]>
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> |
b502bd11 | 17 | #include <linux/magic.h> |
1da177e4 LT |
18 | #include <linux/pipe_fs_i.h> |
19 | #include <linux/uio.h> | |
20 | #include <linux/highmem.h> | |
5274f052 | 21 | #include <linux/pagemap.h> |
db349509 | 22 | #include <linux/audit.h> |
ba719bae | 23 | #include <linux/syscalls.h> |
b492e95b | 24 | #include <linux/fcntl.h> |
d86133bd | 25 | #include <linux/memcontrol.h> |
1da177e4 | 26 | |
7c0f6ba6 | 27 | #include <linux/uaccess.h> |
1da177e4 LT |
28 | #include <asm/ioctls.h> |
29 | ||
599a0ac1 AV |
30 | #include "internal.h" |
31 | ||
b492e95b JA |
32 | /* |
33 | * The max size that a non-root user is allowed to grow the pipe. Can | |
ff9da691 | 34 | * be set by root in /proc/sys/fs/pipe-max-size |
b492e95b | 35 | */ |
ff9da691 JA |
36 | unsigned int pipe_max_size = 1048576; |
37 | ||
759c0114 WT |
38 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
39 | * matches default values. | |
40 | */ | |
41 | unsigned long pipe_user_pages_hard; | |
42 | unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; | |
43 | ||
1da177e4 LT |
44 | /* |
45 | * We use a start+len construction, which provides full use of the | |
46 | * allocated memory. | |
47 | * -- Florian Coosmann (FGC) | |
48 | * | |
49 | * Reads with count = 0 should always return 0. | |
50 | * -- Julian Bradfield 1999-06-07. | |
51 | * | |
52 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
53 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
54 | * | |
55 | * pipe_read & write cleanup | |
56 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
57 | */ | |
58 | ||
61e0d47c MS |
59 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
60 | { | |
6447a3cf | 61 | if (pipe->files) |
72b0d9aa | 62 | mutex_lock_nested(&pipe->mutex, subclass); |
61e0d47c MS |
63 | } |
64 | ||
65 | void pipe_lock(struct pipe_inode_info *pipe) | |
66 | { | |
67 | /* | |
68 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
69 | */ | |
70 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
71 | } | |
72 | EXPORT_SYMBOL(pipe_lock); | |
73 | ||
74 | void pipe_unlock(struct pipe_inode_info *pipe) | |
75 | { | |
6447a3cf | 76 | if (pipe->files) |
72b0d9aa | 77 | mutex_unlock(&pipe->mutex); |
61e0d47c MS |
78 | } |
79 | EXPORT_SYMBOL(pipe_unlock); | |
80 | ||
ebec73f4 AV |
81 | static inline void __pipe_lock(struct pipe_inode_info *pipe) |
82 | { | |
83 | mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); | |
84 | } | |
85 | ||
86 | static inline void __pipe_unlock(struct pipe_inode_info *pipe) | |
87 | { | |
88 | mutex_unlock(&pipe->mutex); | |
89 | } | |
90 | ||
61e0d47c MS |
91 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
92 | struct pipe_inode_info *pipe2) | |
93 | { | |
94 | BUG_ON(pipe1 == pipe2); | |
95 | ||
96 | if (pipe1 < pipe2) { | |
97 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
98 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
99 | } else { | |
023d43c7 PZ |
100 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
101 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
102 | } |
103 | } | |
104 | ||
1da177e4 | 105 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
3a326a2c | 106 | void pipe_wait(struct pipe_inode_info *pipe) |
1da177e4 LT |
107 | { |
108 | DEFINE_WAIT(wait); | |
109 | ||
d79fc0fc IM |
110 | /* |
111 | * Pipes are system-local resources, so sleeping on them | |
112 | * is considered a noninteractive wait: | |
113 | */ | |
af927232 | 114 | prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE); |
61e0d47c | 115 | pipe_unlock(pipe); |
1da177e4 | 116 | schedule(); |
3a326a2c | 117 | finish_wait(&pipe->wait, &wait); |
61e0d47c | 118 | pipe_lock(pipe); |
1da177e4 LT |
119 | } |
120 | ||
341b446b IM |
121 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
122 | struct pipe_buffer *buf) | |
1da177e4 LT |
123 | { |
124 | struct page *page = buf->page; | |
125 | ||
5274f052 JA |
126 | /* |
127 | * If nobody else uses this page, and we don't already have a | |
128 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 129 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 130 | */ |
341b446b | 131 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 132 | pipe->tmp_page = page; |
341b446b | 133 | else |
09cbfeaf | 134 | put_page(page); |
1da177e4 LT |
135 | } |
136 | ||
d86133bd VD |
137 | static int anon_pipe_buf_steal(struct pipe_inode_info *pipe, |
138 | struct pipe_buffer *buf) | |
139 | { | |
140 | struct page *page = buf->page; | |
141 | ||
142 | if (page_count(page) == 1) { | |
c4159a75 | 143 | if (memcg_kmem_enabled()) |
d86133bd | 144 | memcg_kmem_uncharge(page, 0); |
d86133bd VD |
145 | __SetPageLocked(page); |
146 | return 0; | |
147 | } | |
148 | return 1; | |
149 | } | |
150 | ||
0845718d | 151 | /** |
b51d63c6 | 152 | * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer |
0845718d JA |
153 | * @pipe: the pipe that the buffer belongs to |
154 | * @buf: the buffer to attempt to steal | |
155 | * | |
156 | * Description: | |
b51d63c6 | 157 | * This function attempts to steal the &struct page attached to |
0845718d JA |
158 | * @buf. If successful, this function returns 0 and returns with |
159 | * the page locked. The caller may then reuse the page for whatever | |
b51d63c6 | 160 | * he wishes; the typical use is insertion into a different file |
0845718d JA |
161 | * page cache. |
162 | */ | |
330ab716 JA |
163 | int generic_pipe_buf_steal(struct pipe_inode_info *pipe, |
164 | struct pipe_buffer *buf) | |
5abc97aa | 165 | { |
46e678c9 JA |
166 | struct page *page = buf->page; |
167 | ||
0845718d JA |
168 | /* |
169 | * A reference of one is golden, that means that the owner of this | |
170 | * page is the only one holding a reference to it. lock the page | |
171 | * and return OK. | |
172 | */ | |
46e678c9 | 173 | if (page_count(page) == 1) { |
46e678c9 JA |
174 | lock_page(page); |
175 | return 0; | |
176 | } | |
177 | ||
178 | return 1; | |
5abc97aa | 179 | } |
51921cb7 | 180 | EXPORT_SYMBOL(generic_pipe_buf_steal); |
5abc97aa | 181 | |
0845718d | 182 | /** |
b51d63c6 | 183 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
184 | * @pipe: the pipe that the buffer belongs to |
185 | * @buf: the buffer to get a reference to | |
186 | * | |
187 | * Description: | |
188 | * This function grabs an extra reference to @buf. It's used in | |
189 | * in the tee() system call, when we duplicate the buffers in one | |
190 | * pipe into another. | |
191 | */ | |
192 | void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) | |
70524490 | 193 | { |
09cbfeaf | 194 | get_page(buf->page); |
70524490 | 195 | } |
51921cb7 | 196 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 197 | |
0845718d JA |
198 | /** |
199 | * generic_pipe_buf_confirm - verify contents of the pipe buffer | |
79685b8d | 200 | * @info: the pipe that the buffer belongs to |
0845718d JA |
201 | * @buf: the buffer to confirm |
202 | * | |
203 | * Description: | |
204 | * This function does nothing, because the generic pipe code uses | |
205 | * pages that are always good when inserted into the pipe. | |
206 | */ | |
cac36bb0 JA |
207 | int generic_pipe_buf_confirm(struct pipe_inode_info *info, |
208 | struct pipe_buffer *buf) | |
f84d7519 JA |
209 | { |
210 | return 0; | |
211 | } | |
51921cb7 | 212 | EXPORT_SYMBOL(generic_pipe_buf_confirm); |
f84d7519 | 213 | |
6818173b MS |
214 | /** |
215 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
216 | * @pipe: the pipe that the buffer belongs to | |
217 | * @buf: the buffer to put a reference to | |
218 | * | |
219 | * Description: | |
220 | * This function releases a reference to @buf. | |
221 | */ | |
222 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
223 | struct pipe_buffer *buf) | |
224 | { | |
09cbfeaf | 225 | put_page(buf->page); |
6818173b | 226 | } |
51921cb7 | 227 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 228 | |
d4c3cca9 | 229 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
1da177e4 | 230 | .can_merge = 1, |
cac36bb0 | 231 | .confirm = generic_pipe_buf_confirm, |
1da177e4 | 232 | .release = anon_pipe_buf_release, |
d86133bd | 233 | .steal = anon_pipe_buf_steal, |
f84d7519 | 234 | .get = generic_pipe_buf_get, |
1da177e4 LT |
235 | }; |
236 | ||
9883035a LT |
237 | static const struct pipe_buf_operations packet_pipe_buf_ops = { |
238 | .can_merge = 0, | |
9883035a LT |
239 | .confirm = generic_pipe_buf_confirm, |
240 | .release = anon_pipe_buf_release, | |
d86133bd | 241 | .steal = anon_pipe_buf_steal, |
9883035a LT |
242 | .get = generic_pipe_buf_get, |
243 | }; | |
244 | ||
1da177e4 | 245 | static ssize_t |
fb9096a3 | 246 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 247 | { |
fb9096a3 | 248 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 249 | struct file *filp = iocb->ki_filp; |
de32ec4c | 250 | struct pipe_inode_info *pipe = filp->private_data; |
1da177e4 LT |
251 | int do_wakeup; |
252 | ssize_t ret; | |
1da177e4 | 253 | |
1da177e4 LT |
254 | /* Null read succeeds. */ |
255 | if (unlikely(total_len == 0)) | |
256 | return 0; | |
257 | ||
258 | do_wakeup = 0; | |
259 | ret = 0; | |
ebec73f4 | 260 | __pipe_lock(pipe); |
1da177e4 | 261 | for (;;) { |
923f4f23 | 262 | int bufs = pipe->nrbufs; |
1da177e4 | 263 | if (bufs) { |
923f4f23 IM |
264 | int curbuf = pipe->curbuf; |
265 | struct pipe_buffer *buf = pipe->bufs + curbuf; | |
1da177e4 | 266 | size_t chars = buf->len; |
637b58c2 AV |
267 | size_t written; |
268 | int error; | |
1da177e4 LT |
269 | |
270 | if (chars > total_len) | |
271 | chars = total_len; | |
272 | ||
fba597db | 273 | error = pipe_buf_confirm(pipe, buf); |
f84d7519 | 274 | if (error) { |
5274f052 | 275 | if (!ret) |
e5953cbd | 276 | ret = error; |
5274f052 JA |
277 | break; |
278 | } | |
f84d7519 | 279 | |
fb9096a3 | 280 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 281 | if (unlikely(written < chars)) { |
341b446b | 282 | if (!ret) |
637b58c2 | 283 | ret = -EFAULT; |
1da177e4 LT |
284 | break; |
285 | } | |
286 | ret += chars; | |
287 | buf->offset += chars; | |
288 | buf->len -= chars; | |
9883035a LT |
289 | |
290 | /* Was it a packet buffer? Clean up and exit */ | |
291 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
292 | total_len = chars; | |
293 | buf->len = 0; | |
294 | } | |
295 | ||
1da177e4 | 296 | if (!buf->len) { |
a779638c | 297 | pipe_buf_release(pipe, buf); |
35f3d14d | 298 | curbuf = (curbuf + 1) & (pipe->buffers - 1); |
923f4f23 IM |
299 | pipe->curbuf = curbuf; |
300 | pipe->nrbufs = --bufs; | |
1da177e4 LT |
301 | do_wakeup = 1; |
302 | } | |
303 | total_len -= chars; | |
304 | if (!total_len) | |
305 | break; /* common path: read succeeded */ | |
306 | } | |
307 | if (bufs) /* More to do? */ | |
308 | continue; | |
923f4f23 | 309 | if (!pipe->writers) |
1da177e4 | 310 | break; |
923f4f23 | 311 | if (!pipe->waiting_writers) { |
1da177e4 LT |
312 | /* syscall merging: Usually we must not sleep |
313 | * if O_NONBLOCK is set, or if we got some data. | |
314 | * But if a writer sleeps in kernel space, then | |
315 | * we can wait for that data without violating POSIX. | |
316 | */ | |
317 | if (ret) | |
318 | break; | |
319 | if (filp->f_flags & O_NONBLOCK) { | |
320 | ret = -EAGAIN; | |
321 | break; | |
322 | } | |
323 | } | |
324 | if (signal_pending(current)) { | |
341b446b IM |
325 | if (!ret) |
326 | ret = -ERESTARTSYS; | |
1da177e4 LT |
327 | break; |
328 | } | |
329 | if (do_wakeup) { | |
28e58ee8 | 330 | wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM); |
923f4f23 | 331 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 | 332 | } |
923f4f23 | 333 | pipe_wait(pipe); |
1da177e4 | 334 | } |
ebec73f4 | 335 | __pipe_unlock(pipe); |
341b446b IM |
336 | |
337 | /* Signal writers asynchronously that there is more room. */ | |
1da177e4 | 338 | if (do_wakeup) { |
28e58ee8 | 339 | wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM); |
923f4f23 | 340 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 LT |
341 | } |
342 | if (ret > 0) | |
343 | file_accessed(filp); | |
344 | return ret; | |
345 | } | |
346 | ||
9883035a LT |
347 | static inline int is_packetized(struct file *file) |
348 | { | |
349 | return (file->f_flags & O_DIRECT) != 0; | |
350 | } | |
351 | ||
1da177e4 | 352 | static ssize_t |
f0d1bec9 | 353 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 354 | { |
ee0b3e67 | 355 | struct file *filp = iocb->ki_filp; |
de32ec4c | 356 | struct pipe_inode_info *pipe = filp->private_data; |
f0d1bec9 AV |
357 | ssize_t ret = 0; |
358 | int do_wakeup = 0; | |
359 | size_t total_len = iov_iter_count(from); | |
1da177e4 LT |
360 | ssize_t chars; |
361 | ||
1da177e4 LT |
362 | /* Null write succeeds. */ |
363 | if (unlikely(total_len == 0)) | |
364 | return 0; | |
365 | ||
ebec73f4 | 366 | __pipe_lock(pipe); |
1da177e4 | 367 | |
923f4f23 | 368 | if (!pipe->readers) { |
1da177e4 LT |
369 | send_sig(SIGPIPE, current, 0); |
370 | ret = -EPIPE; | |
371 | goto out; | |
372 | } | |
373 | ||
374 | /* We try to merge small writes */ | |
375 | chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */ | |
923f4f23 | 376 | if (pipe->nrbufs && chars != 0) { |
341b446b | 377 | int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) & |
35f3d14d | 378 | (pipe->buffers - 1); |
923f4f23 | 379 | struct pipe_buffer *buf = pipe->bufs + lastbuf; |
1da177e4 | 380 | int offset = buf->offset + buf->len; |
341b446b | 381 | |
fba597db MS |
382 | if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) { |
383 | ret = pipe_buf_confirm(pipe, buf); | |
6ae08069 | 384 | if (ret) |
5274f052 | 385 | goto out; |
f84d7519 | 386 | |
f0d1bec9 AV |
387 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
388 | if (unlikely(ret < chars)) { | |
6ae08069 | 389 | ret = -EFAULT; |
1da177e4 | 390 | goto out; |
f6762b7a | 391 | } |
f0d1bec9 | 392 | do_wakeup = 1; |
6ae08069 | 393 | buf->len += ret; |
f0d1bec9 | 394 | if (!iov_iter_count(from)) |
1da177e4 LT |
395 | goto out; |
396 | } | |
397 | } | |
398 | ||
399 | for (;;) { | |
400 | int bufs; | |
341b446b | 401 | |
923f4f23 | 402 | if (!pipe->readers) { |
1da177e4 | 403 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
404 | if (!ret) |
405 | ret = -EPIPE; | |
1da177e4 LT |
406 | break; |
407 | } | |
923f4f23 | 408 | bufs = pipe->nrbufs; |
35f3d14d JA |
409 | if (bufs < pipe->buffers) { |
410 | int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1); | |
923f4f23 IM |
411 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
412 | struct page *page = pipe->tmp_page; | |
f0d1bec9 | 413 | int copied; |
1da177e4 LT |
414 | |
415 | if (!page) { | |
d86133bd | 416 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
417 | if (unlikely(!page)) { |
418 | ret = ret ? : -ENOMEM; | |
419 | break; | |
420 | } | |
923f4f23 | 421 | pipe->tmp_page = page; |
1da177e4 | 422 | } |
341b446b | 423 | /* Always wake up, even if the copy fails. Otherwise |
1da177e4 LT |
424 | * we lock up (O_NONBLOCK-)readers that sleep due to |
425 | * syscall merging. | |
426 | * FIXME! Is this really true? | |
427 | */ | |
428 | do_wakeup = 1; | |
f0d1bec9 AV |
429 | copied = copy_page_from_iter(page, 0, PAGE_SIZE, from); |
430 | if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) { | |
341b446b | 431 | if (!ret) |
f0d1bec9 | 432 | ret = -EFAULT; |
1da177e4 LT |
433 | break; |
434 | } | |
f0d1bec9 | 435 | ret += copied; |
1da177e4 LT |
436 | |
437 | /* Insert it into the buffer array */ | |
438 | buf->page = page; | |
439 | buf->ops = &anon_pipe_buf_ops; | |
440 | buf->offset = 0; | |
f0d1bec9 | 441 | buf->len = copied; |
9883035a LT |
442 | buf->flags = 0; |
443 | if (is_packetized(filp)) { | |
444 | buf->ops = &packet_pipe_buf_ops; | |
445 | buf->flags = PIPE_BUF_FLAG_PACKET; | |
446 | } | |
923f4f23 IM |
447 | pipe->nrbufs = ++bufs; |
448 | pipe->tmp_page = NULL; | |
1da177e4 | 449 | |
f0d1bec9 | 450 | if (!iov_iter_count(from)) |
1da177e4 LT |
451 | break; |
452 | } | |
35f3d14d | 453 | if (bufs < pipe->buffers) |
1da177e4 LT |
454 | continue; |
455 | if (filp->f_flags & O_NONBLOCK) { | |
341b446b IM |
456 | if (!ret) |
457 | ret = -EAGAIN; | |
1da177e4 LT |
458 | break; |
459 | } | |
460 | if (signal_pending(current)) { | |
341b446b IM |
461 | if (!ret) |
462 | ret = -ERESTARTSYS; | |
1da177e4 LT |
463 | break; |
464 | } | |
465 | if (do_wakeup) { | |
28e58ee8 | 466 | wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM); |
923f4f23 | 467 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 LT |
468 | do_wakeup = 0; |
469 | } | |
923f4f23 IM |
470 | pipe->waiting_writers++; |
471 | pipe_wait(pipe); | |
472 | pipe->waiting_writers--; | |
1da177e4 LT |
473 | } |
474 | out: | |
ebec73f4 | 475 | __pipe_unlock(pipe); |
1da177e4 | 476 | if (do_wakeup) { |
28e58ee8 | 477 | wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM); |
923f4f23 | 478 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 | 479 | } |
7e775f46 | 480 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
481 | int err = file_update_time(filp); |
482 | if (err) | |
483 | ret = err; | |
7e775f46 | 484 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 485 | } |
1da177e4 LT |
486 | return ret; |
487 | } | |
488 | ||
d59d0b1b | 489 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 490 | { |
de32ec4c | 491 | struct pipe_inode_info *pipe = filp->private_data; |
1da177e4 LT |
492 | int count, buf, nrbufs; |
493 | ||
494 | switch (cmd) { | |
495 | case FIONREAD: | |
ebec73f4 | 496 | __pipe_lock(pipe); |
1da177e4 | 497 | count = 0; |
923f4f23 IM |
498 | buf = pipe->curbuf; |
499 | nrbufs = pipe->nrbufs; | |
1da177e4 | 500 | while (--nrbufs >= 0) { |
923f4f23 | 501 | count += pipe->bufs[buf].len; |
35f3d14d | 502 | buf = (buf+1) & (pipe->buffers - 1); |
1da177e4 | 503 | } |
ebec73f4 | 504 | __pipe_unlock(pipe); |
923f4f23 | 505 | |
1da177e4 LT |
506 | return put_user(count, (int __user *)arg); |
507 | default: | |
46ce341b | 508 | return -ENOIOCTLCMD; |
1da177e4 LT |
509 | } |
510 | } | |
511 | ||
512 | /* No kernel lock held - fine */ | |
076ccb76 | 513 | static __poll_t |
1da177e4 LT |
514 | pipe_poll(struct file *filp, poll_table *wait) |
515 | { | |
076ccb76 | 516 | __poll_t mask; |
de32ec4c | 517 | struct pipe_inode_info *pipe = filp->private_data; |
1da177e4 LT |
518 | int nrbufs; |
519 | ||
923f4f23 | 520 | poll_wait(filp, &pipe->wait, wait); |
1da177e4 LT |
521 | |
522 | /* Reading only -- no need for acquiring the semaphore. */ | |
923f4f23 | 523 | nrbufs = pipe->nrbufs; |
1da177e4 LT |
524 | mask = 0; |
525 | if (filp->f_mode & FMODE_READ) { | |
526 | mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0; | |
923f4f23 | 527 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
1da177e4 LT |
528 | mask |= POLLHUP; |
529 | } | |
530 | ||
531 | if (filp->f_mode & FMODE_WRITE) { | |
35f3d14d | 532 | mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0; |
5e5d7a22 PE |
533 | /* |
534 | * Most Unices do not set POLLERR for FIFOs but on Linux they | |
535 | * behave exactly like pipes for poll(). | |
536 | */ | |
923f4f23 | 537 | if (!pipe->readers) |
1da177e4 LT |
538 | mask |= POLLERR; |
539 | } | |
540 | ||
541 | return mask; | |
542 | } | |
543 | ||
b0d8d229 LT |
544 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
545 | { | |
546 | int kill = 0; | |
547 | ||
548 | spin_lock(&inode->i_lock); | |
549 | if (!--pipe->files) { | |
550 | inode->i_pipe = NULL; | |
551 | kill = 1; | |
552 | } | |
553 | spin_unlock(&inode->i_lock); | |
554 | ||
555 | if (kill) | |
556 | free_pipe_info(pipe); | |
557 | } | |
558 | ||
1da177e4 | 559 | static int |
599a0ac1 | 560 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 561 | { |
b0d8d229 | 562 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 563 | |
ebec73f4 | 564 | __pipe_lock(pipe); |
599a0ac1 AV |
565 | if (file->f_mode & FMODE_READ) |
566 | pipe->readers--; | |
567 | if (file->f_mode & FMODE_WRITE) | |
568 | pipe->writers--; | |
341b446b | 569 | |
ba5bb147 | 570 | if (pipe->readers || pipe->writers) { |
28e58ee8 | 571 | wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP); |
923f4f23 IM |
572 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
573 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 574 | } |
ebec73f4 | 575 | __pipe_unlock(pipe); |
ba5bb147 | 576 | |
b0d8d229 | 577 | put_pipe_info(inode, pipe); |
1da177e4 LT |
578 | return 0; |
579 | } | |
580 | ||
581 | static int | |
599a0ac1 | 582 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 583 | { |
de32ec4c | 584 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 585 | int retval = 0; |
1da177e4 | 586 | |
ebec73f4 | 587 | __pipe_lock(pipe); |
599a0ac1 AV |
588 | if (filp->f_mode & FMODE_READ) |
589 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
590 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 591 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
592 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
593 | /* this can happen only if on == T */ | |
e5bc49ba ON |
594 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
595 | } | |
ebec73f4 | 596 | __pipe_unlock(pipe); |
60aa4924 | 597 | return retval; |
1da177e4 LT |
598 | } |
599 | ||
9c87bcf0 | 600 | static unsigned long account_pipe_buffers(struct user_struct *user, |
759c0114 WT |
601 | unsigned long old, unsigned long new) |
602 | { | |
9c87bcf0 | 603 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
604 | } |
605 | ||
9c87bcf0 | 606 | static bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 607 | { |
9903a91c | 608 | return pipe_user_pages_soft && user_bufs > pipe_user_pages_soft; |
759c0114 WT |
609 | } |
610 | ||
9c87bcf0 | 611 | static bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 612 | { |
9903a91c | 613 | return pipe_user_pages_hard && user_bufs > pipe_user_pages_hard; |
759c0114 WT |
614 | } |
615 | ||
85c2dd54 EB |
616 | static bool is_unprivileged_user(void) |
617 | { | |
618 | return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN); | |
619 | } | |
620 | ||
7bee130e | 621 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 622 | { |
923f4f23 | 623 | struct pipe_inode_info *pipe; |
09b4d199 MK |
624 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
625 | struct user_struct *user = get_current_user(); | |
9c87bcf0 | 626 | unsigned long user_bufs; |
3a326a2c | 627 | |
d86133bd | 628 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
09b4d199 MK |
629 | if (pipe == NULL) |
630 | goto out_free_uid; | |
631 | ||
086e774a MK |
632 | if (pipe_bufs * PAGE_SIZE > pipe_max_size && !capable(CAP_SYS_RESOURCE)) |
633 | pipe_bufs = pipe_max_size >> PAGE_SHIFT; | |
634 | ||
9c87bcf0 | 635 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
a005ca0e | 636 | |
85c2dd54 | 637 | if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) { |
9c87bcf0 | 638 | user_bufs = account_pipe_buffers(user, pipe_bufs, 1); |
a005ca0e | 639 | pipe_bufs = 1; |
09b4d199 | 640 | } |
759c0114 | 641 | |
85c2dd54 | 642 | if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user()) |
a005ca0e MK |
643 | goto out_revert_acct; |
644 | ||
645 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), | |
646 | GFP_KERNEL_ACCOUNT); | |
647 | ||
09b4d199 MK |
648 | if (pipe->bufs) { |
649 | init_waitqueue_head(&pipe->wait); | |
650 | pipe->r_counter = pipe->w_counter = 1; | |
651 | pipe->buffers = pipe_bufs; | |
652 | pipe->user = user; | |
09b4d199 MK |
653 | mutex_init(&pipe->mutex); |
654 | return pipe; | |
3a326a2c IM |
655 | } |
656 | ||
a005ca0e | 657 | out_revert_acct: |
9c87bcf0 | 658 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
09b4d199 MK |
659 | kfree(pipe); |
660 | out_free_uid: | |
661 | free_uid(user); | |
35f3d14d | 662 | return NULL; |
3a326a2c IM |
663 | } |
664 | ||
4b8a8f1e | 665 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
666 | { |
667 | int i; | |
1da177e4 | 668 | |
9c87bcf0 | 669 | (void) account_pipe_buffers(pipe->user, pipe->buffers, 0); |
759c0114 | 670 | free_uid(pipe->user); |
35f3d14d | 671 | for (i = 0; i < pipe->buffers; i++) { |
923f4f23 | 672 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 673 | if (buf->ops) |
a779638c | 674 | pipe_buf_release(pipe, buf); |
1da177e4 | 675 | } |
923f4f23 IM |
676 | if (pipe->tmp_page) |
677 | __free_page(pipe->tmp_page); | |
35f3d14d | 678 | kfree(pipe->bufs); |
923f4f23 | 679 | kfree(pipe); |
1da177e4 LT |
680 | } |
681 | ||
fa3536cc | 682 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 683 | |
c23fbb6b ED |
684 | /* |
685 | * pipefs_dname() is called from d_path(). | |
686 | */ | |
687 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
688 | { | |
689 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
75c3cfa8 | 690 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
691 | } |
692 | ||
3ba13d17 | 693 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 694 | .d_dname = pipefs_dname, |
1da177e4 LT |
695 | }; |
696 | ||
697 | static struct inode * get_pipe_inode(void) | |
698 | { | |
a209dfc7 | 699 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 700 | struct pipe_inode_info *pipe; |
1da177e4 LT |
701 | |
702 | if (!inode) | |
703 | goto fail_inode; | |
704 | ||
85fe4025 CH |
705 | inode->i_ino = get_next_ino(); |
706 | ||
7bee130e | 707 | pipe = alloc_pipe_info(); |
923f4f23 | 708 | if (!pipe) |
1da177e4 | 709 | goto fail_iput; |
3a326a2c | 710 | |
ba5bb147 AV |
711 | inode->i_pipe = pipe; |
712 | pipe->files = 2; | |
923f4f23 | 713 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 714 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
715 | |
716 | /* | |
717 | * Mark the inode dirty from the very beginning, | |
718 | * that way it will never be moved to the dirty | |
719 | * list because "mark_inode_dirty()" will think | |
720 | * that it already _is_ on the dirty list. | |
721 | */ | |
722 | inode->i_state = I_DIRTY; | |
723 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
724 | inode->i_uid = current_fsuid(); |
725 | inode->i_gid = current_fsgid(); | |
078cd827 | 726 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
923f4f23 | 727 | |
1da177e4 LT |
728 | return inode; |
729 | ||
730 | fail_iput: | |
731 | iput(inode); | |
341b446b | 732 | |
1da177e4 LT |
733 | fail_inode: |
734 | return NULL; | |
735 | } | |
736 | ||
e4fad8e5 | 737 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 738 | { |
d6cbd281 | 739 | int err; |
e4fad8e5 | 740 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 741 | struct file *f; |
2c48b9c4 | 742 | struct path path; |
1da177e4 | 743 | |
1da177e4 | 744 | if (!inode) |
e4fad8e5 | 745 | return -ENFILE; |
1da177e4 | 746 | |
d6cbd281 | 747 | err = -ENOMEM; |
cdf01226 | 748 | path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &empty_name); |
2c48b9c4 | 749 | if (!path.dentry) |
d6cbd281 | 750 | goto err_inode; |
2c48b9c4 | 751 | path.mnt = mntget(pipe_mnt); |
341b446b | 752 | |
2c48b9c4 | 753 | d_instantiate(path.dentry, inode); |
430e285e | 754 | |
599a0ac1 | 755 | f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops); |
e9bb1f9b EB |
756 | if (IS_ERR(f)) { |
757 | err = PTR_ERR(f); | |
430e285e | 758 | goto err_dentry; |
e9bb1f9b | 759 | } |
341b446b | 760 | |
9883035a | 761 | f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)); |
de32ec4c | 762 | f->private_data = inode->i_pipe; |
d6cbd281 | 763 | |
599a0ac1 | 764 | res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops); |
e9bb1f9b EB |
765 | if (IS_ERR(res[0])) { |
766 | err = PTR_ERR(res[0]); | |
e4fad8e5 | 767 | goto err_file; |
e9bb1f9b | 768 | } |
e4fad8e5 AV |
769 | |
770 | path_get(&path); | |
de32ec4c | 771 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 AV |
772 | res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK); |
773 | res[1] = f; | |
774 | return 0; | |
1da177e4 | 775 | |
e4fad8e5 AV |
776 | err_file: |
777 | put_filp(f); | |
778 | err_dentry: | |
4b8a8f1e | 779 | free_pipe_info(inode->i_pipe); |
2c48b9c4 | 780 | path_put(&path); |
e4fad8e5 | 781 | return err; |
ed152437 | 782 | |
e4fad8e5 | 783 | err_inode: |
4b8a8f1e | 784 | free_pipe_info(inode->i_pipe); |
1da177e4 | 785 | iput(inode); |
e4fad8e5 | 786 | return err; |
d6cbd281 AK |
787 | } |
788 | ||
5b249b1b | 789 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 790 | { |
d6cbd281 AK |
791 | int error; |
792 | int fdw, fdr; | |
793 | ||
9883035a | 794 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT)) |
ed8cae8b UD |
795 | return -EINVAL; |
796 | ||
e4fad8e5 AV |
797 | error = create_pipe_files(files, flags); |
798 | if (error) | |
799 | return error; | |
d6cbd281 | 800 | |
ed8cae8b | 801 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
802 | if (error < 0) |
803 | goto err_read_pipe; | |
804 | fdr = error; | |
805 | ||
ed8cae8b | 806 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
807 | if (error < 0) |
808 | goto err_fdr; | |
809 | fdw = error; | |
810 | ||
157cf649 | 811 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
812 | fd[0] = fdr; |
813 | fd[1] = fdw; | |
d6cbd281 AK |
814 | return 0; |
815 | ||
816 | err_fdr: | |
817 | put_unused_fd(fdr); | |
818 | err_read_pipe: | |
e4fad8e5 AV |
819 | fput(files[0]); |
820 | fput(files[1]); | |
d6cbd281 | 821 | return error; |
1da177e4 LT |
822 | } |
823 | ||
5b249b1b AV |
824 | int do_pipe_flags(int *fd, int flags) |
825 | { | |
826 | struct file *files[2]; | |
827 | int error = __do_pipe_flags(fd, files, flags); | |
828 | if (!error) { | |
829 | fd_install(fd[0], files[0]); | |
830 | fd_install(fd[1], files[1]); | |
831 | } | |
832 | return error; | |
833 | } | |
834 | ||
d35c7b0e UD |
835 | /* |
836 | * sys_pipe() is the normal C calling standard for creating | |
837 | * a pipe. It's not the way Unix traditionally does this, though. | |
838 | */ | |
d4e82042 | 839 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
d35c7b0e | 840 | { |
5b249b1b | 841 | struct file *files[2]; |
d35c7b0e UD |
842 | int fd[2]; |
843 | int error; | |
844 | ||
5b249b1b | 845 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 846 | if (!error) { |
5b249b1b AV |
847 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
848 | fput(files[0]); | |
849 | fput(files[1]); | |
850 | put_unused_fd(fd[0]); | |
851 | put_unused_fd(fd[1]); | |
d35c7b0e | 852 | error = -EFAULT; |
5b249b1b AV |
853 | } else { |
854 | fd_install(fd[0], files[0]); | |
855 | fd_install(fd[1], files[1]); | |
ba719bae | 856 | } |
d35c7b0e UD |
857 | } |
858 | return error; | |
859 | } | |
860 | ||
2b664219 | 861 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b UD |
862 | { |
863 | return sys_pipe2(fildes, 0); | |
864 | } | |
865 | ||
fc7478a2 | 866 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 AV |
867 | { |
868 | int cur = *cnt; | |
869 | ||
870 | while (cur == *cnt) { | |
fc7478a2 | 871 | pipe_wait(pipe); |
f776c738 AV |
872 | if (signal_pending(current)) |
873 | break; | |
874 | } | |
875 | return cur == *cnt ? -ERESTARTSYS : 0; | |
876 | } | |
877 | ||
fc7478a2 | 878 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 879 | { |
fc7478a2 | 880 | wake_up_interruptible(&pipe->wait); |
f776c738 AV |
881 | } |
882 | ||
883 | static int fifo_open(struct inode *inode, struct file *filp) | |
884 | { | |
885 | struct pipe_inode_info *pipe; | |
599a0ac1 | 886 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
887 | int ret; |
888 | ||
ba5bb147 AV |
889 | filp->f_version = 0; |
890 | ||
891 | spin_lock(&inode->i_lock); | |
892 | if (inode->i_pipe) { | |
893 | pipe = inode->i_pipe; | |
894 | pipe->files++; | |
895 | spin_unlock(&inode->i_lock); | |
896 | } else { | |
897 | spin_unlock(&inode->i_lock); | |
7bee130e | 898 | pipe = alloc_pipe_info(); |
f776c738 | 899 | if (!pipe) |
ba5bb147 AV |
900 | return -ENOMEM; |
901 | pipe->files = 1; | |
902 | spin_lock(&inode->i_lock); | |
903 | if (unlikely(inode->i_pipe)) { | |
904 | inode->i_pipe->files++; | |
905 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 906 | free_pipe_info(pipe); |
ba5bb147 AV |
907 | pipe = inode->i_pipe; |
908 | } else { | |
909 | inode->i_pipe = pipe; | |
910 | spin_unlock(&inode->i_lock); | |
911 | } | |
f776c738 | 912 | } |
de32ec4c | 913 | filp->private_data = pipe; |
ba5bb147 AV |
914 | /* OK, we have a pipe and it's pinned down */ |
915 | ||
ebec73f4 | 916 | __pipe_lock(pipe); |
f776c738 AV |
917 | |
918 | /* We can only do regular read/write on fifos */ | |
919 | filp->f_mode &= (FMODE_READ | FMODE_WRITE); | |
920 | ||
921 | switch (filp->f_mode) { | |
922 | case FMODE_READ: | |
923 | /* | |
924 | * O_RDONLY | |
925 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
926 | * opened, even when there is no process writing the FIFO. | |
927 | */ | |
f776c738 AV |
928 | pipe->r_counter++; |
929 | if (pipe->readers++ == 0) | |
fc7478a2 | 930 | wake_up_partner(pipe); |
f776c738 | 931 | |
599a0ac1 | 932 | if (!is_pipe && !pipe->writers) { |
f776c738 AV |
933 | if ((filp->f_flags & O_NONBLOCK)) { |
934 | /* suppress POLLHUP until we have | |
935 | * seen a writer */ | |
936 | filp->f_version = pipe->w_counter; | |
937 | } else { | |
fc7478a2 | 938 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
939 | goto err_rd; |
940 | } | |
941 | } | |
942 | break; | |
943 | ||
944 | case FMODE_WRITE: | |
945 | /* | |
946 | * O_WRONLY | |
947 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
948 | * errno=ENXIO when there is no process reading the FIFO. | |
949 | */ | |
950 | ret = -ENXIO; | |
599a0ac1 | 951 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
952 | goto err; |
953 | ||
f776c738 AV |
954 | pipe->w_counter++; |
955 | if (!pipe->writers++) | |
fc7478a2 | 956 | wake_up_partner(pipe); |
f776c738 | 957 | |
599a0ac1 | 958 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 959 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
960 | goto err_wr; |
961 | } | |
962 | break; | |
963 | ||
964 | case FMODE_READ | FMODE_WRITE: | |
965 | /* | |
966 | * O_RDWR | |
967 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
968 | * This implementation will NEVER block on a O_RDWR open, since | |
969 | * the process can at least talk to itself. | |
970 | */ | |
f776c738 AV |
971 | |
972 | pipe->readers++; | |
973 | pipe->writers++; | |
974 | pipe->r_counter++; | |
975 | pipe->w_counter++; | |
976 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 977 | wake_up_partner(pipe); |
f776c738 AV |
978 | break; |
979 | ||
980 | default: | |
981 | ret = -EINVAL; | |
982 | goto err; | |
983 | } | |
984 | ||
985 | /* Ok! */ | |
ebec73f4 | 986 | __pipe_unlock(pipe); |
f776c738 AV |
987 | return 0; |
988 | ||
989 | err_rd: | |
990 | if (!--pipe->readers) | |
991 | wake_up_interruptible(&pipe->wait); | |
992 | ret = -ERESTARTSYS; | |
993 | goto err; | |
994 | ||
995 | err_wr: | |
996 | if (!--pipe->writers) | |
997 | wake_up_interruptible(&pipe->wait); | |
998 | ret = -ERESTARTSYS; | |
999 | goto err; | |
1000 | ||
1001 | err: | |
ebec73f4 | 1002 | __pipe_unlock(pipe); |
b0d8d229 LT |
1003 | |
1004 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1005 | return ret; |
1006 | } | |
1007 | ||
599a0ac1 AV |
1008 | const struct file_operations pipefifo_fops = { |
1009 | .open = fifo_open, | |
1010 | .llseek = no_llseek, | |
fb9096a3 | 1011 | .read_iter = pipe_read, |
f0d1bec9 | 1012 | .write_iter = pipe_write, |
599a0ac1 AV |
1013 | .poll = pipe_poll, |
1014 | .unlocked_ioctl = pipe_ioctl, | |
1015 | .release = pipe_release, | |
1016 | .fasync = pipe_fasync, | |
f776c738 AV |
1017 | }; |
1018 | ||
f491bd71 MK |
1019 | /* |
1020 | * Currently we rely on the pipe array holding a power-of-2 number | |
d3f14c48 | 1021 | * of pages. Returns 0 on error. |
f491bd71 | 1022 | */ |
7a8d1819 | 1023 | unsigned int round_pipe_size(unsigned int size) |
f491bd71 MK |
1024 | { |
1025 | unsigned long nr_pages; | |
1026 | ||
4c2e4bef EB |
1027 | /* Minimum pipe size, as required by POSIX */ |
1028 | if (size < PAGE_SIZE) | |
1029 | size = PAGE_SIZE; | |
d3f14c48 | 1030 | |
f491bd71 | 1031 | nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; |
d3f14c48 JL |
1032 | if (nr_pages == 0) |
1033 | return 0; | |
1034 | ||
f491bd71 MK |
1035 | return roundup_pow_of_two(nr_pages) << PAGE_SHIFT; |
1036 | } | |
1037 | ||
35f3d14d JA |
1038 | /* |
1039 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1040 | * pipe size if successful, or return -ERROR on error. | |
1041 | */ | |
d37d4166 | 1042 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) |
35f3d14d JA |
1043 | { |
1044 | struct pipe_buffer *bufs; | |
d37d4166 | 1045 | unsigned int size, nr_pages; |
9c87bcf0 | 1046 | unsigned long user_bufs; |
b0b91d18 | 1047 | long ret = 0; |
d37d4166 MK |
1048 | |
1049 | size = round_pipe_size(arg); | |
d3f14c48 JL |
1050 | if (size == 0) |
1051 | return -EINVAL; | |
d37d4166 MK |
1052 | nr_pages = size >> PAGE_SHIFT; |
1053 | ||
1054 | if (!nr_pages) | |
1055 | return -EINVAL; | |
1056 | ||
b0b91d18 MK |
1057 | /* |
1058 | * If trying to increase the pipe capacity, check that an | |
1059 | * unprivileged user is not trying to exceed various limits | |
1060 | * (soft limit check here, hard limit check just below). | |
1061 | * Decreasing the pipe capacity is always permitted, even | |
1062 | * if the user is currently over a limit. | |
1063 | */ | |
1064 | if (nr_pages > pipe->buffers && | |
1065 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) | |
d37d4166 MK |
1066 | return -EPERM; |
1067 | ||
9c87bcf0 | 1068 | user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages); |
b0b91d18 MK |
1069 | |
1070 | if (nr_pages > pipe->buffers && | |
9c87bcf0 MK |
1071 | (too_many_pipe_buffers_hard(user_bufs) || |
1072 | too_many_pipe_buffers_soft(user_bufs)) && | |
85c2dd54 | 1073 | is_unprivileged_user()) { |
b0b91d18 MK |
1074 | ret = -EPERM; |
1075 | goto out_revert_acct; | |
1076 | } | |
35f3d14d | 1077 | |
35f3d14d JA |
1078 | /* |
1079 | * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't | |
1080 | * expect a lot of shrink+grow operations, just free and allocate | |
1081 | * again like we would do for growing. If the pipe currently | |
1082 | * contains more buffers than arg, then return busy. | |
1083 | */ | |
b0b91d18 MK |
1084 | if (nr_pages < pipe->nrbufs) { |
1085 | ret = -EBUSY; | |
1086 | goto out_revert_acct; | |
1087 | } | |
35f3d14d | 1088 | |
d86133bd VD |
1089 | bufs = kcalloc(nr_pages, sizeof(*bufs), |
1090 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); | |
b0b91d18 MK |
1091 | if (unlikely(!bufs)) { |
1092 | ret = -ENOMEM; | |
1093 | goto out_revert_acct; | |
1094 | } | |
35f3d14d JA |
1095 | |
1096 | /* | |
1097 | * The pipe array wraps around, so just start the new one at zero | |
1098 | * and adjust the indexes. | |
1099 | */ | |
1100 | if (pipe->nrbufs) { | |
1d862f41 MS |
1101 | unsigned int tail; |
1102 | unsigned int head; | |
35f3d14d | 1103 | |
1d862f41 MS |
1104 | tail = pipe->curbuf + pipe->nrbufs; |
1105 | if (tail < pipe->buffers) | |
1106 | tail = 0; | |
1107 | else | |
1108 | tail &= (pipe->buffers - 1); | |
1109 | ||
1110 | head = pipe->nrbufs - tail; | |
35f3d14d JA |
1111 | if (head) |
1112 | memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer)); | |
1113 | if (tail) | |
1d862f41 | 1114 | memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer)); |
35f3d14d JA |
1115 | } |
1116 | ||
1117 | pipe->curbuf = 0; | |
1118 | kfree(pipe->bufs); | |
1119 | pipe->bufs = bufs; | |
b9598db3 JA |
1120 | pipe->buffers = nr_pages; |
1121 | return nr_pages * PAGE_SIZE; | |
b0b91d18 MK |
1122 | |
1123 | out_revert_acct: | |
9c87bcf0 | 1124 | (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers); |
b0b91d18 | 1125 | return ret; |
35f3d14d JA |
1126 | } |
1127 | ||
72083646 LT |
1128 | /* |
1129 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1130 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1131 | * pipe. | |
1132 | */ | |
1133 | struct pipe_inode_info *get_pipe_info(struct file *file) | |
1134 | { | |
de32ec4c | 1135 | return file->f_op == &pipefifo_fops ? file->private_data : NULL; |
72083646 LT |
1136 | } |
1137 | ||
35f3d14d JA |
1138 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
1139 | { | |
1140 | struct pipe_inode_info *pipe; | |
1141 | long ret; | |
1142 | ||
c66fb347 | 1143 | pipe = get_pipe_info(file); |
35f3d14d JA |
1144 | if (!pipe) |
1145 | return -EBADF; | |
1146 | ||
ebec73f4 | 1147 | __pipe_lock(pipe); |
35f3d14d JA |
1148 | |
1149 | switch (cmd) { | |
d37d4166 MK |
1150 | case F_SETPIPE_SZ: |
1151 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1152 | break; |
1153 | case F_GETPIPE_SZ: | |
b9598db3 | 1154 | ret = pipe->buffers * PAGE_SIZE; |
35f3d14d JA |
1155 | break; |
1156 | default: | |
1157 | ret = -EINVAL; | |
1158 | break; | |
1159 | } | |
1160 | ||
ebec73f4 | 1161 | __pipe_unlock(pipe); |
35f3d14d JA |
1162 | return ret; |
1163 | } | |
1164 | ||
ff0c7d15 NP |
1165 | static const struct super_operations pipefs_ops = { |
1166 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1167 | .statfs = simple_statfs, |
ff0c7d15 NP |
1168 | }; |
1169 | ||
1da177e4 LT |
1170 | /* |
1171 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1172 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1173 | * any operations on the root directory. However, we need a non-trivial | |
1174 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1175 | */ | |
51139ada AV |
1176 | static struct dentry *pipefs_mount(struct file_system_type *fs_type, |
1177 | int flags, const char *dev_name, void *data) | |
1da177e4 | 1178 | { |
c74a1cbb AV |
1179 | return mount_pseudo(fs_type, "pipe:", &pipefs_ops, |
1180 | &pipefs_dentry_operations, PIPEFS_MAGIC); | |
1da177e4 LT |
1181 | } |
1182 | ||
1183 | static struct file_system_type pipe_fs_type = { | |
1184 | .name = "pipefs", | |
51139ada | 1185 | .mount = pipefs_mount, |
1da177e4 LT |
1186 | .kill_sb = kill_anon_super, |
1187 | }; | |
1188 | ||
1189 | static int __init init_pipe_fs(void) | |
1190 | { | |
1191 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1192 | |
1da177e4 LT |
1193 | if (!err) { |
1194 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1195 | if (IS_ERR(pipe_mnt)) { | |
1196 | err = PTR_ERR(pipe_mnt); | |
1197 | unregister_filesystem(&pipe_fs_type); | |
1198 | } | |
1199 | } | |
1200 | return err; | |
1201 | } | |
1202 | ||
1da177e4 | 1203 | fs_initcall(init_pipe_fs); |