@@ -116,7 +116,7 @@ static const struct rhashtable_params efx_tc_recirc_ht_params = {
.head_offset = offsetof(struct efx_tc_recirc_id, linkage),
};
-static struct efx_tc_mac_pedit_action __maybe_unused *efx_tc_flower_get_mac(struct efx_nic *efx,
+static struct efx_tc_mac_pedit_action *efx_tc_flower_get_mac(struct efx_nic *efx,
unsigned char h_addr[ETH_ALEN],
struct netlink_ext_ack *extack)
{
@@ -155,7 +155,7 @@ static struct efx_tc_mac_pedit_action __maybe_unused *efx_tc_flower_get_mac(stru
return ERR_PTR(rc);
}
-static void __maybe_unused efx_tc_flower_put_mac(struct efx_nic *efx,
+static void efx_tc_flower_put_mac(struct efx_nic *efx,
struct efx_tc_mac_pedit_action *ped)
{
if (!refcount_dec_and_test(&ped->ref))
@@ -191,6 +191,10 @@ static void efx_tc_free_action_set(struct efx_nic *efx,
list_del(&act->encap_user);
efx_tc_flower_release_encap_md(efx, act->encap_md);
}
+ if (act->src_mac)
+ efx_tc_flower_put_mac(efx, act->src_mac);
+ if (act->dst_mac)
+ efx_tc_flower_put_mac(efx, act->dst_mac);
kfree(act);
}
@@ -753,6 +757,7 @@ static const char *efx_tc_encap_type_name(enum efx_encap_type typ)
/* For details of action order constraints refer to SF-123102-TC-1§12.6.1 */
enum efx_tc_action_order {
EFX_TC_AO_DECAP,
+ EFX_TC_AO_PEDIT_MAC_ADDRS,
EFX_TC_AO_VLAN_POP,
EFX_TC_AO_VLAN_PUSH,
EFX_TC_AO_COUNT,
@@ -767,6 +772,11 @@ static bool efx_tc_flower_action_order_ok(const struct efx_tc_action_set *act,
case EFX_TC_AO_DECAP:
if (act->decap)
return false;
+ /* PEDIT_MAC_ADDRS must not happen before DECAP, though it
+ * can wait until much later
+ */
+ if (act->dst_mac || act->src_mac)
+ return false;
fallthrough;
case EFX_TC_AO_VLAN_POP:
if (act->vlan_pop >= 2)
@@ -786,6 +796,7 @@ static bool efx_tc_flower_action_order_ok(const struct efx_tc_action_set *act,
if (act->count)
return false;
fallthrough;
+ case EFX_TC_AO_PEDIT_MAC_ADDRS:
case EFX_TC_AO_ENCAP:
if (act->encap_md)
return false;
@@ -956,6 +967,191 @@ static void efx_tc_flower_release_lhs_actions(struct efx_nic *efx,
efx_tc_flower_put_counter_index(efx, act->count);
}
+/**
+ * struct efx_tc_mangler_state - accumulates 32-bit pedits into fields
+ *
+ * @dst_mac_32: dst_mac[0:3] has been populated
+ * @dst_mac_16: dst_mac[4:5] has been populated
+ * @src_mac_16: src_mac[0:1] has been populated
+ * @src_mac_32: src_mac[2:5] has been populated
+ * @dst_mac: h_dest field of ethhdr
+ * @src_mac: h_source field of ethhdr
+ *
+ * Since FLOW_ACTION_MANGLE comes in 32-bit chunks that do not
+ * necessarily equate to whole fields of the packet header, this
+ * structure is used to hold the cumulative effect of the partial
+ * field pedits that have been processed so far.
+ */
+struct efx_tc_mangler_state {
+ u8 dst_mac_32:1; /* eth->h_dest[0:3] */
+ u8 dst_mac_16:1; /* eth->h_dest[4:5] */
+ u8 src_mac_16:1; /* eth->h_source[0:1] */
+ u8 src_mac_32:1; /* eth->h_source[2:5] */
+ unsigned char dst_mac[ETH_ALEN];
+ unsigned char src_mac[ETH_ALEN];
+};
+
+/** efx_tc_complete_mac_mangle() - pull complete field pedits out of @mung
+ * @efx: NIC we're installing a flow rule on
+ * @act: action set (cursor) to update
+ * @mung: accumulated partial mangles
+ * @extack: netlink extended ack for reporting errors
+ *
+ * Check @mung to find any combinations of partial mangles that can be
+ * combined into a complete packet field edit, add that edit to @act,
+ * and consume the partial mangles from @mung.
+ */
+
+static int efx_tc_complete_mac_mangle(struct efx_nic *efx,
+ struct efx_tc_action_set *act,
+ struct efx_tc_mangler_state *mung,
+ struct netlink_ext_ack *extack)
+{
+ struct efx_tc_mac_pedit_action *ped;
+
+ if (mung->dst_mac_32 && mung->dst_mac_16) {
+ ped = efx_tc_flower_get_mac(efx, mung->dst_mac, extack);
+ if (IS_ERR(ped))
+ return PTR_ERR(ped);
+
+ /* Check that we have not already populated dst_mac */
+ if (act->dst_mac)
+ efx_tc_flower_put_mac(efx, act->dst_mac);
+
+ act->dst_mac = ped;
+
+ /* consume the incomplete state */
+ mung->dst_mac_32 = 0;
+ mung->dst_mac_16 = 0;
+ }
+ if (mung->src_mac_16 && mung->src_mac_32) {
+ ped = efx_tc_flower_get_mac(efx, mung->src_mac, extack);
+ if (IS_ERR(ped))
+ return PTR_ERR(ped);
+
+ /* Check that we have not already populated src_mac */
+ if (act->src_mac)
+ efx_tc_flower_put_mac(efx, act->src_mac);
+
+ act->src_mac = ped;
+
+ /* consume the incomplete state */
+ mung->src_mac_32 = 0;
+ mung->src_mac_16 = 0;
+ }
+ return 0;
+}
+
+/**
+ * efx_tc_mangle() - handle a single 32-bit (or less) pedit
+ * @efx: NIC we're installing a flow rule on
+ * @act: action set (cursor) to update
+ * @fa: FLOW_ACTION_MANGLE action metadata
+ * @mung: accumulator for partial mangles
+ * @extack: netlink extended ack for reporting errors
+ *
+ * Identify the fields written by a FLOW_ACTION_MANGLE, and record
+ * the partial mangle state in @mung. If this mangle completes an
+ * earlier partial mangle, consume and apply to @act by calling
+ * efx_tc_complete_mac_mangle().
+ */
+
+static int efx_tc_mangle(struct efx_nic *efx, struct efx_tc_action_set *act,
+ const struct flow_action_entry *fa,
+ struct efx_tc_mangler_state *mung,
+ struct netlink_ext_ack *extack)
+{
+ __le32 mac32;
+ __le16 mac16;
+
+ switch (fa->mangle.htype) {
+ case FLOW_ACT_MANGLE_HDR_TYPE_ETH:
+ BUILD_BUG_ON(offsetof(struct ethhdr, h_dest) != 0);
+ BUILD_BUG_ON(offsetof(struct ethhdr, h_source) != 6);
+ if (!efx_tc_flower_action_order_ok(act, EFX_TC_AO_PEDIT_MAC_ADDRS)) {
+ NL_SET_ERR_MSG_MOD(extack,
+ "Pedit mangle mac action violates action order");
+ return -EOPNOTSUPP;
+ }
+ switch (fa->mangle.offset) {
+ case 0:
+ if (fa->mangle.mask) {
+ NL_SET_ERR_MSG_FMT_MOD(extack,
+ "Unsupported: mask (%#x) of eth.dst32 mangle",
+ fa->mangle.mask);
+ return -EOPNOTSUPP;
+ }
+ /* Ethernet address is little-endian */
+ mac32 = cpu_to_le32(fa->mangle.val);
+ memcpy(mung->dst_mac, &mac32, sizeof(mac32));
+ mung->dst_mac_32 = 1;
+ return efx_tc_complete_mac_mangle(efx, act, mung, extack);
+ case 4:
+ if (fa->mangle.mask == 0xffff) {
+ mac16 = cpu_to_le16(fa->mangle.val >> 16);
+ memcpy(mung->src_mac, &mac16, sizeof(mac16));
+ mung->src_mac_16 = 1;
+ } else if (fa->mangle.mask == 0xffff0000) {
+ mac16 = cpu_to_le16((u16)fa->mangle.val);
+ memcpy(mung->dst_mac + 4, &mac16, sizeof(mac16));
+ mung->dst_mac_16 = 1;
+ } else {
+ NL_SET_ERR_MSG_FMT_MOD(extack,
+ "Unsupported: mask (%#x) of eth+4 mangle is not high or low 16b",
+ fa->mangle.mask);
+ return -EOPNOTSUPP;
+ }
+ return efx_tc_complete_mac_mangle(efx, act, mung, extack);
+ case 8:
+ if (fa->mangle.mask) {
+ NL_SET_ERR_MSG_FMT_MOD(extack,
+ "Unsupported: mask (%#x) of eth.src32 mangle",
+ fa->mangle.mask);
+ return -EOPNOTSUPP;
+ }
+ mac32 = cpu_to_le32(fa->mangle.val);
+ memcpy(mung->src_mac + 2, &mac32, sizeof(mac32));
+ mung->src_mac_32 = 1;
+ return efx_tc_complete_mac_mangle(efx, act, mung, extack);
+ default:
+ NL_SET_ERR_MSG_FMT_MOD(extack, "Unsupported: mangle eth+%u %x/%x",
+ fa->mangle.offset, fa->mangle.val, fa->mangle.mask);
+ return -EOPNOTSUPP;
+ }
+ break;
+ default:
+ NL_SET_ERR_MSG_FMT_MOD(extack, "Unhandled mangle htype %u for action rule",
+ fa->mangle.htype);
+ return -EOPNOTSUPP;
+ }
+ return 0;
+}
+
+/**
+ * efx_tc_incomplete_mangle() - check for leftover partial pedits
+ * @mung: accumulator for partial mangles
+ * @extack: netlink extended ack for reporting errors
+ *
+ * Since the MAE can only overwrite whole fields, any partial
+ * field mangle left over on reaching packet delivery (mirred or
+ * end of TC actions) cannot be offloaded. Check for any such
+ * and reject them with -%EOPNOTSUPP.
+ */
+
+static int efx_tc_incomplete_mangle(struct efx_tc_mangler_state *mung,
+ struct netlink_ext_ack *extack)
+{
+ if (mung->dst_mac_32 || mung->dst_mac_16) {
+ NL_SET_ERR_MSG_MOD(extack, "Incomplete pedit of destination MAC address");
+ return -EOPNOTSUPP;
+ }
+ if (mung->src_mac_16 || mung->src_mac_32) {
+ NL_SET_ERR_MSG_MOD(extack, "Incomplete pedit of source MAC address");
+ return -EOPNOTSUPP;
+ }
+ return 0;
+}
+
static int efx_tc_flower_replace_foreign(struct efx_nic *efx,
struct net_device *net_dev,
struct flow_cls_offload *tc)
@@ -1351,6 +1547,7 @@ static int efx_tc_flower_replace(struct efx_nic *efx,
struct netlink_ext_ack *extack = tc->common.extack;
const struct ip_tunnel_info *encap_info = NULL;
struct efx_tc_flow_rule *rule = NULL, *old;
+ struct efx_tc_mangler_state mung = {};
struct efx_tc_action_set *act = NULL;
const struct flow_action_entry *fa;
struct efx_rep *from_efv, *to_efv;
@@ -1687,6 +1884,11 @@ static int efx_tc_flower_replace(struct efx_nic *efx,
act->vlan_proto[act->vlan_push] = fa->vlan.proto;
act->vlan_push++;
break;
+ case FLOW_ACTION_MANGLE:
+ rc = efx_tc_mangle(efx, act, fa, &mung, extack);
+ if (rc < 0)
+ goto release;
+ break;
case FLOW_ACTION_TUNNEL_ENCAP:
if (encap_info) {
/* Can't specify encap multiple times.
@@ -1726,6 +1928,9 @@ static int efx_tc_flower_replace(struct efx_nic *efx,
}
}
+ rc = efx_tc_incomplete_mangle(&mung, extack);
+ if (rc < 0)
+ goto release;
if (act) {
/* Not shot/redirected, so deliver to default dest */
if (from_efv == EFX_EFV_PF)