xref: /netbsd-src/sys/dev/raidframe/rf_decluster.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: rf_decluster.c,v 1.6 2001/01/26 04:40:03 oster Exp $	*/
2 /*
3  * Copyright (c) 1995 Carnegie-Mellon University.
4  * All rights reserved.
5  *
6  * Author: Mark Holland
7  *
8  * Permission to use, copy, modify and distribute this software and
9  * its documentation is hereby granted, provided that both the copyright
10  * notice and this permission notice appear in all copies of the
11  * software, derivative works or modified versions, and any portions
12  * thereof, and that both notices appear in supporting documentation.
13  *
14  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17  *
18  * Carnegie Mellon requests users of this software to return to
19  *
20  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
21  *  School of Computer Science
22  *  Carnegie Mellon University
23  *  Pittsburgh PA 15213-3890
24  *
25  * any improvements or extensions that they make and grant Carnegie the
26  * rights to redistribute these changes.
27  */
28 
29 /*----------------------------------------------------------------------
30  *
31  * rf_decluster.c -- code related to the declustered layout
32  *
33  * Created 10-21-92 (MCH)
34  *
35  * Nov 93:  adding support for distributed sparing.  This code is a little
36  *          complex:  the basic layout used is as follows:
37  *          let F = (v-1)/GCD(r,v-1).  The spare space for each set of
38  *          F consecutive fulltables is grouped together and placed after
39  *          that set of tables.
40  *                   +------------------------------+
41  *                   |        F fulltables          |
42  *                   |        Spare Space           |
43  *                   |        F fulltables          |
44  *                   |        Spare Space           |
45  *                   |            ...               |
46  *                   +------------------------------+
47  *
48  *--------------------------------------------------------------------*/
49 
50 #include "rf_archs.h"
51 #include "rf_types.h"
52 #include "rf_raid.h"
53 #include "rf_raidframe.h"
54 #include "rf_configure.h"
55 #include "rf_decluster.h"
56 #include "rf_debugMem.h"
57 #include "rf_utils.h"
58 #include "rf_alloclist.h"
59 #include "rf_general.h"
60 #include "rf_shutdown.h"
61 
62 
63 extern int rf_copyback_in_progress;	/* debug only */
64 
65 /* found in rf_kintf.c */
66 int     rf_GetSpareTableFromDaemon(RF_SparetWait_t * req);
67 
68 #if (RF_INCLUDE_PARITY_DECLUSTERING > 0) || (RF_INCLUDE_PARITY_DECLUSTERING_PQ > 0)
69 
70 /* configuration code */
71 
72 int
73 rf_ConfigureDeclustered(
74     RF_ShutdownList_t ** listp,
75     RF_Raid_t * raidPtr,
76     RF_Config_t * cfgPtr)
77 {
78 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
79 	int     b, v, k, r, lambda;	/* block design params */
80 	int     i, j;
81 	RF_RowCol_t *first_avail_slot;
82 	RF_StripeCount_t complete_FT_count, numCompleteFullTablesPerDisk;
83 	RF_DeclusteredConfigInfo_t *info;
84 	RF_StripeCount_t PUsPerDisk, spareRegionDepthInPUs, numCompleteSpareRegionsPerDisk,
85 	        extraPUsPerDisk;
86 	RF_StripeCount_t totSparePUsPerDisk;
87 	RF_SectorNum_t diskOffsetOfLastFullTableInSUs;
88 	RF_SectorCount_t SpareSpaceInSUs;
89 	char   *cfgBuf = (char *) (cfgPtr->layoutSpecific);
90 	RF_StripeNum_t l, SUID;
91 
92 	SUID = l = 0;
93 	numCompleteSpareRegionsPerDisk = 0;
94 
95 	/* 1. create layout specific structure */
96 	RF_MallocAndAdd(info, sizeof(RF_DeclusteredConfigInfo_t), (RF_DeclusteredConfigInfo_t *), raidPtr->cleanupList);
97 	if (info == NULL)
98 		return (ENOMEM);
99 	layoutPtr->layoutSpecificInfo = (void *) info;
100 	info->SpareTable = NULL;
101 
102 	/* 2. extract parameters from the config structure */
103 	if (layoutPtr->map->flags & RF_DISTRIBUTE_SPARE) {
104 		(void) bcopy(cfgBuf, info->sparemap_fname, RF_SPAREMAP_NAME_LEN);
105 	}
106 	cfgBuf += RF_SPAREMAP_NAME_LEN;
107 
108 	b = *((int *) cfgBuf);
109 	cfgBuf += sizeof(int);
110 	v = *((int *) cfgBuf);
111 	cfgBuf += sizeof(int);
112 	k = *((int *) cfgBuf);
113 	cfgBuf += sizeof(int);
114 	r = *((int *) cfgBuf);
115 	cfgBuf += sizeof(int);
116 	lambda = *((int *) cfgBuf);
117 	cfgBuf += sizeof(int);
118 	raidPtr->noRotate = *((int *) cfgBuf);
119 	cfgBuf += sizeof(int);
120 
121 	/* the sparemaps are generated assuming that parity is rotated, so we
122 	 * issue a warning if both distributed sparing and no-rotate are on at
123 	 * the same time */
124 	if ((layoutPtr->map->flags & RF_DISTRIBUTE_SPARE) && raidPtr->noRotate) {
125 		RF_ERRORMSG("Warning:  distributed sparing specified without parity rotation.\n");
126 	}
127 	if (raidPtr->numCol != v) {
128 		RF_ERRORMSG2("RAID: config error: table element count (%d) not equal to no. of cols (%d)\n", v, raidPtr->numCol);
129 		return (EINVAL);
130 	}
131 	/* 3.  set up the values used in the mapping code */
132 	info->BlocksPerTable = b;
133 	info->Lambda = lambda;
134 	info->NumParityReps = info->groupSize = k;
135 	info->SUsPerTable = b * (k - 1) * layoutPtr->SUsPerPU;	/* b blks, k-1 SUs each */
136 	info->SUsPerFullTable = k * info->SUsPerTable;	/* rot k times */
137 	info->PUsPerBlock = k - 1;
138 	info->SUsPerBlock = info->PUsPerBlock * layoutPtr->SUsPerPU;
139 	info->TableDepthInPUs = (b * k) / v;
140 	info->FullTableDepthInPUs = info->TableDepthInPUs * k;	/* k repetitions */
141 
142 	/* used only in distributed sparing case */
143 	info->FullTablesPerSpareRegion = (v - 1) / rf_gcd(r, v - 1);	/* (v-1)/gcd fulltables */
144 	info->TablesPerSpareRegion = k * info->FullTablesPerSpareRegion;
145 	info->SpareSpaceDepthPerRegionInSUs = (r * info->TablesPerSpareRegion / (v - 1)) * layoutPtr->SUsPerPU;
146 
147 	/* check to make sure the block design is sufficiently small */
148 	if ((raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
149 		if (info->FullTableDepthInPUs * layoutPtr->SUsPerPU + info->SpareSpaceDepthPerRegionInSUs > layoutPtr->stripeUnitsPerDisk) {
150 			RF_ERRORMSG3("RAID: config error: Full Table depth (%d) + Spare Space (%d) larger than disk size (%d) (BD too big)\n",
151 			    (int) info->FullTableDepthInPUs,
152 			    (int) info->SpareSpaceDepthPerRegionInSUs,
153 			    (int) layoutPtr->stripeUnitsPerDisk);
154 			return (EINVAL);
155 		}
156 	} else {
157 		if (info->TableDepthInPUs * layoutPtr->SUsPerPU > layoutPtr->stripeUnitsPerDisk) {
158 			RF_ERRORMSG2("RAID: config error: Table depth (%d) larger than disk size (%d) (BD too big)\n",
159 			    (int) (info->TableDepthInPUs * layoutPtr->SUsPerPU), \
160 			    (int) layoutPtr->stripeUnitsPerDisk);
161 			return (EINVAL);
162 		}
163 	}
164 
165 
166 	/* compute the size of each disk, and the number of tables in the last
167 	 * fulltable (which need not be complete) */
168 	if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
169 
170 		PUsPerDisk = layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerPU;
171 		spareRegionDepthInPUs = (info->TablesPerSpareRegion * info->TableDepthInPUs +
172 		    (info->TablesPerSpareRegion * info->TableDepthInPUs) / (v - 1));
173 		info->SpareRegionDepthInSUs = spareRegionDepthInPUs * layoutPtr->SUsPerPU;
174 
175 		numCompleteSpareRegionsPerDisk = PUsPerDisk / spareRegionDepthInPUs;
176 		info->NumCompleteSRs = numCompleteSpareRegionsPerDisk;
177 		extraPUsPerDisk = PUsPerDisk % spareRegionDepthInPUs;
178 
179 		/* assume conservatively that we need the full amount of spare
180 		 * space in one region in order to provide spares for the
181 		 * partial spare region at the end of the array.  We set "i"
182 		 * to the number of tables in the partial spare region.  This
183 		 * may actually include some fulltables. */
184 		extraPUsPerDisk -= (info->SpareSpaceDepthPerRegionInSUs / layoutPtr->SUsPerPU);
185 		if (extraPUsPerDisk <= 0)
186 			i = 0;
187 		else
188 			i = extraPUsPerDisk / info->TableDepthInPUs;
189 
190 		complete_FT_count = raidPtr->numRow * (numCompleteSpareRegionsPerDisk * (info->TablesPerSpareRegion / k) + i / k);
191 		info->FullTableLimitSUID = complete_FT_count * info->SUsPerFullTable;
192 		info->ExtraTablesPerDisk = i % k;
193 
194 		/* note that in the last spare region, the spare space is
195 		 * complete even though data/parity space is not */
196 		totSparePUsPerDisk = (numCompleteSpareRegionsPerDisk + 1) * (info->SpareSpaceDepthPerRegionInSUs / layoutPtr->SUsPerPU);
197 		info->TotSparePUsPerDisk = totSparePUsPerDisk;
198 
199 		layoutPtr->stripeUnitsPerDisk =
200 		    ((complete_FT_count / raidPtr->numRow) * info->FullTableDepthInPUs +	/* data & parity space */
201 		    info->ExtraTablesPerDisk * info->TableDepthInPUs +
202 		    totSparePUsPerDisk	/* spare space */
203 		    ) * layoutPtr->SUsPerPU;
204 		layoutPtr->dataStripeUnitsPerDisk =
205 		    (complete_FT_count * info->FullTableDepthInPUs + info->ExtraTablesPerDisk * info->TableDepthInPUs)
206 		    * layoutPtr->SUsPerPU * (k - 1) / k;
207 
208 	} else {
209 		/* non-dist spare case:  force each disk to contain an
210 		 * integral number of tables */
211 		layoutPtr->stripeUnitsPerDisk /= (info->TableDepthInPUs * layoutPtr->SUsPerPU);
212 		layoutPtr->stripeUnitsPerDisk *= (info->TableDepthInPUs * layoutPtr->SUsPerPU);
213 
214 		/* compute the number of tables in the last fulltable, which
215 		 * need not be complete */
216 		complete_FT_count =
217 		    ((layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerPU) / info->FullTableDepthInPUs) * raidPtr->numRow;
218 
219 		info->FullTableLimitSUID = complete_FT_count * info->SUsPerFullTable;
220 		info->ExtraTablesPerDisk =
221 		    ((layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerPU) / info->TableDepthInPUs) % k;
222 	}
223 
224 	raidPtr->sectorsPerDisk = layoutPtr->stripeUnitsPerDisk * layoutPtr->sectorsPerStripeUnit;
225 
226 	/* find the disk offset of the stripe unit where the last fulltable
227 	 * starts */
228 	numCompleteFullTablesPerDisk = complete_FT_count / raidPtr->numRow;
229 	diskOffsetOfLastFullTableInSUs = numCompleteFullTablesPerDisk * info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
230 	if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
231 		SpareSpaceInSUs = numCompleteSpareRegionsPerDisk * info->SpareSpaceDepthPerRegionInSUs;
232 		diskOffsetOfLastFullTableInSUs += SpareSpaceInSUs;
233 		info->DiskOffsetOfLastSpareSpaceChunkInSUs =
234 		    diskOffsetOfLastFullTableInSUs + info->ExtraTablesPerDisk * info->TableDepthInPUs * layoutPtr->SUsPerPU;
235 	}
236 	info->DiskOffsetOfLastFullTableInSUs = diskOffsetOfLastFullTableInSUs;
237 	info->numCompleteFullTablesPerDisk = numCompleteFullTablesPerDisk;
238 
239 	/* 4.  create and initialize the lookup tables */
240 	info->LayoutTable = rf_make_2d_array(b, k, raidPtr->cleanupList);
241 	if (info->LayoutTable == NULL)
242 		return (ENOMEM);
243 	info->OffsetTable = rf_make_2d_array(b, k, raidPtr->cleanupList);
244 	if (info->OffsetTable == NULL)
245 		return (ENOMEM);
246 	info->BlockTable = rf_make_2d_array(info->TableDepthInPUs * layoutPtr->SUsPerPU, raidPtr->numCol, raidPtr->cleanupList);
247 	if (info->BlockTable == NULL)
248 		return (ENOMEM);
249 
250 	first_avail_slot = rf_make_1d_array(v, NULL);
251 	if (first_avail_slot == NULL)
252 		return (ENOMEM);
253 
254 	for (i = 0; i < b; i++)
255 		for (j = 0; j < k; j++)
256 			info->LayoutTable[i][j] = *cfgBuf++;
257 
258 	/* initialize offset table */
259 	for (i = 0; i < b; i++)
260 		for (j = 0; j < k; j++) {
261 			info->OffsetTable[i][j] = first_avail_slot[info->LayoutTable[i][j]];
262 			first_avail_slot[info->LayoutTable[i][j]]++;
263 		}
264 
265 	/* initialize block table */
266 	for (SUID = l = 0; l < layoutPtr->SUsPerPU; l++) {
267 		for (i = 0; i < b; i++) {
268 			for (j = 0; j < k; j++) {
269 				info->BlockTable[(info->OffsetTable[i][j] * layoutPtr->SUsPerPU) + l]
270 				    [info->LayoutTable[i][j]] = SUID;
271 			}
272 			SUID++;
273 		}
274 	}
275 
276 	rf_free_1d_array(first_avail_slot, v);
277 
278 	/* 5.  set up the remaining redundant-but-useful parameters */
279 
280 	raidPtr->totalSectors = (k * complete_FT_count + raidPtr->numRow * info->ExtraTablesPerDisk) *
281 	    info->SUsPerTable * layoutPtr->sectorsPerStripeUnit;
282 	layoutPtr->numStripe = (raidPtr->totalSectors / layoutPtr->sectorsPerStripeUnit) / (k - 1);
283 
284 	/* strange evaluation order below to try and minimize overflow
285 	 * problems */
286 
287 	layoutPtr->dataSectorsPerStripe = (k - 1) * layoutPtr->sectorsPerStripeUnit;
288 	layoutPtr->bytesPerStripeUnit = layoutPtr->sectorsPerStripeUnit << raidPtr->logBytesPerSector;
289 	layoutPtr->numDataCol = k - 1;
290 	layoutPtr->numParityCol = 1;
291 
292 	return (0);
293 }
294 /* declustering with distributed sparing */
295 static void rf_ShutdownDeclusteredDS(RF_ThreadArg_t);
296 static void
297 rf_ShutdownDeclusteredDS(arg)
298 	RF_ThreadArg_t arg;
299 {
300 	RF_DeclusteredConfigInfo_t *info;
301 	RF_Raid_t *raidPtr;
302 
303 	raidPtr = (RF_Raid_t *) arg;
304 	info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
305 	if (info->SpareTable)
306 		rf_FreeSpareTable(raidPtr);
307 }
308 
309 int
310 rf_ConfigureDeclusteredDS(
311     RF_ShutdownList_t ** listp,
312     RF_Raid_t * raidPtr,
313     RF_Config_t * cfgPtr)
314 {
315 	int     rc;
316 
317 	rc = rf_ConfigureDeclustered(listp, raidPtr, cfgPtr);
318 	if (rc)
319 		return (rc);
320 	rc = rf_ShutdownCreate(listp, rf_ShutdownDeclusteredDS, raidPtr);
321 	if (rc) {
322 		RF_ERRORMSG1("Got %d adding shutdown event for DeclusteredDS\n", rc);
323 		rf_ShutdownDeclusteredDS(raidPtr);
324 		return (rc);
325 	}
326 	return (0);
327 }
328 
329 void
330 rf_MapSectorDeclustered(raidPtr, raidSector, row, col, diskSector, remap)
331 	RF_Raid_t *raidPtr;
332 	RF_RaidAddr_t raidSector;
333 	RF_RowCol_t *row;
334 	RF_RowCol_t *col;
335 	RF_SectorNum_t *diskSector;
336 	int     remap;
337 {
338 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
339 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
340 	RF_StripeNum_t SUID = raidSector / layoutPtr->sectorsPerStripeUnit;
341 	RF_StripeNum_t FullTableID, FullTableOffset, TableID, TableOffset;
342 	RF_StripeNum_t BlockID, BlockOffset, RepIndex;
343 	RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
344 	RF_StripeCount_t fulltable_depth = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
345 	RF_StripeNum_t base_suid = 0, outSU, SpareRegion = 0, SpareSpace = 0;
346 
347 	rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
348 
349 	FullTableID = SUID / sus_per_fulltable;	/* fulltable ID within array
350 						 * (across rows) */
351 	if (raidPtr->numRow == 1)
352 		*row = 0;	/* avoid a mod and a div in the common case */
353 	else {
354 		*row = FullTableID % raidPtr->numRow;
355 		FullTableID /= raidPtr->numRow;	/* convert to fulltable ID on
356 						 * this disk */
357 	}
358 	if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
359 		SpareRegion = FullTableID / info->FullTablesPerSpareRegion;
360 		SpareSpace = SpareRegion * info->SpareSpaceDepthPerRegionInSUs;
361 	}
362 	FullTableOffset = SUID % sus_per_fulltable;
363 	TableID = FullTableOffset / info->SUsPerTable;
364 	TableOffset = FullTableOffset - TableID * info->SUsPerTable;
365 	BlockID = TableOffset / info->PUsPerBlock;
366 	BlockOffset = TableOffset - BlockID * info->PUsPerBlock;
367 	BlockID %= info->BlocksPerTable;
368 	RepIndex = info->PUsPerBlock - TableID;
369 	if (!raidPtr->noRotate)
370 		BlockOffset += ((BlockOffset >= RepIndex) ? 1 : 0);
371 	*col = info->LayoutTable[BlockID][BlockOffset];
372 
373 	/* remap to distributed spare space if indicated */
374 	if (remap) {
375 		RF_ASSERT(raidPtr->Disks[*row][*col].status == rf_ds_reconstructing || raidPtr->Disks[*row][*col].status == rf_ds_dist_spared ||
376 		    (rf_copyback_in_progress && raidPtr->Disks[*row][*col].status == rf_ds_optimal));
377 		rf_remap_to_spare_space(layoutPtr, info, *row, FullTableID, TableID, BlockID, (base_suid) ? 1 : 0, SpareRegion, col, &outSU);
378 	} else {
379 
380 		outSU = base_suid;
381 		outSU += FullTableID * fulltable_depth;	/* offs to strt of FT */
382 		outSU += SpareSpace;	/* skip rsvd spare space */
383 		outSU += TableID * info->TableDepthInPUs * layoutPtr->SUsPerPU;	/* offs to strt of tble */
384 		outSU += info->OffsetTable[BlockID][BlockOffset] * layoutPtr->SUsPerPU;	/* offs to the PU */
385 	}
386 	outSU += TableOffset / (info->BlocksPerTable * info->PUsPerBlock);	/* offs to the SU within
387 										 * a PU */
388 
389 	/* convert SUs to sectors, and, if not aligned to SU boundary, add in
390 	 * offset to sector.  */
391 	*diskSector = outSU * layoutPtr->sectorsPerStripeUnit + (raidSector % layoutPtr->sectorsPerStripeUnit);
392 
393 	RF_ASSERT(*col != -1);
394 }
395 
396 
397 /* prototyping this inexplicably causes the compile of the layout table (rf_layout.c) to fail */
398 void
399 rf_MapParityDeclustered(
400     RF_Raid_t * raidPtr,
401     RF_RaidAddr_t raidSector,
402     RF_RowCol_t * row,
403     RF_RowCol_t * col,
404     RF_SectorNum_t * diskSector,
405     int remap)
406 {
407 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
408 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
409 	RF_StripeNum_t SUID = raidSector / layoutPtr->sectorsPerStripeUnit;
410 	RF_StripeNum_t FullTableID, FullTableOffset, TableID, TableOffset;
411 	RF_StripeNum_t BlockID, BlockOffset, RepIndex;
412 	RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
413 	RF_StripeCount_t fulltable_depth = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
414 	RF_StripeNum_t base_suid = 0, outSU, SpareRegion = 0, SpareSpace = 0;
415 
416 	rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
417 
418 	/* compute row & (possibly) spare space exactly as before */
419 	FullTableID = SUID / sus_per_fulltable;
420 	if (raidPtr->numRow == 1)
421 		*row = 0;	/* avoid a mod and a div in the common case */
422 	else {
423 		*row = FullTableID % raidPtr->numRow;
424 		FullTableID /= raidPtr->numRow;	/* convert to fulltable ID on
425 						 * this disk */
426 	}
427 	if ((raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
428 		SpareRegion = FullTableID / info->FullTablesPerSpareRegion;
429 		SpareSpace = SpareRegion * info->SpareSpaceDepthPerRegionInSUs;
430 	}
431 	/* compute BlockID and RepIndex exactly as before */
432 	FullTableOffset = SUID % sus_per_fulltable;
433 	TableID = FullTableOffset / info->SUsPerTable;
434 	TableOffset = FullTableOffset - TableID * info->SUsPerTable;
435 	/* TableOffset     = FullTableOffset % info->SUsPerTable; */
436 	/* BlockID         = (TableOffset / info->PUsPerBlock) %
437 	 * info->BlocksPerTable; */
438 	BlockID = TableOffset / info->PUsPerBlock;
439 	/* BlockOffset     = TableOffset % info->PUsPerBlock; */
440 	BlockOffset = TableOffset - BlockID * info->PUsPerBlock;
441 	BlockID %= info->BlocksPerTable;
442 
443 	/* the parity block is in the position indicated by RepIndex */
444 	RepIndex = (raidPtr->noRotate) ? info->PUsPerBlock : info->PUsPerBlock - TableID;
445 	*col = info->LayoutTable[BlockID][RepIndex];
446 
447 	if (remap) {
448 		RF_ASSERT(raidPtr->Disks[*row][*col].status == rf_ds_reconstructing || raidPtr->Disks[*row][*col].status == rf_ds_dist_spared ||
449 		    (rf_copyback_in_progress && raidPtr->Disks[*row][*col].status == rf_ds_optimal));
450 		rf_remap_to_spare_space(layoutPtr, info, *row, FullTableID, TableID, BlockID, (base_suid) ? 1 : 0, SpareRegion, col, &outSU);
451 	} else {
452 
453 		/* compute sector as before, except use RepIndex instead of
454 		 * BlockOffset */
455 		outSU = base_suid;
456 		outSU += FullTableID * fulltable_depth;
457 		outSU += SpareSpace;	/* skip rsvd spare space */
458 		outSU += TableID * info->TableDepthInPUs * layoutPtr->SUsPerPU;
459 		outSU += info->OffsetTable[BlockID][RepIndex] * layoutPtr->SUsPerPU;
460 	}
461 
462 	outSU += TableOffset / (info->BlocksPerTable * info->PUsPerBlock);
463 	*diskSector = outSU * layoutPtr->sectorsPerStripeUnit + (raidSector % layoutPtr->sectorsPerStripeUnit);
464 
465 	RF_ASSERT(*col != -1);
466 }
467 /* returns an array of ints identifying the disks that comprise the stripe containing the indicated address.
468  * the caller must _never_ attempt to modify this array.
469  */
470 void
471 rf_IdentifyStripeDeclustered(
472     RF_Raid_t * raidPtr,
473     RF_RaidAddr_t addr,
474     RF_RowCol_t ** diskids,
475     RF_RowCol_t * outRow)
476 {
477 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
478 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
479 	RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
480 	RF_StripeCount_t fulltable_depth = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
481 	RF_StripeNum_t base_suid = 0;
482 	RF_StripeNum_t SUID = rf_RaidAddressToStripeUnitID(layoutPtr, addr);
483 	RF_StripeNum_t stripeID, FullTableID;
484 	int     tableOffset;
485 
486 	rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
487 	FullTableID = SUID / sus_per_fulltable;	/* fulltable ID within array
488 						 * (across rows) */
489 	*outRow = FullTableID % raidPtr->numRow;
490 	stripeID = rf_StripeUnitIDToStripeID(layoutPtr, SUID);	/* find stripe offset
491 								 * into array */
492 	tableOffset = (stripeID % info->BlocksPerTable);	/* find offset into
493 								 * block design table */
494 	*diskids = info->LayoutTable[tableOffset];
495 }
496 /* This returns the default head-separation limit, which is measured
497  * in "required units for reconstruction".  Each time a disk fetches
498  * a unit, it bumps a counter.  The head-sep code prohibits any disk
499  * from getting more than headSepLimit counter values ahead of any
500  * other.
501  *
502  * We assume here that the number of floating recon buffers is already
503  * set.  There are r stripes to be reconstructed in each table, and so
504  * if we have a total of B buffers, we can have at most B/r tables
505  * under recon at any one time.  In each table, lambda units are required
506  * from each disk, so given B buffers, the head sep limit has to be
507  * (lambda*B)/r units.  We subtract one to avoid weird boundary cases.
508  *
509  * for example, suppose were given 50 buffers, r=19, and lambda=4 as in
510  * the 20.5 design.  There are 19 stripes/table to be reconstructed, so
511  * we can have 50/19 tables concurrently under reconstruction, which means
512  * we can allow the fastest disk to get 50/19 tables ahead of the slower
513  * disk.  There are lambda "required units" for each disk, so the fastest
514  * disk can get 4*50/19 = 10 counter values ahead of the slowest.
515  *
516  * If numBufsToAccumulate is not 1, we need to limit the head sep further
517  * because multiple bufs will be required for each stripe under recon.
518  */
519 RF_HeadSepLimit_t
520 rf_GetDefaultHeadSepLimitDeclustered(
521     RF_Raid_t * raidPtr)
522 {
523 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
524 
525 	return (info->Lambda * raidPtr->numFloatingReconBufs / info->TableDepthInPUs / rf_numBufsToAccumulate);
526 }
527 /* returns the default number of recon buffers to use.  The value
528  * is somewhat arbitrary...it's intended to be large enough to allow
529  * for a reasonably large head-sep limit, but small enough that you
530  * don't use up all your system memory with buffers.
531  */
532 int
533 rf_GetDefaultNumFloatingReconBuffersDeclustered(RF_Raid_t * raidPtr)
534 {
535 	return (100 * rf_numBufsToAccumulate);
536 }
537 /* sectors in the last fulltable of the array need to be handled
538  * specially since this fulltable can be incomplete.  this function
539  * changes the values of certain params to handle this.
540  *
541  * the idea here is that MapSector et. al. figure out which disk the
542  * addressed unit lives on by computing the modulos of the unit number
543  * with the number of units per fulltable, table, etc.  In the last
544  * fulltable, there are fewer units per fulltable, so we need to adjust
545  * the number of user data units per fulltable to reflect this.
546  *
547  * so, we (1) convert the fulltable size and depth parameters to
548  * the size of the partial fulltable at the end, (2) compute the
549  * disk sector offset where this fulltable starts, and (3) convert
550  * the users stripe unit number from an offset into the array to
551  * an offset into the last fulltable.
552  */
553 void
554 rf_decluster_adjust_params(
555     RF_RaidLayout_t * layoutPtr,
556     RF_StripeNum_t * SUID,
557     RF_StripeCount_t * sus_per_fulltable,
558     RF_StripeCount_t * fulltable_depth,
559     RF_StripeNum_t * base_suid)
560 {
561 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
562 
563 	if (*SUID >= info->FullTableLimitSUID) {
564 		/* new full table size is size of last full table on disk */
565 		*sus_per_fulltable = info->ExtraTablesPerDisk * info->SUsPerTable;
566 
567 		/* new full table depth is corresponding depth */
568 		*fulltable_depth = info->ExtraTablesPerDisk * info->TableDepthInPUs * layoutPtr->SUsPerPU;
569 
570 		/* set up the new base offset */
571 		*base_suid = info->DiskOffsetOfLastFullTableInSUs;
572 
573 		/* convert users array address to an offset into the last
574 		 * fulltable */
575 		*SUID -= info->FullTableLimitSUID;
576 	}
577 }
578 /*
579  * map a stripe ID to a parity stripe ID.
580  * See comment above RaidAddressToParityStripeID in layout.c.
581  */
582 void
583 rf_MapSIDToPSIDDeclustered(
584     RF_RaidLayout_t * layoutPtr,
585     RF_StripeNum_t stripeID,
586     RF_StripeNum_t * psID,
587     RF_ReconUnitNum_t * which_ru)
588 {
589 	RF_DeclusteredConfigInfo_t *info;
590 
591 	info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
592 
593 	*psID = (stripeID / (layoutPtr->SUsPerPU * info->BlocksPerTable))
594 	    * info->BlocksPerTable + (stripeID % info->BlocksPerTable);
595 	*which_ru = (stripeID % (info->BlocksPerTable * layoutPtr->SUsPerPU))
596 	    / info->BlocksPerTable;
597 	RF_ASSERT((*which_ru) < layoutPtr->SUsPerPU / layoutPtr->SUsPerRU);
598 }
599 /*
600  * Called from MapSector and MapParity to retarget an access at the spare unit.
601  * Modifies the "col" and "outSU" parameters only.
602  */
603 void
604 rf_remap_to_spare_space(
605     RF_RaidLayout_t * layoutPtr,
606     RF_DeclusteredConfigInfo_t * info,
607     RF_RowCol_t row,
608     RF_StripeNum_t FullTableID,
609     RF_StripeNum_t TableID,
610     RF_SectorNum_t BlockID,
611     RF_StripeNum_t base_suid,
612     RF_StripeNum_t SpareRegion,
613     RF_RowCol_t * outCol,
614     RF_StripeNum_t * outSU)
615 {
616 	RF_StripeNum_t ftID, spareTableStartSU, TableInSpareRegion, lastSROffset,
617 	        which_ft;
618 
619 	/*
620          * note that FullTableID and hence SpareRegion may have gotten
621          * tweaked by rf_decluster_adjust_params. We detect this by
622          * noticing that base_suid is not 0.
623          */
624 	if (base_suid == 0) {
625 		ftID = FullTableID;
626 	} else {
627 		/*
628 	         * There may be > 1.0 full tables in the last (i.e. partial)
629 	         * spare region.  find out which of these we're in.
630 	         */
631 		lastSROffset = info->NumCompleteSRs * info->SpareRegionDepthInSUs;
632 		which_ft = (info->DiskOffsetOfLastFullTableInSUs - lastSROffset) / (info->FullTableDepthInPUs * layoutPtr->SUsPerPU);
633 
634 		/* compute the actual full table ID */
635 		ftID = info->DiskOffsetOfLastFullTableInSUs / (info->FullTableDepthInPUs * layoutPtr->SUsPerPU) + which_ft;
636 		SpareRegion = info->NumCompleteSRs;
637 	}
638 	TableInSpareRegion = (ftID * info->NumParityReps + TableID) % info->TablesPerSpareRegion;
639 
640 	*outCol = info->SpareTable[TableInSpareRegion][BlockID].spareDisk;
641 	RF_ASSERT(*outCol != -1);
642 
643 	spareTableStartSU = (SpareRegion == info->NumCompleteSRs) ?
644 	    info->DiskOffsetOfLastFullTableInSUs + info->ExtraTablesPerDisk * info->TableDepthInPUs * layoutPtr->SUsPerPU :
645 	    (SpareRegion + 1) * info->SpareRegionDepthInSUs - info->SpareSpaceDepthPerRegionInSUs;
646 	*outSU = spareTableStartSU + info->SpareTable[TableInSpareRegion][BlockID].spareBlockOffsetInSUs;
647 	if (*outSU >= layoutPtr->stripeUnitsPerDisk) {
648 		printf("rf_remap_to_spare_space: invalid remapped disk SU offset %ld\n", (long) *outSU);
649 	}
650 }
651 
652 #endif /* (RF_INCLUDE_PARITY_DECLUSTERING > 0)  || (RF_INCLUDE_PARITY_DECLUSTERING_PQ > 0) */
653 
654 
655 int
656 rf_InstallSpareTable(
657     RF_Raid_t * raidPtr,
658     RF_RowCol_t frow,
659     RF_RowCol_t fcol)
660 {
661 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
662 	RF_SparetWait_t *req;
663 	int     retcode;
664 
665 	RF_Malloc(req, sizeof(*req), (RF_SparetWait_t *));
666 	req->C = raidPtr->numCol;
667 	req->G = raidPtr->Layout.numDataCol + raidPtr->Layout.numParityCol;
668 	req->fcol = fcol;
669 	req->SUsPerPU = raidPtr->Layout.SUsPerPU;
670 	req->TablesPerSpareRegion = info->TablesPerSpareRegion;
671 	req->BlocksPerTable = info->BlocksPerTable;
672 	req->TableDepthInPUs = info->TableDepthInPUs;
673 	req->SpareSpaceDepthPerRegionInSUs = info->SpareSpaceDepthPerRegionInSUs;
674 
675 	retcode = rf_GetSpareTableFromDaemon(req);
676 	RF_ASSERT(!retcode);	/* XXX -- fix this to recover gracefully --
677 				 * XXX */
678 	return (retcode);
679 }
680 /*
681  * Invoked via ioctl to install a spare table in the kernel.
682  */
683 int
684 rf_SetSpareTable(raidPtr, data)
685 	RF_Raid_t *raidPtr;
686 	void   *data;
687 {
688 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
689 	RF_SpareTableEntry_t **ptrs;
690 	int     i, retcode;
691 
692 	/* what we need to copyin is a 2-d array, so first copyin the user
693 	 * pointers to the rows in the table */
694 	RF_Malloc(ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *), (RF_SpareTableEntry_t **));
695 	retcode = copyin((caddr_t) data, (caddr_t) ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
696 
697 	if (retcode)
698 		return (retcode);
699 
700 	/* now allocate kernel space for the row pointers */
701 	RF_Malloc(info->SpareTable, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *), (RF_SpareTableEntry_t **));
702 
703 	/* now allocate kernel space for each row in the table, and copy it in
704 	 * from user space */
705 	for (i = 0; i < info->TablesPerSpareRegion; i++) {
706 		RF_Malloc(info->SpareTable[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t), (RF_SpareTableEntry_t *));
707 		retcode = copyin(ptrs[i], info->SpareTable[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t));
708 		if (retcode) {
709 			info->SpareTable = NULL;	/* blow off the memory
710 							 * we've allocated */
711 			return (retcode);
712 		}
713 	}
714 
715 	/* free up the temporary array we used */
716 	RF_Free(ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
717 
718 	return (0);
719 }
720 
721 RF_ReconUnitCount_t
722 rf_GetNumSpareRUsDeclustered(raidPtr)
723 	RF_Raid_t *raidPtr;
724 {
725 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
726 
727 	return (((RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo)->TotSparePUsPerDisk);
728 }
729 
730 void
731 rf_FreeSpareTable(raidPtr)
732 	RF_Raid_t *raidPtr;
733 {
734 	long    i;
735 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
736 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
737 	RF_SpareTableEntry_t **table = info->SpareTable;
738 
739 	for (i = 0; i < info->TablesPerSpareRegion; i++) {
740 		RF_Free(table[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t));
741 	}
742 	RF_Free(table, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
743 	info->SpareTable = (RF_SpareTableEntry_t **) NULL;
744 }
745