svn commit: r192686 - head/sys/nfsclient

[Doug Rabson]

Also, this problem appears to be there in the new experimental NFS  
client code as well. I haven't tried to fix that.


On 24 May 2009, at 14:22, Doug Rabson wrote:

> Author: dfr
> Date: Sun May 24 13:22:00 2009
> New Revision: 192686
> URL: http://svn.freebsd.org/changeset/base/192686
>
> Log:
>  Make sure we feed 32bit align memory to nfsm_dissect otherwise we  
> will fault
>  on platforms with strict alignment requirements. In particular,  
> this fixes the
>  problems with the new RPC transport on the arm platform.
>
>  Note: this adds yet another copy of nfs_realign(). I will attempt  
> to refactor
>  after NFS_LEGACYRPC is removed.
>
>  Submitted by:	sam
>
> Modified:
>  head/sys/nfsclient/nfs_krpc.c
>
> Modified: head/sys/nfsclient/nfs_krpc.c
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> = 
> ======================================================================
> --- head/sys/nfsclient/nfs_krpc.c	Sun May 24 12:39:38 2009	(r192685)
> +++ head/sys/nfsclient/nfs_krpc.c	Sun May 24 13:22:00 2009	(r192686)
> @@ -407,6 +407,65 @@ nfs_feedback(int type, int proc, void *a
> }
>
> /*
> + *	nfs_realign:
> + *
> + *	Check for badly aligned mbuf data and realign by copying the  
> unaligned
> + *	portion of the data into a new mbuf chain and freeing the portions
> + *	of the old chain that were replaced.
> + *
> + *	We cannot simply realign the data within the existing mbuf chain
> + *	because the underlying buffers may contain other rpc commands and
> + *	we cannot afford to overwrite them.
> + *
> + *	We would prefer to avoid this situation entirely.  The situation  
> does
> + *	not occur with NFS/UDP and is supposed to only occassionally occur
> + *	with TCP.  Use vfs.nfs.realign_count and realign_test to check  
> this.
> + *
> + */
> +static int
> +nfs_realign(struct mbuf **pm, int hsiz)
> +{
> +	struct mbuf *m, *n;
> +	int off, space;
> +
> +	++nfs_realign_test;
> +	while ((m = *pm) != NULL) {
> +		if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
> +			/*
> +			 * NB: we can't depend on m_pkthdr.len to help us
> +			 * decide what to do here.  May not be worth doing
> +			 * the m_length calculation as m_copyback will
> +			 * expand the mbuf chain below as needed.
> +			 */
> +			space = m_length(m, NULL);
> +			if (space >= MINCLSIZE) {
> +				/* NB: m_copyback handles space > MCLBYTES */
> +				n = m_getcl(M_DONTWAIT, MT_DATA, 0);
> +			} else
> +				n = m_get(M_DONTWAIT, MT_DATA);
> +			if (n == NULL)
> +				return (ENOMEM);
> +			/*
> +			 * Align the remainder of the mbuf chain.
> +			 */
> +			n->m_len = 0;
> +			off = 0;
> +			while (m != NULL) {
> +				m_copyback(n, off, m->m_len, mtod(m, caddr_t));
> +				off += m->m_len;
> +				m = m->m_next;
> +			}
> +			m_freem(*pm);
> +			*pm = n;
> +			++nfs_realign_count;
> +			break;
> +		}
> +		pm = &m->m_next;
> +	}
> +	return (0);
> +}
> +
> +/*
>  * nfs_request - goes something like this
>  *	- fill in request struct
>  *	- links it into list
> @@ -525,12 +584,25 @@ tryagain:
> 	} else {
> 		error = EACCES;
> 	}
> -	md = mrep;
> 	if (error)
> 		goto nfsmout;
>
> 	KASSERT(mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
>
> +	/*
> +	 * Search for any mbufs that are not a multiple of 4 bytes long
> +	 * or with m_data not longword aligned.
> +	 * These could cause pointer alignment problems, so copy them to
> +	 * well aligned mbufs.
> +	 */
> +	error = nfs_realign(&mrep, 2 * NFSX_UNSIGNED);
> +	if (error == ENOMEM) {
> +		m_freem(mrep);
> +		AUTH_DESTROY(auth);
> +		return (error);
> +	}
> +
> +	md = mrep;
> 	dpos = mtod(mrep, caddr_t);
> 	tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
> 	if (*tl != 0) {