Home | History | Annotate | Line # | Download | only in raidframe
rf_raid1.c revision 1.31.24.1
      1  1.31.24.1     rmind /*	$NetBSD: rf_raid1.c,v 1.31.24.1 2010/05/30 05:17:41 rmind 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  1.31.24.1     rmind __KERNEL_RCSID(0, "$NetBSD: rf_raid1.c,v 1.31.24.1 2010/05/30 05:17:41 rmind 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        1.3     oster 	/* create a RAID level 1 configuration structure */
     69        1.3     oster 	RF_MallocAndAdd(info, sizeof(RF_Raid1ConfigInfo_t), (RF_Raid1ConfigInfo_t *), raidPtr->cleanupList);
     70        1.3     oster 	if (info == NULL)
     71        1.3     oster 		return (ENOMEM);
     72        1.3     oster 	layoutPtr->layoutSpecificInfo = (void *) info;
     73        1.3     oster 
     74        1.3     oster 	/* ... and fill it in. */
     75        1.3     oster 	info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol / 2, 2, raidPtr->cleanupList);
     76        1.3     oster 	if (info->stripeIdentifier == NULL)
     77        1.3     oster 		return (ENOMEM);
     78        1.3     oster 	for (i = 0; i < (raidPtr->numCol / 2); i++) {
     79        1.3     oster 		info->stripeIdentifier[i][0] = (2 * i);
     80        1.3     oster 		info->stripeIdentifier[i][1] = (2 * i) + 1;
     81        1.3     oster 	}
     82        1.3     oster 
     83        1.3     oster 	/* this implementation of RAID level 1 uses one row of numCol disks
     84        1.3     oster 	 * and allows multiple (numCol / 2) stripes per row.  A stripe
     85        1.3     oster 	 * consists of a single data unit and a single parity (mirror) unit.
     86        1.3     oster 	 * stripe id = raidAddr / stripeUnitSize */
     87        1.3     oster 	raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
     88        1.3     oster 	layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
     89        1.3     oster 	layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
     90        1.3     oster 	layoutPtr->numDataCol = 1;
     91        1.3     oster 	layoutPtr->numParityCol = 1;
     92        1.3     oster 	return (0);
     93        1.1     oster }
     94        1.1     oster 
     95        1.1     oster 
     96        1.1     oster /* returns the physical disk location of the primary copy in the mirror pair */
     97       1.25     perry void
     98       1.15     oster rf_MapSectorRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
     99       1.28  christos 		  RF_RowCol_t *col, RF_SectorNum_t *diskSector,
    100       1.29  christos 		  int remap)
    101        1.1     oster {
    102        1.3     oster 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
    103        1.3     oster 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
    104        1.1     oster 
    105        1.3     oster 	*col = 2 * mirrorPair;
    106        1.3     oster 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    107        1.1     oster }
    108        1.1     oster 
    109        1.1     oster 
    110        1.1     oster /* Map Parity
    111        1.1     oster  *
    112        1.1     oster  * returns the physical disk location of the secondary copy in the mirror
    113        1.1     oster  * pair
    114        1.1     oster  */
    115       1.25     perry void
    116       1.15     oster rf_MapParityRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
    117       1.28  christos 		  RF_RowCol_t *col, RF_SectorNum_t *diskSector,
    118       1.29  christos 		  int remap)
    119        1.1     oster {
    120        1.3     oster 	RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
    121        1.3     oster 	RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
    122        1.1     oster 
    123        1.3     oster 	*col = (2 * mirrorPair) + 1;
    124        1.1     oster 
    125        1.3     oster 	*diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
    126        1.1     oster }
    127        1.1     oster 
    128        1.1     oster 
    129        1.1     oster /* IdentifyStripeRAID1
    130        1.1     oster  *
    131        1.1     oster  * returns a list of disks for a given redundancy group
    132        1.1     oster  */
    133       1.25     perry void
    134       1.15     oster rf_IdentifyStripeRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t addr,
    135       1.15     oster 		       RF_RowCol_t **diskids)
    136        1.1     oster {
    137        1.3     oster 	RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
    138        1.3     oster 	RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
    139        1.3     oster 	RF_ASSERT(stripeID >= 0);
    140        1.3     oster 	RF_ASSERT(addr >= 0);
    141        1.3     oster 	*diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
    142        1.3     oster 	RF_ASSERT(*diskids);
    143        1.1     oster }
    144        1.1     oster 
    145        1.1     oster 
    146        1.1     oster /* MapSIDToPSIDRAID1
    147        1.1     oster  *
    148        1.1     oster  * maps a logical stripe to a stripe in the redundant array
    149        1.1     oster  */
    150       1.25     perry void
    151       1.29  christos rf_MapSIDToPSIDRAID1(RF_RaidLayout_t *layoutPtr,
    152       1.28  christos 		     RF_StripeNum_t stripeID,
    153       1.15     oster 		     RF_StripeNum_t *psID, RF_ReconUnitNum_t *which_ru)
    154        1.1     oster {
    155        1.3     oster 	*which_ru = 0;
    156        1.3     oster 	*psID = stripeID;
    157        1.1     oster }
    158        1.1     oster 
    159        1.1     oster 
    160        1.1     oster 
    161        1.1     oster /******************************************************************************
    162        1.1     oster  * select a graph to perform a single-stripe access
    163        1.1     oster  *
    164        1.1     oster  * Parameters:  raidPtr    - description of the physical array
    165        1.1     oster  *              type       - type of operation (read or write) requested
    166        1.1     oster  *              asmap      - logical & physical addresses for this access
    167        1.1     oster  *              createFunc - name of function to use to create the graph
    168        1.1     oster  *****************************************************************************/
    169        1.1     oster 
    170       1.25     perry void
    171       1.15     oster rf_RAID1DagSelect(RF_Raid_t *raidPtr, RF_IoType_t type,
    172       1.15     oster 		  RF_AccessStripeMap_t *asmap, RF_VoidFuncPtr *createFunc)
    173        1.1     oster {
    174       1.14     oster 	RF_RowCol_t fcol, oc;
    175        1.3     oster 	RF_PhysDiskAddr_t *failedPDA;
    176        1.5     oster 	int     prior_recon;
    177        1.3     oster 	RF_RowStatus_t rstat;
    178        1.3     oster 	RF_SectorNum_t oo;
    179        1.3     oster 
    180        1.3     oster 
    181        1.3     oster 	RF_ASSERT(RF_IO_IS_R_OR_W(type));
    182        1.3     oster 
    183        1.3     oster 	if (asmap->numDataFailed + asmap->numParityFailed > 1) {
    184       1.23     oster #if RF_DEBUG_DAG
    185       1.25     perry 		if (rf_dagDebug)
    186       1.16     oster 			RF_ERRORMSG("Multiple disks failed in a single group!  Aborting I/O operation.\n");
    187       1.23     oster #endif
    188        1.3     oster 		*createFunc = NULL;
    189        1.3     oster 		return;
    190        1.3     oster 	}
    191        1.3     oster 	if (asmap->numDataFailed + asmap->numParityFailed) {
    192        1.3     oster 		/*
    193        1.3     oster 	         * We've got a fault. Re-map to spare space, iff applicable.
    194        1.3     oster 	         * Shouldn't the arch-independent code do this for us?
    195        1.3     oster 	         * Anyway, it turns out if we don't do this here, then when
    196        1.3     oster 	         * we're reconstructing, writes go only to the surviving
    197        1.3     oster 	         * original disk, and aren't reflected on the reconstructed
    198        1.3     oster 	         * spare. Oops. --jimz
    199        1.3     oster 	         */
    200        1.3     oster 		failedPDA = asmap->failedPDAs[0];
    201        1.3     oster 		fcol = failedPDA->col;
    202       1.14     oster 		rstat = raidPtr->status;
    203        1.3     oster 		prior_recon = (rstat == rf_rs_reconfigured) || (
    204        1.3     oster 		    (rstat == rf_rs_reconstructing) ?
    205       1.14     oster 		    rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, failedPDA->startSector) : 0
    206        1.3     oster 		    );
    207        1.3     oster 		if (prior_recon) {
    208        1.3     oster 			oc = fcol;
    209        1.3     oster 			oo = failedPDA->startSector;
    210        1.3     oster 			/*
    211        1.3     oster 		         * If we did distributed sparing, we'd monkey with that here.
    212        1.3     oster 		         * But we don't, so we'll
    213        1.3     oster 		         */
    214       1.14     oster 			failedPDA->col = raidPtr->Disks[fcol].spareCol;
    215        1.3     oster 			/*
    216        1.3     oster 		         * Redirect other components, iff necessary. This looks
    217        1.3     oster 		         * pretty suspicious to me, but it's what the raid5
    218        1.3     oster 		         * DAG select does.
    219        1.3     oster 		         */
    220        1.3     oster 			if (asmap->parityInfo->next) {
    221        1.3     oster 				if (failedPDA == asmap->parityInfo) {
    222        1.3     oster 					failedPDA->next->col = failedPDA->col;
    223        1.3     oster 				} else {
    224        1.3     oster 					if (failedPDA == asmap->parityInfo->next) {
    225        1.3     oster 						asmap->parityInfo->col = failedPDA->col;
    226        1.3     oster 					}
    227        1.3     oster 				}
    228        1.3     oster 			}
    229       1.23     oster #if RF_DEBUG_DAG > 0 || RF_DEBUG_MAP > 0
    230        1.3     oster 			if (rf_dagDebug || rf_mapDebug) {
    231       1.14     oster 				printf("raid%d: Redirected type '%c' c %d o %ld -> c %d o %ld\n",
    232       1.25     perry 				       raidPtr->raidid, type, oc,
    233       1.25     perry 				       (long) oo,
    234        1.5     oster 				       failedPDA->col,
    235        1.5     oster 				       (long) failedPDA->startSector);
    236        1.3     oster 			}
    237       1.23     oster #endif
    238        1.3     oster 			asmap->numDataFailed = asmap->numParityFailed = 0;
    239        1.3     oster 		}
    240        1.3     oster 	}
    241        1.3     oster 	if (type == RF_IO_TYPE_READ) {
    242        1.3     oster 		if (asmap->numDataFailed == 0)
    243        1.3     oster 			*createFunc = (RF_VoidFuncPtr) rf_CreateMirrorIdleReadDAG;
    244        1.3     oster 		else
    245        1.3     oster 			*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
    246        1.3     oster 	} else {
    247        1.3     oster 		*createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
    248        1.3     oster 	}
    249        1.1     oster }
    250        1.1     oster 
    251       1.25     perry int
    252       1.15     oster rf_VerifyParityRAID1(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
    253       1.15     oster 		     RF_PhysDiskAddr_t *parityPDA, int correct_it,
    254       1.15     oster 		     RF_RaidAccessFlags_t flags)
    255        1.1     oster {
    256        1.5     oster 	int     nbytes, bcount, stripeWidth, ret, i, j, nbad, *bbufs;
    257       1.13     oster 	RF_DagNode_t *blockNode, *wrBlock;
    258        1.3     oster 	RF_DagHeader_t *rd_dag_h, *wr_dag_h;
    259        1.3     oster 	RF_AccessStripeMapHeader_t *asm_h;
    260        1.3     oster 	RF_AllocListElem_t *allocList;
    261       1.19     oster #if RF_ACC_TRACE > 0
    262        1.3     oster 	RF_AccTraceEntry_t tracerec;
    263       1.19     oster #endif
    264        1.3     oster 	RF_ReconUnitNum_t which_ru;
    265        1.3     oster 	RF_RaidLayout_t *layoutPtr;
    266        1.3     oster 	RF_AccessStripeMap_t *aasm;
    267        1.3     oster 	RF_SectorCount_t nsector;
    268        1.3     oster 	RF_RaidAddr_t startAddr;
    269       1.26  christos 	char   *bf, *buf1, *buf2;
    270        1.3     oster 	RF_PhysDiskAddr_t *pda;
    271        1.3     oster 	RF_StripeNum_t psID;
    272        1.3     oster 	RF_MCPair_t *mcpair;
    273        1.3     oster 
    274        1.3     oster 	layoutPtr = &raidPtr->Layout;
    275        1.3     oster 	startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
    276        1.3     oster 	nsector = parityPDA->numSector;
    277        1.3     oster 	nbytes = rf_RaidAddressToByte(raidPtr, nsector);
    278        1.3     oster 	psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
    279        1.3     oster 
    280        1.3     oster 	asm_h = NULL;
    281        1.3     oster 	rd_dag_h = wr_dag_h = NULL;
    282        1.3     oster 	mcpair = NULL;
    283        1.3     oster 
    284        1.3     oster 	ret = RF_PARITY_COULD_NOT_VERIFY;
    285        1.3     oster 
    286        1.3     oster 	rf_MakeAllocList(allocList);
    287        1.3     oster 	if (allocList == NULL)
    288        1.3     oster 		return (RF_PARITY_COULD_NOT_VERIFY);
    289        1.3     oster 	mcpair = rf_AllocMCPair();
    290        1.3     oster 	if (mcpair == NULL)
    291        1.3     oster 		goto done;
    292        1.3     oster 	RF_ASSERT(layoutPtr->numDataCol == layoutPtr->numParityCol);
    293        1.3     oster 	stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
    294        1.3     oster 	bcount = nbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol);
    295       1.26  christos 	RF_MallocAndAdd(bf, bcount, (char *), allocList);
    296       1.26  christos 	if (bf == NULL)
    297        1.3     oster 		goto done;
    298       1.10     oster #if RF_DEBUG_VERIFYPARITY
    299        1.3     oster 	if (rf_verifyParityDebug) {
    300        1.5     oster 		printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
    301       1.26  christos 		       raidPtr->raidid, (long) bf, bcount, (long) bf,
    302       1.26  christos 		       (long) bf + bcount);
    303        1.3     oster 	}
    304       1.10     oster #endif
    305        1.3     oster 	/*
    306        1.3     oster          * Generate a DAG which will read the entire stripe- then we can
    307        1.3     oster          * just compare data chunks versus "parity" chunks.
    308        1.3     oster          */
    309        1.3     oster 
    310       1.26  christos 	rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, bf,
    311        1.3     oster 	    rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
    312        1.3     oster 	    RF_IO_NORMAL_PRIORITY);
    313        1.3     oster 	if (rd_dag_h == NULL)
    314        1.3     oster 		goto done;
    315        1.3     oster 	blockNode = rd_dag_h->succedents[0];
    316        1.3     oster 
    317        1.3     oster 	/*
    318        1.3     oster          * Map the access to physical disk addresses (PDAs)- this will
    319        1.3     oster          * get us both a list of data addresses, and "parity" addresses
    320        1.3     oster          * (which are really mirror copies).
    321        1.3     oster          */
    322        1.3     oster 	asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
    323       1.26  christos 	    bf, RF_DONT_REMAP);
    324        1.3     oster 	aasm = asm_h->stripeMap;
    325        1.3     oster 
    326       1.26  christos 	buf1 = bf;
    327        1.3     oster 	/*
    328        1.3     oster          * Loop through the data blocks, setting up read nodes for each.
    329        1.3     oster          */
    330        1.3     oster 	for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
    331        1.3     oster 		RF_ASSERT(pda);
    332        1.3     oster 
    333        1.3     oster 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
    334        1.3     oster 
    335        1.3     oster 		RF_ASSERT(pda->numSector != 0);
    336        1.3     oster 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
    337        1.3     oster 			/* cannot verify parity with dead disk */
    338        1.3     oster 			goto done;
    339        1.3     oster 		}
    340        1.3     oster 		pda->bufPtr = buf1;
    341        1.3     oster 		blockNode->succedents[i]->params[0].p = pda;
    342        1.3     oster 		blockNode->succedents[i]->params[1].p = buf1;
    343        1.3     oster 		blockNode->succedents[i]->params[2].v = psID;
    344       1.17     oster 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    345        1.3     oster 		buf1 += nbytes;
    346        1.3     oster 	}
    347        1.3     oster 	RF_ASSERT(pda == NULL);
    348        1.3     oster 	/*
    349        1.3     oster          * keep i, buf1 running
    350        1.3     oster          *
    351        1.3     oster          * Loop through parity blocks, setting up read nodes for each.
    352        1.3     oster          */
    353        1.3     oster 	for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
    354        1.3     oster 		RF_ASSERT(pda);
    355        1.3     oster 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
    356        1.3     oster 		RF_ASSERT(pda->numSector != 0);
    357        1.3     oster 		if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
    358        1.3     oster 			/* cannot verify parity with dead disk */
    359        1.3     oster 			goto done;
    360        1.3     oster 		}
    361        1.3     oster 		pda->bufPtr = buf1;
    362        1.3     oster 		blockNode->succedents[i]->params[0].p = pda;
    363        1.3     oster 		blockNode->succedents[i]->params[1].p = buf1;
    364        1.3     oster 		blockNode->succedents[i]->params[2].v = psID;
    365       1.17     oster 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    366        1.3     oster 		buf1 += nbytes;
    367        1.3     oster 	}
    368        1.3     oster 	RF_ASSERT(pda == NULL);
    369        1.3     oster 
    370       1.19     oster #if RF_ACC_TRACE > 0
    371        1.6   thorpej 	memset((char *) &tracerec, 0, sizeof(tracerec));
    372        1.3     oster 	rd_dag_h->tracerec = &tracerec;
    373       1.19     oster #endif
    374        1.9     oster #if 0
    375        1.3     oster 	if (rf_verifyParityDebug > 1) {
    376       1.25     perry 		printf("raid%d: RAID1 parity verify read dag:\n",
    377        1.5     oster 		       raidPtr->raidid);
    378        1.3     oster 		rf_PrintDAGList(rd_dag_h);
    379        1.3     oster 	}
    380        1.9     oster #endif
    381        1.3     oster 	RF_LOCK_MUTEX(mcpair->mutex);
    382        1.3     oster 	mcpair->flag = 0;
    383       1.18     oster 	RF_UNLOCK_MUTEX(mcpair->mutex);
    384       1.18     oster 
    385        1.3     oster 	rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
    386        1.3     oster 	    (void *) mcpair);
    387       1.18     oster 
    388       1.18     oster 	RF_LOCK_MUTEX(mcpair->mutex);
    389        1.3     oster 	while (mcpair->flag == 0) {
    390        1.3     oster 		RF_WAIT_MCPAIR(mcpair);
    391        1.3     oster 	}
    392        1.3     oster 	RF_UNLOCK_MUTEX(mcpair->mutex);
    393        1.3     oster 
    394        1.3     oster 	if (rd_dag_h->status != rf_enable) {
    395        1.3     oster 		RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
    396        1.3     oster 		ret = RF_PARITY_COULD_NOT_VERIFY;
    397        1.3     oster 		goto done;
    398        1.3     oster 	}
    399        1.3     oster 	/*
    400        1.3     oster          * buf1 is the beginning of the data blocks chunk
    401        1.3     oster          * buf2 is the beginning of the parity blocks chunk
    402        1.3     oster          */
    403       1.26  christos 	buf1 = bf;
    404       1.26  christos 	buf2 = bf + (nbytes * layoutPtr->numDataCol);
    405        1.3     oster 	ret = RF_PARITY_OKAY;
    406        1.3     oster 	/*
    407        1.3     oster          * bbufs is "bad bufs"- an array whose entries are the data
    408        1.3     oster          * column numbers where we had miscompares. (That is, column 0
    409        1.3     oster          * and column 1 of the array are mirror copies, and are considered
    410        1.3     oster          * "data column 0" for this purpose).
    411        1.3     oster          */
    412        1.3     oster 	RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *),
    413        1.3     oster 	    allocList);
    414        1.3     oster 	nbad = 0;
    415        1.3     oster 	/*
    416        1.3     oster          * Check data vs "parity" (mirror copy).
    417        1.3     oster          */
    418        1.3     oster 	for (i = 0; i < layoutPtr->numDataCol; i++) {
    419       1.10     oster #if RF_DEBUG_VERIFYPARITY
    420        1.3     oster 		if (rf_verifyParityDebug) {
    421        1.5     oster 			printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
    422       1.25     perry 			       raidPtr->raidid, nbytes, i, (long) buf1,
    423       1.26  christos 			       (long) buf2, (long) bf);
    424        1.3     oster 		}
    425       1.10     oster #endif
    426        1.7   thorpej 		ret = memcmp(buf1, buf2, nbytes);
    427        1.3     oster 		if (ret) {
    428       1.10     oster #if RF_DEBUG_VERIFYPARITY
    429        1.3     oster 			if (rf_verifyParityDebug > 1) {
    430        1.3     oster 				for (j = 0; j < nbytes; j++) {
    431        1.3     oster 					if (buf1[j] != buf2[j])
    432        1.3     oster 						break;
    433        1.3     oster 				}
    434        1.3     oster 				printf("psid=%ld j=%d\n", (long) psID, j);
    435        1.3     oster 				printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff,
    436        1.3     oster 				    buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff);
    437        1.3     oster 				printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff,
    438        1.3     oster 				    buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff);
    439        1.3     oster 			}
    440        1.3     oster 			if (rf_verifyParityDebug) {
    441        1.5     oster 				printf("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, i);
    442        1.3     oster 			}
    443       1.10     oster #endif
    444        1.3     oster 			/*
    445        1.3     oster 		         * Parity is bad. Keep track of which columns were bad.
    446        1.3     oster 		         */
    447        1.3     oster 			if (bbufs)
    448        1.3     oster 				bbufs[nbad] = i;
    449        1.3     oster 			nbad++;
    450        1.3     oster 			ret = RF_PARITY_BAD;
    451        1.3     oster 		}
    452        1.3     oster 		buf1 += nbytes;
    453        1.3     oster 		buf2 += nbytes;
    454        1.3     oster 	}
    455        1.3     oster 
    456        1.3     oster 	if ((ret != RF_PARITY_OKAY) && correct_it) {
    457        1.3     oster 		ret = RF_PARITY_COULD_NOT_CORRECT;
    458       1.10     oster #if RF_DEBUG_VERIFYPARITY
    459        1.3     oster 		if (rf_verifyParityDebug) {
    460        1.5     oster 			printf("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid);
    461        1.3     oster 		}
    462       1.10     oster #endif
    463        1.3     oster 		if (bbufs == NULL)
    464        1.3     oster 			goto done;
    465        1.3     oster 		/*
    466        1.3     oster 	         * Make a DAG with one write node for each bad unit. We'll simply
    467        1.3     oster 	         * write the contents of the data unit onto the parity unit for
    468        1.3     oster 	         * correction. (It's possible that the mirror copy was the correct
    469        1.3     oster 	         * copy, and that we're spooging good data by writing bad over it,
    470        1.3     oster 	         * but there's no way we can know that.
    471        1.3     oster 	         */
    472       1.26  christos 		wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, bf,
    473        1.3     oster 		    rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags,
    474        1.3     oster 		    RF_IO_NORMAL_PRIORITY);
    475        1.3     oster 		if (wr_dag_h == NULL)
    476        1.3     oster 			goto done;
    477        1.3     oster 		wrBlock = wr_dag_h->succedents[0];
    478        1.3     oster 		/*
    479        1.3     oster 	         * Fill in a write node for each bad compare.
    480        1.3     oster 	         */
    481        1.3     oster 		for (i = 0; i < nbad; i++) {
    482        1.3     oster 			j = i + layoutPtr->numDataCol;
    483        1.3     oster 			pda = blockNode->succedents[j]->params[0].p;
    484        1.3     oster 			pda->bufPtr = blockNode->succedents[i]->params[1].p;
    485        1.3     oster 			wrBlock->succedents[i]->params[0].p = pda;
    486        1.3     oster 			wrBlock->succedents[i]->params[1].p = pda->bufPtr;
    487        1.3     oster 			wrBlock->succedents[i]->params[2].v = psID;
    488  1.31.24.1     rmind 			wrBlock->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    489        1.3     oster 		}
    490       1.19     oster #if RF_ACC_TRACE > 0
    491        1.6   thorpej 		memset((char *) &tracerec, 0, sizeof(tracerec));
    492        1.3     oster 		wr_dag_h->tracerec = &tracerec;
    493       1.19     oster #endif
    494        1.9     oster #if 0
    495        1.3     oster 		if (rf_verifyParityDebug > 1) {
    496        1.3     oster 			printf("Parity verify write dag:\n");
    497        1.3     oster 			rf_PrintDAGList(wr_dag_h);
    498        1.3     oster 		}
    499        1.9     oster #endif
    500        1.3     oster 		RF_LOCK_MUTEX(mcpair->mutex);
    501        1.3     oster 		mcpair->flag = 0;
    502       1.20     oster 		RF_UNLOCK_MUTEX(mcpair->mutex);
    503       1.20     oster 
    504        1.3     oster 		/* fire off the write DAG */
    505        1.3     oster 		rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
    506        1.3     oster 		    (void *) mcpair);
    507       1.20     oster 
    508       1.20     oster 		RF_LOCK_MUTEX(mcpair->mutex);
    509        1.3     oster 		while (!mcpair->flag) {
    510        1.3     oster 			RF_WAIT_COND(mcpair->cond, mcpair->mutex);
    511        1.3     oster 		}
    512        1.3     oster 		RF_UNLOCK_MUTEX(mcpair->mutex);
    513        1.3     oster 		if (wr_dag_h->status != rf_enable) {
    514        1.3     oster 			RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n");
    515        1.3     oster 			goto done;
    516        1.3     oster 		}
    517        1.3     oster 		ret = RF_PARITY_CORRECTED;
    518        1.3     oster 	}
    519        1.1     oster done:
    520        1.3     oster 	/*
    521        1.3     oster          * All done. We might've gotten here without doing part of the function,
    522        1.3     oster          * so cleanup what we have to and return our running status.
    523        1.3     oster          */
    524        1.3     oster 	if (asm_h)
    525        1.3     oster 		rf_FreeAccessStripeMap(asm_h);
    526        1.3     oster 	if (rd_dag_h)
    527        1.3     oster 		rf_FreeDAG(rd_dag_h);
    528        1.3     oster 	if (wr_dag_h)
    529        1.3     oster 		rf_FreeDAG(wr_dag_h);
    530        1.3     oster 	if (mcpair)
    531        1.3     oster 		rf_FreeMCPair(mcpair);
    532        1.3     oster 	rf_FreeAllocList(allocList);
    533       1.10     oster #if RF_DEBUG_VERIFYPARITY
    534        1.3     oster 	if (rf_verifyParityDebug) {
    535       1.25     perry 		printf("raid%d: RAID1 parity verify, returning %d\n",
    536        1.5     oster 		       raidPtr->raidid, ret);
    537        1.3     oster 	}
    538       1.10     oster #endif
    539        1.3     oster 	return (ret);
    540        1.1     oster }
    541        1.1     oster 
    542       1.25     perry /* rbuf          - the recon buffer to submit
    543       1.15     oster  * keep_it       - whether we can keep this buffer or we have to return it
    544       1.15     oster  * use_committed - whether to use a committed or an available recon buffer
    545       1.15     oster  */
    546       1.15     oster 
    547       1.25     perry int
    548       1.15     oster rf_SubmitReconBufferRAID1(RF_ReconBuffer_t *rbuf, int keep_it,
    549       1.15     oster 			  int use_committed)
    550        1.1     oster {
    551        1.3     oster 	RF_ReconParityStripeStatus_t *pssPtr;
    552        1.3     oster 	RF_ReconCtrl_t *reconCtrlPtr;
    553       1.21     oster 	int     retcode;
    554        1.3     oster 	RF_CallbackDesc_t *cb, *p;
    555        1.3     oster 	RF_ReconBuffer_t *t;
    556        1.3     oster 	RF_Raid_t *raidPtr;
    557       1.30  christos 	void *ta;
    558        1.3     oster 
    559        1.3     oster 	retcode = 0;
    560        1.3     oster 
    561        1.3     oster 	raidPtr = rbuf->raidPtr;
    562       1.14     oster 	reconCtrlPtr = raidPtr->reconControl;
    563        1.3     oster 
    564        1.3     oster 	RF_ASSERT(rbuf);
    565        1.3     oster 	RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
    566        1.3     oster 
    567       1.11     oster #if RF_DEBUG_RECON
    568        1.3     oster 	if (rf_reconbufferDebug) {
    569       1.14     oster 		printf("raid%d: RAID1 reconbuffer submission c%d psid %ld ru%d (failed offset %ld)\n",
    570       1.25     perry 		       raidPtr->raidid, rbuf->col,
    571        1.5     oster 		       (long) rbuf->parityStripeID, rbuf->which_ru,
    572        1.5     oster 		       (long) rbuf->failedDiskSectorOffset);
    573        1.3     oster 	}
    574       1.10     oster #endif
    575        1.3     oster 	if (rf_reconDebug) {
    576       1.30  christos 		unsigned char *b = rbuf->buffer;
    577        1.3     oster 		printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
    578        1.3     oster 		    (long) rbuf->parityStripeID, (long) rbuf->buffer);
    579        1.3     oster 		printf("RAID1 psid %ld   %02x %02x %02x %02x %02x\n",
    580       1.30  christos 		    (long)rbuf->parityStripeID, b[0], b[1], b[2], b[3], b[4]);
    581        1.3     oster 	}
    582       1.14     oster 	RF_LOCK_PSS_MUTEX(raidPtr, rbuf->parityStripeID);
    583        1.3     oster 
    584        1.3     oster 	RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
    585       1.24     oster 	while(reconCtrlPtr->rb_lock) {
    586       1.24     oster 		ltsleep(&reconCtrlPtr->rb_lock, PRIBIO, "reconctlcnmhs", 0, &reconCtrlPtr->rb_mutex);
    587       1.24     oster 	}
    588       1.24     oster 	reconCtrlPtr->rb_lock = 1;
    589       1.24     oster 	RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
    590        1.3     oster 
    591        1.3     oster 	pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
    592       1.21     oster 	    rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, NULL);
    593        1.3     oster 	RF_ASSERT(pssPtr);	/* if it didn't exist, we wouldn't have gotten
    594        1.3     oster 				 * an rbuf for it */
    595        1.3     oster 
    596        1.3     oster 	/*
    597        1.3     oster          * Since this is simple mirroring, the first submission for a stripe is also
    598        1.3     oster          * treated as the last.
    599        1.3     oster          */
    600        1.3     oster 
    601        1.3     oster 	t = NULL;
    602        1.3     oster 	if (keep_it) {
    603       1.11     oster #if RF_DEBUG_RECON
    604        1.3     oster 		if (rf_reconbufferDebug) {
    605       1.25     perry 			printf("raid%d: RAID1 rbuf submission: keeping rbuf\n",
    606        1.5     oster 			       raidPtr->raidid);
    607        1.3     oster 		}
    608       1.10     oster #endif
    609        1.3     oster 		t = rbuf;
    610        1.3     oster 	} else {
    611        1.3     oster 		if (use_committed) {
    612       1.11     oster #if RF_DEBUG_RECON
    613        1.3     oster 			if (rf_reconbufferDebug) {
    614        1.5     oster 				printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid);
    615        1.3     oster 			}
    616       1.10     oster #endif
    617        1.3     oster 			t = reconCtrlPtr->committedRbufs;
    618        1.3     oster 			RF_ASSERT(t);
    619        1.3     oster 			reconCtrlPtr->committedRbufs = t->next;
    620        1.3     oster 			t->next = NULL;
    621        1.3     oster 		} else
    622        1.3     oster 			if (reconCtrlPtr->floatingRbufs) {
    623       1.11     oster #if RF_DEBUG_RECON
    624        1.3     oster 				if (rf_reconbufferDebug) {
    625        1.5     oster 					printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid);
    626        1.3     oster 				}
    627       1.10     oster #endif
    628        1.3     oster 				t = reconCtrlPtr->floatingRbufs;
    629        1.3     oster 				reconCtrlPtr->floatingRbufs = t->next;
    630        1.3     oster 				t->next = NULL;
    631        1.3     oster 			}
    632        1.3     oster 	}
    633        1.3     oster 	if (t == NULL) {
    634       1.11     oster #if RF_DEBUG_RECON
    635        1.3     oster 		if (rf_reconbufferDebug) {
    636        1.5     oster 			printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid);
    637        1.3     oster 		}
    638       1.10     oster #endif
    639        1.3     oster 		RF_ASSERT((keep_it == 0) && (use_committed == 0));
    640        1.3     oster 		raidPtr->procsInBufWait++;
    641        1.3     oster 		if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
    642        1.3     oster 		    && (raidPtr->numFullReconBuffers == 0)) {
    643        1.3     oster 			/* ruh-ro */
    644        1.3     oster 			RF_ERRORMSG("Buffer wait deadlock\n");
    645       1.14     oster 			rf_PrintPSStatusTable(raidPtr);
    646        1.3     oster 			RF_PANIC();
    647        1.3     oster 		}
    648        1.3     oster 		pssPtr->flags |= RF_PSS_BUFFERWAIT;
    649        1.3     oster 		cb = rf_AllocCallbackDesc();
    650        1.3     oster 		cb->col = rbuf->col;
    651        1.3     oster 		cb->callbackArg.v = rbuf->parityStripeID;
    652        1.3     oster 		cb->next = NULL;
    653        1.3     oster 		if (reconCtrlPtr->bufferWaitList == NULL) {
    654        1.3     oster 			/* we are the wait list- lucky us */
    655        1.3     oster 			reconCtrlPtr->bufferWaitList = cb;
    656        1.3     oster 		} else {
    657        1.3     oster 			/* append to wait list */
    658        1.3     oster 			for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
    659        1.3     oster 			p->next = cb;
    660        1.3     oster 		}
    661        1.3     oster 		retcode = 1;
    662        1.3     oster 		goto out;
    663        1.3     oster 	}
    664        1.3     oster 	if (t != rbuf) {
    665        1.3     oster 		t->col = reconCtrlPtr->fcol;
    666        1.3     oster 		t->parityStripeID = rbuf->parityStripeID;
    667        1.3     oster 		t->which_ru = rbuf->which_ru;
    668        1.3     oster 		t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
    669        1.3     oster 		t->spCol = rbuf->spCol;
    670        1.3     oster 		t->spOffset = rbuf->spOffset;
    671        1.3     oster 		/* Swap buffers. DANCE! */
    672        1.3     oster 		ta = t->buffer;
    673        1.3     oster 		t->buffer = rbuf->buffer;
    674        1.3     oster 		rbuf->buffer = ta;
    675        1.3     oster 	}
    676        1.3     oster 	/*
    677        1.3     oster          * Use the rbuf we've been given as the target.
    678        1.3     oster          */
    679        1.3     oster 	RF_ASSERT(pssPtr->rbuf == NULL);
    680        1.3     oster 	pssPtr->rbuf = t;
    681        1.3     oster 
    682        1.3     oster 	t->count = 1;
    683        1.3     oster 	/*
    684        1.3     oster          * Below, we use 1 for numDataCol (which is equal to the count in the
    685        1.3     oster          * previous line), so we'll always be done.
    686        1.3     oster          */
    687        1.3     oster 	rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
    688        1.1     oster 
    689        1.1     oster out:
    690       1.14     oster 	RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->parityStripeID);
    691       1.24     oster 	RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
    692       1.24     oster 	reconCtrlPtr->rb_lock = 0;
    693       1.24     oster 	wakeup(&reconCtrlPtr->rb_lock);
    694        1.3     oster 	RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
    695       1.11     oster #if RF_DEBUG_RECON
    696        1.3     oster 	if (rf_reconbufferDebug) {
    697       1.25     perry 		printf("raid%d: RAID1 rbuf submission: returning %d\n",
    698        1.5     oster 		       raidPtr->raidid, retcode);
    699        1.3     oster 	}
    700       1.10     oster #endif
    701        1.3     oster 	return (retcode);
    702        1.1     oster }
    703       1.31     oster 
    704       1.31     oster RF_HeadSepLimit_t
    705       1.31     oster rf_GetDefaultHeadSepLimitRAID1(RF_Raid_t *raidPtr)
    706       1.31     oster {
    707       1.31     oster 	return (10);
    708       1.31     oster }
    709       1.31     oster 
    710