static int
vnop_close_9p(struct vnop_close_args *ap)
{
openfid_9p *op;
node_9p *np;
int e;
TRACE();
e = 0;
np = NTO9P(ap->a_vp);
nlock_9p(np, NODE_LCK_EXCLUSIVE);
op = ofidget(np, ap->a_fflag);
if (op->fid == NOFID) {
e = EBADF;
goto error;
}
if (OSDecrementAtomic(&op->ref) == 1) {
if (ISSET(np->flags, NODE_MMAPPED))
ubc_msync(np->vp, 0, ubc_getsize(np->vp), NULL, UBC_PUSHDIRTY|UBC_SYNC);
else
cluster_push(np->vp, IO_CLOSE);
/* root gets clunk in vfs_unmount_9p() */
if (!ISSET(np->nmp->flags, F_UNMOUNTING))
e = clunk_9p(np->nmp, op->fid);
op->fid = NOFID;
}
error:
nunlock_9p(np);
return e;
}
void
telemetry_task_ctl_locked(task_t task, uint32_t reasons, int enable_disable)
{
uint32_t origflags;
assert((reasons != 0) && ((reasons | TF_TELEMETRY) == TF_TELEMETRY));
task_lock_assert_owned(task);
origflags = task->t_flags;
if (enable_disable == 1) {
task->t_flags |= reasons;
if ((origflags & TF_TELEMETRY) == 0) {
OSIncrementAtomic(&telemetry_active_tasks);
#if TELEMETRY_DEBUG
printf("%s: telemetry OFF -> ON (%d active)\n", proc_name_address(task->bsd_info), telemetry_active_tasks);
#endif
}
} else {
task->t_flags &= ~reasons;
if (((origflags & TF_TELEMETRY) != 0) && ((task->t_flags & TF_TELEMETRY) == 0)) {
/*
* If this task went from having at least one telemetry bit to having none,
* the net change was to disable telemetry for the task.
*/
OSDecrementAtomic(&telemetry_active_tasks);
#if TELEMETRY_DEBUG
printf("%s: telemetry ON -> OFF (%d active)\n", proc_name_address(task->bsd_info), telemetry_active_tasks);
#endif
}
}
}
void OSMetaClass::instanceDestructed() const
{
if ((1 == OSDecrementAtomic((SInt32 *) &instanceCount)) && superClassLink)
superClassLink->instanceDestructed();
if( ((int) instanceCount) < 0)
printf("%s: bad retain(%d)", getClassName(), instanceCount);
}
/*
* Free struct sackhole.
*/
static void
tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole)
{
zfree(sack_hole_zone, hole);
tp->snd_numholes--;
OSDecrementAtomic(&tcp_sack_globalholes);
}
static void
utun_detached(
ifnet_t interface)
{
struct utun_pcb *pcb = ifnet_softc(interface);
utun_free(pcb);
OSDecrementAtomic(&utun_ifcount);
}
void
rw_exit(krwlock_t *rwlp)
{
if (rwlp->rw_owner == current_thread()) {
rwlp->rw_owner = NULL;
lck_rw_unlock_exclusive((lck_rw_t *)&rwlp->rw_lock[0]);
} else {
OSDecrementAtomic((volatile SInt32 *)&rwlp->rw_readers);
lck_rw_unlock_shared((lck_rw_t *)&rwlp->rw_lock[0]);
}
}
static void
ipsec_detached(
ifnet_t interface)
{
struct ipsec_pcb *pcb = ifnet_softc(interface);
ifnet_release(pcb->ipsec_ifp);
ipsec_free(pcb);
OSDecrementAtomic(&ipsec_ifcount);
}
/*
* Routine: lock_set_dereference
*
* Release a reference on a lock set. If this is the last reference,
* the lock set data structure is deallocated.
*/
void
lock_set_dereference(lock_set_t lock_set)
{
int size;
if (1 == OSDecrementAtomic(&((lock_set)->ref_count))) {
ipc_port_dealloc_kernel(lock_set->port);
size = (int)(sizeof(struct lock_set) +
(sizeof(struct ulock) * (lock_set->n_ulocks - 1)));
kfree(lock_set, size);
}
}
void com_VFSFilter0::putClient()
{
//
// do not exchange or add any condition before OSDecrementAtomic as it must be always done!
//
if( 0x1 == OSDecrementAtomic( &this->clientInvocations ) && NULL == this->userClient ){
//
// this was the last invocation
//
wakeup( &this->clientInvocations );
}
}
void KernelPatcher::onKextSummariesUpdated() {
if (that) {
// macOS 10.12 generates an interrupt during this call but unlike 10.11 and below
// it never stops handling interrupts hanging forever inside hndl_allintrs.
// This happens even with cpus=1, and the reason is not fully understood.
//
// For this reason on 10.12 and above the outer function is routed, and so far it
// seems to cause fewer issues. Regarding syncing:
// - the only place modifying gLoadedKextSummaries is updateLoadedKextSummaries;
// - updateLoadedKextSummaries is called from load/unload separately;
// - sKextSummariesLock is not exported or visible.
// As a result no syncing should be necessary but there are guards for future
// changes and in case of any misunderstanding.
if (getKernelVersion() >= KernelVersion::Sierra) {
if (OSIncrementAtomic(&updateSummariesEntryCount) != 0) {
panic("onKextSummariesUpdated entered another time");
}
that->orgUpdateLoadedKextSummaries();
}
DBGLOG("patcher @ invoked at kext loading/unloading");
if (that->khandlers.size() > 0 && that->loadedKextSummaries) {
auto num = (*that->loadedKextSummaries)->numSummaries;
if (num > 0) {
OSKextLoadedKextSummary &last = (*that->loadedKextSummaries)->summaries[num-1];
DBGLOG("patcher @ last kext is %llX and its name is %.*s", last.address, KMOD_MAX_NAME, last.name);
// We may add khandlers items inside the handler
for (size_t i = 0; i < that->khandlers.size(); i++) {
if (!strncmp(that->khandlers[i]->id, last.name, KMOD_MAX_NAME)) {
DBGLOG("patcher @ caught the right kext at %llX, invoking handler", last.address);
that->khandlers[i]->address = last.address;
that->khandlers[i]->size = last.size;
that->khandlers[i]->handler(that->khandlers[i]);
// Remove the item
that->khandlers.erase(i);
break;
}
}
} else {
SYSLOG("patcher @ no kext is currently loaded, this should not happen");
}
}
if (getKernelVersion() >= KernelVersion::Sierra && OSDecrementAtomic(&updateSummariesEntryCount) != 1) {
panic("onKextSummariesUpdated left another time");
}
}
}
void
OSMetaClass::instanceDestructed() const
{
if ((1 == OSDecrementAtomic(&instanceCount)) && superClassLink) {
superClassLink->instanceDestructed();
}
if (((int)instanceCount) < 0) {
OSKext * myKext = (OSKext *)reserved;
OSKextLog(myKext, kOSMetaClassLogSpec,
// xxx - this phrasing is rather cryptic
"OSMetaClass: Class %s - bad retain (%d)",
getClassName(), instanceCount);
}
}
VFSFilter0UserClient* com_VFSFilter0::getClient()
/*
if non NULL is returned the putClient() must be called
*/
{
VFSFilter0UserClient* currentClient;
//
// if ther is no user client, then nobody call for logging
//
if( NULL == this->userClient || this->pendingUnregistration )
return NULL;
OSIncrementAtomic( &this->clientInvocations );
currentClient = (VFSFilter0UserClient*)this->userClient;
//
// if the current client is NULL or can't be atomicaly exchanged
// with the same value then the unregistration is in progress,
// the call to OSCompareAndSwapPtr( NULL, NULL, &this->userClient )
// checks the this->userClient for NULL atomically
//
if( !currentClient ||
!OSCompareAndSwapPtr( currentClient, currentClient, &this->userClient ) ||
OSCompareAndSwapPtr( NULL, NULL, &this->userClient ) ){
//
// the unregistration is in the progress and waiting for all
// invocations to return
//
assert( this->pendingUnregistration );
if( 0x1 == OSDecrementAtomic( &this->clientInvocations ) ){
//
// this was the last invocation
//
wakeup( &this->clientInvocations );
}
return NULL;
}
return currentClient;
}
void throttle_lowpri_io(boolean_t ok_to_sleep)
{
int i;
int max_try_num;
struct uthread *ut;
struct _throttle_io_info_t *info;
ut = get_bsdthread_info(current_thread());
if ((ut->uu_lowpri_window == 0) || (ut->uu_throttle_info == NULL))
goto done;
info = ut->uu_throttle_info;
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_START,
ut->uu_lowpri_window, ok_to_sleep, 0, 0, 0);
if (ok_to_sleep == TRUE) {
max_try_num = lowpri_max_waiting_msecs / LOWPRI_SLEEP_INTERVAL * MAX(1, info->numthreads_throttling);
for (i=0; i<max_try_num; i++) {
if (throttle_io_will_be_throttled_internal(ut->uu_lowpri_window, info)) {
IOSleep(LOWPRI_SLEEP_INTERVAL);
DEBUG_ALLOC_THROTTLE_INFO("sleeping because of info = %p\n", info, info );
} else {
break;
}
}
}
KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, 97)) | DBG_FUNC_END,
ut->uu_lowpri_window, i*5, 0, 0, 0);
SInt32 oldValue;
oldValue = OSDecrementAtomic(&info->numthreads_throttling);
if (oldValue <= 0) {
panic("%s: numthreads negative", __func__);
}
done:
ut->uu_lowpri_window = 0;
if (ut->uu_throttle_info)
throttle_info_rel(ut->uu_throttle_info);
ut->uu_throttle_info = NULL;
}
__private_extern__ void
devtimer_release(devtimer_ref timer)
{
UInt32 old_retain_count;
old_retain_count = OSDecrementAtomic(&timer->dt_retain_count);
switch (old_retain_count) {
case 0:
panic("devtimer_release: retain count is 0\n");
break;
case 1:
devtimer_invalidate(timer);
FREE(timer, M_DEVTIMER);
_devtimer_printf("devtimer: timer released\n");
break;
default:
break;
}
return;
}
int
fuse_filehandle_put(vnode_t vp, vfs_context_t context, fufh_type_t fufh_type)
{
struct fuse_dispatcher fdi;
struct fuse_release_in *fri;
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct fuse_filehandle *fufh = NULL;
fuse_trace_printf("fuse_filehandle_put(vp=%p, fufh_type=%d)\n",
vp, fufh_type);
fufh = &(fvdat->fufh[fufh_type]);
if (FUFH_IS_VALID(fufh)) {
panic("fuse4x: filehandle_put called on a valid fufh (type=%d)",
fufh_type);
/* NOTREACHED */
}
if (fuse_isdeadfs(vp)) {
goto out;
}
int op = vnode_isdir(vp) ? FUSE_RELEASEDIR : FUSE_RELEASE;
fuse_dispatcher_init(&fdi, sizeof(*fri));
fuse_dispatcher_make_vp(&fdi, op, vp, context);
fri = fdi.indata;
fri->fh = fufh->fh_id;
fri->flags = fufh->open_flags;
int err = fuse_dispatcher_wait_answer(&fdi);
if (!err) {
fuse_ticket_drop(fdi.ticket);
}
out:
OSDecrementAtomic((SInt32 *)&fuse_fh_current);
fuse_invalidate_attr(vp);
return err;
}
/*
* Release the reference and if the item was allocated and this is the last
* reference then free it.
*
* This routine always returns the old value.
*/
static int
throttle_info_rel(struct _throttle_io_info_t *info)
{
SInt32 oldValue = OSDecrementAtomic(&info->refcnt);
DEBUG_ALLOC_THROTTLE_INFO("refcnt = %d info = %p\n",
info, (int)(oldValue -1), info );
/* The reference count just went negative, very bad */
if (oldValue == 0)
panic("throttle info ref cnt went negative!");
/*
* Once reference count is zero, no one else should be able to take a
* reference
*/
if ((info->refcnt == 0) && (info->alloc)) {
DEBUG_ALLOC_THROTTLE_INFO("Freeing info = %p\n", info, info );
FREE(info, M_TEMP);
}
return oldValue;
}
/*ARGSUSED*/
static int
zfs_vfs_unmount(struct mount *mp, int mntflags, vfs_context_t context)
{
zfsvfs_t *zfsvfs = vfs_fsprivate(mp);
objset_t *os = zfsvfs->z_os;
znode_t *zp, *nextzp;
int ret, i;
int flags;
/*XXX NOEL: delegation admin stuffs, add back if we use delg. admin */
#if 0
ret = 0; /* UNDEFINED: secpolicy_fs_unmount(cr, vfsp); */
if (ret) {
ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
ZFS_DELEG_PERM_MOUNT, cr);
if (ret)
return (ret);
}
/*
* We purge the parent filesystem's vfsp as the parent filesystem
* and all of its snapshots have their vnode's v_vfsp set to the
* parent's filesystem's vfsp. Note, 'z_parent' is self
* referential for non-snapshots.
*/
(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
#endif
/*
* Unmount any snapshots mounted under .zfs before unmounting the
* dataset itself.
*/
#if 0
if (zfsvfs->z_ctldir != NULL &&
(ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
return (ret);
#endif
flags = SKIPSYSTEM;
if (mntflags & MNT_FORCE)
flags |= FORCECLOSE;
ret = vflush(mp, NULLVP, flags);
/*
* Mac OS X needs a file system modify time
*
* We use the mtime of the "com.apple.system.mtime"
* extended attribute, which is associated with the
* file system root directory.
*
* Here we need to release the ref we took on z_mtime_vp during mount.
*/
if ((ret == 0) || (mntflags & MNT_FORCE)) {
if (zfsvfs->z_mtime_vp != NULL) {
struct vnode *mvp;
mvp = zfsvfs->z_mtime_vp;
zfsvfs->z_mtime_vp = NULL;
if (vnode_get(mvp) == 0) {
vnode_rele(mvp);
vnode_recycle(mvp);
vnode_put(mvp);
}
}
}
if (!(mntflags & MNT_FORCE)) {
/*
* Check the number of active vnodes in the file system.
* Our count is maintained in the vfs structure, but the
* number is off by 1 to indicate a hold on the vfs
* structure itself.
*
* The '.zfs' directory maintains a reference of its
* own, and any active references underneath are
* reflected in the vnode count.
*/
if (ret)
return (EBUSY);
#if 0
if (zfsvfs->z_ctldir == NULL) {
if (vfsp->vfs_count > 1)
return (EBUSY);
} else {
if (vfsp->vfs_count > 2 ||
zfsvfs->z_ctldir->v_count > 1) {
return (EBUSY);
}
}
#endif
}
rw_enter(&zfsvfs->z_unmount_lock, RW_WRITER);
rw_enter(&zfsvfs->z_unmount_inactive_lock, RW_WRITER);
/*
* At this point there are no vops active, and any new vops will
* fail with EIO since we have z_unmount_lock for writer (only
* relavent for forced unmount).
*
* Release all holds on dbufs.
* Note, the dmu can still callback via znode_pageout_func()
* which can zfs_znode_free() the znode. So we lock
* z_all_znodes; search the list for a held dbuf; drop the lock
* (we know zp can't disappear if we hold a dbuf lock) then
* regrab the lock and restart.
*/
mutex_enter(&zfsvfs->z_znodes_lock);
for (zp = list_head(&zfsvfs->z_all_znodes); zp; zp = nextzp) {
nextzp = list_next(&zfsvfs->z_all_znodes, zp);
if (zp->z_dbuf_held) {
/* dbufs should only be held when force unmounting */
zp->z_dbuf_held = 0;
mutex_exit(&zfsvfs->z_znodes_lock);
dmu_buf_rele(zp->z_dbuf, NULL);
/* Start again */
mutex_enter(&zfsvfs->z_znodes_lock);
nextzp = list_head(&zfsvfs->z_all_znodes);
}
}
mutex_exit(&zfsvfs->z_znodes_lock);
/*
* Set the unmounted flag and let new vops unblock.
* zfs_inactive will have the unmounted behavior, and all other
* vops will fail with EIO.
*/
zfsvfs->z_unmounted = B_TRUE;
rw_exit(&zfsvfs->z_unmount_lock);
rw_exit(&zfsvfs->z_unmount_inactive_lock);
/*
* Unregister properties.
*/
#ifndef __APPLE__
if (!dmu_objset_is_snapshot(os))
zfs_unregister_callbacks(zfsvfs);
#endif
/*
* Close the zil. NB: Can't close the zil while zfs_inactive
* threads are blocked as zil_close can call zfs_inactive.
*/
if (zfsvfs->z_log) {
zil_close(zfsvfs->z_log);
zfsvfs->z_log = NULL;
}
/*
* Evict all dbufs so that cached znodes will be freed
*/
if (dmu_objset_evict_dbufs(os, B_TRUE)) {
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
(void) dmu_objset_evict_dbufs(os, B_FALSE);
}
/*
* Finally close the objset
*/
dmu_objset_close(os);
/*
* We can now safely destroy the '.zfs' directory node.
*/
#if 0
if (zfsvfs->z_ctldir != NULL)
zfsctl_destroy(zfsvfs);
#endif
/*
* Note that this work is normally done in zfs_freevfs, but since
* there is no VOP_FREEVFS in OSX, we free VFS items here
*/
OSDecrementAtomic((SInt32 *)&zfs_active_fs_count);
for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
mutex_destroy(&zfsvfs->z_hold_mtx[i]);
mutex_destroy(&zfsvfs->z_znodes_lock);
list_destroy(&zfsvfs->z_all_znodes);
rw_destroy(&zfsvfs->z_unmount_lock);
rw_destroy(&zfsvfs->z_unmount_inactive_lock);
return (0);
}
struct vnode* vnode_getparent(struct vnode *vp); /* sys/vnode_internal.h */
static int
zfs_vget_internal(zfsvfs_t *zfsvfs, ino64_t ino, struct vnode **vpp)
{
struct vnode *vp;
struct vnode *dvp = NULL;
znode_t *zp;
int error;
*vpp = NULL;
/*
* On Mac OS X we always export the root directory id as 2
* and its parent as 1
*/
if (ino == 2 || ino == 1)
ino = zfsvfs->z_root;
if ((error = zfs_zget(zfsvfs, ino, &zp)))
goto out;
/* Don't expose EA objects! */
if (zp->z_phys->zp_flags & ZFS_XATTR) {
vnode_put(ZTOV(zp));
error = ENOENT;
goto out;
}
*vpp = vp = ZTOV(zp);
if (vnode_isvroot(vp))
goto out;
/*
* If this znode didn't just come from the cache then
* it won't have a valid identity (parent and name).
*
* Manually fix its identity here (normally done by namei lookup).
*/
if ((dvp = vnode_getparent(vp)) == NULL) {
if (zp->z_phys->zp_parent != 0 &&
zfs_vget_internal(zfsvfs, zp->z_phys->zp_parent, &dvp)) {
goto out;
}
if ( vnode_isdir(dvp) ) {
char objname[ZAP_MAXNAMELEN]; /* 256 bytes */
int flags = VNODE_UPDATE_PARENT;
/* Look for znode's name in its parent's zap */
if ( zap_value_search(zfsvfs->z_os,
zp->z_phys->zp_parent,
zp->z_id,
ZFS_DIRENT_OBJ(-1ULL),
objname) == 0 ) {
flags |= VNODE_UPDATE_NAME;
}
/* Update the znode's parent and name */
vnode_update_identity(vp, dvp, objname, 0, 0, flags);
}
}
/* All done with znode's parent */
vnode_put(dvp);
out:
return (error);
}
/*
* Get a vnode from a file id (ignoring the generation)
*
* Use by NFS Server (readdirplus) and VFS (build_path)
*/
static int
zfs_vfs_vget(struct mount *mp, ino64_t ino, struct vnode **vpp, __unused vfs_context_t context)
{
zfsvfs_t *zfsvfs = vfs_fsprivate(mp);
int error;
ZFS_ENTER(zfsvfs);
/*
* On Mac OS X we always export the root directory id as 2.
* So we don't expect to see the real root directory id
* from zfs_vfs_vget KPI (unless of course the real id was
* already 2).
*/
if ((ino == zfsvfs->z_root) && (zfsvfs->z_root != 2)) {
ZFS_EXIT(zfsvfs);
return (ENOENT);
}
error = zfs_vget_internal(zfsvfs, ino, vpp);
ZFS_EXIT(zfsvfs);
return (error);
}
static int tcp_cubic_cleanup(struct tcpcb *tp)
{
#pragma unused(tp)
OSDecrementAtomic((volatile SInt32 *)&tcp_cc_cubic.num_sockets);
return (0);
}
static errno_t
fuse_vfsop_mount(mount_t mp, __unused vnode_t devvp, user_addr_t udata,
vfs_context_t context)
{
int err = 0;
int mntopts = 0;
bool mounted = false;
uint32_t max_read = ~0;
size_t len;
fuse_device_t fdev = NULL;
struct fuse_data *data = NULL;
fuse_mount_args fusefs_args;
struct vfsstatfs *vfsstatfsp = vfs_statfs(mp);
fuse_trace_printf_vfsop();
if (vfs_isupdate(mp)) {
return ENOTSUP;
}
err = copyin(udata, &fusefs_args, sizeof(fusefs_args));
if (err) {
return EINVAL;
}
/*
* Interesting flags that we can receive from mount or may want to
* otherwise forcibly set include:
*
* MNT_ASYNC
* MNT_AUTOMOUNTED
* MNT_DEFWRITE
* MNT_DONTBROWSE
* MNT_IGNORE_OWNERSHIP
* MNT_JOURNALED
* MNT_NODEV
* MNT_NOEXEC
* MNT_NOSUID
* MNT_NOUSERXATTR
* MNT_RDONLY
* MNT_SYNCHRONOUS
* MNT_UNION
*/
err = ENOTSUP;
vfs_setlocklocal(mp);
/** Option Processing. **/
if (*fusefs_args.fstypename) {
size_t typenamelen = strlen(fusefs_args.fstypename);
if (typenamelen > FUSE_FSTYPENAME_MAXLEN) {
return EINVAL;
}
snprintf(vfsstatfsp->f_fstypename, MFSTYPENAMELEN, "%s%s",
FUSE_FSTYPENAME_PREFIX, fusefs_args.fstypename);
}
if (!*fusefs_args.fsname)
return EINVAL;
if ((fusefs_args.daemon_timeout > FUSE_MAX_DAEMON_TIMEOUT) ||
(fusefs_args.daemon_timeout < FUSE_MIN_DAEMON_TIMEOUT)) {
return EINVAL;
}
if (fusefs_args.altflags & FUSE_MOPT_SPARSE) {
mntopts |= FSESS_SPARSE;
}
if (fusefs_args.altflags & FUSE_MOPT_AUTO_CACHE) {
mntopts |= FSESS_AUTO_CACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_AUTO_XATTR) {
if (fusefs_args.altflags & FUSE_MOPT_NATIVE_XATTR) {
return EINVAL;
}
mntopts |= FSESS_AUTO_XATTR;
} else if (fusefs_args.altflags & FUSE_MOPT_NATIVE_XATTR) {
mntopts |= FSESS_NATIVE_XATTR;
}
if (fusefs_args.altflags & FUSE_MOPT_JAIL_SYMLINKS) {
mntopts |= FSESS_JAIL_SYMLINKS;
}
/*
* Note that unlike Linux, which keeps allow_root in user-space and
* passes allow_other in that case to the kernel, we let allow_root
* reach the kernel. The 'if' ordering is important here.
*/
if (fusefs_args.altflags & FUSE_MOPT_ALLOW_ROOT) {
int is_member = 0;
if ((kauth_cred_ismember_gid(kauth_cred_get(), fuse_admin_group, &is_member) != 0) || !is_member) {
log("fuse4x: caller is not a member of fuse4x admin group. "
"Either add user (id=%d) to group (id=%d), "
"or set correct '" SYSCTL_FUSE4X_TUNABLES_ADMIN "' sysctl value.\n",
kauth_cred_getuid(kauth_cred_get()), fuse_admin_group);
return EPERM;
}
mntopts |= FSESS_ALLOW_ROOT;
} else if (fusefs_args.altflags & FUSE_MOPT_ALLOW_OTHER) {
if (!fuse_allow_other && !fuse_vfs_context_issuser(context)) {
int is_member = 0;
if ((kauth_cred_ismember_gid(kauth_cred_get(), fuse_admin_group, &is_member) != 0) || !is_member) {
log("fuse4x: caller is not a member of fuse4x admin group. "
"Either add user (id=%d) to group (id=%d), "
"or set correct '" SYSCTL_FUSE4X_TUNABLES_ADMIN "' sysctl value.\n",
kauth_cred_getuid(kauth_cred_get()), fuse_admin_group);
return EPERM;
}
}
mntopts |= FSESS_ALLOW_OTHER;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_APPLEDOUBLE) {
mntopts |= FSESS_NO_APPLEDOUBLE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_APPLEXATTR) {
mntopts |= FSESS_NO_APPLEXATTR;
}
if ((fusefs_args.altflags & FUSE_MOPT_FSID) && (fusefs_args.fsid != 0)) {
fsid_t fsid;
mount_t other_mp;
uint32_t target_dev;
target_dev = FUSE_MAKEDEV(FUSE_CUSTOM_FSID_DEVICE_MAJOR,
fusefs_args.fsid);
fsid.val[0] = target_dev;
fsid.val[1] = FUSE_CUSTOM_FSID_VAL1;
other_mp = vfs_getvfs(&fsid);
if (other_mp != NULL) {
return EPERM;
}
vfsstatfsp->f_fsid.val[0] = target_dev;
vfsstatfsp->f_fsid.val[1] = FUSE_CUSTOM_FSID_VAL1;
} else {
vfs_getnewfsid(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_NO_ATTRCACHE) {
mntopts |= FSESS_NO_ATTRCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_READAHEAD) {
mntopts |= FSESS_NO_READAHEAD;
}
if (fusefs_args.altflags & (FUSE_MOPT_NO_UBC | FUSE_MOPT_DIRECT_IO)) {
mntopts |= FSESS_NO_UBC;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_VNCACHE) {
mntopts |= FSESS_NO_VNCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NEGATIVE_VNCACHE) {
if (mntopts & FSESS_NO_VNCACHE) {
return EINVAL;
}
mntopts |= FSESS_NEGATIVE_VNCACHE;
}
if (fusefs_args.altflags & FUSE_MOPT_NO_SYNCWRITES) {
/* Cannot mix 'nosyncwrites' with 'noubc' or 'noreadahead'. */
if (mntopts & (FSESS_NO_READAHEAD | FSESS_NO_UBC)) {
log("fuse4x: cannot mix 'nosyncwrites' with 'noubc' or 'noreadahead'\n");
return EINVAL;
}
mntopts |= FSESS_NO_SYNCWRITES;
vfs_clearflags(mp, MNT_SYNCHRONOUS);
vfs_setflags(mp, MNT_ASYNC);
/* We check for this only if we have nosyncwrites in the first place. */
if (fusefs_args.altflags & FUSE_MOPT_NO_SYNCONCLOSE) {
mntopts |= FSESS_NO_SYNCONCLOSE;
}
} else {
vfs_clearflags(mp, MNT_ASYNC);
vfs_setflags(mp, MNT_SYNCHRONOUS);
}
if (mntopts & FSESS_NO_UBC) {
/* If no buffer cache, disallow exec from file system. */
vfs_setflags(mp, MNT_NOEXEC);
}
vfs_setauthopaque(mp);
vfs_setauthopaqueaccess(mp);
if ((fusefs_args.altflags & FUSE_MOPT_DEFAULT_PERMISSIONS) &&
(fusefs_args.altflags & FUSE_MOPT_DEFER_PERMISSIONS)) {
return EINVAL;
}
if (fusefs_args.altflags & FUSE_MOPT_DEFAULT_PERMISSIONS) {
mntopts |= FSESS_DEFAULT_PERMISSIONS;
vfs_clearauthopaque(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_DEFER_PERMISSIONS) {
mntopts |= FSESS_DEFER_PERMISSIONS;
}
if (fusefs_args.altflags & FUSE_MOPT_EXTENDED_SECURITY) {
mntopts |= FSESS_EXTENDED_SECURITY;
vfs_setextendedsecurity(mp);
}
if (fusefs_args.altflags & FUSE_MOPT_LOCALVOL) {
vfs_setflags(mp, MNT_LOCAL);
}
/* done checking incoming option bits */
err = 0;
vfs_setfsprivate(mp, NULL);
fdev = fuse_device_get(fusefs_args.rdev);
if (!fdev) {
log("fuse4x: invalid device file (number=%d)\n", fusefs_args.rdev);
return EINVAL;
}
fuse_lck_mtx_lock(fdev->mtx);
data = fdev->data;
if (!data) {
fuse_lck_mtx_unlock(fdev->mtx);
return ENXIO;
}
if (data->mounted) {
fuse_lck_mtx_unlock(fdev->mtx);
return EALREADY;
}
if (!data->opened) {
fuse_lck_mtx_unlock(fdev->mtx);
err = ENXIO;
goto out;
}
data->mounted = true;
OSIncrementAtomic((SInt32 *)&fuse_mount_count);
mounted = true;
if (data->dead) {
fuse_lck_mtx_unlock(fdev->mtx);
err = ENOTCONN;
goto out;
}
if (!data->daemoncred) {
panic("fuse4x: daemon found but identity unknown");
}
if (fuse_vfs_context_issuser(context) &&
kauth_cred_getuid(vfs_context_ucred(context)) != kauth_cred_getuid(data->daemoncred)) {
fuse_lck_mtx_unlock(fdev->mtx);
err = EPERM;
log("fuse4x: fuse daemon running by user_id=%d does not have privileges to mount on directory %s owned by user_id=%d\n",
kauth_cred_getuid(data->daemoncred), vfsstatfsp->f_mntonname, kauth_cred_getuid(vfs_context_ucred(context)));
goto out;
}
data->mp = mp;
data->fdev = fdev;
data->dataflags |= mntopts;
data->daemon_timeout.tv_sec = fusefs_args.daemon_timeout;
data->daemon_timeout.tv_nsec = 0;
if (data->daemon_timeout.tv_sec) {
data->daemon_timeout_p = &(data->daemon_timeout);
} else {
data->daemon_timeout_p = NULL;
}
data->max_read = max_read;
data->fssubtype = fusefs_args.fssubtype;
data->noimplflags = (uint64_t)0;
data->blocksize = fuse_round_size(fusefs_args.blocksize,
FUSE_MIN_BLOCKSIZE, FUSE_MAX_BLOCKSIZE);
data->iosize = fuse_round_size(fusefs_args.iosize,
FUSE_MIN_IOSIZE, FUSE_MAX_IOSIZE);
if (data->iosize < data->blocksize) {
data->iosize = data->blocksize;
}
data->userkernel_bufsize = FUSE_DEFAULT_USERKERNEL_BUFSIZE;
copystr(fusefs_args.fsname, vfsstatfsp->f_mntfromname,
MNAMELEN - 1, &len);
bzero(vfsstatfsp->f_mntfromname + len, MNAMELEN - len);
copystr(fusefs_args.volname, data->volname, MAXPATHLEN - 1, &len);
bzero(data->volname + len, MAXPATHLEN - len);
/* previous location of vfs_setioattr() */
vfs_setfsprivate(mp, data);
fuse_lck_mtx_unlock(fdev->mtx);
/* Send a handshake message to the daemon. */
fuse_send_init(data, context);
struct vfs_attr vfs_attr;
VFSATTR_INIT(&vfs_attr);
// Our vfs_getattr() doesn't look at most *_IS_ACTIVE()'s
err = fuse_vfsop_getattr(mp, &vfs_attr, context);
if (!err) {
vfsstatfsp->f_bsize = vfs_attr.f_bsize;
vfsstatfsp->f_iosize = vfs_attr.f_iosize;
vfsstatfsp->f_blocks = vfs_attr.f_blocks;
vfsstatfsp->f_bfree = vfs_attr.f_bfree;
vfsstatfsp->f_bavail = vfs_attr.f_bavail;
vfsstatfsp->f_bused = vfs_attr.f_bused;
vfsstatfsp->f_files = vfs_attr.f_files;
vfsstatfsp->f_ffree = vfs_attr.f_ffree;
// vfsstatfsp->f_fsid already handled above
vfsstatfsp->f_owner = kauth_cred_getuid(data->daemoncred);
vfsstatfsp->f_flags = vfs_flags(mp);
// vfsstatfsp->f_fstypename already handled above
// vfsstatfsp->f_mntonname handled elsewhere
// vfsstatfsp->f_mnfromname already handled above
vfsstatfsp->f_fssubtype = data->fssubtype;
}
if (fusefs_args.altflags & FUSE_MOPT_BLOCKSIZE) {
vfsstatfsp->f_bsize = data->blocksize;
} else {
//data->blocksize = vfsstatfsp->f_bsize;
}
if (fusefs_args.altflags & FUSE_MOPT_IOSIZE) {
vfsstatfsp->f_iosize = data->iosize;
} else {
//data->iosize = (uint32_t)vfsstatfsp->f_iosize;
vfsstatfsp->f_iosize = data->iosize;
}
out:
if (err) {
vfs_setfsprivate(mp, NULL);
fuse_lck_mtx_lock(fdev->mtx);
data = fdev->data; /* again */
if (mounted) {
OSDecrementAtomic((SInt32 *)&fuse_mount_count);
}
if (data) {
data->mounted = false;
if (!data->opened) {
fuse_device_close_final(fdev);
/* data is gone now */
}
}
fuse_lck_mtx_unlock(fdev->mtx);
} else {
vnode_t fuse_rootvp = NULLVP;
err = fuse_vfsop_root(mp, &fuse_rootvp, context);
if (err) {
goto out; /* go back and follow error path */
}
err = vnode_ref(fuse_rootvp);
(void)vnode_put(fuse_rootvp);
if (err) {
goto out; /* go back and follow error path */
} else {
struct vfsioattr ioattr;
vfs_ioattr(mp, &ioattr);
ioattr.io_devblocksize = data->blocksize;
ioattr.io_maxsegreadsize = ioattr.io_maxsegwritesize =
ioattr.io_maxreadcnt = ioattr.io_maxwritecnt = data->iosize;
ioattr.io_segreadcnt = ioattr.io_segwritecnt = data->iosize / PAGE_SIZE;
vfs_setioattr(mp, &ioattr);
}
}
return err;
}
static errno_t
fuse_vfsop_unmount(mount_t mp, int mntflags, vfs_context_t context)
{
int err = 0;
bool force = false;
fuse_device_t fdev;
struct fuse_data *data;
struct fuse_dispatcher fdi;
vnode_t fuse_rootvp = NULLVP;
fuse_trace_printf_vfsop();
if (mntflags & MNT_FORCE) {
force = true;
}
data = fuse_get_mpdata(mp);
if (!data) {
panic("fuse4x: no mount private data in vfs_unmount");
}
fdev = data->fdev;
if (data->dead) {
/*
* If the file system daemon is dead, it's pointless to try to do
* any unmount-time operations that go out to user space. Therefore,
* we pretend that this is a force unmount. However, this isn't of much
* use. That's because if any non-root vnode is in use, the vflush()
* that the kernel does before calling our VFS_UNMOUNT will fail
* if the original unmount wasn't forcible already. That earlier
* vflush is called with SKIPROOT though, so it wouldn't bail out
* on the root vnode being in use.
*
* If we want, we could set FORCECLOSE here so that a non-forced
* unmount will be "upgraded" to a forced unmount if the root vnode
* is busy (you are cd'd to the mount point, for example). It's not
* quite pure to do that though.
*
* flags |= FORCECLOSE;
* log("fuse4x: forcing unmount on a dead file system\n");
*/
} else if (!data->inited) {
force = true;
log("fuse4x: forcing unmount on not-yet-alive file system\n");
fuse_data_kill(data);
}
fuse_rootvp = data->rootvp;
fuse_trace_printf("%s: Calling vflush(mp, fuse_rootvp, flags=0x%X);\n", __FUNCTION__, force ? FORCECLOSE : 0);
err = vflush(mp, fuse_rootvp, force ? FORCECLOSE : 0);
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
if (err) {
return err;
}
if (vnode_isinuse(fuse_rootvp, 1) && !force) {
return EBUSY;
}
fuse_trace_printf("%s: Calling vnode_rele(fuse_rootp);\n", __FUNCTION__);
vnode_rele(fuse_rootvp); /* We got this reference in fuse_vfsop_mount(). */
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
data->rootvp = NULLVP;
fuse_trace_printf("%s: Calling vflush(mp, NULLVP, FORCECLOSE);\n", __FUNCTION__);
(void)vflush(mp, NULLVP, FORCECLOSE);
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
if (!data->dead) {
fuse_dispatcher_init(&fdi, 0 /* no data to send along */);
fuse_dispatcher_make(&fdi, FUSE_DESTROY, mp, FUSE_ROOT_ID, context);
fuse_trace_printf("%s: Waiting for reply from FUSE_DESTROY.\n", __FUNCTION__);
err = fuse_dispatcher_wait_answer(&fdi);
fuse_trace_printf("%s: Reply received.\n", __FUNCTION__);
if (!err) {
fuse_ticket_drop(fdi.ticket);
}
}
fuse_data_kill(data);
fuse_lck_mtx_lock(fdev->mtx);
vfs_setfsprivate(mp, NULL);
data->mounted = false;
OSDecrementAtomic((SInt32 *)&fuse_mount_count);
if (!data->opened) {
/* fdev->data was left for us to clean up */
fuse_device_close_final(fdev);
/* fdev->data is gone now */
}
fuse_lck_mtx_unlock(fdev->mtx);
return 0;
}
int tcp_newreno_cleanup(struct tcpcb *tp) {
#pragma unused(tp)
OSDecrementAtomic((volatile SInt32 *)&tcp_cc_newreno.num_sockets);
return 0;
}
void throttle_info_update(void *throttle_info, int flags)
{
struct _throttle_io_info_t *info = throttle_info;
struct uthread *ut;
int policy;
int is_throttleable_io = 0;
int is_passive_io = 0;
SInt32 oldValue;
if (!lowpri_IO_initial_window_msecs || (info == NULL))
return;
policy = throttle_get_io_policy(&ut);
switch (policy) {
case IOPOL_DEFAULT:
case IOPOL_NORMAL:
break;
case IOPOL_THROTTLE:
is_throttleable_io = 1;
break;
case IOPOL_PASSIVE:
is_passive_io = 1;
break;
default:
printf("unknown I/O policy %d", policy);
break;
}
if (!is_throttleable_io && ISSET(flags, B_PASSIVE))
is_passive_io |= 1;
if (!is_throttleable_io) {
if (!is_passive_io){
microuptime(&info->last_normal_IO_timestamp);
}
} else if (ut) {
/*
* I'd really like to do the IOSleep here, but
* we may be holding all kinds of filesystem related locks
* and the pages for this I/O marked 'busy'...
* we don't want to cause a normal task to block on
* one of these locks while we're throttling a task marked
* for low priority I/O... we'll mark the uthread and
* do the delay just before we return from the system
* call that triggered this I/O or from vnode_pagein
*/
if (ut->uu_lowpri_window == 0) {
ut->uu_throttle_info = info;
throttle_info_ref(ut->uu_throttle_info);
DEBUG_ALLOC_THROTTLE_INFO("updating info = %p\n", info, info );
oldValue = OSIncrementAtomic(&info->numthreads_throttling);
if (oldValue < 0) {
panic("%s: numthreads negative", __func__);
}
ut->uu_lowpri_window = lowpri_IO_initial_window_msecs;
ut->uu_lowpri_window += lowpri_IO_window_msecs_inc * oldValue;
} else {
/* The thread sends I/Os to different devices within the same system call */
if (ut->uu_throttle_info != info) {
struct _throttle_io_info_t *old_info = ut->uu_throttle_info;
// keep track of the numthreads in the right device
OSDecrementAtomic(&old_info->numthreads_throttling);
OSIncrementAtomic(&info->numthreads_throttling);
DEBUG_ALLOC_THROTTLE_INFO("switching from info = %p\n", old_info, old_info );
DEBUG_ALLOC_THROTTLE_INFO("switching to info = %p\n", info, info );
/* This thread no longer needs a reference on that throttle info */
throttle_info_rel(ut->uu_throttle_info);
ut->uu_throttle_info = info;
/* Need to take a reference on this throttle info */
throttle_info_ref(ut->uu_throttle_info);
}
int numthreads = MAX(1, info->numthreads_throttling);
ut->uu_lowpri_window += lowpri_IO_window_msecs_inc * numthreads;
if (ut->uu_lowpri_window > lowpri_max_window_msecs * numthreads)
ut->uu_lowpri_window = lowpri_max_window_msecs * numthreads;
}
}
}
// Called at hardware interrupt time
bool
AppleUSBUHCI::FilterInterrupt(void)
{
UInt16 activeInterrupts;
Boolean needSignal = false;
UInt64 currentFrame;
uint64_t timeStamp;
// we leave all interrupts enabled, so see which ones are active
activeInterrupts = ioRead16(kUHCI_STS) & kUHCI_STS_INTR_MASK;
if (activeInterrupts != 0)
{
USBTrace( kUSBTUHCIInterrupts, kTPUHCIInterruptsFilterInterrupt , (uintptr_t)this, activeInterrupts, 0, 3 );
if (activeInterrupts & kUHCI_STS_HCPE)
{
// Host Controller Process Error - usually a bad data structure on the list
_hostControllerProcessInterrupt = kUHCI_STS_HCPE;
ioWrite16(kUHCI_STS, kUHCI_STS_HCPE);
needSignal = true;
//USBLog(1, "AppleUSBUHCI[%p]::FilterInterrupt - HCPE error - legacy reg = %p", this, (void*)_device->configRead16(kUHCI_PCI_LEGKEY));
USBTrace( kUSBTUHCIInterrupts, kTPUHCIInterruptsFilterInterrupt, (uintptr_t)this, _hostControllerProcessInterrupt, _device->configRead16(kUHCI_PCI_LEGKEY), 1 );
}
if (activeInterrupts & kUHCI_STS_HSE)
{
// Host System Error - usually a PCI issue
_hostSystemErrorInterrupt = kUHCI_STS_HSE;
ioWrite16(kUHCI_STS, kUHCI_STS_HSE);
needSignal = true;
//USBLog(1, "AppleUSBUHCI[%p]::FilterInterrupt - HSE error - legacy reg = %p", this, (void*)_device->configRead16(kUHCI_PCI_LEGKEY));
USBTrace( kUSBTUHCIInterrupts, kTPUHCIInterruptsFilterInterrupt, (uintptr_t)this, _hostSystemErrorInterrupt, _device->configRead16(kUHCI_PCI_LEGKEY), 2 );
}
if (activeInterrupts & kUHCI_STS_RD)
{
// Resume Detect - remote wakeup
_resumeDetectInterrupt = kUHCI_STS_RD;
ioWrite16(kUHCI_STS, kUHCI_STS_RD);
needSignal = true;
}
if (activeInterrupts & kUHCI_STS_EI)
{
// USB Error Interrupt - transaction error (CRC, timeout, etc)
_usbErrorInterrupt = kUHCI_STS_EI;
ioWrite16(kUHCI_STS, kUHCI_STS_EI);
needSignal = true;
}
if (activeInterrupts & kUHCI_STS_INT)
{
// Normal IOC interrupt - we need to check out low latency Isoch as well
timeStamp = mach_absolute_time();
ioWrite16(kUHCI_STS, kUHCI_STS_INT);
needSignal = true;
// This function will give us the current frame number and check for rollover at the same time
// since we are calling from filterInterrupts we will not be preempted, it will also update the
// cached copy of the frame_number_with_time
GetFrameNumberInternal();
// we need to check the periodic list to see if there are any Isoch TDs which need to come off
// and potentially have their frame lists updated (for Low Latency) we will place them in reverse
// order on a "done queue" which will be looked at by the isoch scavanger
// only do this if the periodic schedule is enabled
if (!_inAbortIsochEP && (_outSlot < kUHCI_NVFRAMES))
{
AppleUHCIIsochTransferDescriptor *cachedHead;
UInt32 cachedProducer;
UInt16 curSlot, testSlot, nextSlot;
curSlot = (ReadFrameNumber() & kUHCI_NVFRAMES_MASK);
cachedHead = (AppleUHCIIsochTransferDescriptor*)_savedDoneQueueHead;
cachedProducer = _producerCount;
testSlot = _outSlot;
while (testSlot != curSlot)
{
IOUSBControllerListElement *thing, *nextThing;
AppleUHCIIsochTransferDescriptor *isochTD;
nextSlot = (testSlot+1) & kUHCI_NVFRAMES_MASK;
thing = _logicalFrameList[testSlot];
while (thing != NULL)
{
nextThing = thing->_logicalNext;
isochTD = OSDynamicCast(AppleUHCIIsochTransferDescriptor, thing);
if (!isochTD)
break; // only care about Isoch in this list - if we get here we are at the interrupt TDs
// need to unlink this TD
_logicalFrameList[testSlot] = nextThing;
_frameList[testSlot] = HostToUSBLong(thing->GetPhysicalLink());
if (isochTD->_lowLatency)
isochTD->frStatus = isochTD->UpdateFrameList(*(AbsoluteTime*)&timeStamp);
// place this guy on the backward done queue
// the reason that we do not use the _logicalNext link is that the done queue is not a null terminated list
// and the element linked "last" in the list might not be a true link - trust me
isochTD->_doneQueueLink = cachedHead;
cachedHead = isochTD;
cachedProducer++;
if (isochTD->_pEndpoint)
{
isochTD->_pEndpoint->onProducerQ++;
OSDecrementAtomic( &(isochTD->_pEndpoint->scheduledTDs));
}
thing = nextThing;
}
testSlot = nextSlot;
_outSlot = testSlot;
}
IOSimpleLockLock( _wdhLock );
_savedDoneQueueHead = cachedHead; // updates the shadow head
_producerCount = cachedProducer; // Validates _producerCount;
IOSimpleLockUnlock( _wdhLock );
}
// 8394970: Make sure we set the flag AFTER we have incremented our producer count.
_usbCompletionInterrupt = kUHCI_STS_INT;
}
}
// We will return false from this filter routine,
// but will indicate that there the action routine should be called
// by calling _filterInterruptSource->signalInterrupt().
// This is needed because IOKit will disable interrupts for a level interrupt
// after the filter interrupt is run, until the action interrupt is called.
// We want to be able to have our filter interrupt routine called
// before the action routine runs, if needed. That is what will enable
// low latency isoch transfers to work, as when the
// system is under heavy load, the action routine can be delayed for tens of ms.
//
if (needSignal)
_interruptSource->signalInterrupt();
return false;
}