4 * Copyright (C) 1991, 1992, 1999 Linus Torvalds
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>
14 #include <linux/log2.h>
15 #include <linux/mount.h>
16 #include <linux/magic.h>
17 #include <linux/pipe_fs_i.h>
18 #include <linux/uio.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/audit.h>
22 #include <linux/syscalls.h>
23 #include <linux/fcntl.h>
25 #include <asm/uaccess.h>
26 #include <asm/ioctls.h>
31 * The max size that a non-root user is allowed to grow the pipe. Can
32 * be set by root in /proc/sys/fs/pipe-max-size
34 unsigned int pipe_max_size
= 1048576;
37 * Minimum pipe size, as required by POSIX
39 unsigned int pipe_min_size
= PAGE_SIZE
;
42 * We use a start+len construction, which provides full use of the
44 * -- Florian Coosmann (FGC)
46 * Reads with count = 0 should always return 0.
47 * -- Julian Bradfield 1999-06-07.
49 * FIFOs and Pipes now generate SIGIO for both readers and writers.
50 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
52 * pipe_read & write cleanup
53 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
56 static void pipe_lock_nested(struct pipe_inode_info
*pipe
, int subclass
)
59 mutex_lock_nested(&pipe
->mutex
, subclass
);
62 void pipe_lock(struct pipe_inode_info
*pipe
)
65 * pipe_lock() nests non-pipe inode locks (for writing to a file)
67 pipe_lock_nested(pipe
, I_MUTEX_PARENT
);
69 EXPORT_SYMBOL(pipe_lock
);
71 void pipe_unlock(struct pipe_inode_info
*pipe
)
74 mutex_unlock(&pipe
->mutex
);
76 EXPORT_SYMBOL(pipe_unlock
);
78 static inline void __pipe_lock(struct pipe_inode_info
*pipe
)
80 mutex_lock_nested(&pipe
->mutex
, I_MUTEX_PARENT
);
83 static inline void __pipe_unlock(struct pipe_inode_info
*pipe
)
85 mutex_unlock(&pipe
->mutex
);
88 void pipe_double_lock(struct pipe_inode_info
*pipe1
,
89 struct pipe_inode_info
*pipe2
)
91 BUG_ON(pipe1
== pipe2
);
94 pipe_lock_nested(pipe1
, I_MUTEX_PARENT
);
95 pipe_lock_nested(pipe2
, I_MUTEX_CHILD
);
97 pipe_lock_nested(pipe2
, I_MUTEX_PARENT
);
98 pipe_lock_nested(pipe1
, I_MUTEX_CHILD
);
102 /* Drop the inode semaphore and wait for a pipe event, atomically */
103 void pipe_wait(struct pipe_inode_info
*pipe
)
108 * Pipes are system-local resources, so sleeping on them
109 * is considered a noninteractive wait:
111 prepare_to_wait(&pipe
->wait
, &wait
, TASK_INTERRUPTIBLE
);
114 finish_wait(&pipe
->wait
, &wait
);
118 static void anon_pipe_buf_release(struct pipe_inode_info
*pipe
,
119 struct pipe_buffer
*buf
)
121 struct page
*page
= buf
->page
;
124 * If nobody else uses this page, and we don't already have a
125 * temporary page, let's keep track of it as a one-deep
126 * allocation cache. (Otherwise just release our reference to it)
128 if (page_count(page
) == 1 && !pipe
->tmp_page
)
129 pipe
->tmp_page
= page
;
131 page_cache_release(page
);
135 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
136 * @pipe: the pipe that the buffer belongs to
137 * @buf: the buffer to attempt to steal
140 * This function attempts to steal the &struct page attached to
141 * @buf. If successful, this function returns 0 and returns with
142 * the page locked. The caller may then reuse the page for whatever
143 * he wishes; the typical use is insertion into a different file
146 int generic_pipe_buf_steal(struct pipe_inode_info
*pipe
,
147 struct pipe_buffer
*buf
)
149 struct page
*page
= buf
->page
;
152 * A reference of one is golden, that means that the owner of this
153 * page is the only one holding a reference to it. lock the page
156 if (page_count(page
) == 1) {
163 EXPORT_SYMBOL(generic_pipe_buf_steal
);
166 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
167 * @pipe: the pipe that the buffer belongs to
168 * @buf: the buffer to get a reference to
171 * This function grabs an extra reference to @buf. It's used in
172 * in the tee() system call, when we duplicate the buffers in one
175 void generic_pipe_buf_get(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
)
177 page_cache_get(buf
->page
);
179 EXPORT_SYMBOL(generic_pipe_buf_get
);
182 * generic_pipe_buf_confirm - verify contents of the pipe buffer
183 * @info: the pipe that the buffer belongs to
184 * @buf: the buffer to confirm
187 * This function does nothing, because the generic pipe code uses
188 * pages that are always good when inserted into the pipe.
190 int generic_pipe_buf_confirm(struct pipe_inode_info
*info
,
191 struct pipe_buffer
*buf
)
195 EXPORT_SYMBOL(generic_pipe_buf_confirm
);
198 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
199 * @pipe: the pipe that the buffer belongs to
200 * @buf: the buffer to put a reference to
203 * This function releases a reference to @buf.
205 void generic_pipe_buf_release(struct pipe_inode_info
*pipe
,
206 struct pipe_buffer
*buf
)
208 page_cache_release(buf
->page
);
210 EXPORT_SYMBOL(generic_pipe_buf_release
);
212 static const struct pipe_buf_operations anon_pipe_buf_ops
= {
214 .confirm
= generic_pipe_buf_confirm
,
215 .release
= anon_pipe_buf_release
,
216 .steal
= generic_pipe_buf_steal
,
217 .get
= generic_pipe_buf_get
,
220 static const struct pipe_buf_operations packet_pipe_buf_ops
= {
222 .confirm
= generic_pipe_buf_confirm
,
223 .release
= anon_pipe_buf_release
,
224 .steal
= generic_pipe_buf_steal
,
225 .get
= generic_pipe_buf_get
,
229 pipe_read(struct kiocb
*iocb
, struct iov_iter
*to
)
231 size_t total_len
= iov_iter_count(to
);
232 struct file
*filp
= iocb
->ki_filp
;
233 struct pipe_inode_info
*pipe
= filp
->private_data
;
237 /* Null read succeeds. */
238 if (unlikely(total_len
== 0))
245 int bufs
= pipe
->nrbufs
;
247 int curbuf
= pipe
->curbuf
;
248 struct pipe_buffer
*buf
= pipe
->bufs
+ curbuf
;
249 const struct pipe_buf_operations
*ops
= buf
->ops
;
250 size_t chars
= buf
->len
;
254 if (chars
> total_len
)
257 error
= ops
->confirm(pipe
, buf
);
264 written
= copy_page_to_iter(buf
->page
, buf
->offset
, chars
, to
);
265 if (unlikely(written
< chars
)) {
271 buf
->offset
+= chars
;
274 /* Was it a packet buffer? Clean up and exit */
275 if (buf
->flags
& PIPE_BUF_FLAG_PACKET
) {
282 ops
->release(pipe
, buf
);
283 curbuf
= (curbuf
+ 1) & (pipe
->buffers
- 1);
284 pipe
->curbuf
= curbuf
;
285 pipe
->nrbufs
= --bufs
;
290 break; /* common path: read succeeded */
292 if (bufs
) /* More to do? */
296 if (!pipe
->waiting_writers
) {
297 /* syscall merging: Usually we must not sleep
298 * if O_NONBLOCK is set, or if we got some data.
299 * But if a writer sleeps in kernel space, then
300 * we can wait for that data without violating POSIX.
304 if (filp
->f_flags
& O_NONBLOCK
) {
309 if (signal_pending(current
)) {
315 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLOUT
| POLLWRNORM
);
316 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
322 /* Signal writers asynchronously that there is more room. */
324 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLOUT
| POLLWRNORM
);
325 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
332 static inline int is_packetized(struct file
*file
)
334 return (file
->f_flags
& O_DIRECT
) != 0;
338 pipe_write(struct kiocb
*iocb
, struct iov_iter
*from
)
340 struct file
*filp
= iocb
->ki_filp
;
341 struct pipe_inode_info
*pipe
= filp
->private_data
;
344 size_t total_len
= iov_iter_count(from
);
347 /* Null write succeeds. */
348 if (unlikely(total_len
== 0))
353 if (!pipe
->readers
) {
354 send_sig(SIGPIPE
, current
, 0);
359 /* We try to merge small writes */
360 chars
= total_len
& (PAGE_SIZE
-1); /* size of the last buffer */
361 if (pipe
->nrbufs
&& chars
!= 0) {
362 int lastbuf
= (pipe
->curbuf
+ pipe
->nrbufs
- 1) &
364 struct pipe_buffer
*buf
= pipe
->bufs
+ lastbuf
;
365 const struct pipe_buf_operations
*ops
= buf
->ops
;
366 int offset
= buf
->offset
+ buf
->len
;
368 if (ops
->can_merge
&& offset
+ chars
<= PAGE_SIZE
) {
369 ret
= ops
->confirm(pipe
, buf
);
373 ret
= copy_page_from_iter(buf
->page
, offset
, chars
, from
);
374 if (unlikely(ret
< chars
)) {
380 if (!iov_iter_count(from
))
388 if (!pipe
->readers
) {
389 send_sig(SIGPIPE
, current
, 0);
395 if (bufs
< pipe
->buffers
) {
396 int newbuf
= (pipe
->curbuf
+ bufs
) & (pipe
->buffers
-1);
397 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
398 struct page
*page
= pipe
->tmp_page
;
402 page
= alloc_page(GFP_HIGHUSER
);
403 if (unlikely(!page
)) {
404 ret
= ret
? : -ENOMEM
;
407 pipe
->tmp_page
= page
;
409 /* Always wake up, even if the copy fails. Otherwise
410 * we lock up (O_NONBLOCK-)readers that sleep due to
412 * FIXME! Is this really true?
415 copied
= copy_page_from_iter(page
, 0, PAGE_SIZE
, from
);
416 if (unlikely(copied
< PAGE_SIZE
&& iov_iter_count(from
))) {
423 /* Insert it into the buffer array */
425 buf
->ops
= &anon_pipe_buf_ops
;
429 if (is_packetized(filp
)) {
430 buf
->ops
= &packet_pipe_buf_ops
;
431 buf
->flags
= PIPE_BUF_FLAG_PACKET
;
433 pipe
->nrbufs
= ++bufs
;
434 pipe
->tmp_page
= NULL
;
436 if (!iov_iter_count(from
))
439 if (bufs
< pipe
->buffers
)
441 if (filp
->f_flags
& O_NONBLOCK
) {
446 if (signal_pending(current
)) {
452 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLIN
| POLLRDNORM
);
453 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
456 pipe
->waiting_writers
++;
458 pipe
->waiting_writers
--;
463 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLIN
| POLLRDNORM
);
464 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
466 if (ret
> 0 && sb_start_write_trylock(file_inode(filp
)->i_sb
)) {
467 int err
= file_update_time(filp
);
470 sb_end_write(file_inode(filp
)->i_sb
);
475 static long pipe_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
477 struct pipe_inode_info
*pipe
= filp
->private_data
;
478 int count
, buf
, nrbufs
;
485 nrbufs
= pipe
->nrbufs
;
486 while (--nrbufs
>= 0) {
487 count
+= pipe
->bufs
[buf
].len
;
488 buf
= (buf
+1) & (pipe
->buffers
- 1);
492 return put_user(count
, (int __user
*)arg
);
498 /* No kernel lock held - fine */
500 pipe_poll(struct file
*filp
, poll_table
*wait
)
503 struct pipe_inode_info
*pipe
= filp
->private_data
;
506 poll_wait(filp
, &pipe
->wait
, wait
);
508 /* Reading only -- no need for acquiring the semaphore. */
509 nrbufs
= pipe
->nrbufs
;
511 if (filp
->f_mode
& FMODE_READ
) {
512 mask
= (nrbufs
> 0) ? POLLIN
| POLLRDNORM
: 0;
513 if (!pipe
->writers
&& filp
->f_version
!= pipe
->w_counter
)
517 if (filp
->f_mode
& FMODE_WRITE
) {
518 mask
|= (nrbufs
< pipe
->buffers
) ? POLLOUT
| POLLWRNORM
: 0;
520 * Most Unices do not set POLLERR for FIFOs but on Linux they
521 * behave exactly like pipes for poll().
530 static void put_pipe_info(struct inode
*inode
, struct pipe_inode_info
*pipe
)
534 spin_lock(&inode
->i_lock
);
535 if (!--pipe
->files
) {
536 inode
->i_pipe
= NULL
;
539 spin_unlock(&inode
->i_lock
);
542 free_pipe_info(pipe
);
546 pipe_release(struct inode
*inode
, struct file
*file
)
548 struct pipe_inode_info
*pipe
= file
->private_data
;
551 if (file
->f_mode
& FMODE_READ
)
553 if (file
->f_mode
& FMODE_WRITE
)
556 if (pipe
->readers
|| pipe
->writers
) {
557 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLIN
| POLLOUT
| POLLRDNORM
| POLLWRNORM
| POLLERR
| POLLHUP
);
558 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
559 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
563 put_pipe_info(inode
, pipe
);
568 pipe_fasync(int fd
, struct file
*filp
, int on
)
570 struct pipe_inode_info
*pipe
= filp
->private_data
;
574 if (filp
->f_mode
& FMODE_READ
)
575 retval
= fasync_helper(fd
, filp
, on
, &pipe
->fasync_readers
);
576 if ((filp
->f_mode
& FMODE_WRITE
) && retval
>= 0) {
577 retval
= fasync_helper(fd
, filp
, on
, &pipe
->fasync_writers
);
578 if (retval
< 0 && (filp
->f_mode
& FMODE_READ
))
579 /* this can happen only if on == T */
580 fasync_helper(-1, filp
, 0, &pipe
->fasync_readers
);
586 struct pipe_inode_info
*alloc_pipe_info(void)
588 struct pipe_inode_info
*pipe
;
590 pipe
= kzalloc(sizeof(struct pipe_inode_info
), GFP_KERNEL
);
592 pipe
->bufs
= kzalloc(sizeof(struct pipe_buffer
) * PIPE_DEF_BUFFERS
, GFP_KERNEL
);
594 init_waitqueue_head(&pipe
->wait
);
595 pipe
->r_counter
= pipe
->w_counter
= 1;
596 pipe
->buffers
= PIPE_DEF_BUFFERS
;
597 mutex_init(&pipe
->mutex
);
606 void free_pipe_info(struct pipe_inode_info
*pipe
)
610 for (i
= 0; i
< pipe
->buffers
; i
++) {
611 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
613 buf
->ops
->release(pipe
, buf
);
616 __free_page(pipe
->tmp_page
);
621 static struct vfsmount
*pipe_mnt __read_mostly
;
624 * pipefs_dname() is called from d_path().
626 static char *pipefs_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
628 return dynamic_dname(dentry
, buffer
, buflen
, "pipe:[%lu]",
629 d_inode(dentry
)->i_ino
);
632 static const struct dentry_operations pipefs_dentry_operations
= {
633 .d_dname
= pipefs_dname
,
636 static struct inode
* get_pipe_inode(void)
638 struct inode
*inode
= new_inode_pseudo(pipe_mnt
->mnt_sb
);
639 struct pipe_inode_info
*pipe
;
644 inode
->i_ino
= get_next_ino();
646 pipe
= alloc_pipe_info();
650 inode
->i_pipe
= pipe
;
652 pipe
->readers
= pipe
->writers
= 1;
653 inode
->i_fop
= &pipefifo_fops
;
656 * Mark the inode dirty from the very beginning,
657 * that way it will never be moved to the dirty
658 * list because "mark_inode_dirty()" will think
659 * that it already _is_ on the dirty list.
661 inode
->i_state
= I_DIRTY
;
662 inode
->i_mode
= S_IFIFO
| S_IRUSR
| S_IWUSR
;
663 inode
->i_uid
= current_fsuid();
664 inode
->i_gid
= current_fsgid();
665 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
676 int create_pipe_files(struct file
**res
, int flags
)
679 struct inode
*inode
= get_pipe_inode();
682 static struct qstr name
= { .name
= "" };
688 path
.dentry
= d_alloc_pseudo(pipe_mnt
->mnt_sb
, &name
);
691 path
.mnt
= mntget(pipe_mnt
);
693 d_instantiate(path
.dentry
, inode
);
695 f
= alloc_file(&path
, FMODE_WRITE
, &pipefifo_fops
);
701 f
->f_flags
= O_WRONLY
| (flags
& (O_NONBLOCK
| O_DIRECT
));
702 f
->private_data
= inode
->i_pipe
;
704 res
[0] = alloc_file(&path
, FMODE_READ
, &pipefifo_fops
);
705 if (IS_ERR(res
[0])) {
706 err
= PTR_ERR(res
[0]);
711 res
[0]->private_data
= inode
->i_pipe
;
712 res
[0]->f_flags
= O_RDONLY
| (flags
& O_NONBLOCK
);
719 free_pipe_info(inode
->i_pipe
);
724 free_pipe_info(inode
->i_pipe
);
729 static int __do_pipe_flags(int *fd
, struct file
**files
, int flags
)
734 if (flags
& ~(O_CLOEXEC
| O_NONBLOCK
| O_DIRECT
))
737 error
= create_pipe_files(files
, flags
);
741 error
= get_unused_fd_flags(flags
);
746 error
= get_unused_fd_flags(flags
);
751 audit_fd_pair(fdr
, fdw
);
764 int do_pipe_flags(int *fd
, int flags
)
766 struct file
*files
[2];
767 int error
= __do_pipe_flags(fd
, files
, flags
);
769 fd_install(fd
[0], files
[0]);
770 fd_install(fd
[1], files
[1]);
776 * sys_pipe() is the normal C calling standard for creating
777 * a pipe. It's not the way Unix traditionally does this, though.
779 SYSCALL_DEFINE2(pipe2
, int __user
*, fildes
, int, flags
)
781 struct file
*files
[2];
785 error
= __do_pipe_flags(fd
, files
, flags
);
787 if (unlikely(copy_to_user(fildes
, fd
, sizeof(fd
)))) {
790 put_unused_fd(fd
[0]);
791 put_unused_fd(fd
[1]);
794 fd_install(fd
[0], files
[0]);
795 fd_install(fd
[1], files
[1]);
801 SYSCALL_DEFINE1(pipe
, int __user
*, fildes
)
803 return sys_pipe2(fildes
, 0);
806 static int wait_for_partner(struct pipe_inode_info
*pipe
, unsigned int *cnt
)
810 while (cur
== *cnt
) {
812 if (signal_pending(current
))
815 return cur
== *cnt
? -ERESTARTSYS
: 0;
818 static void wake_up_partner(struct pipe_inode_info
*pipe
)
820 wake_up_interruptible(&pipe
->wait
);
823 static int fifo_open(struct inode
*inode
, struct file
*filp
)
825 struct pipe_inode_info
*pipe
;
826 bool is_pipe
= inode
->i_sb
->s_magic
== PIPEFS_MAGIC
;
831 spin_lock(&inode
->i_lock
);
833 pipe
= inode
->i_pipe
;
835 spin_unlock(&inode
->i_lock
);
837 spin_unlock(&inode
->i_lock
);
838 pipe
= alloc_pipe_info();
842 spin_lock(&inode
->i_lock
);
843 if (unlikely(inode
->i_pipe
)) {
844 inode
->i_pipe
->files
++;
845 spin_unlock(&inode
->i_lock
);
846 free_pipe_info(pipe
);
847 pipe
= inode
->i_pipe
;
849 inode
->i_pipe
= pipe
;
850 spin_unlock(&inode
->i_lock
);
853 filp
->private_data
= pipe
;
854 /* OK, we have a pipe and it's pinned down */
858 /* We can only do regular read/write on fifos */
859 filp
->f_mode
&= (FMODE_READ
| FMODE_WRITE
);
861 switch (filp
->f_mode
) {
865 * POSIX.1 says that O_NONBLOCK means return with the FIFO
866 * opened, even when there is no process writing the FIFO.
869 if (pipe
->readers
++ == 0)
870 wake_up_partner(pipe
);
872 if (!is_pipe
&& !pipe
->writers
) {
873 if ((filp
->f_flags
& O_NONBLOCK
)) {
874 /* suppress POLLHUP until we have
876 filp
->f_version
= pipe
->w_counter
;
878 if (wait_for_partner(pipe
, &pipe
->w_counter
))
887 * POSIX.1 says that O_NONBLOCK means return -1 with
888 * errno=ENXIO when there is no process reading the FIFO.
891 if (!is_pipe
&& (filp
->f_flags
& O_NONBLOCK
) && !pipe
->readers
)
895 if (!pipe
->writers
++)
896 wake_up_partner(pipe
);
898 if (!is_pipe
&& !pipe
->readers
) {
899 if (wait_for_partner(pipe
, &pipe
->r_counter
))
904 case FMODE_READ
| FMODE_WRITE
:
907 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
908 * This implementation will NEVER block on a O_RDWR open, since
909 * the process can at least talk to itself.
916 if (pipe
->readers
== 1 || pipe
->writers
== 1)
917 wake_up_partner(pipe
);
930 if (!--pipe
->readers
)
931 wake_up_interruptible(&pipe
->wait
);
936 if (!--pipe
->writers
)
937 wake_up_interruptible(&pipe
->wait
);
944 put_pipe_info(inode
, pipe
);
948 const struct file_operations pipefifo_fops
= {
951 .read_iter
= pipe_read
,
952 .write_iter
= pipe_write
,
954 .unlocked_ioctl
= pipe_ioctl
,
955 .release
= pipe_release
,
956 .fasync
= pipe_fasync
,
960 * Allocate a new array of pipe buffers and copy the info over. Returns the
961 * pipe size if successful, or return -ERROR on error.
963 static long pipe_set_size(struct pipe_inode_info
*pipe
, unsigned long nr_pages
)
965 struct pipe_buffer
*bufs
;
968 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
969 * expect a lot of shrink+grow operations, just free and allocate
970 * again like we would do for growing. If the pipe currently
971 * contains more buffers than arg, then return busy.
973 if (nr_pages
< pipe
->nrbufs
)
976 bufs
= kcalloc(nr_pages
, sizeof(*bufs
), GFP_KERNEL
| __GFP_NOWARN
);
981 * The pipe array wraps around, so just start the new one at zero
982 * and adjust the indexes.
988 tail
= pipe
->curbuf
+ pipe
->nrbufs
;
989 if (tail
< pipe
->buffers
)
992 tail
&= (pipe
->buffers
- 1);
994 head
= pipe
->nrbufs
- tail
;
996 memcpy(bufs
, pipe
->bufs
+ pipe
->curbuf
, head
* sizeof(struct pipe_buffer
));
998 memcpy(bufs
+ head
, pipe
->bufs
, tail
* sizeof(struct pipe_buffer
));
1004 pipe
->buffers
= nr_pages
;
1005 return nr_pages
* PAGE_SIZE
;
1009 * Currently we rely on the pipe array holding a power-of-2 number
1012 static inline unsigned int round_pipe_size(unsigned int size
)
1014 unsigned long nr_pages
;
1016 nr_pages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1017 return roundup_pow_of_two(nr_pages
) << PAGE_SHIFT
;
1021 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1022 * will return an error.
1024 int pipe_proc_fn(struct ctl_table
*table
, int write
, void __user
*buf
,
1025 size_t *lenp
, loff_t
*ppos
)
1029 ret
= proc_dointvec_minmax(table
, write
, buf
, lenp
, ppos
);
1030 if (ret
< 0 || !write
)
1033 pipe_max_size
= round_pipe_size(pipe_max_size
);
1038 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1039 * location, so checking ->i_pipe is not enough to verify that this is a
1042 struct pipe_inode_info
*get_pipe_info(struct file
*file
)
1044 return file
->f_op
== &pipefifo_fops
? file
->private_data
: NULL
;
1047 long pipe_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
1049 struct pipe_inode_info
*pipe
;
1052 pipe
= get_pipe_info(file
);
1059 case F_SETPIPE_SZ
: {
1060 unsigned int size
, nr_pages
;
1062 size
= round_pipe_size(arg
);
1063 nr_pages
= size
>> PAGE_SHIFT
;
1069 if (!capable(CAP_SYS_RESOURCE
) && size
> pipe_max_size
) {
1073 ret
= pipe_set_size(pipe
, nr_pages
);
1077 ret
= pipe
->buffers
* PAGE_SIZE
;
1085 __pipe_unlock(pipe
);
1089 static const struct super_operations pipefs_ops
= {
1090 .destroy_inode
= free_inode_nonrcu
,
1091 .statfs
= simple_statfs
,
1095 * pipefs should _never_ be mounted by userland - too much of security hassle,
1096 * no real gain from having the whole whorehouse mounted. So we don't need
1097 * any operations on the root directory. However, we need a non-trivial
1098 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1100 static struct dentry
*pipefs_mount(struct file_system_type
*fs_type
,
1101 int flags
, const char *dev_name
, void *data
)
1103 return mount_pseudo(fs_type
, "pipe:", &pipefs_ops
,
1104 &pipefs_dentry_operations
, PIPEFS_MAGIC
);
1107 static struct file_system_type pipe_fs_type
= {
1109 .mount
= pipefs_mount
,
1110 .kill_sb
= kill_anon_super
,
1113 static int __init
init_pipe_fs(void)
1115 int err
= register_filesystem(&pipe_fs_type
);
1118 pipe_mnt
= kern_mount(&pipe_fs_type
);
1119 if (IS_ERR(pipe_mnt
)) {
1120 err
= PTR_ERR(pipe_mnt
);
1121 unregister_filesystem(&pipe_fs_type
);
1127 fs_initcall(init_pipe_fs
);