/*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);
}
/*
* smbfs mount vfsop
* Set up mount info record and attach it to vfs struct.
*/
static int
smbfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
{
char *data = uap->dataptr;
int error;
smbnode_t *rtnp = NULL; /* root of this fs */
smbmntinfo_t *smi = NULL;
dev_t smbfs_dev;
int version;
int devfd;
zone_t *zone = curproc->p_zone;
zone_t *mntzone = NULL;
smb_share_t *ssp = NULL;
smb_cred_t scred;
int flags, sec;
STRUCT_DECL(smbfs_args, args); /* smbfs mount arguments */
if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
return (error);
if (mvp->v_type != VDIR)
return (ENOTDIR);
/*
* get arguments
*
* uap->datalen might be different from sizeof (args)
* in a compatible situation.
*/
STRUCT_INIT(args, get_udatamodel());
bzero(STRUCT_BUF(args), SIZEOF_STRUCT(smbfs_args, DATAMODEL_NATIVE));
if (copyin(data, STRUCT_BUF(args), MIN(uap->datalen,
SIZEOF_STRUCT(smbfs_args, DATAMODEL_NATIVE))))
return (EFAULT);
/*
* Check mount program version
*/
version = STRUCT_FGET(args, version);
if (version != SMBFS_VERSION) {
cmn_err(CE_WARN, "mount version mismatch:"
" kernel=%d, mount=%d\n",
SMBFS_VERSION, version);
return (EINVAL);
}
/*
* Deal with re-mount requests.
*/
if (uap->flags & MS_REMOUNT) {
cmn_err(CE_WARN, "MS_REMOUNT not implemented");
return (ENOTSUP);
}
/*
* Check for busy
*/
mutex_enter(&mvp->v_lock);
if (!(uap->flags & MS_OVERLAY) &&
(mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
mutex_exit(&mvp->v_lock);
return (EBUSY);
}
mutex_exit(&mvp->v_lock);
/*
* Get the "share" from the netsmb driver (ssp).
* It is returned with a "ref" (hold) for us.
* Release this hold: at errout below, or in
* smbfs_freevfs().
*/
devfd = STRUCT_FGET(args, devfd);
error = smb_dev2share(devfd, &ssp);
if (error) {
cmn_err(CE_WARN, "invalid device handle %d (%d)\n",
devfd, error);
return (error);
}
/*
* Use "goto errout" from here on.
* See: ssp, smi, rtnp, mntzone
*/
/*
* Determine the zone we're being mounted into.
*/
zone_hold(mntzone = zone); /* start with this assumption */
if (getzoneid() == GLOBAL_ZONEID) {
zone_rele(mntzone);
mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
ASSERT(mntzone != NULL);
if (mntzone != zone) {
error = EBUSY;
goto errout;
}
}
/*
* Stop the mount from going any further if the zone is going away.
*/
if (zone_status_get(mntzone) >= ZONE_IS_SHUTTING_DOWN) {
error = EBUSY;
goto errout;
}
/*
* On a Trusted Extensions client, we may have to force read-only
* for read-down mounts.
*/
if (is_system_labeled()) {
void *addr;
int ipvers = 0;
struct smb_vc *vcp;
vcp = SSTOVC(ssp);
addr = smb_vc_getipaddr(vcp, &ipvers);
error = smbfs_mount_label_policy(vfsp, addr, ipvers, cr);
if (error > 0)
goto errout;
if (error == -1) {
/* change mount to read-only to prevent write-down */
vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
}
}
/* Prevent unload. */
atomic_inc_32(&smbfs_mountcount);
/*
* Create a mount record and link it to the vfs struct.
* No more possiblities for errors from here on.
* Tear-down of this stuff is in smbfs_free_smi()
*
* Compare with NFS: nfsrootvp()
*/
smi = kmem_zalloc(sizeof (*smi), KM_SLEEP);
mutex_init(&smi->smi_lock, NULL, MUTEX_DEFAULT, NULL);
cv_init(&smi->smi_statvfs_cv, NULL, CV_DEFAULT, NULL);
rw_init(&smi->smi_hash_lk, NULL, RW_DEFAULT, NULL);
smbfs_init_hash_avl(&smi->smi_hash_avl);
smi->smi_share = ssp;
ssp = NULL;
/*
* Convert the anonymous zone hold acquired via zone_hold() above
* into a zone reference.
*/
zone_init_ref(&smi->smi_zone_ref);
zone_hold_ref(mntzone, &smi->smi_zone_ref, ZONE_REF_SMBFS);
zone_rele(mntzone);
mntzone = NULL;
/*
* Initialize option defaults
*/
smi->smi_flags = SMI_LLOCK;
smi->smi_acregmin = SEC2HR(SMBFS_ACREGMIN);
smi->smi_acregmax = SEC2HR(SMBFS_ACREGMAX);
smi->smi_acdirmin = SEC2HR(SMBFS_ACDIRMIN);
smi->smi_acdirmax = SEC2HR(SMBFS_ACDIRMAX);
/*
* All "generic" mount options have already been
* handled in vfs.c:domount() - see mntopts stuff.
* Query generic options using vfs_optionisset().
*/
if (vfs_optionisset(vfsp, MNTOPT_INTR, NULL))
smi->smi_flags |= SMI_INT;
if (vfs_optionisset(vfsp, MNTOPT_ACL, NULL))
smi->smi_flags |= SMI_ACL;
/*
* Get the mount options that come in as smbfs_args,
* starting with args.flags (SMBFS_MF_xxx)
*/
flags = STRUCT_FGET(args, flags);
smi->smi_uid = STRUCT_FGET(args, uid);
smi->smi_gid = STRUCT_FGET(args, gid);
smi->smi_fmode = STRUCT_FGET(args, file_mode) & 0777;
smi->smi_dmode = STRUCT_FGET(args, dir_mode) & 0777;
/*
* Hande the SMBFS_MF_xxx flags.
*/
if (flags & SMBFS_MF_NOAC)
smi->smi_flags |= SMI_NOAC;
if (flags & SMBFS_MF_ACREGMIN) {
sec = STRUCT_FGET(args, acregmin);
if (sec < 0 || sec > SMBFS_ACMINMAX)
sec = SMBFS_ACMINMAX;
smi->smi_acregmin = SEC2HR(sec);
}
if (flags & SMBFS_MF_ACREGMAX) {
sec = STRUCT_FGET(args, acregmax);
if (sec < 0 || sec > SMBFS_ACMAXMAX)
sec = SMBFS_ACMAXMAX;
smi->smi_acregmax = SEC2HR(sec);
}
if (flags & SMBFS_MF_ACDIRMIN) {
sec = STRUCT_FGET(args, acdirmin);
if (sec < 0 || sec > SMBFS_ACMINMAX)
sec = SMBFS_ACMINMAX;
smi->smi_acdirmin = SEC2HR(sec);
}
if (flags & SMBFS_MF_ACDIRMAX) {
sec = STRUCT_FGET(args, acdirmax);
if (sec < 0 || sec > SMBFS_ACMAXMAX)
sec = SMBFS_ACMAXMAX;
smi->smi_acdirmax = SEC2HR(sec);
}
/*
* Get attributes of the remote file system,
* i.e. ACL support, named streams, etc.
*/
smb_credinit(&scred, cr);
error = smbfs_smb_qfsattr(smi->smi_share, &smi->smi_fsa, &scred);
smb_credrele(&scred);
if (error) {
SMBVDEBUG("smbfs_smb_qfsattr error %d\n", error);
}
/*
* We enable XATTR by default (via smbfs_mntopts)
* but if the share does not support named streams,
* force the NOXATTR option (also clears XATTR).
* Caller will set or clear VFS_XATTR after this.
*/
if ((smi->smi_fsattr & FILE_NAMED_STREAMS) == 0)
vfs_setmntopt(vfsp, MNTOPT_NOXATTR, NULL, 0);
/*
* Ditto ACLs (disable if not supported on this share)
*/
if ((smi->smi_fsattr & FILE_PERSISTENT_ACLS) == 0) {
vfs_setmntopt(vfsp, MNTOPT_NOACL, NULL, 0);
smi->smi_flags &= ~SMI_ACL;
}
/*
* Assign a unique device id to the mount
*/
mutex_enter(&smbfs_minor_lock);
do {
smbfs_minor = (smbfs_minor + 1) & MAXMIN32;
smbfs_dev = makedevice(smbfs_major, smbfs_minor);
} while (vfs_devismounted(smbfs_dev));
mutex_exit(&smbfs_minor_lock);
vfsp->vfs_dev = smbfs_dev;
vfs_make_fsid(&vfsp->vfs_fsid, smbfs_dev, smbfsfstyp);
vfsp->vfs_data = (caddr_t)smi;
vfsp->vfs_fstype = smbfsfstyp;
vfsp->vfs_bsize = MAXBSIZE;
vfsp->vfs_bcount = 0;
smi->smi_vfsp = vfsp;
smbfs_zonelist_add(smi); /* undo in smbfs_freevfs */
/*
* Create the root vnode, which we need in unmount
* for the call to smbfs_check_table(), etc.
* Release this hold in smbfs_unmount.
*/
rtnp = smbfs_node_findcreate(smi, "\\", 1, NULL, 0, 0,
&smbfs_fattr0);
ASSERT(rtnp != NULL);
rtnp->r_vnode->v_type = VDIR;
rtnp->r_vnode->v_flag |= VROOT;
smi->smi_root = rtnp;
/*
* NFS does other stuff here too:
* async worker threads
* init kstats
*
* End of code from NFS nfsrootvp()
*/
return (0);
errout:
vfsp->vfs_data = NULL;
if (smi != NULL)
smbfs_free_smi(smi);
if (mntzone != NULL)
zone_rele(mntzone);
if (ssp != NULL)
smb_share_rele(ssp);
return (error);
}
