/*!
* Create new cell alias entry and update dynroot vnode.
* \param alias
* \param cell
* \return
*/
afs_int32
afs_NewCellAlias(char *alias, char *cell)
{
struct cell_alias *tc;
ObtainSharedLock(&afs_xcell, 681);
if (afs_CellOrAliasExists_nl(alias)) {
ReleaseSharedLock(&afs_xcell);
return EEXIST;
}
UpgradeSToWLock(&afs_xcell, 682);
tc = afs_osi_Alloc(sizeof(struct cell_alias));
osi_Assert(tc != NULL);
tc->alias = afs_strdup(alias);
tc->cell = afs_strdup(cell);
tc->next = afs_cellalias_head;
tc->index = afs_cellalias_index++;
afs_cellalias_head = tc;
ReleaseWriteLock(&afs_xcell);
afs_DynrootInvalidate();
return 0;
}
int
osi_UFSTruncate(struct osi_file *afile, afs_int32 asize)
{
afs_ucred_t *oldCred;
struct vattr tvattr;
afs_int32 code;
struct osi_stat tstat;
AFS_STATCNT(osi_Truncate);
/* This routine only shrinks files, and most systems
* have very slow truncates, even when the file is already
* small enough. Check now and save some time.
*/
code = afs_osi_Stat(afile, &tstat);
if (code || tstat.size <= asize)
return code;
ObtainWriteLock(&afs_xosi, 321);
tvattr.va_mask = AT_SIZE;
tvattr.va_size = asize;
/*
* The only time a flag is used (ATTR_UTIME) is when we're changing the time
*/
AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
{
caller_context_t ct;
code = VOP_SETATTR(afile->vnode, &tvattr, 0, afs_osi_credp, &ct);
}
#else
code = VOP_SETATTR(afile->vnode, &tvattr, 0, afs_osi_credp);
#endif
AFS_GLOCK();
ReleaseWriteLock(&afs_xosi);
return code;
}
int
afs_osi_Stat(struct osi_file *afile, struct osi_stat *astat)
{
afs_int32 code;
struct vattr tvattr;
AFS_STATCNT(osi_Stat);
ObtainWriteLock(&afs_xosi, 320);
/* Ufs doesn't seem to care about the flags so we pass 0 for now */
tvattr.va_mask = AT_ALL;
AFS_GUNLOCK();
#ifdef AFS_SUN511_ENV
code = VOP_GETATTR(afile->vnode, &tvattr, 0, afs_osi_credp, NULL);
#else
code = VOP_GETATTR(afile->vnode, &tvattr, 0, afs_osi_credp);
#endif
AFS_GLOCK();
if (code == 0) {
astat->size = tvattr.va_size;
astat->mtime = tvattr.va_mtime.tv_sec;
astat->atime = tvattr.va_atime.tv_sec;
}
ReleaseWriteLock(&afs_xosi);
return code;
}
/* afs_FlushCBs
* to be used only in dire circumstances, this drops all callbacks on
* the floor, without giving them back to the server. It's ok, the server can
* deal with it, but it is a little bit rude.
*/
void
afs_FlushCBs(void)
{
register int i;
register struct vcache *tvc;
ObtainWriteLock(&afs_xcbhash, 86); /* pretty likely I'm going to remove something */
for (i = 0; i < VCSIZE; i++) /* reset all the vnodes */
for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
tvc->callback = 0;
tvc->dchint = NULL; /* invalidate hints */
tvc->f.states &= ~(CStatd);
if (QPrev(&(tvc->callsort)))
QRemove(&(tvc->callsort));
if (!(tvc->f.states & (CVInit|CVFlushed)) &&
((tvc->f.fid.Fid.Vnode & 1) || (vType(tvc) == VDIR)))
osi_dnlc_purgedp(tvc);
}
afs_InitCBQueue(0);
ReleaseWriteLock(&afs_xcbhash);
}
/**
*
* @param volid Volume ID. If it's 0, get it from the name.
* @param aname Volume name.
* @param ve Volume entry.
* @param tcell The cell containing this volume.
* @param agood
* @param type Type of volume.
* @param areq Request.
* @return Volume or NULL if failure.
*/
static struct volume *
afs_SetupVolume(afs_int32 volid, char *aname, void *ve, struct cell *tcell,
afs_int32 agood, afs_int32 type, struct vrequest *areq)
{
struct volume *tv;
struct vldbentry *ove = (struct vldbentry *)ve;
struct nvldbentry *nve = (struct nvldbentry *)ve;
struct uvldbentry *uve = (struct uvldbentry *)ve;
int whichType; /* which type of volume to look for */
int i, j, err = 0;
if (!volid) {
int len;
/* special hint from file server to use vlserver */
len = strlen(aname);
if (len >= 8 && strcmp(aname + len - 7, ".backup") == 0)
whichType = BACKVOL;
else if (len >= 10 && strcmp(aname + len - 9, ".readonly") == 0)
whichType = ROVOL;
else
whichType = RWVOL;
/* figure out which one we're really interested in (a set is returned) */
volid = afs_vtoi(aname);
if (volid == 0) {
if (type == 2) {
volid = uve->volumeId[whichType];
} else if (type == 1) {
volid = nve->volumeId[whichType];
} else {
volid = ove->volumeId[whichType];
}
} /* end of if (volid == 0) */
} /* end of if (!volid) */
ObtainWriteLock(&afs_xvolume, 108);
i = VHash(volid);
for (tv = afs_volumes[i]; tv; tv = tv->next) {
if (tv->volume == volid && tv->cell == tcell->cellNum) {
break;
}
}
if (!tv) {
struct fvolume *tf = 0;
tv = afs_GetVolSlot();
if (!tv) {
ReleaseWriteLock(&afs_xvolume);
return NULL;
}
memset(tv, 0, sizeof(struct volume));
for (j = fvTable[FVHash(tcell->cellNum, volid)]; j != 0; j = tf->next) {
if (afs_FVIndex != j) {
struct osi_file *tfile;
tfile = osi_UFSOpen(&volumeInode);
err =
afs_osi_Read(tfile, sizeof(struct fvolume) * j,
&staticFVolume, sizeof(struct fvolume));
osi_UFSClose(tfile);
if (err != sizeof(struct fvolume)) {
afs_warn("afs_SetupVolume: error %d reading volumeinfo\n",
(int)err);
/* put tv back on the free list; the data in it is not valid */
tv->next = afs_freeVolList;
afs_freeVolList = tv;
/* staticFVolume contents are not valid */
afs_FVIndex = -1;
ReleaseWriteLock(&afs_xvolume);
return NULL;
}
afs_FVIndex = j;
}
tf = &staticFVolume;
if (tf->cell == tcell->cellNum && tf->volume == volid)
break;
}
tv->cell = tcell->cellNum;
AFS_RWLOCK_INIT(&tv->lock, "volume lock");
tv->next = afs_volumes[i]; /* thread into list */
afs_volumes[i] = tv;
tv->volume = volid;
if (tf && (j != 0)) {
tv->vtix = afs_FVIndex;
tv->mtpoint = tf->mtpoint;
tv->dotdot = tf->dotdot;
tv->rootVnode = tf->rootVnode;
tv->rootUnique = tf->rootUnique;
} else {
tv->vtix = -1;
tv->rootVnode = tv->rootUnique = 0;
afs_GetDynrootMountFid(&tv->dotdot);
afs_GetDynrootMountFid(&tv->mtpoint);
tv->mtpoint.Fid.Vnode =
VNUM_FROM_TYPEID(VN_TYPE_MOUNT, tcell->cellIndex << 2);
tv->mtpoint.Fid.Unique = volid;
}
}
tv->refCount++;
tv->states &= ~VRecheck; /* just checked it */
tv->accessTime = osi_Time();
ReleaseWriteLock(&afs_xvolume);
if (type == 2) {
LockAndInstallUVolumeEntry(tv, uve, tcell->cellNum, tcell, areq);
} else if (type == 1)
LockAndInstallNVolumeEntry(tv, nve, tcell->cellNum);
else
LockAndInstallVolumeEntry(tv, ove, tcell->cellNum);
if (agood) {
if (!tv->name) {
tv->name = afs_osi_Alloc(strlen(aname) + 1);
osi_Assert(tv->name != NULL);
strcpy(tv->name, aname);
}
}
for (i = 0; i < NMAXNSERVERS; i++) {
tv->status[i] = not_busy;
}
ReleaseWriteLock(&tv->lock);
return tv;
}
/*
* This is almost exactly like the PFlush() routine in afs_pioctl.c,
* but that routine is static. We are about to change a file from
* normal caching to bypass it's caching. Therefore, we want to
* free up any cache space in use by the file, and throw out any
* existing VM pages for the file. We keep track of the number of
* times we go back and forth from caching to bypass.
*/
void
afs_TransitionToBypass(struct vcache *avc,
afs_ucred_t *acred, int aflags)
{
afs_int32 code;
struct vrequest treq;
int setDesire = 0;
int setManual = 0;
if (!avc)
return;
if (aflags & TRANSChangeDesiredBit)
setDesire = 1;
if (aflags & TRANSSetManualBit)
setManual = 1;
#ifdef AFS_BOZONLOCK_ENV
afs_BozonLock(&avc->pvnLock, avc); /* Since afs_TryToSmush will do a pvn_vptrunc */
#else
AFS_GLOCK();
#endif
ObtainWriteLock(&avc->lock, 925);
/*
* Someone may have beat us to doing the transition - we had no lock
* when we checked the flag earlier. No cause to panic, just return.
*/
if (avc->cachingStates & FCSBypass)
goto done;
/* If we never cached this, just change state */
if (setDesire && (!(avc->cachingStates & FCSBypass))) {
avc->cachingStates |= FCSBypass;
goto done;
}
/* cg2v, try to store any chunks not written 20071204 */
if (avc->execsOrWriters > 0) {
code = afs_InitReq(&treq, acred);
if (!code)
code = afs_StoreAllSegments(avc, &treq, AFS_SYNC | AFS_LASTSTORE);
}
#if 0
/* also cg2v, don't dequeue the callback */
ObtainWriteLock(&afs_xcbhash, 956);
afs_DequeueCallback(avc);
ReleaseWriteLock(&afs_xcbhash);
#endif
avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat */
/* now find the disk cache entries */
afs_TryToSmush(avc, acred, 1);
osi_dnlc_purgedp(avc);
if (avc->linkData && !(avc->f.states & CCore)) {
afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
avc->linkData = NULL;
}
avc->cachingStates |= FCSBypass; /* Set the bypass flag */
if(setDesire)
avc->cachingStates |= FCSDesireBypass;
if(setManual)
avc->cachingStates |= FCSManuallySet;
avc->cachingTransitions++;
done:
ReleaseWriteLock(&avc->lock);
#ifdef AFS_BOZONLOCK_ENV
afs_BozonUnlock(&avc->pvnLock, avc);
#else
AFS_GUNLOCK();
#endif
}
int
afs_InvalidateAllSegments(struct vcache *avc)
{
struct dcache *tdc;
afs_int32 hash;
afs_int32 index;
struct dcache **dcList;
int i, dcListMax, dcListCount;
AFS_STATCNT(afs_InvalidateAllSegments);
afs_Trace2(afs_iclSetp, CM_TRACE_INVALL, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
hash = DVHash(&avc->f.fid);
avc->f.truncPos = AFS_NOTRUNC; /* don't truncate later */
avc->f.states &= ~CExtendedFile; /* not any more */
ObtainWriteLock(&afs_xcbhash, 459);
afs_DequeueCallback(avc);
avc->f.states &= ~(CStatd | CDirty); /* mark status information as bad, too */
ReleaseWriteLock(&afs_xcbhash);
if (avc->f.fid.Fid.Vnode & 1 || (vType(avc) == VDIR))
osi_dnlc_purgedp(avc);
/* Blow away pages; for now, only for Solaris */
#if (defined(AFS_SUN5_ENV))
if (WriteLocked(&avc->lock))
osi_ReleaseVM(avc, (afs_ucred_t *)0);
#endif
/*
* Block out others from screwing with this table; is a read lock
* sufficient?
*/
ObtainWriteLock(&afs_xdcache, 286);
dcListMax = 0;
for (index = afs_dvhashTbl[hash]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
/* In the case of fatal errors during stores, we MUST
* invalidate all of the relevant chunks. Otherwise, the chunks
* will be left with the 'new' data that was never successfully
* written to the server, but the DV in the dcache is still the
* old DV. So, we may indefinitely serve data to applications
* that is not actually in the file on the fileserver. If we
* cannot afs_GetValidDSlot the appropriate entries, currently
* there is no way to ensure the dcache is invalidated. So for
* now, to avoid risking serving bad data from the cache, panic
* instead. */
osi_Panic("afs_InvalidateAllSegments tdc count");
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid))
dcListMax++;
afs_PutDCache(tdc);
}
index = afs_dvnextTbl[index];
}
dcList = osi_Alloc(dcListMax * sizeof(struct dcache *));
dcListCount = 0;
for (index = afs_dvhashTbl[hash]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
/* We cannot proceed after getting this error; we risk serving
* incorrect data to applications. So panic instead. See the
* above comment next to the previous afs_GetValidDSlot call
* for details. */
osi_Panic("afs_InvalidateAllSegments tdc store");
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
/* same file? we'll zap it */
if (afs_indexFlags[index] & IFDataMod) {
afs_stats_cmperf.cacheCurrDirtyChunks--;
/* don't write it back */
afs_indexFlags[index] &= ~IFDataMod;
}
afs_indexFlags[index] &= ~IFAnyPages;
if (dcListCount < dcListMax)
dcList[dcListCount++] = tdc;
else
afs_PutDCache(tdc);
} else {
afs_PutDCache(tdc);
}
}
index = afs_dvnextTbl[index];
}
ReleaseWriteLock(&afs_xdcache);
for (i = 0; i < dcListCount; i++) {
tdc = dcList[i];
ObtainWriteLock(&tdc->lock, 679);
ZapDCE(tdc);
if (vType(avc) == VDIR)
DZap(tdc);
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
osi_Free(dcList, dcListMax * sizeof(struct dcache *));
return 0;
}
void
unlock_cmdLine(void)
{
ReleaseWriteLock(&cmdLineLock);
}
int
afs_MemRead(struct vcache *avc, struct uio *auio,
afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, tlen;
afs_size_t len = 0;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error, trybusy = 1;
afs_int32 code;
struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop = NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
memset(&tuio, 0, sizeof(tuio));
#endif
AFS_STATCNT(afs_MemRead);
if (avc->vc_error)
return EIO;
/* check that we have the latest status info in the vnode cache */
if ((code = afs_CreateReq(&treq, acred)))
return code;
if (!noLock) {
code = afs_VerifyVCache(avc, treq);
if (code) {
code = afs_CheckCode(code, treq, 8); /* failed to get it */
afs_DestroyReq(treq);
return code;
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
code = afs_CheckCode(EACCES, treq, 9);
afs_DestroyReq(treq);
return code;
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
memset(tvec, 0, sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
/*
* Locks held:
* avc->lock(R)
*/
/* This bit is bogus. We're checking to see if the read goes past the
* end of the file. If so, we should be zeroing out all of the buffers
* that the client has passed into us (there is a danger that we may leak
* kernel memory if we do not). However, this behaviour is disabled by
* not setting len before this segment runs, and by setting len to 0
* immediately we enter it. In addition, we also need to check for a read
* which partially goes off the end of the file in the while loop below.
*/
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->f.chunkBytes - offset;
}
} else {
int versionOk;
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
try_background:
#endif
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw.
*/
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now,
* and we'll need new data */
tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
if (trybusy && !afs_BBusy()) {
#endif
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 665);
if (!(tdc->mflags & DFFetchReq)) {
int dontwait = B_DONTWAIT;
/* start the daemon (may already be running, however) */
UpgradeSToWLock(&tdc->mflock, 666);
tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS)
dontwait = 0;
#endif
bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc, (void *)0, (void *)0);
if (!bp) {
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
/* don't use bp pointer! */
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
len = tdc->validPos - filePos;
versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* no longer fetching, verify data version
* (avoid new GetDCache call) */
if (versionOk && len > 0) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %lld m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
goto try_background;
}
#if 0
printf("afs_read: forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
#endif
}
#endif
/* If we get here, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (!tdc) {
error = EIO;
break;
}
/*
* Locks held:
* avc->lock(R)
* tdc->lock(R)
*/
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the mem cache */
/* mung uio structure to be right for this transfer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
code = afs_MemReadUIO(&tdc->f.inode, tuiop);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
} /* the whole while loop */
/*
* Locks held:
* avc->lock(R)
* tdc->lock(R) if tdc
*/
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
/*
* try to queue prefetch, if needed. If DataVersion is zero there
* should not be any more: files with DV 0 never have been stored
* on the fileserver, symbolic links and directories never require
* more than a single chunk.
*/
if (!noLock && !(hiszero(avc->f.m.DataVersion)) &&
#ifndef AFS_VM_RDWR_ENV
afs_preCache
#else
1
#endif
) {
afs_PrefetchChunk(avc, tdc, acred, treq);
}
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
error = afs_CheckCode(error, treq, 10);
afs_DestroyReq(treq);
return error;
}
/* called with the dcache entry triggering the fetch, the vcache entry involved,
* and a vrequest for the read call. Marks the dcache entry as having already
* triggered a prefetch, starts the prefetch going and sets the DFFetchReq
* flag in the prefetched block, so that the next call to read knows to wait
* for the daemon to start doing things.
*
* This function must be called with the vnode at least read-locked, and
* no locks on the dcache, because it plays around with dcache entries.
*/
void
afs_PrefetchChunk(struct vcache *avc, struct dcache *adc,
afs_ucred_t *acred, struct vrequest *areq)
{
struct dcache *tdc;
afs_size_t offset;
afs_size_t j1, j2; /* junk vbls for GetDCache to trash */
offset = adc->f.chunk + 1; /* next chunk we'll need */
offset = AFS_CHUNKTOBASE(offset); /* base of next chunk */
ObtainReadLock(&adc->lock);
ObtainSharedLock(&adc->mflock, 662);
if (offset < avc->f.m.Length && !(adc->mflags & DFNextStarted)
&& !afs_BBusy()) {
struct brequest *bp;
UpgradeSToWLock(&adc->mflock, 663);
adc->mflags |= DFNextStarted; /* we've tried to prefetch for this guy */
ReleaseWriteLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
tdc = afs_GetDCache(avc, offset, areq, &j1, &j2, 2); /* type 2 never returns 0 */
/*
* In disconnected mode, type 2 can return 0 because it doesn't
* make any sense to allocate a dcache we can never fill
*/
if (tdc == NULL)
return;
ObtainSharedLock(&tdc->mflock, 651);
if (!(tdc->mflags & DFFetchReq)) {
/* ask the daemon to do the work */
UpgradeSToWLock(&tdc->mflock, 652);
tdc->mflags |= DFFetchReq; /* guaranteed to be cleared by BKG or GetDCache */
/* last parm (1) tells bkg daemon to do an afs_PutDCache when it is done,
* since we don't want to wait for it to finish before doing so ourselves.
*/
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
(afs_size_t) offset, (afs_size_t) 1, tdc,
(void *)0, (void *)0);
if (!bp) {
/* Bkg table full; just abort non-important prefetching to avoid deadlocks */
tdc->mflags &= ~DFFetchReq;
ReleaseWriteLock(&tdc->mflock);
afs_PutDCache(tdc);
/*
* DCLOCKXXX: This is a little sketchy, since someone else
* could have already started a prefetch.. In practice,
* this probably doesn't matter; at most it would cause an
* extra slot in the BKG table to be used up when someone
* prefetches this for the second time.
*/
ObtainReadLock(&adc->lock);
ObtainWriteLock(&adc->mflock, 664);
adc->mflags &= ~DFNextStarted;
ReleaseWriteLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
} else {
ReleaseWriteLock(&tdc->mflock);
}
} else {
ReleaseSharedLock(&tdc->mflock);
afs_PutDCache(tdc);
}
} else {
ReleaseSharedLock(&adc->mflock);
ReleaseReadLock(&adc->lock);
}
}
int
afs_UFSRead(struct vcache *avc, struct uio *auio,
afs_ucred_t *acred, daddr_t albn, struct buf **abpp,
int noLock)
{
afs_size_t totalLength;
afs_size_t transferLength;
afs_size_t filePos;
afs_size_t offset, tlen;
afs_size_t len = 0;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error;
struct osi_file *tfile;
afs_int32 code;
int trybusy = 1;
struct vrequest *treq = NULL;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop=NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
memset(&tuio, 0, sizeof(tuio));
#endif
AFS_STATCNT(afs_UFSRead);
if (avc && avc->vc_error)
return EIO;
AFS_DISCON_LOCK();
/* check that we have the latest status info in the vnode cache */
if ((code = afs_CreateReq(&treq, acred)))
return code;
if (!noLock) {
if (!avc)
osi_Panic("null avc in afs_UFSRead");
else {
code = afs_VerifyVCache(avc, treq);
if (code) {
code = afs_CheckCode(code, treq, 11); /* failed to get it */
afs_DestroyReq(treq);
AFS_DISCON_UNLOCK();
return code;
}
}
}
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
if (!afs_AccessOK
(avc, PRSFS_READ, treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
AFS_DISCON_UNLOCK();
code = afs_CheckCode(EACCES, treq, 12);
afs_DestroyReq(treq);
return code;
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(sizeof(struct iovec));
memset(tvec, 0, sizeof(struct iovec));
#endif
totalLength = AFS_UIO_RESID(auio);
filePos = AFS_UIO_OFFSET(auio);
afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
totalLength, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
/* This bit is bogus. We're checking to see if the read goes past the
* end of the file. If so, we should be zeroing out all of the buffers
* that the client has passed into us (there is a danger that we may leak
* kernel memory if we do not). However, this behaviour is disabled by
* not setting len before this segment runs, and by setting len to 0
* immediately we enter it. In addition, we also need to check for a read
* which partially goes off the end of the file in the while loop below.
*/
if (filePos >= avc->f.m.Length) {
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
len = 0;
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
}
while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
if (filePos >= avc->f.m.Length)
break; /* all done */
if (noLock) {
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
}
} else {
int versionOk;
/* a tricky question: does the presence of the DFFetching flag
* mean that we're fetching the latest version of the file? No.
* The server could update the file as soon as the fetch responsible
* for the setting of the DFFetching flag completes.
*
* However, the presence of the DFFetching flag (visible under
* a dcache read lock since it is set and cleared only under a
* dcache write lock) means that we're fetching as good a version
* as was known to this client at the time of the last call to
* afs_VerifyVCache, since the latter updates the stat cache's
* m.DataVersion field under a vcache write lock, and from the
* time that the DFFetching flag goes on in afs_GetDCache (before
* the fetch starts), to the time it goes off (after the fetch
* completes), afs_GetDCache keeps at least a read lock on the
* vcache entry.
*
* This means that if the DFFetching flag is set, we can use that
* data for any reads that must come from the current version of
* the file (current == m.DataVersion).
*
* Another way of looking at this same point is this: if we're
* fetching some data and then try do an afs_VerifyVCache, the
* VerifyVCache operation will not complete until after the
* DFFetching flag is turned off and the dcache entry's f.versionNo
* field is updated.
*
* Note, by the way, that if DFFetching is set,
* m.DataVersion > f.versionNo (the latter is not updated until
* after the fetch completes).
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc); /* before reusing tdc */
}
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
try_background:
#endif
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 2);
if (!tdc) {
error = ENETDOWN;
break;
}
ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw. */
if (!(tdc->dflags & DFFetching)
&& !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now, and we'll need new data */
tagain:
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (trybusy && (!afs_BBusy() || (afs_protocols & VICEP_ACCESS))) {
#else
if (trybusy && !afs_BBusy()) {
#endif
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 667);
if (!(tdc->mflags & DFFetchReq)) {
int dontwait = B_DONTWAIT;
UpgradeSToWLock(&tdc->mflock, 668);
tdc->mflags |= DFFetchReq;
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS)
dontwait = 0;
#endif
bp = afs_BQueue(BOP_FETCH, avc, dontwait, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc, (void *)0, (void *)0);
if (!bp) {
/* Bkg table full; retry deadlocks */
tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
ConvertWToSLock(&tdc->mflock);
}
code = 0;
ConvertSToRLock(&tdc->mflock);
while (!code && tdc->mflags & DFFetchReq) {
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
__LINE__, ICL_TYPE_POINTER, tdc,
ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
ReleaseReadLock(&tdc->mflock);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
ObtainReadLock(&tdc->mflock);
}
ReleaseReadLock(&tdc->mflock);
if (code) {
error = code;
break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
* data is now streaming in, then wait for some interesting stuff.
*/
code = 0;
while (!code && (tdc->dflags & DFFetching)
&& tdc->validPos <= filePos) {
/* too early: wait for DFFetching flag to vanish,
* or data to appear */
afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
__FILE__, ICL_TYPE_INT32, __LINE__,
ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
tdc->dflags);
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
ObtainReadLock(&tdc->lock);
}
if (code) {
error = code;
break;
}
/* fetching flag gone, data is here, or we never tried
* (BBusy for instance) */
len = tdc->validPos - filePos;
versionOk = hsame(avc->f.m.DataVersion, tdc->f.versionNo) ? 1 : 0;
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* no longer fetching, verify data version (avoid new
* GetDCache call) */
if (versionOk && len > 0) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
} else {
/* don't have current data, so get it below */
afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
ICL_TYPE_HYPER, &avc->f.m.DataVersion,
ICL_TYPE_HYPER, &tdc->f.versionNo);
#if 0
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) {
printf("afs_read: DV mismatch? %d instead of %d for %u.%u.%u\n", tdc->f.versionNo.low, avc->f.m.DataVersion.low, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
printf("afs_read: validPos %llu filePos %llu totalLength %d m.Length %llu noLock %d\n", tdc->validPos, filePos, totalLength, avc->f.m.Length, noLock);
printf("afs_read: or len too low? %lld for %u.%u.%u\n", len, avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
}
#endif
#endif
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
#ifdef STRUCT_TASK_STRUCT_HAS_CRED
if (afs_protocols & VICEP_ACCESS) { /* avoid foreground fetch */
if (!versionOk) {
printf("afs_read: avoid forground %u.%u.%u\n", avc->f.fid.Fid.Volume, avc->f.fid.Fid.Vnode, avc->f.fid.Fid.Unique);
goto try_background;
}
}
#endif
/* If we get here, it was not possible to start the
* background daemon. With flag == 1 afs_GetDCache
* does the FetchData rpc synchronously.
*/
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
if (tdc)
ObtainReadLock(&tdc->lock);
}
}
if (!tdc) {
error = EIO;
break;
}
len = tdc->validPos - filePos;
afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (len > totalLength)
len = totalLength; /* will read len bytes */
if (len <= 0) { /* shouldn't get here if DFFetching is on */
afs_Trace4(afs_iclSetp, CM_TRACE_VNODEREAD2, ICL_TYPE_POINTER,
tdc, ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(tdc->validPos),
ICL_TYPE_INT32, tdc->f.chunkBytes, ICL_TYPE_INT32,
tdc->dflags);
/* read past the end of a chunk, may not be at next chunk yet, and yet
* also not at eof, so may have to supply fake zeros */
len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
if (len > totalLength)
len = totalLength; /* and still within xfr request */
tlen = avc->f.m.Length - offset; /* and still within file */
if (len > tlen)
len = tlen;
if (len > AFS_ZEROS)
len = sizeof(afs_zeros); /* and in 0 buffer */
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
#else
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
#endif
AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
} else {
/* get the data from the file */
tfile = (struct osi_file *)osi_UFSOpen(&tdc->f.inode);
#ifdef AFS_DARWIN80_ENV
trimlen = len;
tuiop = afsio_darwin_partialcopy(auio, trimlen);
uio_setoffset(tuiop, offset);
#else
/* mung uio structure to be right for this transfer */
afsio_copy(auio, &tuio, tvec);
trimlen = len;
afsio_trim(&tuio, trimlen);
tuio.afsio_offset = offset;
#endif
#if defined(AFS_AIX41_ENV)
AFS_GUNLOCK();
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
NULL, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_AIX32_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
/* Flush all JFS pages now for big performance gain in big file cases
* If we do something like this, must check to be sure that AFS file
* isn't mmapped... see afs_gn_map() for why.
*/
/*
if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
many different ways to do similar things:
so far, the best performing one is #2, but #1 might match it if we
straighten out the confusion regarding which pages to flush. It
really does matter.
1. vm_flushp(tfile->vnode->v_gnode->gn_seg, 0, len/PAGESIZE - 1);
2. vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
3. vms_inactive(tfile->vnode->v_gnode->gn_seg) Doesn't work correctly
4. vms_delete(tfile->vnode->v_gnode->gn_seg) probably also fails
tfile->vnode->v_gnode->gn_seg = NULL;
5. deletep
6. ipgrlse
7. ifreeseg
Unfortunately, this seems to cause frequent "cache corruption" episodes.
vm_releasep(tfile->vnode->v_gnode->gn_seg, offset/PAGESIZE,
(len + PAGESIZE-1)/PAGESIZE);
}
*/
#elif defined(AFS_AIX_ENV)
code =
VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
&tuio, NULL, NULL, -1);
#elif defined(AFS_SUN5_ENV)
AFS_GUNLOCK();
#ifdef AFS_SUN510_ENV
VOP_RWLOCK(tfile->vnode, 0, NULL);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, NULL);
VOP_RWUNLOCK(tfile->vnode, 0, NULL);
#else
VOP_RWLOCK(tfile->vnode, 0);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_RWUNLOCK(tfile->vnode, 0);
#endif
AFS_GLOCK();
#elif defined(AFS_SGI_ENV)
AFS_GUNLOCK();
AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, afs_osi_credp,
code);
AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
AFS_GLOCK();
#elif defined(AFS_HPUX100_ENV)
AFS_GUNLOCK();
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
AFS_GLOCK();
#elif defined(AFS_LINUX20_ENV)
AFS_GUNLOCK();
code = osi_rdwr(tfile, &tuio, UIO_READ);
AFS_GLOCK();
#elif defined(AFS_DARWIN80_ENV)
AFS_GUNLOCK();
code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, current_proc());
AFS_GLOCK();
#elif defined(AFS_FBSD80_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_FBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curthread);
AFS_GLOCK();
#elif defined(AFS_NBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0);
AFS_GLOCK();
#elif defined(AFS_XBSD_ENV)
AFS_GUNLOCK();
VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp);
VOP_UNLOCK(tfile->vnode, 0, curproc);
AFS_GLOCK();
#else
code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, afs_osi_credp);
#endif
osi_UFSClose(tfile);
if (code) {
error = code;
break;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
trimlen = len;
afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
if (len <= 0)
break; /* surprise eof */
#ifdef AFS_DARWIN80_ENV
if (tuiop) {
uio_free(tuiop);
tuiop = 0;
}
#endif
}
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch, obviously.
*/
if (tdc) {
ReleaseReadLock(&tdc->lock);
#if !defined(AFS_VM_RDWR_ENV)
/*
* try to queue prefetch, if needed. If DataVersion is zero there
* should not be any more: files with DV 0 never have been stored
* on the fileserver, symbolic links and directories never require
* more than a single chunk.
*/
if (!noLock && !(hiszero(avc->f.m.DataVersion))) {
if (!(tdc->mflags & DFNextStarted))
afs_PrefetchChunk(avc, tdc, acred, treq);
}
#endif
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
#ifdef AFS_DARWIN80_ENV
if (tuiop)
uio_free(tuiop);
#else
osi_FreeSmallSpace(tvec);
#endif
AFS_DISCON_UNLOCK();
error = afs_CheckCode(error, treq, 13);
afs_DestroyReq(treq);
return error;
}
/*
* Regenerates the dynroot contents from the current list of
* cells. Useful when the list of cells has changed due to
* an AFSDB lookup, for instance.
*/
static void
afs_RebuildDynroot(void)
{
int cellidx, maxcellidx, i;
int aliasidx, maxaliasidx;
struct cell *c;
struct cell_alias *ca;
int curChunk, curPage;
int dirSize, dotLen;
char *newDir, *dotCell;
struct DirHeader *dirHeader;
int linkCount = 0;
struct afs_dynSymlink *ts;
int newVersion;
ObtainReadLock(&afs_dynrootDirLock);
newVersion = afs_dynrootVersion;
ReleaseReadLock(&afs_dynrootDirLock);
/*
* Compute the amount of space we need for the fake dir
*/
curChunk = 13;
curPage = 0;
/* Reserve space for "." and ".." */
curChunk += 2;
/* Reserve space for the dynamic-mount directory */
afs_dynroot_computeDirEnt(AFS_DYNROOT_MOUNTNAME, &curPage, &curChunk);
for (cellidx = 0;; cellidx++) {
c = afs_GetCellByIndex(cellidx, READ_LOCK);
if (!c)
break;
if (c->cellNum == afs_dynrootCell)
continue;
dotLen = strlen(c->cellName) + 2;
dotCell = afs_osi_Alloc(dotLen);
strcpy(dotCell, ".");
afs_strcat(dotCell, c->cellName);
afs_dynroot_computeDirEnt(c->cellName, &curPage, &curChunk);
afs_dynroot_computeDirEnt(dotCell, &curPage, &curChunk);
afs_osi_Free(dotCell, dotLen);
afs_PutCell(c, READ_LOCK);
}
maxcellidx = cellidx;
for (aliasidx = 0;; aliasidx++) {
ca = afs_GetCellAlias(aliasidx);
if (!ca)
break;
dotLen = strlen(ca->alias) + 2;
dotCell = afs_osi_Alloc(dotLen);
strcpy(dotCell, ".");
afs_strcat(dotCell, ca->alias);
afs_dynroot_computeDirEnt(ca->alias, &curPage, &curChunk);
afs_dynroot_computeDirEnt(dotCell, &curPage, &curChunk);
afs_osi_Free(dotCell, dotLen);
afs_PutCellAlias(ca);
}
maxaliasidx = aliasidx;
ObtainReadLock(&afs_dynSymlinkLock);
ts = afs_dynSymlinkBase;
while (ts) {
afs_dynroot_computeDirEnt(ts->name, &curPage, &curChunk);
ts = ts->next;
}
dirSize = (curPage + 1) * AFS_PAGESIZE;
newDir = afs_osi_Alloc(dirSize);
/*
* Now actually construct the directory.
*/
curChunk = 13;
curPage = 0;
dirHeader = (struct DirHeader *)newDir;
dirHeader->header.pgcount = 0;
dirHeader->header.tag = htons(1234);
dirHeader->header.freecount = 0;
dirHeader->header.freebitmap[0] = 0xff;
dirHeader->header.freebitmap[1] = 0x1f;
for (i = 2; i < EPP / 8; i++)
dirHeader->header.freebitmap[i] = 0;
dirHeader->alloMap[0] = EPP - DHE - 1;
for (i = 1; i < MAXPAGES; i++)
dirHeader->alloMap[i] = EPP;
for (i = 0; i < NHASHENT; i++)
dirHeader->hashTable[i] = 0;
/* Install ".", "..", and the dynamic mount directory */
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk, ".", 1);
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk, "..", 1);
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk,
AFS_DYNROOT_MOUNTNAME, AFS_DYNROOT_MOUNT_VNODE);
linkCount += 3;
for (cellidx = 0; cellidx < maxcellidx; cellidx++) {
c = afs_GetCellByIndex(cellidx, READ_LOCK);
if (!c)
continue;
if (c->cellNum == afs_dynrootCell)
continue;
dotLen = strlen(c->cellName) + 2;
dotCell = afs_osi_Alloc(dotLen);
strcpy(dotCell, ".");
afs_strcat(dotCell, c->cellName);
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk, c->cellName,
VNUM_FROM_CIDX_RW(cellidx, 0));
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk, dotCell,
VNUM_FROM_CIDX_RW(cellidx, 1));
afs_osi_Free(dotCell, dotLen);
linkCount += 2;
afs_PutCell(c, READ_LOCK);
}
for (aliasidx = 0; aliasidx < maxaliasidx; aliasidx++) {
ca = afs_GetCellAlias(aliasidx);
if (!ca)
continue;
dotLen = strlen(ca->alias) + 2;
dotCell = afs_osi_Alloc(dotLen);
strcpy(dotCell, ".");
afs_strcat(dotCell, ca->alias);
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk, ca->alias,
VNUM_FROM_CAIDX_RW(aliasidx, 0));
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk, dotCell,
VNUM_FROM_CAIDX_RW(aliasidx, 1));
afs_osi_Free(dotCell, dotLen);
afs_PutCellAlias(ca);
}
ts = afs_dynSymlinkBase;
while (ts) {
int vnum = VNUM_FROM_TYPEID(VN_TYPE_SYMLINK, ts->index);
afs_dynroot_addDirEnt(dirHeader, &curPage, &curChunk, ts->name, vnum);
ts = ts->next;
}
ReleaseReadLock(&afs_dynSymlinkLock);
ObtainWriteLock(&afs_dynrootDirLock, 549);
if (afs_dynrootDir)
afs_osi_Free(afs_dynrootDir, afs_dynrootDirLen);
afs_dynrootDir = newDir;
afs_dynrootDirLen = dirSize;
afs_dynrootDirLinkcnt = linkCount;
afs_dynrootDirVersion = newVersion;
ReleaseWriteLock(&afs_dynrootDirLock);
}
int
osi_dnlc_enter(struct vcache *adp, char *aname, struct vcache *avc,
afs_hyper_t * avno)
{
struct nc *tnc;
unsigned int key, skey;
char *ts = aname;
int safety;
if (!afs_usednlc)
return 0;
TRACE(osi_dnlc_enterT, 0);
dnlcHash(ts, key); /* leaves ts pointing at the NULL */
if (ts - aname >= AFSNCNAMESIZE) {
return 0;
}
skey = key & (NHSIZE - 1);
dnlcstats.enters++;
retry:
ObtainWriteLock(&afs_xdnlc, 222);
/* Only cache entries from the latest version of the directory */
if (!(adp->f.states & CStatd) || !hsame(*avno, adp->f.m.DataVersion)) {
ReleaseWriteLock(&afs_xdnlc);
return 0;
}
/*
* Make sure each directory entry gets cached no more than once.
*/
for (tnc = nameHash[skey], safety = 0; tnc; tnc = tnc->next, safety++) {
if ((tnc->dirp == adp) && (!strcmp((char *)tnc->name, aname))) {
/* duplicate entry */
break;
} else if (tnc->next == nameHash[skey]) { /* end of list */
tnc = NULL;
break;
} else if (safety > NCSIZE) {
afs_warn("DNLC cycle");
dnlcstats.cycles++;
ReleaseWriteLock(&afs_xdnlc);
osi_dnlc_purge();
goto retry;
}
}
if (tnc == NULL) {
tnc = GetMeAnEntry();
tnc->dirp = adp;
tnc->vp = avc;
tnc->key = key;
memcpy((char *)tnc->name, aname, ts - aname + 1); /* include the NULL */
InsertEntry(tnc);
} else {
/* duplicate */
tnc->vp = avc;
}
ReleaseWriteLock(&afs_xdnlc);
return 0;
}
/* don't set CDirty in here because RPC is called synchronously */
int
afs_symlink(OSI_VC_DECL(adp), char *aname, struct vattr *attrs,
char *atargetName, struct vcache **tvcp, afs_ucred_t *acred)
{
afs_uint32 now = 0;
struct vrequest *treq = NULL;
afs_int32 code = 0;
struct afs_conn *tc;
struct VenusFid newFid;
struct dcache *tdc;
afs_size_t offset, len;
afs_int32 alen;
struct server *hostp = 0;
struct vcache *tvc;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus *OutFidStatus, *OutDirStatus;
struct AFSCallBack CallBack;
struct AFSVolSync tsync;
struct volume *volp = 0;
struct afs_fakestat_state fakestate;
struct rx_connection *rxconn;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_symlink);
afs_Trace2(afs_iclSetp, CM_TRACE_SYMLINK, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname);
OutFidStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
OutDirStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
memset(&InStatus, 0, sizeof(InStatus));
if ((code = afs_CreateReq(&treq, acred)))
goto done2;
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, treq);
if (code)
goto done;
if (strlen(aname) > AFSNAMEMAX || strlen(atargetName) > AFSPATHMAX) {
code = ENAMETOOLONG;
goto done;
}
if (afs_IsDynroot(adp)) {
code = afs_DynrootVOPSymlink(adp, acred, aname, atargetName);
goto done;
}
if (afs_IsDynrootMount(adp)) {
code = EROFS;
goto done;
}
code = afs_VerifyVCache(adp, treq);
if (code) {
code = afs_CheckCode(code, treq, 30);
goto done;
}
/** If the volume is read-only, return error without making an RPC to the
* fileserver
*/
if (adp->f.states & CRO) {
code = EROFS;
goto done;
}
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
InStatus.Mask = AFS_SETMODTIME | AFS_SETMODE;
InStatus.ClientModTime = osi_Time();
alen = strlen(atargetName); /* we want it to include the null */
if ( (*atargetName == '#' || *atargetName == '%') && alen > 1 && atargetName[alen-1] == '.') {
InStatus.UnixModeBits = 0644; /* mt pt: null from "." at end */
if (alen == 1)
alen++; /* Empty string */
} else {
InStatus.UnixModeBits = 0755;
alen++; /* add in the null */
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, treq, &offset, &len, 1);
volp = afs_FindVolume(&adp->f.fid, READ_LOCK); /*parent is also in same vol */
ObtainWriteLock(&adp->lock, 156);
if (tdc)
ObtainWriteLock(&tdc->lock, 636);
/* No further locks: if the SymLink succeeds, it does not matter what happens
* to our local copy of the directory. If somebody tampers with it in the meantime,
* the copy will be invalidated */
if (!AFS_IS_DISCON_RW) {
do {
tc = afs_Conn(&adp->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
hostp = tc->parent->srvr->server;
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_SYMLINK);
if (adp->f.states & CForeign) {
now = osi_Time();
RX_AFS_GUNLOCK();
code =
RXAFS_DFSSymlink(rxconn,
(struct AFSFid *)&adp->f.fid.Fid,
aname, atargetName, &InStatus,
(struct AFSFid *)&newFid.Fid,
OutFidStatus, OutDirStatus,
&CallBack, &tsync);
RX_AFS_GLOCK();
} else {
RX_AFS_GUNLOCK();
code =
RXAFS_Symlink(rxconn, (struct AFSFid *)&adp->f.fid.Fid,
aname, atargetName, &InStatus,
(struct AFSFid *)&newFid.Fid,
OutFidStatus, OutDirStatus, &tsync);
RX_AFS_GLOCK();
}
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &adp->f.fid, treq, AFS_STATS_FS_RPCIDX_SYMLINK,
SHARED_LOCK, NULL));
} else {
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
afs_GenFakeFid(&newFid, VREG, 0);
}
ObtainWriteLock(&afs_xvcache, 40);
if (code) {
if (code < 0) {
afs_StaleVCache(adp);
}
ReleaseWriteLock(&adp->lock);
ReleaseWriteLock(&afs_xvcache);
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto done;
}
/* otherwise, we should see if we can make the change to the dir locally */
if (AFS_IS_DISCON_RW || afs_LocalHero(adp, tdc, OutDirStatus, 1)) {
/* we can do it locally */
ObtainWriteLock(&afs_xdcache, 293);
/* If the following fails because the name has been created in the meantime, the
* directory is out-of-date - the file server knows best! */
code = afs_dir_Create(tdc, aname, &newFid.Fid);
ReleaseWriteLock(&afs_xdcache);
if (code && !AFS_IS_DISCON_RW) {
ZapDCE(tdc); /* surprise error -- use invalid value */
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
ReleaseWriteLock(&adp->lock);
/* now we're done with parent dir, create the link's entry. Note that
* no one can get a pointer to the new cache entry until we release
* the xvcache lock. */
tvc = afs_NewVCache(&newFid, hostp);
if (!tvc)
{
code = -2;
ReleaseWriteLock(&afs_xvcache);
goto done;
}
ObtainWriteLock(&tvc->lock, 157);
ObtainWriteLock(&afs_xcbhash, 500);
tvc->f.states |= CStatd; /* have valid info */
tvc->f.states &= ~CBulkFetching;
if (adp->f.states & CForeign) {
tvc->f.states |= CForeign;
/* We don't have to worry about losing the callback since we're doing it
* under the afs_xvcache lock actually, afs_NewVCache may drop the
* afs_xvcache lock, if it calls afs_FlushVCache */
tvc->cbExpires = CallBack.ExpirationTime + now;
afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), volp);
} else {
tvc->cbExpires = 0x7fffffff; /* never expires, they can't change */
/* since it never expires, we don't have to queue the callback */
}
ReleaseWriteLock(&afs_xcbhash);
if (AFS_IS_DISCON_RW) {
attrs->va_mode = InStatus.UnixModeBits;
afs_GenDisconStatus(adp, tvc, &newFid, attrs, treq, VLNK);
code = afs_DisconCreateSymlink(tvc, atargetName, treq);
if (code) {
/* XXX - When this goes wrong, we need to tidy up the changes we made to
* the parent, and get rid of the vcache we just created */
ReleaseWriteLock(&tvc->lock);
ReleaseWriteLock(&afs_xvcache);
afs_PutVCache(tvc);
goto done;
}
afs_DisconAddDirty(tvc, VDisconCreate, 0);
} else {
afs_ProcessFS(tvc, OutFidStatus, treq);
}
if (!tvc->linkData) {
tvc->linkData = afs_osi_Alloc(alen);
osi_Assert(tvc->linkData != NULL);
strncpy(tvc->linkData, atargetName, alen - 1);
tvc->linkData[alen - 1] = 0;
}
ReleaseWriteLock(&tvc->lock);
ReleaseWriteLock(&afs_xvcache);
if (tvcp)
*tvcp = tvc;
else
afs_PutVCache(tvc);
code = 0;
done:
afs_PutFakeStat(&fakestate);
if (volp)
afs_PutVolume(volp, READ_LOCK);
AFS_DISCON_UNLOCK();
code = afs_CheckCode(code, treq, 31);
afs_DestroyReq(treq);
done2:
osi_FreeSmallSpace(OutFidStatus);
osi_FreeSmallSpace(OutDirStatus);
return code;
}
int
DRead(struct dcache *adc, int page, struct DirBuffer *entry)
{
/* Read a page from the disk. */
struct buffer *tb, *tb2;
struct osi_file *tfile;
int code;
AFS_STATCNT(DRead);
memset(entry, 0, sizeof(struct DirBuffer));
ObtainWriteLock(&afs_bufferLock, 256);
#define bufmatch(tb) (tb->page == page && tb->fid == adc->index)
#define buf_Front(head,parent,p) {(parent)->hashNext = (p)->hashNext; (p)->hashNext= *(head);*(head)=(p);}
/* this apparently-complicated-looking code is simply an example of
* a little bit of loop unrolling, and is a standard linked-list
* traversal trick. It saves a few assignments at the the expense
* of larger code size. This could be simplified by better use of
* macros.
*/
if ((tb = phTable[pHash(adc->index, page)])) {
if (bufmatch(tb)) {
ObtainWriteLock(&tb->lock, 257);
tb->lockers++;
ReleaseWriteLock(&afs_bufferLock);
tb->accesstime = timecounter++;
AFS_STATS(afs_stats_cmperf.bufHits++);
ReleaseWriteLock(&tb->lock);
entry->buffer = tb;
entry->data = tb->data;
return 0;
} else {
struct buffer **bufhead;
bufhead = &(phTable[pHash(adc->index, page)]);
while ((tb2 = tb->hashNext)) {
if (bufmatch(tb2)) {
buf_Front(bufhead, tb, tb2);
ObtainWriteLock(&tb2->lock, 258);
tb2->lockers++;
ReleaseWriteLock(&afs_bufferLock);
tb2->accesstime = timecounter++;
AFS_STATS(afs_stats_cmperf.bufHits++);
ReleaseWriteLock(&tb2->lock);
entry->buffer = tb2;
entry->data = tb2->data;
return 0;
}
if ((tb = tb2->hashNext)) {
if (bufmatch(tb)) {
buf_Front(bufhead, tb2, tb);
ObtainWriteLock(&tb->lock, 259);
tb->lockers++;
ReleaseWriteLock(&afs_bufferLock);
tb->accesstime = timecounter++;
AFS_STATS(afs_stats_cmperf.bufHits++);
ReleaseWriteLock(&tb->lock);
entry->buffer = tb;
entry->data = tb->data;
return 0;
}
} else
break;
}
}
} else
tb2 = NULL;
AFS_STATS(afs_stats_cmperf.bufMisses++);
/* can't find it */
/* The last thing we looked at was either tb or tb2 (or nothing). That
* is at least the oldest buffer on one particular hash chain, so it's
* a pretty good place to start looking for the truly oldest buffer.
*/
tb = afs_newslot(adc, page, (tb ? tb : tb2));
if (!tb) {
ReleaseWriteLock(&afs_bufferLock);
return EIO;
}
ObtainWriteLock(&tb->lock, 260);
tb->lockers++;
ReleaseWriteLock(&afs_bufferLock);
if (page * AFS_BUFFER_PAGESIZE >= adc->f.chunkBytes) {
tb->fid = NULLIDX;
afs_reset_inode(&tb->inode);
tb->lockers--;
ReleaseWriteLock(&tb->lock);
return EIO;
}
tfile = afs_CFileOpen(&adc->f.inode);
code =
afs_CFileRead(tfile, tb->page * AFS_BUFFER_PAGESIZE, tb->data,
AFS_BUFFER_PAGESIZE);
afs_CFileClose(tfile);
if (code < AFS_BUFFER_PAGESIZE) {
tb->fid = NULLIDX;
afs_reset_inode(&tb->inode);
tb->lockers--;
ReleaseWriteLock(&tb->lock);
return EIO;
}
/* Note that findslot sets the page field in the buffer equal to
* what it is searching for. */
ReleaseWriteLock(&tb->lock);
entry->buffer = tb;
entry->data = tb->data;
return 0;
}
/*!
* Handles all the reconnection details:
* - Get all the details about the vnode: name, fid, and parent dir fid.
* - Send data to server.
* - Handle errors.
* - Reorder vhash and dcaches in their hashes, using the newly acquired fid.
*/
int
afs_ProcessOpCreate(struct vcache *avc, struct vrequest *areq,
afs_ucred_t *acred)
{
char *tname = NULL, *ttargetName = NULL;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus OutFidStatus, OutDirStatus;
struct VenusFid pdir_fid, newFid;
struct AFSCallBack CallBack;
struct AFSVolSync tsync;
struct vcache *tdp = NULL, *tvc = NULL;
struct dcache *tdc = NULL;
struct afs_conn *tc;
struct rx_connection *rxconn;
afs_int32 hash, new_hash, index;
afs_size_t tlen;
int code, op = 0;
XSTATS_DECLS;
tname = afs_osi_Alloc(AFSNAMEMAX);
if (!tname)
return ENOMEM;
code = afs_GetParentVCache(avc, 0, &pdir_fid, tname, &tdp);
if (code)
goto end;
/* This data may also be in linkData, but then we have to deal with
* the joy of terminating NULLs and . and file modes. So just get
* it from the dcache where it won't have been fiddled with.
*/
if (vType(avc) == VLNK) {
afs_size_t offset;
struct dcache *tdc;
struct osi_file *tfile;
tdc = afs_GetDCache(avc, 0, areq, &offset, &tlen, 0);
if (!tdc) {
code = ENOENT;
goto end;
}
if (tlen > 1024) {
afs_PutDCache(tdc);
code = EFAULT;
goto end;
}
tlen++; /* space for NULL */
ttargetName = afs_osi_Alloc(tlen);
if (!ttargetName) {
afs_PutDCache(tdc);
return ENOMEM;
}
ObtainReadLock(&tdc->lock);
tfile = afs_CFileOpen(&tdc->f.inode);
code = afs_CFileRead(tfile, 0, ttargetName, tlen);
ttargetName[tlen-1] = '\0';
afs_CFileClose(tfile);
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
}
/* Set status. */
InStatus.Mask = AFS_SETMODTIME | AFS_SETMODE | AFS_SETGROUP;
InStatus.ClientModTime = avc->f.m.Date;
InStatus.Owner = avc->f.m.Owner;
InStatus.Group = (afs_int32) afs_cr_gid(acred);
/* Only care about protection bits. */
InStatus.UnixModeBits = avc->f.m.Mode & 0xffff;
do {
tc = afs_Conn(&tdp->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
switch (vType(avc)) {
case VREG:
/* Make file on server. */
op = AFS_STATS_FS_RPCIDX_CREATEFILE;
XSTATS_START_TIME(op);
RX_AFS_GUNLOCK();
code = RXAFS_CreateFile(tc->id,
(struct AFSFid *)&tdp->f.fid.Fid,
tname, &InStatus,
(struct AFSFid *) &newFid.Fid,
&OutFidStatus, &OutDirStatus,
&CallBack, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
break;
case VDIR:
/* Make dir on server. */
op = AFS_STATS_FS_RPCIDX_MAKEDIR;
XSTATS_START_TIME(op);
RX_AFS_GUNLOCK();
code = RXAFS_MakeDir(rxconn, (struct AFSFid *) &tdp->f.fid.Fid,
tname, &InStatus,
(struct AFSFid *) &newFid.Fid,
&OutFidStatus, &OutDirStatus,
&CallBack, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
break;
case VLNK:
/* Make symlink on server. */
op = AFS_STATS_FS_RPCIDX_SYMLINK;
XSTATS_START_TIME(op);
RX_AFS_GUNLOCK();
code = RXAFS_Symlink(rxconn,
(struct AFSFid *) &tdp->f.fid.Fid,
tname, ttargetName, &InStatus,
(struct AFSFid *) &newFid.Fid,
&OutFidStatus, &OutDirStatus, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
break;
default:
op = AFS_STATS_FS_RPCIDX_CREATEFILE;
code = 1;
break;
}
} else
code = -1;
} while (afs_Analyze(tc, rxconn, code, &tdp->f.fid, areq, op, SHARED_LOCK, NULL));
/* TODO: Handle errors. */
if (code) {
/* printf("afs_ProcessOpCreate: error while creating vnode on server, code=%d .\n", code); */
goto end;
}
/* The rpc doesn't set the cell number. */
newFid.Cell = avc->f.fid.Cell;
/*
* Change the fid in the dir entry.
*/
/* Seek the dir's dcache. */
tdc = afs_FindDCacheByFid(&tdp->f.fid);
if (tdc) {
/* And now change the fid in the parent dir entry. */
afs_dir_ChangeFid(tdc, tname, &avc->f.fid.Fid.Vnode, &newFid.Fid.Vnode);
afs_PutDCache(tdc);
}
if (vType(avc) == VDIR) {
/* Change fid in the dir for the "." entry. ".." has alredy been
* handled by afs_FixChildrenFids when processing the parent dir.
*/
tdc = afs_FindDCacheByFid(&avc->f.fid);
if (tdc) {
afs_dir_ChangeFid(tdc, ".", &avc->f.fid.Fid.Vnode,
&newFid.Fid.Vnode);
if (avc->f.m.LinkCount >= 2)
/* For non empty dirs, fix children's parentVnode and
* parentUnique reference.
*/
afs_FixChildrenFids(&avc->f.fid, &newFid);
afs_PutDCache(tdc);
}
}
/* Recompute hash chain positions for vnode and dcaches.
* Then change to the new FID.
*/
/* The vcache goes first. */
ObtainWriteLock(&afs_xvcache, 735);
/* Old fid hash. */
hash = VCHash(&avc->f.fid);
/* New fid hash. */
new_hash = VCHash(&newFid);
/* Remove hash from old position. */
/* XXX: not checking array element contents. It shouldn't be empty.
* If it oopses, then something else might be wrong.
*/
if (afs_vhashT[hash] == avc) {
/* First in hash chain (might be the only one). */
afs_vhashT[hash] = avc->hnext;
} else {
/* More elements in hash chain. */
for (tvc = afs_vhashT[hash]; tvc; tvc = tvc->hnext) {
if (tvc->hnext == avc) {
tvc->hnext = avc->hnext;
break;
}
}
} /* if (!afs_vhashT[i]->hnext) */
QRemove(&avc->vhashq);
/* Insert hash in new position. */
avc->hnext = afs_vhashT[new_hash];
afs_vhashT[new_hash] = avc;
QAdd(&afs_vhashTV[VCHashV(&newFid)], &avc->vhashq);
ReleaseWriteLock(&afs_xvcache);
/* Do the same thing for all dcaches. */
hash = DVHash(&avc->f.fid);
ObtainWriteLock(&afs_xdcache, 743);
for (index = afs_dvhashTbl[hash]; index != NULLIDX; index = hash) {
hash = afs_dvnextTbl[index];
tdc = afs_GetValidDSlot(index);
ReleaseReadLock(&tdc->tlock);
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
/* Safer but slower. */
afs_HashOutDCache(tdc, 0);
/* Put dcache in new positions in the dchash and dvhash. */
new_hash = DCHash(&newFid, tdc->f.chunk);
afs_dcnextTbl[tdc->index] = afs_dchashTbl[new_hash];
afs_dchashTbl[new_hash] = tdc->index;
new_hash = DVHash(&newFid);
afs_dvnextTbl[tdc->index] = afs_dvhashTbl[new_hash];
afs_dvhashTbl[new_hash] = tdc->index;
afs_indexUnique[tdc->index] = newFid.Fid.Unique;
memcpy(&tdc->f.fid, &newFid, sizeof(struct VenusFid));
} /* if fid match */
} /* if uniquifier match */
if (tdc)
afs_PutDCache(tdc);
} /* for all dcaches in this hash bucket */
ReleaseWriteLock(&afs_xdcache);
/* Now we can set the new fid. */
memcpy(&avc->f.fid, &newFid, sizeof(struct VenusFid));
end:
if (tdp)
afs_PutVCache(tdp);
afs_osi_Free(tname, AFSNAMEMAX);
if (ttargetName)
afs_osi_Free(ttargetName, tlen);
return code;
}
/*!
* All files that have been dirty before disconnection are going to
* be replayed back to the server.
*
* \param areq Request from the user.
* \param acred User credentials.
*
* \return If all files synchronized succesfully, return 0, otherwise
* return error code
*
* \note For now, it's the request from the PDiscon pioctl.
*
*/
int
afs_ResyncDisconFiles(struct vrequest *areq, afs_ucred_t *acred)
{
struct afs_conn *tc;
struct rx_connection *rxconn;
struct vcache *tvc;
struct AFSFetchStatus fstat;
struct AFSCallBack callback;
struct AFSVolSync tsync;
int code = 0;
afs_int32 start = 0;
XSTATS_DECLS;
/*AFS_STATCNT(afs_ResyncDisconFiles);*/
ObtainWriteLock(&afs_disconDirtyLock, 707);
while (!QEmpty(&afs_disconDirty)) {
tvc = QEntry(QPrev(&afs_disconDirty), struct vcache, dirtyq);
/* Can't lock tvc whilst holding the discon dirty lock */
ReleaseWriteLock(&afs_disconDirtyLock);
/* Get local write lock. */
ObtainWriteLock(&tvc->lock, 705);
if (tvc->f.ddirty_flags & VDisconRemove) {
/* Delete the file on the server and just move on
* to the next file. After all, it has been deleted
* we can't replay any other operation it.
*/
code = afs_ProcessOpRemove(tvc, areq);
goto next_file;
} else if (tvc->f.ddirty_flags & VDisconCreate) {
/* For newly created files, we don't need a server lock. */
code = afs_ProcessOpCreate(tvc, areq, acred);
if (code)
goto next_file;
tvc->f.ddirty_flags &= ~VDisconCreate;
tvc->f.ddirty_flags |= VDisconCreated;
}
#if 0
/* Get server write lock. */
do {
tc = afs_Conn(&tvc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_SETLOCK);
RX_AFS_GUNLOCK();
code = RXAFS_SetLock(rxconn,
(struct AFSFid *)&tvc->f.fid.Fid,
LockWrite,
&tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze(tc,
rxconn,
code,
&tvc->f.fid,
areq,
AFS_STATS_FS_RPCIDX_SETLOCK,
SHARED_LOCK,
NULL));
if (code)
goto next_file;
#endif
if (tvc->f.ddirty_flags & VDisconRename) {
/* If we're renaming the file, do so now */
code = afs_ProcessOpRename(tvc, areq);
if (code)
goto unlock_srv_file;
}
/* Issue a FetchStatus to get info about DV and callbacks. */
do {
tc = afs_Conn(&tvc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
tvc->callback = tc->srvr->server;
start = osi_Time();
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHSTATUS);
RX_AFS_GUNLOCK();
code = RXAFS_FetchStatus(rxconn,
(struct AFSFid *)&tvc->f.fid.Fid,
&fstat,
&callback,
&tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze(tc,
rxconn,
code,
&tvc->f.fid,
areq,
AFS_STATS_FS_RPCIDX_FETCHSTATUS,
SHARED_LOCK,
NULL));
if (code) {
goto unlock_srv_file;
}
if ((dv_match(tvc, fstat) && (tvc->f.m.Date == fstat.ServerModTime)) ||
(afs_ConflictPolicy == CLIENT_WINS) ||
(tvc->f.ddirty_flags & VDisconCreated)) {
/*
* Send changes to the server if there's data version match, or
* client wins policy has been selected or file has been created
* but doesn't have it's the contents on to the server yet.
*/
/*
* XXX: Checking server attr changes by timestamp might not the
* most elegant solution, but it's the most viable one that we could find.
*/
afs_UpdateStatus(tvc, &tvc->f.fid, areq, &fstat, &callback, start);
code = afs_SendChanges(tvc, areq);
} else if (afs_ConflictPolicy == SERVER_WINS) {
/* DV mismatch, apply collision resolution policy. */
/* Discard this files chunks and remove from current dir. */
afs_ResetVCache(tvc, acred, 0);
tvc->f.truncPos = AFS_NOTRUNC;
} else {
/* printf("afs_ResyncDisconFiles: no resolution policy selected.\n"); */
} /* if DV match or client wins policy */
unlock_srv_file:
/* Release server write lock. */
#if 0
do {
tc = afs_Conn(&tvc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_RELEASELOCK);
RX_AFS_GUNLOCK();
ucode = RXAFS_ReleaseLock(rxconn,
(struct AFSFid *) &tvc->f.fid.Fid,
&tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
ucode = -1;
} while (afs_Analyze(tc,
rxconn,
ucode,
&tvc->f.fid,
areq,
AFS_STATS_FS_RPCIDX_RELEASELOCK,
SHARED_LOCK,
NULL));
#endif
next_file:
ObtainWriteLock(&afs_disconDirtyLock, 710);
if (code == 0) {
/* Replayed successfully - pull the vcache from the
* disconnected list */
tvc->f.ddirty_flags = 0;
QRemove(&tvc->dirtyq);
afs_PutVCache(tvc);
} else {
if (code == EAGAIN) {
/* Operation was deferred. Pull it from the current place in
* the list, and stick it at the end again */
QRemove(&tvc->dirtyq);
QAdd(&afs_disconDirty, &tvc->dirtyq);
} else {
/* Failed - keep state as is, and let the user know we died */
ReleaseWriteLock(&tvc->lock);
break;
}
}
/* Release local write lock. */
ReleaseWriteLock(&tvc->lock);
} /* while (tvc) */
if (code) {
ReleaseWriteLock(&afs_disconDirtyLock);
return code;
}
/* Dispose of all of the shadow directories */
afs_DisconDiscardAllShadows(0, acred);
ReleaseWriteLock(&afs_disconDirtyLock);
return code;
}
int
afsremove(register struct vcache *adp, register struct dcache *tdc,
register struct vcache *tvc, char *aname, afs_ucred_t *acred,
struct vrequest *treqp)
{
register afs_int32 code = 0;
register struct afs_conn *tc;
struct AFSFetchStatus OutDirStatus;
struct AFSVolSync tsync;
XSTATS_DECLS;
if (!AFS_IS_DISCONNECTED) {
do {
tc = afs_Conn(&adp->f.fid, treqp, SHARED_LOCK);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_REMOVEFILE);
RX_AFS_GUNLOCK();
code =
RXAFS_RemoveFile(tc->id, (struct AFSFid *)&adp->f.fid.Fid,
aname, &OutDirStatus, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, code, &adp->f.fid, treqp, AFS_STATS_FS_RPCIDX_REMOVEFILE,
SHARED_LOCK, NULL));
}
osi_dnlc_remove(adp, aname, tvc);
if (code) {
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
if (tvc)
afs_PutVCache(tvc);
if (code < 0) {
ObtainWriteLock(&afs_xcbhash, 497);
afs_DequeueCallback(adp);
adp->f.states &= ~CStatd;
ReleaseWriteLock(&afs_xcbhash);
osi_dnlc_purgedp(adp);
}
ReleaseWriteLock(&adp->lock);
code = afs_CheckCode(code, treqp, 21);
return code;
}
if (tdc)
UpgradeSToWLock(&tdc->lock, 637);
if (AFS_IS_DISCON_RW || afs_LocalHero(adp, tdc, &OutDirStatus, 1)) {
/* we can do it locally */
code = afs_dir_Delete(tdc, aname);
if (code) {
ZapDCE(tdc); /* surprise error -- invalid value */
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc); /* drop ref count */
}
ReleaseWriteLock(&adp->lock);
/* now, get vnode for unlinked dude, and see if we should force it
* from cache. adp is now the deleted files vnode. Note that we
* call FindVCache instead of GetVCache since if the file's really
* gone, we won't be able to fetch the status info anyway. */
if (tvc) {
afs_MarinerLog("store$Removing", tvc);
#ifdef AFS_BOZONLOCK_ENV
afs_BozonLock(&tvc->pvnLock, tvc);
/* Since afs_TryToSmush will do a pvn_vptrunc */
#endif
ObtainWriteLock(&tvc->lock, 141);
/* note that callback will be broken on the deleted file if there are
* still >0 links left to it, so we'll get the stat right */
tvc->f.m.LinkCount--;
tvc->f.states &= ~CUnique; /* For the dfs xlator */
if (tvc->f.m.LinkCount == 0 && !osi_Active(tvc)) {
if (!AFS_NFSXLATORREQ(acred))
afs_TryToSmush(tvc, acred, 0);
}
ReleaseWriteLock(&tvc->lock);
#ifdef AFS_BOZONLOCK_ENV
afs_BozonUnlock(&tvc->pvnLock, tvc);
#endif
afs_PutVCache(tvc);
}
return (0);
}
int
afsrename(struct vcache *aodp, char *aname1, struct vcache *andp,
char *aname2, struct AFS_UCRED *acred, struct vrequest *areq)
{
register struct afs_conn *tc;
register afs_int32 code = 0;
afs_int32 returnCode;
int oneDir, doLocally;
afs_size_t offset, len;
struct VenusFid unlinkFid, fileFid;
struct vcache *tvc;
struct dcache *tdc1, *tdc2;
struct AFSFetchStatus OutOldDirStatus, OutNewDirStatus;
struct AFSVolSync tsync;
XSTATS_DECLS;
AFS_STATCNT(afs_rename);
afs_Trace4(afs_iclSetp, CM_TRACE_RENAME, ICL_TYPE_POINTER, aodp,
ICL_TYPE_STRING, aname1, ICL_TYPE_POINTER, andp,
ICL_TYPE_STRING, aname2);
if (strlen(aname1) > AFSNAMEMAX || strlen(aname2) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done;
}
/* verify the latest versions of the stat cache entries */
tagain:
code = afs_VerifyVCache(aodp, areq);
if (code)
goto done;
code = afs_VerifyVCache(andp, areq);
if (code)
goto done;
/* lock in appropriate order, after some checks */
if (aodp->f.fid.Cell != andp->f.fid.Cell
|| aodp->f.fid.Fid.Volume != andp->f.fid.Fid.Volume) {
code = EXDEV;
goto done;
}
oneDir = 0;
code = 0;
if (andp->f.fid.Fid.Vnode == aodp->f.fid.Fid.Vnode) {
if (!strcmp(aname1, aname2)) {
/* Same directory and same name; this is a noop and just return success
* to save cycles and follow posix standards */
code = 0;
goto done;
}
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
ObtainWriteLock(&andp->lock, 147);
tdc1 = afs_GetDCache(aodp, (afs_size_t) 0, areq, &offset, &len, 0);
if (!tdc1) {
code = ENOENT;
} else {
ObtainWriteLock(&tdc1->lock, 643);
}
tdc2 = tdc1;
oneDir = 1; /* only one dude locked */
} else if ((andp->f.states & CRO) || (aodp->f.states & CRO)) {
code = EROFS;
goto done;
} else if (andp->f.fid.Fid.Vnode < aodp->f.fid.Fid.Vnode) {
ObtainWriteLock(&andp->lock, 148); /* lock smaller one first */
ObtainWriteLock(&aodp->lock, 149);
tdc2 = afs_FindDCache(andp, (afs_size_t) 0);
if (tdc2)
ObtainWriteLock(&tdc2->lock, 644);
tdc1 = afs_GetDCache(aodp, (afs_size_t) 0, areq, &offset, &len, 0);
if (tdc1)
ObtainWriteLock(&tdc1->lock, 645);
else
code = ENOENT;
} else {
ObtainWriteLock(&aodp->lock, 150); /* lock smaller one first */
ObtainWriteLock(&andp->lock, 557);
tdc1 = afs_GetDCache(aodp, (afs_size_t) 0, areq, &offset, &len, 0);
if (tdc1)
ObtainWriteLock(&tdc1->lock, 646);
else
code = ENOENT;
tdc2 = afs_FindDCache(andp, (afs_size_t) 0);
if (tdc2)
ObtainWriteLock(&tdc2->lock, 647);
}
osi_dnlc_remove(aodp, aname1, 0);
osi_dnlc_remove(andp, aname2, 0);
/*
* Make sure that the data in the cache is current. We may have
* received a callback while we were waiting for the write lock.
*/
if (tdc1) {
if (!(aodp->f.states & CStatd)
|| !hsame(aodp->f.m.DataVersion, tdc1->f.versionNo)) {
ReleaseWriteLock(&aodp->lock);
if (!oneDir) {
if (tdc2) {
ReleaseWriteLock(&tdc2->lock);
afs_PutDCache(tdc2);
}
ReleaseWriteLock(&andp->lock);
}
ReleaseWriteLock(&tdc1->lock);
afs_PutDCache(tdc1);
goto tagain;
}
}
if (code == 0)
code = afs_dir_Lookup(tdc1, aname1, &fileFid.Fid);
if (code) {
if (tdc1) {
ReleaseWriteLock(&tdc1->lock);
afs_PutDCache(tdc1);
}
ReleaseWriteLock(&aodp->lock);
if (!oneDir) {
if (tdc2) {
ReleaseWriteLock(&tdc2->lock);
afs_PutDCache(tdc2);
}
ReleaseWriteLock(&andp->lock);
}
goto done;
}
if (!AFS_IS_DISCON_RW) {
/* Connected. */
do {
tc = afs_Conn(&aodp->f.fid, areq, SHARED_LOCK);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_RENAME);
RX_AFS_GUNLOCK();
code =
RXAFS_Rename(tc->id,
(struct AFSFid *)&aodp->f.fid.Fid,
aname1,
(struct AFSFid *)&andp->f.fid.Fid,
aname2,
&OutOldDirStatus,
&OutNewDirStatus,
&tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, code, &andp->f.fid, areq, AFS_STATS_FS_RPCIDX_RENAME,
SHARED_LOCK, NULL));
} else {
#if defined(AFS_DISCON_ENV)
/* Disconnected. */
/* Seek moved file vcache. */
fileFid.Cell = aodp->f.fid.Cell;
fileFid.Fid.Volume = aodp->f.fid.Fid.Volume;
ObtainSharedLock(&afs_xvcache, 754);
tvc = afs_FindVCache(&fileFid, 0 , 1);
ReleaseSharedLock(&afs_xvcache);
if (tvc) {
/* XXX - We're locking this vcache whilst holding dcaches. Ooops */
ObtainWriteLock(&tvc->lock, 750);
if (!(tvc->f.ddirty_flags & (VDisconRename|VDisconCreate))) {
/* If the vnode was created locally, then we don't care
* about recording the rename - we'll do it automatically
* on replay. If we've already renamed, we've already stored
* the required information about where we came from.
*/
if (!aodp->f.shadow.vnode) {
/* Make shadow copy of parent dir only. */
afs_MakeShadowDir(aodp, tdc1);
}
/* Save old parent dir fid so it will be searchable
* in the shadow dir.
*/
tvc->f.oldParent.vnode = aodp->f.fid.Fid.Vnode;
tvc->f.oldParent.unique = aodp->f.fid.Fid.Unique;
afs_DisconAddDirty(tvc,
VDisconRename
| (oneDir ? VDisconRenameSameDir:0),
1);
}
ReleaseWriteLock(&tvc->lock);
afs_PutVCache(tvc);
} else {
code = ENOENT;
} /* if (tvc) */
#endif
} /* if !(AFS_IS_DISCON_RW)*/
returnCode = code; /* remember for later */
/* Now we try to do things locally. This is really loathsome code. */
unlinkFid.Fid.Vnode = 0;
if (code == 0) {
/* In any event, we don't really care if the data (tdc2) is not
* in the cache; if it isn't, we won't do the update locally. */
/* see if version numbers increased properly */
doLocally = 1;
if (!AFS_IS_DISCON_RW) {
if (oneDir) {
/* number increases by 1 for whole rename operation */
if (!afs_LocalHero(aodp, tdc1, &OutOldDirStatus, 1)) {
doLocally = 0;
}
} else {
/* two separate dirs, each increasing by 1 */
if (!afs_LocalHero(aodp, tdc1, &OutOldDirStatus, 1))
doLocally = 0;
if (!afs_LocalHero(andp, tdc2, &OutNewDirStatus, 1))
doLocally = 0;
if (!doLocally) {
if (tdc1) {
ZapDCE(tdc1);
DZap(tdc1);
}
if (tdc2) {
ZapDCE(tdc2);
DZap(tdc2);
}
}
}
} /* if (!AFS_IS_DISCON_RW) */
/* now really do the work */
if (doLocally) {
/* first lookup the fid of the dude we're moving */
code = afs_dir_Lookup(tdc1, aname1, &fileFid.Fid);
if (code == 0) {
/* delete the source */
code = afs_dir_Delete(tdc1, aname1);
}
/* first see if target is there */
if (code == 0
&& afs_dir_Lookup(tdc2, aname2,
&unlinkFid.Fid) == 0) {
/* target already exists, and will be unlinked by server */
code = afs_dir_Delete(tdc2, aname2);
}
if (code == 0) {
ObtainWriteLock(&afs_xdcache, 292);
code = afs_dir_Create(tdc2, aname2, &fileFid.Fid);
ReleaseWriteLock(&afs_xdcache);
}
if (code != 0) {
ZapDCE(tdc1);
DZap(tdc1);
if (!oneDir) {
ZapDCE(tdc2);
DZap(tdc2);
}
}
}
/* update dir link counts */
if (AFS_IS_DISCON_RW) {
if (!oneDir) {
aodp->f.m.LinkCount--;
andp->f.m.LinkCount++;
}
/* If we're in the same directory, link count doesn't change */
} else {
aodp->f.m.LinkCount = OutOldDirStatus.LinkCount;
if (!oneDir)
andp->f.m.LinkCount = OutNewDirStatus.LinkCount;
}
} else { /* operation failed (code != 0) */
if (code < 0) {
/* if failed, server might have done something anyway, and
* assume that we know about it */
ObtainWriteLock(&afs_xcbhash, 498);
afs_DequeueCallback(aodp);
afs_DequeueCallback(andp);
andp->f.states &= ~CStatd;
aodp->f.states &= ~CStatd;
ReleaseWriteLock(&afs_xcbhash);
osi_dnlc_purgedp(andp);
osi_dnlc_purgedp(aodp);
}
}
/* release locks */
if (tdc1) {
ReleaseWriteLock(&tdc1->lock);
afs_PutDCache(tdc1);
}
if ((!oneDir) && tdc2) {
ReleaseWriteLock(&tdc2->lock);
afs_PutDCache(tdc2);
}
ReleaseWriteLock(&aodp->lock);
if (!oneDir) {
ReleaseWriteLock(&andp->lock);
}
if (returnCode) {
code = returnCode;
goto done;
}
/* now, some more details. if unlinkFid.Fid.Vnode then we should decrement
* the link count on this file. Note that if fileFid is a dir, then we don't
* have to invalidate its ".." entry, since its DataVersion # should have
* changed. However, interface is not good enough to tell us the
* *file*'s new DataVersion, so we're stuck. Our hack: delete mark
* the data as having an "unknown" version (effectively discarding the ".."
* entry */
if (unlinkFid.Fid.Vnode) {
unlinkFid.Fid.Volume = aodp->f.fid.Fid.Volume;
unlinkFid.Cell = aodp->f.fid.Cell;
tvc = NULL;
if (!unlinkFid.Fid.Unique) {
tvc = afs_LookupVCache(&unlinkFid, areq, NULL, aodp, aname1);
}
if (!tvc) /* lookup failed or wasn't called */
tvc = afs_GetVCache(&unlinkFid, areq, NULL, NULL);
if (tvc) {
#ifdef AFS_BOZONLOCK_ENV
afs_BozonLock(&tvc->pvnLock, tvc); /* Since afs_TryToSmush will do a pvn_vptrunc */
#endif
ObtainWriteLock(&tvc->lock, 151);
tvc->f.m.LinkCount--;
tvc->f.states &= ~CUnique; /* For the dfs xlator */
if (tvc->f.m.LinkCount == 0 && !osi_Active(tvc)) {
/* if this was last guy (probably) discard from cache.
* We have to be careful to not get rid of the stat
* information, since otherwise operations will start
* failing even if the file was still open (or
* otherwise active), and the server no longer has the
* info. If the file still has valid links, we'll get
* a break-callback msg from the server, so it doesn't
* matter that we don't discard the status info */
if (!AFS_NFSXLATORREQ(acred))
afs_TryToSmush(tvc, acred, 0);
}
ReleaseWriteLock(&tvc->lock);
#ifdef AFS_BOZONLOCK_ENV
afs_BozonUnlock(&tvc->pvnLock, tvc);
#endif
afs_PutVCache(tvc);
}
}
/* now handle ".." invalidation */
if (!oneDir) {
fileFid.Fid.Volume = aodp->f.fid.Fid.Volume;
fileFid.Cell = aodp->f.fid.Cell;
if (!fileFid.Fid.Unique)
tvc = afs_LookupVCache(&fileFid, areq, NULL, andp, aname2);
else
tvc = afs_GetVCache(&fileFid, areq, NULL, (struct vcache *)0);
if (tvc && (vType(tvc) == VDIR)) {
ObtainWriteLock(&tvc->lock, 152);
tdc1 = afs_FindDCache(tvc, (afs_size_t) 0);
if (tdc1) {
if (AFS_IS_DISCON_RW) {
#if defined(AFS_DISCON_ENV)
/* If disconnected, we need to fix (not discard) the "..".*/
afs_dir_ChangeFid(tdc1,
"..",
&aodp->f.fid.Fid.Vnode,
&andp->f.fid.Fid.Vnode);
#endif
} else {
ObtainWriteLock(&tdc1->lock, 648);
ZapDCE(tdc1); /* mark as unknown */
DZap(tdc1);
ReleaseWriteLock(&tdc1->lock);
afs_PutDCache(tdc1); /* put it back */
}
}
osi_dnlc_remove(tvc, "..", 0);
ReleaseWriteLock(&tvc->lock);
afs_PutVCache(tvc);
} else if (AFS_IS_DISCON_RW && tvc && (vType(tvc) == VREG)) {
/* XXX - Should tvc not get locked here? */
tvc->f.parent.vnode = andp->f.fid.Fid.Vnode;
tvc->f.parent.unique = andp->f.fid.Fid.Unique;
} else if (tvc) {
/* True we shouldn't come here since tvc SHOULD be a dir, but we
* 'syntactically' need to unless we change the 'if' above...
*/
afs_PutVCache(tvc);
}
}
code = returnCode;
done:
return code;
}
static int
VLDB_Same(struct VenusFid *afid, struct vrequest *areq)
{
struct vrequest treq;
struct afs_conn *tconn;
int i, type = 0;
union {
struct vldbentry tve;
struct nvldbentry ntve;
struct uvldbentry utve;
} *v;
struct volume *tvp;
struct cell *tcell;
char *bp, tbuf[CVBS]; /* biggest volume id is 2^32, ~ 4*10^9 */
unsigned int changed;
struct server *(oldhosts[NMAXNSERVERS]);
AFS_STATCNT(CheckVLDB);
afs_FinalizeReq(areq);
if ((i = afs_InitReq(&treq, afs_osi_credp)))
return DUNNO;
v = afs_osi_Alloc(sizeof(*v));
tcell = afs_GetCell(afid->Cell, READ_LOCK);
bp = afs_cv2string(&tbuf[CVBS], afid->Fid.Volume);
do {
VSleep(2); /* Better safe than sorry. */
tconn =
afs_ConnByMHosts(tcell->cellHosts, tcell->vlport, tcell->cellNum,
&treq, SHARED_LOCK);
if (tconn) {
if (tconn->srvr->server->flags & SNO_LHOSTS) {
type = 0;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
RX_AFS_GLOCK();
} else if (tconn->srvr->server->flags & SYES_LHOSTS) {
type = 1;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
RX_AFS_GLOCK();
} else {
type = 2;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameU(tconn->id, bp, &v->utve);
RX_AFS_GLOCK();
if (!(tconn->srvr->server->flags & SVLSRV_UUID)) {
if (i == RXGEN_OPCODE) {
type = 1;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
RX_AFS_GLOCK();
if (i == RXGEN_OPCODE) {
type = 0;
tconn->srvr->server->flags |= SNO_LHOSTS;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
RX_AFS_GLOCK();
} else if (!i)
tconn->srvr->server->flags |= SYES_LHOSTS;
} else if (!i)
tconn->srvr->server->flags |= SVLSRV_UUID;
}
lastcode = i;
}
} else
i = -1;
} while (afs_Analyze(tconn, i, NULL, &treq, -1, /* no op code for this */
SHARED_LOCK, tcell));
afs_PutCell(tcell, READ_LOCK);
afs_Trace2(afs_iclSetp, CM_TRACE_CHECKVLDB, ICL_TYPE_FID, &afid,
ICL_TYPE_INT32, i);
if (i) {
afs_osi_Free(v, sizeof(*v));
return DUNNO;
}
/* have info, copy into serverHost array */
changed = 0;
tvp = afs_FindVolume(afid, WRITE_LOCK);
if (tvp) {
ObtainWriteLock(&tvp->lock, 107);
for (i = 0; i < NMAXNSERVERS && tvp->serverHost[i]; i++) {
oldhosts[i] = tvp->serverHost[i];
}
if (type == 2) {
InstallUVolumeEntry(tvp, &v->utve, afid->Cell, tcell, &treq);
} else if (type == 1) {
InstallNVolumeEntry(tvp, &v->ntve, afid->Cell);
} else {
InstallVolumeEntry(tvp, &v->tve, afid->Cell);
}
if (i < NMAXNSERVERS && tvp->serverHost[i]) {
changed = 1;
}
for (--i; !changed && i >= 0; i--) {
if (tvp->serverHost[i] != oldhosts[i]) {
changed = 1; /* also happens if prefs change. big deal. */
}
}
ReleaseWriteLock(&tvp->lock);
afs_PutVolume(tvp, WRITE_LOCK);
} else { /* can't find volume */
tvp = afs_GetVolume(afid, &treq, WRITE_LOCK);
if (tvp) {
afs_PutVolume(tvp, WRITE_LOCK);
afs_osi_Free(v, sizeof(*v));
return DIFFERENT;
} else {
afs_osi_Free(v, sizeof(*v));
return DUNNO;
}
}
afs_osi_Free(v, sizeof(*v));
return (changed ? DIFFERENT : SAME);
} /*VLDB_Same */
afs_open(struct vcache **avcp, afs_int32 aflags, afs_ucred_t *acred)
#endif
{
afs_int32 code;
struct vrequest treq;
struct vcache *tvc;
int writing;
struct afs_fakestat_state fakestate;
AFS_STATCNT(afs_open);
if ((code = afs_InitReq(&treq, acred)))
return code;
#ifdef AFS_SGI64_ENV
/* avcpp can be, but is not necesarily, bhp's vnode. */
tvc = VTOAFS(BHV_TO_VNODE(bhv));
#else
tvc = *avcp;
#endif
afs_Trace2(afs_iclSetp, CM_TRACE_OPEN, ICL_TYPE_POINTER, tvc,
ICL_TYPE_INT32, aflags);
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&tvc, &fakestate, &treq);
if (code)
goto done;
code = afs_VerifyVCache(tvc, &treq);
if (code)
goto done;
ObtainReadLock(&tvc->lock);
if (AFS_IS_DISCONNECTED && (afs_DCacheMissingChunks(tvc) != 0)) {
ReleaseReadLock(&tvc->lock);
/* printf("Network is down in afs_open: missing chunks\n"); */
code = ENETDOWN;
goto done;
}
ReleaseReadLock(&tvc->lock);
if (aflags & (FWRITE | FTRUNC))
writing = 1;
else
writing = 0;
if (vType(tvc) == VDIR) {
/* directory */
if (writing) {
code = EISDIR;
goto done;
} else {
if (!afs_AccessOK
(tvc, ((tvc->f.states & CForeign) ? PRSFS_READ : PRSFS_LOOKUP),
&treq, CHECK_MODE_BITS)) {
code = EACCES;
/* printf("afs_Open: no access for dir\n"); */
goto done;
}
}
} else {
#ifdef AFS_SUN5_ENV
if (AFS_NFSXLATORREQ(acred) && (aflags & FREAD)) {
if (!afs_AccessOK
(tvc, PRSFS_READ, &treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
code = EACCES;
goto done;
}
}
#endif
#ifdef AFS_AIX41_ENV
if (aflags & FRSHARE) {
/*
* Hack for AIX 4.1:
* Apparently it is possible for a file to get mapped without
* either VNOP_MAP or VNOP_RDWR being called, if (1) it is a
* sharable library, and (2) it has already been loaded. We must
* ensure that the credp is up to date. We detect the situation
* by checking for O_RSHARE at open time.
*/
/*
* We keep the caller's credentials since an async daemon will
* handle the request at some point. We assume that the same
* credentials will be used.
*/
ObtainWriteLock(&tvc->lock, 140);
if (!tvc->credp || (tvc->credp != acred)) {
crhold(acred);
if (tvc->credp) {
struct ucred *crp = tvc->credp;
tvc->credp = NULL;
crfree(crp);
}
tvc->credp = acred;
}
ReleaseWriteLock(&tvc->lock);
}
#endif
/* normal file or symlink */
osi_FlushText(tvc); /* only needed to flush text if text locked last time */
#ifdef AFS_BOZONLOCK_ENV
afs_BozonLock(&tvc->pvnLock, tvc);
#endif
osi_FlushPages(tvc, acred);
#ifdef AFS_BOZONLOCK_ENV
afs_BozonUnlock(&tvc->pvnLock, tvc);
#endif
}
/* set date on file if open in O_TRUNC mode */
if (aflags & FTRUNC) {
/* this fixes touch */
ObtainWriteLock(&tvc->lock, 123);
tvc->f.m.Date = osi_Time();
tvc->f.states |= CDirty;
ReleaseWriteLock(&tvc->lock);
}
ObtainReadLock(&tvc->lock);
if (writing)
tvc->execsOrWriters++;
tvc->opens++;
#if defined(AFS_SGI_ENV) || defined (AFS_LINUX26_ENV)
if (writing && tvc->cred == NULL) {
crhold(acred);
tvc->cred = acred;
}
#endif
ReleaseReadLock(&tvc->lock);
if ((afs_preCache != 0) && (writing == 0) && (vType(tvc) != VDIR) &&
(!afs_BBusy())) {
struct dcache *tdc;
afs_size_t offset, len;
tdc = afs_GetDCache(tvc, 0, &treq, &offset, &len, 1);
ObtainSharedLock(&tdc->mflock, 865);
if (!(tdc->mflags & DFFetchReq)) {
struct brequest *bp;
/* start the daemon (may already be running, however) */
UpgradeSToWLock(&tdc->mflock, 666);
tdc->mflags |= DFFetchReq; /* guaranteed to be cleared by BKG or
GetDCache */
/* last parm (1) tells bkg daemon to do an afs_PutDCache when it
is done, since we don't want to wait for it to finish before
doing so ourselves.
*/
bp = afs_BQueue(BOP_FETCH, tvc, B_DONTWAIT, 0, acred,
(afs_size_t) 0, (afs_size_t) 1, tdc,
(void *)0, (void *)0);
if (!bp) {
tdc->mflags &= ~DFFetchReq;
}
ReleaseWriteLock(&tdc->mflock);
} else {
ReleaseSharedLock(&tdc->mflock);
}
}
done:
afs_PutFakeStat(&fakestate);
AFS_DISCON_UNLOCK();
code = afs_CheckCode(code, &treq, 4); /* avoid AIX -O bug */
afs_Trace2(afs_iclSetp, CM_TRACE_OPEN, ICL_TYPE_POINTER, tvc,
ICL_TYPE_INT32, 999999);
return code;
}
int
SPAGCB_GetCreds(struct rx_call *a_call, afs_int32 a_uid,
CredInfos *a_creds)
{
struct unixuser *tu;
CredInfo *tci;
int bucket, count, i = 0, clen;
char *cellname;
RX_AFS_GLOCK();
memset(a_creds, 0, sizeof(struct CredInfos));
if ((rx_HostOf(rx_PeerOf(rx_ConnectionOf(a_call))) != afs_nfs_server_addr
|| rx_PortOf(rx_PeerOf(rx_ConnectionOf(a_call))) != htons(7001))
#if 0 /* for debugging ONLY! */
&& rx_PortOf(rx_PeerOf(rx_ConnectionOf(a_call))) != htons(7901)
#endif
) {
RX_AFS_GUNLOCK();
return UAEPERM;
}
ObtainWriteLock(&afs_xuser, 823);
/* count them first */
bucket = UHash(a_uid);
for (count = 0, tu = afs_users[bucket]; tu; tu = tu->next) {
if (tu->uid == a_uid) count++;
}
if (!count) {
ReleaseWriteLock(&afs_xuser);
RX_AFS_GUNLOCK();
return UAESRCH;
}
a_creds->CredInfos_val =
(CredInfo *)afs_osi_Alloc(count * sizeof(CredInfo));
if (!a_creds->CredInfos_val)
goto out;
a_creds->CredInfos_len = count;
memset(a_creds->CredInfos_val, 0, count * sizeof(CredInfo));
for (i = 0, tu = afs_users[bucket]; tu; tu = tu->next, i++) {
if (tu->uid == a_uid && tu->cellinfo &&
(tu->states & UHasTokens) && !(tu->states & UTokensBad)) {
tci = &a_creds->CredInfos_val[i];
tci->vid = tu->vid;
tci->ct.AuthHandle = tu->ct.AuthHandle;
memcpy(tci->ct.HandShakeKey, tu->ct.HandShakeKey, 8);
tci->ct.ViceId = tu->ct.ViceId;
tci->ct.BeginTimestamp = tu->ct.BeginTimestamp;
tci->ct.EndTimestamp = tu->ct.EndTimestamp;
cellname = ((struct afspag_cell *)(tu->cellinfo))->cellname;
clen = strlen(cellname) + 1;
tci->cellname = afs_osi_Alloc(clen);
if (!tci->cellname)
goto out;
memcpy(tci->cellname, cellname, clen);
tci->st.st_len = tu->stLen;
tci->st.st_val = afs_osi_Alloc(tu->stLen);
if (!tci->st.st_val) {
afs_osi_Free(tci->cellname, clen);
goto out;
}
memcpy(tci->st.st_val, tu->stp, tu->stLen);
if (tu->states & UPrimary)
tci->states |= UPrimary;
}
}
ReleaseWriteLock(&afs_xuser);
RX_AFS_GUNLOCK();
return 0;
out:
if (a_creds->CredInfos_val) {
while (i-- > 0) {
afs_osi_Free(a_creds->CredInfos_val[i].st.st_val,
a_creds->CredInfos_val[i].st.st_len);
afs_osi_Free(a_creds->CredInfos_val[i].cellname,
strlen(a_creds->CredInfos_val[i].cellname) + 1);
}
afs_osi_Free(a_creds->CredInfos_val, count * sizeof(CredInfo));
}
ReleaseWriteLock(&afs_xuser);
RX_AFS_GUNLOCK();
return UAENOMEM;
}
void
unlock_Status(void)
{
ReleaseWriteLock(&statusQueueLock);
}
/* release a write lock and sleep on an address, atomically */
void
LWP_WaitProcessW(void *addr, struct Lock *alock)
{
ReleaseWriteLock(alock);
LWP_WaitProcess(addr);
}
int
afs_StoreAllSegments(struct vcache *avc, struct vrequest *areq,
int sync)
{
struct dcache *tdc;
afs_int32 code = 0;
afs_int32 index;
afs_int32 origCBs, foreign = 0;
int hash;
afs_hyper_t newDV, oldDV; /* DV when we start, and finish, respectively */
struct dcache **dcList;
unsigned int i, j, minj, moredata, high, off;
afs_size_t maxStoredLength; /* highest offset we've written to server. */
int safety, marineronce = 0;
AFS_STATCNT(afs_StoreAllSegments);
hset(oldDV, avc->f.m.DataVersion);
hset(newDV, avc->f.m.DataVersion);
hash = DVHash(&avc->f.fid);
foreign = (avc->f.states & CForeign);
dcList = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
afs_Trace2(afs_iclSetp, CM_TRACE_STOREALL, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
#if !defined(AFS_AIX32_ENV) && !defined(AFS_SGI65_ENV)
/* In the aix vm implementation we need to do the vm_writep even
* on the memcache case since that's we adjust the file's size
* and finish flushing partial vm pages.
*/
if ((cacheDiskType != AFS_FCACHE_TYPE_MEM) ||
(sync & AFS_VMSYNC_INVAL) || (sync & AFS_VMSYNC) ||
(sync & AFS_LASTSTORE))
#endif /* !AFS_AIX32_ENV && !AFS_SGI65_ENV */
{
/* If we're not diskless, reading a file may stress the VM
* system enough to cause a pageout, and this vnode would be
* locked when the pageout occurs. We can prevent this problem
* by making sure all dirty pages are already flushed. We don't
* do this when diskless because reading a diskless (i.e.
* memory-resident) chunk doesn't require using new VM, and we
* also don't want to dump more dirty data into a diskless cache,
* since they're smaller, and we might exceed its available
* space.
*/
#if defined(AFS_SUN5_ENV)
if (sync & AFS_VMSYNC_INVAL) /* invalidate VM pages */
osi_VM_TryToSmush(avc, CRED(), 1);
else
#endif
osi_VM_StoreAllSegments(avc);
}
if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
/* This will probably make someone sad ... */
/*printf("Net down in afs_StoreSegments\n");*/
return ENETDOWN;
}
ConvertWToSLock(&avc->lock);
/*
* Subsequent code expects a sorted list, and it expects all the
* chunks in the list to be contiguous, so we need a sort and a
* while loop in here, too - but this will work for a first pass...
* 92.10.05 - OK, there's a sort in here now. It's kind of a modified
* bin sort, I guess. Chunk numbers start with 0
*
* - Have to get a write lock on xdcache because GetDSlot might need it (if
* the chunk doesn't have a dcache struct).
* This seems like overkill in most cases.
* - I'm not sure that it's safe to do "index = .hvNextp", then unlock
* xdcache, then relock xdcache and try to use index. It is done
* a lot elsewhere in the CM, but I'm not buying that argument.
* - should be able to check IFDataMod without doing the GetDSlot (just
* hold afs_xdcache). That way, it's easy to do this without the
* writelock on afs_xdcache, and we save unneccessary disk
* operations. I don't think that works, 'cuz the next pointers
* are still on disk.
*/
origCBs = afs_allCBs;
maxStoredLength = 0;
minj = 0;
do {
memset(dcList, 0, NCHUNKSATONCE * sizeof(struct dcache *));
high = 0;
moredata = FALSE;
/* lock and start over from beginning of hash chain
* in order to avoid a race condition. */
ObtainWriteLock(&afs_xdcache, 284);
index = afs_dvhashTbl[hash];
for (j = 0; index != NULLIDX;) {
if ((afs_indexFlags[index] & IFDataMod)
&& (afs_indexUnique[index] == avc->f.fid.Fid.Unique)) {
tdc = afs_GetValidDSlot(index); /* refcount+1. */
if (!tdc) {
ReleaseWriteLock(&afs_xdcache);
code = EIO;
goto done;
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid) && tdc->f.chunk >= minj) {
off = tdc->f.chunk - minj;
if (off < NCHUNKSATONCE) {
if (dcList[off])
osi_Panic("dclist slot already in use!");
if (afs_mariner && !marineronce) {
/* first chunk only */
afs_MarinerLog("store$Storing", avc);
marineronce++;
}
dcList[off] = tdc;
if (off > high)
high = off;
j++;
/* DCLOCKXXX: chunkBytes is protected by tdc->lock which we
* can't grab here, due to lock ordering with afs_xdcache.
* So, disable this shortcut for now. -- kolya 2001-10-13
*/
/* shortcut: big win for little files */
/* tlen -= tdc->f.chunkBytes;
* if (tlen <= 0)
* break;
*/
} else {
moredata = TRUE;
afs_PutDCache(tdc);
if (j == NCHUNKSATONCE)
break;
}
} else {
afs_PutDCache(tdc);
}
}
index = afs_dvnextTbl[index];
}
ReleaseWriteLock(&afs_xdcache);
/* this guy writes chunks, puts back dcache structs, and bumps newDV */
/* "moredata" just says "there are more dirty chunks yet to come".
*/
if (j) {
code =
afs_CacheStoreVCache(dcList, avc, areq, sync,
minj, high, moredata,
&newDV, &maxStoredLength);
/* Release any zero-length dcache entries in our interval
* that we locked but didn't store back above.
*/
for (j = 0; j <= high; j++) {
tdc = dcList[j];
if (tdc) {
osi_Assert(tdc->f.chunkBytes == 0);
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
}
}
/* if (j) */
minj += NCHUNKSATONCE;
} while (!code && moredata);
done:
UpgradeSToWLock(&avc->lock, 29);
/* send a trivial truncation store if did nothing else */
if (code == 0) {
/*
* Call StoreMini if we haven't written enough data to extend the
* file at the fileserver to the client's notion of the file length.
*/
if ((avc->f.truncPos != AFS_NOTRUNC)
|| ((avc->f.states & CExtendedFile)
&& (maxStoredLength < avc->f.m.Length))) {
code = afs_StoreMini(avc, areq);
if (code == 0)
hadd32(newDV, 1); /* just bumped here, too */
}
avc->f.states &= ~CExtendedFile;
}
/*
* Finally, turn off DWriting, turn on DFEntryMod,
* update f.versionNo.
* A lot of this could be integrated into the loop above
*/
if (!code) {
afs_hyper_t h_unset;
hones(h_unset);
minj = 0;
do {
moredata = FALSE;
memset(dcList, 0,
NCHUNKSATONCE * sizeof(struct dcache *));
/* overkill, but it gets the lock in case GetDSlot needs it */
ObtainWriteLock(&afs_xdcache, 285);
for (j = 0, safety = 0, index = afs_dvhashTbl[hash];
index != NULLIDX && safety < afs_cacheFiles + 2;
index = afs_dvnextTbl[index]) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
/* This is okay; since manipulating the dcaches at this
* point is best-effort. We only get a dcache here to
* increment the dv and turn off DWriting. If we were
* supposed to do that for a dcache, but could not
* due to an I/O error, it just means the dv won't
* be updated so we don't be able to use that cached
* chunk in the future. That's inefficient, but not
* an error. */
continue;
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid)
&& tdc->f.chunk >= minj) {
off = tdc->f.chunk - minj;
if (off < NCHUNKSATONCE) {
/* this is the file, and the correct chunk range */
if (j >= NCHUNKSATONCE)
osi_Panic
("Too many dcache entries in range\n");
dcList[j++] = tdc;
} else {
moredata = TRUE;
afs_PutDCache(tdc);
if (j == NCHUNKSATONCE)
break;
}
} else {
afs_PutDCache(tdc);
}
}
}
ReleaseWriteLock(&afs_xdcache);
for (i = 0; i < j; i++) {
/* Iterate over the dcache entries we collected above */
tdc = dcList[i];
ObtainSharedLock(&tdc->lock, 677);
/* was code here to clear IFDataMod, but it should only be done
* in storedcache and storealldcache.
*/
/* Only increase DV if we had up-to-date data to start with.
* Otherwise, we could be falsely upgrading an old chunk
* (that we never read) into one labelled with the current
* DV #. Also note that we check that no intervening stores
* occurred, otherwise we might mislabel cache information
* for a chunk that we didn't store this time
*/
/* Don't update the version number if it's not yet set. */
if (!hsame(tdc->f.versionNo, h_unset)
&& hcmp(tdc->f.versionNo, oldDV) >= 0) {
if ((!(afs_dvhack || foreign)
&& hsame(avc->f.m.DataVersion, newDV))
|| ((afs_dvhack || foreign)
&& (origCBs == afs_allCBs))) {
/* no error, this is the DV */
UpgradeSToWLock(&tdc->lock, 678);
hset(tdc->f.versionNo, avc->f.m.DataVersion);
tdc->dflags |= DFEntryMod;
/* DWriting may not have gotten cleared above, if all
* we did was a StoreMini */
tdc->f.states &= ~DWriting;
ConvertWToSLock(&tdc->lock);
}
}
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
minj += NCHUNKSATONCE;
} while (moredata);
}
if (code) {
/*
* Invalidate chunks after an error for ccores files since
* afs_inactive won't be called for these and they won't be
* invalidated. Also discard data if it's a permanent error from the
* fileserver.
*/
if (areq->permWriteError || (avc->f.states & CCore)) {
afs_InvalidateAllSegments(avc);
}
}
afs_Trace3(afs_iclSetp, CM_TRACE_STOREALLDONE, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->f.m.Length, ICL_TYPE_INT32, code);
/* would like a Trace5, but it doesn't exist... */
afs_Trace3(afs_iclSetp, CM_TRACE_AVCLOCKER, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->lock.wait_states, ICL_TYPE_INT32,
avc->lock.excl_locked);
afs_Trace4(afs_iclSetp, CM_TRACE_AVCLOCKEE, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->lock.wait_states, ICL_TYPE_INT32,
avc->lock.readers_reading, ICL_TYPE_INT32,
avc->lock.num_waiting);
/*
* Finally, if updated DataVersion matches newDV, we did all of the
* stores. If mapDV indicates that the page cache was flushed up
* to when we started the store, then we can relabel them as flushed
* as recently as newDV.
* Turn off CDirty bit because the stored data is now in sync with server.
*/
if (code == 0 && hcmp(avc->mapDV, oldDV) >= 0) {
if ((!(afs_dvhack || foreign) && hsame(avc->f.m.DataVersion, newDV))
|| ((afs_dvhack || foreign) && (origCBs == afs_allCBs))) {
hset(avc->mapDV, newDV);
avc->f.states &= ~CDirty;
}
}
osi_FreeLargeSpace(dcList);
/* If not the final write a temporary error is ok. */
if (code && !areq->permWriteError && !(sync & AFS_LASTSTORE))
code = 0;
return code;
} /*afs_StoreAllSegments (new 03/02/94) */
/*!
* \brief Load the list of cells from given inode.
* \param inode Source inode.
* \param lookupcode
* \return 0 for success. < 0 for error.
*/
int
afs_cellname_init(afs_dcache_id_t *inode, int lookupcode)
{
struct osi_file *tfile;
int cc, off = 0;
ObtainWriteLock(&afs_xcell, 692);
afs_cellnum_next = 1;
afs_cellname_dirty = 0;
if (cacheDiskType == AFS_FCACHE_TYPE_MEM) {
ReleaseWriteLock(&afs_xcell);
return 0;
}
if (lookupcode) {
ReleaseWriteLock(&afs_xcell);
return lookupcode;
}
tfile = osi_UFSOpen(inode);
if (!tfile) {
ReleaseWriteLock(&afs_xcell);
return EIO;
}
afs_copy_inode(&afs_cellname_inode, inode);
afs_cellname_inode_set = 1;
while (1) {
afs_int32 cellnum, clen, magic;
struct cell_name *cn;
char *cellname;
cc = afs_osi_Read(tfile, off, &magic, sizeof(magic));
if (cc != sizeof(magic))
break;
if (magic != AFS_CELLINFO_MAGIC)
break;
off += cc;
cc = afs_osi_Read(tfile, off, &cellnum, sizeof(cellnum));
if (cc != sizeof(cellnum))
break;
off += cc;
cc = afs_osi_Read(tfile, off, &clen, sizeof(clen));
if (cc != sizeof(clen))
break;
off += cc;
cellname = afs_osi_Alloc(clen + 1);
if (!cellname)
break;
cc = afs_osi_Read(tfile, off, cellname, clen);
if (cc != clen) {
afs_osi_Free(cellname, clen + 1);
break;
}
off += cc;
cellname[clen] = '\0';
if (afs_cellname_lookup_name(cellname)
|| afs_cellname_lookup_id(cellnum)) {
afs_osi_Free(cellname, clen + 1);
break;
}
cn = afs_cellname_new(cellname, cellnum);
afs_osi_Free(cellname, clen + 1);
}
osi_UFSClose(tfile);
ReleaseWriteLock(&afs_xcell);
return 0;
}
/*
* afs_TruncateAllSegments
*
* Description:
* Truncate a cache file.
*
* Parameters:
* avc : Ptr to vcache entry to truncate.
* alen : Number of bytes to make the file.
* areq : Ptr to request structure.
*
* Environment:
* Called with avc write-locked; in VFS40 systems, pvnLock is also
* held.
*/
int
afs_TruncateAllSegments(struct vcache *avc, afs_size_t alen,
struct vrequest *areq, afs_ucred_t *acred)
{
struct dcache *tdc;
afs_int32 code;
afs_int32 index;
afs_size_t newSize;
int dcCount, dcPos;
struct dcache **tdcArray = NULL;
AFS_STATCNT(afs_TruncateAllSegments);
avc->f.m.Date = osi_Time();
afs_Trace3(afs_iclSetp, CM_TRACE_TRUNCALL, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(alen));
if (alen >= avc->f.m.Length) {
/*
* Special speedup since Sun's vm extends the file this way;
* we've never written to the file thus we can just set the new
* length and avoid the needless calls below.
* Also used for ftruncate calls which can extend the file.
* To completely minimize the possible extra StoreMini RPC, we really
* should keep the ExtendedPos as well and clear this flag if we
* truncate below that value before we store the file back.
*/
avc->f.states |= CExtendedFile;
avc->f.m.Length = alen;
return 0;
}
#if (defined(AFS_SUN5_ENV))
/* Zero unused portion of last page */
osi_VM_PreTruncate(avc, alen, acred);
#endif
#if (defined(AFS_SUN5_ENV))
ObtainWriteLock(&avc->vlock, 546);
avc->activeV++; /* Block new getpages */
ReleaseWriteLock(&avc->vlock);
#endif
ReleaseWriteLock(&avc->lock);
AFS_GUNLOCK();
/* Flush pages beyond end-of-file. */
osi_VM_Truncate(avc, alen, acred);
AFS_GLOCK();
ObtainWriteLock(&avc->lock, 79);
avc->f.m.Length = alen;
if (alen < avc->f.truncPos)
avc->f.truncPos = alen;
code = DVHash(&avc->f.fid);
/* block out others from screwing with this table */
ObtainWriteLock(&afs_xdcache, 287);
dcCount = 0;
for (index = afs_dvhashTbl[code]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
ReleaseWriteLock(&afs_xdcache);
code = EIO;
goto done;
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid))
dcCount++;
afs_PutDCache(tdc);
}
index = afs_dvnextTbl[index];
}
/* Now allocate space where we can save those dcache entries, and
* do a second pass over them.. Since we're holding xdcache, it
* shouldn't be changing.
*/
tdcArray = osi_Alloc(dcCount * sizeof(struct dcache *));
dcPos = 0;
for (index = afs_dvhashTbl[code]; index != NULLIDX; index = afs_dvnextTbl[index]) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
/* make sure we put back all of the tdcArray members before
* bailing out */
/* remember, the last valid tdc is at dcPos-1, so start at
* dcPos-1, not at dcPos itself. */
for (dcPos = dcPos - 1; dcPos >= 0; dcPos--) {
tdc = tdcArray[dcPos];
afs_PutDCache(tdc);
}
code = EIO;
goto done;
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
/* same file, and modified, we'll store it back */
if (dcPos < dcCount) {
tdcArray[dcPos++] = tdc;
} else {
afs_PutDCache(tdc);
}
} else {
afs_PutDCache(tdc);
}
}
}
ReleaseWriteLock(&afs_xdcache);
/* Now we loop over the array of dcache entries and truncate them */
for (index = 0; index < dcPos; index++) {
struct osi_file *tfile;
tdc = tdcArray[index];
newSize = alen - AFS_CHUNKTOBASE(tdc->f.chunk);
if (newSize < 0)
newSize = 0;
ObtainSharedLock(&tdc->lock, 672);
if (newSize < tdc->f.chunkBytes && newSize < MAX_AFS_UINT32) {
UpgradeSToWLock(&tdc->lock, 673);
tdc->f.states |= DWriting;
tfile = afs_CFileOpen(&tdc->f.inode);
afs_CFileTruncate(tfile, (afs_int32)newSize);
afs_CFileClose(tfile);
afs_AdjustSize(tdc, (afs_int32)newSize);
if (alen < tdc->validPos) {
if (alen < AFS_CHUNKTOBASE(tdc->f.chunk))
tdc->validPos = 0;
else
tdc->validPos = alen;
}
ConvertWToSLock(&tdc->lock);
}
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
code = 0;
done:
if (tdcArray) {
osi_Free(tdcArray, dcCount * sizeof(struct dcache *));
}
#if (defined(AFS_SUN5_ENV))
ObtainWriteLock(&avc->vlock, 547);
if (--avc->activeV == 0 && (avc->vstates & VRevokeWait)) {
avc->vstates &= ~VRevokeWait;
afs_osi_Wakeup((char *)&avc->vstates);
}
ReleaseWriteLock(&avc->vlock);
#endif
return code;
}
/*!
* Create or update a cell entry.
* \param acellName Name of cell.
* \param acellHosts Array of hosts that this cell has.
* \param aflags Cell flags.
* \param linkedcname
* \param fsport File server port.
* \param vlport Volume server port.
* \param timeout Cell timeout value, 0 means static AFSDB entry.
* \return
*/
afs_int32
afs_NewCell(char *acellName, afs_int32 * acellHosts, int aflags,
char *linkedcname, u_short fsport, u_short vlport, int timeout)
{
struct cell *tc, *tcl = 0;
afs_int32 i, newc = 0, code = 0;
AFS_STATCNT(afs_NewCell);
ObtainWriteLock(&afs_xcell, 103);
tc = afs_FindCellByName_nl(acellName, READ_LOCK);
if (tc) {
aflags &= ~CNoSUID;
} else {
tc = afs_osi_Alloc(sizeof(struct cell));
osi_Assert(tc != NULL);
memset(tc, 0, sizeof(*tc));
tc->cellName = afs_strdup(acellName);
tc->fsport = AFS_FSPORT;
tc->vlport = AFS_VLPORT;
AFS_MD5_String(tc->cellHandle, tc->cellName, strlen(tc->cellName));
AFS_RWLOCK_INIT(&tc->lock, "cell lock");
newc = 1;
aflags |= CNoSUID;
}
ObtainWriteLock(&tc->lock, 688);
/* If the cell we've found has the correct name but no timeout,
* and we're called with a non-zero timeout, bail out: never
* override static configuration entries with AFSDB ones.
* One exception: if the original cell entry had no servers,
* it must get servers from AFSDB.
*/
if (timeout && !tc->timeout && tc->cellHosts[0]) {
code = EEXIST; /* This code is checked for in afs_LookupAFSDB */
goto bad;
}
/* we don't want to keep pinging old vlservers which were down,
* since they don't matter any more. It's easier to do this than
* to remove the server from its various hash tables. */
for (i = 0; i < AFS_MAXCELLHOSTS; i++) {
if (!tc->cellHosts[i])
break;
tc->cellHosts[i]->flags &= ~SRVR_ISDOWN;
tc->cellHosts[i]->flags |= SRVR_ISGONE;
}
if (fsport)
tc->fsport = fsport;
if (vlport)
tc->vlport = vlport;
if (aflags & CLinkedCell) {
if (!linkedcname) {
code = EINVAL;
goto bad;
}
tcl = afs_FindCellByName_nl(linkedcname, READ_LOCK);
if (!tcl) {
code = ENOENT;
goto bad;
}
if (tcl->lcellp) { /* XXX Overwriting if one existed before! XXX */
tcl->lcellp->lcellp = (struct cell *)0;
tcl->lcellp->states &= ~CLinkedCell;
}
tc->lcellp = tcl;
tcl->lcellp = tc;
}
tc->states |= aflags;
tc->timeout = timeout;
memset(tc->cellHosts, 0, sizeof(tc->cellHosts));
for (i = 0; i < AFS_MAXCELLHOSTS; i++) {
/* Get server for each host and link this cell in.*/
struct server *ts;
afs_uint32 temp = acellHosts[i];
if (!temp)
break;
ts = afs_GetServer(&temp, 1, 0, tc->vlport, WRITE_LOCK, NULL, 0);
ts->cell = tc;
ts->flags &= ~SRVR_ISGONE;
/* Set the server as a host of the new cell. */
tc->cellHosts[i] = ts;
afs_PutServer(ts, WRITE_LOCK);
}
afs_SortServers(tc->cellHosts, AFS_MAXCELLHOSTS); /* randomize servers */
/* New cell: Build and add to LRU cell queue. */
if (newc) {
struct cell_name *cn;
cn = afs_cellname_lookup_name(acellName);
if (!cn)
cn = afs_cellname_new(acellName, 0);
tc->cnamep = cn;
tc->cellNum = cn->cellnum;
tc->cellIndex = afs_cellindex++;
afs_stats_cmperf.numCellsVisible++;
QAdd(&CellLRU, &tc->lruq);
}
ReleaseWriteLock(&tc->lock);
ReleaseWriteLock(&afs_xcell);
afs_PutCell(tc, 0);
if (!(aflags & CHush))
afs_DynrootInvalidate();
return 0;
bad:
if (newc) {
afs_osi_FreeStr(tc->cellName);
afs_osi_Free(tc, sizeof(struct cell));
}
ReleaseWriteLock(&tc->lock);
ReleaseWriteLock(&afs_xcell);
return code;
}
/**
* Reset volume name to volume id mapping cache.
* @param flags
*/
void
afs_CheckVolumeNames(int flags)
{
afs_int32 i, j;
struct volume *tv;
unsigned int now;
struct vcache *tvc;
afs_int32 *volumeID, *cellID, vsize, nvols;
#ifdef AFS_DARWIN80_ENV
vnode_t tvp;
#endif
AFS_STATCNT(afs_CheckVolumeNames);
nvols = 0;
volumeID = cellID = NULL;
vsize = 0;
ObtainReadLock(&afs_xvolume);
if (flags & AFS_VOLCHECK_EXPIRED) {
/*
* allocate space to hold the volumeIDs and cellIDs, only if
* we will be invalidating the mountpoints later on
*/
for (i = 0; i < NVOLS; i++)
for (tv = afs_volumes[i]; tv; tv = tv->next)
++vsize;
volumeID = afs_osi_Alloc(2 * vsize * sizeof(*volumeID));
cellID = (volumeID) ? volumeID + vsize : 0;
}
now = osi_Time();
for (i = 0; i < NVOLS; i++) {
for (tv = afs_volumes[i]; tv; tv = tv->next) {
if (flags & AFS_VOLCHECK_EXPIRED) {
if (((tv->expireTime < (now + 10)) && (tv->states & VRO))
|| (flags & AFS_VOLCHECK_FORCE)) {
afs_ResetVolumeInfo(tv); /* also resets status */
if (volumeID) {
volumeID[nvols] = tv->volume;
cellID[nvols] = tv->cell;
}
++nvols;
continue;
}
}
/* ??? */
if (flags & (AFS_VOLCHECK_BUSY | AFS_VOLCHECK_FORCE)) {
for (j = 0; j < AFS_MAXHOSTS; j++)
tv->status[j] = not_busy;
}
}
}
ReleaseReadLock(&afs_xvolume);
/* next ensure all mt points are re-evaluated */
if (nvols || (flags & (AFS_VOLCHECK_FORCE | AFS_VOLCHECK_MTPTS))) {
loop:
ObtainReadLock(&afs_xvcache);
for (i = 0; i < VCSIZE; i++) {
for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
/* if the volume of "mvid" of the vcache entry is among the
* ones we found earlier, then we re-evaluate it. Also, if the
* force bit is set or we explicitly asked to reevaluate the
* mt-pts, we clean the cmvalid bit */
if ((flags & (AFS_VOLCHECK_FORCE | AFS_VOLCHECK_MTPTS))
|| (tvc->mvid
&& inVolList(tvc->mvid, nvols, volumeID, cellID)))
tvc->f.states &= ~CMValid;
/* If the volume that this file belongs to was reset earlier,
* then we should remove its callback.
* Again, if forced, always do it.
*/
if ((tvc->f.states & CRO)
&& (inVolList(&tvc->f.fid, nvols, volumeID, cellID)
|| (flags & AFS_VOLCHECK_FORCE))) {
if (tvc->f.states & CVInit) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop;
}
#ifdef AFS_DARWIN80_ENV
if (tvc->f.states & CDeadVnode) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop;
}
tvp = AFSTOV(tvc);
if (vnode_get(tvp))
continue;
if (vnode_ref(tvp)) {
AFS_GUNLOCK();
/* AFSTOV(tvc) may be NULL */
vnode_put(tvp);
AFS_GLOCK();
continue;
}
#else
AFS_FAST_HOLD(tvc);
#endif
ReleaseReadLock(&afs_xvcache);
ObtainWriteLock(&afs_xcbhash, 485);
/* LOCKXXX: We aren't holding tvc write lock? */
afs_DequeueCallback(tvc);
tvc->f.states &= ~CStatd;
ReleaseWriteLock(&afs_xcbhash);
if (tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR))
osi_dnlc_purgedp(tvc);
#ifdef AFS_DARWIN80_ENV
vnode_put(AFSTOV(tvc));
/* our tvc ptr is still good until now */
AFS_FAST_RELE(tvc);
ObtainReadLock(&afs_xvcache);
#else
ObtainReadLock(&afs_xvcache);
/* our tvc ptr is still good until now */
AFS_FAST_RELE(tvc);
#endif
}
}
}
osi_dnlc_purge(); /* definitely overkill, but it's safer this way. */
ReleaseReadLock(&afs_xvcache);
}
if (volumeID)
afs_osi_Free(volumeID, 2 * vsize * sizeof(*volumeID));
} /*afs_CheckVolumeNames */
afs_int32
DumpDB(struct rx_call *call,
int firstcall, /* 1 - init. 0 - no init */
afs_int32 maxLength,
charListT *charListPtr,
afs_int32 *done)
{
#ifdef AFS_PTHREAD_ENV
pthread_t dumperPid, watcherPid;
pthread_attr_t dumperPid_tattr;
pthread_attr_t watcherPid_tattr;
#else
PROCESS dumperPid, watcherPid;
#endif
int readSize;
afs_int32 code = 0;
if (callPermitted(call) == 0)
ERROR(BUDB_NOTPERMITTED);
ObtainWriteLock(&dumpSyncPtr->ds_lock);
/* If asking for zero bytes, then this is a call to reset the timeToLive
* timer. Reset it if there is a dump in progress.
*/
if (maxLength == 0) {
charListPtr->charListT_val = NULL;
charListPtr->charListT_len = 0;
*done = ((dumpSyncPtr->statusFlags == 0) ? 1 : 0);
/* reset the clock on dump timeout */
dumpSyncPtr->timeToLive = time(0) + DUMP_TTL_INC;
goto error_exit;
}
if (dumpSyncPtr->statusFlags == 0) {
if (!firstcall)
ERROR(BUDB_DUMPFAILED);
LogDebug(5, "Setup dump\n");
/* no dump in progress - setup and retake lock */
memset(dumpSyncPtr, 0, sizeof(*dumpSyncPtr));
/* ObtainWriteLock(&dumpSyncPtr->ds_lock); */
/* mark dump in progress */
dumpSyncPtr->statusFlags = 1;
code = pipe(dumpSyncPtr->pipeFid);
if (code)
ERROR(errno);
#ifdef AFS_PTHREAD_ENV
/* Initialize the condition variables and the mutexes we use
* to signal and synchronize the reader and writer threads.
*/
assert(pthread_cond_init(&dumpSyncPtr->ds_readerStatus_cond, (const pthread_condattr_t *)0) == 0);
assert(pthread_cond_init(&dumpSyncPtr->ds_writerStatus_cond, (const pthread_condattr_t *)0) == 0);
assert(pthread_mutex_init(&dumpSyncPtr->ds_readerStatus_mutex, (const pthread_mutexattr_t *)0) == 0);
assert(pthread_mutex_init(&dumpSyncPtr->ds_writerStatus_mutex, (const pthread_mutexattr_t *)0) == 0);
/* Initialize the thread attributes and launch the thread */
assert(pthread_attr_init(&dumperPid_tattr) == 0);
assert(pthread_attr_setdetachstate(&dumperPid_tattr, PTHREAD_CREATE_DETACHED) == 0);
assert(pthread_create(&dumperPid, &dumperPid_tattr, (void *)setupDbDump, NULL) == 0);
#else
code =
LWP_CreateProcess(setupDbDump, 16384, 1,
(void *)(intptr_t)dumpSyncPtr->pipeFid[1],
"Database Dumper", &dumperPid);
if (code)
goto error_exit;
#endif
dumpSyncPtr->dumperPid = dumperPid;
dumpSyncPtr->timeToLive = time(0) + DUMP_TTL_INC;
#ifdef AFS_PTHREAD_ENV
/* Initialize the thread attributes and launch the thread */
assert(pthread_attr_init(&watcherPid_tattr) == 0);
assert(pthread_attr_setdetachstate(&watcherPid_tattr, PTHREAD_CREATE_DETACHED) == 0);
assert(pthread_create(&watcherPid, &watcherPid_tattr, (void *)dumpWatcher, NULL) == 0);
#else
/* now create the watcher thread */
code =
LWP_CreateProcess(dumpWatcher, 16384, 1, 0,
"Database Dump Watchdog", &watcherPid);
#endif
} else if (firstcall)
ERROR(BUDB_LOCKED);
/* now read the database and feed it to the rpc connection */
/* wait for data */
while (dumpSyncPtr->ds_bytes == 0) {
/* if no more data */
if ((dumpSyncPtr->ds_writerStatus == DS_DONE)
|| (dumpSyncPtr->ds_writerStatus == DS_DONE_ERROR)) {
break;
}
if (dumpSyncPtr->ds_writerStatus == DS_WAITING) {
LogDebug(6, "wakup writer\n");
dumpSyncPtr->ds_writerStatus = 0;
#ifdef AFS_PTHREAD_ENV
assert(pthread_cond_broadcast(&dumpSyncPtr->ds_writerStatus_cond) == 0);
#else
code = LWP_SignalProcess(&dumpSyncPtr->ds_writerStatus);
if (code)
LogError(code, "BUDB_DumpDB: signal delivery failed\n");
#endif
}
LogDebug(6, "wait for writer\n");
dumpSyncPtr->ds_readerStatus = DS_WAITING;
ReleaseWriteLock(&dumpSyncPtr->ds_lock);
#ifdef AFS_PTHREAD_ENV
assert(pthread_mutex_lock(&dumpSyncPtr->ds_readerStatus_mutex) == 0);
assert(pthread_cond_wait(&dumpSyncPtr->ds_readerStatus_cond, &dumpSyncPtr->ds_readerStatus_mutex) == 0);
assert(pthread_mutex_unlock(&dumpSyncPtr->ds_readerStatus_mutex) == 0);
#else
LWP_WaitProcess(&dumpSyncPtr->ds_readerStatus);
#endif
ObtainWriteLock(&dumpSyncPtr->ds_lock);
}
charListPtr->charListT_val = (char *)malloc(maxLength);
readSize =
read(dumpSyncPtr->pipeFid[0], charListPtr->charListT_val, maxLength);
/* reset the clock on dump timeout */
dumpSyncPtr->timeToLive = time(0) + DUMP_TTL_INC;
LogDebug(4, "read of len %d returned %d\n", maxLength, readSize);
charListPtr->charListT_len = readSize;
if (readSize == 0) { /* last chunk */
*done = 1;
close(dumpSyncPtr->pipeFid[0]);
dumpSyncPtr->statusFlags = 0;
} else
*done = 0;
dumpSyncPtr->ds_bytes -= readSize;
if (dumpSyncPtr->ds_writerStatus == DS_WAITING) {
dumpSyncPtr->ds_writerStatus = 0;
#ifdef AFS_PTHREAD_ENV
assert(pthread_cond_broadcast(&dumpSyncPtr->ds_writerStatus_cond) == 0);
#else
code = LWP_SignalProcess(&dumpSyncPtr->ds_writerStatus);
if (code)
LogError(code, "BUDB_DumpDB: signal delivery failed\n");
#endif
}
error_exit:
if (!code && (dumpSyncPtr->ds_writerStatus == DS_DONE_ERROR))
code = -1;
ReleaseWriteLock(&dumpSyncPtr->ds_lock);
return (code);
}
/* question: does afs_create need to set CDirty in the adp or the avc?
* I think we can get away without it, but I'm not sure. Note that
* afs_setattr is called in here for truncation.
*/
#ifdef AFS_SGI64_ENV
int
afs_create(OSI_VC_DECL(adp), char *aname, struct vattr *attrs, int flags,
int amode, struct vcache **avcp, afs_ucred_t *acred)
#else /* AFS_SGI64_ENV */
int
afs_create(OSI_VC_DECL(adp), char *aname, struct vattr *attrs,
enum vcexcl aexcl, int amode, struct vcache **avcp,
afs_ucred_t *acred)
#endif /* AFS_SGI64_ENV */
{
afs_int32 origCBs, origZaps, finalZaps;
struct vrequest *treq = NULL;
afs_int32 code;
struct afs_conn *tc;
struct VenusFid newFid;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus *OutFidStatus, *OutDirStatus;
struct AFSVolSync tsync;
struct AFSCallBack CallBack;
afs_int32 now;
struct dcache *tdc;
afs_size_t offset, len;
struct server *hostp = 0;
struct vcache *tvc;
struct volume *volp = 0;
struct afs_fakestat_state fakestate;
struct rx_connection *rxconn;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_create);
OutFidStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
OutDirStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
memset(&InStatus, 0, sizeof(InStatus));
if ((code = afs_CreateReq(&treq, acred)))
goto done2;
afs_Trace3(afs_iclSetp, CM_TRACE_CREATE, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname, ICL_TYPE_INT32, amode);
afs_InitFakeStat(&fakestate);
#ifdef AFS_SGI65_ENV
/* If avcp is passed not null, it's the old reference to this file.
* We can use this to avoid create races. For now, just decrement
* the reference count on it.
*/
if (*avcp) {
AFS_RELE(AFSTOV(*avcp));
*avcp = NULL;
}
#endif
if (strlen(aname) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done3;
}
if (!afs_ENameOK(aname)) {
code = EINVAL;
goto done3;
}
switch (attrs->va_type) {
case VBLK:
case VCHR:
#if !defined(AFS_SUN5_ENV)
case VSOCK:
#endif
case VFIFO:
/* We don't support special devices or FIFOs */
code = EINVAL;
goto done3;
default:
;
}
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, treq);
if (code)
goto done;
tagain:
code = afs_VerifyVCache(adp, treq);
if (code)
goto done;
/** If the volume is read-only, return error without making an RPC to the
* fileserver
*/
if (adp->f.states & CRO) {
code = EROFS;
goto done;
}
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, treq, &offset, &len, 1);
ObtainWriteLock(&adp->lock, 135);
if (tdc)
ObtainSharedLock(&tdc->lock, 630);
/*
* Make sure that the data in the cache is current. We may have
* received a callback while we were waiting for the write lock.
*/
if (!(adp->f.states & CStatd)
|| (tdc && !hsame(adp->f.m.DataVersion, tdc->f.versionNo))) {
ReleaseWriteLock(&adp->lock);
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto tagain;
}
if (tdc) {
/* see if file already exists. If it does, we only set
* the size attributes (to handle O_TRUNC) */
code = afs_dir_Lookup(tdc, aname, &newFid.Fid); /* use dnlc first xxx */
if (code == 0) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
ReleaseWriteLock(&adp->lock);
#ifdef AFS_SGI64_ENV
if (flags & VEXCL) {
#else
if (aexcl != NONEXCL) {
#endif
code = EEXIST; /* file exists in excl mode open */
goto done;
}
/* found the file, so use it */
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
tvc = NULL;
if (newFid.Fid.Unique == 0) {
tvc = afs_LookupVCache(&newFid, treq, NULL, adp, aname);
}
if (!tvc) /* lookup failed or wasn't called */
tvc = afs_GetVCache(&newFid, treq, NULL, NULL);
if (tvc) {
/* if the thing exists, we need the right access to open it.
* we must check that here, since no other checks are
* made by the open system call */
len = attrs->va_size; /* only do the truncate */
/*
* We used to check always for READ access before; the
* problem is that we will fail if the existing file
* has mode -w-w-w, which is wrong.
*/
if ((amode & VREAD)
&& !afs_AccessOK(tvc, PRSFS_READ, treq, CHECK_MODE_BITS)) {
afs_PutVCache(tvc);
code = EACCES;
goto done;
}
#if defined(AFS_DARWIN80_ENV)
if ((amode & VWRITE) || VATTR_IS_ACTIVE(attrs, va_data_size))
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
if ((amode & VWRITE) || (attrs->va_mask & AT_SIZE))
#else
if ((amode & VWRITE) || len != 0xffffffff)
#endif
{
/* needed for write access check */
tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
tvc->f.parent.unique = adp->f.fid.Fid.Unique;
/* need write mode for these guys */
if (!afs_AccessOK
(tvc, PRSFS_WRITE, treq, CHECK_MODE_BITS)) {
afs_PutVCache(tvc);
code = EACCES;
goto done;
}
}
#if defined(AFS_DARWIN80_ENV)
if (VATTR_IS_ACTIVE(attrs, va_data_size))
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
if (attrs->va_mask & AT_SIZE)
#else
if (len != 0xffffffff)
#endif
{
if (vType(tvc) != VREG) {
afs_PutVCache(tvc);
code = EISDIR;
goto done;
}
/* do a truncate */
#if defined(AFS_DARWIN80_ENV)
VATTR_INIT(attrs);
VATTR_SET_SUPPORTED(attrs, va_data_size);
VATTR_SET_ACTIVE(attrs, va_data_size);
#elif defined(UKERNEL)
attrs->va_mask = ATTR_SIZE;
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
attrs->va_mask = AT_SIZE;
#else
VATTR_NULL(attrs);
#endif
attrs->va_size = len;
ObtainWriteLock(&tvc->lock, 136);
tvc->f.states |= CCreating;
ReleaseWriteLock(&tvc->lock);
#if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
#if defined(AFS_SGI64_ENV)
code =
afs_setattr(VNODE_TO_FIRST_BHV((vnode_t *) tvc),
attrs, 0, acred);
#else
code = afs_setattr(tvc, attrs, 0, acred);
#endif /* AFS_SGI64_ENV */
#else /* SUN5 || SGI */
code = afs_setattr(tvc, attrs, acred);
#endif /* SUN5 || SGI */
ObtainWriteLock(&tvc->lock, 137);
tvc->f.states &= ~CCreating;
ReleaseWriteLock(&tvc->lock);
if (code) {
afs_PutVCache(tvc);
goto done;
}
}
*avcp = tvc;
} else
code = ENOENT; /* shouldn't get here */
/* make sure vrefCount bumped only if code == 0 */
goto done;
}
}
/* if we create the file, we don't do any access checks, since
* that's how O_CREAT is supposed to work */
if (adp->f.states & CForeign) {
origCBs = afs_allCBs;
origZaps = afs_allZaps;
} else {
origCBs = afs_evenCBs; /* if changes, we don't really have a callback */
origZaps = afs_evenZaps; /* number of even numbered vnodes discarded */
}
InStatus.Mask = AFS_SETMODTIME | AFS_SETMODE | AFS_SETGROUP;
InStatus.ClientModTime = osi_Time();
InStatus.Group = (afs_int32) afs_cr_gid(acred);
if (AFS_NFSXLATORREQ(acred)) {
/*
* XXX The following is mainly used to fix a bug in the HP-UX
* nfs client where they create files with mode of 0 without
* doing any setattr later on to fix it. * XXX
*/
#if defined(AFS_AIX_ENV)
if (attrs->va_mode != -1) {
#else
#if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
if (attrs->va_mask & AT_MODE) {
#else
if (attrs->va_mode != ((unsigned short)-1)) {
#endif
#endif
if (!attrs->va_mode)
attrs->va_mode = 0x1b6; /* XXX default mode: rw-rw-rw XXX */
}
}
if (!AFS_IS_DISCONNECTED) {
/* If not disconnected, connect to the server.*/
InStatus.UnixModeBits = attrs->va_mode & 0xffff; /* only care about protection bits */
do {
tc = afs_Conn(&adp->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
hostp = tc->srvr->server; /* remember for callback processing */
now = osi_Time();
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_CREATEFILE);
RX_AFS_GUNLOCK();
code =
RXAFS_CreateFile(rxconn, (struct AFSFid *)&adp->f.fid.Fid,
aname, &InStatus, (struct AFSFid *)
&newFid.Fid, OutFidStatus, OutDirStatus,
&CallBack, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
CallBack.ExpirationTime += now;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &adp->f.fid, treq, AFS_STATS_FS_RPCIDX_CREATEFILE,
SHARED_LOCK, NULL));
if ((code == EEXIST || code == UAEEXIST) &&
#ifdef AFS_SGI64_ENV
!(flags & VEXCL)
#else /* AFS_SGI64_ENV */
aexcl == NONEXCL
#endif
) {
/* if we get an EEXIST in nonexcl mode, just do a lookup */
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
ReleaseWriteLock(&adp->lock);
#if defined(AFS_SGI64_ENV)
code = afs_lookup(VNODE_TO_FIRST_BHV((vnode_t *) adp), aname, avcp,
NULL, 0, NULL, acred);
#elif defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
code = afs_lookup(adp, aname, avcp, NULL, 0, NULL, acred);
#elif defined(UKERNEL)
code = afs_lookup(adp, aname, avcp, acred, 0);
#elif !defined(AFS_DARWIN_ENV)
code = afs_lookup(adp, aname, avcp, acred);
#endif
goto done;
}
if (code) {
if (code < 0) {
ObtainWriteLock(&afs_xcbhash, 488);
afs_DequeueCallback(adp);
adp->f.states &= ~CStatd;
ReleaseWriteLock(&afs_xcbhash);
osi_dnlc_purgedp(adp);
}
ReleaseWriteLock(&adp->lock);
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto done;
}
} else {
/* Generate a fake FID for disconnected mode. */
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
afs_GenFakeFid(&newFid, VREG, 1);
} /* if (!AFS_IS_DISCON_RW) */
/* otherwise, we should see if we can make the change to the dir locally */
if (tdc)
UpgradeSToWLock(&tdc->lock, 631);
if (AFS_IS_DISCON_RW || afs_LocalHero(adp, tdc, OutDirStatus, 1)) {
/* we can do it locally */
ObtainWriteLock(&afs_xdcache, 291);
code = afs_dir_Create(tdc, aname, &newFid.Fid);
ReleaseWriteLock(&afs_xdcache);
if (code) {
ZapDCE(tdc);
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
if (AFS_IS_DISCON_RW)
adp->f.m.LinkCount++;
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
ReleaseWriteLock(&adp->lock);
volp = afs_FindVolume(&newFid, READ_LOCK);
/* New tricky optimistic callback handling algorithm for file creation works
* as follows. We create the file essentially with no locks set at all. File
* server may thus handle operations from others cache managers as well as from
* this very own cache manager that reference the file in question before
* we managed to create the cache entry. However, if anyone else changes
* any of the status information for a file, we'll see afs_evenCBs increase
* (files always have even fids). If someone on this workstation manages
* to do something to the file, they'll end up having to create a cache
* entry for the new file. Either we'll find it once we've got the afs_xvcache
* lock set, or it was also *deleted* the vnode before we got there, in which case
* we will find evenZaps has changed, too. Thus, we only assume we have the right
* status information if no callbacks or vnode removals have occurred to even
* numbered files from the time the call started until the time that we got the xvcache
* lock set. Of course, this also assumes that any call that modifies a file first
* gets a write lock on the file's vnode, but if that weren't true, the whole cache manager
* would fail, since no call would be able to update the local vnode status after modifying
* a file on a file server. */
ObtainWriteLock(&afs_xvcache, 138);
if (adp->f.states & CForeign)
finalZaps = afs_allZaps; /* do this before calling newvcache */
else
finalZaps = afs_evenZaps; /* do this before calling newvcache */
/* don't need to call RemoveVCB, since only path leaving a callback is the
* one where we pass through afs_NewVCache. Can't have queued a VCB unless
* we created and freed an entry between file creation time and here, and the
* freeing of the vnode will change evenZaps. Don't need to update the VLRU
* queue, since the find will only succeed in the event of a create race, and
* then the vcache will be at the front of the VLRU queue anyway... */
if (!(tvc = afs_FindVCache(&newFid, 0, DO_STATS))) {
tvc = afs_NewVCache(&newFid, hostp);
if (tvc) {
int finalCBs;
ObtainWriteLock(&tvc->lock, 139);
ObtainWriteLock(&afs_xcbhash, 489);
finalCBs = afs_evenCBs;
/* add the callback in */
if (adp->f.states & CForeign) {
tvc->f.states |= CForeign;
finalCBs = afs_allCBs;
}
if (origCBs == finalCBs && origZaps == finalZaps) {
tvc->f.states |= CStatd; /* we've fake entire thing, so don't stat */
tvc->f.states &= ~CBulkFetching;
if (!AFS_IS_DISCON_RW) {
tvc->cbExpires = CallBack.ExpirationTime;
afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), volp);
}
} else {
afs_DequeueCallback(tvc);
tvc->f.states &= ~(CStatd | CUnique);
tvc->callback = 0;
if (tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR))
osi_dnlc_purgedp(tvc);
}
ReleaseWriteLock(&afs_xcbhash);
if (AFS_IS_DISCON_RW) {
afs_DisconAddDirty(tvc, VDisconCreate, 0);
afs_GenDisconStatus(adp, tvc, &newFid, attrs, treq, VREG);
} else {
afs_ProcessFS(tvc, OutFidStatus, treq);
}
tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
tvc->f.parent.unique = adp->f.fid.Fid.Unique;
#if !defined(UKERNEL)
if (volp && (volp->states & VPartVisible))
tvc->f.states |= CPartVisible;
#endif
ReleaseWriteLock(&tvc->lock);
*avcp = tvc;
code = 0;
} else
code = ENOENT;
} else {
/* otherwise cache entry already exists, someone else must
* have created it. Comments used to say: "don't need write
* lock to *clear* these flags" but we should do it anyway.
* Code used to clear stat bit and callback, but I don't see
* the point -- we didn't have a create race, somebody else just
* snuck into NewVCache before we got here, probably a racing
* lookup.
*/
*avcp = tvc;
code = 0;
}
ReleaseWriteLock(&afs_xvcache);
done:
AFS_DISCON_UNLOCK();
done3:
if (volp)
afs_PutVolume(volp, READ_LOCK);
if (code == 0) {
if (afs_mariner)
afs_AddMarinerName(aname, *avcp);
/* return the new status in vattr */
afs_CopyOutAttrs(*avcp, attrs);
if (afs_mariner)
afs_MarinerLog("store$Creating", *avcp);
}
afs_PutFakeStat(&fakestate);
code = afs_CheckCode(code, treq, 20);
afs_DestroyReq(treq);
done2:
osi_FreeSmallSpace(OutFidStatus);
osi_FreeSmallSpace(OutDirStatus);
return code;
}
/*
* Check to see if we can track the change locally: requires that
* we have sufficiently recent info in data cache. If so, we
* know the new DataVersion number, and place it correctly in both the
* data and stat cache entries. This routine returns 1 if we should
* do the operation locally, and 0 otherwise.
*
* This routine must be called with the stat cache entry write-locked,
* and dcache entry write-locked.
*/
int
afs_LocalHero(struct vcache *avc, struct dcache *adc,
AFSFetchStatus * astat, int aincr)
{
afs_int32 ok;
afs_hyper_t avers;
AFS_STATCNT(afs_LocalHero);
hset64(avers, astat->dataVersionHigh, astat->DataVersion);
/* avers *is* the version number now, no matter what */
if (adc) {
/* does what's in the dcache *now* match what's in the vcache *now*,
* and do we have a valid callback? if not, our local copy is not "ok" */
ok = (hsame(avc->f.m.DataVersion, adc->f.versionNo) && avc->callback
&& (avc->f.states & CStatd) && avc->cbExpires >= osi_Time());
} else {
ok = 0;
}
if (ok) {
/* check that the DV on the server is what we expect it to be */
afs_hyper_t newDV;
hset(newDV, adc->f.versionNo);
hadd32(newDV, aincr);
if (!hsame(avers, newDV)) {
ok = 0;
}
}
#if defined(AFS_SGI_ENV)
osi_Assert(avc->v.v_type == VDIR);
#endif
/* The bulk status code used the length as a sequence number. */
/* Don't update the vcache entry unless the stats are current. */
if (avc->f.states & CStatd) {
hset(avc->f.m.DataVersion, avers);
#ifdef AFS_64BIT_CLIENT
FillInt64(avc->f.m.Length, astat->Length_hi, astat->Length);
#else /* AFS_64BIT_CLIENT */
avc->f.m.Length = astat->Length;
#endif /* AFS_64BIT_CLIENT */
avc->f.m.Date = astat->ClientModTime;
}
if (ok) {
/* we've been tracking things correctly */
adc->dflags |= DFEntryMod;
adc->f.versionNo = avers;
return 1;
} else {
if (adc) {
ZapDCE(adc);
DZap(adc);
}
if (avc->f.states & CStatd) {
osi_dnlc_purgedp(avc);
}
return 0;
}
}
/* Note that we don't set CDirty here, this is OK because the unlink
* RPC is called synchronously */
int
afs_remove(OSI_VC_DECL(adp), char *aname, afs_ucred_t *acred)
{
struct vrequest treq;
register struct dcache *tdc;
struct VenusFid unlinkFid;
register afs_int32 code;
register struct vcache *tvc;
afs_size_t offset, len;
struct afs_fakestat_state fakestate;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_remove);
afs_Trace2(afs_iclSetp, CM_TRACE_REMOVE, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname);
if ((code = afs_InitReq(&treq, acred))) {
return code;
}
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, &treq);
if (code)
goto done;
/* Check if this is dynroot */
if (afs_IsDynroot(adp)) {
code = afs_DynrootVOPRemove(adp, acred, aname);
goto done;
}
if (afs_IsDynrootMount(adp)) {
code = ENOENT;
goto done;
}
if (strlen(aname) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done;
}
tagain:
code = afs_VerifyVCache(adp, &treq);
tvc = NULL;
if (code) {
code = afs_CheckCode(code, &treq, 23);
goto done;
}
/** If the volume is read-only, return error without making an RPC to the
* fileserver
*/
if (adp->f.states & CRO) {
code = EROFS;
goto done;
}
/* If we're running disconnected without logging, go no further... */
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, &treq, &offset, &len, 1); /* test for error below */
ObtainWriteLock(&adp->lock, 142);
if (tdc)
ObtainSharedLock(&tdc->lock, 638);
/*
* Make sure that the data in the cache is current. We may have
* received a callback while we were waiting for the write lock.
*/
if (!(adp->f.states & CStatd)
|| (tdc && !hsame(adp->f.m.DataVersion, tdc->f.versionNo))) {
ReleaseWriteLock(&adp->lock);
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto tagain;
}
unlinkFid.Fid.Vnode = 0;
if (!tvc) {
tvc = osi_dnlc_lookup(adp, aname, WRITE_LOCK);
}
/* This should not be necessary since afs_lookup() has already
* done the work.
*/
if (!tvc)
if (tdc) {
code = afs_dir_Lookup(tdc, aname, &unlinkFid.Fid);
if (code == 0) {
afs_int32 cached = 0;
unlinkFid.Cell = adp->f.fid.Cell;
unlinkFid.Fid.Volume = adp->f.fid.Fid.Volume;
if (unlinkFid.Fid.Unique == 0) {
tvc =
afs_LookupVCache(&unlinkFid, &treq, &cached, adp,
aname);
} else {
ObtainReadLock(&afs_xvcache);
tvc = afs_FindVCache(&unlinkFid, 0, DO_STATS);
ReleaseReadLock(&afs_xvcache);
}
}
}
if (AFS_IS_DISCON_RW) {
if (!adp->f.shadow.vnode && !(adp->f.ddirty_flags & VDisconCreate)) {
/* Make shadow copy of parent dir. */
afs_MakeShadowDir(adp, tdc);
}
/* Can't hold a dcache lock whilst we're getting a vcache one */
if (tdc)
ReleaseSharedLock(&tdc->lock);
/* XXX - We're holding adp->lock still, and we've got no
* guarantee about whether the ordering matches the lock hierarchy */
ObtainWriteLock(&tvc->lock, 713);
/* If we were locally created, then we don't need to do very
* much beyond ensuring that we don't exist anymore */
if (tvc->f.ddirty_flags & VDisconCreate) {
afs_DisconRemoveDirty(tvc);
} else {
/* Add removed file vcache to dirty list. */
afs_DisconAddDirty(tvc, VDisconRemove, 1);
}
adp->f.m.LinkCount--;
ReleaseWriteLock(&tvc->lock);
if (tdc)
ObtainSharedLock(&tdc->lock, 714);
}
if (tvc && osi_Active(tvc)) {
/* about to delete whole file, prefetch it first */
ReleaseWriteLock(&adp->lock);
if (tdc)
ReleaseSharedLock(&tdc->lock);
ObtainWriteLock(&tvc->lock, 143);
FetchWholeEnchilada(tvc, &treq);
ReleaseWriteLock(&tvc->lock);
ObtainWriteLock(&adp->lock, 144);
/* Technically I don't think we need this back, but let's hold it
anyway; The "got" reference should actually be sufficient. */
if (tdc)
ObtainSharedLock(&tdc->lock, 640);
}
osi_dnlc_remove(adp, aname, tvc);
Tadp1 = adp;
#ifndef AFS_DARWIN80_ENV
Tadpr = VREFCOUNT(adp);
#endif
Ttvc = tvc;
Tnam = aname;
Tnam1 = 0;
#ifndef AFS_DARWIN80_ENV
if (tvc)
Ttvcr = VREFCOUNT(tvc);
#endif
#ifdef AFS_AIX_ENV
if (tvc && VREFCOUNT_GT(tvc, 2) && tvc->opens > 0
&& !(tvc->f.states & CUnlinked)) {
#else
if (tvc && VREFCOUNT_GT(tvc, 1) && tvc->opens > 0
&& !(tvc->f.states & CUnlinked)) {
#endif
char *unlname = afs_newname();
ReleaseWriteLock(&adp->lock);
if (tdc)
ReleaseSharedLock(&tdc->lock);
code = afsrename(adp, aname, adp, unlname, acred, &treq);
Tnam1 = unlname;
if (!code) {
struct VenusFid *oldmvid = NULL;
if (tvc->mvid)
oldmvid = tvc->mvid;
tvc->mvid = (struct VenusFid *)unlname;
if (oldmvid)
osi_FreeSmallSpace(oldmvid);
crhold(acred);
if (tvc->uncred) {
crfree(tvc->uncred);
}
tvc->uncred = acred;
tvc->f.states |= CUnlinked;
/* if rename succeeded, remove should not */
ObtainWriteLock(&tvc->lock, 715);
if (tvc->f.ddirty_flags & VDisconRemove) {
tvc->f.ddirty_flags &= ~VDisconRemove;
}
ReleaseWriteLock(&tvc->lock);
} else {
osi_FreeSmallSpace(unlname);
}
if (tdc)
afs_PutDCache(tdc);
afs_PutVCache(tvc);
} else {
code = afsremove(adp, tdc, tvc, aname, acred, &treq);
}
done:
afs_PutFakeStat(&fakestate);
#ifndef AFS_DARWIN80_ENV
/* we can't track by thread, it's not exported in the KPI; only do
this on !macos */
osi_Assert(!WriteLocked(&adp->lock) || (adp->lock.pid_writer != MyPidxx));
#endif
AFS_DISCON_UNLOCK();
return code;
}
/* afs_remunlink -- This tries to delete the file at the server after it has
* been renamed when unlinked locally but now has been finally released.
*
* CAUTION -- may be called with avc unheld. */
int
afs_remunlink(register struct vcache *avc, register int doit)
{
afs_ucred_t *cred;
char *unlname;
struct vcache *adp;
struct vrequest treq;
struct VenusFid dirFid;
register struct dcache *tdc;
afs_int32 code = 0;
if (NBObtainWriteLock(&avc->lock, 423))
return 0;
#if defined(AFS_DARWIN80_ENV)
if (vnode_get(AFSTOV(avc))) {
ReleaseWriteLock(&avc->lock);
return 0;
}
#endif
if (avc->mvid && (doit || (avc->f.states & CUnlinkedDel))) {
if ((code = afs_InitReq(&treq, avc->uncred))) {
ReleaseWriteLock(&avc->lock);
} else {
/* Must bump the refCount because GetVCache may block.
* Also clear mvid so no other thread comes here if we block.
*/
unlname = (char *)avc->mvid;
avc->mvid = NULL;
cred = avc->uncred;
avc->uncred = NULL;
#if defined(AFS_DARWIN_ENV) && !defined(AFS_DARWIN80_ENV)
VREF(AFSTOV(avc));
#else
AFS_FAST_HOLD(avc);
#endif
/* We'll only try this once. If it fails, just release the vnode.
* Clear after doing hold so that NewVCache doesn't find us yet.
*/
avc->f.states &= ~(CUnlinked | CUnlinkedDel);
ReleaseWriteLock(&avc->lock);
dirFid.Cell = avc->f.fid.Cell;
dirFid.Fid.Volume = avc->f.fid.Fid.Volume;
dirFid.Fid.Vnode = avc->f.parent.vnode;
dirFid.Fid.Unique = avc->f.parent.unique;
adp = afs_GetVCache(&dirFid, &treq, NULL, NULL);
if (adp) {
tdc = afs_FindDCache(adp, (afs_size_t) 0);
ObtainWriteLock(&adp->lock, 159);
if (tdc)
ObtainSharedLock(&tdc->lock, 639);
/* afsremove releases the adp & tdc locks, and does vn_rele(avc) */
code = afsremove(adp, tdc, avc, unlname, cred, &treq);
afs_PutVCache(adp);
} else {
/* we failed - and won't be back to try again. */
afs_PutVCache(avc);
}
osi_FreeSmallSpace(unlname);
crfree(cred);
}
} else {
#if defined(AFS_DARWIN80_ENV)
vnode_put(AFSTOV(avc));
#endif
ReleaseWriteLock(&avc->lock);
}
return code;
}
