1 /* $NetBSD: rf_parityscan.c,v 1.23 2004/01/10 00:56:28 oster 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 <sys/cdefs.h> 36 __KERNEL_RCSID(0, "$NetBSD: rf_parityscan.c,v 1.23 2004/01/10 00:56:28 oster Exp $"); 37 38 #include <dev/raidframe/raidframevar.h> 39 40 #include "rf_raid.h" 41 #include "rf_dag.h" 42 #include "rf_dagfuncs.h" 43 #include "rf_dagutils.h" 44 #include "rf_mcpair.h" 45 #include "rf_general.h" 46 #include "rf_engine.h" 47 #include "rf_parityscan.h" 48 #include "rf_map.h" 49 50 /***************************************************************************** 51 * 52 * walk through the entire arry and write new parity. This works by 53 * creating two DAGs, one to read a stripe of data and one to write 54 * new parity. The first is executed, the data is xored together, and 55 * then the second is executed. To avoid constantly building and 56 * tearing down the DAGs, we create them a priori and fill them in 57 * with the mapping information as we go along. 58 * 59 * there should never be more than one thread running this. 60 * 61 ****************************************************************************/ 62 63 int 64 rf_RewriteParity(RF_Raid_t *raidPtr) 65 { 66 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout; 67 RF_AccessStripeMapHeader_t *asm_h; 68 int ret_val; 69 int rc; 70 RF_SectorNum_t i; 71 72 if (raidPtr->Layout.map->faultsTolerated == 0) { 73 /* There isn't any parity. Call it "okay." */ 74 return (RF_PARITY_OKAY); 75 } 76 if (raidPtr->status != rf_rs_optimal) { 77 /* 78 * We're in degraded mode. Don't try to verify parity now! 79 * XXX: this should be a "we don't want to", not a 80 * "we can't" error. 81 */ 82 return (RF_PARITY_COULD_NOT_VERIFY); 83 } 84 85 ret_val = 0; 86 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. we 131 * validate only the range of parity defined by parityPDA, since this 132 * is all we have locked. The way we do this is to create an asm that 133 * 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(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *aasm, 139 int correct_it, RF_RaidAccessFlags_t flags) 140 { 141 RF_PhysDiskAddr_t *parityPDA; 142 RF_AccessStripeMap_t *doasm; 143 const RF_LayoutSW_t *lp; 144 int lrc, rc; 145 146 lp = raidPtr->Layout.map; 147 if (lp->faultsTolerated == 0) { 148 /* 149 * There isn't any parity. Call it "okay." 150 */ 151 return (RF_PARITY_OKAY); 152 } 153 rc = RF_PARITY_OKAY; 154 if (lp->VerifyParity) { 155 for (doasm = aasm; doasm; doasm = doasm->next) { 156 for (parityPDA = doasm->parityInfo; parityPDA; 157 parityPDA = parityPDA->next) { 158 lrc = lp->VerifyParity(raidPtr, 159 doasm->raidAddress, 160 parityPDA, 161 correct_it, flags); 162 if (lrc > rc) { 163 /* see rf_parityscan.h for why this 164 * works */ 165 rc = lrc; 166 } 167 } 168 } 169 } else { 170 rc = RF_PARITY_COULD_NOT_VERIFY; 171 } 172 return (rc); 173 } 174 175 int 176 rf_VerifyParityBasic(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr, 177 RF_PhysDiskAddr_t *parityPDA, int correct_it, 178 RF_RaidAccessFlags_t flags) 179 { 180 RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout); 181 RF_RaidAddr_t startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, 182 raidAddr); 183 RF_SectorCount_t numsector = parityPDA->numSector; 184 int numbytes = rf_RaidAddressToByte(raidPtr, numsector); 185 int bytesPerStripe = numbytes * layoutPtr->numDataCol; 186 RF_DagHeader_t *rd_dag_h, *wr_dag_h; /* read, write dag */ 187 RF_DagNode_t *blockNode, *wrBlock; 188 RF_AccessStripeMapHeader_t *asm_h; 189 RF_AccessStripeMap_t *asmap; 190 RF_AllocListElem_t *alloclist; 191 RF_PhysDiskAddr_t *pda; 192 char *pbuf, *buf, *end_p, *p; 193 int i, retcode; 194 RF_ReconUnitNum_t which_ru; 195 RF_StripeNum_t psID = rf_RaidAddressToParityStripeID(layoutPtr, 196 raidAddr, 197 &which_ru); 198 int stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol; 199 RF_AccTraceEntry_t tracerec; 200 RF_MCPair_t *mcpair; 201 202 retcode = RF_PARITY_OKAY; 203 204 mcpair = rf_AllocMCPair(); 205 rf_MakeAllocList(alloclist); 206 RF_MallocAndAdd(buf, numbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol), (char *), alloclist); 207 RF_MallocAndAdd(pbuf, numbytes, (char *), alloclist); 208 end_p = buf + bytesPerStripe; 209 210 rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, numbytes, buf, rf_DiskReadFunc, rf_DiskReadUndoFunc, 211 "Rod", alloclist, flags, RF_IO_NORMAL_PRIORITY); 212 blockNode = rd_dag_h->succedents[0]; 213 214 /* map the stripe and fill in the PDAs in the dag */ 215 asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe, buf, RF_DONT_REMAP); 216 asmap = asm_h->stripeMap; 217 218 for (pda = asmap->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) { 219 RF_ASSERT(pda); 220 rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1); 221 RF_ASSERT(pda->numSector != 0); 222 if (rf_TryToRedirectPDA(raidPtr, pda, 0)) 223 goto out; /* no way to verify parity if disk is 224 * dead. return w/ good status */ 225 blockNode->succedents[i]->params[0].p = pda; 226 blockNode->succedents[i]->params[2].v = psID; 227 blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); 228 } 229 230 RF_ASSERT(!asmap->parityInfo->next); 231 rf_RangeRestrictPDA(raidPtr, parityPDA, asmap->parityInfo, 0, 1); 232 RF_ASSERT(asmap->parityInfo->numSector != 0); 233 if (rf_TryToRedirectPDA(raidPtr, asmap->parityInfo, 1)) 234 goto out; 235 blockNode->succedents[layoutPtr->numDataCol]->params[0].p = asmap->parityInfo; 236 237 /* fire off the DAG */ 238 memset((char *) &tracerec, 0, sizeof(tracerec)); 239 rd_dag_h->tracerec = &tracerec; 240 #if 0 241 if (rf_verifyParityDebug) { 242 printf("Parity verify read dag:\n"); 243 rf_PrintDAGList(rd_dag_h); 244 } 245 #endif 246 RF_LOCK_MUTEX(mcpair->mutex); 247 mcpair->flag = 0; 248 rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 249 (void *) mcpair); 250 while (!mcpair->flag) 251 RF_WAIT_COND(mcpair->cond, mcpair->mutex); 252 RF_UNLOCK_MUTEX(mcpair->mutex); 253 if (rd_dag_h->status != rf_enable) { 254 RF_ERRORMSG("Unable to verify parity: can't read the stripe\n"); 255 retcode = RF_PARITY_COULD_NOT_VERIFY; 256 goto out; 257 } 258 for (p = buf; p < end_p; p += numbytes) { 259 rf_bxor(p, pbuf, numbytes); 260 } 261 for (i = 0; i < numbytes; i++) { 262 if (pbuf[i] != buf[bytesPerStripe + i]) { 263 if (!correct_it) 264 RF_ERRORMSG3("Parity verify error: byte %d of parity is 0x%x should be 0x%x\n", 265 i, (u_char) buf[bytesPerStripe + i], (u_char) pbuf[i]); 266 retcode = RF_PARITY_BAD; 267 break; 268 } 269 } 270 271 if (retcode && correct_it) { 272 wr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, numbytes, pbuf, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, 273 "Wnp", alloclist, flags, RF_IO_NORMAL_PRIORITY); 274 wrBlock = wr_dag_h->succedents[0]; 275 wrBlock->succedents[0]->params[0].p = asmap->parityInfo; 276 wrBlock->succedents[0]->params[2].v = psID; 277 wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); 278 memset((char *) &tracerec, 0, sizeof(tracerec)); 279 wr_dag_h->tracerec = &tracerec; 280 #if 0 281 if (rf_verifyParityDebug) { 282 printf("Parity verify write dag:\n"); 283 rf_PrintDAGList(wr_dag_h); 284 } 285 #endif 286 RF_LOCK_MUTEX(mcpair->mutex); 287 mcpair->flag = 0; 288 rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc, 289 (void *) mcpair); 290 while (!mcpair->flag) 291 RF_WAIT_COND(mcpair->cond, mcpair->mutex); 292 RF_UNLOCK_MUTEX(mcpair->mutex); 293 if (wr_dag_h->status != rf_enable) { 294 RF_ERRORMSG("Unable to correct parity in VerifyParity: can't write the stripe\n"); 295 retcode = RF_PARITY_COULD_NOT_CORRECT; 296 } 297 rf_FreeDAG(wr_dag_h); 298 if (retcode == RF_PARITY_BAD) 299 retcode = RF_PARITY_CORRECTED; 300 } 301 out: 302 rf_FreeAccessStripeMap(asm_h); 303 rf_FreeAllocList(alloclist); 304 rf_FreeDAG(rd_dag_h); 305 rf_FreeMCPair(mcpair); 306 return (retcode); 307 } 308 309 int 310 rf_TryToRedirectPDA(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda, int parity) 311 { 312 if (raidPtr->Disks[pda->col].status == rf_ds_reconstructing) { 313 if (rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, pda->startSector)) { 314 if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) { 315 #if RF_DEBUG_VERIFYPARITY 316 RF_RowCol_t oc = pda->col; 317 RF_SectorNum_t os = pda->startSector; 318 #endif 319 if (parity) { 320 (raidPtr->Layout.map->MapParity) (raidPtr, pda->raidAddress, &pda->col, &pda->startSector, RF_REMAP); 321 #if RF_DEBUG_VERIFYPARITY 322 if (rf_verifyParityDebug) 323 printf("VerifyParity: Redir P c %d sect %ld -> c %d sect %ld\n", 324 oc, (long) os, pda->col, (long) pda->startSector); 325 #endif 326 } else { 327 (raidPtr->Layout.map->MapSector) (raidPtr, pda->raidAddress, &pda->col, &pda->startSector, RF_REMAP); 328 #if RF_DEBUG_VERIFYPARITY 329 if (rf_verifyParityDebug) 330 printf("VerifyParity: Redir D c %d sect %ld -> c %d sect %ld\n", 331 oc, (long) os, pda->col, (long) pda->startSector); 332 #endif 333 } 334 } else { 335 RF_RowCol_t spCol = raidPtr->Disks[pda->col].spareCol; 336 pda->col = spCol; 337 } 338 } 339 } 340 if (RF_DEAD_DISK(raidPtr->Disks[pda->col].status)) 341 return (1); 342 return (0); 343 } 344 /***************************************************************************** 345 * 346 * currently a stub. 347 * 348 * takes as input an ASM describing a write operation and containing 349 * one failure, and verifies that the parity was correctly updated to 350 * reflect the write. 351 * 352 * if it's a data unit that's failed, we read the other data units in 353 * the stripe and the parity unit, XOR them together, and verify that 354 * we get the data intended for the failed disk. Since it's easy, we 355 * also validate that the right data got written to the surviving data 356 * disks. 357 * 358 * If it's the parity that failed, there's really no validation we can 359 * do except the above verification that the right data got written to 360 * all disks. This is because the new data intended for the failed 361 * disk is supplied in the ASM, but this is of course not the case for 362 * the new parity. 363 * 364 ****************************************************************************/ 365 #if 0 366 int 367 rf_VerifyDegrModeWrite(RF_Raid_t *raidPtr, RF_AccessStripeMapHeader_t *asmh) 368 { 369 return (0); 370 } 371 #endif 372 /* creates a simple DAG with a header, a block-recon node at level 1, 373 * nNodes nodes at level 2, an unblock-recon node at level 3, and a 374 * terminator node at level 4. The stripe address field in the block 375 * and unblock nodes are not touched, nor are the pda fields in the 376 * second-level nodes, so they must be filled in later. 377 * 378 * commit point is established at unblock node - this means that any 379 * failure during dag execution causes the dag to fail 380 * 381 * name - node names at the second level 382 */ 383 RF_DagHeader_t * 384 rf_MakeSimpleDAG(RF_Raid_t *raidPtr, int nNodes, int bytesPerSU, char *databuf, 385 int (*doFunc) (RF_DagNode_t * node), 386 int (*undoFunc) (RF_DagNode_t * node), 387 char *name, RF_AllocListElem_t *alloclist, 388 RF_RaidAccessFlags_t flags, int priority) 389 { 390 RF_DagHeader_t *dag_h; 391 RF_DagNode_t *nodes, *termNode, *blockNode, *unblockNode; 392 int i; 393 394 /* create the nodes, the block & unblock nodes, and the terminator 395 * node */ 396 RF_MallocAndAdd(nodes, (nNodes + 3) * sizeof(RF_DagNode_t), 397 (RF_DagNode_t *), alloclist); 398 blockNode = &nodes[nNodes]; 399 unblockNode = blockNode + 1; 400 termNode = unblockNode + 1; 401 402 dag_h = rf_AllocDAGHeader(); 403 dag_h->raidPtr = (void *) raidPtr; 404 dag_h->allocList = NULL;/* we won't use this alloc list */ 405 dag_h->status = rf_enable; 406 dag_h->numSuccedents = 1; 407 dag_h->creator = "SimpleDAG"; 408 409 /* this dag can not commit until the unblock node is reached errors 410 * prior to the commit point imply the dag has failed */ 411 dag_h->numCommitNodes = 1; 412 dag_h->numCommits = 0; 413 414 dag_h->succedents[0] = blockNode; 415 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nNodes, 0, 0, 0, dag_h, "Nil", alloclist); 416 rf_InitNode(unblockNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, 1, nNodes, 0, 0, dag_h, "Nil", alloclist); 417 unblockNode->succedents[0] = termNode; 418 for (i = 0; i < nNodes; i++) { 419 blockNode->succedents[i] = unblockNode->antecedents[i] = &nodes[i]; 420 unblockNode->antType[i] = rf_control; 421 rf_InitNode(&nodes[i], rf_wait, RF_FALSE, doFunc, undoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, name, alloclist); 422 nodes[i].succedents[0] = unblockNode; 423 nodes[i].antecedents[0] = blockNode; 424 nodes[i].antType[0] = rf_control; 425 nodes[i].params[1].p = (databuf + (i * bytesPerSU)); 426 } 427 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, 1, 0, 0, dag_h, "Trm", alloclist); 428 termNode->antecedents[0] = unblockNode; 429 termNode->antType[0] = rf_control; 430 return (dag_h); 431 } 432