Home | History | Annotate | Line # | Download | only in raidframe
rf_dagffwr.c revision 1.11
      1 /*	$NetBSD: rf_dagffwr.c,v 1.11 2003/07/01 22:43:59 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 1.11 2003/07/01 22:43:59 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(
     80     RF_Raid_t * raidPtr,
     81     RF_AccessStripeMap_t * asmap,
     82     RF_DagHeader_t * dag_h,
     83     void *bp,
     84     RF_RaidAccessFlags_t flags,
     85     RF_AllocListElem_t * allocList,
     86     RF_IoType_t type)
     87 {
     88 	rf_CreateNonredundantDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
     89 	    RF_IO_TYPE_WRITE);
     90 }
     91 
     92 void
     93 rf_CreateRAID0WriteDAG(
     94     RF_Raid_t * raidPtr,
     95     RF_AccessStripeMap_t * asmap,
     96     RF_DagHeader_t * dag_h,
     97     void *bp,
     98     RF_RaidAccessFlags_t flags,
     99     RF_AllocListElem_t * allocList,
    100     RF_IoType_t type)
    101 {
    102 	rf_CreateNonredundantDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
    103 	    RF_IO_TYPE_WRITE);
    104 }
    105 
    106 void
    107 rf_CreateSmallWriteDAG(
    108     RF_Raid_t * raidPtr,
    109     RF_AccessStripeMap_t * asmap,
    110     RF_DagHeader_t * dag_h,
    111     void *bp,
    112     RF_RaidAccessFlags_t flags,
    113     RF_AllocListElem_t * allocList)
    114 {
    115 	/* "normal" rollaway */
    116 	rf_CommonCreateSmallWriteDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
    117 	    &rf_xorFuncs, NULL);
    118 }
    119 
    120 void
    121 rf_CreateLargeWriteDAG(
    122     RF_Raid_t * raidPtr,
    123     RF_AccessStripeMap_t * asmap,
    124     RF_DagHeader_t * dag_h,
    125     void *bp,
    126     RF_RaidAccessFlags_t flags,
    127     RF_AllocListElem_t * allocList)
    128 {
    129 	/* "normal" rollaway */
    130 	rf_CommonCreateLargeWriteDAG(raidPtr, asmap, dag_h, bp, flags, allocList,
    131 	    1, rf_RegularXorFunc, RF_TRUE);
    132 }
    133 
    134 
    135 /******************************************************************************
    136  *
    137  * DAG creation code begins here
    138  */
    139 
    140 
    141 /******************************************************************************
    142  *
    143  * creates a DAG to perform a large-write operation:
    144  *
    145  *           / Rod \           / Wnd \
    146  * H -- block- Rod - Xor - Cmt - Wnd --- T
    147  *           \ Rod /          \  Wnp /
    148  *                             \[Wnq]/
    149  *
    150  * The XOR node also does the Q calculation in the P+Q architecture.
    151  * All nodes are before the commit node (Cmt) are assumed to be atomic and
    152  * undoable - or - they make no changes to permanent state.
    153  *
    154  * Rod = read old data
    155  * Cmt = commit node
    156  * Wnp = write new parity
    157  * Wnd = write new data
    158  * Wnq = write new "q"
    159  * [] denotes optional segments in the graph
    160  *
    161  * Parameters:  raidPtr   - description of the physical array
    162  *              asmap     - logical & physical addresses for this access
    163  *              bp        - buffer ptr (holds write data)
    164  *              flags     - general flags (e.g. disk locking)
    165  *              allocList - list of memory allocated in DAG creation
    166  *              nfaults   - number of faults array can tolerate
    167  *                          (equal to # redundancy units in stripe)
    168  *              redfuncs  - list of redundancy generating functions
    169  *
    170  *****************************************************************************/
    171 
    172 void
    173 rf_CommonCreateLargeWriteDAG(
    174     RF_Raid_t * raidPtr,
    175     RF_AccessStripeMap_t * asmap,
    176     RF_DagHeader_t * dag_h,
    177     void *bp,
    178     RF_RaidAccessFlags_t flags,
    179     RF_AllocListElem_t * allocList,
    180     int nfaults,
    181     int (*redFunc) (RF_DagNode_t *),
    182     int allowBufferRecycle)
    183 {
    184 	RF_DagNode_t *nodes, *wndNodes, *rodNodes, *xorNode, *wnpNode;
    185 	RF_DagNode_t *wnqNode, *blockNode, *commitNode, *termNode;
    186 	int     nWndNodes, nRodNodes, i, nodeNum, asmNum;
    187 	RF_AccessStripeMapHeader_t *new_asm_h[2];
    188 	RF_StripeNum_t parityStripeID;
    189 	char   *sosBuffer, *eosBuffer;
    190 	RF_ReconUnitNum_t which_ru;
    191 	RF_RaidLayout_t *layoutPtr;
    192 	RF_PhysDiskAddr_t *pda;
    193 
    194 	layoutPtr = &(raidPtr->Layout);
    195 	parityStripeID = rf_RaidAddressToParityStripeID(layoutPtr, asmap->raidAddress,
    196 	    &which_ru);
    197 
    198 	if (rf_dagDebug) {
    199 		printf("[Creating large-write DAG]\n");
    200 	}
    201 	dag_h->creator = "LargeWriteDAG";
    202 
    203 	dag_h->numCommitNodes = 1;
    204 	dag_h->numCommits = 0;
    205 	dag_h->numSuccedents = 1;
    206 
    207 	/* alloc the nodes: Wnd, xor, commit, block, term, and  Wnp */
    208 	nWndNodes = asmap->numStripeUnitsAccessed;
    209 	RF_CallocAndAdd(nodes, nWndNodes + 4 + nfaults, sizeof(RF_DagNode_t),
    210 	    (RF_DagNode_t *), allocList);
    211 	i = 0;
    212 	wndNodes = &nodes[i];
    213 	i += nWndNodes;
    214 	xorNode = &nodes[i];
    215 	i += 1;
    216 	wnpNode = &nodes[i];
    217 	i += 1;
    218 	blockNode = &nodes[i];
    219 	i += 1;
    220 	commitNode = &nodes[i];
    221 	i += 1;
    222 	termNode = &nodes[i];
    223 	i += 1;
    224 	if (nfaults == 2) {
    225 		wnqNode = &nodes[i];
    226 		i += 1;
    227 	} else {
    228 		wnqNode = NULL;
    229 	}
    230 	rf_MapUnaccessedPortionOfStripe(raidPtr, layoutPtr, asmap, dag_h, new_asm_h,
    231 	    &nRodNodes, &sosBuffer, &eosBuffer, allocList);
    232 	if (nRodNodes > 0) {
    233 		RF_CallocAndAdd(rodNodes, nRodNodes, sizeof(RF_DagNode_t),
    234 		    (RF_DagNode_t *), allocList);
    235 	} else {
    236 		rodNodes = NULL;
    237 	}
    238 
    239 	/* begin node initialization */
    240 	if (nRodNodes > 0) {
    241 		rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc,
    242 		    NULL, nRodNodes, 0, 0, 0, dag_h, "Nil", allocList);
    243 	} else {
    244 		rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc,
    245 		    NULL, 1, 0, 0, 0, dag_h, "Nil", allocList);
    246 	}
    247 
    248 	rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL,
    249 	    nWndNodes + nfaults, 1, 0, 0, dag_h, "Cmt", allocList);
    250 	rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL,
    251 	    0, nWndNodes + nfaults, 0, 0, dag_h, "Trm", allocList);
    252 
    253 	/* initialize the Rod nodes */
    254 	for (nodeNum = asmNum = 0; asmNum < 2; asmNum++) {
    255 		if (new_asm_h[asmNum]) {
    256 			pda = new_asm_h[asmNum]->stripeMap->physInfo;
    257 			while (pda) {
    258 				rf_InitNode(&rodNodes[nodeNum], rf_wait, RF_FALSE, rf_DiskReadFunc,
    259 				    rf_DiskReadUndoFunc, rf_GenericWakeupFunc, 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 				    0, 0, 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, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
    277 		    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnd", allocList);
    278 		RF_ASSERT(pda != NULL);
    279 		wndNodes[i].params[0].p = pda;
    280 		wndNodes[i].params[1].p = pda->bufPtr;
    281 		wndNodes[i].params[2].v = parityStripeID;
    282 		wndNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
    283 		pda = pda->next;
    284 	}
    285 
    286 	/* initialize the redundancy node */
    287 	if (nRodNodes > 0) {
    288 		rf_InitNode(xorNode, rf_wait, RF_FALSE, redFunc, rf_NullNodeUndoFunc, NULL, 1,
    289 		    nRodNodes, 2 * (nWndNodes + nRodNodes) + 1, nfaults, dag_h,
    290 		    "Xr ", allocList);
    291 	} else {
    292 		rf_InitNode(xorNode, rf_wait, RF_FALSE, redFunc, rf_NullNodeUndoFunc, NULL, 1,
    293 		    1, 2 * (nWndNodes + nRodNodes) + 1, nfaults, dag_h, "Xr ", allocList);
    294 	}
    295 	xorNode->flags |= RF_DAGNODE_FLAG_YIELD;
    296 	for (i = 0; i < nWndNodes; i++) {
    297 		xorNode->params[2 * i + 0] = wndNodes[i].params[0];	/* pda */
    298 		xorNode->params[2 * i + 1] = wndNodes[i].params[1];	/* buf ptr */
    299 	}
    300 	for (i = 0; i < nRodNodes; i++) {
    301 		xorNode->params[2 * (nWndNodes + i) + 0] = rodNodes[i].params[0];	/* pda */
    302 		xorNode->params[2 * (nWndNodes + i) + 1] = rodNodes[i].params[1];	/* buf ptr */
    303 	}
    304 	/* xor node needs to get at RAID information */
    305 	xorNode->params[2 * (nWndNodes + nRodNodes)].p = raidPtr;
    306 
    307 	/*
    308          * Look for an Rod node that reads a complete SU. If none, alloc a buffer
    309          * to receive the parity info. Note that we can't use a new data buffer
    310          * because it will not have gotten written when the xor occurs.
    311          */
    312 	if (allowBufferRecycle) {
    313 		for (i = 0; i < nRodNodes; i++) {
    314 			if (((RF_PhysDiskAddr_t *) rodNodes[i].params[0].p)->numSector == raidPtr->Layout.sectorsPerStripeUnit)
    315 				break;
    316 		}
    317 	}
    318 	if ((!allowBufferRecycle) || (i == nRodNodes)) {
    319 		RF_CallocAndAdd(xorNode->results[0], 1,
    320 		    rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit),
    321 		    (void *), allocList);
    322 	} else {
    323 		xorNode->results[0] = rodNodes[i].params[1].p;
    324 	}
    325 
    326 	/* initialize the Wnp node */
    327 	rf_InitNode(wnpNode, rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
    328 	    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnp", allocList);
    329 	wnpNode->params[0].p = asmap->parityInfo;
    330 	wnpNode->params[1].p = xorNode->results[0];
    331 	wnpNode->params[2].v = parityStripeID;
    332 	wnpNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
    333 	/* parityInfo must describe entire parity unit */
    334 	RF_ASSERT(asmap->parityInfo->next == NULL);
    335 
    336 	if (nfaults == 2) {
    337 		/*
    338 	         * We never try to recycle a buffer for the Q calcuation
    339 	         * in addition to the parity. This would cause two buffers
    340 	         * to get smashed during the P and Q calculation, guaranteeing
    341 	         * one would be wrong.
    342 	         */
    343 		RF_CallocAndAdd(xorNode->results[1], 1,
    344 		    rf_RaidAddressToByte(raidPtr, raidPtr->Layout.sectorsPerStripeUnit),
    345 		    (void *), allocList);
    346 		rf_InitNode(wnqNode, rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
    347 		    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wnq", allocList);
    348 		wnqNode->params[0].p = asmap->qInfo;
    349 		wnqNode->params[1].p = xorNode->results[1];
    350 		wnqNode->params[2].v = parityStripeID;
    351 		wnqNode->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
    352 		/* parityInfo must describe entire parity unit */
    353 		RF_ASSERT(asmap->parityInfo->next == NULL);
    354 	}
    355 	/*
    356          * Connect nodes to form graph.
    357          */
    358 
    359 	/* connect dag header to block node */
    360 	RF_ASSERT(blockNode->numAntecedents == 0);
    361 	dag_h->succedents[0] = blockNode;
    362 
    363 	if (nRodNodes > 0) {
    364 		/* connect the block node to the Rod nodes */
    365 		RF_ASSERT(blockNode->numSuccedents == nRodNodes);
    366 		RF_ASSERT(xorNode->numAntecedents == nRodNodes);
    367 		for (i = 0; i < nRodNodes; i++) {
    368 			RF_ASSERT(rodNodes[i].numAntecedents == 1);
    369 			blockNode->succedents[i] = &rodNodes[i];
    370 			rodNodes[i].antecedents[0] = blockNode;
    371 			rodNodes[i].antType[0] = rf_control;
    372 
    373 			/* connect the Rod nodes to the Xor node */
    374 			RF_ASSERT(rodNodes[i].numSuccedents == 1);
    375 			rodNodes[i].succedents[0] = xorNode;
    376 			xorNode->antecedents[i] = &rodNodes[i];
    377 			xorNode->antType[i] = rf_trueData;
    378 		}
    379 	} else {
    380 		/* connect the block node to the Xor node */
    381 		RF_ASSERT(blockNode->numSuccedents == 1);
    382 		RF_ASSERT(xorNode->numAntecedents == 1);
    383 		blockNode->succedents[0] = xorNode;
    384 		xorNode->antecedents[0] = blockNode;
    385 		xorNode->antType[0] = rf_control;
    386 	}
    387 
    388 	/* connect the xor node to the commit node */
    389 	RF_ASSERT(xorNode->numSuccedents == 1);
    390 	RF_ASSERT(commitNode->numAntecedents == 1);
    391 	xorNode->succedents[0] = commitNode;
    392 	commitNode->antecedents[0] = xorNode;
    393 	commitNode->antType[0] = rf_control;
    394 
    395 	/* connect the commit node to the write nodes */
    396 	RF_ASSERT(commitNode->numSuccedents == nWndNodes + nfaults);
    397 	for (i = 0; i < nWndNodes; i++) {
    398 		RF_ASSERT(wndNodes->numAntecedents == 1);
    399 		commitNode->succedents[i] = &wndNodes[i];
    400 		wndNodes[i].antecedents[0] = commitNode;
    401 		wndNodes[i].antType[0] = rf_control;
    402 	}
    403 	RF_ASSERT(wnpNode->numAntecedents == 1);
    404 	commitNode->succedents[nWndNodes] = wnpNode;
    405 	wnpNode->antecedents[0] = commitNode;
    406 	wnpNode->antType[0] = rf_trueData;
    407 	if (nfaults == 2) {
    408 		RF_ASSERT(wnqNode->numAntecedents == 1);
    409 		commitNode->succedents[nWndNodes + 1] = wnqNode;
    410 		wnqNode->antecedents[0] = commitNode;
    411 		wnqNode->antType[0] = rf_trueData;
    412 	}
    413 	/* connect the write nodes to the term node */
    414 	RF_ASSERT(termNode->numAntecedents == nWndNodes + nfaults);
    415 	RF_ASSERT(termNode->numSuccedents == 0);
    416 	for (i = 0; i < nWndNodes; i++) {
    417 		RF_ASSERT(wndNodes->numSuccedents == 1);
    418 		wndNodes[i].succedents[0] = termNode;
    419 		termNode->antecedents[i] = &wndNodes[i];
    420 		termNode->antType[i] = rf_control;
    421 	}
    422 	RF_ASSERT(wnpNode->numSuccedents == 1);
    423 	wnpNode->succedents[0] = termNode;
    424 	termNode->antecedents[nWndNodes] = wnpNode;
    425 	termNode->antType[nWndNodes] = rf_control;
    426 	if (nfaults == 2) {
    427 		RF_ASSERT(wnqNode->numSuccedents == 1);
    428 		wnqNode->succedents[0] = termNode;
    429 		termNode->antecedents[nWndNodes + 1] = wnqNode;
    430 		termNode->antType[nWndNodes + 1] = rf_control;
    431 	}
    432 }
    433 /******************************************************************************
    434  *
    435  * creates a DAG to perform a small-write operation (either raid 5 or pq),
    436  * which is as follows:
    437  *
    438  * Hdr -> Nil -> Rop -> Xor -> Cmt ----> Wnp [Unp] --> Trm
    439  *            \- Rod X      /     \----> Wnd [Und]-/
    440  *           [\- Rod X     /       \---> Wnd [Und]-/]
    441  *           [\- Roq -> Q /         \--> Wnq [Unq]-/]
    442  *
    443  * Rop = read old parity
    444  * Rod = read old data
    445  * Roq = read old "q"
    446  * Cmt = commit node
    447  * Und = unlock data disk
    448  * Unp = unlock parity disk
    449  * Unq = unlock q disk
    450  * Wnp = write new parity
    451  * Wnd = write new data
    452  * Wnq = write new "q"
    453  * [ ] denotes optional segments in the graph
    454  *
    455  * Parameters:  raidPtr   - description of the physical array
    456  *              asmap     - logical & physical addresses for this access
    457  *              bp        - buffer ptr (holds write data)
    458  *              flags     - general flags (e.g. disk locking)
    459  *              allocList - list of memory allocated in DAG creation
    460  *              pfuncs    - list of parity generating functions
    461  *              qfuncs    - list of q generating functions
    462  *
    463  * A null qfuncs indicates single fault tolerant
    464  *****************************************************************************/
    465 
    466 void
    467 rf_CommonCreateSmallWriteDAG(
    468     RF_Raid_t * raidPtr,
    469     RF_AccessStripeMap_t * asmap,
    470     RF_DagHeader_t * dag_h,
    471     void *bp,
    472     RF_RaidAccessFlags_t flags,
    473     RF_AllocListElem_t * allocList,
    474     const RF_RedFuncs_t * pfuncs,
    475     const RF_RedFuncs_t * qfuncs)
    476 {
    477 	RF_DagNode_t *readDataNodes, *readParityNodes, *readQNodes, *termNode;
    478 	RF_DagNode_t *unlockDataNodes, *unlockParityNodes, *unlockQNodes;
    479 	RF_DagNode_t *xorNodes, *qNodes, *blockNode, *commitNode, *nodes;
    480 	RF_DagNode_t *writeDataNodes, *writeParityNodes, *writeQNodes;
    481 	int     i, j, nNodes, totalNumNodes, lu_flag;
    482 	RF_ReconUnitNum_t which_ru;
    483 	int     (*func) (RF_DagNode_t *), (*undoFunc) (RF_DagNode_t *);
    484 	int     (*qfunc) (RF_DagNode_t *);
    485 	int     numDataNodes, numParityNodes;
    486 	RF_StripeNum_t parityStripeID;
    487 	RF_PhysDiskAddr_t *pda;
    488 	char   *name, *qname;
    489 	long    nfaults;
    490 
    491 	nfaults = qfuncs ? 2 : 1;
    492 	lu_flag = (rf_enableAtomicRMW) ? 1 : 0;	/* lock/unlock flag */
    493 
    494 	parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout),
    495 	    asmap->raidAddress, &which_ru);
    496 	pda = asmap->physInfo;
    497 	numDataNodes = asmap->numStripeUnitsAccessed;
    498 	numParityNodes = (asmap->parityInfo->next) ? 2 : 1;
    499 
    500 	if (rf_dagDebug) {
    501 		printf("[Creating small-write DAG]\n");
    502 	}
    503 	RF_ASSERT(numDataNodes > 0);
    504 	dag_h->creator = "SmallWriteDAG";
    505 
    506 	dag_h->numCommitNodes = 1;
    507 	dag_h->numCommits = 0;
    508 	dag_h->numSuccedents = 1;
    509 
    510 	/*
    511          * DAG creation occurs in four steps:
    512          * 1. count the number of nodes in the DAG
    513          * 2. create the nodes
    514          * 3. initialize the nodes
    515          * 4. connect the nodes
    516          */
    517 
    518 	/*
    519          * Step 1. compute number of nodes in the graph
    520          */
    521 
    522 	/* number of nodes: a read and write for each data unit a redundancy
    523 	 * computation node for each parity node (nfaults * nparity) a read
    524 	 * and write for each parity unit a block and commit node (2) a
    525 	 * terminate node if atomic RMW an unlock node for each data unit,
    526 	 * redundancy unit */
    527 	totalNumNodes = (2 * numDataNodes) + (nfaults * numParityNodes)
    528 	    + (nfaults * 2 * numParityNodes) + 3;
    529 	if (lu_flag) {
    530 		totalNumNodes += (numDataNodes + (nfaults * numParityNodes));
    531 	}
    532 	/*
    533          * Step 2. create the nodes
    534          */
    535 	RF_CallocAndAdd(nodes, totalNumNodes, sizeof(RF_DagNode_t),
    536 	    (RF_DagNode_t *), allocList);
    537 	i = 0;
    538 	blockNode = &nodes[i];
    539 	i += 1;
    540 	commitNode = &nodes[i];
    541 	i += 1;
    542 	readDataNodes = &nodes[i];
    543 	i += numDataNodes;
    544 	readParityNodes = &nodes[i];
    545 	i += numParityNodes;
    546 	writeDataNodes = &nodes[i];
    547 	i += numDataNodes;
    548 	writeParityNodes = &nodes[i];
    549 	i += numParityNodes;
    550 	xorNodes = &nodes[i];
    551 	i += numParityNodes;
    552 	termNode = &nodes[i];
    553 	i += 1;
    554 	if (lu_flag) {
    555 		unlockDataNodes = &nodes[i];
    556 		i += numDataNodes;
    557 		unlockParityNodes = &nodes[i];
    558 		i += numParityNodes;
    559 	} else {
    560 		unlockDataNodes = unlockParityNodes = NULL;
    561 	}
    562 	if (nfaults == 2) {
    563 		readQNodes = &nodes[i];
    564 		i += numParityNodes;
    565 		writeQNodes = &nodes[i];
    566 		i += numParityNodes;
    567 		qNodes = &nodes[i];
    568 		i += numParityNodes;
    569 		if (lu_flag) {
    570 			unlockQNodes = &nodes[i];
    571 			i += numParityNodes;
    572 		} else {
    573 			unlockQNodes = NULL;
    574 		}
    575 	} else {
    576 		readQNodes = writeQNodes = qNodes = unlockQNodes = NULL;
    577 	}
    578 	RF_ASSERT(i == totalNumNodes);
    579 
    580 	/*
    581          * Step 3. initialize the nodes
    582          */
    583 	/* initialize block node (Nil) */
    584 	nNodes = numDataNodes + (nfaults * numParityNodes);
    585 	rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc,
    586 	    NULL, nNodes, 0, 0, 0, dag_h, "Nil", allocList);
    587 
    588 	/* initialize commit node (Cmt) */
    589 	rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc,
    590 	    NULL, nNodes, (nfaults * numParityNodes), 0, 0, dag_h, "Cmt", allocList);
    591 
    592 	/* initialize terminate node (Trm) */
    593 	rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc,
    594 	    NULL, 0, nNodes, 0, 0, dag_h, "Trm", allocList);
    595 
    596 	/* initialize nodes which read old data (Rod) */
    597 	for (i = 0; i < numDataNodes; i++) {
    598 		rf_InitNode(&readDataNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc,
    599 		    rf_GenericWakeupFunc, (nfaults * numParityNodes), 1, 4, 0, dag_h,
    600 		    "Rod", allocList);
    601 		RF_ASSERT(pda != NULL);
    602 		/* physical disk addr desc */
    603 		readDataNodes[i].params[0].p = pda;
    604 		/* buffer to hold old data */
    605 		readDataNodes[i].params[1].p = rf_AllocBuffer(raidPtr,
    606 		    dag_h, pda, allocList);
    607 		readDataNodes[i].params[2].v = parityStripeID;
    608 		readDataNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    609 		    lu_flag, 0, which_ru);
    610 		pda = pda->next;
    611 		for (j = 0; j < readDataNodes[i].numSuccedents; j++) {
    612 			readDataNodes[i].propList[j] = NULL;
    613 		}
    614 	}
    615 
    616 	/* initialize nodes which read old parity (Rop) */
    617 	pda = asmap->parityInfo;
    618 	i = 0;
    619 	for (i = 0; i < numParityNodes; i++) {
    620 		RF_ASSERT(pda != NULL);
    621 		rf_InitNode(&readParityNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc,
    622 		    rf_DiskReadUndoFunc, rf_GenericWakeupFunc, numParityNodes, 1, 4,
    623 		    0, dag_h, "Rop", allocList);
    624 		readParityNodes[i].params[0].p = pda;
    625 		/* buffer to hold old parity */
    626 		readParityNodes[i].params[1].p = rf_AllocBuffer(raidPtr,
    627 		    dag_h, pda, allocList);
    628 		readParityNodes[i].params[2].v = parityStripeID;
    629 		readParityNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    630 		    lu_flag, 0, which_ru);
    631 		pda = pda->next;
    632 		for (j = 0; j < readParityNodes[i].numSuccedents; j++) {
    633 			readParityNodes[i].propList[0] = NULL;
    634 		}
    635 	}
    636 
    637 	/* initialize nodes which read old Q (Roq) */
    638 	if (nfaults == 2) {
    639 		pda = asmap->qInfo;
    640 		for (i = 0; i < numParityNodes; i++) {
    641 			RF_ASSERT(pda != NULL);
    642 			rf_InitNode(&readQNodes[i], rf_wait, RF_FALSE, rf_DiskReadFunc, rf_DiskReadUndoFunc,
    643 			    rf_GenericWakeupFunc, numParityNodes, 1, 4, 0, dag_h, "Roq", allocList);
    644 			readQNodes[i].params[0].p = pda;
    645 			/* buffer to hold old Q */
    646 			readQNodes[i].params[1].p = rf_AllocBuffer(raidPtr, dag_h, pda,
    647 			    allocList);
    648 			readQNodes[i].params[2].v = parityStripeID;
    649 			readQNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    650 			    lu_flag, 0, which_ru);
    651 			pda = pda->next;
    652 			for (j = 0; j < readQNodes[i].numSuccedents; j++) {
    653 				readQNodes[i].propList[0] = NULL;
    654 			}
    655 		}
    656 	}
    657 	/* initialize nodes which write new data (Wnd) */
    658 	pda = asmap->physInfo;
    659 	for (i = 0; i < numDataNodes; i++) {
    660 		RF_ASSERT(pda != NULL);
    661 		rf_InitNode(&writeDataNodes[i], rf_wait, RF_FALSE, rf_DiskWriteFunc,
    662 		    rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h,
    663 		    "Wnd", allocList);
    664 		/* physical disk addr desc */
    665 		writeDataNodes[i].params[0].p = pda;
    666 		/* buffer holding new data to be written */
    667 		writeDataNodes[i].params[1].p = pda->bufPtr;
    668 		writeDataNodes[i].params[2].v = parityStripeID;
    669 		writeDataNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    670 		    0, 0, which_ru);
    671 		if (lu_flag) {
    672 			/* initialize node to unlock the disk queue */
    673 			rf_InitNode(&unlockDataNodes[i], rf_wait, RF_FALSE, rf_DiskUnlockFunc,
    674 			    rf_DiskUnlockUndoFunc, rf_GenericWakeupFunc, 1, 1, 2, 0, dag_h,
    675 			    "Und", allocList);
    676 			/* physical disk addr desc */
    677 			unlockDataNodes[i].params[0].p = pda;
    678 			unlockDataNodes[i].params[1].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    679 			    0, lu_flag, which_ru);
    680 		}
    681 		pda = pda->next;
    682 	}
    683 
    684 	/*
    685          * Initialize nodes which compute new parity and Q.
    686          */
    687 	/*
    688          * We use the simple XOR func in the double-XOR case, and when
    689          * we're accessing only a portion of one stripe unit. The distinction
    690          * between the two is that the regular XOR func assumes that the targbuf
    691          * is a full SU in size, and examines the pda associated with the buffer
    692          * to decide where within the buffer to XOR the data, whereas
    693          * the simple XOR func just XORs the data into the start of the buffer.
    694          */
    695 	if ((numParityNodes == 2) || ((numDataNodes == 1)
    696 		&& (asmap->totalSectorsAccessed < 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, undoFunc, NULL,
    728 			    1, (numDataNodes + numParityNodes), 7, 1, dag_h, name, allocList);
    729 			xorNodes[i].flags |= RF_DAGNODE_FLAG_YIELD;
    730 			xorNodes[i].params[0] = readDataNodes[i].params[0];
    731 			xorNodes[i].params[1] = readDataNodes[i].params[1];
    732 			xorNodes[i].params[2] = readParityNodes[i].params[0];
    733 			xorNodes[i].params[3] = readParityNodes[i].params[1];
    734 			xorNodes[i].params[4] = writeDataNodes[i].params[0];
    735 			xorNodes[i].params[5] = writeDataNodes[i].params[1];
    736 			xorNodes[i].params[6].p = raidPtr;
    737 			/* use old parity buf as target buf */
    738 			xorNodes[i].results[0] = readParityNodes[i].params[1].p;
    739 			if (nfaults == 2) {
    740 				/* note: no wakeup func for qor */
    741 				rf_InitNode(&qNodes[i], rf_wait, RF_FALSE, qfunc, undoFunc, NULL, 1,
    742 				    (numDataNodes + numParityNodes), 7, 1, dag_h, qname, allocList);
    743 				qNodes[i].params[0] = readDataNodes[i].params[0];
    744 				qNodes[i].params[1] = readDataNodes[i].params[1];
    745 				qNodes[i].params[2] = readQNodes[i].params[0];
    746 				qNodes[i].params[3] = readQNodes[i].params[1];
    747 				qNodes[i].params[4] = writeDataNodes[i].params[0];
    748 				qNodes[i].params[5] = writeDataNodes[i].params[1];
    749 				qNodes[i].params[6].p = raidPtr;
    750 				/* use old Q buf as target buf */
    751 				qNodes[i].results[0] = readQNodes[i].params[1].p;
    752 			}
    753 		}
    754 	} else {
    755 		/* there is only one xor node in this case */
    756 		rf_InitNode(&xorNodes[0], rf_wait, RF_FALSE, func, undoFunc, NULL, 1,
    757 		    (numDataNodes + numParityNodes),
    758 		    (2 * (numDataNodes + numDataNodes + 1) + 1), 1, dag_h, name, allocList);
    759 		xorNodes[0].flags |= RF_DAGNODE_FLAG_YIELD;
    760 		for (i = 0; i < numDataNodes + 1; i++) {
    761 			/* set up params related to Rod and Rop nodes */
    762 			xorNodes[0].params[2 * i + 0] = readDataNodes[i].params[0];	/* pda */
    763 			xorNodes[0].params[2 * i + 1] = readDataNodes[i].params[1];	/* buffer ptr */
    764 		}
    765 		for (i = 0; i < numDataNodes; i++) {
    766 			/* set up params related to Wnd and Wnp nodes */
    767 			xorNodes[0].params[2 * (numDataNodes + 1 + i) + 0] =	/* pda */
    768 			    writeDataNodes[i].params[0];
    769 			xorNodes[0].params[2 * (numDataNodes + 1 + i) + 1] =	/* buffer ptr */
    770 			    writeDataNodes[i].params[1];
    771 		}
    772 		/* xor node needs to get at RAID information */
    773 		xorNodes[0].params[2 * (numDataNodes + numDataNodes + 1)].p = raidPtr;
    774 		xorNodes[0].results[0] = readParityNodes[0].params[1].p;
    775 		if (nfaults == 2) {
    776 			rf_InitNode(&qNodes[0], rf_wait, RF_FALSE, qfunc, undoFunc, NULL, 1,
    777 			    (numDataNodes + numParityNodes),
    778 			    (2 * (numDataNodes + numDataNodes + 1) + 1), 1, dag_h,
    779 			    qname, allocList);
    780 			for (i = 0; i < numDataNodes; i++) {
    781 				/* set up params related to Rod */
    782 				qNodes[0].params[2 * i + 0] = readDataNodes[i].params[0];	/* pda */
    783 				qNodes[0].params[2 * i + 1] = readDataNodes[i].params[1];	/* buffer ptr */
    784 			}
    785 			/* and read old q */
    786 			qNodes[0].params[2 * numDataNodes + 0] =	/* pda */
    787 			    readQNodes[0].params[0];
    788 			qNodes[0].params[2 * numDataNodes + 1] =	/* buffer ptr */
    789 			    readQNodes[0].params[1];
    790 			for (i = 0; i < numDataNodes; i++) {
    791 				/* set up params related to Wnd nodes */
    792 				qNodes[0].params[2 * (numDataNodes + 1 + i) + 0] =	/* pda */
    793 				    writeDataNodes[i].params[0];
    794 				qNodes[0].params[2 * (numDataNodes + 1 + i) + 1] =	/* buffer ptr */
    795 				    writeDataNodes[i].params[1];
    796 			}
    797 			/* xor node needs to get at RAID information */
    798 			qNodes[0].params[2 * (numDataNodes + numDataNodes + 1)].p = raidPtr;
    799 			qNodes[0].results[0] = readQNodes[0].params[1].p;
    800 		}
    801 	}
    802 
    803 	/* initialize nodes which write new parity (Wnp) */
    804 	pda = asmap->parityInfo;
    805 	for (i = 0; i < numParityNodes; i++) {
    806 		rf_InitNode(&writeParityNodes[i], rf_wait, RF_FALSE, rf_DiskWriteFunc,
    807 		    rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h,
    808 		    "Wnp", allocList);
    809 		RF_ASSERT(pda != NULL);
    810 		writeParityNodes[i].params[0].p = pda;	/* param 1 (bufPtr)
    811 							 * filled in by xor node */
    812 		writeParityNodes[i].params[1].p = xorNodes[i].results[0];	/* buffer pointer for
    813 										 * parity write
    814 										 * operation */
    815 		writeParityNodes[i].params[2].v = parityStripeID;
    816 		writeParityNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    817 		    0, 0, which_ru);
    818 		if (lu_flag) {
    819 			/* initialize node to unlock the disk queue */
    820 			rf_InitNode(&unlockParityNodes[i], rf_wait, RF_FALSE, rf_DiskUnlockFunc,
    821 			    rf_DiskUnlockUndoFunc, rf_GenericWakeupFunc, 1, 1, 2, 0, dag_h,
    822 			    "Unp", allocList);
    823 			unlockParityNodes[i].params[0].p = pda;	/* physical disk addr
    824 								 * desc */
    825 			unlockParityNodes[i].params[1].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    826 			    0, lu_flag, which_ru);
    827 		}
    828 		pda = pda->next;
    829 	}
    830 
    831 	/* initialize nodes which write new Q (Wnq) */
    832 	if (nfaults == 2) {
    833 		pda = asmap->qInfo;
    834 		for (i = 0; i < numParityNodes; i++) {
    835 			rf_InitNode(&writeQNodes[i], rf_wait, RF_FALSE, rf_DiskWriteFunc,
    836 			    rf_DiskWriteUndoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h,
    837 			    "Wnq", allocList);
    838 			RF_ASSERT(pda != NULL);
    839 			writeQNodes[i].params[0].p = pda;	/* param 1 (bufPtr)
    840 								 * filled in by xor node */
    841 			writeQNodes[i].params[1].p = qNodes[i].results[0];	/* buffer pointer for
    842 										 * parity write
    843 										 * operation */
    844 			writeQNodes[i].params[2].v = parityStripeID;
    845 			writeQNodes[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    846 			    0, 0, which_ru);
    847 			if (lu_flag) {
    848 				/* initialize node to unlock the disk queue */
    849 				rf_InitNode(&unlockQNodes[i], rf_wait, RF_FALSE, rf_DiskUnlockFunc,
    850 				    rf_DiskUnlockUndoFunc, rf_GenericWakeupFunc, 1, 1, 2, 0, dag_h,
    851 				    "Unq", allocList);
    852 				unlockQNodes[i].params[0].p = pda;	/* physical disk addr
    853 									 * desc */
    854 				unlockQNodes[i].params[1].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY,
    855 				    0, lu_flag, which_ru);
    856 			}
    857 			pda = pda->next;
    858 		}
    859 	}
    860 	/*
    861          * Step 4. connect the nodes.
    862          */
    863 
    864 	/* connect header to block node */
    865 	dag_h->succedents[0] = blockNode;
    866 
    867 	/* connect block node to read old data nodes */
    868 	RF_ASSERT(blockNode->numSuccedents == (numDataNodes + (numParityNodes * nfaults)));
    869 	for (i = 0; i < numDataNodes; i++) {
    870 		blockNode->succedents[i] = &readDataNodes[i];
    871 		RF_ASSERT(readDataNodes[i].numAntecedents == 1);
    872 		readDataNodes[i].antecedents[0] = blockNode;
    873 		readDataNodes[i].antType[0] = rf_control;
    874 	}
    875 
    876 	/* connect block node to read old parity nodes */
    877 	for (i = 0; i < numParityNodes; i++) {
    878 		blockNode->succedents[numDataNodes + i] = &readParityNodes[i];
    879 		RF_ASSERT(readParityNodes[i].numAntecedents == 1);
    880 		readParityNodes[i].antecedents[0] = blockNode;
    881 		readParityNodes[i].antType[0] = rf_control;
    882 	}
    883 
    884 	/* connect block node to read old Q nodes */
    885 	if (nfaults == 2) {
    886 		for (i = 0; i < numParityNodes; i++) {
    887 			blockNode->succedents[numDataNodes + numParityNodes + i] = &readQNodes[i];
    888 			RF_ASSERT(readQNodes[i].numAntecedents == 1);
    889 			readQNodes[i].antecedents[0] = blockNode;
    890 			readQNodes[i].antType[0] = rf_control;
    891 		}
    892 	}
    893 	/* connect read old data nodes to xor nodes */
    894 	for (i = 0; i < numDataNodes; i++) {
    895 		RF_ASSERT(readDataNodes[i].numSuccedents == (nfaults * numParityNodes));
    896 		for (j = 0; j < numParityNodes; j++) {
    897 			RF_ASSERT(xorNodes[j].numAntecedents == numDataNodes + numParityNodes);
    898 			readDataNodes[i].succedents[j] = &xorNodes[j];
    899 			xorNodes[j].antecedents[i] = &readDataNodes[i];
    900 			xorNodes[j].antType[i] = rf_trueData;
    901 		}
    902 	}
    903 
    904 	/* connect read old data nodes to q nodes */
    905 	if (nfaults == 2) {
    906 		for (i = 0; i < numDataNodes; i++) {
    907 			for (j = 0; j < numParityNodes; j++) {
    908 				RF_ASSERT(qNodes[j].numAntecedents == numDataNodes + numParityNodes);
    909 				readDataNodes[i].succedents[numParityNodes + j] = &qNodes[j];
    910 				qNodes[j].antecedents[i] = &readDataNodes[i];
    911 				qNodes[j].antType[i] = rf_trueData;
    912 			}
    913 		}
    914 	}
    915 	/* connect read old parity nodes to xor nodes */
    916 	for (i = 0; i < numParityNodes; i++) {
    917 		RF_ASSERT(readParityNodes[i].numSuccedents == numParityNodes);
    918 		for (j = 0; j < numParityNodes; j++) {
    919 			readParityNodes[i].succedents[j] = &xorNodes[j];
    920 			xorNodes[j].antecedents[numDataNodes + i] = &readParityNodes[i];
    921 			xorNodes[j].antType[numDataNodes + i] = rf_trueData;
    922 		}
    923 	}
    924 
    925 	/* connect read old q nodes to q nodes */
    926 	if (nfaults == 2) {
    927 		for (i = 0; i < numParityNodes; i++) {
    928 			RF_ASSERT(readParityNodes[i].numSuccedents == numParityNodes);
    929 			for (j = 0; j < numParityNodes; j++) {
    930 				readQNodes[i].succedents[j] = &qNodes[j];
    931 				qNodes[j].antecedents[numDataNodes + i] = &readQNodes[i];
    932 				qNodes[j].antType[numDataNodes + i] = rf_trueData;
    933 			}
    934 		}
    935 	}
    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 	/* connect q nodes to commit node */
    946 	if (nfaults == 2) {
    947 		for (i = 0; i < numParityNodes; i++) {
    948 			RF_ASSERT(qNodes[i].numSuccedents == 1);
    949 			qNodes[i].succedents[0] = commitNode;
    950 			commitNode->antecedents[i + numParityNodes] = &qNodes[i];
    951 			commitNode->antType[i + numParityNodes] = rf_control;
    952 		}
    953 	}
    954 	/* connect commit node to write nodes */
    955 	RF_ASSERT(commitNode->numSuccedents == (numDataNodes + (nfaults * numParityNodes)));
    956 	for (i = 0; i < numDataNodes; i++) {
    957 		RF_ASSERT(writeDataNodes[i].numAntecedents == 1);
    958 		commitNode->succedents[i] = &writeDataNodes[i];
    959 		writeDataNodes[i].antecedents[0] = commitNode;
    960 		writeDataNodes[i].antType[0] = rf_trueData;
    961 	}
    962 	for (i = 0; i < numParityNodes; i++) {
    963 		RF_ASSERT(writeParityNodes[i].numAntecedents == 1);
    964 		commitNode->succedents[i + numDataNodes] = &writeParityNodes[i];
    965 		writeParityNodes[i].antecedents[0] = commitNode;
    966 		writeParityNodes[i].antType[0] = rf_trueData;
    967 	}
    968 	if (nfaults == 2) {
    969 		for (i = 0; i < numParityNodes; i++) {
    970 			RF_ASSERT(writeQNodes[i].numAntecedents == 1);
    971 			commitNode->succedents[i + numDataNodes + numParityNodes] = &writeQNodes[i];
    972 			writeQNodes[i].antecedents[0] = commitNode;
    973 			writeQNodes[i].antType[0] = rf_trueData;
    974 		}
    975 	}
    976 	RF_ASSERT(termNode->numAntecedents == (numDataNodes + (nfaults * numParityNodes)));
    977 	RF_ASSERT(termNode->numSuccedents == 0);
    978 	for (i = 0; i < numDataNodes; i++) {
    979 		if (lu_flag) {
    980 			/* connect write new data nodes to unlock nodes */
    981 			RF_ASSERT(writeDataNodes[i].numSuccedents == 1);
    982 			RF_ASSERT(unlockDataNodes[i].numAntecedents == 1);
    983 			writeDataNodes[i].succedents[0] = &unlockDataNodes[i];
    984 			unlockDataNodes[i].antecedents[0] = &writeDataNodes[i];
    985 			unlockDataNodes[i].antType[0] = rf_control;
    986 
    987 			/* connect unlock nodes to term node */
    988 			RF_ASSERT(unlockDataNodes[i].numSuccedents == 1);
    989 			unlockDataNodes[i].succedents[0] = termNode;
    990 			termNode->antecedents[i] = &unlockDataNodes[i];
    991 			termNode->antType[i] = rf_control;
    992 		} else {
    993 			/* connect write new data nodes to term node */
    994 			RF_ASSERT(writeDataNodes[i].numSuccedents == 1);
    995 			RF_ASSERT(termNode->numAntecedents == (numDataNodes + (nfaults * numParityNodes)));
    996 			writeDataNodes[i].succedents[0] = termNode;
    997 			termNode->antecedents[i] = &writeDataNodes[i];
    998 			termNode->antType[i] = rf_control;
    999 		}
   1000 	}
   1001 
   1002 	for (i = 0; i < numParityNodes; i++) {
   1003 		if (lu_flag) {
   1004 			/* connect write new parity nodes to unlock nodes */
   1005 			RF_ASSERT(writeParityNodes[i].numSuccedents == 1);
   1006 			RF_ASSERT(unlockParityNodes[i].numAntecedents == 1);
   1007 			writeParityNodes[i].succedents[0] = &unlockParityNodes[i];
   1008 			unlockParityNodes[i].antecedents[0] = &writeParityNodes[i];
   1009 			unlockParityNodes[i].antType[0] = rf_control;
   1010 
   1011 			/* connect unlock nodes to term node */
   1012 			RF_ASSERT(unlockParityNodes[i].numSuccedents == 1);
   1013 			unlockParityNodes[i].succedents[0] = termNode;
   1014 			termNode->antecedents[numDataNodes + i] = &unlockParityNodes[i];
   1015 			termNode->antType[numDataNodes + i] = rf_control;
   1016 		} else {
   1017 			RF_ASSERT(writeParityNodes[i].numSuccedents == 1);
   1018 			writeParityNodes[i].succedents[0] = termNode;
   1019 			termNode->antecedents[numDataNodes + i] = &writeParityNodes[i];
   1020 			termNode->antType[numDataNodes + i] = rf_control;
   1021 		}
   1022 	}
   1023 
   1024 	if (nfaults == 2) {
   1025 		for (i = 0; i < numParityNodes; i++) {
   1026 			if (lu_flag) {
   1027 				/* connect write new Q nodes to unlock nodes */
   1028 				RF_ASSERT(writeQNodes[i].numSuccedents == 1);
   1029 				RF_ASSERT(unlockQNodes[i].numAntecedents == 1);
   1030 				writeQNodes[i].succedents[0] = &unlockQNodes[i];
   1031 				unlockQNodes[i].antecedents[0] = &writeQNodes[i];
   1032 				unlockQNodes[i].antType[0] = rf_control;
   1033 
   1034 				/* connect unlock nodes to unblock node */
   1035 				RF_ASSERT(unlockQNodes[i].numSuccedents == 1);
   1036 				unlockQNodes[i].succedents[0] = termNode;
   1037 				termNode->antecedents[numDataNodes + numParityNodes + i] = &unlockQNodes[i];
   1038 				termNode->antType[numDataNodes + numParityNodes + i] = rf_control;
   1039 			} else {
   1040 				RF_ASSERT(writeQNodes[i].numSuccedents == 1);
   1041 				writeQNodes[i].succedents[0] = termNode;
   1042 				termNode->antecedents[numDataNodes + numParityNodes + i] = &writeQNodes[i];
   1043 				termNode->antType[numDataNodes + numParityNodes + i] = rf_control;
   1044 			}
   1045 		}
   1046 	}
   1047 }
   1048 
   1049 
   1050 /******************************************************************************
   1051  * create a write graph (fault-free or degraded) for RAID level 1
   1052  *
   1053  * Hdr -> Commit -> Wpd -> Nil -> Trm
   1054  *               -> Wsd ->
   1055  *
   1056  * The "Wpd" node writes data to the primary copy in the mirror pair
   1057  * The "Wsd" node writes data to the secondary copy in the mirror pair
   1058  *
   1059  * Parameters:  raidPtr   - description of the physical array
   1060  *              asmap     - logical & physical addresses for this access
   1061  *              bp        - buffer ptr (holds write data)
   1062  *              flags     - general flags (e.g. disk locking)
   1063  *              allocList - list of memory allocated in DAG creation
   1064  *****************************************************************************/
   1065 
   1066 void
   1067 rf_CreateRaidOneWriteDAG(
   1068     RF_Raid_t * raidPtr,
   1069     RF_AccessStripeMap_t * asmap,
   1070     RF_DagHeader_t * dag_h,
   1071     void *bp,
   1072     RF_RaidAccessFlags_t flags,
   1073     RF_AllocListElem_t * allocList)
   1074 {
   1075 	RF_DagNode_t *unblockNode, *termNode, *commitNode;
   1076 	RF_DagNode_t *nodes, *wndNode, *wmirNode;
   1077 	int     nWndNodes, nWmirNodes, i;
   1078 	RF_ReconUnitNum_t which_ru;
   1079 	RF_PhysDiskAddr_t *pda, *pdaP;
   1080 	RF_StripeNum_t parityStripeID;
   1081 
   1082 	parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout),
   1083 	    asmap->raidAddress, &which_ru);
   1084 	if (rf_dagDebug) {
   1085 		printf("[Creating RAID level 1 write DAG]\n");
   1086 	}
   1087 	dag_h->creator = "RaidOneWriteDAG";
   1088 
   1089 	/* 2 implies access not SU aligned */
   1090 	nWmirNodes = (asmap->parityInfo->next) ? 2 : 1;
   1091 	nWndNodes = (asmap->physInfo->next) ? 2 : 1;
   1092 
   1093 	/* alloc the Wnd nodes and the Wmir node */
   1094 	if (asmap->numDataFailed == 1)
   1095 		nWndNodes--;
   1096 	if (asmap->numParityFailed == 1)
   1097 		nWmirNodes--;
   1098 
   1099 	/* total number of nodes = nWndNodes + nWmirNodes + (commit + unblock
   1100 	 * + terminator) */
   1101 	RF_CallocAndAdd(nodes, nWndNodes + nWmirNodes + 3, sizeof(RF_DagNode_t),
   1102 	    (RF_DagNode_t *), allocList);
   1103 	i = 0;
   1104 	wndNode = &nodes[i];
   1105 	i += nWndNodes;
   1106 	wmirNode = &nodes[i];
   1107 	i += nWmirNodes;
   1108 	commitNode = &nodes[i];
   1109 	i += 1;
   1110 	unblockNode = &nodes[i];
   1111 	i += 1;
   1112 	termNode = &nodes[i];
   1113 	i += 1;
   1114 	RF_ASSERT(i == (nWndNodes + nWmirNodes + 3));
   1115 
   1116 	/* this dag can commit immediately */
   1117 	dag_h->numCommitNodes = 1;
   1118 	dag_h->numCommits = 0;
   1119 	dag_h->numSuccedents = 1;
   1120 
   1121 	/* initialize the commit, unblock, and term nodes */
   1122 	rf_InitNode(commitNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc,
   1123 	    NULL, (nWndNodes + nWmirNodes), 0, 0, 0, dag_h, "Cmt", allocList);
   1124 	rf_InitNode(unblockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc,
   1125 	    NULL, 1, (nWndNodes + nWmirNodes), 0, 0, dag_h, "Nil", allocList);
   1126 	rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc,
   1127 	    NULL, 0, 1, 0, 0, dag_h, "Trm", allocList);
   1128 
   1129 	/* initialize the wnd nodes */
   1130 	if (nWndNodes > 0) {
   1131 		pda = asmap->physInfo;
   1132 		for (i = 0; i < nWndNodes; i++) {
   1133 			rf_InitNode(&wndNode[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
   1134 			    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wpd", allocList);
   1135 			RF_ASSERT(pda != NULL);
   1136 			wndNode[i].params[0].p = pda;
   1137 			wndNode[i].params[1].p = pda->bufPtr;
   1138 			wndNode[i].params[2].v = parityStripeID;
   1139 			wndNode[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
   1140 			pda = pda->next;
   1141 		}
   1142 		RF_ASSERT(pda == NULL);
   1143 	}
   1144 	/* initialize the mirror nodes */
   1145 	if (nWmirNodes > 0) {
   1146 		pda = asmap->physInfo;
   1147 		pdaP = asmap->parityInfo;
   1148 		for (i = 0; i < nWmirNodes; i++) {
   1149 			rf_InitNode(&wmirNode[i], rf_wait, RF_FALSE, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
   1150 			    rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, "Wsd", allocList);
   1151 			RF_ASSERT(pda != NULL);
   1152 			wmirNode[i].params[0].p = pdaP;
   1153 			wmirNode[i].params[1].p = pda->bufPtr;
   1154 			wmirNode[i].params[2].v = parityStripeID;
   1155 			wmirNode[i].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
   1156 			pda = pda->next;
   1157 			pdaP = pdaP->next;
   1158 		}
   1159 		RF_ASSERT(pda == NULL);
   1160 		RF_ASSERT(pdaP == NULL);
   1161 	}
   1162 	/* link the header node to the commit node */
   1163 	RF_ASSERT(dag_h->numSuccedents == 1);
   1164 	RF_ASSERT(commitNode->numAntecedents == 0);
   1165 	dag_h->succedents[0] = commitNode;
   1166 
   1167 	/* link the commit node to the write nodes */
   1168 	RF_ASSERT(commitNode->numSuccedents == (nWndNodes + nWmirNodes));
   1169 	for (i = 0; i < nWndNodes; i++) {
   1170 		RF_ASSERT(wndNode[i].numAntecedents == 1);
   1171 		commitNode->succedents[i] = &wndNode[i];
   1172 		wndNode[i].antecedents[0] = commitNode;
   1173 		wndNode[i].antType[0] = rf_control;
   1174 	}
   1175 	for (i = 0; i < nWmirNodes; i++) {
   1176 		RF_ASSERT(wmirNode[i].numAntecedents == 1);
   1177 		commitNode->succedents[i + nWndNodes] = &wmirNode[i];
   1178 		wmirNode[i].antecedents[0] = commitNode;
   1179 		wmirNode[i].antType[0] = rf_control;
   1180 	}
   1181 
   1182 	/* link the write nodes to the unblock node */
   1183 	RF_ASSERT(unblockNode->numAntecedents == (nWndNodes + nWmirNodes));
   1184 	for (i = 0; i < nWndNodes; i++) {
   1185 		RF_ASSERT(wndNode[i].numSuccedents == 1);
   1186 		wndNode[i].succedents[0] = unblockNode;
   1187 		unblockNode->antecedents[i] = &wndNode[i];
   1188 		unblockNode->antType[i] = rf_control;
   1189 	}
   1190 	for (i = 0; i < nWmirNodes; i++) {
   1191 		RF_ASSERT(wmirNode[i].numSuccedents == 1);
   1192 		wmirNode[i].succedents[0] = unblockNode;
   1193 		unblockNode->antecedents[i + nWndNodes] = &wmirNode[i];
   1194 		unblockNode->antType[i + nWndNodes] = rf_control;
   1195 	}
   1196 
   1197 	/* link the unblock node to the term node */
   1198 	RF_ASSERT(unblockNode->numSuccedents == 1);
   1199 	RF_ASSERT(termNode->numAntecedents == 1);
   1200 	RF_ASSERT(termNode->numSuccedents == 0);
   1201 	unblockNode->succedents[0] = termNode;
   1202 	termNode->antecedents[0] = unblockNode;
   1203 	termNode->antType[0] = rf_control;
   1204 }
   1205