]>
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, |
cac36bb0 | 139 | .confirm = page_cache_pipe_buf_confirm, |
5274f052 | 140 | .release = page_cache_pipe_buf_release, |
5abc97aa | 141 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 142 | .get = generic_pipe_buf_get, |
5274f052 JA |
143 | }; |
144 | ||
912d35f8 JA |
145 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
146 | struct pipe_buffer *buf) | |
147 | { | |
7afa6fd0 JA |
148 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
149 | return 1; | |
150 | ||
1432873a | 151 | buf->flags |= PIPE_BUF_FLAG_LRU; |
330ab716 | 152 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
153 | } |
154 | ||
d4c3cca9 | 155 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
912d35f8 | 156 | .can_merge = 0, |
cac36bb0 | 157 | .confirm = generic_pipe_buf_confirm, |
912d35f8 JA |
158 | .release = page_cache_pipe_buf_release, |
159 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 160 | .get = generic_pipe_buf_get, |
912d35f8 JA |
161 | }; |
162 | ||
825cdcb1 NK |
163 | static void wakeup_pipe_readers(struct pipe_inode_info *pipe) |
164 | { | |
165 | smp_mb(); | |
166 | if (waitqueue_active(&pipe->wait)) | |
167 | wake_up_interruptible(&pipe->wait); | |
168 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
169 | } | |
170 | ||
932cc6d4 JA |
171 | /** |
172 | * splice_to_pipe - fill passed data into a pipe | |
173 | * @pipe: pipe to fill | |
174 | * @spd: data to fill | |
175 | * | |
176 | * Description: | |
79685b8d | 177 | * @spd contains a map of pages and len/offset tuples, along with |
932cc6d4 JA |
178 | * the struct pipe_buf_operations associated with these pages. This |
179 | * function will link that data to the pipe. | |
180 | * | |
83f9135b | 181 | */ |
d6b29d7c JA |
182 | ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
183 | struct splice_pipe_desc *spd) | |
5274f052 | 184 | { |
00de00bd | 185 | unsigned int spd_pages = spd->nr_pages; |
912d35f8 | 186 | int ret, do_wakeup, page_nr; |
5274f052 JA |
187 | |
188 | ret = 0; | |
189 | do_wakeup = 0; | |
912d35f8 | 190 | page_nr = 0; |
5274f052 | 191 | |
61e0d47c | 192 | pipe_lock(pipe); |
5274f052 | 193 | |
5274f052 | 194 | for (;;) { |
3a326a2c | 195 | if (!pipe->readers) { |
5274f052 JA |
196 | send_sig(SIGPIPE, current, 0); |
197 | if (!ret) | |
198 | ret = -EPIPE; | |
199 | break; | |
200 | } | |
201 | ||
35f3d14d JA |
202 | if (pipe->nrbufs < pipe->buffers) { |
203 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
3a326a2c | 204 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 205 | |
912d35f8 JA |
206 | buf->page = spd->pages[page_nr]; |
207 | buf->offset = spd->partial[page_nr].offset; | |
208 | buf->len = spd->partial[page_nr].len; | |
497f9625 | 209 | buf->private = spd->partial[page_nr].private; |
912d35f8 | 210 | buf->ops = spd->ops; |
7afa6fd0 JA |
211 | if (spd->flags & SPLICE_F_GIFT) |
212 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
213 | ||
6f767b04 | 214 | pipe->nrbufs++; |
912d35f8 JA |
215 | page_nr++; |
216 | ret += buf->len; | |
217 | ||
6447a3cf | 218 | if (pipe->files) |
6f767b04 | 219 | do_wakeup = 1; |
5274f052 | 220 | |
912d35f8 | 221 | if (!--spd->nr_pages) |
5274f052 | 222 | break; |
35f3d14d | 223 | if (pipe->nrbufs < pipe->buffers) |
5274f052 JA |
224 | continue; |
225 | ||
226 | break; | |
227 | } | |
228 | ||
912d35f8 | 229 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
230 | if (!ret) |
231 | ret = -EAGAIN; | |
232 | break; | |
233 | } | |
234 | ||
5274f052 JA |
235 | if (signal_pending(current)) { |
236 | if (!ret) | |
237 | ret = -ERESTARTSYS; | |
238 | break; | |
239 | } | |
240 | ||
241 | if (do_wakeup) { | |
c0bd1f65 | 242 | smp_mb(); |
3a326a2c IM |
243 | if (waitqueue_active(&pipe->wait)) |
244 | wake_up_interruptible_sync(&pipe->wait); | |
245 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
246 | do_wakeup = 0; |
247 | } | |
248 | ||
3a326a2c IM |
249 | pipe->waiting_writers++; |
250 | pipe_wait(pipe); | |
251 | pipe->waiting_writers--; | |
5274f052 JA |
252 | } |
253 | ||
61e0d47c | 254 | pipe_unlock(pipe); |
5274f052 | 255 | |
825cdcb1 NK |
256 | if (do_wakeup) |
257 | wakeup_pipe_readers(pipe); | |
5274f052 | 258 | |
00de00bd | 259 | while (page_nr < spd_pages) |
bbdfc2f7 | 260 | spd->spd_release(spd, page_nr++); |
5274f052 JA |
261 | |
262 | return ret; | |
263 | } | |
2b514574 | 264 | EXPORT_SYMBOL_GPL(splice_to_pipe); |
5274f052 | 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, | |
c62d2555 | 363 | mapping_gfp_constraint(mapping, 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 | 526 | |
be64f884 BH |
527 | if (IS_DAX(in->f_mapping->host)) |
528 | return default_file_splice_read(in, ppos, pipe, len, flags); | |
529 | ||
d366d398 JA |
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, | |
6818173b MS |
550 | .confirm = generic_pipe_buf_confirm, |
551 | .release = generic_pipe_buf_release, | |
552 | .steal = generic_pipe_buf_steal, | |
553 | .get = generic_pipe_buf_get, | |
554 | }; | |
555 | ||
28a625cb MS |
556 | static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, |
557 | struct pipe_buffer *buf) | |
558 | { | |
559 | return 1; | |
560 | } | |
561 | ||
562 | /* Pipe buffer operations for a socket and similar. */ | |
563 | const struct pipe_buf_operations nosteal_pipe_buf_ops = { | |
564 | .can_merge = 0, | |
28a625cb MS |
565 | .confirm = generic_pipe_buf_confirm, |
566 | .release = generic_pipe_buf_release, | |
567 | .steal = generic_pipe_buf_nosteal, | |
568 | .get = generic_pipe_buf_get, | |
569 | }; | |
570 | EXPORT_SYMBOL(nosteal_pipe_buf_ops); | |
571 | ||
6818173b MS |
572 | static ssize_t kernel_readv(struct file *file, const struct iovec *vec, |
573 | unsigned long vlen, loff_t offset) | |
574 | { | |
575 | mm_segment_t old_fs; | |
576 | loff_t pos = offset; | |
577 | ssize_t res; | |
578 | ||
579 | old_fs = get_fs(); | |
580 | set_fs(get_ds()); | |
581 | /* The cast to a user pointer is valid due to the set_fs() */ | |
582 | res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos); | |
583 | set_fs(old_fs); | |
584 | ||
585 | return res; | |
586 | } | |
587 | ||
7bb307e8 | 588 | ssize_t kernel_write(struct file *file, const char *buf, size_t count, |
b2858d7d | 589 | loff_t pos) |
0b0a47f5 MS |
590 | { |
591 | mm_segment_t old_fs; | |
592 | ssize_t res; | |
593 | ||
594 | old_fs = get_fs(); | |
595 | set_fs(get_ds()); | |
596 | /* The cast to a user pointer is valid due to the set_fs() */ | |
7bb307e8 | 597 | res = vfs_write(file, (__force const char __user *)buf, count, &pos); |
0b0a47f5 MS |
598 | set_fs(old_fs); |
599 | ||
600 | return res; | |
601 | } | |
7bb307e8 | 602 | EXPORT_SYMBOL(kernel_write); |
0b0a47f5 | 603 | |
6818173b MS |
604 | ssize_t default_file_splice_read(struct file *in, loff_t *ppos, |
605 | struct pipe_inode_info *pipe, size_t len, | |
606 | unsigned int flags) | |
607 | { | |
608 | unsigned int nr_pages; | |
609 | unsigned int nr_freed; | |
610 | size_t offset; | |
35f3d14d JA |
611 | struct page *pages[PIPE_DEF_BUFFERS]; |
612 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
613 | struct iovec *vec, __vec[PIPE_DEF_BUFFERS]; | |
6818173b MS |
614 | ssize_t res; |
615 | size_t this_len; | |
616 | int error; | |
617 | int i; | |
618 | struct splice_pipe_desc spd = { | |
619 | .pages = pages, | |
620 | .partial = partial, | |
047fe360 | 621 | .nr_pages_max = PIPE_DEF_BUFFERS, |
6818173b MS |
622 | .flags = flags, |
623 | .ops = &default_pipe_buf_ops, | |
624 | .spd_release = spd_release_page, | |
625 | }; | |
626 | ||
35f3d14d JA |
627 | if (splice_grow_spd(pipe, &spd)) |
628 | return -ENOMEM; | |
629 | ||
630 | res = -ENOMEM; | |
631 | vec = __vec; | |
047fe360 ED |
632 | if (spd.nr_pages_max > PIPE_DEF_BUFFERS) { |
633 | vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL); | |
35f3d14d JA |
634 | if (!vec) |
635 | goto shrink_ret; | |
636 | } | |
637 | ||
6818173b MS |
638 | offset = *ppos & ~PAGE_CACHE_MASK; |
639 | nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
640 | ||
047fe360 | 641 | for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) { |
6818173b MS |
642 | struct page *page; |
643 | ||
4f231228 | 644 | page = alloc_page(GFP_USER); |
6818173b MS |
645 | error = -ENOMEM; |
646 | if (!page) | |
647 | goto err; | |
648 | ||
649 | this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset); | |
4f231228 | 650 | vec[i].iov_base = (void __user *) page_address(page); |
6818173b | 651 | vec[i].iov_len = this_len; |
35f3d14d | 652 | spd.pages[i] = page; |
6818173b MS |
653 | spd.nr_pages++; |
654 | len -= this_len; | |
655 | offset = 0; | |
656 | } | |
657 | ||
658 | res = kernel_readv(in, vec, spd.nr_pages, *ppos); | |
77f6bf57 AM |
659 | if (res < 0) { |
660 | error = res; | |
6818173b | 661 | goto err; |
77f6bf57 | 662 | } |
6818173b MS |
663 | |
664 | error = 0; | |
665 | if (!res) | |
666 | goto err; | |
667 | ||
668 | nr_freed = 0; | |
669 | for (i = 0; i < spd.nr_pages; i++) { | |
6818173b | 670 | this_len = min_t(size_t, vec[i].iov_len, res); |
35f3d14d JA |
671 | spd.partial[i].offset = 0; |
672 | spd.partial[i].len = this_len; | |
6818173b | 673 | if (!this_len) { |
35f3d14d JA |
674 | __free_page(spd.pages[i]); |
675 | spd.pages[i] = NULL; | |
6818173b MS |
676 | nr_freed++; |
677 | } | |
678 | res -= this_len; | |
679 | } | |
680 | spd.nr_pages -= nr_freed; | |
681 | ||
682 | res = splice_to_pipe(pipe, &spd); | |
683 | if (res > 0) | |
684 | *ppos += res; | |
685 | ||
35f3d14d JA |
686 | shrink_ret: |
687 | if (vec != __vec) | |
688 | kfree(vec); | |
047fe360 | 689 | splice_shrink_spd(&spd); |
6818173b MS |
690 | return res; |
691 | ||
692 | err: | |
4f231228 | 693 | for (i = 0; i < spd.nr_pages; i++) |
35f3d14d | 694 | __free_page(spd.pages[i]); |
4f231228 | 695 | |
35f3d14d JA |
696 | res = error; |
697 | goto shrink_ret; | |
6818173b MS |
698 | } |
699 | EXPORT_SYMBOL(default_file_splice_read); | |
700 | ||
5274f052 | 701 | /* |
4f6f0bd2 | 702 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 703 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 704 | */ |
76ad4d11 | 705 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
706 | struct pipe_buffer *buf, struct splice_desc *sd) |
707 | { | |
6a14b90b | 708 | struct file *file = sd->u.file; |
5274f052 | 709 | loff_t pos = sd->pos; |
a8adbe37 | 710 | int more; |
5274f052 | 711 | |
72c2d531 | 712 | if (!likely(file->f_op->sendpage)) |
a8adbe37 MM |
713 | return -EINVAL; |
714 | ||
35f9c09f | 715 | more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; |
ae62ca7b ED |
716 | |
717 | if (sd->len < sd->total_len && pipe->nrbufs > 1) | |
35f9c09f | 718 | more |= MSG_SENDPAGE_NOTLAST; |
ae62ca7b | 719 | |
a8adbe37 MM |
720 | return file->f_op->sendpage(file, buf->page, buf->offset, |
721 | sd->len, &pos, more); | |
5274f052 JA |
722 | } |
723 | ||
b3c2d2dd MS |
724 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
725 | { | |
726 | smp_mb(); | |
727 | if (waitqueue_active(&pipe->wait)) | |
728 | wake_up_interruptible(&pipe->wait); | |
729 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
730 | } | |
731 | ||
932cc6d4 | 732 | /** |
b3c2d2dd | 733 | * splice_from_pipe_feed - feed available data from a pipe to a file |
932cc6d4 JA |
734 | * @pipe: pipe to splice from |
735 | * @sd: information to @actor | |
736 | * @actor: handler that splices the data | |
737 | * | |
738 | * Description: | |
b3c2d2dd MS |
739 | * This function loops over the pipe and calls @actor to do the |
740 | * actual moving of a single struct pipe_buffer to the desired | |
741 | * destination. It returns when there's no more buffers left in | |
742 | * the pipe or if the requested number of bytes (@sd->total_len) | |
743 | * have been copied. It returns a positive number (one) if the | |
744 | * pipe needs to be filled with more data, zero if the required | |
745 | * number of bytes have been copied and -errno on error. | |
932cc6d4 | 746 | * |
b3c2d2dd MS |
747 | * This, together with splice_from_pipe_{begin,end,next}, may be |
748 | * used to implement the functionality of __splice_from_pipe() when | |
749 | * locking is required around copying the pipe buffers to the | |
750 | * destination. | |
83f9135b | 751 | */ |
96f9bc8f | 752 | static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
b3c2d2dd | 753 | splice_actor *actor) |
5274f052 | 754 | { |
b3c2d2dd | 755 | int ret; |
5274f052 | 756 | |
b3c2d2dd MS |
757 | while (pipe->nrbufs) { |
758 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
759 | const struct pipe_buf_operations *ops = buf->ops; | |
5274f052 | 760 | |
b3c2d2dd MS |
761 | sd->len = buf->len; |
762 | if (sd->len > sd->total_len) | |
763 | sd->len = sd->total_len; | |
5274f052 | 764 | |
a8adbe37 MM |
765 | ret = buf->ops->confirm(pipe, buf); |
766 | if (unlikely(ret)) { | |
b3c2d2dd MS |
767 | if (ret == -ENODATA) |
768 | ret = 0; | |
769 | return ret; | |
770 | } | |
a8adbe37 MM |
771 | |
772 | ret = actor(pipe, buf, sd); | |
773 | if (ret <= 0) | |
774 | return ret; | |
775 | ||
b3c2d2dd MS |
776 | buf->offset += ret; |
777 | buf->len -= ret; | |
778 | ||
779 | sd->num_spliced += ret; | |
780 | sd->len -= ret; | |
781 | sd->pos += ret; | |
782 | sd->total_len -= ret; | |
783 | ||
784 | if (!buf->len) { | |
785 | buf->ops = NULL; | |
786 | ops->release(pipe, buf); | |
35f3d14d | 787 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); |
b3c2d2dd | 788 | pipe->nrbufs--; |
6447a3cf | 789 | if (pipe->files) |
b3c2d2dd MS |
790 | sd->need_wakeup = true; |
791 | } | |
5274f052 | 792 | |
b3c2d2dd MS |
793 | if (!sd->total_len) |
794 | return 0; | |
795 | } | |
5274f052 | 796 | |
b3c2d2dd MS |
797 | return 1; |
798 | } | |
5274f052 | 799 | |
b3c2d2dd MS |
800 | /** |
801 | * splice_from_pipe_next - wait for some data to splice from | |
802 | * @pipe: pipe to splice from | |
803 | * @sd: information about the splice operation | |
804 | * | |
805 | * Description: | |
806 | * This function will wait for some data and return a positive | |
807 | * value (one) if pipe buffers are available. It will return zero | |
808 | * or -errno if no more data needs to be spliced. | |
809 | */ | |
96f9bc8f | 810 | static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd | 811 | { |
c725bfce JK |
812 | /* |
813 | * Check for signal early to make process killable when there are | |
814 | * always buffers available | |
815 | */ | |
816 | if (signal_pending(current)) | |
817 | return -ERESTARTSYS; | |
818 | ||
b3c2d2dd MS |
819 | while (!pipe->nrbufs) { |
820 | if (!pipe->writers) | |
821 | return 0; | |
016b661e | 822 | |
b3c2d2dd MS |
823 | if (!pipe->waiting_writers && sd->num_spliced) |
824 | return 0; | |
73d62d83 | 825 | |
b3c2d2dd MS |
826 | if (sd->flags & SPLICE_F_NONBLOCK) |
827 | return -EAGAIN; | |
5274f052 | 828 | |
b3c2d2dd MS |
829 | if (signal_pending(current)) |
830 | return -ERESTARTSYS; | |
5274f052 | 831 | |
b3c2d2dd MS |
832 | if (sd->need_wakeup) { |
833 | wakeup_pipe_writers(pipe); | |
834 | sd->need_wakeup = false; | |
5274f052 JA |
835 | } |
836 | ||
b3c2d2dd MS |
837 | pipe_wait(pipe); |
838 | } | |
29e35094 | 839 | |
b3c2d2dd MS |
840 | return 1; |
841 | } | |
5274f052 | 842 | |
b3c2d2dd MS |
843 | /** |
844 | * splice_from_pipe_begin - start splicing from pipe | |
b80901bb | 845 | * @sd: information about the splice operation |
b3c2d2dd MS |
846 | * |
847 | * Description: | |
848 | * This function should be called before a loop containing | |
849 | * splice_from_pipe_next() and splice_from_pipe_feed() to | |
850 | * initialize the necessary fields of @sd. | |
851 | */ | |
96f9bc8f | 852 | static void splice_from_pipe_begin(struct splice_desc *sd) |
b3c2d2dd MS |
853 | { |
854 | sd->num_spliced = 0; | |
855 | sd->need_wakeup = false; | |
856 | } | |
5274f052 | 857 | |
b3c2d2dd MS |
858 | /** |
859 | * splice_from_pipe_end - finish splicing from pipe | |
860 | * @pipe: pipe to splice from | |
861 | * @sd: information about the splice operation | |
862 | * | |
863 | * Description: | |
864 | * This function will wake up pipe writers if necessary. It should | |
865 | * be called after a loop containing splice_from_pipe_next() and | |
866 | * splice_from_pipe_feed(). | |
867 | */ | |
96f9bc8f | 868 | static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd MS |
869 | { |
870 | if (sd->need_wakeup) | |
871 | wakeup_pipe_writers(pipe); | |
872 | } | |
5274f052 | 873 | |
b3c2d2dd MS |
874 | /** |
875 | * __splice_from_pipe - splice data from a pipe to given actor | |
876 | * @pipe: pipe to splice from | |
877 | * @sd: information to @actor | |
878 | * @actor: handler that splices the data | |
879 | * | |
880 | * Description: | |
881 | * This function does little more than loop over the pipe and call | |
882 | * @actor to do the actual moving of a single struct pipe_buffer to | |
883 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
884 | * pipe_to_user. | |
885 | * | |
886 | */ | |
887 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, | |
888 | splice_actor *actor) | |
889 | { | |
890 | int ret; | |
5274f052 | 891 | |
b3c2d2dd MS |
892 | splice_from_pipe_begin(sd); |
893 | do { | |
894 | ret = splice_from_pipe_next(pipe, sd); | |
895 | if (ret > 0) | |
896 | ret = splice_from_pipe_feed(pipe, sd, actor); | |
897 | } while (ret > 0); | |
898 | splice_from_pipe_end(pipe, sd); | |
899 | ||
900 | return sd->num_spliced ? sd->num_spliced : ret; | |
5274f052 | 901 | } |
40bee44e | 902 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 903 | |
932cc6d4 JA |
904 | /** |
905 | * splice_from_pipe - splice data from a pipe to a file | |
906 | * @pipe: pipe to splice from | |
907 | * @out: file to splice to | |
908 | * @ppos: position in @out | |
909 | * @len: how many bytes to splice | |
910 | * @flags: splice modifier flags | |
911 | * @actor: handler that splices the data | |
912 | * | |
913 | * Description: | |
2933970b | 914 | * See __splice_from_pipe. This function locks the pipe inode, |
932cc6d4 JA |
915 | * otherwise it's identical to __splice_from_pipe(). |
916 | * | |
917 | */ | |
6da61809 MF |
918 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
919 | loff_t *ppos, size_t len, unsigned int flags, | |
920 | splice_actor *actor) | |
921 | { | |
922 | ssize_t ret; | |
c66ab6fa JA |
923 | struct splice_desc sd = { |
924 | .total_len = len, | |
925 | .flags = flags, | |
926 | .pos = *ppos, | |
6a14b90b | 927 | .u.file = out, |
c66ab6fa | 928 | }; |
6da61809 | 929 | |
61e0d47c | 930 | pipe_lock(pipe); |
c66ab6fa | 931 | ret = __splice_from_pipe(pipe, &sd, actor); |
61e0d47c | 932 | pipe_unlock(pipe); |
6da61809 MF |
933 | |
934 | return ret; | |
935 | } | |
936 | ||
8d020765 AV |
937 | /** |
938 | * iter_file_splice_write - splice data from a pipe to a file | |
939 | * @pipe: pipe info | |
940 | * @out: file to write to | |
941 | * @ppos: position in @out | |
942 | * @len: number of bytes to splice | |
943 | * @flags: splice modifier flags | |
944 | * | |
945 | * Description: | |
946 | * Will either move or copy pages (determined by @flags options) from | |
947 | * the given pipe inode to the given file. | |
948 | * This one is ->write_iter-based. | |
949 | * | |
950 | */ | |
951 | ssize_t | |
952 | iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
953 | loff_t *ppos, size_t len, unsigned int flags) | |
954 | { | |
955 | struct splice_desc sd = { | |
956 | .total_len = len, | |
957 | .flags = flags, | |
958 | .pos = *ppos, | |
959 | .u.file = out, | |
960 | }; | |
961 | int nbufs = pipe->buffers; | |
962 | struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), | |
963 | GFP_KERNEL); | |
964 | ssize_t ret; | |
965 | ||
966 | if (unlikely(!array)) | |
967 | return -ENOMEM; | |
968 | ||
969 | pipe_lock(pipe); | |
970 | ||
971 | splice_from_pipe_begin(&sd); | |
972 | while (sd.total_len) { | |
973 | struct iov_iter from; | |
8d020765 AV |
974 | size_t left; |
975 | int n, idx; | |
976 | ||
977 | ret = splice_from_pipe_next(pipe, &sd); | |
978 | if (ret <= 0) | |
979 | break; | |
980 | ||
981 | if (unlikely(nbufs < pipe->buffers)) { | |
982 | kfree(array); | |
983 | nbufs = pipe->buffers; | |
984 | array = kcalloc(nbufs, sizeof(struct bio_vec), | |
985 | GFP_KERNEL); | |
986 | if (!array) { | |
987 | ret = -ENOMEM; | |
988 | break; | |
989 | } | |
990 | } | |
991 | ||
992 | /* build the vector */ | |
993 | left = sd.total_len; | |
994 | for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) { | |
995 | struct pipe_buffer *buf = pipe->bufs + idx; | |
996 | size_t this_len = buf->len; | |
997 | ||
998 | if (this_len > left) | |
999 | this_len = left; | |
1000 | ||
1001 | if (idx == pipe->buffers - 1) | |
1002 | idx = -1; | |
1003 | ||
1004 | ret = buf->ops->confirm(pipe, buf); | |
1005 | if (unlikely(ret)) { | |
1006 | if (ret == -ENODATA) | |
1007 | ret = 0; | |
1008 | goto done; | |
1009 | } | |
1010 | ||
1011 | array[n].bv_page = buf->page; | |
1012 | array[n].bv_len = this_len; | |
1013 | array[n].bv_offset = buf->offset; | |
1014 | left -= this_len; | |
1015 | } | |
1016 | ||
05afcb77 AV |
1017 | iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n, |
1018 | sd.total_len - left); | |
dbe4e192 | 1019 | ret = vfs_iter_write(out, &from, &sd.pos); |
8d020765 AV |
1020 | if (ret <= 0) |
1021 | break; | |
1022 | ||
1023 | sd.num_spliced += ret; | |
1024 | sd.total_len -= ret; | |
dbe4e192 | 1025 | *ppos = sd.pos; |
8d020765 AV |
1026 | |
1027 | /* dismiss the fully eaten buffers, adjust the partial one */ | |
1028 | while (ret) { | |
1029 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
1030 | if (ret >= buf->len) { | |
1031 | const struct pipe_buf_operations *ops = buf->ops; | |
1032 | ret -= buf->len; | |
1033 | buf->len = 0; | |
1034 | buf->ops = NULL; | |
1035 | ops->release(pipe, buf); | |
1036 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); | |
1037 | pipe->nrbufs--; | |
1038 | if (pipe->files) | |
1039 | sd.need_wakeup = true; | |
1040 | } else { | |
1041 | buf->offset += ret; | |
1042 | buf->len -= ret; | |
1043 | ret = 0; | |
1044 | } | |
1045 | } | |
1046 | } | |
1047 | done: | |
1048 | kfree(array); | |
1049 | splice_from_pipe_end(pipe, &sd); | |
1050 | ||
1051 | pipe_unlock(pipe); | |
1052 | ||
1053 | if (sd.num_spliced) | |
1054 | ret = sd.num_spliced; | |
1055 | ||
1056 | return ret; | |
1057 | } | |
1058 | ||
1059 | EXPORT_SYMBOL(iter_file_splice_write); | |
1060 | ||
b2858d7d MS |
1061 | static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1062 | struct splice_desc *sd) | |
0b0a47f5 | 1063 | { |
b2858d7d MS |
1064 | int ret; |
1065 | void *data; | |
06ae43f3 | 1066 | loff_t tmp = sd->pos; |
b2858d7d | 1067 | |
fbb32750 | 1068 | data = kmap(buf->page); |
06ae43f3 | 1069 | ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); |
fbb32750 | 1070 | kunmap(buf->page); |
b2858d7d MS |
1071 | |
1072 | return ret; | |
0b0a47f5 MS |
1073 | } |
1074 | ||
1075 | static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, | |
1076 | struct file *out, loff_t *ppos, | |
1077 | size_t len, unsigned int flags) | |
1078 | { | |
b2858d7d | 1079 | ssize_t ret; |
0b0a47f5 | 1080 | |
b2858d7d MS |
1081 | ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); |
1082 | if (ret > 0) | |
1083 | *ppos += ret; | |
0b0a47f5 | 1084 | |
b2858d7d | 1085 | return ret; |
0b0a47f5 MS |
1086 | } |
1087 | ||
83f9135b JA |
1088 | /** |
1089 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 1090 | * @pipe: pipe to splice from |
83f9135b | 1091 | * @out: socket to write to |
932cc6d4 | 1092 | * @ppos: position in @out |
83f9135b JA |
1093 | * @len: number of bytes to splice |
1094 | * @flags: splice modifier flags | |
1095 | * | |
932cc6d4 JA |
1096 | * Description: |
1097 | * Will send @len bytes from the pipe to a network socket. No data copying | |
1098 | * is involved. | |
83f9135b JA |
1099 | * |
1100 | */ | |
3a326a2c | 1101 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1102 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1103 | { |
00522fb4 | 1104 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
1105 | } |
1106 | ||
059a8f37 | 1107 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 1108 | |
83f9135b JA |
1109 | /* |
1110 | * Attempt to initiate a splice from pipe to file. | |
1111 | */ | |
3a326a2c | 1112 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1113 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1114 | { |
0b0a47f5 MS |
1115 | ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, |
1116 | loff_t *, size_t, unsigned int); | |
5274f052 | 1117 | |
72c2d531 | 1118 | if (out->f_op->splice_write) |
cc56f7de CG |
1119 | splice_write = out->f_op->splice_write; |
1120 | else | |
0b0a47f5 MS |
1121 | splice_write = default_file_splice_write; |
1122 | ||
500368f7 | 1123 | return splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
1124 | } |
1125 | ||
83f9135b JA |
1126 | /* |
1127 | * Attempt to initiate a splice from a file to a pipe. | |
1128 | */ | |
cbb7e577 JA |
1129 | static long do_splice_to(struct file *in, loff_t *ppos, |
1130 | struct pipe_inode_info *pipe, size_t len, | |
1131 | unsigned int flags) | |
5274f052 | 1132 | { |
6818173b MS |
1133 | ssize_t (*splice_read)(struct file *, loff_t *, |
1134 | struct pipe_inode_info *, size_t, unsigned int); | |
5274f052 JA |
1135 | int ret; |
1136 | ||
49570e9b | 1137 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
1138 | return -EBADF; |
1139 | ||
cbb7e577 | 1140 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
1141 | if (unlikely(ret < 0)) |
1142 | return ret; | |
1143 | ||
72c2d531 | 1144 | if (in->f_op->splice_read) |
cc56f7de CG |
1145 | splice_read = in->f_op->splice_read; |
1146 | else | |
6818173b MS |
1147 | splice_read = default_file_splice_read; |
1148 | ||
1149 | return splice_read(in, ppos, pipe, len, flags); | |
5274f052 JA |
1150 | } |
1151 | ||
932cc6d4 JA |
1152 | /** |
1153 | * splice_direct_to_actor - splices data directly between two non-pipes | |
1154 | * @in: file to splice from | |
1155 | * @sd: actor information on where to splice to | |
1156 | * @actor: handles the data splicing | |
1157 | * | |
1158 | * Description: | |
1159 | * This is a special case helper to splice directly between two | |
1160 | * points, without requiring an explicit pipe. Internally an allocated | |
79685b8d | 1161 | * pipe is cached in the process, and reused during the lifetime of |
932cc6d4 JA |
1162 | * that process. |
1163 | * | |
c66ab6fa JA |
1164 | */ |
1165 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1166 | splice_direct_actor *actor) | |
b92ce558 JA |
1167 | { |
1168 | struct pipe_inode_info *pipe; | |
1169 | long ret, bytes; | |
1170 | umode_t i_mode; | |
c66ab6fa | 1171 | size_t len; |
0ff28d9f | 1172 | int i, flags, more; |
b92ce558 JA |
1173 | |
1174 | /* | |
1175 | * We require the input being a regular file, as we don't want to | |
1176 | * randomly drop data for eg socket -> socket splicing. Use the | |
1177 | * piped splicing for that! | |
1178 | */ | |
496ad9aa | 1179 | i_mode = file_inode(in)->i_mode; |
b92ce558 JA |
1180 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1181 | return -EINVAL; | |
1182 | ||
1183 | /* | |
1184 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1185 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1186 | */ | |
1187 | pipe = current->splice_pipe; | |
49570e9b | 1188 | if (unlikely(!pipe)) { |
7bee130e | 1189 | pipe = alloc_pipe_info(); |
b92ce558 JA |
1190 | if (!pipe) |
1191 | return -ENOMEM; | |
1192 | ||
1193 | /* | |
1194 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1195 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1196 | * PIPE_READERS appropriately. |
1197 | */ | |
1198 | pipe->readers = 1; | |
1199 | ||
1200 | current->splice_pipe = pipe; | |
1201 | } | |
1202 | ||
1203 | /* | |
73d62d83 | 1204 | * Do the splice. |
b92ce558 JA |
1205 | */ |
1206 | ret = 0; | |
1207 | bytes = 0; | |
c66ab6fa JA |
1208 | len = sd->total_len; |
1209 | flags = sd->flags; | |
1210 | ||
1211 | /* | |
1212 | * Don't block on output, we have to drain the direct pipe. | |
1213 | */ | |
1214 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
0ff28d9f | 1215 | more = sd->flags & SPLICE_F_MORE; |
b92ce558 JA |
1216 | |
1217 | while (len) { | |
51a92c0f | 1218 | size_t read_len; |
a82c53a0 | 1219 | loff_t pos = sd->pos, prev_pos = pos; |
b92ce558 | 1220 | |
bcd4f3ac | 1221 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1222 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1223 | goto out_release; |
1224 | ||
1225 | read_len = ret; | |
c66ab6fa | 1226 | sd->total_len = read_len; |
b92ce558 | 1227 | |
0ff28d9f CL |
1228 | /* |
1229 | * If more data is pending, set SPLICE_F_MORE | |
1230 | * If this is the last data and SPLICE_F_MORE was not set | |
1231 | * initially, clears it. | |
1232 | */ | |
1233 | if (read_len < len) | |
1234 | sd->flags |= SPLICE_F_MORE; | |
1235 | else if (!more) | |
1236 | sd->flags &= ~SPLICE_F_MORE; | |
b92ce558 JA |
1237 | /* |
1238 | * NOTE: nonblocking mode only applies to the input. We | |
1239 | * must not do the output in nonblocking mode as then we | |
1240 | * could get stuck data in the internal pipe: | |
1241 | */ | |
c66ab6fa | 1242 | ret = actor(pipe, sd); |
a82c53a0 TZ |
1243 | if (unlikely(ret <= 0)) { |
1244 | sd->pos = prev_pos; | |
b92ce558 | 1245 | goto out_release; |
a82c53a0 | 1246 | } |
b92ce558 JA |
1247 | |
1248 | bytes += ret; | |
1249 | len -= ret; | |
bcd4f3ac | 1250 | sd->pos = pos; |
b92ce558 | 1251 | |
a82c53a0 TZ |
1252 | if (ret < read_len) { |
1253 | sd->pos = prev_pos + ret; | |
51a92c0f | 1254 | goto out_release; |
a82c53a0 | 1255 | } |
b92ce558 JA |
1256 | } |
1257 | ||
9e97198d | 1258 | done: |
b92ce558 | 1259 | pipe->nrbufs = pipe->curbuf = 0; |
80848708 | 1260 | file_accessed(in); |
b92ce558 JA |
1261 | return bytes; |
1262 | ||
1263 | out_release: | |
1264 | /* | |
1265 | * If we did an incomplete transfer we must release | |
1266 | * the pipe buffers in question: | |
1267 | */ | |
35f3d14d | 1268 | for (i = 0; i < pipe->buffers; i++) { |
b92ce558 JA |
1269 | struct pipe_buffer *buf = pipe->bufs + i; |
1270 | ||
1271 | if (buf->ops) { | |
1272 | buf->ops->release(pipe, buf); | |
1273 | buf->ops = NULL; | |
1274 | } | |
1275 | } | |
b92ce558 | 1276 | |
9e97198d JA |
1277 | if (!bytes) |
1278 | bytes = ret; | |
c66ab6fa | 1279 | |
9e97198d | 1280 | goto done; |
c66ab6fa JA |
1281 | } |
1282 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1283 | ||
1284 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1285 | struct splice_desc *sd) | |
1286 | { | |
6a14b90b | 1287 | struct file *file = sd->u.file; |
c66ab6fa | 1288 | |
7995bd28 | 1289 | return do_splice_from(pipe, file, sd->opos, sd->total_len, |
2cb4b05e | 1290 | sd->flags); |
c66ab6fa JA |
1291 | } |
1292 | ||
932cc6d4 JA |
1293 | /** |
1294 | * do_splice_direct - splices data directly between two files | |
1295 | * @in: file to splice from | |
1296 | * @ppos: input file offset | |
1297 | * @out: file to splice to | |
acdb37c3 | 1298 | * @opos: output file offset |
932cc6d4 JA |
1299 | * @len: number of bytes to splice |
1300 | * @flags: splice modifier flags | |
1301 | * | |
1302 | * Description: | |
1303 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1304 | * doing it in the application would incur an extra system call | |
1305 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1306 | * can splice directly through a process-private pipe. | |
1307 | * | |
1308 | */ | |
c66ab6fa | 1309 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
7995bd28 | 1310 | loff_t *opos, size_t len, unsigned int flags) |
c66ab6fa JA |
1311 | { |
1312 | struct splice_desc sd = { | |
1313 | .len = len, | |
1314 | .total_len = len, | |
1315 | .flags = flags, | |
1316 | .pos = *ppos, | |
6a14b90b | 1317 | .u.file = out, |
7995bd28 | 1318 | .opos = opos, |
c66ab6fa | 1319 | }; |
51a92c0f | 1320 | long ret; |
c66ab6fa | 1321 | |
18c67cb9 AV |
1322 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1323 | return -EBADF; | |
1324 | ||
1325 | if (unlikely(out->f_flags & O_APPEND)) | |
1326 | return -EINVAL; | |
1327 | ||
1328 | ret = rw_verify_area(WRITE, out, opos, len); | |
1329 | if (unlikely(ret < 0)) | |
1330 | return ret; | |
1331 | ||
c66ab6fa | 1332 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); |
51a92c0f | 1333 | if (ret > 0) |
a82c53a0 | 1334 | *ppos = sd.pos; |
51a92c0f | 1335 | |
c66ab6fa | 1336 | return ret; |
b92ce558 | 1337 | } |
1c118596 | 1338 | EXPORT_SYMBOL(do_splice_direct); |
b92ce558 | 1339 | |
7c77f0b3 MS |
1340 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
1341 | struct pipe_inode_info *opipe, | |
1342 | size_t len, unsigned int flags); | |
ddac0d39 | 1343 | |
83f9135b JA |
1344 | /* |
1345 | * Determine where to splice to/from. | |
1346 | */ | |
529565dc IM |
1347 | static long do_splice(struct file *in, loff_t __user *off_in, |
1348 | struct file *out, loff_t __user *off_out, | |
1349 | size_t len, unsigned int flags) | |
5274f052 | 1350 | { |
7c77f0b3 MS |
1351 | struct pipe_inode_info *ipipe; |
1352 | struct pipe_inode_info *opipe; | |
7995bd28 | 1353 | loff_t offset; |
a4514ebd | 1354 | long ret; |
5274f052 | 1355 | |
71993e62 LT |
1356 | ipipe = get_pipe_info(in); |
1357 | opipe = get_pipe_info(out); | |
7c77f0b3 MS |
1358 | |
1359 | if (ipipe && opipe) { | |
1360 | if (off_in || off_out) | |
1361 | return -ESPIPE; | |
1362 | ||
1363 | if (!(in->f_mode & FMODE_READ)) | |
1364 | return -EBADF; | |
1365 | ||
1366 | if (!(out->f_mode & FMODE_WRITE)) | |
1367 | return -EBADF; | |
1368 | ||
1369 | /* Splicing to self would be fun, but... */ | |
1370 | if (ipipe == opipe) | |
1371 | return -EINVAL; | |
1372 | ||
1373 | return splice_pipe_to_pipe(ipipe, opipe, len, flags); | |
1374 | } | |
1375 | ||
1376 | if (ipipe) { | |
529565dc IM |
1377 | if (off_in) |
1378 | return -ESPIPE; | |
b92ce558 | 1379 | if (off_out) { |
19c9a49b | 1380 | if (!(out->f_mode & FMODE_PWRITE)) |
b92ce558 | 1381 | return -EINVAL; |
cbb7e577 | 1382 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1383 | return -EFAULT; |
7995bd28 AV |
1384 | } else { |
1385 | offset = out->f_pos; | |
1386 | } | |
529565dc | 1387 | |
18c67cb9 AV |
1388 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1389 | return -EBADF; | |
1390 | ||
1391 | if (unlikely(out->f_flags & O_APPEND)) | |
1392 | return -EINVAL; | |
1393 | ||
1394 | ret = rw_verify_area(WRITE, out, &offset, len); | |
1395 | if (unlikely(ret < 0)) | |
1396 | return ret; | |
1397 | ||
500368f7 | 1398 | file_start_write(out); |
7995bd28 | 1399 | ret = do_splice_from(ipipe, out, &offset, len, flags); |
500368f7 | 1400 | file_end_write(out); |
a4514ebd | 1401 | |
7995bd28 AV |
1402 | if (!off_out) |
1403 | out->f_pos = offset; | |
1404 | else if (copy_to_user(off_out, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1405 | ret = -EFAULT; |
1406 | ||
1407 | return ret; | |
529565dc | 1408 | } |
5274f052 | 1409 | |
7c77f0b3 | 1410 | if (opipe) { |
529565dc IM |
1411 | if (off_out) |
1412 | return -ESPIPE; | |
b92ce558 | 1413 | if (off_in) { |
19c9a49b | 1414 | if (!(in->f_mode & FMODE_PREAD)) |
b92ce558 | 1415 | return -EINVAL; |
cbb7e577 | 1416 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1417 | return -EFAULT; |
7995bd28 AV |
1418 | } else { |
1419 | offset = in->f_pos; | |
1420 | } | |
529565dc | 1421 | |
7995bd28 | 1422 | ret = do_splice_to(in, &offset, opipe, len, flags); |
a4514ebd | 1423 | |
7995bd28 AV |
1424 | if (!off_in) |
1425 | in->f_pos = offset; | |
1426 | else if (copy_to_user(off_in, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1427 | ret = -EFAULT; |
1428 | ||
1429 | return ret; | |
529565dc | 1430 | } |
5274f052 JA |
1431 | |
1432 | return -EINVAL; | |
1433 | } | |
1434 | ||
912d35f8 JA |
1435 | /* |
1436 | * Map an iov into an array of pages and offset/length tupples. With the | |
1437 | * partial_page structure, we can map several non-contiguous ranges into | |
1438 | * our ones pages[] map instead of splitting that operation into pieces. | |
1439 | * Could easily be exported as a generic helper for other users, in which | |
1440 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1441 | */ | |
1442 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1443 | unsigned int nr_vecs, struct page **pages, | |
bd1a68b5 | 1444 | struct partial_page *partial, bool aligned, |
35f3d14d | 1445 | unsigned int pipe_buffers) |
912d35f8 JA |
1446 | { |
1447 | int buffers = 0, error = 0; | |
1448 | ||
912d35f8 JA |
1449 | while (nr_vecs) { |
1450 | unsigned long off, npages; | |
75723957 | 1451 | struct iovec entry; |
912d35f8 JA |
1452 | void __user *base; |
1453 | size_t len; | |
1454 | int i; | |
1455 | ||
75723957 | 1456 | error = -EFAULT; |
bc40d73c | 1457 | if (copy_from_user(&entry, iov, sizeof(entry))) |
912d35f8 JA |
1458 | break; |
1459 | ||
75723957 LT |
1460 | base = entry.iov_base; |
1461 | len = entry.iov_len; | |
1462 | ||
912d35f8 JA |
1463 | /* |
1464 | * Sanity check this iovec. 0 read succeeds. | |
1465 | */ | |
75723957 | 1466 | error = 0; |
912d35f8 JA |
1467 | if (unlikely(!len)) |
1468 | break; | |
1469 | error = -EFAULT; | |
712a30e6 | 1470 | if (!access_ok(VERIFY_READ, base, len)) |
912d35f8 JA |
1471 | break; |
1472 | ||
1473 | /* | |
1474 | * Get this base offset and number of pages, then map | |
1475 | * in the user pages. | |
1476 | */ | |
1477 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1478 | |
1479 | /* | |
1480 | * If asked for alignment, the offset must be zero and the | |
1481 | * length a multiple of the PAGE_SIZE. | |
1482 | */ | |
1483 | error = -EINVAL; | |
1484 | if (aligned && (off || len & ~PAGE_MASK)) | |
1485 | break; | |
1486 | ||
912d35f8 | 1487 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
35f3d14d JA |
1488 | if (npages > pipe_buffers - buffers) |
1489 | npages = pipe_buffers - buffers; | |
912d35f8 | 1490 | |
bc40d73c NP |
1491 | error = get_user_pages_fast((unsigned long)base, npages, |
1492 | 0, &pages[buffers]); | |
912d35f8 JA |
1493 | |
1494 | if (unlikely(error <= 0)) | |
1495 | break; | |
1496 | ||
1497 | /* | |
1498 | * Fill this contiguous range into the partial page map. | |
1499 | */ | |
1500 | for (i = 0; i < error; i++) { | |
7591489a | 1501 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1502 | |
1503 | partial[buffers].offset = off; | |
1504 | partial[buffers].len = plen; | |
1505 | ||
1506 | off = 0; | |
1507 | len -= plen; | |
1508 | buffers++; | |
1509 | } | |
1510 | ||
1511 | /* | |
1512 | * We didn't complete this iov, stop here since it probably | |
1513 | * means we have to move some of this into a pipe to | |
1514 | * be able to continue. | |
1515 | */ | |
1516 | if (len) | |
1517 | break; | |
1518 | ||
1519 | /* | |
1520 | * Don't continue if we mapped fewer pages than we asked for, | |
1521 | * or if we mapped the max number of pages that we have | |
1522 | * room for. | |
1523 | */ | |
35f3d14d | 1524 | if (error < npages || buffers == pipe_buffers) |
912d35f8 JA |
1525 | break; |
1526 | ||
1527 | nr_vecs--; | |
1528 | iov++; | |
1529 | } | |
1530 | ||
912d35f8 JA |
1531 | if (buffers) |
1532 | return buffers; | |
1533 | ||
1534 | return error; | |
1535 | } | |
1536 | ||
6a14b90b JA |
1537 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1538 | struct splice_desc *sd) | |
1539 | { | |
6130f531 AV |
1540 | int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); |
1541 | return n == sd->len ? n : -EFAULT; | |
6a14b90b JA |
1542 | } |
1543 | ||
1544 | /* | |
1545 | * For lack of a better implementation, implement vmsplice() to userspace | |
1546 | * as a simple copy of the pipes pages to the user iov. | |
1547 | */ | |
6130f531 | 1548 | static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov, |
6a14b90b JA |
1549 | unsigned long nr_segs, unsigned int flags) |
1550 | { | |
1551 | struct pipe_inode_info *pipe; | |
1552 | struct splice_desc sd; | |
6a14b90b | 1553 | long ret; |
6130f531 AV |
1554 | struct iovec iovstack[UIO_FASTIOV]; |
1555 | struct iovec *iov = iovstack; | |
1556 | struct iov_iter iter; | |
6a14b90b | 1557 | |
71993e62 | 1558 | pipe = get_pipe_info(file); |
6a14b90b JA |
1559 | if (!pipe) |
1560 | return -EBADF; | |
1561 | ||
345995fa AV |
1562 | ret = import_iovec(READ, uiov, nr_segs, |
1563 | ARRAY_SIZE(iovstack), &iov, &iter); | |
1564 | if (ret < 0) | |
1565 | return ret; | |
6a14b90b | 1566 | |
345995fa | 1567 | sd.total_len = iov_iter_count(&iter); |
6130f531 | 1568 | sd.len = 0; |
6130f531 AV |
1569 | sd.flags = flags; |
1570 | sd.u.data = &iter; | |
1571 | sd.pos = 0; | |
6a14b90b | 1572 | |
345995fa AV |
1573 | if (sd.total_len) { |
1574 | pipe_lock(pipe); | |
1575 | ret = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1576 | pipe_unlock(pipe); | |
1577 | } | |
6a14b90b | 1578 | |
345995fa | 1579 | kfree(iov); |
6a14b90b JA |
1580 | return ret; |
1581 | } | |
1582 | ||
912d35f8 JA |
1583 | /* |
1584 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1585 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1586 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1587 | */ |
6a14b90b JA |
1588 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1589 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1590 | { |
ddac0d39 | 1591 | struct pipe_inode_info *pipe; |
35f3d14d JA |
1592 | struct page *pages[PIPE_DEF_BUFFERS]; |
1593 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
912d35f8 JA |
1594 | struct splice_pipe_desc spd = { |
1595 | .pages = pages, | |
1596 | .partial = partial, | |
047fe360 | 1597 | .nr_pages_max = PIPE_DEF_BUFFERS, |
912d35f8 JA |
1598 | .flags = flags, |
1599 | .ops = &user_page_pipe_buf_ops, | |
bbdfc2f7 | 1600 | .spd_release = spd_release_page, |
912d35f8 | 1601 | }; |
35f3d14d | 1602 | long ret; |
912d35f8 | 1603 | |
71993e62 | 1604 | pipe = get_pipe_info(file); |
ddac0d39 | 1605 | if (!pipe) |
912d35f8 | 1606 | return -EBADF; |
912d35f8 | 1607 | |
35f3d14d JA |
1608 | if (splice_grow_spd(pipe, &spd)) |
1609 | return -ENOMEM; | |
1610 | ||
1611 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages, | |
bd1a68b5 | 1612 | spd.partial, false, |
047fe360 | 1613 | spd.nr_pages_max); |
912d35f8 | 1614 | if (spd.nr_pages <= 0) |
35f3d14d JA |
1615 | ret = spd.nr_pages; |
1616 | else | |
1617 | ret = splice_to_pipe(pipe, &spd); | |
912d35f8 | 1618 | |
047fe360 | 1619 | splice_shrink_spd(&spd); |
35f3d14d | 1620 | return ret; |
912d35f8 JA |
1621 | } |
1622 | ||
6a14b90b JA |
1623 | /* |
1624 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1625 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1626 | * operation that can be supported without any funky alignment restrictions | |
1627 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1628 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1629 | * solutions for that: | |
1630 | * | |
1631 | * - memcpy() the data internally, at which point we might as well just | |
1632 | * do a regular read() on the buffer anyway. | |
1633 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1634 | * has restriction limitations on both ends of the pipe). | |
1635 | * | |
1636 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1637 | * | |
1638 | */ | |
836f92ad HC |
1639 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, |
1640 | unsigned long, nr_segs, unsigned int, flags) | |
912d35f8 | 1641 | { |
2903ff01 | 1642 | struct fd f; |
912d35f8 | 1643 | long error; |
912d35f8 | 1644 | |
6a14b90b JA |
1645 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1646 | return -EINVAL; | |
1647 | else if (unlikely(!nr_segs)) | |
1648 | return 0; | |
1649 | ||
912d35f8 | 1650 | error = -EBADF; |
2903ff01 AV |
1651 | f = fdget(fd); |
1652 | if (f.file) { | |
1653 | if (f.file->f_mode & FMODE_WRITE) | |
1654 | error = vmsplice_to_pipe(f.file, iov, nr_segs, flags); | |
1655 | else if (f.file->f_mode & FMODE_READ) | |
1656 | error = vmsplice_to_user(f.file, iov, nr_segs, flags); | |
1657 | ||
1658 | fdput(f); | |
912d35f8 JA |
1659 | } |
1660 | ||
1661 | return error; | |
1662 | } | |
1663 | ||
76b021d0 AV |
1664 | #ifdef CONFIG_COMPAT |
1665 | COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, | |
1666 | unsigned int, nr_segs, unsigned int, flags) | |
1667 | { | |
1668 | unsigned i; | |
1669 | struct iovec __user *iov; | |
1670 | if (nr_segs > UIO_MAXIOV) | |
1671 | return -EINVAL; | |
1672 | iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec)); | |
1673 | for (i = 0; i < nr_segs; i++) { | |
1674 | struct compat_iovec v; | |
1675 | if (get_user(v.iov_base, &iov32[i].iov_base) || | |
1676 | get_user(v.iov_len, &iov32[i].iov_len) || | |
1677 | put_user(compat_ptr(v.iov_base), &iov[i].iov_base) || | |
1678 | put_user(v.iov_len, &iov[i].iov_len)) | |
1679 | return -EFAULT; | |
1680 | } | |
1681 | return sys_vmsplice(fd, iov, nr_segs, flags); | |
1682 | } | |
1683 | #endif | |
1684 | ||
836f92ad HC |
1685 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
1686 | int, fd_out, loff_t __user *, off_out, | |
1687 | size_t, len, unsigned int, flags) | |
5274f052 | 1688 | { |
2903ff01 | 1689 | struct fd in, out; |
5274f052 | 1690 | long error; |
5274f052 JA |
1691 | |
1692 | if (unlikely(!len)) | |
1693 | return 0; | |
1694 | ||
1695 | error = -EBADF; | |
2903ff01 AV |
1696 | in = fdget(fd_in); |
1697 | if (in.file) { | |
1698 | if (in.file->f_mode & FMODE_READ) { | |
1699 | out = fdget(fd_out); | |
1700 | if (out.file) { | |
1701 | if (out.file->f_mode & FMODE_WRITE) | |
1702 | error = do_splice(in.file, off_in, | |
1703 | out.file, off_out, | |
529565dc | 1704 | len, flags); |
2903ff01 | 1705 | fdput(out); |
5274f052 JA |
1706 | } |
1707 | } | |
2903ff01 | 1708 | fdput(in); |
5274f052 | 1709 | } |
5274f052 JA |
1710 | return error; |
1711 | } | |
70524490 | 1712 | |
aadd06e5 JA |
1713 | /* |
1714 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1715 | * return an appropriate error. | |
1716 | */ | |
7c77f0b3 | 1717 | static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1718 | { |
1719 | int ret; | |
1720 | ||
1721 | /* | |
1722 | * Check ->nrbufs without the inode lock first. This function | |
1723 | * is speculative anyways, so missing one is ok. | |
1724 | */ | |
1725 | if (pipe->nrbufs) | |
1726 | return 0; | |
1727 | ||
1728 | ret = 0; | |
61e0d47c | 1729 | pipe_lock(pipe); |
aadd06e5 JA |
1730 | |
1731 | while (!pipe->nrbufs) { | |
1732 | if (signal_pending(current)) { | |
1733 | ret = -ERESTARTSYS; | |
1734 | break; | |
1735 | } | |
1736 | if (!pipe->writers) | |
1737 | break; | |
1738 | if (!pipe->waiting_writers) { | |
1739 | if (flags & SPLICE_F_NONBLOCK) { | |
1740 | ret = -EAGAIN; | |
1741 | break; | |
1742 | } | |
1743 | } | |
1744 | pipe_wait(pipe); | |
1745 | } | |
1746 | ||
61e0d47c | 1747 | pipe_unlock(pipe); |
aadd06e5 JA |
1748 | return ret; |
1749 | } | |
1750 | ||
1751 | /* | |
1752 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1753 | * return an appropriate error. | |
1754 | */ | |
7c77f0b3 | 1755 | static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1756 | { |
1757 | int ret; | |
1758 | ||
1759 | /* | |
1760 | * Check ->nrbufs without the inode lock first. This function | |
1761 | * is speculative anyways, so missing one is ok. | |
1762 | */ | |
35f3d14d | 1763 | if (pipe->nrbufs < pipe->buffers) |
aadd06e5 JA |
1764 | return 0; |
1765 | ||
1766 | ret = 0; | |
61e0d47c | 1767 | pipe_lock(pipe); |
aadd06e5 | 1768 | |
35f3d14d | 1769 | while (pipe->nrbufs >= pipe->buffers) { |
aadd06e5 JA |
1770 | if (!pipe->readers) { |
1771 | send_sig(SIGPIPE, current, 0); | |
1772 | ret = -EPIPE; | |
1773 | break; | |
1774 | } | |
1775 | if (flags & SPLICE_F_NONBLOCK) { | |
1776 | ret = -EAGAIN; | |
1777 | break; | |
1778 | } | |
1779 | if (signal_pending(current)) { | |
1780 | ret = -ERESTARTSYS; | |
1781 | break; | |
1782 | } | |
1783 | pipe->waiting_writers++; | |
1784 | pipe_wait(pipe); | |
1785 | pipe->waiting_writers--; | |
1786 | } | |
1787 | ||
61e0d47c | 1788 | pipe_unlock(pipe); |
aadd06e5 JA |
1789 | return ret; |
1790 | } | |
1791 | ||
7c77f0b3 MS |
1792 | /* |
1793 | * Splice contents of ipipe to opipe. | |
1794 | */ | |
1795 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, | |
1796 | struct pipe_inode_info *opipe, | |
1797 | size_t len, unsigned int flags) | |
1798 | { | |
1799 | struct pipe_buffer *ibuf, *obuf; | |
1800 | int ret = 0, nbuf; | |
1801 | bool input_wakeup = false; | |
1802 | ||
1803 | ||
1804 | retry: | |
1805 | ret = ipipe_prep(ipipe, flags); | |
1806 | if (ret) | |
1807 | return ret; | |
1808 | ||
1809 | ret = opipe_prep(opipe, flags); | |
1810 | if (ret) | |
1811 | return ret; | |
1812 | ||
1813 | /* | |
1814 | * Potential ABBA deadlock, work around it by ordering lock | |
1815 | * grabbing by pipe info address. Otherwise two different processes | |
1816 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1817 | */ | |
1818 | pipe_double_lock(ipipe, opipe); | |
1819 | ||
1820 | do { | |
1821 | if (!opipe->readers) { | |
1822 | send_sig(SIGPIPE, current, 0); | |
1823 | if (!ret) | |
1824 | ret = -EPIPE; | |
1825 | break; | |
1826 | } | |
1827 | ||
1828 | if (!ipipe->nrbufs && !ipipe->writers) | |
1829 | break; | |
1830 | ||
1831 | /* | |
1832 | * Cannot make any progress, because either the input | |
1833 | * pipe is empty or the output pipe is full. | |
1834 | */ | |
35f3d14d | 1835 | if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { |
7c77f0b3 MS |
1836 | /* Already processed some buffers, break */ |
1837 | if (ret) | |
1838 | break; | |
1839 | ||
1840 | if (flags & SPLICE_F_NONBLOCK) { | |
1841 | ret = -EAGAIN; | |
1842 | break; | |
1843 | } | |
1844 | ||
1845 | /* | |
1846 | * We raced with another reader/writer and haven't | |
1847 | * managed to process any buffers. A zero return | |
1848 | * value means EOF, so retry instead. | |
1849 | */ | |
1850 | pipe_unlock(ipipe); | |
1851 | pipe_unlock(opipe); | |
1852 | goto retry; | |
1853 | } | |
1854 | ||
1855 | ibuf = ipipe->bufs + ipipe->curbuf; | |
35f3d14d | 1856 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); |
7c77f0b3 MS |
1857 | obuf = opipe->bufs + nbuf; |
1858 | ||
1859 | if (len >= ibuf->len) { | |
1860 | /* | |
1861 | * Simply move the whole buffer from ipipe to opipe | |
1862 | */ | |
1863 | *obuf = *ibuf; | |
1864 | ibuf->ops = NULL; | |
1865 | opipe->nrbufs++; | |
35f3d14d | 1866 | ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); |
7c77f0b3 MS |
1867 | ipipe->nrbufs--; |
1868 | input_wakeup = true; | |
1869 | } else { | |
1870 | /* | |
1871 | * Get a reference to this pipe buffer, | |
1872 | * so we can copy the contents over. | |
1873 | */ | |
1874 | ibuf->ops->get(ipipe, ibuf); | |
1875 | *obuf = *ibuf; | |
1876 | ||
1877 | /* | |
1878 | * Don't inherit the gift flag, we need to | |
1879 | * prevent multiple steals of this page. | |
1880 | */ | |
1881 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1882 | ||
1883 | obuf->len = len; | |
1884 | opipe->nrbufs++; | |
1885 | ibuf->offset += obuf->len; | |
1886 | ibuf->len -= obuf->len; | |
1887 | } | |
1888 | ret += obuf->len; | |
1889 | len -= obuf->len; | |
1890 | } while (len); | |
1891 | ||
1892 | pipe_unlock(ipipe); | |
1893 | pipe_unlock(opipe); | |
1894 | ||
1895 | /* | |
1896 | * If we put data in the output pipe, wakeup any potential readers. | |
1897 | */ | |
825cdcb1 NK |
1898 | if (ret > 0) |
1899 | wakeup_pipe_readers(opipe); | |
1900 | ||
7c77f0b3 MS |
1901 | if (input_wakeup) |
1902 | wakeup_pipe_writers(ipipe); | |
1903 | ||
1904 | return ret; | |
1905 | } | |
1906 | ||
70524490 JA |
1907 | /* |
1908 | * Link contents of ipipe to opipe. | |
1909 | */ | |
1910 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1911 | struct pipe_inode_info *opipe, | |
1912 | size_t len, unsigned int flags) | |
1913 | { | |
1914 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1915 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1916 | |
1917 | /* | |
1918 | * Potential ABBA deadlock, work around it by ordering lock | |
61e0d47c | 1919 | * grabbing by pipe info address. Otherwise two different processes |
70524490 JA |
1920 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
1921 | */ | |
61e0d47c | 1922 | pipe_double_lock(ipipe, opipe); |
70524490 | 1923 | |
aadd06e5 | 1924 | do { |
70524490 JA |
1925 | if (!opipe->readers) { |
1926 | send_sig(SIGPIPE, current, 0); | |
1927 | if (!ret) | |
1928 | ret = -EPIPE; | |
1929 | break; | |
1930 | } | |
70524490 | 1931 | |
aadd06e5 JA |
1932 | /* |
1933 | * If we have iterated all input buffers or ran out of | |
1934 | * output room, break. | |
1935 | */ | |
35f3d14d | 1936 | if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) |
aadd06e5 | 1937 | break; |
70524490 | 1938 | |
35f3d14d JA |
1939 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); |
1940 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); | |
70524490 JA |
1941 | |
1942 | /* | |
aadd06e5 JA |
1943 | * Get a reference to this pipe buffer, |
1944 | * so we can copy the contents over. | |
70524490 | 1945 | */ |
aadd06e5 JA |
1946 | ibuf->ops->get(ipipe, ibuf); |
1947 | ||
1948 | obuf = opipe->bufs + nbuf; | |
1949 | *obuf = *ibuf; | |
1950 | ||
2a27250e | 1951 | /* |
aadd06e5 JA |
1952 | * Don't inherit the gift flag, we need to |
1953 | * prevent multiple steals of this page. | |
2a27250e | 1954 | */ |
aadd06e5 | 1955 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1956 | |
aadd06e5 JA |
1957 | if (obuf->len > len) |
1958 | obuf->len = len; | |
70524490 | 1959 | |
aadd06e5 JA |
1960 | opipe->nrbufs++; |
1961 | ret += obuf->len; | |
1962 | len -= obuf->len; | |
1963 | i++; | |
1964 | } while (len); | |
70524490 | 1965 | |
02cf01ae JA |
1966 | /* |
1967 | * return EAGAIN if we have the potential of some data in the | |
1968 | * future, otherwise just return 0 | |
1969 | */ | |
1970 | if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) | |
1971 | ret = -EAGAIN; | |
1972 | ||
61e0d47c MS |
1973 | pipe_unlock(ipipe); |
1974 | pipe_unlock(opipe); | |
70524490 | 1975 | |
aadd06e5 JA |
1976 | /* |
1977 | * If we put data in the output pipe, wakeup any potential readers. | |
1978 | */ | |
825cdcb1 NK |
1979 | if (ret > 0) |
1980 | wakeup_pipe_readers(opipe); | |
70524490 JA |
1981 | |
1982 | return ret; | |
1983 | } | |
1984 | ||
1985 | /* | |
1986 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1987 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1988 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1989 | * applicable one is SPLICE_F_NONBLOCK. | |
1990 | */ | |
1991 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1992 | unsigned int flags) | |
1993 | { | |
71993e62 LT |
1994 | struct pipe_inode_info *ipipe = get_pipe_info(in); |
1995 | struct pipe_inode_info *opipe = get_pipe_info(out); | |
aadd06e5 | 1996 | int ret = -EINVAL; |
70524490 JA |
1997 | |
1998 | /* | |
aadd06e5 JA |
1999 | * Duplicate the contents of ipipe to opipe without actually |
2000 | * copying the data. | |
70524490 | 2001 | */ |
aadd06e5 JA |
2002 | if (ipipe && opipe && ipipe != opipe) { |
2003 | /* | |
2004 | * Keep going, unless we encounter an error. The ipipe/opipe | |
2005 | * ordering doesn't really matter. | |
2006 | */ | |
7c77f0b3 | 2007 | ret = ipipe_prep(ipipe, flags); |
aadd06e5 | 2008 | if (!ret) { |
7c77f0b3 | 2009 | ret = opipe_prep(opipe, flags); |
02cf01ae | 2010 | if (!ret) |
aadd06e5 | 2011 | ret = link_pipe(ipipe, opipe, len, flags); |
aadd06e5 JA |
2012 | } |
2013 | } | |
70524490 | 2014 | |
aadd06e5 | 2015 | return ret; |
70524490 JA |
2016 | } |
2017 | ||
836f92ad | 2018 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
70524490 | 2019 | { |
2903ff01 AV |
2020 | struct fd in; |
2021 | int error; | |
70524490 JA |
2022 | |
2023 | if (unlikely(!len)) | |
2024 | return 0; | |
2025 | ||
2026 | error = -EBADF; | |
2903ff01 AV |
2027 | in = fdget(fdin); |
2028 | if (in.file) { | |
2029 | if (in.file->f_mode & FMODE_READ) { | |
2030 | struct fd out = fdget(fdout); | |
2031 | if (out.file) { | |
2032 | if (out.file->f_mode & FMODE_WRITE) | |
2033 | error = do_tee(in.file, out.file, | |
2034 | len, flags); | |
2035 | fdput(out); | |
70524490 JA |
2036 | } |
2037 | } | |
2903ff01 | 2038 | fdput(in); |
70524490 JA |
2039 | } |
2040 | ||
2041 | return error; | |
2042 | } |