xref: /netbsd-src/sys/dev/raidframe/rf_parityscan.c (revision 8a8f936f250a330d54f8a24ed0e92aadf9743a7b)
1 /*	$NetBSD: rf_parityscan.c,v 1.11 2001/10/04 15:58:55 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_parityscan.c -- misc utilities related to parity verification
32  *
33  *****************************************************************************/
34 
35 #include <dev/raidframe/raidframevar.h>
36 
37 #include "rf_raid.h"
38 #include "rf_dag.h"
39 #include "rf_dagfuncs.h"
40 #include "rf_dagutils.h"
41 #include "rf_mcpair.h"
42 #include "rf_general.h"
43 #include "rf_engine.h"
44 #include "rf_parityscan.h"
45 #include "rf_map.h"
46 
47 /*****************************************************************************************
48  *
49  * walk through the entire arry and write new parity.
50  * This works by creating two DAGs, one to read a stripe of data and one to
51  * write new parity.  The first is executed, the data is xored together, and
52  * then the second is executed.  To avoid constantly building and tearing down
53  * the DAGs, we create them a priori and fill them in with the mapping
54  * information as we go along.
55  *
56  * there should never be more than one thread running this.
57  *
58  ****************************************************************************************/
59 
60 int
61 rf_RewriteParity(raidPtr)
62 	RF_Raid_t *raidPtr;
63 {
64 	RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
65 	RF_AccessStripeMapHeader_t *asm_h;
66 	int ret_val;
67 	int rc;
68 	RF_PhysDiskAddr_t pda;
69 	RF_SectorNum_t i;
70 
71 	if (raidPtr->Layout.map->faultsTolerated == 0) {
72 		/* There isn't any parity. Call it "okay." */
73 		return (RF_PARITY_OKAY);
74 	}
75 	if (raidPtr->status[0] != rf_rs_optimal) {
76 		/*
77 		 * We're in degraded mode.  Don't try to verify parity now!
78 		 * XXX: this should be a "we don't want to", not a
79 		 * "we can't" error.
80 		 */
81 		return (RF_PARITY_COULD_NOT_VERIFY);
82 	}
83 
84 	ret_val = 0;
85 
86 	pda.startSector = 0;
87 	pda.numSector = raidPtr->Layout.sectorsPerStripeUnit;
88 	rc = RF_PARITY_OKAY;
89 
90 	for (i = 0; i < raidPtr->totalSectors &&
91 		     rc <= RF_PARITY_CORRECTED;
92 	     i += layoutPtr->dataSectorsPerStripe) {
93 		if (raidPtr->waitShutdown) {
94 			/* Someone is pulling the plug on this set...
95 			   abort the re-write */
96 			return (1);
97 		}
98 		asm_h = rf_MapAccess(raidPtr, i,
99 				     layoutPtr->dataSectorsPerStripe,
100 				     NULL, RF_DONT_REMAP);
101 		raidPtr->parity_rewrite_stripes_done =
102 			i / layoutPtr->dataSectorsPerStripe ;
103 		rc = rf_VerifyParity(raidPtr, asm_h->stripeMap, 1, 0);
104 
105 		switch (rc) {
106 		case RF_PARITY_OKAY:
107 		case RF_PARITY_CORRECTED:
108 			break;
109 		case RF_PARITY_BAD:
110 			printf("Parity bad during correction\n");
111 			ret_val = 1;
112 			break;
113 		case RF_PARITY_COULD_NOT_CORRECT:
114 			printf("Could not correct bad parity\n");
115 			ret_val = 1;
116 			break;
117 		case RF_PARITY_COULD_NOT_VERIFY:
118 			printf("Could not verify parity\n");
119 			ret_val = 1;
120 			break;
121 		default:
122 			printf("Bad rc=%d from VerifyParity in RewriteParity\n", rc);
123 			ret_val = 1;
124 		}
125 		rf_FreeAccessStripeMap(asm_h);
126 	}
127 	return (ret_val);
128 }
129 /*****************************************************************************************
130  *
131  * verify that the parity in a particular stripe is correct.
132  * we validate only the range of parity defined by parityPDA, since
133  * this is all we have locked.  The way we do this is to create an asm
134  * that maps the whole stripe and then range-restrict it to the parity
135  * region defined by the parityPDA.
136  *
137  ****************************************************************************************/
138 int
139 rf_VerifyParity(raidPtr, aasm, correct_it, flags)
140 	RF_Raid_t *raidPtr;
141 	RF_AccessStripeMap_t *aasm;
142 	int     correct_it;
143 	RF_RaidAccessFlags_t flags;
144 {
145 	RF_PhysDiskAddr_t *parityPDA;
146 	RF_AccessStripeMap_t *doasm;
147 	RF_LayoutSW_t *lp;
148 	int     lrc, rc;
149 
150 	lp = raidPtr->Layout.map;
151 	if (lp->faultsTolerated == 0) {
152 		/*
153 	         * There isn't any parity. Call it "okay."
154 	         */
155 		return (RF_PARITY_OKAY);
156 	}
157 	rc = RF_PARITY_OKAY;
158 	if (lp->VerifyParity) {
159 		for (doasm = aasm; doasm; doasm = doasm->next) {
160 			for (parityPDA = doasm->parityInfo; parityPDA;
161 			     parityPDA = parityPDA->next) {
162 				lrc = lp->VerifyParity(raidPtr,
163 						       doasm->raidAddress,
164 						       parityPDA,
165 						       correct_it, flags);
166 				if (lrc > rc) {
167 					/* see rf_parityscan.h for why this
168 					 * works */
169 					rc = lrc;
170 				}
171 			}
172 		}
173 	} else {
174 		rc = RF_PARITY_COULD_NOT_VERIFY;
175 	}
176 	return (rc);
177 }
178 
179 int
180 rf_VerifyParityBasic(raidPtr, raidAddr, parityPDA, correct_it, flags)
181 	RF_Raid_t *raidPtr;
182 	RF_RaidAddr_t raidAddr;
183 	RF_PhysDiskAddr_t *parityPDA;
184 	int     correct_it;
185 	RF_RaidAccessFlags_t flags;
186 {
187 	RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
188 	RF_RaidAddr_t startAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
189 								     raidAddr);
190 	RF_SectorCount_t numsector = parityPDA->numSector;
191 	int     numbytes = rf_RaidAddressToByte(raidPtr, numsector);
192 	int     bytesPerStripe = numbytes * layoutPtr->numDataCol;
193 	RF_DagHeader_t *rd_dag_h, *wr_dag_h;	/* read, write dag */
194 	RF_DagNode_t *blockNode, *unblockNode, *wrBlock, *wrUnblock;
195 	RF_AccessStripeMapHeader_t *asm_h;
196 	RF_AccessStripeMap_t *asmap;
197 	RF_AllocListElem_t *alloclist;
198 	RF_PhysDiskAddr_t *pda;
199 	char   *pbuf, *buf, *end_p, *p;
200 	int     i, retcode;
201 	RF_ReconUnitNum_t which_ru;
202 	RF_StripeNum_t psID = rf_RaidAddressToParityStripeID(layoutPtr,
203 							     raidAddr,
204 							     &which_ru);
205 	int     stripeWidth = layoutPtr->numDataCol + layoutPtr->numParityCol;
206 	RF_AccTraceEntry_t tracerec;
207 	RF_MCPair_t *mcpair;
208 
209 	retcode = RF_PARITY_OKAY;
210 
211 	mcpair = rf_AllocMCPair();
212 	rf_MakeAllocList(alloclist);
213 	RF_MallocAndAdd(buf, numbytes * (layoutPtr->numDataCol + layoutPtr->numParityCol), (char *), alloclist);
214 	RF_CallocAndAdd(pbuf, 1, numbytes, (char *), alloclist);	/* use calloc to make
215 									 * sure buffer is zeroed */
216 	end_p = buf + bytesPerStripe;
217 
218 	rd_dag_h = rf_MakeSimpleDAG(raidPtr, stripeWidth, numbytes, buf, rf_DiskReadFunc, rf_DiskReadUndoFunc,
219 	    "Rod", alloclist, flags, RF_IO_NORMAL_PRIORITY);
220 	blockNode = rd_dag_h->succedents[0];
221 	unblockNode = blockNode->succedents[0]->succedents[0];
222 
223 	/* map the stripe and fill in the PDAs in the dag */
224 	asm_h = rf_MapAccess(raidPtr, startAddr, layoutPtr->dataSectorsPerStripe, buf, RF_DONT_REMAP);
225 	asmap = asm_h->stripeMap;
226 
227 	for (pda = asmap->physInfo, i = 0; i < layoutPtr->numDataCol; i++, pda = pda->next) {
228 		RF_ASSERT(pda);
229 		rf_RangeRestrictPDA(raidPtr, parityPDA, pda, 0, 1);
230 		RF_ASSERT(pda->numSector != 0);
231 		if (rf_TryToRedirectPDA(raidPtr, pda, 0))
232 			goto out;	/* no way to verify parity if disk is
233 					 * dead.  return w/ good status */
234 		blockNode->succedents[i]->params[0].p = pda;
235 		blockNode->succedents[i]->params[2].v = psID;
236 		blockNode->succedents[i]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
237 	}
238 
239 	RF_ASSERT(!asmap->parityInfo->next);
240 	rf_RangeRestrictPDA(raidPtr, parityPDA, asmap->parityInfo, 0, 1);
241 	RF_ASSERT(asmap->parityInfo->numSector != 0);
242 	if (rf_TryToRedirectPDA(raidPtr, asmap->parityInfo, 1))
243 		goto out;
244 	blockNode->succedents[layoutPtr->numDataCol]->params[0].p = asmap->parityInfo;
245 
246 	/* fire off the DAG */
247 	memset((char *) &tracerec, 0, sizeof(tracerec));
248 	rd_dag_h->tracerec = &tracerec;
249 
250 	if (rf_verifyParityDebug) {
251 		printf("Parity verify read dag:\n");
252 		rf_PrintDAGList(rd_dag_h);
253 	}
254 	RF_LOCK_MUTEX(mcpair->mutex);
255 	mcpair->flag = 0;
256 	rf_DispatchDAG(rd_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
257 	    (void *) mcpair);
258 	while (!mcpair->flag)
259 		RF_WAIT_COND(mcpair->cond, mcpair->mutex);
260 	RF_UNLOCK_MUTEX(mcpair->mutex);
261 	if (rd_dag_h->status != rf_enable) {
262 		RF_ERRORMSG("Unable to verify parity:  can't read the stripe\n");
263 		retcode = RF_PARITY_COULD_NOT_VERIFY;
264 		goto out;
265 	}
266 	for (p = buf; p < end_p; p += numbytes) {
267 		rf_bxor(p, pbuf, numbytes, NULL);
268 	}
269 	for (i = 0; i < numbytes; i++) {
270 #if 0
271 		if (pbuf[i] != 0 || buf[bytesPerStripe + i] != 0) {
272 			printf("Bytes: %d %d %d\n", i, pbuf[i], buf[bytesPerStripe + i]);
273 		}
274 #endif
275 		if (pbuf[i] != buf[bytesPerStripe + i]) {
276 			if (!correct_it)
277 				RF_ERRORMSG3("Parity verify error: byte %d of parity is 0x%x should be 0x%x\n",
278 				    i, (u_char) buf[bytesPerStripe + i], (u_char) pbuf[i]);
279 			retcode = RF_PARITY_BAD;
280 			break;
281 		}
282 	}
283 
284 	if (retcode && correct_it) {
285 		wr_dag_h = rf_MakeSimpleDAG(raidPtr, 1, numbytes, pbuf, rf_DiskWriteFunc, rf_DiskWriteUndoFunc,
286 		    "Wnp", alloclist, flags, RF_IO_NORMAL_PRIORITY);
287 		wrBlock = wr_dag_h->succedents[0];
288 		wrUnblock = wrBlock->succedents[0]->succedents[0];
289 		wrBlock->succedents[0]->params[0].p = asmap->parityInfo;
290 		wrBlock->succedents[0]->params[2].v = psID;
291 		wrBlock->succedents[0]->params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
292 		memset((char *) &tracerec, 0, sizeof(tracerec));
293 		wr_dag_h->tracerec = &tracerec;
294 		if (rf_verifyParityDebug) {
295 			printf("Parity verify write dag:\n");
296 			rf_PrintDAGList(wr_dag_h);
297 		}
298 		RF_LOCK_MUTEX(mcpair->mutex);
299 		mcpair->flag = 0;
300 		rf_DispatchDAG(wr_dag_h, (void (*) (void *)) rf_MCPairWakeupFunc,
301 		    (void *) mcpair);
302 		while (!mcpair->flag)
303 			RF_WAIT_COND(mcpair->cond, mcpair->mutex);
304 		RF_UNLOCK_MUTEX(mcpair->mutex);
305 		if (wr_dag_h->status != rf_enable) {
306 			RF_ERRORMSG("Unable to correct parity in VerifyParity:  can't write the stripe\n");
307 			retcode = RF_PARITY_COULD_NOT_CORRECT;
308 		}
309 		rf_FreeDAG(wr_dag_h);
310 		if (retcode == RF_PARITY_BAD)
311 			retcode = RF_PARITY_CORRECTED;
312 	}
313 out:
314 	rf_FreeAccessStripeMap(asm_h);
315 	rf_FreeAllocList(alloclist);
316 	rf_FreeDAG(rd_dag_h);
317 	rf_FreeMCPair(mcpair);
318 	return (retcode);
319 }
320 
321 int
322 rf_TryToRedirectPDA(raidPtr, pda, parity)
323 	RF_Raid_t *raidPtr;
324 	RF_PhysDiskAddr_t *pda;
325 	int     parity;
326 {
327 	if (raidPtr->Disks[pda->row][pda->col].status == rf_ds_reconstructing) {
328 		if (rf_CheckRUReconstructed(raidPtr->reconControl[pda->row]->reconMap, pda->startSector)) {
329 			if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
330 				RF_RowCol_t or = pda->row, oc = pda->col;
331 				RF_SectorNum_t os = pda->startSector;
332 				if (parity) {
333 					(raidPtr->Layout.map->MapParity) (raidPtr, pda->raidAddress, &pda->row, &pda->col, &pda->startSector, RF_REMAP);
334 					if (rf_verifyParityDebug)
335 						printf("VerifyParity: Redir P r %d c %d sect %ld -> r %d c %d sect %ld\n",
336 						    or, oc, (long) os, pda->row, pda->col, (long) pda->startSector);
337 				} else {
338 					(raidPtr->Layout.map->MapSector) (raidPtr, pda->raidAddress, &pda->row, &pda->col, &pda->startSector, RF_REMAP);
339 					if (rf_verifyParityDebug)
340 						printf("VerifyParity: Redir D r %d c %d sect %ld -> r %d c %d sect %ld\n",
341 						    or, oc, (long) os, pda->row, pda->col, (long) pda->startSector);
342 				}
343 			} else {
344 				RF_RowCol_t spRow = raidPtr->Disks[pda->row][pda->col].spareRow;
345 				RF_RowCol_t spCol = raidPtr->Disks[pda->row][pda->col].spareCol;
346 				pda->row = spRow;
347 				pda->col = spCol;
348 			}
349 		}
350 	}
351 	if (RF_DEAD_DISK(raidPtr->Disks[pda->row][pda->col].status))
352 		return (1);
353 	return (0);
354 }
355 /*****************************************************************************************
356  *
357  * currently a stub.
358  *
359  * takes as input an ASM describing a write operation and containing one failure, and
360  * verifies that the parity was correctly updated to reflect the write.
361  *
362  * if it's a data unit that's failed, we read the other data units in the stripe and
363  * the parity unit, XOR them together, and verify that we get the data intended for
364  * the failed disk.  Since it's easy, we also validate that the right data got written
365  * to the surviving data disks.
366  *
367  * If it's the parity that failed, there's really no validation we can do except the
368  * above verification that the right data got written to all disks.  This is because
369  * the new data intended for the failed disk is supplied in the ASM, but this is of
370  * course not the case for the new parity.
371  *
372  ****************************************************************************************/
373 int
374 rf_VerifyDegrModeWrite(raidPtr, asmh)
375 	RF_Raid_t *raidPtr;
376 	RF_AccessStripeMapHeader_t *asmh;
377 {
378 	return (0);
379 }
380 /* creates a simple DAG with a header, a block-recon node at level 1,
381  * nNodes nodes at level 2, an unblock-recon node at level 3, and
382  * a terminator node at level 4.  The stripe address field in
383  * the block and unblock nodes are not touched, nor are the pda
384  * fields in the second-level nodes, so they must be filled in later.
385  *
386  * commit point is established at unblock node - this means that any
387  * failure during dag execution causes the dag to fail
388  */
389 RF_DagHeader_t *
390 rf_MakeSimpleDAG(raidPtr, nNodes, bytesPerSU, databuf, doFunc, undoFunc, name, alloclist, flags, priority)
391 	RF_Raid_t *raidPtr;
392 	int     nNodes;
393 	int     bytesPerSU;
394 	char   *databuf;
395 	int     (*doFunc) (RF_DagNode_t * node);
396 	int     (*undoFunc) (RF_DagNode_t * node);
397 	char   *name;		/* node names at the second level */
398 	RF_AllocListElem_t *alloclist;
399 	RF_RaidAccessFlags_t flags;
400 	int     priority;
401 {
402 	RF_DagHeader_t *dag_h;
403 	RF_DagNode_t *nodes, *termNode, *blockNode, *unblockNode;
404 	int     i;
405 
406 	/* create the nodes, the block & unblock nodes, and the terminator
407 	 * node */
408 	RF_CallocAndAdd(nodes, nNodes + 3, sizeof(RF_DagNode_t), (RF_DagNode_t *), alloclist);
409 	blockNode = &nodes[nNodes];
410 	unblockNode = blockNode + 1;
411 	termNode = unblockNode + 1;
412 
413 	dag_h = rf_AllocDAGHeader();
414 	dag_h->raidPtr = (void *) raidPtr;
415 	dag_h->allocList = NULL;/* we won't use this alloc list */
416 	dag_h->status = rf_enable;
417 	dag_h->numSuccedents = 1;
418 	dag_h->creator = "SimpleDAG";
419 
420 	/* this dag can not commit until the unblock node is reached errors
421 	 * prior to the commit point imply the dag has failed */
422 	dag_h->numCommitNodes = 1;
423 	dag_h->numCommits = 0;
424 
425 	dag_h->succedents[0] = blockNode;
426 	rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nNodes, 0, 0, 0, dag_h, "Nil", alloclist);
427 	rf_InitNode(unblockNode, rf_wait, RF_TRUE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, 1, nNodes, 0, 0, dag_h, "Nil", alloclist);
428 	unblockNode->succedents[0] = termNode;
429 	for (i = 0; i < nNodes; i++) {
430 		blockNode->succedents[i] = unblockNode->antecedents[i] = &nodes[i];
431 		unblockNode->antType[i] = rf_control;
432 		rf_InitNode(&nodes[i], rf_wait, RF_FALSE, doFunc, undoFunc, rf_GenericWakeupFunc, 1, 1, 4, 0, dag_h, name, alloclist);
433 		nodes[i].succedents[0] = unblockNode;
434 		nodes[i].antecedents[0] = blockNode;
435 		nodes[i].antType[0] = rf_control;
436 		nodes[i].params[1].p = (databuf + (i * bytesPerSU));
437 	}
438 	rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, 1, 0, 0, dag_h, "Trm", alloclist);
439 	termNode->antecedents[0] = unblockNode;
440 	termNode->antType[0] = rf_control;
441 	return (dag_h);
442 }
443