Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/pipe.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1999 Linus Torvalds | |
6 | */ | |
7 | ||
8 | #include <linux/mm.h> | |
9 | #include <linux/file.h> | |
10 | #include <linux/poll.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/fs.h> | |
35f3d14d | 15 | #include <linux/log2.h> |
1da177e4 | 16 | #include <linux/mount.h> |
4fa7ec5d | 17 | #include <linux/pseudo_fs.h> |
b502bd11 | 18 | #include <linux/magic.h> |
1da177e4 LT |
19 | #include <linux/pipe_fs_i.h> |
20 | #include <linux/uio.h> | |
21 | #include <linux/highmem.h> | |
5274f052 | 22 | #include <linux/pagemap.h> |
db349509 | 23 | #include <linux/audit.h> |
ba719bae | 24 | #include <linux/syscalls.h> |
b492e95b | 25 | #include <linux/fcntl.h> |
d86133bd | 26 | #include <linux/memcontrol.h> |
1da177e4 | 27 | |
7c0f6ba6 | 28 | #include <linux/uaccess.h> |
1da177e4 LT |
29 | #include <asm/ioctls.h> |
30 | ||
599a0ac1 AV |
31 | #include "internal.h" |
32 | ||
b492e95b JA |
33 | /* |
34 | * The max size that a non-root user is allowed to grow the pipe. Can | |
ff9da691 | 35 | * be set by root in /proc/sys/fs/pipe-max-size |
b492e95b | 36 | */ |
ff9da691 JA |
37 | unsigned int pipe_max_size = 1048576; |
38 | ||
759c0114 WT |
39 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
40 | * matches default values. | |
41 | */ | |
42 | unsigned long pipe_user_pages_hard; | |
43 | unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; | |
44 | ||
1da177e4 | 45 | /* |
8cefc107 DH |
46 | * We use head and tail indices that aren't masked off, except at the point of |
47 | * dereference, but rather they're allowed to wrap naturally. This means there | |
48 | * isn't a dead spot in the buffer, but the ring has to be a power of two and | |
49 | * <= 2^31. | |
50 | * -- David Howells 2019-09-23. | |
51 | * | |
1da177e4 LT |
52 | * Reads with count = 0 should always return 0. |
53 | * -- Julian Bradfield 1999-06-07. | |
54 | * | |
55 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
56 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
57 | * | |
58 | * pipe_read & write cleanup | |
59 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
60 | */ | |
61 | ||
61e0d47c MS |
62 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
63 | { | |
6447a3cf | 64 | if (pipe->files) |
72b0d9aa | 65 | mutex_lock_nested(&pipe->mutex, subclass); |
61e0d47c MS |
66 | } |
67 | ||
68 | void pipe_lock(struct pipe_inode_info *pipe) | |
69 | { | |
70 | /* | |
71 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
72 | */ | |
73 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
74 | } | |
75 | EXPORT_SYMBOL(pipe_lock); | |
76 | ||
77 | void pipe_unlock(struct pipe_inode_info *pipe) | |
78 | { | |
6447a3cf | 79 | if (pipe->files) |
72b0d9aa | 80 | mutex_unlock(&pipe->mutex); |
61e0d47c MS |
81 | } |
82 | EXPORT_SYMBOL(pipe_unlock); | |
83 | ||
ebec73f4 AV |
84 | static inline void __pipe_lock(struct pipe_inode_info *pipe) |
85 | { | |
86 | mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); | |
87 | } | |
88 | ||
89 | static inline void __pipe_unlock(struct pipe_inode_info *pipe) | |
90 | { | |
91 | mutex_unlock(&pipe->mutex); | |
92 | } | |
93 | ||
61e0d47c MS |
94 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
95 | struct pipe_inode_info *pipe2) | |
96 | { | |
97 | BUG_ON(pipe1 == pipe2); | |
98 | ||
99 | if (pipe1 < pipe2) { | |
100 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
101 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
102 | } else { | |
023d43c7 PZ |
103 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
104 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
105 | } |
106 | } | |
107 | ||
1da177e4 | 108 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
3a326a2c | 109 | void pipe_wait(struct pipe_inode_info *pipe) |
1da177e4 | 110 | { |
0ddad21d LT |
111 | DEFINE_WAIT(rdwait); |
112 | DEFINE_WAIT(wrwait); | |
1da177e4 | 113 | |
d79fc0fc IM |
114 | /* |
115 | * Pipes are system-local resources, so sleeping on them | |
116 | * is considered a noninteractive wait: | |
117 | */ | |
0ddad21d LT |
118 | prepare_to_wait(&pipe->rd_wait, &rdwait, TASK_INTERRUPTIBLE); |
119 | prepare_to_wait(&pipe->wr_wait, &wrwait, TASK_INTERRUPTIBLE); | |
61e0d47c | 120 | pipe_unlock(pipe); |
1da177e4 | 121 | schedule(); |
0ddad21d LT |
122 | finish_wait(&pipe->rd_wait, &rdwait); |
123 | finish_wait(&pipe->wr_wait, &wrwait); | |
61e0d47c | 124 | pipe_lock(pipe); |
1da177e4 LT |
125 | } |
126 | ||
341b446b IM |
127 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
128 | struct pipe_buffer *buf) | |
1da177e4 LT |
129 | { |
130 | struct page *page = buf->page; | |
131 | ||
5274f052 JA |
132 | /* |
133 | * If nobody else uses this page, and we don't already have a | |
134 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 135 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 136 | */ |
341b446b | 137 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 138 | pipe->tmp_page = page; |
341b446b | 139 | else |
09cbfeaf | 140 | put_page(page); |
1da177e4 LT |
141 | } |
142 | ||
d86133bd VD |
143 | static int anon_pipe_buf_steal(struct pipe_inode_info *pipe, |
144 | struct pipe_buffer *buf) | |
145 | { | |
146 | struct page *page = buf->page; | |
147 | ||
148 | if (page_count(page) == 1) { | |
f4b00eab | 149 | memcg_kmem_uncharge_page(page, 0); |
d86133bd VD |
150 | __SetPageLocked(page); |
151 | return 0; | |
152 | } | |
153 | return 1; | |
154 | } | |
155 | ||
0845718d | 156 | /** |
b51d63c6 | 157 | * generic_pipe_buf_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 | */ | |
330ab716 JA |
168 | int generic_pipe_buf_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 JA |
179 | lock_page(page); |
180 | return 0; | |
181 | } | |
182 | ||
183 | return 1; | |
5abc97aa | 184 | } |
51921cb7 | 185 | EXPORT_SYMBOL(generic_pipe_buf_steal); |
5abc97aa | 186 | |
0845718d | 187 | /** |
b51d63c6 | 188 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
189 | * @pipe: the pipe that the buffer belongs to |
190 | * @buf: the buffer to get a reference to | |
191 | * | |
192 | * Description: | |
193 | * This function grabs an extra reference to @buf. It's used in | |
194 | * in the tee() system call, when we duplicate the buffers in one | |
195 | * pipe into another. | |
196 | */ | |
15fab63e | 197 | bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
70524490 | 198 | { |
15fab63e | 199 | return try_get_page(buf->page); |
70524490 | 200 | } |
51921cb7 | 201 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 202 | |
6818173b MS |
203 | /** |
204 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
205 | * @pipe: the pipe that the buffer belongs to | |
206 | * @buf: the buffer to put a reference to | |
207 | * | |
208 | * Description: | |
209 | * This function releases a reference to @buf. | |
210 | */ | |
211 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
212 | struct pipe_buffer *buf) | |
213 | { | |
09cbfeaf | 214 | put_page(buf->page); |
6818173b | 215 | } |
51921cb7 | 216 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 217 | |
d4c3cca9 | 218 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
1da177e4 | 219 | .release = anon_pipe_buf_release, |
d86133bd | 220 | .steal = anon_pipe_buf_steal, |
f84d7519 | 221 | .get = generic_pipe_buf_get, |
1da177e4 LT |
222 | }; |
223 | ||
85190d15 LT |
224 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
225 | static inline bool pipe_readable(const struct pipe_inode_info *pipe) | |
226 | { | |
227 | unsigned int head = READ_ONCE(pipe->head); | |
228 | unsigned int tail = READ_ONCE(pipe->tail); | |
229 | unsigned int writers = READ_ONCE(pipe->writers); | |
230 | ||
231 | return !pipe_empty(head, tail) || !writers; | |
232 | } | |
233 | ||
1da177e4 | 234 | static ssize_t |
fb9096a3 | 235 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 236 | { |
fb9096a3 | 237 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 238 | struct file *filp = iocb->ki_filp; |
de32ec4c | 239 | struct pipe_inode_info *pipe = filp->private_data; |
0ddad21d | 240 | bool was_full, wake_next_reader = false; |
1da177e4 | 241 | ssize_t ret; |
1da177e4 | 242 | |
1da177e4 LT |
243 | /* Null read succeeds. */ |
244 | if (unlikely(total_len == 0)) | |
245 | return 0; | |
246 | ||
1da177e4 | 247 | ret = 0; |
ebec73f4 | 248 | __pipe_lock(pipe); |
f467a6a6 LT |
249 | |
250 | /* | |
251 | * We only wake up writers if the pipe was full when we started | |
252 | * reading in order to avoid unnecessary wakeups. | |
253 | * | |
254 | * But when we do wake up writers, we do so using a sync wakeup | |
255 | * (WF_SYNC), because we want them to get going and generate more | |
256 | * data for us. | |
257 | */ | |
258 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); | |
1da177e4 | 259 | for (;;) { |
8cefc107 DH |
260 | unsigned int head = pipe->head; |
261 | unsigned int tail = pipe->tail; | |
262 | unsigned int mask = pipe->ring_size - 1; | |
263 | ||
264 | if (!pipe_empty(head, tail)) { | |
265 | struct pipe_buffer *buf = &pipe->bufs[tail & mask]; | |
1da177e4 | 266 | size_t chars = buf->len; |
637b58c2 AV |
267 | size_t written; |
268 | int error; | |
1da177e4 LT |
269 | |
270 | if (chars > total_len) | |
271 | chars = total_len; | |
272 | ||
fba597db | 273 | error = pipe_buf_confirm(pipe, buf); |
f84d7519 | 274 | if (error) { |
5274f052 | 275 | if (!ret) |
e5953cbd | 276 | ret = error; |
5274f052 JA |
277 | break; |
278 | } | |
f84d7519 | 279 | |
fb9096a3 | 280 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 281 | if (unlikely(written < chars)) { |
341b446b | 282 | if (!ret) |
637b58c2 | 283 | ret = -EFAULT; |
1da177e4 LT |
284 | break; |
285 | } | |
286 | ret += chars; | |
287 | buf->offset += chars; | |
288 | buf->len -= chars; | |
9883035a LT |
289 | |
290 | /* Was it a packet buffer? Clean up and exit */ | |
291 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
292 | total_len = chars; | |
293 | buf->len = 0; | |
294 | } | |
295 | ||
1da177e4 | 296 | if (!buf->len) { |
a779638c | 297 | pipe_buf_release(pipe, buf); |
0ddad21d | 298 | spin_lock_irq(&pipe->rd_wait.lock); |
8cefc107 DH |
299 | tail++; |
300 | pipe->tail = tail; | |
0ddad21d | 301 | spin_unlock_irq(&pipe->rd_wait.lock); |
1da177e4 LT |
302 | } |
303 | total_len -= chars; | |
304 | if (!total_len) | |
305 | break; /* common path: read succeeded */ | |
8cefc107 DH |
306 | if (!pipe_empty(head, tail)) /* More to do? */ |
307 | continue; | |
1da177e4 | 308 | } |
8cefc107 | 309 | |
923f4f23 | 310 | if (!pipe->writers) |
1da177e4 | 311 | break; |
a28c8b9d LT |
312 | if (ret) |
313 | break; | |
314 | if (filp->f_flags & O_NONBLOCK) { | |
315 | ret = -EAGAIN; | |
316 | break; | |
1da177e4 | 317 | } |
85190d15 | 318 | __pipe_unlock(pipe); |
d1c6a2aa LT |
319 | |
320 | /* | |
321 | * We only get here if we didn't actually read anything. | |
322 | * | |
323 | * However, we could have seen (and removed) a zero-sized | |
324 | * pipe buffer, and might have made space in the buffers | |
325 | * that way. | |
326 | * | |
327 | * You can't make zero-sized pipe buffers by doing an empty | |
328 | * write (not even in packet mode), but they can happen if | |
329 | * the writer gets an EFAULT when trying to fill a buffer | |
330 | * that already got allocated and inserted in the buffer | |
331 | * array. | |
332 | * | |
333 | * So we still need to wake up any pending writers in the | |
334 | * _very_ unlikely case that the pipe was full, but we got | |
335 | * no data. | |
336 | */ | |
337 | if (unlikely(was_full)) { | |
0ddad21d | 338 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
f467a6a6 LT |
339 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
340 | } | |
d1c6a2aa LT |
341 | |
342 | /* | |
343 | * But because we didn't read anything, at this point we can | |
344 | * just return directly with -ERESTARTSYS if we're interrupted, | |
345 | * since we've done any required wakeups and there's no need | |
346 | * to mark anything accessed. And we've dropped the lock. | |
347 | */ | |
0ddad21d | 348 | if (wait_event_interruptible_exclusive(pipe->rd_wait, pipe_readable(pipe)) < 0) |
d1c6a2aa LT |
349 | return -ERESTARTSYS; |
350 | ||
85190d15 | 351 | __pipe_lock(pipe); |
f467a6a6 | 352 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); |
0ddad21d | 353 | wake_next_reader = true; |
1da177e4 | 354 | } |
0ddad21d LT |
355 | if (pipe_empty(pipe->head, pipe->tail)) |
356 | wake_next_reader = false; | |
ebec73f4 | 357 | __pipe_unlock(pipe); |
341b446b | 358 | |
f467a6a6 | 359 | if (was_full) { |
0ddad21d | 360 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
923f4f23 | 361 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 | 362 | } |
0ddad21d LT |
363 | if (wake_next_reader) |
364 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); | |
1da177e4 LT |
365 | if (ret > 0) |
366 | file_accessed(filp); | |
367 | return ret; | |
368 | } | |
369 | ||
9883035a LT |
370 | static inline int is_packetized(struct file *file) |
371 | { | |
372 | return (file->f_flags & O_DIRECT) != 0; | |
373 | } | |
374 | ||
85190d15 LT |
375 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
376 | static inline bool pipe_writable(const struct pipe_inode_info *pipe) | |
377 | { | |
378 | unsigned int head = READ_ONCE(pipe->head); | |
379 | unsigned int tail = READ_ONCE(pipe->tail); | |
380 | unsigned int max_usage = READ_ONCE(pipe->max_usage); | |
381 | ||
382 | return !pipe_full(head, tail, max_usage) || | |
383 | !READ_ONCE(pipe->readers); | |
384 | } | |
385 | ||
1da177e4 | 386 | static ssize_t |
f0d1bec9 | 387 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 388 | { |
ee0b3e67 | 389 | struct file *filp = iocb->ki_filp; |
de32ec4c | 390 | struct pipe_inode_info *pipe = filp->private_data; |
8f868d68 | 391 | unsigned int head; |
f0d1bec9 | 392 | ssize_t ret = 0; |
f0d1bec9 | 393 | size_t total_len = iov_iter_count(from); |
1da177e4 | 394 | ssize_t chars; |
1b6b26ae | 395 | bool was_empty = false; |
0ddad21d | 396 | bool wake_next_writer = false; |
1da177e4 | 397 | |
1da177e4 LT |
398 | /* Null write succeeds. */ |
399 | if (unlikely(total_len == 0)) | |
400 | return 0; | |
401 | ||
ebec73f4 | 402 | __pipe_lock(pipe); |
1da177e4 | 403 | |
923f4f23 | 404 | if (!pipe->readers) { |
1da177e4 LT |
405 | send_sig(SIGPIPE, current, 0); |
406 | ret = -EPIPE; | |
407 | goto out; | |
408 | } | |
409 | ||
1b6b26ae LT |
410 | /* |
411 | * Only wake up if the pipe started out empty, since | |
412 | * otherwise there should be no readers waiting. | |
413 | * | |
414 | * If it wasn't empty we try to merge new data into | |
415 | * the last buffer. | |
416 | * | |
417 | * That naturally merges small writes, but it also | |
418 | * page-aligs the rest of the writes for large writes | |
419 | * spanning multiple pages. | |
420 | */ | |
8cefc107 | 421 | head = pipe->head; |
1b6b26ae LT |
422 | was_empty = pipe_empty(head, pipe->tail); |
423 | chars = total_len & (PAGE_SIZE-1); | |
424 | if (chars && !was_empty) { | |
8f868d68 | 425 | unsigned int mask = pipe->ring_size - 1; |
8cefc107 | 426 | struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; |
1da177e4 | 427 | int offset = buf->offset + buf->len; |
341b446b | 428 | |
f6dd9755 CH |
429 | if ((buf->flags & PIPE_BUF_FLAG_CAN_MERGE) && |
430 | offset + chars <= PAGE_SIZE) { | |
fba597db | 431 | ret = pipe_buf_confirm(pipe, buf); |
6ae08069 | 432 | if (ret) |
5274f052 | 433 | goto out; |
f84d7519 | 434 | |
f0d1bec9 AV |
435 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
436 | if (unlikely(ret < chars)) { | |
6ae08069 | 437 | ret = -EFAULT; |
1da177e4 | 438 | goto out; |
f6762b7a | 439 | } |
1b6b26ae | 440 | |
6ae08069 | 441 | buf->len += ret; |
f0d1bec9 | 442 | if (!iov_iter_count(from)) |
1da177e4 LT |
443 | goto out; |
444 | } | |
445 | } | |
446 | ||
447 | for (;;) { | |
923f4f23 | 448 | if (!pipe->readers) { |
1da177e4 | 449 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
450 | if (!ret) |
451 | ret = -EPIPE; | |
1da177e4 LT |
452 | break; |
453 | } | |
8cefc107 | 454 | |
a194dfe6 | 455 | head = pipe->head; |
8f868d68 DH |
456 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) { |
457 | unsigned int mask = pipe->ring_size - 1; | |
8cefc107 | 458 | struct pipe_buffer *buf = &pipe->bufs[head & mask]; |
923f4f23 | 459 | struct page *page = pipe->tmp_page; |
f0d1bec9 | 460 | int copied; |
1da177e4 LT |
461 | |
462 | if (!page) { | |
d86133bd | 463 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
464 | if (unlikely(!page)) { |
465 | ret = ret ? : -ENOMEM; | |
466 | break; | |
467 | } | |
923f4f23 | 468 | pipe->tmp_page = page; |
1da177e4 | 469 | } |
a194dfe6 DH |
470 | |
471 | /* Allocate a slot in the ring in advance and attach an | |
472 | * empty buffer. If we fault or otherwise fail to use | |
473 | * it, either the reader will consume it or it'll still | |
474 | * be there for the next write. | |
475 | */ | |
0ddad21d | 476 | spin_lock_irq(&pipe->rd_wait.lock); |
a194dfe6 DH |
477 | |
478 | head = pipe->head; | |
8f868d68 | 479 | if (pipe_full(head, pipe->tail, pipe->max_usage)) { |
0ddad21d | 480 | spin_unlock_irq(&pipe->rd_wait.lock); |
8df44129 DH |
481 | continue; |
482 | } | |
483 | ||
a194dfe6 | 484 | pipe->head = head + 1; |
0ddad21d | 485 | spin_unlock_irq(&pipe->rd_wait.lock); |
1da177e4 LT |
486 | |
487 | /* Insert it into the buffer array */ | |
a194dfe6 | 488 | buf = &pipe->bufs[head & mask]; |
1da177e4 LT |
489 | buf->page = page; |
490 | buf->ops = &anon_pipe_buf_ops; | |
491 | buf->offset = 0; | |
a194dfe6 | 492 | buf->len = 0; |
f6dd9755 | 493 | if (is_packetized(filp)) |
9883035a | 494 | buf->flags = PIPE_BUF_FLAG_PACKET; |
f6dd9755 CH |
495 | else |
496 | buf->flags = PIPE_BUF_FLAG_CAN_MERGE; | |
923f4f23 | 497 | pipe->tmp_page = NULL; |
1da177e4 | 498 | |
a194dfe6 DH |
499 | copied = copy_page_from_iter(page, 0, PAGE_SIZE, from); |
500 | if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) { | |
501 | if (!ret) | |
502 | ret = -EFAULT; | |
503 | break; | |
504 | } | |
505 | ret += copied; | |
506 | buf->offset = 0; | |
507 | buf->len = copied; | |
508 | ||
f0d1bec9 | 509 | if (!iov_iter_count(from)) |
1da177e4 LT |
510 | break; |
511 | } | |
8cefc107 | 512 | |
8f868d68 | 513 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) |
1da177e4 | 514 | continue; |
8cefc107 DH |
515 | |
516 | /* Wait for buffer space to become available. */ | |
1da177e4 | 517 | if (filp->f_flags & O_NONBLOCK) { |
341b446b IM |
518 | if (!ret) |
519 | ret = -EAGAIN; | |
1da177e4 LT |
520 | break; |
521 | } | |
522 | if (signal_pending(current)) { | |
341b446b IM |
523 | if (!ret) |
524 | ret = -ERESTARTSYS; | |
1da177e4 LT |
525 | break; |
526 | } | |
1b6b26ae LT |
527 | |
528 | /* | |
529 | * We're going to release the pipe lock and wait for more | |
530 | * space. We wake up any readers if necessary, and then | |
531 | * after waiting we need to re-check whether the pipe | |
532 | * become empty while we dropped the lock. | |
533 | */ | |
85190d15 | 534 | __pipe_unlock(pipe); |
1b6b26ae | 535 | if (was_empty) { |
0ddad21d | 536 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
1b6b26ae LT |
537 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
538 | } | |
0ddad21d | 539 | wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); |
85190d15 | 540 | __pipe_lock(pipe); |
0dd1e377 | 541 | was_empty = pipe_empty(pipe->head, pipe->tail); |
0ddad21d | 542 | wake_next_writer = true; |
1da177e4 LT |
543 | } |
544 | out: | |
0ddad21d LT |
545 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) |
546 | wake_next_writer = false; | |
ebec73f4 | 547 | __pipe_unlock(pipe); |
1b6b26ae LT |
548 | |
549 | /* | |
550 | * If we do do a wakeup event, we do a 'sync' wakeup, because we | |
551 | * want the reader to start processing things asap, rather than | |
552 | * leave the data pending. | |
553 | * | |
554 | * This is particularly important for small writes, because of | |
555 | * how (for example) the GNU make jobserver uses small writes to | |
556 | * wake up pending jobs | |
557 | */ | |
558 | if (was_empty) { | |
0ddad21d | 559 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
923f4f23 | 560 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 | 561 | } |
0ddad21d LT |
562 | if (wake_next_writer) |
563 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); | |
7e775f46 | 564 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
565 | int err = file_update_time(filp); |
566 | if (err) | |
567 | ret = err; | |
7e775f46 | 568 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 569 | } |
1da177e4 LT |
570 | return ret; |
571 | } | |
572 | ||
d59d0b1b | 573 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 574 | { |
de32ec4c | 575 | struct pipe_inode_info *pipe = filp->private_data; |
8cefc107 | 576 | int count, head, tail, mask; |
1da177e4 LT |
577 | |
578 | switch (cmd) { | |
579 | case FIONREAD: | |
ebec73f4 | 580 | __pipe_lock(pipe); |
1da177e4 | 581 | count = 0; |
8cefc107 DH |
582 | head = pipe->head; |
583 | tail = pipe->tail; | |
584 | mask = pipe->ring_size - 1; | |
585 | ||
586 | while (tail != head) { | |
587 | count += pipe->bufs[tail & mask].len; | |
588 | tail++; | |
1da177e4 | 589 | } |
ebec73f4 | 590 | __pipe_unlock(pipe); |
923f4f23 | 591 | |
1da177e4 LT |
592 | return put_user(count, (int __user *)arg); |
593 | default: | |
46ce341b | 594 | return -ENOIOCTLCMD; |
1da177e4 LT |
595 | } |
596 | } | |
597 | ||
dd67081b | 598 | /* No kernel lock held - fine */ |
a11e1d43 LT |
599 | static __poll_t |
600 | pipe_poll(struct file *filp, poll_table *wait) | |
dd67081b | 601 | { |
a11e1d43 | 602 | __poll_t mask; |
dd67081b | 603 | struct pipe_inode_info *pipe = filp->private_data; |
ad910e36 | 604 | unsigned int head, tail; |
a11e1d43 | 605 | |
ad910e36 | 606 | /* |
0ddad21d | 607 | * Reading pipe state only -- no need for acquiring the semaphore. |
ad910e36 LT |
608 | * |
609 | * But because this is racy, the code has to add the | |
610 | * entry to the poll table _first_ .. | |
611 | */ | |
0ddad21d LT |
612 | if (filp->f_mode & FMODE_READ) |
613 | poll_wait(filp, &pipe->rd_wait, wait); | |
614 | if (filp->f_mode & FMODE_WRITE) | |
615 | poll_wait(filp, &pipe->wr_wait, wait); | |
1da177e4 | 616 | |
ad910e36 LT |
617 | /* |
618 | * .. and only then can you do the racy tests. That way, | |
619 | * if something changes and you got it wrong, the poll | |
620 | * table entry will wake you up and fix it. | |
621 | */ | |
622 | head = READ_ONCE(pipe->head); | |
623 | tail = READ_ONCE(pipe->tail); | |
624 | ||
a11e1d43 | 625 | mask = 0; |
1da177e4 | 626 | if (filp->f_mode & FMODE_READ) { |
8cefc107 DH |
627 | if (!pipe_empty(head, tail)) |
628 | mask |= EPOLLIN | EPOLLRDNORM; | |
923f4f23 | 629 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
a9a08845 | 630 | mask |= EPOLLHUP; |
1da177e4 LT |
631 | } |
632 | ||
633 | if (filp->f_mode & FMODE_WRITE) { | |
6718b6f8 | 634 | if (!pipe_full(head, tail, pipe->max_usage)) |
8cefc107 | 635 | mask |= EPOLLOUT | EPOLLWRNORM; |
5e5d7a22 | 636 | /* |
a9a08845 | 637 | * Most Unices do not set EPOLLERR for FIFOs but on Linux they |
5e5d7a22 PE |
638 | * behave exactly like pipes for poll(). |
639 | */ | |
923f4f23 | 640 | if (!pipe->readers) |
a9a08845 | 641 | mask |= EPOLLERR; |
1da177e4 LT |
642 | } |
643 | ||
644 | return mask; | |
645 | } | |
646 | ||
b0d8d229 LT |
647 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
648 | { | |
649 | int kill = 0; | |
650 | ||
651 | spin_lock(&inode->i_lock); | |
652 | if (!--pipe->files) { | |
653 | inode->i_pipe = NULL; | |
654 | kill = 1; | |
655 | } | |
656 | spin_unlock(&inode->i_lock); | |
657 | ||
658 | if (kill) | |
659 | free_pipe_info(pipe); | |
660 | } | |
661 | ||
1da177e4 | 662 | static int |
599a0ac1 | 663 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 664 | { |
b0d8d229 | 665 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 666 | |
ebec73f4 | 667 | __pipe_lock(pipe); |
599a0ac1 AV |
668 | if (file->f_mode & FMODE_READ) |
669 | pipe->readers--; | |
670 | if (file->f_mode & FMODE_WRITE) | |
671 | pipe->writers--; | |
341b446b | 672 | |
6551d5c5 LT |
673 | /* Was that the last reader or writer, but not the other side? */ |
674 | if (!pipe->readers != !pipe->writers) { | |
675 | wake_up_interruptible_all(&pipe->rd_wait); | |
676 | wake_up_interruptible_all(&pipe->wr_wait); | |
923f4f23 IM |
677 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
678 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 679 | } |
ebec73f4 | 680 | __pipe_unlock(pipe); |
ba5bb147 | 681 | |
b0d8d229 | 682 | put_pipe_info(inode, pipe); |
1da177e4 LT |
683 | return 0; |
684 | } | |
685 | ||
686 | static int | |
599a0ac1 | 687 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 688 | { |
de32ec4c | 689 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 690 | int retval = 0; |
1da177e4 | 691 | |
ebec73f4 | 692 | __pipe_lock(pipe); |
599a0ac1 AV |
693 | if (filp->f_mode & FMODE_READ) |
694 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
695 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 696 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
697 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
698 | /* this can happen only if on == T */ | |
e5bc49ba ON |
699 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
700 | } | |
ebec73f4 | 701 | __pipe_unlock(pipe); |
60aa4924 | 702 | return retval; |
1da177e4 LT |
703 | } |
704 | ||
9c87bcf0 | 705 | static unsigned long account_pipe_buffers(struct user_struct *user, |
759c0114 WT |
706 | unsigned long old, unsigned long new) |
707 | { | |
9c87bcf0 | 708 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
709 | } |
710 | ||
9c87bcf0 | 711 | static bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 712 | { |
f7340761 EB |
713 | unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft); |
714 | ||
715 | return soft_limit && user_bufs > soft_limit; | |
759c0114 WT |
716 | } |
717 | ||
9c87bcf0 | 718 | static bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 719 | { |
f7340761 EB |
720 | unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard); |
721 | ||
722 | return hard_limit && user_bufs > hard_limit; | |
759c0114 WT |
723 | } |
724 | ||
85c2dd54 EB |
725 | static bool is_unprivileged_user(void) |
726 | { | |
727 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
728 | } | |
729 | ||
7bee130e | 730 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 731 | { |
923f4f23 | 732 | struct pipe_inode_info *pipe; |
09b4d199 MK |
733 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
734 | struct user_struct *user = get_current_user(); | |
9c87bcf0 | 735 | unsigned long user_bufs; |
f7340761 | 736 | unsigned int max_size = READ_ONCE(pipe_max_size); |
3a326a2c | 737 | |
d86133bd | 738 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
09b4d199 MK |
739 | if (pipe == NULL) |
740 | goto out_free_uid; | |
741 | ||
f7340761 EB |
742 | if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE)) |
743 | pipe_bufs = max_size >> PAGE_SHIFT; | |
086e774a | 744 | |
9c87bcf0 | 745 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
a005ca0e | 746 | |
85c2dd54 | 747 | if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) { |
9c87bcf0 | 748 | user_bufs = account_pipe_buffers(user, pipe_bufs, 1); |
a005ca0e | 749 | pipe_bufs = 1; |
09b4d199 | 750 | } |
759c0114 | 751 | |
85c2dd54 | 752 | if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user()) |
a005ca0e MK |
753 | goto out_revert_acct; |
754 | ||
755 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), | |
756 | GFP_KERNEL_ACCOUNT); | |
757 | ||
09b4d199 | 758 | if (pipe->bufs) { |
0ddad21d LT |
759 | init_waitqueue_head(&pipe->rd_wait); |
760 | init_waitqueue_head(&pipe->wr_wait); | |
09b4d199 | 761 | pipe->r_counter = pipe->w_counter = 1; |
6718b6f8 | 762 | pipe->max_usage = pipe_bufs; |
8cefc107 | 763 | pipe->ring_size = pipe_bufs; |
09b4d199 | 764 | pipe->user = user; |
09b4d199 MK |
765 | mutex_init(&pipe->mutex); |
766 | return pipe; | |
3a326a2c IM |
767 | } |
768 | ||
a005ca0e | 769 | out_revert_acct: |
9c87bcf0 | 770 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
09b4d199 MK |
771 | kfree(pipe); |
772 | out_free_uid: | |
773 | free_uid(user); | |
35f3d14d | 774 | return NULL; |
3a326a2c IM |
775 | } |
776 | ||
4b8a8f1e | 777 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
778 | { |
779 | int i; | |
1da177e4 | 780 | |
8cefc107 | 781 | (void) account_pipe_buffers(pipe->user, pipe->ring_size, 0); |
759c0114 | 782 | free_uid(pipe->user); |
8cefc107 | 783 | for (i = 0; i < pipe->ring_size; i++) { |
923f4f23 | 784 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 785 | if (buf->ops) |
a779638c | 786 | pipe_buf_release(pipe, buf); |
1da177e4 | 787 | } |
923f4f23 IM |
788 | if (pipe->tmp_page) |
789 | __free_page(pipe->tmp_page); | |
35f3d14d | 790 | kfree(pipe->bufs); |
923f4f23 | 791 | kfree(pipe); |
1da177e4 LT |
792 | } |
793 | ||
fa3536cc | 794 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 795 | |
c23fbb6b ED |
796 | /* |
797 | * pipefs_dname() is called from d_path(). | |
798 | */ | |
799 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
800 | { | |
801 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
75c3cfa8 | 802 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
803 | } |
804 | ||
3ba13d17 | 805 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 806 | .d_dname = pipefs_dname, |
1da177e4 LT |
807 | }; |
808 | ||
809 | static struct inode * get_pipe_inode(void) | |
810 | { | |
a209dfc7 | 811 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 812 | struct pipe_inode_info *pipe; |
1da177e4 LT |
813 | |
814 | if (!inode) | |
815 | goto fail_inode; | |
816 | ||
85fe4025 CH |
817 | inode->i_ino = get_next_ino(); |
818 | ||
7bee130e | 819 | pipe = alloc_pipe_info(); |
923f4f23 | 820 | if (!pipe) |
1da177e4 | 821 | goto fail_iput; |
3a326a2c | 822 | |
ba5bb147 AV |
823 | inode->i_pipe = pipe; |
824 | pipe->files = 2; | |
923f4f23 | 825 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 826 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
827 | |
828 | /* | |
829 | * Mark the inode dirty from the very beginning, | |
830 | * that way it will never be moved to the dirty | |
831 | * list because "mark_inode_dirty()" will think | |
832 | * that it already _is_ on the dirty list. | |
833 | */ | |
834 | inode->i_state = I_DIRTY; | |
835 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
836 | inode->i_uid = current_fsuid(); |
837 | inode->i_gid = current_fsgid(); | |
078cd827 | 838 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
923f4f23 | 839 | |
1da177e4 LT |
840 | return inode; |
841 | ||
842 | fail_iput: | |
843 | iput(inode); | |
341b446b | 844 | |
1da177e4 LT |
845 | fail_inode: |
846 | return NULL; | |
847 | } | |
848 | ||
e4fad8e5 | 849 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 850 | { |
e4fad8e5 | 851 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 852 | struct file *f; |
1da177e4 | 853 | |
1da177e4 | 854 | if (!inode) |
e4fad8e5 | 855 | return -ENFILE; |
1da177e4 | 856 | |
152b6372 AV |
857 | f = alloc_file_pseudo(inode, pipe_mnt, "", |
858 | O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)), | |
859 | &pipefifo_fops); | |
e9bb1f9b | 860 | if (IS_ERR(f)) { |
152b6372 AV |
861 | free_pipe_info(inode->i_pipe); |
862 | iput(inode); | |
863 | return PTR_ERR(f); | |
e9bb1f9b | 864 | } |
341b446b | 865 | |
de32ec4c | 866 | f->private_data = inode->i_pipe; |
d6cbd281 | 867 | |
183266f2 AV |
868 | res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK), |
869 | &pipefifo_fops); | |
e9bb1f9b | 870 | if (IS_ERR(res[0])) { |
b10a4a9f AV |
871 | put_pipe_info(inode, inode->i_pipe); |
872 | fput(f); | |
873 | return PTR_ERR(res[0]); | |
e9bb1f9b | 874 | } |
de32ec4c | 875 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 | 876 | res[1] = f; |
d8e464ec LT |
877 | stream_open(inode, res[0]); |
878 | stream_open(inode, res[1]); | |
e4fad8e5 | 879 | return 0; |
d6cbd281 AK |
880 | } |
881 | ||
5b249b1b | 882 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 883 | { |
d6cbd281 AK |
884 | int error; |
885 | int fdw, fdr; | |
886 | ||
9883035a | 887 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT)) |
ed8cae8b UD |
888 | return -EINVAL; |
889 | ||
e4fad8e5 AV |
890 | error = create_pipe_files(files, flags); |
891 | if (error) | |
892 | return error; | |
d6cbd281 | 893 | |
ed8cae8b | 894 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
895 | if (error < 0) |
896 | goto err_read_pipe; | |
897 | fdr = error; | |
898 | ||
ed8cae8b | 899 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
900 | if (error < 0) |
901 | goto err_fdr; | |
902 | fdw = error; | |
903 | ||
157cf649 | 904 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
905 | fd[0] = fdr; |
906 | fd[1] = fdw; | |
d6cbd281 AK |
907 | return 0; |
908 | ||
909 | err_fdr: | |
910 | put_unused_fd(fdr); | |
911 | err_read_pipe: | |
e4fad8e5 AV |
912 | fput(files[0]); |
913 | fput(files[1]); | |
d6cbd281 | 914 | return error; |
1da177e4 LT |
915 | } |
916 | ||
5b249b1b AV |
917 | int do_pipe_flags(int *fd, int flags) |
918 | { | |
919 | struct file *files[2]; | |
920 | int error = __do_pipe_flags(fd, files, flags); | |
921 | if (!error) { | |
922 | fd_install(fd[0], files[0]); | |
923 | fd_install(fd[1], files[1]); | |
924 | } | |
925 | return error; | |
926 | } | |
927 | ||
d35c7b0e UD |
928 | /* |
929 | * sys_pipe() is the normal C calling standard for creating | |
930 | * a pipe. It's not the way Unix traditionally does this, though. | |
931 | */ | |
0a216dd1 | 932 | static int do_pipe2(int __user *fildes, int flags) |
d35c7b0e | 933 | { |
5b249b1b | 934 | struct file *files[2]; |
d35c7b0e UD |
935 | int fd[2]; |
936 | int error; | |
937 | ||
5b249b1b | 938 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 939 | if (!error) { |
5b249b1b AV |
940 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
941 | fput(files[0]); | |
942 | fput(files[1]); | |
943 | put_unused_fd(fd[0]); | |
944 | put_unused_fd(fd[1]); | |
d35c7b0e | 945 | error = -EFAULT; |
5b249b1b AV |
946 | } else { |
947 | fd_install(fd[0], files[0]); | |
948 | fd_install(fd[1], files[1]); | |
ba719bae | 949 | } |
d35c7b0e UD |
950 | } |
951 | return error; | |
952 | } | |
953 | ||
0a216dd1 DB |
954 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
955 | { | |
956 | return do_pipe2(fildes, flags); | |
957 | } | |
958 | ||
2b664219 | 959 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b | 960 | { |
0a216dd1 | 961 | return do_pipe2(fildes, 0); |
ed8cae8b UD |
962 | } |
963 | ||
fc7478a2 | 964 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 | 965 | { |
8cefc107 | 966 | int cur = *cnt; |
f776c738 AV |
967 | |
968 | while (cur == *cnt) { | |
fc7478a2 | 969 | pipe_wait(pipe); |
f776c738 AV |
970 | if (signal_pending(current)) |
971 | break; | |
972 | } | |
973 | return cur == *cnt ? -ERESTARTSYS : 0; | |
974 | } | |
975 | ||
fc7478a2 | 976 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 977 | { |
6551d5c5 LT |
978 | wake_up_interruptible_all(&pipe->rd_wait); |
979 | wake_up_interruptible_all(&pipe->wr_wait); | |
f776c738 AV |
980 | } |
981 | ||
982 | static int fifo_open(struct inode *inode, struct file *filp) | |
983 | { | |
984 | struct pipe_inode_info *pipe; | |
599a0ac1 | 985 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
986 | int ret; |
987 | ||
ba5bb147 AV |
988 | filp->f_version = 0; |
989 | ||
990 | spin_lock(&inode->i_lock); | |
991 | if (inode->i_pipe) { | |
992 | pipe = inode->i_pipe; | |
993 | pipe->files++; | |
994 | spin_unlock(&inode->i_lock); | |
995 | } else { | |
996 | spin_unlock(&inode->i_lock); | |
7bee130e | 997 | pipe = alloc_pipe_info(); |
f776c738 | 998 | if (!pipe) |
ba5bb147 AV |
999 | return -ENOMEM; |
1000 | pipe->files = 1; | |
1001 | spin_lock(&inode->i_lock); | |
1002 | if (unlikely(inode->i_pipe)) { | |
1003 | inode->i_pipe->files++; | |
1004 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 1005 | free_pipe_info(pipe); |
ba5bb147 AV |
1006 | pipe = inode->i_pipe; |
1007 | } else { | |
1008 | inode->i_pipe = pipe; | |
1009 | spin_unlock(&inode->i_lock); | |
1010 | } | |
f776c738 | 1011 | } |
de32ec4c | 1012 | filp->private_data = pipe; |
ba5bb147 AV |
1013 | /* OK, we have a pipe and it's pinned down */ |
1014 | ||
ebec73f4 | 1015 | __pipe_lock(pipe); |
f776c738 AV |
1016 | |
1017 | /* We can only do regular read/write on fifos */ | |
d8e464ec | 1018 | stream_open(inode, filp); |
f776c738 | 1019 | |
d8e464ec | 1020 | switch (filp->f_mode & (FMODE_READ | FMODE_WRITE)) { |
f776c738 AV |
1021 | case FMODE_READ: |
1022 | /* | |
1023 | * O_RDONLY | |
1024 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
1025 | * opened, even when there is no process writing the FIFO. | |
1026 | */ | |
f776c738 AV |
1027 | pipe->r_counter++; |
1028 | if (pipe->readers++ == 0) | |
fc7478a2 | 1029 | wake_up_partner(pipe); |
f776c738 | 1030 | |
599a0ac1 | 1031 | if (!is_pipe && !pipe->writers) { |
f776c738 | 1032 | if ((filp->f_flags & O_NONBLOCK)) { |
a9a08845 | 1033 | /* suppress EPOLLHUP until we have |
f776c738 AV |
1034 | * seen a writer */ |
1035 | filp->f_version = pipe->w_counter; | |
1036 | } else { | |
fc7478a2 | 1037 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
1038 | goto err_rd; |
1039 | } | |
1040 | } | |
1041 | break; | |
8cefc107 | 1042 | |
f776c738 AV |
1043 | case FMODE_WRITE: |
1044 | /* | |
1045 | * O_WRONLY | |
1046 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
1047 | * errno=ENXIO when there is no process reading the FIFO. | |
1048 | */ | |
1049 | ret = -ENXIO; | |
599a0ac1 | 1050 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
1051 | goto err; |
1052 | ||
f776c738 AV |
1053 | pipe->w_counter++; |
1054 | if (!pipe->writers++) | |
fc7478a2 | 1055 | wake_up_partner(pipe); |
f776c738 | 1056 | |
599a0ac1 | 1057 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 1058 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
1059 | goto err_wr; |
1060 | } | |
1061 | break; | |
8cefc107 | 1062 | |
f776c738 AV |
1063 | case FMODE_READ | FMODE_WRITE: |
1064 | /* | |
1065 | * O_RDWR | |
1066 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
1067 | * This implementation will NEVER block on a O_RDWR open, since | |
1068 | * the process can at least talk to itself. | |
1069 | */ | |
f776c738 AV |
1070 | |
1071 | pipe->readers++; | |
1072 | pipe->writers++; | |
1073 | pipe->r_counter++; | |
1074 | pipe->w_counter++; | |
1075 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 1076 | wake_up_partner(pipe); |
f776c738 AV |
1077 | break; |
1078 | ||
1079 | default: | |
1080 | ret = -EINVAL; | |
1081 | goto err; | |
1082 | } | |
1083 | ||
1084 | /* Ok! */ | |
ebec73f4 | 1085 | __pipe_unlock(pipe); |
f776c738 AV |
1086 | return 0; |
1087 | ||
1088 | err_rd: | |
1089 | if (!--pipe->readers) | |
0ddad21d | 1090 | wake_up_interruptible(&pipe->wr_wait); |
f776c738 AV |
1091 | ret = -ERESTARTSYS; |
1092 | goto err; | |
1093 | ||
1094 | err_wr: | |
1095 | if (!--pipe->writers) | |
6551d5c5 | 1096 | wake_up_interruptible_all(&pipe->rd_wait); |
f776c738 AV |
1097 | ret = -ERESTARTSYS; |
1098 | goto err; | |
1099 | ||
1100 | err: | |
ebec73f4 | 1101 | __pipe_unlock(pipe); |
b0d8d229 LT |
1102 | |
1103 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1104 | return ret; |
1105 | } | |
1106 | ||
599a0ac1 AV |
1107 | const struct file_operations pipefifo_fops = { |
1108 | .open = fifo_open, | |
1109 | .llseek = no_llseek, | |
fb9096a3 | 1110 | .read_iter = pipe_read, |
f0d1bec9 | 1111 | .write_iter = pipe_write, |
a11e1d43 | 1112 | .poll = pipe_poll, |
599a0ac1 AV |
1113 | .unlocked_ioctl = pipe_ioctl, |
1114 | .release = pipe_release, | |
1115 | .fasync = pipe_fasync, | |
f776c738 AV |
1116 | }; |
1117 | ||
f491bd71 MK |
1118 | /* |
1119 | * Currently we rely on the pipe array holding a power-of-2 number | |
d3f14c48 | 1120 | * of pages. Returns 0 on error. |
f491bd71 | 1121 | */ |
96e99be4 | 1122 | unsigned int round_pipe_size(unsigned long size) |
f491bd71 | 1123 | { |
c4fed5a9 | 1124 | if (size > (1U << 31)) |
96e99be4 EB |
1125 | return 0; |
1126 | ||
4c2e4bef EB |
1127 | /* Minimum pipe size, as required by POSIX */ |
1128 | if (size < PAGE_SIZE) | |
c4fed5a9 | 1129 | return PAGE_SIZE; |
d3f14c48 | 1130 | |
c4fed5a9 | 1131 | return roundup_pow_of_two(size); |
f491bd71 MK |
1132 | } |
1133 | ||
35f3d14d JA |
1134 | /* |
1135 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1136 | * pipe size if successful, or return -ERROR on error. | |
1137 | */ | |
d37d4166 | 1138 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) |
35f3d14d JA |
1139 | { |
1140 | struct pipe_buffer *bufs; | |
8cefc107 | 1141 | unsigned int size, nr_slots, head, tail, mask, n; |
9c87bcf0 | 1142 | unsigned long user_bufs; |
b0b91d18 | 1143 | long ret = 0; |
d37d4166 MK |
1144 | |
1145 | size = round_pipe_size(arg); | |
8cefc107 | 1146 | nr_slots = size >> PAGE_SHIFT; |
d37d4166 | 1147 | |
8cefc107 | 1148 | if (!nr_slots) |
d37d4166 MK |
1149 | return -EINVAL; |
1150 | ||
b0b91d18 MK |
1151 | /* |
1152 | * If trying to increase the pipe capacity, check that an | |
1153 | * unprivileged user is not trying to exceed various limits | |
1154 | * (soft limit check here, hard limit check just below). | |
1155 | * Decreasing the pipe capacity is always permitted, even | |
1156 | * if the user is currently over a limit. | |
1157 | */ | |
8cefc107 | 1158 | if (nr_slots > pipe->ring_size && |
b0b91d18 | 1159 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) |
d37d4166 MK |
1160 | return -EPERM; |
1161 | ||
8cefc107 | 1162 | user_bufs = account_pipe_buffers(pipe->user, pipe->ring_size, nr_slots); |
b0b91d18 | 1163 | |
8cefc107 | 1164 | if (nr_slots > pipe->ring_size && |
9c87bcf0 MK |
1165 | (too_many_pipe_buffers_hard(user_bufs) || |
1166 | too_many_pipe_buffers_soft(user_bufs)) && | |
85c2dd54 | 1167 | is_unprivileged_user()) { |
b0b91d18 MK |
1168 | ret = -EPERM; |
1169 | goto out_revert_acct; | |
1170 | } | |
35f3d14d | 1171 | |
35f3d14d | 1172 | /* |
8cefc107 DH |
1173 | * We can shrink the pipe, if arg is greater than the ring occupancy. |
1174 | * Since we don't expect a lot of shrink+grow operations, just free and | |
1175 | * allocate again like we would do for growing. If the pipe currently | |
35f3d14d JA |
1176 | * contains more buffers than arg, then return busy. |
1177 | */ | |
8cefc107 DH |
1178 | mask = pipe->ring_size - 1; |
1179 | head = pipe->head; | |
1180 | tail = pipe->tail; | |
1181 | n = pipe_occupancy(pipe->head, pipe->tail); | |
1182 | if (nr_slots < n) { | |
b0b91d18 MK |
1183 | ret = -EBUSY; |
1184 | goto out_revert_acct; | |
1185 | } | |
35f3d14d | 1186 | |
8cefc107 | 1187 | bufs = kcalloc(nr_slots, sizeof(*bufs), |
d86133bd | 1188 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); |
b0b91d18 MK |
1189 | if (unlikely(!bufs)) { |
1190 | ret = -ENOMEM; | |
1191 | goto out_revert_acct; | |
1192 | } | |
35f3d14d JA |
1193 | |
1194 | /* | |
1195 | * The pipe array wraps around, so just start the new one at zero | |
8cefc107 | 1196 | * and adjust the indices. |
35f3d14d | 1197 | */ |
8cefc107 DH |
1198 | if (n > 0) { |
1199 | unsigned int h = head & mask; | |
1200 | unsigned int t = tail & mask; | |
1201 | if (h > t) { | |
1202 | memcpy(bufs, pipe->bufs + t, | |
1203 | n * sizeof(struct pipe_buffer)); | |
1204 | } else { | |
1205 | unsigned int tsize = pipe->ring_size - t; | |
1206 | if (h > 0) | |
1207 | memcpy(bufs + tsize, pipe->bufs, | |
1208 | h * sizeof(struct pipe_buffer)); | |
1209 | memcpy(bufs, pipe->bufs + t, | |
1210 | tsize * sizeof(struct pipe_buffer)); | |
1211 | } | |
35f3d14d JA |
1212 | } |
1213 | ||
8cefc107 DH |
1214 | head = n; |
1215 | tail = 0; | |
1216 | ||
35f3d14d JA |
1217 | kfree(pipe->bufs); |
1218 | pipe->bufs = bufs; | |
8cefc107 | 1219 | pipe->ring_size = nr_slots; |
6718b6f8 | 1220 | pipe->max_usage = nr_slots; |
8cefc107 DH |
1221 | pipe->tail = tail; |
1222 | pipe->head = head; | |
6551d5c5 LT |
1223 | |
1224 | /* This might have made more room for writers */ | |
1225 | wake_up_interruptible(&pipe->wr_wait); | |
6718b6f8 | 1226 | return pipe->max_usage * PAGE_SIZE; |
b0b91d18 MK |
1227 | |
1228 | out_revert_acct: | |
8cefc107 | 1229 | (void) account_pipe_buffers(pipe->user, nr_slots, pipe->ring_size); |
b0b91d18 | 1230 | return ret; |
35f3d14d JA |
1231 | } |
1232 | ||
72083646 LT |
1233 | /* |
1234 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1235 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1236 | * pipe. | |
1237 | */ | |
1238 | struct pipe_inode_info *get_pipe_info(struct file *file) | |
1239 | { | |
de32ec4c | 1240 | return file->f_op == &pipefifo_fops ? file->private_data : NULL; |
72083646 LT |
1241 | } |
1242 | ||
35f3d14d JA |
1243 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
1244 | { | |
1245 | struct pipe_inode_info *pipe; | |
1246 | long ret; | |
1247 | ||
c66fb347 | 1248 | pipe = get_pipe_info(file); |
35f3d14d JA |
1249 | if (!pipe) |
1250 | return -EBADF; | |
1251 | ||
ebec73f4 | 1252 | __pipe_lock(pipe); |
35f3d14d JA |
1253 | |
1254 | switch (cmd) { | |
d37d4166 MK |
1255 | case F_SETPIPE_SZ: |
1256 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1257 | break; |
1258 | case F_GETPIPE_SZ: | |
6718b6f8 | 1259 | ret = pipe->max_usage * PAGE_SIZE; |
35f3d14d JA |
1260 | break; |
1261 | default: | |
1262 | ret = -EINVAL; | |
1263 | break; | |
1264 | } | |
1265 | ||
ebec73f4 | 1266 | __pipe_unlock(pipe); |
35f3d14d JA |
1267 | return ret; |
1268 | } | |
1269 | ||
ff0c7d15 NP |
1270 | static const struct super_operations pipefs_ops = { |
1271 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1272 | .statfs = simple_statfs, |
ff0c7d15 NP |
1273 | }; |
1274 | ||
1da177e4 LT |
1275 | /* |
1276 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1277 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1278 | * any operations on the root directory. However, we need a non-trivial | |
1279 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1280 | */ | |
4fa7ec5d DH |
1281 | |
1282 | static int pipefs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1283 | { |
4fa7ec5d DH |
1284 | struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC); |
1285 | if (!ctx) | |
1286 | return -ENOMEM; | |
1287 | ctx->ops = &pipefs_ops; | |
1288 | ctx->dops = &pipefs_dentry_operations; | |
1289 | return 0; | |
1da177e4 LT |
1290 | } |
1291 | ||
1292 | static struct file_system_type pipe_fs_type = { | |
1293 | .name = "pipefs", | |
4fa7ec5d | 1294 | .init_fs_context = pipefs_init_fs_context, |
1da177e4 LT |
1295 | .kill_sb = kill_anon_super, |
1296 | }; | |
1297 | ||
1298 | static int __init init_pipe_fs(void) | |
1299 | { | |
1300 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1301 | |
1da177e4 LT |
1302 | if (!err) { |
1303 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1304 | if (IS_ERR(pipe_mnt)) { | |
1305 | err = PTR_ERR(pipe_mnt); | |
1306 | unregister_filesystem(&pipe_fs_type); | |
1307 | } | |
1308 | } | |
1309 | return err; | |
1310 | } | |
1311 | ||
1da177e4 | 1312 | fs_initcall(init_pipe_fs); |