afs_rename(OSI_VC_DECL(aodp), char *aname1, struct vcache *andp, char *aname2, struct AFS_UCRED *acred)
#endif
{
register afs_int32 code;
struct afs_fakestat_state ofakestate;
struct afs_fakestat_state nfakestate;
struct vrequest treq;
OSI_VC_CONVERT(aodp);
code = afs_InitReq(&treq, acred);
if (code)
return code;
afs_InitFakeStat(&ofakestate);
afs_InitFakeStat(&nfakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&aodp, &ofakestate, &treq);
if (code)
goto done;
code = afs_EvalFakeStat(&andp, &nfakestate, &treq);
if (code)
goto done;
code = afsrename(aodp, aname1, andp, aname2, acred, &treq);
done:
afs_PutFakeStat(&ofakestate);
afs_PutFakeStat(&nfakestate);
AFS_DISCON_UNLOCK();
code = afs_CheckCode(code, &treq, 25);
return code;
}
static void
BStore(struct brequest *ab)
{
struct vcache *tvc;
afs_int32 code;
struct vrequest *treq = NULL;
#if defined(AFS_SGI_ENV)
struct cred *tmpcred;
#endif
AFS_STATCNT(BStore);
if ((code = afs_CreateReq(&treq, ab->cred)))
return;
tvc = ab->vc;
#if defined(AFS_SGI_ENV)
/*
* Since StoreOnLastReference can end up calling osi_SyncVM which
* calls into VM code that assumes that u.u_cred has the
* correct credentials, we set our to theirs for this xaction
*/
tmpcred = OSI_GET_CURRENT_CRED();
OSI_SET_CURRENT_CRED(ab->cred);
/*
* To avoid recursion since the WriteLock may be released during VM
* operations, we hold the VOP_RWLOCK across this transaction as
* do the other callers of StoreOnLastReference
*/
AFS_RWLOCK((vnode_t *) tvc, 1);
#endif
ObtainWriteLock(&tvc->lock, 209);
code = afs_StoreOnLastReference(tvc, treq);
ReleaseWriteLock(&tvc->lock);
#if defined(AFS_SGI_ENV)
OSI_SET_CURRENT_CRED(tmpcred);
AFS_RWUNLOCK((vnode_t *) tvc, 1);
#endif
/* now set final return code, and wakeup anyone waiting */
if ((ab->flags & BUVALID) == 0) {
/* To explain code_raw/code_checkcode:
* Anyone that's waiting won't have our treq, so they won't be able to
* call afs_CheckCode themselves on the return code we provide here.
* But if we give back only the afs_CheckCode value, they won't know
* what the "raw" value was. So give back both values, so the waiter
* can know the "raw" value for interpreting the value internally, as
* well as the afs_CheckCode value to give to the OS. */
ab->code_raw = code;
ab->code_checkcode = afs_CheckCode(code, treq, 430);
ab->flags |= BUVALID;
if (ab->flags & BUWAIT) {
ab->flags &= ~BUWAIT;
afs_osi_Wakeup(ab);
}
}
afs_DestroyReq(treq);
}
static void
BStore(struct brequest *ab)
{
struct vcache *tvc;
afs_int32 code;
struct vrequest treq;
#if defined(AFS_SGI_ENV)
struct cred *tmpcred;
#endif
AFS_STATCNT(BStore);
if ((code = afs_InitReq(&treq, ab->cred)))
return;
code = 0;
tvc = ab->vc;
#if defined(AFS_SGI_ENV)
/*
* Since StoreOnLastReference can end up calling osi_SyncVM which
* calls into VM code that assumes that u.u_cred has the
* correct credentials, we set our to theirs for this xaction
*/
tmpcred = OSI_GET_CURRENT_CRED();
OSI_SET_CURRENT_CRED(ab->cred);
/*
* To avoid recursion since the WriteLock may be released during VM
* operations, we hold the VOP_RWLOCK across this transaction as
* do the other callers of StoreOnLastReference
*/
AFS_RWLOCK((vnode_t *) tvc, 1);
#endif
ObtainWriteLock(&tvc->lock, 209);
code = afs_StoreOnLastReference(tvc, &treq);
ReleaseWriteLock(&tvc->lock);
#if defined(AFS_SGI_ENV)
OSI_SET_CURRENT_CRED(tmpcred);
AFS_RWUNLOCK((vnode_t *) tvc, 1);
#endif
/* now set final return code, and wakeup anyone waiting */
if ((ab->flags & BUVALID) == 0) {
ab->code = afs_CheckCode(code, &treq, 43); /* set final code, since treq doesn't go across processes */
ab->flags |= BUVALID;
if (ab->flags & BUWAIT) {
ab->flags &= ~BUWAIT;
afs_osi_Wakeup(ab);
}
}
}
int
afs_readlink(OSI_VC_DECL(avc), struct uio *auio, afs_ucred_t *acred)
{
afs_int32 code;
struct vrequest *treq = NULL;
char *tp;
struct afs_fakestat_state fakestat;
OSI_VC_CONVERT(avc);
AFS_STATCNT(afs_readlink);
afs_Trace1(afs_iclSetp, CM_TRACE_READLINK, ICL_TYPE_POINTER, avc);
if ((code = afs_CreateReq(&treq, acred)))
return code;
afs_InitFakeStat(&fakestat);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&avc, &fakestat, treq);
if (code)
goto done;
code = afs_VerifyVCache(avc, treq);
if (code)
goto done;
if (vType(avc) != VLNK) {
code = EINVAL;
goto done;
}
ObtainWriteLock(&avc->lock, 158);
code = afs_HandleLink(avc, treq);
/* finally uiomove it to user-land */
if (code == 0) {
tp = avc->linkData;
if (tp)
AFS_UIOMOVE(tp, strlen(tp), UIO_READ, auio, code);
else {
code = EIO;
}
}
ReleaseWriteLock(&avc->lock);
done:
afs_PutFakeStat(&fakestat);
AFS_DISCON_UNLOCK();
code = afs_CheckCode(code, treq, 32);
afs_DestroyReq(treq);
return code;
}
static void
BPartialStore(struct brequest *ab)
{
struct vcache *tvc;
afs_int32 code;
struct vrequest *treq = NULL;
int locked, shared_locked = 0;
AFS_STATCNT(BStore);
if ((code = afs_CreateReq(&treq, ab->cred)))
return;
tvc = ab->vc;
locked = tvc->lock.excl_locked? 1:0;
if (!locked)
ObtainWriteLock(&tvc->lock, 1209);
else if (!(tvc->lock.excl_locked & WRITE_LOCK)) {
shared_locked = 1;
ConvertSToRLock(&tvc->lock);
}
code = afs_StoreAllSegments(tvc, treq, AFS_ASYNC);
if (!locked)
ReleaseWriteLock(&tvc->lock);
else if (shared_locked)
ConvertSToRLock(&tvc->lock);
/* now set final return code, and wakeup anyone waiting */
if ((ab->flags & BUVALID) == 0) {
/* set final code, since treq doesn't go across processes */
ab->code_raw = code;
ab->code_checkcode = afs_CheckCode(code, treq, 43);
ab->flags |= BUVALID;
if (ab->flags & BUWAIT) {
ab->flags &= ~BUWAIT;
afs_osi_Wakeup(ab);
}
}
afs_DestroyReq(treq);
}
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;
}
afs_link(struct vcache *avc, OSI_VC_DECL(adp), char *aname,
afs_ucred_t *acred)
#endif
{
struct vrequest treq;
struct dcache *tdc;
afs_int32 code;
struct afs_conn *tc;
afs_size_t offset, len;
struct AFSFetchStatus OutFidStatus, OutDirStatus;
struct AFSVolSync tsync;
struct afs_fakestat_state vfakestate, dfakestate;
struct rx_connection *rxconn;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_link);
afs_Trace3(afs_iclSetp, CM_TRACE_LINK, ICL_TYPE_POINTER, adp,
ICL_TYPE_POINTER, avc, ICL_TYPE_STRING, aname);
/* create a hard link; new entry is aname in dir adp */
if ((code = afs_InitReq(&treq, acred)))
goto done2;
afs_InitFakeStat(&vfakestate);
afs_InitFakeStat(&dfakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&avc, &vfakestate, &treq);
if (code)
goto done;
code = afs_EvalFakeStat(&adp, &dfakestate, &treq);
if (code)
goto done;
if (avc->f.fid.Cell != adp->f.fid.Cell
|| avc->f.fid.Fid.Volume != adp->f.fid.Fid.Volume) {
code = EXDEV;
goto done;
}
if (strlen(aname) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done;
}
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) {
code = ENETDOWN;
goto done;
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, &treq, &offset, &len, 1); /* test for error below */
ObtainWriteLock(&adp->lock, 145);
do {
tc = afs_Conn(&adp->f.fid, &treq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_LINK);
RX_AFS_GUNLOCK();
code =
RXAFS_Link(rxconn, (struct AFSFid *)&adp->f.fid.Fid, aname,
(struct AFSFid *)&avc->f.fid.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_LINK,
SHARED_LOCK, NULL));
if (code) {
if (tdc)
afs_PutDCache(tdc);
if (code < 0) {
ObtainWriteLock(&afs_xcbhash, 492);
afs_DequeueCallback(adp);
adp->f.states &= ~CStatd;
ReleaseWriteLock(&afs_xcbhash);
osi_dnlc_purgedp(adp);
}
ReleaseWriteLock(&adp->lock);
goto done;
}
if (tdc)
ObtainWriteLock(&tdc->lock, 635);
if (afs_LocalHero(adp, tdc, &OutDirStatus, 1)) {
/* we can do it locally */
ObtainWriteLock(&afs_xdcache, 290);
code = afs_dir_Create(tdc, aname, &avc->f.fid.Fid);
ReleaseWriteLock(&afs_xdcache);
if (code) {
ZapDCE(tdc); /* surprise error -- invalid value */
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc); /* drop ref count */
}
ReleaseWriteLock(&adp->lock);
ObtainWriteLock(&avc->lock, 146); /* correct link count */
/* we could lock both dir and file; since we get the new fid
* status back, you'd think we could put it in the cache status
* entry at that point. Note that if we don't lock the file over
* the rpc call, we have no guarantee that the status info
* returned in ustat is the most recent to store in the file's
* cache entry */
ObtainWriteLock(&afs_xcbhash, 493);
afs_DequeueCallback(avc);
avc->f.states &= ~CStatd; /* don't really know new link count */
ReleaseWriteLock(&afs_xcbhash);
if (avc->f.fid.Fid.Vnode & 1 || (vType(avc) == VDIR))
osi_dnlc_purgedp(avc);
ReleaseWriteLock(&avc->lock);
code = 0;
done:
code = afs_CheckCode(code, &treq, 24);
afs_PutFakeStat(&vfakestate);
afs_PutFakeStat(&dfakestate);
AFS_DISCON_UNLOCK();
done2:
return code;
}
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;
}
/* 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;
}
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);
}
/*
* Given a FID, obtain or construct a dentry, or return an error.
* This should be called with the BKL and AFS_GLOCK held.
*/
static struct dentry *get_dentry_from_fid(cred_t *credp, struct VenusFid *afid)
{
struct vrequest treq;
struct vcache *vcp;
struct vattr vattr;
struct inode *ip;
struct dentry *dp;
afs_int32 code;
code = afs_InitReq(&treq, credp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): afs_InitReq: %d\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique,
code);
#endif
return ERR_PTR(-afs_CheckCode(code, &treq, 101));
}
vcp = afs_GetVCache(afid, &treq, NULL, NULL);
if (vcp == NULL) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): no vcache\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique);
#endif
return NULL;
}
/*
* Now, it might be that we just caused a directory vnode to
* spring into existence, in which case its parent FID is unset.
* We need to do something about that, but only because we care
* in our own get_parent(), below -- the common code never looks
* at parentVnode on directories, except for VIOCGETVCXSTATUS.
* So, if this fails, we don't really care very much.
*/
if (vType(vcp) == VDIR && vcp->mvstat != 2 && !vcp->f.parent.vnode)
update_dir_parent(&treq, vcp);
/*
* If this is a volume root directory and fakestat is enabled,
* we might need to replace the directory by a mount point.
*/
code = UnEvalFakeStat(&treq, &vcp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): UnEvalFakeStat: %d\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique,
code);
#endif
afs_PutVCache(vcp);
return ERR_PTR(-afs_CheckCode(code, &treq, 101));
}
ip = AFSTOV(vcp);
afs_getattr(vcp, &vattr, credp);
afs_fill_inode(ip, &vattr);
/* d_alloc_anon might block, so we shouldn't hold the glock */
AFS_GUNLOCK();
dp = d_alloc_anon(ip);
AFS_GLOCK();
if (!dp) {
iput(ip);
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): out of memory\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique);
#endif
return ERR_PTR(-ENOMEM);
}
dp->d_op = &afs_dentry_operations;
return dp;
}
static int afs_export_get_name(struct dentry *parent, char *name,
struct dentry *child)
{
struct afs_fakestat_state fakestate;
struct get_name_data data;
struct vrequest treq;
struct volume *tvp;
struct vcache *vcp;
struct dcache *tdc;
cred_t *credp;
afs_size_t dirOffset, dirLen;
afs_int32 code = 0;
if (!parent->d_inode) {
#ifdef OSI_EXPORT_DEBUG
/* can't lookup name in a negative dentry */
printk("afs: get_name(%s, %s): no parent inode\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
child->d_name.name ? (char *)child->d_name.name : "?");
#endif
return -EIO;
}
if (!child->d_inode) {
#ifdef OSI_EXPORT_DEBUG
/* can't find the FID of negative dentry */
printk("afs: get_name(%s, %s): no child inode\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
child->d_name.name ? (char *)child->d_name.name : "?");
#endif
return -ENOENT;
}
afs_InitFakeStat(&fakestate);
credp = crref();
AFS_GLOCK();
vcp = VTOAFS(child->d_inode);
/* special case dynamic mount directory */
if (afs_IsDynrootMount(vcp)) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): this is the dynmount dir\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
data.fid = vcp->f.fid;
if (VTOAFS(parent->d_inode) == afs_globalVp)
strcpy(name, AFS_DYNROOT_MOUNTNAME);
else
code = -ENOENT;
goto done;
}
/* Figure out what FID to look for */
if (vcp->mvstat == 2) { /* volume root */
tvp = afs_GetVolume(&vcp->f.fid, 0, READ_LOCK);
if (!tvp) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): no volume for root\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
code = ENOENT;
goto done;
}
data.fid = tvp->mtpoint;
afs_PutVolume(tvp, READ_LOCK);
} else {
data.fid = vcp->f.fid;
}
vcp = VTOAFS(parent->d_inode);
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): parent is 0x%08x/%d/%d.%d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
code = afs_InitReq(&treq, credp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): afs_InitReq: %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique, code);
#endif
goto done;
}
/* a dynamic mount point in the dynamic mount directory */
if (afs_IsDynrootMount(vcp) && afs_IsDynrootAnyFid(&data.fid)
&& VNUM_TO_VNTYPE(data.fid.Fid.Vnode) == VN_TYPE_MOUNT) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): dynamic mount point\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique);
#endif
vcp = afs_GetVCache(&data.fid, &treq, NULL, NULL);
if (vcp) {
ObtainReadLock(&vcp->lock);
if (strlen(vcp->linkData + 1) <= NAME_MAX)
strcpy(name, vcp->linkData + 1);
else
code = ENOENT;
ReleaseReadLock(&vcp->lock);
afs_PutVCache(vcp);
} else {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): no vcache\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique);
#endif
code = ENOENT;
}
goto done;
}
code = afs_EvalFakeStat(&vcp, &fakestate, &treq);
if (code)
goto done;
if (vcp->f.fid.Cell != data.fid.Cell ||
vcp->f.fid.Fid.Volume != data.fid.Fid.Volume) {
/* parent is not the expected cell and volume; thus it
* cannot possibly contain the fid we are looking for */
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): wrong parent 0x%08x/%d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume);
#endif
code = ENOENT;
goto done;
}
redo:
if (!(vcp->f.states & CStatd)) {
if ((code = afs_VerifyVCache2(vcp, &treq))) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): VerifyVCache2(0x%08x/%d/%d.%d): %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
goto done;
}
}
tdc = afs_GetDCache(vcp, (afs_size_t) 0, &treq, &dirOffset, &dirLen, 1);
if (!tdc) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): GetDCache(0x%08x/%d/%d.%d): %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
code = EIO;
goto done;
}
ObtainReadLock(&vcp->lock);
ObtainReadLock(&tdc->lock);
/*
* Make sure that the data in the cache is current. There are two
* cases we need to worry about:
* 1. The cache data is being fetched by another process.
* 2. The cache data is no longer valid
*/
while ((vcp->f.states & CStatd)
&& (tdc->dflags & DFFetching)
&& hsame(vcp->f.m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&vcp->lock);
afs_osi_Sleep(&tdc->validPos);
ObtainReadLock(&vcp->lock);
ObtainReadLock(&tdc->lock);
}
if (!(vcp->f.states & CStatd)
|| !hsame(vcp->f.m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&vcp->lock);
afs_PutDCache(tdc);
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): dir (0x%08x/%d/%d.%d) changed; retrying\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
goto redo;
}
data.name = name;
data.found = 0;
code = afs_dir_EnumerateDir(tdc, get_name_hook, &data);
if (!code && !data.found) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): not found\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique);
#endif
code = ENOENT;
} else if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): Enumeratedir(0x%08x/%d/%d.%d): %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
}
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&vcp->lock);
afs_PutDCache(tdc);
done:
if (!code) {
printk("afs: get_name(%s, 0x%08x/%d/%d.%d) => %s\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique, name);
}
afs_PutFakeStat(&fakestate);
AFS_GUNLOCK();
crfree(credp);
code = afs_CheckCode(code, &treq, 102);
return -code;
}
int
afs_mkdir(OSI_VC_DECL(adp), char *aname, struct vattr *attrs,
struct vcache **avcp, afs_ucred_t *acred)
{
struct vrequest *treq = NULL;
afs_int32 code;
struct afs_conn *tc;
struct rx_connection *rxconn;
struct VenusFid newFid;
struct dcache *tdc;
struct dcache *new_dc;
afs_size_t offset, len;
struct vcache *tvc;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus *OutFidStatus, *OutDirStatus;
struct AFSCallBack CallBack;
struct AFSVolSync tsync;
afs_int32 now;
struct afs_fakestat_state fakestate;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_mkdir);
afs_Trace2(afs_iclSetp, CM_TRACE_MKDIR, 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);
if (strlen(aname) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done3;
}
if (!afs_ENameOK(aname)) {
code = EINVAL;
goto done3;
}
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, treq);
if (code)
goto done;
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) {
/*printf("Network is down in afs_mkdir\n");*/
code = ENETDOWN;
goto done;
}
InStatus.Mask = AFS_SETMODTIME | AFS_SETMODE | AFS_SETGROUP;
InStatus.ClientModTime = osi_Time();
InStatus.UnixModeBits = attrs->va_mode & 0xffff; /* only care about protection bits */
InStatus.Group = (afs_int32) afs_cr_gid(acred);
tdc = afs_GetDCache(adp, (afs_size_t) 0, treq, &offset, &len, 1);
ObtainWriteLock(&adp->lock, 153);
if (!AFS_IS_DISCON_RW) {
do {
tc = afs_Conn(&adp->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_MAKEDIR);
now = osi_Time();
RX_AFS_GUNLOCK();
code =
RXAFS_MakeDir(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;
/* DON'T forget to Set the callback value... */
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &adp->f.fid, treq, AFS_STATS_FS_RPCIDX_MAKEDIR,
SHARED_LOCK, NULL));
if (code) {
if (code < 0) {
afs_StaleVCache(adp);
}
ReleaseWriteLock(&adp->lock);
if (tdc)
afs_PutDCache(tdc);
goto done;
}
} else {
/* Disconnected. */
/* We have the dir entry now, we can use it while disconnected. */
if (adp->mvid.target_root == NULL) {
/* If not mount point, generate a new fid. */
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
afs_GenFakeFid(&newFid, VDIR, 1);
}
/* XXX: If mount point???*/
/* Operations with the actual dir's cache entry are further
* down, where the dir entry gets created.
*/
} /* if (!AFS_IS_DISCON_RW) */
/* otherwise, we should see if we can make the change to the dir locally */
if (tdc)
ObtainWriteLock(&tdc->lock, 632);
if (AFS_IS_DISCON_RW || afs_LocalHero(adp, tdc, OutDirStatus, 1)) {
/* we can do it locally */
ObtainWriteLock(&afs_xdcache, 294);
code = afs_dir_Create(tdc, aname, &newFid.Fid);
ReleaseWriteLock(&afs_xdcache);
if (code) {
ZapDCE(tdc); /* surprise error -- use invalid value */
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
if (AFS_IS_DISCON_RW)
/* We will have to settle with the local link count. */
adp->f.m.LinkCount++;
else
adp->f.m.LinkCount = OutDirStatus->LinkCount;
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
ReleaseWriteLock(&adp->lock);
if (AFS_IS_DISCON_RW) {
/* When disconnected, we have to create the full dir here. */
/* Generate a new vcache and fill it. */
tvc = afs_NewVCache(&newFid, NULL);
if (tvc) {
*avcp = tvc;
} else {
code = EIO;
goto done;
}
ObtainWriteLock(&tvc->lock, 738);
afs_GenDisconStatus(adp, tvc, &newFid, attrs, treq, VDIR);
ReleaseWriteLock(&tvc->lock);
/* And now make an empty dir, containing . and .. : */
/* Get a new dcache for it first. */
new_dc = afs_GetDCache(tvc, (afs_size_t) 0, treq, &offset, &len, 1);
if (!new_dc) {
/* printf("afs_mkdir: can't get new dcache for dir.\n"); */
code = EIO;
goto done;
}
ObtainWriteLock(&afs_xdcache, 739);
code = afs_dir_MakeDir(new_dc,
(afs_int32 *) &newFid.Fid,
(afs_int32 *) &adp->f.fid.Fid);
ReleaseWriteLock(&afs_xdcache);
/* if (code) printf("afs_mkdir: afs_dirMakeDir code = %u\n", code); */
afs_PutDCache(new_dc);
ObtainWriteLock(&tvc->lock, 731);
/* Update length in the vcache. */
tvc->f.m.Length = new_dc->f.chunkBytes;
afs_DisconAddDirty(tvc, VDisconCreate, 1);
ReleaseWriteLock(&tvc->lock);
} else {
/* now we're done with parent dir, create the real dir's cache entry */
tvc = afs_GetVCache(&newFid, treq, NULL, NULL);
if (tvc) {
code = 0;
*avcp = tvc;
} else {
/* For some reason, we cannot fetch the vcache for our
* newly-created dir. */
code = EIO;
}
} /* if (AFS_DISCON_RW) */
done:
AFS_DISCON_UNLOCK();
done3:
afs_PutFakeStat(&fakestate);
code = afs_CheckCode(code, treq, 26);
afs_DestroyReq(treq);
done2:
osi_FreeSmallSpace(OutFidStatus);
osi_FreeSmallSpace(OutDirStatus);
return code;
}
/* 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;
}
/* dispatch a no-cache read request */
afs_int32
afs_ReadNoCache(struct vcache *avc,
struct nocache_read_request *bparms,
afs_ucred_t *acred)
{
afs_int32 code;
afs_int32 bcnt;
struct brequest *breq;
struct vrequest *areq;
/* the reciever will free this */
areq = osi_Alloc(sizeof(struct vrequest));
if (avc->vc_error) {
code = EIO;
afs_warn("afs_ReadNoCache VCache Error!\n");
goto cleanup;
}
if ((code = afs_InitReq(areq, acred))) {
afs_warn("afs_ReadNoCache afs_InitReq error!\n");
goto cleanup;
}
AFS_GLOCK();
code = afs_VerifyVCache(avc, areq);
AFS_GUNLOCK();
if (code) {
code = afs_CheckCode(code, areq, 11); /* failed to get it */
afs_warn("afs_ReadNoCache Failed to verify VCache!\n");
goto cleanup;
}
bparms->areq = areq;
/* and queue this one */
bcnt = 1;
AFS_GLOCK();
while(bcnt < 20) {
breq = afs_BQueue(BOP_FETCH_NOCACHE, avc, B_DONTWAIT, 0, acred, 1, 1,
bparms, (void *)0, (void *)0);
if(breq != 0) {
code = 0;
break;
}
afs_osi_Wait(10 * bcnt, 0, 0);
}
AFS_GUNLOCK();
if(!breq) {
code = EBUSY;
goto cleanup;
}
return code;
cleanup:
/* If there's a problem before we queue the request, we need to
* do everything that would normally happen when the request was
* processed, like unlocking the pages and freeing memory.
*/
unlock_and_release_pages(bparms->auio);
osi_Free(areq, sizeof(struct vrequest));
osi_Free(bparms->auio->uio_iov,
bparms->auio->uio_iovcnt * sizeof(struct iovec));
osi_Free(bparms->auio, sizeof(struct uio));
osi_Free(bparms, sizeof(struct nocache_read_request));
return code;
}
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;
}
int afs_lockctl(struct vcache * avc, struct AFS_FLOCK * af, int acmd,
afs_ucred_t * acred)
#endif
{
struct vrequest treq;
afs_int32 code;
struct afs_fakestat_state fakestate;
AFS_STATCNT(afs_lockctl);
if ((code = afs_InitReq(&treq, acred)))
return code;
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&avc, &fakestate, &treq);
if (code) {
goto done;
}
#if defined(AFS_SGI_ENV)
if ((acmd == F_GETLK) || (acmd == F_RGETLK)) {
#else
if (acmd == F_GETLK) {
#endif
if (af->l_type == F_UNLCK) {
code = 0;
goto done;
}
code = HandleGetLock(avc, af, &treq, clid);
code = afs_CheckCode(code, &treq, 2); /* defeat buggy AIX optimz */
goto done;
} else if ((acmd == F_SETLK) || (acmd == F_SETLKW)
#if defined(AFS_SGI_ENV)
|| (acmd == F_RSETLK) || (acmd == F_RSETLKW)) {
#else
) {
#endif
if ((avc->f.states & CRO)) {
/* for RO volumes, don't do anything for locks; the fileserver doesn't
* even track them. A write lock should not be possible, though. */
if (af->l_type == F_WRLCK) {
code = EBADF;
} else {
code = 0;
}
goto done;
}
/* Java VMs ask for l_len=(long)-1 regardless of OS/CPU */
if ((sizeof(af->l_len) == 8) && (af->l_len == 0x7fffffffffffffffLL))
af->l_len = 0;
/* next line makes byte range locks always succeed,
* even when they should block */
if (af->l_whence != 0 || af->l_start != 0 || af->l_len != 0) {
DoLockWarning(acred);
code = 0;
goto done;
}
/* otherwise we can turn this into a whole-file flock */
if (af->l_type == F_RDLCK)
code = LOCK_SH;
else if (af->l_type == F_WRLCK)
code = LOCK_EX;
else if (af->l_type == F_UNLCK)
code = LOCK_UN;
else {
code = EINVAL; /* unknown lock type */
goto done;
}
if (((acmd == F_SETLK)
#if defined(AFS_SGI_ENV)
|| (acmd == F_RSETLK)
#endif
) && code != LOCK_UN)
code |= LOCK_NB; /* non-blocking, s.v.p. */
#if defined(AFS_DARWIN_ENV)
code = HandleFlock(avc, code, &treq, clid, 0 /*!onlymine */ );
#elif defined(AFS_SGI_ENV)
AFS_RWLOCK((vnode_t *) avc, VRWLOCK_WRITE);
code = HandleFlock(avc, code, &treq, clid, 0 /*!onlymine */ );
AFS_RWUNLOCK((vnode_t *) avc, VRWLOCK_WRITE);
#else
code = HandleFlock(avc, code, &treq, 0, 0 /*!onlymine */ );
#endif
code = afs_CheckCode(code, &treq, 3); /* defeat AIX -O bug */
goto done;
}
code = EINVAL;
done:
afs_PutFakeStat(&fakestate);
AFS_DISCON_UNLOCK();
return code;
}
/*
* Get a description of the first lock which would
* block the lock specified. If the specified lock
* would succeed, fill in the lock structure with 'F_UNLCK'.
*
* To do that, we have to ask the server for the lock
* count if:
* 1. The file is not locked by this machine.
* 2. Asking for write lock, and only the current
* PID has the file read locked.
*/
static int
HandleGetLock(struct vcache *avc, struct AFS_FLOCK *af,
struct vrequest *areq, int clid)
{
afs_int32 code;
struct AFS_FLOCK flock;
lockIdSet(&flock, NULL, clid);
ObtainWriteLock(&avc->lock, 122);
if (avc->flockCount == 0) {
/*
* We don't know ourselves, so ask the server. Unfortunately, we
* don't know the pid. Not even the server knows the pid. Besides,
* the process with the lock is on another machine
*/
code = GetFlockCount(avc, areq);
if (code == 0 || (af->l_type == F_RDLCK && code > 0)) {
af->l_type = F_UNLCK;
goto unlck_leave;
}
if (code > 0)
af->l_type = F_RDLCK;
else
af->l_type = F_WRLCK;
af->l_pid = 0;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = 0;
#endif
goto done;
}
if (af->l_type == F_RDLCK) {
/*
* We want a read lock. If there are only
* read locks, or we are the one with the
* write lock, say it is unlocked.
*/
if (avc->flockCount > 0 || /* only read locks */
!lockIdcmp2(&flock, avc, NULL, 1, clid)) {
af->l_type = F_UNLCK;
goto unlck_leave;
}
/* one write lock, but who? */
af->l_type = F_WRLCK; /* not us, so lock would block */
if (avc->slocks) { /* we know who, so tell */
af->l_pid = avc->slocks->pid;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = avc->slocks->sysid;
#endif
} else {
af->l_pid = 0; /* XXX can't happen?? */
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = 0;
#endif
}
goto done;
}
/*
* Ok, we want a write lock. If there is a write lock
* already, and it is not this process, we fail.
*/
if (avc->flockCount < 0) {
if (lockIdcmp2(&flock, avc, NULL, 1, clid)) {
af->l_type = F_WRLCK;
if (avc->slocks) {
af->l_pid = avc->slocks->pid;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = avc->slocks->sysid;
#endif
} else {
af->l_pid = 0; /* XXX can't happen?? */
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = 0;
#endif
}
goto done;
}
/* we are the one with the write lock */
af->l_type = F_UNLCK;
goto unlck_leave;
}
/*
* Want a write lock, and we know there are read locks.
* If there is more than one, or it isn't us, we cannot lock.
*/
if ((avc->flockCount > 1)
|| lockIdcmp2(&flock, avc, NULL, 1, clid)) {
struct SimpleLocks *slp;
af->l_type = F_RDLCK;
af->l_pid = 0;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = 0;
#endif
/* find a pid that isn't our own */
for (slp = avc->slocks; slp; slp = slp->next) {
if (lockIdcmp2(&flock, NULL, slp, 1, clid)) {
af->l_pid = slp->pid;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = avc->slocks->sysid;
#endif
break;
}
}
goto done;
}
/*
* Ok, we want a write lock. If there is a write lock
* already, and it is not this process, we fail.
*/
if (avc->flockCount < 0) {
if (lockIdcmp2(&flock, avc, NULL, 1, clid)) {
af->l_type = F_WRLCK;
if (avc->slocks) {
af->l_pid = avc->slocks->pid;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = avc->slocks->sysid;
#endif
} else {
af->l_pid = 0; /* XXX can't happen?? */
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = 0;
#endif
}
goto done;
}
/* we are the one with the write lock */
af->l_type = F_UNLCK;
goto unlck_leave;
}
/*
* Want a write lock, and we know there are read locks.
* If there is more than one, or it isn't us, we cannot lock.
*/
if ((avc->flockCount > 1)
|| lockIdcmp2(&flock, avc, NULL, 1, clid)) {
struct SimpleLocks *slp;
af->l_type = F_RDLCK;
af->l_pid = 0;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = 0;
#endif
/* find a pid that isn't our own */
for (slp = avc->slocks; slp; slp = slp->next) {
if (lockIdcmp2(&flock, NULL, slp, 1, clid)) {
af->l_pid = slp->pid;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = avc->slocks->sysid;
#endif
break;
}
}
goto done;
}
/*
* Want a write lock, and there is just one read lock, and it
* is this process with a read lock. Ask the server if there
* are any more processes with the file locked.
*/
code = GetFlockCount(avc, areq);
if (code == 0 || code == 1) {
af->l_type = F_UNLCK;
goto unlck_leave;
}
if (code > 0)
af->l_type = F_RDLCK;
else
af->l_type = F_WRLCK;
af->l_pid = 0;
#if defined(AFS_HAVE_FLOCK_SYSID)
af->l_sysid = 0;
#endif
done:
af->l_whence = 0;
af->l_start = 0;
af->l_len = 0; /* to end of file */
unlck_leave:
ReleaseWriteLock(&avc->lock);
return 0;
}
/* Get the 'flock' count from the server. This comes back in a 'spare'
* field from a GetStatus RPC. If we have any problems with the RPC,
* we lie and say the file is unlocked. If we ask any 'old' fileservers,
* the spare field will be a zero, saying the file is unlocked. This is
* OK, as a further 'lock' request will do the right thing.
*/
static int
GetFlockCount(struct vcache *avc, struct vrequest *areq)
{
struct afs_conn *tc;
afs_int32 code;
struct AFSFetchStatus OutStatus;
struct AFSCallBack CallBack;
struct AFSVolSync tsync;
struct rx_connection *rxconn;
int temp;
XSTATS_DECLS;
temp = areq->flags & O_NONBLOCK;
areq->flags |= O_NONBLOCK;
/* If we're disconnected, lie and say that we've got no locks. Ick */
if (AFS_IS_DISCONNECTED)
return 0;
do {
tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHSTATUS);
RX_AFS_GUNLOCK();
code =
RXAFS_FetchStatus(rxconn, (struct AFSFid *)&avc->f.fid.Fid,
&OutStatus, &CallBack, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &avc->f.fid, areq, AFS_STATS_FS_RPCIDX_FETCHSTATUS,
SHARED_LOCK, NULL));
if (temp)
areq->flags &= ~O_NONBLOCK;
if (code) {
return (0); /* failed, say it is 'unlocked' */
} else {
return ((int)OutStatus.lockCount);
}
}
/* clid - nonzero on sgi sunos osf1 only */
int
HandleFlock(struct vcache *avc, int acom, struct vrequest *areq,
pid_t clid, int onlymine)
{
struct afs_conn *tc;
struct SimpleLocks *slp, *tlp, **slpp;
afs_int32 code;
struct AFSVolSync tsync;
afs_int32 lockType;
struct AFS_FLOCK flock;
XSTATS_DECLS;
AFS_STATCNT(HandleFlock);
code = 0; /* default when we don't make any network calls */
lockIdSet(&flock, NULL, clid);
#if defined(AFS_SGI_ENV)
osi_Assert(valusema(&avc->vc_rwlock) <= 0);
osi_Assert(OSI_GET_LOCKID() == avc->vc_rwlockid);
#endif
ObtainWriteLock(&avc->lock, 118);
if (acom & LOCK_UN) {
int stored_segments = 0;
retry_unlock:
/* defect 3083 */
#ifdef AFS_AIX_ENV
/* If the lock is held exclusive, then only the owning process
* or a child can unlock it. Use pid and ppid because they are
* unique identifiers.
*/
if ((avc->flockCount < 0) && (getpid() != avc->ownslock)) {
#ifdef AFS_AIX41_ENV
if (onlymine || (getppid() != avc->ownslock)) {
#else
if (onlymine || (u.u_procp->p_ppid != avc->ownslock)) {
#endif
ReleaseWriteLock(&avc->lock);
return 0;
}
}
#endif
if (lockIdcmp2(&flock, avc, NULL, onlymine, clid)) {
ReleaseWriteLock(&avc->lock);
return 0;
}
#ifdef AFS_AIX_ENV
avc->ownslock = 0;
#endif
if (avc->flockCount == 0) {
ReleaseWriteLock(&avc->lock);
return 0 /*ENOTTY*/;
/* no lock held */
}
/* unlock the lock */
if (avc->flockCount > 0) {
slpp = &avc->slocks;
for (slp = *slpp; slp;) {
if (!lockIdcmp2(&flock, avc, slp, onlymine, clid)) {
avc->flockCount--;
tlp = *slpp = slp->next;
osi_FreeSmallSpace(slp);
slp = tlp;
} else {
slpp = &slp->next;
slp = *slpp;
}
}
} else if (avc->flockCount == -1) {
if (!stored_segments) {
afs_StoreAllSegments(avc, areq, AFS_SYNC | AFS_VMSYNC); /* fsync file early */
/* afs_StoreAllSegments can drop and reacquire the write lock
* on avc and GLOCK, so the flocks may be completely different
* now. Go back and perform all checks again. */
stored_segments = 1;
goto retry_unlock;
}
avc->flockCount = 0;
/* And remove the (only) exclusive lock entry from the list... */
osi_FreeSmallSpace(avc->slocks);
avc->slocks = 0;
}
if (avc->flockCount == 0) {
if (!AFS_IS_DISCONNECTED) {
struct rx_connection *rxconn;
do {
tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_RELEASELOCK);
RX_AFS_GUNLOCK();
code = RXAFS_ReleaseLock(rxconn, (struct AFSFid *)
&avc->f.fid.Fid, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &avc->f.fid, areq,
AFS_STATS_FS_RPCIDX_RELEASELOCK, SHARED_LOCK, NULL));
} else {
/*printf("Network is dooooooowwwwwwwnnnnnnn\n");*/
code = ENETDOWN;
}
}
} else {
while (1) { /* set a new lock */
/*
* Upgrading from shared locks to Exclusive and vice versa
* is a bit tricky and we don't really support it yet. But
* we try to support the common used one which is upgrade
* a shared lock to an exclusive for the same process...
*/
if ((avc->flockCount > 0 && (acom & LOCK_EX))
|| (avc->flockCount == -1 && (acom & LOCK_SH))) {
/*
* Upgrading from shared locks to an exclusive one:
* For now if all the shared locks belong to the
* same process then we unlock them on the server
* and proceed with the upgrade. Unless we change the
* server's locking interface impl we prohibit from
* unlocking other processes's shared locks...
* Upgrading from an exclusive lock to a shared one:
* Again only allowed to be done by the same process.
*/
slpp = &avc->slocks;
for (slp = *slpp; slp;) {
if (!lockIdcmp2
(&flock, avc, slp, 1 /*!onlymine */ , clid)) {
if (acom & LOCK_EX)
avc->flockCount--;
else
avc->flockCount = 0;
tlp = *slpp = slp->next;
osi_FreeSmallSpace(slp);
slp = tlp;
} else {
code = EWOULDBLOCK;
slpp = &slp->next;
slp = *slpp;
}
}
if (!code && avc->flockCount == 0) {
if (!AFS_IS_DISCONNECTED) {
struct rx_connection *rxconn;
do {
tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME
(AFS_STATS_FS_RPCIDX_RELEASELOCK);
RX_AFS_GUNLOCK();
code =
RXAFS_ReleaseLock(rxconn,
(struct AFSFid *)&avc->
f.fid.Fid, &tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &avc->f.fid, areq,
AFS_STATS_FS_RPCIDX_RELEASELOCK, SHARED_LOCK,
NULL));
}
}
} else if (avc->flockCount == -1 && (acom & LOCK_EX)) {
if (lockIdcmp2(&flock, avc, NULL, 1, clid)) {
code = EWOULDBLOCK;
} else {
code = 0;
/* We've just re-grabbed an exclusive lock, so we don't
* need to contact the fileserver, and we don't need to
* add the lock to avc->slocks (since we already have a
* lock there). So, we are done. */
break;
}
}
if (code == 0) {
/* compatible here, decide if needs to go to file server. If
* we've already got the file locked (and thus read-locked, since
* we've already checked for compatibility), we shouldn't send
* the call through to the server again */
if (avc->flockCount == 0) {
struct rx_connection *rxconn;
/* we're the first on our block, send the call through */
lockType = ((acom & LOCK_EX) ? LockWrite : LockRead);
if (!AFS_IS_DISCONNECTED) {
do {
tc = afs_Conn(&avc->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 *)
&avc->f.fid.Fid, lockType,
&tsync);
RX_AFS_GLOCK();
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &avc->f.fid, areq,
AFS_STATS_FS_RPCIDX_SETLOCK, SHARED_LOCK,
NULL));
if ((lockType == LockWrite) && (code == VREADONLY))
code = EBADF; /* per POSIX; VREADONLY == EROFS */
} else
/* XXX - Should probably try and log this when we're
* XXX - running with logging enabled. But it's horrid
*/
code = 0; /* pretend we worked - ick!!! */
} else
code = 0; /* otherwise, pretend things worked */
}
if (code == 0) {
slp = (struct SimpleLocks *)
osi_AllocSmallSpace(sizeof(struct SimpleLocks));
if (acom & LOCK_EX) {
/* defect 3083 */
#ifdef AFS_AIX_ENV
/* Record unique id of process owning exclusive lock. */
avc->ownslock = getpid();
#endif
slp->type = LockWrite;
slp->next = NULL;
avc->slocks = slp;
avc->flockCount = -1;
} else {
slp->type = LockRead;
slp->next = avc->slocks;
avc->slocks = slp;
avc->flockCount++;
}
lockIdSet(&flock, slp, clid);
break;
}
/* now, if we got EWOULDBLOCK, and we're supposed to wait, we do */
if (((code == EWOULDBLOCK) || (code == EAGAIN) ||
(code == UAEWOULDBLOCK) || (code == UAEAGAIN))
&& !(acom & LOCK_NB)) {
/* sleep for a second, allowing interrupts */
ReleaseWriteLock(&avc->lock);
#if defined(AFS_SGI_ENV)
AFS_RWUNLOCK((vnode_t *) avc, VRWLOCK_WRITE);
#endif
code = afs_osi_Wait(1000, NULL, 1);
#if defined(AFS_SGI_ENV)
AFS_RWLOCK((vnode_t *) avc, VRWLOCK_WRITE);
#endif
ObtainWriteLock(&avc->lock, 120);
if (code) {
code = EINTR; /* return this if ^C typed */
break;
}
} else
break;
} /* while loop */
}
ReleaseWriteLock(&avc->lock);
code = afs_CheckCode(code, areq, 1); /* defeat a buggy AIX optimization */
return code;
}
/* warn a user that a lock has been ignored */
afs_int32 lastWarnTime = 0; /* this is used elsewhere */
static afs_int32 lastWarnPid = 0;
static void
DoLockWarning(afs_ucred_t * acred)
{
afs_int32 now;
pid_t pid = MyPidxx2Pid(MyPidxx);
char *procname;
now = osi_Time();
AFS_STATCNT(DoLockWarning);
/* check if we've already warned this user recently */
if (!((now < lastWarnTime + 120) && (lastWarnPid == pid))) {
procname = afs_osi_Alloc(256);
if (!procname)
return;
/* Copies process name to allocated procname, see osi_machdeps for details of macro */
osi_procname(procname, 256);
procname[255] = '\0';
/* otherwise, it is time to nag the user */
lastWarnTime = now;
lastWarnPid = pid;
#ifdef AFS_LINUX26_ENV
afs_warnuser
("afs: byte-range locks only enforced for processes on this machine (pid %d (%s), user %ld).\n", pid, procname, (long)afs_cr_uid(acred));
#else
afs_warnuser
("afs: byte-range lock/unlock ignored; make sure no one else is running this program (pid %d (%s), user %ld).\n", pid, procname, (long)afs_cr_uid(acred));
#endif
afs_osi_Free(procname, 256);
}
return;
}
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;
}
/* 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;
}
}
afs_rmdir(OSI_VC_DECL(adp), char *aname, afs_ucred_t *acred)
#endif
{
struct vrequest *treq = NULL;
struct dcache *tdc;
struct vcache *tvc = NULL;
afs_int32 code;
struct afs_conn *tc;
afs_size_t offset, len;
struct AFSFetchStatus OutDirStatus;
struct AFSVolSync tsync;
struct afs_fakestat_state fakestate;
struct rx_connection *rxconn;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_rmdir);
afs_Trace2(afs_iclSetp, CM_TRACE_RMDIR, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname);
if ((code = afs_CreateReq(&treq, acred)))
goto done2;
afs_InitFakeStat(&fakestate);
if (strlen(aname) > AFSNAMEMAX) {
code = ENAMETOOLONG;
goto done;
}
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, treq);
if (code)
goto done;
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) {
/* Disconnected read only mode. */
code = ENETDOWN;
goto done;
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, treq, &offset, &len, 1); /* test for error below */
ObtainWriteLock(&adp->lock, 154);
if (tdc)
ObtainSharedLock(&tdc->lock, 633);
if (tdc && (adp->f.states & CForeign)) {
struct VenusFid unlinkFid;
unlinkFid.Fid.Vnode = 0;
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, 1 /* do xstats */ );
ReleaseReadLock(&afs_xvcache);
}
}
}
if (!AFS_IS_DISCON_RW) {
/* Not disconnected, can connect to server. */
do {
tc = afs_Conn(&adp->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_REMOVEDIR);
RX_AFS_GUNLOCK();
code =
RXAFS_RemoveDir(rxconn,
(struct AFSFid *)&adp->f.fid.Fid,
aname,
&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_REMOVEDIR,
SHARED_LOCK, NULL));
if (code) {
if (tdc) {
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
if (code < 0) {
afs_StaleVCache(adp);
}
ReleaseWriteLock(&adp->lock);
goto done;
}
/* here if rpc worked; update the in-core link count */
adp->f.m.LinkCount = OutDirStatus.LinkCount;
} else {
/* Disconnected. */
if (!tdc) {
ReleaseWriteLock(&adp->lock);
/* printf("afs_rmdir: No local dcache!\n"); */
code = ENETDOWN;
goto done;
}
if (!tvc) {
/* Find the vcache. */
struct VenusFid tfid;
tfid.Cell = adp->f.fid.Cell;
tfid.Fid.Volume = adp->f.fid.Fid.Volume;
code = afs_dir_Lookup(tdc, aname, &tfid.Fid);
ObtainSharedLock(&afs_xvcache, 764);
tvc = afs_FindVCache(&tfid, 0, 1 /* do xstats */ );
ReleaseSharedLock(&afs_xvcache);
if (!tvc) {
/* printf("afs_rmdir: Can't find dir's vcache!\n"); */
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc); /* drop ref count */
ReleaseWriteLock(&adp->lock);
code = ENETDOWN;
goto done;
}
}
if (tvc->f.m.LinkCount > 2) {
/* This dir contains more than . and .., thus it can't be
* deleted.
*/
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
afs_PutVCache(tvc);
ReleaseWriteLock(&adp->lock);
code = ENOTEMPTY;
goto done;
}
/* Make a shadow copy of the parent dir (if not done already).
* If we were created locally, then we don't need to have a shadow
* directory (as there's no server state to remember)
*/
if (!adp->f.shadow.vnode && !(adp->f.ddirty_flags & VDisconCreate)) {
afs_MakeShadowDir(adp, tdc);
}
adp->f.m.LinkCount--;
} /* if (!AFS_IS_DISCON_RW) */
if (tdc)
UpgradeSToWLock(&tdc->lock, 634);
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 */
}
if (tvc)
osi_dnlc_purgedp(tvc); /* get rid of any entries for this directory */
else
osi_dnlc_remove(adp, aname, 0);
if (tvc) {
ObtainWriteLock(&tvc->lock, 155);
tvc->f.states &= ~CUnique; /* For the dfs xlator */
if (AFS_IS_DISCON_RW) {
if (tvc->f.ddirty_flags & VDisconCreate) {
/* If we we were created whilst disconnected, removal doesn't
* need to get logged. Just go away gracefully */
afs_DisconRemoveDirty(tvc);
} else {
afs_DisconAddDirty(tvc, VDisconRemove, 1);
}
}
ReleaseWriteLock(&tvc->lock);
afs_PutVCache(tvc);
}
ReleaseWriteLock(&adp->lock);
/* don't worry about link count since dirs can not be hardlinked */
code = 0;
done:
AFS_DISCON_UNLOCK();
afs_PutFakeStat(&fakestate);
code = afs_CheckCode(code, treq, 27);
afs_DestroyReq(treq);
done2:
return code;
}
