/* Invoked when the copyback has completed. */
void
rf_CopybackComplete(RF_CopybackDesc_t *desc, int status)
{
RF_Raid_t *raidPtr = desc->raidPtr;
struct timeval t, diff;
if (!status) {
RF_LOCK_MUTEX(raidPtr->mutex);
if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
RF_ASSERT(raidPtr->Layout.map->parityConfig == 'D');
rf_FreeSpareTable(raidPtr);
} else {
raidPtr->Disks[desc->spRow][desc->spCol].status =
rf_ds_spare;
}
RF_UNLOCK_MUTEX(raidPtr->mutex);
RF_GETTIME(t);
RF_TIMEVAL_DIFF(&desc->starttime, &t, &diff);
printf("Copyback time was %d.%06d seconds.\n",
(int) diff.tv_sec, (int) diff.tv_usec);
} else
printf("COPYBACK: Failure.\n");
RF_Free(desc->databuf, rf_RaidAddressToByte(raidPtr, desc->sectPerSU));
rf_FreeMCPair(desc->mcpair);
RF_Free(desc, sizeof(*desc));
rf_copyback_in_progress = 0;
rf_ResumeNewRequests(raidPtr);
}
void
rf_free_2d_array(RF_RowCol_t **a, int b, int k)
{
RF_RowCol_t i;
for (i = 0; i < b; i++)
RF_Free(a[i], k * sizeof(RF_RowCol_t));
RF_Free(a, b * sizeof(RF_RowCol_t));
}
void
rf_FreeSpareTable(RF_Raid_t *raidPtr)
{
long i;
RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
RF_SpareTableEntry_t **table = info->SpareTable;
for (i = 0; i < info->TablesPerSpareRegion; i++) {
RF_Free(table[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t));
}
RF_Free(table, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
info->SpareTable = (RF_SpareTableEntry_t **) NULL;
}
RF_ReconMap_t *
rf_MakeReconMap(
RF_Raid_t *raidPtr,
RF_SectorCount_t ru_sectors, /*
* Size of reconstruction unit
* in sectors.
*/
RF_SectorCount_t disk_sectors, /* Size of disk in sectors. */
RF_ReconUnitCount_t spareUnitsPerDisk /*
* Zero unless distributed
* sparing.
*/
)
{
RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
RF_ReconUnitCount_t num_rus = layoutPtr->stripeUnitsPerDisk /
layoutPtr->SUsPerRU;
RF_ReconMap_t *p;
int rc;
RF_Malloc(p, sizeof(RF_ReconMap_t), (RF_ReconMap_t *));
p->sectorsPerReconUnit = ru_sectors;
p->sectorsInDisk = disk_sectors;
p->totalRUs = num_rus;
p->spareRUs = spareUnitsPerDisk;
p->unitsLeft = num_rus - spareUnitsPerDisk;
RF_Malloc(p->status, num_rus * sizeof(RF_ReconMapListElem_t *),
(RF_ReconMapListElem_t **));
RF_ASSERT(p->status != (RF_ReconMapListElem_t **) NULL);
(void) bzero((char *) p->status, num_rus *
sizeof(RF_ReconMapListElem_t *));
p->size = sizeof(RF_ReconMap_t) + num_rus *
sizeof(RF_ReconMapListElem_t *);
p->maxSize = p->size;
rc = rf_mutex_init(&p->mutex);
if (rc) {
RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d.\n",
__FILE__, __LINE__, rc);
RF_Free(p->status, num_rus * sizeof(RF_ReconMapListElem_t *));
RF_Free(p, sizeof(RF_ReconMap_t));
return (NULL);
}
return (p);
}
int
rf_ConfigureAccessTrace(RF_ShutdownList_t **listp)
{
int rc;
numTracesSoFar = accessTraceBufCount = rf_stopCollectingTraces = 0;
if (rf_accessTraceBufSize) {
RF_Malloc(access_tracebuf, rf_accessTraceBufSize *
sizeof(RF_AccTraceEntry_t), (RF_AccTraceEntry_t *));
accessTraceBufCount = 0;
}
traceCount = 0;
numTracesSoFar = 0;
rc = rf_mutex_init(&rf_tracing_mutex);
if (rc) {
RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d.\n",
__FILE__, __LINE__, rc);
}
rc = rf_ShutdownCreate(listp, rf_ShutdownAccessTrace, NULL);
if (rc) {
RF_ERRORMSG3("Unable to add to shutdown list file %s line %d"
" rc=%d.\n", __FILE__, __LINE__, rc);
if (rf_accessTraceBufSize) {
RF_Free(access_tracebuf, rf_accessTraceBufSize *
sizeof(RF_AccTraceEntry_t));
rf_mutex_destroy(&rf_tracing_mutex);
}
}
return (rc);
}
/*
* Invoked via ioctl to install a spare table in the kernel.
*/
int
rf_SetSpareTable(RF_Raid_t *raidPtr, void *data)
{
RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
RF_SpareTableEntry_t **ptrs;
int i, retcode;
/* what we need to copyin is a 2-d array, so first copyin the user
* pointers to the rows in the table */
RF_Malloc(ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *), (RF_SpareTableEntry_t **));
retcode = copyin((void *) data, (void *) ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
if (retcode)
return (retcode);
/* now allocate kernel space for the row pointers */
RF_Malloc(info->SpareTable, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *), (RF_SpareTableEntry_t **));
/* now allocate kernel space for each row in the table, and copy it in
* from user space */
for (i = 0; i < info->TablesPerSpareRegion; i++) {
RF_Malloc(info->SpareTable[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t), (RF_SpareTableEntry_t *));
retcode = copyin(ptrs[i], info->SpareTable[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t));
if (retcode) {
info->SpareTable = NULL; /* blow off the memory
* we've allocated */
return (retcode);
}
}
/* free up the temporary array we used */
RF_Free(ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
return (0);
}
/*
* We use the debug_mem_mutex here because we need to lock it anyway to call
* free. This is probably a bug somewhere else in the code, but when I call
* malloc/free outside of any lock, I have endless trouble with malloc
* appearing to return the same pointer twice. Since we have to lock it
* anyway, we might as well use it as the lock around the al_free_list.
* Note that we can't call Free with the debug_mem_mutex locked.
*/
void
rf_FreeAllocList(RF_AllocListElem_t *l)
{
int i;
RF_AllocListElem_t *temp, *p;
for (p = l; p; p = p->next) {
RF_ASSERT(p->numPointers >= 0 &&
p->numPointers <= RF_POINTERS_PER_ALLOC_LIST_ELEMENT);
for (i = 0; i < p->numPointers; i++) {
RF_ASSERT(p->pointers[i]);
RF_Free(p->pointers[i], p->sizes[i]);
}
}
while (l) {
temp = l;
l = l->next;
if (al_free_list_count > RF_AL_FREELIST_MAX) {
DO_FREE(temp, sizeof(*temp));
} else {
temp->next = al_free_list;
al_free_list = temp;
al_free_list_count++;
}
}
}
RF_CommonLogData_t *
rf_AllocParityLogCommonData(RF_Raid_t *raidPtr)
{
RF_CommonLogData_t *common = NULL;
int rc;
/*
* Return a struct for holding common parity log information from the
* free list (rf_parityLogDiskQueue.freeCommonList). If the free list
* is empty, call RF_Malloc to create a new structure. NON-BLOCKING
*/
RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
if (raidPtr->parityLogDiskQueue.freeCommonList) {
common = raidPtr->parityLogDiskQueue.freeCommonList;
raidPtr->parityLogDiskQueue.freeCommonList =
raidPtr->parityLogDiskQueue.freeCommonList->next;
RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
} else {
RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
RF_Malloc(common, sizeof(RF_CommonLogData_t),
(RF_CommonLogData_t *));
rc = rf_mutex_init(&common->mutex);
if (rc) {
RF_ERRORMSG3("Unable to init mutex file %s line %d"
" rc=%d\n", __FILE__, __LINE__, rc);
RF_Free(common, sizeof(RF_CommonLogData_t));
common = NULL;
}
}
common->next = NULL;
return (common);
}
void
rf_ShutdownAccessTrace(void *ignored)
{
if (rf_accessTraceBufSize) {
if (accessTraceBufCount)
rf_FlushAccessTraceBuf();
RF_Free(access_tracebuf, rf_accessTraceBufSize *
sizeof(RF_AccTraceEntry_t));
}
rf_mutex_destroy(&rf_tracing_mutex);
}
void
rf_FreeReconMapListElem(RF_ReconMap_t *mapPtr, RF_ReconMapListElem_t *p)
{
int delta;
if (mapPtr) {
delta = 0 - (int) sizeof(RF_ReconMapListElem_t);
rf_update_size(mapPtr, delta);
}
RF_Free(p, sizeof(*p));
}
int
rf_get_info50(RF_Raid_t *raidPtr, void *data)
{
RF_DeviceConfig50_t **ucfgp = data, *d_cfg;
size_t i, j;
int error;
if (!raidPtr->valid)
return ENODEV;
RF_Malloc(d_cfg, sizeof(RF_DeviceConfig50_t), (RF_DeviceConfig50_t *));
if (d_cfg == NULL)
return ENOMEM;
d_cfg->rows = 1; /* there is only 1 row now */
d_cfg->cols = raidPtr->numCol;
d_cfg->ndevs = raidPtr->numCol;
if (d_cfg->ndevs >= RF_MAX_DISKS)
goto nomem;
d_cfg->nspares = raidPtr->numSpare;
if (d_cfg->nspares >= RF_MAX_DISKS)
goto nomem;
d_cfg->maxqdepth = raidPtr->maxQueueDepth;
for (j = 0; j < d_cfg->cols; j++)
rf_disk_to_disk50(&d_cfg->devs[j], &raidPtr->Disks[j]);
for (j = d_cfg->cols, i = 0; i < d_cfg->nspares; i++, j++)
rf_disk_to_disk50(&d_cfg->spares[i], &raidPtr->Disks[j]);
error = copyout(d_cfg, *ucfgp, sizeof(RF_DeviceConfig50_t));
RF_Free(d_cfg, sizeof(RF_DeviceConfig50_t));
return error;
nomem:
RF_Free(d_cfg, sizeof(RF_DeviceConfig_t));
return ENOMEM;
}
void
rf_FreeReconMap(RF_ReconMap_t *mapPtr)
{
RF_ReconMapListElem_t *p, *q;
RF_ReconUnitCount_t numRUs;
RF_ReconUnitNum_t i;
numRUs = mapPtr->sectorsInDisk / mapPtr->sectorsPerReconUnit;
if (mapPtr->sectorsInDisk % mapPtr->sectorsPerReconUnit)
numRUs++;
for (i = 0; i < numRUs; i++) {
p = mapPtr->status[i];
while (p != RU_NOTHING && p != RU_ALL) {
q = p;
p = p->next;
RF_Free(q, sizeof(*q));
}
}
rf_mutex_destroy(&mapPtr->mutex);
RF_Free(mapPtr->status, mapPtr->totalRUs *
sizeof(RF_ReconMapListElem_t *));
RF_Free(mapPtr, sizeof(RF_ReconMap_t));
}
void
rf_FreeParityStripeStatusTable(RF_Raid_t *raidPtr,
RF_PSStatusHeader_t *pssTable)
{
#if RF_DEBUG_PSS
int i;
if (rf_pssDebug)
RealPrintPSStatusTable(raidPtr, pssTable);
for (i = 0; i < raidPtr->pssTableSize; i++) {
if (pssTable[i].chain) {
printf("ERROR: pss hash chain not null at recon shutdown\n");
}
}
#endif
RF_Free(pssTable, raidPtr->pssTableSize * sizeof(RF_PSStatusHeader_t));
}
void
rf_free_1d_array(RF_RowCol_t *a, int n)
{
RF_Free(a, n * sizeof(RF_RowCol_t));
}
void
rf_doubleEOdecode(
RF_Raid_t * raidPtr,
char **rrdbuf,
char **dest,
RF_RowCol_t * fcol,
char *pbuf,
char *ebuf)
{
RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & (raidPtr->Layout);
int i, j, k, f1, f2, row;
int rrdrow, erow, count = 0;
int bytesPerSector = rf_RaidAddressToByte(raidPtr, 1);
int numRowInEncMatix = (RF_EO_MATRIX_DIM) - 1;
#if 0
int pcol = (RF_EO_MATRIX_DIM) - 1;
#endif
int ecol = (RF_EO_MATRIX_DIM) - 2;
int bytesPerEU = bytesPerSector / numRowInEncMatix;
int numDataCol = layoutPtr->numDataCol;
#if RF_EO_MATRIX_DIM > 17
int shortsPerEU = bytesPerEU / sizeof(short);
short *rrdbuf_current, *pbuf_current, *ebuf_current;
short *dest_smaller, *dest_smaller_current, *dest_larger, *dest_larger_current;
short *temp;
short *P;
RF_ASSERT(bytesPerEU % sizeof(short) == 0);
RF_Malloc(P, bytesPerEU, (short *));
RF_Malloc(temp, bytesPerEU, (short *));
#elif RF_EO_MATRIX_DIM == 17
int longsPerEU = bytesPerEU / sizeof(long);
long *rrdbuf_current, *pbuf_current, *ebuf_current;
long *dest_smaller, *dest_smaller_current, *dest_larger, *dest_larger_current;
long *temp;
long *P;
RF_ASSERT(bytesPerEU % sizeof(long) == 0);
RF_Malloc(P, bytesPerEU, (long *));
RF_Malloc(temp, bytesPerEU, (long *));
#endif
RF_ASSERT(*((long *) dest[0]) == 0);
RF_ASSERT(*((long *) dest[1]) == 0);
memset((char *) P, 0, bytesPerEU);
memset((char *) temp, 0, bytesPerEU);
RF_ASSERT(*P == 0);
/* calculate the 'P' parameter, which, not parity, is the Xor of all
* elements in the last two column, ie. 'E' and 'parity' colume, see
* the Ref. paper by Blaum, et al 1993 */
for (i = 0; i < numRowInEncMatix; i++)
for (k = 0; k < longsPerEU; k++) {
#if RF_EO_MATRIX_DIM > 17
ebuf_current = ((short *) ebuf) + i * shortsPerEU + k;
pbuf_current = ((short *) pbuf) + i * shortsPerEU + k;
#elif RF_EO_MATRIX_DIM == 17
ebuf_current = ((long *) ebuf) + i * longsPerEU + k;
pbuf_current = ((long *) pbuf) + i * longsPerEU + k;
#endif
P[k] ^= *ebuf_current;
P[k] ^= *pbuf_current;
}
RF_ASSERT(fcol[0] != fcol[1]);
if (fcol[0] < fcol[1]) {
#if RF_EO_MATRIX_DIM > 17
dest_smaller = (short *) (dest[0]);
dest_larger = (short *) (dest[1]);
#elif RF_EO_MATRIX_DIM == 17
dest_smaller = (long *) (dest[0]);
dest_larger = (long *) (dest[1]);
#endif
f1 = fcol[0];
f2 = fcol[1];
} else {
#if RF_EO_MATRIX_DIM > 17
dest_smaller = (short *) (dest[1]);
dest_larger = (short *) (dest[0]);
#elif RF_EO_MATRIX_DIM == 17
dest_smaller = (long *) (dest[1]);
dest_larger = (long *) (dest[0]);
#endif
f1 = fcol[1];
f2 = fcol[0];
}
row = (RF_EO_MATRIX_DIM) - 1;
while ((row = rf_EO_Mod((row + f1 - f2), RF_EO_MATRIX_DIM)) != ((RF_EO_MATRIX_DIM) - 1)) {
#if RF_EO_MATRIX_DIM > 17
dest_larger_current = dest_larger + row * shortsPerEU;
dest_smaller_current = dest_smaller + row * shortsPerEU;
#elif RF_EO_MATRIX_DIM == 17
dest_larger_current = dest_larger + row * longsPerEU;
dest_smaller_current = dest_smaller + row * longsPerEU;
#endif
/** Do the diagonal recovery. Initially, temp[k] = (failed 1),
which is the failed data in the colume which has smaller col index. **/
/* step 1: ^(SUM of nonfailed in-diagonal A(rrdrow,0..m-3)) */
for (j = 0; j < numDataCol; j++) {
if (j == f1 || j == f2)
continue;
rrdrow = rf_EO_Mod((row + f2 - j), RF_EO_MATRIX_DIM);
if (rrdrow != (RF_EO_MATRIX_DIM) - 1) {
#if RF_EO_MATRIX_DIM > 17
rrdbuf_current = (short *) (rrdbuf[j]) + rrdrow * shortsPerEU;
for (k = 0; k < shortsPerEU; k++)
temp[k] ^= *(rrdbuf_current + k);
#elif RF_EO_MATRIX_DIM == 17
rrdbuf_current = (long *) (rrdbuf[j]) + rrdrow * longsPerEU;
for (k = 0; k < longsPerEU; k++)
temp[k] ^= *(rrdbuf_current + k);
#endif
}
}
/* step 2: ^E(erow,m-2), If erow is at the buttom row, don't
* Xor into it E(erow,m-2) = (principle diagonal) ^ (failed
* 1) ^ (failed 2) ^ ( SUM of nonfailed in-diagonal
* A(rrdrow,0..m-3) ) After this step, temp[k] = (principle
* diagonal) ^ (failed 2) */
erow = rf_EO_Mod((row + f2 - ecol), (RF_EO_MATRIX_DIM));
if (erow != (RF_EO_MATRIX_DIM) - 1) {
#if RF_EO_MATRIX_DIM > 17
ebuf_current = (short *) ebuf + shortsPerEU * erow;
for (k = 0; k < shortsPerEU; k++)
temp[k] ^= *(ebuf_current + k);
#elif RF_EO_MATRIX_DIM == 17
ebuf_current = (long *) ebuf + longsPerEU * erow;
for (k = 0; k < longsPerEU; k++)
temp[k] ^= *(ebuf_current + k);
#endif
}
/* step 3: ^P to obtain the failed data (failed 2). P can be
* proved to be actually (principle diagonal) After this
* step, temp[k] = (failed 2), the failed data to be recovered */
#if RF_EO_MATRIX_DIM > 17
for (k = 0; k < shortsPerEU; k++)
temp[k] ^= P[k];
/* Put the data to the destination buffer */
for (k = 0; k < shortsPerEU; k++)
dest_larger_current[k] = temp[k];
#elif RF_EO_MATRIX_DIM == 17
for (k = 0; k < longsPerEU; k++)
temp[k] ^= P[k];
/* Put the data to the destination buffer */
for (k = 0; k < longsPerEU; k++)
dest_larger_current[k] = temp[k];
#endif
/** THE FOLLOWING DO THE HORIZONTAL XOR **/
/* step 1: ^(SUM of A(row,0..m-3)), ie. all nonfailed data
* columes */
for (j = 0; j < numDataCol; j++) {
if (j == f1 || j == f2)
continue;
#if RF_EO_MATRIX_DIM > 17
rrdbuf_current = (short *) (rrdbuf[j]) + row * shortsPerEU;
for (k = 0; k < shortsPerEU; k++)
temp[k] ^= *(rrdbuf_current + k);
#elif RF_EO_MATRIX_DIM == 17
rrdbuf_current = (long *) (rrdbuf[j]) + row * longsPerEU;
for (k = 0; k < longsPerEU; k++)
temp[k] ^= *(rrdbuf_current + k);
#endif
}
/* step 2: ^A(row,m-1) */
/* step 3: Put the data to the destination buffer */
#if RF_EO_MATRIX_DIM > 17
pbuf_current = (short *) pbuf + shortsPerEU * row;
for (k = 0; k < shortsPerEU; k++)
temp[k] ^= *(pbuf_current + k);
for (k = 0; k < shortsPerEU; k++)
dest_smaller_current[k] = temp[k];
#elif RF_EO_MATRIX_DIM == 17
pbuf_current = (long *) pbuf + longsPerEU * row;
for (k = 0; k < longsPerEU; k++)
temp[k] ^= *(pbuf_current + k);
for (k = 0; k < longsPerEU; k++)
dest_smaller_current[k] = temp[k];
#endif
count++;
}
/* Check if all Encoding Unit in the data buffer have been decoded,
* according EvenOdd theory, if "RF_EO_MATRIX_DIM" is a prime number,
* this algorithm will covered all buffer */
RF_ASSERT(count == numRowInEncMatix);
RF_Free((char *) P, bytesPerEU);
RF_Free((char *) temp, bytesPerEU);
}
int
rf_config50(RF_Raid_t *raidPtr, int unit, void *data, RF_Config_t **k_cfgp)
{
RF_Config50_t *u50_cfg, *k50_cfg;
RF_Config_t *k_cfg;
size_t i, j;
int error;
if (raidPtr->valid) {
/* There is a valid RAID set running on this unit! */
printf("raid%d: Device already configured!\n", unit);
return EINVAL;
}
/* copy-in the configuration information */
/* data points to a pointer to the configuration structure */
u50_cfg = *((RF_Config50_t **) data);
RF_Malloc(k50_cfg, sizeof(RF_Config50_t), (RF_Config50_t *));
if (k50_cfg == NULL)
return ENOMEM;
error = copyin(u50_cfg, k50_cfg, sizeof(RF_Config50_t));
if (error) {
RF_Free(k50_cfg, sizeof(RF_Config50_t));
return error;
}
RF_Malloc(k_cfg, sizeof(RF_Config_t), (RF_Config_t *));
if (k_cfg == NULL) {
RF_Free(k50_cfg, sizeof(RF_Config50_t));
return ENOMEM;
}
k_cfg->numRow = k50_cfg->numRow;
k_cfg->numCol = k50_cfg->numCol;
k_cfg->numSpare = k50_cfg->numSpare;
for (i = 0; i < RF_MAXROW; i++)
for (j = 0; j < RF_MAXCOL; j++)
k_cfg->devs[i][j] = k50_cfg->devs[i][j];
memcpy(k_cfg->devnames, k50_cfg->devnames,
sizeof(k_cfg->devnames));
for (i = 0; i < RF_MAXSPARE; i++)
k_cfg->spare_devs[i] = k50_cfg->spare_devs[i];
memcpy(k_cfg->spare_names, k50_cfg->spare_names,
sizeof(k_cfg->spare_names));
k_cfg->sectPerSU = k50_cfg->sectPerSU;
k_cfg->SUsPerPU = k50_cfg->SUsPerPU;
k_cfg->SUsPerRU = k50_cfg->SUsPerRU;
k_cfg->parityConfig = k50_cfg->parityConfig;
memcpy(k_cfg->diskQueueType, k50_cfg->diskQueueType,
sizeof(k_cfg->diskQueueType));
k_cfg->maxOutstandingDiskReqs = k50_cfg->maxOutstandingDiskReqs;
memcpy(k_cfg->debugVars, k50_cfg->debugVars,
sizeof(k_cfg->debugVars));
k_cfg->layoutSpecificSize = k50_cfg->layoutSpecificSize;
k_cfg->layoutSpecific = k50_cfg->layoutSpecific;
k_cfg->force = k50_cfg->force;
RF_Free(k50_cfg, sizeof(RF_Config50_t));
*k_cfgp = k_cfg;
return 0;
}