rf_dagdegrd.c revision 1.5
1/* $NetBSD: rf_dagdegrd.c,v 1.5 2000/01/07 03:40:57 oster Exp $ */ 2/* 3 * Copyright (c) 1995 Carnegie-Mellon University. 4 * All rights reserved. 5 * 6 * Author: Mark Holland, Daniel Stodolsky, William V. Courtright II 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 * rf_dagdegrd.c 31 * 32 * code for creating degraded read DAGs 33 */ 34 35#include "rf_types.h" 36#include "rf_raid.h" 37#include "rf_dag.h" 38#include "rf_dagutils.h" 39#include "rf_dagfuncs.h" 40#include "rf_debugMem.h" 41#include "rf_memchunk.h" 42#include "rf_general.h" 43#include "rf_dagdegrd.h" 44 45 46/****************************************************************************** 47 * 48 * General comments on DAG creation: 49 * 50 * All DAGs in this file use roll-away error recovery. Each DAG has a single 51 * commit node, usually called "Cmt." If an error occurs before the Cmt node 52 * is reached, the execution engine will halt forward execution and work 53 * backward through the graph, executing the undo functions. Assuming that 54 * each node in the graph prior to the Cmt node are undoable and atomic - or - 55 * does not make changes to permanent state, the graph will fail atomically. 56 * If an error occurs after the Cmt node executes, the engine will roll-forward 57 * through the graph, blindly executing nodes until it reaches the end. 58 * If a graph reaches the end, it is assumed to have completed successfully. 59 * 60 * A graph has only 1 Cmt node. 61 * 62 */ 63 64 65/****************************************************************************** 66 * 67 * The following wrappers map the standard DAG creation interface to the 68 * DAG creation routines. Additionally, these wrappers enable experimentation 69 * with new DAG structures by providing an extra level of indirection, allowing 70 * the DAG creation routines to be replaced at this single point. 71 */ 72 73void 74rf_CreateRaidFiveDegradedReadDAG( 75 RF_Raid_t * raidPtr, 76 RF_AccessStripeMap_t * asmap, 77 RF_DagHeader_t * dag_h, 78 void *bp, 79 RF_RaidAccessFlags_t flags, 80 RF_AllocListElem_t * allocList) 81{ 82 rf_CreateDegradedReadDAG(raidPtr, asmap, dag_h, bp, flags, allocList, 83 &rf_xorRecoveryFuncs); 84} 85 86 87/****************************************************************************** 88 * 89 * DAG creation code begins here 90 */ 91 92 93/****************************************************************************** 94 * Create a degraded read DAG for RAID level 1 95 * 96 * Hdr -> Nil -> R(p/s)d -> Commit -> Trm 97 * 98 * The "Rd" node reads data from the surviving disk in the mirror pair 99 * Rpd - read of primary copy 100 * Rsd - read of secondary copy 101 * 102 * Parameters: raidPtr - description of the physical array 103 * asmap - logical & physical addresses for this access 104 * bp - buffer ptr (for holding write data) 105 * flags - general flags (e.g. disk locking) 106 * allocList - list of memory allocated in DAG creation 107 *****************************************************************************/ 108 109void 110rf_CreateRaidOneDegradedReadDAG( 111 RF_Raid_t * raidPtr, 112 RF_AccessStripeMap_t * asmap, 113 RF_DagHeader_t * dag_h, 114 void *bp, 115 RF_RaidAccessFlags_t flags, 116 RF_AllocListElem_t * allocList) 117{ 118 RF_DagNode_t *nodes, *rdNode, *blockNode, *commitNode, *termNode; 119 RF_StripeNum_t parityStripeID; 120 RF_ReconUnitNum_t which_ru; 121 RF_PhysDiskAddr_t *pda; 122 int useMirror, i; 123 124 useMirror = 0; 125 parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout), 126 asmap->raidAddress, &which_ru); 127 if (rf_dagDebug) { 128 printf("[Creating RAID level 1 degraded read DAG]\n"); 129 } 130 dag_h->creator = "RaidOneDegradedReadDAG"; 131 /* alloc the Wnd nodes and the Wmir node */ 132 if (asmap->numDataFailed == 0) 133 useMirror = RF_FALSE; 134 else 135 useMirror = RF_TRUE; 136 137 /* total number of nodes = 1 + (block + commit + terminator) */ 138 RF_CallocAndAdd(nodes, 4, sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList); 139 i = 0; 140 rdNode = &nodes[i]; 141 i++; 142 blockNode = &nodes[i]; 143 i++; 144 commitNode = &nodes[i]; 145 i++; 146 termNode = &nodes[i]; 147 i++; 148 149 /* this dag can not commit until the commit node is reached. errors 150 * prior to the commit point imply the dag has failed and must be 151 * retried */ 152 dag_h->numCommitNodes = 1; 153 dag_h->numCommits = 0; 154 dag_h->numSuccedents = 1; 155 156 /* initialize the block, commit, and terminator nodes */ 157 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, 158 NULL, 1, 0, 0, 0, dag_h, "Nil", allocList); 159 rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, 160 NULL, 1, 1, 0, 0, dag_h, "Cmt", allocList); 161 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, 162 NULL, 0, 1, 0, 0, dag_h, "Trm", allocList); 163 164 pda = asmap->physInfo; 165 RF_ASSERT(pda != NULL); 166 /* parityInfo must describe entire parity unit */ 167 RF_ASSERT(asmap->parityInfo->next == NULL); 168 169 /* initialize the data node */ 170 if (!useMirror) { 171 /* read primary copy of data */ 172 rf_InitNode(rdNode, rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, 173 rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Rpd", allocList); 174 rdNode->params[0].p = pda; 175 rdNode->params[1].p = pda->bufPtr; 176 rdNode->params[2].v = parityStripeID; 177 rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 178 } else { 179 /* read secondary copy of data */ 180 rf_InitNode(rdNode, rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, 181 rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Rsd", allocList); 182 rdNode->params[0].p = asmap->parityInfo; 183 rdNode->params[1].p = pda->bufPtr; 184 rdNode->params[2].v = parityStripeID; 185 rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 186 } 187 188 /* connect header to block node */ 189 RF_ASSERT(dag_h->numSuccedents == 1); 190 RF_ASSERT(blockNode->numAntecedents == 0); 191 dag_h->succedents[0] = blockNode; 192 193 /* connect block node to rdnode */ 194 RF_ASSERT(blockNode->numSuccedents == 1); 195 RF_ASSERT(rdNode->numAntecedents == 1); 196 blockNode->succedents[0] = rdNode; 197 rdNode->antecedents[0] = blockNode; 198 rdNode->antType[0] = rf_control; 199 200 /* connect rdnode to commit node */ 201 RF_ASSERT(rdNode->numSuccedents == 1); 202 RF_ASSERT(commitNode->numAntecedents == 1); 203 rdNode->succedents[0] = commitNode; 204 commitNode->antecedents[0] = rdNode; 205 commitNode->antType[0] = rf_control; 206 207 /* connect commit node to terminator */ 208 RF_ASSERT(commitNode->numSuccedents == 1); 209 RF_ASSERT(termNode->numAntecedents == 1); 210 RF_ASSERT(termNode->numSuccedents == 0); 211 commitNode->succedents[0] = termNode; 212 termNode->antecedents[0] = commitNode; 213 termNode->antType[0] = rf_control; 214} 215 216 217 218/****************************************************************************** 219 * 220 * creates a DAG to perform a degraded-mode read of data within one stripe. 221 * This DAG is as follows: 222 * 223 * Hdr -> Block -> Rud -> Xor -> Cmt -> T 224 * -> Rrd -> 225 * -> Rp --> 226 * 227 * Each R node is a successor of the L node 228 * One successor arc from each R node goes to C, and the other to X 229 * There is one Rud for each chunk of surviving user data requested by the 230 * user, and one Rrd for each chunk of surviving user data _not_ being read by 231 * the user 232 * R = read, ud = user data, rd = recovery (surviving) data, p = parity 233 * X = XOR, C = Commit, T = terminate 234 * 235 * The block node guarantees a single source node. 236 * 237 * Note: The target buffer for the XOR node is set to the actual user buffer 238 * where the failed data is supposed to end up. This buffer is zero'd by the 239 * code here. Thus, if you create a degraded read dag, use it, and then 240 * re-use, you have to be sure to zero the target buffer prior to the re-use. 241 * 242 * The recfunc argument at the end specifies the name and function used for 243 * the redundancy 244 * recovery function. 245 * 246 *****************************************************************************/ 247 248void 249rf_CreateDegradedReadDAG( 250 RF_Raid_t * raidPtr, 251 RF_AccessStripeMap_t * asmap, 252 RF_DagHeader_t * dag_h, 253 void *bp, 254 RF_RaidAccessFlags_t flags, 255 RF_AllocListElem_t * allocList, 256 RF_RedFuncs_t * recFunc) 257{ 258 RF_DagNode_t *nodes, *rudNodes, *rrdNodes, *xorNode, *blockNode; 259 RF_DagNode_t *commitNode, *rpNode, *termNode; 260 int nNodes, nRrdNodes, nRudNodes, nXorBufs, i; 261 int j, paramNum; 262 RF_SectorCount_t sectorsPerSU; 263 RF_ReconUnitNum_t which_ru; 264 char *overlappingPDAs;/* a temporary array of flags */ 265 RF_AccessStripeMapHeader_t *new_asm_h[2]; 266 RF_PhysDiskAddr_t *pda, *parityPDA; 267 RF_StripeNum_t parityStripeID; 268 RF_PhysDiskAddr_t *failedPDA; 269 RF_RaidLayout_t *layoutPtr; 270 char *rpBuf; 271 272 layoutPtr = &(raidPtr->Layout); 273 /* failedPDA points to the pda within the asm that targets the failed 274 * disk */ 275 failedPDA = asmap->failedPDAs[0]; 276 parityStripeID = rf_RaidAddressToParityStripeID(layoutPtr, 277 asmap->raidAddress, &which_ru); 278 sectorsPerSU = layoutPtr->sectorsPerStripeUnit; 279 280 if (rf_dagDebug) { 281 printf("[Creating degraded read DAG]\n"); 282 } 283 RF_ASSERT(asmap->numDataFailed == 1); 284 dag_h->creator = "DegradedReadDAG"; 285 286 /* 287 * generate two ASMs identifying the surviving data we need 288 * in order to recover the lost data 289 */ 290 291 /* overlappingPDAs array must be zero'd */ 292 RF_Calloc(overlappingPDAs, asmap->numStripeUnitsAccessed, sizeof(char), (char *)); 293 rf_GenerateFailedAccessASMs(raidPtr, asmap, failedPDA, dag_h, new_asm_h, &nXorBufs, 294 &rpBuf, overlappingPDAs, allocList); 295 296 /* 297 * create all the nodes at once 298 * 299 * -1 because no access is generated for the failed pda 300 */ 301 nRudNodes = asmap->numStripeUnitsAccessed - 1; 302 nRrdNodes = ((new_asm_h[0]) ? new_asm_h[0]->stripeMap->numStripeUnitsAccessed : 0) + 303 ((new_asm_h[1]) ? new_asm_h[1]->stripeMap->numStripeUnitsAccessed : 0); 304 nNodes = 5 + nRudNodes + nRrdNodes; /* lock, unlock, xor, Rp, Rud, 305 * Rrd */ 306 RF_CallocAndAdd(nodes, nNodes, sizeof(RF_DagNode_t), (RF_DagNode_t *), 307 allocList); 308 i = 0; 309 blockNode = &nodes[i]; 310 i++; 311 commitNode = &nodes[i]; 312 i++; 313 xorNode = &nodes[i]; 314 i++; 315 rpNode = &nodes[i]; 316 i++; 317 termNode = &nodes[i]; 318 i++; 319 rudNodes = &nodes[i]; 320 i += nRudNodes; 321 rrdNodes = &nodes[i]; 322 i += nRrdNodes; 323 RF_ASSERT(i == nNodes); 324 325 /* initialize nodes */ 326 dag_h->numCommitNodes = 1; 327 dag_h->numCommits = 0; 328 /* this dag can not commit until the commit node is reached errors 329 * prior to the commit point imply the dag has failed */ 330 dag_h->numSuccedents = 1; 331 332 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, 333 NULL, nRudNodes + nRrdNodes + 1, 0, 0, 0, dag_h, "Nil", allocList); 334 rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, 335 NULL, 1, 1, 0, 0, dag_h, "Cmt", allocList); 336 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, 337 NULL, 0, 1, 0, 0, dag_h, "Trm", allocList); 338 rf_InitNode(xorNode, rf_wait, RF_FALSE, recFunc->simple, rf_NullNodeUndoFunc, 339 NULL, 1, nRudNodes + nRrdNodes + 1, 2 * nXorBufs + 2, 1, dag_h, 340 recFunc->SimpleName, allocList); 341 342 /* fill in the Rud nodes */ 343 for (pda = asmap->physInfo, i = 0; i < nRudNodes; i++, pda = pda->next) { 344 if (pda == failedPDA) { 345 i--; 346 continue; 347 } 348 rf_InitNode(&rudNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc, 349 rf_DiskReadUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, 350 "Rud", allocList); 351 RF_ASSERT(pda); 352 rudNodes[i].params[0].p = pda; 353 rudNodes[i].params[1].p = pda->bufPtr; 354 rudNodes[i].params[2].v = parityStripeID; 355 rudNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 356 } 357 358 /* fill in the Rrd nodes */ 359 i = 0; 360 if (new_asm_h[0]) { 361 for (pda = new_asm_h[0]->stripeMap->physInfo; 362 i < new_asm_h[0]->stripeMap->numStripeUnitsAccessed; 363 i++, pda = pda->next) { 364 rf_InitNode(&rrdNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc, 365 rf_DiskReadUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, 366 dag_h, "Rrd", allocList); 367 RF_ASSERT(pda); 368 rrdNodes[i].params[0].p = pda; 369 rrdNodes[i].params[1].p = pda->bufPtr; 370 rrdNodes[i].params[2].v = parityStripeID; 371 rrdNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 372 } 373 } 374 if (new_asm_h[1]) { 375 for (j = 0, pda = new_asm_h[1]->stripeMap->physInfo; 376 j < new_asm_h[1]->stripeMap->numStripeUnitsAccessed; 377 j++, pda = pda->next) { 378 rf_InitNode(&rrdNodes[i + j], rf_wait, RF_FALSE, rf_DiskReadFunc, 379 rf_DiskReadUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, 380 dag_h, "Rrd", allocList); 381 RF_ASSERT(pda); 382 rrdNodes[i + j].params[0].p = pda; 383 rrdNodes[i + j].params[1].p = pda->bufPtr; 384 rrdNodes[i + j].params[2].v = parityStripeID; 385 rrdNodes[i + j].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 386 } 387 } 388 /* make a PDA for the parity unit */ 389 RF_MallocAndAdd(parityPDA, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t *), allocList); 390 parityPDA->row = asmap->parityInfo->row; 391 parityPDA->col = asmap->parityInfo->col; 392 parityPDA->startSector = ((asmap->parityInfo->startSector / sectorsPerSU) 393 * sectorsPerSU) + (failedPDA->startSector % sectorsPerSU); 394 parityPDA->numSector = failedPDA->numSector; 395 396 /* initialize the Rp node */ 397 rf_InitNode(rpNode, rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, 398 rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Rp ", allocList); 399 rpNode->params[0].p = parityPDA; 400 rpNode->params[1].p = rpBuf; 401 rpNode->params[2].v = parityStripeID; 402 rpNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 403 404 /* 405 * the last and nastiest step is to assign all 406 * the parameters of the Xor node 407 */ 408 paramNum = 0; 409 for (i = 0; i < nRrdNodes; i++) { 410 /* all the Rrd nodes need to be xored together */ 411 xorNode->params[paramNum++] = rrdNodes[i].params[0]; 412 xorNode->params[paramNum++] = rrdNodes[i].params[1]; 413 } 414 for (i = 0; i < nRudNodes; i++) { 415 /* any Rud nodes that overlap the failed access need to be 416 * xored in */ 417 if (overlappingPDAs[i]) { 418 RF_MallocAndAdd(pda, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t *), allocList); 419 bcopy((char *) rudNodes[i].params[0].p, (char *) pda, sizeof(RF_PhysDiskAddr_t)); 420 rf_RangeRestrictPDA(raidPtr, failedPDA, pda, RF_RESTRICT_DOBUFFER, 0); 421 xorNode->params[paramNum++].p = pda; 422 xorNode->params[paramNum++].p = pda->bufPtr; 423 } 424 } 425 RF_Free(overlappingPDAs, asmap->numStripeUnitsAccessed * sizeof(char)); 426 427 /* install parity pda as last set of params to be xor'd */ 428 xorNode->params[paramNum++].p = parityPDA; 429 xorNode->params[paramNum++].p = rpBuf; 430 431 /* 432 * the last 2 params to the recovery xor node are 433 * the failed PDA and the raidPtr 434 */ 435 xorNode->params[paramNum++].p = failedPDA; 436 xorNode->params[paramNum++].p = raidPtr; 437 RF_ASSERT(paramNum == 2 * nXorBufs + 2); 438 439 /* 440 * The xor node uses results[0] as the target buffer. 441 * Set pointer and zero the buffer. In the kernel, this 442 * may be a user buffer in which case we have to remap it. 443 */ 444 xorNode->results[0] = failedPDA->bufPtr; 445 RF_BZERO(bp, failedPDA->bufPtr, rf_RaidAddressToByte(raidPtr, 446 failedPDA->numSector)); 447 448 /* connect nodes to form graph */ 449 /* connect the header to the block node */ 450 RF_ASSERT(dag_h->numSuccedents == 1); 451 RF_ASSERT(blockNode->numAntecedents == 0); 452 dag_h->succedents[0] = blockNode; 453 454 /* connect the block node to the read nodes */ 455 RF_ASSERT(blockNode->numSuccedents == (1 + nRrdNodes + nRudNodes)); 456 RF_ASSERT(rpNode->numAntecedents == 1); 457 blockNode->succedents[0] = rpNode; 458 rpNode->antecedents[0] = blockNode; 459 rpNode->antType[0] = rf_control; 460 for (i = 0; i < nRrdNodes; i++) { 461 RF_ASSERT(rrdNodes[i].numSuccedents == 1); 462 blockNode->succedents[1 + i] = &rrdNodes[i]; 463 rrdNodes[i].antecedents[0] = blockNode; 464 rrdNodes[i].antType[0] = rf_control; 465 } 466 for (i = 0; i < nRudNodes; i++) { 467 RF_ASSERT(rudNodes[i].numSuccedents == 1); 468 blockNode->succedents[1 + nRrdNodes + i] = &rudNodes[i]; 469 rudNodes[i].antecedents[0] = blockNode; 470 rudNodes[i].antType[0] = rf_control; 471 } 472 473 /* connect the read nodes to the xor node */ 474 RF_ASSERT(xorNode->numAntecedents == (1 + nRrdNodes + nRudNodes)); 475 RF_ASSERT(rpNode->numSuccedents == 1); 476 rpNode->succedents[0] = xorNode; 477 xorNode->antecedents[0] = rpNode; 478 xorNode->antType[0] = rf_trueData; 479 for (i = 0; i < nRrdNodes; i++) { 480 RF_ASSERT(rrdNodes[i].numSuccedents == 1); 481 rrdNodes[i].succedents[0] = xorNode; 482 xorNode->antecedents[1 + i] = &rrdNodes[i]; 483 xorNode->antType[1 + i] = rf_trueData; 484 } 485 for (i = 0; i < nRudNodes; i++) { 486 RF_ASSERT(rudNodes[i].numSuccedents == 1); 487 rudNodes[i].succedents[0] = xorNode; 488 xorNode->antecedents[1 + nRrdNodes + i] = &rudNodes[i]; 489 xorNode->antType[1 + nRrdNodes + i] = rf_trueData; 490 } 491 492 /* connect the xor node to the commit node */ 493 RF_ASSERT(xorNode->numSuccedents == 1); 494 RF_ASSERT(commitNode->numAntecedents == 1); 495 xorNode->succedents[0] = commitNode; 496 commitNode->antecedents[0] = xorNode; 497 commitNode->antType[0] = rf_control; 498 499 /* connect the termNode to the commit node */ 500 RF_ASSERT(commitNode->numSuccedents == 1); 501 RF_ASSERT(termNode->numAntecedents == 1); 502 RF_ASSERT(termNode->numSuccedents == 0); 503 commitNode->succedents[0] = termNode; 504 termNode->antType[0] = rf_control; 505 termNode->antecedents[0] = commitNode; 506} 507 508 509/****************************************************************************** 510 * Create a degraded read DAG for Chained Declustering 511 * 512 * Hdr -> Nil -> R(p/s)d -> Cmt -> Trm 513 * 514 * The "Rd" node reads data from the surviving disk in the mirror pair 515 * Rpd - read of primary copy 516 * Rsd - read of secondary copy 517 * 518 * Parameters: raidPtr - description of the physical array 519 * asmap - logical & physical addresses for this access 520 * bp - buffer ptr (for holding write data) 521 * flags - general flags (e.g. disk locking) 522 * allocList - list of memory allocated in DAG creation 523 *****************************************************************************/ 524 525void 526rf_CreateRaidCDegradedReadDAG( 527 RF_Raid_t * raidPtr, 528 RF_AccessStripeMap_t * asmap, 529 RF_DagHeader_t * dag_h, 530 void *bp, 531 RF_RaidAccessFlags_t flags, 532 RF_AllocListElem_t * allocList) 533{ 534 RF_DagNode_t *nodes, *rdNode, *blockNode, *commitNode, *termNode; 535 RF_StripeNum_t parityStripeID; 536 int useMirror, i, shiftable; 537 RF_ReconUnitNum_t which_ru; 538 RF_PhysDiskAddr_t *pda; 539 540 if ((asmap->numDataFailed + asmap->numParityFailed) == 0) { 541 shiftable = RF_TRUE; 542 } else { 543 shiftable = RF_FALSE; 544 } 545 useMirror = 0; 546 parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout), 547 asmap->raidAddress, &which_ru); 548 549 if (rf_dagDebug) { 550 printf("[Creating RAID C degraded read DAG]\n"); 551 } 552 dag_h->creator = "RaidCDegradedReadDAG"; 553 /* alloc the Wnd nodes and the Wmir node */ 554 if (asmap->numDataFailed == 0) 555 useMirror = RF_FALSE; 556 else 557 useMirror = RF_TRUE; 558 559 /* total number of nodes = 1 + (block + commit + terminator) */ 560 RF_CallocAndAdd(nodes, 4, sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList); 561 i = 0; 562 rdNode = &nodes[i]; 563 i++; 564 blockNode = &nodes[i]; 565 i++; 566 commitNode = &nodes[i]; 567 i++; 568 termNode = &nodes[i]; 569 i++; 570 571 /* 572 * This dag can not commit until the commit node is reached. 573 * Errors prior to the commit point imply the dag has failed 574 * and must be retried. 575 */ 576 dag_h->numCommitNodes = 1; 577 dag_h->numCommits = 0; 578 dag_h->numSuccedents = 1; 579 580 /* initialize the block, commit, and terminator nodes */ 581 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, 582 NULL, 1, 0, 0, 0, dag_h, "Nil", allocList); 583 rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, 584 NULL, 1, 1, 0, 0, dag_h, "Cmt", allocList); 585 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, 586 NULL, 0, 1, 0, 0, dag_h, "Trm", allocList); 587 588 pda = asmap->physInfo; 589 RF_ASSERT(pda != NULL); 590 /* parityInfo must describe entire parity unit */ 591 RF_ASSERT(asmap->parityInfo->next == NULL); 592 593 /* initialize the data node */ 594 if (!useMirror) { 595 rf_InitNode(rdNode, rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, 596 rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Rpd", allocList); 597 if (shiftable && rf_compute_workload_shift(raidPtr, pda)) { 598 /* shift this read to the next disk in line */ 599 rdNode->params[0].p = asmap->parityInfo; 600 rdNode->params[1].p = pda->bufPtr; 601 rdNode->params[2].v = parityStripeID; 602 rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 603 } else { 604 /* read primary copy */ 605 rdNode->params[0].p = pda; 606 rdNode->params[1].p = pda->bufPtr; 607 rdNode->params[2].v = parityStripeID; 608 rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 609 } 610 } else { 611 /* read secondary copy of data */ 612 rf_InitNode(rdNode, rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, 613 rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Rsd", allocList); 614 rdNode->params[0].p = asmap->parityInfo; 615 rdNode->params[1].p = pda->bufPtr; 616 rdNode->params[2].v = parityStripeID; 617 rdNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru); 618 } 619 620 /* connect header to block node */ 621 RF_ASSERT(dag_h->numSuccedents == 1); 622 RF_ASSERT(blockNode->numAntecedents == 0); 623 dag_h->succedents[0] = blockNode; 624 625 /* connect block node to rdnode */ 626 RF_ASSERT(blockNode->numSuccedents == 1); 627 RF_ASSERT(rdNode->numAntecedents == 1); 628 blockNode->succedents[0] = rdNode; 629 rdNode->antecedents[0] = blockNode; 630 rdNode->antType[0] = rf_control; 631 632 /* connect rdnode to commit node */ 633 RF_ASSERT(rdNode->numSuccedents == 1); 634 RF_ASSERT(commitNode->numAntecedents == 1); 635 rdNode->succedents[0] = commitNode; 636 commitNode->antecedents[0] = rdNode; 637 commitNode->antType[0] = rf_control; 638 639 /* connect commit node to terminator */ 640 RF_ASSERT(commitNode->numSuccedents == 1); 641 RF_ASSERT(termNode->numAntecedents == 1); 642 RF_ASSERT(termNode->numSuccedents == 0); 643 commitNode->succedents[0] = termNode; 644 termNode->antecedents[0] = commitNode; 645 termNode->antType[0] = rf_control; 646} 647/* 648 * XXX move this elsewhere? 649 */ 650void 651rf_DD_GenerateFailedAccessASMs( 652 RF_Raid_t * raidPtr, 653 RF_AccessStripeMap_t * asmap, 654 RF_PhysDiskAddr_t ** pdap, 655 int *nNodep, 656 RF_PhysDiskAddr_t ** pqpdap, 657 int *nPQNodep, 658 RF_AllocListElem_t * allocList) 659{ 660 RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout); 661 int PDAPerDisk, i; 662 RF_SectorCount_t secPerSU = layoutPtr->sectorsPerStripeUnit; 663 int numDataCol = layoutPtr->numDataCol; 664 int state; 665 RF_SectorNum_t suoff, suend; 666 unsigned firstDataCol, napdas, count; 667 RF_SectorNum_t fone_start, fone_end, ftwo_start = 0, ftwo_end = 0; 668 RF_PhysDiskAddr_t *fone = asmap->failedPDAs[0], *ftwo = asmap->failedPDAs[1]; 669 RF_PhysDiskAddr_t *pda_p; 670 RF_PhysDiskAddr_t *phys_p; 671 RF_RaidAddr_t sosAddr; 672 673 /* determine how many pda's we will have to generate per unaccess 674 * stripe. If there is only one failed data unit, it is one; if two, 675 * possibly two, depending wether they overlap. */ 676 677 fone_start = rf_StripeUnitOffset(layoutPtr, fone->startSector); 678 fone_end = fone_start + fone->numSector; 679 680#define CONS_PDA(if,start,num) \ 681 pda_p->row = asmap->if->row; pda_p->col = asmap->if->col; \ 682 pda_p->startSector = ((asmap->if->startSector / secPerSU) * secPerSU) + start; \ 683 pda_p->numSector = num; \ 684 pda_p->next = NULL; \ 685 RF_MallocAndAdd(pda_p->bufPtr,rf_RaidAddressToByte(raidPtr,num),(char *), allocList) 686 687 if (asmap->numDataFailed == 1) { 688 PDAPerDisk = 1; 689 state = 1; 690 RF_MallocAndAdd(*pqpdap, 2 * sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t *), allocList); 691 pda_p = *pqpdap; 692 /* build p */ 693 CONS_PDA(parityInfo, fone_start, fone->numSector); 694 pda_p->type = RF_PDA_TYPE_PARITY; 695 pda_p++; 696 /* build q */ 697 CONS_PDA(qInfo, fone_start, fone->numSector); 698 pda_p->type = RF_PDA_TYPE_Q; 699 } else { 700 ftwo_start = rf_StripeUnitOffset(layoutPtr, ftwo->startSector); 701 ftwo_end = ftwo_start + ftwo->numSector; 702 if (fone->numSector + ftwo->numSector > secPerSU) { 703 PDAPerDisk = 1; 704 state = 2; 705 RF_MallocAndAdd(*pqpdap, 2 * sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t *), allocList); 706 pda_p = *pqpdap; 707 CONS_PDA(parityInfo, 0, secPerSU); 708 pda_p->type = RF_PDA_TYPE_PARITY; 709 pda_p++; 710 CONS_PDA(qInfo, 0, secPerSU); 711 pda_p->type = RF_PDA_TYPE_Q; 712 } else { 713 PDAPerDisk = 2; 714 state = 3; 715 /* four of them, fone, then ftwo */ 716 RF_MallocAndAdd(*pqpdap, 4 * sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t *), allocList); 717 pda_p = *pqpdap; 718 CONS_PDA(parityInfo, fone_start, fone->numSector); 719 pda_p->type = RF_PDA_TYPE_PARITY; 720 pda_p++; 721 CONS_PDA(qInfo, fone_start, fone->numSector); 722 pda_p->type = RF_PDA_TYPE_Q; 723 pda_p++; 724 CONS_PDA(parityInfo, ftwo_start, ftwo->numSector); 725 pda_p->type = RF_PDA_TYPE_PARITY; 726 pda_p++; 727 CONS_PDA(qInfo, ftwo_start, ftwo->numSector); 728 pda_p->type = RF_PDA_TYPE_Q; 729 } 730 } 731 /* figure out number of nonaccessed pda */ 732 napdas = PDAPerDisk * (numDataCol - asmap->numStripeUnitsAccessed - (ftwo == NULL ? 1 : 0)); 733 *nPQNodep = PDAPerDisk; 734 735 /* sweep over the over accessed pda's, figuring out the number of 736 * additional pda's to generate. Of course, skip the failed ones */ 737 738 count = 0; 739 for (pda_p = asmap->physInfo; pda_p; pda_p = pda_p->next) { 740 if ((pda_p == fone) || (pda_p == ftwo)) 741 continue; 742 suoff = rf_StripeUnitOffset(layoutPtr, pda_p->startSector); 743 suend = suoff + pda_p->numSector; 744 switch (state) { 745 case 1: /* one failed PDA to overlap */ 746 /* if a PDA doesn't contain the failed unit, it can 747 * only miss the start or end, not both */ 748 if ((suoff > fone_start) || (suend < fone_end)) 749 count++; 750 break; 751 case 2: /* whole stripe */ 752 if (suoff) /* leak at begining */ 753 count++; 754 if (suend < numDataCol) /* leak at end */ 755 count++; 756 break; 757 case 3: /* two disjoint units */ 758 if ((suoff > fone_start) || (suend < fone_end)) 759 count++; 760 if ((suoff > ftwo_start) || (suend < ftwo_end)) 761 count++; 762 break; 763 default: 764 RF_PANIC(); 765 } 766 } 767 768 napdas += count; 769 *nNodep = napdas; 770 if (napdas == 0) 771 return; /* short circuit */ 772 773 /* allocate up our list of pda's */ 774 775 RF_CallocAndAdd(pda_p, napdas, sizeof(RF_PhysDiskAddr_t), (RF_PhysDiskAddr_t *), allocList); 776 *pdap = pda_p; 777 778 /* linkem together */ 779 for (i = 0; i < (napdas - 1); i++) 780 pda_p[i].next = pda_p + (i + 1); 781 782 /* march through the one's up to the first accessed disk */ 783 firstDataCol = rf_RaidAddressToStripeUnitID(&(raidPtr->Layout), asmap->physInfo->raidAddress) % numDataCol; 784 sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress); 785 for (i = 0; i < firstDataCol; i++) { 786 if ((pda_p - (*pdap)) == napdas) 787 continue; 788 pda_p->type = RF_PDA_TYPE_DATA; 789 pda_p->raidAddress = sosAddr + (i * secPerSU); 790 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 791 /* skip over dead disks */ 792 if (RF_DEAD_DISK(raidPtr->Disks[pda_p->row][pda_p->col].status)) 793 continue; 794 switch (state) { 795 case 1: /* fone */ 796 pda_p->numSector = fone->numSector; 797 pda_p->raidAddress += fone_start; 798 pda_p->startSector += fone_start; 799 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 800 break; 801 case 2: /* full stripe */ 802 pda_p->numSector = secPerSU; 803 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, secPerSU), (char *), allocList); 804 break; 805 case 3: /* two slabs */ 806 pda_p->numSector = fone->numSector; 807 pda_p->raidAddress += fone_start; 808 pda_p->startSector += fone_start; 809 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 810 pda_p++; 811 pda_p->type = RF_PDA_TYPE_DATA; 812 pda_p->raidAddress = sosAddr + (i * secPerSU); 813 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 814 pda_p->numSector = ftwo->numSector; 815 pda_p->raidAddress += ftwo_start; 816 pda_p->startSector += ftwo_start; 817 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 818 break; 819 default: 820 RF_PANIC(); 821 } 822 pda_p++; 823 } 824 825 /* march through the touched stripe units */ 826 for (phys_p = asmap->physInfo; phys_p; phys_p = phys_p->next, i++) { 827 if ((phys_p == asmap->failedPDAs[0]) || (phys_p == asmap->failedPDAs[1])) 828 continue; 829 suoff = rf_StripeUnitOffset(layoutPtr, phys_p->startSector); 830 suend = suoff + phys_p->numSector; 831 switch (state) { 832 case 1: /* single buffer */ 833 if (suoff > fone_start) { 834 RF_ASSERT(suend >= fone_end); 835 /* The data read starts after the mapped 836 * access, snip off the begining */ 837 pda_p->numSector = suoff - fone_start; 838 pda_p->raidAddress = sosAddr + (i * secPerSU) + fone_start; 839 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 840 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 841 pda_p++; 842 } 843 if (suend < fone_end) { 844 RF_ASSERT(suoff <= fone_start); 845 /* The data read stops before the end of the 846 * failed access, extend */ 847 pda_p->numSector = fone_end - suend; 848 pda_p->raidAddress = sosAddr + (i * secPerSU) + suend; /* off by one? */ 849 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 850 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 851 pda_p++; 852 } 853 break; 854 case 2: /* whole stripe unit */ 855 RF_ASSERT((suoff == 0) || (suend == secPerSU)); 856 if (suend < secPerSU) { /* short read, snip from end 857 * on */ 858 pda_p->numSector = secPerSU - suend; 859 pda_p->raidAddress = sosAddr + (i * secPerSU) + suend; /* off by one? */ 860 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 861 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 862 pda_p++; 863 } else 864 if (suoff > 0) { /* short at front */ 865 pda_p->numSector = suoff; 866 pda_p->raidAddress = sosAddr + (i * secPerSU); 867 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 868 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 869 pda_p++; 870 } 871 break; 872 case 3: /* two nonoverlapping failures */ 873 if ((suoff > fone_start) || (suend < fone_end)) { 874 if (suoff > fone_start) { 875 RF_ASSERT(suend >= fone_end); 876 /* The data read starts after the 877 * mapped access, snip off the 878 * begining */ 879 pda_p->numSector = suoff - fone_start; 880 pda_p->raidAddress = sosAddr + (i * secPerSU) + fone_start; 881 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 882 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 883 pda_p++; 884 } 885 if (suend < fone_end) { 886 RF_ASSERT(suoff <= fone_start); 887 /* The data read stops before the end 888 * of the failed access, extend */ 889 pda_p->numSector = fone_end - suend; 890 pda_p->raidAddress = sosAddr + (i * secPerSU) + suend; /* off by one? */ 891 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 892 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 893 pda_p++; 894 } 895 } 896 if ((suoff > ftwo_start) || (suend < ftwo_end)) { 897 if (suoff > ftwo_start) { 898 RF_ASSERT(suend >= ftwo_end); 899 /* The data read starts after the 900 * mapped access, snip off the 901 * begining */ 902 pda_p->numSector = suoff - ftwo_start; 903 pda_p->raidAddress = sosAddr + (i * secPerSU) + ftwo_start; 904 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 905 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 906 pda_p++; 907 } 908 if (suend < ftwo_end) { 909 RF_ASSERT(suoff <= ftwo_start); 910 /* The data read stops before the end 911 * of the failed access, extend */ 912 pda_p->numSector = ftwo_end - suend; 913 pda_p->raidAddress = sosAddr + (i * secPerSU) + suend; /* off by one? */ 914 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 915 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 916 pda_p++; 917 } 918 } 919 break; 920 default: 921 RF_PANIC(); 922 } 923 } 924 925 /* after the last accessed disk */ 926 for (; i < numDataCol; i++) { 927 if ((pda_p - (*pdap)) == napdas) 928 continue; 929 pda_p->type = RF_PDA_TYPE_DATA; 930 pda_p->raidAddress = sosAddr + (i * secPerSU); 931 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 932 /* skip over dead disks */ 933 if (RF_DEAD_DISK(raidPtr->Disks[pda_p->row][pda_p->col].status)) 934 continue; 935 switch (state) { 936 case 1: /* fone */ 937 pda_p->numSector = fone->numSector; 938 pda_p->raidAddress += fone_start; 939 pda_p->startSector += fone_start; 940 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 941 break; 942 case 2: /* full stripe */ 943 pda_p->numSector = secPerSU; 944 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, secPerSU), (char *), allocList); 945 break; 946 case 3: /* two slabs */ 947 pda_p->numSector = fone->numSector; 948 pda_p->raidAddress += fone_start; 949 pda_p->startSector += fone_start; 950 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 951 pda_p++; 952 pda_p->type = RF_PDA_TYPE_DATA; 953 pda_p->raidAddress = sosAddr + (i * secPerSU); 954 (raidPtr->Layout.map->MapSector) (raidPtr, pda_p->raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), 0); 955 pda_p->numSector = ftwo->numSector; 956 pda_p->raidAddress += ftwo_start; 957 pda_p->startSector += ftwo_start; 958 RF_MallocAndAdd(pda_p->bufPtr, rf_RaidAddressToByte(raidPtr, pda_p->numSector), (char *), allocList); 959 break; 960 default: 961 RF_PANIC(); 962 } 963 pda_p++; 964 } 965 966 RF_ASSERT(pda_p - *pdap == napdas); 967 return; 968} 969#define INIT_DISK_NODE(node,name) \ 970rf_InitNode(node, rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, rf_GenericWakeupFunc, 2,1,4,0, dag_h, name, allocList); \ 971(node)->succedents[0] = unblockNode; \ 972(node)->succedents[1] = recoveryNode; \ 973(node)->antecedents[0] = blockNode; \ 974(node)->antType[0] = rf_control 975 976#define DISK_NODE_PARAMS(_node_,_p_) \ 977 (_node_).params[0].p = _p_ ; \ 978 (_node_).params[1].p = (_p_)->bufPtr; \ 979 (_node_).params[2].v = parityStripeID; \ 980 (_node_).params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru) 981 982void 983rf_DoubleDegRead( 984 RF_Raid_t * raidPtr, 985 RF_AccessStripeMap_t * asmap, 986 RF_DagHeader_t * dag_h, 987 void *bp, 988 RF_RaidAccessFlags_t flags, 989 RF_AllocListElem_t * allocList, 990 char *redundantReadNodeName, 991 char *recoveryNodeName, 992 int (*recovFunc) (RF_DagNode_t *)) 993{ 994 RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout); 995 RF_DagNode_t *nodes, *rudNodes, *rrdNodes, *recoveryNode, *blockNode, 996 *unblockNode, *rpNodes, *rqNodes, *termNode; 997 RF_PhysDiskAddr_t *pda, *pqPDAs; 998 RF_PhysDiskAddr_t *npdas; 999 int nNodes, nRrdNodes, nRudNodes, i; 1000 RF_ReconUnitNum_t which_ru; 1001 int nReadNodes, nPQNodes; 1002 RF_PhysDiskAddr_t *failedPDA = asmap->failedPDAs[0]; 1003 RF_PhysDiskAddr_t *failedPDAtwo = asmap->failedPDAs[1]; 1004 RF_StripeNum_t parityStripeID = rf_RaidAddressToParityStripeID(layoutPtr, asmap->raidAddress, &which_ru); 1005 1006 if (rf_dagDebug) 1007 printf("[Creating Double Degraded Read DAG]\n"); 1008 rf_DD_GenerateFailedAccessASMs(raidPtr, asmap, &npdas, &nRrdNodes, &pqPDAs, &nPQNodes, allocList); 1009 1010 nRudNodes = asmap->numStripeUnitsAccessed - (asmap->numDataFailed); 1011 nReadNodes = nRrdNodes + nRudNodes + 2 * nPQNodes; 1012 nNodes = 4 /* block, unblock, recovery, term */ + nReadNodes; 1013 1014 RF_CallocAndAdd(nodes, nNodes, sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList); 1015 i = 0; 1016 blockNode = &nodes[i]; 1017 i += 1; 1018 unblockNode = &nodes[i]; 1019 i += 1; 1020 recoveryNode = &nodes[i]; 1021 i += 1; 1022 termNode = &nodes[i]; 1023 i += 1; 1024 rudNodes = &nodes[i]; 1025 i += nRudNodes; 1026 rrdNodes = &nodes[i]; 1027 i += nRrdNodes; 1028 rpNodes = &nodes[i]; 1029 i += nPQNodes; 1030 rqNodes = &nodes[i]; 1031 i += nPQNodes; 1032 RF_ASSERT(i == nNodes); 1033 1034 dag_h->numSuccedents = 1; 1035 dag_h->succedents[0] = blockNode; 1036 dag_h->creator = "DoubleDegRead"; 1037 dag_h->numCommits = 0; 1038 dag_h->numCommitNodes = 1; /* unblock */ 1039 1040 rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, 2, 0, 0, dag_h, "Trm", allocList); 1041 termNode->antecedents[0] = unblockNode; 1042 termNode->antType[0] = rf_control; 1043 termNode->antecedents[1] = recoveryNode; 1044 termNode->antType[1] = rf_control; 1045 1046 /* init the block and unblock nodes */ 1047 /* The block node has all nodes except itself, unblock and recovery as 1048 * successors. Similarly for predecessors of the unblock. */ 1049 rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nReadNodes, 0, 0, 0, dag_h, "Nil", allocList); 1050 rf_InitNode(unblockNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, 1, nReadNodes, 0, 0, dag_h, "Nil", allocList); 1051 1052 for (i = 0; i < nReadNodes; i++) { 1053 blockNode->succedents[i] = rudNodes + i; 1054 unblockNode->antecedents[i] = rudNodes + i; 1055 unblockNode->antType[i] = rf_control; 1056 } 1057 unblockNode->succedents[0] = termNode; 1058 1059 /* The recovery node has all the reads as predecessors, and the term 1060 * node as successors. It gets a pda as a param from each of the read 1061 * nodes plus the raidPtr. For each failed unit is has a result pda. */ 1062 rf_InitNode(recoveryNode, rf_wait, RF_FALSE, recovFunc, rf_NullNodeUndoFunc, NULL, 1063 1, /* succesors */ 1064 nReadNodes, /* preds */ 1065 nReadNodes + 2, /* params */ 1066 asmap->numDataFailed, /* results */ 1067 dag_h, recoveryNodeName, allocList); 1068 1069 recoveryNode->succedents[0] = termNode; 1070 for (i = 0; i < nReadNodes; i++) { 1071 recoveryNode->antecedents[i] = rudNodes + i; 1072 recoveryNode->antType[i] = rf_trueData; 1073 } 1074 1075 /* build the read nodes, then come back and fill in recovery params 1076 * and results */ 1077 pda = asmap->physInfo; 1078 for (i = 0; i < nRudNodes; pda = pda->next) { 1079 if ((pda == failedPDA) || (pda == failedPDAtwo)) 1080 continue; 1081 INIT_DISK_NODE(rudNodes + i, "Rud"); 1082 RF_ASSERT(pda); 1083 DISK_NODE_PARAMS(rudNodes[i], pda); 1084 i++; 1085 } 1086 1087 pda = npdas; 1088 for (i = 0; i < nRrdNodes; i++, pda = pda->next) { 1089 INIT_DISK_NODE(rrdNodes + i, "Rrd"); 1090 RF_ASSERT(pda); 1091 DISK_NODE_PARAMS(rrdNodes[i], pda); 1092 } 1093 1094 /* redundancy pdas */ 1095 pda = pqPDAs; 1096 INIT_DISK_NODE(rpNodes, "Rp"); 1097 RF_ASSERT(pda); 1098 DISK_NODE_PARAMS(rpNodes[0], pda); 1099 pda++; 1100 INIT_DISK_NODE(rqNodes, redundantReadNodeName); 1101 RF_ASSERT(pda); 1102 DISK_NODE_PARAMS(rqNodes[0], pda); 1103 if (nPQNodes == 2) { 1104 pda++; 1105 INIT_DISK_NODE(rpNodes + 1, "Rp"); 1106 RF_ASSERT(pda); 1107 DISK_NODE_PARAMS(rpNodes[1], pda); 1108 pda++; 1109 INIT_DISK_NODE(rqNodes + 1, redundantReadNodeName); 1110 RF_ASSERT(pda); 1111 DISK_NODE_PARAMS(rqNodes[1], pda); 1112 } 1113 /* fill in recovery node params */ 1114 for (i = 0; i < nReadNodes; i++) 1115 recoveryNode->params[i] = rudNodes[i].params[0]; /* pda */ 1116 recoveryNode->params[i++].p = (void *) raidPtr; 1117 recoveryNode->params[i++].p = (void *) asmap; 1118 recoveryNode->results[0] = failedPDA; 1119 if (asmap->numDataFailed == 2) 1120 recoveryNode->results[1] = failedPDAtwo; 1121 1122 /* zero fill the target data buffers? */ 1123} 1124