]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/pipe.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1999 Linus Torvalds | |
6 | */ | |
7 | ||
8 | #include <linux/mm.h> | |
9 | #include <linux/file.h> | |
10 | #include <linux/poll.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/fs.h> | |
35f3d14d | 15 | #include <linux/log2.h> |
1da177e4 | 16 | #include <linux/mount.h> |
4fa7ec5d | 17 | #include <linux/pseudo_fs.h> |
b502bd11 | 18 | #include <linux/magic.h> |
1da177e4 LT |
19 | #include <linux/pipe_fs_i.h> |
20 | #include <linux/uio.h> | |
21 | #include <linux/highmem.h> | |
5274f052 | 22 | #include <linux/pagemap.h> |
db349509 | 23 | #include <linux/audit.h> |
ba719bae | 24 | #include <linux/syscalls.h> |
b492e95b | 25 | #include <linux/fcntl.h> |
d86133bd | 26 | #include <linux/memcontrol.h> |
1da177e4 | 27 | |
7c0f6ba6 | 28 | #include <linux/uaccess.h> |
1da177e4 LT |
29 | #include <asm/ioctls.h> |
30 | ||
599a0ac1 AV |
31 | #include "internal.h" |
32 | ||
b492e95b JA |
33 | /* |
34 | * The max size that a non-root user is allowed to grow the pipe. Can | |
ff9da691 | 35 | * be set by root in /proc/sys/fs/pipe-max-size |
b492e95b | 36 | */ |
ff9da691 JA |
37 | unsigned int pipe_max_size = 1048576; |
38 | ||
759c0114 WT |
39 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
40 | * matches default values. | |
41 | */ | |
42 | unsigned long pipe_user_pages_hard; | |
43 | unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; | |
44 | ||
1da177e4 | 45 | /* |
8cefc107 DH |
46 | * We use head and tail indices that aren't masked off, except at the point of |
47 | * dereference, but rather they're allowed to wrap naturally. This means there | |
48 | * isn't a dead spot in the buffer, but the ring has to be a power of two and | |
49 | * <= 2^31. | |
50 | * -- David Howells 2019-09-23. | |
51 | * | |
1da177e4 LT |
52 | * Reads with count = 0 should always return 0. |
53 | * -- Julian Bradfield 1999-06-07. | |
54 | * | |
55 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
56 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
57 | * | |
58 | * pipe_read & write cleanup | |
59 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
60 | */ | |
61 | ||
61e0d47c MS |
62 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
63 | { | |
6447a3cf | 64 | if (pipe->files) |
72b0d9aa | 65 | mutex_lock_nested(&pipe->mutex, subclass); |
61e0d47c MS |
66 | } |
67 | ||
68 | void pipe_lock(struct pipe_inode_info *pipe) | |
69 | { | |
70 | /* | |
71 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
72 | */ | |
73 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
74 | } | |
75 | EXPORT_SYMBOL(pipe_lock); | |
76 | ||
77 | void pipe_unlock(struct pipe_inode_info *pipe) | |
78 | { | |
6447a3cf | 79 | if (pipe->files) |
72b0d9aa | 80 | mutex_unlock(&pipe->mutex); |
61e0d47c MS |
81 | } |
82 | EXPORT_SYMBOL(pipe_unlock); | |
83 | ||
ebec73f4 AV |
84 | static inline void __pipe_lock(struct pipe_inode_info *pipe) |
85 | { | |
86 | mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); | |
87 | } | |
88 | ||
89 | static inline void __pipe_unlock(struct pipe_inode_info *pipe) | |
90 | { | |
91 | mutex_unlock(&pipe->mutex); | |
92 | } | |
93 | ||
61e0d47c MS |
94 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
95 | struct pipe_inode_info *pipe2) | |
96 | { | |
97 | BUG_ON(pipe1 == pipe2); | |
98 | ||
99 | if (pipe1 < pipe2) { | |
100 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
101 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
102 | } else { | |
023d43c7 PZ |
103 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
104 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
105 | } |
106 | } | |
107 | ||
1da177e4 | 108 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
3a326a2c | 109 | void pipe_wait(struct pipe_inode_info *pipe) |
1da177e4 LT |
110 | { |
111 | DEFINE_WAIT(wait); | |
112 | ||
d79fc0fc IM |
113 | /* |
114 | * Pipes are system-local resources, so sleeping on them | |
115 | * is considered a noninteractive wait: | |
116 | */ | |
af927232 | 117 | prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE); |
61e0d47c | 118 | pipe_unlock(pipe); |
1da177e4 | 119 | schedule(); |
3a326a2c | 120 | finish_wait(&pipe->wait, &wait); |
61e0d47c | 121 | pipe_lock(pipe); |
1da177e4 LT |
122 | } |
123 | ||
341b446b IM |
124 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
125 | struct pipe_buffer *buf) | |
1da177e4 LT |
126 | { |
127 | struct page *page = buf->page; | |
128 | ||
5274f052 JA |
129 | /* |
130 | * If nobody else uses this page, and we don't already have a | |
131 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 132 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 133 | */ |
341b446b | 134 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 135 | pipe->tmp_page = page; |
341b446b | 136 | else |
09cbfeaf | 137 | put_page(page); |
1da177e4 LT |
138 | } |
139 | ||
d86133bd VD |
140 | static int anon_pipe_buf_steal(struct pipe_inode_info *pipe, |
141 | struct pipe_buffer *buf) | |
142 | { | |
143 | struct page *page = buf->page; | |
144 | ||
145 | if (page_count(page) == 1) { | |
60cd4bcd | 146 | memcg_kmem_uncharge(page, 0); |
d86133bd VD |
147 | __SetPageLocked(page); |
148 | return 0; | |
149 | } | |
150 | return 1; | |
151 | } | |
152 | ||
0845718d | 153 | /** |
b51d63c6 | 154 | * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer |
0845718d JA |
155 | * @pipe: the pipe that the buffer belongs to |
156 | * @buf: the buffer to attempt to steal | |
157 | * | |
158 | * Description: | |
b51d63c6 | 159 | * This function attempts to steal the &struct page attached to |
0845718d JA |
160 | * @buf. If successful, this function returns 0 and returns with |
161 | * the page locked. The caller may then reuse the page for whatever | |
b51d63c6 | 162 | * he wishes; the typical use is insertion into a different file |
0845718d JA |
163 | * page cache. |
164 | */ | |
330ab716 JA |
165 | int generic_pipe_buf_steal(struct pipe_inode_info *pipe, |
166 | struct pipe_buffer *buf) | |
5abc97aa | 167 | { |
46e678c9 JA |
168 | struct page *page = buf->page; |
169 | ||
0845718d JA |
170 | /* |
171 | * A reference of one is golden, that means that the owner of this | |
172 | * page is the only one holding a reference to it. lock the page | |
173 | * and return OK. | |
174 | */ | |
46e678c9 | 175 | if (page_count(page) == 1) { |
46e678c9 JA |
176 | lock_page(page); |
177 | return 0; | |
178 | } | |
179 | ||
180 | return 1; | |
5abc97aa | 181 | } |
51921cb7 | 182 | EXPORT_SYMBOL(generic_pipe_buf_steal); |
5abc97aa | 183 | |
0845718d | 184 | /** |
b51d63c6 | 185 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
186 | * @pipe: the pipe that the buffer belongs to |
187 | * @buf: the buffer to get a reference to | |
188 | * | |
189 | * Description: | |
190 | * This function grabs an extra reference to @buf. It's used in | |
191 | * in the tee() system call, when we duplicate the buffers in one | |
192 | * pipe into another. | |
193 | */ | |
15fab63e | 194 | bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
70524490 | 195 | { |
15fab63e | 196 | return try_get_page(buf->page); |
70524490 | 197 | } |
51921cb7 | 198 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 199 | |
0845718d JA |
200 | /** |
201 | * generic_pipe_buf_confirm - verify contents of the pipe buffer | |
79685b8d | 202 | * @info: the pipe that the buffer belongs to |
0845718d JA |
203 | * @buf: the buffer to confirm |
204 | * | |
205 | * Description: | |
206 | * This function does nothing, because the generic pipe code uses | |
207 | * pages that are always good when inserted into the pipe. | |
208 | */ | |
cac36bb0 JA |
209 | int generic_pipe_buf_confirm(struct pipe_inode_info *info, |
210 | struct pipe_buffer *buf) | |
f84d7519 JA |
211 | { |
212 | return 0; | |
213 | } | |
51921cb7 | 214 | EXPORT_SYMBOL(generic_pipe_buf_confirm); |
f84d7519 | 215 | |
6818173b MS |
216 | /** |
217 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
218 | * @pipe: the pipe that the buffer belongs to | |
219 | * @buf: the buffer to put a reference to | |
220 | * | |
221 | * Description: | |
222 | * This function releases a reference to @buf. | |
223 | */ | |
224 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
225 | struct pipe_buffer *buf) | |
226 | { | |
09cbfeaf | 227 | put_page(buf->page); |
6818173b | 228 | } |
51921cb7 | 229 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 230 | |
01e7187b | 231 | /* New data written to a pipe may be appended to a buffer with this type. */ |
d4c3cca9 | 232 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
cac36bb0 | 233 | .confirm = generic_pipe_buf_confirm, |
1da177e4 | 234 | .release = anon_pipe_buf_release, |
d86133bd | 235 | .steal = anon_pipe_buf_steal, |
f84d7519 | 236 | .get = generic_pipe_buf_get, |
1da177e4 LT |
237 | }; |
238 | ||
a0ce2f0a | 239 | static const struct pipe_buf_operations anon_pipe_buf_nomerge_ops = { |
cac36bb0 | 240 | .confirm = generic_pipe_buf_confirm, |
1da177e4 | 241 | .release = anon_pipe_buf_release, |
d86133bd | 242 | .steal = anon_pipe_buf_steal, |
f84d7519 | 243 | .get = generic_pipe_buf_get, |
1da177e4 LT |
244 | }; |
245 | ||
9883035a | 246 | static const struct pipe_buf_operations packet_pipe_buf_ops = { |
9883035a LT |
247 | .confirm = generic_pipe_buf_confirm, |
248 | .release = anon_pipe_buf_release, | |
d86133bd | 249 | .steal = anon_pipe_buf_steal, |
9883035a LT |
250 | .get = generic_pipe_buf_get, |
251 | }; | |
252 | ||
01e7187b JH |
253 | /** |
254 | * pipe_buf_mark_unmergeable - mark a &struct pipe_buffer as unmergeable | |
255 | * @buf: the buffer to mark | |
256 | * | |
257 | * Description: | |
258 | * This function ensures that no future writes will be merged into the | |
259 | * given &struct pipe_buffer. This is necessary when multiple pipe buffers | |
260 | * share the same backing page. | |
261 | */ | |
a0ce2f0a JH |
262 | void pipe_buf_mark_unmergeable(struct pipe_buffer *buf) |
263 | { | |
264 | if (buf->ops == &anon_pipe_buf_ops) | |
265 | buf->ops = &anon_pipe_buf_nomerge_ops; | |
266 | } | |
267 | ||
01e7187b JH |
268 | static bool pipe_buf_can_merge(struct pipe_buffer *buf) |
269 | { | |
270 | return buf->ops == &anon_pipe_buf_ops; | |
271 | } | |
272 | ||
1da177e4 | 273 | static ssize_t |
fb9096a3 | 274 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 275 | { |
fb9096a3 | 276 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 277 | struct file *filp = iocb->ki_filp; |
de32ec4c | 278 | struct pipe_inode_info *pipe = filp->private_data; |
1da177e4 LT |
279 | int do_wakeup; |
280 | ssize_t ret; | |
1da177e4 | 281 | |
1da177e4 LT |
282 | /* Null read succeeds. */ |
283 | if (unlikely(total_len == 0)) | |
284 | return 0; | |
285 | ||
286 | do_wakeup = 0; | |
287 | ret = 0; | |
ebec73f4 | 288 | __pipe_lock(pipe); |
1da177e4 | 289 | for (;;) { |
8cefc107 DH |
290 | unsigned int head = pipe->head; |
291 | unsigned int tail = pipe->tail; | |
292 | unsigned int mask = pipe->ring_size - 1; | |
293 | ||
294 | if (!pipe_empty(head, tail)) { | |
295 | struct pipe_buffer *buf = &pipe->bufs[tail & mask]; | |
1da177e4 | 296 | size_t chars = buf->len; |
637b58c2 AV |
297 | size_t written; |
298 | int error; | |
1da177e4 LT |
299 | |
300 | if (chars > total_len) | |
301 | chars = total_len; | |
302 | ||
fba597db | 303 | error = pipe_buf_confirm(pipe, buf); |
f84d7519 | 304 | if (error) { |
5274f052 | 305 | if (!ret) |
e5953cbd | 306 | ret = error; |
5274f052 JA |
307 | break; |
308 | } | |
f84d7519 | 309 | |
fb9096a3 | 310 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 311 | if (unlikely(written < chars)) { |
341b446b | 312 | if (!ret) |
637b58c2 | 313 | ret = -EFAULT; |
1da177e4 LT |
314 | break; |
315 | } | |
316 | ret += chars; | |
317 | buf->offset += chars; | |
318 | buf->len -= chars; | |
9883035a LT |
319 | |
320 | /* Was it a packet buffer? Clean up and exit */ | |
321 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
322 | total_len = chars; | |
323 | buf->len = 0; | |
324 | } | |
325 | ||
1da177e4 | 326 | if (!buf->len) { |
cefa80ce | 327 | bool wake; |
a779638c | 328 | pipe_buf_release(pipe, buf); |
b667b867 | 329 | spin_lock_irq(&pipe->wait.lock); |
8cefc107 DH |
330 | tail++; |
331 | pipe->tail = tail; | |
8446487f | 332 | do_wakeup = 1; |
cefa80ce DH |
333 | wake = head - (tail - 1) == pipe->max_usage / 2; |
334 | if (wake) | |
3c0edea9 | 335 | wake_up_locked_poll( |
8446487f | 336 | &pipe->wait, EPOLLOUT | EPOLLWRNORM); |
b667b867 | 337 | spin_unlock_irq(&pipe->wait.lock); |
cefa80ce | 338 | if (wake) |
8446487f | 339 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 LT |
340 | } |
341 | total_len -= chars; | |
342 | if (!total_len) | |
343 | break; /* common path: read succeeded */ | |
8cefc107 DH |
344 | if (!pipe_empty(head, tail)) /* More to do? */ |
345 | continue; | |
1da177e4 | 346 | } |
8cefc107 | 347 | |
923f4f23 | 348 | if (!pipe->writers) |
1da177e4 | 349 | break; |
923f4f23 | 350 | if (!pipe->waiting_writers) { |
1da177e4 LT |
351 | /* syscall merging: Usually we must not sleep |
352 | * if O_NONBLOCK is set, or if we got some data. | |
353 | * But if a writer sleeps in kernel space, then | |
354 | * we can wait for that data without violating POSIX. | |
355 | */ | |
356 | if (ret) | |
357 | break; | |
358 | if (filp->f_flags & O_NONBLOCK) { | |
359 | ret = -EAGAIN; | |
360 | break; | |
361 | } | |
362 | } | |
363 | if (signal_pending(current)) { | |
341b446b IM |
364 | if (!ret) |
365 | ret = -ERESTARTSYS; | |
1da177e4 LT |
366 | break; |
367 | } | |
923f4f23 | 368 | pipe_wait(pipe); |
1da177e4 | 369 | } |
ebec73f4 | 370 | __pipe_unlock(pipe); |
341b446b IM |
371 | |
372 | /* Signal writers asynchronously that there is more room. */ | |
1da177e4 | 373 | if (do_wakeup) { |
3c0edea9 | 374 | wake_up_interruptible_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM); |
923f4f23 | 375 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 LT |
376 | } |
377 | if (ret > 0) | |
378 | file_accessed(filp); | |
379 | return ret; | |
380 | } | |
381 | ||
9883035a LT |
382 | static inline int is_packetized(struct file *file) |
383 | { | |
384 | return (file->f_flags & O_DIRECT) != 0; | |
385 | } | |
386 | ||
1da177e4 | 387 | static ssize_t |
f0d1bec9 | 388 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 389 | { |
ee0b3e67 | 390 | struct file *filp = iocb->ki_filp; |
de32ec4c | 391 | struct pipe_inode_info *pipe = filp->private_data; |
a194dfe6 | 392 | unsigned int head, max_usage, mask; |
f0d1bec9 AV |
393 | ssize_t ret = 0; |
394 | int do_wakeup = 0; | |
395 | size_t total_len = iov_iter_count(from); | |
1da177e4 LT |
396 | ssize_t chars; |
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 | ||
8cefc107 | 410 | head = pipe->head; |
6718b6f8 | 411 | max_usage = pipe->max_usage; |
8cefc107 DH |
412 | mask = pipe->ring_size - 1; |
413 | ||
1da177e4 LT |
414 | /* We try to merge small writes */ |
415 | chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */ | |
a194dfe6 | 416 | if (!pipe_empty(head, pipe->tail) && chars != 0) { |
8cefc107 | 417 | struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; |
1da177e4 | 418 | int offset = buf->offset + buf->len; |
341b446b | 419 | |
01e7187b | 420 | if (pipe_buf_can_merge(buf) && offset + chars <= PAGE_SIZE) { |
fba597db | 421 | ret = pipe_buf_confirm(pipe, buf); |
6ae08069 | 422 | if (ret) |
5274f052 | 423 | goto out; |
f84d7519 | 424 | |
f0d1bec9 AV |
425 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
426 | if (unlikely(ret < chars)) { | |
6ae08069 | 427 | ret = -EFAULT; |
1da177e4 | 428 | goto out; |
f6762b7a | 429 | } |
f0d1bec9 | 430 | do_wakeup = 1; |
6ae08069 | 431 | buf->len += ret; |
f0d1bec9 | 432 | if (!iov_iter_count(from)) |
1da177e4 LT |
433 | goto out; |
434 | } | |
435 | } | |
436 | ||
437 | for (;;) { | |
923f4f23 | 438 | if (!pipe->readers) { |
1da177e4 | 439 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
440 | if (!ret) |
441 | ret = -EPIPE; | |
1da177e4 LT |
442 | break; |
443 | } | |
8cefc107 | 444 | |
a194dfe6 DH |
445 | head = pipe->head; |
446 | if (!pipe_full(head, pipe->tail, max_usage)) { | |
8cefc107 | 447 | struct pipe_buffer *buf = &pipe->bufs[head & mask]; |
923f4f23 | 448 | struct page *page = pipe->tmp_page; |
f0d1bec9 | 449 | int copied; |
1da177e4 LT |
450 | |
451 | if (!page) { | |
d86133bd | 452 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
453 | if (unlikely(!page)) { |
454 | ret = ret ? : -ENOMEM; | |
455 | break; | |
456 | } | |
923f4f23 | 457 | pipe->tmp_page = page; |
1da177e4 | 458 | } |
a194dfe6 DH |
459 | |
460 | /* Allocate a slot in the ring in advance and attach an | |
461 | * empty buffer. If we fault or otherwise fail to use | |
462 | * it, either the reader will consume it or it'll still | |
463 | * be there for the next write. | |
464 | */ | |
465 | spin_lock_irq(&pipe->wait.lock); | |
466 | ||
467 | head = pipe->head; | |
8df44129 DH |
468 | if (pipe_full(head, pipe->tail, max_usage)) { |
469 | spin_unlock_irq(&pipe->wait.lock); | |
470 | continue; | |
471 | } | |
472 | ||
a194dfe6 DH |
473 | pipe->head = head + 1; |
474 | ||
341b446b | 475 | /* Always wake up, even if the copy fails. Otherwise |
1da177e4 LT |
476 | * we lock up (O_NONBLOCK-)readers that sleep due to |
477 | * syscall merging. | |
478 | * FIXME! Is this really true? | |
479 | */ | |
3c0edea9 | 480 | wake_up_locked_poll( |
a194dfe6 DH |
481 | &pipe->wait, EPOLLIN | EPOLLRDNORM); |
482 | ||
483 | spin_unlock_irq(&pipe->wait.lock); | |
484 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
1da177e4 LT |
485 | |
486 | /* Insert it into the buffer array */ | |
a194dfe6 | 487 | buf = &pipe->bufs[head & mask]; |
1da177e4 LT |
488 | buf->page = page; |
489 | buf->ops = &anon_pipe_buf_ops; | |
490 | buf->offset = 0; | |
a194dfe6 | 491 | buf->len = 0; |
9883035a LT |
492 | buf->flags = 0; |
493 | if (is_packetized(filp)) { | |
494 | buf->ops = &packet_pipe_buf_ops; | |
495 | buf->flags = PIPE_BUF_FLAG_PACKET; | |
496 | } | |
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 | |
a194dfe6 | 513 | if (!pipe_full(head, pipe->tail, 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 | } | |
923f4f23 IM |
527 | pipe->waiting_writers++; |
528 | pipe_wait(pipe); | |
529 | pipe->waiting_writers--; | |
1da177e4 LT |
530 | } |
531 | out: | |
ebec73f4 | 532 | __pipe_unlock(pipe); |
1da177e4 | 533 | if (do_wakeup) { |
3c0edea9 | 534 | wake_up_interruptible_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM); |
923f4f23 | 535 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 | 536 | } |
7e775f46 | 537 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
538 | int err = file_update_time(filp); |
539 | if (err) | |
540 | ret = err; | |
7e775f46 | 541 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 542 | } |
1da177e4 LT |
543 | return ret; |
544 | } | |
545 | ||
d59d0b1b | 546 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 547 | { |
de32ec4c | 548 | struct pipe_inode_info *pipe = filp->private_data; |
8cefc107 | 549 | int count, head, tail, mask; |
1da177e4 LT |
550 | |
551 | switch (cmd) { | |
552 | case FIONREAD: | |
ebec73f4 | 553 | __pipe_lock(pipe); |
1da177e4 | 554 | count = 0; |
8cefc107 DH |
555 | head = pipe->head; |
556 | tail = pipe->tail; | |
557 | mask = pipe->ring_size - 1; | |
558 | ||
559 | while (tail != head) { | |
560 | count += pipe->bufs[tail & mask].len; | |
561 | tail++; | |
1da177e4 | 562 | } |
ebec73f4 | 563 | __pipe_unlock(pipe); |
923f4f23 | 564 | |
1da177e4 LT |
565 | return put_user(count, (int __user *)arg); |
566 | default: | |
46ce341b | 567 | return -ENOIOCTLCMD; |
1da177e4 LT |
568 | } |
569 | } | |
570 | ||
dd67081b | 571 | /* No kernel lock held - fine */ |
a11e1d43 LT |
572 | static __poll_t |
573 | pipe_poll(struct file *filp, poll_table *wait) | |
dd67081b | 574 | { |
a11e1d43 | 575 | __poll_t mask; |
dd67081b | 576 | struct pipe_inode_info *pipe = filp->private_data; |
8cefc107 DH |
577 | unsigned int head = READ_ONCE(pipe->head); |
578 | unsigned int tail = READ_ONCE(pipe->tail); | |
a11e1d43 LT |
579 | |
580 | poll_wait(filp, &pipe->wait, wait); | |
1da177e4 | 581 | |
8cefc107 DH |
582 | BUG_ON(pipe_occupancy(head, tail) > pipe->ring_size); |
583 | ||
1da177e4 | 584 | /* Reading only -- no need for acquiring the semaphore. */ |
a11e1d43 | 585 | mask = 0; |
1da177e4 | 586 | if (filp->f_mode & FMODE_READ) { |
8cefc107 DH |
587 | if (!pipe_empty(head, tail)) |
588 | mask |= EPOLLIN | EPOLLRDNORM; | |
923f4f23 | 589 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
a9a08845 | 590 | mask |= EPOLLHUP; |
1da177e4 LT |
591 | } |
592 | ||
593 | if (filp->f_mode & FMODE_WRITE) { | |
6718b6f8 | 594 | if (!pipe_full(head, tail, pipe->max_usage)) |
8cefc107 | 595 | mask |= EPOLLOUT | EPOLLWRNORM; |
5e5d7a22 | 596 | /* |
a9a08845 | 597 | * Most Unices do not set EPOLLERR for FIFOs but on Linux they |
5e5d7a22 PE |
598 | * behave exactly like pipes for poll(). |
599 | */ | |
923f4f23 | 600 | if (!pipe->readers) |
a9a08845 | 601 | mask |= EPOLLERR; |
1da177e4 LT |
602 | } |
603 | ||
604 | return mask; | |
605 | } | |
606 | ||
b0d8d229 LT |
607 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
608 | { | |
609 | int kill = 0; | |
610 | ||
611 | spin_lock(&inode->i_lock); | |
612 | if (!--pipe->files) { | |
613 | inode->i_pipe = NULL; | |
614 | kill = 1; | |
615 | } | |
616 | spin_unlock(&inode->i_lock); | |
617 | ||
618 | if (kill) | |
619 | free_pipe_info(pipe); | |
620 | } | |
621 | ||
1da177e4 | 622 | static int |
599a0ac1 | 623 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 624 | { |
b0d8d229 | 625 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 626 | |
ebec73f4 | 627 | __pipe_lock(pipe); |
599a0ac1 AV |
628 | if (file->f_mode & FMODE_READ) |
629 | pipe->readers--; | |
630 | if (file->f_mode & FMODE_WRITE) | |
631 | pipe->writers--; | |
341b446b | 632 | |
ba5bb147 | 633 | if (pipe->readers || pipe->writers) { |
a9a08845 | 634 | wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP); |
923f4f23 IM |
635 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
636 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 637 | } |
ebec73f4 | 638 | __pipe_unlock(pipe); |
ba5bb147 | 639 | |
b0d8d229 | 640 | put_pipe_info(inode, pipe); |
1da177e4 LT |
641 | return 0; |
642 | } | |
643 | ||
644 | static int | |
599a0ac1 | 645 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 646 | { |
de32ec4c | 647 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 648 | int retval = 0; |
1da177e4 | 649 | |
ebec73f4 | 650 | __pipe_lock(pipe); |
599a0ac1 AV |
651 | if (filp->f_mode & FMODE_READ) |
652 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
653 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 654 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
655 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
656 | /* this can happen only if on == T */ | |
e5bc49ba ON |
657 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
658 | } | |
ebec73f4 | 659 | __pipe_unlock(pipe); |
60aa4924 | 660 | return retval; |
1da177e4 LT |
661 | } |
662 | ||
9c87bcf0 | 663 | static unsigned long account_pipe_buffers(struct user_struct *user, |
759c0114 WT |
664 | unsigned long old, unsigned long new) |
665 | { | |
9c87bcf0 | 666 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
667 | } |
668 | ||
9c87bcf0 | 669 | static bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 670 | { |
f7340761 EB |
671 | unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft); |
672 | ||
673 | return soft_limit && user_bufs > soft_limit; | |
759c0114 WT |
674 | } |
675 | ||
9c87bcf0 | 676 | static bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 677 | { |
f7340761 EB |
678 | unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard); |
679 | ||
680 | return hard_limit && user_bufs > hard_limit; | |
759c0114 WT |
681 | } |
682 | ||
85c2dd54 EB |
683 | static bool is_unprivileged_user(void) |
684 | { | |
685 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
686 | } | |
687 | ||
7bee130e | 688 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 689 | { |
923f4f23 | 690 | struct pipe_inode_info *pipe; |
09b4d199 MK |
691 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
692 | struct user_struct *user = get_current_user(); | |
9c87bcf0 | 693 | unsigned long user_bufs; |
f7340761 | 694 | unsigned int max_size = READ_ONCE(pipe_max_size); |
3a326a2c | 695 | |
d86133bd | 696 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
09b4d199 MK |
697 | if (pipe == NULL) |
698 | goto out_free_uid; | |
699 | ||
f7340761 EB |
700 | if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE)) |
701 | pipe_bufs = max_size >> PAGE_SHIFT; | |
086e774a | 702 | |
9c87bcf0 | 703 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
a005ca0e | 704 | |
85c2dd54 | 705 | if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) { |
9c87bcf0 | 706 | user_bufs = account_pipe_buffers(user, pipe_bufs, 1); |
a005ca0e | 707 | pipe_bufs = 1; |
09b4d199 | 708 | } |
759c0114 | 709 | |
85c2dd54 | 710 | if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user()) |
a005ca0e MK |
711 | goto out_revert_acct; |
712 | ||
713 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), | |
714 | GFP_KERNEL_ACCOUNT); | |
715 | ||
09b4d199 MK |
716 | if (pipe->bufs) { |
717 | init_waitqueue_head(&pipe->wait); | |
718 | pipe->r_counter = pipe->w_counter = 1; | |
6718b6f8 | 719 | pipe->max_usage = pipe_bufs; |
8cefc107 | 720 | pipe->ring_size = pipe_bufs; |
09b4d199 | 721 | pipe->user = user; |
09b4d199 MK |
722 | mutex_init(&pipe->mutex); |
723 | return pipe; | |
3a326a2c IM |
724 | } |
725 | ||
a005ca0e | 726 | out_revert_acct: |
9c87bcf0 | 727 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
09b4d199 MK |
728 | kfree(pipe); |
729 | out_free_uid: | |
730 | free_uid(user); | |
35f3d14d | 731 | return NULL; |
3a326a2c IM |
732 | } |
733 | ||
4b8a8f1e | 734 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
735 | { |
736 | int i; | |
1da177e4 | 737 | |
8cefc107 | 738 | (void) account_pipe_buffers(pipe->user, pipe->ring_size, 0); |
759c0114 | 739 | free_uid(pipe->user); |
8cefc107 | 740 | for (i = 0; i < pipe->ring_size; i++) { |
923f4f23 | 741 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 742 | if (buf->ops) |
a779638c | 743 | pipe_buf_release(pipe, buf); |
1da177e4 | 744 | } |
923f4f23 IM |
745 | if (pipe->tmp_page) |
746 | __free_page(pipe->tmp_page); | |
35f3d14d | 747 | kfree(pipe->bufs); |
923f4f23 | 748 | kfree(pipe); |
1da177e4 LT |
749 | } |
750 | ||
fa3536cc | 751 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 752 | |
c23fbb6b ED |
753 | /* |
754 | * pipefs_dname() is called from d_path(). | |
755 | */ | |
756 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
757 | { | |
758 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
75c3cfa8 | 759 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
760 | } |
761 | ||
3ba13d17 | 762 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 763 | .d_dname = pipefs_dname, |
1da177e4 LT |
764 | }; |
765 | ||
766 | static struct inode * get_pipe_inode(void) | |
767 | { | |
a209dfc7 | 768 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 769 | struct pipe_inode_info *pipe; |
1da177e4 LT |
770 | |
771 | if (!inode) | |
772 | goto fail_inode; | |
773 | ||
85fe4025 CH |
774 | inode->i_ino = get_next_ino(); |
775 | ||
7bee130e | 776 | pipe = alloc_pipe_info(); |
923f4f23 | 777 | if (!pipe) |
1da177e4 | 778 | goto fail_iput; |
3a326a2c | 779 | |
ba5bb147 AV |
780 | inode->i_pipe = pipe; |
781 | pipe->files = 2; | |
923f4f23 | 782 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 783 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
784 | |
785 | /* | |
786 | * Mark the inode dirty from the very beginning, | |
787 | * that way it will never be moved to the dirty | |
788 | * list because "mark_inode_dirty()" will think | |
789 | * that it already _is_ on the dirty list. | |
790 | */ | |
791 | inode->i_state = I_DIRTY; | |
792 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
793 | inode->i_uid = current_fsuid(); |
794 | inode->i_gid = current_fsgid(); | |
078cd827 | 795 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
923f4f23 | 796 | |
1da177e4 LT |
797 | return inode; |
798 | ||
799 | fail_iput: | |
800 | iput(inode); | |
341b446b | 801 | |
1da177e4 LT |
802 | fail_inode: |
803 | return NULL; | |
804 | } | |
805 | ||
e4fad8e5 | 806 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 807 | { |
e4fad8e5 | 808 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 809 | struct file *f; |
1da177e4 | 810 | |
1da177e4 | 811 | if (!inode) |
e4fad8e5 | 812 | return -ENFILE; |
1da177e4 | 813 | |
152b6372 AV |
814 | f = alloc_file_pseudo(inode, pipe_mnt, "", |
815 | O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)), | |
816 | &pipefifo_fops); | |
e9bb1f9b | 817 | if (IS_ERR(f)) { |
152b6372 AV |
818 | free_pipe_info(inode->i_pipe); |
819 | iput(inode); | |
820 | return PTR_ERR(f); | |
e9bb1f9b | 821 | } |
341b446b | 822 | |
de32ec4c | 823 | f->private_data = inode->i_pipe; |
d6cbd281 | 824 | |
183266f2 AV |
825 | res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK), |
826 | &pipefifo_fops); | |
e9bb1f9b | 827 | if (IS_ERR(res[0])) { |
b10a4a9f AV |
828 | put_pipe_info(inode, inode->i_pipe); |
829 | fput(f); | |
830 | return PTR_ERR(res[0]); | |
e9bb1f9b | 831 | } |
de32ec4c | 832 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 AV |
833 | res[1] = f; |
834 | return 0; | |
d6cbd281 AK |
835 | } |
836 | ||
5b249b1b | 837 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 838 | { |
d6cbd281 AK |
839 | int error; |
840 | int fdw, fdr; | |
841 | ||
9883035a | 842 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT)) |
ed8cae8b UD |
843 | return -EINVAL; |
844 | ||
e4fad8e5 AV |
845 | error = create_pipe_files(files, flags); |
846 | if (error) | |
847 | return error; | |
d6cbd281 | 848 | |
ed8cae8b | 849 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
850 | if (error < 0) |
851 | goto err_read_pipe; | |
852 | fdr = error; | |
853 | ||
ed8cae8b | 854 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
855 | if (error < 0) |
856 | goto err_fdr; | |
857 | fdw = error; | |
858 | ||
157cf649 | 859 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
860 | fd[0] = fdr; |
861 | fd[1] = fdw; | |
d6cbd281 AK |
862 | return 0; |
863 | ||
864 | err_fdr: | |
865 | put_unused_fd(fdr); | |
866 | err_read_pipe: | |
e4fad8e5 AV |
867 | fput(files[0]); |
868 | fput(files[1]); | |
d6cbd281 | 869 | return error; |
1da177e4 LT |
870 | } |
871 | ||
5b249b1b AV |
872 | int do_pipe_flags(int *fd, int flags) |
873 | { | |
874 | struct file *files[2]; | |
875 | int error = __do_pipe_flags(fd, files, flags); | |
876 | if (!error) { | |
877 | fd_install(fd[0], files[0]); | |
878 | fd_install(fd[1], files[1]); | |
879 | } | |
880 | return error; | |
881 | } | |
882 | ||
d35c7b0e UD |
883 | /* |
884 | * sys_pipe() is the normal C calling standard for creating | |
885 | * a pipe. It's not the way Unix traditionally does this, though. | |
886 | */ | |
0a216dd1 | 887 | static int do_pipe2(int __user *fildes, int flags) |
d35c7b0e | 888 | { |
5b249b1b | 889 | struct file *files[2]; |
d35c7b0e UD |
890 | int fd[2]; |
891 | int error; | |
892 | ||
5b249b1b | 893 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 894 | if (!error) { |
5b249b1b AV |
895 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
896 | fput(files[0]); | |
897 | fput(files[1]); | |
898 | put_unused_fd(fd[0]); | |
899 | put_unused_fd(fd[1]); | |
d35c7b0e | 900 | error = -EFAULT; |
5b249b1b AV |
901 | } else { |
902 | fd_install(fd[0], files[0]); | |
903 | fd_install(fd[1], files[1]); | |
ba719bae | 904 | } |
d35c7b0e UD |
905 | } |
906 | return error; | |
907 | } | |
908 | ||
0a216dd1 DB |
909 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
910 | { | |
911 | return do_pipe2(fildes, flags); | |
912 | } | |
913 | ||
2b664219 | 914 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b | 915 | { |
0a216dd1 | 916 | return do_pipe2(fildes, 0); |
ed8cae8b UD |
917 | } |
918 | ||
fc7478a2 | 919 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 | 920 | { |
8cefc107 | 921 | int cur = *cnt; |
f776c738 AV |
922 | |
923 | while (cur == *cnt) { | |
fc7478a2 | 924 | pipe_wait(pipe); |
f776c738 AV |
925 | if (signal_pending(current)) |
926 | break; | |
927 | } | |
928 | return cur == *cnt ? -ERESTARTSYS : 0; | |
929 | } | |
930 | ||
fc7478a2 | 931 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 932 | { |
fc7478a2 | 933 | wake_up_interruptible(&pipe->wait); |
f776c738 AV |
934 | } |
935 | ||
936 | static int fifo_open(struct inode *inode, struct file *filp) | |
937 | { | |
938 | struct pipe_inode_info *pipe; | |
599a0ac1 | 939 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
940 | int ret; |
941 | ||
ba5bb147 AV |
942 | filp->f_version = 0; |
943 | ||
944 | spin_lock(&inode->i_lock); | |
945 | if (inode->i_pipe) { | |
946 | pipe = inode->i_pipe; | |
947 | pipe->files++; | |
948 | spin_unlock(&inode->i_lock); | |
949 | } else { | |
950 | spin_unlock(&inode->i_lock); | |
7bee130e | 951 | pipe = alloc_pipe_info(); |
f776c738 | 952 | if (!pipe) |
ba5bb147 AV |
953 | return -ENOMEM; |
954 | pipe->files = 1; | |
955 | spin_lock(&inode->i_lock); | |
956 | if (unlikely(inode->i_pipe)) { | |
957 | inode->i_pipe->files++; | |
958 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 959 | free_pipe_info(pipe); |
ba5bb147 AV |
960 | pipe = inode->i_pipe; |
961 | } else { | |
962 | inode->i_pipe = pipe; | |
963 | spin_unlock(&inode->i_lock); | |
964 | } | |
f776c738 | 965 | } |
de32ec4c | 966 | filp->private_data = pipe; |
ba5bb147 AV |
967 | /* OK, we have a pipe and it's pinned down */ |
968 | ||
ebec73f4 | 969 | __pipe_lock(pipe); |
f776c738 AV |
970 | |
971 | /* We can only do regular read/write on fifos */ | |
972 | filp->f_mode &= (FMODE_READ | FMODE_WRITE); | |
973 | ||
974 | switch (filp->f_mode) { | |
975 | case FMODE_READ: | |
976 | /* | |
977 | * O_RDONLY | |
978 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
979 | * opened, even when there is no process writing the FIFO. | |
980 | */ | |
f776c738 AV |
981 | pipe->r_counter++; |
982 | if (pipe->readers++ == 0) | |
fc7478a2 | 983 | wake_up_partner(pipe); |
f776c738 | 984 | |
599a0ac1 | 985 | if (!is_pipe && !pipe->writers) { |
f776c738 | 986 | if ((filp->f_flags & O_NONBLOCK)) { |
a9a08845 | 987 | /* suppress EPOLLHUP until we have |
f776c738 AV |
988 | * seen a writer */ |
989 | filp->f_version = pipe->w_counter; | |
990 | } else { | |
fc7478a2 | 991 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
992 | goto err_rd; |
993 | } | |
994 | } | |
995 | break; | |
8cefc107 | 996 | |
f776c738 AV |
997 | case FMODE_WRITE: |
998 | /* | |
999 | * O_WRONLY | |
1000 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
1001 | * errno=ENXIO when there is no process reading the FIFO. | |
1002 | */ | |
1003 | ret = -ENXIO; | |
599a0ac1 | 1004 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
1005 | goto err; |
1006 | ||
f776c738 AV |
1007 | pipe->w_counter++; |
1008 | if (!pipe->writers++) | |
fc7478a2 | 1009 | wake_up_partner(pipe); |
f776c738 | 1010 | |
599a0ac1 | 1011 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 1012 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
1013 | goto err_wr; |
1014 | } | |
1015 | break; | |
8cefc107 | 1016 | |
f776c738 AV |
1017 | case FMODE_READ | FMODE_WRITE: |
1018 | /* | |
1019 | * O_RDWR | |
1020 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
1021 | * This implementation will NEVER block on a O_RDWR open, since | |
1022 | * the process can at least talk to itself. | |
1023 | */ | |
f776c738 AV |
1024 | |
1025 | pipe->readers++; | |
1026 | pipe->writers++; | |
1027 | pipe->r_counter++; | |
1028 | pipe->w_counter++; | |
1029 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 1030 | wake_up_partner(pipe); |
f776c738 AV |
1031 | break; |
1032 | ||
1033 | default: | |
1034 | ret = -EINVAL; | |
1035 | goto err; | |
1036 | } | |
1037 | ||
1038 | /* Ok! */ | |
ebec73f4 | 1039 | __pipe_unlock(pipe); |
f776c738 AV |
1040 | return 0; |
1041 | ||
1042 | err_rd: | |
1043 | if (!--pipe->readers) | |
1044 | wake_up_interruptible(&pipe->wait); | |
1045 | ret = -ERESTARTSYS; | |
1046 | goto err; | |
1047 | ||
1048 | err_wr: | |
1049 | if (!--pipe->writers) | |
1050 | wake_up_interruptible(&pipe->wait); | |
1051 | ret = -ERESTARTSYS; | |
1052 | goto err; | |
1053 | ||
1054 | err: | |
ebec73f4 | 1055 | __pipe_unlock(pipe); |
b0d8d229 LT |
1056 | |
1057 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1058 | return ret; |
1059 | } | |
1060 | ||
599a0ac1 AV |
1061 | const struct file_operations pipefifo_fops = { |
1062 | .open = fifo_open, | |
1063 | .llseek = no_llseek, | |
fb9096a3 | 1064 | .read_iter = pipe_read, |
f0d1bec9 | 1065 | .write_iter = pipe_write, |
a11e1d43 | 1066 | .poll = pipe_poll, |
599a0ac1 AV |
1067 | .unlocked_ioctl = pipe_ioctl, |
1068 | .release = pipe_release, | |
1069 | .fasync = pipe_fasync, | |
f776c738 AV |
1070 | }; |
1071 | ||
f491bd71 MK |
1072 | /* |
1073 | * Currently we rely on the pipe array holding a power-of-2 number | |
d3f14c48 | 1074 | * of pages. Returns 0 on error. |
f491bd71 | 1075 | */ |
96e99be4 | 1076 | unsigned int round_pipe_size(unsigned long size) |
f491bd71 | 1077 | { |
c4fed5a9 | 1078 | if (size > (1U << 31)) |
96e99be4 EB |
1079 | return 0; |
1080 | ||
4c2e4bef EB |
1081 | /* Minimum pipe size, as required by POSIX */ |
1082 | if (size < PAGE_SIZE) | |
c4fed5a9 | 1083 | return PAGE_SIZE; |
d3f14c48 | 1084 | |
c4fed5a9 | 1085 | return roundup_pow_of_two(size); |
f491bd71 MK |
1086 | } |
1087 | ||
35f3d14d JA |
1088 | /* |
1089 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1090 | * pipe size if successful, or return -ERROR on error. | |
1091 | */ | |
d37d4166 | 1092 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) |
35f3d14d JA |
1093 | { |
1094 | struct pipe_buffer *bufs; | |
8cefc107 | 1095 | unsigned int size, nr_slots, head, tail, mask, n; |
9c87bcf0 | 1096 | unsigned long user_bufs; |
b0b91d18 | 1097 | long ret = 0; |
d37d4166 MK |
1098 | |
1099 | size = round_pipe_size(arg); | |
8cefc107 | 1100 | nr_slots = size >> PAGE_SHIFT; |
d37d4166 | 1101 | |
8cefc107 | 1102 | if (!nr_slots) |
d37d4166 MK |
1103 | return -EINVAL; |
1104 | ||
b0b91d18 MK |
1105 | /* |
1106 | * If trying to increase the pipe capacity, check that an | |
1107 | * unprivileged user is not trying to exceed various limits | |
1108 | * (soft limit check here, hard limit check just below). | |
1109 | * Decreasing the pipe capacity is always permitted, even | |
1110 | * if the user is currently over a limit. | |
1111 | */ | |
8cefc107 | 1112 | if (nr_slots > pipe->ring_size && |
b0b91d18 | 1113 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) |
d37d4166 MK |
1114 | return -EPERM; |
1115 | ||
8cefc107 | 1116 | user_bufs = account_pipe_buffers(pipe->user, pipe->ring_size, nr_slots); |
b0b91d18 | 1117 | |
8cefc107 | 1118 | if (nr_slots > pipe->ring_size && |
9c87bcf0 MK |
1119 | (too_many_pipe_buffers_hard(user_bufs) || |
1120 | too_many_pipe_buffers_soft(user_bufs)) && | |
85c2dd54 | 1121 | is_unprivileged_user()) { |
b0b91d18 MK |
1122 | ret = -EPERM; |
1123 | goto out_revert_acct; | |
1124 | } | |
35f3d14d | 1125 | |
35f3d14d | 1126 | /* |
8cefc107 DH |
1127 | * We can shrink the pipe, if arg is greater than the ring occupancy. |
1128 | * Since we don't expect a lot of shrink+grow operations, just free and | |
1129 | * allocate again like we would do for growing. If the pipe currently | |
35f3d14d JA |
1130 | * contains more buffers than arg, then return busy. |
1131 | */ | |
8cefc107 DH |
1132 | mask = pipe->ring_size - 1; |
1133 | head = pipe->head; | |
1134 | tail = pipe->tail; | |
1135 | n = pipe_occupancy(pipe->head, pipe->tail); | |
1136 | if (nr_slots < n) { | |
b0b91d18 MK |
1137 | ret = -EBUSY; |
1138 | goto out_revert_acct; | |
1139 | } | |
35f3d14d | 1140 | |
8cefc107 | 1141 | bufs = kcalloc(nr_slots, sizeof(*bufs), |
d86133bd | 1142 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); |
b0b91d18 MK |
1143 | if (unlikely(!bufs)) { |
1144 | ret = -ENOMEM; | |
1145 | goto out_revert_acct; | |
1146 | } | |
35f3d14d JA |
1147 | |
1148 | /* | |
1149 | * The pipe array wraps around, so just start the new one at zero | |
8cefc107 | 1150 | * and adjust the indices. |
35f3d14d | 1151 | */ |
8cefc107 DH |
1152 | if (n > 0) { |
1153 | unsigned int h = head & mask; | |
1154 | unsigned int t = tail & mask; | |
1155 | if (h > t) { | |
1156 | memcpy(bufs, pipe->bufs + t, | |
1157 | n * sizeof(struct pipe_buffer)); | |
1158 | } else { | |
1159 | unsigned int tsize = pipe->ring_size - t; | |
1160 | if (h > 0) | |
1161 | memcpy(bufs + tsize, pipe->bufs, | |
1162 | h * sizeof(struct pipe_buffer)); | |
1163 | memcpy(bufs, pipe->bufs + t, | |
1164 | tsize * sizeof(struct pipe_buffer)); | |
1165 | } | |
35f3d14d JA |
1166 | } |
1167 | ||
8cefc107 DH |
1168 | head = n; |
1169 | tail = 0; | |
1170 | ||
35f3d14d JA |
1171 | kfree(pipe->bufs); |
1172 | pipe->bufs = bufs; | |
8cefc107 | 1173 | pipe->ring_size = nr_slots; |
6718b6f8 | 1174 | pipe->max_usage = nr_slots; |
8cefc107 DH |
1175 | pipe->tail = tail; |
1176 | pipe->head = head; | |
6718b6f8 | 1177 | return pipe->max_usage * PAGE_SIZE; |
b0b91d18 MK |
1178 | |
1179 | out_revert_acct: | |
8cefc107 | 1180 | (void) account_pipe_buffers(pipe->user, nr_slots, pipe->ring_size); |
b0b91d18 | 1181 | return ret; |
35f3d14d JA |
1182 | } |
1183 | ||
72083646 LT |
1184 | /* |
1185 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1186 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1187 | * pipe. | |
1188 | */ | |
1189 | struct pipe_inode_info *get_pipe_info(struct file *file) | |
1190 | { | |
de32ec4c | 1191 | return file->f_op == &pipefifo_fops ? file->private_data : NULL; |
72083646 LT |
1192 | } |
1193 | ||
35f3d14d JA |
1194 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
1195 | { | |
1196 | struct pipe_inode_info *pipe; | |
1197 | long ret; | |
1198 | ||
c66fb347 | 1199 | pipe = get_pipe_info(file); |
35f3d14d JA |
1200 | if (!pipe) |
1201 | return -EBADF; | |
1202 | ||
ebec73f4 | 1203 | __pipe_lock(pipe); |
35f3d14d JA |
1204 | |
1205 | switch (cmd) { | |
d37d4166 MK |
1206 | case F_SETPIPE_SZ: |
1207 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1208 | break; |
1209 | case F_GETPIPE_SZ: | |
6718b6f8 | 1210 | ret = pipe->max_usage * PAGE_SIZE; |
35f3d14d JA |
1211 | break; |
1212 | default: | |
1213 | ret = -EINVAL; | |
1214 | break; | |
1215 | } | |
1216 | ||
ebec73f4 | 1217 | __pipe_unlock(pipe); |
35f3d14d JA |
1218 | return ret; |
1219 | } | |
1220 | ||
ff0c7d15 NP |
1221 | static const struct super_operations pipefs_ops = { |
1222 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1223 | .statfs = simple_statfs, |
ff0c7d15 NP |
1224 | }; |
1225 | ||
1da177e4 LT |
1226 | /* |
1227 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1228 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1229 | * any operations on the root directory. However, we need a non-trivial | |
1230 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1231 | */ | |
4fa7ec5d DH |
1232 | |
1233 | static int pipefs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1234 | { |
4fa7ec5d DH |
1235 | struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC); |
1236 | if (!ctx) | |
1237 | return -ENOMEM; | |
1238 | ctx->ops = &pipefs_ops; | |
1239 | ctx->dops = &pipefs_dentry_operations; | |
1240 | return 0; | |
1da177e4 LT |
1241 | } |
1242 | ||
1243 | static struct file_system_type pipe_fs_type = { | |
1244 | .name = "pipefs", | |
4fa7ec5d | 1245 | .init_fs_context = pipefs_init_fs_context, |
1da177e4 LT |
1246 | .kill_sb = kill_anon_super, |
1247 | }; | |
1248 | ||
1249 | static int __init init_pipe_fs(void) | |
1250 | { | |
1251 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1252 | |
1da177e4 LT |
1253 | if (!err) { |
1254 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1255 | if (IS_ERR(pipe_mnt)) { | |
1256 | err = PTR_ERR(pipe_mnt); | |
1257 | unregister_filesystem(&pipe_fs_type); | |
1258 | } | |
1259 | } | |
1260 | return err; | |
1261 | } | |
1262 | ||
1da177e4 | 1263 | fs_initcall(init_pipe_fs); |