| 1 | /* $NetBSD: rf_pq.c,v 1.16 2009/03/14 15:36:20 dsl Exp $ */ |
| 2 | /* |
| 3 | * Copyright (c) 1995 Carnegie-Mellon University. |
| 4 | * All rights reserved. |
| 5 | * |
| 6 | * Author: Daniel Stodolsky |
| 7 | * |
| 8 | * Permission to use, copy, modify and distribute this software and |
| 9 | * its documentation is hereby granted, provided that both the copyright |
| 10 | * notice and this permission notice appear in all copies of the |
| 11 | * software, derivative works or modified versions, and any portions |
| 12 | * thereof, and that both notices appear in supporting documentation. |
| 13 | * |
| 14 | * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" |
| 15 | * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND |
| 16 | * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. |
| 17 | * |
| 18 | * Carnegie Mellon requests users of this software to return to |
| 19 | * |
| 20 | * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU |
| 21 | * School of Computer Science |
| 22 | * Carnegie Mellon University |
| 23 | * Pittsburgh PA 15213-3890 |
| 24 | * |
| 25 | * any improvements or extensions that they make and grant Carnegie the |
| 26 | * rights to redistribute these changes. |
| 27 | */ |
| 28 | |
| 29 | /* |
| 30 | * Code for RAID level 6 (P + Q) disk array architecture. |
| 31 | */ |
| 32 | |
| 33 | #include <sys/cdefs.h> |
| 34 | __KERNEL_RCSID(0, "$NetBSD: rf_pq.c,v 1.16 2009/03/14 15:36:20 dsl Exp $" ); |
| 35 | |
| 36 | #include "rf_archs.h" |
| 37 | |
| 38 | #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) || (RF_INCLUDE_EVENODD > 0) |
| 39 | |
| 40 | #include <dev/raidframe/raidframevar.h> |
| 41 | |
| 42 | #include "rf_raid.h" |
| 43 | #include "rf_dag.h" |
| 44 | #include "rf_dagffrd.h" |
| 45 | #include "rf_dagffwr.h" |
| 46 | #include "rf_dagdegrd.h" |
| 47 | #include "rf_dagdegwr.h" |
| 48 | #include "rf_dagutils.h" |
| 49 | #include "rf_dagfuncs.h" |
| 50 | #include "rf_etimer.h" |
| 51 | #include "rf_pqdeg.h" |
| 52 | #include "rf_general.h" |
| 53 | #include "rf_map.h" |
| 54 | #include "rf_pq.h" |
| 55 | |
| 56 | RF_RedFuncs_t rf_pFuncs = {rf_RegularONPFunc, "Regular Old-New P" , rf_SimpleONPFunc, "Simple Old-New P" }; |
| 57 | RF_RedFuncs_t rf_pRecoveryFuncs = {rf_RecoveryPFunc, "Recovery P Func" , rf_RecoveryPFunc, "Recovery P Func" }; |
| 58 | |
| 59 | int |
| 60 | rf_RegularONPFunc(RF_DagNode_t *node) |
| 61 | { |
| 62 | return (rf_RegularXorFunc(node)); |
| 63 | } |
| 64 | /* |
| 65 | same as simpleONQ func, but the coefficient is always 1 |
| 66 | */ |
| 67 | |
| 68 | int |
| 69 | rf_SimpleONPFunc(RF_DagNode_t *node) |
| 70 | { |
| 71 | return (rf_SimpleXorFunc(node)); |
| 72 | } |
| 73 | |
| 74 | int |
| 75 | rf_RecoveryPFunc(RF_DagNode_t *node) |
| 76 | { |
| 77 | return (rf_RecoveryXorFunc(node)); |
| 78 | } |
| 79 | |
| 80 | int |
| 81 | rf_RegularPFunc(RF_DagNode_t *node) |
| 82 | { |
| 83 | return (rf_RegularXorFunc(node)); |
| 84 | } |
| 85 | #endif /* (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) || (RF_INCLUDE_EVENODD > 0) */ |
| 86 | #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) |
| 87 | |
| 88 | static void |
| 89 | QDelta(char *dest, char *obuf, char *nbuf, unsigned length, |
| 90 | unsigned char coeff); |
| 91 | static void |
| 92 | rf_InvertQ(unsigned long *qbuf, unsigned long *abuf, |
| 93 | unsigned length, unsigned coeff); |
| 94 | |
| 95 | RF_RedFuncs_t rf_qFuncs = {rf_RegularONQFunc, "Regular Old-New Q" , rf_SimpleONQFunc, "Simple Old-New Q" }; |
| 96 | RF_RedFuncs_t rf_qRecoveryFuncs = {rf_RecoveryQFunc, "Recovery Q Func" , rf_RecoveryQFunc, "Recovery Q Func" }; |
| 97 | RF_RedFuncs_t rf_pqRecoveryFuncs = {rf_RecoveryPQFunc, "Recovery PQ Func" , rf_RecoveryPQFunc, "Recovery PQ Func" }; |
| 98 | |
| 99 | void |
| 100 | rf_PQDagSelect( |
| 101 | RF_Raid_t * raidPtr, |
| 102 | RF_IoType_t type, |
| 103 | RF_AccessStripeMap_t * asmap, |
| 104 | RF_VoidFuncPtr * createFunc) |
| 105 | { |
| 106 | RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout); |
| 107 | unsigned ndfail = asmap->numDataFailed; |
| 108 | unsigned npfail = asmap->numParityFailed; |
| 109 | unsigned ntfail = npfail + ndfail; |
| 110 | |
| 111 | RF_ASSERT(RF_IO_IS_R_OR_W(type)); |
| 112 | if (ntfail > 2) { |
| 113 | RF_ERRORMSG("more than two disks failed in a single group! Aborting I/O operation.\n" ); |
| 114 | *createFunc = NULL; |
| 115 | return; |
| 116 | } |
| 117 | /* ok, we can do this I/O */ |
| 118 | if (type == RF_IO_TYPE_READ) { |
| 119 | switch (ndfail) { |
| 120 | case 0: |
| 121 | /* fault free read */ |
| 122 | *createFunc = (RF_VoidFuncPtr) rf_CreateFaultFreeReadDAG; /* same as raid 5 */ |
| 123 | break; |
| 124 | case 1: |
| 125 | /* lost a single data unit */ |
| 126 | /* two cases: (1) parity is not lost. do a normal raid |
| 127 | * 5 reconstruct read. (2) parity is lost. do a |
| 128 | * reconstruct read using "q". */ |
| 129 | if (ntfail == 2) { /* also lost redundancy */ |
| 130 | if (asmap->failedPDAs[1]->type == RF_PDA_TYPE_PARITY) |
| 131 | *createFunc = (RF_VoidFuncPtr) rf_PQ_110_CreateReadDAG; |
| 132 | else |
| 133 | *createFunc = (RF_VoidFuncPtr) rf_PQ_101_CreateReadDAG; |
| 134 | } else { |
| 135 | /* P and Q are ok. But is there a failure in |
| 136 | * some unaccessed data unit? */ |
| 137 | if (rf_NumFailedDataUnitsInStripe(raidPtr, asmap) == 2) |
| 138 | *createFunc = (RF_VoidFuncPtr) rf_PQ_200_CreateReadDAG; |
| 139 | else |
| 140 | *createFunc = (RF_VoidFuncPtr) rf_PQ_100_CreateReadDAG; |
| 141 | } |
| 142 | break; |
| 143 | case 2: |
| 144 | /* lost two data units */ |
| 145 | *createFunc = (RF_VoidFuncPtr) rf_PQ_200_CreateReadDAG; |
| 146 | break; |
| 147 | } |
| 148 | return; |
| 149 | } |
| 150 | /* a write */ |
| 151 | switch (ntfail) { |
| 152 | case 0: /* fault free */ |
| 153 | if (rf_suppressLocksAndLargeWrites || |
| 154 | (((asmap->numStripeUnitsAccessed <= (layoutPtr->numDataCol / 2)) && (layoutPtr->numDataCol != 1)) || |
| 155 | (asmap->parityInfo->next != NULL) || (asmap->qInfo->next != NULL) || rf_CheckStripeForFailures(raidPtr, asmap))) { |
| 156 | |
| 157 | *createFunc = (RF_VoidFuncPtr) rf_PQCreateSmallWriteDAG; |
| 158 | } else { |
| 159 | *createFunc = (RF_VoidFuncPtr) rf_PQCreateLargeWriteDAG; |
| 160 | } |
| 161 | break; |
| 162 | |
| 163 | case 1: /* single disk fault */ |
| 164 | if (npfail == 1) { |
| 165 | RF_ASSERT((asmap->failedPDAs[0]->type == RF_PDA_TYPE_PARITY) || (asmap->failedPDAs[0]->type == RF_PDA_TYPE_Q)); |
| 166 | if (asmap->failedPDAs[0]->type == RF_PDA_TYPE_Q) { /* q died, treat like |
| 167 | * normal mode raid5 |
| 168 | * write. */ |
| 169 | if (((asmap->numStripeUnitsAccessed <= (layoutPtr->numDataCol / 2)) || (asmap->numStripeUnitsAccessed == 1)) |
| 170 | || rf_NumFailedDataUnitsInStripe(raidPtr, asmap)) |
| 171 | *createFunc = (RF_VoidFuncPtr) rf_PQ_001_CreateSmallWriteDAG; |
| 172 | else |
| 173 | *createFunc = (RF_VoidFuncPtr) rf_PQ_001_CreateLargeWriteDAG; |
| 174 | } else {/* parity died, small write only updating Q */ |
| 175 | if (((asmap->numStripeUnitsAccessed <= (layoutPtr->numDataCol / 2)) || (asmap->numStripeUnitsAccessed == 1)) |
| 176 | || rf_NumFailedDataUnitsInStripe(raidPtr, asmap)) |
| 177 | *createFunc = (RF_VoidFuncPtr) rf_PQ_010_CreateSmallWriteDAG; |
| 178 | else |
| 179 | *createFunc = (RF_VoidFuncPtr) rf_PQ_010_CreateLargeWriteDAG; |
| 180 | } |
| 181 | } else { /* data missing. Do a P reconstruct write if |
| 182 | * only a single data unit is lost in the |
| 183 | * stripe, otherwise a PQ reconstruct write. */ |
| 184 | if (rf_NumFailedDataUnitsInStripe(raidPtr, asmap) == 2) |
| 185 | *createFunc = (RF_VoidFuncPtr) rf_PQ_200_CreateWriteDAG; |
| 186 | else |
| 187 | *createFunc = (RF_VoidFuncPtr) rf_PQ_100_CreateWriteDAG; |
| 188 | } |
| 189 | break; |
| 190 | |
| 191 | case 2: /* two disk faults */ |
| 192 | switch (npfail) { |
| 193 | case 2: /* both p and q dead */ |
| 194 | *createFunc = (RF_VoidFuncPtr) rf_PQ_011_CreateWriteDAG; |
| 195 | break; |
| 196 | case 1: /* either p or q and dead data */ |
| 197 | RF_ASSERT(asmap->failedPDAs[0]->type == RF_PDA_TYPE_DATA); |
| 198 | RF_ASSERT((asmap->failedPDAs[1]->type == RF_PDA_TYPE_PARITY) || (asmap->failedPDAs[1]->type == RF_PDA_TYPE_Q)); |
| 199 | if (asmap->failedPDAs[1]->type == RF_PDA_TYPE_Q) |
| 200 | *createFunc = (RF_VoidFuncPtr) rf_PQ_101_CreateWriteDAG; |
| 201 | else |
| 202 | *createFunc = (RF_VoidFuncPtr) rf_PQ_110_CreateWriteDAG; |
| 203 | break; |
| 204 | case 0: /* double data loss */ |
| 205 | *createFunc = (RF_VoidFuncPtr) rf_PQ_200_CreateWriteDAG; |
| 206 | break; |
| 207 | } |
| 208 | break; |
| 209 | |
| 210 | default: /* more than 2 disk faults */ |
| 211 | *createFunc = NULL; |
| 212 | RF_PANIC(); |
| 213 | } |
| 214 | return; |
| 215 | } |
| 216 | /* |
| 217 | Used as a stop gap info function |
| 218 | */ |
| 219 | #if 0 |
| 220 | static void |
| 221 | PQOne(RF_Raid_t *raidPtr, int *nSucc, int *nAnte, RF_AccessStripeMap_t *asmap) |
| 222 | { |
| 223 | *nSucc = *nAnte = 1; |
| 224 | } |
| 225 | |
| 226 | static void |
| 227 | PQOneTwo(RF_Raid_t *raidPtr, int *nSucc, int *nAnte, RF_AccessStripeMap_t *asmap) |
| 228 | { |
| 229 | *nSucc = 1; |
| 230 | *nAnte = 2; |
| 231 | } |
| 232 | #endif |
| 233 | |
| 234 | RF_CREATE_DAG_FUNC_DECL(rf_PQCreateLargeWriteDAG) |
| 235 | { |
| 236 | rf_CommonCreateLargeWriteDAG(raidPtr, asmap, dag_h, bp, flags, allocList, 2, |
| 237 | rf_RegularPQFunc, RF_FALSE); |
| 238 | } |
| 239 | |
| 240 | int |
| 241 | rf_RegularONQFunc(RF_DagNode_t *node) |
| 242 | { |
| 243 | int np = node->numParams; |
| 244 | int d; |
| 245 | RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[np - 1].p; |
| 246 | int i; |
| 247 | RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec; |
| 248 | RF_Etimer_t timer; |
| 249 | char *qbuf, *qpbuf; |
| 250 | char *obuf, *nbuf; |
| 251 | RF_PhysDiskAddr_t *old, *new; |
| 252 | unsigned long coeff; |
| 253 | unsigned secPerSU = raidPtr->Layout.sectorsPerStripeUnit; |
| 254 | |
| 255 | RF_ETIMER_START(timer); |
| 256 | |
| 257 | d = (np - 3) / 4; |
| 258 | RF_ASSERT(4 * d + 3 == np); |
| 259 | qbuf = (char *) node->params[2 * d + 1].p; /* q buffer */ |
| 260 | for (i = 0; i < d; i++) { |
| 261 | old = (RF_PhysDiskAddr_t *) node->params[2 * i].p; |
| 262 | obuf = (char *) node->params[2 * i + 1].p; |
| 263 | new = (RF_PhysDiskAddr_t *) node->params[2 * (d + 1 + i)].p; |
| 264 | nbuf = (char *) node->params[2 * (d + 1 + i) + 1].p; |
| 265 | RF_ASSERT(new->numSector == old->numSector); |
| 266 | RF_ASSERT(new->raidAddress == old->raidAddress); |
| 267 | /* the stripe unit within the stripe tells us the coefficient |
| 268 | * to use for the multiply. */ |
| 269 | coeff = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), new->raidAddress); |
| 270 | /* compute the data unit offset within the column, then add |
| 271 | * one */ |
| 272 | coeff = (coeff % raidPtr->Layout.numDataCol); |
| 273 | qpbuf = qbuf + rf_RaidAddressToByte(raidPtr, old->startSector % secPerSU); |
| 274 | QDelta(qpbuf, obuf, nbuf, rf_RaidAddressToByte(raidPtr, old->numSector), coeff); |
| 275 | } |
| 276 | |
| 277 | RF_ETIMER_STOP(timer); |
| 278 | RF_ETIMER_EVAL(timer); |
| 279 | tracerec->q_us += RF_ETIMER_VAL_US(timer); |
| 280 | rf_GenericWakeupFunc(node, 0); /* call wake func explicitly since no |
| 281 | * I/O in this node */ |
| 282 | return (0); |
| 283 | } |
| 284 | /* |
| 285 | See the SimpleXORFunc for the difference between a simple and regular func. |
| 286 | These Q functions should be used for |
| 287 | |
| 288 | new q = Q(data,old data,old q) |
| 289 | |
| 290 | style updates and not for |
| 291 | |
| 292 | q = ( new data, new data, .... ) |
| 293 | |
| 294 | computations. |
| 295 | |
| 296 | The simple q takes 2(2d+1)+1 params, where d is the number |
| 297 | of stripes written. The order of params is |
| 298 | old data pda_0, old data buffer_0, old data pda_1, old data buffer_1, ... old data pda_d, old data buffer_d |
| 299 | [2d] old q pda_0, old q buffer |
| 300 | [2d_2] new data pda_0, new data buffer_0, ... new data pda_d, new data buffer_d |
| 301 | raidPtr |
| 302 | */ |
| 303 | |
| 304 | int |
| 305 | rf_SimpleONQFunc(RF_DagNode_t *node) |
| 306 | { |
| 307 | int np = node->numParams; |
| 308 | int d; |
| 309 | RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[np - 1].p; |
| 310 | int i; |
| 311 | RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec; |
| 312 | RF_Etimer_t timer; |
| 313 | char *qbuf; |
| 314 | char *obuf, *nbuf; |
| 315 | RF_PhysDiskAddr_t *old, *new; |
| 316 | unsigned long coeff; |
| 317 | |
| 318 | RF_ETIMER_START(timer); |
| 319 | |
| 320 | d = (np - 3) / 4; |
| 321 | RF_ASSERT(4 * d + 3 == np); |
| 322 | qbuf = (char *) node->params[2 * d + 1].p; /* q buffer */ |
| 323 | for (i = 0; i < d; i++) { |
| 324 | old = (RF_PhysDiskAddr_t *) node->params[2 * i].p; |
| 325 | obuf = (char *) node->params[2 * i + 1].p; |
| 326 | new = (RF_PhysDiskAddr_t *) node->params[2 * (d + 1 + i)].p; |
| 327 | nbuf = (char *) node->params[2 * (d + 1 + i) + 1].p; |
| 328 | RF_ASSERT(new->numSector == old->numSector); |
| 329 | RF_ASSERT(new->raidAddress == old->raidAddress); |
| 330 | /* the stripe unit within the stripe tells us the coefficient |
| 331 | * to use for the multiply. */ |
| 332 | coeff = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), new->raidAddress); |
| 333 | /* compute the data unit offset within the column, then add |
| 334 | * one */ |
| 335 | coeff = (coeff % raidPtr->Layout.numDataCol); |
| 336 | QDelta(qbuf, obuf, nbuf, rf_RaidAddressToByte(raidPtr, old->numSector), coeff); |
| 337 | } |
| 338 | |
| 339 | RF_ETIMER_STOP(timer); |
| 340 | RF_ETIMER_EVAL(timer); |
| 341 | tracerec->q_us += RF_ETIMER_VAL_US(timer); |
| 342 | rf_GenericWakeupFunc(node, 0); /* call wake func explicitly since no |
| 343 | * I/O in this node */ |
| 344 | return (0); |
| 345 | } |
| 346 | RF_CREATE_DAG_FUNC_DECL(rf_PQCreateSmallWriteDAG) |
| 347 | { |
| 348 | rf_CommonCreateSmallWriteDAG(raidPtr, asmap, dag_h, bp, flags, allocList, &rf_pFuncs, &rf_qFuncs); |
| 349 | } |
| 350 | |
| 351 | static void RegularQSubr(RF_DagNode_t *node, char *qbuf); |
| 352 | |
| 353 | static void |
| 354 | RegularQSubr(RF_DagNode_t *node, char *qbuf) |
| 355 | { |
| 356 | int np = node->numParams; |
| 357 | int d; |
| 358 | RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[np - 1].p; |
| 359 | unsigned secPerSU = raidPtr->Layout.sectorsPerStripeUnit; |
| 360 | int i; |
| 361 | RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec; |
| 362 | RF_Etimer_t timer; |
| 363 | char *obuf, *qpbuf; |
| 364 | RF_PhysDiskAddr_t *old; |
| 365 | unsigned long coeff; |
| 366 | |
| 367 | RF_ETIMER_START(timer); |
| 368 | |
| 369 | d = (np - 1) / 2; |
| 370 | RF_ASSERT(2 * d + 1 == np); |
| 371 | for (i = 0; i < d; i++) { |
| 372 | old = (RF_PhysDiskAddr_t *) node->params[2 * i].p; |
| 373 | obuf = (char *) node->params[2 * i + 1].p; |
| 374 | coeff = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), old->raidAddress); |
| 375 | /* compute the data unit offset within the column, then add |
| 376 | * one */ |
| 377 | coeff = (coeff % raidPtr->Layout.numDataCol); |
| 378 | /* the input buffers may not all be aligned with the start of |
| 379 | * the stripe. so shift by their sector offset within the |
| 380 | * stripe unit */ |
| 381 | qpbuf = qbuf + rf_RaidAddressToByte(raidPtr, old->startSector % secPerSU); |
| 382 | rf_IncQ((unsigned long *) qpbuf, (unsigned long *) obuf, rf_RaidAddressToByte(raidPtr, old->numSector), coeff); |
| 383 | } |
| 384 | |
| 385 | RF_ETIMER_STOP(timer); |
| 386 | RF_ETIMER_EVAL(timer); |
| 387 | tracerec->q_us += RF_ETIMER_VAL_US(timer); |
| 388 | } |
| 389 | /* |
| 390 | used in degraded writes. |
| 391 | */ |
| 392 | |
| 393 | static void DegrQSubr(RF_DagNode_t *node); |
| 394 | |
| 395 | static void |
| 396 | DegrQSubr(RF_DagNode_t *node) |
| 397 | { |
| 398 | int np = node->numParams; |
| 399 | int d; |
| 400 | RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[np - 1].p; |
| 401 | unsigned secPerSU = raidPtr->Layout.sectorsPerStripeUnit; |
| 402 | int i; |
| 403 | RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec; |
| 404 | RF_Etimer_t timer; |
| 405 | char *qbuf = node->results[1]; |
| 406 | char *obuf, *qpbuf; |
| 407 | RF_PhysDiskAddr_t *old; |
| 408 | unsigned long coeff; |
| 409 | unsigned fail_start; |
| 410 | int j; |
| 411 | |
| 412 | old = (RF_PhysDiskAddr_t *) node->params[np - 2].p; |
| 413 | fail_start = old->startSector % secPerSU; |
| 414 | |
| 415 | RF_ETIMER_START(timer); |
| 416 | |
| 417 | d = (np - 2) / 2; |
| 418 | RF_ASSERT(2 * d + 2 == np); |
| 419 | for (i = 0; i < d; i++) { |
| 420 | old = (RF_PhysDiskAddr_t *) node->params[2 * i].p; |
| 421 | obuf = (char *) node->params[2 * i + 1].p; |
| 422 | coeff = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), old->raidAddress); |
| 423 | /* compute the data unit offset within the column, then add |
| 424 | * one */ |
| 425 | coeff = (coeff % raidPtr->Layout.numDataCol); |
| 426 | /* the input buffers may not all be aligned with the start of |
| 427 | * the stripe. so shift by their sector offset within the |
| 428 | * stripe unit */ |
| 429 | j = old->startSector % secPerSU; |
| 430 | RF_ASSERT(j >= fail_start); |
| 431 | qpbuf = qbuf + rf_RaidAddressToByte(raidPtr, j - fail_start); |
| 432 | rf_IncQ((unsigned long *) qpbuf, (unsigned long *) obuf, rf_RaidAddressToByte(raidPtr, old->numSector), coeff); |
| 433 | } |
| 434 | |
| 435 | RF_ETIMER_STOP(timer); |
| 436 | RF_ETIMER_EVAL(timer); |
| 437 | tracerec->q_us += RF_ETIMER_VAL_US(timer); |
| 438 | } |
| 439 | /* |
| 440 | Called by large write code to compute the new parity and the new q. |
| 441 | |
| 442 | structure of the params: |
| 443 | |
| 444 | pda_0, buffer_0, pda_1 , buffer_1, ... , pda_d, buffer_d ( d = numDataCol |
| 445 | raidPtr |
| 446 | |
| 447 | for a total of 2d+1 arguments. |
| 448 | The result buffers results[0], results[1] are the buffers for the p and q, |
| 449 | respectively. |
| 450 | |
| 451 | We compute Q first, then compute P. The P calculation may try to reuse |
| 452 | one of the input buffers for its output, so if we computed P first, we would |
| 453 | corrupt the input for the q calculation. |
| 454 | */ |
| 455 | |
| 456 | int |
| 457 | rf_RegularPQFunc(RF_DagNode_t *node) |
| 458 | { |
| 459 | RegularQSubr(node, node->results[1]); |
| 460 | return (rf_RegularXorFunc(node)); /* does the wakeup */ |
| 461 | } |
| 462 | |
| 463 | int |
| 464 | rf_RegularQFunc(RF_DagNode_t *node) |
| 465 | { |
| 466 | /* Almost ... adjust Qsubr args */ |
| 467 | RegularQSubr(node, node->results[0]); |
| 468 | rf_GenericWakeupFunc(node, 0); /* call wake func explicitly since no |
| 469 | * I/O in this node */ |
| 470 | return (0); |
| 471 | } |
| 472 | /* |
| 473 | Called by singly degraded write code to compute the new parity and the new q. |
| 474 | |
| 475 | structure of the params: |
| 476 | |
| 477 | pda_0, buffer_0, pda_1 , buffer_1, ... , pda_d, buffer_d |
| 478 | failedPDA raidPtr |
| 479 | |
| 480 | for a total of 2d+2 arguments. |
| 481 | The result buffers results[0], results[1] are the buffers for the parity and q, |
| 482 | respectively. |
| 483 | |
| 484 | We compute Q first, then compute parity. The parity calculation may try to reuse |
| 485 | one of the input buffers for its output, so if we computed parity first, we would |
| 486 | corrupt the input for the q calculation. |
| 487 | |
| 488 | We treat this identically to the regularPQ case, ignoring the failedPDA extra argument. |
| 489 | */ |
| 490 | |
| 491 | void |
| 492 | rf_Degraded_100_PQFunc(RF_DagNode_t *node) |
| 493 | { |
| 494 | int np = node->numParams; |
| 495 | |
| 496 | RF_ASSERT(np >= 2); |
| 497 | DegrQSubr(node); |
| 498 | rf_RecoveryXorFunc(node); |
| 499 | } |
| 500 | |
| 501 | |
| 502 | /* |
| 503 | The two below are used when reading a stripe with a single lost data unit. |
| 504 | The parameters are |
| 505 | |
| 506 | pda_0, buffer_0, .... pda_n, buffer_n, P pda, P buffer, failedPDA, raidPtr |
| 507 | |
| 508 | and results[0] contains the data buffer. Which is originally zero-filled. |
| 509 | |
| 510 | */ |
| 511 | |
| 512 | /* this Q func is used by the degraded-mode dag functions to recover lost data. |
| 513 | * the second-to-last parameter is the PDA for the failed portion of the access. |
| 514 | * the code here looks at this PDA and assumes that the xor target buffer is |
| 515 | * equal in size to the number of sectors in the failed PDA. It then uses |
| 516 | * the other PDAs in the parameter list to determine where within the target |
| 517 | * buffer the corresponding data should be xored. |
| 518 | * |
| 519 | * Recall the basic equation is |
| 520 | * |
| 521 | * Q = ( data_1 + 2 * data_2 ... + k * data_k ) mod 256 |
| 522 | * |
| 523 | * so to recover data_j we need |
| 524 | * |
| 525 | * J data_j = (Q - data_1 - 2 data_2 ....- k* data_k) mod 256 |
| 526 | * |
| 527 | * So the coefficient for each buffer is (255 - data_col), and j should be initialized by |
| 528 | * copying Q into it. Then we need to do a table lookup to convert to solve |
| 529 | * data_j /= J |
| 530 | * |
| 531 | * |
| 532 | */ |
| 533 | int |
| 534 | rf_RecoveryQFunc(RF_DagNode_t *node) |
| 535 | { |
| 536 | RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p; |
| 537 | RF_RaidLayout_t *layoutPtr = (RF_RaidLayout_t *) & raidPtr->Layout; |
| 538 | RF_PhysDiskAddr_t *failedPDA = (RF_PhysDiskAddr_t *) node->params[node->numParams - 2].p; |
| 539 | int i; |
| 540 | RF_PhysDiskAddr_t *pda; |
| 541 | RF_RaidAddr_t suoffset, failedSUOffset = rf_StripeUnitOffset(layoutPtr, failedPDA->startSector); |
| 542 | char *srcbuf, *destbuf; |
| 543 | RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec; |
| 544 | RF_Etimer_t timer; |
| 545 | unsigned long coeff; |
| 546 | |
| 547 | RF_ETIMER_START(timer); |
| 548 | /* start by copying Q into the buffer */ |
| 549 | memcpy(node->results[0], node->params[node->numParams - 3].p, |
| 550 | rf_RaidAddressToByte(raidPtr, failedPDA->numSector)); |
| 551 | for (i = 0; i < node->numParams - 4; i += 2) { |
| 552 | RF_ASSERT(node->params[i + 1].p != node->results[0]); |
| 553 | pda = (RF_PhysDiskAddr_t *) node->params[i].p; |
| 554 | srcbuf = (char *) node->params[i + 1].p; |
| 555 | suoffset = rf_StripeUnitOffset(layoutPtr, pda->startSector); |
| 556 | destbuf = ((char *) node->results[0]) + rf_RaidAddressToByte(raidPtr, suoffset - failedSUOffset); |
| 557 | coeff = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), pda->raidAddress); |
| 558 | /* compute the data unit offset within the column */ |
| 559 | coeff = (coeff % raidPtr->Layout.numDataCol); |
| 560 | rf_IncQ((unsigned long *) destbuf, (unsigned long *) srcbuf, rf_RaidAddressToByte(raidPtr, pda->numSector), coeff); |
| 561 | } |
| 562 | /* Do the nasty inversion now */ |
| 563 | coeff = (rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), failedPDA->startSector) % raidPtr->Layout.numDataCol); |
| 564 | rf_InvertQ(node->results[0], node->results[0], rf_RaidAddressToByte(raidPtr, pda->numSector), coeff); |
| 565 | RF_ETIMER_STOP(timer); |
| 566 | RF_ETIMER_EVAL(timer); |
| 567 | tracerec->q_us += RF_ETIMER_VAL_US(timer); |
| 568 | rf_GenericWakeupFunc(node, 0); |
| 569 | return (0); |
| 570 | } |
| 571 | |
| 572 | int |
| 573 | rf_RecoveryPQFunc(RF_DagNode_t *node) |
| 574 | { |
| 575 | RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[node->numParams - 1].p; |
| 576 | printf("raid%d: Recovery from PQ not implemented.\n" ,raidPtr->raidid); |
| 577 | return (1); |
| 578 | } |
| 579 | /* |
| 580 | Degraded write Q subroutine. |
| 581 | Used when P is dead. |
| 582 | Large-write style Q computation. |
| 583 | Parameters |
| 584 | |
| 585 | (pda,buf),(pda,buf),.....,(failedPDA,bufPtr),failedPDA,raidPtr. |
| 586 | |
| 587 | We ignore failedPDA. |
| 588 | |
| 589 | This is a "simple style" recovery func. |
| 590 | */ |
| 591 | |
| 592 | void |
| 593 | rf_PQ_DegradedWriteQFunc(RF_DagNode_t *node) |
| 594 | { |
| 595 | int np = node->numParams; |
| 596 | int d; |
| 597 | RF_Raid_t *raidPtr = (RF_Raid_t *) node->params[np - 1].p; |
| 598 | unsigned secPerSU = raidPtr->Layout.sectorsPerStripeUnit; |
| 599 | int i; |
| 600 | RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec; |
| 601 | RF_Etimer_t timer; |
| 602 | char *qbuf = node->results[0]; |
| 603 | char *obuf, *qpbuf; |
| 604 | RF_PhysDiskAddr_t *old; |
| 605 | unsigned long coeff; |
| 606 | int fail_start, j; |
| 607 | |
| 608 | old = (RF_PhysDiskAddr_t *) node->params[np - 2].p; |
| 609 | fail_start = old->startSector % secPerSU; |
| 610 | |
| 611 | RF_ETIMER_START(timer); |
| 612 | |
| 613 | d = (np - 2) / 2; |
| 614 | RF_ASSERT(2 * d + 2 == np); |
| 615 | |
| 616 | for (i = 0; i < d; i++) { |
| 617 | old = (RF_PhysDiskAddr_t *) node->params[2 * i].p; |
| 618 | obuf = (char *) node->params[2 * i + 1].p; |
| 619 | coeff = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), old->raidAddress); |
| 620 | /* compute the data unit offset within the column, then add |
| 621 | * one */ |
| 622 | coeff = (coeff % raidPtr->Layout.numDataCol); |
| 623 | j = old->startSector % secPerSU; |
| 624 | RF_ASSERT(j >= fail_start); |
| 625 | qpbuf = qbuf + rf_RaidAddressToByte(raidPtr, j - fail_start); |
| 626 | rf_IncQ((unsigned long *) qpbuf, (unsigned long *) obuf, rf_RaidAddressToByte(raidPtr, old->numSector), coeff); |
| 627 | } |
| 628 | |
| 629 | RF_ETIMER_STOP(timer); |
| 630 | RF_ETIMER_EVAL(timer); |
| 631 | tracerec->q_us += RF_ETIMER_VAL_US(timer); |
| 632 | rf_GenericWakeupFunc(node, 0); |
| 633 | } |
| 634 | |
| 635 | |
| 636 | |
| 637 | |
| 638 | /* Q computations */ |
| 639 | |
| 640 | /* |
| 641 | coeff - colummn; |
| 642 | |
| 643 | compute dest ^= qfor[28-coeff][rn[coeff+1] a] |
| 644 | |
| 645 | on 5-bit basis; |
| 646 | length in bytes; |
| 647 | */ |
| 648 | |
| 649 | void |
| 650 | rf_IncQ(unsigned long *dest, unsigned long *buf, unsigned length, unsigned coeff) |
| 651 | { |
| 652 | unsigned long a, d, new; |
| 653 | unsigned long a1, a2; |
| 654 | unsigned int *q = &(rf_qfor[28 - coeff][0]); |
| 655 | unsigned r = rf_rn[coeff + 1]; |
| 656 | |
| 657 | #define EXTRACT(a,i) ((a >> (5L*i)) & 0x1f) |
| 658 | #define INSERT(a,i) (a << (5L*i)) |
| 659 | |
| 660 | length /= 8; |
| 661 | /* 13 5 bit quants in a 64 bit word */ |
| 662 | while (length) { |
| 663 | a = *buf++; |
| 664 | d = *dest; |
| 665 | a1 = EXTRACT(a, 0) ^ r; |
| 666 | a2 = EXTRACT(a, 1) ^ r; |
| 667 | new = INSERT(a2, 1) | a1; |
| 668 | a1 = EXTRACT(a, 2) ^ r; |
| 669 | a2 = EXTRACT(a, 3) ^ r; |
| 670 | a1 = q[a1]; |
| 671 | a2 = q[a2]; |
| 672 | new = new | INSERT(a1, 2) | INSERT(a2, 3); |
| 673 | a1 = EXTRACT(a, 4) ^ r; |
| 674 | a2 = EXTRACT(a, 5) ^ r; |
| 675 | a1 = q[a1]; |
| 676 | a2 = q[a2]; |
| 677 | new = new | INSERT(a1, 4) | INSERT(a2, 5); |
| 678 | a1 = EXTRACT(a, 5) ^ r; |
| 679 | a2 = EXTRACT(a, 6) ^ r; |
| 680 | a1 = q[a1]; |
| 681 | a2 = q[a2]; |
| 682 | new = new | INSERT(a1, 5) | INSERT(a2, 6); |
| 683 | #if RF_LONGSHIFT > 2 |
| 684 | a1 = EXTRACT(a, 7) ^ r; |
| 685 | a2 = EXTRACT(a, 8) ^ r; |
| 686 | a1 = q[a1]; |
| 687 | a2 = q[a2]; |
| 688 | new = new | INSERT(a1, 7) | INSERT(a2, 8); |
| 689 | a1 = EXTRACT(a, 9) ^ r; |
| 690 | a2 = EXTRACT(a, 10) ^ r; |
| 691 | a1 = q[a1]; |
| 692 | a2 = q[a2]; |
| 693 | new = new | INSERT(a1, 9) | INSERT(a2, 10); |
| 694 | a1 = EXTRACT(a, 11) ^ r; |
| 695 | a2 = EXTRACT(a, 12) ^ r; |
| 696 | a1 = q[a1]; |
| 697 | a2 = q[a2]; |
| 698 | new = new | INSERT(a1, 11) | INSERT(a2, 12); |
| 699 | #endif /* RF_LONGSHIFT > 2 */ |
| 700 | d ^= new; |
| 701 | *dest++ = d; |
| 702 | length--; |
| 703 | } |
| 704 | } |
| 705 | /* |
| 706 | compute |
| 707 | |
| 708 | dest ^= rf_qfor[28-coeff][rf_rn[coeff+1] (old^new) ] |
| 709 | |
| 710 | on a five bit basis. |
| 711 | optimization: compute old ^ new on 64 bit basis. |
| 712 | |
| 713 | length in bytes. |
| 714 | */ |
| 715 | |
| 716 | static void |
| 717 | QDelta( |
| 718 | char *dest, |
| 719 | char *obuf, |
| 720 | char *nbuf, |
| 721 | unsigned length, |
| 722 | unsigned char coeff) |
| 723 | { |
| 724 | unsigned long a, d, new; |
| 725 | unsigned long a1, a2; |
| 726 | unsigned int *q = &(rf_qfor[28 - coeff][0]); |
| 727 | unsigned int r = rf_rn[coeff + 1]; |
| 728 | |
| 729 | r = a1 = a2 = new = d = a = 0; /* XXX for now... */ |
| 730 | q = NULL; /* XXX for now */ |
| 731 | |
| 732 | #ifdef _KERNEL |
| 733 | /* PQ in kernel currently not supported because the encoding/decoding |
| 734 | * table is not present */ |
| 735 | memset(dest, 0, length); |
| 736 | #else /* KERNEL */ |
| 737 | /* this code probably doesn't work and should be rewritten -wvcii */ |
| 738 | /* 13 5 bit quants in a 64 bit word */ |
| 739 | length /= 8; |
| 740 | while (length) { |
| 741 | a = *obuf++; /* XXX need to reorg to avoid cache conflicts */ |
| 742 | a ^= *nbuf++; |
| 743 | d = *dest; |
| 744 | a1 = EXTRACT(a, 0) ^ r; |
| 745 | a2 = EXTRACT(a, 1) ^ r; |
| 746 | a1 = q[a1]; |
| 747 | a2 = q[a2]; |
| 748 | new = INSERT(a2, 1) | a1; |
| 749 | a1 = EXTRACT(a, 2) ^ r; |
| 750 | a2 = EXTRACT(a, 3) ^ r; |
| 751 | a1 = q[a1]; |
| 752 | a2 = q[a2]; |
| 753 | new = new | INSERT(a1, 2) | INSERT(a2, 3); |
| 754 | a1 = EXTRACT(a, 4) ^ r; |
| 755 | a2 = EXTRACT(a, 5) ^ r; |
| 756 | a1 = q[a1]; |
| 757 | a2 = q[a2]; |
| 758 | new = new | INSERT(a1, 4) | INSERT(a2, 5); |
| 759 | a1 = EXTRACT(a, 5) ^ r; |
| 760 | a2 = EXTRACT(a, 6) ^ r; |
| 761 | a1 = q[a1]; |
| 762 | a2 = q[a2]; |
| 763 | new = new | INSERT(a1, 5) | INSERT(a2, 6); |
| 764 | #if RF_LONGSHIFT > 2 |
| 765 | a1 = EXTRACT(a, 7) ^ r; |
| 766 | a2 = EXTRACT(a, 8) ^ r; |
| 767 | a1 = q[a1]; |
| 768 | a2 = q[a2]; |
| 769 | new = new | INSERT(a1, 7) | INSERT(a2, 8); |
| 770 | a1 = EXTRACT(a, 9) ^ r; |
| 771 | a2 = EXTRACT(a, 10) ^ r; |
| 772 | a1 = q[a1]; |
| 773 | a2 = q[a2]; |
| 774 | new = new | INSERT(a1, 9) | INSERT(a2, 10); |
| 775 | a1 = EXTRACT(a, 11) ^ r; |
| 776 | a2 = EXTRACT(a, 12) ^ r; |
| 777 | a1 = q[a1]; |
| 778 | a2 = q[a2]; |
| 779 | new = new | INSERT(a1, 11) | INSERT(a2, 12); |
| 780 | #endif /* RF_LONGSHIFT > 2 */ |
| 781 | d ^= new; |
| 782 | *dest++ = d; |
| 783 | length--; |
| 784 | } |
| 785 | #endif /* _KERNEL */ |
| 786 | } |
| 787 | /* |
| 788 | recover columns a and b from the given p and q into |
| 789 | bufs abuf and bbuf. All bufs are word aligned. |
| 790 | Length is in bytes. |
| 791 | */ |
| 792 | |
| 793 | |
| 794 | /* |
| 795 | * XXX |
| 796 | * |
| 797 | * Everything about this seems wrong. |
| 798 | */ |
| 799 | void |
| 800 | rf_PQ_recover(unsigned long *pbuf, unsigned long *qbuf, unsigned long *abuf, unsigned long *bbuf, unsigned length, unsigned coeff_a, unsigned coeff_b) |
| 801 | { |
| 802 | unsigned long p, q, a, a0, a1; |
| 803 | int col = (29 * coeff_a) + coeff_b; |
| 804 | unsigned char *q0 = &(rf_qinv[col][0]); |
| 805 | |
| 806 | length /= 8; |
| 807 | while (length) { |
| 808 | p = *pbuf++; |
| 809 | q = *qbuf++; |
| 810 | a0 = EXTRACT(p, 0); |
| 811 | a1 = EXTRACT(q, 0); |
| 812 | a = q0[a0 << 5 | a1]; |
| 813 | #define MF(i) \ |
| 814 | a0 = EXTRACT(p,i); \ |
| 815 | a1 = EXTRACT(q,i); \ |
| 816 | a = a | INSERT(q0[a0<<5 | a1],i) |
| 817 | |
| 818 | MF(1); |
| 819 | MF(2); |
| 820 | MF(3); |
| 821 | MF(4); |
| 822 | MF(5); |
| 823 | MF(6); |
| 824 | #if 0 |
| 825 | MF(7); |
| 826 | MF(8); |
| 827 | MF(9); |
| 828 | MF(10); |
| 829 | MF(11); |
| 830 | MF(12); |
| 831 | #endif /* 0 */ |
| 832 | *abuf++ = a; |
| 833 | *bbuf++ = a ^ p; |
| 834 | length--; |
| 835 | } |
| 836 | } |
| 837 | /* |
| 838 | Lost parity and a data column. Recover that data column. |
| 839 | Assume col coeff is lost. Let q the contents of Q after |
| 840 | all surviving data columns have been q-xored out of it. |
| 841 | Then we have the equation |
| 842 | |
| 843 | q[28-coeff][a_i ^ r_i+1] = q |
| 844 | |
| 845 | but q is cyclic with period 31. |
| 846 | So q[3+coeff][q[28-coeff][a_i ^ r_{i+1}]] = |
| 847 | q[31][a_i ^ r_{i+1}] = a_i ^ r_{i+1} . |
| 848 | |
| 849 | so a_i = r_{coeff+1} ^ q[3+coeff][q] |
| 850 | |
| 851 | The routine is passed q buffer and the buffer |
| 852 | the data is to be recoverd into. They can be the same. |
| 853 | */ |
| 854 | |
| 855 | |
| 856 | |
| 857 | static void |
| 858 | rf_InvertQ( |
| 859 | unsigned long *qbuf, |
| 860 | unsigned long *abuf, |
| 861 | unsigned length, |
| 862 | unsigned coeff) |
| 863 | { |
| 864 | unsigned long a, new; |
| 865 | unsigned long a1, a2; |
| 866 | unsigned int *q = &(rf_qfor[3 + coeff][0]); |
| 867 | unsigned r = rf_rn[coeff + 1]; |
| 868 | |
| 869 | /* 13 5 bit quants in a 64 bit word */ |
| 870 | length /= 8; |
| 871 | while (length) { |
| 872 | a = *qbuf++; |
| 873 | a1 = EXTRACT(a, 0); |
| 874 | a2 = EXTRACT(a, 1); |
| 875 | a1 = r ^ q[a1]; |
| 876 | a2 = r ^ q[a2]; |
| 877 | new = INSERT(a2, 1) | a1; |
| 878 | #define M(i,j) \ |
| 879 | a1 = EXTRACT(a,i); \ |
| 880 | a2 = EXTRACT(a,j); \ |
| 881 | a1 = r ^ q[a1]; \ |
| 882 | a2 = r ^ q[a2]; \ |
| 883 | new = new | INSERT(a1,i) | INSERT(a2,j) |
| 884 | |
| 885 | M(2, 3); |
| 886 | M(4, 5); |
| 887 | M(5, 6); |
| 888 | #if RF_LONGSHIFT > 2 |
| 889 | M(7, 8); |
| 890 | M(9, 10); |
| 891 | M(11, 12); |
| 892 | #endif /* RF_LONGSHIFT > 2 */ |
| 893 | *abuf++ = new; |
| 894 | length--; |
| 895 | } |
| 896 | } |
| 897 | #endif /* (RF_INCLUDE_DECL_PQ > 0) || |
| 898 | * (RF_INCLUDE_RAID6 > 0) */ |
| 899 | |