]>
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 | * |
c2058e06 JA |
15 | * Copyright (C) 2005-2006 Jens Axboe <axboe@suse.de> |
16 | * Copyright (C) 2005-2006 Linus Torvalds <torvalds@osdl.org> | |
17 | * Copyright (C) 2006 Ingo Molnar <mingo@elte.hu> | |
5274f052 JA |
18 | * |
19 | */ | |
20 | #include <linux/fs.h> | |
21 | #include <linux/file.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/pipe_fs_i.h> | |
24 | #include <linux/mm_inline.h> | |
5abc97aa | 25 | #include <linux/swap.h> |
4f6f0bd2 JA |
26 | #include <linux/writeback.h> |
27 | #include <linux/buffer_head.h> | |
a0f06780 | 28 | #include <linux/module.h> |
4f6f0bd2 | 29 | #include <linux/syscalls.h> |
912d35f8 | 30 | #include <linux/uio.h> |
5274f052 | 31 | |
912d35f8 JA |
32 | struct partial_page { |
33 | unsigned int offset; | |
34 | unsigned int len; | |
35 | }; | |
36 | ||
37 | /* | |
00522fb4 | 38 | * Passed to splice_to_pipe |
912d35f8 JA |
39 | */ |
40 | struct splice_pipe_desc { | |
41 | struct page **pages; /* page map */ | |
42 | struct partial_page *partial; /* pages[] may not be contig */ | |
43 | int nr_pages; /* number of pages in map */ | |
44 | unsigned int flags; /* splice flags */ | |
45 | struct pipe_buf_operations *ops;/* ops associated with output pipe */ | |
46 | }; | |
47 | ||
83f9135b JA |
48 | /* |
49 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
50 | * a vm helper function, it's already simplified quite a bit by the | |
51 | * addition of remove_mapping(). If success is returned, the caller may | |
52 | * attempt to reuse this page for another destination. | |
53 | */ | |
5abc97aa JA |
54 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *info, |
55 | struct pipe_buffer *buf) | |
56 | { | |
57 | struct page *page = buf->page; | |
4f6f0bd2 | 58 | struct address_space *mapping = page_mapping(page); |
5abc97aa | 59 | |
9e0267c2 JA |
60 | lock_page(page); |
61 | ||
5abc97aa JA |
62 | WARN_ON(!PageUptodate(page)); |
63 | ||
ad8d6f0a JA |
64 | /* |
65 | * At least for ext2 with nobh option, we need to wait on writeback | |
66 | * completing on this page, since we'll remove it from the pagecache. | |
67 | * Otherwise truncate wont wait on the page, allowing the disk | |
68 | * blocks to be reused by someone else before we actually wrote our | |
69 | * data to them. fs corruption ensues. | |
70 | */ | |
71 | wait_on_page_writeback(page); | |
72 | ||
4f6f0bd2 JA |
73 | if (PagePrivate(page)) |
74 | try_to_release_page(page, mapping_gfp_mask(mapping)); | |
75 | ||
9e0267c2 JA |
76 | if (!remove_mapping(mapping, page)) { |
77 | unlock_page(page); | |
5abc97aa | 78 | return 1; |
9e0267c2 | 79 | } |
5abc97aa | 80 | |
5abc97aa JA |
81 | return 0; |
82 | } | |
83 | ||
5274f052 JA |
84 | static void page_cache_pipe_buf_release(struct pipe_inode_info *info, |
85 | struct pipe_buffer *buf) | |
86 | { | |
87 | page_cache_release(buf->page); | |
5274f052 JA |
88 | } |
89 | ||
f84d7519 JA |
90 | static int page_cache_pipe_buf_pin(struct pipe_inode_info *info, |
91 | struct pipe_buffer *buf) | |
5274f052 JA |
92 | { |
93 | struct page *page = buf->page; | |
49d0b21b | 94 | int err; |
5274f052 JA |
95 | |
96 | if (!PageUptodate(page)) { | |
49d0b21b JA |
97 | lock_page(page); |
98 | ||
99 | /* | |
100 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 101 | * splice, if this is the first page. |
49d0b21b JA |
102 | */ |
103 | if (!page->mapping) { | |
104 | err = -ENODATA; | |
105 | goto error; | |
106 | } | |
5274f052 | 107 | |
49d0b21b | 108 | /* |
73d62d83 | 109 | * Uh oh, read-error from disk. |
49d0b21b JA |
110 | */ |
111 | if (!PageUptodate(page)) { | |
112 | err = -EIO; | |
113 | goto error; | |
114 | } | |
115 | ||
116 | /* | |
f84d7519 | 117 | * Page is ok afterall, we are done. |
49d0b21b | 118 | */ |
5274f052 | 119 | unlock_page(page); |
5274f052 JA |
120 | } |
121 | ||
f84d7519 | 122 | return 0; |
49d0b21b JA |
123 | error: |
124 | unlock_page(page); | |
f84d7519 | 125 | return err; |
70524490 JA |
126 | } |
127 | ||
5274f052 JA |
128 | static struct pipe_buf_operations page_cache_pipe_buf_ops = { |
129 | .can_merge = 0, | |
f84d7519 JA |
130 | .map = generic_pipe_buf_map, |
131 | .unmap = generic_pipe_buf_unmap, | |
132 | .pin = page_cache_pipe_buf_pin, | |
5274f052 | 133 | .release = page_cache_pipe_buf_release, |
5abc97aa | 134 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 135 | .get = generic_pipe_buf_get, |
5274f052 JA |
136 | }; |
137 | ||
912d35f8 JA |
138 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
139 | struct pipe_buffer *buf) | |
140 | { | |
7afa6fd0 JA |
141 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
142 | return 1; | |
143 | ||
330ab716 | 144 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
145 | } |
146 | ||
147 | static struct pipe_buf_operations user_page_pipe_buf_ops = { | |
148 | .can_merge = 0, | |
f84d7519 JA |
149 | .map = generic_pipe_buf_map, |
150 | .unmap = generic_pipe_buf_unmap, | |
151 | .pin = generic_pipe_buf_pin, | |
912d35f8 JA |
152 | .release = page_cache_pipe_buf_release, |
153 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 154 | .get = generic_pipe_buf_get, |
912d35f8 JA |
155 | }; |
156 | ||
83f9135b JA |
157 | /* |
158 | * Pipe output worker. This sets up our pipe format with the page cache | |
159 | * pipe buffer operations. Otherwise very similar to the regular pipe_writev(). | |
160 | */ | |
00522fb4 JA |
161 | static ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
162 | struct splice_pipe_desc *spd) | |
5274f052 | 163 | { |
912d35f8 | 164 | int ret, do_wakeup, page_nr; |
5274f052 JA |
165 | |
166 | ret = 0; | |
167 | do_wakeup = 0; | |
912d35f8 | 168 | page_nr = 0; |
5274f052 | 169 | |
3a326a2c IM |
170 | if (pipe->inode) |
171 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 172 | |
5274f052 | 173 | for (;;) { |
3a326a2c | 174 | if (!pipe->readers) { |
5274f052 JA |
175 | send_sig(SIGPIPE, current, 0); |
176 | if (!ret) | |
177 | ret = -EPIPE; | |
178 | break; | |
179 | } | |
180 | ||
6f767b04 JA |
181 | if (pipe->nrbufs < PIPE_BUFFERS) { |
182 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (PIPE_BUFFERS - 1); | |
3a326a2c | 183 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
5274f052 | 184 | |
912d35f8 JA |
185 | buf->page = spd->pages[page_nr]; |
186 | buf->offset = spd->partial[page_nr].offset; | |
187 | buf->len = spd->partial[page_nr].len; | |
188 | buf->ops = spd->ops; | |
7afa6fd0 JA |
189 | if (spd->flags & SPLICE_F_GIFT) |
190 | buf->flags |= PIPE_BUF_FLAG_GIFT; | |
191 | ||
6f767b04 | 192 | pipe->nrbufs++; |
912d35f8 JA |
193 | page_nr++; |
194 | ret += buf->len; | |
195 | ||
6f767b04 JA |
196 | if (pipe->inode) |
197 | do_wakeup = 1; | |
5274f052 | 198 | |
912d35f8 | 199 | if (!--spd->nr_pages) |
5274f052 | 200 | break; |
6f767b04 | 201 | if (pipe->nrbufs < PIPE_BUFFERS) |
5274f052 JA |
202 | continue; |
203 | ||
204 | break; | |
205 | } | |
206 | ||
912d35f8 | 207 | if (spd->flags & SPLICE_F_NONBLOCK) { |
29e35094 LT |
208 | if (!ret) |
209 | ret = -EAGAIN; | |
210 | break; | |
211 | } | |
212 | ||
5274f052 JA |
213 | if (signal_pending(current)) { |
214 | if (!ret) | |
215 | ret = -ERESTARTSYS; | |
216 | break; | |
217 | } | |
218 | ||
219 | if (do_wakeup) { | |
c0bd1f65 | 220 | smp_mb(); |
3a326a2c IM |
221 | if (waitqueue_active(&pipe->wait)) |
222 | wake_up_interruptible_sync(&pipe->wait); | |
223 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
224 | do_wakeup = 0; |
225 | } | |
226 | ||
3a326a2c IM |
227 | pipe->waiting_writers++; |
228 | pipe_wait(pipe); | |
229 | pipe->waiting_writers--; | |
5274f052 JA |
230 | } |
231 | ||
3a326a2c IM |
232 | if (pipe->inode) |
233 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
234 | |
235 | if (do_wakeup) { | |
c0bd1f65 | 236 | smp_mb(); |
3a326a2c IM |
237 | if (waitqueue_active(&pipe->wait)) |
238 | wake_up_interruptible(&pipe->wait); | |
239 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
5274f052 JA |
240 | } |
241 | ||
912d35f8 JA |
242 | while (page_nr < spd->nr_pages) |
243 | page_cache_release(spd->pages[page_nr++]); | |
5274f052 JA |
244 | |
245 | return ret; | |
246 | } | |
247 | ||
3a326a2c | 248 | static int |
cbb7e577 JA |
249 | __generic_file_splice_read(struct file *in, loff_t *ppos, |
250 | struct pipe_inode_info *pipe, size_t len, | |
251 | unsigned int flags) | |
5274f052 JA |
252 | { |
253 | struct address_space *mapping = in->f_mapping; | |
912d35f8 | 254 | unsigned int loff, nr_pages; |
16c523dd | 255 | struct page *pages[PIPE_BUFFERS]; |
912d35f8 | 256 | struct partial_page partial[PIPE_BUFFERS]; |
5274f052 | 257 | struct page *page; |
91ad66ef JA |
258 | pgoff_t index, end_index; |
259 | loff_t isize; | |
912d35f8 | 260 | size_t total_len; |
eb20796b | 261 | int error, page_nr; |
912d35f8 JA |
262 | struct splice_pipe_desc spd = { |
263 | .pages = pages, | |
264 | .partial = partial, | |
265 | .flags = flags, | |
266 | .ops = &page_cache_pipe_buf_ops, | |
267 | }; | |
5274f052 | 268 | |
cbb7e577 | 269 | index = *ppos >> PAGE_CACHE_SHIFT; |
912d35f8 JA |
270 | loff = *ppos & ~PAGE_CACHE_MASK; |
271 | nr_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
5274f052 JA |
272 | |
273 | if (nr_pages > PIPE_BUFFERS) | |
274 | nr_pages = PIPE_BUFFERS; | |
275 | ||
276 | /* | |
73d62d83 | 277 | * Initiate read-ahead on this page range. however, don't call into |
0b749ce3 JA |
278 | * read-ahead if this is a non-zero offset (we are likely doing small |
279 | * chunk splice and the page is already there) for a single page. | |
5274f052 | 280 | */ |
eb645a24 JA |
281 | if (!loff || nr_pages > 1) |
282 | page_cache_readahead(mapping, &in->f_ra, in, index, nr_pages); | |
5274f052 | 283 | |
5274f052 | 284 | /* |
73d62d83 | 285 | * Now fill in the holes: |
5274f052 | 286 | */ |
7480a904 | 287 | error = 0; |
912d35f8 | 288 | total_len = 0; |
82aa5d61 | 289 | |
eb20796b JA |
290 | /* |
291 | * Lookup the (hopefully) full range of pages we need. | |
292 | */ | |
293 | spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, pages); | |
82aa5d61 | 294 | |
eb20796b JA |
295 | /* |
296 | * If find_get_pages_contig() returned fewer pages than we needed, | |
297 | * allocate the rest. | |
298 | */ | |
299 | index += spd.nr_pages; | |
300 | while (spd.nr_pages < nr_pages) { | |
82aa5d61 | 301 | /* |
eb20796b JA |
302 | * Page could be there, find_get_pages_contig() breaks on |
303 | * the first hole. | |
5274f052 | 304 | */ |
7480a904 JA |
305 | page = find_get_page(mapping, index); |
306 | if (!page) { | |
e27dedd8 JA |
307 | /* |
308 | * Make sure the read-ahead engine is notified | |
309 | * about this failure. | |
310 | */ | |
311 | handle_ra_miss(mapping, &in->f_ra, index); | |
312 | ||
7480a904 | 313 | /* |
eb20796b | 314 | * page didn't exist, allocate one. |
7480a904 JA |
315 | */ |
316 | page = page_cache_alloc_cold(mapping); | |
317 | if (!page) | |
318 | break; | |
319 | ||
320 | error = add_to_page_cache_lru(page, mapping, index, | |
eb20796b | 321 | mapping_gfp_mask(mapping)); |
7480a904 JA |
322 | if (unlikely(error)) { |
323 | page_cache_release(page); | |
324 | break; | |
325 | } | |
eb20796b JA |
326 | /* |
327 | * add_to_page_cache() locks the page, unlock it | |
328 | * to avoid convoluting the logic below even more. | |
329 | */ | |
330 | unlock_page(page); | |
7480a904 JA |
331 | } |
332 | ||
eb20796b JA |
333 | pages[spd.nr_pages++] = page; |
334 | index++; | |
335 | } | |
336 | ||
337 | /* | |
338 | * Now loop over the map and see if we need to start IO on any | |
339 | * pages, fill in the partial map, etc. | |
340 | */ | |
341 | index = *ppos >> PAGE_CACHE_SHIFT; | |
342 | nr_pages = spd.nr_pages; | |
343 | spd.nr_pages = 0; | |
344 | for (page_nr = 0; page_nr < nr_pages; page_nr++) { | |
345 | unsigned int this_len; | |
346 | ||
347 | if (!len) | |
348 | break; | |
349 | ||
350 | /* | |
351 | * this_len is the max we'll use from this page | |
352 | */ | |
353 | this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); | |
354 | page = pages[page_nr]; | |
355 | ||
7480a904 JA |
356 | /* |
357 | * If the page isn't uptodate, we may need to start io on it | |
358 | */ | |
359 | if (!PageUptodate(page)) { | |
c4f895cb JA |
360 | /* |
361 | * If in nonblock mode then dont block on waiting | |
362 | * for an in-flight io page | |
363 | */ | |
364 | if (flags & SPLICE_F_NONBLOCK) | |
365 | break; | |
366 | ||
7480a904 JA |
367 | lock_page(page); |
368 | ||
369 | /* | |
370 | * page was truncated, stop here. if this isn't the | |
371 | * first page, we'll just complete what we already | |
372 | * added | |
373 | */ | |
374 | if (!page->mapping) { | |
375 | unlock_page(page); | |
7480a904 JA |
376 | break; |
377 | } | |
378 | /* | |
379 | * page was already under io and is now done, great | |
380 | */ | |
381 | if (PageUptodate(page)) { | |
382 | unlock_page(page); | |
383 | goto fill_it; | |
384 | } | |
5274f052 | 385 | |
7480a904 JA |
386 | /* |
387 | * need to read in the page | |
388 | */ | |
389 | error = mapping->a_ops->readpage(in, page); | |
5274f052 | 390 | if (unlikely(error)) { |
eb20796b JA |
391 | /* |
392 | * We really should re-lookup the page here, | |
393 | * but it complicates things a lot. Instead | |
394 | * lets just do what we already stored, and | |
395 | * we'll get it the next time we are called. | |
396 | */ | |
7480a904 | 397 | if (error == AOP_TRUNCATED_PAGE) |
eb20796b JA |
398 | error = 0; |
399 | ||
5274f052 JA |
400 | break; |
401 | } | |
91ad66ef JA |
402 | |
403 | /* | |
404 | * i_size must be checked after ->readpage(). | |
405 | */ | |
406 | isize = i_size_read(mapping->host); | |
407 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
eb20796b | 408 | if (unlikely(!isize || index > end_index)) |
91ad66ef | 409 | break; |
91ad66ef JA |
410 | |
411 | /* | |
412 | * if this is the last page, see if we need to shrink | |
413 | * the length and stop | |
414 | */ | |
415 | if (end_index == index) { | |
416 | loff = PAGE_CACHE_SIZE - (isize & ~PAGE_CACHE_MASK); | |
eb20796b | 417 | if (total_len + loff > isize) |
91ad66ef | 418 | break; |
91ad66ef JA |
419 | /* |
420 | * force quit after adding this page | |
421 | */ | |
eb20796b | 422 | len = this_len; |
82aa5d61 | 423 | this_len = min(this_len, loff); |
912d35f8 | 424 | loff = 0; |
91ad66ef | 425 | } |
5274f052 | 426 | } |
7480a904 | 427 | fill_it: |
eb20796b JA |
428 | partial[page_nr].offset = loff; |
429 | partial[page_nr].len = this_len; | |
82aa5d61 | 430 | len -= this_len; |
912d35f8 | 431 | total_len += this_len; |
91ad66ef | 432 | loff = 0; |
eb20796b JA |
433 | spd.nr_pages++; |
434 | index++; | |
5274f052 JA |
435 | } |
436 | ||
eb20796b JA |
437 | /* |
438 | * Release any pages at the end, if we quit early. 'i' is how far | |
439 | * we got, 'nr_pages' is how many pages are in the map. | |
440 | */ | |
441 | while (page_nr < nr_pages) | |
442 | page_cache_release(pages[page_nr++]); | |
443 | ||
912d35f8 | 444 | if (spd.nr_pages) |
00522fb4 | 445 | return splice_to_pipe(pipe, &spd); |
5274f052 | 446 | |
7480a904 | 447 | return error; |
5274f052 JA |
448 | } |
449 | ||
83f9135b JA |
450 | /** |
451 | * generic_file_splice_read - splice data from file to a pipe | |
452 | * @in: file to splice from | |
453 | * @pipe: pipe to splice to | |
454 | * @len: number of bytes to splice | |
455 | * @flags: splice modifier flags | |
456 | * | |
457 | * Will read pages from given file and fill them into a pipe. | |
83f9135b | 458 | */ |
cbb7e577 JA |
459 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
460 | struct pipe_inode_info *pipe, size_t len, | |
461 | unsigned int flags) | |
5274f052 JA |
462 | { |
463 | ssize_t spliced; | |
464 | int ret; | |
465 | ||
466 | ret = 0; | |
467 | spliced = 0; | |
3a326a2c | 468 | |
5274f052 | 469 | while (len) { |
cbb7e577 | 470 | ret = __generic_file_splice_read(in, ppos, pipe, len, flags); |
5274f052 | 471 | |
c4f895cb | 472 | if (ret < 0) |
5274f052 | 473 | break; |
c4f895cb JA |
474 | else if (!ret) { |
475 | if (spliced) | |
476 | break; | |
477 | if (flags & SPLICE_F_NONBLOCK) { | |
478 | ret = -EAGAIN; | |
479 | break; | |
480 | } | |
481 | } | |
5274f052 | 482 | |
cbb7e577 | 483 | *ppos += ret; |
5274f052 JA |
484 | len -= ret; |
485 | spliced += ret; | |
486 | } | |
487 | ||
488 | if (spliced) | |
489 | return spliced; | |
490 | ||
491 | return ret; | |
492 | } | |
493 | ||
059a8f37 JA |
494 | EXPORT_SYMBOL(generic_file_splice_read); |
495 | ||
5274f052 | 496 | /* |
4f6f0bd2 | 497 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 498 | * using sendpage(). Return the number of bytes sent. |
5274f052 JA |
499 | */ |
500 | static int pipe_to_sendpage(struct pipe_inode_info *info, | |
501 | struct pipe_buffer *buf, struct splice_desc *sd) | |
502 | { | |
503 | struct file *file = sd->file; | |
504 | loff_t pos = sd->pos; | |
f84d7519 | 505 | int ret, more; |
5274f052 | 506 | |
f84d7519 JA |
507 | ret = buf->ops->pin(info, buf); |
508 | if (!ret) { | |
509 | more = (sd->flags & SPLICE_F_MORE) || sd->len < sd->total_len; | |
5274f052 | 510 | |
f84d7519 JA |
511 | ret = file->f_op->sendpage(file, buf->page, buf->offset, |
512 | sd->len, &pos, more); | |
513 | } | |
5274f052 | 514 | |
016b661e | 515 | return ret; |
5274f052 JA |
516 | } |
517 | ||
518 | /* | |
519 | * This is a little more tricky than the file -> pipe splicing. There are | |
520 | * basically three cases: | |
521 | * | |
522 | * - Destination page already exists in the address space and there | |
523 | * are users of it. For that case we have no other option that | |
524 | * copying the data. Tough luck. | |
525 | * - Destination page already exists in the address space, but there | |
526 | * are no users of it. Make sure it's uptodate, then drop it. Fall | |
527 | * through to last case. | |
528 | * - Destination page does not exist, we can add the pipe page to | |
529 | * the page cache and avoid the copy. | |
530 | * | |
83f9135b JA |
531 | * If asked to move pages to the output file (SPLICE_F_MOVE is set in |
532 | * sd->flags), we attempt to migrate pages from the pipe to the output | |
533 | * file address space page cache. This is possible if no one else has | |
534 | * the pipe page referenced outside of the pipe and page cache. If | |
535 | * SPLICE_F_MOVE isn't set, or we cannot move the page, we simply create | |
536 | * a new page in the output file page cache and fill/dirty that. | |
5274f052 JA |
537 | */ |
538 | static int pipe_to_file(struct pipe_inode_info *info, struct pipe_buffer *buf, | |
539 | struct splice_desc *sd) | |
540 | { | |
541 | struct file *file = sd->file; | |
542 | struct address_space *mapping = file->f_mapping; | |
3e7ee3e7 | 543 | gfp_t gfp_mask = mapping_gfp_mask(mapping); |
016b661e | 544 | unsigned int offset, this_len; |
5274f052 | 545 | struct page *page; |
5274f052 | 546 | pgoff_t index; |
3e7ee3e7 | 547 | int ret; |
5274f052 JA |
548 | |
549 | /* | |
49d0b21b | 550 | * make sure the data in this buffer is uptodate |
5274f052 | 551 | */ |
f84d7519 JA |
552 | ret = buf->ops->pin(info, buf); |
553 | if (unlikely(ret)) | |
554 | return ret; | |
5274f052 JA |
555 | |
556 | index = sd->pos >> PAGE_CACHE_SHIFT; | |
557 | offset = sd->pos & ~PAGE_CACHE_MASK; | |
558 | ||
016b661e JA |
559 | this_len = sd->len; |
560 | if (this_len + offset > PAGE_CACHE_SIZE) | |
561 | this_len = PAGE_CACHE_SIZE - offset; | |
562 | ||
5274f052 | 563 | /* |
0568b409 JA |
564 | * Reuse buf page, if SPLICE_F_MOVE is set and we are doing a full |
565 | * page. | |
5274f052 | 566 | */ |
0568b409 | 567 | if ((sd->flags & SPLICE_F_MOVE) && this_len == PAGE_CACHE_SIZE) { |
83f9135b JA |
568 | /* |
569 | * If steal succeeds, buf->page is now pruned from the vm | |
a893b99b JA |
570 | * side (page cache) and we can reuse it. The page will also |
571 | * be locked on successful return. | |
83f9135b | 572 | */ |
5abc97aa JA |
573 | if (buf->ops->steal(info, buf)) |
574 | goto find_page; | |
575 | ||
576 | page = buf->page; | |
a893b99b JA |
577 | page_cache_get(page); |
578 | ||
579 | /* | |
580 | * page must be on the LRU for adding to the pagecache. | |
581 | * Check this without grabbing the zone lock, if it isn't | |
582 | * the do grab the zone lock, recheck, and add if necessary. | |
583 | */ | |
584 | if (!PageLRU(page)) { | |
585 | struct zone *zone = page_zone(page); | |
586 | ||
587 | spin_lock_irq(&zone->lru_lock); | |
588 | if (!PageLRU(page)) { | |
589 | SetPageLRU(page); | |
590 | add_page_to_inactive_list(zone, page); | |
591 | } | |
592 | spin_unlock_irq(&zone->lru_lock); | |
593 | } | |
594 | ||
46e678c9 | 595 | if (add_to_page_cache(page, mapping, index, gfp_mask)) { |
a893b99b | 596 | page_cache_release(page); |
46e678c9 | 597 | unlock_page(page); |
5abc97aa | 598 | goto find_page; |
46e678c9 | 599 | } |
5abc97aa JA |
600 | } else { |
601 | find_page: | |
9e0267c2 JA |
602 | page = find_lock_page(mapping, index); |
603 | if (!page) { | |
604 | ret = -ENOMEM; | |
605 | page = page_cache_alloc_cold(mapping); | |
606 | if (unlikely(!page)) | |
607 | goto out_nomem; | |
608 | ||
609 | /* | |
610 | * This will also lock the page | |
611 | */ | |
612 | ret = add_to_page_cache_lru(page, mapping, index, | |
613 | gfp_mask); | |
614 | if (unlikely(ret)) | |
615 | goto out; | |
616 | } | |
5abc97aa JA |
617 | |
618 | /* | |
9e0267c2 JA |
619 | * We get here with the page locked. If the page is also |
620 | * uptodate, we don't need to do more. If it isn't, we | |
621 | * may need to bring it in if we are not going to overwrite | |
622 | * the full page. | |
5abc97aa JA |
623 | */ |
624 | if (!PageUptodate(page)) { | |
016b661e | 625 | if (this_len < PAGE_CACHE_SIZE) { |
5abc97aa JA |
626 | ret = mapping->a_ops->readpage(file, page); |
627 | if (unlikely(ret)) | |
628 | goto out; | |
629 | ||
630 | lock_page(page); | |
631 | ||
632 | if (!PageUptodate(page)) { | |
633 | /* | |
73d62d83 | 634 | * Page got invalidated, repeat. |
5abc97aa JA |
635 | */ |
636 | if (!page->mapping) { | |
637 | unlock_page(page); | |
638 | page_cache_release(page); | |
639 | goto find_page; | |
640 | } | |
641 | ret = -EIO; | |
642 | goto out; | |
5274f052 | 643 | } |
9e0267c2 | 644 | } else |
5abc97aa | 645 | SetPageUptodate(page); |
5274f052 JA |
646 | } |
647 | } | |
648 | ||
016b661e | 649 | ret = mapping->a_ops->prepare_write(file, page, offset, offset+this_len); |
bfc4ee39 JA |
650 | if (unlikely(ret)) { |
651 | loff_t isize = i_size_read(mapping->host); | |
652 | ||
653 | if (ret != AOP_TRUNCATED_PAGE) | |
654 | unlock_page(page); | |
4f6f0bd2 | 655 | page_cache_release(page); |
bfc4ee39 JA |
656 | if (ret == AOP_TRUNCATED_PAGE) |
657 | goto find_page; | |
658 | ||
659 | /* | |
660 | * prepare_write() may have instantiated a few blocks | |
661 | * outside i_size. Trim these off again. | |
662 | */ | |
663 | if (sd->pos + this_len > isize) | |
664 | vmtruncate(mapping->host, isize); | |
665 | ||
5274f052 | 666 | goto out; |
bfc4ee39 | 667 | } |
5274f052 | 668 | |
0568b409 | 669 | if (buf->page != page) { |
f84d7519 JA |
670 | /* |
671 | * Careful, ->map() uses KM_USER0! | |
672 | */ | |
f6762b7a | 673 | char *src = buf->ops->map(info, buf, 1); |
f84d7519 | 674 | char *dst = kmap_atomic(page, KM_USER1); |
5abc97aa | 675 | |
016b661e | 676 | memcpy(dst + offset, src + buf->offset, this_len); |
5abc97aa | 677 | flush_dcache_page(page); |
f84d7519 | 678 | kunmap_atomic(dst, KM_USER1); |
f6762b7a | 679 | buf->ops->unmap(info, buf, src); |
5abc97aa | 680 | } |
5274f052 | 681 | |
016b661e | 682 | ret = mapping->a_ops->commit_write(file, page, offset, offset+this_len); |
0568b409 JA |
683 | if (!ret) { |
684 | /* | |
685 | * Return the number of bytes written and mark page as | |
686 | * accessed, we are now done! | |
687 | */ | |
688 | ret = this_len; | |
689 | mark_page_accessed(page); | |
690 | balance_dirty_pages_ratelimited(mapping); | |
691 | } else if (ret == AOP_TRUNCATED_PAGE) { | |
4f6f0bd2 JA |
692 | page_cache_release(page); |
693 | goto find_page; | |
0568b409 | 694 | } |
5274f052 | 695 | out: |
0568b409 | 696 | page_cache_release(page); |
9e0267c2 | 697 | unlock_page(page); |
9aefe431 | 698 | out_nomem: |
5274f052 JA |
699 | return ret; |
700 | } | |
701 | ||
83f9135b JA |
702 | /* |
703 | * Pipe input worker. Most of this logic works like a regular pipe, the | |
704 | * key here is the 'actor' worker passed in that actually moves the data | |
705 | * to the wanted destination. See pipe_to_file/pipe_to_sendpage above. | |
706 | */ | |
00522fb4 JA |
707 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
708 | loff_t *ppos, size_t len, unsigned int flags, | |
709 | splice_actor *actor) | |
5274f052 | 710 | { |
5274f052 JA |
711 | int ret, do_wakeup, err; |
712 | struct splice_desc sd; | |
713 | ||
714 | ret = 0; | |
715 | do_wakeup = 0; | |
716 | ||
717 | sd.total_len = len; | |
718 | sd.flags = flags; | |
719 | sd.file = out; | |
cbb7e577 | 720 | sd.pos = *ppos; |
5274f052 | 721 | |
3a326a2c IM |
722 | if (pipe->inode) |
723 | mutex_lock(&pipe->inode->i_mutex); | |
5274f052 | 724 | |
5274f052 | 725 | for (;;) { |
6f767b04 JA |
726 | if (pipe->nrbufs) { |
727 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
5274f052 JA |
728 | struct pipe_buf_operations *ops = buf->ops; |
729 | ||
730 | sd.len = buf->len; | |
731 | if (sd.len > sd.total_len) | |
732 | sd.len = sd.total_len; | |
733 | ||
3a326a2c | 734 | err = actor(pipe, buf, &sd); |
016b661e | 735 | if (err <= 0) { |
5274f052 JA |
736 | if (!ret && err != -ENODATA) |
737 | ret = err; | |
738 | ||
739 | break; | |
740 | } | |
741 | ||
016b661e JA |
742 | ret += err; |
743 | buf->offset += err; | |
744 | buf->len -= err; | |
745 | ||
746 | sd.len -= err; | |
747 | sd.pos += err; | |
748 | sd.total_len -= err; | |
749 | if (sd.len) | |
750 | continue; | |
73d62d83 | 751 | |
5274f052 JA |
752 | if (!buf->len) { |
753 | buf->ops = NULL; | |
3a326a2c | 754 | ops->release(pipe, buf); |
6f767b04 JA |
755 | pipe->curbuf = (pipe->curbuf + 1) & (PIPE_BUFFERS - 1); |
756 | pipe->nrbufs--; | |
757 | if (pipe->inode) | |
758 | do_wakeup = 1; | |
5274f052 JA |
759 | } |
760 | ||
5274f052 JA |
761 | if (!sd.total_len) |
762 | break; | |
763 | } | |
764 | ||
6f767b04 | 765 | if (pipe->nrbufs) |
5274f052 | 766 | continue; |
3a326a2c | 767 | if (!pipe->writers) |
5274f052 | 768 | break; |
3a326a2c | 769 | if (!pipe->waiting_writers) { |
5274f052 JA |
770 | if (ret) |
771 | break; | |
772 | } | |
773 | ||
29e35094 LT |
774 | if (flags & SPLICE_F_NONBLOCK) { |
775 | if (!ret) | |
776 | ret = -EAGAIN; | |
777 | break; | |
778 | } | |
779 | ||
5274f052 JA |
780 | if (signal_pending(current)) { |
781 | if (!ret) | |
782 | ret = -ERESTARTSYS; | |
783 | break; | |
784 | } | |
785 | ||
786 | if (do_wakeup) { | |
c0bd1f65 | 787 | smp_mb(); |
3a326a2c IM |
788 | if (waitqueue_active(&pipe->wait)) |
789 | wake_up_interruptible_sync(&pipe->wait); | |
790 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
791 | do_wakeup = 0; |
792 | } | |
793 | ||
3a326a2c | 794 | pipe_wait(pipe); |
5274f052 JA |
795 | } |
796 | ||
3a326a2c IM |
797 | if (pipe->inode) |
798 | mutex_unlock(&pipe->inode->i_mutex); | |
5274f052 JA |
799 | |
800 | if (do_wakeup) { | |
c0bd1f65 | 801 | smp_mb(); |
3a326a2c IM |
802 | if (waitqueue_active(&pipe->wait)) |
803 | wake_up_interruptible(&pipe->wait); | |
804 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
5274f052 JA |
805 | } |
806 | ||
5274f052 | 807 | return ret; |
5274f052 JA |
808 | } |
809 | ||
83f9135b JA |
810 | /** |
811 | * generic_file_splice_write - splice data from a pipe to a file | |
3a326a2c | 812 | * @pipe: pipe info |
83f9135b JA |
813 | * @out: file to write to |
814 | * @len: number of bytes to splice | |
815 | * @flags: splice modifier flags | |
816 | * | |
817 | * Will either move or copy pages (determined by @flags options) from | |
818 | * the given pipe inode to the given file. | |
819 | * | |
820 | */ | |
3a326a2c IM |
821 | ssize_t |
822 | generic_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
cbb7e577 | 823 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 824 | { |
4f6f0bd2 | 825 | struct address_space *mapping = out->f_mapping; |
3a326a2c IM |
826 | ssize_t ret; |
827 | ||
00522fb4 | 828 | ret = splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_file); |
a4514ebd | 829 | if (ret > 0) { |
4f6f0bd2 | 830 | struct inode *inode = mapping->host; |
4f6f0bd2 | 831 | |
a4514ebd JA |
832 | *ppos += ret; |
833 | ||
834 | /* | |
835 | * If file or inode is SYNC and we actually wrote some data, | |
836 | * sync it. | |
837 | */ | |
838 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
839 | int err; | |
840 | ||
841 | mutex_lock(&inode->i_mutex); | |
842 | err = generic_osync_inode(inode, mapping, | |
843 | OSYNC_METADATA|OSYNC_DATA); | |
844 | mutex_unlock(&inode->i_mutex); | |
4f6f0bd2 | 845 | |
a4514ebd JA |
846 | if (err) |
847 | ret = err; | |
848 | } | |
4f6f0bd2 JA |
849 | } |
850 | ||
851 | return ret; | |
5274f052 JA |
852 | } |
853 | ||
059a8f37 JA |
854 | EXPORT_SYMBOL(generic_file_splice_write); |
855 | ||
83f9135b JA |
856 | /** |
857 | * generic_splice_sendpage - splice data from a pipe to a socket | |
858 | * @inode: pipe inode | |
859 | * @out: socket to write to | |
860 | * @len: number of bytes to splice | |
861 | * @flags: splice modifier flags | |
862 | * | |
863 | * Will send @len bytes from the pipe to a network socket. No data copying | |
864 | * is involved. | |
865 | * | |
866 | */ | |
3a326a2c | 867 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 868 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 869 | { |
00522fb4 | 870 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
871 | } |
872 | ||
059a8f37 | 873 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 874 | |
83f9135b JA |
875 | /* |
876 | * Attempt to initiate a splice from pipe to file. | |
877 | */ | |
3a326a2c | 878 | static long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 879 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 880 | { |
5274f052 JA |
881 | int ret; |
882 | ||
49570e9b | 883 | if (unlikely(!out->f_op || !out->f_op->splice_write)) |
5274f052 JA |
884 | return -EINVAL; |
885 | ||
49570e9b | 886 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
5274f052 JA |
887 | return -EBADF; |
888 | ||
cbb7e577 | 889 | ret = rw_verify_area(WRITE, out, ppos, len); |
5274f052 JA |
890 | if (unlikely(ret < 0)) |
891 | return ret; | |
892 | ||
cbb7e577 | 893 | return out->f_op->splice_write(pipe, out, ppos, len, flags); |
5274f052 JA |
894 | } |
895 | ||
83f9135b JA |
896 | /* |
897 | * Attempt to initiate a splice from a file to a pipe. | |
898 | */ | |
cbb7e577 JA |
899 | static long do_splice_to(struct file *in, loff_t *ppos, |
900 | struct pipe_inode_info *pipe, size_t len, | |
901 | unsigned int flags) | |
5274f052 | 902 | { |
cbb7e577 | 903 | loff_t isize, left; |
5274f052 JA |
904 | int ret; |
905 | ||
49570e9b | 906 | if (unlikely(!in->f_op || !in->f_op->splice_read)) |
5274f052 JA |
907 | return -EINVAL; |
908 | ||
49570e9b | 909 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
910 | return -EBADF; |
911 | ||
cbb7e577 | 912 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
913 | if (unlikely(ret < 0)) |
914 | return ret; | |
915 | ||
916 | isize = i_size_read(in->f_mapping->host); | |
cbb7e577 | 917 | if (unlikely(*ppos >= isize)) |
5274f052 JA |
918 | return 0; |
919 | ||
cbb7e577 | 920 | left = isize - *ppos; |
49570e9b | 921 | if (unlikely(left < len)) |
5274f052 JA |
922 | len = left; |
923 | ||
cbb7e577 | 924 | return in->f_op->splice_read(in, ppos, pipe, len, flags); |
5274f052 JA |
925 | } |
926 | ||
cbb7e577 JA |
927 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
928 | size_t len, unsigned int flags) | |
b92ce558 JA |
929 | { |
930 | struct pipe_inode_info *pipe; | |
931 | long ret, bytes; | |
cbb7e577 | 932 | loff_t out_off; |
b92ce558 JA |
933 | umode_t i_mode; |
934 | int i; | |
935 | ||
936 | /* | |
937 | * We require the input being a regular file, as we don't want to | |
938 | * randomly drop data for eg socket -> socket splicing. Use the | |
939 | * piped splicing for that! | |
940 | */ | |
941 | i_mode = in->f_dentry->d_inode->i_mode; | |
942 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) | |
943 | return -EINVAL; | |
944 | ||
945 | /* | |
946 | * neither in nor out is a pipe, setup an internal pipe attached to | |
947 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
948 | */ | |
949 | pipe = current->splice_pipe; | |
49570e9b | 950 | if (unlikely(!pipe)) { |
b92ce558 JA |
951 | pipe = alloc_pipe_info(NULL); |
952 | if (!pipe) | |
953 | return -ENOMEM; | |
954 | ||
955 | /* | |
956 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 957 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
958 | * PIPE_READERS appropriately. |
959 | */ | |
960 | pipe->readers = 1; | |
961 | ||
962 | current->splice_pipe = pipe; | |
963 | } | |
964 | ||
965 | /* | |
73d62d83 | 966 | * Do the splice. |
b92ce558 JA |
967 | */ |
968 | ret = 0; | |
969 | bytes = 0; | |
cbb7e577 | 970 | out_off = 0; |
b92ce558 JA |
971 | |
972 | while (len) { | |
973 | size_t read_len, max_read_len; | |
974 | ||
975 | /* | |
976 | * Do at most PIPE_BUFFERS pages worth of transfer: | |
977 | */ | |
978 | max_read_len = min(len, (size_t)(PIPE_BUFFERS*PAGE_SIZE)); | |
979 | ||
cbb7e577 | 980 | ret = do_splice_to(in, ppos, pipe, max_read_len, flags); |
b92ce558 JA |
981 | if (unlikely(ret < 0)) |
982 | goto out_release; | |
983 | ||
984 | read_len = ret; | |
985 | ||
986 | /* | |
987 | * NOTE: nonblocking mode only applies to the input. We | |
988 | * must not do the output in nonblocking mode as then we | |
989 | * could get stuck data in the internal pipe: | |
990 | */ | |
cbb7e577 | 991 | ret = do_splice_from(pipe, out, &out_off, read_len, |
b92ce558 JA |
992 | flags & ~SPLICE_F_NONBLOCK); |
993 | if (unlikely(ret < 0)) | |
994 | goto out_release; | |
995 | ||
996 | bytes += ret; | |
997 | len -= ret; | |
998 | ||
999 | /* | |
1000 | * In nonblocking mode, if we got back a short read then | |
1001 | * that was due to either an IO error or due to the | |
1002 | * pagecache entry not being there. In the IO error case | |
1003 | * the _next_ splice attempt will produce a clean IO error | |
1004 | * return value (not a short read), so in both cases it's | |
1005 | * correct to break out of the loop here: | |
1006 | */ | |
1007 | if ((flags & SPLICE_F_NONBLOCK) && (read_len < max_read_len)) | |
1008 | break; | |
1009 | } | |
1010 | ||
1011 | pipe->nrbufs = pipe->curbuf = 0; | |
1012 | ||
1013 | return bytes; | |
1014 | ||
1015 | out_release: | |
1016 | /* | |
1017 | * If we did an incomplete transfer we must release | |
1018 | * the pipe buffers in question: | |
1019 | */ | |
1020 | for (i = 0; i < PIPE_BUFFERS; i++) { | |
1021 | struct pipe_buffer *buf = pipe->bufs + i; | |
1022 | ||
1023 | if (buf->ops) { | |
1024 | buf->ops->release(pipe, buf); | |
1025 | buf->ops = NULL; | |
1026 | } | |
1027 | } | |
1028 | pipe->nrbufs = pipe->curbuf = 0; | |
1029 | ||
1030 | /* | |
1031 | * If we transferred some data, return the number of bytes: | |
1032 | */ | |
1033 | if (bytes > 0) | |
1034 | return bytes; | |
1035 | ||
1036 | return ret; | |
1037 | } | |
1038 | ||
1039 | EXPORT_SYMBOL(do_splice_direct); | |
1040 | ||
83f9135b JA |
1041 | /* |
1042 | * Determine where to splice to/from. | |
1043 | */ | |
529565dc IM |
1044 | static long do_splice(struct file *in, loff_t __user *off_in, |
1045 | struct file *out, loff_t __user *off_out, | |
1046 | size_t len, unsigned int flags) | |
5274f052 | 1047 | { |
3a326a2c | 1048 | struct pipe_inode_info *pipe; |
cbb7e577 | 1049 | loff_t offset, *off; |
a4514ebd | 1050 | long ret; |
5274f052 | 1051 | |
3a326a2c | 1052 | pipe = in->f_dentry->d_inode->i_pipe; |
529565dc IM |
1053 | if (pipe) { |
1054 | if (off_in) | |
1055 | return -ESPIPE; | |
b92ce558 JA |
1056 | if (off_out) { |
1057 | if (out->f_op->llseek == no_llseek) | |
1058 | return -EINVAL; | |
cbb7e577 | 1059 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1060 | return -EFAULT; |
cbb7e577 JA |
1061 | off = &offset; |
1062 | } else | |
1063 | off = &out->f_pos; | |
529565dc | 1064 | |
a4514ebd JA |
1065 | ret = do_splice_from(pipe, out, off, len, flags); |
1066 | ||
1067 | if (off_out && copy_to_user(off_out, off, sizeof(loff_t))) | |
1068 | ret = -EFAULT; | |
1069 | ||
1070 | return ret; | |
529565dc | 1071 | } |
5274f052 | 1072 | |
3a326a2c | 1073 | pipe = out->f_dentry->d_inode->i_pipe; |
529565dc IM |
1074 | if (pipe) { |
1075 | if (off_out) | |
1076 | return -ESPIPE; | |
b92ce558 JA |
1077 | if (off_in) { |
1078 | if (in->f_op->llseek == no_llseek) | |
1079 | return -EINVAL; | |
cbb7e577 | 1080 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1081 | return -EFAULT; |
cbb7e577 JA |
1082 | off = &offset; |
1083 | } else | |
1084 | off = &in->f_pos; | |
529565dc | 1085 | |
a4514ebd JA |
1086 | ret = do_splice_to(in, off, pipe, len, flags); |
1087 | ||
1088 | if (off_in && copy_to_user(off_in, off, sizeof(loff_t))) | |
1089 | ret = -EFAULT; | |
1090 | ||
1091 | return ret; | |
529565dc | 1092 | } |
5274f052 JA |
1093 | |
1094 | return -EINVAL; | |
1095 | } | |
1096 | ||
912d35f8 JA |
1097 | /* |
1098 | * Map an iov into an array of pages and offset/length tupples. With the | |
1099 | * partial_page structure, we can map several non-contiguous ranges into | |
1100 | * our ones pages[] map instead of splitting that operation into pieces. | |
1101 | * Could easily be exported as a generic helper for other users, in which | |
1102 | * case one would probably want to add a 'max_nr_pages' parameter as well. | |
1103 | */ | |
1104 | static int get_iovec_page_array(const struct iovec __user *iov, | |
1105 | unsigned int nr_vecs, struct page **pages, | |
7afa6fd0 | 1106 | struct partial_page *partial, int aligned) |
912d35f8 JA |
1107 | { |
1108 | int buffers = 0, error = 0; | |
1109 | ||
1110 | /* | |
1111 | * It's ok to take the mmap_sem for reading, even | |
1112 | * across a "get_user()". | |
1113 | */ | |
1114 | down_read(¤t->mm->mmap_sem); | |
1115 | ||
1116 | while (nr_vecs) { | |
1117 | unsigned long off, npages; | |
1118 | void __user *base; | |
1119 | size_t len; | |
1120 | int i; | |
1121 | ||
1122 | /* | |
1123 | * Get user address base and length for this iovec. | |
1124 | */ | |
1125 | error = get_user(base, &iov->iov_base); | |
1126 | if (unlikely(error)) | |
1127 | break; | |
1128 | error = get_user(len, &iov->iov_len); | |
1129 | if (unlikely(error)) | |
1130 | break; | |
1131 | ||
1132 | /* | |
1133 | * Sanity check this iovec. 0 read succeeds. | |
1134 | */ | |
1135 | if (unlikely(!len)) | |
1136 | break; | |
1137 | error = -EFAULT; | |
1138 | if (unlikely(!base)) | |
1139 | break; | |
1140 | ||
1141 | /* | |
1142 | * Get this base offset and number of pages, then map | |
1143 | * in the user pages. | |
1144 | */ | |
1145 | off = (unsigned long) base & ~PAGE_MASK; | |
7afa6fd0 JA |
1146 | |
1147 | /* | |
1148 | * If asked for alignment, the offset must be zero and the | |
1149 | * length a multiple of the PAGE_SIZE. | |
1150 | */ | |
1151 | error = -EINVAL; | |
1152 | if (aligned && (off || len & ~PAGE_MASK)) | |
1153 | break; | |
1154 | ||
912d35f8 JA |
1155 | npages = (off + len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
1156 | if (npages > PIPE_BUFFERS - buffers) | |
1157 | npages = PIPE_BUFFERS - buffers; | |
1158 | ||
1159 | error = get_user_pages(current, current->mm, | |
1160 | (unsigned long) base, npages, 0, 0, | |
1161 | &pages[buffers], NULL); | |
1162 | ||
1163 | if (unlikely(error <= 0)) | |
1164 | break; | |
1165 | ||
1166 | /* | |
1167 | * Fill this contiguous range into the partial page map. | |
1168 | */ | |
1169 | for (i = 0; i < error; i++) { | |
7591489a | 1170 | const int plen = min_t(size_t, len, PAGE_SIZE - off); |
912d35f8 JA |
1171 | |
1172 | partial[buffers].offset = off; | |
1173 | partial[buffers].len = plen; | |
1174 | ||
1175 | off = 0; | |
1176 | len -= plen; | |
1177 | buffers++; | |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * We didn't complete this iov, stop here since it probably | |
1182 | * means we have to move some of this into a pipe to | |
1183 | * be able to continue. | |
1184 | */ | |
1185 | if (len) | |
1186 | break; | |
1187 | ||
1188 | /* | |
1189 | * Don't continue if we mapped fewer pages than we asked for, | |
1190 | * or if we mapped the max number of pages that we have | |
1191 | * room for. | |
1192 | */ | |
1193 | if (error < npages || buffers == PIPE_BUFFERS) | |
1194 | break; | |
1195 | ||
1196 | nr_vecs--; | |
1197 | iov++; | |
1198 | } | |
1199 | ||
1200 | up_read(¤t->mm->mmap_sem); | |
1201 | ||
1202 | if (buffers) | |
1203 | return buffers; | |
1204 | ||
1205 | return error; | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1210 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1211 | * to file). In both cases the output is a pipe, naturally. | |
1212 | * | |
1213 | * Note that vmsplice only supports splicing _from_ user memory to a pipe, | |
1214 | * not the other way around. Splicing from user memory is a simple operation | |
1215 | * that can be supported without any funky alignment restrictions or nasty | |
1216 | * vm tricks. We simply map in the user memory and fill them into a pipe. | |
1217 | * The reverse isn't quite as easy, though. There are two possible solutions | |
1218 | * for that: | |
1219 | * | |
1220 | * - memcpy() the data internally, at which point we might as well just | |
1221 | * do a regular read() on the buffer anyway. | |
1222 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1223 | * has restriction limitations on both ends of the pipe). | |
1224 | * | |
1225 | * Alas, it isn't here. | |
1226 | * | |
1227 | */ | |
1228 | static long do_vmsplice(struct file *file, const struct iovec __user *iov, | |
1229 | unsigned long nr_segs, unsigned int flags) | |
1230 | { | |
1231 | struct pipe_inode_info *pipe = file->f_dentry->d_inode->i_pipe; | |
1232 | struct page *pages[PIPE_BUFFERS]; | |
1233 | struct partial_page partial[PIPE_BUFFERS]; | |
1234 | struct splice_pipe_desc spd = { | |
1235 | .pages = pages, | |
1236 | .partial = partial, | |
1237 | .flags = flags, | |
1238 | .ops = &user_page_pipe_buf_ops, | |
1239 | }; | |
1240 | ||
1241 | if (unlikely(!pipe)) | |
1242 | return -EBADF; | |
1243 | if (unlikely(nr_segs > UIO_MAXIOV)) | |
1244 | return -EINVAL; | |
1245 | else if (unlikely(!nr_segs)) | |
1246 | return 0; | |
1247 | ||
7afa6fd0 JA |
1248 | spd.nr_pages = get_iovec_page_array(iov, nr_segs, pages, partial, |
1249 | flags & SPLICE_F_GIFT); | |
912d35f8 JA |
1250 | if (spd.nr_pages <= 0) |
1251 | return spd.nr_pages; | |
1252 | ||
00522fb4 | 1253 | return splice_to_pipe(pipe, &spd); |
912d35f8 JA |
1254 | } |
1255 | ||
1256 | asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov, | |
1257 | unsigned long nr_segs, unsigned int flags) | |
1258 | { | |
1259 | struct file *file; | |
1260 | long error; | |
1261 | int fput; | |
1262 | ||
1263 | error = -EBADF; | |
1264 | file = fget_light(fd, &fput); | |
1265 | if (file) { | |
1266 | if (file->f_mode & FMODE_WRITE) | |
1267 | error = do_vmsplice(file, iov, nr_segs, flags); | |
1268 | ||
1269 | fput_light(file, fput); | |
1270 | } | |
1271 | ||
1272 | return error; | |
1273 | } | |
1274 | ||
529565dc IM |
1275 | asmlinkage long sys_splice(int fd_in, loff_t __user *off_in, |
1276 | int fd_out, loff_t __user *off_out, | |
1277 | size_t len, unsigned int flags) | |
5274f052 JA |
1278 | { |
1279 | long error; | |
1280 | struct file *in, *out; | |
1281 | int fput_in, fput_out; | |
1282 | ||
1283 | if (unlikely(!len)) | |
1284 | return 0; | |
1285 | ||
1286 | error = -EBADF; | |
529565dc | 1287 | in = fget_light(fd_in, &fput_in); |
5274f052 JA |
1288 | if (in) { |
1289 | if (in->f_mode & FMODE_READ) { | |
529565dc | 1290 | out = fget_light(fd_out, &fput_out); |
5274f052 JA |
1291 | if (out) { |
1292 | if (out->f_mode & FMODE_WRITE) | |
529565dc IM |
1293 | error = do_splice(in, off_in, |
1294 | out, off_out, | |
1295 | len, flags); | |
5274f052 JA |
1296 | fput_light(out, fput_out); |
1297 | } | |
1298 | } | |
1299 | ||
1300 | fput_light(in, fput_in); | |
1301 | } | |
1302 | ||
1303 | return error; | |
1304 | } | |
70524490 JA |
1305 | |
1306 | /* | |
1307 | * Link contents of ipipe to opipe. | |
1308 | */ | |
1309 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1310 | struct pipe_inode_info *opipe, | |
1311 | size_t len, unsigned int flags) | |
1312 | { | |
1313 | struct pipe_buffer *ibuf, *obuf; | |
2a27250e JA |
1314 | int ret, do_wakeup, i, ipipe_first; |
1315 | ||
1316 | ret = do_wakeup = ipipe_first = 0; | |
70524490 JA |
1317 | |
1318 | /* | |
1319 | * Potential ABBA deadlock, work around it by ordering lock | |
1320 | * grabbing by inode address. Otherwise two different processes | |
1321 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1322 | */ | |
1323 | if (ipipe->inode < opipe->inode) { | |
2a27250e | 1324 | ipipe_first = 1; |
70524490 JA |
1325 | mutex_lock(&ipipe->inode->i_mutex); |
1326 | mutex_lock(&opipe->inode->i_mutex); | |
1327 | } else { | |
1328 | mutex_lock(&opipe->inode->i_mutex); | |
1329 | mutex_lock(&ipipe->inode->i_mutex); | |
1330 | } | |
1331 | ||
1332 | for (i = 0;; i++) { | |
1333 | if (!opipe->readers) { | |
1334 | send_sig(SIGPIPE, current, 0); | |
1335 | if (!ret) | |
1336 | ret = -EPIPE; | |
1337 | break; | |
1338 | } | |
1339 | if (ipipe->nrbufs - i) { | |
1340 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (PIPE_BUFFERS - 1)); | |
1341 | ||
1342 | /* | |
1343 | * If we have room, fill this buffer | |
1344 | */ | |
1345 | if (opipe->nrbufs < PIPE_BUFFERS) { | |
1346 | int nbuf = (opipe->curbuf + opipe->nrbufs) & (PIPE_BUFFERS - 1); | |
1347 | ||
1348 | /* | |
1349 | * Get a reference to this pipe buffer, | |
1350 | * so we can copy the contents over. | |
1351 | */ | |
1352 | ibuf->ops->get(ipipe, ibuf); | |
1353 | ||
1354 | obuf = opipe->bufs + nbuf; | |
1355 | *obuf = *ibuf; | |
1356 | ||
7afa6fd0 JA |
1357 | /* |
1358 | * Don't inherit the gift flag, we need to | |
1359 | * prevent multiple steals of this page. | |
1360 | */ | |
1361 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1362 | ||
70524490 JA |
1363 | if (obuf->len > len) |
1364 | obuf->len = len; | |
1365 | ||
1366 | opipe->nrbufs++; | |
1367 | do_wakeup = 1; | |
1368 | ret += obuf->len; | |
1369 | len -= obuf->len; | |
1370 | ||
1371 | if (!len) | |
1372 | break; | |
1373 | if (opipe->nrbufs < PIPE_BUFFERS) | |
1374 | continue; | |
1375 | } | |
1376 | ||
1377 | /* | |
1378 | * We have input available, but no output room. | |
2a27250e JA |
1379 | * If we already copied data, return that. If we |
1380 | * need to drop the opipe lock, it must be ordered | |
1381 | * last to avoid deadlocks. | |
70524490 | 1382 | */ |
2a27250e | 1383 | if ((flags & SPLICE_F_NONBLOCK) || !ipipe_first) { |
70524490 JA |
1384 | if (!ret) |
1385 | ret = -EAGAIN; | |
1386 | break; | |
1387 | } | |
1388 | if (signal_pending(current)) { | |
1389 | if (!ret) | |
1390 | ret = -ERESTARTSYS; | |
1391 | break; | |
1392 | } | |
1393 | if (do_wakeup) { | |
1394 | smp_mb(); | |
1395 | if (waitqueue_active(&opipe->wait)) | |
1396 | wake_up_interruptible(&opipe->wait); | |
1397 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1398 | do_wakeup = 0; | |
1399 | } | |
1400 | ||
1401 | opipe->waiting_writers++; | |
1402 | pipe_wait(opipe); | |
1403 | opipe->waiting_writers--; | |
1404 | continue; | |
1405 | } | |
1406 | ||
1407 | /* | |
1408 | * No input buffers, do the usual checks for available | |
1409 | * writers and blocking and wait if necessary | |
1410 | */ | |
1411 | if (!ipipe->writers) | |
1412 | break; | |
1413 | if (!ipipe->waiting_writers) { | |
1414 | if (ret) | |
1415 | break; | |
1416 | } | |
2a27250e JA |
1417 | /* |
1418 | * pipe_wait() drops the ipipe mutex. To avoid deadlocks | |
1419 | * with another process, we can only safely do that if | |
1420 | * the ipipe lock is ordered last. | |
1421 | */ | |
1422 | if ((flags & SPLICE_F_NONBLOCK) || ipipe_first) { | |
70524490 JA |
1423 | if (!ret) |
1424 | ret = -EAGAIN; | |
1425 | break; | |
1426 | } | |
1427 | if (signal_pending(current)) { | |
1428 | if (!ret) | |
1429 | ret = -ERESTARTSYS; | |
1430 | break; | |
1431 | } | |
1432 | ||
1433 | if (waitqueue_active(&ipipe->wait)) | |
1434 | wake_up_interruptible_sync(&ipipe->wait); | |
1435 | kill_fasync(&ipipe->fasync_writers, SIGIO, POLL_OUT); | |
1436 | ||
1437 | pipe_wait(ipipe); | |
1438 | } | |
1439 | ||
1440 | mutex_unlock(&ipipe->inode->i_mutex); | |
1441 | mutex_unlock(&opipe->inode->i_mutex); | |
1442 | ||
1443 | if (do_wakeup) { | |
1444 | smp_mb(); | |
1445 | if (waitqueue_active(&opipe->wait)) | |
1446 | wake_up_interruptible(&opipe->wait); | |
1447 | kill_fasync(&opipe->fasync_readers, SIGIO, POLL_IN); | |
1448 | } | |
1449 | ||
1450 | return ret; | |
1451 | } | |
1452 | ||
1453 | /* | |
1454 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1455 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1456 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1457 | * applicable one is SPLICE_F_NONBLOCK. | |
1458 | */ | |
1459 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1460 | unsigned int flags) | |
1461 | { | |
1462 | struct pipe_inode_info *ipipe = in->f_dentry->d_inode->i_pipe; | |
1463 | struct pipe_inode_info *opipe = out->f_dentry->d_inode->i_pipe; | |
1464 | ||
1465 | /* | |
1466 | * Link ipipe to the two output pipes, consuming as we go along. | |
1467 | */ | |
1468 | if (ipipe && opipe) | |
1469 | return link_pipe(ipipe, opipe, len, flags); | |
1470 | ||
1471 | return -EINVAL; | |
1472 | } | |
1473 | ||
1474 | asmlinkage long sys_tee(int fdin, int fdout, size_t len, unsigned int flags) | |
1475 | { | |
1476 | struct file *in; | |
1477 | int error, fput_in; | |
1478 | ||
1479 | if (unlikely(!len)) | |
1480 | return 0; | |
1481 | ||
1482 | error = -EBADF; | |
1483 | in = fget_light(fdin, &fput_in); | |
1484 | if (in) { | |
1485 | if (in->f_mode & FMODE_READ) { | |
1486 | int fput_out; | |
1487 | struct file *out = fget_light(fdout, &fput_out); | |
1488 | ||
1489 | if (out) { | |
1490 | if (out->f_mode & FMODE_WRITE) | |
1491 | error = do_tee(in, out, len, flags); | |
1492 | fput_light(out, fput_out); | |
1493 | } | |
1494 | } | |
1495 | fput_light(in, fput_in); | |
1496 | } | |
1497 | ||
1498 | return error; | |
1499 | } |