xref: /netbsd-src/sys/dev/raidframe/rf_map.c (revision 7330f729ccf0bd976a06f95fad452fe774fc7fd1)
1 /*	$NetBSD: rf_map.c,v 1.50 2019/05/28 08:59:35 msaitoh Exp $	*/
2 /*
3  * Copyright (c) 1995 Carnegie-Mellon University.
4  * All rights reserved.
5  *
6  * Author: Mark Holland
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@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  * map.c -- main code for mapping RAID addresses to physical disk addresses
32  *
33  **************************************************************************/
34 
35 #include <sys/cdefs.h>
36 __KERNEL_RCSID(0, "$NetBSD: rf_map.c,v 1.50 2019/05/28 08:59:35 msaitoh Exp $");
37 
38 #include <dev/raidframe/raidframevar.h>
39 
40 #include "rf_threadstuff.h"
41 #include "rf_raid.h"
42 #include "rf_general.h"
43 #include "rf_map.h"
44 #include "rf_shutdown.h"
45 
46 static void rf_FreePDAList(RF_PhysDiskAddr_t *pda_list);
47 static void rf_FreeASMList(RF_AccessStripeMap_t *asm_list);
48 
49 /***************************************************************************
50  *
51  * MapAccess -- main 1st order mapping routine.  Maps an access in the
52  * RAID address space to the corresponding set of physical disk
53  * addresses.  The result is returned as a list of AccessStripeMap
54  * structures, one per stripe accessed.  Each ASM structure contains a
55  * pointer to a list of PhysDiskAddr structures, which describe the
56  * physical locations touched by the user access.  Note that this
57  * routine returns only static mapping information, i.e. the list of
58  * physical addresses returned does not necessarily identify the set
59  * of physical locations that will actually be read or written.  The
60  * routine also maps the parity.  The physical disk location returned
61  * always indicates the entire parity unit, even when only a subset of
62  * it is being accessed.  This is because an access that is not stripe
63  * unit aligned but that spans a stripe unit boundary may require
64  * access two distinct portions of the parity unit, and we can't yet
65  * tell which portion(s) we'll actually need.  We leave it up to the
66  * algorithm selection code to decide what subset of the parity unit
67  * to access.  Note that addresses in the RAID address space must
68  * always be maintained as longs, instead of ints.
69  *
70  * This routine returns NULL if numBlocks is 0
71  *
72  * raidAddress - starting address in RAID address space
73  * numBlocks   - number of blocks in RAID address space to access
74  * buffer      - buffer to supply/receive data
75  * remap       - 1 => remap address to spare space
76  ***************************************************************************/
77 
78 RF_AccessStripeMapHeader_t *
79 rf_MapAccess(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddress,
80 	     RF_SectorCount_t numBlocks, void *buffer, int remap)
81 {
82 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
83 	RF_AccessStripeMapHeader_t *asm_hdr = NULL;
84 	RF_AccessStripeMap_t *asm_list = NULL, *asm_p = NULL;
85 	int     faultsTolerated = layoutPtr->map->faultsTolerated;
86 	/* we'll change raidAddress along the way */
87 	RF_RaidAddr_t startAddress = raidAddress;
88 	RF_RaidAddr_t endAddress = raidAddress + numBlocks;
89 	RF_RaidDisk_t *disks = raidPtr->Disks;
90 	RF_PhysDiskAddr_t *pda_p;
91 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
92 	RF_PhysDiskAddr_t *pda_q;
93 #endif
94 	RF_StripeCount_t numStripes = 0;
95 	RF_RaidAddr_t stripeRealEndAddress, stripeEndAddress,
96 		nextStripeUnitAddress;
97 	RF_RaidAddr_t startAddrWithinStripe, lastRaidAddr;
98 	RF_StripeCount_t totStripes;
99 	RF_StripeNum_t stripeID, lastSID, SUID, lastSUID;
100 	RF_AccessStripeMap_t *asmList, *t_asm;
101 	RF_PhysDiskAddr_t *pdaList, *t_pda;
102 
103 	/* allocate all the ASMs and PDAs up front */
104 	lastRaidAddr = raidAddress + numBlocks - 1;
105 	stripeID = rf_RaidAddressToStripeID(layoutPtr, raidAddress);
106 	lastSID = rf_RaidAddressToStripeID(layoutPtr, lastRaidAddr);
107 	totStripes = lastSID - stripeID + 1;
108 	SUID = rf_RaidAddressToStripeUnitID(layoutPtr, raidAddress);
109 	lastSUID = rf_RaidAddressToStripeUnitID(layoutPtr, lastRaidAddr);
110 
111 	asmList = rf_AllocASMList(totStripes);
112 
113 	/* may also need pda(s) per stripe for parity */
114 	pdaList = rf_AllocPDAList(lastSUID - SUID + 1 +
115 				  faultsTolerated * totStripes);
116 
117 
118 	if (raidAddress + numBlocks > raidPtr->totalSectors) {
119 		RF_ERRORMSG1("Unable to map access because offset (%d) was invalid\n",
120 		    (int) raidAddress);
121 		return (NULL);
122 	}
123 #if RF_DEBUG_MAP
124 	if (rf_mapDebug)
125 		rf_PrintRaidAddressInfo(raidPtr, raidAddress, numBlocks);
126 #endif
127 	for (; raidAddress < endAddress;) {
128 		/* make the next stripe structure */
129 		RF_ASSERT(asmList);
130 		t_asm = asmList;
131 		asmList = asmList->next;
132 		memset(t_asm, 0, sizeof(*t_asm));
133 		if (!asm_p)
134 			asm_list = asm_p = t_asm;
135 		else {
136 			asm_p->next = t_asm;
137 			asm_p = asm_p->next;
138 		}
139 		numStripes++;
140 
141 		/* map SUs from current location to the end of the stripe */
142 		asm_p->stripeID =	/* rf_RaidAddressToStripeID(layoutPtr,
143 		        raidAddress) */ stripeID++;
144 		stripeRealEndAddress = rf_RaidAddressOfNextStripeBoundary(layoutPtr, raidAddress);
145 		stripeEndAddress = RF_MIN(endAddress, stripeRealEndAddress);
146 		asm_p->raidAddress = raidAddress;
147 		asm_p->endRaidAddress = stripeEndAddress;
148 
149 		/* map each stripe unit in the stripe */
150 		pda_p = NULL;
151 
152 		/* Raid addr of start of portion of access that is
153                    within this stripe */
154 		startAddrWithinStripe = raidAddress;
155 
156 		for (; raidAddress < stripeEndAddress;) {
157 			RF_ASSERT(pdaList);
158 			t_pda = pdaList;
159 			pdaList = pdaList->next;
160 			memset(t_pda, 0, sizeof(*t_pda));
161 			if (!pda_p)
162 				asm_p->physInfo = pda_p = t_pda;
163 			else {
164 				pda_p->next = t_pda;
165 				pda_p = pda_p->next;
166 			}
167 
168 			pda_p->type = RF_PDA_TYPE_DATA;
169 			(layoutPtr->map->MapSector) (raidPtr, raidAddress,
170 						     &(pda_p->col),
171 						     &(pda_p->startSector),
172 						     remap);
173 
174 			/* mark any failures we find.  failedPDA is
175 			 * don't-care if there is more than one
176 			 * failure */
177 
178 			/* the RAID address corresponding to this
179                            physical diskaddress */
180 			pda_p->raidAddress = raidAddress;
181 			nextStripeUnitAddress = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, raidAddress);
182 			pda_p->numSector = RF_MIN(endAddress, nextStripeUnitAddress) - raidAddress;
183 			RF_ASSERT(pda_p->numSector != 0);
184 			rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 0);
185 			pda_p->bufPtr = (char *)buffer + rf_RaidAddressToByte(raidPtr, (raidAddress - startAddress));
186 			asm_p->totalSectorsAccessed += pda_p->numSector;
187 			asm_p->numStripeUnitsAccessed++;
188 
189 			raidAddress = RF_MIN(endAddress, nextStripeUnitAddress);
190 		}
191 
192 		/* Map the parity. At this stage, the startSector and
193 		 * numSector fields for the parity unit are always set
194 		 * to indicate the entire parity unit. We may modify
195 		 * this after mapping the data portion. */
196 		switch (faultsTolerated) {
197 		case 0:
198 			break;
199 		case 1:	/* single fault tolerant */
200 			RF_ASSERT(pdaList);
201 			t_pda = pdaList;
202 			pdaList = pdaList->next;
203 			memset(t_pda, 0, sizeof(*t_pda));
204 			pda_p = asm_p->parityInfo = t_pda;
205 			pda_p->type = RF_PDA_TYPE_PARITY;
206 			(layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
207 			    &(pda_p->col), &(pda_p->startSector), remap);
208 			pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
209 			/* raidAddr may be needed to find unit to redirect to */
210 			pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
211 			rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
212 			rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
213 
214 			break;
215 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
216 		case 2:	/* two fault tolerant */
217 			RF_ASSERT(pdaList && pdaList->next);
218 			t_pda = pdaList;
219 			pdaList = pdaList->next;
220 			memset(t_pda, 0, sizeof(*t_pda));
221 			pda_p = asm_p->parityInfo = t_pda;
222 			pda_p->type = RF_PDA_TYPE_PARITY;
223 			t_pda = pdaList;
224 			pdaList = pdaList->next;
225 			memset(t_pda, 0, sizeof(*t_pda));
226 			pda_q = asm_p->qInfo = t_pda;
227 			pda_q->type = RF_PDA_TYPE_Q;
228 			(layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
229 			    &(pda_p->col), &(pda_p->startSector), remap);
230 			(layoutPtr->map->MapQ) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
231 			    &(pda_q->col), &(pda_q->startSector), remap);
232 			pda_q->numSector = pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
233 			/* raidAddr may be needed to find unit to redirect to */
234 			pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
235 			pda_q->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
236 			/* failure mode stuff */
237 			rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
238 			rf_ASMCheckStatus(raidPtr, pda_q, asm_p, disks, 1);
239 			rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
240 			rf_ASMParityAdjust(asm_p->qInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
241 			break;
242 #endif
243 		}
244 	}
245 	RF_ASSERT(asmList == NULL && pdaList == NULL);
246 	/* make the header structure */
247 	asm_hdr = rf_AllocAccessStripeMapHeader();
248 	RF_ASSERT(numStripes == totStripes);
249 	asm_hdr->numStripes = numStripes;
250 	asm_hdr->stripeMap = asm_list;
251 
252 #if RF_DEBUG_MAP
253 	if (rf_mapDebug)
254 		rf_PrintAccessStripeMap(asm_hdr);
255 #endif
256 	return (asm_hdr);
257 }
258 
259 /***************************************************************************
260  * This routine walks through an ASM list and marks the PDAs that have
261  * failed.  It's called only when a disk failure causes an in-flight
262  * DAG to fail.  The parity may consist of two components, but we want
263  * to use only one failedPDA pointer.  Thus we set failedPDA to point
264  * to the first parity component, and rely on the rest of the code to
265  * do the right thing with this.
266  ***************************************************************************/
267 
268 void
269 rf_MarkFailuresInASMList(RF_Raid_t *raidPtr,
270 			 RF_AccessStripeMapHeader_t *asm_h)
271 {
272 	RF_RaidDisk_t *disks = raidPtr->Disks;
273 	RF_AccessStripeMap_t *asmap;
274 	RF_PhysDiskAddr_t *pda;
275 
276 	for (asmap = asm_h->stripeMap; asmap; asmap = asmap->next) {
277 		asmap->numDataFailed = 0;
278 		asmap->numParityFailed = 0;
279 		asmap->numQFailed = 0;
280 		asmap->numFailedPDAs = 0;
281 		memset(asmap->failedPDAs, 0,
282 		    RF_MAX_FAILED_PDA * sizeof(*asmap->failedPDAs));
283 		for (pda = asmap->physInfo; pda; pda = pda->next) {
284 			if (RF_DEAD_DISK(disks[pda->col].status)) {
285 				asmap->numDataFailed++;
286 				asmap->failedPDAs[asmap->numFailedPDAs] = pda;
287 				asmap->numFailedPDAs++;
288 			}
289 		}
290 		pda = asmap->parityInfo;
291 		if (pda && RF_DEAD_DISK(disks[pda->col].status)) {
292 			asmap->numParityFailed++;
293 			asmap->failedPDAs[asmap->numFailedPDAs] = pda;
294 			asmap->numFailedPDAs++;
295 		}
296 		pda = asmap->qInfo;
297 		if (pda && RF_DEAD_DISK(disks[pda->col].status)) {
298 			asmap->numQFailed++;
299 			asmap->failedPDAs[asmap->numFailedPDAs] = pda;
300 			asmap->numFailedPDAs++;
301 		}
302 	}
303 }
304 
305 /***************************************************************************
306  *
307  * routines to allocate and free list elements.  All allocation
308  * routines zero the structure before returning it.
309  *
310  * FreePhysDiskAddr is static.  It should never be called directly,
311  * because FreeAccessStripeMap takes care of freeing the PhysDiskAddr
312  * list.
313  *
314  ***************************************************************************/
315 
316 #define RF_MAX_FREE_ASMHDR 128
317 #define RF_MIN_FREE_ASMHDR  32
318 
319 #define RF_MAX_FREE_ASM 192
320 #define RF_MIN_FREE_ASM  64
321 
322 #define RF_MAX_FREE_PDA 192
323 #define RF_MIN_FREE_PDA  64
324 
325 #define RF_MAX_FREE_ASMHLE 64
326 #define RF_MIN_FREE_ASMHLE 16
327 
328 #define RF_MAX_FREE_FSS 128
329 #define RF_MIN_FREE_FSS  32
330 
331 #define RF_MAX_FREE_VFPLE 128
332 #define RF_MIN_FREE_VFPLE  32
333 
334 #define RF_MAX_FREE_VPLE 128
335 #define RF_MIN_FREE_VPLE  32
336 
337 
338 /* called at shutdown time.  So far, all that is necessary is to
339    release all the free lists */
340 static void rf_ShutdownMapModule(void *);
341 static void
342 rf_ShutdownMapModule(void *ignored)
343 {
344 	pool_destroy(&rf_pools.asm_hdr);
345 	pool_destroy(&rf_pools.asmap);
346 	pool_destroy(&rf_pools.asmhle);
347 	pool_destroy(&rf_pools.pda);
348 	pool_destroy(&rf_pools.fss);
349 	pool_destroy(&rf_pools.vfple);
350 	pool_destroy(&rf_pools.vple);
351 }
352 
353 int
354 rf_ConfigureMapModule(RF_ShutdownList_t **listp)
355 {
356 
357 	rf_pool_init(&rf_pools.asm_hdr, sizeof(RF_AccessStripeMapHeader_t),
358 		     "rf_asmhdr_pl", RF_MIN_FREE_ASMHDR, RF_MAX_FREE_ASMHDR);
359 	rf_pool_init(&rf_pools.asmap, sizeof(RF_AccessStripeMap_t),
360 		     "rf_asm_pl", RF_MIN_FREE_ASM, RF_MAX_FREE_ASM);
361 	rf_pool_init(&rf_pools.asmhle, sizeof(RF_ASMHeaderListElem_t),
362 		     "rf_asmhle_pl", RF_MIN_FREE_ASMHLE, RF_MAX_FREE_ASMHLE);
363 	rf_pool_init(&rf_pools.pda, sizeof(RF_PhysDiskAddr_t),
364 		     "rf_pda_pl", RF_MIN_FREE_PDA, RF_MAX_FREE_PDA);
365 	rf_pool_init(&rf_pools.fss, sizeof(RF_FailedStripe_t),
366 		     "rf_fss_pl", RF_MIN_FREE_FSS, RF_MAX_FREE_FSS);
367 	rf_pool_init(&rf_pools.vfple, sizeof(RF_VoidFunctionPointerListElem_t),
368 		     "rf_vfple_pl", RF_MIN_FREE_VFPLE, RF_MAX_FREE_VFPLE);
369 	rf_pool_init(&rf_pools.vple, sizeof(RF_VoidPointerListElem_t),
370 		     "rf_vple_pl", RF_MIN_FREE_VPLE, RF_MAX_FREE_VPLE);
371 	rf_ShutdownCreate(listp, rf_ShutdownMapModule, NULL);
372 
373 	return (0);
374 }
375 
376 RF_AccessStripeMapHeader_t *
377 rf_AllocAccessStripeMapHeader(void)
378 {
379 	return pool_get(&rf_pools.asm_hdr, PR_WAITOK | PR_ZERO);
380 }
381 
382 void
383 rf_FreeAccessStripeMapHeader(RF_AccessStripeMapHeader_t *p)
384 {
385 	pool_put(&rf_pools.asm_hdr, p);
386 }
387 
388 
389 RF_VoidFunctionPointerListElem_t *
390 rf_AllocVFPListElem(void)
391 {
392 	return pool_get(&rf_pools.vfple, PR_WAITOK | PR_ZERO);
393 }
394 
395 void
396 rf_FreeVFPListElem(RF_VoidFunctionPointerListElem_t *p)
397 {
398 
399 	pool_put(&rf_pools.vfple, p);
400 }
401 
402 
403 RF_VoidPointerListElem_t *
404 rf_AllocVPListElem(void)
405 {
406 	return pool_get(&rf_pools.vple, PR_WAITOK | PR_ZERO);
407 }
408 
409 void
410 rf_FreeVPListElem(RF_VoidPointerListElem_t *p)
411 {
412 
413 	pool_put(&rf_pools.vple, p);
414 }
415 
416 RF_ASMHeaderListElem_t *
417 rf_AllocASMHeaderListElem(void)
418 {
419 	return pool_get(&rf_pools.asmhle, PR_WAITOK | PR_ZERO);
420 }
421 
422 void
423 rf_FreeASMHeaderListElem(RF_ASMHeaderListElem_t *p)
424 {
425 
426 	pool_put(&rf_pools.asmhle, p);
427 }
428 
429 RF_FailedStripe_t *
430 rf_AllocFailedStripeStruct(void)
431 {
432 	return pool_get(&rf_pools.fss, PR_WAITOK | PR_ZERO);
433 }
434 
435 void
436 rf_FreeFailedStripeStruct(RF_FailedStripe_t *p)
437 {
438 	pool_put(&rf_pools.fss, p);
439 }
440 
441 
442 
443 
444 
445 RF_PhysDiskAddr_t *
446 rf_AllocPhysDiskAddr(void)
447 {
448 	return pool_get(&rf_pools.pda, PR_WAITOK | PR_ZERO);
449 }
450 /* allocates a list of PDAs, locking the free list only once when we
451  * have to call calloc, we do it one component at a time to simplify
452  * the process of freeing the list at program shutdown.  This should
453  * not be much of a performance hit, because it should be very
454  * infrequently executed.  */
455 RF_PhysDiskAddr_t *
456 rf_AllocPDAList(int count)
457 {
458 	RF_PhysDiskAddr_t *p, *prev;
459 	int i;
460 
461 	p = NULL;
462 	prev = NULL;
463 	for (i = 0; i < count; i++) {
464 		p = pool_get(&rf_pools.pda, PR_WAITOK);
465 		p->next = prev;
466 		prev = p;
467 	}
468 
469 	return (p);
470 }
471 
472 void
473 rf_FreePhysDiskAddr(RF_PhysDiskAddr_t *p)
474 {
475 	pool_put(&rf_pools.pda, p);
476 }
477 
478 static void
479 rf_FreePDAList(RF_PhysDiskAddr_t *pda_list)
480 {
481 	RF_PhysDiskAddr_t *p, *tmp;
482 
483 	p=pda_list;
484 	while (p) {
485 		tmp = p->next;
486 		pool_put(&rf_pools.pda, p);
487 		p = tmp;
488 	}
489 }
490 
491 /* this is essentially identical to AllocPDAList.  I should combine
492  * the two.  when we have to call calloc, we do it one component at a
493  * time to simplify the process of freeing the list at program
494  * shutdown.  This should not be much of a performance hit, because it
495  * should be very infrequently executed.  */
496 RF_AccessStripeMap_t *
497 rf_AllocASMList(int count)
498 {
499 	RF_AccessStripeMap_t *p, *prev;
500 	int i;
501 
502 	p = NULL;
503 	prev = NULL;
504 	for (i = 0; i < count; i++) {
505 		p = pool_get(&rf_pools.asmap, PR_WAITOK);
506 		p->next = prev;
507 		prev = p;
508 	}
509 	return (p);
510 }
511 
512 static void
513 rf_FreeASMList(RF_AccessStripeMap_t *asm_list)
514 {
515 	RF_AccessStripeMap_t *p, *tmp;
516 
517 	p=asm_list;
518 	while (p) {
519 		tmp = p->next;
520 		pool_put(&rf_pools.asmap, p);
521 		p = tmp;
522 	}
523 }
524 
525 void
526 rf_FreeAccessStripeMap(RF_AccessStripeMapHeader_t *hdr)
527 {
528 	RF_AccessStripeMap_t *p;
529 	RF_PhysDiskAddr_t *pdp, *trailer, *pdaList = NULL, *pdaEnd = NULL;
530 	int     count = 0, t;
531 
532 	for (p = hdr->stripeMap; p; p = p->next) {
533 
534 		/* link the 3 pda lists into the accumulating pda list */
535 
536 		if (!pdaList)
537 			pdaList = p->qInfo;
538 		else
539 			pdaEnd->next = p->qInfo;
540 		for (trailer = NULL, pdp = p->qInfo; pdp;) {
541 			trailer = pdp;
542 			pdp = pdp->next;
543 			count++;
544 		}
545 		if (trailer)
546 			pdaEnd = trailer;
547 
548 		if (!pdaList)
549 			pdaList = p->parityInfo;
550 		else
551 			pdaEnd->next = p->parityInfo;
552 		for (trailer = NULL, pdp = p->parityInfo; pdp;) {
553 			trailer = pdp;
554 			pdp = pdp->next;
555 			count++;
556 		}
557 		if (trailer)
558 			pdaEnd = trailer;
559 
560 		if (!pdaList)
561 			pdaList = p->physInfo;
562 		else
563 			pdaEnd->next = p->physInfo;
564 		for (trailer = NULL, pdp = p->physInfo; pdp;) {
565 			trailer = pdp;
566 			pdp = pdp->next;
567 			count++;
568 		}
569 		if (trailer)
570 			pdaEnd = trailer;
571 	}
572 
573 	/* debug only */
574 	for (t = 0, pdp = pdaList; pdp; pdp = pdp->next)
575 		t++;
576 	RF_ASSERT(t == count);
577 
578 	if (pdaList)
579 		rf_FreePDAList(pdaList);
580 	rf_FreeASMList(hdr->stripeMap);
581 	rf_FreeAccessStripeMapHeader(hdr);
582 }
583 /* We can't use the large write optimization if there are any failures
584  * in the stripe.  In the declustered layout, there is no way to
585  * immediately determine what disks constitute a stripe, so we
586  * actually have to hunt through the stripe looking for failures.  The
587  * reason we map the parity instead of just using asm->parityInfo->col
588  * is because the latter may have been already redirected to a spare
589  * drive, which would mess up the computation of the stripe offset.
590  *
591  * ASSUMES AT MOST ONE FAILURE IN THE STRIPE.  */
592 int
593 rf_CheckStripeForFailures(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
594 {
595 	RF_RowCol_t tcol, pcol, *diskids, i;
596 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
597 	RF_StripeCount_t stripeOffset;
598 	int     numFailures;
599 	RF_RaidAddr_t sosAddr;
600 	RF_SectorNum_t diskOffset, poffset;
601 
602 	/* quick out in the fault-free case.  */
603 	rf_lock_mutex2(raidPtr->mutex);
604 	numFailures = raidPtr->numFailures;
605 	rf_unlock_mutex2(raidPtr->mutex);
606 	if (numFailures == 0)
607 		return (0);
608 
609 	sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
610 						     asmap->raidAddress);
611 	(layoutPtr->map->IdentifyStripe) (raidPtr, asmap->raidAddress,
612 					  &diskids);
613 	(layoutPtr->map->MapParity) (raidPtr, asmap->raidAddress,
614 				     &pcol, &poffset, 0);	/* get pcol */
615 
616 	/* this need not be true if we've redirected the access to a
617 	 * spare in another row RF_ASSERT(row == testrow); */
618 	stripeOffset = 0;
619 	for (i = 0; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++) {
620 		if (diskids[i] != pcol) {
621 			if (RF_DEAD_DISK(raidPtr->Disks[diskids[i]].status)) {
622 				if (raidPtr->status != rf_rs_reconstructing)
623 					return (1);
624 				RF_ASSERT(raidPtr->reconControl->fcol == diskids[i]);
625 				layoutPtr->map->MapSector(raidPtr,
626 				    sosAddr + stripeOffset * layoutPtr->sectorsPerStripeUnit,
627 				    &tcol, &diskOffset, 0);
628 				RF_ASSERT(tcol == diskids[i]);
629 				if (!rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, diskOffset))
630 					return (1);
631 				asmap->flags |= RF_ASM_REDIR_LARGE_WRITE;
632 				return (0);
633 			}
634 			stripeOffset++;
635 		}
636 	}
637 	return (0);
638 }
639 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) || (RF_INCLUDE_EVENODD >0)
640 /*
641    return the number of failed data units in the stripe.
642 */
643 
644 int
645 rf_NumFailedDataUnitsInStripe(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
646 {
647 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
648 	RF_RowCol_t tcol, i;
649 	RF_SectorNum_t diskOffset;
650 	RF_RaidAddr_t sosAddr;
651 	int     numFailures;
652 
653 	/* quick out in the fault-free case.  */
654 	rf_lock_mutex2(raidPtr->mutex);
655 	numFailures = raidPtr->numFailures;
656 	rf_unlock_mutex2(raidPtr->mutex);
657 	if (numFailures == 0)
658 		return (0);
659 	numFailures = 0;
660 
661 	sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
662 						     asmap->raidAddress);
663 	for (i = 0; i < layoutPtr->numDataCol; i++) {
664 		(layoutPtr->map->MapSector) (raidPtr, sosAddr + i * layoutPtr->sectorsPerStripeUnit,
665 		    &tcol, &diskOffset, 0);
666 		if (RF_DEAD_DISK(raidPtr->Disks[tcol].status))
667 			numFailures++;
668 	}
669 
670 	return numFailures;
671 }
672 #endif
673 
674 /****************************************************************************
675  *
676  * debug routines
677  *
678  ***************************************************************************/
679 #if RF_DEBUG_MAP
680 void
681 rf_PrintAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h)
682 {
683 	rf_PrintFullAccessStripeMap(asm_h, 0);
684 }
685 #endif
686 
687 /* prbuf - flag to print buffer pointers */
688 void
689 rf_PrintFullAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h, int prbuf)
690 {
691 	int     i;
692 	RF_AccessStripeMap_t *asmap = asm_h->stripeMap;
693 	RF_PhysDiskAddr_t *p;
694 	printf("%d stripes total\n", (int) asm_h->numStripes);
695 	for (; asmap; asmap = asmap->next) {
696 		/* printf("Num failures: %d\n",asmap->numDataFailed); */
697 		/* printf("Num sectors:
698 		 * %d\n",(int)asmap->totalSectorsAccessed); */
699 		printf("Stripe %d (%d sectors), failures: %d data, %d parity: ",
700 		    (int) asmap->stripeID,
701 		    (int) asmap->totalSectorsAccessed,
702 		    (int) asmap->numDataFailed,
703 		    (int) asmap->numParityFailed);
704 		if (asmap->parityInfo) {
705 			printf("Parity [c%d s%d-%d", asmap->parityInfo->col,
706 			    (int) asmap->parityInfo->startSector,
707 			    (int) (asmap->parityInfo->startSector +
708 				asmap->parityInfo->numSector - 1));
709 			if (prbuf)
710 				printf(" b0x%lx", (unsigned long) asmap->parityInfo->bufPtr);
711 			if (asmap->parityInfo->next) {
712 				printf(", c%d s%d-%d", asmap->parityInfo->next->col,
713 				    (int) asmap->parityInfo->next->startSector,
714 				    (int) (asmap->parityInfo->next->startSector +
715 					asmap->parityInfo->next->numSector - 1));
716 				if (prbuf)
717 					printf(" b0x%lx", (unsigned long) asmap->parityInfo->next->bufPtr);
718 				RF_ASSERT(asmap->parityInfo->next->next == NULL);
719 			}
720 			printf("]\n\t");
721 		}
722 		for (i = 0, p = asmap->physInfo; p; p = p->next, i++) {
723 			printf("SU c%d s%d-%d ", p->col, (int) p->startSector,
724 			    (int) (p->startSector + p->numSector - 1));
725 			if (prbuf)
726 				printf("b0x%lx ", (unsigned long) p->bufPtr);
727 			if (i && !(i & 1))
728 				printf("\n\t");
729 		}
730 		printf("\n");
731 		p = asm_h->stripeMap->failedPDAs[0];
732 		if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 1)
733 			printf("[multiple failures]\n");
734 		else
735 			if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 0)
736 				printf("\t[Failed PDA: c%d s%d-%d]\n", p->col,
737 				    (int) p->startSector, (int) (p->startSector + p->numSector - 1));
738 	}
739 }
740 
741 #if RF_MAP_DEBUG
742 void
743 rf_PrintRaidAddressInfo(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
744 			RF_SectorCount_t numBlocks)
745 {
746 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
747 	RF_RaidAddr_t ra, sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
748 
749 	printf("Raid addrs of SU boundaries from start of stripe to end of access:\n\t");
750 	for (ra = sosAddr; ra <= raidAddr + numBlocks; ra += layoutPtr->sectorsPerStripeUnit) {
751 		printf("%d (0x%x), ", (int) ra, (int) ra);
752 	}
753 	printf("\n");
754 	printf("Offset into stripe unit: %d (0x%x)\n",
755 	    (int) (raidAddr % layoutPtr->sectorsPerStripeUnit),
756 	    (int) (raidAddr % layoutPtr->sectorsPerStripeUnit));
757 }
758 #endif
759 /* given a parity descriptor and the starting address within a stripe,
760  * range restrict the parity descriptor to touch only the correct
761  * stuff.  */
762 void
763 rf_ASMParityAdjust(RF_PhysDiskAddr_t *toAdjust,
764 		   RF_StripeNum_t startAddrWithinStripe,
765 		   RF_SectorNum_t endAddress,
766 		   RF_RaidLayout_t *layoutPtr,
767 		   RF_AccessStripeMap_t *asm_p)
768 {
769 	RF_PhysDiskAddr_t *new_pda;
770 
771 	/* when we're accessing only a portion of one stripe unit, we
772 	 * want the parity descriptor to identify only the chunk of
773 	 * parity associated with the data.  When the access spans
774 	 * exactly one stripe unit boundary and is less than a stripe
775 	 * unit in size, it uses two disjoint regions of the parity
776 	 * unit.  When an access spans more than one stripe unit
777 	 * boundary, it uses all of the parity unit.
778 	 *
779 	 * To better handle the case where stripe units are small, we
780 	 * may eventually want to change the 2nd case so that if the
781 	 * SU size is below some threshold, we just read/write the
782 	 * whole thing instead of breaking it up into two accesses. */
783 	if (asm_p->numStripeUnitsAccessed == 1) {
784 		int     x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
785 		toAdjust->startSector += x;
786 		toAdjust->raidAddress += x;
787 		toAdjust->numSector = asm_p->physInfo->numSector;
788 		RF_ASSERT(toAdjust->numSector != 0);
789 	} else
790 		if (asm_p->numStripeUnitsAccessed == 2 && asm_p->totalSectorsAccessed < layoutPtr->sectorsPerStripeUnit) {
791 			int     x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
792 
793 			/* create a second pda and copy the parity map info
794 			 * into it */
795 			RF_ASSERT(toAdjust->next == NULL);
796 			/* the following will get freed in rf_FreeAccessStripeMap() via
797 			   rf_FreePDAList() */
798 			new_pda = toAdjust->next = rf_AllocPhysDiskAddr();
799 			*new_pda = *toAdjust;	/* structure assignment */
800 			new_pda->next = NULL;
801 
802 			/* adjust the start sector & number of blocks for the
803 			 * first parity pda */
804 			toAdjust->startSector += x;
805 			toAdjust->raidAddress += x;
806 			toAdjust->numSector = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, startAddrWithinStripe) - startAddrWithinStripe;
807 			RF_ASSERT(toAdjust->numSector != 0);
808 
809 			/* adjust the second pda */
810 			new_pda->numSector = endAddress - rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, endAddress);
811 			/* new_pda->raidAddress =
812 			 * rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr,
813 			 * toAdjust->raidAddress); */
814 			RF_ASSERT(new_pda->numSector != 0);
815 		}
816 }
817 
818 /* Check if a disk has been spared or failed. If spared, redirect the
819  * I/O.  If it has been failed, record it in the asm pointer.  Fifth
820  * arg is whether data or parity.  */
821 void
822 rf_ASMCheckStatus(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda_p,
823 		  RF_AccessStripeMap_t *asm_p, RF_RaidDisk_t *disks,
824 		  int parity)
825 {
826 	RF_DiskStatus_t dstatus;
827 	RF_RowCol_t fcol;
828 
829 	dstatus = disks[pda_p->col].status;
830 
831 	if (dstatus == rf_ds_spared) {
832 		/* if the disk has been spared, redirect access to the spare */
833 		fcol = pda_p->col;
834 		pda_p->col = disks[fcol].spareCol;
835 	} else
836 		if (dstatus == rf_ds_dist_spared) {
837 			/* ditto if disk has been spared to dist spare space */
838 #if RF_DEBUG_MAP
839 			RF_RowCol_t oc = pda_p->col;
840 			RF_SectorNum_t oo = pda_p->startSector;
841 #endif
842 			if (pda_p->type == RF_PDA_TYPE_DATA)
843 				raidPtr->Layout.map->MapSector(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
844 			else
845 				raidPtr->Layout.map->MapParity(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
846 
847 #if RF_DEBUG_MAP
848 			if (rf_mapDebug) {
849 				printf("Redirected c %d o %d -> c %d o %d\n", oc, (int) oo,
850 				    pda_p->col, (int) pda_p->startSector);
851 			}
852 #endif
853 		} else
854 			if (RF_DEAD_DISK(dstatus)) {
855 				/* if the disk is inaccessible, mark the
856 				 * failure */
857 				if (parity)
858 					asm_p->numParityFailed++;
859 				else {
860 					asm_p->numDataFailed++;
861 				}
862 				asm_p->failedPDAs[asm_p->numFailedPDAs] = pda_p;
863 				asm_p->numFailedPDAs++;
864 #if 0
865 				switch (asm_p->numParityFailed + asm_p->numDataFailed) {
866 				case 1:
867 					asm_p->failedPDAs[0] = pda_p;
868 					break;
869 				case 2:
870 					asm_p->failedPDAs[1] = pda_p;
871 				default:
872 					break;
873 				}
874 #endif
875 			}
876 	/* the redirected access should never span a stripe unit boundary */
877 	RF_ASSERT(rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress) ==
878 	    rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress + pda_p->numSector - 1));
879 	RF_ASSERT(pda_p->col != -1);
880 }
881