Home | History | Annotate | Line # | Download | only in raidframe
rf_raid1.c revision 1.39
      1  1.39     oster /*	$NetBSD: rf_raid1.c,v 1.39 2021/07/23 22:34:12 oster Exp $	*/
      2   1.1     oster /*
      3   1.1     oster  * Copyright (c) 1995 Carnegie-Mellon University.
      4   1.1     oster  * All rights reserved.
      5   1.1     oster  *
      6   1.1     oster  * Author: William V. Courtright II
      7   1.1     oster  *
      8   1.1     oster  * Permission to use, copy, modify and distribute this software and
      9   1.1     oster  * its documentation is hereby granted, provided that both the copyright
     10   1.1     oster  * notice and this permission notice appear in all copies of the
     11   1.1     oster  * software, derivative works or modified versions, and any portions
     12   1.1     oster  * thereof, and that both notices appear in supporting documentation.
     13   1.1     oster  *
     14   1.1     oster  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
     15   1.1     oster  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
     16   1.1     oster  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
     17   1.1     oster  *
     18   1.1     oster  * Carnegie Mellon requests users of this software to return to
     19   1.1     oster  *
     20   1.1     oster  *  Software Distribution Coordinator  or  Software.Distribution (at) CS.CMU.EDU
     21   1.1     oster  *  School of Computer Science
     22   1.1     oster  *  Carnegie Mellon University
     23   1.1     oster  *  Pittsburgh PA 15213-3890
     24   1.1     oster  *
     25   1.1     oster  * any improvements or extensions that they make and grant Carnegie the
     26   1.1     oster  * rights to redistribute these changes.
     27   1.1     oster  */
     28   1.1     oster 
     29   1.1     oster /*****************************************************************************
     30   1.1     oster  *
     31   1.1     oster  * rf_raid1.c -- implements RAID Level 1
     32   1.1     oster  *
     33   1.1     oster  *****************************************************************************/
     34   1.8     lukem 
     35   1.8     lukem #include <sys/cdefs.h>
     36  1.39     oster __KERNEL_RCSID(0, "$NetBSD: rf_raid1.c,v 1.39 2021/07/23 22:34:12 oster Exp $");
     37   1.1     oster 
     38   1.1     oster #include "rf_raid.h"
     39   1.1     oster #include "rf_raid1.h"
     40   1.1     oster #include "rf_dag.h"
     41   1.1     oster #include "rf_dagffrd.h"
     42   1.1     oster #include "rf_dagffwr.h"
     43   1.1     oster #include "rf_dagdegrd.h"
     44   1.1     oster #include "rf_dagutils.h"
     45   1.1     oster #include "rf_dagfuncs.h"
     46   1.1     oster #include "rf_diskqueue.h"
     47   1.1     oster #include "rf_general.h"
     48   1.1     oster #include "rf_utils.h"
     49   1.1     oster #include "rf_parityscan.h"
     50   1.1     oster #include "rf_mcpair.h"
     51   1.1     oster #include "rf_layout.h"
     52   1.1     oster #include "rf_map.h"
     53   1.1     oster #include "rf_engine.h"
     54   1.1     oster #include "rf_reconbuffer.h"
     55   1.1     oster 
     56   1.1     oster typedef struct RF_Raid1ConfigInfo_s {
     57   1.3     oster 	RF_RowCol_t **stripeIdentifier;
     58   1.3     oster }       RF_Raid1ConfigInfo_t;
     59   1.1     oster /* start of day code specific to RAID level 1 */
     60  1.25     perry int
     61  1.29  christos rf_ConfigureRAID1(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
     62  1.29  christos 		  RF_Config_t *cfgPtr)
     63   1.1     oster {
     64   1.3     oster 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
     65   1.3     oster 	RF_Raid1ConfigInfo_t *info;
     66   1.3     oster 	RF_RowCol_t i;
     67   1.3     oster 
     68  1.39     oster 	/* Sanity check the number of columns... */
     69  1.39     oster 	if (raidPtr->numCol < 2 || raidPtr->numCol % 2 != 0) {
     70  1.39     oster 		return (EINVAL);
     71  1.39     oster 	}
     72  1.39     oster 
     73   1.3     oster 	/* create a RAID level 1 configuration structure */
     74  1.36  christos 	info = RF_MallocAndAdd(sizeof(*info), raidPtr->cleanupList);
     75   1.3     oster 	if (info == NULL)
     76   1.3     oster 		return (ENOMEM);
     77   1.3     oster 	layoutPtr->layoutSpecificInfo = (void *) info;
     78   1.3     oster 
     79   1.3     oster 	/* ... and fill it in. */
     80   1.3     oster 	info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol / 2, 2, raidPtr->cleanupList);
     81   1.3     oster 	if (info->stripeIdentifier == NULL)
     82   1.3     oster 		return (ENOMEM);
     83   1.3     oster 	for (i = 0; i < (raidPtr->numCol / 2); i++) {
     84   1.3     oster 		info->stripeIdentifier[i][0] = (2 * i);
     85   1.3     oster 		info->stripeIdentifier[i][1] = (2 * i) + 1;
     86   1.3     oster 	}
     87   1.3     oster 
     88   1.3     oster 	/* this implementation of RAID level 1 uses one row of numCol disks
     89   1.3     oster 	 * and allows multiple (numCol / 2) stripes per row.  A stripe
     90   1.3     oster 	 * consists of a single data unit and a single parity (mirror) unit.
     91   1.3     oster 	 * stripe id = raidAddr / stripeUnitSize */
     92   1.3     oster 	raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
     93   1.3     oster 	layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
     94   1.3     oster 	layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
     95   1.3     oster 	layoutPtr->numDataCol = 1;
     96   1.3     oster 	layoutPtr->numParityCol = 1;
     97   1.3     oster 	return (0);
     98   1.1     oster }
     99   1.1     oster 
    100   1.1     oster 
    101   1.1     oster /* returns the physical disk location of the primary copy in the mirror pair */
    102  1.25     perry void
    103  1.15     oster rf_MapSectorRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
    104  1.28  christos 		  RF_RowCol_t *col, RF_SectorNum_t *diskSector,
    105  1.29  christos 		  int remap)
    106   1.1     oster {
    107   1.3     oster 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
    108   1.3     oster 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
    109   1.1     oster 
    110   1.3     oster 	*col = 2 * mirrorPair;
    111   1.3     oster 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    112   1.1     oster }
    113   1.1     oster 
    114   1.1     oster 
    115   1.1     oster /* Map Parity
    116   1.1     oster  *
    117   1.1     oster  * returns the physical disk location of the secondary copy in the mirror
    118   1.1     oster  * pair
    119   1.1     oster  */
    120  1.25     perry void
    121  1.15     oster rf_MapParityRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
    122  1.28  christos 		  RF_RowCol_t *col, RF_SectorNum_t *diskSector,
    123  1.29  christos 		  int remap)
    124   1.1     oster {
    125   1.3     oster 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
    126   1.3     oster 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
    127   1.1     oster 
    128   1.3     oster 	*col = (2 * mirrorPair) + 1;
    129   1.1     oster 
    130   1.3     oster 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    131   1.1     oster }
    132   1.1     oster 
    133   1.1     oster 
    134   1.1     oster /* IdentifyStripeRAID1
    135   1.1     oster  *
    136   1.1     oster  * returns a list of disks for a given redundancy group
    137   1.1     oster  */
    138  1.25     perry void
    139  1.15     oster rf_IdentifyStripeRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t addr,
    140  1.15     oster 		       RF_RowCol_t **diskids)
    141   1.1     oster {
    142   1.3     oster 	RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
    143   1.3     oster 	RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
    144   1.3     oster 	RF_ASSERT(stripeID >= 0);
    145   1.3     oster 	RF_ASSERT(addr >= 0);
    146   1.3     oster 	*diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
    147   1.3     oster 	RF_ASSERT(*diskids);
    148   1.1     oster }
    149   1.1     oster 
    150   1.1     oster 
    151   1.1     oster /* MapSIDToPSIDRAID1
    152   1.1     oster  *
    153   1.1     oster  * maps a logical stripe to a stripe in the redundant array
    154   1.1     oster  */
    155  1.25     perry void
    156  1.29  christos rf_MapSIDToPSIDRAID1(RF_RaidLayout_t *layoutPtr,
    157  1.28  christos 		     RF_StripeNum_t stripeID,
    158  1.15     oster 		     RF_StripeNum_t *psID, RF_ReconUnitNum_t *which_ru)
    159   1.1     oster {
    160   1.3     oster 	*which_ru = 0;
    161   1.3     oster 	*psID = stripeID;
    162   1.1     oster }
    163   1.1     oster 
    164   1.1     oster 
    165   1.1     oster 
    166   1.1     oster /******************************************************************************
    167   1.1     oster  * select a graph to perform a single-stripe access
    168   1.1     oster  *
    169   1.1     oster  * Parameters:  raidPtr    - description of the physical array
    170   1.1     oster  *              type       - type of operation (read or write) requested
    171   1.1     oster  *              asmap      - logical & physical addresses for this access
    172   1.1     oster  *              createFunc - name of function to use to create the graph
    173   1.1     oster  *****************************************************************************/
    174   1.1     oster 
    175  1.25     perry void
    176  1.15     oster rf_RAID1DagSelect(RF_Raid_t *raidPtr, RF_IoType_t type,
    177  1.15     oster 		  RF_AccessStripeMap_t *asmap, RF_VoidFuncPtr *createFunc)
    178   1.1     oster {
    179  1.35    martin 	RF_RowCol_t fcol, oc __unused;
    180   1.3     oster 	RF_PhysDiskAddr_t *failedPDA;
    181   1.5     oster 	int     prior_recon;
    182   1.3     oster 	RF_RowStatus_t rstat;
    183  1.35    martin 	RF_SectorNum_t oo __unused;
    184   1.3     oster 
    185   1.3     oster 
    186   1.3     oster 	RF_ASSERT(RF_IO_IS_R_OR_W(type));
    187   1.3     oster 
    188   1.3     oster 	if (asmap->numDataFailed + asmap->numParityFailed > 1) {
    189  1.23     oster #if RF_DEBUG_DAG
    190  1.25     perry 		if (rf_dagDebug)
    191  1.16     oster 			RF_ERRORMSG("Multiple disks failed in a single group!  Aborting I/O operation.\n");
    192  1.23     oster #endif
    193   1.3     oster 		*createFunc = NULL;
    194   1.3     oster 		return;
    195   1.3     oster 	}
    196   1.3     oster 	if (asmap->numDataFailed + asmap->numParityFailed) {
    197   1.3     oster 		/*
    198   1.3     oster 	         * We've got a fault. Re-map to spare space, iff applicable.
    199   1.3     oster 	         * Shouldn't the arch-independent code do this for us?
    200   1.3     oster 	         * Anyway, it turns out if we don't do this here, then when
    201   1.3     oster 	         * we're reconstructing, writes go only to the surviving
    202   1.3     oster 	         * original disk, and aren't reflected on the reconstructed
    203   1.3     oster 	         * spare. Oops. --jimz
    204   1.3     oster 	         */
    205   1.3     oster 		failedPDA = asmap->failedPDAs[0];
    206   1.3     oster 		fcol = failedPDA->col;
    207  1.14     oster 		rstat = raidPtr->status;
    208   1.3     oster 		prior_recon = (rstat == rf_rs_reconfigured) || (
    209   1.3     oster 		    (rstat == rf_rs_reconstructing) ?
    210  1.14     oster 		    rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, failedPDA->startSector) : 0
    211   1.3     oster 		    );
    212   1.3     oster 		if (prior_recon) {
    213   1.3     oster 			oc = fcol;
    214   1.3     oster 			oo = failedPDA->startSector;
    215   1.3     oster 			/*
    216   1.3     oster 		         * If we did distributed sparing, we'd monkey with that here.
    217   1.3     oster 		         * But we don't, so we'll
    218   1.3     oster 		         */
    219  1.14     oster 			failedPDA->col = raidPtr->Disks[fcol].spareCol;
    220   1.3     oster 			/*
    221   1.3     oster 		         * Redirect other components, iff necessary. This looks
    222   1.3     oster 		         * pretty suspicious to me, but it's what the raid5
    223   1.3     oster 		         * DAG select does.
    224   1.3     oster 		         */
    225   1.3     oster 			if (asmap->parityInfo->next) {
    226   1.3     oster 				if (failedPDA == asmap->parityInfo) {
    227   1.3     oster 					failedPDA->next->col = failedPDA->col;
    228   1.3     oster 				} else {
    229   1.3     oster 					if (failedPDA == asmap->parityInfo->next) {
    230   1.3     oster 						asmap->parityInfo->col = failedPDA->col;
    231   1.3     oster 					}
    232   1.3     oster 				}
    233   1.3     oster 			}
    234  1.23     oster #if RF_DEBUG_DAG > 0 || RF_DEBUG_MAP > 0
    235   1.3     oster 			if (rf_dagDebug || rf_mapDebug) {
    236  1.14     oster 				printf("raid%d: Redirected type '%c' c %d o %ld -> c %d o %ld\n",
    237  1.25     perry 				       raidPtr->raidid, type, oc,
    238  1.25     perry 				       (long) oo,
    239   1.5     oster 				       failedPDA->col,
    240   1.5     oster 				       (long) failedPDA->startSector);
    241   1.3     oster 			}
    242  1.23     oster #endif
    243   1.3     oster 			asmap->numDataFailed = asmap->numParityFailed = 0;
    244   1.3     oster 		}
    245   1.3     oster 	}
    246   1.3     oster 	if (type == RF_IO_TYPE_READ) {
    247   1.3     oster 		if (asmap->numDataFailed == 0)
    248   1.3     oster 			*createFunc = (RF_VoidFuncPtr) rf_CreateMirrorIdleReadDAG;
    249   1.3     oster 		else
    250   1.3     oster 			*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
    251   1.3     oster 	} else {
    252   1.3     oster 		*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
    253   1.3     oster 	}
    254   1.1     oster }
    255   1.1     oster 
    256  1.25     perry int
    257  1.15     oster rf_VerifyParityRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
    258  1.15     oster 		     RF_PhysDiskAddr_t *parityPDA, int correct_it,
    259  1.15     oster 		     RF_RaidAccessFlags_t flags)
    260   1.1     oster {
    261   1.5     oster 	int     nbytes, bcount, stripeWidth, ret, i, j, nbad, *bbufs;
    262  1.13     oster 	RF_DagNode_t *blockNode, *wrBlock;
    263   1.3     oster 	RF_DagHeader_t *rd_dag_h, *wr_dag_h;
    264   1.3     oster 	RF_AccessStripeMapHeader_t *asm_h;
    265   1.3     oster 	RF_AllocListElem_t *allocList;
    266  1.19     oster #if RF_ACC_TRACE > 0
    267   1.3     oster 	RF_AccTraceEntry_t tracerec;
    268  1.19     oster #endif
    269   1.3     oster 	RF_ReconUnitNum_t which_ru;
    270   1.3     oster 	RF_RaidLayout_t *layoutPtr;
    271   1.3     oster 	RF_AccessStripeMap_t *aasm;
    272   1.3     oster 	RF_SectorCount_t nsector;
    273   1.3     oster 	RF_RaidAddr_t startAddr;
    274  1.26  christos 	char   *bf, *buf1, *buf2;
    275   1.3     oster 	RF_PhysDiskAddr_t *pda;
    276   1.3     oster 	RF_StripeNum_t psID;
    277   1.3     oster 	RF_MCPair_t *mcpair;
    278   1.3     oster 
    279   1.3     oster 	layoutPtr = &raidPtr->Layout;
    280   1.3     oster 	startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
    281   1.3     oster 	nsector = parityPDA->numSector;
    282   1.3     oster 	nbytes = rf_RaidAddressToByte(raidPtr, nsector);
    283   1.3     oster 	psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
    284   1.3     oster 
    285   1.3     oster 	asm_h = NULL;
    286   1.3     oster 	rd_dag_h = wr_dag_h = NULL;
    287   1.3     oster 	mcpair = NULL;
    288   1.3     oster 
    289   1.3     oster 	ret = RF_PARITY_COULD_NOT_VERIFY;
    290   1.3     oster 
    291   1.3     oster 	rf_MakeAllocList(allocList);
    292   1.3     oster 	if (allocList == NULL)
    293   1.3     oster 		return (RF_PARITY_COULD_NOT_VERIFY);
    294  1.38     oster 	mcpair = rf_AllocMCPair(raidPtr);
    295   1.3     oster 	if (mcpair == NULL)
    296   1.3     oster 		goto done;
    297   1.3     oster 	RF_ASSERT(layoutPtr->numDataCol == layoutPtr->numParityCol);
    298   1.3     oster 	stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
    299   1.3     oster 	bcount = nbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol);
    300  1.36  christos 	bf = RF_MallocAndAdd(bcount, allocList);
    301  1.26  christos 	if (bf == NULL)
    302   1.3     oster 		goto done;
    303  1.10     oster #if RF_DEBUG_VERIFYPARITY
    304   1.3     oster 	if (rf_verifyParityDebug) {
    305   1.5     oster 		printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
    306  1.26  christos 		       raidPtr->raidid, (long) bf, bcount, (long) bf,
    307  1.26  christos 		       (long) bf + bcount);
    308   1.3     oster 	}
    309  1.10     oster #endif
    310   1.3     oster 	/*
    311   1.3     oster          * Generate a DAG which will read the entire stripe- then we can
    312   1.3     oster          * just compare data chunks versus "parity" chunks.
    313   1.3     oster          */
    314   1.3     oster 
    315  1.26  christos 	rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, bf,
    316   1.3     oster 	    rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
    317   1.3     oster 	    RF_IO_NORMAL_PRIORITY);
    318   1.3     oster 	if (rd_dag_h == NULL)
    319   1.3     oster 		goto done;
    320   1.3     oster 	blockNode = rd_dag_h->succedents[0];
    321   1.3     oster 
    322   1.3     oster 	/*
    323   1.3     oster          * Map the access to physical disk addresses (PDAs)- this will
    324   1.3     oster          * get us both a list of data addresses, and "parity" addresses
    325   1.3     oster          * (which are really mirror copies).
    326   1.3     oster          */
    327   1.3     oster 	asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
    328  1.26  christos 	    bf, RF_DONT_REMAP);
    329   1.3     oster 	aasm = asm_h->stripeMap;
    330   1.3     oster 
    331  1.26  christos 	buf1 = bf;
    332   1.3     oster 	/*
    333   1.3     oster          * Loop through the data blocks, setting up read nodes for each.
    334   1.3     oster          */
    335   1.3     oster 	for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
    336   1.3     oster 		RF_ASSERT(pda);
    337   1.3     oster 
    338   1.3     oster 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
    339   1.3     oster 
    340   1.3     oster 		RF_ASSERT(pda->numSector != 0);
    341   1.3     oster 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
    342   1.3     oster 			/* cannot verify parity with dead disk */
    343   1.3     oster 			goto done;
    344   1.3     oster 		}
    345   1.3     oster 		pda->bufPtr = buf1;
    346   1.3     oster 		blockNode->succedents[i]->params[0].p = pda;
    347   1.3     oster 		blockNode->succedents[i]->params[1].p = buf1;
    348   1.3     oster 		blockNode->succedents[i]->params[2].v = psID;
    349  1.17     oster 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    350   1.3     oster 		buf1 += nbytes;
    351   1.3     oster 	}
    352   1.3     oster 	RF_ASSERT(pda == NULL);
    353   1.3     oster 	/*
    354   1.3     oster          * keep i, buf1 running
    355   1.3     oster          *
    356   1.3     oster          * Loop through parity blocks, setting up read nodes for each.
    357   1.3     oster          */
    358   1.3     oster 	for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
    359   1.3     oster 		RF_ASSERT(pda);
    360   1.3     oster 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
    361   1.3     oster 		RF_ASSERT(pda->numSector != 0);
    362   1.3     oster 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
    363   1.3     oster 			/* cannot verify parity with dead disk */
    364   1.3     oster 			goto done;
    365   1.3     oster 		}
    366   1.3     oster 		pda->bufPtr = buf1;
    367   1.3     oster 		blockNode->succedents[i]->params[0].p = pda;
    368   1.3     oster 		blockNode->succedents[i]->params[1].p = buf1;
    369   1.3     oster 		blockNode->succedents[i]->params[2].v = psID;
    370  1.17     oster 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    371   1.3     oster 		buf1 += nbytes;
    372   1.3     oster 	}
    373   1.3     oster 	RF_ASSERT(pda == NULL);
    374   1.3     oster 
    375  1.19     oster #if RF_ACC_TRACE > 0
    376  1.36  christos 	memset(&tracerec, 0, sizeof(tracerec));
    377   1.3     oster 	rd_dag_h->tracerec = &tracerec;
    378  1.19     oster #endif
    379   1.9     oster #if 0
    380   1.3     oster 	if (rf_verifyParityDebug > 1) {
    381  1.25     perry 		printf("raid%d: RAID1 parity verify read dag:\n",
    382   1.5     oster 		       raidPtr->raidid);
    383   1.3     oster 		rf_PrintDAGList(rd_dag_h);
    384   1.3     oster 	}
    385   1.9     oster #endif
    386  1.33       mrg 	RF_LOCK_MCPAIR(mcpair);
    387   1.3     oster 	mcpair->flag = 0;
    388  1.33       mrg 	RF_UNLOCK_MCPAIR(mcpair);
    389  1.18     oster 
    390   1.3     oster 	rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
    391   1.3     oster 	    (void *) mcpair);
    392  1.18     oster 
    393  1.33       mrg 	RF_LOCK_MCPAIR(mcpair);
    394   1.3     oster 	while (mcpair->flag == 0) {
    395   1.3     oster 		RF_WAIT_MCPAIR(mcpair);
    396   1.3     oster 	}
    397  1.33       mrg 	RF_UNLOCK_MCPAIR(mcpair);
    398   1.3     oster 
    399   1.3     oster 	if (rd_dag_h->status != rf_enable) {
    400   1.3     oster 		RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
    401   1.3     oster 		ret = RF_PARITY_COULD_NOT_VERIFY;
    402   1.3     oster 		goto done;
    403   1.3     oster 	}
    404   1.3     oster 	/*
    405   1.3     oster          * buf1 is the beginning of the data blocks chunk
    406   1.3     oster          * buf2 is the beginning of the parity blocks chunk
    407   1.3     oster          */
    408  1.26  christos 	buf1 = bf;
    409  1.26  christos 	buf2 = bf + (nbytes * layoutPtr->numDataCol);
    410   1.3     oster 	ret = RF_PARITY_OKAY;
    411   1.3     oster 	/*
    412   1.3     oster          * bbufs is "bad bufs"- an array whose entries are the data
    413   1.3     oster          * column numbers where we had miscompares. (That is, column 0
    414   1.3     oster          * and column 1 of the array are mirror copies, and are considered
    415   1.3     oster          * "data column 0" for this purpose).
    416   1.3     oster          */
    417  1.36  christos 	bbufs = RF_MallocAndAdd(layoutPtr->numParityCol * sizeof(*bbufs),
    418   1.3     oster 	    allocList);
    419   1.3     oster 	nbad = 0;
    420   1.3     oster 	/*
    421   1.3     oster          * Check data vs "parity" (mirror copy).
    422   1.3     oster          */
    423   1.3     oster 	for (i = 0; i < layoutPtr->numDataCol; i++) {
    424  1.10     oster #if RF_DEBUG_VERIFYPARITY
    425   1.3     oster 		if (rf_verifyParityDebug) {
    426   1.5     oster 			printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
    427  1.25     perry 			       raidPtr->raidid, nbytes, i, (long) buf1,
    428  1.26  christos 			       (long) buf2, (long) bf);
    429   1.3     oster 		}
    430  1.10     oster #endif
    431   1.7   thorpej 		ret = memcmp(buf1, buf2, nbytes);
    432   1.3     oster 		if (ret) {
    433  1.10     oster #if RF_DEBUG_VERIFYPARITY
    434   1.3     oster 			if (rf_verifyParityDebug > 1) {
    435   1.3     oster 				for (j = 0; j < nbytes; j++) {
    436   1.3     oster 					if (buf1[j] != buf2[j])
    437   1.3     oster 						break;
    438   1.3     oster 				}
    439   1.3     oster 				printf("psid=%ld j=%d\n", (long) psID, j);
    440   1.3     oster 				printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff,
    441   1.3     oster 				    buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff);
    442   1.3     oster 				printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff,
    443   1.3     oster 				    buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff);
    444   1.3     oster 			}
    445   1.3     oster 			if (rf_verifyParityDebug) {
    446   1.5     oster 				printf("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, i);
    447   1.3     oster 			}
    448  1.10     oster #endif
    449   1.3     oster 			/*
    450   1.3     oster 		         * Parity is bad. Keep track of which columns were bad.
    451   1.3     oster 		         */
    452   1.3     oster 			if (bbufs)
    453   1.3     oster 				bbufs[nbad] = i;
    454   1.3     oster 			nbad++;
    455   1.3     oster 			ret = RF_PARITY_BAD;
    456   1.3     oster 		}
    457   1.3     oster 		buf1 += nbytes;
    458   1.3     oster 		buf2 += nbytes;
    459   1.3     oster 	}
    460   1.3     oster 
    461   1.3     oster 	if ((ret != RF_PARITY_OKAY) && correct_it) {
    462   1.3     oster 		ret = RF_PARITY_COULD_NOT_CORRECT;
    463  1.10     oster #if RF_DEBUG_VERIFYPARITY
    464   1.3     oster 		if (rf_verifyParityDebug) {
    465   1.5     oster 			printf("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid);
    466   1.3     oster 		}
    467  1.10     oster #endif
    468   1.3     oster 		if (bbufs == NULL)
    469   1.3     oster 			goto done;
    470   1.3     oster 		/*
    471   1.3     oster 	         * Make a DAG with one write node for each bad unit. We'll simply
    472   1.3     oster 	         * write the contents of the data unit onto the parity unit for
    473   1.3     oster 	         * correction. (It's possible that the mirror copy was the correct
    474   1.3     oster 	         * copy, and that we're spooging good data by writing bad over it,
    475   1.3     oster 	         * but there's no way we can know that.
    476   1.3     oster 	         */
    477  1.26  christos 		wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, bf,
    478   1.3     oster 		    rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags,
    479   1.3     oster 		    RF_IO_NORMAL_PRIORITY);
    480   1.3     oster 		if (wr_dag_h == NULL)
    481   1.3     oster 			goto done;
    482   1.3     oster 		wrBlock = wr_dag_h->succedents[0];
    483   1.3     oster 		/*
    484   1.3     oster 	         * Fill in a write node for each bad compare.
    485   1.3     oster 	         */
    486   1.3     oster 		for (i = 0; i < nbad; i++) {
    487   1.3     oster 			j = i + layoutPtr->numDataCol;
    488   1.3     oster 			pda = blockNode->succedents[j]->params[0].p;
    489   1.3     oster 			pda->bufPtr = blockNode->succedents[i]->params[1].p;
    490   1.3     oster 			wrBlock->succedents[i]->params[0].p = pda;
    491   1.3     oster 			wrBlock->succedents[i]->params[1].p = pda->bufPtr;
    492   1.3     oster 			wrBlock->succedents[i]->params[2].v = psID;
    493  1.32     oster 			wrBlock->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    494   1.3     oster 		}
    495  1.19     oster #if RF_ACC_TRACE > 0
    496  1.36  christos 		memset(&tracerec, 0, sizeof(tracerec));
    497   1.3     oster 		wr_dag_h->tracerec = &tracerec;
    498  1.19     oster #endif
    499   1.9     oster #if 0
    500   1.3     oster 		if (rf_verifyParityDebug > 1) {
    501   1.3     oster 			printf("Parity verify write dag:\n");
    502   1.3     oster 			rf_PrintDAGList(wr_dag_h);
    503   1.3     oster 		}
    504   1.9     oster #endif
    505  1.33       mrg 		RF_LOCK_MCPAIR(mcpair);
    506   1.3     oster 		mcpair->flag = 0;
    507  1.33       mrg 		RF_UNLOCK_MCPAIR(mcpair);
    508  1.20     oster 
    509   1.3     oster 		/* fire off the write DAG */
    510   1.3     oster 		rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
    511   1.3     oster 		    (void *) mcpair);
    512  1.20     oster 
    513  1.33       mrg 		RF_LOCK_MCPAIR(mcpair);
    514   1.3     oster 		while (!mcpair->flag) {
    515  1.33       mrg 			RF_WAIT_MCPAIR(mcpair);
    516   1.3     oster 		}
    517  1.33       mrg 		RF_UNLOCK_MCPAIR(mcpair);
    518   1.3     oster 		if (wr_dag_h->status != rf_enable) {
    519   1.3     oster 			RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n");
    520   1.3     oster 			goto done;
    521   1.3     oster 		}
    522   1.3     oster 		ret = RF_PARITY_CORRECTED;
    523   1.3     oster 	}
    524   1.1     oster done:
    525   1.3     oster 	/*
    526   1.3     oster          * All done. We might've gotten here without doing part of the function,
    527   1.3     oster          * so cleanup what we have to and return our running status.
    528   1.3     oster          */
    529   1.3     oster 	if (asm_h)
    530  1.38     oster 		rf_FreeAccessStripeMap(raidPtr, asm_h);
    531   1.3     oster 	if (rd_dag_h)
    532   1.3     oster 		rf_FreeDAG(rd_dag_h);
    533   1.3     oster 	if (wr_dag_h)
    534   1.3     oster 		rf_FreeDAG(wr_dag_h);
    535   1.3     oster 	if (mcpair)
    536  1.38     oster 		rf_FreeMCPair(raidPtr, mcpair);
    537   1.3     oster 	rf_FreeAllocList(allocList);
    538  1.10     oster #if RF_DEBUG_VERIFYPARITY
    539   1.3     oster 	if (rf_verifyParityDebug) {
    540  1.25     perry 		printf("raid%d: RAID1 parity verify, returning %d\n",
    541   1.5     oster 		       raidPtr->raidid, ret);
    542   1.3     oster 	}
    543  1.10     oster #endif
    544   1.3     oster 	return (ret);
    545   1.1     oster }
    546   1.1     oster 
    547  1.25     perry /* rbuf          - the recon buffer to submit
    548  1.15     oster  * keep_it       - whether we can keep this buffer or we have to return it
    549  1.15     oster  * use_committed - whether to use a committed or an available recon buffer
    550  1.15     oster  */
    551  1.15     oster 
    552  1.25     perry int
    553  1.15     oster rf_SubmitReconBufferRAID1(RF_ReconBuffer_t *rbuf, int keep_it,
    554  1.15     oster 			  int use_committed)
    555   1.1     oster {
    556   1.3     oster 	RF_ReconParityStripeStatus_t *pssPtr;
    557   1.3     oster 	RF_ReconCtrl_t *reconCtrlPtr;
    558  1.21     oster 	int     retcode;
    559  1.37  christos 	RF_CallbackValueDesc_t *cb, *p;
    560   1.3     oster 	RF_ReconBuffer_t *t;
    561   1.3     oster 	RF_Raid_t *raidPtr;
    562  1.30  christos 	void *ta;
    563   1.3     oster 
    564   1.3     oster 	retcode = 0;
    565   1.3     oster 
    566   1.3     oster 	raidPtr = rbuf->raidPtr;
    567  1.14     oster 	reconCtrlPtr = raidPtr->reconControl;
    568   1.3     oster 
    569   1.3     oster 	RF_ASSERT(rbuf);
    570   1.3     oster 	RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
    571   1.3     oster 
    572  1.11     oster #if RF_DEBUG_RECON
    573   1.3     oster 	if (rf_reconbufferDebug) {
    574  1.14     oster 		printf("raid%d: RAID1 reconbuffer submission c%d psid %ld ru%d (failed offset %ld)\n",
    575  1.25     perry 		       raidPtr->raidid, rbuf->col,
    576   1.5     oster 		       (long) rbuf->parityStripeID, rbuf->which_ru,
    577   1.5     oster 		       (long) rbuf->failedDiskSectorOffset);
    578   1.3     oster 	}
    579  1.10     oster #endif
    580   1.3     oster 	if (rf_reconDebug) {
    581  1.30  christos 		unsigned char *b = rbuf->buffer;
    582   1.3     oster 		printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
    583   1.3     oster 		    (long) rbuf->parityStripeID, (long) rbuf->buffer);
    584   1.3     oster 		printf("RAID1 psid %ld   %02x %02x %02x %02x %02x\n",
    585  1.30  christos 		    (long)rbuf->parityStripeID, b[0], b[1], b[2], b[3], b[4]);
    586   1.3     oster 	}
    587  1.14     oster 	RF_LOCK_PSS_MUTEX(raidPtr, rbuf->parityStripeID);
    588   1.3     oster 
    589  1.34       mrg 	rf_lock_mutex2(reconCtrlPtr->rb_mutex);
    590  1.24     oster 	while(reconCtrlPtr->rb_lock) {
    591  1.34       mrg 		rf_wait_cond2(reconCtrlPtr->rb_cv, reconCtrlPtr->rb_mutex);
    592  1.24     oster 	}
    593  1.24     oster 	reconCtrlPtr->rb_lock = 1;
    594  1.34       mrg 	rf_unlock_mutex2(reconCtrlPtr->rb_mutex);
    595   1.3     oster 
    596   1.3     oster 	pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
    597  1.21     oster 	    rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, NULL);
    598   1.3     oster 	RF_ASSERT(pssPtr);	/* if it didn't exist, we wouldn't have gotten
    599   1.3     oster 				 * an rbuf for it */
    600   1.3     oster 
    601   1.3     oster 	/*
    602   1.3     oster          * Since this is simple mirroring, the first submission for a stripe is also
    603   1.3     oster          * treated as the last.
    604   1.3     oster          */
    605   1.3     oster 
    606   1.3     oster 	t = NULL;
    607   1.3     oster 	if (keep_it) {
    608  1.11     oster #if RF_DEBUG_RECON
    609   1.3     oster 		if (rf_reconbufferDebug) {
    610  1.25     perry 			printf("raid%d: RAID1 rbuf submission: keeping rbuf\n",
    611   1.5     oster 			       raidPtr->raidid);
    612   1.3     oster 		}
    613  1.10     oster #endif
    614   1.3     oster 		t = rbuf;
    615   1.3     oster 	} else {
    616   1.3     oster 		if (use_committed) {
    617  1.11     oster #if RF_DEBUG_RECON
    618   1.3     oster 			if (rf_reconbufferDebug) {
    619   1.5     oster 				printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid);
    620   1.3     oster 			}
    621  1.10     oster #endif
    622   1.3     oster 			t = reconCtrlPtr->committedRbufs;
    623   1.3     oster 			RF_ASSERT(t);
    624   1.3     oster 			reconCtrlPtr->committedRbufs = t->next;
    625   1.3     oster 			t->next = NULL;
    626   1.3     oster 		} else
    627   1.3     oster 			if (reconCtrlPtr->floatingRbufs) {
    628  1.11     oster #if RF_DEBUG_RECON
    629   1.3     oster 				if (rf_reconbufferDebug) {
    630   1.5     oster 					printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid);
    631   1.3     oster 				}
    632  1.10     oster #endif
    633   1.3     oster 				t = reconCtrlPtr->floatingRbufs;
    634   1.3     oster 				reconCtrlPtr->floatingRbufs = t->next;
    635   1.3     oster 				t->next = NULL;
    636   1.3     oster 			}
    637   1.3     oster 	}
    638   1.3     oster 	if (t == NULL) {
    639  1.11     oster #if RF_DEBUG_RECON
    640   1.3     oster 		if (rf_reconbufferDebug) {
    641   1.5     oster 			printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid);
    642   1.3     oster 		}
    643  1.10     oster #endif
    644   1.3     oster 		RF_ASSERT((keep_it == 0) && (use_committed == 0));
    645   1.3     oster 		raidPtr->procsInBufWait++;
    646   1.3     oster 		if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
    647   1.3     oster 		    && (raidPtr->numFullReconBuffers == 0)) {
    648   1.3     oster 			/* ruh-ro */
    649   1.3     oster 			RF_ERRORMSG("Buffer wait deadlock\n");
    650  1.14     oster 			rf_PrintPSStatusTable(raidPtr);
    651   1.3     oster 			RF_PANIC();
    652   1.3     oster 		}
    653   1.3     oster 		pssPtr->flags |= RF_PSS_BUFFERWAIT;
    654  1.38     oster 		cb = rf_AllocCallbackValueDesc(raidPtr);
    655   1.3     oster 		cb->col = rbuf->col;
    656  1.37  christos 		cb->v = rbuf->parityStripeID;
    657   1.3     oster 		cb->next = NULL;
    658   1.3     oster 		if (reconCtrlPtr->bufferWaitList == NULL) {
    659   1.3     oster 			/* we are the wait list- lucky us */
    660   1.3     oster 			reconCtrlPtr->bufferWaitList = cb;
    661   1.3     oster 		} else {
    662   1.3     oster 			/* append to wait list */
    663   1.3     oster 			for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
    664   1.3     oster 			p->next = cb;
    665   1.3     oster 		}
    666   1.3     oster 		retcode = 1;
    667   1.3     oster 		goto out;
    668   1.3     oster 	}
    669   1.3     oster 	if (t != rbuf) {
    670   1.3     oster 		t->col = reconCtrlPtr->fcol;
    671   1.3     oster 		t->parityStripeID = rbuf->parityStripeID;
    672   1.3     oster 		t->which_ru = rbuf->which_ru;
    673   1.3     oster 		t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
    674   1.3     oster 		t->spCol = rbuf->spCol;
    675   1.3     oster 		t->spOffset = rbuf->spOffset;
    676   1.3     oster 		/* Swap buffers. DANCE! */
    677   1.3     oster 		ta = t->buffer;
    678   1.3     oster 		t->buffer = rbuf->buffer;
    679   1.3     oster 		rbuf->buffer = ta;
    680   1.3     oster 	}
    681   1.3     oster 	/*
    682   1.3     oster          * Use the rbuf we've been given as the target.
    683   1.3     oster          */
    684   1.3     oster 	RF_ASSERT(pssPtr->rbuf == NULL);
    685   1.3     oster 	pssPtr->rbuf = t;
    686   1.3     oster 
    687   1.3     oster 	t->count = 1;
    688   1.3     oster 	/*
    689   1.3     oster          * Below, we use 1 for numDataCol (which is equal to the count in the
    690   1.3     oster          * previous line), so we'll always be done.
    691   1.3     oster          */
    692   1.3     oster 	rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
    693   1.1     oster 
    694   1.1     oster out:
    695  1.14     oster 	RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->parityStripeID);
    696  1.34       mrg 	rf_lock_mutex2(reconCtrlPtr->rb_mutex);
    697  1.24     oster 	reconCtrlPtr->rb_lock = 0;
    698  1.34       mrg 	rf_broadcast_cond2(reconCtrlPtr->rb_cv);
    699  1.34       mrg 	rf_unlock_mutex2(reconCtrlPtr->rb_mutex);
    700  1.11     oster #if RF_DEBUG_RECON
    701   1.3     oster 	if (rf_reconbufferDebug) {
    702  1.25     perry 		printf("raid%d: RAID1 rbuf submission: returning %d\n",
    703   1.5     oster 		       raidPtr->raidid, retcode);
    704   1.3     oster 	}
    705  1.10     oster #endif
    706   1.3     oster 	return (retcode);
    707   1.1     oster }
    708  1.31     oster 
    709  1.31     oster RF_HeadSepLimit_t
    710  1.31     oster rf_GetDefaultHeadSepLimitRAID1(RF_Raid_t *raidPtr)
    711  1.31     oster {
    712  1.31     oster 	return (10);
    713  1.31     oster }
    714  1.31     oster 
    715