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>
24 #include <linux/memcontrol.h>
26 #include <asm/uaccess.h>
27 #include <asm/ioctls.h>
32 * The max size that a non-root user is allowed to grow the pipe. Can
33 * be set by root in /proc/sys/fs/pipe-max-size
35 unsigned int pipe_max_size
= 1048576;
38 * Minimum pipe size, as required by POSIX
40 unsigned int pipe_min_size
= PAGE_SIZE
;
42 /* Maximum allocatable pages per user. Hard limit is unset by default, soft
43 * matches default values.
45 unsigned long pipe_user_pages_hard
;
46 unsigned long pipe_user_pages_soft
= PIPE_DEF_BUFFERS
* INR_OPEN_CUR
;
49 * We use a start+len construction, which provides full use of the
51 * -- Florian Coosmann (FGC)
53 * Reads with count = 0 should always return 0.
54 * -- Julian Bradfield 1999-06-07.
56 * FIFOs and Pipes now generate SIGIO for both readers and writers.
57 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
59 * pipe_read & write cleanup
60 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
63 static void pipe_lock_nested(struct pipe_inode_info
*pipe
, int subclass
)
66 mutex_lock_nested(&pipe
->mutex
, subclass
);
69 void pipe_lock(struct pipe_inode_info
*pipe
)
72 * pipe_lock() nests non-pipe inode locks (for writing to a file)
74 pipe_lock_nested(pipe
, I_MUTEX_PARENT
);
76 EXPORT_SYMBOL(pipe_lock
);
78 void pipe_unlock(struct pipe_inode_info
*pipe
)
81 mutex_unlock(&pipe
->mutex
);
83 EXPORT_SYMBOL(pipe_unlock
);
85 static inline void __pipe_lock(struct pipe_inode_info
*pipe
)
87 mutex_lock_nested(&pipe
->mutex
, I_MUTEX_PARENT
);
90 static inline void __pipe_unlock(struct pipe_inode_info
*pipe
)
92 mutex_unlock(&pipe
->mutex
);
95 void pipe_double_lock(struct pipe_inode_info
*pipe1
,
96 struct pipe_inode_info
*pipe2
)
98 BUG_ON(pipe1
== pipe2
);
101 pipe_lock_nested(pipe1
, I_MUTEX_PARENT
);
102 pipe_lock_nested(pipe2
, I_MUTEX_CHILD
);
104 pipe_lock_nested(pipe2
, I_MUTEX_PARENT
);
105 pipe_lock_nested(pipe1
, I_MUTEX_CHILD
);
109 /* Drop the inode semaphore and wait for a pipe event, atomically */
110 void pipe_wait(struct pipe_inode_info
*pipe
)
115 * Pipes are system-local resources, so sleeping on them
116 * is considered a noninteractive wait:
118 prepare_to_wait(&pipe
->wait
, &wait
, TASK_INTERRUPTIBLE
);
121 finish_wait(&pipe
->wait
, &wait
);
125 static void anon_pipe_buf_release(struct pipe_inode_info
*pipe
,
126 struct pipe_buffer
*buf
)
128 struct page
*page
= buf
->page
;
131 * If nobody else uses this page, and we don't already have a
132 * temporary page, let's keep track of it as a one-deep
133 * allocation cache. (Otherwise just release our reference to it)
135 if (page_count(page
) == 1 && !pipe
->tmp_page
)
136 pipe
->tmp_page
= page
;
141 static int anon_pipe_buf_steal(struct pipe_inode_info
*pipe
,
142 struct pipe_buffer
*buf
)
144 struct page
*page
= buf
->page
;
146 if (page_count(page
) == 1) {
147 if (memcg_kmem_enabled())
148 memcg_kmem_uncharge(page
, 0);
149 __SetPageLocked(page
);
156 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
157 * @pipe: the pipe that the buffer belongs to
158 * @buf: the buffer to attempt to steal
161 * This function attempts to steal the &struct page attached to
162 * @buf. If successful, this function returns 0 and returns with
163 * the page locked. The caller may then reuse the page for whatever
164 * he wishes; the typical use is insertion into a different file
167 int generic_pipe_buf_steal(struct pipe_inode_info
*pipe
,
168 struct pipe_buffer
*buf
)
170 struct page
*page
= buf
->page
;
173 * A reference of one is golden, that means that the owner of this
174 * page is the only one holding a reference to it. lock the page
177 if (page_count(page
) == 1) {
184 EXPORT_SYMBOL(generic_pipe_buf_steal
);
187 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
188 * @pipe: the pipe that the buffer belongs to
189 * @buf: the buffer to get a reference to
192 * This function grabs an extra reference to @buf. It's used in
193 * in the tee() system call, when we duplicate the buffers in one
196 void generic_pipe_buf_get(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
)
200 EXPORT_SYMBOL(generic_pipe_buf_get
);
203 * generic_pipe_buf_confirm - verify contents of the pipe buffer
204 * @info: the pipe that the buffer belongs to
205 * @buf: the buffer to confirm
208 * This function does nothing, because the generic pipe code uses
209 * pages that are always good when inserted into the pipe.
211 int generic_pipe_buf_confirm(struct pipe_inode_info
*info
,
212 struct pipe_buffer
*buf
)
216 EXPORT_SYMBOL(generic_pipe_buf_confirm
);
219 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
220 * @pipe: the pipe that the buffer belongs to
221 * @buf: the buffer to put a reference to
224 * This function releases a reference to @buf.
226 void generic_pipe_buf_release(struct pipe_inode_info
*pipe
,
227 struct pipe_buffer
*buf
)
231 EXPORT_SYMBOL(generic_pipe_buf_release
);
233 static const struct pipe_buf_operations anon_pipe_buf_ops
= {
235 .confirm
= generic_pipe_buf_confirm
,
236 .release
= anon_pipe_buf_release
,
237 .steal
= anon_pipe_buf_steal
,
238 .get
= generic_pipe_buf_get
,
241 static const struct pipe_buf_operations packet_pipe_buf_ops
= {
243 .confirm
= generic_pipe_buf_confirm
,
244 .release
= anon_pipe_buf_release
,
245 .steal
= anon_pipe_buf_steal
,
246 .get
= generic_pipe_buf_get
,
250 pipe_read(struct kiocb
*iocb
, struct iov_iter
*to
)
252 size_t total_len
= iov_iter_count(to
);
253 struct file
*filp
= iocb
->ki_filp
;
254 struct pipe_inode_info
*pipe
= filp
->private_data
;
258 /* Null read succeeds. */
259 if (unlikely(total_len
== 0))
266 int bufs
= pipe
->nrbufs
;
268 int curbuf
= pipe
->curbuf
;
269 struct pipe_buffer
*buf
= pipe
->bufs
+ curbuf
;
270 const struct pipe_buf_operations
*ops
= buf
->ops
;
271 size_t chars
= buf
->len
;
275 if (chars
> total_len
)
278 error
= ops
->confirm(pipe
, buf
);
285 written
= copy_page_to_iter(buf
->page
, buf
->offset
, chars
, to
);
286 if (unlikely(written
< chars
)) {
292 buf
->offset
+= chars
;
295 /* Was it a packet buffer? Clean up and exit */
296 if (buf
->flags
& PIPE_BUF_FLAG_PACKET
) {
303 ops
->release(pipe
, buf
);
304 curbuf
= (curbuf
+ 1) & (pipe
->buffers
- 1);
305 pipe
->curbuf
= curbuf
;
306 pipe
->nrbufs
= --bufs
;
311 break; /* common path: read succeeded */
313 if (bufs
) /* More to do? */
317 if (!pipe
->waiting_writers
) {
318 /* syscall merging: Usually we must not sleep
319 * if O_NONBLOCK is set, or if we got some data.
320 * But if a writer sleeps in kernel space, then
321 * we can wait for that data without violating POSIX.
325 if (filp
->f_flags
& O_NONBLOCK
) {
330 if (signal_pending(current
)) {
336 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLOUT
| POLLWRNORM
);
337 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
343 /* Signal writers asynchronously that there is more room. */
345 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLOUT
| POLLWRNORM
);
346 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
353 static inline int is_packetized(struct file
*file
)
355 return (file
->f_flags
& O_DIRECT
) != 0;
359 pipe_write(struct kiocb
*iocb
, struct iov_iter
*from
)
361 struct file
*filp
= iocb
->ki_filp
;
362 struct pipe_inode_info
*pipe
= filp
->private_data
;
365 size_t total_len
= iov_iter_count(from
);
368 /* Null write succeeds. */
369 if (unlikely(total_len
== 0))
374 if (!pipe
->readers
) {
375 send_sig(SIGPIPE
, current
, 0);
380 /* We try to merge small writes */
381 chars
= total_len
& (PAGE_SIZE
-1); /* size of the last buffer */
382 if (pipe
->nrbufs
&& chars
!= 0) {
383 int lastbuf
= (pipe
->curbuf
+ pipe
->nrbufs
- 1) &
385 struct pipe_buffer
*buf
= pipe
->bufs
+ lastbuf
;
386 const struct pipe_buf_operations
*ops
= buf
->ops
;
387 int offset
= buf
->offset
+ buf
->len
;
389 if (ops
->can_merge
&& offset
+ chars
<= PAGE_SIZE
) {
390 ret
= ops
->confirm(pipe
, buf
);
394 ret
= copy_page_from_iter(buf
->page
, offset
, chars
, from
);
395 if (unlikely(ret
< chars
)) {
401 if (!iov_iter_count(from
))
409 if (!pipe
->readers
) {
410 send_sig(SIGPIPE
, current
, 0);
416 if (bufs
< pipe
->buffers
) {
417 int newbuf
= (pipe
->curbuf
+ bufs
) & (pipe
->buffers
-1);
418 struct pipe_buffer
*buf
= pipe
->bufs
+ newbuf
;
419 struct page
*page
= pipe
->tmp_page
;
423 page
= alloc_page(GFP_HIGHUSER
| __GFP_ACCOUNT
);
424 if (unlikely(!page
)) {
425 ret
= ret
? : -ENOMEM
;
428 pipe
->tmp_page
= page
;
430 /* Always wake up, even if the copy fails. Otherwise
431 * we lock up (O_NONBLOCK-)readers that sleep due to
433 * FIXME! Is this really true?
436 copied
= copy_page_from_iter(page
, 0, PAGE_SIZE
, from
);
437 if (unlikely(copied
< PAGE_SIZE
&& iov_iter_count(from
))) {
444 /* Insert it into the buffer array */
446 buf
->ops
= &anon_pipe_buf_ops
;
450 if (is_packetized(filp
)) {
451 buf
->ops
= &packet_pipe_buf_ops
;
452 buf
->flags
= PIPE_BUF_FLAG_PACKET
;
454 pipe
->nrbufs
= ++bufs
;
455 pipe
->tmp_page
= NULL
;
457 if (!iov_iter_count(from
))
460 if (bufs
< pipe
->buffers
)
462 if (filp
->f_flags
& O_NONBLOCK
) {
467 if (signal_pending(current
)) {
473 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLIN
| POLLRDNORM
);
474 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
477 pipe
->waiting_writers
++;
479 pipe
->waiting_writers
--;
484 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLIN
| POLLRDNORM
);
485 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
487 if (ret
> 0 && sb_start_write_trylock(file_inode(filp
)->i_sb
)) {
488 int err
= file_update_time(filp
);
491 sb_end_write(file_inode(filp
)->i_sb
);
496 static long pipe_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
498 struct pipe_inode_info
*pipe
= filp
->private_data
;
499 int count
, buf
, nrbufs
;
506 nrbufs
= pipe
->nrbufs
;
507 while (--nrbufs
>= 0) {
508 count
+= pipe
->bufs
[buf
].len
;
509 buf
= (buf
+1) & (pipe
->buffers
- 1);
513 return put_user(count
, (int __user
*)arg
);
519 /* No kernel lock held - fine */
521 pipe_poll(struct file
*filp
, poll_table
*wait
)
524 struct pipe_inode_info
*pipe
= filp
->private_data
;
527 poll_wait(filp
, &pipe
->wait
, wait
);
529 /* Reading only -- no need for acquiring the semaphore. */
530 nrbufs
= pipe
->nrbufs
;
532 if (filp
->f_mode
& FMODE_READ
) {
533 mask
= (nrbufs
> 0) ? POLLIN
| POLLRDNORM
: 0;
534 if (!pipe
->writers
&& filp
->f_version
!= pipe
->w_counter
)
538 if (filp
->f_mode
& FMODE_WRITE
) {
539 mask
|= (nrbufs
< pipe
->buffers
) ? POLLOUT
| POLLWRNORM
: 0;
541 * Most Unices do not set POLLERR for FIFOs but on Linux they
542 * behave exactly like pipes for poll().
551 static void put_pipe_info(struct inode
*inode
, struct pipe_inode_info
*pipe
)
555 spin_lock(&inode
->i_lock
);
556 if (!--pipe
->files
) {
557 inode
->i_pipe
= NULL
;
560 spin_unlock(&inode
->i_lock
);
563 free_pipe_info(pipe
);
567 pipe_release(struct inode
*inode
, struct file
*file
)
569 struct pipe_inode_info
*pipe
= file
->private_data
;
572 if (file
->f_mode
& FMODE_READ
)
574 if (file
->f_mode
& FMODE_WRITE
)
577 if (pipe
->readers
|| pipe
->writers
) {
578 wake_up_interruptible_sync_poll(&pipe
->wait
, POLLIN
| POLLOUT
| POLLRDNORM
| POLLWRNORM
| POLLERR
| POLLHUP
);
579 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
580 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
584 put_pipe_info(inode
, pipe
);
589 pipe_fasync(int fd
, struct file
*filp
, int on
)
591 struct pipe_inode_info
*pipe
= filp
->private_data
;
595 if (filp
->f_mode
& FMODE_READ
)
596 retval
= fasync_helper(fd
, filp
, on
, &pipe
->fasync_readers
);
597 if ((filp
->f_mode
& FMODE_WRITE
) && retval
>= 0) {
598 retval
= fasync_helper(fd
, filp
, on
, &pipe
->fasync_writers
);
599 if (retval
< 0 && (filp
->f_mode
& FMODE_READ
))
600 /* this can happen only if on == T */
601 fasync_helper(-1, filp
, 0, &pipe
->fasync_readers
);
607 static unsigned long account_pipe_buffers(struct user_struct
*user
,
608 unsigned long old
, unsigned long new)
610 return atomic_long_add_return(new - old
, &user
->pipe_bufs
);
613 static bool too_many_pipe_buffers_soft(unsigned long user_bufs
)
615 return pipe_user_pages_soft
&& user_bufs
>= pipe_user_pages_soft
;
618 static bool too_many_pipe_buffers_hard(unsigned long user_bufs
)
620 return pipe_user_pages_hard
&& user_bufs
>= pipe_user_pages_hard
;
623 struct pipe_inode_info
*alloc_pipe_info(void)
625 struct pipe_inode_info
*pipe
;
626 unsigned long pipe_bufs
= PIPE_DEF_BUFFERS
;
627 struct user_struct
*user
= get_current_user();
628 unsigned long user_bufs
;
630 pipe
= kzalloc(sizeof(struct pipe_inode_info
), GFP_KERNEL_ACCOUNT
);
634 if (pipe_bufs
* PAGE_SIZE
> pipe_max_size
&& !capable(CAP_SYS_RESOURCE
))
635 pipe_bufs
= pipe_max_size
>> PAGE_SHIFT
;
637 user_bufs
= account_pipe_buffers(user
, 0, pipe_bufs
);
639 if (too_many_pipe_buffers_soft(user_bufs
)) {
640 user_bufs
= account_pipe_buffers(user
, pipe_bufs
, 1);
644 if (too_many_pipe_buffers_hard(user_bufs
))
645 goto out_revert_acct
;
647 pipe
->bufs
= kcalloc(pipe_bufs
, sizeof(struct pipe_buffer
),
651 init_waitqueue_head(&pipe
->wait
);
652 pipe
->r_counter
= pipe
->w_counter
= 1;
653 pipe
->buffers
= pipe_bufs
;
655 mutex_init(&pipe
->mutex
);
660 (void) account_pipe_buffers(user
, pipe_bufs
, 0);
667 void free_pipe_info(struct pipe_inode_info
*pipe
)
671 (void) account_pipe_buffers(pipe
->user
, pipe
->buffers
, 0);
672 free_uid(pipe
->user
);
673 for (i
= 0; i
< pipe
->buffers
; i
++) {
674 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
676 buf
->ops
->release(pipe
, buf
);
679 __free_page(pipe
->tmp_page
);
684 static struct vfsmount
*pipe_mnt __read_mostly
;
687 * pipefs_dname() is called from d_path().
689 static char *pipefs_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
691 return dynamic_dname(dentry
, buffer
, buflen
, "pipe:[%lu]",
692 d_inode(dentry
)->i_ino
);
695 static const struct dentry_operations pipefs_dentry_operations
= {
696 .d_dname
= pipefs_dname
,
699 static struct inode
* get_pipe_inode(void)
701 struct inode
*inode
= new_inode_pseudo(pipe_mnt
->mnt_sb
);
702 struct pipe_inode_info
*pipe
;
707 inode
->i_ino
= get_next_ino();
709 pipe
= alloc_pipe_info();
713 inode
->i_pipe
= pipe
;
715 pipe
->readers
= pipe
->writers
= 1;
716 inode
->i_fop
= &pipefifo_fops
;
719 * Mark the inode dirty from the very beginning,
720 * that way it will never be moved to the dirty
721 * list because "mark_inode_dirty()" will think
722 * that it already _is_ on the dirty list.
724 inode
->i_state
= I_DIRTY
;
725 inode
->i_mode
= S_IFIFO
| S_IRUSR
| S_IWUSR
;
726 inode
->i_uid
= current_fsuid();
727 inode
->i_gid
= current_fsgid();
728 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
739 int create_pipe_files(struct file
**res
, int flags
)
742 struct inode
*inode
= get_pipe_inode();
745 static struct qstr name
= { .name
= "" };
751 path
.dentry
= d_alloc_pseudo(pipe_mnt
->mnt_sb
, &name
);
754 path
.mnt
= mntget(pipe_mnt
);
756 d_instantiate(path
.dentry
, inode
);
758 f
= alloc_file(&path
, FMODE_WRITE
, &pipefifo_fops
);
764 f
->f_flags
= O_WRONLY
| (flags
& (O_NONBLOCK
| O_DIRECT
));
765 f
->private_data
= inode
->i_pipe
;
767 res
[0] = alloc_file(&path
, FMODE_READ
, &pipefifo_fops
);
768 if (IS_ERR(res
[0])) {
769 err
= PTR_ERR(res
[0]);
774 res
[0]->private_data
= inode
->i_pipe
;
775 res
[0]->f_flags
= O_RDONLY
| (flags
& O_NONBLOCK
);
782 free_pipe_info(inode
->i_pipe
);
787 free_pipe_info(inode
->i_pipe
);
792 static int __do_pipe_flags(int *fd
, struct file
**files
, int flags
)
797 if (flags
& ~(O_CLOEXEC
| O_NONBLOCK
| O_DIRECT
))
800 error
= create_pipe_files(files
, flags
);
804 error
= get_unused_fd_flags(flags
);
809 error
= get_unused_fd_flags(flags
);
814 audit_fd_pair(fdr
, fdw
);
827 int do_pipe_flags(int *fd
, int flags
)
829 struct file
*files
[2];
830 int error
= __do_pipe_flags(fd
, files
, flags
);
832 fd_install(fd
[0], files
[0]);
833 fd_install(fd
[1], files
[1]);
839 * sys_pipe() is the normal C calling standard for creating
840 * a pipe. It's not the way Unix traditionally does this, though.
842 SYSCALL_DEFINE2(pipe2
, int __user
*, fildes
, int, flags
)
844 struct file
*files
[2];
848 error
= __do_pipe_flags(fd
, files
, flags
);
850 if (unlikely(copy_to_user(fildes
, fd
, sizeof(fd
)))) {
853 put_unused_fd(fd
[0]);
854 put_unused_fd(fd
[1]);
857 fd_install(fd
[0], files
[0]);
858 fd_install(fd
[1], files
[1]);
864 SYSCALL_DEFINE1(pipe
, int __user
*, fildes
)
866 return sys_pipe2(fildes
, 0);
869 static int wait_for_partner(struct pipe_inode_info
*pipe
, unsigned int *cnt
)
873 while (cur
== *cnt
) {
875 if (signal_pending(current
))
878 return cur
== *cnt
? -ERESTARTSYS
: 0;
881 static void wake_up_partner(struct pipe_inode_info
*pipe
)
883 wake_up_interruptible(&pipe
->wait
);
886 static int fifo_open(struct inode
*inode
, struct file
*filp
)
888 struct pipe_inode_info
*pipe
;
889 bool is_pipe
= inode
->i_sb
->s_magic
== PIPEFS_MAGIC
;
894 spin_lock(&inode
->i_lock
);
896 pipe
= inode
->i_pipe
;
898 spin_unlock(&inode
->i_lock
);
900 spin_unlock(&inode
->i_lock
);
901 pipe
= alloc_pipe_info();
905 spin_lock(&inode
->i_lock
);
906 if (unlikely(inode
->i_pipe
)) {
907 inode
->i_pipe
->files
++;
908 spin_unlock(&inode
->i_lock
);
909 free_pipe_info(pipe
);
910 pipe
= inode
->i_pipe
;
912 inode
->i_pipe
= pipe
;
913 spin_unlock(&inode
->i_lock
);
916 filp
->private_data
= pipe
;
917 /* OK, we have a pipe and it's pinned down */
921 /* We can only do regular read/write on fifos */
922 filp
->f_mode
&= (FMODE_READ
| FMODE_WRITE
);
924 switch (filp
->f_mode
) {
928 * POSIX.1 says that O_NONBLOCK means return with the FIFO
929 * opened, even when there is no process writing the FIFO.
932 if (pipe
->readers
++ == 0)
933 wake_up_partner(pipe
);
935 if (!is_pipe
&& !pipe
->writers
) {
936 if ((filp
->f_flags
& O_NONBLOCK
)) {
937 /* suppress POLLHUP until we have
939 filp
->f_version
= pipe
->w_counter
;
941 if (wait_for_partner(pipe
, &pipe
->w_counter
))
950 * POSIX.1 says that O_NONBLOCK means return -1 with
951 * errno=ENXIO when there is no process reading the FIFO.
954 if (!is_pipe
&& (filp
->f_flags
& O_NONBLOCK
) && !pipe
->readers
)
958 if (!pipe
->writers
++)
959 wake_up_partner(pipe
);
961 if (!is_pipe
&& !pipe
->readers
) {
962 if (wait_for_partner(pipe
, &pipe
->r_counter
))
967 case FMODE_READ
| FMODE_WRITE
:
970 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
971 * This implementation will NEVER block on a O_RDWR open, since
972 * the process can at least talk to itself.
979 if (pipe
->readers
== 1 || pipe
->writers
== 1)
980 wake_up_partner(pipe
);
993 if (!--pipe
->readers
)
994 wake_up_interruptible(&pipe
->wait
);
999 if (!--pipe
->writers
)
1000 wake_up_interruptible(&pipe
->wait
);
1005 __pipe_unlock(pipe
);
1007 put_pipe_info(inode
, pipe
);
1011 const struct file_operations pipefifo_fops
= {
1013 .llseek
= no_llseek
,
1014 .read_iter
= pipe_read
,
1015 .write_iter
= pipe_write
,
1017 .unlocked_ioctl
= pipe_ioctl
,
1018 .release
= pipe_release
,
1019 .fasync
= pipe_fasync
,
1023 * Currently we rely on the pipe array holding a power-of-2 number
1026 static inline unsigned int round_pipe_size(unsigned int size
)
1028 unsigned long nr_pages
;
1030 nr_pages
= (size
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1031 return roundup_pow_of_two(nr_pages
) << PAGE_SHIFT
;
1035 * Allocate a new array of pipe buffers and copy the info over. Returns the
1036 * pipe size if successful, or return -ERROR on error.
1038 static long pipe_set_size(struct pipe_inode_info
*pipe
, unsigned long arg
)
1040 struct pipe_buffer
*bufs
;
1041 unsigned int size
, nr_pages
;
1042 unsigned long user_bufs
;
1045 size
= round_pipe_size(arg
);
1046 nr_pages
= size
>> PAGE_SHIFT
;
1052 * If trying to increase the pipe capacity, check that an
1053 * unprivileged user is not trying to exceed various limits
1054 * (soft limit check here, hard limit check just below).
1055 * Decreasing the pipe capacity is always permitted, even
1056 * if the user is currently over a limit.
1058 if (nr_pages
> pipe
->buffers
&&
1059 size
> pipe_max_size
&& !capable(CAP_SYS_RESOURCE
))
1062 user_bufs
= account_pipe_buffers(pipe
->user
, pipe
->buffers
, nr_pages
);
1064 if (nr_pages
> pipe
->buffers
&&
1065 (too_many_pipe_buffers_hard(user_bufs
) ||
1066 too_many_pipe_buffers_soft(user_bufs
)) &&
1067 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
)) {
1069 goto out_revert_acct
;
1073 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1074 * expect a lot of shrink+grow operations, just free and allocate
1075 * again like we would do for growing. If the pipe currently
1076 * contains more buffers than arg, then return busy.
1078 if (nr_pages
< pipe
->nrbufs
) {
1080 goto out_revert_acct
;
1083 bufs
= kcalloc(nr_pages
, sizeof(*bufs
),
1084 GFP_KERNEL_ACCOUNT
| __GFP_NOWARN
);
1085 if (unlikely(!bufs
)) {
1087 goto out_revert_acct
;
1091 * The pipe array wraps around, so just start the new one at zero
1092 * and adjust the indexes.
1098 tail
= pipe
->curbuf
+ pipe
->nrbufs
;
1099 if (tail
< pipe
->buffers
)
1102 tail
&= (pipe
->buffers
- 1);
1104 head
= pipe
->nrbufs
- tail
;
1106 memcpy(bufs
, pipe
->bufs
+ pipe
->curbuf
, head
* sizeof(struct pipe_buffer
));
1108 memcpy(bufs
+ head
, pipe
->bufs
, tail
* sizeof(struct pipe_buffer
));
1114 pipe
->buffers
= nr_pages
;
1115 return nr_pages
* PAGE_SIZE
;
1118 (void) account_pipe_buffers(pipe
->user
, nr_pages
, pipe
->buffers
);
1123 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1124 * will return an error.
1126 int pipe_proc_fn(struct ctl_table
*table
, int write
, void __user
*buf
,
1127 size_t *lenp
, loff_t
*ppos
)
1131 ret
= proc_dointvec_minmax(table
, write
, buf
, lenp
, ppos
);
1132 if (ret
< 0 || !write
)
1135 pipe_max_size
= round_pipe_size(pipe_max_size
);
1140 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1141 * location, so checking ->i_pipe is not enough to verify that this is a
1144 struct pipe_inode_info
*get_pipe_info(struct file
*file
)
1146 return file
->f_op
== &pipefifo_fops
? file
->private_data
: NULL
;
1149 long pipe_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
1151 struct pipe_inode_info
*pipe
;
1154 pipe
= get_pipe_info(file
);
1162 ret
= pipe_set_size(pipe
, arg
);
1165 ret
= pipe
->buffers
* PAGE_SIZE
;
1172 __pipe_unlock(pipe
);
1176 static const struct super_operations pipefs_ops
= {
1177 .destroy_inode
= free_inode_nonrcu
,
1178 .statfs
= simple_statfs
,
1182 * pipefs should _never_ be mounted by userland - too much of security hassle,
1183 * no real gain from having the whole whorehouse mounted. So we don't need
1184 * any operations on the root directory. However, we need a non-trivial
1185 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1187 static struct dentry
*pipefs_mount(struct file_system_type
*fs_type
,
1188 int flags
, const char *dev_name
, void *data
)
1190 return mount_pseudo(fs_type
, "pipe:", &pipefs_ops
,
1191 &pipefs_dentry_operations
, PIPEFS_MAGIC
);
1194 static struct file_system_type pipe_fs_type
= {
1196 .mount
= pipefs_mount
,
1197 .kill_sb
= kill_anon_super
,
1200 static int __init
init_pipe_fs(void)
1202 int err
= register_filesystem(&pipe_fs_type
);
1205 pipe_mnt
= kern_mount(&pipe_fs_type
);
1206 if (IS_ERR(pipe_mnt
)) {
1207 err
= PTR_ERR(pipe_mnt
);
1208 unregister_filesystem(&pipe_fs_type
);
1214 fs_initcall(init_pipe_fs
);