blob: 52c4151148383c2e9cd89f891e322241e110d3c1 [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>
14#include <linux/mount.h>
15#include <linux/pipe_fs_i.h>
16#include <linux/uio.h>
17#include <linux/highmem.h>
Jens Axboe5274f052006-03-30 15:15:30 +020018#include <linux/pagemap.h>
Al Virodb349502007-02-07 01:48:00 -050019#include <linux/audit.h>
Ulrich Drepperba719ba2008-05-06 20:42:38 -070020#include <linux/syscalls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22#include <asm/uaccess.h>
23#include <asm/ioctls.h>
24
25/*
26 * We use a start+len construction, which provides full use of the
27 * allocated memory.
28 * -- Florian Coosmann (FGC)
29 *
30 * Reads with count = 0 should always return 0.
31 * -- Julian Bradfield 1999-06-07.
32 *
33 * FIFOs and Pipes now generate SIGIO for both readers and writers.
34 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
35 *
36 * pipe_read & write cleanup
37 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
38 */
39
Miklos Szeredi61e0d472009-04-14 19:48:41 +020040static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
41{
42 if (pipe->inode)
43 mutex_lock_nested(&pipe->inode->i_mutex, subclass);
44}
45
46void pipe_lock(struct pipe_inode_info *pipe)
47{
48 /*
49 * pipe_lock() nests non-pipe inode locks (for writing to a file)
50 */
51 pipe_lock_nested(pipe, I_MUTEX_PARENT);
52}
53EXPORT_SYMBOL(pipe_lock);
54
55void pipe_unlock(struct pipe_inode_info *pipe)
56{
57 if (pipe->inode)
58 mutex_unlock(&pipe->inode->i_mutex);
59}
60EXPORT_SYMBOL(pipe_unlock);
61
62void pipe_double_lock(struct pipe_inode_info *pipe1,
63 struct pipe_inode_info *pipe2)
64{
65 BUG_ON(pipe1 == pipe2);
66
67 if (pipe1 < pipe2) {
68 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
69 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
70 } else {
Peter Zijlstra023d43c2009-07-21 10:09:23 +020071 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
72 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020073 }
74}
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* Drop the inode semaphore and wait for a pipe event, atomically */
Ingo Molnar3a326a22006-04-10 15:18:35 +020077void pipe_wait(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078{
79 DEFINE_WAIT(wait);
80
Ingo Molnard79fc0f2005-09-10 00:26:12 -070081 /*
82 * Pipes are system-local resources, so sleeping on them
83 * is considered a noninteractive wait:
84 */
Mike Galbraithaf927232007-10-15 17:00:13 +020085 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020086 pipe_unlock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 schedule();
Ingo Molnar3a326a22006-04-10 15:18:35 +020088 finish_wait(&pipe->wait, &wait);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020089 pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090}
91
Arjan van de Ven858119e2006-01-14 13:20:43 -080092static int
Jens Axboef6762b72006-05-01 20:02:05 +020093pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
94 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095{
96 unsigned long copy;
97
98 while (len > 0) {
99 while (!iov->iov_len)
100 iov++;
101 copy = min_t(unsigned long, len, iov->iov_len);
102
Jens Axboef6762b72006-05-01 20:02:05 +0200103 if (atomic) {
104 if (__copy_from_user_inatomic(to, iov->iov_base, copy))
105 return -EFAULT;
106 } else {
107 if (copy_from_user(to, iov->iov_base, copy))
108 return -EFAULT;
109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 to += copy;
111 len -= copy;
112 iov->iov_base += copy;
113 iov->iov_len -= copy;
114 }
115 return 0;
116}
117
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int
Jens Axboef6762b72006-05-01 20:02:05 +0200119pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
120 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
122 unsigned long copy;
123
124 while (len > 0) {
125 while (!iov->iov_len)
126 iov++;
127 copy = min_t(unsigned long, len, iov->iov_len);
128
Jens Axboef6762b72006-05-01 20:02:05 +0200129 if (atomic) {
130 if (__copy_to_user_inatomic(iov->iov_base, from, copy))
131 return -EFAULT;
132 } else {
133 if (copy_to_user(iov->iov_base, from, copy))
134 return -EFAULT;
135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 from += copy;
137 len -= copy;
138 iov->iov_base += copy;
139 iov->iov_len -= copy;
140 }
141 return 0;
142}
143
Jens Axboef6762b72006-05-01 20:02:05 +0200144/*
145 * Attempt to pre-fault in the user memory, so we can use atomic copies.
146 * Returns the number of bytes not faulted in.
147 */
148static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
149{
150 while (!iov->iov_len)
151 iov++;
152
153 while (len > 0) {
154 unsigned long this_len;
155
156 this_len = min_t(unsigned long, len, iov->iov_len);
157 if (fault_in_pages_writeable(iov->iov_base, this_len))
158 break;
159
160 len -= this_len;
161 iov++;
162 }
163
164 return len;
165}
166
167/*
168 * Pre-fault in the user memory, so we can use atomic copies.
169 */
170static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
171{
172 while (!iov->iov_len)
173 iov++;
174
175 while (len > 0) {
176 unsigned long this_len;
177
178 this_len = min_t(unsigned long, len, iov->iov_len);
179 fault_in_pages_readable(iov->iov_base, this_len);
180 len -= this_len;
181 iov++;
182 }
183}
184
Ingo Molnar341b4462006-04-11 13:57:45 +0200185static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
186 struct pipe_buffer *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187{
188 struct page *page = buf->page;
189
Jens Axboe5274f052006-03-30 15:15:30 +0200190 /*
191 * If nobody else uses this page, and we don't already have a
192 * temporary page, let's keep track of it as a one-deep
Ingo Molnar341b4462006-04-11 13:57:45 +0200193 * allocation cache. (Otherwise just release our reference to it)
Jens Axboe5274f052006-03-30 15:15:30 +0200194 */
Ingo Molnar341b4462006-04-11 13:57:45 +0200195 if (page_count(page) == 1 && !pipe->tmp_page)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200196 pipe->tmp_page = page;
Ingo Molnar341b4462006-04-11 13:57:45 +0200197 else
198 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199}
200
Jens Axboe08457182007-06-12 20:51:32 +0200201/**
202 * generic_pipe_buf_map - virtually map a pipe buffer
203 * @pipe: the pipe that the buffer belongs to
204 * @buf: the buffer that should be mapped
205 * @atomic: whether to use an atomic map
206 *
207 * Description:
208 * This function returns a kernel virtual address mapping for the
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800209 * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
Jens Axboe08457182007-06-12 20:51:32 +0200210 * and the caller has to be careful not to fault before calling
211 * the unmap function.
212 *
213 * Note that this function occupies KM_USER0 if @atomic != 0.
214 */
Jens Axboef84d7512006-05-01 19:59:03 +0200215void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200216 struct pipe_buffer *buf, int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
Jens Axboef6762b72006-05-01 20:02:05 +0200218 if (atomic) {
219 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
220 return kmap_atomic(buf->page, KM_USER0);
221 }
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 return kmap(buf->page);
224}
225
Jens Axboe08457182007-06-12 20:51:32 +0200226/**
227 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
228 * @pipe: the pipe that the buffer belongs to
229 * @buf: the buffer that should be unmapped
230 * @map_data: the data that the mapping function returned
231 *
232 * Description:
233 * This function undoes the mapping that ->map() provided.
234 */
Jens Axboef84d7512006-05-01 19:59:03 +0200235void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200236 struct pipe_buffer *buf, void *map_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Jens Axboef6762b72006-05-01 20:02:05 +0200238 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
239 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
240 kunmap_atomic(map_data, KM_USER0);
241 } else
242 kunmap(buf->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
Jens Axboe08457182007-06-12 20:51:32 +0200245/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800246 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200247 * @pipe: the pipe that the buffer belongs to
248 * @buf: the buffer to attempt to steal
249 *
250 * Description:
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800251 * This function attempts to steal the &struct page attached to
Jens Axboe08457182007-06-12 20:51:32 +0200252 * @buf. If successful, this function returns 0 and returns with
253 * the page locked. The caller may then reuse the page for whatever
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800254 * he wishes; the typical use is insertion into a different file
Jens Axboe08457182007-06-12 20:51:32 +0200255 * page cache.
256 */
Jens Axboe330ab712006-05-02 15:29:57 +0200257int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
258 struct pipe_buffer *buf)
Jens Axboe5abc97a2006-03-30 15:16:46 +0200259{
Jens Axboe46e678c2006-04-30 16:36:32 +0200260 struct page *page = buf->page;
261
Jens Axboe08457182007-06-12 20:51:32 +0200262 /*
263 * A reference of one is golden, that means that the owner of this
264 * page is the only one holding a reference to it. lock the page
265 * and return OK.
266 */
Jens Axboe46e678c2006-04-30 16:36:32 +0200267 if (page_count(page) == 1) {
Jens Axboe46e678c2006-04-30 16:36:32 +0200268 lock_page(page);
269 return 0;
270 }
271
272 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200273}
274
Jens Axboe08457182007-06-12 20:51:32 +0200275/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800276 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200277 * @pipe: the pipe that the buffer belongs to
278 * @buf: the buffer to get a reference to
279 *
280 * Description:
281 * This function grabs an extra reference to @buf. It's used in
282 * in the tee() system call, when we duplicate the buffers in one
283 * pipe into another.
284 */
285void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
Jens Axboe70524492006-04-11 15:51:17 +0200286{
287 page_cache_get(buf->page);
288}
289
Jens Axboe08457182007-06-12 20:51:32 +0200290/**
291 * generic_pipe_buf_confirm - verify contents of the pipe buffer
Randy Dunlap79685b82007-07-27 08:08:51 +0200292 * @info: the pipe that the buffer belongs to
Jens Axboe08457182007-06-12 20:51:32 +0200293 * @buf: the buffer to confirm
294 *
295 * Description:
296 * This function does nothing, because the generic pipe code uses
297 * pages that are always good when inserted into the pipe.
298 */
Jens Axboecac36bb2007-06-14 13:10:48 +0200299int generic_pipe_buf_confirm(struct pipe_inode_info *info,
300 struct pipe_buffer *buf)
Jens Axboef84d7512006-05-01 19:59:03 +0200301{
302 return 0;
303}
304
Miklos Szeredi68181732009-05-07 15:37:36 +0200305/**
306 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
307 * @pipe: the pipe that the buffer belongs to
308 * @buf: the buffer to put a reference to
309 *
310 * Description:
311 * This function releases a reference to @buf.
312 */
313void generic_pipe_buf_release(struct pipe_inode_info *pipe,
314 struct pipe_buffer *buf)
315{
316 page_cache_release(buf->page);
317}
318
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800319static const struct pipe_buf_operations anon_pipe_buf_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 .can_merge = 1,
Jens Axboef84d7512006-05-01 19:59:03 +0200321 .map = generic_pipe_buf_map,
322 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb2007-06-14 13:10:48 +0200323 .confirm = generic_pipe_buf_confirm,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 .release = anon_pipe_buf_release,
Jens Axboe330ab712006-05-02 15:29:57 +0200325 .steal = generic_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200326 .get = generic_pipe_buf_get,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327};
328
329static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700330pipe_read(struct kiocb *iocb, const struct iovec *_iov,
331 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700333 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800334 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200335 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 int do_wakeup;
337 ssize_t ret;
338 struct iovec *iov = (struct iovec *)_iov;
339 size_t total_len;
340
341 total_len = iov_length(iov, nr_segs);
342 /* Null read succeeds. */
343 if (unlikely(total_len == 0))
344 return 0;
345
346 do_wakeup = 0;
347 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200348 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200349 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 for (;;) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200351 int bufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 if (bufs) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200353 int curbuf = pipe->curbuf;
354 struct pipe_buffer *buf = pipe->bufs + curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800355 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 void *addr;
357 size_t chars = buf->len;
Jens Axboef6762b72006-05-01 20:02:05 +0200358 int error, atomic;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 if (chars > total_len)
361 chars = total_len;
362
Jens Axboecac36bb2007-06-14 13:10:48 +0200363 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200364 if (error) {
Jens Axboe5274f052006-03-30 15:15:30 +0200365 if (!ret)
Jens Axboef84d7512006-05-01 19:59:03 +0200366 error = ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200367 break;
368 }
Jens Axboef84d7512006-05-01 19:59:03 +0200369
Jens Axboef6762b72006-05-01 20:02:05 +0200370 atomic = !iov_fault_in_pages_write(iov, chars);
371redo:
372 addr = ops->map(pipe, buf, atomic);
373 error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
374 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200376 /*
377 * Just retry with the slow path if we failed.
378 */
379 if (atomic) {
380 atomic = 0;
381 goto redo;
382 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200383 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200384 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 break;
386 }
387 ret += chars;
388 buf->offset += chars;
389 buf->len -= chars;
390 if (!buf->len) {
391 buf->ops = NULL;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200392 ops->release(pipe, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 curbuf = (curbuf + 1) & (PIPE_BUFFERS-1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200394 pipe->curbuf = curbuf;
395 pipe->nrbufs = --bufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 do_wakeup = 1;
397 }
398 total_len -= chars;
399 if (!total_len)
400 break; /* common path: read succeeded */
401 }
402 if (bufs) /* More to do? */
403 continue;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200404 if (!pipe->writers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 break;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200406 if (!pipe->waiting_writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 /* syscall merging: Usually we must not sleep
408 * if O_NONBLOCK is set, or if we got some data.
409 * But if a writer sleeps in kernel space, then
410 * we can wait for that data without violating POSIX.
411 */
412 if (ret)
413 break;
414 if (filp->f_flags & O_NONBLOCK) {
415 ret = -EAGAIN;
416 break;
417 }
418 }
419 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200420 if (!ret)
421 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 break;
423 }
424 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200425 wake_up_interruptible_sync(&pipe->wait);
426 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200428 pipe_wait(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200430 mutex_unlock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200431
432 /* Signal writers asynchronously that there is more room. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200434 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200435 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 }
437 if (ret > 0)
438 file_accessed(filp);
439 return ret;
440}
441
442static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700443pipe_write(struct kiocb *iocb, const struct iovec *_iov,
444 unsigned long nr_segs, loff_t ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700446 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800447 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200448 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 ssize_t ret;
450 int do_wakeup;
451 struct iovec *iov = (struct iovec *)_iov;
452 size_t total_len;
453 ssize_t chars;
454
455 total_len = iov_length(iov, nr_segs);
456 /* Null write succeeds. */
457 if (unlikely(total_len == 0))
458 return 0;
459
460 do_wakeup = 0;
461 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200462 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200463 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Ingo Molnar923f4f22006-04-11 13:53:33 +0200465 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 send_sig(SIGPIPE, current, 0);
467 ret = -EPIPE;
468 goto out;
469 }
470
471 /* We try to merge small writes */
472 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200473 if (pipe->nrbufs && chars != 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200474 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
475 (PIPE_BUFFERS-1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200476 struct pipe_buffer *buf = pipe->bufs + lastbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800477 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 int offset = buf->offset + buf->len;
Ingo Molnar341b4462006-04-11 13:57:45 +0200479
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
Jens Axboef6762b72006-05-01 20:02:05 +0200481 int error, atomic = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200482 void *addr;
Jens Axboe5274f052006-03-30 15:15:30 +0200483
Jens Axboecac36bb2007-06-14 13:10:48 +0200484 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200485 if (error)
Jens Axboe5274f052006-03-30 15:15:30 +0200486 goto out;
Jens Axboef84d7512006-05-01 19:59:03 +0200487
Jens Axboef6762b72006-05-01 20:02:05 +0200488 iov_fault_in_pages_read(iov, chars);
489redo1:
490 addr = ops->map(pipe, buf, atomic);
Jens Axboe5274f052006-03-30 15:15:30 +0200491 error = pipe_iov_copy_from_user(offset + addr, iov,
Jens Axboef6762b72006-05-01 20:02:05 +0200492 chars, atomic);
493 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 ret = error;
495 do_wakeup = 1;
Jens Axboef6762b72006-05-01 20:02:05 +0200496 if (error) {
497 if (atomic) {
498 atomic = 0;
499 goto redo1;
500 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 goto out;
Jens Axboef6762b72006-05-01 20:02:05 +0200502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 buf->len += chars;
504 total_len -= chars;
505 ret = chars;
506 if (!total_len)
507 goto out;
508 }
509 }
510
511 for (;;) {
512 int bufs;
Ingo Molnar341b4462006-04-11 13:57:45 +0200513
Ingo Molnar923f4f22006-04-11 13:53:33 +0200514 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 send_sig(SIGPIPE, current, 0);
Ingo Molnar341b4462006-04-11 13:57:45 +0200516 if (!ret)
517 ret = -EPIPE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 break;
519 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200520 bufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if (bufs < PIPE_BUFFERS) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200522 int newbuf = (pipe->curbuf + bufs) & (PIPE_BUFFERS-1);
523 struct pipe_buffer *buf = pipe->bufs + newbuf;
524 struct page *page = pipe->tmp_page;
Jens Axboef6762b72006-05-01 20:02:05 +0200525 char *src;
526 int error, atomic = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 if (!page) {
529 page = alloc_page(GFP_HIGHUSER);
530 if (unlikely(!page)) {
531 ret = ret ? : -ENOMEM;
532 break;
533 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200534 pipe->tmp_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200536 /* Always wake up, even if the copy fails. Otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 * we lock up (O_NONBLOCK-)readers that sleep due to
538 * syscall merging.
539 * FIXME! Is this really true?
540 */
541 do_wakeup = 1;
542 chars = PAGE_SIZE;
543 if (chars > total_len)
544 chars = total_len;
545
Jens Axboef6762b72006-05-01 20:02:05 +0200546 iov_fault_in_pages_read(iov, chars);
547redo2:
548 if (atomic)
549 src = kmap_atomic(page, KM_USER0);
550 else
551 src = kmap(page);
552
553 error = pipe_iov_copy_from_user(src, iov, chars,
554 atomic);
555 if (atomic)
556 kunmap_atomic(src, KM_USER0);
557 else
558 kunmap(page);
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200561 if (atomic) {
562 atomic = 0;
563 goto redo2;
564 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200565 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200566 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 break;
568 }
569 ret += chars;
570
571 /* Insert it into the buffer array */
572 buf->page = page;
573 buf->ops = &anon_pipe_buf_ops;
574 buf->offset = 0;
575 buf->len = chars;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200576 pipe->nrbufs = ++bufs;
577 pipe->tmp_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
579 total_len -= chars;
580 if (!total_len)
581 break;
582 }
583 if (bufs < PIPE_BUFFERS)
584 continue;
585 if (filp->f_flags & O_NONBLOCK) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200586 if (!ret)
587 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 break;
589 }
590 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200591 if (!ret)
592 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 break;
594 }
595 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200596 wake_up_interruptible_sync(&pipe->wait);
597 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 do_wakeup = 0;
599 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200600 pipe->waiting_writers++;
601 pipe_wait(pipe);
602 pipe->waiting_writers--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 }
604out:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200605 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200607 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200608 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
610 if (ret > 0)
Christoph Hellwig870f4812006-01-09 20:52:01 -0800611 file_update_time(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 return ret;
613}
614
615static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
617{
618 return -EBADF;
619}
620
621static ssize_t
Ingo Molnar341b4462006-04-11 13:57:45 +0200622bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
623 loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
625 return -EBADF;
626}
627
Andi Kleend59d0b12008-02-08 04:21:23 -0800628static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800630 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200631 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 int count, buf, nrbufs;
633
634 switch (cmd) {
635 case FIONREAD:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200636 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200637 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 count = 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200639 buf = pipe->curbuf;
640 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 while (--nrbufs >= 0) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200642 count += pipe->bufs[buf].len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 buf = (buf+1) & (PIPE_BUFFERS-1);
644 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200645 mutex_unlock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 return put_user(count, (int __user *)arg);
648 default:
649 return -EINVAL;
650 }
651}
652
653/* No kernel lock held - fine */
654static unsigned int
655pipe_poll(struct file *filp, poll_table *wait)
656{
657 unsigned int mask;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800658 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200659 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 int nrbufs;
661
Ingo Molnar923f4f22006-04-11 13:53:33 +0200662 poll_wait(filp, &pipe->wait, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 /* Reading only -- no need for acquiring the semaphore. */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200665 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 mask = 0;
667 if (filp->f_mode & FMODE_READ) {
668 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200669 if (!pipe->writers && filp->f_version != pipe->w_counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 mask |= POLLHUP;
671 }
672
673 if (filp->f_mode & FMODE_WRITE) {
674 mask |= (nrbufs < PIPE_BUFFERS) ? POLLOUT | POLLWRNORM : 0;
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700675 /*
676 * Most Unices do not set POLLERR for FIFOs but on Linux they
677 * behave exactly like pipes for poll().
678 */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200679 if (!pipe->readers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 mask |= POLLERR;
681 }
682
683 return mask;
684}
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686static int
687pipe_release(struct inode *inode, int decr, int decw)
688{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200689 struct pipe_inode_info *pipe;
690
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200691 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200692 pipe = inode->i_pipe;
693 pipe->readers -= decr;
694 pipe->writers -= decw;
Ingo Molnar341b4462006-04-11 13:57:45 +0200695
Ingo Molnar923f4f22006-04-11 13:53:33 +0200696 if (!pipe->readers && !pipe->writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 free_pipe_info(inode);
698 } else {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200699 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200700 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
701 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200703 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
705 return 0;
706}
707
708static int
709pipe_read_fasync(int fd, struct file *filp, int on)
710{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800711 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 int retval;
713
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200714 mutex_lock(&inode->i_mutex);
715 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
716 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700718 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719}
720
721
722static int
723pipe_write_fasync(int fd, struct file *filp, int on)
724{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800725 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 int retval;
727
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200728 mutex_lock(&inode->i_mutex);
729 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
730 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700732 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
735
736static int
737pipe_rdwr_fasync(int fd, struct file *filp, int on)
738{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800739 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar341b4462006-04-11 13:57:45 +0200740 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 int retval;
742
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200743 mutex_lock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200744 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700745 if (retval >= 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200746 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700747 if (retval < 0) /* this can happen only if on == T */
748 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
749 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200750 mutex_unlock(&inode->i_mutex);
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700751 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
754
755static int
756pipe_read_release(struct inode *inode, struct file *filp)
757{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return pipe_release(inode, 1, 0);
759}
760
761static int
762pipe_write_release(struct inode *inode, struct file *filp)
763{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 return pipe_release(inode, 0, 1);
765}
766
767static int
768pipe_rdwr_release(struct inode *inode, struct file *filp)
769{
770 int decr, decw;
771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 decr = (filp->f_mode & FMODE_READ) != 0;
773 decw = (filp->f_mode & FMODE_WRITE) != 0;
774 return pipe_release(inode, decr, decw);
775}
776
777static int
778pipe_read_open(struct inode *inode, struct file *filp)
779{
780 /* We could have perhaps used atomic_t, but this and friends
781 below are the only places. So it doesn't seem worthwhile. */
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200782 mutex_lock(&inode->i_mutex);
783 inode->i_pipe->readers++;
784 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
786 return 0;
787}
788
789static int
790pipe_write_open(struct inode *inode, struct file *filp)
791{
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200792 mutex_lock(&inode->i_mutex);
793 inode->i_pipe->writers++;
794 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
796 return 0;
797}
798
799static int
800pipe_rdwr_open(struct inode *inode, struct file *filp)
801{
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200802 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 if (filp->f_mode & FMODE_READ)
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200804 inode->i_pipe->readers++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 if (filp->f_mode & FMODE_WRITE)
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200806 inode->i_pipe->writers++;
807 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 return 0;
810}
811
812/*
813 * The file_operations structs are not static because they
814 * are also used in linux/fs/fifo.c to do operations on FIFOs.
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200815 *
816 * Pipes reuse fifos' file_operations structs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 */
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200818const struct file_operations read_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700820 .read = do_sync_read,
821 .aio_read = pipe_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 .write = bad_pipe_w,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700823 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800824 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 .open = pipe_read_open,
826 .release = pipe_read_release,
827 .fasync = pipe_read_fasync,
828};
829
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200830const struct file_operations write_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 .llseek = no_llseek,
832 .read = bad_pipe_r,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700833 .write = do_sync_write,
834 .aio_write = pipe_write,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700835 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800836 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 .open = pipe_write_open,
838 .release = pipe_write_release,
839 .fasync = pipe_write_fasync,
840};
841
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200842const struct file_operations rdwr_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700844 .read = do_sync_read,
845 .aio_read = pipe_read,
846 .write = do_sync_write,
847 .aio_write = pipe_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800849 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 .open = pipe_rdwr_open,
851 .release = pipe_rdwr_release,
852 .fasync = pipe_rdwr_fasync,
853};
854
Ingo Molnar3a326a22006-04-10 15:18:35 +0200855struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
856{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200857 struct pipe_inode_info *pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200858
Ingo Molnar923f4f22006-04-11 13:53:33 +0200859 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
860 if (pipe) {
861 init_waitqueue_head(&pipe->wait);
862 pipe->r_counter = pipe->w_counter = 1;
863 pipe->inode = inode;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200864 }
865
Ingo Molnar923f4f22006-04-11 13:53:33 +0200866 return pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200867}
868
Ingo Molnar923f4f22006-04-11 13:53:33 +0200869void __free_pipe_info(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
871 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 for (i = 0; i < PIPE_BUFFERS; i++) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200874 struct pipe_buffer *buf = pipe->bufs + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (buf->ops)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200876 buf->ops->release(pipe, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200878 if (pipe->tmp_page)
879 __free_page(pipe->tmp_page);
880 kfree(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881}
882
Jens Axboeb92ce552006-04-11 13:52:07 +0200883void free_pipe_info(struct inode *inode)
884{
885 __free_pipe_info(inode->i_pipe);
886 inode->i_pipe = NULL;
887}
888
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800889static struct vfsmount *pipe_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890static int pipefs_delete_dentry(struct dentry *dentry)
891{
Eric Dumazetd18de5a2006-12-06 20:38:45 -0800892 /*
893 * At creation time, we pretended this dentry was hashed
894 * (by clearing DCACHE_UNHASHED bit in d_flags)
895 * At delete time, we restore the truth : not hashed.
896 * (so that dput() can proceed correctly)
897 */
898 dentry->d_flags |= DCACHE_UNHASHED;
899 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900}
Ingo Molnar341b4462006-04-11 13:57:45 +0200901
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700902/*
903 * pipefs_dname() is called from d_path().
904 */
905static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
906{
907 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
908 dentry->d_inode->i_ino);
909}
910
Al Viro3ba13d12009-02-20 06:02:22 +0000911static const struct dentry_operations pipefs_dentry_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 .d_delete = pipefs_delete_dentry,
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700913 .d_dname = pipefs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914};
915
916static struct inode * get_pipe_inode(void)
917{
918 struct inode *inode = new_inode(pipe_mnt->mnt_sb);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200919 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921 if (!inode)
922 goto fail_inode;
923
Ingo Molnar923f4f22006-04-11 13:53:33 +0200924 pipe = alloc_pipe_info(inode);
925 if (!pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 goto fail_iput;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200927 inode->i_pipe = pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200928
Ingo Molnar923f4f22006-04-11 13:53:33 +0200929 pipe->readers = pipe->writers = 1;
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200930 inode->i_fop = &rdwr_pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 /*
933 * Mark the inode dirty from the very beginning,
934 * that way it will never be moved to the dirty
935 * list because "mark_inode_dirty()" will think
936 * that it already _is_ on the dirty list.
937 */
938 inode->i_state = I_DIRTY;
939 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
David Howellsda9592e2008-11-14 10:39:05 +1100940 inode->i_uid = current_fsuid();
941 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 return inode;
945
946fail_iput:
947 iput(inode);
Ingo Molnar341b4462006-04-11 13:57:45 +0200948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949fail_inode:
950 return NULL;
951}
952
Ulrich Drepperbe61a862008-07-23 21:29:40 -0700953struct file *create_write_pipe(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
Andi Kleend6cbd282006-09-30 23:29:26 -0700955 int err;
956 struct inode *inode;
957 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 struct dentry *dentry;
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700959 struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
Andi Kleend6cbd282006-09-30 23:29:26 -0700961 err = -ENFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 inode = get_pipe_inode();
963 if (!inode)
Dave Hansen430e2852008-02-15 14:37:26 -0800964 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Andi Kleend6cbd282006-09-30 23:29:26 -0700966 err = -ENOMEM;
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700967 dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 if (!dentry)
Andi Kleend6cbd282006-09-30 23:29:26 -0700969 goto err_inode;
Ingo Molnar341b4462006-04-11 13:57:45 +0200970
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 dentry->d_op = &pipefs_dentry_operations;
Eric Dumazetd18de5a2006-12-06 20:38:45 -0800972 /*
973 * We dont want to publish this dentry into global dentry hash table.
974 * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
975 * This permits a working /proc/$pid/fd/XXX on pipes
976 */
977 dentry->d_flags &= ~DCACHE_UNHASHED;
978 d_instantiate(dentry, inode);
Dave Hansen430e2852008-02-15 14:37:26 -0800979
980 err = -ENFILE;
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200981 f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipefifo_fops);
Dave Hansen430e2852008-02-15 14:37:26 -0800982 if (!f)
983 goto err_dentry;
Andi Kleend6cbd282006-09-30 23:29:26 -0700984 f->f_mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
Ulrich Drepperbe61a862008-07-23 21:29:40 -0700986 f->f_flags = O_WRONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -0700987 f->f_version = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Andi Kleend6cbd282006-09-30 23:29:26 -0700989 return f;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
Dave Hansen430e2852008-02-15 14:37:26 -0800991 err_dentry:
Al Viroed152432008-04-22 19:51:27 -0400992 free_pipe_info(inode);
Dave Hansen430e2852008-02-15 14:37:26 -0800993 dput(dentry);
Al Viroed152432008-04-22 19:51:27 -0400994 return ERR_PTR(err);
995
Andi Kleend6cbd282006-09-30 23:29:26 -0700996 err_inode:
997 free_pipe_info(inode);
998 iput(inode);
Dave Hansen430e2852008-02-15 14:37:26 -0800999 err:
Andi Kleend6cbd282006-09-30 23:29:26 -07001000 return ERR_PTR(err);
1001}
1002
1003void free_write_pipe(struct file *f)
1004{
Al Viro5ccac882006-12-18 13:31:18 +00001005 free_pipe_info(f->f_dentry->d_inode);
Jan Blunckc8e7f442008-06-09 16:40:35 -07001006 path_put(&f->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001007 put_filp(f);
1008}
1009
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001010struct file *create_read_pipe(struct file *wrf, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001011{
1012 struct file *f = get_empty_filp();
1013 if (!f)
1014 return ERR_PTR(-ENFILE);
1015
1016 /* Grab pipe from the writer */
Jan Blunckc8e7f442008-06-09 16:40:35 -07001017 f->f_path = wrf->f_path;
1018 path_get(&wrf->f_path);
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001019 f->f_mapping = wrf->f_path.dentry->d_inode->i_mapping;
Andi Kleend6cbd282006-09-30 23:29:26 -07001020
1021 f->f_pos = 0;
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001022 f->f_flags = O_RDONLY | (flags & O_NONBLOCK);
Denys Vlasenkod2d96482008-07-01 14:16:09 +02001023 f->f_op = &read_pipefifo_fops;
Andi Kleend6cbd282006-09-30 23:29:26 -07001024 f->f_mode = FMODE_READ;
1025 f->f_version = 0;
1026
1027 return f;
1028}
1029
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001030int do_pipe_flags(int *fd, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001031{
1032 struct file *fw, *fr;
1033 int error;
1034 int fdw, fdr;
1035
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001036 if (flags & ~(O_CLOEXEC | O_NONBLOCK))
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001037 return -EINVAL;
1038
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001039 fw = create_write_pipe(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001040 if (IS_ERR(fw))
1041 return PTR_ERR(fw);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001042 fr = create_read_pipe(fw, flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001043 error = PTR_ERR(fr);
1044 if (IS_ERR(fr))
1045 goto err_write_pipe;
1046
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001047 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001048 if (error < 0)
1049 goto err_read_pipe;
1050 fdr = error;
1051
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001052 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001053 if (error < 0)
1054 goto err_fdr;
1055 fdw = error;
1056
Al Viro157cf642008-12-14 04:57:47 -05001057 audit_fd_pair(fdr, fdw);
Andi Kleend6cbd282006-09-30 23:29:26 -07001058 fd_install(fdr, fr);
1059 fd_install(fdw, fw);
1060 fd[0] = fdr;
1061 fd[1] = fdw;
Ingo Molnar341b4462006-04-11 13:57:45 +02001062
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 return 0;
1064
Andi Kleend6cbd282006-09-30 23:29:26 -07001065 err_fdr:
1066 put_unused_fd(fdr);
1067 err_read_pipe:
Jan Blunckc8e7f442008-06-09 16:40:35 -07001068 path_put(&fr->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001069 put_filp(fr);
1070 err_write_pipe:
1071 free_write_pipe(fw);
1072 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
1075/*
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001076 * sys_pipe() is the normal C calling standard for creating
1077 * a pipe. It's not the way Unix traditionally does this, though.
1078 */
Heiko Carstensd4e82042009-01-14 14:14:34 +01001079SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001080{
1081 int fd[2];
1082 int error;
1083
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001084 error = do_pipe_flags(fd, flags);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001085 if (!error) {
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001086 if (copy_to_user(fildes, fd, sizeof(fd))) {
1087 sys_close(fd[0]);
1088 sys_close(fd[1]);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001089 error = -EFAULT;
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001090 }
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001091 }
1092 return error;
1093}
1094
Heiko Carstens2b664212009-01-14 14:14:35 +01001095SYSCALL_DEFINE1(pipe, int __user *, fildes)
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001096{
1097 return sys_pipe2(fildes, 0);
1098}
1099
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001100/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 * pipefs should _never_ be mounted by userland - too much of security hassle,
1102 * no real gain from having the whole whorehouse mounted. So we don't need
1103 * any operations on the root directory. However, we need a non-trivial
1104 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1105 */
David Howells454e2392006-06-23 02:02:57 -07001106static int pipefs_get_sb(struct file_system_type *fs_type,
1107 int flags, const char *dev_name, void *data,
1108 struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
David Howells454e2392006-06-23 02:02:57 -07001110 return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111}
1112
1113static struct file_system_type pipe_fs_type = {
1114 .name = "pipefs",
1115 .get_sb = pipefs_get_sb,
1116 .kill_sb = kill_anon_super,
1117};
1118
1119static int __init init_pipe_fs(void)
1120{
1121 int err = register_filesystem(&pipe_fs_type);
Ingo Molnar341b4462006-04-11 13:57:45 +02001122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 if (!err) {
1124 pipe_mnt = kern_mount(&pipe_fs_type);
1125 if (IS_ERR(pipe_mnt)) {
1126 err = PTR_ERR(pipe_mnt);
1127 unregister_filesystem(&pipe_fs_type);
1128 }
1129 }
1130 return err;
1131}
1132
1133static void __exit exit_pipe_fs(void)
1134{
1135 unregister_filesystem(&pipe_fs_type);
1136 mntput(pipe_mnt);
1137}
1138
1139fs_initcall(init_pipe_fs);
1140module_exit(exit_pipe_fs);