/*
* There's no protocol today to obtain the label from the server.
* So we rely on conventions: zones, zone names, and zone paths
* must match across TX servers and their TX clients. Now use
* the exported name to find the equivalent local zone and its
* label. Caller is responsible for doing a label_rele of the
* returned ts_label.
*/
ts_label_t *
getflabel_cipso(vfs_t *vfsp)
{
zone_t *reszone;
zone_t *new_reszone;
char *nfspath, *respath;
refstr_t *resource_ref;
boolean_t treat_abs = B_FALSE;
if (vfsp->vfs_resource == NULL)
return (NULL); /* error */
resource_ref = vfs_getresource(vfsp);
nfspath = (char *)refstr_value(resource_ref);
respath = strchr(nfspath, ':'); /* skip server name */
if (respath)
respath++; /* skip over ":" */
if (*respath != '/') {
/* treat path as absolute but it doesn't have leading '/' */
treat_abs = B_TRUE;
}
reszone = zone_find_by_any_path(respath, treat_abs);
if (reszone == global_zone) {
refstr_rele(resource_ref);
label_hold(l_admin_low);
zone_rele(reszone);
return (l_admin_low);
}
/*
* Skip over zonepath (not including "root"), e.g. /zone/internal
*/
respath += reszone->zone_rootpathlen - 7;
if (treat_abs)
respath--; /* no leading '/' to skip */
if (strncmp(respath, "/root/", 6) == 0) {
/* Check if we now have something like "/zone/public/" */
respath += 5; /* skip "/root" first */
new_reszone = zone_find_by_any_path(respath, B_FALSE);
if (new_reszone != global_zone) {
zone_rele(reszone);
reszone = new_reszone;
} else {
zone_rele(new_reszone);
}
}
refstr_rele(resource_ref);
label_hold(reszone->zone_slabel);
zone_rele(reszone);
return (reszone->zone_slabel);
}
static int
putacct(idtype_t idtype, id_t id, void *buf, size_t bufsize, int flags)
{
int error;
taskid_t tkid;
proc_t *p;
task_t *tk;
void *kbuf;
struct exacct_globals *acg;
if (bufsize == 0 || bufsize > EXACCT_MAX_BUFSIZE)
return (set_errno(EINVAL));
kbuf = kmem_alloc(bufsize, KM_SLEEP);
if (copyin(buf, kbuf, bufsize) != 0) {
error = EFAULT;
goto out;
}
acg = zone_getspecific(exacct_zone_key, curproc->p_zone);
switch (idtype) {
case P_PID:
mutex_enter(&pidlock);
if ((p = prfind(id)) == NULL) {
mutex_exit(&pidlock);
error = ESRCH;
} else {
zone_t *zone = p->p_zone;
tkid = p->p_task->tk_tkid;
zone_hold(zone);
mutex_exit(&pidlock);
error = exacct_tag_proc(&acg->ac_proc, id, tkid, kbuf,
bufsize, flags, zone->zone_nodename);
zone_rele(zone);
}
break;
case P_TASKID:
if ((tk = task_hold_by_id(id)) != NULL) {
error = exacct_tag_task(&acg->ac_task, tk, kbuf,
bufsize, flags);
task_rele(tk);
} else {
error = ESRCH;
}
break;
default:
error = EINVAL;
break;
}
out:
kmem_free(kbuf, bufsize);
return (error == 0 ? error : set_errno(error));
}
/*
* Set the fields in the 'target' clone to the specified values.
* Then, look at all clones to determine which message types are
* currently active and which clone is the primary console queue.
* If the primary console queue changes to or from the backlog
* queue, copy all messages from backlog to primary or vice versa.
*/
void
log_update(log_t *target, queue_t *q, short flags, log_filter_t *filter)
{
log_t *lp;
short active = SL_CONSOLE;
zone_t *zptr = NULL;
log_zone_t *lzp;
zoneid_t zoneid = target->log_zoneid;
int i;
log_enter();
if (q != NULL)
target->log_q = q;
target->log_wanted = filter;
target->log_flags = flags;
target->log_overflow = 0;
/*
* Need to special case the global zone here since this may be
* called before zone_init.
*/
if (zoneid == GLOBAL_ZONEID) {
lzp = &log_global;
} else if ((zptr = zone_find_by_id(zoneid)) == NULL) {
log_exit();
return; /* zone is being destroyed, ignore update */
} else {
lzp = zone_getspecific(log_zone_key, zptr);
}
ASSERT(lzp != NULL);
for (i = LOG_LOGMAXIDX; i >= LOG_LOGMINIDX; i--) {
lp = &lzp->lz_clones[i];
if (zoneid == GLOBAL_ZONEID && (lp->log_flags & SL_CONSOLE))
log_consq = lp->log_q;
active |= lp->log_flags;
}
lzp->lz_active = active;
if (zptr)
zone_rele(zptr);
if (log_consq == target->log_q) {
if (flags & SL_CONSOLE)
log_conswitch(&log_backlog, target);
else
log_conswitch(target, &log_backlog);
}
target->log_q = q;
log_exit();
}
/*
* Find a stack instance given the zoneid.
* Increases the reference count if found; caller must do a
* netstack_rele().
*
* If there is no exact match then assume the shared stack instance
* matches.
*
* Skip the unitialized ones.
*/
netstack_t *
netstack_find_by_zoneid(zoneid_t zoneid)
{
netstack_t *ns;
zone_t *zone;
zone = zone_find_by_id(zoneid);
if (zone == NULL)
return (NULL);
ns = zone->zone_netstack;
ASSERT(ns != NULL);
if (ns->netstack_flags & (NSF_UNINIT|NSF_CLOSING))
ns = NULL;
else
netstack_hold(ns);
zone_rele(zone);
return (ns);
}
static int
pset_bind(psetid_t pset, idtype_t idtype, id_t id, psetid_t *opset)
{
kthread_t *tp;
proc_t *pp;
task_t *tk;
kproject_t *kpj;
contract_t *ct;
zone_t *zptr;
psetid_t oldpset;
int error = 0;
void *projbuf, *zonebuf;
pool_lock();
if ((pset != PS_QUERY) && (pset != PS_SOFT) &&
(pset != PS_HARD) && (pset != PS_QUERY_TYPE)) {
/*
* Check if the set actually exists before checking
* permissions. This is the historical error
* precedence. Note that if pset was PS_MYID, the
* cpupart_get_cpus call will change it to the
* processor set id of the caller (or PS_NONE if the
* caller is not bound to a processor set).
*/
if (pool_state == POOL_ENABLED) {
pool_unlock();
return (set_errno(ENOTSUP));
}
if (cpupart_get_cpus(&pset, NULL, NULL) != 0) {
pool_unlock();
return (set_errno(EINVAL));
} else if (pset != PS_NONE && secpolicy_pset(CRED()) != 0) {
pool_unlock();
return (set_errno(EPERM));
}
}
/*
* Pre-allocate enough buffers for FSS for all active projects
* and for all active zones on the system. Unused buffers will
* be freed later by fss_freebuf().
*/
mutex_enter(&cpu_lock);
projbuf = fss_allocbuf(FSS_NPROJ_BUF, FSS_ALLOC_PROJ);
zonebuf = fss_allocbuf(FSS_NPROJ_BUF, FSS_ALLOC_ZONE);
switch (idtype) {
case P_LWPID:
pp = curproc;
mutex_enter(&pidlock);
mutex_enter(&pp->p_lock);
if (id == P_MYID) {
tp = curthread;
} else {
if ((tp = idtot(pp, id)) == NULL) {
mutex_exit(&pp->p_lock);
mutex_exit(&pidlock);
error = ESRCH;
break;
}
}
error = pset_bind_thread(tp, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pp->p_lock);
mutex_exit(&pidlock);
break;
case P_PID:
mutex_enter(&pidlock);
if (id == P_MYID) {
pp = curproc;
} else if ((pp = prfind(id)) == NULL) {
mutex_exit(&pidlock);
error = ESRCH;
break;
}
error = pset_bind_process(pp, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pidlock);
break;
case P_TASKID:
mutex_enter(&pidlock);
if (id == P_MYID)
id = curproc->p_task->tk_tkid;
if ((tk = task_hold_by_id(id)) == NULL) {
mutex_exit(&pidlock);
error = ESRCH;
break;
}
error = pset_bind_task(tk, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pidlock);
task_rele(tk);
break;
case P_PROJID:
pp = curproc;
if (id == P_MYID)
id = curprojid();
if ((kpj = project_hold_by_id(id, pp->p_zone,
PROJECT_HOLD_FIND)) == NULL) {
error = ESRCH;
break;
}
mutex_enter(&pidlock);
error = pset_bind_project(kpj, pset, &oldpset, projbuf,
zonebuf);
mutex_exit(&pidlock);
project_rele(kpj);
break;
case P_ZONEID:
if (id == P_MYID)
id = getzoneid();
if ((zptr = zone_find_by_id(id)) == NULL) {
error = ESRCH;
break;
}
mutex_enter(&pidlock);
error = pset_bind_zone(zptr, pset, &oldpset, projbuf, zonebuf);
mutex_exit(&pidlock);
zone_rele(zptr);
break;
case P_CTID:
if (id == P_MYID)
id = PRCTID(curproc);
if ((ct = contract_type_ptr(process_type, id,
curproc->p_zone->zone_uniqid)) == NULL) {
error = ESRCH;
break;
}
mutex_enter(&pidlock);
error = pset_bind_contract(ct->ct_data, pset, &oldpset, projbuf,
zonebuf);
mutex_exit(&pidlock);
contract_rele(ct);
break;
case P_PSETID:
if (id == P_MYID || pset != PS_NONE || !INGLOBALZONE(curproc)) {
error = EINVAL;
break;
}
error = pset_unbind(id, projbuf, zonebuf, idtype);
break;
case P_ALL:
if (id == P_MYID || pset != PS_NONE || !INGLOBALZONE(curproc)) {
error = EINVAL;
break;
}
error = pset_unbind(PS_NONE, projbuf, zonebuf, idtype);
break;
default:
error = EINVAL;
break;
}
fss_freebuf(projbuf, FSS_ALLOC_PROJ);
fss_freebuf(zonebuf, FSS_ALLOC_ZONE);
mutex_exit(&cpu_lock);
pool_unlock();
if (error != 0)
return (set_errno(error));
if (opset != NULL) {
if (copyout(&oldpset, opset, sizeof (psetid_t)) != 0)
return (set_errno(EFAULT));
}
return (0);
}
/*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);
}
void
log_sendmsg(mblk_t *mp, zoneid_t zoneid)
{
log_t *lp;
char *src, *dst;
mblk_t *mp2 = mp->b_cont;
log_ctl_t *lc = (log_ctl_t *)mp->b_rptr;
int flags, fac;
off_t facility = 0;
off_t body = 0;
zone_t *zptr = NULL;
log_zone_t *lzp;
int i;
int backlog;
/*
* Need to special case the global zone here since this may be
* called before zone_init.
*/
if (zoneid == GLOBAL_ZONEID) {
lzp = &log_global;
} else if ((zptr = zone_find_by_id(zoneid)) == NULL) {
/* specified zone doesn't exist, free message and return */
log_freemsg(mp);
return;
} else {
lzp = zone_getspecific(log_zone_key, zptr);
}
ASSERT(lzp != NULL);
if ((lc->flags & lzp->lz_active) == 0) {
if (zptr)
zone_rele(zptr);
log_freemsg(mp);
return;
}
if (panicstr) {
/*
* Raise the console queue's q_hiwat to ensure that we
* capture all panic messages.
*/
log_consq->q_hiwat = 2 * LOG_HIWAT;
log_consq->q_flag &= ~QFULL;
/* Message was created while panicking. */
lc->flags |= SL_PANICMSG;
}
src = (char *)mp2->b_rptr;
dst = strstr(src, "FACILITY_AND_PRIORITY] ");
if (dst != NULL) {
facility = dst - src;
body = facility + 23; /* strlen("FACILITY_AND_PRIORITY] ") */
}
log_enter();
/*
* In the early boot phase hrestime is invalid, then timechanged is 0.
* If hrestime is not valid, the ttime is set to 0 here and the correct
* ttime is calculated in log_conswitch() later. The log_conswitch()
* calculation to determine the correct ttime does not use ttime data
* from these log_ctl_t structures; it only uses ttime from log_ctl_t's
* that contain good data.
*
*/
lc->ltime = ddi_get_lbolt();
if (timechanged) {
lc->ttime = gethrestime_sec();
} else {
lc->ttime = 0;
}
flags = lc->flags & lzp->lz_active;
log_seq_no[flags & SL_ERROR]++;
log_seq_no[flags & SL_TRACE]++;
log_seq_no[flags & SL_CONSOLE]++;
/*
* If this is in the global zone, start with the backlog, then
* walk through the clone logs. If not, just do the clone logs.
*/
backlog = (zoneid == GLOBAL_ZONEID);
i = LOG_LOGMINIDX;
while (i <= LOG_LOGMAXIDX) {
if (backlog) {
/*
* Do the backlog this time, then start on the
* others.
*/
backlog = 0;
lp = &log_backlog;
} else {
lp = &lzp->lz_clones[i++];
}
if ((lp->log_flags & flags) && lp->log_wanted(lp, lc)) {
if (canput(lp->log_q)) {
lp->log_overflow = 0;
lc->seq_no = log_seq_no[lp->log_flags];
if ((mp2 = copymsg(mp)) == NULL)
break;
if (facility != 0) {
src = (char *)mp2->b_cont->b_rptr;
dst = src + facility;
fac = (lc->pri & LOG_FACMASK) >> 3;
dst += snprintf(dst,
LOG_FACSIZE + LOG_PRISIZE, "%s.%s",
log_fac[MIN(fac, LOG_NFACILITIES)],
log_pri[lc->pri & LOG_PRIMASK]);
src += body - 2; /* copy "] " too */
while (*src != '\0')
*dst++ = *src++;
*dst++ = '\0';
mp2->b_cont->b_wptr = (uchar_t *)dst;
}
(void) putq(lp->log_q, mp2);
} else if (++lp->log_overflow == 1) {
/*
* getflabel -
*
* Return pointer to the ts_label associated with the specified file,
* or returns NULL if error occurs. Caller is responsible for doing
* a label_rele of the ts_label.
*/
ts_label_t *
getflabel(vnode_t *vp)
{
vfs_t *vfsp, *rvfsp;
vnode_t *rvp, *rvp2;
zone_t *zone;
ts_label_t *zl;
int err;
boolean_t vfs_is_held = B_FALSE;
char vpath[MAXPATHLEN];
ASSERT(vp);
vfsp = vp->v_vfsp;
if (vfsp == NULL)
return (NULL);
rvp = vp;
/*
* Traverse lofs mounts and fattach'es to get the real vnode
*/
if (VOP_REALVP(rvp, &rvp2, NULL) == 0)
rvp = rvp2;
rvfsp = rvp->v_vfsp;
/* rvp/rvfsp now represent the real vnode/vfs we will be using */
/* Go elsewhere to handle all nfs files. */
if (strncmp(vfssw[rvfsp->vfs_fstype].vsw_name, "nfs", 3) == 0)
return (getflabel_nfs(rvfsp));
/*
* Fast path, for objects in a labeled zone: everything except
* for lofs/nfs will be just the label of that zone.
*/
if ((rvfsp->vfs_zone != NULL) && (rvfsp->vfs_zone != global_zone)) {
if ((strcmp(vfssw[rvfsp->vfs_fstype].vsw_name,
"lofs") != 0)) {
zone = rvfsp->vfs_zone;
zone_hold(zone);
goto zone_out; /* return this label */
}
}
/*
* Get the vnode path -- it may be missing or weird for some
* cases, like devices. In those cases use the label of the
* current zone.
*/
err = vnodetopath(rootdir, rvp, vpath, sizeof (vpath), kcred);
if ((err != 0) || (*vpath != '/')) {
zone = curproc->p_zone;
zone_hold(zone);
goto zone_out;
}
/*
* For zfs filesystem, return the explicit label property if a
* meaningful one exists.
*/
if (strncmp(vfssw[rvfsp->vfs_fstype].vsw_name, "zfs", 3) == 0) {
ts_label_t *tsl;
tsl = getflabel_zfs(rvfsp);
/* if label found, return it, otherwise continue... */
if (tsl != NULL)
return (tsl);
}
/*
* If a mountpoint exists, hold the vfs while we reference it.
* Otherwise if mountpoint is NULL it should not be held (e.g.,
* a hold/release on spec_vfs would result in an attempted free
* and panic.)
*/
if (vfsp->vfs_mntpt != NULL) {
VFS_HOLD(vfsp);
vfs_is_held = B_TRUE;
}
zone = zone_find_by_any_path(vpath, B_FALSE);
/*
* If the vnode source zone is properly set to a non-global zone, or
* any zone if the mount is R/W, then use the label of that zone.
*/
if ((zone != global_zone) || ((vfsp->vfs_flag & VFS_RDONLY) != 0))
goto zone_out; /* return this label */
/*
* Otherwise, if we're not in the global zone, use the label of
* our zone.
*/
if ((zone = curproc->p_zone) != global_zone) {
zone_hold(zone);
goto zone_out; /* return this label */
}
/*
* We're in the global zone and the mount is R/W ... so the file
* may actually be in the global zone -- or in the root of any zone.
* Always build our own path for the file, to be sure it's simplified
* (i.e., no ".", "..", "//", and so on).
*/
zone_rele(zone);
zone = zone_find_by_any_path(vpath, B_FALSE);
zone_out:
if ((curproc->p_zone == global_zone) && (zone == global_zone)) {
vfs_t *nvfs;
boolean_t exported = B_FALSE;
refstr_t *mntpt_ref;
char *mntpt;
/*
* File is in the global zone - check whether it's admin_high.
* If it's in a filesys that was exported from the global zone,
* it's admin_low by definition. Otherwise, if it's in a
* filesys that's NOT exported to any zone, it's admin_high.
*
* And for these files if there wasn't a valid mount resource,
* the file must be admin_high (not exported, probably a global
* zone device).
*/
if (!vfs_is_held)
goto out_high;
mntpt_ref = vfs_getmntpoint(vfsp);
mntpt = (char *)refstr_value(mntpt_ref);
if ((mntpt != NULL) && (*mntpt == '/')) {
zone_t *to_zone;
to_zone = zone_find_by_any_path(mntpt, B_FALSE);
zone_rele(to_zone);
if (to_zone != global_zone) {
/* force admin_low */
exported = B_TRUE;
}
}
if (mntpt_ref)
refstr_rele(mntpt_ref);
if (!exported) {
size_t plen = strlen(vpath);
vfs_list_read_lock();
nvfs = vfsp->vfs_next;
while (nvfs != vfsp) {
const char *rstr;
size_t rlen = 0;
/*
* Skip checking this vfs if it's not lofs
* (the only way to export from the global
* zone to a zone).
*/
if (strncmp(vfssw[nvfs->vfs_fstype].vsw_name,
"lofs", 4) != 0) {
nvfs = nvfs->vfs_next;
continue;
}
rstr = refstr_value(nvfs->vfs_resource);
if (rstr != NULL)
rlen = strlen(rstr);
/*
* Check for a match: does this vfs correspond
* to our global zone file path? I.e., check
* if the resource string of this vfs is a
* prefix of our path.
*/
if ((rlen > 0) && (rlen <= plen) &&
(strncmp(rstr, vpath, rlen) == 0) &&
(vpath[rlen] == '/' ||
vpath[rlen] == '\0')) {
/* force admin_low */
exported = B_TRUE;
break;
}
nvfs = nvfs->vfs_next;
}
vfs_list_unlock();
}
if (!exported)
goto out_high;
}
if (vfs_is_held)
VFS_RELE(vfsp);
/*
* Now that we have the "home" zone for the file, return the slabel
* of that zone.
*/
zl = zone->zone_slabel;
label_hold(zl);
zone_rele(zone);
return (zl);
out_high:
if (vfs_is_held)
VFS_RELE(vfsp);
label_hold(l_admin_high);
zone_rele(zone);
return (l_admin_high);
}
/*
* smbfs_mount_label_policy:
* Determine whether the mount is allowed according to MAC check,
* by comparing (where appropriate) label of the remote server
* against the label of the zone being mounted into.
*
* Returns:
* 0 : access allowed
* -1 : read-only access allowed (i.e., read-down)
* >0 : error code, such as EACCES
*
* NB:
* NFS supports Cipso labels by parsing the vfs_resource
* to see what the Solaris server global zone has shared.
* We can't support that for CIFS since resource names
* contain share names, not paths.
*/
static int
smbfs_mount_label_policy(vfs_t *vfsp, void *ipaddr, int addr_type, cred_t *cr)
{
bslabel_t *server_sl, *mntlabel;
zone_t *mntzone = NULL;
ts_label_t *zlabel;
tsol_tpc_t *tp;
ts_label_t *tsl = NULL;
int retv;
/*
* Get the zone's label. Each zone on a labeled system has a label.
*/
mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
zlabel = mntzone->zone_slabel;
ASSERT(zlabel != NULL);
label_hold(zlabel);
retv = EACCES; /* assume the worst */
/*
* Next, get the assigned label of the remote server.
*/
tp = find_tpc(ipaddr, addr_type, B_FALSE);
if (tp == NULL)
goto out; /* error getting host entry */
if (tp->tpc_tp.tp_doi != zlabel->tsl_doi)
goto rel_tpc; /* invalid domain */
if ((tp->tpc_tp.host_type != UNLABELED))
goto rel_tpc; /* invalid hosttype */
server_sl = &tp->tpc_tp.tp_def_label;
mntlabel = label2bslabel(zlabel);
/*
* Now compare labels to complete the MAC check. If the labels
* are equal or if the requestor is in the global zone and has
* NET_MAC_AWARE, then allow read-write access. (Except for
* mounts into the global zone itself; restrict these to
* read-only.)
*
* If the requestor is in some other zone, but his label
* dominates the server, then allow read-down.
*
* Otherwise, access is denied.
*/
if (blequal(mntlabel, server_sl) ||
(crgetzoneid(cr) == GLOBAL_ZONEID &&
getpflags(NET_MAC_AWARE, cr) != 0)) {
if ((mntzone == global_zone) ||
!blequal(mntlabel, server_sl))
retv = -1; /* read-only */
else
retv = 0; /* access OK */
} else if (bldominates(mntlabel, server_sl)) {
retv = -1; /* read-only */
} else {
retv = EACCES;
}
if (tsl != NULL)
label_rele(tsl);
rel_tpc:
/*LINTED*/
TPC_RELE(tp);
out:
if (mntzone)
zone_rele(mntzone);
label_rele(zlabel);
return (retv);
}
/*
* 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);
}