Skip site navigation (1)Skip section navigation (2) 
Date:      Fri, 16 May 2003 13:30:50 -0400 (EDT)
From:      Harikesavan Krishnan <harikesa@cs.sunysb.edu>
To:        freebsd-fs@freebsd.org
Subject:   Problem with Flags on fist  
Message-ID:  <Pine.GSO.4.53.0305161330080.16505@compserv2>
next in thread | raw e-mail | index | archive | help 

Hi,

We're working on porting fistgen (stackable templates) from freebsd 4.x to
FreeBSD 5.0.  I'm facing a small problem while renaming inside any of my
stackable file systems (such as cryptfs, rot13fs, etc.).  To achieve full
stacking transparency, we keep a componentname structure in our layer, as
well as the lower layer.  We have some code that tries to check the
cnp->cn_flags fields in the two layers.  We found that the flags differ in
both 4.x and 5.x: we're not sure why, or whether this is a bug.  It doesn't
cause anything bad that we've noticed, and we ran a large compile, fsx,
bonnie, etc. repeatedly.

Here's the code snippet:

static int cryptfs_rename(ap)
    struct vop_rename_args *ap;
{
  vnode_t *thisfdvp = ap->a_fdvp;
  vnode_t *thistdvp = ap->a_tdvp;

  // some code follows

  thisfcnp = ap->a_fcnp;
  lowerfcnp = cryptfs_new_cnp((thisfdvp), thisfcnp);
  ap->a_fcnp = lowerfcnp;

  error = cryptfs_bypass((struct vop_generic_args *) ap);

  if ((thisfcnp->cn_flags & PARAMASK) != (lowerfcnp->cn_flags & PARAMASK))
  {
    printk("%s: FLAGS CHANGED fthis:0x%x flower:0x%x", __FUNCTION__,
            (int)thisfcnp->cn_flags, (int) lowerfcnp->cn_flags);
  }
...
}


The flag values of the upper and lower layers are as follows:

  thisfcnp->cn_flags = 0x209410
  which is: (PDIRUNLOCK | ISLASTCN |  SAVESTART | HASBUF )

  lowerfcnp->cn_flags = 0x940c
  which is: (ISLASTCN | SAVESTART | HASBUF)

These are the same flag values that we've noticed when running stackable f/s
based on our 4.x templates.  However, in fbsd 4.x, the flags always match,
and the above printk "FLAGS CHANGED" is never printed.  Once we ported the
templates to 5.0, this printk started showing up.  We did notice that the
value of PARAMASX in 5.0 is different.

Fortunately, the bug is quite reproducible: mount the f/s, and rename a file
inside of it (same result whether the file existed or not).

So our questions are:

1. Is it ok to leave the code as is, or is there something wrong we're doing
   that may bite us later?

2. what could be causing the change in flags?  Are we doing something wrong
   in ->rename, or somewhere else we're not setting something right?

3. any advise on how to fix the code?


BTW, this is the last issue we have left before we can release a new version
of fistgen that includes fully working, brand new, templates for freebsd 4.x
and 5.0.

Thanks,
Hari and Erez.

Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.GSO.4.53.0305161330080.16505>