svn commit: r204251 - stable/8/sys/cddl/boot/zfs

[Pawel Jakub Dawidek]

Author: pjd
Date: Tue Feb 23 16:46:34 2010
New Revision: 204251
URL: http://svn.freebsd.org/changeset/base/204251

Log:
  MFC r201684.
  
  Teach the (gpt)zfsboot and zfsloader raidz code to use its buffers
  more efficiently.
  
  Before this patch, in the worst case memory use would increase
  exponentially on the number of drives in the raidz vdev.
  
  Submitted by:	Matt Reimer <mattjreimer at gmail.com>
  Sponsored by:	VPOP Technologies, Inc.
  Silence from:	dfr

Modified:
  stable/8/sys/cddl/boot/zfs/zfssubr.c
Directory Properties:
  stable/8/sys/   (props changed)
  stable/8/sys/amd64/include/xen/   (props changed)
  stable/8/sys/cddl/contrib/opensolaris/   (props changed)
  stable/8/sys/contrib/dev/acpica/   (props changed)
  stable/8/sys/contrib/pf/   (props changed)
  stable/8/sys/dev/xen/xenpci/   (props changed)
  stable/8/sys/netinet/   (props changed)

Modified: stable/8/sys/cddl/boot/zfs/zfssubr.c
==============================================================================
--- stable/8/sys/cddl/boot/zfs/zfssubr.c	Tue Feb 23 16:39:53 2010	(r204250)
+++ stable/8/sys/cddl/boot/zfs/zfssubr.c	Tue Feb 23 16:46:34 2010	(r204251)
@@ -454,7 +454,7 @@ vdev_raidz_reconstruct_q(raidz_col_t *co
 
 static void
 vdev_raidz_reconstruct_pq(raidz_col_t *cols, int nparity, int acols,
-    int x, int y)
+    int x, int y, void *temp_p, void *temp_q)
 {
 	uint8_t *p, *q, *pxy, *qxy, *xd, *yd, tmp, a, b, aexp, bexp;
 	void *pdata, *qdata;
@@ -478,10 +478,8 @@ vdev_raidz_reconstruct_pq(raidz_col_t *c
 	xsize = cols[x].rc_size;
 	ysize = cols[y].rc_size;
 
-	cols[VDEV_RAIDZ_P].rc_data =
-		zfs_alloc_temp(cols[VDEV_RAIDZ_P].rc_size);
-	cols[VDEV_RAIDZ_Q].rc_data =
-		zfs_alloc_temp(cols[VDEV_RAIDZ_Q].rc_size);
+	cols[VDEV_RAIDZ_P].rc_data = temp_p;
+	cols[VDEV_RAIDZ_Q].rc_data = temp_q;
 	cols[x].rc_size = 0;
 	cols[y].rc_size = 0;
 
@@ -551,9 +549,12 @@ vdev_raidz_read(vdev_t *vdev, const blkp
 	uint64_t f = b % dcols;
 	uint64_t o = (b / dcols) << unit_shift;
 	uint64_t q, r, coff;
-	int c, c1, bc, col, acols, devidx, asize, n;
+	int c, c1, bc, col, acols, devidx, asize, n, max_rc_size;
 	static raidz_col_t cols[16];
 	raidz_col_t *rc, *rc1;
+	void *orig, *orig1, *temp_p, *temp_q;
+
+	orig = orig1 = temp_p = temp_q = NULL;
 
 	q = s / (dcols - nparity);
 	r = s - q * (dcols - nparity);
@@ -561,6 +562,7 @@ vdev_raidz_read(vdev_t *vdev, const blkp
 
 	acols = (q == 0 ? bc : dcols);
 	asize = 0;
+	max_rc_size = 0;
 	
 	for (c = 0; c < acols; c++) {
 		col = f + c;
@@ -577,6 +579,8 @@ vdev_raidz_read(vdev_t *vdev, const blkp
 		cols[c].rc_tried = 0;
 		cols[c].rc_skipped = 0;
 		asize += cols[c].rc_size;
+		if (cols[c].rc_size > max_rc_size)
+			max_rc_size = cols[c].rc_size;
 	}
 
 	asize = roundup(asize, (nparity + 1) << unit_shift);
@@ -777,8 +781,13 @@ reconstruct:
 			//ASSERT(c != acols);
 			//ASSERT(!rc->rc_skipped || rc->rc_error == ENXIO || rc->rc_error == ESTALE);
 
+			if (temp_p == NULL)
+				temp_p = zfs_alloc_temp(max_rc_size);
+			if (temp_q == NULL)
+				temp_q = zfs_alloc_temp(max_rc_size);
+
 			vdev_raidz_reconstruct_pq(cols, nparity, acols,
-			    c1, c);
+			    c1, c, temp_p, temp_q);
 
 			if (zio_checksum_error(bp, buf) == 0)
 				return (0);
@@ -845,18 +854,12 @@ reconstruct:
 		return (EIO);
 	}
 
-	asize = 0;
-	for (c = 0; c < acols; c++) {
-		rc = &cols[c];
-		if (rc->rc_size > asize)
-			asize = rc->rc_size;
-	}
 	if (cols[VDEV_RAIDZ_P].rc_error == 0) {
 		/*
 		 * Attempt to reconstruct the data from parity P.
 		 */
-		void *orig;
-		orig = zfs_alloc_temp(asize);
+		if (orig == NULL)
+			orig = zfs_alloc_temp(max_rc_size);
 		for (c = nparity; c < acols; c++) {
 			rc = &cols[c];
 
@@ -874,8 +877,8 @@ reconstruct:
 		/*
 		 * Attempt to reconstruct the data from parity Q.
 		 */
-		void *orig;
-		orig = zfs_alloc_temp(asize);
+		if (orig == NULL)
+			orig = zfs_alloc_temp(max_rc_size);
 		for (c = nparity; c < acols; c++) {
 			rc = &cols[c];
 
@@ -895,9 +898,14 @@ reconstruct:
 		/*
 		 * Attempt to reconstruct the data from both P and Q.
 		 */
-		void *orig, *orig1;
-		orig = zfs_alloc_temp(asize);
-		orig1 = zfs_alloc_temp(asize);
+		if (orig == NULL)
+			orig = zfs_alloc_temp(max_rc_size);
+		if (orig1 == NULL)
+			orig1 = zfs_alloc_temp(max_rc_size);
+		if (temp_p == NULL)
+			temp_p = zfs_alloc_temp(max_rc_size);
+		if (temp_q == NULL)
+			temp_q = zfs_alloc_temp(max_rc_size);
 		for (c = nparity; c < acols - 1; c++) {
 			rc = &cols[c];
 
@@ -909,7 +917,7 @@ reconstruct:
 				memcpy(orig1, rc1->rc_data, rc1->rc_size);
 
 				vdev_raidz_reconstruct_pq(cols, nparity,
-				    acols, c, c1);
+				    acols, c, c1, temp_p, temp_q);
 
 				if (zio_checksum_error(bp, buf) == 0)
 					return (0);