NetBSD/sys/dev/raidframe/rf_raid5.c

306 lines
9.9 KiB
C
Raw Normal View History

/* $NetBSD: rf_raid5.c,v 1.19 2006/11/16 01:33:23 christos Exp $ */
/*
* Copyright (c) 1995 Carnegie-Mellon University.
* All rights reserved.
*
* Author: Mark Holland
*
* Permission to use, copy, modify and distribute this software and
* its documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie the
* rights to redistribute these changes.
*/
/******************************************************************************
*
* rf_raid5.c -- implements RAID Level 5
*
*****************************************************************************/
2001-11-13 10:11:12 +03:00
#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: rf_raid5.c,v 1.19 2006/11/16 01:33:23 christos Exp $");
2001-11-13 10:11:12 +03:00
#include <dev/raidframe/raidframevar.h>
#include "rf_raid.h"
#include "rf_raid5.h"
#include "rf_dag.h"
#include "rf_dagffrd.h"
#include "rf_dagffwr.h"
#include "rf_dagdegrd.h"
#include "rf_dagdegwr.h"
#include "rf_dagutils.h"
#include "rf_general.h"
#include "rf_map.h"
#include "rf_utils.h"
typedef struct RF_Raid5ConfigInfo_s {
RF_RowCol_t **stripeIdentifier; /* filled in at config time and used
* by IdentifyStripe */
} RF_Raid5ConfigInfo_t;
2005-02-27 03:26:58 +03:00
int
rf_ConfigureRAID5(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
RF_Config_t *cfgPtr)
{
RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
RF_Raid5ConfigInfo_t *info;
RF_RowCol_t i, j, startdisk;
/* create a RAID level 5 configuration structure */
RF_MallocAndAdd(info, sizeof(RF_Raid5ConfigInfo_t), (RF_Raid5ConfigInfo_t *), raidPtr->cleanupList);
if (info == NULL)
return (ENOMEM);
layoutPtr->layoutSpecificInfo = (void *) info;
/* the stripe identifier must identify the disks in each stripe, IN
* THE ORDER THAT THEY APPEAR IN THE STRIPE. */
info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol, raidPtr->numCol, raidPtr->cleanupList);
if (info->stripeIdentifier == NULL)
return (ENOMEM);
startdisk = 0;
for (i = 0; i < raidPtr->numCol; i++) {
for (j = 0; j < raidPtr->numCol; j++) {
info->stripeIdentifier[i][j] = (startdisk + j) % raidPtr->numCol;
}
if ((--startdisk) < 0)
startdisk = raidPtr->numCol - 1;
}
/* fill in the remaining layout parameters */
layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk;
layoutPtr->numDataCol = raidPtr->numCol - 1;
layoutPtr->dataSectorsPerStripe = layoutPtr->numDataCol * layoutPtr->sectorsPerStripeUnit;
layoutPtr->numParityCol = 1;
layoutPtr->dataStripeUnitsPerDisk = layoutPtr->stripeUnitsPerDisk;
raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * layoutPtr->numDataCol * layoutPtr->sectorsPerStripeUnit;
return (0);
}
2005-02-27 03:26:58 +03:00
int
rf_GetDefaultNumFloatingReconBuffersRAID5(RF_Raid_t *raidPtr)
{
return (20);
}
2005-02-27 03:26:58 +03:00
RF_HeadSepLimit_t
rf_GetDefaultHeadSepLimitRAID5(RF_Raid_t *raidPtr)
{
return (10);
}
#if !defined(__NetBSD__) && !defined(_KERNEL)
/* not currently used */
2005-02-27 03:26:58 +03:00
int
rf_ShutdownRAID5(RF_Raid_t *raidPtr)
{
return (0);
}
#endif
2005-02-27 03:26:58 +03:00
void
rf_MapSectorRAID5(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
RF_RowCol_t *col, RF_SectorNum_t *diskSector,
int remap)
{
RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
*col = (SUID % raidPtr->numCol);
*diskSector = (SUID / (raidPtr->Layout.numDataCol)) * raidPtr->Layout.sectorsPerStripeUnit +
(raidSector % raidPtr->Layout.sectorsPerStripeUnit);
}
2005-02-27 03:26:58 +03:00
void
rf_MapParityRAID5(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
RF_RowCol_t *col, RF_SectorNum_t *diskSector,
int remap)
{
RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
*col = raidPtr->Layout.numDataCol - (SUID / raidPtr->Layout.numDataCol) % raidPtr->numCol;
*diskSector = (SUID / (raidPtr->Layout.numDataCol)) * raidPtr->Layout.sectorsPerStripeUnit +
(raidSector % raidPtr->Layout.sectorsPerStripeUnit);
}
2005-02-27 03:26:58 +03:00
void
rf_IdentifyStripeRAID5(RF_Raid_t *raidPtr, RF_RaidAddr_t addr,
RF_RowCol_t **diskids)
{
RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
RF_Raid5ConfigInfo_t *info = (RF_Raid5ConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
*diskids = info->stripeIdentifier[stripeID % raidPtr->numCol];
}
2005-02-27 03:26:58 +03:00
void
rf_MapSIDToPSIDRAID5(RF_RaidLayout_t *layoutPtr,
RF_StripeNum_t stripeID,
RF_StripeNum_t *psID, RF_ReconUnitNum_t *which_ru)
{
*which_ru = 0;
*psID = stripeID;
}
/* select an algorithm for performing an access. Returns two pointers,
* one to a function that will return information about the DAG, and
* another to a function that will create the dag.
*/
2005-02-27 03:26:58 +03:00
void
rf_RaidFiveDagSelect(RF_Raid_t *raidPtr, RF_IoType_t type,
RF_AccessStripeMap_t *asmap,
RF_VoidFuncPtr *createFunc)
{
RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
RF_PhysDiskAddr_t *failedPDA = NULL;
RF_RowCol_t fcol;
RF_RowStatus_t rstat;
int prior_recon;
RF_ASSERT(RF_IO_IS_R_OR_W(type));
Fix the "We panic if we can't create a DAG" problem that's existed ~forever. This requires a number of things: 1) If we can't create a DAG, set desc->numStripes to 0 in rf_SelectAlgorithm. This will ensure that we don't attempt to free any dagArray[] elements in rf_StateCleanup. 2) Modify rf_State_CreateDAG() to not panic in the event of a DAG failure. Instead, set the bp->b_flags and bp->b_error, and set things up to skip to rf_State_Cleanup(). 3) Need to mark desc->status as "bad" so that we actually stop looking for a different DAG. (which we won't find... no matter how many times we try). 4) rf_State_LastState() will then do the biodone(), and return EIO for the IO in question. 5) Remove some " || 1 "'s from ProcessNode(). These were for debugging, and we don't need the failure notices spewing over and over again as the failing DAGs are processed. 6) Needed to change if (asmap->numDataFailed + asmap->numParityFailed > 1) to if ((asmap->numDataFailed + asmap->numParityFailed > 1) || (raidPtr->numFailures > 1)){ in rf_raid5.c so that it doesn't try to return rf_CreateNonRedundantWriteDAG as the creation function. 7) Note that we can't apply the above change to the RAID 1 code as with the silly "fake 2-D" RAID 1 sets, it is possible to have 2 failed components in the RAID 1 set, and that would stop them from working. (I really don't know why/how those "fake 2-D" RAID 1 sets even work with all the "single-fault" assumptions present in the rest of the code.) 8) Needed to protect rf_RAID0DagSelect() in a similar way -- it should return NULL as the createFunc. 9) No point printing out "Multiple disks failed..." a zillion times.
2004-01-03 00:41:08 +03:00
if ((asmap->numDataFailed + asmap->numParityFailed > 1) ||
(raidPtr->numFailures > 1)){
#if RF_DEBUG_DAG
2005-02-27 03:26:58 +03:00
if (rf_dagDebug)
Fix the "We panic if we can't create a DAG" problem that's existed ~forever. This requires a number of things: 1) If we can't create a DAG, set desc->numStripes to 0 in rf_SelectAlgorithm. This will ensure that we don't attempt to free any dagArray[] elements in rf_StateCleanup. 2) Modify rf_State_CreateDAG() to not panic in the event of a DAG failure. Instead, set the bp->b_flags and bp->b_error, and set things up to skip to rf_State_Cleanup(). 3) Need to mark desc->status as "bad" so that we actually stop looking for a different DAG. (which we won't find... no matter how many times we try). 4) rf_State_LastState() will then do the biodone(), and return EIO for the IO in question. 5) Remove some " || 1 "'s from ProcessNode(). These were for debugging, and we don't need the failure notices spewing over and over again as the failing DAGs are processed. 6) Needed to change if (asmap->numDataFailed + asmap->numParityFailed > 1) to if ((asmap->numDataFailed + asmap->numParityFailed > 1) || (raidPtr->numFailures > 1)){ in rf_raid5.c so that it doesn't try to return rf_CreateNonRedundantWriteDAG as the creation function. 7) Note that we can't apply the above change to the RAID 1 code as with the silly "fake 2-D" RAID 1 sets, it is possible to have 2 failed components in the RAID 1 set, and that would stop them from working. (I really don't know why/how those "fake 2-D" RAID 1 sets even work with all the "single-fault" assumptions present in the rest of the code.) 8) Needed to protect rf_RAID0DagSelect() in a similar way -- it should return NULL as the createFunc. 9) No point printing out "Multiple disks failed..." a zillion times.
2004-01-03 00:41:08 +03:00
RF_ERRORMSG("Multiple disks failed in a single group! Aborting I/O operation.\n");
#endif
*createFunc = NULL;
return;
}
if (asmap->numDataFailed + asmap->numParityFailed == 1) {
2005-02-27 03:26:58 +03:00
/* if under recon & already reconstructed, redirect
* the access to the spare drive and eliminate the
* failure indication */
failedPDA = asmap->failedPDAs[0];
fcol = failedPDA->col;
rstat = raidPtr->status;
prior_recon = (rstat == rf_rs_reconfigured) || (
(rstat == rf_rs_reconstructing) ?
rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, failedPDA->startSector) : 0
);
if (prior_recon) {
#if RF_DEBUG_DAG > 0 || RF_DEBUG_MAP > 0
RF_RowCol_t oc = failedPDA->col;
RF_SectorNum_t oo = failedPDA->startSector;
2005-02-27 03:26:58 +03:00
#endif
#if RF_INCLUDE_PARITY_DECLUSTERING_DS > 0
if (layoutPtr->map->flags & RF_DISTRIBUTE_SPARE) { /* redirect to dist
* spare space */
2005-02-27 03:26:58 +03:00
if (failedPDA == asmap->parityInfo) {
2005-02-27 03:26:58 +03:00
/* parity has failed */
(layoutPtr->map->MapParity) (raidPtr, failedPDA->raidAddress,
&failedPDA->col, &failedPDA->startSector, RF_REMAP);
2005-02-27 03:26:58 +03:00
if (asmap->parityInfo->next) { /* redir 2nd component,
* if any */
RF_PhysDiskAddr_t *p = asmap->parityInfo->next;
RF_SectorNum_t SUoffs = p->startSector % layoutPtr->sectorsPerStripeUnit;
p->col = failedPDA->col;
p->startSector = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, failedPDA->startSector) +
SUoffs; /* cheating:
* startSector is not
* really a RAID address */
}
} else
if (asmap->parityInfo->next && failedPDA == asmap->parityInfo->next) {
RF_ASSERT(0); /* should not ever
* happen */
} else {
2005-02-27 03:26:58 +03:00
/* data has failed */
(layoutPtr->map->MapSector) (raidPtr, failedPDA->raidAddress,
&failedPDA->col, &failedPDA->startSector, RF_REMAP);
2005-02-27 03:26:58 +03:00
}
2005-02-27 03:26:58 +03:00
} else {
2005-02-27 03:26:58 +03:00
#endif
/* redirect to dedicated spare space */
2005-02-27 03:26:58 +03:00
failedPDA->col = raidPtr->Disks[fcol].spareCol;
2005-02-27 03:26:58 +03:00
/* the parity may have two distinct
* components, both of which may need
* to be redirected */
if (asmap->parityInfo->next) {
if (failedPDA == asmap->parityInfo) {
failedPDA->next->col = failedPDA->col;
} else
if (failedPDA == asmap->parityInfo->next) { /* paranoid: should
* never occur */
asmap->parityInfo->col = failedPDA->col;
}
}
#if RF_INCLUDE_PARITY_DECLUSTERING_DS > 0
}
#endif
RF_ASSERT(failedPDA->col != -1);
2005-02-27 03:26:58 +03:00
#if RF_DEBUG_DAG > 0 || RF_DEBUG_MAP > 0
if (rf_dagDebug || rf_mapDebug) {
printf("raid%d: Redirected type '%c' c %d o %ld -> c %d o %ld\n",
2005-02-27 03:26:58 +03:00
raidPtr->raidid, type, oc,
(long) oo, failedPDA->col,
(long) failedPDA->startSector);
}
#endif
asmap->numDataFailed = asmap->numParityFailed = 0;
}
}
/* all dags begin/end with block/unblock node therefore, hdrSucc &
* termAnt counts should always be 1 also, these counts should not be
* visible outside dag creation routines - manipulating the counts
* here should be removed */
if (type == RF_IO_TYPE_READ) {
if (asmap->numDataFailed == 0)
*createFunc = (RF_VoidFuncPtr) rf_CreateFaultFreeReadDAG;
else
*createFunc = (RF_VoidFuncPtr) rf_CreateRaidFiveDegradedReadDAG;
} else {
/* if mirroring, always use large writes. If the access
* requires two distinct parity updates, always do a small
* write. If the stripe contains a failure but the access
* does not, do a small write. The first conditional
* (numStripeUnitsAccessed <= numDataCol/2) uses a
* less-than-or-equal rather than just a less-than because
* when G is 3 or 4, numDataCol/2 is 1, and I want
* single-stripe-unit updates to use just one disk. */
if ((asmap->numDataFailed + asmap->numParityFailed) == 0) {
if (rf_suppressLocksAndLargeWrites ||
(((asmap->numStripeUnitsAccessed <= (layoutPtr->numDataCol / 2)) && (layoutPtr->numDataCol != 1)) ||
(asmap->parityInfo->next != NULL) || rf_CheckStripeForFailures(raidPtr, asmap))) {
*createFunc = (RF_VoidFuncPtr) rf_CreateSmallWriteDAG;
} else
*createFunc = (RF_VoidFuncPtr) rf_CreateLargeWriteDAG;
} else {
if (asmap->numParityFailed == 1)
*createFunc = (RF_VoidFuncPtr) rf_CreateNonRedundantWriteDAG;
else
2006-04-15 02:43:12 +04:00
if (asmap->numStripeUnitsAccessed != 1 && (failedPDA == NULL || failedPDA->numSector != layoutPtr->sectorsPerStripeUnit))
*createFunc = NULL;
else
*createFunc = (RF_VoidFuncPtr) rf_CreateDegradedWriteDAG;
}
}
}