blob: d2d7566ce68e5cb4fc9713e531670d8f05d52d02 [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
Jens Axboeff9da692010-06-03 14:54:39 +020029 * be set by root in /proc/sys/fs/pipe-max-size
Jens Axboeb492e952010-05-19 21:03:16 +020030 */
Jens Axboeff9da692010-06-03 14:54:39 +020031unsigned int pipe_max_size = 1048576;
32
33/*
34 * Minimum pipe size, as required by POSIX
35 */
36unsigned int pipe_min_size = PAGE_SIZE;
Jens Axboeb492e952010-05-19 21:03:16 +020037
38/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 * We use a start+len construction, which provides full use of the
40 * allocated memory.
41 * -- Florian Coosmann (FGC)
42 *
43 * Reads with count = 0 should always return 0.
44 * -- Julian Bradfield 1999-06-07.
45 *
46 * FIFOs and Pipes now generate SIGIO for both readers and writers.
47 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
48 *
49 * pipe_read & write cleanup
50 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
51 */
52
Miklos Szeredi61e0d472009-04-14 19:48:41 +020053static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
54{
55 if (pipe->inode)
56 mutex_lock_nested(&pipe->inode->i_mutex, subclass);
57}
58
59void pipe_lock(struct pipe_inode_info *pipe)
60{
61 /*
62 * pipe_lock() nests non-pipe inode locks (for writing to a file)
63 */
64 pipe_lock_nested(pipe, I_MUTEX_PARENT);
65}
66EXPORT_SYMBOL(pipe_lock);
67
68void pipe_unlock(struct pipe_inode_info *pipe)
69{
70 if (pipe->inode)
71 mutex_unlock(&pipe->inode->i_mutex);
72}
73EXPORT_SYMBOL(pipe_unlock);
74
75void pipe_double_lock(struct pipe_inode_info *pipe1,
76 struct pipe_inode_info *pipe2)
77{
78 BUG_ON(pipe1 == pipe2);
79
80 if (pipe1 < pipe2) {
81 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
82 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
83 } else {
Peter Zijlstra023d43c2009-07-21 10:09:23 +020084 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
85 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020086 }
87}
88
Linus Torvalds1da177e2005-04-16 15:20:36 -070089/* Drop the inode semaphore and wait for a pipe event, atomically */
Ingo Molnar3a326a22006-04-10 15:18:35 +020090void pipe_wait(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
92 DEFINE_WAIT(wait);
93
Ingo Molnard79fc0f2005-09-10 00:26:12 -070094 /*
95 * Pipes are system-local resources, so sleeping on them
96 * is considered a noninteractive wait:
97 */
Mike Galbraithaf927232007-10-15 17:00:13 +020098 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
Miklos Szeredi61e0d472009-04-14 19:48:41 +020099 pipe_unlock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 schedule();
Ingo Molnar3a326a22006-04-10 15:18:35 +0200101 finish_wait(&pipe->wait, &wait);
Miklos Szeredi61e0d472009-04-14 19:48:41 +0200102 pipe_lock(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103}
104
Arjan van de Ven858119e2006-01-14 13:20:43 -0800105static int
Jens Axboef6762b72006-05-01 20:02:05 +0200106pipe_iov_copy_from_user(void *to, struct iovec *iov, unsigned long len,
107 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
109 unsigned long copy;
110
111 while (len > 0) {
112 while (!iov->iov_len)
113 iov++;
114 copy = min_t(unsigned long, len, iov->iov_len);
115
Jens Axboef6762b72006-05-01 20:02:05 +0200116 if (atomic) {
117 if (__copy_from_user_inatomic(to, iov->iov_base, copy))
118 return -EFAULT;
119 } else {
120 if (copy_from_user(to, iov->iov_base, copy))
121 return -EFAULT;
122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 to += copy;
124 len -= copy;
125 iov->iov_base += copy;
126 iov->iov_len -= copy;
127 }
128 return 0;
129}
130
Arjan van de Ven858119e2006-01-14 13:20:43 -0800131static int
Jens Axboef6762b72006-05-01 20:02:05 +0200132pipe_iov_copy_to_user(struct iovec *iov, const void *from, unsigned long len,
133 int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134{
135 unsigned long copy;
136
137 while (len > 0) {
138 while (!iov->iov_len)
139 iov++;
140 copy = min_t(unsigned long, len, iov->iov_len);
141
Jens Axboef6762b72006-05-01 20:02:05 +0200142 if (atomic) {
143 if (__copy_to_user_inatomic(iov->iov_base, from, copy))
144 return -EFAULT;
145 } else {
146 if (copy_to_user(iov->iov_base, from, copy))
147 return -EFAULT;
148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 from += copy;
150 len -= copy;
151 iov->iov_base += copy;
152 iov->iov_len -= copy;
153 }
154 return 0;
155}
156
Jens Axboef6762b72006-05-01 20:02:05 +0200157/*
158 * Attempt to pre-fault in the user memory, so we can use atomic copies.
159 * Returns the number of bytes not faulted in.
160 */
161static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
162{
163 while (!iov->iov_len)
164 iov++;
165
166 while (len > 0) {
167 unsigned long this_len;
168
169 this_len = min_t(unsigned long, len, iov->iov_len);
170 if (fault_in_pages_writeable(iov->iov_base, this_len))
171 break;
172
173 len -= this_len;
174 iov++;
175 }
176
177 return len;
178}
179
180/*
181 * Pre-fault in the user memory, so we can use atomic copies.
182 */
183static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
184{
185 while (!iov->iov_len)
186 iov++;
187
188 while (len > 0) {
189 unsigned long this_len;
190
191 this_len = min_t(unsigned long, len, iov->iov_len);
192 fault_in_pages_readable(iov->iov_base, this_len);
193 len -= this_len;
194 iov++;
195 }
196}
197
Ingo Molnar341b4462006-04-11 13:57:45 +0200198static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
199 struct pipe_buffer *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
201 struct page *page = buf->page;
202
Jens Axboe5274f052006-03-30 15:15:30 +0200203 /*
204 * If nobody else uses this page, and we don't already have a
205 * temporary page, let's keep track of it as a one-deep
Ingo Molnar341b4462006-04-11 13:57:45 +0200206 * allocation cache. (Otherwise just release our reference to it)
Jens Axboe5274f052006-03-30 15:15:30 +0200207 */
Ingo Molnar341b4462006-04-11 13:57:45 +0200208 if (page_count(page) == 1 && !pipe->tmp_page)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200209 pipe->tmp_page = page;
Ingo Molnar341b4462006-04-11 13:57:45 +0200210 else
211 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
213
Jens Axboe08457182007-06-12 20:51:32 +0200214/**
215 * generic_pipe_buf_map - virtually map a pipe buffer
216 * @pipe: the pipe that the buffer belongs to
217 * @buf: the buffer that should be mapped
218 * @atomic: whether to use an atomic map
219 *
220 * Description:
221 * This function returns a kernel virtual address mapping for the
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800222 * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
Jens Axboe08457182007-06-12 20:51:32 +0200223 * and the caller has to be careful not to fault before calling
224 * the unmap function.
225 *
226 * Note that this function occupies KM_USER0 if @atomic != 0.
227 */
Jens Axboef84d7512006-05-01 19:59:03 +0200228void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200229 struct pipe_buffer *buf, int atomic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
Jens Axboef6762b72006-05-01 20:02:05 +0200231 if (atomic) {
232 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
233 return kmap_atomic(buf->page, KM_USER0);
234 }
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 return kmap(buf->page);
237}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200238EXPORT_SYMBOL(generic_pipe_buf_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Jens Axboe08457182007-06-12 20:51:32 +0200240/**
241 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
242 * @pipe: the pipe that the buffer belongs to
243 * @buf: the buffer that should be unmapped
244 * @map_data: the data that the mapping function returned
245 *
246 * Description:
247 * This function undoes the mapping that ->map() provided.
248 */
Jens Axboef84d7512006-05-01 19:59:03 +0200249void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
Jens Axboef6762b72006-05-01 20:02:05 +0200250 struct pipe_buffer *buf, void *map_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
Jens Axboef6762b72006-05-01 20:02:05 +0200252 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
253 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
254 kunmap_atomic(map_data, KM_USER0);
255 } else
256 kunmap(buf->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200258EXPORT_SYMBOL(generic_pipe_buf_unmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Jens Axboe08457182007-06-12 20:51:32 +0200260/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800261 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200262 * @pipe: the pipe that the buffer belongs to
263 * @buf: the buffer to attempt to steal
264 *
265 * Description:
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800266 * This function attempts to steal the &struct page attached to
Jens Axboe08457182007-06-12 20:51:32 +0200267 * @buf. If successful, this function returns 0 and returns with
268 * the page locked. The caller may then reuse the page for whatever
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800269 * he wishes; the typical use is insertion into a different file
Jens Axboe08457182007-06-12 20:51:32 +0200270 * page cache.
271 */
Jens Axboe330ab712006-05-02 15:29:57 +0200272int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
273 struct pipe_buffer *buf)
Jens Axboe5abc97a2006-03-30 15:16:46 +0200274{
Jens Axboe46e678c2006-04-30 16:36:32 +0200275 struct page *page = buf->page;
276
Jens Axboe08457182007-06-12 20:51:32 +0200277 /*
278 * A reference of one is golden, that means that the owner of this
279 * page is the only one holding a reference to it. lock the page
280 * and return OK.
281 */
Jens Axboe46e678c2006-04-30 16:36:32 +0200282 if (page_count(page) == 1) {
Jens Axboe46e678c2006-04-30 16:36:32 +0200283 lock_page(page);
284 return 0;
285 }
286
287 return 1;
Jens Axboe5abc97a2006-03-30 15:16:46 +0200288}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200289EXPORT_SYMBOL(generic_pipe_buf_steal);
Jens Axboe5abc97a2006-03-30 15:16:46 +0200290
Jens Axboe08457182007-06-12 20:51:32 +0200291/**
Randy Dunlapb51d63c2008-02-13 15:03:22 -0800292 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
Jens Axboe08457182007-06-12 20:51:32 +0200293 * @pipe: the pipe that the buffer belongs to
294 * @buf: the buffer to get a reference to
295 *
296 * Description:
297 * This function grabs an extra reference to @buf. It's used in
298 * in the tee() system call, when we duplicate the buffers in one
299 * pipe into another.
300 */
301void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
Jens Axboe70524492006-04-11 15:51:17 +0200302{
303 page_cache_get(buf->page);
304}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200305EXPORT_SYMBOL(generic_pipe_buf_get);
Jens Axboe70524492006-04-11 15:51:17 +0200306
Jens Axboe08457182007-06-12 20:51:32 +0200307/**
308 * generic_pipe_buf_confirm - verify contents of the pipe buffer
Randy Dunlap79685b82007-07-27 08:08:51 +0200309 * @info: the pipe that the buffer belongs to
Jens Axboe08457182007-06-12 20:51:32 +0200310 * @buf: the buffer to confirm
311 *
312 * Description:
313 * This function does nothing, because the generic pipe code uses
314 * pages that are always good when inserted into the pipe.
315 */
Jens Axboecac36bb02007-06-14 13:10:48 +0200316int generic_pipe_buf_confirm(struct pipe_inode_info *info,
317 struct pipe_buffer *buf)
Jens Axboef84d7512006-05-01 19:59:03 +0200318{
319 return 0;
320}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200321EXPORT_SYMBOL(generic_pipe_buf_confirm);
Jens Axboef84d7512006-05-01 19:59:03 +0200322
Miklos Szeredi68181732009-05-07 15:37:36 +0200323/**
324 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
325 * @pipe: the pipe that the buffer belongs to
326 * @buf: the buffer to put a reference to
327 *
328 * Description:
329 * This function releases a reference to @buf.
330 */
331void generic_pipe_buf_release(struct pipe_inode_info *pipe,
332 struct pipe_buffer *buf)
333{
334 page_cache_release(buf->page);
335}
Miklos Szeredi51921cb2010-05-26 08:44:22 +0200336EXPORT_SYMBOL(generic_pipe_buf_release);
Miklos Szeredi68181732009-05-07 15:37:36 +0200337
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800338static const struct pipe_buf_operations anon_pipe_buf_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 .can_merge = 1,
Jens Axboef84d7512006-05-01 19:59:03 +0200340 .map = generic_pipe_buf_map,
341 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +0200342 .confirm = generic_pipe_buf_confirm,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 .release = anon_pipe_buf_release,
Jens Axboe330ab712006-05-02 15:29:57 +0200344 .steal = generic_pipe_buf_steal,
Jens Axboef84d7512006-05-01 19:59:03 +0200345 .get = generic_pipe_buf_get,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346};
347
348static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700349pipe_read(struct kiocb *iocb, const struct iovec *_iov,
350 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700352 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800353 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200354 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 int do_wakeup;
356 ssize_t ret;
357 struct iovec *iov = (struct iovec *)_iov;
358 size_t total_len;
359
360 total_len = iov_length(iov, nr_segs);
361 /* Null read succeeds. */
362 if (unlikely(total_len == 0))
363 return 0;
364
365 do_wakeup = 0;
366 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200367 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200368 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 for (;;) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200370 int bufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 if (bufs) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200372 int curbuf = pipe->curbuf;
373 struct pipe_buffer *buf = pipe->bufs + curbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800374 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 void *addr;
376 size_t chars = buf->len;
Jens Axboef6762b72006-05-01 20:02:05 +0200377 int error, atomic;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379 if (chars > total_len)
380 chars = total_len;
381
Jens Axboecac36bb02007-06-14 13:10:48 +0200382 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200383 if (error) {
Jens Axboe5274f052006-03-30 15:15:30 +0200384 if (!ret)
Nicolas Kaisere5953cb2010-10-21 14:56:00 +0200385 ret = error;
Jens Axboe5274f052006-03-30 15:15:30 +0200386 break;
387 }
Jens Axboef84d7512006-05-01 19:59:03 +0200388
Jens Axboef6762b72006-05-01 20:02:05 +0200389 atomic = !iov_fault_in_pages_write(iov, chars);
390redo:
391 addr = ops->map(pipe, buf, atomic);
392 error = pipe_iov_copy_to_user(iov, addr + buf->offset, chars, atomic);
393 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200395 /*
396 * Just retry with the slow path if we failed.
397 */
398 if (atomic) {
399 atomic = 0;
400 goto redo;
401 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200402 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200403 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 break;
405 }
406 ret += chars;
407 buf->offset += chars;
408 buf->len -= chars;
409 if (!buf->len) {
410 buf->ops = NULL;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200411 ops->release(pipe, buf);
Jens Axboe35f3d142010-05-20 10:43:18 +0200412 curbuf = (curbuf + 1) & (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200413 pipe->curbuf = curbuf;
414 pipe->nrbufs = --bufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 do_wakeup = 1;
416 }
417 total_len -= chars;
418 if (!total_len)
419 break; /* common path: read succeeded */
420 }
421 if (bufs) /* More to do? */
422 continue;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200423 if (!pipe->writers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 break;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200425 if (!pipe->waiting_writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 /* syscall merging: Usually we must not sleep
427 * if O_NONBLOCK is set, or if we got some data.
428 * But if a writer sleeps in kernel space, then
429 * we can wait for that data without violating POSIX.
430 */
431 if (ret)
432 break;
433 if (filp->f_flags & O_NONBLOCK) {
434 ret = -EAGAIN;
435 break;
436 }
437 }
438 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200439 if (!ret)
440 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 break;
442 }
443 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200444 wake_up_interruptible_sync(&pipe->wait);
445 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200447 pipe_wait(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200449 mutex_unlock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200450
451 /* Signal writers asynchronously that there is more room. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200453 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200454 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
456 if (ret > 0)
457 file_accessed(filp);
458 return ret;
459}
460
461static ssize_t
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700462pipe_write(struct kiocb *iocb, const struct iovec *_iov,
463 unsigned long nr_segs, loff_t ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700465 struct file *filp = iocb->ki_filp;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800466 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200467 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 ssize_t ret;
469 int do_wakeup;
470 struct iovec *iov = (struct iovec *)_iov;
471 size_t total_len;
472 ssize_t chars;
473
474 total_len = iov_length(iov, nr_segs);
475 /* Null write succeeds. */
476 if (unlikely(total_len == 0))
477 return 0;
478
479 do_wakeup = 0;
480 ret = 0;
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200481 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200482 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Ingo Molnar923f4f22006-04-11 13:53:33 +0200484 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 send_sig(SIGPIPE, current, 0);
486 ret = -EPIPE;
487 goto out;
488 }
489
490 /* We try to merge small writes */
491 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200492 if (pipe->nrbufs && chars != 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200493 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
Jens Axboe35f3d142010-05-20 10:43:18 +0200494 (pipe->buffers - 1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200495 struct pipe_buffer *buf = pipe->bufs + lastbuf;
Eric Dumazetd4c3cca2006-12-13 00:34:04 -0800496 const struct pipe_buf_operations *ops = buf->ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 int offset = buf->offset + buf->len;
Ingo Molnar341b4462006-04-11 13:57:45 +0200498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
Jens Axboef6762b72006-05-01 20:02:05 +0200500 int error, atomic = 1;
Jens Axboe5274f052006-03-30 15:15:30 +0200501 void *addr;
Jens Axboe5274f052006-03-30 15:15:30 +0200502
Jens Axboecac36bb02007-06-14 13:10:48 +0200503 error = ops->confirm(pipe, buf);
Jens Axboef84d7512006-05-01 19:59:03 +0200504 if (error)
Jens Axboe5274f052006-03-30 15:15:30 +0200505 goto out;
Jens Axboef84d7512006-05-01 19:59:03 +0200506
Jens Axboef6762b72006-05-01 20:02:05 +0200507 iov_fault_in_pages_read(iov, chars);
508redo1:
509 addr = ops->map(pipe, buf, atomic);
Jens Axboe5274f052006-03-30 15:15:30 +0200510 error = pipe_iov_copy_from_user(offset + addr, iov,
Jens Axboef6762b72006-05-01 20:02:05 +0200511 chars, atomic);
512 ops->unmap(pipe, buf, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 ret = error;
514 do_wakeup = 1;
Jens Axboef6762b72006-05-01 20:02:05 +0200515 if (error) {
516 if (atomic) {
517 atomic = 0;
518 goto redo1;
519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 goto out;
Jens Axboef6762b72006-05-01 20:02:05 +0200521 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 buf->len += chars;
523 total_len -= chars;
524 ret = chars;
525 if (!total_len)
526 goto out;
527 }
528 }
529
530 for (;;) {
531 int bufs;
Ingo Molnar341b4462006-04-11 13:57:45 +0200532
Ingo Molnar923f4f22006-04-11 13:53:33 +0200533 if (!pipe->readers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 send_sig(SIGPIPE, current, 0);
Ingo Molnar341b4462006-04-11 13:57:45 +0200535 if (!ret)
536 ret = -EPIPE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 break;
538 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200539 bufs = pipe->nrbufs;
Jens Axboe35f3d142010-05-20 10:43:18 +0200540 if (bufs < pipe->buffers) {
541 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200542 struct pipe_buffer *buf = pipe->bufs + newbuf;
543 struct page *page = pipe->tmp_page;
Jens Axboef6762b72006-05-01 20:02:05 +0200544 char *src;
545 int error, atomic = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
547 if (!page) {
548 page = alloc_page(GFP_HIGHUSER);
549 if (unlikely(!page)) {
550 ret = ret ? : -ENOMEM;
551 break;
552 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200553 pipe->tmp_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200555 /* Always wake up, even if the copy fails. Otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 * we lock up (O_NONBLOCK-)readers that sleep due to
557 * syscall merging.
558 * FIXME! Is this really true?
559 */
560 do_wakeup = 1;
561 chars = PAGE_SIZE;
562 if (chars > total_len)
563 chars = total_len;
564
Jens Axboef6762b72006-05-01 20:02:05 +0200565 iov_fault_in_pages_read(iov, chars);
566redo2:
567 if (atomic)
568 src = kmap_atomic(page, KM_USER0);
569 else
570 src = kmap(page);
571
572 error = pipe_iov_copy_from_user(src, iov, chars,
573 atomic);
574 if (atomic)
575 kunmap_atomic(src, KM_USER0);
576 else
577 kunmap(page);
578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 if (unlikely(error)) {
Jens Axboef6762b72006-05-01 20:02:05 +0200580 if (atomic) {
581 atomic = 0;
582 goto redo2;
583 }
Ingo Molnar341b4462006-04-11 13:57:45 +0200584 if (!ret)
Jens Axboef6762b72006-05-01 20:02:05 +0200585 ret = error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 break;
587 }
588 ret += chars;
589
590 /* Insert it into the buffer array */
591 buf->page = page;
592 buf->ops = &anon_pipe_buf_ops;
593 buf->offset = 0;
594 buf->len = chars;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200595 pipe->nrbufs = ++bufs;
596 pipe->tmp_page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597
598 total_len -= chars;
599 if (!total_len)
600 break;
601 }
Jens Axboe35f3d142010-05-20 10:43:18 +0200602 if (bufs < pipe->buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 continue;
604 if (filp->f_flags & O_NONBLOCK) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200605 if (!ret)
606 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 break;
608 }
609 if (signal_pending(current)) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200610 if (!ret)
611 ret = -ERESTARTSYS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 break;
613 }
614 if (do_wakeup) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200615 wake_up_interruptible_sync(&pipe->wait);
616 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 do_wakeup = 0;
618 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200619 pipe->waiting_writers++;
620 pipe_wait(pipe);
621 pipe->waiting_writers--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 }
623out:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200624 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 if (do_wakeup) {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200626 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200627 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
629 if (ret > 0)
Christoph Hellwig870f4812006-01-09 20:52:01 -0800630 file_update_time(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 return ret;
632}
633
634static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
636{
637 return -EBADF;
638}
639
640static ssize_t
Ingo Molnar341b4462006-04-11 13:57:45 +0200641bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
642 loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
644 return -EBADF;
645}
646
Andi Kleend59d0b12008-02-08 04:21:23 -0800647static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800649 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200650 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 int count, buf, nrbufs;
652
653 switch (cmd) {
654 case FIONREAD:
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200655 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200656 pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 count = 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200658 buf = pipe->curbuf;
659 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 while (--nrbufs >= 0) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200661 count += pipe->bufs[buf].len;
Jens Axboe35f3d142010-05-20 10:43:18 +0200662 buf = (buf+1) & (pipe->buffers - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200664 mutex_unlock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return put_user(count, (int __user *)arg);
667 default:
668 return -EINVAL;
669 }
670}
671
672/* No kernel lock held - fine */
673static unsigned int
674pipe_poll(struct file *filp, poll_table *wait)
675{
676 unsigned int mask;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800677 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200678 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 int nrbufs;
680
Ingo Molnar923f4f22006-04-11 13:53:33 +0200681 poll_wait(filp, &pipe->wait, wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
683 /* Reading only -- no need for acquiring the semaphore. */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200684 nrbufs = pipe->nrbufs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 mask = 0;
686 if (filp->f_mode & FMODE_READ) {
687 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200688 if (!pipe->writers && filp->f_version != pipe->w_counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 mask |= POLLHUP;
690 }
691
692 if (filp->f_mode & FMODE_WRITE) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200693 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700694 /*
695 * Most Unices do not set POLLERR for FIFOs but on Linux they
696 * behave exactly like pipes for poll().
697 */
Ingo Molnar923f4f22006-04-11 13:53:33 +0200698 if (!pipe->readers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 mask |= POLLERR;
700 }
701
702 return mask;
703}
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705static int
706pipe_release(struct inode *inode, int decr, int decw)
707{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200708 struct pipe_inode_info *pipe;
709
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200710 mutex_lock(&inode->i_mutex);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200711 pipe = inode->i_pipe;
712 pipe->readers -= decr;
713 pipe->writers -= decw;
Ingo Molnar341b4462006-04-11 13:57:45 +0200714
Ingo Molnar923f4f22006-04-11 13:53:33 +0200715 if (!pipe->readers && !pipe->writers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 free_pipe_info(inode);
717 } else {
Ingo Molnar71e20f12007-10-15 17:00:19 +0200718 wake_up_interruptible_sync(&pipe->wait);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200719 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
720 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200722 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
724 return 0;
725}
726
727static int
728pipe_read_fasync(int fd, struct file *filp, int on)
729{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800730 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 int retval;
732
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200733 mutex_lock(&inode->i_mutex);
734 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
735 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700737 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738}
739
740
741static int
742pipe_write_fasync(int fd, struct file *filp, int on)
743{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800744 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 int retval;
746
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200747 mutex_lock(&inode->i_mutex);
748 retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
749 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700751 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
754
755static int
756pipe_rdwr_fasync(int fd, struct file *filp, int on)
757{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800758 struct inode *inode = filp->f_path.dentry->d_inode;
Ingo Molnar341b4462006-04-11 13:57:45 +0200759 struct pipe_inode_info *pipe = inode->i_pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 int retval;
761
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200762 mutex_lock(&inode->i_mutex);
Ingo Molnar341b4462006-04-11 13:57:45 +0200763 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700764 if (retval >= 0) {
Ingo Molnar341b4462006-04-11 13:57:45 +0200765 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
Oleg Nesterove5bc49b2009-03-12 14:31:28 -0700766 if (retval < 0) /* this can happen only if on == T */
767 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
768 }
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200769 mutex_unlock(&inode->i_mutex);
Jonathan Corbet60aa4922009-02-01 14:52:56 -0700770 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771}
772
773
774static int
775pipe_read_release(struct inode *inode, struct file *filp)
776{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 return pipe_release(inode, 1, 0);
778}
779
780static int
781pipe_write_release(struct inode *inode, struct file *filp)
782{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 return pipe_release(inode, 0, 1);
784}
785
786static int
787pipe_rdwr_release(struct inode *inode, struct file *filp)
788{
789 int decr, decw;
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 decr = (filp->f_mode & FMODE_READ) != 0;
792 decw = (filp->f_mode & FMODE_WRITE) != 0;
793 return pipe_release(inode, decr, decw);
794}
795
796static int
797pipe_read_open(struct inode *inode, struct file *filp)
798{
Earl Chewad396022009-10-19 15:55:41 -0700799 int ret = -ENOENT;
800
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200801 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700802
803 if (inode->i_pipe) {
804 ret = 0;
805 inode->i_pipe->readers++;
806 }
807
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200808 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Earl Chewad396022009-10-19 15:55:41 -0700810 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
813static int
814pipe_write_open(struct inode *inode, struct file *filp)
815{
Earl Chewad396022009-10-19 15:55:41 -0700816 int ret = -ENOENT;
817
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200818 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700819
820 if (inode->i_pipe) {
821 ret = 0;
822 inode->i_pipe->writers++;
823 }
824
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200825 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
Earl Chewad396022009-10-19 15:55:41 -0700827 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828}
829
830static int
831pipe_rdwr_open(struct inode *inode, struct file *filp)
832{
Earl Chewad396022009-10-19 15:55:41 -0700833 int ret = -ENOENT;
834
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200835 mutex_lock(&inode->i_mutex);
Earl Chewad396022009-10-19 15:55:41 -0700836
837 if (inode->i_pipe) {
838 ret = 0;
839 if (filp->f_mode & FMODE_READ)
840 inode->i_pipe->readers++;
841 if (filp->f_mode & FMODE_WRITE)
842 inode->i_pipe->writers++;
843 }
844
Ingo Molnar9aeedfc2006-04-11 13:53:10 +0200845 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Earl Chewad396022009-10-19 15:55:41 -0700847 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848}
849
850/*
851 * The file_operations structs are not static because they
852 * are also used in linux/fs/fifo.c to do operations on FIFOs.
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200853 *
854 * Pipes reuse fifos' file_operations structs.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 */
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200856const struct file_operations read_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700858 .read = do_sync_read,
859 .aio_read = pipe_read,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 .write = bad_pipe_w,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700861 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800862 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 .open = pipe_read_open,
864 .release = pipe_read_release,
865 .fasync = pipe_read_fasync,
866};
867
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200868const struct file_operations write_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 .llseek = no_llseek,
870 .read = bad_pipe_r,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700871 .write = do_sync_write,
872 .aio_write = pipe_write,
Pekka Enberg5e5d7a22005-09-06 15:17:48 -0700873 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800874 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 .open = pipe_write_open,
876 .release = pipe_write_release,
877 .fasync = pipe_write_fasync,
878};
879
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200880const struct file_operations rdwr_pipefifo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -0700882 .read = do_sync_read,
883 .aio_read = pipe_read,
884 .write = do_sync_write,
885 .aio_write = pipe_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 .poll = pipe_poll,
Andi Kleend59d0b12008-02-08 04:21:23 -0800887 .unlocked_ioctl = pipe_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 .open = pipe_rdwr_open,
889 .release = pipe_rdwr_release,
890 .fasync = pipe_rdwr_fasync,
891};
892
Ingo Molnar3a326a22006-04-10 15:18:35 +0200893struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
894{
Ingo Molnar923f4f22006-04-11 13:53:33 +0200895 struct pipe_inode_info *pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200896
Ingo Molnar923f4f22006-04-11 13:53:33 +0200897 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
898 if (pipe) {
Jens Axboe35f3d142010-05-20 10:43:18 +0200899 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
900 if (pipe->bufs) {
901 init_waitqueue_head(&pipe->wait);
902 pipe->r_counter = pipe->w_counter = 1;
903 pipe->inode = inode;
904 pipe->buffers = PIPE_DEF_BUFFERS;
905 return pipe;
906 }
907 kfree(pipe);
Ingo Molnar3a326a22006-04-10 15:18:35 +0200908 }
909
Jens Axboe35f3d142010-05-20 10:43:18 +0200910 return NULL;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200911}
912
Ingo Molnar923f4f22006-04-11 13:53:33 +0200913void __free_pipe_info(struct pipe_inode_info *pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914{
915 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Jens Axboe35f3d142010-05-20 10:43:18 +0200917 for (i = 0; i < pipe->buffers; i++) {
Ingo Molnar923f4f22006-04-11 13:53:33 +0200918 struct pipe_buffer *buf = pipe->bufs + i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 if (buf->ops)
Ingo Molnar923f4f22006-04-11 13:53:33 +0200920 buf->ops->release(pipe, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 }
Ingo Molnar923f4f22006-04-11 13:53:33 +0200922 if (pipe->tmp_page)
923 __free_page(pipe->tmp_page);
Jens Axboe35f3d142010-05-20 10:43:18 +0200924 kfree(pipe->bufs);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200925 kfree(pipe);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926}
927
Jens Axboeb92ce552006-04-11 13:52:07 +0200928void free_pipe_info(struct inode *inode)
929{
930 __free_pipe_info(inode->i_pipe);
931 inode->i_pipe = NULL;
932}
933
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800934static struct vfsmount *pipe_mnt __read_mostly;
Ingo Molnar341b4462006-04-11 13:57:45 +0200935
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700936/*
937 * pipefs_dname() is called from d_path().
938 */
939static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
940{
941 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
942 dentry->d_inode->i_ino);
943}
944
Al Viro3ba13d12009-02-20 06:02:22 +0000945static const struct dentry_operations pipefs_dentry_operations = {
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700946 .d_dname = pipefs_dname,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947};
948
949static struct inode * get_pipe_inode(void)
950{
951 struct inode *inode = new_inode(pipe_mnt->mnt_sb);
Ingo Molnar923f4f22006-04-11 13:53:33 +0200952 struct pipe_inode_info *pipe;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954 if (!inode)
955 goto fail_inode;
956
Christoph Hellwig85fe4022010-10-23 11:19:54 -0400957 inode->i_ino = get_next_ino();
958
Ingo Molnar923f4f22006-04-11 13:53:33 +0200959 pipe = alloc_pipe_info(inode);
960 if (!pipe)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 goto fail_iput;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200962 inode->i_pipe = pipe;
Ingo Molnar3a326a22006-04-10 15:18:35 +0200963
Ingo Molnar923f4f22006-04-11 13:53:33 +0200964 pipe->readers = pipe->writers = 1;
Denys Vlasenkod2d96482008-07-01 14:16:09 +0200965 inode->i_fop = &rdwr_pipefifo_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 /*
968 * Mark the inode dirty from the very beginning,
969 * that way it will never be moved to the dirty
970 * list because "mark_inode_dirty()" will think
971 * that it already _is_ on the dirty list.
972 */
973 inode->i_state = I_DIRTY;
974 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
David Howellsda9592e2008-11-14 10:39:05 +1100975 inode->i_uid = current_fsuid();
976 inode->i_gid = current_fsgid();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
Ingo Molnar923f4f22006-04-11 13:53:33 +0200978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 return inode;
980
981fail_iput:
982 iput(inode);
Ingo Molnar341b4462006-04-11 13:57:45 +0200983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984fail_inode:
985 return NULL;
986}
987
Ulrich Drepperbe61a862008-07-23 21:29:40 -0700988struct file *create_write_pipe(int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989{
Andi Kleend6cbd282006-09-30 23:29:26 -0700990 int err;
991 struct inode *inode;
992 struct file *f;
Al Viro2c48b9c2009-08-09 00:52:35 +0400993 struct path path;
Eric Dumazetc23fbb62007-05-08 00:26:18 -0700994 struct qstr name = { .name = "" };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Andi Kleend6cbd282006-09-30 23:29:26 -0700996 err = -ENFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 inode = get_pipe_inode();
998 if (!inode)
Dave Hansen430e2852008-02-15 14:37:26 -0800999 goto err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
Andi Kleend6cbd282006-09-30 23:29:26 -07001001 err = -ENOMEM;
Al Viro2c48b9c2009-08-09 00:52:35 +04001002 path.dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name);
1003 if (!path.dentry)
Andi Kleend6cbd282006-09-30 23:29:26 -07001004 goto err_inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04001005 path.mnt = mntget(pipe_mnt);
Ingo Molnar341b4462006-04-11 13:57:45 +02001006
Al Viro2c48b9c2009-08-09 00:52:35 +04001007 path.dentry->d_op = &pipefs_dentry_operations;
Al Viro2c48b9c2009-08-09 00:52:35 +04001008 d_instantiate(path.dentry, inode);
Dave Hansen430e2852008-02-15 14:37:26 -08001009
1010 err = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04001011 f = alloc_file(&path, FMODE_WRITE, &write_pipefifo_fops);
Dave Hansen430e2852008-02-15 14:37:26 -08001012 if (!f)
1013 goto err_dentry;
Andi Kleend6cbd282006-09-30 23:29:26 -07001014 f->f_mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001016 f->f_flags = O_WRONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -07001017 f->f_version = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
Andi Kleend6cbd282006-09-30 23:29:26 -07001019 return f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
Dave Hansen430e2852008-02-15 14:37:26 -08001021 err_dentry:
Al Viroed152432008-04-22 19:51:27 -04001022 free_pipe_info(inode);
Al Viro2c48b9c2009-08-09 00:52:35 +04001023 path_put(&path);
Al Viroed152432008-04-22 19:51:27 -04001024 return ERR_PTR(err);
1025
Andi Kleend6cbd282006-09-30 23:29:26 -07001026 err_inode:
1027 free_pipe_info(inode);
1028 iput(inode);
Dave Hansen430e2852008-02-15 14:37:26 -08001029 err:
Andi Kleend6cbd282006-09-30 23:29:26 -07001030 return ERR_PTR(err);
1031}
1032
1033void free_write_pipe(struct file *f)
1034{
Al Viro5ccac88e2006-12-18 13:31:18 +00001035 free_pipe_info(f->f_dentry->d_inode);
Jan Blunckc8e7f442008-06-09 16:40:35 -07001036 path_put(&f->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001037 put_filp(f);
1038}
1039
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001040struct file *create_read_pipe(struct file *wrf, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001041{
Al Virod2314122009-08-09 01:01:37 +04001042 /* Grab pipe from the writer */
1043 struct file *f = alloc_file(&wrf->f_path, FMODE_READ,
1044 &read_pipefifo_fops);
Andi Kleend6cbd282006-09-30 23:29:26 -07001045 if (!f)
1046 return ERR_PTR(-ENFILE);
1047
Jan Blunckc8e7f442008-06-09 16:40:35 -07001048 path_get(&wrf->f_path);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001049 f->f_flags = O_RDONLY | (flags & O_NONBLOCK);
Andi Kleend6cbd282006-09-30 23:29:26 -07001050
1051 return f;
1052}
1053
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001054int do_pipe_flags(int *fd, int flags)
Andi Kleend6cbd282006-09-30 23:29:26 -07001055{
1056 struct file *fw, *fr;
1057 int error;
1058 int fdw, fdr;
1059
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001060 if (flags & ~(O_CLOEXEC | O_NONBLOCK))
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001061 return -EINVAL;
1062
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001063 fw = create_write_pipe(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001064 if (IS_ERR(fw))
1065 return PTR_ERR(fw);
Ulrich Drepperbe61a862008-07-23 21:29:40 -07001066 fr = create_read_pipe(fw, flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001067 error = PTR_ERR(fr);
1068 if (IS_ERR(fr))
1069 goto err_write_pipe;
1070
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001071 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001072 if (error < 0)
1073 goto err_read_pipe;
1074 fdr = error;
1075
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001076 error = get_unused_fd_flags(flags);
Andi Kleend6cbd282006-09-30 23:29:26 -07001077 if (error < 0)
1078 goto err_fdr;
1079 fdw = error;
1080
Al Viro157cf642008-12-14 04:57:47 -05001081 audit_fd_pair(fdr, fdw);
Andi Kleend6cbd282006-09-30 23:29:26 -07001082 fd_install(fdr, fr);
1083 fd_install(fdw, fw);
1084 fd[0] = fdr;
1085 fd[1] = fdw;
Ingo Molnar341b4462006-04-11 13:57:45 +02001086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 return 0;
1088
Andi Kleend6cbd282006-09-30 23:29:26 -07001089 err_fdr:
1090 put_unused_fd(fdr);
1091 err_read_pipe:
Jan Blunckc8e7f442008-06-09 16:40:35 -07001092 path_put(&fr->f_path);
Andi Kleend6cbd282006-09-30 23:29:26 -07001093 put_filp(fr);
1094 err_write_pipe:
1095 free_write_pipe(fw);
1096 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097}
1098
1099/*
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001100 * sys_pipe() is the normal C calling standard for creating
1101 * a pipe. It's not the way Unix traditionally does this, though.
1102 */
Heiko Carstensd4e82042009-01-14 14:14:34 +01001103SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001104{
1105 int fd[2];
1106 int error;
1107
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001108 error = do_pipe_flags(fd, flags);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001109 if (!error) {
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001110 if (copy_to_user(fildes, fd, sizeof(fd))) {
1111 sys_close(fd[0]);
1112 sys_close(fd[1]);
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001113 error = -EFAULT;
Ulrich Drepperba719ba2008-05-06 20:42:38 -07001114 }
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001115 }
1116 return error;
1117}
1118
Heiko Carstens2b664212009-01-14 14:14:35 +01001119SYSCALL_DEFINE1(pipe, int __user *, fildes)
Ulrich Dreppered8cae82008-07-23 21:29:30 -07001120{
1121 return sys_pipe2(fildes, 0);
1122}
1123
Ulrich Drepperd35c7b02008-05-03 15:10:37 -04001124/*
Jens Axboe35f3d142010-05-20 10:43:18 +02001125 * Allocate a new array of pipe buffers and copy the info over. Returns the
1126 * pipe size if successful, or return -ERROR on error.
1127 */
Jens Axboeb9598db2010-05-24 19:34:43 +02001128static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +02001129{
1130 struct pipe_buffer *bufs;
1131
1132 /*
Jens Axboe35f3d142010-05-20 10:43:18 +02001133 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1134 * expect a lot of shrink+grow operations, just free and allocate
1135 * again like we would do for growing. If the pipe currently
1136 * contains more buffers than arg, then return busy.
1137 */
Jens Axboeb9598db2010-05-24 19:34:43 +02001138 if (nr_pages < pipe->nrbufs)
Jens Axboe35f3d142010-05-20 10:43:18 +02001139 return -EBUSY;
1140
Jens Axboeb9598db2010-05-24 19:34:43 +02001141 bufs = kcalloc(nr_pages, sizeof(struct pipe_buffer), GFP_KERNEL);
Jens Axboe35f3d142010-05-20 10:43:18 +02001142 if (unlikely(!bufs))
1143 return -ENOMEM;
1144
1145 /*
1146 * The pipe array wraps around, so just start the new one at zero
1147 * and adjust the indexes.
1148 */
1149 if (pipe->nrbufs) {
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001150 unsigned int tail;
1151 unsigned int head;
Jens Axboe35f3d142010-05-20 10:43:18 +02001152
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001153 tail = pipe->curbuf + pipe->nrbufs;
1154 if (tail < pipe->buffers)
1155 tail = 0;
1156 else
1157 tail &= (pipe->buffers - 1);
1158
1159 head = pipe->nrbufs - tail;
Jens Axboe35f3d142010-05-20 10:43:18 +02001160 if (head)
1161 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1162 if (tail)
Miklos Szeredi1d862f42010-06-08 16:28:45 +02001163 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
Jens Axboe35f3d142010-05-20 10:43:18 +02001164 }
1165
1166 pipe->curbuf = 0;
1167 kfree(pipe->bufs);
1168 pipe->bufs = bufs;
Jens Axboeb9598db2010-05-24 19:34:43 +02001169 pipe->buffers = nr_pages;
1170 return nr_pages * PAGE_SIZE;
Jens Axboe35f3d142010-05-20 10:43:18 +02001171}
1172
Jens Axboeff9da692010-06-03 14:54:39 +02001173/*
1174 * Currently we rely on the pipe array holding a power-of-2 number
1175 * of pages.
1176 */
1177static inline unsigned int round_pipe_size(unsigned int size)
1178{
1179 unsigned long nr_pages;
1180
1181 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1182 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1183}
1184
1185/*
1186 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1187 * will return an error.
1188 */
1189int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1190 size_t *lenp, loff_t *ppos)
1191{
1192 int ret;
1193
1194 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1195 if (ret < 0 || !write)
1196 return ret;
1197
1198 pipe_max_size = round_pipe_size(pipe_max_size);
1199 return ret;
1200}
1201
Jens Axboe35f3d142010-05-20 10:43:18 +02001202long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1203{
1204 struct pipe_inode_info *pipe;
1205 long ret;
1206
1207 pipe = file->f_path.dentry->d_inode->i_pipe;
1208 if (!pipe)
1209 return -EBADF;
1210
1211 mutex_lock(&pipe->inode->i_mutex);
1212
1213 switch (cmd) {
Jens Axboeb9598db2010-05-24 19:34:43 +02001214 case F_SETPIPE_SZ: {
Jens Axboeff9da692010-06-03 14:54:39 +02001215 unsigned int size, nr_pages;
Jens Axboeb9598db2010-05-24 19:34:43 +02001216
Jens Axboeff9da692010-06-03 14:54:39 +02001217 size = round_pipe_size(arg);
1218 nr_pages = size >> PAGE_SHIFT;
Jens Axboeb9598db2010-05-24 19:34:43 +02001219
Miklos Szeredi6db40cf2010-06-09 09:27:57 +02001220 ret = -EINVAL;
1221 if (!nr_pages)
1222 goto out;
1223
Jens Axboeff9da692010-06-03 14:54:39 +02001224 if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
Jens Axboeb4ca7612010-06-01 12:42:12 +02001225 ret = -EPERM;
Julia Lawallcc967be2010-05-26 17:54:39 +02001226 goto out;
Julia Lawallcc967be2010-05-26 17:54:39 +02001227 }
Jens Axboeff9da692010-06-03 14:54:39 +02001228 ret = pipe_set_size(pipe, nr_pages);
Jens Axboe35f3d142010-05-20 10:43:18 +02001229 break;
Jens Axboeb9598db2010-05-24 19:34:43 +02001230 }
Jens Axboe35f3d142010-05-20 10:43:18 +02001231 case F_GETPIPE_SZ:
Jens Axboeb9598db2010-05-24 19:34:43 +02001232 ret = pipe->buffers * PAGE_SIZE;
Jens Axboe35f3d142010-05-20 10:43:18 +02001233 break;
1234 default:
1235 ret = -EINVAL;
1236 break;
1237 }
1238
Julia Lawallcc967be2010-05-26 17:54:39 +02001239out:
Jens Axboe35f3d142010-05-20 10:43:18 +02001240 mutex_unlock(&pipe->inode->i_mutex);
1241 return ret;
1242}
1243
1244/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 * pipefs should _never_ be mounted by userland - too much of security hassle,
1246 * no real gain from having the whole whorehouse mounted. So we don't need
1247 * any operations on the root directory. However, we need a non-trivial
1248 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1249 */
David Howells454e2392006-06-23 02:02:57 -07001250static int pipefs_get_sb(struct file_system_type *fs_type,
1251 int flags, const char *dev_name, void *data,
1252 struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253{
David Howells454e2392006-06-23 02:02:57 -07001254 return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255}
1256
1257static struct file_system_type pipe_fs_type = {
1258 .name = "pipefs",
1259 .get_sb = pipefs_get_sb,
1260 .kill_sb = kill_anon_super,
1261};
1262
1263static int __init init_pipe_fs(void)
1264{
1265 int err = register_filesystem(&pipe_fs_type);
Ingo Molnar341b4462006-04-11 13:57:45 +02001266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (!err) {
1268 pipe_mnt = kern_mount(&pipe_fs_type);
1269 if (IS_ERR(pipe_mnt)) {
1270 err = PTR_ERR(pipe_mnt);
1271 unregister_filesystem(&pipe_fs_type);
1272 }
1273 }
1274 return err;
1275}
1276
1277static void __exit exit_pipe_fs(void)
1278{
1279 unregister_filesystem(&pipe_fs_type);
1280 mntput(pipe_mnt);
1281}
1282
1283fs_initcall(init_pipe_fs);
1284module_exit(exit_pipe_fs);