]>
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 <[email protected]> 2001-08-16 | |
58 | * | |
59 | * pipe_read & write cleanup | |
60 | * -- Manfred Spraul <[email protected]> 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 | ||
341b446b IM |
109 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
110 | struct pipe_buffer *buf) | |
1da177e4 LT |
111 | { |
112 | struct page *page = buf->page; | |
113 | ||
5274f052 JA |
114 | /* |
115 | * If nobody else uses this page, and we don't already have a | |
116 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 117 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 118 | */ |
341b446b | 119 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 120 | pipe->tmp_page = page; |
341b446b | 121 | else |
09cbfeaf | 122 | put_page(page); |
1da177e4 LT |
123 | } |
124 | ||
c928f642 CH |
125 | static bool anon_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
126 | struct pipe_buffer *buf) | |
d86133bd VD |
127 | { |
128 | struct page *page = buf->page; | |
129 | ||
c928f642 CH |
130 | if (page_count(page) != 1) |
131 | return false; | |
132 | memcg_kmem_uncharge_page(page, 0); | |
133 | __SetPageLocked(page); | |
134 | return true; | |
d86133bd VD |
135 | } |
136 | ||
0845718d | 137 | /** |
c928f642 | 138 | * generic_pipe_buf_try_steal - attempt to take ownership of a &pipe_buffer |
0845718d JA |
139 | * @pipe: the pipe that the buffer belongs to |
140 | * @buf: the buffer to attempt to steal | |
141 | * | |
142 | * Description: | |
b51d63c6 | 143 | * This function attempts to steal the &struct page attached to |
0845718d JA |
144 | * @buf. If successful, this function returns 0 and returns with |
145 | * the page locked. The caller may then reuse the page for whatever | |
b51d63c6 | 146 | * he wishes; the typical use is insertion into a different file |
0845718d JA |
147 | * page cache. |
148 | */ | |
c928f642 CH |
149 | bool generic_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
150 | struct pipe_buffer *buf) | |
5abc97aa | 151 | { |
46e678c9 JA |
152 | struct page *page = buf->page; |
153 | ||
0845718d JA |
154 | /* |
155 | * A reference of one is golden, that means that the owner of this | |
156 | * page is the only one holding a reference to it. lock the page | |
157 | * and return OK. | |
158 | */ | |
46e678c9 | 159 | if (page_count(page) == 1) { |
46e678c9 | 160 | lock_page(page); |
c928f642 | 161 | return true; |
46e678c9 | 162 | } |
c928f642 | 163 | return false; |
5abc97aa | 164 | } |
c928f642 | 165 | EXPORT_SYMBOL(generic_pipe_buf_try_steal); |
5abc97aa | 166 | |
0845718d | 167 | /** |
b51d63c6 | 168 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
169 | * @pipe: the pipe that the buffer belongs to |
170 | * @buf: the buffer to get a reference to | |
171 | * | |
172 | * Description: | |
173 | * This function grabs an extra reference to @buf. It's used in | |
3d742d4b | 174 | * the tee() system call, when we duplicate the buffers in one |
0845718d JA |
175 | * pipe into another. |
176 | */ | |
15fab63e | 177 | bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
70524490 | 178 | { |
15fab63e | 179 | return try_get_page(buf->page); |
70524490 | 180 | } |
51921cb7 | 181 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 182 | |
6818173b MS |
183 | /** |
184 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
185 | * @pipe: the pipe that the buffer belongs to | |
186 | * @buf: the buffer to put a reference to | |
187 | * | |
188 | * Description: | |
189 | * This function releases a reference to @buf. | |
190 | */ | |
191 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
192 | struct pipe_buffer *buf) | |
193 | { | |
09cbfeaf | 194 | put_page(buf->page); |
6818173b | 195 | } |
51921cb7 | 196 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 197 | |
d4c3cca9 | 198 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
c928f642 CH |
199 | .release = anon_pipe_buf_release, |
200 | .try_steal = anon_pipe_buf_try_steal, | |
201 | .get = generic_pipe_buf_get, | |
1da177e4 LT |
202 | }; |
203 | ||
85190d15 LT |
204 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
205 | static inline bool pipe_readable(const struct pipe_inode_info *pipe) | |
206 | { | |
207 | unsigned int head = READ_ONCE(pipe->head); | |
208 | unsigned int tail = READ_ONCE(pipe->tail); | |
209 | unsigned int writers = READ_ONCE(pipe->writers); | |
210 | ||
211 | return !pipe_empty(head, tail) || !writers; | |
212 | } | |
213 | ||
1da177e4 | 214 | static ssize_t |
fb9096a3 | 215 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 216 | { |
fb9096a3 | 217 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 218 | struct file *filp = iocb->ki_filp; |
de32ec4c | 219 | struct pipe_inode_info *pipe = filp->private_data; |
0ddad21d | 220 | bool was_full, wake_next_reader = false; |
1da177e4 | 221 | ssize_t ret; |
1da177e4 | 222 | |
1da177e4 LT |
223 | /* Null read succeeds. */ |
224 | if (unlikely(total_len == 0)) | |
225 | return 0; | |
226 | ||
1da177e4 | 227 | ret = 0; |
ebec73f4 | 228 | __pipe_lock(pipe); |
f467a6a6 LT |
229 | |
230 | /* | |
231 | * We only wake up writers if the pipe was full when we started | |
232 | * reading in order to avoid unnecessary wakeups. | |
233 | * | |
234 | * But when we do wake up writers, we do so using a sync wakeup | |
235 | * (WF_SYNC), because we want them to get going and generate more | |
236 | * data for us. | |
237 | */ | |
238 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); | |
1da177e4 | 239 | for (;;) { |
8cefc107 DH |
240 | unsigned int head = pipe->head; |
241 | unsigned int tail = pipe->tail; | |
242 | unsigned int mask = pipe->ring_size - 1; | |
243 | ||
e7d553d6 DH |
244 | #ifdef CONFIG_WATCH_QUEUE |
245 | if (pipe->note_loss) { | |
246 | struct watch_notification n; | |
247 | ||
248 | if (total_len < 8) { | |
249 | if (ret == 0) | |
250 | ret = -ENOBUFS; | |
251 | break; | |
252 | } | |
253 | ||
254 | n.type = WATCH_TYPE_META; | |
255 | n.subtype = WATCH_META_LOSS_NOTIFICATION; | |
256 | n.info = watch_sizeof(n); | |
257 | if (copy_to_iter(&n, sizeof(n), to) != sizeof(n)) { | |
258 | if (ret == 0) | |
259 | ret = -EFAULT; | |
260 | break; | |
261 | } | |
262 | ret += sizeof(n); | |
263 | total_len -= sizeof(n); | |
264 | pipe->note_loss = false; | |
265 | } | |
266 | #endif | |
267 | ||
8cefc107 DH |
268 | if (!pipe_empty(head, tail)) { |
269 | struct pipe_buffer *buf = &pipe->bufs[tail & mask]; | |
1da177e4 | 270 | size_t chars = buf->len; |
637b58c2 AV |
271 | size_t written; |
272 | int error; | |
1da177e4 | 273 | |
8cfba763 DH |
274 | if (chars > total_len) { |
275 | if (buf->flags & PIPE_BUF_FLAG_WHOLE) { | |
276 | if (ret == 0) | |
277 | ret = -ENOBUFS; | |
278 | break; | |
279 | } | |
1da177e4 | 280 | chars = total_len; |
8cfba763 | 281 | } |
1da177e4 | 282 | |
fba597db | 283 | error = pipe_buf_confirm(pipe, buf); |
f84d7519 | 284 | if (error) { |
5274f052 | 285 | if (!ret) |
e5953cbd | 286 | ret = error; |
5274f052 JA |
287 | break; |
288 | } | |
f84d7519 | 289 | |
fb9096a3 | 290 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 291 | if (unlikely(written < chars)) { |
341b446b | 292 | if (!ret) |
637b58c2 | 293 | ret = -EFAULT; |
1da177e4 LT |
294 | break; |
295 | } | |
296 | ret += chars; | |
297 | buf->offset += chars; | |
298 | buf->len -= chars; | |
9883035a LT |
299 | |
300 | /* Was it a packet buffer? Clean up and exit */ | |
301 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
302 | total_len = chars; | |
303 | buf->len = 0; | |
304 | } | |
305 | ||
1da177e4 | 306 | if (!buf->len) { |
a779638c | 307 | pipe_buf_release(pipe, buf); |
0ddad21d | 308 | spin_lock_irq(&pipe->rd_wait.lock); |
e7d553d6 DH |
309 | #ifdef CONFIG_WATCH_QUEUE |
310 | if (buf->flags & PIPE_BUF_FLAG_LOSS) | |
311 | pipe->note_loss = true; | |
312 | #endif | |
8cefc107 DH |
313 | tail++; |
314 | pipe->tail = tail; | |
0ddad21d | 315 | spin_unlock_irq(&pipe->rd_wait.lock); |
1da177e4 LT |
316 | } |
317 | total_len -= chars; | |
318 | if (!total_len) | |
319 | break; /* common path: read succeeded */ | |
8cefc107 DH |
320 | if (!pipe_empty(head, tail)) /* More to do? */ |
321 | continue; | |
1da177e4 | 322 | } |
8cefc107 | 323 | |
923f4f23 | 324 | if (!pipe->writers) |
1da177e4 | 325 | break; |
a28c8b9d LT |
326 | if (ret) |
327 | break; | |
328 | if (filp->f_flags & O_NONBLOCK) { | |
329 | ret = -EAGAIN; | |
330 | break; | |
1da177e4 | 331 | } |
85190d15 | 332 | __pipe_unlock(pipe); |
d1c6a2aa LT |
333 | |
334 | /* | |
335 | * We only get here if we didn't actually read anything. | |
336 | * | |
337 | * However, we could have seen (and removed) a zero-sized | |
338 | * pipe buffer, and might have made space in the buffers | |
339 | * that way. | |
340 | * | |
341 | * You can't make zero-sized pipe buffers by doing an empty | |
342 | * write (not even in packet mode), but they can happen if | |
343 | * the writer gets an EFAULT when trying to fill a buffer | |
344 | * that already got allocated and inserted in the buffer | |
345 | * array. | |
346 | * | |
347 | * So we still need to wake up any pending writers in the | |
348 | * _very_ unlikely case that the pipe was full, but we got | |
349 | * no data. | |
350 | */ | |
351 | if (unlikely(was_full)) { | |
0ddad21d | 352 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
f467a6a6 LT |
353 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
354 | } | |
d1c6a2aa LT |
355 | |
356 | /* | |
357 | * But because we didn't read anything, at this point we can | |
358 | * just return directly with -ERESTARTSYS if we're interrupted, | |
359 | * since we've done any required wakeups and there's no need | |
360 | * to mark anything accessed. And we've dropped the lock. | |
361 | */ | |
0ddad21d | 362 | if (wait_event_interruptible_exclusive(pipe->rd_wait, pipe_readable(pipe)) < 0) |
d1c6a2aa LT |
363 | return -ERESTARTSYS; |
364 | ||
85190d15 | 365 | __pipe_lock(pipe); |
f467a6a6 | 366 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); |
0ddad21d | 367 | wake_next_reader = true; |
1da177e4 | 368 | } |
0ddad21d LT |
369 | if (pipe_empty(pipe->head, pipe->tail)) |
370 | wake_next_reader = false; | |
ebec73f4 | 371 | __pipe_unlock(pipe); |
341b446b | 372 | |
f467a6a6 | 373 | if (was_full) { |
0ddad21d | 374 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
923f4f23 | 375 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 | 376 | } |
0ddad21d LT |
377 | if (wake_next_reader) |
378 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); | |
1da177e4 LT |
379 | if (ret > 0) |
380 | file_accessed(filp); | |
381 | return ret; | |
382 | } | |
383 | ||
9883035a LT |
384 | static inline int is_packetized(struct file *file) |
385 | { | |
386 | return (file->f_flags & O_DIRECT) != 0; | |
387 | } | |
388 | ||
85190d15 LT |
389 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
390 | static inline bool pipe_writable(const struct pipe_inode_info *pipe) | |
391 | { | |
392 | unsigned int head = READ_ONCE(pipe->head); | |
393 | unsigned int tail = READ_ONCE(pipe->tail); | |
394 | unsigned int max_usage = READ_ONCE(pipe->max_usage); | |
395 | ||
396 | return !pipe_full(head, tail, max_usage) || | |
397 | !READ_ONCE(pipe->readers); | |
398 | } | |
399 | ||
1da177e4 | 400 | static ssize_t |
f0d1bec9 | 401 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 402 | { |
ee0b3e67 | 403 | struct file *filp = iocb->ki_filp; |
de32ec4c | 404 | struct pipe_inode_info *pipe = filp->private_data; |
8f868d68 | 405 | unsigned int head; |
f0d1bec9 | 406 | ssize_t ret = 0; |
f0d1bec9 | 407 | size_t total_len = iov_iter_count(from); |
1da177e4 | 408 | ssize_t chars; |
1b6b26ae | 409 | bool was_empty = false; |
0ddad21d | 410 | bool wake_next_writer = false; |
1da177e4 | 411 | |
1da177e4 LT |
412 | /* Null write succeeds. */ |
413 | if (unlikely(total_len == 0)) | |
414 | return 0; | |
415 | ||
ebec73f4 | 416 | __pipe_lock(pipe); |
1da177e4 | 417 | |
923f4f23 | 418 | if (!pipe->readers) { |
1da177e4 LT |
419 | send_sig(SIGPIPE, current, 0); |
420 | ret = -EPIPE; | |
421 | goto out; | |
422 | } | |
423 | ||
c73be61c DH |
424 | #ifdef CONFIG_WATCH_QUEUE |
425 | if (pipe->watch_queue) { | |
426 | ret = -EXDEV; | |
427 | goto out; | |
428 | } | |
429 | #endif | |
430 | ||
1b6b26ae | 431 | /* |
3a34b13a LT |
432 | * Epoll nonsensically wants a wakeup whether the pipe |
433 | * was already empty or not. | |
1b6b26ae LT |
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 | |
3a34b13a | 439 | * page-aligns the rest of the writes for large writes |
1b6b26ae LT |
440 | * spanning multiple pages. |
441 | */ | |
8cefc107 | 442 | head = pipe->head; |
3a34b13a | 443 | was_empty = true; |
1b6b26ae | 444 | chars = total_len & (PAGE_SIZE-1); |
3a34b13a | 445 | if (chars && !pipe_empty(head, pipe->tail)) { |
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 | |
f6dd9755 CH |
450 | if ((buf->flags & PIPE_BUF_FLAG_CAN_MERGE) && |
451 | offset + chars <= PAGE_SIZE) { | |
fba597db | 452 | ret = pipe_buf_confirm(pipe, buf); |
6ae08069 | 453 | if (ret) |
5274f052 | 454 | goto out; |
f84d7519 | 455 | |
f0d1bec9 AV |
456 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
457 | if (unlikely(ret < chars)) { | |
6ae08069 | 458 | ret = -EFAULT; |
1da177e4 | 459 | goto out; |
f6762b7a | 460 | } |
1b6b26ae | 461 | |
6ae08069 | 462 | buf->len += ret; |
f0d1bec9 | 463 | if (!iov_iter_count(from)) |
1da177e4 LT |
464 | goto out; |
465 | } | |
466 | } | |
467 | ||
468 | for (;;) { | |
923f4f23 | 469 | if (!pipe->readers) { |
1da177e4 | 470 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
471 | if (!ret) |
472 | ret = -EPIPE; | |
1da177e4 LT |
473 | break; |
474 | } | |
8cefc107 | 475 | |
a194dfe6 | 476 | head = pipe->head; |
8f868d68 DH |
477 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) { |
478 | unsigned int mask = pipe->ring_size - 1; | |
8cefc107 | 479 | struct pipe_buffer *buf = &pipe->bufs[head & mask]; |
923f4f23 | 480 | struct page *page = pipe->tmp_page; |
f0d1bec9 | 481 | int copied; |
1da177e4 LT |
482 | |
483 | if (!page) { | |
d86133bd | 484 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
485 | if (unlikely(!page)) { |
486 | ret = ret ? : -ENOMEM; | |
487 | break; | |
488 | } | |
923f4f23 | 489 | pipe->tmp_page = page; |
1da177e4 | 490 | } |
a194dfe6 DH |
491 | |
492 | /* Allocate a slot in the ring in advance and attach an | |
493 | * empty buffer. If we fault or otherwise fail to use | |
494 | * it, either the reader will consume it or it'll still | |
495 | * be there for the next write. | |
496 | */ | |
0ddad21d | 497 | spin_lock_irq(&pipe->rd_wait.lock); |
a194dfe6 DH |
498 | |
499 | head = pipe->head; | |
8f868d68 | 500 | if (pipe_full(head, pipe->tail, pipe->max_usage)) { |
0ddad21d | 501 | spin_unlock_irq(&pipe->rd_wait.lock); |
8df44129 DH |
502 | continue; |
503 | } | |
504 | ||
a194dfe6 | 505 | pipe->head = head + 1; |
0ddad21d | 506 | spin_unlock_irq(&pipe->rd_wait.lock); |
1da177e4 LT |
507 | |
508 | /* Insert it into the buffer array */ | |
a194dfe6 | 509 | buf = &pipe->bufs[head & mask]; |
1da177e4 LT |
510 | buf->page = page; |
511 | buf->ops = &anon_pipe_buf_ops; | |
512 | buf->offset = 0; | |
a194dfe6 | 513 | buf->len = 0; |
f6dd9755 | 514 | if (is_packetized(filp)) |
9883035a | 515 | buf->flags = PIPE_BUF_FLAG_PACKET; |
f6dd9755 CH |
516 | else |
517 | buf->flags = PIPE_BUF_FLAG_CAN_MERGE; | |
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 | 556 | if (was_empty) { |
0ddad21d | 557 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
1b6b26ae LT |
558 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
559 | } | |
0ddad21d | 560 | wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); |
85190d15 | 561 | __pipe_lock(pipe); |
0dd1e377 | 562 | was_empty = pipe_empty(pipe->head, pipe->tail); |
0ddad21d | 563 | wake_next_writer = true; |
1da177e4 LT |
564 | } |
565 | out: | |
0ddad21d LT |
566 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) |
567 | wake_next_writer = false; | |
ebec73f4 | 568 | __pipe_unlock(pipe); |
1b6b26ae LT |
569 | |
570 | /* | |
571 | * If we do do a wakeup event, we do a 'sync' wakeup, because we | |
572 | * want the reader to start processing things asap, rather than | |
573 | * leave the data pending. | |
574 | * | |
575 | * This is particularly important for small writes, because of | |
576 | * how (for example) the GNU make jobserver uses small writes to | |
577 | * wake up pending jobs | |
578 | */ | |
579 | if (was_empty) { | |
0ddad21d | 580 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
923f4f23 | 581 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 | 582 | } |
0ddad21d LT |
583 | if (wake_next_writer) |
584 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); | |
7e775f46 | 585 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
586 | int err = file_update_time(filp); |
587 | if (err) | |
588 | ret = err; | |
7e775f46 | 589 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 590 | } |
1da177e4 LT |
591 | return ret; |
592 | } | |
593 | ||
d59d0b1b | 594 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 595 | { |
de32ec4c | 596 | struct pipe_inode_info *pipe = filp->private_data; |
8cefc107 | 597 | int count, head, tail, mask; |
1da177e4 LT |
598 | |
599 | switch (cmd) { | |
c73be61c DH |
600 | case FIONREAD: |
601 | __pipe_lock(pipe); | |
602 | count = 0; | |
603 | head = pipe->head; | |
604 | tail = pipe->tail; | |
605 | mask = pipe->ring_size - 1; | |
8cefc107 | 606 | |
c73be61c DH |
607 | while (tail != head) { |
608 | count += pipe->bufs[tail & mask].len; | |
609 | tail++; | |
610 | } | |
611 | __pipe_unlock(pipe); | |
923f4f23 | 612 | |
c73be61c | 613 | return put_user(count, (int __user *)arg); |
923f4f23 | 614 | |
c73be61c DH |
615 | #ifdef CONFIG_WATCH_QUEUE |
616 | case IOC_WATCH_QUEUE_SET_SIZE: { | |
617 | int ret; | |
618 | __pipe_lock(pipe); | |
619 | ret = watch_queue_set_size(pipe, arg); | |
620 | __pipe_unlock(pipe); | |
621 | return ret; | |
622 | } | |
623 | ||
624 | case IOC_WATCH_QUEUE_SET_FILTER: | |
625 | return watch_queue_set_filter( | |
626 | pipe, (struct watch_notification_filter __user *)arg); | |
627 | #endif | |
628 | ||
629 | default: | |
630 | return -ENOIOCTLCMD; | |
1da177e4 LT |
631 | } |
632 | } | |
633 | ||
dd67081b | 634 | /* No kernel lock held - fine */ |
a11e1d43 LT |
635 | static __poll_t |
636 | pipe_poll(struct file *filp, poll_table *wait) | |
dd67081b | 637 | { |
a11e1d43 | 638 | __poll_t mask; |
dd67081b | 639 | struct pipe_inode_info *pipe = filp->private_data; |
ad910e36 | 640 | unsigned int head, tail; |
a11e1d43 | 641 | |
ad910e36 | 642 | /* |
0ddad21d | 643 | * Reading pipe state only -- no need for acquiring the semaphore. |
ad910e36 LT |
644 | * |
645 | * But because this is racy, the code has to add the | |
646 | * entry to the poll table _first_ .. | |
647 | */ | |
0ddad21d LT |
648 | if (filp->f_mode & FMODE_READ) |
649 | poll_wait(filp, &pipe->rd_wait, wait); | |
650 | if (filp->f_mode & FMODE_WRITE) | |
651 | poll_wait(filp, &pipe->wr_wait, wait); | |
1da177e4 | 652 | |
ad910e36 LT |
653 | /* |
654 | * .. and only then can you do the racy tests. That way, | |
655 | * if something changes and you got it wrong, the poll | |
656 | * table entry will wake you up and fix it. | |
657 | */ | |
658 | head = READ_ONCE(pipe->head); | |
659 | tail = READ_ONCE(pipe->tail); | |
660 | ||
a11e1d43 | 661 | mask = 0; |
1da177e4 | 662 | if (filp->f_mode & FMODE_READ) { |
8cefc107 DH |
663 | if (!pipe_empty(head, tail)) |
664 | mask |= EPOLLIN | EPOLLRDNORM; | |
923f4f23 | 665 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
a9a08845 | 666 | mask |= EPOLLHUP; |
1da177e4 LT |
667 | } |
668 | ||
669 | if (filp->f_mode & FMODE_WRITE) { | |
6718b6f8 | 670 | if (!pipe_full(head, tail, pipe->max_usage)) |
8cefc107 | 671 | mask |= EPOLLOUT | EPOLLWRNORM; |
5e5d7a22 | 672 | /* |
a9a08845 | 673 | * Most Unices do not set EPOLLERR for FIFOs but on Linux they |
5e5d7a22 PE |
674 | * behave exactly like pipes for poll(). |
675 | */ | |
923f4f23 | 676 | if (!pipe->readers) |
a9a08845 | 677 | mask |= EPOLLERR; |
1da177e4 LT |
678 | } |
679 | ||
680 | return mask; | |
681 | } | |
682 | ||
b0d8d229 LT |
683 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
684 | { | |
685 | int kill = 0; | |
686 | ||
687 | spin_lock(&inode->i_lock); | |
688 | if (!--pipe->files) { | |
689 | inode->i_pipe = NULL; | |
690 | kill = 1; | |
691 | } | |
692 | spin_unlock(&inode->i_lock); | |
693 | ||
694 | if (kill) | |
695 | free_pipe_info(pipe); | |
696 | } | |
697 | ||
1da177e4 | 698 | static int |
599a0ac1 | 699 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 700 | { |
b0d8d229 | 701 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 702 | |
ebec73f4 | 703 | __pipe_lock(pipe); |
599a0ac1 AV |
704 | if (file->f_mode & FMODE_READ) |
705 | pipe->readers--; | |
706 | if (file->f_mode & FMODE_WRITE) | |
707 | pipe->writers--; | |
341b446b | 708 | |
6551d5c5 LT |
709 | /* Was that the last reader or writer, but not the other side? */ |
710 | if (!pipe->readers != !pipe->writers) { | |
711 | wake_up_interruptible_all(&pipe->rd_wait); | |
712 | wake_up_interruptible_all(&pipe->wr_wait); | |
923f4f23 IM |
713 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
714 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 715 | } |
ebec73f4 | 716 | __pipe_unlock(pipe); |
ba5bb147 | 717 | |
b0d8d229 | 718 | put_pipe_info(inode, pipe); |
1da177e4 LT |
719 | return 0; |
720 | } | |
721 | ||
722 | static int | |
599a0ac1 | 723 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 724 | { |
de32ec4c | 725 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 726 | int retval = 0; |
1da177e4 | 727 | |
ebec73f4 | 728 | __pipe_lock(pipe); |
599a0ac1 AV |
729 | if (filp->f_mode & FMODE_READ) |
730 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
731 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 732 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
733 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
734 | /* this can happen only if on == T */ | |
e5bc49ba ON |
735 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
736 | } | |
ebec73f4 | 737 | __pipe_unlock(pipe); |
60aa4924 | 738 | return retval; |
1da177e4 LT |
739 | } |
740 | ||
c73be61c DH |
741 | unsigned long account_pipe_buffers(struct user_struct *user, |
742 | unsigned long old, unsigned long new) | |
759c0114 | 743 | { |
9c87bcf0 | 744 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
745 | } |
746 | ||
c73be61c | 747 | bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 748 | { |
f7340761 EB |
749 | unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft); |
750 | ||
751 | return soft_limit && user_bufs > soft_limit; | |
759c0114 WT |
752 | } |
753 | ||
c73be61c | 754 | bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 755 | { |
f7340761 EB |
756 | unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard); |
757 | ||
758 | return hard_limit && user_bufs > hard_limit; | |
759c0114 WT |
759 | } |
760 | ||
c73be61c | 761 | bool pipe_is_unprivileged_user(void) |
85c2dd54 EB |
762 | { |
763 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
764 | } | |
765 | ||
7bee130e | 766 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 767 | { |
923f4f23 | 768 | struct pipe_inode_info *pipe; |
09b4d199 MK |
769 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
770 | struct user_struct *user = get_current_user(); | |
9c87bcf0 | 771 | unsigned long user_bufs; |
f7340761 | 772 | unsigned int max_size = READ_ONCE(pipe_max_size); |
3a326a2c | 773 | |
d86133bd | 774 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
09b4d199 MK |
775 | if (pipe == NULL) |
776 | goto out_free_uid; | |
777 | ||
f7340761 EB |
778 | if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE)) |
779 | pipe_bufs = max_size >> PAGE_SHIFT; | |
086e774a | 780 | |
9c87bcf0 | 781 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
a005ca0e | 782 | |
c73be61c | 783 | if (too_many_pipe_buffers_soft(user_bufs) && pipe_is_unprivileged_user()) { |
9c87bcf0 | 784 | user_bufs = account_pipe_buffers(user, pipe_bufs, 1); |
a005ca0e | 785 | pipe_bufs = 1; |
09b4d199 | 786 | } |
759c0114 | 787 | |
c73be61c | 788 | if (too_many_pipe_buffers_hard(user_bufs) && pipe_is_unprivileged_user()) |
a005ca0e MK |
789 | goto out_revert_acct; |
790 | ||
791 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), | |
792 | GFP_KERNEL_ACCOUNT); | |
793 | ||
09b4d199 | 794 | if (pipe->bufs) { |
0ddad21d LT |
795 | init_waitqueue_head(&pipe->rd_wait); |
796 | init_waitqueue_head(&pipe->wr_wait); | |
09b4d199 | 797 | pipe->r_counter = pipe->w_counter = 1; |
6718b6f8 | 798 | pipe->max_usage = pipe_bufs; |
8cefc107 | 799 | pipe->ring_size = pipe_bufs; |
c73be61c | 800 | pipe->nr_accounted = pipe_bufs; |
09b4d199 | 801 | pipe->user = user; |
09b4d199 MK |
802 | mutex_init(&pipe->mutex); |
803 | return pipe; | |
3a326a2c IM |
804 | } |
805 | ||
a005ca0e | 806 | out_revert_acct: |
9c87bcf0 | 807 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
09b4d199 MK |
808 | kfree(pipe); |
809 | out_free_uid: | |
810 | free_uid(user); | |
35f3d14d | 811 | return NULL; |
3a326a2c IM |
812 | } |
813 | ||
4b8a8f1e | 814 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
815 | { |
816 | int i; | |
1da177e4 | 817 | |
c73be61c DH |
818 | #ifdef CONFIG_WATCH_QUEUE |
819 | if (pipe->watch_queue) { | |
820 | watch_queue_clear(pipe->watch_queue); | |
821 | put_watch_queue(pipe->watch_queue); | |
822 | } | |
823 | #endif | |
824 | ||
825 | (void) account_pipe_buffers(pipe->user, pipe->nr_accounted, 0); | |
759c0114 | 826 | free_uid(pipe->user); |
8cefc107 | 827 | for (i = 0; i < pipe->ring_size; i++) { |
923f4f23 | 828 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 829 | if (buf->ops) |
a779638c | 830 | pipe_buf_release(pipe, buf); |
1da177e4 | 831 | } |
923f4f23 IM |
832 | if (pipe->tmp_page) |
833 | __free_page(pipe->tmp_page); | |
35f3d14d | 834 | kfree(pipe->bufs); |
923f4f23 | 835 | kfree(pipe); |
1da177e4 LT |
836 | } |
837 | ||
fa3536cc | 838 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 839 | |
c23fbb6b ED |
840 | /* |
841 | * pipefs_dname() is called from d_path(). | |
842 | */ | |
843 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
844 | { | |
845 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
75c3cfa8 | 846 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
847 | } |
848 | ||
3ba13d17 | 849 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 850 | .d_dname = pipefs_dname, |
1da177e4 LT |
851 | }; |
852 | ||
853 | static struct inode * get_pipe_inode(void) | |
854 | { | |
a209dfc7 | 855 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 856 | struct pipe_inode_info *pipe; |
1da177e4 LT |
857 | |
858 | if (!inode) | |
859 | goto fail_inode; | |
860 | ||
85fe4025 CH |
861 | inode->i_ino = get_next_ino(); |
862 | ||
7bee130e | 863 | pipe = alloc_pipe_info(); |
923f4f23 | 864 | if (!pipe) |
1da177e4 | 865 | goto fail_iput; |
3a326a2c | 866 | |
ba5bb147 AV |
867 | inode->i_pipe = pipe; |
868 | pipe->files = 2; | |
923f4f23 | 869 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 870 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
871 | |
872 | /* | |
873 | * Mark the inode dirty from the very beginning, | |
874 | * that way it will never be moved to the dirty | |
875 | * list because "mark_inode_dirty()" will think | |
876 | * that it already _is_ on the dirty list. | |
877 | */ | |
878 | inode->i_state = I_DIRTY; | |
879 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
880 | inode->i_uid = current_fsuid(); |
881 | inode->i_gid = current_fsgid(); | |
078cd827 | 882 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
923f4f23 | 883 | |
1da177e4 LT |
884 | return inode; |
885 | ||
886 | fail_iput: | |
887 | iput(inode); | |
341b446b | 888 | |
1da177e4 LT |
889 | fail_inode: |
890 | return NULL; | |
891 | } | |
892 | ||
e4fad8e5 | 893 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 894 | { |
e4fad8e5 | 895 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 896 | struct file *f; |
8a018eb5 | 897 | int error; |
1da177e4 | 898 | |
1da177e4 | 899 | if (!inode) |
e4fad8e5 | 900 | return -ENFILE; |
1da177e4 | 901 | |
c73be61c | 902 | if (flags & O_NOTIFICATION_PIPE) { |
8a018eb5 QC |
903 | error = watch_queue_init(inode->i_pipe); |
904 | if (error) { | |
905 | free_pipe_info(inode->i_pipe); | |
c73be61c | 906 | iput(inode); |
8a018eb5 | 907 | return error; |
c73be61c | 908 | } |
c73be61c DH |
909 | } |
910 | ||
152b6372 AV |
911 | f = alloc_file_pseudo(inode, pipe_mnt, "", |
912 | O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)), | |
913 | &pipefifo_fops); | |
e9bb1f9b | 914 | if (IS_ERR(f)) { |
152b6372 AV |
915 | free_pipe_info(inode->i_pipe); |
916 | iput(inode); | |
917 | return PTR_ERR(f); | |
e9bb1f9b | 918 | } |
341b446b | 919 | |
de32ec4c | 920 | f->private_data = inode->i_pipe; |
d6cbd281 | 921 | |
183266f2 AV |
922 | res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK), |
923 | &pipefifo_fops); | |
e9bb1f9b | 924 | if (IS_ERR(res[0])) { |
b10a4a9f AV |
925 | put_pipe_info(inode, inode->i_pipe); |
926 | fput(f); | |
927 | return PTR_ERR(res[0]); | |
e9bb1f9b | 928 | } |
de32ec4c | 929 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 | 930 | res[1] = f; |
d8e464ec LT |
931 | stream_open(inode, res[0]); |
932 | stream_open(inode, res[1]); | |
e4fad8e5 | 933 | return 0; |
d6cbd281 AK |
934 | } |
935 | ||
5b249b1b | 936 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 937 | { |
d6cbd281 AK |
938 | int error; |
939 | int fdw, fdr; | |
940 | ||
c73be61c | 941 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT | O_NOTIFICATION_PIPE)) |
ed8cae8b UD |
942 | return -EINVAL; |
943 | ||
e4fad8e5 AV |
944 | error = create_pipe_files(files, flags); |
945 | if (error) | |
946 | return error; | |
d6cbd281 | 947 | |
ed8cae8b | 948 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
949 | if (error < 0) |
950 | goto err_read_pipe; | |
951 | fdr = error; | |
952 | ||
ed8cae8b | 953 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
954 | if (error < 0) |
955 | goto err_fdr; | |
956 | fdw = error; | |
957 | ||
157cf649 | 958 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
959 | fd[0] = fdr; |
960 | fd[1] = fdw; | |
d6cbd281 AK |
961 | return 0; |
962 | ||
963 | err_fdr: | |
964 | put_unused_fd(fdr); | |
965 | err_read_pipe: | |
e4fad8e5 AV |
966 | fput(files[0]); |
967 | fput(files[1]); | |
d6cbd281 | 968 | return error; |
1da177e4 LT |
969 | } |
970 | ||
5b249b1b AV |
971 | int do_pipe_flags(int *fd, int flags) |
972 | { | |
973 | struct file *files[2]; | |
974 | int error = __do_pipe_flags(fd, files, flags); | |
975 | if (!error) { | |
976 | fd_install(fd[0], files[0]); | |
977 | fd_install(fd[1], files[1]); | |
978 | } | |
979 | return error; | |
980 | } | |
981 | ||
d35c7b0e UD |
982 | /* |
983 | * sys_pipe() is the normal C calling standard for creating | |
984 | * a pipe. It's not the way Unix traditionally does this, though. | |
985 | */ | |
0a216dd1 | 986 | static int do_pipe2(int __user *fildes, int flags) |
d35c7b0e | 987 | { |
5b249b1b | 988 | struct file *files[2]; |
d35c7b0e UD |
989 | int fd[2]; |
990 | int error; | |
991 | ||
5b249b1b | 992 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 993 | if (!error) { |
5b249b1b AV |
994 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
995 | fput(files[0]); | |
996 | fput(files[1]); | |
997 | put_unused_fd(fd[0]); | |
998 | put_unused_fd(fd[1]); | |
d35c7b0e | 999 | error = -EFAULT; |
5b249b1b AV |
1000 | } else { |
1001 | fd_install(fd[0], files[0]); | |
1002 | fd_install(fd[1], files[1]); | |
ba719bae | 1003 | } |
d35c7b0e UD |
1004 | } |
1005 | return error; | |
1006 | } | |
1007 | ||
0a216dd1 DB |
1008 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
1009 | { | |
1010 | return do_pipe2(fildes, flags); | |
1011 | } | |
1012 | ||
2b664219 | 1013 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b | 1014 | { |
0a216dd1 | 1015 | return do_pipe2(fildes, 0); |
ed8cae8b UD |
1016 | } |
1017 | ||
472e5b05 LT |
1018 | /* |
1019 | * This is the stupid "wait for pipe to be readable or writable" | |
1020 | * model. | |
1021 | * | |
1022 | * See pipe_read/write() for the proper kind of exclusive wait, | |
1023 | * but that requires that we wake up any other readers/writers | |
1024 | * if we then do not end up reading everything (ie the whole | |
1025 | * "wake_next_reader/writer" logic in pipe_read/write()). | |
1026 | */ | |
1027 | void pipe_wait_readable(struct pipe_inode_info *pipe) | |
1028 | { | |
1029 | pipe_unlock(pipe); | |
1030 | wait_event_interruptible(pipe->rd_wait, pipe_readable(pipe)); | |
1031 | pipe_lock(pipe); | |
1032 | } | |
1033 | ||
1034 | void pipe_wait_writable(struct pipe_inode_info *pipe) | |
1035 | { | |
1036 | pipe_unlock(pipe); | |
1037 | wait_event_interruptible(pipe->wr_wait, pipe_writable(pipe)); | |
1038 | pipe_lock(pipe); | |
1039 | } | |
1040 | ||
1041 | /* | |
1042 | * This depends on both the wait (here) and the wakeup (wake_up_partner) | |
1043 | * holding the pipe lock, so "*cnt" is stable and we know a wakeup cannot | |
1044 | * race with the count check and waitqueue prep. | |
1045 | * | |
1046 | * Normally in order to avoid races, you'd do the prepare_to_wait() first, | |
1047 | * then check the condition you're waiting for, and only then sleep. But | |
1048 | * because of the pipe lock, we can check the condition before being on | |
1049 | * the wait queue. | |
1050 | * | |
1051 | * We use the 'rd_wait' waitqueue for pipe partner waiting. | |
1052 | */ | |
fc7478a2 | 1053 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 | 1054 | { |
472e5b05 | 1055 | DEFINE_WAIT(rdwait); |
8cefc107 | 1056 | int cur = *cnt; |
f776c738 AV |
1057 | |
1058 | while (cur == *cnt) { | |
472e5b05 LT |
1059 | prepare_to_wait(&pipe->rd_wait, &rdwait, TASK_INTERRUPTIBLE); |
1060 | pipe_unlock(pipe); | |
1061 | schedule(); | |
1062 | finish_wait(&pipe->rd_wait, &rdwait); | |
1063 | pipe_lock(pipe); | |
f776c738 AV |
1064 | if (signal_pending(current)) |
1065 | break; | |
1066 | } | |
1067 | return cur == *cnt ? -ERESTARTSYS : 0; | |
1068 | } | |
1069 | ||
fc7478a2 | 1070 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 1071 | { |
6551d5c5 | 1072 | wake_up_interruptible_all(&pipe->rd_wait); |
f776c738 AV |
1073 | } |
1074 | ||
1075 | static int fifo_open(struct inode *inode, struct file *filp) | |
1076 | { | |
1077 | struct pipe_inode_info *pipe; | |
599a0ac1 | 1078 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
1079 | int ret; |
1080 | ||
ba5bb147 AV |
1081 | filp->f_version = 0; |
1082 | ||
1083 | spin_lock(&inode->i_lock); | |
1084 | if (inode->i_pipe) { | |
1085 | pipe = inode->i_pipe; | |
1086 | pipe->files++; | |
1087 | spin_unlock(&inode->i_lock); | |
1088 | } else { | |
1089 | spin_unlock(&inode->i_lock); | |
7bee130e | 1090 | pipe = alloc_pipe_info(); |
f776c738 | 1091 | if (!pipe) |
ba5bb147 AV |
1092 | return -ENOMEM; |
1093 | pipe->files = 1; | |
1094 | spin_lock(&inode->i_lock); | |
1095 | if (unlikely(inode->i_pipe)) { | |
1096 | inode->i_pipe->files++; | |
1097 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 1098 | free_pipe_info(pipe); |
ba5bb147 AV |
1099 | pipe = inode->i_pipe; |
1100 | } else { | |
1101 | inode->i_pipe = pipe; | |
1102 | spin_unlock(&inode->i_lock); | |
1103 | } | |
f776c738 | 1104 | } |
de32ec4c | 1105 | filp->private_data = pipe; |
ba5bb147 AV |
1106 | /* OK, we have a pipe and it's pinned down */ |
1107 | ||
ebec73f4 | 1108 | __pipe_lock(pipe); |
f776c738 AV |
1109 | |
1110 | /* We can only do regular read/write on fifos */ | |
d8e464ec | 1111 | stream_open(inode, filp); |
f776c738 | 1112 | |
d8e464ec | 1113 | switch (filp->f_mode & (FMODE_READ | FMODE_WRITE)) { |
f776c738 AV |
1114 | case FMODE_READ: |
1115 | /* | |
1116 | * O_RDONLY | |
1117 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
1118 | * opened, even when there is no process writing the FIFO. | |
1119 | */ | |
f776c738 AV |
1120 | pipe->r_counter++; |
1121 | if (pipe->readers++ == 0) | |
fc7478a2 | 1122 | wake_up_partner(pipe); |
f776c738 | 1123 | |
599a0ac1 | 1124 | if (!is_pipe && !pipe->writers) { |
f776c738 | 1125 | if ((filp->f_flags & O_NONBLOCK)) { |
a9a08845 | 1126 | /* suppress EPOLLHUP until we have |
f776c738 AV |
1127 | * seen a writer */ |
1128 | filp->f_version = pipe->w_counter; | |
1129 | } else { | |
fc7478a2 | 1130 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
1131 | goto err_rd; |
1132 | } | |
1133 | } | |
1134 | break; | |
8cefc107 | 1135 | |
f776c738 AV |
1136 | case FMODE_WRITE: |
1137 | /* | |
1138 | * O_WRONLY | |
1139 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
1140 | * errno=ENXIO when there is no process reading the FIFO. | |
1141 | */ | |
1142 | ret = -ENXIO; | |
599a0ac1 | 1143 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
1144 | goto err; |
1145 | ||
f776c738 AV |
1146 | pipe->w_counter++; |
1147 | if (!pipe->writers++) | |
fc7478a2 | 1148 | wake_up_partner(pipe); |
f776c738 | 1149 | |
599a0ac1 | 1150 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 1151 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
1152 | goto err_wr; |
1153 | } | |
1154 | break; | |
8cefc107 | 1155 | |
f776c738 AV |
1156 | case FMODE_READ | FMODE_WRITE: |
1157 | /* | |
1158 | * O_RDWR | |
1159 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
1160 | * This implementation will NEVER block on a O_RDWR open, since | |
1161 | * the process can at least talk to itself. | |
1162 | */ | |
f776c738 AV |
1163 | |
1164 | pipe->readers++; | |
1165 | pipe->writers++; | |
1166 | pipe->r_counter++; | |
1167 | pipe->w_counter++; | |
1168 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 1169 | wake_up_partner(pipe); |
f776c738 AV |
1170 | break; |
1171 | ||
1172 | default: | |
1173 | ret = -EINVAL; | |
1174 | goto err; | |
1175 | } | |
1176 | ||
1177 | /* Ok! */ | |
ebec73f4 | 1178 | __pipe_unlock(pipe); |
f776c738 AV |
1179 | return 0; |
1180 | ||
1181 | err_rd: | |
1182 | if (!--pipe->readers) | |
0ddad21d | 1183 | wake_up_interruptible(&pipe->wr_wait); |
f776c738 AV |
1184 | ret = -ERESTARTSYS; |
1185 | goto err; | |
1186 | ||
1187 | err_wr: | |
1188 | if (!--pipe->writers) | |
6551d5c5 | 1189 | wake_up_interruptible_all(&pipe->rd_wait); |
f776c738 AV |
1190 | ret = -ERESTARTSYS; |
1191 | goto err; | |
1192 | ||
1193 | err: | |
ebec73f4 | 1194 | __pipe_unlock(pipe); |
b0d8d229 LT |
1195 | |
1196 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1197 | return ret; |
1198 | } | |
1199 | ||
599a0ac1 AV |
1200 | const struct file_operations pipefifo_fops = { |
1201 | .open = fifo_open, | |
1202 | .llseek = no_llseek, | |
fb9096a3 | 1203 | .read_iter = pipe_read, |
f0d1bec9 | 1204 | .write_iter = pipe_write, |
a11e1d43 | 1205 | .poll = pipe_poll, |
599a0ac1 AV |
1206 | .unlocked_ioctl = pipe_ioctl, |
1207 | .release = pipe_release, | |
1208 | .fasync = pipe_fasync, | |
f8ad8187 | 1209 | .splice_write = iter_file_splice_write, |
f776c738 AV |
1210 | }; |
1211 | ||
f491bd71 MK |
1212 | /* |
1213 | * Currently we rely on the pipe array holding a power-of-2 number | |
d3f14c48 | 1214 | * of pages. Returns 0 on error. |
f491bd71 | 1215 | */ |
96e99be4 | 1216 | unsigned int round_pipe_size(unsigned long size) |
f491bd71 | 1217 | { |
c4fed5a9 | 1218 | if (size > (1U << 31)) |
96e99be4 EB |
1219 | return 0; |
1220 | ||
4c2e4bef EB |
1221 | /* Minimum pipe size, as required by POSIX */ |
1222 | if (size < PAGE_SIZE) | |
c4fed5a9 | 1223 | return PAGE_SIZE; |
d3f14c48 | 1224 | |
c4fed5a9 | 1225 | return roundup_pow_of_two(size); |
f491bd71 MK |
1226 | } |
1227 | ||
35f3d14d | 1228 | /* |
c73be61c | 1229 | * Resize the pipe ring to a number of slots. |
35f3d14d | 1230 | */ |
c73be61c | 1231 | int pipe_resize_ring(struct pipe_inode_info *pipe, unsigned int nr_slots) |
35f3d14d JA |
1232 | { |
1233 | struct pipe_buffer *bufs; | |
c73be61c | 1234 | unsigned int head, tail, mask, n; |
35f3d14d | 1235 | |
35f3d14d | 1236 | /* |
8cefc107 DH |
1237 | * We can shrink the pipe, if arg is greater than the ring occupancy. |
1238 | * Since we don't expect a lot of shrink+grow operations, just free and | |
1239 | * allocate again like we would do for growing. If the pipe currently | |
35f3d14d JA |
1240 | * contains more buffers than arg, then return busy. |
1241 | */ | |
8cefc107 DH |
1242 | mask = pipe->ring_size - 1; |
1243 | head = pipe->head; | |
1244 | tail = pipe->tail; | |
1245 | n = pipe_occupancy(pipe->head, pipe->tail); | |
c73be61c DH |
1246 | if (nr_slots < n) |
1247 | return -EBUSY; | |
35f3d14d | 1248 | |
8cefc107 | 1249 | bufs = kcalloc(nr_slots, sizeof(*bufs), |
d86133bd | 1250 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); |
c73be61c DH |
1251 | if (unlikely(!bufs)) |
1252 | return -ENOMEM; | |
35f3d14d JA |
1253 | |
1254 | /* | |
1255 | * The pipe array wraps around, so just start the new one at zero | |
8cefc107 | 1256 | * and adjust the indices. |
35f3d14d | 1257 | */ |
8cefc107 DH |
1258 | if (n > 0) { |
1259 | unsigned int h = head & mask; | |
1260 | unsigned int t = tail & mask; | |
1261 | if (h > t) { | |
1262 | memcpy(bufs, pipe->bufs + t, | |
1263 | n * sizeof(struct pipe_buffer)); | |
1264 | } else { | |
1265 | unsigned int tsize = pipe->ring_size - t; | |
1266 | if (h > 0) | |
1267 | memcpy(bufs + tsize, pipe->bufs, | |
1268 | h * sizeof(struct pipe_buffer)); | |
1269 | memcpy(bufs, pipe->bufs + t, | |
1270 | tsize * sizeof(struct pipe_buffer)); | |
1271 | } | |
35f3d14d JA |
1272 | } |
1273 | ||
8cefc107 DH |
1274 | head = n; |
1275 | tail = 0; | |
1276 | ||
35f3d14d JA |
1277 | kfree(pipe->bufs); |
1278 | pipe->bufs = bufs; | |
8cefc107 | 1279 | pipe->ring_size = nr_slots; |
c73be61c DH |
1280 | if (pipe->max_usage > nr_slots) |
1281 | pipe->max_usage = nr_slots; | |
8cefc107 DH |
1282 | pipe->tail = tail; |
1283 | pipe->head = head; | |
6551d5c5 LT |
1284 | |
1285 | /* This might have made more room for writers */ | |
1286 | wake_up_interruptible(&pipe->wr_wait); | |
c73be61c DH |
1287 | return 0; |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1292 | * pipe size if successful, or return -ERROR on error. | |
1293 | */ | |
1294 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) | |
1295 | { | |
1296 | unsigned long user_bufs; | |
1297 | unsigned int nr_slots, size; | |
1298 | long ret = 0; | |
1299 | ||
1300 | #ifdef CONFIG_WATCH_QUEUE | |
1301 | if (pipe->watch_queue) | |
1302 | return -EBUSY; | |
1303 | #endif | |
1304 | ||
1305 | size = round_pipe_size(arg); | |
1306 | nr_slots = size >> PAGE_SHIFT; | |
1307 | ||
1308 | if (!nr_slots) | |
1309 | return -EINVAL; | |
1310 | ||
1311 | /* | |
1312 | * If trying to increase the pipe capacity, check that an | |
1313 | * unprivileged user is not trying to exceed various limits | |
1314 | * (soft limit check here, hard limit check just below). | |
1315 | * Decreasing the pipe capacity is always permitted, even | |
1316 | * if the user is currently over a limit. | |
1317 | */ | |
1318 | if (nr_slots > pipe->max_usage && | |
1319 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) | |
1320 | return -EPERM; | |
1321 | ||
1322 | user_bufs = account_pipe_buffers(pipe->user, pipe->nr_accounted, nr_slots); | |
1323 | ||
1324 | if (nr_slots > pipe->max_usage && | |
1325 | (too_many_pipe_buffers_hard(user_bufs) || | |
1326 | too_many_pipe_buffers_soft(user_bufs)) && | |
1327 | pipe_is_unprivileged_user()) { | |
1328 | ret = -EPERM; | |
1329 | goto out_revert_acct; | |
1330 | } | |
1331 | ||
1332 | ret = pipe_resize_ring(pipe, nr_slots); | |
1333 | if (ret < 0) | |
1334 | goto out_revert_acct; | |
1335 | ||
1336 | pipe->max_usage = nr_slots; | |
1337 | pipe->nr_accounted = nr_slots; | |
6718b6f8 | 1338 | return pipe->max_usage * PAGE_SIZE; |
b0b91d18 MK |
1339 | |
1340 | out_revert_acct: | |
c73be61c | 1341 | (void) account_pipe_buffers(pipe->user, nr_slots, pipe->nr_accounted); |
b0b91d18 | 1342 | return ret; |
35f3d14d JA |
1343 | } |
1344 | ||
72083646 | 1345 | /* |
4e7b5671 CH |
1346 | * Note that i_pipe and i_cdev share the same location, so checking ->i_pipe is |
1347 | * not enough to verify that this is a pipe. | |
72083646 | 1348 | */ |
c73be61c | 1349 | struct pipe_inode_info *get_pipe_info(struct file *file, bool for_splice) |
72083646 | 1350 | { |
c73be61c DH |
1351 | struct pipe_inode_info *pipe = file->private_data; |
1352 | ||
1353 | if (file->f_op != &pipefifo_fops || !pipe) | |
1354 | return NULL; | |
1355 | #ifdef CONFIG_WATCH_QUEUE | |
1356 | if (for_splice && pipe->watch_queue) | |
1357 | return NULL; | |
1358 | #endif | |
1359 | return pipe; | |
72083646 LT |
1360 | } |
1361 | ||
35f3d14d JA |
1362 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
1363 | { | |
1364 | struct pipe_inode_info *pipe; | |
1365 | long ret; | |
1366 | ||
c73be61c | 1367 | pipe = get_pipe_info(file, false); |
35f3d14d JA |
1368 | if (!pipe) |
1369 | return -EBADF; | |
1370 | ||
ebec73f4 | 1371 | __pipe_lock(pipe); |
35f3d14d JA |
1372 | |
1373 | switch (cmd) { | |
d37d4166 MK |
1374 | case F_SETPIPE_SZ: |
1375 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1376 | break; |
1377 | case F_GETPIPE_SZ: | |
6718b6f8 | 1378 | ret = pipe->max_usage * PAGE_SIZE; |
35f3d14d JA |
1379 | break; |
1380 | default: | |
1381 | ret = -EINVAL; | |
1382 | break; | |
1383 | } | |
1384 | ||
ebec73f4 | 1385 | __pipe_unlock(pipe); |
35f3d14d JA |
1386 | return ret; |
1387 | } | |
1388 | ||
ff0c7d15 NP |
1389 | static const struct super_operations pipefs_ops = { |
1390 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1391 | .statfs = simple_statfs, |
ff0c7d15 NP |
1392 | }; |
1393 | ||
1da177e4 LT |
1394 | /* |
1395 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1396 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1397 | * any operations on the root directory. However, we need a non-trivial | |
1398 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1399 | */ | |
4fa7ec5d DH |
1400 | |
1401 | static int pipefs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1402 | { |
4fa7ec5d DH |
1403 | struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC); |
1404 | if (!ctx) | |
1405 | return -ENOMEM; | |
1406 | ctx->ops = &pipefs_ops; | |
1407 | ctx->dops = &pipefs_dentry_operations; | |
1408 | return 0; | |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | static struct file_system_type pipe_fs_type = { | |
1412 | .name = "pipefs", | |
4fa7ec5d | 1413 | .init_fs_context = pipefs_init_fs_context, |
1da177e4 LT |
1414 | .kill_sb = kill_anon_super, |
1415 | }; | |
1416 | ||
1417 | static int __init init_pipe_fs(void) | |
1418 | { | |
1419 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1420 | |
1da177e4 LT |
1421 | if (!err) { |
1422 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1423 | if (IS_ERR(pipe_mnt)) { | |
1424 | err = PTR_ERR(pipe_mnt); | |
1425 | unregister_filesystem(&pipe_fs_type); | |
1426 | } | |
1427 | } | |
1428 | return err; | |
1429 | } | |
1430 | ||
1da177e4 | 1431 | fs_initcall(init_pipe_fs); |