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;
274 pipe_read(struct kiocb *iocb, struct iov_iter *to)
276 size_t total_len = iov_iter_count(to);
277 struct file *filp = iocb->ki_filp;
278 struct pipe_inode_info *pipe = filp->private_data;
282 /* Null read succeeds. */
283 if (unlikely(total_len == 0))
290 unsigned int head = pipe->head;
291 unsigned int tail = pipe->tail;
292 unsigned int mask = pipe->ring_size - 1;
294 if (!pipe_empty(head, tail)) {
295 struct pipe_buffer *buf = &pipe->bufs[tail & mask];
296 size_t chars = buf->len;
300 if (chars > total_len)
303 error = pipe_buf_confirm(pipe, buf);
310 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
311 if (unlikely(written < chars)) {
317 buf->offset += chars;
320 /* Was it a packet buffer? Clean up and exit */
321 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
327 pipe_buf_release(pipe, buf);
328 spin_lock_irq(&pipe->wait.lock);
332 if (head - (tail - 1) == pipe->max_usage)
333 wake_up_interruptible_sync_poll_locked(
334 &pipe->wait, EPOLLOUT | EPOLLWRNORM);
335 spin_unlock_irq(&pipe->wait.lock);
336 if (head - (tail - 1) == pipe->max_usage)
337 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
341 break; /* common path: read succeeded */
342 if (!pipe_empty(head, tail)) /* More to do? */
348 if (!pipe->waiting_writers) {
349 /* syscall merging: Usually we must not sleep
350 * if O_NONBLOCK is set, or if we got some data.
351 * But if a writer sleeps in kernel space, then
352 * we can wait for that data without violating POSIX.
356 if (filp->f_flags & O_NONBLOCK) {
361 if (signal_pending(current)) {
370 /* Signal writers asynchronously that there is more room. */
372 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
373 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
380 static inline int is_packetized(struct file *file)
382 return (file->f_flags & O_DIRECT) != 0;
386 pipe_write(struct kiocb *iocb, struct iov_iter *from)
388 struct file *filp = iocb->ki_filp;
389 struct pipe_inode_info *pipe = filp->private_data;
390 unsigned int head, max_usage, mask;
393 size_t total_len = iov_iter_count(from);
396 /* Null write succeeds. */
397 if (unlikely(total_len == 0))
402 if (!pipe->readers) {
403 send_sig(SIGPIPE, current, 0);
409 max_usage = pipe->max_usage;
410 mask = pipe->ring_size - 1;
412 /* We try to merge small writes */
413 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
414 if (!pipe_empty(head, pipe->tail) && chars != 0) {
415 struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask];
416 int offset = buf->offset + buf->len;
418 if (pipe_buf_can_merge(buf) && offset + chars <= PAGE_SIZE) {
419 ret = pipe_buf_confirm(pipe, buf);
423 ret = copy_page_from_iter(buf->page, offset, chars, from);
424 if (unlikely(ret < chars)) {
430 if (!iov_iter_count(from))
436 if (!pipe->readers) {
437 send_sig(SIGPIPE, current, 0);
444 if (!pipe_full(head, pipe->tail, max_usage)) {
445 struct pipe_buffer *buf = &pipe->bufs[head & mask];
446 struct page *page = pipe->tmp_page;
450 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
451 if (unlikely(!page)) {
452 ret = ret ? : -ENOMEM;
455 pipe->tmp_page = page;
458 /* Allocate a slot in the ring in advance and attach an
459 * empty buffer. If we fault or otherwise fail to use
460 * it, either the reader will consume it or it'll still
461 * be there for the next write.
463 spin_lock_irq(&pipe->wait.lock);
466 pipe->head = head + 1;
468 /* Always wake up, even if the copy fails. Otherwise
469 * we lock up (O_NONBLOCK-)readers that sleep due to
471 * FIXME! Is this really true?
473 wake_up_interruptible_sync_poll_locked(
474 &pipe->wait, EPOLLIN | EPOLLRDNORM);
476 spin_unlock_irq(&pipe->wait.lock);
477 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
479 /* Insert it into the buffer array */
480 buf = &pipe->bufs[head & mask];
482 buf->ops = &anon_pipe_buf_ops;
486 if (is_packetized(filp)) {
487 buf->ops = &packet_pipe_buf_ops;
488 buf->flags = PIPE_BUF_FLAG_PACKET;
490 pipe->tmp_page = NULL;
492 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
493 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
502 if (!iov_iter_count(from))
506 if (!pipe_full(head, pipe->tail, max_usage))
509 /* Wait for buffer space to become available. */
510 if (filp->f_flags & O_NONBLOCK) {
515 if (signal_pending(current)) {
520 pipe->waiting_writers++;
522 pipe->waiting_writers--;
527 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
528 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
530 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
531 int err = file_update_time(filp);
534 sb_end_write(file_inode(filp)->i_sb);
539 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
541 struct pipe_inode_info *pipe = filp->private_data;
542 int count, head, tail, mask;
550 mask = pipe->ring_size - 1;
552 while (tail != head) {
553 count += pipe->bufs[tail & mask].len;
558 return put_user(count, (int __user *)arg);
564 /* No kernel lock held - fine */
566 pipe_poll(struct file *filp, poll_table *wait)
569 struct pipe_inode_info *pipe = filp->private_data;
570 unsigned int head = READ_ONCE(pipe->head);
571 unsigned int tail = READ_ONCE(pipe->tail);
573 poll_wait(filp, &pipe->wait, wait);
575 BUG_ON(pipe_occupancy(head, tail) > pipe->ring_size);
577 /* Reading only -- no need for acquiring the semaphore. */
579 if (filp->f_mode & FMODE_READ) {
580 if (!pipe_empty(head, tail))
581 mask |= EPOLLIN | EPOLLRDNORM;
582 if (!pipe->writers && filp->f_version != pipe->w_counter)
586 if (filp->f_mode & FMODE_WRITE) {
587 if (!pipe_full(head, tail, pipe->max_usage))
588 mask |= EPOLLOUT | EPOLLWRNORM;
590 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
591 * behave exactly like pipes for poll().
600 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
604 spin_lock(&inode->i_lock);
605 if (!--pipe->files) {
606 inode->i_pipe = NULL;
609 spin_unlock(&inode->i_lock);
612 free_pipe_info(pipe);
616 pipe_release(struct inode *inode, struct file *file)
618 struct pipe_inode_info *pipe = file->private_data;
621 if (file->f_mode & FMODE_READ)
623 if (file->f_mode & FMODE_WRITE)
626 if (pipe->readers || pipe->writers) {
627 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP);
628 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
629 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
633 put_pipe_info(inode, pipe);
638 pipe_fasync(int fd, struct file *filp, int on)
640 struct pipe_inode_info *pipe = filp->private_data;
644 if (filp->f_mode & FMODE_READ)
645 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
646 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
647 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
648 if (retval < 0 && (filp->f_mode & FMODE_READ))
649 /* this can happen only if on == T */
650 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
656 static unsigned long account_pipe_buffers(struct user_struct *user,
657 unsigned long old, unsigned long new)
659 return atomic_long_add_return(new - old, &user->pipe_bufs);
662 static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
664 unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft);
666 return soft_limit && user_bufs > soft_limit;
669 static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
671 unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard);
673 return hard_limit && user_bufs > hard_limit;
676 static bool is_unprivileged_user(void)
678 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
681 struct pipe_inode_info *alloc_pipe_info(void)
683 struct pipe_inode_info *pipe;
684 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
685 struct user_struct *user = get_current_user();
686 unsigned long user_bufs;
687 unsigned int max_size = READ_ONCE(pipe_max_size);
689 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
693 if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE))
694 pipe_bufs = max_size >> PAGE_SHIFT;
696 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
698 if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
699 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
703 if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
704 goto out_revert_acct;
706 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
710 init_waitqueue_head(&pipe->wait);
711 pipe->r_counter = pipe->w_counter = 1;
712 pipe->max_usage = pipe_bufs;
713 pipe->ring_size = pipe_bufs;
715 mutex_init(&pipe->mutex);
720 (void) account_pipe_buffers(user, pipe_bufs, 0);
727 void free_pipe_info(struct pipe_inode_info *pipe)
731 (void) account_pipe_buffers(pipe->user, pipe->ring_size, 0);
732 free_uid(pipe->user);
733 for (i = 0; i < pipe->ring_size; i++) {
734 struct pipe_buffer *buf = pipe->bufs + i;
736 pipe_buf_release(pipe, buf);
739 __free_page(pipe->tmp_page);
744 static struct vfsmount *pipe_mnt __read_mostly;
747 * pipefs_dname() is called from d_path().
749 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
751 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
752 d_inode(dentry)->i_ino);
755 static const struct dentry_operations pipefs_dentry_operations = {
756 .d_dname = pipefs_dname,
759 static struct inode * get_pipe_inode(void)
761 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
762 struct pipe_inode_info *pipe;
767 inode->i_ino = get_next_ino();
769 pipe = alloc_pipe_info();
773 inode->i_pipe = pipe;
775 pipe->readers = pipe->writers = 1;
776 inode->i_fop = &pipefifo_fops;
779 * Mark the inode dirty from the very beginning,
780 * that way it will never be moved to the dirty
781 * list because "mark_inode_dirty()" will think
782 * that it already _is_ on the dirty list.
784 inode->i_state = I_DIRTY;
785 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
786 inode->i_uid = current_fsuid();
787 inode->i_gid = current_fsgid();
788 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
799 int create_pipe_files(struct file **res, int flags)
801 struct inode *inode = get_pipe_inode();
807 f = alloc_file_pseudo(inode, pipe_mnt, "",
808 O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)),
811 free_pipe_info(inode->i_pipe);
816 f->private_data = inode->i_pipe;
818 res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK),
820 if (IS_ERR(res[0])) {
821 put_pipe_info(inode, inode->i_pipe);
823 return PTR_ERR(res[0]);
825 res[0]->private_data = inode->i_pipe;
830 static int __do_pipe_flags(int *fd, struct file **files, int flags)
835 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
838 error = create_pipe_files(files, flags);
842 error = get_unused_fd_flags(flags);
847 error = get_unused_fd_flags(flags);
852 audit_fd_pair(fdr, fdw);
865 int do_pipe_flags(int *fd, int flags)
867 struct file *files[2];
868 int error = __do_pipe_flags(fd, files, flags);
870 fd_install(fd[0], files[0]);
871 fd_install(fd[1], files[1]);
877 * sys_pipe() is the normal C calling standard for creating
878 * a pipe. It's not the way Unix traditionally does this, though.
880 static int do_pipe2(int __user *fildes, int flags)
882 struct file *files[2];
886 error = __do_pipe_flags(fd, files, flags);
888 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
891 put_unused_fd(fd[0]);
892 put_unused_fd(fd[1]);
895 fd_install(fd[0], files[0]);
896 fd_install(fd[1], files[1]);
902 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
904 return do_pipe2(fildes, flags);
907 SYSCALL_DEFINE1(pipe, int __user *, fildes)
909 return do_pipe2(fildes, 0);
912 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
916 while (cur == *cnt) {
918 if (signal_pending(current))
921 return cur == *cnt ? -ERESTARTSYS : 0;
924 static void wake_up_partner(struct pipe_inode_info *pipe)
926 wake_up_interruptible(&pipe->wait);
929 static int fifo_open(struct inode *inode, struct file *filp)
931 struct pipe_inode_info *pipe;
932 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
937 spin_lock(&inode->i_lock);
939 pipe = inode->i_pipe;
941 spin_unlock(&inode->i_lock);
943 spin_unlock(&inode->i_lock);
944 pipe = alloc_pipe_info();
948 spin_lock(&inode->i_lock);
949 if (unlikely(inode->i_pipe)) {
950 inode->i_pipe->files++;
951 spin_unlock(&inode->i_lock);
952 free_pipe_info(pipe);
953 pipe = inode->i_pipe;
955 inode->i_pipe = pipe;
956 spin_unlock(&inode->i_lock);
959 filp->private_data = pipe;
960 /* OK, we have a pipe and it's pinned down */
964 /* We can only do regular read/write on fifos */
965 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
967 switch (filp->f_mode) {
971 * POSIX.1 says that O_NONBLOCK means return with the FIFO
972 * opened, even when there is no process writing the FIFO.
975 if (pipe->readers++ == 0)
976 wake_up_partner(pipe);
978 if (!is_pipe && !pipe->writers) {
979 if ((filp->f_flags & O_NONBLOCK)) {
980 /* suppress EPOLLHUP until we have
982 filp->f_version = pipe->w_counter;
984 if (wait_for_partner(pipe, &pipe->w_counter))
993 * POSIX.1 says that O_NONBLOCK means return -1 with
994 * errno=ENXIO when there is no process reading the FIFO.
997 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
1001 if (!pipe->writers++)
1002 wake_up_partner(pipe);
1004 if (!is_pipe && !pipe->readers) {
1005 if (wait_for_partner(pipe, &pipe->r_counter))
1010 case FMODE_READ | FMODE_WRITE:
1013 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
1014 * This implementation will NEVER block on a O_RDWR open, since
1015 * the process can at least talk to itself.
1022 if (pipe->readers == 1 || pipe->writers == 1)
1023 wake_up_partner(pipe);
1032 __pipe_unlock(pipe);
1036 if (!--pipe->readers)
1037 wake_up_interruptible(&pipe->wait);
1042 if (!--pipe->writers)
1043 wake_up_interruptible(&pipe->wait);
1048 __pipe_unlock(pipe);
1050 put_pipe_info(inode, pipe);
1054 const struct file_operations pipefifo_fops = {
1056 .llseek = no_llseek,
1057 .read_iter = pipe_read,
1058 .write_iter = pipe_write,
1060 .unlocked_ioctl = pipe_ioctl,
1061 .release = pipe_release,
1062 .fasync = pipe_fasync,
1066 * Currently we rely on the pipe array holding a power-of-2 number
1067 * of pages. Returns 0 on error.
1069 unsigned int round_pipe_size(unsigned long size)
1071 if (size > (1U << 31))
1074 /* Minimum pipe size, as required by POSIX */
1075 if (size < PAGE_SIZE)
1078 return roundup_pow_of_two(size);
1082 * Allocate a new array of pipe buffers and copy the info over. Returns the
1083 * pipe size if successful, or return -ERROR on error.
1085 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1087 struct pipe_buffer *bufs;
1088 unsigned int size, nr_slots, head, tail, mask, n;
1089 unsigned long user_bufs;
1092 size = round_pipe_size(arg);
1093 nr_slots = size >> PAGE_SHIFT;
1099 * If trying to increase the pipe capacity, check that an
1100 * unprivileged user is not trying to exceed various limits
1101 * (soft limit check here, hard limit check just below).
1102 * Decreasing the pipe capacity is always permitted, even
1103 * if the user is currently over a limit.
1105 if (nr_slots > pipe->ring_size &&
1106 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1109 user_bufs = account_pipe_buffers(pipe->user, pipe->ring_size, nr_slots);
1111 if (nr_slots > pipe->ring_size &&
1112 (too_many_pipe_buffers_hard(user_bufs) ||
1113 too_many_pipe_buffers_soft(user_bufs)) &&
1114 is_unprivileged_user()) {
1116 goto out_revert_acct;
1120 * We can shrink the pipe, if arg is greater than the ring occupancy.
1121 * Since we don't expect a lot of shrink+grow operations, just free and
1122 * allocate again like we would do for growing. If the pipe currently
1123 * contains more buffers than arg, then return busy.
1125 mask = pipe->ring_size - 1;
1128 n = pipe_occupancy(pipe->head, pipe->tail);
1131 goto out_revert_acct;
1134 bufs = kcalloc(nr_slots, sizeof(*bufs),
1135 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1136 if (unlikely(!bufs)) {
1138 goto out_revert_acct;
1142 * The pipe array wraps around, so just start the new one at zero
1143 * and adjust the indices.
1146 unsigned int h = head & mask;
1147 unsigned int t = tail & mask;
1149 memcpy(bufs, pipe->bufs + t,
1150 n * sizeof(struct pipe_buffer));
1152 unsigned int tsize = pipe->ring_size - t;
1154 memcpy(bufs + tsize, pipe->bufs,
1155 h * sizeof(struct pipe_buffer));
1156 memcpy(bufs, pipe->bufs + t,
1157 tsize * sizeof(struct pipe_buffer));
1166 pipe->ring_size = nr_slots;
1167 pipe->max_usage = nr_slots;
1170 return pipe->max_usage * PAGE_SIZE;
1173 (void) account_pipe_buffers(pipe->user, nr_slots, pipe->ring_size);
1178 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1179 * location, so checking ->i_pipe is not enough to verify that this is a
1182 struct pipe_inode_info *get_pipe_info(struct file *file)
1184 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1187 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1189 struct pipe_inode_info *pipe;
1192 pipe = get_pipe_info(file);
1200 ret = pipe_set_size(pipe, arg);
1203 ret = pipe->max_usage * PAGE_SIZE;
1210 __pipe_unlock(pipe);
1214 static const struct super_operations pipefs_ops = {
1215 .destroy_inode = free_inode_nonrcu,
1216 .statfs = simple_statfs,
1220 * pipefs should _never_ be mounted by userland - too much of security hassle,
1221 * no real gain from having the whole whorehouse mounted. So we don't need
1222 * any operations on the root directory. However, we need a non-trivial
1223 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1226 static int pipefs_init_fs_context(struct fs_context *fc)
1228 struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC);
1231 ctx->ops = &pipefs_ops;
1232 ctx->dops = &pipefs_dentry_operations;
1236 static struct file_system_type pipe_fs_type = {
1238 .init_fs_context = pipefs_init_fs_context,
1239 .kill_sb = kill_anon_super,
1242 static int __init init_pipe_fs(void)
1244 int err = register_filesystem(&pipe_fs_type);
1247 pipe_mnt = kern_mount(&pipe_fs_type);
1248 if (IS_ERR(pipe_mnt)) {
1249 err = PTR_ERR(pipe_mnt);
1250 unregister_filesystem(&pipe_fs_type);
1256 fs_initcall(init_pipe_fs);