rf_reconmap.c revision 1.6 1 1.6 oster /* $NetBSD: rf_reconmap.c,v 1.6 1999/08/14 21:44:24 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 1.1 oster
36 1.1 oster #include "rf_raid.h"
37 1.1 oster #include <sys/time.h>
38 1.1 oster #include "rf_general.h"
39 1.1 oster #include "rf_utils.h"
40 1.1 oster
41 1.1 oster /* special pointer values indicating that a reconstruction unit
42 1.1 oster * has been either totally reconstructed or not at all. Both
43 1.1 oster * are illegal pointer values, so you have to be careful not to
44 1.1 oster * dereference through them. RU_NOTHING must be zero, since
45 1.1 oster * MakeReconMap uses bzero to initialize the structure. These are used
46 1.1 oster * only at the head of the list.
47 1.1 oster */
48 1.1 oster #define RU_ALL ((RF_ReconMapListElem_t *) -1)
49 1.1 oster #define RU_NOTHING ((RF_ReconMapListElem_t *) 0)
50 1.1 oster
51 1.1 oster /* used to mark the end of the list */
52 1.1 oster #define RU_NIL ((RF_ReconMapListElem_t *) 0)
53 1.1 oster
54 1.1 oster
55 1.4 oster static void
56 1.4 oster compact_stat_entry(RF_Raid_t * raidPtr, RF_ReconMap_t * mapPtr,
57 1.4 oster int i);
58 1.4 oster static void crunch_list(RF_ReconMap_t * mapPtr, RF_ReconMapListElem_t * listPtr);
59 1.4 oster static RF_ReconMapListElem_t *
60 1.4 oster MakeReconMapListElem(RF_SectorNum_t startSector,
61 1.4 oster RF_SectorNum_t stopSector, RF_ReconMapListElem_t * next);
62 1.4 oster static void
63 1.4 oster FreeReconMapListElem(RF_ReconMap_t * mapPtr,
64 1.4 oster RF_ReconMapListElem_t * p);
65 1.4 oster static void update_size(RF_ReconMap_t * mapPtr, int size);
66 1.4 oster static void PrintList(RF_ReconMapListElem_t * listPtr);
67 1.1 oster
68 1.1 oster /*-----------------------------------------------------------------------------
69 1.1 oster *
70 1.4 oster * Creates and initializes new Reconstruction map
71 1.1 oster *
72 1.1 oster *-----------------------------------------------------------------------------*/
73 1.1 oster
74 1.4 oster RF_ReconMap_t *
75 1.4 oster rf_MakeReconMap(raidPtr, ru_sectors, disk_sectors, spareUnitsPerDisk)
76 1.4 oster RF_Raid_t *raidPtr;
77 1.4 oster RF_SectorCount_t ru_sectors; /* size of reconstruction unit in
78 1.4 oster * sectors */
79 1.4 oster RF_SectorCount_t disk_sectors; /* size of disk in sectors */
80 1.4 oster RF_ReconUnitCount_t spareUnitsPerDisk; /* zero unless distributed
81 1.4 oster * sparing */
82 1.4 oster {
83 1.4 oster RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
84 1.4 oster RF_ReconUnitCount_t num_rus = layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerRU;
85 1.4 oster RF_ReconMap_t *p;
86 1.4 oster int rc;
87 1.4 oster
88 1.4 oster RF_Malloc(p, sizeof(RF_ReconMap_t), (RF_ReconMap_t *));
89 1.4 oster p->sectorsPerReconUnit = ru_sectors;
90 1.4 oster p->sectorsInDisk = disk_sectors;
91 1.4 oster
92 1.4 oster p->totalRUs = num_rus;
93 1.4 oster p->spareRUs = spareUnitsPerDisk;
94 1.4 oster p->unitsLeft = num_rus - spareUnitsPerDisk;
95 1.4 oster
96 1.4 oster RF_Malloc(p->status, num_rus * sizeof(RF_ReconMapListElem_t *), (RF_ReconMapListElem_t **));
97 1.4 oster RF_ASSERT(p->status != (RF_ReconMapListElem_t **) NULL);
98 1.4 oster
99 1.4 oster (void) bzero((char *) p->status, num_rus * sizeof(RF_ReconMapListElem_t *));
100 1.4 oster
101 1.4 oster p->size = sizeof(RF_ReconMap_t) + num_rus * sizeof(RF_ReconMapListElem_t *);
102 1.4 oster p->maxSize = p->size;
103 1.4 oster
104 1.4 oster rc = rf_mutex_init(&p->mutex);
105 1.4 oster if (rc) {
106 1.4 oster RF_ERRORMSG3("Unable to init mutex file %s line %d rc=%d\n", __FILE__,
107 1.4 oster __LINE__, rc);
108 1.4 oster RF_Free(p->status, num_rus * sizeof(RF_ReconMapListElem_t *));
109 1.4 oster RF_Free(p, sizeof(RF_ReconMap_t));
110 1.4 oster return (NULL);
111 1.4 oster }
112 1.4 oster return (p);
113 1.1 oster }
114 1.1 oster
115 1.1 oster
116 1.1 oster /*-----------------------------------------------------------------------------
117 1.1 oster *
118 1.1 oster * marks a new set of sectors as reconstructed. All the possible mergings get
119 1.1 oster * complicated. To simplify matters, the approach I take is to just dump
120 1.1 oster * something into the list, and then clean it up (i.e. merge elements and
121 1.1 oster * eliminate redundant ones) in a second pass over the list (compact_stat_entry()).
122 1.1 oster * Not 100% efficient, since a structure can be allocated and then immediately
123 1.1 oster * freed, but it keeps this code from becoming (more of) a nightmare of
124 1.1 oster * special cases. The only thing that compact_stat_entry() assumes is that the
125 1.1 oster * list is sorted by startSector, and so this is the only condition I maintain
126 1.1 oster * here. (MCH)
127 1.1 oster *
128 1.1 oster *-----------------------------------------------------------------------------*/
129 1.1 oster
130 1.4 oster void
131 1.4 oster rf_ReconMapUpdate(raidPtr, mapPtr, startSector, stopSector)
132 1.4 oster RF_Raid_t *raidPtr;
133 1.4 oster RF_ReconMap_t *mapPtr;
134 1.4 oster RF_SectorNum_t startSector;
135 1.4 oster RF_SectorNum_t stopSector;
136 1.4 oster {
137 1.4 oster RF_SectorCount_t sectorsPerReconUnit = mapPtr->sectorsPerReconUnit;
138 1.4 oster RF_SectorNum_t i, first_in_RU, last_in_RU;
139 1.4 oster RF_ReconMapListElem_t *p, *pt;
140 1.4 oster
141 1.4 oster RF_LOCK_MUTEX(mapPtr->mutex);
142 1.6 oster RF_ASSERT(startSector >= 0 && stopSector < mapPtr->sectorsInDisk && stopSector >= startSector);
143 1.4 oster
144 1.4 oster while (startSector <= stopSector) {
145 1.4 oster i = startSector / mapPtr->sectorsPerReconUnit;
146 1.4 oster first_in_RU = i * sectorsPerReconUnit;
147 1.4 oster last_in_RU = first_in_RU + sectorsPerReconUnit - 1;
148 1.4 oster p = mapPtr->status[i];
149 1.4 oster if (p != RU_ALL) {
150 1.4 oster if (p == RU_NOTHING || p->startSector > startSector) { /* insert at front of
151 1.4 oster * list */
152 1.4 oster
153 1.4 oster mapPtr->status[i] = MakeReconMapListElem(startSector, RF_MIN(stopSector, last_in_RU), (p == RU_NOTHING) ? NULL : p);
154 1.4 oster update_size(mapPtr, sizeof(RF_ReconMapListElem_t));
155 1.4 oster
156 1.4 oster } else {/* general case */
157 1.4 oster do { /* search for place to insert */
158 1.4 oster pt = p;
159 1.4 oster p = p->next;
160 1.4 oster } while (p && (p->startSector < startSector));
161 1.4 oster pt->next = MakeReconMapListElem(startSector, RF_MIN(stopSector, last_in_RU), p);
162 1.4 oster update_size(mapPtr, sizeof(RF_ReconMapListElem_t));
163 1.4 oster }
164 1.4 oster compact_stat_entry(raidPtr, mapPtr, i);
165 1.4 oster }
166 1.4 oster startSector = RF_MIN(stopSector, last_in_RU) + 1;
167 1.4 oster }
168 1.4 oster RF_UNLOCK_MUTEX(mapPtr->mutex);
169 1.1 oster }
170 1.1 oster
171 1.1 oster
172 1.1 oster
173 1.1 oster /*-----------------------------------------------------------------------------
174 1.1 oster *
175 1.1 oster * performs whatever list compactions can be done, and frees any space
176 1.1 oster * that is no longer necessary. Assumes only that the list is sorted
177 1.1 oster * by startSector. crunch_list() compacts a single list as much as possible,
178 1.1 oster * and the second block of code deletes the entire list if possible.
179 1.1 oster * crunch_list() is also called from MakeReconMapAccessList().
180 1.1 oster *
181 1.1 oster * When a recon unit is detected to be fully reconstructed, we set the
182 1.1 oster * corresponding bit in the parity stripe map so that the head follow
183 1.1 oster * code will not select this parity stripe again. This is redundant (but
184 1.1 oster * harmless) when compact_stat_entry is called from the reconstruction code,
185 1.1 oster * but necessary when called from the user-write code.
186 1.1 oster *
187 1.1 oster *-----------------------------------------------------------------------------*/
188 1.1 oster
189 1.4 oster static void
190 1.4 oster compact_stat_entry(raidPtr, mapPtr, i)
191 1.4 oster RF_Raid_t *raidPtr;
192 1.4 oster RF_ReconMap_t *mapPtr;
193 1.4 oster int i;
194 1.4 oster {
195 1.4 oster RF_SectorCount_t sectorsPerReconUnit = mapPtr->sectorsPerReconUnit;
196 1.4 oster RF_ReconMapListElem_t *p = mapPtr->status[i];
197 1.4 oster
198 1.4 oster crunch_list(mapPtr, p);
199 1.4 oster
200 1.4 oster if ((p->startSector == i * sectorsPerReconUnit) &&
201 1.4 oster (p->stopSector == i * sectorsPerReconUnit + sectorsPerReconUnit - 1)) {
202 1.4 oster mapPtr->status[i] = RU_ALL;
203 1.4 oster mapPtr->unitsLeft--;
204 1.4 oster FreeReconMapListElem(mapPtr, p);
205 1.4 oster }
206 1.4 oster }
207 1.4 oster
208 1.4 oster static void
209 1.4 oster crunch_list(mapPtr, listPtr)
210 1.4 oster RF_ReconMap_t *mapPtr;
211 1.4 oster RF_ReconMapListElem_t *listPtr;
212 1.4 oster {
213 1.4 oster RF_ReconMapListElem_t *pt, *p = listPtr;
214 1.4 oster
215 1.4 oster if (!p)
216 1.4 oster return;
217 1.4 oster pt = p;
218 1.4 oster p = p->next;
219 1.4 oster while (p) {
220 1.4 oster if (pt->stopSector >= p->startSector - 1) {
221 1.4 oster pt->stopSector = RF_MAX(pt->stopSector, p->stopSector);
222 1.4 oster pt->next = p->next;
223 1.4 oster FreeReconMapListElem(mapPtr, p);
224 1.4 oster p = pt->next;
225 1.4 oster } else {
226 1.4 oster pt = p;
227 1.4 oster p = p->next;
228 1.4 oster }
229 1.4 oster }
230 1.1 oster }
231 1.1 oster /*-----------------------------------------------------------------------------
232 1.4 oster *
233 1.1 oster * Allocate and fill a new list element
234 1.1 oster *
235 1.1 oster *-----------------------------------------------------------------------------*/
236 1.1 oster
237 1.4 oster static RF_ReconMapListElem_t *
238 1.4 oster MakeReconMapListElem(
239 1.4 oster RF_SectorNum_t startSector,
240 1.4 oster RF_SectorNum_t stopSector,
241 1.4 oster RF_ReconMapListElem_t * next)
242 1.4 oster {
243 1.4 oster RF_ReconMapListElem_t *p;
244 1.4 oster
245 1.4 oster RF_Malloc(p, sizeof(RF_ReconMapListElem_t), (RF_ReconMapListElem_t *));
246 1.4 oster if (p == NULL)
247 1.4 oster return (NULL);
248 1.4 oster p->startSector = startSector;
249 1.4 oster p->stopSector = stopSector;
250 1.4 oster p->next = next;
251 1.4 oster return (p);
252 1.1 oster }
253 1.1 oster /*-----------------------------------------------------------------------------
254 1.4 oster *
255 1.1 oster * Free a list element
256 1.1 oster *
257 1.1 oster *-----------------------------------------------------------------------------*/
258 1.1 oster
259 1.4 oster static void
260 1.4 oster FreeReconMapListElem(mapPtr, p)
261 1.4 oster RF_ReconMap_t *mapPtr;
262 1.4 oster RF_ReconMapListElem_t *p;
263 1.4 oster {
264 1.4 oster int delta;
265 1.4 oster
266 1.4 oster if (mapPtr) {
267 1.4 oster delta = 0 - (int) sizeof(RF_ReconMapListElem_t);
268 1.4 oster update_size(mapPtr, delta);
269 1.4 oster }
270 1.4 oster RF_Free(p, sizeof(*p));
271 1.1 oster }
272 1.1 oster /*-----------------------------------------------------------------------------
273 1.4 oster *
274 1.1 oster * Free an entire status structure. Inefficient, but can be called at any time.
275 1.1 oster *
276 1.1 oster *-----------------------------------------------------------------------------*/
277 1.4 oster void
278 1.4 oster rf_FreeReconMap(mapPtr)
279 1.4 oster RF_ReconMap_t *mapPtr;
280 1.4 oster {
281 1.4 oster RF_ReconMapListElem_t *p, *q;
282 1.4 oster RF_ReconUnitCount_t numRUs;
283 1.4 oster RF_ReconUnitNum_t i;
284 1.4 oster
285 1.4 oster numRUs = mapPtr->sectorsInDisk / mapPtr->sectorsPerReconUnit;
286 1.4 oster if (mapPtr->sectorsInDisk % mapPtr->sectorsPerReconUnit)
287 1.4 oster numRUs++;
288 1.4 oster
289 1.4 oster for (i = 0; i < numRUs; i++) {
290 1.4 oster p = mapPtr->status[i];
291 1.4 oster while (p != RU_NOTHING && p != RU_ALL) {
292 1.4 oster q = p;
293 1.4 oster p = p->next;
294 1.4 oster RF_Free(q, sizeof(*q));
295 1.4 oster }
296 1.4 oster }
297 1.4 oster rf_mutex_destroy(&mapPtr->mutex);
298 1.4 oster RF_Free(mapPtr->status, mapPtr->totalRUs * sizeof(RF_ReconMapListElem_t *));
299 1.4 oster RF_Free(mapPtr, sizeof(RF_ReconMap_t));
300 1.1 oster }
301 1.1 oster /*-----------------------------------------------------------------------------
302 1.1 oster *
303 1.1 oster * returns nonzero if the indicated RU has been reconstructed already
304 1.1 oster *
305 1.1 oster *---------------------------------------------------------------------------*/
306 1.1 oster
307 1.4 oster int
308 1.4 oster rf_CheckRUReconstructed(mapPtr, startSector)
309 1.4 oster RF_ReconMap_t *mapPtr;
310 1.4 oster RF_SectorNum_t startSector;
311 1.1 oster {
312 1.4 oster RF_ReconMapListElem_t *l; /* used for searching */
313 1.4 oster RF_ReconUnitNum_t i;
314 1.1 oster
315 1.4 oster i = startSector / mapPtr->sectorsPerReconUnit;
316 1.4 oster l = mapPtr->status[i];
317 1.4 oster return ((l == RU_ALL) ? 1 : 0);
318 1.1 oster }
319 1.1 oster
320 1.4 oster RF_ReconUnitCount_t
321 1.4 oster rf_UnitsLeftToReconstruct(mapPtr)
322 1.4 oster RF_ReconMap_t *mapPtr;
323 1.1 oster {
324 1.4 oster RF_ASSERT(mapPtr != NULL);
325 1.4 oster return (mapPtr->unitsLeft);
326 1.1 oster }
327 1.1 oster /* updates the size fields of a status descriptor */
328 1.4 oster static void
329 1.4 oster update_size(mapPtr, size)
330 1.4 oster RF_ReconMap_t *mapPtr;
331 1.4 oster int size;
332 1.4 oster {
333 1.4 oster mapPtr->size += size;
334 1.4 oster mapPtr->maxSize = RF_MAX(mapPtr->size, mapPtr->maxSize);
335 1.4 oster }
336 1.4 oster
337 1.4 oster static void
338 1.4 oster PrintList(listPtr)
339 1.4 oster RF_ReconMapListElem_t *listPtr;
340 1.4 oster {
341 1.4 oster while (listPtr) {
342 1.4 oster printf("%d,%d -> ", (int) listPtr->startSector, (int) listPtr->stopSector);
343 1.4 oster listPtr = listPtr->next;
344 1.4 oster }
345 1.4 oster printf("\n");
346 1.4 oster }
347 1.4 oster
348 1.4 oster void
349 1.4 oster rf_PrintReconMap(raidPtr, mapPtr, frow, fcol)
350 1.4 oster RF_Raid_t *raidPtr;
351 1.4 oster RF_ReconMap_t *mapPtr;
352 1.4 oster RF_RowCol_t frow;
353 1.4 oster RF_RowCol_t fcol;
354 1.4 oster {
355 1.4 oster RF_ReconUnitCount_t numRUs;
356 1.4 oster RF_ReconMapListElem_t *p;
357 1.4 oster RF_ReconUnitNum_t i;
358 1.4 oster
359 1.4 oster numRUs = mapPtr->totalRUs;
360 1.4 oster if (mapPtr->sectorsInDisk % mapPtr->sectorsPerReconUnit)
361 1.4 oster numRUs++;
362 1.4 oster
363 1.4 oster for (i = 0; i < numRUs; i++) {
364 1.4 oster p = mapPtr->status[i];
365 1.4 oster if (p == RU_ALL)/* printf("[%d] ALL\n",i) */
366 1.4 oster ;
367 1.4 oster else
368 1.4 oster if (p == RU_NOTHING) {
369 1.4 oster printf("%d: Unreconstructed\n", i);
370 1.4 oster } else {
371 1.4 oster printf("%d: ", i);
372 1.4 oster PrintList(p);
373 1.4 oster }
374 1.4 oster }
375 1.4 oster }
376 1.4 oster
377 1.4 oster void
378 1.4 oster rf_PrintReconSchedule(mapPtr, starttime)
379 1.4 oster RF_ReconMap_t *mapPtr;
380 1.4 oster struct timeval *starttime;
381 1.4 oster {
382 1.4 oster static int old_pctg = -1;
383 1.4 oster struct timeval tv, diff;
384 1.4 oster int new_pctg;
385 1.4 oster
386 1.4 oster new_pctg = 100 - (rf_UnitsLeftToReconstruct(mapPtr) * 100 / mapPtr->totalRUs);
387 1.4 oster if (new_pctg != old_pctg) {
388 1.4 oster RF_GETTIME(tv);
389 1.4 oster RF_TIMEVAL_DIFF(starttime, &tv, &diff);
390 1.4 oster printf("%d %d.%06d\n", (int) new_pctg, (int) diff.tv_sec, (int) diff.tv_usec);
391 1.4 oster old_pctg = new_pctg;
392 1.4 oster }
393 1.1 oster }
394