Module Name: src
Committed By: sborrill
Date: Fri Feb 24 17:58:45 UTC 2012
Modified Files:
src/sys/dev/raidframe [netbsd-5]: rf_reconmap.c rf_reconstruct.c
Log Message:
Pull up the following revisions(s) (requested by oster in ticket #1728):
sys/dev/raidframe/rf_reconmap.c: revision 1.34
sys/dev/raidframe/rf_reconstruct.c: revision 1.118
Remove a DIAGNOSTIC check that is invalid for RAID5_RS.
Add logic to the main reconstruction loop to handle RAID5 with rotated
spares. Correct issue where we were doing one more stripe than necessary.
To generate a diff of this commit:
cvs rdiff -u -r1.31 -r1.31.8.1 src/sys/dev/raidframe/rf_reconmap.c
cvs rdiff -u -r1.105.4.4 -r1.105.4.5 src/sys/dev/raidframe/rf_reconstruct.c
Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.
Modified files:
Index: src/sys/dev/raidframe/rf_reconmap.c
diff -u src/sys/dev/raidframe/rf_reconmap.c:1.31 src/sys/dev/raidframe/rf_reconmap.c:1.31.8.1
--- src/sys/dev/raidframe/rf_reconmap.c:1.31 Mon May 19 19:49:54 2008
+++ src/sys/dev/raidframe/rf_reconmap.c Fri Feb 24 17:58:44 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: rf_reconmap.c,v 1.31 2008/05/19 19:49:54 oster Exp $ */
+/* $NetBSD: rf_reconmap.c,v 1.31.8.1 2012/02/24 17:58:44 sborrill Exp $ */
/*
* Copyright (c) 1995 Carnegie-Mellon University.
* All rights reserved.
@@ -34,7 +34,7 @@
*************************************************************************/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: rf_reconmap.c,v 1.31 2008/05/19 19:49:54 oster Exp $");
+__KERNEL_RCSID(0, "$NetBSD: rf_reconmap.c,v 1.31.8.1 2012/02/24 17:58:44 sborrill Exp $");
#include "rf_raid.h"
#include <sys/time.h>
@@ -156,7 +156,14 @@ rf_ReconMapUpdate(RF_Raid_t *raidPtr, RF
/* do we need to move the queue? */
while (i > mapPtr->high_ru) {
+#if 0
#ifdef DIAGNOSTIC
+ /* XXX: The check below is not valid for
+ * RAID5_RS. It is valid for RAID 1 and RAID 5.
+ * The issue is that we can easily have
+ * RU_NOTHING entries here too, and those are
+ * quite correct.
+ */
if (mapPtr->status[mapPtr->head]!=RU_ALL) {
printf("\nraid%d: reconmap incorrect -- working on i %" PRIu64 "\n",
raidPtr->raidid, i);
@@ -169,6 +176,7 @@ rf_ReconMapUpdate(RF_Raid_t *raidPtr, RF
panic("reconmap incorrect");
}
#endif
+#endif
mapPtr->low_ru++;
mapPtr->high_ru++;
/* initialize "highest" RU status entry, which
Index: src/sys/dev/raidframe/rf_reconstruct.c
diff -u src/sys/dev/raidframe/rf_reconstruct.c:1.105.4.4 src/sys/dev/raidframe/rf_reconstruct.c:1.105.4.5
--- src/sys/dev/raidframe/rf_reconstruct.c:1.105.4.4 Sun Nov 21 22:06:53 2010
+++ src/sys/dev/raidframe/rf_reconstruct.c Fri Feb 24 17:58:44 2012
@@ -1,4 +1,4 @@
-/* $NetBSD: rf_reconstruct.c,v 1.105.4.4 2010/11/21 22:06:53 riz Exp $ */
+/* $NetBSD: rf_reconstruct.c,v 1.105.4.5 2012/02/24 17:58:44 sborrill Exp $ */
/*
* Copyright (c) 1995 Carnegie-Mellon University.
* All rights reserved.
@@ -33,7 +33,7 @@
************************************************************/
#include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: rf_reconstruct.c,v 1.105.4.4 2010/11/21 22:06:53 riz Exp $");
+__KERNEL_RCSID(0, "$NetBSD: rf_reconstruct.c,v 1.105.4.5 2012/02/24 17:58:44 sborrill Exp $");
#include <sys/param.h>
#include <sys/time.h>
@@ -557,6 +557,9 @@ rf_ContinueReconstructFailedDisk(RF_Raid
RF_ReconCtrl_t *tmp_reconctrl;
RF_ReconEvent_t *event;
RF_StripeCount_t incPSID,lastPSID,num_writes,pending_writes,prev;
+#if RF_INCLUDE_RAID5_RS > 0
+ RF_StripeCount_t startPSID,endPSID,aPSID,bPSID,offPSID;
+#endif
RF_ReconUnitCount_t RUsPerPU;
struct timeval etime, elpsd;
unsigned long xor_s, xor_resid_us;
@@ -609,7 +612,17 @@ rf_ContinueReconstructFailedDisk(RF_Raid
recon_error = 0;
write_error = 0;
pending_writes = incPSID;
- raidPtr->reconControl->lastPSID = incPSID;
+ raidPtr->reconControl->lastPSID = incPSID - 1;
+
+ /* bounds check raidPtr->reconControl->lastPSID and
+ pending_writes so that we don't attempt to wait for more IO
+ than can possibly happen */
+
+ if (raidPtr->reconControl->lastPSID > lastPSID)
+ raidPtr->reconControl->lastPSID = lastPSID;
+
+ if (pending_writes > lastPSID)
+ pending_writes = lastPSID;
/* start the actual reconstruction */
@@ -623,6 +636,49 @@ rf_ContinueReconstructFailedDisk(RF_Raid
}
num_writes = 0;
+
+#if RF_INCLUDE_RAID5_RS > 0
+ /* For RAID5 with Rotated Spares we will be 'short'
+ some number of writes since no writes will get
+ issued for stripes where the spare is on the
+ component being rebuilt. Account for the shortage
+ here so that we don't hang indefinitely below
+ waiting for writes to complete that were never
+ scheduled.
+
+ XXX: Should be fixed for PARITY_DECLUSTERING and
+ others too!
+
+ */
+
+ if (raidPtr->Layout.numDataCol <
+ raidPtr->numCol - raidPtr->Layout.numParityCol) {
+ /* numDataCol is at least 2 less than numCol, so
+ should be RAID 5 with Rotated Spares */
+
+ /* XXX need to update for RAID 6 */
+
+ startPSID = raidPtr->reconControl->lastPSID - pending_writes + 1;
+ endPSID = raidPtr->reconControl->lastPSID;
+
+ offPSID = raidPtr->numCol - col - 1;
+
+ aPSID = startPSID - startPSID % raidPtr->numCol + offPSID;
+ if (aPSID < startPSID) {
+ aPSID += raidPtr->numCol;
+ }
+
+ bPSID = endPSID - ((endPSID - offPSID) % raidPtr->numCol);
+
+ if (aPSID < endPSID) {
+ num_writes = ((bPSID - aPSID) / raidPtr->numCol) + 1;
+ }
+
+ if ((aPSID == endPSID) && (bPSID == endPSID)) {
+ num_writes++;
+ }
+ }
+#endif
/* issue a read for each surviving disk */
@@ -701,7 +757,7 @@ rf_ContinueReconstructFailedDisk(RF_Raid
#endif
}
- /* reads done, wakup any waiters, and then wait for writes */
+ /* reads done, wakeup any waiters, and then wait for writes */
rf_WakeupHeadSepCBWaiters(raidPtr);
@@ -1121,7 +1177,7 @@ IssueNextReadRequest(RF_Raid_t *raidPtr,
ctrl->ru_count = 0;
/* code left over from when head-sep was based on
* parity stripe id */
- if (ctrl->curPSID >= raidPtr->reconControl->lastPSID) {
+ if (ctrl->curPSID > raidPtr->reconControl->lastPSID) {
CheckForNewMinHeadSep(raidPtr, ++(ctrl->headSepCounter));
return (RF_RECON_DONE_READS); /* finito! */
}
