rf_map.c revision 1.15 1 /* $NetBSD: rf_map.c,v 1.15 2002/09/19 23:23:19 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 1.15 2002/09/19 23:23:19 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_DEBUG_MAP
125 if (rf_mapDebug)
126 rf_PrintRaidAddressInfo(raidPtr, raidAddress, numBlocks);
127 #endif
128 for (; raidAddress < endAddress;) {
129 /* make the next stripe structure */
130 RF_ASSERT(asmList);
131 t_asm = asmList;
132 asmList = asmList->next;
133 memset((char *) t_asm, 0, sizeof(RF_AccessStripeMap_t));
134 if (!asm_p)
135 asm_list = asm_p = t_asm;
136 else {
137 asm_p->next = t_asm;
138 asm_p = asm_p->next;
139 }
140 numStripes++;
141
142 /* map SUs from current location to the end of the stripe */
143 asm_p->stripeID = /* rf_RaidAddressToStripeID(layoutPtr,
144 raidAddress) */ stripeID++;
145 stripeRealEndAddress = rf_RaidAddressOfNextStripeBoundary(layoutPtr, raidAddress);
146 stripeEndAddress = RF_MIN(endAddress, stripeRealEndAddress);
147 asm_p->raidAddress = raidAddress;
148 asm_p->endRaidAddress = stripeEndAddress;
149
150 /* map each stripe unit in the stripe */
151 pda_p = NULL;
152 startAddrWithinStripe = raidAddress; /* Raid addr of start of
153 * portion of access
154 * that is within this
155 * stripe */
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, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
170
171 /* mark any failures we find. failedPDA is don't-care
172 * if there is more than one failure */
173 pda_p->raidAddress = raidAddress; /* the RAID address
174 * corresponding to this
175 * physical disk address */
176 nextStripeUnitAddress = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, raidAddress);
177 pda_p->numSector = RF_MIN(endAddress, nextStripeUnitAddress) - raidAddress;
178 RF_ASSERT(pda_p->numSector != 0);
179 rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 0);
180 pda_p->bufPtr = buffer + rf_RaidAddressToByte(raidPtr, (raidAddress - startAddress));
181 asm_p->totalSectorsAccessed += pda_p->numSector;
182 asm_p->numStripeUnitsAccessed++;
183 asm_p->origRow = pda_p->row; /* redundant but
184 * harmless to do this
185 * in every loop
186 * iteration */
187
188 raidAddress = RF_MIN(endAddress, nextStripeUnitAddress);
189 }
190
191 /* Map the parity. At this stage, the startSector and
192 * numSector fields for the parity unit are always set to
193 * indicate the entire parity unit. We may modify this after
194 * mapping the data portion. */
195 switch (faultsTolerated) {
196 case 0:
197 break;
198 case 1: /* single fault tolerant */
199 RF_ASSERT(pdaList);
200 t_pda = pdaList;
201 pdaList = pdaList->next;
202 memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
203 pda_p = asm_p->parityInfo = t_pda;
204 pda_p->type = RF_PDA_TYPE_PARITY;
205 (layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
206 &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
207 pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
208 /* raidAddr may be needed to find unit to redirect to */
209 pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
210 rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
211 rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
212
213 break;
214 case 2: /* two fault tolerant */
215 RF_ASSERT(pdaList && pdaList->next);
216 t_pda = pdaList;
217 pdaList = pdaList->next;
218 memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
219 pda_p = asm_p->parityInfo = t_pda;
220 pda_p->type = RF_PDA_TYPE_PARITY;
221 t_pda = pdaList;
222 pdaList = pdaList->next;
223 memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
224 pda_q = asm_p->qInfo = t_pda;
225 pda_q->type = RF_PDA_TYPE_Q;
226 (layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
227 &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
228 (layoutPtr->map->MapQ) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
229 &(pda_q->row), &(pda_q->col), &(pda_q->startSector), remap);
230 pda_q->numSector = pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
231 /* raidAddr may be needed to find unit to redirect to */
232 pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
233 pda_q->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
234 /* failure mode stuff */
235 rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
236 rf_ASMCheckStatus(raidPtr, pda_q, asm_p, disks, 1);
237 rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
238 rf_ASMParityAdjust(asm_p->qInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
239 break;
240 }
241 }
242 RF_ASSERT(asmList == NULL && pdaList == NULL);
243 /* make the header structure */
244 asm_hdr = rf_AllocAccessStripeMapHeader();
245 RF_ASSERT(numStripes == totStripes);
246 asm_hdr->numStripes = numStripes;
247 asm_hdr->stripeMap = asm_list;
248
249 #if RF_DEBUG_MAP
250 if (rf_mapDebug)
251 rf_PrintAccessStripeMap(asm_hdr);
252 #endif
253 return (asm_hdr);
254 }
255 /*****************************************************************************************
256 * This routine walks through an ASM list and marks the PDAs that have failed.
257 * It's called only when a disk failure causes an in-flight DAG to fail.
258 * The parity may consist of two components, but we want to use only one failedPDA
259 * pointer. Thus we set failedPDA to point to the first parity component, and rely
260 * on the rest of the code to do the right thing with this.
261 ****************************************************************************************/
262
263 void
264 rf_MarkFailuresInASMList(raidPtr, asm_h)
265 RF_Raid_t *raidPtr;
266 RF_AccessStripeMapHeader_t *asm_h;
267 {
268 RF_RaidDisk_t **disks = raidPtr->Disks;
269 RF_AccessStripeMap_t *asmap;
270 RF_PhysDiskAddr_t *pda;
271
272 for (asmap = asm_h->stripeMap; asmap; asmap = asmap->next) {
273 asmap->numDataFailed = asmap->numParityFailed = asmap->numQFailed = 0;
274 asmap->numFailedPDAs = 0;
275 memset((char *) asmap->failedPDAs, 0,
276 RF_MAX_FAILED_PDA * sizeof(RF_PhysDiskAddr_t *));
277 for (pda = asmap->physInfo; pda; pda = pda->next) {
278 if (RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
279 asmap->numDataFailed++;
280 asmap->failedPDAs[asmap->numFailedPDAs] = pda;
281 asmap->numFailedPDAs++;
282 }
283 }
284 pda = asmap->parityInfo;
285 if (pda && RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
286 asmap->numParityFailed++;
287 asmap->failedPDAs[asmap->numFailedPDAs] = pda;
288 asmap->numFailedPDAs++;
289 }
290 pda = asmap->qInfo;
291 if (pda && RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
292 asmap->numQFailed++;
293 asmap->failedPDAs[asmap->numFailedPDAs] = pda;
294 asmap->numFailedPDAs++;
295 }
296 }
297 }
298
299 /*****************************************************************************************
300 *
301 * routines to allocate and free list elements. All allocation routines zero the
302 * structure before returning it.
303 *
304 * FreePhysDiskAddr is static. It should never be called directly, because
305 * FreeAccessStripeMap takes care of freeing the PhysDiskAddr list.
306 *
307 ****************************************************************************************/
308
309 static RF_FreeList_t *rf_asmhdr_freelist;
310 #define RF_MAX_FREE_ASMHDR 128
311 #define RF_ASMHDR_INC 16
312 #define RF_ASMHDR_INITIAL 32
313
314 static RF_FreeList_t *rf_asm_freelist;
315 #define RF_MAX_FREE_ASM 192
316 #define RF_ASM_INC 24
317 #define RF_ASM_INITIAL 64
318
319 static RF_FreeList_t *rf_pda_freelist;
320 #define RF_MAX_FREE_PDA 192
321 #define RF_PDA_INC 24
322 #define RF_PDA_INITIAL 64
323
324 /* called at shutdown time. So far, all that is necessary is to release all the free lists */
325 static void rf_ShutdownMapModule(void *);
326 static void
327 rf_ShutdownMapModule(ignored)
328 void *ignored;
329 {
330 RF_FREELIST_DESTROY(rf_asmhdr_freelist, next, (RF_AccessStripeMapHeader_t *));
331 RF_FREELIST_DESTROY(rf_pda_freelist, next, (RF_PhysDiskAddr_t *));
332 RF_FREELIST_DESTROY(rf_asm_freelist, next, (RF_AccessStripeMap_t *));
333 }
334
335 int
336 rf_ConfigureMapModule(listp)
337 RF_ShutdownList_t **listp;
338 {
339 int rc;
340
341 RF_FREELIST_CREATE(rf_asmhdr_freelist, RF_MAX_FREE_ASMHDR,
342 RF_ASMHDR_INC, sizeof(RF_AccessStripeMapHeader_t));
343 if (rf_asmhdr_freelist == NULL) {
344 return (ENOMEM);
345 }
346 RF_FREELIST_CREATE(rf_asm_freelist, RF_MAX_FREE_ASM,
347 RF_ASM_INC, sizeof(RF_AccessStripeMap_t));
348 if (rf_asm_freelist == NULL) {
349 RF_FREELIST_DESTROY(rf_asmhdr_freelist, next, (RF_AccessStripeMapHeader_t *));
350 return (ENOMEM);
351 }
352 RF_FREELIST_CREATE(rf_pda_freelist, RF_MAX_FREE_PDA,
353 RF_PDA_INC, sizeof(RF_PhysDiskAddr_t));
354 if (rf_pda_freelist == NULL) {
355 RF_FREELIST_DESTROY(rf_asmhdr_freelist, next, (RF_AccessStripeMapHeader_t *));
356 RF_FREELIST_DESTROY(rf_pda_freelist, next, (RF_PhysDiskAddr_t *));
357 return (ENOMEM);
358 }
359 rc = rf_ShutdownCreate(listp, rf_ShutdownMapModule, NULL);
360 if (rc) {
361 rf_print_unable_to_add_shutdown(__FILE__, __LINE__, rc);
362 rf_ShutdownMapModule(NULL);
363 return (rc);
364 }
365 RF_FREELIST_PRIME(rf_asmhdr_freelist, RF_ASMHDR_INITIAL, next,
366 (RF_AccessStripeMapHeader_t *));
367 RF_FREELIST_PRIME(rf_asm_freelist, RF_ASM_INITIAL, next,
368 (RF_AccessStripeMap_t *));
369 RF_FREELIST_PRIME(rf_pda_freelist, RF_PDA_INITIAL, next,
370 (RF_PhysDiskAddr_t *));
371
372 return (0);
373 }
374
375 RF_AccessStripeMapHeader_t *
376 rf_AllocAccessStripeMapHeader()
377 {
378 RF_AccessStripeMapHeader_t *p;
379
380 RF_FREELIST_GET(rf_asmhdr_freelist, p, next, (RF_AccessStripeMapHeader_t *));
381 memset((char *) p, 0, sizeof(RF_AccessStripeMapHeader_t));
382
383 return (p);
384 }
385
386
387 void
388 rf_FreeAccessStripeMapHeader(p)
389 RF_AccessStripeMapHeader_t *p;
390 {
391 RF_FREELIST_FREE(rf_asmhdr_freelist, p, next);
392 }
393
394 RF_PhysDiskAddr_t *
395 rf_AllocPhysDiskAddr()
396 {
397 RF_PhysDiskAddr_t *p;
398
399 RF_FREELIST_GET(rf_pda_freelist, p, next, (RF_PhysDiskAddr_t *));
400 memset((char *) p, 0, sizeof(RF_PhysDiskAddr_t));
401
402 return (p);
403 }
404 /* allocates a list of PDAs, locking the free list only once
405 * when we have to call calloc, we do it one component at a time to simplify
406 * the process of freeing the list at program shutdown. This should not be
407 * much of a performance hit, because it should be very infrequently executed.
408 */
409 RF_PhysDiskAddr_t *
410 rf_AllocPDAList(count)
411 int count;
412 {
413 RF_PhysDiskAddr_t *p = NULL;
414
415 RF_FREELIST_GET_N(rf_pda_freelist, p, next, (RF_PhysDiskAddr_t *), count);
416 return (p);
417 }
418
419 void
420 rf_FreePhysDiskAddr(p)
421 RF_PhysDiskAddr_t *p;
422 {
423 RF_FREELIST_FREE(rf_pda_freelist, p, next);
424 }
425
426 static void
427 rf_FreePDAList(l_start, l_end, count)
428 RF_PhysDiskAddr_t *l_start, *l_end; /* pointers to start and end
429 * of list */
430 int count; /* number of elements in list */
431 {
432 RF_FREELIST_FREE_N(rf_pda_freelist, l_start, next, (RF_PhysDiskAddr_t *), count);
433 }
434
435 RF_AccessStripeMap_t *
436 rf_AllocAccessStripeMapComponent()
437 {
438 RF_AccessStripeMap_t *p;
439
440 RF_FREELIST_GET(rf_asm_freelist, p, next, (RF_AccessStripeMap_t *));
441 memset((char *) p, 0, sizeof(RF_AccessStripeMap_t));
442
443 return (p);
444 }
445 /* this is essentially identical to AllocPDAList. I should combine the two.
446 * when we have to call calloc, we do it one component at a time to simplify
447 * the process of freeing the list at program shutdown. This should not be
448 * much of a performance hit, because it should be very infrequently executed.
449 */
450 RF_AccessStripeMap_t *
451 rf_AllocASMList(count)
452 int count;
453 {
454 RF_AccessStripeMap_t *p = NULL;
455
456 RF_FREELIST_GET_N(rf_asm_freelist, p, next, (RF_AccessStripeMap_t *), count);
457 return (p);
458 }
459
460 void
461 rf_FreeAccessStripeMapComponent(p)
462 RF_AccessStripeMap_t *p;
463 {
464 RF_FREELIST_FREE(rf_asm_freelist, p, next);
465 }
466
467 static void
468 rf_FreeASMList(l_start, l_end, count)
469 RF_AccessStripeMap_t *l_start, *l_end;
470 int count;
471 {
472 RF_FREELIST_FREE_N(rf_asm_freelist, l_start, next, (RF_AccessStripeMap_t *), count);
473 }
474
475 void
476 rf_FreeAccessStripeMap(hdr)
477 RF_AccessStripeMapHeader_t *hdr;
478 {
479 RF_AccessStripeMap_t *p, *pt = NULL;
480 RF_PhysDiskAddr_t *pdp, *trailer, *pdaList = NULL, *pdaEnd = NULL;
481 int count = 0, t, asm_count = 0;
482
483 for (p = hdr->stripeMap; p; p = p->next) {
484
485 /* link the 3 pda lists into the accumulating pda list */
486
487 if (!pdaList)
488 pdaList = p->qInfo;
489 else
490 pdaEnd->next = p->qInfo;
491 for (trailer = NULL, pdp = p->qInfo; pdp;) {
492 trailer = pdp;
493 pdp = pdp->next;
494 count++;
495 }
496 if (trailer)
497 pdaEnd = trailer;
498
499 if (!pdaList)
500 pdaList = p->parityInfo;
501 else
502 pdaEnd->next = p->parityInfo;
503 for (trailer = NULL, pdp = p->parityInfo; pdp;) {
504 trailer = pdp;
505 pdp = pdp->next;
506 count++;
507 }
508 if (trailer)
509 pdaEnd = trailer;
510
511 if (!pdaList)
512 pdaList = p->physInfo;
513 else
514 pdaEnd->next = p->physInfo;
515 for (trailer = NULL, pdp = p->physInfo; pdp;) {
516 trailer = pdp;
517 pdp = pdp->next;
518 count++;
519 }
520 if (trailer)
521 pdaEnd = trailer;
522
523 pt = p;
524 asm_count++;
525 }
526
527 /* debug only */
528 for (t = 0, pdp = pdaList; pdp; pdp = pdp->next)
529 t++;
530 RF_ASSERT(t == count);
531
532 if (pdaList)
533 rf_FreePDAList(pdaList, pdaEnd, count);
534 rf_FreeASMList(hdr->stripeMap, pt, asm_count);
535 rf_FreeAccessStripeMapHeader(hdr);
536 }
537 /* We can't use the large write optimization if there are any failures in the stripe.
538 * In the declustered layout, there is no way to immediately determine what disks
539 * constitute a stripe, so we actually have to hunt through the stripe looking for failures.
540 * The reason we map the parity instead of just using asm->parityInfo->col is because
541 * the latter may have been already redirected to a spare drive, which would
542 * mess up the computation of the stripe offset.
543 *
544 * ASSUMES AT MOST ONE FAILURE IN THE STRIPE.
545 */
546 int
547 rf_CheckStripeForFailures(raidPtr, asmap)
548 RF_Raid_t *raidPtr;
549 RF_AccessStripeMap_t *asmap;
550 {
551 RF_RowCol_t trow, tcol, prow, pcol, *diskids, row, i;
552 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
553 RF_StripeCount_t stripeOffset;
554 int numFailures;
555 RF_RaidAddr_t sosAddr;
556 RF_SectorNum_t diskOffset, poffset;
557 RF_RowCol_t testrow;
558
559 /* quick out in the fault-free case. */
560 RF_LOCK_MUTEX(raidPtr->mutex);
561 numFailures = raidPtr->numFailures;
562 RF_UNLOCK_MUTEX(raidPtr->mutex);
563 if (numFailures == 0)
564 return (0);
565
566 sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
567 row = asmap->physInfo->row;
568 (layoutPtr->map->IdentifyStripe) (raidPtr, asmap->raidAddress, &diskids, &testrow);
569 (layoutPtr->map->MapParity) (raidPtr, asmap->raidAddress, &prow, &pcol, &poffset, 0); /* get pcol */
570
571 /* this need not be true if we've redirected the access to a spare in
572 * another row RF_ASSERT(row == testrow); */
573 stripeOffset = 0;
574 for (i = 0; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++) {
575 if (diskids[i] != pcol) {
576 if (RF_DEAD_DISK(raidPtr->Disks[testrow][diskids[i]].status)) {
577 if (raidPtr->status[testrow] != rf_rs_reconstructing)
578 return (1);
579 RF_ASSERT(raidPtr->reconControl[testrow]->fcol == diskids[i]);
580 layoutPtr->map->MapSector(raidPtr,
581 sosAddr + stripeOffset * layoutPtr->sectorsPerStripeUnit,
582 &trow, &tcol, &diskOffset, 0);
583 RF_ASSERT((trow == testrow) && (tcol == diskids[i]));
584 if (!rf_CheckRUReconstructed(raidPtr->reconControl[testrow]->reconMap, diskOffset))
585 return (1);
586 asmap->flags |= RF_ASM_REDIR_LARGE_WRITE;
587 return (0);
588 }
589 stripeOffset++;
590 }
591 }
592 return (0);
593 }
594 /*
595 return the number of failed data units in the stripe.
596 */
597
598 int
599 rf_NumFailedDataUnitsInStripe(raidPtr, asmap)
600 RF_Raid_t *raidPtr;
601 RF_AccessStripeMap_t *asmap;
602 {
603 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
604 RF_RowCol_t trow, tcol, row, i;
605 RF_SectorNum_t diskOffset;
606 RF_RaidAddr_t sosAddr;
607 int numFailures;
608
609 /* quick out in the fault-free case. */
610 RF_LOCK_MUTEX(raidPtr->mutex);
611 numFailures = raidPtr->numFailures;
612 RF_UNLOCK_MUTEX(raidPtr->mutex);
613 if (numFailures == 0)
614 return (0);
615 numFailures = 0;
616
617 sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
618 row = asmap->physInfo->row;
619 for (i = 0; i < layoutPtr->numDataCol; i++) {
620 (layoutPtr->map->MapSector) (raidPtr, sosAddr + i * layoutPtr->sectorsPerStripeUnit,
621 &trow, &tcol, &diskOffset, 0);
622 if (RF_DEAD_DISK(raidPtr->Disks[trow][tcol].status))
623 numFailures++;
624 }
625
626 return numFailures;
627 }
628
629
630 /*****************************************************************************************
631 *
632 * debug routines
633 *
634 ****************************************************************************************/
635
636 void
637 rf_PrintAccessStripeMap(asm_h)
638 RF_AccessStripeMapHeader_t *asm_h;
639 {
640 rf_PrintFullAccessStripeMap(asm_h, 0);
641 }
642
643 void
644 rf_PrintFullAccessStripeMap(asm_h, prbuf)
645 RF_AccessStripeMapHeader_t *asm_h;
646 int prbuf; /* flag to print buffer pointers */
647 {
648 int i;
649 RF_AccessStripeMap_t *asmap = asm_h->stripeMap;
650 RF_PhysDiskAddr_t *p;
651 printf("%d stripes total\n", (int) asm_h->numStripes);
652 for (; asmap; asmap = asmap->next) {
653 /* printf("Num failures: %d\n",asmap->numDataFailed); */
654 /* printf("Num sectors:
655 * %d\n",(int)asmap->totalSectorsAccessed); */
656 printf("Stripe %d (%d sectors), failures: %d data, %d parity: ",
657 (int) asmap->stripeID,
658 (int) asmap->totalSectorsAccessed,
659 (int) asmap->numDataFailed,
660 (int) asmap->numParityFailed);
661 if (asmap->parityInfo) {
662 printf("Parity [r%d c%d s%d-%d", asmap->parityInfo->row, asmap->parityInfo->col,
663 (int) asmap->parityInfo->startSector,
664 (int) (asmap->parityInfo->startSector +
665 asmap->parityInfo->numSector - 1));
666 if (prbuf)
667 printf(" b0x%lx", (unsigned long) asmap->parityInfo->bufPtr);
668 if (asmap->parityInfo->next) {
669 printf(", r%d c%d s%d-%d", asmap->parityInfo->next->row,
670 asmap->parityInfo->next->col,
671 (int) asmap->parityInfo->next->startSector,
672 (int) (asmap->parityInfo->next->startSector +
673 asmap->parityInfo->next->numSector - 1));
674 if (prbuf)
675 printf(" b0x%lx", (unsigned long) asmap->parityInfo->next->bufPtr);
676 RF_ASSERT(asmap->parityInfo->next->next == NULL);
677 }
678 printf("]\n\t");
679 }
680 for (i = 0, p = asmap->physInfo; p; p = p->next, i++) {
681 printf("SU r%d c%d s%d-%d ", p->row, p->col, (int) p->startSector,
682 (int) (p->startSector + p->numSector - 1));
683 if (prbuf)
684 printf("b0x%lx ", (unsigned long) p->bufPtr);
685 if (i && !(i & 1))
686 printf("\n\t");
687 }
688 printf("\n");
689 p = asm_h->stripeMap->failedPDAs[0];
690 if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 1)
691 printf("[multiple failures]\n");
692 else
693 if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 0)
694 printf("\t[Failed PDA: r%d c%d s%d-%d]\n", p->row, p->col,
695 (int) p->startSector, (int) (p->startSector + p->numSector - 1));
696 }
697 }
698
699 #if RF_MAP_DEBUG
700 void
701 rf_PrintRaidAddressInfo(raidPtr, raidAddr, numBlocks)
702 RF_Raid_t *raidPtr;
703 RF_RaidAddr_t raidAddr;
704 RF_SectorCount_t numBlocks;
705 {
706 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
707 RF_RaidAddr_t ra, sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
708
709 printf("Raid addrs of SU boundaries from start of stripe to end of access:\n\t");
710 for (ra = sosAddr; ra <= raidAddr + numBlocks; ra += layoutPtr->sectorsPerStripeUnit) {
711 printf("%d (0x%x), ", (int) ra, (int) ra);
712 }
713 printf("\n");
714 printf("Offset into stripe unit: %d (0x%x)\n",
715 (int) (raidAddr % layoutPtr->sectorsPerStripeUnit),
716 (int) (raidAddr % layoutPtr->sectorsPerStripeUnit));
717 }
718 #endif
719 /*
720 given a parity descriptor and the starting address within a stripe,
721 range restrict the parity descriptor to touch only the correct stuff.
722 */
723 void
724 rf_ASMParityAdjust(
725 RF_PhysDiskAddr_t * toAdjust,
726 RF_StripeNum_t startAddrWithinStripe,
727 RF_SectorNum_t endAddress,
728 RF_RaidLayout_t * layoutPtr,
729 RF_AccessStripeMap_t * asm_p)
730 {
731 RF_PhysDiskAddr_t *new_pda;
732
733 /* when we're accessing only a portion of one stripe unit, we want the
734 * parity descriptor to identify only the chunk of parity associated
735 * with the data. When the access spans exactly one stripe unit
736 * boundary and is less than a stripe unit in size, it uses two
737 * disjoint regions of the parity unit. When an access spans more
738 * than one stripe unit boundary, it uses all of the parity unit.
739 *
740 * To better handle the case where stripe units are small, we may
741 * eventually want to change the 2nd case so that if the SU size is
742 * below some threshold, we just read/write the whole thing instead of
743 * breaking it up into two accesses. */
744 if (asm_p->numStripeUnitsAccessed == 1) {
745 int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
746 toAdjust->startSector += x;
747 toAdjust->raidAddress += x;
748 toAdjust->numSector = asm_p->physInfo->numSector;
749 RF_ASSERT(toAdjust->numSector != 0);
750 } else
751 if (asm_p->numStripeUnitsAccessed == 2 && asm_p->totalSectorsAccessed < layoutPtr->sectorsPerStripeUnit) {
752 int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
753
754 /* create a second pda and copy the parity map info
755 * into it */
756 RF_ASSERT(toAdjust->next == NULL);
757 new_pda = toAdjust->next = rf_AllocPhysDiskAddr();
758 *new_pda = *toAdjust; /* structure assignment */
759 new_pda->next = NULL;
760
761 /* adjust the start sector & number of blocks for the
762 * first parity pda */
763 toAdjust->startSector += x;
764 toAdjust->raidAddress += x;
765 toAdjust->numSector = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, startAddrWithinStripe) - startAddrWithinStripe;
766 RF_ASSERT(toAdjust->numSector != 0);
767
768 /* adjust the second pda */
769 new_pda->numSector = endAddress - rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, endAddress);
770 /* new_pda->raidAddress =
771 * rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr,
772 * toAdjust->raidAddress); */
773 RF_ASSERT(new_pda->numSector != 0);
774 }
775 }
776
777 /*
778 Check if a disk has been spared or failed. If spared,
779 redirect the I/O.
780 If it has been failed, record it in the asm pointer.
781 Fourth arg is whether data or parity.
782 */
783 void
784 rf_ASMCheckStatus(
785 RF_Raid_t * raidPtr,
786 RF_PhysDiskAddr_t * pda_p,
787 RF_AccessStripeMap_t * asm_p,
788 RF_RaidDisk_t ** disks,
789 int parity)
790 {
791 RF_DiskStatus_t dstatus;
792 RF_RowCol_t frow, fcol;
793
794 dstatus = disks[pda_p->row][pda_p->col].status;
795
796 if (dstatus == rf_ds_spared) {
797 /* if the disk has been spared, redirect access to the spare */
798 frow = pda_p->row;
799 fcol = pda_p->col;
800 pda_p->row = disks[frow][fcol].spareRow;
801 pda_p->col = disks[frow][fcol].spareCol;
802 } else
803 if (dstatus == rf_ds_dist_spared) {
804 /* ditto if disk has been spared to dist spare space */
805 #if RF_DEBUG_MAP
806 RF_RowCol_t or = pda_p->row, oc = pda_p->col;
807 RF_SectorNum_t oo = pda_p->startSector;
808 #endif
809 if (pda_p->type == RF_PDA_TYPE_DATA)
810 raidPtr->Layout.map->MapSector(raidPtr, pda_p->raidAddress, &pda_p->row, &pda_p->col, &pda_p->startSector, RF_REMAP);
811 else
812 raidPtr->Layout.map->MapParity(raidPtr, pda_p->raidAddress, &pda_p->row, &pda_p->col, &pda_p->startSector, RF_REMAP);
813
814 #if RF_DEBUG_MAP
815 if (rf_mapDebug) {
816 printf("Redirected r %d c %d o %d -> r%d c %d o %d\n", or, oc, (int) oo,
817 pda_p->row, pda_p->col, (int) pda_p->startSector);
818 }
819 #endif
820 } else
821 if (RF_DEAD_DISK(dstatus)) {
822 /* if the disk is inaccessible, mark the
823 * failure */
824 if (parity)
825 asm_p->numParityFailed++;
826 else {
827 asm_p->numDataFailed++;
828 }
829 asm_p->failedPDAs[asm_p->numFailedPDAs] = pda_p;
830 asm_p->numFailedPDAs++;
831 #if 0
832 switch (asm_p->numParityFailed + asm_p->numDataFailed) {
833 case 1:
834 asm_p->failedPDAs[0] = pda_p;
835 break;
836 case 2:
837 asm_p->failedPDAs[1] = pda_p;
838 default:
839 break;
840 }
841 #endif
842 }
843 /* the redirected access should never span a stripe unit boundary */
844 RF_ASSERT(rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress) ==
845 rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress + pda_p->numSector - 1));
846 RF_ASSERT(pda_p->col != -1);
847 }
848