Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/pipe.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | #include <linux/mm.h> | |
8 | #include <linux/file.h> | |
9 | #include <linux/poll.h> | |
10 | #include <linux/slab.h> | |
11 | #include <linux/module.h> | |
12 | #include <linux/init.h> | |
13 | #include <linux/fs.h> | |
35f3d14d | 14 | #include <linux/log2.h> |
1da177e4 | 15 | #include <linux/mount.h> |
b502bd11 | 16 | #include <linux/magic.h> |
1da177e4 LT |
17 | #include <linux/pipe_fs_i.h> |
18 | #include <linux/uio.h> | |
19 | #include <linux/highmem.h> | |
5274f052 | 20 | #include <linux/pagemap.h> |
db349509 | 21 | #include <linux/audit.h> |
ba719bae | 22 | #include <linux/syscalls.h> |
b492e95b | 23 | #include <linux/fcntl.h> |
d86133bd | 24 | #include <linux/memcontrol.h> |
1da177e4 LT |
25 | |
26 | #include <asm/uaccess.h> | |
27 | #include <asm/ioctls.h> | |
28 | ||
599a0ac1 AV |
29 | #include "internal.h" |
30 | ||
b492e95b JA |
31 | /* |
32 | * The max size that a non-root user is allowed to grow the pipe. Can | |
ff9da691 | 33 | * be set by root in /proc/sys/fs/pipe-max-size |
b492e95b | 34 | */ |
ff9da691 JA |
35 | unsigned int pipe_max_size = 1048576; |
36 | ||
37 | /* | |
38 | * Minimum pipe size, as required by POSIX | |
39 | */ | |
40 | unsigned int pipe_min_size = PAGE_SIZE; | |
b492e95b | 41 | |
759c0114 WT |
42 | /* Maximum allocatable pages per user. Hard limit is unset by default, soft |
43 | * matches default values. | |
44 | */ | |
45 | unsigned long pipe_user_pages_hard; | |
46 | unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR; | |
47 | ||
1da177e4 LT |
48 | /* |
49 | * We use a start+len construction, which provides full use of the | |
50 | * allocated memory. | |
51 | * -- Florian Coosmann (FGC) | |
52 | * | |
53 | * Reads with count = 0 should always return 0. | |
54 | * -- Julian Bradfield 1999-06-07. | |
55 | * | |
56 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
57 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
58 | * | |
59 | * pipe_read & write cleanup | |
60 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
61 | */ | |
62 | ||
61e0d47c MS |
63 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
64 | { | |
6447a3cf | 65 | if (pipe->files) |
72b0d9aa | 66 | mutex_lock_nested(&pipe->mutex, subclass); |
61e0d47c MS |
67 | } |
68 | ||
69 | void pipe_lock(struct pipe_inode_info *pipe) | |
70 | { | |
71 | /* | |
72 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
73 | */ | |
74 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
75 | } | |
76 | EXPORT_SYMBOL(pipe_lock); | |
77 | ||
78 | void pipe_unlock(struct pipe_inode_info *pipe) | |
79 | { | |
6447a3cf | 80 | if (pipe->files) |
72b0d9aa | 81 | mutex_unlock(&pipe->mutex); |
61e0d47c MS |
82 | } |
83 | EXPORT_SYMBOL(pipe_unlock); | |
84 | ||
ebec73f4 AV |
85 | static inline void __pipe_lock(struct pipe_inode_info *pipe) |
86 | { | |
87 | mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT); | |
88 | } | |
89 | ||
90 | static inline void __pipe_unlock(struct pipe_inode_info *pipe) | |
91 | { | |
92 | mutex_unlock(&pipe->mutex); | |
93 | } | |
94 | ||
61e0d47c MS |
95 | void pipe_double_lock(struct pipe_inode_info *pipe1, |
96 | struct pipe_inode_info *pipe2) | |
97 | { | |
98 | BUG_ON(pipe1 == pipe2); | |
99 | ||
100 | if (pipe1 < pipe2) { | |
101 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
102 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
103 | } else { | |
023d43c7 PZ |
104 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
105 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
106 | } |
107 | } | |
108 | ||
1da177e4 | 109 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
3a326a2c | 110 | void pipe_wait(struct pipe_inode_info *pipe) |
1da177e4 LT |
111 | { |
112 | DEFINE_WAIT(wait); | |
113 | ||
d79fc0fc IM |
114 | /* |
115 | * Pipes are system-local resources, so sleeping on them | |
116 | * is considered a noninteractive wait: | |
117 | */ | |
af927232 | 118 | prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE); |
61e0d47c | 119 | pipe_unlock(pipe); |
1da177e4 | 120 | schedule(); |
3a326a2c | 121 | finish_wait(&pipe->wait, &wait); |
61e0d47c | 122 | pipe_lock(pipe); |
1da177e4 LT |
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 | ||
d86133bd VD |
141 | static int anon_pipe_buf_steal(struct pipe_inode_info *pipe, |
142 | struct pipe_buffer *buf) | |
143 | { | |
144 | struct page *page = buf->page; | |
145 | ||
146 | if (page_count(page) == 1) { | |
c4159a75 | 147 | if (memcg_kmem_enabled()) |
d86133bd | 148 | memcg_kmem_uncharge(page, 0); |
d86133bd VD |
149 | __SetPageLocked(page); |
150 | return 0; | |
151 | } | |
152 | return 1; | |
153 | } | |
154 | ||
0845718d | 155 | /** |
b51d63c6 | 156 | * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer |
0845718d JA |
157 | * @pipe: the pipe that the buffer belongs to |
158 | * @buf: the buffer to attempt to steal | |
159 | * | |
160 | * Description: | |
b51d63c6 | 161 | * This function attempts to steal the &struct page attached to |
0845718d JA |
162 | * @buf. If successful, this function returns 0 and returns with |
163 | * the page locked. The caller may then reuse the page for whatever | |
b51d63c6 | 164 | * he wishes; the typical use is insertion into a different file |
0845718d JA |
165 | * page cache. |
166 | */ | |
330ab716 JA |
167 | int generic_pipe_buf_steal(struct pipe_inode_info *pipe, |
168 | struct pipe_buffer *buf) | |
5abc97aa | 169 | { |
46e678c9 JA |
170 | struct page *page = buf->page; |
171 | ||
0845718d JA |
172 | /* |
173 | * A reference of one is golden, that means that the owner of this | |
174 | * page is the only one holding a reference to it. lock the page | |
175 | * and return OK. | |
176 | */ | |
46e678c9 | 177 | if (page_count(page) == 1) { |
46e678c9 JA |
178 | lock_page(page); |
179 | return 0; | |
180 | } | |
181 | ||
182 | return 1; | |
5abc97aa | 183 | } |
51921cb7 | 184 | EXPORT_SYMBOL(generic_pipe_buf_steal); |
5abc97aa | 185 | |
0845718d | 186 | /** |
b51d63c6 | 187 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
188 | * @pipe: the pipe that the buffer belongs to |
189 | * @buf: the buffer to get a reference to | |
190 | * | |
191 | * Description: | |
192 | * This function grabs an extra reference to @buf. It's used in | |
193 | * in the tee() system call, when we duplicate the buffers in one | |
194 | * pipe into another. | |
195 | */ | |
196 | void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) | |
70524490 | 197 | { |
09cbfeaf | 198 | get_page(buf->page); |
70524490 | 199 | } |
51921cb7 | 200 | EXPORT_SYMBOL(generic_pipe_buf_get); |
70524490 | 201 | |
0845718d JA |
202 | /** |
203 | * generic_pipe_buf_confirm - verify contents of the pipe buffer | |
79685b8d | 204 | * @info: the pipe that the buffer belongs to |
0845718d JA |
205 | * @buf: the buffer to confirm |
206 | * | |
207 | * Description: | |
208 | * This function does nothing, because the generic pipe code uses | |
209 | * pages that are always good when inserted into the pipe. | |
210 | */ | |
cac36bb0 JA |
211 | int generic_pipe_buf_confirm(struct pipe_inode_info *info, |
212 | struct pipe_buffer *buf) | |
f84d7519 JA |
213 | { |
214 | return 0; | |
215 | } | |
51921cb7 | 216 | EXPORT_SYMBOL(generic_pipe_buf_confirm); |
f84d7519 | 217 | |
6818173b MS |
218 | /** |
219 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
220 | * @pipe: the pipe that the buffer belongs to | |
221 | * @buf: the buffer to put a reference to | |
222 | * | |
223 | * Description: | |
224 | * This function releases a reference to @buf. | |
225 | */ | |
226 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
227 | struct pipe_buffer *buf) | |
228 | { | |
09cbfeaf | 229 | put_page(buf->page); |
6818173b | 230 | } |
51921cb7 | 231 | EXPORT_SYMBOL(generic_pipe_buf_release); |
6818173b | 232 | |
d4c3cca9 | 233 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
1da177e4 | 234 | .can_merge = 1, |
cac36bb0 | 235 | .confirm = generic_pipe_buf_confirm, |
1da177e4 | 236 | .release = anon_pipe_buf_release, |
d86133bd | 237 | .steal = anon_pipe_buf_steal, |
f84d7519 | 238 | .get = generic_pipe_buf_get, |
1da177e4 LT |
239 | }; |
240 | ||
9883035a LT |
241 | static const struct pipe_buf_operations packet_pipe_buf_ops = { |
242 | .can_merge = 0, | |
9883035a LT |
243 | .confirm = generic_pipe_buf_confirm, |
244 | .release = anon_pipe_buf_release, | |
d86133bd | 245 | .steal = anon_pipe_buf_steal, |
9883035a LT |
246 | .get = generic_pipe_buf_get, |
247 | }; | |
248 | ||
1da177e4 | 249 | static ssize_t |
fb9096a3 | 250 | pipe_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 251 | { |
fb9096a3 | 252 | size_t total_len = iov_iter_count(to); |
ee0b3e67 | 253 | struct file *filp = iocb->ki_filp; |
de32ec4c | 254 | struct pipe_inode_info *pipe = filp->private_data; |
1da177e4 LT |
255 | int do_wakeup; |
256 | ssize_t ret; | |
1da177e4 | 257 | |
1da177e4 LT |
258 | /* Null read succeeds. */ |
259 | if (unlikely(total_len == 0)) | |
260 | return 0; | |
261 | ||
262 | do_wakeup = 0; | |
263 | ret = 0; | |
ebec73f4 | 264 | __pipe_lock(pipe); |
1da177e4 | 265 | for (;;) { |
923f4f23 | 266 | int bufs = pipe->nrbufs; |
1da177e4 | 267 | if (bufs) { |
923f4f23 IM |
268 | int curbuf = pipe->curbuf; |
269 | struct pipe_buffer *buf = pipe->bufs + curbuf; | |
d4c3cca9 | 270 | const struct pipe_buf_operations *ops = buf->ops; |
1da177e4 | 271 | size_t chars = buf->len; |
637b58c2 AV |
272 | size_t written; |
273 | int error; | |
1da177e4 LT |
274 | |
275 | if (chars > total_len) | |
276 | chars = total_len; | |
277 | ||
cac36bb0 | 278 | error = ops->confirm(pipe, buf); |
f84d7519 | 279 | if (error) { |
5274f052 | 280 | if (!ret) |
e5953cbd | 281 | ret = error; |
5274f052 JA |
282 | break; |
283 | } | |
f84d7519 | 284 | |
fb9096a3 | 285 | written = copy_page_to_iter(buf->page, buf->offset, chars, to); |
637b58c2 | 286 | if (unlikely(written < chars)) { |
341b446b | 287 | if (!ret) |
637b58c2 | 288 | ret = -EFAULT; |
1da177e4 LT |
289 | break; |
290 | } | |
291 | ret += chars; | |
292 | buf->offset += chars; | |
293 | buf->len -= chars; | |
9883035a LT |
294 | |
295 | /* Was it a packet buffer? Clean up and exit */ | |
296 | if (buf->flags & PIPE_BUF_FLAG_PACKET) { | |
297 | total_len = chars; | |
298 | buf->len = 0; | |
299 | } | |
300 | ||
1da177e4 LT |
301 | if (!buf->len) { |
302 | buf->ops = NULL; | |
923f4f23 | 303 | ops->release(pipe, buf); |
35f3d14d | 304 | curbuf = (curbuf + 1) & (pipe->buffers - 1); |
923f4f23 IM |
305 | pipe->curbuf = curbuf; |
306 | pipe->nrbufs = --bufs; | |
1da177e4 LT |
307 | do_wakeup = 1; |
308 | } | |
309 | total_len -= chars; | |
310 | if (!total_len) | |
311 | break; /* common path: read succeeded */ | |
312 | } | |
313 | if (bufs) /* More to do? */ | |
314 | continue; | |
923f4f23 | 315 | if (!pipe->writers) |
1da177e4 | 316 | break; |
923f4f23 | 317 | if (!pipe->waiting_writers) { |
1da177e4 LT |
318 | /* syscall merging: Usually we must not sleep |
319 | * if O_NONBLOCK is set, or if we got some data. | |
320 | * But if a writer sleeps in kernel space, then | |
321 | * we can wait for that data without violating POSIX. | |
322 | */ | |
323 | if (ret) | |
324 | break; | |
325 | if (filp->f_flags & O_NONBLOCK) { | |
326 | ret = -EAGAIN; | |
327 | break; | |
328 | } | |
329 | } | |
330 | if (signal_pending(current)) { | |
341b446b IM |
331 | if (!ret) |
332 | ret = -ERESTARTSYS; | |
1da177e4 LT |
333 | break; |
334 | } | |
335 | if (do_wakeup) { | |
28e58ee8 | 336 | wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM); |
923f4f23 | 337 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 | 338 | } |
923f4f23 | 339 | pipe_wait(pipe); |
1da177e4 | 340 | } |
ebec73f4 | 341 | __pipe_unlock(pipe); |
341b446b IM |
342 | |
343 | /* Signal writers asynchronously that there is more room. */ | |
1da177e4 | 344 | if (do_wakeup) { |
28e58ee8 | 345 | wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM); |
923f4f23 | 346 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 LT |
347 | } |
348 | if (ret > 0) | |
349 | file_accessed(filp); | |
350 | return ret; | |
351 | } | |
352 | ||
9883035a LT |
353 | static inline int is_packetized(struct file *file) |
354 | { | |
355 | return (file->f_flags & O_DIRECT) != 0; | |
356 | } | |
357 | ||
1da177e4 | 358 | static ssize_t |
f0d1bec9 | 359 | pipe_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 360 | { |
ee0b3e67 | 361 | struct file *filp = iocb->ki_filp; |
de32ec4c | 362 | struct pipe_inode_info *pipe = filp->private_data; |
f0d1bec9 AV |
363 | ssize_t ret = 0; |
364 | int do_wakeup = 0; | |
365 | size_t total_len = iov_iter_count(from); | |
1da177e4 LT |
366 | ssize_t chars; |
367 | ||
1da177e4 LT |
368 | /* Null write succeeds. */ |
369 | if (unlikely(total_len == 0)) | |
370 | return 0; | |
371 | ||
ebec73f4 | 372 | __pipe_lock(pipe); |
1da177e4 | 373 | |
923f4f23 | 374 | if (!pipe->readers) { |
1da177e4 LT |
375 | send_sig(SIGPIPE, current, 0); |
376 | ret = -EPIPE; | |
377 | goto out; | |
378 | } | |
379 | ||
380 | /* We try to merge small writes */ | |
381 | chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */ | |
923f4f23 | 382 | if (pipe->nrbufs && chars != 0) { |
341b446b | 383 | int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) & |
35f3d14d | 384 | (pipe->buffers - 1); |
923f4f23 | 385 | struct pipe_buffer *buf = pipe->bufs + lastbuf; |
d4c3cca9 | 386 | const struct pipe_buf_operations *ops = buf->ops; |
1da177e4 | 387 | int offset = buf->offset + buf->len; |
341b446b | 388 | |
1da177e4 | 389 | if (ops->can_merge && offset + chars <= PAGE_SIZE) { |
6ae08069 EB |
390 | ret = ops->confirm(pipe, buf); |
391 | if (ret) | |
5274f052 | 392 | goto out; |
f84d7519 | 393 | |
f0d1bec9 AV |
394 | ret = copy_page_from_iter(buf->page, offset, chars, from); |
395 | if (unlikely(ret < chars)) { | |
6ae08069 | 396 | ret = -EFAULT; |
1da177e4 | 397 | goto out; |
f6762b7a | 398 | } |
f0d1bec9 | 399 | do_wakeup = 1; |
6ae08069 | 400 | buf->len += ret; |
f0d1bec9 | 401 | if (!iov_iter_count(from)) |
1da177e4 LT |
402 | goto out; |
403 | } | |
404 | } | |
405 | ||
406 | for (;;) { | |
407 | int bufs; | |
341b446b | 408 | |
923f4f23 | 409 | if (!pipe->readers) { |
1da177e4 | 410 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
411 | if (!ret) |
412 | ret = -EPIPE; | |
1da177e4 LT |
413 | break; |
414 | } | |
923f4f23 | 415 | bufs = pipe->nrbufs; |
35f3d14d JA |
416 | if (bufs < pipe->buffers) { |
417 | int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1); | |
923f4f23 IM |
418 | struct pipe_buffer *buf = pipe->bufs + newbuf; |
419 | struct page *page = pipe->tmp_page; | |
f0d1bec9 | 420 | int copied; |
1da177e4 LT |
421 | |
422 | if (!page) { | |
d86133bd | 423 | page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT); |
1da177e4 LT |
424 | if (unlikely(!page)) { |
425 | ret = ret ? : -ENOMEM; | |
426 | break; | |
427 | } | |
923f4f23 | 428 | pipe->tmp_page = page; |
1da177e4 | 429 | } |
341b446b | 430 | /* Always wake up, even if the copy fails. Otherwise |
1da177e4 LT |
431 | * we lock up (O_NONBLOCK-)readers that sleep due to |
432 | * syscall merging. | |
433 | * FIXME! Is this really true? | |
434 | */ | |
435 | do_wakeup = 1; | |
f0d1bec9 AV |
436 | copied = copy_page_from_iter(page, 0, PAGE_SIZE, from); |
437 | if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) { | |
341b446b | 438 | if (!ret) |
f0d1bec9 | 439 | ret = -EFAULT; |
1da177e4 LT |
440 | break; |
441 | } | |
f0d1bec9 | 442 | ret += copied; |
1da177e4 LT |
443 | |
444 | /* Insert it into the buffer array */ | |
445 | buf->page = page; | |
446 | buf->ops = &anon_pipe_buf_ops; | |
447 | buf->offset = 0; | |
f0d1bec9 | 448 | buf->len = copied; |
9883035a LT |
449 | buf->flags = 0; |
450 | if (is_packetized(filp)) { | |
451 | buf->ops = &packet_pipe_buf_ops; | |
452 | buf->flags = PIPE_BUF_FLAG_PACKET; | |
453 | } | |
923f4f23 IM |
454 | pipe->nrbufs = ++bufs; |
455 | pipe->tmp_page = NULL; | |
1da177e4 | 456 | |
f0d1bec9 | 457 | if (!iov_iter_count(from)) |
1da177e4 LT |
458 | break; |
459 | } | |
35f3d14d | 460 | if (bufs < pipe->buffers) |
1da177e4 LT |
461 | continue; |
462 | if (filp->f_flags & O_NONBLOCK) { | |
341b446b IM |
463 | if (!ret) |
464 | ret = -EAGAIN; | |
1da177e4 LT |
465 | break; |
466 | } | |
467 | if (signal_pending(current)) { | |
341b446b IM |
468 | if (!ret) |
469 | ret = -ERESTARTSYS; | |
1da177e4 LT |
470 | break; |
471 | } | |
472 | if (do_wakeup) { | |
28e58ee8 | 473 | wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM); |
923f4f23 | 474 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 LT |
475 | do_wakeup = 0; |
476 | } | |
923f4f23 IM |
477 | pipe->waiting_writers++; |
478 | pipe_wait(pipe); | |
479 | pipe->waiting_writers--; | |
1da177e4 LT |
480 | } |
481 | out: | |
ebec73f4 | 482 | __pipe_unlock(pipe); |
1da177e4 | 483 | if (do_wakeup) { |
28e58ee8 | 484 | wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM); |
923f4f23 | 485 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 | 486 | } |
7e775f46 | 487 | if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) { |
c3b2da31 JB |
488 | int err = file_update_time(filp); |
489 | if (err) | |
490 | ret = err; | |
7e775f46 | 491 | sb_end_write(file_inode(filp)->i_sb); |
c3b2da31 | 492 | } |
1da177e4 LT |
493 | return ret; |
494 | } | |
495 | ||
d59d0b1b | 496 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 497 | { |
de32ec4c | 498 | struct pipe_inode_info *pipe = filp->private_data; |
1da177e4 LT |
499 | int count, buf, nrbufs; |
500 | ||
501 | switch (cmd) { | |
502 | case FIONREAD: | |
ebec73f4 | 503 | __pipe_lock(pipe); |
1da177e4 | 504 | count = 0; |
923f4f23 IM |
505 | buf = pipe->curbuf; |
506 | nrbufs = pipe->nrbufs; | |
1da177e4 | 507 | while (--nrbufs >= 0) { |
923f4f23 | 508 | count += pipe->bufs[buf].len; |
35f3d14d | 509 | buf = (buf+1) & (pipe->buffers - 1); |
1da177e4 | 510 | } |
ebec73f4 | 511 | __pipe_unlock(pipe); |
923f4f23 | 512 | |
1da177e4 LT |
513 | return put_user(count, (int __user *)arg); |
514 | default: | |
46ce341b | 515 | return -ENOIOCTLCMD; |
1da177e4 LT |
516 | } |
517 | } | |
518 | ||
519 | /* No kernel lock held - fine */ | |
520 | static unsigned int | |
521 | pipe_poll(struct file *filp, poll_table *wait) | |
522 | { | |
523 | unsigned int mask; | |
de32ec4c | 524 | struct pipe_inode_info *pipe = filp->private_data; |
1da177e4 LT |
525 | int nrbufs; |
526 | ||
923f4f23 | 527 | poll_wait(filp, &pipe->wait, wait); |
1da177e4 LT |
528 | |
529 | /* Reading only -- no need for acquiring the semaphore. */ | |
923f4f23 | 530 | nrbufs = pipe->nrbufs; |
1da177e4 LT |
531 | mask = 0; |
532 | if (filp->f_mode & FMODE_READ) { | |
533 | mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0; | |
923f4f23 | 534 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
1da177e4 LT |
535 | mask |= POLLHUP; |
536 | } | |
537 | ||
538 | if (filp->f_mode & FMODE_WRITE) { | |
35f3d14d | 539 | mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0; |
5e5d7a22 PE |
540 | /* |
541 | * Most Unices do not set POLLERR for FIFOs but on Linux they | |
542 | * behave exactly like pipes for poll(). | |
543 | */ | |
923f4f23 | 544 | if (!pipe->readers) |
1da177e4 LT |
545 | mask |= POLLERR; |
546 | } | |
547 | ||
548 | return mask; | |
549 | } | |
550 | ||
b0d8d229 LT |
551 | static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe) |
552 | { | |
553 | int kill = 0; | |
554 | ||
555 | spin_lock(&inode->i_lock); | |
556 | if (!--pipe->files) { | |
557 | inode->i_pipe = NULL; | |
558 | kill = 1; | |
559 | } | |
560 | spin_unlock(&inode->i_lock); | |
561 | ||
562 | if (kill) | |
563 | free_pipe_info(pipe); | |
564 | } | |
565 | ||
1da177e4 | 566 | static int |
599a0ac1 | 567 | pipe_release(struct inode *inode, struct file *file) |
1da177e4 | 568 | { |
b0d8d229 | 569 | struct pipe_inode_info *pipe = file->private_data; |
923f4f23 | 570 | |
ebec73f4 | 571 | __pipe_lock(pipe); |
599a0ac1 AV |
572 | if (file->f_mode & FMODE_READ) |
573 | pipe->readers--; | |
574 | if (file->f_mode & FMODE_WRITE) | |
575 | pipe->writers--; | |
341b446b | 576 | |
ba5bb147 | 577 | if (pipe->readers || pipe->writers) { |
28e58ee8 | 578 | wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP); |
923f4f23 IM |
579 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
580 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 581 | } |
ebec73f4 | 582 | __pipe_unlock(pipe); |
ba5bb147 | 583 | |
b0d8d229 | 584 | put_pipe_info(inode, pipe); |
1da177e4 LT |
585 | return 0; |
586 | } | |
587 | ||
588 | static int | |
599a0ac1 | 589 | pipe_fasync(int fd, struct file *filp, int on) |
1da177e4 | 590 | { |
de32ec4c | 591 | struct pipe_inode_info *pipe = filp->private_data; |
599a0ac1 | 592 | int retval = 0; |
1da177e4 | 593 | |
ebec73f4 | 594 | __pipe_lock(pipe); |
599a0ac1 AV |
595 | if (filp->f_mode & FMODE_READ) |
596 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); | |
597 | if ((filp->f_mode & FMODE_WRITE) && retval >= 0) { | |
341b446b | 598 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
599a0ac1 AV |
599 | if (retval < 0 && (filp->f_mode & FMODE_READ)) |
600 | /* this can happen only if on == T */ | |
e5bc49ba ON |
601 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); |
602 | } | |
ebec73f4 | 603 | __pipe_unlock(pipe); |
60aa4924 | 604 | return retval; |
1da177e4 LT |
605 | } |
606 | ||
6ab1c8f0 | 607 | static unsigned long account_pipe_buffers(struct user_struct *user, |
759c0114 WT |
608 | unsigned long old, unsigned long new) |
609 | { | |
6ab1c8f0 | 610 | return atomic_long_add_return(new - old, &user->pipe_bufs); |
759c0114 WT |
611 | } |
612 | ||
6ab1c8f0 | 613 | static bool too_many_pipe_buffers_soft(unsigned long user_bufs) |
759c0114 | 614 | { |
6ab1c8f0 | 615 | return pipe_user_pages_soft && user_bufs >= pipe_user_pages_soft; |
759c0114 WT |
616 | } |
617 | ||
6ab1c8f0 | 618 | static bool too_many_pipe_buffers_hard(unsigned long user_bufs) |
759c0114 | 619 | { |
6ab1c8f0 | 620 | return pipe_user_pages_hard && user_bufs >= pipe_user_pages_hard; |
759c0114 WT |
621 | } |
622 | ||
7bee130e | 623 | struct pipe_inode_info *alloc_pipe_info(void) |
3a326a2c | 624 | { |
923f4f23 | 625 | struct pipe_inode_info *pipe; |
7bb5e326 MK |
626 | unsigned long pipe_bufs = PIPE_DEF_BUFFERS; |
627 | struct user_struct *user = get_current_user(); | |
6ab1c8f0 | 628 | unsigned long user_bufs; |
3a326a2c | 629 | |
d86133bd | 630 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT); |
7bb5e326 MK |
631 | if (pipe == NULL) |
632 | goto out_free_uid; | |
633 | ||
68a5369f MK |
634 | if (pipe_bufs * PAGE_SIZE > pipe_max_size && !capable(CAP_SYS_RESOURCE)) |
635 | pipe_bufs = pipe_max_size >> PAGE_SHIFT; | |
636 | ||
6ab1c8f0 | 637 | user_bufs = account_pipe_buffers(user, 0, pipe_bufs); |
71a9509e | 638 | |
6ab1c8f0 MK |
639 | if (too_many_pipe_buffers_soft(user_bufs)) { |
640 | user_bufs = account_pipe_buffers(user, pipe_bufs, 1); | |
71a9509e | 641 | pipe_bufs = 1; |
7bb5e326 | 642 | } |
759c0114 | 643 | |
6ab1c8f0 | 644 | if (too_many_pipe_buffers_hard(user_bufs)) |
71a9509e MK |
645 | goto out_revert_acct; |
646 | ||
647 | pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer), | |
648 | GFP_KERNEL_ACCOUNT); | |
649 | ||
7bb5e326 MK |
650 | if (pipe->bufs) { |
651 | init_waitqueue_head(&pipe->wait); | |
652 | pipe->r_counter = pipe->w_counter = 1; | |
653 | pipe->buffers = pipe_bufs; | |
654 | pipe->user = user; | |
7bb5e326 MK |
655 | mutex_init(&pipe->mutex); |
656 | return pipe; | |
3a326a2c IM |
657 | } |
658 | ||
71a9509e | 659 | out_revert_acct: |
6ab1c8f0 | 660 | (void) account_pipe_buffers(user, pipe_bufs, 0); |
7bb5e326 MK |
661 | kfree(pipe); |
662 | out_free_uid: | |
663 | free_uid(user); | |
35f3d14d | 664 | return NULL; |
3a326a2c IM |
665 | } |
666 | ||
4b8a8f1e | 667 | void free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
668 | { |
669 | int i; | |
1da177e4 | 670 | |
6ab1c8f0 | 671 | (void) account_pipe_buffers(pipe->user, pipe->buffers, 0); |
759c0114 | 672 | free_uid(pipe->user); |
35f3d14d | 673 | for (i = 0; i < pipe->buffers; i++) { |
923f4f23 | 674 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 675 | if (buf->ops) |
923f4f23 | 676 | buf->ops->release(pipe, buf); |
1da177e4 | 677 | } |
923f4f23 IM |
678 | if (pipe->tmp_page) |
679 | __free_page(pipe->tmp_page); | |
35f3d14d | 680 | kfree(pipe->bufs); |
923f4f23 | 681 | kfree(pipe); |
1da177e4 LT |
682 | } |
683 | ||
fa3536cc | 684 | static struct vfsmount *pipe_mnt __read_mostly; |
341b446b | 685 | |
c23fbb6b ED |
686 | /* |
687 | * pipefs_dname() is called from d_path(). | |
688 | */ | |
689 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
690 | { | |
691 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
75c3cfa8 | 692 | d_inode(dentry)->i_ino); |
c23fbb6b ED |
693 | } |
694 | ||
3ba13d17 | 695 | static const struct dentry_operations pipefs_dentry_operations = { |
c23fbb6b | 696 | .d_dname = pipefs_dname, |
1da177e4 LT |
697 | }; |
698 | ||
699 | static struct inode * get_pipe_inode(void) | |
700 | { | |
a209dfc7 | 701 | struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb); |
923f4f23 | 702 | struct pipe_inode_info *pipe; |
1da177e4 LT |
703 | |
704 | if (!inode) | |
705 | goto fail_inode; | |
706 | ||
85fe4025 CH |
707 | inode->i_ino = get_next_ino(); |
708 | ||
7bee130e | 709 | pipe = alloc_pipe_info(); |
923f4f23 | 710 | if (!pipe) |
1da177e4 | 711 | goto fail_iput; |
3a326a2c | 712 | |
ba5bb147 AV |
713 | inode->i_pipe = pipe; |
714 | pipe->files = 2; | |
923f4f23 | 715 | pipe->readers = pipe->writers = 1; |
599a0ac1 | 716 | inode->i_fop = &pipefifo_fops; |
1da177e4 LT |
717 | |
718 | /* | |
719 | * Mark the inode dirty from the very beginning, | |
720 | * that way it will never be moved to the dirty | |
721 | * list because "mark_inode_dirty()" will think | |
722 | * that it already _is_ on the dirty list. | |
723 | */ | |
724 | inode->i_state = I_DIRTY; | |
725 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
726 | inode->i_uid = current_fsuid(); |
727 | inode->i_gid = current_fsgid(); | |
1da177e4 | 728 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
923f4f23 | 729 | |
1da177e4 LT |
730 | return inode; |
731 | ||
732 | fail_iput: | |
733 | iput(inode); | |
341b446b | 734 | |
1da177e4 LT |
735 | fail_inode: |
736 | return NULL; | |
737 | } | |
738 | ||
e4fad8e5 | 739 | int create_pipe_files(struct file **res, int flags) |
1da177e4 | 740 | { |
d6cbd281 | 741 | int err; |
e4fad8e5 | 742 | struct inode *inode = get_pipe_inode(); |
d6cbd281 | 743 | struct file *f; |
2c48b9c4 | 744 | struct path path; |
e4fad8e5 | 745 | static struct qstr name = { .name = "" }; |
1da177e4 | 746 | |
1da177e4 | 747 | if (!inode) |
e4fad8e5 | 748 | return -ENFILE; |
1da177e4 | 749 | |
d6cbd281 | 750 | err = -ENOMEM; |
4b936885 | 751 | path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name); |
2c48b9c4 | 752 | if (!path.dentry) |
d6cbd281 | 753 | goto err_inode; |
2c48b9c4 | 754 | path.mnt = mntget(pipe_mnt); |
341b446b | 755 | |
2c48b9c4 | 756 | d_instantiate(path.dentry, inode); |
430e285e | 757 | |
599a0ac1 | 758 | f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops); |
e9bb1f9b EB |
759 | if (IS_ERR(f)) { |
760 | err = PTR_ERR(f); | |
430e285e | 761 | goto err_dentry; |
e9bb1f9b | 762 | } |
341b446b | 763 | |
9883035a | 764 | f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)); |
de32ec4c | 765 | f->private_data = inode->i_pipe; |
d6cbd281 | 766 | |
599a0ac1 | 767 | res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops); |
e9bb1f9b EB |
768 | if (IS_ERR(res[0])) { |
769 | err = PTR_ERR(res[0]); | |
e4fad8e5 | 770 | goto err_file; |
e9bb1f9b | 771 | } |
e4fad8e5 AV |
772 | |
773 | path_get(&path); | |
de32ec4c | 774 | res[0]->private_data = inode->i_pipe; |
e4fad8e5 AV |
775 | res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK); |
776 | res[1] = f; | |
777 | return 0; | |
1da177e4 | 778 | |
e4fad8e5 AV |
779 | err_file: |
780 | put_filp(f); | |
781 | err_dentry: | |
4b8a8f1e | 782 | free_pipe_info(inode->i_pipe); |
2c48b9c4 | 783 | path_put(&path); |
e4fad8e5 | 784 | return err; |
ed152437 | 785 | |
e4fad8e5 | 786 | err_inode: |
4b8a8f1e | 787 | free_pipe_info(inode->i_pipe); |
1da177e4 | 788 | iput(inode); |
e4fad8e5 | 789 | return err; |
d6cbd281 AK |
790 | } |
791 | ||
5b249b1b | 792 | static int __do_pipe_flags(int *fd, struct file **files, int flags) |
d6cbd281 | 793 | { |
d6cbd281 AK |
794 | int error; |
795 | int fdw, fdr; | |
796 | ||
9883035a | 797 | if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT)) |
ed8cae8b UD |
798 | return -EINVAL; |
799 | ||
e4fad8e5 AV |
800 | error = create_pipe_files(files, flags); |
801 | if (error) | |
802 | return error; | |
d6cbd281 | 803 | |
ed8cae8b | 804 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
805 | if (error < 0) |
806 | goto err_read_pipe; | |
807 | fdr = error; | |
808 | ||
ed8cae8b | 809 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
810 | if (error < 0) |
811 | goto err_fdr; | |
812 | fdw = error; | |
813 | ||
157cf649 | 814 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
815 | fd[0] = fdr; |
816 | fd[1] = fdw; | |
d6cbd281 AK |
817 | return 0; |
818 | ||
819 | err_fdr: | |
820 | put_unused_fd(fdr); | |
821 | err_read_pipe: | |
e4fad8e5 AV |
822 | fput(files[0]); |
823 | fput(files[1]); | |
d6cbd281 | 824 | return error; |
1da177e4 LT |
825 | } |
826 | ||
5b249b1b AV |
827 | int do_pipe_flags(int *fd, int flags) |
828 | { | |
829 | struct file *files[2]; | |
830 | int error = __do_pipe_flags(fd, files, flags); | |
831 | if (!error) { | |
832 | fd_install(fd[0], files[0]); | |
833 | fd_install(fd[1], files[1]); | |
834 | } | |
835 | return error; | |
836 | } | |
837 | ||
d35c7b0e UD |
838 | /* |
839 | * sys_pipe() is the normal C calling standard for creating | |
840 | * a pipe. It's not the way Unix traditionally does this, though. | |
841 | */ | |
d4e82042 | 842 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
d35c7b0e | 843 | { |
5b249b1b | 844 | struct file *files[2]; |
d35c7b0e UD |
845 | int fd[2]; |
846 | int error; | |
847 | ||
5b249b1b | 848 | error = __do_pipe_flags(fd, files, flags); |
d35c7b0e | 849 | if (!error) { |
5b249b1b AV |
850 | if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) { |
851 | fput(files[0]); | |
852 | fput(files[1]); | |
853 | put_unused_fd(fd[0]); | |
854 | put_unused_fd(fd[1]); | |
d35c7b0e | 855 | error = -EFAULT; |
5b249b1b AV |
856 | } else { |
857 | fd_install(fd[0], files[0]); | |
858 | fd_install(fd[1], files[1]); | |
ba719bae | 859 | } |
d35c7b0e UD |
860 | } |
861 | return error; | |
862 | } | |
863 | ||
2b664219 | 864 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b UD |
865 | { |
866 | return sys_pipe2(fildes, 0); | |
867 | } | |
868 | ||
fc7478a2 | 869 | static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt) |
f776c738 AV |
870 | { |
871 | int cur = *cnt; | |
872 | ||
873 | while (cur == *cnt) { | |
fc7478a2 | 874 | pipe_wait(pipe); |
f776c738 AV |
875 | if (signal_pending(current)) |
876 | break; | |
877 | } | |
878 | return cur == *cnt ? -ERESTARTSYS : 0; | |
879 | } | |
880 | ||
fc7478a2 | 881 | static void wake_up_partner(struct pipe_inode_info *pipe) |
f776c738 | 882 | { |
fc7478a2 | 883 | wake_up_interruptible(&pipe->wait); |
f776c738 AV |
884 | } |
885 | ||
886 | static int fifo_open(struct inode *inode, struct file *filp) | |
887 | { | |
888 | struct pipe_inode_info *pipe; | |
599a0ac1 | 889 | bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC; |
f776c738 AV |
890 | int ret; |
891 | ||
ba5bb147 AV |
892 | filp->f_version = 0; |
893 | ||
894 | spin_lock(&inode->i_lock); | |
895 | if (inode->i_pipe) { | |
896 | pipe = inode->i_pipe; | |
897 | pipe->files++; | |
898 | spin_unlock(&inode->i_lock); | |
899 | } else { | |
900 | spin_unlock(&inode->i_lock); | |
7bee130e | 901 | pipe = alloc_pipe_info(); |
f776c738 | 902 | if (!pipe) |
ba5bb147 AV |
903 | return -ENOMEM; |
904 | pipe->files = 1; | |
905 | spin_lock(&inode->i_lock); | |
906 | if (unlikely(inode->i_pipe)) { | |
907 | inode->i_pipe->files++; | |
908 | spin_unlock(&inode->i_lock); | |
4b8a8f1e | 909 | free_pipe_info(pipe); |
ba5bb147 AV |
910 | pipe = inode->i_pipe; |
911 | } else { | |
912 | inode->i_pipe = pipe; | |
913 | spin_unlock(&inode->i_lock); | |
914 | } | |
f776c738 | 915 | } |
de32ec4c | 916 | filp->private_data = pipe; |
ba5bb147 AV |
917 | /* OK, we have a pipe and it's pinned down */ |
918 | ||
ebec73f4 | 919 | __pipe_lock(pipe); |
f776c738 AV |
920 | |
921 | /* We can only do regular read/write on fifos */ | |
922 | filp->f_mode &= (FMODE_READ | FMODE_WRITE); | |
923 | ||
924 | switch (filp->f_mode) { | |
925 | case FMODE_READ: | |
926 | /* | |
927 | * O_RDONLY | |
928 | * POSIX.1 says that O_NONBLOCK means return with the FIFO | |
929 | * opened, even when there is no process writing the FIFO. | |
930 | */ | |
f776c738 AV |
931 | pipe->r_counter++; |
932 | if (pipe->readers++ == 0) | |
fc7478a2 | 933 | wake_up_partner(pipe); |
f776c738 | 934 | |
599a0ac1 | 935 | if (!is_pipe && !pipe->writers) { |
f776c738 AV |
936 | if ((filp->f_flags & O_NONBLOCK)) { |
937 | /* suppress POLLHUP until we have | |
938 | * seen a writer */ | |
939 | filp->f_version = pipe->w_counter; | |
940 | } else { | |
fc7478a2 | 941 | if (wait_for_partner(pipe, &pipe->w_counter)) |
f776c738 AV |
942 | goto err_rd; |
943 | } | |
944 | } | |
945 | break; | |
946 | ||
947 | case FMODE_WRITE: | |
948 | /* | |
949 | * O_WRONLY | |
950 | * POSIX.1 says that O_NONBLOCK means return -1 with | |
951 | * errno=ENXIO when there is no process reading the FIFO. | |
952 | */ | |
953 | ret = -ENXIO; | |
599a0ac1 | 954 | if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers) |
f776c738 AV |
955 | goto err; |
956 | ||
f776c738 AV |
957 | pipe->w_counter++; |
958 | if (!pipe->writers++) | |
fc7478a2 | 959 | wake_up_partner(pipe); |
f776c738 | 960 | |
599a0ac1 | 961 | if (!is_pipe && !pipe->readers) { |
fc7478a2 | 962 | if (wait_for_partner(pipe, &pipe->r_counter)) |
f776c738 AV |
963 | goto err_wr; |
964 | } | |
965 | break; | |
966 | ||
967 | case FMODE_READ | FMODE_WRITE: | |
968 | /* | |
969 | * O_RDWR | |
970 | * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set. | |
971 | * This implementation will NEVER block on a O_RDWR open, since | |
972 | * the process can at least talk to itself. | |
973 | */ | |
f776c738 AV |
974 | |
975 | pipe->readers++; | |
976 | pipe->writers++; | |
977 | pipe->r_counter++; | |
978 | pipe->w_counter++; | |
979 | if (pipe->readers == 1 || pipe->writers == 1) | |
fc7478a2 | 980 | wake_up_partner(pipe); |
f776c738 AV |
981 | break; |
982 | ||
983 | default: | |
984 | ret = -EINVAL; | |
985 | goto err; | |
986 | } | |
987 | ||
988 | /* Ok! */ | |
ebec73f4 | 989 | __pipe_unlock(pipe); |
f776c738 AV |
990 | return 0; |
991 | ||
992 | err_rd: | |
993 | if (!--pipe->readers) | |
994 | wake_up_interruptible(&pipe->wait); | |
995 | ret = -ERESTARTSYS; | |
996 | goto err; | |
997 | ||
998 | err_wr: | |
999 | if (!--pipe->writers) | |
1000 | wake_up_interruptible(&pipe->wait); | |
1001 | ret = -ERESTARTSYS; | |
1002 | goto err; | |
1003 | ||
1004 | err: | |
ebec73f4 | 1005 | __pipe_unlock(pipe); |
b0d8d229 LT |
1006 | |
1007 | put_pipe_info(inode, pipe); | |
f776c738 AV |
1008 | return ret; |
1009 | } | |
1010 | ||
599a0ac1 AV |
1011 | const struct file_operations pipefifo_fops = { |
1012 | .open = fifo_open, | |
1013 | .llseek = no_llseek, | |
fb9096a3 | 1014 | .read_iter = pipe_read, |
f0d1bec9 | 1015 | .write_iter = pipe_write, |
599a0ac1 AV |
1016 | .poll = pipe_poll, |
1017 | .unlocked_ioctl = pipe_ioctl, | |
1018 | .release = pipe_release, | |
1019 | .fasync = pipe_fasync, | |
f776c738 AV |
1020 | }; |
1021 | ||
3a7bb597 MK |
1022 | /* |
1023 | * Currently we rely on the pipe array holding a power-of-2 number | |
1024 | * of pages. | |
1025 | */ | |
1026 | static inline unsigned int round_pipe_size(unsigned int size) | |
1027 | { | |
1028 | unsigned long nr_pages; | |
1029 | ||
1030 | nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1031 | return roundup_pow_of_two(nr_pages) << PAGE_SHIFT; | |
1032 | } | |
1033 | ||
35f3d14d JA |
1034 | /* |
1035 | * Allocate a new array of pipe buffers and copy the info over. Returns the | |
1036 | * pipe size if successful, or return -ERROR on error. | |
1037 | */ | |
a2a77da5 | 1038 | static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg) |
35f3d14d JA |
1039 | { |
1040 | struct pipe_buffer *bufs; | |
a2a77da5 | 1041 | unsigned int size, nr_pages; |
6ab1c8f0 | 1042 | unsigned long user_bufs; |
3a6b3159 | 1043 | long ret = 0; |
a2a77da5 MK |
1044 | |
1045 | size = round_pipe_size(arg); | |
1046 | nr_pages = size >> PAGE_SHIFT; | |
1047 | ||
1048 | if (!nr_pages) | |
1049 | return -EINVAL; | |
1050 | ||
3a6b3159 MK |
1051 | /* |
1052 | * If trying to increase the pipe capacity, check that an | |
1053 | * unprivileged user is not trying to exceed various limits | |
1054 | * (soft limit check here, hard limit check just below). | |
1055 | * Decreasing the pipe capacity is always permitted, even | |
1056 | * if the user is currently over a limit. | |
1057 | */ | |
1058 | if (nr_pages > pipe->buffers && | |
1059 | size > pipe_max_size && !capable(CAP_SYS_RESOURCE)) | |
a2a77da5 MK |
1060 | return -EPERM; |
1061 | ||
6ab1c8f0 | 1062 | user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages); |
3a6b3159 MK |
1063 | |
1064 | if (nr_pages > pipe->buffers && | |
6ab1c8f0 MK |
1065 | (too_many_pipe_buffers_hard(user_bufs) || |
1066 | too_many_pipe_buffers_soft(user_bufs)) && | |
3a6b3159 MK |
1067 | !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) { |
1068 | ret = -EPERM; | |
1069 | goto out_revert_acct; | |
1070 | } | |
35f3d14d | 1071 | |
35f3d14d JA |
1072 | /* |
1073 | * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't | |
1074 | * expect a lot of shrink+grow operations, just free and allocate | |
1075 | * again like we would do for growing. If the pipe currently | |
1076 | * contains more buffers than arg, then return busy. | |
1077 | */ | |
3a6b3159 MK |
1078 | if (nr_pages < pipe->nrbufs) { |
1079 | ret = -EBUSY; | |
1080 | goto out_revert_acct; | |
1081 | } | |
35f3d14d | 1082 | |
d86133bd VD |
1083 | bufs = kcalloc(nr_pages, sizeof(*bufs), |
1084 | GFP_KERNEL_ACCOUNT | __GFP_NOWARN); | |
3a6b3159 MK |
1085 | if (unlikely(!bufs)) { |
1086 | ret = -ENOMEM; | |
1087 | goto out_revert_acct; | |
1088 | } | |
35f3d14d JA |
1089 | |
1090 | /* | |
1091 | * The pipe array wraps around, so just start the new one at zero | |
1092 | * and adjust the indexes. | |
1093 | */ | |
1094 | if (pipe->nrbufs) { | |
1d862f41 MS |
1095 | unsigned int tail; |
1096 | unsigned int head; | |
35f3d14d | 1097 | |
1d862f41 MS |
1098 | tail = pipe->curbuf + pipe->nrbufs; |
1099 | if (tail < pipe->buffers) | |
1100 | tail = 0; | |
1101 | else | |
1102 | tail &= (pipe->buffers - 1); | |
1103 | ||
1104 | head = pipe->nrbufs - tail; | |
35f3d14d JA |
1105 | if (head) |
1106 | memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer)); | |
1107 | if (tail) | |
1d862f41 | 1108 | memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer)); |
35f3d14d JA |
1109 | } |
1110 | ||
1111 | pipe->curbuf = 0; | |
1112 | kfree(pipe->bufs); | |
1113 | pipe->bufs = bufs; | |
b9598db3 JA |
1114 | pipe->buffers = nr_pages; |
1115 | return nr_pages * PAGE_SIZE; | |
3a6b3159 MK |
1116 | |
1117 | out_revert_acct: | |
6ab1c8f0 | 1118 | (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers); |
3a6b3159 | 1119 | return ret; |
35f3d14d JA |
1120 | } |
1121 | ||
ff9da691 JA |
1122 | /* |
1123 | * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax | |
1124 | * will return an error. | |
1125 | */ | |
1126 | int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf, | |
1127 | size_t *lenp, loff_t *ppos) | |
1128 | { | |
1129 | int ret; | |
1130 | ||
1131 | ret = proc_dointvec_minmax(table, write, buf, lenp, ppos); | |
1132 | if (ret < 0 || !write) | |
1133 | return ret; | |
1134 | ||
1135 | pipe_max_size = round_pipe_size(pipe_max_size); | |
1136 | return ret; | |
1137 | } | |
1138 | ||
72083646 LT |
1139 | /* |
1140 | * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same | |
1141 | * location, so checking ->i_pipe is not enough to verify that this is a | |
1142 | * pipe. | |
1143 | */ | |
1144 | struct pipe_inode_info *get_pipe_info(struct file *file) | |
1145 | { | |
de32ec4c | 1146 | return file->f_op == &pipefifo_fops ? file->private_data : NULL; |
72083646 LT |
1147 | } |
1148 | ||
35f3d14d JA |
1149 | long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg) |
1150 | { | |
1151 | struct pipe_inode_info *pipe; | |
1152 | long ret; | |
1153 | ||
c66fb347 | 1154 | pipe = get_pipe_info(file); |
35f3d14d JA |
1155 | if (!pipe) |
1156 | return -EBADF; | |
1157 | ||
ebec73f4 | 1158 | __pipe_lock(pipe); |
35f3d14d JA |
1159 | |
1160 | switch (cmd) { | |
a2a77da5 MK |
1161 | case F_SETPIPE_SZ: |
1162 | ret = pipe_set_size(pipe, arg); | |
35f3d14d JA |
1163 | break; |
1164 | case F_GETPIPE_SZ: | |
b9598db3 | 1165 | ret = pipe->buffers * PAGE_SIZE; |
35f3d14d JA |
1166 | break; |
1167 | default: | |
1168 | ret = -EINVAL; | |
1169 | break; | |
1170 | } | |
1171 | ||
ebec73f4 | 1172 | __pipe_unlock(pipe); |
35f3d14d JA |
1173 | return ret; |
1174 | } | |
1175 | ||
ff0c7d15 NP |
1176 | static const struct super_operations pipefs_ops = { |
1177 | .destroy_inode = free_inode_nonrcu, | |
d70ef97b | 1178 | .statfs = simple_statfs, |
ff0c7d15 NP |
1179 | }; |
1180 | ||
1da177e4 LT |
1181 | /* |
1182 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1183 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1184 | * any operations on the root directory. However, we need a non-trivial | |
1185 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1186 | */ | |
51139ada AV |
1187 | static struct dentry *pipefs_mount(struct file_system_type *fs_type, |
1188 | int flags, const char *dev_name, void *data) | |
1da177e4 | 1189 | { |
c74a1cbb AV |
1190 | return mount_pseudo(fs_type, "pipe:", &pipefs_ops, |
1191 | &pipefs_dentry_operations, PIPEFS_MAGIC); | |
1da177e4 LT |
1192 | } |
1193 | ||
1194 | static struct file_system_type pipe_fs_type = { | |
1195 | .name = "pipefs", | |
51139ada | 1196 | .mount = pipefs_mount, |
1da177e4 LT |
1197 | .kill_sb = kill_anon_super, |
1198 | }; | |
1199 | ||
1200 | static int __init init_pipe_fs(void) | |
1201 | { | |
1202 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1203 | |
1da177e4 LT |
1204 | if (!err) { |
1205 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1206 | if (IS_ERR(pipe_mnt)) { | |
1207 | err = PTR_ERR(pipe_mnt); | |
1208 | unregister_filesystem(&pipe_fs_type); | |
1209 | } | |
1210 | } | |
1211 | return err; | |
1212 | } | |
1213 | ||
1da177e4 | 1214 | fs_initcall(init_pipe_fs); |