Align the offset in vn_rdwr_inchunks() so that at most the first and

the last chunk are misaligned relative to a MAXBSIZE byte boundary.
vn_rdwr_inchunks() is used mainly for elf core dumps, and elf sections
are usually perfectly misaligned relative to MAXBSIZE, and chunking
prevents the file system from doing much realigning.

This gives a surprisingly large speedup for core dumps -- from 50 to
13 seconds for a 512MB core dump here.  The pessimization was mostly
from an interaction of the misalignment with IO_DIRECT.  It increased
the number of i/o's for each chunk by a factor of 5 (3 writes and 2
read-before-writes instead of 1 write).

1 files changed, 11 insertions, 1 deletions

diff --git a/sys/kern/vfs_vnops.c b/sys/kern/vfs_vnops.c
index aa49d6f..e67e22e 100644
--- a/sys/kern/vfs_vnops.c
+++ b/sys/kern/vfs_vnops.c
@@ -476,8 +476,18 @@ vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
 	int error = 0;
 
 	do {
-		int chunk = (len > MAXBSIZE) ? MAXBSIZE : len;
+		int chunk;
 
+		/*
+		 * Force `offset' to a multiple of MAXBSIZE except possibly
+		 * for the first chunk, so that filesystems only need to
+		 * write full blocks except possibly for the first and last
+		 * chunks.
+		 */
+		chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
+
+		if (chunk > len)
+			chunk = len;
 		if (rw != UIO_READ && vp->v_type == VREG)
 			bwillwrite();
 		error = vn_rdwr(rw, vp, base, chunk, offset, segflg,