/* $NetBSD: rf_dagffwr.c,v 1.21 2004/03/06 23:52:20 oster Exp $ */ /* * Copyright (c) 1995 Carnegie-Mellon University. * All rights reserved. * * Author: Mark Holland, Daniel Stodolsky, William V. Courtright II * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * rf_dagff.c * * code for creating fault-free DAGs * */ #include __KERNEL_RCSID(0, "$NetBSD: rf_dagffwr.c,v 1.21 2004/03/06 23:52:20 oster Exp $"); #include #include "rf_raid.h" #include "rf_dag.h" #include "rf_dagutils.h" #include "rf_dagfuncs.h" #include "rf_debugMem.h" #include "rf_dagffrd.h" #include "rf_general.h" #include "rf_dagffwr.h" /****************************************************************************** * * General comments on DAG creation: * * All DAGs in this file use roll-away error recovery. Each DAG has a single * commit node, usually called "Cmt." If an error occurs before the Cmt node * is reached, the execution engine will halt forward execution and work * backward through the graph, executing the undo functions. Assuming that * each node in the graph prior to the Cmt node are undoable and atomic - or - * does not make changes to permanent state, the graph will fail atomically. * If an error occurs after the Cmt node executes, the engine will roll-forward * through the graph, blindly executing nodes until it reaches the end. * If a graph reaches the end, it is assumed to have completed successfully. * * A graph has only 1 Cmt node. * */ /****************************************************************************** * * The following wrappers map the standard DAG creation interface to the * DAG creation routines. Additionally, these wrappers enable experimentation * with new DAG structures by providing an extra level of indirection, allowing * the DAG creation routines to be replaced at this single point. */ void rf_CreateNonRedundantWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap, RF_DagHeader_t *dag_h, void *bp, RF_RaidAccessFlags_t flags, RF_AllocListElem_t *allocList, RF_IoType_t type) { rf_CreateNonredundantDAG(raidPtr, asmap, dag_h, bp, flags, allocList, RF_IO_TYPE_WRITE); } void rf_CreateRAID0WriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap, RF_DagHeader_t *dag_h, void *bp, RF_RaidAccessFlags_t flags, RF_AllocListElem_t *allocList, RF_IoType_t type) { rf_CreateNonredundantDAG(raidPtr, asmap, dag_h, bp, flags, allocList, RF_IO_TYPE_WRITE); } void rf_CreateSmallWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap, RF_DagHeader_t *dag_h, void *bp, RF_RaidAccessFlags_t flags, RF_AllocListElem_t *allocList) { /* "normal" rollaway */ rf_CommonCreateSmallWriteDAG(raidPtr, asmap, dag_h, bp, flags, allocList, &rf_xorFuncs, NULL); } void rf_CreateLargeWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap, RF_DagHeader_t *dag_h, void *bp, RF_RaidAccessFlags_t flags, RF_AllocListElem_t *allocList) { /* "normal" rollaway */ rf_CommonCreateLargeWriteDAG(raidPtr, asmap, dag_h, bp, flags, allocList, 1, rf_RegularXorFunc, RF_TRUE); } /****************************************************************************** * * DAG creation code begins here */ /****************************************************************************** * * creates a DAG to perform a large-write operation: * * / Rod \ / Wnd \ * H -- block- Rod - Xor - Cmt - Wnd --- T * \ Rod / \ Wnp / * \[Wnq]/ * * The XOR node also does the Q calculation in the P+Q architecture. * All nodes are before the commit node (Cmt) are assumed to be atomic and * undoable - or - they make no changes to permanent state. * * Rod = read old data * Cmt = commit node * Wnp = write new parity * Wnd = write new data * Wnq = write new "q" * [] denotes optional segments in the graph * * Parameters: raidPtr - description of the physical array * asmap - logical & physical addresses for this access * bp - buffer ptr (holds write data) * flags - general flags (e.g. disk locking) * allocList - list of memory allocated in DAG creation * nfaults - number of faults array can tolerate * (equal to # redundancy units in stripe) * redfuncs - list of redundancy generating functions * *****************************************************************************/ void rf_CommonCreateLargeWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap, RF_DagHeader_t *dag_h, void *bp, RF_RaidAccessFlags_t flags, RF_AllocListElem_t *allocList, int nfaults, int (*redFunc) (RF_DagNode_t *), int allowBufferRecycle) { RF_DagNode_t *nodes, *wndNodes, *rodNodes, *xorNode, *wnpNode; RF_DagNode_t *wnqNode, *blockNode, *commitNode, *termNode; int nWndNodes, nRodNodes, i, nodeNum, asmNum; RF_AccessStripeMapHeader_t *new_asm_h[2]; RF_StripeNum_t parityStripeID; char *sosBuffer, *eosBuffer; RF_ReconUnitNum_t which_ru; RF_RaidLayout_t *layoutPtr; RF_PhysDiskAddr_t *pda; layoutPtr = &(raidPtr->Layout); parityStripeID = rf_RaidAddressToParityStripeID(layoutPtr, asmap->raidAddress, &which_ru); #if RF_DEBUG_DAG if (rf_dagDebug) { printf("[Creating large-write DAG]\n"); } #endif dag_h->creator = "LargeWriteDAG"; dag_h->numCommitNodes = 1; dag_h->numCommits = 0; dag_h->numSuccedents = 1; /* alloc the nodes: Wnd, xor, commit, block, term, and Wnp */ nWndNodes = asmap->numStripeUnitsAccessed; RF_MallocAndAdd(nodes, (nWndNodes + 4 + nfaults) * sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList); i = 0; wndNodes = &nodes[i]; i += nWndNodes; xorNode = &nodes[i]; i += 1; wnpNode = &nodes[i]; i += 1; blockNode = &nodes[i]; i += 1; commitNode = &nodes[i]; i += 1; termNode = &nodes[i]; i += 1; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { wnqNode = &nodes[i]; i += 1; } else { #endif wnqNode = NULL; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) } #endif rf_MapUnaccessedPortionOfStripe(raidPtr, layoutPtr, asmap, dag_h, new_asm_h, &nRodNodes, &sosBuffer, &eosBuffer, allocList); if (nRodNodes > 0) { RF_MallocAndAdd(rodNodes, nRodNodes * sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList); } else { rodNodes = NULL; } /* begin node initialization */ if (nRodNodes > 0) { rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nRodNodes, 0, 0, 0, dag_h, "Nil", allocList); } else { rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, 1, 0, 0, 0, dag_h, "Nil", allocList); } rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nWndNodes + nfaults, 1, 0, 0, dag_h, "Cmt", allocList); rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, nWndNodes + nfaults, 0, 0, dag_h, "Trm", allocList); /* initialize the Rod nodes */ for (nodeNum = asmNum = 0; asmNum < 2; asmNum++) { if (new_asm_h[asmNum]) { pda = new_asm_h[asmNum]->stripeMap->physInfo; while (pda) { rf_InitNode(&rodNodes[nodeNum], rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Rod", allocList); rodNodes[nodeNum].params[0].p = pda; rodNodes[nodeNum].params[1].p = pda->bufPtr; rodNodes[nodeNum].params[2].v = parityStripeID; rodNodes[nodeNum].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); nodeNum++; pda = pda->next; } } } RF_ASSERT(nodeNum == nRodNodes); /* initialize the wnd nodes */ pda = asmap->physInfo; for (i = 0; i < nWndNodes; i++) { rf_InitNode(&wndNodes[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnd", allocList); RF_ASSERT(pda != NULL); wndNodes[i].params[0].p = pda; wndNodes[i].params[1].p = pda->bufPtr; wndNodes[i].params[2].v = parityStripeID; wndNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; } /* initialize the redundancy node */ if (nRodNodes > 0) { rf_InitNode(xorNode, rf_wait, RF_FALSE, redFunc, rf_NullNodeUndoFunc, NULL, 1, nRodNodes, 2 * (nWndNodes + nRodNodes) + 1, nfaults, dag_h, "Xr ", allocList); } else { rf_InitNode(xorNode, rf_wait, RF_FALSE, redFunc, rf_NullNodeUndoFunc, NULL, 1, 1, 2 * (nWndNodes + nRodNodes) + 1, nfaults, dag_h, "Xr ", allocList); } xorNode->flags |= RF_DAGNODE_FLAG_YIELD; for (i = 0; i < nWndNodes; i++) { /* pda */ xorNode->params[2 * i + 0] = wndNodes[i].params[0]; /* buf ptr */ xorNode->params[2 * i + 1] = wndNodes[i].params[1]; } for (i = 0; i < nRodNodes; i++) { /* pda */ xorNode->params[2 * (nWndNodes + i) + 0] = rodNodes[i].params[0]; /* buf ptr */ xorNode->params[2 * (nWndNodes + i) + 1] = rodNodes[i].params[1]; } /* xor node needs to get at RAID information */ xorNode->params[2 * (nWndNodes + nRodNodes)].p = raidPtr; /* * Look for an Rod node that reads a complete SU. If none, * alloc a buffer to receive the parity info. Note that we * can't use a new data buffer because it will not have gotten * written when the xor occurs. */ if (allowBufferRecycle) { for (i = 0; i < nRodNodes; i++) { if (((RF_PhysDiskAddr_t *) rodNodes[i].params[0].p)->numSector == raidPtr->Layout.sectorsPerStripeUnit) break; } } if ((!allowBufferRecycle) || (i == nRodNodes)) { RF_MallocAndAdd(xorNode->results[0], rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit), (void *), allocList); } else { xorNode->results[0] = rodNodes[i].params[1].p; } /* initialize the Wnp node */ rf_InitNode(wnpNode, rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnp", allocList); wnpNode->params[0].p = asmap->parityInfo; wnpNode->params[1].p = xorNode->results[0]; wnpNode->params[2].v = parityStripeID; wnpNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); /* parityInfo must describe entire parity unit */ RF_ASSERT(asmap->parityInfo->next == NULL); #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { /* * We never try to recycle a buffer for the Q calcuation * in addition to the parity. This would cause two buffers * to get smashed during the P and Q calculation, guaranteeing * one would be wrong. */ RF_MallocAndAdd(xorNode->results[1], rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit), (void *), allocList); rf_InitNode(wnqNode, rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnq", allocList); wnqNode->params[0].p = asmap->qInfo; wnqNode->params[1].p = xorNode->results[1]; wnqNode->params[2].v = parityStripeID; wnqNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); /* parityInfo must describe entire parity unit */ RF_ASSERT(asmap->parityInfo->next == NULL); } #endif /* * Connect nodes to form graph. */ /* connect dag header to block node */ RF_ASSERT(blockNode->numAntecedents == 0); dag_h->succedents[0] = blockNode; if (nRodNodes > 0) { /* connect the block node to the Rod nodes */ RF_ASSERT(blockNode->numSuccedents == nRodNodes); RF_ASSERT(xorNode->numAntecedents == nRodNodes); for (i = 0; i < nRodNodes; i++) { RF_ASSERT(rodNodes[i].numAntecedents == 1); blockNode->succedents[i] = &rodNodes[i]; rodNodes[i].antecedents[0] = blockNode; rodNodes[i].antType[0] = rf_control; /* connect the Rod nodes to the Xor node */ RF_ASSERT(rodNodes[i].numSuccedents == 1); rodNodes[i].succedents[0] = xorNode; xorNode->antecedents[i] = &rodNodes[i]; xorNode->antType[i] = rf_trueData; } } else { /* connect the block node to the Xor node */ RF_ASSERT(blockNode->numSuccedents == 1); RF_ASSERT(xorNode->numAntecedents == 1); blockNode->succedents[0] = xorNode; xorNode->antecedents[0] = blockNode; xorNode->antType[0] = rf_control; } /* connect the xor node to the commit node */ RF_ASSERT(xorNode->numSuccedents == 1); RF_ASSERT(commitNode->numAntecedents == 1); xorNode->succedents[0] = commitNode; commitNode->antecedents[0] = xorNode; commitNode->antType[0] = rf_control; /* connect the commit node to the write nodes */ RF_ASSERT(commitNode->numSuccedents == nWndNodes + nfaults); for (i = 0; i < nWndNodes; i++) { RF_ASSERT(wndNodes->numAntecedents == 1); commitNode->succedents[i] = &wndNodes[i]; wndNodes[i].antecedents[0] = commitNode; wndNodes[i].antType[0] = rf_control; } RF_ASSERT(wnpNode->numAntecedents == 1); commitNode->succedents[nWndNodes] = wnpNode; wnpNode->antecedents[0] = commitNode; wnpNode->antType[0] = rf_trueData; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { RF_ASSERT(wnqNode->numAntecedents == 1); commitNode->succedents[nWndNodes + 1] = wnqNode; wnqNode->antecedents[0] = commitNode; wnqNode->antType[0] = rf_trueData; } #endif /* connect the write nodes to the term node */ RF_ASSERT(termNode->numAntecedents == nWndNodes + nfaults); RF_ASSERT(termNode->numSuccedents == 0); for (i = 0; i < nWndNodes; i++) { RF_ASSERT(wndNodes->numSuccedents == 1); wndNodes[i].succedents[0] = termNode; termNode->antecedents[i] = &wndNodes[i]; termNode->antType[i] = rf_control; } RF_ASSERT(wnpNode->numSuccedents == 1); wnpNode->succedents[0] = termNode; termNode->antecedents[nWndNodes] = wnpNode; termNode->antType[nWndNodes] = rf_control; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { RF_ASSERT(wnqNode->numSuccedents == 1); wnqNode->succedents[0] = termNode; termNode->antecedents[nWndNodes + 1] = wnqNode; termNode->antType[nWndNodes + 1] = rf_control; } #endif } /****************************************************************************** * * creates a DAG to perform a small-write operation (either raid 5 or pq), * which is as follows: * * Hdr -> Nil -> Rop -> Xor -> Cmt ----> Wnp [Unp] --> Trm * \- Rod X / \----> Wnd [Und]-/ * [\- Rod X / \---> Wnd [Und]-/] * [\- Roq -> Q / \--> Wnq [Unq]-/] * * Rop = read old parity * Rod = read old data * Roq = read old "q" * Cmt = commit node * Und = unlock data disk * Unp = unlock parity disk * Unq = unlock q disk * Wnp = write new parity * Wnd = write new data * Wnq = write new "q" * [ ] denotes optional segments in the graph * * Parameters: raidPtr - description of the physical array * asmap - logical & physical addresses for this access * bp - buffer ptr (holds write data) * flags - general flags (e.g. disk locking) * allocList - list of memory allocated in DAG creation * pfuncs - list of parity generating functions * qfuncs - list of q generating functions * * A null qfuncs indicates single fault tolerant *****************************************************************************/ void rf_CommonCreateSmallWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap, RF_DagHeader_t *dag_h, void *bp, RF_RaidAccessFlags_t flags, RF_AllocListElem_t *allocList, const RF_RedFuncs_t *pfuncs, const RF_RedFuncs_t *qfuncs) { RF_DagNode_t *readDataNodes, *readParityNodes, *readQNodes, *termNode; RF_DagNode_t *xorNodes, *qNodes, *blockNode, *commitNode, *nodes; RF_DagNode_t *writeDataNodes, *writeParityNodes, *writeQNodes; int i, j, nNodes, totalNumNodes; RF_ReconUnitNum_t which_ru; int (*func) (RF_DagNode_t *), (*undoFunc) (RF_DagNode_t *); int (*qfunc) (RF_DagNode_t *); int numDataNodes, numParityNodes; RF_StripeNum_t parityStripeID; RF_PhysDiskAddr_t *pda; char *name, *qname; long nfaults; nfaults = qfuncs ? 2 : 1; parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout), asmap->raidAddress, &which_ru); pda = asmap->physInfo; numDataNodes = asmap->numStripeUnitsAccessed; numParityNodes = (asmap->parityInfo->next) ? 2 : 1; #if RF_DEBUG_DAG if (rf_dagDebug) { printf("[Creating small-write DAG]\n"); } #endif RF_ASSERT(numDataNodes > 0); dag_h->creator = "SmallWriteDAG"; dag_h->numCommitNodes = 1; dag_h->numCommits = 0; dag_h->numSuccedents = 1; /* * DAG creation occurs in four steps: * 1. count the number of nodes in the DAG * 2. create the nodes * 3. initialize the nodes * 4. connect the nodes */ /* * Step 1. compute number of nodes in the graph */ /* number of nodes: a read and write for each data unit a * redundancy computation node for each parity node (nfaults * * nparity) a read and write for each parity unit a block and * commit node (2) a terminate node if atomic RMW an unlock * node for each data unit, redundancy unit */ totalNumNodes = (2 * numDataNodes) + (nfaults * numParityNodes) + (nfaults * 2 * numParityNodes) + 3; /* * Step 2. create the nodes */ RF_MallocAndAdd(nodes, totalNumNodes * sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList); i = 0; blockNode = &nodes[i]; i += 1; commitNode = &nodes[i]; i += 1; readDataNodes = &nodes[i]; i += numDataNodes; readParityNodes = &nodes[i]; i += numParityNodes; writeDataNodes = &nodes[i]; i += numDataNodes; writeParityNodes = &nodes[i]; i += numParityNodes; xorNodes = &nodes[i]; i += numParityNodes; termNode = &nodes[i]; i += 1; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { readQNodes = &nodes[i]; i += numParityNodes; writeQNodes = &nodes[i]; i += numParityNodes; qNodes = &nodes[i]; i += numParityNodes; } else { #endif readQNodes = writeQNodes = qNodes = NULL; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) } #endif RF_ASSERT(i == totalNumNodes); /* * Step 3. initialize the nodes */ /* initialize block node (Nil) */ nNodes = numDataNodes + (nfaults * numParityNodes); rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nNodes, 0, 0, 0, dag_h, "Nil", allocList); /* initialize commit node (Cmt) */ rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nNodes, (nfaults * numParityNodes), 0, 0, dag_h, "Cmt", allocList); /* initialize terminate node (Trm) */ rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, nNodes, 0, 0, dag_h, "Trm", allocList); /* initialize nodes which read old data (Rod) */ for (i = 0; i < numDataNodes; i++) { rf_InitNode(&readDataNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, rf_GenericWakeupFunc, (nfaults * numParityNodes), 1, 4, 0, dag_h, "Rod", allocList); RF_ASSERT(pda != NULL); /* physical disk addr desc */ readDataNodes[i].params[0].p = pda; /* buffer to hold old data */ readDataNodes[i].params[1].p = rf_AllocBuffer(raidPtr, pda, allocList); readDataNodes[i].params[2].v = parityStripeID; readDataNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; for (j = 0; j < readDataNodes[i].numSuccedents; j++) { readDataNodes[i].propList[j] = NULL; } } /* initialize nodes which read old parity (Rop) */ pda = asmap->parityInfo; i = 0; for (i = 0; i < numParityNodes; i++) { RF_ASSERT(pda != NULL); rf_InitNode(&readParityNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, rf_GenericWakeupFunc, numParityNodes, 1, 4, 0, dag_h, "Rop", allocList); readParityNodes[i].params[0].p = pda; /* buffer to hold old parity */ readParityNodes[i].params[1].p = rf_AllocBuffer(raidPtr, pda, allocList); readParityNodes[i].params[2].v = parityStripeID; readParityNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; for (j = 0; j < readParityNodes[i].numSuccedents; j++) { readParityNodes[i].propList[0] = NULL; } } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) /* initialize nodes which read old Q (Roq) */ if (nfaults == 2) { pda = asmap->qInfo; for (i = 0; i < numParityNodes; i++) { RF_ASSERT(pda != NULL); rf_InitNode(&readQNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc, rf_GenericWakeupFunc, numParityNodes, 1, 4, 0, dag_h, "Roq", allocList); readQNodes[i].params[0].p = pda; /* buffer to hold old Q */ readQNodes[i].params[1].p = rf_AllocBuffer(raidPtr, pda, allocList); readQNodes[i].params[2].v = parityStripeID; readQNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; for (j = 0; j < readQNodes[i].numSuccedents; j++) { readQNodes[i].propList[0] = NULL; } } } #endif /* initialize nodes which write new data (Wnd) */ pda = asmap->physInfo; for (i = 0; i < numDataNodes; i++) { RF_ASSERT(pda != NULL); rf_InitNode(&writeDataNodes[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnd", allocList); /* physical disk addr desc */ writeDataNodes[i].params[0].p = pda; /* buffer holding new data to be written */ writeDataNodes[i].params[1].p = pda->bufPtr; writeDataNodes[i].params[2].v = parityStripeID; writeDataNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; } /* * Initialize nodes which compute new parity and Q. */ /* * We use the simple XOR func in the double-XOR case, and when * we're accessing only a portion of one stripe unit. The * distinction between the two is that the regular XOR func * assumes that the targbuf is a full SU in size, and examines * the pda associated with the buffer to decide where within * the buffer to XOR the data, whereas the simple XOR func * just XORs the data into the start of the buffer. */ if ((numParityNodes == 2) || ((numDataNodes == 1) && (asmap->totalSectorsAccessed < raidPtr->Layout.sectorsPerStripeUnit))) { func = pfuncs->simple; undoFunc = rf_NullNodeUndoFunc; name = pfuncs->SimpleName; if (qfuncs) { qfunc = qfuncs->simple; qname = qfuncs->SimpleName; } else { qfunc = NULL; qname = NULL; } } else { func = pfuncs->regular; undoFunc = rf_NullNodeUndoFunc; name = pfuncs->RegularName; if (qfuncs) { qfunc = qfuncs->regular; qname = qfuncs->RegularName; } else { qfunc = NULL; qname = NULL; } } /* * Initialize the xor nodes: params are {pda,buf} * from {Rod,Wnd,Rop} nodes, and raidPtr */ if (numParityNodes == 2) { /* double-xor case */ for (i = 0; i < numParityNodes; i++) { /* note: no wakeup func for xor */ rf_InitNode(&xorNodes[i], rf_wait, RF_FALSE, func, undoFunc, NULL, 1, (numDataNodes + numParityNodes), 7, 1, dag_h, name, allocList); xorNodes[i].flags |= RF_DAGNODE_FLAG_YIELD; xorNodes[i].params[0] = readDataNodes[i].params[0]; xorNodes[i].params[1] = readDataNodes[i].params[1]; xorNodes[i].params[2] = readParityNodes[i].params[0]; xorNodes[i].params[3] = readParityNodes[i].params[1]; xorNodes[i].params[4] = writeDataNodes[i].params[0]; xorNodes[i].params[5] = writeDataNodes[i].params[1]; xorNodes[i].params[6].p = raidPtr; /* use old parity buf as target buf */ xorNodes[i].results[0] = readParityNodes[i].params[1].p; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { /* note: no wakeup func for qor */ rf_InitNode(&qNodes[i], rf_wait, RF_FALSE, qfunc, undoFunc, NULL, 1, (numDataNodes + numParityNodes), 7, 1, dag_h, qname, allocList); qNodes[i].params[0] = readDataNodes[i].params[0]; qNodes[i].params[1] = readDataNodes[i].params[1]; qNodes[i].params[2] = readQNodes[i].params[0]; qNodes[i].params[3] = readQNodes[i].params[1]; qNodes[i].params[4] = writeDataNodes[i].params[0]; qNodes[i].params[5] = writeDataNodes[i].params[1]; qNodes[i].params[6].p = raidPtr; /* use old Q buf as target buf */ qNodes[i].results[0] = readQNodes[i].params[1].p; } #endif } } else { /* there is only one xor node in this case */ rf_InitNode(&xorNodes[0], rf_wait, RF_FALSE, func, undoFunc, NULL, 1, (numDataNodes + numParityNodes), (2 * (numDataNodes + numDataNodes + 1) + 1), 1, dag_h, name, allocList); xorNodes[0].flags |= RF_DAGNODE_FLAG_YIELD; for (i = 0; i < numDataNodes + 1; i++) { /* set up params related to Rod and Rop nodes */ xorNodes[0].params[2 * i + 0] = readDataNodes[i].params[0]; /* pda */ xorNodes[0].params[2 * i + 1] = readDataNodes[i].params[1]; /* buffer ptr */ } for (i = 0; i < numDataNodes; i++) { /* set up params related to Wnd and Wnp nodes */ xorNodes[0].params[2 * (numDataNodes + 1 + i) + 0] = /* pda */ writeDataNodes[i].params[0]; xorNodes[0].params[2 * (numDataNodes + 1 + i) + 1] = /* buffer ptr */ writeDataNodes[i].params[1]; } /* xor node needs to get at RAID information */ xorNodes[0].params[2 * (numDataNodes + numDataNodes + 1)].p = raidPtr; xorNodes[0].results[0] = readParityNodes[0].params[1].p; #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { rf_InitNode(&qNodes[0], rf_wait, RF_FALSE, qfunc, undoFunc, NULL, 1, (numDataNodes + numParityNodes), (2 * (numDataNodes + numDataNodes + 1) + 1), 1, dag_h, qname, allocList); for (i = 0; i < numDataNodes; i++) { /* set up params related to Rod */ qNodes[0].params[2 * i + 0] = readDataNodes[i].params[0]; /* pda */ qNodes[0].params[2 * i + 1] = readDataNodes[i].params[1]; /* buffer ptr */ } /* and read old q */ qNodes[0].params[2 * numDataNodes + 0] = /* pda */ readQNodes[0].params[0]; qNodes[0].params[2 * numDataNodes + 1] = /* buffer ptr */ readQNodes[0].params[1]; for (i = 0; i < numDataNodes; i++) { /* set up params related to Wnd nodes */ qNodes[0].params[2 * (numDataNodes + 1 + i) + 0] = /* pda */ writeDataNodes[i].params[0]; qNodes[0].params[2 * (numDataNodes + 1 + i) + 1] = /* buffer ptr */ writeDataNodes[i].params[1]; } /* xor node needs to get at RAID information */ qNodes[0].params[2 * (numDataNodes + numDataNodes + 1)].p = raidPtr; qNodes[0].results[0] = readQNodes[0].params[1].p; } #endif } /* initialize nodes which write new parity (Wnp) */ pda = asmap->parityInfo; for (i = 0; i < numParityNodes; i++) { rf_InitNode(&writeParityNodes[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnp", allocList); RF_ASSERT(pda != NULL); writeParityNodes[i].params[0].p = pda; /* param 1 (bufPtr) * filled in by xor node */ writeParityNodes[i].params[1].p = xorNodes[i].results[0]; /* buffer pointer for * parity write * operation */ writeParityNodes[i].params[2].v = parityStripeID; writeParityNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) /* initialize nodes which write new Q (Wnq) */ if (nfaults == 2) { pda = asmap->qInfo; for (i = 0; i < numParityNodes; i++) { rf_InitNode(&writeQNodes[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnq", allocList); RF_ASSERT(pda != NULL); writeQNodes[i].params[0].p = pda; /* param 1 (bufPtr) * filled in by xor node */ writeQNodes[i].params[1].p = qNodes[i].results[0]; /* buffer pointer for * parity write * operation */ writeQNodes[i].params[2].v = parityStripeID; writeQNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; } } #endif /* * Step 4. connect the nodes. */ /* connect header to block node */ dag_h->succedents[0] = blockNode; /* connect block node to read old data nodes */ RF_ASSERT(blockNode->numSuccedents == (numDataNodes + (numParityNodes * nfaults))); for (i = 0; i < numDataNodes; i++) { blockNode->succedents[i] = &readDataNodes[i]; RF_ASSERT(readDataNodes[i].numAntecedents == 1); readDataNodes[i].antecedents[0] = blockNode; readDataNodes[i].antType[0] = rf_control; } /* connect block node to read old parity nodes */ for (i = 0; i < numParityNodes; i++) { blockNode->succedents[numDataNodes + i] = &readParityNodes[i]; RF_ASSERT(readParityNodes[i].numAntecedents == 1); readParityNodes[i].antecedents[0] = blockNode; readParityNodes[i].antType[0] = rf_control; } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) /* connect block node to read old Q nodes */ if (nfaults == 2) { for (i = 0; i < numParityNodes; i++) { blockNode->succedents[numDataNodes + numParityNodes + i] = &readQNodes[i]; RF_ASSERT(readQNodes[i].numAntecedents == 1); readQNodes[i].antecedents[0] = blockNode; readQNodes[i].antType[0] = rf_control; } } #endif /* connect read old data nodes to xor nodes */ for (i = 0; i < numDataNodes; i++) { RF_ASSERT(readDataNodes[i].numSuccedents == (nfaults * numParityNodes)); for (j = 0; j < numParityNodes; j++) { RF_ASSERT(xorNodes[j].numAntecedents == numDataNodes + numParityNodes); readDataNodes[i].succedents[j] = &xorNodes[j]; xorNodes[j].antecedents[i] = &readDataNodes[i]; xorNodes[j].antType[i] = rf_trueData; } } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) /* connect read old data nodes to q nodes */ if (nfaults == 2) { for (i = 0; i < numDataNodes; i++) { for (j = 0; j < numParityNodes; j++) { RF_ASSERT(qNodes[j].numAntecedents == numDataNodes + numParityNodes); readDataNodes[i].succedents[numParityNodes + j] = &qNodes[j]; qNodes[j].antecedents[i] = &readDataNodes[i]; qNodes[j].antType[i] = rf_trueData; } } } #endif /* connect read old parity nodes to xor nodes */ for (i = 0; i < numParityNodes; i++) { RF_ASSERT(readParityNodes[i].numSuccedents == numParityNodes); for (j = 0; j < numParityNodes; j++) { readParityNodes[i].succedents[j] = &xorNodes[j]; xorNodes[j].antecedents[numDataNodes + i] = &readParityNodes[i]; xorNodes[j].antType[numDataNodes + i] = rf_trueData; } } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) /* connect read old q nodes to q nodes */ if (nfaults == 2) { for (i = 0; i < numParityNodes; i++) { RF_ASSERT(readParityNodes[i].numSuccedents == numParityNodes); for (j = 0; j < numParityNodes; j++) { readQNodes[i].succedents[j] = &qNodes[j]; qNodes[j].antecedents[numDataNodes + i] = &readQNodes[i]; qNodes[j].antType[numDataNodes + i] = rf_trueData; } } } #endif /* connect xor nodes to commit node */ RF_ASSERT(commitNode->numAntecedents == (nfaults * numParityNodes)); for (i = 0; i < numParityNodes; i++) { RF_ASSERT(xorNodes[i].numSuccedents == 1); xorNodes[i].succedents[0] = commitNode; commitNode->antecedents[i] = &xorNodes[i]; commitNode->antType[i] = rf_control; } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) /* connect q nodes to commit node */ if (nfaults == 2) { for (i = 0; i < numParityNodes; i++) { RF_ASSERT(qNodes[i].numSuccedents == 1); qNodes[i].succedents[0] = commitNode; commitNode->antecedents[i + numParityNodes] = &qNodes[i]; commitNode->antType[i + numParityNodes] = rf_control; } } #endif /* connect commit node to write nodes */ RF_ASSERT(commitNode->numSuccedents == (numDataNodes + (nfaults * numParityNodes))); for (i = 0; i < numDataNodes; i++) { RF_ASSERT(writeDataNodes[i].numAntecedents == 1); commitNode->succedents[i] = &writeDataNodes[i]; writeDataNodes[i].antecedents[0] = commitNode; writeDataNodes[i].antType[0] = rf_trueData; } for (i = 0; i < numParityNodes; i++) { RF_ASSERT(writeParityNodes[i].numAntecedents == 1); commitNode->succedents[i + numDataNodes] = &writeParityNodes[i]; writeParityNodes[i].antecedents[0] = commitNode; writeParityNodes[i].antType[0] = rf_trueData; } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { for (i = 0; i < numParityNodes; i++) { RF_ASSERT(writeQNodes[i].numAntecedents == 1); commitNode->succedents[i + numDataNodes + numParityNodes] = &writeQNodes[i]; writeQNodes[i].antecedents[0] = commitNode; writeQNodes[i].antType[0] = rf_trueData; } } #endif RF_ASSERT(termNode->numAntecedents == (numDataNodes + (nfaults * numParityNodes))); RF_ASSERT(termNode->numSuccedents == 0); for (i = 0; i < numDataNodes; i++) { /* connect write new data nodes to term node */ RF_ASSERT(writeDataNodes[i].numSuccedents == 1); RF_ASSERT(termNode->numAntecedents == (numDataNodes + (nfaults * numParityNodes))); writeDataNodes[i].succedents[0] = termNode; termNode->antecedents[i] = &writeDataNodes[i]; termNode->antType[i] = rf_control; } for (i = 0; i < numParityNodes; i++) { RF_ASSERT(writeParityNodes[i].numSuccedents == 1); writeParityNodes[i].succedents[0] = termNode; termNode->antecedents[numDataNodes + i] = &writeParityNodes[i]; termNode->antType[numDataNodes + i] = rf_control; } #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) if (nfaults == 2) { for (i = 0; i < numParityNodes; i++) { RF_ASSERT(writeQNodes[i].numSuccedents == 1); writeQNodes[i].succedents[0] = termNode; termNode->antecedents[numDataNodes + numParityNodes + i] = &writeQNodes[i]; termNode->antType[numDataNodes + numParityNodes + i] = rf_control; } } #endif } /****************************************************************************** * create a write graph (fault-free or degraded) for RAID level 1 * * Hdr -> Commit -> Wpd -> Nil -> Trm * -> Wsd -> * * The "Wpd" node writes data to the primary copy in the mirror pair * The "Wsd" node writes data to the secondary copy in the mirror pair * * Parameters: raidPtr - description of the physical array * asmap - logical & physical addresses for this access * bp - buffer ptr (holds write data) * flags - general flags (e.g. disk locking) * allocList - list of memory allocated in DAG creation *****************************************************************************/ void rf_CreateRaidOneWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap, RF_DagHeader_t *dag_h, void *bp, RF_RaidAccessFlags_t flags, RF_AllocListElem_t *allocList) { RF_DagNode_t *unblockNode, *termNode, *commitNode; RF_DagNode_t *nodes, *wndNode, *wmirNode; int nWndNodes, nWmirNodes, i; RF_ReconUnitNum_t which_ru; RF_PhysDiskAddr_t *pda, *pdaP; RF_StripeNum_t parityStripeID; parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout), asmap->raidAddress, &which_ru); #if RF_DEBUG_DAG if (rf_dagDebug) { printf("[Creating RAID level 1 write DAG]\n"); } #endif dag_h->creator = "RaidOneWriteDAG"; /* 2 implies access not SU aligned */ nWmirNodes = (asmap->parityInfo->next) ? 2 : 1; nWndNodes = (asmap->physInfo->next) ? 2 : 1; /* alloc the Wnd nodes and the Wmir node */ if (asmap->numDataFailed == 1) nWndNodes--; if (asmap->numParityFailed == 1) nWmirNodes--; /* total number of nodes = nWndNodes + nWmirNodes + (commit + unblock * + terminator) */ RF_MallocAndAdd(nodes, (nWndNodes + nWmirNodes + 3) * sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList); i = 0; wndNode = &nodes[i]; i += nWndNodes; wmirNode = &nodes[i]; i += nWmirNodes; commitNode = &nodes[i]; i += 1; unblockNode = &nodes[i]; i += 1; termNode = &nodes[i]; i += 1; RF_ASSERT(i == (nWndNodes + nWmirNodes + 3)); /* this dag can commit immediately */ dag_h->numCommitNodes = 1; dag_h->numCommits = 0; dag_h->numSuccedents = 1; /* initialize the commit, unblock, and term nodes */ rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, (nWndNodes + nWmirNodes), 0, 0, 0, dag_h, "Cmt", allocList); rf_InitNode(unblockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, 1, (nWndNodes + nWmirNodes), 0, 0, dag_h, "Nil", allocList); rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, 1, 0, 0, dag_h, "Trm", allocList); /* initialize the wnd nodes */ if (nWndNodes > 0) { pda = asmap->physInfo; for (i = 0; i < nWndNodes; i++) { rf_InitNode(&wndNode[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wpd", allocList); RF_ASSERT(pda != NULL); wndNode[i].params[0].p = pda; wndNode[i].params[1].p = pda->bufPtr; wndNode[i].params[2].v = parityStripeID; wndNode[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; } RF_ASSERT(pda == NULL); } /* initialize the mirror nodes */ if (nWmirNodes > 0) { pda = asmap->physInfo; pdaP = asmap->parityInfo; for (i = 0; i < nWmirNodes; i++) { rf_InitNode(&wmirNode[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wsd", allocList); RF_ASSERT(pda != NULL); wmirNode[i].params[0].p = pdaP; wmirNode[i].params[1].p = pda->bufPtr; wmirNode[i].params[2].v = parityStripeID; wmirNode[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru); pda = pda->next; pdaP = pdaP->next; } RF_ASSERT(pda == NULL); RF_ASSERT(pdaP == NULL); } /* link the header node to the commit node */ RF_ASSERT(dag_h->numSuccedents == 1); RF_ASSERT(commitNode->numAntecedents == 0); dag_h->succedents[0] = commitNode; /* link the commit node to the write nodes */ RF_ASSERT(commitNode->numSuccedents == (nWndNodes + nWmirNodes)); for (i = 0; i < nWndNodes; i++) { RF_ASSERT(wndNode[i].numAntecedents == 1); commitNode->succedents[i] = &wndNode[i]; wndNode[i].antecedents[0] = commitNode; wndNode[i].antType[0] = rf_control; } for (i = 0; i < nWmirNodes; i++) { RF_ASSERT(wmirNode[i].numAntecedents == 1); commitNode->succedents[i + nWndNodes] = &wmirNode[i]; wmirNode[i].antecedents[0] = commitNode; wmirNode[i].antType[0] = rf_control; } /* link the write nodes to the unblock node */ RF_ASSERT(unblockNode->numAntecedents == (nWndNodes + nWmirNodes)); for (i = 0; i < nWndNodes; i++) { RF_ASSERT(wndNode[i].numSuccedents == 1); wndNode[i].succedents[0] = unblockNode; unblockNode->antecedents[i] = &wndNode[i]; unblockNode->antType[i] = rf_control; } for (i = 0; i < nWmirNodes; i++) { RF_ASSERT(wmirNode[i].numSuccedents == 1); wmirNode[i].succedents[0] = unblockNode; unblockNode->antecedents[i + nWndNodes] = &wmirNode[i]; unblockNode->antType[i + nWndNodes] = rf_control; } /* link the unblock node to the term node */ RF_ASSERT(unblockNode->numSuccedents == 1); RF_ASSERT(termNode->numAntecedents == 1); RF_ASSERT(termNode->numSuccedents == 0); unblockNode->succedents[0] = termNode; termNode->antecedents[0] = unblockNode; termNode->antType[0] = rf_control; }