2002-01-10 14:31:15

by Craig Christophel

Subject: buffer.c lock_kernel() -- removal

Included is a patch against buffer.c that removes the
lock_kernel()/unlock_kernel() pairs in the few functions that they exist.

This is for 2.5>2.5.2-pre10.

I beleive that I checked all of the member functions for locking schematics
but if you can prove me wrong -- go for it. The only issue I can see would
be if DQUOT_SYNC() didn't do a lock_kernel() inside itself; but it does so
that is taken care of (utilizing BKL only if dquot is compiled in). Well
after all of that, here is the patch for review.


Craig.

Index: linux/fs/buffer.c
diff -u linux/fs/buffer.c:1.3 linux/fs/buffer.c:1.4
--- linux/fs/buffer.c:1.3 Wed Jan 9 02:18:26 2002
+++ linux/fs/buffer.c Thu Jan 10 09:14:10 2002
@@ -322,14 +322,12 @@
kdev_t dev = sb->s_dev;
sync_buffers(dev, 0);

- lock_kernel();
sync_inodes_sb(sb);
DQUOT_SYNC(sb);
lock_super(sb);
if (sb->s_dirt && sb->s_op && sb->s_op->write_super)
sb->s_op->write_super(sb);
unlock_super(sb);
- unlock_kernel();

return sync_buffers(dev, 1);
}
@@ -345,7 +343,6 @@
{
sync_buffers(dev, 0);

- lock_kernel();
sync_inodes(dev);
if (!kdev_none(dev)) {
struct super_block *sb = get_super(dev);
@@ -356,7 +353,6 @@
} else
DQUOT_SYNC(NULL);
sync_supers(dev);
- unlock_kernel();

return sync_buffers(dev, 1);
}
@@ -387,7 +383,6 @@
kdev_t dev;
int ret;

- lock_kernel();
/* sync the inode to buffers */
write_inode_now(inode, 0);

@@ -401,7 +396,6 @@
/* .. finally sync the buffers to disk */
dev = inode->i_dev;
ret = sync_buffers(dev, 1);
- unlock_kernel();
return ret;
}

@@ -2559,10 +2553,8 @@

static int sync_old_buffers(void)
{
- lock_kernel();
sync_unlocked_inodes();
sync_supers(NODEV);
- unlock_kernel();

for (;;) {
struct buffer_head *bh;

2002-01-13 03:45:53

by Alexander Viro

Subject: Re: buffer.c lock_kernel() -- removal

On Thu, 10 Jan 2002, Craig Christophel wrote:

> Included is a patch against buffer.c that removes the
> lock_kernel()/unlock_kernel() pairs in the few functions that they exist.
>
> This is for 2.5>2.5.2-pre10.
>
> I beleive that I checked all of the member functions for locking schematics
> but if you can prove me wrong -- go for it. The only issue I can see would

... and in the very first chunk we have

> - lock_kernel();
> sync_inodes_sb(sb);
> DQUOT_SYNC(sb);
> lock_super(sb);
> if (sb->s_dirt && sb->s_op && sb->s_op->write_super)
> sb->s_op->write_super(sb);
> unlock_super(sb);
> - unlock_kernel();

i.e. method that used to have BKL loses it. Unless you are willing to post
the results of audit for all filesystems - sorry, no go.