The VFS/BIO subsystem contained a number of hacks in order to optimize

piecemeal, middle-of-file writes for NFS.  These hacks have caused no
end of trouble, especially when combined with mmap().  I've removed
them.  Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write.  NFS does, however,
optimize piecemeal appends to files.  For most common file operations,
you will not notice the difference.  The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations.  NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write.  There is quite a bit of room for further
optimization in these areas.

The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault.  This
is not correct operation.  The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid.  A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid.  This operation is
necessary to properly support mmap().  The zeroing occurs most often
when dealing with file-EOF situations.  Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.

getblk() and allocbuf() have been rewritten.  B_CACHE operation is now
formally defined in comments and more straightforward in
implementation.  B_CACHE for VMIO buffers is based on the validity of
the backing store.  B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa).  biodone() is now responsible for setting B_CACHE
when a successful read completes.  B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated.  VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE.  This means that bowrite() and bawrite() also
set B_CACHE indirectly.

There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount.  These have been fixed.  getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.

Major fixes to NFS/TCP have been made.  A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain.  The server's kernel must be
recompiled to get the benefit of the fixes.

Submitted by:	Matthew Dillon <dillon@apollo.backplane.com>

3 files changed, 59 insertions, 87 deletions

diff --git a/sys/nfsserver/nfs.h b/sys/nfsserver/nfs.h
index bc15a7c..78a54a2 100644
--- a/sys/nfsserver/nfs.h
+++ b/sys/nfsserver/nfs.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.44 1998/09/07 05:42:15 bde Exp $
+ * $Id: nfs.h,v 1.45 1999/02/25 00:03:50 peter Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -651,8 +651,7 @@ void	nfs_disconnect __P((struct nfsmount *));
 void	nfs_safedisconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *, struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **, char **, const char *, long));
-int	nfs_bioread __P((struct vnode *, struct uio *, int, struct ucred *,
-			 int));
+int	nfs_bioread __P((struct vnode *, struct uio *, int, struct ucred *));
 int	nfsm_uiotombuf __P((struct uio *, struct mbuf **, int, caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));
diff --git a/sys/nfsserver/nfs_srvsock.c b/sys/nfsserver/nfs_srvsock.c
index 1490f72..2267629 100644
--- a/sys/nfsserver/nfs_srvsock.c
+++ b/sys/nfsserver/nfs_srvsock.c
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs_socket.c	8.5 (Berkeley) 3/30/95
- * $Id: nfs_socket.c,v 1.50 1999/02/25 00:03:51 peter Exp $
+ * $Id: nfs_socket.c,v 1.51 1999/04/24 11:29:48 dt Exp $
  */
 
 /*
@@ -54,6 +54,7 @@
 #include <sys/socketvar.h>
 #include <sys/syslog.h>
 #include <sys/tprintf.h>
+#include <sys/sysctl.h>
 
 #include <netinet/in.h>
 #include <netinet/tcp.h>
@@ -115,6 +116,15 @@ static int proct[NFS_NPROCS] = {
 	0, 0, 0,
 };
 
+static int nfs_realign_test;
+static int nfs_realign_count;
+
+SYSCTL_DECL(_vfs_nfs);
+
+SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RD, &nfs_realign_test, 0, "");
+SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RD, &nfs_realign_count, 0, "");
+
+
 /*
  * There is a congestion window for outstanding rpcs maintained per mount
  * point. The cwnd size is adjusted in roughly the way that:
@@ -138,7 +148,7 @@ struct callout_handle	nfs_timer_handle;
 static int	nfs_msg __P((struct proc *,char *,char *));
 static int	nfs_rcvlock __P((struct nfsreq *));
 static void	nfs_rcvunlock __P((struct nfsreq *));
-static void	nfs_realign __P((struct mbuf *m, int hsiz));
+static void	nfs_realign __P((struct mbuf **pm, int hsiz));
 static int	nfs_receive __P((struct nfsreq *rep, struct sockaddr **aname,
 				 struct mbuf **mp));
 static int	nfs_reconnect __P((struct nfsreq *rep));
@@ -702,7 +712,7 @@ errout:
 	 * These could cause pointer alignment problems, so copy them to
 	 * well aligned mbufs.
 	 */
-	nfs_realign(*mp, 5 * NFSX_UNSIGNED);
+	nfs_realign(mp, 5 * NFSX_UNSIGNED);
 	return (error);
 }
 
@@ -1589,92 +1599,56 @@ nfs_rcvunlock(rep)
 }
 
 /*
- * Check for badly aligned mbuf data areas and
- * realign data in an mbuf list by copying the data areas up, as required.
+ *	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 void
-nfs_realign(m, hsiz)
-	register struct mbuf *m;
+nfs_realign(pm, hsiz)
+	register struct mbuf **pm;
 	int hsiz;
 {
-	register struct mbuf *m2;
-	register int siz, mlen, olen;
-	register caddr_t tcp, fcp;
-	struct mbuf *mnew;
+	struct mbuf *m;
+	struct mbuf *n = NULL;
+	int off = 0;
 
-	while (m) {
-	    /*
-	     * This never happens for UDP, rarely happens for TCP
-	     * but frequently happens for iso transport.
-	     */
-	    if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
-		olen = m->m_len;
-		fcp = mtod(m, caddr_t);
-		if ((intptr_t)fcp & 0x3) {
-			m->m_flags &= ~M_PKTHDR;
-			if (m->m_flags & M_EXT)
-				m->m_data = m->m_ext.ext_buf +
-					((m->m_ext.ext_size - olen) & ~0x3);
-			else
-				m->m_data = m->m_dat;
-		}
-		m->m_len = 0;
-		tcp = mtod(m, caddr_t);
-		mnew = m;
-		m2 = m->m_next;
+	++nfs_realign_test;
 
-		/*
-		 * If possible, only put the first invariant part
-		 * of the RPC header in the first mbuf.
-		 */
-		mlen = M_TRAILINGSPACE(m);
-		if (olen <= hsiz && mlen > hsiz)
-			mlen = hsiz;
-
-		/*
-		 * Loop through the mbuf list consolidating data.
-		 */
-		while (m) {
-			while (olen > 0) {
-				if (mlen == 0) {
-					m2->m_flags &= ~M_PKTHDR;
-					if (m2->m_flags & M_EXT)
-						m2->m_data = m2->m_ext.ext_buf;
-					else
-						m2->m_data = m2->m_dat;
-					m2->m_len = 0;
-					mlen = M_TRAILINGSPACE(m2);
-					tcp = mtod(m2, caddr_t);
-					mnew = m2;
-					m2 = m2->m_next;
-				}
-				siz = min(mlen, olen);
-				if (tcp != fcp)
-					bcopy(fcp, tcp, siz);
-				mnew->m_len += siz;
-				mlen -= siz;
-				olen -= siz;
-				tcp += siz;
-				fcp += siz;
-			}
-			m = m->m_next;
-			if (m) {
-				olen = m->m_len;
-				fcp = mtod(m, caddr_t);
+	while ((m = *pm) != NULL) {
+		if ((m->m_len & 0x3) || (mtod(m, intptr_t) & 0x3)) {
+			MGET(n, M_WAIT, MT_DATA);
+			if (m->m_len >= MINCLSIZE) {
+				MCLGET(n, M_WAIT);
 			}
+			n->m_len = 0;
+			break;
 		}
+		pm = &m->m_next;
+	}
 
-		/*
-		 * Finally, set m_len == 0 for any trailing mbufs that have
-		 * been copied out of.
-		 */
-		while (m2) {
-			m2->m_len = 0;
-			m2 = m2->m_next;
+	/*
+	 * If n is non-NULL, loop on m copying data, then replace the
+	 * portion of the chain that had to be realigned.
+	 */
+	if (n != NULL) {
+		++nfs_realign_count;
+		while (m) {
+			m_copyback(n, off, m->m_len, mtod(m, caddr_t));
+			off += m->m_len;
+			m = m->m_next;
 		}
-		return;
-	    }
-	    m = m->m_next;
+		m_freem(*pm);
+		*pm = n;
 	}
 }
 
@@ -2040,7 +2014,7 @@ nfsrv_rcv(so, arg, waitflag)
 					m_freem(mp);
 					continue;
 				}
-				nfs_realign(mp, 10 * NFSX_UNSIGNED);
+				nfs_realign(&mp, 10 * NFSX_UNSIGNED);
 				rec->nr_address = nam;
 				rec->nr_packet = mp;
 				STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
@@ -2182,7 +2156,7 @@ nfsrv_getstream(slp, waitflag)
 		if (!rec) {
 		    m_freem(slp->ns_frag);
 		} else {
-		    nfs_realign(slp->ns_frag, 10 * NFSX_UNSIGNED);
+		    nfs_realign(&slp->ns_frag, 10 * NFSX_UNSIGNED);
 		    rec->nr_address = (struct sockaddr *)0;
 		    rec->nr_packet = slp->ns_frag;
 		    STAILQ_INSERT_TAIL(&slp->ns_rec, rec, nr_link);
diff --git a/sys/nfsserver/nfsrvstats.h b/sys/nfsserver/nfsrvstats.h
index bc15a7c..78a54a2 100644
--- a/sys/nfsserver/nfsrvstats.h
+++ b/sys/nfsserver/nfsrvstats.h
@@ -34,7 +34,7 @@
  * SUCH DAMAGE.
  *
  *	@(#)nfs.h	8.4 (Berkeley) 5/1/95
- * $Id: nfs.h,v 1.44 1998/09/07 05:42:15 bde Exp $
+ * $Id: nfs.h,v 1.45 1999/02/25 00:03:50 peter Exp $
  */
 
 #ifndef _NFS_NFS_H_
@@ -651,8 +651,7 @@ void	nfs_disconnect __P((struct nfsmount *));
 void	nfs_safedisconnect __P((struct nfsmount *));
 int	nfs_getattrcache __P((struct vnode *, struct vattr *));
 int	nfsm_strtmbuf __P((struct mbuf **, char **, const char *, long));
-int	nfs_bioread __P((struct vnode *, struct uio *, int, struct ucred *,
-			 int));
+int	nfs_bioread __P((struct vnode *, struct uio *, int, struct ucred *));
 int	nfsm_uiotombuf __P((struct uio *, struct mbuf **, int, caddr_t *));
 void	nfsrv_init __P((int));
 void	nfs_clearcommit __P((struct mount *));