Do not promote large async writes to sync.

Present implementation of large sync writes is too strict and so can be
quite slow.  Instead of doing that, execute large async write in chunks,
syncing each chunk separately.

It would be good to fix large sync writes too, but I leave it to somebody
with more skills in this area.

Reviewed by:	rmacklem
MFC after:	1 week

1 files changed, 15 insertions, 34 deletions

diff --git a/sys/fs/nfsclient/nfs_clbio.c b/sys/fs/nfsclient/nfs_clbio.c
index ea42537..e8de4a5 100644
--- a/sys/fs/nfsclient/nfs_clbio.c
+++ b/sys/fs/nfsclient/nfs_clbio.c
@@ -864,7 +864,7 @@ ncl_write(struct vop_write_args *ap)
 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 	daddr_t lbn;
 	int bcount, noncontig_write, obcount;
-	int bp_cached, n, on, error = 0, error1;
+	int bp_cached, n, on, error = 0, error1, wouldcommit;
 	size_t orig_resid, local_resid;
 	off_t orig_size, tmp_off;
 
@@ -908,7 +908,6 @@ ncl_write(struct vop_write_args *ap)
 			if (ioflag & IO_NDELAY)
 				return (EAGAIN);
 #endif
-flush_and_restart:
 			np->n_attrstamp = 0;
 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
@@ -965,27 +964,14 @@ flush_and_restart:
 	 * IO_UNIT -- we just make all writes atomic anyway, as there's
 	 * no point optimizing for something that really won't ever happen.
 	 */
+	wouldcommit = 0;
 	if (!(ioflag & IO_SYNC)) {
 		int nflag;
 
 		mtx_lock(&np->n_mtx);
 		nflag = np->n_flag;
 		mtx_unlock(&np->n_mtx);
-		int needrestart = 0;
-		if (nmp->nm_wcommitsize < uio->uio_resid) {
-			/*
-			 * If this request could not possibly be completed
-			 * without exceeding the maximum outstanding write
-			 * commit size, see if we can convert it into a
-			 * synchronous write operation.
-			 */
-			if (ioflag & IO_NDELAY)
-				return (EAGAIN);
-			ioflag |= IO_SYNC;
-			if (nflag & NMODIFIED)
-				needrestart = 1;
-		} else if (nflag & NMODIFIED) {
-			int wouldcommit = 0;
+		if (nflag & NMODIFIED) {
 			BO_LOCK(&vp->v_bufobj);
 			if (vp->v_bufobj.bo_dirty.bv_cnt != 0) {
 				TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd,
@@ -995,27 +981,22 @@ flush_and_restart:
 				}
 			}
 			BO_UNLOCK(&vp->v_bufobj);
-			/*
-			 * Since we're not operating synchronously and
-			 * bypassing the buffer cache, we are in a commit
-			 * and holding all of these buffers whether
-			 * transmitted or not.  If not limited, this
-			 * will lead to the buffer cache deadlocking,
-			 * as no one else can flush our uncommitted buffers.
-			 */
-			wouldcommit += uio->uio_resid;
-			/*
-			 * If we would initially exceed the maximum
-			 * outstanding write commit size, flush and restart.
-			 */
-			if (wouldcommit > nmp->nm_wcommitsize)
-				needrestart = 1;
 		}
-		if (needrestart)
-			goto flush_and_restart;
 	}
 
 	do {
+		if (!(ioflag & IO_SYNC)) {
+			wouldcommit += biosize;
+			if (wouldcommit > nmp->nm_wcommitsize) {
+				np->n_attrstamp = 0;
+				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
+				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
+				if (error)
+					return (error);
+				wouldcommit = biosize;
+			}
+		}
+
 		NFSINCRGLOBAL(newnfsstats.biocache_writes);
 		lbn = uio->uio_offset / biosize;
 		on = uio->uio_offset - (lbn * biosize);