1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992, 1999 Linus Torvalds
9 #include <linux/file.h>
10 #include <linux/poll.h>
11 #include <linux/slab.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
15 #include <linux/log2.h>
16 #include <linux/mount.h>
17 #include <linux/pseudo_fs.h>
18 #include <linux/magic.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/uio.h>
21 #include <linux/highmem.h>
22 #include <linux/pagemap.h>
23 #include <linux/audit.h>
24 #include <linux/syscalls.h>
25 #include <linux/fcntl.h>
26 #include <linux/memcontrol.h>
28 #include <linux/uaccess.h>
29 #include <asm/ioctls.h>
34 * The max size that a non-root user is allowed to grow the pipe. Can
35 * be set by root in /proc/sys/fs/pipe-max-size
37 unsigned int pipe_max_size
= 1048576;
39 /* Maximum allocatable pages per user. Hard limit is unset by default, soft
40 * matches default values.
42 unsigned long pipe_user_pages_hard
;
43 unsigned long pipe_user_pages_soft
= PIPE_DEF_BUFFERS
* INR_OPEN_CUR
;
46 * We use head and tail indices that aren't masked off, except at the point of
47 * dereference, but rather they're allowed to wrap naturally. This means there
48 * isn't a dead spot in the buffer, but the ring has to be a power of two and
50 * -- David Howells 2019-09-23.
52 * Reads with count = 0 should always return 0.
53 * -- Julian Bradfield 1999-06-07.
55 * FIFOs and Pipes now generate SIGIO for both readers and writers.
56 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
58 * pipe_read & write cleanup
59 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
62 static void pipe_lock_nested(struct pipe_inode_info
*pipe
, int subclass
)
65 mutex_lock_nested(&pipe
->mutex
, subclass
);
68 void pipe_lock(struct pipe_inode_info
*pipe
)
71 * pipe_lock() nests non-pipe inode locks (for writing to a file)
73 pipe_lock_nested(pipe
, I_MUTEX_PARENT
);
75 EXPORT_SYMBOL(pipe_lock
);
77 void pipe_unlock(struct pipe_inode_info
*pipe
)
80 mutex_unlock(&pipe
->mutex
);
82 EXPORT_SYMBOL(pipe_unlock
);
84 static inline void __pipe_lock(struct pipe_inode_info
*pipe
)
86 mutex_lock_nested(&pipe
->mutex
, I_MUTEX_PARENT
);
89 static inline void __pipe_unlock(struct pipe_inode_info
*pipe
)
91 mutex_unlock(&pipe
->mutex
);
94 void pipe_double_lock(struct pipe_inode_info
*pipe1
,
95 struct pipe_inode_info
*pipe2
)
97 BUG_ON(pipe1
== pipe2
);
100 pipe_lock_nested(pipe1
, I_MUTEX_PARENT
);
101 pipe_lock_nested(pipe2
, I_MUTEX_CHILD
);
103 pipe_lock_nested(pipe2
, I_MUTEX_PARENT
);
104 pipe_lock_nested(pipe1
, I_MUTEX_CHILD
);
108 /* Drop the inode semaphore and wait for a pipe event, atomically */
109 void pipe_wait(struct pipe_inode_info
*pipe
)
114 * Pipes are system-local resources, so sleeping on them
115 * is considered a noninteractive wait:
117 prepare_to_wait(&pipe
->wait
, &wait
, TASK_INTERRUPTIBLE
);
120 finish_wait(&pipe
->wait
, &wait
);
124 static void anon_pipe_buf_release(struct pipe_inode_info
*pipe
,
125 struct pipe_buffer
*buf
)
127 struct page
*page
= buf
->page
;
130 * If nobody else uses this page, and we don't already have a
131 * temporary page, let's keep track of it as a one-deep
132 * allocation cache. (Otherwise just release our reference to it)
134 if (page_count(page
) == 1 && !pipe
->tmp_page
)
135 pipe
->tmp_page
= page
;
140 static int anon_pipe_buf_steal(struct pipe_inode_info
*pipe
,
141 struct pipe_buffer
*buf
)
143 struct page
*page
= buf
->page
;
145 if (page_count(page
) == 1) {
146 memcg_kmem_uncharge(page
, 0);
147 __SetPageLocked(page
);
154 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
155 * @pipe: the pipe that the buffer belongs to
156 * @buf: the buffer to attempt to steal
159 * This function attempts to steal the &struct page attached to
160 * @buf. If successful, this function returns 0 and returns with
161 * the page locked. The caller may then reuse the page for whatever
162 * he wishes; the typical use is insertion into a different file
165 int generic_pipe_buf_steal(struct pipe_inode_info
*pipe
,
166 struct pipe_buffer
*buf
)
168 struct page
*page
= buf
->page
;
171 * A reference of one is golden, that means that the owner of this
172 * page is the only one holding a reference to it. lock the page
175 if (page_count(page
) == 1) {
182 EXPORT_SYMBOL(generic_pipe_buf_steal
);
185 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
186 * @pipe: the pipe that the buffer belongs to
187 * @buf: the buffer to get a reference to
190 * This function grabs an extra reference to @buf. It's used in
191 * in the tee() system call, when we duplicate the buffers in one
194 bool generic_pipe_buf_get(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
)
196 return try_get_page(buf
->page
);
198 EXPORT_SYMBOL(generic_pipe_buf_get
);
201 * generic_pipe_buf_confirm - verify contents of the pipe buffer
202 * @info: the pipe that the buffer belongs to
203 * @buf: the buffer to confirm
206 * This function does nothing, because the generic pipe code uses
207 * pages that are always good when inserted into the pipe.
209 int generic_pipe_buf_confirm(struct pipe_inode_info
*info
,
210 struct pipe_buffer
*buf
)
214 EXPORT_SYMBOL(generic_pipe_buf_confirm
);
217 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
218 * @pipe: the pipe that the buffer belongs to
219 * @buf: the buffer to put a reference to
222 * This function releases a reference to @buf.
224 void generic_pipe_buf_release(struct pipe_inode_info
*pipe
,
225 struct pipe_buffer
*buf
)
229 EXPORT_SYMBOL(generic_pipe_buf_release
);
231 /* New data written to a pipe may be appended to a buffer with this type. */
232 static const struct pipe_buf_operations anon_pipe_buf_ops
= {
233 .confirm
= generic_pipe_buf_confirm
,
234 .release
= anon_pipe_buf_release
,
235 .steal
= anon_pipe_buf_steal
,
236 .get
= generic_pipe_buf_get
,
239 static const struct pipe_buf_operations anon_pipe_buf_nomerge_ops
= {
240 .confirm
= generic_pipe_buf_confirm
,
241 .release
= anon_pipe_buf_release
,
242 .steal
= anon_pipe_buf_steal
,
243 .get
= generic_pipe_buf_get
,
246 static const struct pipe_buf_operations packet_pipe_buf_ops
= {
247 .confirm
= generic_pipe_buf_confirm
,
248 .release
= anon_pipe_buf_release
,
249 .steal
= anon_pipe_buf_steal
,
250 .get
= generic_pipe_buf_get
,
254 * pipe_buf_mark_unmergeable - mark a &struct pipe_buffer as unmergeable
255 * @buf: the buffer to mark
258 * This function ensures that no future writes will be merged into the
259 * given &struct pipe_buffer. This is necessary when multiple pipe buffers
260 * share the same backing page.
262 void pipe_buf_mark_unmergeable(struct pipe_buffer
*buf
)
264 if (buf
->ops
== &anon_pipe_buf_ops
)
265 buf
->ops
= &anon_pipe_buf_nomerge_ops
;
268 static bool pipe_buf_can_merge(struct pipe_buffer
*buf
)
270 return buf
->ops
== &anon_pipe_buf_ops
;
273 /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */
274 static inline bool pipe_readable(const struct pipe_inode_info
*pipe
)
276 unsigned int head
= READ_ONCE(pipe
->head
);
277 unsigned int tail
= READ_ONCE(pipe
->tail
);
278 unsigned int writers
= READ_ONCE(pipe
->writers
);
280 return !pipe_empty(head
, tail
) || !writers
;
284 pipe_read(struct kiocb
*iocb
, struct iov_iter
*to
)
286 size_t total_len
= iov_iter_count(to
);
287 struct file
*filp
= iocb
->ki_filp
;
288 struct pipe_inode_info
*pipe
= filp
->private_data
;
292 /* Null read succeeds. */
293 if (unlikely(total_len
== 0))
300 * We only wake up writers if the pipe was full when we started
301 * reading in order to avoid unnecessary wakeups.
303 * But when we do wake up writers, we do so using a sync wakeup
304 * (WF_SYNC), because we want them to get going and generate more
307 was_full
= pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
);
309 unsigned int head
= pipe
->head
;
310 unsigned int tail
= pipe
->tail
;
311 unsigned int mask
= pipe
->ring_size
- 1;
313 if (!pipe_empty(head
, tail
)) {
314 struct pipe_buffer
*buf
= &pipe
->bufs
[tail
& mask
];
315 size_t chars
= buf
->len
;
319 if (chars
> total_len
)
322 error
= pipe_buf_confirm(pipe
, buf
);
329 written
= copy_page_to_iter(buf
->page
, buf
->offset
, chars
, to
);
330 if (unlikely(written
< chars
)) {
336 buf
->offset
+= chars
;
339 /* Was it a packet buffer? Clean up and exit */
340 if (buf
->flags
& PIPE_BUF_FLAG_PACKET
) {
346 pipe_buf_release(pipe
, buf
);
347 spin_lock_irq(&pipe
->wait
.lock
);
350 spin_unlock_irq(&pipe
->wait
.lock
);
354 break; /* common path: read succeeded */
355 if (!pipe_empty(head
, tail
)) /* More to do? */
363 if (filp
->f_flags
& O_NONBLOCK
) {
367 if (signal_pending(current
)) {
374 wake_up_interruptible_sync_poll(&pipe
->wait
, EPOLLOUT
| EPOLLWRNORM
);
375 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
377 wait_event_interruptible(pipe
->wait
, pipe_readable(pipe
));
379 was_full
= pipe_full(pipe
->head
, pipe
->tail
, pipe
->max_usage
);
384 wake_up_interruptible_sync_poll(&pipe
->wait
, EPOLLOUT
| EPOLLWRNORM
);
385 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
392 static inline int is_packetized(struct file
*file
)
394 return (file
->f_flags
& O_DIRECT
) != 0;
397 /* Done while waiting without holding the pipe lock - thus the READ_ONCE() */
398 static inline bool pipe_writable(const struct pipe_inode_info
*pipe
)
400 unsigned int head
= READ_ONCE(pipe
->head
);
401 unsigned int tail
= READ_ONCE(pipe
->tail
);
402 unsigned int max_usage
= READ_ONCE(pipe
->max_usage
);
404 return !pipe_full(head
, tail
, max_usage
) ||
405 !READ_ONCE(pipe
->readers
);
409 pipe_write(struct kiocb
*iocb
, struct iov_iter
*from
)
411 struct file
*filp
= iocb
->ki_filp
;
412 struct pipe_inode_info
*pipe
= filp
->private_data
;
415 size_t total_len
= iov_iter_count(from
);
417 bool was_empty
= false;
419 /* Null write succeeds. */
420 if (unlikely(total_len
== 0))
425 if (!pipe
->readers
) {
426 send_sig(SIGPIPE
, current
, 0);
432 * Only wake up if the pipe started out empty, since
433 * otherwise there should be no readers waiting.
435 * If it wasn't empty we try to merge new data into
438 * That naturally merges small writes, but it also
439 * page-aligs the rest of the writes for large writes
440 * spanning multiple pages.
443 was_empty
= pipe_empty(head
, pipe
->tail
);
444 chars
= total_len
& (PAGE_SIZE
-1);
445 if (chars
&& !was_empty
) {
446 unsigned int mask
= pipe
->ring_size
- 1;
447 struct pipe_buffer
*buf
= &pipe
->bufs
[(head
- 1) & mask
];
448 int offset
= buf
->offset
+ buf
->len
;
450 if (pipe_buf_can_merge(buf
) && offset
+ chars
<= PAGE_SIZE
) {
451 ret
= pipe_buf_confirm(pipe
, buf
);
455 ret
= copy_page_from_iter(buf
->page
, offset
, chars
, from
);
456 if (unlikely(ret
< chars
)) {
462 if (!iov_iter_count(from
))
468 if (!pipe
->readers
) {
469 send_sig(SIGPIPE
, current
, 0);
476 if (!pipe_full(head
, pipe
->tail
, pipe
->max_usage
)) {
477 unsigned int mask
= pipe
->ring_size
- 1;
478 struct pipe_buffer
*buf
= &pipe
->bufs
[head
& mask
];
479 struct page
*page
= pipe
->tmp_page
;
483 page
= alloc_page(GFP_HIGHUSER
| __GFP_ACCOUNT
);
484 if (unlikely(!page
)) {
485 ret
= ret
? : -ENOMEM
;
488 pipe
->tmp_page
= page
;
491 /* Allocate a slot in the ring in advance and attach an
492 * empty buffer. If we fault or otherwise fail to use
493 * it, either the reader will consume it or it'll still
494 * be there for the next write.
496 spin_lock_irq(&pipe
->wait
.lock
);
499 if (pipe_full(head
, pipe
->tail
, pipe
->max_usage
)) {
500 spin_unlock_irq(&pipe
->wait
.lock
);
504 pipe
->head
= head
+ 1;
505 spin_unlock_irq(&pipe
->wait
.lock
);
507 /* Insert it into the buffer array */
508 buf
= &pipe
->bufs
[head
& mask
];
510 buf
->ops
= &anon_pipe_buf_ops
;
514 if (is_packetized(filp
)) {
515 buf
->ops
= &packet_pipe_buf_ops
;
516 buf
->flags
= PIPE_BUF_FLAG_PACKET
;
518 pipe
->tmp_page
= NULL
;
520 copied
= copy_page_from_iter(page
, 0, PAGE_SIZE
, from
);
521 if (unlikely(copied
< PAGE_SIZE
&& iov_iter_count(from
))) {
530 if (!iov_iter_count(from
))
534 if (!pipe_full(head
, pipe
->tail
, pipe
->max_usage
))
537 /* Wait for buffer space to become available. */
538 if (filp
->f_flags
& O_NONBLOCK
) {
543 if (signal_pending(current
)) {
550 * We're going to release the pipe lock and wait for more
551 * space. We wake up any readers if necessary, and then
552 * after waiting we need to re-check whether the pipe
553 * become empty while we dropped the lock.
557 wake_up_interruptible_sync_poll(&pipe
->wait
, EPOLLIN
| EPOLLRDNORM
);
558 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
560 wait_event_interruptible(pipe
->wait
, pipe_writable(pipe
));
562 was_empty
= pipe_empty(head
, pipe
->tail
);
568 * If we do do a wakeup event, we do a 'sync' wakeup, because we
569 * want the reader to start processing things asap, rather than
570 * leave the data pending.
572 * This is particularly important for small writes, because of
573 * how (for example) the GNU make jobserver uses small writes to
574 * wake up pending jobs
577 wake_up_interruptible_sync_poll(&pipe
->wait
, EPOLLIN
| EPOLLRDNORM
);
578 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
580 if (ret
> 0 && sb_start_write_trylock(file_inode(filp
)->i_sb
)) {
581 int err
= file_update_time(filp
);
584 sb_end_write(file_inode(filp
)->i_sb
);
589 static long pipe_ioctl(struct file
*filp
, unsigned int cmd
, unsigned long arg
)
591 struct pipe_inode_info
*pipe
= filp
->private_data
;
592 int count
, head
, tail
, mask
;
600 mask
= pipe
->ring_size
- 1;
602 while (tail
!= head
) {
603 count
+= pipe
->bufs
[tail
& mask
].len
;
608 return put_user(count
, (int __user
*)arg
);
614 /* No kernel lock held - fine */
616 pipe_poll(struct file
*filp
, poll_table
*wait
)
619 struct pipe_inode_info
*pipe
= filp
->private_data
;
620 unsigned int head
, tail
;
623 * Reading only -- no need for acquiring the semaphore.
625 * But because this is racy, the code has to add the
626 * entry to the poll table _first_ ..
628 poll_wait(filp
, &pipe
->wait
, wait
);
631 * .. and only then can you do the racy tests. That way,
632 * if something changes and you got it wrong, the poll
633 * table entry will wake you up and fix it.
635 head
= READ_ONCE(pipe
->head
);
636 tail
= READ_ONCE(pipe
->tail
);
639 if (filp
->f_mode
& FMODE_READ
) {
640 if (!pipe_empty(head
, tail
))
641 mask
|= EPOLLIN
| EPOLLRDNORM
;
642 if (!pipe
->writers
&& filp
->f_version
!= pipe
->w_counter
)
646 if (filp
->f_mode
& FMODE_WRITE
) {
647 if (!pipe_full(head
, tail
, pipe
->max_usage
))
648 mask
|= EPOLLOUT
| EPOLLWRNORM
;
650 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
651 * behave exactly like pipes for poll().
660 static void put_pipe_info(struct inode
*inode
, struct pipe_inode_info
*pipe
)
664 spin_lock(&inode
->i_lock
);
665 if (!--pipe
->files
) {
666 inode
->i_pipe
= NULL
;
669 spin_unlock(&inode
->i_lock
);
672 free_pipe_info(pipe
);
676 pipe_release(struct inode
*inode
, struct file
*file
)
678 struct pipe_inode_info
*pipe
= file
->private_data
;
681 if (file
->f_mode
& FMODE_READ
)
683 if (file
->f_mode
& FMODE_WRITE
)
686 if (pipe
->readers
|| pipe
->writers
) {
687 wake_up_interruptible_sync_poll(&pipe
->wait
, EPOLLIN
| EPOLLOUT
| EPOLLRDNORM
| EPOLLWRNORM
| EPOLLERR
| EPOLLHUP
);
688 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
689 kill_fasync(&pipe
->fasync_writers
, SIGIO
, POLL_OUT
);
693 put_pipe_info(inode
, pipe
);
698 pipe_fasync(int fd
, struct file
*filp
, int on
)
700 struct pipe_inode_info
*pipe
= filp
->private_data
;
704 if (filp
->f_mode
& FMODE_READ
)
705 retval
= fasync_helper(fd
, filp
, on
, &pipe
->fasync_readers
);
706 if ((filp
->f_mode
& FMODE_WRITE
) && retval
>= 0) {
707 retval
= fasync_helper(fd
, filp
, on
, &pipe
->fasync_writers
);
708 if (retval
< 0 && (filp
->f_mode
& FMODE_READ
))
709 /* this can happen only if on == T */
710 fasync_helper(-1, filp
, 0, &pipe
->fasync_readers
);
716 static unsigned long account_pipe_buffers(struct user_struct
*user
,
717 unsigned long old
, unsigned long new)
719 return atomic_long_add_return(new - old
, &user
->pipe_bufs
);
722 static bool too_many_pipe_buffers_soft(unsigned long user_bufs
)
724 unsigned long soft_limit
= READ_ONCE(pipe_user_pages_soft
);
726 return soft_limit
&& user_bufs
> soft_limit
;
729 static bool too_many_pipe_buffers_hard(unsigned long user_bufs
)
731 unsigned long hard_limit
= READ_ONCE(pipe_user_pages_hard
);
733 return hard_limit
&& user_bufs
> hard_limit
;
736 static bool is_unprivileged_user(void)
738 return !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
);
741 struct pipe_inode_info
*alloc_pipe_info(void)
743 struct pipe_inode_info
*pipe
;
744 unsigned long pipe_bufs
= PIPE_DEF_BUFFERS
;
745 struct user_struct
*user
= get_current_user();
746 unsigned long user_bufs
;
747 unsigned int max_size
= READ_ONCE(pipe_max_size
);
749 pipe
= kzalloc(sizeof(struct pipe_inode_info
), GFP_KERNEL_ACCOUNT
);
753 if (pipe_bufs
* PAGE_SIZE
> max_size
&& !capable(CAP_SYS_RESOURCE
))
754 pipe_bufs
= max_size
>> PAGE_SHIFT
;
756 user_bufs
= account_pipe_buffers(user
, 0, pipe_bufs
);
758 if (too_many_pipe_buffers_soft(user_bufs
) && is_unprivileged_user()) {
759 user_bufs
= account_pipe_buffers(user
, pipe_bufs
, 1);
763 if (too_many_pipe_buffers_hard(user_bufs
) && is_unprivileged_user())
764 goto out_revert_acct
;
766 pipe
->bufs
= kcalloc(pipe_bufs
, sizeof(struct pipe_buffer
),
770 init_waitqueue_head(&pipe
->wait
);
771 pipe
->r_counter
= pipe
->w_counter
= 1;
772 pipe
->max_usage
= pipe_bufs
;
773 pipe
->ring_size
= pipe_bufs
;
775 mutex_init(&pipe
->mutex
);
780 (void) account_pipe_buffers(user
, pipe_bufs
, 0);
787 void free_pipe_info(struct pipe_inode_info
*pipe
)
791 (void) account_pipe_buffers(pipe
->user
, pipe
->ring_size
, 0);
792 free_uid(pipe
->user
);
793 for (i
= 0; i
< pipe
->ring_size
; i
++) {
794 struct pipe_buffer
*buf
= pipe
->bufs
+ i
;
796 pipe_buf_release(pipe
, buf
);
799 __free_page(pipe
->tmp_page
);
804 static struct vfsmount
*pipe_mnt __read_mostly
;
807 * pipefs_dname() is called from d_path().
809 static char *pipefs_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
811 return dynamic_dname(dentry
, buffer
, buflen
, "pipe:[%lu]",
812 d_inode(dentry
)->i_ino
);
815 static const struct dentry_operations pipefs_dentry_operations
= {
816 .d_dname
= pipefs_dname
,
819 static struct inode
* get_pipe_inode(void)
821 struct inode
*inode
= new_inode_pseudo(pipe_mnt
->mnt_sb
);
822 struct pipe_inode_info
*pipe
;
827 inode
->i_ino
= get_next_ino();
829 pipe
= alloc_pipe_info();
833 inode
->i_pipe
= pipe
;
835 pipe
->readers
= pipe
->writers
= 1;
836 inode
->i_fop
= &pipefifo_fops
;
839 * Mark the inode dirty from the very beginning,
840 * that way it will never be moved to the dirty
841 * list because "mark_inode_dirty()" will think
842 * that it already _is_ on the dirty list.
844 inode
->i_state
= I_DIRTY
;
845 inode
->i_mode
= S_IFIFO
| S_IRUSR
| S_IWUSR
;
846 inode
->i_uid
= current_fsuid();
847 inode
->i_gid
= current_fsgid();
848 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
859 int create_pipe_files(struct file
**res
, int flags
)
861 struct inode
*inode
= get_pipe_inode();
867 f
= alloc_file_pseudo(inode
, pipe_mnt
, "",
868 O_WRONLY
| (flags
& (O_NONBLOCK
| O_DIRECT
)),
871 free_pipe_info(inode
->i_pipe
);
876 f
->private_data
= inode
->i_pipe
;
878 res
[0] = alloc_file_clone(f
, O_RDONLY
| (flags
& O_NONBLOCK
),
880 if (IS_ERR(res
[0])) {
881 put_pipe_info(inode
, inode
->i_pipe
);
883 return PTR_ERR(res
[0]);
885 res
[0]->private_data
= inode
->i_pipe
;
887 stream_open(inode
, res
[0]);
888 stream_open(inode
, res
[1]);
892 static int __do_pipe_flags(int *fd
, struct file
**files
, int flags
)
897 if (flags
& ~(O_CLOEXEC
| O_NONBLOCK
| O_DIRECT
))
900 error
= create_pipe_files(files
, flags
);
904 error
= get_unused_fd_flags(flags
);
909 error
= get_unused_fd_flags(flags
);
914 audit_fd_pair(fdr
, fdw
);
927 int do_pipe_flags(int *fd
, int flags
)
929 struct file
*files
[2];
930 int error
= __do_pipe_flags(fd
, files
, flags
);
932 fd_install(fd
[0], files
[0]);
933 fd_install(fd
[1], files
[1]);
939 * sys_pipe() is the normal C calling standard for creating
940 * a pipe. It's not the way Unix traditionally does this, though.
942 static int do_pipe2(int __user
*fildes
, int flags
)
944 struct file
*files
[2];
948 error
= __do_pipe_flags(fd
, files
, flags
);
950 if (unlikely(copy_to_user(fildes
, fd
, sizeof(fd
)))) {
953 put_unused_fd(fd
[0]);
954 put_unused_fd(fd
[1]);
957 fd_install(fd
[0], files
[0]);
958 fd_install(fd
[1], files
[1]);
964 SYSCALL_DEFINE2(pipe2
, int __user
*, fildes
, int, flags
)
966 return do_pipe2(fildes
, flags
);
969 SYSCALL_DEFINE1(pipe
, int __user
*, fildes
)
971 return do_pipe2(fildes
, 0);
974 static int wait_for_partner(struct pipe_inode_info
*pipe
, unsigned int *cnt
)
978 while (cur
== *cnt
) {
980 if (signal_pending(current
))
983 return cur
== *cnt
? -ERESTARTSYS
: 0;
986 static void wake_up_partner(struct pipe_inode_info
*pipe
)
988 wake_up_interruptible(&pipe
->wait
);
991 static int fifo_open(struct inode
*inode
, struct file
*filp
)
993 struct pipe_inode_info
*pipe
;
994 bool is_pipe
= inode
->i_sb
->s_magic
== PIPEFS_MAGIC
;
999 spin_lock(&inode
->i_lock
);
1000 if (inode
->i_pipe
) {
1001 pipe
= inode
->i_pipe
;
1003 spin_unlock(&inode
->i_lock
);
1005 spin_unlock(&inode
->i_lock
);
1006 pipe
= alloc_pipe_info();
1010 spin_lock(&inode
->i_lock
);
1011 if (unlikely(inode
->i_pipe
)) {
1012 inode
->i_pipe
->files
++;
1013 spin_unlock(&inode
->i_lock
);
1014 free_pipe_info(pipe
);
1015 pipe
= inode
->i_pipe
;
1017 inode
->i_pipe
= pipe
;
1018 spin_unlock(&inode
->i_lock
);
1021 filp
->private_data
= pipe
;
1022 /* OK, we have a pipe and it's pinned down */
1026 /* We can only do regular read/write on fifos */
1027 stream_open(inode
, filp
);
1029 switch (filp
->f_mode
& (FMODE_READ
| FMODE_WRITE
)) {
1033 * POSIX.1 says that O_NONBLOCK means return with the FIFO
1034 * opened, even when there is no process writing the FIFO.
1037 if (pipe
->readers
++ == 0)
1038 wake_up_partner(pipe
);
1040 if (!is_pipe
&& !pipe
->writers
) {
1041 if ((filp
->f_flags
& O_NONBLOCK
)) {
1042 /* suppress EPOLLHUP until we have
1044 filp
->f_version
= pipe
->w_counter
;
1046 if (wait_for_partner(pipe
, &pipe
->w_counter
))
1055 * POSIX.1 says that O_NONBLOCK means return -1 with
1056 * errno=ENXIO when there is no process reading the FIFO.
1059 if (!is_pipe
&& (filp
->f_flags
& O_NONBLOCK
) && !pipe
->readers
)
1063 if (!pipe
->writers
++)
1064 wake_up_partner(pipe
);
1066 if (!is_pipe
&& !pipe
->readers
) {
1067 if (wait_for_partner(pipe
, &pipe
->r_counter
))
1072 case FMODE_READ
| FMODE_WRITE
:
1075 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
1076 * This implementation will NEVER block on a O_RDWR open, since
1077 * the process can at least talk to itself.
1084 if (pipe
->readers
== 1 || pipe
->writers
== 1)
1085 wake_up_partner(pipe
);
1094 __pipe_unlock(pipe
);
1098 if (!--pipe
->readers
)
1099 wake_up_interruptible(&pipe
->wait
);
1104 if (!--pipe
->writers
)
1105 wake_up_interruptible(&pipe
->wait
);
1110 __pipe_unlock(pipe
);
1112 put_pipe_info(inode
, pipe
);
1116 const struct file_operations pipefifo_fops
= {
1118 .llseek
= no_llseek
,
1119 .read_iter
= pipe_read
,
1120 .write_iter
= pipe_write
,
1122 .unlocked_ioctl
= pipe_ioctl
,
1123 .release
= pipe_release
,
1124 .fasync
= pipe_fasync
,
1128 * Currently we rely on the pipe array holding a power-of-2 number
1129 * of pages. Returns 0 on error.
1131 unsigned int round_pipe_size(unsigned long size
)
1133 if (size
> (1U << 31))
1136 /* Minimum pipe size, as required by POSIX */
1137 if (size
< PAGE_SIZE
)
1140 return roundup_pow_of_two(size
);
1144 * Allocate a new array of pipe buffers and copy the info over. Returns the
1145 * pipe size if successful, or return -ERROR on error.
1147 static long pipe_set_size(struct pipe_inode_info
*pipe
, unsigned long arg
)
1149 struct pipe_buffer
*bufs
;
1150 unsigned int size
, nr_slots
, head
, tail
, mask
, n
;
1151 unsigned long user_bufs
;
1154 size
= round_pipe_size(arg
);
1155 nr_slots
= size
>> PAGE_SHIFT
;
1161 * If trying to increase the pipe capacity, check that an
1162 * unprivileged user is not trying to exceed various limits
1163 * (soft limit check here, hard limit check just below).
1164 * Decreasing the pipe capacity is always permitted, even
1165 * if the user is currently over a limit.
1167 if (nr_slots
> pipe
->ring_size
&&
1168 size
> pipe_max_size
&& !capable(CAP_SYS_RESOURCE
))
1171 user_bufs
= account_pipe_buffers(pipe
->user
, pipe
->ring_size
, nr_slots
);
1173 if (nr_slots
> pipe
->ring_size
&&
1174 (too_many_pipe_buffers_hard(user_bufs
) ||
1175 too_many_pipe_buffers_soft(user_bufs
)) &&
1176 is_unprivileged_user()) {
1178 goto out_revert_acct
;
1182 * We can shrink the pipe, if arg is greater than the ring occupancy.
1183 * Since we don't expect a lot of shrink+grow operations, just free and
1184 * allocate again like we would do for growing. If the pipe currently
1185 * contains more buffers than arg, then return busy.
1187 mask
= pipe
->ring_size
- 1;
1190 n
= pipe_occupancy(pipe
->head
, pipe
->tail
);
1193 goto out_revert_acct
;
1196 bufs
= kcalloc(nr_slots
, sizeof(*bufs
),
1197 GFP_KERNEL_ACCOUNT
| __GFP_NOWARN
);
1198 if (unlikely(!bufs
)) {
1200 goto out_revert_acct
;
1204 * The pipe array wraps around, so just start the new one at zero
1205 * and adjust the indices.
1208 unsigned int h
= head
& mask
;
1209 unsigned int t
= tail
& mask
;
1211 memcpy(bufs
, pipe
->bufs
+ t
,
1212 n
* sizeof(struct pipe_buffer
));
1214 unsigned int tsize
= pipe
->ring_size
- t
;
1216 memcpy(bufs
+ tsize
, pipe
->bufs
,
1217 h
* sizeof(struct pipe_buffer
));
1218 memcpy(bufs
, pipe
->bufs
+ t
,
1219 tsize
* sizeof(struct pipe_buffer
));
1228 pipe
->ring_size
= nr_slots
;
1229 pipe
->max_usage
= nr_slots
;
1232 wake_up_interruptible_all(&pipe
->wait
);
1233 return pipe
->max_usage
* PAGE_SIZE
;
1236 (void) account_pipe_buffers(pipe
->user
, nr_slots
, pipe
->ring_size
);
1241 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1242 * location, so checking ->i_pipe is not enough to verify that this is a
1245 struct pipe_inode_info
*get_pipe_info(struct file
*file
)
1247 return file
->f_op
== &pipefifo_fops
? file
->private_data
: NULL
;
1250 long pipe_fcntl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
1252 struct pipe_inode_info
*pipe
;
1255 pipe
= get_pipe_info(file
);
1263 ret
= pipe_set_size(pipe
, arg
);
1266 ret
= pipe
->max_usage
* PAGE_SIZE
;
1273 __pipe_unlock(pipe
);
1277 static const struct super_operations pipefs_ops
= {
1278 .destroy_inode
= free_inode_nonrcu
,
1279 .statfs
= simple_statfs
,
1283 * pipefs should _never_ be mounted by userland - too much of security hassle,
1284 * no real gain from having the whole whorehouse mounted. So we don't need
1285 * any operations on the root directory. However, we need a non-trivial
1286 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1289 static int pipefs_init_fs_context(struct fs_context
*fc
)
1291 struct pseudo_fs_context
*ctx
= init_pseudo(fc
, PIPEFS_MAGIC
);
1294 ctx
->ops
= &pipefs_ops
;
1295 ctx
->dops
= &pipefs_dentry_operations
;
1299 static struct file_system_type pipe_fs_type
= {
1301 .init_fs_context
= pipefs_init_fs_context
,
1302 .kill_sb
= kill_anon_super
,
1305 static int __init
init_pipe_fs(void)
1307 int err
= register_filesystem(&pipe_fs_type
);
1310 pipe_mnt
= kern_mount(&pipe_fs_type
);
1311 if (IS_ERR(pipe_mnt
)) {
1312 err
= PTR_ERR(pipe_mnt
);
1313 unregister_filesystem(&pipe_fs_type
);
1319 fs_initcall(init_pipe_fs
);