diff options
Diffstat (limited to 'drivers/net/ethernet/sfc/tc.c')
| -rw-r--r-- | drivers/net/ethernet/sfc/tc.c | 476 |
1 files changed, 476 insertions, 0 deletions
diff --git a/drivers/net/ethernet/sfc/tc.c b/drivers/net/ethernet/sfc/tc.c index 039180c61c83..047322b04d4f 100644 --- a/drivers/net/ethernet/sfc/tc.c +++ b/drivers/net/ethernet/sfc/tc.c @@ -31,6 +31,9 @@ enum efx_encap_type efx_tc_indr_netdev_type(struct net_device *net_dev) return EFX_ENCAP_TYPE_NONE; } +#define EFX_TC_HDR_TYPE_TTL_MASK ((u32)0xff) +/* Hoplimit is stored in the most significant byte in the pedit ipv6 header action */ +#define EFX_TC_HDR_TYPE_HLIMIT_MASK ~((u32)0xff000000) #define EFX_EFV_PF NULL /* Look up the representor information (efv) for a device. * May return NULL for the PF (us), or an error pointer for a device that @@ -86,6 +89,12 @@ s64 efx_tc_flower_external_mport(struct efx_nic *efx, struct efx_rep *efv) return mport; } +static const struct rhashtable_params efx_tc_mac_ht_params = { + .key_len = offsetofend(struct efx_tc_mac_pedit_action, h_addr), + .key_offset = 0, + .head_offset = offsetof(struct efx_tc_mac_pedit_action, linkage), +}; + static const struct rhashtable_params efx_tc_encap_match_ht_params = { .key_len = offsetof(struct efx_tc_encap_match, linkage), .key_offset = 0, @@ -110,6 +119,56 @@ 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 *efx_tc_flower_get_mac(struct efx_nic *efx, + unsigned char h_addr[ETH_ALEN], + struct netlink_ext_ack *extack) +{ + struct efx_tc_mac_pedit_action *ped, *old; + int rc; + + ped = kzalloc(sizeof(*ped), GFP_USER); + if (!ped) + return ERR_PTR(-ENOMEM); + memcpy(ped->h_addr, h_addr, ETH_ALEN); + old = rhashtable_lookup_get_insert_fast(&efx->tc->mac_ht, + &ped->linkage, + efx_tc_mac_ht_params); + if (old) { + /* don't need our new entry */ + kfree(ped); + if (!refcount_inc_not_zero(&old->ref)) + return ERR_PTR(-EAGAIN); + /* existing entry found, ref taken */ + return old; + } + + rc = efx_mae_allocate_pedit_mac(efx, ped); + if (rc < 0) { + NL_SET_ERR_MSG_MOD(extack, "Failed to store pedit MAC address in hw"); + goto out_remove; + } + + /* ref and return */ + refcount_set(&ped->ref, 1); + return ped; +out_remove: + rhashtable_remove_fast(&efx->tc->mac_ht, &ped->linkage, + efx_tc_mac_ht_params); + kfree(ped); + return ERR_PTR(rc); +} + +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)) + return; /* still in use */ + rhashtable_remove_fast(&efx->tc->mac_ht, &ped->linkage, + efx_tc_mac_ht_params); + efx_mae_free_pedit_mac(efx, ped); + kfree(ped); +} + static void efx_tc_free_action_set(struct efx_nic *efx, struct efx_tc_action_set *act, bool in_hw) { @@ -135,6 +194,10 @@ 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); } @@ -697,6 +760,8 @@ static const char *efx_tc_encap_type_name(enum efx_encap_type typ) /* For details of action order constraints refer to SF-123102-TC-1ยง12.6.1 */ enum efx_tc_action_order { EFX_TC_AO_DECAP, + EFX_TC_AO_DEC_TTL, + EFX_TC_AO_PEDIT_MAC_ADDRS, EFX_TC_AO_VLAN_POP, EFX_TC_AO_VLAN_PUSH, EFX_TC_AO_COUNT, @@ -711,6 +776,15 @@ 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; + + /* Decrementing ttl must not happen before DECAP */ + if (act->do_ttl_dec) + return false; fallthrough; case EFX_TC_AO_VLAN_POP: if (act->vlan_pop >= 2) @@ -730,12 +804,17 @@ 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; fallthrough; case EFX_TC_AO_DELIVER: return !act->deliver; + case EFX_TC_AO_DEC_TTL: + if (act->encap_md) + return false; + return !act->do_ttl_dec; default: /* Bad caller. Whatever they wanted to do, say they can't. */ WARN_ON_ONCE(1); @@ -900,6 +979,375 @@ 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; +} + +static int efx_tc_pedit_add(struct efx_nic *efx, struct efx_tc_action_set *act, + const struct flow_action_entry *fa, + struct netlink_ext_ack *extack) +{ + switch (fa->mangle.htype) { + case FLOW_ACT_MANGLE_HDR_TYPE_IP4: + switch (fa->mangle.offset) { + case offsetof(struct iphdr, ttl): + /* check that pedit applies to ttl only */ + if (fa->mangle.mask != ~EFX_TC_HDR_TYPE_TTL_MASK) + break; + + /* Adding 0xff is equivalent to decrementing the ttl. + * Other added values are not supported. + */ + if ((fa->mangle.val & EFX_TC_HDR_TYPE_TTL_MASK) != U8_MAX) + break; + + /* check that we do not decrement ttl twice */ + if (!efx_tc_flower_action_order_ok(act, + EFX_TC_AO_DEC_TTL)) { + NL_SET_ERR_MSG_MOD(extack, "Unsupported: multiple dec ttl"); + return -EOPNOTSUPP; + } + act->do_ttl_dec = 1; + return 0; + default: + break; + } + break; + case FLOW_ACT_MANGLE_HDR_TYPE_IP6: + switch (fa->mangle.offset) { + case round_down(offsetof(struct ipv6hdr, hop_limit), 4): + /* check that pedit applies to hoplimit only */ + if (fa->mangle.mask != EFX_TC_HDR_TYPE_HLIMIT_MASK) + break; + + /* Adding 0xff is equivalent to decrementing the hoplimit. + * Other added values are not supported. + */ + if ((fa->mangle.val >> 24) != U8_MAX) + break; + + /* check that we do not decrement hoplimit twice */ + if (!efx_tc_flower_action_order_ok(act, + EFX_TC_AO_DEC_TTL)) { + NL_SET_ERR_MSG_MOD(extack, "Unsupported: multiple dec ttl"); + return -EOPNOTSUPP; + } + act->do_ttl_dec = 1; + return 0; + default: + break; + } + break; + default: + break; + } + + NL_SET_ERR_MSG_FMT_MOD(extack, + "Unsupported: ttl add action type %x %x %x/%x", + fa->mangle.htype, fa->mangle.offset, + fa->mangle.val, fa->mangle.mask); + return -EOPNOTSUPP; +} + +/** + * 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 + * @match: original match used along with the mangle action + * + * 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, + struct efx_tc_match *match) +{ + __le32 mac32; + __le16 mac16; + u8 tr_ttl; + + 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; + case FLOW_ACT_MANGLE_HDR_TYPE_IP4: + switch (fa->mangle.offset) { + case offsetof(struct iphdr, ttl): + /* we currently only support pedit IP4 when it applies + * to TTL and then only when it can be achieved with a + * decrement ttl action + */ + + /* check that pedit applies to ttl only */ + if (fa->mangle.mask != ~EFX_TC_HDR_TYPE_TTL_MASK) { + NL_SET_ERR_MSG_FMT_MOD(extack, + "Unsupported: mask (%#x) out of range, only support mangle action on ipv4.ttl", + fa->mangle.mask); + return -EOPNOTSUPP; + } + + /* we can only convert to a dec ttl when we have an + * exact match on the ttl field + */ + if (match->mask.ip_ttl != U8_MAX) { + NL_SET_ERR_MSG_FMT_MOD(extack, + "Unsupported: only support mangle ipv4.ttl when we have an exact match on ttl, mask used for match (%#x)", + match->mask.ip_ttl); + return -EOPNOTSUPP; + } + + /* check that we don't try to decrement 0, which equates + * to setting the ttl to 0xff + */ + if (match->value.ip_ttl == 0) { + NL_SET_ERR_MSG_MOD(extack, + "Unsupported: we cannot decrement ttl past 0"); + return -EOPNOTSUPP; + } + + /* check that we do not decrement ttl twice */ + if (!efx_tc_flower_action_order_ok(act, + EFX_TC_AO_DEC_TTL)) { + NL_SET_ERR_MSG_MOD(extack, + "Unsupported: multiple dec ttl"); + return -EOPNOTSUPP; + } + + /* check pedit can be achieved with decrement action */ + tr_ttl = match->value.ip_ttl - 1; + if ((fa->mangle.val & EFX_TC_HDR_TYPE_TTL_MASK) == tr_ttl) { + act->do_ttl_dec = 1; + return 0; + } + + fallthrough; + default: + NL_SET_ERR_MSG_FMT_MOD(extack, + "Unsupported: only support mangle on the ttl field (offset is %u)", + fa->mangle.offset); + return -EOPNOTSUPP; + } + break; + case FLOW_ACT_MANGLE_HDR_TYPE_IP6: + switch (fa->mangle.offset) { + case round_down(offsetof(struct ipv6hdr, hop_limit), 4): + /* we currently only support pedit IP6 when it applies + * to the hoplimit and then only when it can be achieved + * with a decrement hoplimit action + */ + + /* check that pedit applies to ttl only */ + if (fa->mangle.mask != EFX_TC_HDR_TYPE_HLIMIT_MASK) { + NL_SET_ERR_MSG_FMT_MOD(extack, + "Unsupported: mask (%#x) out of range, only support mangle action on ipv6.hop_limit", + fa->mangle.mask); + + return -EOPNOTSUPP; + } + + /* we can only convert to a dec ttl when we have an + * exact match on the ttl field + */ + if (match->mask.ip_ttl != U8_MAX) { + NL_SET_ERR_MSG_FMT_MOD(extack, + "Unsupported: only support mangle ipv6.hop_limit when we have an exact match on ttl, mask used for match (%#x)", + match->mask.ip_ttl); + return -EOPNOTSUPP; + } + + /* check that we don't try to decrement 0, which equates + * to setting the ttl to 0xff + */ + if (match->value.ip_ttl == 0) { + NL_SET_ERR_MSG_MOD(extack, + "Unsupported: we cannot decrement hop_limit past 0"); + return -EOPNOTSUPP; + } + + /* check that we do not decrement hoplimit twice */ + if (!efx_tc_flower_action_order_ok(act, + EFX_TC_AO_DEC_TTL)) { + NL_SET_ERR_MSG_MOD(extack, + "Unsupported: multiple dec ttl"); + return -EOPNOTSUPP; + } + + /* check pedit can be achieved with decrement action */ + tr_ttl = match->value.ip_ttl - 1; + if ((fa->mangle.val >> 24) == tr_ttl) { + act->do_ttl_dec = 1; + return 0; + } + + fallthrough; + default: + NL_SET_ERR_MSG_FMT_MOD(extack, + "Unsupported: only support mangle on the hop_limit field"); + return -EOPNOTSUPP; + } + 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) @@ -1295,6 +1743,7 @@ 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; @@ -1631,6 +2080,16 @@ 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_ADD: + rc = efx_tc_pedit_add(efx, act, fa, extack); + if (rc < 0) + goto release; + break; + case FLOW_ACTION_MANGLE: + rc = efx_tc_mangle(efx, act, fa, &mung, extack, &match); + if (rc < 0) + goto release; + break; case FLOW_ACTION_TUNNEL_ENCAP: if (encap_info) { /* Can't specify encap multiple times. @@ -1670,6 +2129,9 @@ 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) @@ -2156,6 +2618,14 @@ static void efx_tc_lhs_free(void *ptr, void *arg) kfree(rule); } +static void efx_tc_mac_free(void *ptr, void *__unused) +{ + struct efx_tc_mac_pedit_action *ped = ptr; + + WARN_ON(refcount_read(&ped->ref)); + kfree(ped); +} + static void efx_tc_flow_free(void *ptr, void *arg) { struct efx_tc_flow_rule *rule = ptr; @@ -2196,6 +2666,9 @@ int efx_init_struct_tc(struct efx_nic *efx) rc = efx_tc_init_counters(efx); if (rc < 0) goto fail_counters; + rc = rhashtable_init(&efx->tc->mac_ht, &efx_tc_mac_ht_params); + if (rc < 0) + goto fail_mac_ht; rc = rhashtable_init(&efx->tc->encap_match_ht, &efx_tc_encap_match_ht_params); if (rc < 0) goto fail_encap_match_ht; @@ -2233,6 +2706,8 @@ fail_lhs_rule_ht: fail_match_action_ht: rhashtable_destroy(&efx->tc->encap_match_ht); fail_encap_match_ht: + rhashtable_destroy(&efx->tc->mac_ht); +fail_mac_ht: efx_tc_destroy_counters(efx); fail_counters: efx_tc_destroy_encap_actions(efx); @@ -2268,6 +2743,7 @@ void efx_fini_struct_tc(struct efx_nic *efx) rhashtable_free_and_destroy(&efx->tc->recirc_ht, efx_tc_recirc_free, efx); WARN_ON(!ida_is_empty(&efx->tc->recirc_ida)); ida_destroy(&efx->tc->recirc_ida); + rhashtable_free_and_destroy(&efx->tc->mac_ht, efx_tc_mac_free, NULL); efx_tc_fini_counters(efx); efx_tc_fini_encap_actions(efx); mutex_unlock(&efx->tc->mutex); |