int
rf_PQDoubleRecoveryFunc(RF_DagNode_t *node)
{
int np = node->numParams;
RF_AccessStripeMap_t *asmap =
(RF_AccessStripeMap_t *) node->params[np - 1].p;
RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[np - 2].p;
RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & (raidPtr->Layout);
int d, i;
unsigned coeff;
RF_RaidAddr_t sosAddr, suoffset;
RF_SectorCount_t len, secPerSU = layoutPtr->sectorsPerStripeUnit;
int two = 0;
RF_PhysDiskAddr_t *ppda, *ppda2, *qpda, *qpda2, *pda, npda;
char *buf;
int numDataCol = layoutPtr->numDataCol;
RF_Etimer_t timer;
RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
RF_ETIMER_START(timer);
if (asmap->failedPDAs[1] &&
(asmap->failedPDAs[1]->numSector +
asmap->failedPDAs[0]->numSector < secPerSU)) {
RF_ASSERT(0);
ppda = node->params[np - 6].p;
ppda2 = node->params[np - 5].p;
qpda = node->params[np - 4].p;
qpda2 = node->params[np - 3].p;
d = (np - 6);
two = 1;
} else {
ppda = node->params[np - 4].p;
qpda = node->params[np - 3].p;
d = (np - 4);
}
for (i = 0; i < d; i++) {
pda = node->params[i].p;
buf = pda->bufPtr;
suoffset = rf_StripeUnitOffset(layoutPtr, pda->startSector);
len = pda->numSector;
coeff = rf_RaidAddressToStripeUnitID(layoutPtr,
pda->raidAddress);
/* Compute the data unit offset within the column. */
coeff = (coeff % raidPtr->Layout.numDataCol);
/* See if pda intersects a recovery pda. */
rf_applyPDA(raidPtr, pda, ppda, qpda, node->dagHdr->bp);
if (two)
rf_applyPDA(raidPtr, pda, ppda, qpda, node->dagHdr->bp);
}
/*
* Ok, we got the parity back to the point where we can recover. We
* now need to determine the coeff of the columns that need to be
* recovered. We can also only need to recover a single stripe unit.
*/
if (asmap->failedPDAs[1] == NULL) { /*
* Only a single stripe unit
* to recover.
*/
pda = asmap->failedPDAs[0];
sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
asmap->raidAddress);
/* Need to determine the column of the other failed disk. */
coeff = rf_RaidAddressToStripeUnitID(layoutPtr,
pda->raidAddress);
/* Compute the data unit offset within the column. */
coeff = (coeff % raidPtr->Layout.numDataCol);
for (i = 0; i < numDataCol; i++) {
npda.raidAddress = sosAddr + (i * secPerSU);
(raidPtr->Layout.map->MapSector) (raidPtr,
npda.raidAddress, &(npda.row), &(npda.col),
&(npda.startSector), 0);
/* Skip over dead disks. */
if (RF_DEAD_DISK(raidPtr->Disks[npda.row][npda.col]
.status))
if (i != coeff)
break;
}
RF_ASSERT(i < numDataCol);
RF_ASSERT(two == 0);
/*
* Recover the data. Since we need only to recover one
* column, we overwrite the parity with the other one.
*/
if (coeff < i) /* Recovering 'a'. */
rf_PQ_recover((unsigned long *) ppda->bufPtr,
(unsigned long *) qpda->bufPtr,
(unsigned long *) pda->bufPtr,
(unsigned long *) ppda->bufPtr,
rf_RaidAddressToByte(raidPtr, pda->numSector),
coeff, i);
else /* Recovering 'b'. */
rf_PQ_recover((unsigned long *) ppda->bufPtr,
(unsigned long *) qpda->bufPtr,
(unsigned long *) ppda->bufPtr,
(unsigned long *) pda->bufPtr,
rf_RaidAddressToByte(raidPtr, pda->numSector),
i, coeff);
} else
RF_PANIC();
RF_ETIMER_STOP(timer);
RF_ETIMER_EVAL(timer);
if (tracerec)
tracerec->q_us += RF_ETIMER_VAL_US(timer);
rf_GenericWakeupFunc(node, 0);
return (0);
}
int
rf_PQWriteDoubleRecoveryFunc(RF_DagNode_t *node)
{
/*
* The situation:
*
* We are doing a write that hits only one failed data unit. The other
* failed data unit is not being overwritten, so we need to generate
* it.
*
* For the moment, we assume all the nonfailed data being written is in
* the shadow of the failed data unit. (i.e., either a single data
* unit write or the entire failed stripe unit is being overwritten.)
*
* Recovery strategy: apply the recovery data to the parity and Q.
* Use P & Q to recover the second failed data unit in P. Zero fill
* Q, then apply the recovered data to P. Then apply the data being
* written to the failed drive. Then walk through the surviving drives,
* applying new data when it exists, othewise the recovery data.
* Quite a mess.
*
*
* The params:
*
* read pda0, read pda1, ..., read pda (numDataCol-3),
* write pda0, ..., write pda (numStripeUnitAccess - numDataFailed),
* failed pda, raidPtr, asmap
*/
int np = node->numParams;
RF_AccessStripeMap_t *asmap = (RF_AccessStripeMap_t *)
node->params[np - 1].p;
RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[np - 2].p;
RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & (raidPtr->Layout);
int i;
RF_RaidAddr_t sosAddr;
unsigned coeff;
RF_StripeCount_t secPerSU = layoutPtr->sectorsPerStripeUnit;
RF_PhysDiskAddr_t *ppda, *qpda, *pda, npda;
int numDataCol = layoutPtr->numDataCol;
RF_Etimer_t timer;
RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
RF_ASSERT(node->numResults == 2);
RF_ASSERT(asmap->failedPDAs[1] == NULL);
RF_ETIMER_START(timer);
ppda = node->results[0];
qpda = node->results[1];
/* apply the recovery data */
for (i = 0; i < numDataCol - 2; i++)
rf_applyPDA(raidPtr, node->params[i].p, ppda, qpda,
node->dagHdr->bp);
/* Determine the other failed data unit. */
pda = asmap->failedPDAs[0];
sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
asmap->raidAddress);
/* Need to determine the column of the other failed disk. */
coeff = rf_RaidAddressToStripeUnitID(layoutPtr, pda->raidAddress);
/* Compute the data unit offset within the column. */
coeff = (coeff % raidPtr->Layout.numDataCol);
for (i = 0; i < numDataCol; i++) {
npda.raidAddress = sosAddr + (i * secPerSU);
(raidPtr->Layout.map->MapSector) (raidPtr, npda.raidAddress,
&(npda.row), &(npda.col), &(npda.startSector), 0);
/* Skip over dead disks. */
if (RF_DEAD_DISK(raidPtr->Disks[npda.row][npda.col].status))
if (i != coeff)
break;
}
RF_ASSERT(i < numDataCol);
/*
* Recover the data. The column we want to recover, we write over the
* parity. The column we don't care about, we dump in q.
*/
if (coeff < i) /* Recovering 'a'. */
rf_PQ_recover((unsigned long *) ppda->bufPtr,
(unsigned long *) qpda->bufPtr,
(unsigned long *) ppda->bufPtr,
(unsigned long *) qpda->bufPtr,
rf_RaidAddressToByte(raidPtr, pda->numSector), coeff, i);
else /* Recovering 'b'. */
rf_PQ_recover((unsigned long *) ppda->bufPtr,
(unsigned long *) qpda->bufPtr,
(unsigned long *) qpda->bufPtr,
(unsigned long *) ppda->bufPtr,
rf_RaidAddressToByte(raidPtr, pda->numSector), i, coeff);
/* OK. The valid data is in P. Zero fill Q, then inc it into it. */
bzero(qpda->bufPtr, rf_RaidAddressToByte(raidPtr, qpda->numSector));
rf_IncQ((unsigned long *) qpda->bufPtr, (unsigned long *) ppda->bufPtr,
rf_RaidAddressToByte(raidPtr, qpda->numSector), i);
/* Now apply all the write data to the buffer. */
/*
* Single stripe unit write case: The failed data is the only thing
* we are writing.
*/
RF_ASSERT(asmap->numStripeUnitsAccessed == 1);
/* Dest, src, len, coeff. */
rf_IncQ((unsigned long *) qpda->bufPtr,
(unsigned long *) asmap->failedPDAs[0]->bufPtr,
rf_RaidAddressToByte(raidPtr, qpda->numSector), coeff);
rf_bxor(asmap->failedPDAs[0]->bufPtr, ppda->bufPtr,
rf_RaidAddressToByte(raidPtr, ppda->numSector), node->dagHdr->bp);
/* Now apply all the recovery data. */
for (i = 0; i < numDataCol - 2; i++)
rf_applyPDA(raidPtr, node->params[i].p, ppda, qpda,
node->dagHdr->bp);
RF_ETIMER_STOP(timer);
RF_ETIMER_EVAL(timer);
if (tracerec)
tracerec->q_us += RF_ETIMER_VAL_US(timer);
rf_GenericWakeupFunc(node, 0);
return (0);
}
int
rf_Configure(RF_Raid_t *raidPtr, RF_Config_t *cfgPtr, RF_AutoConfig_t *ac)
{
RF_RowCol_t col;
int rc;
RF_LOCK_LKMGR_MUTEX(configureMutex);
configureCount++;
if (isconfigged == 0) {
rf_mutex_init(&rf_printf_mutex);
/* initialize globals */
DO_INIT_CONFIGURE(rf_ConfigureAllocList);
/*
* Yes, this does make debugging general to the whole
* system instead of being array specific. Bummer, drag.
*/
rf_ConfigureDebug(cfgPtr);
DO_INIT_CONFIGURE(rf_ConfigureDebugMem);
#if RF_ACC_TRACE > 0
DO_INIT_CONFIGURE(rf_ConfigureAccessTrace);
#endif
DO_INIT_CONFIGURE(rf_ConfigureMapModule);
DO_INIT_CONFIGURE(rf_ConfigureReconEvent);
DO_INIT_CONFIGURE(rf_ConfigureCallback);
DO_INIT_CONFIGURE(rf_ConfigureRDFreeList);
DO_INIT_CONFIGURE(rf_ConfigureNWayXor);
DO_INIT_CONFIGURE(rf_ConfigureStripeLockFreeList);
DO_INIT_CONFIGURE(rf_ConfigureMCPair);
DO_INIT_CONFIGURE(rf_ConfigureDAGs);
DO_INIT_CONFIGURE(rf_ConfigureDAGFuncs);
DO_INIT_CONFIGURE(rf_ConfigureReconstruction);
DO_INIT_CONFIGURE(rf_ConfigureCopyback);
DO_INIT_CONFIGURE(rf_ConfigureDiskQueueSystem);
DO_INIT_CONFIGURE(rf_ConfigurePSStatus);
isconfigged = 1;
}
RF_UNLOCK_LKMGR_MUTEX(configureMutex);
DO_RAID_MUTEX(&raidPtr->mutex);
/* set up the cleanup list. Do this after ConfigureDebug so that
* value of memDebug will be set */
rf_MakeAllocList(raidPtr->cleanupList);
if (raidPtr->cleanupList == NULL) {
DO_RAID_FAIL();
return (ENOMEM);
}
rf_ShutdownCreate(&raidPtr->shutdownList,
(void (*) (void *)) rf_FreeAllocList,
raidPtr->cleanupList);
raidPtr->numCol = cfgPtr->numCol;
raidPtr->numSpare = cfgPtr->numSpare;
raidPtr->status = rf_rs_optimal;
raidPtr->reconControl = NULL;
TAILQ_INIT(&(raidPtr->iodone));
simple_lock_init(&(raidPtr->iodone_lock));
DO_RAID_INIT_CONFIGURE(rf_ConfigureEngine);
DO_RAID_INIT_CONFIGURE(rf_ConfigureStripeLocks);
raidPtr->outstandingCond = 0;
raidPtr->nAccOutstanding = 0;
raidPtr->waitShutdown = 0;
DO_RAID_MUTEX(&raidPtr->access_suspend_mutex);
raidPtr->waitForReconCond = 0;
if (ac!=NULL) {
/* We have an AutoConfig structure.. Don't do the
normal disk configuration... call the auto config
stuff */
rf_AutoConfigureDisks(raidPtr, cfgPtr, ac);
} else {
DO_RAID_INIT_CONFIGURE(rf_ConfigureDisks);
DO_RAID_INIT_CONFIGURE(rf_ConfigureSpareDisks);
}
/* do this after ConfigureDisks & ConfigureSpareDisks to be sure dev
* no. is set */
DO_RAID_INIT_CONFIGURE(rf_ConfigureDiskQueues);
DO_RAID_INIT_CONFIGURE(rf_ConfigureLayout);
/* Initialize per-RAID PSS bits */
rf_InitPSStatus(raidPtr);
#if RF_INCLUDE_CHAINDECLUSTER > 0
for (col = 0; col < raidPtr->numCol; col++) {
/*
* XXX better distribution
*/
raidPtr->hist_diskreq[col] = 0;
}
#endif
raidPtr->numNewFailures = 0;
raidPtr->copyback_in_progress = 0;
raidPtr->parity_rewrite_in_progress = 0;
raidPtr->adding_hot_spare = 0;
raidPtr->recon_in_progress = 0;
raidPtr->maxOutstanding = cfgPtr->maxOutstandingDiskReqs;
/* autoconfigure and root_partition will actually get filled in
after the config is done */
raidPtr->autoconfigure = 0;
raidPtr->root_partition = 0;
raidPtr->last_unit = raidPtr->raidid;
raidPtr->config_order = 0;
if (rf_keepAccTotals) {
raidPtr->keep_acc_totals = 1;
}
/* Allocate a bunch of buffers to be used in low-memory conditions */
raidPtr->iobuf = NULL;
rc = rf_AllocEmergBuffers(raidPtr);
if (rc) {
printf("raid%d: Unable to allocate emergency buffers.\n",
raidPtr->raidid);
DO_RAID_FAIL();
return(rc);
}
/* Set up parity map stuff, if applicable. */
#ifndef RF_NO_PARITY_MAP
rf_paritymap_attach(raidPtr, cfgPtr->force);
#endif
raidPtr->valid = 1;
printf("raid%d: %s\n", raidPtr->raidid,
raidPtr->Layout.map->configName);
printf("raid%d: Components:", raidPtr->raidid);
for (col = 0; col < raidPtr->numCol; col++) {
printf(" %s", raidPtr->Disks[col].devname);
if (RF_DEAD_DISK(raidPtr->Disks[col].status)) {
printf("[**FAILED**]");
}
}
printf("\n");
printf("raid%d: Total Sectors: %" PRIu64 " (%" PRIu64 " MB)\n",
raidPtr->raidid,
raidPtr->totalSectors,
(raidPtr->totalSectors / 1024 *
(1 << raidPtr->logBytesPerSector) / 1024));
return (0);
}
