rf_raid1.c revision 1.7 1 1.7 thorpej /* $NetBSD: rf_raid1.c,v 1.7 2001/07/18 06:46:46 thorpej 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: William V. Courtright II
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_raid1.c -- implements RAID Level 1
32 1.1 oster *
33 1.1 oster *****************************************************************************/
34 1.1 oster
35 1.1 oster #include "rf_raid.h"
36 1.1 oster #include "rf_raid1.h"
37 1.1 oster #include "rf_dag.h"
38 1.1 oster #include "rf_dagffrd.h"
39 1.1 oster #include "rf_dagffwr.h"
40 1.1 oster #include "rf_dagdegrd.h"
41 1.1 oster #include "rf_dagutils.h"
42 1.1 oster #include "rf_dagfuncs.h"
43 1.1 oster #include "rf_diskqueue.h"
44 1.1 oster #include "rf_general.h"
45 1.1 oster #include "rf_utils.h"
46 1.1 oster #include "rf_parityscan.h"
47 1.1 oster #include "rf_mcpair.h"
48 1.1 oster #include "rf_layout.h"
49 1.1 oster #include "rf_map.h"
50 1.1 oster #include "rf_engine.h"
51 1.1 oster #include "rf_reconbuffer.h"
52 1.1 oster
53 1.1 oster typedef struct RF_Raid1ConfigInfo_s {
54 1.3 oster RF_RowCol_t **stripeIdentifier;
55 1.3 oster } RF_Raid1ConfigInfo_t;
56 1.1 oster /* start of day code specific to RAID level 1 */
57 1.3 oster int
58 1.3 oster rf_ConfigureRAID1(
59 1.3 oster RF_ShutdownList_t ** listp,
60 1.3 oster RF_Raid_t * raidPtr,
61 1.3 oster RF_Config_t * cfgPtr)
62 1.1 oster {
63 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
64 1.3 oster RF_Raid1ConfigInfo_t *info;
65 1.3 oster RF_RowCol_t i;
66 1.3 oster
67 1.3 oster /* create a RAID level 1 configuration structure */
68 1.3 oster RF_MallocAndAdd(info, sizeof(RF_Raid1ConfigInfo_t), (RF_Raid1ConfigInfo_t *), raidPtr->cleanupList);
69 1.3 oster if (info == NULL)
70 1.3 oster return (ENOMEM);
71 1.3 oster layoutPtr->layoutSpecificInfo = (void *) info;
72 1.3 oster
73 1.3 oster /* ... and fill it in. */
74 1.3 oster info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol / 2, 2, raidPtr->cleanupList);
75 1.3 oster if (info->stripeIdentifier == NULL)
76 1.3 oster return (ENOMEM);
77 1.3 oster for (i = 0; i < (raidPtr->numCol / 2); i++) {
78 1.3 oster info->stripeIdentifier[i][0] = (2 * i);
79 1.3 oster info->stripeIdentifier[i][1] = (2 * i) + 1;
80 1.3 oster }
81 1.3 oster
82 1.3 oster RF_ASSERT(raidPtr->numRow == 1);
83 1.3 oster
84 1.3 oster /* this implementation of RAID level 1 uses one row of numCol disks
85 1.3 oster * and allows multiple (numCol / 2) stripes per row. A stripe
86 1.3 oster * consists of a single data unit and a single parity (mirror) unit.
87 1.3 oster * stripe id = raidAddr / stripeUnitSize */
88 1.3 oster raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
89 1.3 oster layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
90 1.3 oster layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
91 1.3 oster layoutPtr->bytesPerStripeUnit = layoutPtr->sectorsPerStripeUnit << raidPtr->logBytesPerSector;
92 1.3 oster layoutPtr->numDataCol = 1;
93 1.3 oster layoutPtr->numParityCol = 1;
94 1.3 oster return (0);
95 1.1 oster }
96 1.1 oster
97 1.1 oster
98 1.1 oster /* returns the physical disk location of the primary copy in the mirror pair */
99 1.3 oster void
100 1.3 oster rf_MapSectorRAID1(
101 1.3 oster RF_Raid_t * raidPtr,
102 1.3 oster RF_RaidAddr_t raidSector,
103 1.3 oster RF_RowCol_t * row,
104 1.3 oster RF_RowCol_t * col,
105 1.3 oster RF_SectorNum_t * diskSector,
106 1.3 oster int remap)
107 1.1 oster {
108 1.3 oster RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
109 1.3 oster RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
110 1.1 oster
111 1.3 oster *row = 0;
112 1.3 oster *col = 2 * mirrorPair;
113 1.3 oster *diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
114 1.1 oster }
115 1.1 oster
116 1.1 oster
117 1.1 oster /* Map Parity
118 1.1 oster *
119 1.1 oster * returns the physical disk location of the secondary copy in the mirror
120 1.1 oster * pair
121 1.1 oster */
122 1.3 oster void
123 1.3 oster rf_MapParityRAID1(
124 1.3 oster RF_Raid_t * raidPtr,
125 1.3 oster RF_RaidAddr_t raidSector,
126 1.3 oster RF_RowCol_t * row,
127 1.3 oster RF_RowCol_t * col,
128 1.3 oster RF_SectorNum_t * diskSector,
129 1.3 oster int remap)
130 1.1 oster {
131 1.3 oster RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
132 1.3 oster RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
133 1.1 oster
134 1.3 oster *row = 0;
135 1.3 oster *col = (2 * mirrorPair) + 1;
136 1.1 oster
137 1.3 oster *diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
138 1.1 oster }
139 1.1 oster
140 1.1 oster
141 1.1 oster /* IdentifyStripeRAID1
142 1.1 oster *
143 1.1 oster * returns a list of disks for a given redundancy group
144 1.1 oster */
145 1.3 oster void
146 1.3 oster rf_IdentifyStripeRAID1(
147 1.3 oster RF_Raid_t * raidPtr,
148 1.3 oster RF_RaidAddr_t addr,
149 1.3 oster RF_RowCol_t ** diskids,
150 1.3 oster RF_RowCol_t * outRow)
151 1.1 oster {
152 1.3 oster RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
153 1.3 oster RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
154 1.3 oster RF_ASSERT(stripeID >= 0);
155 1.3 oster RF_ASSERT(addr >= 0);
156 1.3 oster *outRow = 0;
157 1.3 oster *diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
158 1.3 oster RF_ASSERT(*diskids);
159 1.1 oster }
160 1.1 oster
161 1.1 oster
162 1.1 oster /* MapSIDToPSIDRAID1
163 1.1 oster *
164 1.1 oster * maps a logical stripe to a stripe in the redundant array
165 1.1 oster */
166 1.3 oster void
167 1.3 oster rf_MapSIDToPSIDRAID1(
168 1.3 oster RF_RaidLayout_t * layoutPtr,
169 1.3 oster RF_StripeNum_t stripeID,
170 1.3 oster RF_StripeNum_t * psID,
171 1.3 oster RF_ReconUnitNum_t * which_ru)
172 1.1 oster {
173 1.3 oster *which_ru = 0;
174 1.3 oster *psID = stripeID;
175 1.1 oster }
176 1.1 oster
177 1.1 oster
178 1.1 oster
179 1.1 oster /******************************************************************************
180 1.1 oster * select a graph to perform a single-stripe access
181 1.1 oster *
182 1.1 oster * Parameters: raidPtr - description of the physical array
183 1.1 oster * type - type of operation (read or write) requested
184 1.1 oster * asmap - logical & physical addresses for this access
185 1.1 oster * createFunc - name of function to use to create the graph
186 1.1 oster *****************************************************************************/
187 1.1 oster
188 1.3 oster void
189 1.3 oster rf_RAID1DagSelect(
190 1.3 oster RF_Raid_t * raidPtr,
191 1.3 oster RF_IoType_t type,
192 1.3 oster RF_AccessStripeMap_t * asmap,
193 1.3 oster RF_VoidFuncPtr * createFunc)
194 1.1 oster {
195 1.3 oster RF_RowCol_t frow, fcol, or, oc;
196 1.3 oster RF_PhysDiskAddr_t *failedPDA;
197 1.5 oster int prior_recon;
198 1.3 oster RF_RowStatus_t rstat;
199 1.3 oster RF_SectorNum_t oo;
200 1.3 oster
201 1.3 oster
202 1.3 oster RF_ASSERT(RF_IO_IS_R_OR_W(type));
203 1.3 oster
204 1.3 oster if (asmap->numDataFailed + asmap->numParityFailed > 1) {
205 1.3 oster RF_ERRORMSG("Multiple disks failed in a single group! Aborting I/O operation.\n");
206 1.3 oster *createFunc = NULL;
207 1.3 oster return;
208 1.3 oster }
209 1.3 oster if (asmap->numDataFailed + asmap->numParityFailed) {
210 1.3 oster /*
211 1.3 oster * We've got a fault. Re-map to spare space, iff applicable.
212 1.3 oster * Shouldn't the arch-independent code do this for us?
213 1.3 oster * Anyway, it turns out if we don't do this here, then when
214 1.3 oster * we're reconstructing, writes go only to the surviving
215 1.3 oster * original disk, and aren't reflected on the reconstructed
216 1.3 oster * spare. Oops. --jimz
217 1.3 oster */
218 1.3 oster failedPDA = asmap->failedPDAs[0];
219 1.3 oster frow = failedPDA->row;
220 1.3 oster fcol = failedPDA->col;
221 1.3 oster rstat = raidPtr->status[frow];
222 1.3 oster prior_recon = (rstat == rf_rs_reconfigured) || (
223 1.3 oster (rstat == rf_rs_reconstructing) ?
224 1.3 oster rf_CheckRUReconstructed(raidPtr->reconControl[frow]->reconMap, failedPDA->startSector) : 0
225 1.3 oster );
226 1.3 oster if (prior_recon) {
227 1.3 oster or = frow;
228 1.3 oster oc = fcol;
229 1.3 oster oo = failedPDA->startSector;
230 1.3 oster /*
231 1.3 oster * If we did distributed sparing, we'd monkey with that here.
232 1.3 oster * But we don't, so we'll
233 1.3 oster */
234 1.3 oster failedPDA->row = raidPtr->Disks[frow][fcol].spareRow;
235 1.3 oster failedPDA->col = raidPtr->Disks[frow][fcol].spareCol;
236 1.3 oster /*
237 1.3 oster * Redirect other components, iff necessary. This looks
238 1.3 oster * pretty suspicious to me, but it's what the raid5
239 1.3 oster * DAG select does.
240 1.3 oster */
241 1.3 oster if (asmap->parityInfo->next) {
242 1.3 oster if (failedPDA == asmap->parityInfo) {
243 1.3 oster failedPDA->next->row = failedPDA->row;
244 1.3 oster failedPDA->next->col = failedPDA->col;
245 1.3 oster } else {
246 1.3 oster if (failedPDA == asmap->parityInfo->next) {
247 1.3 oster asmap->parityInfo->row = failedPDA->row;
248 1.3 oster asmap->parityInfo->col = failedPDA->col;
249 1.3 oster }
250 1.3 oster }
251 1.3 oster }
252 1.3 oster if (rf_dagDebug || rf_mapDebug) {
253 1.5 oster printf("raid%d: Redirected type '%c' r %d c %d o %ld -> r %d c %d o %ld\n",
254 1.5 oster raidPtr->raidid, type, or, oc,
255 1.5 oster (long) oo, failedPDA->row,
256 1.5 oster failedPDA->col,
257 1.5 oster (long) failedPDA->startSector);
258 1.3 oster }
259 1.3 oster asmap->numDataFailed = asmap->numParityFailed = 0;
260 1.3 oster }
261 1.3 oster }
262 1.3 oster if (type == RF_IO_TYPE_READ) {
263 1.3 oster if (asmap->numDataFailed == 0)
264 1.3 oster *createFunc = (RF_VoidFuncPtr) rf_CreateMirrorIdleReadDAG;
265 1.3 oster else
266 1.3 oster *createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
267 1.3 oster } else {
268 1.3 oster *createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
269 1.3 oster }
270 1.1 oster }
271 1.1 oster
272 1.3 oster int
273 1.3 oster rf_VerifyParityRAID1(
274 1.3 oster RF_Raid_t * raidPtr,
275 1.3 oster RF_RaidAddr_t raidAddr,
276 1.3 oster RF_PhysDiskAddr_t * parityPDA,
277 1.3 oster int correct_it,
278 1.3 oster RF_RaidAccessFlags_t flags)
279 1.1 oster {
280 1.5 oster int nbytes, bcount, stripeWidth, ret, i, j, nbad, *bbufs;
281 1.3 oster RF_DagNode_t *blockNode, *unblockNode, *wrBlock;
282 1.3 oster RF_DagHeader_t *rd_dag_h, *wr_dag_h;
283 1.3 oster RF_AccessStripeMapHeader_t *asm_h;
284 1.3 oster RF_AllocListElem_t *allocList;
285 1.3 oster RF_AccTraceEntry_t tracerec;
286 1.3 oster RF_ReconUnitNum_t which_ru;
287 1.3 oster RF_RaidLayout_t *layoutPtr;
288 1.3 oster RF_AccessStripeMap_t *aasm;
289 1.3 oster RF_SectorCount_t nsector;
290 1.3 oster RF_RaidAddr_t startAddr;
291 1.3 oster char *buf, *buf1, *buf2;
292 1.3 oster RF_PhysDiskAddr_t *pda;
293 1.3 oster RF_StripeNum_t psID;
294 1.3 oster RF_MCPair_t *mcpair;
295 1.3 oster
296 1.3 oster layoutPtr = &raidPtr->Layout;
297 1.3 oster startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
298 1.3 oster nsector = parityPDA->numSector;
299 1.3 oster nbytes = rf_RaidAddressToByte(raidPtr, nsector);
300 1.3 oster psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
301 1.3 oster
302 1.3 oster asm_h = NULL;
303 1.3 oster rd_dag_h = wr_dag_h = NULL;
304 1.3 oster mcpair = NULL;
305 1.3 oster
306 1.3 oster ret = RF_PARITY_COULD_NOT_VERIFY;
307 1.3 oster
308 1.3 oster rf_MakeAllocList(allocList);
309 1.3 oster if (allocList == NULL)
310 1.3 oster return (RF_PARITY_COULD_NOT_VERIFY);
311 1.3 oster mcpair = rf_AllocMCPair();
312 1.3 oster if (mcpair == NULL)
313 1.3 oster goto done;
314 1.3 oster RF_ASSERT(layoutPtr->numDataCol == layoutPtr->numParityCol);
315 1.3 oster stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
316 1.3 oster bcount = nbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol);
317 1.3 oster RF_MallocAndAdd(buf, bcount, (char *), allocList);
318 1.3 oster if (buf == NULL)
319 1.3 oster goto done;
320 1.3 oster if (rf_verifyParityDebug) {
321 1.5 oster printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
322 1.5 oster raidPtr->raidid, (long) buf, bcount, (long) buf,
323 1.5 oster (long) buf + bcount);
324 1.3 oster }
325 1.3 oster /*
326 1.3 oster * Generate a DAG which will read the entire stripe- then we can
327 1.3 oster * just compare data chunks versus "parity" chunks.
328 1.3 oster */
329 1.3 oster
330 1.3 oster rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, buf,
331 1.3 oster rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
332 1.3 oster RF_IO_NORMAL_PRIORITY);
333 1.3 oster if (rd_dag_h == NULL)
334 1.3 oster goto done;
335 1.3 oster blockNode = rd_dag_h->succedents[0];
336 1.3 oster unblockNode = blockNode->succedents[0]->succedents[0];
337 1.3 oster
338 1.3 oster /*
339 1.3 oster * Map the access to physical disk addresses (PDAs)- this will
340 1.3 oster * get us both a list of data addresses, and "parity" addresses
341 1.3 oster * (which are really mirror copies).
342 1.3 oster */
343 1.3 oster asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
344 1.3 oster buf, RF_DONT_REMAP);
345 1.3 oster aasm = asm_h->stripeMap;
346 1.3 oster
347 1.3 oster buf1 = buf;
348 1.3 oster /*
349 1.3 oster * Loop through the data blocks, setting up read nodes for each.
350 1.3 oster */
351 1.3 oster for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
352 1.3 oster RF_ASSERT(pda);
353 1.3 oster
354 1.3 oster rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
355 1.3 oster
356 1.3 oster RF_ASSERT(pda->numSector != 0);
357 1.3 oster if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
358 1.3 oster /* cannot verify parity with dead disk */
359 1.3 oster goto done;
360 1.3 oster }
361 1.3 oster pda->bufPtr = buf1;
362 1.3 oster blockNode->succedents[i]->params[0].p = pda;
363 1.3 oster blockNode->succedents[i]->params[1].p = buf1;
364 1.3 oster blockNode->succedents[i]->params[2].v = psID;
365 1.3 oster blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
366 1.3 oster buf1 += nbytes;
367 1.3 oster }
368 1.3 oster RF_ASSERT(pda == NULL);
369 1.3 oster /*
370 1.3 oster * keep i, buf1 running
371 1.3 oster *
372 1.3 oster * Loop through parity blocks, setting up read nodes for each.
373 1.3 oster */
374 1.3 oster for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
375 1.3 oster RF_ASSERT(pda);
376 1.3 oster rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
377 1.3 oster RF_ASSERT(pda->numSector != 0);
378 1.3 oster if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
379 1.3 oster /* cannot verify parity with dead disk */
380 1.3 oster goto done;
381 1.3 oster }
382 1.3 oster pda->bufPtr = buf1;
383 1.3 oster blockNode->succedents[i]->params[0].p = pda;
384 1.3 oster blockNode->succedents[i]->params[1].p = buf1;
385 1.3 oster blockNode->succedents[i]->params[2].v = psID;
386 1.3 oster blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
387 1.3 oster buf1 += nbytes;
388 1.3 oster }
389 1.3 oster RF_ASSERT(pda == NULL);
390 1.3 oster
391 1.6 thorpej memset((char *) &tracerec, 0, sizeof(tracerec));
392 1.3 oster rd_dag_h->tracerec = &tracerec;
393 1.3 oster
394 1.3 oster if (rf_verifyParityDebug > 1) {
395 1.5 oster printf("raid%d: RAID1 parity verify read dag:\n",
396 1.5 oster raidPtr->raidid);
397 1.3 oster rf_PrintDAGList(rd_dag_h);
398 1.3 oster }
399 1.3 oster RF_LOCK_MUTEX(mcpair->mutex);
400 1.3 oster mcpair->flag = 0;
401 1.3 oster rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
402 1.3 oster (void *) mcpair);
403 1.3 oster while (mcpair->flag == 0) {
404 1.3 oster RF_WAIT_MCPAIR(mcpair);
405 1.3 oster }
406 1.3 oster RF_UNLOCK_MUTEX(mcpair->mutex);
407 1.3 oster
408 1.3 oster if (rd_dag_h->status != rf_enable) {
409 1.3 oster RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
410 1.3 oster ret = RF_PARITY_COULD_NOT_VERIFY;
411 1.3 oster goto done;
412 1.3 oster }
413 1.3 oster /*
414 1.3 oster * buf1 is the beginning of the data blocks chunk
415 1.3 oster * buf2 is the beginning of the parity blocks chunk
416 1.3 oster */
417 1.3 oster buf1 = buf;
418 1.3 oster buf2 = buf + (nbytes * layoutPtr->numDataCol);
419 1.3 oster ret = RF_PARITY_OKAY;
420 1.3 oster /*
421 1.3 oster * bbufs is "bad bufs"- an array whose entries are the data
422 1.3 oster * column numbers where we had miscompares. (That is, column 0
423 1.3 oster * and column 1 of the array are mirror copies, and are considered
424 1.3 oster * "data column 0" for this purpose).
425 1.3 oster */
426 1.3 oster RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *),
427 1.3 oster allocList);
428 1.3 oster nbad = 0;
429 1.3 oster /*
430 1.3 oster * Check data vs "parity" (mirror copy).
431 1.3 oster */
432 1.3 oster for (i = 0; i < layoutPtr->numDataCol; i++) {
433 1.3 oster if (rf_verifyParityDebug) {
434 1.5 oster printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
435 1.5 oster raidPtr->raidid, nbytes, i, (long) buf1,
436 1.5 oster (long) buf2, (long) buf);
437 1.3 oster }
438 1.7 thorpej ret = memcmp(buf1, buf2, nbytes);
439 1.3 oster if (ret) {
440 1.3 oster if (rf_verifyParityDebug > 1) {
441 1.3 oster for (j = 0; j < nbytes; j++) {
442 1.3 oster if (buf1[j] != buf2[j])
443 1.3 oster break;
444 1.3 oster }
445 1.3 oster printf("psid=%ld j=%d\n", (long) psID, j);
446 1.3 oster printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff,
447 1.3 oster buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff);
448 1.3 oster printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff,
449 1.3 oster buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff);
450 1.3 oster }
451 1.3 oster if (rf_verifyParityDebug) {
452 1.5 oster printf("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, i);
453 1.3 oster }
454 1.3 oster /*
455 1.3 oster * Parity is bad. Keep track of which columns were bad.
456 1.3 oster */
457 1.3 oster if (bbufs)
458 1.3 oster bbufs[nbad] = i;
459 1.3 oster nbad++;
460 1.3 oster ret = RF_PARITY_BAD;
461 1.3 oster }
462 1.3 oster buf1 += nbytes;
463 1.3 oster buf2 += nbytes;
464 1.3 oster }
465 1.3 oster
466 1.3 oster if ((ret != RF_PARITY_OKAY) && correct_it) {
467 1.3 oster ret = RF_PARITY_COULD_NOT_CORRECT;
468 1.3 oster if (rf_verifyParityDebug) {
469 1.5 oster printf("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid);
470 1.3 oster }
471 1.3 oster if (bbufs == NULL)
472 1.3 oster goto done;
473 1.3 oster /*
474 1.3 oster * Make a DAG with one write node for each bad unit. We'll simply
475 1.3 oster * write the contents of the data unit onto the parity unit for
476 1.3 oster * correction. (It's possible that the mirror copy was the correct
477 1.3 oster * copy, and that we're spooging good data by writing bad over it,
478 1.3 oster * but there's no way we can know that.
479 1.3 oster */
480 1.3 oster wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, buf,
481 1.3 oster rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags,
482 1.3 oster RF_IO_NORMAL_PRIORITY);
483 1.3 oster if (wr_dag_h == NULL)
484 1.3 oster goto done;
485 1.3 oster wrBlock = wr_dag_h->succedents[0];
486 1.3 oster /*
487 1.3 oster * Fill in a write node for each bad compare.
488 1.3 oster */
489 1.3 oster for (i = 0; i < nbad; i++) {
490 1.3 oster j = i + layoutPtr->numDataCol;
491 1.3 oster pda = blockNode->succedents[j]->params[0].p;
492 1.3 oster pda->bufPtr = blockNode->succedents[i]->params[1].p;
493 1.3 oster wrBlock->succedents[i]->params[0].p = pda;
494 1.3 oster wrBlock->succedents[i]->params[1].p = pda->bufPtr;
495 1.3 oster wrBlock->succedents[i]->params[2].v = psID;
496 1.3 oster wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
497 1.3 oster }
498 1.6 thorpej memset((char *) &tracerec, 0, sizeof(tracerec));
499 1.3 oster wr_dag_h->tracerec = &tracerec;
500 1.3 oster if (rf_verifyParityDebug > 1) {
501 1.3 oster printf("Parity verify write dag:\n");
502 1.3 oster rf_PrintDAGList(wr_dag_h);
503 1.3 oster }
504 1.3 oster RF_LOCK_MUTEX(mcpair->mutex);
505 1.3 oster mcpair->flag = 0;
506 1.3 oster /* fire off the write DAG */
507 1.3 oster rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
508 1.3 oster (void *) mcpair);
509 1.3 oster while (!mcpair->flag) {
510 1.3 oster RF_WAIT_COND(mcpair->cond, mcpair->mutex);
511 1.3 oster }
512 1.3 oster RF_UNLOCK_MUTEX(mcpair->mutex);
513 1.3 oster if (wr_dag_h->status != rf_enable) {
514 1.3 oster RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n");
515 1.3 oster goto done;
516 1.3 oster }
517 1.3 oster ret = RF_PARITY_CORRECTED;
518 1.3 oster }
519 1.1 oster done:
520 1.3 oster /*
521 1.3 oster * All done. We might've gotten here without doing part of the function,
522 1.3 oster * so cleanup what we have to and return our running status.
523 1.3 oster */
524 1.3 oster if (asm_h)
525 1.3 oster rf_FreeAccessStripeMap(asm_h);
526 1.3 oster if (rd_dag_h)
527 1.3 oster rf_FreeDAG(rd_dag_h);
528 1.3 oster if (wr_dag_h)
529 1.3 oster rf_FreeDAG(wr_dag_h);
530 1.3 oster if (mcpair)
531 1.3 oster rf_FreeMCPair(mcpair);
532 1.3 oster rf_FreeAllocList(allocList);
533 1.3 oster if (rf_verifyParityDebug) {
534 1.5 oster printf("raid%d: RAID1 parity verify, returning %d\n",
535 1.5 oster raidPtr->raidid, ret);
536 1.3 oster }
537 1.3 oster return (ret);
538 1.1 oster }
539 1.1 oster
540 1.3 oster int
541 1.3 oster rf_SubmitReconBufferRAID1(rbuf, keep_it, use_committed)
542 1.3 oster RF_ReconBuffer_t *rbuf; /* the recon buffer to submit */
543 1.3 oster int keep_it; /* whether we can keep this buffer or we have
544 1.3 oster * to return it */
545 1.3 oster int use_committed; /* whether to use a committed or an available
546 1.3 oster * recon buffer */
547 1.1 oster {
548 1.3 oster RF_ReconParityStripeStatus_t *pssPtr;
549 1.3 oster RF_ReconCtrl_t *reconCtrlPtr;
550 1.3 oster RF_RaidLayout_t *layoutPtr;
551 1.5 oster int retcode, created;
552 1.3 oster RF_CallbackDesc_t *cb, *p;
553 1.3 oster RF_ReconBuffer_t *t;
554 1.3 oster RF_Raid_t *raidPtr;
555 1.3 oster caddr_t ta;
556 1.3 oster
557 1.3 oster retcode = 0;
558 1.3 oster created = 0;
559 1.3 oster
560 1.3 oster raidPtr = rbuf->raidPtr;
561 1.3 oster layoutPtr = &raidPtr->Layout;
562 1.3 oster reconCtrlPtr = raidPtr->reconControl[rbuf->row];
563 1.3 oster
564 1.3 oster RF_ASSERT(rbuf);
565 1.3 oster RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
566 1.3 oster
567 1.3 oster if (rf_reconbufferDebug) {
568 1.5 oster printf("raid%d: RAID1 reconbuffer submission r%d c%d psid %ld ru%d (failed offset %ld)\n",
569 1.5 oster raidPtr->raidid, rbuf->row, rbuf->col,
570 1.5 oster (long) rbuf->parityStripeID, rbuf->which_ru,
571 1.5 oster (long) rbuf->failedDiskSectorOffset);
572 1.3 oster }
573 1.3 oster if (rf_reconDebug) {
574 1.3 oster printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
575 1.3 oster (long) rbuf->parityStripeID, (long) rbuf->buffer);
576 1.3 oster printf("RAID1 psid %ld %02x %02x %02x %02x %02x\n",
577 1.3 oster (long) rbuf->parityStripeID,
578 1.3 oster rbuf->buffer[0], rbuf->buffer[1], rbuf->buffer[2], rbuf->buffer[3],
579 1.3 oster rbuf->buffer[4]);
580 1.3 oster }
581 1.3 oster RF_LOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
582 1.3 oster
583 1.3 oster RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
584 1.3 oster
585 1.3 oster pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
586 1.3 oster rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, &created);
587 1.3 oster RF_ASSERT(pssPtr); /* if it didn't exist, we wouldn't have gotten
588 1.3 oster * an rbuf for it */
589 1.3 oster
590 1.3 oster /*
591 1.3 oster * Since this is simple mirroring, the first submission for a stripe is also
592 1.3 oster * treated as the last.
593 1.3 oster */
594 1.3 oster
595 1.3 oster t = NULL;
596 1.3 oster if (keep_it) {
597 1.3 oster if (rf_reconbufferDebug) {
598 1.5 oster printf("raid%d: RAID1 rbuf submission: keeping rbuf\n",
599 1.5 oster raidPtr->raidid);
600 1.3 oster }
601 1.3 oster t = rbuf;
602 1.3 oster } else {
603 1.3 oster if (use_committed) {
604 1.3 oster if (rf_reconbufferDebug) {
605 1.5 oster printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid);
606 1.3 oster }
607 1.3 oster t = reconCtrlPtr->committedRbufs;
608 1.3 oster RF_ASSERT(t);
609 1.3 oster reconCtrlPtr->committedRbufs = t->next;
610 1.3 oster t->next = NULL;
611 1.3 oster } else
612 1.3 oster if (reconCtrlPtr->floatingRbufs) {
613 1.3 oster if (rf_reconbufferDebug) {
614 1.5 oster printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid);
615 1.3 oster }
616 1.3 oster t = reconCtrlPtr->floatingRbufs;
617 1.3 oster reconCtrlPtr->floatingRbufs = t->next;
618 1.3 oster t->next = NULL;
619 1.3 oster }
620 1.3 oster }
621 1.3 oster if (t == NULL) {
622 1.3 oster if (rf_reconbufferDebug) {
623 1.5 oster printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid);
624 1.3 oster }
625 1.3 oster RF_ASSERT((keep_it == 0) && (use_committed == 0));
626 1.3 oster raidPtr->procsInBufWait++;
627 1.3 oster if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
628 1.3 oster && (raidPtr->numFullReconBuffers == 0)) {
629 1.3 oster /* ruh-ro */
630 1.3 oster RF_ERRORMSG("Buffer wait deadlock\n");
631 1.3 oster rf_PrintPSStatusTable(raidPtr, rbuf->row);
632 1.3 oster RF_PANIC();
633 1.3 oster }
634 1.3 oster pssPtr->flags |= RF_PSS_BUFFERWAIT;
635 1.3 oster cb = rf_AllocCallbackDesc();
636 1.3 oster cb->row = rbuf->row;
637 1.3 oster cb->col = rbuf->col;
638 1.3 oster cb->callbackArg.v = rbuf->parityStripeID;
639 1.3 oster cb->callbackArg2.v = rbuf->which_ru;
640 1.3 oster cb->next = NULL;
641 1.3 oster if (reconCtrlPtr->bufferWaitList == NULL) {
642 1.3 oster /* we are the wait list- lucky us */
643 1.3 oster reconCtrlPtr->bufferWaitList = cb;
644 1.3 oster } else {
645 1.3 oster /* append to wait list */
646 1.3 oster for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
647 1.3 oster p->next = cb;
648 1.3 oster }
649 1.3 oster retcode = 1;
650 1.3 oster goto out;
651 1.3 oster }
652 1.3 oster if (t != rbuf) {
653 1.3 oster t->row = rbuf->row;
654 1.3 oster t->col = reconCtrlPtr->fcol;
655 1.3 oster t->parityStripeID = rbuf->parityStripeID;
656 1.3 oster t->which_ru = rbuf->which_ru;
657 1.3 oster t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
658 1.3 oster t->spRow = rbuf->spRow;
659 1.3 oster t->spCol = rbuf->spCol;
660 1.3 oster t->spOffset = rbuf->spOffset;
661 1.3 oster /* Swap buffers. DANCE! */
662 1.3 oster ta = t->buffer;
663 1.3 oster t->buffer = rbuf->buffer;
664 1.3 oster rbuf->buffer = ta;
665 1.3 oster }
666 1.3 oster /*
667 1.3 oster * Use the rbuf we've been given as the target.
668 1.3 oster */
669 1.3 oster RF_ASSERT(pssPtr->rbuf == NULL);
670 1.3 oster pssPtr->rbuf = t;
671 1.3 oster
672 1.3 oster t->count = 1;
673 1.3 oster /*
674 1.3 oster * Below, we use 1 for numDataCol (which is equal to the count in the
675 1.3 oster * previous line), so we'll always be done.
676 1.3 oster */
677 1.3 oster rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
678 1.1 oster
679 1.1 oster out:
680 1.3 oster RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
681 1.3 oster RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
682 1.3 oster if (rf_reconbufferDebug) {
683 1.5 oster printf("raid%d: RAID1 rbuf submission: returning %d\n",
684 1.5 oster raidPtr->raidid, retcode);
685 1.3 oster }
686 1.3 oster return (retcode);
687 1.1 oster }
688