From: Felix Fietkau Date: Sun, 17 Mar 2019 18:11:30 +0100 Subject: [PATCH] mac80211: optimize skb resizing When forwarding unicast packets from ethernet to batman-adv over 802.11s (with forwarding disabled), the typical required headroom to transmit encrypted packets on mt76 is 32 (802.11) + 6 (802.11s) + 8 (CCMP) + 2 (padding) + 6 (LLC) + 18 (batman-adv) - 14 (old ethernet header) = 58 bytes. On systems where NET_SKB_PAD is 64 this leads to a call to pskb_expand_head for every packet, since mac80211 also tries to allocate 16 bytes status headroom for radiotap headers. This patch fixes these unnecessary reallocations by only requiring the extra status headroom in ieee80211_tx_monitor() If however a reallocation happens before that call, the status headroom gets added there as well, in order to avoid double reallocation. The patch also cleans up the code by moving the headroom calculation to ieee80211_skb_resize. Signed-off-by: Felix Fietkau --- --- a/net/mac80211/ieee80211_i.h +++ b/net/mac80211/ieee80211_i.h @@ -1779,6 +1779,9 @@ int ieee80211_tx_control_port(struct wip const u8 *dest, __be16 proto, bool unencrypted); int ieee80211_probe_mesh_link(struct wiphy *wiphy, struct net_device *dev, const u8 *buf, size_t len); +int ieee80211_skb_resize(struct ieee80211_local *local, + struct ieee80211_sub_if_data *sdata, + struct sk_buff *skb, int hdrlen, int hdr_add); /* HT */ void ieee80211_apply_htcap_overrides(struct ieee80211_sub_if_data *sdata, --- a/net/mac80211/status.c +++ b/net/mac80211/status.c @@ -655,6 +655,11 @@ void ieee80211_tx_monitor(struct ieee802 struct net_device *prev_dev = NULL; int rtap_len; + if (ieee80211_skb_resize(local, NULL, skb, 0, 0)) { + dev_kfree_skb(skb); + return; + } + /* send frame to monitor interfaces now */ rtap_len = ieee80211_tx_radiotap_len(info); if (WARN_ON_ONCE(skb_headroom(skb) < rtap_len)) { --- a/net/mac80211/tx.c +++ b/net/mac80211/tx.c @@ -1936,37 +1936,53 @@ static bool ieee80211_tx(struct ieee8021 } /* device xmit handlers */ - -static int ieee80211_skb_resize(struct ieee80211_sub_if_data *sdata, - struct sk_buff *skb, - int head_need, bool may_encrypt) +int ieee80211_skb_resize(struct ieee80211_local *local, + struct ieee80211_sub_if_data *sdata, + struct sk_buff *skb, int hdr_len, int hdr_extra) { - struct ieee80211_local *local = sdata->local; + struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); struct ieee80211_hdr *hdr; - bool enc_tailroom; - int tail_need = 0; - - hdr = (struct ieee80211_hdr *) skb->data; - enc_tailroom = may_encrypt && - (sdata->crypto_tx_tailroom_needed_cnt || - ieee80211_is_mgmt(hdr->frame_control)); - - if (enc_tailroom) { - tail_need = IEEE80211_ENCRYPT_TAILROOM; - tail_need -= skb_tailroom(skb); - tail_need = max_t(int, tail_need, 0); + int head_need, head_max; + int tail_need, tail_max; + bool enc_tailroom = false; + + if (sdata && !hdr_len && + !(info->flags & IEEE80211_TX_INTFL_DONT_ENCRYPT)) { + hdr = (struct ieee80211_hdr *) skb->data; + enc_tailroom = (sdata->crypto_tx_tailroom_needed_cnt || + ieee80211_is_mgmt(hdr->frame_control)); + hdr_len += sdata->encrypt_headroom; + } + + head_need = head_max = hdr_len; + tail_need = tail_max = 0; + if (!sdata) { + head_need = head_max = local->tx_headroom; + } else { + head_max += hdr_extra; + head_max += max_t(int, local->tx_headroom, + local->hw.extra_tx_headroom); + head_need += local->hw.extra_tx_headroom; + + tail_max = IEEE80211_ENCRYPT_TAILROOM; + if (enc_tailroom) + tail_need = tail_max; } if (skb_cloned(skb) && (!ieee80211_hw_check(&local->hw, SUPPORTS_CLONED_SKBS) || !skb_clone_writable(skb, ETH_HLEN) || enc_tailroom)) I802_DEBUG_INC(local->tx_expand_skb_head_cloned); - else if (head_need || tail_need) + else if (head_need > skb_headroom(skb) || + tail_need > skb_tailroom(skb)) I802_DEBUG_INC(local->tx_expand_skb_head); else return 0; - if (pskb_expand_head(skb, head_need, tail_need, GFP_ATOMIC)) { + head_max = max_t(int, 0, head_max - skb_headroom(skb)); + tail_max = max_t(int, 0, tail_max - skb_tailroom(skb)); + + if (pskb_expand_head(skb, head_max, tail_max, GFP_ATOMIC)) { wiphy_debug(local->hw.wiphy, "failed to reallocate TX buffer\n"); return -ENOMEM; @@ -1982,18 +1998,8 @@ void ieee80211_xmit(struct ieee80211_sub struct ieee80211_local *local = sdata->local; struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb); struct ieee80211_hdr *hdr; - int headroom; - bool may_encrypt; - - may_encrypt = !(info->flags & IEEE80211_TX_INTFL_DONT_ENCRYPT); - headroom = local->tx_headroom; - if (may_encrypt) - headroom += sdata->encrypt_headroom; - headroom -= skb_headroom(skb); - headroom = max_t(int, 0, headroom); - - if (ieee80211_skb_resize(sdata, skb, headroom, may_encrypt)) { + if (ieee80211_skb_resize(local, sdata, skb, 0, 0)) { ieee80211_free_txskb(&local->hw, skb); return; } @@ -2774,29 +2780,13 @@ static struct sk_buff *ieee80211_build_h } skb_pull(skb, skip_header_bytes); - head_need = hdrlen + encaps_len + meshhdrlen - skb_headroom(skb); + head_need = hdrlen + encaps_len + meshhdrlen; - /* - * So we need to modify the skb header and hence need a copy of - * that. The head_need variable above doesn't, so far, include - * the needed header space that we don't need right away. If we - * can, then we don't reallocate right now but only after the - * frame arrives at the master device (if it does...) - * - * If we cannot, however, then we will reallocate to include all - * the ever needed space. Also, if we need to reallocate it anyway, - * make it big enough for everything we may ever need. - */ - - if (head_need > 0 || skb_cloned(skb)) { - head_need += sdata->encrypt_headroom; - head_need += local->tx_headroom; - head_need = max_t(int, 0, head_need); - if (ieee80211_skb_resize(sdata, skb, head_need, true)) { - ieee80211_free_txskb(&local->hw, skb); - skb = NULL; - return ERR_PTR(-ENOMEM); - } + if (ieee80211_skb_resize(local, sdata, skb, head_need, + sdata->encrypt_headroom)) { + ieee80211_free_txskb(&local->hw, skb); + skb = NULL; + return ERR_PTR(-ENOMEM); } if (encaps_data) @@ -3411,7 +3401,6 @@ static bool ieee80211_xmit_fast(struct i struct ieee80211_local *local = sdata->local; u16 ethertype = (skb->data[12] << 8) | skb->data[13]; int extra_head = fast_tx->hdr_len - (ETH_HLEN - 2); - int hw_headroom = sdata->local->hw.extra_tx_headroom; struct ethhdr eth; struct ieee80211_tx_info *info; struct ieee80211_hdr *hdr = (void *)fast_tx->hdr; @@ -3463,10 +3452,7 @@ static bool ieee80211_xmit_fast(struct i * as the may-encrypt argument for the resize to not account for * more room than we already have in 'extra_head' */ - if (unlikely(ieee80211_skb_resize(sdata, skb, - max_t(int, extra_head + hw_headroom - - skb_headroom(skb), 0), - false))) { + if (unlikely(ieee80211_skb_resize(local, sdata, skb, extra_head, 0))) { kfree_skb(skb); return true; } id='n135' href='#n135'>135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362