/*ARGSUSED*/
static int
lo_mount(struct vfs *vfsp,
struct vnode *vp,
struct mounta *uap,
struct cred *cr)
{
int error;
struct vnode *srootvp = NULL; /* the server's root */
struct vnode *realrootvp;
struct loinfo *li;
int nodev;
nodev = vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL);
if ((error = secpolicy_fs_mount(cr, vp, vfsp)) != 0)
return (EPERM);
/*
* Loopback devices which get "nodevices" added can be done without
* "nodevices" set because we cannot import devices into a zone
* with loopback. Note that we have all zone privileges when
* this happens; if not, we'd have gotten "nosuid".
*/
if (!nodev && vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
vfs_setmntopt(vfsp, MNTOPT_DEVICES, NULL, VFS_NODISPLAY);
mutex_enter(&vp->v_lock);
if (!(uap->flags & MS_OVERLAY) &&
(vp->v_count != 1 || (vp->v_flag & VROOT))) {
mutex_exit(&vp->v_lock);
return (EBUSY);
}
mutex_exit(&vp->v_lock);
/*
* Find real root, and make vfs point to real vfs
*/
if (error = lookupname(uap->spec, (uap->flags & MS_SYSSPACE) ?
UIO_SYSSPACE : UIO_USERSPACE, FOLLOW, NULLVPP, &realrootvp))
return (error);
/*
* Enforce MAC policy if needed.
*
* Loopback mounts must not allow writing up. The dominance test
* is intended to prevent a global zone caller from accidentally
* creating write-up conditions between two labeled zones.
* Local zones can't violate MAC on their own without help from
* the global zone because they can't name a pathname that
* they don't already have.
*
* The special case check for the NET_MAC_AWARE process flag is
* to support the case of the automounter in the global zone. We
* permit automounting of local zone directories such as home
* directories, into the global zone as required by setlabel,
* zonecopy, and saving of desktop sessions. Such mounts are
* trusted not to expose the contents of one zone's directories
* to another by leaking them through the global zone.
*/
if (is_system_labeled() && crgetzoneid(cr) == GLOBAL_ZONEID) {
char specname[MAXPATHLEN];
zone_t *from_zptr;
zone_t *to_zptr;
if (vnodetopath(NULL, realrootvp, specname,
sizeof (specname), CRED()) != 0) {
VN_RELE(realrootvp);
return (EACCES);
}
from_zptr = zone_find_by_path(specname);
to_zptr = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
/*
* Special case for zone devfs: the zone for /dev will
* incorrectly appear as the global zone since it's not
* under the zone rootpath. So for zone devfs check allow
* read-write mounts.
*
* Second special case for scratch zones used for Live Upgrade:
* this is used to mount the zone's root from /root to /a in
* the scratch zone. As with the other special case, this
* appears to be outside of the zone because it's not under
* the zone rootpath, which is $ZONEPATH/lu in the scratch
* zone case.
*/
if (from_zptr != to_zptr &&
!(to_zptr->zone_flags & ZF_IS_SCRATCH)) {
/*
* We know at this point that the labels aren't equal
* because the zone pointers aren't equal, and zones
* can't share a label.
*
* If the source is the global zone then making
* it available to a local zone must be done in
* read-only mode as the label will become admin_low.
*
* If it is a mount between local zones then if
* the current process is in the global zone and has
* the NET_MAC_AWARE flag, then regular read-write
* access is allowed. If it's in some other zone, but
* the label on the mount point dominates the original
* source, then allow the mount as read-only
* ("read-down").
*/
if (from_zptr->zone_id == GLOBAL_ZONEID) {
/* make the mount read-only */
vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
} else { /* cross-zone mount */
if (to_zptr->zone_id == GLOBAL_ZONEID &&
/* LINTED: no consequent */
getpflags(NET_MAC_AWARE, cr) != 0) {
/* Allow the mount as read-write */
} else if (bldominates(
label2bslabel(to_zptr->zone_slabel),
label2bslabel(from_zptr->zone_slabel))) {
/* make the mount read-only */
vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
} else {
VN_RELE(realrootvp);
zone_rele(to_zptr);
zone_rele(from_zptr);
return (EACCES);
}
}
}
zone_rele(to_zptr);
zone_rele(from_zptr);
}
/*
* realrootvp may be an AUTOFS node, in which case we
* perform a VOP_ACCESS() to trigger the mount of the
* intended filesystem, so we loopback mount the intended
* filesystem instead of the AUTOFS filesystem.
*/
(void) VOP_ACCESS(realrootvp, 0, 0, cr, NULL);
/*
* We're interested in the top most filesystem.
* This is specially important when uap->spec is a trigger
* AUTOFS node, since we're really interested in mounting the
* filesystem AUTOFS mounted as result of the VOP_ACCESS()
* call not the AUTOFS node itself.
*/
if (vn_mountedvfs(realrootvp) != NULL) {
if (error = traverse(&realrootvp)) {
VN_RELE(realrootvp);
return (error);
}
}
/*
* Allocate a vfs info struct and attach it
*/
li = kmem_zalloc(sizeof (struct loinfo), KM_SLEEP);
li->li_realvfs = realrootvp->v_vfsp;
li->li_mountvfs = vfsp;
/*
* Set mount flags to be inherited by loopback vfs's
*/
if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
li->li_mflag |= VFS_RDONLY;
}
if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
li->li_mflag |= (VFS_NOSETUID|VFS_NODEVICES);
}
if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
li->li_mflag |= VFS_NODEVICES;
}
if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
li->li_mflag |= VFS_NOSETUID;
}
/*
* Permissive flags are added to the "deny" bitmap.
*/
if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
li->li_dflag |= VFS_XATTR;
}
if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
li->li_dflag |= VFS_NBMAND;
}
/*
* Propagate inheritable mount flags from the real vfs.
*/
if ((li->li_realvfs->vfs_flag & VFS_RDONLY) &&
!vfs_optionisset(vfsp, MNTOPT_RO, NULL))
vfs_setmntopt(vfsp, MNTOPT_RO, NULL,
VFS_NODISPLAY);
if ((li->li_realvfs->vfs_flag & VFS_NOSETUID) &&
!vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL))
vfs_setmntopt(vfsp, MNTOPT_NOSETUID, NULL,
VFS_NODISPLAY);
if ((li->li_realvfs->vfs_flag & VFS_NODEVICES) &&
!vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
vfs_setmntopt(vfsp, MNTOPT_NODEVICES, NULL,
VFS_NODISPLAY);
/*
* Permissive flags such as VFS_XATTR, as opposed to restrictive flags
* such as VFS_RDONLY, are handled differently. An explicit
* MNTOPT_NOXATTR should override the underlying filesystem's VFS_XATTR.
*/
if ((li->li_realvfs->vfs_flag & VFS_XATTR) &&
!vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL) &&
!vfs_optionisset(vfsp, MNTOPT_XATTR, NULL))
vfs_setmntopt(vfsp, MNTOPT_XATTR, NULL,
VFS_NODISPLAY);
if ((li->li_realvfs->vfs_flag & VFS_NBMAND) &&
!vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL) &&
!vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL))
vfs_setmntopt(vfsp, MNTOPT_NBMAND, NULL,
VFS_NODISPLAY);
li->li_refct = 0;
vfsp->vfs_data = (caddr_t)li;
vfsp->vfs_bcount = 0;
vfsp->vfs_fstype = lofsfstype;
vfsp->vfs_bsize = li->li_realvfs->vfs_bsize;
vfsp->vfs_dev = li->li_realvfs->vfs_dev;
vfsp->vfs_fsid.val[0] = li->li_realvfs->vfs_fsid.val[0];
vfsp->vfs_fsid.val[1] = li->li_realvfs->vfs_fsid.val[1];
if (vfs_optionisset(vfsp, MNTOPT_LOFS_NOSUB, NULL)) {
li->li_flag |= LO_NOSUB;
}
/*
* Propagate any VFS features
*/
vfs_propagate_features(li->li_realvfs, vfsp);
/*
* Setup the hashtable. If the root of this mount isn't a directory,
* there's no point in allocating a large hashtable. A table with one
* bucket is sufficient.
*/
if (realrootvp->v_type != VDIR)
lsetup(li, 1);
else
lsetup(li, 0);
/*
* Make the root vnode
*/
srootvp = makelonode(realrootvp, li, 0);
srootvp->v_flag |= VROOT;
li->li_rootvp = srootvp;
#ifdef LODEBUG
lo_dprint(4, "lo_mount: vfs %p realvfs %p root %p realroot %p li %p\n",
vfsp, li->li_realvfs, srootvp, realrootvp, li);
#endif
return (0);
}
/*
* Get/Make vfs structure for given real vfs
*/
static struct vfs *
makelfsnode(struct vfs *vfsp, struct loinfo *li)
{
struct lfsnode *lfs;
struct lfsnode *tlfs;
/*
* Don't grab any locks for the fast (common) case.
*/
if (vfsp == li->li_realvfs)
return (li->li_mountvfs);
ASSERT(li->li_refct > 0);
mutex_enter(&li->li_lfslock);
if ((lfs = lfsfind(vfsp, li)) == NULL) {
mutex_exit(&li->li_lfslock);
lfs = kmem_zalloc(sizeof (*lfs), KM_SLEEP);
mutex_enter(&li->li_lfslock);
if ((tlfs = lfsfind(vfsp, li)) != NULL) {
kmem_free(lfs, sizeof (*lfs));
lfs = tlfs;
goto found_lfs;
}
lfs->lfs_realvfs = vfsp;
/*
* Even though the lfsnode is strictly speaking a private
* implementation detail of lofs, it should behave as a regular
* vfs_t for the benefit of the rest of the kernel.
*/
VFS_INIT(&lfs->lfs_vfs, lo_vfsops, (caddr_t)li);
lfs->lfs_vfs.vfs_fstype = li->li_mountvfs->vfs_fstype;
lfs->lfs_vfs.vfs_flag =
((vfsp->vfs_flag | li->li_mflag) & ~li->li_dflag) &
INHERIT_VFS_FLAG;
lfs->lfs_vfs.vfs_bsize = vfsp->vfs_bsize;
lfs->lfs_vfs.vfs_dev = vfsp->vfs_dev;
lfs->lfs_vfs.vfs_fsid = vfsp->vfs_fsid;
if (vfsp->vfs_mntpt != NULL) {
lfs->lfs_vfs.vfs_mntpt = vfs_getmntpoint(vfsp);
/* Leave a reference to the mountpoint */
}
(void) VFS_ROOT(vfsp, &lfs->lfs_realrootvp);
/*
* We use 1 instead of 0 as the value to associate with
* an idle lfs_vfs. This is to prevent VFS_RELE()
* trying to kmem_free() our lfs_t (which is the wrong
* size).
*/
VFS_HOLD(&lfs->lfs_vfs);
lfs->lfs_next = li->li_lfs;
li->li_lfs = lfs;
vfs_propagate_features(vfsp, &lfs->lfs_vfs);
}
found_lfs:
VFS_HOLD(&lfs->lfs_vfs);
mutex_exit(&li->li_lfslock);
return (&lfs->lfs_vfs);
}
