]>
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> | |
14 | #include <linux/mount.h> | |
15 | #include <linux/pipe_fs_i.h> | |
16 | #include <linux/uio.h> | |
17 | #include <linux/highmem.h> | |
5274f052 | 18 | #include <linux/pagemap.h> |
db349509 | 19 | #include <linux/audit.h> |
ba719bae | 20 | #include <linux/syscalls.h> |
1da177e4 LT |
21 | |
22 | #include <asm/uaccess.h> | |
23 | #include <asm/ioctls.h> | |
24 | ||
25 | /* | |
26 | * We use a start+len construction, which provides full use of the | |
27 | * allocated memory. | |
28 | * -- Florian Coosmann (FGC) | |
29 | * | |
30 | * Reads with count = 0 should always return 0. | |
31 | * -- Julian Bradfield 1999-06-07. | |
32 | * | |
33 | * FIFOs and Pipes now generate SIGIO for both readers and writers. | |
34 | * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16 | |
35 | * | |
36 | * pipe_read & write cleanup | |
37 | * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09 | |
38 | */ | |
39 | ||
61e0d47c MS |
40 | static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass) |
41 | { | |
42 | if (pipe->inode) | |
43 | mutex_lock_nested(&pipe->inode->i_mutex, subclass); | |
44 | } | |
45 | ||
46 | void pipe_lock(struct pipe_inode_info *pipe) | |
47 | { | |
48 | /* | |
49 | * pipe_lock() nests non-pipe inode locks (for writing to a file) | |
50 | */ | |
51 | pipe_lock_nested(pipe, I_MUTEX_PARENT); | |
52 | } | |
53 | EXPORT_SYMBOL(pipe_lock); | |
54 | ||
55 | void pipe_unlock(struct pipe_inode_info *pipe) | |
56 | { | |
57 | if (pipe->inode) | |
58 | mutex_unlock(&pipe->inode->i_mutex); | |
59 | } | |
60 | EXPORT_SYMBOL(pipe_unlock); | |
61 | ||
62 | void pipe_double_lock(struct pipe_inode_info *pipe1, | |
63 | struct pipe_inode_info *pipe2) | |
64 | { | |
65 | BUG_ON(pipe1 == pipe2); | |
66 | ||
67 | if (pipe1 < pipe2) { | |
68 | pipe_lock_nested(pipe1, I_MUTEX_PARENT); | |
69 | pipe_lock_nested(pipe2, I_MUTEX_CHILD); | |
70 | } else { | |
023d43c7 PZ |
71 | pipe_lock_nested(pipe2, I_MUTEX_PARENT); |
72 | pipe_lock_nested(pipe1, I_MUTEX_CHILD); | |
61e0d47c MS |
73 | } |
74 | } | |
75 | ||
1da177e4 | 76 | /* Drop the inode semaphore and wait for a pipe event, atomically */ |
3a326a2c | 77 | void pipe_wait(struct pipe_inode_info *pipe) |
1da177e4 LT |
78 | { |
79 | DEFINE_WAIT(wait); | |
80 | ||
d79fc0fc IM |
81 | /* |
82 | * Pipes are system-local resources, so sleeping on them | |
83 | * is considered a noninteractive wait: | |
84 | */ | |
af927232 | 85 | prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE); |
61e0d47c | 86 | pipe_unlock(pipe); |
1da177e4 | 87 | schedule(); |
3a326a2c | 88 | finish_wait(&pipe->wait, &wait); |
61e0d47c | 89 | pipe_lock(pipe); |
1da177e4 LT |
90 | } |
91 | ||
858119e1 | 92 | static int |
f6762b7a JA |
93 | pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len, |
94 | int atomic) | |
1da177e4 LT |
95 | { |
96 | unsigned long copy; | |
97 | ||
98 | while (len > 0) { | |
99 | while (!iov->iov_len) | |
100 | iov++; | |
101 | copy = min_t(unsigned long, len, iov->iov_len); | |
102 | ||
f6762b7a JA |
103 | if (atomic) { |
104 | if (__copy_from_user_inatomic(to, iov->iov_base, copy)) | |
105 | return -EFAULT; | |
106 | } else { | |
107 | if (copy_from_user(to, iov->iov_base, copy)) | |
108 | return -EFAULT; | |
109 | } | |
1da177e4 LT |
110 | to += copy; |
111 | len -= copy; | |
112 | iov->iov_base += copy; | |
113 | iov->iov_len -= copy; | |
114 | } | |
115 | return 0; | |
116 | } | |
117 | ||
858119e1 | 118 | static int |
f6762b7a JA |
119 | pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len, |
120 | int atomic) | |
1da177e4 LT |
121 | { |
122 | unsigned long copy; | |
123 | ||
124 | while (len > 0) { | |
125 | while (!iov->iov_len) | |
126 | iov++; | |
127 | copy = min_t(unsigned long, len, iov->iov_len); | |
128 | ||
f6762b7a JA |
129 | if (atomic) { |
130 | if (__copy_to_user_inatomic(iov->iov_base, from, copy)) | |
131 | return -EFAULT; | |
132 | } else { | |
133 | if (copy_to_user(iov->iov_base, from, copy)) | |
134 | return -EFAULT; | |
135 | } | |
1da177e4 LT |
136 | from += copy; |
137 | len -= copy; | |
138 | iov->iov_base += copy; | |
139 | iov->iov_len -= copy; | |
140 | } | |
141 | return 0; | |
142 | } | |
143 | ||
f6762b7a JA |
144 | /* |
145 | * Attempt to pre-fault in the user memory, so we can use atomic copies. | |
146 | * Returns the number of bytes not faulted in. | |
147 | */ | |
148 | static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len) | |
149 | { | |
150 | while (!iov->iov_len) | |
151 | iov++; | |
152 | ||
153 | while (len > 0) { | |
154 | unsigned long this_len; | |
155 | ||
156 | this_len = min_t(unsigned long, len, iov->iov_len); | |
157 | if (fault_in_pages_writeable(iov->iov_base, this_len)) | |
158 | break; | |
159 | ||
160 | len -= this_len; | |
161 | iov++; | |
162 | } | |
163 | ||
164 | return len; | |
165 | } | |
166 | ||
167 | /* | |
168 | * Pre-fault in the user memory, so we can use atomic copies. | |
169 | */ | |
170 | static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len) | |
171 | { | |
172 | while (!iov->iov_len) | |
173 | iov++; | |
174 | ||
175 | while (len > 0) { | |
176 | unsigned long this_len; | |
177 | ||
178 | this_len = min_t(unsigned long, len, iov->iov_len); | |
179 | fault_in_pages_readable(iov->iov_base, this_len); | |
180 | len -= this_len; | |
181 | iov++; | |
182 | } | |
183 | } | |
184 | ||
341b446b IM |
185 | static void anon_pipe_buf_release(struct pipe_inode_info *pipe, |
186 | struct pipe_buffer *buf) | |
1da177e4 LT |
187 | { |
188 | struct page *page = buf->page; | |
189 | ||
5274f052 JA |
190 | /* |
191 | * If nobody else uses this page, and we don't already have a | |
192 | * temporary page, let's keep track of it as a one-deep | |
341b446b | 193 | * allocation cache. (Otherwise just release our reference to it) |
5274f052 | 194 | */ |
341b446b | 195 | if (page_count(page) == 1 && !pipe->tmp_page) |
923f4f23 | 196 | pipe->tmp_page = page; |
341b446b IM |
197 | else |
198 | page_cache_release(page); | |
1da177e4 LT |
199 | } |
200 | ||
0845718d JA |
201 | /** |
202 | * generic_pipe_buf_map - virtually map a pipe buffer | |
203 | * @pipe: the pipe that the buffer belongs to | |
204 | * @buf: the buffer that should be mapped | |
205 | * @atomic: whether to use an atomic map | |
206 | * | |
207 | * Description: | |
208 | * This function returns a kernel virtual address mapping for the | |
b51d63c6 | 209 | * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided |
0845718d JA |
210 | * and the caller has to be careful not to fault before calling |
211 | * the unmap function. | |
212 | * | |
213 | * Note that this function occupies KM_USER0 if @atomic != 0. | |
214 | */ | |
f84d7519 | 215 | void *generic_pipe_buf_map(struct pipe_inode_info *pipe, |
f6762b7a | 216 | struct pipe_buffer *buf, int atomic) |
1da177e4 | 217 | { |
f6762b7a JA |
218 | if (atomic) { |
219 | buf->flags |= PIPE_BUF_FLAG_ATOMIC; | |
220 | return kmap_atomic(buf->page, KM_USER0); | |
221 | } | |
222 | ||
1da177e4 LT |
223 | return kmap(buf->page); |
224 | } | |
225 | ||
0845718d JA |
226 | /** |
227 | * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer | |
228 | * @pipe: the pipe that the buffer belongs to | |
229 | * @buf: the buffer that should be unmapped | |
230 | * @map_data: the data that the mapping function returned | |
231 | * | |
232 | * Description: | |
233 | * This function undoes the mapping that ->map() provided. | |
234 | */ | |
f84d7519 | 235 | void generic_pipe_buf_unmap(struct pipe_inode_info *pipe, |
f6762b7a | 236 | struct pipe_buffer *buf, void *map_data) |
1da177e4 | 237 | { |
f6762b7a JA |
238 | if (buf->flags & PIPE_BUF_FLAG_ATOMIC) { |
239 | buf->flags &= ~PIPE_BUF_FLAG_ATOMIC; | |
240 | kunmap_atomic(map_data, KM_USER0); | |
241 | } else | |
242 | kunmap(buf->page); | |
1da177e4 LT |
243 | } |
244 | ||
0845718d | 245 | /** |
b51d63c6 | 246 | * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer |
0845718d JA |
247 | * @pipe: the pipe that the buffer belongs to |
248 | * @buf: the buffer to attempt to steal | |
249 | * | |
250 | * Description: | |
b51d63c6 | 251 | * This function attempts to steal the &struct page attached to |
0845718d JA |
252 | * @buf. If successful, this function returns 0 and returns with |
253 | * the page locked. The caller may then reuse the page for whatever | |
b51d63c6 | 254 | * he wishes; the typical use is insertion into a different file |
0845718d JA |
255 | * page cache. |
256 | */ | |
330ab716 JA |
257 | int generic_pipe_buf_steal(struct pipe_inode_info *pipe, |
258 | struct pipe_buffer *buf) | |
5abc97aa | 259 | { |
46e678c9 JA |
260 | struct page *page = buf->page; |
261 | ||
0845718d JA |
262 | /* |
263 | * A reference of one is golden, that means that the owner of this | |
264 | * page is the only one holding a reference to it. lock the page | |
265 | * and return OK. | |
266 | */ | |
46e678c9 | 267 | if (page_count(page) == 1) { |
46e678c9 JA |
268 | lock_page(page); |
269 | return 0; | |
270 | } | |
271 | ||
272 | return 1; | |
5abc97aa JA |
273 | } |
274 | ||
0845718d | 275 | /** |
b51d63c6 | 276 | * generic_pipe_buf_get - get a reference to a &struct pipe_buffer |
0845718d JA |
277 | * @pipe: the pipe that the buffer belongs to |
278 | * @buf: the buffer to get a reference to | |
279 | * | |
280 | * Description: | |
281 | * This function grabs an extra reference to @buf. It's used in | |
282 | * in the tee() system call, when we duplicate the buffers in one | |
283 | * pipe into another. | |
284 | */ | |
285 | void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf) | |
70524490 JA |
286 | { |
287 | page_cache_get(buf->page); | |
288 | } | |
289 | ||
0845718d JA |
290 | /** |
291 | * generic_pipe_buf_confirm - verify contents of the pipe buffer | |
79685b8d | 292 | * @info: the pipe that the buffer belongs to |
0845718d JA |
293 | * @buf: the buffer to confirm |
294 | * | |
295 | * Description: | |
296 | * This function does nothing, because the generic pipe code uses | |
297 | * pages that are always good when inserted into the pipe. | |
298 | */ | |
cac36bb0 JA |
299 | int generic_pipe_buf_confirm(struct pipe_inode_info *info, |
300 | struct pipe_buffer *buf) | |
f84d7519 JA |
301 | { |
302 | return 0; | |
303 | } | |
304 | ||
6818173b MS |
305 | /** |
306 | * generic_pipe_buf_release - put a reference to a &struct pipe_buffer | |
307 | * @pipe: the pipe that the buffer belongs to | |
308 | * @buf: the buffer to put a reference to | |
309 | * | |
310 | * Description: | |
311 | * This function releases a reference to @buf. | |
312 | */ | |
313 | void generic_pipe_buf_release(struct pipe_inode_info *pipe, | |
314 | struct pipe_buffer *buf) | |
315 | { | |
316 | page_cache_release(buf->page); | |
317 | } | |
318 | ||
d4c3cca9 | 319 | static const struct pipe_buf_operations anon_pipe_buf_ops = { |
1da177e4 | 320 | .can_merge = 1, |
f84d7519 JA |
321 | .map = generic_pipe_buf_map, |
322 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 323 | .confirm = generic_pipe_buf_confirm, |
1da177e4 | 324 | .release = anon_pipe_buf_release, |
330ab716 | 325 | .steal = generic_pipe_buf_steal, |
f84d7519 | 326 | .get = generic_pipe_buf_get, |
1da177e4 LT |
327 | }; |
328 | ||
329 | static ssize_t | |
ee0b3e67 BP |
330 | pipe_read(struct kiocb *iocb, const struct iovec *_iov, |
331 | unsigned long nr_segs, loff_t pos) | |
1da177e4 | 332 | { |
ee0b3e67 | 333 | struct file *filp = iocb->ki_filp; |
0f7fc9e4 | 334 | struct inode *inode = filp->f_path.dentry->d_inode; |
923f4f23 | 335 | struct pipe_inode_info *pipe; |
1da177e4 LT |
336 | int do_wakeup; |
337 | ssize_t ret; | |
338 | struct iovec *iov = (struct iovec *)_iov; | |
339 | size_t total_len; | |
340 | ||
341 | total_len = iov_length(iov, nr_segs); | |
342 | /* Null read succeeds. */ | |
343 | if (unlikely(total_len == 0)) | |
344 | return 0; | |
345 | ||
346 | do_wakeup = 0; | |
347 | ret = 0; | |
9aeedfc4 | 348 | mutex_lock(&inode->i_mutex); |
923f4f23 | 349 | pipe = inode->i_pipe; |
1da177e4 | 350 | for (;;) { |
923f4f23 | 351 | int bufs = pipe->nrbufs; |
1da177e4 | 352 | if (bufs) { |
923f4f23 IM |
353 | int curbuf = pipe->curbuf; |
354 | struct pipe_buffer *buf = pipe->bufs + curbuf; | |
d4c3cca9 | 355 | const struct pipe_buf_operations *ops = buf->ops; |
1da177e4 LT |
356 | void *addr; |
357 | size_t chars = buf->len; | |
f6762b7a | 358 | int error, atomic; |
1da177e4 LT |
359 | |
360 | if (chars > total_len) | |
361 | chars = total_len; | |
362 | ||
cac36bb0 | 363 | error = ops->confirm(pipe, buf); |
f84d7519 | 364 | if (error) { |
5274f052 | 365 | if (!ret) |
f84d7519 | 366 | error = ret; |
5274f052 JA |
367 | break; |
368 | } | |
f84d7519 | 369 | |
f6762b7a JA |
370 | atomic = !iov_fault_in_pages_write(iov, chars); |
371 | redo: | |
372 | addr = ops->map(pipe, buf, atomic); | |
373 | error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic); | |
374 | ops->unmap(pipe, buf, addr); | |
1da177e4 | 375 | if (unlikely(error)) { |
f6762b7a JA |
376 | /* |
377 | * Just retry with the slow path if we failed. | |
378 | */ | |
379 | if (atomic) { | |
380 | atomic = 0; | |
381 | goto redo; | |
382 | } | |
341b446b | 383 | if (!ret) |
f6762b7a | 384 | ret = error; |
1da177e4 LT |
385 | break; |
386 | } | |
387 | ret += chars; | |
388 | buf->offset += chars; | |
389 | buf->len -= chars; | |
390 | if (!buf->len) { | |
391 | buf->ops = NULL; | |
923f4f23 | 392 | ops->release(pipe, buf); |
1da177e4 | 393 | curbuf = (curbuf + 1) & (PIPE_BUFFERS-1); |
923f4f23 IM |
394 | pipe->curbuf = curbuf; |
395 | pipe->nrbufs = --bufs; | |
1da177e4 LT |
396 | do_wakeup = 1; |
397 | } | |
398 | total_len -= chars; | |
399 | if (!total_len) | |
400 | break; /* common path: read succeeded */ | |
401 | } | |
402 | if (bufs) /* More to do? */ | |
403 | continue; | |
923f4f23 | 404 | if (!pipe->writers) |
1da177e4 | 405 | break; |
923f4f23 | 406 | if (!pipe->waiting_writers) { |
1da177e4 LT |
407 | /* syscall merging: Usually we must not sleep |
408 | * if O_NONBLOCK is set, or if we got some data. | |
409 | * But if a writer sleeps in kernel space, then | |
410 | * we can wait for that data without violating POSIX. | |
411 | */ | |
412 | if (ret) | |
413 | break; | |
414 | if (filp->f_flags & O_NONBLOCK) { | |
415 | ret = -EAGAIN; | |
416 | break; | |
417 | } | |
418 | } | |
419 | if (signal_pending(current)) { | |
341b446b IM |
420 | if (!ret) |
421 | ret = -ERESTARTSYS; | |
1da177e4 LT |
422 | break; |
423 | } | |
424 | if (do_wakeup) { | |
923f4f23 IM |
425 | wake_up_interruptible_sync(&pipe->wait); |
426 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 427 | } |
923f4f23 | 428 | pipe_wait(pipe); |
1da177e4 | 429 | } |
9aeedfc4 | 430 | mutex_unlock(&inode->i_mutex); |
341b446b IM |
431 | |
432 | /* Signal writers asynchronously that there is more room. */ | |
1da177e4 | 433 | if (do_wakeup) { |
71e20f18 | 434 | wake_up_interruptible_sync(&pipe->wait); |
923f4f23 | 435 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); |
1da177e4 LT |
436 | } |
437 | if (ret > 0) | |
438 | file_accessed(filp); | |
439 | return ret; | |
440 | } | |
441 | ||
442 | static ssize_t | |
ee0b3e67 BP |
443 | pipe_write(struct kiocb *iocb, const struct iovec *_iov, |
444 | unsigned long nr_segs, loff_t ppos) | |
1da177e4 | 445 | { |
ee0b3e67 | 446 | struct file *filp = iocb->ki_filp; |
0f7fc9e4 | 447 | struct inode *inode = filp->f_path.dentry->d_inode; |
923f4f23 | 448 | struct pipe_inode_info *pipe; |
1da177e4 LT |
449 | ssize_t ret; |
450 | int do_wakeup; | |
451 | struct iovec *iov = (struct iovec *)_iov; | |
452 | size_t total_len; | |
453 | ssize_t chars; | |
454 | ||
455 | total_len = iov_length(iov, nr_segs); | |
456 | /* Null write succeeds. */ | |
457 | if (unlikely(total_len == 0)) | |
458 | return 0; | |
459 | ||
460 | do_wakeup = 0; | |
461 | ret = 0; | |
9aeedfc4 | 462 | mutex_lock(&inode->i_mutex); |
923f4f23 | 463 | pipe = inode->i_pipe; |
1da177e4 | 464 | |
923f4f23 | 465 | if (!pipe->readers) { |
1da177e4 LT |
466 | send_sig(SIGPIPE, current, 0); |
467 | ret = -EPIPE; | |
468 | goto out; | |
469 | } | |
470 | ||
471 | /* We try to merge small writes */ | |
472 | chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */ | |
923f4f23 | 473 | if (pipe->nrbufs && chars != 0) { |
341b446b IM |
474 | int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) & |
475 | (PIPE_BUFFERS-1); | |
923f4f23 | 476 | struct pipe_buffer *buf = pipe->bufs + lastbuf; |
d4c3cca9 | 477 | const struct pipe_buf_operations *ops = buf->ops; |
1da177e4 | 478 | int offset = buf->offset + buf->len; |
341b446b | 479 | |
1da177e4 | 480 | if (ops->can_merge && offset + chars <= PAGE_SIZE) { |
f6762b7a | 481 | int error, atomic = 1; |
5274f052 | 482 | void *addr; |
5274f052 | 483 | |
cac36bb0 | 484 | error = ops->confirm(pipe, buf); |
f84d7519 | 485 | if (error) |
5274f052 | 486 | goto out; |
f84d7519 | 487 | |
f6762b7a JA |
488 | iov_fault_in_pages_read(iov, chars); |
489 | redo1: | |
490 | addr = ops->map(pipe, buf, atomic); | |
5274f052 | 491 | error = pipe_iov_copy_from_user(offset + addr, iov, |
f6762b7a JA |
492 | chars, atomic); |
493 | ops->unmap(pipe, buf, addr); | |
1da177e4 LT |
494 | ret = error; |
495 | do_wakeup = 1; | |
f6762b7a JA |
496 | if (error) { |
497 | if (atomic) { | |
498 | atomic = 0; | |
499 | goto redo1; | |
500 | } | |
1da177e4 | 501 | goto out; |
f6762b7a | 502 | } |
1da177e4 LT |
503 | buf->len += chars; |
504 | total_len -= chars; | |
505 | ret = chars; | |
506 | if (!total_len) | |
507 | goto out; | |
508 | } | |
509 | } | |
510 | ||
511 | for (;;) { | |
512 | int bufs; | |
341b446b | 513 | |
923f4f23 | 514 | if (!pipe->readers) { |
1da177e4 | 515 | send_sig(SIGPIPE, current, 0); |
341b446b IM |
516 | if (!ret) |
517 | ret = -EPIPE; | |
1da177e4 LT |
518 | break; |
519 | } | |
923f4f23 | 520 | bufs = pipe->nrbufs; |
1da177e4 | 521 | if (bufs < PIPE_BUFFERS) { |
923f4f23 IM |
522 | int newbuf = (pipe->curbuf + bufs) & (PIPE_BUFFERS-1); |
523 | struct pipe_buffer *buf = pipe->bufs + newbuf; | |
524 | struct page *page = pipe->tmp_page; | |
f6762b7a JA |
525 | char *src; |
526 | int error, atomic = 1; | |
1da177e4 LT |
527 | |
528 | if (!page) { | |
529 | page = alloc_page(GFP_HIGHUSER); | |
530 | if (unlikely(!page)) { | |
531 | ret = ret ? : -ENOMEM; | |
532 | break; | |
533 | } | |
923f4f23 | 534 | pipe->tmp_page = page; |
1da177e4 | 535 | } |
341b446b | 536 | /* Always wake up, even if the copy fails. Otherwise |
1da177e4 LT |
537 | * we lock up (O_NONBLOCK-)readers that sleep due to |
538 | * syscall merging. | |
539 | * FIXME! Is this really true? | |
540 | */ | |
541 | do_wakeup = 1; | |
542 | chars = PAGE_SIZE; | |
543 | if (chars > total_len) | |
544 | chars = total_len; | |
545 | ||
f6762b7a JA |
546 | iov_fault_in_pages_read(iov, chars); |
547 | redo2: | |
548 | if (atomic) | |
549 | src = kmap_atomic(page, KM_USER0); | |
550 | else | |
551 | src = kmap(page); | |
552 | ||
553 | error = pipe_iov_copy_from_user(src, iov, chars, | |
554 | atomic); | |
555 | if (atomic) | |
556 | kunmap_atomic(src, KM_USER0); | |
557 | else | |
558 | kunmap(page); | |
559 | ||
1da177e4 | 560 | if (unlikely(error)) { |
f6762b7a JA |
561 | if (atomic) { |
562 | atomic = 0; | |
563 | goto redo2; | |
564 | } | |
341b446b | 565 | if (!ret) |
f6762b7a | 566 | ret = error; |
1da177e4 LT |
567 | break; |
568 | } | |
569 | ret += chars; | |
570 | ||
571 | /* Insert it into the buffer array */ | |
572 | buf->page = page; | |
573 | buf->ops = &anon_pipe_buf_ops; | |
574 | buf->offset = 0; | |
575 | buf->len = chars; | |
923f4f23 IM |
576 | pipe->nrbufs = ++bufs; |
577 | pipe->tmp_page = NULL; | |
1da177e4 LT |
578 | |
579 | total_len -= chars; | |
580 | if (!total_len) | |
581 | break; | |
582 | } | |
583 | if (bufs < PIPE_BUFFERS) | |
584 | continue; | |
585 | if (filp->f_flags & O_NONBLOCK) { | |
341b446b IM |
586 | if (!ret) |
587 | ret = -EAGAIN; | |
1da177e4 LT |
588 | break; |
589 | } | |
590 | if (signal_pending(current)) { | |
341b446b IM |
591 | if (!ret) |
592 | ret = -ERESTARTSYS; | |
1da177e4 LT |
593 | break; |
594 | } | |
595 | if (do_wakeup) { | |
923f4f23 IM |
596 | wake_up_interruptible_sync(&pipe->wait); |
597 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); | |
1da177e4 LT |
598 | do_wakeup = 0; |
599 | } | |
923f4f23 IM |
600 | pipe->waiting_writers++; |
601 | pipe_wait(pipe); | |
602 | pipe->waiting_writers--; | |
1da177e4 LT |
603 | } |
604 | out: | |
9aeedfc4 | 605 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 606 | if (do_wakeup) { |
71e20f18 | 607 | wake_up_interruptible_sync(&pipe->wait); |
923f4f23 | 608 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
1da177e4 LT |
609 | } |
610 | if (ret > 0) | |
870f4817 | 611 | file_update_time(filp); |
1da177e4 LT |
612 | return ret; |
613 | } | |
614 | ||
1da177e4 LT |
615 | static ssize_t |
616 | bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos) | |
617 | { | |
618 | return -EBADF; | |
619 | } | |
620 | ||
621 | static ssize_t | |
341b446b IM |
622 | bad_pipe_w(struct file *filp, const char __user *buf, size_t count, |
623 | loff_t *ppos) | |
1da177e4 LT |
624 | { |
625 | return -EBADF; | |
626 | } | |
627 | ||
d59d0b1b | 628 | static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) |
1da177e4 | 629 | { |
0f7fc9e4 | 630 | struct inode *inode = filp->f_path.dentry->d_inode; |
923f4f23 | 631 | struct pipe_inode_info *pipe; |
1da177e4 LT |
632 | int count, buf, nrbufs; |
633 | ||
634 | switch (cmd) { | |
635 | case FIONREAD: | |
9aeedfc4 | 636 | mutex_lock(&inode->i_mutex); |
923f4f23 | 637 | pipe = inode->i_pipe; |
1da177e4 | 638 | count = 0; |
923f4f23 IM |
639 | buf = pipe->curbuf; |
640 | nrbufs = pipe->nrbufs; | |
1da177e4 | 641 | while (--nrbufs >= 0) { |
923f4f23 | 642 | count += pipe->bufs[buf].len; |
1da177e4 LT |
643 | buf = (buf+1) & (PIPE_BUFFERS-1); |
644 | } | |
9aeedfc4 | 645 | mutex_unlock(&inode->i_mutex); |
923f4f23 | 646 | |
1da177e4 LT |
647 | return put_user(count, (int __user *)arg); |
648 | default: | |
649 | return -EINVAL; | |
650 | } | |
651 | } | |
652 | ||
653 | /* No kernel lock held - fine */ | |
654 | static unsigned int | |
655 | pipe_poll(struct file *filp, poll_table *wait) | |
656 | { | |
657 | unsigned int mask; | |
0f7fc9e4 | 658 | struct inode *inode = filp->f_path.dentry->d_inode; |
923f4f23 | 659 | struct pipe_inode_info *pipe = inode->i_pipe; |
1da177e4 LT |
660 | int nrbufs; |
661 | ||
923f4f23 | 662 | poll_wait(filp, &pipe->wait, wait); |
1da177e4 LT |
663 | |
664 | /* Reading only -- no need for acquiring the semaphore. */ | |
923f4f23 | 665 | nrbufs = pipe->nrbufs; |
1da177e4 LT |
666 | mask = 0; |
667 | if (filp->f_mode & FMODE_READ) { | |
668 | mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0; | |
923f4f23 | 669 | if (!pipe->writers && filp->f_version != pipe->w_counter) |
1da177e4 LT |
670 | mask |= POLLHUP; |
671 | } | |
672 | ||
673 | if (filp->f_mode & FMODE_WRITE) { | |
674 | mask |= (nrbufs < PIPE_BUFFERS) ? POLLOUT | POLLWRNORM : 0; | |
5e5d7a22 PE |
675 | /* |
676 | * Most Unices do not set POLLERR for FIFOs but on Linux they | |
677 | * behave exactly like pipes for poll(). | |
678 | */ | |
923f4f23 | 679 | if (!pipe->readers) |
1da177e4 LT |
680 | mask |= POLLERR; |
681 | } | |
682 | ||
683 | return mask; | |
684 | } | |
685 | ||
1da177e4 LT |
686 | static int |
687 | pipe_release(struct inode *inode, int decr, int decw) | |
688 | { | |
923f4f23 IM |
689 | struct pipe_inode_info *pipe; |
690 | ||
9aeedfc4 | 691 | mutex_lock(&inode->i_mutex); |
923f4f23 IM |
692 | pipe = inode->i_pipe; |
693 | pipe->readers -= decr; | |
694 | pipe->writers -= decw; | |
341b446b | 695 | |
923f4f23 | 696 | if (!pipe->readers && !pipe->writers) { |
1da177e4 LT |
697 | free_pipe_info(inode); |
698 | } else { | |
71e20f18 | 699 | wake_up_interruptible_sync(&pipe->wait); |
923f4f23 IM |
700 | kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN); |
701 | kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT); | |
1da177e4 | 702 | } |
9aeedfc4 | 703 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
704 | |
705 | return 0; | |
706 | } | |
707 | ||
708 | static int | |
709 | pipe_read_fasync(int fd, struct file *filp, int on) | |
710 | { | |
0f7fc9e4 | 711 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
712 | int retval; |
713 | ||
9aeedfc4 IM |
714 | mutex_lock(&inode->i_mutex); |
715 | retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers); | |
716 | mutex_unlock(&inode->i_mutex); | |
1da177e4 | 717 | |
60aa4924 | 718 | return retval; |
1da177e4 LT |
719 | } |
720 | ||
721 | ||
722 | static int | |
723 | pipe_write_fasync(int fd, struct file *filp, int on) | |
724 | { | |
0f7fc9e4 | 725 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
726 | int retval; |
727 | ||
9aeedfc4 IM |
728 | mutex_lock(&inode->i_mutex); |
729 | retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers); | |
730 | mutex_unlock(&inode->i_mutex); | |
1da177e4 | 731 | |
60aa4924 | 732 | return retval; |
1da177e4 LT |
733 | } |
734 | ||
735 | ||
736 | static int | |
737 | pipe_rdwr_fasync(int fd, struct file *filp, int on) | |
738 | { | |
0f7fc9e4 | 739 | struct inode *inode = filp->f_path.dentry->d_inode; |
341b446b | 740 | struct pipe_inode_info *pipe = inode->i_pipe; |
1da177e4 LT |
741 | int retval; |
742 | ||
9aeedfc4 | 743 | mutex_lock(&inode->i_mutex); |
341b446b | 744 | retval = fasync_helper(fd, filp, on, &pipe->fasync_readers); |
e5bc49ba | 745 | if (retval >= 0) { |
341b446b | 746 | retval = fasync_helper(fd, filp, on, &pipe->fasync_writers); |
e5bc49ba ON |
747 | if (retval < 0) /* this can happen only if on == T */ |
748 | fasync_helper(-1, filp, 0, &pipe->fasync_readers); | |
749 | } | |
9aeedfc4 | 750 | mutex_unlock(&inode->i_mutex); |
60aa4924 | 751 | return retval; |
1da177e4 LT |
752 | } |
753 | ||
754 | ||
755 | static int | |
756 | pipe_read_release(struct inode *inode, struct file *filp) | |
757 | { | |
1da177e4 LT |
758 | return pipe_release(inode, 1, 0); |
759 | } | |
760 | ||
761 | static int | |
762 | pipe_write_release(struct inode *inode, struct file *filp) | |
763 | { | |
1da177e4 LT |
764 | return pipe_release(inode, 0, 1); |
765 | } | |
766 | ||
767 | static int | |
768 | pipe_rdwr_release(struct inode *inode, struct file *filp) | |
769 | { | |
770 | int decr, decw; | |
771 | ||
1da177e4 LT |
772 | decr = (filp->f_mode & FMODE_READ) != 0; |
773 | decw = (filp->f_mode & FMODE_WRITE) != 0; | |
774 | return pipe_release(inode, decr, decw); | |
775 | } | |
776 | ||
777 | static int | |
778 | pipe_read_open(struct inode *inode, struct file *filp) | |
779 | { | |
ad396024 EC |
780 | int ret = -ENOENT; |
781 | ||
9aeedfc4 | 782 | mutex_lock(&inode->i_mutex); |
ad396024 EC |
783 | |
784 | if (inode->i_pipe) { | |
785 | ret = 0; | |
786 | inode->i_pipe->readers++; | |
787 | } | |
788 | ||
9aeedfc4 | 789 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 790 | |
ad396024 | 791 | return ret; |
1da177e4 LT |
792 | } |
793 | ||
794 | static int | |
795 | pipe_write_open(struct inode *inode, struct file *filp) | |
796 | { | |
ad396024 EC |
797 | int ret = -ENOENT; |
798 | ||
9aeedfc4 | 799 | mutex_lock(&inode->i_mutex); |
ad396024 EC |
800 | |
801 | if (inode->i_pipe) { | |
802 | ret = 0; | |
803 | inode->i_pipe->writers++; | |
804 | } | |
805 | ||
9aeedfc4 | 806 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 807 | |
ad396024 | 808 | return ret; |
1da177e4 LT |
809 | } |
810 | ||
811 | static int | |
812 | pipe_rdwr_open(struct inode *inode, struct file *filp) | |
813 | { | |
ad396024 EC |
814 | int ret = -ENOENT; |
815 | ||
9aeedfc4 | 816 | mutex_lock(&inode->i_mutex); |
ad396024 EC |
817 | |
818 | if (inode->i_pipe) { | |
819 | ret = 0; | |
820 | if (filp->f_mode & FMODE_READ) | |
821 | inode->i_pipe->readers++; | |
822 | if (filp->f_mode & FMODE_WRITE) | |
823 | inode->i_pipe->writers++; | |
824 | } | |
825 | ||
9aeedfc4 | 826 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 827 | |
ad396024 | 828 | return ret; |
1da177e4 LT |
829 | } |
830 | ||
831 | /* | |
832 | * The file_operations structs are not static because they | |
833 | * are also used in linux/fs/fifo.c to do operations on FIFOs. | |
d2d9648e DV |
834 | * |
835 | * Pipes reuse fifos' file_operations structs. | |
1da177e4 | 836 | */ |
d2d9648e | 837 | const struct file_operations read_pipefifo_fops = { |
1da177e4 | 838 | .llseek = no_llseek, |
ee0b3e67 BP |
839 | .read = do_sync_read, |
840 | .aio_read = pipe_read, | |
1da177e4 LT |
841 | .write = bad_pipe_w, |
842 | .poll = pipe_poll, | |
d59d0b1b | 843 | .unlocked_ioctl = pipe_ioctl, |
1da177e4 LT |
844 | .open = pipe_read_open, |
845 | .release = pipe_read_release, | |
846 | .fasync = pipe_read_fasync, | |
847 | }; | |
848 | ||
d2d9648e | 849 | const struct file_operations write_pipefifo_fops = { |
1da177e4 LT |
850 | .llseek = no_llseek, |
851 | .read = bad_pipe_r, | |
ee0b3e67 BP |
852 | .write = do_sync_write, |
853 | .aio_write = pipe_write, | |
1da177e4 | 854 | .poll = pipe_poll, |
d59d0b1b | 855 | .unlocked_ioctl = pipe_ioctl, |
1da177e4 LT |
856 | .open = pipe_write_open, |
857 | .release = pipe_write_release, | |
858 | .fasync = pipe_write_fasync, | |
859 | }; | |
860 | ||
d2d9648e | 861 | const struct file_operations rdwr_pipefifo_fops = { |
1da177e4 | 862 | .llseek = no_llseek, |
ee0b3e67 BP |
863 | .read = do_sync_read, |
864 | .aio_read = pipe_read, | |
865 | .write = do_sync_write, | |
866 | .aio_write = pipe_write, | |
1da177e4 | 867 | .poll = pipe_poll, |
d59d0b1b | 868 | .unlocked_ioctl = pipe_ioctl, |
1da177e4 LT |
869 | .open = pipe_rdwr_open, |
870 | .release = pipe_rdwr_release, | |
871 | .fasync = pipe_rdwr_fasync, | |
872 | }; | |
873 | ||
3a326a2c IM |
874 | struct pipe_inode_info * alloc_pipe_info(struct inode *inode) |
875 | { | |
923f4f23 | 876 | struct pipe_inode_info *pipe; |
3a326a2c | 877 | |
923f4f23 IM |
878 | pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL); |
879 | if (pipe) { | |
880 | init_waitqueue_head(&pipe->wait); | |
881 | pipe->r_counter = pipe->w_counter = 1; | |
882 | pipe->inode = inode; | |
3a326a2c IM |
883 | } |
884 | ||
923f4f23 | 885 | return pipe; |
3a326a2c IM |
886 | } |
887 | ||
923f4f23 | 888 | void __free_pipe_info(struct pipe_inode_info *pipe) |
1da177e4 LT |
889 | { |
890 | int i; | |
1da177e4 | 891 | |
1da177e4 | 892 | for (i = 0; i < PIPE_BUFFERS; i++) { |
923f4f23 | 893 | struct pipe_buffer *buf = pipe->bufs + i; |
1da177e4 | 894 | if (buf->ops) |
923f4f23 | 895 | buf->ops->release(pipe, buf); |
1da177e4 | 896 | } |
923f4f23 IM |
897 | if (pipe->tmp_page) |
898 | __free_page(pipe->tmp_page); | |
899 | kfree(pipe); | |
1da177e4 LT |
900 | } |
901 | ||
b92ce558 JA |
902 | void free_pipe_info(struct inode *inode) |
903 | { | |
904 | __free_pipe_info(inode->i_pipe); | |
905 | inode->i_pipe = NULL; | |
906 | } | |
907 | ||
fa3536cc | 908 | static struct vfsmount *pipe_mnt __read_mostly; |
1da177e4 LT |
909 | static int pipefs_delete_dentry(struct dentry *dentry) |
910 | { | |
d18de5a2 ED |
911 | /* |
912 | * At creation time, we pretended this dentry was hashed | |
913 | * (by clearing DCACHE_UNHASHED bit in d_flags) | |
914 | * At delete time, we restore the truth : not hashed. | |
915 | * (so that dput() can proceed correctly) | |
916 | */ | |
917 | dentry->d_flags |= DCACHE_UNHASHED; | |
918 | return 0; | |
1da177e4 | 919 | } |
341b446b | 920 | |
c23fbb6b ED |
921 | /* |
922 | * pipefs_dname() is called from d_path(). | |
923 | */ | |
924 | static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen) | |
925 | { | |
926 | return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", | |
927 | dentry->d_inode->i_ino); | |
928 | } | |
929 | ||
3ba13d17 | 930 | static const struct dentry_operations pipefs_dentry_operations = { |
1da177e4 | 931 | .d_delete = pipefs_delete_dentry, |
c23fbb6b | 932 | .d_dname = pipefs_dname, |
1da177e4 LT |
933 | }; |
934 | ||
935 | static struct inode * get_pipe_inode(void) | |
936 | { | |
937 | struct inode *inode = new_inode(pipe_mnt->mnt_sb); | |
923f4f23 | 938 | struct pipe_inode_info *pipe; |
1da177e4 LT |
939 | |
940 | if (!inode) | |
941 | goto fail_inode; | |
942 | ||
923f4f23 IM |
943 | pipe = alloc_pipe_info(inode); |
944 | if (!pipe) | |
1da177e4 | 945 | goto fail_iput; |
923f4f23 | 946 | inode->i_pipe = pipe; |
3a326a2c | 947 | |
923f4f23 | 948 | pipe->readers = pipe->writers = 1; |
d2d9648e | 949 | inode->i_fop = &rdwr_pipefifo_fops; |
1da177e4 LT |
950 | |
951 | /* | |
952 | * Mark the inode dirty from the very beginning, | |
953 | * that way it will never be moved to the dirty | |
954 | * list because "mark_inode_dirty()" will think | |
955 | * that it already _is_ on the dirty list. | |
956 | */ | |
957 | inode->i_state = I_DIRTY; | |
958 | inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR; | |
da9592ed DH |
959 | inode->i_uid = current_fsuid(); |
960 | inode->i_gid = current_fsgid(); | |
1da177e4 | 961 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
923f4f23 | 962 | |
1da177e4 LT |
963 | return inode; |
964 | ||
965 | fail_iput: | |
966 | iput(inode); | |
341b446b | 967 | |
1da177e4 LT |
968 | fail_inode: |
969 | return NULL; | |
970 | } | |
971 | ||
be61a86d | 972 | struct file *create_write_pipe(int flags) |
1da177e4 | 973 | { |
d6cbd281 AK |
974 | int err; |
975 | struct inode *inode; | |
976 | struct file *f; | |
1da177e4 | 977 | struct dentry *dentry; |
c23fbb6b | 978 | struct qstr name = { .name = "" }; |
1da177e4 | 979 | |
d6cbd281 | 980 | err = -ENFILE; |
1da177e4 LT |
981 | inode = get_pipe_inode(); |
982 | if (!inode) | |
430e285e | 983 | goto err; |
1da177e4 | 984 | |
d6cbd281 | 985 | err = -ENOMEM; |
c23fbb6b | 986 | dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name); |
1da177e4 | 987 | if (!dentry) |
d6cbd281 | 988 | goto err_inode; |
341b446b | 989 | |
1da177e4 | 990 | dentry->d_op = &pipefs_dentry_operations; |
d18de5a2 ED |
991 | /* |
992 | * We dont want to publish this dentry into global dentry hash table. | |
993 | * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED | |
994 | * This permits a working /proc/$pid/fd/XXX on pipes | |
995 | */ | |
996 | dentry->d_flags &= ~DCACHE_UNHASHED; | |
997 | d_instantiate(dentry, inode); | |
430e285e DH |
998 | |
999 | err = -ENFILE; | |
d2d9648e | 1000 | f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipefifo_fops); |
430e285e DH |
1001 | if (!f) |
1002 | goto err_dentry; | |
d6cbd281 | 1003 | f->f_mapping = inode->i_mapping; |
341b446b | 1004 | |
be61a86d | 1005 | f->f_flags = O_WRONLY | (flags & O_NONBLOCK); |
d6cbd281 AK |
1006 | f->f_version = 0; |
1007 | ||
1008 | return f; | |
1da177e4 | 1009 | |
430e285e | 1010 | err_dentry: |
ed152437 | 1011 | free_pipe_info(inode); |
430e285e | 1012 | dput(dentry); |
ed152437 AV |
1013 | return ERR_PTR(err); |
1014 | ||
d6cbd281 | 1015 | err_inode: |
1da177e4 LT |
1016 | free_pipe_info(inode); |
1017 | iput(inode); | |
430e285e | 1018 | err: |
d6cbd281 AK |
1019 | return ERR_PTR(err); |
1020 | } | |
1021 | ||
1022 | void free_write_pipe(struct file *f) | |
1023 | { | |
5ccac88e | 1024 | free_pipe_info(f->f_dentry->d_inode); |
c8e7f449 | 1025 | path_put(&f->f_path); |
d6cbd281 AK |
1026 | put_filp(f); |
1027 | } | |
1028 | ||
be61a86d | 1029 | struct file *create_read_pipe(struct file *wrf, int flags) |
d6cbd281 AK |
1030 | { |
1031 | struct file *f = get_empty_filp(); | |
1032 | if (!f) | |
1033 | return ERR_PTR(-ENFILE); | |
1034 | ||
1035 | /* Grab pipe from the writer */ | |
c8e7f449 JB |
1036 | f->f_path = wrf->f_path; |
1037 | path_get(&wrf->f_path); | |
0f7fc9e4 | 1038 | f->f_mapping = wrf->f_path.dentry->d_inode->i_mapping; |
d6cbd281 AK |
1039 | |
1040 | f->f_pos = 0; | |
be61a86d | 1041 | f->f_flags = O_RDONLY | (flags & O_NONBLOCK); |
d2d9648e | 1042 | f->f_op = &read_pipefifo_fops; |
d6cbd281 AK |
1043 | f->f_mode = FMODE_READ; |
1044 | f->f_version = 0; | |
1045 | ||
1046 | return f; | |
1047 | } | |
1048 | ||
ed8cae8b | 1049 | int do_pipe_flags(int *fd, int flags) |
d6cbd281 AK |
1050 | { |
1051 | struct file *fw, *fr; | |
1052 | int error; | |
1053 | int fdw, fdr; | |
1054 | ||
be61a86d | 1055 | if (flags & ~(O_CLOEXEC | O_NONBLOCK)) |
ed8cae8b UD |
1056 | return -EINVAL; |
1057 | ||
be61a86d | 1058 | fw = create_write_pipe(flags); |
d6cbd281 AK |
1059 | if (IS_ERR(fw)) |
1060 | return PTR_ERR(fw); | |
be61a86d | 1061 | fr = create_read_pipe(fw, flags); |
d6cbd281 AK |
1062 | error = PTR_ERR(fr); |
1063 | if (IS_ERR(fr)) | |
1064 | goto err_write_pipe; | |
1065 | ||
ed8cae8b | 1066 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
1067 | if (error < 0) |
1068 | goto err_read_pipe; | |
1069 | fdr = error; | |
1070 | ||
ed8cae8b | 1071 | error = get_unused_fd_flags(flags); |
d6cbd281 AK |
1072 | if (error < 0) |
1073 | goto err_fdr; | |
1074 | fdw = error; | |
1075 | ||
157cf649 | 1076 | audit_fd_pair(fdr, fdw); |
d6cbd281 AK |
1077 | fd_install(fdr, fr); |
1078 | fd_install(fdw, fw); | |
1079 | fd[0] = fdr; | |
1080 | fd[1] = fdw; | |
1081 | ||
1082 | return 0; | |
1083 | ||
1084 | err_fdr: | |
1085 | put_unused_fd(fdr); | |
1086 | err_read_pipe: | |
c8e7f449 | 1087 | path_put(&fr->f_path); |
d6cbd281 AK |
1088 | put_filp(fr); |
1089 | err_write_pipe: | |
1090 | free_write_pipe(fw); | |
1091 | return error; | |
1da177e4 LT |
1092 | } |
1093 | ||
d35c7b0e UD |
1094 | /* |
1095 | * sys_pipe() is the normal C calling standard for creating | |
1096 | * a pipe. It's not the way Unix traditionally does this, though. | |
1097 | */ | |
d4e82042 | 1098 | SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags) |
d35c7b0e UD |
1099 | { |
1100 | int fd[2]; | |
1101 | int error; | |
1102 | ||
ed8cae8b | 1103 | error = do_pipe_flags(fd, flags); |
d35c7b0e | 1104 | if (!error) { |
ba719bae UD |
1105 | if (copy_to_user(fildes, fd, sizeof(fd))) { |
1106 | sys_close(fd[0]); | |
1107 | sys_close(fd[1]); | |
d35c7b0e | 1108 | error = -EFAULT; |
ba719bae | 1109 | } |
d35c7b0e UD |
1110 | } |
1111 | return error; | |
1112 | } | |
1113 | ||
2b664219 | 1114 | SYSCALL_DEFINE1(pipe, int __user *, fildes) |
ed8cae8b UD |
1115 | { |
1116 | return sys_pipe2(fildes, 0); | |
1117 | } | |
1118 | ||
1da177e4 LT |
1119 | /* |
1120 | * pipefs should _never_ be mounted by userland - too much of security hassle, | |
1121 | * no real gain from having the whole whorehouse mounted. So we don't need | |
1122 | * any operations on the root directory. However, we need a non-trivial | |
1123 | * d_name - pipe: will go nicely and kill the special-casing in procfs. | |
1124 | */ | |
454e2398 DH |
1125 | static int pipefs_get_sb(struct file_system_type *fs_type, |
1126 | int flags, const char *dev_name, void *data, | |
1127 | struct vfsmount *mnt) | |
1da177e4 | 1128 | { |
454e2398 | 1129 | return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt); |
1da177e4 LT |
1130 | } |
1131 | ||
1132 | static struct file_system_type pipe_fs_type = { | |
1133 | .name = "pipefs", | |
1134 | .get_sb = pipefs_get_sb, | |
1135 | .kill_sb = kill_anon_super, | |
1136 | }; | |
1137 | ||
1138 | static int __init init_pipe_fs(void) | |
1139 | { | |
1140 | int err = register_filesystem(&pipe_fs_type); | |
341b446b | 1141 | |
1da177e4 LT |
1142 | if (!err) { |
1143 | pipe_mnt = kern_mount(&pipe_fs_type); | |
1144 | if (IS_ERR(pipe_mnt)) { | |
1145 | err = PTR_ERR(pipe_mnt); | |
1146 | unregister_filesystem(&pipe_fs_type); | |
1147 | } | |
1148 | } | |
1149 | return err; | |
1150 | } | |
1151 | ||
1152 | static void __exit exit_pipe_fs(void) | |
1153 | { | |
1154 | unregister_filesystem(&pipe_fs_type); | |
1155 | mntput(pipe_mnt); | |
1156 | } | |
1157 | ||
1158 | fs_initcall(init_pipe_fs); | |
1159 | module_exit(exit_pipe_fs); |