static int
afs_StoreMini(struct vcache *avc, struct vrequest *areq)
{
struct afs_conn *tc;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus OutStatus;
struct AFSVolSync tsync;
afs_int32 code;
struct rx_call *tcall;
struct rx_connection *rxconn;
afs_size_t tlen, xlen = 0;
XSTATS_DECLS;
AFS_STATCNT(afs_StoreMini);
afs_Trace2(afs_iclSetp, CM_TRACE_STOREMINI, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->f.m.Length);
tlen = avc->f.m.Length;
if (avc->f.truncPos < tlen)
tlen = avc->f.truncPos;
avc->f.truncPos = AFS_NOTRUNC;
avc->f.states &= ~CExtendedFile;
do {
tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
#ifdef AFS_64BIT_CLIENT
retry:
#endif
RX_AFS_GUNLOCK();
tcall = rx_NewCall(rxconn);
RX_AFS_GLOCK();
/* Set the client mod time since we always want the file
* to have the client's mod time and not the server's one
* (to avoid problems with make, etc.) It almost always
* works fine with standard afs because them server/client
* times are in sync and more importantly this storemini
* it's a special call that would typically be followed by
* the proper store-data or store-status calls.
*/
InStatus.Mask = AFS_SETMODTIME;
InStatus.ClientModTime = avc->f.m.Date;
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_STOREDATA);
afs_Trace4(afs_iclSetp, CM_TRACE_STOREDATA64, ICL_TYPE_FID,
&avc->f.fid.Fid, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length), ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(xlen), ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(tlen));
RX_AFS_GUNLOCK();
#ifdef AFS_64BIT_CLIENT
if (!afs_serverHasNo64Bit(tc)) {
code =
StartRXAFS_StoreData64(tcall,
(struct AFSFid *)&avc->f.fid.Fid,
&InStatus, avc->f.m.Length,
(afs_size_t) 0, tlen);
} else {
afs_int32 l1, l2;
l1 = avc->f.m.Length;
l2 = tlen;
if ((avc->f.m.Length > 0x7fffffff) ||
(tlen > 0x7fffffff) ||
((0x7fffffff - tlen) < avc->f.m.Length)) {
code = EFBIG;
goto error;
}
code =
StartRXAFS_StoreData(tcall,
(struct AFSFid *)&avc->f.fid.Fid,
&InStatus, l1, 0, l2);
}
#else /* AFS_64BIT_CLIENT */
code =
StartRXAFS_StoreData(tcall, (struct AFSFid *)&avc->f.fid.Fid,
&InStatus, avc->f.m.Length, 0, tlen);
#endif /* AFS_64BIT_CLIENT */
if (code == 0) {
code = EndRXAFS_StoreData(tcall, &OutStatus, &tsync);
}
#ifdef AFS_64BIT_CLIENT
error:
#endif
code = rx_EndCall(tcall, code);
RX_AFS_GLOCK();
XSTATS_END_TIME;
#ifdef AFS_64BIT_CLIENT
if (code == RXGEN_OPCODE && !afs_serverHasNo64Bit(tc)) {
afs_serverSetNo64Bit(tc);
goto retry;
}
#endif /* AFS_64BIT_CLIENT */
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &avc->f.fid, areq, AFS_STATS_FS_RPCIDX_STOREDATA,
SHARED_LOCK, NULL));
if (code == 0)
afs_ProcessFS(avc, &OutStatus, areq);
return code;
} /*afs_StoreMini */
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;
}
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) */
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;
}
/*------------------------------------------------------------------------
* EXPORTED afs_Analyze
*
* Description:
* Analyze the outcome of an RPC operation, taking whatever support
* actions are necessary.
*
* Arguments:
* aconn : Ptr to the relevant connection on which the call was made.
* acode : The return code experienced by the RPC.
* afid : The FID of the file involved in the action. This argument
* may be null if none was involved.
* areq : The request record associated with this operation.
* op : which RPC we are analyzing.
* cellp : pointer to a cell struct. Must provide either fid or cell.
*
* Returns:
* Non-zero value if the related RPC operation should be retried,
* zero otherwise.
*
* Environment:
* This routine is typically called in a do-while loop, causing the
* embedded RPC operation to be called repeatedly if appropriate
* until whatever error condition (if any) is intolerable.
*
* Side Effects:
* As advertised.
*
* NOTE:
* The retry return value is used by afs_StoreAllSegments to determine
* if this is a temporary or permanent error.
*------------------------------------------------------------------------*/
int
afs_Analyze(register struct afs_conn *aconn, afs_int32 acode,
struct VenusFid *afid, register struct vrequest *areq, int op,
afs_int32 locktype, struct cell *cellp)
{
afs_int32 i;
struct srvAddr *sa;
struct server *tsp;
struct volume *tvp;
afs_int32 shouldRetry = 0;
afs_int32 serversleft = 1;
struct afs_stats_RPCErrors *aerrP;
afs_int32 markeddown;
if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
/* On reconnection, act as connected. XXX: for now.... */
/* SXW - This may get very tired after a while. We should try and
* intercept all RPCs before they get here ... */
/*printf("afs_Analyze: disconnected\n");*/
afs_FinalizeReq(areq);
if (aconn) {
/* SXW - I suspect that this will _never_ happen - we shouldn't
* get a connection because we're disconnected !!!*/
afs_PutConn(aconn, locktype);
}
return 0;
}
AFS_STATCNT(afs_Analyze);
afs_Trace4(afs_iclSetp, CM_TRACE_ANALYZE, ICL_TYPE_INT32, op,
ICL_TYPE_POINTER, aconn, ICL_TYPE_INT32, acode, ICL_TYPE_LONG,
areq->uid);
aerrP = (struct afs_stats_RPCErrors *)0;
if ((op >= 0) && (op < AFS_STATS_NUM_FS_RPC_OPS))
aerrP = &(afs_stats_cmfullperf.rpc.fsRPCErrors[op]);
afs_FinalizeReq(areq);
if (!aconn && areq->busyCount) { /* one RPC or more got VBUSY/VRESTARTING */
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
afs_warnuser("afs: Waiting for busy volume %u (%s) in cell %s\n",
(afid ? afid->Fid.Volume : 0),
(tvp->name ? tvp->name : ""),
((tvp->serverHost[0]
&& tvp->serverHost[0]->cell) ? tvp->serverHost[0]->
cell->cellName : ""));
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->status[i] != not_busy && tvp->status[i] != offline) {
tvp->status[i] = not_busy;
}
if (tvp->status[i] == not_busy)
shouldRetry = 1;
}
afs_PutVolume(tvp, READ_LOCK);
} else {
afs_warnuser("afs: Waiting for busy volume %u\n",
(afid ? afid->Fid.Volume : 0));
}
if (areq->busyCount > 100) {
if (aerrP)
(aerrP->err_Volume)++;
areq->volumeError = VOLBUSY;
shouldRetry = 0;
} else {
VSleep(afs_BusyWaitPeriod); /* poll periodically */
}
if (shouldRetry != 0)
areq->busyCount++;
return shouldRetry; /* should retry */
}
if (!aconn || !aconn->srvr) {
if (!areq->volumeError) {
if (aerrP)
(aerrP->err_Network)++;
if (hm_retry_int && !(areq->flags & O_NONBLOCK) && /* "hard" mount */
((afid && afs_IsPrimaryCellNum(afid->Cell))
|| (cellp && afs_IsPrimaryCell(cellp)))) {
if (!afid) {
afs_warnuser
("afs: hard-mount waiting for a vlserver to return to service\n");
VSleep(hm_retry_int);
afs_CheckServers(1, cellp);
shouldRetry = 1;
} else {
tvp = afs_FindVolume(afid, READ_LOCK);
if (!tvp || (tvp->states & VRO)) {
shouldRetry = hm_retry_RO;
} else {
shouldRetry = hm_retry_RW;
}
if (tvp)
afs_PutVolume(tvp, READ_LOCK);
if (shouldRetry) {
afs_warnuser
("afs: hard-mount waiting for volume %u\n",
afid->Fid.Volume);
VSleep(hm_retry_int);
afs_CheckServers(1, cellp);
}
}
} /* if (hm_retry_int ... */
else {
areq->networkError = 1;
}
}
return shouldRetry;
}
/* Find server associated with this connection. */
sa = aconn->srvr;
tsp = sa->server;
/* Before we do anything with acode, make sure we translate it back to
* a system error */
if ((acode & ~0xff) == ERROR_TABLE_BASE_uae)
acode = et_to_sys_error(acode);
if (acode == 0) {
/* If we previously took an error, mark this volume not busy */
if (areq->volumeError) {
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->serverHost[i] == tsp) {
tvp->status[i] = not_busy;
}
}
afs_PutVolume(tvp, READ_LOCK);
}
}
afs_PutConn(aconn, locktype);
return 0;
}
/* If network troubles, mark server as having bogued out again. */
/* VRESTARTING is < 0 because of backward compatibility issues
* with 3.4 file servers and older cache managers */
#ifdef AFS_64BIT_CLIENT
if (acode == -455)
acode = 455;
#endif /* AFS_64BIT_CLIENT */
if ((acode < 0) && (acode != VRESTARTING)) {
if (acode == RX_CALL_TIMEOUT) {
serversleft = afs_BlackListOnce(areq, afid, tsp);
areq->idleError++;
if (serversleft) {
shouldRetry = 1;
} else {
shouldRetry = 0;
}
/* By doing this, we avoid ever marking a server down
* in an idle timeout case. That's because the server is
* still responding and may only be letting a single vnode
* time out. We otherwise risk having the server continually
* be marked down, then up, then down again...
*/
goto out;
}
markeddown = afs_ServerDown(sa);
ForceNewConnections(sa); /**multi homed clients lock:afs_xsrvAddr? */
if (aerrP)
(aerrP->err_Server)++;
#if 0
/* retry *once* when the server is timed out in case of NAT */
if (markeddown && acode == RX_CALL_DEAD) {
aconn->forceConnectFS = 1;
shouldRetry = 1;
}
#endif
}
if (acode == VBUSY || acode == VRESTARTING) {
if (acode == VBUSY) {
areq->busyCount++;
if (aerrP)
(aerrP->err_VolumeBusies)++;
} else
areq->busyCount = 1;
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
for (i = 0; i < MAXHOSTS; i++) {
if (tvp->serverHost[i] == tsp) {
tvp->status[i] = rdwr_busy; /* can't tell which yet */
/* to tell which, have to look at the op code. */
}
}
afs_PutVolume(tvp, READ_LOCK);
} else {
afs_warnuser("afs: Waiting for busy volume %u in cell %s\n",
(afid ? afid->Fid.Volume : 0), tsp->cell->cellName);
VSleep(afs_BusyWaitPeriod); /* poll periodically */
}
shouldRetry = 1;
acode = 0;
} else if (acode == VICETOKENDEAD
|| (acode & ~0xff) == ERROR_TABLE_BASE_RXK) {
/* any rxkad error is treated as token expiration */
struct unixuser *tu;
/*
* I'm calling these errors protection errors, since they involve
* faulty authentication.
*/
if (aerrP)
(aerrP->err_Protection)++;
tu = afs_FindUser(areq->uid, tsp->cell->cellNum, READ_LOCK);
if (tu) {
if (acode == VICETOKENDEAD) {
aconn->forceConnectFS = 1;
} else if (acode == RXKADEXPIRED) {
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user of AFS id %d for cell %s have expired\n",
tu->vid, aconn->srvr->server->cell->cellName);
} else {
serversleft = afs_BlackListOnce(areq, afid, tsp);
areq->tokenError++;
if (serversleft) {
afs_warnuser
("afs: Tokens for user of AFS id %d for cell %s: rxkad error=%d\n",
tu->vid, aconn->srvr->server->cell->cellName, acode);
shouldRetry = 1;
} else {
areq->tokenError = 0;
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user of AFS id %d for cell %s are discarded (rxkad error=%d)\n",
tu->vid, aconn->srvr->server->cell->cellName, acode);
}
}
afs_PutUser(tu, READ_LOCK);
} else {
/* The else case shouldn't be possible and should probably be replaced by a panic? */
if (acode == VICETOKENDEAD) {
aconn->forceConnectFS = 1;
} else if (acode == RXKADEXPIRED) {
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user %d for cell %s have expired\n",
areq->uid, aconn->srvr->server->cell->cellName);
} else {
aconn->forceConnectFS = 0; /* don't check until new tokens set */
aconn->user->states |= UTokensBad;
afs_warnuser
("afs: Tokens for user %d for cell %s are discarded (rxkad error = %d)\n",
areq->uid, aconn->srvr->server->cell->cellName, acode);
}
}
shouldRetry = 1; /* Try again (as root). */
}
/* Check for access violation. */
else if (acode == EACCES) {
/* should mark access error in non-existent per-user global structure */
if (aerrP)
(aerrP->err_Protection)++;
areq->accessError = 1;
if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
areq->permWriteError = 1;
shouldRetry = 0;
}
/* check for ubik errors; treat them like crashed servers */
else if (acode >= ERROR_TABLE_BASE_U && acode < ERROR_TABLE_BASE_U + 255) {
afs_ServerDown(sa);
if (aerrP)
(aerrP->err_Server)++;
shouldRetry = 1; /* retryable (maybe one is working) */
VSleep(1); /* just in case */
}
/* Check for bad volume data base / missing volume. */
else if (acode == VSALVAGE || acode == VOFFLINE || acode == VNOVOL
|| acode == VNOSERVICE || acode == VMOVED) {
struct cell *tcell;
int same;
shouldRetry = 1;
areq->volumeError = VOLMISSING;
if (aerrP)
(aerrP->err_Volume)++;
if (afid && (tcell = afs_GetCell(afid->Cell, 0))) {
same = VLDB_Same(afid, areq);
tvp = afs_FindVolume(afid, READ_LOCK);
if (tvp) {
for (i = 0; i < MAXHOSTS && tvp->serverHost[i]; i++) {
if (tvp->serverHost[i] == tsp) {
if (tvp->status[i] == end_not_busy)
tvp->status[i] = offline;
else
tvp->status[i]++;
} else if (!same) {
tvp->status[i] = not_busy; /* reset the others */
}
}
afs_PutVolume(tvp, READ_LOCK);
}
}
} else if (acode >= ERROR_TABLE_BASE_VL && acode <= ERROR_TABLE_BASE_VL + 255) { /* vlserver errors */
shouldRetry = 0;
areq->volumeError = VOLMISSING;
} else if (acode >= 0) {
if (aerrP)
(aerrP->err_Other)++;
if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
areq->permWriteError = 1;
shouldRetry = 0; /* Other random Vice error. */
} else if (acode == RX_MSGSIZE) { /* same meaning as EMSGSIZE... */
VSleep(1); /* Just a hack for desperate times. */
if (aerrP)
(aerrP->err_Other)++;
shouldRetry = 1; /* packet was too big, please retry call */
}
if (acode < 0 && acode != RX_MSGSIZE && acode != VRESTARTING) {
/* If we get here, code < 0 and we have network/Server troubles.
* areq->networkError is not set here, since we always
* retry in case there is another server. However, if we find
* no connection (aconn == 0) we set the networkError flag.
*/
afs_MarkServerUpOrDown(sa, SRVR_ISDOWN);
if (aerrP)
(aerrP->err_Server)++;
VSleep(1); /* Just a hack for desperate times. */
shouldRetry = 1;
}
out:
/* now unlock the connection and return */
afs_PutConn(aconn, locktype);
return (shouldRetry);
} /*afs_Analyze */
int
afs_BioDaemon(afs_int32 nbiods)
{
afs_int32 code, s, pflg = 0;
label_t jmpbuf;
struct buf *bp, *bp1, *tbp1, *tbp2; /* temp pointers only */
caddr_t tmpaddr;
struct vnode *vp;
struct vcache *vcp;
char tmperr;
if (!afs_initbiod) {
/* XXX ###1 XXX */
afs_initbiod = 1;
/* pin lock, since we'll be using it in an interrupt. */
lock_alloc(&afs_asyncbuf_lock, LOCK_ALLOC_PIN, 2, 1);
simple_lock_init(&afs_asyncbuf_lock);
pin(&afs_asyncbuf, sizeof(struct buf *));
pin(&afs_asyncbuf_cv, sizeof(afs_int32));
}
/* Ignore HUP signals... */
{
sigset_t sigbits, osigbits;
/*
* add SIGHUP to the set of already masked signals
*/
SIGFILLSET(sigbits); /* allow all signals */
SIGDELSET(sigbits, SIGHUP); /* except SIGHUP */
limit_sigs(&sigbits, &osigbits); /* and already masked */
}
/* Main body starts here -- this is an intentional infinite loop, and
* should NEVER exit
*
* Now, the loop will exit if get_bioreq() returns NULL, indicating
* that we've been interrupted.
*/
while (1) {
bp = afs_get_bioreq();
if (!bp)
break; /* we were interrupted */
if (code = setjmpx(&jmpbuf)) {
/* This should not have happend, maybe a lack of resources */
AFS_GUNLOCK();
s = disable_lock(INTMAX, &afs_asyncbuf_lock);
for (bp1 = bp; bp; bp = bp1) {
if (bp1)
bp1 = (struct buf *)bp1->b_work;
bp->b_actf = 0;
bp->b_error = code;
bp->b_flags |= B_ERROR;
iodone(bp);
}
unlock_enable(s, &afs_asyncbuf_lock);
AFS_GLOCK();
continue;
}
vcp = VTOAFS(bp->b_vp);
if (bp->b_flags & B_PFSTORE) { /* XXXX */
ObtainWriteLock(&vcp->lock, 404);
if (vcp->v.v_gnode->gn_mwrcnt) {
afs_offs_t newlength =
(afs_offs_t) dbtob(bp->b_blkno) + bp->b_bcount;
if (vcp->f.m.Length < newlength) {
afs_Trace4(afs_iclSetp, CM_TRACE_SETLENGTH,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_LONG,
__LINE__, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(vcp->f.m.Length),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(newlength));
vcp->f.m.Length = newlength;
}
}
ReleaseWriteLock(&vcp->lock);
}
/* If the buffer represents a protection violation, rather than
* an actual request for I/O, no special action need be taken.
*/
if (bp->b_flags & B_PFPROT) {
iodone(bp); /* Notify all users of the buffer that we're done */
clrjmpx(&jmpbuf);
continue;
}
if (DOvmlock)
ObtainWriteLock(&vcp->pvmlock, 211);
/*
* First map its data area to a region in the current address space
* by calling vm_att with the subspace identifier, and a pointer to
* the data area. vm_att returns a new data area pointer, but we
* also want to hang onto the old one.
*/
tmpaddr = bp->b_baddr;
bp->b_baddr = (caddr_t) vm_att(bp->b_xmemd.subspace_id, tmpaddr);
tmperr = afs_ustrategy(bp); /* temp variable saves offset calculation */
if (tmperr) { /* in non-error case */
bp->b_flags |= B_ERROR; /* should other flags remain set ??? */
bp->b_error = tmperr;
}
/* Unmap the buffer's data area by calling vm_det. Reset data area
* to the value that we saved above.
*/
vm_det(bp->b_baddr);
bp->b_baddr = tmpaddr;
/*
* buffer may be linked with other buffers via the b_work field.
* See also afs_gn_strategy. For each buffer in the chain (including
* bp) notify all users of the buffer that the daemon is finished
* using it by calling iodone.
* assumes iodone can modify the b_work field.
*/
for (tbp1 = bp;;) {
tbp2 = (struct buf *)tbp1->b_work;
iodone(tbp1);
if (!tbp2)
break;
tbp1 = (struct buf *)tbp2->b_work;
iodone(tbp2);
if (!tbp1)
break;
}
if (DOvmlock)
ReleaseWriteLock(&vcp->pvmlock); /* Unlock the vnode. */
clrjmpx(&jmpbuf);
} /* infinite loop (unless we're interrupted) */
} /* end of afs_BioDaemon() */
int
afs_MemRead(register struct vcache *avc, struct uio *auio,
struct AFS_UCRED *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, len, tlen;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error, trybusy = 1;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop = NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
#endif
afs_int32 code;
struct vrequest treq;
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_InitReq(&treq, acred)))
return code;
if (!noLock) {
code = afs_VerifyVCache(avc, &treq);
if (code) {
code = afs_CheckCode(code, &treq, 8); /* failed to get it */
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)) {
return afs_CheckCode(EACCES, &treq, 9);
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(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)
*/
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 {
/* 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 */
}
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:
if (trybusy && !afs_BBusy()) {
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 665);
if (!(tdc->mflags & DFFetchReq)) {
/* start the daemon (may already be running, however) */
UpgradeSToWLock(&tdc->mflock, 666);
tdc->mflags |= DFFetchReq;
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc);
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) */
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
} else {
/* no longer fetching, verify data version
* (avoid new GetDCache call) */
if (hsame(avc->f.m.DataVersion, tdc->f.versionNo)
&& ((len = tdc->validPos - filePos) > 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);
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
/* If we get, 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 (!noLock &&
#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);
return error;
}
int
afs_UFSRead(register struct vcache *avc, struct uio *auio,
struct AFS_UCRED *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, len, tlen;
afs_int32 trimlen;
struct dcache *tdc = 0;
afs_int32 error;
#ifdef AFS_DARWIN80_ENV
uio_t tuiop=NULL;
#else
struct uio tuio;
struct uio *tuiop = &tuio;
struct iovec *tvec;
#endif
struct osi_file *tfile;
afs_int32 code;
int trybusy = 1;
struct vrequest treq;
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_InitReq(&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_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();
return afs_CheckCode(EACCES, &treq, 12);
}
}
#endif
#ifndef AFS_DARWIN80_ENV
tvec = (struct iovec *)osi_AllocSmallSpace(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
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 {
/* 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 */
}
tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 2);
#ifdef AFS_DISCON_ENV
if (!tdc) {
printf("Network down in afs_read");
error = ENETDOWN;
break;
}
#endif /* AFS_DISCON_ENV */
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:
if (trybusy && !afs_BBusy()) {
struct brequest *bp;
/* daemon is not busy */
ObtainSharedLock(&tdc->mflock, 667);
if (!(tdc->mflags & DFFetchReq)) {
UpgradeSToWLock(&tdc->mflock, 668);
tdc->mflags |= DFFetchReq;
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
(afs_size_t) filePos, (afs_size_t) 0,
tdc);
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) */
if (tdc->dflags & DFFetching) {
/* still fetching, some new data is here:
* compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
} else {
/* no longer fetching, verify data version (avoid new
* GetDCache call) */
if (hsame(avc->f.m.DataVersion, tdc->f.versionNo)
&& ((len = tdc->validPos - filePos) > 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);
ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
tdc = NULL;
}
}
if (!tdc) {
/* If we get, 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 */
#ifdef IHINT
if (tfile = tdc->ihint) {
if (tdc->f.inode != tfile->inum) {
afs_warn("afs_UFSRead: %x hint mismatch tdc %d inum %d\n",
tdc, tdc->f.inode, tfile->inum);
osi_UFSClose(tfile);
tdc->ihint = tfile = 0;
nihints--;
}
}
if (tfile != 0) {
usedihint++;
} else
#endif /* IHINT */
#if defined(LINUX_USE_FH)
tfile = (struct osi_file *)osi_UFSOpen_fh(&tdc->f.fh, tdc->f.fh_type);
#else
tfile = (struct osi_file *)osi_UFSOpen(tdc->f.inode);
#endif
#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
{
caller_context_t ct;
VOP_RWLOCK(tfile->vnode, 0, &ct);
code = VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, &ct);
VOP_RWUNLOCK(tfile->vnode, 0, &ct);
}
#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_OSF_ENV)
tuio.uio_rw = UIO_READ;
AFS_GUNLOCK();
VOP_READ(tfile->vnode, &tuio, 0, afs_osi_credp, code);
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_FBSD50_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_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
#ifdef IHINT
if (!tdc->ihint && nihints < maxIHint) {
tdc->ihint = tfile;
nihints++;
} else
#endif /* IHINT */
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 (!noLock) {
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);
return error;
}
