Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/pipe.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1999 Linus Torvalds | |
6 | */ | |
7 | ||
8 | #include <linux/mm.h> | |
9 | #include <linux/file.h> | |
10 | #include <linux/poll.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/fs.h> | |
35f3d14d | 15 | #include <linux/log2.h> |
1da177e4 | 16 | #include <linux/mount.h> |
4fa7ec5d | 17 | #include <linux/pseudo_fs.h> |
b502bd11 | 18 | #include <linux/magic.h> |
1da177e4 LT |
19 | #include <linux/pipe_fs_i.h> |
20 | #include <linux/uio.h> | |
21 | #include <linux/highmem.h> | |
5274f052 | 22 | #include <linux/pagemap.h> |
db349509 | 23 | #include <linux/audit.h> |
ba719bae | 24 | #include <linux/syscalls.h> |
b492e95b | 25 | #include <linux/fcntl.h> |
d86133bd | 26 | #include <linux/memcontrol.h> |
1da177e4 | 27 | |
7c0f6ba6 | 28 | #include <linux/uaccess.h> |
1da177e4 LT |
29 | #include <asm/ioctls.h> |
30 | ||
599a0ac1 AV |
31 | #include "internal.h" |
32 | ||
b492e95b JA |
33 | /* |
34 | * The max size that a non-root user is allowed to grow the pipe. Can | |
ff9da691 | 35 | * be set by root in /proc/sys/fs/pipe-max-size |
b492e95b | 36 | */ |
ff9da691 JA |
37 | unsigned int pipe_max_size = 1048576; |
38 | ||
759c0114 WT |
39 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
40 | * matches default values. | |
41 | */ | |
42 | unsigned long pipe_user_pages_hard; | |
43 | unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; | |
44 | ||
1da177e4 | 45 | /* |
8cefc107 DH |
46 | * We use head and tail indices that aren't masked off, except at the point of |
47 | * dereference, but rather they're allowed to wrap naturally. This means there | |
48 | * isn't a dead spot in the buffer, but the ring has to be a power of two and | |
49 | * <= 2^31. | |
50 | * -- David Howells 2019-09-23. | |
51 | * | |
1da177e4 LT |
52 | * Reads with count = 0 should always return 0. |
53 | * -- Julian Bradfield 1999-06-07. | |
54 | * | |
55 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
56 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
57 | * | |
58 | * pipe_read & write cleanup | |
59 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
60 | */ | |
61 | ||
61e0d47c MS |
62 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
63 | { | |
6447a3cf | 64 | if (pipe->files) |
72b0d9aa | 65 | mutex_lock_nested(&pipe->mutex, subclass); |
61e0d47c MS |
66 | } |
67 | ||
68 | void pipe_lock(struct pipe_inode_info *pipe) | |
69 | { | |
70 | /* | |
71 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
72 | */ | |
73 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
74 | } | |
75 | EXPORT_SYMBOL(pipe_lock); | |
76 | ||
77 | void pipe_unlock(struct pipe_inode_info *pipe) | |
78 | { | |
6447a3cf | 79 | if (pipe->files) |
72b0d9aa | 80 | mutex_unlock(&pipe->mutex); |
61e0d47c MS |
81 | } |
82 | EXPORT_SYMBOL(pipe_unlock); | |
83 | ||
ebec73f4 AV |
84 | static inline void __pipe_lock(struct pipe_inode_info *pipe) |
85 | { | |
86 | mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); | |
87 | } | |
88 | ||
89 | static inline void __pipe_unlock(struct pipe_inode_info *pipe) | |
90 | { | |
91 | mutex_unlock(&pipe->mutex); | |
92 | } | |
93 | ||
61e0d47c MS |
94 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
95 | struct pipe_inode_info *pipe2) | |
96 | { | |
97 | BUG_ON(pipe1 == pipe2); | |
98 | ||
99 | if (pipe1 < pipe2) { | |
100 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
101 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
102 | } else { | |
023d43c7 PZ |
103 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
104 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
105 | } |
106 | } | |
107 | ||
1da177e4 | 108 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
3a326a2c | 109 | void pipe_wait(struct pipe_inode_info *pipe) |
1da177e4 LT |
110 | { |
111 | DEFINE_WAIT(wait); | |
112 | ||
d79fc0fc IM |
113 | /* |
114 | * Pipes are system-local resources, so sleeping on them | |
115 | * is considered a noninteractive wait: | |
116 | */ | |
af927232 | 117 | prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE); |
61e0d47c | 118 | pipe_unlock(pipe); |
1da177e4 | 119 | schedule(); |
3a326a2c | 120 | finish_wait(&pipe->wait, &wait); |
61e0d47c | 121 | pipe_lock(pipe); |
1da177e4 LT |
122 | } |
123 | ||
341b446b IM |
124 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
125 | struct pipe_buffer *buf) | |
1da177e4 LT |
126 | { |
127 | struct page *page = buf->page; | |
128 | ||
5274f052 JA |
129 | /* |
130 | * If nobody else uses this page, and we don't already have a | |
131 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 132 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 133 | */ |
341b446b | 134 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 135 | pipe->tmp_page = page; |
341b446b | 136 | else |
09cbfeaf | 137 | put_page(page); |
1da177e4 LT |
138 | } |
139 | ||
d86133bd VD |
140 | static int anon_pipe_buf_steal(struct pipe_inode_info *pipe, |
141 | struct pipe_buffer *buf) | |
142 | { | |
143 | struct page *page = buf->page; | |
144 | ||
145 | if (page_count(page) == 1) { | |
60cd4bcd | 146 | memcg_kmem_uncharge(page, 0); |
d86133bd VD |
147 | __SetPageLocked(page); |
148 | return 0; | |
149 | } | |
150 | return 1; | |
151 | } | |
152 | ||
0845718d | 153 | /** |
b51d63c6 | 154 | * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer |
0845718d JA |
155 | * @pipe: the pipe that the buffer belongs to |
156 | * @buf: the buffer to attempt to steal | |
157 | * | |
158 | * Description: | |
b51d63c6 | 159 | * This function attempts to steal the &struct page attached to |
0845718d JA |
160 | * @buf. If successful, this function returns 0 and returns with |
161 | * the page locked. The caller may then reuse the page for whatever | |
b51d63c6 | 162 | * he wishes; the typical use is insertion into a different file |
0845718d JA |
163 | * page cache. |
164 | */ | |
330ab716 JA |
165 | int generic_pipe_buf_steal(struct pipe_inode_info *pipe, |
166 | struct pipe_buffer *buf) | |
5abc97aa | 167 | { |
46e678c9 JA |
168 | struct page *page = buf->page; |
169 | ||
0845718d JA |
170 | /* |
171 | * A reference of one is golden, that means that the owner of this | |
172 | * page is the only one holding a reference to it. lock the page | |
173 | * and return OK. | |
174 | */ | |
46e678c9 | 175 | if (page_count(page) == 1) { |
46e678c9 JA |
176 | lock_page(page); |
177 | return 0; | |
178 | } | |
179 | ||
180 | return 1; | |
5abc97aa | 181 | } |
51921cb7 | 182 | EXPORT_SYMBOL(generic_pipe_buf_steal); |
5abc97aa | 183 | |
0845718d | 184 | /** |
b51d63c6 | 185 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
186 | * @pipe: the pipe that the buffer belongs to |
187 | * @buf: the buffer to get a reference to | |
188 | * | |
189 | * Description: | |
190 | * This function grabs an extra reference to @buf. It's used in | |
191 | * in the tee() system call, when we duplicate the buffers in one | |
192 | * pipe into another. | |
193 | */ | |
15fab63e | 194 | bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
70524490 | 195 | { |
15fab63e | 196 | return try_get_page(buf->page); |
70524490 | 197 | } |
51921cb7 | 198 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 199 | |
0845718d JA |
200 | /** |
201 | * generic_pipe_buf_confirm - verify contents of the pipe buffer | |
79685b8d | 202 | * @info: the pipe that the buffer belongs to |
0845718d JA |
203 | * @buf: the buffer to confirm |
204 | * | |
205 | * Description: | |
206 | * This function does nothing, because the generic pipe code uses | |
207 | * pages that are always good when inserted into the pipe. | |
208 | */ | |
cac36bb0 JA |
209 | int generic_pipe_buf_confirm(struct pipe_inode_info *info, |
210 | struct pipe_buffer *buf) | |
f84d7519 JA |
211 | { |
212 | return 0; | |
213 | } | |
51921cb7 | 214 | EXPORT_SYMBOL(generic_pipe_buf_confirm); |
f84d7519 | 215 | |
6818173b MS |
216 | /** |
217 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
218 | * @pipe: the pipe that the buffer belongs to | |
219 | * @buf: the buffer to put a reference to | |
220 | * | |
221 | * Description: | |
222 | * This function releases a reference to @buf. | |
223 | */ | |
224 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
225 | struct pipe_buffer *buf) | |
226 | { | |
09cbfeaf | 227 | put_page(buf->page); |
6818173b | 228 | } |
51921cb7 | 229 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 230 | |
01e7187b | 231 | /* New data written to a pipe may be appended to a buffer with this type. */ |
d4c3cca9 | 232 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
cac36bb0 | 233 | .confirm = generic_pipe_buf_confirm, |
1da177e4 | 234 | .release = anon_pipe_buf_release, |
d86133bd | 235 | .steal = anon_pipe_buf_steal, |
f84d7519 | 236 | .get = generic_pipe_buf_get, |
1da177e4 LT |
237 | }; |
238 | ||
a0ce2f0a | 239 | static const struct pipe_buf_operations anon_pipe_buf_nomerge_ops = { |
cac36bb0 | 240 | .confirm = generic_pipe_buf_confirm, |
1da177e4 | 241 | .release = anon_pipe_buf_release, |
d86133bd | 242 | .steal = anon_pipe_buf_steal, |
f84d7519 | 243 | .get = generic_pipe_buf_get, |
1da177e4 LT |
244 | }; |
245 | ||
9883035a | 246 | static const struct pipe_buf_operations packet_pipe_buf_ops = { |
9883035a LT |
247 | .confirm = generic_pipe_buf_confirm, |
248 | .release = anon_pipe_buf_release, | |
d86133bd | 249 | .steal = anon_pipe_buf_steal, |
9883035a LT |
250 | .get = generic_pipe_buf_get, |
251 | }; | |
252 | ||
01e7187b JH |
253 | /** |
254 | * pipe_buf_mark_unmergeable - mark a &struct pipe_buffer as unmergeable | |
255 | * @buf: the buffer to mark | |
256 | * | |
257 | * Description: | |
258 | * This function ensures that no future writes will be merged into the | |
259 | * given &struct pipe_buffer. This is necessary when multiple pipe buffers | |
260 | * share the same backing page. | |
261 | */ | |
a0ce2f0a JH |
262 | void pipe_buf_mark_unmergeable(struct pipe_buffer *buf) |
263 | { | |
264 | if (buf->ops == &anon_pipe_buf_ops) | |
265 | buf->ops = &anon_pipe_buf_nomerge_ops; | |
266 | } | |
267 | ||
01e7187b JH |
268 | static bool pipe_buf_can_merge(struct pipe_buffer *buf) |
269 | { | |
270 | return buf->ops == &anon_pipe_buf_ops; | |
271 | } | |
272 | ||
85190d15 LT |
273 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
274 | static inline bool pipe_readable(const struct pipe_inode_info *pipe) | |
275 | { | |
276 | unsigned int head = READ_ONCE(pipe->head); | |
277 | unsigned int tail = READ_ONCE(pipe->tail); | |
278 | unsigned int writers = READ_ONCE(pipe->writers); | |
279 | ||
280 | return !pipe_empty(head, tail) || !writers; | |
281 | } | |
282 | ||
1da177e4 | 283 | static ssize_t |
fb9096a3 | 284 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 285 | { |
fb9096a3 | 286 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 287 | struct file *filp = iocb->ki_filp; |
de32ec4c | 288 | struct pipe_inode_info *pipe = filp->private_data; |
f467a6a6 | 289 | bool was_full; |
1da177e4 | 290 | ssize_t ret; |
1da177e4 | 291 | |
1da177e4 LT |
292 | /* Null read succeeds. */ |
293 | if (unlikely(total_len == 0)) | |
294 | return 0; | |
295 | ||
1da177e4 | 296 | ret = 0; |
ebec73f4 | 297 | __pipe_lock(pipe); |
f467a6a6 LT |
298 | |
299 | /* | |
300 | * We only wake up writers if the pipe was full when we started | |
301 | * reading in order to avoid unnecessary wakeups. | |
302 | * | |
303 | * But when we do wake up writers, we do so using a sync wakeup | |
304 | * (WF_SYNC), because we want them to get going and generate more | |
305 | * data for us. | |
306 | */ | |
307 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); | |
1da177e4 | 308 | for (;;) { |
8cefc107 DH |
309 | unsigned int head = pipe->head; |
310 | unsigned int tail = pipe->tail; | |
311 | unsigned int mask = pipe->ring_size - 1; | |
312 | ||
313 | if (!pipe_empty(head, tail)) { | |
314 | struct pipe_buffer *buf = &pipe->bufs[tail & mask]; | |
1da177e4 | 315 | size_t chars = buf->len; |
637b58c2 AV |
316 | size_t written; |
317 | int error; | |
1da177e4 LT |
318 | |
319 | if (chars > total_len) | |
320 | chars = total_len; | |
321 | ||
fba597db | 322 | error = pipe_buf_confirm(pipe, buf); |
f84d7519 | 323 | if (error) { |
5274f052 | 324 | if (!ret) |
e5953cbd | 325 | ret = error; |
5274f052 JA |
326 | break; |
327 | } | |
f84d7519 | 328 | |
fb9096a3 | 329 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 330 | if (unlikely(written < chars)) { |
341b446b | 331 | if (!ret) |
637b58c2 | 332 | ret = -EFAULT; |
1da177e4 LT |
333 | break; |
334 | } | |
335 | ret += chars; | |
336 | buf->offset += chars; | |
337 | buf->len -= chars; | |
9883035a LT |
338 | |
339 | /* Was it a packet buffer? Clean up and exit */ | |
340 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
341 | total_len = chars; | |
342 | buf->len = 0; | |
343 | } | |
344 | ||
1da177e4 | 345 | if (!buf->len) { |
a779638c | 346 | pipe_buf_release(pipe, buf); |
b667b867 | 347 | spin_lock_irq(&pipe->wait.lock); |
8cefc107 DH |
348 | tail++; |
349 | pipe->tail = tail; | |
b667b867 | 350 | spin_unlock_irq(&pipe->wait.lock); |
1da177e4 LT |
351 | } |
352 | total_len -= chars; | |
353 | if (!total_len) | |
354 | break; /* common path: read succeeded */ | |
8cefc107 DH |
355 | if (!pipe_empty(head, tail)) /* More to do? */ |
356 | continue; | |
1da177e4 | 357 | } |
8cefc107 | 358 | |
923f4f23 | 359 | if (!pipe->writers) |
1da177e4 | 360 | break; |
a28c8b9d LT |
361 | if (ret) |
362 | break; | |
363 | if (filp->f_flags & O_NONBLOCK) { | |
364 | ret = -EAGAIN; | |
365 | break; | |
1da177e4 LT |
366 | } |
367 | if (signal_pending(current)) { | |
341b446b IM |
368 | if (!ret) |
369 | ret = -ERESTARTSYS; | |
1da177e4 LT |
370 | break; |
371 | } | |
85190d15 | 372 | __pipe_unlock(pipe); |
f467a6a6 LT |
373 | if (was_full) { |
374 | wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM); | |
375 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
376 | } | |
85190d15 LT |
377 | wait_event_interruptible(pipe->wait, pipe_readable(pipe)); |
378 | __pipe_lock(pipe); | |
f467a6a6 | 379 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); |
1da177e4 | 380 | } |
ebec73f4 | 381 | __pipe_unlock(pipe); |
341b446b | 382 | |
f467a6a6 LT |
383 | if (was_full) { |
384 | wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM); | |
923f4f23 | 385 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 LT |
386 | } |
387 | if (ret > 0) | |
388 | file_accessed(filp); | |
389 | return ret; | |
390 | } | |
391 | ||
9883035a LT |
392 | static inline int is_packetized(struct file *file) |
393 | { | |
394 | return (file->f_flags & O_DIRECT) != 0; | |
395 | } | |
396 | ||
85190d15 LT |
397 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
398 | static inline bool pipe_writable(const struct pipe_inode_info *pipe) | |
399 | { | |
400 | unsigned int head = READ_ONCE(pipe->head); | |
401 | unsigned int tail = READ_ONCE(pipe->tail); | |
402 | unsigned int max_usage = READ_ONCE(pipe->max_usage); | |
403 | ||
404 | return !pipe_full(head, tail, max_usage) || | |
405 | !READ_ONCE(pipe->readers); | |
406 | } | |
407 | ||
1da177e4 | 408 | static ssize_t |
f0d1bec9 | 409 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 410 | { |
ee0b3e67 | 411 | struct file *filp = iocb->ki_filp; |
de32ec4c | 412 | struct pipe_inode_info *pipe = filp->private_data; |
8f868d68 | 413 | unsigned int head; |
f0d1bec9 | 414 | ssize_t ret = 0; |
f0d1bec9 | 415 | size_t total_len = iov_iter_count(from); |
1da177e4 | 416 | ssize_t chars; |
1b6b26ae | 417 | bool was_empty = false; |
1da177e4 | 418 | |
1da177e4 LT |
419 | /* Null write succeeds. */ |
420 | if (unlikely(total_len == 0)) | |
421 | return 0; | |
422 | ||
ebec73f4 | 423 | __pipe_lock(pipe); |
1da177e4 | 424 | |
923f4f23 | 425 | if (!pipe->readers) { |
1da177e4 LT |
426 | send_sig(SIGPIPE, current, 0); |
427 | ret = -EPIPE; | |
428 | goto out; | |
429 | } | |
430 | ||
1b6b26ae LT |
431 | /* |
432 | * Only wake up if the pipe started out empty, since | |
433 | * otherwise there should be no readers waiting. | |
434 | * | |
435 | * If it wasn't empty we try to merge new data into | |
436 | * the last buffer. | |
437 | * | |
438 | * That naturally merges small writes, but it also | |
439 | * page-aligs the rest of the writes for large writes | |
440 | * spanning multiple pages. | |
441 | */ | |
8cefc107 | 442 | head = pipe->head; |
1b6b26ae LT |
443 | was_empty = pipe_empty(head, pipe->tail); |
444 | chars = total_len & (PAGE_SIZE-1); | |
445 | if (chars && !was_empty) { | |
8f868d68 | 446 | unsigned int mask = pipe->ring_size - 1; |
8cefc107 | 447 | struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; |
1da177e4 | 448 | int offset = buf->offset + buf->len; |
341b446b | 449 | |
01e7187b | 450 | if (pipe_buf_can_merge(buf) && offset + chars <= PAGE_SIZE) { |
fba597db | 451 | ret = pipe_buf_confirm(pipe, buf); |
6ae08069 | 452 | if (ret) |
5274f052 | 453 | goto out; |
f84d7519 | 454 | |
f0d1bec9 AV |
455 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
456 | if (unlikely(ret < chars)) { | |
6ae08069 | 457 | ret = -EFAULT; |
1da177e4 | 458 | goto out; |
f6762b7a | 459 | } |
1b6b26ae | 460 | |
6ae08069 | 461 | buf->len += ret; |
f0d1bec9 | 462 | if (!iov_iter_count(from)) |
1da177e4 LT |
463 | goto out; |
464 | } | |
465 | } | |
466 | ||
467 | for (;;) { | |
923f4f23 | 468 | if (!pipe->readers) { |
1da177e4 | 469 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
470 | if (!ret) |
471 | ret = -EPIPE; | |
1da177e4 LT |
472 | break; |
473 | } | |
8cefc107 | 474 | |
a194dfe6 | 475 | head = pipe->head; |
8f868d68 DH |
476 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) { |
477 | unsigned int mask = pipe->ring_size - 1; | |
8cefc107 | 478 | struct pipe_buffer *buf = &pipe->bufs[head & mask]; |
923f4f23 | 479 | struct page *page = pipe->tmp_page; |
f0d1bec9 | 480 | int copied; |
1da177e4 LT |
481 | |
482 | if (!page) { | |
d86133bd | 483 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
484 | if (unlikely(!page)) { |
485 | ret = ret ? : -ENOMEM; | |
486 | break; | |
487 | } | |
923f4f23 | 488 | pipe->tmp_page = page; |
1da177e4 | 489 | } |
a194dfe6 DH |
490 | |
491 | /* Allocate a slot in the ring in advance and attach an | |
492 | * empty buffer. If we fault or otherwise fail to use | |
493 | * it, either the reader will consume it or it'll still | |
494 | * be there for the next write. | |
495 | */ | |
496 | spin_lock_irq(&pipe->wait.lock); | |
497 | ||
498 | head = pipe->head; | |
8f868d68 | 499 | if (pipe_full(head, pipe->tail, pipe->max_usage)) { |
8df44129 DH |
500 | spin_unlock_irq(&pipe->wait.lock); |
501 | continue; | |
502 | } | |
503 | ||
a194dfe6 | 504 | pipe->head = head + 1; |
a194dfe6 | 505 | spin_unlock_irq(&pipe->wait.lock); |
1da177e4 LT |
506 | |
507 | /* Insert it into the buffer array */ | |
a194dfe6 | 508 | buf = &pipe->bufs[head & mask]; |
1da177e4 LT |
509 | buf->page = page; |
510 | buf->ops = &anon_pipe_buf_ops; | |
511 | buf->offset = 0; | |
a194dfe6 | 512 | buf->len = 0; |
9883035a LT |
513 | buf->flags = 0; |
514 | if (is_packetized(filp)) { | |
515 | buf->ops = &packet_pipe_buf_ops; | |
516 | buf->flags = PIPE_BUF_FLAG_PACKET; | |
517 | } | |
923f4f23 | 518 | pipe->tmp_page = NULL; |
1da177e4 | 519 | |
a194dfe6 DH |
520 | copied = copy_page_from_iter(page, 0, PAGE_SIZE, from); |
521 | if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) { | |
522 | if (!ret) | |
523 | ret = -EFAULT; | |
524 | break; | |
525 | } | |
526 | ret += copied; | |
527 | buf->offset = 0; | |
528 | buf->len = copied; | |
529 | ||
f0d1bec9 | 530 | if (!iov_iter_count(from)) |
1da177e4 LT |
531 | break; |
532 | } | |
8cefc107 | 533 | |
8f868d68 | 534 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) |
1da177e4 | 535 | continue; |
8cefc107 DH |
536 | |
537 | /* Wait for buffer space to become available. */ | |
1da177e4 | 538 | if (filp->f_flags & O_NONBLOCK) { |
341b446b IM |
539 | if (!ret) |
540 | ret = -EAGAIN; | |
1da177e4 LT |
541 | break; |
542 | } | |
543 | if (signal_pending(current)) { | |
341b446b IM |
544 | if (!ret) |
545 | ret = -ERESTARTSYS; | |
1da177e4 LT |
546 | break; |
547 | } | |
1b6b26ae LT |
548 | |
549 | /* | |
550 | * We're going to release the pipe lock and wait for more | |
551 | * space. We wake up any readers if necessary, and then | |
552 | * after waiting we need to re-check whether the pipe | |
553 | * become empty while we dropped the lock. | |
554 | */ | |
85190d15 | 555 | __pipe_unlock(pipe); |
1b6b26ae LT |
556 | if (was_empty) { |
557 | wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM); | |
558 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
559 | } | |
85190d15 LT |
560 | wait_event_interruptible(pipe->wait, pipe_writable(pipe)); |
561 | __pipe_lock(pipe); | |
1b6b26ae | 562 | was_empty = pipe_empty(head, pipe->tail); |
1da177e4 LT |
563 | } |
564 | out: | |
ebec73f4 | 565 | __pipe_unlock(pipe); |
1b6b26ae LT |
566 | |
567 | /* | |
568 | * If we do do a wakeup event, we do a 'sync' wakeup, because we | |
569 | * want the reader to start processing things asap, rather than | |
570 | * leave the data pending. | |
571 | * | |
572 | * This is particularly important for small writes, because of | |
573 | * how (for example) the GNU make jobserver uses small writes to | |
574 | * wake up pending jobs | |
575 | */ | |
576 | if (was_empty) { | |
577 | wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM); | |
923f4f23 | 578 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 | 579 | } |
7e775f46 | 580 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
581 | int err = file_update_time(filp); |
582 | if (err) | |
583 | ret = err; | |
7e775f46 | 584 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 585 | } |
1da177e4 LT |
586 | return ret; |
587 | } | |
588 | ||
d59d0b1b | 589 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 590 | { |
de32ec4c | 591 | struct pipe_inode_info *pipe = filp->private_data; |
8cefc107 | 592 | int count, head, tail, mask; |
1da177e4 LT |
593 | |
594 | switch (cmd) { | |
595 | case FIONREAD: | |
ebec73f4 | 596 | __pipe_lock(pipe); |
1da177e4 | 597 | count = 0; |
8cefc107 DH |
598 | head = pipe->head; |
599 | tail = pipe->tail; | |
600 | mask = pipe->ring_size - 1; | |
601 | ||
602 | while (tail != head) { | |
603 | count += pipe->bufs[tail & mask].len; | |
604 | tail++; | |
1da177e4 | 605 | } |
ebec73f4 | 606 | __pipe_unlock(pipe); |
923f4f23 | 607 | |
1da177e4 LT |
608 | return put_user(count, (int __user *)arg); |
609 | default: | |
46ce341b | 610 | return -ENOIOCTLCMD; |
1da177e4 LT |
611 | } |
612 | } | |
613 | ||
dd67081b | 614 | /* No kernel lock held - fine */ |
a11e1d43 LT |
615 | static __poll_t |
616 | pipe_poll(struct file *filp, poll_table *wait) | |
dd67081b | 617 | { |
a11e1d43 | 618 | __poll_t mask; |
dd67081b | 619 | struct pipe_inode_info *pipe = filp->private_data; |
ad910e36 | 620 | unsigned int head, tail; |
a11e1d43 | 621 | |
ad910e36 LT |
622 | /* |
623 | * Reading only -- no need for acquiring the semaphore. | |
624 | * | |
625 | * But because this is racy, the code has to add the | |
626 | * entry to the poll table _first_ .. | |
627 | */ | |
a11e1d43 | 628 | poll_wait(filp, &pipe->wait, wait); |
1da177e4 | 629 | |
ad910e36 LT |
630 | /* |
631 | * .. and only then can you do the racy tests. That way, | |
632 | * if something changes and you got it wrong, the poll | |
633 | * table entry will wake you up and fix it. | |
634 | */ | |
635 | head = READ_ONCE(pipe->head); | |
636 | tail = READ_ONCE(pipe->tail); | |
637 | ||
a11e1d43 | 638 | mask = 0; |
1da177e4 | 639 | if (filp->f_mode & FMODE_READ) { |
8cefc107 DH |
640 | if (!pipe_empty(head, tail)) |
641 | mask |= EPOLLIN | EPOLLRDNORM; | |
923f4f23 | 642 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
a9a08845 | 643 | mask |= EPOLLHUP; |
1da177e4 LT |
644 | } |
645 | ||
646 | if (filp->f_mode & FMODE_WRITE) { | |
6718b6f8 | 647 | if (!pipe_full(head, tail, pipe->max_usage)) |
8cefc107 | 648 | mask |= EPOLLOUT | EPOLLWRNORM; |
5e5d7a22 | 649 | /* |
a9a08845 | 650 | * Most Unices do not set EPOLLERR for FIFOs but on Linux they |
5e5d7a22 PE |
651 | * behave exactly like pipes for poll(). |
652 | */ | |
923f4f23 | 653 | if (!pipe->readers) |
a9a08845 | 654 | mask |= EPOLLERR; |
1da177e4 LT |
655 | } |
656 | ||
657 | return mask; | |
658 | } | |
659 | ||
b0d8d229 LT |
660 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
661 | { | |
662 | int kill = 0; | |
663 | ||
664 | spin_lock(&inode->i_lock); | |
665 | if (!--pipe->files) { | |
666 | inode->i_pipe = NULL; | |
667 | kill = 1; | |
668 | } | |
669 | spin_unlock(&inode->i_lock); | |
670 | ||
671 | if (kill) | |
672 | free_pipe_info(pipe); | |
673 | } | |
674 | ||
1da177e4 | 675 | static int |
599a0ac1 | 676 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 677 | { |
b0d8d229 | 678 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 679 | |
ebec73f4 | 680 | __pipe_lock(pipe); |
599a0ac1 AV |
681 | if (file->f_mode & FMODE_READ) |
682 | pipe->readers--; | |
683 | if (file->f_mode & FMODE_WRITE) | |
684 | pipe->writers--; | |
341b446b | 685 | |
ba5bb147 | 686 | if (pipe->readers || pipe->writers) { |
a9a08845 | 687 | wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP); |
923f4f23 IM |
688 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
689 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 690 | } |
ebec73f4 | 691 | __pipe_unlock(pipe); |
ba5bb147 | 692 | |
b0d8d229 | 693 | put_pipe_info(inode, pipe); |
1da177e4 LT |
694 | return 0; |
695 | } | |
696 | ||
697 | static int | |
599a0ac1 | 698 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 699 | { |
de32ec4c | 700 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 701 | int retval = 0; |
1da177e4 | 702 | |
ebec73f4 | 703 | __pipe_lock(pipe); |
599a0ac1 AV |
704 | if (filp->f_mode & FMODE_READ) |
705 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
706 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 707 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
708 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
709 | /* this can happen only if on == T */ | |
e5bc49ba ON |
710 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
711 | } | |
ebec73f4 | 712 | __pipe_unlock(pipe); |
60aa4924 | 713 | return retval; |
1da177e4 LT |
714 | } |
715 | ||
9c87bcf0 | 716 | static unsigned long account_pipe_buffers(struct user_struct *user, |
759c0114 WT |
717 | unsigned long old, unsigned long new) |
718 | { | |
9c87bcf0 | 719 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
720 | } |
721 | ||
9c87bcf0 | 722 | static bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 723 | { |
f7340761 EB |
724 | unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft); |
725 | ||
726 | return soft_limit && user_bufs > soft_limit; | |
759c0114 WT |
727 | } |
728 | ||
9c87bcf0 | 729 | static bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 730 | { |
f7340761 EB |
731 | unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard); |
732 | ||
733 | return hard_limit && user_bufs > hard_limit; | |
759c0114 WT |
734 | } |
735 | ||
85c2dd54 EB |
736 | static bool is_unprivileged_user(void) |
737 | { | |
738 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
739 | } | |
740 | ||
7bee130e | 741 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 742 | { |
923f4f23 | 743 | struct pipe_inode_info *pipe; |
09b4d199 MK |
744 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
745 | struct user_struct *user = get_current_user(); | |
9c87bcf0 | 746 | unsigned long user_bufs; |
f7340761 | 747 | unsigned int max_size = READ_ONCE(pipe_max_size); |
3a326a2c | 748 | |
d86133bd | 749 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
09b4d199 MK |
750 | if (pipe == NULL) |
751 | goto out_free_uid; | |
752 | ||
f7340761 EB |
753 | if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE)) |
754 | pipe_bufs = max_size >> PAGE_SHIFT; | |
086e774a | 755 | |
9c87bcf0 | 756 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
a005ca0e | 757 | |
85c2dd54 | 758 | if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) { |
9c87bcf0 | 759 | user_bufs = account_pipe_buffers(user, pipe_bufs, 1); |
a005ca0e | 760 | pipe_bufs = 1; |
09b4d199 | 761 | } |
759c0114 | 762 | |
85c2dd54 | 763 | if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user()) |
a005ca0e MK |
764 | goto out_revert_acct; |
765 | ||
766 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), | |
767 | GFP_KERNEL_ACCOUNT); | |
768 | ||
09b4d199 MK |
769 | if (pipe->bufs) { |
770 | init_waitqueue_head(&pipe->wait); | |
771 | pipe->r_counter = pipe->w_counter = 1; | |
6718b6f8 | 772 | pipe->max_usage = pipe_bufs; |
8cefc107 | 773 | pipe->ring_size = pipe_bufs; |
09b4d199 | 774 | pipe->user = user; |
09b4d199 MK |
775 | mutex_init(&pipe->mutex); |
776 | return pipe; | |
3a326a2c IM |
777 | } |
778 | ||
a005ca0e | 779 | out_revert_acct: |
9c87bcf0 | 780 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
09b4d199 MK |
781 | kfree(pipe); |
782 | out_free_uid: | |
783 | free_uid(user); | |
35f3d14d | 784 | return NULL; |
3a326a2c IM |
785 | } |
786 | ||
4b8a8f1e | 787 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
788 | { |
789 | int i; | |
1da177e4 | 790 | |
8cefc107 | 791 | (void) account_pipe_buffers(pipe->user, pipe->ring_size, 0); |
759c0114 | 792 | free_uid(pipe->user); |
8cefc107 | 793 | for (i = 0; i < pipe->ring_size; i++) { |
923f4f23 | 794 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 795 | if (buf->ops) |
a779638c | 796 | pipe_buf_release(pipe, buf); |
1da177e4 | 797 | } |
923f4f23 IM |
798 | if (pipe->tmp_page) |
799 | __free_page(pipe->tmp_page); | |
35f3d14d | 800 | kfree(pipe->bufs); |
923f4f23 | 801 | kfree(pipe); |
1da177e4 LT |
802 | } |
803 | ||
fa3536cc | 804 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 805 | |
c23fbb6b ED |
806 | /* |
807 | * pipefs_dname() is called from d_path(). | |
808 | */ | |
809 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
810 | { | |
811 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
75c3cfa8 | 812 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
813 | } |
814 | ||
3ba13d17 | 815 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 816 | .d_dname = pipefs_dname, |
1da177e4 LT |
817 | }; |
818 | ||
819 | static struct inode * get_pipe_inode(void) | |
820 | { | |
a209dfc7 | 821 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 822 | struct pipe_inode_info *pipe; |
1da177e4 LT |
823 | |
824 | if (!inode) | |
825 | goto fail_inode; | |
826 | ||
85fe4025 CH |
827 | inode->i_ino = get_next_ino(); |
828 | ||
7bee130e | 829 | pipe = alloc_pipe_info(); |
923f4f23 | 830 | if (!pipe) |
1da177e4 | 831 | goto fail_iput; |
3a326a2c | 832 | |
ba5bb147 AV |
833 | inode->i_pipe = pipe; |
834 | pipe->files = 2; | |
923f4f23 | 835 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 836 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
837 | |
838 | /* | |
839 | * Mark the inode dirty from the very beginning, | |
840 | * that way it will never be moved to the dirty | |
841 | * list because "mark_inode_dirty()" will think | |
842 | * that it already _is_ on the dirty list. | |
843 | */ | |
844 | inode->i_state = I_DIRTY; | |
845 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
846 | inode->i_uid = current_fsuid(); |
847 | inode->i_gid = current_fsgid(); | |
078cd827 | 848 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
923f4f23 | 849 | |
1da177e4 LT |
850 | return inode; |
851 | ||
852 | fail_iput: | |
853 | iput(inode); | |
341b446b | 854 | |
1da177e4 LT |
855 | fail_inode: |
856 | return NULL; | |
857 | } | |
858 | ||
e4fad8e5 | 859 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 860 | { |
e4fad8e5 | 861 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 862 | struct file *f; |
1da177e4 | 863 | |
1da177e4 | 864 | if (!inode) |
e4fad8e5 | 865 | return -ENFILE; |
1da177e4 | 866 | |
152b6372 AV |
867 | f = alloc_file_pseudo(inode, pipe_mnt, "", |
868 | O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)), | |
869 | &pipefifo_fops); | |
e9bb1f9b | 870 | if (IS_ERR(f)) { |
152b6372 AV |
871 | free_pipe_info(inode->i_pipe); |
872 | iput(inode); | |
873 | return PTR_ERR(f); | |
e9bb1f9b | 874 | } |
341b446b | 875 | |
de32ec4c | 876 | f->private_data = inode->i_pipe; |
d6cbd281 | 877 | |
183266f2 AV |
878 | res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK), |
879 | &pipefifo_fops); | |
e9bb1f9b | 880 | if (IS_ERR(res[0])) { |
b10a4a9f AV |
881 | put_pipe_info(inode, inode->i_pipe); |
882 | fput(f); | |
883 | return PTR_ERR(res[0]); | |
e9bb1f9b | 884 | } |
de32ec4c | 885 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 | 886 | res[1] = f; |
d8e464ec LT |
887 | stream_open(inode, res[0]); |
888 | stream_open(inode, res[1]); | |
e4fad8e5 | 889 | return 0; |
d6cbd281 AK |
890 | } |
891 | ||
5b249b1b | 892 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 893 | { |
d6cbd281 AK |
894 | int error; |
895 | int fdw, fdr; | |
896 | ||
9883035a | 897 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT)) |
ed8cae8b UD |
898 | return -EINVAL; |
899 | ||
e4fad8e5 AV |
900 | error = create_pipe_files(files, flags); |
901 | if (error) | |
902 | return error; | |
d6cbd281 | 903 | |
ed8cae8b | 904 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
905 | if (error < 0) |
906 | goto err_read_pipe; | |
907 | fdr = error; | |
908 | ||
ed8cae8b | 909 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
910 | if (error < 0) |
911 | goto err_fdr; | |
912 | fdw = error; | |
913 | ||
157cf649 | 914 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
915 | fd[0] = fdr; |
916 | fd[1] = fdw; | |
d6cbd281 AK |
917 | return 0; |
918 | ||
919 | err_fdr: | |
920 | put_unused_fd(fdr); | |
921 | err_read_pipe: | |
e4fad8e5 AV |
922 | fput(files[0]); |
923 | fput(files[1]); | |
d6cbd281 | 924 | return error; |
1da177e4 LT |
925 | } |
926 | ||
5b249b1b AV |
927 | int do_pipe_flags(int *fd, int flags) |
928 | { | |
929 | struct file *files[2]; | |
930 | int error = __do_pipe_flags(fd, files, flags); | |
931 | if (!error) { | |
932 | fd_install(fd[0], files[0]); | |
933 | fd_install(fd[1], files[1]); | |
934 | } | |
935 | return error; | |
936 | } | |
937 | ||
d35c7b0e UD |
938 | /* |
939 | * sys_pipe() is the normal C calling standard for creating | |
940 | * a pipe. It's not the way Unix traditionally does this, though. | |
941 | */ | |
0a216dd1 | 942 | static int do_pipe2(int __user *fildes, int flags) |
d35c7b0e | 943 | { |
5b249b1b | 944 | struct file *files[2]; |
d35c7b0e UD |
945 | int fd[2]; |
946 | int error; | |
947 | ||
5b249b1b | 948 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 949 | if (!error) { |
5b249b1b AV |
950 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
951 | fput(files[0]); | |
952 | fput(files[1]); | |
953 | put_unused_fd(fd[0]); | |
954 | put_unused_fd(fd[1]); | |
d35c7b0e | 955 | error = -EFAULT; |
5b249b1b AV |
956 | } else { |
957 | fd_install(fd[0], files[0]); | |
958 | fd_install(fd[1], files[1]); | |
ba719bae | 959 | } |
d35c7b0e UD |
960 | } |
961 | return error; | |
962 | } | |
963 | ||
0a216dd1 DB |
964 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
965 | { | |
966 | return do_pipe2(fildes, flags); | |
967 | } | |
968 | ||
2b664219 | 969 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b | 970 | { |
0a216dd1 | 971 | return do_pipe2(fildes, 0); |
ed8cae8b UD |
972 | } |
973 | ||
fc7478a2 | 974 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 | 975 | { |
8cefc107 | 976 | int cur = *cnt; |
f776c738 AV |
977 | |
978 | while (cur == *cnt) { | |
fc7478a2 | 979 | pipe_wait(pipe); |
f776c738 AV |
980 | if (signal_pending(current)) |
981 | break; | |
982 | } | |
983 | return cur == *cnt ? -ERESTARTSYS : 0; | |
984 | } | |
985 | ||
fc7478a2 | 986 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 987 | { |
fc7478a2 | 988 | wake_up_interruptible(&pipe->wait); |
f776c738 AV |
989 | } |
990 | ||
991 | static int fifo_open(struct inode *inode, struct file *filp) | |
992 | { | |
993 | struct pipe_inode_info *pipe; | |
599a0ac1 | 994 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
995 | int ret; |
996 | ||
ba5bb147 AV |
997 | filp->f_version = 0; |
998 | ||
999 | spin_lock(&inode->i_lock); | |
1000 | if (inode->i_pipe) { | |
1001 | pipe = inode->i_pipe; | |
1002 | pipe->files++; | |
1003 | spin_unlock(&inode->i_lock); | |
1004 | } else { | |
1005 | spin_unlock(&inode->i_lock); | |
7bee130e | 1006 | pipe = alloc_pipe_info(); |
f776c738 | 1007 | if (!pipe) |
ba5bb147 AV |
1008 | return -ENOMEM; |
1009 | pipe->files = 1; | |
1010 | spin_lock(&inode->i_lock); | |
1011 | if (unlikely(inode->i_pipe)) { | |
1012 | inode->i_pipe->files++; | |
1013 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 1014 | free_pipe_info(pipe); |
ba5bb147 AV |
1015 | pipe = inode->i_pipe; |
1016 | } else { | |
1017 | inode->i_pipe = pipe; | |
1018 | spin_unlock(&inode->i_lock); | |
1019 | } | |
f776c738 | 1020 | } |
de32ec4c | 1021 | filp->private_data = pipe; |
ba5bb147 AV |
1022 | /* OK, we have a pipe and it's pinned down */ |
1023 | ||
ebec73f4 | 1024 | __pipe_lock(pipe); |
f776c738 AV |
1025 | |
1026 | /* We can only do regular read/write on fifos */ | |
d8e464ec | 1027 | stream_open(inode, filp); |
f776c738 | 1028 | |
d8e464ec | 1029 | switch (filp->f_mode & (FMODE_READ | FMODE_WRITE)) { |
f776c738 AV |
1030 | case FMODE_READ: |
1031 | /* | |
1032 | * O_RDONLY | |
1033 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
1034 | * opened, even when there is no process writing the FIFO. | |
1035 | */ | |
f776c738 AV |
1036 | pipe->r_counter++; |
1037 | if (pipe->readers++ == 0) | |
fc7478a2 | 1038 | wake_up_partner(pipe); |
f776c738 | 1039 | |
599a0ac1 | 1040 | if (!is_pipe && !pipe->writers) { |
f776c738 | 1041 | if ((filp->f_flags & O_NONBLOCK)) { |
a9a08845 | 1042 | /* suppress EPOLLHUP until we have |
f776c738 AV |
1043 | * seen a writer */ |
1044 | filp->f_version = pipe->w_counter; | |
1045 | } else { | |
fc7478a2 | 1046 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
1047 | goto err_rd; |
1048 | } | |
1049 | } | |
1050 | break; | |
8cefc107 | 1051 | |
f776c738 AV |
1052 | case FMODE_WRITE: |
1053 | /* | |
1054 | * O_WRONLY | |
1055 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
1056 | * errno=ENXIO when there is no process reading the FIFO. | |
1057 | */ | |
1058 | ret = -ENXIO; | |
599a0ac1 | 1059 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
1060 | goto err; |
1061 | ||
f776c738 AV |
1062 | pipe->w_counter++; |
1063 | if (!pipe->writers++) | |
fc7478a2 | 1064 | wake_up_partner(pipe); |
f776c738 | 1065 | |
599a0ac1 | 1066 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 1067 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
1068 | goto err_wr; |
1069 | } | |
1070 | break; | |
8cefc107 | 1071 | |
f776c738 AV |
1072 | case FMODE_READ | FMODE_WRITE: |
1073 | /* | |
1074 | * O_RDWR | |
1075 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
1076 | * This implementation will NEVER block on a O_RDWR open, since | |
1077 | * the process can at least talk to itself. | |
1078 | */ | |
f776c738 AV |
1079 | |
1080 | pipe->readers++; | |
1081 | pipe->writers++; | |
1082 | pipe->r_counter++; | |
1083 | pipe->w_counter++; | |
1084 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 1085 | wake_up_partner(pipe); |
f776c738 AV |
1086 | break; |
1087 | ||
1088 | default: | |
1089 | ret = -EINVAL; | |
1090 | goto err; | |
1091 | } | |
1092 | ||
1093 | /* Ok! */ | |
ebec73f4 | 1094 | __pipe_unlock(pipe); |
f776c738 AV |
1095 | return 0; |
1096 | ||
1097 | err_rd: | |
1098 | if (!--pipe->readers) | |
1099 | wake_up_interruptible(&pipe->wait); | |
1100 | ret = -ERESTARTSYS; | |
1101 | goto err; | |
1102 | ||
1103 | err_wr: | |
1104 | if (!--pipe->writers) | |
1105 | wake_up_interruptible(&pipe->wait); | |
1106 | ret = -ERESTARTSYS; | |
1107 | goto err; | |
1108 | ||
1109 | err: | |
ebec73f4 | 1110 | __pipe_unlock(pipe); |
b0d8d229 LT |
1111 | |
1112 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1113 | return ret; |
1114 | } | |
1115 | ||
599a0ac1 AV |
1116 | const struct file_operations pipefifo_fops = { |
1117 | .open = fifo_open, | |
1118 | .llseek = no_llseek, | |
fb9096a3 | 1119 | .read_iter = pipe_read, |
f0d1bec9 | 1120 | .write_iter = pipe_write, |
a11e1d43 | 1121 | .poll = pipe_poll, |
599a0ac1 AV |
1122 | .unlocked_ioctl = pipe_ioctl, |
1123 | .release = pipe_release, | |
1124 | .fasync = pipe_fasync, | |
f776c738 AV |
1125 | }; |
1126 | ||
f491bd71 MK |
1127 | /* |
1128 | * Currently we rely on the pipe array holding a power-of-2 number | |
d3f14c48 | 1129 | * of pages. Returns 0 on error. |
f491bd71 | 1130 | */ |
96e99be4 | 1131 | unsigned int round_pipe_size(unsigned long size) |
f491bd71 | 1132 | { |
c4fed5a9 | 1133 | if (size > (1U << 31)) |
96e99be4 EB |
1134 | return 0; |
1135 | ||
4c2e4bef EB |
1136 | /* Minimum pipe size, as required by POSIX */ |
1137 | if (size < PAGE_SIZE) | |
c4fed5a9 | 1138 | return PAGE_SIZE; |
d3f14c48 | 1139 | |
c4fed5a9 | 1140 | return roundup_pow_of_two(size); |
f491bd71 MK |
1141 | } |
1142 | ||
35f3d14d JA |
1143 | /* |
1144 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1145 | * pipe size if successful, or return -ERROR on error. | |
1146 | */ | |
d37d4166 | 1147 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) |
35f3d14d JA |
1148 | { |
1149 | struct pipe_buffer *bufs; | |
8cefc107 | 1150 | unsigned int size, nr_slots, head, tail, mask, n; |
9c87bcf0 | 1151 | unsigned long user_bufs; |
b0b91d18 | 1152 | long ret = 0; |
d37d4166 MK |
1153 | |
1154 | size = round_pipe_size(arg); | |
8cefc107 | 1155 | nr_slots = size >> PAGE_SHIFT; |
d37d4166 | 1156 | |
8cefc107 | 1157 | if (!nr_slots) |
d37d4166 MK |
1158 | return -EINVAL; |
1159 | ||
b0b91d18 MK |
1160 | /* |
1161 | * If trying to increase the pipe capacity, check that an | |
1162 | * unprivileged user is not trying to exceed various limits | |
1163 | * (soft limit check here, hard limit check just below). | |
1164 | * Decreasing the pipe capacity is always permitted, even | |
1165 | * if the user is currently over a limit. | |
1166 | */ | |
8cefc107 | 1167 | if (nr_slots > pipe->ring_size && |
b0b91d18 | 1168 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) |
d37d4166 MK |
1169 | return -EPERM; |
1170 | ||
8cefc107 | 1171 | user_bufs = account_pipe_buffers(pipe->user, pipe->ring_size, nr_slots); |
b0b91d18 | 1172 | |
8cefc107 | 1173 | if (nr_slots > pipe->ring_size && |
9c87bcf0 MK |
1174 | (too_many_pipe_buffers_hard(user_bufs) || |
1175 | too_many_pipe_buffers_soft(user_bufs)) && | |
85c2dd54 | 1176 | is_unprivileged_user()) { |
b0b91d18 MK |
1177 | ret = -EPERM; |
1178 | goto out_revert_acct; | |
1179 | } | |
35f3d14d | 1180 | |
35f3d14d | 1181 | /* |
8cefc107 DH |
1182 | * We can shrink the pipe, if arg is greater than the ring occupancy. |
1183 | * Since we don't expect a lot of shrink+grow operations, just free and | |
1184 | * allocate again like we would do for growing. If the pipe currently | |
35f3d14d JA |
1185 | * contains more buffers than arg, then return busy. |
1186 | */ | |
8cefc107 DH |
1187 | mask = pipe->ring_size - 1; |
1188 | head = pipe->head; | |
1189 | tail = pipe->tail; | |
1190 | n = pipe_occupancy(pipe->head, pipe->tail); | |
1191 | if (nr_slots < n) { | |
b0b91d18 MK |
1192 | ret = -EBUSY; |
1193 | goto out_revert_acct; | |
1194 | } | |
35f3d14d | 1195 | |
8cefc107 | 1196 | bufs = kcalloc(nr_slots, sizeof(*bufs), |
d86133bd | 1197 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); |
b0b91d18 MK |
1198 | if (unlikely(!bufs)) { |
1199 | ret = -ENOMEM; | |
1200 | goto out_revert_acct; | |
1201 | } | |
35f3d14d JA |
1202 | |
1203 | /* | |
1204 | * The pipe array wraps around, so just start the new one at zero | |
8cefc107 | 1205 | * and adjust the indices. |
35f3d14d | 1206 | */ |
8cefc107 DH |
1207 | if (n > 0) { |
1208 | unsigned int h = head & mask; | |
1209 | unsigned int t = tail & mask; | |
1210 | if (h > t) { | |
1211 | memcpy(bufs, pipe->bufs + t, | |
1212 | n * sizeof(struct pipe_buffer)); | |
1213 | } else { | |
1214 | unsigned int tsize = pipe->ring_size - t; | |
1215 | if (h > 0) | |
1216 | memcpy(bufs + tsize, pipe->bufs, | |
1217 | h * sizeof(struct pipe_buffer)); | |
1218 | memcpy(bufs, pipe->bufs + t, | |
1219 | tsize * sizeof(struct pipe_buffer)); | |
1220 | } | |
35f3d14d JA |
1221 | } |
1222 | ||
8cefc107 DH |
1223 | head = n; |
1224 | tail = 0; | |
1225 | ||
35f3d14d JA |
1226 | kfree(pipe->bufs); |
1227 | pipe->bufs = bufs; | |
8cefc107 | 1228 | pipe->ring_size = nr_slots; |
6718b6f8 | 1229 | pipe->max_usage = nr_slots; |
8cefc107 DH |
1230 | pipe->tail = tail; |
1231 | pipe->head = head; | |
8c7b8c34 | 1232 | wake_up_interruptible_all(&pipe->wait); |
6718b6f8 | 1233 | return pipe->max_usage * PAGE_SIZE; |
b0b91d18 MK |
1234 | |
1235 | out_revert_acct: | |
8cefc107 | 1236 | (void) account_pipe_buffers(pipe->user, nr_slots, pipe->ring_size); |
b0b91d18 | 1237 | return ret; |
35f3d14d JA |
1238 | } |
1239 | ||
72083646 LT |
1240 | /* |
1241 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1242 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1243 | * pipe. | |
1244 | */ | |
1245 | struct pipe_inode_info *get_pipe_info(struct file *file) | |
1246 | { | |
de32ec4c | 1247 | return file->f_op == &pipefifo_fops ? file->private_data : NULL; |
72083646 LT |
1248 | } |
1249 | ||
35f3d14d JA |
1250 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
1251 | { | |
1252 | struct pipe_inode_info *pipe; | |
1253 | long ret; | |
1254 | ||
c66fb347 | 1255 | pipe = get_pipe_info(file); |
35f3d14d JA |
1256 | if (!pipe) |
1257 | return -EBADF; | |
1258 | ||
ebec73f4 | 1259 | __pipe_lock(pipe); |
35f3d14d JA |
1260 | |
1261 | switch (cmd) { | |
d37d4166 MK |
1262 | case F_SETPIPE_SZ: |
1263 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1264 | break; |
1265 | case F_GETPIPE_SZ: | |
6718b6f8 | 1266 | ret = pipe->max_usage * PAGE_SIZE; |
35f3d14d JA |
1267 | break; |
1268 | default: | |
1269 | ret = -EINVAL; | |
1270 | break; | |
1271 | } | |
1272 | ||
ebec73f4 | 1273 | __pipe_unlock(pipe); |
35f3d14d JA |
1274 | return ret; |
1275 | } | |
1276 | ||
ff0c7d15 NP |
1277 | static const struct super_operations pipefs_ops = { |
1278 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1279 | .statfs = simple_statfs, |
ff0c7d15 NP |
1280 | }; |
1281 | ||
1da177e4 LT |
1282 | /* |
1283 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1284 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1285 | * any operations on the root directory. However, we need a non-trivial | |
1286 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1287 | */ | |
4fa7ec5d DH |
1288 | |
1289 | static int pipefs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1290 | { |
4fa7ec5d DH |
1291 | struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC); |
1292 | if (!ctx) | |
1293 | return -ENOMEM; | |
1294 | ctx->ops = &pipefs_ops; | |
1295 | ctx->dops = &pipefs_dentry_operations; | |
1296 | return 0; | |
1da177e4 LT |
1297 | } |
1298 | ||
1299 | static struct file_system_type pipe_fs_type = { | |
1300 | .name = "pipefs", | |
4fa7ec5d | 1301 | .init_fs_context = pipefs_init_fs_context, |
1da177e4 LT |
1302 | .kill_sb = kill_anon_super, |
1303 | }; | |
1304 | ||
1305 | static int __init init_pipe_fs(void) | |
1306 | { | |
1307 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1308 | |
1da177e4 LT |
1309 | if (!err) { |
1310 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1311 | if (IS_ERR(pipe_mnt)) { | |
1312 | err = PTR_ERR(pipe_mnt); | |
1313 | unregister_filesystem(&pipe_fs_type); | |
1314 | } | |
1315 | } | |
1316 | return err; | |
1317 | } | |
1318 | ||
1da177e4 | 1319 | fs_initcall(init_pipe_fs); |