rf_evenodd.c revision 1.16.10.2 1 1.16.10.2 yamt /* $NetBSD: rf_evenodd.c,v 1.16.10.2 2006/12/10 07:18:11 yamt Exp $ */
2 1.1 oster /*
3 1.1 oster * Copyright (c) 1995 Carnegie-Mellon University.
4 1.1 oster * All rights reserved.
5 1.1 oster *
6 1.1 oster * Author: Chang-Ming Wu
7 1.1 oster *
8 1.1 oster * Permission to use, copy, modify and distribute this software and
9 1.1 oster * its documentation is hereby granted, provided that both the copyright
10 1.1 oster * notice and this permission notice appear in all copies of the
11 1.1 oster * software, derivative works or modified versions, and any portions
12 1.1 oster * thereof, and that both notices appear in supporting documentation.
13 1.1 oster *
14 1.1 oster * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
15 1.1 oster * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
16 1.1 oster * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
17 1.1 oster *
18 1.1 oster * Carnegie Mellon requests users of this software to return to
19 1.1 oster *
20 1.1 oster * Software Distribution Coordinator or Software.Distribution (at) CS.CMU.EDU
21 1.1 oster * School of Computer Science
22 1.1 oster * Carnegie Mellon University
23 1.1 oster * Pittsburgh PA 15213-3890
24 1.1 oster *
25 1.1 oster * any improvements or extensions that they make and grant Carnegie the
26 1.1 oster * rights to redistribute these changes.
27 1.1 oster */
28 1.1 oster
29 1.1 oster /*****************************************************************************************
30 1.1 oster *
31 1.1 oster * rf_evenodd.c -- implements EVENODD array architecture
32 1.1 oster *
33 1.1 oster ****************************************************************************************/
34 1.7 lukem
35 1.7 lukem #include <sys/cdefs.h>
36 1.16.10.2 yamt __KERNEL_RCSID(0, "$NetBSD: rf_evenodd.c,v 1.16.10.2 2006/12/10 07:18:11 yamt Exp $");
37 1.1 oster
38 1.1 oster #include "rf_archs.h"
39 1.1 oster
40 1.1 oster #if RF_INCLUDE_EVENODD > 0
41 1.1 oster
42 1.6 oster #include <dev/raidframe/raidframevar.h>
43 1.6 oster
44 1.1 oster #include "rf_raid.h"
45 1.1 oster #include "rf_dag.h"
46 1.1 oster #include "rf_dagffrd.h"
47 1.1 oster #include "rf_dagffwr.h"
48 1.1 oster #include "rf_dagdegrd.h"
49 1.1 oster #include "rf_dagdegwr.h"
50 1.1 oster #include "rf_dagutils.h"
51 1.1 oster #include "rf_dagfuncs.h"
52 1.1 oster #include "rf_etimer.h"
53 1.1 oster #include "rf_general.h"
54 1.1 oster #include "rf_evenodd.h"
55 1.1 oster #include "rf_parityscan.h"
56 1.1 oster #include "rf_utils.h"
57 1.1 oster #include "rf_map.h"
58 1.1 oster #include "rf_pq.h"
59 1.1 oster #include "rf_mcpair.h"
60 1.1 oster #include "rf_evenodd.h"
61 1.1 oster #include "rf_evenodd_dagfuncs.h"
62 1.1 oster #include "rf_evenodd_dags.h"
63 1.1 oster #include "rf_engine.h"
64 1.1 oster
65 1.1 oster typedef struct RF_EvenOddConfigInfo_s {
66 1.2 oster RF_RowCol_t **stripeIdentifier; /* filled in at config time & used by
67 1.2 oster * IdentifyStripe */
68 1.2 oster } RF_EvenOddConfigInfo_t;
69 1.2 oster
70 1.14 perry int
71 1.16.10.2 yamt rf_ConfigureEvenOdd(RF_ShutdownList_t **listp, RF_Raid_t *raidPtr,
72 1.16.10.2 yamt RF_Config_t *cfgPtr)
73 1.2 oster {
74 1.2 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
75 1.2 oster RF_EvenOddConfigInfo_t *info;
76 1.2 oster RF_RowCol_t i, j, startdisk;
77 1.2 oster
78 1.2 oster RF_MallocAndAdd(info, sizeof(RF_EvenOddConfigInfo_t), (RF_EvenOddConfigInfo_t *), raidPtr->cleanupList);
79 1.2 oster layoutPtr->layoutSpecificInfo = (void *) info;
80 1.2 oster
81 1.2 oster info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol, raidPtr->numCol, raidPtr->cleanupList);
82 1.2 oster startdisk = 0;
83 1.2 oster for (i = 0; i < raidPtr->numCol; i++) {
84 1.2 oster for (j = 0; j < raidPtr->numCol; j++) {
85 1.2 oster info->stripeIdentifier[i][j] = (startdisk + j) % raidPtr->numCol;
86 1.2 oster }
87 1.2 oster if ((startdisk -= 2) < 0)
88 1.2 oster startdisk += raidPtr->numCol;
89 1.2 oster }
90 1.1 oster
91 1.2 oster /* fill in the remaining layout parameters */
92 1.2 oster layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk;
93 1.2 oster layoutPtr->numDataCol = raidPtr->numCol - 2; /* ORIG:
94 1.2 oster * layoutPtr->numDataCol
95 1.2 oster * = raidPtr->numCol-1; */
96 1.1 oster #if RF_EO_MATRIX_DIM > 17
97 1.2 oster if (raidPtr->numCol <= 17) {
98 1.2 oster printf("Number of stripe units in a parity stripe is smaller than 17. Please\n");
99 1.2 oster printf("define the macro RF_EO_MATRIX_DIM in file rf_evenodd_dagfuncs.h to \n");
100 1.2 oster printf("be 17 to increase performance. \n");
101 1.2 oster return (EINVAL);
102 1.2 oster }
103 1.1 oster #elif RF_EO_MATRIX_DIM == 17
104 1.2 oster if (raidPtr->numCol > 17) {
105 1.2 oster printf("Number of stripe units in a parity stripe is bigger than 17. Please\n");
106 1.2 oster printf("define the macro RF_EO_MATRIX_DIM in file rf_evenodd_dagfuncs.h to \n");
107 1.2 oster printf("be 257 for encoding and decoding functions to work. \n");
108 1.2 oster return (EINVAL);
109 1.2 oster }
110 1.1 oster #endif
111 1.2 oster layoutPtr->dataSectorsPerStripe = layoutPtr->numDataCol * layoutPtr->sectorsPerStripeUnit;
112 1.2 oster layoutPtr->numParityCol = 2;
113 1.2 oster layoutPtr->dataStripeUnitsPerDisk = layoutPtr->stripeUnitsPerDisk;
114 1.2 oster raidPtr->sectorsPerDisk = layoutPtr->stripeUnitsPerDisk * layoutPtr->sectorsPerStripeUnit;
115 1.1 oster
116 1.2 oster raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * layoutPtr->numDataCol * layoutPtr->sectorsPerStripeUnit;
117 1.1 oster
118 1.2 oster return (0);
119 1.1 oster }
120 1.1 oster
121 1.14 perry int
122 1.16.10.2 yamt rf_GetDefaultNumFloatingReconBuffersEvenOdd(RF_Raid_t *raidPtr)
123 1.1 oster {
124 1.2 oster return (20);
125 1.1 oster }
126 1.1 oster
127 1.14 perry RF_HeadSepLimit_t
128 1.16.10.2 yamt rf_GetDefaultHeadSepLimitEvenOdd(RF_Raid_t *raidPtr)
129 1.1 oster {
130 1.2 oster return (10);
131 1.1 oster }
132 1.1 oster
133 1.14 perry void
134 1.12 oster rf_IdentifyStripeEvenOdd(RF_Raid_t *raidPtr, RF_RaidAddr_t addr,
135 1.16 oster RF_RowCol_t **diskids)
136 1.1 oster {
137 1.2 oster RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
138 1.2 oster RF_EvenOddConfigInfo_t *info = (RF_EvenOddConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
139 1.1 oster
140 1.2 oster *diskids = info->stripeIdentifier[stripeID % raidPtr->numCol];
141 1.1 oster }
142 1.2 oster /* The layout of stripe unit on the disks are: c0 c1 c2 c3 c4
143 1.1 oster
144 1.1 oster 0 1 2 E P
145 1.1 oster 5 E P 3 4
146 1.1 oster P 6 7 8 E
147 1.1 oster 10 11 E P 9
148 1.1 oster E P 12 13 14
149 1.1 oster ....
150 1.1 oster
151 1.2 oster We use the MapSectorRAID5 to map data information because the routine can be shown to map exactly
152 1.1 oster the layout of data stripe unit as shown above although we have 2 redundant information now.
153 1.1 oster But for E and P, we use rf_MapEEvenOdd and rf_MapParityEvenOdd which are different method from raid-5.
154 1.1 oster */
155 1.1 oster
156 1.1 oster
157 1.14 perry void
158 1.12 oster rf_MapParityEvenOdd(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
159 1.16 oster RF_RowCol_t *col,
160 1.16.10.2 yamt RF_SectorNum_t *diskSector, int remap)
161 1.1 oster {
162 1.2 oster RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
163 1.2 oster RF_StripeNum_t endSUIDofthisStrip = (SUID / raidPtr->Layout.numDataCol + 1) * raidPtr->Layout.numDataCol - 1;
164 1.1 oster
165 1.2 oster *col = (endSUIDofthisStrip + 2) % raidPtr->numCol;
166 1.2 oster *diskSector = (SUID / (raidPtr->Layout.numDataCol)) * raidPtr->Layout.sectorsPerStripeUnit +
167 1.2 oster (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
168 1.1 oster }
169 1.1 oster
170 1.14 perry void
171 1.12 oster rf_MapEEvenOdd(RF_Raid_t *raidPtr, RF_RaidAddr_t raidSector,
172 1.16 oster RF_RowCol_t *col, RF_SectorNum_t *diskSector,
173 1.16.10.2 yamt int remap)
174 1.1 oster {
175 1.2 oster RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
176 1.2 oster RF_StripeNum_t endSUIDofthisStrip = (SUID / raidPtr->Layout.numDataCol + 1) * raidPtr->Layout.numDataCol - 1;
177 1.1 oster
178 1.2 oster *col = (endSUIDofthisStrip + 1) % raidPtr->numCol;
179 1.2 oster *diskSector = (SUID / (raidPtr->Layout.numDataCol)) * raidPtr->Layout.sectorsPerStripeUnit +
180 1.2 oster (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
181 1.1 oster }
182 1.1 oster
183 1.14 perry void
184 1.12 oster rf_EODagSelect(RF_Raid_t *raidPtr, RF_IoType_t type,
185 1.12 oster RF_AccessStripeMap_t *asmap, RF_VoidFuncPtr *createFunc)
186 1.1 oster {
187 1.2 oster RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
188 1.2 oster unsigned ndfail = asmap->numDataFailed;
189 1.2 oster unsigned npfail = asmap->numParityFailed + asmap->numQFailed;
190 1.2 oster unsigned ntfail = npfail + ndfail;
191 1.2 oster
192 1.2 oster RF_ASSERT(RF_IO_IS_R_OR_W(type));
193 1.2 oster if (ntfail > 2) {
194 1.2 oster RF_ERRORMSG("more than two disks failed in a single group! Aborting I/O operation.\n");
195 1.10 oster *createFunc = NULL;
196 1.2 oster return;
197 1.2 oster }
198 1.2 oster /* ok, we can do this I/O */
199 1.2 oster if (type == RF_IO_TYPE_READ) {
200 1.2 oster switch (ndfail) {
201 1.2 oster case 0:
202 1.2 oster /* fault free read */
203 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_CreateFaultFreeReadDAG; /* same as raid 5 */
204 1.2 oster break;
205 1.2 oster case 1:
206 1.2 oster /* lost a single data unit */
207 1.2 oster /* two cases: (1) parity is not lost. do a normal raid
208 1.2 oster * 5 reconstruct read. (2) parity is lost. do a
209 1.2 oster * reconstruct read using "e". */
210 1.2 oster if (ntfail == 2) { /* also lost redundancy */
211 1.2 oster if (asmap->failedPDAs[1]->type == RF_PDA_TYPE_PARITY)
212 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_110_CreateReadDAG;
213 1.2 oster else
214 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_101_CreateReadDAG;
215 1.2 oster } else {
216 1.2 oster /* P and E are ok. But is there a failure in
217 1.2 oster * some unaccessed data unit? */
218 1.2 oster if (rf_NumFailedDataUnitsInStripe(raidPtr, asmap) == 2)
219 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_200_CreateReadDAG;
220 1.2 oster else
221 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_100_CreateReadDAG;
222 1.2 oster }
223 1.2 oster break;
224 1.2 oster case 2:
225 1.2 oster /* *createFunc = rf_EO_200_CreateReadDAG; */
226 1.2 oster *createFunc = NULL;
227 1.2 oster break;
228 1.2 oster }
229 1.2 oster return;
230 1.2 oster }
231 1.2 oster /* a write */
232 1.2 oster switch (ntfail) {
233 1.2 oster case 0: /* fault free */
234 1.2 oster if (rf_suppressLocksAndLargeWrites ||
235 1.2 oster (((asmap->numStripeUnitsAccessed <= (layoutPtr->numDataCol / 2)) && (layoutPtr->numDataCol != 1)) ||
236 1.2 oster (asmap->parityInfo->next != NULL) || (asmap->qInfo->next != NULL) || rf_CheckStripeForFailures(raidPtr, asmap))) {
237 1.2 oster
238 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EOCreateSmallWriteDAG;
239 1.2 oster } else {
240 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EOCreateLargeWriteDAG;
241 1.2 oster }
242 1.2 oster break;
243 1.2 oster
244 1.2 oster case 1: /* single disk fault */
245 1.2 oster if (npfail == 1) {
246 1.2 oster RF_ASSERT((asmap->failedPDAs[0]->type == RF_PDA_TYPE_PARITY) || (asmap->failedPDAs[0]->type == RF_PDA_TYPE_Q));
247 1.2 oster if (asmap->failedPDAs[0]->type == RF_PDA_TYPE_Q) { /* q died, treat like
248 1.2 oster * normal mode raid5
249 1.2 oster * write. */
250 1.2 oster if (((asmap->numStripeUnitsAccessed <= (layoutPtr->numDataCol / 2)) || (asmap->numStripeUnitsAccessed == 1))
251 1.2 oster || (asmap->parityInfo->next != NULL) || rf_NumFailedDataUnitsInStripe(raidPtr, asmap))
252 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_001_CreateSmallWriteDAG;
253 1.2 oster else
254 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_001_CreateLargeWriteDAG;
255 1.2 oster } else {/* parity died, small write only updating Q */
256 1.2 oster if (((asmap->numStripeUnitsAccessed <= (layoutPtr->numDataCol / 2)) || (asmap->numStripeUnitsAccessed == 1))
257 1.2 oster || (asmap->qInfo->next != NULL) || rf_NumFailedDataUnitsInStripe(raidPtr, asmap))
258 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_010_CreateSmallWriteDAG;
259 1.2 oster else
260 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_010_CreateLargeWriteDAG;
261 1.2 oster }
262 1.2 oster } else { /* data missing. Do a P reconstruct write if
263 1.2 oster * only a single data unit is lost in the
264 1.2 oster * stripe, otherwise a reconstruct write which
265 1.2 oster * employnig both P and E units. */
266 1.2 oster if (rf_NumFailedDataUnitsInStripe(raidPtr, asmap) == 2) {
267 1.2 oster if (asmap->numStripeUnitsAccessed == 1)
268 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_200_CreateWriteDAG;
269 1.2 oster else
270 1.2 oster *createFunc = NULL; /* No direct support for
271 1.2 oster * this case now, like
272 1.2 oster * that in Raid-5 */
273 1.2 oster } else {
274 1.2 oster if (asmap->numStripeUnitsAccessed != 1 && asmap->failedPDAs[0]->numSector != layoutPtr->sectorsPerStripeUnit)
275 1.2 oster *createFunc = NULL; /* No direct support for
276 1.2 oster * this case now, like
277 1.2 oster * that in Raid-5 */
278 1.2 oster else
279 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_100_CreateWriteDAG;
280 1.2 oster }
281 1.2 oster }
282 1.2 oster break;
283 1.2 oster
284 1.2 oster case 2: /* two disk faults */
285 1.2 oster switch (npfail) {
286 1.2 oster case 2: /* both p and q dead */
287 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_011_CreateWriteDAG;
288 1.2 oster break;
289 1.2 oster case 1: /* either p or q and dead data */
290 1.2 oster RF_ASSERT(asmap->failedPDAs[0]->type == RF_PDA_TYPE_DATA);
291 1.2 oster RF_ASSERT((asmap->failedPDAs[1]->type == RF_PDA_TYPE_PARITY) || (asmap->failedPDAs[1]->type == RF_PDA_TYPE_Q));
292 1.2 oster if (asmap->failedPDAs[1]->type == RF_PDA_TYPE_Q) {
293 1.2 oster if (asmap->numStripeUnitsAccessed != 1 && asmap->failedPDAs[0]->numSector != layoutPtr->sectorsPerStripeUnit)
294 1.2 oster *createFunc = NULL; /* In both PQ and
295 1.2 oster * EvenOdd, no direct
296 1.2 oster * support for this case
297 1.2 oster * now, like that in
298 1.2 oster * Raid-5 */
299 1.2 oster else
300 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_101_CreateWriteDAG;
301 1.2 oster } else {
302 1.2 oster if (asmap->numStripeUnitsAccessed != 1 && asmap->failedPDAs[0]->numSector != layoutPtr->sectorsPerStripeUnit)
303 1.2 oster *createFunc = NULL; /* No direct support for
304 1.2 oster * this case, like that
305 1.2 oster * in Raid-5 */
306 1.2 oster else
307 1.2 oster *createFunc = (RF_VoidFuncPtr) rf_EO_110_CreateWriteDAG;
308 1.2 oster }
309 1.2 oster break;
310 1.2 oster case 0: /* double data loss */
311 1.2 oster /* if(asmap->failedPDAs[0]->numSector +
312 1.2 oster * asmap->failedPDAs[1]->numSector == 2 *
313 1.2 oster * layoutPtr->sectorsPerStripeUnit ) createFunc =
314 1.2 oster * rf_EOCreateLargeWriteDAG; else */
315 1.2 oster *createFunc = NULL; /* currently, in Evenodd, No
316 1.2 oster * support for simultaneous
317 1.2 oster * access of both failed SUs */
318 1.2 oster break;
319 1.2 oster }
320 1.2 oster break;
321 1.2 oster
322 1.2 oster default: /* more than 2 disk faults */
323 1.2 oster *createFunc = NULL;
324 1.2 oster RF_PANIC();
325 1.2 oster }
326 1.2 oster return;
327 1.1 oster }
328 1.1 oster
329 1.1 oster
330 1.14 perry int
331 1.12 oster rf_VerifyParityEvenOdd(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
332 1.12 oster RF_PhysDiskAddr_t *parityPDA, int correct_it,
333 1.12 oster RF_RaidAccessFlags_t flags)
334 1.1 oster {
335 1.2 oster RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
336 1.2 oster RF_RaidAddr_t startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
337 1.2 oster RF_SectorCount_t numsector = parityPDA->numSector;
338 1.2 oster int numbytes = rf_RaidAddressToByte(raidPtr, numsector);
339 1.2 oster int bytesPerStripe = numbytes * layoutPtr->numDataCol;
340 1.2 oster RF_DagHeader_t *rd_dag_h, *wr_dag_h; /* read, write dag */
341 1.2 oster RF_DagNode_t *blockNode, *unblockNode, *wrBlock, *wrUnblock;
342 1.2 oster RF_AccessStripeMapHeader_t *asm_h;
343 1.2 oster RF_AccessStripeMap_t *asmap;
344 1.2 oster RF_AllocListElem_t *alloclist;
345 1.2 oster RF_PhysDiskAddr_t *pda;
346 1.2 oster char *pbuf, *buf, *end_p, *p;
347 1.2 oster char *redundantbuf2;
348 1.2 oster int redundantTwoErr = 0, redundantOneErr = 0;
349 1.2 oster int parity_cant_correct = RF_FALSE, red2_cant_correct = RF_FALSE,
350 1.2 oster parity_corrected = RF_FALSE, red2_corrected = RF_FALSE;
351 1.2 oster int i, retcode;
352 1.2 oster RF_ReconUnitNum_t which_ru;
353 1.2 oster RF_StripeNum_t psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
354 1.2 oster int stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
355 1.2 oster RF_AccTraceEntry_t tracerec;
356 1.2 oster RF_MCPair_t *mcpair;
357 1.2 oster
358 1.2 oster retcode = RF_PARITY_OKAY;
359 1.2 oster
360 1.2 oster mcpair = rf_AllocMCPair();
361 1.2 oster rf_MakeAllocList(alloclist);
362 1.2 oster RF_MallocAndAdd(buf, numbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol), (char *), alloclist);
363 1.11 oster RF_MallocAndAdd(pbuf, numbytes, (char *), alloclist);
364 1.2 oster end_p = buf + bytesPerStripe;
365 1.11 oster RF_MallocAndAdd(redundantbuf2, numbytes, (char *), alloclist);
366 1.2 oster
367 1.2 oster rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, numbytes, buf, rf_DiskReadFunc, rf_DiskReadUndoFunc,
368 1.2 oster "Rod", alloclist, flags, RF_IO_NORMAL_PRIORITY);
369 1.2 oster blockNode = rd_dag_h->succedents[0];
370 1.2 oster unblockNode = blockNode->succedents[0]->succedents[0];
371 1.2 oster
372 1.2 oster /* map the stripe and fill in the PDAs in the dag */
373 1.2 oster asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe, buf, RF_DONT_REMAP);
374 1.2 oster asmap = asm_h->stripeMap;
375 1.2 oster
376 1.2 oster for (pda = asmap->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
377 1.2 oster RF_ASSERT(pda);
378 1.2 oster rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
379 1.2 oster RF_ASSERT(pda->numSector != 0);
380 1.2 oster if (rf_TryToRedirectPDA(raidPtr, pda, 0))
381 1.2 oster goto out; /* no way to verify parity if disk is
382 1.2 oster * dead. return w/ good status */
383 1.2 oster blockNode->succedents[i]->params[0].p = pda;
384 1.2 oster blockNode->succedents[i]->params[2].v = psID;
385 1.13 oster blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
386 1.2 oster }
387 1.2 oster
388 1.2 oster RF_ASSERT(!asmap->parityInfo->next);
389 1.2 oster rf_RangeRestrictPDA(raidPtr, parityPDA, asmap->parityInfo, 0, 1);
390 1.2 oster RF_ASSERT(asmap->parityInfo->numSector != 0);
391 1.2 oster if (rf_TryToRedirectPDA(raidPtr, asmap->parityInfo, 1))
392 1.2 oster goto out;
393 1.2 oster blockNode->succedents[layoutPtr->numDataCol]->params[0].p = asmap->parityInfo;
394 1.2 oster
395 1.2 oster RF_ASSERT(!asmap->qInfo->next);
396 1.2 oster rf_RangeRestrictPDA(raidPtr, parityPDA, asmap->qInfo, 0, 1);
397 1.2 oster RF_ASSERT(asmap->qInfo->numSector != 0);
398 1.2 oster if (rf_TryToRedirectPDA(raidPtr, asmap->qInfo, 1))
399 1.2 oster goto out;
400 1.2 oster /* if disk is dead, b/c no reconstruction is implemented right now,
401 1.2 oster * the function "rf_TryToRedirectPDA" always return one, which cause
402 1.2 oster * go to out and return w/ good status */
403 1.2 oster blockNode->succedents[layoutPtr->numDataCol + 1]->params[0].p = asmap->qInfo;
404 1.2 oster
405 1.2 oster /* fire off the DAG */
406 1.5 thorpej memset((char *) &tracerec, 0, sizeof(tracerec));
407 1.2 oster rd_dag_h->tracerec = &tracerec;
408 1.2 oster
409 1.8 oster #if RF_DEBUG_VALIDATE_DAG
410 1.2 oster if (rf_verifyParityDebug) {
411 1.2 oster printf("Parity verify read dag:\n");
412 1.2 oster rf_PrintDAGList(rd_dag_h);
413 1.2 oster }
414 1.8 oster #endif
415 1.2 oster RF_LOCK_MUTEX(mcpair->mutex);
416 1.2 oster mcpair->flag = 0;
417 1.2 oster rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
418 1.2 oster (void *) mcpair);
419 1.2 oster while (!mcpair->flag)
420 1.2 oster RF_WAIT_COND(mcpair->cond, mcpair->mutex);
421 1.2 oster RF_UNLOCK_MUTEX(mcpair->mutex);
422 1.2 oster if (rd_dag_h->status != rf_enable) {
423 1.2 oster RF_ERRORMSG("Unable to verify parity: can't read the stripe\n");
424 1.2 oster retcode = RF_PARITY_COULD_NOT_VERIFY;
425 1.2 oster goto out;
426 1.2 oster }
427 1.2 oster for (p = buf, i = 0; p < end_p; p += numbytes, i++) {
428 1.2 oster rf_e_encToBuf(raidPtr, i, p, RF_EO_MATRIX_DIM - 2, redundantbuf2, numsector);
429 1.2 oster /* the corresponding columes in EvenOdd encoding Matrix for
430 1.2 oster * these p pointers which point to the databuffer in a full
431 1.2 oster * stripe are sequentially from 0 to layoutPtr->numDataCol-1 */
432 1.16 oster rf_bxor(p, pbuf, numbytes);
433 1.2 oster }
434 1.2 oster RF_ASSERT(i == layoutPtr->numDataCol);
435 1.2 oster
436 1.2 oster for (i = 0; i < numbytes; i++) {
437 1.2 oster if (pbuf[i] != buf[bytesPerStripe + i]) {
438 1.2 oster if (!correct_it) {
439 1.2 oster RF_ERRORMSG3("Parity verify error: byte %d of parity is 0x%x should be 0x%x\n",
440 1.2 oster i, (u_char) buf[bytesPerStripe + i], (u_char) pbuf[i]);
441 1.2 oster }
442 1.2 oster }
443 1.2 oster redundantOneErr = 1;
444 1.2 oster break;
445 1.2 oster }
446 1.2 oster
447 1.2 oster for (i = 0; i < numbytes; i++) {
448 1.2 oster if (redundantbuf2[i] != buf[bytesPerStripe + numbytes + i]) {
449 1.2 oster if (!correct_it) {
450 1.2 oster RF_ERRORMSG3("Parity verify error: byte %d of second redundant information is 0x%x should be 0x%x\n",
451 1.2 oster i, (u_char) buf[bytesPerStripe + numbytes + i], (u_char) redundantbuf2[i]);
452 1.2 oster }
453 1.2 oster redundantTwoErr = 1;
454 1.2 oster break;
455 1.2 oster }
456 1.2 oster }
457 1.2 oster if (redundantOneErr || redundantTwoErr)
458 1.2 oster retcode = RF_PARITY_BAD;
459 1.2 oster
460 1.2 oster /* correct the first redundant disk, ie parity if it is error */
461 1.2 oster if (redundantOneErr && correct_it) {
462 1.2 oster wr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, numbytes, pbuf, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
463 1.2 oster "Wnp", alloclist, flags, RF_IO_NORMAL_PRIORITY);
464 1.2 oster wrBlock = wr_dag_h->succedents[0];
465 1.2 oster wrUnblock = wrBlock->succedents[0]->succedents[0];
466 1.2 oster wrBlock->succedents[0]->params[0].p = asmap->parityInfo;
467 1.2 oster wrBlock->succedents[0]->params[2].v = psID;
468 1.13 oster wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
469 1.5 thorpej memset((char *) &tracerec, 0, sizeof(tracerec));
470 1.2 oster wr_dag_h->tracerec = &tracerec;
471 1.8 oster #if RF_DEBUG_VALIDATE_DAG
472 1.2 oster if (rf_verifyParityDebug) {
473 1.2 oster printf("Parity verify write dag:\n");
474 1.2 oster rf_PrintDAGList(wr_dag_h);
475 1.2 oster }
476 1.8 oster #endif
477 1.2 oster RF_LOCK_MUTEX(mcpair->mutex);
478 1.2 oster mcpair->flag = 0;
479 1.2 oster rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
480 1.2 oster (void *) mcpair);
481 1.2 oster while (!mcpair->flag)
482 1.2 oster RF_WAIT_COND(mcpair->cond, mcpair->mutex);
483 1.2 oster RF_UNLOCK_MUTEX(mcpair->mutex);
484 1.2 oster if (wr_dag_h->status != rf_enable) {
485 1.2 oster RF_ERRORMSG("Unable to correct parity in VerifyParity: can't write the stripe\n");
486 1.2 oster parity_cant_correct = RF_TRUE;
487 1.2 oster } else {
488 1.2 oster parity_corrected = RF_TRUE;
489 1.2 oster }
490 1.2 oster rf_FreeDAG(wr_dag_h);
491 1.2 oster }
492 1.2 oster if (redundantTwoErr && correct_it) {
493 1.2 oster wr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, numbytes, redundantbuf2, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
494 1.2 oster "Wnred2", alloclist, flags, RF_IO_NORMAL_PRIORITY);
495 1.2 oster wrBlock = wr_dag_h->succedents[0];
496 1.2 oster wrUnblock = wrBlock->succedents[0]->succedents[0];
497 1.2 oster wrBlock->succedents[0]->params[0].p = asmap->qInfo;
498 1.2 oster wrBlock->succedents[0]->params[2].v = psID;
499 1.13 oster wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, which_ru);
500 1.5 thorpej memset((char *) &tracerec, 0, sizeof(tracerec));
501 1.2 oster wr_dag_h->tracerec = &tracerec;
502 1.8 oster #if RF_DEBUG_VALIDATE_DAG
503 1.2 oster if (rf_verifyParityDebug) {
504 1.2 oster printf("Dag of write new second redundant information in parity verify :\n");
505 1.2 oster rf_PrintDAGList(wr_dag_h);
506 1.2 oster }
507 1.8 oster #endif
508 1.2 oster RF_LOCK_MUTEX(mcpair->mutex);
509 1.2 oster mcpair->flag = 0;
510 1.2 oster rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
511 1.2 oster (void *) mcpair);
512 1.2 oster while (!mcpair->flag)
513 1.2 oster RF_WAIT_COND(mcpair->cond, mcpair->mutex);
514 1.2 oster RF_UNLOCK_MUTEX(mcpair->mutex);
515 1.2 oster if (wr_dag_h->status != rf_enable) {
516 1.2 oster RF_ERRORMSG("Unable to correct second redundant information in VerifyParity: can't write the stripe\n");
517 1.2 oster red2_cant_correct = RF_TRUE;
518 1.2 oster } else {
519 1.2 oster red2_corrected = RF_TRUE;
520 1.2 oster }
521 1.2 oster rf_FreeDAG(wr_dag_h);
522 1.2 oster }
523 1.2 oster if ((redundantOneErr && parity_cant_correct) ||
524 1.2 oster (redundantTwoErr && red2_cant_correct))
525 1.2 oster retcode = RF_PARITY_COULD_NOT_CORRECT;
526 1.2 oster if ((retcode = RF_PARITY_BAD) && parity_corrected && red2_corrected)
527 1.2 oster retcode = RF_PARITY_CORRECTED;
528 1.1 oster
529 1.1 oster
530 1.1 oster out:
531 1.2 oster rf_FreeAccessStripeMap(asm_h);
532 1.2 oster rf_FreeAllocList(alloclist);
533 1.2 oster rf_FreeDAG(rd_dag_h);
534 1.2 oster rf_FreeMCPair(mcpair);
535 1.2 oster return (retcode);
536 1.1 oster }
537 1.2 oster #endif /* RF_INCLUDE_EVENODD > 0 */
538