/*
* Destroy a process structure that resulted from a call to forkproc(), but
* which must be returned to the system because of a subsequent failure
* preventing it from becoming active.
*
* Parameters: p The incomplete process from forkproc()
*
* Returns: (void)
*
* Note: This function should only be used in an error handler following
* a call to forkproc().
*
* Operations occur in reverse order of those in forkproc().
*/
void
forkproc_free(proc_t p)
{
/* We held signal and a transition locks; drop them */
proc_signalend(p, 0);
proc_transend(p, 0);
/*
* If we have our own copy of the resource limits structure, we
* need to free it. If it's a shared copy, we need to drop our
* reference on it.
*/
proc_limitdrop(p, 0);
p->p_limit = NULL;
#if SYSV_SHM
/* Need to drop references to the shared memory segment(s), if any */
if (p->vm_shm) {
/*
* Use shmexec(): we have no address space, so no mappings
*
* XXX Yes, the routine is badly named.
*/
shmexec(p);
}
#endif
/* Need to undo the effects of the fdcopy(), if any */
fdfree(p);
/*
* Drop the reference on a text vnode pointer, if any
* XXX This code is broken in forkproc(); see <rdar://4256419>;
* XXX if anyone ever uses this field, we will be extremely unhappy.
*/
if (p->p_textvp) {
vnode_rele(p->p_textvp);
p->p_textvp = NULL;
}
/* Stop the profiling clock */
stopprofclock(p);
/* Release the credential reference */
kauth_cred_unref(&p->p_ucred);
proc_list_lock();
/* Decrement the count of processes in the system */
nprocs--;
proc_list_unlock();
thread_call_free(p->p_rcall);
/* Free allocated memory */
FREE_ZONE(p->p_sigacts, sizeof *p->p_sigacts, M_SIGACTS);
FREE_ZONE(p->p_stats, sizeof *p->p_stats, M_PSTATS);
proc_checkdeadrefs(p);
FREE_ZONE(p, sizeof *p, M_PROC);
}
int
imageboot_setup()
{
dev_t dev;
int error = 0;
char *root_path = NULL;
DBG_TRACE("%s: entry\n", __FUNCTION__);
MALLOC_ZONE(root_path, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
if (root_path == NULL)
return (ENOMEM);
if(PE_parse_boot_argn("rp", root_path, MAXPATHLEN) == FALSE) {
error = ENOENT;
goto done;
}
printf("%s: root image url is %s\n", __FUNCTION__, root_path);
error = di_root_image(root_path, rootdevice, &dev);
if(error) {
printf("%s: di_root_image failed: %d\n", __FUNCTION__, error);
goto done;
}
rootdev = dev;
mountroot = NULL;
printf("%s: root device 0x%x\n", __FUNCTION__, rootdev);
error = vfs_mountroot();
if (error == 0 && rootvnode != NULL) {
struct vnode *tvp;
struct vnode *newdp;
/*
* Get the vnode for '/'.
* Set fdp->fd_fd.fd_cdir to reference it.
*/
if (VFS_ROOT(TAILQ_LAST(&mountlist,mntlist), &newdp, vfs_context_kernel()))
panic("%s: cannot find root vnode", __FUNCTION__);
vnode_ref(newdp);
vnode_put(newdp);
tvp = rootvnode;
vnode_rele(tvp);
filedesc0.fd_cdir = newdp;
rootvnode = newdp;
mount_list_lock();
TAILQ_REMOVE(&mountlist, TAILQ_FIRST(&mountlist), mnt_list);
mount_list_unlock();
mountlist.tqh_first->mnt_flag |= MNT_ROOTFS;
DBG_TRACE("%s: root switched\n", __FUNCTION__);
}
done:
FREE_ZONE(root_path, MAXPATHLEN, M_NAMEI);
DBG_TRACE("%s: exit\n", __FUNCTION__);
return (error);
}
/*
* Make a new or get existing nullfs node.
* Vp is the alias vnode, lowervp is the lower vnode.
*
* lowervp is assumed to have an iocount on it from the caller
*/
int
null_nodeget(
struct mount * mp, struct vnode * lowervp, struct vnode * dvp, struct vnode ** vpp, struct componentname * cnp, int root)
{
struct vnode * vp;
int error;
/* Lookup the hash firstly. */
error = null_hashget(mp, lowervp, vpp);
/* ENOENT means it wasn't found, EIO is a failure we should bail from, 0 is it
* was found */
if (error != ENOENT) {
/* null_hashget checked the vid, so if we got something here its legit to
* the best of our knowledge*/
/* if we found something then there is an iocount on vpp,
if we didn't find something then vpp shouldn't be used by the caller */
return error;
}
/*
* We do not serialize vnode creation, instead we will check for
* duplicates later, when adding new vnode to hash.
*/
error = vnode_ref(lowervp); // take a ref on lowervp so we let the system know we care about it
if(error)
{
// Failed to get a reference on the lower vp so bail. Lowervp may be gone already.
return error;
}
error = null_getnewvnode(mp, lowervp, dvp, &vp, cnp, root);
if (error) {
vnode_rele(lowervp);
return (error);
}
/*
* Atomically insert our new node into the hash or vget existing
* if someone else has beaten us to it.
*/
error = null_hashins(mp, VTONULL(vp), vpp);
if (error || *vpp != NULL) {
/* recycle will call reclaim which will get rid of the internals */
vnode_recycle(vp);
vnode_put(vp);
/* if we found vpp, then null_hashins put an iocount on it */
return error;
}
/* vp has an iocount from null_getnewvnode */
*vpp = vp;
return (0);
}
static int
null_reclaim(struct vnop_reclaim_args * ap)
{
struct vnode * vp;
struct null_node * xp;
struct vnode * lowervp;
struct null_mount * null_mp = MOUNTTONULLMOUNT(vnode_mount(ap->a_vp));
NULLFSDEBUG("%s %p\n", __FUNCTION__, ap->a_vp);
vp = ap->a_vp;
xp = VTONULL(vp);
lowervp = xp->null_lowervp;
lck_mtx_lock(&null_mp->nullm_lock);
vnode_removefsref(vp);
if (lowervp != NULL) {
/* root and second don't have a lowervp, so nothing to release and nothing
* got hashed */
if (xp->null_flags & NULL_FLAG_HASHED) {
/* only call this if we actually made it into the hash list. reclaim gets
called also to
clean up a vnode that got created when it didn't need to under race
conditions */
null_hashrem(xp);
}
vnode_getwithref(lowervp);
vnode_rele(lowervp);
vnode_put(lowervp);
}
if (vp == null_mp->nullm_rootvp) {
null_mp->nullm_rootvp = NULL;
} else if (vp == null_mp->nullm_secondvp) {
null_mp->nullm_secondvp = NULL;
} else if (vp == null_mp->nullm_thirdcovervp) {
null_mp->nullm_thirdcovervp = NULL;
}
lck_mtx_unlock(&null_mp->nullm_lock);
cache_purge(vp);
vnode_clearfsnode(vp);
FREE(xp, M_TEMP);
return 0;
}
static int
vfs_unmount_9p(mount_t mp, int mntflags, __unused vfs_context_t ctx)
{
mount_9p *nmp;
vnode_t vp;
int e, flags;
TRACE();
nmp = MTO9P(mp);
flags = 0;
if(ISSET(mntflags,MNT_FORCE))
SET(flags, FORCECLOSE);
OSBitOrAtomic(F_UNMOUNTING, &nmp->flags);
vp = nmp->root;
if ((e=vflush(mp, vp, flags)))
goto error;
if (vnode_isinuse(vp, 1) && !ISSET(flags, FORCECLOSE)) {
e = EBUSY;
goto error;
}
clunk_9p(nmp, NTO9P(vp)->fid);
vnode_rele(vp);
vflush(mp, NULL, FORCECLOSE);
vfs_setfsprivate(mp, NULL);
disconnect_9p(nmp);
cancelrpcs_9p(nmp);
freemount_9p(nmp);
return 0;
error:
OSBitAndAtomic(~F_UNMOUNTING, &nmp->flags);
return e;
}
static errno_t
fuse_vfsop_unmount(mount_t mp, int mntflags, vfs_context_t context)
{
int err = 0;
int flags = 0;
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) {
flags |= FORCECLOSE;
}
data = fuse_get_mpdata(mp);
if (!data) {
panic("fuse4x: no mount private data in vfs_unmount");
}
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fdev = data->fdev;
if (fdata_dead_get(data)) {
/*
* 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->dataflags & FSESS_INITED)) {
flags |= FORCECLOSE;
log("fuse4x: forcing unmount on not-yet-alive file system\n");
fdata_set_dead(data);
}
fuse_rootvp = data->rootvp;
fuse_trace_printf("%s: Calling vflush(mp, fuse_rootvp, flags=0x%X);\n", __FUNCTION__, flags);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
err = vflush(mp, fuse_rootvp, flags);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
if (err) {
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
return err;
}
if (vnode_isinuse(fuse_rootvp, 1) && !(flags & FORCECLOSE)) {
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
return EBUSY;
}
if (fdata_dead_get(data)) {
goto alreadydead;
}
fdisp_init(&fdi, 0 /* no data to send along */);
fdisp_make(&fdi, FUSE_DESTROY, mp, FUSE_ROOT_ID, context);
fuse_trace_printf("%s: Waiting for reply from FUSE_DESTROY.\n", __FUNCTION__);
err = fdisp_wait_answ(&fdi);
fuse_trace_printf("%s: Reply received.\n", __FUNCTION__);
if (!err) {
fuse_ticket_drop(fdi.tick);
}
/*
* Note that dounmount() signals a VQ_UNMOUNT VFS event.
*/
fdata_set_dead(data);
alreadydead:
fuse_trace_printf("%s: Calling vnode_rele(fuse_rootp);\n", __FUNCTION__);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
vnode_rele(fuse_rootvp); /* We got this reference in fuse_vfsop_mount(). */
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
data->rootvp = NULLVP;
fuse_trace_printf("%s: Calling vflush(mp, NULLVP, FORCECLOSE);\n", __FUNCTION__);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
(void)vflush(mp, NULLVP, FORCECLOSE);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_lock(data->biglock);
#endif
fuse_trace_printf("%s: Done.\n", __FUNCTION__);
fuse_lck_mtx_lock(fdev->mtx);
vfs_setfsprivate(mp, NULL);
data->dataflags &= ~FSESS_MOUNTED;
OSAddAtomic(-1, (SInt32 *)&fuse_mount_count);
#if M_FUSE4X_ENABLE_BIGLOCK
fuse_biglock_unlock(data->biglock);
#endif
if (!(data->dataflags & FSESS_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;
}
/*
* This routine frees all the BSD context in uthread except the credential.
* It does not free the uthread structure as well
*/
void
uthread_cleanup(task_t task, void *uthread, void * bsd_info)
{
struct _select *sel;
uthread_t uth = (uthread_t)uthread;
proc_t p = (proc_t)bsd_info;
if (uth->uu_lowpri_window || uth->uu_throttle_info) {
/*
* task is marked as a low priority I/O type
* and we've somehow managed to not dismiss the throttle
* through the normal exit paths back to user space...
* no need to throttle this thread since its going away
* but we do need to update our bookeeping w/r to throttled threads
*
* Calling this routine will clean up any throttle info reference
* still inuse by the thread.
*/
throttle_lowpri_io(FALSE);
}
/*
* Per-thread audit state should never last beyond system
* call return. Since we don't audit the thread creation/
* removal, the thread state pointer should never be
* non-NULL when we get here.
*/
assert(uth->uu_ar == NULL);
sel = &uth->uu_select;
/* cleanup the select bit space */
if (sel->nbytes) {
FREE(sel->ibits, M_TEMP);
FREE(sel->obits, M_TEMP);
sel->nbytes = 0;
}
if (uth->uu_cdir) {
vnode_rele(uth->uu_cdir);
uth->uu_cdir = NULLVP;
}
if (uth->uu_allocsize && uth->uu_wqset){
kfree(uth->uu_wqset, uth->uu_allocsize);
sel->count = 0;
uth->uu_allocsize = 0;
uth->uu_wqset = 0;
sel->wql = 0;
}
if(uth->pth_name != NULL)
{
kfree(uth->pth_name, MAXTHREADNAMESIZE);
uth->pth_name = 0;
}
if ((task != kernel_task) && p) {
if (((uth->uu_flag & UT_VFORK) == UT_VFORK) && (uth->uu_proc != PROC_NULL)) {
vfork_exit_internal(uth->uu_proc, 0, 1);
}
/*
* Remove the thread from the process list and
* transfer [appropriate] pending signals to the process.
*/
if (get_bsdtask_info(task) == p) {
proc_lock(p);
TAILQ_REMOVE(&p->p_uthlist, uth, uu_list);
p->p_siglist |= (uth->uu_siglist & execmask & (~p->p_sigignore | sigcantmask));
proc_unlock(p);
}
#if CONFIG_DTRACE
struct dtrace_ptss_page_entry *tmpptr = uth->t_dtrace_scratch;
uth->t_dtrace_scratch = NULL;
if (tmpptr != NULL) {
dtrace_ptss_release_entry(p, tmpptr);
}
#endif
}
}
/*
* Routine: macx_swapoff
* Function:
* Syscall interface to remove a file from backing store
*/
int
macx_swapoff(
struct macx_swapoff_args *args)
{
__unused int flags = args->flags;
kern_return_t kr;
mach_port_t backing_store;
struct vnode *vp = 0;
struct nameidata nd, *ndp;
struct proc *p = current_proc();
int i;
int error;
boolean_t funnel_state;
vfs_context_t ctx = vfs_context_current();
AUDIT_MACH_SYSCALL_ENTER(AUE_SWAPOFF);
funnel_state = thread_funnel_set(kernel_flock, TRUE);
backing_store = NULL;
ndp = &nd;
if ((error = suser(kauth_cred_get(), 0)))
goto swapoff_bailout;
/*
* Get the vnode for the paging area.
*/
NDINIT(ndp, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNPATH1,
((IS_64BIT_PROCESS(p)) ? UIO_USERSPACE64 : UIO_USERSPACE32),
(user_addr_t) args->filename, ctx);
if ((error = namei(ndp)))
goto swapoff_bailout;
nameidone(ndp);
vp = ndp->ni_vp;
if (vp->v_type != VREG) {
error = EINVAL;
goto swapoff_bailout;
}
#if CONFIG_MACF
vnode_lock(vp);
error = mac_system_check_swapoff(vfs_context_ucred(ctx), vp);
vnode_unlock(vp);
if (error)
goto swapoff_bailout;
#endif
for(i = 0; i < MAX_BACKING_STORE; i++) {
if(bs_port_table[i].vp == vp) {
break;
}
}
if (i == MAX_BACKING_STORE) {
error = EINVAL;
goto swapoff_bailout;
}
backing_store = (mach_port_t)bs_port_table[i].bs;
kr = default_pager_backing_store_delete(backing_store);
switch (kr) {
case KERN_SUCCESS:
error = 0;
bs_port_table[i].vp = 0;
/* This vnode is no longer used for swapfile */
vnode_lock_spin(vp);
CLR(vp->v_flag, VSWAP);
vnode_unlock(vp);
/* get rid of macx_swapon() "long term" reference */
vnode_rele(vp);
break;
case KERN_FAILURE:
error = EAGAIN;
break;
default:
error = EAGAIN;
break;
}
swapoff_bailout:
/* get rid of macx_swapoff() namei() reference */
if (vp)
vnode_put(vp);
(void) thread_funnel_set(kernel_flock, FALSE);
AUDIT_MACH_SYSCALL_EXIT(error);
return(error);
}
/*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);
}
/*
* Device close routine
*/
int
spec_close(struct vnop_close_args *ap)
{
struct vnode *vp = ap->a_vp;
dev_t dev = vp->v_rdev;
int (*devclose)(dev_t, int, int, struct proc *);
int mode, error;
int flags = ap->a_fflag;
struct proc *p = vfs_context_proc(ap->a_context);
struct session *sessp;
switch (vp->v_type) {
case VCHR:
/*
* Hack: a tty device that is a controlling terminal
* has a reference from the session structure.
* We cannot easily tell that a character device is
* a controlling terminal, unless it is the closing
* process' controlling terminal. In that case,
* if the reference count is 1 (this is the very
* last close)
*/
sessp = proc_session(p);
if (sessp != SESSION_NULL) {
if ((vcount(vp) == 1) &&
(vp == sessp->s_ttyvp)) {
session_lock(sessp);
sessp->s_ttyvp = NULL;
sessp->s_ttyvid = 0;
sessp->s_ttyp = TTY_NULL;
sessp->s_ttypgrpid = NO_PID;
session_unlock(sessp);
vnode_rele(vp);
}
session_rele(sessp);
}
devclose = cdevsw[major(dev)].d_close;
mode = S_IFCHR;
/*
* close on last reference or on vnode revoke call
*/
if ((flags & IO_REVOKE) != 0)
break;
if (vcount(vp) > 0)
return (0);
break;
case VBLK:
/*
* Since every use (buffer, vnode, swap, blockmap)
* holds a reference to the vnode, and because we mark
* any other vnodes that alias this device, when the
* sum of the reference counts on all the aliased
* vnodes descends to zero, we are on last close.
*/
if (vcount(vp) > 0)
return (0);
/*
* On last close of a block device (that isn't mounted)
* we must invalidate any in core blocks, so that
* we can, for instance, change floppy disks.
*/
if ((error = spec_fsync_internal(vp, MNT_WAIT, ap->a_context)))
return (error);
error = buf_invalidateblks(vp, BUF_WRITE_DATA, 0, 0);
if (error)
return (error);
devclose = bdevsw[major(dev)].d_close;
mode = S_IFBLK;
break;
default:
panic("spec_close: not special");
return(EBADF);
}
return ((*devclose)(dev, flags, mode, p));
}
/*
* Free reference to null layer
*/
static int
nullfs_unmount(struct mount * mp, int mntflags, __unused vfs_context_t ctx)
{
struct null_mount * mntdata;
struct vnode * vp;
int error, flags;
NULLFSDEBUG("nullfs_unmount: mp = %p\n", (void *)mp);
/* check entitlement or superuser*/
if (!IOTaskHasEntitlement(current_task(), NULLFS_ENTITLEMENT) &&
vfs_context_suser(ctx) != 0) {
return EPERM;
}
if (mntflags & MNT_FORCE) {
flags = FORCECLOSE;
} else {
flags = 0;
}
mntdata = MOUNTTONULLMOUNT(mp);
vp = mntdata->nullm_rootvp;
// release our reference on the root before flushing.
// it will get pulled out of the mount structure by reclaim
vnode_getalways(vp);
error = vflush(mp, vp, flags);
if (error)
{
vnode_put(vp);
return (error);
}
if (vnode_isinuse(vp,1) && flags == 0)
{
vnode_put(vp);
return EBUSY;
}
vnode_rele(vp); // Drop reference taken by nullfs_mount
vnode_put(vp); // Drop ref taken above
//Force close to get rid of the last vnode
(void)vflush(mp, NULL, FORCECLOSE);
/* no more vnodes, so tear down the mountpoint */
lck_mtx_lock(&mntdata->nullm_lock);
vfs_setfsprivate(mp, NULL);
vnode_getalways(mntdata->nullm_lowerrootvp);
vnode_rele(mntdata->nullm_lowerrootvp);
vnode_put(mntdata->nullm_lowerrootvp);
lck_mtx_unlock(&mntdata->nullm_lock);
nullfs_destroy_lck(&mntdata->nullm_lock);
FREE(mntdata, M_TEMP);
uint64_t vflags = vfs_flags(mp);
vfs_setflags(mp, vflags & ~MNT_LOCAL);
return (0);
}
/*
* Mount null layer
*/
static int
nullfs_mount(struct mount * mp, __unused vnode_t devvp, user_addr_t user_data, vfs_context_t ctx)
{
int error = 0;
struct vnode *lowerrootvp = NULL, *vp = NULL;
struct vfsstatfs * sp = NULL;
struct null_mount * xmp = NULL;
char data[MAXPATHLEN];
size_t count;
struct vfs_attr vfa;
/* set defaults (arbitrary since this file system is readonly) */
uint32_t bsize = BLKDEV_IOSIZE;
size_t iosize = BLKDEV_IOSIZE;
uint64_t blocks = 4711 * 4711;
uint64_t bfree = 0;
uint64_t bavail = 0;
uint64_t bused = 4711;
uint64_t files = 4711;
uint64_t ffree = 0;
kauth_cred_t cred = vfs_context_ucred(ctx);
NULLFSDEBUG("nullfs_mount(mp = %p) %llx\n", (void *)mp, vfs_flags(mp));
if (vfs_flags(mp) & MNT_ROOTFS)
return (EOPNOTSUPP);
/*
* Update is a no-op
*/
if (vfs_isupdate(mp)) {
return ENOTSUP;
}
/* check entitlement */
if (!IOTaskHasEntitlement(current_task(), NULLFS_ENTITLEMENT)) {
return EPERM;
}
/*
* Get argument
*/
error = copyinstr(user_data, data, MAXPATHLEN - 1, &count);
if (error) {
NULLFSDEBUG("nullfs: error copying data form user %d\n", error);
goto error;
}
/* This could happen if the system is configured for 32 bit inodes instead of
* 64 bit */
if (count > MAX_MNT_FROM_LENGTH) {
error = EINVAL;
NULLFSDEBUG("nullfs: path to translocate too large for this system %d vs %d\n", count, MAX_MNT_FROM_LENGTH);
goto error;
}
error = vnode_lookup(data, 0, &lowerrootvp, ctx);
if (error) {
NULLFSDEBUG("lookup %s -> %d\n", data, error);
goto error;
}
/* lowervrootvp has an iocount after vnode_lookup, drop that for a usecount.
Keep this to signal what we want to keep around the thing we are mirroring.
Drop it in unmount.*/
error = vnode_ref(lowerrootvp);
vnode_put(lowerrootvp);
if (error)
{
// If vnode_ref failed, then null it out so it can't be used anymore in cleanup.
lowerrootvp = NULL;
goto error;
}
NULLFSDEBUG("mount %s\n", data);
MALLOC(xmp, struct null_mount *, sizeof(*xmp), M_TEMP, M_WAITOK | M_ZERO);
if (xmp == NULL) {
error = ENOMEM;
goto error;
}
/*
* Save reference to underlying FS
*/
xmp->nullm_lowerrootvp = lowerrootvp;
xmp->nullm_lowerrootvid = vnode_vid(lowerrootvp);
error = null_getnewvnode(mp, NULL, NULL, &vp, NULL, 1);
if (error) {
goto error;
}
/* vp has an iocount on it from vnode_create. drop that for a usecount. This
* is our root vnode so we drop the ref in unmount
*
* Assuming for now that because we created this vnode and we aren't finished mounting we can get a ref*/
vnode_ref(vp);
vnode_put(vp);
error = nullfs_init_lck(&xmp->nullm_lock);
if (error) {
goto error;
}
xmp->nullm_rootvp = vp;
/* read the flags the user set, but then ignore some of them, we will only
allow them if they are set on the lower file system */
uint64_t flags = vfs_flags(mp) & (~(MNT_IGNORE_OWNERSHIP | MNT_LOCAL));
uint64_t lowerflags = vfs_flags(vnode_mount(lowerrootvp)) & (MNT_LOCAL | MNT_QUARANTINE | MNT_IGNORE_OWNERSHIP | MNT_NOEXEC);
if (lowerflags) {
flags |= lowerflags;
}
/* force these flags */
flags |= (MNT_DONTBROWSE | MNT_MULTILABEL | MNT_NOSUID | MNT_RDONLY);
vfs_setflags(mp, flags);
vfs_setfsprivate(mp, xmp);
vfs_getnewfsid(mp);
vfs_setlocklocal(mp);
/* fill in the stat block */
sp = vfs_statfs(mp);
strlcpy(sp->f_mntfromname, data, MAX_MNT_FROM_LENGTH);
sp->f_flags = flags;
xmp->nullm_flags = NULLM_CASEINSENSITIVE; /* default to case insensitive */
error = nullfs_vfs_getlowerattr(vnode_mount(lowerrootvp), &vfa, ctx);
if (error == 0) {
if (VFSATTR_IS_SUPPORTED(&vfa, f_bsize)) {
bsize = vfa.f_bsize;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_iosize)) {
iosize = vfa.f_iosize;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_blocks)) {
blocks = vfa.f_blocks;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_bfree)) {
bfree = vfa.f_bfree;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_bavail)) {
bavail = vfa.f_bavail;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_bused)) {
bused = vfa.f_bused;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_files)) {
files = vfa.f_files;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_ffree)) {
ffree = vfa.f_ffree;
}
if (VFSATTR_IS_SUPPORTED(&vfa, f_capabilities)) {
if ((vfa.f_capabilities.capabilities[VOL_CAPABILITIES_FORMAT] & (VOL_CAP_FMT_CASE_SENSITIVE)) &&
(vfa.f_capabilities.valid[VOL_CAPABILITIES_FORMAT] & (VOL_CAP_FMT_CASE_SENSITIVE))) {
xmp->nullm_flags &= ~NULLM_CASEINSENSITIVE;
}
}
} else {
goto error;
}
sp->f_bsize = bsize;
sp->f_iosize = iosize;
sp->f_blocks = blocks;
sp->f_bfree = bfree;
sp->f_bavail = bavail;
sp->f_bused = bused;
sp->f_files = files;
sp->f_ffree = ffree;
/* Associate the mac label information from the mirrored filesystem with the
* mirror */
MAC_PERFORM(mount_label_associate, cred, vnode_mount(lowerrootvp), vfs_mntlabel(mp));
NULLFSDEBUG("nullfs_mount: lower %s, alias at %s\n", sp->f_mntfromname, sp->f_mntonname);
return (0);
error:
if (xmp) {
FREE(xmp, M_TEMP);
}
if (lowerrootvp) {
vnode_getwithref(lowerrootvp);
vnode_rele(lowerrootvp);
vnode_put(lowerrootvp);
}
if (vp) {
/* we made the root vnode but the mount is failed, so clean it up */
vnode_getwithref(vp);
vnode_rele(vp);
/* give vp back */
vnode_recycle(vp);
vnode_put(vp);
}
return error;
}
