/*ARGSUSED*/
static int
nca_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
{
/* Reopen supported */
if (q->q_ptr != NULL)
return (0);
/*
* NCA is not supported in non-global zones; we enforce this restriction
* here.
*/
if (credp != NULL && crgetzoneid(credp) != GLOBAL_ZONEID) {
return (ENOTSUP);
}
if (! (sflag & MODOPEN)) {
/* Device instance */
RD(q)->q_ptr = (void *)B_TRUE;
WR(q)->q_ptr = (void *)B_TRUE;
} else {
/* Modopen just pass through */
RD(q)->q_ptr = (void *)B_FALSE;
WR(q)->q_ptr = (void *)B_FALSE;
}
qprocson(q);
return (0);
}
/*
* Implements flow walk ioctl.
* Retrieves a specific flow or a list of flows from the specified link.
* ENOSPC is returned a bigger buffer is needed.
*/
int
dld_walk_flow(dld_ioc_walkflow_t *wf, intptr_t uaddr, cred_t *credp)
{
flowinfo_state_t state;
mac_flowinfo_t finfo;
int err = 0;
/* For now, one can only view flows from the global zone. */
if (crgetzoneid(credp) != GLOBAL_ZONEID)
return (EPERM);
state.fi_bufsize = wf->wf_len;
state.fi_fl = (uchar_t *)uaddr + sizeof (*wf);
state.fi_nflows = 0;
if (wf->wf_name[0] == '\0') {
err = mac_link_flow_walk(wf->wf_linkid, dld_walk_flow_cb,
&state);
} else {
err = mac_link_flow_info(wf->wf_name, &finfo);
if (err != 0)
return (err);
err = dld_walk_flow_cb(&finfo, &state);
}
wf->wf_nflows = state.fi_nflows;
return (err);
}
/*
* Find a stack instance given the cred.
* This is used by the modules to potentially allow for a future when
* something other than the zoneid is used to determine the stack.
*/
netstack_t *
netstack_find_by_cred(const cred_t *cr)
{
zoneid_t zoneid = crgetzoneid(cr);
/* Handle the case when cr_zone is NULL */
if (zoneid == (zoneid_t)-1)
zoneid = GLOBAL_ZONEID;
/* For performance ... */
if (curproc->p_zone->zone_id == zoneid)
return (netstack_get_current());
else
return (netstack_find_by_zoneid(zoneid));
}
/*
* pfp_open_index is an internal function used to open a MAC device by
* its index. Both a mac_handle_t and mac_client_handle_t are acquired
* because some of the interfaces provided by the mac layer require either
* only the mac_handle_t or both it and mac_handle_t.
*
* Whilst inside the kernel we can access data structures supporting any
* zone, access to interfaces from non-global zones is restricted to those
* interfaces (if any) that are exclusively assigned to a zone.
*/
static int
pfp_open_index(int index, mac_handle_t *mhp, mac_client_handle_t *mcip,
cred_t *cred)
{
mac_client_handle_t mch;
zoneid_t ifzoneid;
mac_handle_t mh;
zoneid_t zoneid;
int error;
mh = 0;
mch = 0;
error = mac_open_by_linkid(index, &mh);
if (error != 0)
goto bad_open;
error = mac_client_open(mh, &mch, NULL,
MAC_OPEN_FLAGS_USE_DATALINK_NAME);
if (error != 0)
goto bad_open;
zoneid = crgetzoneid(cred);
if (zoneid != GLOBAL_ZONEID) {
mac_perim_handle_t perim;
mac_perim_enter_by_mh(mh, &perim);
error = dls_link_getzid(mac_client_name(mch), &ifzoneid);
mac_perim_exit(perim);
if (error != 0)
goto bad_open;
if (ifzoneid != zoneid) {
error = EACCES;
goto bad_open;
}
}
*mcip = mch;
*mhp = mh;
return (0);
bad_open:
if (mch != 0)
mac_client_close(mch, 0);
if (mh != 0)
mac_close(mh);
return (error);
}
/*
* Convert a credential into a "ucred". Allow the caller to specify
* and aligned buffer, e.g., in an mblk, so we don't have to allocate
* memory and copy it twice.
*
* This function may call cred2ucaud(), which calls CRED(). Since this
* can be called from an interrupt thread, receiver's cred (rcr) is needed
* to determine whether audit info should be included.
*/
struct ucred_s *
cred2ucred(const cred_t *cr, pid_t pid, void *buf, const cred_t *rcr)
{
struct ucred_s *uc;
uint32_t realsz = ucredminsize(cr);
ts_label_t *tslp = is_system_labeled() ? crgetlabel(cr) : NULL;
/* The structure isn't always completely filled in, so zero it */
if (buf == NULL) {
uc = kmem_zalloc(realsz, KM_SLEEP);
} else {
bzero(buf, realsz);
uc = buf;
}
uc->uc_size = realsz;
uc->uc_pid = pid;
uc->uc_projid = cr->cr_projid;
uc->uc_zoneid = crgetzoneid(cr);
if (REMOTE_PEER_CRED(cr)) {
/*
* Other than label, the rest of cred info about a
* remote peer isn't available. Copy the label directly
* after the header where we generally copy the prcred.
* That's why we use sizeof (struct ucred_s). The other
* offset fields are initialized to 0.
*/
uc->uc_labeloff = tslp == NULL ? 0 : sizeof (struct ucred_s);
} else {
uc->uc_credoff = UCRED_CRED_OFF;
uc->uc_privoff = UCRED_PRIV_OFF;
uc->uc_audoff = UCRED_AUD_OFF;
uc->uc_labeloff = tslp == NULL ? 0 : UCRED_LABEL_OFF;
cred2prcred(cr, UCCRED(uc));
cred2prpriv(cr, UCPRIV(uc));
if (audoff == 0 || cred2ucaud(cr, UCAUD(uc), rcr) != 0)
uc->uc_audoff = 0;
}
if (tslp != NULL)
bcopy(&tslp->tsl_label, UCLABEL(uc), sizeof (bslabel_t));
return (uc);
}
/*ARGSUSED*/
static int
nsmb_open(dev_t *dev, int flags, int otyp, cred_t *cr)
{
smb_dev_t *sdp;
minor_t m;
mutex_enter(&dev_lck);
for (m = last_minor + 1; m != last_minor; m++) {
if (m > NSMB_MAX_MINOR)
m = NSMB_MIN_MINOR;
if (ddi_get_soft_state(statep, m) == NULL) {
last_minor = m;
goto found;
}
}
/* No available minor units. */
mutex_exit(&dev_lck);
return (ENXIO);
found:
/* NB: dev_lck still held */
if (ddi_soft_state_zalloc(statep, m) == DDI_FAILURE) {
mutex_exit(&dev_lck);
return (ENXIO);
}
if ((sdp = ddi_get_soft_state(statep, m)) == NULL) {
mutex_exit(&dev_lck);
return (ENXIO);
}
*dev = makedevice(nsmb_major, m);
mutex_exit(&dev_lck);
sdp->sd_cred = cr;
sdp->sd_smbfid = -1;
sdp->sd_flags |= NSMBFL_OPEN;
sdp->zoneid = crgetzoneid(cr);
return (0);
}
int
vnic_info(vnic_info_t *info, cred_t *credp)
{
vnic_t *vnic;
int err;
/* Make sure that the VNIC link is visible from the caller's zone. */
if (!dls_devnet_islinkvisible(info->vn_vnic_id, crgetzoneid(credp)))
return (ENOENT);
rw_enter(&vnic_lock, RW_WRITER);
err = mod_hash_find(vnic_hash, VNIC_HASH_KEY(info->vn_vnic_id),
(mod_hash_val_t *)&vnic);
if (err != 0) {
rw_exit(&vnic_lock);
return (ENOENT);
}
info->vn_link_id = vnic->vn_link_id;
info->vn_mac_addr_type = vnic->vn_addr_type;
info->vn_mac_len = vnic->vn_addr_len;
bcopy(vnic->vn_addr, info->vn_mac_addr, MAXMACADDRLEN);
info->vn_mac_slot = vnic->vn_slot_id;
info->vn_mac_prefix_len = 0;
info->vn_vid = vnic->vn_vid;
info->vn_force = vnic->vn_force;
info->vn_vrid = vnic->vn_vrid;
info->vn_af = vnic->vn_af;
bzero(&info->vn_resource_props, sizeof (mac_resource_props_t));
if (vnic->vn_mch != NULL)
mac_client_get_resources(vnic->vn_mch,
&info->vn_resource_props);
rw_exit(&vnic_lock);
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);
}
/* ARGSUSED */
int
vnic_dev_delete(datalink_id_t vnic_id, uint32_t flags, cred_t *credp)
{
vnic_t *vnic = NULL;
mod_hash_val_t val;
datalink_id_t tmpid;
int rc;
rw_enter(&vnic_lock, RW_WRITER);
if (mod_hash_find(vnic_hash, VNIC_HASH_KEY(vnic_id),
(mod_hash_val_t *)&vnic) != 0) {
rw_exit(&vnic_lock);
return (ENOENT);
}
if ((rc = dls_devnet_destroy(vnic->vn_mh, &tmpid, B_TRUE)) != 0) {
rw_exit(&vnic_lock);
return (rc);
}
ASSERT(vnic_id == tmpid);
/*
* We cannot unregister the MAC yet. Unregistering would
* free up mac_impl_t which should not happen at this time.
* So disable mac_impl_t by calling mac_disable(). This will prevent
* any new claims on mac_impl_t.
*/
if ((rc = mac_disable(vnic->vn_mh)) != 0) {
(void) dls_devnet_create(vnic->vn_mh, vnic_id,
crgetzoneid(credp));
rw_exit(&vnic_lock);
return (rc);
}
vnic_cleanup_secondary_macs(vnic, vnic->vn_nhandles);
vnic->vn_enabled = B_FALSE;
(void) mod_hash_remove(vnic_hash, VNIC_HASH_KEY(vnic_id), &val);
ASSERT(vnic == (vnic_t *)val);
vnic_count--;
rw_exit(&vnic_lock);
/*
* XXX-nicolas shouldn't have a void cast here, if it's
* expected that the function will never fail, then we should
* have an ASSERT().
*/
(void) mac_unregister(vnic->vn_mh);
if (vnic->vn_lower_mh != NULL) {
/*
* Check if MAC address for the vnic was obtained from the
* factory MAC addresses. If yes, release it.
*/
if (vnic->vn_addr_type == VNIC_MAC_ADDR_TYPE_FACTORY) {
(void) mac_addr_factory_release(vnic->vn_mch,
vnic->vn_slot_id);
}
(void) mac_margin_remove(vnic->vn_lower_mh, vnic->vn_margin);
(void) mac_mtu_remove(vnic->vn_lower_mh, vnic->vn_mtu);
(void) mac_notify_remove(vnic->vn_mnh, B_TRUE);
(void) mac_unicast_remove(vnic->vn_mch, vnic->vn_muh);
mac_client_close(vnic->vn_mch, MAC_CLOSE_FLAGS_IS_VNIC);
mac_close(vnic->vn_lower_mh);
}
kmem_cache_free(vnic_cache, vnic);
return (0);
}
/* ARGSUSED */
int
vnic_dev_create(datalink_id_t vnic_id, datalink_id_t linkid,
vnic_mac_addr_type_t *vnic_addr_type, int *mac_len, uchar_t *mac_addr,
int *mac_slot, uint_t mac_prefix_len, uint16_t vid, vrid_t vrid,
int af, mac_resource_props_t *mrp, uint32_t flags, vnic_ioc_diag_t *diag,
cred_t *credp)
{
vnic_t *vnic;
mac_register_t *mac;
int err;
boolean_t is_anchor = ((flags & VNIC_IOC_CREATE_ANCHOR) != 0);
char vnic_name[MAXNAMELEN];
const mac_info_t *minfop;
uint32_t req_hwgrp_flag = B_FALSE;
*diag = VNIC_IOC_DIAG_NONE;
rw_enter(&vnic_lock, RW_WRITER);
/* does a VNIC with the same id already exist? */
err = mod_hash_find(vnic_hash, VNIC_HASH_KEY(vnic_id),
(mod_hash_val_t *)&vnic);
if (err == 0) {
rw_exit(&vnic_lock);
return (EEXIST);
}
vnic = kmem_cache_alloc(vnic_cache, KM_NOSLEEP);
if (vnic == NULL) {
rw_exit(&vnic_lock);
return (ENOMEM);
}
bzero(vnic, sizeof (*vnic));
vnic->vn_id = vnic_id;
vnic->vn_link_id = linkid;
vnic->vn_vrid = vrid;
vnic->vn_af = af;
if (!is_anchor) {
if (linkid == DATALINK_INVALID_LINKID) {
err = EINVAL;
goto bail;
}
/*
* Open the lower MAC and assign its initial bandwidth and
* MAC address. We do this here during VNIC creation and
* do not wait until the upper MAC client open so that we
* can validate the VNIC creation parameters (bandwidth,
* MAC address, etc) and reserve a factory MAC address if
* one was requested.
*/
err = mac_open_by_linkid(linkid, &vnic->vn_lower_mh);
if (err != 0)
goto bail;
/*
* VNIC(vlan) over VNICs(vlans) is not supported.
*/
if (mac_is_vnic(vnic->vn_lower_mh)) {
err = EINVAL;
goto bail;
}
/* only ethernet support for now */
minfop = mac_info(vnic->vn_lower_mh);
if (minfop->mi_nativemedia != DL_ETHER) {
err = ENOTSUP;
goto bail;
}
(void) dls_mgmt_get_linkinfo(vnic_id, vnic_name, NULL, NULL,
NULL);
err = mac_client_open(vnic->vn_lower_mh, &vnic->vn_mch,
vnic_name, MAC_OPEN_FLAGS_IS_VNIC);
if (err != 0)
goto bail;
/* assign a MAC address to the VNIC */
err = vnic_unicast_add(vnic, *vnic_addr_type, mac_slot,
mac_prefix_len, mac_len, mac_addr, flags, diag, vid,
req_hwgrp_flag);
if (err != 0) {
vnic->vn_muh = NULL;
if (diag != NULL && req_hwgrp_flag)
*diag = VNIC_IOC_DIAG_NO_HWRINGS;
goto bail;
}
/* register to receive notification from underlying MAC */
vnic->vn_mnh = mac_notify_add(vnic->vn_lower_mh, vnic_notify_cb,
vnic);
*vnic_addr_type = vnic->vn_addr_type;
vnic->vn_addr_len = *mac_len;
vnic->vn_vid = vid;
bcopy(mac_addr, vnic->vn_addr, vnic->vn_addr_len);
if (vnic->vn_addr_type == VNIC_MAC_ADDR_TYPE_FACTORY)
vnic->vn_slot_id = *mac_slot;
/*
* Set the initial VNIC capabilities. If the VNIC is created
* over MACs which does not support nactive vlan, disable
* VNIC's hardware checksum capability if its VID is not 0,
* since the underlying MAC would get the hardware checksum
* offset wrong in case of VLAN packets.
*/
if (vid == 0 || !mac_capab_get(vnic->vn_lower_mh,
MAC_CAPAB_NO_NATIVEVLAN, NULL)) {
if (!mac_capab_get(vnic->vn_lower_mh, MAC_CAPAB_HCKSUM,
&vnic->vn_hcksum_txflags))
vnic->vn_hcksum_txflags = 0;
} else {
vnic->vn_hcksum_txflags = 0;
}
}
/* register with the MAC module */
if ((mac = mac_alloc(MAC_VERSION)) == NULL)
goto bail;
mac->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
mac->m_driver = vnic;
mac->m_dip = vnic_get_dip();
mac->m_instance = (uint_t)-1;
mac->m_src_addr = vnic->vn_addr;
mac->m_callbacks = &vnic_m_callbacks;
if (!is_anchor) {
/*
* If this is a VNIC based VLAN, then we check for the
* margin unless it has been created with the force
* flag. If we are configuring a VLAN over an etherstub,
* we don't check the margin even if force is not set.
*/
if (vid == 0 || (flags & VNIC_IOC_CREATE_FORCE) != 0) {
if (vid != VLAN_ID_NONE)
vnic->vn_force = B_TRUE;
/*
* As the current margin size of the underlying mac is
* used to determine the margin size of the VNIC
* itself, request the underlying mac not to change
* to a smaller margin size.
*/
err = mac_margin_add(vnic->vn_lower_mh,
&vnic->vn_margin, B_TRUE);
ASSERT(err == 0);
} else {
vnic->vn_margin = VLAN_TAGSZ;
err = mac_margin_add(vnic->vn_lower_mh,
&vnic->vn_margin, B_FALSE);
if (err != 0) {
mac_free(mac);
if (diag != NULL)
*diag = VNIC_IOC_DIAG_MACMARGIN_INVALID;
goto bail;
}
}
mac_sdu_get(vnic->vn_lower_mh, &mac->m_min_sdu,
&mac->m_max_sdu);
err = mac_mtu_add(vnic->vn_lower_mh, &mac->m_max_sdu, B_FALSE);
if (err != 0) {
VERIFY(mac_margin_remove(vnic->vn_lower_mh,
vnic->vn_margin) == 0);
mac_free(mac);
if (diag != NULL)
*diag = VNIC_IOC_DIAG_MACMTU_INVALID;
goto bail;
}
vnic->vn_mtu = mac->m_max_sdu;
} else {
vnic->vn_margin = VLAN_TAGSZ;
mac->m_min_sdu = 1;
mac->m_max_sdu = ANCHOR_VNIC_MAX_MTU;
vnic->vn_mtu = ANCHOR_VNIC_MAX_MTU;
}
mac->m_margin = vnic->vn_margin;
err = mac_register(mac, &vnic->vn_mh);
mac_free(mac);
if (err != 0) {
if (!is_anchor) {
VERIFY(mac_mtu_remove(vnic->vn_lower_mh,
vnic->vn_mtu) == 0);
VERIFY(mac_margin_remove(vnic->vn_lower_mh,
vnic->vn_margin) == 0);
}
goto bail;
}
/* Set the VNIC's MAC in the client */
if (!is_anchor) {
mac_set_upper_mac(vnic->vn_mch, vnic->vn_mh, mrp);
if (mrp != NULL) {
if ((mrp->mrp_mask & MRP_RX_RINGS) != 0 ||
(mrp->mrp_mask & MRP_TX_RINGS) != 0) {
req_hwgrp_flag = B_TRUE;
}
err = mac_client_set_resources(vnic->vn_mch, mrp);
if (err != 0) {
VERIFY(mac_mtu_remove(vnic->vn_lower_mh,
vnic->vn_mtu) == 0);
VERIFY(mac_margin_remove(vnic->vn_lower_mh,
vnic->vn_margin) == 0);
(void) mac_unregister(vnic->vn_mh);
goto bail;
}
}
}
err = dls_devnet_create(vnic->vn_mh, vnic->vn_id, crgetzoneid(credp));
if (err != 0) {
VERIFY(is_anchor || mac_margin_remove(vnic->vn_lower_mh,
vnic->vn_margin) == 0);
if (!is_anchor) {
VERIFY(mac_mtu_remove(vnic->vn_lower_mh,
vnic->vn_mtu) == 0);
VERIFY(mac_margin_remove(vnic->vn_lower_mh,
vnic->vn_margin) == 0);
}
(void) mac_unregister(vnic->vn_mh);
goto bail;
}
/* add new VNIC to hash table */
err = mod_hash_insert(vnic_hash, VNIC_HASH_KEY(vnic_id),
(mod_hash_val_t)vnic);
ASSERT(err == 0);
vnic_count++;
/*
* Now that we've enabled this VNIC, we should go through and update the
* link state by setting it to our parents.
*/
vnic->vn_enabled = B_TRUE;
if (is_anchor) {
mac_link_update(vnic->vn_mh, LINK_STATE_UP);
} else {
mac_link_update(vnic->vn_mh,
mac_client_stat_get(vnic->vn_mch, MAC_STAT_LINK_STATE));
}
rw_exit(&vnic_lock);
return (0);
bail:
rw_exit(&vnic_lock);
if (!is_anchor) {
if (vnic->vn_mnh != NULL)
(void) mac_notify_remove(vnic->vn_mnh, B_TRUE);
if (vnic->vn_muh != NULL)
(void) mac_unicast_remove(vnic->vn_mch, vnic->vn_muh);
if (vnic->vn_mch != NULL)
mac_client_close(vnic->vn_mch, MAC_CLOSE_FLAGS_IS_VNIC);
if (vnic->vn_lower_mh != NULL)
mac_close(vnic->vn_lower_mh);
}
kmem_cache_free(vnic_cache, vnic);
return (err);
}
static int
log_wput(queue_t *q, mblk_t *mp)
{
log_t *lp = (log_t *)q->q_ptr;
struct iocblk *iocp;
mblk_t *mp2;
cred_t *cr = DB_CRED(mp);
zoneid_t zoneid;
/*
* Default to global zone if dblk doesn't have a valid cred.
* Calls to syslog() go through putmsg(), which does set up
* the cred.
*/
zoneid = (cr != NULL) ? crgetzoneid(cr) : GLOBAL_ZONEID;
switch (DB_TYPE(mp)) {
case M_FLUSH:
if (*mp->b_rptr & FLUSHW) {
flushq(q, FLUSHALL);
*mp->b_rptr &= ~FLUSHW;
}
if (*mp->b_rptr & FLUSHR) {
flushq(RD(q), FLUSHALL);
qreply(q, mp);
return (0);
}
break;
case M_IOCTL:
iocp = (struct iocblk *)mp->b_rptr;
if (lp->log_major != LOG_LOGMIN) {
/* write-only device */
miocnak(q, mp, 0, EINVAL);
return (0);
}
if (iocp->ioc_count == TRANSPARENT) {
miocnak(q, mp, 0, EINVAL);
return (0);
}
if (lp->log_flags) {
miocnak(q, mp, 0, EBUSY);
return (0);
}
freemsg(lp->log_data);
lp->log_data = mp->b_cont;
mp->b_cont = NULL;
switch (iocp->ioc_cmd) {
case I_CONSLOG:
log_update(lp, RD(q), SL_CONSOLE, log_console);
break;
case I_TRCLOG:
if (lp->log_data == NULL) {
miocnak(q, mp, 0, EINVAL);
return (0);
}
log_update(lp, RD(q), SL_TRACE, log_trace);
break;
case I_ERRLOG:
log_update(lp, RD(q), SL_ERROR, log_error);
break;
default:
miocnak(q, mp, 0, EINVAL);
return (0);
}
miocack(q, mp, 0, 0);
return (0);
case M_PROTO:
if (MBLKL(mp) == sizeof (log_ctl_t) && mp->b_cont != NULL) {
log_ctl_t *lc = (log_ctl_t *)mp->b_rptr;
/* This code is used by savecore to log dump msgs */
if (mp->b_band != 0 &&
secpolicy_sys_config(CRED(), B_FALSE) == 0) {
(void) putq(log_consq, mp);
return (0);
}
if ((lc->pri & LOG_FACMASK) == LOG_KERN)
lc->pri |= LOG_USER;
mp2 = log_makemsg(LOG_MID, LOG_CONSMIN, lc->level,
lc->flags, lc->pri, mp->b_cont->b_rptr,
MBLKL(mp->b_cont) + 1, 0);
if (mp2 != NULL)
log_sendmsg(mp2, zoneid);
}
break;
case M_DATA:
mp2 = log_makemsg(LOG_MID, LOG_CONSMIN, 0, SL_CONSOLE,
LOG_USER | LOG_INFO, mp->b_rptr, MBLKL(mp) + 1, 0);
if (mp2 != NULL)
log_sendmsg(mp2, zoneid);
break;
}
freemsg(mp);
return (0);
}
/*ARGSUSED*/
static int
nsmb_open(dev_t *dev, int flags, int otyp, cred_t *cr)
{
major_t new_major;
smb_dev_t *sdp, *sdv;
mutex_enter(&dev_lck);
for (; ; ) {
minor_t start = nsmb_minor;
do {
if (nsmb_minor >= MAXMIN32) {
if (nsmb_major == getmajor(*dev))
nsmb_minor = NSMB_MIN_MINOR;
else
nsmb_minor = 0;
} else {
nsmb_minor++;
}
sdv = ddi_get_soft_state(statep, nsmb_minor);
} while ((sdv != NULL) && (nsmb_minor != start));
if (nsmb_minor == start) {
/*
* The condition we need to solve here is all the
* MAXMIN32(~262000) minors numbers are reached. We
* need to create a new major number.
* zfs uses getudev() to create a new major number.
*/
if ((new_major = getudev()) == (major_t)-1) {
cmn_err(CE_WARN,
"nsmb: Can't get unique major "
"device number.");
mutex_exit(&dev_lck);
return (-1);
}
nsmb_major = new_major;
nsmb_minor = 0;
} else {
break;
}
}
/*
* This is called by mount or open call.
* The open() routine is passed a pointer to a device number so
* that the driver can change the minor number. This allows
* drivers to dynamically create minor instances of the dev-
* ice. An example of this might be a pseudo-terminal driver
* that creates a new pseudo-terminal whenever it is opened.
* A driver that chooses the minor number dynamically, normally
* creates only one minor device node in attach(9E) with
* ddi_create_minor_node(9F) then changes the minor number com-
* ponent of *devp using makedevice(9F) and getmajor(9F) The
* driver needs to keep track of available minor numbers inter-
* nally.
* Stuff the structure smb_dev.
* return.
*/
if (ddi_soft_state_zalloc(statep, nsmb_minor) == DDI_FAILURE) {
mutex_exit(&dev_lck);
return (ENXIO);
}
if ((sdp = ddi_get_soft_state(statep, nsmb_minor)) == NULL) {
mutex_exit(&dev_lck);
return (ENXIO);
}
sdp->sd_seq = nsmb_minor;
sdp->sd_cred = cr;
sdp->sd_smbfid = -1;
sdp->sd_flags |= NSMBFL_OPEN;
sdp->zoneid = crgetzoneid(cr);
mutex_exit(&dev_lck);
*dev = makedevice(nsmb_major, nsmb_minor);
return (0);
}
/*
* Remove zombie children from the process table.
*/
void
freeproc(proc_t *p)
{
proc_t *q;
task_t *tk;
ASSERT(p->p_stat == SZOMB);
ASSERT(p->p_tlist == NULL);
ASSERT(MUTEX_HELD(&pidlock));
sigdelq(p, NULL, 0);
if (p->p_killsqp) {
siginfofree(p->p_killsqp);
p->p_killsqp = NULL;
}
prfree(p); /* inform /proc */
/*
* Don't free the init processes.
* Other dying processes will access it.
*/
if (p == proc_init)
return;
/*
* We wait until now to free the cred structure because a
* zombie process's credentials may be examined by /proc.
* No cred locking needed because there are no threads at this point.
*/
upcount_dec(crgetruid(p->p_cred), crgetzoneid(p->p_cred));
crfree(p->p_cred);
if (p->p_corefile != NULL) {
corectl_path_rele(p->p_corefile);
p->p_corefile = NULL;
}
if (p->p_content != NULL) {
corectl_content_rele(p->p_content);
p->p_content = NULL;
}
if (p->p_nextofkin && !((p->p_nextofkin->p_flag & SNOWAIT) ||
(PTOU(p->p_nextofkin)->u_signal[SIGCLD - 1] == SIG_IGN))) {
/*
* This should still do the right thing since p_utime/stime
* get set to the correct value on process exit, so it
* should get properly updated
*/
p->p_nextofkin->p_cutime += p->p_utime;
p->p_nextofkin->p_cstime += p->p_stime;
p->p_nextofkin->p_cacct[LMS_USER] += p->p_acct[LMS_USER];
p->p_nextofkin->p_cacct[LMS_SYSTEM] += p->p_acct[LMS_SYSTEM];
p->p_nextofkin->p_cacct[LMS_TRAP] += p->p_acct[LMS_TRAP];
p->p_nextofkin->p_cacct[LMS_TFAULT] += p->p_acct[LMS_TFAULT];
p->p_nextofkin->p_cacct[LMS_DFAULT] += p->p_acct[LMS_DFAULT];
p->p_nextofkin->p_cacct[LMS_KFAULT] += p->p_acct[LMS_KFAULT];
p->p_nextofkin->p_cacct[LMS_USER_LOCK]
+= p->p_acct[LMS_USER_LOCK];
p->p_nextofkin->p_cacct[LMS_SLEEP] += p->p_acct[LMS_SLEEP];
p->p_nextofkin->p_cacct[LMS_WAIT_CPU]
+= p->p_acct[LMS_WAIT_CPU];
p->p_nextofkin->p_cacct[LMS_STOPPED] += p->p_acct[LMS_STOPPED];
p->p_nextofkin->p_cru.minflt += p->p_ru.minflt;
p->p_nextofkin->p_cru.majflt += p->p_ru.majflt;
p->p_nextofkin->p_cru.nswap += p->p_ru.nswap;
p->p_nextofkin->p_cru.inblock += p->p_ru.inblock;
p->p_nextofkin->p_cru.oublock += p->p_ru.oublock;
p->p_nextofkin->p_cru.msgsnd += p->p_ru.msgsnd;
p->p_nextofkin->p_cru.msgrcv += p->p_ru.msgrcv;
p->p_nextofkin->p_cru.nsignals += p->p_ru.nsignals;
p->p_nextofkin->p_cru.nvcsw += p->p_ru.nvcsw;
p->p_nextofkin->p_cru.nivcsw += p->p_ru.nivcsw;
p->p_nextofkin->p_cru.sysc += p->p_ru.sysc;
p->p_nextofkin->p_cru.ioch += p->p_ru.ioch;
}
q = p->p_nextofkin;
if (q && q->p_orphan == p)
q->p_orphan = p->p_nextorph;
else if (q) {
for (q = q->p_orphan; q; q = q->p_nextorph)
if (q->p_nextorph == p)
break;
ASSERT(q && q->p_nextorph == p);
q->p_nextorph = p->p_nextorph;
}
/*
* The process table slot is being freed, so it is now safe to give up
* task and project membership.
*/
mutex_enter(&p->p_lock);
tk = p->p_task;
task_detach(p);
mutex_exit(&p->p_lock);
proc_detach(p);
pid_exit(p, tk); /* frees pid and proc structure */
task_rele(tk);
}
/*
* sppp_dl_attach_upper()
*
* MT-Perimeters:
* exclusive inner, exclusive outer.
*
* Description:
* Called by qwriter (INNER) from sppp_dlattachreq as the result of
* receiving a DL_ATTACH_REQ message.
*/
static void
sppp_dl_attach_upper(queue_t *q, mblk_t *mp)
{
sppa_t *ppa;
spppstr_t *sps = q->q_ptr;
union DL_primitives *dlp;
int err = ENOMEM;
cred_t *cr;
zoneid_t zoneid;
ASSERT(!IS_SPS_PIOATTACH(sps));
dlp = (union DL_primitives *)mp->b_rptr;
/* If there's something here, it's detached. */
if (sps->sps_ppa != NULL) {
sppp_remove_ppa(sps);
}
if ((cr = msg_getcred(mp, NULL)) == NULL)
zoneid = sps->sps_zoneid;
else
zoneid = crgetzoneid(cr);
ppa = sppp_find_ppa(dlp->attach_req.dl_ppa);
if (ppa == NULL) {
ppa = sppp_create_ppa(dlp->attach_req.dl_ppa, zoneid);
} else if (ppa->ppa_zoneid != zoneid) {
ppa = NULL;
err = EPERM;
}
/*
* If we can't find or create it, then it's either because we're out of
* memory or because the requested PPA is owned by a different zone.
*/
if (ppa == NULL) {
DBGERROR((CE_CONT, "DLPI attach: cannot create ppa %u\n",
dlp->attach_req.dl_ppa));
dlerrorack(q, mp, dlp->dl_primitive, DL_SYSERR, err);
return;
}
/*
* Preallocate the hangup message so that we're always able to
* send this upstream in the event of a catastrophic failure.
*/
if ((sps->sps_hangup = allocb(1, BPRI_MED)) == NULL) {
DBGERROR((CE_CONT, "DLPI attach: cannot allocate hangup\n"));
dlerrorack(q, mp, dlp->dl_primitive, DL_SYSERR, ENOSR);
return;
}
sps->sps_dlstate = DL_UNBOUND;
sps->sps_ppa = ppa;
/*
* Add this stream to the head of the list of sibling streams
* which belong to the specified ppa.
*/
rw_enter(&ppa->ppa_sib_lock, RW_WRITER);
ppa->ppa_refcnt++;
sps->sps_nextsib = ppa->ppa_streams;
ppa->ppa_streams = sps;
/*
* And if this stream was marked as promiscuous (SPS_PROMISC), then we
* need to update the promiscuous streams count. This should only
* happen when DL_PROMISCON_REQ was issued prior to attachment.
*/
if (IS_SPS_PROMISC(sps)) {
ppa->ppa_promicnt++;
}
rw_exit(&ppa->ppa_sib_lock);
DBGDLPI((CE_CONT, "/%d: attached to ppa %d\n", sps->sps_mn_id,
ppa->ppa_ppa_id));
dlokack(q, mp, DL_ATTACH_REQ);
}
/*
* 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);
}