Commit | Line | Data |
---|---|---|
5274f052 JA |
1 | /* |
2 | * "splice": joining two ropes together by interweaving their strands. | |
3 | * | |
4 | * This is the "extended pipe" functionality, where a pipe is used as | |
5 | * an arbitrary in-memory buffer. Think of a pipe as a small kernel | |
6 | * buffer that you can use to transfer data from one end to the other. | |
7 | * | |
8 | * The traditional unix read/write is extended with a "splice()" operation | |
9 | * that transfers data buffers to or from a pipe buffer. | |
10 | * | |
11 | * Named by Larry McVoy, original implementation from Linus, extended by | |
c2058e06 JA |
12 | * Jens to support splicing to files, network, direct splicing, etc and |
13 | * fixing lots of bugs. | |
5274f052 | 14 | * |
0fe23479 | 15 | * Copyright (C) 2005-2006 Jens Axboe <axboe@kernel.dk> |
c2058e06 JA |
16 | * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> |
17 | * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> | |
5274f052 JA |
18 | * |
19 | */ | |
20 | #include <linux/fs.h> | |
21 | #include <linux/file.h> | |
22 | #include <linux/pagemap.h> | |
d6b29d7c | 23 | #include <linux/splice.h> |
08e552c6 | 24 | #include <linux/memcontrol.h> |
5274f052 | 25 | #include <linux/mm_inline.h> |
5abc97aa | 26 | #include <linux/swap.h> |
4f6f0bd2 | 27 | #include <linux/writeback.h> |
630d9c47 | 28 | #include <linux/export.h> |
4f6f0bd2 | 29 | #include <linux/syscalls.h> |
912d35f8 | 30 | #include <linux/uio.h> |
29ce2058 | 31 | #include <linux/security.h> |
5a0e3ad6 | 32 | #include <linux/gfp.h> |
35f9c09f | 33 | #include <linux/socket.h> |
76b021d0 | 34 | #include <linux/compat.h> |
06ae43f3 | 35 | #include "internal.h" |
5274f052 | 36 | |
83f9135b JA |
37 | /* |
38 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
39 | * a vm helper function, it's already simplified quite a bit by the | |
40 | * addition of remove_mapping(). If success is returned, the caller may | |
41 | * attempt to reuse this page for another destination. | |
42 | */ | |
76ad4d11 | 43 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, |
5abc97aa JA |
44 | struct pipe_buffer *buf) |
45 | { | |
46 | struct page *page = buf->page; | |
9e94cd4f | 47 | struct address_space *mapping; |
5abc97aa | 48 | |
9e0267c2 JA |
49 | lock_page(page); |
50 | ||
9e94cd4f JA |
51 | mapping = page_mapping(page); |
52 | if (mapping) { | |
53 | WARN_ON(!PageUptodate(page)); | |
5abc97aa | 54 | |
9e94cd4f JA |
55 | /* |
56 | * At least for ext2 with nobh option, we need to wait on | |
57 | * writeback completing on this page, since we'll remove it | |
58 | * from the pagecache. Otherwise truncate wont wait on the | |
59 | * page, allowing the disk blocks to be reused by someone else | |
60 | * before we actually wrote our data to them. fs corruption | |
61 | * ensues. | |
62 | */ | |
63 | wait_on_page_writeback(page); | |
ad8d6f0a | 64 | |
266cf658 DH |
65 | if (page_has_private(page) && |
66 | !try_to_release_page(page, GFP_KERNEL)) | |
ca39d651 | 67 | goto out_unlock; |
4f6f0bd2 | 68 | |
9e94cd4f JA |
69 | /* |
70 | * If we succeeded in removing the mapping, set LRU flag | |
71 | * and return good. | |
72 | */ | |
73 | if (remove_mapping(mapping, page)) { | |
74 | buf->flags |= PIPE_BUF_FLAG_LRU; | |
75 | return 0; | |
76 | } | |
9e0267c2 | 77 | } |
5abc97aa | 78 | |
9e94cd4f JA |
79 | /* |
80 | * Raced with truncate or failed to remove page from current | |
81 | * address space, unlock and return failure. | |
82 | */ | |
ca39d651 | 83 | out_unlock: |
9e94cd4f JA |
84 | unlock_page(page); |
85 | return 1; | |
5abc97aa JA |
86 | } |
87 | ||
76ad4d11 | 88 | static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, |
5274f052 JA |
89 | struct pipe_buffer *buf) |
90 | { | |
91 | page_cache_release(buf->page); | |
1432873a | 92 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
5274f052 JA |
93 | } |
94 | ||
0845718d JA |
95 | /* |
96 | * Check whether the contents of buf is OK to access. Since the content | |
97 | * is a page cache page, IO may be in flight. | |
98 | */ | |
cac36bb0 JA |
99 | static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, |
100 | struct pipe_buffer *buf) | |
5274f052 JA |
101 | { |
102 | struct page *page = buf->page; | |
49d0b21b | 103 | int err; |
5274f052 JA |
104 | |
105 | if (!PageUptodate(page)) { | |
49d0b21b JA |
106 | lock_page(page); |
107 | ||
108 | /* | |
109 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 110 | * splice, if this is the first page. |
49d0b21b JA |
111 | */ |
112 | if (!page->mapping) { | |
113 | err = -ENODATA; | |
114 | goto error; | |
115 | } | |
5274f052 | 116 | |
49d0b21b | 117 | /* |
73d62d83 | 118 | * Uh oh, read-error from disk. |
49d0b21b JA |
119 | */ |
120 | if (!PageUptodate(page)) { | |
121 | err = -EIO; | |
122 | goto error; | |
123 | } | |
124 | ||
125 | /* | |
f84d7519 | 126 | * Page is ok afterall, we are done. |
49d0b21b | 127 | */ |
5274f052 | 128 | unlock_page(page); |
5274f052 JA |
129 | } |
130 | ||
f84d7519 | 131 | return 0; |
49d0b21b JA |
132 | error: |
133 | unlock_page(page); | |
f84d7519 | 134 | return err; |
70524490 JA |
135 | } |
136 | ||
708e3508 | 137 | const struct pipe_buf_operations page_cache_pipe_buf_ops = { |
5274f052 | 138 | .can_merge = 0, |
f84d7519 JA |
139 | .map = generic_pipe_buf_map, |
140 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 141 | .confirm = page_cache_pipe_buf_confirm, |
5274f052 | 142 | .release = page_cache_pipe_buf_release, |
5abc97aa | 143 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 144 | .get = generic_pipe_buf_get, |
5274f052 JA |
145 | }; |
146 | ||
912d35f8 JA |
147 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
148 | struct pipe_buffer *buf) | |
149 | { | |
7afa6fd0 JA |
150 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
151 | return 1; | |
152 | ||
1432873a | 153 | buf->flags |= PIPE_BUF_FLAG_LRU; |
330ab716 | 154 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
155 | } |
156 | ||
d4c3cca9 | 157 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
912d35f8 | 158 | .can_merge = 0, |
f84d7519 JA |
159 | .map = generic_pipe_buf_map, |
160 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 161 | .confirm = generic_pipe_buf_confirm, |
912d35f8 JA |
162 | .release = page_cache_pipe_buf_release, |
163 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 164 | .get = generic_pipe_buf_get, |
912d35f8 JA |
165 | }; |
166 | ||
825cdcb1 NK |
167 | static void wakeup_pipe_readers(struct pipe_inode_info *pipe) |
168 | { | |
169 | smp_mb(); | |
170 | if (waitqueue_active(&pipe->wait)) | |
171 | wake_up_interruptible(&pipe->wait); | |
172 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
173 | } | |
174 | ||
932cc6d4 JA |
175 | /** |
176 | * splice_to_pipe - fill passed data into a pipe | |
177 | * @pipe: pipe to fill | |
178 | * @spd: data to fill | |
179 | * | |
180 | * Description: | |
79685b8d | 181 | * @spd contains a map of pages and len/offset tuples, along with |
932cc6d4 JA |
182 | * the struct pipe_buf_operations associated with these pages. This |
183 | * function will link that data to the pipe. | |
184 | * | |
83f9135b | 185 | */ |
d6b29d7c JA |
186 | ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
187 | struct splice_pipe_desc *spd) | |
5274f052 | 188 | { |
00de00bd | 189 | unsigned int spd_pages = spd->nr_pages; |
912d35f8 | 190 | int ret, do_wakeup, page_nr; |
5274f052 JA |
191 | |
192 | ret = 0; | |
193 | do_wakeup = 0; | |
912d35f8 | 194 | page_nr = 0; |
5274f052 | 195 | |
61e0d47c | 196 | pipe_lock(pipe); |
5274f052 | 197 | |
5274f052 | 198 | for (;;) { |
3a326a2c | 199 | if (!pipe->readers) { |
5274f052 JA |
200 | send_sig(SIGPIPE, current, 0); |
201 | if (!ret) | |
202 | ret = -EPIPE; | |
203 | break; | |
204 | } | |
205 | ||
35f3d14d JA |
206 | if (pipe->nrbufs < pipe->buffers) { |
207 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
3a326a2c | 208 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 209 | |
912d35f8 JA |
210 | buf->page = spd->pages[page_nr]; |
211 | buf->offset = spd->partial[page_nr].offset; | |
212 | buf->len = spd->partial[page_nr].len; | |
497f9625 | 213 | buf->private = spd->partial[page_nr].private; |
912d35f8 | 214 | buf->ops = spd->ops; |
7afa6fd0 JA |
215 | if (spd->flags & SPLICE_F_GIFT) |
216 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
217 | ||
6f767b04 | 218 | pipe->nrbufs++; |
912d35f8 JA |
219 | page_nr++; |
220 | ret += buf->len; | |
221 | ||
6447a3cf | 222 | if (pipe->files) |
6f767b04 | 223 | do_wakeup = 1; |
5274f052 | 224 | |
912d35f8 | 225 | if (!--spd->nr_pages) |
5274f052 | 226 | break; |
35f3d14d | 227 | if (pipe->nrbufs < pipe->buffers) |
5274f052 JA |
228 | continue; |
229 | ||
230 | break; | |
231 | } | |
232 | ||
912d35f8 | 233 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
234 | if (!ret) |
235 | ret = -EAGAIN; | |
236 | break; | |
237 | } | |
238 | ||
5274f052 JA |
239 | if (signal_pending(current)) { |
240 | if (!ret) | |
241 | ret = -ERESTARTSYS; | |
242 | break; | |
243 | } | |
244 | ||
245 | if (do_wakeup) { | |
c0bd1f65 | 246 | smp_mb(); |
3a326a2c IM |
247 | if (waitqueue_active(&pipe->wait)) |
248 | wake_up_interruptible_sync(&pipe->wait); | |
249 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
250 | do_wakeup = 0; |
251 | } | |
252 | ||
3a326a2c IM |
253 | pipe->waiting_writers++; |
254 | pipe_wait(pipe); | |
255 | pipe->waiting_writers--; | |
5274f052 JA |
256 | } |
257 | ||
61e0d47c | 258 | pipe_unlock(pipe); |
5274f052 | 259 | |
825cdcb1 NK |
260 | if (do_wakeup) |
261 | wakeup_pipe_readers(pipe); | |
5274f052 | 262 | |
00de00bd | 263 | while (page_nr < spd_pages) |
bbdfc2f7 | 264 | spd->spd_release(spd, page_nr++); |
5274f052 JA |
265 | |
266 | return ret; | |
267 | } | |
268 | ||
708e3508 | 269 | void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) |
bbdfc2f7 JA |
270 | { |
271 | page_cache_release(spd->pages[i]); | |
272 | } | |
273 | ||
35f3d14d JA |
274 | /* |
275 | * Check if we need to grow the arrays holding pages and partial page | |
276 | * descriptions. | |
277 | */ | |
047fe360 | 278 | int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) |
35f3d14d | 279 | { |
047fe360 ED |
280 | unsigned int buffers = ACCESS_ONCE(pipe->buffers); |
281 | ||
282 | spd->nr_pages_max = buffers; | |
283 | if (buffers <= PIPE_DEF_BUFFERS) | |
35f3d14d JA |
284 | return 0; |
285 | ||
047fe360 ED |
286 | spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL); |
287 | spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL); | |
35f3d14d JA |
288 | |
289 | if (spd->pages && spd->partial) | |
290 | return 0; | |
291 | ||
292 | kfree(spd->pages); | |
293 | kfree(spd->partial); | |
294 | return -ENOMEM; | |
295 | } | |
296 | ||
047fe360 | 297 | void splice_shrink_spd(struct splice_pipe_desc *spd) |
35f3d14d | 298 | { |
047fe360 | 299 | if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) |
35f3d14d JA |
300 | return; |
301 | ||
302 | kfree(spd->pages); | |
303 | kfree(spd->partial); | |
304 | } | |
305 | ||
3a326a2c | 306 | static int |
cbb7e577 JA |
307 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
308 | struct pipe_inode_info *pipe, size_t len, | |
309 | unsigned int flags) | |
5274f052 JA |
310 | { |
311 | struct address_space *mapping = in->f_mapping; | |
d8983910 | 312 | unsigned int loff, nr_pages, req_pages; |
35f3d14d JA |
313 | struct page *pages[PIPE_DEF_BUFFERS]; |
314 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
5274f052 | 315 | struct page *page; |
91ad66ef JA |
316 | pgoff_t index, end_index; |
317 | loff_t isize; | |
eb20796b | 318 | int error, page_nr; |
912d35f8 JA |
319 | struct splice_pipe_desc spd = { |
320 | .pages = pages, | |
321 | .partial = partial, | |
047fe360 | 322 | .nr_pages_max = PIPE_DEF_BUFFERS, |
912d35f8 JA |
323 | .flags = flags, |
324 | .ops = &page_cache_pipe_buf_ops, | |
bbdfc2f7 | 325 | .spd_release = spd_release_page, |
912d35f8 | 326 | }; |
5274f052 | 327 | |
35f3d14d JA |
328 | if (splice_grow_spd(pipe, &spd)) |
329 | return -ENOMEM; | |
330 | ||
cbb7e577 | 331 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 | 332 | loff = *ppos & ~PAGE_CACHE_MASK; |
d8983910 | 333 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
047fe360 | 334 | nr_pages = min(req_pages, spd.nr_pages_max); |
5274f052 | 335 | |
eb20796b JA |
336 | /* |
337 | * Lookup the (hopefully) full range of pages we need. | |
338 | */ | |
35f3d14d | 339 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages); |
431a4820 | 340 | index += spd.nr_pages; |
82aa5d61 | 341 | |
eb20796b JA |
342 | /* |
343 | * If find_get_pages_contig() returned fewer pages than we needed, | |
431a4820 | 344 | * readahead/allocate the rest and fill in the holes. |
eb20796b | 345 | */ |
431a4820 | 346 | if (spd.nr_pages < nr_pages) |
cf914a7d RR |
347 | page_cache_sync_readahead(mapping, &in->f_ra, in, |
348 | index, req_pages - spd.nr_pages); | |
431a4820 | 349 | |
932cc6d4 | 350 | error = 0; |
eb20796b | 351 | while (spd.nr_pages < nr_pages) { |
82aa5d61 | 352 | /* |
eb20796b JA |
353 | * Page could be there, find_get_pages_contig() breaks on |
354 | * the first hole. | |
5274f052 | 355 | */ |
7480a904 JA |
356 | page = find_get_page(mapping, index); |
357 | if (!page) { | |
7480a904 | 358 | /* |
eb20796b | 359 | * page didn't exist, allocate one. |
7480a904 JA |
360 | */ |
361 | page = page_cache_alloc_cold(mapping); | |
362 | if (!page) | |
363 | break; | |
364 | ||
365 | error = add_to_page_cache_lru(page, mapping, index, | |
0ae0b5d0 | 366 | GFP_KERNEL); |
7480a904 JA |
367 | if (unlikely(error)) { |
368 | page_cache_release(page); | |
a0548871 JA |
369 | if (error == -EEXIST) |
370 | continue; | |
7480a904 JA |
371 | break; |
372 | } | |
eb20796b JA |
373 | /* |
374 | * add_to_page_cache() locks the page, unlock it | |
375 | * to avoid convoluting the logic below even more. | |
376 | */ | |
377 | unlock_page(page); | |
7480a904 JA |
378 | } |
379 | ||
35f3d14d | 380 | spd.pages[spd.nr_pages++] = page; |
eb20796b JA |
381 | index++; |
382 | } | |
383 | ||
384 | /* | |
385 | * Now loop over the map and see if we need to start IO on any | |
386 | * pages, fill in the partial map, etc. | |
387 | */ | |
388 | index = *ppos >> PAGE_CACHE_SHIFT; | |
389 | nr_pages = spd.nr_pages; | |
390 | spd.nr_pages = 0; | |
391 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
392 | unsigned int this_len; | |
393 | ||
394 | if (!len) | |
395 | break; | |
396 | ||
397 | /* | |
398 | * this_len is the max we'll use from this page | |
399 | */ | |
400 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
35f3d14d | 401 | page = spd.pages[page_nr]; |
eb20796b | 402 | |
a08a166f | 403 | if (PageReadahead(page)) |
cf914a7d | 404 | page_cache_async_readahead(mapping, &in->f_ra, in, |
d8983910 | 405 | page, index, req_pages - page_nr); |
a08a166f | 406 | |
7480a904 JA |
407 | /* |
408 | * If the page isn't uptodate, we may need to start io on it | |
409 | */ | |
410 | if (!PageUptodate(page)) { | |
6965031d | 411 | lock_page(page); |
7480a904 JA |
412 | |
413 | /* | |
32502b84 MS |
414 | * Page was truncated, or invalidated by the |
415 | * filesystem. Redo the find/create, but this time the | |
416 | * page is kept locked, so there's no chance of another | |
417 | * race with truncate/invalidate. | |
7480a904 JA |
418 | */ |
419 | if (!page->mapping) { | |
420 | unlock_page(page); | |
32502b84 MS |
421 | page = find_or_create_page(mapping, index, |
422 | mapping_gfp_mask(mapping)); | |
423 | ||
424 | if (!page) { | |
425 | error = -ENOMEM; | |
426 | break; | |
427 | } | |
35f3d14d JA |
428 | page_cache_release(spd.pages[page_nr]); |
429 | spd.pages[page_nr] = page; | |
7480a904 JA |
430 | } |
431 | /* | |
432 | * page was already under io and is now done, great | |
433 | */ | |
434 | if (PageUptodate(page)) { | |
435 | unlock_page(page); | |
436 | goto fill_it; | |
437 | } | |
5274f052 | 438 | |
7480a904 JA |
439 | /* |
440 | * need to read in the page | |
441 | */ | |
442 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 443 | if (unlikely(error)) { |
eb20796b JA |
444 | /* |
445 | * We really should re-lookup the page here, | |
446 | * but it complicates things a lot. Instead | |
447 | * lets just do what we already stored, and | |
448 | * we'll get it the next time we are called. | |
449 | */ | |
7480a904 | 450 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
451 | error = 0; |
452 | ||
5274f052 JA |
453 | break; |
454 | } | |
620a324b JA |
455 | } |
456 | fill_it: | |
457 | /* | |
458 | * i_size must be checked after PageUptodate. | |
459 | */ | |
460 | isize = i_size_read(mapping->host); | |
461 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
462 | if (unlikely(!isize || index > end_index)) | |
463 | break; | |
464 | ||
465 | /* | |
466 | * if this is the last page, see if we need to shrink | |
467 | * the length and stop | |
468 | */ | |
469 | if (end_index == index) { | |
470 | unsigned int plen; | |
91ad66ef JA |
471 | |
472 | /* | |
620a324b | 473 | * max good bytes in this page |
91ad66ef | 474 | */ |
620a324b JA |
475 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
476 | if (plen <= loff) | |
91ad66ef | 477 | break; |
91ad66ef JA |
478 | |
479 | /* | |
620a324b | 480 | * force quit after adding this page |
91ad66ef | 481 | */ |
620a324b JA |
482 | this_len = min(this_len, plen - loff); |
483 | len = this_len; | |
5274f052 | 484 | } |
620a324b | 485 | |
35f3d14d JA |
486 | spd.partial[page_nr].offset = loff; |
487 | spd.partial[page_nr].len = this_len; | |
82aa5d61 | 488 | len -= this_len; |
91ad66ef | 489 | loff = 0; |
eb20796b JA |
490 | spd.nr_pages++; |
491 | index++; | |
5274f052 JA |
492 | } |
493 | ||
eb20796b | 494 | /* |
475ecade | 495 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
496 | * we got, 'nr_pages' is how many pages are in the map. |
497 | */ | |
498 | while (page_nr < nr_pages) | |
35f3d14d | 499 | page_cache_release(spd.pages[page_nr++]); |
f4e6b498 | 500 | in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; |
eb20796b | 501 | |
912d35f8 | 502 | if (spd.nr_pages) |
35f3d14d | 503 | error = splice_to_pipe(pipe, &spd); |
5274f052 | 504 | |
047fe360 | 505 | splice_shrink_spd(&spd); |
7480a904 | 506 | return error; |
5274f052 JA |
507 | } |
508 | ||
83f9135b JA |
509 | /** |
510 | * generic_file_splice_read - splice data from file to a pipe | |
511 | * @in: file to splice from | |
932cc6d4 | 512 | * @ppos: position in @in |
83f9135b JA |
513 | * @pipe: pipe to splice to |
514 | * @len: number of bytes to splice | |
515 | * @flags: splice modifier flags | |
516 | * | |
932cc6d4 JA |
517 | * Description: |
518 | * Will read pages from given file and fill them into a pipe. Can be | |
519 | * used as long as the address_space operations for the source implements | |
520 | * a readpage() hook. | |
521 | * | |
83f9135b | 522 | */ |
cbb7e577 JA |
523 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
524 | struct pipe_inode_info *pipe, size_t len, | |
525 | unsigned int flags) | |
5274f052 | 526 | { |
d366d398 | 527 | loff_t isize, left; |
8191ecd1 | 528 | int ret; |
d366d398 JA |
529 | |
530 | isize = i_size_read(in->f_mapping->host); | |
531 | if (unlikely(*ppos >= isize)) | |
532 | return 0; | |
533 | ||
534 | left = isize - *ppos; | |
535 | if (unlikely(left < len)) | |
536 | len = left; | |
5274f052 | 537 | |
8191ecd1 | 538 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
723590ed | 539 | if (ret > 0) { |
cbb7e577 | 540 | *ppos += ret; |
723590ed MS |
541 | file_accessed(in); |
542 | } | |
5274f052 JA |
543 | |
544 | return ret; | |
545 | } | |
059a8f37 JA |
546 | EXPORT_SYMBOL(generic_file_splice_read); |
547 | ||
6818173b MS |
548 | static const struct pipe_buf_operations default_pipe_buf_ops = { |
549 | .can_merge = 0, | |
550 | .map = generic_pipe_buf_map, | |
551 | .unmap = generic_pipe_buf_unmap, | |
552 | .confirm = generic_pipe_buf_confirm, | |
553 | .release = generic_pipe_buf_release, | |
554 | .steal = generic_pipe_buf_steal, | |
555 | .get = generic_pipe_buf_get, | |
556 | }; | |
557 | ||
d840f989 MS |
558 | static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, |
559 | struct pipe_buffer *buf) | |
560 | { | |
561 | return 1; | |
562 | } | |
563 | ||
564 | /* Pipe buffer operations for a socket and similar. */ | |
565 | const struct pipe_buf_operations nosteal_pipe_buf_ops = { | |
566 | .can_merge = 0, | |
567 | .map = generic_pipe_buf_map, | |
568 | .unmap = generic_pipe_buf_unmap, | |
569 | .confirm = generic_pipe_buf_confirm, | |
570 | .release = generic_pipe_buf_release, | |
571 | .steal = generic_pipe_buf_nosteal, | |
572 | .get = generic_pipe_buf_get, | |
573 | }; | |
574 | EXPORT_SYMBOL(nosteal_pipe_buf_ops); | |
575 | ||
6818173b MS |
576 | static ssize_t kernel_readv(struct file *file, const struct iovec *vec, |
577 | unsigned long vlen, loff_t offset) | |
578 | { | |
579 | mm_segment_t old_fs; | |
580 | loff_t pos = offset; | |
581 | ssize_t res; | |
582 | ||
583 | old_fs = get_fs(); | |
584 | set_fs(get_ds()); | |
585 | /* The cast to a user pointer is valid due to the set_fs() */ | |
586 | res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos); | |
587 | set_fs(old_fs); | |
588 | ||
589 | return res; | |
590 | } | |
591 | ||
7bb307e8 | 592 | ssize_t kernel_write(struct file *file, const char *buf, size_t count, |
b2858d7d | 593 | loff_t pos) |
0b0a47f5 MS |
594 | { |
595 | mm_segment_t old_fs; | |
596 | ssize_t res; | |
597 | ||
598 | old_fs = get_fs(); | |
599 | set_fs(get_ds()); | |
600 | /* The cast to a user pointer is valid due to the set_fs() */ | |
7bb307e8 | 601 | res = vfs_write(file, (__force const char __user *)buf, count, &pos); |
0b0a47f5 MS |
602 | set_fs(old_fs); |
603 | ||
604 | return res; | |
605 | } | |
7bb307e8 | 606 | EXPORT_SYMBOL(kernel_write); |
0b0a47f5 | 607 | |
6818173b MS |
608 | ssize_t default_file_splice_read(struct file *in, loff_t *ppos, |
609 | struct pipe_inode_info *pipe, size_t len, | |
610 | unsigned int flags) | |
611 | { | |
612 | unsigned int nr_pages; | |
613 | unsigned int nr_freed; | |
614 | size_t offset; | |
35f3d14d JA |
615 | struct page *pages[PIPE_DEF_BUFFERS]; |
616 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
617 | struct iovec *vec, __vec[PIPE_DEF_BUFFERS]; | |
6818173b MS |
618 | ssize_t res; |
619 | size_t this_len; | |
620 | int error; | |
621 | int i; | |
622 | struct splice_pipe_desc spd = { | |
623 | .pages = pages, | |
624 | .partial = partial, | |
047fe360 | 625 | .nr_pages_max = PIPE_DEF_BUFFERS, |
6818173b MS |
626 | .flags = flags, |
627 | .ops = &default_pipe_buf_ops, | |
628 | .spd_release = spd_release_page, | |
629 | }; | |
630 | ||
35f3d14d JA |
631 | if (splice_grow_spd(pipe, &spd)) |
632 | return -ENOMEM; | |
633 | ||
634 | res = -ENOMEM; | |
635 | vec = __vec; | |
047fe360 ED |
636 | if (spd.nr_pages_max > PIPE_DEF_BUFFERS) { |
637 | vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL); | |
35f3d14d JA |
638 | if (!vec) |
639 | goto shrink_ret; | |
640 | } | |
641 | ||
6818173b MS |
642 | offset = *ppos & ~PAGE_CACHE_MASK; |
643 | nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
644 | ||
047fe360 | 645 | for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) { |
6818173b MS |
646 | struct page *page; |
647 | ||
4f231228 | 648 | page = alloc_page(GFP_USER); |
6818173b MS |
649 | error = -ENOMEM; |
650 | if (!page) | |
651 | goto err; | |
652 | ||
653 | this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset); | |
4f231228 | 654 | vec[i].iov_base = (void __user *) page_address(page); |
6818173b | 655 | vec[i].iov_len = this_len; |
35f3d14d | 656 | spd.pages[i] = page; |
6818173b MS |
657 | spd.nr_pages++; |
658 | len -= this_len; | |
659 | offset = 0; | |
660 | } | |
661 | ||
662 | res = kernel_readv(in, vec, spd.nr_pages, *ppos); | |
77f6bf57 AM |
663 | if (res < 0) { |
664 | error = res; | |
6818173b | 665 | goto err; |
77f6bf57 | 666 | } |
6818173b MS |
667 | |
668 | error = 0; | |
669 | if (!res) | |
670 | goto err; | |
671 | ||
672 | nr_freed = 0; | |
673 | for (i = 0; i < spd.nr_pages; i++) { | |
6818173b | 674 | this_len = min_t(size_t, vec[i].iov_len, res); |
35f3d14d JA |
675 | spd.partial[i].offset = 0; |
676 | spd.partial[i].len = this_len; | |
6818173b | 677 | if (!this_len) { |
35f3d14d JA |
678 | __free_page(spd.pages[i]); |
679 | spd.pages[i] = NULL; | |
6818173b MS |
680 | nr_freed++; |
681 | } | |
682 | res -= this_len; | |
683 | } | |
684 | spd.nr_pages -= nr_freed; | |
685 | ||
686 | res = splice_to_pipe(pipe, &spd); | |
687 | if (res > 0) | |
688 | *ppos += res; | |
689 | ||
35f3d14d JA |
690 | shrink_ret: |
691 | if (vec != __vec) | |
692 | kfree(vec); | |
047fe360 | 693 | splice_shrink_spd(&spd); |
6818173b MS |
694 | return res; |
695 | ||
696 | err: | |
4f231228 | 697 | for (i = 0; i < spd.nr_pages; i++) |
35f3d14d | 698 | __free_page(spd.pages[i]); |
4f231228 | 699 | |
35f3d14d JA |
700 | res = error; |
701 | goto shrink_ret; | |
6818173b MS |
702 | } |
703 | EXPORT_SYMBOL(default_file_splice_read); | |
704 | ||
5274f052 | 705 | /* |
4f6f0bd2 | 706 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 707 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 708 | */ |
76ad4d11 | 709 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
710 | struct pipe_buffer *buf, struct splice_desc *sd) |
711 | { | |
6a14b90b | 712 | struct file *file = sd->u.file; |
5274f052 | 713 | loff_t pos = sd->pos; |
a8adbe37 | 714 | int more; |
5274f052 | 715 | |
a8adbe37 MM |
716 | if (!likely(file->f_op && file->f_op->sendpage)) |
717 | return -EINVAL; | |
718 | ||
35f9c09f | 719 | more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; |
ae62ca7b ED |
720 | |
721 | if (sd->len < sd->total_len && pipe->nrbufs > 1) | |
35f9c09f | 722 | more |= MSG_SENDPAGE_NOTLAST; |
ae62ca7b | 723 | |
a8adbe37 MM |
724 | return file->f_op->sendpage(file, buf->page, buf->offset, |
725 | sd->len, &pos, more); | |
5274f052 JA |
726 | } |
727 | ||
728 | /* | |
729 | * This is a little more tricky than the file -> pipe splicing. There are | |
730 | * basically three cases: | |
731 | * | |
732 | * - Destination page already exists in the address space and there | |
733 | * are users of it. For that case we have no other option that | |
734 | * copying the data. Tough luck. | |
735 | * - Destination page already exists in the address space, but there | |
736 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
737 | * through to last case. | |
738 | * - Destination page does not exist, we can add the pipe page to | |
739 | * the page cache and avoid the copy. | |
740 | * | |
83f9135b JA |
741 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
742 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
743 | * file address space page cache. This is possible if no one else has | |
744 | * the pipe page referenced outside of the pipe and page cache. If | |
745 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
746 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 | 747 | */ |
328eaaba MS |
748 | int pipe_to_file(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
749 | struct splice_desc *sd) | |
5274f052 | 750 | { |
6a14b90b | 751 | struct file *file = sd->u.file; |
5274f052 | 752 | struct address_space *mapping = file->f_mapping; |
016b661e | 753 | unsigned int offset, this_len; |
5274f052 | 754 | struct page *page; |
afddba49 | 755 | void *fsdata; |
3e7ee3e7 | 756 | int ret; |
5274f052 | 757 | |
5274f052 JA |
758 | offset = sd->pos & ~PAGE_CACHE_MASK; |
759 | ||
016b661e JA |
760 | this_len = sd->len; |
761 | if (this_len + offset > PAGE_CACHE_SIZE) | |
762 | this_len = PAGE_CACHE_SIZE - offset; | |
763 | ||
afddba49 NP |
764 | ret = pagecache_write_begin(file, mapping, sd->pos, this_len, |
765 | AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata); | |
766 | if (unlikely(ret)) | |
767 | goto out; | |
5274f052 | 768 | |
0568b409 | 769 | if (buf->page != page) { |
76ad4d11 | 770 | char *src = buf->ops->map(pipe, buf, 1); |
e8e3c3d6 | 771 | char *dst = kmap_atomic(page); |
5abc97aa | 772 | |
016b661e | 773 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 774 | flush_dcache_page(page); |
e8e3c3d6 | 775 | kunmap_atomic(dst); |
76ad4d11 | 776 | buf->ops->unmap(pipe, buf, src); |
5abc97aa | 777 | } |
afddba49 NP |
778 | ret = pagecache_write_end(file, mapping, sd->pos, this_len, this_len, |
779 | page, fsdata); | |
5274f052 | 780 | out: |
5274f052 JA |
781 | return ret; |
782 | } | |
328eaaba | 783 | EXPORT_SYMBOL(pipe_to_file); |
5274f052 | 784 | |
b3c2d2dd MS |
785 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
786 | { | |
787 | smp_mb(); | |
788 | if (waitqueue_active(&pipe->wait)) | |
789 | wake_up_interruptible(&pipe->wait); | |
790 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
791 | } | |
792 | ||
932cc6d4 | 793 | /** |
b3c2d2dd | 794 | * splice_from_pipe_feed - feed available data from a pipe to a file |
932cc6d4 JA |
795 | * @pipe: pipe to splice from |
796 | * @sd: information to @actor | |
797 | * @actor: handler that splices the data | |
798 | * | |
799 | * Description: | |
b3c2d2dd MS |
800 | * This function loops over the pipe and calls @actor to do the |
801 | * actual moving of a single struct pipe_buffer to the desired | |
802 | * destination. It returns when there's no more buffers left in | |
803 | * the pipe or if the requested number of bytes (@sd->total_len) | |
804 | * have been copied. It returns a positive number (one) if the | |
805 | * pipe needs to be filled with more data, zero if the required | |
806 | * number of bytes have been copied and -errno on error. | |
932cc6d4 | 807 | * |
b3c2d2dd MS |
808 | * This, together with splice_from_pipe_{begin,end,next}, may be |
809 | * used to implement the functionality of __splice_from_pipe() when | |
810 | * locking is required around copying the pipe buffers to the | |
811 | * destination. | |
83f9135b | 812 | */ |
b3c2d2dd MS |
813 | int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
814 | splice_actor *actor) | |
5274f052 | 815 | { |
b3c2d2dd | 816 | int ret; |
5274f052 | 817 | |
b3c2d2dd MS |
818 | while (pipe->nrbufs) { |
819 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
820 | const struct pipe_buf_operations *ops = buf->ops; | |
5274f052 | 821 | |
b3c2d2dd MS |
822 | sd->len = buf->len; |
823 | if (sd->len > sd->total_len) | |
824 | sd->len = sd->total_len; | |
5274f052 | 825 | |
a8adbe37 MM |
826 | ret = buf->ops->confirm(pipe, buf); |
827 | if (unlikely(ret)) { | |
b3c2d2dd MS |
828 | if (ret == -ENODATA) |
829 | ret = 0; | |
830 | return ret; | |
831 | } | |
a8adbe37 MM |
832 | |
833 | ret = actor(pipe, buf, sd); | |
834 | if (ret <= 0) | |
835 | return ret; | |
836 | ||
b3c2d2dd MS |
837 | buf->offset += ret; |
838 | buf->len -= ret; | |
839 | ||
840 | sd->num_spliced += ret; | |
841 | sd->len -= ret; | |
842 | sd->pos += ret; | |
843 | sd->total_len -= ret; | |
844 | ||
845 | if (!buf->len) { | |
846 | buf->ops = NULL; | |
847 | ops->release(pipe, buf); | |
35f3d14d | 848 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); |
b3c2d2dd | 849 | pipe->nrbufs--; |
6447a3cf | 850 | if (pipe->files) |
b3c2d2dd MS |
851 | sd->need_wakeup = true; |
852 | } | |
5274f052 | 853 | |
b3c2d2dd MS |
854 | if (!sd->total_len) |
855 | return 0; | |
856 | } | |
5274f052 | 857 | |
b3c2d2dd MS |
858 | return 1; |
859 | } | |
860 | EXPORT_SYMBOL(splice_from_pipe_feed); | |
5274f052 | 861 | |
b3c2d2dd MS |
862 | /** |
863 | * splice_from_pipe_next - wait for some data to splice from | |
864 | * @pipe: pipe to splice from | |
865 | * @sd: information about the splice operation | |
866 | * | |
867 | * Description: | |
868 | * This function will wait for some data and return a positive | |
869 | * value (one) if pipe buffers are available. It will return zero | |
870 | * or -errno if no more data needs to be spliced. | |
871 | */ | |
872 | int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
873 | { | |
874 | while (!pipe->nrbufs) { | |
875 | if (!pipe->writers) | |
876 | return 0; | |
016b661e | 877 | |
b3c2d2dd MS |
878 | if (!pipe->waiting_writers && sd->num_spliced) |
879 | return 0; | |
73d62d83 | 880 | |
b3c2d2dd MS |
881 | if (sd->flags & SPLICE_F_NONBLOCK) |
882 | return -EAGAIN; | |
5274f052 | 883 | |
b3c2d2dd MS |
884 | if (signal_pending(current)) |
885 | return -ERESTARTSYS; | |
5274f052 | 886 | |
b3c2d2dd MS |
887 | if (sd->need_wakeup) { |
888 | wakeup_pipe_writers(pipe); | |
889 | sd->need_wakeup = false; | |
5274f052 JA |
890 | } |
891 | ||
b3c2d2dd MS |
892 | pipe_wait(pipe); |
893 | } | |
29e35094 | 894 | |
b3c2d2dd MS |
895 | return 1; |
896 | } | |
897 | EXPORT_SYMBOL(splice_from_pipe_next); | |
5274f052 | 898 | |
b3c2d2dd MS |
899 | /** |
900 | * splice_from_pipe_begin - start splicing from pipe | |
b80901bb | 901 | * @sd: information about the splice operation |
b3c2d2dd MS |
902 | * |
903 | * Description: | |
904 | * This function should be called before a loop containing | |
905 | * splice_from_pipe_next() and splice_from_pipe_feed() to | |
906 | * initialize the necessary fields of @sd. | |
907 | */ | |
908 | void splice_from_pipe_begin(struct splice_desc *sd) | |
909 | { | |
910 | sd->num_spliced = 0; | |
911 | sd->need_wakeup = false; | |
912 | } | |
913 | EXPORT_SYMBOL(splice_from_pipe_begin); | |
5274f052 | 914 | |
b3c2d2dd MS |
915 | /** |
916 | * splice_from_pipe_end - finish splicing from pipe | |
917 | * @pipe: pipe to splice from | |
918 | * @sd: information about the splice operation | |
919 | * | |
920 | * Description: | |
921 | * This function will wake up pipe writers if necessary. It should | |
922 | * be called after a loop containing splice_from_pipe_next() and | |
923 | * splice_from_pipe_feed(). | |
924 | */ | |
925 | void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) | |
926 | { | |
927 | if (sd->need_wakeup) | |
928 | wakeup_pipe_writers(pipe); | |
929 | } | |
930 | EXPORT_SYMBOL(splice_from_pipe_end); | |
5274f052 | 931 | |
b3c2d2dd MS |
932 | /** |
933 | * __splice_from_pipe - splice data from a pipe to given actor | |
934 | * @pipe: pipe to splice from | |
935 | * @sd: information to @actor | |
936 | * @actor: handler that splices the data | |
937 | * | |
938 | * Description: | |
939 | * This function does little more than loop over the pipe and call | |
940 | * @actor to do the actual moving of a single struct pipe_buffer to | |
941 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
942 | * pipe_to_user. | |
943 | * | |
944 | */ | |
945 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, | |
946 | splice_actor *actor) | |
947 | { | |
948 | int ret; | |
5274f052 | 949 | |
b3c2d2dd MS |
950 | splice_from_pipe_begin(sd); |
951 | do { | |
b63a96fa | 952 | cond_resched(); |
b3c2d2dd MS |
953 | ret = splice_from_pipe_next(pipe, sd); |
954 | if (ret > 0) | |
955 | ret = splice_from_pipe_feed(pipe, sd, actor); | |
956 | } while (ret > 0); | |
957 | splice_from_pipe_end(pipe, sd); | |
958 | ||
959 | return sd->num_spliced ? sd->num_spliced : ret; | |
5274f052 | 960 | } |
40bee44e | 961 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 962 | |
932cc6d4 JA |
963 | /** |
964 | * splice_from_pipe - splice data from a pipe to a file | |
965 | * @pipe: pipe to splice from | |
966 | * @out: file to splice to | |
967 | * @ppos: position in @out | |
968 | * @len: how many bytes to splice | |
969 | * @flags: splice modifier flags | |
970 | * @actor: handler that splices the data | |
971 | * | |
972 | * Description: | |
2933970b | 973 | * See __splice_from_pipe. This function locks the pipe inode, |
932cc6d4 JA |
974 | * otherwise it's identical to __splice_from_pipe(). |
975 | * | |
976 | */ | |
6da61809 MF |
977 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
978 | loff_t *ppos, size_t len, unsigned int flags, | |
979 | splice_actor *actor) | |
980 | { | |
981 | ssize_t ret; | |
c66ab6fa JA |
982 | struct splice_desc sd = { |
983 | .total_len = len, | |
984 | .flags = flags, | |
985 | .pos = *ppos, | |
6a14b90b | 986 | .u.file = out, |
c66ab6fa | 987 | }; |
6da61809 | 988 | |
61e0d47c | 989 | pipe_lock(pipe); |
c66ab6fa | 990 | ret = __splice_from_pipe(pipe, &sd, actor); |
61e0d47c | 991 | pipe_unlock(pipe); |
6da61809 MF |
992 | |
993 | return ret; | |
994 | } | |
995 | ||
83f9135b JA |
996 | /** |
997 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 998 | * @pipe: pipe info |
83f9135b | 999 | * @out: file to write to |
932cc6d4 | 1000 | * @ppos: position in @out |
83f9135b JA |
1001 | * @len: number of bytes to splice |
1002 | * @flags: splice modifier flags | |
1003 | * | |
932cc6d4 JA |
1004 | * Description: |
1005 | * Will either move or copy pages (determined by @flags options) from | |
1006 | * the given pipe inode to the given file. | |
83f9135b JA |
1007 | * |
1008 | */ | |
3a326a2c IM |
1009 | ssize_t |
1010 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 1011 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1012 | { |
4f6f0bd2 | 1013 | struct address_space *mapping = out->f_mapping; |
8c34e2d6 | 1014 | struct inode *inode = mapping->host; |
7f3d4ee1 | 1015 | struct splice_desc sd = { |
7f3d4ee1 | 1016 | .flags = flags, |
7f3d4ee1 MS |
1017 | .u.file = out, |
1018 | }; | |
3a326a2c IM |
1019 | ssize_t ret; |
1020 | ||
13d32f27 BH |
1021 | ret = generic_write_checks(out, ppos, &len, S_ISBLK(inode->i_mode)); |
1022 | if (ret) | |
1023 | return ret; | |
1024 | sd.total_len = len; | |
1025 | sd.pos = *ppos; | |
1026 | ||
61e0d47c | 1027 | pipe_lock(pipe); |
eb443e5a MS |
1028 | |
1029 | splice_from_pipe_begin(&sd); | |
1030 | do { | |
1031 | ret = splice_from_pipe_next(pipe, &sd); | |
1032 | if (ret <= 0) | |
1033 | break; | |
1034 | ||
1035 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); | |
1036 | ret = file_remove_suid(out); | |
723590ed | 1037 | if (!ret) { |
c3b2da31 JB |
1038 | ret = file_update_time(out); |
1039 | if (!ret) | |
1040 | ret = splice_from_pipe_feed(pipe, &sd, | |
1041 | pipe_to_file); | |
723590ed | 1042 | } |
eb443e5a MS |
1043 | mutex_unlock(&inode->i_mutex); |
1044 | } while (ret > 0); | |
1045 | splice_from_pipe_end(pipe, &sd); | |
1046 | ||
61e0d47c | 1047 | pipe_unlock(pipe); |
eb443e5a MS |
1048 | |
1049 | if (sd.num_spliced) | |
1050 | ret = sd.num_spliced; | |
1051 | ||
a4514ebd | 1052 | if (ret > 0) { |
148f948b | 1053 | int err; |
17ee4f49 | 1054 | |
148f948b JK |
1055 | err = generic_write_sync(out, *ppos, ret); |
1056 | if (err) | |
1057 | ret = err; | |
1058 | else | |
1059 | *ppos += ret; | |
d0e1d66b | 1060 | balance_dirty_pages_ratelimited(mapping); |
4f6f0bd2 JA |
1061 | } |
1062 | ||
1063 | return ret; | |
5274f052 JA |
1064 | } |
1065 | ||
059a8f37 JA |
1066 | EXPORT_SYMBOL(generic_file_splice_write); |
1067 | ||
b2858d7d MS |
1068 | static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1069 | struct splice_desc *sd) | |
0b0a47f5 | 1070 | { |
b2858d7d MS |
1071 | int ret; |
1072 | void *data; | |
06ae43f3 | 1073 | loff_t tmp = sd->pos; |
b2858d7d | 1074 | |
b2858d7d | 1075 | data = buf->ops->map(pipe, buf, 0); |
06ae43f3 | 1076 | ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); |
b2858d7d MS |
1077 | buf->ops->unmap(pipe, buf, data); |
1078 | ||
1079 | return ret; | |
0b0a47f5 MS |
1080 | } |
1081 | ||
1082 | static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, | |
1083 | struct file *out, loff_t *ppos, | |
1084 | size_t len, unsigned int flags) | |
1085 | { | |
b2858d7d | 1086 | ssize_t ret; |
0b0a47f5 | 1087 | |
b2858d7d MS |
1088 | ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); |
1089 | if (ret > 0) | |
1090 | *ppos += ret; | |
0b0a47f5 | 1091 | |
b2858d7d | 1092 | return ret; |
0b0a47f5 MS |
1093 | } |
1094 | ||
83f9135b JA |
1095 | /** |
1096 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 1097 | * @pipe: pipe to splice from |
83f9135b | 1098 | * @out: socket to write to |
932cc6d4 | 1099 | * @ppos: position in @out |
83f9135b JA |
1100 | * @len: number of bytes to splice |
1101 | * @flags: splice modifier flags | |
1102 | * | |
932cc6d4 JA |
1103 | * Description: |
1104 | * Will send @len bytes from the pipe to a network socket. No data copying | |
1105 | * is involved. | |
83f9135b JA |
1106 | * |
1107 | */ | |
3a326a2c | 1108 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1109 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1110 | { |
00522fb4 | 1111 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
1112 | } |
1113 | ||
059a8f37 | 1114 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 1115 | |
83f9135b JA |
1116 | /* |
1117 | * Attempt to initiate a splice from pipe to file. | |
1118 | */ | |
3a326a2c | 1119 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1120 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1121 | { |
0b0a47f5 MS |
1122 | ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, |
1123 | loff_t *, size_t, unsigned int); | |
5274f052 JA |
1124 | int ret; |
1125 | ||
49570e9b | 1126 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
1127 | return -EBADF; |
1128 | ||
efc968d4 LT |
1129 | if (unlikely(out->f_flags & O_APPEND)) |
1130 | return -EINVAL; | |
1131 | ||
cbb7e577 | 1132 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
1133 | if (unlikely(ret < 0)) |
1134 | return ret; | |
1135 | ||
cc56f7de CG |
1136 | if (out->f_op && out->f_op->splice_write) |
1137 | splice_write = out->f_op->splice_write; | |
1138 | else | |
0b0a47f5 MS |
1139 | splice_write = default_file_splice_write; |
1140 | ||
2dd8c9ad AV |
1141 | file_start_write(out); |
1142 | ret = splice_write(pipe, out, ppos, len, flags); | |
1143 | file_end_write(out); | |
1144 | return ret; | |
5274f052 JA |
1145 | } |
1146 | ||
83f9135b JA |
1147 | /* |
1148 | * Attempt to initiate a splice from a file to a pipe. | |
1149 | */ | |
cbb7e577 JA |
1150 | static long do_splice_to(struct file *in, loff_t *ppos, |
1151 | struct pipe_inode_info *pipe, size_t len, | |
1152 | unsigned int flags) | |
5274f052 | 1153 | { |
6818173b MS |
1154 | ssize_t (*splice_read)(struct file *, loff_t *, |
1155 | struct pipe_inode_info *, size_t, unsigned int); | |
5274f052 JA |
1156 | int ret; |
1157 | ||
49570e9b | 1158 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
1159 | return -EBADF; |
1160 | ||
cbb7e577 | 1161 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
1162 | if (unlikely(ret < 0)) |
1163 | return ret; | |
1164 | ||
cc56f7de CG |
1165 | if (in->f_op && in->f_op->splice_read) |
1166 | splice_read = in->f_op->splice_read; | |
1167 | else | |
6818173b MS |
1168 | splice_read = default_file_splice_read; |
1169 | ||
1170 | return splice_read(in, ppos, pipe, len, flags); | |
5274f052 JA |
1171 | } |
1172 | ||
932cc6d4 JA |
1173 | /** |
1174 | * splice_direct_to_actor - splices data directly between two non-pipes | |
1175 | * @in: file to splice from | |
1176 | * @sd: actor information on where to splice to | |
1177 | * @actor: handles the data splicing | |
1178 | * | |
1179 | * Description: | |
1180 | * This is a special case helper to splice directly between two | |
1181 | * points, without requiring an explicit pipe. Internally an allocated | |
79685b8d | 1182 | * pipe is cached in the process, and reused during the lifetime of |
932cc6d4 JA |
1183 | * that process. |
1184 | * | |
c66ab6fa JA |
1185 | */ |
1186 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1187 | splice_direct_actor *actor) | |
b92ce558 JA |
1188 | { |
1189 | struct pipe_inode_info *pipe; | |
1190 | long ret, bytes; | |
1191 | umode_t i_mode; | |
c66ab6fa | 1192 | size_t len; |
4c67196f | 1193 | int i, flags, more; |
b92ce558 JA |
1194 | |
1195 | /* | |
1196 | * We require the input being a regular file, as we don't want to | |
1197 | * randomly drop data for eg socket -> socket splicing. Use the | |
1198 | * piped splicing for that! | |
1199 | */ | |
496ad9aa | 1200 | i_mode = file_inode(in)->i_mode; |
b92ce558 JA |
1201 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1202 | return -EINVAL; | |
1203 | ||
1204 | /* | |
1205 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1206 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1207 | */ | |
1208 | pipe = current->splice_pipe; | |
49570e9b | 1209 | if (unlikely(!pipe)) { |
7bee130e | 1210 | pipe = alloc_pipe_info(); |
b92ce558 JA |
1211 | if (!pipe) |
1212 | return -ENOMEM; | |
1213 | ||
1214 | /* | |
1215 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1216 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1217 | * PIPE_READERS appropriately. |
1218 | */ | |
1219 | pipe->readers = 1; | |
1220 | ||
1221 | current->splice_pipe = pipe; | |
1222 | } | |
1223 | ||
1224 | /* | |
73d62d83 | 1225 | * Do the splice. |
b92ce558 JA |
1226 | */ |
1227 | ret = 0; | |
1228 | bytes = 0; | |
c66ab6fa JA |
1229 | len = sd->total_len; |
1230 | flags = sd->flags; | |
1231 | ||
1232 | /* | |
1233 | * Don't block on output, we have to drain the direct pipe. | |
1234 | */ | |
1235 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
4c67196f | 1236 | more = sd->flags & SPLICE_F_MORE; |
b92ce558 JA |
1237 | |
1238 | while (len) { | |
51a92c0f | 1239 | size_t read_len; |
a82c53a0 | 1240 | loff_t pos = sd->pos, prev_pos = pos; |
b92ce558 | 1241 | |
bcd4f3ac | 1242 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1243 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1244 | goto out_release; |
1245 | ||
1246 | read_len = ret; | |
c66ab6fa | 1247 | sd->total_len = read_len; |
b92ce558 | 1248 | |
4c67196f CL |
1249 | /* |
1250 | * If more data is pending, set SPLICE_F_MORE | |
1251 | * If this is the last data and SPLICE_F_MORE was not set | |
1252 | * initially, clears it. | |
1253 | */ | |
1254 | if (read_len < len) | |
1255 | sd->flags |= SPLICE_F_MORE; | |
1256 | else if (!more) | |
1257 | sd->flags &= ~SPLICE_F_MORE; | |
b92ce558 JA |
1258 | /* |
1259 | * NOTE: nonblocking mode only applies to the input. We | |
1260 | * must not do the output in nonblocking mode as then we | |
1261 | * could get stuck data in the internal pipe: | |
1262 | */ | |
c66ab6fa | 1263 | ret = actor(pipe, sd); |
a82c53a0 TZ |
1264 | if (unlikely(ret <= 0)) { |
1265 | sd->pos = prev_pos; | |
b92ce558 | 1266 | goto out_release; |
a82c53a0 | 1267 | } |
b92ce558 JA |
1268 | |
1269 | bytes += ret; | |
1270 | len -= ret; | |
bcd4f3ac | 1271 | sd->pos = pos; |
b92ce558 | 1272 | |
a82c53a0 TZ |
1273 | if (ret < read_len) { |
1274 | sd->pos = prev_pos + ret; | |
51a92c0f | 1275 | goto out_release; |
a82c53a0 | 1276 | } |
b92ce558 JA |
1277 | } |
1278 | ||
9e97198d | 1279 | done: |
b92ce558 | 1280 | pipe->nrbufs = pipe->curbuf = 0; |
80848708 | 1281 | file_accessed(in); |
b92ce558 JA |
1282 | return bytes; |
1283 | ||
1284 | out_release: | |
1285 | /* | |
1286 | * If we did an incomplete transfer we must release | |
1287 | * the pipe buffers in question: | |
1288 | */ | |
35f3d14d | 1289 | for (i = 0; i < pipe->buffers; i++) { |
b92ce558 JA |
1290 | struct pipe_buffer *buf = pipe->bufs + i; |
1291 | ||
1292 | if (buf->ops) { | |
1293 | buf->ops->release(pipe, buf); | |
1294 | buf->ops = NULL; | |
1295 | } | |
1296 | } | |
b92ce558 | 1297 | |
9e97198d JA |
1298 | if (!bytes) |
1299 | bytes = ret; | |
c66ab6fa | 1300 | |
9e97198d | 1301 | goto done; |
c66ab6fa JA |
1302 | } |
1303 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1304 | ||
1305 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1306 | struct splice_desc *sd) | |
1307 | { | |
6a14b90b | 1308 | struct file *file = sd->u.file; |
c66ab6fa | 1309 | |
7995bd28 | 1310 | return do_splice_from(pipe, file, sd->opos, sd->total_len, |
2cb4b05e | 1311 | sd->flags); |
c66ab6fa JA |
1312 | } |
1313 | ||
932cc6d4 JA |
1314 | /** |
1315 | * do_splice_direct - splices data directly between two files | |
1316 | * @in: file to splice from | |
1317 | * @ppos: input file offset | |
1318 | * @out: file to splice to | |
acdb37c3 | 1319 | * @opos: output file offset |
932cc6d4 JA |
1320 | * @len: number of bytes to splice |
1321 | * @flags: splice modifier flags | |
1322 | * | |
1323 | * Description: | |
1324 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1325 | * doing it in the application would incur an extra system call | |
1326 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1327 | * can splice directly through a process-private pipe. | |
1328 | * | |
1329 | */ | |
c66ab6fa | 1330 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
7995bd28 | 1331 | loff_t *opos, size_t len, unsigned int flags) |
c66ab6fa JA |
1332 | { |
1333 | struct splice_desc sd = { | |
1334 | .len = len, | |
1335 | .total_len = len, | |
1336 | .flags = flags, | |
1337 | .pos = *ppos, | |
6a14b90b | 1338 | .u.file = out, |
7995bd28 | 1339 | .opos = opos, |
c66ab6fa | 1340 | }; |
51a92c0f | 1341 | long ret; |
c66ab6fa JA |
1342 | |
1343 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); | |
51a92c0f | 1344 | if (ret > 0) |
a82c53a0 | 1345 | *ppos = sd.pos; |
51a92c0f | 1346 | |
c66ab6fa | 1347 | return ret; |
b92ce558 JA |
1348 | } |
1349 | ||
7c77f0b3 MS |
1350 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
1351 | struct pipe_inode_info *opipe, | |
1352 | size_t len, unsigned int flags); | |
ddac0d39 | 1353 | |
83f9135b JA |
1354 | /* |
1355 | * Determine where to splice to/from. | |
1356 | */ | |
529565dc IM |
1357 | static long do_splice(struct file *in, loff_t __user *off_in, |
1358 | struct file *out, loff_t __user *off_out, | |
1359 | size_t len, unsigned int flags) | |
5274f052 | 1360 | { |
7c77f0b3 MS |
1361 | struct pipe_inode_info *ipipe; |
1362 | struct pipe_inode_info *opipe; | |
7995bd28 | 1363 | loff_t offset; |
a4514ebd | 1364 | long ret; |
5274f052 | 1365 | |
71993e62 LT |
1366 | ipipe = get_pipe_info(in); |
1367 | opipe = get_pipe_info(out); | |
7c77f0b3 MS |
1368 | |
1369 | if (ipipe && opipe) { | |
1370 | if (off_in || off_out) | |
1371 | return -ESPIPE; | |
1372 | ||
1373 | if (!(in->f_mode & FMODE_READ)) | |
1374 | return -EBADF; | |
1375 | ||
1376 | if (!(out->f_mode & FMODE_WRITE)) | |
1377 | return -EBADF; | |
1378 | ||
1379 | /* Splicing to self would be fun, but... */ | |
1380 | if (ipipe == opipe) | |
1381 | return -EINVAL; | |
1382 | ||
1383 | return splice_pipe_to_pipe(ipipe, opipe, len, flags); | |
1384 | } | |
1385 | ||
1386 | if (ipipe) { | |
529565dc IM |
1387 | if (off_in) |
1388 | return -ESPIPE; | |
b92ce558 | 1389 | if (off_out) { |
19c9a49b | 1390 | if (!(out->f_mode & FMODE_PWRITE)) |
b92ce558 | 1391 | return -EINVAL; |
cbb7e577 | 1392 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1393 | return -EFAULT; |
7995bd28 AV |
1394 | } else { |
1395 | offset = out->f_pos; | |
1396 | } | |
529565dc | 1397 | |
7995bd28 | 1398 | ret = do_splice_from(ipipe, out, &offset, len, flags); |
a4514ebd | 1399 | |
7995bd28 AV |
1400 | if (!off_out) |
1401 | out->f_pos = offset; | |
1402 | else if (copy_to_user(off_out, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1403 | ret = -EFAULT; |
1404 | ||
1405 | return ret; | |
529565dc | 1406 | } |
5274f052 | 1407 | |
7c77f0b3 | 1408 | if (opipe) { |
529565dc IM |
1409 | if (off_out) |
1410 | return -ESPIPE; | |
b92ce558 | 1411 | if (off_in) { |
19c9a49b | 1412 | if (!(in->f_mode & FMODE_PREAD)) |
b92ce558 | 1413 | return -EINVAL; |
cbb7e577 | 1414 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1415 | return -EFAULT; |
7995bd28 AV |
1416 | } else { |
1417 | offset = in->f_pos; | |
1418 | } | |
529565dc | 1419 | |
7995bd28 | 1420 | ret = do_splice_to(in, &offset, opipe, len, flags); |
a4514ebd | 1421 | |
7995bd28 AV |
1422 | if (!off_in) |
1423 | in->f_pos = offset; | |
1424 | else if (copy_to_user(off_in, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1425 | ret = -EFAULT; |
1426 | ||
1427 | return ret; | |
529565dc | 1428 | } |
5274f052 JA |
1429 | |
1430 | return -EINVAL; | |
1431 | } | |
1432 | ||
912d35f8 JA |
1433 | /* |
1434 | * Map an iov into an array of pages and offset/length tupples. With the | |
1435 | * partial_page structure, we can map several non-contiguous ranges into | |
1436 | * our ones pages[] map instead of splitting that operation into pieces. | |
1437 | * Could easily be exported as a generic helper for other users, in which | |
1438 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1439 | */ | |
1440 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1441 | unsigned int nr_vecs, struct page **pages, | |
bd1a68b5 | 1442 | struct partial_page *partial, bool aligned, |
35f3d14d | 1443 | unsigned int pipe_buffers) |
912d35f8 JA |
1444 | { |
1445 | int buffers = 0, error = 0; | |
1446 | ||
912d35f8 JA |
1447 | while (nr_vecs) { |
1448 | unsigned long off, npages; | |
75723957 | 1449 | struct iovec entry; |
912d35f8 JA |
1450 | void __user *base; |
1451 | size_t len; | |
1452 | int i; | |
1453 | ||
75723957 | 1454 | error = -EFAULT; |
bc40d73c | 1455 | if (copy_from_user(&entry, iov, sizeof(entry))) |
912d35f8 JA |
1456 | break; |
1457 | ||
75723957 LT |
1458 | base = entry.iov_base; |
1459 | len = entry.iov_len; | |
1460 | ||
912d35f8 JA |
1461 | /* |
1462 | * Sanity check this iovec. 0 read succeeds. | |
1463 | */ | |
75723957 | 1464 | error = 0; |
912d35f8 JA |
1465 | if (unlikely(!len)) |
1466 | break; | |
1467 | error = -EFAULT; | |
712a30e6 | 1468 | if (!access_ok(VERIFY_READ, base, len)) |
912d35f8 JA |
1469 | break; |
1470 | ||
1471 | /* | |
1472 | * Get this base offset and number of pages, then map | |
1473 | * in the user pages. | |
1474 | */ | |
1475 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1476 | |
1477 | /* | |
1478 | * If asked for alignment, the offset must be zero and the | |
1479 | * length a multiple of the PAGE_SIZE. | |
1480 | */ | |
1481 | error = -EINVAL; | |
1482 | if (aligned && (off || len & ~PAGE_MASK)) | |
1483 | break; | |
1484 | ||
912d35f8 | 1485 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
35f3d14d JA |
1486 | if (npages > pipe_buffers - buffers) |
1487 | npages = pipe_buffers - buffers; | |
912d35f8 | 1488 | |
bc40d73c NP |
1489 | error = get_user_pages_fast((unsigned long)base, npages, |
1490 | 0, &pages[buffers]); | |
912d35f8 JA |
1491 | |
1492 | if (unlikely(error <= 0)) | |
1493 | break; | |
1494 | ||
1495 | /* | |
1496 | * Fill this contiguous range into the partial page map. | |
1497 | */ | |
1498 | for (i = 0; i < error; i++) { | |
7591489a | 1499 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1500 | |
1501 | partial[buffers].offset = off; | |
1502 | partial[buffers].len = plen; | |
1503 | ||
1504 | off = 0; | |
1505 | len -= plen; | |
1506 | buffers++; | |
1507 | } | |
1508 | ||
1509 | /* | |
1510 | * We didn't complete this iov, stop here since it probably | |
1511 | * means we have to move some of this into a pipe to | |
1512 | * be able to continue. | |
1513 | */ | |
1514 | if (len) | |
1515 | break; | |
1516 | ||
1517 | /* | |
1518 | * Don't continue if we mapped fewer pages than we asked for, | |
1519 | * or if we mapped the max number of pages that we have | |
1520 | * room for. | |
1521 | */ | |
35f3d14d | 1522 | if (error < npages || buffers == pipe_buffers) |
912d35f8 JA |
1523 | break; |
1524 | ||
1525 | nr_vecs--; | |
1526 | iov++; | |
1527 | } | |
1528 | ||
912d35f8 JA |
1529 | if (buffers) |
1530 | return buffers; | |
1531 | ||
1532 | return error; | |
1533 | } | |
1534 | ||
6a14b90b JA |
1535 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1536 | struct splice_desc *sd) | |
1537 | { | |
1538 | char *src; | |
1539 | int ret; | |
1540 | ||
6a14b90b JA |
1541 | /* |
1542 | * See if we can use the atomic maps, by prefaulting in the | |
1543 | * pages and doing an atomic copy | |
1544 | */ | |
1545 | if (!fault_in_pages_writeable(sd->u.userptr, sd->len)) { | |
1546 | src = buf->ops->map(pipe, buf, 1); | |
1547 | ret = __copy_to_user_inatomic(sd->u.userptr, src + buf->offset, | |
1548 | sd->len); | |
1549 | buf->ops->unmap(pipe, buf, src); | |
1550 | if (!ret) { | |
1551 | ret = sd->len; | |
1552 | goto out; | |
1553 | } | |
1554 | } | |
1555 | ||
1556 | /* | |
1557 | * No dice, use slow non-atomic map and copy | |
1558 | */ | |
1559 | src = buf->ops->map(pipe, buf, 0); | |
1560 | ||
1561 | ret = sd->len; | |
1562 | if (copy_to_user(sd->u.userptr, src + buf->offset, sd->len)) | |
1563 | ret = -EFAULT; | |
1564 | ||
6866bef4 | 1565 | buf->ops->unmap(pipe, buf, src); |
6a14b90b JA |
1566 | out: |
1567 | if (ret > 0) | |
1568 | sd->u.userptr += ret; | |
6a14b90b JA |
1569 | return ret; |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * For lack of a better implementation, implement vmsplice() to userspace | |
1574 | * as a simple copy of the pipes pages to the user iov. | |
1575 | */ | |
1576 | static long vmsplice_to_user(struct file *file, const struct iovec __user *iov, | |
1577 | unsigned long nr_segs, unsigned int flags) | |
1578 | { | |
1579 | struct pipe_inode_info *pipe; | |
1580 | struct splice_desc sd; | |
1581 | ssize_t size; | |
1582 | int error; | |
1583 | long ret; | |
1584 | ||
71993e62 | 1585 | pipe = get_pipe_info(file); |
6a14b90b JA |
1586 | if (!pipe) |
1587 | return -EBADF; | |
1588 | ||
61e0d47c | 1589 | pipe_lock(pipe); |
6a14b90b JA |
1590 | |
1591 | error = ret = 0; | |
1592 | while (nr_segs) { | |
1593 | void __user *base; | |
1594 | size_t len; | |
1595 | ||
1596 | /* | |
1597 | * Get user address base and length for this iovec. | |
1598 | */ | |
1599 | error = get_user(base, &iov->iov_base); | |
1600 | if (unlikely(error)) | |
1601 | break; | |
1602 | error = get_user(len, &iov->iov_len); | |
1603 | if (unlikely(error)) | |
1604 | break; | |
1605 | ||
1606 | /* | |
1607 | * Sanity check this iovec. 0 read succeeds. | |
1608 | */ | |
1609 | if (unlikely(!len)) | |
1610 | break; | |
1611 | if (unlikely(!base)) { | |
1612 | error = -EFAULT; | |
1613 | break; | |
1614 | } | |
1615 | ||
8811930d JA |
1616 | if (unlikely(!access_ok(VERIFY_WRITE, base, len))) { |
1617 | error = -EFAULT; | |
1618 | break; | |
1619 | } | |
1620 | ||
6a14b90b JA |
1621 | sd.len = 0; |
1622 | sd.total_len = len; | |
1623 | sd.flags = flags; | |
1624 | sd.u.userptr = base; | |
1625 | sd.pos = 0; | |
1626 | ||
1627 | size = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1628 | if (size < 0) { | |
1629 | if (!ret) | |
1630 | ret = size; | |
1631 | ||
1632 | break; | |
1633 | } | |
1634 | ||
1635 | ret += size; | |
1636 | ||
1637 | if (size < len) | |
1638 | break; | |
1639 | ||
1640 | nr_segs--; | |
1641 | iov++; | |
1642 | } | |
1643 | ||
61e0d47c | 1644 | pipe_unlock(pipe); |
6a14b90b JA |
1645 | |
1646 | if (!ret) | |
1647 | ret = error; | |
1648 | ||
1649 | return ret; | |
1650 | } | |
1651 | ||
912d35f8 JA |
1652 | /* |
1653 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1654 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1655 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1656 | */ |
6a14b90b JA |
1657 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1658 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1659 | { |
ddac0d39 | 1660 | struct pipe_inode_info *pipe; |
35f3d14d JA |
1661 | struct page *pages[PIPE_DEF_BUFFERS]; |
1662 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
912d35f8 JA |
1663 | struct splice_pipe_desc spd = { |
1664 | .pages = pages, | |
1665 | .partial = partial, | |
047fe360 | 1666 | .nr_pages_max = PIPE_DEF_BUFFERS, |
912d35f8 JA |
1667 | .flags = flags, |
1668 | .ops = &user_page_pipe_buf_ops, | |
bbdfc2f7 | 1669 | .spd_release = spd_release_page, |
912d35f8 | 1670 | }; |
35f3d14d | 1671 | long ret; |
912d35f8 | 1672 | |
71993e62 | 1673 | pipe = get_pipe_info(file); |
ddac0d39 | 1674 | if (!pipe) |
912d35f8 | 1675 | return -EBADF; |
912d35f8 | 1676 | |
35f3d14d JA |
1677 | if (splice_grow_spd(pipe, &spd)) |
1678 | return -ENOMEM; | |
1679 | ||
1680 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages, | |
bd1a68b5 | 1681 | spd.partial, false, |
047fe360 | 1682 | spd.nr_pages_max); |
912d35f8 | 1683 | if (spd.nr_pages <= 0) |
35f3d14d JA |
1684 | ret = spd.nr_pages; |
1685 | else | |
1686 | ret = splice_to_pipe(pipe, &spd); | |
912d35f8 | 1687 | |
047fe360 | 1688 | splice_shrink_spd(&spd); |
35f3d14d | 1689 | return ret; |
912d35f8 JA |
1690 | } |
1691 | ||
6a14b90b JA |
1692 | /* |
1693 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1694 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1695 | * operation that can be supported without any funky alignment restrictions | |
1696 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1697 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1698 | * solutions for that: | |
1699 | * | |
1700 | * - memcpy() the data internally, at which point we might as well just | |
1701 | * do a regular read() on the buffer anyway. | |
1702 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1703 | * has restriction limitations on both ends of the pipe). | |
1704 | * | |
1705 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1706 | * | |
1707 | */ | |
836f92ad HC |
1708 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, |
1709 | unsigned long, nr_segs, unsigned int, flags) | |
912d35f8 | 1710 | { |
2903ff01 | 1711 | struct fd f; |
912d35f8 | 1712 | long error; |
912d35f8 | 1713 | |
6a14b90b JA |
1714 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1715 | return -EINVAL; | |
1716 | else if (unlikely(!nr_segs)) | |
1717 | return 0; | |
1718 | ||
912d35f8 | 1719 | error = -EBADF; |
2903ff01 AV |
1720 | f = fdget(fd); |
1721 | if (f.file) { | |
1722 | if (f.file->f_mode & FMODE_WRITE) | |
1723 | error = vmsplice_to_pipe(f.file, iov, nr_segs, flags); | |
1724 | else if (f.file->f_mode & FMODE_READ) | |
1725 | error = vmsplice_to_user(f.file, iov, nr_segs, flags); | |
1726 | ||
1727 | fdput(f); | |
912d35f8 JA |
1728 | } |
1729 | ||
1730 | return error; | |
1731 | } | |
1732 | ||
76b021d0 AV |
1733 | #ifdef CONFIG_COMPAT |
1734 | COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, | |
1735 | unsigned int, nr_segs, unsigned int, flags) | |
1736 | { | |
1737 | unsigned i; | |
1738 | struct iovec __user *iov; | |
1739 | if (nr_segs > UIO_MAXIOV) | |
1740 | return -EINVAL; | |
1741 | iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec)); | |
1742 | for (i = 0; i < nr_segs; i++) { | |
1743 | struct compat_iovec v; | |
1744 | if (get_user(v.iov_base, &iov32[i].iov_base) || | |
1745 | get_user(v.iov_len, &iov32[i].iov_len) || | |
1746 | put_user(compat_ptr(v.iov_base), &iov[i].iov_base) || | |
1747 | put_user(v.iov_len, &iov[i].iov_len)) | |
1748 | return -EFAULT; | |
1749 | } | |
1750 | return sys_vmsplice(fd, iov, nr_segs, flags); | |
1751 | } | |
1752 | #endif | |
1753 | ||
836f92ad HC |
1754 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
1755 | int, fd_out, loff_t __user *, off_out, | |
1756 | size_t, len, unsigned int, flags) | |
5274f052 | 1757 | { |
2903ff01 | 1758 | struct fd in, out; |
5274f052 | 1759 | long error; |
5274f052 JA |
1760 | |
1761 | if (unlikely(!len)) | |
1762 | return 0; | |
1763 | ||
1764 | error = -EBADF; | |
2903ff01 AV |
1765 | in = fdget(fd_in); |
1766 | if (in.file) { | |
1767 | if (in.file->f_mode & FMODE_READ) { | |
1768 | out = fdget(fd_out); | |
1769 | if (out.file) { | |
1770 | if (out.file->f_mode & FMODE_WRITE) | |
1771 | error = do_splice(in.file, off_in, | |
1772 | out.file, off_out, | |
529565dc | 1773 | len, flags); |
2903ff01 | 1774 | fdput(out); |
5274f052 JA |
1775 | } |
1776 | } | |
2903ff01 | 1777 | fdput(in); |
5274f052 | 1778 | } |
5274f052 JA |
1779 | return error; |
1780 | } | |
70524490 | 1781 | |
aadd06e5 JA |
1782 | /* |
1783 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1784 | * return an appropriate error. | |
1785 | */ | |
7c77f0b3 | 1786 | static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1787 | { |
1788 | int ret; | |
1789 | ||
1790 | /* | |
1791 | * Check ->nrbufs without the inode lock first. This function | |
1792 | * is speculative anyways, so missing one is ok. | |
1793 | */ | |
1794 | if (pipe->nrbufs) | |
1795 | return 0; | |
1796 | ||
1797 | ret = 0; | |
61e0d47c | 1798 | pipe_lock(pipe); |
aadd06e5 JA |
1799 | |
1800 | while (!pipe->nrbufs) { | |
1801 | if (signal_pending(current)) { | |
1802 | ret = -ERESTARTSYS; | |
1803 | break; | |
1804 | } | |
1805 | if (!pipe->writers) | |
1806 | break; | |
1807 | if (!pipe->waiting_writers) { | |
1808 | if (flags & SPLICE_F_NONBLOCK) { | |
1809 | ret = -EAGAIN; | |
1810 | break; | |
1811 | } | |
1812 | } | |
1813 | pipe_wait(pipe); | |
1814 | } | |
1815 | ||
61e0d47c | 1816 | pipe_unlock(pipe); |
aadd06e5 JA |
1817 | return ret; |
1818 | } | |
1819 | ||
1820 | /* | |
1821 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1822 | * return an appropriate error. | |
1823 | */ | |
7c77f0b3 | 1824 | static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1825 | { |
1826 | int ret; | |
1827 | ||
1828 | /* | |
1829 | * Check ->nrbufs without the inode lock first. This function | |
1830 | * is speculative anyways, so missing one is ok. | |
1831 | */ | |
35f3d14d | 1832 | if (pipe->nrbufs < pipe->buffers) |
aadd06e5 JA |
1833 | return 0; |
1834 | ||
1835 | ret = 0; | |
61e0d47c | 1836 | pipe_lock(pipe); |
aadd06e5 | 1837 | |
35f3d14d | 1838 | while (pipe->nrbufs >= pipe->buffers) { |
aadd06e5 JA |
1839 | if (!pipe->readers) { |
1840 | send_sig(SIGPIPE, current, 0); | |
1841 | ret = -EPIPE; | |
1842 | break; | |
1843 | } | |
1844 | if (flags & SPLICE_F_NONBLOCK) { | |
1845 | ret = -EAGAIN; | |
1846 | break; | |
1847 | } | |
1848 | if (signal_pending(current)) { | |
1849 | ret = -ERESTARTSYS; | |
1850 | break; | |
1851 | } | |
1852 | pipe->waiting_writers++; | |
1853 | pipe_wait(pipe); | |
1854 | pipe->waiting_writers--; | |
1855 | } | |
1856 | ||
61e0d47c | 1857 | pipe_unlock(pipe); |
aadd06e5 JA |
1858 | return ret; |
1859 | } | |
1860 | ||
7c77f0b3 MS |
1861 | /* |
1862 | * Splice contents of ipipe to opipe. | |
1863 | */ | |
1864 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, | |
1865 | struct pipe_inode_info *opipe, | |
1866 | size_t len, unsigned int flags) | |
1867 | { | |
1868 | struct pipe_buffer *ibuf, *obuf; | |
1869 | int ret = 0, nbuf; | |
1870 | bool input_wakeup = false; | |
1871 | ||
1872 | ||
1873 | retry: | |
1874 | ret = ipipe_prep(ipipe, flags); | |
1875 | if (ret) | |
1876 | return ret; | |
1877 | ||
1878 | ret = opipe_prep(opipe, flags); | |
1879 | if (ret) | |
1880 | return ret; | |
1881 | ||
1882 | /* | |
1883 | * Potential ABBA deadlock, work around it by ordering lock | |
1884 | * grabbing by pipe info address. Otherwise two different processes | |
1885 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1886 | */ | |
1887 | pipe_double_lock(ipipe, opipe); | |
1888 | ||
1889 | do { | |
1890 | if (!opipe->readers) { | |
1891 | send_sig(SIGPIPE, current, 0); | |
1892 | if (!ret) | |
1893 | ret = -EPIPE; | |
1894 | break; | |
1895 | } | |
1896 | ||
1897 | if (!ipipe->nrbufs && !ipipe->writers) | |
1898 | break; | |
1899 | ||
1900 | /* | |
1901 | * Cannot make any progress, because either the input | |
1902 | * pipe is empty or the output pipe is full. | |
1903 | */ | |
35f3d14d | 1904 | if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { |
7c77f0b3 MS |
1905 | /* Already processed some buffers, break */ |
1906 | if (ret) | |
1907 | break; | |
1908 | ||
1909 | if (flags & SPLICE_F_NONBLOCK) { | |
1910 | ret = -EAGAIN; | |
1911 | break; | |
1912 | } | |
1913 | ||
1914 | /* | |
1915 | * We raced with another reader/writer and haven't | |
1916 | * managed to process any buffers. A zero return | |
1917 | * value means EOF, so retry instead. | |
1918 | */ | |
1919 | pipe_unlock(ipipe); | |
1920 | pipe_unlock(opipe); | |
1921 | goto retry; | |
1922 | } | |
1923 | ||
1924 | ibuf = ipipe->bufs + ipipe->curbuf; | |
35f3d14d | 1925 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); |
7c77f0b3 MS |
1926 | obuf = opipe->bufs + nbuf; |
1927 | ||
1928 | if (len >= ibuf->len) { | |
1929 | /* | |
1930 | * Simply move the whole buffer from ipipe to opipe | |
1931 | */ | |
1932 | *obuf = *ibuf; | |
1933 | ibuf->ops = NULL; | |
1934 | opipe->nrbufs++; | |
35f3d14d | 1935 | ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); |
7c77f0b3 MS |
1936 | ipipe->nrbufs--; |
1937 | input_wakeup = true; | |
1938 | } else { | |
1939 | /* | |
1940 | * Get a reference to this pipe buffer, | |
1941 | * so we can copy the contents over. | |
1942 | */ | |
1943 | ibuf->ops->get(ipipe, ibuf); | |
1944 | *obuf = *ibuf; | |
1945 | ||
1946 | /* | |
1947 | * Don't inherit the gift flag, we need to | |
1948 | * prevent multiple steals of this page. | |
1949 | */ | |
1950 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1951 | ||
1952 | obuf->len = len; | |
1953 | opipe->nrbufs++; | |
1954 | ibuf->offset += obuf->len; | |
1955 | ibuf->len -= obuf->len; | |
1956 | } | |
1957 | ret += obuf->len; | |
1958 | len -= obuf->len; | |
1959 | } while (len); | |
1960 | ||
1961 | pipe_unlock(ipipe); | |
1962 | pipe_unlock(opipe); | |
1963 | ||
1964 | /* | |
1965 | * If we put data in the output pipe, wakeup any potential readers. | |
1966 | */ | |
825cdcb1 NK |
1967 | if (ret > 0) |
1968 | wakeup_pipe_readers(opipe); | |
1969 | ||
7c77f0b3 MS |
1970 | if (input_wakeup) |
1971 | wakeup_pipe_writers(ipipe); | |
1972 | ||
1973 | return ret; | |
1974 | } | |
1975 | ||
70524490 JA |
1976 | /* |
1977 | * Link contents of ipipe to opipe. | |
1978 | */ | |
1979 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1980 | struct pipe_inode_info *opipe, | |
1981 | size_t len, unsigned int flags) | |
1982 | { | |
1983 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1984 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1985 | |
1986 | /* | |
1987 | * Potential ABBA deadlock, work around it by ordering lock | |
61e0d47c | 1988 | * grabbing by pipe info address. Otherwise two different processes |
70524490 JA |
1989 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
1990 | */ | |
61e0d47c | 1991 | pipe_double_lock(ipipe, opipe); |
70524490 | 1992 | |
aadd06e5 | 1993 | do { |
70524490 JA |
1994 | if (!opipe->readers) { |
1995 | send_sig(SIGPIPE, current, 0); | |
1996 | if (!ret) | |
1997 | ret = -EPIPE; | |
1998 | break; | |
1999 | } | |
70524490 | 2000 | |
aadd06e5 JA |
2001 | /* |
2002 | * If we have iterated all input buffers or ran out of | |
2003 | * output room, break. | |
2004 | */ | |
35f3d14d | 2005 | if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) |
aadd06e5 | 2006 | break; |
70524490 | 2007 | |
35f3d14d JA |
2008 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); |
2009 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); | |
70524490 JA |
2010 | |
2011 | /* | |
aadd06e5 JA |
2012 | * Get a reference to this pipe buffer, |
2013 | * so we can copy the contents over. | |
70524490 | 2014 | */ |
aadd06e5 JA |
2015 | ibuf->ops->get(ipipe, ibuf); |
2016 | ||
2017 | obuf = opipe->bufs + nbuf; | |
2018 | *obuf = *ibuf; | |
2019 | ||
2a27250e | 2020 | /* |
aadd06e5 JA |
2021 | * Don't inherit the gift flag, we need to |
2022 | * prevent multiple steals of this page. | |
2a27250e | 2023 | */ |
aadd06e5 | 2024 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 2025 | |
aadd06e5 JA |
2026 | if (obuf->len > len) |
2027 | obuf->len = len; | |
70524490 | 2028 | |
aadd06e5 JA |
2029 | opipe->nrbufs++; |
2030 | ret += obuf->len; | |
2031 | len -= obuf->len; | |
2032 | i++; | |
2033 | } while (len); | |
70524490 | 2034 | |
02cf01ae JA |
2035 | /* |
2036 | * return EAGAIN if we have the potential of some data in the | |
2037 | * future, otherwise just return 0 | |
2038 | */ | |
2039 | if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) | |
2040 | ret = -EAGAIN; | |
2041 | ||
61e0d47c MS |
2042 | pipe_unlock(ipipe); |
2043 | pipe_unlock(opipe); | |
70524490 | 2044 | |
aadd06e5 JA |
2045 | /* |
2046 | * If we put data in the output pipe, wakeup any potential readers. | |
2047 | */ | |
825cdcb1 NK |
2048 | if (ret > 0) |
2049 | wakeup_pipe_readers(opipe); | |
70524490 JA |
2050 | |
2051 | return ret; | |
2052 | } | |
2053 | ||
2054 | /* | |
2055 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
2056 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
2057 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
2058 | * applicable one is SPLICE_F_NONBLOCK. | |
2059 | */ | |
2060 | static long do_tee(struct file *in, struct file *out, size_t len, | |
2061 | unsigned int flags) | |
2062 | { | |
71993e62 LT |
2063 | struct pipe_inode_info *ipipe = get_pipe_info(in); |
2064 | struct pipe_inode_info *opipe = get_pipe_info(out); | |
aadd06e5 | 2065 | int ret = -EINVAL; |
70524490 JA |
2066 | |
2067 | /* | |
aadd06e5 JA |
2068 | * Duplicate the contents of ipipe to opipe without actually |
2069 | * copying the data. | |
70524490 | 2070 | */ |
aadd06e5 JA |
2071 | if (ipipe && opipe && ipipe != opipe) { |
2072 | /* | |
2073 | * Keep going, unless we encounter an error. The ipipe/opipe | |
2074 | * ordering doesn't really matter. | |
2075 | */ | |
7c77f0b3 | 2076 | ret = ipipe_prep(ipipe, flags); |
aadd06e5 | 2077 | if (!ret) { |
7c77f0b3 | 2078 | ret = opipe_prep(opipe, flags); |
02cf01ae | 2079 | if (!ret) |
aadd06e5 | 2080 | ret = link_pipe(ipipe, opipe, len, flags); |
aadd06e5 JA |
2081 | } |
2082 | } | |
70524490 | 2083 | |
aadd06e5 | 2084 | return ret; |
70524490 JA |
2085 | } |
2086 | ||
836f92ad | 2087 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
70524490 | 2088 | { |
2903ff01 AV |
2089 | struct fd in; |
2090 | int error; | |
70524490 JA |
2091 | |
2092 | if (unlikely(!len)) | |
2093 | return 0; | |
2094 | ||
2095 | error = -EBADF; | |
2903ff01 AV |
2096 | in = fdget(fdin); |
2097 | if (in.file) { | |
2098 | if (in.file->f_mode & FMODE_READ) { | |
2099 | struct fd out = fdget(fdout); | |
2100 | if (out.file) { | |
2101 | if (out.file->f_mode & FMODE_WRITE) | |
2102 | error = do_tee(in.file, out.file, | |
2103 | len, flags); | |
2104 | fdput(out); | |
70524490 JA |
2105 | } |
2106 | } | |
2903ff01 | 2107 | fdput(in); |
70524490 JA |
2108 | } |
2109 | ||
2110 | return error; | |
2111 | } |