rf_raid1.c revision 1.13 1 1.13 oster /* $NetBSD: rf_raid1.c,v 1.13 2002/09/23 03:38:51 oster 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 1.8 lukem
35 1.8 lukem #include <sys/cdefs.h>
36 1.13 oster __KERNEL_RCSID(0, "$NetBSD: rf_raid1.c,v 1.13 2002/09/23 03:38:51 oster Exp $");
37 1.1 oster
38 1.1 oster #include "rf_raid.h"
39 1.1 oster #include "rf_raid1.h"
40 1.1 oster #include "rf_dag.h"
41 1.1 oster #include "rf_dagffrd.h"
42 1.1 oster #include "rf_dagffwr.h"
43 1.1 oster #include "rf_dagdegrd.h"
44 1.1 oster #include "rf_dagutils.h"
45 1.1 oster #include "rf_dagfuncs.h"
46 1.1 oster #include "rf_diskqueue.h"
47 1.1 oster #include "rf_general.h"
48 1.1 oster #include "rf_utils.h"
49 1.1 oster #include "rf_parityscan.h"
50 1.1 oster #include "rf_mcpair.h"
51 1.1 oster #include "rf_layout.h"
52 1.1 oster #include "rf_map.h"
53 1.1 oster #include "rf_engine.h"
54 1.1 oster #include "rf_reconbuffer.h"
55 1.1 oster
56 1.1 oster typedef struct RF_Raid1ConfigInfo_s {
57 1.3 oster RF_RowCol_t **stripeIdentifier;
58 1.3 oster } RF_Raid1ConfigInfo_t;
59 1.1 oster /* start of day code specific to RAID level 1 */
60 1.3 oster int
61 1.3 oster rf_ConfigureRAID1(
62 1.3 oster RF_ShutdownList_t ** listp,
63 1.3 oster RF_Raid_t * raidPtr,
64 1.3 oster RF_Config_t * cfgPtr)
65 1.1 oster {
66 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
67 1.3 oster RF_Raid1ConfigInfo_t *info;
68 1.3 oster RF_RowCol_t i;
69 1.3 oster
70 1.3 oster /* create a RAID level 1 configuration structure */
71 1.3 oster RF_MallocAndAdd(info, sizeof(RF_Raid1ConfigInfo_t), (RF_Raid1ConfigInfo_t *), raidPtr->cleanupList);
72 1.3 oster if (info == NULL)
73 1.3 oster return (ENOMEM);
74 1.3 oster layoutPtr->layoutSpecificInfo = (void *) info;
75 1.3 oster
76 1.3 oster /* ... and fill it in. */
77 1.3 oster info->stripeIdentifier = rf_make_2d_array(raidPtr->numCol / 2, 2, raidPtr->cleanupList);
78 1.3 oster if (info->stripeIdentifier == NULL)
79 1.3 oster return (ENOMEM);
80 1.3 oster for (i = 0; i < (raidPtr->numCol / 2); i++) {
81 1.3 oster info->stripeIdentifier[i][0] = (2 * i);
82 1.3 oster info->stripeIdentifier[i][1] = (2 * i) + 1;
83 1.3 oster }
84 1.3 oster
85 1.3 oster RF_ASSERT(raidPtr->numRow == 1);
86 1.3 oster
87 1.3 oster /* this implementation of RAID level 1 uses one row of numCol disks
88 1.3 oster * and allows multiple (numCol / 2) stripes per row. A stripe
89 1.3 oster * consists of a single data unit and a single parity (mirror) unit.
90 1.3 oster * stripe id = raidAddr / stripeUnitSize */
91 1.3 oster raidPtr->totalSectors = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2) * layoutPtr->sectorsPerStripeUnit;
92 1.3 oster layoutPtr->numStripe = layoutPtr->stripeUnitsPerDisk * (raidPtr->numCol / 2);
93 1.3 oster layoutPtr->dataSectorsPerStripe = layoutPtr->sectorsPerStripeUnit;
94 1.3 oster layoutPtr->numDataCol = 1;
95 1.3 oster layoutPtr->numParityCol = 1;
96 1.3 oster return (0);
97 1.1 oster }
98 1.1 oster
99 1.1 oster
100 1.1 oster /* returns the physical disk location of the primary copy in the mirror pair */
101 1.3 oster void
102 1.3 oster rf_MapSectorRAID1(
103 1.3 oster RF_Raid_t * raidPtr,
104 1.3 oster RF_RaidAddr_t raidSector,
105 1.3 oster RF_RowCol_t * row,
106 1.3 oster RF_RowCol_t * col,
107 1.3 oster RF_SectorNum_t * diskSector,
108 1.3 oster int remap)
109 1.1 oster {
110 1.3 oster RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
111 1.3 oster RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
112 1.1 oster
113 1.3 oster *row = 0;
114 1.3 oster *col = 2 * mirrorPair;
115 1.3 oster *diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
116 1.1 oster }
117 1.1 oster
118 1.1 oster
119 1.1 oster /* Map Parity
120 1.1 oster *
121 1.1 oster * returns the physical disk location of the secondary copy in the mirror
122 1.1 oster * pair
123 1.1 oster */
124 1.3 oster void
125 1.3 oster rf_MapParityRAID1(
126 1.3 oster RF_Raid_t * raidPtr,
127 1.3 oster RF_RaidAddr_t raidSector,
128 1.3 oster RF_RowCol_t * row,
129 1.3 oster RF_RowCol_t * col,
130 1.3 oster RF_SectorNum_t * diskSector,
131 1.3 oster int remap)
132 1.1 oster {
133 1.3 oster RF_StripeNum_t SUID = raidSector / raidPtr->Layout.sectorsPerStripeUnit;
134 1.3 oster RF_RowCol_t mirrorPair = SUID % (raidPtr->numCol / 2);
135 1.1 oster
136 1.3 oster *row = 0;
137 1.3 oster *col = (2 * mirrorPair) + 1;
138 1.1 oster
139 1.3 oster *diskSector = ((SUID / (raidPtr->numCol / 2)) * raidPtr->Layout.sectorsPerStripeUnit) + (raidSector % raidPtr->Layout.sectorsPerStripeUnit);
140 1.1 oster }
141 1.1 oster
142 1.1 oster
143 1.1 oster /* IdentifyStripeRAID1
144 1.1 oster *
145 1.1 oster * returns a list of disks for a given redundancy group
146 1.1 oster */
147 1.3 oster void
148 1.3 oster rf_IdentifyStripeRAID1(
149 1.3 oster RF_Raid_t * raidPtr,
150 1.3 oster RF_RaidAddr_t addr,
151 1.3 oster RF_RowCol_t ** diskids,
152 1.3 oster RF_RowCol_t * outRow)
153 1.1 oster {
154 1.3 oster RF_StripeNum_t stripeID = rf_RaidAddressToStripeID(&raidPtr->Layout, addr);
155 1.3 oster RF_Raid1ConfigInfo_t *info = raidPtr->Layout.layoutSpecificInfo;
156 1.3 oster RF_ASSERT(stripeID >= 0);
157 1.3 oster RF_ASSERT(addr >= 0);
158 1.3 oster *outRow = 0;
159 1.3 oster *diskids = info->stripeIdentifier[stripeID % (raidPtr->numCol / 2)];
160 1.3 oster RF_ASSERT(*diskids);
161 1.1 oster }
162 1.1 oster
163 1.1 oster
164 1.1 oster /* MapSIDToPSIDRAID1
165 1.1 oster *
166 1.1 oster * maps a logical stripe to a stripe in the redundant array
167 1.1 oster */
168 1.3 oster void
169 1.3 oster rf_MapSIDToPSIDRAID1(
170 1.3 oster RF_RaidLayout_t * layoutPtr,
171 1.3 oster RF_StripeNum_t stripeID,
172 1.3 oster RF_StripeNum_t * psID,
173 1.3 oster RF_ReconUnitNum_t * which_ru)
174 1.1 oster {
175 1.3 oster *which_ru = 0;
176 1.3 oster *psID = stripeID;
177 1.1 oster }
178 1.1 oster
179 1.1 oster
180 1.1 oster
181 1.1 oster /******************************************************************************
182 1.1 oster * select a graph to perform a single-stripe access
183 1.1 oster *
184 1.1 oster * Parameters: raidPtr - description of the physical array
185 1.1 oster * type - type of operation (read or write) requested
186 1.1 oster * asmap - logical & physical addresses for this access
187 1.1 oster * createFunc - name of function to use to create the graph
188 1.1 oster *****************************************************************************/
189 1.1 oster
190 1.3 oster void
191 1.3 oster rf_RAID1DagSelect(
192 1.3 oster RF_Raid_t * raidPtr,
193 1.3 oster RF_IoType_t type,
194 1.3 oster RF_AccessStripeMap_t * asmap,
195 1.3 oster RF_VoidFuncPtr * createFunc)
196 1.1 oster {
197 1.3 oster RF_RowCol_t frow, fcol, or, oc;
198 1.3 oster RF_PhysDiskAddr_t *failedPDA;
199 1.5 oster int prior_recon;
200 1.3 oster RF_RowStatus_t rstat;
201 1.3 oster RF_SectorNum_t oo;
202 1.3 oster
203 1.3 oster
204 1.3 oster RF_ASSERT(RF_IO_IS_R_OR_W(type));
205 1.3 oster
206 1.3 oster if (asmap->numDataFailed + asmap->numParityFailed > 1) {
207 1.3 oster RF_ERRORMSG("Multiple disks failed in a single group! Aborting I/O operation.\n");
208 1.3 oster *createFunc = NULL;
209 1.3 oster return;
210 1.3 oster }
211 1.3 oster if (asmap->numDataFailed + asmap->numParityFailed) {
212 1.3 oster /*
213 1.3 oster * We've got a fault. Re-map to spare space, iff applicable.
214 1.3 oster * Shouldn't the arch-independent code do this for us?
215 1.3 oster * Anyway, it turns out if we don't do this here, then when
216 1.3 oster * we're reconstructing, writes go only to the surviving
217 1.3 oster * original disk, and aren't reflected on the reconstructed
218 1.3 oster * spare. Oops. --jimz
219 1.3 oster */
220 1.3 oster failedPDA = asmap->failedPDAs[0];
221 1.3 oster frow = failedPDA->row;
222 1.3 oster fcol = failedPDA->col;
223 1.3 oster rstat = raidPtr->status[frow];
224 1.3 oster prior_recon = (rstat == rf_rs_reconfigured) || (
225 1.3 oster (rstat == rf_rs_reconstructing) ?
226 1.3 oster rf_CheckRUReconstructed(raidPtr->reconControl[frow]->reconMap, failedPDA->startSector) : 0
227 1.3 oster );
228 1.3 oster if (prior_recon) {
229 1.3 oster or = frow;
230 1.3 oster oc = fcol;
231 1.3 oster oo = failedPDA->startSector;
232 1.3 oster /*
233 1.3 oster * If we did distributed sparing, we'd monkey with that here.
234 1.3 oster * But we don't, so we'll
235 1.3 oster */
236 1.3 oster failedPDA->row = raidPtr->Disks[frow][fcol].spareRow;
237 1.3 oster failedPDA->col = raidPtr->Disks[frow][fcol].spareCol;
238 1.3 oster /*
239 1.3 oster * Redirect other components, iff necessary. This looks
240 1.3 oster * pretty suspicious to me, but it's what the raid5
241 1.3 oster * DAG select does.
242 1.3 oster */
243 1.3 oster if (asmap->parityInfo->next) {
244 1.3 oster if (failedPDA == asmap->parityInfo) {
245 1.3 oster failedPDA->next->row = failedPDA->row;
246 1.3 oster failedPDA->next->col = failedPDA->col;
247 1.3 oster } else {
248 1.3 oster if (failedPDA == asmap->parityInfo->next) {
249 1.3 oster asmap->parityInfo->row = failedPDA->row;
250 1.3 oster asmap->parityInfo->col = failedPDA->col;
251 1.3 oster }
252 1.3 oster }
253 1.3 oster }
254 1.3 oster if (rf_dagDebug || rf_mapDebug) {
255 1.5 oster printf("raid%d: Redirected type '%c' r %d c %d o %ld -> r %d c %d o %ld\n",
256 1.5 oster raidPtr->raidid, type, or, oc,
257 1.5 oster (long) oo, failedPDA->row,
258 1.5 oster failedPDA->col,
259 1.5 oster (long) failedPDA->startSector);
260 1.3 oster }
261 1.3 oster asmap->numDataFailed = asmap->numParityFailed = 0;
262 1.3 oster }
263 1.3 oster }
264 1.3 oster if (type == RF_IO_TYPE_READ) {
265 1.3 oster if (asmap->numDataFailed == 0)
266 1.3 oster *createFunc = (RF_VoidFuncPtr) rf_CreateMirrorIdleReadDAG;
267 1.3 oster else
268 1.3 oster *createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneDegradedReadDAG;
269 1.3 oster } else {
270 1.3 oster *createFunc = (RF_VoidFuncPtr) rf_CreateRaidOneWriteDAG;
271 1.3 oster }
272 1.1 oster }
273 1.1 oster
274 1.3 oster int
275 1.3 oster rf_VerifyParityRAID1(
276 1.3 oster RF_Raid_t * raidPtr,
277 1.3 oster RF_RaidAddr_t raidAddr,
278 1.3 oster RF_PhysDiskAddr_t * parityPDA,
279 1.3 oster int correct_it,
280 1.3 oster RF_RaidAccessFlags_t flags)
281 1.1 oster {
282 1.5 oster int nbytes, bcount, stripeWidth, ret, i, j, nbad, *bbufs;
283 1.13 oster RF_DagNode_t *blockNode, *wrBlock;
284 1.3 oster RF_DagHeader_t *rd_dag_h, *wr_dag_h;
285 1.3 oster RF_AccessStripeMapHeader_t *asm_h;
286 1.3 oster RF_AllocListElem_t *allocList;
287 1.3 oster RF_AccTraceEntry_t tracerec;
288 1.3 oster RF_ReconUnitNum_t which_ru;
289 1.3 oster RF_RaidLayout_t *layoutPtr;
290 1.3 oster RF_AccessStripeMap_t *aasm;
291 1.3 oster RF_SectorCount_t nsector;
292 1.3 oster RF_RaidAddr_t startAddr;
293 1.3 oster char *buf, *buf1, *buf2;
294 1.3 oster RF_PhysDiskAddr_t *pda;
295 1.3 oster RF_StripeNum_t psID;
296 1.3 oster RF_MCPair_t *mcpair;
297 1.3 oster
298 1.3 oster layoutPtr = &raidPtr->Layout;
299 1.3 oster startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
300 1.3 oster nsector = parityPDA->numSector;
301 1.3 oster nbytes = rf_RaidAddressToByte(raidPtr, nsector);
302 1.3 oster psID = rf_RaidAddressToParityStripeID(layoutPtr, raidAddr, &which_ru);
303 1.3 oster
304 1.3 oster asm_h = NULL;
305 1.3 oster rd_dag_h = wr_dag_h = NULL;
306 1.3 oster mcpair = NULL;
307 1.3 oster
308 1.3 oster ret = RF_PARITY_COULD_NOT_VERIFY;
309 1.3 oster
310 1.3 oster rf_MakeAllocList(allocList);
311 1.3 oster if (allocList == NULL)
312 1.3 oster return (RF_PARITY_COULD_NOT_VERIFY);
313 1.3 oster mcpair = rf_AllocMCPair();
314 1.3 oster if (mcpair == NULL)
315 1.3 oster goto done;
316 1.3 oster RF_ASSERT(layoutPtr->numDataCol == layoutPtr->numParityCol);
317 1.3 oster stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
318 1.3 oster bcount = nbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol);
319 1.3 oster RF_MallocAndAdd(buf, bcount, (char *), allocList);
320 1.3 oster if (buf == NULL)
321 1.3 oster goto done;
322 1.10 oster #if RF_DEBUG_VERIFYPARITY
323 1.3 oster if (rf_verifyParityDebug) {
324 1.5 oster printf("raid%d: RAID1 parity verify: buf=%lx bcount=%d (%lx - %lx)\n",
325 1.5 oster raidPtr->raidid, (long) buf, bcount, (long) buf,
326 1.5 oster (long) buf + bcount);
327 1.3 oster }
328 1.10 oster #endif
329 1.3 oster /*
330 1.3 oster * Generate a DAG which will read the entire stripe- then we can
331 1.3 oster * just compare data chunks versus "parity" chunks.
332 1.3 oster */
333 1.3 oster
334 1.3 oster rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, nbytes, buf,
335 1.3 oster rf_DiskReadFunc, rf_DiskReadUndoFunc, "Rod", allocList, flags,
336 1.3 oster RF_IO_NORMAL_PRIORITY);
337 1.3 oster if (rd_dag_h == NULL)
338 1.3 oster goto done;
339 1.3 oster blockNode = rd_dag_h->succedents[0];
340 1.3 oster
341 1.3 oster /*
342 1.3 oster * Map the access to physical disk addresses (PDAs)- this will
343 1.3 oster * get us both a list of data addresses, and "parity" addresses
344 1.3 oster * (which are really mirror copies).
345 1.3 oster */
346 1.3 oster asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe,
347 1.3 oster buf, RF_DONT_REMAP);
348 1.3 oster aasm = asm_h->stripeMap;
349 1.3 oster
350 1.3 oster buf1 = buf;
351 1.3 oster /*
352 1.3 oster * Loop through the data blocks, setting up read nodes for each.
353 1.3 oster */
354 1.3 oster for (pda = aasm->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
355 1.3 oster RF_ASSERT(pda);
356 1.3 oster
357 1.3 oster rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
358 1.3 oster
359 1.3 oster RF_ASSERT(pda->numSector != 0);
360 1.3 oster if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
361 1.3 oster /* cannot verify parity with dead disk */
362 1.3 oster goto done;
363 1.3 oster }
364 1.3 oster pda->bufPtr = buf1;
365 1.3 oster blockNode->succedents[i]->params[0].p = pda;
366 1.3 oster blockNode->succedents[i]->params[1].p = buf1;
367 1.3 oster blockNode->succedents[i]->params[2].v = psID;
368 1.3 oster blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
369 1.3 oster buf1 += nbytes;
370 1.3 oster }
371 1.3 oster RF_ASSERT(pda == NULL);
372 1.3 oster /*
373 1.3 oster * keep i, buf1 running
374 1.3 oster *
375 1.3 oster * Loop through parity blocks, setting up read nodes for each.
376 1.3 oster */
377 1.3 oster for (pda = aasm->parityInfo; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++, pda = pda->next) {
378 1.3 oster RF_ASSERT(pda);
379 1.3 oster rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
380 1.3 oster RF_ASSERT(pda->numSector != 0);
381 1.3 oster if (rf_TryToRedirectPDA(raidPtr, pda, 0)) {
382 1.3 oster /* cannot verify parity with dead disk */
383 1.3 oster goto done;
384 1.3 oster }
385 1.3 oster pda->bufPtr = buf1;
386 1.3 oster blockNode->succedents[i]->params[0].p = pda;
387 1.3 oster blockNode->succedents[i]->params[1].p = buf1;
388 1.3 oster blockNode->succedents[i]->params[2].v = psID;
389 1.3 oster blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
390 1.3 oster buf1 += nbytes;
391 1.3 oster }
392 1.3 oster RF_ASSERT(pda == NULL);
393 1.3 oster
394 1.6 thorpej memset((char *) &tracerec, 0, sizeof(tracerec));
395 1.3 oster rd_dag_h->tracerec = &tracerec;
396 1.3 oster
397 1.9 oster #if 0
398 1.3 oster if (rf_verifyParityDebug > 1) {
399 1.5 oster printf("raid%d: RAID1 parity verify read dag:\n",
400 1.5 oster raidPtr->raidid);
401 1.3 oster rf_PrintDAGList(rd_dag_h);
402 1.3 oster }
403 1.9 oster #endif
404 1.3 oster RF_LOCK_MUTEX(mcpair->mutex);
405 1.3 oster mcpair->flag = 0;
406 1.3 oster rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
407 1.3 oster (void *) mcpair);
408 1.3 oster while (mcpair->flag == 0) {
409 1.3 oster RF_WAIT_MCPAIR(mcpair);
410 1.3 oster }
411 1.3 oster RF_UNLOCK_MUTEX(mcpair->mutex);
412 1.3 oster
413 1.3 oster if (rd_dag_h->status != rf_enable) {
414 1.3 oster RF_ERRORMSG("Unable to verify raid1 parity: can't read stripe\n");
415 1.3 oster ret = RF_PARITY_COULD_NOT_VERIFY;
416 1.3 oster goto done;
417 1.3 oster }
418 1.3 oster /*
419 1.3 oster * buf1 is the beginning of the data blocks chunk
420 1.3 oster * buf2 is the beginning of the parity blocks chunk
421 1.3 oster */
422 1.3 oster buf1 = buf;
423 1.3 oster buf2 = buf + (nbytes * layoutPtr->numDataCol);
424 1.3 oster ret = RF_PARITY_OKAY;
425 1.3 oster /*
426 1.3 oster * bbufs is "bad bufs"- an array whose entries are the data
427 1.3 oster * column numbers where we had miscompares. (That is, column 0
428 1.3 oster * and column 1 of the array are mirror copies, and are considered
429 1.3 oster * "data column 0" for this purpose).
430 1.3 oster */
431 1.3 oster RF_MallocAndAdd(bbufs, layoutPtr->numParityCol * sizeof(int), (int *),
432 1.3 oster allocList);
433 1.3 oster nbad = 0;
434 1.3 oster /*
435 1.3 oster * Check data vs "parity" (mirror copy).
436 1.3 oster */
437 1.3 oster for (i = 0; i < layoutPtr->numDataCol; i++) {
438 1.10 oster #if RF_DEBUG_VERIFYPARITY
439 1.3 oster if (rf_verifyParityDebug) {
440 1.5 oster printf("raid%d: RAID1 parity verify %d bytes: i=%d buf1=%lx buf2=%lx buf=%lx\n",
441 1.5 oster raidPtr->raidid, nbytes, i, (long) buf1,
442 1.5 oster (long) buf2, (long) buf);
443 1.3 oster }
444 1.10 oster #endif
445 1.7 thorpej ret = memcmp(buf1, buf2, nbytes);
446 1.3 oster if (ret) {
447 1.10 oster #if RF_DEBUG_VERIFYPARITY
448 1.3 oster if (rf_verifyParityDebug > 1) {
449 1.3 oster for (j = 0; j < nbytes; j++) {
450 1.3 oster if (buf1[j] != buf2[j])
451 1.3 oster break;
452 1.3 oster }
453 1.3 oster printf("psid=%ld j=%d\n", (long) psID, j);
454 1.3 oster printf("buf1 %02x %02x %02x %02x %02x\n", buf1[0] & 0xff,
455 1.3 oster buf1[1] & 0xff, buf1[2] & 0xff, buf1[3] & 0xff, buf1[4] & 0xff);
456 1.3 oster printf("buf2 %02x %02x %02x %02x %02x\n", buf2[0] & 0xff,
457 1.3 oster buf2[1] & 0xff, buf2[2] & 0xff, buf2[3] & 0xff, buf2[4] & 0xff);
458 1.3 oster }
459 1.3 oster if (rf_verifyParityDebug) {
460 1.5 oster printf("raid%d: RAID1: found bad parity, i=%d\n", raidPtr->raidid, i);
461 1.3 oster }
462 1.10 oster #endif
463 1.3 oster /*
464 1.3 oster * Parity is bad. Keep track of which columns were bad.
465 1.3 oster */
466 1.3 oster if (bbufs)
467 1.3 oster bbufs[nbad] = i;
468 1.3 oster nbad++;
469 1.3 oster ret = RF_PARITY_BAD;
470 1.3 oster }
471 1.3 oster buf1 += nbytes;
472 1.3 oster buf2 += nbytes;
473 1.3 oster }
474 1.3 oster
475 1.3 oster if ((ret != RF_PARITY_OKAY) && correct_it) {
476 1.3 oster ret = RF_PARITY_COULD_NOT_CORRECT;
477 1.10 oster #if RF_DEBUG_VERIFYPARITY
478 1.3 oster if (rf_verifyParityDebug) {
479 1.5 oster printf("raid%d: RAID1 parity verify: parity not correct\n", raidPtr->raidid);
480 1.3 oster }
481 1.10 oster #endif
482 1.3 oster if (bbufs == NULL)
483 1.3 oster goto done;
484 1.3 oster /*
485 1.3 oster * Make a DAG with one write node for each bad unit. We'll simply
486 1.3 oster * write the contents of the data unit onto the parity unit for
487 1.3 oster * correction. (It's possible that the mirror copy was the correct
488 1.3 oster * copy, and that we're spooging good data by writing bad over it,
489 1.3 oster * but there's no way we can know that.
490 1.3 oster */
491 1.3 oster wr_dag_h = rf_MakeSimpleDAG(raidPtr, nbad, nbytes, buf,
492 1.3 oster rf_DiskWriteFunc, rf_DiskWriteUndoFunc, "Wnp", allocList, flags,
493 1.3 oster RF_IO_NORMAL_PRIORITY);
494 1.3 oster if (wr_dag_h == NULL)
495 1.3 oster goto done;
496 1.3 oster wrBlock = wr_dag_h->succedents[0];
497 1.3 oster /*
498 1.3 oster * Fill in a write node for each bad compare.
499 1.3 oster */
500 1.3 oster for (i = 0; i < nbad; i++) {
501 1.3 oster j = i + layoutPtr->numDataCol;
502 1.3 oster pda = blockNode->succedents[j]->params[0].p;
503 1.3 oster pda->bufPtr = blockNode->succedents[i]->params[1].p;
504 1.3 oster wrBlock->succedents[i]->params[0].p = pda;
505 1.3 oster wrBlock->succedents[i]->params[1].p = pda->bufPtr;
506 1.3 oster wrBlock->succedents[i]->params[2].v = psID;
507 1.3 oster wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
508 1.3 oster }
509 1.6 thorpej memset((char *) &tracerec, 0, sizeof(tracerec));
510 1.3 oster wr_dag_h->tracerec = &tracerec;
511 1.9 oster #if 0
512 1.3 oster if (rf_verifyParityDebug > 1) {
513 1.3 oster printf("Parity verify write dag:\n");
514 1.3 oster rf_PrintDAGList(wr_dag_h);
515 1.3 oster }
516 1.9 oster #endif
517 1.3 oster RF_LOCK_MUTEX(mcpair->mutex);
518 1.3 oster mcpair->flag = 0;
519 1.3 oster /* fire off the write DAG */
520 1.3 oster rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
521 1.3 oster (void *) mcpair);
522 1.3 oster while (!mcpair->flag) {
523 1.3 oster RF_WAIT_COND(mcpair->cond, mcpair->mutex);
524 1.3 oster }
525 1.3 oster RF_UNLOCK_MUTEX(mcpair->mutex);
526 1.3 oster if (wr_dag_h->status != rf_enable) {
527 1.3 oster RF_ERRORMSG("Unable to correct RAID1 parity in VerifyParity\n");
528 1.3 oster goto done;
529 1.3 oster }
530 1.3 oster ret = RF_PARITY_CORRECTED;
531 1.3 oster }
532 1.1 oster done:
533 1.3 oster /*
534 1.3 oster * All done. We might've gotten here without doing part of the function,
535 1.3 oster * so cleanup what we have to and return our running status.
536 1.3 oster */
537 1.3 oster if (asm_h)
538 1.3 oster rf_FreeAccessStripeMap(asm_h);
539 1.3 oster if (rd_dag_h)
540 1.3 oster rf_FreeDAG(rd_dag_h);
541 1.3 oster if (wr_dag_h)
542 1.3 oster rf_FreeDAG(wr_dag_h);
543 1.3 oster if (mcpair)
544 1.3 oster rf_FreeMCPair(mcpair);
545 1.3 oster rf_FreeAllocList(allocList);
546 1.10 oster #if RF_DEBUG_VERIFYPARITY
547 1.3 oster if (rf_verifyParityDebug) {
548 1.5 oster printf("raid%d: RAID1 parity verify, returning %d\n",
549 1.5 oster raidPtr->raidid, ret);
550 1.3 oster }
551 1.10 oster #endif
552 1.3 oster return (ret);
553 1.1 oster }
554 1.1 oster
555 1.3 oster int
556 1.3 oster rf_SubmitReconBufferRAID1(rbuf, keep_it, use_committed)
557 1.3 oster RF_ReconBuffer_t *rbuf; /* the recon buffer to submit */
558 1.3 oster int keep_it; /* whether we can keep this buffer or we have
559 1.3 oster * to return it */
560 1.3 oster int use_committed; /* whether to use a committed or an available
561 1.3 oster * recon buffer */
562 1.1 oster {
563 1.3 oster RF_ReconParityStripeStatus_t *pssPtr;
564 1.3 oster RF_ReconCtrl_t *reconCtrlPtr;
565 1.5 oster int retcode, created;
566 1.3 oster RF_CallbackDesc_t *cb, *p;
567 1.3 oster RF_ReconBuffer_t *t;
568 1.3 oster RF_Raid_t *raidPtr;
569 1.3 oster caddr_t ta;
570 1.3 oster
571 1.3 oster retcode = 0;
572 1.3 oster created = 0;
573 1.3 oster
574 1.3 oster raidPtr = rbuf->raidPtr;
575 1.3 oster reconCtrlPtr = raidPtr->reconControl[rbuf->row];
576 1.3 oster
577 1.3 oster RF_ASSERT(rbuf);
578 1.3 oster RF_ASSERT(rbuf->col != reconCtrlPtr->fcol);
579 1.3 oster
580 1.11 oster #if RF_DEBUG_RECON
581 1.3 oster if (rf_reconbufferDebug) {
582 1.5 oster printf("raid%d: RAID1 reconbuffer submission r%d c%d psid %ld ru%d (failed offset %ld)\n",
583 1.5 oster raidPtr->raidid, rbuf->row, rbuf->col,
584 1.5 oster (long) rbuf->parityStripeID, rbuf->which_ru,
585 1.5 oster (long) rbuf->failedDiskSectorOffset);
586 1.3 oster }
587 1.10 oster #endif
588 1.3 oster if (rf_reconDebug) {
589 1.3 oster printf("RAID1 reconbuffer submit psid %ld buf %lx\n",
590 1.3 oster (long) rbuf->parityStripeID, (long) rbuf->buffer);
591 1.3 oster printf("RAID1 psid %ld %02x %02x %02x %02x %02x\n",
592 1.3 oster (long) rbuf->parityStripeID,
593 1.3 oster rbuf->buffer[0], rbuf->buffer[1], rbuf->buffer[2], rbuf->buffer[3],
594 1.3 oster rbuf->buffer[4]);
595 1.3 oster }
596 1.3 oster RF_LOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
597 1.3 oster
598 1.3 oster RF_LOCK_MUTEX(reconCtrlPtr->rb_mutex);
599 1.3 oster
600 1.3 oster pssPtr = rf_LookupRUStatus(raidPtr, reconCtrlPtr->pssTable,
601 1.3 oster rbuf->parityStripeID, rbuf->which_ru, RF_PSS_NONE, &created);
602 1.3 oster RF_ASSERT(pssPtr); /* if it didn't exist, we wouldn't have gotten
603 1.3 oster * an rbuf for it */
604 1.3 oster
605 1.3 oster /*
606 1.3 oster * Since this is simple mirroring, the first submission for a stripe is also
607 1.3 oster * treated as the last.
608 1.3 oster */
609 1.3 oster
610 1.3 oster t = NULL;
611 1.3 oster if (keep_it) {
612 1.11 oster #if RF_DEBUG_RECON
613 1.3 oster if (rf_reconbufferDebug) {
614 1.5 oster printf("raid%d: RAID1 rbuf submission: keeping rbuf\n",
615 1.5 oster raidPtr->raidid);
616 1.3 oster }
617 1.10 oster #endif
618 1.3 oster t = rbuf;
619 1.3 oster } else {
620 1.3 oster if (use_committed) {
621 1.11 oster #if RF_DEBUG_RECON
622 1.3 oster if (rf_reconbufferDebug) {
623 1.5 oster printf("raid%d: RAID1 rbuf submission: using committed rbuf\n", raidPtr->raidid);
624 1.3 oster }
625 1.10 oster #endif
626 1.3 oster t = reconCtrlPtr->committedRbufs;
627 1.3 oster RF_ASSERT(t);
628 1.3 oster reconCtrlPtr->committedRbufs = t->next;
629 1.3 oster t->next = NULL;
630 1.3 oster } else
631 1.3 oster if (reconCtrlPtr->floatingRbufs) {
632 1.11 oster #if RF_DEBUG_RECON
633 1.3 oster if (rf_reconbufferDebug) {
634 1.5 oster printf("raid%d: RAID1 rbuf submission: using floating rbuf\n", raidPtr->raidid);
635 1.3 oster }
636 1.10 oster #endif
637 1.3 oster t = reconCtrlPtr->floatingRbufs;
638 1.3 oster reconCtrlPtr->floatingRbufs = t->next;
639 1.3 oster t->next = NULL;
640 1.3 oster }
641 1.3 oster }
642 1.3 oster if (t == NULL) {
643 1.11 oster #if RF_DEBUG_RECON
644 1.3 oster if (rf_reconbufferDebug) {
645 1.5 oster printf("raid%d: RAID1 rbuf submission: waiting for rbuf\n", raidPtr->raidid);
646 1.3 oster }
647 1.10 oster #endif
648 1.3 oster RF_ASSERT((keep_it == 0) && (use_committed == 0));
649 1.3 oster raidPtr->procsInBufWait++;
650 1.3 oster if ((raidPtr->procsInBufWait == (raidPtr->numCol - 1))
651 1.3 oster && (raidPtr->numFullReconBuffers == 0)) {
652 1.3 oster /* ruh-ro */
653 1.3 oster RF_ERRORMSG("Buffer wait deadlock\n");
654 1.3 oster rf_PrintPSStatusTable(raidPtr, rbuf->row);
655 1.3 oster RF_PANIC();
656 1.3 oster }
657 1.3 oster pssPtr->flags |= RF_PSS_BUFFERWAIT;
658 1.3 oster cb = rf_AllocCallbackDesc();
659 1.3 oster cb->row = rbuf->row;
660 1.3 oster cb->col = rbuf->col;
661 1.3 oster cb->callbackArg.v = rbuf->parityStripeID;
662 1.3 oster cb->callbackArg2.v = rbuf->which_ru;
663 1.3 oster cb->next = NULL;
664 1.3 oster if (reconCtrlPtr->bufferWaitList == NULL) {
665 1.3 oster /* we are the wait list- lucky us */
666 1.3 oster reconCtrlPtr->bufferWaitList = cb;
667 1.3 oster } else {
668 1.3 oster /* append to wait list */
669 1.3 oster for (p = reconCtrlPtr->bufferWaitList; p->next; p = p->next);
670 1.3 oster p->next = cb;
671 1.3 oster }
672 1.3 oster retcode = 1;
673 1.3 oster goto out;
674 1.3 oster }
675 1.3 oster if (t != rbuf) {
676 1.3 oster t->row = rbuf->row;
677 1.3 oster t->col = reconCtrlPtr->fcol;
678 1.3 oster t->parityStripeID = rbuf->parityStripeID;
679 1.3 oster t->which_ru = rbuf->which_ru;
680 1.3 oster t->failedDiskSectorOffset = rbuf->failedDiskSectorOffset;
681 1.3 oster t->spRow = rbuf->spRow;
682 1.3 oster t->spCol = rbuf->spCol;
683 1.3 oster t->spOffset = rbuf->spOffset;
684 1.3 oster /* Swap buffers. DANCE! */
685 1.3 oster ta = t->buffer;
686 1.3 oster t->buffer = rbuf->buffer;
687 1.3 oster rbuf->buffer = ta;
688 1.3 oster }
689 1.3 oster /*
690 1.3 oster * Use the rbuf we've been given as the target.
691 1.3 oster */
692 1.3 oster RF_ASSERT(pssPtr->rbuf == NULL);
693 1.3 oster pssPtr->rbuf = t;
694 1.3 oster
695 1.3 oster t->count = 1;
696 1.3 oster /*
697 1.3 oster * Below, we use 1 for numDataCol (which is equal to the count in the
698 1.3 oster * previous line), so we'll always be done.
699 1.3 oster */
700 1.3 oster rf_CheckForFullRbuf(raidPtr, reconCtrlPtr, pssPtr, 1);
701 1.1 oster
702 1.1 oster out:
703 1.3 oster RF_UNLOCK_PSS_MUTEX(raidPtr, rbuf->row, rbuf->parityStripeID);
704 1.3 oster RF_UNLOCK_MUTEX(reconCtrlPtr->rb_mutex);
705 1.11 oster #if RF_DEBUG_RECON
706 1.3 oster if (rf_reconbufferDebug) {
707 1.5 oster printf("raid%d: RAID1 rbuf submission: returning %d\n",
708 1.5 oster raidPtr->raidid, retcode);
709 1.3 oster }
710 1.10 oster #endif
711 1.3 oster return (retcode);
712 1.1 oster }
713