Home | History | Annotate | Line # | Download | only in raidframe
rf_states.c revision 1.28
      1 /*	$NetBSD: rf_states.c,v 1.28 2004/02/29 22:11:54 oster Exp $	*/
      2 /*
      3  * Copyright (c) 1995 Carnegie-Mellon University.
      4  * All rights reserved.
      5  *
      6  * Author: Mark Holland, William V. Courtright II, Robby Findler
      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 #include <sys/cdefs.h>
     30 __KERNEL_RCSID(0, "$NetBSD: rf_states.c,v 1.28 2004/02/29 22:11:54 oster Exp $");
     31 
     32 #include <sys/errno.h>
     33 
     34 #include "rf_archs.h"
     35 #include "rf_threadstuff.h"
     36 #include "rf_raid.h"
     37 #include "rf_dag.h"
     38 #include "rf_desc.h"
     39 #include "rf_aselect.h"
     40 #include "rf_general.h"
     41 #include "rf_states.h"
     42 #include "rf_dagutils.h"
     43 #include "rf_driver.h"
     44 #include "rf_engine.h"
     45 #include "rf_map.h"
     46 #include "rf_etimer.h"
     47 #include "rf_kintf.h"
     48 
     49 #ifndef RF_DEBUG_STATES
     50 #define RF_DEBUG_STATES 0
     51 #endif
     52 
     53 /* prototypes for some of the available states.
     54 
     55    States must:
     56 
     57      - not block.
     58 
     59      - either schedule rf_ContinueRaidAccess as a callback and return
     60        RF_TRUE, or complete all of their work and return RF_FALSE.
     61 
     62      - increment desc->state when they have finished their work.
     63 */
     64 
     65 #if RF_DEBUG_STATES
     66 static char *
     67 StateName(RF_AccessState_t state)
     68 {
     69 	switch (state) {
     70 		case rf_QuiesceState:return "QuiesceState";
     71 	case rf_MapState:
     72 		return "MapState";
     73 	case rf_LockState:
     74 		return "LockState";
     75 	case rf_CreateDAGState:
     76 		return "CreateDAGState";
     77 	case rf_ExecuteDAGState:
     78 		return "ExecuteDAGState";
     79 	case rf_ProcessDAGState:
     80 		return "ProcessDAGState";
     81 	case rf_CleanupState:
     82 		return "CleanupState";
     83 	case rf_LastState:
     84 		return "LastState";
     85 	case rf_IncrAccessesCountState:
     86 		return "IncrAccessesCountState";
     87 	case rf_DecrAccessesCountState:
     88 		return "DecrAccessesCountState";
     89 	default:
     90 		return "!!! UnnamedState !!!";
     91 	}
     92 }
     93 #endif
     94 
     95 void
     96 rf_ContinueRaidAccess(RF_RaidAccessDesc_t *desc)
     97 {
     98 	int     suspended = RF_FALSE;
     99 	int     current_state_index = desc->state;
    100 	RF_AccessState_t current_state = desc->states[current_state_index];
    101 #if RF_DEBUG_STATES
    102 	int     unit = desc->raidPtr->raidid;
    103 #endif
    104 
    105 	do {
    106 
    107 		current_state_index = desc->state;
    108 		current_state = desc->states[current_state_index];
    109 
    110 		switch (current_state) {
    111 
    112 		case rf_QuiesceState:
    113 			suspended = rf_State_Quiesce(desc);
    114 			break;
    115 		case rf_IncrAccessesCountState:
    116 			suspended = rf_State_IncrAccessCount(desc);
    117 			break;
    118 		case rf_MapState:
    119 			suspended = rf_State_Map(desc);
    120 			break;
    121 		case rf_LockState:
    122 			suspended = rf_State_Lock(desc);
    123 			break;
    124 		case rf_CreateDAGState:
    125 			suspended = rf_State_CreateDAG(desc);
    126 			break;
    127 		case rf_ExecuteDAGState:
    128 			suspended = rf_State_ExecuteDAG(desc);
    129 			break;
    130 		case rf_ProcessDAGState:
    131 			suspended = rf_State_ProcessDAG(desc);
    132 			break;
    133 		case rf_CleanupState:
    134 			suspended = rf_State_Cleanup(desc);
    135 			break;
    136 		case rf_DecrAccessesCountState:
    137 			suspended = rf_State_DecrAccessCount(desc);
    138 			break;
    139 		case rf_LastState:
    140 			suspended = rf_State_LastState(desc);
    141 			break;
    142 		}
    143 
    144 		/* after this point, we cannot dereference desc since
    145 		 * desc may have been freed. desc is only freed in
    146 		 * LastState, so if we renter this function or loop
    147 		 * back up, desc should be valid. */
    148 
    149 #if RF_DEBUG_STATES
    150 		if (rf_printStatesDebug) {
    151 			printf("raid%d: State: %-24s StateIndex: %3i desc: 0x%ld %s\n",
    152 			       unit, StateName(current_state),
    153 			       current_state_index, (long) desc,
    154 			       suspended ? "callback scheduled" : "looping");
    155 		}
    156 #endif
    157 	} while (!suspended && current_state != rf_LastState);
    158 
    159 	return;
    160 }
    161 
    162 
    163 void
    164 rf_ContinueDagAccess(RF_DagList_t *dagList)
    165 {
    166 #if RF_ACC_TRACE > 0
    167 	RF_AccTraceEntry_t *tracerec = &(dagList->desc->tracerec);
    168 	RF_Etimer_t timer;
    169 #endif
    170 	RF_RaidAccessDesc_t *desc;
    171 	RF_DagHeader_t *dag_h;
    172 	int     i;
    173 
    174 	desc = dagList->desc;
    175 
    176 #if RF_ACC_TRACE > 0
    177 	timer = tracerec->timer;
    178 	RF_ETIMER_STOP(timer);
    179 	RF_ETIMER_EVAL(timer);
    180 	tracerec->specific.user.exec_us = RF_ETIMER_VAL_US(timer);
    181 	RF_ETIMER_START(tracerec->timer);
    182 #endif
    183 
    184 	/* skip to dag which just finished */
    185 	dag_h = dagList->dags;
    186 	for (i = 0; i < dagList->numDagsDone; i++) {
    187 		dag_h = dag_h->next;
    188 	}
    189 
    190 	/* check to see if retry is required */
    191 	if (dag_h->status == rf_rollBackward) {
    192 		/* when a dag fails, mark desc status as bad and allow
    193 		 * all other dags in the desc to execute to
    194 		 * completion.  then, free all dags and start over */
    195 		desc->status = 1;	/* bad status */
    196 #if 0
    197 		printf("raid%d: DAG failure: %c addr 0x%lx "
    198 		       "(%ld) nblk 0x%x (%d) buf 0x%lx state %d\n",
    199 		       desc->raidPtr->raidid, desc->type,
    200 		       (long) desc->raidAddress,
    201 		       (long) desc->raidAddress, (int) desc->numBlocks,
    202 		       (int) desc->numBlocks,
    203 		       (unsigned long) (desc->bufPtr), desc->state);
    204 #endif
    205 	}
    206 	dagList->numDagsDone++;
    207 	rf_ContinueRaidAccess(desc);
    208 }
    209 
    210 int
    211 rf_State_LastState(RF_RaidAccessDesc_t *desc)
    212 {
    213 	void    (*callbackFunc) (RF_CBParam_t) = desc->callbackFunc;
    214 	RF_CBParam_t callbackArg;
    215 
    216 	callbackArg.p = desc->callbackArg;
    217 
    218 	/*
    219 	 * If this is not an async request, wake up the caller
    220 	 */
    221 	if (desc->async_flag == 0)
    222 		wakeup(desc->bp);
    223 
    224 	/*
    225 	 * That's all the IO for this one... unbusy the 'disk'.
    226 	 */
    227 
    228 	rf_disk_unbusy(desc);
    229 
    230 	/*
    231 	 * Wakeup any requests waiting to go.
    232 	 */
    233 
    234 	RF_LOCK_MUTEX(((RF_Raid_t *) desc->raidPtr)->mutex);
    235 	((RF_Raid_t *) desc->raidPtr)->openings++;
    236 	RF_UNLOCK_MUTEX(((RF_Raid_t *) desc->raidPtr)->mutex);
    237 
    238 	/* wake up any pending IO */
    239 	raidstart(((RF_Raid_t *) desc->raidPtr));
    240 
    241 	/* printf("Calling biodone on 0x%x\n",desc->bp); */
    242 	biodone(desc->bp);	/* access came through ioctl */
    243 
    244 	if (callbackFunc)
    245 		callbackFunc(callbackArg);
    246 	rf_FreeRaidAccDesc(desc);
    247 
    248 	return RF_FALSE;
    249 }
    250 
    251 int
    252 rf_State_IncrAccessCount(RF_RaidAccessDesc_t *desc)
    253 {
    254 	RF_Raid_t *raidPtr;
    255 
    256 	raidPtr = desc->raidPtr;
    257 	/* Bummer. We have to do this to be 100% safe w.r.t. the increment
    258 	 * below */
    259 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
    260 	raidPtr->accs_in_flight++;	/* used to detect quiescence */
    261 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
    262 
    263 	desc->state++;
    264 	return RF_FALSE;
    265 }
    266 
    267 int
    268 rf_State_DecrAccessCount(RF_RaidAccessDesc_t *desc)
    269 {
    270 	RF_Raid_t *raidPtr;
    271 
    272 	raidPtr = desc->raidPtr;
    273 
    274 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
    275 	raidPtr->accs_in_flight--;
    276 	if (raidPtr->accesses_suspended && raidPtr->accs_in_flight == 0) {
    277 		rf_SignalQuiescenceLock(raidPtr);
    278 	}
    279 	rf_UpdateUserStats(raidPtr, RF_ETIMER_VAL_US(desc->timer), desc->numBlocks);
    280 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
    281 
    282 	desc->state++;
    283 	return RF_FALSE;
    284 }
    285 
    286 int
    287 rf_State_Quiesce(RF_RaidAccessDesc_t *desc)
    288 {
    289 #if RF_ACC_TRACE > 0
    290 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
    291 	RF_Etimer_t timer;
    292 #endif
    293 	int     suspended = RF_FALSE;
    294 	RF_Raid_t *raidPtr;
    295 
    296 	raidPtr = desc->raidPtr;
    297 
    298 #if RF_ACC_TRACE > 0
    299 	RF_ETIMER_START(timer);
    300 	RF_ETIMER_START(desc->timer);
    301 #endif
    302 
    303 	RF_LOCK_MUTEX(raidPtr->access_suspend_mutex);
    304 	if (raidPtr->accesses_suspended) {
    305 		RF_CallbackDesc_t *cb;
    306 		cb = rf_AllocCallbackDesc();
    307 
    308 		cb->callbackFunc = (void (*) (RF_CBParam_t)) rf_ContinueRaidAccess;
    309 		cb->callbackArg.p = (void *) desc;
    310 		cb->next = raidPtr->quiesce_wait_list;
    311 		raidPtr->quiesce_wait_list = cb;
    312 		suspended = RF_TRUE;
    313 	}
    314 	RF_UNLOCK_MUTEX(raidPtr->access_suspend_mutex);
    315 
    316 #if RF_ACC_TRACE > 0
    317 	RF_ETIMER_STOP(timer);
    318 	RF_ETIMER_EVAL(timer);
    319 	tracerec->specific.user.suspend_ovhd_us += RF_ETIMER_VAL_US(timer);
    320 #endif
    321 
    322 #if RF_DEBUG_QUIESCE
    323 	if (suspended && rf_quiesceDebug)
    324 		printf("Stalling access due to quiescence lock\n");
    325 #endif
    326 	desc->state++;
    327 	return suspended;
    328 }
    329 
    330 int
    331 rf_State_Map(RF_RaidAccessDesc_t *desc)
    332 {
    333 	RF_Raid_t *raidPtr = desc->raidPtr;
    334 #if RF_ACC_TRACE > 0
    335 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
    336 	RF_Etimer_t timer;
    337 
    338 	RF_ETIMER_START(timer);
    339 #endif
    340 
    341 	if (!(desc->asmap = rf_MapAccess(raidPtr, desc->raidAddress, desc->numBlocks,
    342 		    desc->bufPtr, RF_DONT_REMAP)))
    343 		RF_PANIC();
    344 
    345 #if RF_ACC_TRACE > 0
    346 	RF_ETIMER_STOP(timer);
    347 	RF_ETIMER_EVAL(timer);
    348 	tracerec->specific.user.map_us = RF_ETIMER_VAL_US(timer);
    349 #endif
    350 
    351 	desc->state++;
    352 	return RF_FALSE;
    353 }
    354 
    355 int
    356 rf_State_Lock(RF_RaidAccessDesc_t *desc)
    357 {
    358 #if RF_ACC_TRACE > 0
    359 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
    360 	RF_Etimer_t timer;
    361 #endif
    362 	RF_Raid_t *raidPtr = desc->raidPtr;
    363 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
    364 	RF_AccessStripeMap_t *asm_p;
    365 	int     suspended = RF_FALSE;
    366 
    367 #if RF_ACC_TRACE > 0
    368 	RF_ETIMER_START(timer);
    369 #endif
    370 	if (!(raidPtr->Layout.map->flags & RF_NO_STRIPE_LOCKS)) {
    371 		RF_StripeNum_t lastStripeID = -1;
    372 
    373 		/* acquire each lock that we don't already hold */
    374 		for (asm_p = asmh->stripeMap; asm_p; asm_p = asm_p->next) {
    375 			RF_ASSERT(RF_IO_IS_R_OR_W(desc->type));
    376 			if (!rf_suppressLocksAndLargeWrites &&
    377 			    asm_p->parityInfo &&
    378 			    !(desc->flags & RF_DAG_SUPPRESS_LOCKS) &&
    379 			    !(asm_p->flags & RF_ASM_FLAGS_LOCK_TRIED)) {
    380 				asm_p->flags |= RF_ASM_FLAGS_LOCK_TRIED;
    381 				/* locks must be acquired hierarchically */
    382 				RF_ASSERT(asm_p->stripeID > lastStripeID);
    383 				lastStripeID = asm_p->stripeID;
    384 
    385 				RF_INIT_LOCK_REQ_DESC(asm_p->lockReqDesc, desc->type,
    386 				    (void (*) (struct buf *)) rf_ContinueRaidAccess, desc, asm_p,
    387 				    raidPtr->Layout.dataSectorsPerStripe);
    388 				if (rf_AcquireStripeLock(raidPtr->lockTable, asm_p->stripeID,
    389 					&asm_p->lockReqDesc)) {
    390 					suspended = RF_TRUE;
    391 					break;
    392 				}
    393 			}
    394 			if (desc->type == RF_IO_TYPE_WRITE &&
    395 			    raidPtr->status == rf_rs_reconstructing) {
    396 				if (!(asm_p->flags & RF_ASM_FLAGS_FORCE_TRIED)) {
    397 					int     val;
    398 
    399 					asm_p->flags |= RF_ASM_FLAGS_FORCE_TRIED;
    400 					val = rf_ForceOrBlockRecon(raidPtr, asm_p,
    401 					    (void (*) (RF_Raid_t *, void *)) rf_ContinueRaidAccess, desc);
    402 					if (val == 0) {
    403 						asm_p->flags |= RF_ASM_FLAGS_RECON_BLOCKED;
    404 					} else {
    405 						suspended = RF_TRUE;
    406 						break;
    407 					}
    408 				} else {
    409 					if (rf_pssDebug) {
    410 						printf("raid%d: skipping force/block because already done, psid %ld\n",
    411 						       desc->raidPtr->raidid,
    412 						       (long) asm_p->stripeID);
    413 					}
    414 				}
    415 			} else {
    416 				if (rf_pssDebug) {
    417 					printf("raid%d: skipping force/block because not write or not under recon, psid %ld\n",
    418 					       desc->raidPtr->raidid,
    419 					       (long) asm_p->stripeID);
    420 				}
    421 			}
    422 		}
    423 #if RF_ACC_TRACE > 0
    424 		RF_ETIMER_STOP(timer);
    425 		RF_ETIMER_EVAL(timer);
    426 		tracerec->specific.user.lock_us += RF_ETIMER_VAL_US(timer);
    427 #endif
    428 		if (suspended)
    429 			return (RF_TRUE);
    430 	}
    431 	desc->state++;
    432 	return (RF_FALSE);
    433 }
    434 /*
    435  * the following three states create, execute, and post-process dags
    436  * the error recovery unit is a single dag.
    437  * by default, SelectAlgorithm creates an array of dags, one per parity stripe
    438  * in some tricky cases, multiple dags per stripe are created
    439  *   - dags within a parity stripe are executed sequentially (arbitrary order)
    440  *   - dags for distinct parity stripes are executed concurrently
    441  *
    442  * repeat until all dags complete successfully -or- dag selection fails
    443  *
    444  * while !done
    445  *   create dag(s) (SelectAlgorithm)
    446  *   if dag
    447  *     execute dag (DispatchDAG)
    448  *     if dag successful
    449  *       done (SUCCESS)
    450  *     else
    451  *       !done (RETRY - start over with new dags)
    452  *   else
    453  *     done (FAIL)
    454  */
    455 int
    456 rf_State_CreateDAG(RF_RaidAccessDesc_t *desc)
    457 {
    458 #if RF_ACC_TRACE > 0
    459 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
    460 	RF_Etimer_t timer;
    461 #endif
    462 	RF_DagHeader_t *dag_h;
    463 	RF_DagList_t *dagList;
    464 	struct buf *bp;
    465 	int     i, selectStatus;
    466 
    467 	/* generate a dag for the access, and fire it off.  When the dag
    468 	 * completes, we'll get re-invoked in the next state. */
    469 #if RF_ACC_TRACE > 0
    470 	RF_ETIMER_START(timer);
    471 #endif
    472 	/* SelectAlgorithm returns one or more dags */
    473 	selectStatus = rf_SelectAlgorithm(desc, desc->flags | RF_DAG_SUPPRESS_LOCKS);
    474 #if RF_DEBUG_VALIDATE_DAG
    475 	if (rf_printDAGsDebug) {
    476 		dagList = desc->dagList;
    477 		for (i = 0; i < desc->numStripes; i++) {
    478 			rf_PrintDAGList(dagList.dags);
    479 			dagList = dagList->next;
    480 		}
    481 	}
    482 #endif /* RF_DEBUG_VALIDATE_DAG */
    483 #if RF_ACC_TRACE > 0
    484 	RF_ETIMER_STOP(timer);
    485 	RF_ETIMER_EVAL(timer);
    486 	/* update time to create all dags */
    487 	tracerec->specific.user.dag_create_us = RF_ETIMER_VAL_US(timer);
    488 #endif
    489 
    490 	desc->status = 0;	/* good status */
    491 
    492 	if (selectStatus) {
    493 		/* failed to create a dag */
    494 		/* this happens when there are too many faults or incomplete
    495 		 * dag libraries */
    496 		printf("raid%d: failed to create a dag. "
    497 		       "Too many component failures.\n",
    498 		       desc->raidPtr->raidid);
    499 
    500 		desc->status = 1; /* bad status */
    501 		/* skip straight to rf_State_Cleanup() */
    502 		desc->state = rf_CleanupState;
    503 		bp = (struct buf *)desc->bp;
    504 		bp->b_flags |= B_ERROR;
    505 		bp->b_error = EIO;
    506 	} else {
    507 		/* bind dags to desc */
    508 		dagList = desc->dagList;
    509 		for (i = 0; i < desc->numStripes; i++) {
    510 			dag_h = dagList->dags;
    511 			while (dag_h) {
    512 				dag_h->bp = (struct buf *) desc->bp;
    513 #if RF_ACC_TRACE > 0
    514 				dag_h->tracerec = tracerec;
    515 #endif
    516 				dag_h = dag_h->next;
    517 			}
    518 			dagList = dagList->next;
    519 		}
    520 		desc->flags |= RF_DAG_DISPATCH_RETURNED;
    521 		desc->state++;	/* next state should be rf_State_ExecuteDAG */
    522 	}
    523 	return RF_FALSE;
    524 }
    525 
    526 
    527 
    528 /* the access has an list of dagLists, one dagList per parity stripe.
    529  * fire the first dag in each parity stripe (dagList).
    530  * dags within a stripe (dagList) must be executed sequentially
    531  *  - this preserves atomic parity update
    532  * dags for independents parity groups (stripes) are fired concurrently */
    533 
    534 int
    535 rf_State_ExecuteDAG(RF_RaidAccessDesc_t *desc)
    536 {
    537 	int     i;
    538 	RF_DagHeader_t *dag_h;
    539 	RF_DagList_t *dagList;
    540 
    541 	/* next state is always rf_State_ProcessDAG important to do
    542 	 * this before firing the first dag (it may finish before we
    543 	 * leave this routine) */
    544 	desc->state++;
    545 
    546 	/* sweep dag array, a stripe at a time, firing the first dag
    547 	 * in each stripe */
    548 	dagList = desc->dagList;
    549 	for (i = 0; i < desc->numStripes; i++) {
    550 		RF_ASSERT(dagList->numDags > 0);
    551 		RF_ASSERT(dagList->numDagsDone == 0);
    552 		RF_ASSERT(dagList->numDagsFired == 0);
    553 #if RF_ACC_TRACE > 0
    554 		RF_ETIMER_START(dagList->tracerec.timer);
    555 #endif
    556 		/* fire first dag in this stripe */
    557 		dag_h = dagList->dags;
    558 		RF_ASSERT(dag_h);
    559 		dagList->numDagsFired++;
    560 		rf_DispatchDAG(dag_h, (void (*) (void *)) rf_ContinueDagAccess, dagList);
    561 		dagList = dagList->next;
    562 	}
    563 
    564 	/* the DAG will always call the callback, even if there was no
    565 	 * blocking, so we are always suspended in this state */
    566 	return RF_TRUE;
    567 }
    568 
    569 
    570 
    571 /* rf_State_ProcessDAG is entered when a dag completes.
    572  * first, check to all dags in the access have completed
    573  * if not, fire as many dags as possible */
    574 
    575 int
    576 rf_State_ProcessDAG(RF_RaidAccessDesc_t *desc)
    577 {
    578 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
    579 	RF_Raid_t *raidPtr = desc->raidPtr;
    580 	RF_DagHeader_t *dag_h;
    581 	int     i, j, done = RF_TRUE;
    582 	RF_DagList_t *dagList, *temp;
    583 	RF_Etimer_t timer;
    584 
    585 	/* check to see if this is the last dag */
    586 	dagList = desc->dagList;
    587 	for (i = 0; i < desc->numStripes; i++) {
    588 		if (dagList->numDags != dagList->numDagsDone)
    589 			done = RF_FALSE;
    590 		dagList = dagList->next;
    591 	}
    592 
    593 	if (done) {
    594 		if (desc->status) {
    595 			/* a dag failed, retry */
    596 			RF_ETIMER_START(timer);
    597 			/* free all dags */
    598 			dagList = desc->dagList;
    599 			for (i = 0; i < desc->numStripes; i++) {
    600 				rf_FreeDAG(dagList->dags);
    601 				temp = dagList;
    602 				dagList = dagList->next;
    603 			}
    604 			rf_MarkFailuresInASMList(raidPtr, asmh);
    605 			/* back up to rf_State_CreateDAG */
    606 			desc->state = desc->state - 2;
    607 			return RF_FALSE;
    608 		} else {
    609 			/* move on to rf_State_Cleanup */
    610 			desc->state++;
    611 		}
    612 		return RF_FALSE;
    613 	} else {
    614 		/* more dags to execute */
    615 		/* see if any are ready to be fired.  if so, fire them */
    616 		/* don't fire the initial dag in a list, it's fired in
    617 		 * rf_State_ExecuteDAG */
    618 		dagList = desc->dagList;
    619 		for (i = 0; i < desc->numStripes; i++) {
    620 			if ((dagList->numDagsDone < dagList->numDags)
    621 			    && (dagList->numDagsDone == dagList->numDagsFired)
    622 			    && (dagList->numDagsFired > 0)) {
    623 #if RF_ACC_TRACE > 0
    624 				RF_ETIMER_START(dagList->tracerec.timer);
    625 #endif
    626 				/* fire next dag in this stripe */
    627 				/* first, skip to next dag awaiting execution */
    628 				dag_h = dagList->dags;
    629 				for (j = 0; j < dagList->numDagsDone; j++)
    630 					dag_h = dag_h->next;
    631 				dagList->numDagsFired++;
    632 				rf_DispatchDAG(dag_h, (void (*) (void *)) rf_ContinueDagAccess,
    633 				    dagList);
    634 			}
    635 			dagList = dagList->next;
    636 		}
    637 		return RF_TRUE;
    638 	}
    639 }
    640 /* only make it this far if all dags complete successfully */
    641 int
    642 rf_State_Cleanup(RF_RaidAccessDesc_t *desc)
    643 {
    644 #if RF_ACC_TRACE > 0
    645 	RF_AccTraceEntry_t *tracerec = &desc->tracerec;
    646 	RF_Etimer_t timer;
    647 #endif
    648 	RF_AccessStripeMapHeader_t *asmh = desc->asmap;
    649 	RF_Raid_t *raidPtr = desc->raidPtr;
    650 	RF_AccessStripeMap_t *asm_p;
    651 	RF_DagList_t *dagList;
    652 	int i;
    653 
    654 	desc->state++;
    655 
    656 #if RF_ACC_TRACE > 0
    657 	timer = tracerec->timer;
    658 	RF_ETIMER_STOP(timer);
    659 	RF_ETIMER_EVAL(timer);
    660 	tracerec->specific.user.dag_retry_us = RF_ETIMER_VAL_US(timer);
    661 
    662 	/* the RAID I/O is complete.  Clean up. */
    663 	tracerec->specific.user.dag_retry_us = 0;
    664 
    665 	RF_ETIMER_START(timer);
    666 #endif
    667 	/* free all dags */
    668 	dagList = desc->dagList;
    669 	for (i = 0; i < desc->numStripes; i++) {
    670 		rf_FreeDAG(dagList->dags);
    671 		dagList = dagList->next;
    672 	}
    673 #if RF_ACC_TRACE > 0
    674 	RF_ETIMER_STOP(timer);
    675 	RF_ETIMER_EVAL(timer);
    676 	tracerec->specific.user.cleanup_us = RF_ETIMER_VAL_US(timer);
    677 
    678 	RF_ETIMER_START(timer);
    679 #endif
    680 	if (!(raidPtr->Layout.map->flags & RF_NO_STRIPE_LOCKS)) {
    681 		for (asm_p = asmh->stripeMap; asm_p; asm_p = asm_p->next) {
    682 			if (!rf_suppressLocksAndLargeWrites &&
    683 			    asm_p->parityInfo &&
    684 			    !(desc->flags & RF_DAG_SUPPRESS_LOCKS)) {
    685 				RF_ASSERT_VALID_LOCKREQ(&asm_p->lockReqDesc);
    686 				rf_ReleaseStripeLock(raidPtr->lockTable,
    687 						     asm_p->stripeID,
    688 						     &asm_p->lockReqDesc);
    689 			}
    690 			if (asm_p->flags & RF_ASM_FLAGS_RECON_BLOCKED) {
    691 				rf_UnblockRecon(raidPtr, asm_p);
    692 			}
    693 		}
    694 	}
    695 #if RF_ACC_TRACE > 0
    696 	RF_ETIMER_STOP(timer);
    697 	RF_ETIMER_EVAL(timer);
    698 	tracerec->specific.user.lock_us += RF_ETIMER_VAL_US(timer);
    699 
    700 	RF_ETIMER_START(timer);
    701 #endif
    702 	rf_FreeAccessStripeMap(asmh);
    703 #if RF_ACC_TRACE > 0
    704 	RF_ETIMER_STOP(timer);
    705 	RF_ETIMER_EVAL(timer);
    706 	tracerec->specific.user.cleanup_us += RF_ETIMER_VAL_US(timer);
    707 
    708 	RF_ETIMER_STOP(desc->timer);
    709 	RF_ETIMER_EVAL(desc->timer);
    710 
    711 	timer = desc->tracerec.tot_timer;
    712 	RF_ETIMER_STOP(timer);
    713 	RF_ETIMER_EVAL(timer);
    714 	desc->tracerec.total_us = RF_ETIMER_VAL_US(timer);
    715 
    716 	rf_LogTraceRec(raidPtr, tracerec);
    717 #endif
    718 	desc->flags |= RF_DAG_ACCESS_COMPLETE;
    719 
    720 	return RF_FALSE;
    721 }
    722