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/aio.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;
43 * We use a start+len construction, which provides full use of the
45 * -- Florian Coosmann (FGC)
47 * Reads with count = 0 should always return 0.
48 * -- Julian Bradfield 1999-06-07.
50 * FIFOs and Pipes now generate SIGIO for both readers and writers.
51 * -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
53 * pipe_read & write cleanup
54 * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
57 static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
60 mutex_lock_nested(&pipe->mutex, subclass);
63 void pipe_lock(struct pipe_inode_info *pipe)
66 * pipe_lock() nests non-pipe inode locks (for writing to a file)
68 pipe_lock_nested(pipe, I_MUTEX_PARENT);
70 EXPORT_SYMBOL(pipe_lock);
72 void pipe_unlock(struct pipe_inode_info *pipe)
75 mutex_unlock(&pipe->mutex);
77 EXPORT_SYMBOL(pipe_unlock);
79 static inline void __pipe_lock(struct pipe_inode_info *pipe)
81 mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
84 static inline void __pipe_unlock(struct pipe_inode_info *pipe)
86 mutex_unlock(&pipe->mutex);
89 void pipe_double_lock(struct pipe_inode_info *pipe1,
90 struct pipe_inode_info *pipe2)
92 BUG_ON(pipe1 == pipe2);
95 pipe_lock_nested(pipe1, I_MUTEX_PARENT);
96 pipe_lock_nested(pipe2, I_MUTEX_CHILD);
98 pipe_lock_nested(pipe2, I_MUTEX_PARENT);
99 pipe_lock_nested(pipe1, I_MUTEX_CHILD);
103 /* Drop the inode semaphore and wait for a pipe event, atomically */
104 void pipe_wait(struct pipe_inode_info *pipe)
109 * Pipes are system-local resources, so sleeping on them
110 * is considered a noninteractive wait:
112 prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
115 finish_wait(&pipe->wait, &wait);
120 pipe_iov_copy_from_user(void *addr, int *offset, struct iovec *iov,
121 size_t *remaining, int atomic)
125 while (*remaining > 0) {
126 while (!iov->iov_len)
128 copy = min_t(unsigned long, *remaining, iov->iov_len);
131 if (__copy_from_user_inatomic(addr + *offset,
132 iov->iov_base, copy))
135 if (copy_from_user(addr + *offset,
136 iov->iov_base, copy))
141 iov->iov_base += copy;
142 iov->iov_len -= copy;
148 pipe_iov_copy_to_user(struct iovec *iov, void *addr, int *offset,
149 size_t *remaining, int atomic)
153 while (*remaining > 0) {
154 while (!iov->iov_len)
156 copy = min_t(unsigned long, *remaining, iov->iov_len);
159 if (__copy_to_user_inatomic(iov->iov_base,
160 addr + *offset, copy))
163 if (copy_to_user(iov->iov_base,
164 addr + *offset, copy))
169 iov->iov_base += copy;
170 iov->iov_len -= copy;
176 * Attempt to pre-fault in the user memory, so we can use atomic copies.
177 * Returns the number of bytes not faulted in.
179 static int iov_fault_in_pages_write(struct iovec *iov, unsigned long len)
181 while (!iov->iov_len)
185 unsigned long this_len;
187 this_len = min_t(unsigned long, len, iov->iov_len);
188 if (fault_in_pages_writeable(iov->iov_base, this_len))
199 * Pre-fault in the user memory, so we can use atomic copies.
201 static void iov_fault_in_pages_read(struct iovec *iov, unsigned long len)
203 while (!iov->iov_len)
207 unsigned long this_len;
209 this_len = min_t(unsigned long, len, iov->iov_len);
210 fault_in_pages_readable(iov->iov_base, this_len);
216 static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
217 struct pipe_buffer *buf)
219 struct page *page = buf->page;
222 * If nobody else uses this page, and we don't already have a
223 * temporary page, let's keep track of it as a one-deep
224 * allocation cache. (Otherwise just release our reference to it)
226 if (page_count(page) == 1 && !pipe->tmp_page)
227 pipe->tmp_page = page;
229 page_cache_release(page);
233 * generic_pipe_buf_map - virtually map a pipe buffer
234 * @pipe: the pipe that the buffer belongs to
235 * @buf: the buffer that should be mapped
236 * @atomic: whether to use an atomic map
239 * This function returns a kernel virtual address mapping for the
240 * pipe_buffer passed in @buf. If @atomic is set, an atomic map is provided
241 * and the caller has to be careful not to fault before calling
242 * the unmap function.
244 * Note that this function calls kmap_atomic() if @atomic != 0.
246 void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
247 struct pipe_buffer *buf, int atomic)
250 buf->flags |= PIPE_BUF_FLAG_ATOMIC;
251 return kmap_atomic(buf->page);
254 return kmap(buf->page);
256 EXPORT_SYMBOL(generic_pipe_buf_map);
259 * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
260 * @pipe: the pipe that the buffer belongs to
261 * @buf: the buffer that should be unmapped
262 * @map_data: the data that the mapping function returned
265 * This function undoes the mapping that ->map() provided.
267 void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
268 struct pipe_buffer *buf, void *map_data)
270 if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
271 buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
272 kunmap_atomic(map_data);
276 EXPORT_SYMBOL(generic_pipe_buf_unmap);
279 * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
280 * @pipe: the pipe that the buffer belongs to
281 * @buf: the buffer to attempt to steal
284 * This function attempts to steal the &struct page attached to
285 * @buf. If successful, this function returns 0 and returns with
286 * the page locked. The caller may then reuse the page for whatever
287 * he wishes; the typical use is insertion into a different file
290 int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
291 struct pipe_buffer *buf)
293 struct page *page = buf->page;
296 * A reference of one is golden, that means that the owner of this
297 * page is the only one holding a reference to it. lock the page
300 if (page_count(page) == 1) {
307 EXPORT_SYMBOL(generic_pipe_buf_steal);
310 * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
311 * @pipe: the pipe that the buffer belongs to
312 * @buf: the buffer to get a reference to
315 * This function grabs an extra reference to @buf. It's used in
316 * in the tee() system call, when we duplicate the buffers in one
319 void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
321 page_cache_get(buf->page);
323 EXPORT_SYMBOL(generic_pipe_buf_get);
326 * generic_pipe_buf_confirm - verify contents of the pipe buffer
327 * @info: the pipe that the buffer belongs to
328 * @buf: the buffer to confirm
331 * This function does nothing, because the generic pipe code uses
332 * pages that are always good when inserted into the pipe.
334 int generic_pipe_buf_confirm(struct pipe_inode_info *info,
335 struct pipe_buffer *buf)
339 EXPORT_SYMBOL(generic_pipe_buf_confirm);
342 * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
343 * @pipe: the pipe that the buffer belongs to
344 * @buf: the buffer to put a reference to
347 * This function releases a reference to @buf.
349 void generic_pipe_buf_release(struct pipe_inode_info *pipe,
350 struct pipe_buffer *buf)
352 page_cache_release(buf->page);
354 EXPORT_SYMBOL(generic_pipe_buf_release);
356 static const struct pipe_buf_operations anon_pipe_buf_ops = {
358 .map = generic_pipe_buf_map,
359 .unmap = generic_pipe_buf_unmap,
360 .confirm = generic_pipe_buf_confirm,
361 .release = anon_pipe_buf_release,
362 .steal = generic_pipe_buf_steal,
363 .get = generic_pipe_buf_get,
366 static const struct pipe_buf_operations packet_pipe_buf_ops = {
368 .map = generic_pipe_buf_map,
369 .unmap = generic_pipe_buf_unmap,
370 .confirm = generic_pipe_buf_confirm,
371 .release = anon_pipe_buf_release,
372 .steal = generic_pipe_buf_steal,
373 .get = generic_pipe_buf_get,
377 pipe_read(struct kiocb *iocb, const struct iovec *_iov,
378 unsigned long nr_segs, loff_t pos)
380 struct file *filp = iocb->ki_filp;
381 struct pipe_inode_info *pipe = filp->private_data;
384 struct iovec *iov = (struct iovec *)_iov;
387 total_len = iov_length(iov, nr_segs);
388 /* Null read succeeds. */
389 if (unlikely(total_len == 0))
396 int bufs = pipe->nrbufs;
398 int curbuf = pipe->curbuf;
399 struct pipe_buffer *buf = pipe->bufs + curbuf;
400 const struct pipe_buf_operations *ops = buf->ops;
402 size_t chars = buf->len, remaining;
405 if (chars > total_len)
408 error = ops->confirm(pipe, buf);
415 atomic = !iov_fault_in_pages_write(iov, chars);
418 addr = ops->map(pipe, buf, atomic);
419 error = pipe_iov_copy_to_user(iov, addr, &buf->offset,
421 ops->unmap(pipe, buf, addr);
422 if (unlikely(error)) {
424 * Just retry with the slow path if we failed.
437 /* Was it a packet buffer? Clean up and exit */
438 if (buf->flags & PIPE_BUF_FLAG_PACKET) {
445 ops->release(pipe, buf);
446 curbuf = (curbuf + 1) & (pipe->buffers - 1);
447 pipe->curbuf = curbuf;
448 pipe->nrbufs = --bufs;
453 break; /* common path: read succeeded */
455 if (bufs) /* More to do? */
459 if (!pipe->waiting_writers) {
460 /* syscall merging: Usually we must not sleep
461 * if O_NONBLOCK is set, or if we got some data.
462 * But if a writer sleeps in kernel space, then
463 * we can wait for that data without violating POSIX.
467 if (filp->f_flags & O_NONBLOCK) {
472 if (signal_pending(current)) {
478 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
479 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
485 /* Signal writers asynchronously that there is more room. */
487 wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
488 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
495 static inline int is_packetized(struct file *file)
497 return (file->f_flags & O_DIRECT) != 0;
501 pipe_write(struct kiocb *iocb, const struct iovec *_iov,
502 unsigned long nr_segs, loff_t ppos)
504 struct file *filp = iocb->ki_filp;
505 struct pipe_inode_info *pipe = filp->private_data;
508 struct iovec *iov = (struct iovec *)_iov;
512 total_len = iov_length(iov, nr_segs);
513 /* Null write succeeds. */
514 if (unlikely(total_len == 0))
521 if (!pipe->readers) {
522 send_sig(SIGPIPE, current, 0);
527 /* We try to merge small writes */
528 chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
529 if (pipe->nrbufs && chars != 0) {
530 int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
532 struct pipe_buffer *buf = pipe->bufs + lastbuf;
533 const struct pipe_buf_operations *ops = buf->ops;
534 int offset = buf->offset + buf->len;
536 if (ops->can_merge && offset + chars <= PAGE_SIZE) {
537 int error, atomic = 1;
539 size_t remaining = chars;
541 error = ops->confirm(pipe, buf);
545 iov_fault_in_pages_read(iov, chars);
547 addr = ops->map(pipe, buf, atomic);
548 error = pipe_iov_copy_from_user(addr, &offset, iov,
550 ops->unmap(pipe, buf, addr);
571 if (!pipe->readers) {
572 send_sig(SIGPIPE, current, 0);
578 if (bufs < pipe->buffers) {
579 int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
580 struct pipe_buffer *buf = pipe->bufs + newbuf;
581 struct page *page = pipe->tmp_page;
583 int error, atomic = 1;
588 #ifndef CONFIG_SPRD_PAGERECORDER
589 page = alloc_page(GFP_HIGHUSER);
591 page = alloc_page_nopagedebug(GFP_HIGHUSER);
593 if (unlikely(!page)) {
594 ret = ret ? : -ENOMEM;
597 pipe->tmp_page = page;
599 /* Always wake up, even if the copy fails. Otherwise
600 * we lock up (O_NONBLOCK-)readers that sleep due to
602 * FIXME! Is this really true?
606 if (chars > total_len)
609 iov_fault_in_pages_read(iov, chars);
613 src = kmap_atomic(page);
617 error = pipe_iov_copy_from_user(src, &offset, iov,
624 if (unlikely(error)) {
635 /* Insert it into the buffer array */
637 buf->ops = &anon_pipe_buf_ops;
641 if (is_packetized(filp)) {
642 buf->ops = &packet_pipe_buf_ops;
643 buf->flags = PIPE_BUF_FLAG_PACKET;
645 pipe->nrbufs = ++bufs;
646 pipe->tmp_page = NULL;
652 if (bufs < pipe->buffers)
654 if (filp->f_flags & O_NONBLOCK) {
659 if (signal_pending(current)) {
665 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
666 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
669 pipe->waiting_writers++;
671 pipe->waiting_writers--;
676 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
677 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
680 int err = file_update_time(filp);
687 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
689 struct pipe_inode_info *pipe = filp->private_data;
690 int count, buf, nrbufs;
697 nrbufs = pipe->nrbufs;
698 while (--nrbufs >= 0) {
699 count += pipe->bufs[buf].len;
700 buf = (buf+1) & (pipe->buffers - 1);
704 return put_user(count, (int __user *)arg);
710 /* No kernel lock held - fine */
712 pipe_poll(struct file *filp, poll_table *wait)
715 struct pipe_inode_info *pipe = filp->private_data;
718 poll_wait(filp, &pipe->wait, wait);
720 /* Reading only -- no need for acquiring the semaphore. */
721 nrbufs = pipe->nrbufs;
723 if (filp->f_mode & FMODE_READ) {
724 mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
725 if (!pipe->writers && filp->f_version != pipe->w_counter)
729 if (filp->f_mode & FMODE_WRITE) {
730 mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
732 * Most Unices do not set POLLERR for FIFOs but on Linux they
733 * behave exactly like pipes for poll().
742 static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
746 spin_lock(&inode->i_lock);
747 if (!--pipe->files) {
748 inode->i_pipe = NULL;
751 spin_unlock(&inode->i_lock);
754 free_pipe_info(pipe);
758 pipe_release(struct inode *inode, struct file *file)
760 struct pipe_inode_info *pipe = file->private_data;
763 if (file->f_mode & FMODE_READ)
765 if (file->f_mode & FMODE_WRITE)
768 if (pipe->readers || pipe->writers) {
769 wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
770 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
771 kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
775 put_pipe_info(inode, pipe);
780 pipe_fasync(int fd, struct file *filp, int on)
782 struct pipe_inode_info *pipe = filp->private_data;
786 if (filp->f_mode & FMODE_READ)
787 retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
788 if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
789 retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
790 if (retval < 0 && (filp->f_mode & FMODE_READ))
791 /* this can happen only if on == T */
792 fasync_helper(-1, filp, 0, &pipe->fasync_readers);
798 struct pipe_inode_info *alloc_pipe_info(void)
800 struct pipe_inode_info *pipe;
802 pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
804 pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
806 init_waitqueue_head(&pipe->wait);
807 pipe->r_counter = pipe->w_counter = 1;
808 pipe->buffers = PIPE_DEF_BUFFERS;
809 mutex_init(&pipe->mutex);
818 void free_pipe_info(struct pipe_inode_info *pipe)
822 for (i = 0; i < pipe->buffers; i++) {
823 struct pipe_buffer *buf = pipe->bufs + i;
825 buf->ops->release(pipe, buf);
828 __free_page(pipe->tmp_page);
833 static struct vfsmount *pipe_mnt __read_mostly;
836 * pipefs_dname() is called from d_path().
838 static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
840 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
841 dentry->d_inode->i_ino);
844 static const struct dentry_operations pipefs_dentry_operations = {
845 .d_dname = pipefs_dname,
848 static struct inode * get_pipe_inode(void)
850 struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
851 struct pipe_inode_info *pipe;
856 inode->i_ino = get_next_ino();
858 pipe = alloc_pipe_info();
862 inode->i_pipe = pipe;
864 pipe->readers = pipe->writers = 1;
865 inode->i_fop = &pipefifo_fops;
868 * Mark the inode dirty from the very beginning,
869 * that way it will never be moved to the dirty
870 * list because "mark_inode_dirty()" will think
871 * that it already _is_ on the dirty list.
873 inode->i_state = I_DIRTY;
874 inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
875 inode->i_uid = current_fsuid();
876 inode->i_gid = current_fsgid();
877 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
888 int create_pipe_files(struct file **res, int flags)
891 struct inode *inode = get_pipe_inode();
894 static struct qstr name = { .name = "" };
900 path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
903 path.mnt = mntget(pipe_mnt);
905 d_instantiate(path.dentry, inode);
908 f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
912 f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
913 f->private_data = inode->i_pipe;
915 res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
920 res[0]->private_data = inode->i_pipe;
921 res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
928 free_pipe_info(inode->i_pipe);
933 free_pipe_info(inode->i_pipe);
938 static int __do_pipe_flags(int *fd, struct file **files, int flags)
943 if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
946 error = create_pipe_files(files, flags);
950 error = get_unused_fd_flags(flags);
955 error = get_unused_fd_flags(flags);
960 audit_fd_pair(fdr, fdw);
973 int do_pipe_flags(int *fd, int flags)
975 struct file *files[2];
976 int error = __do_pipe_flags(fd, files, flags);
978 fd_install(fd[0], files[0]);
979 fd_install(fd[1], files[1]);
985 * sys_pipe() is the normal C calling standard for creating
986 * a pipe. It's not the way Unix traditionally does this, though.
988 SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
990 struct file *files[2];
994 error = __do_pipe_flags(fd, files, flags);
996 if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
999 put_unused_fd(fd[0]);
1000 put_unused_fd(fd[1]);
1003 fd_install(fd[0], files[0]);
1004 fd_install(fd[1], files[1]);
1010 SYSCALL_DEFINE1(pipe, int __user *, fildes)
1012 return sys_pipe2(fildes, 0);
1015 static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
1019 while (cur == *cnt) {
1021 if (signal_pending(current))
1024 return cur == *cnt ? -ERESTARTSYS : 0;
1027 static void wake_up_partner(struct pipe_inode_info *pipe)
1029 wake_up_interruptible(&pipe->wait);
1032 static int fifo_open(struct inode *inode, struct file *filp)
1034 struct pipe_inode_info *pipe;
1035 bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
1038 filp->f_version = 0;
1040 spin_lock(&inode->i_lock);
1041 if (inode->i_pipe) {
1042 pipe = inode->i_pipe;
1044 spin_unlock(&inode->i_lock);
1046 spin_unlock(&inode->i_lock);
1047 pipe = alloc_pipe_info();
1051 spin_lock(&inode->i_lock);
1052 if (unlikely(inode->i_pipe)) {
1053 inode->i_pipe->files++;
1054 spin_unlock(&inode->i_lock);
1055 free_pipe_info(pipe);
1056 pipe = inode->i_pipe;
1058 inode->i_pipe = pipe;
1059 spin_unlock(&inode->i_lock);
1062 filp->private_data = pipe;
1063 /* OK, we have a pipe and it's pinned down */
1067 /* We can only do regular read/write on fifos */
1068 filp->f_mode &= (FMODE_READ | FMODE_WRITE);
1070 switch (filp->f_mode) {
1074 * POSIX.1 says that O_NONBLOCK means return with the FIFO
1075 * opened, even when there is no process writing the FIFO.
1078 if (pipe->readers++ == 0)
1079 wake_up_partner(pipe);
1081 if (!is_pipe && !pipe->writers) {
1082 if ((filp->f_flags & O_NONBLOCK)) {
1083 /* suppress POLLHUP until we have
1085 filp->f_version = pipe->w_counter;
1087 if (wait_for_partner(pipe, &pipe->w_counter))
1096 * POSIX.1 says that O_NONBLOCK means return -1 with
1097 * errno=ENXIO when there is no process reading the FIFO.
1100 if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
1104 if (!pipe->writers++)
1105 wake_up_partner(pipe);
1107 if (!is_pipe && !pipe->readers) {
1108 if (wait_for_partner(pipe, &pipe->r_counter))
1113 case FMODE_READ | FMODE_WRITE:
1116 * POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
1117 * This implementation will NEVER block on a O_RDWR open, since
1118 * the process can at least talk to itself.
1125 if (pipe->readers == 1 || pipe->writers == 1)
1126 wake_up_partner(pipe);
1135 __pipe_unlock(pipe);
1139 if (!--pipe->readers)
1140 wake_up_interruptible(&pipe->wait);
1145 if (!--pipe->writers)
1146 wake_up_interruptible(&pipe->wait);
1151 __pipe_unlock(pipe);
1153 put_pipe_info(inode, pipe);
1157 const struct file_operations pipefifo_fops = {
1159 .llseek = no_llseek,
1160 .read = do_sync_read,
1161 .aio_read = pipe_read,
1162 .write = do_sync_write,
1163 .aio_write = pipe_write,
1165 .unlocked_ioctl = pipe_ioctl,
1166 .release = pipe_release,
1167 .fasync = pipe_fasync,
1171 * Allocate a new array of pipe buffers and copy the info over. Returns the
1172 * pipe size if successful, or return -ERROR on error.
1174 static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
1176 struct pipe_buffer *bufs;
1179 * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
1180 * expect a lot of shrink+grow operations, just free and allocate
1181 * again like we would do for growing. If the pipe currently
1182 * contains more buffers than arg, then return busy.
1184 if (nr_pages < pipe->nrbufs)
1187 bufs = kcalloc(nr_pages, sizeof(*bufs), GFP_KERNEL | __GFP_NOWARN);
1188 if (unlikely(!bufs))
1192 * The pipe array wraps around, so just start the new one at zero
1193 * and adjust the indexes.
1199 tail = pipe->curbuf + pipe->nrbufs;
1200 if (tail < pipe->buffers)
1203 tail &= (pipe->buffers - 1);
1205 head = pipe->nrbufs - tail;
1207 memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
1209 memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
1215 pipe->buffers = nr_pages;
1216 return nr_pages * PAGE_SIZE;
1220 * Currently we rely on the pipe array holding a power-of-2 number
1223 static inline unsigned int round_pipe_size(unsigned int size)
1225 unsigned long nr_pages;
1227 nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
1228 return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
1232 * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
1233 * will return an error.
1235 int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
1236 size_t *lenp, loff_t *ppos)
1240 ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
1241 if (ret < 0 || !write)
1244 pipe_max_size = round_pipe_size(pipe_max_size);
1249 * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
1250 * location, so checking ->i_pipe is not enough to verify that this is a
1253 struct pipe_inode_info *get_pipe_info(struct file *file)
1255 return file->f_op == &pipefifo_fops ? file->private_data : NULL;
1258 long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
1260 struct pipe_inode_info *pipe;
1263 pipe = get_pipe_info(file);
1270 case F_SETPIPE_SZ: {
1271 unsigned int size, nr_pages;
1273 size = round_pipe_size(arg);
1274 nr_pages = size >> PAGE_SHIFT;
1280 if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
1284 ret = pipe_set_size(pipe, nr_pages);
1288 ret = pipe->buffers * PAGE_SIZE;
1296 __pipe_unlock(pipe);
1300 static const struct super_operations pipefs_ops = {
1301 .destroy_inode = free_inode_nonrcu,
1302 .statfs = simple_statfs,
1306 * pipefs should _never_ be mounted by userland - too much of security hassle,
1307 * no real gain from having the whole whorehouse mounted. So we don't need
1308 * any operations on the root directory. However, we need a non-trivial
1309 * d_name - pipe: will go nicely and kill the special-casing in procfs.
1311 static struct dentry *pipefs_mount(struct file_system_type *fs_type,
1312 int flags, const char *dev_name, void *data)
1314 return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
1315 &pipefs_dentry_operations, PIPEFS_MAGIC);
1318 static struct file_system_type pipe_fs_type = {
1320 .mount = pipefs_mount,
1321 .kill_sb = kill_anon_super,
1324 static int __init init_pipe_fs(void)
1326 int err = register_filesystem(&pipe_fs_type);
1329 pipe_mnt = kern_mount(&pipe_fs_type);
1330 if (IS_ERR(pipe_mnt)) {
1331 err = PTR_ERR(pipe_mnt);
1332 unregister_filesystem(&pipe_fs_type);
1338 fs_initcall(init_pipe_fs);