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> |
8d020765 | 35 | #include <linux/aio.h> |
06ae43f3 | 36 | #include "internal.h" |
5274f052 | 37 | |
83f9135b JA |
38 | /* |
39 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
40 | * a vm helper function, it's already simplified quite a bit by the | |
41 | * addition of remove_mapping(). If success is returned, the caller may | |
42 | * attempt to reuse this page for another destination. | |
43 | */ | |
76ad4d11 | 44 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, |
5abc97aa JA |
45 | struct pipe_buffer *buf) |
46 | { | |
47 | struct page *page = buf->page; | |
9e94cd4f | 48 | struct address_space *mapping; |
5abc97aa | 49 | |
9e0267c2 JA |
50 | lock_page(page); |
51 | ||
9e94cd4f JA |
52 | mapping = page_mapping(page); |
53 | if (mapping) { | |
54 | WARN_ON(!PageUptodate(page)); | |
5abc97aa | 55 | |
9e94cd4f JA |
56 | /* |
57 | * At least for ext2 with nobh option, we need to wait on | |
58 | * writeback completing on this page, since we'll remove it | |
59 | * from the pagecache. Otherwise truncate wont wait on the | |
60 | * page, allowing the disk blocks to be reused by someone else | |
61 | * before we actually wrote our data to them. fs corruption | |
62 | * ensues. | |
63 | */ | |
64 | wait_on_page_writeback(page); | |
ad8d6f0a | 65 | |
266cf658 DH |
66 | if (page_has_private(page) && |
67 | !try_to_release_page(page, GFP_KERNEL)) | |
ca39d651 | 68 | goto out_unlock; |
4f6f0bd2 | 69 | |
9e94cd4f JA |
70 | /* |
71 | * If we succeeded in removing the mapping, set LRU flag | |
72 | * and return good. | |
73 | */ | |
74 | if (remove_mapping(mapping, page)) { | |
75 | buf->flags |= PIPE_BUF_FLAG_LRU; | |
76 | return 0; | |
77 | } | |
9e0267c2 | 78 | } |
5abc97aa | 79 | |
9e94cd4f JA |
80 | /* |
81 | * Raced with truncate or failed to remove page from current | |
82 | * address space, unlock and return failure. | |
83 | */ | |
ca39d651 | 84 | out_unlock: |
9e94cd4f JA |
85 | unlock_page(page); |
86 | return 1; | |
5abc97aa JA |
87 | } |
88 | ||
76ad4d11 | 89 | static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, |
5274f052 JA |
90 | struct pipe_buffer *buf) |
91 | { | |
92 | page_cache_release(buf->page); | |
1432873a | 93 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
5274f052 JA |
94 | } |
95 | ||
0845718d JA |
96 | /* |
97 | * Check whether the contents of buf is OK to access. Since the content | |
98 | * is a page cache page, IO may be in flight. | |
99 | */ | |
cac36bb0 JA |
100 | static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, |
101 | struct pipe_buffer *buf) | |
5274f052 JA |
102 | { |
103 | struct page *page = buf->page; | |
49d0b21b | 104 | int err; |
5274f052 JA |
105 | |
106 | if (!PageUptodate(page)) { | |
49d0b21b JA |
107 | lock_page(page); |
108 | ||
109 | /* | |
110 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 111 | * splice, if this is the first page. |
49d0b21b JA |
112 | */ |
113 | if (!page->mapping) { | |
114 | err = -ENODATA; | |
115 | goto error; | |
116 | } | |
5274f052 | 117 | |
49d0b21b | 118 | /* |
73d62d83 | 119 | * Uh oh, read-error from disk. |
49d0b21b JA |
120 | */ |
121 | if (!PageUptodate(page)) { | |
122 | err = -EIO; | |
123 | goto error; | |
124 | } | |
125 | ||
126 | /* | |
f84d7519 | 127 | * Page is ok afterall, we are done. |
49d0b21b | 128 | */ |
5274f052 | 129 | unlock_page(page); |
5274f052 JA |
130 | } |
131 | ||
f84d7519 | 132 | return 0; |
49d0b21b JA |
133 | error: |
134 | unlock_page(page); | |
f84d7519 | 135 | return err; |
70524490 JA |
136 | } |
137 | ||
708e3508 | 138 | const struct pipe_buf_operations page_cache_pipe_buf_ops = { |
5274f052 | 139 | .can_merge = 0, |
cac36bb0 | 140 | .confirm = page_cache_pipe_buf_confirm, |
5274f052 | 141 | .release = page_cache_pipe_buf_release, |
5abc97aa | 142 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 143 | .get = generic_pipe_buf_get, |
5274f052 JA |
144 | }; |
145 | ||
912d35f8 JA |
146 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
147 | struct pipe_buffer *buf) | |
148 | { | |
7afa6fd0 JA |
149 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
150 | return 1; | |
151 | ||
1432873a | 152 | buf->flags |= PIPE_BUF_FLAG_LRU; |
330ab716 | 153 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
154 | } |
155 | ||
d4c3cca9 | 156 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
912d35f8 | 157 | .can_merge = 0, |
cac36bb0 | 158 | .confirm = generic_pipe_buf_confirm, |
912d35f8 JA |
159 | .release = page_cache_pipe_buf_release, |
160 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 161 | .get = generic_pipe_buf_get, |
912d35f8 JA |
162 | }; |
163 | ||
825cdcb1 NK |
164 | static void wakeup_pipe_readers(struct pipe_inode_info *pipe) |
165 | { | |
166 | smp_mb(); | |
167 | if (waitqueue_active(&pipe->wait)) | |
168 | wake_up_interruptible(&pipe->wait); | |
169 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
170 | } | |
171 | ||
932cc6d4 JA |
172 | /** |
173 | * splice_to_pipe - fill passed data into a pipe | |
174 | * @pipe: pipe to fill | |
175 | * @spd: data to fill | |
176 | * | |
177 | * Description: | |
79685b8d | 178 | * @spd contains a map of pages and len/offset tuples, along with |
932cc6d4 JA |
179 | * the struct pipe_buf_operations associated with these pages. This |
180 | * function will link that data to the pipe. | |
181 | * | |
83f9135b | 182 | */ |
d6b29d7c JA |
183 | ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
184 | struct splice_pipe_desc *spd) | |
5274f052 | 185 | { |
00de00bd | 186 | unsigned int spd_pages = spd->nr_pages; |
912d35f8 | 187 | int ret, do_wakeup, page_nr; |
5274f052 JA |
188 | |
189 | ret = 0; | |
190 | do_wakeup = 0; | |
912d35f8 | 191 | page_nr = 0; |
5274f052 | 192 | |
61e0d47c | 193 | pipe_lock(pipe); |
5274f052 | 194 | |
5274f052 | 195 | for (;;) { |
3a326a2c | 196 | if (!pipe->readers) { |
5274f052 JA |
197 | send_sig(SIGPIPE, current, 0); |
198 | if (!ret) | |
199 | ret = -EPIPE; | |
200 | break; | |
201 | } | |
202 | ||
35f3d14d JA |
203 | if (pipe->nrbufs < pipe->buffers) { |
204 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
3a326a2c | 205 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 206 | |
912d35f8 JA |
207 | buf->page = spd->pages[page_nr]; |
208 | buf->offset = spd->partial[page_nr].offset; | |
209 | buf->len = spd->partial[page_nr].len; | |
497f9625 | 210 | buf->private = spd->partial[page_nr].private; |
912d35f8 | 211 | buf->ops = spd->ops; |
7afa6fd0 JA |
212 | if (spd->flags & SPLICE_F_GIFT) |
213 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
214 | ||
6f767b04 | 215 | pipe->nrbufs++; |
912d35f8 JA |
216 | page_nr++; |
217 | ret += buf->len; | |
218 | ||
6447a3cf | 219 | if (pipe->files) |
6f767b04 | 220 | do_wakeup = 1; |
5274f052 | 221 | |
912d35f8 | 222 | if (!--spd->nr_pages) |
5274f052 | 223 | break; |
35f3d14d | 224 | if (pipe->nrbufs < pipe->buffers) |
5274f052 JA |
225 | continue; |
226 | ||
227 | break; | |
228 | } | |
229 | ||
912d35f8 | 230 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
231 | if (!ret) |
232 | ret = -EAGAIN; | |
233 | break; | |
234 | } | |
235 | ||
5274f052 JA |
236 | if (signal_pending(current)) { |
237 | if (!ret) | |
238 | ret = -ERESTARTSYS; | |
239 | break; | |
240 | } | |
241 | ||
242 | if (do_wakeup) { | |
c0bd1f65 | 243 | smp_mb(); |
3a326a2c IM |
244 | if (waitqueue_active(&pipe->wait)) |
245 | wake_up_interruptible_sync(&pipe->wait); | |
246 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
247 | do_wakeup = 0; |
248 | } | |
249 | ||
3a326a2c IM |
250 | pipe->waiting_writers++; |
251 | pipe_wait(pipe); | |
252 | pipe->waiting_writers--; | |
5274f052 JA |
253 | } |
254 | ||
61e0d47c | 255 | pipe_unlock(pipe); |
5274f052 | 256 | |
825cdcb1 NK |
257 | if (do_wakeup) |
258 | wakeup_pipe_readers(pipe); | |
5274f052 | 259 | |
00de00bd | 260 | while (page_nr < spd_pages) |
bbdfc2f7 | 261 | spd->spd_release(spd, page_nr++); |
5274f052 JA |
262 | |
263 | return ret; | |
264 | } | |
265 | ||
708e3508 | 266 | void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) |
bbdfc2f7 JA |
267 | { |
268 | page_cache_release(spd->pages[i]); | |
269 | } | |
270 | ||
35f3d14d JA |
271 | /* |
272 | * Check if we need to grow the arrays holding pages and partial page | |
273 | * descriptions. | |
274 | */ | |
047fe360 | 275 | int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) |
35f3d14d | 276 | { |
047fe360 ED |
277 | unsigned int buffers = ACCESS_ONCE(pipe->buffers); |
278 | ||
279 | spd->nr_pages_max = buffers; | |
280 | if (buffers <= PIPE_DEF_BUFFERS) | |
35f3d14d JA |
281 | return 0; |
282 | ||
047fe360 ED |
283 | spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL); |
284 | spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL); | |
35f3d14d JA |
285 | |
286 | if (spd->pages && spd->partial) | |
287 | return 0; | |
288 | ||
289 | kfree(spd->pages); | |
290 | kfree(spd->partial); | |
291 | return -ENOMEM; | |
292 | } | |
293 | ||
047fe360 | 294 | void splice_shrink_spd(struct splice_pipe_desc *spd) |
35f3d14d | 295 | { |
047fe360 | 296 | if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) |
35f3d14d JA |
297 | return; |
298 | ||
299 | kfree(spd->pages); | |
300 | kfree(spd->partial); | |
301 | } | |
302 | ||
3a326a2c | 303 | static int |
cbb7e577 JA |
304 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
305 | struct pipe_inode_info *pipe, size_t len, | |
306 | unsigned int flags) | |
5274f052 JA |
307 | { |
308 | struct address_space *mapping = in->f_mapping; | |
d8983910 | 309 | unsigned int loff, nr_pages, req_pages; |
35f3d14d JA |
310 | struct page *pages[PIPE_DEF_BUFFERS]; |
311 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
5274f052 | 312 | struct page *page; |
91ad66ef JA |
313 | pgoff_t index, end_index; |
314 | loff_t isize; | |
eb20796b | 315 | int error, page_nr; |
912d35f8 JA |
316 | struct splice_pipe_desc spd = { |
317 | .pages = pages, | |
318 | .partial = partial, | |
047fe360 | 319 | .nr_pages_max = PIPE_DEF_BUFFERS, |
912d35f8 JA |
320 | .flags = flags, |
321 | .ops = &page_cache_pipe_buf_ops, | |
bbdfc2f7 | 322 | .spd_release = spd_release_page, |
912d35f8 | 323 | }; |
5274f052 | 324 | |
35f3d14d JA |
325 | if (splice_grow_spd(pipe, &spd)) |
326 | return -ENOMEM; | |
327 | ||
cbb7e577 | 328 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 | 329 | loff = *ppos & ~PAGE_CACHE_MASK; |
d8983910 | 330 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
047fe360 | 331 | nr_pages = min(req_pages, spd.nr_pages_max); |
5274f052 | 332 | |
eb20796b JA |
333 | /* |
334 | * Lookup the (hopefully) full range of pages we need. | |
335 | */ | |
35f3d14d | 336 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages); |
431a4820 | 337 | index += spd.nr_pages; |
82aa5d61 | 338 | |
eb20796b JA |
339 | /* |
340 | * If find_get_pages_contig() returned fewer pages than we needed, | |
431a4820 | 341 | * readahead/allocate the rest and fill in the holes. |
eb20796b | 342 | */ |
431a4820 | 343 | if (spd.nr_pages < nr_pages) |
cf914a7d RR |
344 | page_cache_sync_readahead(mapping, &in->f_ra, in, |
345 | index, req_pages - spd.nr_pages); | |
431a4820 | 346 | |
932cc6d4 | 347 | error = 0; |
eb20796b | 348 | while (spd.nr_pages < nr_pages) { |
82aa5d61 | 349 | /* |
eb20796b JA |
350 | * Page could be there, find_get_pages_contig() breaks on |
351 | * the first hole. | |
5274f052 | 352 | */ |
7480a904 JA |
353 | page = find_get_page(mapping, index); |
354 | if (!page) { | |
7480a904 | 355 | /* |
eb20796b | 356 | * page didn't exist, allocate one. |
7480a904 JA |
357 | */ |
358 | page = page_cache_alloc_cold(mapping); | |
359 | if (!page) | |
360 | break; | |
361 | ||
362 | error = add_to_page_cache_lru(page, mapping, index, | |
0ae0b5d0 | 363 | GFP_KERNEL); |
7480a904 JA |
364 | if (unlikely(error)) { |
365 | page_cache_release(page); | |
a0548871 JA |
366 | if (error == -EEXIST) |
367 | continue; | |
7480a904 JA |
368 | break; |
369 | } | |
eb20796b JA |
370 | /* |
371 | * add_to_page_cache() locks the page, unlock it | |
372 | * to avoid convoluting the logic below even more. | |
373 | */ | |
374 | unlock_page(page); | |
7480a904 JA |
375 | } |
376 | ||
35f3d14d | 377 | spd.pages[spd.nr_pages++] = page; |
eb20796b JA |
378 | index++; |
379 | } | |
380 | ||
381 | /* | |
382 | * Now loop over the map and see if we need to start IO on any | |
383 | * pages, fill in the partial map, etc. | |
384 | */ | |
385 | index = *ppos >> PAGE_CACHE_SHIFT; | |
386 | nr_pages = spd.nr_pages; | |
387 | spd.nr_pages = 0; | |
388 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
389 | unsigned int this_len; | |
390 | ||
391 | if (!len) | |
392 | break; | |
393 | ||
394 | /* | |
395 | * this_len is the max we'll use from this page | |
396 | */ | |
397 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
35f3d14d | 398 | page = spd.pages[page_nr]; |
eb20796b | 399 | |
a08a166f | 400 | if (PageReadahead(page)) |
cf914a7d | 401 | page_cache_async_readahead(mapping, &in->f_ra, in, |
d8983910 | 402 | page, index, req_pages - page_nr); |
a08a166f | 403 | |
7480a904 JA |
404 | /* |
405 | * If the page isn't uptodate, we may need to start io on it | |
406 | */ | |
407 | if (!PageUptodate(page)) { | |
6965031d | 408 | lock_page(page); |
7480a904 JA |
409 | |
410 | /* | |
32502b84 MS |
411 | * Page was truncated, or invalidated by the |
412 | * filesystem. Redo the find/create, but this time the | |
413 | * page is kept locked, so there's no chance of another | |
414 | * race with truncate/invalidate. | |
7480a904 JA |
415 | */ |
416 | if (!page->mapping) { | |
417 | unlock_page(page); | |
32502b84 MS |
418 | page = find_or_create_page(mapping, index, |
419 | mapping_gfp_mask(mapping)); | |
420 | ||
421 | if (!page) { | |
422 | error = -ENOMEM; | |
423 | break; | |
424 | } | |
35f3d14d JA |
425 | page_cache_release(spd.pages[page_nr]); |
426 | spd.pages[page_nr] = page; | |
7480a904 JA |
427 | } |
428 | /* | |
429 | * page was already under io and is now done, great | |
430 | */ | |
431 | if (PageUptodate(page)) { | |
432 | unlock_page(page); | |
433 | goto fill_it; | |
434 | } | |
5274f052 | 435 | |
7480a904 JA |
436 | /* |
437 | * need to read in the page | |
438 | */ | |
439 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 440 | if (unlikely(error)) { |
eb20796b JA |
441 | /* |
442 | * We really should re-lookup the page here, | |
443 | * but it complicates things a lot. Instead | |
444 | * lets just do what we already stored, and | |
445 | * we'll get it the next time we are called. | |
446 | */ | |
7480a904 | 447 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
448 | error = 0; |
449 | ||
5274f052 JA |
450 | break; |
451 | } | |
620a324b JA |
452 | } |
453 | fill_it: | |
454 | /* | |
455 | * i_size must be checked after PageUptodate. | |
456 | */ | |
457 | isize = i_size_read(mapping->host); | |
458 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
459 | if (unlikely(!isize || index > end_index)) | |
460 | break; | |
461 | ||
462 | /* | |
463 | * if this is the last page, see if we need to shrink | |
464 | * the length and stop | |
465 | */ | |
466 | if (end_index == index) { | |
467 | unsigned int plen; | |
91ad66ef JA |
468 | |
469 | /* | |
620a324b | 470 | * max good bytes in this page |
91ad66ef | 471 | */ |
620a324b JA |
472 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
473 | if (plen <= loff) | |
91ad66ef | 474 | break; |
91ad66ef JA |
475 | |
476 | /* | |
620a324b | 477 | * force quit after adding this page |
91ad66ef | 478 | */ |
620a324b JA |
479 | this_len = min(this_len, plen - loff); |
480 | len = this_len; | |
5274f052 | 481 | } |
620a324b | 482 | |
35f3d14d JA |
483 | spd.partial[page_nr].offset = loff; |
484 | spd.partial[page_nr].len = this_len; | |
82aa5d61 | 485 | len -= this_len; |
91ad66ef | 486 | loff = 0; |
eb20796b JA |
487 | spd.nr_pages++; |
488 | index++; | |
5274f052 JA |
489 | } |
490 | ||
eb20796b | 491 | /* |
475ecade | 492 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
493 | * we got, 'nr_pages' is how many pages are in the map. |
494 | */ | |
495 | while (page_nr < nr_pages) | |
35f3d14d | 496 | page_cache_release(spd.pages[page_nr++]); |
f4e6b498 | 497 | in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; |
eb20796b | 498 | |
912d35f8 | 499 | if (spd.nr_pages) |
35f3d14d | 500 | error = splice_to_pipe(pipe, &spd); |
5274f052 | 501 | |
047fe360 | 502 | splice_shrink_spd(&spd); |
7480a904 | 503 | return error; |
5274f052 JA |
504 | } |
505 | ||
83f9135b JA |
506 | /** |
507 | * generic_file_splice_read - splice data from file to a pipe | |
508 | * @in: file to splice from | |
932cc6d4 | 509 | * @ppos: position in @in |
83f9135b JA |
510 | * @pipe: pipe to splice to |
511 | * @len: number of bytes to splice | |
512 | * @flags: splice modifier flags | |
513 | * | |
932cc6d4 JA |
514 | * Description: |
515 | * Will read pages from given file and fill them into a pipe. Can be | |
516 | * used as long as the address_space operations for the source implements | |
517 | * a readpage() hook. | |
518 | * | |
83f9135b | 519 | */ |
cbb7e577 JA |
520 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
521 | struct pipe_inode_info *pipe, size_t len, | |
522 | unsigned int flags) | |
5274f052 | 523 | { |
d366d398 | 524 | loff_t isize, left; |
8191ecd1 | 525 | int ret; |
d366d398 JA |
526 | |
527 | isize = i_size_read(in->f_mapping->host); | |
528 | if (unlikely(*ppos >= isize)) | |
529 | return 0; | |
530 | ||
531 | left = isize - *ppos; | |
532 | if (unlikely(left < len)) | |
533 | len = left; | |
5274f052 | 534 | |
8191ecd1 | 535 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
723590ed | 536 | if (ret > 0) { |
cbb7e577 | 537 | *ppos += ret; |
723590ed MS |
538 | file_accessed(in); |
539 | } | |
5274f052 JA |
540 | |
541 | return ret; | |
542 | } | |
059a8f37 JA |
543 | EXPORT_SYMBOL(generic_file_splice_read); |
544 | ||
6818173b MS |
545 | static const struct pipe_buf_operations default_pipe_buf_ops = { |
546 | .can_merge = 0, | |
6818173b MS |
547 | .confirm = generic_pipe_buf_confirm, |
548 | .release = generic_pipe_buf_release, | |
549 | .steal = generic_pipe_buf_steal, | |
550 | .get = generic_pipe_buf_get, | |
551 | }; | |
552 | ||
28a625cb MS |
553 | static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, |
554 | struct pipe_buffer *buf) | |
555 | { | |
556 | return 1; | |
557 | } | |
558 | ||
559 | /* Pipe buffer operations for a socket and similar. */ | |
560 | const struct pipe_buf_operations nosteal_pipe_buf_ops = { | |
561 | .can_merge = 0, | |
28a625cb MS |
562 | .confirm = generic_pipe_buf_confirm, |
563 | .release = generic_pipe_buf_release, | |
564 | .steal = generic_pipe_buf_nosteal, | |
565 | .get = generic_pipe_buf_get, | |
566 | }; | |
567 | EXPORT_SYMBOL(nosteal_pipe_buf_ops); | |
568 | ||
6818173b MS |
569 | static ssize_t kernel_readv(struct file *file, const struct iovec *vec, |
570 | unsigned long vlen, loff_t offset) | |
571 | { | |
572 | mm_segment_t old_fs; | |
573 | loff_t pos = offset; | |
574 | ssize_t res; | |
575 | ||
576 | old_fs = get_fs(); | |
577 | set_fs(get_ds()); | |
578 | /* The cast to a user pointer is valid due to the set_fs() */ | |
579 | res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos); | |
580 | set_fs(old_fs); | |
581 | ||
582 | return res; | |
583 | } | |
584 | ||
7bb307e8 | 585 | ssize_t kernel_write(struct file *file, const char *buf, size_t count, |
b2858d7d | 586 | loff_t pos) |
0b0a47f5 MS |
587 | { |
588 | mm_segment_t old_fs; | |
589 | ssize_t res; | |
590 | ||
591 | old_fs = get_fs(); | |
592 | set_fs(get_ds()); | |
593 | /* The cast to a user pointer is valid due to the set_fs() */ | |
7bb307e8 | 594 | res = vfs_write(file, (__force const char __user *)buf, count, &pos); |
0b0a47f5 MS |
595 | set_fs(old_fs); |
596 | ||
597 | return res; | |
598 | } | |
7bb307e8 | 599 | EXPORT_SYMBOL(kernel_write); |
0b0a47f5 | 600 | |
6818173b MS |
601 | ssize_t default_file_splice_read(struct file *in, loff_t *ppos, |
602 | struct pipe_inode_info *pipe, size_t len, | |
603 | unsigned int flags) | |
604 | { | |
605 | unsigned int nr_pages; | |
606 | unsigned int nr_freed; | |
607 | size_t offset; | |
35f3d14d JA |
608 | struct page *pages[PIPE_DEF_BUFFERS]; |
609 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
610 | struct iovec *vec, __vec[PIPE_DEF_BUFFERS]; | |
6818173b MS |
611 | ssize_t res; |
612 | size_t this_len; | |
613 | int error; | |
614 | int i; | |
615 | struct splice_pipe_desc spd = { | |
616 | .pages = pages, | |
617 | .partial = partial, | |
047fe360 | 618 | .nr_pages_max = PIPE_DEF_BUFFERS, |
6818173b MS |
619 | .flags = flags, |
620 | .ops = &default_pipe_buf_ops, | |
621 | .spd_release = spd_release_page, | |
622 | }; | |
623 | ||
35f3d14d JA |
624 | if (splice_grow_spd(pipe, &spd)) |
625 | return -ENOMEM; | |
626 | ||
627 | res = -ENOMEM; | |
628 | vec = __vec; | |
047fe360 ED |
629 | if (spd.nr_pages_max > PIPE_DEF_BUFFERS) { |
630 | vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL); | |
35f3d14d JA |
631 | if (!vec) |
632 | goto shrink_ret; | |
633 | } | |
634 | ||
6818173b MS |
635 | offset = *ppos & ~PAGE_CACHE_MASK; |
636 | nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
637 | ||
047fe360 | 638 | for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) { |
6818173b MS |
639 | struct page *page; |
640 | ||
4f231228 | 641 | page = alloc_page(GFP_USER); |
6818173b MS |
642 | error = -ENOMEM; |
643 | if (!page) | |
644 | goto err; | |
645 | ||
646 | this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset); | |
4f231228 | 647 | vec[i].iov_base = (void __user *) page_address(page); |
6818173b | 648 | vec[i].iov_len = this_len; |
35f3d14d | 649 | spd.pages[i] = page; |
6818173b MS |
650 | spd.nr_pages++; |
651 | len -= this_len; | |
652 | offset = 0; | |
653 | } | |
654 | ||
655 | res = kernel_readv(in, vec, spd.nr_pages, *ppos); | |
77f6bf57 AM |
656 | if (res < 0) { |
657 | error = res; | |
6818173b | 658 | goto err; |
77f6bf57 | 659 | } |
6818173b MS |
660 | |
661 | error = 0; | |
662 | if (!res) | |
663 | goto err; | |
664 | ||
665 | nr_freed = 0; | |
666 | for (i = 0; i < spd.nr_pages; i++) { | |
6818173b | 667 | this_len = min_t(size_t, vec[i].iov_len, res); |
35f3d14d JA |
668 | spd.partial[i].offset = 0; |
669 | spd.partial[i].len = this_len; | |
6818173b | 670 | if (!this_len) { |
35f3d14d JA |
671 | __free_page(spd.pages[i]); |
672 | spd.pages[i] = NULL; | |
6818173b MS |
673 | nr_freed++; |
674 | } | |
675 | res -= this_len; | |
676 | } | |
677 | spd.nr_pages -= nr_freed; | |
678 | ||
679 | res = splice_to_pipe(pipe, &spd); | |
680 | if (res > 0) | |
681 | *ppos += res; | |
682 | ||
35f3d14d JA |
683 | shrink_ret: |
684 | if (vec != __vec) | |
685 | kfree(vec); | |
047fe360 | 686 | splice_shrink_spd(&spd); |
6818173b MS |
687 | return res; |
688 | ||
689 | err: | |
4f231228 | 690 | for (i = 0; i < spd.nr_pages; i++) |
35f3d14d | 691 | __free_page(spd.pages[i]); |
4f231228 | 692 | |
35f3d14d JA |
693 | res = error; |
694 | goto shrink_ret; | |
6818173b MS |
695 | } |
696 | EXPORT_SYMBOL(default_file_splice_read); | |
697 | ||
5274f052 | 698 | /* |
4f6f0bd2 | 699 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 700 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 701 | */ |
76ad4d11 | 702 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
703 | struct pipe_buffer *buf, struct splice_desc *sd) |
704 | { | |
6a14b90b | 705 | struct file *file = sd->u.file; |
5274f052 | 706 | loff_t pos = sd->pos; |
a8adbe37 | 707 | int more; |
5274f052 | 708 | |
72c2d531 | 709 | if (!likely(file->f_op->sendpage)) |
a8adbe37 MM |
710 | return -EINVAL; |
711 | ||
35f9c09f | 712 | more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; |
ae62ca7b ED |
713 | |
714 | if (sd->len < sd->total_len && pipe->nrbufs > 1) | |
35f9c09f | 715 | more |= MSG_SENDPAGE_NOTLAST; |
ae62ca7b | 716 | |
a8adbe37 MM |
717 | return file->f_op->sendpage(file, buf->page, buf->offset, |
718 | sd->len, &pos, more); | |
5274f052 JA |
719 | } |
720 | ||
b3c2d2dd MS |
721 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
722 | { | |
723 | smp_mb(); | |
724 | if (waitqueue_active(&pipe->wait)) | |
725 | wake_up_interruptible(&pipe->wait); | |
726 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
727 | } | |
728 | ||
932cc6d4 | 729 | /** |
b3c2d2dd | 730 | * splice_from_pipe_feed - feed available data from a pipe to a file |
932cc6d4 JA |
731 | * @pipe: pipe to splice from |
732 | * @sd: information to @actor | |
733 | * @actor: handler that splices the data | |
734 | * | |
735 | * Description: | |
b3c2d2dd MS |
736 | * This function loops over the pipe and calls @actor to do the |
737 | * actual moving of a single struct pipe_buffer to the desired | |
738 | * destination. It returns when there's no more buffers left in | |
739 | * the pipe or if the requested number of bytes (@sd->total_len) | |
740 | * have been copied. It returns a positive number (one) if the | |
741 | * pipe needs to be filled with more data, zero if the required | |
742 | * number of bytes have been copied and -errno on error. | |
932cc6d4 | 743 | * |
b3c2d2dd MS |
744 | * This, together with splice_from_pipe_{begin,end,next}, may be |
745 | * used to implement the functionality of __splice_from_pipe() when | |
746 | * locking is required around copying the pipe buffers to the | |
747 | * destination. | |
83f9135b | 748 | */ |
96f9bc8f | 749 | static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
b3c2d2dd | 750 | splice_actor *actor) |
5274f052 | 751 | { |
b3c2d2dd | 752 | int ret; |
5274f052 | 753 | |
b3c2d2dd MS |
754 | while (pipe->nrbufs) { |
755 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
756 | const struct pipe_buf_operations *ops = buf->ops; | |
5274f052 | 757 | |
b3c2d2dd MS |
758 | sd->len = buf->len; |
759 | if (sd->len > sd->total_len) | |
760 | sd->len = sd->total_len; | |
5274f052 | 761 | |
a8adbe37 MM |
762 | ret = buf->ops->confirm(pipe, buf); |
763 | if (unlikely(ret)) { | |
b3c2d2dd MS |
764 | if (ret == -ENODATA) |
765 | ret = 0; | |
766 | return ret; | |
767 | } | |
a8adbe37 MM |
768 | |
769 | ret = actor(pipe, buf, sd); | |
770 | if (ret <= 0) | |
771 | return ret; | |
772 | ||
b3c2d2dd MS |
773 | buf->offset += ret; |
774 | buf->len -= ret; | |
775 | ||
776 | sd->num_spliced += ret; | |
777 | sd->len -= ret; | |
778 | sd->pos += ret; | |
779 | sd->total_len -= ret; | |
780 | ||
781 | if (!buf->len) { | |
782 | buf->ops = NULL; | |
783 | ops->release(pipe, buf); | |
35f3d14d | 784 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); |
b3c2d2dd | 785 | pipe->nrbufs--; |
6447a3cf | 786 | if (pipe->files) |
b3c2d2dd MS |
787 | sd->need_wakeup = true; |
788 | } | |
5274f052 | 789 | |
b3c2d2dd MS |
790 | if (!sd->total_len) |
791 | return 0; | |
792 | } | |
5274f052 | 793 | |
b3c2d2dd MS |
794 | return 1; |
795 | } | |
5274f052 | 796 | |
b3c2d2dd MS |
797 | /** |
798 | * splice_from_pipe_next - wait for some data to splice from | |
799 | * @pipe: pipe to splice from | |
800 | * @sd: information about the splice operation | |
801 | * | |
802 | * Description: | |
803 | * This function will wait for some data and return a positive | |
804 | * value (one) if pipe buffers are available. It will return zero | |
805 | * or -errno if no more data needs to be spliced. | |
806 | */ | |
96f9bc8f | 807 | static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd MS |
808 | { |
809 | while (!pipe->nrbufs) { | |
810 | if (!pipe->writers) | |
811 | return 0; | |
016b661e | 812 | |
b3c2d2dd MS |
813 | if (!pipe->waiting_writers && sd->num_spliced) |
814 | return 0; | |
73d62d83 | 815 | |
b3c2d2dd MS |
816 | if (sd->flags & SPLICE_F_NONBLOCK) |
817 | return -EAGAIN; | |
5274f052 | 818 | |
b3c2d2dd MS |
819 | if (signal_pending(current)) |
820 | return -ERESTARTSYS; | |
5274f052 | 821 | |
b3c2d2dd MS |
822 | if (sd->need_wakeup) { |
823 | wakeup_pipe_writers(pipe); | |
824 | sd->need_wakeup = false; | |
5274f052 JA |
825 | } |
826 | ||
b3c2d2dd MS |
827 | pipe_wait(pipe); |
828 | } | |
29e35094 | 829 | |
b3c2d2dd MS |
830 | return 1; |
831 | } | |
5274f052 | 832 | |
b3c2d2dd MS |
833 | /** |
834 | * splice_from_pipe_begin - start splicing from pipe | |
b80901bb | 835 | * @sd: information about the splice operation |
b3c2d2dd MS |
836 | * |
837 | * Description: | |
838 | * This function should be called before a loop containing | |
839 | * splice_from_pipe_next() and splice_from_pipe_feed() to | |
840 | * initialize the necessary fields of @sd. | |
841 | */ | |
96f9bc8f | 842 | static void splice_from_pipe_begin(struct splice_desc *sd) |
b3c2d2dd MS |
843 | { |
844 | sd->num_spliced = 0; | |
845 | sd->need_wakeup = false; | |
846 | } | |
5274f052 | 847 | |
b3c2d2dd MS |
848 | /** |
849 | * splice_from_pipe_end - finish splicing from pipe | |
850 | * @pipe: pipe to splice from | |
851 | * @sd: information about the splice operation | |
852 | * | |
853 | * Description: | |
854 | * This function will wake up pipe writers if necessary. It should | |
855 | * be called after a loop containing splice_from_pipe_next() and | |
856 | * splice_from_pipe_feed(). | |
857 | */ | |
96f9bc8f | 858 | static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd MS |
859 | { |
860 | if (sd->need_wakeup) | |
861 | wakeup_pipe_writers(pipe); | |
862 | } | |
5274f052 | 863 | |
b3c2d2dd MS |
864 | /** |
865 | * __splice_from_pipe - splice data from a pipe to given actor | |
866 | * @pipe: pipe to splice from | |
867 | * @sd: information to @actor | |
868 | * @actor: handler that splices the data | |
869 | * | |
870 | * Description: | |
871 | * This function does little more than loop over the pipe and call | |
872 | * @actor to do the actual moving of a single struct pipe_buffer to | |
873 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
874 | * pipe_to_user. | |
875 | * | |
876 | */ | |
877 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, | |
878 | splice_actor *actor) | |
879 | { | |
880 | int ret; | |
5274f052 | 881 | |
b3c2d2dd MS |
882 | splice_from_pipe_begin(sd); |
883 | do { | |
884 | ret = splice_from_pipe_next(pipe, sd); | |
885 | if (ret > 0) | |
886 | ret = splice_from_pipe_feed(pipe, sd, actor); | |
887 | } while (ret > 0); | |
888 | splice_from_pipe_end(pipe, sd); | |
889 | ||
890 | return sd->num_spliced ? sd->num_spliced : ret; | |
5274f052 | 891 | } |
40bee44e | 892 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 893 | |
932cc6d4 JA |
894 | /** |
895 | * splice_from_pipe - splice data from a pipe to a file | |
896 | * @pipe: pipe to splice from | |
897 | * @out: file to splice to | |
898 | * @ppos: position in @out | |
899 | * @len: how many bytes to splice | |
900 | * @flags: splice modifier flags | |
901 | * @actor: handler that splices the data | |
902 | * | |
903 | * Description: | |
2933970b | 904 | * See __splice_from_pipe. This function locks the pipe inode, |
932cc6d4 JA |
905 | * otherwise it's identical to __splice_from_pipe(). |
906 | * | |
907 | */ | |
6da61809 MF |
908 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
909 | loff_t *ppos, size_t len, unsigned int flags, | |
910 | splice_actor *actor) | |
911 | { | |
912 | ssize_t ret; | |
c66ab6fa JA |
913 | struct splice_desc sd = { |
914 | .total_len = len, | |
915 | .flags = flags, | |
916 | .pos = *ppos, | |
6a14b90b | 917 | .u.file = out, |
c66ab6fa | 918 | }; |
6da61809 | 919 | |
61e0d47c | 920 | pipe_lock(pipe); |
c66ab6fa | 921 | ret = __splice_from_pipe(pipe, &sd, actor); |
61e0d47c | 922 | pipe_unlock(pipe); |
6da61809 MF |
923 | |
924 | return ret; | |
925 | } | |
926 | ||
8d020765 AV |
927 | /** |
928 | * iter_file_splice_write - splice data from a pipe to a file | |
929 | * @pipe: pipe info | |
930 | * @out: file to write to | |
931 | * @ppos: position in @out | |
932 | * @len: number of bytes to splice | |
933 | * @flags: splice modifier flags | |
934 | * | |
935 | * Description: | |
936 | * Will either move or copy pages (determined by @flags options) from | |
937 | * the given pipe inode to the given file. | |
938 | * This one is ->write_iter-based. | |
939 | * | |
940 | */ | |
941 | ssize_t | |
942 | iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
943 | loff_t *ppos, size_t len, unsigned int flags) | |
944 | { | |
945 | struct splice_desc sd = { | |
946 | .total_len = len, | |
947 | .flags = flags, | |
948 | .pos = *ppos, | |
949 | .u.file = out, | |
950 | }; | |
951 | int nbufs = pipe->buffers; | |
952 | struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), | |
953 | GFP_KERNEL); | |
954 | ssize_t ret; | |
955 | ||
956 | if (unlikely(!array)) | |
957 | return -ENOMEM; | |
958 | ||
959 | pipe_lock(pipe); | |
960 | ||
961 | splice_from_pipe_begin(&sd); | |
962 | while (sd.total_len) { | |
963 | struct iov_iter from; | |
964 | struct kiocb kiocb; | |
965 | size_t left; | |
966 | int n, idx; | |
967 | ||
968 | ret = splice_from_pipe_next(pipe, &sd); | |
969 | if (ret <= 0) | |
970 | break; | |
971 | ||
972 | if (unlikely(nbufs < pipe->buffers)) { | |
973 | kfree(array); | |
974 | nbufs = pipe->buffers; | |
975 | array = kcalloc(nbufs, sizeof(struct bio_vec), | |
976 | GFP_KERNEL); | |
977 | if (!array) { | |
978 | ret = -ENOMEM; | |
979 | break; | |
980 | } | |
981 | } | |
982 | ||
983 | /* build the vector */ | |
984 | left = sd.total_len; | |
985 | for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) { | |
986 | struct pipe_buffer *buf = pipe->bufs + idx; | |
987 | size_t this_len = buf->len; | |
988 | ||
989 | if (this_len > left) | |
990 | this_len = left; | |
991 | ||
992 | if (idx == pipe->buffers - 1) | |
993 | idx = -1; | |
994 | ||
995 | ret = buf->ops->confirm(pipe, buf); | |
996 | if (unlikely(ret)) { | |
997 | if (ret == -ENODATA) | |
998 | ret = 0; | |
999 | goto done; | |
1000 | } | |
1001 | ||
1002 | array[n].bv_page = buf->page; | |
1003 | array[n].bv_len = this_len; | |
1004 | array[n].bv_offset = buf->offset; | |
1005 | left -= this_len; | |
1006 | } | |
1007 | ||
1008 | /* ... iov_iter */ | |
05afcb77 AV |
1009 | iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n, |
1010 | sd.total_len - left); | |
8d020765 AV |
1011 | |
1012 | /* ... and iocb */ | |
1013 | init_sync_kiocb(&kiocb, out); | |
1014 | kiocb.ki_pos = sd.pos; | |
1015 | kiocb.ki_nbytes = sd.total_len - left; | |
1016 | ||
1017 | /* now, send it */ | |
1018 | ret = out->f_op->write_iter(&kiocb, &from); | |
1019 | if (-EIOCBQUEUED == ret) | |
1020 | ret = wait_on_sync_kiocb(&kiocb); | |
1021 | ||
1022 | if (ret <= 0) | |
1023 | break; | |
1024 | ||
1025 | sd.num_spliced += ret; | |
1026 | sd.total_len -= ret; | |
1027 | *ppos = sd.pos = kiocb.ki_pos; | |
1028 | ||
1029 | /* dismiss the fully eaten buffers, adjust the partial one */ | |
1030 | while (ret) { | |
1031 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
1032 | if (ret >= buf->len) { | |
1033 | const struct pipe_buf_operations *ops = buf->ops; | |
1034 | ret -= buf->len; | |
1035 | buf->len = 0; | |
1036 | buf->ops = NULL; | |
1037 | ops->release(pipe, buf); | |
1038 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); | |
1039 | pipe->nrbufs--; | |
1040 | if (pipe->files) | |
1041 | sd.need_wakeup = true; | |
1042 | } else { | |
1043 | buf->offset += ret; | |
1044 | buf->len -= ret; | |
1045 | ret = 0; | |
1046 | } | |
1047 | } | |
1048 | } | |
1049 | done: | |
1050 | kfree(array); | |
1051 | splice_from_pipe_end(pipe, &sd); | |
1052 | ||
1053 | pipe_unlock(pipe); | |
1054 | ||
1055 | if (sd.num_spliced) | |
1056 | ret = sd.num_spliced; | |
1057 | ||
1058 | return ret; | |
1059 | } | |
1060 | ||
1061 | EXPORT_SYMBOL(iter_file_splice_write); | |
1062 | ||
b2858d7d MS |
1063 | static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1064 | struct splice_desc *sd) | |
0b0a47f5 | 1065 | { |
b2858d7d MS |
1066 | int ret; |
1067 | void *data; | |
06ae43f3 | 1068 | loff_t tmp = sd->pos; |
b2858d7d | 1069 | |
fbb32750 | 1070 | data = kmap(buf->page); |
06ae43f3 | 1071 | ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); |
fbb32750 | 1072 | kunmap(buf->page); |
b2858d7d MS |
1073 | |
1074 | return ret; | |
0b0a47f5 MS |
1075 | } |
1076 | ||
1077 | static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, | |
1078 | struct file *out, loff_t *ppos, | |
1079 | size_t len, unsigned int flags) | |
1080 | { | |
b2858d7d | 1081 | ssize_t ret; |
0b0a47f5 | 1082 | |
b2858d7d MS |
1083 | ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); |
1084 | if (ret > 0) | |
1085 | *ppos += ret; | |
0b0a47f5 | 1086 | |
b2858d7d | 1087 | return ret; |
0b0a47f5 MS |
1088 | } |
1089 | ||
83f9135b JA |
1090 | /** |
1091 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 1092 | * @pipe: pipe to splice from |
83f9135b | 1093 | * @out: socket to write to |
932cc6d4 | 1094 | * @ppos: position in @out |
83f9135b JA |
1095 | * @len: number of bytes to splice |
1096 | * @flags: splice modifier flags | |
1097 | * | |
932cc6d4 JA |
1098 | * Description: |
1099 | * Will send @len bytes from the pipe to a network socket. No data copying | |
1100 | * is involved. | |
83f9135b JA |
1101 | * |
1102 | */ | |
3a326a2c | 1103 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1104 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1105 | { |
00522fb4 | 1106 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
1107 | } |
1108 | ||
059a8f37 | 1109 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 1110 | |
83f9135b JA |
1111 | /* |
1112 | * Attempt to initiate a splice from pipe to file. | |
1113 | */ | |
3a326a2c | 1114 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1115 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1116 | { |
0b0a47f5 MS |
1117 | ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, |
1118 | loff_t *, size_t, unsigned int); | |
5274f052 | 1119 | |
72c2d531 | 1120 | if (out->f_op->splice_write) |
cc56f7de CG |
1121 | splice_write = out->f_op->splice_write; |
1122 | else | |
0b0a47f5 MS |
1123 | splice_write = default_file_splice_write; |
1124 | ||
500368f7 | 1125 | return splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
1126 | } |
1127 | ||
83f9135b JA |
1128 | /* |
1129 | * Attempt to initiate a splice from a file to a pipe. | |
1130 | */ | |
cbb7e577 JA |
1131 | static long do_splice_to(struct file *in, loff_t *ppos, |
1132 | struct pipe_inode_info *pipe, size_t len, | |
1133 | unsigned int flags) | |
5274f052 | 1134 | { |
6818173b MS |
1135 | ssize_t (*splice_read)(struct file *, loff_t *, |
1136 | struct pipe_inode_info *, size_t, unsigned int); | |
5274f052 JA |
1137 | int ret; |
1138 | ||
49570e9b | 1139 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
1140 | return -EBADF; |
1141 | ||
cbb7e577 | 1142 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
1143 | if (unlikely(ret < 0)) |
1144 | return ret; | |
1145 | ||
72c2d531 | 1146 | if (in->f_op->splice_read) |
cc56f7de CG |
1147 | splice_read = in->f_op->splice_read; |
1148 | else | |
6818173b MS |
1149 | splice_read = default_file_splice_read; |
1150 | ||
1151 | return splice_read(in, ppos, pipe, len, flags); | |
5274f052 JA |
1152 | } |
1153 | ||
932cc6d4 JA |
1154 | /** |
1155 | * splice_direct_to_actor - splices data directly between two non-pipes | |
1156 | * @in: file to splice from | |
1157 | * @sd: actor information on where to splice to | |
1158 | * @actor: handles the data splicing | |
1159 | * | |
1160 | * Description: | |
1161 | * This is a special case helper to splice directly between two | |
1162 | * points, without requiring an explicit pipe. Internally an allocated | |
79685b8d | 1163 | * pipe is cached in the process, and reused during the lifetime of |
932cc6d4 JA |
1164 | * that process. |
1165 | * | |
c66ab6fa JA |
1166 | */ |
1167 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1168 | splice_direct_actor *actor) | |
b92ce558 JA |
1169 | { |
1170 | struct pipe_inode_info *pipe; | |
1171 | long ret, bytes; | |
1172 | umode_t i_mode; | |
c66ab6fa JA |
1173 | size_t len; |
1174 | int i, flags; | |
b92ce558 JA |
1175 | |
1176 | /* | |
1177 | * We require the input being a regular file, as we don't want to | |
1178 | * randomly drop data for eg socket -> socket splicing. Use the | |
1179 | * piped splicing for that! | |
1180 | */ | |
496ad9aa | 1181 | i_mode = file_inode(in)->i_mode; |
b92ce558 JA |
1182 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1183 | return -EINVAL; | |
1184 | ||
1185 | /* | |
1186 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1187 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1188 | */ | |
1189 | pipe = current->splice_pipe; | |
49570e9b | 1190 | if (unlikely(!pipe)) { |
7bee130e | 1191 | pipe = alloc_pipe_info(); |
b92ce558 JA |
1192 | if (!pipe) |
1193 | return -ENOMEM; | |
1194 | ||
1195 | /* | |
1196 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1197 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1198 | * PIPE_READERS appropriately. |
1199 | */ | |
1200 | pipe->readers = 1; | |
1201 | ||
1202 | current->splice_pipe = pipe; | |
1203 | } | |
1204 | ||
1205 | /* | |
73d62d83 | 1206 | * Do the splice. |
b92ce558 JA |
1207 | */ |
1208 | ret = 0; | |
1209 | bytes = 0; | |
c66ab6fa JA |
1210 | len = sd->total_len; |
1211 | flags = sd->flags; | |
1212 | ||
1213 | /* | |
1214 | * Don't block on output, we have to drain the direct pipe. | |
1215 | */ | |
1216 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
b92ce558 JA |
1217 | |
1218 | while (len) { | |
51a92c0f | 1219 | size_t read_len; |
a82c53a0 | 1220 | loff_t pos = sd->pos, prev_pos = pos; |
b92ce558 | 1221 | |
bcd4f3ac | 1222 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1223 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1224 | goto out_release; |
1225 | ||
1226 | read_len = ret; | |
c66ab6fa | 1227 | sd->total_len = read_len; |
b92ce558 JA |
1228 | |
1229 | /* | |
1230 | * NOTE: nonblocking mode only applies to the input. We | |
1231 | * must not do the output in nonblocking mode as then we | |
1232 | * could get stuck data in the internal pipe: | |
1233 | */ | |
c66ab6fa | 1234 | ret = actor(pipe, sd); |
a82c53a0 TZ |
1235 | if (unlikely(ret <= 0)) { |
1236 | sd->pos = prev_pos; | |
b92ce558 | 1237 | goto out_release; |
a82c53a0 | 1238 | } |
b92ce558 JA |
1239 | |
1240 | bytes += ret; | |
1241 | len -= ret; | |
bcd4f3ac | 1242 | sd->pos = pos; |
b92ce558 | 1243 | |
a82c53a0 TZ |
1244 | if (ret < read_len) { |
1245 | sd->pos = prev_pos + ret; | |
51a92c0f | 1246 | goto out_release; |
a82c53a0 | 1247 | } |
b92ce558 JA |
1248 | } |
1249 | ||
9e97198d | 1250 | done: |
b92ce558 | 1251 | pipe->nrbufs = pipe->curbuf = 0; |
80848708 | 1252 | file_accessed(in); |
b92ce558 JA |
1253 | return bytes; |
1254 | ||
1255 | out_release: | |
1256 | /* | |
1257 | * If we did an incomplete transfer we must release | |
1258 | * the pipe buffers in question: | |
1259 | */ | |
35f3d14d | 1260 | for (i = 0; i < pipe->buffers; i++) { |
b92ce558 JA |
1261 | struct pipe_buffer *buf = pipe->bufs + i; |
1262 | ||
1263 | if (buf->ops) { | |
1264 | buf->ops->release(pipe, buf); | |
1265 | buf->ops = NULL; | |
1266 | } | |
1267 | } | |
b92ce558 | 1268 | |
9e97198d JA |
1269 | if (!bytes) |
1270 | bytes = ret; | |
c66ab6fa | 1271 | |
9e97198d | 1272 | goto done; |
c66ab6fa JA |
1273 | } |
1274 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1275 | ||
1276 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1277 | struct splice_desc *sd) | |
1278 | { | |
6a14b90b | 1279 | struct file *file = sd->u.file; |
c66ab6fa | 1280 | |
7995bd28 | 1281 | return do_splice_from(pipe, file, sd->opos, sd->total_len, |
2cb4b05e | 1282 | sd->flags); |
c66ab6fa JA |
1283 | } |
1284 | ||
932cc6d4 JA |
1285 | /** |
1286 | * do_splice_direct - splices data directly between two files | |
1287 | * @in: file to splice from | |
1288 | * @ppos: input file offset | |
1289 | * @out: file to splice to | |
acdb37c3 | 1290 | * @opos: output file offset |
932cc6d4 JA |
1291 | * @len: number of bytes to splice |
1292 | * @flags: splice modifier flags | |
1293 | * | |
1294 | * Description: | |
1295 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1296 | * doing it in the application would incur an extra system call | |
1297 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1298 | * can splice directly through a process-private pipe. | |
1299 | * | |
1300 | */ | |
c66ab6fa | 1301 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
7995bd28 | 1302 | loff_t *opos, size_t len, unsigned int flags) |
c66ab6fa JA |
1303 | { |
1304 | struct splice_desc sd = { | |
1305 | .len = len, | |
1306 | .total_len = len, | |
1307 | .flags = flags, | |
1308 | .pos = *ppos, | |
6a14b90b | 1309 | .u.file = out, |
7995bd28 | 1310 | .opos = opos, |
c66ab6fa | 1311 | }; |
51a92c0f | 1312 | long ret; |
c66ab6fa | 1313 | |
18c67cb9 AV |
1314 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1315 | return -EBADF; | |
1316 | ||
1317 | if (unlikely(out->f_flags & O_APPEND)) | |
1318 | return -EINVAL; | |
1319 | ||
1320 | ret = rw_verify_area(WRITE, out, opos, len); | |
1321 | if (unlikely(ret < 0)) | |
1322 | return ret; | |
1323 | ||
c66ab6fa | 1324 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); |
51a92c0f | 1325 | if (ret > 0) |
a82c53a0 | 1326 | *ppos = sd.pos; |
51a92c0f | 1327 | |
c66ab6fa | 1328 | return ret; |
b92ce558 | 1329 | } |
1c118596 | 1330 | EXPORT_SYMBOL(do_splice_direct); |
b92ce558 | 1331 | |
7c77f0b3 MS |
1332 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
1333 | struct pipe_inode_info *opipe, | |
1334 | size_t len, unsigned int flags); | |
ddac0d39 | 1335 | |
83f9135b JA |
1336 | /* |
1337 | * Determine where to splice to/from. | |
1338 | */ | |
529565dc IM |
1339 | static long do_splice(struct file *in, loff_t __user *off_in, |
1340 | struct file *out, loff_t __user *off_out, | |
1341 | size_t len, unsigned int flags) | |
5274f052 | 1342 | { |
7c77f0b3 MS |
1343 | struct pipe_inode_info *ipipe; |
1344 | struct pipe_inode_info *opipe; | |
7995bd28 | 1345 | loff_t offset; |
a4514ebd | 1346 | long ret; |
5274f052 | 1347 | |
71993e62 LT |
1348 | ipipe = get_pipe_info(in); |
1349 | opipe = get_pipe_info(out); | |
7c77f0b3 MS |
1350 | |
1351 | if (ipipe && opipe) { | |
1352 | if (off_in || off_out) | |
1353 | return -ESPIPE; | |
1354 | ||
1355 | if (!(in->f_mode & FMODE_READ)) | |
1356 | return -EBADF; | |
1357 | ||
1358 | if (!(out->f_mode & FMODE_WRITE)) | |
1359 | return -EBADF; | |
1360 | ||
1361 | /* Splicing to self would be fun, but... */ | |
1362 | if (ipipe == opipe) | |
1363 | return -EINVAL; | |
1364 | ||
1365 | return splice_pipe_to_pipe(ipipe, opipe, len, flags); | |
1366 | } | |
1367 | ||
1368 | if (ipipe) { | |
529565dc IM |
1369 | if (off_in) |
1370 | return -ESPIPE; | |
b92ce558 | 1371 | if (off_out) { |
19c9a49b | 1372 | if (!(out->f_mode & FMODE_PWRITE)) |
b92ce558 | 1373 | return -EINVAL; |
cbb7e577 | 1374 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1375 | return -EFAULT; |
7995bd28 AV |
1376 | } else { |
1377 | offset = out->f_pos; | |
1378 | } | |
529565dc | 1379 | |
18c67cb9 AV |
1380 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1381 | return -EBADF; | |
1382 | ||
1383 | if (unlikely(out->f_flags & O_APPEND)) | |
1384 | return -EINVAL; | |
1385 | ||
1386 | ret = rw_verify_area(WRITE, out, &offset, len); | |
1387 | if (unlikely(ret < 0)) | |
1388 | return ret; | |
1389 | ||
500368f7 | 1390 | file_start_write(out); |
7995bd28 | 1391 | ret = do_splice_from(ipipe, out, &offset, len, flags); |
500368f7 | 1392 | file_end_write(out); |
a4514ebd | 1393 | |
7995bd28 AV |
1394 | if (!off_out) |
1395 | out->f_pos = offset; | |
1396 | else if (copy_to_user(off_out, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1397 | ret = -EFAULT; |
1398 | ||
1399 | return ret; | |
529565dc | 1400 | } |
5274f052 | 1401 | |
7c77f0b3 | 1402 | if (opipe) { |
529565dc IM |
1403 | if (off_out) |
1404 | return -ESPIPE; | |
b92ce558 | 1405 | if (off_in) { |
19c9a49b | 1406 | if (!(in->f_mode & FMODE_PREAD)) |
b92ce558 | 1407 | return -EINVAL; |
cbb7e577 | 1408 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1409 | return -EFAULT; |
7995bd28 AV |
1410 | } else { |
1411 | offset = in->f_pos; | |
1412 | } | |
529565dc | 1413 | |
7995bd28 | 1414 | ret = do_splice_to(in, &offset, opipe, len, flags); |
a4514ebd | 1415 | |
7995bd28 AV |
1416 | if (!off_in) |
1417 | in->f_pos = offset; | |
1418 | else if (copy_to_user(off_in, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1419 | ret = -EFAULT; |
1420 | ||
1421 | return ret; | |
529565dc | 1422 | } |
5274f052 JA |
1423 | |
1424 | return -EINVAL; | |
1425 | } | |
1426 | ||
912d35f8 JA |
1427 | /* |
1428 | * Map an iov into an array of pages and offset/length tupples. With the | |
1429 | * partial_page structure, we can map several non-contiguous ranges into | |
1430 | * our ones pages[] map instead of splitting that operation into pieces. | |
1431 | * Could easily be exported as a generic helper for other users, in which | |
1432 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1433 | */ | |
1434 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1435 | unsigned int nr_vecs, struct page **pages, | |
bd1a68b5 | 1436 | struct partial_page *partial, bool aligned, |
35f3d14d | 1437 | unsigned int pipe_buffers) |
912d35f8 JA |
1438 | { |
1439 | int buffers = 0, error = 0; | |
1440 | ||
912d35f8 JA |
1441 | while (nr_vecs) { |
1442 | unsigned long off, npages; | |
75723957 | 1443 | struct iovec entry; |
912d35f8 JA |
1444 | void __user *base; |
1445 | size_t len; | |
1446 | int i; | |
1447 | ||
75723957 | 1448 | error = -EFAULT; |
bc40d73c | 1449 | if (copy_from_user(&entry, iov, sizeof(entry))) |
912d35f8 JA |
1450 | break; |
1451 | ||
75723957 LT |
1452 | base = entry.iov_base; |
1453 | len = entry.iov_len; | |
1454 | ||
912d35f8 JA |
1455 | /* |
1456 | * Sanity check this iovec. 0 read succeeds. | |
1457 | */ | |
75723957 | 1458 | error = 0; |
912d35f8 JA |
1459 | if (unlikely(!len)) |
1460 | break; | |
1461 | error = -EFAULT; | |
712a30e6 | 1462 | if (!access_ok(VERIFY_READ, base, len)) |
912d35f8 JA |
1463 | break; |
1464 | ||
1465 | /* | |
1466 | * Get this base offset and number of pages, then map | |
1467 | * in the user pages. | |
1468 | */ | |
1469 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1470 | |
1471 | /* | |
1472 | * If asked for alignment, the offset must be zero and the | |
1473 | * length a multiple of the PAGE_SIZE. | |
1474 | */ | |
1475 | error = -EINVAL; | |
1476 | if (aligned && (off || len & ~PAGE_MASK)) | |
1477 | break; | |
1478 | ||
912d35f8 | 1479 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
35f3d14d JA |
1480 | if (npages > pipe_buffers - buffers) |
1481 | npages = pipe_buffers - buffers; | |
912d35f8 | 1482 | |
bc40d73c NP |
1483 | error = get_user_pages_fast((unsigned long)base, npages, |
1484 | 0, &pages[buffers]); | |
912d35f8 JA |
1485 | |
1486 | if (unlikely(error <= 0)) | |
1487 | break; | |
1488 | ||
1489 | /* | |
1490 | * Fill this contiguous range into the partial page map. | |
1491 | */ | |
1492 | for (i = 0; i < error; i++) { | |
7591489a | 1493 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1494 | |
1495 | partial[buffers].offset = off; | |
1496 | partial[buffers].len = plen; | |
1497 | ||
1498 | off = 0; | |
1499 | len -= plen; | |
1500 | buffers++; | |
1501 | } | |
1502 | ||
1503 | /* | |
1504 | * We didn't complete this iov, stop here since it probably | |
1505 | * means we have to move some of this into a pipe to | |
1506 | * be able to continue. | |
1507 | */ | |
1508 | if (len) | |
1509 | break; | |
1510 | ||
1511 | /* | |
1512 | * Don't continue if we mapped fewer pages than we asked for, | |
1513 | * or if we mapped the max number of pages that we have | |
1514 | * room for. | |
1515 | */ | |
35f3d14d | 1516 | if (error < npages || buffers == pipe_buffers) |
912d35f8 JA |
1517 | break; |
1518 | ||
1519 | nr_vecs--; | |
1520 | iov++; | |
1521 | } | |
1522 | ||
912d35f8 JA |
1523 | if (buffers) |
1524 | return buffers; | |
1525 | ||
1526 | return error; | |
1527 | } | |
1528 | ||
6a14b90b JA |
1529 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1530 | struct splice_desc *sd) | |
1531 | { | |
6130f531 AV |
1532 | int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); |
1533 | return n == sd->len ? n : -EFAULT; | |
6a14b90b JA |
1534 | } |
1535 | ||
1536 | /* | |
1537 | * For lack of a better implementation, implement vmsplice() to userspace | |
1538 | * as a simple copy of the pipes pages to the user iov. | |
1539 | */ | |
6130f531 | 1540 | static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov, |
6a14b90b JA |
1541 | unsigned long nr_segs, unsigned int flags) |
1542 | { | |
1543 | struct pipe_inode_info *pipe; | |
1544 | struct splice_desc sd; | |
6a14b90b | 1545 | long ret; |
6130f531 AV |
1546 | struct iovec iovstack[UIO_FASTIOV]; |
1547 | struct iovec *iov = iovstack; | |
1548 | struct iov_iter iter; | |
b6dd6f47 | 1549 | ssize_t count; |
6a14b90b | 1550 | |
71993e62 | 1551 | pipe = get_pipe_info(file); |
6a14b90b JA |
1552 | if (!pipe) |
1553 | return -EBADF; | |
1554 | ||
6130f531 AV |
1555 | ret = rw_copy_check_uvector(READ, uiov, nr_segs, |
1556 | ARRAY_SIZE(iovstack), iovstack, &iov); | |
1557 | if (ret <= 0) | |
b6dd6f47 | 1558 | goto out; |
6a14b90b | 1559 | |
b6dd6f47 | 1560 | count = ret; |
71d8e532 | 1561 | iov_iter_init(&iter, READ, iov, nr_segs, count); |
6a14b90b | 1562 | |
6130f531 AV |
1563 | sd.len = 0; |
1564 | sd.total_len = count; | |
1565 | sd.flags = flags; | |
1566 | sd.u.data = &iter; | |
1567 | sd.pos = 0; | |
6a14b90b | 1568 | |
6130f531 AV |
1569 | pipe_lock(pipe); |
1570 | ret = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
61e0d47c | 1571 | pipe_unlock(pipe); |
6a14b90b | 1572 | |
b6dd6f47 | 1573 | out: |
6130f531 AV |
1574 | if (iov != iovstack) |
1575 | kfree(iov); | |
6a14b90b JA |
1576 | |
1577 | return ret; | |
1578 | } | |
1579 | ||
912d35f8 JA |
1580 | /* |
1581 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1582 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1583 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1584 | */ |
6a14b90b JA |
1585 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1586 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1587 | { |
ddac0d39 | 1588 | struct pipe_inode_info *pipe; |
35f3d14d JA |
1589 | struct page *pages[PIPE_DEF_BUFFERS]; |
1590 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
912d35f8 JA |
1591 | struct splice_pipe_desc spd = { |
1592 | .pages = pages, | |
1593 | .partial = partial, | |
047fe360 | 1594 | .nr_pages_max = PIPE_DEF_BUFFERS, |
912d35f8 JA |
1595 | .flags = flags, |
1596 | .ops = &user_page_pipe_buf_ops, | |
bbdfc2f7 | 1597 | .spd_release = spd_release_page, |
912d35f8 | 1598 | }; |
35f3d14d | 1599 | long ret; |
912d35f8 | 1600 | |
71993e62 | 1601 | pipe = get_pipe_info(file); |
ddac0d39 | 1602 | if (!pipe) |
912d35f8 | 1603 | return -EBADF; |
912d35f8 | 1604 | |
35f3d14d JA |
1605 | if (splice_grow_spd(pipe, &spd)) |
1606 | return -ENOMEM; | |
1607 | ||
1608 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages, | |
bd1a68b5 | 1609 | spd.partial, false, |
047fe360 | 1610 | spd.nr_pages_max); |
912d35f8 | 1611 | if (spd.nr_pages <= 0) |
35f3d14d JA |
1612 | ret = spd.nr_pages; |
1613 | else | |
1614 | ret = splice_to_pipe(pipe, &spd); | |
912d35f8 | 1615 | |
047fe360 | 1616 | splice_shrink_spd(&spd); |
35f3d14d | 1617 | return ret; |
912d35f8 JA |
1618 | } |
1619 | ||
6a14b90b JA |
1620 | /* |
1621 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1622 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1623 | * operation that can be supported without any funky alignment restrictions | |
1624 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1625 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1626 | * solutions for that: | |
1627 | * | |
1628 | * - memcpy() the data internally, at which point we might as well just | |
1629 | * do a regular read() on the buffer anyway. | |
1630 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1631 | * has restriction limitations on both ends of the pipe). | |
1632 | * | |
1633 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1634 | * | |
1635 | */ | |
836f92ad HC |
1636 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, |
1637 | unsigned long, nr_segs, unsigned int, flags) | |
912d35f8 | 1638 | { |
2903ff01 | 1639 | struct fd f; |
912d35f8 | 1640 | long error; |
912d35f8 | 1641 | |
6a14b90b JA |
1642 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1643 | return -EINVAL; | |
1644 | else if (unlikely(!nr_segs)) | |
1645 | return 0; | |
1646 | ||
912d35f8 | 1647 | error = -EBADF; |
2903ff01 AV |
1648 | f = fdget(fd); |
1649 | if (f.file) { | |
1650 | if (f.file->f_mode & FMODE_WRITE) | |
1651 | error = vmsplice_to_pipe(f.file, iov, nr_segs, flags); | |
1652 | else if (f.file->f_mode & FMODE_READ) | |
1653 | error = vmsplice_to_user(f.file, iov, nr_segs, flags); | |
1654 | ||
1655 | fdput(f); | |
912d35f8 JA |
1656 | } |
1657 | ||
1658 | return error; | |
1659 | } | |
1660 | ||
76b021d0 AV |
1661 | #ifdef CONFIG_COMPAT |
1662 | COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, | |
1663 | unsigned int, nr_segs, unsigned int, flags) | |
1664 | { | |
1665 | unsigned i; | |
1666 | struct iovec __user *iov; | |
1667 | if (nr_segs > UIO_MAXIOV) | |
1668 | return -EINVAL; | |
1669 | iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec)); | |
1670 | for (i = 0; i < nr_segs; i++) { | |
1671 | struct compat_iovec v; | |
1672 | if (get_user(v.iov_base, &iov32[i].iov_base) || | |
1673 | get_user(v.iov_len, &iov32[i].iov_len) || | |
1674 | put_user(compat_ptr(v.iov_base), &iov[i].iov_base) || | |
1675 | put_user(v.iov_len, &iov[i].iov_len)) | |
1676 | return -EFAULT; | |
1677 | } | |
1678 | return sys_vmsplice(fd, iov, nr_segs, flags); | |
1679 | } | |
1680 | #endif | |
1681 | ||
836f92ad HC |
1682 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
1683 | int, fd_out, loff_t __user *, off_out, | |
1684 | size_t, len, unsigned int, flags) | |
5274f052 | 1685 | { |
2903ff01 | 1686 | struct fd in, out; |
5274f052 | 1687 | long error; |
5274f052 JA |
1688 | |
1689 | if (unlikely(!len)) | |
1690 | return 0; | |
1691 | ||
1692 | error = -EBADF; | |
2903ff01 AV |
1693 | in = fdget(fd_in); |
1694 | if (in.file) { | |
1695 | if (in.file->f_mode & FMODE_READ) { | |
1696 | out = fdget(fd_out); | |
1697 | if (out.file) { | |
1698 | if (out.file->f_mode & FMODE_WRITE) | |
1699 | error = do_splice(in.file, off_in, | |
1700 | out.file, off_out, | |
529565dc | 1701 | len, flags); |
2903ff01 | 1702 | fdput(out); |
5274f052 JA |
1703 | } |
1704 | } | |
2903ff01 | 1705 | fdput(in); |
5274f052 | 1706 | } |
5274f052 JA |
1707 | return error; |
1708 | } | |
70524490 | 1709 | |
aadd06e5 JA |
1710 | /* |
1711 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1712 | * return an appropriate error. | |
1713 | */ | |
7c77f0b3 | 1714 | static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1715 | { |
1716 | int ret; | |
1717 | ||
1718 | /* | |
1719 | * Check ->nrbufs without the inode lock first. This function | |
1720 | * is speculative anyways, so missing one is ok. | |
1721 | */ | |
1722 | if (pipe->nrbufs) | |
1723 | return 0; | |
1724 | ||
1725 | ret = 0; | |
61e0d47c | 1726 | pipe_lock(pipe); |
aadd06e5 JA |
1727 | |
1728 | while (!pipe->nrbufs) { | |
1729 | if (signal_pending(current)) { | |
1730 | ret = -ERESTARTSYS; | |
1731 | break; | |
1732 | } | |
1733 | if (!pipe->writers) | |
1734 | break; | |
1735 | if (!pipe->waiting_writers) { | |
1736 | if (flags & SPLICE_F_NONBLOCK) { | |
1737 | ret = -EAGAIN; | |
1738 | break; | |
1739 | } | |
1740 | } | |
1741 | pipe_wait(pipe); | |
1742 | } | |
1743 | ||
61e0d47c | 1744 | pipe_unlock(pipe); |
aadd06e5 JA |
1745 | return ret; |
1746 | } | |
1747 | ||
1748 | /* | |
1749 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1750 | * return an appropriate error. | |
1751 | */ | |
7c77f0b3 | 1752 | static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1753 | { |
1754 | int ret; | |
1755 | ||
1756 | /* | |
1757 | * Check ->nrbufs without the inode lock first. This function | |
1758 | * is speculative anyways, so missing one is ok. | |
1759 | */ | |
35f3d14d | 1760 | if (pipe->nrbufs < pipe->buffers) |
aadd06e5 JA |
1761 | return 0; |
1762 | ||
1763 | ret = 0; | |
61e0d47c | 1764 | pipe_lock(pipe); |
aadd06e5 | 1765 | |
35f3d14d | 1766 | while (pipe->nrbufs >= pipe->buffers) { |
aadd06e5 JA |
1767 | if (!pipe->readers) { |
1768 | send_sig(SIGPIPE, current, 0); | |
1769 | ret = -EPIPE; | |
1770 | break; | |
1771 | } | |
1772 | if (flags & SPLICE_F_NONBLOCK) { | |
1773 | ret = -EAGAIN; | |
1774 | break; | |
1775 | } | |
1776 | if (signal_pending(current)) { | |
1777 | ret = -ERESTARTSYS; | |
1778 | break; | |
1779 | } | |
1780 | pipe->waiting_writers++; | |
1781 | pipe_wait(pipe); | |
1782 | pipe->waiting_writers--; | |
1783 | } | |
1784 | ||
61e0d47c | 1785 | pipe_unlock(pipe); |
aadd06e5 JA |
1786 | return ret; |
1787 | } | |
1788 | ||
7c77f0b3 MS |
1789 | /* |
1790 | * Splice contents of ipipe to opipe. | |
1791 | */ | |
1792 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, | |
1793 | struct pipe_inode_info *opipe, | |
1794 | size_t len, unsigned int flags) | |
1795 | { | |
1796 | struct pipe_buffer *ibuf, *obuf; | |
1797 | int ret = 0, nbuf; | |
1798 | bool input_wakeup = false; | |
1799 | ||
1800 | ||
1801 | retry: | |
1802 | ret = ipipe_prep(ipipe, flags); | |
1803 | if (ret) | |
1804 | return ret; | |
1805 | ||
1806 | ret = opipe_prep(opipe, flags); | |
1807 | if (ret) | |
1808 | return ret; | |
1809 | ||
1810 | /* | |
1811 | * Potential ABBA deadlock, work around it by ordering lock | |
1812 | * grabbing by pipe info address. Otherwise two different processes | |
1813 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1814 | */ | |
1815 | pipe_double_lock(ipipe, opipe); | |
1816 | ||
1817 | do { | |
1818 | if (!opipe->readers) { | |
1819 | send_sig(SIGPIPE, current, 0); | |
1820 | if (!ret) | |
1821 | ret = -EPIPE; | |
1822 | break; | |
1823 | } | |
1824 | ||
1825 | if (!ipipe->nrbufs && !ipipe->writers) | |
1826 | break; | |
1827 | ||
1828 | /* | |
1829 | * Cannot make any progress, because either the input | |
1830 | * pipe is empty or the output pipe is full. | |
1831 | */ | |
35f3d14d | 1832 | if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { |
7c77f0b3 MS |
1833 | /* Already processed some buffers, break */ |
1834 | if (ret) | |
1835 | break; | |
1836 | ||
1837 | if (flags & SPLICE_F_NONBLOCK) { | |
1838 | ret = -EAGAIN; | |
1839 | break; | |
1840 | } | |
1841 | ||
1842 | /* | |
1843 | * We raced with another reader/writer and haven't | |
1844 | * managed to process any buffers. A zero return | |
1845 | * value means EOF, so retry instead. | |
1846 | */ | |
1847 | pipe_unlock(ipipe); | |
1848 | pipe_unlock(opipe); | |
1849 | goto retry; | |
1850 | } | |
1851 | ||
1852 | ibuf = ipipe->bufs + ipipe->curbuf; | |
35f3d14d | 1853 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); |
7c77f0b3 MS |
1854 | obuf = opipe->bufs + nbuf; |
1855 | ||
1856 | if (len >= ibuf->len) { | |
1857 | /* | |
1858 | * Simply move the whole buffer from ipipe to opipe | |
1859 | */ | |
1860 | *obuf = *ibuf; | |
1861 | ibuf->ops = NULL; | |
1862 | opipe->nrbufs++; | |
35f3d14d | 1863 | ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); |
7c77f0b3 MS |
1864 | ipipe->nrbufs--; |
1865 | input_wakeup = true; | |
1866 | } else { | |
1867 | /* | |
1868 | * Get a reference to this pipe buffer, | |
1869 | * so we can copy the contents over. | |
1870 | */ | |
1871 | ibuf->ops->get(ipipe, ibuf); | |
1872 | *obuf = *ibuf; | |
1873 | ||
1874 | /* | |
1875 | * Don't inherit the gift flag, we need to | |
1876 | * prevent multiple steals of this page. | |
1877 | */ | |
1878 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1879 | ||
1880 | obuf->len = len; | |
1881 | opipe->nrbufs++; | |
1882 | ibuf->offset += obuf->len; | |
1883 | ibuf->len -= obuf->len; | |
1884 | } | |
1885 | ret += obuf->len; | |
1886 | len -= obuf->len; | |
1887 | } while (len); | |
1888 | ||
1889 | pipe_unlock(ipipe); | |
1890 | pipe_unlock(opipe); | |
1891 | ||
1892 | /* | |
1893 | * If we put data in the output pipe, wakeup any potential readers. | |
1894 | */ | |
825cdcb1 NK |
1895 | if (ret > 0) |
1896 | wakeup_pipe_readers(opipe); | |
1897 | ||
7c77f0b3 MS |
1898 | if (input_wakeup) |
1899 | wakeup_pipe_writers(ipipe); | |
1900 | ||
1901 | return ret; | |
1902 | } | |
1903 | ||
70524490 JA |
1904 | /* |
1905 | * Link contents of ipipe to opipe. | |
1906 | */ | |
1907 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1908 | struct pipe_inode_info *opipe, | |
1909 | size_t len, unsigned int flags) | |
1910 | { | |
1911 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1912 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1913 | |
1914 | /* | |
1915 | * Potential ABBA deadlock, work around it by ordering lock | |
61e0d47c | 1916 | * grabbing by pipe info address. Otherwise two different processes |
70524490 JA |
1917 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
1918 | */ | |
61e0d47c | 1919 | pipe_double_lock(ipipe, opipe); |
70524490 | 1920 | |
aadd06e5 | 1921 | do { |
70524490 JA |
1922 | if (!opipe->readers) { |
1923 | send_sig(SIGPIPE, current, 0); | |
1924 | if (!ret) | |
1925 | ret = -EPIPE; | |
1926 | break; | |
1927 | } | |
70524490 | 1928 | |
aadd06e5 JA |
1929 | /* |
1930 | * If we have iterated all input buffers or ran out of | |
1931 | * output room, break. | |
1932 | */ | |
35f3d14d | 1933 | if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) |
aadd06e5 | 1934 | break; |
70524490 | 1935 | |
35f3d14d JA |
1936 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); |
1937 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); | |
70524490 JA |
1938 | |
1939 | /* | |
aadd06e5 JA |
1940 | * Get a reference to this pipe buffer, |
1941 | * so we can copy the contents over. | |
70524490 | 1942 | */ |
aadd06e5 JA |
1943 | ibuf->ops->get(ipipe, ibuf); |
1944 | ||
1945 | obuf = opipe->bufs + nbuf; | |
1946 | *obuf = *ibuf; | |
1947 | ||
2a27250e | 1948 | /* |
aadd06e5 JA |
1949 | * Don't inherit the gift flag, we need to |
1950 | * prevent multiple steals of this page. | |
2a27250e | 1951 | */ |
aadd06e5 | 1952 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1953 | |
aadd06e5 JA |
1954 | if (obuf->len > len) |
1955 | obuf->len = len; | |
70524490 | 1956 | |
aadd06e5 JA |
1957 | opipe->nrbufs++; |
1958 | ret += obuf->len; | |
1959 | len -= obuf->len; | |
1960 | i++; | |
1961 | } while (len); | |
70524490 | 1962 | |
02cf01ae JA |
1963 | /* |
1964 | * return EAGAIN if we have the potential of some data in the | |
1965 | * future, otherwise just return 0 | |
1966 | */ | |
1967 | if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) | |
1968 | ret = -EAGAIN; | |
1969 | ||
61e0d47c MS |
1970 | pipe_unlock(ipipe); |
1971 | pipe_unlock(opipe); | |
70524490 | 1972 | |
aadd06e5 JA |
1973 | /* |
1974 | * If we put data in the output pipe, wakeup any potential readers. | |
1975 | */ | |
825cdcb1 NK |
1976 | if (ret > 0) |
1977 | wakeup_pipe_readers(opipe); | |
70524490 JA |
1978 | |
1979 | return ret; | |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1984 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1985 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1986 | * applicable one is SPLICE_F_NONBLOCK. | |
1987 | */ | |
1988 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1989 | unsigned int flags) | |
1990 | { | |
71993e62 LT |
1991 | struct pipe_inode_info *ipipe = get_pipe_info(in); |
1992 | struct pipe_inode_info *opipe = get_pipe_info(out); | |
aadd06e5 | 1993 | int ret = -EINVAL; |
70524490 JA |
1994 | |
1995 | /* | |
aadd06e5 JA |
1996 | * Duplicate the contents of ipipe to opipe without actually |
1997 | * copying the data. | |
70524490 | 1998 | */ |
aadd06e5 JA |
1999 | if (ipipe && opipe && ipipe != opipe) { |
2000 | /* | |
2001 | * Keep going, unless we encounter an error. The ipipe/opipe | |
2002 | * ordering doesn't really matter. | |
2003 | */ | |
7c77f0b3 | 2004 | ret = ipipe_prep(ipipe, flags); |
aadd06e5 | 2005 | if (!ret) { |
7c77f0b3 | 2006 | ret = opipe_prep(opipe, flags); |
02cf01ae | 2007 | if (!ret) |
aadd06e5 | 2008 | ret = link_pipe(ipipe, opipe, len, flags); |
aadd06e5 JA |
2009 | } |
2010 | } | |
70524490 | 2011 | |
aadd06e5 | 2012 | return ret; |
70524490 JA |
2013 | } |
2014 | ||
836f92ad | 2015 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
70524490 | 2016 | { |
2903ff01 AV |
2017 | struct fd in; |
2018 | int error; | |
70524490 JA |
2019 | |
2020 | if (unlikely(!len)) | |
2021 | return 0; | |
2022 | ||
2023 | error = -EBADF; | |
2903ff01 AV |
2024 | in = fdget(fdin); |
2025 | if (in.file) { | |
2026 | if (in.file->f_mode & FMODE_READ) { | |
2027 | struct fd out = fdget(fdout); | |
2028 | if (out.file) { | |
2029 | if (out.file->f_mode & FMODE_WRITE) | |
2030 | error = do_tee(in.file, out.file, | |
2031 | len, flags); | |
2032 | fdput(out); | |
70524490 JA |
2033 | } |
2034 | } | |
2903ff01 | 2035 | fdput(in); |
70524490 JA |
2036 | } |
2037 | ||
2038 | return error; | |
2039 | } |