blob: db6eaaba0dd81fb777f2b64f1e385d40eb3f1819 [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}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200233EXPORT_SYMBOL(generic_pipe_buf_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Jens Axboe08457182007-06-12 20:51:32 +0200235/**
236 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
237 * @pipe: the pipe that the buffer belongs to
238 * @buf: the buffer that should be unmapped
239 * @map_data: the data that the mapping function returned
240 *
241 * Description:
242 * This function undoes the mapping that ->map() provided.
243 */
Jens Axboef84d7512006-05-01 19:59:03 +0200244void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200245 struct pipe_buffer *buf, void *map_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Jens Axboef6762b72006-05-01 20:02:05 +0200247 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
248 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
249 kunmap_atomic(map_data, KM_USER0);
250 } else
251 kunmap(buf->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200253EXPORT_SYMBOL(generic_pipe_buf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Jens Axboe08457182007-06-12 20:51:32 +0200255/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800256 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200257 * @pipe: the pipe that the buffer belongs to
258 * @buf: the buffer to attempt to steal
259 *
260 * Description:
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800261 * This function attempts to steal the &struct page attached to
Jens Axboe08457182007-06-12 20:51:32 +0200262 * @buf. If successful, this function returns 0 and returns with
263 * the page locked. The caller may then reuse the page for whatever
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800264 * he wishes; the typical use is insertion into a different file
Jens Axboe08457182007-06-12 20:51:32 +0200265 * page cache.
266 */
Jens Axboe330ab712006-05-02 15:29:57 +0200267int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
268 struct pipe_buffer *buf)
Jens Axboe5abc97a2006-03-30 15:16:46 +0200269{
Jens Axboe46e678c2006-04-30 16:36:32 +0200270 struct page *page = buf->page;
271
Jens Axboe08457182007-06-12 20:51:32 +0200272 /*
273 * A reference of one is golden, that means that the owner of this
274 * page is the only one holding a reference to it. lock the page
275 * and return OK.
276 */
Jens Axboe46e678c2006-04-30 16:36:32 +0200277 if (page_count(page) == 1) {
Jens Axboe46e678c2006-04-30 16:36:32 +0200278 lock_page(page);
279 return 0;
280 }
281
282 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200283}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200284EXPORT_SYMBOL(generic_pipe_buf_steal);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200285
Jens Axboe08457182007-06-12 20:51:32 +0200286/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800287 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200288 * @pipe: the pipe that the buffer belongs to
289 * @buf: the buffer to get a reference to
290 *
291 * Description:
292 * This function grabs an extra reference to @buf. It's used in
293 * in the tee() system call, when we duplicate the buffers in one
294 * pipe into another.
295 */
296void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
Jens Axboe70524492006-04-11 15:51:17 +0200297{
298 page_cache_get(buf->page);
299}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200300EXPORT_SYMBOL(generic_pipe_buf_get);
Jens Axboe70524492006-04-11 15:51:17 +0200301
Jens Axboe08457182007-06-12 20:51:32 +0200302/**
303 * generic_pipe_buf_confirm - verify contents of the pipe buffer
Randy Dunlap79685b82007-07-27 08:08:51 +0200304 * @info: the pipe that the buffer belongs to
Jens Axboe08457182007-06-12 20:51:32 +0200305 * @buf: the buffer to confirm
306 *
307 * Description:
308 * This function does nothing, because the generic pipe code uses
309 * pages that are always good when inserted into the pipe.
310 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200311int generic_pipe_buf_confirm(struct pipe_inode_info *info,
312 struct pipe_buffer *buf)
Jens Axboef84d7512006-05-01 19:59:03 +0200313{
314 return 0;
315}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200316EXPORT_SYMBOL(generic_pipe_buf_confirm);
Jens Axboef84d7512006-05-01 19:59:03 +0200317
Miklos Szeredi68181732009-05-07 15:37:36 +0200318/**
319 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
320 * @pipe: the pipe that the buffer belongs to
321 * @buf: the buffer to put a reference to
322 *
323 * Description:
324 * This function releases a reference to @buf.
325 */
326void generic_pipe_buf_release(struct pipe_inode_info *pipe,
327 struct pipe_buffer *buf)
328{
329 page_cache_release(buf->page);
330}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200331EXPORT_SYMBOL(generic_pipe_buf_release);
Miklos Szeredi68181732009-05-07 15:37:36 +0200332
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800333static const struct pipe_buf_operations anon_pipe_buf_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 .can_merge = 1,
Jens Axboef84d7512006-05-01 19:59:03 +0200335 .map = generic_pipe_buf_map,
336 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200337 .confirm = generic_pipe_buf_confirm,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 .release = anon_pipe_buf_release,
Jens Axboe330ab712006-05-02 15:29:57 +0200339 .steal = generic_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200340 .get = generic_pipe_buf_get,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341};
342
343static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700344pipe_read(struct kiocb *iocb, const struct iovec *_iov,
345 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700347 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800348 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200349 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 int do_wakeup;
351 ssize_t ret;
352 struct iovec *iov = (struct iovec *)_iov;
353 size_t total_len;
354
355 total_len = iov_length(iov, nr_segs);
356 /* Null read succeeds. */
357 if (unlikely(total_len == 0))
358 return 0;
359
360 do_wakeup = 0;
361 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200362 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200363 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 for (;;) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200365 int bufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 if (bufs) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200367 int curbuf = pipe->curbuf;
368 struct pipe_buffer *buf = pipe->bufs + curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800369 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 void *addr;
371 size_t chars = buf->len;
Jens Axboef6762b72006-05-01 20:02:05 +0200372 int error, atomic;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
374 if (chars > total_len)
375 chars = total_len;
376
Jens Axboecac36bb02007-06-14 13:10:48 +0200377 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200378 if (error) {
Jens Axboe5274f052006-03-30 15:15:30 +0200379 if (!ret)
Jens Axboef84d7512006-05-01 19:59:03 +0200380 error = ret;
Jens Axboe5274f052006-03-30 15:15:30 +0200381 break;
382 }
Jens Axboef84d7512006-05-01 19:59:03 +0200383
Jens Axboef6762b72006-05-01 20:02:05 +0200384 atomic = !iov_fault_in_pages_write(iov, chars);
385redo:
386 addr = ops->map(pipe, buf, atomic);
387 error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
388 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200390 /*
391 * Just retry with the slow path if we failed.
392 */
393 if (atomic) {
394 atomic = 0;
395 goto redo;
396 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200397 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200398 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 break;
400 }
401 ret += chars;
402 buf->offset += chars;
403 buf->len -= chars;
404 if (!buf->len) {
405 buf->ops = NULL;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200406 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200407 curbuf = (curbuf + 1) & (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200408 pipe->curbuf = curbuf;
409 pipe->nrbufs = --bufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 do_wakeup = 1;
411 }
412 total_len -= chars;
413 if (!total_len)
414 break; /* common path: read succeeded */
415 }
416 if (bufs) /* More to do? */
417 continue;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200418 if (!pipe->writers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 break;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200420 if (!pipe->waiting_writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 /* syscall merging: Usually we must not sleep
422 * if O_NONBLOCK is set, or if we got some data.
423 * But if a writer sleeps in kernel space, then
424 * we can wait for that data without violating POSIX.
425 */
426 if (ret)
427 break;
428 if (filp->f_flags & O_NONBLOCK) {
429 ret = -EAGAIN;
430 break;
431 }
432 }
433 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200434 if (!ret)
435 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 break;
437 }
438 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200439 wake_up_interruptible_sync(&pipe->wait);
440 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200442 pipe_wait(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200444 mutex_unlock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200445
446 /* Signal writers asynchronously that there is more room. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200448 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200449 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 }
451 if (ret > 0)
452 file_accessed(filp);
453 return ret;
454}
455
456static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700457pipe_write(struct kiocb *iocb, const struct iovec *_iov,
458 unsigned long nr_segs, loff_t ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700460 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800461 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200462 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 ssize_t ret;
464 int do_wakeup;
465 struct iovec *iov = (struct iovec *)_iov;
466 size_t total_len;
467 ssize_t chars;
468
469 total_len = iov_length(iov, nr_segs);
470 /* Null write succeeds. */
471 if (unlikely(total_len == 0))
472 return 0;
473
474 do_wakeup = 0;
475 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200476 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200477 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Ingo Molnar923f4f22006-04-11 13:53:33 +0200479 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 send_sig(SIGPIPE, current, 0);
481 ret = -EPIPE;
482 goto out;
483 }
484
485 /* We try to merge small writes */
486 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200487 if (pipe->nrbufs && chars != 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200488 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
Jens Axboe35f3d142010-05-20 10:43:18 +0200489 (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200490 struct pipe_buffer *buf = pipe->bufs + lastbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800491 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 int offset = buf->offset + buf->len;
Ingo Molnar341b4462006-04-11 13:57:45 +0200493
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
Jens Axboef6762b72006-05-01 20:02:05 +0200495 int error, atomic = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200496 void *addr;
Jens Axboe5274f052006-03-30 15:15:30 +0200497
Jens Axboecac36bb02007-06-14 13:10:48 +0200498 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200499 if (error)
Jens Axboe5274f052006-03-30 15:15:30 +0200500 goto out;
Jens Axboef84d7512006-05-01 19:59:03 +0200501
Jens Axboef6762b72006-05-01 20:02:05 +0200502 iov_fault_in_pages_read(iov, chars);
503redo1:
504 addr = ops->map(pipe, buf, atomic);
Jens Axboe5274f052006-03-30 15:15:30 +0200505 error = pipe_iov_copy_from_user(offset + addr, iov,
Jens Axboef6762b72006-05-01 20:02:05 +0200506 chars, atomic);
507 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 ret = error;
509 do_wakeup = 1;
Jens Axboef6762b72006-05-01 20:02:05 +0200510 if (error) {
511 if (atomic) {
512 atomic = 0;
513 goto redo1;
514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 goto out;
Jens Axboef6762b72006-05-01 20:02:05 +0200516 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 buf->len += chars;
518 total_len -= chars;
519 ret = chars;
520 if (!total_len)
521 goto out;
522 }
523 }
524
525 for (;;) {
526 int bufs;
Ingo Molnar341b4462006-04-11 13:57:45 +0200527
Ingo Molnar923f4f22006-04-11 13:53:33 +0200528 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 send_sig(SIGPIPE, current, 0);
Ingo Molnar341b4462006-04-11 13:57:45 +0200530 if (!ret)
531 ret = -EPIPE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 break;
533 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200534 bufs = pipe->nrbufs;
Jens Axboe35f3d142010-05-20 10:43:18 +0200535 if (bufs < pipe->buffers) {
536 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200537 struct pipe_buffer *buf = pipe->bufs + newbuf;
538 struct page *page = pipe->tmp_page;
Jens Axboef6762b72006-05-01 20:02:05 +0200539 char *src;
540 int error, atomic = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 if (!page) {
543 page = alloc_page(GFP_HIGHUSER);
544 if (unlikely(!page)) {
545 ret = ret ? : -ENOMEM;
546 break;
547 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200548 pipe->tmp_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200550 /* Always wake up, even if the copy fails. Otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 * we lock up (O_NONBLOCK-)readers that sleep due to
552 * syscall merging.
553 * FIXME! Is this really true?
554 */
555 do_wakeup = 1;
556 chars = PAGE_SIZE;
557 if (chars > total_len)
558 chars = total_len;
559
Jens Axboef6762b72006-05-01 20:02:05 +0200560 iov_fault_in_pages_read(iov, chars);
561redo2:
562 if (atomic)
563 src = kmap_atomic(page, KM_USER0);
564 else
565 src = kmap(page);
566
567 error = pipe_iov_copy_from_user(src, iov, chars,
568 atomic);
569 if (atomic)
570 kunmap_atomic(src, KM_USER0);
571 else
572 kunmap(page);
573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200575 if (atomic) {
576 atomic = 0;
577 goto redo2;
578 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200579 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200580 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 break;
582 }
583 ret += chars;
584
585 /* Insert it into the buffer array */
586 buf->page = page;
587 buf->ops = &anon_pipe_buf_ops;
588 buf->offset = 0;
589 buf->len = chars;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200590 pipe->nrbufs = ++bufs;
591 pipe->tmp_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
593 total_len -= chars;
594 if (!total_len)
595 break;
596 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200597 if (bufs < pipe->buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 continue;
599 if (filp->f_flags & O_NONBLOCK) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200600 if (!ret)
601 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 break;
603 }
604 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200605 if (!ret)
606 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 break;
608 }
609 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200610 wake_up_interruptible_sync(&pipe->wait);
611 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 do_wakeup = 0;
613 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200614 pipe->waiting_writers++;
615 pipe_wait(pipe);
616 pipe->waiting_writers--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618out:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200619 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200621 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200622 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 }
624 if (ret > 0)
Christoph Hellwig870f4812006-01-09 20:52:01 -0800625 file_update_time(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 return ret;
627}
628
629static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
631{
632 return -EBADF;
633}
634
635static ssize_t
Ingo Molnar341b4462006-04-11 13:57:45 +0200636bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
637 loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
639 return -EBADF;
640}
641
Andi Kleend59d0b12008-02-08 04:21:23 -0800642static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800644 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200645 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 int count, buf, nrbufs;
647
648 switch (cmd) {
649 case FIONREAD:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200650 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200651 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 count = 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200653 buf = pipe->curbuf;
654 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 while (--nrbufs >= 0) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200656 count += pipe->bufs[buf].len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200657 buf = (buf+1) & (pipe->buffers - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200659 mutex_unlock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 return put_user(count, (int __user *)arg);
662 default:
663 return -EINVAL;
664 }
665}
666
667/* No kernel lock held - fine */
668static unsigned int
669pipe_poll(struct file *filp, poll_table *wait)
670{
671 unsigned int mask;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800672 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200673 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 int nrbufs;
675
Ingo Molnar923f4f22006-04-11 13:53:33 +0200676 poll_wait(filp, &pipe->wait, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678 /* Reading only -- no need for acquiring the semaphore. */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200679 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 mask = 0;
681 if (filp->f_mode & FMODE_READ) {
682 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200683 if (!pipe->writers && filp->f_version != pipe->w_counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 mask |= POLLHUP;
685 }
686
687 if (filp->f_mode & FMODE_WRITE) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200688 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700689 /*
690 * Most Unices do not set POLLERR for FIFOs but on Linux they
691 * behave exactly like pipes for poll().
692 */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200693 if (!pipe->readers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 mask |= POLLERR;
695 }
696
697 return mask;
698}
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700static int
701pipe_release(struct inode *inode, int decr, int decw)
702{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200703 struct pipe_inode_info *pipe;
704
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200705 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200706 pipe = inode->i_pipe;
707 pipe->readers -= decr;
708 pipe->writers -= decw;
Ingo Molnar341b4462006-04-11 13:57:45 +0200709
Ingo Molnar923f4f22006-04-11 13:53:33 +0200710 if (!pipe->readers && !pipe->writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 free_pipe_info(inode);
712 } else {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200713 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200714 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
715 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200717 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719 return 0;
720}
721
722static int
723pipe_read_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_readers);
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_write_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 int retval;
741
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200742 mutex_lock(&inode->i_mutex);
743 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
744 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700746 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
748
749
750static int
751pipe_rdwr_fasync(int fd, struct file *filp, int on)
752{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800753 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar341b4462006-04-11 13:57:45 +0200754 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 int retval;
756
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200757 mutex_lock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200758 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700759 if (retval >= 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200760 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700761 if (retval < 0) /* this can happen only if on == T */
762 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
763 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200764 mutex_unlock(&inode->i_mutex);
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700765 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
768
769static int
770pipe_read_release(struct inode *inode, struct file *filp)
771{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 return pipe_release(inode, 1, 0);
773}
774
775static int
776pipe_write_release(struct inode *inode, struct file *filp)
777{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 return pipe_release(inode, 0, 1);
779}
780
781static int
782pipe_rdwr_release(struct inode *inode, struct file *filp)
783{
784 int decr, decw;
785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 decr = (filp->f_mode & FMODE_READ) != 0;
787 decw = (filp->f_mode & FMODE_WRITE) != 0;
788 return pipe_release(inode, decr, decw);
789}
790
791static int
792pipe_read_open(struct inode *inode, struct file *filp)
793{
Earl Chewad396022009-10-19 15:55:41 -0700794 int ret = -ENOENT;
795
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200796 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700797
798 if (inode->i_pipe) {
799 ret = 0;
800 inode->i_pipe->readers++;
801 }
802
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200803 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Earl Chewad396022009-10-19 15:55:41 -0700805 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
808static int
809pipe_write_open(struct inode *inode, struct file *filp)
810{
Earl Chewad396022009-10-19 15:55:41 -0700811 int ret = -ENOENT;
812
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200813 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700814
815 if (inode->i_pipe) {
816 ret = 0;
817 inode->i_pipe->writers++;
818 }
819
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200820 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Earl Chewad396022009-10-19 15:55:41 -0700822 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823}
824
825static int
826pipe_rdwr_open(struct inode *inode, struct file *filp)
827{
Earl Chewad396022009-10-19 15:55:41 -0700828 int ret = -ENOENT;
829
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200830 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700831
832 if (inode->i_pipe) {
833 ret = 0;
834 if (filp->f_mode & FMODE_READ)
835 inode->i_pipe->readers++;
836 if (filp->f_mode & FMODE_WRITE)
837 inode->i_pipe->writers++;
838 }
839
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200840 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Earl Chewad396022009-10-19 15:55:41 -0700842 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843}
844
845/*
846 * The file_operations structs are not static because they
847 * are also used in linux/fs/fifo.c to do operations on FIFOs.
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200848 *
849 * Pipes reuse fifos' file_operations structs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 */
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200851const struct file_operations read_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700853 .read = do_sync_read,
854 .aio_read = pipe_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 .write = bad_pipe_w,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700856 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800857 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 .open = pipe_read_open,
859 .release = pipe_read_release,
860 .fasync = pipe_read_fasync,
861};
862
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200863const struct file_operations write_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 .llseek = no_llseek,
865 .read = bad_pipe_r,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700866 .write = do_sync_write,
867 .aio_write = pipe_write,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700868 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800869 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 .open = pipe_write_open,
871 .release = pipe_write_release,
872 .fasync = pipe_write_fasync,
873};
874
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200875const struct file_operations rdwr_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700877 .read = do_sync_read,
878 .aio_read = pipe_read,
879 .write = do_sync_write,
880 .aio_write = pipe_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800882 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 .open = pipe_rdwr_open,
884 .release = pipe_rdwr_release,
885 .fasync = pipe_rdwr_fasync,
886};
887
Ingo Molnar3a326a22006-04-10 15:18:35 +0200888struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
889{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200890 struct pipe_inode_info *pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200891
Ingo Molnar923f4f22006-04-11 13:53:33 +0200892 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
893 if (pipe) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200894 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
895 if (pipe->bufs) {
896 init_waitqueue_head(&pipe->wait);
897 pipe->r_counter = pipe->w_counter = 1;
898 pipe->inode = inode;
899 pipe->buffers = PIPE_DEF_BUFFERS;
900 return pipe;
901 }
902 kfree(pipe);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200903 }
904
Jens Axboe35f3d142010-05-20 10:43:18 +0200905 return NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200906}
907
Ingo Molnar923f4f22006-04-11 13:53:33 +0200908void __free_pipe_info(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
910 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Jens Axboe35f3d142010-05-20 10:43:18 +0200912 for (i = 0; i < pipe->buffers; i++) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200913 struct pipe_buffer *buf = pipe->bufs + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 if (buf->ops)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200915 buf->ops->release(pipe, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200917 if (pipe->tmp_page)
918 __free_page(pipe->tmp_page);
Jens Axboe35f3d142010-05-20 10:43:18 +0200919 kfree(pipe->bufs);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200920 kfree(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921}
922
Jens Axboeb92ce552006-04-11 13:52:07 +0200923void free_pipe_info(struct inode *inode)
924{
925 __free_pipe_info(inode->i_pipe);
926 inode->i_pipe = NULL;
927}
928
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800929static struct vfsmount *pipe_mnt __read_mostly;
Ingo Molnar341b4462006-04-11 13:57:45 +0200930
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700931/*
932 * pipefs_dname() is called from d_path().
933 */
934static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
935{
936 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
937 dentry->d_inode->i_ino);
938}
939
Al Viro3ba13d12009-02-20 06:02:22 +0000940static const struct dentry_operations pipefs_dentry_operations = {
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700941 .d_dname = pipefs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942};
943
944static struct inode * get_pipe_inode(void)
945{
946 struct inode *inode = new_inode(pipe_mnt->mnt_sb);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200947 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
949 if (!inode)
950 goto fail_inode;
951
Ingo Molnar923f4f22006-04-11 13:53:33 +0200952 pipe = alloc_pipe_info(inode);
953 if (!pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 goto fail_iput;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200955 inode->i_pipe = pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200956
Ingo Molnar923f4f22006-04-11 13:53:33 +0200957 pipe->readers = pipe->writers = 1;
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200958 inode->i_fop = &rdwr_pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 /*
961 * Mark the inode dirty from the very beginning,
962 * that way it will never be moved to the dirty
963 * list because "mark_inode_dirty()" will think
964 * that it already _is_ on the dirty list.
965 */
966 inode->i_state = I_DIRTY;
967 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
David Howellsda9592e2008-11-14 10:39:05 +1100968 inode->i_uid = current_fsuid();
969 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 return inode;
973
974fail_iput:
975 iput(inode);
Ingo Molnar341b4462006-04-11 13:57:45 +0200976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977fail_inode:
978 return NULL;
979}
980
Ulrich Drepperbe61a862008-07-23 21:29:40 -0700981struct file *create_write_pipe(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
Andi Kleend6cbd282006-09-30 23:29:26 -0700983 int err;
984 struct inode *inode;
985 struct file *f;
Al Viro2c48b9c2009-08-09 00:52:35 +0400986 struct path path;
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700987 struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Andi Kleend6cbd282006-09-30 23:29:26 -0700989 err = -ENFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 inode = get_pipe_inode();
991 if (!inode)
Dave Hansen430e2852008-02-15 14:37:26 -0800992 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Andi Kleend6cbd282006-09-30 23:29:26 -0700994 err = -ENOMEM;
Al Viro2c48b9c2009-08-09 00:52:35 +0400995 path.dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name);
996 if (!path.dentry)
Andi Kleend6cbd282006-09-30 23:29:26 -0700997 goto err_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +0400998 path.mnt = mntget(pipe_mnt);
Ingo Molnar341b4462006-04-11 13:57:45 +0200999
Al Viro2c48b9c2009-08-09 00:52:35 +04001000 path.dentry->d_op = &pipefs_dentry_operations;
Al Viro2c48b9c2009-08-09 00:52:35 +04001001 d_instantiate(path.dentry, inode);
Dave Hansen430e2852008-02-15 14:37:26 -08001002
1003 err = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04001004 f = alloc_file(&path, FMODE_WRITE, &write_pipefifo_fops);
Dave Hansen430e2852008-02-15 14:37:26 -08001005 if (!f)
1006 goto err_dentry;
Andi Kleend6cbd282006-09-30 23:29:26 -07001007 f->f_mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001009 f->f_flags = O_WRONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -07001010 f->f_version = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
Andi Kleend6cbd282006-09-30 23:29:26 -07001012 return f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
Dave Hansen430e2852008-02-15 14:37:26 -08001014 err_dentry:
Al Viroed152432008-04-22 19:51:27 -04001015 free_pipe_info(inode);
Al Viro2c48b9c2009-08-09 00:52:35 +04001016 path_put(&path);
Al Viroed152432008-04-22 19:51:27 -04001017 return ERR_PTR(err);
1018
Andi Kleend6cbd282006-09-30 23:29:26 -07001019 err_inode:
1020 free_pipe_info(inode);
1021 iput(inode);
Dave Hansen430e2852008-02-15 14:37:26 -08001022 err:
Andi Kleend6cbd282006-09-30 23:29:26 -07001023 return ERR_PTR(err);
1024}
1025
1026void free_write_pipe(struct file *f)
1027{
Al Viro5ccac88e2006-12-18 13:31:18 +00001028 free_pipe_info(f->f_dentry->d_inode);
Jan Blunckc8e7f442008-06-09 16:40:35 -07001029 path_put(&f->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001030 put_filp(f);
1031}
1032
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001033struct file *create_read_pipe(struct file *wrf, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001034{
Al Virod2314122009-08-09 01:01:37 +04001035 /* Grab pipe from the writer */
1036 struct file *f = alloc_file(&wrf->f_path, FMODE_READ,
1037 &read_pipefifo_fops);
Andi Kleend6cbd282006-09-30 23:29:26 -07001038 if (!f)
1039 return ERR_PTR(-ENFILE);
1040
Jan Blunckc8e7f442008-06-09 16:40:35 -07001041 path_get(&wrf->f_path);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001042 f->f_flags = O_RDONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -07001043
1044 return f;
1045}
1046
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001047int do_pipe_flags(int *fd, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001048{
1049 struct file *fw, *fr;
1050 int error;
1051 int fdw, fdr;
1052
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001053 if (flags & ~(O_CLOEXEC | O_NONBLOCK))
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001054 return -EINVAL;
1055
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001056 fw = create_write_pipe(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001057 if (IS_ERR(fw))
1058 return PTR_ERR(fw);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001059 fr = create_read_pipe(fw, flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001060 error = PTR_ERR(fr);
1061 if (IS_ERR(fr))
1062 goto err_write_pipe;
1063
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001064 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001065 if (error < 0)
1066 goto err_read_pipe;
1067 fdr = error;
1068
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001069 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001070 if (error < 0)
1071 goto err_fdr;
1072 fdw = error;
1073
Al Viro157cf642008-12-14 04:57:47 -05001074 audit_fd_pair(fdr, fdw);
Andi Kleend6cbd282006-09-30 23:29:26 -07001075 fd_install(fdr, fr);
1076 fd_install(fdw, fw);
1077 fd[0] = fdr;
1078 fd[1] = fdw;
Ingo Molnar341b4462006-04-11 13:57:45 +02001079
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 return 0;
1081
Andi Kleend6cbd282006-09-30 23:29:26 -07001082 err_fdr:
1083 put_unused_fd(fdr);
1084 err_read_pipe:
Jan Blunckc8e7f442008-06-09 16:40:35 -07001085 path_put(&fr->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001086 put_filp(fr);
1087 err_write_pipe:
1088 free_write_pipe(fw);
1089 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090}
1091
1092/*
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001093 * sys_pipe() is the normal C calling standard for creating
1094 * a pipe. It's not the way Unix traditionally does this, though.
1095 */
Heiko Carstensd4e82042009-01-14 14:14:34 +01001096SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001097{
1098 int fd[2];
1099 int error;
1100
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001101 error = do_pipe_flags(fd, flags);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001102 if (!error) {
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001103 if (copy_to_user(fildes, fd, sizeof(fd))) {
1104 sys_close(fd[0]);
1105 sys_close(fd[1]);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001106 error = -EFAULT;
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001107 }
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001108 }
1109 return error;
1110}
1111
Heiko Carstens2b664212009-01-14 14:14:35 +01001112SYSCALL_DEFINE1(pipe, int __user *, fildes)
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001113{
1114 return sys_pipe2(fildes, 0);
1115}
1116
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001117/*
Jens Axboe35f3d142010-05-20 10:43:18 +02001118 * Allocate a new array of pipe buffers and copy the info over. Returns the
1119 * pipe size if successful, or return -ERROR on error.
1120 */
1121static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1122{
1123 struct pipe_buffer *bufs;
1124
1125 /*
1126 * Must be a power-of-2 currently
1127 */
1128 if (!is_power_of_2(arg))
1129 return -EINVAL;
1130
1131 /*
1132 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1133 * expect a lot of shrink+grow operations, just free and allocate
1134 * again like we would do for growing. If the pipe currently
1135 * contains more buffers than arg, then return busy.
1136 */
1137 if (arg < pipe->nrbufs)
1138 return -EBUSY;
1139
1140 bufs = kcalloc(arg, sizeof(struct pipe_buffer), GFP_KERNEL);
1141 if (unlikely(!bufs))
1142 return -ENOMEM;
1143
1144 /*
1145 * The pipe array wraps around, so just start the new one at zero
1146 * and adjust the indexes.
1147 */
1148 if (pipe->nrbufs) {
1149 const unsigned int tail = pipe->nrbufs & (pipe->buffers - 1);
1150 const unsigned int head = pipe->nrbufs - tail;
1151
1152 if (head)
1153 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1154 if (tail)
1155 memcpy(bufs + head, pipe->bufs + pipe->curbuf, tail * sizeof(struct pipe_buffer));
1156 }
1157
1158 pipe->curbuf = 0;
1159 kfree(pipe->bufs);
1160 pipe->bufs = bufs;
1161 pipe->buffers = arg;
1162 return arg;
1163}
1164
1165long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1166{
1167 struct pipe_inode_info *pipe;
1168 long ret;
1169
1170 pipe = file->f_path.dentry->d_inode->i_pipe;
1171 if (!pipe)
1172 return -EBADF;
1173
1174 mutex_lock(&pipe->inode->i_mutex);
1175
1176 switch (cmd) {
1177 case F_SETPIPE_SZ:
Julia Lawallcc967be2010-05-26 17:54:39 +02001178 if (!capable(CAP_SYS_ADMIN) && arg > pipe_max_pages) {
1179 ret = -EINVAL;
1180 goto out;
1181 }
Jens Axboeb492e952010-05-19 21:03:16 +02001182 /*
1183 * The pipe needs to be at least 2 pages large to
1184 * guarantee POSIX behaviour.
1185 */
Julia Lawallcc967be2010-05-26 17:54:39 +02001186 if (arg < 2) {
1187 ret = -EINVAL;
1188 goto out;
1189 }
Jens Axboe35f3d142010-05-20 10:43:18 +02001190 ret = pipe_set_size(pipe, arg);
1191 break;
1192 case F_GETPIPE_SZ:
1193 ret = pipe->buffers;
1194 break;
1195 default:
1196 ret = -EINVAL;
1197 break;
1198 }
1199
Julia Lawallcc967be2010-05-26 17:54:39 +02001200out:
Jens Axboe35f3d142010-05-20 10:43:18 +02001201 mutex_unlock(&pipe->inode->i_mutex);
1202 return ret;
1203}
1204
1205/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 * pipefs should _never_ be mounted by userland - too much of security hassle,
1207 * no real gain from having the whole whorehouse mounted. So we don't need
1208 * any operations on the root directory. However, we need a non-trivial
1209 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1210 */
David Howells454e2392006-06-23 02:02:57 -07001211static int pipefs_get_sb(struct file_system_type *fs_type,
1212 int flags, const char *dev_name, void *data,
1213 struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214{
David Howells454e2392006-06-23 02:02:57 -07001215 return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216}
1217
1218static struct file_system_type pipe_fs_type = {
1219 .name = "pipefs",
1220 .get_sb = pipefs_get_sb,
1221 .kill_sb = kill_anon_super,
1222};
1223
1224static int __init init_pipe_fs(void)
1225{
1226 int err = register_filesystem(&pipe_fs_type);
Ingo Molnar341b4462006-04-11 13:57:45 +02001227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (!err) {
1229 pipe_mnt = kern_mount(&pipe_fs_type);
1230 if (IS_ERR(pipe_mnt)) {
1231 err = PTR_ERR(pipe_mnt);
1232 unregister_filesystem(&pipe_fs_type);
1233 }
1234 }
1235 return err;
1236}
1237
1238static void __exit exit_pipe_fs(void)
1239{
1240 unregister_filesystem(&pipe_fs_type);
1241 mntput(pipe_mnt);
1242}
1243
1244fs_initcall(init_pipe_fs);
1245module_exit(exit_pipe_fs);