rf_paritylogging.c revision 1.13 1 /* $NetBSD: rf_paritylogging.c,v 1.13 2002/09/14 17:53:58 oster Exp $ */
2 /*
3 * Copyright (c) 1995 Carnegie-Mellon University.
4 * All rights reserved.
5 *
6 * Author: William V. Courtright II
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 parity logging configuration, dag selection, and mapping is implemented here
32 */
33
34 #include <sys/cdefs.h>
35 __KERNEL_RCSID(0, "$NetBSD: rf_paritylogging.c,v 1.13 2002/09/14 17:53:58 oster Exp $");
36
37 #include "rf_archs.h"
38
39 #if RF_INCLUDE_PARITYLOGGING > 0
40
41 #include <dev/raidframe/raidframevar.h>
42
43 #include "rf_raid.h"
44 #include "rf_dag.h"
45 #include "rf_dagutils.h"
46 #include "rf_dagfuncs.h"
47 #include "rf_dagffrd.h"
48 #include "rf_dagffwr.h"
49 #include "rf_dagdegrd.h"
50 #include "rf_dagdegwr.h"
51 #include "rf_paritylog.h"
52 #include "rf_paritylogDiskMgr.h"
53 #include "rf_paritylogging.h"
54 #include "rf_parityloggingdags.h"
55 #include "rf_general.h"
56 #include "rf_map.h"
57 #include "rf_utils.h"
58 #include "rf_shutdown.h"
59
60 typedef struct RF_ParityLoggingConfigInfo_s {
61 RF_RowCol_t **stripeIdentifier; /* filled in at config time & used by
62 * IdentifyStripe */
63 } RF_ParityLoggingConfigInfo_t;
64
65 static void FreeRegionInfo(RF_Raid_t * raidPtr, RF_RegionId_t regionID);
66 static void rf_ShutdownParityLogging(RF_ThreadArg_t arg);
67 static void rf_ShutdownParityLoggingRegionInfo(RF_ThreadArg_t arg);
68 static void rf_ShutdownParityLoggingPool(RF_ThreadArg_t arg);
69 static void rf_ShutdownParityLoggingRegionBufferPool(RF_ThreadArg_t arg);
70 static void rf_ShutdownParityLoggingParityBufferPool(RF_ThreadArg_t arg);
71 static void rf_ShutdownParityLoggingDiskQueue(RF_ThreadArg_t arg);
72
73 int
74 rf_ConfigureParityLogging(
75 RF_ShutdownList_t ** listp,
76 RF_Raid_t * raidPtr,
77 RF_Config_t * cfgPtr)
78 {
79 int i, j, startdisk, rc;
80 RF_SectorCount_t totalLogCapacity, fragmentation, lastRegionCapacity;
81 RF_SectorCount_t parityBufferCapacity, maxRegionParityRange;
82 RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
83 RF_ParityLoggingConfigInfo_t *info;
84 RF_ParityLog_t *l = NULL, *next;
85 caddr_t lHeapPtr;
86
87 if (rf_numParityRegions <= 0)
88 return(EINVAL);
89
90 /*
91 * We create multiple entries on the shutdown list here, since
92 * this configuration routine is fairly complicated in and of
93 * itself, and this makes backing out of a failed configuration
94 * much simpler.
95 */
96
97 raidPtr->numSectorsPerLog = RF_DEFAULT_NUM_SECTORS_PER_LOG;
98
99 /* create a parity logging configuration structure */
100 RF_MallocAndAdd(info, sizeof(RF_ParityLoggingConfigInfo_t),
101 (RF_ParityLoggingConfigInfo_t *),
102 raidPtr->cleanupList);
103 if (info == NULL)
104 return (ENOMEM);
105 layoutPtr->layoutSpecificInfo = (void *) info;
106
107 RF_ASSERT(raidPtr->numRow == 1);
108
109 /* the stripe identifier must identify the disks in each stripe, IN
110 * THE ORDER THAT THEY APPEAR IN THE STRIPE. */
111 info->stripeIdentifier = rf_make_2d_array((raidPtr->numCol),
112 (raidPtr->numCol),
113 raidPtr->cleanupList);
114 if (info->stripeIdentifier == NULL)
115 return (ENOMEM);
116
117 startdisk = 0;
118 for (i = 0; i < (raidPtr->numCol); i++) {
119 for (j = 0; j < (raidPtr->numCol); j++) {
120 info->stripeIdentifier[i][j] = (startdisk + j) %
121 (raidPtr->numCol - 1);
122 }
123 if ((--startdisk) < 0)
124 startdisk = raidPtr->numCol - 1 - 1;
125 }
126
127 /* fill in the remaining layout parameters */
128 layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk;
129 layoutPtr->bytesPerStripeUnit = layoutPtr->sectorsPerStripeUnit <<
130 raidPtr->logBytesPerSector;
131 layoutPtr->numParityCol = 1;
132 layoutPtr->numParityLogCol = 1;
133 layoutPtr->numDataCol = raidPtr->numCol - layoutPtr->numParityCol -
134 layoutPtr->numParityLogCol;
135 layoutPtr->dataSectorsPerStripe = layoutPtr->numDataCol *
136 layoutPtr->sectorsPerStripeUnit;
137 layoutPtr->dataStripeUnitsPerDisk = layoutPtr->stripeUnitsPerDisk;
138 raidPtr->sectorsPerDisk = layoutPtr->stripeUnitsPerDisk *
139 layoutPtr->sectorsPerStripeUnit;
140
141 raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk *
142 layoutPtr->numDataCol * layoutPtr->sectorsPerStripeUnit;
143
144 /* configure parity log parameters
145 *
146 * parameter comment/constraints
147 * -------------------------------------------
148 * numParityRegions* all regions (except possibly last)
149 * of equal size
150 * totalInCoreLogCapacity* amount of memory in bytes available
151 * for in-core logs (default 1 MB)
152 * numSectorsPerLog# capacity of an in-core log in sectors
153 * (1 * disk track)
154 * numParityLogs total number of in-core logs,
155 * should be at least numParityRegions
156 * regionLogCapacity size of a region log (except possibly
157 * last one) in sectors
158 * totalLogCapacity total amount of log space in sectors
159 *
160 * where '*' denotes a user settable parameter.
161 * Note that logs are fixed to be the size of a disk track,
162 * value #defined in rf_paritylog.h
163 *
164 */
165
166 totalLogCapacity = layoutPtr->stripeUnitsPerDisk * layoutPtr->sectorsPerStripeUnit * layoutPtr->numParityLogCol;
167 raidPtr->regionLogCapacity = totalLogCapacity / rf_numParityRegions;
168 if (rf_parityLogDebug)
169 printf("bytes per sector %d\n", raidPtr->bytesPerSector);
170
171 /* reduce fragmentation within a disk region by adjusting the number
172 * of regions in an attempt to allow an integral number of logs to fit
173 * into a disk region */
174 fragmentation = raidPtr->regionLogCapacity % raidPtr->numSectorsPerLog;
175 if (fragmentation > 0)
176 for (i = 1; i < (raidPtr->numSectorsPerLog / 2); i++) {
177 if (((totalLogCapacity / (rf_numParityRegions + i)) %
178 raidPtr->numSectorsPerLog) < fragmentation) {
179 rf_numParityRegions++;
180 raidPtr->regionLogCapacity = totalLogCapacity /
181 rf_numParityRegions;
182 fragmentation = raidPtr->regionLogCapacity %
183 raidPtr->numSectorsPerLog;
184 }
185 if (((totalLogCapacity / (rf_numParityRegions - i)) %
186 raidPtr->numSectorsPerLog) < fragmentation) {
187 rf_numParityRegions--;
188 raidPtr->regionLogCapacity = totalLogCapacity /
189 rf_numParityRegions;
190 fragmentation = raidPtr->regionLogCapacity %
191 raidPtr->numSectorsPerLog;
192 }
193 }
194 /* ensure integral number of regions per log */
195 raidPtr->regionLogCapacity = (raidPtr->regionLogCapacity /
196 raidPtr->numSectorsPerLog) *
197 raidPtr->numSectorsPerLog;
198
199 raidPtr->numParityLogs = rf_totalInCoreLogCapacity /
200 (raidPtr->bytesPerSector * raidPtr->numSectorsPerLog);
201 /* to avoid deadlock, must ensure that enough logs exist for each
202 * region to have one simultaneously */
203 if (raidPtr->numParityLogs < rf_numParityRegions)
204 raidPtr->numParityLogs = rf_numParityRegions;
205
206 /* create region information structs */
207 printf("Allocating %d bytes for in-core parity region info\n",
208 (int) (rf_numParityRegions * sizeof(RF_RegionInfo_t)));
209 RF_Malloc(raidPtr->regionInfo,
210 (rf_numParityRegions * sizeof(RF_RegionInfo_t)),
211 (RF_RegionInfo_t *));
212 if (raidPtr->regionInfo == NULL)
213 return (ENOMEM);
214
215 /* last region may not be full capacity */
216 lastRegionCapacity = raidPtr->regionLogCapacity;
217 while ((rf_numParityRegions - 1) * raidPtr->regionLogCapacity +
218 lastRegionCapacity > totalLogCapacity)
219 lastRegionCapacity = lastRegionCapacity -
220 raidPtr->numSectorsPerLog;
221
222 raidPtr->regionParityRange = raidPtr->sectorsPerDisk /
223 rf_numParityRegions;
224 maxRegionParityRange = raidPtr->regionParityRange;
225
226 /* i can't remember why this line is in the code -wvcii 6/30/95 */
227 /* if (raidPtr->sectorsPerDisk % rf_numParityRegions > 0)
228 regionParityRange++; */
229
230 /* build pool of unused parity logs */
231 printf("Allocating %d bytes for %d parity logs\n",
232 raidPtr->numParityLogs * raidPtr->numSectorsPerLog *
233 raidPtr->bytesPerSector,
234 raidPtr->numParityLogs);
235 RF_Malloc(raidPtr->parityLogBufferHeap, raidPtr->numParityLogs *
236 raidPtr->numSectorsPerLog * raidPtr->bytesPerSector,
237 (caddr_t));
238 if (raidPtr->parityLogBufferHeap == NULL)
239 return (ENOMEM);
240 lHeapPtr = raidPtr->parityLogBufferHeap;
241 rc = rf_mutex_init(&raidPtr->parityLogPool.mutex);
242 if (rc) {
243 rf_print_unable_to_init_mutex(__FILE__, __LINE__, rc);
244 RF_Free(raidPtr->parityLogBufferHeap, raidPtr->numParityLogs *
245 raidPtr->numSectorsPerLog * raidPtr->bytesPerSector);
246 return (ENOMEM);
247 }
248 for (i = 0; i < raidPtr->numParityLogs; i++) {
249 if (i == 0) {
250 RF_Calloc(raidPtr->parityLogPool.parityLogs, 1,
251 sizeof(RF_ParityLog_t), (RF_ParityLog_t *));
252 if (raidPtr->parityLogPool.parityLogs == NULL) {
253 RF_Free(raidPtr->parityLogBufferHeap,
254 raidPtr->numParityLogs *
255 raidPtr->numSectorsPerLog *
256 raidPtr->bytesPerSector);
257 return (ENOMEM);
258 }
259 l = raidPtr->parityLogPool.parityLogs;
260 } else {
261 RF_Calloc(l->next, 1, sizeof(RF_ParityLog_t),
262 (RF_ParityLog_t *));
263 if (l->next == NULL) {
264 RF_Free(raidPtr->parityLogBufferHeap,
265 raidPtr->numParityLogs *
266 raidPtr->numSectorsPerLog *
267 raidPtr->bytesPerSector);
268 for (l = raidPtr->parityLogPool.parityLogs;
269 l;
270 l = next) {
271 next = l->next;
272 if (l->records)
273 RF_Free(l->records, (raidPtr->numSectorsPerLog * sizeof(RF_ParityLogRecord_t)));
274 RF_Free(l, sizeof(RF_ParityLog_t));
275 }
276 return (ENOMEM);
277 }
278 l = l->next;
279 }
280 l->bufPtr = lHeapPtr;
281 lHeapPtr += raidPtr->numSectorsPerLog *
282 raidPtr->bytesPerSector;
283 RF_Malloc(l->records, (raidPtr->numSectorsPerLog *
284 sizeof(RF_ParityLogRecord_t)),
285 (RF_ParityLogRecord_t *));
286 if (l->records == NULL) {
287 RF_Free(raidPtr->parityLogBufferHeap,
288 raidPtr->numParityLogs *
289 raidPtr->numSectorsPerLog *
290 raidPtr->bytesPerSector);
291 for (l = raidPtr->parityLogPool.parityLogs;
292 l;
293 l = next) {
294 next = l->next;
295 if (l->records)
296 RF_Free(l->records,
297 (raidPtr->numSectorsPerLog *
298 sizeof(RF_ParityLogRecord_t)));
299 RF_Free(l, sizeof(RF_ParityLog_t));
300 }
301 return (ENOMEM);
302 }
303 }
304 rc = rf_ShutdownCreate(listp, rf_ShutdownParityLoggingPool, raidPtr);
305 if (rc) {
306 RF_ERRORMSG3("Unable to create shutdown entry file %s line %d rc=%d\n", __FILE__,
307 __LINE__, rc);
308 rf_ShutdownParityLoggingPool(raidPtr);
309 return (rc);
310 }
311 /* build pool of region buffers */
312 rc = rf_mutex_init(&raidPtr->regionBufferPool.mutex);
313 if (rc) {
314 rf_print_unable_to_init_mutex(__FILE__, __LINE__, rc);
315 return (ENOMEM);
316 }
317 rc = rf_cond_init(&raidPtr->regionBufferPool.cond);
318 if (rc) {
319 rf_print_unable_to_init_cond(__FILE__, __LINE__, rc);
320 rf_mutex_destroy(&raidPtr->regionBufferPool.mutex);
321 return (ENOMEM);
322 }
323 raidPtr->regionBufferPool.bufferSize = raidPtr->regionLogCapacity *
324 raidPtr->bytesPerSector;
325 printf("regionBufferPool.bufferSize %d\n",
326 raidPtr->regionBufferPool.bufferSize);
327
328 /* for now, only one region at a time may be reintegrated */
329 raidPtr->regionBufferPool.totalBuffers = 1;
330
331 raidPtr->regionBufferPool.availableBuffers =
332 raidPtr->regionBufferPool.totalBuffers;
333 raidPtr->regionBufferPool.availBuffersIndex = 0;
334 raidPtr->regionBufferPool.emptyBuffersIndex = 0;
335 printf("Allocating %d bytes for regionBufferPool\n",
336 (int) (raidPtr->regionBufferPool.totalBuffers *
337 sizeof(caddr_t)));
338 RF_Malloc(raidPtr->regionBufferPool.buffers,
339 raidPtr->regionBufferPool.totalBuffers * sizeof(caddr_t),
340 (caddr_t *));
341 if (raidPtr->regionBufferPool.buffers == NULL) {
342 rf_mutex_destroy(&raidPtr->regionBufferPool.mutex);
343 rf_cond_destroy(&raidPtr->regionBufferPool.cond);
344 return (ENOMEM);
345 }
346 for (i = 0; i < raidPtr->regionBufferPool.totalBuffers; i++) {
347 printf("Allocating %d bytes for regionBufferPool#%d\n",
348 (int) (raidPtr->regionBufferPool.bufferSize *
349 sizeof(char)), i);
350 RF_Malloc(raidPtr->regionBufferPool.buffers[i],
351 raidPtr->regionBufferPool.bufferSize * sizeof(char),
352 (caddr_t));
353 if (raidPtr->regionBufferPool.buffers[i] == NULL) {
354 rf_mutex_destroy(&raidPtr->regionBufferPool.mutex);
355 rf_cond_destroy(&raidPtr->regionBufferPool.cond);
356 for (j = 0; j < i; j++) {
357 RF_Free(raidPtr->regionBufferPool.buffers[i],
358 raidPtr->regionBufferPool.bufferSize *
359 sizeof(char));
360 }
361 RF_Free(raidPtr->regionBufferPool.buffers,
362 raidPtr->regionBufferPool.totalBuffers *
363 sizeof(caddr_t));
364 return (ENOMEM);
365 }
366 printf("raidPtr->regionBufferPool.buffers[%d] = %lx\n", i,
367 (long) raidPtr->regionBufferPool.buffers[i]);
368 }
369 rc = rf_ShutdownCreate(listp,
370 rf_ShutdownParityLoggingRegionBufferPool,
371 raidPtr);
372 if (rc) {
373 RF_ERRORMSG3("Unable to create shutdown entry file %s line %d rc=%d\n", __FILE__,
374 __LINE__, rc);
375 rf_ShutdownParityLoggingRegionBufferPool(raidPtr);
376 return (rc);
377 }
378 /* build pool of parity buffers */
379 parityBufferCapacity = maxRegionParityRange;
380 rc = rf_mutex_init(&raidPtr->parityBufferPool.mutex);
381 if (rc) {
382 rf_print_unable_to_init_mutex(__FILE__, __LINE__, rc);
383 return (rc);
384 }
385 rc = rf_cond_init(&raidPtr->parityBufferPool.cond);
386 if (rc) {
387 rf_print_unable_to_init_cond(__FILE__, __LINE__, rc);
388 rf_mutex_destroy(&raidPtr->parityBufferPool.mutex);
389 return (ENOMEM);
390 }
391 raidPtr->parityBufferPool.bufferSize = parityBufferCapacity *
392 raidPtr->bytesPerSector;
393 printf("parityBufferPool.bufferSize %d\n",
394 raidPtr->parityBufferPool.bufferSize);
395
396 /* for now, only one region at a time may be reintegrated */
397 raidPtr->parityBufferPool.totalBuffers = 1;
398
399 raidPtr->parityBufferPool.availableBuffers =
400 raidPtr->parityBufferPool.totalBuffers;
401 raidPtr->parityBufferPool.availBuffersIndex = 0;
402 raidPtr->parityBufferPool.emptyBuffersIndex = 0;
403 printf("Allocating %d bytes for parityBufferPool of %d units\n",
404 (int) (raidPtr->parityBufferPool.totalBuffers *
405 sizeof(caddr_t)),
406 raidPtr->parityBufferPool.totalBuffers );
407 RF_Malloc(raidPtr->parityBufferPool.buffers,
408 raidPtr->parityBufferPool.totalBuffers * sizeof(caddr_t),
409 (caddr_t *));
410 if (raidPtr->parityBufferPool.buffers == NULL) {
411 rf_mutex_destroy(&raidPtr->parityBufferPool.mutex);
412 rf_cond_destroy(&raidPtr->parityBufferPool.cond);
413 return (ENOMEM);
414 }
415 for (i = 0; i < raidPtr->parityBufferPool.totalBuffers; i++) {
416 printf("Allocating %d bytes for parityBufferPool#%d\n",
417 (int) (raidPtr->parityBufferPool.bufferSize *
418 sizeof(char)),i);
419 RF_Malloc(raidPtr->parityBufferPool.buffers[i],
420 raidPtr->parityBufferPool.bufferSize * sizeof(char),
421 (caddr_t));
422 if (raidPtr->parityBufferPool.buffers == NULL) {
423 rf_mutex_destroy(&raidPtr->parityBufferPool.mutex);
424 rf_cond_destroy(&raidPtr->parityBufferPool.cond);
425 for (j = 0; j < i; j++) {
426 RF_Free(raidPtr->parityBufferPool.buffers[i],
427 raidPtr->regionBufferPool.bufferSize *
428 sizeof(char));
429 }
430 RF_Free(raidPtr->parityBufferPool.buffers,
431 raidPtr->regionBufferPool.totalBuffers *
432 sizeof(caddr_t));
433 return (ENOMEM);
434 }
435 printf("parityBufferPool.buffers[%d] = %lx\n", i,
436 (long) raidPtr->parityBufferPool.buffers[i]);
437 }
438 rc = rf_ShutdownCreate(listp,
439 rf_ShutdownParityLoggingParityBufferPool,
440 raidPtr);
441 if (rc) {
442 RF_ERRORMSG3("Unable to create shutdown entry file %s line %d rc=%d\n", __FILE__,
443 __LINE__, rc);
444 rf_ShutdownParityLoggingParityBufferPool(raidPtr);
445 return (rc);
446 }
447 /* initialize parityLogDiskQueue */
448 rc = rf_create_managed_mutex(listp,
449 &raidPtr->parityLogDiskQueue.mutex);
450 if (rc) {
451 rf_print_unable_to_init_mutex(__FILE__, __LINE__, rc);
452 return (rc);
453 }
454 rc = rf_create_managed_cond(listp, &raidPtr->parityLogDiskQueue.cond);
455 if (rc) {
456 rf_print_unable_to_init_cond(__FILE__, __LINE__, rc);
457 return (rc);
458 }
459 raidPtr->parityLogDiskQueue.flushQueue = NULL;
460 raidPtr->parityLogDiskQueue.reintQueue = NULL;
461 raidPtr->parityLogDiskQueue.bufHead = NULL;
462 raidPtr->parityLogDiskQueue.bufTail = NULL;
463 raidPtr->parityLogDiskQueue.reintHead = NULL;
464 raidPtr->parityLogDiskQueue.reintTail = NULL;
465 raidPtr->parityLogDiskQueue.logBlockHead = NULL;
466 raidPtr->parityLogDiskQueue.logBlockTail = NULL;
467 raidPtr->parityLogDiskQueue.reintBlockHead = NULL;
468 raidPtr->parityLogDiskQueue.reintBlockTail = NULL;
469 raidPtr->parityLogDiskQueue.freeDataList = NULL;
470 raidPtr->parityLogDiskQueue.freeCommonList = NULL;
471
472 rc = rf_ShutdownCreate(listp,
473 rf_ShutdownParityLoggingDiskQueue,
474 raidPtr);
475 if (rc) {
476 RF_ERRORMSG3("Unable to create shutdown entry file %s line %d rc=%d\n", __FILE__,
477 __LINE__, rc);
478 return (rc);
479 }
480 for (i = 0; i < rf_numParityRegions; i++) {
481 rc = rf_mutex_init(&raidPtr->regionInfo[i].mutex);
482 if (rc) {
483 rf_print_unable_to_init_mutex(__FILE__, __LINE__, rc);
484 for (j = 0; j < i; j++)
485 FreeRegionInfo(raidPtr, j);
486 RF_Free(raidPtr->regionInfo,
487 (rf_numParityRegions *
488 sizeof(RF_RegionInfo_t)));
489 return (ENOMEM);
490 }
491 rc = rf_mutex_init(&raidPtr->regionInfo[i].reintMutex);
492 if (rc) {
493 rf_print_unable_to_init_mutex(__FILE__, __LINE__, rc);
494 rf_mutex_destroy(&raidPtr->regionInfo[i].mutex);
495 for (j = 0; j < i; j++)
496 FreeRegionInfo(raidPtr, j);
497 RF_Free(raidPtr->regionInfo,
498 (rf_numParityRegions *
499 sizeof(RF_RegionInfo_t)));
500 return (ENOMEM);
501 }
502 raidPtr->regionInfo[i].reintInProgress = RF_FALSE;
503 raidPtr->regionInfo[i].regionStartAddr =
504 raidPtr->regionLogCapacity * i;
505 raidPtr->regionInfo[i].parityStartAddr =
506 raidPtr->regionParityRange * i;
507 if (i < rf_numParityRegions - 1) {
508 raidPtr->regionInfo[i].capacity =
509 raidPtr->regionLogCapacity;
510 raidPtr->regionInfo[i].numSectorsParity =
511 raidPtr->regionParityRange;
512 } else {
513 raidPtr->regionInfo[i].capacity =
514 lastRegionCapacity;
515 raidPtr->regionInfo[i].numSectorsParity =
516 raidPtr->sectorsPerDisk -
517 raidPtr->regionParityRange * i;
518 if (raidPtr->regionInfo[i].numSectorsParity >
519 maxRegionParityRange)
520 maxRegionParityRange =
521 raidPtr->regionInfo[i].numSectorsParity;
522 }
523 raidPtr->regionInfo[i].diskCount = 0;
524 RF_ASSERT(raidPtr->regionInfo[i].capacity +
525 raidPtr->regionInfo[i].regionStartAddr <=
526 totalLogCapacity);
527 RF_ASSERT(raidPtr->regionInfo[i].parityStartAddr +
528 raidPtr->regionInfo[i].numSectorsParity <=
529 raidPtr->sectorsPerDisk);
530 printf("Allocating %d bytes for region %d\n",
531 (int) (raidPtr->regionInfo[i].capacity *
532 sizeof(RF_DiskMap_t)), i);
533 RF_Malloc(raidPtr->regionInfo[i].diskMap,
534 (raidPtr->regionInfo[i].capacity *
535 sizeof(RF_DiskMap_t)),
536 (RF_DiskMap_t *));
537 if (raidPtr->regionInfo[i].diskMap == NULL) {
538 rf_mutex_destroy(&raidPtr->regionInfo[i].mutex);
539 rf_mutex_destroy(&raidPtr->regionInfo[i].reintMutex);
540 for (j = 0; j < i; j++)
541 FreeRegionInfo(raidPtr, j);
542 RF_Free(raidPtr->regionInfo,
543 (rf_numParityRegions *
544 sizeof(RF_RegionInfo_t)));
545 return (ENOMEM);
546 }
547 raidPtr->regionInfo[i].loggingEnabled = RF_FALSE;
548 raidPtr->regionInfo[i].coreLog = NULL;
549 }
550 rc = rf_ShutdownCreate(listp,
551 rf_ShutdownParityLoggingRegionInfo,
552 raidPtr);
553 if (rc) {
554 RF_ERRORMSG3("Unable to create shutdown entry file %s line %d rc=%d\n", __FILE__,
555 __LINE__, rc);
556 rf_ShutdownParityLoggingRegionInfo(raidPtr);
557 return (rc);
558 }
559 RF_ASSERT(raidPtr->parityLogDiskQueue.threadState == 0);
560 raidPtr->parityLogDiskQueue.threadState = RF_PLOG_CREATED;
561 rc = RF_CREATE_THREAD(raidPtr->pLogDiskThreadHandle,
562 rf_ParityLoggingDiskManager, raidPtr,"rf_log");
563 if (rc) {
564 raidPtr->parityLogDiskQueue.threadState = 0;
565 RF_ERRORMSG3("Unable to create parity logging disk thread file %s line %d rc=%d\n",
566 __FILE__, __LINE__, rc);
567 return (ENOMEM);
568 }
569 /* wait for thread to start */
570 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
571 while (!(raidPtr->parityLogDiskQueue.threadState & RF_PLOG_RUNNING)) {
572 RF_WAIT_COND(raidPtr->parityLogDiskQueue.cond,
573 raidPtr->parityLogDiskQueue.mutex);
574 }
575 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
576
577 rc = rf_ShutdownCreate(listp, rf_ShutdownParityLogging, raidPtr);
578 if (rc) {
579 RF_ERRORMSG1("Got rc=%d adding parity logging shutdown event\n", rc);
580 rf_ShutdownParityLogging(raidPtr);
581 return (rc);
582 }
583 if (rf_parityLogDebug) {
584 printf(" size of disk log in sectors: %d\n",
585 (int) totalLogCapacity);
586 printf(" total number of parity regions is %d\n", (int) rf_numParityRegions);
587 printf(" nominal sectors of log per parity region is %d\n", (int) raidPtr->regionLogCapacity);
588 printf(" nominal region fragmentation is %d sectors\n", (int) fragmentation);
589 printf(" total number of parity logs is %d\n", raidPtr->numParityLogs);
590 printf(" parity log size is %d sectors\n", raidPtr->numSectorsPerLog);
591 printf(" total in-core log space is %d bytes\n", (int) rf_totalInCoreLogCapacity);
592 }
593 rf_EnableParityLogging(raidPtr);
594
595 return (0);
596 }
597
598 static void
599 FreeRegionInfo(
600 RF_Raid_t * raidPtr,
601 RF_RegionId_t regionID)
602 {
603 RF_LOCK_MUTEX(raidPtr->regionInfo[regionID].mutex);
604 RF_Free(raidPtr->regionInfo[regionID].diskMap,
605 (raidPtr->regionInfo[regionID].capacity *
606 sizeof(RF_DiskMap_t)));
607 if (!rf_forceParityLogReint && raidPtr->regionInfo[regionID].coreLog) {
608 rf_ReleaseParityLogs(raidPtr,
609 raidPtr->regionInfo[regionID].coreLog);
610 raidPtr->regionInfo[regionID].coreLog = NULL;
611 } else {
612 RF_ASSERT(raidPtr->regionInfo[regionID].coreLog == NULL);
613 RF_ASSERT(raidPtr->regionInfo[regionID].diskCount == 0);
614 }
615 RF_UNLOCK_MUTEX(raidPtr->regionInfo[regionID].mutex);
616 rf_mutex_destroy(&raidPtr->regionInfo[regionID].mutex);
617 rf_mutex_destroy(&raidPtr->regionInfo[regionID].reintMutex);
618 }
619
620
621 static void
622 FreeParityLogQueue(
623 RF_Raid_t * raidPtr,
624 RF_ParityLogQueue_t * queue)
625 {
626 RF_ParityLog_t *l1, *l2;
627
628 RF_LOCK_MUTEX(queue->mutex);
629 l1 = queue->parityLogs;
630 while (l1) {
631 l2 = l1;
632 l1 = l2->next;
633 RF_Free(l2->records, (raidPtr->numSectorsPerLog *
634 sizeof(RF_ParityLogRecord_t)));
635 RF_Free(l2, sizeof(RF_ParityLog_t));
636 }
637 RF_UNLOCK_MUTEX(queue->mutex);
638 rf_mutex_destroy(&queue->mutex);
639 }
640
641
642 static void
643 FreeRegionBufferQueue(RF_RegionBufferQueue_t * queue)
644 {
645 int i;
646
647 RF_LOCK_MUTEX(queue->mutex);
648 if (queue->availableBuffers != queue->totalBuffers) {
649 printf("Attempt to free region queue which is still in use!\n");
650 RF_ASSERT(0);
651 }
652 for (i = 0; i < queue->totalBuffers; i++)
653 RF_Free(queue->buffers[i], queue->bufferSize);
654 RF_Free(queue->buffers, queue->totalBuffers * sizeof(caddr_t));
655 RF_UNLOCK_MUTEX(queue->mutex);
656 rf_mutex_destroy(&queue->mutex);
657 }
658
659 static void
660 rf_ShutdownParityLoggingRegionInfo(RF_ThreadArg_t arg)
661 {
662 RF_Raid_t *raidPtr;
663 RF_RegionId_t i;
664
665 raidPtr = (RF_Raid_t *) arg;
666 if (rf_parityLogDebug) {
667 printf("raid%d: ShutdownParityLoggingRegionInfo\n",
668 raidPtr->raidid);
669 }
670 /* free region information structs */
671 for (i = 0; i < rf_numParityRegions; i++)
672 FreeRegionInfo(raidPtr, i);
673 RF_Free(raidPtr->regionInfo, (rf_numParityRegions *
674 sizeof(raidPtr->regionInfo)));
675 raidPtr->regionInfo = NULL;
676 }
677
678 static void
679 rf_ShutdownParityLoggingPool(RF_ThreadArg_t arg)
680 {
681 RF_Raid_t *raidPtr;
682
683 raidPtr = (RF_Raid_t *) arg;
684 if (rf_parityLogDebug) {
685 printf("raid%d: ShutdownParityLoggingPool\n", raidPtr->raidid);
686 }
687 /* free contents of parityLogPool */
688 FreeParityLogQueue(raidPtr, &raidPtr->parityLogPool);
689 RF_Free(raidPtr->parityLogBufferHeap, raidPtr->numParityLogs *
690 raidPtr->numSectorsPerLog * raidPtr->bytesPerSector);
691 }
692
693 static void
694 rf_ShutdownParityLoggingRegionBufferPool(RF_ThreadArg_t arg)
695 {
696 RF_Raid_t *raidPtr;
697
698 raidPtr = (RF_Raid_t *) arg;
699 if (rf_parityLogDebug) {
700 printf("raid%d: ShutdownParityLoggingRegionBufferPool\n",
701 raidPtr->raidid);
702 }
703 FreeRegionBufferQueue(&raidPtr->regionBufferPool);
704 }
705
706 static void
707 rf_ShutdownParityLoggingParityBufferPool(RF_ThreadArg_t arg)
708 {
709 RF_Raid_t *raidPtr;
710
711 raidPtr = (RF_Raid_t *) arg;
712 if (rf_parityLogDebug) {
713 printf("raid%d: ShutdownParityLoggingParityBufferPool\n",
714 raidPtr->raidid);
715 }
716 FreeRegionBufferQueue(&raidPtr->parityBufferPool);
717 }
718
719 static void
720 rf_ShutdownParityLoggingDiskQueue(RF_ThreadArg_t arg)
721 {
722 RF_ParityLogData_t *d;
723 RF_CommonLogData_t *c;
724 RF_Raid_t *raidPtr;
725
726 raidPtr = (RF_Raid_t *) arg;
727 if (rf_parityLogDebug) {
728 printf("raid%d: ShutdownParityLoggingDiskQueue\n",
729 raidPtr->raidid);
730 }
731 /* free disk manager stuff */
732 RF_ASSERT(raidPtr->parityLogDiskQueue.bufHead == NULL);
733 RF_ASSERT(raidPtr->parityLogDiskQueue.bufTail == NULL);
734 RF_ASSERT(raidPtr->parityLogDiskQueue.reintHead == NULL);
735 RF_ASSERT(raidPtr->parityLogDiskQueue.reintTail == NULL);
736 while (raidPtr->parityLogDiskQueue.freeDataList) {
737 d = raidPtr->parityLogDiskQueue.freeDataList;
738 raidPtr->parityLogDiskQueue.freeDataList =
739 raidPtr->parityLogDiskQueue.freeDataList->next;
740 RF_Free(d, sizeof(RF_ParityLogData_t));
741 }
742 while (raidPtr->parityLogDiskQueue.freeCommonList) {
743 c = raidPtr->parityLogDiskQueue.freeCommonList;
744 rf_mutex_destroy(&c->mutex);
745 raidPtr->parityLogDiskQueue.freeCommonList =
746 raidPtr->parityLogDiskQueue.freeCommonList->next;
747 RF_Free(c, sizeof(RF_CommonLogData_t));
748 }
749 }
750
751 static void
752 rf_ShutdownParityLogging(RF_ThreadArg_t arg)
753 {
754 RF_Raid_t *raidPtr;
755
756 raidPtr = (RF_Raid_t *) arg;
757 if (rf_parityLogDebug) {
758 printf("raid%d: ShutdownParityLogging\n", raidPtr->raidid);
759 }
760 /* shutdown disk thread */
761 /* This has the desirable side-effect of forcing all regions to be
762 * reintegrated. This is necessary since all parity log maps are
763 * currently held in volatile memory. */
764
765 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
766 raidPtr->parityLogDiskQueue.threadState |= RF_PLOG_TERMINATE;
767 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
768 RF_SIGNAL_COND(raidPtr->parityLogDiskQueue.cond);
769 /*
770 * pLogDiskThread will now terminate when queues are cleared
771 * now wait for it to be done
772 */
773 RF_LOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
774 while (!(raidPtr->parityLogDiskQueue.threadState & RF_PLOG_SHUTDOWN)) {
775 RF_WAIT_COND(raidPtr->parityLogDiskQueue.cond,
776 raidPtr->parityLogDiskQueue.mutex);
777 }
778 RF_UNLOCK_MUTEX(raidPtr->parityLogDiskQueue.mutex);
779 if (rf_parityLogDebug) {
780 printf("raid%d: ShutdownParityLogging done (thread completed)\n", raidPtr->raidid);
781 }
782 }
783
784 int
785 rf_GetDefaultNumFloatingReconBuffersParityLogging(RF_Raid_t * raidPtr)
786 {
787 return (20);
788 }
789
790 RF_HeadSepLimit_t
791 rf_GetDefaultHeadSepLimitParityLogging(RF_Raid_t * raidPtr)
792 {
793 return (10);
794 }
795 /* return the region ID for a given RAID address */
796 RF_RegionId_t
797 rf_MapRegionIDParityLogging(
798 RF_Raid_t * raidPtr,
799 RF_SectorNum_t address)
800 {
801 RF_RegionId_t regionID;
802
803 /* regionID = address / (raidPtr->regionParityRange * raidPtr->Layout.numDataCol); */
804 regionID = address / raidPtr->regionParityRange;
805 if (regionID == rf_numParityRegions) {
806 /* last region may be larger than other regions */
807 regionID--;
808 }
809 RF_ASSERT(address >= raidPtr->regionInfo[regionID].parityStartAddr);
810 RF_ASSERT(address < raidPtr->regionInfo[regionID].parityStartAddr +
811 raidPtr->regionInfo[regionID].numSectorsParity);
812 RF_ASSERT(regionID < rf_numParityRegions);
813 return (regionID);
814 }
815
816
817 /* given a logical RAID sector, determine physical disk address of data */
818 void
819 rf_MapSectorParityLogging(
820 RF_Raid_t * raidPtr,
821 RF_RaidAddr_t raidSector,
822 RF_RowCol_t * row,
823 RF_RowCol_t * col,
824 RF_SectorNum_t * diskSector,
825 int remap)
826 {
827 RF_StripeNum_t SUID = raidSector /
828 raidPtr->Layout.sectorsPerStripeUnit;
829 *row = 0;
830 /* *col = (SUID % (raidPtr->numCol -
831 * raidPtr->Layout.numParityLogCol)); */
832 *col = SUID % raidPtr->Layout.numDataCol;
833 *diskSector = (SUID / (raidPtr->Layout.numDataCol)) *
834 raidPtr->Layout.sectorsPerStripeUnit +
835 (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
836 }
837
838
839 /* given a logical RAID sector, determine physical disk address of parity */
840 void
841 rf_MapParityParityLogging(
842 RF_Raid_t * raidPtr,
843 RF_RaidAddr_t raidSector,
844 RF_RowCol_t * row,
845 RF_RowCol_t * col,
846 RF_SectorNum_t * diskSector,
847 int remap)
848 {
849 RF_StripeNum_t SUID = raidSector /
850 raidPtr->Layout.sectorsPerStripeUnit;
851
852 *row = 0;
853 /* *col =
854 * raidPtr->Layout.numDataCol-(SUID/raidPtr->Layout.numDataCol)%(raidPt
855 * r->numCol - raidPtr->Layout.numParityLogCol); */
856 *col = raidPtr->Layout.numDataCol;
857 *diskSector = (SUID / (raidPtr->Layout.numDataCol)) *
858 raidPtr->Layout.sectorsPerStripeUnit +
859 (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
860 }
861
862
863 /* given a regionID and sector offset, determine the physical disk address of the parity log */
864 void
865 rf_MapLogParityLogging(
866 RF_Raid_t * raidPtr,
867 RF_RegionId_t regionID,
868 RF_SectorNum_t regionOffset,
869 RF_RowCol_t * row,
870 RF_RowCol_t * col,
871 RF_SectorNum_t * startSector)
872 {
873 *row = 0;
874 *col = raidPtr->numCol - 1;
875 *startSector = raidPtr->regionInfo[regionID].regionStartAddr + regionOffset;
876 }
877
878
879 /* given a regionID, determine the physical disk address of the logged
880 parity for that region */
881 void
882 rf_MapRegionParity(
883 RF_Raid_t * raidPtr,
884 RF_RegionId_t regionID,
885 RF_RowCol_t * row,
886 RF_RowCol_t * col,
887 RF_SectorNum_t * startSector,
888 RF_SectorCount_t * numSector)
889 {
890 *row = 0;
891 *col = raidPtr->numCol - 2;
892 *startSector = raidPtr->regionInfo[regionID].parityStartAddr;
893 *numSector = raidPtr->regionInfo[regionID].numSectorsParity;
894 }
895
896
897 /* given a logical RAID address, determine the participating disks in
898 the stripe */
899 void
900 rf_IdentifyStripeParityLogging(
901 RF_Raid_t * raidPtr,
902 RF_RaidAddr_t addr,
903 RF_RowCol_t ** diskids,
904 RF_RowCol_t * outRow)
905 {
906 RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout,
907 addr);
908 RF_ParityLoggingConfigInfo_t *info = (RF_ParityLoggingConfigInfo_t *)
909 raidPtr->Layout.layoutSpecificInfo;
910 *outRow = 0;
911 *diskids = info->stripeIdentifier[stripeID % raidPtr->numCol];
912 }
913
914
915 void
916 rf_MapSIDToPSIDParityLogging(
917 RF_RaidLayout_t * layoutPtr,
918 RF_StripeNum_t stripeID,
919 RF_StripeNum_t * psID,
920 RF_ReconUnitNum_t * which_ru)
921 {
922 *which_ru = 0;
923 *psID = stripeID;
924 }
925
926
927 /* select an algorithm for performing an access. Returns two pointers,
928 * one to a function that will return information about the DAG, and
929 * another to a function that will create the dag.
930 */
931 void
932 rf_ParityLoggingDagSelect(
933 RF_Raid_t * raidPtr,
934 RF_IoType_t type,
935 RF_AccessStripeMap_t * asmp,
936 RF_VoidFuncPtr * createFunc)
937 {
938 RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
939 RF_PhysDiskAddr_t *failedPDA = NULL;
940 RF_RowCol_t frow, fcol;
941 RF_RowStatus_t rstat;
942 int prior_recon;
943
944 RF_ASSERT(RF_IO_IS_R_OR_W(type));
945
946 if (asmp->numDataFailed + asmp->numParityFailed > 1) {
947 RF_ERRORMSG("Multiple disks failed in a single group! Aborting I/O operation.\n");
948 /* *infoFunc = */ *createFunc = NULL;
949 return;
950 } else
951 if (asmp->numDataFailed + asmp->numParityFailed == 1) {
952
953 /* if under recon & already reconstructed, redirect
954 * the access to the spare drive and eliminate the
955 * failure indication */
956 failedPDA = asmp->failedPDAs[0];
957 frow = failedPDA->row;
958 fcol = failedPDA->col;
959 rstat = raidPtr->status[failedPDA->row];
960 prior_recon = (rstat == rf_rs_reconfigured) || (
961 (rstat == rf_rs_reconstructing) ?
962 rf_CheckRUReconstructed(raidPtr->reconControl[frow]->reconMap, failedPDA->startSector) : 0
963 );
964 if (prior_recon) {
965 RF_RowCol_t or = failedPDA->row, oc = failedPDA->col;
966 RF_SectorNum_t oo = failedPDA->startSector;
967 if (layoutPtr->map->flags &
968 RF_DISTRIBUTE_SPARE) {
969 /* redirect to dist spare space */
970
971 if (failedPDA == asmp->parityInfo) {
972
973 /* parity has failed */
974 (layoutPtr->map->MapParity) (raidPtr, failedPDA->raidAddress, &failedPDA->row,
975 &failedPDA->col, &failedPDA->startSector, RF_REMAP);
976
977 if (asmp->parityInfo->next) { /* redir 2nd component,
978 * if any */
979 RF_PhysDiskAddr_t *p = asmp->parityInfo->next;
980 RF_SectorNum_t SUoffs = p->startSector % layoutPtr->sectorsPerStripeUnit;
981 p->row = failedPDA->row;
982 p->col = failedPDA->col;
983 p->startSector = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, failedPDA->startSector) +
984 SUoffs; /* cheating:
985 * startSector is not
986 * really a RAID address */
987 }
988 } else
989 if (asmp->parityInfo->next && failedPDA == asmp->parityInfo->next) {
990 RF_ASSERT(0); /* should not ever
991 * happen */
992 } else {
993
994 /* data has failed */
995 (layoutPtr->map->MapSector) (raidPtr, failedPDA->raidAddress, &failedPDA->row,
996 &failedPDA->col, &failedPDA->startSector, RF_REMAP);
997
998 }
999
1000 } else {
1001 /* redirect to dedicated spare space */
1002
1003 failedPDA->row = raidPtr->Disks[frow][fcol].spareRow;
1004 failedPDA->col = raidPtr->Disks[frow][fcol].spareCol;
1005
1006 /* the parity may have two distinct
1007 * components, both of which may need
1008 * to be redirected */
1009 if (asmp->parityInfo->next) {
1010 if (failedPDA == asmp->parityInfo) {
1011 failedPDA->next->row = failedPDA->row;
1012 failedPDA->next->col = failedPDA->col;
1013 } else
1014 if (failedPDA == asmp->parityInfo->next) { /* paranoid: should never occur */
1015 asmp->parityInfo->row = failedPDA->row;
1016 asmp->parityInfo->col = failedPDA->col;
1017 }
1018 }
1019 }
1020
1021 RF_ASSERT(failedPDA->col != -1);
1022
1023 if (rf_dagDebug || rf_mapDebug) {
1024 printf("raid%d: Redirected type '%c' r %d c %d o %ld -> r %d c %d o %ld\n",
1025 raidPtr->raidid, type, or, oc, (long) oo, failedPDA->row, failedPDA->col, (long) failedPDA->startSector);
1026 }
1027 asmp->numDataFailed = asmp->numParityFailed = 0;
1028 }
1029 }
1030 if (type == RF_IO_TYPE_READ) {
1031
1032 if (asmp->numDataFailed == 0)
1033 *createFunc = (RF_VoidFuncPtr) rf_CreateFaultFreeReadDAG;
1034 else
1035 *createFunc = (RF_VoidFuncPtr) rf_CreateRaidFiveDegradedReadDAG;
1036
1037 } else {
1038
1039
1040 /* if mirroring, always use large writes. If the access
1041 * requires two distinct parity updates, always do a small
1042 * write. If the stripe contains a failure but the access
1043 * does not, do a small write. The first conditional
1044 * (numStripeUnitsAccessed <= numDataCol/2) uses a
1045 * less-than-or-equal rather than just a less-than because
1046 * when G is 3 or 4, numDataCol/2 is 1, and I want
1047 * single-stripe-unit updates to use just one disk. */
1048 if ((asmp->numDataFailed + asmp->numParityFailed) == 0) {
1049 if (((asmp->numStripeUnitsAccessed <=
1050 (layoutPtr->numDataCol / 2)) &&
1051 (layoutPtr->numDataCol != 1)) ||
1052 (asmp->parityInfo->next != NULL) ||
1053 rf_CheckStripeForFailures(raidPtr, asmp)) {
1054 *createFunc = (RF_VoidFuncPtr) rf_CreateParityLoggingSmallWriteDAG;
1055 } else
1056 *createFunc = (RF_VoidFuncPtr) rf_CreateParityLoggingLargeWriteDAG;
1057 } else
1058 if (asmp->numParityFailed == 1)
1059 *createFunc = (RF_VoidFuncPtr) rf_CreateNonRedundantWriteDAG;
1060 else
1061 if (asmp->numStripeUnitsAccessed != 1 && failedPDA->numSector != layoutPtr->sectorsPerStripeUnit)
1062 *createFunc = NULL;
1063 else
1064 *createFunc = (RF_VoidFuncPtr) rf_CreateDegradedWriteDAG;
1065 }
1066 }
1067 #endif /* RF_INCLUDE_PARITYLOGGING > 0 */
1068