xref: /netbsd-src/sys/dev/raidframe/rf_map.c (revision 08c81a9c2dc8c7300e893321eb65c0925d60871c)
1 /*	$NetBSD: rf_map.c,v 1.14 2002/09/14 17:53:59 oster 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 1.14 2002/09/14 17:53:59 oster 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_freelist.h"
45 #include "rf_shutdown.h"
46 
47 static void rf_FreePDAList(RF_PhysDiskAddr_t * start, RF_PhysDiskAddr_t * end, int count);
48 static void
49 rf_FreeASMList(RF_AccessStripeMap_t * start, RF_AccessStripeMap_t * end,
50     int count);
51 
52 /*****************************************************************************************
53  *
54  * MapAccess -- main 1st order mapping routine.
55  *
56  * Maps an access in the RAID address space to the corresponding set of physical disk
57  * addresses.  The result is returned as a list of AccessStripeMap structures, one per
58  * stripe accessed.  Each ASM structure contains a pointer to a list of PhysDiskAddr
59  * structures, which describe the physical locations touched by the user access.  Note
60  * that this routine returns only static mapping information, i.e. the list of physical
61  * addresses returned does not necessarily identify the set of physical locations that
62  * will actually be read or written.
63  *
64  * The routine also maps the parity.  The physical disk location returned always
65  * indicates the entire parity unit, even when only a subset of it is being accessed.
66  * This is because an access that is not stripe unit aligned but that spans a stripe
67  * unit boundary may require access two distinct portions of the parity unit, and we
68  * can't yet tell which portion(s) we'll actually need.  We leave it up to the algorithm
69  * selection code to decide what subset of the parity unit to access.
70  *
71  * Note that addresses in the RAID address space must always be maintained as
72  * longs, instead of ints.
73  *
74  * This routine returns NULL if numBlocks is 0
75  *
76  ****************************************************************************************/
77 
78 RF_AccessStripeMapHeader_t *
79 rf_MapAccess(raidPtr, raidAddress, numBlocks, buffer, remap)
80 	RF_Raid_t *raidPtr;
81 	RF_RaidAddr_t raidAddress;	/* starting address in RAID address
82 					 * space */
83 	RF_SectorCount_t numBlocks;	/* number of blocks in RAID address
84 					 * space to access */
85 	caddr_t buffer;		/* buffer to supply/receive data */
86 	int     remap;		/* 1 => remap addresses to spare space */
87 {
88 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
89 	RF_AccessStripeMapHeader_t *asm_hdr = NULL;
90 	RF_AccessStripeMap_t *asm_list = NULL, *asm_p = NULL;
91 	int     faultsTolerated = layoutPtr->map->faultsTolerated;
92 	RF_RaidAddr_t startAddress = raidAddress;	/* we'll change
93 							 * raidAddress along the
94 							 * way */
95 	RF_RaidAddr_t endAddress = raidAddress + numBlocks;
96 	RF_RaidDisk_t **disks = raidPtr->Disks;
97 
98 	RF_PhysDiskAddr_t *pda_p, *pda_q;
99 	RF_StripeCount_t numStripes = 0;
100 	RF_RaidAddr_t stripeRealEndAddress, stripeEndAddress, nextStripeUnitAddress;
101 	RF_RaidAddr_t startAddrWithinStripe, lastRaidAddr;
102 	RF_StripeCount_t totStripes;
103 	RF_StripeNum_t stripeID, lastSID, SUID, lastSUID;
104 	RF_AccessStripeMap_t *asmList, *t_asm;
105 	RF_PhysDiskAddr_t *pdaList, *t_pda;
106 
107 	/* allocate all the ASMs and PDAs up front */
108 	lastRaidAddr = raidAddress + numBlocks - 1;
109 	stripeID = rf_RaidAddressToStripeID(layoutPtr, raidAddress);
110 	lastSID = rf_RaidAddressToStripeID(layoutPtr, lastRaidAddr);
111 	totStripes = lastSID - stripeID + 1;
112 	SUID = rf_RaidAddressToStripeUnitID(layoutPtr, raidAddress);
113 	lastSUID = rf_RaidAddressToStripeUnitID(layoutPtr, lastRaidAddr);
114 
115 	asmList = rf_AllocASMList(totStripes);
116 	pdaList = rf_AllocPDAList(lastSUID - SUID + 1 + faultsTolerated * totStripes);	/* may also need pda(s)
117 											 * per stripe for parity */
118 
119 	if (raidAddress + numBlocks > raidPtr->totalSectors) {
120 		RF_ERRORMSG1("Unable to map access because offset (%d) was invalid\n",
121 		    (int) raidAddress);
122 		return (NULL);
123 	}
124 	if (rf_mapDebug)
125 		rf_PrintRaidAddressInfo(raidPtr, raidAddress, numBlocks);
126 	for (; raidAddress < endAddress;) {
127 		/* make the next stripe structure */
128 		RF_ASSERT(asmList);
129 		t_asm = asmList;
130 		asmList = asmList->next;
131 		memset((char *) t_asm, 0, sizeof(RF_AccessStripeMap_t));
132 		if (!asm_p)
133 			asm_list = asm_p = t_asm;
134 		else {
135 			asm_p->next = t_asm;
136 			asm_p = asm_p->next;
137 		}
138 		numStripes++;
139 
140 		/* map SUs from current location to the end of the stripe */
141 		asm_p->stripeID =	/* rf_RaidAddressToStripeID(layoutPtr,
142 		        raidAddress) */ stripeID++;
143 		stripeRealEndAddress = rf_RaidAddressOfNextStripeBoundary(layoutPtr, raidAddress);
144 		stripeEndAddress = RF_MIN(endAddress, stripeRealEndAddress);
145 		asm_p->raidAddress = raidAddress;
146 		asm_p->endRaidAddress = stripeEndAddress;
147 
148 		/* map each stripe unit in the stripe */
149 		pda_p = NULL;
150 		startAddrWithinStripe = raidAddress;	/* Raid addr of start of
151 							 * portion of access
152 							 * that is within this
153 							 * stripe */
154 		for (; raidAddress < stripeEndAddress;) {
155 			RF_ASSERT(pdaList);
156 			t_pda = pdaList;
157 			pdaList = pdaList->next;
158 			memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
159 			if (!pda_p)
160 				asm_p->physInfo = pda_p = t_pda;
161 			else {
162 				pda_p->next = t_pda;
163 				pda_p = pda_p->next;
164 			}
165 
166 			pda_p->type = RF_PDA_TYPE_DATA;
167 			(layoutPtr->map->MapSector) (raidPtr, raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
168 
169 			/* mark any failures we find.  failedPDA is don't-care
170 			 * if there is more than one failure */
171 			pda_p->raidAddress = raidAddress;	/* the RAID address
172 								 * corresponding to this
173 								 * physical disk address */
174 			nextStripeUnitAddress = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, raidAddress);
175 			pda_p->numSector = RF_MIN(endAddress, nextStripeUnitAddress) - raidAddress;
176 			RF_ASSERT(pda_p->numSector != 0);
177 			rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 0);
178 			pda_p->bufPtr = buffer + rf_RaidAddressToByte(raidPtr, (raidAddress - startAddress));
179 			asm_p->totalSectorsAccessed += pda_p->numSector;
180 			asm_p->numStripeUnitsAccessed++;
181 			asm_p->origRow = pda_p->row;	/* redundant but
182 							 * harmless to do this
183 							 * in every loop
184 							 * iteration */
185 
186 			raidAddress = RF_MIN(endAddress, nextStripeUnitAddress);
187 		}
188 
189 		/* Map the parity. At this stage, the startSector and
190 		 * numSector fields for the parity unit are always set to
191 		 * indicate the entire parity unit. We may modify this after
192 		 * mapping the data portion. */
193 		switch (faultsTolerated) {
194 		case 0:
195 			break;
196 		case 1:	/* single fault tolerant */
197 			RF_ASSERT(pdaList);
198 			t_pda = pdaList;
199 			pdaList = pdaList->next;
200 			memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
201 			pda_p = asm_p->parityInfo = t_pda;
202 			pda_p->type = RF_PDA_TYPE_PARITY;
203 			(layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
204 			    &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
205 			pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
206 			/* raidAddr may be needed to find unit to redirect to */
207 			pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
208 			rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
209 			rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
210 
211 			break;
212 		case 2:	/* two fault tolerant */
213 			RF_ASSERT(pdaList && pdaList->next);
214 			t_pda = pdaList;
215 			pdaList = pdaList->next;
216 			memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
217 			pda_p = asm_p->parityInfo = t_pda;
218 			pda_p->type = RF_PDA_TYPE_PARITY;
219 			t_pda = pdaList;
220 			pdaList = pdaList->next;
221 			memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
222 			pda_q = asm_p->qInfo = t_pda;
223 			pda_q->type = RF_PDA_TYPE_Q;
224 			(layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
225 			    &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
226 			(layoutPtr->map->MapQ) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
227 			    &(pda_q->row), &(pda_q->col), &(pda_q->startSector), remap);
228 			pda_q->numSector = pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
229 			/* raidAddr may be needed to find unit to redirect to */
230 			pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
231 			pda_q->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
232 			/* failure mode stuff */
233 			rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
234 			rf_ASMCheckStatus(raidPtr, pda_q, asm_p, disks, 1);
235 			rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
236 			rf_ASMParityAdjust(asm_p->qInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
237 			break;
238 		}
239 	}
240 	RF_ASSERT(asmList == NULL && pdaList == NULL);
241 	/* make the header structure */
242 	asm_hdr = rf_AllocAccessStripeMapHeader();
243 	RF_ASSERT(numStripes == totStripes);
244 	asm_hdr->numStripes = numStripes;
245 	asm_hdr->stripeMap = asm_list;
246 
247 	if (rf_mapDebug)
248 		rf_PrintAccessStripeMap(asm_hdr);
249 	return (asm_hdr);
250 }
251 /*****************************************************************************************
252  * This routine walks through an ASM list and marks the PDAs that have failed.
253  * It's called only when a disk failure causes an in-flight DAG to fail.
254  * The parity may consist of two components, but we want to use only one failedPDA
255  * pointer.  Thus we set failedPDA to point to the first parity component, and rely
256  * on the rest of the code to do the right thing with this.
257  ****************************************************************************************/
258 
259 void
260 rf_MarkFailuresInASMList(raidPtr, asm_h)
261 	RF_Raid_t *raidPtr;
262 	RF_AccessStripeMapHeader_t *asm_h;
263 {
264 	RF_RaidDisk_t **disks = raidPtr->Disks;
265 	RF_AccessStripeMap_t *asmap;
266 	RF_PhysDiskAddr_t *pda;
267 
268 	for (asmap = asm_h->stripeMap; asmap; asmap = asmap->next) {
269 		asmap->numDataFailed = asmap->numParityFailed = asmap->numQFailed = 0;
270 		asmap->numFailedPDAs = 0;
271 		memset((char *) asmap->failedPDAs, 0,
272 		    RF_MAX_FAILED_PDA * sizeof(RF_PhysDiskAddr_t *));
273 		for (pda = asmap->physInfo; pda; pda = pda->next) {
274 			if (RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
275 				asmap->numDataFailed++;
276 				asmap->failedPDAs[asmap->numFailedPDAs] = pda;
277 				asmap->numFailedPDAs++;
278 			}
279 		}
280 		pda = asmap->parityInfo;
281 		if (pda && RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
282 			asmap->numParityFailed++;
283 			asmap->failedPDAs[asmap->numFailedPDAs] = pda;
284 			asmap->numFailedPDAs++;
285 		}
286 		pda = asmap->qInfo;
287 		if (pda && RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
288 			asmap->numQFailed++;
289 			asmap->failedPDAs[asmap->numFailedPDAs] = pda;
290 			asmap->numFailedPDAs++;
291 		}
292 	}
293 }
294 
295 /*****************************************************************************************
296  *
297  * routines to allocate and free list elements.  All allocation routines zero the
298  * structure before returning it.
299  *
300  * FreePhysDiskAddr is static.  It should never be called directly, because
301  * FreeAccessStripeMap takes care of freeing the PhysDiskAddr list.
302  *
303  ****************************************************************************************/
304 
305 static RF_FreeList_t *rf_asmhdr_freelist;
306 #define RF_MAX_FREE_ASMHDR 128
307 #define RF_ASMHDR_INC       16
308 #define RF_ASMHDR_INITIAL   32
309 
310 static RF_FreeList_t *rf_asm_freelist;
311 #define RF_MAX_FREE_ASM 192
312 #define RF_ASM_INC       24
313 #define RF_ASM_INITIAL   64
314 
315 static RF_FreeList_t *rf_pda_freelist;
316 #define RF_MAX_FREE_PDA 192
317 #define RF_PDA_INC       24
318 #define RF_PDA_INITIAL   64
319 
320 /* called at shutdown time.  So far, all that is necessary is to release all the free lists */
321 static void rf_ShutdownMapModule(void *);
322 static void
323 rf_ShutdownMapModule(ignored)
324 	void   *ignored;
325 {
326 	RF_FREELIST_DESTROY(rf_asmhdr_freelist, next, (RF_AccessStripeMapHeader_t *));
327 	RF_FREELIST_DESTROY(rf_pda_freelist, next, (RF_PhysDiskAddr_t *));
328 	RF_FREELIST_DESTROY(rf_asm_freelist, next, (RF_AccessStripeMap_t *));
329 }
330 
331 int
332 rf_ConfigureMapModule(listp)
333 	RF_ShutdownList_t **listp;
334 {
335 	int     rc;
336 
337 	RF_FREELIST_CREATE(rf_asmhdr_freelist, RF_MAX_FREE_ASMHDR,
338 	    RF_ASMHDR_INC, sizeof(RF_AccessStripeMapHeader_t));
339 	if (rf_asmhdr_freelist == NULL) {
340 		return (ENOMEM);
341 	}
342 	RF_FREELIST_CREATE(rf_asm_freelist, RF_MAX_FREE_ASM,
343 	    RF_ASM_INC, sizeof(RF_AccessStripeMap_t));
344 	if (rf_asm_freelist == NULL) {
345 		RF_FREELIST_DESTROY(rf_asmhdr_freelist, next, (RF_AccessStripeMapHeader_t *));
346 		return (ENOMEM);
347 	}
348 	RF_FREELIST_CREATE(rf_pda_freelist, RF_MAX_FREE_PDA,
349 	    RF_PDA_INC, sizeof(RF_PhysDiskAddr_t));
350 	if (rf_pda_freelist == NULL) {
351 		RF_FREELIST_DESTROY(rf_asmhdr_freelist, next, (RF_AccessStripeMapHeader_t *));
352 		RF_FREELIST_DESTROY(rf_pda_freelist, next, (RF_PhysDiskAddr_t *));
353 		return (ENOMEM);
354 	}
355 	rc = rf_ShutdownCreate(listp, rf_ShutdownMapModule, NULL);
356 	if (rc) {
357 		rf_print_unable_to_add_shutdown(__FILE__, __LINE__, rc);
358 		rf_ShutdownMapModule(NULL);
359 		return (rc);
360 	}
361 	RF_FREELIST_PRIME(rf_asmhdr_freelist, RF_ASMHDR_INITIAL, next,
362 	    (RF_AccessStripeMapHeader_t *));
363 	RF_FREELIST_PRIME(rf_asm_freelist, RF_ASM_INITIAL, next,
364 	    (RF_AccessStripeMap_t *));
365 	RF_FREELIST_PRIME(rf_pda_freelist, RF_PDA_INITIAL, next,
366 	    (RF_PhysDiskAddr_t *));
367 
368 	return (0);
369 }
370 
371 RF_AccessStripeMapHeader_t *
372 rf_AllocAccessStripeMapHeader()
373 {
374 	RF_AccessStripeMapHeader_t *p;
375 
376 	RF_FREELIST_GET(rf_asmhdr_freelist, p, next, (RF_AccessStripeMapHeader_t *));
377 	memset((char *) p, 0, sizeof(RF_AccessStripeMapHeader_t));
378 
379 	return (p);
380 }
381 
382 
383 void
384 rf_FreeAccessStripeMapHeader(p)
385 	RF_AccessStripeMapHeader_t *p;
386 {
387 	RF_FREELIST_FREE(rf_asmhdr_freelist, p, next);
388 }
389 
390 RF_PhysDiskAddr_t *
391 rf_AllocPhysDiskAddr()
392 {
393 	RF_PhysDiskAddr_t *p;
394 
395 	RF_FREELIST_GET(rf_pda_freelist, p, next, (RF_PhysDiskAddr_t *));
396 	memset((char *) p, 0, sizeof(RF_PhysDiskAddr_t));
397 
398 	return (p);
399 }
400 /* allocates a list of PDAs, locking the free list only once
401  * when we have to call calloc, we do it one component at a time to simplify
402  * the process of freeing the list at program shutdown.  This should not be
403  * much of a performance hit, because it should be very infrequently executed.
404  */
405 RF_PhysDiskAddr_t *
406 rf_AllocPDAList(count)
407 	int     count;
408 {
409 	RF_PhysDiskAddr_t *p = NULL;
410 
411 	RF_FREELIST_GET_N(rf_pda_freelist, p, next, (RF_PhysDiskAddr_t *), count);
412 	return (p);
413 }
414 
415 void
416 rf_FreePhysDiskAddr(p)
417 	RF_PhysDiskAddr_t *p;
418 {
419 	RF_FREELIST_FREE(rf_pda_freelist, p, next);
420 }
421 
422 static void
423 rf_FreePDAList(l_start, l_end, count)
424 	RF_PhysDiskAddr_t *l_start, *l_end;	/* pointers to start and end
425 						 * of list */
426 	int     count;		/* number of elements in list */
427 {
428 	RF_FREELIST_FREE_N(rf_pda_freelist, l_start, next, (RF_PhysDiskAddr_t *), count);
429 }
430 
431 RF_AccessStripeMap_t *
432 rf_AllocAccessStripeMapComponent()
433 {
434 	RF_AccessStripeMap_t *p;
435 
436 	RF_FREELIST_GET(rf_asm_freelist, p, next, (RF_AccessStripeMap_t *));
437 	memset((char *) p, 0, sizeof(RF_AccessStripeMap_t));
438 
439 	return (p);
440 }
441 /* this is essentially identical to AllocPDAList.  I should combine the two.
442  * when we have to call calloc, we do it one component at a time to simplify
443  * the process of freeing the list at program shutdown.  This should not be
444  * much of a performance hit, because it should be very infrequently executed.
445  */
446 RF_AccessStripeMap_t *
447 rf_AllocASMList(count)
448 	int     count;
449 {
450 	RF_AccessStripeMap_t *p = NULL;
451 
452 	RF_FREELIST_GET_N(rf_asm_freelist, p, next, (RF_AccessStripeMap_t *), count);
453 	return (p);
454 }
455 
456 void
457 rf_FreeAccessStripeMapComponent(p)
458 	RF_AccessStripeMap_t *p;
459 {
460 	RF_FREELIST_FREE(rf_asm_freelist, p, next);
461 }
462 
463 static void
464 rf_FreeASMList(l_start, l_end, count)
465 	RF_AccessStripeMap_t *l_start, *l_end;
466 	int     count;
467 {
468 	RF_FREELIST_FREE_N(rf_asm_freelist, l_start, next, (RF_AccessStripeMap_t *), count);
469 }
470 
471 void
472 rf_FreeAccessStripeMap(hdr)
473 	RF_AccessStripeMapHeader_t *hdr;
474 {
475 	RF_AccessStripeMap_t *p, *pt = NULL;
476 	RF_PhysDiskAddr_t *pdp, *trailer, *pdaList = NULL, *pdaEnd = NULL;
477 	int     count = 0, t, asm_count = 0;
478 
479 	for (p = hdr->stripeMap; p; p = p->next) {
480 
481 		/* link the 3 pda lists into the accumulating pda list */
482 
483 		if (!pdaList)
484 			pdaList = p->qInfo;
485 		else
486 			pdaEnd->next = p->qInfo;
487 		for (trailer = NULL, pdp = p->qInfo; pdp;) {
488 			trailer = pdp;
489 			pdp = pdp->next;
490 			count++;
491 		}
492 		if (trailer)
493 			pdaEnd = trailer;
494 
495 		if (!pdaList)
496 			pdaList = p->parityInfo;
497 		else
498 			pdaEnd->next = p->parityInfo;
499 		for (trailer = NULL, pdp = p->parityInfo; pdp;) {
500 			trailer = pdp;
501 			pdp = pdp->next;
502 			count++;
503 		}
504 		if (trailer)
505 			pdaEnd = trailer;
506 
507 		if (!pdaList)
508 			pdaList = p->physInfo;
509 		else
510 			pdaEnd->next = p->physInfo;
511 		for (trailer = NULL, pdp = p->physInfo; pdp;) {
512 			trailer = pdp;
513 			pdp = pdp->next;
514 			count++;
515 		}
516 		if (trailer)
517 			pdaEnd = trailer;
518 
519 		pt = p;
520 		asm_count++;
521 	}
522 
523 	/* debug only */
524 	for (t = 0, pdp = pdaList; pdp; pdp = pdp->next)
525 		t++;
526 	RF_ASSERT(t == count);
527 
528 	if (pdaList)
529 		rf_FreePDAList(pdaList, pdaEnd, count);
530 	rf_FreeASMList(hdr->stripeMap, pt, asm_count);
531 	rf_FreeAccessStripeMapHeader(hdr);
532 }
533 /* We can't use the large write optimization if there are any failures in the stripe.
534  * In the declustered layout, there is no way to immediately determine what disks
535  * constitute a stripe, so we actually have to hunt through the stripe looking for failures.
536  * The reason we map the parity instead of just using asm->parityInfo->col is because
537  * the latter may have been already redirected to a spare drive, which would
538  * mess up the computation of the stripe offset.
539  *
540  * ASSUMES AT MOST ONE FAILURE IN THE STRIPE.
541  */
542 int
543 rf_CheckStripeForFailures(raidPtr, asmap)
544 	RF_Raid_t *raidPtr;
545 	RF_AccessStripeMap_t *asmap;
546 {
547 	RF_RowCol_t trow, tcol, prow, pcol, *diskids, row, i;
548 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
549 	RF_StripeCount_t stripeOffset;
550 	int     numFailures;
551 	RF_RaidAddr_t sosAddr;
552 	RF_SectorNum_t diskOffset, poffset;
553 	RF_RowCol_t testrow;
554 
555 	/* quick out in the fault-free case.  */
556 	RF_LOCK_MUTEX(raidPtr->mutex);
557 	numFailures = raidPtr->numFailures;
558 	RF_UNLOCK_MUTEX(raidPtr->mutex);
559 	if (numFailures == 0)
560 		return (0);
561 
562 	sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
563 	row = asmap->physInfo->row;
564 	(layoutPtr->map->IdentifyStripe) (raidPtr, asmap->raidAddress, &diskids, &testrow);
565 	(layoutPtr->map->MapParity) (raidPtr, asmap->raidAddress, &prow, &pcol, &poffset, 0);	/* get pcol */
566 
567 	/* this need not be true if we've redirected the access to a spare in
568 	 * another row RF_ASSERT(row == testrow); */
569 	stripeOffset = 0;
570 	for (i = 0; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++) {
571 		if (diskids[i] != pcol) {
572 			if (RF_DEAD_DISK(raidPtr->Disks[testrow][diskids[i]].status)) {
573 				if (raidPtr->status[testrow] != rf_rs_reconstructing)
574 					return (1);
575 				RF_ASSERT(raidPtr->reconControl[testrow]->fcol == diskids[i]);
576 				layoutPtr->map->MapSector(raidPtr,
577 				    sosAddr + stripeOffset * layoutPtr->sectorsPerStripeUnit,
578 				    &trow, &tcol, &diskOffset, 0);
579 				RF_ASSERT((trow == testrow) && (tcol == diskids[i]));
580 				if (!rf_CheckRUReconstructed(raidPtr->reconControl[testrow]->reconMap, diskOffset))
581 					return (1);
582 				asmap->flags |= RF_ASM_REDIR_LARGE_WRITE;
583 				return (0);
584 			}
585 			stripeOffset++;
586 		}
587 	}
588 	return (0);
589 }
590 /*
591    return the number of failed data units in the stripe.
592 */
593 
594 int
595 rf_NumFailedDataUnitsInStripe(raidPtr, asmap)
596 	RF_Raid_t *raidPtr;
597 	RF_AccessStripeMap_t *asmap;
598 {
599 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
600 	RF_RowCol_t trow, tcol, row, i;
601 	RF_SectorNum_t diskOffset;
602 	RF_RaidAddr_t sosAddr;
603 	int     numFailures;
604 
605 	/* quick out in the fault-free case.  */
606 	RF_LOCK_MUTEX(raidPtr->mutex);
607 	numFailures = raidPtr->numFailures;
608 	RF_UNLOCK_MUTEX(raidPtr->mutex);
609 	if (numFailures == 0)
610 		return (0);
611 	numFailures = 0;
612 
613 	sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
614 	row = asmap->physInfo->row;
615 	for (i = 0; i < layoutPtr->numDataCol; i++) {
616 		(layoutPtr->map->MapSector) (raidPtr, sosAddr + i * layoutPtr->sectorsPerStripeUnit,
617 		    &trow, &tcol, &diskOffset, 0);
618 		if (RF_DEAD_DISK(raidPtr->Disks[trow][tcol].status))
619 			numFailures++;
620 	}
621 
622 	return numFailures;
623 }
624 
625 
626 /*****************************************************************************************
627  *
628  * debug routines
629  *
630  ****************************************************************************************/
631 
632 void
633 rf_PrintAccessStripeMap(asm_h)
634 	RF_AccessStripeMapHeader_t *asm_h;
635 {
636 	rf_PrintFullAccessStripeMap(asm_h, 0);
637 }
638 
639 void
640 rf_PrintFullAccessStripeMap(asm_h, prbuf)
641 	RF_AccessStripeMapHeader_t *asm_h;
642 	int     prbuf;		/* flag to print buffer pointers */
643 {
644 	int     i;
645 	RF_AccessStripeMap_t *asmap = asm_h->stripeMap;
646 	RF_PhysDiskAddr_t *p;
647 	printf("%d stripes total\n", (int) asm_h->numStripes);
648 	for (; asmap; asmap = asmap->next) {
649 		/* printf("Num failures: %d\n",asmap->numDataFailed); */
650 		/* printf("Num sectors:
651 		 * %d\n",(int)asmap->totalSectorsAccessed); */
652 		printf("Stripe %d (%d sectors), failures: %d data, %d parity: ",
653 		    (int) asmap->stripeID,
654 		    (int) asmap->totalSectorsAccessed,
655 		    (int) asmap->numDataFailed,
656 		    (int) asmap->numParityFailed);
657 		if (asmap->parityInfo) {
658 			printf("Parity [r%d c%d s%d-%d", asmap->parityInfo->row, asmap->parityInfo->col,
659 			    (int) asmap->parityInfo->startSector,
660 			    (int) (asmap->parityInfo->startSector +
661 				asmap->parityInfo->numSector - 1));
662 			if (prbuf)
663 				printf(" b0x%lx", (unsigned long) asmap->parityInfo->bufPtr);
664 			if (asmap->parityInfo->next) {
665 				printf(", r%d c%d s%d-%d", asmap->parityInfo->next->row,
666 				    asmap->parityInfo->next->col,
667 				    (int) asmap->parityInfo->next->startSector,
668 				    (int) (asmap->parityInfo->next->startSector +
669 					asmap->parityInfo->next->numSector - 1));
670 				if (prbuf)
671 					printf(" b0x%lx", (unsigned long) asmap->parityInfo->next->bufPtr);
672 				RF_ASSERT(asmap->parityInfo->next->next == NULL);
673 			}
674 			printf("]\n\t");
675 		}
676 		for (i = 0, p = asmap->physInfo; p; p = p->next, i++) {
677 			printf("SU r%d c%d s%d-%d ", p->row, p->col, (int) p->startSector,
678 			    (int) (p->startSector + p->numSector - 1));
679 			if (prbuf)
680 				printf("b0x%lx ", (unsigned long) p->bufPtr);
681 			if (i && !(i & 1))
682 				printf("\n\t");
683 		}
684 		printf("\n");
685 		p = asm_h->stripeMap->failedPDAs[0];
686 		if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 1)
687 			printf("[multiple failures]\n");
688 		else
689 			if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 0)
690 				printf("\t[Failed PDA: r%d c%d s%d-%d]\n", p->row, p->col,
691 				    (int) p->startSector, (int) (p->startSector + p->numSector - 1));
692 	}
693 }
694 
695 void
696 rf_PrintRaidAddressInfo(raidPtr, raidAddr, numBlocks)
697 	RF_Raid_t *raidPtr;
698 	RF_RaidAddr_t raidAddr;
699 	RF_SectorCount_t numBlocks;
700 {
701 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
702 	RF_RaidAddr_t ra, sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
703 
704 	printf("Raid addrs of SU boundaries from start of stripe to end of access:\n\t");
705 	for (ra = sosAddr; ra <= raidAddr + numBlocks; ra += layoutPtr->sectorsPerStripeUnit) {
706 		printf("%d (0x%x), ", (int) ra, (int) ra);
707 	}
708 	printf("\n");
709 	printf("Offset into stripe unit: %d (0x%x)\n",
710 	    (int) (raidAddr % layoutPtr->sectorsPerStripeUnit),
711 	    (int) (raidAddr % layoutPtr->sectorsPerStripeUnit));
712 }
713 /*
714    given a parity descriptor and the starting address within a stripe,
715    range restrict the parity descriptor to touch only the correct stuff.
716 */
717 void
718 rf_ASMParityAdjust(
719     RF_PhysDiskAddr_t * toAdjust,
720     RF_StripeNum_t startAddrWithinStripe,
721     RF_SectorNum_t endAddress,
722     RF_RaidLayout_t * layoutPtr,
723     RF_AccessStripeMap_t * asm_p)
724 {
725 	RF_PhysDiskAddr_t *new_pda;
726 
727 	/* when we're accessing only a portion of one stripe unit, we want the
728 	 * parity descriptor to identify only the chunk of parity associated
729 	 * with the data.  When the access spans exactly one stripe unit
730 	 * boundary and is less than a stripe unit in size, it uses two
731 	 * disjoint regions of the parity unit.  When an access spans more
732 	 * than one stripe unit boundary, it uses all of the parity unit.
733 	 *
734 	 * To better handle the case where stripe units are small, we may
735 	 * eventually want to change the 2nd case so that if the SU size is
736 	 * below some threshold, we just read/write the whole thing instead of
737 	 * breaking it up into two accesses. */
738 	if (asm_p->numStripeUnitsAccessed == 1) {
739 		int     x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
740 		toAdjust->startSector += x;
741 		toAdjust->raidAddress += x;
742 		toAdjust->numSector = asm_p->physInfo->numSector;
743 		RF_ASSERT(toAdjust->numSector != 0);
744 	} else
745 		if (asm_p->numStripeUnitsAccessed == 2 && asm_p->totalSectorsAccessed < layoutPtr->sectorsPerStripeUnit) {
746 			int     x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
747 
748 			/* create a second pda and copy the parity map info
749 			 * into it */
750 			RF_ASSERT(toAdjust->next == NULL);
751 			new_pda = toAdjust->next = rf_AllocPhysDiskAddr();
752 			*new_pda = *toAdjust;	/* structure assignment */
753 			new_pda->next = NULL;
754 
755 			/* adjust the start sector & number of blocks for the
756 			 * first parity pda */
757 			toAdjust->startSector += x;
758 			toAdjust->raidAddress += x;
759 			toAdjust->numSector = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, startAddrWithinStripe) - startAddrWithinStripe;
760 			RF_ASSERT(toAdjust->numSector != 0);
761 
762 			/* adjust the second pda */
763 			new_pda->numSector = endAddress - rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, endAddress);
764 			/* new_pda->raidAddress =
765 			 * rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr,
766 			 * toAdjust->raidAddress); */
767 			RF_ASSERT(new_pda->numSector != 0);
768 		}
769 }
770 
771 /*
772    Check if a disk has been spared or failed. If spared,
773    redirect the I/O.
774    If it has been failed, record it in the asm pointer.
775    Fourth arg is whether data or parity.
776 */
777 void
778 rf_ASMCheckStatus(
779     RF_Raid_t * raidPtr,
780     RF_PhysDiskAddr_t * pda_p,
781     RF_AccessStripeMap_t * asm_p,
782     RF_RaidDisk_t ** disks,
783     int parity)
784 {
785 	RF_DiskStatus_t dstatus;
786 	RF_RowCol_t frow, fcol;
787 
788 	dstatus = disks[pda_p->row][pda_p->col].status;
789 
790 	if (dstatus == rf_ds_spared) {
791 		/* if the disk has been spared, redirect access to the spare */
792 		frow = pda_p->row;
793 		fcol = pda_p->col;
794 		pda_p->row = disks[frow][fcol].spareRow;
795 		pda_p->col = disks[frow][fcol].spareCol;
796 	} else
797 		if (dstatus == rf_ds_dist_spared) {
798 			/* ditto if disk has been spared to dist spare space */
799 			RF_RowCol_t or = pda_p->row, oc = pda_p->col;
800 			RF_SectorNum_t oo = pda_p->startSector;
801 
802 			if (pda_p->type == RF_PDA_TYPE_DATA)
803 				raidPtr->Layout.map->MapSector(raidPtr, pda_p->raidAddress, &pda_p->row, &pda_p->col, &pda_p->startSector, RF_REMAP);
804 			else
805 				raidPtr->Layout.map->MapParity(raidPtr, pda_p->raidAddress, &pda_p->row, &pda_p->col, &pda_p->startSector, RF_REMAP);
806 
807 			if (rf_mapDebug) {
808 				printf("Redirected r %d c %d o %d -> r%d c %d o %d\n", or, oc, (int) oo,
809 				    pda_p->row, pda_p->col, (int) pda_p->startSector);
810 			}
811 		} else
812 			if (RF_DEAD_DISK(dstatus)) {
813 				/* if the disk is inaccessible, mark the
814 				 * failure */
815 				if (parity)
816 					asm_p->numParityFailed++;
817 				else {
818 					asm_p->numDataFailed++;
819 				}
820 				asm_p->failedPDAs[asm_p->numFailedPDAs] = pda_p;
821 				asm_p->numFailedPDAs++;
822 #if 0
823 				switch (asm_p->numParityFailed + asm_p->numDataFailed) {
824 				case 1:
825 					asm_p->failedPDAs[0] = pda_p;
826 					break;
827 				case 2:
828 					asm_p->failedPDAs[1] = pda_p;
829 				default:
830 					break;
831 				}
832 #endif
833 			}
834 	/* the redirected access should never span a stripe unit boundary */
835 	RF_ASSERT(rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress) ==
836 	    rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress + pda_p->numSector - 1));
837 	RF_ASSERT(pda_p->col != -1);
838 }
839