l2tp: take a reference for kernel sockets in l2tp_tunnel_sock_lookup
[linux-2.6-block.git] / fs / pipe.c
CommitLineData
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>
1da177e4
LT
24
25#include <asm/uaccess.h>
26#include <asm/ioctls.h>
27
b492e95b
JA
28/*
29 * The max size that a non-root user is allowed to grow the pipe. Can
ff9da691 30 * be set by root in /proc/sys/fs/pipe-max-size
b492e95b 31 */
ff9da691
JA
32unsigned int pipe_max_size = 1048576;
33
34/*
35 * Minimum pipe size, as required by POSIX
36 */
37unsigned int pipe_min_size = PAGE_SIZE;
b492e95b 38
1da177e4
LT
39/*
40 * We use a start+len construction, which provides full use of the
41 * allocated memory.
42 * -- Florian Coosmann (FGC)
43 *
44 * Reads with count = 0 should always return 0.
45 * -- Julian Bradfield 1999-06-07.
46 *
47 * FIFOs and Pipes now generate SIGIO for both readers and writers.
48 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
49 *
50 * pipe_read & write cleanup
51 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
52 */
53
61e0d47c
MS
54static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
55{
56 if (pipe->inode)
57 mutex_lock_nested(&pipe->inode->i_mutex, subclass);
58}
59
60void pipe_lock(struct pipe_inode_info *pipe)
61{
62 /*
63 * pipe_lock() nests non-pipe inode locks (for writing to a file)
64 */
65 pipe_lock_nested(pipe, I_MUTEX_PARENT);
66}
67EXPORT_SYMBOL(pipe_lock);
68
69void pipe_unlock(struct pipe_inode_info *pipe)
70{
71 if (pipe->inode)
72 mutex_unlock(&pipe->inode->i_mutex);
73}
74EXPORT_SYMBOL(pipe_unlock);
75
76void pipe_double_lock(struct pipe_inode_info *pipe1,
77 struct pipe_inode_info *pipe2)
78{
79 BUG_ON(pipe1 == pipe2);
80
81 if (pipe1 < pipe2) {
82 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
83 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
84 } else {
023d43c7
PZ
85 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
86 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
61e0d47c
MS
87 }
88}
89
1da177e4 90/* Drop the inode semaphore and wait for a pipe event, atomically */
3a326a2c 91void pipe_wait(struct pipe_inode_info *pipe)
1da177e4
LT
92{
93 DEFINE_WAIT(wait);
94
d79fc0fc
IM
95 /*
96 * Pipes are system-local resources, so sleeping on them
97 * is considered a noninteractive wait:
98 */
af927232 99 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
61e0d47c 100 pipe_unlock(pipe);
1da177e4 101 schedule();
3a326a2c 102 finish_wait(&pipe->wait, &wait);
61e0d47c 103 pipe_lock(pipe);
1da177e4
LT
104}
105
858119e1 106static int
f6762b7a
JA
107pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
108 int atomic)
1da177e4
LT
109{
110 unsigned long copy;
111
112 while (len > 0) {
113 while (!iov->iov_len)
114 iov++;
115 copy = min_t(unsigned long, len, iov->iov_len);
116
f6762b7a
JA
117 if (atomic) {
118 if (__copy_from_user_inatomic(to, iov->iov_base, copy))
119 return -EFAULT;
120 } else {
121 if (copy_from_user(to, iov->iov_base, copy))
122 return -EFAULT;
123 }
1da177e4
LT
124 to += copy;
125 len -= copy;
126 iov->iov_base += copy;
127 iov->iov_len -= copy;
128 }
129 return 0;
130}
131
858119e1 132static int
f6762b7a
JA
133pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
134 int atomic)
1da177e4
LT
135{
136 unsigned long copy;
137
138 while (len > 0) {
139 while (!iov->iov_len)
140 iov++;
141 copy = min_t(unsigned long, len, iov->iov_len);
142
f6762b7a
JA
143 if (atomic) {
144 if (__copy_to_user_inatomic(iov->iov_base, from, copy))
145 return -EFAULT;
146 } else {
147 if (copy_to_user(iov->iov_base, from, copy))
148 return -EFAULT;
149 }
1da177e4
LT
150 from += copy;
151 len -= copy;
152 iov->iov_base += copy;
153 iov->iov_len -= copy;
154 }
155 return 0;
156}
157
f6762b7a
JA
158/*
159 * Attempt to pre-fault in the user memory, so we can use atomic copies.
160 * Returns the number of bytes not faulted in.
161 */
162static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
163{
164 while (!iov->iov_len)
165 iov++;
166
167 while (len > 0) {
168 unsigned long this_len;
169
170 this_len = min_t(unsigned long, len, iov->iov_len);
171 if (fault_in_pages_writeable(iov->iov_base, this_len))
172 break;
173
174 len -= this_len;
175 iov++;
176 }
177
178 return len;
179}
180
181/*
182 * Pre-fault in the user memory, so we can use atomic copies.
183 */
184static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
185{
186 while (!iov->iov_len)
187 iov++;
188
189 while (len > 0) {
190 unsigned long this_len;
191
192 this_len = min_t(unsigned long, len, iov->iov_len);
193 fault_in_pages_readable(iov->iov_base, this_len);
194 len -= this_len;
195 iov++;
196 }
197}
198
341b446b
IM
199static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
200 struct pipe_buffer *buf)
1da177e4
LT
201{
202 struct page *page = buf->page;
203
5274f052
JA
204 /*
205 * If nobody else uses this page, and we don't already have a
206 * temporary page, let's keep track of it as a one-deep
341b446b 207 * allocation cache. (Otherwise just release our reference to it)
5274f052 208 */
341b446b 209 if (page_count(page) == 1 && !pipe->tmp_page)
923f4f23 210 pipe->tmp_page = page;
341b446b
IM
211 else
212 page_cache_release(page);
1da177e4
LT
213}
214
0845718d
JA
215/**
216 * generic_pipe_buf_map - virtually map a pipe buffer
217 * @pipe: the pipe that the buffer belongs to
218 * @buf: the buffer that should be mapped
219 * @atomic: whether to use an atomic map
220 *
221 * Description:
222 * This function returns a kernel virtual address mapping for the
b51d63c6 223 * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
0845718d
JA
224 * and the caller has to be careful not to fault before calling
225 * the unmap function.
226 *
2164d334 227 * Note that this function calls kmap_atomic() if @atomic != 0.
0845718d 228 */
f84d7519 229void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
f6762b7a 230 struct pipe_buffer *buf, int atomic)
1da177e4 231{
f6762b7a
JA
232 if (atomic) {
233 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
e8e3c3d6 234 return kmap_atomic(buf->page);
f6762b7a
JA
235 }
236
1da177e4
LT
237 return kmap(buf->page);
238}
51921cb7 239EXPORT_SYMBOL(generic_pipe_buf_map);
1da177e4 240
0845718d
JA
241/**
242 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
243 * @pipe: the pipe that the buffer belongs to
244 * @buf: the buffer that should be unmapped
245 * @map_data: the data that the mapping function returned
246 *
247 * Description:
248 * This function undoes the mapping that ->map() provided.
249 */
f84d7519 250void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
f6762b7a 251 struct pipe_buffer *buf, void *map_data)
1da177e4 252{
f6762b7a
JA
253 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
254 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
e8e3c3d6 255 kunmap_atomic(map_data);
f6762b7a
JA
256 } else
257 kunmap(buf->page);
1da177e4 258}
51921cb7 259EXPORT_SYMBOL(generic_pipe_buf_unmap);
1da177e4 260
0845718d 261/**
b51d63c6 262 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
0845718d
JA
263 * @pipe: the pipe that the buffer belongs to
264 * @buf: the buffer to attempt to steal
265 *
266 * Description:
b51d63c6 267 * This function attempts to steal the &struct page attached to
0845718d
JA
268 * @buf. If successful, this function returns 0 and returns with
269 * the page locked. The caller may then reuse the page for whatever
b51d63c6 270 * he wishes; the typical use is insertion into a different file
0845718d
JA
271 * page cache.
272 */
330ab716
JA
273int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
274 struct pipe_buffer *buf)
5abc97aa 275{
46e678c9
JA
276 struct page *page = buf->page;
277
0845718d
JA
278 /*
279 * A reference of one is golden, that means that the owner of this
280 * page is the only one holding a reference to it. lock the page
281 * and return OK.
282 */
46e678c9 283 if (page_count(page) == 1) {
46e678c9
JA
284 lock_page(page);
285 return 0;
286 }
287
288 return 1;
5abc97aa 289}
51921cb7 290EXPORT_SYMBOL(generic_pipe_buf_steal);
5abc97aa 291
0845718d 292/**
b51d63c6 293 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
0845718d
JA
294 * @pipe: the pipe that the buffer belongs to
295 * @buf: the buffer to get a reference to
296 *
297 * Description:
298 * This function grabs an extra reference to @buf. It's used in
299 * in the tee() system call, when we duplicate the buffers in one
300 * pipe into another.
301 */
302void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
70524490
JA
303{
304 page_cache_get(buf->page);
305}
51921cb7 306EXPORT_SYMBOL(generic_pipe_buf_get);
70524490 307
0845718d
JA
308/**
309 * generic_pipe_buf_confirm - verify contents of the pipe buffer
79685b8d 310 * @info: the pipe that the buffer belongs to
0845718d
JA
311 * @buf: the buffer to confirm
312 *
313 * Description:
314 * This function does nothing, because the generic pipe code uses
315 * pages that are always good when inserted into the pipe.
316 */
cac36bb0
JA
317int generic_pipe_buf_confirm(struct pipe_inode_info *info,
318 struct pipe_buffer *buf)
f84d7519
JA
319{
320 return 0;
321}
51921cb7 322EXPORT_SYMBOL(generic_pipe_buf_confirm);
f84d7519 323
6818173b
MS
324/**
325 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
326 * @pipe: the pipe that the buffer belongs to
327 * @buf: the buffer to put a reference to
328 *
329 * Description:
330 * This function releases a reference to @buf.
331 */
332void generic_pipe_buf_release(struct pipe_inode_info *pipe,
333 struct pipe_buffer *buf)
334{
335 page_cache_release(buf->page);
336}
51921cb7 337EXPORT_SYMBOL(generic_pipe_buf_release);
6818173b 338
d4c3cca9 339static const struct pipe_buf_operations anon_pipe_buf_ops = {
1da177e4 340 .can_merge = 1,
f84d7519
JA
341 .map = generic_pipe_buf_map,
342 .unmap = generic_pipe_buf_unmap,
cac36bb0 343 .confirm = generic_pipe_buf_confirm,
1da177e4 344 .release = anon_pipe_buf_release,
330ab716 345 .steal = generic_pipe_buf_steal,
f84d7519 346 .get = generic_pipe_buf_get,
1da177e4
LT
347};
348
9883035a
LT
349static const struct pipe_buf_operations packet_pipe_buf_ops = {
350 .can_merge = 0,
351 .map = generic_pipe_buf_map,
352 .unmap = generic_pipe_buf_unmap,
353 .confirm = generic_pipe_buf_confirm,
354 .release = anon_pipe_buf_release,
355 .steal = generic_pipe_buf_steal,
356 .get = generic_pipe_buf_get,
357};
358
1da177e4 359static ssize_t
ee0b3e67
BP
360pipe_read(struct kiocb *iocb, const struct iovec *_iov,
361 unsigned long nr_segs, loff_t pos)
1da177e4 362{
ee0b3e67 363 struct file *filp = iocb->ki_filp;
496ad9aa 364 struct inode *inode = file_inode(filp);
923f4f23 365 struct pipe_inode_info *pipe;
1da177e4
LT
366 int do_wakeup;
367 ssize_t ret;
368 struct iovec *iov = (struct iovec *)_iov;
369 size_t total_len;
370
371 total_len = iov_length(iov, nr_segs);
372 /* Null read succeeds. */
373 if (unlikely(total_len == 0))
374 return 0;
375
376 do_wakeup = 0;
377 ret = 0;
9aeedfc4 378 mutex_lock(&inode->i_mutex);
923f4f23 379 pipe = inode->i_pipe;
1da177e4 380 for (;;) {
923f4f23 381 int bufs = pipe->nrbufs;
1da177e4 382 if (bufs) {
923f4f23
IM
383 int curbuf = pipe->curbuf;
384 struct pipe_buffer *buf = pipe->bufs + curbuf;
d4c3cca9 385 const struct pipe_buf_operations *ops = buf->ops;
1da177e4
LT
386 void *addr;
387 size_t chars = buf->len;
f6762b7a 388 int error, atomic;
1da177e4
LT
389
390 if (chars > total_len)
391 chars = total_len;
392
cac36bb0 393 error = ops->confirm(pipe, buf);
f84d7519 394 if (error) {
5274f052 395 if (!ret)
e5953cbd 396 ret = error;
5274f052
JA
397 break;
398 }
f84d7519 399
f6762b7a
JA
400 atomic = !iov_fault_in_pages_write(iov, chars);
401redo:
402 addr = ops->map(pipe, buf, atomic);
403 error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
404 ops->unmap(pipe, buf, addr);
1da177e4 405 if (unlikely(error)) {
f6762b7a
JA
406 /*
407 * Just retry with the slow path if we failed.
408 */
409 if (atomic) {
410 atomic = 0;
411 goto redo;
412 }
341b446b 413 if (!ret)
f6762b7a 414 ret = error;
1da177e4
LT
415 break;
416 }
417 ret += chars;
418 buf->offset += chars;
419 buf->len -= chars;
9883035a
LT
420
421 /* Was it a packet buffer? Clean up and exit */
422 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
423 total_len = chars;
424 buf->len = 0;
425 }
426
1da177e4
LT
427 if (!buf->len) {
428 buf->ops = NULL;
923f4f23 429 ops->release(pipe, buf);
35f3d14d 430 curbuf = (curbuf + 1) & (pipe->buffers - 1);
923f4f23
IM
431 pipe->curbuf = curbuf;
432 pipe->nrbufs = --bufs;
1da177e4
LT
433 do_wakeup = 1;
434 }
435 total_len -= chars;
436 if (!total_len)
437 break; /* common path: read succeeded */
438 }
439 if (bufs) /* More to do? */
440 continue;
923f4f23 441 if (!pipe->writers)
1da177e4 442 break;
923f4f23 443 if (!pipe->waiting_writers) {
1da177e4
LT
444 /* syscall merging: Usually we must not sleep
445 * if O_NONBLOCK is set, or if we got some data.
446 * But if a writer sleeps in kernel space, then
447 * we can wait for that data without violating POSIX.
448 */
449 if (ret)
450 break;
451 if (filp->f_flags & O_NONBLOCK) {
452 ret = -EAGAIN;
453 break;
454 }
455 }
456 if (signal_pending(current)) {
341b446b
IM
457 if (!ret)
458 ret = -ERESTARTSYS;
1da177e4
LT
459 break;
460 }
461 if (do_wakeup) {
28e58ee8 462 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
923f4f23 463 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
1da177e4 464 }
923f4f23 465 pipe_wait(pipe);
1da177e4 466 }
9aeedfc4 467 mutex_unlock(&inode->i_mutex);
341b446b
IM
468
469 /* Signal writers asynchronously that there is more room. */
1da177e4 470 if (do_wakeup) {
28e58ee8 471 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
923f4f23 472 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
1da177e4
LT
473 }
474 if (ret > 0)
475 file_accessed(filp);
476 return ret;
477}
478
9883035a
LT
479static inline int is_packetized(struct file *file)
480{
481 return (file->f_flags & O_DIRECT) != 0;
482}
483
1da177e4 484static ssize_t
ee0b3e67
BP
485pipe_write(struct kiocb *iocb, const struct iovec *_iov,
486 unsigned long nr_segs, loff_t ppos)
1da177e4 487{
ee0b3e67 488 struct file *filp = iocb->ki_filp;
496ad9aa 489 struct inode *inode = file_inode(filp);
923f4f23 490 struct pipe_inode_info *pipe;
1da177e4
LT
491 ssize_t ret;
492 int do_wakeup;
493 struct iovec *iov = (struct iovec *)_iov;
494 size_t total_len;
495 ssize_t chars;
496
497 total_len = iov_length(iov, nr_segs);
498 /* Null write succeeds. */
499 if (unlikely(total_len == 0))
500 return 0;
501
502 do_wakeup = 0;
503 ret = 0;
9aeedfc4 504 mutex_lock(&inode->i_mutex);
923f4f23 505 pipe = inode->i_pipe;
1da177e4 506
923f4f23 507 if (!pipe->readers) {
1da177e4
LT
508 send_sig(SIGPIPE, current, 0);
509 ret = -EPIPE;
510 goto out;
511 }
512
513 /* We try to merge small writes */
514 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
923f4f23 515 if (pipe->nrbufs && chars != 0) {
341b446b 516 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
35f3d14d 517 (pipe->buffers - 1);
923f4f23 518 struct pipe_buffer *buf = pipe->bufs + lastbuf;
d4c3cca9 519 const struct pipe_buf_operations *ops = buf->ops;
1da177e4 520 int offset = buf->offset + buf->len;
341b446b 521
1da177e4 522 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
f6762b7a 523 int error, atomic = 1;
5274f052 524 void *addr;
5274f052 525
cac36bb0 526 error = ops->confirm(pipe, buf);
f84d7519 527 if (error)
5274f052 528 goto out;
f84d7519 529
f6762b7a
JA
530 iov_fault_in_pages_read(iov, chars);
531redo1:
532 addr = ops->map(pipe, buf, atomic);
5274f052 533 error = pipe_iov_copy_from_user(offset + addr, iov,
f6762b7a
JA
534 chars, atomic);
535 ops->unmap(pipe, buf, addr);
1da177e4
LT
536 ret = error;
537 do_wakeup = 1;
f6762b7a
JA
538 if (error) {
539 if (atomic) {
540 atomic = 0;
541 goto redo1;
542 }
1da177e4 543 goto out;
f6762b7a 544 }
1da177e4
LT
545 buf->len += chars;
546 total_len -= chars;
547 ret = chars;
548 if (!total_len)
549 goto out;
550 }
551 }
552
553 for (;;) {
554 int bufs;
341b446b 555
923f4f23 556 if (!pipe->readers) {
1da177e4 557 send_sig(SIGPIPE, current, 0);
341b446b
IM
558 if (!ret)
559 ret = -EPIPE;
1da177e4
LT
560 break;
561 }
923f4f23 562 bufs = pipe->nrbufs;
35f3d14d
JA
563 if (bufs < pipe->buffers) {
564 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
923f4f23
IM
565 struct pipe_buffer *buf = pipe->bufs + newbuf;
566 struct page *page = pipe->tmp_page;
f6762b7a
JA
567 char *src;
568 int error, atomic = 1;
1da177e4
LT
569
570 if (!page) {
571 page = alloc_page(GFP_HIGHUSER);
572 if (unlikely(!page)) {
573 ret = ret ? : -ENOMEM;
574 break;
575 }
923f4f23 576 pipe->tmp_page = page;
1da177e4 577 }
341b446b 578 /* Always wake up, even if the copy fails. Otherwise
1da177e4
LT
579 * we lock up (O_NONBLOCK-)readers that sleep due to
580 * syscall merging.
581 * FIXME! Is this really true?
582 */
583 do_wakeup = 1;
584 chars = PAGE_SIZE;
585 if (chars > total_len)
586 chars = total_len;
587
f6762b7a
JA
588 iov_fault_in_pages_read(iov, chars);
589redo2:
590 if (atomic)
e8e3c3d6 591 src = kmap_atomic(page);
f6762b7a
JA
592 else
593 src = kmap(page);
594
595 error = pipe_iov_copy_from_user(src, iov, chars,
596 atomic);
597 if (atomic)
e8e3c3d6 598 kunmap_atomic(src);
f6762b7a
JA
599 else
600 kunmap(page);
601
1da177e4 602 if (unlikely(error)) {
f6762b7a
JA
603 if (atomic) {
604 atomic = 0;
605 goto redo2;
606 }
341b446b 607 if (!ret)
f6762b7a 608 ret = error;
1da177e4
LT
609 break;
610 }
611 ret += chars;
612
613 /* Insert it into the buffer array */
614 buf->page = page;
615 buf->ops = &anon_pipe_buf_ops;
616 buf->offset = 0;
617 buf->len = chars;
9883035a
LT
618 buf->flags = 0;
619 if (is_packetized(filp)) {
620 buf->ops = &packet_pipe_buf_ops;
621 buf->flags = PIPE_BUF_FLAG_PACKET;
622 }
923f4f23
IM
623 pipe->nrbufs = ++bufs;
624 pipe->tmp_page = NULL;
1da177e4
LT
625
626 total_len -= chars;
627 if (!total_len)
628 break;
629 }
35f3d14d 630 if (bufs < pipe->buffers)
1da177e4
LT
631 continue;
632 if (filp->f_flags & O_NONBLOCK) {
341b446b
IM
633 if (!ret)
634 ret = -EAGAIN;
1da177e4
LT
635 break;
636 }
637 if (signal_pending(current)) {
341b446b
IM
638 if (!ret)
639 ret = -ERESTARTSYS;
1da177e4
LT
640 break;
641 }
642 if (do_wakeup) {
28e58ee8 643 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
923f4f23 644 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1da177e4
LT
645 do_wakeup = 0;
646 }
923f4f23
IM
647 pipe->waiting_writers++;
648 pipe_wait(pipe);
649 pipe->waiting_writers--;
1da177e4
LT
650 }
651out:
9aeedfc4 652 mutex_unlock(&inode->i_mutex);
1da177e4 653 if (do_wakeup) {
28e58ee8 654 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
923f4f23 655 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1da177e4 656 }
c3b2da31
JB
657 if (ret > 0) {
658 int err = file_update_time(filp);
659 if (err)
660 ret = err;
661 }
1da177e4
LT
662 return ret;
663}
664
1da177e4
LT
665static ssize_t
666bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
667{
668 return -EBADF;
669}
670
671static ssize_t
341b446b
IM
672bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
673 loff_t *ppos)
1da177e4
LT
674{
675 return -EBADF;
676}
677
d59d0b1b 678static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1da177e4 679{
496ad9aa 680 struct inode *inode = file_inode(filp);
923f4f23 681 struct pipe_inode_info *pipe;
1da177e4
LT
682 int count, buf, nrbufs;
683
684 switch (cmd) {
685 case FIONREAD:
9aeedfc4 686 mutex_lock(&inode->i_mutex);
923f4f23 687 pipe = inode->i_pipe;
1da177e4 688 count = 0;
923f4f23
IM
689 buf = pipe->curbuf;
690 nrbufs = pipe->nrbufs;
1da177e4 691 while (--nrbufs >= 0) {
923f4f23 692 count += pipe->bufs[buf].len;
35f3d14d 693 buf = (buf+1) & (pipe->buffers - 1);
1da177e4 694 }
9aeedfc4 695 mutex_unlock(&inode->i_mutex);
923f4f23 696
1da177e4
LT
697 return put_user(count, (int __user *)arg);
698 default:
46ce341b 699 return -ENOIOCTLCMD;
1da177e4
LT
700 }
701}
702
703/* No kernel lock held - fine */
704static unsigned int
705pipe_poll(struct file *filp, poll_table *wait)
706{
707 unsigned int mask;
496ad9aa 708 struct inode *inode = file_inode(filp);
923f4f23 709 struct pipe_inode_info *pipe = inode->i_pipe;
1da177e4
LT
710 int nrbufs;
711
923f4f23 712 poll_wait(filp, &pipe->wait, wait);
1da177e4
LT
713
714 /* Reading only -- no need for acquiring the semaphore. */
923f4f23 715 nrbufs = pipe->nrbufs;
1da177e4
LT
716 mask = 0;
717 if (filp->f_mode & FMODE_READ) {
718 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
923f4f23 719 if (!pipe->writers && filp->f_version != pipe->w_counter)
1da177e4
LT
720 mask |= POLLHUP;
721 }
722
723 if (filp->f_mode & FMODE_WRITE) {
35f3d14d 724 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
5e5d7a22
PE
725 /*
726 * Most Unices do not set POLLERR for FIFOs but on Linux they
727 * behave exactly like pipes for poll().
728 */
923f4f23 729 if (!pipe->readers)
1da177e4
LT
730 mask |= POLLERR;
731 }
732
733 return mask;
734}
735
1da177e4
LT
736static int
737pipe_release(struct inode *inode, int decr, int decw)
738{
923f4f23
IM
739 struct pipe_inode_info *pipe;
740
9aeedfc4 741 mutex_lock(&inode->i_mutex);
923f4f23
IM
742 pipe = inode->i_pipe;
743 pipe->readers -= decr;
744 pipe->writers -= decw;
341b446b 745
923f4f23 746 if (!pipe->readers && !pipe->writers) {
1da177e4
LT
747 free_pipe_info(inode);
748 } else {
28e58ee8 749 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
923f4f23
IM
750 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
751 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
1da177e4 752 }
9aeedfc4 753 mutex_unlock(&inode->i_mutex);
1da177e4
LT
754
755 return 0;
756}
757
758static int
759pipe_read_fasync(int fd, struct file *filp, int on)
760{
496ad9aa 761 struct inode *inode = file_inode(filp);
1da177e4
LT
762 int retval;
763
9aeedfc4
IM
764 mutex_lock(&inode->i_mutex);
765 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
766 mutex_unlock(&inode->i_mutex);
1da177e4 767
60aa4924 768 return retval;
1da177e4
LT
769}
770
771
772static int
773pipe_write_fasync(int fd, struct file *filp, int on)
774{
496ad9aa 775 struct inode *inode = file_inode(filp);
1da177e4
LT
776 int retval;
777
9aeedfc4
IM
778 mutex_lock(&inode->i_mutex);
779 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
780 mutex_unlock(&inode->i_mutex);
1da177e4 781
60aa4924 782 return retval;
1da177e4
LT
783}
784
785
786static int
787pipe_rdwr_fasync(int fd, struct file *filp, int on)
788{
496ad9aa 789 struct inode *inode = file_inode(filp);
341b446b 790 struct pipe_inode_info *pipe = inode->i_pipe;
1da177e4
LT
791 int retval;
792
9aeedfc4 793 mutex_lock(&inode->i_mutex);
341b446b 794 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
e5bc49ba 795 if (retval >= 0) {
341b446b 796 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
e5bc49ba
ON
797 if (retval < 0) /* this can happen only if on == T */
798 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
799 }
9aeedfc4 800 mutex_unlock(&inode->i_mutex);
60aa4924 801 return retval;
1da177e4
LT
802}
803
804
805static int
806pipe_read_release(struct inode *inode, struct file *filp)
807{
1da177e4
LT
808 return pipe_release(inode, 1, 0);
809}
810
811static int
812pipe_write_release(struct inode *inode, struct file *filp)
813{
1da177e4
LT
814 return pipe_release(inode, 0, 1);
815}
816
817static int
818pipe_rdwr_release(struct inode *inode, struct file *filp)
819{
820 int decr, decw;
821
1da177e4
LT
822 decr = (filp->f_mode & FMODE_READ) != 0;
823 decw = (filp->f_mode & FMODE_WRITE) != 0;
824 return pipe_release(inode, decr, decw);
825}
826
827static int
828pipe_read_open(struct inode *inode, struct file *filp)
829{
ad396024
EC
830 int ret = -ENOENT;
831
9aeedfc4 832 mutex_lock(&inode->i_mutex);
ad396024
EC
833
834 if (inode->i_pipe) {
835 ret = 0;
836 inode->i_pipe->readers++;
837 }
838
9aeedfc4 839 mutex_unlock(&inode->i_mutex);
1da177e4 840
ad396024 841 return ret;
1da177e4
LT
842}
843
844static int
845pipe_write_open(struct inode *inode, struct file *filp)
846{
ad396024
EC
847 int ret = -ENOENT;
848
9aeedfc4 849 mutex_lock(&inode->i_mutex);
ad396024
EC
850
851 if (inode->i_pipe) {
852 ret = 0;
853 inode->i_pipe->writers++;
854 }
855
9aeedfc4 856 mutex_unlock(&inode->i_mutex);
1da177e4 857
ad396024 858 return ret;
1da177e4
LT
859}
860
861static int
862pipe_rdwr_open(struct inode *inode, struct file *filp)
863{
ad396024
EC
864 int ret = -ENOENT;
865
a930d879
AV
866 if (!(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
867 return -EINVAL;
868
9aeedfc4 869 mutex_lock(&inode->i_mutex);
ad396024
EC
870
871 if (inode->i_pipe) {
872 ret = 0;
873 if (filp->f_mode & FMODE_READ)
874 inode->i_pipe->readers++;
875 if (filp->f_mode & FMODE_WRITE)
876 inode->i_pipe->writers++;
877 }
878
9aeedfc4 879 mutex_unlock(&inode->i_mutex);
1da177e4 880
ad396024 881 return ret;
1da177e4
LT
882}
883
884/*
885 * The file_operations structs are not static because they
886 * are also used in linux/fs/fifo.c to do operations on FIFOs.
d2d9648e
DV
887 *
888 * Pipes reuse fifos' file_operations structs.
1da177e4 889 */
d2d9648e 890const struct file_operations read_pipefifo_fops = {
1da177e4 891 .llseek = no_llseek,
ee0b3e67
BP
892 .read = do_sync_read,
893 .aio_read = pipe_read,
1da177e4
LT
894 .write = bad_pipe_w,
895 .poll = pipe_poll,
d59d0b1b 896 .unlocked_ioctl = pipe_ioctl,
1da177e4
LT
897 .open = pipe_read_open,
898 .release = pipe_read_release,
899 .fasync = pipe_read_fasync,
900};
901
d2d9648e 902const struct file_operations write_pipefifo_fops = {
1da177e4
LT
903 .llseek = no_llseek,
904 .read = bad_pipe_r,
ee0b3e67
BP
905 .write = do_sync_write,
906 .aio_write = pipe_write,
1da177e4 907 .poll = pipe_poll,
d59d0b1b 908 .unlocked_ioctl = pipe_ioctl,
1da177e4
LT
909 .open = pipe_write_open,
910 .release = pipe_write_release,
911 .fasync = pipe_write_fasync,
912};
913
d2d9648e 914const struct file_operations rdwr_pipefifo_fops = {
1da177e4 915 .llseek = no_llseek,
ee0b3e67
BP
916 .read = do_sync_read,
917 .aio_read = pipe_read,
918 .write = do_sync_write,
919 .aio_write = pipe_write,
1da177e4 920 .poll = pipe_poll,
d59d0b1b 921 .unlocked_ioctl = pipe_ioctl,
1da177e4
LT
922 .open = pipe_rdwr_open,
923 .release = pipe_rdwr_release,
924 .fasync = pipe_rdwr_fasync,
925};
926
3a326a2c
IM
927struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
928{
923f4f23 929 struct pipe_inode_info *pipe;
3a326a2c 930
923f4f23
IM
931 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
932 if (pipe) {
35f3d14d
JA
933 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
934 if (pipe->bufs) {
935 init_waitqueue_head(&pipe->wait);
936 pipe->r_counter = pipe->w_counter = 1;
937 pipe->inode = inode;
938 pipe->buffers = PIPE_DEF_BUFFERS;
939 return pipe;
940 }
941 kfree(pipe);
3a326a2c
IM
942 }
943
35f3d14d 944 return NULL;
3a326a2c
IM
945}
946
923f4f23 947void __free_pipe_info(struct pipe_inode_info *pipe)
1da177e4
LT
948{
949 int i;
1da177e4 950
35f3d14d 951 for (i = 0; i < pipe->buffers; i++) {
923f4f23 952 struct pipe_buffer *buf = pipe->bufs + i;
1da177e4 953 if (buf->ops)
923f4f23 954 buf->ops->release(pipe, buf);
1da177e4 955 }
923f4f23
IM
956 if (pipe->tmp_page)
957 __free_page(pipe->tmp_page);
35f3d14d 958 kfree(pipe->bufs);
923f4f23 959 kfree(pipe);
1da177e4
LT
960}
961
b92ce558
JA
962void free_pipe_info(struct inode *inode)
963{
964 __free_pipe_info(inode->i_pipe);
965 inode->i_pipe = NULL;
966}
967
fa3536cc 968static struct vfsmount *pipe_mnt __read_mostly;
341b446b 969
c23fbb6b
ED
970/*
971 * pipefs_dname() is called from d_path().
972 */
973static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
974{
975 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
976 dentry->d_inode->i_ino);
977}
978
3ba13d17 979static const struct dentry_operations pipefs_dentry_operations = {
c23fbb6b 980 .d_dname = pipefs_dname,
1da177e4
LT
981};
982
983static struct inode * get_pipe_inode(void)
984{
a209dfc7 985 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
923f4f23 986 struct pipe_inode_info *pipe;
1da177e4
LT
987
988 if (!inode)
989 goto fail_inode;
990
85fe4025
CH
991 inode->i_ino = get_next_ino();
992
923f4f23
IM
993 pipe = alloc_pipe_info(inode);
994 if (!pipe)
1da177e4 995 goto fail_iput;
923f4f23 996 inode->i_pipe = pipe;
3a326a2c 997
923f4f23 998 pipe->readers = pipe->writers = 1;
d2d9648e 999 inode->i_fop = &rdwr_pipefifo_fops;
1da177e4
LT
1000
1001 /*
1002 * Mark the inode dirty from the very beginning,
1003 * that way it will never be moved to the dirty
1004 * list because "mark_inode_dirty()" will think
1005 * that it already _is_ on the dirty list.
1006 */
1007 inode->i_state = I_DIRTY;
1008 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
da9592ed
DH
1009 inode->i_uid = current_fsuid();
1010 inode->i_gid = current_fsgid();
1da177e4 1011 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
923f4f23 1012
1da177e4
LT
1013 return inode;
1014
1015fail_iput:
1016 iput(inode);
341b446b 1017
1da177e4
LT
1018fail_inode:
1019 return NULL;
1020}
1021
e4fad8e5 1022int create_pipe_files(struct file **res, int flags)
1da177e4 1023{
d6cbd281 1024 int err;
e4fad8e5 1025 struct inode *inode = get_pipe_inode();
d6cbd281 1026 struct file *f;
2c48b9c4 1027 struct path path;
e4fad8e5 1028 static struct qstr name = { .name = "" };
1da177e4 1029
1da177e4 1030 if (!inode)
e4fad8e5 1031 return -ENFILE;
1da177e4 1032
d6cbd281 1033 err = -ENOMEM;
4b936885 1034 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
2c48b9c4 1035 if (!path.dentry)
d6cbd281 1036 goto err_inode;
2c48b9c4 1037 path.mnt = mntget(pipe_mnt);
341b446b 1038
2c48b9c4 1039 d_instantiate(path.dentry, inode);
430e285e
DH
1040
1041 err = -ENFILE;
2c48b9c4 1042 f = alloc_file(&path, FMODE_WRITE, &write_pipefifo_fops);
39b65252 1043 if (IS_ERR(f))
430e285e 1044 goto err_dentry;
341b446b 1045
9883035a 1046 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
d6cbd281 1047
e4fad8e5 1048 res[0] = alloc_file(&path, FMODE_READ, &read_pipefifo_fops);
39b65252 1049 if (IS_ERR(res[0]))
e4fad8e5
AV
1050 goto err_file;
1051
1052 path_get(&path);
1053 res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
1054 res[1] = f;
1055 return 0;
1da177e4 1056
e4fad8e5
AV
1057err_file:
1058 put_filp(f);
1059err_dentry:
ed152437 1060 free_pipe_info(inode);
2c48b9c4 1061 path_put(&path);
e4fad8e5 1062 return err;
ed152437 1063
e4fad8e5 1064err_inode:
1da177e4
LT
1065 free_pipe_info(inode);
1066 iput(inode);
e4fad8e5 1067 return err;
d6cbd281
AK
1068}
1069
5b249b1b 1070static int __do_pipe_flags(int *fd, struct file **files, int flags)
d6cbd281 1071{
d6cbd281
AK
1072 int error;
1073 int fdw, fdr;
1074
9883035a 1075 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
ed8cae8b
UD
1076 return -EINVAL;
1077
e4fad8e5
AV
1078 error = create_pipe_files(files, flags);
1079 if (error)
1080 return error;
d6cbd281 1081
ed8cae8b 1082 error = get_unused_fd_flags(flags);
d6cbd281
AK
1083 if (error < 0)
1084 goto err_read_pipe;
1085 fdr = error;
1086
ed8cae8b 1087 error = get_unused_fd_flags(flags);
d6cbd281
AK
1088 if (error < 0)
1089 goto err_fdr;
1090 fdw = error;
1091
157cf649 1092 audit_fd_pair(fdr, fdw);
d6cbd281
AK
1093 fd[0] = fdr;
1094 fd[1] = fdw;
d6cbd281
AK
1095 return 0;
1096
1097 err_fdr:
1098 put_unused_fd(fdr);
1099 err_read_pipe:
e4fad8e5
AV
1100 fput(files[0]);
1101 fput(files[1]);
d6cbd281 1102 return error;
1da177e4
LT
1103}
1104
5b249b1b
AV
1105int do_pipe_flags(int *fd, int flags)
1106{
1107 struct file *files[2];
1108 int error = __do_pipe_flags(fd, files, flags);
1109 if (!error) {
1110 fd_install(fd[0], files[0]);
1111 fd_install(fd[1], files[1]);
1112 }
1113 return error;
1114}
1115
d35c7b0e
UD
1116/*
1117 * sys_pipe() is the normal C calling standard for creating
1118 * a pipe. It's not the way Unix traditionally does this, though.
1119 */
d4e82042 1120SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
d35c7b0e 1121{
5b249b1b 1122 struct file *files[2];
d35c7b0e
UD
1123 int fd[2];
1124 int error;
1125
5b249b1b 1126 error = __do_pipe_flags(fd, files, flags);
d35c7b0e 1127 if (!error) {
5b249b1b
AV
1128 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
1129 fput(files[0]);
1130 fput(files[1]);
1131 put_unused_fd(fd[0]);
1132 put_unused_fd(fd[1]);
d35c7b0e 1133 error = -EFAULT;
5b249b1b
AV
1134 } else {
1135 fd_install(fd[0], files[0]);
1136 fd_install(fd[1], files[1]);
ba719bae 1137 }
d35c7b0e
UD
1138 }
1139 return error;
1140}
1141
2b664219 1142SYSCALL_DEFINE1(pipe, int __user *, fildes)
ed8cae8b
UD
1143{
1144 return sys_pipe2(fildes, 0);
1145}
1146
35f3d14d
JA
1147/*
1148 * Allocate a new array of pipe buffers and copy the info over. Returns the
1149 * pipe size if successful, or return -ERROR on error.
1150 */
b9598db3 1151static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
35f3d14d
JA
1152{
1153 struct pipe_buffer *bufs;
1154
35f3d14d
JA
1155 /*
1156 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1157 * expect a lot of shrink+grow operations, just free and allocate
1158 * again like we would do for growing. If the pipe currently
1159 * contains more buffers than arg, then return busy.
1160 */
b9598db3 1161 if (nr_pages < pipe->nrbufs)
35f3d14d
JA
1162 return -EBUSY;
1163
2ccd4f4d 1164 bufs = kcalloc(nr_pages, sizeof(*bufs), GFP_KERNEL | __GFP_NOWARN);
35f3d14d
JA
1165 if (unlikely(!bufs))
1166 return -ENOMEM;
1167
1168 /*
1169 * The pipe array wraps around, so just start the new one at zero
1170 * and adjust the indexes.
1171 */
1172 if (pipe->nrbufs) {
1d862f41
MS
1173 unsigned int tail;
1174 unsigned int head;
35f3d14d 1175
1d862f41
MS
1176 tail = pipe->curbuf + pipe->nrbufs;
1177 if (tail < pipe->buffers)
1178 tail = 0;
1179 else
1180 tail &= (pipe->buffers - 1);
1181
1182 head = pipe->nrbufs - tail;
35f3d14d
JA
1183 if (head)
1184 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1185 if (tail)
1d862f41 1186 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
35f3d14d
JA
1187 }
1188
1189 pipe->curbuf = 0;
1190 kfree(pipe->bufs);
1191 pipe->bufs = bufs;
b9598db3
JA
1192 pipe->buffers = nr_pages;
1193 return nr_pages * PAGE_SIZE;
35f3d14d
JA
1194}
1195
ff9da691
JA
1196/*
1197 * Currently we rely on the pipe array holding a power-of-2 number
1198 * of pages.
1199 */
1200static inline unsigned int round_pipe_size(unsigned int size)
1201{
1202 unsigned long nr_pages;
1203
1204 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1205 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1206}
1207
1208/*
1209 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1210 * will return an error.
1211 */
1212int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1213 size_t *lenp, loff_t *ppos)
1214{
1215 int ret;
1216
1217 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1218 if (ret < 0 || !write)
1219 return ret;
1220
1221 pipe_max_size = round_pipe_size(pipe_max_size);
1222 return ret;
1223}
1224
72083646
LT
1225/*
1226 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1227 * location, so checking ->i_pipe is not enough to verify that this is a
1228 * pipe.
1229 */
1230struct pipe_inode_info *get_pipe_info(struct file *file)
1231{
496ad9aa 1232 struct inode *i = file_inode(file);
72083646
LT
1233
1234 return S_ISFIFO(i->i_mode) ? i->i_pipe : NULL;
1235}
1236
35f3d14d
JA
1237long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1238{
1239 struct pipe_inode_info *pipe;
1240 long ret;
1241
c66fb347 1242 pipe = get_pipe_info(file);
35f3d14d
JA
1243 if (!pipe)
1244 return -EBADF;
1245
1246 mutex_lock(&pipe->inode->i_mutex);
1247
1248 switch (cmd) {
b9598db3 1249 case F_SETPIPE_SZ: {
ff9da691 1250 unsigned int size, nr_pages;
b9598db3 1251
ff9da691
JA
1252 size = round_pipe_size(arg);
1253 nr_pages = size >> PAGE_SHIFT;
b9598db3 1254
6db40cf0
MS
1255 ret = -EINVAL;
1256 if (!nr_pages)
1257 goto out;
1258
ff9da691 1259 if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
b4ca7615 1260 ret = -EPERM;
cc967be5 1261 goto out;
cc967be5 1262 }
ff9da691 1263 ret = pipe_set_size(pipe, nr_pages);
35f3d14d 1264 break;
b9598db3 1265 }
35f3d14d 1266 case F_GETPIPE_SZ:
b9598db3 1267 ret = pipe->buffers * PAGE_SIZE;
35f3d14d
JA
1268 break;
1269 default:
1270 ret = -EINVAL;
1271 break;
1272 }
1273
cc967be5 1274out:
35f3d14d
JA
1275 mutex_unlock(&pipe->inode->i_mutex);
1276 return ret;
1277}
1278
ff0c7d15
NP
1279static const struct super_operations pipefs_ops = {
1280 .destroy_inode = free_inode_nonrcu,
d70ef97b 1281 .statfs = simple_statfs,
ff0c7d15
NP
1282};
1283
1da177e4
LT
1284/*
1285 * pipefs should _never_ be mounted by userland - too much of security hassle,
1286 * no real gain from having the whole whorehouse mounted. So we don't need
1287 * any operations on the root directory. However, we need a non-trivial
1288 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1289 */
51139ada
AV
1290static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1291 int flags, const char *dev_name, void *data)
1da177e4 1292{
c74a1cbb
AV
1293 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1294 &pipefs_dentry_operations, PIPEFS_MAGIC);
1da177e4
LT
1295}
1296
1297static struct file_system_type pipe_fs_type = {
1298 .name = "pipefs",
51139ada 1299 .mount = pipefs_mount,
1da177e4
LT
1300 .kill_sb = kill_anon_super,
1301};
1302
1303static int __init init_pipe_fs(void)
1304{
1305 int err = register_filesystem(&pipe_fs_type);
341b446b 1306
1da177e4
LT
1307 if (!err) {
1308 pipe_mnt = kern_mount(&pipe_fs_type);
1309 if (IS_ERR(pipe_mnt)) {
1310 err = PTR_ERR(pipe_mnt);
1311 unregister_filesystem(&pipe_fs_type);
1312 }
1313 }
1314 return err;
1315}
1316
1da177e4 1317fs_initcall(init_pipe_fs);