Add support for ext_pgs mbufs to nfsrvd_readdir() and nfsrvd_readdirplus().

This patch code that optionally (based on ND_TLS, never set yet) generates
readdir replies in ext_pgs mbufs.
To trim the list back, a new function that is ext_pgs aware called
nfsm_trimtrailing() replaces newnfs_trimtrailing().
newnfs_trimtrailing() is no longer used, but will be removed in a future
commit, since its removal does modify the internal kpi between the NFS
modules.

This is another in the series of commits that add support to the NFS client
and server for building RPC messages in ext_pgs mbufs with anonymous pages.
This is useful so that the entire mbuf list does not need to be
copied before calling sosend() when NFS over TLS is enabled.

Use of ext_pgs mbufs will not be enabled until the kernel RPC is updated
to handle TLS.

1 files changed, 74 insertions, 2 deletions

diff --git a/sys/fs/nfsserver/nfs_nfsdport.c b/sys/fs/nfsserver/nfs_nfsdport.c
index 34066585c3a3..eb971d73d534 100644
--- a/sys/fs/nfsserver/nfs_nfsdport.c
+++ b/sys/fs/nfsserver/nfs_nfsdport.c
@@ -144,6 +144,8 @@ static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
     NFSPROC_T *);
 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
+static void nfsm_trimtrailing(struct nfsrv_descript *, struct mbuf *,
+    char *, int, int);
 
 int nfs_pnfsio(task_fn_t *, void *);
 
@@ -2043,6 +2045,17 @@ again:
 	vput(vp);
 
 	/*
+	 * If cnt > MCLBYTES and the reply will not be saved, use
+	 * ext_pgs mbufs for TLS.
+	 * For NFSv4.0, we do not know for sure if the reply will
+	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
+	 */
+	if (cnt > MCLBYTES && siz > MCLBYTES &&
+	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
+	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
+		nd->nd_flag |= ND_EXTPG;
+
+	/*
 	 * dirlen is the size of the reply, including all XDR and must
 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
 	 * if the XDR should be included in "count", but to be safe, we do.
@@ -2146,6 +2159,7 @@ nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
 	struct mount *mp, *new_mp;
 	uint64_t mounted_on_fileno;
 	struct thread *p = curthread;
+	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
 
 	if (nd->nd_repstat) {
 		nfsrv_postopattr(nd, getret, &at);
@@ -2359,11 +2373,27 @@ again:
 	}
 
 	/*
+	 * If the reply is likely to exceed MCLBYTES and the reply will
+	 * not be saved, use ext_pgs mbufs for TLS.
+	 * It is difficult to predict how large each entry will be and
+	 * how many entries have been read, so just assume the directory
+	 * entries grow by a factor of 4 when attributes are included.
+	 * For NFSv4.0, we do not know for sure if the reply will
+	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
+	 */
+	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
+	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
+	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
+		nd->nd_flag |= ND_EXTPG;
+
+	/*
 	 * Save this position, in case there is an error before one entry
 	 * is created.
 	 */
 	mb0 = nd->nd_mb;
 	bpos0 = nd->nd_bpos;
+	bextpg0 = nd->nd_bextpg;
+	bextpgsiz0 = nd->nd_bextpgsiz;
 
 	/*
 	 * Fill in the first part of the reply.
@@ -2385,6 +2415,8 @@ again:
 	 */
 	mb1 = nd->nd_mb;
 	bpos1 = nd->nd_bpos;
+	bextpg1 = nd->nd_bextpg;
+	bextpgsiz1 = nd->nd_bextpgsiz;
 
 	/* Loop through the records and build reply */
 	entrycnt = 0;
@@ -2401,6 +2433,8 @@ again:
 			 */
 			mb1 = nd->nd_mb;
 			bpos1 = nd->nd_bpos;
+			bextpg1 = nd->nd_bextpg;
+			bextpgsiz1 = nd->nd_bextpgsiz;
 	
 			/*
 			 * For readdir_and_lookup get the vnode using
@@ -2626,11 +2660,11 @@ invalid:
 		if (!nd->nd_repstat && entrycnt == 0)
 			nd->nd_repstat = NFSERR_TOOSMALL;
 		if (nd->nd_repstat) {
-			newnfs_trimtrailing(nd, mb0, bpos0);
+			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
 			if (nd->nd_flag & ND_NFSV3)
 				nfsrv_postopattr(nd, getret, &at);
 		} else
-			newnfs_trimtrailing(nd, mb1, bpos1);
+			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
 		eofflag = 0;
 	} else if (cpos < cend)
 		eofflag = 0;
@@ -6418,6 +6452,44 @@ out:
 	return (error);
 }
 
+/*
+ * Trim trailing data off the mbuf list being built.
+ */
+static void
+nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
+    int bextpg, int bextpgsiz)
+{
+	vm_page_t pg;
+	int fullpgsiz, i;
+
+	if (mb->m_next != NULL) {
+		m_freem(mb->m_next);
+		mb->m_next = NULL;
+	}
+	if ((mb->m_flags & M_EXTPG) != 0) {
+		/* First, get rid of any pages after this position. */
+		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
+			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
+			vm_page_unwire_noq(pg);
+			vm_page_free(pg);
+		}
+		mb->m_epg_npgs = bextpg + 1;
+		if (bextpg == 0)
+			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
+		else
+			fullpgsiz = PAGE_SIZE;
+		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
+		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
+		for (i = 1; i < mb->m_epg_npgs; i++)
+			mb->m_len += m_epg_pagelen(mb, i, 0);
+		nd->nd_bextpgsiz = bextpgsiz;
+		nd->nd_bextpg = bextpg;
+	} else
+		mb->m_len = bpos - mtod(mb, char *);
+	nd->nd_mb = mb;
+	nd->nd_bpos = bpos;
+}
+
 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
 
 /*