Home | History | Annotate | Line # | Download | only in raidframe
rf_dagffwr.c revision 1.21
      1 /*	$NetBSD: rf_dagffwr.c,v 1.21 2004/03/06 23:52:20 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 (at) 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_dagff.c
     31  *
     32  * code for creating fault-free DAGs
     33  *
     34  */
     35 
     36 #include <sys/cdefs.h>
     37 __KERNEL_RCSID(0, "$NetBSD: rf_dagffwr.c,v 1.21 2004/03/06 23:52:20 oster Exp $");
     38 
     39 #include <dev/raidframe/raidframevar.h>
     40 
     41 #include "rf_raid.h"
     42 #include "rf_dag.h"
     43 #include "rf_dagutils.h"
     44 #include "rf_dagfuncs.h"
     45 #include "rf_debugMem.h"
     46 #include "rf_dagffrd.h"
     47 #include "rf_general.h"
     48 #include "rf_dagffwr.h"
     49 
     50 /******************************************************************************
     51  *
     52  * General comments on DAG creation:
     53  *
     54  * All DAGs in this file use roll-away error recovery.  Each DAG has a single
     55  * commit node, usually called "Cmt."  If an error occurs before the Cmt node
     56  * is reached, the execution engine will halt forward execution and work
     57  * backward through the graph, executing the undo functions.  Assuming that
     58  * each node in the graph prior to the Cmt node are undoable and atomic - or -
     59  * does not make changes to permanent state, the graph will fail atomically.
     60  * If an error occurs after the Cmt node executes, the engine will roll-forward
     61  * through the graph, blindly executing nodes until it reaches the end.
     62  * If a graph reaches the end, it is assumed to have completed successfully.
     63  *
     64  * A graph has only 1 Cmt node.
     65  *
     66  */
     67 
     68 
     69 /******************************************************************************
     70  *
     71  * The following wrappers map the standard DAG creation interface to the
     72  * DAG creation routines.  Additionally, these wrappers enable experimentation
     73  * with new DAG structures by providing an extra level of indirection, allowing
     74  * the DAG creation routines to be replaced at this single point.
     75  */
     76 
     77 
     78 void
     79 rf_CreateNonRedundantWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
     80 			      RF_DagHeader_t *dag_h, void *bp,
     81 			      RF_RaidAccessFlags_t flags,
     82 			      RF_AllocListElem_t *allocList,
     83 			      RF_IoType_t type)
     84 {
     85 	rf_CreateNonredundantDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
     86 				 RF_IO_TYPE_WRITE);
     87 }
     88 
     89 void
     90 rf_CreateRAID0WriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
     91 		       RF_DagHeader_t *dag_h, void *bp,
     92 		       RF_RaidAccessFlags_t flags,
     93 		       RF_AllocListElem_t *allocList,
     94 		       RF_IoType_t type)
     95 {
     96 	rf_CreateNonredundantDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
     97 				 RF_IO_TYPE_WRITE);
     98 }
     99 
    100 void
    101 rf_CreateSmallWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
    102 		       RF_DagHeader_t *dag_h, void *bp,
    103 		       RF_RaidAccessFlags_t flags,
    104 		       RF_AllocListElem_t *allocList)
    105 {
    106 	/* "normal" rollaway */
    107 	rf_CommonCreateSmallWriteDAG(raidPtr, asmap, dag_h, bp, flags,
    108 				     allocList, &rf_xorFuncs, NULL);
    109 }
    110 
    111 void
    112 rf_CreateLargeWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
    113 		       RF_DagHeader_t *dag_h, void *bp,
    114 		       RF_RaidAccessFlags_t flags,
    115 		       RF_AllocListElem_t *allocList)
    116 {
    117 	/* "normal" rollaway */
    118 	rf_CommonCreateLargeWriteDAG(raidPtr, asmap, dag_h, bp, flags,
    119 				     allocList, 1, rf_RegularXorFunc, RF_TRUE);
    120 }
    121 
    122 
    123 /******************************************************************************
    124  *
    125  * DAG creation code begins here
    126  */
    127 
    128 
    129 /******************************************************************************
    130  *
    131  * creates a DAG to perform a large-write operation:
    132  *
    133  *           / Rod \           / Wnd \
    134  * H -- block- Rod - Xor - Cmt - Wnd --- T
    135  *           \ Rod /          \  Wnp /
    136  *                             \[Wnq]/
    137  *
    138  * The XOR node also does the Q calculation in the P+Q architecture.
    139  * All nodes are before the commit node (Cmt) are assumed to be atomic and
    140  * undoable - or - they make no changes to permanent state.
    141  *
    142  * Rod = read old data
    143  * Cmt = commit node
    144  * Wnp = write new parity
    145  * Wnd = write new data
    146  * Wnq = write new "q"
    147  * [] denotes optional segments in the graph
    148  *
    149  * Parameters:  raidPtr   - description of the physical array
    150  *              asmap     - logical & physical addresses for this access
    151  *              bp        - buffer ptr (holds write data)
    152  *              flags     - general flags (e.g. disk locking)
    153  *              allocList - list of memory allocated in DAG creation
    154  *              nfaults   - number of faults array can tolerate
    155  *                          (equal to # redundancy units in stripe)
    156  *              redfuncs  - list of redundancy generating functions
    157  *
    158  *****************************************************************************/
    159 
    160 void
    161 rf_CommonCreateLargeWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
    162 			     RF_DagHeader_t *dag_h, void *bp,
    163 			     RF_RaidAccessFlags_t flags,
    164 			     RF_AllocListElem_t *allocList,
    165 			     int nfaults, int (*redFunc) (RF_DagNode_t *),
    166 			     int allowBufferRecycle)
    167 {
    168 	RF_DagNode_t *nodes, *wndNodes, *rodNodes, *xorNode, *wnpNode;
    169 	RF_DagNode_t *wnqNode, *blockNode, *commitNode, *termNode;
    170 	int     nWndNodes, nRodNodes, i, nodeNum, asmNum;
    171 	RF_AccessStripeMapHeader_t *new_asm_h[2];
    172 	RF_StripeNum_t parityStripeID;
    173 	char   *sosBuffer, *eosBuffer;
    174 	RF_ReconUnitNum_t which_ru;
    175 	RF_RaidLayout_t *layoutPtr;
    176 	RF_PhysDiskAddr_t *pda;
    177 
    178 	layoutPtr = &(raidPtr->Layout);
    179 	parityStripeID = rf_RaidAddressToParityStripeID(layoutPtr,
    180 							asmap->raidAddress,
    181 							&which_ru);
    182 
    183 #if RF_DEBUG_DAG
    184 	if (rf_dagDebug) {
    185 		printf("[Creating large-write DAG]\n");
    186 	}
    187 #endif
    188 	dag_h->creator = "LargeWriteDAG";
    189 
    190 	dag_h->numCommitNodes = 1;
    191 	dag_h->numCommits = 0;
    192 	dag_h->numSuccedents = 1;
    193 
    194 	/* alloc the nodes: Wnd, xor, commit, block, term, and  Wnp */
    195 	nWndNodes = asmap->numStripeUnitsAccessed;
    196 	RF_MallocAndAdd(nodes,
    197 			(nWndNodes + 4 + nfaults) * sizeof(RF_DagNode_t),
    198 			(RF_DagNode_t *), allocList);
    199 	i = 0;
    200 	wndNodes = &nodes[i];
    201 	i += nWndNodes;
    202 	xorNode = &nodes[i];
    203 	i += 1;
    204 	wnpNode = &nodes[i];
    205 	i += 1;
    206 	blockNode = &nodes[i];
    207 	i += 1;
    208 	commitNode = &nodes[i];
    209 	i += 1;
    210 	termNode = &nodes[i];
    211 	i += 1;
    212 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    213 	if (nfaults == 2) {
    214 		wnqNode = &nodes[i];
    215 		i += 1;
    216 	} else {
    217 #endif
    218 		wnqNode = NULL;
    219 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    220 	}
    221 #endif
    222 	rf_MapUnaccessedPortionOfStripe(raidPtr, layoutPtr, asmap, dag_h,
    223 					new_asm_h, &nRodNodes, &sosBuffer,
    224 					&eosBuffer, allocList);
    225 	if (nRodNodes > 0) {
    226 		RF_MallocAndAdd(rodNodes, nRodNodes * sizeof(RF_DagNode_t),
    227 				(RF_DagNode_t *), allocList);
    228 	} else {
    229 		rodNodes = NULL;
    230 	}
    231 
    232 	/* begin node initialization */
    233 	if (nRodNodes > 0) {
    234 		rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc,
    235 			    rf_NullNodeUndoFunc, NULL, nRodNodes, 0, 0, 0,
    236 			    dag_h, "Nil", allocList);
    237 	} else {
    238 		rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc,
    239 			    rf_NullNodeUndoFunc, NULL, 1, 0, 0, 0,
    240 			    dag_h, "Nil", allocList);
    241 	}
    242 
    243 	rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc,
    244 		    rf_NullNodeUndoFunc, NULL, nWndNodes + nfaults, 1, 0, 0,
    245 		    dag_h, "Cmt", allocList);
    246 	rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc,
    247 		    rf_TerminateUndoFunc, NULL, 0, nWndNodes + nfaults, 0, 0,
    248 		    dag_h, "Trm", allocList);
    249 
    250 	/* initialize the Rod nodes */
    251 	for (nodeNum = asmNum = 0; asmNum < 2; asmNum++) {
    252 		if (new_asm_h[asmNum]) {
    253 			pda = new_asm_h[asmNum]->stripeMap->physInfo;
    254 			while (pda) {
    255 				rf_InitNode(&rodNodes[nodeNum], rf_wait,
    256 					    RF_FALSE, rf_DiskReadFunc,
    257 					    rf_DiskReadUndoFunc,
    258 					    rf_GenericWakeupFunc,
    259 					    1, 1, 4, 0, dag_h,
    260 					    "Rod", allocList);
    261 				rodNodes[nodeNum].params[0].p = pda;
    262 				rodNodes[nodeNum].params[1].p = pda->bufPtr;
    263 				rodNodes[nodeNum].params[2].v = parityStripeID;
    264 				rodNodes[nodeNum].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    265 				    which_ru);
    266 				nodeNum++;
    267 				pda = pda->next;
    268 			}
    269 		}
    270 	}
    271 	RF_ASSERT(nodeNum == nRodNodes);
    272 
    273 	/* initialize the wnd nodes */
    274 	pda = asmap->physInfo;
    275 	for (i = 0; i < nWndNodes; i++) {
    276 		rf_InitNode(&wndNodes[i], rf_wait, RF_FALSE,
    277 			    rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
    278 			    rf_GenericWakeupFunc, 1, 1, 4, 0,
    279 			    dag_h, "Wnd", allocList);
    280 		RF_ASSERT(pda != NULL);
    281 		wndNodes[i].params[0].p = pda;
    282 		wndNodes[i].params[1].p = pda->bufPtr;
    283 		wndNodes[i].params[2].v = parityStripeID;
    284 		wndNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    285 		pda = pda->next;
    286 	}
    287 
    288 	/* initialize the redundancy node */
    289 	if (nRodNodes > 0) {
    290 		rf_InitNode(xorNode, rf_wait, RF_FALSE, redFunc,
    291 			    rf_NullNodeUndoFunc, NULL, 1,
    292 			    nRodNodes, 2 * (nWndNodes + nRodNodes) + 1,
    293 			    nfaults, dag_h, "Xr ", allocList);
    294 	} else {
    295 		rf_InitNode(xorNode, rf_wait, RF_FALSE, redFunc,
    296 			    rf_NullNodeUndoFunc, NULL, 1,
    297 			    1, 2 * (nWndNodes + nRodNodes) + 1,
    298 			    nfaults, dag_h, "Xr ", allocList);
    299 	}
    300 	xorNode->flags |= RF_DAGNODE_FLAG_YIELD;
    301 	for (i = 0; i < nWndNodes; i++) {
    302 		/* pda */
    303 		xorNode->params[2 * i + 0] = wndNodes[i].params[0];
    304 		/* buf ptr */
    305 		xorNode->params[2 * i + 1] = wndNodes[i].params[1];
    306 	}
    307 	for (i = 0; i < nRodNodes; i++) {
    308 		/* pda */
    309 		xorNode->params[2 * (nWndNodes + i) + 0] = rodNodes[i].params[0];
    310 		/* buf ptr */
    311 		xorNode->params[2 * (nWndNodes + i) + 1] = rodNodes[i].params[1];
    312 	}
    313 	/* xor node needs to get at RAID information */
    314 	xorNode->params[2 * (nWndNodes + nRodNodes)].p = raidPtr;
    315 
    316 	/*
    317          * Look for an Rod node that reads a complete SU. If none,
    318          * alloc a buffer to receive the parity info. Note that we
    319          * can't use a new data buffer because it will not have gotten
    320          * written when the xor occurs.  */
    321 	if (allowBufferRecycle) {
    322 		for (i = 0; i < nRodNodes; i++) {
    323 			if (((RF_PhysDiskAddr_t *) rodNodes[i].params[0].p)->numSector == raidPtr->Layout.sectorsPerStripeUnit)
    324 				break;
    325 		}
    326 	}
    327 	if ((!allowBufferRecycle) || (i == nRodNodes)) {
    328 		RF_MallocAndAdd(xorNode->results[0],
    329 				rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit),
    330 				(void *), allocList);
    331 	} else {
    332 		xorNode->results[0] = rodNodes[i].params[1].p;
    333 	}
    334 
    335 	/* initialize the Wnp node */
    336 	rf_InitNode(wnpNode, rf_wait, RF_FALSE, rf_DiskWriteFunc,
    337 		    rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0,
    338 		    dag_h, "Wnp", allocList);
    339 	wnpNode->params[0].p = asmap->parityInfo;
    340 	wnpNode->params[1].p = xorNode->results[0];
    341 	wnpNode->params[2].v = parityStripeID;
    342 	wnpNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    343 	/* parityInfo must describe entire parity unit */
    344 	RF_ASSERT(asmap->parityInfo->next == NULL);
    345 
    346 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    347 	if (nfaults == 2) {
    348 		/*
    349 	         * We never try to recycle a buffer for the Q calcuation
    350 	         * in addition to the parity. This would cause two buffers
    351 	         * to get smashed during the P and Q calculation, guaranteeing
    352 	         * one would be wrong.
    353 	         */
    354 		RF_MallocAndAdd(xorNode->results[1],
    355 				rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit),
    356 				(void *), allocList);
    357 		rf_InitNode(wnqNode, rf_wait, RF_FALSE, rf_DiskWriteFunc,
    358 			    rf_DiskWriteUndoFunc, rf_GenericWakeupFunc,
    359 			    1, 1, 4, 0, dag_h, "Wnq", allocList);
    360 		wnqNode->params[0].p = asmap->qInfo;
    361 		wnqNode->params[1].p = xorNode->results[1];
    362 		wnqNode->params[2].v = parityStripeID;
    363 		wnqNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
    364 		/* parityInfo must describe entire parity unit */
    365 		RF_ASSERT(asmap->parityInfo->next == NULL);
    366 	}
    367 #endif
    368 	/*
    369          * Connect nodes to form graph.
    370          */
    371 
    372 	/* connect dag header to block node */
    373 	RF_ASSERT(blockNode->numAntecedents == 0);
    374 	dag_h->succedents[0] = blockNode;
    375 
    376 	if (nRodNodes > 0) {
    377 		/* connect the block node to the Rod nodes */
    378 		RF_ASSERT(blockNode->numSuccedents == nRodNodes);
    379 		RF_ASSERT(xorNode->numAntecedents == nRodNodes);
    380 		for (i = 0; i < nRodNodes; i++) {
    381 			RF_ASSERT(rodNodes[i].numAntecedents == 1);
    382 			blockNode->succedents[i] = &rodNodes[i];
    383 			rodNodes[i].antecedents[0] = blockNode;
    384 			rodNodes[i].antType[0] = rf_control;
    385 
    386 			/* connect the Rod nodes to the Xor node */
    387 			RF_ASSERT(rodNodes[i].numSuccedents == 1);
    388 			rodNodes[i].succedents[0] = xorNode;
    389 			xorNode->antecedents[i] = &rodNodes[i];
    390 			xorNode->antType[i] = rf_trueData;
    391 		}
    392 	} else {
    393 		/* connect the block node to the Xor node */
    394 		RF_ASSERT(blockNode->numSuccedents == 1);
    395 		RF_ASSERT(xorNode->numAntecedents == 1);
    396 		blockNode->succedents[0] = xorNode;
    397 		xorNode->antecedents[0] = blockNode;
    398 		xorNode->antType[0] = rf_control;
    399 	}
    400 
    401 	/* connect the xor node to the commit node */
    402 	RF_ASSERT(xorNode->numSuccedents == 1);
    403 	RF_ASSERT(commitNode->numAntecedents == 1);
    404 	xorNode->succedents[0] = commitNode;
    405 	commitNode->antecedents[0] = xorNode;
    406 	commitNode->antType[0] = rf_control;
    407 
    408 	/* connect the commit node to the write nodes */
    409 	RF_ASSERT(commitNode->numSuccedents == nWndNodes + nfaults);
    410 	for (i = 0; i < nWndNodes; i++) {
    411 		RF_ASSERT(wndNodes->numAntecedents == 1);
    412 		commitNode->succedents[i] = &wndNodes[i];
    413 		wndNodes[i].antecedents[0] = commitNode;
    414 		wndNodes[i].antType[0] = rf_control;
    415 	}
    416 	RF_ASSERT(wnpNode->numAntecedents == 1);
    417 	commitNode->succedents[nWndNodes] = wnpNode;
    418 	wnpNode->antecedents[0] = commitNode;
    419 	wnpNode->antType[0] = rf_trueData;
    420 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    421 	if (nfaults == 2) {
    422 		RF_ASSERT(wnqNode->numAntecedents == 1);
    423 		commitNode->succedents[nWndNodes + 1] = wnqNode;
    424 		wnqNode->antecedents[0] = commitNode;
    425 		wnqNode->antType[0] = rf_trueData;
    426 	}
    427 #endif
    428 	/* connect the write nodes to the term node */
    429 	RF_ASSERT(termNode->numAntecedents == nWndNodes + nfaults);
    430 	RF_ASSERT(termNode->numSuccedents == 0);
    431 	for (i = 0; i < nWndNodes; i++) {
    432 		RF_ASSERT(wndNodes->numSuccedents == 1);
    433 		wndNodes[i].succedents[0] = termNode;
    434 		termNode->antecedents[i] = &wndNodes[i];
    435 		termNode->antType[i] = rf_control;
    436 	}
    437 	RF_ASSERT(wnpNode->numSuccedents == 1);
    438 	wnpNode->succedents[0] = termNode;
    439 	termNode->antecedents[nWndNodes] = wnpNode;
    440 	termNode->antType[nWndNodes] = rf_control;
    441 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    442 	if (nfaults == 2) {
    443 		RF_ASSERT(wnqNode->numSuccedents == 1);
    444 		wnqNode->succedents[0] = termNode;
    445 		termNode->antecedents[nWndNodes + 1] = wnqNode;
    446 		termNode->antType[nWndNodes + 1] = rf_control;
    447 	}
    448 #endif
    449 }
    450 /******************************************************************************
    451  *
    452  * creates a DAG to perform a small-write operation (either raid 5 or pq),
    453  * which is as follows:
    454  *
    455  * Hdr -> Nil -> Rop -> Xor -> Cmt ----> Wnp [Unp] --> Trm
    456  *            \- Rod X      /     \----> Wnd [Und]-/
    457  *           [\- Rod X     /       \---> Wnd [Und]-/]
    458  *           [\- Roq -> Q /         \--> Wnq [Unq]-/]
    459  *
    460  * Rop = read old parity
    461  * Rod = read old data
    462  * Roq = read old "q"
    463  * Cmt = commit node
    464  * Und = unlock data disk
    465  * Unp = unlock parity disk
    466  * Unq = unlock q disk
    467  * Wnp = write new parity
    468  * Wnd = write new data
    469  * Wnq = write new "q"
    470  * [ ] denotes optional segments in the graph
    471  *
    472  * Parameters:  raidPtr   - description of the physical array
    473  *              asmap     - logical & physical addresses for this access
    474  *              bp        - buffer ptr (holds write data)
    475  *              flags     - general flags (e.g. disk locking)
    476  *              allocList - list of memory allocated in DAG creation
    477  *              pfuncs    - list of parity generating functions
    478  *              qfuncs    - list of q generating functions
    479  *
    480  * A null qfuncs indicates single fault tolerant
    481  *****************************************************************************/
    482 
    483 void
    484 rf_CommonCreateSmallWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
    485 			     RF_DagHeader_t *dag_h, void *bp,
    486 			     RF_RaidAccessFlags_t flags,
    487 			     RF_AllocListElem_t *allocList,
    488 			     const RF_RedFuncs_t *pfuncs,
    489 			     const RF_RedFuncs_t *qfuncs)
    490 {
    491 	RF_DagNode_t *readDataNodes, *readParityNodes, *readQNodes, *termNode;
    492 	RF_DagNode_t *xorNodes, *qNodes, *blockNode, *commitNode, *nodes;
    493 	RF_DagNode_t *writeDataNodes, *writeParityNodes, *writeQNodes;
    494 	int     i, j, nNodes, totalNumNodes;
    495 	RF_ReconUnitNum_t which_ru;
    496 	int     (*func) (RF_DagNode_t *), (*undoFunc) (RF_DagNode_t *);
    497 	int     (*qfunc) (RF_DagNode_t *);
    498 	int     numDataNodes, numParityNodes;
    499 	RF_StripeNum_t parityStripeID;
    500 	RF_PhysDiskAddr_t *pda;
    501 	char   *name, *qname;
    502 	long    nfaults;
    503 
    504 	nfaults = qfuncs ? 2 : 1;
    505 
    506 	parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout),
    507 	    asmap->raidAddress, &which_ru);
    508 	pda = asmap->physInfo;
    509 	numDataNodes = asmap->numStripeUnitsAccessed;
    510 	numParityNodes = (asmap->parityInfo->next) ? 2 : 1;
    511 
    512 #if RF_DEBUG_DAG
    513 	if (rf_dagDebug) {
    514 		printf("[Creating small-write DAG]\n");
    515 	}
    516 #endif
    517 	RF_ASSERT(numDataNodes > 0);
    518 	dag_h->creator = "SmallWriteDAG";
    519 
    520 	dag_h->numCommitNodes = 1;
    521 	dag_h->numCommits = 0;
    522 	dag_h->numSuccedents = 1;
    523 
    524 	/*
    525          * DAG creation occurs in four steps:
    526          * 1. count the number of nodes in the DAG
    527          * 2. create the nodes
    528          * 3. initialize the nodes
    529          * 4. connect the nodes
    530          */
    531 
    532 	/*
    533          * Step 1. compute number of nodes in the graph
    534          */
    535 
    536 	/* number of nodes: a read and write for each data unit a
    537 	 * redundancy computation node for each parity node (nfaults *
    538 	 * nparity) a read and write for each parity unit a block and
    539 	 * commit node (2) a terminate node if atomic RMW an unlock
    540 	 * node for each data unit, redundancy unit */
    541 	totalNumNodes = (2 * numDataNodes) + (nfaults * numParityNodes)
    542 	    + (nfaults * 2 * numParityNodes) + 3;
    543 	/*
    544          * Step 2. create the nodes
    545          */
    546 	RF_MallocAndAdd(nodes, totalNumNodes * sizeof(RF_DagNode_t),
    547 			(RF_DagNode_t *), allocList);
    548 	i = 0;
    549 	blockNode = &nodes[i];
    550 	i += 1;
    551 	commitNode = &nodes[i];
    552 	i += 1;
    553 	readDataNodes = &nodes[i];
    554 	i += numDataNodes;
    555 	readParityNodes = &nodes[i];
    556 	i += numParityNodes;
    557 	writeDataNodes = &nodes[i];
    558 	i += numDataNodes;
    559 	writeParityNodes = &nodes[i];
    560 	i += numParityNodes;
    561 	xorNodes = &nodes[i];
    562 	i += numParityNodes;
    563 	termNode = &nodes[i];
    564 	i += 1;
    565 
    566 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    567 	if (nfaults == 2) {
    568 		readQNodes = &nodes[i];
    569 		i += numParityNodes;
    570 		writeQNodes = &nodes[i];
    571 		i += numParityNodes;
    572 		qNodes = &nodes[i];
    573 		i += numParityNodes;
    574 	} else {
    575 #endif
    576 		readQNodes = writeQNodes = qNodes = NULL;
    577 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    578 	}
    579 #endif
    580 	RF_ASSERT(i == totalNumNodes);
    581 
    582 	/*
    583          * Step 3. initialize the nodes
    584          */
    585 	/* initialize block node (Nil) */
    586 	nNodes = numDataNodes + (nfaults * numParityNodes);
    587 	rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc,
    588 		    rf_NullNodeUndoFunc, NULL, nNodes, 0, 0, 0,
    589 		    dag_h, "Nil", allocList);
    590 
    591 	/* initialize commit node (Cmt) */
    592 	rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc,
    593 		    rf_NullNodeUndoFunc, NULL, nNodes,
    594 		    (nfaults * numParityNodes), 0, 0, dag_h, "Cmt", allocList);
    595 
    596 	/* initialize terminate node (Trm) */
    597 	rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc,
    598 		    rf_TerminateUndoFunc, NULL, 0, nNodes, 0, 0,
    599 		    dag_h, "Trm", allocList);
    600 
    601 	/* initialize nodes which read old data (Rod) */
    602 	for (i = 0; i < numDataNodes; i++) {
    603 		rf_InitNode(&readDataNodes[i], rf_wait, RF_FALSE,
    604 			    rf_DiskReadFunc, rf_DiskReadUndoFunc,
    605 			    rf_GenericWakeupFunc, (nfaults * numParityNodes),
    606 			    1, 4, 0, dag_h, "Rod", allocList);
    607 		RF_ASSERT(pda != NULL);
    608 		/* physical disk addr desc */
    609 		readDataNodes[i].params[0].p = pda;
    610 		/* buffer to hold old data */
    611 		readDataNodes[i].params[1].p = rf_AllocBuffer(raidPtr, pda, allocList);
    612 		readDataNodes[i].params[2].v = parityStripeID;
    613 		readDataNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    614 		    which_ru);
    615 		pda = pda->next;
    616 		for (j = 0; j < readDataNodes[i].numSuccedents; j++) {
    617 			readDataNodes[i].propList[j] = NULL;
    618 		}
    619 	}
    620 
    621 	/* initialize nodes which read old parity (Rop) */
    622 	pda = asmap->parityInfo;
    623 	i = 0;
    624 	for (i = 0; i < numParityNodes; i++) {
    625 		RF_ASSERT(pda != NULL);
    626 		rf_InitNode(&readParityNodes[i], rf_wait, RF_FALSE,
    627 			    rf_DiskReadFunc, rf_DiskReadUndoFunc,
    628 			    rf_GenericWakeupFunc, numParityNodes, 1, 4, 0,
    629 			    dag_h, "Rop", allocList);
    630 		readParityNodes[i].params[0].p = pda;
    631 		/* buffer to hold old parity */
    632 		readParityNodes[i].params[1].p = rf_AllocBuffer(raidPtr, pda, allocList);
    633 		readParityNodes[i].params[2].v = parityStripeID;
    634 		readParityNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    635 		    which_ru);
    636 		pda = pda->next;
    637 		for (j = 0; j < readParityNodes[i].numSuccedents; j++) {
    638 			readParityNodes[i].propList[0] = NULL;
    639 		}
    640 	}
    641 
    642 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    643 	/* initialize nodes which read old Q (Roq) */
    644 	if (nfaults == 2) {
    645 		pda = asmap->qInfo;
    646 		for (i = 0; i < numParityNodes; i++) {
    647 			RF_ASSERT(pda != NULL);
    648 			rf_InitNode(&readQNodes[i], rf_wait, RF_FALSE,
    649 				    rf_DiskReadFunc, rf_DiskReadUndoFunc,
    650 				    rf_GenericWakeupFunc, numParityNodes,
    651 				    1, 4, 0, dag_h, "Roq", allocList);
    652 			readQNodes[i].params[0].p = pda;
    653 			/* buffer to hold old Q */
    654 			readQNodes[i].params[1].p = rf_AllocBuffer(raidPtr, pda, allocList);
    655 			readQNodes[i].params[2].v = parityStripeID;
    656 			readQNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    657 			    which_ru);
    658 			pda = pda->next;
    659 			for (j = 0; j < readQNodes[i].numSuccedents; j++) {
    660 				readQNodes[i].propList[0] = NULL;
    661 			}
    662 		}
    663 	}
    664 #endif
    665 	/* initialize nodes which write new data (Wnd) */
    666 	pda = asmap->physInfo;
    667 	for (i = 0; i < numDataNodes; i++) {
    668 		RF_ASSERT(pda != NULL);
    669 		rf_InitNode(&writeDataNodes[i], rf_wait, RF_FALSE,
    670 			    rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
    671 			    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h,
    672 			    "Wnd", allocList);
    673 		/* physical disk addr desc */
    674 		writeDataNodes[i].params[0].p = pda;
    675 		/* buffer holding new data to be written */
    676 		writeDataNodes[i].params[1].p = pda->bufPtr;
    677 		writeDataNodes[i].params[2].v = parityStripeID;
    678 		writeDataNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    679 		    which_ru);
    680 		pda = pda->next;
    681 	}
    682 
    683 	/*
    684          * Initialize nodes which compute new parity and Q.
    685          */
    686 	/*
    687          * We use the simple XOR func in the double-XOR case, and when
    688          * we're accessing only a portion of one stripe unit. The
    689          * distinction between the two is that the regular XOR func
    690          * assumes that the targbuf is a full SU in size, and examines
    691          * the pda associated with the buffer to decide where within
    692          * the buffer to XOR the data, whereas the simple XOR func
    693          * just XORs the data into the start of the buffer.  */
    694 	if ((numParityNodes == 2) || ((numDataNodes == 1)
    695 		&& (asmap->totalSectorsAccessed <
    696 		    raidPtr->Layout.sectorsPerStripeUnit))) {
    697 		func = pfuncs->simple;
    698 		undoFunc = rf_NullNodeUndoFunc;
    699 		name = pfuncs->SimpleName;
    700 		if (qfuncs) {
    701 			qfunc = qfuncs->simple;
    702 			qname = qfuncs->SimpleName;
    703 		} else {
    704 			qfunc = NULL;
    705 			qname = NULL;
    706 		}
    707 	} else {
    708 		func = pfuncs->regular;
    709 		undoFunc = rf_NullNodeUndoFunc;
    710 		name = pfuncs->RegularName;
    711 		if (qfuncs) {
    712 			qfunc = qfuncs->regular;
    713 			qname = qfuncs->RegularName;
    714 		} else {
    715 			qfunc = NULL;
    716 			qname = NULL;
    717 		}
    718 	}
    719 	/*
    720          * Initialize the xor nodes: params are {pda,buf}
    721          * from {Rod,Wnd,Rop} nodes, and raidPtr
    722          */
    723 	if (numParityNodes == 2) {
    724 		/* double-xor case */
    725 		for (i = 0; i < numParityNodes; i++) {
    726 			/* note: no wakeup func for xor */
    727 			rf_InitNode(&xorNodes[i], rf_wait, RF_FALSE, func,
    728 				    undoFunc, NULL, 1,
    729 				    (numDataNodes + numParityNodes),
    730 				    7, 1, dag_h, name, allocList);
    731 			xorNodes[i].flags |= RF_DAGNODE_FLAG_YIELD;
    732 			xorNodes[i].params[0] = readDataNodes[i].params[0];
    733 			xorNodes[i].params[1] = readDataNodes[i].params[1];
    734 			xorNodes[i].params[2] = readParityNodes[i].params[0];
    735 			xorNodes[i].params[3] = readParityNodes[i].params[1];
    736 			xorNodes[i].params[4] = writeDataNodes[i].params[0];
    737 			xorNodes[i].params[5] = writeDataNodes[i].params[1];
    738 			xorNodes[i].params[6].p = raidPtr;
    739 			/* use old parity buf as target buf */
    740 			xorNodes[i].results[0] = readParityNodes[i].params[1].p;
    741 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    742 			if (nfaults == 2) {
    743 				/* note: no wakeup func for qor */
    744 				rf_InitNode(&qNodes[i], rf_wait, RF_FALSE,
    745 					    qfunc, undoFunc, NULL, 1,
    746 					    (numDataNodes + numParityNodes),
    747 					    7, 1, dag_h, qname, allocList);
    748 				qNodes[i].params[0] = readDataNodes[i].params[0];
    749 				qNodes[i].params[1] = readDataNodes[i].params[1];
    750 				qNodes[i].params[2] = readQNodes[i].params[0];
    751 				qNodes[i].params[3] = readQNodes[i].params[1];
    752 				qNodes[i].params[4] = writeDataNodes[i].params[0];
    753 				qNodes[i].params[5] = writeDataNodes[i].params[1];
    754 				qNodes[i].params[6].p = raidPtr;
    755 				/* use old Q buf as target buf */
    756 				qNodes[i].results[0] = readQNodes[i].params[1].p;
    757 			}
    758 #endif
    759 		}
    760 	} else {
    761 		/* there is only one xor node in this case */
    762 		rf_InitNode(&xorNodes[0], rf_wait, RF_FALSE, func,
    763 			    undoFunc, NULL, 1, (numDataNodes + numParityNodes),
    764 			    (2 * (numDataNodes + numDataNodes + 1) + 1), 1,
    765 			    dag_h, name, allocList);
    766 		xorNodes[0].flags |= RF_DAGNODE_FLAG_YIELD;
    767 		for (i = 0; i < numDataNodes + 1; i++) {
    768 			/* set up params related to Rod and Rop nodes */
    769 			xorNodes[0].params[2 * i + 0] = readDataNodes[i].params[0];	/* pda */
    770 			xorNodes[0].params[2 * i + 1] = readDataNodes[i].params[1];	/* buffer ptr */
    771 		}
    772 		for (i = 0; i < numDataNodes; i++) {
    773 			/* set up params related to Wnd and Wnp nodes */
    774 			xorNodes[0].params[2 * (numDataNodes + 1 + i) + 0] =	/* pda */
    775 			    writeDataNodes[i].params[0];
    776 			xorNodes[0].params[2 * (numDataNodes + 1 + i) + 1] =	/* buffer ptr */
    777 			    writeDataNodes[i].params[1];
    778 		}
    779 		/* xor node needs to get at RAID information */
    780 		xorNodes[0].params[2 * (numDataNodes + numDataNodes + 1)].p = raidPtr;
    781 		xorNodes[0].results[0] = readParityNodes[0].params[1].p;
    782 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    783 		if (nfaults == 2) {
    784 			rf_InitNode(&qNodes[0], rf_wait, RF_FALSE, qfunc,
    785 				    undoFunc, NULL, 1,
    786 				    (numDataNodes + numParityNodes),
    787 				    (2 * (numDataNodes + numDataNodes + 1) + 1), 1,
    788 				    dag_h, qname, allocList);
    789 			for (i = 0; i < numDataNodes; i++) {
    790 				/* set up params related to Rod */
    791 				qNodes[0].params[2 * i + 0] = readDataNodes[i].params[0];	/* pda */
    792 				qNodes[0].params[2 * i + 1] = readDataNodes[i].params[1];	/* buffer ptr */
    793 			}
    794 			/* and read old q */
    795 			qNodes[0].params[2 * numDataNodes + 0] =	/* pda */
    796 			    readQNodes[0].params[0];
    797 			qNodes[0].params[2 * numDataNodes + 1] =	/* buffer ptr */
    798 			    readQNodes[0].params[1];
    799 			for (i = 0; i < numDataNodes; i++) {
    800 				/* set up params related to Wnd nodes */
    801 				qNodes[0].params[2 * (numDataNodes + 1 + i) + 0] =	/* pda */
    802 				    writeDataNodes[i].params[0];
    803 				qNodes[0].params[2 * (numDataNodes + 1 + i) + 1] =	/* buffer ptr */
    804 				    writeDataNodes[i].params[1];
    805 			}
    806 			/* xor node needs to get at RAID information */
    807 			qNodes[0].params[2 * (numDataNodes + numDataNodes + 1)].p = raidPtr;
    808 			qNodes[0].results[0] = readQNodes[0].params[1].p;
    809 		}
    810 #endif
    811 	}
    812 
    813 	/* initialize nodes which write new parity (Wnp) */
    814 	pda = asmap->parityInfo;
    815 	for (i = 0; i < numParityNodes; i++) {
    816 		rf_InitNode(&writeParityNodes[i], rf_wait, RF_FALSE,
    817 			    rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
    818 			    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h,
    819 			    "Wnp", allocList);
    820 		RF_ASSERT(pda != NULL);
    821 		writeParityNodes[i].params[0].p = pda;	/* param 1 (bufPtr)
    822 							 * filled in by xor node */
    823 		writeParityNodes[i].params[1].p = xorNodes[i].results[0];	/* buffer pointer for
    824 										 * parity write
    825 										 * operation */
    826 		writeParityNodes[i].params[2].v = parityStripeID;
    827 		writeParityNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    828 		    which_ru);
    829 		pda = pda->next;
    830 	}
    831 
    832 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    833 	/* initialize nodes which write new Q (Wnq) */
    834 	if (nfaults == 2) {
    835 		pda = asmap->qInfo;
    836 		for (i = 0; i < numParityNodes; i++) {
    837 			rf_InitNode(&writeQNodes[i], rf_wait, RF_FALSE,
    838 				    rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
    839 				    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h,
    840 				    "Wnq", allocList);
    841 			RF_ASSERT(pda != NULL);
    842 			writeQNodes[i].params[0].p = pda;	/* param 1 (bufPtr)
    843 								 * filled in by xor node */
    844 			writeQNodes[i].params[1].p = qNodes[i].results[0];	/* buffer pointer for
    845 										 * parity write
    846 										 * operation */
    847 			writeQNodes[i].params[2].v = parityStripeID;
    848 			writeQNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    849 			    which_ru);
    850 			pda = pda->next;
    851 		}
    852 	}
    853 #endif
    854 	/*
    855          * Step 4. connect the nodes.
    856          */
    857 
    858 	/* connect header to block node */
    859 	dag_h->succedents[0] = blockNode;
    860 
    861 	/* connect block node to read old data nodes */
    862 	RF_ASSERT(blockNode->numSuccedents == (numDataNodes + (numParityNodes * nfaults)));
    863 	for (i = 0; i < numDataNodes; i++) {
    864 		blockNode->succedents[i] = &readDataNodes[i];
    865 		RF_ASSERT(readDataNodes[i].numAntecedents == 1);
    866 		readDataNodes[i].antecedents[0] = blockNode;
    867 		readDataNodes[i].antType[0] = rf_control;
    868 	}
    869 
    870 	/* connect block node to read old parity nodes */
    871 	for (i = 0; i < numParityNodes; i++) {
    872 		blockNode->succedents[numDataNodes + i] = &readParityNodes[i];
    873 		RF_ASSERT(readParityNodes[i].numAntecedents == 1);
    874 		readParityNodes[i].antecedents[0] = blockNode;
    875 		readParityNodes[i].antType[0] = rf_control;
    876 	}
    877 
    878 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    879 	/* connect block node to read old Q nodes */
    880 	if (nfaults == 2) {
    881 		for (i = 0; i < numParityNodes; i++) {
    882 			blockNode->succedents[numDataNodes + numParityNodes + i] = &readQNodes[i];
    883 			RF_ASSERT(readQNodes[i].numAntecedents == 1);
    884 			readQNodes[i].antecedents[0] = blockNode;
    885 			readQNodes[i].antType[0] = rf_control;
    886 		}
    887 	}
    888 #endif
    889 	/* connect read old data nodes to xor nodes */
    890 	for (i = 0; i < numDataNodes; i++) {
    891 		RF_ASSERT(readDataNodes[i].numSuccedents == (nfaults * numParityNodes));
    892 		for (j = 0; j < numParityNodes; j++) {
    893 			RF_ASSERT(xorNodes[j].numAntecedents == numDataNodes + numParityNodes);
    894 			readDataNodes[i].succedents[j] = &xorNodes[j];
    895 			xorNodes[j].antecedents[i] = &readDataNodes[i];
    896 			xorNodes[j].antType[i] = rf_trueData;
    897 		}
    898 	}
    899 
    900 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    901 	/* connect read old data nodes to q nodes */
    902 	if (nfaults == 2) {
    903 		for (i = 0; i < numDataNodes; i++) {
    904 			for (j = 0; j < numParityNodes; j++) {
    905 				RF_ASSERT(qNodes[j].numAntecedents == numDataNodes + numParityNodes);
    906 				readDataNodes[i].succedents[numParityNodes + j] = &qNodes[j];
    907 				qNodes[j].antecedents[i] = &readDataNodes[i];
    908 				qNodes[j].antType[i] = rf_trueData;
    909 			}
    910 		}
    911 	}
    912 #endif
    913 	/* connect read old parity nodes to xor nodes */
    914 	for (i = 0; i < numParityNodes; i++) {
    915 		RF_ASSERT(readParityNodes[i].numSuccedents == numParityNodes);
    916 		for (j = 0; j < numParityNodes; j++) {
    917 			readParityNodes[i].succedents[j] = &xorNodes[j];
    918 			xorNodes[j].antecedents[numDataNodes + i] = &readParityNodes[i];
    919 			xorNodes[j].antType[numDataNodes + i] = rf_trueData;
    920 		}
    921 	}
    922 
    923 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    924 	/* connect read old q nodes to q nodes */
    925 	if (nfaults == 2) {
    926 		for (i = 0; i < numParityNodes; i++) {
    927 			RF_ASSERT(readParityNodes[i].numSuccedents == numParityNodes);
    928 			for (j = 0; j < numParityNodes; j++) {
    929 				readQNodes[i].succedents[j] = &qNodes[j];
    930 				qNodes[j].antecedents[numDataNodes + i] = &readQNodes[i];
    931 				qNodes[j].antType[numDataNodes + i] = rf_trueData;
    932 			}
    933 		}
    934 	}
    935 #endif
    936 	/* connect xor nodes to commit node */
    937 	RF_ASSERT(commitNode->numAntecedents == (nfaults * numParityNodes));
    938 	for (i = 0; i < numParityNodes; i++) {
    939 		RF_ASSERT(xorNodes[i].numSuccedents == 1);
    940 		xorNodes[i].succedents[0] = commitNode;
    941 		commitNode->antecedents[i] = &xorNodes[i];
    942 		commitNode->antType[i] = rf_control;
    943 	}
    944 
    945 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    946 	/* connect q nodes to commit node */
    947 	if (nfaults == 2) {
    948 		for (i = 0; i < numParityNodes; i++) {
    949 			RF_ASSERT(qNodes[i].numSuccedents == 1);
    950 			qNodes[i].succedents[0] = commitNode;
    951 			commitNode->antecedents[i + numParityNodes] = &qNodes[i];
    952 			commitNode->antType[i + numParityNodes] = rf_control;
    953 		}
    954 	}
    955 #endif
    956 	/* connect commit node to write nodes */
    957 	RF_ASSERT(commitNode->numSuccedents == (numDataNodes + (nfaults * numParityNodes)));
    958 	for (i = 0; i < numDataNodes; i++) {
    959 		RF_ASSERT(writeDataNodes[i].numAntecedents == 1);
    960 		commitNode->succedents[i] = &writeDataNodes[i];
    961 		writeDataNodes[i].antecedents[0] = commitNode;
    962 		writeDataNodes[i].antType[0] = rf_trueData;
    963 	}
    964 	for (i = 0; i < numParityNodes; i++) {
    965 		RF_ASSERT(writeParityNodes[i].numAntecedents == 1);
    966 		commitNode->succedents[i + numDataNodes] = &writeParityNodes[i];
    967 		writeParityNodes[i].antecedents[0] = commitNode;
    968 		writeParityNodes[i].antType[0] = rf_trueData;
    969 	}
    970 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    971 	if (nfaults == 2) {
    972 		for (i = 0; i < numParityNodes; i++) {
    973 			RF_ASSERT(writeQNodes[i].numAntecedents == 1);
    974 			commitNode->succedents[i + numDataNodes + numParityNodes] = &writeQNodes[i];
    975 			writeQNodes[i].antecedents[0] = commitNode;
    976 			writeQNodes[i].antType[0] = rf_trueData;
    977 		}
    978 	}
    979 #endif
    980 	RF_ASSERT(termNode->numAntecedents == (numDataNodes + (nfaults * numParityNodes)));
    981 	RF_ASSERT(termNode->numSuccedents == 0);
    982 	for (i = 0; i < numDataNodes; i++) {
    983 		/* connect write new data nodes to term node */
    984 		RF_ASSERT(writeDataNodes[i].numSuccedents == 1);
    985 		RF_ASSERT(termNode->numAntecedents == (numDataNodes + (nfaults * numParityNodes)));
    986 		writeDataNodes[i].succedents[0] = termNode;
    987 		termNode->antecedents[i] = &writeDataNodes[i];
    988 		termNode->antType[i] = rf_control;
    989 	}
    990 
    991 	for (i = 0; i < numParityNodes; i++) {
    992 		RF_ASSERT(writeParityNodes[i].numSuccedents == 1);
    993 		writeParityNodes[i].succedents[0] = termNode;
    994 		termNode->antecedents[numDataNodes + i] = &writeParityNodes[i];
    995 		termNode->antType[numDataNodes + i] = rf_control;
    996 	}
    997 
    998 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
    999 	if (nfaults == 2) {
   1000 		for (i = 0; i < numParityNodes; i++) {
   1001 			RF_ASSERT(writeQNodes[i].numSuccedents == 1);
   1002 			writeQNodes[i].succedents[0] = termNode;
   1003 			termNode->antecedents[numDataNodes + numParityNodes + i] = &writeQNodes[i];
   1004 			termNode->antType[numDataNodes + numParityNodes + i] = rf_control;
   1005 		}
   1006 	}
   1007 #endif
   1008 }
   1009 
   1010 
   1011 /******************************************************************************
   1012  * create a write graph (fault-free or degraded) for RAID level 1
   1013  *
   1014  * Hdr -> Commit -> Wpd -> Nil -> Trm
   1015  *               -> Wsd ->
   1016  *
   1017  * The "Wpd" node writes data to the primary copy in the mirror pair
   1018  * The "Wsd" node writes data to the secondary copy in the mirror pair
   1019  *
   1020  * Parameters:  raidPtr   - description of the physical array
   1021  *              asmap     - logical & physical addresses for this access
   1022  *              bp        - buffer ptr (holds write data)
   1023  *              flags     - general flags (e.g. disk locking)
   1024  *              allocList - list of memory allocated in DAG creation
   1025  *****************************************************************************/
   1026 
   1027 void
   1028 rf_CreateRaidOneWriteDAG(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
   1029 			 RF_DagHeader_t *dag_h, void *bp,
   1030 			 RF_RaidAccessFlags_t flags,
   1031 			 RF_AllocListElem_t *allocList)
   1032 {
   1033 	RF_DagNode_t *unblockNode, *termNode, *commitNode;
   1034 	RF_DagNode_t *nodes, *wndNode, *wmirNode;
   1035 	int     nWndNodes, nWmirNodes, i;
   1036 	RF_ReconUnitNum_t which_ru;
   1037 	RF_PhysDiskAddr_t *pda, *pdaP;
   1038 	RF_StripeNum_t parityStripeID;
   1039 
   1040 	parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout),
   1041 	    asmap->raidAddress, &which_ru);
   1042 #if RF_DEBUG_DAG
   1043 	if (rf_dagDebug) {
   1044 		printf("[Creating RAID level 1 write DAG]\n");
   1045 	}
   1046 #endif
   1047 	dag_h->creator = "RaidOneWriteDAG";
   1048 
   1049 	/* 2 implies access not SU aligned */
   1050 	nWmirNodes = (asmap->parityInfo->next) ? 2 : 1;
   1051 	nWndNodes = (asmap->physInfo->next) ? 2 : 1;
   1052 
   1053 	/* alloc the Wnd nodes and the Wmir node */
   1054 	if (asmap->numDataFailed == 1)
   1055 		nWndNodes--;
   1056 	if (asmap->numParityFailed == 1)
   1057 		nWmirNodes--;
   1058 
   1059 	/* total number of nodes = nWndNodes + nWmirNodes + (commit + unblock
   1060 	 * + terminator) */
   1061 	RF_MallocAndAdd(nodes,
   1062 			(nWndNodes + nWmirNodes + 3) * sizeof(RF_DagNode_t),
   1063 			(RF_DagNode_t *), allocList);
   1064 	i = 0;
   1065 	wndNode = &nodes[i];
   1066 	i += nWndNodes;
   1067 	wmirNode = &nodes[i];
   1068 	i += nWmirNodes;
   1069 	commitNode = &nodes[i];
   1070 	i += 1;
   1071 	unblockNode = &nodes[i];
   1072 	i += 1;
   1073 	termNode = &nodes[i];
   1074 	i += 1;
   1075 	RF_ASSERT(i == (nWndNodes + nWmirNodes + 3));
   1076 
   1077 	/* this dag can commit immediately */
   1078 	dag_h->numCommitNodes = 1;
   1079 	dag_h->numCommits = 0;
   1080 	dag_h->numSuccedents = 1;
   1081 
   1082 	/* initialize the commit, unblock, and term nodes */
   1083 	rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc,
   1084 		    rf_NullNodeUndoFunc, NULL, (nWndNodes + nWmirNodes),
   1085 		    0, 0, 0, dag_h, "Cmt", allocList);
   1086 	rf_InitNode(unblockNode, rf_wait, RF_FALSE, rf_NullNodeFunc,
   1087 		    rf_NullNodeUndoFunc, NULL, 1, (nWndNodes + nWmirNodes),
   1088 		    0, 0, dag_h, "Nil", allocList);
   1089 	rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc,
   1090 		    rf_TerminateUndoFunc, NULL, 0, 1, 0, 0,
   1091 		    dag_h, "Trm", allocList);
   1092 
   1093 	/* initialize the wnd nodes */
   1094 	if (nWndNodes > 0) {
   1095 		pda = asmap->physInfo;
   1096 		for (i = 0; i < nWndNodes; i++) {
   1097 			rf_InitNode(&wndNode[i], rf_wait, RF_FALSE,
   1098 				    rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
   1099 				    rf_GenericWakeupFunc, 1, 1, 4, 0,
   1100 				    dag_h, "Wpd", allocList);
   1101 			RF_ASSERT(pda != NULL);
   1102 			wndNode[i].params[0].p = pda;
   1103 			wndNode[i].params[1].p = pda->bufPtr;
   1104 			wndNode[i].params[2].v = parityStripeID;
   1105 			wndNode[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
   1106 			pda = pda->next;
   1107 		}
   1108 		RF_ASSERT(pda == NULL);
   1109 	}
   1110 	/* initialize the mirror nodes */
   1111 	if (nWmirNodes > 0) {
   1112 		pda = asmap->physInfo;
   1113 		pdaP = asmap->parityInfo;
   1114 		for (i = 0; i < nWmirNodes; i++) {
   1115 			rf_InitNode(&wmirNode[i], rf_wait, RF_FALSE,
   1116 				    rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
   1117 				    rf_GenericWakeupFunc, 1, 1, 4, 0,
   1118 				    dag_h, "Wsd", allocList);
   1119 			RF_ASSERT(pda != NULL);
   1120 			wmirNode[i].params[0].p = pdaP;
   1121 			wmirNode[i].params[1].p = pda->bufPtr;
   1122 			wmirNode[i].params[2].v = parityStripeID;
   1123 			wmirNode[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
   1124 			pda = pda->next;
   1125 			pdaP = pdaP->next;
   1126 		}
   1127 		RF_ASSERT(pda == NULL);
   1128 		RF_ASSERT(pdaP == NULL);
   1129 	}
   1130 	/* link the header node to the commit node */
   1131 	RF_ASSERT(dag_h->numSuccedents == 1);
   1132 	RF_ASSERT(commitNode->numAntecedents == 0);
   1133 	dag_h->succedents[0] = commitNode;
   1134 
   1135 	/* link the commit node to the write nodes */
   1136 	RF_ASSERT(commitNode->numSuccedents == (nWndNodes + nWmirNodes));
   1137 	for (i = 0; i < nWndNodes; i++) {
   1138 		RF_ASSERT(wndNode[i].numAntecedents == 1);
   1139 		commitNode->succedents[i] = &wndNode[i];
   1140 		wndNode[i].antecedents[0] = commitNode;
   1141 		wndNode[i].antType[0] = rf_control;
   1142 	}
   1143 	for (i = 0; i < nWmirNodes; i++) {
   1144 		RF_ASSERT(wmirNode[i].numAntecedents == 1);
   1145 		commitNode->succedents[i + nWndNodes] = &wmirNode[i];
   1146 		wmirNode[i].antecedents[0] = commitNode;
   1147 		wmirNode[i].antType[0] = rf_control;
   1148 	}
   1149 
   1150 	/* link the write nodes to the unblock node */
   1151 	RF_ASSERT(unblockNode->numAntecedents == (nWndNodes + nWmirNodes));
   1152 	for (i = 0; i < nWndNodes; i++) {
   1153 		RF_ASSERT(wndNode[i].numSuccedents == 1);
   1154 		wndNode[i].succedents[0] = unblockNode;
   1155 		unblockNode->antecedents[i] = &wndNode[i];
   1156 		unblockNode->antType[i] = rf_control;
   1157 	}
   1158 	for (i = 0; i < nWmirNodes; i++) {
   1159 		RF_ASSERT(wmirNode[i].numSuccedents == 1);
   1160 		wmirNode[i].succedents[0] = unblockNode;
   1161 		unblockNode->antecedents[i + nWndNodes] = &wmirNode[i];
   1162 		unblockNode->antType[i + nWndNodes] = rf_control;
   1163 	}
   1164 
   1165 	/* link the unblock node to the term node */
   1166 	RF_ASSERT(unblockNode->numSuccedents == 1);
   1167 	RF_ASSERT(termNode->numAntecedents == 1);
   1168 	RF_ASSERT(termNode->numSuccedents == 0);
   1169 	unblockNode->succedents[0] = termNode;
   1170 	termNode->antecedents[0] = unblockNode;
   1171 	termNode->antType[0] = rf_control;
   1172 }
   1173