Revert "input: touch: raydium: V61.7 code drop"
[linux-3.10.git] / fs / pipe.c
index 68e8206..0e0752e 100644 (file)
--- a/fs/pipe.c
+++ b/fs/pipe.c
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/fs.h>
+#include <linux/log2.h>
 #include <linux/mount.h>
+#include <linux/magic.h>
 #include <linux/pipe_fs_i.h>
 #include <linux/uio.h>
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
 #include <linux/audit.h>
 #include <linux/syscalls.h>
+#include <linux/fcntl.h>
+#include <linux/aio.h>
 
 #include <asm/uaccess.h>
 #include <asm/ioctls.h>
 
+#include "internal.h"
+
+/*
+ * The max size that a non-root user is allowed to grow the pipe. Can
+ * be set by root in /proc/sys/fs/pipe-max-size
+ */
+unsigned int pipe_max_size = 1048576;
+
+/*
+ * Minimum pipe size, as required by POSIX
+ */
+unsigned int pipe_min_size = PAGE_SIZE;
+
 /*
  * We use a start+len construction, which provides full use of the 
  * allocated memory.
  * -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
  */
 
+static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
+{
+       if (pipe->files)
+               mutex_lock_nested(&pipe->mutex, subclass);
+}
+
+void pipe_lock(struct pipe_inode_info *pipe)
+{
+       /*
+        * pipe_lock() nests non-pipe inode locks (for writing to a file)
+        */
+       pipe_lock_nested(pipe, I_MUTEX_PARENT);
+}
+EXPORT_SYMBOL(pipe_lock);
+
+void pipe_unlock(struct pipe_inode_info *pipe)
+{
+       if (pipe->files)
+               mutex_unlock(&pipe->mutex);
+}
+EXPORT_SYMBOL(pipe_unlock);
+
+static inline void __pipe_lock(struct pipe_inode_info *pipe)
+{
+       mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
+}
+
+static inline void __pipe_unlock(struct pipe_inode_info *pipe)
+{
+       mutex_unlock(&pipe->mutex);
+}
+
+void pipe_double_lock(struct pipe_inode_info *pipe1,
+                     struct pipe_inode_info *pipe2)
+{
+       BUG_ON(pipe1 == pipe2);
+
+       if (pipe1 < pipe2) {
+               pipe_lock_nested(pipe1, I_MUTEX_PARENT);
+               pipe_lock_nested(pipe2, I_MUTEX_CHILD);
+       } else {
+               pipe_lock_nested(pipe2, I_MUTEX_PARENT);
+               pipe_lock_nested(pipe1, I_MUTEX_CHILD);
+       }
+}
+
 /* Drop the inode semaphore and wait for a pipe event, atomically */
 void pipe_wait(struct pipe_inode_info *pipe)
 {
@@ -47,12 +110,10 @@ void pipe_wait(struct pipe_inode_info *pipe)
         * is considered a noninteractive wait:
         */
        prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
-       if (pipe->inode)
-               mutex_unlock(&pipe->inode->i_mutex);
+       pipe_unlock(pipe);
        schedule();
        finish_wait(&pipe->wait, &wait);
-       if (pipe->inode)
-               mutex_lock(&pipe->inode->i_mutex);
+       pipe_lock(pipe);
 }
 
 static int
@@ -176,18 +237,19 @@ static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
  *     and the caller has to be careful not to fault before calling
  *     the unmap function.
  *
- *     Note that this function occupies KM_USER0 if @atomic != 0.
+ *     Note that this function calls kmap_atomic() if @atomic != 0.
  */
 void *generic_pipe_buf_map(struct pipe_inode_info *pipe,
                           struct pipe_buffer *buf, int atomic)
 {
        if (atomic) {
                buf->flags |= PIPE_BUF_FLAG_ATOMIC;
-               return kmap_atomic(buf->page, KM_USER0);
+               return kmap_atomic(buf->page);
        }
 
        return kmap(buf->page);
 }
+EXPORT_SYMBOL(generic_pipe_buf_map);
 
 /**
  * generic_pipe_buf_unmap - unmap a previously mapped pipe buffer
@@ -203,10 +265,11 @@ void generic_pipe_buf_unmap(struct pipe_inode_info *pipe,
 {
        if (buf->flags & PIPE_BUF_FLAG_ATOMIC) {
                buf->flags &= ~PIPE_BUF_FLAG_ATOMIC;
-               kunmap_atomic(map_data, KM_USER0);
+               kunmap_atomic(map_data);
        } else
                kunmap(buf->page);
 }
+EXPORT_SYMBOL(generic_pipe_buf_unmap);
 
 /**
  * generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
@@ -237,6 +300,7 @@ int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
 
        return 1;
 }
+EXPORT_SYMBOL(generic_pipe_buf_steal);
 
 /**
  * generic_pipe_buf_get - get a reference to a &struct pipe_buffer
@@ -252,6 +316,7 @@ void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
 {
        page_cache_get(buf->page);
 }
+EXPORT_SYMBOL(generic_pipe_buf_get);
 
 /**
  * generic_pipe_buf_confirm - verify contents of the pipe buffer
@@ -267,6 +332,22 @@ int generic_pipe_buf_confirm(struct pipe_inode_info *info,
 {
        return 0;
 }
+EXPORT_SYMBOL(generic_pipe_buf_confirm);
+
+/**
+ * generic_pipe_buf_release - put a reference to a &struct pipe_buffer
+ * @pipe:      the pipe that the buffer belongs to
+ * @buf:       the buffer to put a reference to
+ *
+ * Description:
+ *     This function releases a reference to @buf.
+ */
+void generic_pipe_buf_release(struct pipe_inode_info *pipe,
+                             struct pipe_buffer *buf)
+{
+       page_cache_release(buf->page);
+}
+EXPORT_SYMBOL(generic_pipe_buf_release);
 
 static const struct pipe_buf_operations anon_pipe_buf_ops = {
        .can_merge = 1,
@@ -278,13 +359,22 @@ static const struct pipe_buf_operations anon_pipe_buf_ops = {
        .get = generic_pipe_buf_get,
 };
 
+static const struct pipe_buf_operations packet_pipe_buf_ops = {
+       .can_merge = 0,
+       .map = generic_pipe_buf_map,
+       .unmap = generic_pipe_buf_unmap,
+       .confirm = generic_pipe_buf_confirm,
+       .release = anon_pipe_buf_release,
+       .steal = generic_pipe_buf_steal,
+       .get = generic_pipe_buf_get,
+};
+
 static ssize_t
 pipe_read(struct kiocb *iocb, const struct iovec *_iov,
           unsigned long nr_segs, loff_t pos)
 {
        struct file *filp = iocb->ki_filp;
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       struct pipe_inode_info *pipe;
+       struct pipe_inode_info *pipe = filp->private_data;
        int do_wakeup;
        ssize_t ret;
        struct iovec *iov = (struct iovec *)_iov;
@@ -297,8 +387,7 @@ pipe_read(struct kiocb *iocb, const struct iovec *_iov,
 
        do_wakeup = 0;
        ret = 0;
-       mutex_lock(&inode->i_mutex);
-       pipe = inode->i_pipe;
+       __pipe_lock(pipe);
        for (;;) {
                int bufs = pipe->nrbufs;
                if (bufs) {
@@ -315,7 +404,7 @@ pipe_read(struct kiocb *iocb, const struct iovec *_iov,
                        error = ops->confirm(pipe, buf);
                        if (error) {
                                if (!ret)
-                                       error = ret;
+                                       ret = error;
                                break;
                        }
 
@@ -339,10 +428,17 @@ redo:
                        ret += chars;
                        buf->offset += chars;
                        buf->len -= chars;
+
+                       /* Was it a packet buffer? Clean up and exit */
+                       if (buf->flags & PIPE_BUF_FLAG_PACKET) {
+                               total_len = chars;
+                               buf->len = 0;
+                       }
+
                        if (!buf->len) {
                                buf->ops = NULL;
                                ops->release(pipe, buf);
-                               curbuf = (curbuf + 1) & (PIPE_BUFFERS-1);
+                               curbuf = (curbuf + 1) & (pipe->buffers - 1);
                                pipe->curbuf = curbuf;
                                pipe->nrbufs = --bufs;
                                do_wakeup = 1;
@@ -374,16 +470,16 @@ redo:
                        break;
                }
                if (do_wakeup) {
-                       wake_up_interruptible_sync(&pipe->wait);
+                       wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
                        kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
                }
                pipe_wait(pipe);
        }
-       mutex_unlock(&inode->i_mutex);
+       __pipe_unlock(pipe);
 
        /* Signal writers asynchronously that there is more room. */
        if (do_wakeup) {
-               wake_up_interruptible_sync(&pipe->wait);
+               wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
                kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
        }
        if (ret > 0)
@@ -391,13 +487,17 @@ redo:
        return ret;
 }
 
+static inline int is_packetized(struct file *file)
+{
+       return (file->f_flags & O_DIRECT) != 0;
+}
+
 static ssize_t
 pipe_write(struct kiocb *iocb, const struct iovec *_iov,
            unsigned long nr_segs, loff_t ppos)
 {
        struct file *filp = iocb->ki_filp;
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       struct pipe_inode_info *pipe;
+       struct pipe_inode_info *pipe = filp->private_data;
        ssize_t ret;
        int do_wakeup;
        struct iovec *iov = (struct iovec *)_iov;
@@ -411,8 +511,7 @@ pipe_write(struct kiocb *iocb, const struct iovec *_iov,
 
        do_wakeup = 0;
        ret = 0;
-       mutex_lock(&inode->i_mutex);
-       pipe = inode->i_pipe;
+       __pipe_lock(pipe);
 
        if (!pipe->readers) {
                send_sig(SIGPIPE, current, 0);
@@ -424,7 +523,7 @@ pipe_write(struct kiocb *iocb, const struct iovec *_iov,
        chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
        if (pipe->nrbufs && chars != 0) {
                int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
-                                                       (PIPE_BUFFERS-1);
+                                                       (pipe->buffers - 1);
                struct pipe_buffer *buf = pipe->bufs + lastbuf;
                const struct pipe_buf_operations *ops = buf->ops;
                int offset = buf->offset + buf->len;
@@ -470,8 +569,8 @@ redo1:
                        break;
                }
                bufs = pipe->nrbufs;
-               if (bufs < PIPE_BUFFERS) {
-                       int newbuf = (pipe->curbuf + bufs) & (PIPE_BUFFERS-1);
+               if (bufs < pipe->buffers) {
+                       int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
                        struct pipe_buffer *buf = pipe->bufs + newbuf;
                        struct page *page = pipe->tmp_page;
                        char *src;
@@ -498,14 +597,14 @@ redo1:
                        iov_fault_in_pages_read(iov, chars);
 redo2:
                        if (atomic)
-                               src = kmap_atomic(page, KM_USER0);
+                               src = kmap_atomic(page);
                        else
                                src = kmap(page);
 
                        error = pipe_iov_copy_from_user(src, iov, chars,
                                                        atomic);
                        if (atomic)
-                               kunmap_atomic(src, KM_USER0);
+                               kunmap_atomic(src);
                        else
                                kunmap(page);
 
@@ -525,6 +624,11 @@ redo2:
                        buf->ops = &anon_pipe_buf_ops;
                        buf->offset = 0;
                        buf->len = chars;
+                       buf->flags = 0;
+                       if (is_packetized(filp)) {
+                               buf->ops = &packet_pipe_buf_ops;
+                               buf->flags = PIPE_BUF_FLAG_PACKET;
+                       }
                        pipe->nrbufs = ++bufs;
                        pipe->tmp_page = NULL;
 
@@ -532,7 +636,7 @@ redo2:
                        if (!total_len)
                                break;
                }
-               if (bufs < PIPE_BUFFERS)
+               if (bufs < pipe->buffers)
                        continue;
                if (filp->f_flags & O_NONBLOCK) {
                        if (!ret)
@@ -545,7 +649,7 @@ redo2:
                        break;
                }
                if (do_wakeup) {
-                       wake_up_interruptible_sync(&pipe->wait);
+                       wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
                        kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
                        do_wakeup = 0;
                }
@@ -554,51 +658,39 @@ redo2:
                pipe->waiting_writers--;
        }
 out:
-       mutex_unlock(&inode->i_mutex);
+       __pipe_unlock(pipe);
        if (do_wakeup) {
-               wake_up_interruptible_sync(&pipe->wait);
+               wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
                kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
        }
-       if (ret > 0)
-               file_update_time(filp);
+       if (ret > 0) {
+               int err = file_update_time(filp);
+               if (err)
+                       ret = err;
+       }
        return ret;
 }
 
-static ssize_t
-bad_pipe_r(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
-{
-       return -EBADF;
-}
-
-static ssize_t
-bad_pipe_w(struct file *filp, const char __user *buf, size_t count,
-          loff_t *ppos)
-{
-       return -EBADF;
-}
-
 static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 {
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       struct pipe_inode_info *pipe;
+       struct pipe_inode_info *pipe = filp->private_data;
        int count, buf, nrbufs;
 
        switch (cmd) {
                case FIONREAD:
-                       mutex_lock(&inode->i_mutex);
-                       pipe = inode->i_pipe;
+                       __pipe_lock(pipe);
                        count = 0;
                        buf = pipe->curbuf;
                        nrbufs = pipe->nrbufs;
                        while (--nrbufs >= 0) {
                                count += pipe->bufs[buf].len;
-                               buf = (buf+1) & (PIPE_BUFFERS-1);
+                               buf = (buf+1) & (pipe->buffers - 1);
                        }
-                       mutex_unlock(&inode->i_mutex);
+                       __pipe_unlock(pipe);
 
                        return put_user(count, (int __user *)arg);
                default:
-                       return -EINVAL;
+                       return -ENOIOCTLCMD;
        }
 }
 
@@ -607,8 +699,7 @@ static unsigned int
 pipe_poll(struct file *filp, poll_table *wait)
 {
        unsigned int mask;
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       struct pipe_inode_info *pipe = inode->i_pipe;
+       struct pipe_inode_info *pipe = filp->private_data;
        int nrbufs;
 
        poll_wait(filp, &pipe->wait, wait);
@@ -623,7 +714,7 @@ pipe_poll(struct file *filp, poll_table *wait)
        }
 
        if (filp->f_mode & FMODE_WRITE) {
-               mask |= (nrbufs < PIPE_BUFFERS) ? POLLOUT | POLLWRNORM : 0;
+               mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
                /*
                 * Most Unices do not set POLLERR for FIFOs but on Linux they
                 * behave exactly like pipes for poll().
@@ -635,269 +726,98 @@ pipe_poll(struct file *filp, poll_table *wait)
        return mask;
 }
 
-static int
-pipe_release(struct inode *inode, int decr, int decw)
+static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
 {
-       struct pipe_inode_info *pipe;
-
-       mutex_lock(&inode->i_mutex);
-       pipe = inode->i_pipe;
-       pipe->readers -= decr;
-       pipe->writers -= decw;
+       int kill = 0;
 
-       if (!pipe->readers && !pipe->writers) {
-               free_pipe_info(inode);
-       } else {
-               wake_up_interruptible_sync(&pipe->wait);
-               kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
-               kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
+       spin_lock(&inode->i_lock);
+       if (!--pipe->files) {
+               inode->i_pipe = NULL;
+               kill = 1;
        }
-       mutex_unlock(&inode->i_mutex);
-
-       return 0;
-}
-
-static int
-pipe_read_fasync(int fd, struct file *filp, int on)
-{
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       int retval;
-
-       mutex_lock(&inode->i_mutex);
-       retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_readers);
-       mutex_unlock(&inode->i_mutex);
+       spin_unlock(&inode->i_lock);
 
-       if (retval < 0)
-               return retval;
-
-       return 0;
+       if (kill)
+               free_pipe_info(pipe);
 }
 
-
 static int
-pipe_write_fasync(int fd, struct file *filp, int on)
+pipe_release(struct inode *inode, struct file *file)
 {
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       int retval;
-
-       mutex_lock(&inode->i_mutex);
-       retval = fasync_helper(fd, filp, on, &inode->i_pipe->fasync_writers);
-       mutex_unlock(&inode->i_mutex);
-
-       if (retval < 0)
-               return retval;
-
-       return 0;
-}
-
-
-static int
-pipe_rdwr_fasync(int fd, struct file *filp, int on)
-{
-       struct inode *inode = filp->f_path.dentry->d_inode;
-       struct pipe_inode_info *pipe = inode->i_pipe;
-       int retval;
-
-       mutex_lock(&inode->i_mutex);
-
-       retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
-
-       if (retval >= 0)
-               retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
-
-       mutex_unlock(&inode->i_mutex);
-
-       if (retval < 0)
-               return retval;
-
-       return 0;
-}
+       struct pipe_inode_info *pipe = file->private_data;
 
+       __pipe_lock(pipe);
+       if (file->f_mode & FMODE_READ)
+               pipe->readers--;
+       if (file->f_mode & FMODE_WRITE)
+               pipe->writers--;
 
-static int
-pipe_read_release(struct inode *inode, struct file *filp)
-{
-       pipe_read_fasync(-1, filp, 0);
-       return pipe_release(inode, 1, 0);
-}
-
-static int
-pipe_write_release(struct inode *inode, struct file *filp)
-{
-       pipe_write_fasync(-1, filp, 0);
-       return pipe_release(inode, 0, 1);
-}
-
-static int
-pipe_rdwr_release(struct inode *inode, struct file *filp)
-{
-       int decr, decw;
-
-       pipe_rdwr_fasync(-1, filp, 0);
-       decr = (filp->f_mode & FMODE_READ) != 0;
-       decw = (filp->f_mode & FMODE_WRITE) != 0;
-       return pipe_release(inode, decr, decw);
-}
-
-static int
-pipe_read_open(struct inode *inode, struct file *filp)
-{
-       /* We could have perhaps used atomic_t, but this and friends
-          below are the only places.  So it doesn't seem worthwhile.  */
-       mutex_lock(&inode->i_mutex);
-       inode->i_pipe->readers++;
-       mutex_unlock(&inode->i_mutex);
+       if (pipe->readers || pipe->writers) {
+               wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
+               kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
+               kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
+       }
+       __pipe_unlock(pipe);
 
+       put_pipe_info(inode, pipe);
        return 0;
 }
 
 static int
-pipe_write_open(struct inode *inode, struct file *filp)
+pipe_fasync(int fd, struct file *filp, int on)
 {
-       mutex_lock(&inode->i_mutex);
-       inode->i_pipe->writers++;
-       mutex_unlock(&inode->i_mutex);
+       struct pipe_inode_info *pipe = filp->private_data;
+       int retval = 0;
 
-       return 0;
-}
-
-static int
-pipe_rdwr_open(struct inode *inode, struct file *filp)
-{
-       mutex_lock(&inode->i_mutex);
+       __pipe_lock(pipe);
        if (filp->f_mode & FMODE_READ)
-               inode->i_pipe->readers++;
-       if (filp->f_mode & FMODE_WRITE)
-               inode->i_pipe->writers++;
-       mutex_unlock(&inode->i_mutex);
-
-       return 0;
+               retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
+       if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
+               retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
+               if (retval < 0 && (filp->f_mode & FMODE_READ))
+                       /* this can happen only if on == T */
+                       fasync_helper(-1, filp, 0, &pipe->fasync_readers);
+       }
+       __pipe_unlock(pipe);
+       return retval;
 }
 
-/*
- * The file_operations structs are not static because they
- * are also used in linux/fs/fifo.c to do operations on FIFOs.
- */
-const struct file_operations read_fifo_fops = {
-       .llseek         = no_llseek,
-       .read           = do_sync_read,
-       .aio_read       = pipe_read,
-       .write          = bad_pipe_w,
-       .poll           = pipe_poll,
-       .unlocked_ioctl = pipe_ioctl,
-       .open           = pipe_read_open,
-       .release        = pipe_read_release,
-       .fasync         = pipe_read_fasync,
-};
-
-const struct file_operations write_fifo_fops = {
-       .llseek         = no_llseek,
-       .read           = bad_pipe_r,
-       .write          = do_sync_write,
-       .aio_write      = pipe_write,
-       .poll           = pipe_poll,
-       .unlocked_ioctl = pipe_ioctl,
-       .open           = pipe_write_open,
-       .release        = pipe_write_release,
-       .fasync         = pipe_write_fasync,
-};
-
-const struct file_operations rdwr_fifo_fops = {
-       .llseek         = no_llseek,
-       .read           = do_sync_read,
-       .aio_read       = pipe_read,
-       .write          = do_sync_write,
-       .aio_write      = pipe_write,
-       .poll           = pipe_poll,
-       .unlocked_ioctl = pipe_ioctl,
-       .open           = pipe_rdwr_open,
-       .release        = pipe_rdwr_release,
-       .fasync         = pipe_rdwr_fasync,
-};
-
-static const struct file_operations read_pipe_fops = {
-       .llseek         = no_llseek,
-       .read           = do_sync_read,
-       .aio_read       = pipe_read,
-       .write          = bad_pipe_w,
-       .poll           = pipe_poll,
-       .unlocked_ioctl = pipe_ioctl,
-       .open           = pipe_read_open,
-       .release        = pipe_read_release,
-       .fasync         = pipe_read_fasync,
-};
-
-static const struct file_operations write_pipe_fops = {
-       .llseek         = no_llseek,
-       .read           = bad_pipe_r,
-       .write          = do_sync_write,
-       .aio_write      = pipe_write,
-       .poll           = pipe_poll,
-       .unlocked_ioctl = pipe_ioctl,
-       .open           = pipe_write_open,
-       .release        = pipe_write_release,
-       .fasync         = pipe_write_fasync,
-};
-
-static const struct file_operations rdwr_pipe_fops = {
-       .llseek         = no_llseek,
-       .read           = do_sync_read,
-       .aio_read       = pipe_read,
-       .write          = do_sync_write,
-       .aio_write      = pipe_write,
-       .poll           = pipe_poll,
-       .unlocked_ioctl = pipe_ioctl,
-       .open           = pipe_rdwr_open,
-       .release        = pipe_rdwr_release,
-       .fasync         = pipe_rdwr_fasync,
-};
-
-struct pipe_inode_info * alloc_pipe_info(struct inode *inode)
+struct pipe_inode_info *alloc_pipe_info(void)
 {
        struct pipe_inode_info *pipe;
 
        pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL);
        if (pipe) {
-               init_waitqueue_head(&pipe->wait);
-               pipe->r_counter = pipe->w_counter = 1;
-               pipe->inode = inode;
+               pipe->bufs = kzalloc(sizeof(struct pipe_buffer) * PIPE_DEF_BUFFERS, GFP_KERNEL);
+               if (pipe->bufs) {
+                       init_waitqueue_head(&pipe->wait);
+                       pipe->r_counter = pipe->w_counter = 1;
+                       pipe->buffers = PIPE_DEF_BUFFERS;
+                       mutex_init(&pipe->mutex);
+                       return pipe;
+               }
+               kfree(pipe);
        }
 
-       return pipe;
+       return NULL;
 }
 
-void __free_pipe_info(struct pipe_inode_info *pipe)
+void free_pipe_info(struct pipe_inode_info *pipe)
 {
        int i;
 
-       for (i = 0; i < PIPE_BUFFERS; i++) {
+       for (i = 0; i < pipe->buffers; i++) {
                struct pipe_buffer *buf = pipe->bufs + i;
                if (buf->ops)
                        buf->ops->release(pipe, buf);
        }
        if (pipe->tmp_page)
                __free_page(pipe->tmp_page);
+       kfree(pipe->bufs);
        kfree(pipe);
 }
 
-void free_pipe_info(struct inode *inode)
-{
-       __free_pipe_info(inode->i_pipe);
-       inode->i_pipe = NULL;
-}
-
 static struct vfsmount *pipe_mnt __read_mostly;
-static int pipefs_delete_dentry(struct dentry *dentry)
-{
-       /*
-        * At creation time, we pretended this dentry was hashed
-        * (by clearing DCACHE_UNHASHED bit in d_flags)
-        * At delete time, we restore the truth : not hashed.
-        * (so that dput() can proceed correctly)
-        */
-       dentry->d_flags |= DCACHE_UNHASHED;
-       return 0;
-}
 
 /*
  * pipefs_dname() is called from d_path().
@@ -908,26 +828,28 @@ static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
                                dentry->d_inode->i_ino);
 }
 
-static struct dentry_operations pipefs_dentry_operations = {
-       .d_delete       = pipefs_delete_dentry,
+static const struct dentry_operations pipefs_dentry_operations = {
        .d_dname        = pipefs_dname,
 };
 
 static struct inode * get_pipe_inode(void)
 {
-       struct inode *inode = new_inode(pipe_mnt->mnt_sb);
+       struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
        struct pipe_inode_info *pipe;
 
        if (!inode)
                goto fail_inode;
 
-       pipe = alloc_pipe_info(inode);
+       inode->i_ino = get_next_ino();
+
+       pipe = alloc_pipe_info();
        if (!pipe)
                goto fail_iput;
-       inode->i_pipe = pipe;
 
+       inode->i_pipe = pipe;
+       pipe->files = 2;
        pipe->readers = pipe->writers = 1;
-       inode->i_fop = &rdwr_pipe_fops;
+       inode->i_fop = &pipefifo_fops;
 
        /*
         * Mark the inode dirty from the very beginning,
@@ -937,8 +859,8 @@ static struct inode * get_pipe_inode(void)
         */
        inode->i_state = I_DIRTY;
        inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
-       inode->i_uid = current->fsuid;
-       inode->i_gid = current->fsgid;
+       inode->i_uid = current_fsuid();
+       inode->i_gid = current_fsgid();
        inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 
        return inode;
@@ -950,99 +872,67 @@ fail_inode:
        return NULL;
 }
 
-struct file *create_write_pipe(void)
+int create_pipe_files(struct file **res, int flags)
 {
        int err;
-       struct inode *inode;
+       struct inode *inode = get_pipe_inode();
        struct file *f;
-       struct dentry *dentry;
-       struct qstr name = { .name = "" };
+       struct path path;
+       static struct qstr name = { .name = "" };
 
-       err = -ENFILE;
-       inode = get_pipe_inode();
        if (!inode)
-               goto err;
+               return -ENFILE;
 
        err = -ENOMEM;
-       dentry = d_alloc(pipe_mnt->mnt_sb->s_root, &name);
-       if (!dentry)
+       path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
+       if (!path.dentry)
                goto err_inode;
+       path.mnt = mntget(pipe_mnt);
 
-       dentry->d_op = &pipefs_dentry_operations;
-       /*
-        * We dont want to publish this dentry into global dentry hash table.
-        * We pretend dentry is already hashed, by unsetting DCACHE_UNHASHED
-        * This permits a working /proc/$pid/fd/XXX on pipes
-        */
-       dentry->d_flags &= ~DCACHE_UNHASHED;
-       d_instantiate(dentry, inode);
+       d_instantiate(path.dentry, inode);
 
        err = -ENFILE;
-       f = alloc_file(pipe_mnt, dentry, FMODE_WRITE, &write_pipe_fops);
-       if (!f)
+       f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
+       if (IS_ERR(f))
                goto err_dentry;
-       f->f_mapping = inode->i_mapping;
-
-       f->f_flags = O_WRONLY;
-       f->f_version = 0;
 
-       return f;
+       f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
+       f->private_data = inode->i_pipe;
 
- err_dentry:
-       free_pipe_info(inode);
-       dput(dentry);
-       return ERR_PTR(err);
+       res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
+       if (IS_ERR(res[0]))
+               goto err_file;
 
- err_inode:
-       free_pipe_info(inode);
-       iput(inode);
- err:
-       return ERR_PTR(err);
-}
+       path_get(&path);
+       res[0]->private_data = inode->i_pipe;
+       res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
+       res[1] = f;
+       return 0;
 
-void free_write_pipe(struct file *f)
-{
-       free_pipe_info(f->f_dentry->d_inode);
-       path_put(&f->f_path);
+err_file:
        put_filp(f);
-}
+err_dentry:
+       free_pipe_info(inode->i_pipe);
+       path_put(&path);
+       return err;
 
-struct file *create_read_pipe(struct file *wrf)
-{
-       struct file *f = get_empty_filp();
-       if (!f)
-               return ERR_PTR(-ENFILE);
-
-       /* Grab pipe from the writer */
-       f->f_path = wrf->f_path;
-       path_get(&wrf->f_path);
-       f->f_mapping = wrf->f_path.dentry->d_inode->i_mapping;
-
-       f->f_pos = 0;
-       f->f_flags = O_RDONLY;
-       f->f_op = &read_pipe_fops;
-       f->f_mode = FMODE_READ;
-       f->f_version = 0;
-
-       return f;
+err_inode:
+       free_pipe_info(inode->i_pipe);
+       iput(inode);
+       return err;
 }
 
-int do_pipe_flags(int *fd, int flags)
+static int __do_pipe_flags(int *fd, struct file **files, int flags)
 {
-       struct file *fw, *fr;
        int error;
        int fdw, fdr;
 
-       if (flags & ~O_CLOEXEC)
+       if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
                return -EINVAL;
 
-       fw = create_write_pipe();
-       if (IS_ERR(fw))
-               return PTR_ERR(fw);
-       fr = create_read_pipe(fw);
-       error = PTR_ERR(fr);
-       if (IS_ERR(fr))
-               goto err_write_pipe;
+       error = create_pipe_files(files, flags);
+       if (error)
+               return error;
 
        error = get_unused_fd_flags(flags);
        if (error < 0)
@@ -1054,75 +944,367 @@ int do_pipe_flags(int *fd, int flags)
                goto err_fdr;
        fdw = error;
 
-       error = audit_fd_pair(fdr, fdw);
-       if (error < 0)
-               goto err_fdw;
-
-       fd_install(fdr, fr);
-       fd_install(fdw, fw);
+       audit_fd_pair(fdr, fdw);
        fd[0] = fdr;
        fd[1] = fdw;
-
        return 0;
 
- err_fdw:
-       put_unused_fd(fdw);
  err_fdr:
        put_unused_fd(fdr);
  err_read_pipe:
-       path_put(&fr->f_path);
-       put_filp(fr);
- err_write_pipe:
-       free_write_pipe(fw);
+       fput(files[0]);
+       fput(files[1]);
        return error;
 }
 
-int do_pipe(int *fd)
+int do_pipe_flags(int *fd, int flags)
 {
-       return do_pipe_flags(fd, 0);
+       struct file *files[2];
+       int error = __do_pipe_flags(fd, files, flags);
+       if (!error) {
+               fd_install(fd[0], files[0]);
+               fd_install(fd[1], files[1]);
+       }
+       return error;
 }
 
 /*
  * sys_pipe() is the normal C calling standard for creating
  * a pipe. It's not the way Unix traditionally does this, though.
  */
-asmlinkage long __weak sys_pipe2(int __user *fildes, int flags)
+SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
 {
+       struct file *files[2];
        int fd[2];
        int error;
 
-       error = do_pipe_flags(fd, flags);
+       error = __do_pipe_flags(fd, files, flags);
        if (!error) {
-               if (copy_to_user(fildes, fd, sizeof(fd))) {
-                       sys_close(fd[0]);
-                       sys_close(fd[1]);
+               if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
+                       fput(files[0]);
+                       fput(files[1]);
+                       put_unused_fd(fd[0]);
+                       put_unused_fd(fd[1]);
                        error = -EFAULT;
+               } else {
+                       fd_install(fd[0], files[0]);
+                       fd_install(fd[1], files[1]);
                }
        }
        return error;
 }
 
-asmlinkage long __weak sys_pipe(int __user *fildes)
+SYSCALL_DEFINE1(pipe, int __user *, fildes)
 {
        return sys_pipe2(fildes, 0);
 }
 
+static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
+{
+       int cur = *cnt; 
+
+       while (cur == *cnt) {
+               pipe_wait(pipe);
+               if (signal_pending(current))
+                       break;
+       }
+       return cur == *cnt ? -ERESTARTSYS : 0;
+}
+
+static void wake_up_partner(struct pipe_inode_info *pipe)
+{
+       wake_up_interruptible(&pipe->wait);
+}
+
+static int fifo_open(struct inode *inode, struct file *filp)
+{
+       struct pipe_inode_info *pipe;
+       bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
+       int ret;
+
+       filp->f_version = 0;
+
+       spin_lock(&inode->i_lock);
+       if (inode->i_pipe) {
+               pipe = inode->i_pipe;
+               pipe->files++;
+               spin_unlock(&inode->i_lock);
+       } else {
+               spin_unlock(&inode->i_lock);
+               pipe = alloc_pipe_info();
+               if (!pipe)
+                       return -ENOMEM;
+               pipe->files = 1;
+               spin_lock(&inode->i_lock);
+               if (unlikely(inode->i_pipe)) {
+                       inode->i_pipe->files++;
+                       spin_unlock(&inode->i_lock);
+                       free_pipe_info(pipe);
+                       pipe = inode->i_pipe;
+               } else {
+                       inode->i_pipe = pipe;
+                       spin_unlock(&inode->i_lock);
+               }
+       }
+       filp->private_data = pipe;
+       /* OK, we have a pipe and it's pinned down */
+
+       __pipe_lock(pipe);
+
+       /* We can only do regular read/write on fifos */
+       filp->f_mode &= (FMODE_READ | FMODE_WRITE);
+
+       switch (filp->f_mode) {
+       case FMODE_READ:
+       /*
+        *  O_RDONLY
+        *  POSIX.1 says that O_NONBLOCK means return with the FIFO
+        *  opened, even when there is no process writing the FIFO.
+        */
+               pipe->r_counter++;
+               if (pipe->readers++ == 0)
+                       wake_up_partner(pipe);
+
+               if (!is_pipe && !pipe->writers) {
+                       if ((filp->f_flags & O_NONBLOCK)) {
+                               /* suppress POLLHUP until we have
+                                * seen a writer */
+                               filp->f_version = pipe->w_counter;
+                       } else {
+                               if (wait_for_partner(pipe, &pipe->w_counter))
+                                       goto err_rd;
+                       }
+               }
+               break;
+       
+       case FMODE_WRITE:
+       /*
+        *  O_WRONLY
+        *  POSIX.1 says that O_NONBLOCK means return -1 with
+        *  errno=ENXIO when there is no process reading the FIFO.
+        */
+               ret = -ENXIO;
+               if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
+                       goto err;
+
+               pipe->w_counter++;
+               if (!pipe->writers++)
+                       wake_up_partner(pipe);
+
+               if (!is_pipe && !pipe->readers) {
+                       if (wait_for_partner(pipe, &pipe->r_counter))
+                               goto err_wr;
+               }
+               break;
+       
+       case FMODE_READ | FMODE_WRITE:
+       /*
+        *  O_RDWR
+        *  POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
+        *  This implementation will NEVER block on a O_RDWR open, since
+        *  the process can at least talk to itself.
+        */
+
+               pipe->readers++;
+               pipe->writers++;
+               pipe->r_counter++;
+               pipe->w_counter++;
+               if (pipe->readers == 1 || pipe->writers == 1)
+                       wake_up_partner(pipe);
+               break;
+
+       default:
+               ret = -EINVAL;
+               goto err;
+       }
+
+       /* Ok! */
+       __pipe_unlock(pipe);
+       return 0;
+
+err_rd:
+       if (!--pipe->readers)
+               wake_up_interruptible(&pipe->wait);
+       ret = -ERESTARTSYS;
+       goto err;
+
+err_wr:
+       if (!--pipe->writers)
+               wake_up_interruptible(&pipe->wait);
+       ret = -ERESTARTSYS;
+       goto err;
+
+err:
+       __pipe_unlock(pipe);
+
+       put_pipe_info(inode, pipe);
+       return ret;
+}
+
+const struct file_operations pipefifo_fops = {
+       .open           = fifo_open,
+       .llseek         = no_llseek,
+       .read           = do_sync_read,
+       .aio_read       = pipe_read,
+       .write          = do_sync_write,
+       .aio_write      = pipe_write,
+       .poll           = pipe_poll,
+       .unlocked_ioctl = pipe_ioctl,
+       .release        = pipe_release,
+       .fasync         = pipe_fasync,
+};
+
+/*
+ * Allocate a new array of pipe buffers and copy the info over. Returns the
+ * pipe size if successful, or return -ERROR on error.
+ */
+static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long nr_pages)
+{
+       struct pipe_buffer *bufs;
+
+       /*
+        * We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
+        * expect a lot of shrink+grow operations, just free and allocate
+        * again like we would do for growing. If the pipe currently
+        * contains more buffers than arg, then return busy.
+        */
+       if (nr_pages < pipe->nrbufs)
+               return -EBUSY;
+
+       bufs = kcalloc(nr_pages, sizeof(*bufs), GFP_KERNEL | __GFP_NOWARN);
+       if (unlikely(!bufs))
+               return -ENOMEM;
+
+       /*
+        * The pipe array wraps around, so just start the new one at zero
+        * and adjust the indexes.
+        */
+       if (pipe->nrbufs) {
+               unsigned int tail;
+               unsigned int head;
+
+               tail = pipe->curbuf + pipe->nrbufs;
+               if (tail < pipe->buffers)
+                       tail = 0;
+               else
+                       tail &= (pipe->buffers - 1);
+
+               head = pipe->nrbufs - tail;
+               if (head)
+                       memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
+               if (tail)
+                       memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
+       }
+
+       pipe->curbuf = 0;
+       kfree(pipe->bufs);
+       pipe->bufs = bufs;
+       pipe->buffers = nr_pages;
+       return nr_pages * PAGE_SIZE;
+}
+
+/*
+ * Currently we rely on the pipe array holding a power-of-2 number
+ * of pages.
+ */
+static inline unsigned int round_pipe_size(unsigned int size)
+{
+       unsigned long nr_pages;
+
+       nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
+       return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
+}
+
+/*
+ * This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
+ * will return an error.
+ */
+int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
+                size_t *lenp, loff_t *ppos)
+{
+       int ret;
+
+       ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
+       if (ret < 0 || !write)
+               return ret;
+
+       pipe_max_size = round_pipe_size(pipe_max_size);
+       return ret;
+}
+
+/*
+ * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
+ * location, so checking ->i_pipe is not enough to verify that this is a
+ * pipe.
+ */
+struct pipe_inode_info *get_pipe_info(struct file *file)
+{
+       return file->f_op == &pipefifo_fops ? file->private_data : NULL;
+}
+
+long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+       struct pipe_inode_info *pipe;
+       long ret;
+
+       pipe = get_pipe_info(file);
+       if (!pipe)
+               return -EBADF;
+
+       __pipe_lock(pipe);
+
+       switch (cmd) {
+       case F_SETPIPE_SZ: {
+               unsigned int size, nr_pages;
+
+               size = round_pipe_size(arg);
+               nr_pages = size >> PAGE_SHIFT;
+
+               ret = -EINVAL;
+               if (!nr_pages)
+                       goto out;
+
+               if (!capable(CAP_SYS_RESOURCE) && size > pipe_max_size) {
+                       ret = -EPERM;
+                       goto out;
+               }
+               ret = pipe_set_size(pipe, nr_pages);
+               break;
+               }
+       case F_GETPIPE_SZ:
+               ret = pipe->buffers * PAGE_SIZE;
+               break;
+       default:
+               ret = -EINVAL;
+               break;
+       }
+
+out:
+       __pipe_unlock(pipe);
+       return ret;
+}
+
+static const struct super_operations pipefs_ops = {
+       .destroy_inode = free_inode_nonrcu,
+       .statfs = simple_statfs,
+};
+
 /*
  * pipefs should _never_ be mounted by userland - too much of security hassle,
  * no real gain from having the whole whorehouse mounted. So we don't need
  * any operations on the root directory. However, we need a non-trivial
  * d_name - pipe: will go nicely and kill the special-casing in procfs.
  */
-static int pipefs_get_sb(struct file_system_type *fs_type,
-                        int flags, const char *dev_name, void *data,
-                        struct vfsmount *mnt)
+static struct dentry *pipefs_mount(struct file_system_type *fs_type,
+                        int flags, const char *dev_name, void *data)
 {
-       return get_sb_pseudo(fs_type, "pipe:", NULL, PIPEFS_MAGIC, mnt);
+       return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
+                       &pipefs_dentry_operations, PIPEFS_MAGIC);
 }
 
 static struct file_system_type pipe_fs_type = {
        .name           = "pipefs",
-       .get_sb         = pipefs_get_sb,
+       .mount          = pipefs_mount,
        .kill_sb        = kill_anon_super,
 };
 
@@ -1140,11 +1322,4 @@ static int __init init_pipe_fs(void)
        return err;
 }
 
-static void __exit exit_pipe_fs(void)
-{
-       unregister_filesystem(&pipe_fs_type);
-       mntput(pipe_mnt);
-}
-
 fs_initcall(init_pipe_fs);
-module_exit(exit_pipe_fs);