/* Generic read interface */
int
afs_osi_Read(struct osi_file *afile, int offset, void *aptr,
afs_int32 asize)
{
afs_ucred_t *oldCred;
unsigned int resid;
afs_int32 code;
afs_int32 cnt1 = 0;
AFS_STATCNT(osi_Read);
/**
* If the osi_file passed in is NULL, panic only if AFS is not shutting
* down. No point in crashing when we are already shutting down
*/
if (!afile) {
if (!afs_shuttingdown)
osi_Panic("osi_Read called with null param");
else
return -EIO;
}
if (offset != -1)
afile->offset = offset;
retry_IO:
/* Note the difference in the way the afile->offset is passed (see comments in gop_rdwr() in afs_aix_subr.c for comments) */
AFS_GUNLOCK();
#ifdef AFS_64BIT_KERNEL
code =
gop_rdwr(UIO_READ, afile->vnode, (caddr_t) aptr, asize,
&afile->offset, AFS_UIOSYS, NULL, &resid);
#else
code =
gop_rdwr(UIO_READ, afile->vnode, (caddr_t) aptr, asize,
(off_t) & afile->offset, AFS_UIOSYS, NULL, &resid);
#endif
AFS_GLOCK();
if (code == 0) {
code = asize - resid;
afile->offset += code;
osi_DisableAtimes(afile->vnode);
} else {
afs_Trace2(afs_iclSetp, CM_TRACE_READFAILED, ICL_TYPE_INT32, resid,
ICL_TYPE_INT32, code);
/*
* To handle periodic low-level EFAULT failures that we've seen with the
* Weitek chip; in all observed failed cases a second read succeeded.
*/
if ((code == EFAULT) && (cnt1++ < 5)) {
afs_stats_cmperf.osiread_efaults++;
goto retry_IO;
}
setuerror(code);
if (code > 0) {
code = -code;
}
}
return code;
}
/*!
* Look up AFSDB for given cell name and create locally.
* \param acellName Cell name.
*/
void
afs_LookupAFSDB(char *acellName)
{
int code;
char *cellName = afs_strdup(acellName);
code = afs_GetCellHostsAFSDB(cellName);
afs_Trace2(afs_iclSetp, CM_TRACE_AFSDB, ICL_TYPE_STRING, cellName,
ICL_TYPE_INT32, code);
afs_osi_FreeStr(cellName);
}
/* afs_root - stat the root of the file system. AFS global held on entry. */
static int
afs_root(struct super_block *afsp)
{
afs_int32 code = 0;
struct vcache *tvp = 0;
AFS_STATCNT(afs_root);
if (afs_globalVp && (afs_globalVp->f.states & CStatd)) {
tvp = afs_globalVp;
} else {
struct vrequest *treq = NULL;
cred_t *credp = crref();
if (afs_globalVp) {
afs_PutVCache(afs_globalVp);
afs_globalVp = NULL;
}
if (!(code = afs_CreateReq(&treq, credp)) && !(code = afs_CheckInit())) {
tvp = afs_GetVCache(&afs_rootFid, treq, NULL, NULL);
if (tvp) {
struct inode *ip = AFSTOV(tvp);
struct vattr *vattr = NULL;
code = afs_CreateAttr(&vattr);
if (!code) {
afs_getattr(tvp, vattr, credp);
afs_fill_inode(ip, vattr);
/* setup super_block and mount point inode. */
afs_globalVp = tvp;
#if defined(HAVE_LINUX_D_MAKE_ROOT)
afsp->s_root = d_make_root(ip);
#else
afsp->s_root = d_alloc_root(ip);
#endif
#if !defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
afsp->s_root->d_op = &afs_dentry_operations;
#endif
afs_DestroyAttr(vattr);
}
} else
code = EIO;
}
crfree(credp);
afs_DestroyReq(treq);
}
afs_Trace2(afs_iclSetp, CM_TRACE_VFSROOT, ICL_TYPE_POINTER, afs_globalVp,
ICL_TYPE_INT32, code);
return code;
}
/* Generic read interface */
int
afs_osi_Read(struct osi_file *afile, int offset, void *aptr,
afs_int32 asize)
{
afs_ucred_t *oldCred;
afs_size_t resid;
afs_int32 code;
#ifdef AFS_DARWIN80_ENV
uio_t uio;
#endif
AFS_STATCNT(osi_Read);
/**
* If the osi_file passed in is NULL, panic only if AFS is not shutting
* down. No point in crashing when we are already shutting down
*/
if (!afile) {
if (afs_shuttingdown == AFS_RUNNING)
osi_Panic("osi_Read called with null param");
else
return -EIO;
}
if (offset != -1)
afile->offset = offset;
AFS_GUNLOCK();
#ifdef AFS_DARWIN80_ENV
uio=uio_create(1, afile->offset, AFS_UIOSYS, UIO_READ);
uio_addiov(uio, CAST_USER_ADDR_T(aptr), asize);
code = VNOP_READ(afile->vnode, uio, IO_UNIT, afs_osi_ctxtp);
resid = AFS_UIO_RESID(uio);
uio_free(uio);
#else
code =
gop_rdwr(UIO_READ, afile->vnode, (caddr_t) aptr, asize, afile->offset,
AFS_UIOSYS, IO_UNIT, &afs_osi_cred, &resid);
#endif
AFS_GLOCK();
if (code == 0) {
code = asize - resid;
afile->offset += code;
osi_DisableAtimes(afile->vnode);
} else {
afs_Trace2(afs_iclSetp, CM_TRACE_READFAILED, ICL_TYPE_INT32, resid,
ICL_TYPE_INT32, code);
if (code > 0) {
code *= -1;
}
}
return code;
}
/* Generic read interface */
int
afs_osi_Read(struct osi_file *afile, int offset, void *aptr,
afs_int32 asize)
{
afs_ucred_t *oldCred;
long resid;
afs_int32 code;
afs_int32 cnt1 = 0;
AFS_STATCNT(osi_Read);
/**
* If the osi_file passed in is NULL, panic only if AFS is not shutting
* down. No point in crashing when we are already shutting down
*/
if (!afile) {
if (afs_shuttingdown == AFS_RUNNING)
osi_Panic("osi_Read called with null param");
else
return -EIO;
}
if (offset != -1)
afile->offset = offset;
retry_IO:
AFS_GUNLOCK();
code =
gop_rdwr(UIO_READ, afile->vnode, (caddr_t) aptr, asize, afile->offset,
AFS_UIOSYS, IO_UNIT, &resid);
AFS_GLOCK();
if (code == 0) {
code = asize - resid;
afile->offset += code;
osi_DisableAtimes(afile->vnode);
} else {
afs_Trace2(afs_iclSetp, CM_TRACE_READFAILED, ICL_TYPE_INT32,
(afs_int32) resid, ICL_TYPE_INT32, code);
/*
* To handle periodic low-level EFAULT failures that we've seen with the
* Weitek chip; in all observed failed cases a second read succeeded.
*/
if ((code == EFAULT) && (cnt1++ < 5)) {
afs_stats_cmperf.osiread_efaults++;
goto retry_IO;
}
setuerror(code);
if (code > 0) {
code *= -1;
}
}
return code;
}
/* afs_root - stat the root of the file system. AFS global held on entry. */
static int
afs_root(struct super_block *afsp)
{
afs_int32 code = 0;
struct vrequest treq;
struct vcache *tvp = 0;
AFS_STATCNT(afs_root);
if (afs_globalVp && (afs_globalVp->f.states & CStatd)) {
tvp = afs_globalVp;
} else {
cred_t *credp = crref();
if (afs_globalVp) {
afs_PutVCache(afs_globalVp);
afs_globalVp = NULL;
}
if (!(code = afs_InitReq(&treq, credp)) && !(code = afs_CheckInit())) {
tvp = afs_GetVCache(&afs_rootFid, &treq, NULL, NULL);
if (tvp) {
struct inode *ip = AFSTOV(tvp);
struct vattr vattr;
afs_getattr(tvp, &vattr, credp);
afs_fill_inode(ip, &vattr);
/* setup super_block and mount point inode. */
afs_globalVp = tvp;
#if defined(AFS_LINUX24_ENV)
afsp->s_root = d_alloc_root(ip);
#else
afsp->s_root = d_alloc_root(ip, NULL);
#endif
afsp->s_root->d_op = &afs_dentry_operations;
} else
code = ENOENT;
}
crfree(credp);
}
afs_Trace2(afs_iclSetp, CM_TRACE_VFSROOT, ICL_TYPE_POINTER, afs_globalVp,
ICL_TYPE_INT32, code);
return code;
}
/* Generic read interface */
int
afs_osi_Read(struct osi_file *afile, int offset, void *aptr,
afs_int32 asize)
{
struct uio auio;
struct iovec iov;
afs_int32 code;
memset(&auio, 0, sizeof(auio));
memset(&iov, 0, sizeof(iov));
AFS_STATCNT(osi_Read);
/*
* If the osi_file passed in is NULL, panic only if AFS is not shutting
* down. No point in crashing when we are already shutting down
*/
if (!afile) {
if (afs_shuttingdown == AFS_RUNNING)
osi_Panic("osi_Read called with null param");
else
return -EIO;
}
if (offset != -1)
afile->offset = offset;
setup_uio(&auio, &iov, aptr, afile->offset, asize, UIO_READ, AFS_UIOSYS);
AFS_GUNLOCK();
code = osi_rdwr(afile, &auio, UIO_READ);
AFS_GLOCK();
if (code == 0) {
code = asize - auio.uio_resid;
afile->offset += code;
} else {
afs_Trace2(afs_iclSetp, CM_TRACE_READFAILED, ICL_TYPE_INT32, auio.uio_resid,
ICL_TYPE_INT32, code);
if (code > 0) {
code = -code;
}
}
return code;
}
/* Generic read interface */
int
afs_osi_Read(struct osi_file *afile, int offset, void *aptr,
afs_int32 asize)
{
afs_ucred_t *oldCred;
ssize_t resid;
afs_int32 code;
afs_int32 cnt1 = 0;
AFS_STATCNT(osi_Read);
/**
* If the osi_file passed in is NULL, panic only if AFS is not shutting
* down. No point in crashing when we are already shutting down
*/
if (!afile) {
if (!afs_shuttingdown)
osi_Panic("osi_Read called with null param");
else
return -EIO;
}
if (offset != -1)
afile->offset = offset;
AFS_GUNLOCK();
code =
gop_rdwr(UIO_READ, afile->vnode, (caddr_t) aptr, asize, afile->offset,
AFS_UIOSYS, 0, 0, afs_osi_credp, &resid);
AFS_GLOCK();
if (code == 0) {
code = asize - resid;
afile->offset += code;
osi_DisableAtimes(afile->vnode);
} else {
afs_Trace2(afs_iclSetp, CM_TRACE_READFAILED, ICL_TYPE_INT32, resid,
ICL_TYPE_INT32, code);
if (code > 0) {
code = -code;
}
}
return code;
}
int
afs_CheckCode(afs_int32 acode, struct vrequest *areq, int where)
{
AFS_STATCNT(afs_CheckCode);
if (acode) {
afs_Trace2(afs_iclSetp, CM_TRACE_CHECKCODE, ICL_TYPE_INT32, acode,
ICL_TYPE_INT32, where);
}
if ((acode & ~0xff) == ERROR_TABLE_BASE_uae)
acode = et_to_sys_error(acode);
if (!areq || !areq->initd)
return acode;
if (areq->networkError)
return ETIMEDOUT;
if (acode == 0)
return 0;
if (areq->accessError)
return EACCES;
if (areq->volumeError == VOLMISSING)
return ENODEV;
if (areq->volumeError == VOLBUSY)
return EWOULDBLOCK;
if (acode == VNOVNODE)
return ENOENT;
if (acode == VDISKFULL)
return ENOSPC;
if (acode == VOVERQUOTA)
return
#ifdef EDQUOT
EDQUOT
#else
ENOSPC
#endif
;
return acode;
} /*afs_CheckCode */
afs_root(struct mount *mp, struct vnode **vpp)
#endif
{
int error;
struct vrequest treq;
register struct vcache *tvp = 0;
#ifdef AFS_FBSD50_ENV
#ifndef AFS_FBSD53_ENV
struct thread *td = curthread;
#endif
struct ucred *cr = td->td_ucred;
#else
struct proc *p = curproc;
struct ucred *cr = p->p_cred->pc_ucred;
#endif
AFS_GLOCK();
AFS_STATCNT(afs_root);
crhold(cr);
if (afs_globalVp && (afs_globalVp->f.states & CStatd)) {
tvp = afs_globalVp;
error = 0;
} else {
tryagain:
#ifndef AFS_FBSD80_ENV
if (afs_globalVp) {
afs_PutVCache(afs_globalVp);
/* vrele() needed here or not? */
afs_globalVp = NULL;
}
#endif
if (!(error = afs_InitReq(&treq, cr)) && !(error = afs_CheckInit())) {
tvp = afs_GetVCache(&afs_rootFid, &treq, NULL, NULL);
/* we really want this to stay around */
if (tvp)
afs_globalVp = tvp;
else
error = ENOENT;
}
}
if (tvp) {
struct vnode *vp = AFSTOV(tvp);
#ifdef AFS_FBSD50_ENV
ASSERT_VI_UNLOCKED(vp, "afs_root");
#endif
AFS_GUNLOCK();
/*
* I'm uncomfortable about this. Shouldn't this happen at a
* higher level, and shouldn't we busy the top-level directory
* to prevent recycling?
*/
#ifdef AFS_FBSD50_ENV
error = vget(vp, LK_EXCLUSIVE | LK_RETRY, td);
vp->v_vflag |= VV_ROOT;
#else
error = vget(vp, LK_EXCLUSIVE | LK_RETRY, p);
vp->v_flag |= VROOT;
#endif
AFS_GLOCK();
if (error != 0)
goto tryagain;
afs_globalVFS = mp;
*vpp = vp;
}
afs_Trace2(afs_iclSetp, CM_TRACE_VFSROOT, ICL_TYPE_POINTER, tvp ? AFSTOV(tvp) : NULL,
ICL_TYPE_INT32, error);
AFS_GUNLOCK();
crfree(cr);
return error;
}
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;
}
static int
afs_StoreMini(struct vcache *avc, struct vrequest *areq)
{
struct afs_conn *tc;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus OutStatus;
struct AFSVolSync tsync;
afs_int32 code;
struct rx_call *tcall;
struct rx_connection *rxconn;
afs_size_t tlen, xlen = 0;
XSTATS_DECLS;
AFS_STATCNT(afs_StoreMini);
afs_Trace2(afs_iclSetp, CM_TRACE_STOREMINI, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->f.m.Length);
tlen = avc->f.m.Length;
if (avc->f.truncPos < tlen)
tlen = avc->f.truncPos;
avc->f.truncPos = AFS_NOTRUNC;
avc->f.states &= ~CExtendedFile;
do {
tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK, &rxconn);
if (tc) {
#ifdef AFS_64BIT_CLIENT
retry:
#endif
RX_AFS_GUNLOCK();
tcall = rx_NewCall(rxconn);
RX_AFS_GLOCK();
/* Set the client mod time since we always want the file
* to have the client's mod time and not the server's one
* (to avoid problems with make, etc.) It almost always
* works fine with standard afs because them server/client
* times are in sync and more importantly this storemini
* it's a special call that would typically be followed by
* the proper store-data or store-status calls.
*/
InStatus.Mask = AFS_SETMODTIME;
InStatus.ClientModTime = avc->f.m.Date;
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_STOREDATA);
afs_Trace4(afs_iclSetp, CM_TRACE_STOREDATA64, ICL_TYPE_FID,
&avc->f.fid.Fid, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(avc->f.m.Length), ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(xlen), ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(tlen));
RX_AFS_GUNLOCK();
#ifdef AFS_64BIT_CLIENT
if (!afs_serverHasNo64Bit(tc)) {
code =
StartRXAFS_StoreData64(tcall,
(struct AFSFid *)&avc->f.fid.Fid,
&InStatus, avc->f.m.Length,
(afs_size_t) 0, tlen);
} else {
afs_int32 l1, l2;
l1 = avc->f.m.Length;
l2 = tlen;
if ((avc->f.m.Length > 0x7fffffff) ||
(tlen > 0x7fffffff) ||
((0x7fffffff - tlen) < avc->f.m.Length)) {
code = EFBIG;
goto error;
}
code =
StartRXAFS_StoreData(tcall,
(struct AFSFid *)&avc->f.fid.Fid,
&InStatus, l1, 0, l2);
}
#else /* AFS_64BIT_CLIENT */
code =
StartRXAFS_StoreData(tcall, (struct AFSFid *)&avc->f.fid.Fid,
&InStatus, avc->f.m.Length, 0, tlen);
#endif /* AFS_64BIT_CLIENT */
if (code == 0) {
code = EndRXAFS_StoreData(tcall, &OutStatus, &tsync);
}
#ifdef AFS_64BIT_CLIENT
error:
#endif
code = rx_EndCall(tcall, code);
RX_AFS_GLOCK();
XSTATS_END_TIME;
#ifdef AFS_64BIT_CLIENT
if (code == RXGEN_OPCODE && !afs_serverHasNo64Bit(tc)) {
afs_serverSetNo64Bit(tc);
goto retry;
}
#endif /* AFS_64BIT_CLIENT */
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &avc->f.fid, areq, AFS_STATS_FS_RPCIDX_STOREDATA,
SHARED_LOCK, NULL));
if (code == 0)
afs_ProcessFS(avc, &OutStatus, areq);
return code;
} /*afs_StoreMini */
afs_open(struct vcache **avcp, afs_int32 aflags, afs_ucred_t *acred)
#endif
{
afs_int32 code;
struct vrequest treq;
struct vcache *tvc;
int writing;
struct afs_fakestat_state fakestate;
AFS_STATCNT(afs_open);
if ((code = afs_InitReq(&treq, acred)))
return code;
#ifdef AFS_SGI64_ENV
/* avcpp can be, but is not necesarily, bhp's vnode. */
tvc = VTOAFS(BHV_TO_VNODE(bhv));
#else
tvc = *avcp;
#endif
afs_Trace2(afs_iclSetp, CM_TRACE_OPEN, ICL_TYPE_POINTER, tvc,
ICL_TYPE_INT32, aflags);
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&tvc, &fakestate, &treq);
if (code)
goto done;
code = afs_VerifyVCache(tvc, &treq);
if (code)
goto done;
ObtainReadLock(&tvc->lock);
if (AFS_IS_DISCONNECTED && (afs_DCacheMissingChunks(tvc) != 0)) {
ReleaseReadLock(&tvc->lock);
/* printf("Network is down in afs_open: missing chunks\n"); */
code = ENETDOWN;
goto done;
}
ReleaseReadLock(&tvc->lock);
if (aflags & (FWRITE | FTRUNC))
writing = 1;
else
writing = 0;
if (vType(tvc) == VDIR) {
/* directory */
if (writing) {
code = EISDIR;
goto done;
} else {
if (!afs_AccessOK
(tvc, ((tvc->f.states & CForeign) ? PRSFS_READ : PRSFS_LOOKUP),
&treq, CHECK_MODE_BITS)) {
code = EACCES;
/* printf("afs_Open: no access for dir\n"); */
goto done;
}
}
} else {
#ifdef AFS_SUN5_ENV
if (AFS_NFSXLATORREQ(acred) && (aflags & FREAD)) {
if (!afs_AccessOK
(tvc, PRSFS_READ, &treq,
CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
code = EACCES;
goto done;
}
}
#endif
#ifdef AFS_AIX41_ENV
if (aflags & FRSHARE) {
/*
* Hack for AIX 4.1:
* Apparently it is possible for a file to get mapped without
* either VNOP_MAP or VNOP_RDWR being called, if (1) it is a
* sharable library, and (2) it has already been loaded. We must
* ensure that the credp is up to date. We detect the situation
* by checking for O_RSHARE at open time.
*/
/*
* We keep the caller's credentials since an async daemon will
* handle the request at some point. We assume that the same
* credentials will be used.
*/
ObtainWriteLock(&tvc->lock, 140);
if (!tvc->credp || (tvc->credp != acred)) {
crhold(acred);
if (tvc->credp) {
struct ucred *crp = tvc->credp;
tvc->credp = NULL;
crfree(crp);
}
tvc->credp = acred;
}
ReleaseWriteLock(&tvc->lock);
}
#endif
/* normal file or symlink */
osi_FlushText(tvc); /* only needed to flush text if text locked last time */
#ifdef AFS_BOZONLOCK_ENV
afs_BozonLock(&tvc->pvnLock, tvc);
#endif
osi_FlushPages(tvc, acred);
#ifdef AFS_BOZONLOCK_ENV
afs_BozonUnlock(&tvc->pvnLock, tvc);
#endif
}
/* set date on file if open in O_TRUNC mode */
if (aflags & FTRUNC) {
/* this fixes touch */
ObtainWriteLock(&tvc->lock, 123);
tvc->f.m.Date = osi_Time();
tvc->f.states |= CDirty;
ReleaseWriteLock(&tvc->lock);
}
ObtainReadLock(&tvc->lock);
if (writing)
tvc->execsOrWriters++;
tvc->opens++;
#if defined(AFS_SGI_ENV) || defined (AFS_LINUX26_ENV)
if (writing && tvc->cred == NULL) {
crhold(acred);
tvc->cred = acred;
}
#endif
ReleaseReadLock(&tvc->lock);
if ((afs_preCache != 0) && (writing == 0) && (vType(tvc) != VDIR) &&
(!afs_BBusy())) {
struct dcache *tdc;
afs_size_t offset, len;
tdc = afs_GetDCache(tvc, 0, &treq, &offset, &len, 1);
ObtainSharedLock(&tdc->mflock, 865);
if (!(tdc->mflags & DFFetchReq)) {
struct brequest *bp;
/* start the daemon (may already be running, however) */
UpgradeSToWLock(&tdc->mflock, 666);
tdc->mflags |= DFFetchReq; /* guaranteed to be cleared by BKG or
GetDCache */
/* last parm (1) tells bkg daemon to do an afs_PutDCache when it
is done, since we don't want to wait for it to finish before
doing so ourselves.
*/
bp = afs_BQueue(BOP_FETCH, tvc, B_DONTWAIT, 0, acred,
(afs_size_t) 0, (afs_size_t) 1, tdc,
(void *)0, (void *)0);
if (!bp) {
tdc->mflags &= ~DFFetchReq;
}
ReleaseWriteLock(&tdc->mflock);
} else {
ReleaseSharedLock(&tdc->mflock);
}
}
done:
afs_PutFakeStat(&fakestate);
AFS_DISCON_UNLOCK();
code = afs_CheckCode(code, &treq, 4); /* avoid AIX -O bug */
afs_Trace2(afs_iclSetp, CM_TRACE_OPEN, ICL_TYPE_POINTER, tvc,
ICL_TYPE_INT32, 999999);
return code;
}
int
afs_InvalidateAllSegments(struct vcache *avc)
{
struct dcache *tdc;
afs_int32 hash;
afs_int32 index;
struct dcache **dcList;
int i, dcListMax, dcListCount;
AFS_STATCNT(afs_InvalidateAllSegments);
afs_Trace2(afs_iclSetp, CM_TRACE_INVALL, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
hash = DVHash(&avc->f.fid);
avc->f.truncPos = AFS_NOTRUNC; /* don't truncate later */
avc->f.states &= ~CExtendedFile; /* not any more */
ObtainWriteLock(&afs_xcbhash, 459);
afs_DequeueCallback(avc);
avc->f.states &= ~(CStatd | CDirty); /* mark status information as bad, too */
ReleaseWriteLock(&afs_xcbhash);
if (avc->f.fid.Fid.Vnode & 1 || (vType(avc) == VDIR))
osi_dnlc_purgedp(avc);
/* Blow away pages; for now, only for Solaris */
#if (defined(AFS_SUN5_ENV))
if (WriteLocked(&avc->lock))
osi_ReleaseVM(avc, (afs_ucred_t *)0);
#endif
/*
* Block out others from screwing with this table; is a read lock
* sufficient?
*/
ObtainWriteLock(&afs_xdcache, 286);
dcListMax = 0;
for (index = afs_dvhashTbl[hash]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetDSlot(index, 0);
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid))
dcListMax++;
afs_PutDCache(tdc);
}
index = afs_dvnextTbl[index];
}
dcList = osi_Alloc(dcListMax * sizeof(struct dcache *));
dcListCount = 0;
for (index = afs_dvhashTbl[hash]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetDSlot(index, 0);
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
/* same file? we'll zap it */
if (afs_indexFlags[index] & IFDataMod) {
afs_stats_cmperf.cacheCurrDirtyChunks--;
/* don't write it back */
afs_indexFlags[index] &= ~IFDataMod;
}
afs_indexFlags[index] &= ~IFAnyPages;
if (dcListCount < dcListMax)
dcList[dcListCount++] = tdc;
else
afs_PutDCache(tdc);
} else {
afs_PutDCache(tdc);
}
}
index = afs_dvnextTbl[index];
}
ReleaseWriteLock(&afs_xdcache);
for (i = 0; i < dcListCount; i++) {
tdc = dcList[i];
ObtainWriteLock(&tdc->lock, 679);
ZapDCE(tdc);
if (vType(avc) == VDIR)
DZap(tdc);
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
osi_Free(dcList, dcListMax * sizeof(struct dcache *));
return 0;
}
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;
}
/* 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
afs_getattr(OSI_VC_DECL(avc), struct vattr *attrs, afs_ucred_t *acred)
#endif
{
afs_int32 code;
struct vrequest *treq = NULL;
struct unixuser *au;
int inited = 0;
OSI_VC_CONVERT(avc);
AFS_STATCNT(afs_getattr);
afs_Trace2(afs_iclSetp, CM_TRACE_GETATTR, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
if (afs_fakestat_enable && avc->mvstat == AFS_MVSTAT_MTPT) {
struct afs_fakestat_state fakestat;
struct vrequest *ureq = NULL;
code = afs_CreateReq(&ureq, acred);
if (code) {
return code;
}
afs_InitFakeStat(&fakestat);
code = afs_TryEvalFakeStat(&avc, &fakestat, ureq);
if (code) {
afs_PutFakeStat(&fakestat);
afs_DestroyReq(ureq);
return code;
}
code = afs_CopyOutAttrs(avc, attrs);
afs_PutFakeStat(&fakestat);
afs_DestroyReq(ureq);
return code;
}
#if defined(AFS_SUN5_ENV)
if (flags & ATTR_HINT) {
code = afs_CopyOutAttrs(avc, attrs);
return code;
}
#endif
#if defined(AFS_DARWIN_ENV) && !defined(AFS_DARWIN80_ENV)
if (avc->f.states & CUBCinit) {
code = afs_CopyOutAttrs(avc, attrs);
return code;
}
#endif
AFS_DISCON_LOCK();
if (afs_shuttingdown != AFS_RUNNING) {
AFS_DISCON_UNLOCK();
return EIO;
}
if (!(avc->f.states & CStatd)) {
if (!(code = afs_CreateReq(&treq, acred))) {
code = afs_VerifyVCache2(avc, treq);
inited = 1;
}
} else
code = 0;
#if defined(AFS_SUN5_ENV)
if (code == 0)
osi_FlushPages(avc, acred);
#endif
if (code == 0) {
osi_FlushText(avc); /* only needed to flush text if text locked last time */
code = afs_CopyOutAttrs(avc, attrs);
if (afs_nfsexporter) {
if (!inited) {
if ((code = afs_CreateReq(&treq, acred))) {
return code;
}
inited = 1;
}
if (AFS_NFSXLATORREQ(acred)) {
if ((vType(avc) != VDIR)
&& !afs_AccessOK(avc, PRSFS_READ, treq,
CHECK_MODE_BITS |
CMB_ALLOW_EXEC_AS_READ)) {
afs_DestroyReq(treq);
return EACCES;
}
}
if ((au = afs_FindUser(treq->uid, -1, READ_LOCK))) {
struct afs_exporter *exporter = au->exporter;
if (exporter && !(afs_nfsexporter->exp_states & EXP_UNIXMODE)) {
unsigned int ubits;
/*
* If the remote user wishes to enforce default Unix mode semantics,
* like in the nfs exporter case, we OR in the user bits
* into the group and other bits. We need to do this
* because there is no RFS_ACCESS call and thus nfs
* clients implement nfs_access by interpreting the
* mode bits in the traditional way, which of course
* loses with afs.
*/
ubits = (attrs->va_mode & 0700) >> 6;
attrs->va_mode = attrs->va_mode | ubits | (ubits << 3);
/* If it's the root of AFS, replace the inode number with the
* inode number of the mounted on directory; otherwise this
* confuses getwd()... */
#ifdef AFS_LINUX22_ENV
if (avc == afs_globalVp) {
struct inode *ip = AFSTOV(avc)->i_sb->s_root->d_inode;
attrs->va_nodeid = ip->i_ino; /* VTOI()? */
}
#else
if (
#if defined(AFS_DARWIN_ENV)
vnode_isvroot(AFSTOV(avc))
#elif defined(AFS_NBSD50_ENV)
AFSTOV(avc)->v_vflag & VV_ROOT
#else
AFSTOV(avc)->v_flag & VROOT
#endif
) {
struct vnode *vp = AFSTOV(avc);
#ifdef AFS_DARWIN80_ENV
/* XXX vp = vnode_mount(vp)->mnt_vnodecovered; */
vp = 0;
#else
vp = vp->v_vfsp->vfs_vnodecovered;
if (vp) { /* Ignore weird failures */
#ifdef AFS_SGI62_ENV
attrs->va_nodeid = VnodeToIno(vp);
#else
struct inode *ip;
ip = (struct inode *)VTOI(vp);
if (ip) /* Ignore weird failures */
attrs->va_nodeid = ip->i_number;
#endif
}
#endif
}
#endif /* AFS_LINUX22_ENV */
}
afs_PutUser(au, READ_LOCK);
}
}
int
afs_StoreAllSegments(struct vcache *avc, struct vrequest *areq,
int sync)
{
struct dcache *tdc;
afs_int32 code = 0;
afs_int32 index;
afs_int32 origCBs, foreign = 0;
int hash;
afs_hyper_t newDV, oldDV; /* DV when we start, and finish, respectively */
struct dcache **dcList;
unsigned int i, j, minj, moredata, high, off;
afs_size_t maxStoredLength; /* highest offset we've written to server. */
int safety, marineronce = 0;
AFS_STATCNT(afs_StoreAllSegments);
hset(oldDV, avc->f.m.DataVersion);
hset(newDV, avc->f.m.DataVersion);
hash = DVHash(&avc->f.fid);
foreign = (avc->f.states & CForeign);
dcList = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
afs_Trace2(afs_iclSetp, CM_TRACE_STOREALL, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
#if !defined(AFS_AIX32_ENV) && !defined(AFS_SGI65_ENV)
/* In the aix vm implementation we need to do the vm_writep even
* on the memcache case since that's we adjust the file's size
* and finish flushing partial vm pages.
*/
if ((cacheDiskType != AFS_FCACHE_TYPE_MEM) ||
(sync & AFS_VMSYNC_INVAL) || (sync & AFS_VMSYNC) ||
(sync & AFS_LASTSTORE))
#endif /* !AFS_AIX32_ENV && !AFS_SGI65_ENV */
{
/* If we're not diskless, reading a file may stress the VM
* system enough to cause a pageout, and this vnode would be
* locked when the pageout occurs. We can prevent this problem
* by making sure all dirty pages are already flushed. We don't
* do this when diskless because reading a diskless (i.e.
* memory-resident) chunk doesn't require using new VM, and we
* also don't want to dump more dirty data into a diskless cache,
* since they're smaller, and we might exceed its available
* space.
*/
#if defined(AFS_SUN5_ENV)
if (sync & AFS_VMSYNC_INVAL) /* invalidate VM pages */
osi_VM_TryToSmush(avc, CRED(), 1);
else
#endif
osi_VM_StoreAllSegments(avc);
}
if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
/* This will probably make someone sad ... */
/*printf("Net down in afs_StoreSegments\n");*/
return ENETDOWN;
}
ConvertWToSLock(&avc->lock);
/*
* Subsequent code expects a sorted list, and it expects all the
* chunks in the list to be contiguous, so we need a sort and a
* while loop in here, too - but this will work for a first pass...
* 92.10.05 - OK, there's a sort in here now. It's kind of a modified
* bin sort, I guess. Chunk numbers start with 0
*
* - Have to get a write lock on xdcache because GetDSlot might need it (if
* the chunk doesn't have a dcache struct).
* This seems like overkill in most cases.
* - I'm not sure that it's safe to do "index = .hvNextp", then unlock
* xdcache, then relock xdcache and try to use index. It is done
* a lot elsewhere in the CM, but I'm not buying that argument.
* - should be able to check IFDataMod without doing the GetDSlot (just
* hold afs_xdcache). That way, it's easy to do this without the
* writelock on afs_xdcache, and we save unneccessary disk
* operations. I don't think that works, 'cuz the next pointers
* are still on disk.
*/
origCBs = afs_allCBs;
maxStoredLength = 0;
minj = 0;
do {
memset(dcList, 0, NCHUNKSATONCE * sizeof(struct dcache *));
high = 0;
moredata = FALSE;
/* lock and start over from beginning of hash chain
* in order to avoid a race condition. */
ObtainWriteLock(&afs_xdcache, 284);
index = afs_dvhashTbl[hash];
for (j = 0; index != NULLIDX;) {
if ((afs_indexFlags[index] & IFDataMod)
&& (afs_indexUnique[index] == avc->f.fid.Fid.Unique)) {
tdc = afs_GetValidDSlot(index); /* refcount+1. */
if (!tdc) {
ReleaseWriteLock(&afs_xdcache);
code = EIO;
goto done;
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid) && tdc->f.chunk >= minj) {
off = tdc->f.chunk - minj;
if (off < NCHUNKSATONCE) {
if (dcList[off])
osi_Panic("dclist slot already in use!");
if (afs_mariner && !marineronce) {
/* first chunk only */
afs_MarinerLog("store$Storing", avc);
marineronce++;
}
dcList[off] = tdc;
if (off > high)
high = off;
j++;
/* DCLOCKXXX: chunkBytes is protected by tdc->lock which we
* can't grab here, due to lock ordering with afs_xdcache.
* So, disable this shortcut for now. -- kolya 2001-10-13
*/
/* shortcut: big win for little files */
/* tlen -= tdc->f.chunkBytes;
* if (tlen <= 0)
* break;
*/
} else {
moredata = TRUE;
afs_PutDCache(tdc);
if (j == NCHUNKSATONCE)
break;
}
} else {
afs_PutDCache(tdc);
}
}
index = afs_dvnextTbl[index];
}
ReleaseWriteLock(&afs_xdcache);
/* this guy writes chunks, puts back dcache structs, and bumps newDV */
/* "moredata" just says "there are more dirty chunks yet to come".
*/
if (j) {
code =
afs_CacheStoreVCache(dcList, avc, areq, sync,
minj, high, moredata,
&newDV, &maxStoredLength);
/* Release any zero-length dcache entries in our interval
* that we locked but didn't store back above.
*/
for (j = 0; j <= high; j++) {
tdc = dcList[j];
if (tdc) {
osi_Assert(tdc->f.chunkBytes == 0);
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
}
}
/* if (j) */
minj += NCHUNKSATONCE;
} while (!code && moredata);
done:
UpgradeSToWLock(&avc->lock, 29);
/* send a trivial truncation store if did nothing else */
if (code == 0) {
/*
* Call StoreMini if we haven't written enough data to extend the
* file at the fileserver to the client's notion of the file length.
*/
if ((avc->f.truncPos != AFS_NOTRUNC)
|| ((avc->f.states & CExtendedFile)
&& (maxStoredLength < avc->f.m.Length))) {
code = afs_StoreMini(avc, areq);
if (code == 0)
hadd32(newDV, 1); /* just bumped here, too */
}
avc->f.states &= ~CExtendedFile;
}
/*
* Finally, turn off DWriting, turn on DFEntryMod,
* update f.versionNo.
* A lot of this could be integrated into the loop above
*/
if (!code) {
afs_hyper_t h_unset;
hones(h_unset);
minj = 0;
do {
moredata = FALSE;
memset(dcList, 0,
NCHUNKSATONCE * sizeof(struct dcache *));
/* overkill, but it gets the lock in case GetDSlot needs it */
ObtainWriteLock(&afs_xdcache, 285);
for (j = 0, safety = 0, index = afs_dvhashTbl[hash];
index != NULLIDX && safety < afs_cacheFiles + 2;
index = afs_dvnextTbl[index]) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
/* This is okay; since manipulating the dcaches at this
* point is best-effort. We only get a dcache here to
* increment the dv and turn off DWriting. If we were
* supposed to do that for a dcache, but could not
* due to an I/O error, it just means the dv won't
* be updated so we don't be able to use that cached
* chunk in the future. That's inefficient, but not
* an error. */
continue;
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid)
&& tdc->f.chunk >= minj) {
off = tdc->f.chunk - minj;
if (off < NCHUNKSATONCE) {
/* this is the file, and the correct chunk range */
if (j >= NCHUNKSATONCE)
osi_Panic
("Too many dcache entries in range\n");
dcList[j++] = tdc;
} else {
moredata = TRUE;
afs_PutDCache(tdc);
if (j == NCHUNKSATONCE)
break;
}
} else {
afs_PutDCache(tdc);
}
}
}
ReleaseWriteLock(&afs_xdcache);
for (i = 0; i < j; i++) {
/* Iterate over the dcache entries we collected above */
tdc = dcList[i];
ObtainSharedLock(&tdc->lock, 677);
/* was code here to clear IFDataMod, but it should only be done
* in storedcache and storealldcache.
*/
/* Only increase DV if we had up-to-date data to start with.
* Otherwise, we could be falsely upgrading an old chunk
* (that we never read) into one labelled with the current
* DV #. Also note that we check that no intervening stores
* occurred, otherwise we might mislabel cache information
* for a chunk that we didn't store this time
*/
/* Don't update the version number if it's not yet set. */
if (!hsame(tdc->f.versionNo, h_unset)
&& hcmp(tdc->f.versionNo, oldDV) >= 0) {
if ((!(afs_dvhack || foreign)
&& hsame(avc->f.m.DataVersion, newDV))
|| ((afs_dvhack || foreign)
&& (origCBs == afs_allCBs))) {
/* no error, this is the DV */
UpgradeSToWLock(&tdc->lock, 678);
hset(tdc->f.versionNo, avc->f.m.DataVersion);
tdc->dflags |= DFEntryMod;
/* DWriting may not have gotten cleared above, if all
* we did was a StoreMini */
tdc->f.states &= ~DWriting;
ConvertWToSLock(&tdc->lock);
}
}
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
minj += NCHUNKSATONCE;
} while (moredata);
}
if (code) {
/*
* Invalidate chunks after an error for ccores files since
* afs_inactive won't be called for these and they won't be
* invalidated. Also discard data if it's a permanent error from the
* fileserver.
*/
if (areq->permWriteError || (avc->f.states & CCore)) {
afs_InvalidateAllSegments(avc);
}
}
afs_Trace3(afs_iclSetp, CM_TRACE_STOREALLDONE, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->f.m.Length, ICL_TYPE_INT32, code);
/* would like a Trace5, but it doesn't exist... */
afs_Trace3(afs_iclSetp, CM_TRACE_AVCLOCKER, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->lock.wait_states, ICL_TYPE_INT32,
avc->lock.excl_locked);
afs_Trace4(afs_iclSetp, CM_TRACE_AVCLOCKEE, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, avc->lock.wait_states, ICL_TYPE_INT32,
avc->lock.readers_reading, ICL_TYPE_INT32,
avc->lock.num_waiting);
/*
* Finally, if updated DataVersion matches newDV, we did all of the
* stores. If mapDV indicates that the page cache was flushed up
* to when we started the store, then we can relabel them as flushed
* as recently as newDV.
* Turn off CDirty bit because the stored data is now in sync with server.
*/
if (code == 0 && hcmp(avc->mapDV, oldDV) >= 0) {
if ((!(afs_dvhack || foreign) && hsame(avc->f.m.DataVersion, newDV))
|| ((afs_dvhack || foreign) && (origCBs == afs_allCBs))) {
hset(avc->mapDV, newDV);
avc->f.states &= ~CDirty;
}
}
osi_FreeLargeSpace(dcList);
/* If not the final write a temporary error is ok. */
if (code && !areq->permWriteError && !(sync & AFS_LASTSTORE))
code = 0;
return code;
} /*afs_StoreAllSegments (new 03/02/94) */
/* don't set CDirty in here because RPC is called synchronously */
int
afs_symlink(OSI_VC_DECL(adp), char *aname, struct vattr *attrs,
char *atargetName, struct vcache **tvcp, afs_ucred_t *acred)
{
afs_uint32 now = 0;
struct vrequest *treq = NULL;
afs_int32 code = 0;
struct afs_conn *tc;
struct VenusFid newFid;
struct dcache *tdc;
afs_size_t offset, len;
afs_int32 alen;
struct server *hostp = 0;
struct vcache *tvc;
struct AFSStoreStatus InStatus;
struct AFSFetchStatus *OutFidStatus, *OutDirStatus;
struct AFSCallBack CallBack;
struct AFSVolSync tsync;
struct volume *volp = 0;
struct afs_fakestat_state fakestate;
struct rx_connection *rxconn;
XSTATS_DECLS;
OSI_VC_CONVERT(adp);
AFS_STATCNT(afs_symlink);
afs_Trace2(afs_iclSetp, CM_TRACE_SYMLINK, ICL_TYPE_POINTER, adp,
ICL_TYPE_STRING, aname);
OutFidStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
OutDirStatus = osi_AllocSmallSpace(sizeof(struct AFSFetchStatus));
memset(&InStatus, 0, sizeof(InStatus));
if ((code = afs_CreateReq(&treq, acred)))
goto done2;
afs_InitFakeStat(&fakestate);
AFS_DISCON_LOCK();
code = afs_EvalFakeStat(&adp, &fakestate, treq);
if (code)
goto done;
if (strlen(aname) > AFSNAMEMAX || strlen(atargetName) > AFSPATHMAX) {
code = ENAMETOOLONG;
goto done;
}
if (afs_IsDynroot(adp)) {
code = afs_DynrootVOPSymlink(adp, acred, aname, atargetName);
goto done;
}
if (afs_IsDynrootMount(adp)) {
code = EROFS;
goto done;
}
code = afs_VerifyVCache(adp, treq);
if (code) {
code = afs_CheckCode(code, treq, 30);
goto done;
}
/** If the volume is read-only, return error without making an RPC to the
* fileserver
*/
if (adp->f.states & CRO) {
code = EROFS;
goto done;
}
if (AFS_IS_DISCONNECTED && !AFS_IS_DISCON_RW) {
code = ENETDOWN;
goto done;
}
InStatus.Mask = AFS_SETMODTIME | AFS_SETMODE;
InStatus.ClientModTime = osi_Time();
alen = strlen(atargetName); /* we want it to include the null */
if ( (*atargetName == '#' || *atargetName == '%') && alen > 1 && atargetName[alen-1] == '.') {
InStatus.UnixModeBits = 0644; /* mt pt: null from "." at end */
if (alen == 1)
alen++; /* Empty string */
} else {
InStatus.UnixModeBits = 0755;
alen++; /* add in the null */
}
tdc = afs_GetDCache(adp, (afs_size_t) 0, treq, &offset, &len, 1);
volp = afs_FindVolume(&adp->f.fid, READ_LOCK); /*parent is also in same vol */
ObtainWriteLock(&adp->lock, 156);
if (tdc)
ObtainWriteLock(&tdc->lock, 636);
/* No further locks: if the SymLink succeeds, it does not matter what happens
* to our local copy of the directory. If somebody tampers with it in the meantime,
* the copy will be invalidated */
if (!AFS_IS_DISCON_RW) {
do {
tc = afs_Conn(&adp->f.fid, treq, SHARED_LOCK, &rxconn);
if (tc) {
hostp = tc->parent->srvr->server;
XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_SYMLINK);
if (adp->f.states & CForeign) {
now = osi_Time();
RX_AFS_GUNLOCK();
code =
RXAFS_DFSSymlink(rxconn,
(struct AFSFid *)&adp->f.fid.Fid,
aname, atargetName, &InStatus,
(struct AFSFid *)&newFid.Fid,
OutFidStatus, OutDirStatus,
&CallBack, &tsync);
RX_AFS_GLOCK();
} else {
RX_AFS_GUNLOCK();
code =
RXAFS_Symlink(rxconn, (struct AFSFid *)&adp->f.fid.Fid,
aname, atargetName, &InStatus,
(struct AFSFid *)&newFid.Fid,
OutFidStatus, OutDirStatus, &tsync);
RX_AFS_GLOCK();
}
XSTATS_END_TIME;
} else
code = -1;
} while (afs_Analyze
(tc, rxconn, code, &adp->f.fid, treq, AFS_STATS_FS_RPCIDX_SYMLINK,
SHARED_LOCK, NULL));
} else {
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
afs_GenFakeFid(&newFid, VREG, 0);
}
ObtainWriteLock(&afs_xvcache, 40);
if (code) {
if (code < 0) {
afs_StaleVCache(adp);
}
ReleaseWriteLock(&adp->lock);
ReleaseWriteLock(&afs_xvcache);
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
goto done;
}
/* otherwise, we should see if we can make the change to the dir locally */
if (AFS_IS_DISCON_RW || afs_LocalHero(adp, tdc, OutDirStatus, 1)) {
/* we can do it locally */
ObtainWriteLock(&afs_xdcache, 293);
/* If the following fails because the name has been created in the meantime, the
* directory is out-of-date - the file server knows best! */
code = afs_dir_Create(tdc, aname, &newFid.Fid);
ReleaseWriteLock(&afs_xdcache);
if (code && !AFS_IS_DISCON_RW) {
ZapDCE(tdc); /* surprise error -- use invalid value */
DZap(tdc);
}
}
if (tdc) {
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
newFid.Cell = adp->f.fid.Cell;
newFid.Fid.Volume = adp->f.fid.Fid.Volume;
ReleaseWriteLock(&adp->lock);
/* now we're done with parent dir, create the link's entry. Note that
* no one can get a pointer to the new cache entry until we release
* the xvcache lock. */
tvc = afs_NewVCache(&newFid, hostp);
if (!tvc)
{
code = -2;
ReleaseWriteLock(&afs_xvcache);
goto done;
}
ObtainWriteLock(&tvc->lock, 157);
ObtainWriteLock(&afs_xcbhash, 500);
tvc->f.states |= CStatd; /* have valid info */
tvc->f.states &= ~CBulkFetching;
if (adp->f.states & CForeign) {
tvc->f.states |= CForeign;
/* We don't have to worry about losing the callback since we're doing it
* under the afs_xvcache lock actually, afs_NewVCache may drop the
* afs_xvcache lock, if it calls afs_FlushVCache */
tvc->cbExpires = CallBack.ExpirationTime + now;
afs_QueueCallback(tvc, CBHash(CallBack.ExpirationTime), volp);
} else {
tvc->cbExpires = 0x7fffffff; /* never expires, they can't change */
/* since it never expires, we don't have to queue the callback */
}
ReleaseWriteLock(&afs_xcbhash);
if (AFS_IS_DISCON_RW) {
attrs->va_mode = InStatus.UnixModeBits;
afs_GenDisconStatus(adp, tvc, &newFid, attrs, treq, VLNK);
code = afs_DisconCreateSymlink(tvc, atargetName, treq);
if (code) {
/* XXX - When this goes wrong, we need to tidy up the changes we made to
* the parent, and get rid of the vcache we just created */
ReleaseWriteLock(&tvc->lock);
ReleaseWriteLock(&afs_xvcache);
afs_PutVCache(tvc);
goto done;
}
afs_DisconAddDirty(tvc, VDisconCreate, 0);
} else {
afs_ProcessFS(tvc, OutFidStatus, treq);
}
if (!tvc->linkData) {
tvc->linkData = afs_osi_Alloc(alen);
osi_Assert(tvc->linkData != NULL);
strncpy(tvc->linkData, atargetName, alen - 1);
tvc->linkData[alen - 1] = 0;
}
ReleaseWriteLock(&tvc->lock);
ReleaseWriteLock(&afs_xvcache);
if (tvcp)
*tvcp = tvc;
else
afs_PutVCache(tvc);
code = 0;
done:
afs_PutFakeStat(&fakestate);
if (volp)
afs_PutVolume(volp, READ_LOCK);
AFS_DISCON_UNLOCK();
code = afs_CheckCode(code, treq, 31);
afs_DestroyReq(treq);
done2:
osi_FreeSmallSpace(OutFidStatus);
osi_FreeSmallSpace(OutDirStatus);
return code;
}
int
afs_InvalidateAllSegments(struct vcache *avc)
{
struct dcache *tdc;
afs_int32 hash;
afs_int32 index;
struct dcache **dcList;
int i, dcListMax, dcListCount;
AFS_STATCNT(afs_InvalidateAllSegments);
afs_Trace2(afs_iclSetp, CM_TRACE_INVALL, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length));
hash = DVHash(&avc->f.fid);
avc->f.truncPos = AFS_NOTRUNC; /* don't truncate later */
avc->f.states &= ~CExtendedFile; /* not any more */
ObtainWriteLock(&afs_xcbhash, 459);
afs_DequeueCallback(avc);
avc->f.states &= ~(CStatd | CDirty); /* mark status information as bad, too */
ReleaseWriteLock(&afs_xcbhash);
if (avc->f.fid.Fid.Vnode & 1 || (vType(avc) == VDIR))
osi_dnlc_purgedp(avc);
/* Blow away pages; for now, only for Solaris */
#if (defined(AFS_SUN5_ENV))
if (WriteLocked(&avc->lock))
osi_ReleaseVM(avc, (afs_ucred_t *)0);
#endif
/*
* Block out others from screwing with this table; is a read lock
* sufficient?
*/
ObtainWriteLock(&afs_xdcache, 286);
dcListMax = 0;
for (index = afs_dvhashTbl[hash]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
/* In the case of fatal errors during stores, we MUST
* invalidate all of the relevant chunks. Otherwise, the chunks
* will be left with the 'new' data that was never successfully
* written to the server, but the DV in the dcache is still the
* old DV. So, we may indefinitely serve data to applications
* that is not actually in the file on the fileserver. If we
* cannot afs_GetValidDSlot the appropriate entries, currently
* there is no way to ensure the dcache is invalidated. So for
* now, to avoid risking serving bad data from the cache, panic
* instead. */
osi_Panic("afs_InvalidateAllSegments tdc count");
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid))
dcListMax++;
afs_PutDCache(tdc);
}
index = afs_dvnextTbl[index];
}
dcList = osi_Alloc(dcListMax * sizeof(struct dcache *));
dcListCount = 0;
for (index = afs_dvhashTbl[hash]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetValidDSlot(index);
if (!tdc) {
/* We cannot proceed after getting this error; we risk serving
* incorrect data to applications. So panic instead. See the
* above comment next to the previous afs_GetValidDSlot call
* for details. */
osi_Panic("afs_InvalidateAllSegments tdc store");
}
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid)) {
/* same file? we'll zap it */
if (afs_indexFlags[index] & IFDataMod) {
afs_stats_cmperf.cacheCurrDirtyChunks--;
/* don't write it back */
afs_indexFlags[index] &= ~IFDataMod;
}
afs_indexFlags[index] &= ~IFAnyPages;
if (dcListCount < dcListMax)
dcList[dcListCount++] = tdc;
else
afs_PutDCache(tdc);
} else {
afs_PutDCache(tdc);
}
}
index = afs_dvnextTbl[index];
}
ReleaseWriteLock(&afs_xdcache);
for (i = 0; i < dcListCount; i++) {
tdc = dcList[i];
ObtainWriteLock(&tdc->lock, 679);
ZapDCE(tdc);
if (vType(avc) == VDIR)
DZap(tdc);
ReleaseWriteLock(&tdc->lock);
afs_PutDCache(tdc);
}
osi_Free(dcList, dcListMax * sizeof(struct dcache *));
return 0;
}
static int
VLDB_Same(struct VenusFid *afid, struct vrequest *areq)
{
struct vrequest treq;
struct afs_conn *tconn;
int i, type = 0;
union {
struct vldbentry tve;
struct nvldbentry ntve;
struct uvldbentry utve;
} *v;
struct volume *tvp;
struct cell *tcell;
char *bp, tbuf[CVBS]; /* biggest volume id is 2^32, ~ 4*10^9 */
unsigned int changed;
struct server *(oldhosts[NMAXNSERVERS]);
AFS_STATCNT(CheckVLDB);
afs_FinalizeReq(areq);
if ((i = afs_InitReq(&treq, afs_osi_credp)))
return DUNNO;
v = afs_osi_Alloc(sizeof(*v));
tcell = afs_GetCell(afid->Cell, READ_LOCK);
bp = afs_cv2string(&tbuf[CVBS], afid->Fid.Volume);
do {
VSleep(2); /* Better safe than sorry. */
tconn =
afs_ConnByMHosts(tcell->cellHosts, tcell->vlport, tcell->cellNum,
&treq, SHARED_LOCK);
if (tconn) {
if (tconn->srvr->server->flags & SNO_LHOSTS) {
type = 0;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
RX_AFS_GLOCK();
} else if (tconn->srvr->server->flags & SYES_LHOSTS) {
type = 1;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
RX_AFS_GLOCK();
} else {
type = 2;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameU(tconn->id, bp, &v->utve);
RX_AFS_GLOCK();
if (!(tconn->srvr->server->flags & SVLSRV_UUID)) {
if (i == RXGEN_OPCODE) {
type = 1;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
RX_AFS_GLOCK();
if (i == RXGEN_OPCODE) {
type = 0;
tconn->srvr->server->flags |= SNO_LHOSTS;
RX_AFS_GUNLOCK();
i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
RX_AFS_GLOCK();
} else if (!i)
tconn->srvr->server->flags |= SYES_LHOSTS;
} else if (!i)
tconn->srvr->server->flags |= SVLSRV_UUID;
}
lastcode = i;
}
} else
i = -1;
} while (afs_Analyze(tconn, i, NULL, &treq, -1, /* no op code for this */
SHARED_LOCK, tcell));
afs_PutCell(tcell, READ_LOCK);
afs_Trace2(afs_iclSetp, CM_TRACE_CHECKVLDB, ICL_TYPE_FID, &afid,
ICL_TYPE_INT32, i);
if (i) {
afs_osi_Free(v, sizeof(*v));
return DUNNO;
}
/* have info, copy into serverHost array */
changed = 0;
tvp = afs_FindVolume(afid, WRITE_LOCK);
if (tvp) {
ObtainWriteLock(&tvp->lock, 107);
for (i = 0; i < NMAXNSERVERS && tvp->serverHost[i]; i++) {
oldhosts[i] = tvp->serverHost[i];
}
if (type == 2) {
InstallUVolumeEntry(tvp, &v->utve, afid->Cell, tcell, &treq);
} else if (type == 1) {
InstallNVolumeEntry(tvp, &v->ntve, afid->Cell);
} else {
InstallVolumeEntry(tvp, &v->tve, afid->Cell);
}
if (i < NMAXNSERVERS && tvp->serverHost[i]) {
changed = 1;
}
for (--i; !changed && i >= 0; i--) {
if (tvp->serverHost[i] != oldhosts[i]) {
changed = 1; /* also happens if prefs change. big deal. */
}
}
ReleaseWriteLock(&tvp->lock);
afs_PutVolume(tvp, WRITE_LOCK);
} else { /* can't find volume */
tvp = afs_GetVolume(afid, &treq, WRITE_LOCK);
if (tvp) {
afs_PutVolume(tvp, WRITE_LOCK);
afs_osi_Free(v, sizeof(*v));
return DIFFERENT;
} else {
afs_osi_Free(v, sizeof(*v));
return DUNNO;
}
}
afs_osi_Free(v, sizeof(*v));
return (changed ? DIFFERENT : SAME);
} /*VLDB_Same */
/**
* @param aname Volume name.
* @param acell Cell id.
* @param agood
* @param areq Request type.
* @param locktype Type of lock to be used.
* @return Volume or NULL if failure.
*/
static struct volume *
afs_NewVolumeByName(char *aname, afs_int32 acell, int agood,
struct vrequest *areq, afs_int32 locktype)
{
afs_int32 code, type = 0;
struct volume *tv, *tv1;
struct vldbentry *tve;
struct nvldbentry *ntve;
struct uvldbentry *utve;
struct cell *tcell;
char *tbuffer, *ve;
struct afs_conn *tconn;
struct vrequest treq;
struct rx_connection *rxconn;
if (strlen(aname) > VL_MAXNAMELEN) /* Invalid volume name */
return NULL;
tcell = afs_GetCell(acell, READ_LOCK);
if (!tcell) {
return NULL;
}
/* allow null request if we don't care about ENODEV/ETIMEDOUT distinction */
if (!areq)
areq = &treq;
afs_Trace2(afs_iclSetp, CM_TRACE_GETVOL, ICL_TYPE_STRING, aname,
ICL_TYPE_POINTER, aname);
tbuffer = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
tve = (struct vldbentry *)(tbuffer + 1024);
ntve = (struct nvldbentry *)tve;
utve = (struct uvldbentry *)tve;
afs_InitReq(&treq, afs_osi_credp); /* *must* be unauth for vldb */
do {
tconn =
afs_ConnByMHosts(tcell->cellHosts, tcell->vlport, tcell->cellNum,
&treq, SHARED_LOCK, 0, &rxconn);
if (tconn) {
if (tconn->srvr->server->flags & SNO_LHOSTS) {
type = 0;
RX_AFS_GUNLOCK();
code = VL_GetEntryByNameO(rxconn, aname, tve);
RX_AFS_GLOCK();
} else if (tconn->srvr->server->flags & SYES_LHOSTS) {
type = 1;
RX_AFS_GUNLOCK();
code = VL_GetEntryByNameN(rxconn, aname, ntve);
RX_AFS_GLOCK();
} else {
type = 2;
RX_AFS_GUNLOCK();
code = VL_GetEntryByNameU(rxconn, aname, utve);
RX_AFS_GLOCK();
if (!(tconn->srvr->server->flags & SVLSRV_UUID)) {
if (code == RXGEN_OPCODE) {
type = 1;
RX_AFS_GUNLOCK();
code = VL_GetEntryByNameN(rxconn, aname, ntve);
RX_AFS_GLOCK();
if (code == RXGEN_OPCODE) {
type = 0;
tconn->srvr->server->flags |= SNO_LHOSTS;
RX_AFS_GUNLOCK();
code = VL_GetEntryByNameO(rxconn, aname, tve);
RX_AFS_GLOCK();
} else if (!code)
tconn->srvr->server->flags |= SYES_LHOSTS;
} else if (!code)
tconn->srvr->server->flags |= SVLSRV_UUID;
}
lastnvcode = code;
}
} else
code = -1;
} while (afs_Analyze(tconn, rxconn, code, NULL, &treq, -1, /* no op code for this */
SHARED_LOCK, tcell));
if (code) {
/* If the client has yet to contact this cell and contact failed due
* to network errors, mark the VLDB servers as back up.
* That the client tried and failed can be determined from the
* fact that there was a downtime incident, but CHasVolRef is not set.
*/
/* RT 48959 - unclear if this should really go */
#if 0
if (areq->networkError && !(tcell->states & CHasVolRef)) {
int i;
struct server *sp;
struct srvAddr *sap;
for (i = 0; i < AFS_MAXCELLHOSTS; i++) {
if ((sp = tcell->cellHosts[i]) == NULL)
break;
for (sap = sp->addr; sap; sap = sap->next_sa)
afs_MarkServerUpOrDown(sap, 0);
}
}
#endif
afs_CopyError(&treq, areq);
osi_FreeLargeSpace(tbuffer);
afs_PutCell(tcell, READ_LOCK);
return NULL;
}
/*
* Check to see if this cell has not yet referenced a volume. If
* it hasn't, it's just about to change its status, and we need to mark
* this fact down. Note that it is remotely possible that afs_SetupVolume
* could fail and we would still not have a volume reference.
*/
if (!(tcell->states & CHasVolRef)) {
tcell->states |= CHasVolRef;
afs_stats_cmperf.numCellsContacted++;
}
/*First time a volume in this cell has been referenced */
if (type == 2)
ve = (char *)utve;
else if (type == 1)
ve = (char *)ntve;
else
ve = (char *)tve;
tv = afs_SetupVolume(0, aname, ve, tcell, agood, type, &treq);
if ((agood == 3) && tv && tv->backVol) {
/*
* This means that very soon we'll ask for the BK volume so
* we'll prefetch it (well we did already.)
*/
tv1 =
afs_SetupVolume(tv->backVol, (char *)0, ve, tcell, 0, type, &treq);
if (tv1) {
tv1->refCount--;
}
}
if ((agood >= 2) && tv && tv->roVol) {
/*
* This means that very soon we'll ask for the RO volume so
* we'll prefetch it (well we did already.)
*/
tv1 = afs_SetupVolume(tv->roVol, NULL, ve, tcell, 0, type, &treq);
if (tv1) {
tv1->refCount--;
}
}
osi_FreeLargeSpace(tbuffer);
afs_PutCell(tcell, READ_LOCK);
return tv;
} /*afs_NewVolumeByName */
afs_root(struct mount *mp, struct vnode **vpp)
#endif
{
int error;
struct vrequest treq;
struct vcache *tvp = 0;
struct vcache *gvp;
#if !defined(AFS_FBSD53_ENV) || defined(AFS_FBSD80_ENV)
struct thread *td = curthread;
#endif
struct ucred *cr = osi_curcred();
AFS_GLOCK();
AFS_STATCNT(afs_root);
crhold(cr);
tryagain:
if (afs_globalVp && (afs_globalVp->f.states & CStatd)) {
tvp = afs_globalVp;
error = 0;
} else {
if (!(error = afs_InitReq(&treq, cr)) && !(error = afs_CheckInit())) {
tvp = afs_GetVCache(&afs_rootFid, &treq, NULL, NULL);
/* we really want this to stay around */
if (tvp) {
gvp = afs_globalVp;
afs_globalVp = tvp;
if (gvp) {
afs_PutVCache(gvp);
if (tvp != afs_globalVp) {
/* someone raced us and won */
afs_PutVCache(tvp);
goto tryagain;
}
}
} else
error = ENOENT;
}
}
if (tvp) {
struct vnode *vp = AFSTOV(tvp);
ASSERT_VI_UNLOCKED(vp, "afs_root");
AFS_GUNLOCK();
error = vget(vp, LK_EXCLUSIVE | LK_RETRY, td);
AFS_GLOCK();
/* we dropped the glock, so re-check everything it had serialized */
if (!afs_globalVp || !(afs_globalVp->f.states & CStatd) ||
tvp != afs_globalVp) {
vput(vp);
afs_PutVCache(tvp);
goto tryagain;
}
if (error != 0)
goto tryagain;
/*
* I'm uncomfortable about this. Shouldn't this happen at a
* higher level, and shouldn't we busy the top-level directory
* to prevent recycling?
*/
vp->v_vflag |= VV_ROOT;
afs_globalVFS = mp;
*vpp = vp;
}
afs_Trace2(afs_iclSetp, CM_TRACE_VFSROOT, ICL_TYPE_POINTER, tvp ? AFSTOV(tvp) : NULL,
ICL_TYPE_INT32, error);
AFS_GUNLOCK();
crfree(cr);
return error;
}