]>
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> |
c73be61c | 27 | #include <linux/watch_queue.h> |
1da177e4 | 28 | |
7c0f6ba6 | 29 | #include <linux/uaccess.h> |
1da177e4 LT |
30 | #include <asm/ioctls.h> |
31 | ||
599a0ac1 AV |
32 | #include "internal.h" |
33 | ||
b492e95b JA |
34 | /* |
35 | * The max size that a non-root user is allowed to grow the pipe. Can | |
ff9da691 | 36 | * be set by root in /proc/sys/fs/pipe-max-size |
b492e95b | 37 | */ |
ff9da691 JA |
38 | unsigned int pipe_max_size = 1048576; |
39 | ||
759c0114 WT |
40 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
41 | * matches default values. | |
42 | */ | |
43 | unsigned long pipe_user_pages_hard; | |
44 | unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; | |
45 | ||
1da177e4 | 46 | /* |
8cefc107 DH |
47 | * We use head and tail indices that aren't masked off, except at the point of |
48 | * dereference, but rather they're allowed to wrap naturally. This means there | |
49 | * isn't a dead spot in the buffer, but the ring has to be a power of two and | |
50 | * <= 2^31. | |
51 | * -- David Howells 2019-09-23. | |
52 | * | |
1da177e4 LT |
53 | * Reads with count = 0 should always return 0. |
54 | * -- Julian Bradfield 1999-06-07. | |
55 | * | |
56 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
57 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
58 | * | |
59 | * pipe_read & write cleanup | |
60 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
61 | */ | |
62 | ||
61e0d47c MS |
63 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
64 | { | |
6447a3cf | 65 | if (pipe->files) |
72b0d9aa | 66 | mutex_lock_nested(&pipe->mutex, subclass); |
61e0d47c MS |
67 | } |
68 | ||
69 | void pipe_lock(struct pipe_inode_info *pipe) | |
70 | { | |
71 | /* | |
72 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
73 | */ | |
74 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
75 | } | |
76 | EXPORT_SYMBOL(pipe_lock); | |
77 | ||
78 | void pipe_unlock(struct pipe_inode_info *pipe) | |
79 | { | |
6447a3cf | 80 | if (pipe->files) |
72b0d9aa | 81 | mutex_unlock(&pipe->mutex); |
61e0d47c MS |
82 | } |
83 | EXPORT_SYMBOL(pipe_unlock); | |
84 | ||
ebec73f4 AV |
85 | static inline void __pipe_lock(struct pipe_inode_info *pipe) |
86 | { | |
87 | mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); | |
88 | } | |
89 | ||
90 | static inline void __pipe_unlock(struct pipe_inode_info *pipe) | |
91 | { | |
92 | mutex_unlock(&pipe->mutex); | |
93 | } | |
94 | ||
61e0d47c MS |
95 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
96 | struct pipe_inode_info *pipe2) | |
97 | { | |
98 | BUG_ON(pipe1 == pipe2); | |
99 | ||
100 | if (pipe1 < pipe2) { | |
101 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
102 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
103 | } else { | |
023d43c7 PZ |
104 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
105 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
106 | } |
107 | } | |
108 | ||
1da177e4 | 109 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
3a326a2c | 110 | void pipe_wait(struct pipe_inode_info *pipe) |
1da177e4 | 111 | { |
0ddad21d LT |
112 | DEFINE_WAIT(rdwait); |
113 | DEFINE_WAIT(wrwait); | |
1da177e4 | 114 | |
d79fc0fc IM |
115 | /* |
116 | * Pipes are system-local resources, so sleeping on them | |
117 | * is considered a noninteractive wait: | |
118 | */ | |
0ddad21d LT |
119 | prepare_to_wait(&pipe->rd_wait, &rdwait, TASK_INTERRUPTIBLE); |
120 | prepare_to_wait(&pipe->wr_wait, &wrwait, TASK_INTERRUPTIBLE); | |
61e0d47c | 121 | pipe_unlock(pipe); |
1da177e4 | 122 | schedule(); |
0ddad21d LT |
123 | finish_wait(&pipe->rd_wait, &rdwait); |
124 | finish_wait(&pipe->wr_wait, &wrwait); | |
61e0d47c | 125 | pipe_lock(pipe); |
1da177e4 LT |
126 | } |
127 | ||
341b446b IM |
128 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
129 | struct pipe_buffer *buf) | |
1da177e4 LT |
130 | { |
131 | struct page *page = buf->page; | |
132 | ||
5274f052 JA |
133 | /* |
134 | * If nobody else uses this page, and we don't already have a | |
135 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 136 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 137 | */ |
341b446b | 138 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 139 | pipe->tmp_page = page; |
341b446b | 140 | else |
09cbfeaf | 141 | put_page(page); |
1da177e4 LT |
142 | } |
143 | ||
c928f642 CH |
144 | static bool anon_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
145 | struct pipe_buffer *buf) | |
d86133bd VD |
146 | { |
147 | struct page *page = buf->page; | |
148 | ||
c928f642 CH |
149 | if (page_count(page) != 1) |
150 | return false; | |
151 | memcg_kmem_uncharge_page(page, 0); | |
152 | __SetPageLocked(page); | |
153 | return true; | |
d86133bd VD |
154 | } |
155 | ||
0845718d | 156 | /** |
c928f642 | 157 | * generic_pipe_buf_try_steal - attempt to take ownership of a &pipe_buffer |
0845718d JA |
158 | * @pipe: the pipe that the buffer belongs to |
159 | * @buf: the buffer to attempt to steal | |
160 | * | |
161 | * Description: | |
b51d63c6 | 162 | * This function attempts to steal the &struct page attached to |
0845718d JA |
163 | * @buf. If successful, this function returns 0 and returns with |
164 | * the page locked. The caller may then reuse the page for whatever | |
b51d63c6 | 165 | * he wishes; the typical use is insertion into a different file |
0845718d JA |
166 | * page cache. |
167 | */ | |
c928f642 CH |
168 | bool generic_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
169 | struct pipe_buffer *buf) | |
5abc97aa | 170 | { |
46e678c9 JA |
171 | struct page *page = buf->page; |
172 | ||
0845718d JA |
173 | /* |
174 | * A reference of one is golden, that means that the owner of this | |
175 | * page is the only one holding a reference to it. lock the page | |
176 | * and return OK. | |
177 | */ | |
46e678c9 | 178 | if (page_count(page) == 1) { |
46e678c9 | 179 | lock_page(page); |
c928f642 | 180 | return true; |
46e678c9 | 181 | } |
c928f642 | 182 | return false; |
5abc97aa | 183 | } |
c928f642 | 184 | EXPORT_SYMBOL(generic_pipe_buf_try_steal); |
5abc97aa | 185 | |
0845718d | 186 | /** |
b51d63c6 | 187 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
188 | * @pipe: the pipe that the buffer belongs to |
189 | * @buf: the buffer to get a reference to | |
190 | * | |
191 | * Description: | |
192 | * This function grabs an extra reference to @buf. It's used in | |
193 | * in the tee() system call, when we duplicate the buffers in one | |
194 | * pipe into another. | |
195 | */ | |
15fab63e | 196 | bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
70524490 | 197 | { |
15fab63e | 198 | return try_get_page(buf->page); |
70524490 | 199 | } |
51921cb7 | 200 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 201 | |
6818173b MS |
202 | /** |
203 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
204 | * @pipe: the pipe that the buffer belongs to | |
205 | * @buf: the buffer to put a reference to | |
206 | * | |
207 | * Description: | |
208 | * This function releases a reference to @buf. | |
209 | */ | |
210 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
211 | struct pipe_buffer *buf) | |
212 | { | |
09cbfeaf | 213 | put_page(buf->page); |
6818173b | 214 | } |
51921cb7 | 215 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 216 | |
d4c3cca9 | 217 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
c928f642 CH |
218 | .release = anon_pipe_buf_release, |
219 | .try_steal = anon_pipe_buf_try_steal, | |
220 | .get = generic_pipe_buf_get, | |
1da177e4 LT |
221 | }; |
222 | ||
85190d15 LT |
223 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
224 | static inline bool pipe_readable(const struct pipe_inode_info *pipe) | |
225 | { | |
226 | unsigned int head = READ_ONCE(pipe->head); | |
227 | unsigned int tail = READ_ONCE(pipe->tail); | |
228 | unsigned int writers = READ_ONCE(pipe->writers); | |
229 | ||
230 | return !pipe_empty(head, tail) || !writers; | |
231 | } | |
232 | ||
1da177e4 | 233 | static ssize_t |
fb9096a3 | 234 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 235 | { |
fb9096a3 | 236 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 237 | struct file *filp = iocb->ki_filp; |
de32ec4c | 238 | struct pipe_inode_info *pipe = filp->private_data; |
0ddad21d | 239 | bool was_full, wake_next_reader = false; |
1da177e4 | 240 | ssize_t ret; |
1da177e4 | 241 | |
1da177e4 LT |
242 | /* Null read succeeds. */ |
243 | if (unlikely(total_len == 0)) | |
244 | return 0; | |
245 | ||
1da177e4 | 246 | ret = 0; |
ebec73f4 | 247 | __pipe_lock(pipe); |
f467a6a6 LT |
248 | |
249 | /* | |
250 | * We only wake up writers if the pipe was full when we started | |
251 | * reading in order to avoid unnecessary wakeups. | |
252 | * | |
253 | * But when we do wake up writers, we do so using a sync wakeup | |
254 | * (WF_SYNC), because we want them to get going and generate more | |
255 | * data for us. | |
256 | */ | |
257 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); | |
1da177e4 | 258 | for (;;) { |
8cefc107 DH |
259 | unsigned int head = pipe->head; |
260 | unsigned int tail = pipe->tail; | |
261 | unsigned int mask = pipe->ring_size - 1; | |
262 | ||
e7d553d6 DH |
263 | #ifdef CONFIG_WATCH_QUEUE |
264 | if (pipe->note_loss) { | |
265 | struct watch_notification n; | |
266 | ||
267 | if (total_len < 8) { | |
268 | if (ret == 0) | |
269 | ret = -ENOBUFS; | |
270 | break; | |
271 | } | |
272 | ||
273 | n.type = WATCH_TYPE_META; | |
274 | n.subtype = WATCH_META_LOSS_NOTIFICATION; | |
275 | n.info = watch_sizeof(n); | |
276 | if (copy_to_iter(&n, sizeof(n), to) != sizeof(n)) { | |
277 | if (ret == 0) | |
278 | ret = -EFAULT; | |
279 | break; | |
280 | } | |
281 | ret += sizeof(n); | |
282 | total_len -= sizeof(n); | |
283 | pipe->note_loss = false; | |
284 | } | |
285 | #endif | |
286 | ||
8cefc107 DH |
287 | if (!pipe_empty(head, tail)) { |
288 | struct pipe_buffer *buf = &pipe->bufs[tail & mask]; | |
1da177e4 | 289 | size_t chars = buf->len; |
637b58c2 AV |
290 | size_t written; |
291 | int error; | |
1da177e4 | 292 | |
8cfba763 DH |
293 | if (chars > total_len) { |
294 | if (buf->flags & PIPE_BUF_FLAG_WHOLE) { | |
295 | if (ret == 0) | |
296 | ret = -ENOBUFS; | |
297 | break; | |
298 | } | |
1da177e4 | 299 | chars = total_len; |
8cfba763 | 300 | } |
1da177e4 | 301 | |
fba597db | 302 | error = pipe_buf_confirm(pipe, buf); |
f84d7519 | 303 | if (error) { |
5274f052 | 304 | if (!ret) |
e5953cbd | 305 | ret = error; |
5274f052 JA |
306 | break; |
307 | } | |
f84d7519 | 308 | |
fb9096a3 | 309 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 310 | if (unlikely(written < chars)) { |
341b446b | 311 | if (!ret) |
637b58c2 | 312 | ret = -EFAULT; |
1da177e4 LT |
313 | break; |
314 | } | |
315 | ret += chars; | |
316 | buf->offset += chars; | |
317 | buf->len -= chars; | |
9883035a LT |
318 | |
319 | /* Was it a packet buffer? Clean up and exit */ | |
320 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
321 | total_len = chars; | |
322 | buf->len = 0; | |
323 | } | |
324 | ||
1da177e4 | 325 | if (!buf->len) { |
a779638c | 326 | pipe_buf_release(pipe, buf); |
0ddad21d | 327 | spin_lock_irq(&pipe->rd_wait.lock); |
e7d553d6 DH |
328 | #ifdef CONFIG_WATCH_QUEUE |
329 | if (buf->flags & PIPE_BUF_FLAG_LOSS) | |
330 | pipe->note_loss = true; | |
331 | #endif | |
8cefc107 DH |
332 | tail++; |
333 | pipe->tail = tail; | |
0ddad21d | 334 | spin_unlock_irq(&pipe->rd_wait.lock); |
1da177e4 LT |
335 | } |
336 | total_len -= chars; | |
337 | if (!total_len) | |
338 | break; /* common path: read succeeded */ | |
8cefc107 DH |
339 | if (!pipe_empty(head, tail)) /* More to do? */ |
340 | continue; | |
1da177e4 | 341 | } |
8cefc107 | 342 | |
923f4f23 | 343 | if (!pipe->writers) |
1da177e4 | 344 | break; |
a28c8b9d LT |
345 | if (ret) |
346 | break; | |
347 | if (filp->f_flags & O_NONBLOCK) { | |
348 | ret = -EAGAIN; | |
349 | break; | |
1da177e4 | 350 | } |
85190d15 | 351 | __pipe_unlock(pipe); |
d1c6a2aa LT |
352 | |
353 | /* | |
354 | * We only get here if we didn't actually read anything. | |
355 | * | |
356 | * However, we could have seen (and removed) a zero-sized | |
357 | * pipe buffer, and might have made space in the buffers | |
358 | * that way. | |
359 | * | |
360 | * You can't make zero-sized pipe buffers by doing an empty | |
361 | * write (not even in packet mode), but they can happen if | |
362 | * the writer gets an EFAULT when trying to fill a buffer | |
363 | * that already got allocated and inserted in the buffer | |
364 | * array. | |
365 | * | |
366 | * So we still need to wake up any pending writers in the | |
367 | * _very_ unlikely case that the pipe was full, but we got | |
368 | * no data. | |
369 | */ | |
370 | if (unlikely(was_full)) { | |
0ddad21d | 371 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
f467a6a6 LT |
372 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
373 | } | |
d1c6a2aa LT |
374 | |
375 | /* | |
376 | * But because we didn't read anything, at this point we can | |
377 | * just return directly with -ERESTARTSYS if we're interrupted, | |
378 | * since we've done any required wakeups and there's no need | |
379 | * to mark anything accessed. And we've dropped the lock. | |
380 | */ | |
0ddad21d | 381 | if (wait_event_interruptible_exclusive(pipe->rd_wait, pipe_readable(pipe)) < 0) |
d1c6a2aa LT |
382 | return -ERESTARTSYS; |
383 | ||
85190d15 | 384 | __pipe_lock(pipe); |
f467a6a6 | 385 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); |
0ddad21d | 386 | wake_next_reader = true; |
1da177e4 | 387 | } |
0ddad21d LT |
388 | if (pipe_empty(pipe->head, pipe->tail)) |
389 | wake_next_reader = false; | |
ebec73f4 | 390 | __pipe_unlock(pipe); |
341b446b | 391 | |
f467a6a6 | 392 | if (was_full) { |
0ddad21d | 393 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
923f4f23 | 394 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 | 395 | } |
0ddad21d LT |
396 | if (wake_next_reader) |
397 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); | |
1da177e4 LT |
398 | if (ret > 0) |
399 | file_accessed(filp); | |
400 | return ret; | |
401 | } | |
402 | ||
9883035a LT |
403 | static inline int is_packetized(struct file *file) |
404 | { | |
405 | return (file->f_flags & O_DIRECT) != 0; | |
406 | } | |
407 | ||
85190d15 LT |
408 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
409 | static inline bool pipe_writable(const struct pipe_inode_info *pipe) | |
410 | { | |
411 | unsigned int head = READ_ONCE(pipe->head); | |
412 | unsigned int tail = READ_ONCE(pipe->tail); | |
413 | unsigned int max_usage = READ_ONCE(pipe->max_usage); | |
414 | ||
415 | return !pipe_full(head, tail, max_usage) || | |
416 | !READ_ONCE(pipe->readers); | |
417 | } | |
418 | ||
1da177e4 | 419 | static ssize_t |
f0d1bec9 | 420 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 421 | { |
ee0b3e67 | 422 | struct file *filp = iocb->ki_filp; |
de32ec4c | 423 | struct pipe_inode_info *pipe = filp->private_data; |
8f868d68 | 424 | unsigned int head; |
f0d1bec9 | 425 | ssize_t ret = 0; |
f0d1bec9 | 426 | size_t total_len = iov_iter_count(from); |
1da177e4 | 427 | ssize_t chars; |
1b6b26ae | 428 | bool was_empty = false; |
0ddad21d | 429 | bool wake_next_writer = false; |
1da177e4 | 430 | |
1da177e4 LT |
431 | /* Null write succeeds. */ |
432 | if (unlikely(total_len == 0)) | |
433 | return 0; | |
434 | ||
ebec73f4 | 435 | __pipe_lock(pipe); |
1da177e4 | 436 | |
923f4f23 | 437 | if (!pipe->readers) { |
1da177e4 LT |
438 | send_sig(SIGPIPE, current, 0); |
439 | ret = -EPIPE; | |
440 | goto out; | |
441 | } | |
442 | ||
c73be61c DH |
443 | #ifdef CONFIG_WATCH_QUEUE |
444 | if (pipe->watch_queue) { | |
445 | ret = -EXDEV; | |
446 | goto out; | |
447 | } | |
448 | #endif | |
449 | ||
1b6b26ae LT |
450 | /* |
451 | * Only wake up if the pipe started out empty, since | |
452 | * otherwise there should be no readers waiting. | |
453 | * | |
454 | * If it wasn't empty we try to merge new data into | |
455 | * the last buffer. | |
456 | * | |
457 | * That naturally merges small writes, but it also | |
458 | * page-aligs the rest of the writes for large writes | |
459 | * spanning multiple pages. | |
460 | */ | |
8cefc107 | 461 | head = pipe->head; |
1b6b26ae LT |
462 | was_empty = pipe_empty(head, pipe->tail); |
463 | chars = total_len & (PAGE_SIZE-1); | |
464 | if (chars && !was_empty) { | |
8f868d68 | 465 | unsigned int mask = pipe->ring_size - 1; |
8cefc107 | 466 | struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; |
1da177e4 | 467 | int offset = buf->offset + buf->len; |
341b446b | 468 | |
f6dd9755 CH |
469 | if ((buf->flags & PIPE_BUF_FLAG_CAN_MERGE) && |
470 | offset + chars <= PAGE_SIZE) { | |
fba597db | 471 | ret = pipe_buf_confirm(pipe, buf); |
6ae08069 | 472 | if (ret) |
5274f052 | 473 | goto out; |
f84d7519 | 474 | |
f0d1bec9 AV |
475 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
476 | if (unlikely(ret < chars)) { | |
6ae08069 | 477 | ret = -EFAULT; |
1da177e4 | 478 | goto out; |
f6762b7a | 479 | } |
1b6b26ae | 480 | |
6ae08069 | 481 | buf->len += ret; |
f0d1bec9 | 482 | if (!iov_iter_count(from)) |
1da177e4 LT |
483 | goto out; |
484 | } | |
485 | } | |
486 | ||
487 | for (;;) { | |
923f4f23 | 488 | if (!pipe->readers) { |
1da177e4 | 489 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
490 | if (!ret) |
491 | ret = -EPIPE; | |
1da177e4 LT |
492 | break; |
493 | } | |
8cefc107 | 494 | |
a194dfe6 | 495 | head = pipe->head; |
8f868d68 DH |
496 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) { |
497 | unsigned int mask = pipe->ring_size - 1; | |
8cefc107 | 498 | struct pipe_buffer *buf = &pipe->bufs[head & mask]; |
923f4f23 | 499 | struct page *page = pipe->tmp_page; |
f0d1bec9 | 500 | int copied; |
1da177e4 LT |
501 | |
502 | if (!page) { | |
d86133bd | 503 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
504 | if (unlikely(!page)) { |
505 | ret = ret ? : -ENOMEM; | |
506 | break; | |
507 | } | |
923f4f23 | 508 | pipe->tmp_page = page; |
1da177e4 | 509 | } |
a194dfe6 DH |
510 | |
511 | /* Allocate a slot in the ring in advance and attach an | |
512 | * empty buffer. If we fault or otherwise fail to use | |
513 | * it, either the reader will consume it or it'll still | |
514 | * be there for the next write. | |
515 | */ | |
0ddad21d | 516 | spin_lock_irq(&pipe->rd_wait.lock); |
a194dfe6 DH |
517 | |
518 | head = pipe->head; | |
8f868d68 | 519 | if (pipe_full(head, pipe->tail, pipe->max_usage)) { |
0ddad21d | 520 | spin_unlock_irq(&pipe->rd_wait.lock); |
8df44129 DH |
521 | continue; |
522 | } | |
523 | ||
a194dfe6 | 524 | pipe->head = head + 1; |
0ddad21d | 525 | spin_unlock_irq(&pipe->rd_wait.lock); |
1da177e4 LT |
526 | |
527 | /* Insert it into the buffer array */ | |
a194dfe6 | 528 | buf = &pipe->bufs[head & mask]; |
1da177e4 LT |
529 | buf->page = page; |
530 | buf->ops = &anon_pipe_buf_ops; | |
531 | buf->offset = 0; | |
a194dfe6 | 532 | buf->len = 0; |
f6dd9755 | 533 | if (is_packetized(filp)) |
9883035a | 534 | buf->flags = PIPE_BUF_FLAG_PACKET; |
f6dd9755 CH |
535 | else |
536 | buf->flags = PIPE_BUF_FLAG_CAN_MERGE; | |
923f4f23 | 537 | pipe->tmp_page = NULL; |
1da177e4 | 538 | |
a194dfe6 DH |
539 | copied = copy_page_from_iter(page, 0, PAGE_SIZE, from); |
540 | if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) { | |
541 | if (!ret) | |
542 | ret = -EFAULT; | |
543 | break; | |
544 | } | |
545 | ret += copied; | |
546 | buf->offset = 0; | |
547 | buf->len = copied; | |
548 | ||
f0d1bec9 | 549 | if (!iov_iter_count(from)) |
1da177e4 LT |
550 | break; |
551 | } | |
8cefc107 | 552 | |
8f868d68 | 553 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) |
1da177e4 | 554 | continue; |
8cefc107 DH |
555 | |
556 | /* Wait for buffer space to become available. */ | |
1da177e4 | 557 | if (filp->f_flags & O_NONBLOCK) { |
341b446b IM |
558 | if (!ret) |
559 | ret = -EAGAIN; | |
1da177e4 LT |
560 | break; |
561 | } | |
562 | if (signal_pending(current)) { | |
341b446b IM |
563 | if (!ret) |
564 | ret = -ERESTARTSYS; | |
1da177e4 LT |
565 | break; |
566 | } | |
1b6b26ae LT |
567 | |
568 | /* | |
569 | * We're going to release the pipe lock and wait for more | |
570 | * space. We wake up any readers if necessary, and then | |
571 | * after waiting we need to re-check whether the pipe | |
572 | * become empty while we dropped the lock. | |
573 | */ | |
85190d15 | 574 | __pipe_unlock(pipe); |
1b6b26ae | 575 | if (was_empty) { |
0ddad21d | 576 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
1b6b26ae LT |
577 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
578 | } | |
0ddad21d | 579 | wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); |
85190d15 | 580 | __pipe_lock(pipe); |
0dd1e377 | 581 | was_empty = pipe_empty(pipe->head, pipe->tail); |
0ddad21d | 582 | wake_next_writer = true; |
1da177e4 LT |
583 | } |
584 | out: | |
0ddad21d LT |
585 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) |
586 | wake_next_writer = false; | |
ebec73f4 | 587 | __pipe_unlock(pipe); |
1b6b26ae LT |
588 | |
589 | /* | |
590 | * If we do do a wakeup event, we do a 'sync' wakeup, because we | |
591 | * want the reader to start processing things asap, rather than | |
592 | * leave the data pending. | |
593 | * | |
594 | * This is particularly important for small writes, because of | |
595 | * how (for example) the GNU make jobserver uses small writes to | |
596 | * wake up pending jobs | |
597 | */ | |
598 | if (was_empty) { | |
0ddad21d | 599 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
923f4f23 | 600 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 | 601 | } |
0ddad21d LT |
602 | if (wake_next_writer) |
603 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); | |
7e775f46 | 604 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
605 | int err = file_update_time(filp); |
606 | if (err) | |
607 | ret = err; | |
7e775f46 | 608 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 609 | } |
1da177e4 LT |
610 | return ret; |
611 | } | |
612 | ||
d59d0b1b | 613 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 614 | { |
de32ec4c | 615 | struct pipe_inode_info *pipe = filp->private_data; |
8cefc107 | 616 | int count, head, tail, mask; |
1da177e4 LT |
617 | |
618 | switch (cmd) { | |
c73be61c DH |
619 | case FIONREAD: |
620 | __pipe_lock(pipe); | |
621 | count = 0; | |
622 | head = pipe->head; | |
623 | tail = pipe->tail; | |
624 | mask = pipe->ring_size - 1; | |
8cefc107 | 625 | |
c73be61c DH |
626 | while (tail != head) { |
627 | count += pipe->bufs[tail & mask].len; | |
628 | tail++; | |
629 | } | |
630 | __pipe_unlock(pipe); | |
923f4f23 | 631 | |
c73be61c | 632 | return put_user(count, (int __user *)arg); |
923f4f23 | 633 | |
c73be61c DH |
634 | #ifdef CONFIG_WATCH_QUEUE |
635 | case IOC_WATCH_QUEUE_SET_SIZE: { | |
636 | int ret; | |
637 | __pipe_lock(pipe); | |
638 | ret = watch_queue_set_size(pipe, arg); | |
639 | __pipe_unlock(pipe); | |
640 | return ret; | |
641 | } | |
642 | ||
643 | case IOC_WATCH_QUEUE_SET_FILTER: | |
644 | return watch_queue_set_filter( | |
645 | pipe, (struct watch_notification_filter __user *)arg); | |
646 | #endif | |
647 | ||
648 | default: | |
649 | return -ENOIOCTLCMD; | |
1da177e4 LT |
650 | } |
651 | } | |
652 | ||
dd67081b | 653 | /* No kernel lock held - fine */ |
a11e1d43 LT |
654 | static __poll_t |
655 | pipe_poll(struct file *filp, poll_table *wait) | |
dd67081b | 656 | { |
a11e1d43 | 657 | __poll_t mask; |
dd67081b | 658 | struct pipe_inode_info *pipe = filp->private_data; |
ad910e36 | 659 | unsigned int head, tail; |
a11e1d43 | 660 | |
ad910e36 | 661 | /* |
0ddad21d | 662 | * Reading pipe state only -- no need for acquiring the semaphore. |
ad910e36 LT |
663 | * |
664 | * But because this is racy, the code has to add the | |
665 | * entry to the poll table _first_ .. | |
666 | */ | |
0ddad21d LT |
667 | if (filp->f_mode & FMODE_READ) |
668 | poll_wait(filp, &pipe->rd_wait, wait); | |
669 | if (filp->f_mode & FMODE_WRITE) | |
670 | poll_wait(filp, &pipe->wr_wait, wait); | |
1da177e4 | 671 | |
ad910e36 LT |
672 | /* |
673 | * .. and only then can you do the racy tests. That way, | |
674 | * if something changes and you got it wrong, the poll | |
675 | * table entry will wake you up and fix it. | |
676 | */ | |
677 | head = READ_ONCE(pipe->head); | |
678 | tail = READ_ONCE(pipe->tail); | |
679 | ||
a11e1d43 | 680 | mask = 0; |
1da177e4 | 681 | if (filp->f_mode & FMODE_READ) { |
8cefc107 DH |
682 | if (!pipe_empty(head, tail)) |
683 | mask |= EPOLLIN | EPOLLRDNORM; | |
923f4f23 | 684 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
a9a08845 | 685 | mask |= EPOLLHUP; |
1da177e4 LT |
686 | } |
687 | ||
688 | if (filp->f_mode & FMODE_WRITE) { | |
6718b6f8 | 689 | if (!pipe_full(head, tail, pipe->max_usage)) |
8cefc107 | 690 | mask |= EPOLLOUT | EPOLLWRNORM; |
5e5d7a22 | 691 | /* |
a9a08845 | 692 | * Most Unices do not set EPOLLERR for FIFOs but on Linux they |
5e5d7a22 PE |
693 | * behave exactly like pipes for poll(). |
694 | */ | |
923f4f23 | 695 | if (!pipe->readers) |
a9a08845 | 696 | mask |= EPOLLERR; |
1da177e4 LT |
697 | } |
698 | ||
699 | return mask; | |
700 | } | |
701 | ||
b0d8d229 LT |
702 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
703 | { | |
704 | int kill = 0; | |
705 | ||
706 | spin_lock(&inode->i_lock); | |
707 | if (!--pipe->files) { | |
708 | inode->i_pipe = NULL; | |
709 | kill = 1; | |
710 | } | |
711 | spin_unlock(&inode->i_lock); | |
712 | ||
713 | if (kill) | |
714 | free_pipe_info(pipe); | |
715 | } | |
716 | ||
1da177e4 | 717 | static int |
599a0ac1 | 718 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 719 | { |
b0d8d229 | 720 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 721 | |
ebec73f4 | 722 | __pipe_lock(pipe); |
599a0ac1 AV |
723 | if (file->f_mode & FMODE_READ) |
724 | pipe->readers--; | |
725 | if (file->f_mode & FMODE_WRITE) | |
726 | pipe->writers--; | |
341b446b | 727 | |
6551d5c5 LT |
728 | /* Was that the last reader or writer, but not the other side? */ |
729 | if (!pipe->readers != !pipe->writers) { | |
730 | wake_up_interruptible_all(&pipe->rd_wait); | |
731 | wake_up_interruptible_all(&pipe->wr_wait); | |
923f4f23 IM |
732 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
733 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 734 | } |
ebec73f4 | 735 | __pipe_unlock(pipe); |
ba5bb147 | 736 | |
b0d8d229 | 737 | put_pipe_info(inode, pipe); |
1da177e4 LT |
738 | return 0; |
739 | } | |
740 | ||
741 | static int | |
599a0ac1 | 742 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 743 | { |
de32ec4c | 744 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 745 | int retval = 0; |
1da177e4 | 746 | |
ebec73f4 | 747 | __pipe_lock(pipe); |
599a0ac1 AV |
748 | if (filp->f_mode & FMODE_READ) |
749 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
750 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 751 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
752 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
753 | /* this can happen only if on == T */ | |
e5bc49ba ON |
754 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
755 | } | |
ebec73f4 | 756 | __pipe_unlock(pipe); |
60aa4924 | 757 | return retval; |
1da177e4 LT |
758 | } |
759 | ||
c73be61c DH |
760 | unsigned long account_pipe_buffers(struct user_struct *user, |
761 | unsigned long old, unsigned long new) | |
759c0114 | 762 | { |
9c87bcf0 | 763 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
764 | } |
765 | ||
c73be61c | 766 | bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 767 | { |
f7340761 EB |
768 | unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft); |
769 | ||
770 | return soft_limit && user_bufs > soft_limit; | |
759c0114 WT |
771 | } |
772 | ||
c73be61c | 773 | bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 774 | { |
f7340761 EB |
775 | unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard); |
776 | ||
777 | return hard_limit && user_bufs > hard_limit; | |
759c0114 WT |
778 | } |
779 | ||
c73be61c | 780 | bool pipe_is_unprivileged_user(void) |
85c2dd54 EB |
781 | { |
782 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
783 | } | |
784 | ||
7bee130e | 785 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 786 | { |
923f4f23 | 787 | struct pipe_inode_info *pipe; |
09b4d199 MK |
788 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
789 | struct user_struct *user = get_current_user(); | |
9c87bcf0 | 790 | unsigned long user_bufs; |
f7340761 | 791 | unsigned int max_size = READ_ONCE(pipe_max_size); |
3a326a2c | 792 | |
d86133bd | 793 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
09b4d199 MK |
794 | if (pipe == NULL) |
795 | goto out_free_uid; | |
796 | ||
f7340761 EB |
797 | if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE)) |
798 | pipe_bufs = max_size >> PAGE_SHIFT; | |
086e774a | 799 | |
9c87bcf0 | 800 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
a005ca0e | 801 | |
c73be61c | 802 | if (too_many_pipe_buffers_soft(user_bufs) && pipe_is_unprivileged_user()) { |
9c87bcf0 | 803 | user_bufs = account_pipe_buffers(user, pipe_bufs, 1); |
a005ca0e | 804 | pipe_bufs = 1; |
09b4d199 | 805 | } |
759c0114 | 806 | |
c73be61c | 807 | if (too_many_pipe_buffers_hard(user_bufs) && pipe_is_unprivileged_user()) |
a005ca0e MK |
808 | goto out_revert_acct; |
809 | ||
810 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), | |
811 | GFP_KERNEL_ACCOUNT); | |
812 | ||
09b4d199 | 813 | if (pipe->bufs) { |
0ddad21d LT |
814 | init_waitqueue_head(&pipe->rd_wait); |
815 | init_waitqueue_head(&pipe->wr_wait); | |
09b4d199 | 816 | pipe->r_counter = pipe->w_counter = 1; |
6718b6f8 | 817 | pipe->max_usage = pipe_bufs; |
8cefc107 | 818 | pipe->ring_size = pipe_bufs; |
c73be61c | 819 | pipe->nr_accounted = pipe_bufs; |
09b4d199 | 820 | pipe->user = user; |
09b4d199 MK |
821 | mutex_init(&pipe->mutex); |
822 | return pipe; | |
3a326a2c IM |
823 | } |
824 | ||
a005ca0e | 825 | out_revert_acct: |
9c87bcf0 | 826 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
09b4d199 MK |
827 | kfree(pipe); |
828 | out_free_uid: | |
829 | free_uid(user); | |
35f3d14d | 830 | return NULL; |
3a326a2c IM |
831 | } |
832 | ||
4b8a8f1e | 833 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
834 | { |
835 | int i; | |
1da177e4 | 836 | |
c73be61c DH |
837 | #ifdef CONFIG_WATCH_QUEUE |
838 | if (pipe->watch_queue) { | |
839 | watch_queue_clear(pipe->watch_queue); | |
840 | put_watch_queue(pipe->watch_queue); | |
841 | } | |
842 | #endif | |
843 | ||
844 | (void) account_pipe_buffers(pipe->user, pipe->nr_accounted, 0); | |
759c0114 | 845 | free_uid(pipe->user); |
8cefc107 | 846 | for (i = 0; i < pipe->ring_size; i++) { |
923f4f23 | 847 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 848 | if (buf->ops) |
a779638c | 849 | pipe_buf_release(pipe, buf); |
1da177e4 | 850 | } |
923f4f23 IM |
851 | if (pipe->tmp_page) |
852 | __free_page(pipe->tmp_page); | |
35f3d14d | 853 | kfree(pipe->bufs); |
923f4f23 | 854 | kfree(pipe); |
1da177e4 LT |
855 | } |
856 | ||
fa3536cc | 857 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 858 | |
c23fbb6b ED |
859 | /* |
860 | * pipefs_dname() is called from d_path(). | |
861 | */ | |
862 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
863 | { | |
864 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
75c3cfa8 | 865 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
866 | } |
867 | ||
3ba13d17 | 868 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 869 | .d_dname = pipefs_dname, |
1da177e4 LT |
870 | }; |
871 | ||
872 | static struct inode * get_pipe_inode(void) | |
873 | { | |
a209dfc7 | 874 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 875 | struct pipe_inode_info *pipe; |
1da177e4 LT |
876 | |
877 | if (!inode) | |
878 | goto fail_inode; | |
879 | ||
85fe4025 CH |
880 | inode->i_ino = get_next_ino(); |
881 | ||
7bee130e | 882 | pipe = alloc_pipe_info(); |
923f4f23 | 883 | if (!pipe) |
1da177e4 | 884 | goto fail_iput; |
3a326a2c | 885 | |
ba5bb147 AV |
886 | inode->i_pipe = pipe; |
887 | pipe->files = 2; | |
923f4f23 | 888 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 889 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
890 | |
891 | /* | |
892 | * Mark the inode dirty from the very beginning, | |
893 | * that way it will never be moved to the dirty | |
894 | * list because "mark_inode_dirty()" will think | |
895 | * that it already _is_ on the dirty list. | |
896 | */ | |
897 | inode->i_state = I_DIRTY; | |
898 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
899 | inode->i_uid = current_fsuid(); |
900 | inode->i_gid = current_fsgid(); | |
078cd827 | 901 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
923f4f23 | 902 | |
1da177e4 LT |
903 | return inode; |
904 | ||
905 | fail_iput: | |
906 | iput(inode); | |
341b446b | 907 | |
1da177e4 LT |
908 | fail_inode: |
909 | return NULL; | |
910 | } | |
911 | ||
e4fad8e5 | 912 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 913 | { |
e4fad8e5 | 914 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 915 | struct file *f; |
1da177e4 | 916 | |
1da177e4 | 917 | if (!inode) |
e4fad8e5 | 918 | return -ENFILE; |
1da177e4 | 919 | |
c73be61c DH |
920 | if (flags & O_NOTIFICATION_PIPE) { |
921 | #ifdef CONFIG_WATCH_QUEUE | |
922 | if (watch_queue_init(inode->i_pipe) < 0) { | |
923 | iput(inode); | |
924 | return -ENOMEM; | |
925 | } | |
926 | #else | |
927 | return -ENOPKG; | |
928 | #endif | |
929 | } | |
930 | ||
152b6372 AV |
931 | f = alloc_file_pseudo(inode, pipe_mnt, "", |
932 | O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)), | |
933 | &pipefifo_fops); | |
e9bb1f9b | 934 | if (IS_ERR(f)) { |
152b6372 AV |
935 | free_pipe_info(inode->i_pipe); |
936 | iput(inode); | |
937 | return PTR_ERR(f); | |
e9bb1f9b | 938 | } |
341b446b | 939 | |
de32ec4c | 940 | f->private_data = inode->i_pipe; |
d6cbd281 | 941 | |
183266f2 AV |
942 | res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK), |
943 | &pipefifo_fops); | |
e9bb1f9b | 944 | if (IS_ERR(res[0])) { |
b10a4a9f AV |
945 | put_pipe_info(inode, inode->i_pipe); |
946 | fput(f); | |
947 | return PTR_ERR(res[0]); | |
e9bb1f9b | 948 | } |
de32ec4c | 949 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 | 950 | res[1] = f; |
d8e464ec LT |
951 | stream_open(inode, res[0]); |
952 | stream_open(inode, res[1]); | |
e4fad8e5 | 953 | return 0; |
d6cbd281 AK |
954 | } |
955 | ||
5b249b1b | 956 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 957 | { |
d6cbd281 AK |
958 | int error; |
959 | int fdw, fdr; | |
960 | ||
c73be61c | 961 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT | O_NOTIFICATION_PIPE)) |
ed8cae8b UD |
962 | return -EINVAL; |
963 | ||
e4fad8e5 AV |
964 | error = create_pipe_files(files, flags); |
965 | if (error) | |
966 | return error; | |
d6cbd281 | 967 | |
ed8cae8b | 968 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
969 | if (error < 0) |
970 | goto err_read_pipe; | |
971 | fdr = error; | |
972 | ||
ed8cae8b | 973 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
974 | if (error < 0) |
975 | goto err_fdr; | |
976 | fdw = error; | |
977 | ||
157cf649 | 978 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
979 | fd[0] = fdr; |
980 | fd[1] = fdw; | |
d6cbd281 AK |
981 | return 0; |
982 | ||
983 | err_fdr: | |
984 | put_unused_fd(fdr); | |
985 | err_read_pipe: | |
e4fad8e5 AV |
986 | fput(files[0]); |
987 | fput(files[1]); | |
d6cbd281 | 988 | return error; |
1da177e4 LT |
989 | } |
990 | ||
5b249b1b AV |
991 | int do_pipe_flags(int *fd, int flags) |
992 | { | |
993 | struct file *files[2]; | |
994 | int error = __do_pipe_flags(fd, files, flags); | |
995 | if (!error) { | |
996 | fd_install(fd[0], files[0]); | |
997 | fd_install(fd[1], files[1]); | |
998 | } | |
999 | return error; | |
1000 | } | |
1001 | ||
d35c7b0e UD |
1002 | /* |
1003 | * sys_pipe() is the normal C calling standard for creating | |
1004 | * a pipe. It's not the way Unix traditionally does this, though. | |
1005 | */ | |
0a216dd1 | 1006 | static int do_pipe2(int __user *fildes, int flags) |
d35c7b0e | 1007 | { |
5b249b1b | 1008 | struct file *files[2]; |
d35c7b0e UD |
1009 | int fd[2]; |
1010 | int error; | |
1011 | ||
5b249b1b | 1012 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 1013 | if (!error) { |
5b249b1b AV |
1014 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
1015 | fput(files[0]); | |
1016 | fput(files[1]); | |
1017 | put_unused_fd(fd[0]); | |
1018 | put_unused_fd(fd[1]); | |
d35c7b0e | 1019 | error = -EFAULT; |
5b249b1b AV |
1020 | } else { |
1021 | fd_install(fd[0], files[0]); | |
1022 | fd_install(fd[1], files[1]); | |
ba719bae | 1023 | } |
d35c7b0e UD |
1024 | } |
1025 | return error; | |
1026 | } | |
1027 | ||
0a216dd1 DB |
1028 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
1029 | { | |
1030 | return do_pipe2(fildes, flags); | |
1031 | } | |
1032 | ||
2b664219 | 1033 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b | 1034 | { |
0a216dd1 | 1035 | return do_pipe2(fildes, 0); |
ed8cae8b UD |
1036 | } |
1037 | ||
fc7478a2 | 1038 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 | 1039 | { |
8cefc107 | 1040 | int cur = *cnt; |
f776c738 AV |
1041 | |
1042 | while (cur == *cnt) { | |
fc7478a2 | 1043 | pipe_wait(pipe); |
f776c738 AV |
1044 | if (signal_pending(current)) |
1045 | break; | |
1046 | } | |
1047 | return cur == *cnt ? -ERESTARTSYS : 0; | |
1048 | } | |
1049 | ||
fc7478a2 | 1050 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 1051 | { |
6551d5c5 LT |
1052 | wake_up_interruptible_all(&pipe->rd_wait); |
1053 | wake_up_interruptible_all(&pipe->wr_wait); | |
f776c738 AV |
1054 | } |
1055 | ||
1056 | static int fifo_open(struct inode *inode, struct file *filp) | |
1057 | { | |
1058 | struct pipe_inode_info *pipe; | |
599a0ac1 | 1059 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
1060 | int ret; |
1061 | ||
ba5bb147 AV |
1062 | filp->f_version = 0; |
1063 | ||
1064 | spin_lock(&inode->i_lock); | |
1065 | if (inode->i_pipe) { | |
1066 | pipe = inode->i_pipe; | |
1067 | pipe->files++; | |
1068 | spin_unlock(&inode->i_lock); | |
1069 | } else { | |
1070 | spin_unlock(&inode->i_lock); | |
7bee130e | 1071 | pipe = alloc_pipe_info(); |
f776c738 | 1072 | if (!pipe) |
ba5bb147 AV |
1073 | return -ENOMEM; |
1074 | pipe->files = 1; | |
1075 | spin_lock(&inode->i_lock); | |
1076 | if (unlikely(inode->i_pipe)) { | |
1077 | inode->i_pipe->files++; | |
1078 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 1079 | free_pipe_info(pipe); |
ba5bb147 AV |
1080 | pipe = inode->i_pipe; |
1081 | } else { | |
1082 | inode->i_pipe = pipe; | |
1083 | spin_unlock(&inode->i_lock); | |
1084 | } | |
f776c738 | 1085 | } |
de32ec4c | 1086 | filp->private_data = pipe; |
ba5bb147 AV |
1087 | /* OK, we have a pipe and it's pinned down */ |
1088 | ||
ebec73f4 | 1089 | __pipe_lock(pipe); |
f776c738 AV |
1090 | |
1091 | /* We can only do regular read/write on fifos */ | |
d8e464ec | 1092 | stream_open(inode, filp); |
f776c738 | 1093 | |
d8e464ec | 1094 | switch (filp->f_mode & (FMODE_READ | FMODE_WRITE)) { |
f776c738 AV |
1095 | case FMODE_READ: |
1096 | /* | |
1097 | * O_RDONLY | |
1098 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
1099 | * opened, even when there is no process writing the FIFO. | |
1100 | */ | |
f776c738 AV |
1101 | pipe->r_counter++; |
1102 | if (pipe->readers++ == 0) | |
fc7478a2 | 1103 | wake_up_partner(pipe); |
f776c738 | 1104 | |
599a0ac1 | 1105 | if (!is_pipe && !pipe->writers) { |
f776c738 | 1106 | if ((filp->f_flags & O_NONBLOCK)) { |
a9a08845 | 1107 | /* suppress EPOLLHUP until we have |
f776c738 AV |
1108 | * seen a writer */ |
1109 | filp->f_version = pipe->w_counter; | |
1110 | } else { | |
fc7478a2 | 1111 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
1112 | goto err_rd; |
1113 | } | |
1114 | } | |
1115 | break; | |
8cefc107 | 1116 | |
f776c738 AV |
1117 | case FMODE_WRITE: |
1118 | /* | |
1119 | * O_WRONLY | |
1120 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
1121 | * errno=ENXIO when there is no process reading the FIFO. | |
1122 | */ | |
1123 | ret = -ENXIO; | |
599a0ac1 | 1124 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
1125 | goto err; |
1126 | ||
f776c738 AV |
1127 | pipe->w_counter++; |
1128 | if (!pipe->writers++) | |
fc7478a2 | 1129 | wake_up_partner(pipe); |
f776c738 | 1130 | |
599a0ac1 | 1131 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 1132 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
1133 | goto err_wr; |
1134 | } | |
1135 | break; | |
8cefc107 | 1136 | |
f776c738 AV |
1137 | case FMODE_READ | FMODE_WRITE: |
1138 | /* | |
1139 | * O_RDWR | |
1140 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
1141 | * This implementation will NEVER block on a O_RDWR open, since | |
1142 | * the process can at least talk to itself. | |
1143 | */ | |
f776c738 AV |
1144 | |
1145 | pipe->readers++; | |
1146 | pipe->writers++; | |
1147 | pipe->r_counter++; | |
1148 | pipe->w_counter++; | |
1149 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 1150 | wake_up_partner(pipe); |
f776c738 AV |
1151 | break; |
1152 | ||
1153 | default: | |
1154 | ret = -EINVAL; | |
1155 | goto err; | |
1156 | } | |
1157 | ||
1158 | /* Ok! */ | |
ebec73f4 | 1159 | __pipe_unlock(pipe); |
f776c738 AV |
1160 | return 0; |
1161 | ||
1162 | err_rd: | |
1163 | if (!--pipe->readers) | |
0ddad21d | 1164 | wake_up_interruptible(&pipe->wr_wait); |
f776c738 AV |
1165 | ret = -ERESTARTSYS; |
1166 | goto err; | |
1167 | ||
1168 | err_wr: | |
1169 | if (!--pipe->writers) | |
6551d5c5 | 1170 | wake_up_interruptible_all(&pipe->rd_wait); |
f776c738 AV |
1171 | ret = -ERESTARTSYS; |
1172 | goto err; | |
1173 | ||
1174 | err: | |
ebec73f4 | 1175 | __pipe_unlock(pipe); |
b0d8d229 LT |
1176 | |
1177 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1178 | return ret; |
1179 | } | |
1180 | ||
599a0ac1 AV |
1181 | const struct file_operations pipefifo_fops = { |
1182 | .open = fifo_open, | |
1183 | .llseek = no_llseek, | |
fb9096a3 | 1184 | .read_iter = pipe_read, |
f0d1bec9 | 1185 | .write_iter = pipe_write, |
a11e1d43 | 1186 | .poll = pipe_poll, |
599a0ac1 AV |
1187 | .unlocked_ioctl = pipe_ioctl, |
1188 | .release = pipe_release, | |
1189 | .fasync = pipe_fasync, | |
f776c738 AV |
1190 | }; |
1191 | ||
f491bd71 MK |
1192 | /* |
1193 | * Currently we rely on the pipe array holding a power-of-2 number | |
d3f14c48 | 1194 | * of pages. Returns 0 on error. |
f491bd71 | 1195 | */ |
96e99be4 | 1196 | unsigned int round_pipe_size(unsigned long size) |
f491bd71 | 1197 | { |
c4fed5a9 | 1198 | if (size > (1U << 31)) |
96e99be4 EB |
1199 | return 0; |
1200 | ||
4c2e4bef EB |
1201 | /* Minimum pipe size, as required by POSIX */ |
1202 | if (size < PAGE_SIZE) | |
c4fed5a9 | 1203 | return PAGE_SIZE; |
d3f14c48 | 1204 | |
c4fed5a9 | 1205 | return roundup_pow_of_two(size); |
f491bd71 MK |
1206 | } |
1207 | ||
35f3d14d | 1208 | /* |
c73be61c | 1209 | * Resize the pipe ring to a number of slots. |
35f3d14d | 1210 | */ |
c73be61c | 1211 | int pipe_resize_ring(struct pipe_inode_info *pipe, unsigned int nr_slots) |
35f3d14d JA |
1212 | { |
1213 | struct pipe_buffer *bufs; | |
c73be61c | 1214 | unsigned int head, tail, mask, n; |
35f3d14d | 1215 | |
35f3d14d | 1216 | /* |
8cefc107 DH |
1217 | * We can shrink the pipe, if arg is greater than the ring occupancy. |
1218 | * Since we don't expect a lot of shrink+grow operations, just free and | |
1219 | * allocate again like we would do for growing. If the pipe currently | |
35f3d14d JA |
1220 | * contains more buffers than arg, then return busy. |
1221 | */ | |
8cefc107 DH |
1222 | mask = pipe->ring_size - 1; |
1223 | head = pipe->head; | |
1224 | tail = pipe->tail; | |
1225 | n = pipe_occupancy(pipe->head, pipe->tail); | |
c73be61c DH |
1226 | if (nr_slots < n) |
1227 | return -EBUSY; | |
35f3d14d | 1228 | |
8cefc107 | 1229 | bufs = kcalloc(nr_slots, sizeof(*bufs), |
d86133bd | 1230 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); |
c73be61c DH |
1231 | if (unlikely(!bufs)) |
1232 | return -ENOMEM; | |
35f3d14d JA |
1233 | |
1234 | /* | |
1235 | * The pipe array wraps around, so just start the new one at zero | |
8cefc107 | 1236 | * and adjust the indices. |
35f3d14d | 1237 | */ |
8cefc107 DH |
1238 | if (n > 0) { |
1239 | unsigned int h = head & mask; | |
1240 | unsigned int t = tail & mask; | |
1241 | if (h > t) { | |
1242 | memcpy(bufs, pipe->bufs + t, | |
1243 | n * sizeof(struct pipe_buffer)); | |
1244 | } else { | |
1245 | unsigned int tsize = pipe->ring_size - t; | |
1246 | if (h > 0) | |
1247 | memcpy(bufs + tsize, pipe->bufs, | |
1248 | h * sizeof(struct pipe_buffer)); | |
1249 | memcpy(bufs, pipe->bufs + t, | |
1250 | tsize * sizeof(struct pipe_buffer)); | |
1251 | } | |
35f3d14d JA |
1252 | } |
1253 | ||
8cefc107 DH |
1254 | head = n; |
1255 | tail = 0; | |
1256 | ||
35f3d14d JA |
1257 | kfree(pipe->bufs); |
1258 | pipe->bufs = bufs; | |
8cefc107 | 1259 | pipe->ring_size = nr_slots; |
c73be61c DH |
1260 | if (pipe->max_usage > nr_slots) |
1261 | pipe->max_usage = nr_slots; | |
8cefc107 DH |
1262 | pipe->tail = tail; |
1263 | pipe->head = head; | |
6551d5c5 LT |
1264 | |
1265 | /* This might have made more room for writers */ | |
1266 | wake_up_interruptible(&pipe->wr_wait); | |
c73be61c DH |
1267 | return 0; |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1272 | * pipe size if successful, or return -ERROR on error. | |
1273 | */ | |
1274 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) | |
1275 | { | |
1276 | unsigned long user_bufs; | |
1277 | unsigned int nr_slots, size; | |
1278 | long ret = 0; | |
1279 | ||
1280 | #ifdef CONFIG_WATCH_QUEUE | |
1281 | if (pipe->watch_queue) | |
1282 | return -EBUSY; | |
1283 | #endif | |
1284 | ||
1285 | size = round_pipe_size(arg); | |
1286 | nr_slots = size >> PAGE_SHIFT; | |
1287 | ||
1288 | if (!nr_slots) | |
1289 | return -EINVAL; | |
1290 | ||
1291 | /* | |
1292 | * If trying to increase the pipe capacity, check that an | |
1293 | * unprivileged user is not trying to exceed various limits | |
1294 | * (soft limit check here, hard limit check just below). | |
1295 | * Decreasing the pipe capacity is always permitted, even | |
1296 | * if the user is currently over a limit. | |
1297 | */ | |
1298 | if (nr_slots > pipe->max_usage && | |
1299 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) | |
1300 | return -EPERM; | |
1301 | ||
1302 | user_bufs = account_pipe_buffers(pipe->user, pipe->nr_accounted, nr_slots); | |
1303 | ||
1304 | if (nr_slots > pipe->max_usage && | |
1305 | (too_many_pipe_buffers_hard(user_bufs) || | |
1306 | too_many_pipe_buffers_soft(user_bufs)) && | |
1307 | pipe_is_unprivileged_user()) { | |
1308 | ret = -EPERM; | |
1309 | goto out_revert_acct; | |
1310 | } | |
1311 | ||
1312 | ret = pipe_resize_ring(pipe, nr_slots); | |
1313 | if (ret < 0) | |
1314 | goto out_revert_acct; | |
1315 | ||
1316 | pipe->max_usage = nr_slots; | |
1317 | pipe->nr_accounted = nr_slots; | |
6718b6f8 | 1318 | return pipe->max_usage * PAGE_SIZE; |
b0b91d18 MK |
1319 | |
1320 | out_revert_acct: | |
c73be61c | 1321 | (void) account_pipe_buffers(pipe->user, nr_slots, pipe->nr_accounted); |
b0b91d18 | 1322 | return ret; |
35f3d14d JA |
1323 | } |
1324 | ||
72083646 LT |
1325 | /* |
1326 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1327 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1328 | * pipe. | |
1329 | */ | |
c73be61c | 1330 | struct pipe_inode_info *get_pipe_info(struct file *file, bool for_splice) |
72083646 | 1331 | { |
c73be61c DH |
1332 | struct pipe_inode_info *pipe = file->private_data; |
1333 | ||
1334 | if (file->f_op != &pipefifo_fops || !pipe) | |
1335 | return NULL; | |
1336 | #ifdef CONFIG_WATCH_QUEUE | |
1337 | if (for_splice && pipe->watch_queue) | |
1338 | return NULL; | |
1339 | #endif | |
1340 | return pipe; | |
72083646 LT |
1341 | } |
1342 | ||
35f3d14d JA |
1343 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
1344 | { | |
1345 | struct pipe_inode_info *pipe; | |
1346 | long ret; | |
1347 | ||
c73be61c | 1348 | pipe = get_pipe_info(file, false); |
35f3d14d JA |
1349 | if (!pipe) |
1350 | return -EBADF; | |
1351 | ||
ebec73f4 | 1352 | __pipe_lock(pipe); |
35f3d14d JA |
1353 | |
1354 | switch (cmd) { | |
d37d4166 MK |
1355 | case F_SETPIPE_SZ: |
1356 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1357 | break; |
1358 | case F_GETPIPE_SZ: | |
6718b6f8 | 1359 | ret = pipe->max_usage * PAGE_SIZE; |
35f3d14d JA |
1360 | break; |
1361 | default: | |
1362 | ret = -EINVAL; | |
1363 | break; | |
1364 | } | |
1365 | ||
ebec73f4 | 1366 | __pipe_unlock(pipe); |
35f3d14d JA |
1367 | return ret; |
1368 | } | |
1369 | ||
ff0c7d15 NP |
1370 | static const struct super_operations pipefs_ops = { |
1371 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1372 | .statfs = simple_statfs, |
ff0c7d15 NP |
1373 | }; |
1374 | ||
1da177e4 LT |
1375 | /* |
1376 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1377 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1378 | * any operations on the root directory. However, we need a non-trivial | |
1379 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1380 | */ | |
4fa7ec5d DH |
1381 | |
1382 | static int pipefs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1383 | { |
4fa7ec5d DH |
1384 | struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC); |
1385 | if (!ctx) | |
1386 | return -ENOMEM; | |
1387 | ctx->ops = &pipefs_ops; | |
1388 | ctx->dops = &pipefs_dentry_operations; | |
1389 | return 0; | |
1da177e4 LT |
1390 | } |
1391 | ||
1392 | static struct file_system_type pipe_fs_type = { | |
1393 | .name = "pipefs", | |
4fa7ec5d | 1394 | .init_fs_context = pipefs_init_fs_context, |
1da177e4 LT |
1395 | .kill_sb = kill_anon_super, |
1396 | }; | |
1397 | ||
1398 | static int __init init_pipe_fs(void) | |
1399 | { | |
1400 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1401 | |
1da177e4 LT |
1402 | if (!err) { |
1403 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1404 | if (IS_ERR(pipe_mnt)) { | |
1405 | err = PTR_ERR(pipe_mnt); | |
1406 | unregister_filesystem(&pipe_fs_type); | |
1407 | } | |
1408 | } | |
1409 | return err; | |
1410 | } | |
1411 | ||
1da177e4 | 1412 | fs_initcall(init_pipe_fs); |