- Move nfs_realign() from the NFS client to the shared NFS code and

  remove the NFS server version in order to reduce code duplication.
  The shared version now uses a second parameter how, which is passed
  on to m_get(9) and m_getcl(9) as the server used M_WAIT while the
  client requires M_DONTWAIT, and replaces the the previously unused
  parameter hsiz.
- Change nfs_realign() to use nfsm_aligned() so as with other NFS code
  the alignment check isn't actually performed on platforms without
  strict alignment requirements for performance reasons because as the
  comment suggests unaligned data only occasionally occurs with TCP.
- Change fha_extract_info() to use nfs_realign() with M_DONTWAIT rather
  than M_WAIT because it's called with the RPC sp_lock held.

Reviewed by:	jhb, rmacklem
MFC after:	1 week

6 files changed, 73 insertions, 126 deletions

diff --git a/sys/nfs/nfs_common.c b/sys/nfs/nfs_common.c
index 6f42645..abf575f 100644
--- a/sys/nfs/nfs_common.c
+++ b/sys/nfs/nfs_common.c
@@ -56,6 +56,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/malloc.h>
 #include <sys/sysent.h>
 #include <sys/syscall.h>
+#include <sys/sysctl.h>
 
 #include <vm/vm.h>
 #include <vm/vm_object.h>
@@ -78,6 +79,16 @@ nfstype nfsv3_type[9] = {
 static void *nfsm_dissect_xx_sub(int s, struct mbuf **md, caddr_t *dpos,
     int how);
 
+SYSCTL_DECL(_vfs_nfs);
+
+static int nfs_realign_test;
+SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RD, &nfs_realign_test,
+    0, "Number of realign tests done");
+
+static int nfs_realign_count;
+SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RD, &nfs_realign_count,
+    0, "Number of mbuf realignments done");
+
 u_quad_t
 nfs_curusec(void)
 {
@@ -337,3 +348,59 @@ nfsm_adv_xx(int s, struct mbuf **md, caddr_t *dpos)
 		return (t1);
 	return (0);
 }
+
+/*
+ * 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.
+ */
+int
+nfs_realign(struct mbuf **pm, int how)
+{
+	struct mbuf *m, *n;
+	int off;
+
+	++nfs_realign_test;
+	while ((m = *pm) != NULL) {
+		if (!nfsm_aligned(m->m_len, u_int32_t) ||
+		    !nfsm_aligned(mtod(m, intptr_t), u_int32_t)) {
+			/*
+			 * 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.
+			 */
+			if (m_length(m, NULL) >= MINCLSIZE) {
+				/* NB: m_copyback handles space > MCLBYTES */
+				n = m_getcl(how, MT_DATA, 0);
+			} else
+				n = m_get(how, 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);
+}
diff --git a/sys/nfs/nfs_common.h b/sys/nfs/nfs_common.h
index 9ad1c11..1c0b315 100644
--- a/sys/nfs/nfs_common.h
+++ b/sys/nfs/nfs_common.h
@@ -48,6 +48,7 @@ extern nfstype nfsv3_type[];
 int	nfs_adv(struct mbuf **, caddr_t *, int, int);
 u_quad_t nfs_curusec(void);
 void	*nfsm_disct(struct mbuf **, caddr_t *, int, int, int);
+int	nfs_realign(struct mbuf **, int);
 
 /* ****************************** */
 /* Build request/reply phase macros */
diff --git a/sys/nfsclient/nfs_krpc.c b/sys/nfsclient/nfs_krpc.c
index 8b23682..bfb1ea1 100644
--- a/sys/nfsclient/nfs_krpc.c
+++ b/sys/nfsclient/nfs_krpc.c
@@ -87,8 +87,6 @@ uint32_t	nfsclient_nfs3_start_probes[NFS_NPROCS];
 uint32_t	nfsclient_nfs3_done_probes[NFS_NPROCS];
 #endif
 
-static int	nfs_realign_test;
-static int	nfs_realign_count;
 static int	nfs_bufpackets = 4;
 static int	nfs_reconnects;
 static int	nfs3_jukebox_delay = 10;
@@ -97,10 +95,6 @@ static int	fake_wchan;
 
 SYSCTL_DECL(_vfs_nfs);
 
-SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_test, CTLFLAG_RW, &nfs_realign_test, 0,
-    "Number of realign tests done");
-SYSCTL_INT(_vfs_nfs, OID_AUTO, realign_count, CTLFLAG_RW, &nfs_realign_count, 0,
-    "Number of mbuf realignments done");
 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
     "Buffer reservation size 2 < x < 64");
 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
@@ -403,65 +397,6 @@ nfs_feedback(int type, int proc, void *arg)
 }
 
 /*
- *	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
@@ -589,7 +524,7 @@ tryagain:
 	 * These could cause pointer alignment problems, so copy them to
 	 * well aligned mbufs.
 	 */
-	error = nfs_realign(&mrep, 2 * NFSX_UNSIGNED);
+	error = nfs_realign(&mrep, M_DONTWAIT);
 	if (error == ENOMEM) {
 		m_freem(mrep);
 		AUTH_DESTROY(auth);
diff --git a/sys/nfsserver/nfs.h b/sys/nfsserver/nfs.h
index 00dbe5b..bb1893e 100644
--- a/sys/nfsserver/nfs.h
+++ b/sys/nfsserver/nfs.h
@@ -239,7 +239,6 @@ extern int nfs_debug;
 
 #endif
 
-void	nfs_realign(struct mbuf **);
 struct mbuf *nfs_rephead(int, struct nfsrv_descript *, int, struct mbuf **,
 	    caddr_t *);
 void	nfsm_srvfattr(struct nfsrv_descript *, struct vattr *,
diff --git a/sys/nfsserver/nfs_fha.c b/sys/nfsserver/nfs_fha.c
index 25e930f..76537d7 100644
--- a/sys/nfsserver/nfs_fha.c
+++ b/sys/nfsserver/nfs_fha.c
@@ -202,7 +202,9 @@ fha_extract_info(struct svc_req *req, struct fha_info *i)
 	    procnum == NFSPROC_NULL)
 		goto out;
 
-	nfs_realign(&req->rq_args);
+	error = nfs_realign(&req->rq_args, M_DONTWAIT);
+	if (error)
+		goto out;
 	md = req->rq_args;
 	dpos = mtod(md, caddr_t);
 
diff --git a/sys/nfsserver/nfs_srvkrpc.c b/sys/nfsserver/nfs_srvkrpc.c
index bdfe424..512373b 100644
--- a/sys/nfsserver/nfs_srvkrpc.c
+++ b/sys/nfsserver/nfs_srvkrpc.c
@@ -96,8 +96,6 @@ SYSCTL_DECL(_vfs_nfsrv);
 SVCPOOL		*nfsrv_pool;
 int		nfsd_waiting = 0;
 int		nfsrv_numnfsd = 0;
-static int	nfs_realign_test;
-static int	nfs_realign_count;
 struct callout	nfsrv_callout;
 static eventhandler_tag nfsrv_nmbclusters_tag;
 
@@ -111,10 +109,6 @@ SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay, CTLFLAG_RW,
 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, gatherdelay_v3, CTLFLAG_RW,
     &nfsrvw_procrastinate_v3, 0,
     "Delay in seconds for NFSv3 write gathering");
-SYSCTL_INT(_vfs_nfsrv, OID_AUTO, realign_test, CTLFLAG_RW,
-	    &nfs_realign_test, 0, "");
-SYSCTL_INT(_vfs_nfsrv, OID_AUTO, realign_count, CTLFLAG_RW,
-	    &nfs_realign_count, 0, "");
 
 static int	nfssvc_addsock(struct file *, struct thread *);
 static int	nfssvc_nfsd(struct thread *, struct nfsd_nfsd_args *);
@@ -250,57 +244,6 @@ nfs_rephead(int siz, struct nfsrv_descript *nd, int err,
 	return (mreq);
 }
 
-/*
- *	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.
- */
-void
-nfs_realign(struct mbuf **pm)	/* XXX COMMON */
-{
-	struct mbuf *m;
-	struct mbuf *n = NULL;
-	int off = 0;
-
-	++nfs_realign_test;
-	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;
-	}
-
-	/*
-	 * 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;
-		}
-		m_freem(*pm);
-		*pm = n;
-	}
-}
-
 static void
 nfssvc_program(struct svc_req *rqst, SVCXPRT *xprt)
 {
@@ -334,7 +277,7 @@ nfssvc_program(struct svc_req *rqst, SVCXPRT *xprt)
 	mreq = mrep = NULL;
 	mreq = rqst->rq_args;
 	rqst->rq_args = NULL;
-	nfs_realign(&mreq);
+	(void)nfs_realign(&mreq, M_WAIT);
 
 	/*
 	 * Note: we want rq_addr, not svc_getrpccaller for nd_nam2 -