/*!
* Print list of disconnected files.
*
* \note Call with afs_DDirtyVCListLock read locked.
*/
void
afs_DbgDisconFiles(void)
{
struct vcache *tvc;
struct afs_q *q;
int i = 0;
afs_warn("List of dirty files: \n");
ObtainReadLock(&afs_disconDirtyLock);
for (q = QPrev(&afs_disconDirty); q != &afs_disconDirty; q = QPrev(q)) {
tvc = QEntry(q, struct vcache, dirtyq);
afs_warn("Cell=%u Volume=%u VNode=%u Unique=%u\n",
tvc->f.fid.Cell,
tvc->f.fid.Fid.Volume,
tvc->f.fid.Fid.Vnode,
tvc->f.fid.Fid.Unique);
i++;
if (i >= 30)
osi_Panic("afs_DbgDisconFiles: loop in dirty list\n");
}
ReleaseReadLock(&afs_disconDirtyLock);
}
/* Try to invalidate pages, for "fs flush" or "fs flushv"; or
* try to free pages, when deleting a file.
*
* Locking: the vcache entry's lock is held. It may be dropped and
* re-obtained.
*
* Since we drop and re-obtain the lock, we can't guarantee that there won't
* be some pages around when we return, newly created by concurrent activity.
*/
void
osi_VM_TryToSmush(struct vcache *avc, afs_ucred_t *acred, int sync)
{
struct vnode *vp;
int tries, code;
int islocked;
vp = AFSTOV(avc);
VI_LOCK(vp);
if (vp->v_iflag & VI_DOOMED) {
VI_UNLOCK(vp);
return;
}
VI_UNLOCK(vp);
islocked = islocked_vnode(vp);
if (islocked == LK_EXCLOTHER)
panic("Trying to Smush over someone else's lock");
else if (islocked == LK_SHARED) {
afs_warn("Trying to Smush with a shared lock");
lock_vnode(vp, LK_UPGRADE);
} else if (!islocked)
lock_vnode(vp, LK_EXCLUSIVE);
if (vp->v_bufobj.bo_object != NULL) {
AFS_VM_OBJECT_WLOCK(vp->v_bufobj.bo_object);
/*
* Do we really want OBJPC_SYNC? OBJPC_INVAL would be
* faster, if invalidation is really what we are being
* asked to do. (It would make more sense, too, since
* otherwise this function is practically identical to
* osi_VM_StoreAllSegments().) -GAW
*/
/*
* Dunno. We no longer resemble osi_VM_StoreAllSegments,
* though maybe that's wrong, now. And OBJPC_SYNC is the
* common thing in 70 file systems, it seems. Matt.
*/
vm_object_page_clean(vp->v_bufobj.bo_object, 0, 0, OBJPC_SYNC);
AFS_VM_OBJECT_WUNLOCK(vp->v_bufobj.bo_object);
}
tries = 5;
code = osi_vinvalbuf(vp, V_SAVE, PCATCH, 0);
while (code && (tries > 0)) {
afs_warn("TryToSmush retrying vinvalbuf");
code = osi_vinvalbuf(vp, V_SAVE, PCATCH, 0);
--tries;
}
if (islocked == LK_SHARED)
lock_vnode(vp, LK_DOWNGRADE);
else if (!islocked)
unlock_vnode(vp);
}
void
cleanup_module(void)
#endif
{
#if defined(AFS_CACHE_BYPASS)
afs_warn("Cache bypass patched libafs module cleaning up.\n");
#endif
afs_shutdown_pagecopy();
#ifdef LINUX_KEYRING_SUPPORT
osi_keyring_shutdown();
#endif
osi_sysctl_clean();
#ifndef LINUX_KEYRING_SUPPORT
osi_syscall_clean();
#endif
unregister_filesystem(&afs_fs_type);
afs_destroy_inodecache();
osi_linux_free_afs_memory();
#ifdef AFS_LINUX24_ENV
osi_ioctl_clean();
osi_proc_clean();
#endif
return;
}
int
afs_MemCacheTruncate(struct osi_file *fP, int size)
{
struct memCacheEntry *mceP = (struct memCacheEntry *)fP;
AFS_STATCNT(afs_MemCacheTruncate);
ObtainWriteLock(&mceP->afs_memLock, 313);
/* old directory entry; g.c. */
if (size == 0 && mceP->dataSize > memCacheBlkSize) {
char *oldData = mceP->data;
mceP->data = afs_osi_Alloc(memCacheBlkSize);
if (mceP->data == NULL) { /* no available memory */
mceP->data = oldData;
ReleaseWriteLock(&mceP->afs_memLock);
afs_warn("afs: afs_MemWriteBlk mem alloc failure (%d bytes)\n",
memCacheBlkSize);
} else {
afs_osi_Free(oldData, mceP->dataSize);
mceP->dataSize = memCacheBlkSize;
}
}
if (size < mceP->size)
mceP->size = size;
ReleaseWriteLock(&mceP->afs_memLock);
return 0;
}
void
shutdown_osinet(void)
{
AFS_STATCNT(shutdown_osinet);
#ifndef AFS_PRIVATE_OSI_ALLOCSPACES
if (afs_cold_shutdown) {
struct osi_packet *tp;
while ((tp = freePacketList)) {
freePacketList = tp->next;
afs_osi_Free(tp, AFS_LRALLOCSIZ);
#ifdef KERNEL_HAVE_PIN
unpin(tp, AFS_LRALLOCSIZ);
#endif
}
while ((tp = freeSmallList)) {
freeSmallList = tp->next;
afs_osi_Free(tp, AFS_SMALLOCSIZ);
#ifdef KERNEL_HAVE_PIN
unpin(tp, AFS_SMALLOCSIZ);
#endif
}
LOCK_INIT(&osi_fsplock, "osi_fsplock");
LOCK_INIT(&osi_flplock, "osi_flplock");
}
#endif /* AFS_PRIVATE_OSI_ALLOCSPACES */
if (afs_stats_cmperf.LargeBlocksActive ||
afs_stats_cmperf.SmallBlocksActive)
{
afs_warn("WARNING: not all blocks freed: large %d small %d\n",
afs_stats_cmperf.LargeBlocksActive,
afs_stats_cmperf.SmallBlocksActive);
}
}
struct file *
afs_linux_raw_open(afs_dcache_id_t *ainode)
{
struct inode *tip = NULL;
struct dentry *dp = NULL;
struct file* filp;
dp = afs_get_dentry_from_fh(afs_cacheSBp, ainode, cache_fh_len, cache_fh_type,
afs_fh_acceptable);
if ((!dp) || IS_ERR(dp))
osi_Panic("Can't get dentry\n");
tip = dp->d_inode;
tip->i_flags |= S_NOATIME; /* Disable updating access times. */
#if defined(STRUCT_TASK_STRUCT_HAS_CRED)
/* Use stashed credentials - prevent selinux/apparmor problems */
filp = afs_dentry_open(dp, afs_cacheMnt, O_RDWR, cache_creds);
if (IS_ERR(filp))
filp = afs_dentry_open(dp, afs_cacheMnt, O_RDWR, current_cred());
#else
filp = dentry_open(dget(dp), mntget(afs_cacheMnt), O_RDWR);
#endif
if (IS_ERR(filp)) {
afs_warn("afs: Cannot open cache file (code %d). Trying to continue, "
"but AFS accesses may return errors or panic the system\n",
(int) PTR_ERR(filp));
filp = NULL;
}
dput(dp);
return filp;
}
/* DARWIN uses locking, and so must provide its own */
void
shutdown_osisleep(void)
{
afs_event_t *tmp;
int i;
for (i=0;i<AFS_EVHASHSIZE;i++) {
while ((tmp = afs_evhasht[i]) != NULL) {
afs_evhasht[i] = tmp->next;
if (tmp->refcount > 0) {
afs_warn("nonzero refcount in shutdown_osisleep()\n");
} else {
#if defined(AFS_AIX_ENV)
xmfree(tmp);
#elif defined(AFS_FBSD_ENV)
afs_osi_Free(tmp, sizeof(*tmp));
#elif defined(AFS_SGI_ENV) || defined(AFS_XBSD_ENV) || defined(AFS_SUN5_ENV)
osi_FreeSmallSpace(tmp);
#elif defined(AFS_LINUX26_ENV)
kfree(tmp);
#elif defined(AFS_LINUX20_ENV)
osi_linux_free(tmp);
#endif
}
}
}
}
int
init_module(void)
#endif
{
int err;
AFS_RWLOCK_INIT(&afs_xosi, "afs_xosi");
#if !defined(AFS_LINUX24_ENV)
/* obtain PAGE_OFFSET value */
afs_linux_page_offset = get_page_offset();
#ifndef AFS_S390_LINUX22_ENV
if (afs_linux_page_offset == 0) {
/* couldn't obtain page offset so can't continue */
printf("afs: Unable to obtain PAGE_OFFSET. Exiting..");
return -EIO;
}
#endif /* AFS_S390_LINUX22_ENV */
#endif /* !defined(AFS_LINUX24_ENV) */
osi_Init();
#ifndef LINUX_KEYRING_SUPPORT
err = osi_syscall_init();
if (err)
return err;
#endif
err = afs_init_inodecache();
if (err) {
#ifndef LINUX_KEYRING_SUPPORT
osi_syscall_clean();
#endif
return err;
}
err = register_filesystem(&afs_fs_type);
if (err) {
afs_destroy_inodecache();
#ifndef LINUX_KEYRING_SUPPORT
osi_syscall_clean();
#endif
return err;
}
osi_sysctl_init();
#ifdef LINUX_KEYRING_SUPPORT
osi_keyring_init();
#endif
#ifdef AFS_LINUX24_ENV
osi_proc_init();
osi_ioctl_init();
#endif
#if defined(AFS_CACHE_BYPASS)
afs_warn("Cache bypass patched libafs module init.\n");
#endif
afs_init_pagecopy();
return 0;
}
/**
* Find a connection with call slots available, allocating one
* if nothing is available and we find an allocated slot
* @param xcv A connection vector
* @param create If set, a new connection may be created
*/
static struct afs_conn *
find_preferred_connection(struct sa_conn_vector *xcv, int create)
{
afs_int32 cix, bix;
struct afs_conn *tc = NULL;
bix = -1;
for(cix = 0; cix < CVEC_LEN; ++cix) {
tc = &(xcv->cvec[cix]);
if (!tc->id) {
if (create) {
tc->parent = xcv;
tc->forceConnectFS = 1;
tc->activated = 1;
bix = cix;
break;
} /* create */
} else {
if (tc->refCount < (RX_MAXCALLS-1)) {
bix = cix;
goto f_conn;
} else if (cix == (CVEC_LEN-1))
conn_vec_select_conn(xcv, bix, tc);
} /* tc->id */
} /* for cix < CVEC_LEN */
if (bix < 0) {
afs_warn("find_preferred_connection: no connection and !create\n");
tc = NULL;
goto out;
}
f_conn:
tc->refCount++;
xcv->refCount++;
#if REPORT_CONNECTIONS_ISSUED
afs_warn("Issuing conn %d refCount=%d parent refCount=%d\n", bix,
tc->refCount, xcv->refCount);
#endif
out:
return (tc);
} /* find_preferred_connection */
void
osi_StopListener(void)
{
struct proc *p;
/*
* Have to drop global lock to safely do this.
* soclose() is currently protected by Giant,
* but pfind and psignal are MPSAFE.
*/
int haveGlock = ISAFS_GLOCK();
if (haveGlock)
AFS_GUNLOCK();
soshutdown(rx_socket, 2);
#ifndef AFS_FBSD70_ENV
soclose(rx_socket);
#endif
p = pfind(rxk_ListenerPid);
afs_warn("osi_StopListener: rxk_ListenerPid %lx\n", p);
if (p)
psignal(p, SIGUSR1);
#ifdef AFS_FBSD50_ENV
PROC_UNLOCK(p);
#endif
#ifdef AFS_FBSD70_ENV
{
/* Avoid destroying socket until osi_NetReceive has
* had a chance to clean up */
int tries;
struct mtx s_mtx;
MUTEX_INIT(&s_mtx, "rx_shutdown_mutex", MUTEX_DEFAULT, 0);
MUTEX_ENTER(&s_mtx);
tries = 3;
while ((tries > 0) && (!so_is_disconn(rx_socket))) {
msleep(&osi_StopListener, &s_mtx, PSOCK | PCATCH,
"rx_shutdown_timedwait", 1 * hz);
--tries;
}
if (so_is_disconn(rx_socket))
soclose(rx_socket);
MUTEX_EXIT(&s_mtx);
MUTEX_DESTROY(&s_mtx);
}
#endif
if (haveGlock)
AFS_GLOCK();
}
/*XXX: this extends a block arbitrarily to support big directories */
int
afs_MemWritevBlk(struct memCacheEntry *mceP, int offset,
struct iovec *iov, int nio, int size)
{
int i;
int bytesWritten;
int bytesToWrite;
AFS_STATCNT(afs_MemWriteBlk);
ObtainWriteLock(&mceP->afs_memLock, 561);
if (offset + size > mceP->dataSize) {
char *oldData = mceP->data;
mceP->data = afs_osi_Alloc(size + offset);
if (mceP->data == NULL) { /* no available memory */
mceP->data = oldData; /* revert back change that was made */
ReleaseWriteLock(&mceP->afs_memLock);
afs_warn("afs: afs_MemWriteBlk mem alloc failure (%d bytes)\n",
size + offset);
return -ENOMEM;
}
/* may overlap, but this is OK */
AFS_GUNLOCK();
memcpy(mceP->data, oldData, mceP->size);
AFS_GLOCK();
afs_osi_Free(oldData, mceP->dataSize);
mceP->dataSize = size + offset;
}
AFS_GUNLOCK();
if (mceP->size < offset)
memset(mceP->data + mceP->size, 0, offset - mceP->size);
for (bytesWritten = 0, i = 0; i < nio && size > 0; i++) {
bytesToWrite = (size < iov[i].iov_len) ? size : iov[i].iov_len;
memcpy(mceP->data + offset, iov[i].iov_base, bytesToWrite);
offset += bytesToWrite;
bytesWritten += bytesToWrite;
size -= bytesToWrite;
}
mceP->size = (offset < mceP->size) ? mceP->size : offset;
AFS_GLOCK();
ReleaseWriteLock(&mceP->afs_memLock);
return bytesWritten;
}
int
afs_InitMemCache(int blkCount, int blkSize, int flags)
{
int index;
AFS_STATCNT(afs_InitMemCache);
if (blkSize)
memCacheBlkSize = blkSize;
memMaxBlkNumber = blkCount;
memCache =
afs_osi_Alloc(memMaxBlkNumber * sizeof(struct memCacheEntry));
osi_Assert(memCache != NULL);
for (index = 0; index < memMaxBlkNumber; index++) {
char *blk;
(memCache + index)->size = 0;
(memCache + index)->dataSize = memCacheBlkSize;
LOCK_INIT(&((memCache + index)->afs_memLock), "afs_memLock");
blk = afs_osi_Alloc(memCacheBlkSize);
if (blk == NULL)
goto nomem;
(memCache + index)->data = blk;
memset((memCache + index)->data, 0, memCacheBlkSize);
}
#if defined(AFS_HAVE_VXFS)
afs_InitDualFSCacheOps((struct vnode *)0);
#endif
for (index = 0; index < blkCount; index++)
afs_InitCacheFile(NULL, 0);
return 0;
nomem:
afs_warn("afsd: memCache allocation failure at %d KB.\n",
(index * memCacheBlkSize) / 1024);
while (--index >= 0) {
afs_osi_Free((memCache + index)->data, memCacheBlkSize);
(memCache + index)->data = NULL;
}
return ENOMEM;
}
void
osi_AttachVnode(struct vcache *avc, int seq) {
struct vnode *vp;
struct thread *p = curthread;
ReleaseWriteLock(&afs_xvcache);
AFS_GUNLOCK();
#if defined(AFS_FBSD60_ENV)
if (getnewvnode(MOUNT_AFS, afs_globalVFS, &afs_vnodeops, &vp))
#else
if (getnewvnode(MOUNT_AFS, afs_globalVFS, afs_vnodeop_p, &vp))
#endif
panic("afs getnewvnode"); /* can't happen */
#ifdef AFS_FBSD70_ENV
/* XXX verified on 80--TODO check on 7x */
if (!vp->v_mount) {
ma_vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); /* !glocked */
insmntque(vp, afs_globalVFS);
MA_VOP_UNLOCK(vp, 0, p);
}
#endif
AFS_GLOCK();
ObtainWriteLock(&afs_xvcache,339);
if (avc->v != NULL) {
/* I'd like to know if this ever happens...
* We don't drop global for the rest of this function,
* so if we do lose the race, the other thread should
* have found the same vnode and finished initializing
* the vcache entry. Is it conceivable that this vcache
* entry could be recycled during this interval? If so,
* then there probably needs to be some sort of additional
* mutual exclusion (an Embryonic flag would suffice).
* -GAW */
afs_warn("afs_NewVCache: lost the race\n");
return;
}
avc->v = vp;
avc->v->v_data = avc;
lockinit(&avc->rwlock, PINOD, "vcache", 0, 0);
}
int
afs_MemWriteUIO(afs_dcache_id_t *ainode, struct uio *uioP)
{
struct memCacheEntry *mceP =
(struct memCacheEntry *)afs_MemCacheOpen(ainode);
afs_int32 code;
AFS_STATCNT(afs_MemWriteUIO);
ObtainWriteLock(&mceP->afs_memLock, 312);
if (AFS_UIO_RESID(uioP) + AFS_UIO_OFFSET(uioP) > mceP->dataSize) {
char *oldData = mceP->data;
mceP->data = afs_osi_Alloc(AFS_UIO_RESID(uioP) + AFS_UIO_OFFSET(uioP));
if (mceP->data == NULL) { /* no available memory */
mceP->data = oldData; /* revert back change that was made */
ReleaseWriteLock(&mceP->afs_memLock);
afs_warn("afs: afs_MemWriteBlk mem alloc failure (%ld bytes)\n",
(long)(AFS_UIO_RESID(uioP) + AFS_UIO_OFFSET(uioP)));
return -ENOMEM;
}
AFS_GUNLOCK();
memcpy(mceP->data, oldData, mceP->size);
AFS_GLOCK();
afs_osi_Free(oldData, mceP->dataSize);
mceP->dataSize = AFS_UIO_RESID(uioP) + AFS_UIO_OFFSET(uioP);
}
if (mceP->size < AFS_UIO_OFFSET(uioP))
memset(mceP->data + mceP->size, 0,
(int)(AFS_UIO_OFFSET(uioP) - mceP->size));
AFS_UIOMOVE(mceP->data + AFS_UIO_OFFSET(uioP), AFS_UIO_RESID(uioP), UIO_WRITE,
uioP, code);
if (AFS_UIO_OFFSET(uioP) > mceP->size)
mceP->size = AFS_UIO_OFFSET(uioP);
ReleaseWriteLock(&mceP->afs_memLock);
return code;
}
void
shutdown_osisleep(void) {
struct afs_event *evp, *nevp, **pevpp;
int i;
for (i=0; i < AFS_EVHASHSIZE; i++) {
evp = afs_evhasht[i];
pevpp = &afs_evhasht[i];
while (evp) {
EVTLOCK_LOCK(evp);
nevp = evp->next;
if (evp->refcount == 0) {
EVTLOCK_DESTROY(evp);
*pevpp = evp->next;
osi_FreeSmallSpace(evp);
afs_evhashcnt--;
} else {
afs_warn("nonzero refcount in shutdown_osisleep()\n");
EVTLOCK_UNLOCK(evp);
pevpp = &evp->next;
}
evp = nevp;
}
}
}
/* Cannot have static linkage--called from BPrefetch (afs_daemons) */
afs_int32
afs_PrefetchNoCache(struct vcache *avc,
afs_ucred_t *acred,
struct nocache_read_request *bparms)
{
struct uio *auio;
#ifndef UKERNEL
struct iovec *iovecp;
#endif
struct vrequest *areq;
afs_int32 code = 0;
struct rx_connection *rxconn;
#ifdef AFS_64BIT_CLIENT
afs_int32 length_hi, bytes, locked;
#endif
struct afs_conn *tc;
struct rx_call *tcall;
struct tlocal1 {
struct AFSVolSync tsync;
struct AFSFetchStatus OutStatus;
struct AFSCallBack CallBack;
};
struct tlocal1 *tcallspec;
auio = bparms->auio;
areq = bparms->areq;
#ifndef UKERNEL
iovecp = auio->uio_iov;
#endif
tcallspec = osi_Alloc(sizeof(struct tlocal1));
do {
tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK /* ignored */, &rxconn);
if (tc) {
avc->callback = tc->parent->srvr->server;
tcall = rx_NewCall(rxconn);
#ifdef AFS_64BIT_CLIENT
if (!afs_serverHasNo64Bit(tc)) {
code = StartRXAFS_FetchData64(tcall,
(struct AFSFid *) &avc->f.fid.Fid,
auio->uio_offset,
bparms->length);
if (code == 0) {
COND_GUNLOCK(locked);
bytes = rx_Read(tcall, (char *)&length_hi,
sizeof(afs_int32));
COND_RE_GLOCK(locked);
if (bytes != sizeof(afs_int32)) {
length_hi = 0;
code = rx_Error(tcall);
COND_GUNLOCK(locked);
code = rx_EndCall(tcall, code);
COND_RE_GLOCK(locked);
tcall = NULL;
}
}
} /* afs_serverHasNo64Bit */
if (code == RXGEN_OPCODE || afs_serverHasNo64Bit(tc)) {
if (auio->uio_offset > 0x7FFFFFFF) {
code = EFBIG;
} else {
afs_int32 pos;
pos = auio->uio_offset;
COND_GUNLOCK(locked);
if (!tcall)
tcall = rx_NewCall(rxconn);
code = StartRXAFS_FetchData(tcall,
(struct AFSFid *) &avc->f.fid.Fid,
pos, bparms->length);
COND_RE_GLOCK(locked);
}
afs_serverSetNo64Bit(tc);
}
#else
code = StartRXAFS_FetchData(tcall,
(struct AFSFid *) &avc->f.fid.Fid,
auio->uio_offset, bparms->length);
#endif
if (code == 0) {
code = afs_NoCacheFetchProc(tcall, avc, auio,
1 /* release_pages */,
bparms->length);
} else {
afs_warn("BYPASS: StartRXAFS_FetchData failed: %d\n", code);
unlock_and_release_pages(auio);
goto done;
}
if (code == 0) {
code = EndRXAFS_FetchData(tcall, &tcallspec->OutStatus,
&tcallspec->CallBack,
&tcallspec->tsync);
} else {
afs_warn("BYPASS: NoCacheFetchProc failed: %d\n", code);
}
code = rx_EndCall(tcall, code);
} else {
afs_warn("BYPASS: No connection.\n");
code = -1;
unlock_and_release_pages(auio);
goto done;
}
} while (afs_Analyze(tc, rxconn, code, &avc->f.fid, areq,
AFS_STATS_FS_RPCIDX_FETCHDATA,
SHARED_LOCK,0));
done:
/*
* Copy appropriate fields into vcache
*/
if (!code)
afs_ProcessFS(avc, &tcallspec->OutStatus, areq);
osi_Free(areq, sizeof(struct vrequest));
osi_Free(tcallspec, sizeof(struct tlocal1));
osi_Free(bparms, sizeof(struct nocache_read_request));
#ifndef UKERNEL
/* in UKERNEL, the "pages" are passed in */
osi_Free(iovecp, auio->uio_iovcnt * sizeof(struct iovec));
osi_Free(auio, sizeof(struct uio));
#endif
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;
}
/* no-cache prefetch routine */
static afs_int32
afs_NoCacheFetchProc(struct rx_call *acall,
struct vcache *avc,
struct uio *auio,
afs_int32 release_pages,
afs_int32 size)
{
afs_int32 length;
afs_int32 code;
int moredata, iovno, iovoff, iovmax, result, locked;
struct iovec *ciov;
struct iovec *rxiov;
int nio = 0;
bypass_page_t pp;
int curpage, bytes;
int pageoff;
rxiov = osi_AllocSmallSpace(sizeof(struct iovec) * RX_MAXIOVECS);
ciov = auio->uio_iov;
pp = (bypass_page_t) ciov->iov_base;
iovmax = auio->uio_iovcnt - 1;
iovno = iovoff = result = 0;
do {
COND_GUNLOCK(locked);
code = rx_Read(acall, (char *)&length, sizeof(afs_int32));
COND_RE_GLOCK(locked);
if (code != sizeof(afs_int32)) {
result = EIO;
afs_warn("Preread error. code: %d instead of %d\n",
code, (int)sizeof(afs_int32));
unlock_and_release_pages(auio);
goto done;
} else
length = ntohl(length);
if (length > size) {
result = EIO;
afs_warn("Preread error. Got length %d, which is greater than size %d\n",
length, size);
unlock_and_release_pages(auio);
goto done;
}
/* If we get a 0 length reply, time to cleanup and return */
if (length == 0) {
unlock_and_release_pages(auio);
result = 0;
goto done;
}
/*
* The fetch protocol is extended for the AFS/DFS translator
* to allow multiple blocks of data, each with its own length,
* to be returned. As long as the top bit is set, there are more
* blocks expected.
*
* We do not do this for AFS file servers because they sometimes
* return large negative numbers as the transfer size.
*/
if (avc->f.states & CForeign) {
moredata = length & 0x80000000;
length &= ~0x80000000;
} else {
moredata = 0;
}
for (curpage = 0; curpage <= iovmax; curpage++) {
pageoff = 0;
/* properly, this should track uio_resid, not a fixed page size! */
while (pageoff < auio->uio_iov[curpage].iov_len) {
/* If no more iovs, issue new read. */
if (iovno >= nio) {
COND_GUNLOCK(locked);
bytes = rx_Readv(acall, rxiov, &nio, RX_MAXIOVECS, length);
COND_RE_GLOCK(locked);
if (bytes < 0) {
afs_warn("afs_NoCacheFetchProc: rx_Read error. Return code was %d\n", bytes);
result = bytes;
unlock_and_release_pages(auio);
goto done;
} else if (bytes == 0) {
/* we failed to read the full length */
result = EIO;
afs_warn("afs_NoCacheFetchProc: rx_Read returned zero. Aborting.\n");
unlock_and_release_pages(auio);
goto done;
}
size -= bytes;
auio->uio_resid -= bytes;
iovno = 0;
}
pp = (bypass_page_t)auio->uio_iov[curpage].iov_base;
if (pageoff + (rxiov[iovno].iov_len - iovoff) <= auio->uio_iov[curpage].iov_len) {
/* Copy entire (or rest of) current iovec into current page */
if (pp)
afs_bypass_copy_page(pp, pageoff, rxiov, iovno, iovoff, auio, curpage, 0);
length -= (rxiov[iovno].iov_len - iovoff);
pageoff += rxiov[iovno].iov_len - iovoff;
iovno++;
iovoff = 0;
} else {
/* Copy only what's needed to fill current page */
if (pp)
afs_bypass_copy_page(pp, pageoff, rxiov, iovno, iovoff, auio, curpage, 1);
length -= (auio->uio_iov[curpage].iov_len - pageoff);
iovoff += auio->uio_iov[curpage].iov_len - pageoff;
pageoff = auio->uio_iov[curpage].iov_len;
}
/* we filled a page, or this is the last page. conditionally release it */
if (pp && ((pageoff == auio->uio_iov[curpage].iov_len &&
release_pages) || (length == 0 && iovno >= nio)))
release_full_page(pp, pageoff);
if (length == 0 && iovno >= nio)
goto done;
}
}
} while (moredata);
done:
osi_FreeSmallSpace(rxiov);
return result;
}
/*
* Inform dynroot that a new vnode is being created. Return value
* is non-zero if this vnode is handled by dynroot, in which case
* FetchStatus will be filled in.
*/
int
afs_DynrootNewVnode(struct vcache *avc, struct AFSFetchStatus *status)
{
char *bp, tbuf[CVBS];
if (_afs_IsDynrootFid(&avc->f.fid)) {
if (!afs_dynrootEnable)
return 0;
afs_GetDynroot(0, 0, status);
afs_PutDynroot();
return 1;
}
if (afs_IsDynrootMount(avc)) {
afs_GetDynrootMount(0, 0, status);
afs_PutDynroot();
return 1;
}
/*
* Check if this is an entry under /afs, e.g. /afs/cellname.
*/
if (avc->f.fid.Cell == afs_dynrootCell
&& avc->f.fid.Fid.Volume == AFS_DYNROOT_VOLUME) {
struct cell *c;
struct cell_alias *ca;
int namelen, linklen, cellidx, rw;
memset(status, 0, sizeof(struct AFSFetchStatus));
status->FileType = SymbolicLink;
status->LinkCount = 1;
status->DataVersion = 1;
status->CallerAccess = PRSFS_LOOKUP | PRSFS_READ;
status->AnonymousAccess = PRSFS_LOOKUP | PRSFS_READ;
status->ParentVnode = 1;
status->ParentUnique = 1;
if (VNUM_TO_VNTYPE(avc->f.fid.Fid.Vnode) == VN_TYPE_SYMLINK) {
struct afs_dynSymlink *ts;
int index = VNUM_TO_VNID(avc->f.fid.Fid.Vnode);
ObtainReadLock(&afs_dynSymlinkLock);
ts = afs_dynSymlinkBase;
while (ts) {
if (ts->index == index)
break;
ts = ts->next;
}
if (ts) {
linklen = strlen(ts->target);
avc->linkData = afs_osi_Alloc(linklen + 1);
strcpy(avc->linkData, ts->target);
status->Length = linklen;
status->UnixModeBits = 0755;
}
ReleaseReadLock(&afs_dynSymlinkLock);
return ts ? 1 : 0;
}
if (VNUM_TO_VNTYPE(avc->f.fid.Fid.Vnode) != VN_TYPE_CELL
&& VNUM_TO_VNTYPE(avc->f.fid.Fid.Vnode) != VN_TYPE_ALIAS
&& VNUM_TO_VNTYPE(avc->f.fid.Fid.Vnode) != VN_TYPE_MOUNT) {
afs_warn("dynroot vnode inconsistency, unknown VNTYPE %d\n",
VNUM_TO_VNTYPE(avc->f.fid.Fid.Vnode));
return 0;
}
cellidx = VNUM_TO_CIDX(avc->f.fid.Fid.Vnode);
rw = VNUM_TO_RW(avc->f.fid.Fid.Vnode);
if (VNUM_TO_VNTYPE(avc->f.fid.Fid.Vnode) == VN_TYPE_ALIAS) {
char *realName;
ca = afs_GetCellAlias(cellidx);
if (!ca) {
afs_warn("dynroot vnode inconsistency, can't find alias %d\n",
cellidx);
return 0;
}
/*
* linkData needs to contain the name of the cell
* we're aliasing for.
*/
realName = ca->cell;
if (!realName) {
afs_warn("dynroot: alias %s missing real cell name\n",
ca->alias);
avc->linkData = afs_strdup("unknown");
linklen = 7;
} else {
int namelen = strlen(realName);
linklen = rw + namelen;
avc->linkData = afs_osi_Alloc(linklen + 1);
strcpy(avc->linkData, rw ? "." : "");
afs_strcat(avc->linkData, realName);
}
status->UnixModeBits = 0755;
afs_PutCellAlias(ca);
} else if (VNUM_TO_VNTYPE(avc->f.fid.Fid.Vnode) == VN_TYPE_MOUNT) {
c = afs_GetCellByIndex(cellidx, READ_LOCK);
if (!c) {
afs_warn("dynroot vnode inconsistency, can't find cell %d\n",
cellidx);
return 0;
}
/*
* linkData needs to contain "%cell:volumeid"
*/
namelen = strlen(c->cellName);
bp = afs_cv2string(&tbuf[CVBS], avc->f.fid.Fid.Unique);
linklen = 2 + namelen + strlen(bp);
avc->linkData = afs_osi_Alloc(linklen + 1);
strcpy(avc->linkData, "%");
afs_strcat(avc->linkData, c->cellName);
afs_strcat(avc->linkData, ":");
afs_strcat(avc->linkData, bp);
status->UnixModeBits = 0644;
status->ParentVnode = AFS_DYNROOT_MOUNT_VNODE;
afs_PutCell(c, READ_LOCK);
} else {
c = afs_GetCellByIndex(cellidx, READ_LOCK);
if (!c) {
afs_warn("dynroot vnode inconsistency, can't find cell %d\n",
cellidx);
return 0;
}
/*
* linkData needs to contain "#cell:root.cell" or "%cell:root.cell"
*/
namelen = strlen(c->cellName);
linklen = 1 + namelen + 10;
avc->linkData = afs_osi_Alloc(linklen + 1);
strcpy(avc->linkData, rw ? "%" : "#");
afs_strcat(avc->linkData, c->cellName);
afs_strcat(avc->linkData, ":root.cell");
status->UnixModeBits = 0644;
afs_PutCell(c, READ_LOCK);
}
status->Length = linklen;
return 1;
}
return 0;
}
void
afs_CheckCallbacks(unsigned int secs)
{
struct vcache *tvc;
register struct afs_q *tq;
struct afs_q *uq;
afs_uint32 now;
struct volume *tvp;
register int safety;
ObtainWriteLock(&afs_xcbhash, 85); /* pretty likely I'm going to remove something */
now = osi_Time();
for (safety = 0, tq = cbHashT[base].head.prev;
(safety <= CBQ_LIMIT) && (tq != &(cbHashT[base].head));
tq = uq, safety++) {
uq = QPrev(tq);
tvc = CBQTOV(tq);
if (tvc->cbExpires < now + secs) { /* race #1 here */
/* Get the volume, and if its callback expiration time is more than secs
* seconds into the future, update this vcache entry and requeue it below
*/
if ((tvc->f.states & CRO)
&& (tvp = afs_FindVolume(&(tvc->f.fid), READ_LOCK))) {
if (tvp->expireTime > now + secs) {
tvc->cbExpires = tvp->expireTime; /* XXX race here */
} else {
int i;
for (i = 0; i < MAXHOSTS && tvp->serverHost[i]; i++) {
if (!(tvp->serverHost[i]->flags & SRVR_ISDOWN)) {
/* What about locking xvcache or vrefcount++ or
* write locking tvc? */
QRemove(tq);
tvc->f.states &= ~(CStatd | CMValid | CUnique);
if (!(tvc->f.states & (CVInit|CVFlushed)) &&
(tvc->f.fid.Fid.Vnode & 1 ||
(vType(tvc) == VDIR)))
osi_dnlc_purgedp(tvc);
tvc->dchint = NULL; /*invalidate em */
afs_ResetVolumeInfo(tvp);
break;
}
}
}
afs_PutVolume(tvp, READ_LOCK);
} else {
/* Do I need to worry about things like execsorwriters?
* What about locking xvcache or vrefcount++ or write locking tvc?
*/
QRemove(tq);
tvc->f.states &= ~(CStatd | CMValid | CUnique);
if (!(tvc->f.states & (CVInit|CVFlushed)) &&
(tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR)))
osi_dnlc_purgedp(tvc);
}
}
if ((tvc->cbExpires > basetime) && CBHash(tvc->cbExpires - basetime)) {
/* it's been renewed on us. Have to be careful not to put it back
* into this slot, or we may never get out of here.
*/
int slot;
slot = (CBHash(tvc->cbExpires - basetime) + base) % CBHTSIZE;
if (slot != base) {
if (QPrev(tq))
QRemove(&(tvc->callsort));
QAdd(&(cbHashT[slot].head), &(tvc->callsort));
/* XXX remember to update volume expiration time */
/* -- not needed for correctness, though */
}
}
}
if (safety > CBQ_LIMIT) {
afs_stats_cmperf.cbloops++;
if (afs_paniconwarn)
osi_Panic("CheckCallbacks");
afs_warn
("AFS Internal Error (minor): please contact AFS Product Support.\n");
ReleaseWriteLock(&afs_xcbhash);
afs_FlushCBs();
return;
} else
ReleaseWriteLock(&afs_xcbhash);
/* XXX future optimization:
if this item has been recently accessed, queue up a stat for it.
{
struct dcache * adc;
ObtainReadLock(&afs_xdcache);
if ((adc = tvc->quick.dc) && (adc->stamp == tvc->quick.stamp)
&& (afs_indexTimes[adc->index] > afs_indexCounter - 20)) {
queue up the stat request
}
ReleaseReadLock(&afs_xdcache);
}
*/
return;
} /* afs_CheckCallback */
struct vcache *
osi_dnlc_lookup(struct vcache *adp, char *aname, int locktype)
{
struct vcache *tvc;
unsigned int key, skey;
char *ts = aname;
struct nc *tnc;
int safety;
#ifdef AFS_DARWIN80_ENV
vnode_t tvp;
#endif
if (!afs_usednlc)
return 0;
dnlcHash(ts, key); /* leaves ts pointing at the NULL */
if (ts - aname >= AFSNCNAMESIZE)
return 0;
skey = key & (NHSIZE - 1);
TRACE(osi_dnlc_lookupT, skey);
dnlcstats.lookups++;
ObtainReadLock(&afs_xvcache);
ObtainReadLock(&afs_xdnlc);
for (tvc = NULL, tnc = nameHash[skey], safety = 0; tnc;
tnc = tnc->next, safety++) {
if ( /* (tnc->key == key) && */ (tnc->dirp == adp)
&& (!strcmp((char *)tnc->name, aname))) {
tvc = tnc->vp;
break;
} else if (tnc->next == nameHash[skey]) { /* end of list */
break;
} else if (safety > NCSIZE) {
afs_warn("DNLC cycle");
dnlcstats.cycles++;
ReleaseReadLock(&afs_xdnlc);
ReleaseReadLock(&afs_xvcache);
osi_dnlc_purge();
return (0);
}
}
ReleaseReadLock(&afs_xdnlc);
if (!tvc) {
ReleaseReadLock(&afs_xvcache);
dnlcstats.misses++;
} else {
if ((tvc->f.states & CVInit)
#ifdef AFS_DARWIN80_ENV
||(tvc->f.states & CDeadVnode)
#endif
)
{
ReleaseReadLock(&afs_xvcache);
dnlcstats.misses++;
osi_dnlc_remove(adp, aname, tvc);
return 0;
}
#if defined(AFS_DARWIN80_ENV)
tvp = AFSTOV(tvc);
if (vnode_get(tvp)) {
ReleaseReadLock(&afs_xvcache);
dnlcstats.misses++;
osi_dnlc_remove(adp, aname, tvc);
return 0;
}
if (vnode_ref(tvp)) {
ReleaseReadLock(&afs_xvcache);
AFS_GUNLOCK();
vnode_put(tvp);
AFS_GLOCK();
dnlcstats.misses++;
osi_dnlc_remove(adp, aname, tvc);
return 0;
}
#else
osi_vnhold(tvc, 0);
#endif
ReleaseReadLock(&afs_xvcache);
}
return tvc;
}
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;
}
struct vcache *
osi_dnlc_lookup(struct vcache *adp, char *aname, int locktype)
{
struct vcache *tvc;
int LRUme;
unsigned int key, skey;
char *ts = aname;
struct nc *tnc, *tnc1 = 0;
int safety;
#ifdef AFS_DARWIN80_ENV
vnode_t tvp;
#endif
ma_critical_enter();
if (!afs_usednlc) {
ma_critical_exit();
return 0;
}
dnlcHash(ts, key); /* leaves ts pointing at the NULL */
if (ts - aname >= AFSNCNAMESIZE) {
ma_critical_exit();
return 0;
}
skey = key & (NHSIZE - 1);
TRACE(osi_dnlc_lookupT, skey);
dnlcstats.lookups++;
ObtainReadLock(&afs_xvcache);
ObtainReadLock(&afs_xdnlc);
for (tvc = NULL, tnc = nameHash[skey], safety = 0; tnc;
tnc = tnc->next, safety++) {
if ( /* (tnc->key == key) && */ (tnc->dirp == adp)
&& (!strcmp((char *)tnc->name, aname))) {
tvc = tnc->vp;
tnc1 = tnc;
break;
} else if (tnc->next == nameHash[skey]) { /* end of list */
break;
} else if (safety > NCSIZE) {
afs_warn("DNLC cycle");
dnlcstats.cycles++;
ReleaseReadLock(&afs_xdnlc);
ReleaseReadLock(&afs_xvcache);
osi_dnlc_purge();
ma_critical_exit();
return (0);
}
}
LRUme = 0; /* (tnc != nameHash[skey]); */
ReleaseReadLock(&afs_xdnlc);
if (!tvc) {
ReleaseReadLock(&afs_xvcache);
dnlcstats.misses++;
} else {
if ((tvc->f.states & CVInit)
#ifdef AFS_DARWIN80_ENV
||(tvc->f.states & CDeadVnode)
#endif
)
{
ReleaseReadLock(&afs_xvcache);
dnlcstats.misses++;
osi_dnlc_remove(adp, aname, tvc);
ma_critical_exit();
return 0;
}
#if defined(AFS_DARWIN80_ENV)
tvp = AFSTOV(tvc);
if (vnode_get(tvp)) {
ReleaseReadLock(&afs_xvcache);
dnlcstats.misses++;
osi_dnlc_remove(adp, aname, tvc);
ma_critical_exit();
return 0;
}
if (vnode_ref(tvp)) {
ReleaseReadLock(&afs_xvcache);
AFS_GUNLOCK();
vnode_put(tvp);
AFS_GLOCK();
dnlcstats.misses++;
osi_dnlc_remove(adp, aname, tvc);
ma_critical_exit();
return 0;
}
#elif defined(AFS_FBSD_ENV)
/* can't sleep in a critical section */
ma_critical_exit();
osi_vnhold(tvc, 0);
ma_critical_enter();
#else
osi_vnhold(tvc, 0);
#endif
ReleaseReadLock(&afs_xvcache);
#ifdef notdef
/*
* XX If LRUme ever is non-zero change the if statement around because
* aix's cc with optimizer on won't necessarily check things in order XX
*/
if (LRUme && (0 == NBObtainWriteLock(&afs_xdnlc))) {
/* don't block to do this */
/* tnc might have been moved during race condition, */
/* but it's always in a legit hash chain when a lock is granted,
* or else it's on the freelist so prev == NULL,
* so at worst this is redundant */
/* Now that we've got it held, and a lock on the dnlc, we
* should check to be sure that there was no race, and
* bail out if there was. */
if (tnc->prev) {
/* special case for only two elements on list - relative ordering
* doesn't change */
if (tnc->prev != tnc->next) {
/* remove from old location */
tnc->prev->next = tnc->next;
tnc->next->prev = tnc->prev;
/* insert into new location */
tnc->next = nameHash[skey];
tnc->prev = tnc->next->prev;
tnc->next->prev = tnc;
tnc->prev->next = tnc;
}
nameHash[skey] = tnc;
}
ReleaseWriteLock(&afs_xdnlc);
}
#endif
}
ma_critical_exit();
return tvc;
}
/* This function always holds the GLOCK whilst it is running. The caller
* gets the GLOCK before invoking it, and afs_osi_Sleep drops the GLOCK
* whilst we are sleeping, and regains it when we're woken up.
*/
void
afs_Daemon(void)
{
afs_int32 code;
struct afs_exporter *exporter;
afs_int32 now;
afs_int32 last3MinCheck, last10MinCheck, last60MinCheck, lastNMinCheck;
afs_int32 last1MinCheck, last5MinCheck;
afs_uint32 lastCBSlotBump;
char cs_warned = 0;
AFS_STATCNT(afs_Daemon);
afs_rootFid.Fid.Volume = 0;
while (afs_initState < 101)
afs_osi_Sleep(&afs_initState);
#ifdef AFS_DARWIN80_ENV
if (afs_osi_ctxtp_initialized)
osi_Panic("vfs context already initialized");
while (afs_osi_ctxtp && vfs_context_ref)
afs_osi_Sleep(&afs_osi_ctxtp);
if (afs_osi_ctxtp && !vfs_context_ref)
vfs_context_rele(afs_osi_ctxtp);
afs_osi_ctxtp = vfs_context_create(NULL);
afs_osi_ctxtp_initialized = 1;
#endif
now = osi_Time();
lastCBSlotBump = now;
/* when a lot of clients are booted simultaneously, they develop
* annoying synchronous VL server bashing behaviors. So we stagger them.
*/
last1MinCheck = now + ((afs_random() & 0x7fffffff) % 60); /* an extra 30 */
last3MinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
last60MinCheck = now - 1800 + ((afs_random() & 0x7fffffff) % 3600);
last10MinCheck = now - 300 + ((afs_random() & 0x7fffffff) % 600);
last5MinCheck = now - 150 + ((afs_random() & 0x7fffffff) % 300);
lastNMinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
/* start off with afs_initState >= 101 (basic init done) */
while (1) {
afs_CheckCallbacks(20); /* unstat anything which will expire soon */
/* things to do every 20 seconds or less - required by protocol spec */
if (afs_nfsexporter)
afs_FlushActiveVcaches(0); /* flush NFS writes */
afs_FlushVCBs(1); /* flush queued callbacks */
afs_MaybeWakeupTruncateDaemon(); /* free cache space if have too */
rx_CheckPackets(); /* Does RX need more packets? */
now = osi_Time();
if (lastCBSlotBump + CBHTSLOTLEN < now) { /* pretty time-dependant */
lastCBSlotBump = now;
if (afs_BumpBase()) {
afs_CheckCallbacks(20); /* unstat anything which will expire soon */
}
}
if (last1MinCheck + 60 < now) {
/* things to do every minute */
DFlush(); /* write out dir buffers */
afs_WriteThroughDSlots(); /* write through cacheinfo entries */
ObtainWriteLock(&afs_xvcache, 736);
afs_FlushReclaimedVcaches();
ReleaseWriteLock(&afs_xvcache);
afs_FlushActiveVcaches(1); /* keep flocks held & flush nfs writes */
#if 0
afs_StoreDirtyVcaches();
#endif
afs_CheckRXEpoch();
last1MinCheck = now;
}
if (last3MinCheck + 180 < now) {
afs_CheckTokenCache(); /* check for access cache resets due to expired
* tickets */
last3MinCheck = now;
}
if (afsd_dynamic_vcaches && (last5MinCheck + 300 < now)) {
/* start with trying to drop us back to our base usage */
int anumber = VCACHE_FREE + (afs_vcount - afs_cacheStats);
if (anumber > 0) {
ObtainWriteLock(&afs_xvcache, 734);
afs_ShakeLooseVCaches(anumber);
ReleaseWriteLock(&afs_xvcache);
}
last5MinCheck = now;
}
if (!afs_CheckServerDaemonStarted) {
/* Do the check here if the correct afsd is not installed. */
if (!cs_warned) {
cs_warned = 1;
afs_warn("Please install afsd with check server daemon.\n");
}
if (lastNMinCheck + afs_probe_interval < now) {
/* only check down servers */
afs_CheckServers(1, NULL);
lastNMinCheck = now;
}
}
if (last10MinCheck + 600 < now) {
#ifdef AFS_USERSPACE_IP_ADDR
extern int rxi_GetcbiInfo(void);
#endif
afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, 600);
#ifdef AFS_USERSPACE_IP_ADDR
if (rxi_GetcbiInfo()) { /* addresses changed from last time */
afs_FlushCBs();
}
#else /* AFS_USERSPACE_IP_ADDR */
if (rxi_GetIFInfo()) { /* addresses changed from last time */
afs_FlushCBs();
}
#endif /* else AFS_USERSPACE_IP_ADDR */
if (!afs_CheckServerDaemonStarted)
afs_CheckServers(0, NULL);
afs_GCUserData(0); /* gc old conns */
/* This is probably the wrong way of doing GC for the various exporters but it will suffice for a while */
for (exporter = root_exported; exporter;
exporter = exporter->exp_next) {
(void)EXP_GC(exporter, 0); /* Generalize params */
}
{
static int cnt = 0;
if (++cnt < 12) {
afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
AFS_VOLCHECK_BUSY);
} else {
cnt = 0;
afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
AFS_VOLCHECK_BUSY |
AFS_VOLCHECK_MTPTS);
}
}
last10MinCheck = now;
}
if (last60MinCheck + 3600 < now) {
afs_Trace1(afs_iclSetp, CM_TRACE_PROBEVOLUME, ICL_TYPE_INT32,
3600);
afs_CheckRootVolume();
#if AFS_GCPAGS
if (afs_gcpags == AFS_GCPAGS_OK) {
afs_int32 didany;
afs_GCPAGs(&didany);
}
#endif
last60MinCheck = now;
}
if (afs_initState < 300) { /* while things ain't rosy */
code = afs_CheckRootVolume();
if (code == 0)
afs_initState = 300; /* succeeded */
if (afs_initState < 200)
afs_initState = 200; /* tried once */
afs_osi_Wakeup(&afs_initState);
}
/* 18285 is because we're trying to divide evenly into 128, that is,
* CBSlotLen, while staying just under 20 seconds. If CBSlotLen
* changes, should probably change this interval, too.
* Some of the preceding actions may take quite some time, so we
* might not want to wait the entire interval */
now = 18285 - (osi_Time() - now);
if (now > 0) {
afs_osi_Wait(now, &AFS_WaitHandler, 0);
}
if (afs_termState == AFSOP_STOP_AFS) {
if (afs_CheckServerDaemonStarted)
afs_termState = AFSOP_STOP_CS;
else
afs_termState = AFSOP_STOP_TRUNCDAEMON;
afs_osi_Wakeup(&afs_termState);
return;
}
}
}
/* lp is pointer to a fairly-old buffer */
static struct buffer *
afs_newslot(struct dcache *adc, afs_int32 apage, struct buffer *lp)
{
/* Find a usable buffer slot */
afs_int32 i;
afs_int32 lt = 0;
struct buffer *tp;
struct osi_file *tfile;
AFS_STATCNT(afs_newslot);
/* we take a pointer here to a buffer which was at the end of an
* LRU hash chain. Odds are, it's one of the older buffers, not
* one of the newer. Having an older buffer to start with may
* permit us to avoid a few of the assignments in the "typical
* case" for loop below.
*/
if (lp && (lp->lockers == 0)) {
lt = lp->accesstime;
} else {
lp = NULL;
}
/* timecounter might have wrapped, if machine is very very busy
* and stays up for a long time. Timecounter mustn't wrap twice
* (positive->negative->positive) before calling newslot, but that
* would require 2 billion consecutive cache hits... Anyway, the
* penalty is only that the cache replacement policy will be
* almost MRU for the next ~2 billion DReads... newslot doesn't
* get called nearly as often as DRead, so in order to avoid the
* performance penalty of using the hypers, it's worth doing the
* extra check here every time. It's probably cheaper than doing
* hcmp, anyway. There is a little performance hit resulting from
* resetting all the access times to 0, but it only happens once
* every month or so, and the access times will rapidly sort
* themselves back out after just a few more DReads.
*/
if (timecounter < 0) {
timecounter = 1;
tp = Buffers;
for (i = 0; i < nbuffers; i++, tp++) {
tp->accesstime = 0;
if (!lp && !tp->lockers) /* one is as good as the rest, I guess */
lp = tp;
}
} else {
/* this is the typical case */
tp = Buffers;
for (i = 0; i < nbuffers; i++, tp++) {
if (tp->lockers == 0) {
if (!lp || tp->accesstime < lt) {
lp = tp;
lt = tp->accesstime;
}
}
}
}
if (lp == 0) {
/* No unlocked buffers. If still possible, allocate a new increment */
if (nbuffers + NPB > afs_max_buffers) {
/* There are no unlocked buffers -- this used to panic, but that
* seems extreme. To the best of my knowledge, all the callers
* of DRead are prepared to handle a zero return. Some of them
* just panic directly, but not all of them. */
afs_warn("afs: all buffers locked\n");
return 0;
}
BufferData = afs_osi_Alloc(AFS_BUFFER_PAGESIZE * NPB);
osi_Assert(BufferData != NULL);
for (i = 0; i< NPB; i++) {
/* Fill in each buffer with an empty indication. */
tp = &Buffers[i + nbuffers];
tp->fid = NULLIDX;
afs_reset_inode(&tp->inode);
tp->accesstime = 0;
tp->lockers = 0;
tp->data = &BufferData[AFS_BUFFER_PAGESIZE * i];
tp->hashIndex = 0;
tp->dirty = 0;
AFS_RWLOCK_INIT(&tp->lock, "buffer lock");
}
lp = &Buffers[nbuffers];
nbuffers += NPB;
}
if (lp->dirty) {
/* see DFlush for rationale for not getting and locking the dcache */
tfile = afs_CFileOpen(&lp->inode);
afs_CFileWrite(tfile, lp->page * AFS_BUFFER_PAGESIZE, lp->data,
AFS_BUFFER_PAGESIZE);
lp->dirty = 0;
afs_CFileClose(tfile);
AFS_STATS(afs_stats_cmperf.bufFlushDirty++);
}
/* Now fill in the header. */
lp->fid = adc->index;
afs_copy_inode(&lp->inode, &adc->f.inode);
lp->page = apage;
lp->accesstime = timecounter++;
FixupBucket(lp); /* move to the right hash bucket */
return lp;
}
/**
* UFS specific version of afs_GetVolSlot
* @return
*/
struct volume *
afs_UFSGetVolSlot(void)
{
struct volume *tv = NULL, **lv;
struct osi_file *tfile;
afs_int32 i = -1, code;
afs_int32 bestTime;
struct volume *bestVp, *oldLp = NULL, **bestLp = NULL;
char *oldname = NULL;
afs_int32 oldvtix = -2; /* Initialize to a value that doesn't occur */
AFS_STATCNT(afs_UFSGetVolSlot);
if (!afs_freeVolList) {
/* get free slot */
bestTime = 0x7fffffff;
bestVp = 0;
bestLp = 0;
for (i = 0; i < NVOLS; i++) {
lv = &afs_volumes[i];
for (tv = *lv; tv; lv = &tv->next, tv = *lv) {
if (tv->refCount == 0) { /* is this one available? */
if (tv->accessTime < bestTime) { /* best one available? */
bestTime = tv->accessTime;
bestLp = lv;
bestVp = tv;
}
}
}
}
if (!bestVp) {
afs_warn("afs_UFSGetVolSlot: no vol slots available\n");
goto error;
}
tv = bestVp;
oldLp = *bestLp;
*bestLp = tv->next;
oldname = tv->name;
tv->name = NULL;
oldvtix = tv->vtix;
/* now write out volume structure to file */
if (tv->vtix < 0) {
tv->vtix = afs_volCounter++;
/* now put on hash chain */
i = FVHash(tv->cell, tv->volume);
staticFVolume.next = fvTable[i];
fvTable[i] = tv->vtix;
} else {
/*
* Haul the guy in from disk so we don't overwrite hash table
* next chain
*/
if (afs_FVIndex != tv->vtix) {
tfile = osi_UFSOpen(&volumeInode);
code =
afs_osi_Read(tfile, sizeof(struct fvolume) * tv->vtix,
&staticFVolume, sizeof(struct fvolume));
osi_UFSClose(tfile);
if (code != sizeof(struct fvolume)) {
afs_warn("afs_UFSGetVolSlot: error %d reading volumeinfo\n",
(int)code);
goto error;
}
afs_FVIndex = tv->vtix;
}
}
afs_FVIndex = tv->vtix;
staticFVolume.volume = tv->volume;
staticFVolume.cell = tv->cell;
staticFVolume.mtpoint = tv->mtpoint;
staticFVolume.dotdot = tv->dotdot;
staticFVolume.rootVnode = tv->rootVnode;
staticFVolume.rootUnique = tv->rootUnique;
tfile = osi_UFSOpen(&volumeInode);
code =
afs_osi_Write(tfile, sizeof(struct fvolume) * afs_FVIndex,
&staticFVolume, sizeof(struct fvolume));
osi_UFSClose(tfile);
if (code != sizeof(struct fvolume)) {
afs_warn("afs_UFSGetVolSlot: error %d writing volumeinfo\n",
(int)code);
goto error;
}
if (oldname) {
afs_osi_Free(oldname, strlen(oldname) + 1);
oldname = NULL;
}
} else {
tv = afs_freeVolList;
afs_freeVolList = tv->next;
}
return tv;
error:
if (tv) {
if (oldvtix == -2) {
afs_warn("afs_UFSGetVolSlot: oldvtix is uninitialized\n");
return NULL;
}
if (oldname) {
tv->name = oldname;
oldname = NULL;
}
if (oldvtix < 0) {
afs_volCounter--;
fvTable[i] = staticFVolume.next;
}
if (bestLp) {
*bestLp = oldLp;
}
tv->vtix = oldvtix;
/* we messed with staticFVolume, so make sure someone else
* doesn't think it's fine to use */
afs_FVIndex = -1;
}
return NULL;
} /*afs_UFSGetVolSlot */
/**
*
* @param volid Volume ID. If it's 0, get it from the name.
* @param aname Volume name.
* @param ve Volume entry.
* @param tcell The cell containing this volume.
* @param agood
* @param type Type of volume.
* @param areq Request.
* @return Volume or NULL if failure.
*/
static struct volume *
afs_SetupVolume(afs_int32 volid, char *aname, void *ve, struct cell *tcell,
afs_int32 agood, afs_int32 type, struct vrequest *areq)
{
struct volume *tv;
struct vldbentry *ove = (struct vldbentry *)ve;
struct nvldbentry *nve = (struct nvldbentry *)ve;
struct uvldbentry *uve = (struct uvldbentry *)ve;
int whichType; /* which type of volume to look for */
int i, j, err = 0;
if (!volid) {
int len;
/* special hint from file server to use vlserver */
len = strlen(aname);
if (len >= 8 && strcmp(aname + len - 7, ".backup") == 0)
whichType = BACKVOL;
else if (len >= 10 && strcmp(aname + len - 9, ".readonly") == 0)
whichType = ROVOL;
else
whichType = RWVOL;
/* figure out which one we're really interested in (a set is returned) */
volid = afs_vtoi(aname);
if (volid == 0) {
if (type == 2) {
volid = uve->volumeId[whichType];
} else if (type == 1) {
volid = nve->volumeId[whichType];
} else {
volid = ove->volumeId[whichType];
}
} /* end of if (volid == 0) */
} /* end of if (!volid) */
ObtainWriteLock(&afs_xvolume, 108);
i = VHash(volid);
for (tv = afs_volumes[i]; tv; tv = tv->next) {
if (tv->volume == volid && tv->cell == tcell->cellNum) {
break;
}
}
if (!tv) {
struct fvolume *tf = 0;
tv = afs_GetVolSlot();
if (!tv) {
ReleaseWriteLock(&afs_xvolume);
return NULL;
}
memset(tv, 0, sizeof(struct volume));
for (j = fvTable[FVHash(tcell->cellNum, volid)]; j != 0; j = tf->next) {
if (afs_FVIndex != j) {
struct osi_file *tfile;
tfile = osi_UFSOpen(&volumeInode);
err =
afs_osi_Read(tfile, sizeof(struct fvolume) * j,
&staticFVolume, sizeof(struct fvolume));
osi_UFSClose(tfile);
if (err != sizeof(struct fvolume)) {
afs_warn("afs_SetupVolume: error %d reading volumeinfo\n",
(int)err);
/* put tv back on the free list; the data in it is not valid */
tv->next = afs_freeVolList;
afs_freeVolList = tv;
/* staticFVolume contents are not valid */
afs_FVIndex = -1;
ReleaseWriteLock(&afs_xvolume);
return NULL;
}
afs_FVIndex = j;
}
tf = &staticFVolume;
if (tf->cell == tcell->cellNum && tf->volume == volid)
break;
}
tv->cell = tcell->cellNum;
AFS_RWLOCK_INIT(&tv->lock, "volume lock");
tv->next = afs_volumes[i]; /* thread into list */
afs_volumes[i] = tv;
tv->volume = volid;
if (tf && (j != 0)) {
tv->vtix = afs_FVIndex;
tv->mtpoint = tf->mtpoint;
tv->dotdot = tf->dotdot;
tv->rootVnode = tf->rootVnode;
tv->rootUnique = tf->rootUnique;
} else {
tv->vtix = -1;
tv->rootVnode = tv->rootUnique = 0;
afs_GetDynrootMountFid(&tv->dotdot);
afs_GetDynrootMountFid(&tv->mtpoint);
tv->mtpoint.Fid.Vnode =
VNUM_FROM_TYPEID(VN_TYPE_MOUNT, tcell->cellIndex << 2);
tv->mtpoint.Fid.Unique = volid;
}
}
tv->refCount++;
tv->states &= ~VRecheck; /* just checked it */
tv->accessTime = osi_Time();
ReleaseWriteLock(&afs_xvolume);
if (type == 2) {
LockAndInstallUVolumeEntry(tv, uve, tcell->cellNum, tcell, areq);
} else if (type == 1)
LockAndInstallNVolumeEntry(tv, nve, tcell->cellNum);
else
LockAndInstallVolumeEntry(tv, ove, tcell->cellNum);
if (agood) {
if (!tv->name) {
tv->name = afs_osi_Alloc(strlen(aname) + 1);
osi_Assert(tv->name != NULL);
strcpy(tv->name, aname);
}
}
for (i = 0; i < NMAXNSERVERS; i++) {
tv->status[i] = not_busy;
}
ReleaseWriteLock(&tv->lock);
return tv;
}
int
osi_dnlc_enter(struct vcache *adp, char *aname, struct vcache *avc,
afs_hyper_t * avno)
{
struct nc *tnc;
unsigned int key, skey;
char *ts = aname;
int safety;
if (!afs_usednlc)
return 0;
TRACE(osi_dnlc_enterT, 0);
dnlcHash(ts, key); /* leaves ts pointing at the NULL */
if (ts - aname >= AFSNCNAMESIZE) {
return 0;
}
skey = key & (NHSIZE - 1);
dnlcstats.enters++;
retry:
ObtainWriteLock(&afs_xdnlc, 222);
/* Only cache entries from the latest version of the directory */
if (!(adp->f.states & CStatd) || !hsame(*avno, adp->f.m.DataVersion)) {
ReleaseWriteLock(&afs_xdnlc);
return 0;
}
/*
* Make sure each directory entry gets cached no more than once.
*/
for (tnc = nameHash[skey], safety = 0; tnc; tnc = tnc->next, safety++) {
if ((tnc->dirp == adp) && (!strcmp((char *)tnc->name, aname))) {
/* duplicate entry */
break;
} else if (tnc->next == nameHash[skey]) { /* end of list */
tnc = NULL;
break;
} else if (safety > NCSIZE) {
afs_warn("DNLC cycle");
dnlcstats.cycles++;
ReleaseWriteLock(&afs_xdnlc);
osi_dnlc_purge();
goto retry;
}
}
if (tnc == NULL) {
tnc = GetMeAnEntry();
tnc->dirp = adp;
tnc->vp = avc;
tnc->key = key;
memcpy((char *)tnc->name, aname, ts - aname + 1); /* include the NULL */
InsertEntry(tnc);
} else {
/* duplicate */
tnc->vp = avc;
}
ReleaseWriteLock(&afs_xdnlc);
return 0;
}
