]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/fs/pipe.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1999 Linus Torvalds | |
6 | */ | |
7 | ||
8 | #include <linux/mm.h> | |
9 | #include <linux/file.h> | |
10 | #include <linux/poll.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/fs.h> | |
35f3d14d | 15 | #include <linux/log2.h> |
1da177e4 | 16 | #include <linux/mount.h> |
4fa7ec5d | 17 | #include <linux/pseudo_fs.h> |
b502bd11 | 18 | #include <linux/magic.h> |
1da177e4 LT |
19 | #include <linux/pipe_fs_i.h> |
20 | #include <linux/uio.h> | |
21 | #include <linux/highmem.h> | |
5274f052 | 22 | #include <linux/pagemap.h> |
db349509 | 23 | #include <linux/audit.h> |
ba719bae | 24 | #include <linux/syscalls.h> |
b492e95b | 25 | #include <linux/fcntl.h> |
d86133bd | 26 | #include <linux/memcontrol.h> |
c73be61c | 27 | #include <linux/watch_queue.h> |
1998f193 | 28 | #include <linux/sysctl.h> |
1da177e4 | 29 | |
7c0f6ba6 | 30 | #include <linux/uaccess.h> |
1da177e4 LT |
31 | #include <asm/ioctls.h> |
32 | ||
599a0ac1 AV |
33 | #include "internal.h" |
34 | ||
46c4c9d1 AXH |
35 | /* |
36 | * New pipe buffers will be restricted to this size while the user is exceeding | |
37 | * their pipe buffer quota. The general pipe use case needs at least two | |
38 | * buffers: one for data yet to be read, and one for new data. If this is less | |
39 | * than two, then a write to a non-empty pipe may block even if the pipe is not | |
40 | * full. This can occur with GNU make jobserver or similar uses of pipes as | |
41 | * semaphores: multiple processes may be waiting to write tokens back to the | |
42 | * pipe before reading tokens: https://lore.kernel.org/lkml/1628086770.5rn8p04n6j.none@localhost/. | |
43 | * | |
44 | * Users can reduce their pipe buffers with F_SETPIPE_SZ below this at their | |
45 | * own risk, namely: pipe writes to non-full pipes may block until the pipe is | |
46 | * emptied. | |
47 | */ | |
48 | #define PIPE_MIN_DEF_BUFFERS 2 | |
49 | ||
b492e95b JA |
50 | /* |
51 | * The max size that a non-root user is allowed to grow the pipe. Can | |
ff9da691 | 52 | * be set by root in /proc/sys/fs/pipe-max-size |
b492e95b | 53 | */ |
1998f193 | 54 | static unsigned int pipe_max_size = 1048576; |
ff9da691 | 55 | |
759c0114 WT |
56 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
57 | * matches default values. | |
58 | */ | |
1998f193 LC |
59 | static unsigned long pipe_user_pages_hard; |
60 | static unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; | |
759c0114 | 61 | |
1da177e4 | 62 | /* |
8cefc107 DH |
63 | * We use head and tail indices that aren't masked off, except at the point of |
64 | * dereference, but rather they're allowed to wrap naturally. This means there | |
65 | * isn't a dead spot in the buffer, but the ring has to be a power of two and | |
66 | * <= 2^31. | |
67 | * -- David Howells 2019-09-23. | |
68 | * | |
1da177e4 LT |
69 | * Reads with count = 0 should always return 0. |
70 | * -- Julian Bradfield 1999-06-07. | |
71 | * | |
72 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
73 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
74 | * | |
75 | * pipe_read & write cleanup | |
76 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
77 | */ | |
78 | ||
61e0d47c MS |
79 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
80 | { | |
6447a3cf | 81 | if (pipe->files) |
72b0d9aa | 82 | mutex_lock_nested(&pipe->mutex, subclass); |
61e0d47c MS |
83 | } |
84 | ||
85 | void pipe_lock(struct pipe_inode_info *pipe) | |
86 | { | |
87 | /* | |
88 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
89 | */ | |
90 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
91 | } | |
92 | EXPORT_SYMBOL(pipe_lock); | |
93 | ||
94 | void pipe_unlock(struct pipe_inode_info *pipe) | |
95 | { | |
6447a3cf | 96 | if (pipe->files) |
72b0d9aa | 97 | mutex_unlock(&pipe->mutex); |
61e0d47c MS |
98 | } |
99 | EXPORT_SYMBOL(pipe_unlock); | |
100 | ||
ebec73f4 AV |
101 | static inline void __pipe_lock(struct pipe_inode_info *pipe) |
102 | { | |
103 | mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); | |
104 | } | |
105 | ||
106 | static inline void __pipe_unlock(struct pipe_inode_info *pipe) | |
107 | { | |
108 | mutex_unlock(&pipe->mutex); | |
109 | } | |
110 | ||
61e0d47c MS |
111 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
112 | struct pipe_inode_info *pipe2) | |
113 | { | |
114 | BUG_ON(pipe1 == pipe2); | |
115 | ||
116 | if (pipe1 < pipe2) { | |
117 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
118 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
119 | } else { | |
023d43c7 PZ |
120 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
121 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
122 | } |
123 | } | |
124 | ||
341b446b IM |
125 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
126 | struct pipe_buffer *buf) | |
1da177e4 LT |
127 | { |
128 | struct page *page = buf->page; | |
129 | ||
5274f052 JA |
130 | /* |
131 | * If nobody else uses this page, and we don't already have a | |
132 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 133 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 134 | */ |
341b446b | 135 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 136 | pipe->tmp_page = page; |
341b446b | 137 | else |
09cbfeaf | 138 | put_page(page); |
1da177e4 LT |
139 | } |
140 | ||
c928f642 CH |
141 | static bool anon_pipe_buf_try_steal(struct pipe_inode_info *pipe, |
142 | struct pipe_buffer *buf) | |
d86133bd VD |
143 | { |
144 | struct page *page = buf->page; | |
145 | ||
c928f642 CH |
146 | if (page_count(page) != 1) |
147 | return false; | |
148 | memcg_kmem_uncharge_page(page, 0); | |
149 | __SetPageLocked(page); | |
150 | return true; | |
d86133bd VD |
151 | } |
152 | ||
0845718d | 153 | /** |
c928f642 | 154 | * generic_pipe_buf_try_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 | */ | |
c928f642 CH |
165 | bool generic_pipe_buf_try_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 | 176 | lock_page(page); |
c928f642 | 177 | return true; |
46e678c9 | 178 | } |
c928f642 | 179 | return false; |
5abc97aa | 180 | } |
c928f642 | 181 | EXPORT_SYMBOL(generic_pipe_buf_try_steal); |
5abc97aa | 182 | |
0845718d | 183 | /** |
b51d63c6 | 184 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
185 | * @pipe: the pipe that the buffer belongs to |
186 | * @buf: the buffer to get a reference to | |
187 | * | |
188 | * Description: | |
189 | * This function grabs an extra reference to @buf. It's used in | |
3d742d4b | 190 | * the tee() system call, when we duplicate the buffers in one |
0845718d JA |
191 | * pipe into another. |
192 | */ | |
15fab63e | 193 | bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) |
70524490 | 194 | { |
cd1adf1b | 195 | return try_get_page(buf->page); |
70524490 | 196 | } |
51921cb7 | 197 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 198 | |
6818173b MS |
199 | /** |
200 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
201 | * @pipe: the pipe that the buffer belongs to | |
202 | * @buf: the buffer to put a reference to | |
203 | * | |
204 | * Description: | |
205 | * This function releases a reference to @buf. | |
206 | */ | |
207 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
208 | struct pipe_buffer *buf) | |
209 | { | |
09cbfeaf | 210 | put_page(buf->page); |
6818173b | 211 | } |
51921cb7 | 212 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 213 | |
d4c3cca9 | 214 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
c928f642 CH |
215 | .release = anon_pipe_buf_release, |
216 | .try_steal = anon_pipe_buf_try_steal, | |
217 | .get = generic_pipe_buf_get, | |
1da177e4 LT |
218 | }; |
219 | ||
85190d15 LT |
220 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
221 | static inline bool pipe_readable(const struct pipe_inode_info *pipe) | |
222 | { | |
223 | unsigned int head = READ_ONCE(pipe->head); | |
224 | unsigned int tail = READ_ONCE(pipe->tail); | |
225 | unsigned int writers = READ_ONCE(pipe->writers); | |
226 | ||
227 | return !pipe_empty(head, tail) || !writers; | |
228 | } | |
229 | ||
1da177e4 | 230 | static ssize_t |
fb9096a3 | 231 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 232 | { |
fb9096a3 | 233 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 234 | struct file *filp = iocb->ki_filp; |
de32ec4c | 235 | struct pipe_inode_info *pipe = filp->private_data; |
0ddad21d | 236 | bool was_full, wake_next_reader = false; |
1da177e4 | 237 | ssize_t ret; |
1da177e4 | 238 | |
1da177e4 LT |
239 | /* Null read succeeds. */ |
240 | if (unlikely(total_len == 0)) | |
241 | return 0; | |
242 | ||
1da177e4 | 243 | ret = 0; |
ebec73f4 | 244 | __pipe_lock(pipe); |
f467a6a6 LT |
245 | |
246 | /* | |
247 | * We only wake up writers if the pipe was full when we started | |
248 | * reading in order to avoid unnecessary wakeups. | |
249 | * | |
250 | * But when we do wake up writers, we do so using a sync wakeup | |
251 | * (WF_SYNC), because we want them to get going and generate more | |
252 | * data for us. | |
253 | */ | |
254 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); | |
1da177e4 | 255 | for (;;) { |
2ed147f0 DH |
256 | /* Read ->head with a barrier vs post_one_notification() */ |
257 | unsigned int head = smp_load_acquire(&pipe->head); | |
8cefc107 DH |
258 | unsigned int tail = pipe->tail; |
259 | unsigned int mask = pipe->ring_size - 1; | |
260 | ||
e7d553d6 DH |
261 | #ifdef CONFIG_WATCH_QUEUE |
262 | if (pipe->note_loss) { | |
263 | struct watch_notification n; | |
264 | ||
265 | if (total_len < 8) { | |
266 | if (ret == 0) | |
267 | ret = -ENOBUFS; | |
268 | break; | |
269 | } | |
270 | ||
271 | n.type = WATCH_TYPE_META; | |
272 | n.subtype = WATCH_META_LOSS_NOTIFICATION; | |
273 | n.info = watch_sizeof(n); | |
274 | if (copy_to_iter(&n, sizeof(n), to) != sizeof(n)) { | |
275 | if (ret == 0) | |
276 | ret = -EFAULT; | |
277 | break; | |
278 | } | |
279 | ret += sizeof(n); | |
280 | total_len -= sizeof(n); | |
281 | pipe->note_loss = false; | |
282 | } | |
283 | #endif | |
284 | ||
8cefc107 DH |
285 | if (!pipe_empty(head, tail)) { |
286 | struct pipe_buffer *buf = &pipe->bufs[tail & mask]; | |
1da177e4 | 287 | size_t chars = buf->len; |
637b58c2 AV |
288 | size_t written; |
289 | int error; | |
1da177e4 | 290 | |
8cfba763 DH |
291 | if (chars > total_len) { |
292 | if (buf->flags & PIPE_BUF_FLAG_WHOLE) { | |
293 | if (ret == 0) | |
294 | ret = -ENOBUFS; | |
295 | break; | |
296 | } | |
1da177e4 | 297 | chars = total_len; |
8cfba763 | 298 | } |
1da177e4 | 299 | |
fba597db | 300 | error = pipe_buf_confirm(pipe, buf); |
f84d7519 | 301 | if (error) { |
5274f052 | 302 | if (!ret) |
e5953cbd | 303 | ret = error; |
5274f052 JA |
304 | break; |
305 | } | |
f84d7519 | 306 | |
fb9096a3 | 307 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 308 | if (unlikely(written < chars)) { |
341b446b | 309 | if (!ret) |
637b58c2 | 310 | ret = -EFAULT; |
1da177e4 LT |
311 | break; |
312 | } | |
313 | ret += chars; | |
314 | buf->offset += chars; | |
315 | buf->len -= chars; | |
9883035a LT |
316 | |
317 | /* Was it a packet buffer? Clean up and exit */ | |
318 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
319 | total_len = chars; | |
320 | buf->len = 0; | |
321 | } | |
322 | ||
1da177e4 | 323 | if (!buf->len) { |
a779638c | 324 | pipe_buf_release(pipe, buf); |
0ddad21d | 325 | spin_lock_irq(&pipe->rd_wait.lock); |
e7d553d6 DH |
326 | #ifdef CONFIG_WATCH_QUEUE |
327 | if (buf->flags & PIPE_BUF_FLAG_LOSS) | |
328 | pipe->note_loss = true; | |
329 | #endif | |
8cefc107 DH |
330 | tail++; |
331 | pipe->tail = tail; | |
0ddad21d | 332 | spin_unlock_irq(&pipe->rd_wait.lock); |
1da177e4 LT |
333 | } |
334 | total_len -= chars; | |
335 | if (!total_len) | |
336 | break; /* common path: read succeeded */ | |
8cefc107 DH |
337 | if (!pipe_empty(head, tail)) /* More to do? */ |
338 | continue; | |
1da177e4 | 339 | } |
8cefc107 | 340 | |
923f4f23 | 341 | if (!pipe->writers) |
1da177e4 | 342 | break; |
a28c8b9d LT |
343 | if (ret) |
344 | break; | |
c04fe8e3 JA |
345 | if ((filp->f_flags & O_NONBLOCK) || |
346 | (iocb->ki_flags & IOCB_NOWAIT)) { | |
a28c8b9d LT |
347 | ret = -EAGAIN; |
348 | break; | |
1da177e4 | 349 | } |
85190d15 | 350 | __pipe_unlock(pipe); |
d1c6a2aa LT |
351 | |
352 | /* | |
353 | * We only get here if we didn't actually read anything. | |
354 | * | |
355 | * However, we could have seen (and removed) a zero-sized | |
356 | * pipe buffer, and might have made space in the buffers | |
357 | * that way. | |
358 | * | |
359 | * You can't make zero-sized pipe buffers by doing an empty | |
360 | * write (not even in packet mode), but they can happen if | |
361 | * the writer gets an EFAULT when trying to fill a buffer | |
362 | * that already got allocated and inserted in the buffer | |
363 | * array. | |
364 | * | |
365 | * So we still need to wake up any pending writers in the | |
366 | * _very_ unlikely case that the pipe was full, but we got | |
367 | * no data. | |
368 | */ | |
fe67f4dd | 369 | if (unlikely(was_full)) |
0ddad21d | 370 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
fe67f4dd | 371 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
d1c6a2aa LT |
372 | |
373 | /* | |
374 | * But because we didn't read anything, at this point we can | |
375 | * just return directly with -ERESTARTSYS if we're interrupted, | |
376 | * since we've done any required wakeups and there's no need | |
377 | * to mark anything accessed. And we've dropped the lock. | |
378 | */ | |
0ddad21d | 379 | if (wait_event_interruptible_exclusive(pipe->rd_wait, pipe_readable(pipe)) < 0) |
d1c6a2aa LT |
380 | return -ERESTARTSYS; |
381 | ||
85190d15 | 382 | __pipe_lock(pipe); |
f467a6a6 | 383 | was_full = pipe_full(pipe->head, pipe->tail, pipe->max_usage); |
0ddad21d | 384 | wake_next_reader = true; |
1da177e4 | 385 | } |
0ddad21d LT |
386 | if (pipe_empty(pipe->head, pipe->tail)) |
387 | wake_next_reader = false; | |
ebec73f4 | 388 | __pipe_unlock(pipe); |
341b446b | 389 | |
fe67f4dd | 390 | if (was_full) |
0ddad21d | 391 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); |
0ddad21d LT |
392 | if (wake_next_reader) |
393 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); | |
fe67f4dd | 394 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 LT |
395 | if (ret > 0) |
396 | file_accessed(filp); | |
397 | return ret; | |
398 | } | |
399 | ||
9883035a LT |
400 | static inline int is_packetized(struct file *file) |
401 | { | |
402 | return (file->f_flags & O_DIRECT) != 0; | |
403 | } | |
404 | ||
85190d15 LT |
405 | /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */ |
406 | static inline bool pipe_writable(const struct pipe_inode_info *pipe) | |
407 | { | |
408 | unsigned int head = READ_ONCE(pipe->head); | |
409 | unsigned int tail = READ_ONCE(pipe->tail); | |
410 | unsigned int max_usage = READ_ONCE(pipe->max_usage); | |
411 | ||
412 | return !pipe_full(head, tail, max_usage) || | |
413 | !READ_ONCE(pipe->readers); | |
414 | } | |
415 | ||
1da177e4 | 416 | static ssize_t |
f0d1bec9 | 417 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 418 | { |
ee0b3e67 | 419 | struct file *filp = iocb->ki_filp; |
de32ec4c | 420 | struct pipe_inode_info *pipe = filp->private_data; |
8f868d68 | 421 | unsigned int head; |
f0d1bec9 | 422 | ssize_t ret = 0; |
f0d1bec9 | 423 | size_t total_len = iov_iter_count(from); |
1da177e4 | 424 | ssize_t chars; |
1b6b26ae | 425 | bool was_empty = false; |
0ddad21d | 426 | bool wake_next_writer = false; |
1da177e4 | 427 | |
1da177e4 LT |
428 | /* Null write succeeds. */ |
429 | if (unlikely(total_len == 0)) | |
430 | return 0; | |
431 | ||
ebec73f4 | 432 | __pipe_lock(pipe); |
1da177e4 | 433 | |
923f4f23 | 434 | if (!pipe->readers) { |
1da177e4 LT |
435 | send_sig(SIGPIPE, current, 0); |
436 | ret = -EPIPE; | |
437 | goto out; | |
438 | } | |
439 | ||
b4bd6b4b | 440 | if (pipe_has_watch_queue(pipe)) { |
c73be61c DH |
441 | ret = -EXDEV; |
442 | goto out; | |
443 | } | |
c73be61c | 444 | |
1b6b26ae | 445 | /* |
1b6b26ae LT |
446 | * If it wasn't empty we try to merge new data into |
447 | * the last buffer. | |
448 | * | |
449 | * That naturally merges small writes, but it also | |
3a34b13a | 450 | * page-aligns the rest of the writes for large writes |
1b6b26ae LT |
451 | * spanning multiple pages. |
452 | */ | |
8cefc107 | 453 | head = pipe->head; |
3b844826 | 454 | was_empty = pipe_empty(head, pipe->tail); |
1b6b26ae | 455 | chars = total_len & (PAGE_SIZE-1); |
3b844826 | 456 | if (chars && !was_empty) { |
8f868d68 | 457 | unsigned int mask = pipe->ring_size - 1; |
8cefc107 | 458 | struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask]; |
1da177e4 | 459 | int offset = buf->offset + buf->len; |
341b446b | 460 | |
f6dd9755 CH |
461 | if ((buf->flags & PIPE_BUF_FLAG_CAN_MERGE) && |
462 | offset + chars <= PAGE_SIZE) { | |
fba597db | 463 | ret = pipe_buf_confirm(pipe, buf); |
6ae08069 | 464 | if (ret) |
5274f052 | 465 | goto out; |
f84d7519 | 466 | |
f0d1bec9 AV |
467 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
468 | if (unlikely(ret < chars)) { | |
6ae08069 | 469 | ret = -EFAULT; |
1da177e4 | 470 | goto out; |
f6762b7a | 471 | } |
1b6b26ae | 472 | |
6ae08069 | 473 | buf->len += ret; |
f0d1bec9 | 474 | if (!iov_iter_count(from)) |
1da177e4 LT |
475 | goto out; |
476 | } | |
477 | } | |
478 | ||
479 | for (;;) { | |
923f4f23 | 480 | if (!pipe->readers) { |
1da177e4 | 481 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
482 | if (!ret) |
483 | ret = -EPIPE; | |
1da177e4 LT |
484 | break; |
485 | } | |
8cefc107 | 486 | |
a194dfe6 | 487 | head = pipe->head; |
8f868d68 DH |
488 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) { |
489 | unsigned int mask = pipe->ring_size - 1; | |
fbaa530e | 490 | struct pipe_buffer *buf; |
923f4f23 | 491 | struct page *page = pipe->tmp_page; |
f0d1bec9 | 492 | int copied; |
1da177e4 LT |
493 | |
494 | if (!page) { | |
d86133bd | 495 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
496 | if (unlikely(!page)) { |
497 | ret = ret ? : -ENOMEM; | |
498 | break; | |
499 | } | |
923f4f23 | 500 | pipe->tmp_page = page; |
1da177e4 | 501 | } |
a194dfe6 DH |
502 | |
503 | /* Allocate a slot in the ring in advance and attach an | |
504 | * empty buffer. If we fault or otherwise fail to use | |
505 | * it, either the reader will consume it or it'll still | |
506 | * be there for the next write. | |
507 | */ | |
a194dfe6 | 508 | pipe->head = head + 1; |
1da177e4 LT |
509 | |
510 | /* Insert it into the buffer array */ | |
a194dfe6 | 511 | buf = &pipe->bufs[head & mask]; |
1da177e4 LT |
512 | buf->page = page; |
513 | buf->ops = &anon_pipe_buf_ops; | |
514 | buf->offset = 0; | |
a194dfe6 | 515 | buf->len = 0; |
f6dd9755 | 516 | if (is_packetized(filp)) |
9883035a | 517 | buf->flags = PIPE_BUF_FLAG_PACKET; |
f6dd9755 CH |
518 | else |
519 | buf->flags = PIPE_BUF_FLAG_CAN_MERGE; | |
923f4f23 | 520 | pipe->tmp_page = NULL; |
1da177e4 | 521 | |
a194dfe6 DH |
522 | copied = copy_page_from_iter(page, 0, PAGE_SIZE, from); |
523 | if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) { | |
524 | if (!ret) | |
525 | ret = -EFAULT; | |
526 | break; | |
527 | } | |
528 | ret += copied; | |
529 | buf->offset = 0; | |
530 | buf->len = copied; | |
531 | ||
f0d1bec9 | 532 | if (!iov_iter_count(from)) |
1da177e4 LT |
533 | break; |
534 | } | |
8cefc107 | 535 | |
8f868d68 | 536 | if (!pipe_full(head, pipe->tail, pipe->max_usage)) |
1da177e4 | 537 | continue; |
8cefc107 DH |
538 | |
539 | /* Wait for buffer space to become available. */ | |
c04fe8e3 JA |
540 | if ((filp->f_flags & O_NONBLOCK) || |
541 | (iocb->ki_flags & IOCB_NOWAIT)) { | |
341b446b IM |
542 | if (!ret) |
543 | ret = -EAGAIN; | |
1da177e4 LT |
544 | break; |
545 | } | |
546 | if (signal_pending(current)) { | |
341b446b IM |
547 | if (!ret) |
548 | ret = -ERESTARTSYS; | |
1da177e4 LT |
549 | break; |
550 | } | |
1b6b26ae LT |
551 | |
552 | /* | |
553 | * We're going to release the pipe lock and wait for more | |
554 | * space. We wake up any readers if necessary, and then | |
555 | * after waiting we need to re-check whether the pipe | |
556 | * become empty while we dropped the lock. | |
557 | */ | |
85190d15 | 558 | __pipe_unlock(pipe); |
fe67f4dd | 559 | if (was_empty) |
0ddad21d | 560 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
fe67f4dd | 561 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
0ddad21d | 562 | wait_event_interruptible_exclusive(pipe->wr_wait, pipe_writable(pipe)); |
85190d15 | 563 | __pipe_lock(pipe); |
0dd1e377 | 564 | was_empty = pipe_empty(pipe->head, pipe->tail); |
0ddad21d | 565 | wake_next_writer = true; |
1da177e4 LT |
566 | } |
567 | out: | |
0ddad21d LT |
568 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) |
569 | wake_next_writer = false; | |
ebec73f4 | 570 | __pipe_unlock(pipe); |
1b6b26ae LT |
571 | |
572 | /* | |
573 | * If we do do a wakeup event, we do a 'sync' wakeup, because we | |
574 | * want the reader to start processing things asap, rather than | |
575 | * leave the data pending. | |
576 | * | |
577 | * This is particularly important for small writes, because of | |
578 | * how (for example) the GNU make jobserver uses small writes to | |
579 | * wake up pending jobs | |
3b844826 LT |
580 | * |
581 | * Epoll nonsensically wants a wakeup whether the pipe | |
582 | * was already empty or not. | |
1b6b26ae | 583 | */ |
fe67f4dd | 584 | if (was_empty || pipe->poll_usage) |
0ddad21d | 585 | wake_up_interruptible_sync_poll(&pipe->rd_wait, EPOLLIN | EPOLLRDNORM); |
fe67f4dd | 586 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
0ddad21d LT |
587 | if (wake_next_writer) |
588 | wake_up_interruptible_sync_poll(&pipe->wr_wait, EPOLLOUT | EPOLLWRNORM); | |
7e775f46 | 589 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
590 | int err = file_update_time(filp); |
591 | if (err) | |
592 | ret = err; | |
7e775f46 | 593 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 594 | } |
1da177e4 LT |
595 | return ret; |
596 | } | |
597 | ||
d59d0b1b | 598 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 599 | { |
de32ec4c | 600 | struct pipe_inode_info *pipe = filp->private_data; |
aeb213cd | 601 | unsigned int count, head, tail, mask; |
1da177e4 LT |
602 | |
603 | switch (cmd) { | |
c73be61c DH |
604 | case FIONREAD: |
605 | __pipe_lock(pipe); | |
606 | count = 0; | |
607 | head = pipe->head; | |
608 | tail = pipe->tail; | |
609 | mask = pipe->ring_size - 1; | |
8cefc107 | 610 | |
c73be61c DH |
611 | while (tail != head) { |
612 | count += pipe->bufs[tail & mask].len; | |
613 | tail++; | |
614 | } | |
615 | __pipe_unlock(pipe); | |
923f4f23 | 616 | |
c73be61c | 617 | return put_user(count, (int __user *)arg); |
923f4f23 | 618 | |
c73be61c DH |
619 | #ifdef CONFIG_WATCH_QUEUE |
620 | case IOC_WATCH_QUEUE_SET_SIZE: { | |
621 | int ret; | |
622 | __pipe_lock(pipe); | |
623 | ret = watch_queue_set_size(pipe, arg); | |
624 | __pipe_unlock(pipe); | |
625 | return ret; | |
626 | } | |
627 | ||
628 | case IOC_WATCH_QUEUE_SET_FILTER: | |
629 | return watch_queue_set_filter( | |
630 | pipe, (struct watch_notification_filter __user *)arg); | |
631 | #endif | |
632 | ||
633 | default: | |
634 | return -ENOIOCTLCMD; | |
1da177e4 LT |
635 | } |
636 | } | |
637 | ||
dd67081b | 638 | /* No kernel lock held - fine */ |
a11e1d43 LT |
639 | static __poll_t |
640 | pipe_poll(struct file *filp, poll_table *wait) | |
dd67081b | 641 | { |
a11e1d43 | 642 | __poll_t mask; |
dd67081b | 643 | struct pipe_inode_info *pipe = filp->private_data; |
ad910e36 | 644 | unsigned int head, tail; |
a11e1d43 | 645 | |
3b844826 | 646 | /* Epoll has some historical nasty semantics, this enables them */ |
f485922d | 647 | WRITE_ONCE(pipe->poll_usage, true); |
3b844826 | 648 | |
ad910e36 | 649 | /* |
0ddad21d | 650 | * Reading pipe state only -- no need for acquiring the semaphore. |
ad910e36 LT |
651 | * |
652 | * But because this is racy, the code has to add the | |
653 | * entry to the poll table _first_ .. | |
654 | */ | |
0ddad21d LT |
655 | if (filp->f_mode & FMODE_READ) |
656 | poll_wait(filp, &pipe->rd_wait, wait); | |
657 | if (filp->f_mode & FMODE_WRITE) | |
658 | poll_wait(filp, &pipe->wr_wait, wait); | |
1da177e4 | 659 | |
ad910e36 LT |
660 | /* |
661 | * .. and only then can you do the racy tests. That way, | |
662 | * if something changes and you got it wrong, the poll | |
663 | * table entry will wake you up and fix it. | |
664 | */ | |
665 | head = READ_ONCE(pipe->head); | |
666 | tail = READ_ONCE(pipe->tail); | |
667 | ||
a11e1d43 | 668 | mask = 0; |
1da177e4 | 669 | if (filp->f_mode & FMODE_READ) { |
8cefc107 DH |
670 | if (!pipe_empty(head, tail)) |
671 | mask |= EPOLLIN | EPOLLRDNORM; | |
923f4f23 | 672 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
a9a08845 | 673 | mask |= EPOLLHUP; |
1da177e4 LT |
674 | } |
675 | ||
676 | if (filp->f_mode & FMODE_WRITE) { | |
6718b6f8 | 677 | if (!pipe_full(head, tail, pipe->max_usage)) |
8cefc107 | 678 | mask |= EPOLLOUT | EPOLLWRNORM; |
5e5d7a22 | 679 | /* |
a9a08845 | 680 | * Most Unices do not set EPOLLERR for FIFOs but on Linux they |
5e5d7a22 PE |
681 | * behave exactly like pipes for poll(). |
682 | */ | |
923f4f23 | 683 | if (!pipe->readers) |
a9a08845 | 684 | mask |= EPOLLERR; |
1da177e4 LT |
685 | } |
686 | ||
687 | return mask; | |
688 | } | |
689 | ||
b0d8d229 LT |
690 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
691 | { | |
692 | int kill = 0; | |
693 | ||
694 | spin_lock(&inode->i_lock); | |
695 | if (!--pipe->files) { | |
696 | inode->i_pipe = NULL; | |
697 | kill = 1; | |
698 | } | |
699 | spin_unlock(&inode->i_lock); | |
700 | ||
701 | if (kill) | |
702 | free_pipe_info(pipe); | |
703 | } | |
704 | ||
1da177e4 | 705 | static int |
599a0ac1 | 706 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 707 | { |
b0d8d229 | 708 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 709 | |
ebec73f4 | 710 | __pipe_lock(pipe); |
599a0ac1 AV |
711 | if (file->f_mode & FMODE_READ) |
712 | pipe->readers--; | |
713 | if (file->f_mode & FMODE_WRITE) | |
714 | pipe->writers--; | |
341b446b | 715 | |
6551d5c5 LT |
716 | /* Was that the last reader or writer, but not the other side? */ |
717 | if (!pipe->readers != !pipe->writers) { | |
718 | wake_up_interruptible_all(&pipe->rd_wait); | |
719 | wake_up_interruptible_all(&pipe->wr_wait); | |
923f4f23 IM |
720 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
721 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 722 | } |
ebec73f4 | 723 | __pipe_unlock(pipe); |
ba5bb147 | 724 | |
b0d8d229 | 725 | put_pipe_info(inode, pipe); |
1da177e4 LT |
726 | return 0; |
727 | } | |
728 | ||
729 | static int | |
599a0ac1 | 730 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 731 | { |
de32ec4c | 732 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 733 | int retval = 0; |
1da177e4 | 734 | |
ebec73f4 | 735 | __pipe_lock(pipe); |
599a0ac1 AV |
736 | if (filp->f_mode & FMODE_READ) |
737 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
738 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 739 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
740 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
741 | /* this can happen only if on == T */ | |
e5bc49ba ON |
742 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
743 | } | |
ebec73f4 | 744 | __pipe_unlock(pipe); |
60aa4924 | 745 | return retval; |
1da177e4 LT |
746 | } |
747 | ||
c73be61c DH |
748 | unsigned long account_pipe_buffers(struct user_struct *user, |
749 | unsigned long old, unsigned long new) | |
759c0114 | 750 | { |
9c87bcf0 | 751 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
752 | } |
753 | ||
c73be61c | 754 | bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 755 | { |
f7340761 EB |
756 | unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft); |
757 | ||
758 | return soft_limit && user_bufs > soft_limit; | |
759c0114 WT |
759 | } |
760 | ||
c73be61c | 761 | bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 762 | { |
f7340761 EB |
763 | unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard); |
764 | ||
765 | return hard_limit && user_bufs > hard_limit; | |
759c0114 WT |
766 | } |
767 | ||
c73be61c | 768 | bool pipe_is_unprivileged_user(void) |
85c2dd54 EB |
769 | { |
770 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
771 | } | |
772 | ||
7bee130e | 773 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 774 | { |
923f4f23 | 775 | struct pipe_inode_info *pipe; |
09b4d199 MK |
776 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
777 | struct user_struct *user = get_current_user(); | |
9c87bcf0 | 778 | unsigned long user_bufs; |
f7340761 | 779 | unsigned int max_size = READ_ONCE(pipe_max_size); |
3a326a2c | 780 | |
d86133bd | 781 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
09b4d199 MK |
782 | if (pipe == NULL) |
783 | goto out_free_uid; | |
784 | ||
f7340761 EB |
785 | if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE)) |
786 | pipe_bufs = max_size >> PAGE_SHIFT; | |
086e774a | 787 | |
9c87bcf0 | 788 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
a005ca0e | 789 | |
c73be61c | 790 | if (too_many_pipe_buffers_soft(user_bufs) && pipe_is_unprivileged_user()) { |
46c4c9d1 AXH |
791 | user_bufs = account_pipe_buffers(user, pipe_bufs, PIPE_MIN_DEF_BUFFERS); |
792 | pipe_bufs = PIPE_MIN_DEF_BUFFERS; | |
09b4d199 | 793 | } |
759c0114 | 794 | |
c73be61c | 795 | if (too_many_pipe_buffers_hard(user_bufs) && pipe_is_unprivileged_user()) |
a005ca0e MK |
796 | goto out_revert_acct; |
797 | ||
906f9040 | 798 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), |
a005ca0e MK |
799 | GFP_KERNEL_ACCOUNT); |
800 | ||
09b4d199 | 801 | if (pipe->bufs) { |
0ddad21d LT |
802 | init_waitqueue_head(&pipe->rd_wait); |
803 | init_waitqueue_head(&pipe->wr_wait); | |
09b4d199 | 804 | pipe->r_counter = pipe->w_counter = 1; |
6718b6f8 | 805 | pipe->max_usage = pipe_bufs; |
8cefc107 | 806 | pipe->ring_size = pipe_bufs; |
c73be61c | 807 | pipe->nr_accounted = pipe_bufs; |
09b4d199 | 808 | pipe->user = user; |
09b4d199 MK |
809 | mutex_init(&pipe->mutex); |
810 | return pipe; | |
3a326a2c IM |
811 | } |
812 | ||
a005ca0e | 813 | out_revert_acct: |
9c87bcf0 | 814 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
09b4d199 MK |
815 | kfree(pipe); |
816 | out_free_uid: | |
817 | free_uid(user); | |
35f3d14d | 818 | return NULL; |
3a326a2c IM |
819 | } |
820 | ||
4b8a8f1e | 821 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 | 822 | { |
aeb213cd | 823 | unsigned int i; |
1da177e4 | 824 | |
c73be61c | 825 | #ifdef CONFIG_WATCH_QUEUE |
db8facfc | 826 | if (pipe->watch_queue) |
c73be61c | 827 | watch_queue_clear(pipe->watch_queue); |
c73be61c DH |
828 | #endif |
829 | ||
830 | (void) account_pipe_buffers(pipe->user, pipe->nr_accounted, 0); | |
759c0114 | 831 | free_uid(pipe->user); |
8cefc107 | 832 | for (i = 0; i < pipe->ring_size; i++) { |
923f4f23 | 833 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 834 | if (buf->ops) |
a779638c | 835 | pipe_buf_release(pipe, buf); |
1da177e4 | 836 | } |
db8facfc DH |
837 | #ifdef CONFIG_WATCH_QUEUE |
838 | if (pipe->watch_queue) | |
839 | put_watch_queue(pipe->watch_queue); | |
840 | #endif | |
923f4f23 IM |
841 | if (pipe->tmp_page) |
842 | __free_page(pipe->tmp_page); | |
906f9040 | 843 | kfree(pipe->bufs); |
923f4f23 | 844 | kfree(pipe); |
1da177e4 LT |
845 | } |
846 | ||
fa3536cc | 847 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 848 | |
c23fbb6b ED |
849 | /* |
850 | * pipefs_dname() is called from d_path(). | |
851 | */ | |
852 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
853 | { | |
0f60d288 | 854 | return dynamic_dname(buffer, buflen, "pipe:[%lu]", |
75c3cfa8 | 855 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
856 | } |
857 | ||
3ba13d17 | 858 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 859 | .d_dname = pipefs_dname, |
1da177e4 LT |
860 | }; |
861 | ||
862 | static struct inode * get_pipe_inode(void) | |
863 | { | |
a209dfc7 | 864 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 865 | struct pipe_inode_info *pipe; |
1da177e4 LT |
866 | |
867 | if (!inode) | |
868 | goto fail_inode; | |
869 | ||
85fe4025 CH |
870 | inode->i_ino = get_next_ino(); |
871 | ||
7bee130e | 872 | pipe = alloc_pipe_info(); |
923f4f23 | 873 | if (!pipe) |
1da177e4 | 874 | goto fail_iput; |
3a326a2c | 875 | |
ba5bb147 AV |
876 | inode->i_pipe = pipe; |
877 | pipe->files = 2; | |
923f4f23 | 878 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 879 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
880 | |
881 | /* | |
882 | * Mark the inode dirty from the very beginning, | |
883 | * that way it will never be moved to the dirty | |
884 | * list because "mark_inode_dirty()" will think | |
885 | * that it already _is_ on the dirty list. | |
886 | */ | |
887 | inode->i_state = I_DIRTY; | |
888 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
889 | inode->i_uid = current_fsuid(); |
890 | inode->i_gid = current_fsgid(); | |
2276e5ba | 891 | inode->i_atime = inode->i_mtime = inode_set_ctime_current(inode); |
923f4f23 | 892 | |
1da177e4 LT |
893 | return inode; |
894 | ||
895 | fail_iput: | |
896 | iput(inode); | |
341b446b | 897 | |
1da177e4 LT |
898 | fail_inode: |
899 | return NULL; | |
900 | } | |
901 | ||
e4fad8e5 | 902 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 903 | { |
e4fad8e5 | 904 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 905 | struct file *f; |
8a018eb5 | 906 | int error; |
1da177e4 | 907 | |
1da177e4 | 908 | if (!inode) |
e4fad8e5 | 909 | return -ENFILE; |
1da177e4 | 910 | |
c73be61c | 911 | if (flags & O_NOTIFICATION_PIPE) { |
8a018eb5 QC |
912 | error = watch_queue_init(inode->i_pipe); |
913 | if (error) { | |
914 | free_pipe_info(inode->i_pipe); | |
c73be61c | 915 | iput(inode); |
8a018eb5 | 916 | return error; |
c73be61c | 917 | } |
c73be61c DH |
918 | } |
919 | ||
152b6372 AV |
920 | f = alloc_file_pseudo(inode, pipe_mnt, "", |
921 | O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)), | |
922 | &pipefifo_fops); | |
e9bb1f9b | 923 | if (IS_ERR(f)) { |
152b6372 AV |
924 | free_pipe_info(inode->i_pipe); |
925 | iput(inode); | |
926 | return PTR_ERR(f); | |
e9bb1f9b | 927 | } |
341b446b | 928 | |
de32ec4c | 929 | f->private_data = inode->i_pipe; |
d6cbd281 | 930 | |
183266f2 AV |
931 | res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK), |
932 | &pipefifo_fops); | |
e9bb1f9b | 933 | if (IS_ERR(res[0])) { |
b10a4a9f AV |
934 | put_pipe_info(inode, inode->i_pipe); |
935 | fput(f); | |
936 | return PTR_ERR(res[0]); | |
e9bb1f9b | 937 | } |
de32ec4c | 938 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 | 939 | res[1] = f; |
d8e464ec LT |
940 | stream_open(inode, res[0]); |
941 | stream_open(inode, res[1]); | |
e4fad8e5 | 942 | return 0; |
d6cbd281 AK |
943 | } |
944 | ||
5b249b1b | 945 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 946 | { |
d6cbd281 AK |
947 | int error; |
948 | int fdw, fdr; | |
949 | ||
c73be61c | 950 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT | O_NOTIFICATION_PIPE)) |
ed8cae8b UD |
951 | return -EINVAL; |
952 | ||
e4fad8e5 AV |
953 | error = create_pipe_files(files, flags); |
954 | if (error) | |
955 | return error; | |
d6cbd281 | 956 | |
ed8cae8b | 957 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
958 | if (error < 0) |
959 | goto err_read_pipe; | |
960 | fdr = error; | |
961 | ||
ed8cae8b | 962 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
963 | if (error < 0) |
964 | goto err_fdr; | |
965 | fdw = error; | |
966 | ||
157cf649 | 967 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
968 | fd[0] = fdr; |
969 | fd[1] = fdw; | |
afed6271 JA |
970 | /* pipe groks IOCB_NOWAIT */ |
971 | files[0]->f_mode |= FMODE_NOWAIT; | |
972 | files[1]->f_mode |= FMODE_NOWAIT; | |
d6cbd281 AK |
973 | return 0; |
974 | ||
975 | err_fdr: | |
976 | put_unused_fd(fdr); | |
977 | err_read_pipe: | |
e4fad8e5 AV |
978 | fput(files[0]); |
979 | fput(files[1]); | |
d6cbd281 | 980 | return error; |
1da177e4 LT |
981 | } |
982 | ||
5b249b1b AV |
983 | int do_pipe_flags(int *fd, int flags) |
984 | { | |
985 | struct file *files[2]; | |
986 | int error = __do_pipe_flags(fd, files, flags); | |
987 | if (!error) { | |
988 | fd_install(fd[0], files[0]); | |
989 | fd_install(fd[1], files[1]); | |
990 | } | |
991 | return error; | |
992 | } | |
993 | ||
d35c7b0e UD |
994 | /* |
995 | * sys_pipe() is the normal C calling standard for creating | |
996 | * a pipe. It's not the way Unix traditionally does this, though. | |
997 | */ | |
0a216dd1 | 998 | static int do_pipe2(int __user *fildes, int flags) |
d35c7b0e | 999 | { |
5b249b1b | 1000 | struct file *files[2]; |
d35c7b0e UD |
1001 | int fd[2]; |
1002 | int error; | |
1003 | ||
5b249b1b | 1004 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 1005 | if (!error) { |
5b249b1b AV |
1006 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
1007 | fput(files[0]); | |
1008 | fput(files[1]); | |
1009 | put_unused_fd(fd[0]); | |
1010 | put_unused_fd(fd[1]); | |
d35c7b0e | 1011 | error = -EFAULT; |
5b249b1b AV |
1012 | } else { |
1013 | fd_install(fd[0], files[0]); | |
1014 | fd_install(fd[1], files[1]); | |
ba719bae | 1015 | } |
d35c7b0e UD |
1016 | } |
1017 | return error; | |
1018 | } | |
1019 | ||
0a216dd1 DB |
1020 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
1021 | { | |
1022 | return do_pipe2(fildes, flags); | |
1023 | } | |
1024 | ||
2b664219 | 1025 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b | 1026 | { |
0a216dd1 | 1027 | return do_pipe2(fildes, 0); |
ed8cae8b UD |
1028 | } |
1029 | ||
472e5b05 LT |
1030 | /* |
1031 | * This is the stupid "wait for pipe to be readable or writable" | |
1032 | * model. | |
1033 | * | |
1034 | * See pipe_read/write() for the proper kind of exclusive wait, | |
1035 | * but that requires that we wake up any other readers/writers | |
1036 | * if we then do not end up reading everything (ie the whole | |
1037 | * "wake_next_reader/writer" logic in pipe_read/write()). | |
1038 | */ | |
1039 | void pipe_wait_readable(struct pipe_inode_info *pipe) | |
1040 | { | |
1041 | pipe_unlock(pipe); | |
1042 | wait_event_interruptible(pipe->rd_wait, pipe_readable(pipe)); | |
1043 | pipe_lock(pipe); | |
1044 | } | |
1045 | ||
1046 | void pipe_wait_writable(struct pipe_inode_info *pipe) | |
1047 | { | |
1048 | pipe_unlock(pipe); | |
1049 | wait_event_interruptible(pipe->wr_wait, pipe_writable(pipe)); | |
1050 | pipe_lock(pipe); | |
1051 | } | |
1052 | ||
1053 | /* | |
1054 | * This depends on both the wait (here) and the wakeup (wake_up_partner) | |
1055 | * holding the pipe lock, so "*cnt" is stable and we know a wakeup cannot | |
1056 | * race with the count check and waitqueue prep. | |
1057 | * | |
1058 | * Normally in order to avoid races, you'd do the prepare_to_wait() first, | |
1059 | * then check the condition you're waiting for, and only then sleep. But | |
1060 | * because of the pipe lock, we can check the condition before being on | |
1061 | * the wait queue. | |
1062 | * | |
1063 | * We use the 'rd_wait' waitqueue for pipe partner waiting. | |
1064 | */ | |
fc7478a2 | 1065 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 | 1066 | { |
472e5b05 | 1067 | DEFINE_WAIT(rdwait); |
8cefc107 | 1068 | int cur = *cnt; |
f776c738 AV |
1069 | |
1070 | while (cur == *cnt) { | |
472e5b05 LT |
1071 | prepare_to_wait(&pipe->rd_wait, &rdwait, TASK_INTERRUPTIBLE); |
1072 | pipe_unlock(pipe); | |
1073 | schedule(); | |
1074 | finish_wait(&pipe->rd_wait, &rdwait); | |
1075 | pipe_lock(pipe); | |
f776c738 AV |
1076 | if (signal_pending(current)) |
1077 | break; | |
1078 | } | |
1079 | return cur == *cnt ? -ERESTARTSYS : 0; | |
1080 | } | |
1081 | ||
fc7478a2 | 1082 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 1083 | { |
6551d5c5 | 1084 | wake_up_interruptible_all(&pipe->rd_wait); |
f776c738 AV |
1085 | } |
1086 | ||
1087 | static int fifo_open(struct inode *inode, struct file *filp) | |
1088 | { | |
1089 | struct pipe_inode_info *pipe; | |
599a0ac1 | 1090 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
1091 | int ret; |
1092 | ||
ba5bb147 AV |
1093 | filp->f_version = 0; |
1094 | ||
1095 | spin_lock(&inode->i_lock); | |
1096 | if (inode->i_pipe) { | |
1097 | pipe = inode->i_pipe; | |
1098 | pipe->files++; | |
1099 | spin_unlock(&inode->i_lock); | |
1100 | } else { | |
1101 | spin_unlock(&inode->i_lock); | |
7bee130e | 1102 | pipe = alloc_pipe_info(); |
f776c738 | 1103 | if (!pipe) |
ba5bb147 AV |
1104 | return -ENOMEM; |
1105 | pipe->files = 1; | |
1106 | spin_lock(&inode->i_lock); | |
1107 | if (unlikely(inode->i_pipe)) { | |
1108 | inode->i_pipe->files++; | |
1109 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 1110 | free_pipe_info(pipe); |
ba5bb147 AV |
1111 | pipe = inode->i_pipe; |
1112 | } else { | |
1113 | inode->i_pipe = pipe; | |
1114 | spin_unlock(&inode->i_lock); | |
1115 | } | |
f776c738 | 1116 | } |
de32ec4c | 1117 | filp->private_data = pipe; |
ba5bb147 AV |
1118 | /* OK, we have a pipe and it's pinned down */ |
1119 | ||
ebec73f4 | 1120 | __pipe_lock(pipe); |
f776c738 AV |
1121 | |
1122 | /* We can only do regular read/write on fifos */ | |
d8e464ec | 1123 | stream_open(inode, filp); |
f776c738 | 1124 | |
d8e464ec | 1125 | switch (filp->f_mode & (FMODE_READ | FMODE_WRITE)) { |
f776c738 AV |
1126 | case FMODE_READ: |
1127 | /* | |
1128 | * O_RDONLY | |
1129 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
1130 | * opened, even when there is no process writing the FIFO. | |
1131 | */ | |
f776c738 AV |
1132 | pipe->r_counter++; |
1133 | if (pipe->readers++ == 0) | |
fc7478a2 | 1134 | wake_up_partner(pipe); |
f776c738 | 1135 | |
599a0ac1 | 1136 | if (!is_pipe && !pipe->writers) { |
f776c738 | 1137 | if ((filp->f_flags & O_NONBLOCK)) { |
a9a08845 | 1138 | /* suppress EPOLLHUP until we have |
f776c738 AV |
1139 | * seen a writer */ |
1140 | filp->f_version = pipe->w_counter; | |
1141 | } else { | |
fc7478a2 | 1142 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
1143 | goto err_rd; |
1144 | } | |
1145 | } | |
1146 | break; | |
8cefc107 | 1147 | |
f776c738 AV |
1148 | case FMODE_WRITE: |
1149 | /* | |
1150 | * O_WRONLY | |
1151 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
1152 | * errno=ENXIO when there is no process reading the FIFO. | |
1153 | */ | |
1154 | ret = -ENXIO; | |
599a0ac1 | 1155 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
1156 | goto err; |
1157 | ||
f776c738 AV |
1158 | pipe->w_counter++; |
1159 | if (!pipe->writers++) | |
fc7478a2 | 1160 | wake_up_partner(pipe); |
f776c738 | 1161 | |
599a0ac1 | 1162 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 1163 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
1164 | goto err_wr; |
1165 | } | |
1166 | break; | |
8cefc107 | 1167 | |
f776c738 AV |
1168 | case FMODE_READ | FMODE_WRITE: |
1169 | /* | |
1170 | * O_RDWR | |
1171 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
1172 | * This implementation will NEVER block on a O_RDWR open, since | |
1173 | * the process can at least talk to itself. | |
1174 | */ | |
f776c738 AV |
1175 | |
1176 | pipe->readers++; | |
1177 | pipe->writers++; | |
1178 | pipe->r_counter++; | |
1179 | pipe->w_counter++; | |
1180 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 1181 | wake_up_partner(pipe); |
f776c738 AV |
1182 | break; |
1183 | ||
1184 | default: | |
1185 | ret = -EINVAL; | |
1186 | goto err; | |
1187 | } | |
1188 | ||
1189 | /* Ok! */ | |
ebec73f4 | 1190 | __pipe_unlock(pipe); |
f776c738 AV |
1191 | return 0; |
1192 | ||
1193 | err_rd: | |
1194 | if (!--pipe->readers) | |
0ddad21d | 1195 | wake_up_interruptible(&pipe->wr_wait); |
f776c738 AV |
1196 | ret = -ERESTARTSYS; |
1197 | goto err; | |
1198 | ||
1199 | err_wr: | |
1200 | if (!--pipe->writers) | |
6551d5c5 | 1201 | wake_up_interruptible_all(&pipe->rd_wait); |
f776c738 AV |
1202 | ret = -ERESTARTSYS; |
1203 | goto err; | |
1204 | ||
1205 | err: | |
ebec73f4 | 1206 | __pipe_unlock(pipe); |
b0d8d229 LT |
1207 | |
1208 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1209 | return ret; |
1210 | } | |
1211 | ||
599a0ac1 AV |
1212 | const struct file_operations pipefifo_fops = { |
1213 | .open = fifo_open, | |
1214 | .llseek = no_llseek, | |
fb9096a3 | 1215 | .read_iter = pipe_read, |
f0d1bec9 | 1216 | .write_iter = pipe_write, |
a11e1d43 | 1217 | .poll = pipe_poll, |
599a0ac1 AV |
1218 | .unlocked_ioctl = pipe_ioctl, |
1219 | .release = pipe_release, | |
1220 | .fasync = pipe_fasync, | |
f8ad8187 | 1221 | .splice_write = iter_file_splice_write, |
f776c738 AV |
1222 | }; |
1223 | ||
f491bd71 MK |
1224 | /* |
1225 | * Currently we rely on the pipe array holding a power-of-2 number | |
d3f14c48 | 1226 | * of pages. Returns 0 on error. |
f491bd71 | 1227 | */ |
515c5046 | 1228 | unsigned int round_pipe_size(unsigned int size) |
f491bd71 | 1229 | { |
c4fed5a9 | 1230 | if (size > (1U << 31)) |
96e99be4 EB |
1231 | return 0; |
1232 | ||
4c2e4bef EB |
1233 | /* Minimum pipe size, as required by POSIX */ |
1234 | if (size < PAGE_SIZE) | |
c4fed5a9 | 1235 | return PAGE_SIZE; |
d3f14c48 | 1236 | |
c4fed5a9 | 1237 | return roundup_pow_of_two(size); |
f491bd71 MK |
1238 | } |
1239 | ||
35f3d14d | 1240 | /* |
c73be61c | 1241 | * Resize the pipe ring to a number of slots. |
189b0ddc DH |
1242 | * |
1243 | * Note the pipe can be reduced in capacity, but only if the current | |
1244 | * occupancy doesn't exceed nr_slots; if it does, EBUSY will be | |
1245 | * returned instead. | |
35f3d14d | 1246 | */ |
c73be61c | 1247 | int pipe_resize_ring(struct pipe_inode_info *pipe, unsigned int nr_slots) |
35f3d14d JA |
1248 | { |
1249 | struct pipe_buffer *bufs; | |
c73be61c | 1250 | unsigned int head, tail, mask, n; |
35f3d14d | 1251 | |
906f9040 LT |
1252 | bufs = kcalloc(nr_slots, sizeof(*bufs), |
1253 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); | |
c73be61c DH |
1254 | if (unlikely(!bufs)) |
1255 | return -ENOMEM; | |
35f3d14d | 1256 | |
189b0ddc DH |
1257 | spin_lock_irq(&pipe->rd_wait.lock); |
1258 | mask = pipe->ring_size - 1; | |
1259 | head = pipe->head; | |
1260 | tail = pipe->tail; | |
1261 | ||
1262 | n = pipe_occupancy(head, tail); | |
1263 | if (nr_slots < n) { | |
1264 | spin_unlock_irq(&pipe->rd_wait.lock); | |
1265 | kfree(bufs); | |
1266 | return -EBUSY; | |
1267 | } | |
1268 | ||
35f3d14d JA |
1269 | /* |
1270 | * The pipe array wraps around, so just start the new one at zero | |
8cefc107 | 1271 | * and adjust the indices. |
35f3d14d | 1272 | */ |
8cefc107 DH |
1273 | if (n > 0) { |
1274 | unsigned int h = head & mask; | |
1275 | unsigned int t = tail & mask; | |
1276 | if (h > t) { | |
1277 | memcpy(bufs, pipe->bufs + t, | |
1278 | n * sizeof(struct pipe_buffer)); | |
1279 | } else { | |
1280 | unsigned int tsize = pipe->ring_size - t; | |
1281 | if (h > 0) | |
1282 | memcpy(bufs + tsize, pipe->bufs, | |
1283 | h * sizeof(struct pipe_buffer)); | |
1284 | memcpy(bufs, pipe->bufs + t, | |
1285 | tsize * sizeof(struct pipe_buffer)); | |
1286 | } | |
35f3d14d JA |
1287 | } |
1288 | ||
8cefc107 DH |
1289 | head = n; |
1290 | tail = 0; | |
1291 | ||
906f9040 | 1292 | kfree(pipe->bufs); |
35f3d14d | 1293 | pipe->bufs = bufs; |
8cefc107 | 1294 | pipe->ring_size = nr_slots; |
c73be61c DH |
1295 | if (pipe->max_usage > nr_slots) |
1296 | pipe->max_usage = nr_slots; | |
8cefc107 DH |
1297 | pipe->tail = tail; |
1298 | pipe->head = head; | |
6551d5c5 | 1299 | |
189b0ddc DH |
1300 | spin_unlock_irq(&pipe->rd_wait.lock); |
1301 | ||
6551d5c5 LT |
1302 | /* This might have made more room for writers */ |
1303 | wake_up_interruptible(&pipe->wr_wait); | |
c73be61c DH |
1304 | return 0; |
1305 | } | |
1306 | ||
1307 | /* | |
1308 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1309 | * pipe size if successful, or return -ERROR on error. | |
1310 | */ | |
515c5046 | 1311 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned int arg) |
c73be61c DH |
1312 | { |
1313 | unsigned long user_bufs; | |
1314 | unsigned int nr_slots, size; | |
1315 | long ret = 0; | |
1316 | ||
b4bd6b4b | 1317 | if (pipe_has_watch_queue(pipe)) |
c73be61c | 1318 | return -EBUSY; |
c73be61c DH |
1319 | |
1320 | size = round_pipe_size(arg); | |
1321 | nr_slots = size >> PAGE_SHIFT; | |
1322 | ||
1323 | if (!nr_slots) | |
1324 | return -EINVAL; | |
1325 | ||
1326 | /* | |
1327 | * If trying to increase the pipe capacity, check that an | |
1328 | * unprivileged user is not trying to exceed various limits | |
1329 | * (soft limit check here, hard limit check just below). | |
1330 | * Decreasing the pipe capacity is always permitted, even | |
1331 | * if the user is currently over a limit. | |
1332 | */ | |
1333 | if (nr_slots > pipe->max_usage && | |
1334 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) | |
1335 | return -EPERM; | |
1336 | ||
1337 | user_bufs = account_pipe_buffers(pipe->user, pipe->nr_accounted, nr_slots); | |
1338 | ||
1339 | if (nr_slots > pipe->max_usage && | |
1340 | (too_many_pipe_buffers_hard(user_bufs) || | |
1341 | too_many_pipe_buffers_soft(user_bufs)) && | |
1342 | pipe_is_unprivileged_user()) { | |
1343 | ret = -EPERM; | |
1344 | goto out_revert_acct; | |
1345 | } | |
1346 | ||
1347 | ret = pipe_resize_ring(pipe, nr_slots); | |
1348 | if (ret < 0) | |
1349 | goto out_revert_acct; | |
1350 | ||
1351 | pipe->max_usage = nr_slots; | |
1352 | pipe->nr_accounted = nr_slots; | |
6718b6f8 | 1353 | return pipe->max_usage * PAGE_SIZE; |
b0b91d18 MK |
1354 | |
1355 | out_revert_acct: | |
c73be61c | 1356 | (void) account_pipe_buffers(pipe->user, nr_slots, pipe->nr_accounted); |
b0b91d18 | 1357 | return ret; |
35f3d14d JA |
1358 | } |
1359 | ||
72083646 | 1360 | /* |
4e7b5671 CH |
1361 | * Note that i_pipe and i_cdev share the same location, so checking ->i_pipe is |
1362 | * not enough to verify that this is a pipe. | |
72083646 | 1363 | */ |
c73be61c | 1364 | struct pipe_inode_info *get_pipe_info(struct file *file, bool for_splice) |
72083646 | 1365 | { |
c73be61c DH |
1366 | struct pipe_inode_info *pipe = file->private_data; |
1367 | ||
1368 | if (file->f_op != &pipefifo_fops || !pipe) | |
1369 | return NULL; | |
b4bd6b4b | 1370 | if (for_splice && pipe_has_watch_queue(pipe)) |
c73be61c | 1371 | return NULL; |
c73be61c | 1372 | return pipe; |
72083646 LT |
1373 | } |
1374 | ||
515c5046 | 1375 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned int arg) |
35f3d14d JA |
1376 | { |
1377 | struct pipe_inode_info *pipe; | |
1378 | long ret; | |
1379 | ||
c73be61c | 1380 | pipe = get_pipe_info(file, false); |
35f3d14d JA |
1381 | if (!pipe) |
1382 | return -EBADF; | |
1383 | ||
ebec73f4 | 1384 | __pipe_lock(pipe); |
35f3d14d JA |
1385 | |
1386 | switch (cmd) { | |
d37d4166 MK |
1387 | case F_SETPIPE_SZ: |
1388 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1389 | break; |
1390 | case F_GETPIPE_SZ: | |
6718b6f8 | 1391 | ret = pipe->max_usage * PAGE_SIZE; |
35f3d14d JA |
1392 | break; |
1393 | default: | |
1394 | ret = -EINVAL; | |
1395 | break; | |
1396 | } | |
1397 | ||
ebec73f4 | 1398 | __pipe_unlock(pipe); |
35f3d14d JA |
1399 | return ret; |
1400 | } | |
1401 | ||
ff0c7d15 NP |
1402 | static const struct super_operations pipefs_ops = { |
1403 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1404 | .statfs = simple_statfs, |
ff0c7d15 NP |
1405 | }; |
1406 | ||
1da177e4 LT |
1407 | /* |
1408 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1409 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1410 | * any operations on the root directory. However, we need a non-trivial | |
1411 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1412 | */ | |
4fa7ec5d DH |
1413 | |
1414 | static int pipefs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1415 | { |
4fa7ec5d DH |
1416 | struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC); |
1417 | if (!ctx) | |
1418 | return -ENOMEM; | |
1419 | ctx->ops = &pipefs_ops; | |
1420 | ctx->dops = &pipefs_dentry_operations; | |
1421 | return 0; | |
1da177e4 LT |
1422 | } |
1423 | ||
1424 | static struct file_system_type pipe_fs_type = { | |
1425 | .name = "pipefs", | |
4fa7ec5d | 1426 | .init_fs_context = pipefs_init_fs_context, |
1da177e4 LT |
1427 | .kill_sb = kill_anon_super, |
1428 | }; | |
1429 | ||
1998f193 LC |
1430 | #ifdef CONFIG_SYSCTL |
1431 | static int do_proc_dopipe_max_size_conv(unsigned long *lvalp, | |
1432 | unsigned int *valp, | |
1433 | int write, void *data) | |
1434 | { | |
1435 | if (write) { | |
1436 | unsigned int val; | |
1437 | ||
1438 | val = round_pipe_size(*lvalp); | |
1439 | if (val == 0) | |
1440 | return -EINVAL; | |
1441 | ||
1442 | *valp = val; | |
1443 | } else { | |
1444 | unsigned int val = *valp; | |
1445 | *lvalp = (unsigned long) val; | |
1446 | } | |
1447 | ||
1448 | return 0; | |
1449 | } | |
1450 | ||
1451 | static int proc_dopipe_max_size(struct ctl_table *table, int write, | |
1452 | void *buffer, size_t *lenp, loff_t *ppos) | |
1453 | { | |
1454 | return do_proc_douintvec(table, write, buffer, lenp, ppos, | |
1455 | do_proc_dopipe_max_size_conv, NULL); | |
1456 | } | |
1457 | ||
1458 | static struct ctl_table fs_pipe_sysctls[] = { | |
1459 | { | |
1460 | .procname = "pipe-max-size", | |
1461 | .data = &pipe_max_size, | |
1462 | .maxlen = sizeof(pipe_max_size), | |
1463 | .mode = 0644, | |
1464 | .proc_handler = proc_dopipe_max_size, | |
1465 | }, | |
1466 | { | |
1467 | .procname = "pipe-user-pages-hard", | |
1468 | .data = &pipe_user_pages_hard, | |
1469 | .maxlen = sizeof(pipe_user_pages_hard), | |
1470 | .mode = 0644, | |
1471 | .proc_handler = proc_doulongvec_minmax, | |
1472 | }, | |
1473 | { | |
1474 | .procname = "pipe-user-pages-soft", | |
1475 | .data = &pipe_user_pages_soft, | |
1476 | .maxlen = sizeof(pipe_user_pages_soft), | |
1477 | .mode = 0644, | |
1478 | .proc_handler = proc_doulongvec_minmax, | |
1479 | }, | |
1480 | { } | |
1481 | }; | |
1482 | #endif | |
1483 | ||
1da177e4 LT |
1484 | static int __init init_pipe_fs(void) |
1485 | { | |
1486 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1487 | |
1da177e4 LT |
1488 | if (!err) { |
1489 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1490 | if (IS_ERR(pipe_mnt)) { | |
1491 | err = PTR_ERR(pipe_mnt); | |
1492 | unregister_filesystem(&pipe_fs_type); | |
1493 | } | |
1494 | } | |
1998f193 LC |
1495 | #ifdef CONFIG_SYSCTL |
1496 | register_sysctl_init("fs", fs_pipe_sysctls); | |
1497 | #endif | |
1da177e4 LT |
1498 | return err; |
1499 | } | |
1500 | ||
1da177e4 | 1501 | fs_initcall(init_pipe_fs); |