blob: d79872eba09a56a03ade25f28a9d9cc8c8b86cb8 [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>
16#include <linux/pipe_fs_i.h>
17#include <linux/uio.h>
18#include <linux/highmem.h>
Jens Axboe5274f052006-03-30 15:15:30 +020019#include <linux/pagemap.h>
Al Virodb349502007-02-07 01:48:00 -050020#include <linux/audit.h>
Ulrich Drepperba719ba2008-05-06 20:42:38 -070021#include <linux/syscalls.h>
Jens Axboeb492e952010-05-19 21:03:16 +020022#include <linux/fcntl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24#include <asm/uaccess.h>
25#include <asm/ioctls.h>
26
27/*
Jens Axboeb492e952010-05-19 21:03:16 +020028 * The max size that a non-root user is allowed to grow the pipe. Can
29 * be set by root in /proc/sys/fs/pipe-max-pages
30 */
31unsigned int pipe_max_pages = PIPE_DEF_BUFFERS * 16;
32
33/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * We use a start+len construction, which provides full use of the
35 * allocated memory.
36 * -- Florian Coosmann (FGC)
37 *
38 * Reads with count = 0 should always return 0.
39 * -- Julian Bradfield 1999-06-07.
40 *
41 * FIFOs and Pipes now generate SIGIO for both readers and writers.
42 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
43 *
44 * pipe_read & write cleanup
45 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
46 */
47
Miklos Szeredi61e0d472009-04-14 19:48:41 +020048static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
49{
50 if (pipe->inode)
51 mutex_lock_nested(&pipe->inode->i_mutex, subclass);
52}
53
54void pipe_lock(struct pipe_inode_info *pipe)
55{
56 /*
57 * pipe_lock() nests non-pipe inode locks (for writing to a file)
58 */
59 pipe_lock_nested(pipe, I_MUTEX_PARENT);
60}
61EXPORT_SYMBOL(pipe_lock);
62
63void pipe_unlock(struct pipe_inode_info *pipe)
64{
65 if (pipe->inode)
66 mutex_unlock(&pipe->inode->i_mutex);
67}
68EXPORT_SYMBOL(pipe_unlock);
69
70void pipe_double_lock(struct pipe_inode_info *pipe1,
71 struct pipe_inode_info *pipe2)
72{
73 BUG_ON(pipe1 == pipe2);
74
75 if (pipe1 < pipe2) {
76 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
77 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
78 } else {
Peter Zijlstra023d43c2009-07-21 10:09:23 +020079 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
80 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020081 }
82}
83
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/* Drop the inode semaphore and wait for a pipe event, atomically */
Ingo Molnar3a326a22006-04-10 15:18:35 +020085void pipe_wait(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
87 DEFINE_WAIT(wait);
88
Ingo Molnard79fc0f2005-09-10 00:26:12 -070089 /*
90 * Pipes are system-local resources, so sleeping on them
91 * is considered a noninteractive wait:
92 */
Mike Galbraithaf927232007-10-15 17:00:13 +020093 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020094 pipe_unlock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 schedule();
Ingo Molnar3a326a22006-04-10 15:18:35 +020096 finish_wait(&pipe->wait, &wait);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020097 pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098}
99
Arjan van de Ven858119e2006-01-14 13:20:43 -0800100static int
Jens Axboef6762b72006-05-01 20:02:05 +0200101pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
102 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103{
104 unsigned long copy;
105
106 while (len > 0) {
107 while (!iov->iov_len)
108 iov++;
109 copy = min_t(unsigned long, len, iov->iov_len);
110
Jens Axboef6762b72006-05-01 20:02:05 +0200111 if (atomic) {
112 if (__copy_from_user_inatomic(to, iov->iov_base, copy))
113 return -EFAULT;
114 } else {
115 if (copy_from_user(to, iov->iov_base, copy))
116 return -EFAULT;
117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 to += copy;
119 len -= copy;
120 iov->iov_base += copy;
121 iov->iov_len -= copy;
122 }
123 return 0;
124}
125
Arjan van de Ven858119e2006-01-14 13:20:43 -0800126static int
Jens Axboef6762b72006-05-01 20:02:05 +0200127pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
128 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129{
130 unsigned long copy;
131
132 while (len > 0) {
133 while (!iov->iov_len)
134 iov++;
135 copy = min_t(unsigned long, len, iov->iov_len);
136
Jens Axboef6762b72006-05-01 20:02:05 +0200137 if (atomic) {
138 if (__copy_to_user_inatomic(iov->iov_base, from, copy))
139 return -EFAULT;
140 } else {
141 if (copy_to_user(iov->iov_base, from, copy))
142 return -EFAULT;
143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 from += copy;
145 len -= copy;
146 iov->iov_base += copy;
147 iov->iov_len -= copy;
148 }
149 return 0;
150}
151
Jens Axboef6762b72006-05-01 20:02:05 +0200152/*
153 * Attempt to pre-fault in the user memory, so we can use atomic copies.
154 * Returns the number of bytes not faulted in.
155 */
156static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
157{
158 while (!iov->iov_len)
159 iov++;
160
161 while (len > 0) {
162 unsigned long this_len;
163
164 this_len = min_t(unsigned long, len, iov->iov_len);
165 if (fault_in_pages_writeable(iov->iov_base, this_len))
166 break;
167
168 len -= this_len;
169 iov++;
170 }
171
172 return len;
173}
174
175/*
176 * Pre-fault in the user memory, so we can use atomic copies.
177 */
178static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
179{
180 while (!iov->iov_len)
181 iov++;
182
183 while (len > 0) {
184 unsigned long this_len;
185
186 this_len = min_t(unsigned long, len, iov->iov_len);
187 fault_in_pages_readable(iov->iov_base, this_len);
188 len -= this_len;
189 iov++;
190 }
191}
192
Ingo Molnar341b4462006-04-11 13:57:45 +0200193static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
194 struct pipe_buffer *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
196 struct page *page = buf->page;
197
Jens Axboe5274f052006-03-30 15:15:30 +0200198 /*
199 * If nobody else uses this page, and we don't already have a
200 * temporary page, let's keep track of it as a one-deep
Ingo Molnar341b4462006-04-11 13:57:45 +0200201 * allocation cache. (Otherwise just release our reference to it)
Jens Axboe5274f052006-03-30 15:15:30 +0200202 */
Ingo Molnar341b4462006-04-11 13:57:45 +0200203 if (page_count(page) == 1 && !pipe->tmp_page)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200204 pipe->tmp_page = page;
Ingo Molnar341b4462006-04-11 13:57:45 +0200205 else
206 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
Jens Axboe08457182007-06-12 20:51:32 +0200209/**
210 * generic_pipe_buf_map - virtually map a pipe buffer
211 * @pipe: the pipe that the buffer belongs to
212 * @buf: the buffer that should be mapped
213 * @atomic: whether to use an atomic map
214 *
215 * Description:
216 * This function returns a kernel virtual address mapping for the
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800217 * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
Jens Axboe08457182007-06-12 20:51:32 +0200218 * and the caller has to be careful not to fault before calling
219 * the unmap function.
220 *
221 * Note that this function occupies KM_USER0 if @atomic != 0.
222 */
Jens Axboef84d7512006-05-01 19:59:03 +0200223void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200224 struct pipe_buffer *buf, int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
Jens Axboef6762b72006-05-01 20:02:05 +0200226 if (atomic) {
227 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
228 return kmap_atomic(buf->page, KM_USER0);
229 }
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 return kmap(buf->page);
232}
233
Jens Axboe08457182007-06-12 20:51:32 +0200234/**
235 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
236 * @pipe: the pipe that the buffer belongs to
237 * @buf: the buffer that should be unmapped
238 * @map_data: the data that the mapping function returned
239 *
240 * Description:
241 * This function undoes the mapping that ->map() provided.
242 */
Jens Axboef84d7512006-05-01 19:59:03 +0200243void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200244 struct pipe_buffer *buf, void *map_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Jens Axboef6762b72006-05-01 20:02:05 +0200246 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
247 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
248 kunmap_atomic(map_data, KM_USER0);
249 } else
250 kunmap(buf->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
Jens Axboe08457182007-06-12 20:51:32 +0200253/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800254 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200255 * @pipe: the pipe that the buffer belongs to
256 * @buf: the buffer to attempt to steal
257 *
258 * Description:
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800259 * This function attempts to steal the &struct page attached to
Jens Axboe08457182007-06-12 20:51:32 +0200260 * @buf. If successful, this function returns 0 and returns with
261 * the page locked. The caller may then reuse the page for whatever
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800262 * he wishes; the typical use is insertion into a different file
Jens Axboe08457182007-06-12 20:51:32 +0200263 * page cache.
264 */
Jens Axboe330ab712006-05-02 15:29:57 +0200265int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
266 struct pipe_buffer *buf)
Jens Axboe5abc97a2006-03-30 15:16:46 +0200267{
Jens Axboe46e678c2006-04-30 16:36:32 +0200268 struct page *page = buf->page;
269
Jens Axboe08457182007-06-12 20:51:32 +0200270 /*
271 * A reference of one is golden, that means that the owner of this
272 * page is the only one holding a reference to it. lock the page
273 * and return OK.
274 */
Jens Axboe46e678c2006-04-30 16:36:32 +0200275 if (page_count(page) == 1) {
Jens Axboe46e678c2006-04-30 16:36:32 +0200276 lock_page(page);
277 return 0;
278 }
279
280 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200281}
282
Jens Axboe08457182007-06-12 20:51:32 +0200283/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800284 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200285 * @pipe: the pipe that the buffer belongs to
286 * @buf: the buffer to get a reference to
287 *
288 * Description:
289 * This function grabs an extra reference to @buf. It's used in
290 * in the tee() system call, when we duplicate the buffers in one
291 * pipe into another.
292 */
293void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
Jens Axboe70524492006-04-11 15:51:17 +0200294{
295 page_cache_get(buf->page);
296}
297
Jens Axboe08457182007-06-12 20:51:32 +0200298/**
299 * generic_pipe_buf_confirm - verify contents of the pipe buffer
Randy Dunlap79685b82007-07-27 08:08:51 +0200300 * @info: the pipe that the buffer belongs to
Jens Axboe08457182007-06-12 20:51:32 +0200301 * @buf: the buffer to confirm
302 *
303 * Description:
304 * This function does nothing, because the generic pipe code uses
305 * pages that are always good when inserted into the pipe.
306 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200307int generic_pipe_buf_confirm(struct pipe_inode_info *info,
308 struct pipe_buffer *buf)
Jens Axboef84d7512006-05-01 19:59:03 +0200309{
310 return 0;
311}
312
Miklos Szeredi68181732009-05-07 15:37:36 +0200313/**
314 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
315 * @pipe: the pipe that the buffer belongs to
316 * @buf: the buffer to put a reference to
317 *
318 * Description:
319 * This function releases a reference to @buf.
320 */
321void generic_pipe_buf_release(struct pipe_inode_info *pipe,
322 struct pipe_buffer *buf)
323{
324 page_cache_release(buf->page);
325}
326
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800327static const struct pipe_buf_operations anon_pipe_buf_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 .can_merge = 1,
Jens Axboef84d7512006-05-01 19:59:03 +0200329 .map = generic_pipe_buf_map,
330 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200331 .confirm = generic_pipe_buf_confirm,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 .release = anon_pipe_buf_release,
Jens Axboe330ab712006-05-02 15:29:57 +0200333 .steal = generic_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200334 .get = generic_pipe_buf_get,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335};
336
337static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700338pipe_read(struct kiocb *iocb, const struct iovec *_iov,
339 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700341 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800342 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200343 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 int do_wakeup;
345 ssize_t ret;
346 struct iovec *iov = (struct iovec *)_iov;
347 size_t total_len;
348
349 total_len = iov_length(iov, nr_segs);
350 /* Null read succeeds. */
351 if (unlikely(total_len == 0))
352 return 0;
353
354 do_wakeup = 0;
355 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200356 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200357 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 for (;;) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200359 int bufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 if (bufs) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200361 int curbuf = pipe->curbuf;
362 struct pipe_buffer *buf = pipe->bufs + curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800363 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 void *addr;
365 size_t chars = buf->len;
Jens Axboef6762b72006-05-01 20:02:05 +0200366 int error, atomic;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
368 if (chars > total_len)
369 chars = total_len;
370
Jens Axboecac36bb02007-06-14 13:10:48 +0200371 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200372 if (error) {
Jens Axboe5274f052006-03-30 15:15:30 +0200373 if (!ret)
Jens Axboef84d7512006-05-01 19:59:03 +0200374 error = ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200375 break;
376 }
Jens Axboef84d7512006-05-01 19:59:03 +0200377
Jens Axboef6762b72006-05-01 20:02:05 +0200378 atomic = !iov_fault_in_pages_write(iov, chars);
379redo:
380 addr = ops->map(pipe, buf, atomic);
381 error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
382 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200384 /*
385 * Just retry with the slow path if we failed.
386 */
387 if (atomic) {
388 atomic = 0;
389 goto redo;
390 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200391 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200392 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 break;
394 }
395 ret += chars;
396 buf->offset += chars;
397 buf->len -= chars;
398 if (!buf->len) {
399 buf->ops = NULL;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200400 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200401 curbuf = (curbuf + 1) & (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200402 pipe->curbuf = curbuf;
403 pipe->nrbufs = --bufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 do_wakeup = 1;
405 }
406 total_len -= chars;
407 if (!total_len)
408 break; /* common path: read succeeded */
409 }
410 if (bufs) /* More to do? */
411 continue;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200412 if (!pipe->writers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 break;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200414 if (!pipe->waiting_writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 /* syscall merging: Usually we must not sleep
416 * if O_NONBLOCK is set, or if we got some data.
417 * But if a writer sleeps in kernel space, then
418 * we can wait for that data without violating POSIX.
419 */
420 if (ret)
421 break;
422 if (filp->f_flags & O_NONBLOCK) {
423 ret = -EAGAIN;
424 break;
425 }
426 }
427 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200428 if (!ret)
429 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 break;
431 }
432 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200433 wake_up_interruptible_sync(&pipe->wait);
434 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200436 pipe_wait(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200438 mutex_unlock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200439
440 /* Signal writers asynchronously that there is more room. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200442 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200443 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 }
445 if (ret > 0)
446 file_accessed(filp);
447 return ret;
448}
449
450static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700451pipe_write(struct kiocb *iocb, const struct iovec *_iov,
452 unsigned long nr_segs, loff_t ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700454 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800455 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200456 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 ssize_t ret;
458 int do_wakeup;
459 struct iovec *iov = (struct iovec *)_iov;
460 size_t total_len;
461 ssize_t chars;
462
463 total_len = iov_length(iov, nr_segs);
464 /* Null write succeeds. */
465 if (unlikely(total_len == 0))
466 return 0;
467
468 do_wakeup = 0;
469 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200470 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200471 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Ingo Molnar923f4f22006-04-11 13:53:33 +0200473 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 send_sig(SIGPIPE, current, 0);
475 ret = -EPIPE;
476 goto out;
477 }
478
479 /* We try to merge small writes */
480 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200481 if (pipe->nrbufs && chars != 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200482 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
Jens Axboe35f3d142010-05-20 10:43:18 +0200483 (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200484 struct pipe_buffer *buf = pipe->bufs + lastbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800485 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 int offset = buf->offset + buf->len;
Ingo Molnar341b4462006-04-11 13:57:45 +0200487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
Jens Axboef6762b72006-05-01 20:02:05 +0200489 int error, atomic = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200490 void *addr;
Jens Axboe5274f052006-03-30 15:15:30 +0200491
Jens Axboecac36bb02007-06-14 13:10:48 +0200492 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200493 if (error)
Jens Axboe5274f052006-03-30 15:15:30 +0200494 goto out;
Jens Axboef84d7512006-05-01 19:59:03 +0200495
Jens Axboef6762b72006-05-01 20:02:05 +0200496 iov_fault_in_pages_read(iov, chars);
497redo1:
498 addr = ops->map(pipe, buf, atomic);
Jens Axboe5274f052006-03-30 15:15:30 +0200499 error = pipe_iov_copy_from_user(offset + addr, iov,
Jens Axboef6762b72006-05-01 20:02:05 +0200500 chars, atomic);
501 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 ret = error;
503 do_wakeup = 1;
Jens Axboef6762b72006-05-01 20:02:05 +0200504 if (error) {
505 if (atomic) {
506 atomic = 0;
507 goto redo1;
508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 goto out;
Jens Axboef6762b72006-05-01 20:02:05 +0200510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 buf->len += chars;
512 total_len -= chars;
513 ret = chars;
514 if (!total_len)
515 goto out;
516 }
517 }
518
519 for (;;) {
520 int bufs;
Ingo Molnar341b4462006-04-11 13:57:45 +0200521
Ingo Molnar923f4f22006-04-11 13:53:33 +0200522 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 send_sig(SIGPIPE, current, 0);
Ingo Molnar341b4462006-04-11 13:57:45 +0200524 if (!ret)
525 ret = -EPIPE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 break;
527 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200528 bufs = pipe->nrbufs;
Jens Axboe35f3d142010-05-20 10:43:18 +0200529 if (bufs < pipe->buffers) {
530 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200531 struct pipe_buffer *buf = pipe->bufs + newbuf;
532 struct page *page = pipe->tmp_page;
Jens Axboef6762b72006-05-01 20:02:05 +0200533 char *src;
534 int error, atomic = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 if (!page) {
537 page = alloc_page(GFP_HIGHUSER);
538 if (unlikely(!page)) {
539 ret = ret ? : -ENOMEM;
540 break;
541 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200542 pipe->tmp_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200544 /* Always wake up, even if the copy fails. Otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 * we lock up (O_NONBLOCK-)readers that sleep due to
546 * syscall merging.
547 * FIXME! Is this really true?
548 */
549 do_wakeup = 1;
550 chars = PAGE_SIZE;
551 if (chars > total_len)
552 chars = total_len;
553
Jens Axboef6762b72006-05-01 20:02:05 +0200554 iov_fault_in_pages_read(iov, chars);
555redo2:
556 if (atomic)
557 src = kmap_atomic(page, KM_USER0);
558 else
559 src = kmap(page);
560
561 error = pipe_iov_copy_from_user(src, iov, chars,
562 atomic);
563 if (atomic)
564 kunmap_atomic(src, KM_USER0);
565 else
566 kunmap(page);
567
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200569 if (atomic) {
570 atomic = 0;
571 goto redo2;
572 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200573 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200574 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 break;
576 }
577 ret += chars;
578
579 /* Insert it into the buffer array */
580 buf->page = page;
581 buf->ops = &anon_pipe_buf_ops;
582 buf->offset = 0;
583 buf->len = chars;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200584 pipe->nrbufs = ++bufs;
585 pipe->tmp_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 total_len -= chars;
588 if (!total_len)
589 break;
590 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200591 if (bufs < pipe->buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 continue;
593 if (filp->f_flags & O_NONBLOCK) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200594 if (!ret)
595 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 break;
597 }
598 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200599 if (!ret)
600 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 break;
602 }
603 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200604 wake_up_interruptible_sync(&pipe->wait);
605 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 do_wakeup = 0;
607 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200608 pipe->waiting_writers++;
609 pipe_wait(pipe);
610 pipe->waiting_writers--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 }
612out:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200613 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200615 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200616 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618 if (ret > 0)
Christoph Hellwig870f4812006-01-09 20:52:01 -0800619 file_update_time(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 return ret;
621}
622
623static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
625{
626 return -EBADF;
627}
628
629static ssize_t
Ingo Molnar341b4462006-04-11 13:57:45 +0200630bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
631 loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 return -EBADF;
634}
635
Andi Kleend59d0b12008-02-08 04:21:23 -0800636static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800638 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200639 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 int count, buf, nrbufs;
641
642 switch (cmd) {
643 case FIONREAD:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200644 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200645 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 count = 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200647 buf = pipe->curbuf;
648 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 while (--nrbufs >= 0) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200650 count += pipe->bufs[buf].len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200651 buf = (buf+1) & (pipe->buffers - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200653 mutex_unlock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200654
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 return put_user(count, (int __user *)arg);
656 default:
657 return -EINVAL;
658 }
659}
660
661/* No kernel lock held - fine */
662static unsigned int
663pipe_poll(struct file *filp, poll_table *wait)
664{
665 unsigned int mask;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800666 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200667 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 int nrbufs;
669
Ingo Molnar923f4f22006-04-11 13:53:33 +0200670 poll_wait(filp, &pipe->wait, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
672 /* Reading only -- no need for acquiring the semaphore. */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200673 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 mask = 0;
675 if (filp->f_mode & FMODE_READ) {
676 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200677 if (!pipe->writers && filp->f_version != pipe->w_counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 mask |= POLLHUP;
679 }
680
681 if (filp->f_mode & FMODE_WRITE) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200682 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700683 /*
684 * Most Unices do not set POLLERR for FIFOs but on Linux they
685 * behave exactly like pipes for poll().
686 */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200687 if (!pipe->readers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 mask |= POLLERR;
689 }
690
691 return mask;
692}
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694static int
695pipe_release(struct inode *inode, int decr, int decw)
696{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200697 struct pipe_inode_info *pipe;
698
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200699 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200700 pipe = inode->i_pipe;
701 pipe->readers -= decr;
702 pipe->writers -= decw;
Ingo Molnar341b4462006-04-11 13:57:45 +0200703
Ingo Molnar923f4f22006-04-11 13:53:33 +0200704 if (!pipe->readers && !pipe->writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 free_pipe_info(inode);
706 } else {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200707 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200708 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
709 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200711 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
713 return 0;
714}
715
716static int
717pipe_read_fasync(int fd, struct file *filp, int on)
718{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800719 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 int retval;
721
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200722 mutex_lock(&inode->i_mutex);
723 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
724 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700726 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727}
728
729
730static int
731pipe_write_fasync(int fd, struct file *filp, int on)
732{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800733 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 int retval;
735
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200736 mutex_lock(&inode->i_mutex);
737 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
738 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700740 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741}
742
743
744static int
745pipe_rdwr_fasync(int fd, struct file *filp, int on)
746{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800747 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar341b4462006-04-11 13:57:45 +0200748 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 int retval;
750
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200751 mutex_lock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200752 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700753 if (retval >= 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200754 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700755 if (retval < 0) /* this can happen only if on == T */
756 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
757 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200758 mutex_unlock(&inode->i_mutex);
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700759 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760}
761
762
763static int
764pipe_read_release(struct inode *inode, struct file *filp)
765{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 return pipe_release(inode, 1, 0);
767}
768
769static int
770pipe_write_release(struct inode *inode, struct file *filp)
771{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 return pipe_release(inode, 0, 1);
773}
774
775static int
776pipe_rdwr_release(struct inode *inode, struct file *filp)
777{
778 int decr, decw;
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 decr = (filp->f_mode & FMODE_READ) != 0;
781 decw = (filp->f_mode & FMODE_WRITE) != 0;
782 return pipe_release(inode, decr, decw);
783}
784
785static int
786pipe_read_open(struct inode *inode, struct file *filp)
787{
Earl Chewad396022009-10-19 15:55:41 -0700788 int ret = -ENOENT;
789
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200790 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700791
792 if (inode->i_pipe) {
793 ret = 0;
794 inode->i_pipe->readers++;
795 }
796
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200797 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Earl Chewad396022009-10-19 15:55:41 -0700799 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
802static int
803pipe_write_open(struct inode *inode, struct file *filp)
804{
Earl Chewad396022009-10-19 15:55:41 -0700805 int ret = -ENOENT;
806
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200807 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700808
809 if (inode->i_pipe) {
810 ret = 0;
811 inode->i_pipe->writers++;
812 }
813
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200814 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
Earl Chewad396022009-10-19 15:55:41 -0700816 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
819static int
820pipe_rdwr_open(struct inode *inode, struct file *filp)
821{
Earl Chewad396022009-10-19 15:55:41 -0700822 int ret = -ENOENT;
823
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200824 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700825
826 if (inode->i_pipe) {
827 ret = 0;
828 if (filp->f_mode & FMODE_READ)
829 inode->i_pipe->readers++;
830 if (filp->f_mode & FMODE_WRITE)
831 inode->i_pipe->writers++;
832 }
833
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200834 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Earl Chewad396022009-10-19 15:55:41 -0700836 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837}
838
839/*
840 * The file_operations structs are not static because they
841 * are also used in linux/fs/fifo.c to do operations on FIFOs.
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200842 *
843 * Pipes reuse fifos' file_operations structs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 */
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200845const struct file_operations read_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700847 .read = do_sync_read,
848 .aio_read = pipe_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 .write = bad_pipe_w,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700850 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800851 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 .open = pipe_read_open,
853 .release = pipe_read_release,
854 .fasync = pipe_read_fasync,
855};
856
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200857const struct file_operations write_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 .llseek = no_llseek,
859 .read = bad_pipe_r,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700860 .write = do_sync_write,
861 .aio_write = pipe_write,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700862 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800863 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 .open = pipe_write_open,
865 .release = pipe_write_release,
866 .fasync = pipe_write_fasync,
867};
868
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200869const struct file_operations rdwr_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700871 .read = do_sync_read,
872 .aio_read = pipe_read,
873 .write = do_sync_write,
874 .aio_write = pipe_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800876 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 .open = pipe_rdwr_open,
878 .release = pipe_rdwr_release,
879 .fasync = pipe_rdwr_fasync,
880};
881
Ingo Molnar3a326a22006-04-10 15:18:35 +0200882struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
883{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200884 struct pipe_inode_info *pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200885
Ingo Molnar923f4f22006-04-11 13:53:33 +0200886 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
887 if (pipe) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200888 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
889 if (pipe->bufs) {
890 init_waitqueue_head(&pipe->wait);
891 pipe->r_counter = pipe->w_counter = 1;
892 pipe->inode = inode;
893 pipe->buffers = PIPE_DEF_BUFFERS;
894 return pipe;
895 }
896 kfree(pipe);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200897 }
898
Jens Axboe35f3d142010-05-20 10:43:18 +0200899 return NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200900}
901
Ingo Molnar923f4f22006-04-11 13:53:33 +0200902void __free_pipe_info(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
904 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
Jens Axboe35f3d142010-05-20 10:43:18 +0200906 for (i = 0; i < pipe->buffers; i++) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200907 struct pipe_buffer *buf = pipe->bufs + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (buf->ops)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200909 buf->ops->release(pipe, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200911 if (pipe->tmp_page)
912 __free_page(pipe->tmp_page);
Jens Axboe35f3d142010-05-20 10:43:18 +0200913 kfree(pipe->bufs);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200914 kfree(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
916
Jens Axboeb92ce552006-04-11 13:52:07 +0200917void free_pipe_info(struct inode *inode)
918{
919 __free_pipe_info(inode->i_pipe);
920 inode->i_pipe = NULL;
921}
922
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800923static struct vfsmount *pipe_mnt __read_mostly;
Ingo Molnar341b4462006-04-11 13:57:45 +0200924
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700925/*
926 * pipefs_dname() is called from d_path().
927 */
928static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
929{
930 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
931 dentry->d_inode->i_ino);
932}
933
Al Viro3ba13d12009-02-20 06:02:22 +0000934static const struct dentry_operations pipefs_dentry_operations = {
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700935 .d_dname = pipefs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936};
937
938static struct inode * get_pipe_inode(void)
939{
940 struct inode *inode = new_inode(pipe_mnt->mnt_sb);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200941 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943 if (!inode)
944 goto fail_inode;
945
Ingo Molnar923f4f22006-04-11 13:53:33 +0200946 pipe = alloc_pipe_info(inode);
947 if (!pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 goto fail_iput;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200949 inode->i_pipe = pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200950
Ingo Molnar923f4f22006-04-11 13:53:33 +0200951 pipe->readers = pipe->writers = 1;
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200952 inode->i_fop = &rdwr_pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954 /*
955 * Mark the inode dirty from the very beginning,
956 * that way it will never be moved to the dirty
957 * list because "mark_inode_dirty()" will think
958 * that it already _is_ on the dirty list.
959 */
960 inode->i_state = I_DIRTY;
961 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
David Howellsda9592e2008-11-14 10:39:05 +1100962 inode->i_uid = current_fsuid();
963 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 return inode;
967
968fail_iput:
969 iput(inode);
Ingo Molnar341b4462006-04-11 13:57:45 +0200970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971fail_inode:
972 return NULL;
973}
974
Ulrich Drepperbe61a862008-07-23 21:29:40 -0700975struct file *create_write_pipe(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976{
Andi Kleend6cbd282006-09-30 23:29:26 -0700977 int err;
978 struct inode *inode;
979 struct file *f;
Al Viro2c48b9c2009-08-09 00:52:35 +0400980 struct path path;
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700981 struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Andi Kleend6cbd282006-09-30 23:29:26 -0700983 err = -ENFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 inode = get_pipe_inode();
985 if (!inode)
Dave Hansen430e2852008-02-15 14:37:26 -0800986 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987
Andi Kleend6cbd282006-09-30 23:29:26 -0700988 err = -ENOMEM;
Al Viro2c48b9c2009-08-09 00:52:35 +0400989 path.dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name);
990 if (!path.dentry)
Andi Kleend6cbd282006-09-30 23:29:26 -0700991 goto err_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400992 path.mnt = mntget(pipe_mnt);
Ingo Molnar341b4462006-04-11 13:57:45 +0200993
Al Viro2c48b9c2009-08-09 00:52:35 +0400994 path.dentry->d_op = &pipefs_dentry_operations;
Al Viro2c48b9c2009-08-09 00:52:35 +0400995 d_instantiate(path.dentry, inode);
Dave Hansen430e2852008-02-15 14:37:26 -0800996
997 err = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +0400998 f = alloc_file(&path, FMODE_WRITE, &write_pipefifo_fops);
Dave Hansen430e2852008-02-15 14:37:26 -0800999 if (!f)
1000 goto err_dentry;
Andi Kleend6cbd282006-09-30 23:29:26 -07001001 f->f_mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001003 f->f_flags = O_WRONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -07001004 f->f_version = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Andi Kleend6cbd282006-09-30 23:29:26 -07001006 return f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Dave Hansen430e2852008-02-15 14:37:26 -08001008 err_dentry:
Al Viroed152432008-04-22 19:51:27 -04001009 free_pipe_info(inode);
Al Viro2c48b9c2009-08-09 00:52:35 +04001010 path_put(&path);
Al Viroed152432008-04-22 19:51:27 -04001011 return ERR_PTR(err);
1012
Andi Kleend6cbd282006-09-30 23:29:26 -07001013 err_inode:
1014 free_pipe_info(inode);
1015 iput(inode);
Dave Hansen430e2852008-02-15 14:37:26 -08001016 err:
Andi Kleend6cbd282006-09-30 23:29:26 -07001017 return ERR_PTR(err);
1018}
1019
1020void free_write_pipe(struct file *f)
1021{
Al Viro5ccac88e2006-12-18 13:31:18 +00001022 free_pipe_info(f->f_dentry->d_inode);
Jan Blunckc8e7f442008-06-09 16:40:35 -07001023 path_put(&f->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001024 put_filp(f);
1025}
1026
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001027struct file *create_read_pipe(struct file *wrf, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001028{
Al Virod2314122009-08-09 01:01:37 +04001029 /* Grab pipe from the writer */
1030 struct file *f = alloc_file(&wrf->f_path, FMODE_READ,
1031 &read_pipefifo_fops);
Andi Kleend6cbd282006-09-30 23:29:26 -07001032 if (!f)
1033 return ERR_PTR(-ENFILE);
1034
Jan Blunckc8e7f442008-06-09 16:40:35 -07001035 path_get(&wrf->f_path);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001036 f->f_flags = O_RDONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -07001037
1038 return f;
1039}
1040
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001041int do_pipe_flags(int *fd, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001042{
1043 struct file *fw, *fr;
1044 int error;
1045 int fdw, fdr;
1046
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001047 if (flags & ~(O_CLOEXEC | O_NONBLOCK))
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001048 return -EINVAL;
1049
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001050 fw = create_write_pipe(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001051 if (IS_ERR(fw))
1052 return PTR_ERR(fw);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001053 fr = create_read_pipe(fw, flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001054 error = PTR_ERR(fr);
1055 if (IS_ERR(fr))
1056 goto err_write_pipe;
1057
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001058 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001059 if (error < 0)
1060 goto err_read_pipe;
1061 fdr = error;
1062
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001063 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001064 if (error < 0)
1065 goto err_fdr;
1066 fdw = error;
1067
Al Viro157cf642008-12-14 04:57:47 -05001068 audit_fd_pair(fdr, fdw);
Andi Kleend6cbd282006-09-30 23:29:26 -07001069 fd_install(fdr, fr);
1070 fd_install(fdw, fw);
1071 fd[0] = fdr;
1072 fd[1] = fdw;
Ingo Molnar341b4462006-04-11 13:57:45 +02001073
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 return 0;
1075
Andi Kleend6cbd282006-09-30 23:29:26 -07001076 err_fdr:
1077 put_unused_fd(fdr);
1078 err_read_pipe:
Jan Blunckc8e7f442008-06-09 16:40:35 -07001079 path_put(&fr->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001080 put_filp(fr);
1081 err_write_pipe:
1082 free_write_pipe(fw);
1083 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084}
1085
1086/*
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001087 * sys_pipe() is the normal C calling standard for creating
1088 * a pipe. It's not the way Unix traditionally does this, though.
1089 */
Heiko Carstensd4e82042009-01-14 14:14:34 +01001090SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001091{
1092 int fd[2];
1093 int error;
1094
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001095 error = do_pipe_flags(fd, flags);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001096 if (!error) {
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001097 if (copy_to_user(fildes, fd, sizeof(fd))) {
1098 sys_close(fd[0]);
1099 sys_close(fd[1]);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001100 error = -EFAULT;
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001101 }
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001102 }
1103 return error;
1104}
1105
Heiko Carstens2b664212009-01-14 14:14:35 +01001106SYSCALL_DEFINE1(pipe, int __user *, fildes)
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001107{
1108 return sys_pipe2(fildes, 0);
1109}
1110
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001111/*
Jens Axboe35f3d142010-05-20 10:43:18 +02001112 * Allocate a new array of pipe buffers and copy the info over. Returns the
1113 * pipe size if successful, or return -ERROR on error.
1114 */
1115static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1116{
1117 struct pipe_buffer *bufs;
1118
1119 /*
1120 * Must be a power-of-2 currently
1121 */
1122 if (!is_power_of_2(arg))
1123 return -EINVAL;
1124
1125 /*
1126 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1127 * expect a lot of shrink+grow operations, just free and allocate
1128 * again like we would do for growing. If the pipe currently
1129 * contains more buffers than arg, then return busy.
1130 */
1131 if (arg < pipe->nrbufs)
1132 return -EBUSY;
1133
1134 bufs = kcalloc(arg, sizeof(struct pipe_buffer), GFP_KERNEL);
1135 if (unlikely(!bufs))
1136 return -ENOMEM;
1137
1138 /*
1139 * The pipe array wraps around, so just start the new one at zero
1140 * and adjust the indexes.
1141 */
1142 if (pipe->nrbufs) {
1143 const unsigned int tail = pipe->nrbufs & (pipe->buffers - 1);
1144 const unsigned int head = pipe->nrbufs - tail;
1145
1146 if (head)
1147 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1148 if (tail)
1149 memcpy(bufs + head, pipe->bufs + pipe->curbuf, tail * sizeof(struct pipe_buffer));
1150 }
1151
1152 pipe->curbuf = 0;
1153 kfree(pipe->bufs);
1154 pipe->bufs = bufs;
1155 pipe->buffers = arg;
1156 return arg;
1157}
1158
1159long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1160{
1161 struct pipe_inode_info *pipe;
1162 long ret;
1163
1164 pipe = file->f_path.dentry->d_inode->i_pipe;
1165 if (!pipe)
1166 return -EBADF;
1167
1168 mutex_lock(&pipe->inode->i_mutex);
1169
1170 switch (cmd) {
1171 case F_SETPIPE_SZ:
Jens Axboeb492e952010-05-19 21:03:16 +02001172 if (!capable(CAP_SYS_ADMIN) && arg > pipe_max_pages)
1173 return -EINVAL;
1174 /*
1175 * The pipe needs to be at least 2 pages large to
1176 * guarantee POSIX behaviour.
1177 */
1178 if (arg < 2)
1179 return -EINVAL;
Jens Axboe35f3d142010-05-20 10:43:18 +02001180 ret = pipe_set_size(pipe, arg);
1181 break;
1182 case F_GETPIPE_SZ:
1183 ret = pipe->buffers;
1184 break;
1185 default:
1186 ret = -EINVAL;
1187 break;
1188 }
1189
1190 mutex_unlock(&pipe->inode->i_mutex);
1191 return ret;
1192}
1193
1194/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 * pipefs should _never_ be mounted by userland - too much of security hassle,
1196 * no real gain from having the whole whorehouse mounted. So we don't need
1197 * any operations on the root directory. However, we need a non-trivial
1198 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1199 */
David Howells454e2392006-06-23 02:02:57 -07001200static int pipefs_get_sb(struct file_system_type *fs_type,
1201 int flags, const char *dev_name, void *data,
1202 struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203{
David Howells454e2392006-06-23 02:02:57 -07001204 return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
1207static struct file_system_type pipe_fs_type = {
1208 .name = "pipefs",
1209 .get_sb = pipefs_get_sb,
1210 .kill_sb = kill_anon_super,
1211};
1212
1213static int __init init_pipe_fs(void)
1214{
1215 int err = register_filesystem(&pipe_fs_type);
Ingo Molnar341b4462006-04-11 13:57:45 +02001216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (!err) {
1218 pipe_mnt = kern_mount(&pipe_fs_type);
1219 if (IS_ERR(pipe_mnt)) {
1220 err = PTR_ERR(pipe_mnt);
1221 unregister_filesystem(&pipe_fs_type);
1222 }
1223 }
1224 return err;
1225}
1226
1227static void __exit exit_pipe_fs(void)
1228{
1229 unregister_filesystem(&pipe_fs_type);
1230 mntput(pipe_mnt);
1231}
1232
1233fs_initcall(init_pipe_fs);
1234module_exit(exit_pipe_fs);