rf_reconmap.c revision 1.14 1 1.14 oster /* $NetBSD: rf_reconmap.c,v 1.14 2002/10/05 22:45:46 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 * rf_reconmap.c
31 1.1 oster *
32 1.1 oster * code to maintain a map of what sectors have/have not been reconstructed
33 1.1 oster *
34 1.1 oster *************************************************************************/
35 1.8 lukem
36 1.8 lukem #include <sys/cdefs.h>
37 1.14 oster __KERNEL_RCSID(0, "$NetBSD: rf_reconmap.c,v 1.14 2002/10/05 22:45:46 oster Exp $");
38 1.1 oster
39 1.1 oster #include "rf_raid.h"
40 1.1 oster #include <sys/time.h>
41 1.1 oster #include "rf_general.h"
42 1.1 oster #include "rf_utils.h"
43 1.1 oster
44 1.1 oster /* special pointer values indicating that a reconstruction unit
45 1.1 oster * has been either totally reconstructed or not at all. Both
46 1.1 oster * are illegal pointer values, so you have to be careful not to
47 1.1 oster * dereference through them. RU_NOTHING must be zero, since
48 1.7 thorpej * MakeReconMap uses memset to initialize the structure. These are used
49 1.1 oster * only at the head of the list.
50 1.1 oster */
51 1.1 oster #define RU_ALL ((RF_ReconMapListElem_t *) -1)
52 1.1 oster #define RU_NOTHING ((RF_ReconMapListElem_t *) 0)
53 1.1 oster
54 1.4 oster static void
55 1.4 oster compact_stat_entry(RF_Raid_t * raidPtr, RF_ReconMap_t * mapPtr,
56 1.4 oster int i);
57 1.14 oster static void crunch_list(RF_ReconMapListElem_t * listPtr);
58 1.4 oster static RF_ReconMapListElem_t *
59 1.4 oster MakeReconMapListElem(RF_SectorNum_t startSector,
60 1.4 oster RF_SectorNum_t stopSector, RF_ReconMapListElem_t * next);
61 1.4 oster static void
62 1.14 oster FreeReconMapListElem(RF_ReconMapListElem_t * p);
63 1.10 oster #if 0
64 1.4 oster static void PrintList(RF_ReconMapListElem_t * listPtr);
65 1.10 oster #endif
66 1.1 oster
67 1.12 oster /*---------------------------------------------------------------------------
68 1.1 oster *
69 1.4 oster * Creates and initializes new Reconstruction map
70 1.1 oster *
71 1.12 oster *-------------------------------------------------------------------------*/
72 1.1 oster
73 1.4 oster RF_ReconMap_t *
74 1.4 oster rf_MakeReconMap(raidPtr, ru_sectors, disk_sectors, spareUnitsPerDisk)
75 1.4 oster RF_Raid_t *raidPtr;
76 1.4 oster RF_SectorCount_t ru_sectors; /* size of reconstruction unit in
77 1.4 oster * sectors */
78 1.4 oster RF_SectorCount_t disk_sectors; /* size of disk in sectors */
79 1.4 oster RF_ReconUnitCount_t spareUnitsPerDisk; /* zero unless distributed
80 1.4 oster * sparing */
81 1.4 oster {
82 1.4 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
83 1.4 oster RF_ReconUnitCount_t num_rus = layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerRU;
84 1.4 oster RF_ReconMap_t *p;
85 1.4 oster int rc;
86 1.4 oster
87 1.4 oster RF_Malloc(p, sizeof(RF_ReconMap_t), (RF_ReconMap_t *));
88 1.4 oster p->sectorsPerReconUnit = ru_sectors;
89 1.4 oster p->sectorsInDisk = disk_sectors;
90 1.4 oster
91 1.4 oster p->totalRUs = num_rus;
92 1.4 oster p->spareRUs = spareUnitsPerDisk;
93 1.4 oster p->unitsLeft = num_rus - spareUnitsPerDisk;
94 1.4 oster
95 1.4 oster RF_Malloc(p->status, num_rus * sizeof(RF_ReconMapListElem_t *), (RF_ReconMapListElem_t **));
96 1.4 oster RF_ASSERT(p->status != (RF_ReconMapListElem_t **) NULL);
97 1.4 oster
98 1.7 thorpej (void) memset((char *) p->status, 0,
99 1.7 thorpej num_rus * sizeof(RF_ReconMapListElem_t *));
100 1.4 oster
101 1.4 oster rc = rf_mutex_init(&p->mutex);
102 1.4 oster if (rc) {
103 1.9 oster rf_print_unable_to_init_mutex(__FILE__, __LINE__, rc);
104 1.4 oster RF_Free(p->status, num_rus * sizeof(RF_ReconMapListElem_t *));
105 1.4 oster RF_Free(p, sizeof(RF_ReconMap_t));
106 1.4 oster return (NULL);
107 1.4 oster }
108 1.4 oster return (p);
109 1.1 oster }
110 1.1 oster
111 1.1 oster
112 1.12 oster /*---------------------------------------------------------------------------
113 1.1 oster *
114 1.12 oster * marks a new set of sectors as reconstructed. All the possible
115 1.12 oster * mergings get complicated. To simplify matters, the approach I take
116 1.12 oster * is to just dump something into the list, and then clean it up
117 1.12 oster * (i.e. merge elements and eliminate redundant ones) in a second pass
118 1.12 oster * over the list (compact_stat_entry()). Not 100% efficient, since a
119 1.12 oster * structure can be allocated and then immediately freed, but it keeps
120 1.12 oster * this code from becoming (more of) a nightmare of special cases.
121 1.12 oster * The only thing that compact_stat_entry() assumes is that the list
122 1.12 oster * is sorted by startSector, and so this is the only condition I
123 1.12 oster * maintain here. (MCH)
124 1.1 oster *
125 1.12 oster *-------------------------------------------------------------------------*/
126 1.1 oster
127 1.4 oster void
128 1.4 oster rf_ReconMapUpdate(raidPtr, mapPtr, startSector, stopSector)
129 1.4 oster RF_Raid_t *raidPtr;
130 1.4 oster RF_ReconMap_t *mapPtr;
131 1.4 oster RF_SectorNum_t startSector;
132 1.4 oster RF_SectorNum_t stopSector;
133 1.4 oster {
134 1.4 oster RF_SectorCount_t sectorsPerReconUnit = mapPtr->sectorsPerReconUnit;
135 1.4 oster RF_SectorNum_t i, first_in_RU, last_in_RU;
136 1.4 oster RF_ReconMapListElem_t *p, *pt;
137 1.4 oster
138 1.4 oster RF_LOCK_MUTEX(mapPtr->mutex);
139 1.12 oster RF_ASSERT(startSector >= 0 && stopSector < mapPtr->sectorsInDisk &&
140 1.12 oster stopSector >= startSector);
141 1.4 oster
142 1.4 oster while (startSector <= stopSector) {
143 1.4 oster i = startSector / mapPtr->sectorsPerReconUnit;
144 1.4 oster first_in_RU = i * sectorsPerReconUnit;
145 1.4 oster last_in_RU = first_in_RU + sectorsPerReconUnit - 1;
146 1.4 oster p = mapPtr->status[i];
147 1.4 oster if (p != RU_ALL) {
148 1.12 oster if (p == RU_NOTHING || p->startSector > startSector) {
149 1.12 oster /* insert at front of list */
150 1.4 oster
151 1.4 oster mapPtr->status[i] = MakeReconMapListElem(startSector, RF_MIN(stopSector, last_in_RU), (p == RU_NOTHING) ? NULL : p);
152 1.4 oster
153 1.4 oster } else {/* general case */
154 1.4 oster do { /* search for place to insert */
155 1.4 oster pt = p;
156 1.4 oster p = p->next;
157 1.4 oster } while (p && (p->startSector < startSector));
158 1.4 oster pt->next = MakeReconMapListElem(startSector, RF_MIN(stopSector, last_in_RU), p);
159 1.14 oster
160 1.4 oster }
161 1.4 oster compact_stat_entry(raidPtr, mapPtr, i);
162 1.4 oster }
163 1.4 oster startSector = RF_MIN(stopSector, last_in_RU) + 1;
164 1.4 oster }
165 1.4 oster RF_UNLOCK_MUTEX(mapPtr->mutex);
166 1.1 oster }
167 1.1 oster
168 1.1 oster
169 1.1 oster
170 1.12 oster /*---------------------------------------------------------------------------
171 1.1 oster *
172 1.1 oster * performs whatever list compactions can be done, and frees any space
173 1.1 oster * that is no longer necessary. Assumes only that the list is sorted
174 1.12 oster * by startSector. crunch_list() compacts a single list as much as
175 1.12 oster * possible, and the second block of code deletes the entire list if
176 1.12 oster * possible. crunch_list() is also called from
177 1.12 oster * MakeReconMapAccessList().
178 1.1 oster *
179 1.1 oster * When a recon unit is detected to be fully reconstructed, we set the
180 1.1 oster * corresponding bit in the parity stripe map so that the head follow
181 1.12 oster * code will not select this parity stripe again. This is redundant
182 1.12 oster * (but harmless) when compact_stat_entry is called from the
183 1.12 oster * reconstruction code, but necessary when called from the user-write
184 1.12 oster * code.
185 1.1 oster *
186 1.12 oster *-------------------------------------------------------------------------*/
187 1.1 oster
188 1.4 oster static void
189 1.4 oster compact_stat_entry(raidPtr, mapPtr, i)
190 1.4 oster RF_Raid_t *raidPtr;
191 1.4 oster RF_ReconMap_t *mapPtr;
192 1.4 oster int i;
193 1.4 oster {
194 1.4 oster RF_SectorCount_t sectorsPerReconUnit = mapPtr->sectorsPerReconUnit;
195 1.4 oster RF_ReconMapListElem_t *p = mapPtr->status[i];
196 1.4 oster
197 1.14 oster crunch_list(p);
198 1.4 oster
199 1.4 oster if ((p->startSector == i * sectorsPerReconUnit) &&
200 1.12 oster (p->stopSector == i * sectorsPerReconUnit +
201 1.12 oster sectorsPerReconUnit - 1)) {
202 1.4 oster mapPtr->status[i] = RU_ALL;
203 1.4 oster mapPtr->unitsLeft--;
204 1.14 oster FreeReconMapListElem(p);
205 1.4 oster }
206 1.4 oster }
207 1.4 oster
208 1.4 oster static void
209 1.14 oster crunch_list(listPtr)
210 1.4 oster RF_ReconMapListElem_t *listPtr;
211 1.4 oster {
212 1.4 oster RF_ReconMapListElem_t *pt, *p = listPtr;
213 1.4 oster
214 1.4 oster if (!p)
215 1.4 oster return;
216 1.4 oster pt = p;
217 1.4 oster p = p->next;
218 1.4 oster while (p) {
219 1.4 oster if (pt->stopSector >= p->startSector - 1) {
220 1.4 oster pt->stopSector = RF_MAX(pt->stopSector, p->stopSector);
221 1.4 oster pt->next = p->next;
222 1.14 oster FreeReconMapListElem(p);
223 1.4 oster p = pt->next;
224 1.4 oster } else {
225 1.4 oster pt = p;
226 1.4 oster p = p->next;
227 1.4 oster }
228 1.4 oster }
229 1.1 oster }
230 1.12 oster /*---------------------------------------------------------------------------
231 1.4 oster *
232 1.1 oster * Allocate and fill a new list element
233 1.1 oster *
234 1.12 oster *-------------------------------------------------------------------------*/
235 1.1 oster
236 1.4 oster static RF_ReconMapListElem_t *
237 1.4 oster MakeReconMapListElem(
238 1.4 oster RF_SectorNum_t startSector,
239 1.4 oster RF_SectorNum_t stopSector,
240 1.4 oster RF_ReconMapListElem_t * next)
241 1.4 oster {
242 1.4 oster RF_ReconMapListElem_t *p;
243 1.4 oster
244 1.4 oster RF_Malloc(p, sizeof(RF_ReconMapListElem_t), (RF_ReconMapListElem_t *));
245 1.4 oster if (p == NULL)
246 1.4 oster return (NULL);
247 1.4 oster p->startSector = startSector;
248 1.4 oster p->stopSector = stopSector;
249 1.4 oster p->next = next;
250 1.4 oster return (p);
251 1.1 oster }
252 1.12 oster /*---------------------------------------------------------------------------
253 1.4 oster *
254 1.1 oster * Free a list element
255 1.1 oster *
256 1.12 oster *-------------------------------------------------------------------------*/
257 1.1 oster
258 1.4 oster static void
259 1.14 oster FreeReconMapListElem(p)
260 1.4 oster RF_ReconMapListElem_t *p;
261 1.4 oster {
262 1.4 oster RF_Free(p, sizeof(*p));
263 1.1 oster }
264 1.12 oster /*---------------------------------------------------------------------------
265 1.4 oster *
266 1.12 oster * Free an entire status structure. Inefficient, but can be called at
267 1.12 oster * any time.
268 1.1 oster *
269 1.12 oster *-------------------------------------------------------------------------*/
270 1.4 oster void
271 1.4 oster rf_FreeReconMap(mapPtr)
272 1.4 oster RF_ReconMap_t *mapPtr;
273 1.4 oster {
274 1.4 oster RF_ReconMapListElem_t *p, *q;
275 1.4 oster RF_ReconUnitCount_t numRUs;
276 1.4 oster RF_ReconUnitNum_t i;
277 1.4 oster
278 1.4 oster numRUs = mapPtr->sectorsInDisk / mapPtr->sectorsPerReconUnit;
279 1.4 oster if (mapPtr->sectorsInDisk % mapPtr->sectorsPerReconUnit)
280 1.4 oster numRUs++;
281 1.4 oster
282 1.4 oster for (i = 0; i < numRUs; i++) {
283 1.4 oster p = mapPtr->status[i];
284 1.4 oster while (p != RU_NOTHING && p != RU_ALL) {
285 1.4 oster q = p;
286 1.4 oster p = p->next;
287 1.4 oster RF_Free(q, sizeof(*q));
288 1.4 oster }
289 1.4 oster }
290 1.4 oster rf_mutex_destroy(&mapPtr->mutex);
291 1.12 oster RF_Free(mapPtr->status, mapPtr->totalRUs *
292 1.12 oster sizeof(RF_ReconMapListElem_t *));
293 1.4 oster RF_Free(mapPtr, sizeof(RF_ReconMap_t));
294 1.1 oster }
295 1.12 oster /*---------------------------------------------------------------------------
296 1.1 oster *
297 1.1 oster * returns nonzero if the indicated RU has been reconstructed already
298 1.1 oster *
299 1.12 oster *-------------------------------------------------------------------------*/
300 1.1 oster
301 1.4 oster int
302 1.4 oster rf_CheckRUReconstructed(mapPtr, startSector)
303 1.4 oster RF_ReconMap_t *mapPtr;
304 1.4 oster RF_SectorNum_t startSector;
305 1.1 oster {
306 1.4 oster RF_ReconMapListElem_t *l; /* used for searching */
307 1.4 oster RF_ReconUnitNum_t i;
308 1.1 oster
309 1.4 oster i = startSector / mapPtr->sectorsPerReconUnit;
310 1.4 oster l = mapPtr->status[i];
311 1.4 oster return ((l == RU_ALL) ? 1 : 0);
312 1.1 oster }
313 1.1 oster
314 1.4 oster RF_ReconUnitCount_t
315 1.4 oster rf_UnitsLeftToReconstruct(mapPtr)
316 1.4 oster RF_ReconMap_t *mapPtr;
317 1.1 oster {
318 1.4 oster RF_ASSERT(mapPtr != NULL);
319 1.4 oster return (mapPtr->unitsLeft);
320 1.4 oster }
321 1.4 oster
322 1.10 oster #if 0
323 1.4 oster static void
324 1.4 oster PrintList(listPtr)
325 1.4 oster RF_ReconMapListElem_t *listPtr;
326 1.4 oster {
327 1.4 oster while (listPtr) {
328 1.12 oster printf("%d,%d -> ", (int) listPtr->startSector,
329 1.12 oster (int) listPtr->stopSector);
330 1.4 oster listPtr = listPtr->next;
331 1.4 oster }
332 1.4 oster printf("\n");
333 1.4 oster }
334 1.4 oster
335 1.4 oster void
336 1.4 oster rf_PrintReconMap(raidPtr, mapPtr, frow, fcol)
337 1.4 oster RF_Raid_t *raidPtr;
338 1.4 oster RF_ReconMap_t *mapPtr;
339 1.4 oster RF_RowCol_t frow;
340 1.4 oster RF_RowCol_t fcol;
341 1.4 oster {
342 1.4 oster RF_ReconUnitCount_t numRUs;
343 1.4 oster RF_ReconMapListElem_t *p;
344 1.4 oster RF_ReconUnitNum_t i;
345 1.4 oster
346 1.4 oster numRUs = mapPtr->totalRUs;
347 1.4 oster if (mapPtr->sectorsInDisk % mapPtr->sectorsPerReconUnit)
348 1.4 oster numRUs++;
349 1.4 oster
350 1.4 oster for (i = 0; i < numRUs; i++) {
351 1.4 oster p = mapPtr->status[i];
352 1.4 oster if (p == RU_ALL)/* printf("[%d] ALL\n",i) */
353 1.4 oster ;
354 1.4 oster else
355 1.4 oster if (p == RU_NOTHING) {
356 1.4 oster printf("%d: Unreconstructed\n", i);
357 1.4 oster } else {
358 1.4 oster printf("%d: ", i);
359 1.4 oster PrintList(p);
360 1.4 oster }
361 1.4 oster }
362 1.4 oster }
363 1.10 oster #endif
364 1.4 oster
365 1.11 oster #if RF_DEBUG_RECON
366 1.4 oster void
367 1.4 oster rf_PrintReconSchedule(mapPtr, starttime)
368 1.4 oster RF_ReconMap_t *mapPtr;
369 1.4 oster struct timeval *starttime;
370 1.4 oster {
371 1.4 oster static int old_pctg = -1;
372 1.4 oster struct timeval tv, diff;
373 1.4 oster int new_pctg;
374 1.4 oster
375 1.12 oster new_pctg = 100 - (rf_UnitsLeftToReconstruct(mapPtr) *
376 1.12 oster 100 / mapPtr->totalRUs);
377 1.4 oster if (new_pctg != old_pctg) {
378 1.4 oster RF_GETTIME(tv);
379 1.4 oster RF_TIMEVAL_DIFF(starttime, &tv, &diff);
380 1.12 oster printf("%d %d.%06d\n", (int) new_pctg, (int) diff.tv_sec,
381 1.12 oster (int) diff.tv_usec);
382 1.4 oster old_pctg = new_pctg;
383 1.4 oster }
384 1.1 oster }
385 1.11 oster #endif
386 1.11 oster
387