From 49ddf8e6e2347cffdcf83d1ca2d04ff929820178 Mon Sep 17 00:00:00 2001 From: Johannes Berg Date: Thu, 31 Mar 2016 20:02:10 +0300 Subject: [PATCH] mac80211: add fast-rx path The regular RX path has a lot of code, but with a few assumptions on the hardware it's possible to reduce the amount of code significantly. Currently the assumptions on the driver are the following: * hardware/driver reordering buffer (if supporting aggregation) * hardware/driver decryption & PN checking (if using encryption) * hardware/driver did de-duplication * hardware/driver did A-MSDU deaggregation * AP_LINK_PS is used (in AP mode) * no client powersave handling in mac80211 (in client mode) of which some are actually checked per packet: * de-duplication * PN checking * decryption and additionally packets must * not be A-MSDU (have been deaggregated by driver/device) * be data packets * not be fragmented * be unicast * have RFC 1042 header Additionally dynamically we assume: * no encryption or CCMP/GCMP, TKIP/WEP/other not allowed * station must be authorized * 4-addr format not enabled Some data needed for the RX path is cached in a new per-station "fast_rx" structure, so that we only need to look at this and the packet, no other memory when processing packets on the fast RX path. After doing the above per-packet checks, the data path collapses down to a pretty simple conversion function taking advantage of the data cached in the small fast_rx struct. This should speed up the RX processing, and will make it easier to reason about parallelizing RX (for which statistics will need to be per-CPU still.) Signed-off-by: Johannes Berg --- include/linux/ieee80211.h | 10 ++ net/mac80211/cfg.c | 10 +- net/mac80211/ieee80211_i.h | 5 + net/mac80211/key.c | 1 + net/mac80211/mlme.c | 9 ++ net/mac80211/rx.c | 351 +++++++++++++++++++++++++++++++++++++++++++++ net/mac80211/sta_info.c | 2 + net/mac80211/sta_info.h | 34 +++++ 8 files changed, 419 insertions(+), 3 deletions(-) diff --git a/include/linux/ieee80211.h b/include/linux/ieee80211.h index bf9706c..113bfc4 100644 --- a/include/linux/ieee80211.h +++ b/include/linux/ieee80211.h @@ -638,6 +638,16 @@ static inline bool ieee80211_is_first_frag(__le16 seq_ctrl) return (seq_ctrl & cpu_to_le16(IEEE80211_SCTL_FRAG)) == 0; } +/** + * ieee80211_is_frag - check if a frame is a fragment + * @hdr: 802.11 header of the frame + */ +static inline bool ieee80211_is_frag(struct ieee80211_hdr *hdr) +{ + return ieee80211_has_morefrags(hdr->frame_control) || + hdr->seq_ctrl & cpu_to_le16(IEEE80211_SCTL_FRAG); +} + struct ieee80211s_hdr { u8 flags; u8 ttl; diff --git a/net/mac80211/cfg.c b/net/mac80211/cfg.c index 62a90f2..fc4730b 100644 --- a/net/mac80211/cfg.c +++ b/net/mac80211/cfg.c @@ -65,11 +65,13 @@ static int ieee80211_change_iface(struct wiphy *wiphy, return ret; if (type == NL80211_IFTYPE_AP_VLAN && - params && params->use_4addr == 0) + params && params->use_4addr == 0) { RCU_INIT_POINTER(sdata->u.vlan.sta, NULL); - else if (type == NL80211_IFTYPE_STATION && - params && params->use_4addr >= 0) + ieee80211_check_fast_rx_iface(sdata); + } else if (type == NL80211_IFTYPE_STATION && + params && params->use_4addr >= 0) { sdata->u.mgd.use_4addr = params->use_4addr; + } if (sdata->vif.type == NL80211_IFTYPE_MONITOR && flags) { struct ieee80211_local *local = sdata->local; @@ -1367,6 +1369,7 @@ static int ieee80211_change_station(struct wiphy *wiphy, rcu_assign_pointer(vlansdata->u.vlan.sta, sta); new_4addr = true; + __ieee80211_check_fast_rx_iface(vlansdata); } if (sta->sdata->vif.type == NL80211_IFTYPE_AP_VLAN && @@ -1889,6 +1892,7 @@ static int ieee80211_change_bss(struct wiphy *wiphy, sdata->flags |= IEEE80211_SDATA_DONT_BRIDGE_PACKETS; else sdata->flags &= ~IEEE80211_SDATA_DONT_BRIDGE_PACKETS; + ieee80211_check_fast_rx_iface(sdata); } if (params->ht_opmode >= 0) { diff --git a/net/mac80211/ieee80211_i.h b/net/mac80211/ieee80211_i.h index c8945e2..6243109 100644 --- a/net/mac80211/ieee80211_i.h +++ b/net/mac80211/ieee80211_i.h @@ -1494,6 +1494,11 @@ u64 ieee80211_mgmt_tx_cookie(struct ieee80211_local *local); int ieee80211_attach_ack_skb(struct ieee80211_local *local, struct sk_buff *skb, u64 *cookie, gfp_t gfp); +void ieee80211_check_fast_rx(struct sta_info *sta); +void __ieee80211_check_fast_rx_iface(struct ieee80211_sub_if_data *sdata); +void ieee80211_check_fast_rx_iface(struct ieee80211_sub_if_data *sdata); +void ieee80211_clear_fast_rx(struct sta_info *sta); + /* STA code */ void ieee80211_sta_setup_sdata(struct ieee80211_sub_if_data *sdata); int ieee80211_mgd_auth(struct ieee80211_sub_if_data *sdata, diff --git a/net/mac80211/key.c b/net/mac80211/key.c index 3df7b03..edd6f29 100644 --- a/net/mac80211/key.c +++ b/net/mac80211/key.c @@ -338,6 +338,7 @@ static void ieee80211_key_replace(struct ieee80211_sub_if_data *sdata, } else { rcu_assign_pointer(sta->gtk[idx], new); } + ieee80211_check_fast_rx(sta); } else { defunikey = old && old == key_mtx_dereference(sdata->local, diff --git a/net/mac80211/mlme.c b/net/mac80211/mlme.c index 2112df4..d3c75ac 100644 --- a/net/mac80211/mlme.c +++ b/net/mac80211/mlme.c @@ -2217,6 +2217,7 @@ static void ieee80211_mgd_probe_ap_send(struct ieee80211_sub_if_data *sdata) const u8 *ssid; u8 *dst = ifmgd->associated->bssid; u8 unicast_limit = max(1, max_probe_tries - 3); + struct sta_info *sta; /* * Try sending broadcast probe requests for the last three @@ -2235,6 +2236,14 @@ static void ieee80211_mgd_probe_ap_send(struct ieee80211_sub_if_data *sdata) */ ifmgd->probe_send_count++; + if (dst) { + mutex_lock(&sdata->local->sta_mtx); + sta = sta_info_get(sdata, dst); + if (!WARN_ON(!sta)) + ieee80211_check_fast_rx(sta); + mutex_unlock(&sdata->local->sta_mtx); + } + if (ieee80211_hw_check(&sdata->local->hw, REPORTS_TX_ACK_STATUS)) { ifmgd->nullfunc_failed = false; ieee80211_send_nullfunc(sdata->local, sdata, false); diff --git a/net/mac80211/rx.c b/net/mac80211/rx.c index 2863832..96f8bbf 100644 --- a/net/mac80211/rx.c +++ b/net/mac80211/rx.c @@ -3508,6 +3508,342 @@ static bool ieee80211_accept_frame(struct ieee80211_rx_data *rx) return false; } +void ieee80211_check_fast_rx(struct sta_info *sta) +{ + struct ieee80211_sub_if_data *sdata = sta->sdata; + struct ieee80211_local *local = sdata->local; + struct ieee80211_key *key; + struct ieee80211_fast_rx fastrx = { + .dev = sdata->dev, + .vif_type = sdata->vif.type, + .control_port_protocol = sdata->control_port_protocol, + }, *old, *new = NULL; + bool assign = false; + + /* use sparse to check that we don't return without updating */ + __acquire(check_fast_rx); + + BUILD_BUG_ON(sizeof(fastrx.rfc1042_hdr) != sizeof(rfc1042_header)); + BUILD_BUG_ON(sizeof(fastrx.rfc1042_hdr) != ETH_ALEN); + ether_addr_copy(fastrx.rfc1042_hdr, rfc1042_header); + ether_addr_copy(fastrx.vif_addr, sdata->vif.addr); + + /* fast-rx doesn't do reordering */ + if (ieee80211_hw_check(&local->hw, AMPDU_AGGREGATION) && + !ieee80211_hw_check(&local->hw, SUPPORTS_REORDERING_BUFFER)) + goto clear; + + switch (sdata->vif.type) { + case NL80211_IFTYPE_STATION: + /* 4-addr is harder to deal with, later maybe */ + if (sdata->u.mgd.use_4addr) + goto clear; + /* software powersave is a huge mess, avoid all of it */ + if (ieee80211_hw_check(&local->hw, PS_NULLFUNC_STACK)) + goto clear; + if (ieee80211_hw_check(&local->hw, SUPPORTS_PS) && + !ieee80211_hw_check(&local->hw, SUPPORTS_DYNAMIC_PS)) + goto clear; + if (sta->sta.tdls) { + fastrx.da_offs = offsetof(struct ieee80211_hdr, addr1); + fastrx.sa_offs = offsetof(struct ieee80211_hdr, addr2); + fastrx.expected_ds_bits = 0; + } else { + fastrx.sta_notify = sdata->u.mgd.probe_send_count > 0; + fastrx.da_offs = offsetof(struct ieee80211_hdr, addr1); + fastrx.sa_offs = offsetof(struct ieee80211_hdr, addr3); + fastrx.expected_ds_bits = + cpu_to_le16(IEEE80211_FCTL_FROMDS); + } + break; + case NL80211_IFTYPE_AP_VLAN: + case NL80211_IFTYPE_AP: + /* parallel-rx requires this, at least with calls to + * ieee80211_sta_ps_transition() + */ + if (!ieee80211_hw_check(&local->hw, AP_LINK_PS)) + goto clear; + fastrx.da_offs = offsetof(struct ieee80211_hdr, addr3); + fastrx.sa_offs = offsetof(struct ieee80211_hdr, addr2); + fastrx.expected_ds_bits = cpu_to_le16(IEEE80211_FCTL_TODS); + + fastrx.internal_forward = + !(sdata->flags & IEEE80211_SDATA_DONT_BRIDGE_PACKETS) && + (sdata->vif.type != NL80211_IFTYPE_AP_VLAN || + !sdata->u.vlan.sta); + break; + default: + goto clear; + } + + if (!test_sta_flag(sta, WLAN_STA_AUTHORIZED)) + goto clear; + + rcu_read_lock(); + key = rcu_dereference(sta->ptk[sta->ptk_idx]); + if (key) { + switch (key->conf.cipher) { + case WLAN_CIPHER_SUITE_TKIP: + /* we don't want to deal with MMIC in fast-rx */ + goto clear_rcu; + case WLAN_CIPHER_SUITE_CCMP: + case WLAN_CIPHER_SUITE_CCMP_256: + case WLAN_CIPHER_SUITE_GCMP: + case WLAN_CIPHER_SUITE_GCMP_256: + break; + default: + /* we also don't want to deal with WEP or cipher scheme + * since those require looking up the key idx in the + * frame, rather than assuming the PTK is used + * (we need to revisit this once we implement the real + * PTK index, which is now valid in the spec, but we + * haven't implemented that part yet) + */ + goto clear_rcu; + } + + fastrx.key = true; + fastrx.icv_len = key->conf.icv_len; + } + + assign = true; + clear_rcu: + rcu_read_unlock(); + clear: + __release(check_fast_rx); + + if (assign) + new = kmemdup(&fastrx, sizeof(fastrx), GFP_KERNEL); + + spin_lock_bh(&sta->lock); + old = rcu_dereference_protected(sta->fast_rx, true); + rcu_assign_pointer(sta->fast_rx, new); + spin_unlock_bh(&sta->lock); + + if (old) + kfree_rcu(old, rcu_head); +} + +void ieee80211_clear_fast_rx(struct sta_info *sta) +{ + struct ieee80211_fast_rx *old; + + spin_lock_bh(&sta->lock); + old = rcu_dereference_protected(sta->fast_rx, true); + RCU_INIT_POINTER(sta->fast_rx, NULL); + spin_unlock_bh(&sta->lock); + + if (old) + kfree_rcu(old, rcu_head); +} + +void __ieee80211_check_fast_rx_iface(struct ieee80211_sub_if_data *sdata) +{ + struct ieee80211_local *local = sdata->local; + struct sta_info *sta; + + lockdep_assert_held(&local->sta_mtx); + + list_for_each_entry_rcu(sta, &local->sta_list, list) { + if (sdata != sta->sdata && + (!sta->sdata->bss || sta->sdata->bss != sdata->bss)) + continue; + ieee80211_check_fast_rx(sta); + } +} + +void ieee80211_check_fast_rx_iface(struct ieee80211_sub_if_data *sdata) +{ + struct ieee80211_local *local = sdata->local; + + mutex_lock(&local->sta_mtx); + __ieee80211_check_fast_rx_iface(sdata); + mutex_unlock(&local->sta_mtx); +} + +static bool ieee80211_invoke_fast_rx(struct ieee80211_rx_data *rx, + struct ieee80211_fast_rx *fast_rx) +{ + struct sk_buff *skb = rx->skb; + struct ieee80211_hdr *hdr = (void *)skb->data; + struct ieee80211_rx_status *status = IEEE80211_SKB_RXCB(skb); + struct sta_info *sta = rx->sta; + int orig_len = skb->len; + int snap_offs = ieee80211_hdrlen(hdr->frame_control); + struct { + u8 snap[sizeof(rfc1042_header)]; + __be16 proto; + } *payload __aligned(2); + struct { + u8 da[ETH_ALEN]; + u8 sa[ETH_ALEN]; + } addrs __aligned(2); + + /* for parallel-rx, we need to have DUP_VALIDATED, otherwise we write + * to a common data structure; drivers can implement that per queue + * but we don't have that information in mac80211 + */ + if (!(status->flag & RX_FLAG_DUP_VALIDATED)) + return false; + +#define FAST_RX_CRYPT_FLAGS (RX_FLAG_PN_VALIDATED | RX_FLAG_DECRYPTED) + + /* If using encryption, we also need to have: + * - PN_VALIDATED: similar, but the implementation is tricky + * - DECRYPTED: necessary for PN_VALIDATED + */ + if (fast_rx->key && + (status->flag & FAST_RX_CRYPT_FLAGS) != FAST_RX_CRYPT_FLAGS) + return false; + + /* we don't deal with A-MSDU deaggregation here */ + if (status->rx_flags & IEEE80211_RX_AMSDU) + return false; + + if (unlikely(!ieee80211_is_data_present(hdr->frame_control))) + return false; + + if (unlikely(ieee80211_is_frag(hdr))) + return false; + + /* Since our interface address cannot be multicast, this + * implicitly also rejects multicast frames without the + * explicit check. + * + * We shouldn't get any *data* frames not addressed to us + * (AP mode will accept multicast *management* frames), but + * punting here will make it go through the full checks in + * ieee80211_accept_frame(). + */ + if (!ether_addr_equal(fast_rx->vif_addr, hdr->addr1)) + return false; + + if ((hdr->frame_control & cpu_to_le16(IEEE80211_FCTL_FROMDS | + IEEE80211_FCTL_TODS)) != + fast_rx->expected_ds_bits) + goto drop; + + /* assign the key to drop unencrypted frames (later) + * and strip the IV/MIC if necessary + */ + if (fast_rx->key && !(status->flag & RX_FLAG_IV_STRIPPED)) { + /* GCMP header length is the same */ + snap_offs += IEEE80211_CCMP_HDR_LEN; + } + + if (!pskb_may_pull(skb, snap_offs + sizeof(*payload))) + goto drop; + payload = (void *)(skb->data + snap_offs); + + if (!ether_addr_equal(payload->snap, fast_rx->rfc1042_hdr)) + return false; + + /* Don't handle these here since they require special code. + * Accept AARP and IPX even though they should come with a + * bridge-tunnel header - but if we get them this way then + * there's little point in discarding them. + */ + if (unlikely(payload->proto == cpu_to_be16(ETH_P_TDLS) || + payload->proto == fast_rx->control_port_protocol)) + return false; + + /* after this point, don't punt to the slowpath! */ + + if (rx->key && !(status->flag & RX_FLAG_MIC_STRIPPED) && + pskb_trim(skb, skb->len - fast_rx->icv_len)) + goto drop; + + if (unlikely(fast_rx->sta_notify)) { + ieee80211_sta_rx_notify(rx->sdata, hdr); + fast_rx->sta_notify = false; + } + + /* statistics part of ieee80211_rx_h_sta_process() */ + sta->rx_stats.last_rx = jiffies; + sta->rx_stats.last_rate = sta_stats_encode_rate(status); + + sta->rx_stats.fragments++; + + if (!(status->flag & RX_FLAG_NO_SIGNAL_VAL)) { + sta->rx_stats.last_signal = status->signal; + ewma_signal_add(&sta->rx_stats_avg.signal, -status->signal); + } + + if (status->chains) { + int i; + + sta->rx_stats.chains = status->chains; + for (i = 0; i < ARRAY_SIZE(status->chain_signal); i++) { + int signal = status->chain_signal[i]; + + if (!(status->chains & BIT(i))) + continue; + + sta->rx_stats.chain_signal_last[i] = signal; + ewma_signal_add(&sta->rx_stats_avg.chain_signal[i], + -signal); + } + } + /* end of statistics */ + + if (rx->key && !ieee80211_has_protected(hdr->frame_control)) + goto drop; + + /* do the header conversion - first grab the addresses */ + ether_addr_copy(addrs.da, skb->data + fast_rx->da_offs); + ether_addr_copy(addrs.sa, skb->data + fast_rx->sa_offs); + /* remove the SNAP but leave the ethertype */ + skb_pull(skb, snap_offs + sizeof(rfc1042_header)); + /* push the addresses in front */ + memcpy(skb_push(skb, sizeof(addrs)), &addrs, sizeof(addrs)); + + skb->dev = fast_rx->dev; + + ieee80211_rx_stats(fast_rx->dev, skb->len); + + /* The seqno index has the same property as needed + * for the rx_msdu field, i.e. it is IEEE80211_NUM_TIDS + * for non-QoS-data frames. Here we know it's a data + * frame, so count MSDUs. + */ + u64_stats_update_begin(&sta->rx_stats.syncp); + sta->rx_stats.msdu[rx->seqno_idx]++; + sta->rx_stats.bytes += orig_len; + u64_stats_update_end(&sta->rx_stats.syncp); + + if (fast_rx->internal_forward) { + struct sta_info *dsta = sta_info_get(rx->sdata, skb->data); + + if (dsta) { + /* + * Send to wireless media and increase priority by 256 + * to keep the received priority instead of + * reclassifying the frame (see cfg80211_classify8021d). + */ + skb->priority += 256; + skb->protocol = htons(ETH_P_802_3); + skb_reset_network_header(skb); + skb_reset_mac_header(skb); + dev_queue_xmit(skb); + return true; + } + } + + /* deliver to local stack */ + skb->protocol = eth_type_trans(skb, fast_rx->dev); + memset(skb->cb, 0, sizeof(skb->cb)); + if (rx->napi) + napi_gro_receive(rx->napi, skb); + else + netif_receive_skb(skb); + + return true; + drop: + dev_kfree_skb(skb); + sta->rx_stats.dropped++; + return true; +} + /* * This function returns whether or not the SKB * was destined for RX processing or not, which, @@ -3522,6 +3858,21 @@ static bool ieee80211_prepare_and_rx_handle(struct ieee80211_rx_data *rx, rx->skb = skb; + /* See if we can do fast-rx; if we have to copy we already lost, + * so punt in that case. We should never have to deliver a data + * frame to multiple interfaces anyway. + * + * We skip the ieee80211_accept_frame() call and do the necessary + * checking inside ieee80211_invoke_fast_rx(). + */ + if (consume && rx->sta) { + struct ieee80211_fast_rx *fast_rx; + + fast_rx = rcu_dereference(rx->sta->fast_rx); + if (fast_rx && ieee80211_invoke_fast_rx(rx, fast_rx)) + return true; + } + if (!ieee80211_accept_frame(rx)) return false; diff --git a/net/mac80211/sta_info.c b/net/mac80211/sta_info.c index bdd303e..a0ce7e4 100644 --- a/net/mac80211/sta_info.c +++ b/net/mac80211/sta_info.c @@ -1874,6 +1874,7 @@ int sta_info_move_state(struct sta_info *sta, atomic_dec(&sta->sdata->bss->num_mcast_sta); clear_bit(WLAN_STA_AUTHORIZED, &sta->_flags); ieee80211_clear_fast_xmit(sta); + ieee80211_clear_fast_rx(sta); } break; case IEEE80211_STA_AUTHORIZED: @@ -1884,6 +1885,7 @@ int sta_info_move_state(struct sta_info *sta, atomic_inc(&sta->sdata->bss->num_mcast_sta); set_bit(WLAN_STA_AUTHORIZED, &sta->_flags); ieee80211_check_fast_xmit(sta); + ieee80211_check_fast_rx(sta); } break; default: diff --git a/net/mac80211/sta_info.h b/net/mac80211/sta_info.h index 7c23b57..a0a0660 100644 --- a/net/mac80211/sta_info.h +++ b/net/mac80211/sta_info.h @@ -286,6 +286,38 @@ struct ieee80211_fast_tx { }; /** + * struct ieee80211_fast_rx - RX fastpath information + * @dev: netdevice for reporting the SKB + * @vif_type: (P2P-less) interface type of the original sdata (sdata->vif.type) + * @vif_addr: interface address + * @rfc1042_hdr: copy of the RFC 1042 SNAP header (to have in cache) + * @control_port_protocol: control port protocol copied from sdata + * @expected_ds_bits: from/to DS bits expected + * @icv_len: length of the MIC if present + * @key: bool indicating encryption is expected (key is set) + * @sta_notify: notify the MLME code (once) + * @internal_forward: forward froms internally on AP/VLAN type interfaces + * @da_offs: offset of the DA in the header (for header conversion) + * @sa_offs: offset of the SA in the header (for header conversion) + * @rcu_head: RCU head for freeing this structure + */ +struct ieee80211_fast_rx { + struct net_device *dev; + enum nl80211_iftype vif_type; + u8 vif_addr[ETH_ALEN] __aligned(2); + u8 rfc1042_hdr[6] __aligned(2); + __be16 control_port_protocol; + __le16 expected_ds_bits; + u8 icv_len; + u8 key:1, + sta_notify:1, + internal_forward:1; + u8 da_offs, sa_offs; + + struct rcu_head rcu_head; +}; + +/** * struct mesh_sta - mesh STA information * @plink_lock: serialize access to plink fields * @llid: Local link ID @@ -391,6 +423,7 @@ DECLARE_EWMA(signal, 1024, 8) * @cipher_scheme: optional cipher scheme for this station * @reserved_tid: reserved TID (if any, otherwise IEEE80211_TID_UNRESERVED) * @fast_tx: TX fastpath information + * @fast_rx: RX fastpath information * @tdls_chandef: a TDLS peer can have a wider chandef that is compatible to * the BSS one. * @tx_stats: TX statistics @@ -414,6 +447,7 @@ struct sta_info { spinlock_t lock; struct ieee80211_fast_tx __rcu *fast_tx; + struct ieee80211_fast_rx __rcu *fast_rx; #ifdef CONFIG_MAC80211_MESH struct mesh_sta *mesh; -- 2.7.4