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 wake_up_interruptible_sync_poll_locked(
333 &pipe->wait, EPOLLOUT | EPOLLWRNORM);
334 spin_unlock_irq(&pipe->wait.lock);
335 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
339 break; /* common path: read succeeded */
340 if (!pipe_empty(head, tail)) /* More to do? */
346 if (!pipe->waiting_writers) {
347 /* syscall merging: Usually we must not sleep
348 * if O_NONBLOCK is set, or if we got some data.
349 * But if a writer sleeps in kernel space, then
350 * we can wait for that data without violating POSIX.
354 if (filp->f_flags & O_NONBLOCK) {
359 if (signal_pending(current)) {
365 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
366 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
373 /* Signal writers asynchronously that there is more room. */
375 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLOUT | EPOLLWRNORM);
376 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
383 static inline int is_packetized(struct file *file)
385 return (file->f_flags & O_DIRECT) != 0;
389 pipe_write(struct kiocb *iocb, struct iov_iter *from)
391 struct file *filp = iocb->ki_filp;
392 struct pipe_inode_info *pipe = filp->private_data;
393 unsigned int head, tail, max_usage, mask;
396 size_t total_len = iov_iter_count(from);
399 /* Null write succeeds. */
400 if (unlikely(total_len == 0))
405 if (!pipe->readers) {
406 send_sig(SIGPIPE, current, 0);
413 max_usage = pipe->max_usage;
414 mask = pipe->ring_size - 1;
416 /* We try to merge small writes */
417 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
418 if (!pipe_empty(head, tail) && chars != 0) {
419 struct pipe_buffer *buf = &pipe->bufs[(head - 1) & mask];
420 int offset = buf->offset + buf->len;
422 if (pipe_buf_can_merge(buf) && offset + chars <= PAGE_SIZE) {
423 ret = pipe_buf_confirm(pipe, buf);
427 ret = copy_page_from_iter(buf->page, offset, chars, from);
428 if (unlikely(ret < chars)) {
434 if (!iov_iter_count(from))
440 if (!pipe->readers) {
441 send_sig(SIGPIPE, current, 0);
448 if (!pipe_full(head, tail, max_usage)) {
449 struct pipe_buffer *buf = &pipe->bufs[head & mask];
450 struct page *page = pipe->tmp_page;
454 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
455 if (unlikely(!page)) {
456 ret = ret ? : -ENOMEM;
459 pipe->tmp_page = page;
461 /* Always wake up, even if the copy fails. Otherwise
462 * we lock up (O_NONBLOCK-)readers that sleep due to
464 * FIXME! Is this really true?
467 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
468 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
475 /* Insert it into the buffer array */
477 buf->ops = &anon_pipe_buf_ops;
481 if (is_packetized(filp)) {
482 buf->ops = &packet_pipe_buf_ops;
483 buf->flags = PIPE_BUF_FLAG_PACKET;
488 pipe->tmp_page = NULL;
490 if (!iov_iter_count(from))
494 if (!pipe_full(head, tail, max_usage))
497 /* Wait for buffer space to become available. */
498 if (filp->f_flags & O_NONBLOCK) {
503 if (signal_pending(current)) {
509 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
510 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
513 pipe->waiting_writers++;
515 pipe->waiting_writers--;
520 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLRDNORM);
521 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
523 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
524 int err = file_update_time(filp);
527 sb_end_write(file_inode(filp)->i_sb);
532 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
534 struct pipe_inode_info *pipe = filp->private_data;
535 int count, head, tail, mask;
543 mask = pipe->ring_size - 1;
545 while (tail != head) {
546 count += pipe->bufs[tail & mask].len;
551 return put_user(count, (int __user *)arg);
557 /* No kernel lock held - fine */
559 pipe_poll(struct file *filp, poll_table *wait)
562 struct pipe_inode_info *pipe = filp->private_data;
563 unsigned int head = READ_ONCE(pipe->head);
564 unsigned int tail = READ_ONCE(pipe->tail);
566 poll_wait(filp, &pipe->wait, wait);
568 BUG_ON(pipe_occupancy(head, tail) > pipe->ring_size);
570 /* Reading only -- no need for acquiring the semaphore. */
572 if (filp->f_mode & FMODE_READ) {
573 if (!pipe_empty(head, tail))
574 mask |= EPOLLIN | EPOLLRDNORM;
575 if (!pipe->writers && filp->f_version != pipe->w_counter)
579 if (filp->f_mode & FMODE_WRITE) {
580 if (!pipe_full(head, tail, pipe->max_usage))
581 mask |= EPOLLOUT | EPOLLWRNORM;
583 * Most Unices do not set EPOLLERR for FIFOs but on Linux they
584 * behave exactly like pipes for poll().
593 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
597 spin_lock(&inode->i_lock);
598 if (!--pipe->files) {
599 inode->i_pipe = NULL;
602 spin_unlock(&inode->i_lock);
605 free_pipe_info(pipe);
609 pipe_release(struct inode *inode, struct file *file)
611 struct pipe_inode_info *pipe = file->private_data;
614 if (file->f_mode & FMODE_READ)
616 if (file->f_mode & FMODE_WRITE)
619 if (pipe->readers || pipe->writers) {
620 wake_up_interruptible_sync_poll(&pipe->wait, EPOLLIN | EPOLLOUT | EPOLLRDNORM | EPOLLWRNORM | EPOLLERR | EPOLLHUP);
621 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
622 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
626 put_pipe_info(inode, pipe);
631 pipe_fasync(int fd, struct file *filp, int on)
633 struct pipe_inode_info *pipe = filp->private_data;
637 if (filp->f_mode & FMODE_READ)
638 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
639 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
640 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
641 if (retval < 0 && (filp->f_mode & FMODE_READ))
642 /* this can happen only if on == T */
643 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
649 static unsigned long account_pipe_buffers(struct user_struct *user,
650 unsigned long old, unsigned long new)
652 return atomic_long_add_return(new - old, &user->pipe_bufs);
655 static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
657 unsigned long soft_limit = READ_ONCE(pipe_user_pages_soft);
659 return soft_limit && user_bufs > soft_limit;
662 static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
664 unsigned long hard_limit = READ_ONCE(pipe_user_pages_hard);
666 return hard_limit && user_bufs > hard_limit;
669 static bool is_unprivileged_user(void)
671 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
674 struct pipe_inode_info *alloc_pipe_info(void)
676 struct pipe_inode_info *pipe;
677 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
678 struct user_struct *user = get_current_user();
679 unsigned long user_bufs;
680 unsigned int max_size = READ_ONCE(pipe_max_size);
682 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
686 if (pipe_bufs * PAGE_SIZE > max_size && !capable(CAP_SYS_RESOURCE))
687 pipe_bufs = max_size >> PAGE_SHIFT;
689 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
691 if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
692 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
696 if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
697 goto out_revert_acct;
699 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
703 init_waitqueue_head(&pipe->wait);
704 pipe->r_counter = pipe->w_counter = 1;
705 pipe->max_usage = pipe_bufs;
706 pipe->ring_size = pipe_bufs;
708 mutex_init(&pipe->mutex);
713 (void) account_pipe_buffers(user, pipe_bufs, 0);
720 void free_pipe_info(struct pipe_inode_info *pipe)
724 (void) account_pipe_buffers(pipe->user, pipe->ring_size, 0);
725 free_uid(pipe->user);
726 for (i = 0; i < pipe->ring_size; i++) {
727 struct pipe_buffer *buf = pipe->bufs + i;
729 pipe_buf_release(pipe, buf);
732 __free_page(pipe->tmp_page);
737 static struct vfsmount *pipe_mnt __read_mostly;
740 * pipefs_dname() is called from d_path().
742 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
744 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
745 d_inode(dentry)->i_ino);
748 static const struct dentry_operations pipefs_dentry_operations = {
749 .d_dname = pipefs_dname,
752 static struct inode * get_pipe_inode(void)
754 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
755 struct pipe_inode_info *pipe;
760 inode->i_ino = get_next_ino();
762 pipe = alloc_pipe_info();
766 inode->i_pipe = pipe;
768 pipe->readers = pipe->writers = 1;
769 inode->i_fop = &pipefifo_fops;
772 * Mark the inode dirty from the very beginning,
773 * that way it will never be moved to the dirty
774 * list because "mark_inode_dirty()" will think
775 * that it already _is_ on the dirty list.
777 inode->i_state = I_DIRTY;
778 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
779 inode->i_uid = current_fsuid();
780 inode->i_gid = current_fsgid();
781 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
792 int create_pipe_files(struct file **res, int flags)
794 struct inode *inode = get_pipe_inode();
800 f = alloc_file_pseudo(inode, pipe_mnt, "",
801 O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT)),
804 free_pipe_info(inode->i_pipe);
809 f->private_data = inode->i_pipe;
811 res[0] = alloc_file_clone(f, O_RDONLY | (flags & O_NONBLOCK),
813 if (IS_ERR(res[0])) {
814 put_pipe_info(inode, inode->i_pipe);
816 return PTR_ERR(res[0]);
818 res[0]->private_data = inode->i_pipe;
823 static int __do_pipe_flags(int *fd, struct file **files, int flags)
828 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
831 error = create_pipe_files(files, flags);
835 error = get_unused_fd_flags(flags);
840 error = get_unused_fd_flags(flags);
845 audit_fd_pair(fdr, fdw);
858 int do_pipe_flags(int *fd, int flags)
860 struct file *files[2];
861 int error = __do_pipe_flags(fd, files, flags);
863 fd_install(fd[0], files[0]);
864 fd_install(fd[1], files[1]);
870 * sys_pipe() is the normal C calling standard for creating
871 * a pipe. It's not the way Unix traditionally does this, though.
873 static int do_pipe2(int __user *fildes, int flags)
875 struct file *files[2];
879 error = __do_pipe_flags(fd, files, flags);
881 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
884 put_unused_fd(fd[0]);
885 put_unused_fd(fd[1]);
888 fd_install(fd[0], files[0]);
889 fd_install(fd[1], files[1]);
895 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
897 return do_pipe2(fildes, flags);
900 SYSCALL_DEFINE1(pipe, int __user *, fildes)
902 return do_pipe2(fildes, 0);
905 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
909 while (cur == *cnt) {
911 if (signal_pending(current))
914 return cur == *cnt ? -ERESTARTSYS : 0;
917 static void wake_up_partner(struct pipe_inode_info *pipe)
919 wake_up_interruptible(&pipe->wait);
922 static int fifo_open(struct inode *inode, struct file *filp)
924 struct pipe_inode_info *pipe;
925 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
930 spin_lock(&inode->i_lock);
932 pipe = inode->i_pipe;
934 spin_unlock(&inode->i_lock);
936 spin_unlock(&inode->i_lock);
937 pipe = alloc_pipe_info();
941 spin_lock(&inode->i_lock);
942 if (unlikely(inode->i_pipe)) {
943 inode->i_pipe->files++;
944 spin_unlock(&inode->i_lock);
945 free_pipe_info(pipe);
946 pipe = inode->i_pipe;
948 inode->i_pipe = pipe;
949 spin_unlock(&inode->i_lock);
952 filp->private_data = pipe;
953 /* OK, we have a pipe and it's pinned down */
957 /* We can only do regular read/write on fifos */
958 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
960 switch (filp->f_mode) {
964 * POSIX.1 says that O_NONBLOCK means return with the FIFO
965 * opened, even when there is no process writing the FIFO.
968 if (pipe->readers++ == 0)
969 wake_up_partner(pipe);
971 if (!is_pipe && !pipe->writers) {
972 if ((filp->f_flags & O_NONBLOCK)) {
973 /* suppress EPOLLHUP until we have
975 filp->f_version = pipe->w_counter;
977 if (wait_for_partner(pipe, &pipe->w_counter))
986 * POSIX.1 says that O_NONBLOCK means return -1 with
987 * errno=ENXIO when there is no process reading the FIFO.
990 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
994 if (!pipe->writers++)
995 wake_up_partner(pipe);
997 if (!is_pipe && !pipe->readers) {
998 if (wait_for_partner(pipe, &pipe->r_counter))
1003 case FMODE_READ | FMODE_WRITE:
1006 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
1007 * This implementation will NEVER block on a O_RDWR open, since
1008 * the process can at least talk to itself.
1015 if (pipe->readers == 1 || pipe->writers == 1)
1016 wake_up_partner(pipe);
1025 __pipe_unlock(pipe);
1029 if (!--pipe->readers)
1030 wake_up_interruptible(&pipe->wait);
1035 if (!--pipe->writers)
1036 wake_up_interruptible(&pipe->wait);
1041 __pipe_unlock(pipe);
1043 put_pipe_info(inode, pipe);
1047 const struct file_operations pipefifo_fops = {
1049 .llseek = no_llseek,
1050 .read_iter = pipe_read,
1051 .write_iter = pipe_write,
1053 .unlocked_ioctl = pipe_ioctl,
1054 .release = pipe_release,
1055 .fasync = pipe_fasync,
1059 * Currently we rely on the pipe array holding a power-of-2 number
1060 * of pages. Returns 0 on error.
1062 unsigned int round_pipe_size(unsigned long size)
1064 if (size > (1U << 31))
1067 /* Minimum pipe size, as required by POSIX */
1068 if (size < PAGE_SIZE)
1071 return roundup_pow_of_two(size);
1075 * Allocate a new array of pipe buffers and copy the info over. Returns the
1076 * pipe size if successful, or return -ERROR on error.
1078 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1080 struct pipe_buffer *bufs;
1081 unsigned int size, nr_slots, head, tail, mask, n;
1082 unsigned long user_bufs;
1085 size = round_pipe_size(arg);
1086 nr_slots = size >> PAGE_SHIFT;
1092 * If trying to increase the pipe capacity, check that an
1093 * unprivileged user is not trying to exceed various limits
1094 * (soft limit check here, hard limit check just below).
1095 * Decreasing the pipe capacity is always permitted, even
1096 * if the user is currently over a limit.
1098 if (nr_slots > pipe->ring_size &&
1099 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1102 user_bufs = account_pipe_buffers(pipe->user, pipe->ring_size, nr_slots);
1104 if (nr_slots > pipe->ring_size &&
1105 (too_many_pipe_buffers_hard(user_bufs) ||
1106 too_many_pipe_buffers_soft(user_bufs)) &&
1107 is_unprivileged_user()) {
1109 goto out_revert_acct;
1113 * We can shrink the pipe, if arg is greater than the ring occupancy.
1114 * Since we don't expect a lot of shrink+grow operations, just free and
1115 * allocate again like we would do for growing. If the pipe currently
1116 * contains more buffers than arg, then return busy.
1118 mask = pipe->ring_size - 1;
1121 n = pipe_occupancy(pipe->head, pipe->tail);
1124 goto out_revert_acct;
1127 bufs = kcalloc(nr_slots, sizeof(*bufs),
1128 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1129 if (unlikely(!bufs)) {
1131 goto out_revert_acct;
1135 * The pipe array wraps around, so just start the new one at zero
1136 * and adjust the indices.
1139 unsigned int h = head & mask;
1140 unsigned int t = tail & mask;
1142 memcpy(bufs, pipe->bufs + t,
1143 n * sizeof(struct pipe_buffer));
1145 unsigned int tsize = pipe->ring_size - t;
1147 memcpy(bufs + tsize, pipe->bufs,
1148 h * sizeof(struct pipe_buffer));
1149 memcpy(bufs, pipe->bufs + t,
1150 tsize * sizeof(struct pipe_buffer));
1159 pipe->ring_size = nr_slots;
1160 pipe->max_usage = nr_slots;
1163 return pipe->max_usage * PAGE_SIZE;
1166 (void) account_pipe_buffers(pipe->user, nr_slots, pipe->ring_size);
1171 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1172 * location, so checking ->i_pipe is not enough to verify that this is a
1175 struct pipe_inode_info *get_pipe_info(struct file *file)
1177 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1180 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1182 struct pipe_inode_info *pipe;
1185 pipe = get_pipe_info(file);
1193 ret = pipe_set_size(pipe, arg);
1196 ret = pipe->max_usage * PAGE_SIZE;
1203 __pipe_unlock(pipe);
1207 static const struct super_operations pipefs_ops = {
1208 .destroy_inode = free_inode_nonrcu,
1209 .statfs = simple_statfs,
1213 * pipefs should _never_ be mounted by userland - too much of security hassle,
1214 * no real gain from having the whole whorehouse mounted. So we don't need
1215 * any operations on the root directory. However, we need a non-trivial
1216 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1219 static int pipefs_init_fs_context(struct fs_context *fc)
1221 struct pseudo_fs_context *ctx = init_pseudo(fc, PIPEFS_MAGIC);
1224 ctx->ops = &pipefs_ops;
1225 ctx->dops = &pipefs_dentry_operations;
1229 static struct file_system_type pipe_fs_type = {
1231 .init_fs_context = pipefs_init_fs_context,
1232 .kill_sb = kill_anon_super,
1235 static int __init init_pipe_fs(void)
1237 int err = register_filesystem(&pipe_fs_type);
1240 pipe_mnt = kern_mount(&pipe_fs_type);
1241 if (IS_ERR(pipe_mnt)) {
1242 err = PTR_ERR(pipe_mnt);
1243 unregister_filesystem(&pipe_fs_type);
1249 fs_initcall(init_pipe_fs);