/*
* Copyright 2000, International Business Machines Corporation and others.
- *$All Rights Reserved.
+ * All Rights Reserved.
*
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
/* Forward declarations. */
static void afs_GetDownD(int anumber, int *aneedSpace, afs_int32 buckethint);
-static void afs_FreeDiscardedDCache(void);
+static int afs_FreeDiscardedDCache(void);
static void afs_DiscardDCache(struct dcache *);
static void afs_FreeDCache(struct dcache *);
/* For split cache */
afs_osi_Read,
afs_osi_Write,
osi_UFSClose,
- afs_UFSRead,
- afs_UFSWrite,
+ afs_UFSReadUIO,
+ afs_UFSWriteUIO,
afs_UFSGetDSlot,
afs_UFSGetVolSlot,
afs_UFSHandleLink,
.fread = afs_osi_Read,
.fwrite = afs_osi_Write,
.close = osi_UFSClose,
- .vread = afs_UFSRead,
- .vwrite = afs_UFSWrite,
+ .vreadUIO = afs_UFSReadUIO,
+ .vwriteUIO = afs_UFSWriteUIO,
.GetDSlot = afs_UFSGetDSlot,
.GetVolSlot = afs_UFSGetVolSlot,
.HandleLink = afs_UFSHandleLink,
afs_MemReadBlk,
afs_MemWriteBlk,
afs_MemCacheClose,
- afs_MemRead,
- afs_MemWrite,
+ afs_MemReadUIO,
+ afs_MemWriteUIO,
afs_MemGetDSlot,
afs_MemGetVolSlot,
afs_MemHandleLink,
.fread = afs_MemReadBlk,
.fwrite = afs_MemWriteBlk,
.close = afs_MemCacheClose,
- .vread = afs_MemRead,
- .vwrite = afs_MemWrite,
+ .vreadUIO = afs_MemReadUIO,
+ .vwriteUIO = afs_MemWriteUIO,
.GetDSlot = afs_MemGetDSlot,
.GetVolSlot = afs_MemGetVolSlot,
.HandleLink = afs_MemHandleLink,
u_int afs_min_cache = 0;
/*!
+ * If there are waiters for the cache to drain, wake them if
+ * the number of free or discarded cache blocks reaches the
+ * CM_CACHESIZEDDRAINEDPCT limit.
+ *
+ * \note Environment:
+ * This routine must be called with the afs_xdcache lock held
+ * (in write mode).
+ */
+static void
+afs_WakeCacheWaitersIfDrained(void)
+{
+ if (afs_WaitForCacheDrain) {
+ if ((afs_blocksUsed - afs_blocksDiscarded) <=
+ PERCENT(CM_CACHESIZEDRAINEDPCT, afs_cacheBlocks)) {
+ afs_WaitForCacheDrain = 0;
+ afs_osi_Wakeup(&afs_WaitForCacheDrain);
+ }
+ }
+}
+
+/*!
* Keeps the cache clean and free by truncating uneeded files, when used.
* \param
* \return
while (1) {
cb_lowat = PERCENT((CM_DCACHESPACEFREEPCT - CM_DCACHEEXTRAPCT), afs_cacheBlocks);
ObtainWriteLock(&afs_xdcache, 266);
- if (afs_CacheTooFull) {
+ if (afs_CacheTooFull || afs_WaitForCacheDrain) {
int space_needed, slots_needed;
/* if we get woken up, we should try to clean something out */
for (counter = 0; counter < 10; counter++) {
space_needed =
afs_blocksUsed - afs_blocksDiscarded - cb_lowat;
+ if (space_needed < 0)
+ space_needed = 0;
slots_needed =
dc_hiwat - afs_freeDCCount - afs_discardDCCount;
- afs_GetDownD(slots_needed, &space_needed, 0);
+ if (slots_needed < 0)
+ slots_needed = 0;
+ if (slots_needed || space_needed)
+ afs_GetDownD(slots_needed, &space_needed, 0);
if ((space_needed <= 0) && (slots_needed <= 0)) {
break;
}
if (afs_termState == AFSOP_STOP_TRUNCDAEMON)
break;
}
- if (!afs_CacheIsTooFull())
+ if (!afs_CacheIsTooFull()) {
afs_CacheTooFull = 0;
+ afs_WakeCacheWaitersIfDrained();
+ }
} /* end of cache cleanup */
ReleaseWriteLock(&afs_xdcache);
*/
while (afs_blocksDiscarded && !afs_WaitForCacheDrain
&& (afs_termState != AFSOP_STOP_TRUNCDAEMON)) {
- afs_FreeDiscardedDCache();
+ int code = afs_FreeDiscardedDCache();
+ if (code) {
+ /* If we can't free any discarded dcache entries, that's okay.
+ * We're just doing this in the background; if someone needs
+ * discarded entries freed, they will try it themselves and/or
+ * signal us that the cache is too full. In any case, we'll
+ * try doing this again the next time we run through the loop.
+ */
+ break;
+ }
}
/* See if we need to continue to run. Someone may have
afs_uint32 maxVictimPtr; /* where it is */
int discard;
int curbucket;
-#if defined(AFS_FBSD80_ENV) && !defined(UKERNEL)
- int vfslocked;
-#endif
-
-#if defined(AFS_FBSD80_ENV) && !defined(UKERNEL)
- vfslocked = VFS_LOCK_GIANT(afs_globalVFS);
-#endif
AFS_STATCNT(afs_GetDownD);
if (!aneedSpace || *aneedSpace <= 0) {
anumber -= afs_freeDCCount;
if (anumber <= 0) {
-#if defined(AFS_FBSD80_ENV) && !defined(UKERNEL)
- VFS_UNLOCK_GIANT(vfslocked);
-#endif
return; /* enough already free */
}
}
* we don't reclaim active entries, or other than target bucket.
* Set to 1, we reclaim even active ones in target bucket.
* Set to 2, we reclaim any inactive one.
- * Set to 3, we reclaim even active ones.
+ * Set to 3, we reclaim even active ones. On Solaris, we also reclaim
+ * entries whose corresponding vcache has a nonempty multiPage list, when
+ * possible.
*/
if (splitdcache) {
phase = 0;
* during the truncate operation.
*/
for (i = 0; i < victimPtr; i++) {
- tdc = afs_GetDSlot(victims[i], 0);
+ tdc = afs_GetValidDSlot(victims[i]);
/* We got tdc->tlock(R) here */
- if (tdc->refCount == 1)
+ if (tdc && tdc->refCount == 1)
victimDCs[i] = tdc;
else
victimDCs[i] = 0;
- ReleaseReadLock(&tdc->tlock);
- if (!victimDCs[i])
- afs_PutDCache(tdc);
+ if (tdc) {
+ ReleaseReadLock(&tdc->tlock);
+ if (!victimDCs[i])
+ afs_PutDCache(tdc);
+ }
}
for (i = 0; i < victimPtr; i++) {
/* q is first elt in dcache entry */
* have to verify, before proceeding, that there are no other
* references to this dcache entry, even now. Note that we
* compare with 1, since we bumped it above when we called
- * afs_GetDSlot to preserve the entry's identity.
+ * afs_GetValidDSlot to preserve the entry's identity.
*/
if (tdc && tdc->refCount == 1) {
unsigned char chunkFlags;
ReleaseWriteLock(&afs_xdcache);
ObtainWriteLock(&tvc->vlock, 543);
- if (tvc->multiPage) {
- skip = 1;
- goto endmultipage;
+ if (!QEmpty(&tvc->multiPage)) {
+ if (phase < 3 || osi_VM_MultiPageConflict(tvc, tdc)) {
+ skip = 1;
+ goto endmultipage;
+ }
}
/* block locking pages */
tvc->vstates |= VPageCleaning;
j = 1; /* we reclaimed at least one victim */
}
}
- afs_PutDCache(tdc);
+ if (tdc)
+ afs_PutDCache(tdc);
} /* end of for victims loop */
if (phase < 5) {
}
} /* big while loop */
-#if defined(AFS_FBSD80_ENV) && !defined(UKERNEL)
- VFS_UNLOCK_GIANT(vfslocked);
-#endif
-
return;
} /*afs_GetDownD */
} else {
afs_FreeDCache(adc);
}
-
- if (afs_WaitForCacheDrain) {
- if (afs_blocksUsed <=
- PERCENT(CM_CACHESIZEDRAINEDPCT, afs_cacheBlocks)) {
- afs_WaitForCacheDrain = 0;
- afs_osi_Wakeup(&afs_WaitForCacheDrain);
- }
- }
} /*afs_FlushDCache */
afs_indexFlags[adc->index] |= IFFree;
adc->dflags |= DFEntryMod;
- if (afs_WaitForCacheDrain) {
- if ((afs_blocksUsed - afs_blocksDiscarded) <=
- PERCENT(CM_CACHESIZEDRAINEDPCT, afs_cacheBlocks)) {
- afs_WaitForCacheDrain = 0;
- afs_osi_Wakeup(&afs_WaitForCacheDrain);
- }
- }
+ afs_WakeCacheWaitersIfDrained();
} /* afs_FreeDCache */
/*!
adc->dflags |= DFEntryMod;
afs_indexFlags[adc->index] |= IFDiscarded;
- if (afs_WaitForCacheDrain) {
- if ((afs_blocksUsed - afs_blocksDiscarded) <=
- PERCENT(CM_CACHESIZEDRAINEDPCT, afs_cacheBlocks)) {
- afs_WaitForCacheDrain = 0;
- afs_osi_Wakeup(&afs_WaitForCacheDrain);
+ afs_WakeCacheWaitersIfDrained();
+} /*afs_DiscardDCache */
+
+/**
+ * Get a dcache entry from the discard or free list
+ *
+ * @param[in] indexp A pointer to the head of the dcache free list or discard
+ * list (afs_freeDCList, or afs_discardDCList)
+ *
+ * @return A dcache from that list, or NULL if none could be retrieved.
+ *
+ * @pre afs_xdcache is write-locked
+ */
+static struct dcache *
+afs_GetDSlotFromList(afs_int32 *indexp)
+{
+ struct dcache *tdc;
+
+ for ( ; *indexp != NULLIDX; indexp = &afs_dvnextTbl[*indexp]) {
+ tdc = afs_GetUnusedDSlot(*indexp);
+ if (tdc) {
+ osi_Assert(tdc->refCount == 1);
+ ReleaseReadLock(&tdc->tlock);
+ *indexp = afs_dvnextTbl[tdc->index];
+ afs_dvnextTbl[tdc->index] = NULLIDX;
+ return tdc;
}
}
-
-} /*afs_DiscardDCache */
+ return NULL;
+}
/*!
* Free the next element on the list of discarded cache elements.
+ *
+ * Returns -1 if we encountered an error preventing us from freeing a
+ * discarded dcache, or 0 on success.
*/
-static void
+static int
afs_FreeDiscardedDCache(void)
{
struct dcache *tdc;
ObtainWriteLock(&afs_xdcache, 510);
if (!afs_blocksDiscarded) {
ReleaseWriteLock(&afs_xdcache);
- return;
+ return 0;
}
/*
* Get an entry from the list of discarded cache elements
*/
- tdc = afs_GetDSlot(afs_discardDCList, 0);
- osi_Assert(tdc->refCount == 1);
- ReleaseReadLock(&tdc->tlock);
+ tdc = afs_GetDSlotFromList(&afs_discardDCList);
+ if (!tdc) {
+ ReleaseWriteLock(&afs_xdcache);
+ return -1;
+ }
- afs_discardDCList = afs_dvnextTbl[tdc->index];
- afs_dvnextTbl[tdc->index] = NULLIDX;
afs_discardDCCount--;
size = ((tdc->f.chunkBytes + afs_fsfragsize) ^ afs_fsfragsize) >> 10; /* round up */
afs_blocksDiscarded -= size;
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
ReleaseWriteLock(&afs_xdcache);
+
+ return 0;
}
/*!
while (afs_blocksDiscarded
&& (afs_blocksUsed >
PERCENT(CM_WAITFORDRAINPCT, afs_cacheBlocks))) {
- afs_FreeDiscardedDCache();
+ int code = afs_FreeDiscardedDCache();
+ if (code) {
+ /* Callers depend on us to get the afs_blocksDiscarded count down.
+ * If we cannot do that, the callers can spin by calling us over
+ * and over. Panic for now until we can figure out something
+ * better. */
+ osi_Panic("Error freeing discarded dcache");
+ }
}
return 0;
}
}
#else
tdc->dflags &= ~DFEntryMod;
- afs_WriteDCache(tdc, 1);
+ osi_Assert(afs_WriteDCache(tdc, 1) == 0);
#endif
}
i = afs_dvnextTbl[index]; /* next pointer this hash table */
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
int releaseTlock = 1;
- tdc = afs_GetDSlot(index, NULL);
+ tdc = afs_GetValidDSlot(index);
+ if (!tdc) {
+ /* afs_TryToSmush is best-effort; we may not actually discard
+ * everything, so failure to discard a dcache due to an i/o
+ * error is okay. */
+ continue;
+ }
if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
if (sync) {
if ((afs_indexFlags[index] & IFDataMod) == 0
for (index = afs_dvhashTbl[i]; index != NULLIDX; index = i) {
i = afs_dvnextTbl[index];
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
- tdc = afs_GetDSlot(index, NULL);
- if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
- totalChunks--;
- }
- ReleaseReadLock(&tdc->tlock);
- afs_PutDCache(tdc);
+ tdc = afs_GetValidDSlot(index);
+ if (tdc) {
+ if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
+ totalChunks--;
+ }
+ ReleaseReadLock(&tdc->tlock);
+ afs_PutDCache(tdc);
+ }
}
}
ReleaseWriteLock(&afs_xdcache);
*/
i = DCHash(&avc->f.fid, chunk);
ObtainWriteLock(&afs_xdcache, 278);
- for (index = afs_dchashTbl[i]; index != NULLIDX;) {
+ for (index = afs_dchashTbl[i]; index != NULLIDX; index = afs_dcnextTbl[index]) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
- tdc = afs_GetDSlot(index, NULL);
+ tdc = afs_GetValidDSlot(index);
+ if (!tdc) {
+ /* afs_FindDCache is best-effort; we may not find the given
+ * file/offset, so if we cannot find the given dcache due to
+ * i/o errors, that is okay. */
+ continue;
+ }
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid) && chunk == tdc->f.chunk) {
break; /* leaving refCount high for caller */
}
afs_PutDCache(tdc);
}
- index = afs_dcnextTbl[index];
}
if (index != NULLIDX) {
hset(afs_indexTimes[tdc->index], afs_indexCounter);
return NULL;
} /*afs_FindDCache */
+/* only call these from afs_AllocDCache() */
+static struct dcache *
+afs_AllocFreeDSlot(void)
+{
+ struct dcache *tdc;
+
+ tdc = afs_GetDSlotFromList(&afs_freeDCList);
+ if (!tdc) {
+ return NULL;
+ }
+ afs_indexFlags[tdc->index] &= ~IFFree;
+ ObtainWriteLock(&tdc->lock, 604);
+ afs_freeDCCount--;
+
+ return tdc;
+}
+static struct dcache *
+afs_AllocDiscardDSlot(afs_int32 lock)
+{
+ struct dcache *tdc;
+ afs_uint32 size = 0;
+ struct osi_file *file;
+
+ tdc = afs_GetDSlotFromList(&afs_discardDCList);
+ if (!tdc) {
+ return NULL;
+ }
+ afs_indexFlags[tdc->index] &= ~IFDiscarded;
+ ObtainWriteLock(&tdc->lock, 605);
+ afs_discardDCCount--;
+ size =
+ ((tdc->f.chunkBytes +
+ afs_fsfragsize) ^ afs_fsfragsize) >> 10;
+ tdc->f.states &= ~(DRO|DBackup|DRW);
+ afs_DCMoveBucket(tdc, size, 0);
+ afs_blocksDiscarded -= size;
+ afs_stats_cmperf.cacheBlocksDiscarded = afs_blocksDiscarded;
+ if ((lock & 2)) {
+ /* Truncate the chunk so zeroes get filled properly */
+ file = afs_CFileOpen(&tdc->f.inode);
+ afs_CFileTruncate(file, 0);
+ afs_CFileClose(file);
+ afs_AdjustSize(tdc, 0);
+ }
+
+ return tdc;
+}
/*!
* Get a fresh dcache from the free or discarded list.
struct VenusFid *ashFid)
{
struct dcache *tdc = NULL;
- afs_uint32 size = 0;
- struct osi_file *file;
-
- if (afs_discardDCList == NULLIDX
- || ((lock & 2) && afs_freeDCList != NULLIDX)) {
- afs_indexFlags[afs_freeDCList] &= ~IFFree;
- tdc = afs_GetDSlot(afs_freeDCList, 0);
- osi_Assert(tdc->refCount == 1);
- ReleaseReadLock(&tdc->tlock);
- ObtainWriteLock(&tdc->lock, 604);
- afs_freeDCList = afs_dvnextTbl[tdc->index];
- afs_freeDCCount--;
+ /* if (lock & 2), prefer 'free' dcaches; otherwise, prefer 'discard'
+ * dcaches. In either case, try both if our first choice doesn't work. */
+ if ((lock & 2)) {
+ tdc = afs_AllocFreeDSlot();
+ if (!tdc) {
+ tdc = afs_AllocDiscardDSlot(lock);
+ }
} else {
- afs_indexFlags[afs_discardDCList] &= ~IFDiscarded;
- tdc = afs_GetDSlot(afs_discardDCList, 0);
- osi_Assert(tdc->refCount == 1);
- ReleaseReadLock(&tdc->tlock);
- ObtainWriteLock(&tdc->lock, 605);
- afs_discardDCList = afs_dvnextTbl[tdc->index];
- afs_discardDCCount--;
- size =
- ((tdc->f.chunkBytes +
- afs_fsfragsize) ^ afs_fsfragsize) >> 10;
- tdc->f.states &= ~(DRO|DBackup|DRW);
- afs_DCMoveBucket(tdc, size, 0);
- afs_blocksDiscarded -= size;
- afs_stats_cmperf.cacheBlocksDiscarded = afs_blocksDiscarded;
- if (lock & 2) {
- /* Truncate the chunk so zeroes get filled properly */
- file = afs_CFileOpen(&tdc->f.inode);
- afs_CFileTruncate(file, 0);
- afs_CFileClose(file);
- afs_AdjustSize(tdc, 0);
+ tdc = afs_AllocDiscardDSlot(lock);
+ if (!tdc) {
+ tdc = afs_AllocFreeDSlot();
}
}
+ if (!tdc) {
+ return NULL;
+ }
/*
* Locks held:
afs_int32 index;
afs_int32 us;
afs_int32 chunk;
- afs_size_t maxGoodLength; /* amount of good data at server */
afs_size_t Position = 0;
afs_int32 size, tlen; /* size of segment to transfer */
struct afs_FetchOutput *tsmall = 0;
int doAdjustSize = 0;
int doReallyAdjustSize = 0;
int overWriteWholeChunk = 0;
+ struct rx_connection *rxconn;
#ifndef AFS_NOSTATS
struct afs_stats_AccessInfo *accP; /*Ptr to access record in stats */
*/
if (!tdc) { /* If the hint wasn't the right dcache entry */
+ int dslot_error = 0;
/*
* Hash on the [fid, chunk] and get the corresponding dcache index
* after write-locking the dcache.
ObtainWriteLock(&afs_xdcache, 280);
us = NULLIDX;
- for (index = afs_dchashTbl[i]; index != NULLIDX;) {
+ for (index = afs_dchashTbl[i]; index != NULLIDX; us = index, index = afs_dcnextTbl[index]) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
- tdc = afs_GetDSlot(index, NULL);
+ tdc = afs_GetValidDSlot(index);
+ if (!tdc) {
+ /* we got an i/o error when trying to get the given dslot,
+ * but do not bail out just yet; it is possible the dcache
+ * we're looking for is elsewhere, so it doesn't matter if
+ * we can't load this one. */
+ dslot_error = 1;
+ continue;
+ }
ReleaseReadLock(&tdc->tlock);
/*
* Locks held:
afs_PutDCache(tdc);
tdc = 0;
}
- us = index;
- index = afs_dcnextTbl[index];
}
/*
afs_Trace2(afs_iclSetp, CM_TRACE_GETDCACHE1, ICL_TYPE_POINTER,
avc, ICL_TYPE_INT32, chunk);
- /* Make sure there is a free dcache entry for us to use */
- if (afs_discardDCList == NULLIDX && afs_freeDCList == NULLIDX) {
- while (1) {
- if (!setLocks)
- avc->f.states |= CDCLock;
- /* just need slots */
- afs_GetDownD(5, (int *)0, afs_DCGetBucket(avc));
- if (!setLocks)
- avc->f.states &= ~CDCLock;
- if (afs_discardDCList != NULLIDX
- || afs_freeDCList != NULLIDX)
- break;
- /* If we can't get space for 5 mins we give up and panic */
- if (++downDCount > 300) {
- osi_Panic("getdcache");
- }
- ReleaseWriteLock(&afs_xdcache);
- /*
- * Locks held:
- * avc->lock(R) if setLocks
- * avc->lock(W) if !setLocks
- */
- afs_osi_Wait(1000, 0, 0);
- goto RetryLookup;
- }
+ if (dslot_error) {
+ /* We couldn't find the dcache we want, but we hit some i/o
+ * errors when trying to find it, so we're not sure if the
+ * dcache we want is in the cache or not. Error out, so we
+ * don't try to possibly create 2 separate dcaches for the
+ * same exact data. */
+ ReleaseWriteLock(&afs_xdcache);
+ goto done;
}
+ if (afs_discardDCList == NULLIDX && afs_freeDCList == NULLIDX) {
+ if (!setLocks)
+ avc->f.states |= CDCLock;
+ /* just need slots */
+ afs_GetDownD(5, (int *)0, afs_DCGetBucket(avc));
+ if (!setLocks)
+ avc->f.states &= ~CDCLock;
+ }
tdc = afs_AllocDCache(avc, chunk, aflags, NULL);
+ if (!tdc) {
+ /* If we can't get space for 5 mins we give up and panic */
+ if (++downDCount > 300)
+ osi_Panic("getdcache");
+ ReleaseWriteLock(&afs_xdcache);
+ /*
+ * Locks held:
+ * avc->lock(R) if setLocks
+ * avc->lock(W) if !setLocks
+ */
+ afs_osi_Wait(1000, 0, 0);
+ goto RetryLookup;
+ }
+
+ /*
+ * Locks held:
+ * avc->lock(R) if setLocks
+ * avc->lock(W) if !setLocks
+ * tdc->lock(W)
+ * afs_xdcache(W)
+ */
/*
* Now add to the two hash chains - note that i is still set
goto RetryGetDCache;
}
- /* Do not fetch data beyond truncPos. */
- maxGoodLength = avc->f.m.Length;
- if (avc->f.truncPos < maxGoodLength)
- maxGoodLength = avc->f.truncPos;
Position = AFS_CHUNKBASE(abyte);
if (vType(avc) == VDIR) {
size = avc->f.m.Length;
}
size = 999999999; /* max size for transfer */
} else {
+ afs_size_t maxGoodLength;
+
+ /* estimate how much data we're expecting back from the server,
+ * and reserve space in the dcache entry for it */
+
+ maxGoodLength = avc->f.m.Length;
+ if (avc->f.truncPos < maxGoodLength)
+ maxGoodLength = avc->f.truncPos;
+
size = AFS_CHUNKSIZE(abyte); /* expected max size */
- /* don't read past end of good data on server */
if (Position + size > maxGoodLength)
size = maxGoodLength - Position;
if (size < 0)
size = 0; /* Handle random races */
if (size > tdc->f.chunkBytes) {
- /* pre-reserve space for file */
+ /* pre-reserve estimated space for file */
afs_AdjustSize(tdc, size); /* changes chunkBytes */
- /* max size for transfer still in size */
+ }
+
+ if (size) {
+ /* For the actual fetch, do not limit the request to the
+ * length of the file. If this results in a read past EOF on
+ * the server, the server will just reply with less data than
+ * requested. If we limit ourselves to only requesting data up
+ * to the avc file length, we open ourselves up to races if the
+ * file is extended on the server at about the same time.
+ *
+ * However, we must restrict ourselves to the avc->f.truncPos
+ * length, since this represents an outstanding local
+ * truncation of the file that will be committed to the
+ * fileserver when we actually write the fileserver contents.
+ * If we do not restrict the fetch length based on
+ * avc->f.truncPos, a different truncate operation extending
+ * the file length could cause the old data after
+ * avc->f.truncPos to reappear, instead of extending the file
+ * with NUL bytes. */
+ size = AFS_CHUNKSIZE(abyte);
+ if (Position + size > avc->f.truncPos) {
+ size = avc->f.truncPos - Position;
+ }
+ if (size < 0) {
+ size = 0;
+ }
}
}
if (afs_mariner && !tdc->f.chunk)
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
}
- tsmall =
- (struct afs_FetchOutput *)osi_AllocLargeSpace(sizeof(struct afs_FetchOutput));
+ tsmall = osi_AllocLargeSpace(sizeof(struct afs_FetchOutput));
setVcacheStatus = 0;
#ifndef AFS_NOSTATS
/*
* tdc->lock(W)
*/
- tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK);
+ tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
#ifndef AFS_NOSTATS
numFetchLoops++;
#endif /* AFS_NOSTATS */
if (!setLocks || slowPass) {
- avc->callback = tc->srvr->server;
+ avc->callback = tc->parent->srvr->server;
} else {
- newCallback = tc->srvr->server;
+ newCallback = tc->parent->srvr->server;
setNewCallback = 1;
}
i = osi_Time();
- code = afs_CacheFetchProc(tc, file, Position, tdc,
+ code = afs_CacheFetchProc(tc, rxconn, file, Position, tdc,
avc, size, tsmall);
} else
code = -1;
}
} while (afs_Analyze
- (tc, code, &avc->f.fid, areq,
+ (tc, rxconn, code, &avc->f.fid, areq,
AFS_STATS_FS_RPCIDX_FETCHDATA, SHARED_LOCK, NULL));
/*
if (wrLock && (tdc->dflags & DFEntryMod)) {
tdc->dflags &= ~DFEntryMod;
ObtainWriteLock(&afs_xdcache, 620);
- afs_WriteDCache(tdc, 1);
+ osi_Assert(afs_WriteDCache(tdc, 1) == 0);
ReleaseWriteLock(&afs_xdcache);
touchedit = 1;
}
*
* Parameters:
* aslot : Dcache slot to look at.
- * tmpdc : Ptr to dcache entry.
+ * needvalid : Whether the specified slot should already exist
*
* Environment:
* Must be called with afs_xdcache write-locked.
*/
struct dcache *
-afs_MemGetDSlot(afs_int32 aslot, struct dcache *tmpdc)
+afs_MemGetDSlot(afs_int32 aslot, int indexvalid, int datavalid)
{
struct dcache *tdc;
int existing = 0;
ConvertWToRLock(&tdc->tlock);
return tdc;
}
- if (tmpdc == NULL) {
- if (!afs_freeDSList)
- afs_GetDownDSlot(4);
- if (!afs_freeDSList) {
- /* none free, making one is better than a panic */
- afs_stats_cmperf.dcacheXAllocs++; /* count in case we have a leak */
- tdc = (struct dcache *)afs_osi_Alloc(sizeof(struct dcache));
+
+ /* if 'indexvalid' is true, the slot must already exist and be populated
+ * somewhere. for memcache, the only place that dcache entries exist is
+ * in memory, so if we did not find it above, something is very wrong. */
+ osi_Assert(!indexvalid);
+
+ if (!afs_freeDSList)
+ afs_GetDownDSlot(4);
+ if (!afs_freeDSList) {
+ /* none free, making one is better than a panic */
+ afs_stats_cmperf.dcacheXAllocs++; /* count in case we have a leak */
+ tdc = afs_osi_Alloc(sizeof(struct dcache));
+ osi_Assert(tdc != NULL);
#ifdef KERNEL_HAVE_PIN
- pin((char *)tdc, sizeof(struct dcache)); /* XXX */
+ pin((char *)tdc, sizeof(struct dcache)); /* XXX */
#endif
- } else {
- tdc = afs_freeDSList;
- afs_freeDSList = (struct dcache *)tdc->lruq.next;
- existing = 1;
- }
- tdc->dflags = 0; /* up-to-date, not in free q */
- tdc->mflags = 0;
- QAdd(&afs_DLRU, &tdc->lruq);
- if (tdc->lruq.prev == &tdc->lruq)
- osi_Panic("lruq 3");
} else {
- tdc = tmpdc;
- tdc->f.states = 0;
+ tdc = afs_freeDSList;
+ afs_freeDSList = (struct dcache *)tdc->lruq.next;
+ existing = 1;
}
+ tdc->dflags = 0; /* up-to-date, not in free q */
+ tdc->mflags = 0;
+ QAdd(&afs_DLRU, &tdc->lruq);
+ if (tdc->lruq.prev == &tdc->lruq)
+ osi_Panic("lruq 3");
/* initialize entry */
tdc->f.fid.Cell = 0;
AFS_RWLOCK_INIT(&tdc->mflock, "dcache flock");
ObtainReadLock(&tdc->tlock);
- if (tmpdc == NULL)
- afs_indexTable[aslot] = tdc;
+ afs_indexTable[aslot] = tdc;
return tdc;
} /*afs_MemGetDSlot */
*
* Parameters:
* aslot : Dcache slot to look at.
- * tmpdc : Ptr to dcache entry.
+ * indexvalid : 1 if we know the slot we're giving is valid, and thus
+ * reading the dcache from the disk index should succeed. 0
+ * if we are initializing a new dcache, and so reading from
+ * the disk index may fail.
+ * datavalid : 0 if we are loading a dcache entry from the free or
+ * discard list, so we know the data in the given dcache is
+ * not valid. 1 if we are loading a known used dcache, so the
+ * data in the dcache must be valid.
*
* Environment:
* afs_xdcache lock write-locked.
*/
struct dcache *
-afs_UFSGetDSlot(afs_int32 aslot, struct dcache *tmpdc)
+afs_UFSGetDSlot(afs_int32 aslot, int indexvalid, int datavalid)
{
afs_int32 code;
struct dcache *tdc;
int existing = 0;
int entryok;
+ int off;
AFS_STATCNT(afs_UFSGetDSlot);
if (CheckLock(&afs_xdcache) != -1)
ConvertWToRLock(&tdc->tlock);
return tdc;
}
+
/* otherwise we should read it in from the cache file */
- /*
- * If we weren't passed an in-memory region to place the file info,
- * we have to allocate one.
- */
- if (tmpdc == NULL) {
- if (!afs_freeDSList)
- afs_GetDownDSlot(4);
- if (!afs_freeDSList) {
- /* none free, making one is better than a panic */
- afs_stats_cmperf.dcacheXAllocs++; /* count in case we have a leak */
- tdc = (struct dcache *)afs_osi_Alloc(sizeof(struct dcache));
+ if (!afs_freeDSList)
+ afs_GetDownDSlot(4);
+ if (!afs_freeDSList) {
+ /* none free, making one is better than a panic */
+ afs_stats_cmperf.dcacheXAllocs++; /* count in case we have a leak */
+ tdc = afs_osi_Alloc(sizeof(struct dcache));
+ osi_Assert(tdc != NULL);
#ifdef KERNEL_HAVE_PIN
- pin((char *)tdc, sizeof(struct dcache)); /* XXX */
+ pin((char *)tdc, sizeof(struct dcache)); /* XXX */
#endif
- } else {
- tdc = afs_freeDSList;
- afs_freeDSList = (struct dcache *)tdc->lruq.next;
- existing = 1;
- }
- tdc->dflags = 0; /* up-to-date, not in free q */
- tdc->mflags = 0;
- QAdd(&afs_DLRU, &tdc->lruq);
- if (tdc->lruq.prev == &tdc->lruq)
- osi_Panic("lruq 3");
} else {
- tdc = tmpdc;
- tdc->f.states = 0;
+ tdc = afs_freeDSList;
+ afs_freeDSList = (struct dcache *)tdc->lruq.next;
+ existing = 1;
}
+ tdc->dflags = 0; /* up-to-date, not in free q */
+ tdc->mflags = 0;
+ QAdd(&afs_DLRU, &tdc->lruq);
+ if (tdc->lruq.prev == &tdc->lruq)
+ osi_Panic("lruq 3");
/*
* Seek to the aslot'th entry and read it in.
*/
+ off = sizeof(struct fcache)*aslot + sizeof(struct afs_fheader);
code =
afs_osi_Read(afs_cacheInodep,
- sizeof(struct fcache) * aslot +
- sizeof(struct afs_fheader), (char *)(&tdc->f),
+ off, (char *)(&tdc->f),
sizeof(struct fcache));
entryok = 1;
- if (code != sizeof(struct fcache))
+ if (code != sizeof(struct fcache)) {
entryok = 0;
- if (!afs_CellNumValid(tdc->f.fid.Cell))
+#if defined(KERNEL_HAVE_UERROR)
+ last_error = getuerror();
+#else
+ last_error = code;
+#endif
+ lasterrtime = osi_Time();
+ if (indexvalid) {
+ struct osi_stat tstat;
+ if (afs_osi_Stat(afs_cacheInodep, &tstat)) {
+ tstat.size = -1;
+ }
+ afs_warn("afs: disk cache read error in CacheItems slot %d "
+ "off %d/%d code %d/%d\n",
+ (int)aslot,
+ off, (int)tstat.size,
+ (int)code, (int)sizeof(struct fcache));
+ /* put tdc back on the free dslot list */
+ QRemove(&tdc->lruq);
+ tdc->index = NULLIDX;
+ tdc->lruq.next = (struct afs_q *)afs_freeDSList;
+ afs_freeDSList = tdc;
+ return NULL;
+ }
+ }
+ if (!afs_CellNumValid(tdc->f.fid.Cell)) {
+ entryok = 0;
+ if (datavalid) {
+ osi_Panic("afs: needed valid dcache but index %d off %d has "
+ "invalid cell num %d\n",
+ (int)aslot, off, (int)tdc->f.fid.Cell);
+ }
+ }
+
+ if (datavalid && tdc->f.fid.Fid.Volume == 0) {
+ osi_Panic("afs: invalid zero-volume dcache entry at slot %d off %d",
+ (int)aslot, off);
+ }
+
+ if (indexvalid && !datavalid) {
+ /* we know that the given dslot does exist, but the data in it is not
+ * valid. this only occurs when we pull a dslot from the free or
+ * discard list, so be sure not to re-use the data; force invalidation.
+ */
entryok = 0;
+ }
if (!entryok) {
tdc->f.fid.Cell = 0;
tdc->f.chunk = -1;
hones(tdc->f.versionNo);
tdc->dflags |= DFEntryMod;
-#if defined(KERNEL_HAVE_UERROR)
- last_error = getuerror();
-#endif
- lasterrtime = osi_Time();
afs_indexUnique[aslot] = tdc->f.fid.Fid.Unique;
tdc->f.states &= ~(DRO|DBackup|DRW);
afs_DCMoveBucket(tdc, 0, 0);
* If we didn't read into a temporary dcache region, update the
* slot pointer table.
*/
- if (tmpdc == NULL)
- afs_indexTable[aslot] = tdc;
+ afs_indexTable[aslot] = tdc;
return tdc;
} /*afs_UFSGetDSlot */
osi_Assert(WriteLocked(&afs_xdcache));
if (atime)
adc->f.modTime = osi_Time();
+
+ if ((afs_indexFlags[adc->index] & (IFFree | IFDiscarded)) == 0 &&
+ adc->f.fid.Fid.Volume == 0) {
+ /* If a dcache slot is not on the free or discard list, it must be
+ * in the hash table. Thus, the volume must be non-zero, since that
+ * is how we determine whether or not to unhash the entry when kicking
+ * it out of the cache. Do this check now, since otherwise this can
+ * cause hash table corruption and a panic later on after we read the
+ * entry back in. */
+ osi_Panic("afs_WriteDCache zero volume index %d flags 0x%x\n",
+ adc->index, (unsigned)afs_indexFlags[adc->index]);
+ }
+
/*
* Seek to the right dcache slot and write the in-memory image out to disk.
*/
sizeof(struct fcache) * adc->index +
sizeof(struct afs_fheader), (char *)(&adc->f),
sizeof(struct fcache));
- if (code != sizeof(struct fcache))
+ if (code != sizeof(struct fcache)) {
+ afs_warn("afs: failed to write to CacheItems off %ld code %d/%d\n",
+ (long)(sizeof(struct fcache) * adc->index + sizeof(struct afs_fheader)),
+ (int)code, (int)sizeof(struct fcache));
return EIO;
+ }
return 0;
}
* is already being handled by the higher-level code.
*/
if ((avc->f.states & CSafeStore) == 0) {
- tb->code = 0;
+ tb->code_raw = tb->code_checkcode = 0;
tb->flags |= BUVALID;
if (tb->flags & BUWAIT) {
tb->flags &= ~BUWAIT;
return 0;
}
-
/*!
* Given a file name and inode, set up that file to be an
* active member in the AFS cache. This also involves checking
return EINVAL;
ObtainWriteLock(&afs_xdcache, 282);
- tdc = afs_GetDSlot(index, NULL);
+ tdc = afs_GetNewDSlot(index);
ReleaseReadLock(&tdc->tlock);
ReleaseWriteLock(&afs_xdcache);
ObtainWriteLock(&tdc->lock, 621);
ObtainWriteLock(&afs_xdcache, 622);
- if (afile) {
- code = afs_LookupInodeByPath(afile, &tdc->f.inode.ufs, NULL);
- if (code) {
- ReleaseWriteLock(&afs_xdcache);
- ReleaseWriteLock(&tdc->lock);
- afs_PutDCache(tdc);
- return code;
- }
+ if (!afile && !ainode) {
+ tfile = NULL;
+ fileIsBad = 1;
} else {
- /* Add any other 'complex' inode types here ... */
+ if (afile) {
+ code = afs_LookupInodeByPath(afile, &tdc->f.inode.ufs, NULL);
+ if (code) {
+ ReleaseWriteLock(&afs_xdcache);
+ ReleaseWriteLock(&tdc->lock);
+ afs_PutDCache(tdc);
+ return code;
+ }
+ } else {
+ /* Add any other 'complex' inode types here ... */
#if !defined(AFS_LINUX26_ENV) && !defined(AFS_CACHE_VNODE_PATH)
- tdc->f.inode.ufs = ainode;
+ tdc->f.inode.ufs = ainode;
#else
- osi_Panic("Can't init cache with inode numbers when complex inodes are "
- "in use\n");
+ osi_Panic("Can't init cache with inode numbers when complex inodes are "
+ "in use\n");
#endif
- }
- fileIsBad = 0;
- if ((tdc->f.states & DWriting) || tdc->f.fid.Fid.Volume == 0)
- fileIsBad = 1;
- tfile = osi_UFSOpen(&tdc->f.inode);
- code = afs_osi_Stat(tfile, &tstat);
- if (code)
- osi_Panic("initcachefile stat");
+ }
+ fileIsBad = 0;
+ if ((tdc->f.states & DWriting) || tdc->f.fid.Fid.Volume == 0)
+ fileIsBad = 1;
+ tfile = osi_UFSOpen(&tdc->f.inode);
+ code = afs_osi_Stat(tfile, &tstat);
+ if (code)
+ osi_Panic("initcachefile stat");
- /*
- * If file size doesn't match the cache info file, it's probably bad.
- */
- if (tdc->f.chunkBytes != tstat.size)
- fileIsBad = 1;
+ /*
+ * If file size doesn't match the cache info file, it's probably bad.
+ */
+ if (tdc->f.chunkBytes != tstat.size)
+ fileIsBad = 1;
+ /*
+ * If file changed within T (120?) seconds of cache info file, it's
+ * probably bad. In addition, if slot changed within last T seconds,
+ * the cache info file may be incorrectly identified, and so slot
+ * may be bad.
+ */
+ if (cacheInfoModTime < tstat.mtime + 120)
+ fileIsBad = 1;
+ if (cacheInfoModTime < tdc->f.modTime + 120)
+ fileIsBad = 1;
+ /* In case write through is behind, make sure cache items entry is
+ * at least as new as the chunk.
+ */
+ if (tdc->f.modTime < tstat.mtime)
+ fileIsBad = 1;
+ }
tdc->f.chunkBytes = 0;
- /*
- * If file changed within T (120?) seconds of cache info file, it's
- * probably bad. In addition, if slot changed within last T seconds,
- * the cache info file may be incorrectly identified, and so slot
- * may be bad.
- */
- if (cacheInfoModTime < tstat.mtime + 120)
- fileIsBad = 1;
- if (cacheInfoModTime < tdc->f.modTime + 120)
- fileIsBad = 1;
- /* In case write through is behind, make sure cache items entry is
- * at least as new as the chunk.
- */
- if (tdc->f.modTime < tstat.mtime)
- fileIsBad = 1;
if (fileIsBad) {
tdc->f.fid.Fid.Volume = 0; /* not in the hash table */
- if (tstat.size != 0)
+ if (tfile && tstat.size != 0)
osi_UFSTruncate(tfile, 0);
tdc->f.states &= ~(DRO|DBackup|DRW);
afs_DCMoveBucket(tdc, 0, 0);
afs_indexUnique[index] = tdc->f.fid.Fid.Unique;
} /*File is not bad */
- osi_UFSClose(tfile);
+ if (tfile)
+ osi_UFSClose(tfile);
tdc->f.states &= ~DWriting;
tdc->dflags &= ~DFEntryMod;
/* don't set f.modTime; we're just cleaning up */
- afs_WriteDCache(tdc, 0);
+ osi_Assert(afs_WriteDCache(tdc, 0) == 0);
ReleaseWriteLock(&afs_xdcache);
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
if (!aDentries)
aDentries = DDSIZE;
- if (aflags & AFSCALL_INIT_MEMCACHE) {
- /*
- * Use a memory cache instead of a disk cache
- */
- cacheDiskType = AFS_FCACHE_TYPE_MEM;
- afs_cacheType = &afs_MemCacheOps;
- afiles = (afiles < aDentries) ? afiles : aDentries; /* min */
- ablocks = afiles * (AFS_FIRSTCSIZE / 1024);
- /* ablocks is reported in 1K blocks */
- code = afs_InitMemCache(afiles, AFS_FIRSTCSIZE, aflags);
- if (code != 0) {
- afs_warn("afsd: memory cache too large for available memory.\n");
- afs_warn("afsd: AFS files cannot be accessed.\n\n");
- dcacheDisabled = 1;
- afiles = ablocks = 0;
- } else
- afs_warn("Memory cache: Allocating %d dcache entries...",
- aDentries);
- } else {
- cacheDiskType = AFS_FCACHE_TYPE_UFS;
- afs_cacheType = &afs_UfsCacheOps;
- }
-
if (aDentries > 512)
afs_dhashsize = 2048;
/* initialize hash tables */
- afs_dvhashTbl =
- (afs_int32 *) afs_osi_Alloc(afs_dhashsize * sizeof(afs_int32));
- afs_dchashTbl =
- (afs_int32 *) afs_osi_Alloc(afs_dhashsize * sizeof(afs_int32));
+ afs_dvhashTbl = afs_osi_Alloc(afs_dhashsize * sizeof(afs_int32));
+ osi_Assert(afs_dvhashTbl != NULL);
+ afs_dchashTbl = afs_osi_Alloc(afs_dhashsize * sizeof(afs_int32));
+ osi_Assert(afs_dchashTbl != NULL);
for (i = 0; i < afs_dhashsize; i++) {
afs_dvhashTbl[i] = NULLIDX;
afs_dchashTbl[i] = NULLIDX;
}
- afs_dvnextTbl = (afs_int32 *) afs_osi_Alloc(afiles * sizeof(afs_int32));
- afs_dcnextTbl = (afs_int32 *) afs_osi_Alloc(afiles * sizeof(afs_int32));
+ afs_dvnextTbl = afs_osi_Alloc(afiles * sizeof(afs_int32));
+ osi_Assert(afs_dvnextTbl != NULL);
+ afs_dcnextTbl = afs_osi_Alloc(afiles * sizeof(afs_int32));
+ osi_Assert(afs_dcnextTbl != NULL);
for (i = 0; i < afiles; i++) {
afs_dvnextTbl[i] = NULLIDX;
afs_dcnextTbl[i] = NULLIDX;
}
/* Allocate and zero the pointer array to the dcache entries */
- afs_indexTable = (struct dcache **)
- afs_osi_Alloc(sizeof(struct dcache *) * afiles);
+ afs_indexTable = afs_osi_Alloc(sizeof(struct dcache *) * afiles);
+ osi_Assert(afs_indexTable != NULL);
memset(afs_indexTable, 0, sizeof(struct dcache *) * afiles);
- afs_indexTimes =
- (afs_hyper_t *) afs_osi_Alloc(afiles * sizeof(afs_hyper_t));
+ afs_indexTimes = afs_osi_Alloc(afiles * sizeof(afs_hyper_t));
+ osi_Assert(afs_indexTimes != NULL);
memset(afs_indexTimes, 0, afiles * sizeof(afs_hyper_t));
- afs_indexUnique =
- (afs_int32 *) afs_osi_Alloc(afiles * sizeof(afs_uint32));
+ afs_indexUnique = afs_osi_Alloc(afiles * sizeof(afs_uint32));
+ osi_Assert(afs_indexUnique != NULL);
memset(afs_indexUnique, 0, afiles * sizeof(afs_uint32));
- afs_indexFlags = (u_char *) afs_osi_Alloc(afiles * sizeof(u_char));
+ afs_indexFlags = afs_osi_Alloc(afiles * sizeof(u_char));
+ osi_Assert(afs_indexFlags != NULL);
memset(afs_indexFlags, 0, afiles * sizeof(char));
/* Allocate and thread the struct dcache entries themselves */
tdp = afs_Initial_freeDSList =
- (struct dcache *)afs_osi_Alloc(aDentries * sizeof(struct dcache));
+ afs_osi_Alloc(aDentries * sizeof(struct dcache));
+ osi_Assert(tdp != NULL);
memset(tdp, 0, aDentries * sizeof(struct dcache));
#ifdef KERNEL_HAVE_PIN
pin((char *)afs_indexTable, sizeof(struct dcache *) * afiles); /* XXX */
afs_stats_cmperf.cacheBucket2_Discarded = 0;
afs_DCSizeInit();
QInit(&afs_DLRU);
+
+ if (aflags & AFSCALL_INIT_MEMCACHE) {
+ /*
+ * Use a memory cache instead of a disk cache
+ */
+ cacheDiskType = AFS_FCACHE_TYPE_MEM;
+ afs_cacheType = &afs_MemCacheOps;
+ afiles = (afiles < aDentries) ? afiles : aDentries; /* min */
+ ablocks = afiles * (AFS_FIRSTCSIZE / 1024);
+ /* ablocks is reported in 1K blocks */
+ code = afs_InitMemCache(afiles, AFS_FIRSTCSIZE, aflags);
+ if (code != 0) {
+ afs_warn("afsd: memory cache too large for available memory.\n");
+ afs_warn("afsd: AFS files cannot be accessed.\n\n");
+ dcacheDisabled = 1;
+ } else
+ afs_warn("Memory cache: Allocating %d dcache entries...",
+ aDentries);
+ } else {
+ cacheDiskType = AFS_FCACHE_TYPE_UFS;
+ afs_cacheType = &afs_UfsCacheOps;
+ }
}
/*!
/* Get a fresh dcache. */
new_dc = afs_AllocDCache(avc, 0, 0, &shadow_fid);
+ osi_Assert(new_dc);
ObtainReadLock(&adc->mflock);
ReleaseWriteLock(&afs_xdcache);
/* Alloc a 4k block. */
- data = (char *) afs_osi_Alloc(4096);
+ data = afs_osi_Alloc(4096);
if (!data) {
afs_warn("afs_MakeShadowDir: could not alloc data\n");
ret_code = ENOMEM;
git://git.openafs.org / openafs.git / blobdiff