xref: /netbsd-src/sys/dev/raidframe/rf_decluster.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: rf_decluster.c,v 1.16 2004/02/29 04:03:50 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 <sys/cdefs.h>
51 __KERNEL_RCSID(0, "$NetBSD: rf_decluster.c,v 1.16 2004/02/29 04:03:50 oster Exp $");
52 
53 #include <dev/raidframe/raidframevar.h>
54 
55 #include "rf_archs.h"
56 #include "rf_raid.h"
57 #include "rf_decluster.h"
58 #include "rf_debugMem.h"
59 #include "rf_utils.h"
60 #include "rf_alloclist.h"
61 #include "rf_general.h"
62 #include "rf_kintf.h"
63 #include "rf_shutdown.h"
64 
65 #if (RF_INCLUDE_PARITY_DECLUSTERING > 0) || (RF_INCLUDE_PARITY_DECLUSTERING_PQ > 0)
66 
67 /* configuration code */
68 
69 int
70 rf_ConfigureDeclustered(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
71 			RF_Config_t *cfgPtr)
72 {
73 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
74 	int     b, v, k, r, lambda;	/* block design params */
75 	int     i, j;
76 	RF_RowCol_t *first_avail_slot;
77 	RF_StripeCount_t complete_FT_count, numCompleteFullTablesPerDisk;
78 	RF_DeclusteredConfigInfo_t *info;
79 	RF_StripeCount_t PUsPerDisk, spareRegionDepthInPUs, numCompleteSpareRegionsPerDisk,
80 	        extraPUsPerDisk;
81 	RF_StripeCount_t totSparePUsPerDisk;
82 	RF_SectorNum_t diskOffsetOfLastFullTableInSUs;
83 	RF_SectorCount_t SpareSpaceInSUs;
84 	char   *cfgBuf = (char *) (cfgPtr->layoutSpecific);
85 	RF_StripeNum_t l, SUID;
86 
87 	SUID = l = 0;
88 	numCompleteSpareRegionsPerDisk = 0;
89 
90 	/* 1. create layout specific structure */
91 	RF_MallocAndAdd(info, sizeof(RF_DeclusteredConfigInfo_t), (RF_DeclusteredConfigInfo_t *), raidPtr->cleanupList);
92 	if (info == NULL)
93 		return (ENOMEM);
94 	layoutPtr->layoutSpecificInfo = (void *) info;
95 	info->SpareTable = NULL;
96 
97 	/* 2. extract parameters from the config structure */
98 	if (layoutPtr->map->flags & RF_DISTRIBUTE_SPARE) {
99 		(void)memcpy(info->sparemap_fname, cfgBuf, RF_SPAREMAP_NAME_LEN);
100 	}
101 	cfgBuf += RF_SPAREMAP_NAME_LEN;
102 
103 	b = *((int *) cfgBuf);
104 	cfgBuf += sizeof(int);
105 	v = *((int *) cfgBuf);
106 	cfgBuf += sizeof(int);
107 	k = *((int *) cfgBuf);
108 	cfgBuf += sizeof(int);
109 	r = *((int *) cfgBuf);
110 	cfgBuf += sizeof(int);
111 	lambda = *((int *) cfgBuf);
112 	cfgBuf += sizeof(int);
113 	raidPtr->noRotate = *((int *) cfgBuf);
114 	cfgBuf += sizeof(int);
115 
116 	/* the sparemaps are generated assuming that parity is rotated, so we
117 	 * issue a warning if both distributed sparing and no-rotate are on at
118 	 * the same time */
119 	if ((layoutPtr->map->flags & RF_DISTRIBUTE_SPARE) && raidPtr->noRotate) {
120 		RF_ERRORMSG("Warning:  distributed sparing specified without parity rotation.\n");
121 	}
122 	if (raidPtr->numCol != v) {
123 		RF_ERRORMSG2("RAID: config error: table element count (%d) not equal to no. of cols (%d)\n", v, raidPtr->numCol);
124 		return (EINVAL);
125 	}
126 	/* 3.  set up the values used in the mapping code */
127 	info->BlocksPerTable = b;
128 	info->Lambda = lambda;
129 	info->NumParityReps = info->groupSize = k;
130 	info->SUsPerTable = b * (k - 1) * layoutPtr->SUsPerPU;	/* b blks, k-1 SUs each */
131 	info->SUsPerFullTable = k * info->SUsPerTable;	/* rot k times */
132 	info->PUsPerBlock = k - 1;
133 	info->SUsPerBlock = info->PUsPerBlock * layoutPtr->SUsPerPU;
134 	info->TableDepthInPUs = (b * k) / v;
135 	info->FullTableDepthInPUs = info->TableDepthInPUs * k;	/* k repetitions */
136 
137 	/* used only in distributed sparing case */
138 	info->FullTablesPerSpareRegion = (v - 1) / rf_gcd(r, v - 1);	/* (v-1)/gcd fulltables */
139 	info->TablesPerSpareRegion = k * info->FullTablesPerSpareRegion;
140 	info->SpareSpaceDepthPerRegionInSUs = (r * info->TablesPerSpareRegion / (v - 1)) * layoutPtr->SUsPerPU;
141 
142 	/* check to make sure the block design is sufficiently small */
143 	if ((raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
144 		if (info->FullTableDepthInPUs * layoutPtr->SUsPerPU + info->SpareSpaceDepthPerRegionInSUs > layoutPtr->stripeUnitsPerDisk) {
145 			RF_ERRORMSG3("RAID: config error: Full Table depth (%d) + Spare Space (%d) larger than disk size (%d) (BD too big)\n",
146 			    (int) info->FullTableDepthInPUs,
147 			    (int) info->SpareSpaceDepthPerRegionInSUs,
148 			    (int) layoutPtr->stripeUnitsPerDisk);
149 			return (EINVAL);
150 		}
151 	} else {
152 		if (info->TableDepthInPUs * layoutPtr->SUsPerPU > layoutPtr->stripeUnitsPerDisk) {
153 			RF_ERRORMSG2("RAID: config error: Table depth (%d) larger than disk size (%d) (BD too big)\n",
154 			    (int) (info->TableDepthInPUs * layoutPtr->SUsPerPU), \
155 			    (int) layoutPtr->stripeUnitsPerDisk);
156 			return (EINVAL);
157 		}
158 	}
159 
160 
161 	/* compute the size of each disk, and the number of tables in the last
162 	 * fulltable (which need not be complete) */
163 	if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
164 
165 		PUsPerDisk = layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerPU;
166 		spareRegionDepthInPUs = (info->TablesPerSpareRegion * info->TableDepthInPUs +
167 		    (info->TablesPerSpareRegion * info->TableDepthInPUs) / (v - 1));
168 		info->SpareRegionDepthInSUs = spareRegionDepthInPUs * layoutPtr->SUsPerPU;
169 
170 		numCompleteSpareRegionsPerDisk = PUsPerDisk / spareRegionDepthInPUs;
171 		info->NumCompleteSRs = numCompleteSpareRegionsPerDisk;
172 		extraPUsPerDisk = PUsPerDisk % spareRegionDepthInPUs;
173 
174 		/* assume conservatively that we need the full amount of spare
175 		 * space in one region in order to provide spares for the
176 		 * partial spare region at the end of the array.  We set "i"
177 		 * to the number of tables in the partial spare region.  This
178 		 * may actually include some fulltables. */
179 		extraPUsPerDisk -= (info->SpareSpaceDepthPerRegionInSUs / layoutPtr->SUsPerPU);
180 		if (extraPUsPerDisk <= 0)
181 			i = 0;
182 		else
183 			i = extraPUsPerDisk / info->TableDepthInPUs;
184 
185 		complete_FT_count = raidPtr->numRow * (numCompleteSpareRegionsPerDisk * (info->TablesPerSpareRegion / k) + i / k);
186 		info->FullTableLimitSUID = complete_FT_count * info->SUsPerFullTable;
187 		info->ExtraTablesPerDisk = i % k;
188 
189 		/* note that in the last spare region, the spare space is
190 		 * complete even though data/parity space is not */
191 		totSparePUsPerDisk = (numCompleteSpareRegionsPerDisk + 1) * (info->SpareSpaceDepthPerRegionInSUs / layoutPtr->SUsPerPU);
192 		info->TotSparePUsPerDisk = totSparePUsPerDisk;
193 
194 		layoutPtr->stripeUnitsPerDisk =
195 		    ((complete_FT_count / raidPtr->numRow) * info->FullTableDepthInPUs +	/* data & parity space */
196 		    info->ExtraTablesPerDisk * info->TableDepthInPUs +
197 		    totSparePUsPerDisk	/* spare space */
198 		    ) * layoutPtr->SUsPerPU;
199 		layoutPtr->dataStripeUnitsPerDisk =
200 		    (complete_FT_count * info->FullTableDepthInPUs + info->ExtraTablesPerDisk * info->TableDepthInPUs)
201 		    * layoutPtr->SUsPerPU * (k - 1) / k;
202 
203 	} else {
204 		/* non-dist spare case:  force each disk to contain an
205 		 * integral number of tables */
206 		layoutPtr->stripeUnitsPerDisk /= (info->TableDepthInPUs * layoutPtr->SUsPerPU);
207 		layoutPtr->stripeUnitsPerDisk *= (info->TableDepthInPUs * layoutPtr->SUsPerPU);
208 
209 		/* compute the number of tables in the last fulltable, which
210 		 * need not be complete */
211 		complete_FT_count =
212 		    ((layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerPU) / info->FullTableDepthInPUs) * raidPtr->numRow;
213 
214 		info->FullTableLimitSUID = complete_FT_count * info->SUsPerFullTable;
215 		info->ExtraTablesPerDisk =
216 		    ((layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerPU) / info->TableDepthInPUs) % k;
217 	}
218 
219 	raidPtr->sectorsPerDisk = layoutPtr->stripeUnitsPerDisk * layoutPtr->sectorsPerStripeUnit;
220 
221 	/* find the disk offset of the stripe unit where the last fulltable
222 	 * starts */
223 	numCompleteFullTablesPerDisk = complete_FT_count / raidPtr->numRow;
224 	diskOffsetOfLastFullTableInSUs = numCompleteFullTablesPerDisk * info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
225 	if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
226 		SpareSpaceInSUs = numCompleteSpareRegionsPerDisk * info->SpareSpaceDepthPerRegionInSUs;
227 		diskOffsetOfLastFullTableInSUs += SpareSpaceInSUs;
228 		info->DiskOffsetOfLastSpareSpaceChunkInSUs =
229 		    diskOffsetOfLastFullTableInSUs + info->ExtraTablesPerDisk * info->TableDepthInPUs * layoutPtr->SUsPerPU;
230 	}
231 	info->DiskOffsetOfLastFullTableInSUs = diskOffsetOfLastFullTableInSUs;
232 	info->numCompleteFullTablesPerDisk = numCompleteFullTablesPerDisk;
233 
234 	/* 4.  create and initialize the lookup tables */
235 	info->LayoutTable = rf_make_2d_array(b, k, raidPtr->cleanupList);
236 	if (info->LayoutTable == NULL)
237 		return (ENOMEM);
238 	info->OffsetTable = rf_make_2d_array(b, k, raidPtr->cleanupList);
239 	if (info->OffsetTable == NULL)
240 		return (ENOMEM);
241 	info->BlockTable = rf_make_2d_array(info->TableDepthInPUs * layoutPtr->SUsPerPU, raidPtr->numCol, raidPtr->cleanupList);
242 	if (info->BlockTable == NULL)
243 		return (ENOMEM);
244 
245 	first_avail_slot = rf_make_1d_array(v, NULL);
246 	if (first_avail_slot == NULL)
247 		return (ENOMEM);
248 
249 	for (i = 0; i < b; i++)
250 		for (j = 0; j < k; j++)
251 			info->LayoutTable[i][j] = *cfgBuf++;
252 
253 	/* initialize offset table */
254 	for (i = 0; i < b; i++)
255 		for (j = 0; j < k; j++) {
256 			info->OffsetTable[i][j] = first_avail_slot[info->LayoutTable[i][j]];
257 			first_avail_slot[info->LayoutTable[i][j]]++;
258 		}
259 
260 	/* initialize block table */
261 	for (SUID = l = 0; l < layoutPtr->SUsPerPU; l++) {
262 		for (i = 0; i < b; i++) {
263 			for (j = 0; j < k; j++) {
264 				info->BlockTable[(info->OffsetTable[i][j] * layoutPtr->SUsPerPU) + l]
265 				    [info->LayoutTable[i][j]] = SUID;
266 			}
267 			SUID++;
268 		}
269 	}
270 
271 	rf_free_1d_array(first_avail_slot, v);
272 
273 	/* 5.  set up the remaining redundant-but-useful parameters */
274 
275 	raidPtr->totalSectors = (k * complete_FT_count + raidPtr->numRow * info->ExtraTablesPerDisk) *
276 	    info->SUsPerTable * layoutPtr->sectorsPerStripeUnit;
277 	layoutPtr->numStripe = (raidPtr->totalSectors / layoutPtr->sectorsPerStripeUnit) / (k - 1);
278 
279 	/* strange evaluation order below to try and minimize overflow
280 	 * problems */
281 
282 	layoutPtr->dataSectorsPerStripe = (k - 1) * layoutPtr->sectorsPerStripeUnit;
283 	layoutPtr->numDataCol = k - 1;
284 	layoutPtr->numParityCol = 1;
285 
286 	return (0);
287 }
288 /* declustering with distributed sparing */
289 static void rf_ShutdownDeclusteredDS(RF_ThreadArg_t);
290 static void
291 rf_ShutdownDeclusteredDS(RF_ThreadArg_t arg)
292 {
293 	RF_DeclusteredConfigInfo_t *info;
294 	RF_Raid_t *raidPtr;
295 
296 	raidPtr = (RF_Raid_t *) arg;
297 	info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
298 	if (info->SpareTable)
299 		rf_FreeSpareTable(raidPtr);
300 }
301 
302 int
303 rf_ConfigureDeclusteredDS(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
304 			  RF_Config_t *cfgPtr)
305 {
306 	int     rc;
307 
308 	rc = rf_ConfigureDeclustered(listp, raidPtr, cfgPtr);
309 	if (rc)
310 		return (rc);
311 	rf_ShutdownCreate(listp, rf_ShutdownDeclusteredDS, raidPtr);
312 
313 	return (0);
314 }
315 
316 void
317 rf_MapSectorDeclustered(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
318 			RF_RowCol_t *row, RF_RowCol_t *col,
319 			RF_SectorNum_t *diskSector, int remap)
320 {
321 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
322 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
323 	RF_StripeNum_t SUID = raidSector / layoutPtr->sectorsPerStripeUnit;
324 	RF_StripeNum_t FullTableID, FullTableOffset, TableID, TableOffset;
325 	RF_StripeNum_t BlockID, BlockOffset, RepIndex;
326 	RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
327 	RF_StripeCount_t fulltable_depth = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
328 	RF_StripeNum_t base_suid = 0, outSU, SpareRegion = 0, SpareSpace = 0;
329 
330 	rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
331 
332 	FullTableID = SUID / sus_per_fulltable;	/* fulltable ID within array
333 						 * (across rows) */
334 	if (raidPtr->numRow == 1)
335 		*row = 0;	/* avoid a mod and a div in the common case */
336 	else {
337 		*row = FullTableID % raidPtr->numRow;
338 		FullTableID /= raidPtr->numRow;	/* convert to fulltable ID on
339 						 * this disk */
340 	}
341 	if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
342 		SpareRegion = FullTableID / info->FullTablesPerSpareRegion;
343 		SpareSpace = SpareRegion * info->SpareSpaceDepthPerRegionInSUs;
344 	}
345 	FullTableOffset = SUID % sus_per_fulltable;
346 	TableID = FullTableOffset / info->SUsPerTable;
347 	TableOffset = FullTableOffset - TableID * info->SUsPerTable;
348 	BlockID = TableOffset / info->PUsPerBlock;
349 	BlockOffset = TableOffset - BlockID * info->PUsPerBlock;
350 	BlockID %= info->BlocksPerTable;
351 	RepIndex = info->PUsPerBlock - TableID;
352 	if (!raidPtr->noRotate)
353 		BlockOffset += ((BlockOffset >= RepIndex) ? 1 : 0);
354 	*col = info->LayoutTable[BlockID][BlockOffset];
355 
356 	/* remap to distributed spare space if indicated */
357 	if (remap) {
358 		RF_ASSERT(raidPtr->Disks[*row][*col].status == rf_ds_reconstructing || raidPtr->Disks[*row][*col].status == rf_ds_dist_spared ||
359 		    (rf_copyback_in_progress && raidPtr->Disks[*row][*col].status == rf_ds_optimal));
360 		rf_remap_to_spare_space(layoutPtr, info, *row, FullTableID, TableID, BlockID, (base_suid) ? 1 : 0, SpareRegion, col, &outSU);
361 	} else {
362 
363 		outSU = base_suid;
364 		outSU += FullTableID * fulltable_depth;	/* offs to strt of FT */
365 		outSU += SpareSpace;	/* skip rsvd spare space */
366 		outSU += TableID * info->TableDepthInPUs * layoutPtr->SUsPerPU;	/* offs to strt of tble */
367 		outSU += info->OffsetTable[BlockID][BlockOffset] * layoutPtr->SUsPerPU;	/* offs to the PU */
368 	}
369 	outSU += TableOffset / (info->BlocksPerTable * info->PUsPerBlock);	/* offs to the SU within
370 										 * a PU */
371 
372 	/* convert SUs to sectors, and, if not aligned to SU boundary, add in
373 	 * offset to sector.  */
374 	*diskSector = outSU * layoutPtr->sectorsPerStripeUnit + (raidSector % layoutPtr->sectorsPerStripeUnit);
375 
376 	RF_ASSERT(*col != -1);
377 }
378 
379 
380 /* prototyping this inexplicably causes the compile of the layout table (rf_layout.c) to fail */
381 void
382 rf_MapParityDeclustered(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
383 			RF_RowCol_t *row, RF_RowCol_t *col,
384 			RF_SectorNum_t *diskSector, int remap)
385 {
386 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
387 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
388 	RF_StripeNum_t SUID = raidSector / layoutPtr->sectorsPerStripeUnit;
389 	RF_StripeNum_t FullTableID, FullTableOffset, TableID, TableOffset;
390 	RF_StripeNum_t BlockID, BlockOffset, RepIndex;
391 	RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
392 	RF_StripeCount_t fulltable_depth = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
393 	RF_StripeNum_t base_suid = 0, outSU, SpareRegion = 0, SpareSpace = 0;
394 
395 	rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
396 
397 	/* compute row & (possibly) spare space exactly as before */
398 	FullTableID = SUID / sus_per_fulltable;
399 	if (raidPtr->numRow == 1)
400 		*row = 0;	/* avoid a mod and a div in the common case */
401 	else {
402 		*row = FullTableID % raidPtr->numRow;
403 		FullTableID /= raidPtr->numRow;	/* convert to fulltable ID on
404 						 * this disk */
405 	}
406 	if ((raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
407 		SpareRegion = FullTableID / info->FullTablesPerSpareRegion;
408 		SpareSpace = SpareRegion * info->SpareSpaceDepthPerRegionInSUs;
409 	}
410 	/* compute BlockID and RepIndex exactly as before */
411 	FullTableOffset = SUID % sus_per_fulltable;
412 	TableID = FullTableOffset / info->SUsPerTable;
413 	TableOffset = FullTableOffset - TableID * info->SUsPerTable;
414 	/* TableOffset     = FullTableOffset % info->SUsPerTable; */
415 	/* BlockID         = (TableOffset / info->PUsPerBlock) %
416 	 * info->BlocksPerTable; */
417 	BlockID = TableOffset / info->PUsPerBlock;
418 	/* BlockOffset     = TableOffset % info->PUsPerBlock; */
419 	BlockOffset = TableOffset - BlockID * info->PUsPerBlock;
420 	BlockID %= info->BlocksPerTable;
421 
422 	/* the parity block is in the position indicated by RepIndex */
423 	RepIndex = (raidPtr->noRotate) ? info->PUsPerBlock : info->PUsPerBlock - TableID;
424 	*col = info->LayoutTable[BlockID][RepIndex];
425 
426 	if (remap) {
427 		RF_ASSERT(raidPtr->Disks[*row][*col].status == rf_ds_reconstructing || raidPtr->Disks[*row][*col].status == rf_ds_dist_spared ||
428 		    (rf_copyback_in_progress && raidPtr->Disks[*row][*col].status == rf_ds_optimal));
429 		rf_remap_to_spare_space(layoutPtr, info, *row, FullTableID, TableID, BlockID, (base_suid) ? 1 : 0, SpareRegion, col, &outSU);
430 	} else {
431 
432 		/* compute sector as before, except use RepIndex instead of
433 		 * BlockOffset */
434 		outSU = base_suid;
435 		outSU += FullTableID * fulltable_depth;
436 		outSU += SpareSpace;	/* skip rsvd spare space */
437 		outSU += TableID * info->TableDepthInPUs * layoutPtr->SUsPerPU;
438 		outSU += info->OffsetTable[BlockID][RepIndex] * layoutPtr->SUsPerPU;
439 	}
440 
441 	outSU += TableOffset / (info->BlocksPerTable * info->PUsPerBlock);
442 	*diskSector = outSU * layoutPtr->sectorsPerStripeUnit + (raidSector % layoutPtr->sectorsPerStripeUnit);
443 
444 	RF_ASSERT(*col != -1);
445 }
446 /* returns an array of ints identifying the disks that comprise the stripe containing the indicated address.
447  * the caller must _never_ attempt to modify this array.
448  */
449 void
450 rf_IdentifyStripeDeclustered(RF_Raid_t *raidPtr, RF_RaidAddr_t addr,
451 			     RF_RowCol_t **diskids, RF_RowCol_t *outRow)
452 {
453 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
454 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
455 	RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
456 	RF_StripeCount_t fulltable_depth = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
457 	RF_StripeNum_t base_suid = 0;
458 	RF_StripeNum_t SUID = rf_RaidAddressToStripeUnitID(layoutPtr, addr);
459 	RF_StripeNum_t stripeID, FullTableID;
460 	int     tableOffset;
461 
462 	rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
463 	FullTableID = SUID / sus_per_fulltable;	/* fulltable ID within array
464 						 * (across rows) */
465 	*outRow = FullTableID % raidPtr->numRow;
466 	stripeID = rf_StripeUnitIDToStripeID(layoutPtr, SUID);	/* find stripe offset
467 								 * into array */
468 	tableOffset = (stripeID % info->BlocksPerTable);	/* find offset into
469 								 * block design table */
470 	*diskids = info->LayoutTable[tableOffset];
471 }
472 /* This returns the default head-separation limit, which is measured
473  * in "required units for reconstruction".  Each time a disk fetches
474  * a unit, it bumps a counter.  The head-sep code prohibits any disk
475  * from getting more than headSepLimit counter values ahead of any
476  * other.
477  *
478  * We assume here that the number of floating recon buffers is already
479  * set.  There are r stripes to be reconstructed in each table, and so
480  * if we have a total of B buffers, we can have at most B/r tables
481  * under recon at any one time.  In each table, lambda units are required
482  * from each disk, so given B buffers, the head sep limit has to be
483  * (lambda*B)/r units.  We subtract one to avoid weird boundary cases.
484  *
485  * for example, suppose were given 50 buffers, r=19, and lambda=4 as in
486  * the 20.5 design.  There are 19 stripes/table to be reconstructed, so
487  * we can have 50/19 tables concurrently under reconstruction, which means
488  * we can allow the fastest disk to get 50/19 tables ahead of the slower
489  * disk.  There are lambda "required units" for each disk, so the fastest
490  * disk can get 4*50/19 = 10 counter values ahead of the slowest.
491  *
492  * If numBufsToAccumulate is not 1, we need to limit the head sep further
493  * because multiple bufs will be required for each stripe under recon.
494  */
495 RF_HeadSepLimit_t
496 rf_GetDefaultHeadSepLimitDeclustered(RF_Raid_t *raidPtr)
497 {
498 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
499 
500 	return (info->Lambda * raidPtr->numFloatingReconBufs / info->TableDepthInPUs / rf_numBufsToAccumulate);
501 }
502 /* returns the default number of recon buffers to use.  The value
503  * is somewhat arbitrary...it's intended to be large enough to allow
504  * for a reasonably large head-sep limit, but small enough that you
505  * don't use up all your system memory with buffers.
506  */
507 int
508 rf_GetDefaultNumFloatingReconBuffersDeclustered(RF_Raid_t * raidPtr)
509 {
510 	return (100 * rf_numBufsToAccumulate);
511 }
512 /* sectors in the last fulltable of the array need to be handled
513  * specially since this fulltable can be incomplete.  this function
514  * changes the values of certain params to handle this.
515  *
516  * the idea here is that MapSector et. al. figure out which disk the
517  * addressed unit lives on by computing the modulos of the unit number
518  * with the number of units per fulltable, table, etc.  In the last
519  * fulltable, there are fewer units per fulltable, so we need to adjust
520  * the number of user data units per fulltable to reflect this.
521  *
522  * so, we (1) convert the fulltable size and depth parameters to
523  * the size of the partial fulltable at the end, (2) compute the
524  * disk sector offset where this fulltable starts, and (3) convert
525  * the users stripe unit number from an offset into the array to
526  * an offset into the last fulltable.
527  */
528 void
529 rf_decluster_adjust_params(RF_RaidLayout_t *layoutPtr,
530 			   RF_StripeNum_t *SUID,
531 			   RF_StripeCount_t *sus_per_fulltable,
532 			   RF_StripeCount_t *fulltable_depth,
533 			   RF_StripeNum_t *base_suid)
534 {
535 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
536 
537 	if (*SUID >= info->FullTableLimitSUID) {
538 		/* new full table size is size of last full table on disk */
539 		*sus_per_fulltable = info->ExtraTablesPerDisk * info->SUsPerTable;
540 
541 		/* new full table depth is corresponding depth */
542 		*fulltable_depth = info->ExtraTablesPerDisk * info->TableDepthInPUs * layoutPtr->SUsPerPU;
543 
544 		/* set up the new base offset */
545 		*base_suid = info->DiskOffsetOfLastFullTableInSUs;
546 
547 		/* convert users array address to an offset into the last
548 		 * fulltable */
549 		*SUID -= info->FullTableLimitSUID;
550 	}
551 }
552 /*
553  * map a stripe ID to a parity stripe ID.
554  * See comment above RaidAddressToParityStripeID in layout.c.
555  */
556 void
557 rf_MapSIDToPSIDDeclustered(RF_RaidLayout_t *layoutPtr,
558 			   RF_StripeNum_t stripeID,
559 			   RF_StripeNum_t *psID,
560 			   RF_ReconUnitNum_t *which_ru)
561 {
562 	RF_DeclusteredConfigInfo_t *info;
563 
564 	info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
565 
566 	*psID = (stripeID / (layoutPtr->SUsPerPU * info->BlocksPerTable))
567 	    * info->BlocksPerTable + (stripeID % info->BlocksPerTable);
568 	*which_ru = (stripeID % (info->BlocksPerTable * layoutPtr->SUsPerPU))
569 	    / info->BlocksPerTable;
570 	RF_ASSERT((*which_ru) < layoutPtr->SUsPerPU / layoutPtr->SUsPerRU);
571 }
572 /*
573  * Called from MapSector and MapParity to retarget an access at the spare unit.
574  * Modifies the "col" and "outSU" parameters only.
575  */
576 void
577 rf_remap_to_spare_space(RF_RaidLayout_t *layoutPtr,
578 			RF_DeclusteredConfigInfo_t *info,
579 			RF_RowCol_t row,
580 			RF_StripeNum_t FullTableID,
581 			RF_StripeNum_t TableID,
582 			RF_SectorNum_t BlockID,
583 			RF_StripeNum_t base_suid,
584 			RF_StripeNum_t SpareRegion,
585 			RF_RowCol_t *outCol,
586 			RF_StripeNum_t *outSU)
587 {
588 	RF_StripeNum_t ftID, spareTableStartSU, TableInSpareRegion, lastSROffset,
589 	        which_ft;
590 
591 	/*
592          * note that FullTableID and hence SpareRegion may have gotten
593          * tweaked by rf_decluster_adjust_params. We detect this by
594          * noticing that base_suid is not 0.
595          */
596 	if (base_suid == 0) {
597 		ftID = FullTableID;
598 	} else {
599 		/*
600 	         * There may be > 1.0 full tables in the last (i.e. partial)
601 	         * spare region.  find out which of these we're in.
602 	         */
603 		lastSROffset = info->NumCompleteSRs * info->SpareRegionDepthInSUs;
604 		which_ft = (info->DiskOffsetOfLastFullTableInSUs - lastSROffset) / (info->FullTableDepthInPUs * layoutPtr->SUsPerPU);
605 
606 		/* compute the actual full table ID */
607 		ftID = info->DiskOffsetOfLastFullTableInSUs / (info->FullTableDepthInPUs * layoutPtr->SUsPerPU) + which_ft;
608 		SpareRegion = info->NumCompleteSRs;
609 	}
610 	TableInSpareRegion = (ftID * info->NumParityReps + TableID) % info->TablesPerSpareRegion;
611 
612 	*outCol = info->SpareTable[TableInSpareRegion][BlockID].spareDisk;
613 	RF_ASSERT(*outCol != -1);
614 
615 	spareTableStartSU = (SpareRegion == info->NumCompleteSRs) ?
616 	    info->DiskOffsetOfLastFullTableInSUs + info->ExtraTablesPerDisk * info->TableDepthInPUs * layoutPtr->SUsPerPU :
617 	    (SpareRegion + 1) * info->SpareRegionDepthInSUs - info->SpareSpaceDepthPerRegionInSUs;
618 	*outSU = spareTableStartSU + info->SpareTable[TableInSpareRegion][BlockID].spareBlockOffsetInSUs;
619 	if (*outSU >= layoutPtr->stripeUnitsPerDisk) {
620 		printf("rf_remap_to_spare_space: invalid remapped disk SU offset %ld\n", (long) *outSU);
621 	}
622 }
623 
624 #endif /* (RF_INCLUDE_PARITY_DECLUSTERING > 0)  || (RF_INCLUDE_PARITY_DECLUSTERING_PQ > 0) */
625 
626 #if (RF_INCLUDE_PARITY_DECLUSTERING_DS > 0)
627 int
628 rf_InstallSpareTable(RF_Raid_t *raidPtr, RF_RowCol_t frow, RF_RowCol_t fcol)
629 {
630 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
631 	RF_SparetWait_t *req;
632 	int     retcode;
633 
634 	RF_Malloc(req, sizeof(*req), (RF_SparetWait_t *));
635 	req->C = raidPtr->numCol;
636 	req->G = raidPtr->Layout.numDataCol + raidPtr->Layout.numParityCol;
637 	req->fcol = fcol;
638 	req->SUsPerPU = raidPtr->Layout.SUsPerPU;
639 	req->TablesPerSpareRegion = info->TablesPerSpareRegion;
640 	req->BlocksPerTable = info->BlocksPerTable;
641 	req->TableDepthInPUs = info->TableDepthInPUs;
642 	req->SpareSpaceDepthPerRegionInSUs = info->SpareSpaceDepthPerRegionInSUs;
643 
644 	retcode = rf_GetSpareTableFromDaemon(req);
645 	RF_ASSERT(!retcode);	/* XXX -- fix this to recover gracefully --
646 				 * XXX */
647 	return (retcode);
648 }
649 #endif
650 #if (RF_INCLUDE_PARITY_DECLUSTERING > 0) || (RF_INCLUDE_PARITY_DECLUSTERING_PQ > 0)
651 /*
652  * Invoked via ioctl to install a spare table in the kernel.
653  */
654 int
655 rf_SetSpareTable(RF_Raid_t *raidPtr, void *data)
656 {
657 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
658 	RF_SpareTableEntry_t **ptrs;
659 	int     i, retcode;
660 
661 	/* what we need to copyin is a 2-d array, so first copyin the user
662 	 * pointers to the rows in the table */
663 	RF_Malloc(ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *), (RF_SpareTableEntry_t **));
664 	retcode = copyin((caddr_t) data, (caddr_t) ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
665 
666 	if (retcode)
667 		return (retcode);
668 
669 	/* now allocate kernel space for the row pointers */
670 	RF_Malloc(info->SpareTable, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *), (RF_SpareTableEntry_t **));
671 
672 	/* now allocate kernel space for each row in the table, and copy it in
673 	 * from user space */
674 	for (i = 0; i < info->TablesPerSpareRegion; i++) {
675 		RF_Malloc(info->SpareTable[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t), (RF_SpareTableEntry_t *));
676 		retcode = copyin(ptrs[i], info->SpareTable[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t));
677 		if (retcode) {
678 			info->SpareTable = NULL;	/* blow off the memory
679 							 * we've allocated */
680 			return (retcode);
681 		}
682 	}
683 
684 	/* free up the temporary array we used */
685 	RF_Free(ptrs, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
686 
687 	return (0);
688 }
689 
690 RF_ReconUnitCount_t
691 rf_GetNumSpareRUsDeclustered(RF_Raid_t *raidPtr)
692 {
693 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
694 
695 	return (((RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo)->TotSparePUsPerDisk);
696 }
697 #endif /* (RF_INCLUDE_PARITY_DECLUSTERING > 0)  || (RF_INCLUDE_PARITY_DECLUSTERING_PQ > 0) */
698 
699 void
700 rf_FreeSpareTable(RF_Raid_t *raidPtr)
701 {
702 	long    i;
703 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
704 	RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
705 	RF_SpareTableEntry_t **table = info->SpareTable;
706 
707 	for (i = 0; i < info->TablesPerSpareRegion; i++) {
708 		RF_Free(table[i], info->BlocksPerTable * sizeof(RF_SpareTableEntry_t));
709 	}
710 	RF_Free(table, info->TablesPerSpareRegion * sizeof(RF_SpareTableEntry_t *));
711 	info->SpareTable = (RF_SpareTableEntry_t **) NULL;
712 }
713