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 <asm/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 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)
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 anon_pipe_buf_nomerge_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,
249 static const struct pipe_buf_operations packet_pipe_buf_ops = {
251 .confirm = generic_pipe_buf_confirm,
252 .release = anon_pipe_buf_release,
253 .steal = anon_pipe_buf_steal,
254 .get = generic_pipe_buf_get,
257 void pipe_buf_mark_unmergeable(struct pipe_buffer *buf)
259 if (buf->ops == &anon_pipe_buf_ops)
260 buf->ops = &anon_pipe_buf_nomerge_ops;
264 pipe_read(struct kiocb *iocb, struct iov_iter *to)
266 size_t total_len = iov_iter_count(to);
267 struct file *filp = iocb->ki_filp;
268 struct pipe_inode_info *pipe = filp->private_data;
272 /* Null read succeeds. */
273 if (unlikely(total_len == 0))
280 int bufs = pipe->nrbufs;
282 int curbuf = pipe->curbuf;
283 struct pipe_buffer *buf = pipe->bufs + curbuf;
284 size_t chars = buf->len;
288 if (chars > total_len)
291 error = pipe_buf_confirm(pipe, buf);
298 written = copy_page_to_iter(buf->page, buf->offset, chars, to);
299 if (unlikely(written < chars)) {
305 buf->offset += chars;
308 /* Was it a packet buffer? Clean up and exit */
309 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
315 pipe_buf_release(pipe, buf);
316 curbuf = (curbuf + 1) & (pipe->buffers - 1);
317 pipe->curbuf = curbuf;
318 pipe->nrbufs = --bufs;
323 break; /* common path: read succeeded */
325 if (bufs) /* More to do? */
329 if (!pipe->waiting_writers) {
330 /* syscall merging: Usually we must not sleep
331 * if O_NONBLOCK is set, or if we got some data.
332 * But if a writer sleeps in kernel space, then
333 * we can wait for that data without violating POSIX.
337 if (filp->f_flags & O_NONBLOCK) {
342 if (signal_pending(current)) {
348 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
349 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
355 /* Signal writers asynchronously that there is more room. */
357 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
358 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
365 static inline int is_packetized(struct file *file)
367 return (file->f_flags & O_DIRECT) != 0;
371 pipe_write(struct kiocb *iocb, struct iov_iter *from)
373 struct file *filp = iocb->ki_filp;
374 struct pipe_inode_info *pipe = filp->private_data;
377 size_t total_len = iov_iter_count(from);
380 /* Null write succeeds. */
381 if (unlikely(total_len == 0))
386 if (!pipe->readers) {
387 send_sig(SIGPIPE, current, 0);
392 /* We try to merge small writes */
393 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
394 if (pipe->nrbufs && chars != 0) {
395 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
397 struct pipe_buffer *buf = pipe->bufs + lastbuf;
398 int offset = buf->offset + buf->len;
400 if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
401 ret = pipe_buf_confirm(pipe, buf);
405 ret = copy_page_from_iter(buf->page, offset, chars, from);
406 if (unlikely(ret < chars)) {
412 if (!iov_iter_count(from))
420 if (!pipe->readers) {
421 send_sig(SIGPIPE, current, 0);
427 if (bufs < pipe->buffers) {
428 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
429 struct pipe_buffer *buf = pipe->bufs + newbuf;
430 struct page *page = pipe->tmp_page;
434 page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
435 if (unlikely(!page)) {
436 ret = ret ? : -ENOMEM;
439 pipe->tmp_page = page;
441 /* Always wake up, even if the copy fails. Otherwise
442 * we lock up (O_NONBLOCK-)readers that sleep due to
444 * FIXME! Is this really true?
447 copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
448 if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
455 /* Insert it into the buffer array */
457 buf->ops = &anon_pipe_buf_ops;
461 if (is_packetized(filp)) {
462 buf->ops = &packet_pipe_buf_ops;
463 buf->flags = PIPE_BUF_FLAG_PACKET;
465 pipe->nrbufs = ++bufs;
466 pipe->tmp_page = NULL;
468 if (!iov_iter_count(from))
471 if (bufs < pipe->buffers)
473 if (filp->f_flags & O_NONBLOCK) {
478 if (signal_pending(current)) {
484 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
485 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
488 pipe->waiting_writers++;
490 pipe->waiting_writers--;
495 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
496 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
498 if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
499 int err = file_update_time(filp);
502 sb_end_write(file_inode(filp)->i_sb);
507 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
509 struct pipe_inode_info *pipe = filp->private_data;
510 int count, buf, nrbufs;
517 nrbufs = pipe->nrbufs;
518 while (--nrbufs >= 0) {
519 count += pipe->bufs[buf].len;
520 buf = (buf+1) & (pipe->buffers - 1);
524 return put_user(count, (int __user *)arg);
530 /* No kernel lock held - fine */
532 pipe_poll(struct file *filp, poll_table *wait)
535 struct pipe_inode_info *pipe = filp->private_data;
538 poll_wait(filp, &pipe->wait, wait);
540 /* Reading only -- no need for acquiring the semaphore. */
541 nrbufs = pipe->nrbufs;
543 if (filp->f_mode & FMODE_READ) {
544 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
545 if (!pipe->writers && filp->f_version != pipe->w_counter)
549 if (filp->f_mode & FMODE_WRITE) {
550 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
552 * Most Unices do not set POLLERR for FIFOs but on Linux they
553 * behave exactly like pipes for poll().
562 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
566 spin_lock(&inode->i_lock);
567 if (!--pipe->files) {
568 inode->i_pipe = NULL;
571 spin_unlock(&inode->i_lock);
574 free_pipe_info(pipe);
578 pipe_release(struct inode *inode, struct file *file)
580 struct pipe_inode_info *pipe = file->private_data;
583 if (file->f_mode & FMODE_READ)
585 if (file->f_mode & FMODE_WRITE)
588 if (pipe->readers || pipe->writers) {
589 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
590 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
591 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
595 put_pipe_info(inode, pipe);
600 pipe_fasync(int fd, struct file *filp, int on)
602 struct pipe_inode_info *pipe = filp->private_data;
606 if (filp->f_mode & FMODE_READ)
607 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
608 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
609 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
610 if (retval < 0 && (filp->f_mode & FMODE_READ))
611 /* this can happen only if on == T */
612 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
618 static unsigned long account_pipe_buffers(struct user_struct *user,
619 unsigned long old, unsigned long new)
621 return atomic_long_add_return(new - old, &user->pipe_bufs);
624 static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
626 return pipe_user_pages_soft && user_bufs > pipe_user_pages_soft;
629 static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
631 return pipe_user_pages_hard && user_bufs > pipe_user_pages_hard;
634 static bool is_unprivileged_user(void)
636 return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
639 struct pipe_inode_info *alloc_pipe_info(void)
641 struct pipe_inode_info *pipe;
642 unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
643 struct user_struct *user = get_current_user();
644 unsigned long user_bufs;
646 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
650 if (pipe_bufs * PAGE_SIZE > pipe_max_size && !capable(CAP_SYS_RESOURCE))
651 pipe_bufs = pipe_max_size >> PAGE_SHIFT;
653 user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
655 if (too_many_pipe_buffers_soft(user_bufs) && is_unprivileged_user()) {
656 user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
660 if (too_many_pipe_buffers_hard(user_bufs) && is_unprivileged_user())
661 goto out_revert_acct;
663 pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
667 init_waitqueue_head(&pipe->wait);
668 pipe->r_counter = pipe->w_counter = 1;
669 pipe->buffers = pipe_bufs;
671 mutex_init(&pipe->mutex);
676 (void) account_pipe_buffers(user, pipe_bufs, 0);
683 void free_pipe_info(struct pipe_inode_info *pipe)
687 (void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
688 free_uid(pipe->user);
689 for (i = 0; i < pipe->buffers; i++) {
690 struct pipe_buffer *buf = pipe->bufs + i;
692 pipe_buf_release(pipe, buf);
695 __free_page(pipe->tmp_page);
700 static struct vfsmount *pipe_mnt __read_mostly;
703 * pipefs_dname() is called from d_path().
705 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
707 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
708 d_inode(dentry)->i_ino);
711 static const struct dentry_operations pipefs_dentry_operations = {
712 .d_dname = pipefs_dname,
715 static struct inode * get_pipe_inode(void)
717 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
718 struct pipe_inode_info *pipe;
723 inode->i_ino = get_next_ino();
725 pipe = alloc_pipe_info();
729 inode->i_pipe = pipe;
731 pipe->readers = pipe->writers = 1;
732 inode->i_fop = &pipefifo_fops;
735 * Mark the inode dirty from the very beginning,
736 * that way it will never be moved to the dirty
737 * list because "mark_inode_dirty()" will think
738 * that it already _is_ on the dirty list.
740 inode->i_state = I_DIRTY;
741 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
742 inode->i_uid = current_fsuid();
743 inode->i_gid = current_fsgid();
744 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
755 int create_pipe_files(struct file **res, int flags)
758 struct inode *inode = get_pipe_inode();
761 static struct qstr name = { .name = "" };
767 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
770 path.mnt = mntget(pipe_mnt);
772 d_instantiate(path.dentry, inode);
774 f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
780 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
781 f->private_data = inode->i_pipe;
783 res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
784 if (IS_ERR(res[0])) {
785 err = PTR_ERR(res[0]);
790 res[0]->private_data = inode->i_pipe;
791 res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
798 free_pipe_info(inode->i_pipe);
803 free_pipe_info(inode->i_pipe);
808 static int __do_pipe_flags(int *fd, struct file **files, int flags)
813 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
816 error = create_pipe_files(files, flags);
820 error = get_unused_fd_flags(flags);
825 error = get_unused_fd_flags(flags);
830 audit_fd_pair(fdr, fdw);
843 int do_pipe_flags(int *fd, int flags)
845 struct file *files[2];
846 int error = __do_pipe_flags(fd, files, flags);
848 fd_install(fd[0], files[0]);
849 fd_install(fd[1], files[1]);
855 * sys_pipe() is the normal C calling standard for creating
856 * a pipe. It's not the way Unix traditionally does this, though.
858 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
860 struct file *files[2];
864 error = __do_pipe_flags(fd, files, flags);
866 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
869 put_unused_fd(fd[0]);
870 put_unused_fd(fd[1]);
873 fd_install(fd[0], files[0]);
874 fd_install(fd[1], files[1]);
880 SYSCALL_DEFINE1(pipe, int __user *, fildes)
882 return sys_pipe2(fildes, 0);
885 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
889 while (cur == *cnt) {
891 if (signal_pending(current))
894 return cur == *cnt ? -ERESTARTSYS : 0;
897 static void wake_up_partner(struct pipe_inode_info *pipe)
899 wake_up_interruptible(&pipe->wait);
902 static int fifo_open(struct inode *inode, struct file *filp)
904 struct pipe_inode_info *pipe;
905 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
910 spin_lock(&inode->i_lock);
912 pipe = inode->i_pipe;
914 spin_unlock(&inode->i_lock);
916 spin_unlock(&inode->i_lock);
917 pipe = alloc_pipe_info();
921 spin_lock(&inode->i_lock);
922 if (unlikely(inode->i_pipe)) {
923 inode->i_pipe->files++;
924 spin_unlock(&inode->i_lock);
925 free_pipe_info(pipe);
926 pipe = inode->i_pipe;
928 inode->i_pipe = pipe;
929 spin_unlock(&inode->i_lock);
932 filp->private_data = pipe;
933 /* OK, we have a pipe and it's pinned down */
937 /* We can only do regular read/write on fifos */
938 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
940 switch (filp->f_mode) {
944 * POSIX.1 says that O_NONBLOCK means return with the FIFO
945 * opened, even when there is no process writing the FIFO.
948 if (pipe->readers++ == 0)
949 wake_up_partner(pipe);
951 if (!is_pipe && !pipe->writers) {
952 if ((filp->f_flags & O_NONBLOCK)) {
953 /* suppress POLLHUP until we have
955 filp->f_version = pipe->w_counter;
957 if (wait_for_partner(pipe, &pipe->w_counter))
966 * POSIX.1 says that O_NONBLOCK means return -1 with
967 * errno=ENXIO when there is no process reading the FIFO.
970 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
974 if (!pipe->writers++)
975 wake_up_partner(pipe);
977 if (!is_pipe && !pipe->readers) {
978 if (wait_for_partner(pipe, &pipe->r_counter))
983 case FMODE_READ | FMODE_WRITE:
986 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
987 * This implementation will NEVER block on a O_RDWR open, since
988 * the process can at least talk to itself.
995 if (pipe->readers == 1 || pipe->writers == 1)
996 wake_up_partner(pipe);
1005 __pipe_unlock(pipe);
1009 if (!--pipe->readers)
1010 wake_up_interruptible(&pipe->wait);
1015 if (!--pipe->writers)
1016 wake_up_interruptible(&pipe->wait);
1021 __pipe_unlock(pipe);
1023 put_pipe_info(inode, pipe);
1027 const struct file_operations pipefifo_fops = {
1029 .llseek = no_llseek,
1030 .read_iter = pipe_read,
1031 .write_iter = pipe_write,
1033 .unlocked_ioctl = pipe_ioctl,
1034 .release = pipe_release,
1035 .fasync = pipe_fasync,
1039 * Currently we rely on the pipe array holding a power-of-2 number
1040 * of pages. Returns 0 on error.
1042 static inline unsigned int round_pipe_size(unsigned int size)
1044 unsigned long nr_pages;
1046 if (size < pipe_min_size)
1047 size = pipe_min_size;
1049 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1053 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1057 * Allocate a new array of pipe buffers and copy the info over. Returns the
1058 * pipe size if successful, or return -ERROR on error.
1060 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
1062 struct pipe_buffer *bufs;
1063 unsigned int size, nr_pages;
1064 unsigned long user_bufs;
1067 size = round_pipe_size(arg);
1070 nr_pages = size >> PAGE_SHIFT;
1076 * If trying to increase the pipe capacity, check that an
1077 * unprivileged user is not trying to exceed various limits
1078 * (soft limit check here, hard limit check just below).
1079 * Decreasing the pipe capacity is always permitted, even
1080 * if the user is currently over a limit.
1082 if (nr_pages > pipe->buffers &&
1083 size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
1086 user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
1088 if (nr_pages > pipe->buffers &&
1089 (too_many_pipe_buffers_hard(user_bufs) ||
1090 too_many_pipe_buffers_soft(user_bufs)) &&
1091 is_unprivileged_user()) {
1093 goto out_revert_acct;
1097 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1098 * expect a lot of shrink+grow operations, just free and allocate
1099 * again like we would do for growing. If the pipe currently
1100 * contains more buffers than arg, then return busy.
1102 if (nr_pages < pipe->nrbufs) {
1104 goto out_revert_acct;
1107 bufs = kcalloc(nr_pages, sizeof(*bufs),
1108 GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
1109 if (unlikely(!bufs)) {
1111 goto out_revert_acct;
1115 * The pipe array wraps around, so just start the new one at zero
1116 * and adjust the indexes.
1122 tail = pipe->curbuf + pipe->nrbufs;
1123 if (tail < pipe->buffers)
1126 tail &= (pipe->buffers - 1);
1128 head = pipe->nrbufs - tail;
1130 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1132 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1138 pipe->buffers = nr_pages;
1139 return nr_pages * PAGE_SIZE;
1142 (void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
1147 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1148 * will return an error.
1150 int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1151 size_t *lenp, loff_t *ppos)
1153 unsigned int rounded_pipe_max_size;
1156 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1157 if (ret < 0 || !write)
1160 rounded_pipe_max_size = round_pipe_size(pipe_max_size);
1161 if (rounded_pipe_max_size == 0)
1164 pipe_max_size = rounded_pipe_max_size;
1169 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1170 * location, so checking ->i_pipe is not enough to verify that this is a
1173 struct pipe_inode_info *get_pipe_info(struct file *file)
1175 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1178 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1180 struct pipe_inode_info *pipe;
1183 pipe = get_pipe_info(file);
1191 ret = pipe_set_size(pipe, arg);
1194 ret = pipe->buffers * PAGE_SIZE;
1201 __pipe_unlock(pipe);
1205 static const struct super_operations pipefs_ops = {
1206 .destroy_inode = free_inode_nonrcu,
1207 .statfs = simple_statfs,
1211 * pipefs should _never_ be mounted by userland - too much of security hassle,
1212 * no real gain from having the whole whorehouse mounted. So we don't need
1213 * any operations on the root directory. However, we need a non-trivial
1214 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1216 static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1217 int flags, const char *dev_name, void *data)
1219 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1220 &pipefs_dentry_operations, PIPEFS_MAGIC);
1223 static struct file_system_type pipe_fs_type = {
1225 .mount = pipefs_mount,
1226 .kill_sb = kill_anon_super,
1229 static int __init init_pipe_fs(void)
1231 int err = register_filesystem(&pipe_fs_type);
1234 pipe_mnt = kern_mount(&pipe_fs_type);
1235 if (IS_ERR(pipe_mnt)) {
1236 err = PTR_ERR(pipe_mnt);
1237 unregister_filesystem(&pipe_fs_type);
1243 fs_initcall(init_pipe_fs);