blob: 49c1065256fd10d9d5fdca3cf449b1e56bd58a0a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Jens Axboe35f3d142010-05-20 10:43:18 +020014#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/mount.h>
Muthu Kumarb502bd12012-03-23 15:01:50 -070016#include <linux/magic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/pipe_fs_i.h>
18#include <linux/uio.h>
19#include <linux/highmem.h>
Jens Axboe5274f052006-03-30 15:15:30 +020020#include <linux/pagemap.h>
Al Virodb349502007-02-07 01:48:00 -050021#include <linux/audit.h>
Ulrich Drepperba719ba2008-05-06 20:42:38 -070022#include <linux/syscalls.h>
Jens Axboeb492e952010-05-19 21:03:16 +020023#include <linux/fcntl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
25#include <asm/uaccess.h>
26#include <asm/ioctls.h>
27
28/*
Jens Axboeb492e952010-05-19 21:03:16 +020029 * The max size that a non-root user is allowed to grow the pipe. Can
Jens Axboeff9da692010-06-03 14:54:39 +020030 * be set by root in /proc/sys/fs/pipe-max-size
Jens Axboeb492e952010-05-19 21:03:16 +020031 */
Jens Axboeff9da692010-06-03 14:54:39 +020032unsigned int pipe_max_size = 1048576;
33
34/*
35 * Minimum pipe size, as required by POSIX
36 */
37unsigned int pipe_min_size = PAGE_SIZE;
Jens Axboeb492e952010-05-19 21:03:16 +020038
39/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * 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
Miklos Szeredi61e0d472009-04-14 19:48:41 +020054static 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 {
Peter Zijlstra023d43c2009-07-21 10:09:23 +020085 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
86 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020087 }
88}
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/* Drop the inode semaphore and wait for a pipe event, atomically */
Ingo Molnar3a326a22006-04-10 15:18:35 +020091void pipe_wait(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 DEFINE_WAIT(wait);
94
Ingo Molnard79fc0f2005-09-10 00:26:12 -070095 /*
96 * Pipes are system-local resources, so sleeping on them
97 * is considered a noninteractive wait:
98 */
Mike Galbraithaf927232007-10-15 17:00:13 +020099 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200100 pipe_unlock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 schedule();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200102 finish_wait(&pipe->wait, &wait);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200103 pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104}
105
Arjan van de Ven858119e2006-01-14 13:20:43 -0800106static int
Jens Axboef6762b72006-05-01 20:02:05 +0200107pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
108 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
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
Jens Axboef6762b72006-05-01 20:02:05 +0200117 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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 to += copy;
125 len -= copy;
126 iov->iov_base += copy;
127 iov->iov_len -= copy;
128 }
129 return 0;
130}
131
Arjan van de Ven858119e2006-01-14 13:20:43 -0800132static int
Jens Axboef6762b72006-05-01 20:02:05 +0200133pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
134 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
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
Jens Axboef6762b72006-05-01 20:02:05 +0200143 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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 from += copy;
151 len -= copy;
152 iov->iov_base += copy;
153 iov->iov_len -= copy;
154 }
155 return 0;
156}
157
Jens Axboef6762b72006-05-01 20:02:05 +0200158/*
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
Ingo Molnar341b4462006-04-11 13:57:45 +0200199static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
200 struct pipe_buffer *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201{
202 struct page *page = buf->page;
203
Jens Axboe5274f052006-03-30 15:15:30 +0200204 /*
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
Ingo Molnar341b4462006-04-11 13:57:45 +0200207 * allocation cache. (Otherwise just release our reference to it)
Jens Axboe5274f052006-03-30 15:15:30 +0200208 */
Ingo Molnar341b4462006-04-11 13:57:45 +0200209 if (page_count(page) == 1 && !pipe->tmp_page)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200210 pipe->tmp_page = page;
Ingo Molnar341b4462006-04-11 13:57:45 +0200211 else
212 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Jens Axboe08457182007-06-12 20:51:32 +0200215/**
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
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800223 * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
Jens Axboe08457182007-06-12 20:51:32 +0200224 * and the caller has to be careful not to fault before calling
225 * the unmap function.
226 *
227 * Note that this function occupies KM_USER0 if @atomic != 0.
228 */
Jens Axboef84d7512006-05-01 19:59:03 +0200229void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200230 struct pipe_buffer *buf, int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
Jens Axboef6762b72006-05-01 20:02:05 +0200232 if (atomic) {
233 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
Cong Wange8e3c3d2011-11-25 23:14:27 +0800234 return kmap_atomic(buf->page);
Jens Axboef6762b72006-05-01 20:02:05 +0200235 }
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 return kmap(buf->page);
238}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200239EXPORT_SYMBOL(generic_pipe_buf_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Jens Axboe08457182007-06-12 20:51:32 +0200241/**
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 */
Jens Axboef84d7512006-05-01 19:59:03 +0200250void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200251 struct pipe_buffer *buf, void *map_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
Jens Axboef6762b72006-05-01 20:02:05 +0200253 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
254 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
Cong Wange8e3c3d2011-11-25 23:14:27 +0800255 kunmap_atomic(map_data);
Jens Axboef6762b72006-05-01 20:02:05 +0200256 } else
257 kunmap(buf->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200259EXPORT_SYMBOL(generic_pipe_buf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Jens Axboe08457182007-06-12 20:51:32 +0200261/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800262 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200263 * @pipe: the pipe that the buffer belongs to
264 * @buf: the buffer to attempt to steal
265 *
266 * Description:
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800267 * This function attempts to steal the &struct page attached to
Jens Axboe08457182007-06-12 20:51:32 +0200268 * @buf. If successful, this function returns 0 and returns with
269 * the page locked. The caller may then reuse the page for whatever
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800270 * he wishes; the typical use is insertion into a different file
Jens Axboe08457182007-06-12 20:51:32 +0200271 * page cache.
272 */
Jens Axboe330ab712006-05-02 15:29:57 +0200273int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
274 struct pipe_buffer *buf)
Jens Axboe5abc97a2006-03-30 15:16:46 +0200275{
Jens Axboe46e678c2006-04-30 16:36:32 +0200276 struct page *page = buf->page;
277
Jens Axboe08457182007-06-12 20:51:32 +0200278 /*
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 */
Jens Axboe46e678c2006-04-30 16:36:32 +0200283 if (page_count(page) == 1) {
Jens Axboe46e678c2006-04-30 16:36:32 +0200284 lock_page(page);
285 return 0;
286 }
287
288 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200289}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200290EXPORT_SYMBOL(generic_pipe_buf_steal);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200291
Jens Axboe08457182007-06-12 20:51:32 +0200292/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800293 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200294 * @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)
Jens Axboe70524492006-04-11 15:51:17 +0200303{
304 page_cache_get(buf->page);
305}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200306EXPORT_SYMBOL(generic_pipe_buf_get);
Jens Axboe70524492006-04-11 15:51:17 +0200307
Jens Axboe08457182007-06-12 20:51:32 +0200308/**
309 * generic_pipe_buf_confirm - verify contents of the pipe buffer
Randy Dunlap79685b82007-07-27 08:08:51 +0200310 * @info: the pipe that the buffer belongs to
Jens Axboe08457182007-06-12 20:51:32 +0200311 * @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 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200317int generic_pipe_buf_confirm(struct pipe_inode_info *info,
318 struct pipe_buffer *buf)
Jens Axboef84d7512006-05-01 19:59:03 +0200319{
320 return 0;
321}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200322EXPORT_SYMBOL(generic_pipe_buf_confirm);
Jens Axboef84d7512006-05-01 19:59:03 +0200323
Miklos Szeredi68181732009-05-07 15:37:36 +0200324/**
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}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200337EXPORT_SYMBOL(generic_pipe_buf_release);
Miklos Szeredi68181732009-05-07 15:37:36 +0200338
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800339static const struct pipe_buf_operations anon_pipe_buf_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 .can_merge = 1,
Jens Axboef84d7512006-05-01 19:59:03 +0200341 .map = generic_pipe_buf_map,
342 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200343 .confirm = generic_pipe_buf_confirm,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 .release = anon_pipe_buf_release,
Jens Axboe330ab712006-05-02 15:29:57 +0200345 .steal = generic_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200346 .get = generic_pipe_buf_get,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347};
348
Linus Torvalds98830352012-04-29 13:12:42 -0700349static 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700360pipe_read(struct kiocb *iocb, const struct iovec *_iov,
361 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700363 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800364 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200365 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 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;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200378 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200379 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 for (;;) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200381 int bufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (bufs) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200383 int curbuf = pipe->curbuf;
384 struct pipe_buffer *buf = pipe->bufs + curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800385 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 void *addr;
387 size_t chars = buf->len;
Jens Axboef6762b72006-05-01 20:02:05 +0200388 int error, atomic;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390 if (chars > total_len)
391 chars = total_len;
392
Jens Axboecac36bb02007-06-14 13:10:48 +0200393 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200394 if (error) {
Jens Axboe5274f052006-03-30 15:15:30 +0200395 if (!ret)
Nicolas Kaisere5953cb2010-10-21 14:56:00 +0200396 ret = error;
Jens Axboe5274f052006-03-30 15:15:30 +0200397 break;
398 }
Jens Axboef84d7512006-05-01 19:59:03 +0200399
Jens Axboef6762b72006-05-01 20:02:05 +0200400 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200406 /*
407 * Just retry with the slow path if we failed.
408 */
409 if (atomic) {
410 atomic = 0;
411 goto redo;
412 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200413 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200414 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 break;
416 }
417 ret += chars;
418 buf->offset += chars;
419 buf->len -= chars;
Linus Torvalds98830352012-04-29 13:12:42 -0700420
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (!buf->len) {
428 buf->ops = NULL;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200429 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200430 curbuf = (curbuf + 1) & (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200431 pipe->curbuf = curbuf;
432 pipe->nrbufs = --bufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 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;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200441 if (!pipe->writers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 break;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200443 if (!pipe->waiting_writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 /* 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)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200457 if (!ret)
458 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 break;
460 }
461 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800462 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200463 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200465 pipe_wait(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200467 mutex_unlock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200468
469 /* Signal writers asynchronously that there is more room. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800471 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200472 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474 if (ret > 0)
475 file_accessed(filp);
476 return ret;
477}
478
Linus Torvalds98830352012-04-29 13:12:42 -0700479static inline int is_packetized(struct file *file)
480{
481 return (file->f_flags & O_DIRECT) != 0;
482}
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700485pipe_write(struct kiocb *iocb, const struct iovec *_iov,
486 unsigned long nr_segs, loff_t ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700488 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800489 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200490 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 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;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200504 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200505 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Ingo Molnar923f4f22006-04-11 13:53:33 +0200507 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 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 */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200515 if (pipe->nrbufs && chars != 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200516 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
Jens Axboe35f3d142010-05-20 10:43:18 +0200517 (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200518 struct pipe_buffer *buf = pipe->bufs + lastbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800519 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 int offset = buf->offset + buf->len;
Ingo Molnar341b4462006-04-11 13:57:45 +0200521
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
Jens Axboef6762b72006-05-01 20:02:05 +0200523 int error, atomic = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200524 void *addr;
Jens Axboe5274f052006-03-30 15:15:30 +0200525
Jens Axboecac36bb02007-06-14 13:10:48 +0200526 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200527 if (error)
Jens Axboe5274f052006-03-30 15:15:30 +0200528 goto out;
Jens Axboef84d7512006-05-01 19:59:03 +0200529
Jens Axboef6762b72006-05-01 20:02:05 +0200530 iov_fault_in_pages_read(iov, chars);
531redo1:
532 addr = ops->map(pipe, buf, atomic);
Jens Axboe5274f052006-03-30 15:15:30 +0200533 error = pipe_iov_copy_from_user(offset + addr, iov,
Jens Axboef6762b72006-05-01 20:02:05 +0200534 chars, atomic);
535 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 ret = error;
537 do_wakeup = 1;
Jens Axboef6762b72006-05-01 20:02:05 +0200538 if (error) {
539 if (atomic) {
540 atomic = 0;
541 goto redo1;
542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 goto out;
Jens Axboef6762b72006-05-01 20:02:05 +0200544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 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;
Ingo Molnar341b4462006-04-11 13:57:45 +0200555
Ingo Molnar923f4f22006-04-11 13:53:33 +0200556 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 send_sig(SIGPIPE, current, 0);
Ingo Molnar341b4462006-04-11 13:57:45 +0200558 if (!ret)
559 ret = -EPIPE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 break;
561 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200562 bufs = pipe->nrbufs;
Jens Axboe35f3d142010-05-20 10:43:18 +0200563 if (bufs < pipe->buffers) {
564 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200565 struct pipe_buffer *buf = pipe->bufs + newbuf;
566 struct page *page = pipe->tmp_page;
Jens Axboef6762b72006-05-01 20:02:05 +0200567 char *src;
568 int error, atomic = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 if (!page) {
571 page = alloc_page(GFP_HIGHUSER);
572 if (unlikely(!page)) {
573 ret = ret ? : -ENOMEM;
574 break;
575 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200576 pipe->tmp_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200578 /* Always wake up, even if the copy fails. Otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * 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
Jens Axboef6762b72006-05-01 20:02:05 +0200588 iov_fault_in_pages_read(iov, chars);
589redo2:
590 if (atomic)
Cong Wange8e3c3d2011-11-25 23:14:27 +0800591 src = kmap_atomic(page);
Jens Axboef6762b72006-05-01 20:02:05 +0200592 else
593 src = kmap(page);
594
595 error = pipe_iov_copy_from_user(src, iov, chars,
596 atomic);
597 if (atomic)
Cong Wange8e3c3d2011-11-25 23:14:27 +0800598 kunmap_atomic(src);
Jens Axboef6762b72006-05-01 20:02:05 +0200599 else
600 kunmap(page);
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200603 if (atomic) {
604 atomic = 0;
605 goto redo2;
606 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200607 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200608 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 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;
Linus Torvalds98830352012-04-29 13:12:42 -0700618 buf->flags = 0;
619 if (is_packetized(filp)) {
620 buf->ops = &packet_pipe_buf_ops;
621 buf->flags = PIPE_BUF_FLAG_PACKET;
622 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200623 pipe->nrbufs = ++bufs;
624 pipe->tmp_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
626 total_len -= chars;
627 if (!total_len)
628 break;
629 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200630 if (bufs < pipe->buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 continue;
632 if (filp->f_flags & O_NONBLOCK) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200633 if (!ret)
634 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 break;
636 }
637 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200638 if (!ret)
639 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 break;
641 }
642 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800643 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200644 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 do_wakeup = 0;
646 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200647 pipe->waiting_writers++;
648 pipe_wait(pipe);
649 pipe->waiting_writers--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 }
651out:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200652 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800654 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200655 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 }
Josef Bacikc3b2da32012-03-26 09:59:21 -0400657 if (ret > 0) {
658 int err = file_update_time(filp);
659 if (err)
660 ret = err;
661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return ret;
663}
664
665static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
667{
668 return -EBADF;
669}
670
671static ssize_t
Ingo Molnar341b4462006-04-11 13:57:45 +0200672bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
673 loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674{
675 return -EBADF;
676}
677
Andi Kleend59d0b12008-02-08 04:21:23 -0800678static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800680 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200681 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 int count, buf, nrbufs;
683
684 switch (cmd) {
685 case FIONREAD:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200686 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200687 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 count = 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200689 buf = pipe->curbuf;
690 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 while (--nrbufs >= 0) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200692 count += pipe->bufs[buf].len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200693 buf = (buf+1) & (pipe->buffers - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200695 mutex_unlock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200696
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 return put_user(count, (int __user *)arg);
698 default:
Will Deacon46ce3412012-05-25 11:39:13 +0100699 return -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
701}
702
703/* No kernel lock held - fine */
704static unsigned int
705pipe_poll(struct file *filp, poll_table *wait)
706{
707 unsigned int mask;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800708 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200709 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 int nrbufs;
711
Ingo Molnar923f4f22006-04-11 13:53:33 +0200712 poll_wait(filp, &pipe->wait, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 /* Reading only -- no need for acquiring the semaphore. */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200715 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 mask = 0;
717 if (filp->f_mode & FMODE_READ) {
718 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200719 if (!pipe->writers && filp->f_version != pipe->w_counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 mask |= POLLHUP;
721 }
722
723 if (filp->f_mode & FMODE_WRITE) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200724 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700725 /*
726 * Most Unices do not set POLLERR for FIFOs but on Linux they
727 * behave exactly like pipes for poll().
728 */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200729 if (!pipe->readers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 mask |= POLLERR;
731 }
732
733 return mask;
734}
735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736static int
737pipe_release(struct inode *inode, int decr, int decw)
738{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200739 struct pipe_inode_info *pipe;
740
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200741 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200742 pipe = inode->i_pipe;
743 pipe->readers -= decr;
744 pipe->writers -= decw;
Ingo Molnar341b4462006-04-11 13:57:45 +0200745
Ingo Molnar923f4f22006-04-11 13:53:33 +0200746 if (!pipe->readers && !pipe->writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 free_pipe_info(inode);
748 } else {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800749 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200750 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
751 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200753 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
755 return 0;
756}
757
758static int
759pipe_read_fasync(int fd, struct file *filp, int on)
760{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800761 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 int retval;
763
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200764 mutex_lock(&inode->i_mutex);
765 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
766 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700768 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769}
770
771
772static int
773pipe_write_fasync(int fd, struct file *filp, int on)
774{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800775 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 int retval;
777
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200778 mutex_lock(&inode->i_mutex);
779 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
780 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700782 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783}
784
785
786static int
787pipe_rdwr_fasync(int fd, struct file *filp, int on)
788{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800789 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar341b4462006-04-11 13:57:45 +0200790 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 int retval;
792
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200793 mutex_lock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200794 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700795 if (retval >= 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200796 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700797 if (retval < 0) /* this can happen only if on == T */
798 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
799 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200800 mutex_unlock(&inode->i_mutex);
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700801 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802}
803
804
805static int
806pipe_read_release(struct inode *inode, struct file *filp)
807{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 return pipe_release(inode, 1, 0);
809}
810
811static int
812pipe_write_release(struct inode *inode, struct file *filp)
813{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 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{
Earl Chewad396022009-10-19 15:55:41 -0700830 int ret = -ENOENT;
831
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200832 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700833
834 if (inode->i_pipe) {
835 ret = 0;
836 inode->i_pipe->readers++;
837 }
838
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200839 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Earl Chewad396022009-10-19 15:55:41 -0700841 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842}
843
844static int
845pipe_write_open(struct inode *inode, struct file *filp)
846{
Earl Chewad396022009-10-19 15:55:41 -0700847 int ret = -ENOENT;
848
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200849 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700850
851 if (inode->i_pipe) {
852 ret = 0;
853 inode->i_pipe->writers++;
854 }
855
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200856 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Earl Chewad396022009-10-19 15:55:41 -0700858 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859}
860
861static int
862pipe_rdwr_open(struct inode *inode, struct file *filp)
863{
Earl Chewad396022009-10-19 15:55:41 -0700864 int ret = -ENOENT;
865
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200866 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700867
868 if (inode->i_pipe) {
869 ret = 0;
870 if (filp->f_mode & FMODE_READ)
871 inode->i_pipe->readers++;
872 if (filp->f_mode & FMODE_WRITE)
873 inode->i_pipe->writers++;
874 }
875
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200876 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
Earl Chewad396022009-10-19 15:55:41 -0700878 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879}
880
881/*
882 * The file_operations structs are not static because they
883 * are also used in linux/fs/fifo.c to do operations on FIFOs.
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200884 *
885 * Pipes reuse fifos' file_operations structs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 */
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200887const struct file_operations read_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700889 .read = do_sync_read,
890 .aio_read = pipe_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 .write = bad_pipe_w,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700892 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800893 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 .open = pipe_read_open,
895 .release = pipe_read_release,
896 .fasync = pipe_read_fasync,
897};
898
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200899const struct file_operations write_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 .llseek = no_llseek,
901 .read = bad_pipe_r,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700902 .write = do_sync_write,
903 .aio_write = pipe_write,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700904 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800905 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 .open = pipe_write_open,
907 .release = pipe_write_release,
908 .fasync = pipe_write_fasync,
909};
910
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200911const struct file_operations rdwr_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700913 .read = do_sync_read,
914 .aio_read = pipe_read,
915 .write = do_sync_write,
916 .aio_write = pipe_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800918 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 .open = pipe_rdwr_open,
920 .release = pipe_rdwr_release,
921 .fasync = pipe_rdwr_fasync,
922};
923
Ingo Molnar3a326a22006-04-10 15:18:35 +0200924struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
925{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200926 struct pipe_inode_info *pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200927
Ingo Molnar923f4f22006-04-11 13:53:33 +0200928 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
929 if (pipe) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200930 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
931 if (pipe->bufs) {
932 init_waitqueue_head(&pipe->wait);
933 pipe->r_counter = pipe->w_counter = 1;
934 pipe->inode = inode;
935 pipe->buffers = PIPE_DEF_BUFFERS;
936 return pipe;
937 }
938 kfree(pipe);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200939 }
940
Jens Axboe35f3d142010-05-20 10:43:18 +0200941 return NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200942}
943
Ingo Molnar923f4f22006-04-11 13:53:33 +0200944void __free_pipe_info(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945{
946 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
Jens Axboe35f3d142010-05-20 10:43:18 +0200948 for (i = 0; i < pipe->buffers; i++) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200949 struct pipe_buffer *buf = pipe->bufs + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 if (buf->ops)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200951 buf->ops->release(pipe, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200953 if (pipe->tmp_page)
954 __free_page(pipe->tmp_page);
Jens Axboe35f3d142010-05-20 10:43:18 +0200955 kfree(pipe->bufs);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200956 kfree(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
Jens Axboeb92ce552006-04-11 13:52:07 +0200959void free_pipe_info(struct inode *inode)
960{
961 __free_pipe_info(inode->i_pipe);
962 inode->i_pipe = NULL;
963}
964
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800965static struct vfsmount *pipe_mnt __read_mostly;
Ingo Molnar341b4462006-04-11 13:57:45 +0200966
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700967/*
968 * pipefs_dname() is called from d_path().
969 */
970static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
971{
972 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
973 dentry->d_inode->i_ino);
974}
975
Al Viro3ba13d12009-02-20 06:02:22 +0000976static const struct dentry_operations pipefs_dentry_operations = {
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700977 .d_dname = pipefs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978};
979
980static struct inode * get_pipe_inode(void)
981{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200982 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200983 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
985 if (!inode)
986 goto fail_inode;
987
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400988 inode->i_ino = get_next_ino();
989
Ingo Molnar923f4f22006-04-11 13:53:33 +0200990 pipe = alloc_pipe_info(inode);
991 if (!pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 goto fail_iput;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200993 inode->i_pipe = pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200994
Ingo Molnar923f4f22006-04-11 13:53:33 +0200995 pipe->readers = pipe->writers = 1;
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200996 inode->i_fop = &rdwr_pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 /*
999 * Mark the inode dirty from the very beginning,
1000 * that way it will never be moved to the dirty
1001 * list because "mark_inode_dirty()" will think
1002 * that it already _is_ on the dirty list.
1003 */
1004 inode->i_state = I_DIRTY;
1005 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
David Howellsda9592e2008-11-14 10:39:05 +11001006 inode->i_uid = current_fsuid();
1007 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Ingo Molnar923f4f22006-04-11 13:53:33 +02001009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 return inode;
1011
1012fail_iput:
1013 iput(inode);
Ingo Molnar341b4462006-04-11 13:57:45 +02001014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015fail_inode:
1016 return NULL;
1017}
1018
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001019struct file *create_write_pipe(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
Andi Kleend6cbd282006-09-30 23:29:26 -07001021 int err;
1022 struct inode *inode;
1023 struct file *f;
Al Viro2c48b9c2009-08-09 00:52:35 +04001024 struct path path;
Eric Dumazetc23fbb62007-05-08 00:26:18 -07001025 struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
Andi Kleend6cbd282006-09-30 23:29:26 -07001027 err = -ENFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 inode = get_pipe_inode();
1029 if (!inode)
Dave Hansen430e2852008-02-15 14:37:26 -08001030 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
Andi Kleend6cbd282006-09-30 23:29:26 -07001032 err = -ENOMEM;
Nick Piggin4b936882011-01-07 17:50:07 +11001033 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
Al Viro2c48b9c2009-08-09 00:52:35 +04001034 if (!path.dentry)
Andi Kleend6cbd282006-09-30 23:29:26 -07001035 goto err_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04001036 path.mnt = mntget(pipe_mnt);
Ingo Molnar341b4462006-04-11 13:57:45 +02001037
Al Viro2c48b9c2009-08-09 00:52:35 +04001038 d_instantiate(path.dentry, inode);
Dave Hansen430e2852008-02-15 14:37:26 -08001039
1040 err = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04001041 f = alloc_file(&path, FMODE_WRITE, &write_pipefifo_fops);
Dave Hansen430e2852008-02-15 14:37:26 -08001042 if (!f)
1043 goto err_dentry;
Andi Kleend6cbd282006-09-30 23:29:26 -07001044 f->f_mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Linus Torvalds98830352012-04-29 13:12:42 -07001046 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
Andi Kleend6cbd282006-09-30 23:29:26 -07001047 f->f_version = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Andi Kleend6cbd282006-09-30 23:29:26 -07001049 return f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
Dave Hansen430e2852008-02-15 14:37:26 -08001051 err_dentry:
Al Viroed152432008-04-22 19:51:27 -04001052 free_pipe_info(inode);
Al Viro2c48b9c2009-08-09 00:52:35 +04001053 path_put(&path);
Al Viroed152432008-04-22 19:51:27 -04001054 return ERR_PTR(err);
1055
Andi Kleend6cbd282006-09-30 23:29:26 -07001056 err_inode:
1057 free_pipe_info(inode);
1058 iput(inode);
Dave Hansen430e2852008-02-15 14:37:26 -08001059 err:
Andi Kleend6cbd282006-09-30 23:29:26 -07001060 return ERR_PTR(err);
1061}
1062
1063void free_write_pipe(struct file *f)
1064{
Al Viro5ccac88e2006-12-18 13:31:18 +00001065 free_pipe_info(f->f_dentry->d_inode);
Jan Blunckc8e7f442008-06-09 16:40:35 -07001066 path_put(&f->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001067 put_filp(f);
1068}
1069
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001070struct file *create_read_pipe(struct file *wrf, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001071{
Al Virod2314122009-08-09 01:01:37 +04001072 /* Grab pipe from the writer */
1073 struct file *f = alloc_file(&wrf->f_path, FMODE_READ,
1074 &read_pipefifo_fops);
Andi Kleend6cbd282006-09-30 23:29:26 -07001075 if (!f)
1076 return ERR_PTR(-ENFILE);
1077
Jan Blunckc8e7f442008-06-09 16:40:35 -07001078 path_get(&wrf->f_path);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001079 f->f_flags = O_RDONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -07001080
1081 return f;
1082}
1083
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001084int do_pipe_flags(int *fd, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001085{
1086 struct file *fw, *fr;
1087 int error;
1088 int fdw, fdr;
1089
Linus Torvalds98830352012-04-29 13:12:42 -07001090 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001091 return -EINVAL;
1092
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001093 fw = create_write_pipe(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001094 if (IS_ERR(fw))
1095 return PTR_ERR(fw);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001096 fr = create_read_pipe(fw, flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001097 error = PTR_ERR(fr);
1098 if (IS_ERR(fr))
1099 goto err_write_pipe;
1100
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001101 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001102 if (error < 0)
1103 goto err_read_pipe;
1104 fdr = error;
1105
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001106 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001107 if (error < 0)
1108 goto err_fdr;
1109 fdw = error;
1110
Al Viro157cf642008-12-14 04:57:47 -05001111 audit_fd_pair(fdr, fdw);
Andi Kleend6cbd282006-09-30 23:29:26 -07001112 fd_install(fdr, fr);
1113 fd_install(fdw, fw);
1114 fd[0] = fdr;
1115 fd[1] = fdw;
Ingo Molnar341b4462006-04-11 13:57:45 +02001116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 return 0;
1118
Andi Kleend6cbd282006-09-30 23:29:26 -07001119 err_fdr:
1120 put_unused_fd(fdr);
1121 err_read_pipe:
Jan Blunckc8e7f442008-06-09 16:40:35 -07001122 path_put(&fr->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001123 put_filp(fr);
1124 err_write_pipe:
1125 free_write_pipe(fw);
1126 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127}
1128
1129/*
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001130 * sys_pipe() is the normal C calling standard for creating
1131 * a pipe. It's not the way Unix traditionally does this, though.
1132 */
Heiko Carstensd4e82042009-01-14 14:14:34 +01001133SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001134{
1135 int fd[2];
1136 int error;
1137
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001138 error = do_pipe_flags(fd, flags);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001139 if (!error) {
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001140 if (copy_to_user(fildes, fd, sizeof(fd))) {
1141 sys_close(fd[0]);
1142 sys_close(fd[1]);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001143 error = -EFAULT;
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001144 }
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001145 }
1146 return error;
1147}
1148
Heiko Carstens2b664212009-01-14 14:14:35 +01001149SYSCALL_DEFINE1(pipe, int __user *, fildes)
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001150{
1151 return sys_pipe2(fildes, 0);
1152}
1153
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001154/*
Jens Axboe35f3d142010-05-20 10:43:18 +02001155 * Allocate a new array of pipe buffers and copy the info over. Returns the
1156 * pipe size if successful, or return -ERROR on error.
1157 */
Jens Axboeb9598db2010-05-24 19:34:43 +02001158static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +02001159{
1160 struct pipe_buffer *bufs;
1161
1162 /*
Jens Axboe35f3d142010-05-20 10:43:18 +02001163 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1164 * expect a lot of shrink+grow operations, just free and allocate
1165 * again like we would do for growing. If the pipe currently
1166 * contains more buffers than arg, then return busy.
1167 */
Jens Axboeb9598db2010-05-24 19:34:43 +02001168 if (nr_pages < pipe->nrbufs)
Jens Axboe35f3d142010-05-20 10:43:18 +02001169 return -EBUSY;
1170
Sasha Levin2ccd4f42012-01-12 17:17:40 -08001171 bufs = kcalloc(nr_pages, sizeof(*bufs), GFP_KERNEL | __GFP_NOWARN);
Jens Axboe35f3d142010-05-20 10:43:18 +02001172 if (unlikely(!bufs))
1173 return -ENOMEM;
1174
1175 /*
1176 * The pipe array wraps around, so just start the new one at zero
1177 * and adjust the indexes.
1178 */
1179 if (pipe->nrbufs) {
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001180 unsigned int tail;
1181 unsigned int head;
Jens Axboe35f3d142010-05-20 10:43:18 +02001182
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001183 tail = pipe->curbuf + pipe->nrbufs;
1184 if (tail < pipe->buffers)
1185 tail = 0;
1186 else
1187 tail &= (pipe->buffers - 1);
1188
1189 head = pipe->nrbufs - tail;
Jens Axboe35f3d142010-05-20 10:43:18 +02001190 if (head)
1191 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1192 if (tail)
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001193 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
Jens Axboe35f3d142010-05-20 10:43:18 +02001194 }
1195
1196 pipe->curbuf = 0;
1197 kfree(pipe->bufs);
1198 pipe->bufs = bufs;
Jens Axboeb9598db2010-05-24 19:34:43 +02001199 pipe->buffers = nr_pages;
1200 return nr_pages * PAGE_SIZE;
Jens Axboe35f3d142010-05-20 10:43:18 +02001201}
1202
Jens Axboeff9da692010-06-03 14:54:39 +02001203/*
1204 * Currently we rely on the pipe array holding a power-of-2 number
1205 * of pages.
1206 */
1207static inline unsigned int round_pipe_size(unsigned int size)
1208{
1209 unsigned long nr_pages;
1210
1211 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1212 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1213}
1214
1215/*
1216 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1217 * will return an error.
1218 */
1219int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1220 size_t *lenp, loff_t *ppos)
1221{
1222 int ret;
1223
1224 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1225 if (ret < 0 || !write)
1226 return ret;
1227
1228 pipe_max_size = round_pipe_size(pipe_max_size);
1229 return ret;
1230}
1231
Linus Torvalds72083642010-11-28 16:27:19 -08001232/*
1233 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1234 * location, so checking ->i_pipe is not enough to verify that this is a
1235 * pipe.
1236 */
1237struct pipe_inode_info *get_pipe_info(struct file *file)
1238{
1239 struct inode *i = file->f_path.dentry->d_inode;
1240
1241 return S_ISFIFO(i->i_mode) ? i->i_pipe : NULL;
1242}
1243
Jens Axboe35f3d142010-05-20 10:43:18 +02001244long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1245{
1246 struct pipe_inode_info *pipe;
1247 long ret;
1248
Linus Torvaldsc66fb342010-11-28 14:09:57 -08001249 pipe = get_pipe_info(file);
Jens Axboe35f3d142010-05-20 10:43:18 +02001250 if (!pipe)
1251 return -EBADF;
1252
1253 mutex_lock(&pipe->inode->i_mutex);
1254
1255 switch (cmd) {
Jens Axboeb9598db2010-05-24 19:34:43 +02001256 case F_SETPIPE_SZ: {
Jens Axboeff9da692010-06-03 14:54:39 +02001257 unsigned int size, nr_pages;
Jens Axboeb9598db2010-05-24 19:34:43 +02001258
Jens Axboeff9da692010-06-03 14:54:39 +02001259 size = round_pipe_size(arg);
1260 nr_pages = size >> PAGE_SHIFT;
Jens Axboeb9598db2010-05-24 19:34:43 +02001261
Miklos Szeredi6db40cf2010-06-09 09:27:57 +02001262 ret = -EINVAL;
1263 if (!nr_pages)
1264 goto out;
1265
Jens Axboeff9da692010-06-03 14:54:39 +02001266 if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
Jens Axboeb4ca7612010-06-01 12:42:12 +02001267 ret = -EPERM;
Julia Lawallcc967be2010-05-26 17:54:39 +02001268 goto out;
Julia Lawallcc967be2010-05-26 17:54:39 +02001269 }
Jens Axboeff9da692010-06-03 14:54:39 +02001270 ret = pipe_set_size(pipe, nr_pages);
Jens Axboe35f3d142010-05-20 10:43:18 +02001271 break;
Jens Axboeb9598db2010-05-24 19:34:43 +02001272 }
Jens Axboe35f3d142010-05-20 10:43:18 +02001273 case F_GETPIPE_SZ:
Jens Axboeb9598db2010-05-24 19:34:43 +02001274 ret = pipe->buffers * PAGE_SIZE;
Jens Axboe35f3d142010-05-20 10:43:18 +02001275 break;
1276 default:
1277 ret = -EINVAL;
1278 break;
1279 }
1280
Julia Lawallcc967be2010-05-26 17:54:39 +02001281out:
Jens Axboe35f3d142010-05-20 10:43:18 +02001282 mutex_unlock(&pipe->inode->i_mutex);
1283 return ret;
1284}
1285
Nick Pigginff0c7d12011-01-07 17:49:50 +11001286static const struct super_operations pipefs_ops = {
1287 .destroy_inode = free_inode_nonrcu,
Pavel Emelyanovd70ef972011-10-31 17:10:04 -07001288 .statfs = simple_statfs,
Nick Pigginff0c7d12011-01-07 17:49:50 +11001289};
1290
Jens Axboe35f3d142010-05-20 10:43:18 +02001291/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 * pipefs should _never_ be mounted by userland - too much of security hassle,
1293 * no real gain from having the whole whorehouse mounted. So we don't need
1294 * any operations on the root directory. However, we need a non-trivial
1295 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1296 */
Al Viro51139ad2010-07-25 23:47:46 +04001297static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1298 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299{
Al Viroc74a1cb2011-01-12 16:59:34 -05001300 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1301 &pipefs_dentry_operations, PIPEFS_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
1304static struct file_system_type pipe_fs_type = {
1305 .name = "pipefs",
Al Viro51139ad2010-07-25 23:47:46 +04001306 .mount = pipefs_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 .kill_sb = kill_anon_super,
1308};
1309
1310static int __init init_pipe_fs(void)
1311{
1312 int err = register_filesystem(&pipe_fs_type);
Ingo Molnar341b4462006-04-11 13:57:45 +02001313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (!err) {
1315 pipe_mnt = kern_mount(&pipe_fs_type);
1316 if (IS_ERR(pipe_mnt)) {
1317 err = PTR_ERR(pipe_mnt);
1318 unregister_filesystem(&pipe_fs_type);
1319 }
1320 }
1321 return err;
1322}
1323
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324fs_initcall(init_pipe_fs);