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>
25 #include <asm/uaccess.h>
26 #include <asm/ioctls.h>
31 * New pipe buffers will be restricted to this size while the user is exceeding
32 * their pipe buffer quota. The general pipe use case needs at least two
33 * buffers: one for data yet to be read, and one for new data. If this is less
34 * than two, then a write to a non-empty pipe may block even if the pipe is not
35 * full. This can occur with GNU make jobserver or similar uses of pipes as
36 * semaphores: multiple processes may be waiting to write tokens back to the
37 * pipe before reading tokens: https://lore.kernel.org/lkml/1628086770.5rn8p04n6j.none@localhost/.
39 * Users can reduce their pipe buffers with F_SETPIPE_SZ below this at their
40 * own risk, namely: pipe writes to non-full pipes may block until the pipe is
43 #define PIPE_MIN_DEF_BUFFERS 2
46 * The max size that a non-root user is allowed to grow the pipe. Can
47 * be set by root in /proc/sys/fs/pipe-max-size
49 unsigned int pipe_max_size = 1048576;
52 * Minimum pipe size, as required by POSIX
54 unsigned int pipe_min_size = PAGE_SIZE;
56 /* Maximum allocatable pages per user. Hard limit is unset by default, soft
57 * matches default values.
59 unsigned long pipe_user_pages_hard;
60 unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
63 * We use a start+len construction, which provides full use of the
65 * -- Florian Coosmann (FGC)
67 * Reads with count = 0 should always return 0.
68 * -- Julian Bradfield 1999-06-07.
70 * FIFOs and Pipes now generate SIGIO for both readers and writers.
71 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
73 * pipe_read & write cleanup
74 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
77 static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
80 mutex_lock_nested(&pipe->mutex, subclass);
83 void pipe_lock(struct pipe_inode_info *pipe)
86 * pipe_lock() nests non-pipe inode locks (for writing to a file)
88 pipe_lock_nested(pipe, I_MUTEX_PARENT);
90 EXPORT_SYMBOL(pipe_lock);
92 void pipe_unlock(struct pipe_inode_info *pipe)
95 mutex_unlock(&pipe->mutex);
97 EXPORT_SYMBOL(pipe_unlock);
99 static inline void __pipe_lock(struct pipe_inode_info *pipe)
101 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
104 static inline void __pipe_unlock(struct pipe_inode_info *pipe)
106 mutex_unlock(&pipe->mutex);
109 void pipe_double_lock(struct pipe_inode_info *pipe1,
110 struct pipe_inode_info *pipe2)
112 BUG_ON(pipe1 == pipe2);
115 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
116 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
118 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
119 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
123 /* Drop the inode semaphore and wait for a pipe event, atomically */
124 void pipe_wait(struct pipe_inode_info *pipe)
129 * Pipes are system-local resources, so sleeping on them
130 * is considered a noninteractive wait:
132 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
135 finish_wait(&pipe->wait, &wait);
139 static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
140 struct pipe_buffer *buf)
142 struct page *page = buf->page;
145 * If nobody else uses this page, and we don't already have a
146 * temporary page, let's keep track of it as a one-deep
147 * allocation cache. (Otherwise just release our reference to it)
149 if (page_count(page) == 1 && !pipe->tmp_page)
150 pipe->tmp_page = page;
152 page_cache_release(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 bool generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
198 return try_get_page(buf->page);
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)
229 page_cache_release(buf->page);
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 = generic_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 = generic_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 const struct pipe_buf_operations *ops = buf->ops;
271 size_t chars = buf->len;
275 if (chars > total_len)
278 error = ops->confirm(pipe, buf);
285 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
286 if (unlikely(written < chars)) {
292 buf->offset += chars;
295 /* Was it a packet buffer? Clean up and exit */
296 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
303 ops->release(pipe, buf);
304 curbuf = (curbuf + 1) & (pipe->buffers - 1);
305 pipe->curbuf = curbuf;
306 pipe->nrbufs = --bufs;
311 break; /* common path: read succeeded */
313 if (bufs) /* More to do? */
317 if (!pipe->waiting_writers) {
318 /* syscall merging: Usually we must not sleep
319 * if O_NONBLOCK is set, or if we got some data.
320 * But if a writer sleeps in kernel space, then
321 * we can wait for that data without violating POSIX.
325 if (filp->f_flags & O_NONBLOCK) {
330 if (signal_pending(current)) {
336 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
337 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
343 /* Signal writers asynchronously that there is more room. */
345 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
346 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
353 static inline int is_packetized(struct file *file)
355 return (file->f_flags & O_DIRECT) != 0;
359 pipe_write(struct kiocb *iocb, struct iov_iter *from)
361 struct file *filp = iocb->ki_filp;
362 struct pipe_inode_info *pipe = filp->private_data;
365 size_t total_len = iov_iter_count(from);
368 /* Null write succeeds. */
369 if (unlikely(total_len == 0))
374 if (!pipe->readers) {
375 send_sig(SIGPIPE, current, 0);
380 /* We try to merge small writes */
381 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
382 if (pipe->nrbufs && chars != 0) {
383 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
385 struct pipe_buffer *buf = pipe->bufs + lastbuf;
386 const struct pipe_buf_operations *ops = buf->ops;
387 int offset = buf->offset + buf->len;
389 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
390 ret = ops->confirm(pipe, buf);
394 ret = copy_page_from_iter(buf->page, offset, chars, from);
395 if (unlikely(ret < chars)) {
401 if (!iov_iter_count(from))
409 if (!pipe->readers) {
410 send_sig(SIGPIPE, current, 0);
416 if (bufs < pipe->buffers) {
417 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
418 struct pipe_buffer *buf = pipe->bufs + newbuf;
419 struct page *page = pipe->tmp_page;
423 page = alloc_page(GFP_HIGHUSER);
424 if (unlikely(!page)) {
425 ret = ret ? : -ENOMEM;
428 pipe->tmp_page = page;
430 /* Always wake up, even if the copy fails. Otherwise
431 * we lock up (O_NONBLOCK-)readers that sleep due to
433 * FIXME! Is this really true?
436 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
437 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
444 /* Insert it into the buffer array */
446 buf->ops = &anon_pipe_buf_ops;
450 if (is_packetized(filp)) {
451 buf->ops = &packet_pipe_buf_ops;
452 buf->flags = PIPE_BUF_FLAG_PACKET;
454 pipe->nrbufs = ++bufs;
455 pipe->tmp_page = NULL;
457 if (!iov_iter_count(from))
460 if (bufs < pipe->buffers)
462 if (filp->f_flags & O_NONBLOCK) {
467 if (signal_pending(current)) {
473 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
474 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
477 pipe->waiting_writers++;
479 pipe->waiting_writers--;
484 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
485 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
487 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
488 int err = file_update_time(filp);
491 sb_end_write(file_inode(filp)->i_sb);
496 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
498 struct pipe_inode_info *pipe = filp->private_data;
499 int count, buf, nrbufs;
506 nrbufs = pipe->nrbufs;
507 while (--nrbufs >= 0) {
508 count += pipe->bufs[buf].len;
509 buf = (buf+1) & (pipe->buffers - 1);
513 return put_user(count, (int __user *)arg);
519 /* No kernel lock held - fine */
521 pipe_poll(struct file *filp, poll_table *wait)
524 struct pipe_inode_info *pipe = filp->private_data;
527 poll_wait(filp, &pipe->wait, wait);
529 /* Reading only -- no need for acquiring the semaphore. */
530 nrbufs = pipe->nrbufs;
532 if (filp->f_mode & FMODE_READ) {
533 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
534 if (!pipe->writers && filp->f_version != pipe->w_counter)
538 if (filp->f_mode & FMODE_WRITE) {
539 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
541 * Most Unices do not set POLLERR for FIFOs but on Linux they
542 * behave exactly like pipes for poll().
551 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
555 spin_lock(&inode->i_lock);
556 if (!--pipe->files) {
557 inode->i_pipe = NULL;
560 spin_unlock(&inode->i_lock);
563 free_pipe_info(pipe);
567 pipe_release(struct inode *inode, struct file *file)
569 struct pipe_inode_info *pipe = file->private_data;
572 if (file->f_mode & FMODE_READ)
574 if (file->f_mode & FMODE_WRITE)
577 if (pipe->readers || pipe->writers) {
578 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
579 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
580 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
584 put_pipe_info(inode, pipe);
589 pipe_fasync(int fd, struct file *filp, int on)
591 struct pipe_inode_info *pipe = filp->private_data;
595 if (filp->f_mode & FMODE_READ)
596 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
597 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
598 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
599 if (retval < 0 && (filp->f_mode & FMODE_READ))
600 /* this can happen only if on == T */
601 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
607 static void account_pipe_buffers(struct pipe_inode_info *pipe,
608 unsigned long old, unsigned long new)
610 atomic_long_add(new - old, &pipe->user->pipe_bufs);
613 static bool too_many_pipe_buffers_soft(struct user_struct *user)
615 return pipe_user_pages_soft &&
616 atomic_long_read(&user->pipe_bufs) >= pipe_user_pages_soft;
619 static bool too_many_pipe_buffers_hard(struct user_struct *user)
621 return pipe_user_pages_hard &&
622 atomic_long_read(&user->pipe_bufs) >= pipe_user_pages_hard;
625 struct pipe_inode_info *alloc_pipe_info(void)
627 struct pipe_inode_info *pipe;
629 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
631 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
632 struct user_struct *user = get_current_user();
634 if (pipe_bufs * PAGE_SIZE > pipe_max_size && !capable(CAP_SYS_RESOURCE))
635 pipe_bufs = pipe_max_size >> PAGE_SHIFT;
637 if (!too_many_pipe_buffers_hard(user)) {
638 if (too_many_pipe_buffers_soft(user))
639 pipe_bufs = PIPE_MIN_DEF_BUFFERS;
640 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * pipe_bufs, GFP_KERNEL);
644 init_waitqueue_head(&pipe->wait);
645 pipe->r_counter = pipe->w_counter = 1;
646 pipe->buffers = pipe_bufs;
648 account_pipe_buffers(pipe, 0, pipe_bufs);
649 mutex_init(&pipe->mutex);
659 void free_pipe_info(struct pipe_inode_info *pipe)
663 account_pipe_buffers(pipe, pipe->buffers, 0);
664 free_uid(pipe->user);
665 for (i = 0; i < pipe->buffers; i++) {
666 struct pipe_buffer *buf = pipe->bufs + i;
668 buf->ops->release(pipe, buf);
671 __free_page(pipe->tmp_page);
676 static struct vfsmount *pipe_mnt __read_mostly;
679 * pipefs_dname() is called from d_path().
681 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
683 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
684 d_inode(dentry)->i_ino);
687 static const struct dentry_operations pipefs_dentry_operations = {
688 .d_dname = pipefs_dname,
691 static struct inode * get_pipe_inode(void)
693 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
694 struct pipe_inode_info *pipe;
699 inode->i_ino = get_next_ino();
701 pipe = alloc_pipe_info();
705 inode->i_pipe = pipe;
707 pipe->readers = pipe->writers = 1;
708 inode->i_fop = &pipefifo_fops;
711 * Mark the inode dirty from the very beginning,
712 * that way it will never be moved to the dirty
713 * list because "mark_inode_dirty()" will think
714 * that it already _is_ on the dirty list.
716 inode->i_state = I_DIRTY;
717 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
718 inode->i_uid = current_fsuid();
719 inode->i_gid = current_fsgid();
720 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
731 int create_pipe_files(struct file **res, int flags)
734 struct inode *inode = get_pipe_inode();
737 static struct qstr name = { .name = "" };
743 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
746 path.mnt = mntget(pipe_mnt);
748 d_instantiate(path.dentry, inode);
750 f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
756 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
757 f->private_data = inode->i_pipe;
759 res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
760 if (IS_ERR(res[0])) {
761 err = PTR_ERR(res[0]);
766 res[0]->private_data = inode->i_pipe;
767 res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
774 free_pipe_info(inode->i_pipe);
779 free_pipe_info(inode->i_pipe);
784 static int __do_pipe_flags(int *fd, struct file **files, int flags)
789 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
792 error = create_pipe_files(files, flags);
796 error = get_unused_fd_flags(flags);
801 error = get_unused_fd_flags(flags);
806 audit_fd_pair(fdr, fdw);
819 int do_pipe_flags(int *fd, int flags)
821 struct file *files[2];
822 int error = __do_pipe_flags(fd, files, flags);
824 fd_install(fd[0], files[0]);
825 fd_install(fd[1], files[1]);
831 * sys_pipe() is the normal C calling standard for creating
832 * a pipe. It's not the way Unix traditionally does this, though.
834 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
836 struct file *files[2];
840 error = __do_pipe_flags(fd, files, flags);
842 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
845 put_unused_fd(fd[0]);
846 put_unused_fd(fd[1]);
849 fd_install(fd[0], files[0]);
850 fd_install(fd[1], files[1]);
856 SYSCALL_DEFINE1(pipe, int __user *, fildes)
858 return sys_pipe2(fildes, 0);
861 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
865 while (cur == *cnt) {
867 if (signal_pending(current))
870 return cur == *cnt ? -ERESTARTSYS : 0;
873 static void wake_up_partner(struct pipe_inode_info *pipe)
875 wake_up_interruptible(&pipe->wait);
878 static int fifo_open(struct inode *inode, struct file *filp)
880 struct pipe_inode_info *pipe;
881 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
886 spin_lock(&inode->i_lock);
888 pipe = inode->i_pipe;
890 spin_unlock(&inode->i_lock);
892 spin_unlock(&inode->i_lock);
893 pipe = alloc_pipe_info();
897 spin_lock(&inode->i_lock);
898 if (unlikely(inode->i_pipe)) {
899 inode->i_pipe->files++;
900 spin_unlock(&inode->i_lock);
901 free_pipe_info(pipe);
902 pipe = inode->i_pipe;
904 inode->i_pipe = pipe;
905 spin_unlock(&inode->i_lock);
908 filp->private_data = pipe;
909 /* OK, we have a pipe and it's pinned down */
913 /* We can only do regular read/write on fifos */
914 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
916 switch (filp->f_mode) {
920 * POSIX.1 says that O_NONBLOCK means return with the FIFO
921 * opened, even when there is no process writing the FIFO.
924 if (pipe->readers++ == 0)
925 wake_up_partner(pipe);
927 if (!is_pipe && !pipe->writers) {
928 if ((filp->f_flags & O_NONBLOCK)) {
929 /* suppress POLLHUP until we have
931 filp->f_version = pipe->w_counter;
933 if (wait_for_partner(pipe, &pipe->w_counter))
942 * POSIX.1 says that O_NONBLOCK means return -1 with
943 * errno=ENXIO when there is no process reading the FIFO.
946 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
950 if (!pipe->writers++)
951 wake_up_partner(pipe);
953 if (!is_pipe && !pipe->readers) {
954 if (wait_for_partner(pipe, &pipe->r_counter))
959 case FMODE_READ | FMODE_WRITE:
962 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
963 * This implementation will NEVER block on a O_RDWR open, since
964 * the process can at least talk to itself.
971 if (pipe->readers == 1 || pipe->writers == 1)
972 wake_up_partner(pipe);
985 if (!--pipe->readers)
986 wake_up_interruptible(&pipe->wait);
991 if (!--pipe->writers)
992 wake_up_interruptible(&pipe->wait);
999 put_pipe_info(inode, pipe);
1003 const struct file_operations pipefifo_fops = {
1005 .llseek = no_llseek,
1006 .read_iter = pipe_read,
1007 .write_iter = pipe_write,
1009 .unlocked_ioctl = pipe_ioctl,
1010 .release = pipe_release,
1011 .fasync = pipe_fasync,
1015 * Allocate a new array of pipe buffers and copy the info over. Returns the
1016 * pipe size if successful, or return -ERROR on error.
1018 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
1020 struct pipe_buffer *bufs;
1026 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1027 * expect a lot of shrink+grow operations, just free and allocate
1028 * again like we would do for growing. If the pipe currently
1029 * contains more buffers than arg, then return busy.
1031 if (nr_pages < pipe->nrbufs)
1034 bufs = kcalloc(nr_pages, sizeof(*bufs), GFP_KERNEL | __GFP_NOWARN);
1035 if (unlikely(!bufs))
1039 * The pipe array wraps around, so just start the new one at zero
1040 * and adjust the indexes.
1046 tail = pipe->curbuf + pipe->nrbufs;
1047 if (tail < pipe->buffers)
1050 tail &= (pipe->buffers - 1);
1052 head = pipe->nrbufs - tail;
1054 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1056 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1059 account_pipe_buffers(pipe, pipe->buffers, nr_pages);
1063 pipe->buffers = nr_pages;
1064 return nr_pages * PAGE_SIZE;
1068 * Currently we rely on the pipe array holding a power-of-2 number
1069 * of pages. Returns 0 on error.
1071 static inline unsigned int round_pipe_size(unsigned int size)
1073 unsigned long nr_pages;
1075 if (size < pipe_min_size)
1076 size = pipe_min_size;
1078 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1082 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1086 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1087 * will return an error.
1089 int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1090 size_t *lenp, loff_t *ppos)
1092 unsigned int rounded_pipe_max_size;
1095 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1096 if (ret < 0 || !write)
1099 rounded_pipe_max_size = round_pipe_size(pipe_max_size);
1100 if (rounded_pipe_max_size == 0)
1103 pipe_max_size = rounded_pipe_max_size;
1108 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1109 * location, so checking ->i_pipe is not enough to verify that this is a
1112 struct pipe_inode_info *get_pipe_info(struct file *file)
1114 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1117 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1119 struct pipe_inode_info *pipe;
1122 pipe = get_pipe_info(file);
1129 case F_SETPIPE_SZ: {
1130 unsigned int size, nr_pages;
1132 size = round_pipe_size(arg);
1133 nr_pages = size >> PAGE_SHIFT;
1139 if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
1142 } else if ((too_many_pipe_buffers_hard(pipe->user) ||
1143 too_many_pipe_buffers_soft(pipe->user)) &&
1144 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) {
1148 ret = pipe_set_size(pipe, nr_pages);
1152 ret = pipe->buffers * PAGE_SIZE;
1160 __pipe_unlock(pipe);
1164 static const struct super_operations pipefs_ops = {
1165 .destroy_inode = free_inode_nonrcu,
1166 .statfs = simple_statfs,
1170 * pipefs should _never_ be mounted by userland - too much of security hassle,
1171 * no real gain from having the whole whorehouse mounted. So we don't need
1172 * any operations on the root directory. However, we need a non-trivial
1173 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1175 static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1176 int flags, const char *dev_name, void *data)
1178 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1179 &pipefs_dentry_operations, PIPEFS_MAGIC);
1182 static struct file_system_type pipe_fs_type = {
1184 .mount = pipefs_mount,
1185 .kill_sb = kill_anon_super,
1188 static int __init init_pipe_fs(void)
1190 int err = register_filesystem(&pipe_fs_type);
1193 pipe_mnt = kern_mount(&pipe_fs_type);
1194 if (IS_ERR(pipe_mnt)) {
1195 err = PTR_ERR(pipe_mnt);
1196 unregister_filesystem(&pipe_fs_type);
1202 fs_initcall(init_pipe_fs);