rf_map.c revision 1.32 1 /* $NetBSD: rf_map.c,v 1.32 2004/03/07 21:57:44 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 (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 * 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 1.32 2004/03/07 21:57:44 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_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/recieve 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, caddr_t 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((char *) t_asm, 0, sizeof(RF_AccessStripeMap_t));
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((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
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 = 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((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
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((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
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((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
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((char *) asmap->failedPDAs, 0,
282 RF_MAX_FAILED_PDA * sizeof(RF_PhysDiskAddr_t *));
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 /* called at shutdown time. So far, all that is necessary is to
326 release all the free lists */
327 static void rf_ShutdownMapModule(void *);
328 static void
329 rf_ShutdownMapModule(void *ignored)
330 {
331 pool_destroy(&rf_pools.asm_hdr);
332 pool_destroy(&rf_pools.asmap);
333 pool_destroy(&rf_pools.pda);
334 }
335
336 int
337 rf_ConfigureMapModule(RF_ShutdownList_t **listp)
338 {
339
340 rf_pool_init(&rf_pools.asm_hdr, sizeof(RF_AccessStripeMapHeader_t),
341 "rf_asmhdr_pl", RF_MIN_FREE_ASMHDR, RF_MAX_FREE_ASMHDR);
342 rf_pool_init(&rf_pools.asmap, sizeof(RF_AccessStripeMap_t),
343 "rf_asm_pl", RF_MIN_FREE_ASM, RF_MAX_FREE_ASM);
344 rf_pool_init(&rf_pools.pda, sizeof(RF_PhysDiskAddr_t),
345 "rf_pda_pl", RF_MIN_FREE_PDA, RF_MAX_FREE_PDA);
346 rf_ShutdownCreate(listp, rf_ShutdownMapModule, NULL);
347
348 return (0);
349 }
350
351 RF_AccessStripeMapHeader_t *
352 rf_AllocAccessStripeMapHeader()
353 {
354 RF_AccessStripeMapHeader_t *p;
355
356 p = pool_get(&rf_pools.asm_hdr, PR_WAITOK);
357 memset((char *) p, 0, sizeof(RF_AccessStripeMapHeader_t));
358
359 return (p);
360 }
361
362 void
363 rf_FreeAccessStripeMapHeader(RF_AccessStripeMapHeader_t *p)
364 {
365 pool_put(&rf_pools.asm_hdr, p);
366 }
367
368 RF_PhysDiskAddr_t *
369 rf_AllocPhysDiskAddr()
370 {
371 RF_PhysDiskAddr_t *p;
372
373 p = pool_get(&rf_pools.pda, PR_WAITOK);
374 memset((char *) p, 0, sizeof(RF_PhysDiskAddr_t));
375
376 return (p);
377 }
378 /* allocates a list of PDAs, locking the free list only once when we
379 * have to call calloc, we do it one component at a time to simplify
380 * the process of freeing the list at program shutdown. This should
381 * not be much of a performance hit, because it should be very
382 * infrequently executed. */
383 RF_PhysDiskAddr_t *
384 rf_AllocPDAList(int count)
385 {
386 RF_PhysDiskAddr_t *p, *prev;
387 int i;
388
389 p = NULL;
390 prev = NULL;
391 for (i = 0; i < count; i++) {
392 p = pool_get(&rf_pools.pda, PR_WAITOK);
393 p->next = prev;
394 prev = p;
395 }
396
397 return (p);
398 }
399
400 #if RF_INCLUDE_PARITYLOGGING > 0
401 void
402 rf_FreePhysDiskAddr(RF_PhysDiskAddr_t *p)
403 {
404 pool_put(&rf_pools.pda, p);
405 }
406 #endif
407
408 static void
409 rf_FreePDAList(RF_PhysDiskAddr_t *pda_list)
410 {
411 RF_PhysDiskAddr_t *p, *tmp;
412
413 p=pda_list;
414 while (p) {
415 tmp = p->next;
416 pool_put(&rf_pools.pda, p);
417 p = tmp;
418 }
419 }
420
421 /* this is essentially identical to AllocPDAList. I should combine
422 * the two. when we have to call calloc, we do it one component at a
423 * time to simplify the process of freeing the list at program
424 * shutdown. This should not be much of a performance hit, because it
425 * should be very infrequently executed. */
426 RF_AccessStripeMap_t *
427 rf_AllocASMList(int count)
428 {
429 RF_AccessStripeMap_t *p, *prev;
430 int i;
431
432 p = NULL;
433 prev = NULL;
434 for (i = 0; i < count; i++) {
435 p = pool_get(&rf_pools.asmap, PR_WAITOK);
436 p->next = prev;
437 prev = p;
438 }
439 return (p);
440 }
441
442 static void
443 rf_FreeASMList(RF_AccessStripeMap_t *asm_list)
444 {
445 RF_AccessStripeMap_t *p, *tmp;
446
447 p=asm_list;
448 while (p) {
449 tmp = p->next;
450 pool_put(&rf_pools.asmap, p);
451 p = tmp;
452 }
453 }
454
455 void
456 rf_FreeAccessStripeMap(RF_AccessStripeMapHeader_t *hdr)
457 {
458 RF_AccessStripeMap_t *p;
459 RF_PhysDiskAddr_t *pdp, *trailer, *pdaList = NULL, *pdaEnd = NULL;
460 int count = 0, t, asm_count = 0;
461
462 for (p = hdr->stripeMap; p; p = p->next) {
463
464 /* link the 3 pda lists into the accumulating pda list */
465
466 if (!pdaList)
467 pdaList = p->qInfo;
468 else
469 pdaEnd->next = p->qInfo;
470 for (trailer = NULL, pdp = p->qInfo; pdp;) {
471 trailer = pdp;
472 pdp = pdp->next;
473 count++;
474 }
475 if (trailer)
476 pdaEnd = trailer;
477
478 if (!pdaList)
479 pdaList = p->parityInfo;
480 else
481 pdaEnd->next = p->parityInfo;
482 for (trailer = NULL, pdp = p->parityInfo; pdp;) {
483 trailer = pdp;
484 pdp = pdp->next;
485 count++;
486 }
487 if (trailer)
488 pdaEnd = trailer;
489
490 if (!pdaList)
491 pdaList = p->physInfo;
492 else
493 pdaEnd->next = p->physInfo;
494 for (trailer = NULL, pdp = p->physInfo; pdp;) {
495 trailer = pdp;
496 pdp = pdp->next;
497 count++;
498 }
499 if (trailer)
500 pdaEnd = trailer;
501
502 asm_count++;
503 }
504
505 /* debug only */
506 for (t = 0, pdp = pdaList; pdp; pdp = pdp->next)
507 t++;
508 RF_ASSERT(t == count);
509
510 if (pdaList)
511 rf_FreePDAList(pdaList);
512 rf_FreeASMList(hdr->stripeMap);
513 rf_FreeAccessStripeMapHeader(hdr);
514 }
515 /* We can't use the large write optimization if there are any failures
516 * in the stripe. In the declustered layout, there is no way to
517 * immediately determine what disks constitute a stripe, so we
518 * actually have to hunt through the stripe looking for failures. The
519 * reason we map the parity instead of just using asm->parityInfo->col
520 * is because the latter may have been already redirected to a spare
521 * drive, which would mess up the computation of the stripe offset.
522 *
523 * ASSUMES AT MOST ONE FAILURE IN THE STRIPE. */
524 int
525 rf_CheckStripeForFailures(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
526 {
527 RF_RowCol_t tcol, pcol, *diskids, i;
528 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
529 RF_StripeCount_t stripeOffset;
530 int numFailures;
531 RF_RaidAddr_t sosAddr;
532 RF_SectorNum_t diskOffset, poffset;
533
534 /* quick out in the fault-free case. */
535 RF_LOCK_MUTEX(raidPtr->mutex);
536 numFailures = raidPtr->numFailures;
537 RF_UNLOCK_MUTEX(raidPtr->mutex);
538 if (numFailures == 0)
539 return (0);
540
541 sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
542 asmap->raidAddress);
543 (layoutPtr->map->IdentifyStripe) (raidPtr, asmap->raidAddress,
544 &diskids);
545 (layoutPtr->map->MapParity) (raidPtr, asmap->raidAddress,
546 &pcol, &poffset, 0); /* get pcol */
547
548 /* this need not be true if we've redirected the access to a
549 * spare in another row RF_ASSERT(row == testrow); */
550 stripeOffset = 0;
551 for (i = 0; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++) {
552 if (diskids[i] != pcol) {
553 if (RF_DEAD_DISK(raidPtr->Disks[diskids[i]].status)) {
554 if (raidPtr->status != rf_rs_reconstructing)
555 return (1);
556 RF_ASSERT(raidPtr->reconControl->fcol == diskids[i]);
557 layoutPtr->map->MapSector(raidPtr,
558 sosAddr + stripeOffset * layoutPtr->sectorsPerStripeUnit,
559 &tcol, &diskOffset, 0);
560 RF_ASSERT(tcol == diskids[i]);
561 if (!rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, diskOffset))
562 return (1);
563 asmap->flags |= RF_ASM_REDIR_LARGE_WRITE;
564 return (0);
565 }
566 stripeOffset++;
567 }
568 }
569 return (0);
570 }
571 #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) || (RF_INCLUDE_EVENODD >0)
572 /*
573 return the number of failed data units in the stripe.
574 */
575
576 int
577 rf_NumFailedDataUnitsInStripe(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
578 {
579 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
580 RF_RowCol_t tcol, i;
581 RF_SectorNum_t diskOffset;
582 RF_RaidAddr_t sosAddr;
583 int numFailures;
584
585 /* quick out in the fault-free case. */
586 RF_LOCK_MUTEX(raidPtr->mutex);
587 numFailures = raidPtr->numFailures;
588 RF_UNLOCK_MUTEX(raidPtr->mutex);
589 if (numFailures == 0)
590 return (0);
591 numFailures = 0;
592
593 sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
594 asmap->raidAddress);
595 for (i = 0; i < layoutPtr->numDataCol; i++) {
596 (layoutPtr->map->MapSector) (raidPtr, sosAddr + i * layoutPtr->sectorsPerStripeUnit,
597 &trow, &tcol, &diskOffset, 0);
598 if (RF_DEAD_DISK(raidPtr->Disks[tcol].status))
599 numFailures++;
600 }
601
602 return numFailures;
603 }
604 #endif
605
606 /****************************************************************************
607 *
608 * debug routines
609 *
610 ***************************************************************************/
611 #if RF_DEBUG_MAP
612 void
613 rf_PrintAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h)
614 {
615 rf_PrintFullAccessStripeMap(asm_h, 0);
616 }
617 #endif
618
619 /* prbuf - flag to print buffer pointers */
620 void
621 rf_PrintFullAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h, int prbuf)
622 {
623 int i;
624 RF_AccessStripeMap_t *asmap = asm_h->stripeMap;
625 RF_PhysDiskAddr_t *p;
626 printf("%d stripes total\n", (int) asm_h->numStripes);
627 for (; asmap; asmap = asmap->next) {
628 /* printf("Num failures: %d\n",asmap->numDataFailed); */
629 /* printf("Num sectors:
630 * %d\n",(int)asmap->totalSectorsAccessed); */
631 printf("Stripe %d (%d sectors), failures: %d data, %d parity: ",
632 (int) asmap->stripeID,
633 (int) asmap->totalSectorsAccessed,
634 (int) asmap->numDataFailed,
635 (int) asmap->numParityFailed);
636 if (asmap->parityInfo) {
637 printf("Parity [c%d s%d-%d", asmap->parityInfo->col,
638 (int) asmap->parityInfo->startSector,
639 (int) (asmap->parityInfo->startSector +
640 asmap->parityInfo->numSector - 1));
641 if (prbuf)
642 printf(" b0x%lx", (unsigned long) asmap->parityInfo->bufPtr);
643 if (asmap->parityInfo->next) {
644 printf(", c%d s%d-%d", asmap->parityInfo->next->col,
645 (int) asmap->parityInfo->next->startSector,
646 (int) (asmap->parityInfo->next->startSector +
647 asmap->parityInfo->next->numSector - 1));
648 if (prbuf)
649 printf(" b0x%lx", (unsigned long) asmap->parityInfo->next->bufPtr);
650 RF_ASSERT(asmap->parityInfo->next->next == NULL);
651 }
652 printf("]\n\t");
653 }
654 for (i = 0, p = asmap->physInfo; p; p = p->next, i++) {
655 printf("SU c%d s%d-%d ", p->col, (int) p->startSector,
656 (int) (p->startSector + p->numSector - 1));
657 if (prbuf)
658 printf("b0x%lx ", (unsigned long) p->bufPtr);
659 if (i && !(i & 1))
660 printf("\n\t");
661 }
662 printf("\n");
663 p = asm_h->stripeMap->failedPDAs[0];
664 if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 1)
665 printf("[multiple failures]\n");
666 else
667 if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 0)
668 printf("\t[Failed PDA: c%d s%d-%d]\n", p->col,
669 (int) p->startSector, (int) (p->startSector + p->numSector - 1));
670 }
671 }
672
673 #if RF_MAP_DEBUG
674 void
675 rf_PrintRaidAddressInfo(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
676 RF_SectorCount_t numBlocks)
677 {
678 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
679 RF_RaidAddr_t ra, sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
680
681 printf("Raid addrs of SU boundaries from start of stripe to end of access:\n\t");
682 for (ra = sosAddr; ra <= raidAddr + numBlocks; ra += layoutPtr->sectorsPerStripeUnit) {
683 printf("%d (0x%x), ", (int) ra, (int) ra);
684 }
685 printf("\n");
686 printf("Offset into stripe unit: %d (0x%x)\n",
687 (int) (raidAddr % layoutPtr->sectorsPerStripeUnit),
688 (int) (raidAddr % layoutPtr->sectorsPerStripeUnit));
689 }
690 #endif
691 /* given a parity descriptor and the starting address within a stripe,
692 * range restrict the parity descriptor to touch only the correct
693 * stuff. */
694 void
695 rf_ASMParityAdjust(RF_PhysDiskAddr_t *toAdjust,
696 RF_StripeNum_t startAddrWithinStripe,
697 RF_SectorNum_t endAddress,
698 RF_RaidLayout_t *layoutPtr,
699 RF_AccessStripeMap_t *asm_p)
700 {
701 RF_PhysDiskAddr_t *new_pda;
702
703 /* when we're accessing only a portion of one stripe unit, we
704 * want the parity descriptor to identify only the chunk of
705 * parity associated with the data. When the access spans
706 * exactly one stripe unit boundary and is less than a stripe
707 * unit in size, it uses two disjoint regions of the parity
708 * unit. When an access spans more than one stripe unit
709 * boundary, it uses all of the parity unit.
710 *
711 * To better handle the case where stripe units are small, we
712 * may eventually want to change the 2nd case so that if the
713 * SU size is below some threshold, we just read/write the
714 * whole thing instead of breaking it up into two accesses. */
715 if (asm_p->numStripeUnitsAccessed == 1) {
716 int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
717 toAdjust->startSector += x;
718 toAdjust->raidAddress += x;
719 toAdjust->numSector = asm_p->physInfo->numSector;
720 RF_ASSERT(toAdjust->numSector != 0);
721 } else
722 if (asm_p->numStripeUnitsAccessed == 2 && asm_p->totalSectorsAccessed < layoutPtr->sectorsPerStripeUnit) {
723 int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
724
725 /* create a second pda and copy the parity map info
726 * into it */
727 RF_ASSERT(toAdjust->next == NULL);
728 new_pda = toAdjust->next = rf_AllocPhysDiskAddr();
729 *new_pda = *toAdjust; /* structure assignment */
730 new_pda->next = NULL;
731
732 /* adjust the start sector & number of blocks for the
733 * first parity pda */
734 toAdjust->startSector += x;
735 toAdjust->raidAddress += x;
736 toAdjust->numSector = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, startAddrWithinStripe) - startAddrWithinStripe;
737 RF_ASSERT(toAdjust->numSector != 0);
738
739 /* adjust the second pda */
740 new_pda->numSector = endAddress - rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, endAddress);
741 /* new_pda->raidAddress =
742 * rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr,
743 * toAdjust->raidAddress); */
744 RF_ASSERT(new_pda->numSector != 0);
745 }
746 }
747
748 /* Check if a disk has been spared or failed. If spared, redirect the
749 * I/O. If it has been failed, record it in the asm pointer. Fourth
750 * arg is whether data or parity. */
751 void
752 rf_ASMCheckStatus(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda_p,
753 RF_AccessStripeMap_t *asm_p, RF_RaidDisk_t *disks,
754 int parity)
755 {
756 RF_DiskStatus_t dstatus;
757 RF_RowCol_t fcol;
758
759 dstatus = disks[pda_p->col].status;
760
761 if (dstatus == rf_ds_spared) {
762 /* if the disk has been spared, redirect access to the spare */
763 fcol = pda_p->col;
764 pda_p->col = disks[fcol].spareCol;
765 } else
766 if (dstatus == rf_ds_dist_spared) {
767 /* ditto if disk has been spared to dist spare space */
768 #if RF_DEBUG_MAP
769 RF_RowCol_t oc = pda_p->col;
770 RF_SectorNum_t oo = pda_p->startSector;
771 #endif
772 if (pda_p->type == RF_PDA_TYPE_DATA)
773 raidPtr->Layout.map->MapSector(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
774 else
775 raidPtr->Layout.map->MapParity(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
776
777 #if RF_DEBUG_MAP
778 if (rf_mapDebug) {
779 printf("Redirected c %d o %d -> c %d o %d\n", oc, (int) oo,
780 pda_p->col, (int) pda_p->startSector);
781 }
782 #endif
783 } else
784 if (RF_DEAD_DISK(dstatus)) {
785 /* if the disk is inaccessible, mark the
786 * failure */
787 if (parity)
788 asm_p->numParityFailed++;
789 else {
790 asm_p->numDataFailed++;
791 }
792 asm_p->failedPDAs[asm_p->numFailedPDAs] = pda_p;
793 asm_p->numFailedPDAs++;
794 #if 0
795 switch (asm_p->numParityFailed + asm_p->numDataFailed) {
796 case 1:
797 asm_p->failedPDAs[0] = pda_p;
798 break;
799 case 2:
800 asm_p->failedPDAs[1] = pda_p;
801 default:
802 break;
803 }
804 #endif
805 }
806 /* the redirected access should never span a stripe unit boundary */
807 RF_ASSERT(rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress) ==
808 rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress + pda_p->numSector - 1));
809 RF_ASSERT(pda_p->col != -1);
810 }
811