kern/111831: [nfs] page fault while in kernel mode with samba in vfs_vmio_release

[Steve Sears]

I hit this bug and fixed it thusly in nfsclient/nfs_bio.c, around line 1735:

            } else {
            if (error) {
            bp->b_ioflags |= BIO_ERROR;
            /* Mark buffer invalid which will result in invalidating
             * its pages and other buffer cleanup in brelse().
             * Cannot set BIO_ERROR without marking buffer B_INVAL.
             */
+++         bp->b_flags |= B_INVAL;
            bp->b_error = np->n_error = error;
            np->n_flag |= NWRITEERR;
+++         np->n_attrstamp = 0;
            }
            bp->b_dirtyoff = bp->b_dirtyend = 0;

Part of the problem is that the transport is returning an error that is not
being dealt with gracefully. In my case it was EAGAIN. I also fixed the
socket code to retry in the session layer if it gets EAGAIN. I'm using a
specialized transport, so that part of the fix is not relevant.

I just looked at the file in the CVS repository and version 1.152.2.5,
2007/07/17 21:02:08, is the same fix. If this change is MFC'd to 6.2, the
submitter should be happy.


 -Steve