lkml.org 
[lkml]   [2009]   [Jan]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
SubjectRe: nfsd stuckage
From
On Tue, Jan 06, 2009 at 06:05:51PM -0500, bfields wrote:
> On Wed, Jan 07, 2009 at 12:03:56AM +0100, Christoph Hellwig wrote:
> > On Tue, Jan 06, 2009 at 06:02:44PM -0500, J. Bruce Fields wrote:
> > > On Tue, Jan 06, 2009 at 02:56:12PM -0800, Andrew Morton wrote:
> > > >
> > > > I just built current mainline plus the just-sent 266 -mm patches.
> > > >
> > > > The machine failed to power off when hit with `halt -pfn'. dmesg output:
> > >
> > > Christoph, can you live with this for now?
> >
> > nfsd part is well, livable. But the fs/sync.c is buggy as stackable
> > filesystems can call vfs_fsync with a NULL pointer due to nfs calling it
> > that way, so please drop that hunk.
>
> Whoops, OK, I didn't understand that. I'll drop that hunk, retest, then
> submit--should take a few minutes.

So this works for me--and I guess I may as well submit it in a pull
request. But: it sounds like we still have a regression for ecryptfs?
(Since on ecryptfs export nfsd will still try to get the mutex twice on
create, unlink, etc.) Maybe we should just revert
4c728ef583b3d82266584da5cb068294c09df31e entirely for now?

--b.

commit 3fbc5c762bd9f9ff52fe7b5b09398a3cff0e8415
Author: J. Bruce Fields <bfields@citi.umich.edu>
Date: Tue Jan 6 13:37:03 2009 -0500

nfsd: fix double-locks of directory mutex

A number of nfsd operations depend on the i_mutex to cover more code
than just the fsync, so the approach of 4c728ef583b3d8 "add a vfs_fsync
helper" doesn't work for nfsd. Revert the parts of those patches that
touch nfsd.

Note: we can't, however, remove the logic from vfs_fsync that was needed
only for the special case of nfsd, because a vfs_fsync(NULL,...) call
can still result indirectly from a stackable filesystem that was called
by nfsd.

Reported-by: Eric Sesterhenn <snakebyte@gmx.de>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>

diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
index 44aa92a..6e50aaa 100644
--- a/fs/nfsd/vfs.c
+++ b/fs/nfsd/vfs.c
@@ -744,16 +744,44 @@ nfsd_close(struct file *filp)
fput(filp);
}

+/*
+ * Sync a file
+ * As this calls fsync (not fdatasync) there is no need for a write_inode
+ * after it.
+ */
+static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
+ const struct file_operations *fop)
+{
+ struct inode *inode = dp->d_inode;
+ int (*fsync) (struct file *, struct dentry *, int);
+ int err;
+
+ err = filemap_fdatawrite(inode->i_mapping);
+ if (err == 0 && fop && (fsync = fop->fsync))
+ err = fsync(filp, dp, 0);
+ if (err == 0)
+ err = filemap_fdatawait(inode->i_mapping);
+
+ return err;
+}
+
static int
nfsd_sync(struct file *filp)
{
- return vfs_fsync(filp, filp->f_path.dentry, 0);
+ int err;
+ struct inode *inode = filp->f_path.dentry->d_inode;
+ dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
+ mutex_lock(&inode->i_mutex);
+ err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
+ mutex_unlock(&inode->i_mutex);
+
+ return err;
}

int
-nfsd_sync_dir(struct dentry *dentry)
+nfsd_sync_dir(struct dentry *dp)
{
- return vfs_fsync(NULL, dentry, 0);
+ return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
}

/*

\
 
 \ /
  Last update: 2009-01-07 01:17    [W:0.037 / U:0.132 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site