0) { | |
853 | *ppos += ret; | |
854 | ||
855 | /* | |
856 | * If file or inode is SYNC and we actually wrote some data, | |
857 | * sync it. | |
858 | */ | |
859 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
860 | err = generic_osync_inode(inode, mapping, | |
861 | OSYNC_METADATA|OSYNC_DATA); | |
862 | ||
863 | if (err) | |
864 | ret = err; | |
865 | } | |
866 | } | |
867 | ||
868 | return ret; | |
869 | } | |
870 | ||
871 | EXPORT_SYMBOL(generic_file_splice_write_nolock); | |
872 | ||
83f9135b JA |
873 | /** |
874 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 875 | * @pipe: pipe info |
83f9135b JA |
876 | * @out: file to write to |
877 | * @len: number of bytes to splice | |
878 | * @flags: splice modifier flags | |
879 | * | |
880 | * Will either move or copy pages (determined by @flags options) from | |
881 | * the given pipe inode to the given file. | |
882 | * | |
883 | */ | |
3a326a2c IM |
884 | ssize_t |
885 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 886 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 887 | { |
4f6f0bd2 | 888 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 889 | struct inode *inode = mapping->host; |
3a326a2c | 890 | ssize_t ret; |
8c34e2d6 JA |
891 | int err; |
892 | ||
0f7fc9e4 | 893 | err = should_remove_suid(out->f_path.dentry); |
8c34e2d6 JA |
894 | if (unlikely(err)) { |
895 | mutex_lock(&inode->i_mutex); | |
0f7fc9e4 | 896 | err = __remove_suid(out->f_path.dentry, err); |
8c34e2d6 JA |
897 | mutex_unlock(&inode->i_mutex); |
898 | if (err) | |
899 | return err; | |
900 | } | |
3a326a2c | 901 | |
00522fb4 | 902 | ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
a4514ebd | 903 | if (ret > 0) { |
a4514ebd JA |
904 | *ppos += ret; |
905 | ||
906 | /* | |
907 | * If file or inode is SYNC and we actually wrote some data, | |
908 | * sync it. | |
909 | */ | |
910 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
a4514ebd JA |
911 | mutex_lock(&inode->i_mutex); |
912 | err = generic_osync_inode(inode, mapping, | |
913 | OSYNC_METADATA|OSYNC_DATA); | |
914 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 915 | |
a4514ebd JA |
916 | if (err) |
917 | ret = err; | |
918 | } | |
4f6f0bd2 JA |
919 | } |
920 | ||
921 | return ret; | |
5274f052 JA |
922 | } |
923 | ||
059a8f37 JA |
924 | EXPORT_SYMBOL(generic_file_splice_write); |
925 | ||
83f9135b JA |
926 | /** |
927 | * generic_splice_sendpage - splice data from a pipe to a socket | |
928 | * @inode: pipe inode | |
929 | * @out: socket to write to | |
930 | * @len: number of bytes to splice | |
931 | * @flags: splice modifier flags | |
932 | * | |
933 | * Will send @len bytes from the pipe to a network socket. No data copying | |
934 | * is involved. | |
935 | * | |
936 | */ | |
3a326a2c | 937 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 938 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 939 | { |
00522fb4 | 940 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
941 | } |
942 | ||
059a8f37 | 943 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 944 | |
83f9135b JA |
945 | /* |
946 | * Attempt to initiate a splice from pipe to file. | |
947 | */ | |
3a326a2c | 948 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 949 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 950 | { |
5274f052 JA |
951 | int ret; |
952 | ||
49570e9b | 953 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
954 | return -EINVAL; |
955 | ||
49570e9b | 956 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
957 | return -EBADF; |
958 | ||
cbb7e577 | 959 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
960 | if (unlikely(ret < 0)) |
961 | return ret; | |
962 | ||
cbb7e577 | 963 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
964 | } |
965 | ||
83f9135b JA |
966 | /* |
967 | * Attempt to initiate a splice from a file to a pipe. | |
968 | */ | |
cbb7e577 JA |
969 | static long do_splice_to(struct file *in, loff_t *ppos, |
970 | struct pipe_inode_info *pipe, size_t len, | |
971 | unsigned int flags) | |
5274f052 | 972 | { |
cbb7e577 | 973 | loff_t isize, left; |
5274f052 JA |
974 | int ret; |
975 | ||
49570e9b | 976 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
977 | return -EINVAL; |
978 | ||
49570e9b | 979 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
980 | return -EBADF; |
981 | ||
cbb7e577 | 982 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
983 | if (unlikely(ret < 0)) |
984 | return ret; | |
985 | ||
986 | isize = i_size_read(in->f_mapping->host); | |
cbb7e577 | 987 | if (unlikely(*ppos >= isize)) |
5274f052 JA |
988 | return 0; |
989 | ||
cbb7e577 | 990 | left = isize - *ppos; |
49570e9b | 991 | if (unlikely(left < len)) |
5274f052 JA |
992 | len = left; |
993 | ||
cbb7e577 | 994 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
995 | } |
996 | ||
cbb7e577 JA |
997 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
998 | size_t len, unsigned int flags) | |
b92ce558 JA |
999 | { |
1000 | struct pipe_inode_info *pipe; | |
1001 | long ret, bytes; | |
cbb7e577 | 1002 | loff_t out_off; |
b92ce558 JA |
1003 | umode_t i_mode; |
1004 | int i; | |
1005 | ||
1006 | /* | |
1007 | * We require the input being a regular file, as we don't want to | |
1008 | * randomly drop data for eg socket -> socket splicing. Use the | |
1009 | * piped splicing for that! | |
1010 | */ | |
0f7fc9e4 | 1011 | i_mode = in->f_path.dentry->d_inode->i_mode; |
b92ce558 JA |
1012 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1013 | return -EINVAL; | |
1014 | ||
1015 | /* | |
1016 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1017 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1018 | */ | |
1019 | pipe = current->splice_pipe; | |
49570e9b | 1020 | if (unlikely(!pipe)) { |
b92ce558 JA |
1021 | pipe = alloc_pipe_info(NULL); |
1022 | if (!pipe) | |
1023 | return -ENOMEM; | |
1024 | ||
1025 | /* | |
1026 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1027 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1028 | * PIPE_READERS appropriately. |
1029 | */ | |
1030 | pipe->readers = 1; | |
1031 | ||
1032 | current->splice_pipe = pipe; | |
1033 | } | |
1034 | ||
1035 | /* | |
73d62d83 | 1036 | * Do the splice. |
b92ce558 JA |
1037 | */ |
1038 | ret = 0; | |
1039 | bytes = 0; | |
cbb7e577 | 1040 | out_off = 0; |
b92ce558 JA |
1041 | |
1042 | while (len) { | |
1043 | size_t read_len, max_read_len; | |
1044 | ||
1045 | /* | |
1046 | * Do at most PIPE_BUFFERS pages worth of transfer: | |
1047 | */ | |
1048 | max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE)); | |
1049 | ||
cbb7e577 | 1050 | ret = do_splice_to(in, ppos, pipe, max_read_len, flags); |
b92ce558 JA |
1051 | if (unlikely(ret < 0)) |
1052 | goto out_release; | |
1053 | ||
1054 | read_len = ret; | |
1055 | ||
1056 | /* | |
1057 | * NOTE: nonblocking mode only applies to the input. We | |
1058 | * must not do the output in nonblocking mode as then we | |
1059 | * could get stuck data in the internal pipe: | |
1060 | */ | |
cbb7e577 | 1061 | ret = do_splice_from(pipe, out, &out_off, read_len, |
b92ce558 JA |
1062 | flags & ~SPLICE_F_NONBLOCK); |
1063 | if (unlikely(ret < 0)) | |
1064 | goto out_release; | |
1065 | ||
1066 | bytes += ret; | |
1067 | len -= ret; | |
1068 | ||
1069 | /* | |
1070 | * In nonblocking mode, if we got back a short read then | |
1071 | * that was due to either an IO error or due to the | |
1072 | * pagecache entry not being there. In the IO error case | |
1073 | * the _next_ splice attempt will produce a clean IO error | |
1074 | * return value (not a short read), so in both cases it's | |
1075 | * correct to break out of the loop here: | |
1076 | */ | |
1077 | if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len)) | |
1078 | break; | |
1079 | } | |
1080 | ||
1081 | pipe->nrbufs = pipe->curbuf = 0; | |
1082 | ||
1083 | return bytes; | |
1084 | ||
1085 | out_release: | |
1086 | /* | |
1087 | * If we did an incomplete transfer we must release | |
1088 | * the pipe buffers in question: | |
1089 | */ | |
1090 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1091 | struct pipe_buffer *buf = pipe->bufs + i; | |
1092 | ||
1093 | if (buf->ops) { | |
1094 | buf->ops->release(pipe, buf); | |
1095 | buf->ops = NULL; | |
1096 | } | |
1097 | } | |
1098 | pipe->nrbufs = pipe->curbuf = 0; | |
1099 | ||
1100 | /* | |
1101 | * If we transferred some data, return the number of bytes: | |
1102 | */ | |
1103 | if (bytes > 0) | |
1104 | return bytes; | |
1105 | ||
1106 | return ret; | |
1107 | } | |
1108 | ||
1109 | EXPORT_SYMBOL(do_splice_direct); | |
1110 | ||
ddac0d39 JA |
1111 | /* |
1112 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1113 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1114 | * pipe. | |
1115 | */ | |
1116 | static inline struct pipe_inode_info *pipe_info(struct inode *inode) | |
1117 | { | |
1118 | if (S_ISFIFO(inode->i_mode)) | |
1119 | return inode->i_pipe; | |
1120 | ||
1121 | return NULL; | |
1122 | } | |
1123 | ||
83f9135b JA |
1124 | /* |
1125 | * Determine where to splice to/from. | |
1126 | */ | |
529565dc IM |
1127 | static long do_splice(struct file *in, loff_t __user *off_in, |
1128 | struct file *out, loff_t __user *off_out, | |
1129 | size_t len, unsigned int flags) | |
5274f052 | 1130 | { |
3a326a2c | 1131 | struct pipe_inode_info *pipe; |
cbb7e577 | 1132 | loff_t offset, *off; |
a4514ebd | 1133 | long ret; |
5274f052 | 1134 | |
0f7fc9e4 | 1135 | pipe = pipe_info(in->f_path.dentry->d_inode); |
529565dc IM |
1136 | if (pipe) { |
1137 | if (off_in) | |
1138 | return -ESPIPE; | |
b92ce558 JA |
1139 | if (off_out) { |
1140 | if (out->f_op->llseek == no_llseek) | |
1141 | return -EINVAL; | |
cbb7e577 | 1142 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1143 | return -EFAULT; |
cbb7e577 JA |
1144 | off = &offset; |
1145 | } else | |
1146 | off = &out->f_pos; | |
529565dc | 1147 | |
a4514ebd JA |
1148 | ret = do_splice_from(pipe, out, off, len, flags); |
1149 | ||
1150 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1151 | ret = -EFAULT; | |
1152 | ||
1153 | return ret; | |
529565dc | 1154 | } |
5274f052 | 1155 | |
0f7fc9e4 | 1156 | pipe = pipe_info(out->f_path.dentry->d_inode); |
529565dc IM |
1157 | if (pipe) { |
1158 | if (off_out) | |
1159 | return -ESPIPE; | |
b92ce558 JA |
1160 | if (off_in) { |
1161 | if (in->f_op->llseek == no_llseek) | |
1162 | return -EINVAL; | |
cbb7e577 | 1163 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1164 | return -EFAULT; |
cbb7e577 JA |
1165 | off = &offset; |
1166 | } else | |
1167 | off = &in->f_pos; | |
529565dc | 1168 | |
a4514ebd JA |
1169 | ret = do_splice_to(in, off, pipe, len, flags); |
1170 | ||
1171 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1172 | ret = -EFAULT; | |
1173 | ||
1174 | return ret; | |
529565dc | 1175 | } |
5274f052 JA |
1176 | |
1177 | return -EINVAL; | |
1178 | } | |
1179 | ||
912d35f8 JA |
1180 | /* |
1181 | * Map an iov into an array of pages and offset/length tupples. With the | |
1182 | * partial_page structure, we can map several non-contiguous ranges into | |
1183 | * our ones pages[] map instead of splitting that operation into pieces. | |
1184 | * Could easily be exported as a generic helper for other users, in which | |
1185 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1186 | */ | |
1187 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1188 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1189 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1190 | { |
1191 | int buffers = 0, error = 0; | |
1192 | ||
1193 | /* | |
1194 | * It's ok to take the mmap_sem for reading, even | |
1195 | * across a "get_user()". | |
1196 | */ | |
1197 | down_read(¤t->mm->mmap_sem); | |
1198 | ||
1199 | while (nr_vecs) { | |
1200 | unsigned long off, npages; | |
1201 | void __user *base; | |
1202 | size_t len; | |
1203 | int i; | |
1204 | ||
1205 | /* | |
1206 | * Get user address base and length for this iovec. | |
1207 | */ | |
1208 | error = get_user(base, &iov->iov_base); | |
1209 | if (unlikely(error)) | |
1210 | break; | |
1211 | error = get_user(len, &iov->iov_len); | |
1212 | if (unlikely(error)) | |
1213 | break; | |
1214 | ||
1215 | /* | |
1216 | * Sanity check this iovec. 0 read succeeds. | |
1217 | */ | |
1218 | if (unlikely(!len)) | |
1219 | break; | |
1220 | error = -EFAULT; | |
1221 | if (unlikely(!base)) | |
1222 | break; | |
1223 | ||
1224 | /* | |
1225 | * Get this base offset and number of pages, then map | |
1226 | * in the user pages. | |
1227 | */ | |
1228 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1229 | |
1230 | /* | |
1231 | * If asked for alignment, the offset must be zero and the | |
1232 | * length a multiple of the PAGE_SIZE. | |
1233 | */ | |
1234 | error = -EINVAL; | |
1235 | if (aligned && (off || len & ~PAGE_MASK)) | |
1236 | break; | |
1237 | ||
912d35f8 JA |
1238 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1239 | if (npages > PIPE_BUFFERS - buffers) | |
1240 | npages = PIPE_BUFFERS - buffers; | |
1241 | ||
1242 | error = get_user_pages(current, current->mm, | |
1243 | (unsigned long) base, npages, 0, 0, | |
1244 | &pages[buffers], NULL); | |
1245 | ||
1246 | if (unlikely(error <= 0)) | |
1247 | break; | |
1248 | ||
1249 | /* | |
1250 | * Fill this contiguous range into the partial page map. | |
1251 | */ | |
1252 | for (i = 0; i < error; i++) { | |
7591489a | 1253 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1254 | |
1255 | partial[buffers].offset = off; | |
1256 | partial[buffers].len = plen; | |
1257 | ||
1258 | off = 0; | |
1259 | len -= plen; | |
1260 | buffers++; | |
1261 | } | |
1262 | ||
1263 | /* | |
1264 | * We didn't complete this iov, stop here since it probably | |
1265 | * means we have to move some of this into a pipe to | |
1266 | * be able to continue. | |
1267 | */ | |
1268 | if (len) | |
1269 | break; | |
1270 | ||
1271 | /* | |
1272 | * Don't continue if we mapped fewer pages than we asked for, | |
1273 | * or if we mapped the max number of pages that we have | |
1274 | * room for. | |
1275 | */ | |
1276 | if (error < npages || buffers == PIPE_BUFFERS) | |
1277 | break; | |
1278 | ||
1279 | nr_vecs--; | |
1280 | iov++; | |
1281 | } | |
1282 | ||
1283 | up_read(¤t->mm->mmap_sem); | |
1284 | ||
1285 | if (buffers) | |
1286 | return buffers; | |
1287 | ||
1288 | return error; | |
1289 | } | |
1290 | ||
1291 | /* | |
1292 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1293 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1294 | * to file). In both cases the output is a pipe, naturally. | |
1295 | * | |
1296 | * Note that vmsplice only supports splicing _from_ user memory to a pipe, | |
1297 | * not the other way around. Splicing from user memory is a simple operation | |
1298 | * that can be supported without any funky alignment restrictions or nasty | |
1299 | * vm tricks. We simply map in the user memory and fill them into a pipe. | |
1300 | * The reverse isn't quite as easy, though. There are two possible solutions | |
1301 | * for that: | |
1302 | * | |
1303 | * - memcpy() the data internally, at which point we might as well just | |
1304 | * do a regular read() on the buffer anyway. | |
1305 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1306 | * has restriction limitations on both ends of the pipe). | |
1307 | * | |
1308 | * Alas, it isn't here. | |
1309 | * | |
1310 | */ | |
1311 | static long do_vmsplice(struct file *file, const struct iovec __user *iov, | |
1312 | unsigned long nr_segs, unsigned int flags) | |
1313 | { | |
ddac0d39 | 1314 | struct pipe_inode_info *pipe; |
912d35f8 JA |
1315 | struct page *pages[PIPE_BUFFERS]; |
1316 | struct partial_page partial[PIPE_BUFFERS]; | |
1317 | struct splice_pipe_desc spd = { | |
1318 | .pages = pages, | |
1319 | .partial = partial, | |
1320 | .flags = flags, | |
1321 | .ops = &user_page_pipe_buf_ops, | |
1322 | }; | |
1323 | ||
0f7fc9e4 | 1324 | pipe = pipe_info(file->f_path.dentry->d_inode); |
ddac0d39 | 1325 | if (!pipe) |
912d35f8 JA |
1326 | return -EBADF; |
1327 | if (unlikely(nr_segs > UIO_MAXIOV)) | |
1328 | return -EINVAL; | |
1329 | else if (unlikely(!nr_segs)) | |
1330 | return 0; | |
1331 | ||
7afa6fd0 JA |
1332 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1333 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1334 | if (spd.nr_pages <= 0) |
1335 | return spd.nr_pages; | |
1336 | ||
00522fb4 | 1337 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1338 | } |
1339 | ||
1340 | asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, | |
1341 | unsigned long nr_segs, unsigned int flags) | |
1342 | { | |
1343 | struct file *file; | |
1344 | long error; | |
1345 | int fput; | |
1346 | ||
1347 | error = -EBADF; | |
1348 | file = fget_light(fd, &fput); | |
1349 | if (file) { | |
1350 | if (file->f_mode & FMODE_WRITE) | |
1351 | error = do_vmsplice(file, iov, nr_segs, flags); | |
1352 | ||
1353 | fput_light(file, fput); | |
1354 | } | |
1355 | ||
1356 | return error; | |
1357 | } | |
1358 | ||
529565dc IM |
1359 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
1360 | int fd_out, loff_t __user *off_out, | |
1361 | size_t len, unsigned int flags) | |
5274f052 JA |
1362 | { |
1363 | long error; | |
1364 | struct file *in, *out; | |
1365 | int fput_in, fput_out; | |
1366 | ||
1367 | if (unlikely(!len)) | |
1368 | return 0; | |
1369 | ||
1370 | error = -EBADF; | |
529565dc | 1371 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1372 | if (in) { |
1373 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1374 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1375 | if (out) { |
1376 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1377 | error = do_splice(in, off_in, |
1378 | out, off_out, | |
1379 | len, flags); | |
5274f052 JA |
1380 | fput_light(out, fput_out); |
1381 | } | |
1382 | } | |
1383 | ||
1384 | fput_light(in, fput_in); | |
1385 | } | |
1386 | ||
1387 | return error; | |
1388 | } | |
70524490 | 1389 | |
aadd06e5 JA |
1390 | /* |
1391 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1392 | * return an appropriate error. | |
1393 | */ | |
1394 | static int link_ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1395 | { | |
1396 | int ret; | |
1397 | ||
1398 | /* | |
1399 | * Check ->nrbufs without the inode lock first. This function | |
1400 | * is speculative anyways, so missing one is ok. | |
1401 | */ | |
1402 | if (pipe->nrbufs) | |
1403 | return 0; | |
1404 | ||
1405 | ret = 0; | |
1406 | mutex_lock(&pipe->inode->i_mutex); | |
1407 | ||
1408 | while (!pipe->nrbufs) { | |
1409 | if (signal_pending(current)) { | |
1410 | ret = -ERESTARTSYS; | |
1411 | break; | |
1412 | } | |
1413 | if (!pipe->writers) | |
1414 | break; | |
1415 | if (!pipe->waiting_writers) { | |
1416 | if (flags & SPLICE_F_NONBLOCK) { | |
1417 | ret = -EAGAIN; | |
1418 | break; | |
1419 | } | |
1420 | } | |
1421 | pipe_wait(pipe); | |
1422 | } | |
1423 | ||
1424 | mutex_unlock(&pipe->inode->i_mutex); | |
1425 | return ret; | |
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1430 | * return an appropriate error. | |
1431 | */ | |
1432 | static int link_opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) | |
1433 | { | |
1434 | int ret; | |
1435 | ||
1436 | /* | |
1437 | * Check ->nrbufs without the inode lock first. This function | |
1438 | * is speculative anyways, so missing one is ok. | |
1439 | */ | |
1440 | if (pipe->nrbufs < PIPE_BUFFERS) | |
1441 | return 0; | |
1442 | ||
1443 | ret = 0; | |
1444 | mutex_lock(&pipe->inode->i_mutex); | |
1445 | ||
1446 | while (pipe->nrbufs >= PIPE_BUFFERS) { | |
1447 | if (!pipe->readers) { | |
1448 | send_sig(SIGPIPE, current, 0); | |
1449 | ret = -EPIPE; | |
1450 | break; | |
1451 | } | |
1452 | if (flags & SPLICE_F_NONBLOCK) { | |
1453 | ret = -EAGAIN; | |
1454 | break; | |
1455 | } | |
1456 | if (signal_pending(current)) { | |
1457 | ret = -ERESTARTSYS; | |
1458 | break; | |
1459 | } | |
1460 | pipe->waiting_writers++; | |
1461 | pipe_wait(pipe); | |
1462 | pipe->waiting_writers--; | |
1463 | } | |
1464 | ||
1465 | mutex_unlock(&pipe->inode->i_mutex); | |
1466 | return ret; | |
1467 | } | |
1468 | ||
70524490 JA |
1469 | /* |
1470 | * Link contents of ipipe to opipe. | |
1471 | */ | |
1472 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1473 | struct pipe_inode_info *opipe, | |
1474 | size_t len, unsigned int flags) | |
1475 | { | |
1476 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1477 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1478 | |
1479 | /* | |
1480 | * Potential ABBA deadlock, work around it by ordering lock | |
1481 | * grabbing by inode address. Otherwise two different processes | |
1482 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1483 | */ | |
62752ee1 | 1484 | inode_double_lock(ipipe->inode, opipe->inode); |
70524490 | 1485 | |
aadd06e5 | 1486 | do { |
70524490 JA |
1487 | if (!opipe->readers) { |
1488 | send_sig(SIGPIPE, current, 0); | |
1489 | if (!ret) | |
1490 | ret = -EPIPE; | |
1491 | break; | |
1492 | } | |
70524490 | 1493 | |
aadd06e5 JA |
1494 | /* |
1495 | * If we have iterated all input buffers or ran out of | |
1496 | * output room, break. | |
1497 | */ | |
1498 | if (i >= ipipe->nrbufs || opipe->nrbufs >= PIPE_BUFFERS) | |
1499 | break; | |
70524490 | 1500 | |
aadd06e5 JA |
1501 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); |
1502 | nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
70524490 JA |
1503 | |
1504 | /* | |
aadd06e5 JA |
1505 | * Get a reference to this pipe buffer, |
1506 | * so we can copy the contents over. | |
70524490 | 1507 | */ |
aadd06e5 JA |
1508 | ibuf->ops->get(ipipe, ibuf); |
1509 | ||
1510 | obuf = opipe->bufs + nbuf; | |
1511 | *obuf = *ibuf; | |
1512 | ||
2a27250e | 1513 | /* |
aadd06e5 JA |
1514 | * Don't inherit the gift flag, we need to |
1515 | * prevent multiple steals of this page. | |
2a27250e | 1516 | */ |
aadd06e5 | 1517 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1518 | |
aadd06e5 JA |
1519 | if (obuf->len > len) |
1520 | obuf->len = len; | |
70524490 | 1521 | |
aadd06e5 JA |
1522 | opipe->nrbufs++; |
1523 | ret += obuf->len; | |
1524 | len -= obuf->len; | |
1525 | i++; | |
1526 | } while (len); | |
70524490 | 1527 | |
62752ee1 | 1528 | inode_double_unlock(ipipe->inode, opipe->inode); |
70524490 | 1529 | |
aadd06e5 JA |
1530 | /* |
1531 | * If we put data in the output pipe, wakeup any potential readers. | |
1532 | */ | |
1533 | if (ret > 0) { | |
70524490 JA |
1534 | smp_mb(); |
1535 | if (waitqueue_active(&opipe->wait)) | |
1536 | wake_up_interruptible(&opipe->wait); | |
1537 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1538 | } | |
1539 | ||
1540 | return ret; | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1545 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1546 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1547 | * applicable one is SPLICE_F_NONBLOCK. | |
1548 | */ | |
1549 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1550 | unsigned int flags) | |
1551 | { | |
0f7fc9e4 JJS |
1552 | struct pipe_inode_info *ipipe = pipe_info(in->f_path.dentry->d_inode); |
1553 | struct pipe_inode_info *opipe = pipe_info(out->f_path.dentry->d_inode); | |
aadd06e5 | 1554 | int ret = -EINVAL; |
70524490 JA |
1555 | |
1556 | /* | |
aadd06e5 JA |
1557 | * Duplicate the contents of ipipe to opipe without actually |
1558 | * copying the data. | |
70524490 | 1559 | */ |
aadd06e5 JA |
1560 | if (ipipe && opipe && ipipe != opipe) { |
1561 | /* | |
1562 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1563 | * ordering doesn't really matter. | |
1564 | */ | |
1565 | ret = link_ipipe_prep(ipipe, flags); | |
1566 | if (!ret) { | |
1567 | ret = link_opipe_prep(opipe, flags); | |
1568 | if (!ret) { | |
1569 | ret = link_pipe(ipipe, opipe, len, flags); | |
1570 | if (!ret && (flags & SPLICE_F_NONBLOCK)) | |
1571 | ret = -EAGAIN; | |
1572 | } | |
1573 | } | |
1574 | } | |
70524490 | 1575 | |
aadd06e5 | 1576 | return ret; |
70524490 JA |
1577 | } |
1578 | ||
1579 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1580 | { | |
1581 | struct file *in; | |
1582 | int error, fput_in; | |
1583 | ||
1584 | if (unlikely(!len)) | |
1585 | return 0; | |
1586 | ||
1587 | error = -EBADF; | |
1588 | in = fget_light(fdin, &fput_in); | |
1589 | if (in) { | |
1590 | if (in->f_mode & FMODE_READ) { | |
1591 | int fput_out; | |
1592 | struct file *out = fget_light(fdout, &fput_out); | |
1593 | ||
1594 | if (out) { | |
1595 | if (out->f_mode & FMODE_WRITE) | |
1596 | error = do_tee(in, out, len, flags); | |
1597 | fput_light(out, fput_out); | |
1598 | } | |
1599 | } | |
1600 | fput_light(in, fput_in); | |
1601 | } | |
1602 | ||
1603 | return error; | |
1604 | } |