2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
13 #include <afsconfig.h>
14 #include "../afs/param.h"
18 #include "../afs/sysincludes.h" /*Standard vendor system headers*/
19 #include "../afs/afsincludes.h" /*AFS-based standard headers*/
20 #include "../afs/afs_stats.h" /* statistics */
21 #include "../afs/afs_cbqueue.h"
22 #include "../afs/afs_osidnlc.h"
24 /* Forward declarations. */
25 static void afs_GetDownD(int anumber, int *aneedSpace);
26 static void afs_FreeDiscardedDCache(void);
27 static void afs_DiscardDCache(struct dcache *);
28 static void afs_FreeDCache(struct dcache *);
31 * --------------------- Exported definitions ---------------------
33 afs_lock_t afs_xdcache; /*Lock: alloc new disk cache entries*/
34 afs_int32 afs_freeDCList; /*Free list for disk cache entries*/
35 afs_int32 afs_freeDCCount; /*Count of elts in freeDCList*/
36 afs_int32 afs_discardDCList; /*Discarded disk cache entries*/
37 afs_int32 afs_discardDCCount; /*Count of elts in discardDCList*/
38 struct dcache *afs_freeDSList; /*Free list for disk slots */
39 struct dcache *afs_Initial_freeDSList; /*Initial list for above*/
40 ino_t cacheInode; /*Inode for CacheItems file*/
41 struct osi_file *afs_cacheInodep = 0; /* file for CacheItems inode */
42 struct afs_q afs_DLRU; /*dcache LRU*/
43 afs_int32 afs_dhashsize = 1024;
44 afs_int32 *afs_dvhashTbl; /*Data cache hash table*/
45 afs_int32 *afs_dchashTbl; /*Data cache hash table*/
46 afs_int32 *afs_dvnextTbl; /*Dcache hash table links */
47 afs_int32 *afs_dcnextTbl; /*Dcache hash table links */
48 struct dcache **afs_indexTable; /*Pointers to dcache entries*/
49 afs_hyper_t *afs_indexTimes; /*Dcache entry Access times*/
50 afs_int32 *afs_indexUnique; /*dcache entry Fid.Unique */
51 unsigned char *afs_indexFlags; /*(only one) Is there data there?*/
52 afs_hyper_t afs_indexCounter; /*Fake time for marking index
54 afs_int32 afs_cacheFiles =0; /*Size of afs_indexTable*/
55 afs_int32 afs_cacheBlocks; /*1K blocks in cache*/
56 afs_int32 afs_cacheStats; /*Stat entries in cache*/
57 afs_int32 afs_blocksUsed; /*Number of blocks in use*/
58 afs_int32 afs_blocksDiscarded; /*Blocks freed but not truncated */
59 afs_int32 afs_fsfragsize = 1023; /*Underlying Filesystem minimum unit
60 *of disk allocation usually 1K
61 *this value is (truefrag -1 ) to
62 *save a bunch of subtracts... */
63 #ifdef AFS_64BIT_CLIENT
64 #ifdef AFS_VM_RDWR_ENV
65 afs_size_t afs_vmMappingEnd; /* for large files (>= 2GB) the VM
66 * mapping an 32bit addressing machines
67 * can only be used below the 2 GB
68 * line. From this point upwards we
69 * must do direct I/O into the cache
70 * files. The value should be on a
72 #endif /* AFS_VM_RDWR_ENV */
73 #endif /* AFS_64BIT_CLIENT */
75 /* The following is used to ensure that new dcache's aren't obtained when
76 * the cache is nearly full.
78 int afs_WaitForCacheDrain = 0;
79 int afs_TruncateDaemonRunning = 0;
80 int afs_CacheTooFull = 0;
82 afs_int32 afs_dcentries; /* In-memory dcache entries */
85 int dcacheDisabled = 0;
87 extern struct dcache *afs_UFSGetDSlot();
88 extern struct volume *afs_UFSGetVolSlot();
89 extern int osi_UFSTruncate(), afs_osi_Read(), afs_osi_Write(), osi_UFSClose();
90 extern int afs_UFSRead(), afs_UFSWrite();
91 extern int afs_UFSHandleLink();
92 static int afs_UFSCacheFetchProc(), afs_UFSCacheStoreProc();
93 struct afs_cacheOps afs_UfsCacheOps = {
101 afs_UFSCacheFetchProc,
102 afs_UFSCacheStoreProc,
108 struct afs_cacheOps afs_MemCacheOps = {
110 afs_MemCacheTruncate,
116 afs_MemCacheFetchProc,
117 afs_MemCacheStoreProc,
123 int cacheDiskType; /*Type of backing disk for cache*/
124 struct afs_cacheOps *afs_cacheType;
133 * Warn about failing to store a file.
136 * acode : Associated error code.
137 * avolume : Volume involved.
138 * aflags : How to handle the output:
139 * aflags & 1: Print out on console
140 * aflags & 2: Print out on controlling tty
143 * Call this from close call when vnodeops is RCS unlocked.
146 void afs_StoreWarn(register afs_int32 acode, afs_int32 avolume, register afs_int32 aflags)
148 static char problem_fmt[] =
149 "afs: failed to store file in volume %d (%s)\n";
150 static char problem_fmt_w_error[] =
151 "afs: failed to store file in volume %d (error %d)\n";
152 static char netproblems[] = "network problems";
153 static char partfull[] = "partition full";
154 static char overquota[] = "over quota";
156 AFS_STATCNT(afs_StoreWarn);
162 afs_warn(problem_fmt, avolume, netproblems);
164 afs_warnuser(problem_fmt, avolume, netproblems);
167 if (acode == ENOSPC) {
172 afs_warn(problem_fmt, avolume, partfull);
174 afs_warnuser(problem_fmt, avolume, partfull);
178 /* EDQUOT doesn't exist on solaris and won't be sent by the server.
179 * Instead ENOSPC will be sent...
181 if (acode == EDQUOT) {
186 afs_warn(problem_fmt, avolume, overquota);
188 afs_warnuser(problem_fmt, avolume, overquota);
196 afs_warn(problem_fmt_w_error, avolume, acode);
198 afs_warnuser(problem_fmt_w_error, avolume, acode);
202 void afs_MaybeWakeupTruncateDaemon(void)
204 if (!afs_CacheTooFull && afs_CacheIsTooFull()) {
205 afs_CacheTooFull = 1;
206 if (!afs_TruncateDaemonRunning)
207 afs_osi_Wakeup((int *)afs_CacheTruncateDaemon);
208 } else if (!afs_TruncateDaemonRunning &&
209 afs_blocksDiscarded > CM_MAXDISCARDEDCHUNKS) {
210 afs_osi_Wakeup((int *)afs_CacheTruncateDaemon);
214 /* Keep statistics on run time for afs_CacheTruncateDaemon. This is a
215 * struct so we need only export one symbol for AIX.
217 static struct CTD_stats {
218 osi_timeval_t CTD_beforeSleep;
219 osi_timeval_t CTD_afterSleep;
220 osi_timeval_t CTD_sleepTime;
221 osi_timeval_t CTD_runTime;
225 u_int afs_min_cache = 0;
226 void afs_CacheTruncateDaemon(void)
228 osi_timeval_t CTD_tmpTime;
231 u_int dc_hiwat = (100-CM_DCACHECOUNTFREEPCT+CM_DCACHEEXTRAPCT)*afs_cacheFiles/100;
232 afs_min_cache = (((10 * AFS_CHUNKSIZE(0)) + afs_fsfragsize) & ~afs_fsfragsize)>>10;
234 osi_GetuTime(&CTD_stats.CTD_afterSleep);
235 afs_TruncateDaemonRunning = 1;
237 cb_lowat = ((CM_DCACHESPACEFREEPCT-CM_DCACHEEXTRAPCT)
238 * afs_cacheBlocks) / 100;
239 MObtainWriteLock(&afs_xdcache,266);
240 if (afs_CacheTooFull) {
241 int space_needed, slots_needed;
242 /* if we get woken up, we should try to clean something out */
243 for (counter = 0; counter < 10; counter++) {
244 space_needed = afs_blocksUsed - afs_blocksDiscarded - cb_lowat;
245 slots_needed = dc_hiwat - afs_freeDCCount - afs_discardDCCount;
246 afs_GetDownD(slots_needed, &space_needed);
247 if ((space_needed <= 0) && (slots_needed <= 0)) {
250 if (afs_termState == AFSOP_STOP_TRUNCDAEMON)
253 if (!afs_CacheIsTooFull())
254 afs_CacheTooFull = 0;
256 MReleaseWriteLock(&afs_xdcache);
259 * This is a defensive check to try to avoid starving threads
260 * that may need the global lock so thay can help free some
261 * cache space. If this thread won't be sleeping or truncating
262 * any cache files then give up the global lock so other
263 * threads get a chance to run.
265 if ((afs_termState!=AFSOP_STOP_TRUNCDAEMON) && afs_CacheTooFull &&
266 (!afs_blocksDiscarded || afs_WaitForCacheDrain)) {
267 afs_osi_Wait(100, 0, 0); /* 100 milliseconds */
271 * This is where we free the discarded cache elements.
273 while(afs_blocksDiscarded && !afs_WaitForCacheDrain &&
274 (afs_termState!=AFSOP_STOP_TRUNCDAEMON))
276 afs_FreeDiscardedDCache();
279 /* See if we need to continue to run. Someone may have
280 * signalled us while we were executing.
282 if (!afs_WaitForCacheDrain && !afs_CacheTooFull &&
283 (afs_termState!=AFSOP_STOP_TRUNCDAEMON))
285 /* Collect statistics on truncate daemon. */
286 CTD_stats.CTD_nSleeps++;
287 osi_GetuTime(&CTD_stats.CTD_beforeSleep);
288 afs_stats_GetDiff(CTD_tmpTime, CTD_stats.CTD_afterSleep,
289 CTD_stats.CTD_beforeSleep);
290 afs_stats_AddTo(CTD_stats.CTD_runTime, CTD_tmpTime);
292 afs_TruncateDaemonRunning = 0;
293 afs_osi_Sleep((int *)afs_CacheTruncateDaemon);
294 afs_TruncateDaemonRunning = 1;
296 osi_GetuTime(&CTD_stats.CTD_afterSleep);
297 afs_stats_GetDiff(CTD_tmpTime, CTD_stats.CTD_beforeSleep,
298 CTD_stats.CTD_afterSleep);
299 afs_stats_AddTo(CTD_stats.CTD_sleepTime, CTD_tmpTime);
301 if (afs_termState == AFSOP_STOP_TRUNCDAEMON) {
303 afs_termState = AFSOP_STOP_AFSDB;
305 afs_termState = AFSOP_STOP_RXEVENT;
307 afs_osi_Wakeup(&afs_termState);
318 * Make adjustment for the new size in the disk cache entry
320 * Major Assumptions Here:
321 * Assumes that frag size is an integral power of two, less one,
322 * and that this is a two's complement machine. I don't
323 * know of any filesystems which violate this assumption...
326 * adc : Ptr to dcache entry.
327 * anewsize : New size desired.
330 void afs_AdjustSize(register struct dcache *adc, register afs_int32 newSize)
332 register afs_int32 oldSize;
334 AFS_STATCNT(afs_AdjustSize);
336 adc->dflags |= DFEntryMod;
337 oldSize = ((adc->f.chunkBytes + afs_fsfragsize)^afs_fsfragsize)>>10;/* round up */
338 adc->f.chunkBytes = newSize;
339 newSize = ((newSize + afs_fsfragsize)^afs_fsfragsize)>>10;/* round up */
340 if (newSize > oldSize) {
341 /* We're growing the file, wakeup the daemon */
342 afs_MaybeWakeupTruncateDaemon();
344 afs_blocksUsed += (newSize - oldSize);
345 afs_stats_cmperf.cacheBlocksInUse = afs_blocksUsed; /* XXX */
353 * This routine is responsible for moving at least one entry (but up
354 * to some number of them) from the LRU queue to the free queue.
357 * anumber : Number of entries that should ideally be moved.
358 * aneedSpace : How much space we need (1K blocks);
361 * The anumber parameter is just a hint; at least one entry MUST be
362 * moved, or we'll panic. We must be called with afs_xdcache
363 * write-locked. We should try to satisfy both anumber and aneedspace,
364 * whichever is more demanding - need to do several things:
365 * 1. only grab up to anumber victims if aneedSpace <= 0, not
366 * the whole set of MAXATONCE.
367 * 2. dynamically choose MAXATONCE to reflect severity of
368 * demand: something like (*aneedSpace >> (logChunk - 9))
369 * N.B. if we're called with aneedSpace <= 0 and anumber > 0, that
370 * indicates that the cache is not properly configured/tuned or
371 * something. We should be able to automatically correct that problem.
374 #define MAXATONCE 16 /* max we can obtain at once */
375 static void afs_GetDownD(int anumber, int *aneedSpace)
379 struct VenusFid *afid;
383 register struct vcache *tvc;
384 afs_uint32 victims[MAXATONCE];
385 struct dcache *victimDCs[MAXATONCE];
386 afs_hyper_t victimTimes[MAXATONCE];/* youngest (largest LRU time) first */
387 afs_uint32 victimPtr; /* next free item in victim arrays */
388 afs_hyper_t maxVictimTime; /* youngest (largest LRU time) victim */
389 afs_uint32 maxVictimPtr; /* where it is */
392 AFS_STATCNT(afs_GetDownD);
393 if (CheckLock(&afs_xdcache) != -1)
394 osi_Panic("getdownd nolock");
395 /* decrement anumber first for all dudes in free list */
396 /* SHOULD always decrement anumber first, even if aneedSpace >0,
397 * because we should try to free space even if anumber <=0 */
398 if (!aneedSpace || *aneedSpace <= 0) {
399 anumber -= afs_freeDCCount;
400 if (anumber <= 0) return; /* enough already free */
402 /* bounds check parameter */
403 if (anumber > MAXATONCE)
404 anumber = MAXATONCE; /* all we can do */
407 * The phase variable manages reclaims. Set to 0, the first pass,
408 * we don't reclaim active entries. Set to 1, we reclaim even active
412 for (i = 0; i < afs_cacheFiles; i++)
413 /* turn off all flags */
414 afs_indexFlags[i] &= ~IFFlag;
416 while (anumber > 0 || (aneedSpace && *aneedSpace >0)) {
417 /* find oldest entries for reclamation */
418 maxVictimPtr = victimPtr = 0;
419 hzero(maxVictimTime);
420 /* select victims from access time array */
421 for (i = 0; i < afs_cacheFiles; i++) {
422 if (afs_indexFlags[i] & (IFDataMod | IFFree | IFDiscarded)) {
423 /* skip if dirty or already free */
426 tdc = afs_indexTable[i];
427 if (tdc && (tdc->refCount != 0)) {
428 /* Referenced; can't use it! */
431 hset(vtime, afs_indexTimes[i]);
433 /* if we've already looked at this one, skip it */
434 if (afs_indexFlags[i] & IFFlag) continue;
436 if (victimPtr < MAXATONCE) {
437 /* if there's at least one free victim slot left */
438 victims[victimPtr] = i;
439 hset(victimTimes[victimPtr], vtime);
440 if (hcmp(vtime, maxVictimTime) > 0) {
441 hset(maxVictimTime, vtime);
442 maxVictimPtr = victimPtr;
446 else if (hcmp(vtime, maxVictimTime) < 0) {
448 * We're older than youngest victim, so we replace at
451 /* find youngest (largest LRU) victim */
453 if (j == victimPtr) osi_Panic("getdownd local");
455 hset(victimTimes[j], vtime);
456 /* recompute maxVictimTime */
457 hset(maxVictimTime, vtime);
458 for(j = 0; j < victimPtr; j++)
459 if (hcmp(maxVictimTime, victimTimes[j]) < 0) {
460 hset(maxVictimTime, victimTimes[j]);
466 /* now really reclaim the victims */
467 j = 0; /* flag to track if we actually got any of the victims */
468 /* first, hold all the victims, since we're going to release the lock
469 * during the truncate operation.
471 for(i=0; i < victimPtr; i++) {
472 tdc = afs_GetDSlot(victims[i], 0);
473 /* We got tdc->tlock(R) here */
474 if (tdc->refCount == 1)
478 ReleaseReadLock(&tdc->tlock);
479 if (!victimDCs[i]) afs_PutDCache(tdc);
481 for(i = 0; i < victimPtr; i++) {
482 /* q is first elt in dcache entry */
484 /* now, since we're dropping the afs_xdcache lock below, we
485 * have to verify, before proceeding, that there are no other
486 * references to this dcache entry, even now. Note that we
487 * compare with 1, since we bumped it above when we called
488 * afs_GetDSlot to preserve the entry's identity.
490 if (tdc && tdc->refCount == 1) {
491 unsigned char chunkFlags;
492 afs_size_t tchunkoffset;
494 /* xdcache is lower than the xvcache lock */
495 MReleaseWriteLock(&afs_xdcache);
496 MObtainReadLock(&afs_xvcache);
497 tvc = afs_FindVCache(afid, 0, 0 /* no stats, no vlru */ );
498 MReleaseReadLock(&afs_xvcache);
499 MObtainWriteLock(&afs_xdcache, 527);
501 if (tdc->refCount > 1) skip = 1;
503 tchunkoffset = AFS_CHUNKTOBASE(tdc->f.chunk);
504 chunkFlags = afs_indexFlags[tdc->index];
505 if (phase == 0 && osi_Active(tvc)) skip = 1;
506 if (phase > 0 && osi_Active(tvc) && (tvc->states & CDCLock)
507 && (chunkFlags & IFAnyPages)) skip = 1;
508 if (chunkFlags & IFDataMod) skip = 1;
509 afs_Trace4(afs_iclSetp, CM_TRACE_GETDOWND,
510 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, skip,
511 ICL_TYPE_INT32, tdc->index,
513 ICL_HANDLE_OFFSET(tchunkoffset));
515 #if defined(AFS_SUN5_ENV)
517 * Now we try to invalidate pages. We do this only for
518 * Solaris. For other platforms, it's OK to recycle a
519 * dcache entry out from under a page, because the strategy
520 * function can call afs_GetDCache().
522 if (!skip && (chunkFlags & IFAnyPages)) {
525 MReleaseWriteLock(&afs_xdcache);
526 MObtainWriteLock(&tvc->vlock, 543);
527 if (tvc->multiPage) {
531 /* block locking pages */
532 tvc->vstates |= VPageCleaning;
533 /* block getting new pages */
535 MReleaseWriteLock(&tvc->vlock);
536 /* One last recheck */
537 MObtainWriteLock(&afs_xdcache, 333);
538 chunkFlags = afs_indexFlags[tdc->index];
539 if (tdc->refCount > 1
540 || (chunkFlags & IFDataMod)
541 || (osi_Active(tvc) && (tvc->states & CDCLock)
542 && (chunkFlags & IFAnyPages))) {
544 MReleaseWriteLock(&afs_xdcache);
547 MReleaseWriteLock(&afs_xdcache);
549 code = osi_VM_GetDownD(tvc, tdc);
551 MObtainWriteLock(&afs_xdcache,269);
552 /* we actually removed all pages, clean and dirty */
554 afs_indexFlags[tdc->index] &= ~(IFDirtyPages| IFAnyPages);
557 MReleaseWriteLock(&afs_xdcache);
559 MObtainWriteLock(&tvc->vlock, 544);
560 if (--tvc->activeV == 0 && (tvc->vstates & VRevokeWait)) {
561 tvc->vstates &= ~VRevokeWait;
562 afs_osi_Wakeup((char *)&tvc->vstates);
565 if (tvc->vstates & VPageCleaning) {
566 tvc->vstates &= ~VPageCleaning;
567 afs_osi_Wakeup((char *)&tvc->vstates);
570 MReleaseWriteLock(&tvc->vlock);
572 #endif /* AFS_SUN5_ENV */
574 MReleaseWriteLock(&afs_xdcache);
578 MObtainWriteLock(&afs_xdcache, 528);
579 if (afs_indexFlags[tdc->index] &
580 (IFDataMod | IFDirtyPages | IFAnyPages)) skip = 1;
581 if (tdc->refCount > 1) skip = 1;
583 #if defined(AFS_SUN5_ENV)
585 /* no vnode, so IFDirtyPages is spurious (we don't
586 * sweep dcaches on vnode recycling, so we can have
587 * DIRTYPAGES set even when all pages are gone). Just
589 * Hold vcache lock to prevent vnode from being
590 * created while we're clearing IFDirtyPages.
592 afs_indexFlags[tdc->index] &= ~(IFDirtyPages | IFAnyPages);
596 /* skip this guy and mark him as recently used */
597 afs_indexFlags[tdc->index] |= IFFlag;
598 afs_Trace4(afs_iclSetp, CM_TRACE_GETDOWND,
599 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, 2,
600 ICL_TYPE_INT32, tdc->index,
602 ICL_HANDLE_OFFSET(tchunkoffset));
605 /* flush this dude from the data cache and reclaim;
606 * first, make sure no one will care that we damage
607 * it, by removing it from all hash tables. Then,
608 * melt it down for parts. Note that any concurrent
609 * (new possibility!) calls to GetDownD won't touch
610 * this guy because his reference count is > 0. */
611 afs_Trace4(afs_iclSetp, CM_TRACE_GETDOWND,
612 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, 3,
613 ICL_TYPE_INT32, tdc->index,
615 ICL_HANDLE_OFFSET(tchunkoffset));
617 AFS_STATCNT(afs_gget);
619 afs_HashOutDCache(tdc);
620 if (tdc->f.chunkBytes != 0) {
623 *aneedSpace -= (tdc->f.chunkBytes + afs_fsfragsize) >> 10;
628 afs_DiscardDCache(tdc);
633 j = 1; /* we reclaimed at least one victim */
640 /* Phase is 0 and no one was found, so try phase 1 (ignore
641 * osi_Active flag) */
644 for (i = 0; i < afs_cacheFiles; i++)
645 /* turn off all flags */
646 afs_indexFlags[i] &= ~IFFlag;
650 /* found no one in phase 1, we're hosed */
651 if (victimPtr == 0) break;
653 } /* big while loop */
660 * Description: remove adc from any hash tables that would allow it to be located
661 * again by afs_FindDCache or afs_GetDCache.
663 * Parameters: adc -- pointer to dcache entry to remove from hash tables.
665 * Locks: Must have the afs_xdcache lock write-locked to call this function.
667 int afs_HashOutDCache(struct dcache *adc)
672 AFS_STATCNT(afs_glink);
674 /* we know this guy's in the LRUQ. We'll move dude into DCQ below */
676 /* if this guy is in the hash table, pull him out */
677 if (adc->f.fid.Fid.Volume != 0) {
678 /* remove entry from first hash chains */
679 i = DCHash(&adc->f.fid, adc->f.chunk);
680 us = afs_dchashTbl[i];
681 if (us == adc->index) {
682 /* first dude in the list */
683 afs_dchashTbl[i] = afs_dcnextTbl[adc->index];
686 /* somewhere on the chain */
687 while (us != NULLIDX) {
688 if (afs_dcnextTbl[us] == adc->index) {
689 /* found item pointing at the one to delete */
690 afs_dcnextTbl[us] = afs_dcnextTbl[adc->index];
693 us = afs_dcnextTbl[us];
695 if (us == NULLIDX) osi_Panic("dcache hc");
697 /* remove entry from *other* hash chain */
698 i = DVHash(&adc->f.fid);
699 us = afs_dvhashTbl[i];
700 if (us == adc->index) {
701 /* first dude in the list */
702 afs_dvhashTbl[i] = afs_dvnextTbl[adc->index];
705 /* somewhere on the chain */
706 while (us != NULLIDX) {
707 if (afs_dvnextTbl[us] == adc->index) {
708 /* found item pointing at the one to delete */
709 afs_dvnextTbl[us] = afs_dvnextTbl[adc->index];
712 us = afs_dvnextTbl[us];
714 if (us == NULLIDX) osi_Panic("dcache hv");
718 /* prevent entry from being found on a reboot (it is already out of
719 * the hash table, but after a crash, we just look at fid fields of
720 * stable (old) entries).
722 adc->f.fid.Fid.Volume = 0; /* invalid */
724 /* mark entry as modified */
725 adc->dflags |= DFEntryMod;
729 } /*afs_HashOutDCache */
736 * Flush the given dcache entry, pulling it from hash chains
737 * and truncating the associated cache file.
740 * adc: Ptr to dcache entry to flush.
743 * This routine must be called with the afs_xdcache lock held
747 void afs_FlushDCache(register struct dcache *adc)
749 AFS_STATCNT(afs_FlushDCache);
751 * Bump the number of cache files flushed.
753 afs_stats_cmperf.cacheFlushes++;
755 /* remove from all hash tables */
756 afs_HashOutDCache(adc);
758 /* Free its space; special case null operation, since truncate operation
759 * in UFS is slow even in this case, and this allows us to pre-truncate
760 * these files at more convenient times with fewer locks set
761 * (see afs_GetDownD).
763 if (adc->f.chunkBytes != 0) {
764 afs_DiscardDCache(adc);
765 afs_MaybeWakeupTruncateDaemon();
770 if (afs_WaitForCacheDrain) {
771 if (afs_blocksUsed <=
772 (CM_CACHESIZEDRAINEDPCT*afs_cacheBlocks)/100) {
773 afs_WaitForCacheDrain = 0;
774 afs_osi_Wakeup(&afs_WaitForCacheDrain);
777 } /*afs_FlushDCache*/
783 * Description: put a dcache entry on the free dcache entry list.
785 * Parameters: adc -- dcache entry to free
787 * Environment: called with afs_xdcache lock write-locked.
789 static void afs_FreeDCache(register struct dcache *adc)
791 /* Thread on free list, update free list count and mark entry as
792 * freed in its indexFlags element. Also, ensure DCache entry gets
793 * written out (set DFEntryMod).
796 afs_dvnextTbl[adc->index] = afs_freeDCList;
797 afs_freeDCList = adc->index;
799 afs_indexFlags[adc->index] |= IFFree;
800 adc->dflags |= DFEntryMod;
802 if (afs_WaitForCacheDrain) {
803 if ((afs_blocksUsed - afs_blocksDiscarded) <=
804 (CM_CACHESIZEDRAINEDPCT*afs_cacheBlocks)/100) {
805 afs_WaitForCacheDrain = 0;
806 afs_osi_Wakeup(&afs_WaitForCacheDrain);
815 * Discard the cache element by moving it to the discardDCList.
816 * This puts the cache element into a quasi-freed state, where
817 * the space may be reused, but the file has not been truncated.
819 * Major Assumptions Here:
820 * Assumes that frag size is an integral power of two, less one,
821 * and that this is a two's complement machine. I don't
822 * know of any filesystems which violate this assumption...
825 * adc : Ptr to dcache entry.
828 * Must be called with afs_xdcache write-locked.
831 static void afs_DiscardDCache(register struct dcache *adc)
833 register afs_int32 size;
835 AFS_STATCNT(afs_DiscardDCache);
837 osi_Assert(adc->refCount == 1);
839 size = ((adc->f.chunkBytes + afs_fsfragsize)^afs_fsfragsize)>>10;/* round up */
840 afs_blocksDiscarded += size;
841 afs_stats_cmperf.cacheBlocksDiscarded = afs_blocksDiscarded;
843 afs_dvnextTbl[adc->index] = afs_discardDCList;
844 afs_discardDCList = adc->index;
845 afs_discardDCCount++;
847 adc->f.fid.Fid.Volume = 0;
848 adc->dflags |= DFEntryMod;
849 afs_indexFlags[adc->index] |= IFDiscarded;
851 if (afs_WaitForCacheDrain) {
852 if ((afs_blocksUsed - afs_blocksDiscarded) <=
853 (CM_CACHESIZEDRAINEDPCT*afs_cacheBlocks)/100) {
854 afs_WaitForCacheDrain = 0;
855 afs_osi_Wakeup(&afs_WaitForCacheDrain);
859 } /*afs_DiscardDCache*/
862 * afs_FreeDiscardedDCache
865 * Free the next element on the list of discarded cache elements.
867 static void afs_FreeDiscardedDCache(void)
869 register struct dcache *tdc;
870 register struct osi_file *tfile;
871 register afs_int32 size;
873 AFS_STATCNT(afs_FreeDiscardedDCache);
875 MObtainWriteLock(&afs_xdcache,510);
876 if (!afs_blocksDiscarded) {
877 MReleaseWriteLock(&afs_xdcache);
882 * Get an entry from the list of discarded cache elements
884 tdc = afs_GetDSlot(afs_discardDCList, 0);
885 osi_Assert(tdc->refCount == 1);
886 ReleaseReadLock(&tdc->tlock);
888 afs_discardDCList = afs_dvnextTbl[tdc->index];
889 afs_dvnextTbl[tdc->index] = NULLIDX;
890 afs_discardDCCount--;
891 size = ((tdc->f.chunkBytes + afs_fsfragsize)^afs_fsfragsize)>>10;/* round up */
892 afs_blocksDiscarded -= size;
893 afs_stats_cmperf.cacheBlocksDiscarded = afs_blocksDiscarded;
894 /* We can lock because we just took it off the free list */
895 ObtainWriteLock(&tdc->lock, 626);
896 MReleaseWriteLock(&afs_xdcache);
899 * Truncate the element to reclaim its space
901 tfile = afs_CFileOpen(tdc->f.inode);
902 afs_CFileTruncate(tfile, 0);
903 afs_CFileClose(tfile);
904 afs_AdjustSize(tdc, 0);
907 * Free the element we just truncated
909 MObtainWriteLock(&afs_xdcache,511);
910 afs_indexFlags[tdc->index] &= ~IFDiscarded;
912 ReleaseWriteLock(&tdc->lock);
914 MReleaseWriteLock(&afs_xdcache);
918 * afs_MaybeFreeDiscardedDCache
921 * Free as many entries from the list of discarded cache elements
922 * as we need to get the free space down below CM_WAITFORDRAINPCT (98%).
927 int afs_MaybeFreeDiscardedDCache(void)
930 AFS_STATCNT(afs_MaybeFreeDiscardedDCache);
932 while (afs_blocksDiscarded &&
933 (afs_blocksUsed > (CM_WAITFORDRAINPCT*afs_cacheBlocks)/100)) {
934 afs_FreeDiscardedDCache();
943 * Try to free up a certain number of disk slots.
946 * anumber : Targeted number of disk slots to free up.
949 * Must be called with afs_xdcache write-locked.
951 static void afs_GetDownDSlot(int anumber)
953 struct afs_q *tq, *nq;
958 AFS_STATCNT(afs_GetDownDSlot);
959 if (cacheDiskType == AFS_FCACHE_TYPE_MEM)
960 osi_Panic("diskless getdowndslot");
962 if (CheckLock(&afs_xdcache) != -1)
963 osi_Panic("getdowndslot nolock");
965 /* decrement anumber first for all dudes in free list */
966 for(tdc = afs_freeDSList; tdc; tdc = (struct dcache *)tdc->lruq.next)
969 return; /* enough already free */
971 for(cnt=0, tq = afs_DLRU.prev; tq != &afs_DLRU && anumber > 0;
973 tdc = (struct dcache *) tq; /* q is first elt in dcache entry */
974 nq = QPrev(tq); /* in case we remove it */
975 if (tdc->refCount == 0) {
976 if ((ix=tdc->index) == NULLIDX) osi_Panic("getdowndslot");
977 /* pull the entry out of the lruq and put it on the free list */
980 /* write-through if modified */
981 if (tdc->dflags & DFEntryMod) {
982 #if defined(AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
984 * ask proxy to do this for us - we don't have the stack space
986 while (tdc->dflags & DFEntryMod) {
989 s = SPLOCK(afs_sgibklock);
990 if (afs_sgibklist == NULL) {
991 /* if slot is free, grab it. */
993 SV_SIGNAL(&afs_sgibksync);
995 /* wait for daemon to (start, then) finish. */
996 SP_WAIT(afs_sgibklock, s, &afs_sgibkwait, PINOD);
1000 tdc->dflags &= ~DFEntryMod;
1001 afs_WriteDCache(tdc, 1);
1008 struct osi_file * f = (struct osi_file *)tdc->ihint;
1016 /* finally put the entry in the free list */
1017 afs_indexTable[ix] = NULL;
1018 afs_indexFlags[ix] &= ~IFEverUsed;
1019 tdc->index = NULLIDX;
1020 tdc->lruq.next = (struct afs_q *) afs_freeDSList;
1021 afs_freeDSList = tdc;
1025 } /*afs_GetDownDSlot*/
1032 * Increment the reference count on a disk cache entry,
1033 * which already has a non-zero refcount. In order to
1034 * increment the refcount of a zero-reference entry, you
1035 * have to hold afs_xdcache.
1038 * adc : Pointer to the dcache entry to increment.
1041 * Nothing interesting.
1043 int afs_RefDCache(struct dcache *adc)
1045 ObtainWriteLock(&adc->tlock, 627);
1046 if (adc->refCount < 0)
1047 osi_Panic("RefDCache: negative refcount");
1049 ReleaseWriteLock(&adc->tlock);
1058 * Decrement the reference count on a disk cache entry.
1061 * ad : Ptr to the dcache entry to decrement.
1064 * Nothing interesting.
1066 int afs_PutDCache(register struct dcache *adc)
1068 AFS_STATCNT(afs_PutDCache);
1069 ObtainWriteLock(&adc->tlock, 276);
1070 if (adc->refCount <= 0)
1071 osi_Panic("putdcache");
1073 ReleaseWriteLock(&adc->tlock);
1082 * Try to discard all data associated with this file from the
1086 * avc : Pointer to the cache info for the file.
1089 * Both pvnLock and lock are write held.
1091 void afs_TryToSmush(register struct vcache *avc, struct AFS_UCRED *acred,
1094 register struct dcache *tdc;
1097 AFS_STATCNT(afs_TryToSmush);
1098 afs_Trace2(afs_iclSetp, CM_TRACE_TRYTOSMUSH, ICL_TYPE_POINTER, avc,
1099 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->m.Length));
1100 sync = 1; /* XX Temp testing XX*/
1102 #if defined(AFS_SUN5_ENV)
1103 ObtainWriteLock(&avc->vlock, 573);
1104 avc->activeV++; /* block new getpages */
1105 ReleaseWriteLock(&avc->vlock);
1108 /* Flush VM pages */
1109 osi_VM_TryToSmush(avc, acred, sync);
1112 * Get the hash chain containing all dce's for this fid
1114 i = DVHash(&avc->fid);
1115 MObtainWriteLock(&afs_xdcache,277);
1116 for(index = afs_dvhashTbl[i]; index != NULLIDX; index=i) {
1117 i = afs_dvnextTbl[index]; /* next pointer this hash table */
1118 if (afs_indexUnique[index] == avc->fid.Fid.Unique) {
1119 int releaseTlock = 1;
1120 tdc = afs_GetDSlot(index, NULL);
1121 if (!FidCmp(&tdc->f.fid, &avc->fid)) {
1123 if ((afs_indexFlags[index] & IFDataMod) == 0 &&
1124 tdc->refCount == 1) {
1125 ReleaseReadLock(&tdc->tlock);
1127 afs_FlushDCache(tdc);
1130 afs_indexTable[index] = 0;
1132 if (releaseTlock) ReleaseReadLock(&tdc->tlock);
1136 #if defined(AFS_SUN5_ENV)
1137 ObtainWriteLock(&avc->vlock, 545);
1138 if (--avc->activeV == 0 && (avc->vstates & VRevokeWait)) {
1139 avc->vstates &= ~VRevokeWait;
1140 afs_osi_Wakeup((char *)&avc->vstates);
1142 ReleaseWriteLock(&avc->vlock);
1144 MReleaseWriteLock(&afs_xdcache);
1146 * It's treated like a callback so that when we do lookups we'll invalidate the unique bit if any
1147 * trytoSmush occured during the lookup call
1156 * Given the cached info for a file and a byte offset into the
1157 * file, make sure the dcache entry for that file and containing
1158 * the given byte is available, returning it to our caller.
1161 * avc : Pointer to the (held) vcache entry to look in.
1162 * abyte : Which byte we want to get to.
1165 * Pointer to the dcache entry covering the file & desired byte,
1166 * or NULL if not found.
1169 * The vcache entry is held upon entry.
1172 struct dcache *afs_FindDCache(register struct vcache *avc, afs_size_t abyte)
1175 register afs_int32 i, index;
1176 register struct dcache *tdc;
1178 AFS_STATCNT(afs_FindDCache);
1179 chunk = AFS_CHUNK(abyte);
1182 * Hash on the [fid, chunk] and get the corresponding dcache index
1183 * after write-locking the dcache.
1185 i = DCHash(&avc->fid, chunk);
1186 MObtainWriteLock(&afs_xdcache,278);
1187 for(index = afs_dchashTbl[i]; index != NULLIDX;) {
1188 if (afs_indexUnique[index] == avc->fid.Fid.Unique) {
1189 tdc = afs_GetDSlot(index, NULL);
1190 ReleaseReadLock(&tdc->tlock);
1191 if (!FidCmp(&tdc->f.fid, &avc->fid) && chunk == tdc->f.chunk) {
1192 break; /* leaving refCount high for caller */
1196 index = afs_dcnextTbl[index];
1198 MReleaseWriteLock(&afs_xdcache);
1199 if (index != NULLIDX) {
1200 hset(afs_indexTimes[tdc->index], afs_indexCounter);
1201 hadd32(afs_indexCounter, 1);
1207 } /*afs_FindDCache*/
1211 * afs_UFSCacheStoreProc
1214 * Called upon store.
1217 * acall : Ptr to the Rx call structure involved.
1218 * afile : Ptr to the related file descriptor.
1219 * alen : Size of the file in bytes.
1220 * avc : Ptr to the vcache entry.
1221 * shouldWake : is it "safe" to return early from close() ?
1222 * abytesToXferP : Set to the number of bytes to xfer.
1223 * NOTE: This parameter is only used if AFS_NOSTATS
1225 * abytesXferredP : Set to the number of bytes actually xferred.
1226 * NOTE: This parameter is only used if AFS_NOSTATS
1230 * Nothing interesting.
1232 static int afs_UFSCacheStoreProc(register struct rx_call *acall,
1233 struct osi_file *afile, register afs_int32 alen, struct vcache *avc,
1234 int *shouldWake, afs_size_t *abytesToXferP, afs_size_t *abytesXferredP)
1236 afs_int32 code, got;
1237 register char *tbuffer;
1240 AFS_STATCNT(UFS_CacheStoreProc);
1244 * In this case, alen is *always* the amount of data we'll be trying
1247 (*abytesToXferP) = alen;
1248 (*abytesXferredP) = 0;
1249 #endif /* AFS_NOSTATS */
1251 afs_Trace4(afs_iclSetp, CM_TRACE_STOREPROC, ICL_TYPE_POINTER, avc,
1252 ICL_TYPE_FID, &(avc->fid),
1253 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->m.Length),
1254 ICL_TYPE_INT32, alen);
1255 tbuffer = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
1257 tlen = (alen > AFS_LRALLOCSIZ ? AFS_LRALLOCSIZ : alen);
1258 got = afs_osi_Read(afile, -1, tbuffer, tlen);
1260 #if defined(KERNEL_HAVE_UERROR)
1261 || (got != tlen && getuerror())
1264 osi_FreeLargeSpace(tbuffer);
1267 afs_Trace2(afs_iclSetp, CM_TRACE_STOREPROC2,
1268 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(*tbuffer),
1269 ICL_TYPE_INT32, got);
1271 code = rx_Write(acall, tbuffer, got); /* writing 0 bytes will
1272 * push a short packet. Is that really what we want, just because the
1273 * data didn't come back from the disk yet? Let's try it and see. */
1276 (*abytesXferredP) += code;
1277 #endif /* AFS_NOSTATS */
1279 osi_FreeLargeSpace(tbuffer);
1284 * If file has been locked on server, we can allow the store
1287 if (shouldWake && *shouldWake && (rx_GetRemoteStatus(acall) & 1)) {
1288 *shouldWake = 0; /* only do this once */
1292 afs_Trace4(afs_iclSetp, CM_TRACE_STOREPROC, ICL_TYPE_POINTER, avc,
1293 ICL_TYPE_FID, &(avc->fid),
1294 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->m.Length),
1295 ICL_TYPE_INT32, alen);
1296 osi_FreeLargeSpace(tbuffer);
1299 } /* afs_UFSCacheStoreProc*/
1303 * afs_UFSCacheFetchProc
1306 * Routine called on fetch; also tells people waiting for data
1307 * that more has arrived.
1310 * acall : Ptr to the Rx call structure.
1311 * afile : File descriptor for the cache file.
1312 * abase : Base offset to fetch.
1313 * adc : Ptr to the dcache entry for the file, write-locked.
1314 * avc : Ptr to the vcache entry for the file.
1315 * abytesToXferP : Set to the number of bytes to xfer.
1316 * NOTE: This parameter is only used if AFS_NOSTATS
1318 * abytesXferredP : Set to the number of bytes actually xferred.
1319 * NOTE: This parameter is only used if AFS_NOSTATS
1323 * Nothing interesting.
1326 static int afs_UFSCacheFetchProc(register struct rx_call *acall,
1327 struct osi_file *afile, afs_size_t abase, struct dcache *adc,
1328 struct vcache *avc, afs_size_t *abytesToXferP,
1329 afs_size_t *abytesXferredP, afs_int32 lengthFound)
1332 register afs_int32 code;
1333 register char *tbuffer;
1337 AFS_STATCNT(UFS_CacheFetchProc);
1338 osi_Assert(WriteLocked(&adc->lock));
1339 afile->offset = 0; /* Each time start from the beginning */
1340 length = lengthFound;
1342 (*abytesToXferP) = 0;
1343 (*abytesXferredP) = 0;
1344 #endif /* AFS_NOSTATS */
1345 tbuffer = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
1346 adc->validPos = abase;
1350 code = rx_Read(acall, (char *)&length, sizeof(afs_int32));
1352 length = ntohl(length);
1353 if (code != sizeof(afs_int32)) {
1354 osi_FreeLargeSpace(tbuffer);
1355 code = rx_Error(acall);
1356 return (code?code:-1); /* try to return code, not -1 */
1360 * The fetch protocol is extended for the AFS/DFS translator
1361 * to allow multiple blocks of data, each with its own length,
1362 * to be returned. As long as the top bit is set, there are more
1365 * We do not do this for AFS file servers because they sometimes
1366 * return large negative numbers as the transfer size.
1368 if (avc->states & CForeign) {
1369 moredata = length & 0x80000000;
1370 length &= ~0x80000000;
1375 (*abytesToXferP) += length;
1376 #endif /* AFS_NOSTATS */
1377 while (length > 0) {
1378 tlen = (length > AFS_LRALLOCSIZ ? AFS_LRALLOCSIZ : length);
1379 #ifdef RX_KERNEL_TRACE
1380 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP,
1381 ICL_TYPE_STRING, "before rx_Read");
1384 code = rx_Read(acall, tbuffer, tlen);
1386 #ifdef RX_KERNEL_TRACE
1387 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP,
1388 ICL_TYPE_STRING, "after rx_Read");
1391 (*abytesXferredP) += code;
1392 #endif /* AFS_NOSTATS */
1394 osi_FreeLargeSpace(tbuffer);
1395 afs_Trace3(afs_iclSetp, CM_TRACE_FETCH64READ,
1396 ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, code,
1397 ICL_TYPE_INT32, length);
1400 code = afs_osi_Write(afile, -1, tbuffer, tlen);
1402 osi_FreeLargeSpace(tbuffer);
1407 adc->validPos = abase;
1408 if (afs_osi_Wakeup(&adc->validPos) == 0)
1409 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAKE,
1410 ICL_TYPE_STRING, __FILE__,
1411 ICL_TYPE_INT32, __LINE__,
1412 ICL_TYPE_POINTER, adc,
1413 ICL_TYPE_INT32, adc->dflags);
1416 osi_FreeLargeSpace(tbuffer);
1419 } /* afs_UFSCacheFetchProc*/
1425 * This function is called to obtain a reference to data stored in
1426 * the disk cache, locating a chunk of data containing the desired
1427 * byte and returning a reference to the disk cache entry, with its
1428 * reference count incremented.
1432 * avc : Ptr to a vcache entry (unlocked)
1433 * abyte : Byte position in the file desired
1434 * areq : Request structure identifying the requesting user.
1435 * aflags : Settings as follows:
1437 * 2 : Return after creating entry.
1438 * 4 : called from afs_vnop_write.c
1439 * *alen contains length of data to be written.
1441 * aoffset : Set to the offset within the chunk where the resident
1443 * alen : Set to the number of bytes of data after the desired
1444 * byte (including the byte itself) which can be read
1448 * The vcache entry pointed to by avc is unlocked upon entry.
1452 struct AFSVolSync tsync;
1453 struct AFSFetchStatus OutStatus;
1454 struct AFSCallBack CallBack;
1458 * Update the vnode-to-dcache hint if we can get the vnode lock
1459 * right away. Assumes dcache entry is at least read-locked.
1461 void updateV2DC(int lockVc, struct vcache *v, struct dcache *d, int src)
1463 if (!lockVc || 0 == NBObtainWriteLock(&v->lock,src)) {
1464 if (hsame(v->m.DataVersion, d->f.versionNo) && v->callback) {
1466 v->quick.stamp = d->stamp = MakeStamp();
1467 v->quick.minLoc = AFS_CHUNKTOBASE(d->f.chunk);
1468 /* Don't think I need these next two lines forever */
1469 v->quick.len = d->f.chunkBytes;
1472 if (lockVc) ReleaseWriteLock(&v->lock);
1476 /* avc - Write-locked unless aflags & 1 */
1477 struct dcache *afs_GetDCache(register struct vcache *avc, afs_size_t abyte,
1478 register struct vrequest *areq, afs_size_t *aoffset, afs_size_t *alen,
1481 register afs_int32 i, code, code1=0, shortcut;
1482 #if defined(AFS_AIX32_ENV) || defined(AFS_SGI_ENV)
1483 register afs_int32 adjustsize = 0;
1489 afs_size_t maxGoodLength; /* amount of good data at server */
1490 struct rx_call *tcall;
1491 afs_size_t Position = 0;
1492 #ifdef AFS_64BIT_CLIENT
1494 afs_size_t lengthFound; /* as returned from server */
1495 #endif /* AFS_64BIT_CLIENT */
1496 afs_int32 size, tlen; /* size of segment to transfer */
1497 struct tlocal1 *tsmall = 0;
1498 register struct dcache *tdc;
1499 register struct osi_file *file;
1500 register struct conn *tc;
1502 struct server *newCallback;
1503 char setNewCallback;
1504 char setVcacheStatus;
1505 char doVcacheUpdate;
1507 int doAdjustSize = 0;
1508 int doReallyAdjustSize = 0;
1509 int overWriteWholeChunk = 0;
1513 struct afs_stats_xferData *xferP; /* Ptr to this op's xfer struct */
1514 osi_timeval_t xferStartTime, /*FS xfer start time*/
1515 xferStopTime; /*FS xfer stop time*/
1516 afs_size_t bytesToXfer; /* # bytes to xfer*/
1517 afs_size_t bytesXferred; /* # bytes actually xferred*/
1518 struct afs_stats_AccessInfo *accP; /*Ptr to access record in stats*/
1519 int fromReplica; /*Are we reading from a replica?*/
1520 int numFetchLoops; /*# times around the fetch/analyze loop*/
1521 #endif /* AFS_NOSTATS */
1523 AFS_STATCNT(afs_GetDCache);
1528 setLocks = aflags & 1;
1531 * Determine the chunk number and offset within the chunk corresponding
1532 * to the desired byte.
1534 if (avc->fid.Fid.Vnode & 1) { /* if (vType(avc) == VDIR) */
1538 chunk = AFS_CHUNK(abyte);
1541 /* come back to here if we waited for the cache to drain. */
1544 setNewCallback = setVcacheStatus = 0;
1548 ObtainWriteLock(&avc->lock, 616);
1550 ObtainReadLock(&avc->lock);
1555 * avc->lock(R) if setLocks && !slowPass
1556 * avc->lock(W) if !setLocks || slowPass
1561 /* check hints first! (might could use bcmp or some such...) */
1562 if ((tdc = avc->h1.dchint)) {
1566 * The locking order between afs_xdcache and dcache lock matters.
1567 * The hint dcache entry could be anywhere, even on the free list.
1568 * Locking afs_xdcache ensures that noone is trying to pull dcache
1569 * entries from the free list, and thereby assuming them to be not
1570 * referenced and not locked.
1572 MObtainReadLock(&afs_xdcache);
1573 dcLocked = (0 == NBObtainSharedLock(&tdc->lock, 601));
1576 (tdc->index != NULLIDX) && !FidCmp(&tdc->f.fid, &avc->fid) &&
1577 chunk == tdc->f.chunk &&
1578 !(afs_indexFlags[tdc->index] & (IFFree|IFDiscarded))) {
1579 /* got the right one. It might not be the right version, and it
1580 * might be fetching, but it's the right dcache entry.
1582 /* All this code should be integrated better with what follows:
1583 * I can save a good bit more time under a write lock if I do..
1585 ObtainWriteLock(&tdc->tlock, 603);
1587 ReleaseWriteLock(&tdc->tlock);
1589 MReleaseReadLock(&afs_xdcache);
1592 if (hsame(tdc->f.versionNo, avc->m.DataVersion) &&
1593 !(tdc->dflags & DFFetching)) {
1595 afs_stats_cmperf.dcacheHits++;
1596 MObtainWriteLock(&afs_xdcache, 559);
1597 QRemove(&tdc->lruq);
1598 QAdd(&afs_DLRU, &tdc->lruq);
1599 MReleaseWriteLock(&afs_xdcache);
1602 * avc->lock(R) if setLocks && !slowPass
1603 * avc->lock(W) if !setLocks || slowPass
1609 if (dcLocked) ReleaseSharedLock(&tdc->lock);
1610 MReleaseReadLock(&afs_xdcache);
1618 * avc->lock(R) if setLocks && !slowPass
1619 * avc->lock(W) if !setLocks || slowPass
1620 * tdc->lock(S) if tdc
1623 if (!tdc) { /* If the hint wasn't the right dcache entry */
1625 * Hash on the [fid, chunk] and get the corresponding dcache index
1626 * after write-locking the dcache.
1631 * avc->lock(R) if setLocks && !slowPass
1632 * avc->lock(W) if !setLocks || slowPass
1635 i = DCHash(&avc->fid, chunk);
1636 /* check to make sure our space is fine */
1637 afs_MaybeWakeupTruncateDaemon();
1639 MObtainWriteLock(&afs_xdcache,280);
1641 for (index = afs_dchashTbl[i]; index != NULLIDX; ) {
1642 if (afs_indexUnique[index] == avc->fid.Fid.Unique) {
1643 tdc = afs_GetDSlot(index, NULL);
1644 ReleaseReadLock(&tdc->tlock);
1647 * avc->lock(R) if setLocks && !slowPass
1648 * avc->lock(W) if !setLocks || slowPass
1651 if (!FidCmp(&tdc->f.fid, &avc->fid) && chunk == tdc->f.chunk) {
1652 /* Move it up in the beginning of the list */
1653 if (afs_dchashTbl[i] != index) {
1654 afs_dcnextTbl[us] = afs_dcnextTbl[index];
1655 afs_dcnextTbl[index] = afs_dchashTbl[i];
1656 afs_dchashTbl[i] = index;
1658 MReleaseWriteLock(&afs_xdcache);
1659 ObtainSharedLock(&tdc->lock, 606);
1660 break; /* leaving refCount high for caller */
1666 index = afs_dcnextTbl[index];
1670 * If we didn't find the entry, we'll create one.
1672 if (index == NULLIDX) {
1675 * avc->lock(R) if setLocks
1676 * avc->lock(W) if !setLocks
1679 afs_Trace2(afs_iclSetp, CM_TRACE_GETDCACHE1, ICL_TYPE_POINTER,
1680 avc, ICL_TYPE_INT32, chunk);
1682 /* Make sure there is a free dcache entry for us to use */
1683 if (afs_discardDCList == NULLIDX && afs_freeDCList == NULLIDX) {
1685 if (!setLocks) avc->states |= CDCLock;
1686 afs_GetDownD(5, (int*)0); /* just need slots */
1687 if (!setLocks) avc->states &= ~CDCLock;
1688 if (afs_discardDCList != NULLIDX || afs_freeDCList != NULLIDX)
1690 /* If we can't get space for 5 mins we give up and panic */
1691 if (++downDCount > 300)
1692 osi_Panic("getdcache");
1693 MReleaseWriteLock(&afs_xdcache);
1696 * avc->lock(R) if setLocks
1697 * avc->lock(W) if !setLocks
1699 afs_osi_Wait(1000, 0, 0);
1704 if (afs_discardDCList == NULLIDX ||
1705 ((aflags & 2) && afs_freeDCList != NULLIDX)) {
1707 afs_indexFlags[afs_freeDCList] &= ~IFFree;
1708 tdc = afs_GetDSlot(afs_freeDCList, 0);
1709 osi_Assert(tdc->refCount == 1);
1710 ReleaseReadLock(&tdc->tlock);
1711 ObtainWriteLock(&tdc->lock, 604);
1712 afs_freeDCList = afs_dvnextTbl[tdc->index];
1715 afs_indexFlags[afs_discardDCList] &= ~IFDiscarded;
1716 tdc = afs_GetDSlot(afs_discardDCList, 0);
1717 osi_Assert(tdc->refCount == 1);
1718 ReleaseReadLock(&tdc->tlock);
1719 ObtainWriteLock(&tdc->lock, 605);
1720 afs_discardDCList = afs_dvnextTbl[tdc->index];
1721 afs_discardDCCount--;
1722 size = ((tdc->f.chunkBytes + afs_fsfragsize)^afs_fsfragsize)>>10;
1723 afs_blocksDiscarded -= size;
1724 afs_stats_cmperf.cacheBlocksDiscarded = afs_blocksDiscarded;
1726 /* Truncate the chunk so zeroes get filled properly */
1727 file = afs_CFileOpen(tdc->f.inode);
1728 afs_CFileTruncate(file, 0);
1729 afs_CFileClose(file);
1730 afs_AdjustSize(tdc, 0);
1736 * avc->lock(R) if setLocks
1737 * avc->lock(W) if !setLocks
1743 * Fill in the newly-allocated dcache record.
1745 afs_indexFlags[tdc->index] &= ~(IFDirtyPages | IFAnyPages);
1746 tdc->f.fid = avc->fid;
1747 afs_indexUnique[tdc->index] = tdc->f.fid.Fid.Unique;
1748 hones(tdc->f.versionNo); /* invalid value */
1749 tdc->f.chunk = chunk;
1750 tdc->validPos = AFS_CHUNKTOBASE(chunk);
1752 if (tdc->lruq.prev == &tdc->lruq) osi_Panic("lruq 1");
1755 * Now add to the two hash chains - note that i is still set
1756 * from the above DCHash call.
1758 afs_dcnextTbl[tdc->index] = afs_dchashTbl[i];
1759 afs_dchashTbl[i] = tdc->index;
1760 i = DVHash(&avc->fid);
1761 afs_dvnextTbl[tdc->index] = afs_dvhashTbl[i];
1762 afs_dvhashTbl[i] = tdc->index;
1763 tdc->dflags = DFEntryMod;
1766 afs_MaybeWakeupTruncateDaemon();
1767 MReleaseWriteLock(&afs_xdcache);
1768 ConvertWToSLock(&tdc->lock);
1770 } /* vcache->dcache hint failed */
1774 * avc->lock(R) if setLocks && !slowPass
1775 * avc->lock(W) if !setLocks || slowPass
1779 afs_Trace4(afs_iclSetp, CM_TRACE_GETDCACHE2, ICL_TYPE_POINTER, avc,
1780 ICL_TYPE_POINTER, tdc,
1781 ICL_TYPE_INT32, hgetlo(tdc->f.versionNo),
1782 ICL_TYPE_INT32, hgetlo(avc->m.DataVersion));
1784 * Here we have the entry in tdc, with its refCount incremented.
1785 * Note: we don't use the S-lock on avc; it costs concurrency when
1786 * storing a file back to the server.
1790 * Not a newly created file so we need to check the file's length and
1791 * compare data versions since someone could have changed the data or we're
1792 * reading a file written elsewhere. We only want to bypass doing no-op
1793 * read rpcs on newly created files (dv of 0) since only then we guarantee
1794 * that this chunk's data hasn't been filled by another client.
1796 size = AFS_CHUNKSIZE(abyte);
1797 if (aflags & 4) /* called from write */
1799 else /* called from read */
1800 tlen = tdc->validPos - abyte;
1801 Position = AFS_CHUNKTOBASE(chunk);
1802 afs_Trace4(afs_iclSetp, CM_TRACE_GETDCACHE3,
1803 ICL_TYPE_INT32, tlen,
1804 ICL_TYPE_INT32, aflags,
1805 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(abyte),
1806 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(Position));
1807 if ((aflags & 4) && (hiszero(avc->m.DataVersion)))
1809 if ((aflags & 4) && (abyte == Position) && (tlen >= size))
1810 overWriteWholeChunk = 1;
1811 if (doAdjustSize || overWriteWholeChunk) {
1812 #if defined(AFS_AIX32_ENV) || defined(AFS_SGI_ENV)
1814 #ifdef AFS_SGI64_ENV
1815 if (doAdjustSize) adjustsize = NBPP;
1816 #else /* AFS_SGI64_ENV */
1817 if (doAdjustSize) adjustsize = 8192;
1818 #endif /* AFS_SGI64_ENV */
1819 #else /* AFS_SGI_ENV */
1820 if (doAdjustSize) adjustsize = 4096;
1821 #endif /* AFS_SGI_ENV */
1822 if (AFS_CHUNKTOBASE(chunk)+adjustsize >= avc->m.Length &&
1823 #else /* defined(AFS_AIX32_ENV) || defined(AFS_SGI_ENV) */
1824 #if defined(AFS_SUN_ENV) || defined(AFS_OSF_ENV)
1825 if ((doAdjustSize || (AFS_CHUNKTOBASE(chunk) >= avc->m.Length)) &&
1827 if (AFS_CHUNKTOBASE(chunk) >= avc->m.Length &&
1829 #endif /* defined(AFS_AIX32_ENV) || defined(AFS_SGI_ENV) */
1830 !hsame(avc->m.DataVersion, tdc->f.versionNo))
1831 doReallyAdjustSize = 1;
1833 if (doReallyAdjustSize || overWriteWholeChunk) {
1834 /* no data in file to read at this position */
1835 UpgradeSToWLock(&tdc->lock, 607);
1837 file = afs_CFileOpen(tdc->f.inode);
1838 afs_CFileTruncate(file, 0);
1839 afs_CFileClose(file);
1840 afs_AdjustSize(tdc, 0);
1841 hset(tdc->f.versionNo, avc->m.DataVersion);
1842 tdc->dflags |= DFEntryMod;
1844 ConvertWToSLock(&tdc->lock);
1849 * We must read in the whole chunk if the version number doesn't
1853 /* don't need data, just a unique dcache entry */
1854 ObtainWriteLock(&afs_xdcache, 608);
1855 hset(afs_indexTimes[tdc->index], afs_indexCounter);
1856 hadd32(afs_indexCounter, 1);
1857 ReleaseWriteLock(&afs_xdcache);
1859 updateV2DC(setLocks, avc, tdc, 553);
1860 if (vType(avc) == VDIR)
1863 *aoffset = AFS_CHUNKOFFSET(abyte);
1864 if (tdc->validPos < abyte)
1865 *alen = (afs_size_t) 0;
1867 *alen = tdc->validPos - abyte;
1868 ReleaseSharedLock(&tdc->lock);
1871 ReleaseWriteLock(&avc->lock);
1873 ReleaseReadLock(&avc->lock);
1875 return tdc; /* check if we're done */
1880 * avc->lock(R) if setLocks && !slowPass
1881 * avc->lock(W) if !setLocks || slowPass
1884 osi_Assert((setLocks && !slowPass) || WriteLocked(&avc->lock));
1886 setNewCallback = setVcacheStatus = 0;
1890 * avc->lock(R) if setLocks && !slowPass
1891 * avc->lock(W) if !setLocks || slowPass
1894 if (!hsame(avc->m.DataVersion, tdc->f.versionNo) && !overWriteWholeChunk) {
1896 * Version number mismatch.
1898 UpgradeSToWLock(&tdc->lock, 609);
1901 * If data ever existed for this vnode, and this is a text object,
1902 * do some clearing. Now, you'd think you need only do the flush
1903 * when VTEXT is on, but VTEXT is turned off when the text object
1904 * is freed, while pages are left lying around in memory marked
1905 * with this vnode. If we would reactivate (create a new text
1906 * object from) this vnode, we could easily stumble upon some of
1907 * these old pages in pagein. So, we always flush these guys.
1908 * Sun has a wonderful lack of useful invariants in this system.
1910 * avc->flushDV is the data version # of the file at the last text
1911 * flush. Clearly, at least, we don't have to flush the file more
1912 * often than it changes
1914 if (hcmp(avc->flushDV, avc->m.DataVersion) < 0) {
1916 * By here, the cache entry is always write-locked. We can
1917 * deadlock if we call osi_Flush with the cache entry locked...
1918 * Unlock the dcache too.
1920 ReleaseWriteLock(&tdc->lock);
1921 if (setLocks && !slowPass)
1922 ReleaseReadLock(&avc->lock);
1924 ReleaseWriteLock(&avc->lock);
1928 * Call osi_FlushPages in open, read/write, and map, since it
1929 * is too hard here to figure out if we should lock the
1932 if (setLocks && !slowPass)
1933 ObtainReadLock(&avc->lock);
1935 ObtainWriteLock(&avc->lock, 66);
1936 ObtainWriteLock(&tdc->lock, 610);
1941 * avc->lock(R) if setLocks && !slowPass
1942 * avc->lock(W) if !setLocks || slowPass
1946 /* Watch for standard race condition around osi_FlushText */
1947 if (hsame(avc->m.DataVersion, tdc->f.versionNo)) {
1948 updateV2DC(setLocks, avc, tdc, 569); /* set hint */
1949 afs_stats_cmperf.dcacheHits++;
1950 ConvertWToSLock(&tdc->lock);
1954 /* Sleep here when cache needs to be drained. */
1955 if (setLocks && !slowPass &&
1956 (afs_blocksUsed > (CM_WAITFORDRAINPCT*afs_cacheBlocks)/100)) {
1957 /* Make sure truncate daemon is running */
1958 afs_MaybeWakeupTruncateDaemon();
1959 ObtainWriteLock(&tdc->tlock, 614);
1960 tdc->refCount--; /* we'll re-obtain the dcache when we re-try. */
1961 ReleaseWriteLock(&tdc->tlock);
1962 ReleaseWriteLock(&tdc->lock);
1963 ReleaseReadLock(&avc->lock);
1964 while ((afs_blocksUsed-afs_blocksDiscarded) >
1965 (CM_WAITFORDRAINPCT*afs_cacheBlocks)/100) {
1966 afs_WaitForCacheDrain = 1;
1967 afs_osi_Sleep(&afs_WaitForCacheDrain);
1969 afs_MaybeFreeDiscardedDCache();
1970 /* need to check if someone else got the chunk first. */
1971 goto RetryGetDCache;
1974 /* Do not fetch data beyond truncPos. */
1975 maxGoodLength = avc->m.Length;
1976 if (avc->truncPos < maxGoodLength) maxGoodLength = avc->truncPos;
1977 Position = AFS_CHUNKBASE(abyte);
1978 if (vType(avc) == VDIR) {
1979 size = avc->m.Length;
1980 if (size > tdc->f.chunkBytes) {
1981 /* pre-reserve space for file */
1982 afs_AdjustSize(tdc, size);
1984 size = 999999999; /* max size for transfer */
1987 size = AFS_CHUNKSIZE(abyte); /* expected max size */
1988 /* don't read past end of good data on server */
1989 if (Position + size > maxGoodLength)
1990 size = maxGoodLength - Position;
1991 if (size < 0) size = 0; /* Handle random races */
1992 if (size > tdc->f.chunkBytes) {
1993 /* pre-reserve space for file */
1994 afs_AdjustSize(tdc, size); /* changes chunkBytes */
1995 /* max size for transfer still in size */
1998 if (afs_mariner && !tdc->f.chunk)
1999 afs_MarinerLog("fetch$Fetching", avc); /* , Position, size, afs_indexCounter );*/
2001 * Right now, we only have one tool, and it's a hammer. So, we
2002 * fetch the whole file.
2004 DZap(&tdc->f.inode); /* pages in cache may be old */
2006 if (file = tdc->ihint) {
2007 if (tdc->f.inode == file->inum )
2014 file = osi_UFSOpen(tdc->f.inode);
2019 file = afs_CFileOpen(tdc->f.inode);
2020 afs_RemoveVCB(&avc->fid);
2021 tdc->f.states |= DWriting;
2022 tdc->dflags |= DFFetching;
2023 tdc->validPos = Position; /* which is AFS_CHUNKBASE(abyte) */
2024 if (tdc->mflags & DFFetchReq) {
2025 tdc->mflags &= ~DFFetchReq;
2026 if (afs_osi_Wakeup(&tdc->validPos) == 0)
2027 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAKE,
2028 ICL_TYPE_STRING, __FILE__,
2029 ICL_TYPE_INT32, __LINE__,
2030 ICL_TYPE_POINTER, tdc,
2031 ICL_TYPE_INT32, tdc->dflags);
2033 tsmall = (struct tlocal1 *) osi_AllocLargeSpace(sizeof(struct tlocal1));
2034 setVcacheStatus = 0;
2037 * Remember if we are doing the reading from a replicated volume,
2038 * and how many times we've zipped around the fetch/analyze loop.
2040 fromReplica = (avc->states & CRO) ? 1 : 0;
2042 accP = &(afs_stats_cmfullperf.accessinf);
2044 (accP->replicatedRefs)++;
2046 (accP->unreplicatedRefs)++;
2047 #endif /* AFS_NOSTATS */
2048 /* this is a cache miss */
2049 afs_Trace4(afs_iclSetp, CM_TRACE_FETCHPROC, ICL_TYPE_POINTER, avc,
2050 ICL_TYPE_FID, &(avc->fid),
2051 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(Position),
2052 ICL_TYPE_INT32, size);
2054 if (size) afs_stats_cmperf.dcacheMisses++;
2057 * Dynamic root support: fetch data from local memory.
2059 if (afs_IsDynroot(avc)) {
2063 afs_GetDynroot(&dynrootDir, &dynrootLen, &tsmall->OutStatus);
2065 dynrootDir += Position;
2066 dynrootLen -= Position;
2067 if (size > dynrootLen)
2069 if (size < 0) size = 0;
2070 code = afs_CFileWrite(file, 0, dynrootDir, size);
2078 tdc->validPos = Position + size;
2079 afs_CFileTruncate(file, size); /* prune it */
2082 * Not a dynamic vnode: do the real fetch.
2087 * avc->lock(R) if setLocks && !slowPass
2088 * avc->lock(W) if !setLocks || slowPass
2092 tc = afs_Conn(&avc->fid, areq, SHARED_LOCK);
2094 afs_int32 length_hi, length, bytes;
2098 (accP->numReplicasAccessed)++;
2100 #endif /* AFS_NOSTATS */
2101 if (!setLocks || slowPass) {
2102 avc->callback = tc->srvr->server;
2104 newCallback = tc->srvr->server;
2109 tcall = rx_NewCall(tc->id);
2112 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHDATA);
2113 #ifdef AFS_64BIT_CLIENT
2114 length_hi = code = 0;
2115 if (!afs_serverHasNo64Bit(tc)) {
2118 code = StartRXAFS_FetchData64(tcall,
2119 (struct AFSFid *) &avc->fid.Fid,
2123 afs_Trace2(afs_iclSetp, CM_TRACE_FETCH64CODE,
2124 ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, code);
2126 bytes = rx_Read(tcall, (char *)&length_hi, sizeof(afs_int32));
2128 if (bytes == sizeof(afs_int32)) {
2129 length_hi = ntohl(length_hi);
2132 code = rx_Error(tcall);
2134 code1 = rx_EndCall(tcall, code);
2136 tcall = (struct rx_call *) 0;
2140 if (code == RXGEN_OPCODE || afs_serverHasNo64Bit(tc)) {
2141 if (Position > 0x7FFFFFFF) {
2148 tcall = rx_NewCall(tc->id);
2149 code = StartRXAFS_FetchData(tcall,
2150 (struct AFSFid *) &avc->fid.Fid, pos, size);
2153 afs_serverSetNo64Bit(tc);
2157 bytes = rx_Read(tcall, (char *)&length, sizeof(afs_int32));
2159 if (bytes == sizeof(afs_int32)) {
2160 length = ntohl(length);
2162 code = rx_Error(tcall);
2165 FillInt64(lengthFound, length_hi, length);
2166 afs_Trace3(afs_iclSetp, CM_TRACE_FETCH64LENG,
2167 ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, code,
2168 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(lengthFound));
2169 #else /* AFS_64BIT_CLIENT */
2171 code = StartRXAFS_FetchData(tcall,
2172 (struct AFSFid *) &avc->fid.Fid,
2177 bytes = rx_Read(tcall, (char *)&length, sizeof(afs_int32));
2179 if (bytes == sizeof(afs_int32)) {
2180 length = ntohl(length);
2182 code = rx_Error(tcall);
2185 #endif /* AFS_64BIT_CLIENT */
2189 xferP = &(afs_stats_cmfullperf.rpc.fsXferTimes[AFS_STATS_FS_XFERIDX_FETCHDATA]);
2190 osi_GetuTime(&xferStartTime);
2192 code = afs_CacheFetchProc(tcall, file,
2193 (afs_size_t) Position, tdc, avc,
2194 &bytesToXfer, &bytesXferred, length);
2196 osi_GetuTime(&xferStopTime);
2197 (xferP->numXfers)++;
2199 (xferP->numSuccesses)++;
2200 afs_stats_XferSumBytes[AFS_STATS_FS_XFERIDX_FETCHDATA] += bytesXferred;
2201 (xferP->sumBytes) += (afs_stats_XferSumBytes[AFS_STATS_FS_XFERIDX_FETCHDATA] >> 10);
2202 afs_stats_XferSumBytes[AFS_STATS_FS_XFERIDX_FETCHDATA] &= 0x3FF;
2203 if (bytesXferred < xferP->minBytes)
2204 xferP->minBytes = bytesXferred;
2205 if (bytesXferred > xferP->maxBytes)
2206 xferP->maxBytes = bytesXferred;
2209 * Tally the size of the object. Note: we tally the actual size,
2210 * NOT the number of bytes that made it out over the wire.
2212 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET0)
2213 (xferP->count[0])++;
2215 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET1)
2216 (xferP->count[1])++;
2218 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET2)
2219 (xferP->count[2])++;
2221 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET3)
2222 (xferP->count[3])++;
2224 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET4)
2225 (xferP->count[4])++;
2227 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET5)
2228 (xferP->count[5])++;
2230 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET6)
2231 (xferP->count[6])++;
2233 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET7)
2234 (xferP->count[7])++;
2236 (xferP->count[8])++;
2238 afs_stats_GetDiff(elapsedTime, xferStartTime, xferStopTime);
2239 afs_stats_AddTo((xferP->sumTime), elapsedTime);
2240 afs_stats_SquareAddTo((xferP->sqrTime), elapsedTime);
2241 if (afs_stats_TimeLessThan(elapsedTime, (xferP->minTime))) {
2242 afs_stats_TimeAssign((xferP->minTime), elapsedTime);
2244 if (afs_stats_TimeGreaterThan(elapsedTime, (xferP->maxTime))) {
2245 afs_stats_TimeAssign((xferP->maxTime), elapsedTime);
2249 code = afs_CacheFetchProc(tcall, file, Position, tdc, avc, 0, 0, length);
2250 #endif /* AFS_NOSTATS */
2254 code = EndRXAFS_FetchData(tcall,
2263 code1 = rx_EndCall(tcall, code);
2269 if ( !code && code1 )
2273 /* callback could have been broken (or expired) in a race here,
2274 * but we return the data anyway. It's as good as we knew about
2275 * when we started. */
2277 * validPos is updated by CacheFetchProc, and can only be
2278 * modifed under a dcache write lock, which we've blocked out
2280 size = tdc->validPos - Position; /* actual segment size */
2281 if (size < 0) size = 0;
2282 afs_CFileTruncate(file, size); /* prune it */
2285 if (!setLocks || slowPass) {
2286 ObtainWriteLock(&afs_xcbhash, 453);
2287 afs_DequeueCallback(avc);
2288 avc->states &= ~(CStatd | CUnique);
2289 avc->callback = NULL;
2290 ReleaseWriteLock(&afs_xcbhash);
2291 if (avc->fid.Fid.Vnode & 1 || (vType(avc) == VDIR))
2292 osi_dnlc_purgedp(avc);
2294 /* Something lost. Forget about performance, and go
2295 * back with a vcache write lock.
2297 afs_CFileTruncate(file, 0);
2298 afs_AdjustSize(tdc, 0);
2299 afs_CFileClose(file);
2300 osi_FreeLargeSpace(tsmall);
2302 ReleaseWriteLock(&tdc->lock);
2305 ReleaseReadLock(&avc->lock);
2307 goto RetryGetDCache;
2312 (afs_Analyze(tc, code, &avc->fid, areq,
2313 AFS_STATS_FS_RPCIDX_FETCHDATA,
2314 SHARED_LOCK, NULL));
2318 * avc->lock(R) if setLocks && !slowPass
2319 * avc->lock(W) if !setLocks || slowPass
2325 * In the case of replicated access, jot down info on the number of
2326 * attempts it took before we got through or gave up.
2329 if (numFetchLoops <= 1)
2330 (accP->refFirstReplicaOK)++;
2331 if (numFetchLoops > accP->maxReplicasPerRef)
2332 accP->maxReplicasPerRef = numFetchLoops;
2334 #endif /* AFS_NOSTATS */
2336 tdc->dflags &= ~DFFetching;
2337 if (afs_osi_Wakeup(&tdc->validPos) == 0)
2338 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAKE,
2339 ICL_TYPE_STRING, __FILE__,
2340 ICL_TYPE_INT32, __LINE__,
2341 ICL_TYPE_POINTER, tdc,
2342 ICL_TYPE_INT32, tdc->dflags);
2343 if (avc->execsOrWriters == 0) tdc->f.states &= ~DWriting;
2345 /* now, if code != 0, we have an error and should punt.
2346 * note that we have the vcache write lock, either because
2347 * !setLocks or slowPass.
2350 afs_CFileTruncate(file, 0);
2351 afs_AdjustSize(tdc, 0);
2352 afs_CFileClose(file);
2353 ZapDCE(tdc); /* sets DFEntryMod */
2354 if (vType(avc) == VDIR) {
2355 DZap(&tdc->f.inode);
2357 ReleaseWriteLock(&tdc->lock);
2359 ObtainWriteLock(&afs_xcbhash, 454);
2360 afs_DequeueCallback(avc);
2361 avc->states &= ~( CStatd | CUnique );
2362 ReleaseWriteLock(&afs_xcbhash);
2363 if (avc->fid.Fid.Vnode & 1 || (vType(avc) == VDIR))
2364 osi_dnlc_purgedp(avc);
2367 * avc->lock(W); assert(!setLocks || slowPass)
2369 osi_Assert(!setLocks || slowPass);
2374 /* otherwise we copy in the just-fetched info */
2375 afs_CFileClose(file);
2376 afs_AdjustSize(tdc, size); /* new size */
2378 * Copy appropriate fields into vcache. Status is
2379 * copied later where we selectively acquire the
2380 * vcache write lock.
2383 afs_ProcessFS(avc, &tsmall->OutStatus, areq);
2385 setVcacheStatus = 1;
2386 hset64(tdc->f.versionNo, tsmall->OutStatus.dataVersionHigh, tsmall->OutStatus.DataVersion);
2387 tdc->dflags |= DFEntryMod;
2388 afs_indexFlags[tdc->index] |= IFEverUsed;
2389 ConvertWToSLock(&tdc->lock);
2390 } /*Data version numbers don't match*/
2393 * Data version numbers match.
2395 afs_stats_cmperf.dcacheHits++;
2396 } /*Data version numbers match*/
2398 updateV2DC(setLocks, avc, tdc, 335); /* set hint */
2402 * avc->lock(R) if setLocks && !slowPass
2403 * avc->lock(W) if !setLocks || slowPass
2404 * tdc->lock(S) if tdc
2408 * See if this was a reference to a file in the local cell.
2410 if (afs_IsPrimaryCellNum(avc->fid.Cell))
2411 afs_stats_cmperf.dlocalAccesses++;
2413 afs_stats_cmperf.dremoteAccesses++;
2415 /* Fix up LRU info */
2418 MObtainWriteLock(&afs_xdcache, 602);
2419 hset(afs_indexTimes[tdc->index], afs_indexCounter);
2420 hadd32(afs_indexCounter, 1);
2421 MReleaseWriteLock(&afs_xdcache);
2423 /* return the data */
2424 if (vType(avc) == VDIR)
2427 *aoffset = AFS_CHUNKOFFSET(abyte);
2428 *alen = (tdc->f.chunkBytes - *aoffset);
2429 ReleaseSharedLock(&tdc->lock);
2434 * avc->lock(R) if setLocks && !slowPass
2435 * avc->lock(W) if !setLocks || slowPass
2438 /* Fix up the callback and status values in the vcache */
2440 if (setLocks && !slowPass) {
2443 * This is our dirty little secret to parallel fetches.
2444 * We don't write-lock the vcache while doing the fetch,
2445 * but potentially we'll need to update the vcache after
2446 * the fetch is done.
2448 * Drop the read lock and try to re-obtain the write
2449 * lock. If the vcache still has the same DV, it's
2450 * ok to go ahead and install the new data.
2452 afs_hyper_t currentDV, statusDV;
2454 hset(currentDV, avc->m.DataVersion);
2456 if (setNewCallback && avc->callback != newCallback)
2460 hset64(statusDV, tsmall->OutStatus.dataVersionHigh,
2461 tsmall->OutStatus.DataVersion);
2463 if (setVcacheStatus && avc->m.Length != tsmall->OutStatus.Length)
2465 if (setVcacheStatus && !hsame(currentDV, statusDV))
2469 ReleaseReadLock(&avc->lock);
2471 if (doVcacheUpdate) {
2472 ObtainWriteLock(&avc->lock, 615);
2473 if (!hsame(avc->m.DataVersion, currentDV)) {
2474 /* We lose. Someone will beat us to it. */
2476 ReleaseWriteLock(&avc->lock);
2481 /* With slow pass, we've already done all the updates */
2483 ReleaseWriteLock(&avc->lock);
2486 /* Check if we need to perform any last-minute fixes with a write-lock */
2487 if (!setLocks || doVcacheUpdate) {
2488 if (setNewCallback) avc->callback = newCallback;
2489 if (tsmall && setVcacheStatus) afs_ProcessFS(avc, &tsmall->OutStatus, areq);
2490 if (setLocks) ReleaseWriteLock(&avc->lock);
2493 if (tsmall) osi_FreeLargeSpace(tsmall);
2500 * afs_WriteThroughDSlots
2503 * Sweep through the dcache slots and write out any modified
2504 * in-memory data back on to our caching store.
2510 * The afs_xdcache is write-locked through this whole affair.
2512 void afs_WriteThroughDSlots(void)
2514 register struct dcache *tdc;
2515 register afs_int32 i, touchedit=0;
2516 struct dcache **ents;
2517 int entmax, entcount;
2519 AFS_STATCNT(afs_WriteThroughDSlots);
2522 * Because of lock ordering, we can't grab dcache locks while
2523 * holding afs_xdcache. So we enter xdcache, get a reference
2524 * for every dcache entry, and exit xdcache.
2526 MObtainWriteLock(&afs_xdcache,283);
2527 entmax = afs_cacheFiles;
2528 ents = afs_osi_Alloc(entmax * sizeof(struct dcache *));
2530 for(i = 0; i < afs_cacheFiles; i++) {
2531 tdc = afs_indexTable[i];
2533 /* Grab tlock in case the existing refcount isn't zero */
2534 if (tdc && !(afs_indexFlags[i] & (IFFree | IFDiscarded))) {
2535 ObtainWriteLock(&tdc->tlock, 623);
2537 ReleaseWriteLock(&tdc->tlock);
2539 ents[entcount++] = tdc;
2542 MReleaseWriteLock(&afs_xdcache);
2545 * Now, for each dcache entry we found, check if it's dirty.
2546 * If so, get write-lock, get afs_xdcache, which protects
2547 * afs_cacheInodep, and flush it. Don't forget to put back
2550 for (i = 0; i < entcount; i++) {
2553 if (tdc->dflags & DFEntryMod) {
2556 wrLock = (0 == NBObtainWriteLock(&tdc->lock, 619));
2558 /* Now that we have the write lock, double-check */
2559 if (wrLock && (tdc->dflags & DFEntryMod)) {
2560 tdc->dflags &= ~DFEntryMod;
2561 MObtainWriteLock(&afs_xdcache, 620);
2562 afs_WriteDCache(tdc, 1);
2563 MReleaseWriteLock(&afs_xdcache);
2566 if (wrLock) ReleaseWriteLock(&tdc->lock);
2571 afs_osi_Free(ents, entmax * sizeof(struct dcache *));
2573 MObtainWriteLock(&afs_xdcache, 617);
2574 if (!touchedit && (cacheDiskType != AFS_FCACHE_TYPE_MEM)) {
2575 /* Touch the file to make sure that the mtime on the file is kept
2576 * up-to-date to avoid losing cached files on cold starts because
2577 * their mtime seems old...
2579 struct afs_fheader theader;
2581 theader.magic = AFS_FHMAGIC;
2582 theader.firstCSize = AFS_FIRSTCSIZE;
2583 theader.otherCSize = AFS_OTHERCSIZE;
2584 theader.version = AFS_CI_VERSION;
2585 afs_osi_Write(afs_cacheInodep, 0, &theader, sizeof(theader));
2587 MReleaseWriteLock(&afs_xdcache);
2594 * Return a pointer to an freshly initialized dcache entry using
2595 * a memory-based cache. The tlock will be read-locked.
2598 * aslot : Dcache slot to look at.
2599 * tmpdc : Ptr to dcache entry.
2602 * Must be called with afs_xdcache write-locked.
2605 struct dcache *afs_MemGetDSlot(register afs_int32 aslot, register struct dcache *tmpdc)
2607 register struct dcache *tdc;
2610 AFS_STATCNT(afs_MemGetDSlot);
2611 if (CheckLock(&afs_xdcache) != -1) osi_Panic("getdslot nolock");
2612 if (aslot < 0 || aslot >= afs_cacheFiles) osi_Panic("getdslot slot");
2613 tdc = afs_indexTable[aslot];
2615 QRemove(&tdc->lruq); /* move to queue head */
2616 QAdd(&afs_DLRU, &tdc->lruq);
2617 /* We're holding afs_xdcache, but get tlock in case refCount != 0 */
2618 ObtainWriteLock(&tdc->tlock, 624);
2620 ConvertWToRLock(&tdc->tlock);
2623 if (tmpdc == NULL) {
2624 if (!afs_freeDSList) afs_GetDownDSlot(4);
2625 if (!afs_freeDSList) {
2626 /* none free, making one is better than a panic */
2627 afs_stats_cmperf.dcacheXAllocs++; /* count in case we have a leak */
2628 tdc = (struct dcache *) afs_osi_Alloc(sizeof (struct dcache));
2629 #ifdef KERNEL_HAVE_PIN
2630 pin((char *)tdc, sizeof(struct dcache)); /* XXX */
2633 tdc = afs_freeDSList;
2634 afs_freeDSList = (struct dcache *) tdc->lruq.next;
2637 tdc->dflags = 0; /* up-to-date, not in free q */
2639 QAdd(&afs_DLRU, &tdc->lruq);
2640 if (tdc->lruq.prev == &tdc->lruq) osi_Panic("lruq 3");
2647 /* initialize entry */
2648 tdc->f.fid.Cell = 0;
2649 tdc->f.fid.Fid.Volume = 0;
2651 hones(tdc->f.versionNo);
2652 tdc->f.inode = aslot;
2653 tdc->dflags |= DFEntryMod;
2656 afs_indexUnique[aslot] = tdc->f.fid.Fid.Unique;
2659 osi_Assert(0 == NBObtainWriteLock(&tdc->lock, 674));
2660 osi_Assert(0 == NBObtainWriteLock(&tdc->mflock, 675));
2661 osi_Assert(0 == NBObtainWriteLock(&tdc->tlock, 676));
2664 RWLOCK_INIT(&tdc->lock, "dcache lock");
2665 RWLOCK_INIT(&tdc->tlock, "dcache tlock");
2666 RWLOCK_INIT(&tdc->mflock, "dcache flock");
2667 ObtainReadLock(&tdc->tlock);
2670 afs_indexTable[aslot] = tdc;
2673 } /*afs_MemGetDSlot*/
2675 unsigned int last_error = 0, lasterrtime = 0;
2681 * Return a pointer to an freshly initialized dcache entry using
2682 * a UFS-based disk cache. The dcache tlock will be read-locked.
2685 * aslot : Dcache slot to look at.
2686 * tmpdc : Ptr to dcache entry.
2689 * afs_xdcache lock write-locked.
2691 struct dcache *afs_UFSGetDSlot(register afs_int32 aslot, register struct dcache *tmpdc)
2693 register afs_int32 code;
2694 register struct dcache *tdc;
2698 AFS_STATCNT(afs_UFSGetDSlot);
2699 if (CheckLock(&afs_xdcache) != -1) osi_Panic("getdslot nolock");
2700 if (aslot < 0 || aslot >= afs_cacheFiles) osi_Panic("getdslot slot");
2701 tdc = afs_indexTable[aslot];
2703 QRemove(&tdc->lruq); /* move to queue head */
2704 QAdd(&afs_DLRU, &tdc->lruq);
2705 /* Grab tlock in case refCount != 0 */
2706 ObtainWriteLock(&tdc->tlock, 625);
2708 ConvertWToRLock(&tdc->tlock);
2711 /* otherwise we should read it in from the cache file */
2713 * If we weren't passed an in-memory region to place the file info,
2714 * we have to allocate one.
2716 if (tmpdc == NULL) {
2717 if (!afs_freeDSList) afs_GetDownDSlot(4);
2718 if (!afs_freeDSList) {
2719 /* none free, making one is better than a panic */
2720 afs_stats_cmperf.dcacheXAllocs++; /* count in case we have a leak */
2721 tdc = (struct dcache *) afs_osi_Alloc(sizeof (struct dcache));
2722 #ifdef KERNEL_HAVE_PIN
2723 pin((char *)tdc, sizeof(struct dcache)); /* XXX */
2726 tdc = afs_freeDSList;
2727 afs_freeDSList = (struct dcache *) tdc->lruq.next;
2730 tdc->dflags = 0; /* up-to-date, not in free q */
2732 QAdd(&afs_DLRU, &tdc->lruq);
2733 if (tdc->lruq.prev == &tdc->lruq) osi_Panic("lruq 3");
2742 * Seek to the aslot'th entry and read it in.
2744 code = afs_osi_Read(afs_cacheInodep, sizeof(struct fcache) * aslot +
2745 sizeof(struct afs_fheader),
2746 (char *)(&tdc->f), sizeof(struct fcache));
2748 if (code != sizeof(struct fcache))
2750 if (!afs_CellNumValid(tdc->f.fid.Cell))
2754 tdc->f.fid.Cell = 0;
2755 tdc->f.fid.Fid.Volume = 0;
2757 hones(tdc->f.versionNo);
2758 tdc->dflags |= DFEntryMod;
2759 #if defined(KERNEL_HAVE_UERROR)
2760 last_error = getuerror();
2762 lasterrtime = osi_Time();
2763 afs_indexUnique[aslot] = tdc->f.fid.Fid.Unique;
2769 osi_Assert(0 == NBObtainWriteLock(&tdc->lock, 674));
2770 osi_Assert(0 == NBObtainWriteLock(&tdc->mflock, 675));
2771 osi_Assert(0 == NBObtainWriteLock(&tdc->tlock, 676));
2774 RWLOCK_INIT(&tdc->lock, "dcache lock");
2775 RWLOCK_INIT(&tdc->tlock, "dcache tlock");
2776 RWLOCK_INIT(&tdc->mflock, "dcache flock");
2777 ObtainReadLock(&tdc->tlock);
2780 * If we didn't read into a temporary dcache region, update the
2781 * slot pointer table.
2784 afs_indexTable[aslot] = tdc;
2787 } /*afs_UFSGetDSlot*/
2795 * write a particular dcache entry back to its home in the
2799 * adc : Pointer to the dcache entry to write.
2800 * atime : If true, set the modtime on the file to the current time.
2803 * Must be called with the afs_xdcache lock at least read-locked,
2804 * and dcache entry at least read-locked.
2805 * The reference count is not changed.
2808 int afs_WriteDCache(register struct dcache *adc, int atime)
2810 register afs_int32 code;
2812 if (cacheDiskType == AFS_FCACHE_TYPE_MEM) return 0;
2813 AFS_STATCNT(afs_WriteDCache);
2815 adc->f.modTime = osi_Time();
2817 * Seek to the right dcache slot and write the in-memory image out to disk.
2819 afs_cellname_write();
2820 code = afs_osi_Write(afs_cacheInodep, sizeof(struct fcache) * adc->index +
2821 sizeof(struct afs_fheader),
2822 (char *)(&adc->f), sizeof(struct fcache));
2823 if (code != sizeof(struct fcache)) return EIO;
2833 * Wake up users of a particular file waiting for stores to take
2837 * avc : Ptr to related vcache entry.
2840 * Nothing interesting.
2843 int afs_wakeup(register struct vcache *avc)
2846 register struct brequest *tb;
2848 AFS_STATCNT(afs_wakeup);
2849 for (i = 0; i < NBRS; i++, tb++) {
2850 /* if request is valid and for this file, we've found it */
2851 if (tb->refCount > 0 && avc == tb->vnode) {
2854 * If CSafeStore is on, then we don't awaken the guy
2855 * waiting for the store until the whole store has finished.
2856 * Otherwise, we do it now. Note that if CSafeStore is on,
2857 * the BStore routine actually wakes up the user, instead
2859 * I think this is redundant now because this sort of thing
2860 * is already being handled by the higher-level code.
2862 if ((avc->states & CSafeStore) == 0) {
2864 tb->flags |= BUVALID;
2865 if (tb->flags & BUWAIT) {
2866 tb->flags &= ~BUWAIT;
2881 * Given a file name and inode, set up that file to be an
2882 * active member in the AFS cache. This also involves checking
2883 * the usability of its data.
2886 * afile : Name of the cache file to initialize.
2887 * ainode : Inode of the file.
2890 * This function is called only during initialization.
2893 int afs_InitCacheFile(char *afile, ino_t ainode)
2895 register afs_int32 code;
2896 #if defined(AFS_LINUX22_ENV)
2897 struct dentry *filevp;
2899 struct vnode *filevp;
2903 struct osi_file *tfile;
2904 struct osi_stat tstat;
2905 register struct dcache *tdc;
2907 AFS_STATCNT(afs_InitCacheFile);
2908 index = afs_stats_cmperf.cacheNumEntries;
2909 if (index >= afs_cacheFiles) return EINVAL;
2911 MObtainWriteLock(&afs_xdcache,282);
2912 tdc = afs_GetDSlot(index, NULL);
2913 ReleaseReadLock(&tdc->tlock);
2914 MReleaseWriteLock(&afs_xdcache);
2916 ObtainWriteLock(&tdc->lock, 621);
2917 MObtainWriteLock(&afs_xdcache, 622);
2919 code = gop_lookupname(afile,
2925 ReleaseWriteLock(&afs_xdcache);
2926 ReleaseWriteLock(&tdc->lock);
2931 * We have a VN_HOLD on filevp. Get the useful info out and
2932 * return. We make use of the fact that the cache is in the
2933 * UFS file system, and just record the inode number.
2935 #ifdef AFS_LINUX22_ENV
2936 tdc->f.inode = VTOI(filevp->d_inode)->i_number;
2939 tdc->f.inode = afs_vnodeToInumber(filevp);
2943 AFS_RELE((struct vnode *)filevp);
2945 #endif /* AFS_LINUX22_ENV */
2948 tdc->f.inode = ainode;
2951 if ((tdc->f.states & DWriting) ||
2952 tdc->f.fid.Fid.Volume == 0) fileIsBad = 1;
2953 tfile = osi_UFSOpen(tdc->f.inode);
2954 code = afs_osi_Stat(tfile, &tstat);
2955 if (code) osi_Panic("initcachefile stat");
2958 * If file size doesn't match the cache info file, it's probably bad.
2960 if (tdc->f.chunkBytes != tstat.size) fileIsBad = 1;
2961 tdc->f.chunkBytes = 0;
2964 * If file changed within T (120?) seconds of cache info file, it's
2965 * probably bad. In addition, if slot changed within last T seconds,
2966 * the cache info file may be incorrectly identified, and so slot
2969 if (cacheInfoModTime < tstat.mtime + 120) fileIsBad = 1;
2970 if (cacheInfoModTime < tdc->f.modTime + 120) fileIsBad = 1;
2971 /* In case write through is behind, make sure cache items entry is
2972 * at least as new as the chunk.
2974 if (tdc->f.modTime < tstat.mtime) fileIsBad = 1;
2976 tdc->f.fid.Fid.Volume = 0; /* not in the hash table */
2977 if (tstat.size != 0)
2978 osi_UFSTruncate(tfile, 0);
2979 /* put entry in free cache slot list */
2980 afs_dvnextTbl[tdc->index] = afs_freeDCList;
2981 afs_freeDCList = index;
2983 afs_indexFlags[index] |= IFFree;
2984 afs_indexUnique[index] = 0;
2988 * We must put this entry in the appropriate hash tables.
2989 * Note that i is still set from the above DCHash call
2991 code = DCHash(&tdc->f.fid, tdc->f.chunk);
2992 afs_dcnextTbl[tdc->index] = afs_dchashTbl[code];
2993 afs_dchashTbl[code] = tdc->index;
2994 code = DVHash(&tdc->f.fid);
2995 afs_dvnextTbl[tdc->index] = afs_dvhashTbl[code];
2996 afs_dvhashTbl[code] = tdc->index;
2997 afs_AdjustSize(tdc, tstat.size); /* adjust to new size */
2999 /* has nontrivial amt of data */
3000 afs_indexFlags[index] |= IFEverUsed;
3001 afs_stats_cmperf.cacheFilesReused++;
3003 * Initialize index times to file's mod times; init indexCounter
3006 hset32(afs_indexTimes[index], tstat.atime);
3007 if (hgetlo(afs_indexCounter) < tstat.atime) {
3008 hset32(afs_indexCounter, tstat.atime);
3010 afs_indexUnique[index] = tdc->f.fid.Fid.Unique;
3011 } /*File is not bad*/
3013 osi_UFSClose(tfile);
3014 tdc->f.states &= ~DWriting;
3015 tdc->dflags &= ~DFEntryMod;
3016 /* don't set f.modTime; we're just cleaning up */
3017 afs_WriteDCache(tdc, 0);
3018 ReleaseWriteLock(&afs_xdcache);
3019 ReleaseWriteLock(&tdc->lock);
3021 afs_stats_cmperf.cacheNumEntries++;
3026 /*Max # of struct dcache's resident at any time*/
3028 * If 'dchint' is enabled then in-memory dcache min is increased because of
3037 * Initialize dcache related variables.
3039 void afs_dcacheInit(int afiles, int ablocks, int aDentries, int achunk,
3042 register struct dcache *tdp;
3046 afs_freeDCList = NULLIDX;
3047 afs_discardDCList = NULLIDX;
3048 afs_freeDCCount = 0;
3049 afs_freeDSList = NULL;
3050 hzero(afs_indexCounter);
3052 LOCK_INIT(&afs_xdcache, "afs_xdcache");
3058 if (achunk < 0 || achunk > 30)
3059 achunk = 13; /* Use default */
3060 AFS_SETCHUNKSIZE(achunk);
3066 if(aflags & AFSCALL_INIT_MEMCACHE) {
3068 * Use a memory cache instead of a disk cache
3070 cacheDiskType = AFS_FCACHE_TYPE_MEM;
3071 afs_cacheType = &afs_MemCacheOps;
3072 afiles = (afiles < aDentries) ? afiles : aDentries; /* min */
3073 ablocks = afiles * (AFS_FIRSTCSIZE/1024);
3074 /* ablocks is reported in 1K blocks */
3075 code = afs_InitMemCache(afiles * AFS_FIRSTCSIZE, AFS_FIRSTCSIZE, aflags);
3077 printf("afsd: memory cache too large for available memory.\n");
3078 printf("afsd: AFS files cannot be accessed.\n\n");
3080 afiles = ablocks = 0;
3083 printf("Memory cache: Allocating %d dcache entries...", aDentries);
3085 cacheDiskType = AFS_FCACHE_TYPE_UFS;
3086 afs_cacheType = &afs_UfsCacheOps;
3089 if (aDentries > 512)
3090 afs_dhashsize = 2048;
3091 /* initialize hash tables */
3092 afs_dvhashTbl = (afs_int32 *) afs_osi_Alloc(afs_dhashsize * sizeof(afs_int32));
3093 afs_dchashTbl = (afs_int32 *) afs_osi_Alloc(afs_dhashsize * sizeof(afs_int32));
3094 for(i=0;i< afs_dhashsize;i++) {
3095 afs_dvhashTbl[i] = NULLIDX;
3096 afs_dchashTbl[i] = NULLIDX;
3098 afs_dvnextTbl = (afs_int32 *) afs_osi_Alloc(afiles * sizeof(afs_int32));
3099 afs_dcnextTbl = (afs_int32 *) afs_osi_Alloc(afiles * sizeof(afs_int32));
3100 for(i=0;i< afiles;i++) {
3101 afs_dvnextTbl[i] = NULLIDX;
3102 afs_dcnextTbl[i] = NULLIDX;
3105 /* Allocate and zero the pointer array to the dcache entries */
3106 afs_indexTable = (struct dcache **)
3107 afs_osi_Alloc(sizeof(struct dcache *) * afiles);
3108 memset((char *)afs_indexTable, 0, sizeof(struct dcache *) * afiles);
3109 afs_indexTimes = (afs_hyper_t *) afs_osi_Alloc(afiles * sizeof(afs_hyper_t));
3110 memset((char *)afs_indexTimes, 0, afiles * sizeof(afs_hyper_t));
3111 afs_indexUnique = (afs_int32 *) afs_osi_Alloc(afiles * sizeof(afs_uint32));
3112 memset((char *)afs_indexUnique, 0, afiles * sizeof(afs_uint32));
3113 afs_indexFlags = (u_char *) afs_osi_Alloc(afiles * sizeof(u_char));
3114 memset((char *)afs_indexFlags, 0, afiles * sizeof(char));
3116 /* Allocate and thread the struct dcache entries themselves */
3117 tdp = afs_Initial_freeDSList =
3118 (struct dcache *) afs_osi_Alloc(aDentries * sizeof(struct dcache));
3119 memset((char *)tdp, 0, aDentries * sizeof(struct dcache));
3120 #ifdef KERNEL_HAVE_PIN
3121 pin((char *)afs_indexTable, sizeof(struct dcache *) * afiles);/* XXX */
3122 pin((char *)afs_indexTimes, sizeof(afs_hyper_t) * afiles); /* XXX */
3123 pin((char *)afs_indexFlags, sizeof(char) * afiles); /* XXX */
3124 pin((char *)afs_indexUnique, sizeof(afs_int32) * afiles); /* XXX */
3125 pin((char *)tdp, aDentries * sizeof(struct dcache)); /* XXX */
3126 pin((char *)afs_dvhashTbl, sizeof(afs_int32) * afs_dhashsize); /* XXX */
3127 pin((char *)afs_dchashTbl, sizeof(afs_int32) * afs_dhashsize); /* XXX */
3128 pin((char *)afs_dcnextTbl, sizeof(afs_int32) * afiles); /* XXX */
3129 pin((char *)afs_dvnextTbl, sizeof(afs_int32) * afiles); /* XXX */
3132 afs_freeDSList = &tdp[0];
3133 for(i=0; i < aDentries-1; i++) {
3134 tdp[i].lruq.next = (struct afs_q *) (&tdp[i+1]);
3136 tdp[aDentries-1].lruq.next = (struct afs_q *) 0;
3138 afs_stats_cmperf.cacheBlocksOrig = afs_stats_cmperf.cacheBlocksTotal = afs_cacheBlocks = ablocks;
3139 afs_ComputeCacheParms(); /* compute parms based on cache size */
3141 afs_dcentries = aDentries;
3150 void shutdown_dcache(void)
3154 afs_osi_Free(afs_dvnextTbl, afs_cacheFiles * sizeof(afs_int32));
3155 afs_osi_Free(afs_dcnextTbl, afs_cacheFiles * sizeof(afs_int32));
3156 afs_osi_Free(afs_indexTable, afs_cacheFiles * sizeof(struct dcache *));
3157 afs_osi_Free(afs_indexTimes, afs_cacheFiles * sizeof(afs_hyper_t));
3158 afs_osi_Free(afs_indexUnique, afs_cacheFiles * sizeof(afs_uint32));
3159 afs_osi_Free(afs_indexFlags, afs_cacheFiles * sizeof(u_char));
3160 afs_osi_Free(afs_Initial_freeDSList, afs_dcentries * sizeof(struct dcache));
3161 #ifdef KERNEL_HAVE_PIN
3162 unpin((char *)afs_dcnextTbl, afs_cacheFiles * sizeof(afs_int32));
3163 unpin((char *)afs_dvnextTbl, afs_cacheFiles * sizeof(afs_int32));
3164 unpin((char *)afs_indexTable, afs_cacheFiles * sizeof(struct dcache *));
3165 unpin((char *)afs_indexTimes, afs_cacheFiles * sizeof(afs_hyper_t));
3166 unpin((char *)afs_indexUnique, afs_cacheFiles * sizeof(afs_uint32));
3167 unpin((u_char *)afs_indexFlags, afs_cacheFiles * sizeof(u_char));
3168 unpin(afs_Initial_freeDSList, afs_dcentries * sizeof(struct dcache));
3172 for(i=0;i< afs_dhashsize;i++) {
3173 afs_dvhashTbl[i] = NULLIDX;
3174 afs_dchashTbl[i] = NULLIDX;
3178 afs_blocksUsed = afs_dcentries = 0;
3179 hzero(afs_indexCounter);
3181 afs_freeDCCount = 0;
3182 afs_freeDCList = NULLIDX;
3183 afs_discardDCList = NULLIDX;
3184 afs_freeDSList = afs_Initial_freeDSList = 0;
3186 LOCK_INIT(&afs_xdcache, "afs_xdcache");