For software builds, the NFS client does many small

synchronous (with FILE_SYNC) writes because non-contiguous
byte ranges in the same buffer cache block are being
written. This patch adds a new mount option "noncontigwr"
which allows the non-contiguous byte ranges to be combined,
with the dirty byte range becoming the superset of the bytes
that are dirty, if the file has not been file locked.
This reduces the number of writes significantly for software
builds. The only case where this change might break existing
applications is where an application is writing
non-overlapping byte ranges within the same buffer cache block
of a file from multiple clients concurrently.
Since such an application would normally do file locking on
the file, avoiding the byte range merge for files that have
been file locked should be sufficient for most (maybe all?) cases.

Submitted by:	jhb (earlier version)
Reviewed by:	kib
MFC after:	3 weeks

4 files changed, 42 insertions, 11 deletions

diff --git a/sys/fs/nfsclient/nfs_clbio.c b/sys/fs/nfsclient/nfs_clbio.c
index 42421ad..dc106ee 100644
--- a/sys/fs/nfsclient/nfs_clbio.c
+++ b/sys/fs/nfsclient/nfs_clbio.c
@@ -872,7 +872,7 @@ ncl_write(struct vop_write_args *ap)
 	struct vattr vattr;
 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 	daddr_t lbn;
-	int bcount;
+	int bcount, noncontig_write, obcount;
 	int bp_cached, n, on, error = 0, error1;
 	size_t orig_resid, local_resid;
 	off_t orig_size, tmp_off;
@@ -1035,7 +1035,15 @@ again:
 		 * unaligned buffer size.
 		 */
 		mtx_lock(&np->n_mtx);
-		if (uio->uio_offset == np->n_size && n) {
+		if ((np->n_flag & NHASBEENLOCKED) == 0 &&
+		    (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0)
+			noncontig_write = 1;
+		else
+			noncontig_write = 0;
+		if ((uio->uio_offset == np->n_size ||
+		    (noncontig_write != 0 &&
+		    lbn == (np->n_size / biosize) &&
+		    uio->uio_offset + n > np->n_size)) && n) {
 			mtx_unlock(&np->n_mtx);
 			/*
 			 * Get the buffer (in its pre-append state to maintain
@@ -1043,8 +1051,8 @@ again:
 			 * nfsnode after we have locked the buffer to prevent
 			 * readers from reading garbage.
 			 */
-			bcount = on;
-			bp = nfs_getcacheblk(vp, lbn, bcount, td);
+			obcount = np->n_size - (lbn * biosize);
+			bp = nfs_getcacheblk(vp, lbn, obcount, td);
 
 			if (bp != NULL) {
 				long save;
@@ -1056,9 +1064,12 @@ again:
 				mtx_unlock(&np->n_mtx);
 
 				save = bp->b_flags & B_CACHE;
-				bcount += n;
+				bcount = on + n;
 				allocbuf(bp, bcount);
 				bp->b_flags |= save;
+				if (noncontig_write != 0 && on > obcount)
+					vfs_bio_bzero_buf(bp, obcount, on -
+					    obcount);
 			}
 		} else {
 			/*
@@ -1157,19 +1168,23 @@ again:
 		 * area, just update the b_dirtyoff and b_dirtyend,
 		 * otherwise force a write rpc of the old dirty area.
 		 *
+		 * If there has been a file lock applied to this file
+		 * or vfs.nfs.old_noncontig_writing is set, do the following:
 		 * While it is possible to merge discontiguous writes due to
 		 * our having a B_CACHE buffer ( and thus valid read data
 		 * for the hole), we don't because it could lead to
 		 * significant cache coherency problems with multiple clients,
 		 * especially if locking is implemented later on.
 		 *
-		 * As an optimization we could theoretically maintain
-		 * a linked list of discontinuous areas, but we would still
-		 * have to commit them separately so there isn't much
-		 * advantage to it except perhaps a bit of asynchronization.
+		 * If vfs.nfs.old_noncontig_writing is not set and there has
+		 * not been file locking done on this file:
+		 * Relax coherency a bit for the sake of performance and
+		 * expand the current dirty region to contain the new
+		 * write even if it means we mark some non-dirty data as
+		 * dirty.
 		 */
 
-		if (bp->b_dirtyend > 0 &&
+		if (noncontig_write == 0 && bp->b_dirtyend > 0 &&
 		    (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
 			if (bwrite(bp) == EINTR) {
 				error = EINTR;
diff --git a/sys/fs/nfsclient/nfs_clvfsops.c b/sys/fs/nfsclient/nfs_clvfsops.c
index 4a180c5..6b09356 100644
--- a/sys/fs/nfsclient/nfs_clvfsops.c
+++ b/sys/fs/nfsclient/nfs_clvfsops.c
@@ -719,7 +719,8 @@ static const char *nfs_opts[] = { "from", "nfs_args",
     "retrans", "acregmin", "acregmax", "acdirmin", "acdirmax", "resvport",
     "readahead", "hostname", "timeout", "addr", "fh", "nfsv3", "sec",
     "principal", "nfsv4", "gssname", "allgssname", "dirpath", "minorversion",
-    "nametimeo", "negnametimeo", "nocto", "pnfs", "wcommitsize",
+    "nametimeo", "negnametimeo", "nocto", "noncontigwr", "pnfs",
+    "wcommitsize",
     NULL };
 
 /*
@@ -840,6 +841,8 @@ nfs_mount(struct mount *mp)
 		args.flags |= NFSMNT_ALLGSSNAME;
 	if (vfs_getopt(mp->mnt_optnew, "nocto", NULL, NULL) == 0)
 		args.flags |= NFSMNT_NOCTO;
+	if (vfs_getopt(mp->mnt_optnew, "noncontigwr", NULL, NULL) == 0)
+		args.flags |= NFSMNT_NONCONTIGWR;
 	if (vfs_getopt(mp->mnt_optnew, "pnfs", NULL, NULL) == 0)
 		args.flags |= NFSMNT_PNFS;
 	if (vfs_getopt(mp->mnt_optnew, "readdirsize", (void **)&opt, NULL) == 0) {
@@ -1792,6 +1795,8 @@ void nfscl_retopts(struct nfsmount *nmp, char *buffer, size_t buflen)
 	    &blen);
 	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NOCTO) != 0, ",nocto", &buf,
 	    &blen);
+	nfscl_printopt(nmp, (nmp->nm_flag & NFSMNT_NONCONTIGWR) != 0,
+	    ",noncontigwr", &buf, &blen);
 	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
 	    0, ",lockd", &buf, &blen);
 	nfscl_printopt(nmp, (nmp->nm_flag & (NFSMNT_NOLOCKD | NFSMNT_NFSV4)) ==
diff --git a/sys/fs/nfsclient/nfs_clvnops.c b/sys/fs/nfsclient/nfs_clvnops.c
index 73616d3..73e5ed9 100644
--- a/sys/fs/nfsclient/nfs_clvnops.c
+++ b/sys/fs/nfsclient/nfs_clvnops.c
@@ -3078,6 +3078,10 @@ nfs_advlock(struct vop_advlock_args *ap)
 					np->n_change = va.va_filerev;
 				}
 			}
+			/* Mark that a file lock has been acquired. */
+			mtx_lock(&np->n_mtx);
+			np->n_flag |= NHASBEENLOCKED;
+			mtx_unlock(&np->n_mtx);
 		}
 		NFSVOPUNLOCK(vp, 0);
 		return (0);
@@ -3097,6 +3101,12 @@ nfs_advlock(struct vop_advlock_args *ap)
 				error = ENOLCK;
 			}
 		}
+		if (error == 0 && ap->a_op == F_SETLK) {
+			/* Mark that a file lock has been acquired. */
+			mtx_lock(&np->n_mtx);
+			np->n_flag |= NHASBEENLOCKED;
+			mtx_unlock(&np->n_mtx);
+		}
 	}
 	return (error);
 }
diff --git a/sys/fs/nfsclient/nfsnode.h b/sys/fs/nfsclient/nfsnode.h
index bbb67d7..d5cb756 100644
--- a/sys/fs/nfsclient/nfsnode.h
+++ b/sys/fs/nfsclient/nfsnode.h
@@ -157,6 +157,7 @@ struct nfsnode {
 #define	NLOCKWANT	0x00010000  /* Want the sleep lock */
 #define	NNOLAYOUT	0x00020000  /* Can't get a layout for this file */
 #define	NWRITEOPENED	0x00040000  /* Has been opened for writing */
+#define	NHASBEENLOCKED	0x00080000  /* Has been file locked. */
 
 /*
  * Convert between nfsnode pointers and vnode pointers