/*******************************************************************************************
* This degraded function allow only two case:
* 1. when write access the full failed stripe unit, then the access can be more than
* one tripe units.
* 2. when write access only part of the failed SU, we assume accesses of more than
* one stripe unit is not allowed so that the write can be dealt with like a
* large write.
* The following function is based on these assumptions. So except in the second case,
* it looks the same as a large write encodeing function. But this is not exactly the
* normal way for doing a degraded write, since raidframe have to break cases of access
* other than the above two into smaller accesses. We may have to change
* DegrESubroutin in the future.
*******************************************************************************************/
void
rf_DegrESubroutine(RF_DagNode_t *node, char *ebuf)
{
RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p;
RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & raidPtr->Layout;
RF_PhysDiskAddr_t *failedPDA = (RF_PhysDiskAddr_t *) node->params[node->numParams - 2].p;
RF_PhysDiskAddr_t *pda;
int i, suoffset, failedSUOffset = rf_StripeUnitOffset(layoutPtr, failedPDA->startSector);
RF_RowCol_t scol;
char *srcbuf, *destbuf;
RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
RF_Etimer_t timer;
RF_ETIMER_START(timer);
for (i = 0; i < node->numParams - 2; i += 2) {
RF_ASSERT(node->params[i + 1].p != ebuf);
pda = (RF_PhysDiskAddr_t *) node->params[i].p;
suoffset = rf_StripeUnitOffset(layoutPtr, pda->startSector);
scol = rf_EUCol(layoutPtr, pda->raidAddress);
srcbuf = (char *) node->params[i + 1].p;
destbuf = ebuf + rf_RaidAddressToByte(raidPtr, suoffset - failedSUOffset);
rf_e_encToBuf(raidPtr, scol, srcbuf, RF_EO_MATRIX_DIM - 2, destbuf, pda->numSector);
}
RF_ETIMER_STOP(timer);
RF_ETIMER_EVAL(timer);
tracerec->q_us += RF_ETIMER_VAL_US(timer);
}
/* Algorithm:
1. Store the difference of old data and new data in the Rod buffer.
2. then encode this buffer into the buffer which already have old 'E' information inside it,
the result can be shown to be the new 'E' information.
3. xor the Wnd buffer into the difference buffer to recover the original old data.
Here we have another alternative: to allocate a temporary buffer for storing the difference of
old data and new data, then encode temp buf into old 'E' buf to form new 'E', but this approach
take the same speed as the previous, and need more memory.
*/
int
rf_RegularONEFunc(RF_DagNode_t *node)
{
RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p;
RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & raidPtr->Layout;
int EpdaIndex = (node->numParams - 1) / 2 - 1; /* the parameter of node
* where you can find
* e-pda */
int i, k;
int suoffset, length;
RF_RowCol_t scol;
char *srcbuf, *destbuf;
RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
RF_Etimer_t timer;
RF_PhysDiskAddr_t *pda;
#ifdef RAID_DIAGNOSTIC
RF_PhysDiskAddr_t *EPDA =
(RF_PhysDiskAddr_t *) node->params[EpdaIndex].p;
int ESUOffset = rf_StripeUnitOffset(layoutPtr, EPDA->startSector);
RF_ASSERT(EPDA->type == RF_PDA_TYPE_Q);
RF_ASSERT(ESUOffset == 0);
#endif /* RAID_DIAGNOSTIC */
RF_ETIMER_START(timer);
/* Xor the Wnd buffer into Rod buffer, the difference of old data and
* new data is stored in Rod buffer */
for (k = 0; k < EpdaIndex; k += 2) {
length = rf_RaidAddressToByte(raidPtr, ((RF_PhysDiskAddr_t *) node->params[k].p)->numSector);
rf_bxor(node->params[k + EpdaIndex + 3].p, node->params[k + 1].p, length);
}
/* Start to encoding the buffer storing the difference of old data and
* new data into 'E' buffer */
for (i = 0; i < EpdaIndex; i += 2)
if (node->params[i + 1].p != node->results[0]) { /* results[0] is buf ptr
* of E */
pda = (RF_PhysDiskAddr_t *) node->params[i].p;
srcbuf = (char *) node->params[i + 1].p;
scol = rf_EUCol(layoutPtr, pda->raidAddress);
suoffset = rf_StripeUnitOffset(layoutPtr, pda->startSector);
destbuf = ((char *) node->results[0]) + rf_RaidAddressToByte(raidPtr, suoffset);
rf_e_encToBuf(raidPtr, scol, srcbuf, RF_EO_MATRIX_DIM - 2, destbuf, pda->numSector);
}
/* Recover the original old data to be used by parity encoding
* function in XorNode */
for (k = 0; k < EpdaIndex; k += 2) {
length = rf_RaidAddressToByte(raidPtr, ((RF_PhysDiskAddr_t *) node->params[k].p)->numSector);
rf_bxor(node->params[k + EpdaIndex + 3].p, node->params[k + 1].p, length);
}
RF_ETIMER_STOP(timer);
RF_ETIMER_EVAL(timer);
tracerec->q_us += RF_ETIMER_VAL_US(timer);
rf_GenericWakeupFunc(node, 0);
#if 1
return (0); /* XXX this was missing.. GO */
#endif
}
void
rf_applyPDA(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda,
RF_PhysDiskAddr_t *ppda, RF_PhysDiskAddr_t *qpda, void *bp)
{
RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
RF_RaidAddr_t s0off = rf_StripeUnitOffset(layoutPtr, ppda->startSector);
RF_SectorCount_t s0len = ppda->numSector, len;
RF_SectorNum_t suoffset;
unsigned coeff;
char *pbuf = ppda->bufPtr;
char *qbuf = qpda->bufPtr;
char *buf;
int delta;
suoffset = rf_StripeUnitOffset(layoutPtr, pda->startSector);
len = pda->numSector;
/* See if pda intersects a recovery pda. */
if ((suoffset < s0off + s0len) && (suoffset + len > s0off)) {
buf = pda->bufPtr;
coeff = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout),
pda->raidAddress);
coeff = (coeff % raidPtr->Layout.numDataCol);
if (suoffset < s0off) {
delta = s0off - suoffset;
buf += rf_RaidAddressToStripeUnitID(&(raidPtr->Layout),
delta);
suoffset = s0off;
len -= delta;
}
if (suoffset > s0off) {
delta = suoffset - s0off;
pbuf += rf_RaidAddressToStripeUnitID(&(raidPtr->Layout),
delta);
qbuf += rf_RaidAddressToStripeUnitID(&(raidPtr->Layout),
delta);
}
if ((suoffset + len) > (s0len + s0off))
len = s0len + s0off - suoffset;
/* Src, dest, len. */
rf_bxor(buf, pbuf, rf_RaidAddressToByte(raidPtr, len), bp);
/* Dest, src, len, coeff. */
rf_IncQ((unsigned long *) qbuf, (unsigned long *) buf,
rf_RaidAddressToByte(raidPtr, len), coeff);
}
}
/**************************************************************************************
* when parity die and one data die, We use second redundant information, 'E',
* to recover the data in dead disk. This function is used in the recovery node of
* for EO_110_CreateReadDAG
**************************************************************************************/
int
rf_RecoveryEFunc(RF_DagNode_t *node)
{
RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p;
RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & raidPtr->Layout;
RF_PhysDiskAddr_t *failedPDA = (RF_PhysDiskAddr_t *) node->params[node->numParams - 2].p;
RF_RowCol_t scol, /* source logical column */
fcol = rf_EUCol(layoutPtr, failedPDA->raidAddress); /* logical column of
* failed SU */
int i;
RF_PhysDiskAddr_t *pda;
int suoffset, failedSUOffset = rf_StripeUnitOffset(layoutPtr, failedPDA->startSector);
char *srcbuf, *destbuf;
RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
RF_Etimer_t timer;
memset((char *) node->results[0], 0,
rf_RaidAddressToByte(raidPtr, failedPDA->numSector));
if (node->dagHdr->status == rf_enable) {
RF_ETIMER_START(timer);
for (i = 0; i < node->numParams - 2; i += 2)
if (node->params[i + 1].p != node->results[0]) {
pda = (RF_PhysDiskAddr_t *) node->params[i].p;
if (i == node->numParams - 4)
scol = RF_EO_MATRIX_DIM - 2; /* the colume of
* redundant E */
else
scol = rf_EUCol(layoutPtr, pda->raidAddress);
srcbuf = (char *) node->params[i + 1].p;
suoffset = rf_StripeUnitOffset(layoutPtr, pda->startSector);
destbuf = ((char *) node->results[0]) + rf_RaidAddressToByte(raidPtr, suoffset - failedSUOffset);
rf_e_encToBuf(raidPtr, scol, srcbuf, fcol, destbuf, pda->numSector);
}
RF_ETIMER_STOP(timer);
RF_ETIMER_EVAL(timer);
tracerec->xor_us += RF_ETIMER_VAL_US(timer);
}
return (rf_GenericWakeupFunc(node, 0)); /* node execute successfully */
}
/*
* XXX move this elsewhere?
*/
void
rf_DD_GenerateFailedAccessASMs(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
RF_PhysDiskAddr_t **pdap, int *nNodep,
RF_PhysDiskAddr_t **pqpdap, int *nPQNodep,
RF_AllocListElem_t *allocList)
{
RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
int PDAPerDisk, i;
RF_SectorCount_t secPerSU = layoutPtr->sectorsPerStripeUnit;
int numDataCol = layoutPtr->numDataCol;
int state;
RF_SectorNum_t suoff, suend;
unsigned firstDataCol, napdas, count;
RF_SectorNum_t fone_start, fone_end, ftwo_start = 0, ftwo_end = 0;
RF_PhysDiskAddr_t *fone = asmap->failedPDAs[0], *ftwo = asmap->failedPDAs[1];
RF_PhysDiskAddr_t *pda_p;
RF_PhysDiskAddr_t *phys_p;
RF_RaidAddr_t sosAddr;
/* determine how many pda's we will have to generate per unaccess
* stripe. If there is only one failed data unit, it is one; if two,
* possibly two, depending whether they overlap. */
fone_start = rf_StripeUnitOffset(layoutPtr, fone->startSector);
fone_end = fone_start + fone->numSector;
#define CONS_PDA(if,start,num) \
pda_p->col = asmap->if->col; \
pda_p->startSector = ((asmap->if->startSector / secPerSU) * secPerSU) + start; \
pda_p->numSector = num; \
pda_p->next = NULL; \
RF_MallocAndAdd(pda_p->bufPtr,rf_RaidAddressToByte(raidPtr,num),(char *), allocList)
if (asmap->numDataFailed == 1) {
PDAPerDisk = 1;
state = 1;
RF_MallocAndAdd(*pqpdap, 2 * sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t *), allocList);
pda_p = *pqpdap;
/* build p */
CONS_PDA(parityInfo, fone_start, fone->numSector);
pda_p->type = RF_PDA_TYPE_PARITY;
pda_p++;
/* build q */
CONS_PDA(qInfo, fone_start, fone->numSector);
pda_p->type = RF_PDA_TYPE_Q;
} else {
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);
}
