1 /* $NetBSD: rf_parityscan.c,v 1.10 2001/07/18 06:45:34 thorpej Exp $ */ 2 /* 3 * Copyright (c) 1995 Carnegie-Mellon University. 4 * All rights reserved. 5 * 6 * Author: Mark Holland 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 * 31 * rf_parityscan.c -- misc utilities related to parity verification 32 * 33 *****************************************************************************/ 34 35 #include "rf_types.h" 36 #include "rf_raid.h" 37 #include "rf_dag.h" 38 #include "rf_dagfuncs.h" 39 #include "rf_dagutils.h" 40 #include "rf_mcpair.h" 41 #include "rf_general.h" 42 #include "rf_engine.h" 43 #include "rf_parityscan.h" 44 #include "rf_map.h" 45 46 /***************************************************************************************** 47 * 48 * walk through the entire arry and write new parity. 49 * This works by creating two DAGs, one to read a stripe of data and one to 50 * write new parity. The first is executed, the data is xored together, and 51 * then the second is executed. To avoid constantly building and tearing down 52 * the DAGs, we create them a priori and fill them in with the mapping 53 * information as we go along. 54 * 55 * there should never be more than one thread running this. 56 * 57 ****************************************************************************************/ 58 59 int 60 rf_RewriteParity(raidPtr) 61 RF_Raid_t *raidPtr; 62 { 63 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout; 64 RF_AccessStripeMapHeader_t *asm_h; 65 int ret_val; 66 int rc; 67 RF_PhysDiskAddr_t pda; 68 RF_SectorNum_t i; 69 70 if (raidPtr->Layout.map->faultsTolerated == 0) { 71 /* There isn't any parity. Call it "okay." */ 72 return (RF_PARITY_OKAY); 73 } 74 if (raidPtr->status[0] != rf_rs_optimal) { 75 /* 76 * We're in degraded mode. Don't try to verify parity now! 77 * XXX: this should be a "we don't want to", not a 78 * "we can't" error. 79 */ 80 return (RF_PARITY_COULD_NOT_VERIFY); 81 } 82 83 ret_val = 0; 84 85 pda.startSector = 0; 86 pda.numSector = raidPtr->Layout.sectorsPerStripeUnit; 87 rc = RF_PARITY_OKAY; 88 89 for (i = 0; i < raidPtr->totalSectors && 90 rc <= RF_PARITY_CORRECTED; 91 i += layoutPtr->dataSectorsPerStripe) { 92 if (raidPtr->waitShutdown) { 93 /* Someone is pulling the plug on this set... 94 abort the re-write */ 95 return (1); 96 } 97 asm_h = rf_MapAccess(raidPtr, i, 98 layoutPtr->dataSectorsPerStripe, 99 NULL, RF_DONT_REMAP); 100 raidPtr->parity_rewrite_stripes_done = 101 i / layoutPtr->dataSectorsPerStripe ; 102 rc = rf_VerifyParity(raidPtr, asm_h->stripeMap, 1, 0); 103 104 switch (rc) { 105 case RF_PARITY_OKAY: 106 case RF_PARITY_CORRECTED: 107 break; 108 case RF_PARITY_BAD: 109 printf("Parity bad during correction\n"); 110 ret_val = 1; 111 break; 112 case RF_PARITY_COULD_NOT_CORRECT: 113 printf("Could not correct bad parity\n"); 114 ret_val = 1; 115 break; 116 case RF_PARITY_COULD_NOT_VERIFY: 117 printf("Could not verify parity\n"); 118 ret_val = 1; 119 break; 120 default: 121 printf("Bad rc=%d from VerifyParity in RewriteParity\n", rc); 122 ret_val = 1; 123 } 124 rf_FreeAccessStripeMap(asm_h); 125 } 126 return (ret_val); 127 } 128 /***************************************************************************************** 129 * 130 * verify that the parity in a particular stripe is correct. 131 * we validate only the range of parity defined by parityPDA, since 132 * this is all we have locked. The way we do this is to create an asm 133 * that maps the whole stripe and then range-restrict it to the parity 134 * region defined by the parityPDA. 135 * 136 ****************************************************************************************/ 137 int 138 rf_VerifyParity(raidPtr, aasm, correct_it, flags) 139 RF_Raid_t *raidPtr; 140 RF_AccessStripeMap_t *aasm; 141 int correct_it; 142 RF_RaidAccessFlags_t flags; 143 { 144 RF_PhysDiskAddr_t *parityPDA; 145 RF_AccessStripeMap_t *doasm; 146 RF_LayoutSW_t *lp; 147 int lrc, rc; 148 149 lp = raidPtr->Layout.map; 150 if (lp->faultsTolerated == 0) { 151 /* 152 * There isn't any parity. Call it "okay." 153 */ 154 return (RF_PARITY_OKAY); 155 } 156 rc = RF_PARITY_OKAY; 157 if (lp->VerifyParity) { 158 for (doasm = aasm; doasm; doasm = doasm->next) { 159 for (parityPDA = doasm->parityInfo; parityPDA; 160 parityPDA = parityPDA->next) { 161 lrc = lp->VerifyParity(raidPtr, 162 doasm->raidAddress, 163 parityPDA, 164 correct_it, flags); 165 if (lrc > rc) { 166 /* see rf_parityscan.h for why this 167 * works */ 168 rc = lrc; 169 } 170 } 171 } 172 } else { 173 rc = RF_PARITY_COULD_NOT_VERIFY; 174 } 175 return (rc); 176 } 177 178 int 179 rf_VerifyParityBasic(raidPtr, raidAddr, parityPDA, correct_it, flags) 180 RF_Raid_t *raidPtr; 181 RF_RaidAddr_t raidAddr; 182 RF_PhysDiskAddr_t *parityPDA; 183 int correct_it; 184 RF_RaidAccessFlags_t flags; 185 { 186 RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout); 187 RF_RaidAddr_t startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, 188 raidAddr); 189 RF_SectorCount_t numsector = parityPDA->numSector; 190 int numbytes = rf_RaidAddressToByte(raidPtr, numsector); 191 int bytesPerStripe = numbytes * layoutPtr->numDataCol; 192 RF_DagHeader_t *rd_dag_h, *wr_dag_h; /* read, write dag */ 193 RF_DagNode_t *blockNode, *unblockNode, *wrBlock, *wrUnblock; 194 RF_AccessStripeMapHeader_t *asm_h; 195 RF_AccessStripeMap_t *asmap; 196 RF_AllocListElem_t *alloclist; 197 RF_PhysDiskAddr_t *pda; 198 char *pbuf, *buf, *end_p, *p; 199 int i, retcode; 200 RF_ReconUnitNum_t which_ru; 201 RF_StripeNum_t psID = rf_RaidAddressToParityStripeID(layoutPtr, 202 raidAddr, 203 &which_ru); 204 int stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol; 205 RF_AccTraceEntry_t tracerec; 206 RF_MCPair_t *mcpair; 207 208 retcode = RF_PARITY_OKAY; 209 210 mcpair = rf_AllocMCPair(); 211 rf_MakeAllocList(alloclist); 212 RF_MallocAndAdd(buf, numbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol), (char *), alloclist); 213 RF_CallocAndAdd(pbuf, 1, numbytes, (char *), alloclist); /* use calloc to make 214 * sure buffer is zeroed */ 215 end_p = buf + bytesPerStripe; 216 217 rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, numbytes, buf, rf_DiskReadFunc, rf_DiskReadUndoFunc, 218 "Rod", alloclist, flags, RF_IO_NORMAL_PRIORITY); 219 blockNode = rd_dag_h->succedents[0]; 220 unblockNode = blockNode->succedents[0]->succedents[0]; 221 222 /* map the stripe and fill in the PDAs in the dag */ 223 asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe, buf, RF_DONT_REMAP); 224 asmap = asm_h->stripeMap; 225 226 for (pda = asmap->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) { 227 RF_ASSERT(pda); 228 rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1); 229 RF_ASSERT(pda->numSector != 0); 230 if (rf_TryToRedirectPDA(raidPtr, pda, 0)) 231 goto out; /* no way to verify parity if disk is 232 * dead. return w/ good status */ 233 blockNode->succedents[i]->params[0].p = pda; 234 blockNode->succedents[i]->params[2].v = psID; 235 blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 236 } 237 238 RF_ASSERT(!asmap->parityInfo->next); 239 rf_RangeRestrictPDA(raidPtr, parityPDA, asmap->parityInfo, 0, 1); 240 RF_ASSERT(asmap->parityInfo->numSector != 0); 241 if (rf_TryToRedirectPDA(raidPtr, asmap->parityInfo, 1)) 242 goto out; 243 blockNode->succedents[layoutPtr->numDataCol]->params[0].p = asmap->parityInfo; 244 245 /* fire off the DAG */ 246 memset((char *) &tracerec, 0, sizeof(tracerec)); 247 rd_dag_h->tracerec = &tracerec; 248 249 if (rf_verifyParityDebug) { 250 printf("Parity verify read dag:\n"); 251 rf_PrintDAGList(rd_dag_h); 252 } 253 RF_LOCK_MUTEX(mcpair->mutex); 254 mcpair->flag = 0; 255 rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 256 (void *) mcpair); 257 while (!mcpair->flag) 258 RF_WAIT_COND(mcpair->cond, mcpair->mutex); 259 RF_UNLOCK_MUTEX(mcpair->mutex); 260 if (rd_dag_h->status != rf_enable) { 261 RF_ERRORMSG("Unable to verify parity: can't read the stripe\n"); 262 retcode = RF_PARITY_COULD_NOT_VERIFY; 263 goto out; 264 } 265 for (p = buf; p < end_p; p += numbytes) { 266 rf_bxor(p, pbuf, numbytes, NULL); 267 } 268 for (i = 0; i < numbytes; i++) { 269 #if 0 270 if (pbuf[i] != 0 || buf[bytesPerStripe + i] != 0) { 271 printf("Bytes: %d %d %d\n", i, pbuf[i], buf[bytesPerStripe + i]); 272 } 273 #endif 274 if (pbuf[i] != buf[bytesPerStripe + i]) { 275 if (!correct_it) 276 RF_ERRORMSG3("Parity verify error: byte %d of parity is 0x%x should be 0x%x\n", 277 i, (u_char) buf[bytesPerStripe + i], (u_char) pbuf[i]); 278 retcode = RF_PARITY_BAD; 279 break; 280 } 281 } 282 283 if (retcode && correct_it) { 284 wr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, numbytes, pbuf, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, 285 "Wnp", alloclist, flags, RF_IO_NORMAL_PRIORITY); 286 wrBlock = wr_dag_h->succedents[0]; 287 wrUnblock = wrBlock->succedents[0]->succedents[0]; 288 wrBlock->succedents[0]->params[0].p = asmap->parityInfo; 289 wrBlock->succedents[0]->params[2].v = psID; 290 wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 291 memset((char *) &tracerec, 0, sizeof(tracerec)); 292 wr_dag_h->tracerec = &tracerec; 293 if (rf_verifyParityDebug) { 294 printf("Parity verify write dag:\n"); 295 rf_PrintDAGList(wr_dag_h); 296 } 297 RF_LOCK_MUTEX(mcpair->mutex); 298 mcpair->flag = 0; 299 rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 300 (void *) mcpair); 301 while (!mcpair->flag) 302 RF_WAIT_COND(mcpair->cond, mcpair->mutex); 303 RF_UNLOCK_MUTEX(mcpair->mutex); 304 if (wr_dag_h->status != rf_enable) { 305 RF_ERRORMSG("Unable to correct parity in VerifyParity: can't write the stripe\n"); 306 retcode = RF_PARITY_COULD_NOT_CORRECT; 307 } 308 rf_FreeDAG(wr_dag_h); 309 if (retcode == RF_PARITY_BAD) 310 retcode = RF_PARITY_CORRECTED; 311 } 312 out: 313 rf_FreeAccessStripeMap(asm_h); 314 rf_FreeAllocList(alloclist); 315 rf_FreeDAG(rd_dag_h); 316 rf_FreeMCPair(mcpair); 317 return (retcode); 318 } 319 320 int 321 rf_TryToRedirectPDA(raidPtr, pda, parity) 322 RF_Raid_t *raidPtr; 323 RF_PhysDiskAddr_t *pda; 324 int parity; 325 { 326 if (raidPtr->Disks[pda->row][pda->col].status == rf_ds_reconstructing) { 327 if (rf_CheckRUReconstructed(raidPtr->reconControl[pda->row]->reconMap, pda->startSector)) { 328 if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) { 329 RF_RowCol_t or = pda->row, oc = pda->col; 330 RF_SectorNum_t os = pda->startSector; 331 if (parity) { 332 (raidPtr->Layout.map->MapParity) (raidPtr, pda->raidAddress, &pda->row, &pda->col, &pda->startSector, RF_REMAP); 333 if (rf_verifyParityDebug) 334 printf("VerifyParity: Redir P r %d c %d sect %ld -> r %d c %d sect %ld\n", 335 or, oc, (long) os, pda->row, pda->col, (long) pda->startSector); 336 } else { 337 (raidPtr->Layout.map->MapSector) (raidPtr, pda->raidAddress, &pda->row, &pda->col, &pda->startSector, RF_REMAP); 338 if (rf_verifyParityDebug) 339 printf("VerifyParity: Redir D r %d c %d sect %ld -> r %d c %d sect %ld\n", 340 or, oc, (long) os, pda->row, pda->col, (long) pda->startSector); 341 } 342 } else { 343 RF_RowCol_t spRow = raidPtr->Disks[pda->row][pda->col].spareRow; 344 RF_RowCol_t spCol = raidPtr->Disks[pda->row][pda->col].spareCol; 345 pda->row = spRow; 346 pda->col = spCol; 347 } 348 } 349 } 350 if (RF_DEAD_DISK(raidPtr->Disks[pda->row][pda->col].status)) 351 return (1); 352 return (0); 353 } 354 /***************************************************************************************** 355 * 356 * currently a stub. 357 * 358 * takes as input an ASM describing a write operation and containing one failure, and 359 * verifies that the parity was correctly updated to reflect the write. 360 * 361 * if it's a data unit that's failed, we read the other data units in the stripe and 362 * the parity unit, XOR them together, and verify that we get the data intended for 363 * the failed disk. Since it's easy, we also validate that the right data got written 364 * to the surviving data disks. 365 * 366 * If it's the parity that failed, there's really no validation we can do except the 367 * above verification that the right data got written to all disks. This is because 368 * the new data intended for the failed disk is supplied in the ASM, but this is of 369 * course not the case for the new parity. 370 * 371 ****************************************************************************************/ 372 int 373 rf_VerifyDegrModeWrite(raidPtr, asmh) 374 RF_Raid_t *raidPtr; 375 RF_AccessStripeMapHeader_t *asmh; 376 { 377 return (0); 378 } 379 /* creates a simple DAG with a header, a block-recon node at level 1, 380 * nNodes nodes at level 2, an unblock-recon node at level 3, and 381 * a terminator node at level 4. The stripe address field in 382 * the block and unblock nodes are not touched, nor are the pda 383 * fields in the second-level nodes, so they must be filled in later. 384 * 385 * commit point is established at unblock node - this means that any 386 * failure during dag execution causes the dag to fail 387 */ 388 RF_DagHeader_t * 389 rf_MakeSimpleDAG(raidPtr, nNodes, bytesPerSU, databuf, doFunc, undoFunc, name, alloclist, flags, priority) 390 RF_Raid_t *raidPtr; 391 int nNodes; 392 int bytesPerSU; 393 char *databuf; 394 int (*doFunc) (RF_DagNode_t * node); 395 int (*undoFunc) (RF_DagNode_t * node); 396 char *name; /* node names at the second level */ 397 RF_AllocListElem_t *alloclist; 398 RF_RaidAccessFlags_t flags; 399 int priority; 400 { 401 RF_DagHeader_t *dag_h; 402 RF_DagNode_t *nodes, *termNode, *blockNode, *unblockNode; 403 int i; 404 405 /* create the nodes, the block & unblock nodes, and the terminator 406 * node */ 407 RF_CallocAndAdd(nodes, nNodes + 3, sizeof(RF_DagNode_t), (RF_DagNode_t *), alloclist); 408 blockNode = &nodes[nNodes]; 409 unblockNode = blockNode + 1; 410 termNode = unblockNode + 1; 411 412 dag_h = rf_AllocDAGHeader(); 413 dag_h->raidPtr = (void *) raidPtr; 414 dag_h->allocList = NULL;/* we won't use this alloc list */ 415 dag_h->status = rf_enable; 416 dag_h->numSuccedents = 1; 417 dag_h->creator = "SimpleDAG"; 418 419 /* this dag can not commit until the unblock node is reached errors 420 * prior to the commit point imply the dag has failed */ 421 dag_h->numCommitNodes = 1; 422 dag_h->numCommits = 0; 423 424 dag_h->succedents[0] = blockNode; 425 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nNodes, 0, 0, 0, dag_h, "Nil", alloclist); 426 rf_InitNode(unblockNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, 1, nNodes, 0, 0, dag_h, "Nil", alloclist); 427 unblockNode->succedents[0] = termNode; 428 for (i = 0; i < nNodes; i++) { 429 blockNode->succedents[i] = unblockNode->antecedents[i] = &nodes[i]; 430 unblockNode->antType[i] = rf_control; 431 rf_InitNode(&nodes[i], rf_wait, RF_FALSE, doFunc, undoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, name, alloclist); 432 nodes[i].succedents[0] = unblockNode; 433 nodes[i].antecedents[0] = blockNode; 434 nodes[i].antType[0] = rf_control; 435 nodes[i].params[1].p = (databuf + (i * bytesPerSU)); 436 } 437 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, 1, 0, 0, dag_h, "Trm", alloclist); 438 termNode->antecedents[0] = unblockNode; 439 termNode->antType[0] = rf_control; 440 return (dag_h); 441 } 442