VFS: we need to set LOOKUP_JUMPED on mountpoint crossing
authorAl Viro <viro@ZenIV.linux.org.uk>
Mon, 7 Nov 2011 21:21:26 +0000 (21:21 +0000)
committerGreg Kroah-Hartman <gregkh@suse.de>
Fri, 11 Nov 2011 17:44:36 +0000 (09:44 -0800)
commitedfaaac6ea69b97b993d7a88f027b3d0f8cc2c63
treeae2bebb71cb7e751bbc6990fb70981c82d2eed03
parente94d1d00463f6fdcef956828ff698a2bce3f7a8f
VFS: we need to set LOOKUP_JUMPED on mountpoint crossing

commit a3fbbde70a0cec017f2431e8f8de208708c76acc upstream.

Mountpoint crossing is similar to following procfs symlinks - we do
not get ->d_revalidate() called for dentry we have arrived at, with
unpleasant consequences for NFS4.

Simple way to reproduce the problem in mainline:

    cat >/tmp/a.c <<'EOF'
    #include <unistd.h>
    #include <fcntl.h>
    #include <stdio.h>
    main()
    {
            struct flock fl = {.l_type = F_RDLCK, .l_whence = SEEK_SET, .l_len = 1};
            if (fcntl(0, F_SETLK, &fl))
                    perror("setlk");
    }
    EOF
cc /tmp/a.c -o /tmp/test

then on nfs4:

    mount --bind file1 file2
    /tmp/test < file1 # ok
    /tmp/test < file2 # spews "setlk: No locks available"...

What happens is the missing call of ->d_revalidate() after mountpoint
crossing and that's where NFS4 would issue OPEN request to server.

The fix is simple - treat mountpoint crossing the same way we deal with
following procfs-style symlinks.  I.e.  set LOOKUP_JUMPED...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
fs/namei.c