4 * Copyright (C) 1991, 1992, 1999 Linus Torvalds
8 #include <linux/file.h>
9 #include <linux/poll.h>
10 #include <linux/slab.h>
11 #include <linux/module.h>
12 #include <linux/init.h>
14 #include <linux/log2.h>
15 #include <linux/mount.h>
16 #include <linux/magic.h>
17 #include <linux/pipe_fs_i.h>
18 #include <linux/uio.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/audit.h>
22 #include <linux/syscalls.h>
23 #include <linux/fcntl.h>
24 #include <linux/memcontrol.h>
26 #include <linux/uaccess.h>
27 #include <asm/ioctls.h>
32 * The max size that a non-root user is allowed to grow the pipe. Can
33 * be set by root in /proc/sys/fs/pipe-max-size
35 unsigned int pipe_max_size = 1048576;
38 * Minimum pipe size, as required by POSIX
40 unsigned int pipe_min_size = PAGE_SIZE;
42 /* Maximum allocatable pages per user. Hard limit is unset by default, soft
43 * matches default values.
45 unsigned long pipe_user_pages_hard;
46 unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
49 * We use a start+len construction, which provides full use of the
51 * -- Florian Coosmann (FGC)
53 * Reads with count = 0 should always return 0.
54 * -- Julian Bradfield 1999-06-07.
56 * FIFOs and Pipes now generate SIGIO for both readers and writers.
57 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
59 * pipe_read & write cleanup
60 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
63 static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
66 mutex_lock_nested(&pipe->mutex, subclass);
69 void pipe_lock(struct pipe_inode_info *pipe)
72 * pipe_lock() nests non-pipe inode locks (for writing to a file)
74 pipe_lock_nested(pipe, I_MUTEX_PARENT);
76 EXPORT_SYMBOL(pipe_lock);
78 void pipe_unlock(struct pipe_inode_info *pipe)
81 mutex_unlock(&pipe->mutex);
83 EXPORT_SYMBOL(pipe_unlock);
85 static inline void __pipe_lock(struct pipe_inode_info *pipe)
87 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
90 static inline void __pipe_unlock(struct pipe_inode_info *pipe)
92 mutex_unlock(&pipe->mutex);
95 void pipe_double_lock(struct pipe_inode_info *pipe1,
96 struct pipe_inode_info *pipe2)
98 BUG_ON(pipe1 == pipe2);
101 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
102 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
104 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
105 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
109 /* Drop the inode semaphore and wait for a pipe event, atomically */
110 void pipe_wait(struct pipe_inode_info *pipe)
115 * Pipes are system-local resources, so sleeping on them
116 * is considered a noninteractive wait:
118 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
121 finish_wait(&pipe->wait, &wait);
125 static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
126 struct pipe_buffer *buf)
128 struct page *page = buf->page;
131 * If nobody else uses this page, and we don't already have a
132 * temporary page, let's keep track of it as a one-deep
133 * allocation cache. (Otherwise just release our reference to it)
135 if (page_count(page) == 1 && !pipe->tmp_page)
136 pipe->tmp_page = page;
141 static int anon_pipe_buf_steal(struct pipe_inode_info *pipe,
142 struct pipe_buffer *buf)
144 struct page *page = buf->page;
146 if (page_count(page) == 1) {
147 if (memcg_kmem_enabled())
148 memcg_kmem_uncharge(page, 0);
149 __SetPageLocked(page);
156 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
157 * @pipe: the pipe that the buffer belongs to
158 * @buf: the buffer to attempt to steal
161 * This function attempts to steal the &struct page attached to
162 * @buf. If successful, this function returns 0 and returns with
163 * the page locked. The caller may then reuse the page for whatever
164 * he wishes; the typical use is insertion into a different file
167 int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
168 struct pipe_buffer *buf)
170 struct page *page = buf->page;
173 * A reference of one is golden, that means that the owner of this
174 * page is the only one holding a reference to it. lock the page
177 if (page_count(page) == 1) {
184 EXPORT_SYMBOL(generic_pipe_buf_steal);
187 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
188 * @pipe: the pipe that the buffer belongs to
189 * @buf: the buffer to get a reference to
192 * This function grabs an extra reference to @buf. It's used in
193 * in the tee() system call, when we duplicate the buffers in one
196 void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
200 EXPORT_SYMBOL(generic_pipe_buf_get);
203 * generic_pipe_buf_confirm - verify contents of the pipe buffer
204 * @info: the pipe that the buffer belongs to
205 * @buf: the buffer to confirm
208 * This function does nothing, because the generic pipe code uses
209 * pages that are always good when inserted into the pipe.
211 int generic_pipe_buf_confirm(struct pipe_inode_info *info,
212 struct pipe_buffer *buf)
216 EXPORT_SYMBOL(generic_pipe_buf_confirm);
219 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
220 * @pipe: the pipe that the buffer belongs to
221 * @buf: the buffer to put a reference to
224 * This function releases a reference to @buf.
226 void generic_pipe_buf_release(struct pipe_inode_info *pipe,
227 struct pipe_buffer *buf)
231 EXPORT_SYMBOL(generic_pipe_buf_release);
233 static const struct pipe_buf_operations anon_pipe_buf_ops = {
235 .confirm = generic_pipe_buf_confirm,
236 .release = anon_pipe_buf_release,
237 .steal = anon_pipe_buf_steal,
238 .get = generic_pipe_buf_get,
241 static const struct pipe_buf_operations packet_pipe_buf_ops = {
243 .confirm = generic_pipe_buf_confirm,
244 .release = anon_pipe_buf_release,
245 .steal = anon_pipe_buf_steal,
246 .get = generic_pipe_buf_get,
250 pipe_read(struct kiocb *iocb, struct iov_iter *to)
252 size_t total_len = iov_iter_count(to);
253 struct file *filp = iocb->ki_filp;
254 struct pipe_inode_info *pipe = filp->private_data;
258 /* Null read succeeds. */
259 if (unlikely(total_len == 0))
266 int bufs = pipe->nrbufs;
268 int curbuf = pipe->curbuf;
269 struct pipe_buffer *buf = pipe->bufs + curbuf;
270 size_t chars = buf->len;
274 if (chars > total_len)
277 error = pipe_buf_confirm(pipe, buf);
284 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
285 if (unlikely(written < chars)) {
291 buf->offset += chars;
294 /* Was it a packet buffer? Clean up and exit */
295 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
301 pipe_buf_release(pipe, buf);
302 curbuf = (curbuf + 1) & (pipe->buffers - 1);
303 pipe->curbuf = curbuf;
304 pipe->nrbufs = --bufs;
309 break; /* common path: read succeeded */
311 if (bufs) /* More to do? */
315 if (!pipe->waiting_writers) {
316 /* syscall merging: Usually we must not sleep
317 * if O_NONBLOCK is set, or if we got some data.
318 * But if a writer sleeps in kernel space, then
319 * we can wait for that data without violating POSIX.
323 if (filp->f_flags & O_NONBLOCK) {
328 if (signal_pending(current)) {
334 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
335 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
341 /* Signal writers asynchronously that there is more room. */
343 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
344 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
351 static inline int is_packetized(struct file *file)
353 return (file->f_flags & O_DIRECT) != 0;
357 pipe_write(struct kiocb *iocb, struct iov_iter *from)
359 struct file *filp = iocb->ki_filp;
360 struct pipe_inode_info *pipe = filp->private_data;
363 size_t total_len = iov_iter_count(from);
366 /* Null write succeeds. */
367 if (unlikely(total_len == 0))
372 if (!pipe->readers) {
373 send_sig(SIGPIPE, current, 0);
378 /* We try to merge small writes */
379 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
380 if (pipe->nrbufs && chars != 0) {
381 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
383 struct pipe_buffer *buf = pipe->bufs + lastbuf;
384 int offset = buf->offset + buf->len;
386 if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
387 ret = pipe_buf_confirm(pipe, buf);
391 ret = copy_page_from_iter(buf->page, offset, chars, from);
392 if (unlikely(ret < chars)) {
398 if (!iov_iter_count(from))
406 if (!pipe->readers) {
407 send_sig(SIGPIPE, current, 0);
413 if (bufs < pipe->buffers) {
414 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
415 struct pipe_buffer *buf = pipe->bufs + newbuf;
416 struct page *page = pipe->tmp_page;
420 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
421 if (unlikely(!page)) {
422 ret = ret ? : -ENOMEM;
425 pipe->tmp_page = page;
427 /* Always wake up, even if the copy fails. Otherwise
428 * we lock up (O_NONBLOCK-)readers that sleep due to
430 * FIXME! Is this really true?
433 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
434 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
441 /* Insert it into the buffer array */
443 buf->ops = &anon_pipe_buf_ops;
447 if (is_packetized(filp)) {
448 buf->ops = &packet_pipe_buf_ops;
449 buf->flags = PIPE_BUF_FLAG_PACKET;
451 pipe->nrbufs = ++bufs;
452 pipe->tmp_page = NULL;
454 if (!iov_iter_count(from))
457 if (bufs < pipe->buffers)
459 if (filp->f_flags & O_NONBLOCK) {
464 if (signal_pending(current)) {
470 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
471 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
474 pipe->waiting_writers++;
476 pipe->waiting_writers--;
481 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
482 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
484 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
485 int err = file_update_time(filp);
488 sb_end_write(file_inode(filp)->i_sb);
493 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
495 struct pipe_inode_info *pipe = filp->private_data;
496 int count, buf, nrbufs;
503 nrbufs = pipe->nrbufs;
504 while (--nrbufs >= 0) {
505 count += pipe->bufs[buf].len;
506 buf = (buf+1) & (pipe->buffers - 1);
510 return put_user(count, (int __user *)arg);
516 /* No kernel lock held - fine */
518 pipe_poll(struct file *filp, poll_table *wait)
521 struct pipe_inode_info *pipe = filp->private_data;
524 poll_wait(filp, &pipe->wait, wait);
526 /* Reading only -- no need for acquiring the semaphore. */
527 nrbufs = pipe->nrbufs;
529 if (filp->f_mode & FMODE_READ) {
530 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
531 if (!pipe->writers && filp->f_version != pipe->w_counter)
535 if (filp->f_mode & FMODE_WRITE) {
536 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
538 * Most Unices do not set POLLERR for FIFOs but on Linux they
539 * behave exactly like pipes for poll().
548 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
552 spin_lock(&inode->i_lock);
553 if (!--pipe->files) {
554 inode->i_pipe = NULL;
557 spin_unlock(&inode->i_lock);
560 free_pipe_info(pipe);
564 pipe_release(struct inode *inode, struct file *file)
566 struct pipe_inode_info *pipe = file->private_data;
569 if (file->f_mode & FMODE_READ)
571 if (file->f_mode & FMODE_WRITE)
574 if (pipe->readers || pipe->writers) {
575 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
576 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
577 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
581 put_pipe_info(inode, pipe);
586 pipe_fasync(int fd, struct file *filp, int on)
588 struct pipe_inode_info *pipe = filp->private_data;
592 if (filp->f_mode & FMODE_READ)
593 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
594 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
595 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
596 if (retval < 0 && (filp->f_mode & FMODE_READ))
597 /* this can happen only if on == T */
598 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
604 static unsigned long account_pipe_buffers(struct user_struct *user,
605 unsigned long old, unsigned long new)
607 return atomic_long_add_return(new - old, &user->pipe_bufs);
610 static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
612 return pipe_user_pages_soft && user_bufs >= pipe_user_pages_soft;
615 static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
617 return pipe_user_pages_hard && user_bufs >= pipe_user_pages_hard;
620 struct pipe_inode_info *alloc_pipe_info(void)
622 struct pipe_inode_info *pipe;
623 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
624 struct user_struct *user = get_current_user();
625 unsigned long user_bufs;
627 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
631 if (pipe_bufs * PAGE_SIZE > pipe_max_size && !capable(CAP_SYS_RESOURCE))
632 pipe_bufs = pipe_max_size >> PAGE_SHIFT;
634 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
636 if (too_many_pipe_buffers_soft(user_bufs)) {
637 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
641 if (too_many_pipe_buffers_hard(user_bufs))
642 goto out_revert_acct;
644 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
648 init_waitqueue_head(&pipe->wait);
649 pipe->r_counter = pipe->w_counter = 1;
650 pipe->buffers = pipe_bufs;
652 mutex_init(&pipe->mutex);
657 (void) account_pipe_buffers(user, pipe_bufs, 0);
664 void free_pipe_info(struct pipe_inode_info *pipe)
668 (void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
669 free_uid(pipe->user);
670 for (i = 0; i < pipe->buffers; i++) {
671 struct pipe_buffer *buf = pipe->bufs + i;
673 pipe_buf_release(pipe, buf);
676 __free_page(pipe->tmp_page);
681 static struct vfsmount *pipe_mnt __read_mostly;
684 * pipefs_dname() is called from d_path().
686 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
688 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
689 d_inode(dentry)->i_ino);
692 static const struct dentry_operations pipefs_dentry_operations = {
693 .d_dname = pipefs_dname,
696 static struct inode * get_pipe_inode(void)
698 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
699 struct pipe_inode_info *pipe;
704 inode->i_ino = get_next_ino();
706 pipe = alloc_pipe_info();
710 inode->i_pipe = pipe;
712 pipe->readers = pipe->writers = 1;
713 inode->i_fop = &pipefifo_fops;
716 * Mark the inode dirty from the very beginning,
717 * that way it will never be moved to the dirty
718 * list because "mark_inode_dirty()" will think
719 * that it already _is_ on the dirty list.
721 inode->i_state = I_DIRTY;
722 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
723 inode->i_uid = current_fsuid();
724 inode->i_gid = current_fsgid();
725 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
736 int create_pipe_files(struct file **res, int flags)
739 struct inode *inode = get_pipe_inode();
742 static struct qstr name = { .name = "" };
748 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
751 path.mnt = mntget(pipe_mnt);
753 d_instantiate(path.dentry, inode);
755 f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
761 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
762 f->private_data = inode->i_pipe;
764 res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
765 if (IS_ERR(res[0])) {
766 err = PTR_ERR(res[0]);
771 res[0]->private_data = inode->i_pipe;
772 res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
779 free_pipe_info(inode->i_pipe);
784 free_pipe_info(inode->i_pipe);
789 static int __do_pipe_flags(int *fd, struct file **files, int flags)
794 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
797 error = create_pipe_files(files, flags);
801 error = get_unused_fd_flags(flags);
806 error = get_unused_fd_flags(flags);
811 audit_fd_pair(fdr, fdw);
824 int do_pipe_flags(int *fd, int flags)
826 struct file *files[2];
827 int error = __do_pipe_flags(fd, files, flags);
829 fd_install(fd[0], files[0]);
830 fd_install(fd[1], files[1]);
836 * sys_pipe() is the normal C calling standard for creating
837 * a pipe. It's not the way Unix traditionally does this, though.
839 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
841 struct file *files[2];
845 error = __do_pipe_flags(fd, files, flags);
847 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
850 put_unused_fd(fd[0]);
851 put_unused_fd(fd[1]);
854 fd_install(fd[0], files[0]);
855 fd_install(fd[1], files[1]);
861 SYSCALL_DEFINE1(pipe, int __user *, fildes)
863 return sys_pipe2(fildes, 0);
866 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
870 while (cur == *cnt) {
872 if (signal_pending(current))
875 return cur == *cnt ? -ERESTARTSYS : 0;
878 static void wake_up_partner(struct pipe_inode_info *pipe)
880 wake_up_interruptible(&pipe->wait);
883 static int fifo_open(struct inode *inode, struct file *filp)
885 struct pipe_inode_info *pipe;
886 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
891 spin_lock(&inode->i_lock);
893 pipe = inode->i_pipe;
895 spin_unlock(&inode->i_lock);
897 spin_unlock(&inode->i_lock);
898 pipe = alloc_pipe_info();
902 spin_lock(&inode->i_lock);
903 if (unlikely(inode->i_pipe)) {
904 inode->i_pipe->files++;
905 spin_unlock(&inode->i_lock);
906 free_pipe_info(pipe);
907 pipe = inode->i_pipe;
909 inode->i_pipe = pipe;
910 spin_unlock(&inode->i_lock);
913 filp->private_data = pipe;
914 /* OK, we have a pipe and it's pinned down */
918 /* We can only do regular read/write on fifos */
919 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
921 switch (filp->f_mode) {
925 * POSIX.1 says that O_NONBLOCK means return with the FIFO
926 * opened, even when there is no process writing the FIFO.
929 if (pipe->readers++ == 0)
930 wake_up_partner(pipe);
932 if (!is_pipe && !pipe->writers) {
933 if ((filp->f_flags & O_NONBLOCK)) {
934 /* suppress POLLHUP until we have
936 filp->f_version = pipe->w_counter;
938 if (wait_for_partner(pipe, &pipe->w_counter))
947 * POSIX.1 says that O_NONBLOCK means return -1 with
948 * errno=ENXIO when there is no process reading the FIFO.
951 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
955 if (!pipe->writers++)
956 wake_up_partner(pipe);
958 if (!is_pipe && !pipe->readers) {
959 if (wait_for_partner(pipe, &pipe->r_counter))
964 case FMODE_READ | FMODE_WRITE:
967 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
968 * This implementation will NEVER block on a O_RDWR open, since
969 * the process can at least talk to itself.
976 if (pipe->readers == 1 || pipe->writers == 1)
977 wake_up_partner(pipe);
990 if (!--pipe->readers)
991 wake_up_interruptible(&pipe->wait);
996 if (!--pipe->writers)
997 wake_up_interruptible(&pipe->wait);
1002 __pipe_unlock(pipe);
1004 put_pipe_info(inode, pipe);
1008 const struct file_operations pipefifo_fops = {
1010 .llseek = no_llseek,
1011 .read_iter = pipe_read,
1012 .write_iter = pipe_write,
1014 .unlocked_ioctl = pipe_ioctl,
1015 .release = pipe_release,
1016 .fasync = pipe_fasync,
1020 * Currently we rely on the pipe array holding a power-of-2 number
1023 static inline unsigned int round_pipe_size(unsigned int size)
1025 unsigned long nr_pages;
1027 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1028 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1032 * Allocate a new array of pipe buffers and copy the info over. Returns the
1033 * pipe size if successful, or return -ERROR on error.
1035 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1037 struct pipe_buffer *bufs;
1038 unsigned int size, nr_pages;
1039 unsigned long user_bufs;
1042 size = round_pipe_size(arg);
1043 nr_pages = size >> PAGE_SHIFT;
1049 * If trying to increase the pipe capacity, check that an
1050 * unprivileged user is not trying to exceed various limits
1051 * (soft limit check here, hard limit check just below).
1052 * Decreasing the pipe capacity is always permitted, even
1053 * if the user is currently over a limit.
1055 if (nr_pages > pipe->buffers &&
1056 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1059 user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
1061 if (nr_pages > pipe->buffers &&
1062 (too_many_pipe_buffers_hard(user_bufs) ||
1063 too_many_pipe_buffers_soft(user_bufs)) &&
1064 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) {
1066 goto out_revert_acct;
1070 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1071 * expect a lot of shrink+grow operations, just free and allocate
1072 * again like we would do for growing. If the pipe currently
1073 * contains more buffers than arg, then return busy.
1075 if (nr_pages < pipe->nrbufs) {
1077 goto out_revert_acct;
1080 bufs = kcalloc(nr_pages, sizeof(*bufs),
1081 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1082 if (unlikely(!bufs)) {
1084 goto out_revert_acct;
1088 * The pipe array wraps around, so just start the new one at zero
1089 * and adjust the indexes.
1095 tail = pipe->curbuf + pipe->nrbufs;
1096 if (tail < pipe->buffers)
1099 tail &= (pipe->buffers - 1);
1101 head = pipe->nrbufs - tail;
1103 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1105 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1111 pipe->buffers = nr_pages;
1112 return nr_pages * PAGE_SIZE;
1115 (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
1120 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1121 * will return an error.
1123 int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1124 size_t *lenp, loff_t *ppos)
1128 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1129 if (ret < 0 || !write)
1132 pipe_max_size = round_pipe_size(pipe_max_size);
1137 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1138 * location, so checking ->i_pipe is not enough to verify that this is a
1141 struct pipe_inode_info *get_pipe_info(struct file *file)
1143 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1146 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1148 struct pipe_inode_info *pipe;
1151 pipe = get_pipe_info(file);
1159 ret = pipe_set_size(pipe, arg);
1162 ret = pipe->buffers * PAGE_SIZE;
1169 __pipe_unlock(pipe);
1173 static const struct super_operations pipefs_ops = {
1174 .destroy_inode = free_inode_nonrcu,
1175 .statfs = simple_statfs,
1179 * pipefs should _never_ be mounted by userland - too much of security hassle,
1180 * no real gain from having the whole whorehouse mounted. So we don't need
1181 * any operations on the root directory. However, we need a non-trivial
1182 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1184 static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1185 int flags, const char *dev_name, void *data)
1187 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1188 &pipefs_dentry_operations, PIPEFS_MAGIC);
1191 static struct file_system_type pipe_fs_type = {
1193 .mount = pipefs_mount,
1194 .kill_sb = kill_anon_super,
1197 static int __init init_pipe_fs(void)
1199 int err = register_filesystem(&pipe_fs_type);
1202 pipe_mnt = kern_mount(&pipe_fs_type);
1203 if (IS_ERR(pipe_mnt)) {
1204 err = PTR_ERR(pipe_mnt);
1205 unregister_filesystem(&pipe_fs_type);
1211 fs_initcall(init_pipe_fs);