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