Home | History | Annotate | Line # | Download | only in raidframe
rf_dagutils.c revision 1.44
      1 /*	$NetBSD: rf_dagutils.c,v 1.44 2004/04/09 23:10:16 oster Exp $	*/
      2 /*
      3  * Copyright (c) 1995 Carnegie-Mellon University.
      4  * All rights reserved.
      5  *
      6  * Authors: Mark Holland, William V. Courtright II, Jim Zelenka
      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  *
     31  * rf_dagutils.c -- utility routines for manipulating dags
     32  *
     33  *****************************************************************************/
     34 
     35 #include <sys/cdefs.h>
     36 __KERNEL_RCSID(0, "$NetBSD: rf_dagutils.c,v 1.44 2004/04/09 23:10:16 oster Exp $");
     37 
     38 #include <dev/raidframe/raidframevar.h>
     39 
     40 #include "rf_archs.h"
     41 #include "rf_threadstuff.h"
     42 #include "rf_raid.h"
     43 #include "rf_dag.h"
     44 #include "rf_dagutils.h"
     45 #include "rf_dagfuncs.h"
     46 #include "rf_general.h"
     47 #include "rf_map.h"
     48 #include "rf_shutdown.h"
     49 
     50 #define SNUM_DIFF(_a_,_b_) (((_a_)>(_b_))?((_a_)-(_b_)):((_b_)-(_a_)))
     51 
     52 const RF_RedFuncs_t rf_xorFuncs = {
     53 	rf_RegularXorFunc, "Reg Xr",
     54 	rf_SimpleXorFunc, "Simple Xr"};
     55 
     56 const RF_RedFuncs_t rf_xorRecoveryFuncs = {
     57 	rf_RecoveryXorFunc, "Recovery Xr",
     58 	rf_RecoveryXorFunc, "Recovery Xr"};
     59 
     60 #if RF_DEBUG_VALIDATE_DAG
     61 static void rf_RecurPrintDAG(RF_DagNode_t *, int, int);
     62 static void rf_PrintDAG(RF_DagHeader_t *);
     63 static int rf_ValidateBranch(RF_DagNode_t *, int *, int *,
     64 			     RF_DagNode_t **, int);
     65 static void rf_ValidateBranchVisitedBits(RF_DagNode_t *, int, int);
     66 static void rf_ValidateVisitedBits(RF_DagHeader_t *);
     67 #endif /* RF_DEBUG_VALIDATE_DAG */
     68 
     69 /* The maximum number of nodes in a DAG is bounded by
     70 
     71 (2 * raidPtr->Layout->numDataCol) + (1 * layoutPtr->numParityCol) +
     72 	(1 * 2 * layoutPtr->numParityCol) + 3
     73 
     74 which is:  2*RF_MAXCOL+1*2+1*2*2+3
     75 
     76 For RF_MAXCOL of 40, this works out to 89.  We use this value to provide an estimate
     77 on the maximum size needed for RF_DAGPCACHE_SIZE.  For RF_MAXCOL of 40, this structure
     78 would be 534 bytes.  Too much to have on-hand in a RF_DagNode_t, but should be ok to
     79 have a few kicking around.
     80 */
     81 #define RF_DAGPCACHE_SIZE ((2*RF_MAXCOL+1*2+1*2*2+3) *(RF_MAX(sizeof(RF_DagParam_t), sizeof(RF_DagNode_t *))))
     82 
     83 
     84 /******************************************************************************
     85  *
     86  * InitNode - initialize a dag node
     87  *
     88  * the size of the propList array is always the same as that of the
     89  * successors array.
     90  *
     91  *****************************************************************************/
     92 void
     93 rf_InitNode(RF_DagNode_t *node, RF_NodeStatus_t initstatus, int commit,
     94     int (*doFunc) (RF_DagNode_t *node),
     95     int (*undoFunc) (RF_DagNode_t *node),
     96     int (*wakeFunc) (RF_DagNode_t *node, int status),
     97     int nSucc, int nAnte, int nParam, int nResult,
     98     RF_DagHeader_t *hdr, char *name, RF_AllocListElem_t *alist)
     99 {
    100 	void  **ptrs;
    101 	int     nptrs;
    102 
    103 	if (nAnte > RF_MAX_ANTECEDENTS)
    104 		RF_PANIC();
    105 	node->status = initstatus;
    106 	node->commitNode = commit;
    107 	node->doFunc = doFunc;
    108 	node->undoFunc = undoFunc;
    109 	node->wakeFunc = wakeFunc;
    110 	node->numParams = nParam;
    111 	node->numResults = nResult;
    112 	node->numAntecedents = nAnte;
    113 	node->numAntDone = 0;
    114 	node->next = NULL;
    115 	/* node->list_next = NULL */  /* Don't touch this here!
    116 	                                 It may already be
    117 					 in use by the caller! */
    118 	node->numSuccedents = nSucc;
    119 	node->name = name;
    120 	node->dagHdr = hdr;
    121 	node->big_dag_ptrs = NULL;
    122 	node->big_dag_params = NULL;
    123 	node->visited = 0;
    124 
    125 	/* allocate all the pointers with one call to malloc */
    126 	nptrs = nSucc + nAnte + nResult + nSucc;
    127 
    128 	if (nptrs <= RF_DAG_PTRCACHESIZE) {
    129 		/*
    130 	         * The dag_ptrs field of the node is basically some scribble
    131 	         * space to be used here. We could get rid of it, and always
    132 	         * allocate the range of pointers, but that's expensive. So,
    133 	         * we pick a "common case" size for the pointer cache. Hopefully,
    134 	         * we'll find that:
    135 	         * (1) Generally, nptrs doesn't exceed RF_DAG_PTRCACHESIZE by
    136 	         *     only a little bit (least efficient case)
    137 	         * (2) Generally, ntprs isn't a lot less than RF_DAG_PTRCACHESIZE
    138 	         *     (wasted memory)
    139 	         */
    140 		ptrs = (void **) node->dag_ptrs;
    141 	} else if (nptrs <= (RF_DAGPCACHE_SIZE / sizeof(RF_DagNode_t *))) {
    142 		node->big_dag_ptrs = rf_AllocDAGPCache();
    143 		ptrs = (void **) node->big_dag_ptrs;
    144 	} else {
    145 		RF_MallocAndAdd(ptrs, nptrs * sizeof(void *),
    146 				(void **), alist);
    147 	}
    148 	node->succedents = (nSucc) ? (RF_DagNode_t **) ptrs : NULL;
    149 	node->antecedents = (nAnte) ? (RF_DagNode_t **) (ptrs + nSucc) : NULL;
    150 	node->results = (nResult) ? (void **) (ptrs + nSucc + nAnte) : NULL;
    151 	node->propList = (nSucc) ? (RF_PropHeader_t **) (ptrs + nSucc + nAnte + nResult) : NULL;
    152 
    153 	if (nParam) {
    154 		if (nParam <= RF_DAG_PARAMCACHESIZE) {
    155 			node->params = (RF_DagParam_t *) node->dag_params;
    156 		} else if (nParam <= (RF_DAGPCACHE_SIZE / sizeof(RF_DagParam_t))) {
    157 			node->big_dag_params = rf_AllocDAGPCache();
    158 			node->params = node->big_dag_params;
    159 		} else {
    160 			RF_MallocAndAdd(node->params,
    161 					nParam * sizeof(RF_DagParam_t),
    162 					(RF_DagParam_t *), alist);
    163 		}
    164 	} else {
    165 		node->params = NULL;
    166 	}
    167 }
    168 
    169 
    170 
    171 /******************************************************************************
    172  *
    173  * allocation and deallocation routines
    174  *
    175  *****************************************************************************/
    176 
    177 void
    178 rf_FreeDAG(RF_DagHeader_t *dag_h)
    179 {
    180 	RF_AccessStripeMapHeader_t *asmap, *t_asmap;
    181 	RF_PhysDiskAddr_t *pda;
    182 	RF_DagNode_t *tmpnode;
    183 	RF_DagHeader_t *nextDag;
    184 
    185 	while (dag_h) {
    186 		nextDag = dag_h->next;
    187 		rf_FreeAllocList(dag_h->allocList);
    188 		for (asmap = dag_h->asmList; asmap;) {
    189 			t_asmap = asmap;
    190 			asmap = asmap->next;
    191 			rf_FreeAccessStripeMap(t_asmap);
    192 		}
    193 		while (dag_h->pda_cleanup_list) {
    194 			pda = dag_h->pda_cleanup_list;
    195 			dag_h->pda_cleanup_list = dag_h->pda_cleanup_list->next;
    196 			rf_FreePhysDiskAddr(pda);
    197 		}
    198 		while (dag_h->nodes) {
    199 			tmpnode = dag_h->nodes;
    200 			dag_h->nodes = dag_h->nodes->list_next;
    201 			rf_FreeDAGNode(tmpnode);
    202 		}
    203 		rf_FreeDAGHeader(dag_h);
    204 		dag_h = nextDag;
    205 	}
    206 }
    207 
    208 #define RF_MAX_FREE_DAGH 128
    209 #define RF_MIN_FREE_DAGH  32
    210 
    211 #define RF_MAX_FREE_DAGNODE 512 /* XXX Tune this... */
    212 #define RF_MIN_FREE_DAGNODE 128 /* XXX Tune this... */
    213 
    214 #define RF_MAX_FREE_DAGLIST 128
    215 #define RF_MIN_FREE_DAGLIST  32
    216 
    217 #define RF_MAX_FREE_DAGPCACHE 128
    218 #define RF_MIN_FREE_DAGPCACHE   8
    219 
    220 #define RF_MAX_FREE_FUNCLIST 128
    221 #define RF_MIN_FREE_FUNCLIST  32
    222 
    223 #define RF_MAX_FREE_BUFFERS 128
    224 #define RF_MIN_FREE_BUFFERS  32
    225 
    226 static void rf_ShutdownDAGs(void *);
    227 static void
    228 rf_ShutdownDAGs(void *ignored)
    229 {
    230 	pool_destroy(&rf_pools.dagh);
    231 	pool_destroy(&rf_pools.dagnode);
    232 	pool_destroy(&rf_pools.daglist);
    233 	pool_destroy(&rf_pools.dagpcache);
    234 	pool_destroy(&rf_pools.funclist);
    235 }
    236 
    237 int
    238 rf_ConfigureDAGs(RF_ShutdownList_t **listp)
    239 {
    240 
    241 	rf_pool_init(&rf_pools.dagnode, sizeof(RF_DagNode_t),
    242 		     "rf_dagnode_pl", RF_MIN_FREE_DAGNODE, RF_MAX_FREE_DAGNODE);
    243 	rf_pool_init(&rf_pools.dagh, sizeof(RF_DagHeader_t),
    244 		     "rf_dagh_pl", RF_MIN_FREE_DAGH, RF_MAX_FREE_DAGH);
    245 	rf_pool_init(&rf_pools.daglist, sizeof(RF_DagList_t),
    246 		     "rf_daglist_pl", RF_MIN_FREE_DAGLIST, RF_MAX_FREE_DAGLIST);
    247 	rf_pool_init(&rf_pools.dagpcache, RF_DAGPCACHE_SIZE,
    248 		     "rf_dagpcache_pl", RF_MIN_FREE_DAGPCACHE, RF_MAX_FREE_DAGPCACHE);
    249 	rf_pool_init(&rf_pools.funclist, sizeof(RF_FuncList_t),
    250 		     "rf_funclist_pl", RF_MIN_FREE_FUNCLIST, RF_MAX_FREE_FUNCLIST);
    251 	rf_ShutdownCreate(listp, rf_ShutdownDAGs, NULL);
    252 
    253 	return (0);
    254 }
    255 
    256 RF_DagHeader_t *
    257 rf_AllocDAGHeader()
    258 {
    259 	RF_DagHeader_t *dh;
    260 
    261 	dh = pool_get(&rf_pools.dagh, PR_WAITOK);
    262 	memset((char *) dh, 0, sizeof(RF_DagHeader_t));
    263 	return (dh);
    264 }
    265 
    266 void
    267 rf_FreeDAGHeader(RF_DagHeader_t * dh)
    268 {
    269 	pool_put(&rf_pools.dagh, dh);
    270 }
    271 
    272 RF_DagNode_t *
    273 rf_AllocDAGNode()
    274 {
    275 	RF_DagNode_t *node;
    276 
    277 	node = pool_get(&rf_pools.dagnode, PR_WAITOK);
    278 	memset(node, 0, sizeof(RF_DagNode_t));
    279 	return (node);
    280 }
    281 
    282 void
    283 rf_FreeDAGNode(RF_DagNode_t *node)
    284 {
    285 	if (node->big_dag_ptrs) {
    286 		rf_FreeDAGPCache(node->big_dag_ptrs);
    287 	}
    288 	if (node->big_dag_params) {
    289 		rf_FreeDAGPCache(node->big_dag_params);
    290 	}
    291 	pool_put(&rf_pools.dagnode, node);
    292 }
    293 
    294 RF_DagList_t *
    295 rf_AllocDAGList()
    296 {
    297 	RF_DagList_t *dagList;
    298 
    299 	dagList = pool_get(&rf_pools.daglist, PR_WAITOK);
    300 	memset(dagList, 0, sizeof(RF_DagList_t));
    301 
    302 	return (dagList);
    303 }
    304 
    305 void
    306 rf_FreeDAGList(RF_DagList_t *dagList)
    307 {
    308 	pool_put(&rf_pools.daglist, dagList);
    309 }
    310 
    311 void *
    312 rf_AllocDAGPCache()
    313 {
    314 	void *p;
    315 	p = pool_get(&rf_pools.dagpcache, PR_WAITOK);
    316 	memset(p, 0, RF_DAGPCACHE_SIZE);
    317 
    318 	return (p);
    319 }
    320 
    321 void
    322 rf_FreeDAGPCache(void *p)
    323 {
    324 	pool_put(&rf_pools.dagpcache, p);
    325 }
    326 
    327 RF_FuncList_t *
    328 rf_AllocFuncList()
    329 {
    330 	RF_FuncList_t *funcList;
    331 
    332 	funcList = pool_get(&rf_pools.funclist, PR_WAITOK);
    333 	memset(funcList, 0, sizeof(RF_FuncList_t));
    334 
    335 	return (funcList);
    336 }
    337 
    338 void
    339 rf_FreeFuncList(RF_FuncList_t *funcList)
    340 {
    341 	pool_put(&rf_pools.funclist, funcList);
    342 }
    343 
    344 /* allocates a stripe buffer -- a buffer large enough to hold all the data
    345    in an entire stripe.
    346 */
    347 
    348 void *
    349 rf_AllocStripeBuffer(RF_Raid_t *raidPtr, RF_DagHeader_t *dag_h, int size)
    350 {
    351 	RF_VoidPointerListElem_t *vple;
    352 	void *p;
    353 
    354 	RF_ASSERT((size <= (raidPtr->numCol * (raidPtr->Layout.sectorsPerStripeUnit <<
    355 					       raidPtr->logBytesPerSector))));
    356 
    357 	p =  malloc( raidPtr->numCol * (raidPtr->Layout.sectorsPerStripeUnit <<
    358 					raidPtr->logBytesPerSector),
    359 		     M_RAIDFRAME, M_NOWAIT);
    360 	if (!p) {
    361 		RF_LOCK_MUTEX(raidPtr->mutex);
    362 		if (raidPtr->stripebuf_count > 0) {
    363 			vple = raidPtr->stripebuf;
    364 			raidPtr->stripebuf = vple->next;
    365 			p = vple->p;
    366 			rf_FreeVPListElem(vple);
    367 			raidPtr->stripebuf_count--;
    368 		} else {
    369 #ifdef DIAGNOSTIC
    370 			printf("raid%d: Help!  Out of emergency full-stripe buffers!\n", raidPtr->raidid);
    371 #endif
    372 		}
    373 		RF_UNLOCK_MUTEX(raidPtr->mutex);
    374 		if (!p) {
    375 			/* We didn't get a buffer... not much we can do other than wait,
    376 			   and hope that someone frees up memory for us.. */
    377 			p = malloc( raidPtr->numCol * (raidPtr->Layout.sectorsPerStripeUnit <<
    378 						       raidPtr->logBytesPerSector), M_RAIDFRAME, M_WAITOK);
    379 		}
    380 	}
    381 	memset(p, 0, raidPtr->numCol * (raidPtr->Layout.sectorsPerStripeUnit << raidPtr->logBytesPerSector));
    382 
    383 	vple = rf_AllocVPListElem();
    384 	vple->p = p;
    385         vple->next = dag_h->desc->stripebufs;
    386         dag_h->desc->stripebufs = vple;
    387 
    388 	return (p);
    389 }
    390 
    391 
    392 void
    393 rf_FreeStripeBuffer(RF_Raid_t *raidPtr, RF_VoidPointerListElem_t *vple)
    394 {
    395 	RF_LOCK_MUTEX(raidPtr->mutex);
    396 	if (raidPtr->stripebuf_count < raidPtr->numEmergencyStripeBuffers) {
    397 		/* just tack it in */
    398 		vple->next = raidPtr->stripebuf;
    399 		raidPtr->stripebuf = vple;
    400 		raidPtr->stripebuf_count++;
    401 	} else {
    402 		free(vple->p, M_RAIDFRAME);
    403 		rf_FreeVPListElem(vple);
    404 	}
    405 	RF_UNLOCK_MUTEX(raidPtr->mutex);
    406 }
    407 
    408 /* allocates a buffer big enough to hold the data described by the
    409 caller (ie. the data of the associated PDA).  Glue this buffer
    410 into our dag_h cleanup structure. */
    411 
    412 void *
    413 rf_AllocBuffer(RF_Raid_t *raidPtr, RF_DagHeader_t *dag_h, int size)
    414 {
    415 	RF_VoidPointerListElem_t *vple;
    416 	void *p;
    417 
    418 	p = rf_AllocIOBuffer(raidPtr, size);
    419 	vple = rf_AllocVPListElem();
    420 	vple->p = p;
    421 	vple->next = dag_h->desc->iobufs;
    422 	dag_h->desc->iobufs = vple;
    423 
    424 	return (p);
    425 }
    426 
    427 void *
    428 rf_AllocIOBuffer(RF_Raid_t *raidPtr, int size)
    429 {
    430 	RF_VoidPointerListElem_t *vple;
    431 	void *p;
    432 
    433 	RF_ASSERT((size <= (raidPtr->Layout.sectorsPerStripeUnit <<
    434 			   raidPtr->logBytesPerSector)));
    435 
    436 	p =  malloc( raidPtr->Layout.sectorsPerStripeUnit <<
    437 				 raidPtr->logBytesPerSector,
    438 				 M_RAIDFRAME, M_NOWAIT);
    439 	if (!p) {
    440 		RF_LOCK_MUTEX(raidPtr->mutex);
    441 		if (raidPtr->iobuf_count > 0) {
    442 			vple = raidPtr->iobuf;
    443 			raidPtr->iobuf = vple->next;
    444 			p = vple->p;
    445 			rf_FreeVPListElem(vple);
    446 			raidPtr->iobuf_count--;
    447 		} else {
    448 #ifdef DIAGNOSTIC
    449 			printf("raid%d: Help!  Out of emergency buffers!\n", raidPtr->raidid);
    450 #endif
    451 		}
    452 		RF_UNLOCK_MUTEX(raidPtr->mutex);
    453 		if (!p) {
    454 			/* We didn't get a buffer... not much we can do other than wait,
    455 			   and hope that someone frees up memory for us.. */
    456 			p = malloc( raidPtr->Layout.sectorsPerStripeUnit <<
    457 				    raidPtr->logBytesPerSector,
    458 				    M_RAIDFRAME, M_WAITOK);
    459 		}
    460 	}
    461 	memset(p, 0, raidPtr->Layout.sectorsPerStripeUnit << raidPtr->logBytesPerSector);
    462 	return (p);
    463 }
    464 
    465 void
    466 rf_FreeIOBuffer(RF_Raid_t *raidPtr, RF_VoidPointerListElem_t *vple)
    467 {
    468 	RF_LOCK_MUTEX(raidPtr->mutex);
    469 	if (raidPtr->iobuf_count < raidPtr->numEmergencyBuffers) {
    470 		/* just tack it in */
    471 		vple->next = raidPtr->iobuf;
    472 		raidPtr->iobuf = vple;
    473 		raidPtr->iobuf_count++;
    474 	} else {
    475 		free(vple->p, M_RAIDFRAME);
    476 		rf_FreeVPListElem(vple);
    477 	}
    478 	RF_UNLOCK_MUTEX(raidPtr->mutex);
    479 }
    480 
    481 
    482 
    483 #if RF_DEBUG_VALIDATE_DAG
    484 /******************************************************************************
    485  *
    486  * debug routines
    487  *
    488  *****************************************************************************/
    489 
    490 char   *
    491 rf_NodeStatusString(RF_DagNode_t *node)
    492 {
    493 	switch (node->status) {
    494 	case rf_wait:
    495 		return ("wait");
    496 	case rf_fired:
    497 		return ("fired");
    498 	case rf_good:
    499 		return ("good");
    500 	case rf_bad:
    501 		return ("bad");
    502 	default:
    503 		return ("?");
    504 	}
    505 }
    506 
    507 void
    508 rf_PrintNodeInfoString(RF_DagNode_t *node)
    509 {
    510 	RF_PhysDiskAddr_t *pda;
    511 	int     (*df) (RF_DagNode_t *) = node->doFunc;
    512 	int     i, lk, unlk;
    513 	void   *bufPtr;
    514 
    515 	if ((df == rf_DiskReadFunc) || (df == rf_DiskWriteFunc)
    516 	    || (df == rf_DiskReadMirrorIdleFunc)
    517 	    || (df == rf_DiskReadMirrorPartitionFunc)) {
    518 		pda = (RF_PhysDiskAddr_t *) node->params[0].p;
    519 		bufPtr = (void *) node->params[1].p;
    520 		lk = 0;
    521 		unlk = 0;
    522 		RF_ASSERT(!(lk && unlk));
    523 		printf("c %d offs %ld nsect %d buf 0x%lx %s\n", pda->col,
    524 		    (long) pda->startSector, (int) pda->numSector, (long) bufPtr,
    525 		    (lk) ? "LOCK" : ((unlk) ? "UNLK" : " "));
    526 		return;
    527 	}
    528 	if (df == rf_DiskUnlockFunc) {
    529 		pda = (RF_PhysDiskAddr_t *) node->params[0].p;
    530 		lk = 0;
    531 		unlk = 0;
    532 		RF_ASSERT(!(lk && unlk));
    533 		printf("c %d %s\n", pda->col,
    534 		    (lk) ? "LOCK" : ((unlk) ? "UNLK" : "nop"));
    535 		return;
    536 	}
    537 	if ((df == rf_SimpleXorFunc) || (df == rf_RegularXorFunc)
    538 	    || (df == rf_RecoveryXorFunc)) {
    539 		printf("result buf 0x%lx\n", (long) node->results[0]);
    540 		for (i = 0; i < node->numParams - 1; i += 2) {
    541 			pda = (RF_PhysDiskAddr_t *) node->params[i].p;
    542 			bufPtr = (RF_PhysDiskAddr_t *) node->params[i + 1].p;
    543 			printf("    buf 0x%lx c%d offs %ld nsect %d\n",
    544 			    (long) bufPtr, pda->col,
    545 			    (long) pda->startSector, (int) pda->numSector);
    546 		}
    547 		return;
    548 	}
    549 #if RF_INCLUDE_PARITYLOGGING > 0
    550 	if (df == rf_ParityLogOverwriteFunc || df == rf_ParityLogUpdateFunc) {
    551 		for (i = 0; i < node->numParams - 1; i += 2) {
    552 			pda = (RF_PhysDiskAddr_t *) node->params[i].p;
    553 			bufPtr = (RF_PhysDiskAddr_t *) node->params[i + 1].p;
    554 			printf(" c%d offs %ld nsect %d buf 0x%lx\n",
    555 			    pda->col, (long) pda->startSector,
    556 			    (int) pda->numSector, (long) bufPtr);
    557 		}
    558 		return;
    559 	}
    560 #endif				/* RF_INCLUDE_PARITYLOGGING > 0 */
    561 
    562 	if ((df == rf_TerminateFunc) || (df == rf_NullNodeFunc)) {
    563 		printf("\n");
    564 		return;
    565 	}
    566 	printf("?\n");
    567 }
    568 #ifdef DEBUG
    569 static void
    570 rf_RecurPrintDAG(RF_DagNode_t *node, int depth, int unvisited)
    571 {
    572 	char   *anttype;
    573 	int     i;
    574 
    575 	node->visited = (unvisited) ? 0 : 1;
    576 	printf("(%d) %d C%d %s: %s,s%d %d/%d,a%d/%d,p%d,r%d S{", depth,
    577 	    node->nodeNum, node->commitNode, node->name, rf_NodeStatusString(node),
    578 	    node->numSuccedents, node->numSuccFired, node->numSuccDone,
    579 	    node->numAntecedents, node->numAntDone, node->numParams, node->numResults);
    580 	for (i = 0; i < node->numSuccedents; i++) {
    581 		printf("%d%s", node->succedents[i]->nodeNum,
    582 		    ((i == node->numSuccedents - 1) ? "\0" : " "));
    583 	}
    584 	printf("} A{");
    585 	for (i = 0; i < node->numAntecedents; i++) {
    586 		switch (node->antType[i]) {
    587 		case rf_trueData:
    588 			anttype = "T";
    589 			break;
    590 		case rf_antiData:
    591 			anttype = "A";
    592 			break;
    593 		case rf_outputData:
    594 			anttype = "O";
    595 			break;
    596 		case rf_control:
    597 			anttype = "C";
    598 			break;
    599 		default:
    600 			anttype = "?";
    601 			break;
    602 		}
    603 		printf("%d(%s)%s", node->antecedents[i]->nodeNum, anttype, (i == node->numAntecedents - 1) ? "\0" : " ");
    604 	}
    605 	printf("}; ");
    606 	rf_PrintNodeInfoString(node);
    607 	for (i = 0; i < node->numSuccedents; i++) {
    608 		if (node->succedents[i]->visited == unvisited)
    609 			rf_RecurPrintDAG(node->succedents[i], depth + 1, unvisited);
    610 	}
    611 }
    612 
    613 static void
    614 rf_PrintDAG(RF_DagHeader_t *dag_h)
    615 {
    616 	int     unvisited, i;
    617 	char   *status;
    618 
    619 	/* set dag status */
    620 	switch (dag_h->status) {
    621 	case rf_enable:
    622 		status = "enable";
    623 		break;
    624 	case rf_rollForward:
    625 		status = "rollForward";
    626 		break;
    627 	case rf_rollBackward:
    628 		status = "rollBackward";
    629 		break;
    630 	default:
    631 		status = "illegal!";
    632 		break;
    633 	}
    634 	/* find out if visited bits are currently set or clear */
    635 	unvisited = dag_h->succedents[0]->visited;
    636 
    637 	printf("DAG type:  %s\n", dag_h->creator);
    638 	printf("format is (depth) num commit type: status,nSucc nSuccFired/nSuccDone,nAnte/nAnteDone,nParam,nResult S{x} A{x(type)};  info\n");
    639 	printf("(0) %d Hdr: %s, s%d, (commit %d/%d) S{", dag_h->nodeNum,
    640 	    status, dag_h->numSuccedents, dag_h->numCommitNodes, dag_h->numCommits);
    641 	for (i = 0; i < dag_h->numSuccedents; i++) {
    642 		printf("%d%s", dag_h->succedents[i]->nodeNum,
    643 		    ((i == dag_h->numSuccedents - 1) ? "\0" : " "));
    644 	}
    645 	printf("};\n");
    646 	for (i = 0; i < dag_h->numSuccedents; i++) {
    647 		if (dag_h->succedents[i]->visited == unvisited)
    648 			rf_RecurPrintDAG(dag_h->succedents[i], 1, unvisited);
    649 	}
    650 }
    651 #endif
    652 /* assigns node numbers */
    653 int
    654 rf_AssignNodeNums(RF_DagHeader_t * dag_h)
    655 {
    656 	int     unvisited, i, nnum;
    657 	RF_DagNode_t *node;
    658 
    659 	nnum = 0;
    660 	unvisited = dag_h->succedents[0]->visited;
    661 
    662 	dag_h->nodeNum = nnum++;
    663 	for (i = 0; i < dag_h->numSuccedents; i++) {
    664 		node = dag_h->succedents[i];
    665 		if (node->visited == unvisited) {
    666 			nnum = rf_RecurAssignNodeNums(dag_h->succedents[i], nnum, unvisited);
    667 		}
    668 	}
    669 	return (nnum);
    670 }
    671 
    672 int
    673 rf_RecurAssignNodeNums(RF_DagNode_t *node, int num, int unvisited)
    674 {
    675 	int     i;
    676 
    677 	node->visited = (unvisited) ? 0 : 1;
    678 
    679 	node->nodeNum = num++;
    680 	for (i = 0; i < node->numSuccedents; i++) {
    681 		if (node->succedents[i]->visited == unvisited) {
    682 			num = rf_RecurAssignNodeNums(node->succedents[i], num, unvisited);
    683 		}
    684 	}
    685 	return (num);
    686 }
    687 /* set the header pointers in each node to "newptr" */
    688 void
    689 rf_ResetDAGHeaderPointers(RF_DagHeader_t *dag_h, RF_DagHeader_t *newptr)
    690 {
    691 	int     i;
    692 	for (i = 0; i < dag_h->numSuccedents; i++)
    693 		if (dag_h->succedents[i]->dagHdr != newptr)
    694 			rf_RecurResetDAGHeaderPointers(dag_h->succedents[i], newptr);
    695 }
    696 
    697 void
    698 rf_RecurResetDAGHeaderPointers(RF_DagNode_t *node, RF_DagHeader_t *newptr)
    699 {
    700 	int     i;
    701 	node->dagHdr = newptr;
    702 	for (i = 0; i < node->numSuccedents; i++)
    703 		if (node->succedents[i]->dagHdr != newptr)
    704 			rf_RecurResetDAGHeaderPointers(node->succedents[i], newptr);
    705 }
    706 
    707 
    708 void
    709 rf_PrintDAGList(RF_DagHeader_t * dag_h)
    710 {
    711 	int     i = 0;
    712 
    713 	for (; dag_h; dag_h = dag_h->next) {
    714 		rf_AssignNodeNums(dag_h);
    715 		printf("\n\nDAG %d IN LIST:\n", i++);
    716 		rf_PrintDAG(dag_h);
    717 	}
    718 }
    719 
    720 static int
    721 rf_ValidateBranch(RF_DagNode_t *node, int *scount, int *acount,
    722 		  RF_DagNode_t **nodes, int unvisited)
    723 {
    724 	int     i, retcode = 0;
    725 
    726 	/* construct an array of node pointers indexed by node num */
    727 	node->visited = (unvisited) ? 0 : 1;
    728 	nodes[node->nodeNum] = node;
    729 
    730 	if (node->next != NULL) {
    731 		printf("INVALID DAG: next pointer in node is not NULL\n");
    732 		retcode = 1;
    733 	}
    734 	if (node->status != rf_wait) {
    735 		printf("INVALID DAG: Node status is not wait\n");
    736 		retcode = 1;
    737 	}
    738 	if (node->numAntDone != 0) {
    739 		printf("INVALID DAG: numAntDone is not zero\n");
    740 		retcode = 1;
    741 	}
    742 	if (node->doFunc == rf_TerminateFunc) {
    743 		if (node->numSuccedents != 0) {
    744 			printf("INVALID DAG: Terminator node has succedents\n");
    745 			retcode = 1;
    746 		}
    747 	} else {
    748 		if (node->numSuccedents == 0) {
    749 			printf("INVALID DAG: Non-terminator node has no succedents\n");
    750 			retcode = 1;
    751 		}
    752 	}
    753 	for (i = 0; i < node->numSuccedents; i++) {
    754 		if (!node->succedents[i]) {
    755 			printf("INVALID DAG: succedent %d of node %s is NULL\n", i, node->name);
    756 			retcode = 1;
    757 		}
    758 		scount[node->succedents[i]->nodeNum]++;
    759 	}
    760 	for (i = 0; i < node->numAntecedents; i++) {
    761 		if (!node->antecedents[i]) {
    762 			printf("INVALID DAG: antecedent %d of node %s is NULL\n", i, node->name);
    763 			retcode = 1;
    764 		}
    765 		acount[node->antecedents[i]->nodeNum]++;
    766 	}
    767 	for (i = 0; i < node->numSuccedents; i++) {
    768 		if (node->succedents[i]->visited == unvisited) {
    769 			if (rf_ValidateBranch(node->succedents[i], scount,
    770 				acount, nodes, unvisited)) {
    771 				retcode = 1;
    772 			}
    773 		}
    774 	}
    775 	return (retcode);
    776 }
    777 
    778 static void
    779 rf_ValidateBranchVisitedBits(RF_DagNode_t *node, int unvisited, int rl)
    780 {
    781 	int     i;
    782 
    783 	RF_ASSERT(node->visited == unvisited);
    784 	for (i = 0; i < node->numSuccedents; i++) {
    785 		if (node->succedents[i] == NULL) {
    786 			printf("node=%lx node->succedents[%d] is NULL\n", (long) node, i);
    787 			RF_ASSERT(0);
    788 		}
    789 		rf_ValidateBranchVisitedBits(node->succedents[i], unvisited, rl + 1);
    790 	}
    791 }
    792 /* NOTE:  never call this on a big dag, because it is exponential
    793  * in execution time
    794  */
    795 static void
    796 rf_ValidateVisitedBits(RF_DagHeader_t *dag)
    797 {
    798 	int     i, unvisited;
    799 
    800 	unvisited = dag->succedents[0]->visited;
    801 
    802 	for (i = 0; i < dag->numSuccedents; i++) {
    803 		if (dag->succedents[i] == NULL) {
    804 			printf("dag=%lx dag->succedents[%d] is NULL\n", (long) dag, i);
    805 			RF_ASSERT(0);
    806 		}
    807 		rf_ValidateBranchVisitedBits(dag->succedents[i], unvisited, 0);
    808 	}
    809 }
    810 /* validate a DAG.  _at entry_ verify that:
    811  *   -- numNodesCompleted is zero
    812  *   -- node queue is null
    813  *   -- dag status is rf_enable
    814  *   -- next pointer is null on every node
    815  *   -- all nodes have status wait
    816  *   -- numAntDone is zero in all nodes
    817  *   -- terminator node has zero successors
    818  *   -- no other node besides terminator has zero successors
    819  *   -- no successor or antecedent pointer in a node is NULL
    820  *   -- number of times that each node appears as a successor of another node
    821  *      is equal to the antecedent count on that node
    822  *   -- number of times that each node appears as an antecedent of another node
    823  *      is equal to the succedent count on that node
    824  *   -- what else?
    825  */
    826 int
    827 rf_ValidateDAG(RF_DagHeader_t *dag_h)
    828 {
    829 	int     i, nodecount;
    830 	int    *scount, *acount;/* per-node successor and antecedent counts */
    831 	RF_DagNode_t **nodes;	/* array of ptrs to nodes in dag */
    832 	int     retcode = 0;
    833 	int     unvisited;
    834 	int     commitNodeCount = 0;
    835 
    836 	if (rf_validateVisitedDebug)
    837 		rf_ValidateVisitedBits(dag_h);
    838 
    839 	if (dag_h->numNodesCompleted != 0) {
    840 		printf("INVALID DAG: num nodes completed is %d, should be 0\n", dag_h->numNodesCompleted);
    841 		retcode = 1;
    842 		goto validate_dag_bad;
    843 	}
    844 	if (dag_h->status != rf_enable) {
    845 		printf("INVALID DAG: not enabled\n");
    846 		retcode = 1;
    847 		goto validate_dag_bad;
    848 	}
    849 	if (dag_h->numCommits != 0) {
    850 		printf("INVALID DAG: numCommits != 0 (%d)\n", dag_h->numCommits);
    851 		retcode = 1;
    852 		goto validate_dag_bad;
    853 	}
    854 	if (dag_h->numSuccedents != 1) {
    855 		/* currently, all dags must have only one succedent */
    856 		printf("INVALID DAG: numSuccedents !1 (%d)\n", dag_h->numSuccedents);
    857 		retcode = 1;
    858 		goto validate_dag_bad;
    859 	}
    860 	nodecount = rf_AssignNodeNums(dag_h);
    861 
    862 	unvisited = dag_h->succedents[0]->visited;
    863 
    864 	RF_Malloc(scount, nodecount * sizeof(int), (int *));
    865 	RF_Malloc(acount, nodecount * sizeof(int), (int *));
    866 	RF_Malloc(nodes, nodecount * sizeof(RF_DagNode_t *),
    867 		  (RF_DagNode_t **));
    868 	for (i = 0; i < dag_h->numSuccedents; i++) {
    869 		if ((dag_h->succedents[i]->visited == unvisited)
    870 		    && rf_ValidateBranch(dag_h->succedents[i], scount,
    871 			acount, nodes, unvisited)) {
    872 			retcode = 1;
    873 		}
    874 	}
    875 	/* start at 1 to skip the header node */
    876 	for (i = 1; i < nodecount; i++) {
    877 		if (nodes[i]->commitNode)
    878 			commitNodeCount++;
    879 		if (nodes[i]->doFunc == NULL) {
    880 			printf("INVALID DAG: node %s has an undefined doFunc\n", nodes[i]->name);
    881 			retcode = 1;
    882 			goto validate_dag_out;
    883 		}
    884 		if (nodes[i]->undoFunc == NULL) {
    885 			printf("INVALID DAG: node %s has an undefined doFunc\n", nodes[i]->name);
    886 			retcode = 1;
    887 			goto validate_dag_out;
    888 		}
    889 		if (nodes[i]->numAntecedents != scount[nodes[i]->nodeNum]) {
    890 			printf("INVALID DAG: node %s has %d antecedents but appears as a succedent %d times\n",
    891 			    nodes[i]->name, nodes[i]->numAntecedents, scount[nodes[i]->nodeNum]);
    892 			retcode = 1;
    893 			goto validate_dag_out;
    894 		}
    895 		if (nodes[i]->numSuccedents != acount[nodes[i]->nodeNum]) {
    896 			printf("INVALID DAG: node %s has %d succedents but appears as an antecedent %d times\n",
    897 			    nodes[i]->name, nodes[i]->numSuccedents, acount[nodes[i]->nodeNum]);
    898 			retcode = 1;
    899 			goto validate_dag_out;
    900 		}
    901 	}
    902 
    903 	if (dag_h->numCommitNodes != commitNodeCount) {
    904 		printf("INVALID DAG: incorrect commit node count.  hdr->numCommitNodes (%d) found (%d) commit nodes in graph\n",
    905 		    dag_h->numCommitNodes, commitNodeCount);
    906 		retcode = 1;
    907 		goto validate_dag_out;
    908 	}
    909 validate_dag_out:
    910 	RF_Free(scount, nodecount * sizeof(int));
    911 	RF_Free(acount, nodecount * sizeof(int));
    912 	RF_Free(nodes, nodecount * sizeof(RF_DagNode_t *));
    913 	if (retcode)
    914 		rf_PrintDAGList(dag_h);
    915 
    916 	if (rf_validateVisitedDebug)
    917 		rf_ValidateVisitedBits(dag_h);
    918 
    919 	return (retcode);
    920 
    921 validate_dag_bad:
    922 	rf_PrintDAGList(dag_h);
    923 	return (retcode);
    924 }
    925 
    926 #endif /* RF_DEBUG_VALIDATE_DAG */
    927 
    928 /******************************************************************************
    929  *
    930  * misc construction routines
    931  *
    932  *****************************************************************************/
    933 
    934 void
    935 rf_redirect_asm(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
    936 {
    937 	int     ds = (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) ? 1 : 0;
    938 	int     fcol = raidPtr->reconControl->fcol;
    939 	int     scol = raidPtr->reconControl->spareCol;
    940 	RF_PhysDiskAddr_t *pda;
    941 
    942 	RF_ASSERT(raidPtr->status == rf_rs_reconstructing);
    943 	for (pda = asmap->physInfo; pda; pda = pda->next) {
    944 		if (pda->col == fcol) {
    945 #if RF_DEBUG_DAG
    946 			if (rf_dagDebug) {
    947 				if (!rf_CheckRUReconstructed(raidPtr->reconControl->reconMap,
    948 					pda->startSector)) {
    949 					RF_PANIC();
    950 				}
    951 			}
    952 #endif
    953 			/* printf("Remapped data for large write\n"); */
    954 			if (ds) {
    955 				raidPtr->Layout.map->MapSector(raidPtr, pda->raidAddress,
    956 				    &pda->col, &pda->startSector, RF_REMAP);
    957 			} else {
    958 				pda->col = scol;
    959 			}
    960 		}
    961 	}
    962 	for (pda = asmap->parityInfo; pda; pda = pda->next) {
    963 		if (pda->col == fcol) {
    964 #if RF_DEBUG_DAG
    965 			if (rf_dagDebug) {
    966 				if (!rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, pda->startSector)) {
    967 					RF_PANIC();
    968 				}
    969 			}
    970 #endif
    971 		}
    972 		if (ds) {
    973 			(raidPtr->Layout.map->MapParity) (raidPtr, pda->raidAddress, &pda->col, &pda->startSector, RF_REMAP);
    974 		} else {
    975 			pda->col = scol;
    976 		}
    977 	}
    978 }
    979 
    980 
    981 /* this routine allocates read buffers and generates stripe maps for the
    982  * regions of the array from the start of the stripe to the start of the
    983  * access, and from the end of the access to the end of the stripe.  It also
    984  * computes and returns the number of DAG nodes needed to read all this data.
    985  * Note that this routine does the wrong thing if the access is fully
    986  * contained within one stripe unit, so we RF_ASSERT against this case at the
    987  * start.
    988  *
    989  * layoutPtr - in: layout information
    990  * asmap     - in: access stripe map
    991  * dag_h     - in: header of the dag to create
    992  * new_asm_h - in: ptr to array of 2 headers.  to be filled in
    993  * nRodNodes - out: num nodes to be generated to read unaccessed data
    994  * sosBuffer, eosBuffer - out: pointers to newly allocated buffer
    995  */
    996 void
    997 rf_MapUnaccessedPortionOfStripe(RF_Raid_t *raidPtr,
    998 				RF_RaidLayout_t *layoutPtr,
    999 				RF_AccessStripeMap_t *asmap,
   1000 				RF_DagHeader_t *dag_h,
   1001 				RF_AccessStripeMapHeader_t **new_asm_h,
   1002 				int *nRodNodes,
   1003 				char **sosBuffer, char **eosBuffer,
   1004 				RF_AllocListElem_t *allocList)
   1005 {
   1006 	RF_RaidAddr_t sosRaidAddress, eosRaidAddress;
   1007 	RF_SectorNum_t sosNumSector, eosNumSector;
   1008 
   1009 	RF_ASSERT(asmap->numStripeUnitsAccessed > (layoutPtr->numDataCol / 2));
   1010 	/* generate an access map for the region of the array from start of
   1011 	 * stripe to start of access */
   1012 	new_asm_h[0] = new_asm_h[1] = NULL;
   1013 	*nRodNodes = 0;
   1014 	if (!rf_RaidAddressStripeAligned(layoutPtr, asmap->raidAddress)) {
   1015 		sosRaidAddress = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
   1016 		sosNumSector = asmap->raidAddress - sosRaidAddress;
   1017 		*sosBuffer = rf_AllocStripeBuffer(raidPtr, dag_h, rf_RaidAddressToByte(raidPtr, sosNumSector));
   1018 		new_asm_h[0] = rf_MapAccess(raidPtr, sosRaidAddress, sosNumSector, *sosBuffer, RF_DONT_REMAP);
   1019 		new_asm_h[0]->next = dag_h->asmList;
   1020 		dag_h->asmList = new_asm_h[0];
   1021 		*nRodNodes += new_asm_h[0]->stripeMap->numStripeUnitsAccessed;
   1022 
   1023 		RF_ASSERT(new_asm_h[0]->stripeMap->next == NULL);
   1024 		/* we're totally within one stripe here */
   1025 		if (asmap->flags & RF_ASM_REDIR_LARGE_WRITE)
   1026 			rf_redirect_asm(raidPtr, new_asm_h[0]->stripeMap);
   1027 	}
   1028 	/* generate an access map for the region of the array from end of
   1029 	 * access to end of stripe */
   1030 	if (!rf_RaidAddressStripeAligned(layoutPtr, asmap->endRaidAddress)) {
   1031 		eosRaidAddress = asmap->endRaidAddress;
   1032 		eosNumSector = rf_RaidAddressOfNextStripeBoundary(layoutPtr, eosRaidAddress) - eosRaidAddress;
   1033 		*eosBuffer = rf_AllocStripeBuffer(raidPtr, dag_h, rf_RaidAddressToByte(raidPtr, eosNumSector));
   1034 		new_asm_h[1] = rf_MapAccess(raidPtr, eosRaidAddress, eosNumSector, *eosBuffer, RF_DONT_REMAP);
   1035 		new_asm_h[1]->next = dag_h->asmList;
   1036 		dag_h->asmList = new_asm_h[1];
   1037 		*nRodNodes += new_asm_h[1]->stripeMap->numStripeUnitsAccessed;
   1038 
   1039 		RF_ASSERT(new_asm_h[1]->stripeMap->next == NULL);
   1040 		/* we're totally within one stripe here */
   1041 		if (asmap->flags & RF_ASM_REDIR_LARGE_WRITE)
   1042 			rf_redirect_asm(raidPtr, new_asm_h[1]->stripeMap);
   1043 	}
   1044 }
   1045 
   1046 
   1047 
   1048 /* returns non-zero if the indicated ranges of stripe unit offsets overlap */
   1049 int
   1050 rf_PDAOverlap(RF_RaidLayout_t *layoutPtr,
   1051 	      RF_PhysDiskAddr_t *src, RF_PhysDiskAddr_t *dest)
   1052 {
   1053 	RF_SectorNum_t soffs = rf_StripeUnitOffset(layoutPtr, src->startSector);
   1054 	RF_SectorNum_t doffs = rf_StripeUnitOffset(layoutPtr, dest->startSector);
   1055 	/* use -1 to be sure we stay within SU */
   1056 	RF_SectorNum_t send = rf_StripeUnitOffset(layoutPtr, src->startSector + src->numSector - 1);
   1057 	RF_SectorNum_t dend = rf_StripeUnitOffset(layoutPtr, dest->startSector + dest->numSector - 1);
   1058 	return ((RF_MAX(soffs, doffs) <= RF_MIN(send, dend)) ? 1 : 0);
   1059 }
   1060 
   1061 
   1062 /* GenerateFailedAccessASMs
   1063  *
   1064  * this routine figures out what portion of the stripe needs to be read
   1065  * to effect the degraded read or write operation.  It's primary function
   1066  * is to identify everything required to recover the data, and then
   1067  * eliminate anything that is already being accessed by the user.
   1068  *
   1069  * The main result is two new ASMs, one for the region from the start of the
   1070  * stripe to the start of the access, and one for the region from the end of
   1071  * the access to the end of the stripe.  These ASMs describe everything that
   1072  * needs to be read to effect the degraded access.  Other results are:
   1073  *    nXorBufs -- the total number of buffers that need to be XORed together to
   1074  *                recover the lost data,
   1075  *    rpBufPtr -- ptr to a newly-allocated buffer to hold the parity.  If NULL
   1076  *                at entry, not allocated.
   1077  *    overlappingPDAs --
   1078  *                describes which of the non-failed PDAs in the user access
   1079  *                overlap data that needs to be read to effect recovery.
   1080  *                overlappingPDAs[i]==1 if and only if, neglecting the failed
   1081  *                PDA, the ith pda in the input asm overlaps data that needs
   1082  *                to be read for recovery.
   1083  */
   1084  /* in: asm - ASM for the actual access, one stripe only */
   1085  /* in: failedPDA - which component of the access has failed */
   1086  /* in: dag_h - header of the DAG we're going to create */
   1087  /* out: new_asm_h - the two new ASMs */
   1088  /* out: nXorBufs - the total number of xor bufs required */
   1089  /* out: rpBufPtr - a buffer for the parity read */
   1090 void
   1091 rf_GenerateFailedAccessASMs(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap,
   1092 			    RF_PhysDiskAddr_t *failedPDA,
   1093 			    RF_DagHeader_t *dag_h,
   1094 			    RF_AccessStripeMapHeader_t **new_asm_h,
   1095 			    int *nXorBufs, char **rpBufPtr,
   1096 			    char *overlappingPDAs,
   1097 			    RF_AllocListElem_t *allocList)
   1098 {
   1099 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
   1100 
   1101 	/* s=start, e=end, s=stripe, a=access, f=failed, su=stripe unit */
   1102 	RF_RaidAddr_t sosAddr, sosEndAddr, eosStartAddr, eosAddr;
   1103 	RF_PhysDiskAddr_t *pda;
   1104 	int     foundit, i;
   1105 
   1106 	foundit = 0;
   1107 	/* first compute the following raid addresses: start of stripe,
   1108 	 * (sosAddr) MIN(start of access, start of failed SU),   (sosEndAddr)
   1109 	 * MAX(end of access, end of failed SU),       (eosStartAddr) end of
   1110 	 * stripe (i.e. start of next stripe)   (eosAddr) */
   1111 	sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
   1112 	sosEndAddr = RF_MIN(asmap->raidAddress, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, failedPDA->raidAddress));
   1113 	eosStartAddr = RF_MAX(asmap->endRaidAddress, rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, failedPDA->raidAddress));
   1114 	eosAddr = rf_RaidAddressOfNextStripeBoundary(layoutPtr, asmap->raidAddress);
   1115 
   1116 	/* now generate access stripe maps for each of the above regions of
   1117 	 * the stripe.  Use a dummy (NULL) buf ptr for now */
   1118 
   1119 	new_asm_h[0] = (sosAddr != sosEndAddr) ? rf_MapAccess(raidPtr, sosAddr, sosEndAddr - sosAddr, NULL, RF_DONT_REMAP) : NULL;
   1120 	new_asm_h[1] = (eosStartAddr != eosAddr) ? rf_MapAccess(raidPtr, eosStartAddr, eosAddr - eosStartAddr, NULL, RF_DONT_REMAP) : NULL;
   1121 
   1122 	/* walk through the PDAs and range-restrict each SU to the region of
   1123 	 * the SU touched on the failed PDA.  also compute total data buffer
   1124 	 * space requirements in this step.  Ignore the parity for now. */
   1125 	/* Also count nodes to find out how many bufs need to be xored together */
   1126 	(*nXorBufs) = 1;	/* in read case, 1 is for parity.  In write
   1127 				 * case, 1 is for failed data */
   1128 
   1129 	if (new_asm_h[0]) {
   1130 		new_asm_h[0]->next = dag_h->asmList;
   1131 		dag_h->asmList = new_asm_h[0];
   1132 		for (pda = new_asm_h[0]->stripeMap->physInfo; pda; pda = pda->next) {
   1133 			rf_RangeRestrictPDA(raidPtr, failedPDA, pda, RF_RESTRICT_NOBUFFER, 0);
   1134 			pda->bufPtr = rf_AllocBuffer(raidPtr, dag_h, pda->numSector << raidPtr->logBytesPerSector);
   1135 		}
   1136 		(*nXorBufs) += new_asm_h[0]->stripeMap->numStripeUnitsAccessed;
   1137 	}
   1138 	if (new_asm_h[1]) {
   1139 		new_asm_h[1]->next = dag_h->asmList;
   1140 		dag_h->asmList = new_asm_h[1];
   1141 		for (pda = new_asm_h[1]->stripeMap->physInfo; pda; pda = pda->next) {
   1142 			rf_RangeRestrictPDA(raidPtr, failedPDA, pda, RF_RESTRICT_NOBUFFER, 0);
   1143 			pda->bufPtr = rf_AllocBuffer(raidPtr, dag_h, pda->numSector << raidPtr->logBytesPerSector);
   1144 		}
   1145 		(*nXorBufs) += new_asm_h[1]->stripeMap->numStripeUnitsAccessed;
   1146 	}
   1147 
   1148 	/* allocate a buffer for parity */
   1149 	if (rpBufPtr)
   1150 		*rpBufPtr = rf_AllocBuffer(raidPtr, dag_h, failedPDA->numSector << raidPtr->logBytesPerSector);
   1151 
   1152 	/* the last step is to figure out how many more distinct buffers need
   1153 	 * to get xor'd to produce the missing unit.  there's one for each
   1154 	 * user-data read node that overlaps the portion of the failed unit
   1155 	 * being accessed */
   1156 
   1157 	for (foundit = i = 0, pda = asmap->physInfo; pda; i++, pda = pda->next) {
   1158 		if (pda == failedPDA) {
   1159 			i--;
   1160 			foundit = 1;
   1161 			continue;
   1162 		}
   1163 		if (rf_PDAOverlap(layoutPtr, pda, failedPDA)) {
   1164 			overlappingPDAs[i] = 1;
   1165 			(*nXorBufs)++;
   1166 		}
   1167 	}
   1168 	if (!foundit) {
   1169 		RF_ERRORMSG("GenerateFailedAccessASMs: did not find failedPDA in asm list\n");
   1170 		RF_ASSERT(0);
   1171 	}
   1172 #if RF_DEBUG_DAG
   1173 	if (rf_degDagDebug) {
   1174 		if (new_asm_h[0]) {
   1175 			printf("First asm:\n");
   1176 			rf_PrintFullAccessStripeMap(new_asm_h[0], 1);
   1177 		}
   1178 		if (new_asm_h[1]) {
   1179 			printf("Second asm:\n");
   1180 			rf_PrintFullAccessStripeMap(new_asm_h[1], 1);
   1181 		}
   1182 	}
   1183 #endif
   1184 }
   1185 
   1186 
   1187 /* adjusts the offset and number of sectors in the destination pda so that
   1188  * it covers at most the region of the SU covered by the source PDA.  This
   1189  * is exclusively a restriction:  the number of sectors indicated by the
   1190  * target PDA can only shrink.
   1191  *
   1192  * For example:  s = sectors within SU indicated by source PDA
   1193  *               d = sectors within SU indicated by dest PDA
   1194  *               r = results, stored in dest PDA
   1195  *
   1196  * |--------------- one stripe unit ---------------------|
   1197  * |           sssssssssssssssssssssssssssssssss         |
   1198  * |    ddddddddddddddddddddddddddddddddddddddddddddd    |
   1199  * |           rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr         |
   1200  *
   1201  * Another example:
   1202  *
   1203  * |--------------- one stripe unit ---------------------|
   1204  * |           sssssssssssssssssssssssssssssssss         |
   1205  * |    ddddddddddddddddddddddd                          |
   1206  * |           rrrrrrrrrrrrrrrr                          |
   1207  *
   1208  */
   1209 void
   1210 rf_RangeRestrictPDA(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *src,
   1211 		    RF_PhysDiskAddr_t *dest, int dobuffer, int doraidaddr)
   1212 {
   1213 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
   1214 	RF_SectorNum_t soffs = rf_StripeUnitOffset(layoutPtr, src->startSector);
   1215 	RF_SectorNum_t doffs = rf_StripeUnitOffset(layoutPtr, dest->startSector);
   1216 	RF_SectorNum_t send = rf_StripeUnitOffset(layoutPtr, src->startSector + src->numSector - 1);	/* use -1 to be sure we
   1217 													 * stay within SU */
   1218 	RF_SectorNum_t dend = rf_StripeUnitOffset(layoutPtr, dest->startSector + dest->numSector - 1);
   1219 	RF_SectorNum_t subAddr = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, dest->startSector);	/* stripe unit boundary */
   1220 
   1221 	dest->startSector = subAddr + RF_MAX(soffs, doffs);
   1222 	dest->numSector = subAddr + RF_MIN(send, dend) + 1 - dest->startSector;
   1223 
   1224 	if (dobuffer)
   1225 		dest->bufPtr += (soffs > doffs) ? rf_RaidAddressToByte(raidPtr, soffs - doffs) : 0;
   1226 	if (doraidaddr) {
   1227 		dest->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, dest->raidAddress) +
   1228 		    rf_StripeUnitOffset(layoutPtr, dest->startSector);
   1229 	}
   1230 }
   1231 
   1232 #if (RF_INCLUDE_CHAINDECLUSTER > 0)
   1233 
   1234 /*
   1235  * Want the highest of these primes to be the largest one
   1236  * less than the max expected number of columns (won't hurt
   1237  * to be too small or too large, but won't be optimal, either)
   1238  * --jimz
   1239  */
   1240 #define NLOWPRIMES 8
   1241 static int lowprimes[NLOWPRIMES] = {2, 3, 5, 7, 11, 13, 17, 19};
   1242 /*****************************************************************************
   1243  * compute the workload shift factor.  (chained declustering)
   1244  *
   1245  * return nonzero if access should shift to secondary, otherwise,
   1246  * access is to primary
   1247  *****************************************************************************/
   1248 int
   1249 rf_compute_workload_shift(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda)
   1250 {
   1251 	/*
   1252          * variables:
   1253          *  d   = column of disk containing primary
   1254          *  f   = column of failed disk
   1255          *  n   = number of disks in array
   1256          *  sd  = "shift distance" (number of columns that d is to the right of f)
   1257          *  v   = numerator of redirection ratio
   1258          *  k   = denominator of redirection ratio
   1259          */
   1260 	RF_RowCol_t d, f, sd, n;
   1261 	int     k, v, ret, i;
   1262 
   1263 	n = raidPtr->numCol;
   1264 
   1265 	/* assign column of primary copy to d */
   1266 	d = pda->col;
   1267 
   1268 	/* assign column of dead disk to f */
   1269 	for (f = 0; ((!RF_DEAD_DISK(raidPtr->Disks[f].status)) && (f < n)); f++);
   1270 
   1271 	RF_ASSERT(f < n);
   1272 	RF_ASSERT(f != d);
   1273 
   1274 	sd = (f > d) ? (n + d - f) : (d - f);
   1275 	RF_ASSERT(sd < n);
   1276 
   1277 	/*
   1278          * v of every k accesses should be redirected
   1279          *
   1280          * v/k := (n-1-sd)/(n-1)
   1281          */
   1282 	v = (n - 1 - sd);
   1283 	k = (n - 1);
   1284 
   1285 #if 1
   1286 	/*
   1287          * XXX
   1288          * Is this worth it?
   1289          *
   1290          * Now reduce the fraction, by repeatedly factoring
   1291          * out primes (just like they teach in elementary school!)
   1292          */
   1293 	for (i = 0; i < NLOWPRIMES; i++) {
   1294 		if (lowprimes[i] > v)
   1295 			break;
   1296 		while (((v % lowprimes[i]) == 0) && ((k % lowprimes[i]) == 0)) {
   1297 			v /= lowprimes[i];
   1298 			k /= lowprimes[i];
   1299 		}
   1300 	}
   1301 #endif
   1302 
   1303 	raidPtr->hist_diskreq[d]++;
   1304 	if (raidPtr->hist_diskreq[d] > v) {
   1305 		ret = 0;	/* do not redirect */
   1306 	} else {
   1307 		ret = 1;	/* redirect */
   1308 	}
   1309 
   1310 #if 0
   1311 	printf("d=%d f=%d sd=%d v=%d k=%d ret=%d h=%d\n", d, f, sd, v, k, ret,
   1312 	    raidPtr->hist_diskreq[d]);
   1313 #endif
   1314 
   1315 	if (raidPtr->hist_diskreq[d] >= k) {
   1316 		/* reset counter */
   1317 		raidPtr->hist_diskreq[d] = 0;
   1318 	}
   1319 	return (ret);
   1320 }
   1321 #endif /* (RF_INCLUDE_CHAINDECLUSTER > 0) */
   1322 
   1323 /*
   1324  * Disk selection routines
   1325  */
   1326 
   1327 /*
   1328  * Selects the disk with the shortest queue from a mirror pair.
   1329  * Both the disk I/Os queued in RAIDframe as well as those at the physical
   1330  * disk are counted as members of the "queue"
   1331  */
   1332 void
   1333 rf_SelectMirrorDiskIdle(RF_DagNode_t * node)
   1334 {
   1335 	RF_Raid_t *raidPtr = (RF_Raid_t *) node->dagHdr->raidPtr;
   1336 	RF_RowCol_t colData, colMirror;
   1337 	int     dataQueueLength, mirrorQueueLength, usemirror;
   1338 	RF_PhysDiskAddr_t *data_pda = (RF_PhysDiskAddr_t *) node->params[0].p;
   1339 	RF_PhysDiskAddr_t *mirror_pda = (RF_PhysDiskAddr_t *) node->params[4].p;
   1340 	RF_PhysDiskAddr_t *tmp_pda;
   1341 	RF_RaidDisk_t *disks = raidPtr->Disks;
   1342 	RF_DiskQueue_t *dqs = raidPtr->Queues, *dataQueue, *mirrorQueue;
   1343 
   1344 	/* return the [row col] of the disk with the shortest queue */
   1345 	colData = data_pda->col;
   1346 	colMirror = mirror_pda->col;
   1347 	dataQueue = &(dqs[colData]);
   1348 	mirrorQueue = &(dqs[colMirror]);
   1349 
   1350 #ifdef RF_LOCK_QUEUES_TO_READ_LEN
   1351 	RF_LOCK_QUEUE_MUTEX(dataQueue, "SelectMirrorDiskIdle");
   1352 #endif				/* RF_LOCK_QUEUES_TO_READ_LEN */
   1353 	dataQueueLength = dataQueue->queueLength + dataQueue->numOutstanding;
   1354 #ifdef RF_LOCK_QUEUES_TO_READ_LEN
   1355 	RF_UNLOCK_QUEUE_MUTEX(dataQueue, "SelectMirrorDiskIdle");
   1356 	RF_LOCK_QUEUE_MUTEX(mirrorQueue, "SelectMirrorDiskIdle");
   1357 #endif				/* RF_LOCK_QUEUES_TO_READ_LEN */
   1358 	mirrorQueueLength = mirrorQueue->queueLength + mirrorQueue->numOutstanding;
   1359 #ifdef RF_LOCK_QUEUES_TO_READ_LEN
   1360 	RF_UNLOCK_QUEUE_MUTEX(mirrorQueue, "SelectMirrorDiskIdle");
   1361 #endif				/* RF_LOCK_QUEUES_TO_READ_LEN */
   1362 
   1363 	usemirror = 0;
   1364 	if (RF_DEAD_DISK(disks[colMirror].status)) {
   1365 		usemirror = 0;
   1366 	} else
   1367 		if (RF_DEAD_DISK(disks[colData].status)) {
   1368 			usemirror = 1;
   1369 		} else
   1370 			if (raidPtr->parity_good == RF_RAID_DIRTY) {
   1371 				/* Trust only the main disk */
   1372 				usemirror = 0;
   1373 			} else
   1374 				if (dataQueueLength < mirrorQueueLength) {
   1375 					usemirror = 0;
   1376 				} else
   1377 					if (mirrorQueueLength < dataQueueLength) {
   1378 						usemirror = 1;
   1379 					} else {
   1380 						/* queues are equal length. attempt
   1381 						 * cleverness. */
   1382 						if (SNUM_DIFF(dataQueue->last_deq_sector, data_pda->startSector)
   1383 						    <= SNUM_DIFF(mirrorQueue->last_deq_sector, mirror_pda->startSector)) {
   1384 							usemirror = 0;
   1385 						} else {
   1386 							usemirror = 1;
   1387 						}
   1388 					}
   1389 
   1390 	if (usemirror) {
   1391 		/* use mirror (parity) disk, swap params 0 & 4 */
   1392 		tmp_pda = data_pda;
   1393 		node->params[0].p = mirror_pda;
   1394 		node->params[4].p = tmp_pda;
   1395 	} else {
   1396 		/* use data disk, leave param 0 unchanged */
   1397 	}
   1398 	/* printf("dataQueueLength %d, mirrorQueueLength
   1399 	 * %d\n",dataQueueLength, mirrorQueueLength); */
   1400 }
   1401 #if (RF_INCLUDE_CHAINDECLUSTER > 0) || (RF_INCLUDE_INTERDECLUSTER > 0) || (RF_DEBUG_VALIDATE_DAG > 0)
   1402 /*
   1403  * Do simple partitioning. This assumes that
   1404  * the data and parity disks are laid out identically.
   1405  */
   1406 void
   1407 rf_SelectMirrorDiskPartition(RF_DagNode_t * node)
   1408 {
   1409 	RF_Raid_t *raidPtr = (RF_Raid_t *) node->dagHdr->raidPtr;
   1410 	RF_RowCol_t colData, colMirror;
   1411 	RF_PhysDiskAddr_t *data_pda = (RF_PhysDiskAddr_t *) node->params[0].p;
   1412 	RF_PhysDiskAddr_t *mirror_pda = (RF_PhysDiskAddr_t *) node->params[4].p;
   1413 	RF_PhysDiskAddr_t *tmp_pda;
   1414 	RF_RaidDisk_t *disks = raidPtr->Disks;
   1415 	int     usemirror;
   1416 
   1417 	/* return the [row col] of the disk with the shortest queue */
   1418 	colData = data_pda->col;
   1419 	colMirror = mirror_pda->col;
   1420 
   1421 	usemirror = 0;
   1422 	if (RF_DEAD_DISK(disks[colMirror].status)) {
   1423 		usemirror = 0;
   1424 	} else
   1425 		if (RF_DEAD_DISK(disks[colData].status)) {
   1426 			usemirror = 1;
   1427 		} else
   1428 			if (raidPtr->parity_good == RF_RAID_DIRTY) {
   1429 				/* Trust only the main disk */
   1430 				usemirror = 0;
   1431 			} else
   1432 				if (data_pda->startSector <
   1433 				    (disks[colData].numBlocks / 2)) {
   1434 					usemirror = 0;
   1435 				} else {
   1436 					usemirror = 1;
   1437 				}
   1438 
   1439 	if (usemirror) {
   1440 		/* use mirror (parity) disk, swap params 0 & 4 */
   1441 		tmp_pda = data_pda;
   1442 		node->params[0].p = mirror_pda;
   1443 		node->params[4].p = tmp_pda;
   1444 	} else {
   1445 		/* use data disk, leave param 0 unchanged */
   1446 	}
   1447 }
   1448 #endif
   1449