void
afspag_Init(afs_int32 nfs_server_addr)
{
struct clientcred ccred;
struct rmtbulk idata, odata;
afs_int32 code, err, addr, obuf;
int i;
afs_uuid_create(&afs_cb_interface.uuid);
AFS_GLOCK();
afs_InitStats();
rx_Init(htons(7001));
AFS_STATCNT(afs_ResourceInit);
AFS_RWLOCK_INIT(&afs_xuser, "afs_xuser");
AFS_RWLOCK_INIT(&afs_xpagcell, "afs_xpagcell");
AFS_RWLOCK_INIT(&afs_xpagsys, "afs_xpagsys");
AFS_RWLOCK_INIT(&afs_icl_lock, "afs_icl_lock");
afs_resourceinit_flag = 1;
afs_nfs_server_addr = nfs_server_addr;
for (i = 0; i < MAXNUMSYSNAMES; i++) {
afs_sysnamelist[i] = afs_osi_Alloc(MAXSYSNAME);
osi_Assert(afs_sysnamelist[i] != NULL);
}
afs_sysname = afs_sysnamelist[0];
strcpy(afs_sysname, SYS_NAME);
afs_sysnamecount = 1;
afs_sysnamegen++;
srv_secobj = rxnull_NewServerSecurityObject();
stats_svc = rx_NewService(0, RX_STATS_SERVICE_ID, "rpcstats", &srv_secobj,
1, RXSTATS_ExecuteRequest);
pagcb_svc = rx_NewService(0, PAGCB_SERVICEID, "pagcb", &srv_secobj,
1, PAGCB_ExecuteRequest);
rx_StartServer(0);
clt_secobj = rxnull_NewClientSecurityObject();
rmtsys_conn = rx_NewConnection(nfs_server_addr, htons(7009),
RMTSYS_SERVICEID, clt_secobj, 0);
#ifdef RXK_LISTENER_ENV
afs_start_thread(rxk_Listener, "Rx Listener");
#endif
afs_start_thread((void *)(void *)rx_ServerProc, "Rx Server Thread");
afs_start_thread(afs_rxevent_daemon, "Rx Event Daemon");
afs_start_thread(afs_Daemon, "AFS PAG Daemon");
afs_icl_InitLogs();
AFS_GUNLOCK();
/* If it's reachable, tell the translator to nuke our creds.
* We should be more agressive about making sure this gets done,
* even if the translator is unreachable when we boot.
*/
addr = obuf = err = 0;
idata.rmtbulk_len = sizeof(addr);
idata.rmtbulk_val = (char *)&addr;
odata.rmtbulk_len = sizeof(obuf);
odata.rmtbulk_val = (char *)&obuf;
memset(&ccred, 0, sizeof(ccred));
code = RMTSYS_Pioctl(rmtsys_conn, &ccred, NIL_PATHP, 0x4F01, 0,
&idata, &odata, &err);
} /*afs_ResourceInit */
struct buf *
afs_get_bioreq()
{
struct buf *bp = NULL;
struct buf *bestbp;
struct buf **bestlbpP, **lbpP;
long bestage, stop;
struct buf *t1P, *t2P; /* temp pointers for list manipulation */
int oldPriority;
afs_uint32 wait_ret;
struct afs_bioqueue *s;
/* ??? Does the forward pointer of the returned buffer need to be NULL?
*/
/* Disable interrupts from the strategy function, and save the
* prior priority level and lock access to the afs_asyncbuf.
*/
AFS_GUNLOCK();
oldPriority = disable_lock(INTMAX, &afs_asyncbuf_lock);
while (1) {
if (afs_asyncbuf) {
/* look for oldest buffer */
bp = bestbp = afs_asyncbuf;
bestage = (long)bestbp->av_back;
bestlbpP = &afs_asyncbuf;
while (1) {
lbpP = &bp->av_forw;
bp = *lbpP;
if (!bp)
break;
if ((long)bp->av_back - bestage < 0) {
bestbp = bp;
bestlbpP = lbpP;
bestage = (long)bp->av_back;
}
}
bp = bestbp;
*bestlbpP = bp->av_forw;
break;
} else {
/* If afs_asyncbuf is null, it is necessary to go to sleep.
* e_wakeup_one() ensures that only one thread wakes.
*/
int interrupted;
/* The LOCK_HANDLER indicates to e_sleep_thread to only drop the
* lock on an MP machine.
*/
interrupted =
e_sleep_thread(&afs_asyncbuf_cv, &afs_asyncbuf_lock,
LOCK_HANDLER | INTERRUPTIBLE);
if (interrupted == THREAD_INTERRUPTED) {
/* re-enable interrupts from strategy */
unlock_enable(oldPriority, &afs_asyncbuf_lock);
AFS_GLOCK();
return (NULL);
}
} /* end of "else asyncbuf is empty" */
} /* end of "inner loop" */
/*assert (bp); */
unlock_enable(oldPriority, &afs_asyncbuf_lock);
AFS_GLOCK();
/* For the convenience of other code, replace the gnodes in
* the b_vp field of bp and the other buffers on the b_work
* chain with the corresponding vnodes.
*
* ??? what happens to the gnodes? They're not just cut loose,
* are they?
*/
for (t1P = bp;;) {
t2P = (struct buf *)t1P->b_work;
t1P->b_vp = ((struct gnode *)t1P->b_vp)->gn_vnode;
if (!t2P)
break;
t1P = (struct buf *)t2P->b_work;
t2P->b_vp = ((struct gnode *)t2P->b_vp)->gn_vnode;
if (!t1P)
break;
}
/* If the buffer does not specify I/O, it may immediately
* be returned to the caller. This condition is detected
* by examining the buffer's flags (the b_flags field). If
* the B_PFPROT bit is set, the buffer represents a protection
* violation, rather than a request for I/O. The remainder
* of the outer loop handles the case where the B_PFPROT bit is clear.
*/
if (bp->b_flags & B_PFPROT) {
return (bp);
}
return (bp);
} /* end of function get_bioreq() */
/* this call, unlike osi_FlushText, is supposed to discard caches that may
contain invalid information if a file is written remotely, but that may
contain valid information that needs to be written back if the file is
being written locally. It doesn't subsume osi_FlushText, since the latter
function may be needed to flush caches that are invalidated by local writes.
avc->pvnLock is already held, avc->lock is guaranteed not to be held (by
us, of course).
*/
void
osi_FlushPages(struct vcache *avc, afs_ucred_t *credp)
{
afs_hyper_t origDV;
#if defined(AFS_CACHE_BYPASS)
/* The optimization to check DV under read lock below is identical a
* change in CITI cache bypass work. The problem CITI found in 1999
* was that this code and background daemon doing prefetching competed
* for the vcache entry shared lock. It's not clear to me from the
* tech report, but it looks like CITI fixed the general prefetch code
* path as a bonus when experimenting on prefetch for cache bypass, see
* citi-tr-01-3.
*/
#endif
if (vType(avc) == VDIR) {
/* not applicable to directories; they're never mapped or stored in
* pages */
return;
}
ObtainReadLock(&avc->lock);
/* If we've already purged this version, or if we're the ones
* writing this version, don't flush it (could lose the
* data we're writing). */
if ((hcmp((avc->f.m.DataVersion), (avc->mapDV)) <= 0)
|| ((avc->execsOrWriters > 0) && afs_DirtyPages(avc))) {
ReleaseReadLock(&avc->lock);
return;
}
ReleaseReadLock(&avc->lock);
ObtainWriteLock(&avc->lock, 10);
/* Check again */
if ((hcmp((avc->f.m.DataVersion), (avc->mapDV)) <= 0)
|| ((avc->execsOrWriters > 0) && afs_DirtyPages(avc))) {
ReleaseWriteLock(&avc->lock);
return;
}
/* At this point, you might think that we can skip trying to flush pages
* if mapDV is zero, since a file with a zero DV will not have any data in
* it. However, some platforms (notably Linux 2.6.22+) will keep a page
* full of zeroes around for an empty file. So play it safe and always
* flush pages. */
AFS_STATCNT(osi_FlushPages);
hset(origDV, avc->f.m.DataVersion);
afs_Trace3(afs_iclSetp, CM_TRACE_FLUSHPAGES, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, origDV.low, ICL_TYPE_INT32, avc->f.m.Length);
ReleaseWriteLock(&avc->lock);
#ifndef AFS_FBSD70_ENV
AFS_GUNLOCK();
#endif
osi_VM_FlushPages(avc, credp);
#ifndef AFS_FBSD70_ENV
AFS_GLOCK();
#endif
ObtainWriteLock(&avc->lock, 88);
/* do this last, and to original version, since stores may occur
* while executing above PUTPAGE call */
hset(avc->mapDV, origDV);
ReleaseWriteLock(&avc->lock);
}
osi_UFSOpen(afs_int32 ainode)
#endif
{
register struct osi_file *afile = NULL;
extern int cacheDiskType;
struct inode *tip = NULL;
struct dentry *dp = NULL;
struct file *filp = NULL;
#if !defined(HAVE_IGET) || defined(LINUX_USE_FH)
struct fid fid;
#endif
AFS_STATCNT(osi_UFSOpen);
if (cacheDiskType != AFS_FCACHE_TYPE_UFS) {
osi_Panic("UFSOpen called for non-UFS cache\n");
}
if (!afs_osicred_initialized) {
/* valid for alpha_osf, SunOS, Ultrix */
memset((char *)&afs_osi_cred, 0, sizeof(struct AFS_UCRED));
crhold(&afs_osi_cred); /* don't let it evaporate, since it is static */
afs_osicred_initialized = 1;
}
afile = (struct osi_file *)osi_AllocLargeSpace(sizeof(struct osi_file));
AFS_GUNLOCK();
if (!afile) {
osi_Panic("osi_UFSOpen: Failed to allocate %d bytes for osi_file.\n",
sizeof(struct osi_file));
}
memset(afile, 0, sizeof(struct osi_file));
#if defined(HAVE_IGET)
tip = iget(afs_cacheSBp, (u_long) ainode);
if (!tip)
osi_Panic("Can't get inode %d\n", ainode);
dp = d_alloc_anon(tip);
#else
#if defined(LINUX_USE_FH)
dp = afs_cacheSBp->s_export_op->fh_to_dentry(afs_cacheSBp, fh, sizeof(struct fid), fh_type);
#else
fid.i32.ino = ainode;
fid.i32.gen = 0;
dp = afs_cacheSBp->s_export_op->fh_to_dentry(afs_cacheSBp, &fid, sizeof(fid), FILEID_INO32_GEN);
#endif
if (!dp)
osi_Panic("Can't get dentry\n");
tip = dp->d_inode;
#endif
tip->i_flags |= MS_NOATIME; /* Disable updating access times. */
#if defined(STRUCT_TASK_HAS_CRED)
filp = dentry_open(dp, mntget(afs_cacheMnt), O_RDWR, current_cred());
#else
filp = dentry_open(dp, mntget(afs_cacheMnt), O_RDWR);
#endif
if (IS_ERR(filp))
#if defined(LINUX_USE_FH)
osi_Panic("Can't open file\n");
#else
osi_Panic("Can't open inode %d\n", ainode);
#endif
afile->filp = filp;
afile->size = i_size_read(FILE_INODE(filp));
AFS_GLOCK();
afile->offset = 0;
afile->proc = (int (*)())0;
#if defined(LINUX_USE_FH)
afile->inum = tip->i_ino; /* for hint validity checking */
#else
afile->inum = ainode; /* for hint validity checking */
#endif
return (void *)afile;
}
/* this call, unlike osi_FlushText, is supposed to discard caches that may
contain invalid information if a file is written remotely, but that may
contain valid information that needs to be written back if the file is
being written locally. It doesn't subsume osi_FlushText, since the latter
function may be needed to flush caches that are invalidated by local writes.
avc->pvnLock is already held, avc->lock is guaranteed not to be held (by
us, of course).
*/
void
osi_FlushPages(struct vcache *avc, afs_ucred_t *credp)
{
#ifdef AFS_FBSD70_ENV
int vfslocked;
#endif
afs_hyper_t origDV;
#if defined(AFS_CACHE_BYPASS)
/* The optimization to check DV under read lock below is identical a
* change in CITI cache bypass work. The problem CITI found in 1999
* was that this code and background daemon doing prefetching competed
* for the vcache entry shared lock. It's not clear to me from the
* tech report, but it looks like CITI fixed the general prefetch code
* path as a bonus when experimenting on prefetch for cache bypass, see
* citi-tr-01-3.
*/
#endif
ObtainReadLock(&avc->lock);
/* If we've already purged this version, or if we're the ones
* writing this version, don't flush it (could lose the
* data we're writing). */
if ((hcmp((avc->f.m.DataVersion), (avc->mapDV)) <= 0)
|| ((avc->execsOrWriters > 0) && afs_DirtyPages(avc))) {
ReleaseReadLock(&avc->lock);
return;
}
ReleaseReadLock(&avc->lock);
ObtainWriteLock(&avc->lock, 10);
/* Check again */
if ((hcmp((avc->f.m.DataVersion), (avc->mapDV)) <= 0)
|| ((avc->execsOrWriters > 0) && afs_DirtyPages(avc))) {
ReleaseWriteLock(&avc->lock);
return;
}
if (hiszero(avc->mapDV)) {
hset(avc->mapDV, avc->f.m.DataVersion);
ReleaseWriteLock(&avc->lock);
return;
}
AFS_STATCNT(osi_FlushPages);
hset(origDV, avc->f.m.DataVersion);
afs_Trace3(afs_iclSetp, CM_TRACE_FLUSHPAGES, ICL_TYPE_POINTER, avc,
ICL_TYPE_INT32, origDV.low, ICL_TYPE_INT32, avc->f.m.Length);
ReleaseWriteLock(&avc->lock);
#ifdef AFS_FBSD70_ENV
vfslocked = VFS_LOCK_GIANT(AFSTOV(avc)->v_mount);
#endif
#ifndef AFS_FBSD70_ENV
AFS_GUNLOCK();
#endif
osi_VM_FlushPages(avc, credp);
#ifndef AFS_FBSD70_ENV
AFS_GLOCK();
#endif
#ifdef AFS_FBSD70_ENV
VFS_UNLOCK_GIANT(vfslocked);
#endif
ObtainWriteLock(&avc->lock, 88);
/* do this last, and to original version, since stores may occur
* while executing above PUTPAGE call */
hset(avc->mapDV, origDV);
ReleaseWriteLock(&avc->lock);
}
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;
}
void *
osi_UfsOpen(afs_dcache_id_t *ainode)
{
#ifdef AFS_CACHE_VNODE_PATH
struct vnode *vp;
#else
struct inode *ip;
#endif
struct osi_file *afile = NULL;
afs_int32 code = 0;
int dummy;
#ifdef AFS_CACHE_VNODE_PATH
char namebuf[1024];
struct pathname lookpn;
#endif
struct osi_stat tstat;
afile = osi_AllocSmallSpace(sizeof(struct osi_file));
AFS_GUNLOCK();
/*
* AFS_CACHE_VNODE_PATH can be used with any file system, including ZFS or tmpfs.
* The ainode is not an inode number but a path.
*/
#ifdef AFS_CACHE_VNODE_PATH
/* Can not use vn_open or lookupname, they use user's CRED()
* We need to run as root So must use low level lookuppnvp
* assume fname starts with /
*/
code = pn_get_buf(ainode->ufs, AFS_UIOSYS, &lookpn, namebuf, sizeof(namebuf));
if (code != 0)
osi_Panic("UfsOpen: pn_get_buf failed %ld %s", code, ainode->ufs);
VN_HOLD(rootdir); /* released in loopuppnvp */
code = lookuppnvp(&lookpn, NULL, FOLLOW, NULL, &vp,
rootdir, rootdir, afs_osi_credp);
if (code != 0)
osi_Panic("UfsOpen: lookuppnvp failed %ld %s", code, ainode->ufs);
#ifdef AFS_SUN511_ENV
code = VOP_OPEN(&vp, FREAD|FWRITE, afs_osi_credp, NULL);
#else
code = VOP_OPEN(&vp, FREAD|FWRITE, afs_osi_credp);
#endif
if (code != 0)
osi_Panic("UfsOpen: VOP_OPEN failed %ld %s", code, ainode->ufs);
#else
code =
igetinode(afs_cacheVfsp, (dev_t) cacheDev.dev, ainode->ufs, &ip,
CRED(), &dummy);
#endif
AFS_GLOCK();
if (code) {
osi_FreeSmallSpace(afile);
osi_Panic("UfsOpen: igetinode failed %ld %s", code, ainode->ufs);
}
#ifdef AFS_CACHE_VNODE_PATH
afile->vnode = vp;
code = afs_osi_Stat(afile, &tstat);
afile->size = tstat.size;
#else
afile->vnode = ITOV(ip);
afile->size = VTOI(afile->vnode)->i_size;
#endif
afile->offset = 0;
afile->proc = (int (*)())0;
return (void *)afile;
}
static int afs_export_get_name(struct dentry *parent, char *name,
struct dentry *child)
{
struct afs_fakestat_state fakestate;
struct get_name_data data;
struct vrequest *treq = NULL;
struct volume *tvp;
struct vcache *vcp;
struct dcache *tdc;
cred_t *credp;
afs_size_t dirOffset, dirLen;
afs_int32 code = 0;
if (!parent->d_inode) {
#ifdef OSI_EXPORT_DEBUG
/* can't lookup name in a negative dentry */
printk("afs: get_name(%s, %s): no parent inode\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
child->d_name.name ? (char *)child->d_name.name : "?");
#endif
return -EIO;
}
if (!child->d_inode) {
#ifdef OSI_EXPORT_DEBUG
/* can't find the FID of negative dentry */
printk("afs: get_name(%s, %s): no child inode\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
child->d_name.name ? (char *)child->d_name.name : "?");
#endif
return -ENOENT;
}
afs_InitFakeStat(&fakestate);
credp = crref();
AFS_GLOCK();
vcp = VTOAFS(child->d_inode);
/* special case dynamic mount directory */
if (afs_IsDynrootMount(vcp)) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): this is the dynmount dir\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
data.fid = vcp->f.fid;
if (VTOAFS(parent->d_inode) == afs_globalVp)
strcpy(name, AFS_DYNROOT_MOUNTNAME);
else
code = -ENOENT;
goto done;
}
/* Figure out what FID to look for */
if (vcp->mvstat == AFS_MVSTAT_ROOT) { /* volume root */
tvp = afs_GetVolume(&vcp->f.fid, 0, READ_LOCK);
if (!tvp) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): no volume for root\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
code = ENOENT;
goto done;
}
data.fid = tvp->mtpoint;
afs_PutVolume(tvp, READ_LOCK);
} else {
data.fid = vcp->f.fid;
}
vcp = VTOAFS(parent->d_inode);
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): parent is 0x%08x/%d/%d.%d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
code = afs_CreateReq(&treq, credp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): afs_CreateReq: %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique, code);
#endif
goto done;
}
/* a dynamic mount point in the dynamic mount directory */
if (afs_IsDynrootMount(vcp) && afs_IsDynrootAnyFid(&data.fid)
&& VNUM_TO_VNTYPE(data.fid.Fid.Vnode) == VN_TYPE_MOUNT) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): dynamic mount point\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique);
#endif
vcp = afs_GetVCache(&data.fid, treq, NULL, NULL);
if (vcp) {
ObtainReadLock(&vcp->lock);
if (strlen(vcp->linkData + 1) <= NAME_MAX)
strcpy(name, vcp->linkData + 1);
else
code = ENOENT;
ReleaseReadLock(&vcp->lock);
afs_PutVCache(vcp);
} else {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): no vcache\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique);
#endif
code = ENOENT;
}
goto done;
}
code = afs_EvalFakeStat(&vcp, &fakestate, treq);
if (code)
goto done;
if (vcp->f.fid.Cell != data.fid.Cell ||
vcp->f.fid.Fid.Volume != data.fid.Fid.Volume) {
/* parent is not the expected cell and volume; thus it
* cannot possibly contain the fid we are looking for */
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): wrong parent 0x%08x/%d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume);
#endif
code = ENOENT;
goto done;
}
redo:
if (!(vcp->f.states & CStatd)) {
if ((code = afs_VerifyVCache2(vcp, treq))) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): VerifyVCache2(0x%08x/%d/%d.%d): %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
goto done;
}
}
tdc = afs_GetDCache(vcp, (afs_size_t) 0, treq, &dirOffset, &dirLen, 1);
if (!tdc) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): GetDCache(0x%08x/%d/%d.%d): %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
code = EIO;
goto done;
}
ObtainReadLock(&vcp->lock);
ObtainReadLock(&tdc->lock);
/*
* Make sure that the data in the cache is current. There are two
* cases we need to worry about:
* 1. The cache data is being fetched by another process.
* 2. The cache data is no longer valid
*/
while ((vcp->f.states & CStatd)
&& (tdc->dflags & DFFetching)
&& hsame(vcp->f.m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&vcp->lock);
afs_osi_Sleep(&tdc->validPos);
ObtainReadLock(&vcp->lock);
ObtainReadLock(&tdc->lock);
}
if (!(vcp->f.states & CStatd)
|| !hsame(vcp->f.m.DataVersion, tdc->f.versionNo)) {
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&vcp->lock);
afs_PutDCache(tdc);
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): dir (0x%08x/%d/%d.%d) changed; retrying\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
goto redo;
}
data.name = name;
data.found = 0;
code = afs_dir_EnumerateDir(tdc, get_name_hook, &data);
if (!code && !data.found) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): not found\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique);
#endif
code = ENOENT;
} else if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_name(%s, 0x%08x/%d/%d.%d): Enumeratedir(0x%08x/%d/%d.%d): %d\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique,
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
}
ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&vcp->lock);
afs_PutDCache(tdc);
done:
if (!code) {
printk("afs: get_name(%s, 0x%08x/%d/%d.%d) => %s\n",
parent->d_name.name ? (char *)parent->d_name.name : "?",
data.fid.Cell, data.fid.Fid.Volume,
data.fid.Fid.Vnode, data.fid.Fid.Unique, name);
}
afs_PutFakeStat(&fakestate);
code = afs_CheckCode(code, treq, 102);
afs_DestroyReq(treq);
AFS_GUNLOCK();
crfree(credp);
return -code;
}
static int afs_encode_fh(struct dentry *de, __u32 *fh, int *max_len,
int connectable)
{
struct vcache *tvc;
struct cell *tc;
int vntype;
if (!de->d_inode) /* encode a negative dentry?! */
return 255;
if (*max_len < 4) /* not enough space */
return 255;
tvc = VTOAFS(de->d_inode);
#ifdef OSI_EXPORT_DEBUG
printk("afs: encode_fh(0x%08x/%d/%d.%d)\n",
tvc->f.fid.Cell, tvc->f.fid.Fid.Volume,
tvc->f.fid.Fid.Vnode, tvc->f.fid.Fid.Unique);
#endif
if (afs_IsDynrootAnyFid(&tvc->f.fid)) {
vntype = VNUM_TO_VNTYPE(tvc->f.fid.Fid.Vnode);
switch (vntype) {
case 0:
/* encode as a normal filehandle */
break;
case VN_TYPE_MOUNT:
if (*max_len < 5) {
return 255;
}
/* fall through */
case VN_TYPE_CELL:
case VN_TYPE_ALIAS:
AFS_GLOCK();
tc = afs_GetCellByIndex(VNUM_TO_CIDX(tvc->f.fid.Fid.Vnode),
READ_LOCK);
if (!tc) {
AFS_GUNLOCK();
return 255;
}
memcpy((void *)fh, tc->cellHandle, 16);
afs_PutCell(tc, READ_LOCK);
AFS_GUNLOCK();
if (vntype == VN_TYPE_MOUNT) {
fh[4] = htonl(tvc->f.fid.Fid.Unique);
*max_len = 5;
return AFSFH_DYN_MOUNT;
}
*max_len = 4;
if (vntype == VN_TYPE_CELL) {
return AFSFH_DYN_RO_CELL | VNUM_TO_RW(tvc->f.fid.Fid.Vnode);
} else {
return AFSFH_DYN_RO_LINK | VNUM_TO_RW(tvc->f.fid.Fid.Vnode);
}
case VN_TYPE_SYMLINK:
/* XXX fill in filehandle for dynroot symlink */
/* XXX return AFSFH_DYN_SYMLINK; */
default:
return 255;
}
}
if (*max_len < 7) {
/* not big enough for a migratable filehandle */
/* always encode in network order */
fh[0] = htonl(tvc->f.fid.Cell);
fh[1] = htonl(tvc->f.fid.Fid.Volume);
fh[2] = htonl(tvc->f.fid.Fid.Vnode);
fh[3] = htonl(tvc->f.fid.Fid.Unique);
*max_len = 4;
return AFSFH_NET_VENUSFID;
}
AFS_GLOCK();
tc = afs_GetCell(tvc->f.fid.Cell, READ_LOCK);
if (!tc) {
AFS_GUNLOCK();
return 255;
}
memcpy((void *)fh, tc->cellHandle, 16);
afs_PutCell(tc, READ_LOCK);
AFS_GUNLOCK();
/* always encode in network order */
fh[4] = htonl(tvc->f.fid.Fid.Volume);
fh[5] = htonl(tvc->f.fid.Fid.Vnode);
fh[6] = htonl(tvc->f.fid.Fid.Unique);
*max_len = 7;
return AFSFH_NET_CELLFID;
}
static struct dentry *afs_decode_fh(struct super_block *sb, __u32 *fh,
int fh_len, int fh_type,
int (*acceptable)(void *, struct dentry *),
void *context)
#endif
{
struct VenusFid fid;
struct cell *tc;
struct dentry *result;
#if defined(NEW_EXPORT_OPS)
__u32 *fh = (__u32 *)fh_fid->raw;
#endif
switch (fh_type) {
case AFSFH_VENUSFID:
if (fh_len != 4)
return NULL;
fid.Cell = fh[0];
fid.Fid.Volume = fh[1];
fid.Fid.Vnode = fh[2];
fid.Fid.Unique = fh[3];
break;
case AFSFH_CELLFID:
if (fh_len != 7)
return NULL;
AFS_GLOCK();
tc = afs_GetCellByHandle((void *)fh, READ_LOCK);
if (!tc) {
AFS_GUNLOCK();
return NULL;
}
fid.Cell = tc->cellNum;
fid.Fid.Volume = fh[4];
fid.Fid.Vnode = fh[5];
fid.Fid.Unique = fh[6];
afs_PutCell(tc, READ_LOCK);
AFS_GUNLOCK();
break;
case AFSFH_NET_VENUSFID:
fid.Cell = ntohl(fh[0]);
fid.Fid.Volume = ntohl(fh[1]);
fid.Fid.Vnode = ntohl(fh[2]);
fid.Fid.Unique = ntohl(fh[3]);
break;
case AFSFH_NET_CELLFID:
if (fh_len != 7)
return NULL;
AFS_GLOCK();
tc = afs_GetCellByHandle((void *)fh, READ_LOCK);
if (!tc) {
AFS_GUNLOCK();
return NULL;
}
fid.Cell = tc->cellNum;
fid.Fid.Volume = ntohl(fh[4]);
fid.Fid.Vnode = ntohl(fh[5]);
fid.Fid.Unique = ntohl(fh[6]);
afs_PutCell(tc, READ_LOCK);
AFS_GUNLOCK();
break;
case AFSFH_DYN_RO_CELL:
case AFSFH_DYN_RW_CELL:
if (fh_len != 4)
return NULL;
AFS_GLOCK();
tc = afs_GetCellByHandle((void *)fh, READ_LOCK);
if (!tc) {
AFS_GUNLOCK();
return NULL;
}
afs_GetDynrootFid(&fid);
fid.Fid.Vnode = VNUM_FROM_CIDX_RW(tc->cellIndex, fh_type & 1);
fid.Fid.Unique = 1;
afs_PutCell(tc, READ_LOCK);
AFS_GUNLOCK();
break;
case AFSFH_DYN_RO_LINK:
case AFSFH_DYN_RW_LINK:
if (fh_len != 4)
return NULL;
AFS_GLOCK();
tc = afs_GetCellByHandle((void *)fh, READ_LOCK);
if (!tc) {
AFS_GUNLOCK();
return NULL;
}
afs_GetDynrootFid(&fid);
fid.Fid.Vnode = VNUM_FROM_CAIDX_RW(tc->cellIndex, fh_type & 1);
fid.Fid.Unique = 1;
afs_PutCell(tc, READ_LOCK);
AFS_GUNLOCK();
break;
case AFSFH_DYN_MOUNT:
if (fh_len != 5)
return NULL;
AFS_GLOCK();
tc = afs_GetCellByHandle((void *)fh, READ_LOCK);
if (!tc) {
AFS_GUNLOCK();
return NULL;
}
afs_GetDynrootFid(&fid);
fid.Fid.Vnode = VNUM_FROM_TYPEID(VN_TYPE_MOUNT,
tc->cellIndex << 2);
fid.Fid.Unique = ntohl(fh[4]);
afs_PutCell(tc, READ_LOCK);
AFS_GUNLOCK();
break;
case AFSFH_DYN_SYMLINK:
/* XXX parse dynroot symlink filehandle */
/* break; */
default:
return NULL;
}
#if defined(NEW_EXPORT_OPS)
result = afs_export_get_dentry(sb, &fid);
#else
result = sb->s_export_op->find_exported_dentry(sb, &fid, 0,
acceptable, context);
#endif
#ifdef OSI_EXPORT_DEBUG
if (!result) {
printk("afs: decode_fh(0x%08x/%d/%d.%d): no dentry\n",
fid.Cell, fid.Fid.Volume,
fid.Fid.Vnode, fid.Fid.Unique);
} else if (IS_ERR(result)) {
printk("afs: decode_fh(0x%08x/%d/%d.%d): error %ld\n",
fid.Cell, fid.Fid.Volume,
fid.Fid.Vnode, fid.Fid.Unique, PTR_ERR(result));
}
#endif
return result;
}
/*
* Given a FID, obtain or construct a dentry, or return an error.
* This should be called with the BKL and AFS_GLOCK held.
*/
static struct dentry *get_dentry_from_fid(cred_t *credp, struct VenusFid *afid)
{
struct vrequest *treq = NULL;
struct vcache *vcp;
struct vattr *vattr = NULL;
struct inode *ip;
struct dentry *dp;
afs_int32 code;
code = afs_CreateAttr(&vattr);
if (code) {
return ERR_PTR(-afs_CheckCode(code, NULL, 104));
}
code = afs_CreateReq(&treq, credp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): afs_CreateReq: %d\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique,
code);
#endif
afs_DestroyAttr(vattr);
return ERR_PTR(-afs_CheckCode(code, NULL, 101));
}
vcp = afs_GetVCache(afid, treq, NULL, NULL);
if (vcp == NULL) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): no vcache\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique);
#endif
afs_DestroyReq(treq);
afs_DestroyAttr(vattr);
return NULL;
}
/*
* Now, it might be that we just caused a directory vnode to
* spring into existence, in which case its parent FID is unset.
* We need to do something about that, but only because we care
* in our own get_parent(), below -- the common code never looks
* at parentVnode on directories, except for VIOCGETVCXSTATUS.
* So, if this fails, we don't really care very much.
*/
if (vType(vcp) == VDIR && vcp->mvstat != AFS_MVSTAT_ROOT && !vcp->f.parent.vnode)
update_dir_parent(treq, vcp);
/*
* If this is a volume root directory and fakestat is enabled,
* we might need to replace the directory by a mount point.
*/
code = UnEvalFakeStat(treq, &vcp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): UnEvalFakeStat: %d\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique,
code);
#endif
afs_PutVCache(vcp);
code = afs_CheckCode(code, treq, 103);
afs_DestroyReq(treq);
afs_DestroyAttr(vattr);
return ERR_PTR(-code);
}
ip = AFSTOV(vcp);
afs_getattr(vcp, vattr, credp);
afs_fill_inode(ip, vattr);
/* d_alloc_anon might block, so we shouldn't hold the glock */
AFS_GUNLOCK();
dp = d_alloc_anon(ip);
AFS_GLOCK();
if (!dp) {
iput(ip);
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_dentry_from_fid(0x%08x/%d/%d.%d): out of memory\n",
afid->Cell, afid->Fid.Volume, afid->Fid.Vnode, afid->Fid.Unique);
#endif
afs_DestroyReq(treq);
afs_DestroyAttr(vattr);
return ERR_PTR(-ENOMEM);
}
dp->d_op = &afs_dentry_operations;
afs_DestroyReq(treq);
afs_DestroyAttr(vattr);
return dp;
}
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;
}
/* 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;
}
/*
* This is almost exactly like the PFlush() routine in afs_pioctl.c,
* but that routine is static. We are about to change a file from
* normal caching to bypass it's caching. Therefore, we want to
* free up any cache space in use by the file, and throw out any
* existing VM pages for the file. We keep track of the number of
* times we go back and forth from caching to bypass.
*/
void
afs_TransitionToBypass(struct vcache *avc,
afs_ucred_t *acred, int aflags)
{
afs_int32 code;
struct vrequest treq;
int setDesire = 0;
int setManual = 0;
if (!avc)
return;
if (aflags & TRANSChangeDesiredBit)
setDesire = 1;
if (aflags & TRANSSetManualBit)
setManual = 1;
#ifdef AFS_BOZONLOCK_ENV
afs_BozonLock(&avc->pvnLock, avc); /* Since afs_TryToSmush will do a pvn_vptrunc */
#else
AFS_GLOCK();
#endif
ObtainWriteLock(&avc->lock, 925);
/*
* Someone may have beat us to doing the transition - we had no lock
* when we checked the flag earlier. No cause to panic, just return.
*/
if (avc->cachingStates & FCSBypass)
goto done;
/* If we never cached this, just change state */
if (setDesire && (!(avc->cachingStates & FCSBypass))) {
avc->cachingStates |= FCSBypass;
goto done;
}
/* cg2v, try to store any chunks not written 20071204 */
if (avc->execsOrWriters > 0) {
code = afs_InitReq(&treq, acred);
if (!code)
code = afs_StoreAllSegments(avc, &treq, AFS_SYNC | AFS_LASTSTORE);
}
#if 0
/* also cg2v, don't dequeue the callback */
ObtainWriteLock(&afs_xcbhash, 956);
afs_DequeueCallback(avc);
ReleaseWriteLock(&afs_xcbhash);
#endif
avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat */
/* now find the disk cache entries */
afs_TryToSmush(avc, acred, 1);
osi_dnlc_purgedp(avc);
if (avc->linkData && !(avc->f.states & CCore)) {
afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
avc->linkData = NULL;
}
avc->cachingStates |= FCSBypass; /* Set the bypass flag */
if(setDesire)
avc->cachingStates |= FCSDesireBypass;
if(setManual)
avc->cachingStates |= FCSManuallySet;
avc->cachingTransitions++;
done:
ReleaseWriteLock(&avc->lock);
#ifdef AFS_BOZONLOCK_ENV
afs_BozonUnlock(&avc->pvnLock, avc);
#else
AFS_GUNLOCK();
#endif
}
int
osi_NetReceive(osi_socket so, struct sockaddr_in *addr, struct iovec *dvec,
int nvecs, int *alength)
{
#ifdef AFS_DARWIN80_ENV
socket_t asocket = (socket_t)so;
struct msghdr msg;
struct sockaddr_storage ss;
int rlen;
mbuf_t m;
#else
struct socket *asocket = (struct socket *)so;
struct uio u;
#endif
int i;
struct iovec iov[RX_MAXIOVECS];
struct sockaddr *sa = NULL;
int code;
size_t resid;
int haveGlock = ISAFS_GLOCK();
/*AFS_STATCNT(osi_NetReceive); */
if (nvecs > RX_MAXIOVECS)
osi_Panic("osi_NetReceive: %d: Too many iovecs.\n", nvecs);
for (i = 0; i < nvecs; i++)
iov[i] = dvec[i];
if ((afs_termState == AFSOP_STOP_RXK_LISTENER) ||
(afs_termState == AFSOP_STOP_COMPLETE))
return -1;
if (haveGlock)
AFS_GUNLOCK();
#if defined(KERNEL_FUNNEL)
thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
#endif
#ifdef AFS_DARWIN80_ENV
resid = *alength;
memset(&msg, 0, sizeof(struct msghdr));
msg.msg_name = &ss;
msg.msg_namelen = sizeof(struct sockaddr_storage);
sa =(struct sockaddr *) &ss;
code = sock_receivembuf(asocket, &msg, &m, 0, alength);
if (!code) {
size_t offset=0,sz;
resid = *alength;
for (i=0;i<nvecs && resid;i++) {
sz=MIN(resid, iov[i].iov_len);
code = mbuf_copydata(m, offset, sz, iov[i].iov_base);
if (code)
break;
resid-=sz;
offset+=sz;
}
}
mbuf_freem(m);
#else
u.uio_iov = &iov[0];
u.uio_iovcnt = nvecs;
u.uio_offset = 0;
u.uio_resid = *alength;
u.uio_segflg = UIO_SYSSPACE;
u.uio_rw = UIO_READ;
u.uio_procp = NULL;
code = soreceive(asocket, &sa, &u, NULL, NULL, NULL);
resid = u.uio_resid;
#endif
#if defined(KERNEL_FUNNEL)
thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
#endif
if (haveGlock)
AFS_GLOCK();
if (code)
return code;
*alength -= resid;
if (sa) {
if (sa->sa_family == AF_INET) {
if (addr)
*addr = *(struct sockaddr_in *)sa;
} else
printf("Unknown socket family %d in NetReceive\n", sa->sa_family);
#ifndef AFS_DARWIN80_ENV
FREE(sa, M_SONAME);
#endif
}
return code;
}
static struct dentry *afs_export_get_parent(struct dentry *child)
{
struct VenusFid tfid;
struct vrequest *treq = NULL;
struct cell *tcell;
struct vcache *vcp;
struct dentry *dp = NULL;
cred_t *credp;
afs_uint32 cellidx;
int code;
if (!child->d_inode) {
/* can't find the parent of a negative dentry */
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_parent(%s): no inode\n",
child->d_name.name ? (char *)child->d_name.name : "?");
#endif
return ERR_PTR(-EIO);
}
credp = crref();
AFS_GLOCK();
vcp = VTOAFS(child->d_inode);
if (afs_IsDynrootMount(vcp)) {
/* the dynmount directory; parent is always the AFS root */
tfid = afs_globalVp->f.fid;
} else if (afs_IsDynrootAny(vcp) &&
VNUM_TO_VNTYPE(vcp->f.fid.Fid.Vnode) == VN_TYPE_MOUNT) {
/* a mount point in the dynmount directory */
afs_GetDynrootMountFid(&tfid);
} else if (vcp->mvstat == AFS_MVSTAT_ROOT) {
/* volume root */
ObtainReadLock(&vcp->lock);
if (vcp->mvid.parent && vcp->mvid.parent->Fid.Volume) {
tfid = *vcp->mvid.parent;
ReleaseReadLock(&vcp->lock);
} else {
ReleaseReadLock(&vcp->lock);
tcell = afs_GetCell(vcp->f.fid.Cell, READ_LOCK);
if (!tcell) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_parent(0x%08x/%d/%d.%d): no cell\n",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
dp = ERR_PTR(-ENOENT);
goto done;
}
cellidx = tcell->cellIndex;
afs_PutCell(tcell, READ_LOCK);
afs_GetDynrootMountFid(&tfid);
tfid.Fid.Vnode = VNUM_FROM_TYPEID(VN_TYPE_MOUNT, cellidx << 2);
tfid.Fid.Unique = vcp->f.fid.Fid.Volume;
}
} else {
/* any other vnode */
if (vType(vcp) == VDIR && !vcp->f.parent.vnode && vcp->mvstat != AFS_MVSTAT_MTPT) {
code = afs_CreateReq(&treq, credp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_parent(0x%08x/%d/%d.%d): afs_CreateReq: %d\n",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
dp = ERR_PTR(-ENOENT);
goto done;
} else {
code = update_dir_parent(treq, vcp);
if (code) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_parent(0x%08x/%d/%d.%d): update_dir_parent: %d\n",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique, code);
#endif
dp = ERR_PTR(-ENOENT);
goto done;
}
}
}
tfid.Cell = vcp->f.fid.Cell;
tfid.Fid.Volume = vcp->f.fid.Fid.Volume;
tfid.Fid.Vnode = vcp->f.parent.vnode;
tfid.Fid.Unique = vcp->f.parent.unique;
}
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_parent(0x%08x/%d/%d.%d): => 0x%08x/%d/%d.%d\n",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique,
tfid.Cell, tfid.Fid.Volume, tfid.Fid.Vnode, tfid.Fid.Unique);
#endif
dp = get_dentry_from_fid(credp, &tfid);
if (!dp) {
#ifdef OSI_EXPORT_DEBUG
printk("afs: get_parent(0x%08x/%d/%d.%d): no dentry\n",
vcp->f.fid.Cell, vcp->f.fid.Fid.Volume,
vcp->f.fid.Fid.Vnode, vcp->f.fid.Fid.Unique);
#endif
dp = ERR_PTR(-ENOENT);
}
done:
afs_DestroyReq(treq);
AFS_GUNLOCK();
crfree(credp);
return dp;
}
int
osi_NetSend(osi_socket so, struct sockaddr_in *addr, struct iovec *dvec,
int nvecs, afs_int32 alength, int istack)
{
#ifdef AFS_DARWIN80_ENV
socket_t asocket = (socket_t)so;
struct msghdr msg;
size_t slen;
#else
struct socket *asocket = (struct socket *)so;
struct uio u;
#endif
afs_int32 code;
int i;
struct iovec iov[RX_MAXIOVECS];
int haveGlock = ISAFS_GLOCK();
AFS_STATCNT(osi_NetSend);
if (nvecs > RX_MAXIOVECS)
osi_Panic("osi_NetSend: %d: Too many iovecs.\n", nvecs);
for (i = 0; i < nvecs; i++)
iov[i] = dvec[i];
addr->sin_len = sizeof(struct sockaddr_in);
if ((afs_termState == AFSOP_STOP_RXK_LISTENER) ||
(afs_termState == AFSOP_STOP_COMPLETE))
return -1;
if (haveGlock)
AFS_GUNLOCK();
#if defined(KERNEL_FUNNEL)
thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
#endif
#ifdef AFS_DARWIN80_ENV
memset(&msg, 0, sizeof(struct msghdr));
msg.msg_name = addr;
msg.msg_namelen = ((struct sockaddr *)addr)->sa_len;
msg.msg_iov = &iov[0];
msg.msg_iovlen = nvecs;
code = sock_send(asocket, &msg, 0, &slen);
#else
u.uio_iov = &iov[0];
u.uio_iovcnt = nvecs;
u.uio_offset = 0;
u.uio_resid = alength;
u.uio_segflg = UIO_SYSSPACE;
u.uio_rw = UIO_WRITE;
u.uio_procp = NULL;
code = sosend(asocket, (struct sockaddr *)addr, &u, NULL, NULL, 0);
#endif
#if defined(KERNEL_FUNNEL)
thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
#endif
if (haveGlock)
AFS_GLOCK();
return code;
}
afs_omount(struct mount *mp, char *path, caddr_t data, struct nameidata *ndp,
struct thread *p)
#endif
{
/* ndp contains the mounted-from device. Just ignore it.
* we also don't care about our thread struct. */
size_t size;
if (mp->mnt_flag & MNT_UPDATE)
return EINVAL;
AFS_GLOCK();
AFS_STATCNT(afs_mount);
if (afs_globalVFS) { /* Don't allow remounts. */
AFS_GUNLOCK();
return EBUSY;
}
afs_globalVFS = mp;
mp->vfs_bsize = 8192;
vfs_getnewfsid(mp);
/*
* This is kind of ugly, as the interlock has grown to encompass
* more fields over time and there's not a good way to group the
* code without duplication.
*/
#ifdef AFS_FBSD62_ENV
MNT_ILOCK(mp);
#endif
mp->mnt_flag &= ~MNT_LOCAL;
#if defined(AFS_FBSD61_ENV) && !defined(AFS_FBSD62_ENV)
MNT_ILOCK(mp);
#endif
#if __FreeBSD_version < 1000021
mp->mnt_kern_flag |= MNTK_MPSAFE; /* solid steel */
#endif
#ifndef AFS_FBSD61_ENV
MNT_ILOCK(mp);
#endif
/*
* XXX mnt_stat "is considered stable as long as a ref is held".
* We should check that we hold the only ref.
*/
mp->mnt_stat.f_iosize = 8192;
if (path != NULL)
copyinstr(path, mp->mnt_stat.f_mntonname, MNAMELEN - 1, &size);
else
bcopy("/afs", mp->mnt_stat.f_mntonname, size = 4);
memset(mp->mnt_stat.f_mntonname + size, 0, MNAMELEN - size);
memset(mp->mnt_stat.f_mntfromname, 0, MNAMELEN);
strcpy(mp->mnt_stat.f_mntfromname, "AFS");
/* null terminated string "AFS" will fit, just leave it be. */
strcpy(mp->mnt_stat.f_fstypename, "afs");
MNT_IUNLOCK(mp);
AFS_GUNLOCK();
#ifdef AFS_FBSD80_ENV
afs_statfs(mp, &mp->mnt_stat);
#else
afs_statfs(mp, &mp->mnt_stat, p);
#endif
return 0;
}
/* set the real time */
void
afs_osi_SetTime(osi_timeval_t * atv)
{
#if defined(AFS_AIX32_ENV)
struct timestruc_t t;
t.tv_sec = atv->tv_sec;
t.tv_nsec = atv->tv_usec * 1000;
ksettimer(&t); /* Was -> settimer(TIMEOFDAY, &t); */
#elif defined(AFS_SUN55_ENV)
stime(atv->tv_sec);
#elif defined(AFS_SUN5_ENV)
/*
* To get more than second resolution we can use adjtime. The problem
* is that the usecs from the server are wrong (by now) so it isn't
* worth complicating the following code.
*/
struct stimea {
time_t time;
} sta;
sta.time = atv->tv_sec;
stime(&sta, NULL);
#elif defined(AFS_SGI_ENV)
struct stimea {
sysarg_t time;
} sta;
AFS_GUNLOCK();
sta.time = atv->tv_sec;
stime(&sta);
AFS_GLOCK();
#elif defined(AFS_DARWIN_ENV)
#ifndef AFS_DARWIN80_ENV
AFS_GUNLOCK();
setthetime(atv);
AFS_GLOCK();
#endif
#else
/* stolen from kern_time.c */
#ifndef AFS_AUX_ENV
boottime.tv_sec += atv->tv_sec - time.tv_sec;
#endif
#ifdef AFS_HPUX_ENV
{
#if !defined(AFS_HPUX1122_ENV)
/* drop the setting of the clock for now. spl7 is not
* known on hpux11.22
*/
register ulong_t s;
struct timeval t;
t.tv_sec = atv->tv_sec;
t.tv_usec = atv->tv_usec;
s = spl7();
time = t;
(void)splx(s);
resettodr(atv);
#endif
}
#else
{
register int s;
s = splclock();
time = *atv;
(void)splx(s);
}
resettodr();
#endif
#ifdef AFS_AUX_ENV
logtchg(atv->tv_sec);
#endif
#endif /* AFS_DARWIN_ENV */
AFS_STATCNT(osi_SetTime);
}
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;
}
/**
* Reset volume name to volume id mapping cache.
* @param flags
*/
void
afs_CheckVolumeNames(int flags)
{
afs_int32 i, j;
struct volume *tv;
unsigned int now;
struct vcache *tvc;
afs_int32 *volumeID, *cellID, vsize, nvols;
#ifdef AFS_DARWIN80_ENV
vnode_t tvp;
#endif
AFS_STATCNT(afs_CheckVolumeNames);
nvols = 0;
volumeID = cellID = NULL;
vsize = 0;
ObtainReadLock(&afs_xvolume);
if (flags & AFS_VOLCHECK_EXPIRED) {
/*
* allocate space to hold the volumeIDs and cellIDs, only if
* we will be invalidating the mountpoints later on
*/
for (i = 0; i < NVOLS; i++)
for (tv = afs_volumes[i]; tv; tv = tv->next)
++vsize;
volumeID = (afs_int32 *) afs_osi_Alloc(2 * vsize * sizeof(*volumeID));
cellID = (volumeID) ? volumeID + vsize : 0;
}
now = osi_Time();
for (i = 0; i < NVOLS; i++) {
for (tv = afs_volumes[i]; tv; tv = tv->next) {
if (flags & AFS_VOLCHECK_EXPIRED) {
if (((tv->expireTime < (now + 10)) && (tv->states & VRO))
|| (flags & AFS_VOLCHECK_FORCE)) {
afs_ResetVolumeInfo(tv); /* also resets status */
if (volumeID) {
volumeID[nvols] = tv->volume;
cellID[nvols] = tv->cell;
}
++nvols;
continue;
}
}
/* ??? */
if (flags & (AFS_VOLCHECK_BUSY | AFS_VOLCHECK_FORCE)) {
for (j = 0; j < MAXHOSTS; j++)
tv->status[j] = not_busy;
}
}
}
ReleaseReadLock(&afs_xvolume);
/* next ensure all mt points are re-evaluated */
if (nvols || (flags & (AFS_VOLCHECK_FORCE | AFS_VOLCHECK_MTPTS))) {
loop:
ObtainReadLock(&afs_xvcache);
for (i = 0; i < VCSIZE; i++) {
for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
/* if the volume of "mvid" of the vcache entry is among the
* ones we found earlier, then we re-evaluate it. Also, if the
* force bit is set or we explicitly asked to reevaluate the
* mt-pts, we clean the cmvalid bit */
if ((flags & (AFS_VOLCHECK_FORCE | AFS_VOLCHECK_MTPTS))
|| (tvc->mvid
&& inVolList(tvc->mvid, nvols, volumeID, cellID)))
tvc->f.states &= ~CMValid;
/* If the volume that this file belongs to was reset earlier,
* then we should remove its callback.
* Again, if forced, always do it.
*/
if ((tvc->f.states & CRO)
&& (inVolList(&tvc->f.fid, nvols, volumeID, cellID)
|| (flags & AFS_VOLCHECK_FORCE))) {
if (tvc->f.states & CVInit) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop;
}
#ifdef AFS_DARWIN80_ENV
if (tvc->f.states & CDeadVnode) {
ReleaseReadLock(&afs_xvcache);
afs_osi_Sleep(&tvc->f.states);
goto loop;
}
tvp = AFSTOV(tvc);
if (vnode_get(tvp))
continue;
if (vnode_ref(tvp)) {
AFS_GUNLOCK();
/* AFSTOV(tvc) may be NULL */
vnode_put(tvp);
AFS_GLOCK();
continue;
}
#else
AFS_FAST_HOLD(tvc);
#endif
ReleaseReadLock(&afs_xvcache);
ObtainWriteLock(&afs_xcbhash, 485);
/* LOCKXXX: We aren't holding tvc write lock? */
afs_DequeueCallback(tvc);
tvc->f.states &= ~CStatd;
ReleaseWriteLock(&afs_xcbhash);
if (tvc->f.fid.Fid.Vnode & 1 || (vType(tvc) == VDIR))
osi_dnlc_purgedp(tvc);
#ifdef AFS_DARWIN80_ENV
vnode_put(AFSTOV(tvc));
/* our tvc ptr is still good until now */
AFS_FAST_RELE(tvc);
ObtainReadLock(&afs_xvcache);
#else
ObtainReadLock(&afs_xvcache);
/* our tvc ptr is still good until now */
AFS_FAST_RELE(tvc);
#endif
}
}
}
osi_dnlc_purge(); /* definitely overkill, but it's safer this way. */
ReleaseReadLock(&afs_xvcache);
}
if (volumeID)
afs_osi_Free(volumeID, 2 * vsize * sizeof(*volumeID));
} /*afs_CheckVolumeNames */
/* works like PFlushVolumeData */
void
darwin_notify_perms(struct unixuser *auser, int event)
{
int i;
struct afs_q *tq, *uq = NULL;
struct vcache *tvc, *hnext;
int isglock = ISAFS_GLOCK();
struct vnode *vp;
struct vnode_attr va;
int isctxtowner = 0;
if (!afs_darwin_fsevents)
return;
VATTR_INIT(&va);
VATTR_SET(&va, va_mode, 0777);
if (event & UTokensObtained)
VATTR_SET(&va, va_uid, auser->uid);
else
VATTR_SET(&va, va_uid, -2); /* nobody */
if (!isglock)
AFS_GLOCK();
if (!(vfs_context_owner == current_thread())) {
get_vfs_context();
isctxtowner = 1;
}
loop:
ObtainReadLock(&afs_xvcache);
for (i = 0; i < VCSIZE; i++) {
for (tq = afs_vhashTV[i].prev; tq != &afs_vhashTV[i]; tq = uq) {
uq = QPrev(tq);
tvc = QTOVH(tq);
if (tvc->f.states & CDeadVnode) {
/* we can afford to be best-effort */
continue;
}
/* no per-file acls, so only notify on directories */
if (!(vp = AFSTOV(tvc)) || !vnode_isdir(AFSTOV(tvc)))
continue;
/* dynroot object. no callbacks. anonymous ACL. just no. */
if (afs_IsDynrootFid(&tvc->f.fid))
continue;
/* no fake fsevents on mount point sources. leaks refs */
if (tvc->mvstat == 1)
continue;
/* if it's being reclaimed, just pass */
if (vnode_get(vp))
continue;
if (vnode_ref(vp)) {
AFS_GUNLOCK();
vnode_put(vp);
AFS_GLOCK();
continue;
}
ReleaseReadLock(&afs_xvcache);
/* Avoid potentially re-entering on this lock */
if (0 == NBObtainWriteLock(&tvc->lock, 234)) {
tvc->f.states |= CEvent;
AFS_GUNLOCK();
vnode_setattr(vp, &va, afs_osi_ctxtp);
tvc->f.states &= ~CEvent;
vnode_put(vp);
AFS_GLOCK();
ReleaseWriteLock(&tvc->lock);
}
ObtainReadLock(&afs_xvcache);
uq = QPrev(tq);
/* our tvc ptr is still good until now */
AFS_FAST_RELE(tvc);
}
}
ReleaseReadLock(&afs_xvcache);
if (isctxtowner)
put_vfs_context();
if (!isglock)
AFS_GUNLOCK();
}
int
afs_fill_super(struct super_block *sb, void *data, int silent)
{
int code = 0;
#if defined(HAVE_LINUX_BDI_INIT)
int bdi_init_done = 0;
#endif
AFS_GLOCK();
if (afs_was_mounted) {
printf
("You must reload the AFS kernel extensions before remounting AFS.\n");
AFS_GUNLOCK();
return -EINVAL;
}
afs_was_mounted = 1;
/* Set basics of super_block */
__module_get(THIS_MODULE);
afs_globalVFS = sb;
sb->s_flags |= MS_NOATIME;
sb->s_blocksize = 1024;
sb->s_blocksize_bits = 10;
sb->s_magic = AFS_VFSMAGIC;
sb->s_op = &afs_sops; /* Super block (vfs) ops */
#if defined(STRUCT_SUPER_BLOCK_HAS_S_D_OP)
sb->s_d_op = &afs_dentry_operations;
#endif
/* used for inodes backing_dev_info field, also */
afs_backing_dev_info = kzalloc(sizeof(struct backing_dev_info), GFP_NOFS);
#if defined(HAVE_LINUX_BDI_INIT)
code = bdi_init(afs_backing_dev_info);
if (code)
goto out;
bdi_init_done = 1;
#endif
#if defined(STRUCT_BACKING_DEV_INFO_HAS_NAME)
afs_backing_dev_info->name = "openafs";
#endif
afs_backing_dev_info->ra_pages = 32;
#if defined (STRUCT_SUPER_BLOCK_HAS_S_BDI)
sb->s_bdi = afs_backing_dev_info;
/* The name specified here will appear in the flushing thread name - flush-afs */
bdi_register(afs_backing_dev_info, NULL, "afs");
#endif
#if !defined(AFS_NONFSTRANS)
sb->s_export_op = &afs_export_ops;
#endif
#if defined(MAX_NON_LFS)
#ifdef AFS_64BIT_CLIENT
#if !defined(MAX_LFS_FILESIZE)
#if BITS_PER_LONG==32
#define MAX_LFS_FILESIZE (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
#elif BITS_PER_LONG==64
#define MAX_LFS_FILESIZE 0x7fffffffffffffff
#endif
#endif
sb->s_maxbytes = MAX_LFS_FILESIZE;
#else
sb->s_maxbytes = MAX_NON_LFS;
#endif
#endif
code = afs_root(sb);
out:
if (code) {
afs_globalVFS = NULL;
afs_FlushAllVCaches();
#if defined(HAVE_LINUX_BDI_INIT)
if (bdi_init_done)
bdi_destroy(afs_backing_dev_info);
#endif
kfree(afs_backing_dev_info);
module_put(THIS_MODULE);
}
AFS_GUNLOCK();
return code ? -EINVAL : 0;
}
/*
* afs_TruncateAllSegments
*
* Description:
* Truncate a cache file.
*
* Parameters:
* avc : Ptr to vcache entry to truncate.
* alen : Number of bytes to make the file.
* areq : Ptr to request structure.
*
* Environment:
* Called with avc write-locked; in VFS40 systems, pvnLock is also
* held.
*/
int
afs_TruncateAllSegments(register struct vcache *avc, afs_size_t alen,
struct vrequest *areq, afs_ucred_t *acred)
{
register struct dcache *tdc;
register afs_int32 code;
register afs_int32 index;
afs_size_t newSize;
int dcCount, dcPos;
struct dcache **tdcArray;
AFS_STATCNT(afs_TruncateAllSegments);
avc->f.m.Date = osi_Time();
afs_Trace3(afs_iclSetp, CM_TRACE_TRUNCALL, ICL_TYPE_POINTER, avc,
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(avc->f.m.Length),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(alen));
if (alen >= avc->f.m.Length) {
/*
* Special speedup since Sun's vm extends the file this way;
* we've never written to the file thus we can just set the new
* length and avoid the needless calls below.
* Also used for ftruncate calls which can extend the file.
* To completely minimize the possible extra StoreMini RPC, we really
* should keep the ExtendedPos as well and clear this flag if we
* truncate below that value before we store the file back.
*/
avc->f.states |= CExtendedFile;
avc->f.m.Length = alen;
return 0;
}
#if (defined(AFS_SUN5_ENV))
/* Zero unused portion of last page */
osi_VM_PreTruncate(avc, alen, acred);
#endif
#if (defined(AFS_SUN5_ENV))
ObtainWriteLock(&avc->vlock, 546);
avc->activeV++; /* Block new getpages */
ReleaseWriteLock(&avc->vlock);
#endif
ReleaseWriteLock(&avc->lock);
AFS_GUNLOCK();
/* Flush pages beyond end-of-file. */
osi_VM_Truncate(avc, alen, acred);
AFS_GLOCK();
ObtainWriteLock(&avc->lock, 79);
avc->f.m.Length = alen;
if (alen < avc->f.truncPos)
avc->f.truncPos = alen;
code = DVHash(&avc->f.fid);
/* block out others from screwing with this table */
ObtainWriteLock(&afs_xdcache, 287);
dcCount = 0;
for (index = afs_dvhashTbl[code]; index != NULLIDX;) {
if (afs_indexUnique[index] == avc->f.fid.Fid.Unique) {
tdc = afs_GetDSlot(index, 0);
ReleaseReadLock(&tdc->tlock);
if (!FidCmp(&tdc->f.fid, &avc->f.fid))
dcCount++;
afs_PutDCache(tdc);
}
index = afs_dvnextTbl[index];
}
/* Now allocate space where we can save those dcache entries, and
* do a second pass over them.. Since we're holding xdcache, it
* shouldn't be changing.
*/
tdcArray = osi_Alloc(dcCount * sizeof(struct dcache *));
dcPos = 0;
for (index = afs_dvhashTbl[code]; index != NULLIDX;) {
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, and modified, we'll store it back */
if (dcPos < dcCount) {
tdcArray[dcPos++] = tdc;
} else {
afs_PutDCache(tdc);
}
} else {
afs_PutDCache(tdc);
}
}
index = afs_dvnextTbl[index];
}
ReleaseWriteLock(&afs_xdcache);
/* Now we loop over the array of dcache entries and truncate them */
for (index = 0; index < dcPos; index++) {
struct osi_file *tfile;
tdc = tdcArray[index];
newSize = alen - AFS_CHUNKTOBASE(tdc->f.chunk);
if (newSize < 0)
newSize = 0;
ObtainSharedLock(&tdc->lock, 672);
if (newSize < tdc->f.chunkBytes && newSize < MAX_AFS_UINT32) {
UpgradeSToWLock(&tdc->lock, 673);
tfile = afs_CFileOpen(&tdc->f.inode);
afs_CFileTruncate(tfile, (afs_int32)newSize);
afs_CFileClose(tfile);
afs_AdjustSize(tdc, (afs_int32)newSize);
if (alen < tdc->validPos) {
if (alen < AFS_CHUNKTOBASE(tdc->f.chunk))
tdc->validPos = 0;
else
tdc->validPos = alen;
}
ConvertWToSLock(&tdc->lock);
}
ReleaseSharedLock(&tdc->lock);
afs_PutDCache(tdc);
}
osi_Free(tdcArray, dcCount * sizeof(struct dcache *));
#if (defined(AFS_SUN5_ENV))
ObtainWriteLock(&avc->vlock, 547);
if (--avc->activeV == 0 && (avc->vstates & VRevokeWait)) {
avc->vstates &= ~VRevokeWait;
afs_osi_Wakeup((char *)&avc->vstates);
}
ReleaseWriteLock(&avc->vlock);
#endif
return 0;
}
int
osi_UFSTruncate(struct osi_file *afile, afs_int32 asize)
{
afs_int32 code;
struct osi_stat tstat;
struct iattr newattrs;
struct inode *inode = OSIFILE_INODE(afile);
AFS_STATCNT(osi_Truncate);
/* This routine only shrinks files, and most systems
* have very slow truncates, even when the file is already
* small enough. Check now and save some time.
*/
code = afs_osi_Stat(afile, &tstat);
if (code || tstat.size <= asize)
return code;
ObtainWriteLock(&afs_xosi, 321);
AFS_GUNLOCK();
#ifdef STRUCT_INODE_HAS_I_ALLOC_SEM
down_write(&inode->i_alloc_sem);
#endif
#ifdef STRUCT_INODE_HAS_I_MUTEX
mutex_lock(&inode->i_mutex);
#else
down(&inode->i_sem);
#endif
newattrs.ia_size = asize;
newattrs.ia_valid = ATTR_SIZE | ATTR_CTIME;
#if defined(AFS_LINUX24_ENV)
newattrs.ia_ctime = CURRENT_TIME;
/* avoid notify_change() since it wants to update dentry->d_parent */
lock_kernel();
code = inode_change_ok(inode, &newattrs);
if (!code) {
#ifdef INODE_SETATTR_NOT_VOID
code = inode_setattr(inode, &newattrs);
#else
inode_setattr(inode, &newattrs);
#endif
}
unlock_kernel();
if (!code)
truncate_inode_pages(&inode->i_data, asize);
#else
i_size_write(inode, asize);
if (inode->i_sb->s_op && inode->i_sb->s_op->notify_change) {
code = inode->i_sb->s_op->notify_change(&afile->dentry, &newattrs);
}
if (!code) {
truncate_inode_pages(inode, asize);
if (inode->i_op && inode->i_op->truncate)
inode->i_op->truncate(inode);
}
#endif
code = -code;
#ifdef STRUCT_INODE_HAS_I_MUTEX
mutex_unlock(&inode->i_mutex);
#else
up(&inode->i_sem);
#endif
#ifdef STRUCT_INODE_HAS_I_ALLOC_SEM
up_write(&inode->i_alloc_sem);
#endif
AFS_GLOCK();
ReleaseWriteLock(&afs_xosi);
return code;
}
void *
osi_UFSOpen(afs_dcache_id_t *ainode)
{
struct vnode *vp;
struct vattr va;
struct osi_file *afile = NULL;
extern int cacheDiskType;
afs_int32 code = 0;
int dummy;
char fname[1024];
struct osi_stat tstat;
AFS_STATCNT(osi_UFSOpen);
if (cacheDiskType != AFS_FCACHE_TYPE_UFS) {
osi_Panic("UFSOpen called for non-UFS cache\n");
}
if (!afs_osicred_initialized) {
memset(&afs_osi_cred, 0, sizeof(afs_ucred_t));
afs_osi_cred.cr_ref++;
#ifndef AFS_DARWIN110_ENV
afs_osi_cred.cr_ngroups = 1;
#endif
afs_osicred_initialized = 1;
}
afile = osi_AllocSmallSpace(sizeof(struct osi_file));
AFS_GUNLOCK();
#ifdef AFS_CACHE_VNODE_PATH
if (!ainode->ufs) {
osi_Panic("No cache inode\n");
}
code = vnode_open(ainode->ufs, O_RDWR, 0, 0, &vp, afs_osi_ctxtp);
#else
#ifndef AFS_DARWIN80_ENV
if (afs_CacheFSType == AFS_APPL_HFS_CACHE)
code = igetinode(afs_cacheVfsp, (dev_t) cacheDev.dev, &ainode->ufs, &vp, &va, &dummy); /* XXX hfs is broken */
else if (afs_CacheFSType == AFS_APPL_UFS_CACHE)
#endif
code =
igetinode(afs_cacheVfsp, (dev_t) cacheDev.dev, (ino_t) ainode->ufs,
&vp, &va, &dummy);
#ifndef AFS_DARWIN80_ENV
else
panic("osi_UFSOpen called before cacheops initialized\n");
#endif
#endif
AFS_GLOCK();
if (code) {
osi_FreeSmallSpace(afile);
osi_Panic("UFSOpen: igetinode failed");
}
afile->vnode = vp;
afile->offset = 0;
afile->proc = (int (*)())0;
#ifndef AFS_CACHE_VNODE_PATH
afile->size = va.va_size;
#else
code = afs_osi_Stat(afile, &tstat);
afile->size = tstat.size;
#endif
return (void *)afile;
}
int
afs_BioDaemon(afs_int32 nbiods)
{
afs_int32 code, s, pflg = 0;
label_t jmpbuf;
struct buf *bp, *bp1, *tbp1, *tbp2; /* temp pointers only */
caddr_t tmpaddr;
struct vnode *vp;
struct vcache *vcp;
char tmperr;
if (!afs_initbiod) {
/* XXX ###1 XXX */
afs_initbiod = 1;
/* pin lock, since we'll be using it in an interrupt. */
lock_alloc(&afs_asyncbuf_lock, LOCK_ALLOC_PIN, 2, 1);
simple_lock_init(&afs_asyncbuf_lock);
pin(&afs_asyncbuf, sizeof(struct buf *));
pin(&afs_asyncbuf_cv, sizeof(afs_int32));
}
/* Ignore HUP signals... */
{
sigset_t sigbits, osigbits;
/*
* add SIGHUP to the set of already masked signals
*/
SIGFILLSET(sigbits); /* allow all signals */
SIGDELSET(sigbits, SIGHUP); /* except SIGHUP */
limit_sigs(&sigbits, &osigbits); /* and already masked */
}
/* Main body starts here -- this is an intentional infinite loop, and
* should NEVER exit
*
* Now, the loop will exit if get_bioreq() returns NULL, indicating
* that we've been interrupted.
*/
while (1) {
bp = afs_get_bioreq();
if (!bp)
break; /* we were interrupted */
if (code = setjmpx(&jmpbuf)) {
/* This should not have happend, maybe a lack of resources */
AFS_GUNLOCK();
s = disable_lock(INTMAX, &afs_asyncbuf_lock);
for (bp1 = bp; bp; bp = bp1) {
if (bp1)
bp1 = (struct buf *)bp1->b_work;
bp->b_actf = 0;
bp->b_error = code;
bp->b_flags |= B_ERROR;
iodone(bp);
}
unlock_enable(s, &afs_asyncbuf_lock);
AFS_GLOCK();
continue;
}
vcp = VTOAFS(bp->b_vp);
if (bp->b_flags & B_PFSTORE) { /* XXXX */
ObtainWriteLock(&vcp->lock, 404);
if (vcp->v.v_gnode->gn_mwrcnt) {
afs_offs_t newlength =
(afs_offs_t) dbtob(bp->b_blkno) + bp->b_bcount;
if (vcp->f.m.Length < newlength) {
afs_Trace4(afs_iclSetp, CM_TRACE_SETLENGTH,
ICL_TYPE_STRING, __FILE__, ICL_TYPE_LONG,
__LINE__, ICL_TYPE_OFFSET,
ICL_HANDLE_OFFSET(vcp->f.m.Length),
ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(newlength));
vcp->f.m.Length = newlength;
}
}
ReleaseWriteLock(&vcp->lock);
}
/* If the buffer represents a protection violation, rather than
* an actual request for I/O, no special action need be taken.
*/
if (bp->b_flags & B_PFPROT) {
iodone(bp); /* Notify all users of the buffer that we're done */
clrjmpx(&jmpbuf);
continue;
}
if (DOvmlock)
ObtainWriteLock(&vcp->pvmlock, 211);
/*
* First map its data area to a region in the current address space
* by calling vm_att with the subspace identifier, and a pointer to
* the data area. vm_att returns a new data area pointer, but we
* also want to hang onto the old one.
*/
tmpaddr = bp->b_baddr;
bp->b_baddr = (caddr_t) vm_att(bp->b_xmemd.subspace_id, tmpaddr);
tmperr = afs_ustrategy(bp); /* temp variable saves offset calculation */
if (tmperr) { /* in non-error case */
bp->b_flags |= B_ERROR; /* should other flags remain set ??? */
bp->b_error = tmperr;
}
/* Unmap the buffer's data area by calling vm_det. Reset data area
* to the value that we saved above.
*/
vm_det(bp->b_baddr);
bp->b_baddr = tmpaddr;
/*
* buffer may be linked with other buffers via the b_work field.
* See also afs_gn_strategy. For each buffer in the chain (including
* bp) notify all users of the buffer that the daemon is finished
* using it by calling iodone.
* assumes iodone can modify the b_work field.
*/
for (tbp1 = bp;;) {
tbp2 = (struct buf *)tbp1->b_work;
iodone(tbp1);
if (!tbp2)
break;
tbp1 = (struct buf *)tbp2->b_work;
iodone(tbp2);
if (!tbp1)
break;
}
if (DOvmlock)
ReleaseWriteLock(&vcp->pvmlock); /* Unlock the vnode. */
clrjmpx(&jmpbuf);
} /* infinite loop (unless we're interrupted) */
} /* end of afs_BioDaemon() */
int
afs_mount(struct mount *mp, char *path, caddr_t data, struct nameidata *ndp, CTX_TYPE ctx)
#endif
{
/* ndp contains the mounted-from device. Just ignore it.
* we also don't care about our proc struct. */
size_t size;
int error;
#ifdef AFS_DARWIN80_ENV
struct vfsioattr ioattr;
/* vfs_statfs advertised as RO, but isn't */
/* new api will be needed to initialize this information (nfs needs to
set mntfromname too) */
#endif
STATFS_TYPE *mnt_stat = vfs_statfs(mp);
if (vfs_isupdate(mp))
return EINVAL;
AFS_GLOCK();
AFS_STATCNT(afs_mount);
if (data == 0 && afs_globalVFS) { /* Don't allow remounts. */
AFS_GUNLOCK();
return (EBUSY);
}
afs_globalVFS = mp;
#ifdef AFS_DARWIN80_ENV
vfs_ioattr(mp, &ioattr);
ioattr.io_devblocksize = (16 * 32768);
vfs_setioattr(mp, &ioattr);
/* f_iosize is handled in VFS_GETATTR */
#else
mp->vfs_bsize = 8192;
mp->mnt_stat.f_iosize = 8192;
#endif
vfs_getnewfsid(mp);
#ifndef AFS_DARWIN80_ENV
(void)copyinstr(path, mp->mnt_stat.f_mntonname, MNAMELEN - 1, &size);
memset(mp->mnt_stat.f_mntonname + size, 0, MNAMELEN - size);
#endif
memset(mnt_stat->f_mntfromname, 0, MNAMELEN);
if (data == 0) {
strcpy(mnt_stat->f_mntfromname, "AFS");
/* null terminated string "AFS" will fit, just leave it be. */
vfs_setfsprivate(mp, NULL);
} else {
struct VenusFid *rootFid = NULL;
struct volume *tvp;
char volName[MNAMELEN];
(void)copyinstr(data, volName, MNAMELEN - 1, &size);
memset(volName + size, 0, MNAMELEN - size);
if (volName[0] == 0) {
strcpy(mnt_stat->f_mntfromname, "AFS");
vfs_setfsprivate(mp, &afs_rootFid);
} else {
struct cell *localcell = afs_GetPrimaryCell(READ_LOCK);
if (localcell == NULL) {
AFS_GUNLOCK();
return ENODEV;
}
/* Set the volume identifier to "AFS:volume.name" */
snprintf(mnt_stat->f_mntfromname, MNAMELEN - 1, "AFS:%s",
volName);
tvp =
afs_GetVolumeByName(volName, localcell->cellNum, 1,
(struct vrequest *)0, READ_LOCK);
if (tvp) {
int volid = (tvp->roVol ? tvp->roVol : tvp->volume);
MALLOC(rootFid, struct VenusFid *, sizeof(*rootFid), M_UFSMNT,
M_WAITOK);
rootFid->Cell = localcell->cellNum;
rootFid->Fid.Volume = volid;
rootFid->Fid.Vnode = 1;
rootFid->Fid.Unique = 1;
} else {
AFS_GUNLOCK();
return ENODEV;
}
vfs_setfsprivate(mp, &rootFid);
}
}
int
osi_TryEvictVCache(struct vcache *avc, int *slept, int defersleep) {
int code;
struct dentry *dentry;
struct inode *inode = AFSTOV(avc);
struct list_head *cur, *head;
/* First, see if we can evict the inode from the dcache */
if (defersleep && avc != afs_globalVp && VREFCOUNT(avc) > 1 && avc->opens == 0) {
*slept = 1;
ReleaseWriteLock(&afs_xvcache);
AFS_GUNLOCK();
#if defined(HAVE_DCACHE_LOCK)
spin_lock(&dcache_lock);
head = &inode->i_dentry;
restart:
cur = head;
while ((cur = cur->next) != head) {
dentry = list_entry(cur, struct dentry, d_alias);
if (d_unhashed(dentry))
continue;
dget_locked(dentry);
spin_unlock(&dcache_lock);
if (d_invalidate(dentry) == -EBUSY) {
dput(dentry);
/* perhaps lock and try to continue? (use cur as head?) */
goto inuse;
}
dput(dentry);
spin_lock(&dcache_lock);
goto restart;
}
spin_unlock(&dcache_lock);
#else /* HAVE_DCACHE_LOCK */
spin_lock(&inode->i_lock);
head = &inode->i_dentry;
restart:
cur = head;
while ((cur = cur->next) != head) {
dentry = list_entry(cur, struct dentry, d_alias);
spin_lock(&dentry->d_lock);
if (d_unhashed(dentry)) {
spin_unlock(&dentry->d_lock);
continue;
}
spin_unlock(&dentry->d_lock);
dget(dentry);
spin_unlock(&inode->i_lock);
if (d_invalidate(dentry) == -EBUSY) {
dput(dentry);
/* perhaps lock and try to continue? (use cur as head?) */
goto inuse;
}
dput(dentry);
spin_lock(&inode->i_lock);
goto restart;
}
spin_unlock(&inode->i_lock);
#endif /* HAVE_DCACHE_LOCK */
inuse:
AFS_GLOCK();
ObtainWriteLock(&afs_xvcache, 733);
}
/* called with the GLOCK held */
int
afs_syscall_pioctl(char *path, unsigned int com, caddr_t cmarg, int follow)
{
cred_t *credp = crref(); /* don't free until done! */
struct afs_ioctl data;
struct clientcred ccred;
struct rmtbulk idata, odata;
short in_size, out_size;
afs_int32 code = 0, pag, err;
gid_t g0, g1;
char *abspath, *pathbuf = 0;
AFS_STATCNT(afs_syscall_pioctl);
if (follow)
follow = 1; /* compat. with old venus */
code = copyin_afs_ioctl(cmarg, &data);
if (code) goto out;
if ((com & 0xff) == 90) {
/* PSetClientContext, in any space */
code = EINVAL;
goto out;
}
/* Special handling for a few pioctls */
switch (com & 0xffff) {
case (0x5600 | 3): /* VIOCSETTOK */
code = afspag_PSetTokens(data.in, data.in_size, &credp);
if (code) goto out;
break;
case (0x5600 | 9): /* VIOCUNLOG */
case (0x5600 | 21): /* VIOCUNPAG */
code = afspag_PUnlog(data.in, data.in_size, &credp);
if (code) goto out;
break;
case (0x5600 | 38): /* VIOC_AFS_SYSNAME */
code = afspag_PSetSysName(data.in, data.in_size, &credp);
if (code) goto out;
break;
}
/* Set up credentials */
memset(&ccred, 0, sizeof(ccred));
pag = PagInCred(credp);
ccred.uid = afs_cr_uid(credp);
if (pag != NOPAG) {
afs_get_groups_from_pag(pag, &g0, &g1);
ccred.group0 = g0;
ccred.group1 = g1;
}
/*
* Copy the path and convert to absolute, if one was given.
* NB: We can only use osI_AllocLargeSpace here as long as
* RMTSYS_MAXPATHLEN is less than AFS_LRALLOCSIZ.
*/
if (path) {
pathbuf = osi_AllocLargeSpace(RMTSYS_MAXPATHLEN);
if (!pathbuf) {
code = ENOMEM;
goto out;
}
code = osi_abspath(path, pathbuf, RMTSYS_MAXPATHLEN, 0, &abspath);
if (code)
goto out_path;
} else {
abspath = NIL_PATHP;
}
/* Allocate, copy, and convert incoming data */
idata.rmtbulk_len = in_size = data.in_size;
if (in_size < 0 || in_size > MAXBUFFERLEN) {
code = EINVAL;
goto out_path;
}
if (in_size > AFS_LRALLOCSIZ)
idata.rmtbulk_val = osi_Alloc(in_size);
else
idata.rmtbulk_val = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
if (!idata.rmtbulk_val) {
code = ENOMEM;
goto out_path;
}
if (in_size) {
AFS_COPYIN(data.in, idata.rmtbulk_val, in_size, code);
if (code)
goto out_idata;
inparam_conversion(com, idata.rmtbulk_val, in_size, 0);
}
/* Allocate space for outgoing data */
odata.rmtbulk_len = out_size = data.out_size;
if (out_size < 0 || out_size > MAXBUFFERLEN) {
code = EINVAL;
goto out_idata;
}
if (out_size > AFS_LRALLOCSIZ)
odata.rmtbulk_val = osi_Alloc(out_size);
else
odata.rmtbulk_val = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
if (!odata.rmtbulk_val) {
code = ENOMEM;
goto out_idata;
}
AFS_GUNLOCK();
code = RMTSYS_Pioctl(rmtsys_conn, &ccred, abspath, com, follow,
&idata, &odata, &err);
AFS_GLOCK();
if (code)
goto out_odata;
/* Convert and copy out the result */
if (odata.rmtbulk_len > out_size) {
code = E2BIG;
goto out_odata;
}
if (odata.rmtbulk_len) {
outparam_conversion(com, odata.rmtbulk_val, odata.rmtbulk_len, 1);
AFS_COPYOUT(odata.rmtbulk_val, data.out, odata.rmtbulk_len, code);
}
if (!code)
code = err;
out_odata:
if (out_size > AFS_LRALLOCSIZ)
osi_Free(odata.rmtbulk_val, out_size);
else
osi_FreeLargeSpace(odata.rmtbulk_val);
out_idata:
if (in_size > AFS_LRALLOCSIZ)
osi_Free(idata.rmtbulk_val, in_size);
else
osi_FreeLargeSpace(idata.rmtbulk_val);
out_path:
if (path)
osi_FreeLargeSpace(pathbuf);
out:
crfree(credp);
#if defined(KERNEL_HAVE_UERROR)
if (!getuerror())
setuerror(code);
return (getuerror());
#else
return (code);
#endif
}
