rf_map.c revision 1.2 1 /* $NetBSD: rf_map.c,v 1.2 1999/01/26 02:33:58 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 "rf_types.h"
36 #include "rf_threadstuff.h"
37 #include "rf_raid.h"
38 #include "rf_general.h"
39 #include "rf_map.h"
40 #include "rf_freelist.h"
41 #include "rf_shutdown.h"
42 #include "rf_sys.h"
43
44 static void rf_FreePDAList(RF_PhysDiskAddr_t *start, RF_PhysDiskAddr_t *end, int count);
45 static void rf_FreeASMList(RF_AccessStripeMap_t *start, RF_AccessStripeMap_t *end,
46 int count);
47
48 /*****************************************************************************************
49 *
50 * MapAccess -- main 1st order mapping routine.
51 *
52 * Maps an access in the RAID address space to the corresponding set of physical disk
53 * addresses. The result is returned as a list of AccessStripeMap structures, one per
54 * stripe accessed. Each ASM structure contains a pointer to a list of PhysDiskAddr
55 * structures, which describe the physical locations touched by the user access. Note
56 * that this routine returns only static mapping information, i.e. the list of physical
57 * addresses returned does not necessarily identify the set of physical locations that
58 * will actually be read or written.
59 *
60 * The routine also maps the parity. The physical disk location returned always
61 * indicates the entire parity unit, even when only a subset of it is being accessed.
62 * This is because an access that is not stripe unit aligned but that spans a stripe
63 * unit boundary may require access two distinct portions of the parity unit, and we
64 * can't yet tell which portion(s) we'll actually need. We leave it up to the algorithm
65 * selection code to decide what subset of the parity unit to access.
66 *
67 * Note that addresses in the RAID address space must always be maintained as
68 * longs, instead of ints.
69 *
70 * This routine returns NULL if numBlocks is 0
71 *
72 ****************************************************************************************/
73
74 RF_AccessStripeMapHeader_t *rf_MapAccess(raidPtr, raidAddress, numBlocks, buffer, remap)
75 RF_Raid_t *raidPtr;
76 RF_RaidAddr_t raidAddress; /* starting address in RAID address space */
77 RF_SectorCount_t numBlocks; /* number of blocks in RAID address space to access */
78 caddr_t buffer; /* buffer to supply/receive data */
79 int remap; /* 1 => remap addresses to spare space */
80 {
81 RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
82 RF_AccessStripeMapHeader_t *asm_hdr = NULL;
83 RF_AccessStripeMap_t *asm_list = NULL, *asm_p = NULL;
84 int faultsTolerated = layoutPtr->map->faultsTolerated;
85 RF_RaidAddr_t startAddress = raidAddress; /* we'll change raidAddress along the way */
86 RF_RaidAddr_t endAddress = raidAddress + numBlocks;
87 RF_RaidDisk_t **disks = raidPtr->Disks;
88
89 RF_PhysDiskAddr_t *pda_p, *pda_q;
90 RF_StripeCount_t numStripes = 0;
91 RF_RaidAddr_t stripeRealEndAddress, stripeEndAddress, nextStripeUnitAddress;
92 RF_RaidAddr_t startAddrWithinStripe, lastRaidAddr;
93 RF_StripeCount_t totStripes;
94 RF_StripeNum_t stripeID, lastSID, SUID, lastSUID;
95 RF_AccessStripeMap_t *asmList, *t_asm;
96 RF_PhysDiskAddr_t *pdaList, *t_pda;
97
98 /* allocate all the ASMs and PDAs up front */
99 lastRaidAddr = raidAddress + numBlocks - 1 ;
100 stripeID = rf_RaidAddressToStripeID(layoutPtr, raidAddress);
101 lastSID = rf_RaidAddressToStripeID(layoutPtr, lastRaidAddr);
102 totStripes = lastSID - stripeID + 1;
103 SUID = rf_RaidAddressToStripeUnitID(layoutPtr, raidAddress);
104 lastSUID = rf_RaidAddressToStripeUnitID(layoutPtr, lastRaidAddr);
105
106 asmList = rf_AllocASMList(totStripes);
107 pdaList = rf_AllocPDAList(lastSUID - SUID + 1 + faultsTolerated * totStripes); /* may also need pda(s) per stripe for parity */
108
109 if (raidAddress+numBlocks > raidPtr->totalSectors) {
110 RF_ERRORMSG1("Unable to map access because offset (%d) was invalid\n",
111 (int)raidAddress);
112 return(NULL);
113 }
114
115 if (rf_mapDebug)
116 rf_PrintRaidAddressInfo(raidPtr, raidAddress, numBlocks);
117 for (; raidAddress < endAddress; ) {
118 /* make the next stripe structure */
119 RF_ASSERT(asmList);
120 t_asm = asmList;
121 asmList = asmList->next;
122 bzero((char *)t_asm, sizeof(RF_AccessStripeMap_t));
123 if (!asm_p)
124 asm_list = asm_p = t_asm;
125 else {
126 asm_p->next = t_asm;
127 asm_p = asm_p->next;
128 }
129 numStripes++;
130
131 /* map SUs from current location to the end of the stripe */
132 asm_p->stripeID = /*rf_RaidAddressToStripeID(layoutPtr, raidAddress)*/ stripeID++;
133 stripeRealEndAddress = rf_RaidAddressOfNextStripeBoundary(layoutPtr, raidAddress);
134 stripeEndAddress = RF_MIN(endAddress,stripeRealEndAddress );
135 asm_p->raidAddress = raidAddress;
136 asm_p->endRaidAddress = stripeEndAddress;
137
138 /* map each stripe unit in the stripe */
139 pda_p = NULL;
140 startAddrWithinStripe = raidAddress; /* Raid addr of start of portion of access that is within this stripe */
141 for (; raidAddress < stripeEndAddress; ) {
142 RF_ASSERT(pdaList);
143 t_pda = pdaList;
144 pdaList = pdaList->next;
145 bzero((char *)t_pda, sizeof(RF_PhysDiskAddr_t));
146 if (!pda_p)
147 asm_p->physInfo = pda_p = t_pda;
148 else {
149 pda_p->next = t_pda;
150 pda_p = pda_p->next;
151 }
152
153 pda_p->type = RF_PDA_TYPE_DATA;
154 (layoutPtr->map->MapSector)(raidPtr, raidAddress, &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
155
156 /* mark any failures we find. failedPDA is don't-care if there is more than one failure */
157 pda_p->raidAddress = raidAddress; /* the RAID address corresponding to this physical disk address */
158 nextStripeUnitAddress = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, raidAddress);
159 pda_p->numSector = RF_MIN(endAddress, nextStripeUnitAddress) - raidAddress;
160 RF_ASSERT(pda_p->numSector != 0);
161 rf_ASMCheckStatus(raidPtr,pda_p,asm_p,disks,0);
162 pda_p->bufPtr = buffer + rf_RaidAddressToByte(raidPtr, (raidAddress - startAddress));
163 asm_p->totalSectorsAccessed += pda_p->numSector;
164 asm_p->numStripeUnitsAccessed++;
165 asm_p->origRow = pda_p->row; /* redundant but harmless to do this in every loop iteration */
166
167 raidAddress = RF_MIN(endAddress, nextStripeUnitAddress);
168 }
169
170 /* Map the parity. At this stage, the startSector and numSector fields
171 * for the parity unit are always set to indicate the entire parity unit.
172 * We may modify this after mapping the data portion.
173 */
174 switch (faultsTolerated)
175 {
176 case 0:
177 break;
178 case 1: /* single fault tolerant */
179 RF_ASSERT(pdaList);
180 t_pda = pdaList;
181 pdaList = pdaList->next;
182 bzero((char *)t_pda, sizeof(RF_PhysDiskAddr_t));
183 pda_p = asm_p->parityInfo = t_pda;
184 pda_p->type = RF_PDA_TYPE_PARITY;
185 (layoutPtr->map->MapParity)(raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
186 &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
187 pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
188 /* raidAddr may be needed to find unit to redirect to */
189 pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
190 rf_ASMCheckStatus(raidPtr,pda_p,asm_p,disks,1);
191 rf_ASMParityAdjust(asm_p->parityInfo,startAddrWithinStripe,endAddress,layoutPtr,asm_p);
192
193 break;
194 case 2: /* two fault tolerant */
195 RF_ASSERT(pdaList && pdaList->next);
196 t_pda = pdaList;
197 pdaList = pdaList->next;
198 bzero((char *)t_pda, sizeof(RF_PhysDiskAddr_t));
199 pda_p = asm_p->parityInfo = t_pda;
200 pda_p->type = RF_PDA_TYPE_PARITY;
201 t_pda = pdaList;
202 pdaList = pdaList->next;
203 bzero((char *)t_pda, sizeof(RF_PhysDiskAddr_t));
204 pda_q = asm_p->qInfo = t_pda;
205 pda_q->type = RF_PDA_TYPE_Q;
206 (layoutPtr->map->MapParity)(raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
207 &(pda_p->row), &(pda_p->col), &(pda_p->startSector), remap);
208 (layoutPtr->map->MapQ)(raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
209 &(pda_q->row), &(pda_q->col), &(pda_q->startSector), remap);
210 pda_q->numSector = pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
211 /* raidAddr may be needed to find unit to redirect to */
212 pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
213 pda_q->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
214 /* failure mode stuff */
215 rf_ASMCheckStatus(raidPtr,pda_p,asm_p,disks,1);
216 rf_ASMCheckStatus(raidPtr,pda_q,asm_p,disks,1);
217 rf_ASMParityAdjust(asm_p->parityInfo,startAddrWithinStripe,endAddress,layoutPtr,asm_p);
218 rf_ASMParityAdjust(asm_p->qInfo,startAddrWithinStripe,endAddress,layoutPtr,asm_p);
219 break;
220 }
221 }
222 RF_ASSERT(asmList == NULL && pdaList == NULL);
223 /* make the header structure */
224 asm_hdr = rf_AllocAccessStripeMapHeader();
225 RF_ASSERT(numStripes == totStripes);
226 asm_hdr->numStripes = numStripes;
227 asm_hdr->stripeMap = asm_list;
228
229 if (rf_mapDebug)
230 rf_PrintAccessStripeMap(asm_hdr);
231 return(asm_hdr);
232 }
233
234 /*****************************************************************************************
235 * This routine walks through an ASM list and marks the PDAs that have failed.
236 * It's called only when a disk failure causes an in-flight DAG to fail.
237 * The parity may consist of two components, but we want to use only one failedPDA
238 * pointer. Thus we set failedPDA to point to the first parity component, and rely
239 * on the rest of the code to do the right thing with this.
240 ****************************************************************************************/
241
242 void rf_MarkFailuresInASMList(raidPtr, asm_h)
243 RF_Raid_t *raidPtr;
244 RF_AccessStripeMapHeader_t *asm_h;
245 {
246 RF_RaidDisk_t **disks = raidPtr->Disks;
247 RF_AccessStripeMap_t *asmap;
248 RF_PhysDiskAddr_t *pda;
249
250 for (asmap = asm_h->stripeMap; asmap; asmap = asmap->next) {
251 asmap->numDataFailed = asmap->numParityFailed = asmap->numQFailed = 0;
252 asmap->numFailedPDAs = 0;
253 bzero((char *)asmap->failedPDAs,
254 RF_MAX_FAILED_PDA*sizeof(RF_PhysDiskAddr_t *));
255 for (pda = asmap->physInfo; pda; pda=pda->next) {
256 if (RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
257 printf("DEAD DISK BOGUSLY DETECTED!!\n");
258 asmap->numDataFailed++;
259 asmap->failedPDAs[asmap->numFailedPDAs] = pda;
260 asmap->numFailedPDAs++;
261 }
262 }
263 pda = asmap->parityInfo;
264 if (pda && RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
265 asmap->numParityFailed++;
266 asmap->failedPDAs[asmap->numFailedPDAs] = pda;
267 asmap->numFailedPDAs++;
268 }
269 pda = asmap->qInfo;
270 if (pda && RF_DEAD_DISK(disks[pda->row][pda->col].status)) {
271 asmap->numQFailed++;
272 asmap->failedPDAs[asmap->numFailedPDAs] = pda;
273 asmap->numFailedPDAs++;
274 }
275 }
276 }
277
278 /*****************************************************************************************
279 *
280 * DuplicateASM -- duplicates an ASM and returns the new one
281 *
282 ****************************************************************************************/
283 RF_AccessStripeMap_t *rf_DuplicateASM(asmap)
284 RF_AccessStripeMap_t *asmap;
285 {
286 RF_AccessStripeMap_t *new_asm;
287 RF_PhysDiskAddr_t *pda, *new_pda, *t_pda;
288
289 new_pda = NULL;
290 new_asm = rf_AllocAccessStripeMapComponent();
291 bcopy((char *)asmap, (char *)new_asm, sizeof(RF_AccessStripeMap_t));
292 new_asm->numFailedPDAs = 0; /* ??? */
293 new_asm->failedPDAs[0] = NULL;
294 new_asm->physInfo = NULL;
295 new_asm->parityInfo = NULL;
296 new_asm->next = NULL;
297
298 for (pda = asmap->physInfo; pda; pda=pda->next) { /* copy the physInfo list */
299 t_pda = rf_AllocPhysDiskAddr();
300 bcopy((char *)pda, (char *)t_pda, sizeof(RF_PhysDiskAddr_t));
301 t_pda->next = NULL;
302 if (!new_asm->physInfo) {new_asm->physInfo = t_pda; new_pda = t_pda;}
303 else {new_pda->next = t_pda; new_pda = new_pda->next;}
304 if (pda == asmap->failedPDAs[0])
305 new_asm->failedPDAs[0] = t_pda;
306 }
307 for (pda = asmap->parityInfo; pda; pda=pda->next) { /* copy the parityInfo list */
308 t_pda = rf_AllocPhysDiskAddr();
309 bcopy((char *)pda, (char *)t_pda, sizeof(RF_PhysDiskAddr_t));
310 t_pda->next = NULL;
311 if (!new_asm->parityInfo) {new_asm->parityInfo = t_pda; new_pda = t_pda;}
312 else {new_pda->next = t_pda; new_pda = new_pda->next;}
313 if (pda == asmap->failedPDAs[0])
314 new_asm->failedPDAs[0] = t_pda;
315 }
316 return(new_asm);
317 }
318
319 /*****************************************************************************************
320 *
321 * DuplicatePDA -- duplicates a PDA and returns the new one
322 *
323 ****************************************************************************************/
324 RF_PhysDiskAddr_t *rf_DuplicatePDA(pda)
325 RF_PhysDiskAddr_t *pda;
326 {
327 RF_PhysDiskAddr_t *new;
328
329 new = rf_AllocPhysDiskAddr();
330 bcopy((char *)pda, (char *)new, sizeof(RF_PhysDiskAddr_t));
331 return(new);
332 }
333
334 /*****************************************************************************************
335 *
336 * routines to allocate and free list elements. All allocation routines zero the
337 * structure before returning it.
338 *
339 * FreePhysDiskAddr is static. It should never be called directly, because
340 * FreeAccessStripeMap takes care of freeing the PhysDiskAddr list.
341 *
342 ****************************************************************************************/
343
344 static RF_FreeList_t *rf_asmhdr_freelist;
345 #define RF_MAX_FREE_ASMHDR 128
346 #define RF_ASMHDR_INC 16
347 #define RF_ASMHDR_INITIAL 32
348
349 static RF_FreeList_t *rf_asm_freelist;
350 #define RF_MAX_FREE_ASM 192
351 #define RF_ASM_INC 24
352 #define RF_ASM_INITIAL 64
353
354 static RF_FreeList_t *rf_pda_freelist;
355 #define RF_MAX_FREE_PDA 192
356 #define RF_PDA_INC 24
357 #define RF_PDA_INITIAL 64
358
359 /* called at shutdown time. So far, all that is necessary is to release all the free lists */
360 static void rf_ShutdownMapModule(void *);
361 static void rf_ShutdownMapModule(ignored)
362 void *ignored;
363 {
364 RF_FREELIST_DESTROY(rf_asmhdr_freelist,next,(RF_AccessStripeMapHeader_t *));
365 RF_FREELIST_DESTROY(rf_pda_freelist,next,(RF_PhysDiskAddr_t *));
366 RF_FREELIST_DESTROY(rf_asm_freelist,next,(RF_AccessStripeMap_t *));
367 }
368
369 int rf_ConfigureMapModule(listp)
370 RF_ShutdownList_t **listp;
371 {
372 int rc;
373
374 RF_FREELIST_CREATE(rf_asmhdr_freelist, RF_MAX_FREE_ASMHDR,
375 RF_ASMHDR_INC, sizeof(RF_AccessStripeMapHeader_t));
376 if (rf_asmhdr_freelist == NULL) {
377 return(ENOMEM);
378 }
379 RF_FREELIST_CREATE(rf_asm_freelist, RF_MAX_FREE_ASM,
380 RF_ASM_INC, sizeof(RF_AccessStripeMap_t));
381 if (rf_asm_freelist == NULL) {
382 RF_FREELIST_DESTROY(rf_asmhdr_freelist,next,(RF_AccessStripeMapHeader_t *));
383 return(ENOMEM);
384 }
385 RF_FREELIST_CREATE(rf_pda_freelist, RF_MAX_FREE_PDA,
386 RF_PDA_INC, sizeof(RF_PhysDiskAddr_t));
387 if (rf_pda_freelist == NULL) {
388 RF_FREELIST_DESTROY(rf_asmhdr_freelist,next,(RF_AccessStripeMapHeader_t *));
389 RF_FREELIST_DESTROY(rf_pda_freelist,next,(RF_PhysDiskAddr_t *));
390 return(ENOMEM);
391 }
392
393 rc = rf_ShutdownCreate(listp, rf_ShutdownMapModule, NULL);
394 if (rc) {
395 RF_ERRORMSG3("Unable to add to shutdown list file %s line %d rc=%d\n", __FILE__,
396 __LINE__, rc);
397 rf_ShutdownMapModule(NULL);
398 return(rc);
399 }
400
401 RF_FREELIST_PRIME(rf_asmhdr_freelist, RF_ASMHDR_INITIAL,next,
402 (RF_AccessStripeMapHeader_t *));
403 RF_FREELIST_PRIME(rf_asm_freelist, RF_ASM_INITIAL,next,
404 (RF_AccessStripeMap_t *));
405 RF_FREELIST_PRIME(rf_pda_freelist, RF_PDA_INITIAL,next,
406 (RF_PhysDiskAddr_t *));
407
408 return(0);
409 }
410
411 RF_AccessStripeMapHeader_t *rf_AllocAccessStripeMapHeader()
412 {
413 RF_AccessStripeMapHeader_t *p;
414
415 RF_FREELIST_GET(rf_asmhdr_freelist,p,next,(RF_AccessStripeMapHeader_t *));
416 bzero((char *)p, sizeof(RF_AccessStripeMapHeader_t));
417
418 return(p);
419 }
420
421
422 void rf_FreeAccessStripeMapHeader(p)
423 RF_AccessStripeMapHeader_t *p;
424 {
425 RF_FREELIST_FREE(rf_asmhdr_freelist,p,next);
426 }
427
428 RF_PhysDiskAddr_t *rf_AllocPhysDiskAddr()
429 {
430 RF_PhysDiskAddr_t *p;
431
432 RF_FREELIST_GET(rf_pda_freelist,p,next,(RF_PhysDiskAddr_t *));
433 bzero((char *)p, sizeof(RF_PhysDiskAddr_t));
434
435 return(p);
436 }
437
438 /* allocates a list of PDAs, locking the free list only once
439 * when we have to call calloc, we do it one component at a time to simplify
440 * the process of freeing the list at program shutdown. This should not be
441 * much of a performance hit, because it should be very infrequently executed.
442 */
443 RF_PhysDiskAddr_t *rf_AllocPDAList(count)
444 int count;
445 {
446 RF_PhysDiskAddr_t *p = NULL;
447
448 RF_FREELIST_GET_N(rf_pda_freelist,p,next,(RF_PhysDiskAddr_t *),count);
449 return(p);
450 }
451
452 void rf_FreePhysDiskAddr(p)
453 RF_PhysDiskAddr_t *p;
454 {
455 RF_FREELIST_FREE(rf_pda_freelist,p,next);
456 }
457
458 static void rf_FreePDAList(l_start, l_end, count)
459 RF_PhysDiskAddr_t *l_start, *l_end; /* pointers to start and end of list */
460 int count; /* number of elements in list */
461 {
462 RF_FREELIST_FREE_N(rf_pda_freelist,l_start,next,(RF_PhysDiskAddr_t *),count);
463 }
464
465 RF_AccessStripeMap_t *rf_AllocAccessStripeMapComponent()
466 {
467 RF_AccessStripeMap_t *p;
468
469 RF_FREELIST_GET(rf_asm_freelist,p,next,(RF_AccessStripeMap_t *));
470 bzero((char *)p, sizeof(RF_AccessStripeMap_t));
471
472 return(p);
473 }
474
475 /* this is essentially identical to AllocPDAList. I should combine the two.
476 * when we have to call calloc, we do it one component at a time to simplify
477 * the process of freeing the list at program shutdown. This should not be
478 * much of a performance hit, because it should be very infrequently executed.
479 */
480 RF_AccessStripeMap_t *rf_AllocASMList(count)
481 int count;
482 {
483 RF_AccessStripeMap_t *p = NULL;
484
485 RF_FREELIST_GET_N(rf_asm_freelist,p,next,(RF_AccessStripeMap_t *),count);
486 return(p);
487 }
488
489 void rf_FreeAccessStripeMapComponent(p)
490 RF_AccessStripeMap_t *p;
491 {
492 RF_FREELIST_FREE(rf_asm_freelist,p,next);
493 }
494
495 static void rf_FreeASMList(l_start, l_end, count)
496 RF_AccessStripeMap_t *l_start, *l_end;
497 int count;
498 {
499 RF_FREELIST_FREE_N(rf_asm_freelist,l_start,next,(RF_AccessStripeMap_t *),count);
500 }
501
502 void rf_FreeAccessStripeMap(hdr)
503 RF_AccessStripeMapHeader_t *hdr;
504 {
505 RF_AccessStripeMap_t *p, *pt = NULL;
506 RF_PhysDiskAddr_t *pdp, *trailer, *pdaList = NULL, *pdaEnd = NULL;
507 int count = 0, t, asm_count = 0;
508
509 for (p = hdr->stripeMap; p; p=p->next) {
510
511 /* link the 3 pda lists into the accumulating pda list */
512
513 if (!pdaList) pdaList = p->qInfo; else pdaEnd->next = p->qInfo;
514 for (trailer=NULL,pdp=p->qInfo; pdp; ) {trailer = pdp; pdp=pdp->next; count++;}
515 if (trailer) pdaEnd = trailer;
516
517 if (!pdaList) pdaList = p->parityInfo; else pdaEnd->next = p->parityInfo;
518 for (trailer=NULL,pdp=p->parityInfo; pdp; ) {trailer = pdp; pdp=pdp->next; count++;}
519 if (trailer) pdaEnd = trailer;
520
521 if (!pdaList) pdaList = p->physInfo; else pdaEnd->next = p->physInfo;
522 for (trailer=NULL,pdp=p->physInfo; pdp; ) {trailer = pdp; pdp=pdp->next; count++;}
523 if (trailer) pdaEnd = trailer;
524
525 pt = p;
526 asm_count++;
527 }
528
529 /* debug only */
530 for (t=0,pdp=pdaList; pdp; pdp=pdp->next)
531 t++;
532 RF_ASSERT(t == count);
533
534 if (pdaList)
535 rf_FreePDAList(pdaList, pdaEnd, count);
536 rf_FreeASMList(hdr->stripeMap, pt, asm_count);
537 rf_FreeAccessStripeMapHeader(hdr);
538 }
539
540 /* We can't use the large write optimization if there are any failures in the stripe.
541 * In the declustered layout, there is no way to immediately determine what disks
542 * constitute a stripe, so we actually have to hunt through the stripe looking for failures.
543 * The reason we map the parity instead of just using asm->parityInfo->col is because
544 * the latter may have been already redirected to a spare drive, which would
545 * mess up the computation of the stripe offset.
546 *
547 * ASSUMES AT MOST ONE FAILURE IN THE STRIPE.
548 */
549 int rf_CheckStripeForFailures(raidPtr, asmap)
550 RF_Raid_t *raidPtr;
551 RF_AccessStripeMap_t *asmap;
552 {
553 RF_RowCol_t trow, tcol, prow, pcol, *diskids, row, i;
554 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
555 RF_StripeCount_t stripeOffset;
556 int numFailures;
557 RF_RaidAddr_t sosAddr;
558 RF_SectorNum_t diskOffset, poffset;
559 RF_RowCol_t testrow;
560
561 /* quick out in the fault-free case. */
562 RF_LOCK_MUTEX(raidPtr->mutex);
563 numFailures = raidPtr->numFailures;
564 RF_UNLOCK_MUTEX(raidPtr->mutex);
565 if (numFailures == 0) return(0);
566
567 sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
568 row = asmap->physInfo->row;
569 (layoutPtr->map->IdentifyStripe)(raidPtr, asmap->raidAddress, &diskids, &testrow);
570 (layoutPtr->map->MapParity)(raidPtr, asmap->raidAddress, &prow, &pcol, &poffset, 0); /* get pcol */
571
572 /* this need not be true if we've redirected the access to a spare in another row
573 RF_ASSERT(row == testrow);
574 */
575 stripeOffset = 0;
576 for (i=0; i<layoutPtr->numDataCol+layoutPtr->numParityCol; i++) {
577 if (diskids[i] != pcol) {
578 if (RF_DEAD_DISK(raidPtr->Disks[testrow][diskids[i]].status)) {
579 if (raidPtr->status[testrow] != rf_rs_reconstructing)
580 return(1);
581 RF_ASSERT(raidPtr->reconControl[testrow]->fcol == diskids[i]);
582 layoutPtr->map->MapSector(raidPtr,
583 sosAddr + stripeOffset * layoutPtr->sectorsPerStripeUnit,
584 &trow, &tcol, &diskOffset, 0);
585 RF_ASSERT( (trow == testrow) && (tcol == diskids[i]) );
586 if (!rf_CheckRUReconstructed(raidPtr->reconControl[testrow]->reconMap, diskOffset))
587 return(1);
588 asmap->flags |= RF_ASM_REDIR_LARGE_WRITE;
589 return(0);
590 }
591 stripeOffset++;
592 }
593 }
594 return(0);
595 }
596
597 /*
598 return the number of failed data units in the stripe.
599 */
600
601 int rf_NumFailedDataUnitsInStripe(raidPtr, asmap)
602 RF_Raid_t *raidPtr;
603 RF_AccessStripeMap_t *asmap;
604 {
605 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
606 RF_RowCol_t trow, tcol, row, i;
607 RF_SectorNum_t diskOffset;
608 RF_RaidAddr_t sosAddr;
609 int numFailures;
610
611 /* quick out in the fault-free case. */
612 RF_LOCK_MUTEX(raidPtr->mutex);
613 numFailures = raidPtr->numFailures;
614 RF_UNLOCK_MUTEX(raidPtr->mutex);
615 if (numFailures == 0) return(0);
616 numFailures = 0;
617
618 sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, asmap->raidAddress);
619 row = asmap->physInfo->row;
620 for (i=0; i<layoutPtr->numDataCol; i++)
621 {
622 (layoutPtr->map->MapSector)(raidPtr, sosAddr + i * layoutPtr->sectorsPerStripeUnit,
623 &trow, &tcol, &diskOffset, 0);
624 if (RF_DEAD_DISK(raidPtr->Disks[trow][tcol].status))
625 numFailures++;
626 }
627
628 return numFailures;
629 }
630
631
632 /*****************************************************************************************
633 *
634 * debug routines
635 *
636 ****************************************************************************************/
637
638 void rf_PrintAccessStripeMap(asm_h)
639 RF_AccessStripeMapHeader_t *asm_h;
640 {
641 rf_PrintFullAccessStripeMap(asm_h, 0);
642 }
643
644 void 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: %d\n",(int)asmap->totalSectorsAccessed); */
655 printf("Stripe %d (%d sectors), failures: %d data, %d parity: ",
656 (int) asmap->stripeID,
657 (int) asmap->totalSectorsAccessed,
658 (int) asmap->numDataFailed,
659 (int) asmap->numParityFailed);
660 if (asmap->parityInfo) {
661 printf("Parity [r%d c%d s%d-%d", asmap->parityInfo->row, asmap->parityInfo->col,
662 (int)asmap->parityInfo->startSector,
663 (int)(asmap->parityInfo->startSector +
664 asmap->parityInfo->numSector - 1));
665 if (prbuf) printf(" b0x%lx",(unsigned long) asmap->parityInfo->bufPtr);
666 if (asmap->parityInfo->next) {
667 printf(", r%d c%d s%d-%d", asmap->parityInfo->next->row,
668 asmap->parityInfo->next->col,
669 (int) asmap->parityInfo->next->startSector,
670 (int)(asmap->parityInfo->next->startSector +
671 asmap->parityInfo->next->numSector - 1));
672 if (prbuf) printf(" b0x%lx",(unsigned long) asmap->parityInfo->next->bufPtr);
673 RF_ASSERT(asmap->parityInfo->next->next == NULL);
674 }
675 printf("]\n\t");
676 }
677 for (i=0,p=asmap->physInfo; p; p=p->next,i++) {
678 printf("SU r%d c%d s%d-%d ", p->row, p->col, (int)p->startSector,
679 (int)(p->startSector + p->numSector - 1));
680 if (prbuf) printf("b0x%lx ", (unsigned long) p->bufPtr);
681 if (i && !(i&1)) printf("\n\t");
682 }
683 printf("\n");
684 p = asm_h->stripeMap->failedPDAs[0];
685 if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 1) printf("[multiple failures]\n");
686 else if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 0)
687 printf("\t[Failed PDA: r%d c%d s%d-%d]\n",p->row, p->col,
688 (int)p->startSector, (int)(p->startSector + p->numSector-1));
689 }
690 }
691
692 void rf_PrintRaidAddressInfo(raidPtr, raidAddr, numBlocks)
693 RF_Raid_t *raidPtr;
694 RF_RaidAddr_t raidAddr;
695 RF_SectorCount_t numBlocks;
696 {
697 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
698 RF_RaidAddr_t ra, sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
699
700 printf("Raid addrs of SU boundaries from start of stripe to end of access:\n\t");
701 for (ra = sosAddr; ra <= raidAddr + numBlocks; ra += layoutPtr->sectorsPerStripeUnit) {
702 printf("%d (0x%x), ",(int)ra, (int)ra);
703 }
704 printf("\n");
705 printf("Offset into stripe unit: %d (0x%x)\n",
706 (int)(raidAddr % layoutPtr->sectorsPerStripeUnit),
707 (int)(raidAddr % layoutPtr->sectorsPerStripeUnit));
708 }
709
710 /*
711 given a parity descriptor and the starting address within a stripe,
712 range restrict the parity descriptor to touch only the correct stuff.
713 */
714 void rf_ASMParityAdjust(
715 RF_PhysDiskAddr_t *toAdjust,
716 RF_StripeNum_t startAddrWithinStripe,
717 RF_SectorNum_t endAddress,
718 RF_RaidLayout_t *layoutPtr,
719 RF_AccessStripeMap_t *asm_p)
720 {
721 RF_PhysDiskAddr_t *new_pda;
722
723 /* when we're accessing only a portion of one stripe unit, we want the parity descriptor
724 * to identify only the chunk of parity associated with the data. When the access spans
725 * exactly one stripe unit boundary and is less than a stripe unit in size, it uses two disjoint
726 * regions of the parity unit. When an access spans more than one stripe unit boundary, it
727 * uses all of the parity unit.
728 *
729 * To better handle the case where stripe units are small, we may eventually want to change
730 * the 2nd case so that if the SU size is below some threshold, we just read/write the whole
731 * thing instead of breaking it up into two accesses.
732 */
733 if (asm_p->numStripeUnitsAccessed == 1)
734 {
735 int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
736 toAdjust->startSector += x;
737 toAdjust->raidAddress += x;
738 toAdjust->numSector = asm_p->physInfo->numSector;
739 RF_ASSERT(toAdjust->numSector != 0);
740 }
741 else
742 if (asm_p->numStripeUnitsAccessed == 2 && asm_p->totalSectorsAccessed < layoutPtr->sectorsPerStripeUnit)
743 {
744 int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
745
746 /* create a second pda and copy the parity map info into it */
747 RF_ASSERT(toAdjust->next == NULL);
748 new_pda = toAdjust->next = rf_AllocPhysDiskAddr();
749 *new_pda = *toAdjust; /* structure assignment */
750 new_pda->next = NULL;
751
752 /* adjust the start sector & number of blocks for the first parity pda */
753 toAdjust->startSector += x;
754 toAdjust->raidAddress += x;
755 toAdjust->numSector = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, startAddrWithinStripe) - startAddrWithinStripe;
756 RF_ASSERT(toAdjust->numSector != 0);
757
758 /* adjust the second pda */
759 new_pda->numSector = endAddress - rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, endAddress);
760 /*new_pda->raidAddress = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, toAdjust->raidAddress);*/
761 RF_ASSERT(new_pda->numSector != 0);
762 }
763 }
764
765 /*
766 Check if a disk has been spared or failed. If spared,
767 redirect the I/O.
768 If it has been failed, record it in the asm pointer.
769 Fourth arg is whether data or parity.
770 */
771 void rf_ASMCheckStatus(
772 RF_Raid_t *raidPtr,
773 RF_PhysDiskAddr_t *pda_p,
774 RF_AccessStripeMap_t *asm_p,
775 RF_RaidDisk_t **disks,
776 int parity)
777 {
778 RF_DiskStatus_t dstatus;
779 RF_RowCol_t frow, fcol;
780
781 dstatus = disks[pda_p->row][pda_p->col].status;
782
783 if (dstatus == rf_ds_spared) {
784 /* if the disk has been spared, redirect access to the spare */
785 frow = pda_p->row; fcol = pda_p->col;
786 pda_p->row = disks[frow][fcol].spareRow;
787 pda_p->col = disks[frow][fcol].spareCol;
788 }
789 else if (dstatus == rf_ds_dist_spared) {
790 /* ditto if disk has been spared to dist spare space */
791 RF_RowCol_t or = pda_p->row, oc=pda_p->col;
792 RF_SectorNum_t oo = pda_p->startSector;
793
794 if (pda_p -> type == RF_PDA_TYPE_DATA)
795 raidPtr->Layout.map->MapSector(raidPtr, pda_p->raidAddress, &pda_p->row, &pda_p->col, &pda_p->startSector, RF_REMAP);
796 else
797 raidPtr->Layout.map->MapParity(raidPtr, pda_p->raidAddress, &pda_p->row, &pda_p->col, &pda_p->startSector, RF_REMAP);
798
799 if (rf_mapDebug) {
800 printf("Redirected r %d c %d o %d -> r%d c %d o %d\n",or,oc,(int)oo,
801 pda_p->row,pda_p->col,(int)pda_p->startSector);
802 }
803 } else if (RF_DEAD_DISK(dstatus)) {
804 /* if the disk is inaccessible, mark the failure */
805 if (parity)
806 asm_p->numParityFailed++;
807 else {
808 asm_p->numDataFailed++;
809 #if 0
810 /* XXX Do we really want this spewing out on the console? GO */
811 printf("DATA_FAILED!\n");
812 #endif
813 }
814 asm_p->failedPDAs[asm_p->numFailedPDAs] = pda_p;
815 asm_p->numFailedPDAs++;
816 #if 0
817 switch (asm_p->numParityFailed + asm_p->numDataFailed)
818 {
819 case 1:
820 asm_p->failedPDAs[0] = pda_p;
821 break;
822 case 2:
823 asm_p->failedPDAs[1] = pda_p;
824 default:
825 break;
826 }
827 #endif
828 }
829 /* the redirected access should never span a stripe unit boundary */
830 RF_ASSERT(rf_RaidAddressToStripeUnitID(&raidPtr->Layout,pda_p->raidAddress) ==
831 rf_RaidAddressToStripeUnitID(&raidPtr->Layout,pda_p->raidAddress + pda_p->numSector -1));
832 RF_ASSERT(pda_p->col != -1);
833 }
834