/*
* 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);
}
/* ARGSUSED */
static int
sdpfp_ioctl(sock_lower_handle_t handle, int cmd, intptr_t arg, int mod,
int32_t *rval, struct cred *cr)
{
#if defined(_SYSCALL32)
struct timeval32 tival;
#else
struct timeval tival;
#endif
mac_client_promisc_type_t mtype;
datalink_id_t linkid;
struct lifreq lifreq;
struct ifreq ifreq;
struct pfpsock *ps;
mac_handle_t mh;
timespec_t tv;
int error;
switch (cmd) {
/*
* ioctls that work on "struct lifreq"
*/
case SIOCSLIFFLAGS :
case SIOCGLIFINDEX :
case SIOCGLIFFLAGS :
case SIOCGLIFMTU :
error = pfp_lifreq_getlinkid(arg, &lifreq, &linkid);
if (error != 0)
return (error);
break;
/*
* ioctls that work on "struct ifreq".
* Not all of these have a "struct lifreq" partner, for example
* SIOCGIFHWADDR, for the simple reason that the logical interface
* does not have a hardware address.
*/
case SIOCSIFFLAGS :
case SIOCGIFINDEX :
case SIOCGIFFLAGS :
case SIOCGIFMTU :
case SIOCGIFHWADDR :
error = pfp_ifreq_getlinkid(arg, &ifreq, &linkid);
if (error != 0)
return (error);
break;
}
error = mac_open_by_linkid(linkid, &mh);
if (error != 0)
return (error);
ps = (struct pfpsock *)handle;
switch (cmd) {
case SIOCGLIFINDEX :
lifreq.lifr_index = linkid;
break;
case SIOCGIFINDEX :
ifreq.ifr_index = linkid;
break;
case SIOCGIFFLAGS :
ifreq.ifr_flags = IFF_RUNNING;
if (ps->ps_promisc == MAC_CLIENT_PROMISC_ALL)
ifreq.ifr_flags |= IFF_PROMISC;
break;
case SIOCGLIFFLAGS :
lifreq.lifr_flags = IFF_RUNNING;
if (ps->ps_promisc == MAC_CLIENT_PROMISC_ALL)
lifreq.lifr_flags |= IFF_PROMISC;
break;
case SIOCSIFFLAGS :
if (linkid != ps->ps_linkid) {
error = EINVAL;
} else {
if ((ifreq.ifr_flags & IFF_PROMISC) != 0)
mtype = MAC_CLIENT_PROMISC_ALL;
else
mtype = MAC_CLIENT_PROMISC_FILTERED;
error = pfp_set_promisc(ps, mtype);
}
break;
case SIOCSLIFFLAGS :
if (linkid != ps->ps_linkid) {
error = EINVAL;
} else {
if ((lifreq.lifr_flags & IFF_PROMISC) != 0)
mtype = MAC_CLIENT_PROMISC_ALL;
else
mtype = MAC_CLIENT_PROMISC_FILTERED;
error = pfp_set_promisc(ps, mtype);
}
break;
case SIOCGIFMTU :
mac_sdu_get(mh, NULL, &ifreq.ifr_mtu);
break;
case SIOCGLIFMTU :
mac_sdu_get(mh, NULL, &lifreq.lifr_mtu);
break;
case SIOCGIFHWADDR :
mac_unicast_primary_get(mh, (uint8_t *)ifreq.ifr_addr.sa_data);
ifreq.ifr_addr.sa_family = pfp_dl_to_arphrd(mac_type(mh));
break;
case SIOCGSTAMP :
(void) gethrestime(&tv);
tival.tv_sec = (time_t)tv.tv_sec;
tival.tv_usec = tv.tv_nsec / 1000;
error = ddi_copyout(&tival, (void *)arg, sizeof (tival), 0);
break;
default :
break;
}
mac_close(mh);
if (error == 0) {
/*
* Only the "GET" ioctls need to copy data back to userace.
*/
switch (cmd) {
case SIOCGLIFINDEX :
case SIOCGLIFFLAGS :
case SIOCGLIFMTU :
error = ddi_copyout(&lifreq, (void *)arg,
sizeof (lifreq), 0);
break;
case SIOCGIFINDEX :
case SIOCGIFFLAGS :
case SIOCGIFMTU :
case SIOCGIFHWADDR :
error = ddi_copyout(&ifreq, (void *)arg,
sizeof (ifreq), 0);
break;
default :
break;
}
}
return (error);
}
/* 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);
}
/* ARGSUSED */
static int
sdpfp_ioctl(sock_lower_handle_t handle, int cmd, intptr_t arg, int mod,
int32_t *rval, struct cred *cr)
{
struct timeval tival;
mac_client_promisc_type_t mtype;
struct sockaddr_dl *sock;
datalink_id_t linkid;
struct lifreq lifreq;
struct ifreq ifreq;
struct pfpsock *ps;
mac_handle_t mh;
int error;
ps = (struct pfpsock *)handle;
switch (cmd) {
/*
* ioctls that work on "struct lifreq"
*/
case SIOCSLIFFLAGS :
case SIOCGLIFINDEX :
case SIOCGLIFFLAGS :
case SIOCGLIFMTU :
case SIOCGLIFHWADDR :
error = pfp_lifreq_getlinkid(arg, &lifreq, &linkid, mod);
if (error != 0)
return (error);
break;
/*
* ioctls that work on "struct ifreq".
* Not all of these have a "struct lifreq" partner, for example
* SIOCGIFHWADDR, for the simple reason that the logical interface
* does not have a hardware address.
*/
case SIOCSIFFLAGS :
case SIOCGIFINDEX :
case SIOCGIFFLAGS :
case SIOCGIFMTU :
case SIOCGIFHWADDR :
error = pfp_ifreq_getlinkid(arg, &ifreq, &linkid, mod);
if (error != 0)
return (error);
break;
case SIOCGSTAMP :
tival.tv_sec = (time_t)ps->ps_timestamp.tv_sec;
tival.tv_usec = ps->ps_timestamp.tv_nsec / 1000;
if (get_udatamodel() == DATAMODEL_NATIVE) {
error = ddi_copyout(&tival, (void *)arg,
sizeof (tival), mod);
}
#ifdef _SYSCALL32_IMPL
else {
struct timeval32 tv32;
TIMEVAL_TO_TIMEVAL32(&tv32, &tival);
error = ddi_copyout(&tv32, (void *)arg,
sizeof (tv32), mod);
}
#endif
return (error);
}
error = mac_open_by_linkid(linkid, &mh);
if (error != 0)
return (error);
switch (cmd) {
case SIOCGLIFINDEX :
lifreq.lifr_index = linkid;
break;
case SIOCGIFINDEX :
ifreq.ifr_index = linkid;
break;
case SIOCGIFFLAGS :
ifreq.ifr_flags = IFF_RUNNING;
if (ps->ps_promisc == MAC_CLIENT_PROMISC_ALL)
ifreq.ifr_flags |= IFF_PROMISC;
break;
case SIOCGLIFFLAGS :
lifreq.lifr_flags = IFF_RUNNING;
if (ps->ps_promisc == MAC_CLIENT_PROMISC_ALL)
lifreq.lifr_flags |= IFF_PROMISC;
break;
case SIOCSIFFLAGS :
if (linkid != ps->ps_linkid) {
error = EINVAL;
} else {
if ((ifreq.ifr_flags & IFF_PROMISC) != 0)
mtype = MAC_CLIENT_PROMISC_ALL;
else
mtype = MAC_CLIENT_PROMISC_FILTERED;
error = pfp_set_promisc(ps, mtype);
}
break;
case SIOCSLIFFLAGS :
if (linkid != ps->ps_linkid) {
error = EINVAL;
} else {
if ((lifreq.lifr_flags & IFF_PROMISC) != 0)
mtype = MAC_CLIENT_PROMISC_ALL;
else
mtype = MAC_CLIENT_PROMISC_FILTERED;
error = pfp_set_promisc(ps, mtype);
}
break;
case SIOCGIFMTU :
mac_sdu_get(mh, NULL, &ifreq.ifr_mtu);
break;
case SIOCGLIFMTU :
mac_sdu_get(mh, NULL, &lifreq.lifr_mtu);
break;
case SIOCGIFHWADDR :
if (mac_addr_len(mh) > sizeof (ifreq.ifr_addr.sa_data)) {
error = EPFNOSUPPORT;
break;
}
if (mac_addr_len(mh) == 0) {
(void) memset(ifreq.ifr_addr.sa_data, 0,
sizeof (ifreq.ifr_addr.sa_data));
} else {
mac_unicast_primary_get(mh,
(uint8_t *)ifreq.ifr_addr.sa_data);
}
/*
* The behaviour here in setting sa_family is consistent
* with what applications such as tcpdump would expect
* for a Linux PF_PACKET socket.
*/
ifreq.ifr_addr.sa_family = pfp_dl_to_arphrd(mac_type(mh));
break;
case SIOCGLIFHWADDR :
lifreq.lifr_type = 0;
sock = (struct sockaddr_dl *)&lifreq.lifr_addr;
if (mac_addr_len(mh) > sizeof (sock->sdl_data)) {
error = EPFNOSUPPORT;
break;
}
/*
* Fill in the sockaddr_dl with link layer details. Of note,
* the index is returned as 0 for a couple of reasons:
* (1) there is no public API that uses or requires it
* (2) the MAC index is currently 32bits and sdl_index is 16.
*/
sock->sdl_family = AF_LINK;
sock->sdl_index = 0;
sock->sdl_type = mac_type(mh);
sock->sdl_nlen = 0;
sock->sdl_alen = mac_addr_len(mh);
sock->sdl_slen = 0;
if (mac_addr_len(mh) == 0) {
(void) memset(sock->sdl_data, 0,
sizeof (sock->sdl_data));
} else {
mac_unicast_primary_get(mh, (uint8_t *)sock->sdl_data);
}
break;
default :
break;
}
mac_close(mh);
if (error == 0) {
/*
* Only the "GET" ioctls need to copy data back to userace.
*/
switch (cmd) {
case SIOCGLIFINDEX :
case SIOCGLIFFLAGS :
case SIOCGLIFMTU :
case SIOCGLIFHWADDR :
error = ddi_copyout(&lifreq, (void *)arg,
sizeof (lifreq), mod);
break;
case SIOCGIFINDEX :
case SIOCGIFFLAGS :
case SIOCGIFMTU :
case SIOCGIFHWADDR :
error = ddi_copyout(&ifreq, (void *)arg,
sizeof (ifreq), mod);
break;
default :
break;
}
}
return (error);
}
