]>
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 | 187 | |
d5088a25 RV |
188 | if (!spd_pages) |
189 | return 0; | |
190 | ||
5274f052 JA |
191 | ret = 0; |
192 | do_wakeup = 0; | |
912d35f8 | 193 | page_nr = 0; |
5274f052 | 194 | |
61e0d47c | 195 | pipe_lock(pipe); |
5274f052 | 196 | |
5274f052 | 197 | for (;;) { |
3a326a2c | 198 | if (!pipe->readers) { |
5274f052 JA |
199 | send_sig(SIGPIPE, current, 0); |
200 | if (!ret) | |
201 | ret = -EPIPE; | |
202 | break; | |
203 | } | |
204 | ||
35f3d14d JA |
205 | if (pipe->nrbufs < pipe->buffers) { |
206 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
3a326a2c | 207 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 208 | |
912d35f8 JA |
209 | buf->page = spd->pages[page_nr]; |
210 | buf->offset = spd->partial[page_nr].offset; | |
211 | buf->len = spd->partial[page_nr].len; | |
497f9625 | 212 | buf->private = spd->partial[page_nr].private; |
912d35f8 | 213 | buf->ops = spd->ops; |
13edda5f | 214 | buf->flags = 0; |
7afa6fd0 JA |
215 | if (spd->flags & SPLICE_F_GIFT) |
216 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
217 | ||
6f767b04 | 218 | pipe->nrbufs++; |
912d35f8 JA |
219 | page_nr++; |
220 | ret += buf->len; | |
221 | ||
6447a3cf | 222 | if (pipe->files) |
6f767b04 | 223 | do_wakeup = 1; |
5274f052 | 224 | |
912d35f8 | 225 | if (!--spd->nr_pages) |
5274f052 | 226 | break; |
35f3d14d | 227 | if (pipe->nrbufs < pipe->buffers) |
5274f052 JA |
228 | continue; |
229 | ||
230 | break; | |
231 | } | |
232 | ||
912d35f8 | 233 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
234 | if (!ret) |
235 | ret = -EAGAIN; | |
236 | break; | |
237 | } | |
238 | ||
5274f052 JA |
239 | if (signal_pending(current)) { |
240 | if (!ret) | |
241 | ret = -ERESTARTSYS; | |
242 | break; | |
243 | } | |
244 | ||
245 | if (do_wakeup) { | |
c0bd1f65 | 246 | smp_mb(); |
3a326a2c IM |
247 | if (waitqueue_active(&pipe->wait)) |
248 | wake_up_interruptible_sync(&pipe->wait); | |
249 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
250 | do_wakeup = 0; |
251 | } | |
252 | ||
3a326a2c IM |
253 | pipe->waiting_writers++; |
254 | pipe_wait(pipe); | |
255 | pipe->waiting_writers--; | |
5274f052 JA |
256 | } |
257 | ||
61e0d47c | 258 | pipe_unlock(pipe); |
5274f052 | 259 | |
825cdcb1 NK |
260 | if (do_wakeup) |
261 | wakeup_pipe_readers(pipe); | |
5274f052 | 262 | |
00de00bd | 263 | while (page_nr < spd_pages) |
bbdfc2f7 | 264 | spd->spd_release(spd, page_nr++); |
5274f052 JA |
265 | |
266 | return ret; | |
267 | } | |
2b514574 | 268 | EXPORT_SYMBOL_GPL(splice_to_pipe); |
5274f052 | 269 | |
708e3508 | 270 | void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) |
bbdfc2f7 JA |
271 | { |
272 | page_cache_release(spd->pages[i]); | |
273 | } | |
274 | ||
35f3d14d JA |
275 | /* |
276 | * Check if we need to grow the arrays holding pages and partial page | |
277 | * descriptions. | |
278 | */ | |
047fe360 | 279 | int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) |
35f3d14d | 280 | { |
047fe360 ED |
281 | unsigned int buffers = ACCESS_ONCE(pipe->buffers); |
282 | ||
283 | spd->nr_pages_max = buffers; | |
284 | if (buffers <= PIPE_DEF_BUFFERS) | |
35f3d14d JA |
285 | return 0; |
286 | ||
047fe360 ED |
287 | spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL); |
288 | spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL); | |
35f3d14d JA |
289 | |
290 | if (spd->pages && spd->partial) | |
291 | return 0; | |
292 | ||
293 | kfree(spd->pages); | |
294 | kfree(spd->partial); | |
295 | return -ENOMEM; | |
296 | } | |
297 | ||
047fe360 | 298 | void splice_shrink_spd(struct splice_pipe_desc *spd) |
35f3d14d | 299 | { |
047fe360 | 300 | if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) |
35f3d14d JA |
301 | return; |
302 | ||
303 | kfree(spd->pages); | |
304 | kfree(spd->partial); | |
305 | } | |
306 | ||
3a326a2c | 307 | static int |
cbb7e577 JA |
308 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
309 | struct pipe_inode_info *pipe, size_t len, | |
310 | unsigned int flags) | |
5274f052 JA |
311 | { |
312 | struct address_space *mapping = in->f_mapping; | |
d8983910 | 313 | unsigned int loff, nr_pages, req_pages; |
35f3d14d JA |
314 | struct page *pages[PIPE_DEF_BUFFERS]; |
315 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
5274f052 | 316 | struct page *page; |
91ad66ef JA |
317 | pgoff_t index, end_index; |
318 | loff_t isize; | |
eb20796b | 319 | int error, page_nr; |
912d35f8 JA |
320 | struct splice_pipe_desc spd = { |
321 | .pages = pages, | |
322 | .partial = partial, | |
047fe360 | 323 | .nr_pages_max = PIPE_DEF_BUFFERS, |
912d35f8 JA |
324 | .flags = flags, |
325 | .ops = &page_cache_pipe_buf_ops, | |
bbdfc2f7 | 326 | .spd_release = spd_release_page, |
912d35f8 | 327 | }; |
5274f052 | 328 | |
35f3d14d JA |
329 | if (splice_grow_spd(pipe, &spd)) |
330 | return -ENOMEM; | |
331 | ||
cbb7e577 | 332 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 | 333 | loff = *ppos & ~PAGE_CACHE_MASK; |
d8983910 | 334 | req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
047fe360 | 335 | nr_pages = min(req_pages, spd.nr_pages_max); |
5274f052 | 336 | |
eb20796b JA |
337 | /* |
338 | * Lookup the (hopefully) full range of pages we need. | |
339 | */ | |
35f3d14d | 340 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages); |
431a4820 | 341 | index += spd.nr_pages; |
82aa5d61 | 342 | |
eb20796b JA |
343 | /* |
344 | * If find_get_pages_contig() returned fewer pages than we needed, | |
431a4820 | 345 | * readahead/allocate the rest and fill in the holes. |
eb20796b | 346 | */ |
431a4820 | 347 | if (spd.nr_pages < nr_pages) |
cf914a7d RR |
348 | page_cache_sync_readahead(mapping, &in->f_ra, in, |
349 | index, req_pages - spd.nr_pages); | |
431a4820 | 350 | |
932cc6d4 | 351 | error = 0; |
eb20796b | 352 | while (spd.nr_pages < nr_pages) { |
82aa5d61 | 353 | /* |
eb20796b JA |
354 | * Page could be there, find_get_pages_contig() breaks on |
355 | * the first hole. | |
5274f052 | 356 | */ |
7480a904 JA |
357 | page = find_get_page(mapping, index); |
358 | if (!page) { | |
7480a904 | 359 | /* |
eb20796b | 360 | * page didn't exist, allocate one. |
7480a904 JA |
361 | */ |
362 | page = page_cache_alloc_cold(mapping); | |
363 | if (!page) | |
364 | break; | |
365 | ||
366 | error = add_to_page_cache_lru(page, mapping, index, | |
c62d2555 | 367 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
7480a904 JA |
368 | if (unlikely(error)) { |
369 | page_cache_release(page); | |
a0548871 JA |
370 | if (error == -EEXIST) |
371 | continue; | |
7480a904 JA |
372 | break; |
373 | } | |
eb20796b JA |
374 | /* |
375 | * add_to_page_cache() locks the page, unlock it | |
376 | * to avoid convoluting the logic below even more. | |
377 | */ | |
378 | unlock_page(page); | |
7480a904 JA |
379 | } |
380 | ||
35f3d14d | 381 | spd.pages[spd.nr_pages++] = page; |
eb20796b JA |
382 | index++; |
383 | } | |
384 | ||
385 | /* | |
386 | * Now loop over the map and see if we need to start IO on any | |
387 | * pages, fill in the partial map, etc. | |
388 | */ | |
389 | index = *ppos >> PAGE_CACHE_SHIFT; | |
390 | nr_pages = spd.nr_pages; | |
391 | spd.nr_pages = 0; | |
392 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
393 | unsigned int this_len; | |
394 | ||
395 | if (!len) | |
396 | break; | |
397 | ||
398 | /* | |
399 | * this_len is the max we'll use from this page | |
400 | */ | |
401 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
35f3d14d | 402 | page = spd.pages[page_nr]; |
eb20796b | 403 | |
a08a166f | 404 | if (PageReadahead(page)) |
cf914a7d | 405 | page_cache_async_readahead(mapping, &in->f_ra, in, |
d8983910 | 406 | page, index, req_pages - page_nr); |
a08a166f | 407 | |
7480a904 JA |
408 | /* |
409 | * If the page isn't uptodate, we may need to start io on it | |
410 | */ | |
411 | if (!PageUptodate(page)) { | |
6965031d | 412 | lock_page(page); |
7480a904 JA |
413 | |
414 | /* | |
32502b84 MS |
415 | * Page was truncated, or invalidated by the |
416 | * filesystem. Redo the find/create, but this time the | |
417 | * page is kept locked, so there's no chance of another | |
418 | * race with truncate/invalidate. | |
7480a904 JA |
419 | */ |
420 | if (!page->mapping) { | |
421 | unlock_page(page); | |
32502b84 MS |
422 | page = find_or_create_page(mapping, index, |
423 | mapping_gfp_mask(mapping)); | |
424 | ||
425 | if (!page) { | |
426 | error = -ENOMEM; | |
427 | break; | |
428 | } | |
35f3d14d JA |
429 | page_cache_release(spd.pages[page_nr]); |
430 | spd.pages[page_nr] = page; | |
7480a904 JA |
431 | } |
432 | /* | |
433 | * page was already under io and is now done, great | |
434 | */ | |
435 | if (PageUptodate(page)) { | |
436 | unlock_page(page); | |
437 | goto fill_it; | |
438 | } | |
5274f052 | 439 | |
7480a904 JA |
440 | /* |
441 | * need to read in the page | |
442 | */ | |
443 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 444 | if (unlikely(error)) { |
eb20796b JA |
445 | /* |
446 | * We really should re-lookup the page here, | |
447 | * but it complicates things a lot. Instead | |
448 | * lets just do what we already stored, and | |
449 | * we'll get it the next time we are called. | |
450 | */ | |
7480a904 | 451 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
452 | error = 0; |
453 | ||
5274f052 JA |
454 | break; |
455 | } | |
620a324b JA |
456 | } |
457 | fill_it: | |
458 | /* | |
459 | * i_size must be checked after PageUptodate. | |
460 | */ | |
461 | isize = i_size_read(mapping->host); | |
462 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
463 | if (unlikely(!isize || index > end_index)) | |
464 | break; | |
465 | ||
466 | /* | |
467 | * if this is the last page, see if we need to shrink | |
468 | * the length and stop | |
469 | */ | |
470 | if (end_index == index) { | |
471 | unsigned int plen; | |
91ad66ef JA |
472 | |
473 | /* | |
620a324b | 474 | * max good bytes in this page |
91ad66ef | 475 | */ |
620a324b JA |
476 | plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
477 | if (plen <= loff) | |
91ad66ef | 478 | break; |
91ad66ef JA |
479 | |
480 | /* | |
620a324b | 481 | * force quit after adding this page |
91ad66ef | 482 | */ |
620a324b JA |
483 | this_len = min(this_len, plen - loff); |
484 | len = this_len; | |
5274f052 | 485 | } |
620a324b | 486 | |
35f3d14d JA |
487 | spd.partial[page_nr].offset = loff; |
488 | spd.partial[page_nr].len = this_len; | |
82aa5d61 | 489 | len -= this_len; |
91ad66ef | 490 | loff = 0; |
eb20796b JA |
491 | spd.nr_pages++; |
492 | index++; | |
5274f052 JA |
493 | } |
494 | ||
eb20796b | 495 | /* |
475ecade | 496 | * Release any pages at the end, if we quit early. 'page_nr' is how far |
eb20796b JA |
497 | * we got, 'nr_pages' is how many pages are in the map. |
498 | */ | |
499 | while (page_nr < nr_pages) | |
35f3d14d | 500 | page_cache_release(spd.pages[page_nr++]); |
f4e6b498 | 501 | in->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; |
eb20796b | 502 | |
912d35f8 | 503 | if (spd.nr_pages) |
35f3d14d | 504 | error = splice_to_pipe(pipe, &spd); |
5274f052 | 505 | |
047fe360 | 506 | splice_shrink_spd(&spd); |
7480a904 | 507 | return error; |
5274f052 JA |
508 | } |
509 | ||
83f9135b JA |
510 | /** |
511 | * generic_file_splice_read - splice data from file to a pipe | |
512 | * @in: file to splice from | |
932cc6d4 | 513 | * @ppos: position in @in |
83f9135b JA |
514 | * @pipe: pipe to splice to |
515 | * @len: number of bytes to splice | |
516 | * @flags: splice modifier flags | |
517 | * | |
932cc6d4 JA |
518 | * Description: |
519 | * Will read pages from given file and fill them into a pipe. Can be | |
520 | * used as long as the address_space operations for the source implements | |
521 | * a readpage() hook. | |
522 | * | |
83f9135b | 523 | */ |
cbb7e577 JA |
524 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
525 | struct pipe_inode_info *pipe, size_t len, | |
526 | unsigned int flags) | |
5274f052 | 527 | { |
d366d398 | 528 | loff_t isize, left; |
8191ecd1 | 529 | int ret; |
d366d398 | 530 | |
be64f884 BH |
531 | if (IS_DAX(in->f_mapping->host)) |
532 | return default_file_splice_read(in, ppos, pipe, len, flags); | |
533 | ||
d366d398 JA |
534 | isize = i_size_read(in->f_mapping->host); |
535 | if (unlikely(*ppos >= isize)) | |
536 | return 0; | |
537 | ||
538 | left = isize - *ppos; | |
539 | if (unlikely(left < len)) | |
540 | len = left; | |
5274f052 | 541 | |
8191ecd1 | 542 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
723590ed | 543 | if (ret > 0) { |
cbb7e577 | 544 | *ppos += ret; |
723590ed MS |
545 | file_accessed(in); |
546 | } | |
5274f052 JA |
547 | |
548 | return ret; | |
549 | } | |
059a8f37 JA |
550 | EXPORT_SYMBOL(generic_file_splice_read); |
551 | ||
6818173b MS |
552 | static const struct pipe_buf_operations default_pipe_buf_ops = { |
553 | .can_merge = 0, | |
6818173b MS |
554 | .confirm = generic_pipe_buf_confirm, |
555 | .release = generic_pipe_buf_release, | |
556 | .steal = generic_pipe_buf_steal, | |
557 | .get = generic_pipe_buf_get, | |
558 | }; | |
559 | ||
28a625cb MS |
560 | static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, |
561 | struct pipe_buffer *buf) | |
562 | { | |
563 | return 1; | |
564 | } | |
565 | ||
566 | /* Pipe buffer operations for a socket and similar. */ | |
567 | const struct pipe_buf_operations nosteal_pipe_buf_ops = { | |
568 | .can_merge = 0, | |
28a625cb MS |
569 | .confirm = generic_pipe_buf_confirm, |
570 | .release = generic_pipe_buf_release, | |
571 | .steal = generic_pipe_buf_nosteal, | |
572 | .get = generic_pipe_buf_get, | |
573 | }; | |
574 | EXPORT_SYMBOL(nosteal_pipe_buf_ops); | |
575 | ||
6818173b MS |
576 | static ssize_t kernel_readv(struct file *file, const struct iovec *vec, |
577 | unsigned long vlen, loff_t offset) | |
578 | { | |
579 | mm_segment_t old_fs; | |
580 | loff_t pos = offset; | |
581 | ssize_t res; | |
582 | ||
583 | old_fs = get_fs(); | |
584 | set_fs(get_ds()); | |
585 | /* The cast to a user pointer is valid due to the set_fs() */ | |
586 | res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos); | |
587 | set_fs(old_fs); | |
588 | ||
589 | return res; | |
590 | } | |
591 | ||
7bb307e8 | 592 | ssize_t kernel_write(struct file *file, const char *buf, size_t count, |
b2858d7d | 593 | loff_t pos) |
0b0a47f5 MS |
594 | { |
595 | mm_segment_t old_fs; | |
596 | ssize_t res; | |
597 | ||
598 | old_fs = get_fs(); | |
599 | set_fs(get_ds()); | |
600 | /* The cast to a user pointer is valid due to the set_fs() */ | |
7bb307e8 | 601 | res = vfs_write(file, (__force const char __user *)buf, count, &pos); |
0b0a47f5 MS |
602 | set_fs(old_fs); |
603 | ||
604 | return res; | |
605 | } | |
7bb307e8 | 606 | EXPORT_SYMBOL(kernel_write); |
0b0a47f5 | 607 | |
6818173b MS |
608 | ssize_t default_file_splice_read(struct file *in, loff_t *ppos, |
609 | struct pipe_inode_info *pipe, size_t len, | |
610 | unsigned int flags) | |
611 | { | |
612 | unsigned int nr_pages; | |
613 | unsigned int nr_freed; | |
614 | size_t offset; | |
35f3d14d JA |
615 | struct page *pages[PIPE_DEF_BUFFERS]; |
616 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
617 | struct iovec *vec, __vec[PIPE_DEF_BUFFERS]; | |
6818173b MS |
618 | ssize_t res; |
619 | size_t this_len; | |
620 | int error; | |
621 | int i; | |
622 | struct splice_pipe_desc spd = { | |
623 | .pages = pages, | |
624 | .partial = partial, | |
047fe360 | 625 | .nr_pages_max = PIPE_DEF_BUFFERS, |
6818173b MS |
626 | .flags = flags, |
627 | .ops = &default_pipe_buf_ops, | |
628 | .spd_release = spd_release_page, | |
629 | }; | |
630 | ||
35f3d14d JA |
631 | if (splice_grow_spd(pipe, &spd)) |
632 | return -ENOMEM; | |
633 | ||
634 | res = -ENOMEM; | |
635 | vec = __vec; | |
047fe360 ED |
636 | if (spd.nr_pages_max > PIPE_DEF_BUFFERS) { |
637 | vec = kmalloc(spd.nr_pages_max * sizeof(struct iovec), GFP_KERNEL); | |
35f3d14d JA |
638 | if (!vec) |
639 | goto shrink_ret; | |
640 | } | |
641 | ||
6818173b MS |
642 | offset = *ppos & ~PAGE_CACHE_MASK; |
643 | nr_pages = (len + offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
644 | ||
047fe360 | 645 | for (i = 0; i < nr_pages && i < spd.nr_pages_max && len; i++) { |
6818173b MS |
646 | struct page *page; |
647 | ||
4f231228 | 648 | page = alloc_page(GFP_USER); |
6818173b MS |
649 | error = -ENOMEM; |
650 | if (!page) | |
651 | goto err; | |
652 | ||
653 | this_len = min_t(size_t, len, PAGE_CACHE_SIZE - offset); | |
4f231228 | 654 | vec[i].iov_base = (void __user *) page_address(page); |
6818173b | 655 | vec[i].iov_len = this_len; |
35f3d14d | 656 | spd.pages[i] = page; |
6818173b MS |
657 | spd.nr_pages++; |
658 | len -= this_len; | |
659 | offset = 0; | |
660 | } | |
661 | ||
662 | res = kernel_readv(in, vec, spd.nr_pages, *ppos); | |
77f6bf57 AM |
663 | if (res < 0) { |
664 | error = res; | |
6818173b | 665 | goto err; |
77f6bf57 | 666 | } |
6818173b MS |
667 | |
668 | error = 0; | |
669 | if (!res) | |
670 | goto err; | |
671 | ||
672 | nr_freed = 0; | |
673 | for (i = 0; i < spd.nr_pages; i++) { | |
6818173b | 674 | this_len = min_t(size_t, vec[i].iov_len, res); |
35f3d14d JA |
675 | spd.partial[i].offset = 0; |
676 | spd.partial[i].len = this_len; | |
6818173b | 677 | if (!this_len) { |
35f3d14d JA |
678 | __free_page(spd.pages[i]); |
679 | spd.pages[i] = NULL; | |
6818173b MS |
680 | nr_freed++; |
681 | } | |
682 | res -= this_len; | |
683 | } | |
684 | spd.nr_pages -= nr_freed; | |
685 | ||
686 | res = splice_to_pipe(pipe, &spd); | |
687 | if (res > 0) | |
688 | *ppos += res; | |
689 | ||
35f3d14d JA |
690 | shrink_ret: |
691 | if (vec != __vec) | |
692 | kfree(vec); | |
047fe360 | 693 | splice_shrink_spd(&spd); |
6818173b MS |
694 | return res; |
695 | ||
696 | err: | |
4f231228 | 697 | for (i = 0; i < spd.nr_pages; i++) |
35f3d14d | 698 | __free_page(spd.pages[i]); |
4f231228 | 699 | |
35f3d14d JA |
700 | res = error; |
701 | goto shrink_ret; | |
6818173b MS |
702 | } |
703 | EXPORT_SYMBOL(default_file_splice_read); | |
704 | ||
5274f052 | 705 | /* |
4f6f0bd2 | 706 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 707 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 708 | */ |
76ad4d11 | 709 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
710 | struct pipe_buffer *buf, struct splice_desc *sd) |
711 | { | |
6a14b90b | 712 | struct file *file = sd->u.file; |
5274f052 | 713 | loff_t pos = sd->pos; |
a8adbe37 | 714 | int more; |
5274f052 | 715 | |
72c2d531 | 716 | if (!likely(file->f_op->sendpage)) |
a8adbe37 MM |
717 | return -EINVAL; |
718 | ||
35f9c09f | 719 | more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; |
ae62ca7b ED |
720 | |
721 | if (sd->len < sd->total_len && pipe->nrbufs > 1) | |
35f9c09f | 722 | more |= MSG_SENDPAGE_NOTLAST; |
ae62ca7b | 723 | |
a8adbe37 MM |
724 | return file->f_op->sendpage(file, buf->page, buf->offset, |
725 | sd->len, &pos, more); | |
5274f052 JA |
726 | } |
727 | ||
b3c2d2dd MS |
728 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
729 | { | |
730 | smp_mb(); | |
731 | if (waitqueue_active(&pipe->wait)) | |
732 | wake_up_interruptible(&pipe->wait); | |
733 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
734 | } | |
735 | ||
932cc6d4 | 736 | /** |
b3c2d2dd | 737 | * splice_from_pipe_feed - feed available data from a pipe to a file |
932cc6d4 JA |
738 | * @pipe: pipe to splice from |
739 | * @sd: information to @actor | |
740 | * @actor: handler that splices the data | |
741 | * | |
742 | * Description: | |
b3c2d2dd MS |
743 | * This function loops over the pipe and calls @actor to do the |
744 | * actual moving of a single struct pipe_buffer to the desired | |
745 | * destination. It returns when there's no more buffers left in | |
746 | * the pipe or if the requested number of bytes (@sd->total_len) | |
747 | * have been copied. It returns a positive number (one) if the | |
748 | * pipe needs to be filled with more data, zero if the required | |
749 | * number of bytes have been copied and -errno on error. | |
932cc6d4 | 750 | * |
b3c2d2dd MS |
751 | * This, together with splice_from_pipe_{begin,end,next}, may be |
752 | * used to implement the functionality of __splice_from_pipe() when | |
753 | * locking is required around copying the pipe buffers to the | |
754 | * destination. | |
83f9135b | 755 | */ |
96f9bc8f | 756 | static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
b3c2d2dd | 757 | splice_actor *actor) |
5274f052 | 758 | { |
b3c2d2dd | 759 | int ret; |
5274f052 | 760 | |
b3c2d2dd MS |
761 | while (pipe->nrbufs) { |
762 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
763 | const struct pipe_buf_operations *ops = buf->ops; | |
5274f052 | 764 | |
b3c2d2dd MS |
765 | sd->len = buf->len; |
766 | if (sd->len > sd->total_len) | |
767 | sd->len = sd->total_len; | |
5274f052 | 768 | |
a8adbe37 MM |
769 | ret = buf->ops->confirm(pipe, buf); |
770 | if (unlikely(ret)) { | |
b3c2d2dd MS |
771 | if (ret == -ENODATA) |
772 | ret = 0; | |
773 | return ret; | |
774 | } | |
a8adbe37 MM |
775 | |
776 | ret = actor(pipe, buf, sd); | |
777 | if (ret <= 0) | |
778 | return ret; | |
779 | ||
b3c2d2dd MS |
780 | buf->offset += ret; |
781 | buf->len -= ret; | |
782 | ||
783 | sd->num_spliced += ret; | |
784 | sd->len -= ret; | |
785 | sd->pos += ret; | |
786 | sd->total_len -= ret; | |
787 | ||
788 | if (!buf->len) { | |
789 | buf->ops = NULL; | |
790 | ops->release(pipe, buf); | |
35f3d14d | 791 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); |
b3c2d2dd | 792 | pipe->nrbufs--; |
6447a3cf | 793 | if (pipe->files) |
b3c2d2dd MS |
794 | sd->need_wakeup = true; |
795 | } | |
5274f052 | 796 | |
b3c2d2dd MS |
797 | if (!sd->total_len) |
798 | return 0; | |
799 | } | |
5274f052 | 800 | |
b3c2d2dd MS |
801 | return 1; |
802 | } | |
5274f052 | 803 | |
b3c2d2dd MS |
804 | /** |
805 | * splice_from_pipe_next - wait for some data to splice from | |
806 | * @pipe: pipe to splice from | |
807 | * @sd: information about the splice operation | |
808 | * | |
809 | * Description: | |
810 | * This function will wait for some data and return a positive | |
811 | * value (one) if pipe buffers are available. It will return zero | |
812 | * or -errno if no more data needs to be spliced. | |
813 | */ | |
96f9bc8f | 814 | static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd | 815 | { |
c725bfce JK |
816 | /* |
817 | * Check for signal early to make process killable when there are | |
818 | * always buffers available | |
819 | */ | |
820 | if (signal_pending(current)) | |
821 | return -ERESTARTSYS; | |
822 | ||
b3c2d2dd MS |
823 | while (!pipe->nrbufs) { |
824 | if (!pipe->writers) | |
825 | return 0; | |
016b661e | 826 | |
b3c2d2dd MS |
827 | if (!pipe->waiting_writers && sd->num_spliced) |
828 | return 0; | |
73d62d83 | 829 | |
b3c2d2dd MS |
830 | if (sd->flags & SPLICE_F_NONBLOCK) |
831 | return -EAGAIN; | |
5274f052 | 832 | |
b3c2d2dd MS |
833 | if (signal_pending(current)) |
834 | return -ERESTARTSYS; | |
5274f052 | 835 | |
b3c2d2dd MS |
836 | if (sd->need_wakeup) { |
837 | wakeup_pipe_writers(pipe); | |
838 | sd->need_wakeup = false; | |
5274f052 JA |
839 | } |
840 | ||
b3c2d2dd MS |
841 | pipe_wait(pipe); |
842 | } | |
29e35094 | 843 | |
b3c2d2dd MS |
844 | return 1; |
845 | } | |
5274f052 | 846 | |
b3c2d2dd MS |
847 | /** |
848 | * splice_from_pipe_begin - start splicing from pipe | |
b80901bb | 849 | * @sd: information about the splice operation |
b3c2d2dd MS |
850 | * |
851 | * Description: | |
852 | * This function should be called before a loop containing | |
853 | * splice_from_pipe_next() and splice_from_pipe_feed() to | |
854 | * initialize the necessary fields of @sd. | |
855 | */ | |
96f9bc8f | 856 | static void splice_from_pipe_begin(struct splice_desc *sd) |
b3c2d2dd MS |
857 | { |
858 | sd->num_spliced = 0; | |
859 | sd->need_wakeup = false; | |
860 | } | |
5274f052 | 861 | |
b3c2d2dd MS |
862 | /** |
863 | * splice_from_pipe_end - finish splicing from pipe | |
864 | * @pipe: pipe to splice from | |
865 | * @sd: information about the splice operation | |
866 | * | |
867 | * Description: | |
868 | * This function will wake up pipe writers if necessary. It should | |
869 | * be called after a loop containing splice_from_pipe_next() and | |
870 | * splice_from_pipe_feed(). | |
871 | */ | |
96f9bc8f | 872 | static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd MS |
873 | { |
874 | if (sd->need_wakeup) | |
875 | wakeup_pipe_writers(pipe); | |
876 | } | |
5274f052 | 877 | |
b3c2d2dd MS |
878 | /** |
879 | * __splice_from_pipe - splice data from a pipe to given actor | |
880 | * @pipe: pipe to splice from | |
881 | * @sd: information to @actor | |
882 | * @actor: handler that splices the data | |
883 | * | |
884 | * Description: | |
885 | * This function does little more than loop over the pipe and call | |
886 | * @actor to do the actual moving of a single struct pipe_buffer to | |
887 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
888 | * pipe_to_user. | |
889 | * | |
890 | */ | |
891 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, | |
892 | splice_actor *actor) | |
893 | { | |
894 | int ret; | |
5274f052 | 895 | |
b3c2d2dd MS |
896 | splice_from_pipe_begin(sd); |
897 | do { | |
c2489e07 | 898 | cond_resched(); |
b3c2d2dd MS |
899 | ret = splice_from_pipe_next(pipe, sd); |
900 | if (ret > 0) | |
901 | ret = splice_from_pipe_feed(pipe, sd, actor); | |
902 | } while (ret > 0); | |
903 | splice_from_pipe_end(pipe, sd); | |
904 | ||
905 | return sd->num_spliced ? sd->num_spliced : ret; | |
5274f052 | 906 | } |
40bee44e | 907 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 908 | |
932cc6d4 JA |
909 | /** |
910 | * splice_from_pipe - splice data from a pipe to a file | |
911 | * @pipe: pipe to splice from | |
912 | * @out: file to splice to | |
913 | * @ppos: position in @out | |
914 | * @len: how many bytes to splice | |
915 | * @flags: splice modifier flags | |
916 | * @actor: handler that splices the data | |
917 | * | |
918 | * Description: | |
2933970b | 919 | * See __splice_from_pipe. This function locks the pipe inode, |
932cc6d4 JA |
920 | * otherwise it's identical to __splice_from_pipe(). |
921 | * | |
922 | */ | |
6da61809 MF |
923 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
924 | loff_t *ppos, size_t len, unsigned int flags, | |
925 | splice_actor *actor) | |
926 | { | |
927 | ssize_t ret; | |
c66ab6fa JA |
928 | struct splice_desc sd = { |
929 | .total_len = len, | |
930 | .flags = flags, | |
931 | .pos = *ppos, | |
6a14b90b | 932 | .u.file = out, |
c66ab6fa | 933 | }; |
6da61809 | 934 | |
61e0d47c | 935 | pipe_lock(pipe); |
c66ab6fa | 936 | ret = __splice_from_pipe(pipe, &sd, actor); |
61e0d47c | 937 | pipe_unlock(pipe); |
6da61809 MF |
938 | |
939 | return ret; | |
940 | } | |
941 | ||
8d020765 AV |
942 | /** |
943 | * iter_file_splice_write - splice data from a pipe to a file | |
944 | * @pipe: pipe info | |
945 | * @out: file to write to | |
946 | * @ppos: position in @out | |
947 | * @len: number of bytes to splice | |
948 | * @flags: splice modifier flags | |
949 | * | |
950 | * Description: | |
951 | * Will either move or copy pages (determined by @flags options) from | |
952 | * the given pipe inode to the given file. | |
953 | * This one is ->write_iter-based. | |
954 | * | |
955 | */ | |
956 | ssize_t | |
957 | iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
958 | loff_t *ppos, size_t len, unsigned int flags) | |
959 | { | |
960 | struct splice_desc sd = { | |
961 | .total_len = len, | |
962 | .flags = flags, | |
963 | .pos = *ppos, | |
964 | .u.file = out, | |
965 | }; | |
966 | int nbufs = pipe->buffers; | |
967 | struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), | |
968 | GFP_KERNEL); | |
969 | ssize_t ret; | |
970 | ||
971 | if (unlikely(!array)) | |
972 | return -ENOMEM; | |
973 | ||
974 | pipe_lock(pipe); | |
975 | ||
976 | splice_from_pipe_begin(&sd); | |
977 | while (sd.total_len) { | |
978 | struct iov_iter from; | |
8d020765 AV |
979 | size_t left; |
980 | int n, idx; | |
981 | ||
982 | ret = splice_from_pipe_next(pipe, &sd); | |
983 | if (ret <= 0) | |
984 | break; | |
985 | ||
986 | if (unlikely(nbufs < pipe->buffers)) { | |
987 | kfree(array); | |
988 | nbufs = pipe->buffers; | |
989 | array = kcalloc(nbufs, sizeof(struct bio_vec), | |
990 | GFP_KERNEL); | |
991 | if (!array) { | |
992 | ret = -ENOMEM; | |
993 | break; | |
994 | } | |
995 | } | |
996 | ||
997 | /* build the vector */ | |
998 | left = sd.total_len; | |
999 | for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) { | |
1000 | struct pipe_buffer *buf = pipe->bufs + idx; | |
1001 | size_t this_len = buf->len; | |
1002 | ||
1003 | if (this_len > left) | |
1004 | this_len = left; | |
1005 | ||
1006 | if (idx == pipe->buffers - 1) | |
1007 | idx = -1; | |
1008 | ||
1009 | ret = buf->ops->confirm(pipe, buf); | |
1010 | if (unlikely(ret)) { | |
1011 | if (ret == -ENODATA) | |
1012 | ret = 0; | |
1013 | goto done; | |
1014 | } | |
1015 | ||
1016 | array[n].bv_page = buf->page; | |
1017 | array[n].bv_len = this_len; | |
1018 | array[n].bv_offset = buf->offset; | |
1019 | left -= this_len; | |
1020 | } | |
1021 | ||
05afcb77 AV |
1022 | iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n, |
1023 | sd.total_len - left); | |
dbe4e192 | 1024 | ret = vfs_iter_write(out, &from, &sd.pos); |
8d020765 AV |
1025 | if (ret <= 0) |
1026 | break; | |
1027 | ||
1028 | sd.num_spliced += ret; | |
1029 | sd.total_len -= ret; | |
dbe4e192 | 1030 | *ppos = sd.pos; |
8d020765 AV |
1031 | |
1032 | /* dismiss the fully eaten buffers, adjust the partial one */ | |
1033 | while (ret) { | |
1034 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
1035 | if (ret >= buf->len) { | |
1036 | const struct pipe_buf_operations *ops = buf->ops; | |
1037 | ret -= buf->len; | |
1038 | buf->len = 0; | |
1039 | buf->ops = NULL; | |
1040 | ops->release(pipe, buf); | |
1041 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); | |
1042 | pipe->nrbufs--; | |
1043 | if (pipe->files) | |
1044 | sd.need_wakeup = true; | |
1045 | } else { | |
1046 | buf->offset += ret; | |
1047 | buf->len -= ret; | |
1048 | ret = 0; | |
1049 | } | |
1050 | } | |
1051 | } | |
1052 | done: | |
1053 | kfree(array); | |
1054 | splice_from_pipe_end(pipe, &sd); | |
1055 | ||
1056 | pipe_unlock(pipe); | |
1057 | ||
1058 | if (sd.num_spliced) | |
1059 | ret = sd.num_spliced; | |
1060 | ||
1061 | return ret; | |
1062 | } | |
1063 | ||
1064 | EXPORT_SYMBOL(iter_file_splice_write); | |
1065 | ||
b2858d7d MS |
1066 | static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1067 | struct splice_desc *sd) | |
0b0a47f5 | 1068 | { |
b2858d7d MS |
1069 | int ret; |
1070 | void *data; | |
06ae43f3 | 1071 | loff_t tmp = sd->pos; |
b2858d7d | 1072 | |
fbb32750 | 1073 | data = kmap(buf->page); |
06ae43f3 | 1074 | ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); |
fbb32750 | 1075 | kunmap(buf->page); |
b2858d7d MS |
1076 | |
1077 | return ret; | |
0b0a47f5 MS |
1078 | } |
1079 | ||
1080 | static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, | |
1081 | struct file *out, loff_t *ppos, | |
1082 | size_t len, unsigned int flags) | |
1083 | { | |
b2858d7d | 1084 | ssize_t ret; |
0b0a47f5 | 1085 | |
b2858d7d MS |
1086 | ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); |
1087 | if (ret > 0) | |
1088 | *ppos += ret; | |
0b0a47f5 | 1089 | |
b2858d7d | 1090 | return ret; |
0b0a47f5 MS |
1091 | } |
1092 | ||
83f9135b JA |
1093 | /** |
1094 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 1095 | * @pipe: pipe to splice from |
83f9135b | 1096 | * @out: socket to write to |
932cc6d4 | 1097 | * @ppos: position in @out |
83f9135b JA |
1098 | * @len: number of bytes to splice |
1099 | * @flags: splice modifier flags | |
1100 | * | |
932cc6d4 JA |
1101 | * Description: |
1102 | * Will send @len bytes from the pipe to a network socket. No data copying | |
1103 | * is involved. | |
83f9135b JA |
1104 | * |
1105 | */ | |
3a326a2c | 1106 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 1107 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 1108 | { |
00522fb4 | 1109 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
1110 | } |
1111 | ||
059a8f37 | 1112 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 1113 | |
83f9135b JA |
1114 | /* |
1115 | * Attempt to initiate a splice from pipe to file. | |
1116 | */ | |
dee5220c TG |
1117 | long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
1118 | loff_t *ppos, size_t len, unsigned int flags) | |
5274f052 | 1119 | { |
0b0a47f5 MS |
1120 | ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, |
1121 | loff_t *, size_t, unsigned int); | |
5274f052 | 1122 | |
72c2d531 | 1123 | if (out->f_op->splice_write) |
cc56f7de CG |
1124 | splice_write = out->f_op->splice_write; |
1125 | else | |
0b0a47f5 MS |
1126 | splice_write = default_file_splice_write; |
1127 | ||
500368f7 | 1128 | return splice_write(pipe, out, ppos, len, flags); |
5274f052 | 1129 | } |
dee5220c | 1130 | EXPORT_SYMBOL(do_splice_from); |
5274f052 | 1131 | |
83f9135b JA |
1132 | /* |
1133 | * Attempt to initiate a splice from a file to a pipe. | |
1134 | */ | |
dee5220c TG |
1135 | long do_splice_to(struct file *in, loff_t *ppos, |
1136 | struct pipe_inode_info *pipe, size_t len, | |
1137 | unsigned int flags) | |
5274f052 | 1138 | { |
6818173b MS |
1139 | ssize_t (*splice_read)(struct file *, loff_t *, |
1140 | struct pipe_inode_info *, size_t, unsigned int); | |
5274f052 JA |
1141 | int ret; |
1142 | ||
49570e9b | 1143 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
1144 | return -EBADF; |
1145 | ||
cbb7e577 | 1146 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
1147 | if (unlikely(ret < 0)) |
1148 | return ret; | |
1149 | ||
72c2d531 | 1150 | if (in->f_op->splice_read) |
cc56f7de CG |
1151 | splice_read = in->f_op->splice_read; |
1152 | else | |
6818173b MS |
1153 | splice_read = default_file_splice_read; |
1154 | ||
1155 | return splice_read(in, ppos, pipe, len, flags); | |
5274f052 | 1156 | } |
dee5220c | 1157 | EXPORT_SYMBOL(do_splice_to); |
5274f052 | 1158 | |
932cc6d4 JA |
1159 | /** |
1160 | * splice_direct_to_actor - splices data directly between two non-pipes | |
1161 | * @in: file to splice from | |
1162 | * @sd: actor information on where to splice to | |
1163 | * @actor: handles the data splicing | |
1164 | * | |
1165 | * Description: | |
1166 | * This is a special case helper to splice directly between two | |
1167 | * points, without requiring an explicit pipe. Internally an allocated | |
79685b8d | 1168 | * pipe is cached in the process, and reused during the lifetime of |
932cc6d4 JA |
1169 | * that process. |
1170 | * | |
c66ab6fa JA |
1171 | */ |
1172 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
1173 | splice_direct_actor *actor) | |
b92ce558 JA |
1174 | { |
1175 | struct pipe_inode_info *pipe; | |
1176 | long ret, bytes; | |
1177 | umode_t i_mode; | |
c66ab6fa | 1178 | size_t len; |
0ff28d9f | 1179 | int i, flags, more; |
b92ce558 JA |
1180 | |
1181 | /* | |
1182 | * We require the input being a regular file, as we don't want to | |
1183 | * randomly drop data for eg socket -> socket splicing. Use the | |
1184 | * piped splicing for that! | |
1185 | */ | |
496ad9aa | 1186 | i_mode = file_inode(in)->i_mode; |
b92ce558 JA |
1187 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
1188 | return -EINVAL; | |
1189 | ||
1190 | /* | |
1191 | * neither in nor out is a pipe, setup an internal pipe attached to | |
1192 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
1193 | */ | |
1194 | pipe = current->splice_pipe; | |
49570e9b | 1195 | if (unlikely(!pipe)) { |
7bee130e | 1196 | pipe = alloc_pipe_info(); |
b92ce558 JA |
1197 | if (!pipe) |
1198 | return -ENOMEM; | |
1199 | ||
1200 | /* | |
1201 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 1202 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
1203 | * PIPE_READERS appropriately. |
1204 | */ | |
1205 | pipe->readers = 1; | |
1206 | ||
1207 | current->splice_pipe = pipe; | |
1208 | } | |
1209 | ||
1210 | /* | |
73d62d83 | 1211 | * Do the splice. |
b92ce558 JA |
1212 | */ |
1213 | ret = 0; | |
1214 | bytes = 0; | |
c66ab6fa JA |
1215 | len = sd->total_len; |
1216 | flags = sd->flags; | |
1217 | ||
1218 | /* | |
1219 | * Don't block on output, we have to drain the direct pipe. | |
1220 | */ | |
1221 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
0ff28d9f | 1222 | more = sd->flags & SPLICE_F_MORE; |
b92ce558 JA |
1223 | |
1224 | while (len) { | |
51a92c0f | 1225 | size_t read_len; |
a82c53a0 | 1226 | loff_t pos = sd->pos, prev_pos = pos; |
b92ce558 | 1227 | |
bcd4f3ac | 1228 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 1229 | if (unlikely(ret <= 0)) |
b92ce558 JA |
1230 | goto out_release; |
1231 | ||
1232 | read_len = ret; | |
c66ab6fa | 1233 | sd->total_len = read_len; |
b92ce558 | 1234 | |
0ff28d9f CL |
1235 | /* |
1236 | * If more data is pending, set SPLICE_F_MORE | |
1237 | * If this is the last data and SPLICE_F_MORE was not set | |
1238 | * initially, clears it. | |
1239 | */ | |
1240 | if (read_len < len) | |
1241 | sd->flags |= SPLICE_F_MORE; | |
1242 | else if (!more) | |
1243 | sd->flags &= ~SPLICE_F_MORE; | |
b92ce558 JA |
1244 | /* |
1245 | * NOTE: nonblocking mode only applies to the input. We | |
1246 | * must not do the output in nonblocking mode as then we | |
1247 | * could get stuck data in the internal pipe: | |
1248 | */ | |
c66ab6fa | 1249 | ret = actor(pipe, sd); |
a82c53a0 TZ |
1250 | if (unlikely(ret <= 0)) { |
1251 | sd->pos = prev_pos; | |
b92ce558 | 1252 | goto out_release; |
a82c53a0 | 1253 | } |
b92ce558 JA |
1254 | |
1255 | bytes += ret; | |
1256 | len -= ret; | |
bcd4f3ac | 1257 | sd->pos = pos; |
b92ce558 | 1258 | |
a82c53a0 TZ |
1259 | if (ret < read_len) { |
1260 | sd->pos = prev_pos + ret; | |
51a92c0f | 1261 | goto out_release; |
a82c53a0 | 1262 | } |
b92ce558 JA |
1263 | } |
1264 | ||
9e97198d | 1265 | done: |
b92ce558 | 1266 | pipe->nrbufs = pipe->curbuf = 0; |
80848708 | 1267 | file_accessed(in); |
b92ce558 JA |
1268 | return bytes; |
1269 | ||
1270 | out_release: | |
1271 | /* | |
1272 | * If we did an incomplete transfer we must release | |
1273 | * the pipe buffers in question: | |
1274 | */ | |
35f3d14d | 1275 | for (i = 0; i < pipe->buffers; i++) { |
b92ce558 JA |
1276 | struct pipe_buffer *buf = pipe->bufs + i; |
1277 | ||
1278 | if (buf->ops) { | |
1279 | buf->ops->release(pipe, buf); | |
1280 | buf->ops = NULL; | |
1281 | } | |
1282 | } | |
b92ce558 | 1283 | |
9e97198d JA |
1284 | if (!bytes) |
1285 | bytes = ret; | |
c66ab6fa | 1286 | |
9e97198d | 1287 | goto done; |
c66ab6fa JA |
1288 | } |
1289 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1290 | ||
1291 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1292 | struct splice_desc *sd) | |
1293 | { | |
6a14b90b | 1294 | struct file *file = sd->u.file; |
c66ab6fa | 1295 | |
7995bd28 | 1296 | return do_splice_from(pipe, file, sd->opos, sd->total_len, |
2cb4b05e | 1297 | sd->flags); |
c66ab6fa JA |
1298 | } |
1299 | ||
932cc6d4 JA |
1300 | /** |
1301 | * do_splice_direct - splices data directly between two files | |
1302 | * @in: file to splice from | |
1303 | * @ppos: input file offset | |
1304 | * @out: file to splice to | |
acdb37c3 | 1305 | * @opos: output file offset |
932cc6d4 JA |
1306 | * @len: number of bytes to splice |
1307 | * @flags: splice modifier flags | |
1308 | * | |
1309 | * Description: | |
1310 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1311 | * doing it in the application would incur an extra system call | |
1312 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1313 | * can splice directly through a process-private pipe. | |
1314 | * | |
1315 | */ | |
c66ab6fa | 1316 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
7995bd28 | 1317 | loff_t *opos, size_t len, unsigned int flags) |
c66ab6fa JA |
1318 | { |
1319 | struct splice_desc sd = { | |
1320 | .len = len, | |
1321 | .total_len = len, | |
1322 | .flags = flags, | |
1323 | .pos = *ppos, | |
6a14b90b | 1324 | .u.file = out, |
7995bd28 | 1325 | .opos = opos, |
c66ab6fa | 1326 | }; |
51a92c0f | 1327 | long ret; |
c66ab6fa | 1328 | |
18c67cb9 AV |
1329 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1330 | return -EBADF; | |
1331 | ||
1332 | if (unlikely(out->f_flags & O_APPEND)) | |
1333 | return -EINVAL; | |
1334 | ||
1335 | ret = rw_verify_area(WRITE, out, opos, len); | |
1336 | if (unlikely(ret < 0)) | |
1337 | return ret; | |
1338 | ||
c66ab6fa | 1339 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); |
51a92c0f | 1340 | if (ret > 0) |
a82c53a0 | 1341 | *ppos = sd.pos; |
51a92c0f | 1342 | |
c66ab6fa | 1343 | return ret; |
b92ce558 | 1344 | } |
1c118596 | 1345 | EXPORT_SYMBOL(do_splice_direct); |
b92ce558 | 1346 | |
7c77f0b3 MS |
1347 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
1348 | struct pipe_inode_info *opipe, | |
1349 | size_t len, unsigned int flags); | |
ddac0d39 | 1350 | |
83f9135b JA |
1351 | /* |
1352 | * Determine where to splice to/from. | |
1353 | */ | |
529565dc IM |
1354 | static long do_splice(struct file *in, loff_t __user *off_in, |
1355 | struct file *out, loff_t __user *off_out, | |
1356 | size_t len, unsigned int flags) | |
5274f052 | 1357 | { |
7c77f0b3 MS |
1358 | struct pipe_inode_info *ipipe; |
1359 | struct pipe_inode_info *opipe; | |
7995bd28 | 1360 | loff_t offset; |
a4514ebd | 1361 | long ret; |
5274f052 | 1362 | |
71993e62 LT |
1363 | ipipe = get_pipe_info(in); |
1364 | opipe = get_pipe_info(out); | |
7c77f0b3 MS |
1365 | |
1366 | if (ipipe && opipe) { | |
1367 | if (off_in || off_out) | |
1368 | return -ESPIPE; | |
1369 | ||
1370 | if (!(in->f_mode & FMODE_READ)) | |
1371 | return -EBADF; | |
1372 | ||
1373 | if (!(out->f_mode & FMODE_WRITE)) | |
1374 | return -EBADF; | |
1375 | ||
1376 | /* Splicing to self would be fun, but... */ | |
1377 | if (ipipe == opipe) | |
1378 | return -EINVAL; | |
1379 | ||
1380 | return splice_pipe_to_pipe(ipipe, opipe, len, flags); | |
1381 | } | |
1382 | ||
1383 | if (ipipe) { | |
529565dc IM |
1384 | if (off_in) |
1385 | return -ESPIPE; | |
b92ce558 | 1386 | if (off_out) { |
19c9a49b | 1387 | if (!(out->f_mode & FMODE_PWRITE)) |
b92ce558 | 1388 | return -EINVAL; |
cbb7e577 | 1389 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1390 | return -EFAULT; |
7995bd28 AV |
1391 | } else { |
1392 | offset = out->f_pos; | |
1393 | } | |
529565dc | 1394 | |
18c67cb9 AV |
1395 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1396 | return -EBADF; | |
1397 | ||
1398 | if (unlikely(out->f_flags & O_APPEND)) | |
1399 | return -EINVAL; | |
1400 | ||
1401 | ret = rw_verify_area(WRITE, out, &offset, len); | |
1402 | if (unlikely(ret < 0)) | |
1403 | return ret; | |
1404 | ||
500368f7 | 1405 | file_start_write(out); |
7995bd28 | 1406 | ret = do_splice_from(ipipe, out, &offset, len, flags); |
500368f7 | 1407 | file_end_write(out); |
a4514ebd | 1408 | |
7995bd28 AV |
1409 | if (!off_out) |
1410 | out->f_pos = offset; | |
1411 | else if (copy_to_user(off_out, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1412 | ret = -EFAULT; |
1413 | ||
1414 | return ret; | |
529565dc | 1415 | } |
5274f052 | 1416 | |
7c77f0b3 | 1417 | if (opipe) { |
529565dc IM |
1418 | if (off_out) |
1419 | return -ESPIPE; | |
b92ce558 | 1420 | if (off_in) { |
19c9a49b | 1421 | if (!(in->f_mode & FMODE_PREAD)) |
b92ce558 | 1422 | return -EINVAL; |
cbb7e577 | 1423 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1424 | return -EFAULT; |
7995bd28 AV |
1425 | } else { |
1426 | offset = in->f_pos; | |
1427 | } | |
529565dc | 1428 | |
7995bd28 | 1429 | ret = do_splice_to(in, &offset, opipe, len, flags); |
a4514ebd | 1430 | |
7995bd28 AV |
1431 | if (!off_in) |
1432 | in->f_pos = offset; | |
1433 | else if (copy_to_user(off_in, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1434 | ret = -EFAULT; |
1435 | ||
1436 | return ret; | |
529565dc | 1437 | } |
5274f052 JA |
1438 | |
1439 | return -EINVAL; | |
1440 | } | |
1441 | ||
912d35f8 JA |
1442 | /* |
1443 | * Map an iov into an array of pages and offset/length tupples. With the | |
1444 | * partial_page structure, we can map several non-contiguous ranges into | |
1445 | * our ones pages[] map instead of splitting that operation into pieces. | |
1446 | * Could easily be exported as a generic helper for other users, in which | |
1447 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1448 | */ | |
1449 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1450 | unsigned int nr_vecs, struct page **pages, | |
bd1a68b5 | 1451 | struct partial_page *partial, bool aligned, |
35f3d14d | 1452 | unsigned int pipe_buffers) |
912d35f8 JA |
1453 | { |
1454 | int buffers = 0, error = 0; | |
1455 | ||
912d35f8 JA |
1456 | while (nr_vecs) { |
1457 | unsigned long off, npages; | |
75723957 | 1458 | struct iovec entry; |
912d35f8 JA |
1459 | void __user *base; |
1460 | size_t len; | |
1461 | int i; | |
1462 | ||
75723957 | 1463 | error = -EFAULT; |
bc40d73c | 1464 | if (copy_from_user(&entry, iov, sizeof(entry))) |
912d35f8 JA |
1465 | break; |
1466 | ||
75723957 LT |
1467 | base = entry.iov_base; |
1468 | len = entry.iov_len; | |
1469 | ||
912d35f8 JA |
1470 | /* |
1471 | * Sanity check this iovec. 0 read succeeds. | |
1472 | */ | |
75723957 | 1473 | error = 0; |
912d35f8 JA |
1474 | if (unlikely(!len)) |
1475 | break; | |
1476 | error = -EFAULT; | |
712a30e6 | 1477 | if (!access_ok(VERIFY_READ, base, len)) |
912d35f8 JA |
1478 | break; |
1479 | ||
1480 | /* | |
1481 | * Get this base offset and number of pages, then map | |
1482 | * in the user pages. | |
1483 | */ | |
1484 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1485 | |
1486 | /* | |
1487 | * If asked for alignment, the offset must be zero and the | |
1488 | * length a multiple of the PAGE_SIZE. | |
1489 | */ | |
1490 | error = -EINVAL; | |
1491 | if (aligned && (off || len & ~PAGE_MASK)) | |
1492 | break; | |
1493 | ||
912d35f8 | 1494 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
35f3d14d JA |
1495 | if (npages > pipe_buffers - buffers) |
1496 | npages = pipe_buffers - buffers; | |
912d35f8 | 1497 | |
bc40d73c NP |
1498 | error = get_user_pages_fast((unsigned long)base, npages, |
1499 | 0, &pages[buffers]); | |
912d35f8 JA |
1500 | |
1501 | if (unlikely(error <= 0)) | |
1502 | break; | |
1503 | ||
1504 | /* | |
1505 | * Fill this contiguous range into the partial page map. | |
1506 | */ | |
1507 | for (i = 0; i < error; i++) { | |
7591489a | 1508 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1509 | |
1510 | partial[buffers].offset = off; | |
1511 | partial[buffers].len = plen; | |
1512 | ||
1513 | off = 0; | |
1514 | len -= plen; | |
1515 | buffers++; | |
1516 | } | |
1517 | ||
1518 | /* | |
1519 | * We didn't complete this iov, stop here since it probably | |
1520 | * means we have to move some of this into a pipe to | |
1521 | * be able to continue. | |
1522 | */ | |
1523 | if (len) | |
1524 | break; | |
1525 | ||
1526 | /* | |
1527 | * Don't continue if we mapped fewer pages than we asked for, | |
1528 | * or if we mapped the max number of pages that we have | |
1529 | * room for. | |
1530 | */ | |
35f3d14d | 1531 | if (error < npages || buffers == pipe_buffers) |
912d35f8 JA |
1532 | break; |
1533 | ||
1534 | nr_vecs--; | |
1535 | iov++; | |
1536 | } | |
1537 | ||
912d35f8 JA |
1538 | if (buffers) |
1539 | return buffers; | |
1540 | ||
1541 | return error; | |
1542 | } | |
1543 | ||
6a14b90b JA |
1544 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1545 | struct splice_desc *sd) | |
1546 | { | |
6130f531 AV |
1547 | int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); |
1548 | return n == sd->len ? n : -EFAULT; | |
6a14b90b JA |
1549 | } |
1550 | ||
1551 | /* | |
1552 | * For lack of a better implementation, implement vmsplice() to userspace | |
1553 | * as a simple copy of the pipes pages to the user iov. | |
1554 | */ | |
6130f531 | 1555 | static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov, |
6a14b90b JA |
1556 | unsigned long nr_segs, unsigned int flags) |
1557 | { | |
1558 | struct pipe_inode_info *pipe; | |
1559 | struct splice_desc sd; | |
6a14b90b | 1560 | long ret; |
6130f531 AV |
1561 | struct iovec iovstack[UIO_FASTIOV]; |
1562 | struct iovec *iov = iovstack; | |
1563 | struct iov_iter iter; | |
6a14b90b | 1564 | |
71993e62 | 1565 | pipe = get_pipe_info(file); |
6a14b90b JA |
1566 | if (!pipe) |
1567 | return -EBADF; | |
1568 | ||
345995fa AV |
1569 | ret = import_iovec(READ, uiov, nr_segs, |
1570 | ARRAY_SIZE(iovstack), &iov, &iter); | |
1571 | if (ret < 0) | |
1572 | return ret; | |
6a14b90b | 1573 | |
345995fa | 1574 | sd.total_len = iov_iter_count(&iter); |
6130f531 | 1575 | sd.len = 0; |
6130f531 AV |
1576 | sd.flags = flags; |
1577 | sd.u.data = &iter; | |
1578 | sd.pos = 0; | |
6a14b90b | 1579 | |
345995fa AV |
1580 | if (sd.total_len) { |
1581 | pipe_lock(pipe); | |
1582 | ret = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1583 | pipe_unlock(pipe); | |
1584 | } | |
6a14b90b | 1585 | |
345995fa | 1586 | kfree(iov); |
6a14b90b JA |
1587 | return ret; |
1588 | } | |
1589 | ||
912d35f8 JA |
1590 | /* |
1591 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1592 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1593 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1594 | */ |
6a14b90b JA |
1595 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *iov, |
1596 | unsigned long nr_segs, unsigned int flags) | |
912d35f8 | 1597 | { |
ddac0d39 | 1598 | struct pipe_inode_info *pipe; |
35f3d14d JA |
1599 | struct page *pages[PIPE_DEF_BUFFERS]; |
1600 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
912d35f8 JA |
1601 | struct splice_pipe_desc spd = { |
1602 | .pages = pages, | |
1603 | .partial = partial, | |
047fe360 | 1604 | .nr_pages_max = PIPE_DEF_BUFFERS, |
912d35f8 JA |
1605 | .flags = flags, |
1606 | .ops = &user_page_pipe_buf_ops, | |
bbdfc2f7 | 1607 | .spd_release = spd_release_page, |
912d35f8 | 1608 | }; |
35f3d14d | 1609 | long ret; |
912d35f8 | 1610 | |
71993e62 | 1611 | pipe = get_pipe_info(file); |
ddac0d39 | 1612 | if (!pipe) |
912d35f8 | 1613 | return -EBADF; |
912d35f8 | 1614 | |
35f3d14d JA |
1615 | if (splice_grow_spd(pipe, &spd)) |
1616 | return -ENOMEM; | |
1617 | ||
1618 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, spd.pages, | |
bd1a68b5 | 1619 | spd.partial, false, |
047fe360 | 1620 | spd.nr_pages_max); |
912d35f8 | 1621 | if (spd.nr_pages <= 0) |
35f3d14d JA |
1622 | ret = spd.nr_pages; |
1623 | else | |
1624 | ret = splice_to_pipe(pipe, &spd); | |
912d35f8 | 1625 | |
047fe360 | 1626 | splice_shrink_spd(&spd); |
35f3d14d | 1627 | return ret; |
912d35f8 JA |
1628 | } |
1629 | ||
6a14b90b JA |
1630 | /* |
1631 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1632 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1633 | * operation that can be supported without any funky alignment restrictions | |
1634 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1635 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1636 | * solutions for that: | |
1637 | * | |
1638 | * - memcpy() the data internally, at which point we might as well just | |
1639 | * do a regular read() on the buffer anyway. | |
1640 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1641 | * has restriction limitations on both ends of the pipe). | |
1642 | * | |
1643 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1644 | * | |
1645 | */ | |
836f92ad HC |
1646 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, |
1647 | unsigned long, nr_segs, unsigned int, flags) | |
912d35f8 | 1648 | { |
2903ff01 | 1649 | struct fd f; |
912d35f8 | 1650 | long error; |
912d35f8 | 1651 | |
6a14b90b JA |
1652 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1653 | return -EINVAL; | |
1654 | else if (unlikely(!nr_segs)) | |
1655 | return 0; | |
1656 | ||
912d35f8 | 1657 | error = -EBADF; |
2903ff01 AV |
1658 | f = fdget(fd); |
1659 | if (f.file) { | |
1660 | if (f.file->f_mode & FMODE_WRITE) | |
1661 | error = vmsplice_to_pipe(f.file, iov, nr_segs, flags); | |
1662 | else if (f.file->f_mode & FMODE_READ) | |
1663 | error = vmsplice_to_user(f.file, iov, nr_segs, flags); | |
1664 | ||
1665 | fdput(f); | |
912d35f8 JA |
1666 | } |
1667 | ||
1668 | return error; | |
1669 | } | |
1670 | ||
76b021d0 AV |
1671 | #ifdef CONFIG_COMPAT |
1672 | COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, | |
1673 | unsigned int, nr_segs, unsigned int, flags) | |
1674 | { | |
1675 | unsigned i; | |
1676 | struct iovec __user *iov; | |
1677 | if (nr_segs > UIO_MAXIOV) | |
1678 | return -EINVAL; | |
1679 | iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec)); | |
1680 | for (i = 0; i < nr_segs; i++) { | |
1681 | struct compat_iovec v; | |
1682 | if (get_user(v.iov_base, &iov32[i].iov_base) || | |
1683 | get_user(v.iov_len, &iov32[i].iov_len) || | |
1684 | put_user(compat_ptr(v.iov_base), &iov[i].iov_base) || | |
1685 | put_user(v.iov_len, &iov[i].iov_len)) | |
1686 | return -EFAULT; | |
1687 | } | |
1688 | return sys_vmsplice(fd, iov, nr_segs, flags); | |
1689 | } | |
1690 | #endif | |
1691 | ||
836f92ad HC |
1692 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
1693 | int, fd_out, loff_t __user *, off_out, | |
1694 | size_t, len, unsigned int, flags) | |
5274f052 | 1695 | { |
2903ff01 | 1696 | struct fd in, out; |
5274f052 | 1697 | long error; |
5274f052 JA |
1698 | |
1699 | if (unlikely(!len)) | |
1700 | return 0; | |
1701 | ||
1702 | error = -EBADF; | |
2903ff01 AV |
1703 | in = fdget(fd_in); |
1704 | if (in.file) { | |
1705 | if (in.file->f_mode & FMODE_READ) { | |
1706 | out = fdget(fd_out); | |
1707 | if (out.file) { | |
1708 | if (out.file->f_mode & FMODE_WRITE) | |
1709 | error = do_splice(in.file, off_in, | |
1710 | out.file, off_out, | |
529565dc | 1711 | len, flags); |
2903ff01 | 1712 | fdput(out); |
5274f052 JA |
1713 | } |
1714 | } | |
2903ff01 | 1715 | fdput(in); |
5274f052 | 1716 | } |
5274f052 JA |
1717 | return error; |
1718 | } | |
70524490 | 1719 | |
aadd06e5 JA |
1720 | /* |
1721 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1722 | * return an appropriate error. | |
1723 | */ | |
7c77f0b3 | 1724 | static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1725 | { |
1726 | int ret; | |
1727 | ||
1728 | /* | |
1729 | * Check ->nrbufs without the inode lock first. This function | |
1730 | * is speculative anyways, so missing one is ok. | |
1731 | */ | |
1732 | if (pipe->nrbufs) | |
1733 | return 0; | |
1734 | ||
1735 | ret = 0; | |
61e0d47c | 1736 | pipe_lock(pipe); |
aadd06e5 JA |
1737 | |
1738 | while (!pipe->nrbufs) { | |
1739 | if (signal_pending(current)) { | |
1740 | ret = -ERESTARTSYS; | |
1741 | break; | |
1742 | } | |
1743 | if (!pipe->writers) | |
1744 | break; | |
1745 | if (!pipe->waiting_writers) { | |
1746 | if (flags & SPLICE_F_NONBLOCK) { | |
1747 | ret = -EAGAIN; | |
1748 | break; | |
1749 | } | |
1750 | } | |
1751 | pipe_wait(pipe); | |
1752 | } | |
1753 | ||
61e0d47c | 1754 | pipe_unlock(pipe); |
aadd06e5 JA |
1755 | return ret; |
1756 | } | |
1757 | ||
1758 | /* | |
1759 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1760 | * return an appropriate error. | |
1761 | */ | |
7c77f0b3 | 1762 | static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1763 | { |
1764 | int ret; | |
1765 | ||
1766 | /* | |
1767 | * Check ->nrbufs without the inode lock first. This function | |
1768 | * is speculative anyways, so missing one is ok. | |
1769 | */ | |
35f3d14d | 1770 | if (pipe->nrbufs < pipe->buffers) |
aadd06e5 JA |
1771 | return 0; |
1772 | ||
1773 | ret = 0; | |
61e0d47c | 1774 | pipe_lock(pipe); |
aadd06e5 | 1775 | |
35f3d14d | 1776 | while (pipe->nrbufs >= pipe->buffers) { |
aadd06e5 JA |
1777 | if (!pipe->readers) { |
1778 | send_sig(SIGPIPE, current, 0); | |
1779 | ret = -EPIPE; | |
1780 | break; | |
1781 | } | |
1782 | if (flags & SPLICE_F_NONBLOCK) { | |
1783 | ret = -EAGAIN; | |
1784 | break; | |
1785 | } | |
1786 | if (signal_pending(current)) { | |
1787 | ret = -ERESTARTSYS; | |
1788 | break; | |
1789 | } | |
1790 | pipe->waiting_writers++; | |
1791 | pipe_wait(pipe); | |
1792 | pipe->waiting_writers--; | |
1793 | } | |
1794 | ||
61e0d47c | 1795 | pipe_unlock(pipe); |
aadd06e5 JA |
1796 | return ret; |
1797 | } | |
1798 | ||
7c77f0b3 MS |
1799 | /* |
1800 | * Splice contents of ipipe to opipe. | |
1801 | */ | |
1802 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, | |
1803 | struct pipe_inode_info *opipe, | |
1804 | size_t len, unsigned int flags) | |
1805 | { | |
1806 | struct pipe_buffer *ibuf, *obuf; | |
1807 | int ret = 0, nbuf; | |
1808 | bool input_wakeup = false; | |
1809 | ||
1810 | ||
1811 | retry: | |
1812 | ret = ipipe_prep(ipipe, flags); | |
1813 | if (ret) | |
1814 | return ret; | |
1815 | ||
1816 | ret = opipe_prep(opipe, flags); | |
1817 | if (ret) | |
1818 | return ret; | |
1819 | ||
1820 | /* | |
1821 | * Potential ABBA deadlock, work around it by ordering lock | |
1822 | * grabbing by pipe info address. Otherwise two different processes | |
1823 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1824 | */ | |
1825 | pipe_double_lock(ipipe, opipe); | |
1826 | ||
1827 | do { | |
1828 | if (!opipe->readers) { | |
1829 | send_sig(SIGPIPE, current, 0); | |
1830 | if (!ret) | |
1831 | ret = -EPIPE; | |
1832 | break; | |
1833 | } | |
1834 | ||
1835 | if (!ipipe->nrbufs && !ipipe->writers) | |
1836 | break; | |
1837 | ||
1838 | /* | |
1839 | * Cannot make any progress, because either the input | |
1840 | * pipe is empty or the output pipe is full. | |
1841 | */ | |
35f3d14d | 1842 | if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { |
7c77f0b3 MS |
1843 | /* Already processed some buffers, break */ |
1844 | if (ret) | |
1845 | break; | |
1846 | ||
1847 | if (flags & SPLICE_F_NONBLOCK) { | |
1848 | ret = -EAGAIN; | |
1849 | break; | |
1850 | } | |
1851 | ||
1852 | /* | |
1853 | * We raced with another reader/writer and haven't | |
1854 | * managed to process any buffers. A zero return | |
1855 | * value means EOF, so retry instead. | |
1856 | */ | |
1857 | pipe_unlock(ipipe); | |
1858 | pipe_unlock(opipe); | |
1859 | goto retry; | |
1860 | } | |
1861 | ||
1862 | ibuf = ipipe->bufs + ipipe->curbuf; | |
35f3d14d | 1863 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); |
7c77f0b3 MS |
1864 | obuf = opipe->bufs + nbuf; |
1865 | ||
1866 | if (len >= ibuf->len) { | |
1867 | /* | |
1868 | * Simply move the whole buffer from ipipe to opipe | |
1869 | */ | |
1870 | *obuf = *ibuf; | |
1871 | ibuf->ops = NULL; | |
1872 | opipe->nrbufs++; | |
35f3d14d | 1873 | ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); |
7c77f0b3 MS |
1874 | ipipe->nrbufs--; |
1875 | input_wakeup = true; | |
1876 | } else { | |
1877 | /* | |
1878 | * Get a reference to this pipe buffer, | |
1879 | * so we can copy the contents over. | |
1880 | */ | |
1881 | ibuf->ops->get(ipipe, ibuf); | |
1882 | *obuf = *ibuf; | |
1883 | ||
1884 | /* | |
1885 | * Don't inherit the gift flag, we need to | |
1886 | * prevent multiple steals of this page. | |
1887 | */ | |
1888 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1889 | ||
1890 | obuf->len = len; | |
1891 | opipe->nrbufs++; | |
1892 | ibuf->offset += obuf->len; | |
1893 | ibuf->len -= obuf->len; | |
1894 | } | |
1895 | ret += obuf->len; | |
1896 | len -= obuf->len; | |
1897 | } while (len); | |
1898 | ||
1899 | pipe_unlock(ipipe); | |
1900 | pipe_unlock(opipe); | |
1901 | ||
1902 | /* | |
1903 | * If we put data in the output pipe, wakeup any potential readers. | |
1904 | */ | |
825cdcb1 NK |
1905 | if (ret > 0) |
1906 | wakeup_pipe_readers(opipe); | |
1907 | ||
7c77f0b3 MS |
1908 | if (input_wakeup) |
1909 | wakeup_pipe_writers(ipipe); | |
1910 | ||
1911 | return ret; | |
1912 | } | |
1913 | ||
70524490 JA |
1914 | /* |
1915 | * Link contents of ipipe to opipe. | |
1916 | */ | |
1917 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1918 | struct pipe_inode_info *opipe, | |
1919 | size_t len, unsigned int flags) | |
1920 | { | |
1921 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1922 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1923 | |
1924 | /* | |
1925 | * Potential ABBA deadlock, work around it by ordering lock | |
61e0d47c | 1926 | * grabbing by pipe info address. Otherwise two different processes |
70524490 JA |
1927 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
1928 | */ | |
61e0d47c | 1929 | pipe_double_lock(ipipe, opipe); |
70524490 | 1930 | |
aadd06e5 | 1931 | do { |
70524490 JA |
1932 | if (!opipe->readers) { |
1933 | send_sig(SIGPIPE, current, 0); | |
1934 | if (!ret) | |
1935 | ret = -EPIPE; | |
1936 | break; | |
1937 | } | |
70524490 | 1938 | |
aadd06e5 JA |
1939 | /* |
1940 | * If we have iterated all input buffers or ran out of | |
1941 | * output room, break. | |
1942 | */ | |
35f3d14d | 1943 | if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) |
aadd06e5 | 1944 | break; |
70524490 | 1945 | |
35f3d14d JA |
1946 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); |
1947 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); | |
70524490 JA |
1948 | |
1949 | /* | |
aadd06e5 JA |
1950 | * Get a reference to this pipe buffer, |
1951 | * so we can copy the contents over. | |
70524490 | 1952 | */ |
aadd06e5 JA |
1953 | ibuf->ops->get(ipipe, ibuf); |
1954 | ||
1955 | obuf = opipe->bufs + nbuf; | |
1956 | *obuf = *ibuf; | |
1957 | ||
2a27250e | 1958 | /* |
aadd06e5 JA |
1959 | * Don't inherit the gift flag, we need to |
1960 | * prevent multiple steals of this page. | |
2a27250e | 1961 | */ |
aadd06e5 | 1962 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1963 | |
aadd06e5 JA |
1964 | if (obuf->len > len) |
1965 | obuf->len = len; | |
70524490 | 1966 | |
aadd06e5 JA |
1967 | opipe->nrbufs++; |
1968 | ret += obuf->len; | |
1969 | len -= obuf->len; | |
1970 | i++; | |
1971 | } while (len); | |
70524490 | 1972 | |
02cf01ae JA |
1973 | /* |
1974 | * return EAGAIN if we have the potential of some data in the | |
1975 | * future, otherwise just return 0 | |
1976 | */ | |
1977 | if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) | |
1978 | ret = -EAGAIN; | |
1979 | ||
61e0d47c MS |
1980 | pipe_unlock(ipipe); |
1981 | pipe_unlock(opipe); | |
70524490 | 1982 | |
aadd06e5 JA |
1983 | /* |
1984 | * If we put data in the output pipe, wakeup any potential readers. | |
1985 | */ | |
825cdcb1 NK |
1986 | if (ret > 0) |
1987 | wakeup_pipe_readers(opipe); | |
70524490 JA |
1988 | |
1989 | return ret; | |
1990 | } | |
1991 | ||
1992 | /* | |
1993 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1994 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1995 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1996 | * applicable one is SPLICE_F_NONBLOCK. | |
1997 | */ | |
1998 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1999 | unsigned int flags) | |
2000 | { | |
71993e62 LT |
2001 | struct pipe_inode_info *ipipe = get_pipe_info(in); |
2002 | struct pipe_inode_info *opipe = get_pipe_info(out); | |
aadd06e5 | 2003 | int ret = -EINVAL; |
70524490 JA |
2004 | |
2005 | /* | |
aadd06e5 JA |
2006 | * Duplicate the contents of ipipe to opipe without actually |
2007 | * copying the data. | |
70524490 | 2008 | */ |
aadd06e5 JA |
2009 | if (ipipe && opipe && ipipe != opipe) { |
2010 | /* | |
2011 | * Keep going, unless we encounter an error. The ipipe/opipe | |
2012 | * ordering doesn't really matter. | |
2013 | */ | |
7c77f0b3 | 2014 | ret = ipipe_prep(ipipe, flags); |
aadd06e5 | 2015 | if (!ret) { |
7c77f0b3 | 2016 | ret = opipe_prep(opipe, flags); |
02cf01ae | 2017 | if (!ret) |
aadd06e5 | 2018 | ret = link_pipe(ipipe, opipe, len, flags); |
aadd06e5 JA |
2019 | } |
2020 | } | |
70524490 | 2021 | |
aadd06e5 | 2022 | return ret; |
70524490 JA |
2023 | } |
2024 | ||
836f92ad | 2025 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
70524490 | 2026 | { |
2903ff01 AV |
2027 | struct fd in; |
2028 | int error; | |
70524490 JA |
2029 | |
2030 | if (unlikely(!len)) | |
2031 | return 0; | |
2032 | ||
2033 | error = -EBADF; | |
2903ff01 AV |
2034 | in = fdget(fdin); |
2035 | if (in.file) { | |
2036 | if (in.file->f_mode & FMODE_READ) { | |
2037 | struct fd out = fdget(fdout); | |
2038 | if (out.file) { | |
2039 | if (out.file->f_mode & FMODE_WRITE) | |
2040 | error = do_tee(in.file, out.file, | |
2041 | len, flags); | |
2042 | fdput(out); | |
70524490 JA |
2043 | } |
2044 | } | |
2903ff01 | 2045 | fdput(in); |
70524490 JA |
2046 | } |
2047 | ||
2048 | return error; | |
2049 | } |