rf_map.c revision 1.32 1 1.32 oster /* $NetBSD: rf_map.c,v 1.32 2004/03/07 21:57:44 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: Mark Holland
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 * map.c -- main code for mapping RAID addresses to physical disk addresses
32 1.1 oster *
33 1.1 oster **************************************************************************/
34 1.9 lukem
35 1.9 lukem #include <sys/cdefs.h>
36 1.32 oster __KERNEL_RCSID(0, "$NetBSD: rf_map.c,v 1.32 2004/03/07 21:57:44 oster Exp $");
37 1.1 oster
38 1.7 oster #include <dev/raidframe/raidframevar.h>
39 1.7 oster
40 1.1 oster #include "rf_threadstuff.h"
41 1.1 oster #include "rf_raid.h"
42 1.1 oster #include "rf_general.h"
43 1.1 oster #include "rf_map.h"
44 1.1 oster #include "rf_shutdown.h"
45 1.1 oster
46 1.27 oster static void rf_FreePDAList(RF_PhysDiskAddr_t *pda_list);
47 1.27 oster static void rf_FreeASMList(RF_AccessStripeMap_t *asm_list);
48 1.1 oster
49 1.21 oster /***************************************************************************
50 1.1 oster *
51 1.21 oster * MapAccess -- main 1st order mapping routine. Maps an access in the
52 1.21 oster * RAID address space to the corresponding set of physical disk
53 1.21 oster * addresses. The result is returned as a list of AccessStripeMap
54 1.21 oster * structures, one per stripe accessed. Each ASM structure contains a
55 1.21 oster * pointer to a list of PhysDiskAddr structures, which describe the
56 1.25 oster * physical locations touched by the user access. Note that this
57 1.25 oster * routine returns only static mapping information, i.e. the list of
58 1.25 oster * physical addresses returned does not necessarily identify the set
59 1.25 oster * of physical locations that will actually be read or written. The
60 1.25 oster * routine also maps the parity. The physical disk location returned
61 1.25 oster * always indicates the entire parity unit, even when only a subset of
62 1.25 oster * it is being accessed. This is because an access that is not stripe
63 1.25 oster * unit aligned but that spans a stripe unit boundary may require
64 1.25 oster * access two distinct portions of the parity unit, and we can't yet
65 1.25 oster * tell which portion(s) we'll actually need. We leave it up to the
66 1.25 oster * algorithm selection code to decide what subset of the parity unit
67 1.25 oster * to access. Note that addresses in the RAID address space must
68 1.25 oster * always be maintained as longs, instead of ints.
69 1.21 oster *
70 1.1 oster * This routine returns NULL if numBlocks is 0
71 1.1 oster *
72 1.26 oster * raidAddress - starting address in RAID address space
73 1.26 oster * numBlocks - number of blocks in RAID address space to access
74 1.26 oster * buffer - buffer to supply/recieve data
75 1.26 oster * remap - 1 => remap address to spare space
76 1.21 oster ***************************************************************************/
77 1.1 oster
78 1.3 oster RF_AccessStripeMapHeader_t *
79 1.26 oster rf_MapAccess(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddress,
80 1.26 oster RF_SectorCount_t numBlocks, caddr_t buffer, int remap)
81 1.3 oster {
82 1.3 oster RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
83 1.3 oster RF_AccessStripeMapHeader_t *asm_hdr = NULL;
84 1.3 oster RF_AccessStripeMap_t *asm_list = NULL, *asm_p = NULL;
85 1.3 oster int faultsTolerated = layoutPtr->map->faultsTolerated;
86 1.25 oster /* we'll change raidAddress along the way */
87 1.25 oster RF_RaidAddr_t startAddress = raidAddress;
88 1.3 oster RF_RaidAddr_t endAddress = raidAddress + numBlocks;
89 1.23 oster RF_RaidDisk_t *disks = raidPtr->Disks;
90 1.31 oster RF_PhysDiskAddr_t *pda_p;
91 1.31 oster #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
92 1.31 oster RF_PhysDiskAddr_t *pda_q;
93 1.31 oster #endif
94 1.3 oster RF_StripeCount_t numStripes = 0;
95 1.25 oster RF_RaidAddr_t stripeRealEndAddress, stripeEndAddress,
96 1.25 oster nextStripeUnitAddress;
97 1.3 oster RF_RaidAddr_t startAddrWithinStripe, lastRaidAddr;
98 1.3 oster RF_StripeCount_t totStripes;
99 1.3 oster RF_StripeNum_t stripeID, lastSID, SUID, lastSUID;
100 1.3 oster RF_AccessStripeMap_t *asmList, *t_asm;
101 1.3 oster RF_PhysDiskAddr_t *pdaList, *t_pda;
102 1.3 oster
103 1.3 oster /* allocate all the ASMs and PDAs up front */
104 1.3 oster lastRaidAddr = raidAddress + numBlocks - 1;
105 1.3 oster stripeID = rf_RaidAddressToStripeID(layoutPtr, raidAddress);
106 1.3 oster lastSID = rf_RaidAddressToStripeID(layoutPtr, lastRaidAddr);
107 1.3 oster totStripes = lastSID - stripeID + 1;
108 1.3 oster SUID = rf_RaidAddressToStripeUnitID(layoutPtr, raidAddress);
109 1.3 oster lastSUID = rf_RaidAddressToStripeUnitID(layoutPtr, lastRaidAddr);
110 1.3 oster
111 1.3 oster asmList = rf_AllocASMList(totStripes);
112 1.25 oster
113 1.25 oster /* may also need pda(s) per stripe for parity */
114 1.25 oster pdaList = rf_AllocPDAList(lastSUID - SUID + 1 +
115 1.25 oster faultsTolerated * totStripes);
116 1.25 oster
117 1.3 oster
118 1.3 oster if (raidAddress + numBlocks > raidPtr->totalSectors) {
119 1.3 oster RF_ERRORMSG1("Unable to map access because offset (%d) was invalid\n",
120 1.3 oster (int) raidAddress);
121 1.3 oster return (NULL);
122 1.3 oster }
123 1.15 oster #if RF_DEBUG_MAP
124 1.3 oster if (rf_mapDebug)
125 1.3 oster rf_PrintRaidAddressInfo(raidPtr, raidAddress, numBlocks);
126 1.15 oster #endif
127 1.3 oster for (; raidAddress < endAddress;) {
128 1.3 oster /* make the next stripe structure */
129 1.3 oster RF_ASSERT(asmList);
130 1.3 oster t_asm = asmList;
131 1.3 oster asmList = asmList->next;
132 1.6 thorpej memset((char *) t_asm, 0, sizeof(RF_AccessStripeMap_t));
133 1.3 oster if (!asm_p)
134 1.3 oster asm_list = asm_p = t_asm;
135 1.3 oster else {
136 1.3 oster asm_p->next = t_asm;
137 1.3 oster asm_p = asm_p->next;
138 1.3 oster }
139 1.3 oster numStripes++;
140 1.3 oster
141 1.3 oster /* map SUs from current location to the end of the stripe */
142 1.3 oster asm_p->stripeID = /* rf_RaidAddressToStripeID(layoutPtr,
143 1.3 oster raidAddress) */ stripeID++;
144 1.3 oster stripeRealEndAddress = rf_RaidAddressOfNextStripeBoundary(layoutPtr, raidAddress);
145 1.3 oster stripeEndAddress = RF_MIN(endAddress, stripeRealEndAddress);
146 1.3 oster asm_p->raidAddress = raidAddress;
147 1.3 oster asm_p->endRaidAddress = stripeEndAddress;
148 1.3 oster
149 1.3 oster /* map each stripe unit in the stripe */
150 1.3 oster pda_p = NULL;
151 1.25 oster
152 1.25 oster /* Raid addr of start of portion of access that is
153 1.25 oster within this stripe */
154 1.25 oster startAddrWithinStripe = raidAddress;
155 1.25 oster
156 1.3 oster for (; raidAddress < stripeEndAddress;) {
157 1.3 oster RF_ASSERT(pdaList);
158 1.3 oster t_pda = pdaList;
159 1.3 oster pdaList = pdaList->next;
160 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
161 1.3 oster if (!pda_p)
162 1.3 oster asm_p->physInfo = pda_p = t_pda;
163 1.3 oster else {
164 1.3 oster pda_p->next = t_pda;
165 1.3 oster pda_p = pda_p->next;
166 1.3 oster }
167 1.3 oster
168 1.3 oster pda_p->type = RF_PDA_TYPE_DATA;
169 1.25 oster (layoutPtr->map->MapSector) (raidPtr, raidAddress,
170 1.25 oster &(pda_p->col),
171 1.25 oster &(pda_p->startSector),
172 1.25 oster remap);
173 1.25 oster
174 1.25 oster /* mark any failures we find. failedPDA is
175 1.25 oster * don't-care if there is more than one
176 1.25 oster * failure */
177 1.25 oster
178 1.25 oster /* the RAID address corresponding to this
179 1.25 oster physical diskaddress */
180 1.25 oster pda_p->raidAddress = raidAddress;
181 1.3 oster nextStripeUnitAddress = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, raidAddress);
182 1.3 oster pda_p->numSector = RF_MIN(endAddress, nextStripeUnitAddress) - raidAddress;
183 1.3 oster RF_ASSERT(pda_p->numSector != 0);
184 1.3 oster rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 0);
185 1.3 oster pda_p->bufPtr = buffer + rf_RaidAddressToByte(raidPtr, (raidAddress - startAddress));
186 1.3 oster asm_p->totalSectorsAccessed += pda_p->numSector;
187 1.3 oster asm_p->numStripeUnitsAccessed++;
188 1.3 oster
189 1.3 oster raidAddress = RF_MIN(endAddress, nextStripeUnitAddress);
190 1.3 oster }
191 1.3 oster
192 1.3 oster /* Map the parity. At this stage, the startSector and
193 1.25 oster * numSector fields for the parity unit are always set
194 1.25 oster * to indicate the entire parity unit. We may modify
195 1.25 oster * this after mapping the data portion. */
196 1.3 oster switch (faultsTolerated) {
197 1.3 oster case 0:
198 1.3 oster break;
199 1.3 oster case 1: /* single fault tolerant */
200 1.3 oster RF_ASSERT(pdaList);
201 1.3 oster t_pda = pdaList;
202 1.3 oster pdaList = pdaList->next;
203 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
204 1.3 oster pda_p = asm_p->parityInfo = t_pda;
205 1.3 oster pda_p->type = RF_PDA_TYPE_PARITY;
206 1.3 oster (layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
207 1.23 oster &(pda_p->col), &(pda_p->startSector), remap);
208 1.3 oster pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
209 1.3 oster /* raidAddr may be needed to find unit to redirect to */
210 1.3 oster pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
211 1.3 oster rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
212 1.3 oster rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
213 1.3 oster
214 1.3 oster break;
215 1.31 oster #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0)
216 1.3 oster case 2: /* two fault tolerant */
217 1.3 oster RF_ASSERT(pdaList && pdaList->next);
218 1.3 oster t_pda = pdaList;
219 1.3 oster pdaList = pdaList->next;
220 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
221 1.3 oster pda_p = asm_p->parityInfo = t_pda;
222 1.3 oster pda_p->type = RF_PDA_TYPE_PARITY;
223 1.3 oster t_pda = pdaList;
224 1.3 oster pdaList = pdaList->next;
225 1.6 thorpej memset((char *) t_pda, 0, sizeof(RF_PhysDiskAddr_t));
226 1.3 oster pda_q = asm_p->qInfo = t_pda;
227 1.3 oster pda_q->type = RF_PDA_TYPE_Q;
228 1.3 oster (layoutPtr->map->MapParity) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
229 1.23 oster &(pda_p->col), &(pda_p->startSector), remap);
230 1.3 oster (layoutPtr->map->MapQ) (raidPtr, rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe),
231 1.23 oster &(pda_q->col), &(pda_q->startSector), remap);
232 1.3 oster pda_q->numSector = pda_p->numSector = layoutPtr->sectorsPerStripeUnit;
233 1.3 oster /* raidAddr may be needed to find unit to redirect to */
234 1.3 oster pda_p->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
235 1.3 oster pda_q->raidAddress = rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, startAddrWithinStripe);
236 1.3 oster /* failure mode stuff */
237 1.3 oster rf_ASMCheckStatus(raidPtr, pda_p, asm_p, disks, 1);
238 1.3 oster rf_ASMCheckStatus(raidPtr, pda_q, asm_p, disks, 1);
239 1.3 oster rf_ASMParityAdjust(asm_p->parityInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
240 1.3 oster rf_ASMParityAdjust(asm_p->qInfo, startAddrWithinStripe, endAddress, layoutPtr, asm_p);
241 1.3 oster break;
242 1.31 oster #endif
243 1.3 oster }
244 1.3 oster }
245 1.3 oster RF_ASSERT(asmList == NULL && pdaList == NULL);
246 1.3 oster /* make the header structure */
247 1.3 oster asm_hdr = rf_AllocAccessStripeMapHeader();
248 1.3 oster RF_ASSERT(numStripes == totStripes);
249 1.3 oster asm_hdr->numStripes = numStripes;
250 1.3 oster asm_hdr->stripeMap = asm_list;
251 1.3 oster
252 1.15 oster #if RF_DEBUG_MAP
253 1.3 oster if (rf_mapDebug)
254 1.3 oster rf_PrintAccessStripeMap(asm_hdr);
255 1.15 oster #endif
256 1.3 oster return (asm_hdr);
257 1.1 oster }
258 1.21 oster
259 1.21 oster /***************************************************************************
260 1.21 oster * This routine walks through an ASM list and marks the PDAs that have
261 1.21 oster * failed. It's called only when a disk failure causes an in-flight
262 1.21 oster * DAG to fail. The parity may consist of two components, but we want
263 1.21 oster * to use only one failedPDA pointer. Thus we set failedPDA to point
264 1.21 oster * to the first parity component, and rely on the rest of the code to
265 1.21 oster * do the right thing with this.
266 1.21 oster ***************************************************************************/
267 1.1 oster
268 1.3 oster void
269 1.26 oster rf_MarkFailuresInASMList(RF_Raid_t *raidPtr,
270 1.26 oster RF_AccessStripeMapHeader_t *asm_h)
271 1.3 oster {
272 1.23 oster RF_RaidDisk_t *disks = raidPtr->Disks;
273 1.3 oster RF_AccessStripeMap_t *asmap;
274 1.3 oster RF_PhysDiskAddr_t *pda;
275 1.3 oster
276 1.3 oster for (asmap = asm_h->stripeMap; asmap; asmap = asmap->next) {
277 1.25 oster asmap->numDataFailed = 0;
278 1.25 oster asmap->numParityFailed = 0;
279 1.25 oster asmap->numQFailed = 0;
280 1.3 oster asmap->numFailedPDAs = 0;
281 1.6 thorpej memset((char *) asmap->failedPDAs, 0,
282 1.3 oster RF_MAX_FAILED_PDA * sizeof(RF_PhysDiskAddr_t *));
283 1.3 oster for (pda = asmap->physInfo; pda; pda = pda->next) {
284 1.23 oster if (RF_DEAD_DISK(disks[pda->col].status)) {
285 1.3 oster asmap->numDataFailed++;
286 1.3 oster asmap->failedPDAs[asmap->numFailedPDAs] = pda;
287 1.3 oster asmap->numFailedPDAs++;
288 1.3 oster }
289 1.3 oster }
290 1.3 oster pda = asmap->parityInfo;
291 1.23 oster if (pda && RF_DEAD_DISK(disks[pda->col].status)) {
292 1.3 oster asmap->numParityFailed++;
293 1.3 oster asmap->failedPDAs[asmap->numFailedPDAs] = pda;
294 1.3 oster asmap->numFailedPDAs++;
295 1.3 oster }
296 1.3 oster pda = asmap->qInfo;
297 1.23 oster if (pda && RF_DEAD_DISK(disks[pda->col].status)) {
298 1.3 oster asmap->numQFailed++;
299 1.3 oster asmap->failedPDAs[asmap->numFailedPDAs] = pda;
300 1.3 oster asmap->numFailedPDAs++;
301 1.3 oster }
302 1.3 oster }
303 1.1 oster }
304 1.3 oster
305 1.21 oster /***************************************************************************
306 1.1 oster *
307 1.21 oster * routines to allocate and free list elements. All allocation
308 1.21 oster * routines zero the structure before returning it.
309 1.1 oster *
310 1.21 oster * FreePhysDiskAddr is static. It should never be called directly,
311 1.21 oster * because FreeAccessStripeMap takes care of freeing the PhysDiskAddr
312 1.21 oster * list.
313 1.1 oster *
314 1.21 oster ***************************************************************************/
315 1.1 oster
316 1.1 oster #define RF_MAX_FREE_ASMHDR 128
317 1.30 oster #define RF_MIN_FREE_ASMHDR 32
318 1.1 oster
319 1.1 oster #define RF_MAX_FREE_ASM 192
320 1.30 oster #define RF_MIN_FREE_ASM 64
321 1.1 oster
322 1.1 oster #define RF_MAX_FREE_PDA 192
323 1.30 oster #define RF_MIN_FREE_PDA 64
324 1.1 oster
325 1.25 oster /* called at shutdown time. So far, all that is necessary is to
326 1.25 oster release all the free lists */
327 1.1 oster static void rf_ShutdownMapModule(void *);
328 1.3 oster static void
329 1.26 oster rf_ShutdownMapModule(void *ignored)
330 1.1 oster {
331 1.32 oster pool_destroy(&rf_pools.asm_hdr);
332 1.32 oster pool_destroy(&rf_pools.asmap);
333 1.32 oster pool_destroy(&rf_pools.pda);
334 1.1 oster }
335 1.1 oster
336 1.3 oster int
337 1.26 oster rf_ConfigureMapModule(RF_ShutdownList_t **listp)
338 1.1 oster {
339 1.1 oster
340 1.32 oster rf_pool_init(&rf_pools.asm_hdr, sizeof(RF_AccessStripeMapHeader_t),
341 1.32 oster "rf_asmhdr_pl", RF_MIN_FREE_ASMHDR, RF_MAX_FREE_ASMHDR);
342 1.32 oster rf_pool_init(&rf_pools.asmap, sizeof(RF_AccessStripeMap_t),
343 1.32 oster "rf_asm_pl", RF_MIN_FREE_ASM, RF_MAX_FREE_ASM);
344 1.32 oster rf_pool_init(&rf_pools.pda, sizeof(RF_PhysDiskAddr_t),
345 1.32 oster "rf_pda_pl", RF_MIN_FREE_PDA, RF_MAX_FREE_PDA);
346 1.29 oster rf_ShutdownCreate(listp, rf_ShutdownMapModule, NULL);
347 1.29 oster
348 1.3 oster return (0);
349 1.1 oster }
350 1.1 oster
351 1.3 oster RF_AccessStripeMapHeader_t *
352 1.3 oster rf_AllocAccessStripeMapHeader()
353 1.1 oster {
354 1.1 oster RF_AccessStripeMapHeader_t *p;
355 1.1 oster
356 1.32 oster p = pool_get(&rf_pools.asm_hdr, PR_WAITOK);
357 1.6 thorpej memset((char *) p, 0, sizeof(RF_AccessStripeMapHeader_t));
358 1.1 oster
359 1.3 oster return (p);
360 1.1 oster }
361 1.1 oster
362 1.3 oster void
363 1.26 oster rf_FreeAccessStripeMapHeader(RF_AccessStripeMapHeader_t *p)
364 1.1 oster {
365 1.32 oster pool_put(&rf_pools.asm_hdr, p);
366 1.1 oster }
367 1.1 oster
368 1.3 oster RF_PhysDiskAddr_t *
369 1.3 oster rf_AllocPhysDiskAddr()
370 1.1 oster {
371 1.1 oster RF_PhysDiskAddr_t *p;
372 1.1 oster
373 1.32 oster p = pool_get(&rf_pools.pda, PR_WAITOK);
374 1.6 thorpej memset((char *) p, 0, sizeof(RF_PhysDiskAddr_t));
375 1.1 oster
376 1.3 oster return (p);
377 1.1 oster }
378 1.25 oster /* allocates a list of PDAs, locking the free list only once when we
379 1.25 oster * have to call calloc, we do it one component at a time to simplify
380 1.25 oster * the process of freeing the list at program shutdown. This should
381 1.25 oster * not be much of a performance hit, because it should be very
382 1.25 oster * infrequently executed. */
383 1.3 oster RF_PhysDiskAddr_t *
384 1.26 oster rf_AllocPDAList(int count)
385 1.1 oster {
386 1.24 oster RF_PhysDiskAddr_t *p, *prev;
387 1.24 oster int i;
388 1.24 oster
389 1.24 oster p = NULL;
390 1.24 oster prev = NULL;
391 1.24 oster for (i = 0; i < count; i++) {
392 1.32 oster p = pool_get(&rf_pools.pda, PR_WAITOK);
393 1.24 oster p->next = prev;
394 1.24 oster prev = p;
395 1.24 oster }
396 1.1 oster
397 1.3 oster return (p);
398 1.1 oster }
399 1.1 oster
400 1.18 oster #if RF_INCLUDE_PARITYLOGGING > 0
401 1.3 oster void
402 1.26 oster rf_FreePhysDiskAddr(RF_PhysDiskAddr_t *p)
403 1.1 oster {
404 1.32 oster pool_put(&rf_pools.pda, p);
405 1.1 oster }
406 1.18 oster #endif
407 1.1 oster
408 1.3 oster static void
409 1.27 oster rf_FreePDAList(RF_PhysDiskAddr_t *pda_list)
410 1.1 oster {
411 1.24 oster RF_PhysDiskAddr_t *p, *tmp;
412 1.24 oster
413 1.27 oster p=pda_list;
414 1.24 oster while (p) {
415 1.24 oster tmp = p->next;
416 1.32 oster pool_put(&rf_pools.pda, p);
417 1.24 oster p = tmp;
418 1.24 oster }
419 1.1 oster }
420 1.1 oster
421 1.25 oster /* this is essentially identical to AllocPDAList. I should combine
422 1.25 oster * the two. when we have to call calloc, we do it one component at a
423 1.25 oster * time to simplify the process of freeing the list at program
424 1.25 oster * shutdown. This should not be much of a performance hit, because it
425 1.25 oster * should be very infrequently executed. */
426 1.3 oster RF_AccessStripeMap_t *
427 1.26 oster rf_AllocASMList(int count)
428 1.1 oster {
429 1.24 oster RF_AccessStripeMap_t *p, *prev;
430 1.24 oster int i;
431 1.1 oster
432 1.24 oster p = NULL;
433 1.24 oster prev = NULL;
434 1.24 oster for (i = 0; i < count; i++) {
435 1.32 oster p = pool_get(&rf_pools.asmap, PR_WAITOK);
436 1.24 oster p->next = prev;
437 1.24 oster prev = p;
438 1.24 oster }
439 1.3 oster return (p);
440 1.1 oster }
441 1.1 oster
442 1.3 oster static void
443 1.27 oster rf_FreeASMList(RF_AccessStripeMap_t *asm_list)
444 1.3 oster {
445 1.24 oster RF_AccessStripeMap_t *p, *tmp;
446 1.24 oster
447 1.27 oster p=asm_list;
448 1.24 oster while (p) {
449 1.24 oster tmp = p->next;
450 1.32 oster pool_put(&rf_pools.asmap, p);
451 1.24 oster p = tmp;
452 1.24 oster }
453 1.3 oster }
454 1.3 oster
455 1.3 oster void
456 1.26 oster rf_FreeAccessStripeMap(RF_AccessStripeMapHeader_t *hdr)
457 1.3 oster {
458 1.28 oster RF_AccessStripeMap_t *p;
459 1.3 oster RF_PhysDiskAddr_t *pdp, *trailer, *pdaList = NULL, *pdaEnd = NULL;
460 1.3 oster int count = 0, t, asm_count = 0;
461 1.3 oster
462 1.3 oster for (p = hdr->stripeMap; p; p = p->next) {
463 1.3 oster
464 1.3 oster /* link the 3 pda lists into the accumulating pda list */
465 1.3 oster
466 1.3 oster if (!pdaList)
467 1.3 oster pdaList = p->qInfo;
468 1.3 oster else
469 1.3 oster pdaEnd->next = p->qInfo;
470 1.3 oster for (trailer = NULL, pdp = p->qInfo; pdp;) {
471 1.3 oster trailer = pdp;
472 1.3 oster pdp = pdp->next;
473 1.3 oster count++;
474 1.3 oster }
475 1.3 oster if (trailer)
476 1.3 oster pdaEnd = trailer;
477 1.3 oster
478 1.3 oster if (!pdaList)
479 1.3 oster pdaList = p->parityInfo;
480 1.3 oster else
481 1.3 oster pdaEnd->next = p->parityInfo;
482 1.3 oster for (trailer = NULL, pdp = p->parityInfo; pdp;) {
483 1.3 oster trailer = pdp;
484 1.3 oster pdp = pdp->next;
485 1.3 oster count++;
486 1.3 oster }
487 1.3 oster if (trailer)
488 1.3 oster pdaEnd = trailer;
489 1.3 oster
490 1.3 oster if (!pdaList)
491 1.3 oster pdaList = p->physInfo;
492 1.3 oster else
493 1.3 oster pdaEnd->next = p->physInfo;
494 1.3 oster for (trailer = NULL, pdp = p->physInfo; pdp;) {
495 1.3 oster trailer = pdp;
496 1.3 oster pdp = pdp->next;
497 1.3 oster count++;
498 1.3 oster }
499 1.3 oster if (trailer)
500 1.3 oster pdaEnd = trailer;
501 1.3 oster
502 1.3 oster asm_count++;
503 1.3 oster }
504 1.3 oster
505 1.3 oster /* debug only */
506 1.3 oster for (t = 0, pdp = pdaList; pdp; pdp = pdp->next)
507 1.3 oster t++;
508 1.3 oster RF_ASSERT(t == count);
509 1.3 oster
510 1.3 oster if (pdaList)
511 1.27 oster rf_FreePDAList(pdaList);
512 1.27 oster rf_FreeASMList(hdr->stripeMap);
513 1.3 oster rf_FreeAccessStripeMapHeader(hdr);
514 1.1 oster }
515 1.21 oster /* We can't use the large write optimization if there are any failures
516 1.21 oster * in the stripe. In the declustered layout, there is no way to
517 1.21 oster * immediately determine what disks constitute a stripe, so we
518 1.21 oster * actually have to hunt through the stripe looking for failures. The
519 1.21 oster * reason we map the parity instead of just using asm->parityInfo->col
520 1.21 oster * is because the latter may have been already redirected to a spare
521 1.21 oster * drive, which would mess up the computation of the stripe offset.
522 1.1 oster *
523 1.21 oster * ASSUMES AT MOST ONE FAILURE IN THE STRIPE. */
524 1.3 oster int
525 1.26 oster rf_CheckStripeForFailures(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
526 1.3 oster {
527 1.23 oster RF_RowCol_t tcol, pcol, *diskids, i;
528 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
529 1.3 oster RF_StripeCount_t stripeOffset;
530 1.3 oster int numFailures;
531 1.3 oster RF_RaidAddr_t sosAddr;
532 1.3 oster RF_SectorNum_t diskOffset, poffset;
533 1.3 oster
534 1.3 oster /* quick out in the fault-free case. */
535 1.3 oster RF_LOCK_MUTEX(raidPtr->mutex);
536 1.3 oster numFailures = raidPtr->numFailures;
537 1.3 oster RF_UNLOCK_MUTEX(raidPtr->mutex);
538 1.3 oster if (numFailures == 0)
539 1.3 oster return (0);
540 1.3 oster
541 1.25 oster sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
542 1.25 oster asmap->raidAddress);
543 1.25 oster (layoutPtr->map->IdentifyStripe) (raidPtr, asmap->raidAddress,
544 1.25 oster &diskids);
545 1.25 oster (layoutPtr->map->MapParity) (raidPtr, asmap->raidAddress,
546 1.25 oster &pcol, &poffset, 0); /* get pcol */
547 1.3 oster
548 1.25 oster /* this need not be true if we've redirected the access to a
549 1.25 oster * spare in another row RF_ASSERT(row == testrow); */
550 1.3 oster stripeOffset = 0;
551 1.3 oster for (i = 0; i < layoutPtr->numDataCol + layoutPtr->numParityCol; i++) {
552 1.3 oster if (diskids[i] != pcol) {
553 1.23 oster if (RF_DEAD_DISK(raidPtr->Disks[diskids[i]].status)) {
554 1.23 oster if (raidPtr->status != rf_rs_reconstructing)
555 1.3 oster return (1);
556 1.23 oster RF_ASSERT(raidPtr->reconControl->fcol == diskids[i]);
557 1.3 oster layoutPtr->map->MapSector(raidPtr,
558 1.3 oster sosAddr + stripeOffset * layoutPtr->sectorsPerStripeUnit,
559 1.23 oster &tcol, &diskOffset, 0);
560 1.23 oster RF_ASSERT(tcol == diskids[i]);
561 1.23 oster if (!rf_CheckRUReconstructed(raidPtr->reconControl->reconMap, diskOffset))
562 1.3 oster return (1);
563 1.3 oster asmap->flags |= RF_ASM_REDIR_LARGE_WRITE;
564 1.3 oster return (0);
565 1.3 oster }
566 1.3 oster stripeOffset++;
567 1.3 oster }
568 1.3 oster }
569 1.3 oster return (0);
570 1.1 oster }
571 1.18 oster #if (RF_INCLUDE_DECL_PQ > 0) || (RF_INCLUDE_RAID6 > 0) || (RF_INCLUDE_EVENODD >0)
572 1.1 oster /*
573 1.1 oster return the number of failed data units in the stripe.
574 1.1 oster */
575 1.1 oster
576 1.3 oster int
577 1.26 oster rf_NumFailedDataUnitsInStripe(RF_Raid_t *raidPtr, RF_AccessStripeMap_t *asmap)
578 1.3 oster {
579 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
580 1.23 oster RF_RowCol_t tcol, i;
581 1.3 oster RF_SectorNum_t diskOffset;
582 1.3 oster RF_RaidAddr_t sosAddr;
583 1.3 oster int numFailures;
584 1.3 oster
585 1.3 oster /* quick out in the fault-free case. */
586 1.3 oster RF_LOCK_MUTEX(raidPtr->mutex);
587 1.3 oster numFailures = raidPtr->numFailures;
588 1.3 oster RF_UNLOCK_MUTEX(raidPtr->mutex);
589 1.3 oster if (numFailures == 0)
590 1.3 oster return (0);
591 1.3 oster numFailures = 0;
592 1.3 oster
593 1.25 oster sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr,
594 1.25 oster asmap->raidAddress);
595 1.3 oster for (i = 0; i < layoutPtr->numDataCol; i++) {
596 1.3 oster (layoutPtr->map->MapSector) (raidPtr, sosAddr + i * layoutPtr->sectorsPerStripeUnit,
597 1.3 oster &trow, &tcol, &diskOffset, 0);
598 1.23 oster if (RF_DEAD_DISK(raidPtr->Disks[tcol].status))
599 1.3 oster numFailures++;
600 1.3 oster }
601 1.1 oster
602 1.3 oster return numFailures;
603 1.1 oster }
604 1.18 oster #endif
605 1.1 oster
606 1.25 oster /****************************************************************************
607 1.1 oster *
608 1.1 oster * debug routines
609 1.1 oster *
610 1.25 oster ***************************************************************************/
611 1.18 oster #if RF_DEBUG_MAP
612 1.3 oster void
613 1.26 oster rf_PrintAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h)
614 1.1 oster {
615 1.3 oster rf_PrintFullAccessStripeMap(asm_h, 0);
616 1.1 oster }
617 1.18 oster #endif
618 1.1 oster
619 1.26 oster /* prbuf - flag to print buffer pointers */
620 1.3 oster void
621 1.26 oster rf_PrintFullAccessStripeMap(RF_AccessStripeMapHeader_t *asm_h, int prbuf)
622 1.3 oster {
623 1.3 oster int i;
624 1.3 oster RF_AccessStripeMap_t *asmap = asm_h->stripeMap;
625 1.3 oster RF_PhysDiskAddr_t *p;
626 1.3 oster printf("%d stripes total\n", (int) asm_h->numStripes);
627 1.3 oster for (; asmap; asmap = asmap->next) {
628 1.3 oster /* printf("Num failures: %d\n",asmap->numDataFailed); */
629 1.3 oster /* printf("Num sectors:
630 1.3 oster * %d\n",(int)asmap->totalSectorsAccessed); */
631 1.3 oster printf("Stripe %d (%d sectors), failures: %d data, %d parity: ",
632 1.3 oster (int) asmap->stripeID,
633 1.3 oster (int) asmap->totalSectorsAccessed,
634 1.3 oster (int) asmap->numDataFailed,
635 1.3 oster (int) asmap->numParityFailed);
636 1.3 oster if (asmap->parityInfo) {
637 1.23 oster printf("Parity [c%d s%d-%d", asmap->parityInfo->col,
638 1.3 oster (int) asmap->parityInfo->startSector,
639 1.3 oster (int) (asmap->parityInfo->startSector +
640 1.3 oster asmap->parityInfo->numSector - 1));
641 1.3 oster if (prbuf)
642 1.3 oster printf(" b0x%lx", (unsigned long) asmap->parityInfo->bufPtr);
643 1.3 oster if (asmap->parityInfo->next) {
644 1.23 oster printf(", c%d s%d-%d", asmap->parityInfo->next->col,
645 1.3 oster (int) asmap->parityInfo->next->startSector,
646 1.3 oster (int) (asmap->parityInfo->next->startSector +
647 1.3 oster asmap->parityInfo->next->numSector - 1));
648 1.3 oster if (prbuf)
649 1.3 oster printf(" b0x%lx", (unsigned long) asmap->parityInfo->next->bufPtr);
650 1.3 oster RF_ASSERT(asmap->parityInfo->next->next == NULL);
651 1.3 oster }
652 1.3 oster printf("]\n\t");
653 1.3 oster }
654 1.3 oster for (i = 0, p = asmap->physInfo; p; p = p->next, i++) {
655 1.23 oster printf("SU c%d s%d-%d ", p->col, (int) p->startSector,
656 1.3 oster (int) (p->startSector + p->numSector - 1));
657 1.3 oster if (prbuf)
658 1.3 oster printf("b0x%lx ", (unsigned long) p->bufPtr);
659 1.3 oster if (i && !(i & 1))
660 1.3 oster printf("\n\t");
661 1.3 oster }
662 1.3 oster printf("\n");
663 1.3 oster p = asm_h->stripeMap->failedPDAs[0];
664 1.3 oster if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 1)
665 1.3 oster printf("[multiple failures]\n");
666 1.3 oster else
667 1.3 oster if (asm_h->stripeMap->numDataFailed + asm_h->stripeMap->numParityFailed > 0)
668 1.23 oster printf("\t[Failed PDA: c%d s%d-%d]\n", p->col,
669 1.3 oster (int) p->startSector, (int) (p->startSector + p->numSector - 1));
670 1.3 oster }
671 1.1 oster }
672 1.1 oster
673 1.15 oster #if RF_MAP_DEBUG
674 1.3 oster void
675 1.26 oster rf_PrintRaidAddressInfo(RF_Raid_t *raidPtr, RF_RaidAddr_t raidAddr,
676 1.26 oster RF_SectorCount_t numBlocks)
677 1.3 oster {
678 1.3 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
679 1.3 oster RF_RaidAddr_t ra, sosAddr = rf_RaidAddressOfPrevStripeBoundary(layoutPtr, raidAddr);
680 1.3 oster
681 1.3 oster printf("Raid addrs of SU boundaries from start of stripe to end of access:\n\t");
682 1.3 oster for (ra = sosAddr; ra <= raidAddr + numBlocks; ra += layoutPtr->sectorsPerStripeUnit) {
683 1.3 oster printf("%d (0x%x), ", (int) ra, (int) ra);
684 1.3 oster }
685 1.3 oster printf("\n");
686 1.3 oster printf("Offset into stripe unit: %d (0x%x)\n",
687 1.3 oster (int) (raidAddr % layoutPtr->sectorsPerStripeUnit),
688 1.3 oster (int) (raidAddr % layoutPtr->sectorsPerStripeUnit));
689 1.3 oster }
690 1.15 oster #endif
691 1.25 oster /* given a parity descriptor and the starting address within a stripe,
692 1.25 oster * range restrict the parity descriptor to touch only the correct
693 1.25 oster * stuff. */
694 1.3 oster void
695 1.26 oster rf_ASMParityAdjust(RF_PhysDiskAddr_t *toAdjust,
696 1.26 oster RF_StripeNum_t startAddrWithinStripe,
697 1.26 oster RF_SectorNum_t endAddress,
698 1.26 oster RF_RaidLayout_t *layoutPtr,
699 1.26 oster RF_AccessStripeMap_t *asm_p)
700 1.3 oster {
701 1.3 oster RF_PhysDiskAddr_t *new_pda;
702 1.3 oster
703 1.25 oster /* when we're accessing only a portion of one stripe unit, we
704 1.25 oster * want the parity descriptor to identify only the chunk of
705 1.25 oster * parity associated with the data. When the access spans
706 1.25 oster * exactly one stripe unit boundary and is less than a stripe
707 1.25 oster * unit in size, it uses two disjoint regions of the parity
708 1.25 oster * unit. When an access spans more than one stripe unit
709 1.25 oster * boundary, it uses all of the parity unit.
710 1.3 oster *
711 1.25 oster * To better handle the case where stripe units are small, we
712 1.25 oster * may eventually want to change the 2nd case so that if the
713 1.25 oster * SU size is below some threshold, we just read/write the
714 1.25 oster * whole thing instead of breaking it up into two accesses. */
715 1.3 oster if (asm_p->numStripeUnitsAccessed == 1) {
716 1.3 oster int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
717 1.3 oster toAdjust->startSector += x;
718 1.3 oster toAdjust->raidAddress += x;
719 1.3 oster toAdjust->numSector = asm_p->physInfo->numSector;
720 1.3 oster RF_ASSERT(toAdjust->numSector != 0);
721 1.3 oster } else
722 1.3 oster if (asm_p->numStripeUnitsAccessed == 2 && asm_p->totalSectorsAccessed < layoutPtr->sectorsPerStripeUnit) {
723 1.3 oster int x = (startAddrWithinStripe % layoutPtr->sectorsPerStripeUnit);
724 1.3 oster
725 1.3 oster /* create a second pda and copy the parity map info
726 1.3 oster * into it */
727 1.3 oster RF_ASSERT(toAdjust->next == NULL);
728 1.3 oster new_pda = toAdjust->next = rf_AllocPhysDiskAddr();
729 1.3 oster *new_pda = *toAdjust; /* structure assignment */
730 1.3 oster new_pda->next = NULL;
731 1.3 oster
732 1.3 oster /* adjust the start sector & number of blocks for the
733 1.3 oster * first parity pda */
734 1.3 oster toAdjust->startSector += x;
735 1.3 oster toAdjust->raidAddress += x;
736 1.3 oster toAdjust->numSector = rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr, startAddrWithinStripe) - startAddrWithinStripe;
737 1.3 oster RF_ASSERT(toAdjust->numSector != 0);
738 1.3 oster
739 1.3 oster /* adjust the second pda */
740 1.3 oster new_pda->numSector = endAddress - rf_RaidAddressOfPrevStripeUnitBoundary(layoutPtr, endAddress);
741 1.3 oster /* new_pda->raidAddress =
742 1.3 oster * rf_RaidAddressOfNextStripeUnitBoundary(layoutPtr,
743 1.3 oster * toAdjust->raidAddress); */
744 1.3 oster RF_ASSERT(new_pda->numSector != 0);
745 1.3 oster }
746 1.1 oster }
747 1.11 oster
748 1.25 oster /* Check if a disk has been spared or failed. If spared, redirect the
749 1.25 oster * I/O. If it has been failed, record it in the asm pointer. Fourth
750 1.25 oster * arg is whether data or parity. */
751 1.3 oster void
752 1.26 oster rf_ASMCheckStatus(RF_Raid_t *raidPtr, RF_PhysDiskAddr_t *pda_p,
753 1.26 oster RF_AccessStripeMap_t *asm_p, RF_RaidDisk_t *disks,
754 1.26 oster int parity)
755 1.3 oster {
756 1.3 oster RF_DiskStatus_t dstatus;
757 1.23 oster RF_RowCol_t fcol;
758 1.3 oster
759 1.23 oster dstatus = disks[pda_p->col].status;
760 1.3 oster
761 1.3 oster if (dstatus == rf_ds_spared) {
762 1.3 oster /* if the disk has been spared, redirect access to the spare */
763 1.3 oster fcol = pda_p->col;
764 1.23 oster pda_p->col = disks[fcol].spareCol;
765 1.3 oster } else
766 1.3 oster if (dstatus == rf_ds_dist_spared) {
767 1.3 oster /* ditto if disk has been spared to dist spare space */
768 1.15 oster #if RF_DEBUG_MAP
769 1.23 oster RF_RowCol_t oc = pda_p->col;
770 1.3 oster RF_SectorNum_t oo = pda_p->startSector;
771 1.15 oster #endif
772 1.3 oster if (pda_p->type == RF_PDA_TYPE_DATA)
773 1.23 oster raidPtr->Layout.map->MapSector(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
774 1.3 oster else
775 1.23 oster raidPtr->Layout.map->MapParity(raidPtr, pda_p->raidAddress, &pda_p->col, &pda_p->startSector, RF_REMAP);
776 1.3 oster
777 1.15 oster #if RF_DEBUG_MAP
778 1.3 oster if (rf_mapDebug) {
779 1.23 oster printf("Redirected c %d o %d -> c %d o %d\n", oc, (int) oo,
780 1.23 oster pda_p->col, (int) pda_p->startSector);
781 1.3 oster }
782 1.15 oster #endif
783 1.3 oster } else
784 1.3 oster if (RF_DEAD_DISK(dstatus)) {
785 1.3 oster /* if the disk is inaccessible, mark the
786 1.3 oster * failure */
787 1.3 oster if (parity)
788 1.3 oster asm_p->numParityFailed++;
789 1.3 oster else {
790 1.3 oster asm_p->numDataFailed++;
791 1.3 oster }
792 1.3 oster asm_p->failedPDAs[asm_p->numFailedPDAs] = pda_p;
793 1.3 oster asm_p->numFailedPDAs++;
794 1.1 oster #if 0
795 1.3 oster switch (asm_p->numParityFailed + asm_p->numDataFailed) {
796 1.3 oster case 1:
797 1.3 oster asm_p->failedPDAs[0] = pda_p;
798 1.3 oster break;
799 1.3 oster case 2:
800 1.3 oster asm_p->failedPDAs[1] = pda_p;
801 1.3 oster default:
802 1.3 oster break;
803 1.3 oster }
804 1.1 oster #endif
805 1.3 oster }
806 1.3 oster /* the redirected access should never span a stripe unit boundary */
807 1.3 oster RF_ASSERT(rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress) ==
808 1.3 oster rf_RaidAddressToStripeUnitID(&raidPtr->Layout, pda_p->raidAddress + pda_p->numSector - 1));
809 1.3 oster RF_ASSERT(pda_p->col != -1);
810 1.1 oster }
811