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/magic.h>
18 #include <linux/pipe_fs_i.h>
19 #include <linux/uio.h>
20 #include <linux/highmem.h>
21 #include <linux/pagemap.h>
22 #include <linux/audit.h>
23 #include <linux/syscalls.h>
24 #include <linux/fcntl.h>
25 #include <linux/memcontrol.h>
27 #include <linux/uaccess.h>
28 #include <asm/ioctls.h>
33 * The max size that a non-root user is allowed to grow the pipe. Can
34 * be set by root in /proc/sys/fs/pipe-max-size
36 unsigned int pipe_max_size = 1048576;
38 /* Maximum allocatable pages per user. Hard limit is unset by default, soft
39 * matches default values.
41 unsigned long pipe_user_pages_hard;
42 unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
45 * We use a start+len construction, which provides full use of the
47 * -- Florian Coosmann (FGC)
49 * Reads with count = 0 should always return 0.
50 * -- Julian Bradfield 1999-06-07.
52 * FIFOs and Pipes now generate SIGIO for both readers and writers.
53 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
55 * pipe_read & write cleanup
56 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
59 static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
62 mutex_lock_nested(&pipe->mutex, subclass);
65 void pipe_lock(struct pipe_inode_info *pipe)
68 * pipe_lock() nests non-pipe inode locks (for writing to a file)
70 pipe_lock_nested(pipe, I_MUTEX_PARENT);
72 EXPORT_SYMBOL(pipe_lock);
74 void pipe_unlock(struct pipe_inode_info *pipe)
77 mutex_unlock(&pipe->mutex);
79 EXPORT_SYMBOL(pipe_unlock);
81 static inline void __pipe_lock(struct pipe_inode_info *pipe)
83 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
86 static inline void __pipe_unlock(struct pipe_inode_info *pipe)
88 mutex_unlock(&pipe->mutex);
91 void pipe_double_lock(struct pipe_inode_info *pipe1,
92 struct pipe_inode_info *pipe2)
94 BUG_ON(pipe1 == pipe2);
97 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
98 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
100 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
101 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
105 /* Drop the inode semaphore and wait for a pipe event, atomically */
106 void pipe_wait(struct pipe_inode_info *pipe)
111 * Pipes are system-local resources, so sleeping on them
112 * is considered a noninteractive wait:
114 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
117 finish_wait(&pipe->wait, &wait);
121 static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
122 struct pipe_buffer *buf)
124 struct page *page = buf->page;
127 * If nobody else uses this page, and we don't already have a
128 * temporary page, let's keep track of it as a one-deep
129 * allocation cache. (Otherwise just release our reference to it)
131 if (page_count(page) == 1 && !pipe->tmp_page)
132 pipe->tmp_page = page;
137 static int anon_pipe_buf_steal(struct pipe_inode_info *pipe,
138 struct pipe_buffer *buf)
140 struct page *page = buf->page;
142 if (page_count(page) == 1) {
143 if (memcg_kmem_enabled())
144 memcg_kmem_uncharge(page, 0);
145 __SetPageLocked(page);
152 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
153 * @pipe: the pipe that the buffer belongs to
154 * @buf: the buffer to attempt to steal
157 * This function attempts to steal the &struct page attached to
158 * @buf. If successful, this function returns 0 and returns with
159 * the page locked. The caller may then reuse the page for whatever
160 * he wishes; the typical use is insertion into a different file
163 int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
164 struct pipe_buffer *buf)
166 struct page *page = buf->page;
169 * A reference of one is golden, that means that the owner of this
170 * page is the only one holding a reference to it. lock the page
173 if (page_count(page) == 1) {
180 EXPORT_SYMBOL(generic_pipe_buf_steal);
183 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
184 * @pipe: the pipe that the buffer belongs to
185 * @buf: the buffer to get a reference to
188 * This function grabs an extra reference to @buf. It's used in
189 * in the tee() system call, when we duplicate the buffers in one
192 bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
194 return try_get_page(buf->page);
196 EXPORT_SYMBOL(generic_pipe_buf_get);
199 * generic_pipe_buf_confirm - verify contents of the pipe buffer
200 * @info: the pipe that the buffer belongs to
201 * @buf: the buffer to confirm
204 * This function does nothing, because the generic pipe code uses
205 * pages that are always good when inserted into the pipe.
207 int generic_pipe_buf_confirm(struct pipe_inode_info *info,
208 struct pipe_buffer *buf)
212 EXPORT_SYMBOL(generic_pipe_buf_confirm);
215 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
216 * @pipe: the pipe that the buffer belongs to
217 * @buf: the buffer to put a reference to
220 * This function releases a reference to @buf.
222 void generic_pipe_buf_release(struct pipe_inode_info *pipe,
223 struct pipe_buffer *buf)
227 EXPORT_SYMBOL(generic_pipe_buf_release);
229 static const struct pipe_buf_operations anon_pipe_buf_ops = {
231 .confirm = generic_pipe_buf_confirm,
232 .release = anon_pipe_buf_release,
233 .steal = anon_pipe_buf_steal,
234 .get = generic_pipe_buf_get,
237 static const struct pipe_buf_operations anon_pipe_buf_nomerge_ops = {
239 .confirm = generic_pipe_buf_confirm,
240 .release = anon_pipe_buf_release,
241 .steal = anon_pipe_buf_steal,
242 .get = generic_pipe_buf_get,
245 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,
253 void pipe_buf_mark_unmergeable(struct pipe_buffer *buf)
255 if (buf->ops == &anon_pipe_buf_ops)
256 buf->ops = &anon_pipe_buf_nomerge_ops;
260 pipe_read(struct kiocb *iocb, struct iov_iter *to)
262 size_t total_len = iov_iter_count(to);
263 struct file *filp = iocb->ki_filp;
264 struct pipe_inode_info *pipe = filp->private_data;
268 /* Null read succeeds. */
269 if (unlikely(total_len == 0))
276 int bufs = pipe->nrbufs;
278 int curbuf = pipe->curbuf;
279 struct pipe_buffer *buf = pipe->bufs + curbuf;
280 size_t chars = buf->len;
284 if (chars > total_len)
287 error = pipe_buf_confirm(pipe, buf);
294 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
295 if (unlikely(written < chars)) {
301 buf->offset += chars;
304 /* Was it a packet buffer? Clean up and exit */
305 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
311 pipe_buf_release(pipe, buf);
312 curbuf = (curbuf + 1) & (pipe->buffers - 1);
313 pipe->curbuf = curbuf;
314 pipe->nrbufs = --bufs;
319 break; /* common path: read succeeded */
321 if (bufs) /* More to do? */
325 if (!pipe->waiting_writers) {
326 /* syscall merging: Usually we must not sleep
327 * if O_NONBLOCK is set, or if we got some data.
328 * But if a writer sleeps in kernel space, then
329 * we can wait for that data without violating POSIX.
333 if (filp->f_flags & O_NONBLOCK) {
338 if (signal_pending(current)) {
344 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
345 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
351 /* Signal writers asynchronously that there is more room. */
353 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
354 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
361 static inline int is_packetized(struct file *file)
363 return (file->f_flags & O_DIRECT) != 0;
367 pipe_write(struct kiocb *iocb, struct iov_iter *from)
369 struct file *filp = iocb->ki_filp;
370 struct pipe_inode_info *pipe = filp->private_data;
373 size_t total_len = iov_iter_count(from);
376 /* Null write succeeds. */
377 if (unlikely(total_len == 0))
382 if (!pipe->readers) {
383 send_sig(SIGPIPE, current, 0);
388 /* We try to merge small writes */
389 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
390 if (pipe->nrbufs && chars != 0) {
391 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
393 struct pipe_buffer *buf = pipe->bufs + lastbuf;
394 int offset = buf->offset + buf->len;
396 if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
397 ret = pipe_buf_confirm(pipe, buf);
401 ret = copy_page_from_iter(buf->page, offset, chars, from);
402 if (unlikely(ret < chars)) {
408 if (!iov_iter_count(from))
416 if (!pipe->readers) {
417 send_sig(SIGPIPE, current, 0);
423 if (bufs < pipe->buffers) {
424 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
425 struct pipe_buffer *buf = pipe->bufs + newbuf;
426 struct page *page = pipe->tmp_page;
430 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
431 if (unlikely(!page)) {
432 ret = ret ? : -ENOMEM;
435 pipe->tmp_page = page;
437 /* Always wake up, even if the copy fails. Otherwise
438 * we lock up (O_NONBLOCK-)readers that sleep due to
440 * FIXME! Is this really true?
443 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
444 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
451 /* Insert it into the buffer array */
453 buf->ops = &anon_pipe_buf_ops;
457 if (is_packetized(filp)) {
458 buf->ops = &packet_pipe_buf_ops;
459 buf->flags = PIPE_BUF_FLAG_PACKET;
461 pipe->nrbufs = ++bufs;
462 pipe->tmp_page = NULL;
464 if (!iov_iter_count(from))
467 if (bufs < pipe->buffers)
469 if (filp->f_flags & O_NONBLOCK) {
474 if (signal_pending(current)) {
480 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
481 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
484 pipe->waiting_writers++;
486 pipe->waiting_writers--;
491 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
492 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
494 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
495 int err = file_update_time(filp);
498 sb_end_write(file_inode(filp)->i_sb);
503 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
505 struct pipe_inode_info *pipe = filp->private_data;
506 int count, buf, nrbufs;
513 nrbufs = pipe->nrbufs;
514 while (--nrbufs >= 0) {
515 count += pipe->bufs[buf].len;
516 buf = (buf+1) & (pipe->buffers - 1);
520 return put_user(count, (int __user *)arg);
526 /* No kernel lock held - fine */
528 pipe_poll(struct file *filp, poll_table *wait)
531 struct pipe_inode_info *pipe = filp->private_data;
534 poll_wait(filp, &pipe->wait, wait);
536 /* Reading only -- no need for acquiring the semaphore. */
537 nrbufs = pipe->nrbufs;
539 if (filp->f_mode & FMODE_READ) {
540 mask = (nrbufs > 0) ? EPOLLIN | EPOLLRDNORM : 0;
541 if (!pipe->writers && filp->f_version != pipe->w_counter)
545 if (filp->f_mode & FMODE_WRITE) {
546 mask |= (nrbufs < pipe->buffers) ? EPOLLOUT | EPOLLWRNORM : 0;
548 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
549 * behave exactly like pipes for poll().
558 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
562 spin_lock(&inode->i_lock);
563 if (!--pipe->files) {
564 inode->i_pipe = NULL;
567 spin_unlock(&inode->i_lock);
570 free_pipe_info(pipe);
574 pipe_release(struct inode *inode, struct file *file)
576 struct pipe_inode_info *pipe = file->private_data;
579 if (file->f_mode & FMODE_READ)
581 if (file->f_mode & FMODE_WRITE)
584 if (pipe->readers || pipe->writers) {
585 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP);
586 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
587 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
591 put_pipe_info(inode, pipe);
596 pipe_fasync(int fd, struct file *filp, int on)
598 struct pipe_inode_info *pipe = filp->private_data;
602 if (filp->f_mode & FMODE_READ)
603 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
604 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
605 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
606 if (retval < 0 && (filp->f_mode & FMODE_READ))
607 /* this can happen only if on == T */
608 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
614 static unsigned long account_pipe_buffers(struct user_struct *user,
615 unsigned long old, unsigned long new)
617 return atomic_long_add_return(new - old, &user->pipe_bufs);
620 static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
622 unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft);
624 return soft_limit && user_bufs > soft_limit;
627 static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
629 unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard);
631 return hard_limit && user_bufs > hard_limit;
634 static bool is_unprivileged_user(void)
636 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
639 struct pipe_inode_info *alloc_pipe_info(void)
641 struct pipe_inode_info *pipe;
642 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
643 struct user_struct *user = get_current_user();
644 unsigned long user_bufs;
645 unsigned int max_size = READ_ONCE(pipe_max_size);
647 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
651 if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE))
652 pipe_bufs = max_size >> PAGE_SHIFT;
654 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
656 if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
657 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
661 if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
662 goto out_revert_acct;
664 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
668 init_waitqueue_head(&pipe->wait);
669 pipe->r_counter = pipe->w_counter = 1;
670 pipe->buffers = pipe_bufs;
672 mutex_init(&pipe->mutex);
677 (void) account_pipe_buffers(user, pipe_bufs, 0);
684 void free_pipe_info(struct pipe_inode_info *pipe)
688 (void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
689 free_uid(pipe->user);
690 for (i = 0; i < pipe->buffers; i++) {
691 struct pipe_buffer *buf = pipe->bufs + i;
693 pipe_buf_release(pipe, buf);
696 __free_page(pipe->tmp_page);
701 static struct vfsmount *pipe_mnt __read_mostly;
704 * pipefs_dname() is called from d_path().
706 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
708 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
709 d_inode(dentry)->i_ino);
712 static const struct dentry_operations pipefs_dentry_operations = {
713 .d_dname = pipefs_dname,
716 static struct inode * get_pipe_inode(void)
718 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
719 struct pipe_inode_info *pipe;
724 inode->i_ino = get_next_ino();
726 pipe = alloc_pipe_info();
730 inode->i_pipe = pipe;
732 pipe->readers = pipe->writers = 1;
733 inode->i_fop = &pipefifo_fops;
736 * Mark the inode dirty from the very beginning,
737 * that way it will never be moved to the dirty
738 * list because "mark_inode_dirty()" will think
739 * that it already _is_ on the dirty list.
741 inode->i_state = I_DIRTY;
742 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
743 inode->i_uid = current_fsuid();
744 inode->i_gid = current_fsgid();
745 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
756 int create_pipe_files(struct file **res, int flags)
758 struct inode *inode = get_pipe_inode();
764 f = alloc_file_pseudo(inode, pipe_mnt, "",
765 O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)),
768 free_pipe_info(inode->i_pipe);
773 f->private_data = inode->i_pipe;
775 res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK),
777 if (IS_ERR(res[0])) {
778 put_pipe_info(inode, inode->i_pipe);
780 return PTR_ERR(res[0]);
782 res[0]->private_data = inode->i_pipe;
787 static int __do_pipe_flags(int *fd, struct file **files, int flags)
792 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
795 error = create_pipe_files(files, flags);
799 error = get_unused_fd_flags(flags);
804 error = get_unused_fd_flags(flags);
809 audit_fd_pair(fdr, fdw);
822 int do_pipe_flags(int *fd, int flags)
824 struct file *files[2];
825 int error = __do_pipe_flags(fd, files, flags);
827 fd_install(fd[0], files[0]);
828 fd_install(fd[1], files[1]);
834 * sys_pipe() is the normal C calling standard for creating
835 * a pipe. It's not the way Unix traditionally does this, though.
837 static int do_pipe2(int __user *fildes, int flags)
839 struct file *files[2];
843 error = __do_pipe_flags(fd, files, flags);
845 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
848 put_unused_fd(fd[0]);
849 put_unused_fd(fd[1]);
852 fd_install(fd[0], files[0]);
853 fd_install(fd[1], files[1]);
859 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
861 return do_pipe2(fildes, flags);
864 SYSCALL_DEFINE1(pipe, int __user *, fildes)
866 return do_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 EPOLLHUP 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
1024 * of pages. Returns 0 on error.
1026 unsigned int round_pipe_size(unsigned long size)
1028 if (size > (1U << 31))
1031 /* Minimum pipe size, as required by POSIX */
1032 if (size < PAGE_SIZE)
1035 return roundup_pow_of_two(size);
1039 * Allocate a new array of pipe buffers and copy the info over. Returns the
1040 * pipe size if successful, or return -ERROR on error.
1042 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1044 struct pipe_buffer *bufs;
1045 unsigned int size, nr_pages;
1046 unsigned long user_bufs;
1049 size = round_pipe_size(arg);
1050 nr_pages = size >> PAGE_SHIFT;
1056 * If trying to increase the pipe capacity, check that an
1057 * unprivileged user is not trying to exceed various limits
1058 * (soft limit check here, hard limit check just below).
1059 * Decreasing the pipe capacity is always permitted, even
1060 * if the user is currently over a limit.
1062 if (nr_pages > pipe->buffers &&
1063 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1066 user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
1068 if (nr_pages > pipe->buffers &&
1069 (too_many_pipe_buffers_hard(user_bufs) ||
1070 too_many_pipe_buffers_soft(user_bufs)) &&
1071 is_unprivileged_user()) {
1073 goto out_revert_acct;
1077 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1078 * expect a lot of shrink+grow operations, just free and allocate
1079 * again like we would do for growing. If the pipe currently
1080 * contains more buffers than arg, then return busy.
1082 if (nr_pages < pipe->nrbufs) {
1084 goto out_revert_acct;
1087 bufs = kcalloc(nr_pages, sizeof(*bufs),
1088 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1089 if (unlikely(!bufs)) {
1091 goto out_revert_acct;
1095 * The pipe array wraps around, so just start the new one at zero
1096 * and adjust the indexes.
1102 tail = pipe->curbuf + pipe->nrbufs;
1103 if (tail < pipe->buffers)
1106 tail &= (pipe->buffers - 1);
1108 head = pipe->nrbufs - tail;
1110 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1112 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1118 pipe->buffers = nr_pages;
1119 return nr_pages * PAGE_SIZE;
1122 (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
1127 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1128 * location, so checking ->i_pipe is not enough to verify that this is a
1131 struct pipe_inode_info *get_pipe_info(struct file *file)
1133 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1136 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1138 struct pipe_inode_info *pipe;
1141 pipe = get_pipe_info(file);
1149 ret = pipe_set_size(pipe, arg);
1152 ret = pipe->buffers * PAGE_SIZE;
1159 __pipe_unlock(pipe);
1163 static const struct super_operations pipefs_ops = {
1164 .destroy_inode = free_inode_nonrcu,
1165 .statfs = simple_statfs,
1169 * pipefs should _never_ be mounted by userland - too much of security hassle,
1170 * no real gain from having the whole whorehouse mounted. So we don't need
1171 * any operations on the root directory. However, we need a non-trivial
1172 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1174 static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1175 int flags, const char *dev_name, void *data)
1177 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1178 &pipefs_dentry_operations, PIPEFS_MAGIC);
1181 static struct file_system_type pipe_fs_type = {
1183 .mount = pipefs_mount,
1184 .kill_sb = kill_anon_super,
1187 static int __init init_pipe_fs(void)
1189 int err = register_filesystem(&pipe_fs_type);
1192 pipe_mnt = kern_mount(&pipe_fs_type);
1193 if (IS_ERR(pipe_mnt)) {
1194 err = PTR_ERR(pipe_mnt);
1195 unregister_filesystem(&pipe_fs_type);
1201 fs_initcall(init_pipe_fs);