]>
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 | */ | |
be297968 | 20 | #include <linux/bvec.h> |
5274f052 JA |
21 | #include <linux/fs.h> |
22 | #include <linux/file.h> | |
23 | #include <linux/pagemap.h> | |
d6b29d7c | 24 | #include <linux/splice.h> |
08e552c6 | 25 | #include <linux/memcontrol.h> |
5274f052 | 26 | #include <linux/mm_inline.h> |
5abc97aa | 27 | #include <linux/swap.h> |
4f6f0bd2 | 28 | #include <linux/writeback.h> |
630d9c47 | 29 | #include <linux/export.h> |
4f6f0bd2 | 30 | #include <linux/syscalls.h> |
912d35f8 | 31 | #include <linux/uio.h> |
29ce2058 | 32 | #include <linux/security.h> |
5a0e3ad6 | 33 | #include <linux/gfp.h> |
35f9c09f | 34 | #include <linux/socket.h> |
76b021d0 | 35 | #include <linux/compat.h> |
06ae43f3 | 36 | #include "internal.h" |
5274f052 | 37 | |
83f9135b JA |
38 | /* |
39 | * Attempt to steal a page from a pipe buffer. This should perhaps go into | |
40 | * a vm helper function, it's already simplified quite a bit by the | |
41 | * addition of remove_mapping(). If success is returned, the caller may | |
42 | * attempt to reuse this page for another destination. | |
43 | */ | |
76ad4d11 | 44 | static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe, |
5abc97aa JA |
45 | struct pipe_buffer *buf) |
46 | { | |
47 | struct page *page = buf->page; | |
9e94cd4f | 48 | struct address_space *mapping; |
5abc97aa | 49 | |
9e0267c2 JA |
50 | lock_page(page); |
51 | ||
9e94cd4f JA |
52 | mapping = page_mapping(page); |
53 | if (mapping) { | |
54 | WARN_ON(!PageUptodate(page)); | |
5abc97aa | 55 | |
9e94cd4f JA |
56 | /* |
57 | * At least for ext2 with nobh option, we need to wait on | |
58 | * writeback completing on this page, since we'll remove it | |
59 | * from the pagecache. Otherwise truncate wont wait on the | |
60 | * page, allowing the disk blocks to be reused by someone else | |
61 | * before we actually wrote our data to them. fs corruption | |
62 | * ensues. | |
63 | */ | |
64 | wait_on_page_writeback(page); | |
ad8d6f0a | 65 | |
266cf658 DH |
66 | if (page_has_private(page) && |
67 | !try_to_release_page(page, GFP_KERNEL)) | |
ca39d651 | 68 | goto out_unlock; |
4f6f0bd2 | 69 | |
9e94cd4f JA |
70 | /* |
71 | * If we succeeded in removing the mapping, set LRU flag | |
72 | * and return good. | |
73 | */ | |
74 | if (remove_mapping(mapping, page)) { | |
75 | buf->flags |= PIPE_BUF_FLAG_LRU; | |
76 | return 0; | |
77 | } | |
9e0267c2 | 78 | } |
5abc97aa | 79 | |
9e94cd4f JA |
80 | /* |
81 | * Raced with truncate or failed to remove page from current | |
82 | * address space, unlock and return failure. | |
83 | */ | |
ca39d651 | 84 | out_unlock: |
9e94cd4f JA |
85 | unlock_page(page); |
86 | return 1; | |
5abc97aa JA |
87 | } |
88 | ||
76ad4d11 | 89 | static void page_cache_pipe_buf_release(struct pipe_inode_info *pipe, |
5274f052 JA |
90 | struct pipe_buffer *buf) |
91 | { | |
09cbfeaf | 92 | put_page(buf->page); |
1432873a | 93 | buf->flags &= ~PIPE_BUF_FLAG_LRU; |
5274f052 JA |
94 | } |
95 | ||
0845718d JA |
96 | /* |
97 | * Check whether the contents of buf is OK to access. Since the content | |
98 | * is a page cache page, IO may be in flight. | |
99 | */ | |
cac36bb0 JA |
100 | static int page_cache_pipe_buf_confirm(struct pipe_inode_info *pipe, |
101 | struct pipe_buffer *buf) | |
5274f052 JA |
102 | { |
103 | struct page *page = buf->page; | |
49d0b21b | 104 | int err; |
5274f052 JA |
105 | |
106 | if (!PageUptodate(page)) { | |
49d0b21b JA |
107 | lock_page(page); |
108 | ||
109 | /* | |
110 | * Page got truncated/unhashed. This will cause a 0-byte | |
73d62d83 | 111 | * splice, if this is the first page. |
49d0b21b JA |
112 | */ |
113 | if (!page->mapping) { | |
114 | err = -ENODATA; | |
115 | goto error; | |
116 | } | |
5274f052 | 117 | |
49d0b21b | 118 | /* |
73d62d83 | 119 | * Uh oh, read-error from disk. |
49d0b21b JA |
120 | */ |
121 | if (!PageUptodate(page)) { | |
122 | err = -EIO; | |
123 | goto error; | |
124 | } | |
125 | ||
126 | /* | |
f84d7519 | 127 | * Page is ok afterall, we are done. |
49d0b21b | 128 | */ |
5274f052 | 129 | unlock_page(page); |
5274f052 JA |
130 | } |
131 | ||
f84d7519 | 132 | return 0; |
49d0b21b JA |
133 | error: |
134 | unlock_page(page); | |
f84d7519 | 135 | return err; |
70524490 JA |
136 | } |
137 | ||
708e3508 | 138 | const struct pipe_buf_operations page_cache_pipe_buf_ops = { |
5274f052 | 139 | .can_merge = 0, |
cac36bb0 | 140 | .confirm = page_cache_pipe_buf_confirm, |
5274f052 | 141 | .release = page_cache_pipe_buf_release, |
5abc97aa | 142 | .steal = page_cache_pipe_buf_steal, |
f84d7519 | 143 | .get = generic_pipe_buf_get, |
5274f052 JA |
144 | }; |
145 | ||
912d35f8 JA |
146 | static int user_page_pipe_buf_steal(struct pipe_inode_info *pipe, |
147 | struct pipe_buffer *buf) | |
148 | { | |
7afa6fd0 JA |
149 | if (!(buf->flags & PIPE_BUF_FLAG_GIFT)) |
150 | return 1; | |
151 | ||
1432873a | 152 | buf->flags |= PIPE_BUF_FLAG_LRU; |
330ab716 | 153 | return generic_pipe_buf_steal(pipe, buf); |
912d35f8 JA |
154 | } |
155 | ||
d4c3cca9 | 156 | static const struct pipe_buf_operations user_page_pipe_buf_ops = { |
912d35f8 | 157 | .can_merge = 0, |
cac36bb0 | 158 | .confirm = generic_pipe_buf_confirm, |
912d35f8 JA |
159 | .release = page_cache_pipe_buf_release, |
160 | .steal = user_page_pipe_buf_steal, | |
f84d7519 | 161 | .get = generic_pipe_buf_get, |
912d35f8 JA |
162 | }; |
163 | ||
825cdcb1 NK |
164 | static void wakeup_pipe_readers(struct pipe_inode_info *pipe) |
165 | { | |
166 | smp_mb(); | |
167 | if (waitqueue_active(&pipe->wait)) | |
168 | wake_up_interruptible(&pipe->wait); | |
169 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
170 | } | |
171 | ||
932cc6d4 JA |
172 | /** |
173 | * splice_to_pipe - fill passed data into a pipe | |
174 | * @pipe: pipe to fill | |
175 | * @spd: data to fill | |
176 | * | |
177 | * Description: | |
79685b8d | 178 | * @spd contains a map of pages and len/offset tuples, along with |
932cc6d4 JA |
179 | * the struct pipe_buf_operations associated with these pages. This |
180 | * function will link that data to the pipe. | |
181 | * | |
83f9135b | 182 | */ |
d6b29d7c JA |
183 | ssize_t splice_to_pipe(struct pipe_inode_info *pipe, |
184 | struct splice_pipe_desc *spd) | |
5274f052 | 185 | { |
00de00bd | 186 | unsigned int spd_pages = spd->nr_pages; |
8924feff | 187 | int ret = 0, page_nr = 0; |
5274f052 | 188 | |
d6785d91 RV |
189 | if (!spd_pages) |
190 | return 0; | |
191 | ||
8924feff AV |
192 | if (unlikely(!pipe->readers)) { |
193 | send_sig(SIGPIPE, current, 0); | |
194 | ret = -EPIPE; | |
195 | goto out; | |
196 | } | |
5274f052 | 197 | |
8924feff AV |
198 | while (pipe->nrbufs < pipe->buffers) { |
199 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
200 | struct pipe_buffer *buf = pipe->bufs + newbuf; | |
5274f052 | 201 | |
8924feff AV |
202 | buf->page = spd->pages[page_nr]; |
203 | buf->offset = spd->partial[page_nr].offset; | |
204 | buf->len = spd->partial[page_nr].len; | |
205 | buf->private = spd->partial[page_nr].private; | |
206 | buf->ops = spd->ops; | |
5a81e6a1 | 207 | buf->flags = 0; |
5274f052 | 208 | |
8924feff AV |
209 | pipe->nrbufs++; |
210 | page_nr++; | |
211 | ret += buf->len; | |
29e35094 | 212 | |
8924feff | 213 | if (!--spd->nr_pages) |
5274f052 | 214 | break; |
5274f052 JA |
215 | } |
216 | ||
8924feff AV |
217 | if (!ret) |
218 | ret = -EAGAIN; | |
5274f052 | 219 | |
8924feff | 220 | out: |
00de00bd | 221 | while (page_nr < spd_pages) |
bbdfc2f7 | 222 | spd->spd_release(spd, page_nr++); |
5274f052 JA |
223 | |
224 | return ret; | |
225 | } | |
2b514574 | 226 | EXPORT_SYMBOL_GPL(splice_to_pipe); |
5274f052 | 227 | |
79fddc4e AV |
228 | ssize_t add_to_pipe(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
229 | { | |
230 | int ret; | |
231 | ||
232 | if (unlikely(!pipe->readers)) { | |
233 | send_sig(SIGPIPE, current, 0); | |
234 | ret = -EPIPE; | |
235 | } else if (pipe->nrbufs == pipe->buffers) { | |
236 | ret = -EAGAIN; | |
237 | } else { | |
238 | int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1); | |
239 | pipe->bufs[newbuf] = *buf; | |
240 | pipe->nrbufs++; | |
241 | return buf->len; | |
242 | } | |
a779638c | 243 | pipe_buf_release(pipe, buf); |
79fddc4e AV |
244 | return ret; |
245 | } | |
246 | EXPORT_SYMBOL(add_to_pipe); | |
247 | ||
708e3508 | 248 | void spd_release_page(struct splice_pipe_desc *spd, unsigned int i) |
bbdfc2f7 | 249 | { |
09cbfeaf | 250 | put_page(spd->pages[i]); |
bbdfc2f7 JA |
251 | } |
252 | ||
35f3d14d JA |
253 | /* |
254 | * Check if we need to grow the arrays holding pages and partial page | |
255 | * descriptions. | |
256 | */ | |
047fe360 | 257 | int splice_grow_spd(const struct pipe_inode_info *pipe, struct splice_pipe_desc *spd) |
35f3d14d | 258 | { |
047fe360 ED |
259 | unsigned int buffers = ACCESS_ONCE(pipe->buffers); |
260 | ||
261 | spd->nr_pages_max = buffers; | |
262 | if (buffers <= PIPE_DEF_BUFFERS) | |
35f3d14d JA |
263 | return 0; |
264 | ||
047fe360 ED |
265 | spd->pages = kmalloc(buffers * sizeof(struct page *), GFP_KERNEL); |
266 | spd->partial = kmalloc(buffers * sizeof(struct partial_page), GFP_KERNEL); | |
35f3d14d JA |
267 | |
268 | if (spd->pages && spd->partial) | |
269 | return 0; | |
270 | ||
271 | kfree(spd->pages); | |
272 | kfree(spd->partial); | |
273 | return -ENOMEM; | |
274 | } | |
275 | ||
047fe360 | 276 | void splice_shrink_spd(struct splice_pipe_desc *spd) |
35f3d14d | 277 | { |
047fe360 | 278 | if (spd->nr_pages_max <= PIPE_DEF_BUFFERS) |
35f3d14d JA |
279 | return; |
280 | ||
281 | kfree(spd->pages); | |
282 | kfree(spd->partial); | |
283 | } | |
284 | ||
83f9135b JA |
285 | /** |
286 | * generic_file_splice_read - splice data from file to a pipe | |
287 | * @in: file to splice from | |
932cc6d4 | 288 | * @ppos: position in @in |
83f9135b JA |
289 | * @pipe: pipe to splice to |
290 | * @len: number of bytes to splice | |
291 | * @flags: splice modifier flags | |
292 | * | |
932cc6d4 JA |
293 | * Description: |
294 | * Will read pages from given file and fill them into a pipe. Can be | |
82c156f8 | 295 | * used as long as it has more or less sane ->read_iter(). |
932cc6d4 | 296 | * |
83f9135b | 297 | */ |
cbb7e577 JA |
298 | ssize_t generic_file_splice_read(struct file *in, loff_t *ppos, |
299 | struct pipe_inode_info *pipe, size_t len, | |
300 | unsigned int flags) | |
5274f052 | 301 | { |
82c156f8 AV |
302 | struct iov_iter to; |
303 | struct kiocb kiocb; | |
82c156f8 | 304 | int idx, ret; |
be64f884 | 305 | |
82c156f8 AV |
306 | iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len); |
307 | idx = to.idx; | |
308 | init_sync_kiocb(&kiocb, in); | |
309 | kiocb.ki_pos = *ppos; | |
310 | ret = in->f_op->read_iter(&kiocb, &to); | |
723590ed | 311 | if (ret > 0) { |
82c156f8 | 312 | *ppos = kiocb.ki_pos; |
723590ed | 313 | file_accessed(in); |
82c156f8 | 314 | } else if (ret < 0) { |
c3a69024 AV |
315 | to.idx = idx; |
316 | to.iov_offset = 0; | |
317 | iov_iter_advance(&to, 0); /* to free what was emitted */ | |
82c156f8 AV |
318 | /* |
319 | * callers of ->splice_read() expect -EAGAIN on | |
320 | * "can't put anything in there", rather than -EFAULT. | |
321 | */ | |
322 | if (ret == -EFAULT) | |
323 | ret = -EAGAIN; | |
723590ed | 324 | } |
5274f052 JA |
325 | |
326 | return ret; | |
327 | } | |
059a8f37 JA |
328 | EXPORT_SYMBOL(generic_file_splice_read); |
329 | ||
241699cd | 330 | const struct pipe_buf_operations default_pipe_buf_ops = { |
6818173b | 331 | .can_merge = 0, |
6818173b MS |
332 | .confirm = generic_pipe_buf_confirm, |
333 | .release = generic_pipe_buf_release, | |
334 | .steal = generic_pipe_buf_steal, | |
335 | .get = generic_pipe_buf_get, | |
336 | }; | |
337 | ||
28a625cb MS |
338 | static int generic_pipe_buf_nosteal(struct pipe_inode_info *pipe, |
339 | struct pipe_buffer *buf) | |
340 | { | |
341 | return 1; | |
342 | } | |
343 | ||
344 | /* Pipe buffer operations for a socket and similar. */ | |
345 | const struct pipe_buf_operations nosteal_pipe_buf_ops = { | |
346 | .can_merge = 0, | |
28a625cb MS |
347 | .confirm = generic_pipe_buf_confirm, |
348 | .release = generic_pipe_buf_release, | |
349 | .steal = generic_pipe_buf_nosteal, | |
350 | .get = generic_pipe_buf_get, | |
351 | }; | |
352 | EXPORT_SYMBOL(nosteal_pipe_buf_ops); | |
353 | ||
523ac9af | 354 | static ssize_t kernel_readv(struct file *file, const struct kvec *vec, |
6818173b MS |
355 | unsigned long vlen, loff_t offset) |
356 | { | |
357 | mm_segment_t old_fs; | |
358 | loff_t pos = offset; | |
359 | ssize_t res; | |
360 | ||
361 | old_fs = get_fs(); | |
362 | set_fs(get_ds()); | |
363 | /* The cast to a user pointer is valid due to the set_fs() */ | |
793b80ef | 364 | res = vfs_readv(file, (const struct iovec __user *)vec, vlen, &pos, 0); |
6818173b MS |
365 | set_fs(old_fs); |
366 | ||
367 | return res; | |
368 | } | |
369 | ||
7bb307e8 | 370 | ssize_t kernel_write(struct file *file, const char *buf, size_t count, |
b2858d7d | 371 | loff_t pos) |
0b0a47f5 MS |
372 | { |
373 | mm_segment_t old_fs; | |
374 | ssize_t res; | |
375 | ||
376 | old_fs = get_fs(); | |
377 | set_fs(get_ds()); | |
378 | /* The cast to a user pointer is valid due to the set_fs() */ | |
7bb307e8 | 379 | res = vfs_write(file, (__force const char __user *)buf, count, &pos); |
0b0a47f5 MS |
380 | set_fs(old_fs); |
381 | ||
382 | return res; | |
383 | } | |
7bb307e8 | 384 | EXPORT_SYMBOL(kernel_write); |
0b0a47f5 | 385 | |
82c156f8 | 386 | static ssize_t default_file_splice_read(struct file *in, loff_t *ppos, |
6818173b MS |
387 | struct pipe_inode_info *pipe, size_t len, |
388 | unsigned int flags) | |
389 | { | |
523ac9af AV |
390 | struct kvec *vec, __vec[PIPE_DEF_BUFFERS]; |
391 | struct iov_iter to; | |
392 | struct page **pages; | |
6818173b | 393 | unsigned int nr_pages; |
523ac9af | 394 | size_t offset, dummy, copied = 0; |
6818173b | 395 | ssize_t res; |
6818173b | 396 | int i; |
35f3d14d | 397 | |
523ac9af AV |
398 | if (pipe->nrbufs == pipe->buffers) |
399 | return -EAGAIN; | |
35f3d14d | 400 | |
523ac9af AV |
401 | /* |
402 | * Try to keep page boundaries matching to source pagecache ones - | |
403 | * it probably won't be much help, but... | |
404 | */ | |
09cbfeaf | 405 | offset = *ppos & ~PAGE_MASK; |
6818173b | 406 | |
523ac9af | 407 | iov_iter_pipe(&to, ITER_PIPE | READ, pipe, len + offset); |
6818173b | 408 | |
523ac9af AV |
409 | res = iov_iter_get_pages_alloc(&to, &pages, len + offset, &dummy); |
410 | if (res <= 0) | |
411 | return -ENOMEM; | |
6818173b | 412 | |
8e54cada AV |
413 | BUG_ON(dummy); |
414 | nr_pages = DIV_ROUND_UP(res, PAGE_SIZE); | |
6818173b | 415 | |
523ac9af AV |
416 | vec = __vec; |
417 | if (nr_pages > PIPE_DEF_BUFFERS) { | |
418 | vec = kmalloc(nr_pages * sizeof(struct kvec), GFP_KERNEL); | |
419 | if (unlikely(!vec)) { | |
420 | res = -ENOMEM; | |
421 | goto out; | |
422 | } | |
77f6bf57 | 423 | } |
6818173b | 424 | |
523ac9af AV |
425 | pipe->bufs[to.idx].offset = offset; |
426 | pipe->bufs[to.idx].len -= offset; | |
427 | ||
428 | for (i = 0; i < nr_pages; i++) { | |
429 | size_t this_len = min_t(size_t, len, PAGE_SIZE - offset); | |
430 | vec[i].iov_base = page_address(pages[i]) + offset; | |
431 | vec[i].iov_len = this_len; | |
432 | len -= this_len; | |
433 | offset = 0; | |
6818173b | 434 | } |
6818173b | 435 | |
523ac9af AV |
436 | res = kernel_readv(in, vec, nr_pages, *ppos); |
437 | if (res > 0) { | |
438 | copied = res; | |
6818173b | 439 | *ppos += res; |
523ac9af | 440 | } |
6818173b | 441 | |
35f3d14d JA |
442 | if (vec != __vec) |
443 | kfree(vec); | |
523ac9af AV |
444 | out: |
445 | for (i = 0; i < nr_pages; i++) | |
446 | put_page(pages[i]); | |
447 | kvfree(pages); | |
448 | iov_iter_advance(&to, copied); /* truncates and discards */ | |
6818173b | 449 | return res; |
6818173b | 450 | } |
6818173b | 451 | |
5274f052 | 452 | /* |
4f6f0bd2 | 453 | * Send 'sd->len' bytes to socket from 'sd->file' at position 'sd->pos' |
016b661e | 454 | * using sendpage(). Return the number of bytes sent. |
5274f052 | 455 | */ |
76ad4d11 | 456 | static int pipe_to_sendpage(struct pipe_inode_info *pipe, |
5274f052 JA |
457 | struct pipe_buffer *buf, struct splice_desc *sd) |
458 | { | |
6a14b90b | 459 | struct file *file = sd->u.file; |
5274f052 | 460 | loff_t pos = sd->pos; |
a8adbe37 | 461 | int more; |
5274f052 | 462 | |
72c2d531 | 463 | if (!likely(file->f_op->sendpage)) |
a8adbe37 MM |
464 | return -EINVAL; |
465 | ||
35f9c09f | 466 | more = (sd->flags & SPLICE_F_MORE) ? MSG_MORE : 0; |
ae62ca7b ED |
467 | |
468 | if (sd->len < sd->total_len && pipe->nrbufs > 1) | |
35f9c09f | 469 | more |= MSG_SENDPAGE_NOTLAST; |
ae62ca7b | 470 | |
a8adbe37 MM |
471 | return file->f_op->sendpage(file, buf->page, buf->offset, |
472 | sd->len, &pos, more); | |
5274f052 JA |
473 | } |
474 | ||
b3c2d2dd MS |
475 | static void wakeup_pipe_writers(struct pipe_inode_info *pipe) |
476 | { | |
477 | smp_mb(); | |
478 | if (waitqueue_active(&pipe->wait)) | |
479 | wake_up_interruptible(&pipe->wait); | |
480 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
481 | } | |
482 | ||
932cc6d4 | 483 | /** |
b3c2d2dd | 484 | * splice_from_pipe_feed - feed available data from a pipe to a file |
932cc6d4 JA |
485 | * @pipe: pipe to splice from |
486 | * @sd: information to @actor | |
487 | * @actor: handler that splices the data | |
488 | * | |
489 | * Description: | |
b3c2d2dd MS |
490 | * This function loops over the pipe and calls @actor to do the |
491 | * actual moving of a single struct pipe_buffer to the desired | |
492 | * destination. It returns when there's no more buffers left in | |
493 | * the pipe or if the requested number of bytes (@sd->total_len) | |
494 | * have been copied. It returns a positive number (one) if the | |
495 | * pipe needs to be filled with more data, zero if the required | |
496 | * number of bytes have been copied and -errno on error. | |
932cc6d4 | 497 | * |
b3c2d2dd MS |
498 | * This, together with splice_from_pipe_{begin,end,next}, may be |
499 | * used to implement the functionality of __splice_from_pipe() when | |
500 | * locking is required around copying the pipe buffers to the | |
501 | * destination. | |
83f9135b | 502 | */ |
96f9bc8f | 503 | static int splice_from_pipe_feed(struct pipe_inode_info *pipe, struct splice_desc *sd, |
b3c2d2dd | 504 | splice_actor *actor) |
5274f052 | 505 | { |
b3c2d2dd | 506 | int ret; |
5274f052 | 507 | |
b3c2d2dd MS |
508 | while (pipe->nrbufs) { |
509 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
5274f052 | 510 | |
b3c2d2dd MS |
511 | sd->len = buf->len; |
512 | if (sd->len > sd->total_len) | |
513 | sd->len = sd->total_len; | |
5274f052 | 514 | |
fba597db | 515 | ret = pipe_buf_confirm(pipe, buf); |
a8adbe37 | 516 | if (unlikely(ret)) { |
b3c2d2dd MS |
517 | if (ret == -ENODATA) |
518 | ret = 0; | |
519 | return ret; | |
520 | } | |
a8adbe37 MM |
521 | |
522 | ret = actor(pipe, buf, sd); | |
523 | if (ret <= 0) | |
524 | return ret; | |
525 | ||
b3c2d2dd MS |
526 | buf->offset += ret; |
527 | buf->len -= ret; | |
528 | ||
529 | sd->num_spliced += ret; | |
530 | sd->len -= ret; | |
531 | sd->pos += ret; | |
532 | sd->total_len -= ret; | |
533 | ||
534 | if (!buf->len) { | |
a779638c | 535 | pipe_buf_release(pipe, buf); |
35f3d14d | 536 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); |
b3c2d2dd | 537 | pipe->nrbufs--; |
6447a3cf | 538 | if (pipe->files) |
b3c2d2dd MS |
539 | sd->need_wakeup = true; |
540 | } | |
5274f052 | 541 | |
b3c2d2dd MS |
542 | if (!sd->total_len) |
543 | return 0; | |
544 | } | |
5274f052 | 545 | |
b3c2d2dd MS |
546 | return 1; |
547 | } | |
5274f052 | 548 | |
b3c2d2dd MS |
549 | /** |
550 | * splice_from_pipe_next - wait for some data to splice from | |
551 | * @pipe: pipe to splice from | |
552 | * @sd: information about the splice operation | |
553 | * | |
554 | * Description: | |
555 | * This function will wait for some data and return a positive | |
556 | * value (one) if pipe buffers are available. It will return zero | |
557 | * or -errno if no more data needs to be spliced. | |
558 | */ | |
96f9bc8f | 559 | static int splice_from_pipe_next(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd | 560 | { |
c725bfce JK |
561 | /* |
562 | * Check for signal early to make process killable when there are | |
563 | * always buffers available | |
564 | */ | |
565 | if (signal_pending(current)) | |
566 | return -ERESTARTSYS; | |
567 | ||
b3c2d2dd MS |
568 | while (!pipe->nrbufs) { |
569 | if (!pipe->writers) | |
570 | return 0; | |
016b661e | 571 | |
b3c2d2dd MS |
572 | if (!pipe->waiting_writers && sd->num_spliced) |
573 | return 0; | |
73d62d83 | 574 | |
b3c2d2dd MS |
575 | if (sd->flags & SPLICE_F_NONBLOCK) |
576 | return -EAGAIN; | |
5274f052 | 577 | |
b3c2d2dd MS |
578 | if (signal_pending(current)) |
579 | return -ERESTARTSYS; | |
5274f052 | 580 | |
b3c2d2dd MS |
581 | if (sd->need_wakeup) { |
582 | wakeup_pipe_writers(pipe); | |
583 | sd->need_wakeup = false; | |
5274f052 JA |
584 | } |
585 | ||
b3c2d2dd MS |
586 | pipe_wait(pipe); |
587 | } | |
29e35094 | 588 | |
b3c2d2dd MS |
589 | return 1; |
590 | } | |
5274f052 | 591 | |
b3c2d2dd MS |
592 | /** |
593 | * splice_from_pipe_begin - start splicing from pipe | |
b80901bb | 594 | * @sd: information about the splice operation |
b3c2d2dd MS |
595 | * |
596 | * Description: | |
597 | * This function should be called before a loop containing | |
598 | * splice_from_pipe_next() and splice_from_pipe_feed() to | |
599 | * initialize the necessary fields of @sd. | |
600 | */ | |
96f9bc8f | 601 | static void splice_from_pipe_begin(struct splice_desc *sd) |
b3c2d2dd MS |
602 | { |
603 | sd->num_spliced = 0; | |
604 | sd->need_wakeup = false; | |
605 | } | |
5274f052 | 606 | |
b3c2d2dd MS |
607 | /** |
608 | * splice_from_pipe_end - finish splicing from pipe | |
609 | * @pipe: pipe to splice from | |
610 | * @sd: information about the splice operation | |
611 | * | |
612 | * Description: | |
613 | * This function will wake up pipe writers if necessary. It should | |
614 | * be called after a loop containing splice_from_pipe_next() and | |
615 | * splice_from_pipe_feed(). | |
616 | */ | |
96f9bc8f | 617 | static void splice_from_pipe_end(struct pipe_inode_info *pipe, struct splice_desc *sd) |
b3c2d2dd MS |
618 | { |
619 | if (sd->need_wakeup) | |
620 | wakeup_pipe_writers(pipe); | |
621 | } | |
5274f052 | 622 | |
b3c2d2dd MS |
623 | /** |
624 | * __splice_from_pipe - splice data from a pipe to given actor | |
625 | * @pipe: pipe to splice from | |
626 | * @sd: information to @actor | |
627 | * @actor: handler that splices the data | |
628 | * | |
629 | * Description: | |
630 | * This function does little more than loop over the pipe and call | |
631 | * @actor to do the actual moving of a single struct pipe_buffer to | |
632 | * the desired destination. See pipe_to_file, pipe_to_sendpage, or | |
633 | * pipe_to_user. | |
634 | * | |
635 | */ | |
636 | ssize_t __splice_from_pipe(struct pipe_inode_info *pipe, struct splice_desc *sd, | |
637 | splice_actor *actor) | |
638 | { | |
639 | int ret; | |
5274f052 | 640 | |
b3c2d2dd MS |
641 | splice_from_pipe_begin(sd); |
642 | do { | |
c2489e07 | 643 | cond_resched(); |
b3c2d2dd MS |
644 | ret = splice_from_pipe_next(pipe, sd); |
645 | if (ret > 0) | |
646 | ret = splice_from_pipe_feed(pipe, sd, actor); | |
647 | } while (ret > 0); | |
648 | splice_from_pipe_end(pipe, sd); | |
649 | ||
650 | return sd->num_spliced ? sd->num_spliced : ret; | |
5274f052 | 651 | } |
40bee44e | 652 | EXPORT_SYMBOL(__splice_from_pipe); |
5274f052 | 653 | |
932cc6d4 JA |
654 | /** |
655 | * splice_from_pipe - splice data from a pipe to a file | |
656 | * @pipe: pipe to splice from | |
657 | * @out: file to splice to | |
658 | * @ppos: position in @out | |
659 | * @len: how many bytes to splice | |
660 | * @flags: splice modifier flags | |
661 | * @actor: handler that splices the data | |
662 | * | |
663 | * Description: | |
2933970b | 664 | * See __splice_from_pipe. This function locks the pipe inode, |
932cc6d4 JA |
665 | * otherwise it's identical to __splice_from_pipe(). |
666 | * | |
667 | */ | |
6da61809 MF |
668 | ssize_t splice_from_pipe(struct pipe_inode_info *pipe, struct file *out, |
669 | loff_t *ppos, size_t len, unsigned int flags, | |
670 | splice_actor *actor) | |
671 | { | |
672 | ssize_t ret; | |
c66ab6fa JA |
673 | struct splice_desc sd = { |
674 | .total_len = len, | |
675 | .flags = flags, | |
676 | .pos = *ppos, | |
6a14b90b | 677 | .u.file = out, |
c66ab6fa | 678 | }; |
6da61809 | 679 | |
61e0d47c | 680 | pipe_lock(pipe); |
c66ab6fa | 681 | ret = __splice_from_pipe(pipe, &sd, actor); |
61e0d47c | 682 | pipe_unlock(pipe); |
6da61809 MF |
683 | |
684 | return ret; | |
685 | } | |
686 | ||
8d020765 AV |
687 | /** |
688 | * iter_file_splice_write - splice data from a pipe to a file | |
689 | * @pipe: pipe info | |
690 | * @out: file to write to | |
691 | * @ppos: position in @out | |
692 | * @len: number of bytes to splice | |
693 | * @flags: splice modifier flags | |
694 | * | |
695 | * Description: | |
696 | * Will either move or copy pages (determined by @flags options) from | |
697 | * the given pipe inode to the given file. | |
698 | * This one is ->write_iter-based. | |
699 | * | |
700 | */ | |
701 | ssize_t | |
702 | iter_file_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
703 | loff_t *ppos, size_t len, unsigned int flags) | |
704 | { | |
705 | struct splice_desc sd = { | |
706 | .total_len = len, | |
707 | .flags = flags, | |
708 | .pos = *ppos, | |
709 | .u.file = out, | |
710 | }; | |
711 | int nbufs = pipe->buffers; | |
712 | struct bio_vec *array = kcalloc(nbufs, sizeof(struct bio_vec), | |
713 | GFP_KERNEL); | |
714 | ssize_t ret; | |
715 | ||
716 | if (unlikely(!array)) | |
717 | return -ENOMEM; | |
718 | ||
719 | pipe_lock(pipe); | |
720 | ||
721 | splice_from_pipe_begin(&sd); | |
722 | while (sd.total_len) { | |
723 | struct iov_iter from; | |
8d020765 AV |
724 | size_t left; |
725 | int n, idx; | |
726 | ||
727 | ret = splice_from_pipe_next(pipe, &sd); | |
728 | if (ret <= 0) | |
729 | break; | |
730 | ||
731 | if (unlikely(nbufs < pipe->buffers)) { | |
732 | kfree(array); | |
733 | nbufs = pipe->buffers; | |
734 | array = kcalloc(nbufs, sizeof(struct bio_vec), | |
735 | GFP_KERNEL); | |
736 | if (!array) { | |
737 | ret = -ENOMEM; | |
738 | break; | |
739 | } | |
740 | } | |
741 | ||
742 | /* build the vector */ | |
743 | left = sd.total_len; | |
744 | for (n = 0, idx = pipe->curbuf; left && n < pipe->nrbufs; n++, idx++) { | |
745 | struct pipe_buffer *buf = pipe->bufs + idx; | |
746 | size_t this_len = buf->len; | |
747 | ||
748 | if (this_len > left) | |
749 | this_len = left; | |
750 | ||
751 | if (idx == pipe->buffers - 1) | |
752 | idx = -1; | |
753 | ||
fba597db | 754 | ret = pipe_buf_confirm(pipe, buf); |
8d020765 AV |
755 | if (unlikely(ret)) { |
756 | if (ret == -ENODATA) | |
757 | ret = 0; | |
758 | goto done; | |
759 | } | |
760 | ||
761 | array[n].bv_page = buf->page; | |
762 | array[n].bv_len = this_len; | |
763 | array[n].bv_offset = buf->offset; | |
764 | left -= this_len; | |
765 | } | |
766 | ||
05afcb77 AV |
767 | iov_iter_bvec(&from, ITER_BVEC | WRITE, array, n, |
768 | sd.total_len - left); | |
dbe4e192 | 769 | ret = vfs_iter_write(out, &from, &sd.pos); |
8d020765 AV |
770 | if (ret <= 0) |
771 | break; | |
772 | ||
773 | sd.num_spliced += ret; | |
774 | sd.total_len -= ret; | |
dbe4e192 | 775 | *ppos = sd.pos; |
8d020765 AV |
776 | |
777 | /* dismiss the fully eaten buffers, adjust the partial one */ | |
778 | while (ret) { | |
779 | struct pipe_buffer *buf = pipe->bufs + pipe->curbuf; | |
780 | if (ret >= buf->len) { | |
8d020765 AV |
781 | ret -= buf->len; |
782 | buf->len = 0; | |
a779638c | 783 | pipe_buf_release(pipe, buf); |
8d020765 AV |
784 | pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1); |
785 | pipe->nrbufs--; | |
786 | if (pipe->files) | |
787 | sd.need_wakeup = true; | |
788 | } else { | |
789 | buf->offset += ret; | |
790 | buf->len -= ret; | |
791 | ret = 0; | |
792 | } | |
793 | } | |
794 | } | |
795 | done: | |
796 | kfree(array); | |
797 | splice_from_pipe_end(pipe, &sd); | |
798 | ||
799 | pipe_unlock(pipe); | |
800 | ||
801 | if (sd.num_spliced) | |
802 | ret = sd.num_spliced; | |
803 | ||
804 | return ret; | |
805 | } | |
806 | ||
807 | EXPORT_SYMBOL(iter_file_splice_write); | |
808 | ||
b2858d7d MS |
809 | static int write_pipe_buf(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
810 | struct splice_desc *sd) | |
0b0a47f5 | 811 | { |
b2858d7d MS |
812 | int ret; |
813 | void *data; | |
06ae43f3 | 814 | loff_t tmp = sd->pos; |
b2858d7d | 815 | |
fbb32750 | 816 | data = kmap(buf->page); |
06ae43f3 | 817 | ret = __kernel_write(sd->u.file, data + buf->offset, sd->len, &tmp); |
fbb32750 | 818 | kunmap(buf->page); |
b2858d7d MS |
819 | |
820 | return ret; | |
0b0a47f5 MS |
821 | } |
822 | ||
823 | static ssize_t default_file_splice_write(struct pipe_inode_info *pipe, | |
824 | struct file *out, loff_t *ppos, | |
825 | size_t len, unsigned int flags) | |
826 | { | |
b2858d7d | 827 | ssize_t ret; |
0b0a47f5 | 828 | |
b2858d7d MS |
829 | ret = splice_from_pipe(pipe, out, ppos, len, flags, write_pipe_buf); |
830 | if (ret > 0) | |
831 | *ppos += ret; | |
0b0a47f5 | 832 | |
b2858d7d | 833 | return ret; |
0b0a47f5 MS |
834 | } |
835 | ||
83f9135b JA |
836 | /** |
837 | * generic_splice_sendpage - splice data from a pipe to a socket | |
932cc6d4 | 838 | * @pipe: pipe to splice from |
83f9135b | 839 | * @out: socket to write to |
932cc6d4 | 840 | * @ppos: position in @out |
83f9135b JA |
841 | * @len: number of bytes to splice |
842 | * @flags: splice modifier flags | |
843 | * | |
932cc6d4 JA |
844 | * Description: |
845 | * Will send @len bytes from the pipe to a network socket. No data copying | |
846 | * is involved. | |
83f9135b JA |
847 | * |
848 | */ | |
3a326a2c | 849 | ssize_t generic_splice_sendpage(struct pipe_inode_info *pipe, struct file *out, |
cbb7e577 | 850 | loff_t *ppos, size_t len, unsigned int flags) |
5274f052 | 851 | { |
00522fb4 | 852 | return splice_from_pipe(pipe, out, ppos, len, flags, pipe_to_sendpage); |
5274f052 JA |
853 | } |
854 | ||
059a8f37 | 855 | EXPORT_SYMBOL(generic_splice_sendpage); |
a0f06780 | 856 | |
83f9135b JA |
857 | /* |
858 | * Attempt to initiate a splice from pipe to file. | |
859 | */ | |
e14748e8 SF |
860 | long do_splice_from(struct pipe_inode_info *pipe, struct file *out, |
861 | loff_t *ppos, size_t len, unsigned int flags) | |
5274f052 | 862 | { |
0b0a47f5 MS |
863 | ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, |
864 | loff_t *, size_t, unsigned int); | |
5274f052 | 865 | |
72c2d531 | 866 | if (out->f_op->splice_write) |
cc56f7de CG |
867 | splice_write = out->f_op->splice_write; |
868 | else | |
0b0a47f5 MS |
869 | splice_write = default_file_splice_write; |
870 | ||
500368f7 | 871 | return splice_write(pipe, out, ppos, len, flags); |
5274f052 | 872 | } |
e14748e8 | 873 | EXPORT_SYMBOL_GPL(do_splice_from); |
5274f052 | 874 | |
83f9135b JA |
875 | /* |
876 | * Attempt to initiate a splice from a file to a pipe. | |
877 | */ | |
e14748e8 SF |
878 | long do_splice_to(struct file *in, loff_t *ppos, |
879 | struct pipe_inode_info *pipe, size_t len, | |
880 | unsigned int flags) | |
5274f052 | 881 | { |
6818173b MS |
882 | ssize_t (*splice_read)(struct file *, loff_t *, |
883 | struct pipe_inode_info *, size_t, unsigned int); | |
5274f052 JA |
884 | int ret; |
885 | ||
49570e9b | 886 | if (unlikely(!(in->f_mode & FMODE_READ))) |
5274f052 JA |
887 | return -EBADF; |
888 | ||
cbb7e577 | 889 | ret = rw_verify_area(READ, in, ppos, len); |
5274f052 JA |
890 | if (unlikely(ret < 0)) |
891 | return ret; | |
892 | ||
03cc0789 AV |
893 | if (unlikely(len > MAX_RW_COUNT)) |
894 | len = MAX_RW_COUNT; | |
895 | ||
72c2d531 | 896 | if (in->f_op->splice_read) |
cc56f7de CG |
897 | splice_read = in->f_op->splice_read; |
898 | else | |
6818173b MS |
899 | splice_read = default_file_splice_read; |
900 | ||
901 | return splice_read(in, ppos, pipe, len, flags); | |
5274f052 | 902 | } |
e14748e8 | 903 | EXPORT_SYMBOL_GPL(do_splice_to); |
5274f052 | 904 | |
932cc6d4 JA |
905 | /** |
906 | * splice_direct_to_actor - splices data directly between two non-pipes | |
907 | * @in: file to splice from | |
908 | * @sd: actor information on where to splice to | |
909 | * @actor: handles the data splicing | |
910 | * | |
911 | * Description: | |
912 | * This is a special case helper to splice directly between two | |
913 | * points, without requiring an explicit pipe. Internally an allocated | |
79685b8d | 914 | * pipe is cached in the process, and reused during the lifetime of |
932cc6d4 JA |
915 | * that process. |
916 | * | |
c66ab6fa JA |
917 | */ |
918 | ssize_t splice_direct_to_actor(struct file *in, struct splice_desc *sd, | |
919 | splice_direct_actor *actor) | |
b92ce558 JA |
920 | { |
921 | struct pipe_inode_info *pipe; | |
922 | long ret, bytes; | |
923 | umode_t i_mode; | |
c66ab6fa | 924 | size_t len; |
0ff28d9f | 925 | int i, flags, more; |
b92ce558 JA |
926 | |
927 | /* | |
928 | * We require the input being a regular file, as we don't want to | |
929 | * randomly drop data for eg socket -> socket splicing. Use the | |
930 | * piped splicing for that! | |
931 | */ | |
496ad9aa | 932 | i_mode = file_inode(in)->i_mode; |
b92ce558 JA |
933 | if (unlikely(!S_ISREG(i_mode) && !S_ISBLK(i_mode))) |
934 | return -EINVAL; | |
935 | ||
936 | /* | |
937 | * neither in nor out is a pipe, setup an internal pipe attached to | |
938 | * 'out' and transfer the wanted data from 'in' to 'out' through that | |
939 | */ | |
940 | pipe = current->splice_pipe; | |
49570e9b | 941 | if (unlikely(!pipe)) { |
7bee130e | 942 | pipe = alloc_pipe_info(); |
b92ce558 JA |
943 | if (!pipe) |
944 | return -ENOMEM; | |
945 | ||
946 | /* | |
947 | * We don't have an immediate reader, but we'll read the stuff | |
00522fb4 | 948 | * out of the pipe right after the splice_to_pipe(). So set |
b92ce558 JA |
949 | * PIPE_READERS appropriately. |
950 | */ | |
951 | pipe->readers = 1; | |
952 | ||
953 | current->splice_pipe = pipe; | |
954 | } | |
955 | ||
956 | /* | |
73d62d83 | 957 | * Do the splice. |
b92ce558 JA |
958 | */ |
959 | ret = 0; | |
960 | bytes = 0; | |
c66ab6fa JA |
961 | len = sd->total_len; |
962 | flags = sd->flags; | |
963 | ||
964 | /* | |
965 | * Don't block on output, we have to drain the direct pipe. | |
966 | */ | |
967 | sd->flags &= ~SPLICE_F_NONBLOCK; | |
0ff28d9f | 968 | more = sd->flags & SPLICE_F_MORE; |
b92ce558 JA |
969 | |
970 | while (len) { | |
51a92c0f | 971 | size_t read_len; |
a82c53a0 | 972 | loff_t pos = sd->pos, prev_pos = pos; |
b92ce558 | 973 | |
bcd4f3ac | 974 | ret = do_splice_to(in, &pos, pipe, len, flags); |
51a92c0f | 975 | if (unlikely(ret <= 0)) |
b92ce558 JA |
976 | goto out_release; |
977 | ||
978 | read_len = ret; | |
c66ab6fa | 979 | sd->total_len = read_len; |
b92ce558 | 980 | |
0ff28d9f CL |
981 | /* |
982 | * If more data is pending, set SPLICE_F_MORE | |
983 | * If this is the last data and SPLICE_F_MORE was not set | |
984 | * initially, clears it. | |
985 | */ | |
986 | if (read_len < len) | |
987 | sd->flags |= SPLICE_F_MORE; | |
988 | else if (!more) | |
989 | sd->flags &= ~SPLICE_F_MORE; | |
b92ce558 JA |
990 | /* |
991 | * NOTE: nonblocking mode only applies to the input. We | |
992 | * must not do the output in nonblocking mode as then we | |
993 | * could get stuck data in the internal pipe: | |
994 | */ | |
c66ab6fa | 995 | ret = actor(pipe, sd); |
a82c53a0 TZ |
996 | if (unlikely(ret <= 0)) { |
997 | sd->pos = prev_pos; | |
b92ce558 | 998 | goto out_release; |
a82c53a0 | 999 | } |
b92ce558 JA |
1000 | |
1001 | bytes += ret; | |
1002 | len -= ret; | |
bcd4f3ac | 1003 | sd->pos = pos; |
b92ce558 | 1004 | |
a82c53a0 TZ |
1005 | if (ret < read_len) { |
1006 | sd->pos = prev_pos + ret; | |
51a92c0f | 1007 | goto out_release; |
a82c53a0 | 1008 | } |
b92ce558 JA |
1009 | } |
1010 | ||
9e97198d | 1011 | done: |
b92ce558 | 1012 | pipe->nrbufs = pipe->curbuf = 0; |
80848708 | 1013 | file_accessed(in); |
b92ce558 JA |
1014 | return bytes; |
1015 | ||
1016 | out_release: | |
1017 | /* | |
1018 | * If we did an incomplete transfer we must release | |
1019 | * the pipe buffers in question: | |
1020 | */ | |
35f3d14d | 1021 | for (i = 0; i < pipe->buffers; i++) { |
b92ce558 JA |
1022 | struct pipe_buffer *buf = pipe->bufs + i; |
1023 | ||
a779638c MS |
1024 | if (buf->ops) |
1025 | pipe_buf_release(pipe, buf); | |
b92ce558 | 1026 | } |
b92ce558 | 1027 | |
9e97198d JA |
1028 | if (!bytes) |
1029 | bytes = ret; | |
c66ab6fa | 1030 | |
9e97198d | 1031 | goto done; |
c66ab6fa JA |
1032 | } |
1033 | EXPORT_SYMBOL(splice_direct_to_actor); | |
1034 | ||
1035 | static int direct_splice_actor(struct pipe_inode_info *pipe, | |
1036 | struct splice_desc *sd) | |
1037 | { | |
6a14b90b | 1038 | struct file *file = sd->u.file; |
c66ab6fa | 1039 | |
7995bd28 | 1040 | return do_splice_from(pipe, file, sd->opos, sd->total_len, |
2cb4b05e | 1041 | sd->flags); |
c66ab6fa JA |
1042 | } |
1043 | ||
932cc6d4 JA |
1044 | /** |
1045 | * do_splice_direct - splices data directly between two files | |
1046 | * @in: file to splice from | |
1047 | * @ppos: input file offset | |
1048 | * @out: file to splice to | |
acdb37c3 | 1049 | * @opos: output file offset |
932cc6d4 JA |
1050 | * @len: number of bytes to splice |
1051 | * @flags: splice modifier flags | |
1052 | * | |
1053 | * Description: | |
1054 | * For use by do_sendfile(). splice can easily emulate sendfile, but | |
1055 | * doing it in the application would incur an extra system call | |
1056 | * (splice in + splice out, as compared to just sendfile()). So this helper | |
1057 | * can splice directly through a process-private pipe. | |
1058 | * | |
1059 | */ | |
c66ab6fa | 1060 | long do_splice_direct(struct file *in, loff_t *ppos, struct file *out, |
7995bd28 | 1061 | loff_t *opos, size_t len, unsigned int flags) |
c66ab6fa JA |
1062 | { |
1063 | struct splice_desc sd = { | |
1064 | .len = len, | |
1065 | .total_len = len, | |
1066 | .flags = flags, | |
1067 | .pos = *ppos, | |
6a14b90b | 1068 | .u.file = out, |
7995bd28 | 1069 | .opos = opos, |
c66ab6fa | 1070 | }; |
51a92c0f | 1071 | long ret; |
c66ab6fa | 1072 | |
18c67cb9 AV |
1073 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1074 | return -EBADF; | |
1075 | ||
1076 | if (unlikely(out->f_flags & O_APPEND)) | |
1077 | return -EINVAL; | |
1078 | ||
1079 | ret = rw_verify_area(WRITE, out, opos, len); | |
1080 | if (unlikely(ret < 0)) | |
1081 | return ret; | |
1082 | ||
c66ab6fa | 1083 | ret = splice_direct_to_actor(in, &sd, direct_splice_actor); |
51a92c0f | 1084 | if (ret > 0) |
a82c53a0 | 1085 | *ppos = sd.pos; |
51a92c0f | 1086 | |
c66ab6fa | 1087 | return ret; |
b92ce558 | 1088 | } |
1c118596 | 1089 | EXPORT_SYMBOL(do_splice_direct); |
b92ce558 | 1090 | |
8924feff AV |
1091 | static int wait_for_space(struct pipe_inode_info *pipe, unsigned flags) |
1092 | { | |
52bce911 LT |
1093 | for (;;) { |
1094 | if (unlikely(!pipe->readers)) { | |
1095 | send_sig(SIGPIPE, current, 0); | |
1096 | return -EPIPE; | |
1097 | } | |
1098 | if (pipe->nrbufs != pipe->buffers) | |
1099 | return 0; | |
8924feff AV |
1100 | if (flags & SPLICE_F_NONBLOCK) |
1101 | return -EAGAIN; | |
1102 | if (signal_pending(current)) | |
1103 | return -ERESTARTSYS; | |
1104 | pipe->waiting_writers++; | |
1105 | pipe_wait(pipe); | |
1106 | pipe->waiting_writers--; | |
1107 | } | |
8924feff AV |
1108 | } |
1109 | ||
7c77f0b3 MS |
1110 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, |
1111 | struct pipe_inode_info *opipe, | |
1112 | size_t len, unsigned int flags); | |
ddac0d39 | 1113 | |
83f9135b JA |
1114 | /* |
1115 | * Determine where to splice to/from. | |
1116 | */ | |
529565dc IM |
1117 | static long do_splice(struct file *in, loff_t __user *off_in, |
1118 | struct file *out, loff_t __user *off_out, | |
1119 | size_t len, unsigned int flags) | |
5274f052 | 1120 | { |
7c77f0b3 MS |
1121 | struct pipe_inode_info *ipipe; |
1122 | struct pipe_inode_info *opipe; | |
7995bd28 | 1123 | loff_t offset; |
a4514ebd | 1124 | long ret; |
5274f052 | 1125 | |
71993e62 LT |
1126 | ipipe = get_pipe_info(in); |
1127 | opipe = get_pipe_info(out); | |
7c77f0b3 MS |
1128 | |
1129 | if (ipipe && opipe) { | |
1130 | if (off_in || off_out) | |
1131 | return -ESPIPE; | |
1132 | ||
1133 | if (!(in->f_mode & FMODE_READ)) | |
1134 | return -EBADF; | |
1135 | ||
1136 | if (!(out->f_mode & FMODE_WRITE)) | |
1137 | return -EBADF; | |
1138 | ||
1139 | /* Splicing to self would be fun, but... */ | |
1140 | if (ipipe == opipe) | |
1141 | return -EINVAL; | |
1142 | ||
1143 | return splice_pipe_to_pipe(ipipe, opipe, len, flags); | |
1144 | } | |
1145 | ||
1146 | if (ipipe) { | |
529565dc IM |
1147 | if (off_in) |
1148 | return -ESPIPE; | |
b92ce558 | 1149 | if (off_out) { |
19c9a49b | 1150 | if (!(out->f_mode & FMODE_PWRITE)) |
b92ce558 | 1151 | return -EINVAL; |
cbb7e577 | 1152 | if (copy_from_user(&offset, off_out, sizeof(loff_t))) |
b92ce558 | 1153 | return -EFAULT; |
7995bd28 AV |
1154 | } else { |
1155 | offset = out->f_pos; | |
1156 | } | |
529565dc | 1157 | |
18c67cb9 AV |
1158 | if (unlikely(!(out->f_mode & FMODE_WRITE))) |
1159 | return -EBADF; | |
1160 | ||
1161 | if (unlikely(out->f_flags & O_APPEND)) | |
1162 | return -EINVAL; | |
1163 | ||
1164 | ret = rw_verify_area(WRITE, out, &offset, len); | |
1165 | if (unlikely(ret < 0)) | |
1166 | return ret; | |
1167 | ||
500368f7 | 1168 | file_start_write(out); |
7995bd28 | 1169 | ret = do_splice_from(ipipe, out, &offset, len, flags); |
500368f7 | 1170 | file_end_write(out); |
a4514ebd | 1171 | |
7995bd28 AV |
1172 | if (!off_out) |
1173 | out->f_pos = offset; | |
1174 | else if (copy_to_user(off_out, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1175 | ret = -EFAULT; |
1176 | ||
1177 | return ret; | |
529565dc | 1178 | } |
5274f052 | 1179 | |
7c77f0b3 | 1180 | if (opipe) { |
529565dc IM |
1181 | if (off_out) |
1182 | return -ESPIPE; | |
b92ce558 | 1183 | if (off_in) { |
19c9a49b | 1184 | if (!(in->f_mode & FMODE_PREAD)) |
b92ce558 | 1185 | return -EINVAL; |
cbb7e577 | 1186 | if (copy_from_user(&offset, off_in, sizeof(loff_t))) |
b92ce558 | 1187 | return -EFAULT; |
7995bd28 AV |
1188 | } else { |
1189 | offset = in->f_pos; | |
1190 | } | |
529565dc | 1191 | |
8924feff AV |
1192 | pipe_lock(opipe); |
1193 | ret = wait_for_space(opipe, flags); | |
1194 | if (!ret) | |
1195 | ret = do_splice_to(in, &offset, opipe, len, flags); | |
1196 | pipe_unlock(opipe); | |
1197 | if (ret > 0) | |
1198 | wakeup_pipe_readers(opipe); | |
7995bd28 AV |
1199 | if (!off_in) |
1200 | in->f_pos = offset; | |
1201 | else if (copy_to_user(off_in, &offset, sizeof(loff_t))) | |
a4514ebd JA |
1202 | ret = -EFAULT; |
1203 | ||
1204 | return ret; | |
529565dc | 1205 | } |
5274f052 JA |
1206 | |
1207 | return -EINVAL; | |
1208 | } | |
1209 | ||
79fddc4e AV |
1210 | static int iter_to_pipe(struct iov_iter *from, |
1211 | struct pipe_inode_info *pipe, | |
1212 | unsigned flags) | |
912d35f8 | 1213 | { |
79fddc4e AV |
1214 | struct pipe_buffer buf = { |
1215 | .ops = &user_page_pipe_buf_ops, | |
1216 | .flags = flags | |
1217 | }; | |
1218 | size_t total = 0; | |
1219 | int ret = 0; | |
1220 | bool failed = false; | |
1221 | ||
1222 | while (iov_iter_count(from) && !failed) { | |
1223 | struct page *pages[16]; | |
db85a9eb AV |
1224 | ssize_t copied; |
1225 | size_t start; | |
79fddc4e | 1226 | int n; |
db85a9eb | 1227 | |
79fddc4e AV |
1228 | copied = iov_iter_get_pages(from, pages, ~0UL, 16, &start); |
1229 | if (copied <= 0) { | |
1230 | ret = copied; | |
1231 | break; | |
1232 | } | |
db85a9eb | 1233 | |
79fddc4e | 1234 | for (n = 0; copied; n++, start = 0) { |
db85a9eb | 1235 | int size = min_t(int, copied, PAGE_SIZE - start); |
79fddc4e AV |
1236 | if (!failed) { |
1237 | buf.page = pages[n]; | |
1238 | buf.offset = start; | |
1239 | buf.len = size; | |
1240 | ret = add_to_pipe(pipe, &buf); | |
1241 | if (unlikely(ret < 0)) { | |
1242 | failed = true; | |
1243 | } else { | |
1244 | iov_iter_advance(from, ret); | |
1245 | total += ret; | |
1246 | } | |
1247 | } else { | |
1248 | put_page(pages[n]); | |
1249 | } | |
db85a9eb | 1250 | copied -= size; |
912d35f8 | 1251 | } |
912d35f8 | 1252 | } |
79fddc4e | 1253 | return total ? total : ret; |
912d35f8 JA |
1254 | } |
1255 | ||
6a14b90b JA |
1256 | static int pipe_to_user(struct pipe_inode_info *pipe, struct pipe_buffer *buf, |
1257 | struct splice_desc *sd) | |
1258 | { | |
6130f531 AV |
1259 | int n = copy_page_to_iter(buf->page, buf->offset, sd->len, sd->u.data); |
1260 | return n == sd->len ? n : -EFAULT; | |
6a14b90b JA |
1261 | } |
1262 | ||
1263 | /* | |
1264 | * For lack of a better implementation, implement vmsplice() to userspace | |
1265 | * as a simple copy of the pipes pages to the user iov. | |
1266 | */ | |
6130f531 | 1267 | static long vmsplice_to_user(struct file *file, const struct iovec __user *uiov, |
6a14b90b JA |
1268 | unsigned long nr_segs, unsigned int flags) |
1269 | { | |
1270 | struct pipe_inode_info *pipe; | |
1271 | struct splice_desc sd; | |
6a14b90b | 1272 | long ret; |
6130f531 AV |
1273 | struct iovec iovstack[UIO_FASTIOV]; |
1274 | struct iovec *iov = iovstack; | |
1275 | struct iov_iter iter; | |
6a14b90b | 1276 | |
71993e62 | 1277 | pipe = get_pipe_info(file); |
6a14b90b JA |
1278 | if (!pipe) |
1279 | return -EBADF; | |
1280 | ||
345995fa AV |
1281 | ret = import_iovec(READ, uiov, nr_segs, |
1282 | ARRAY_SIZE(iovstack), &iov, &iter); | |
1283 | if (ret < 0) | |
1284 | return ret; | |
6a14b90b | 1285 | |
345995fa | 1286 | sd.total_len = iov_iter_count(&iter); |
6130f531 | 1287 | sd.len = 0; |
6130f531 AV |
1288 | sd.flags = flags; |
1289 | sd.u.data = &iter; | |
1290 | sd.pos = 0; | |
6a14b90b | 1291 | |
345995fa AV |
1292 | if (sd.total_len) { |
1293 | pipe_lock(pipe); | |
1294 | ret = __splice_from_pipe(pipe, &sd, pipe_to_user); | |
1295 | pipe_unlock(pipe); | |
1296 | } | |
6a14b90b | 1297 | |
345995fa | 1298 | kfree(iov); |
6a14b90b JA |
1299 | return ret; |
1300 | } | |
1301 | ||
912d35f8 JA |
1302 | /* |
1303 | * vmsplice splices a user address range into a pipe. It can be thought of | |
1304 | * as splice-from-memory, where the regular splice is splice-from-file (or | |
1305 | * to file). In both cases the output is a pipe, naturally. | |
912d35f8 | 1306 | */ |
db85a9eb | 1307 | static long vmsplice_to_pipe(struct file *file, const struct iovec __user *uiov, |
6a14b90b | 1308 | unsigned long nr_segs, unsigned int flags) |
912d35f8 | 1309 | { |
ddac0d39 | 1310 | struct pipe_inode_info *pipe; |
db85a9eb AV |
1311 | struct iovec iovstack[UIO_FASTIOV]; |
1312 | struct iovec *iov = iovstack; | |
1313 | struct iov_iter from; | |
35f3d14d | 1314 | long ret; |
79fddc4e AV |
1315 | unsigned buf_flag = 0; |
1316 | ||
1317 | if (flags & SPLICE_F_GIFT) | |
1318 | buf_flag = PIPE_BUF_FLAG_GIFT; | |
912d35f8 | 1319 | |
71993e62 | 1320 | pipe = get_pipe_info(file); |
ddac0d39 | 1321 | if (!pipe) |
912d35f8 | 1322 | return -EBADF; |
912d35f8 | 1323 | |
db85a9eb AV |
1324 | ret = import_iovec(WRITE, uiov, nr_segs, |
1325 | ARRAY_SIZE(iovstack), &iov, &from); | |
1326 | if (ret < 0) | |
1327 | return ret; | |
1328 | ||
8924feff AV |
1329 | pipe_lock(pipe); |
1330 | ret = wait_for_space(pipe, flags); | |
79fddc4e AV |
1331 | if (!ret) |
1332 | ret = iter_to_pipe(&from, pipe, buf_flag); | |
8924feff AV |
1333 | pipe_unlock(pipe); |
1334 | if (ret > 0) | |
1335 | wakeup_pipe_readers(pipe); | |
db85a9eb | 1336 | kfree(iov); |
35f3d14d | 1337 | return ret; |
912d35f8 JA |
1338 | } |
1339 | ||
6a14b90b JA |
1340 | /* |
1341 | * Note that vmsplice only really supports true splicing _from_ user memory | |
1342 | * to a pipe, not the other way around. Splicing from user memory is a simple | |
1343 | * operation that can be supported without any funky alignment restrictions | |
1344 | * or nasty vm tricks. We simply map in the user memory and fill them into | |
1345 | * a pipe. The reverse isn't quite as easy, though. There are two possible | |
1346 | * solutions for that: | |
1347 | * | |
1348 | * - memcpy() the data internally, at which point we might as well just | |
1349 | * do a regular read() on the buffer anyway. | |
1350 | * - Lots of nasty vm tricks, that are neither fast nor flexible (it | |
1351 | * has restriction limitations on both ends of the pipe). | |
1352 | * | |
1353 | * Currently we punt and implement it as a normal copy, see pipe_to_user(). | |
1354 | * | |
1355 | */ | |
836f92ad HC |
1356 | SYSCALL_DEFINE4(vmsplice, int, fd, const struct iovec __user *, iov, |
1357 | unsigned long, nr_segs, unsigned int, flags) | |
912d35f8 | 1358 | { |
2903ff01 | 1359 | struct fd f; |
912d35f8 | 1360 | long error; |
912d35f8 | 1361 | |
6a14b90b JA |
1362 | if (unlikely(nr_segs > UIO_MAXIOV)) |
1363 | return -EINVAL; | |
1364 | else if (unlikely(!nr_segs)) | |
1365 | return 0; | |
1366 | ||
912d35f8 | 1367 | error = -EBADF; |
2903ff01 AV |
1368 | f = fdget(fd); |
1369 | if (f.file) { | |
1370 | if (f.file->f_mode & FMODE_WRITE) | |
1371 | error = vmsplice_to_pipe(f.file, iov, nr_segs, flags); | |
1372 | else if (f.file->f_mode & FMODE_READ) | |
1373 | error = vmsplice_to_user(f.file, iov, nr_segs, flags); | |
1374 | ||
1375 | fdput(f); | |
912d35f8 JA |
1376 | } |
1377 | ||
1378 | return error; | |
1379 | } | |
1380 | ||
76b021d0 AV |
1381 | #ifdef CONFIG_COMPAT |
1382 | COMPAT_SYSCALL_DEFINE4(vmsplice, int, fd, const struct compat_iovec __user *, iov32, | |
1383 | unsigned int, nr_segs, unsigned int, flags) | |
1384 | { | |
1385 | unsigned i; | |
1386 | struct iovec __user *iov; | |
1387 | if (nr_segs > UIO_MAXIOV) | |
1388 | return -EINVAL; | |
1389 | iov = compat_alloc_user_space(nr_segs * sizeof(struct iovec)); | |
1390 | for (i = 0; i < nr_segs; i++) { | |
1391 | struct compat_iovec v; | |
1392 | if (get_user(v.iov_base, &iov32[i].iov_base) || | |
1393 | get_user(v.iov_len, &iov32[i].iov_len) || | |
1394 | put_user(compat_ptr(v.iov_base), &iov[i].iov_base) || | |
1395 | put_user(v.iov_len, &iov[i].iov_len)) | |
1396 | return -EFAULT; | |
1397 | } | |
1398 | return sys_vmsplice(fd, iov, nr_segs, flags); | |
1399 | } | |
1400 | #endif | |
1401 | ||
836f92ad HC |
1402 | SYSCALL_DEFINE6(splice, int, fd_in, loff_t __user *, off_in, |
1403 | int, fd_out, loff_t __user *, off_out, | |
1404 | size_t, len, unsigned int, flags) | |
5274f052 | 1405 | { |
2903ff01 | 1406 | struct fd in, out; |
5274f052 | 1407 | long error; |
5274f052 JA |
1408 | |
1409 | if (unlikely(!len)) | |
1410 | return 0; | |
1411 | ||
1412 | error = -EBADF; | |
2903ff01 AV |
1413 | in = fdget(fd_in); |
1414 | if (in.file) { | |
1415 | if (in.file->f_mode & FMODE_READ) { | |
1416 | out = fdget(fd_out); | |
1417 | if (out.file) { | |
1418 | if (out.file->f_mode & FMODE_WRITE) | |
1419 | error = do_splice(in.file, off_in, | |
1420 | out.file, off_out, | |
529565dc | 1421 | len, flags); |
2903ff01 | 1422 | fdput(out); |
5274f052 JA |
1423 | } |
1424 | } | |
2903ff01 | 1425 | fdput(in); |
5274f052 | 1426 | } |
5274f052 JA |
1427 | return error; |
1428 | } | |
70524490 | 1429 | |
aadd06e5 JA |
1430 | /* |
1431 | * Make sure there's data to read. Wait for input if we can, otherwise | |
1432 | * return an appropriate error. | |
1433 | */ | |
7c77f0b3 | 1434 | static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1435 | { |
1436 | int ret; | |
1437 | ||
1438 | /* | |
1439 | * Check ->nrbufs without the inode lock first. This function | |
1440 | * is speculative anyways, so missing one is ok. | |
1441 | */ | |
1442 | if (pipe->nrbufs) | |
1443 | return 0; | |
1444 | ||
1445 | ret = 0; | |
61e0d47c | 1446 | pipe_lock(pipe); |
aadd06e5 JA |
1447 | |
1448 | while (!pipe->nrbufs) { | |
1449 | if (signal_pending(current)) { | |
1450 | ret = -ERESTARTSYS; | |
1451 | break; | |
1452 | } | |
1453 | if (!pipe->writers) | |
1454 | break; | |
1455 | if (!pipe->waiting_writers) { | |
1456 | if (flags & SPLICE_F_NONBLOCK) { | |
1457 | ret = -EAGAIN; | |
1458 | break; | |
1459 | } | |
1460 | } | |
1461 | pipe_wait(pipe); | |
1462 | } | |
1463 | ||
61e0d47c | 1464 | pipe_unlock(pipe); |
aadd06e5 JA |
1465 | return ret; |
1466 | } | |
1467 | ||
1468 | /* | |
1469 | * Make sure there's writeable room. Wait for room if we can, otherwise | |
1470 | * return an appropriate error. | |
1471 | */ | |
7c77f0b3 | 1472 | static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags) |
aadd06e5 JA |
1473 | { |
1474 | int ret; | |
1475 | ||
1476 | /* | |
1477 | * Check ->nrbufs without the inode lock first. This function | |
1478 | * is speculative anyways, so missing one is ok. | |
1479 | */ | |
35f3d14d | 1480 | if (pipe->nrbufs < pipe->buffers) |
aadd06e5 JA |
1481 | return 0; |
1482 | ||
1483 | ret = 0; | |
61e0d47c | 1484 | pipe_lock(pipe); |
aadd06e5 | 1485 | |
35f3d14d | 1486 | while (pipe->nrbufs >= pipe->buffers) { |
aadd06e5 JA |
1487 | if (!pipe->readers) { |
1488 | send_sig(SIGPIPE, current, 0); | |
1489 | ret = -EPIPE; | |
1490 | break; | |
1491 | } | |
1492 | if (flags & SPLICE_F_NONBLOCK) { | |
1493 | ret = -EAGAIN; | |
1494 | break; | |
1495 | } | |
1496 | if (signal_pending(current)) { | |
1497 | ret = -ERESTARTSYS; | |
1498 | break; | |
1499 | } | |
1500 | pipe->waiting_writers++; | |
1501 | pipe_wait(pipe); | |
1502 | pipe->waiting_writers--; | |
1503 | } | |
1504 | ||
61e0d47c | 1505 | pipe_unlock(pipe); |
aadd06e5 JA |
1506 | return ret; |
1507 | } | |
1508 | ||
7c77f0b3 MS |
1509 | /* |
1510 | * Splice contents of ipipe to opipe. | |
1511 | */ | |
1512 | static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe, | |
1513 | struct pipe_inode_info *opipe, | |
1514 | size_t len, unsigned int flags) | |
1515 | { | |
1516 | struct pipe_buffer *ibuf, *obuf; | |
1517 | int ret = 0, nbuf; | |
1518 | bool input_wakeup = false; | |
1519 | ||
1520 | ||
1521 | retry: | |
1522 | ret = ipipe_prep(ipipe, flags); | |
1523 | if (ret) | |
1524 | return ret; | |
1525 | ||
1526 | ret = opipe_prep(opipe, flags); | |
1527 | if (ret) | |
1528 | return ret; | |
1529 | ||
1530 | /* | |
1531 | * Potential ABBA deadlock, work around it by ordering lock | |
1532 | * grabbing by pipe info address. Otherwise two different processes | |
1533 | * could deadlock (one doing tee from A -> B, the other from B -> A). | |
1534 | */ | |
1535 | pipe_double_lock(ipipe, opipe); | |
1536 | ||
1537 | do { | |
1538 | if (!opipe->readers) { | |
1539 | send_sig(SIGPIPE, current, 0); | |
1540 | if (!ret) | |
1541 | ret = -EPIPE; | |
1542 | break; | |
1543 | } | |
1544 | ||
1545 | if (!ipipe->nrbufs && !ipipe->writers) | |
1546 | break; | |
1547 | ||
1548 | /* | |
1549 | * Cannot make any progress, because either the input | |
1550 | * pipe is empty or the output pipe is full. | |
1551 | */ | |
35f3d14d | 1552 | if (!ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) { |
7c77f0b3 MS |
1553 | /* Already processed some buffers, break */ |
1554 | if (ret) | |
1555 | break; | |
1556 | ||
1557 | if (flags & SPLICE_F_NONBLOCK) { | |
1558 | ret = -EAGAIN; | |
1559 | break; | |
1560 | } | |
1561 | ||
1562 | /* | |
1563 | * We raced with another reader/writer and haven't | |
1564 | * managed to process any buffers. A zero return | |
1565 | * value means EOF, so retry instead. | |
1566 | */ | |
1567 | pipe_unlock(ipipe); | |
1568 | pipe_unlock(opipe); | |
1569 | goto retry; | |
1570 | } | |
1571 | ||
1572 | ibuf = ipipe->bufs + ipipe->curbuf; | |
35f3d14d | 1573 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); |
7c77f0b3 MS |
1574 | obuf = opipe->bufs + nbuf; |
1575 | ||
1576 | if (len >= ibuf->len) { | |
1577 | /* | |
1578 | * Simply move the whole buffer from ipipe to opipe | |
1579 | */ | |
1580 | *obuf = *ibuf; | |
1581 | ibuf->ops = NULL; | |
1582 | opipe->nrbufs++; | |
35f3d14d | 1583 | ipipe->curbuf = (ipipe->curbuf + 1) & (ipipe->buffers - 1); |
7c77f0b3 MS |
1584 | ipipe->nrbufs--; |
1585 | input_wakeup = true; | |
1586 | } else { | |
1587 | /* | |
1588 | * Get a reference to this pipe buffer, | |
1589 | * so we can copy the contents over. | |
1590 | */ | |
7bf2d1df | 1591 | pipe_buf_get(ipipe, ibuf); |
7c77f0b3 MS |
1592 | *obuf = *ibuf; |
1593 | ||
1594 | /* | |
1595 | * Don't inherit the gift flag, we need to | |
1596 | * prevent multiple steals of this page. | |
1597 | */ | |
1598 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; | |
1599 | ||
1600 | obuf->len = len; | |
1601 | opipe->nrbufs++; | |
1602 | ibuf->offset += obuf->len; | |
1603 | ibuf->len -= obuf->len; | |
1604 | } | |
1605 | ret += obuf->len; | |
1606 | len -= obuf->len; | |
1607 | } while (len); | |
1608 | ||
1609 | pipe_unlock(ipipe); | |
1610 | pipe_unlock(opipe); | |
1611 | ||
1612 | /* | |
1613 | * If we put data in the output pipe, wakeup any potential readers. | |
1614 | */ | |
825cdcb1 NK |
1615 | if (ret > 0) |
1616 | wakeup_pipe_readers(opipe); | |
1617 | ||
7c77f0b3 MS |
1618 | if (input_wakeup) |
1619 | wakeup_pipe_writers(ipipe); | |
1620 | ||
1621 | return ret; | |
1622 | } | |
1623 | ||
70524490 JA |
1624 | /* |
1625 | * Link contents of ipipe to opipe. | |
1626 | */ | |
1627 | static int link_pipe(struct pipe_inode_info *ipipe, | |
1628 | struct pipe_inode_info *opipe, | |
1629 | size_t len, unsigned int flags) | |
1630 | { | |
1631 | struct pipe_buffer *ibuf, *obuf; | |
aadd06e5 | 1632 | int ret = 0, i = 0, nbuf; |
70524490 JA |
1633 | |
1634 | /* | |
1635 | * Potential ABBA deadlock, work around it by ordering lock | |
61e0d47c | 1636 | * grabbing by pipe info address. Otherwise two different processes |
70524490 JA |
1637 | * could deadlock (one doing tee from A -> B, the other from B -> A). |
1638 | */ | |
61e0d47c | 1639 | pipe_double_lock(ipipe, opipe); |
70524490 | 1640 | |
aadd06e5 | 1641 | do { |
70524490 JA |
1642 | if (!opipe->readers) { |
1643 | send_sig(SIGPIPE, current, 0); | |
1644 | if (!ret) | |
1645 | ret = -EPIPE; | |
1646 | break; | |
1647 | } | |
70524490 | 1648 | |
aadd06e5 JA |
1649 | /* |
1650 | * If we have iterated all input buffers or ran out of | |
1651 | * output room, break. | |
1652 | */ | |
35f3d14d | 1653 | if (i >= ipipe->nrbufs || opipe->nrbufs >= opipe->buffers) |
aadd06e5 | 1654 | break; |
70524490 | 1655 | |
35f3d14d JA |
1656 | ibuf = ipipe->bufs + ((ipipe->curbuf + i) & (ipipe->buffers-1)); |
1657 | nbuf = (opipe->curbuf + opipe->nrbufs) & (opipe->buffers - 1); | |
70524490 JA |
1658 | |
1659 | /* | |
aadd06e5 JA |
1660 | * Get a reference to this pipe buffer, |
1661 | * so we can copy the contents over. | |
70524490 | 1662 | */ |
7bf2d1df | 1663 | pipe_buf_get(ipipe, ibuf); |
aadd06e5 JA |
1664 | |
1665 | obuf = opipe->bufs + nbuf; | |
1666 | *obuf = *ibuf; | |
1667 | ||
2a27250e | 1668 | /* |
aadd06e5 JA |
1669 | * Don't inherit the gift flag, we need to |
1670 | * prevent multiple steals of this page. | |
2a27250e | 1671 | */ |
aadd06e5 | 1672 | obuf->flags &= ~PIPE_BUF_FLAG_GIFT; |
70524490 | 1673 | |
aadd06e5 JA |
1674 | if (obuf->len > len) |
1675 | obuf->len = len; | |
70524490 | 1676 | |
aadd06e5 JA |
1677 | opipe->nrbufs++; |
1678 | ret += obuf->len; | |
1679 | len -= obuf->len; | |
1680 | i++; | |
1681 | } while (len); | |
70524490 | 1682 | |
02cf01ae JA |
1683 | /* |
1684 | * return EAGAIN if we have the potential of some data in the | |
1685 | * future, otherwise just return 0 | |
1686 | */ | |
1687 | if (!ret && ipipe->waiting_writers && (flags & SPLICE_F_NONBLOCK)) | |
1688 | ret = -EAGAIN; | |
1689 | ||
61e0d47c MS |
1690 | pipe_unlock(ipipe); |
1691 | pipe_unlock(opipe); | |
70524490 | 1692 | |
aadd06e5 JA |
1693 | /* |
1694 | * If we put data in the output pipe, wakeup any potential readers. | |
1695 | */ | |
825cdcb1 NK |
1696 | if (ret > 0) |
1697 | wakeup_pipe_readers(opipe); | |
70524490 JA |
1698 | |
1699 | return ret; | |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * This is a tee(1) implementation that works on pipes. It doesn't copy | |
1704 | * any data, it simply references the 'in' pages on the 'out' pipe. | |
1705 | * The 'flags' used are the SPLICE_F_* variants, currently the only | |
1706 | * applicable one is SPLICE_F_NONBLOCK. | |
1707 | */ | |
1708 | static long do_tee(struct file *in, struct file *out, size_t len, | |
1709 | unsigned int flags) | |
1710 | { | |
71993e62 LT |
1711 | struct pipe_inode_info *ipipe = get_pipe_info(in); |
1712 | struct pipe_inode_info *opipe = get_pipe_info(out); | |
aadd06e5 | 1713 | int ret = -EINVAL; |
70524490 JA |
1714 | |
1715 | /* | |
aadd06e5 JA |
1716 | * Duplicate the contents of ipipe to opipe without actually |
1717 | * copying the data. | |
70524490 | 1718 | */ |
aadd06e5 JA |
1719 | if (ipipe && opipe && ipipe != opipe) { |
1720 | /* | |
1721 | * Keep going, unless we encounter an error. The ipipe/opipe | |
1722 | * ordering doesn't really matter. | |
1723 | */ | |
7c77f0b3 | 1724 | ret = ipipe_prep(ipipe, flags); |
aadd06e5 | 1725 | if (!ret) { |
7c77f0b3 | 1726 | ret = opipe_prep(opipe, flags); |
02cf01ae | 1727 | if (!ret) |
aadd06e5 | 1728 | ret = link_pipe(ipipe, opipe, len, flags); |
aadd06e5 JA |
1729 | } |
1730 | } | |
70524490 | 1731 | |
aadd06e5 | 1732 | return ret; |
70524490 JA |
1733 | } |
1734 | ||
836f92ad | 1735 | SYSCALL_DEFINE4(tee, int, fdin, int, fdout, size_t, len, unsigned int, flags) |
70524490 | 1736 | { |
2903ff01 AV |
1737 | struct fd in; |
1738 | int error; | |
70524490 JA |
1739 | |
1740 | if (unlikely(!len)) | |
1741 | return 0; | |
1742 | ||
1743 | error = -EBADF; | |
2903ff01 AV |
1744 | in = fdget(fdin); |
1745 | if (in.file) { | |
1746 | if (in.file->f_mode & FMODE_READ) { | |
1747 | struct fd out = fdget(fdout); | |
1748 | if (out.file) { | |
1749 | if (out.file->f_mode & FMODE_WRITE) | |
1750 | error = do_tee(in.file, out.file, | |
1751 | len, flags); | |
1752 | fdput(out); | |
70524490 JA |
1753 | } |
1754 | } | |
2903ff01 | 1755 | fdput(in); |
70524490 JA |
1756 | } |
1757 | ||
1758 | return error; | |
1759 | } |