blob: 78fd0d0788dbc465c3d7a6ba674bb64a6ebea25d [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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070024#include <linux/aio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26#include <asm/uaccess.h>
27#include <asm/ioctls.h>
28
Al Viro599a0ac2013-03-12 09:58:10 -040029#include "internal.h"
30
Linus Torvalds1da177e2005-04-16 15:20:36 -070031/*
Jens Axboeb492e952010-05-19 21:03:16 +020032 * The max size that a non-root user is allowed to grow the pipe. Can
Jens Axboeff9da692010-06-03 14:54:39 +020033 * be set by root in /proc/sys/fs/pipe-max-size
Jens Axboeb492e952010-05-19 21:03:16 +020034 */
Jens Axboeff9da692010-06-03 14:54:39 +020035unsigned int pipe_max_size = 1048576;
36
37/*
38 * Minimum pipe size, as required by POSIX
39 */
40unsigned int pipe_min_size = PAGE_SIZE;
Jens Axboeb492e952010-05-19 21:03:16 +020041
42/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 * We use a start+len construction, which provides full use of the
44 * allocated memory.
45 * -- Florian Coosmann (FGC)
46 *
47 * Reads with count = 0 should always return 0.
48 * -- Julian Bradfield 1999-06-07.
49 *
50 * FIFOs and Pipes now generate SIGIO for both readers and writers.
51 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
52 *
53 * pipe_read & write cleanup
54 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
55 */
56
Miklos Szeredi61e0d472009-04-14 19:48:41 +020057static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
58{
Al Viro6447a3c2013-03-21 11:01:38 -040059 if (pipe->files)
Al Viro72b0d9a2013-03-21 02:32:24 -040060 mutex_lock_nested(&pipe->mutex, subclass);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020061}
62
63void pipe_lock(struct pipe_inode_info *pipe)
64{
65 /*
66 * pipe_lock() nests non-pipe inode locks (for writing to a file)
67 */
68 pipe_lock_nested(pipe, I_MUTEX_PARENT);
69}
70EXPORT_SYMBOL(pipe_lock);
71
72void pipe_unlock(struct pipe_inode_info *pipe)
73{
Al Viro6447a3c2013-03-21 11:01:38 -040074 if (pipe->files)
Al Viro72b0d9a2013-03-21 02:32:24 -040075 mutex_unlock(&pipe->mutex);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020076}
77EXPORT_SYMBOL(pipe_unlock);
78
Al Viroebec73f2013-03-21 12:24:01 -040079static inline void __pipe_lock(struct pipe_inode_info *pipe)
80{
81 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
82}
83
84static inline void __pipe_unlock(struct pipe_inode_info *pipe)
85{
86 mutex_unlock(&pipe->mutex);
87}
88
Miklos Szeredi61e0d472009-04-14 19:48:41 +020089void pipe_double_lock(struct pipe_inode_info *pipe1,
90 struct pipe_inode_info *pipe2)
91{
92 BUG_ON(pipe1 == pipe2);
93
94 if (pipe1 < pipe2) {
95 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
96 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
97 } else {
Peter Zijlstra023d43c2009-07-21 10:09:23 +020098 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
99 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200100 }
101}
102
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103/* Drop the inode semaphore and wait for a pipe event, atomically */
Ingo Molnar3a326a22006-04-10 15:18:35 +0200104void pipe_wait(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105{
106 DEFINE_WAIT(wait);
107
Ingo Molnard79fc0f2005-09-10 00:26:12 -0700108 /*
109 * Pipes are system-local resources, so sleeping on them
110 * is considered a noninteractive wait:
111 */
Mike Galbraithaf927232007-10-15 17:00:13 +0200112 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200113 pipe_unlock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 schedule();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200115 finish_wait(&pipe->wait, &wait);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200116 pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117}
118
Arjan van de Ven858119e2006-01-14 13:20:43 -0800119static int
Jens Axboef6762b72006-05-01 20:02:05 +0200120pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
121 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
123 unsigned long copy;
124
125 while (len > 0) {
126 while (!iov->iov_len)
127 iov++;
128 copy = min_t(unsigned long, len, iov->iov_len);
129
Jens Axboef6762b72006-05-01 20:02:05 +0200130 if (atomic) {
131 if (__copy_from_user_inatomic(to, iov->iov_base, copy))
132 return -EFAULT;
133 } else {
134 if (copy_from_user(to, iov->iov_base, copy))
135 return -EFAULT;
136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 to += copy;
138 len -= copy;
139 iov->iov_base += copy;
140 iov->iov_len -= copy;
141 }
142 return 0;
143}
144
Arjan van de Ven858119e2006-01-14 13:20:43 -0800145static int
Jens Axboef6762b72006-05-01 20:02:05 +0200146pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
147 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 unsigned long copy;
150
151 while (len > 0) {
152 while (!iov->iov_len)
153 iov++;
154 copy = min_t(unsigned long, len, iov->iov_len);
155
Jens Axboef6762b72006-05-01 20:02:05 +0200156 if (atomic) {
157 if (__copy_to_user_inatomic(iov->iov_base, from, copy))
158 return -EFAULT;
159 } else {
160 if (copy_to_user(iov->iov_base, from, copy))
161 return -EFAULT;
162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 from += copy;
164 len -= copy;
165 iov->iov_base += copy;
166 iov->iov_len -= copy;
167 }
168 return 0;
169}
170
Jens Axboef6762b72006-05-01 20:02:05 +0200171/*
172 * Attempt to pre-fault in the user memory, so we can use atomic copies.
173 * Returns the number of bytes not faulted in.
174 */
175static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
176{
177 while (!iov->iov_len)
178 iov++;
179
180 while (len > 0) {
181 unsigned long this_len;
182
183 this_len = min_t(unsigned long, len, iov->iov_len);
184 if (fault_in_pages_writeable(iov->iov_base, this_len))
185 break;
186
187 len -= this_len;
188 iov++;
189 }
190
191 return len;
192}
193
194/*
195 * Pre-fault in the user memory, so we can use atomic copies.
196 */
197static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
198{
199 while (!iov->iov_len)
200 iov++;
201
202 while (len > 0) {
203 unsigned long this_len;
204
205 this_len = min_t(unsigned long, len, iov->iov_len);
206 fault_in_pages_readable(iov->iov_base, this_len);
207 len -= this_len;
208 iov++;
209 }
210}
211
Ingo Molnar341b4462006-04-11 13:57:45 +0200212static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
213 struct pipe_buffer *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 struct page *page = buf->page;
216
Jens Axboe5274f052006-03-30 15:15:30 +0200217 /*
218 * If nobody else uses this page, and we don't already have a
219 * temporary page, let's keep track of it as a one-deep
Ingo Molnar341b4462006-04-11 13:57:45 +0200220 * allocation cache. (Otherwise just release our reference to it)
Jens Axboe5274f052006-03-30 15:15:30 +0200221 */
Ingo Molnar341b4462006-04-11 13:57:45 +0200222 if (page_count(page) == 1 && !pipe->tmp_page)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200223 pipe->tmp_page = page;
Ingo Molnar341b4462006-04-11 13:57:45 +0200224 else
225 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
Jens Axboe08457182007-06-12 20:51:32 +0200228/**
229 * generic_pipe_buf_map - virtually map a pipe buffer
230 * @pipe: the pipe that the buffer belongs to
231 * @buf: the buffer that should be mapped
232 * @atomic: whether to use an atomic map
233 *
234 * Description:
235 * This function returns a kernel virtual address mapping for the
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800236 * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
Jens Axboe08457182007-06-12 20:51:32 +0200237 * and the caller has to be careful not to fault before calling
238 * the unmap function.
239 *
Cong Wang2164d332012-06-23 11:33:51 +0800240 * Note that this function calls kmap_atomic() if @atomic != 0.
Jens Axboe08457182007-06-12 20:51:32 +0200241 */
Jens Axboef84d7512006-05-01 19:59:03 +0200242void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200243 struct pipe_buffer *buf, int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
Jens Axboef6762b72006-05-01 20:02:05 +0200245 if (atomic) {
246 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
Cong Wange8e3c3d2011-11-25 23:14:27 +0800247 return kmap_atomic(buf->page);
Jens Axboef6762b72006-05-01 20:02:05 +0200248 }
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 return kmap(buf->page);
251}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200252EXPORT_SYMBOL(generic_pipe_buf_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Jens Axboe08457182007-06-12 20:51:32 +0200254/**
255 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
256 * @pipe: the pipe that the buffer belongs to
257 * @buf: the buffer that should be unmapped
258 * @map_data: the data that the mapping function returned
259 *
260 * Description:
261 * This function undoes the mapping that ->map() provided.
262 */
Jens Axboef84d7512006-05-01 19:59:03 +0200263void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200264 struct pipe_buffer *buf, void *map_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Jens Axboef6762b72006-05-01 20:02:05 +0200266 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
267 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
Cong Wange8e3c3d2011-11-25 23:14:27 +0800268 kunmap_atomic(map_data);
Jens Axboef6762b72006-05-01 20:02:05 +0200269 } else
270 kunmap(buf->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200272EXPORT_SYMBOL(generic_pipe_buf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Jens Axboe08457182007-06-12 20:51:32 +0200274/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800275 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200276 * @pipe: the pipe that the buffer belongs to
277 * @buf: the buffer to attempt to steal
278 *
279 * Description:
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800280 * This function attempts to steal the &struct page attached to
Jens Axboe08457182007-06-12 20:51:32 +0200281 * @buf. If successful, this function returns 0 and returns with
282 * the page locked. The caller may then reuse the page for whatever
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800283 * he wishes; the typical use is insertion into a different file
Jens Axboe08457182007-06-12 20:51:32 +0200284 * page cache.
285 */
Jens Axboe330ab712006-05-02 15:29:57 +0200286int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
287 struct pipe_buffer *buf)
Jens Axboe5abc97a2006-03-30 15:16:46 +0200288{
Jens Axboe46e678c2006-04-30 16:36:32 +0200289 struct page *page = buf->page;
290
Jens Axboe08457182007-06-12 20:51:32 +0200291 /*
292 * A reference of one is golden, that means that the owner of this
293 * page is the only one holding a reference to it. lock the page
294 * and return OK.
295 */
Jens Axboe46e678c2006-04-30 16:36:32 +0200296 if (page_count(page) == 1) {
Jens Axboe46e678c2006-04-30 16:36:32 +0200297 lock_page(page);
298 return 0;
299 }
300
301 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200302}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200303EXPORT_SYMBOL(generic_pipe_buf_steal);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200304
Jens Axboe08457182007-06-12 20:51:32 +0200305/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800306 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200307 * @pipe: the pipe that the buffer belongs to
308 * @buf: the buffer to get a reference to
309 *
310 * Description:
311 * This function grabs an extra reference to @buf. It's used in
312 * in the tee() system call, when we duplicate the buffers in one
313 * pipe into another.
314 */
315void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
Jens Axboe70524492006-04-11 15:51:17 +0200316{
317 page_cache_get(buf->page);
318}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200319EXPORT_SYMBOL(generic_pipe_buf_get);
Jens Axboe70524492006-04-11 15:51:17 +0200320
Jens Axboe08457182007-06-12 20:51:32 +0200321/**
322 * generic_pipe_buf_confirm - verify contents of the pipe buffer
Randy Dunlap79685b82007-07-27 08:08:51 +0200323 * @info: the pipe that the buffer belongs to
Jens Axboe08457182007-06-12 20:51:32 +0200324 * @buf: the buffer to confirm
325 *
326 * Description:
327 * This function does nothing, because the generic pipe code uses
328 * pages that are always good when inserted into the pipe.
329 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200330int generic_pipe_buf_confirm(struct pipe_inode_info *info,
331 struct pipe_buffer *buf)
Jens Axboef84d7512006-05-01 19:59:03 +0200332{
333 return 0;
334}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200335EXPORT_SYMBOL(generic_pipe_buf_confirm);
Jens Axboef84d7512006-05-01 19:59:03 +0200336
Miklos Szeredi68181732009-05-07 15:37:36 +0200337/**
338 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
339 * @pipe: the pipe that the buffer belongs to
340 * @buf: the buffer to put a reference to
341 *
342 * Description:
343 * This function releases a reference to @buf.
344 */
345void generic_pipe_buf_release(struct pipe_inode_info *pipe,
346 struct pipe_buffer *buf)
347{
348 page_cache_release(buf->page);
349}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200350EXPORT_SYMBOL(generic_pipe_buf_release);
Miklos Szeredi68181732009-05-07 15:37:36 +0200351
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800352static const struct pipe_buf_operations anon_pipe_buf_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 .can_merge = 1,
Jens Axboef84d7512006-05-01 19:59:03 +0200354 .map = generic_pipe_buf_map,
355 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200356 .confirm = generic_pipe_buf_confirm,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 .release = anon_pipe_buf_release,
Jens Axboe330ab712006-05-02 15:29:57 +0200358 .steal = generic_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200359 .get = generic_pipe_buf_get,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360};
361
Linus Torvalds98830352012-04-29 13:12:42 -0700362static const struct pipe_buf_operations packet_pipe_buf_ops = {
363 .can_merge = 0,
364 .map = generic_pipe_buf_map,
365 .unmap = generic_pipe_buf_unmap,
366 .confirm = generic_pipe_buf_confirm,
367 .release = anon_pipe_buf_release,
368 .steal = generic_pipe_buf_steal,
369 .get = generic_pipe_buf_get,
370};
371
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700373pipe_read(struct kiocb *iocb, const struct iovec *_iov,
374 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700376 struct file *filp = iocb->ki_filp;
Al Virode32ec42013-03-21 11:16:56 -0400377 struct pipe_inode_info *pipe = filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 int do_wakeup;
379 ssize_t ret;
380 struct iovec *iov = (struct iovec *)_iov;
381 size_t total_len;
382
383 total_len = iov_length(iov, nr_segs);
384 /* Null read succeeds. */
385 if (unlikely(total_len == 0))
386 return 0;
387
388 do_wakeup = 0;
389 ret = 0;
Al Viroebec73f2013-03-21 12:24:01 -0400390 __pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 for (;;) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200392 int bufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 if (bufs) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200394 int curbuf = pipe->curbuf;
395 struct pipe_buffer *buf = pipe->bufs + curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800396 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 void *addr;
398 size_t chars = buf->len;
Jens Axboef6762b72006-05-01 20:02:05 +0200399 int error, atomic;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
401 if (chars > total_len)
402 chars = total_len;
403
Jens Axboecac36bb02007-06-14 13:10:48 +0200404 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200405 if (error) {
Jens Axboe5274f052006-03-30 15:15:30 +0200406 if (!ret)
Nicolas Kaisere5953cb2010-10-21 14:56:00 +0200407 ret = error;
Jens Axboe5274f052006-03-30 15:15:30 +0200408 break;
409 }
Jens Axboef84d7512006-05-01 19:59:03 +0200410
Jens Axboef6762b72006-05-01 20:02:05 +0200411 atomic = !iov_fault_in_pages_write(iov, chars);
412redo:
413 addr = ops->map(pipe, buf, atomic);
414 error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
415 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200417 /*
418 * Just retry with the slow path if we failed.
419 */
420 if (atomic) {
421 atomic = 0;
422 goto redo;
423 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200424 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200425 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 break;
427 }
428 ret += chars;
429 buf->offset += chars;
430 buf->len -= chars;
Linus Torvalds98830352012-04-29 13:12:42 -0700431
432 /* Was it a packet buffer? Clean up and exit */
433 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
434 total_len = chars;
435 buf->len = 0;
436 }
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 if (!buf->len) {
439 buf->ops = NULL;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200440 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200441 curbuf = (curbuf + 1) & (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200442 pipe->curbuf = curbuf;
443 pipe->nrbufs = --bufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 do_wakeup = 1;
445 }
446 total_len -= chars;
447 if (!total_len)
448 break; /* common path: read succeeded */
449 }
450 if (bufs) /* More to do? */
451 continue;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200452 if (!pipe->writers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 break;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200454 if (!pipe->waiting_writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 /* syscall merging: Usually we must not sleep
456 * if O_NONBLOCK is set, or if we got some data.
457 * But if a writer sleeps in kernel space, then
458 * we can wait for that data without violating POSIX.
459 */
460 if (ret)
461 break;
462 if (filp->f_flags & O_NONBLOCK) {
463 ret = -EAGAIN;
464 break;
465 }
466 }
467 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200468 if (!ret)
469 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 break;
471 }
472 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800473 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200474 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200476 pipe_wait(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 }
Al Viroebec73f2013-03-21 12:24:01 -0400478 __pipe_unlock(pipe);
Ingo Molnar341b4462006-04-11 13:57:45 +0200479
480 /* Signal writers asynchronously that there is more room. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800482 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200483 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
485 if (ret > 0)
486 file_accessed(filp);
487 return ret;
488}
489
Linus Torvalds98830352012-04-29 13:12:42 -0700490static inline int is_packetized(struct file *file)
491{
492 return (file->f_flags & O_DIRECT) != 0;
493}
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700496pipe_write(struct kiocb *iocb, const struct iovec *_iov,
497 unsigned long nr_segs, loff_t ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700499 struct file *filp = iocb->ki_filp;
Al Virode32ec42013-03-21 11:16:56 -0400500 struct pipe_inode_info *pipe = filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 ssize_t ret;
502 int do_wakeup;
503 struct iovec *iov = (struct iovec *)_iov;
504 size_t total_len;
505 ssize_t chars;
506
507 total_len = iov_length(iov, nr_segs);
508 /* Null write succeeds. */
509 if (unlikely(total_len == 0))
510 return 0;
511
512 do_wakeup = 0;
513 ret = 0;
Al Viroebec73f2013-03-21 12:24:01 -0400514 __pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Ingo Molnar923f4f22006-04-11 13:53:33 +0200516 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 send_sig(SIGPIPE, current, 0);
518 ret = -EPIPE;
519 goto out;
520 }
521
522 /* We try to merge small writes */
523 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200524 if (pipe->nrbufs && chars != 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200525 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
Jens Axboe35f3d142010-05-20 10:43:18 +0200526 (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200527 struct pipe_buffer *buf = pipe->bufs + lastbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800528 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 int offset = buf->offset + buf->len;
Ingo Molnar341b4462006-04-11 13:57:45 +0200530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
Jens Axboef6762b72006-05-01 20:02:05 +0200532 int error, atomic = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200533 void *addr;
Jens Axboe5274f052006-03-30 15:15:30 +0200534
Jens Axboecac36bb02007-06-14 13:10:48 +0200535 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200536 if (error)
Jens Axboe5274f052006-03-30 15:15:30 +0200537 goto out;
Jens Axboef84d7512006-05-01 19:59:03 +0200538
Jens Axboef6762b72006-05-01 20:02:05 +0200539 iov_fault_in_pages_read(iov, chars);
540redo1:
541 addr = ops->map(pipe, buf, atomic);
Jens Axboe5274f052006-03-30 15:15:30 +0200542 error = pipe_iov_copy_from_user(offset + addr, iov,
Jens Axboef6762b72006-05-01 20:02:05 +0200543 chars, atomic);
544 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 ret = error;
546 do_wakeup = 1;
Jens Axboef6762b72006-05-01 20:02:05 +0200547 if (error) {
548 if (atomic) {
549 atomic = 0;
550 goto redo1;
551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 goto out;
Jens Axboef6762b72006-05-01 20:02:05 +0200553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 buf->len += chars;
555 total_len -= chars;
556 ret = chars;
557 if (!total_len)
558 goto out;
559 }
560 }
561
562 for (;;) {
563 int bufs;
Ingo Molnar341b4462006-04-11 13:57:45 +0200564
Ingo Molnar923f4f22006-04-11 13:53:33 +0200565 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 send_sig(SIGPIPE, current, 0);
Ingo Molnar341b4462006-04-11 13:57:45 +0200567 if (!ret)
568 ret = -EPIPE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 break;
570 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200571 bufs = pipe->nrbufs;
Jens Axboe35f3d142010-05-20 10:43:18 +0200572 if (bufs < pipe->buffers) {
573 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200574 struct pipe_buffer *buf = pipe->bufs + newbuf;
575 struct page *page = pipe->tmp_page;
Jens Axboef6762b72006-05-01 20:02:05 +0200576 char *src;
577 int error, atomic = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579 if (!page) {
580 page = alloc_page(GFP_HIGHUSER);
581 if (unlikely(!page)) {
582 ret = ret ? : -ENOMEM;
583 break;
584 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200585 pipe->tmp_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200587 /* Always wake up, even if the copy fails. Otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 * we lock up (O_NONBLOCK-)readers that sleep due to
589 * syscall merging.
590 * FIXME! Is this really true?
591 */
592 do_wakeup = 1;
593 chars = PAGE_SIZE;
594 if (chars > total_len)
595 chars = total_len;
596
Jens Axboef6762b72006-05-01 20:02:05 +0200597 iov_fault_in_pages_read(iov, chars);
598redo2:
599 if (atomic)
Cong Wange8e3c3d2011-11-25 23:14:27 +0800600 src = kmap_atomic(page);
Jens Axboef6762b72006-05-01 20:02:05 +0200601 else
602 src = kmap(page);
603
604 error = pipe_iov_copy_from_user(src, iov, chars,
605 atomic);
606 if (atomic)
Cong Wange8e3c3d2011-11-25 23:14:27 +0800607 kunmap_atomic(src);
Jens Axboef6762b72006-05-01 20:02:05 +0200608 else
609 kunmap(page);
610
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200612 if (atomic) {
613 atomic = 0;
614 goto redo2;
615 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200616 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200617 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 break;
619 }
620 ret += chars;
621
622 /* Insert it into the buffer array */
623 buf->page = page;
624 buf->ops = &anon_pipe_buf_ops;
625 buf->offset = 0;
626 buf->len = chars;
Linus Torvalds98830352012-04-29 13:12:42 -0700627 buf->flags = 0;
628 if (is_packetized(filp)) {
629 buf->ops = &packet_pipe_buf_ops;
630 buf->flags = PIPE_BUF_FLAG_PACKET;
631 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200632 pipe->nrbufs = ++bufs;
633 pipe->tmp_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635 total_len -= chars;
636 if (!total_len)
637 break;
638 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200639 if (bufs < pipe->buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 continue;
641 if (filp->f_flags & O_NONBLOCK) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200642 if (!ret)
643 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 break;
645 }
646 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200647 if (!ret)
648 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 break;
650 }
651 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800652 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200653 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 do_wakeup = 0;
655 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200656 pipe->waiting_writers++;
657 pipe_wait(pipe);
658 pipe->waiting_writers--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 }
660out:
Al Viroebec73f2013-03-21 12:24:01 -0400661 __pipe_unlock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 if (do_wakeup) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800663 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200664 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 }
Dmitry Monakhov7e775f42014-01-23 15:55:21 -0800666 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
Josef Bacikc3b2da32012-03-26 09:59:21 -0400667 int err = file_update_time(filp);
668 if (err)
669 ret = err;
Dmitry Monakhov7e775f42014-01-23 15:55:21 -0800670 sb_end_write(file_inode(filp)->i_sb);
Josef Bacikc3b2da32012-03-26 09:59:21 -0400671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 return ret;
673}
674
Andi Kleend59d0b12008-02-08 04:21:23 -0800675static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
Al Virode32ec42013-03-21 11:16:56 -0400677 struct pipe_inode_info *pipe = filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 int count, buf, nrbufs;
679
680 switch (cmd) {
681 case FIONREAD:
Al Viroebec73f2013-03-21 12:24:01 -0400682 __pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 count = 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200684 buf = pipe->curbuf;
685 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 while (--nrbufs >= 0) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200687 count += pipe->bufs[buf].len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200688 buf = (buf+1) & (pipe->buffers - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
Al Viroebec73f2013-03-21 12:24:01 -0400690 __pipe_unlock(pipe);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 return put_user(count, (int __user *)arg);
693 default:
Will Deacon46ce3412012-05-25 11:39:13 +0100694 return -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
696}
697
698/* No kernel lock held - fine */
699static unsigned int
700pipe_poll(struct file *filp, poll_table *wait)
701{
702 unsigned int mask;
Al Virode32ec42013-03-21 11:16:56 -0400703 struct pipe_inode_info *pipe = filp->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 int nrbufs;
705
Ingo Molnar923f4f22006-04-11 13:53:33 +0200706 poll_wait(filp, &pipe->wait, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
708 /* Reading only -- no need for acquiring the semaphore. */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200709 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 mask = 0;
711 if (filp->f_mode & FMODE_READ) {
712 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200713 if (!pipe->writers && filp->f_version != pipe->w_counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 mask |= POLLHUP;
715 }
716
717 if (filp->f_mode & FMODE_WRITE) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200718 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700719 /*
720 * Most Unices do not set POLLERR for FIFOs but on Linux they
721 * behave exactly like pipes for poll().
722 */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200723 if (!pipe->readers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 mask |= POLLERR;
725 }
726
727 return mask;
728}
729
Linus Torvaldsb0d8d222013-12-02 09:44:51 -0800730static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
731{
732 int kill = 0;
733
734 spin_lock(&inode->i_lock);
735 if (!--pipe->files) {
736 inode->i_pipe = NULL;
737 kill = 1;
738 }
739 spin_unlock(&inode->i_lock);
740
741 if (kill)
742 free_pipe_info(pipe);
743}
744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745static int
Al Viro599a0ac2013-03-12 09:58:10 -0400746pipe_release(struct inode *inode, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
Linus Torvaldsb0d8d222013-12-02 09:44:51 -0800748 struct pipe_inode_info *pipe = file->private_data;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200749
Al Viroebec73f2013-03-21 12:24:01 -0400750 __pipe_lock(pipe);
Al Viro599a0ac2013-03-12 09:58:10 -0400751 if (file->f_mode & FMODE_READ)
752 pipe->readers--;
753 if (file->f_mode & FMODE_WRITE)
754 pipe->writers--;
Ingo Molnar341b4462006-04-11 13:57:45 +0200755
Al Viroba5bb142013-03-21 02:21:19 -0400756 if (pipe->readers || pipe->writers) {
Linus Torvalds28e58ee2011-01-20 16:21:59 -0800757 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200758 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
759 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
Al Viroebec73f2013-03-21 12:24:01 -0400761 __pipe_unlock(pipe);
Al Viroba5bb142013-03-21 02:21:19 -0400762
Linus Torvaldsb0d8d222013-12-02 09:44:51 -0800763 put_pipe_info(inode, pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 return 0;
765}
766
767static int
Al Viro599a0ac2013-03-12 09:58:10 -0400768pipe_fasync(int fd, struct file *filp, int on)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769{
Al Virode32ec42013-03-21 11:16:56 -0400770 struct pipe_inode_info *pipe = filp->private_data;
Al Viro599a0ac2013-03-12 09:58:10 -0400771 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Al Viroebec73f2013-03-21 12:24:01 -0400773 __pipe_lock(pipe);
Al Viro599a0ac2013-03-12 09:58:10 -0400774 if (filp->f_mode & FMODE_READ)
775 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
776 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200777 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
Al Viro599a0ac2013-03-12 09:58:10 -0400778 if (retval < 0 && (filp->f_mode & FMODE_READ))
779 /* this can happen only if on == T */
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700780 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
781 }
Al Viroebec73f2013-03-21 12:24:01 -0400782 __pipe_unlock(pipe);
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700783 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
Al Viro7bee1302013-03-21 11:04:15 -0400786struct pipe_inode_info *alloc_pipe_info(void)
Ingo Molnar3a326a22006-04-10 15:18:35 +0200787{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200788 struct pipe_inode_info *pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200789
Ingo Molnar923f4f22006-04-11 13:53:33 +0200790 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
791 if (pipe) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200792 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
793 if (pipe->bufs) {
794 init_waitqueue_head(&pipe->wait);
795 pipe->r_counter = pipe->w_counter = 1;
Jens Axboe35f3d142010-05-20 10:43:18 +0200796 pipe->buffers = PIPE_DEF_BUFFERS;
Al Viro72b0d9a2013-03-21 02:32:24 -0400797 mutex_init(&pipe->mutex);
Jens Axboe35f3d142010-05-20 10:43:18 +0200798 return pipe;
799 }
800 kfree(pipe);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200801 }
802
Jens Axboe35f3d142010-05-20 10:43:18 +0200803 return NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200804}
805
Al Viro4b8a8f12013-03-21 11:06:46 -0400806void free_pipe_info(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
808 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Jens Axboe35f3d142010-05-20 10:43:18 +0200810 for (i = 0; i < pipe->buffers; i++) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200811 struct pipe_buffer *buf = pipe->bufs + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 if (buf->ops)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200813 buf->ops->release(pipe, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200815 if (pipe->tmp_page)
816 __free_page(pipe->tmp_page);
Jens Axboe35f3d142010-05-20 10:43:18 +0200817 kfree(pipe->bufs);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200818 kfree(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800821static struct vfsmount *pipe_mnt __read_mostly;
Ingo Molnar341b4462006-04-11 13:57:45 +0200822
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700823/*
824 * pipefs_dname() is called from d_path().
825 */
826static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
827{
828 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
829 dentry->d_inode->i_ino);
830}
831
Al Viro3ba13d12009-02-20 06:02:22 +0000832static const struct dentry_operations pipefs_dentry_operations = {
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700833 .d_dname = pipefs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834};
835
836static struct inode * get_pipe_inode(void)
837{
Eric Dumazeta209dfc2011-07-26 11:36:34 +0200838 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200839 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 if (!inode)
842 goto fail_inode;
843
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400844 inode->i_ino = get_next_ino();
845
Al Viro7bee1302013-03-21 11:04:15 -0400846 pipe = alloc_pipe_info();
Ingo Molnar923f4f22006-04-11 13:53:33 +0200847 if (!pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 goto fail_iput;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200849
Al Viroba5bb142013-03-21 02:21:19 -0400850 inode->i_pipe = pipe;
851 pipe->files = 2;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200852 pipe->readers = pipe->writers = 1;
Al Viro599a0ac2013-03-12 09:58:10 -0400853 inode->i_fop = &pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 /*
856 * Mark the inode dirty from the very beginning,
857 * that way it will never be moved to the dirty
858 * list because "mark_inode_dirty()" will think
859 * that it already _is_ on the dirty list.
860 */
861 inode->i_state = I_DIRTY;
862 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
David Howellsda9592e2008-11-14 10:39:05 +1100863 inode->i_uid = current_fsuid();
864 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200866
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 return inode;
868
869fail_iput:
870 iput(inode);
Ingo Molnar341b4462006-04-11 13:57:45 +0200871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872fail_inode:
873 return NULL;
874}
875
Al Viroe4fad8e2012-07-21 15:33:25 +0400876int create_pipe_files(struct file **res, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
Andi Kleend6cbd282006-09-30 23:29:26 -0700878 int err;
Al Viroe4fad8e2012-07-21 15:33:25 +0400879 struct inode *inode = get_pipe_inode();
Andi Kleend6cbd282006-09-30 23:29:26 -0700880 struct file *f;
Al Viro2c48b9c2009-08-09 00:52:35 +0400881 struct path path;
Al Viroe4fad8e2012-07-21 15:33:25 +0400882 static struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (!inode)
Al Viroe4fad8e2012-07-21 15:33:25 +0400885 return -ENFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
Andi Kleend6cbd282006-09-30 23:29:26 -0700887 err = -ENOMEM;
Nick Piggin4b936882011-01-07 17:50:07 +1100888 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
Al Viro2c48b9c2009-08-09 00:52:35 +0400889 if (!path.dentry)
Andi Kleend6cbd282006-09-30 23:29:26 -0700890 goto err_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400891 path.mnt = mntget(pipe_mnt);
Ingo Molnar341b4462006-04-11 13:57:45 +0200892
Al Viro2c48b9c2009-08-09 00:52:35 +0400893 d_instantiate(path.dentry, inode);
Dave Hansen430e2852008-02-15 14:37:26 -0800894
895 err = -ENFILE;
Al Viro599a0ac2013-03-12 09:58:10 -0400896 f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
Anatol Pomozov39b65252012-09-12 20:11:55 -0700897 if (IS_ERR(f))
Dave Hansen430e2852008-02-15 14:37:26 -0800898 goto err_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
Linus Torvalds98830352012-04-29 13:12:42 -0700900 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
Al Virode32ec42013-03-21 11:16:56 -0400901 f->private_data = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Al Viro599a0ac2013-03-12 09:58:10 -0400903 res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
Anatol Pomozov39b65252012-09-12 20:11:55 -0700904 if (IS_ERR(res[0]))
Al Viroe4fad8e2012-07-21 15:33:25 +0400905 goto err_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Al Viroe4fad8e2012-07-21 15:33:25 +0400907 path_get(&path);
Al Virode32ec42013-03-21 11:16:56 -0400908 res[0]->private_data = inode->i_pipe;
Al Viroe4fad8e2012-07-21 15:33:25 +0400909 res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
910 res[1] = f;
911 return 0;
912
913err_file:
914 put_filp(f);
915err_dentry:
Al Viro4b8a8f12013-03-21 11:06:46 -0400916 free_pipe_info(inode->i_pipe);
Al Viro2c48b9c2009-08-09 00:52:35 +0400917 path_put(&path);
Al Viroe4fad8e2012-07-21 15:33:25 +0400918 return err;
Al Viroed152432008-04-22 19:51:27 -0400919
Al Viroe4fad8e2012-07-21 15:33:25 +0400920err_inode:
Al Viro4b8a8f12013-03-21 11:06:46 -0400921 free_pipe_info(inode->i_pipe);
Andi Kleend6cbd282006-09-30 23:29:26 -0700922 iput(inode);
Al Viroe4fad8e2012-07-21 15:33:25 +0400923 return err;
Andi Kleend6cbd282006-09-30 23:29:26 -0700924}
925
Al Viro5b249b12012-08-19 12:17:29 -0400926static int __do_pipe_flags(int *fd, struct file **files, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -0700927{
Andi Kleend6cbd282006-09-30 23:29:26 -0700928 int error;
929 int fdw, fdr;
930
Linus Torvalds98830352012-04-29 13:12:42 -0700931 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
Ulrich Dreppered8cae82008-07-23 21:29:30 -0700932 return -EINVAL;
933
Al Viroe4fad8e2012-07-21 15:33:25 +0400934 error = create_pipe_files(files, flags);
935 if (error)
936 return error;
Andi Kleend6cbd282006-09-30 23:29:26 -0700937
Ulrich Dreppered8cae82008-07-23 21:29:30 -0700938 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -0700939 if (error < 0)
940 goto err_read_pipe;
941 fdr = error;
942
Ulrich Dreppered8cae82008-07-23 21:29:30 -0700943 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -0700944 if (error < 0)
945 goto err_fdr;
946 fdw = error;
947
Al Viro157cf642008-12-14 04:57:47 -0500948 audit_fd_pair(fdr, fdw);
Andi Kleend6cbd282006-09-30 23:29:26 -0700949 fd[0] = fdr;
950 fd[1] = fdw;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 return 0;
952
Andi Kleend6cbd282006-09-30 23:29:26 -0700953 err_fdr:
954 put_unused_fd(fdr);
955 err_read_pipe:
Al Viroe4fad8e2012-07-21 15:33:25 +0400956 fput(files[0]);
957 fput(files[1]);
Andi Kleend6cbd282006-09-30 23:29:26 -0700958 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959}
960
Al Viro5b249b12012-08-19 12:17:29 -0400961int do_pipe_flags(int *fd, int flags)
962{
963 struct file *files[2];
964 int error = __do_pipe_flags(fd, files, flags);
965 if (!error) {
966 fd_install(fd[0], files[0]);
967 fd_install(fd[1], files[1]);
968 }
969 return error;
970}
971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972/*
Ulrich Drepperd35c7b02008-05-03 15:10:37 -0400973 * sys_pipe() is the normal C calling standard for creating
974 * a pipe. It's not the way Unix traditionally does this, though.
975 */
Heiko Carstensd4e82042009-01-14 14:14:34 +0100976SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
Ulrich Drepperd35c7b02008-05-03 15:10:37 -0400977{
Al Viro5b249b12012-08-19 12:17:29 -0400978 struct file *files[2];
Ulrich Drepperd35c7b02008-05-03 15:10:37 -0400979 int fd[2];
980 int error;
981
Al Viro5b249b12012-08-19 12:17:29 -0400982 error = __do_pipe_flags(fd, files, flags);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -0400983 if (!error) {
Al Viro5b249b12012-08-19 12:17:29 -0400984 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
985 fput(files[0]);
986 fput(files[1]);
987 put_unused_fd(fd[0]);
988 put_unused_fd(fd[1]);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -0400989 error = -EFAULT;
Al Viro5b249b12012-08-19 12:17:29 -0400990 } else {
991 fd_install(fd[0], files[0]);
992 fd_install(fd[1], files[1]);
Ulrich Drepperba719ba2008-05-06 20:42:38 -0700993 }
Ulrich Drepperd35c7b02008-05-03 15:10:37 -0400994 }
995 return error;
996}
997
Heiko Carstens2b664212009-01-14 14:14:35 +0100998SYSCALL_DEFINE1(pipe, int __user *, fildes)
Ulrich Dreppered8cae82008-07-23 21:29:30 -0700999{
1000 return sys_pipe2(fildes, 0);
1001}
1002
Al Virofc7478a2013-03-21 02:07:59 -04001003static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
Al Virof776c732013-03-12 09:46:27 -04001004{
1005 int cur = *cnt;
1006
1007 while (cur == *cnt) {
Al Virofc7478a2013-03-21 02:07:59 -04001008 pipe_wait(pipe);
Al Virof776c732013-03-12 09:46:27 -04001009 if (signal_pending(current))
1010 break;
1011 }
1012 return cur == *cnt ? -ERESTARTSYS : 0;
1013}
1014
Al Virofc7478a2013-03-21 02:07:59 -04001015static void wake_up_partner(struct pipe_inode_info *pipe)
Al Virof776c732013-03-12 09:46:27 -04001016{
Al Virofc7478a2013-03-21 02:07:59 -04001017 wake_up_interruptible(&pipe->wait);
Al Virof776c732013-03-12 09:46:27 -04001018}
1019
1020static int fifo_open(struct inode *inode, struct file *filp)
1021{
1022 struct pipe_inode_info *pipe;
Al Viro599a0ac2013-03-12 09:58:10 -04001023 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
Al Virof776c732013-03-12 09:46:27 -04001024 int ret;
1025
Al Viroba5bb142013-03-21 02:21:19 -04001026 filp->f_version = 0;
1027
1028 spin_lock(&inode->i_lock);
1029 if (inode->i_pipe) {
1030 pipe = inode->i_pipe;
1031 pipe->files++;
1032 spin_unlock(&inode->i_lock);
1033 } else {
1034 spin_unlock(&inode->i_lock);
Al Viro7bee1302013-03-21 11:04:15 -04001035 pipe = alloc_pipe_info();
Al Virof776c732013-03-12 09:46:27 -04001036 if (!pipe)
Al Viroba5bb142013-03-21 02:21:19 -04001037 return -ENOMEM;
1038 pipe->files = 1;
1039 spin_lock(&inode->i_lock);
1040 if (unlikely(inode->i_pipe)) {
1041 inode->i_pipe->files++;
1042 spin_unlock(&inode->i_lock);
Al Viro4b8a8f12013-03-21 11:06:46 -04001043 free_pipe_info(pipe);
Al Viroba5bb142013-03-21 02:21:19 -04001044 pipe = inode->i_pipe;
1045 } else {
1046 inode->i_pipe = pipe;
1047 spin_unlock(&inode->i_lock);
1048 }
Al Virof776c732013-03-12 09:46:27 -04001049 }
Al Virode32ec42013-03-21 11:16:56 -04001050 filp->private_data = pipe;
Al Viroba5bb142013-03-21 02:21:19 -04001051 /* OK, we have a pipe and it's pinned down */
1052
Al Viroebec73f2013-03-21 12:24:01 -04001053 __pipe_lock(pipe);
Al Virof776c732013-03-12 09:46:27 -04001054
1055 /* We can only do regular read/write on fifos */
1056 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
1057
1058 switch (filp->f_mode) {
1059 case FMODE_READ:
1060 /*
1061 * O_RDONLY
1062 * POSIX.1 says that O_NONBLOCK means return with the FIFO
1063 * opened, even when there is no process writing the FIFO.
1064 */
Al Virof776c732013-03-12 09:46:27 -04001065 pipe->r_counter++;
1066 if (pipe->readers++ == 0)
Al Virofc7478a2013-03-21 02:07:59 -04001067 wake_up_partner(pipe);
Al Virof776c732013-03-12 09:46:27 -04001068
Al Viro599a0ac2013-03-12 09:58:10 -04001069 if (!is_pipe && !pipe->writers) {
Al Virof776c732013-03-12 09:46:27 -04001070 if ((filp->f_flags & O_NONBLOCK)) {
1071 /* suppress POLLHUP until we have
1072 * seen a writer */
1073 filp->f_version = pipe->w_counter;
1074 } else {
Al Virofc7478a2013-03-21 02:07:59 -04001075 if (wait_for_partner(pipe, &pipe->w_counter))
Al Virof776c732013-03-12 09:46:27 -04001076 goto err_rd;
1077 }
1078 }
1079 break;
1080
1081 case FMODE_WRITE:
1082 /*
1083 * O_WRONLY
1084 * POSIX.1 says that O_NONBLOCK means return -1 with
1085 * errno=ENXIO when there is no process reading the FIFO.
1086 */
1087 ret = -ENXIO;
Al Viro599a0ac2013-03-12 09:58:10 -04001088 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
Al Virof776c732013-03-12 09:46:27 -04001089 goto err;
1090
Al Virof776c732013-03-12 09:46:27 -04001091 pipe->w_counter++;
1092 if (!pipe->writers++)
Al Virofc7478a2013-03-21 02:07:59 -04001093 wake_up_partner(pipe);
Al Virof776c732013-03-12 09:46:27 -04001094
Al Viro599a0ac2013-03-12 09:58:10 -04001095 if (!is_pipe && !pipe->readers) {
Al Virofc7478a2013-03-21 02:07:59 -04001096 if (wait_for_partner(pipe, &pipe->r_counter))
Al Virof776c732013-03-12 09:46:27 -04001097 goto err_wr;
1098 }
1099 break;
1100
1101 case FMODE_READ | FMODE_WRITE:
1102 /*
1103 * O_RDWR
1104 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
1105 * This implementation will NEVER block on a O_RDWR open, since
1106 * the process can at least talk to itself.
1107 */
Al Virof776c732013-03-12 09:46:27 -04001108
1109 pipe->readers++;
1110 pipe->writers++;
1111 pipe->r_counter++;
1112 pipe->w_counter++;
1113 if (pipe->readers == 1 || pipe->writers == 1)
Al Virofc7478a2013-03-21 02:07:59 -04001114 wake_up_partner(pipe);
Al Virof776c732013-03-12 09:46:27 -04001115 break;
1116
1117 default:
1118 ret = -EINVAL;
1119 goto err;
1120 }
1121
1122 /* Ok! */
Al Viroebec73f2013-03-21 12:24:01 -04001123 __pipe_unlock(pipe);
Al Virof776c732013-03-12 09:46:27 -04001124 return 0;
1125
1126err_rd:
1127 if (!--pipe->readers)
1128 wake_up_interruptible(&pipe->wait);
1129 ret = -ERESTARTSYS;
1130 goto err;
1131
1132err_wr:
1133 if (!--pipe->writers)
1134 wake_up_interruptible(&pipe->wait);
1135 ret = -ERESTARTSYS;
1136 goto err;
1137
1138err:
Al Viroebec73f2013-03-21 12:24:01 -04001139 __pipe_unlock(pipe);
Linus Torvaldsb0d8d222013-12-02 09:44:51 -08001140
1141 put_pipe_info(inode, pipe);
Al Virof776c732013-03-12 09:46:27 -04001142 return ret;
1143}
1144
Al Viro599a0ac2013-03-12 09:58:10 -04001145const struct file_operations pipefifo_fops = {
1146 .open = fifo_open,
1147 .llseek = no_llseek,
1148 .read = do_sync_read,
1149 .aio_read = pipe_read,
1150 .write = do_sync_write,
1151 .aio_write = pipe_write,
1152 .poll = pipe_poll,
1153 .unlocked_ioctl = pipe_ioctl,
1154 .release = pipe_release,
1155 .fasync = pipe_fasync,
Al Virof776c732013-03-12 09:46:27 -04001156};
1157
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001158/*
Jens Axboe35f3d142010-05-20 10:43:18 +02001159 * Allocate a new array of pipe buffers and copy the info over. Returns the
1160 * pipe size if successful, or return -ERROR on error.
1161 */
Jens Axboeb9598db2010-05-24 19:34:43 +02001162static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +02001163{
1164 struct pipe_buffer *bufs;
1165
1166 /*
Jens Axboe35f3d142010-05-20 10:43:18 +02001167 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1168 * expect a lot of shrink+grow operations, just free and allocate
1169 * again like we would do for growing. If the pipe currently
1170 * contains more buffers than arg, then return busy.
1171 */
Jens Axboeb9598db2010-05-24 19:34:43 +02001172 if (nr_pages < pipe->nrbufs)
Jens Axboe35f3d142010-05-20 10:43:18 +02001173 return -EBUSY;
1174
Sasha Levin2ccd4f42012-01-12 17:17:40 -08001175 bufs = kcalloc(nr_pages, sizeof(*bufs), GFP_KERNEL | __GFP_NOWARN);
Jens Axboe35f3d142010-05-20 10:43:18 +02001176 if (unlikely(!bufs))
1177 return -ENOMEM;
1178
1179 /*
1180 * The pipe array wraps around, so just start the new one at zero
1181 * and adjust the indexes.
1182 */
1183 if (pipe->nrbufs) {
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001184 unsigned int tail;
1185 unsigned int head;
Jens Axboe35f3d142010-05-20 10:43:18 +02001186
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001187 tail = pipe->curbuf + pipe->nrbufs;
1188 if (tail < pipe->buffers)
1189 tail = 0;
1190 else
1191 tail &= (pipe->buffers - 1);
1192
1193 head = pipe->nrbufs - tail;
Jens Axboe35f3d142010-05-20 10:43:18 +02001194 if (head)
1195 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1196 if (tail)
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001197 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
Jens Axboe35f3d142010-05-20 10:43:18 +02001198 }
1199
1200 pipe->curbuf = 0;
1201 kfree(pipe->bufs);
1202 pipe->bufs = bufs;
Jens Axboeb9598db2010-05-24 19:34:43 +02001203 pipe->buffers = nr_pages;
1204 return nr_pages * PAGE_SIZE;
Jens Axboe35f3d142010-05-20 10:43:18 +02001205}
1206
Jens Axboeff9da692010-06-03 14:54:39 +02001207/*
1208 * Currently we rely on the pipe array holding a power-of-2 number
1209 * of pages.
1210 */
1211static inline unsigned int round_pipe_size(unsigned int size)
1212{
1213 unsigned long nr_pages;
1214
1215 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1216 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1217}
1218
1219/*
1220 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1221 * will return an error.
1222 */
1223int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1224 size_t *lenp, loff_t *ppos)
1225{
1226 int ret;
1227
1228 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1229 if (ret < 0 || !write)
1230 return ret;
1231
1232 pipe_max_size = round_pipe_size(pipe_max_size);
1233 return ret;
1234}
1235
Linus Torvalds72083642010-11-28 16:27:19 -08001236/*
1237 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1238 * location, so checking ->i_pipe is not enough to verify that this is a
1239 * pipe.
1240 */
1241struct pipe_inode_info *get_pipe_info(struct file *file)
1242{
Al Virode32ec42013-03-21 11:16:56 -04001243 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
Linus Torvalds72083642010-11-28 16:27:19 -08001244}
1245
Jens Axboe35f3d142010-05-20 10:43:18 +02001246long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1247{
1248 struct pipe_inode_info *pipe;
1249 long ret;
1250
Linus Torvaldsc66fb342010-11-28 14:09:57 -08001251 pipe = get_pipe_info(file);
Jens Axboe35f3d142010-05-20 10:43:18 +02001252 if (!pipe)
1253 return -EBADF;
1254
Al Viroebec73f2013-03-21 12:24:01 -04001255 __pipe_lock(pipe);
Jens Axboe35f3d142010-05-20 10:43:18 +02001256
1257 switch (cmd) {
Jens Axboeb9598db2010-05-24 19:34:43 +02001258 case F_SETPIPE_SZ: {
Jens Axboeff9da692010-06-03 14:54:39 +02001259 unsigned int size, nr_pages;
Jens Axboeb9598db2010-05-24 19:34:43 +02001260
Jens Axboeff9da692010-06-03 14:54:39 +02001261 size = round_pipe_size(arg);
1262 nr_pages = size >> PAGE_SHIFT;
Jens Axboeb9598db2010-05-24 19:34:43 +02001263
Miklos Szeredi6db40cf2010-06-09 09:27:57 +02001264 ret = -EINVAL;
1265 if (!nr_pages)
1266 goto out;
1267
Jens Axboeff9da692010-06-03 14:54:39 +02001268 if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
Jens Axboeb4ca7612010-06-01 12:42:12 +02001269 ret = -EPERM;
Julia Lawallcc967be2010-05-26 17:54:39 +02001270 goto out;
Julia Lawallcc967be2010-05-26 17:54:39 +02001271 }
Jens Axboeff9da692010-06-03 14:54:39 +02001272 ret = pipe_set_size(pipe, nr_pages);
Jens Axboe35f3d142010-05-20 10:43:18 +02001273 break;
Jens Axboeb9598db2010-05-24 19:34:43 +02001274 }
Jens Axboe35f3d142010-05-20 10:43:18 +02001275 case F_GETPIPE_SZ:
Jens Axboeb9598db2010-05-24 19:34:43 +02001276 ret = pipe->buffers * PAGE_SIZE;
Jens Axboe35f3d142010-05-20 10:43:18 +02001277 break;
1278 default:
1279 ret = -EINVAL;
1280 break;
1281 }
1282
Julia Lawallcc967be2010-05-26 17:54:39 +02001283out:
Al Viroebec73f2013-03-21 12:24:01 -04001284 __pipe_unlock(pipe);
Jens Axboe35f3d142010-05-20 10:43:18 +02001285 return ret;
1286}
1287
Nick Pigginff0c7d12011-01-07 17:49:50 +11001288static const struct super_operations pipefs_ops = {
1289 .destroy_inode = free_inode_nonrcu,
Pavel Emelyanovd70ef972011-10-31 17:10:04 -07001290 .statfs = simple_statfs,
Nick Pigginff0c7d12011-01-07 17:49:50 +11001291};
1292
Jens Axboe35f3d142010-05-20 10:43:18 +02001293/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 * pipefs should _never_ be mounted by userland - too much of security hassle,
1295 * no real gain from having the whole whorehouse mounted. So we don't need
1296 * any operations on the root directory. However, we need a non-trivial
1297 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1298 */
Al Viro51139ad2010-07-25 23:47:46 +04001299static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1300 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
Al Viroc74a1cb2011-01-12 16:59:34 -05001302 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1303 &pipefs_dentry_operations, PIPEFS_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304}
1305
1306static struct file_system_type pipe_fs_type = {
1307 .name = "pipefs",
Al Viro51139ad2010-07-25 23:47:46 +04001308 .mount = pipefs_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 .kill_sb = kill_anon_super,
1310};
1311
1312static int __init init_pipe_fs(void)
1313{
1314 int err = register_filesystem(&pipe_fs_type);
Ingo Molnar341b4462006-04-11 13:57:45 +02001315
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if (!err) {
1317 pipe_mnt = kern_mount(&pipe_fs_type);
1318 if (IS_ERR(pipe_mnt)) {
1319 err = PTR_ERR(pipe_mnt);
1320 unregister_filesystem(&pipe_fs_type);
1321 }
1322 }
1323 return err;
1324}
1325
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326fs_initcall(init_pipe_fs);