1 /* 2 * IPV4 GSO/GRO offload support 3 * Linux INET implementation 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License 7 * as published by the Free Software Foundation; either version 8 * 2 of the License, or (at your option) any later version. 9 * 10 * TCPv4 GSO/GRO support 11 */ 12 13 #include <linux/skbuff.h> 14 #include <net/tcp.h> 15 #include <net/protocol.h> 16 17 static void tcp_gso_tstamp(struct sk_buff *skb, unsigned int ts_seq, 18 unsigned int seq, unsigned int mss) 19 { 20 while (skb) { 21 if (before(ts_seq, seq + mss)) { 22 skb_shinfo(skb)->tx_flags |= SKBTX_SW_TSTAMP; 23 skb_shinfo(skb)->tskey = ts_seq; 24 return; 25 } 26 27 skb = skb->next; 28 seq += mss; 29 } 30 } 31 32 struct sk_buff *tcp_gso_segment(struct sk_buff *skb, 33 netdev_features_t features) 34 { 35 struct sk_buff *segs = ERR_PTR(-EINVAL); 36 unsigned int sum_truesize = 0; 37 struct tcphdr *th; 38 unsigned int thlen; 39 unsigned int seq; 40 __be32 delta; 41 unsigned int oldlen; 42 unsigned int mss; 43 struct sk_buff *gso_skb = skb; 44 __sum16 newcheck; 45 bool ooo_okay, copy_destructor; 46 47 if (!pskb_may_pull(skb, sizeof(*th))) 48 goto out; 49 50 th = tcp_hdr(skb); 51 thlen = th->doff * 4; 52 if (thlen < sizeof(*th)) 53 goto out; 54 55 if (!pskb_may_pull(skb, thlen)) 56 goto out; 57 58 oldlen = (u16)~skb->len; 59 __skb_pull(skb, thlen); 60 61 mss = tcp_skb_mss(skb); 62 if (unlikely(skb->len <= mss)) 63 goto out; 64 65 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { 66 /* Packet is from an untrusted source, reset gso_segs. */ 67 int type = skb_shinfo(skb)->gso_type; 68 69 if (unlikely(type & 70 ~(SKB_GSO_TCPV4 | 71 SKB_GSO_DODGY | 72 SKB_GSO_TCP_ECN | 73 SKB_GSO_TCPV6 | 74 SKB_GSO_GRE | 75 SKB_GSO_GRE_CSUM | 76 SKB_GSO_IPIP | 77 SKB_GSO_SIT | 78 SKB_GSO_MPLS | 79 SKB_GSO_UDP_TUNNEL | 80 SKB_GSO_UDP_TUNNEL_CSUM | 81 0) || 82 !(type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))) 83 goto out; 84 85 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); 86 87 segs = NULL; 88 goto out; 89 } 90 91 copy_destructor = gso_skb->destructor == tcp_wfree; 92 ooo_okay = gso_skb->ooo_okay; 93 /* All segments but the first should have ooo_okay cleared */ 94 skb->ooo_okay = 0; 95 96 segs = skb_segment(skb, features); 97 if (IS_ERR(segs)) 98 goto out; 99 100 /* Only first segment might have ooo_okay set */ 101 segs->ooo_okay = ooo_okay; 102 103 delta = htonl(oldlen + (thlen + mss)); 104 105 skb = segs; 106 th = tcp_hdr(skb); 107 seq = ntohl(th->seq); 108 109 if (unlikely(skb_shinfo(gso_skb)->tx_flags & SKBTX_SW_TSTAMP)) 110 tcp_gso_tstamp(segs, skb_shinfo(gso_skb)->tskey, seq, mss); 111 112 newcheck = ~csum_fold((__force __wsum)((__force u32)th->check + 113 (__force u32)delta)); 114 115 do { 116 th->fin = th->psh = 0; 117 th->check = newcheck; 118 119 if (skb->ip_summed != CHECKSUM_PARTIAL) 120 th->check = gso_make_checksum(skb, ~th->check); 121 122 seq += mss; 123 if (copy_destructor) { 124 skb->destructor = gso_skb->destructor; 125 skb->sk = gso_skb->sk; 126 sum_truesize += skb->truesize; 127 } 128 skb = skb->next; 129 th = tcp_hdr(skb); 130 131 th->seq = htonl(seq); 132 th->cwr = 0; 133 } while (skb->next); 134 135 /* Following permits TCP Small Queues to work well with GSO : 136 * The callback to TCP stack will be called at the time last frag 137 * is freed at TX completion, and not right now when gso_skb 138 * is freed by GSO engine 139 */ 140 if (copy_destructor) { 141 swap(gso_skb->sk, skb->sk); 142 swap(gso_skb->destructor, skb->destructor); 143 sum_truesize += skb->truesize; 144 atomic_add(sum_truesize - gso_skb->truesize, 145 &skb->sk->sk_wmem_alloc); 146 } 147 148 delta = htonl(oldlen + (skb_tail_pointer(skb) - 149 skb_transport_header(skb)) + 150 skb->data_len); 151 th->check = ~csum_fold((__force __wsum)((__force u32)th->check + 152 (__force u32)delta)); 153 if (skb->ip_summed != CHECKSUM_PARTIAL) 154 th->check = gso_make_checksum(skb, ~th->check); 155 out: 156 return segs; 157 } 158 159 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb) 160 { 161 struct sk_buff **pp = NULL; 162 struct sk_buff *p; 163 struct tcphdr *th; 164 struct tcphdr *th2; 165 unsigned int len; 166 unsigned int thlen; 167 __be32 flags; 168 unsigned int mss = 1; 169 unsigned int hlen; 170 unsigned int off; 171 int flush = 1; 172 int i; 173 174 off = skb_gro_offset(skb); 175 hlen = off + sizeof(*th); 176 th = skb_gro_header_fast(skb, off); 177 if (skb_gro_header_hard(skb, hlen)) { 178 th = skb_gro_header_slow(skb, hlen, off); 179 if (unlikely(!th)) 180 goto out; 181 } 182 183 thlen = th->doff * 4; 184 if (thlen < sizeof(*th)) 185 goto out; 186 187 hlen = off + thlen; 188 if (skb_gro_header_hard(skb, hlen)) { 189 th = skb_gro_header_slow(skb, hlen, off); 190 if (unlikely(!th)) 191 goto out; 192 } 193 194 skb_gro_pull(skb, thlen); 195 196 len = skb_gro_len(skb); 197 flags = tcp_flag_word(th); 198 199 for (; (p = *head); head = &p->next) { 200 if (!NAPI_GRO_CB(p)->same_flow) 201 continue; 202 203 th2 = tcp_hdr(p); 204 205 if (*(u32 *)&th->source ^ *(u32 *)&th2->source) { 206 NAPI_GRO_CB(p)->same_flow = 0; 207 continue; 208 } 209 210 goto found; 211 } 212 213 goto out_check_final; 214 215 found: 216 /* Include the IP ID check below from the inner most IP hdr */ 217 flush = NAPI_GRO_CB(p)->flush | NAPI_GRO_CB(p)->flush_id; 218 flush |= (__force int)(flags & TCP_FLAG_CWR); 219 flush |= (__force int)((flags ^ tcp_flag_word(th2)) & 220 ~(TCP_FLAG_CWR | TCP_FLAG_FIN | TCP_FLAG_PSH)); 221 flush |= (__force int)(th->ack_seq ^ th2->ack_seq); 222 for (i = sizeof(*th); i < thlen; i += 4) 223 flush |= *(u32 *)((u8 *)th + i) ^ 224 *(u32 *)((u8 *)th2 + i); 225 226 mss = tcp_skb_mss(p); 227 228 flush |= (len - 1) >= mss; 229 flush |= (ntohl(th2->seq) + skb_gro_len(p)) ^ ntohl(th->seq); 230 231 if (flush || skb_gro_receive(head, skb)) { 232 mss = 1; 233 goto out_check_final; 234 } 235 236 p = *head; 237 th2 = tcp_hdr(p); 238 tcp_flag_word(th2) |= flags & (TCP_FLAG_FIN | TCP_FLAG_PSH); 239 240 out_check_final: 241 flush = len < mss; 242 flush |= (__force int)(flags & (TCP_FLAG_URG | TCP_FLAG_PSH | 243 TCP_FLAG_RST | TCP_FLAG_SYN | 244 TCP_FLAG_FIN)); 245 246 if (p && (!NAPI_GRO_CB(skb)->same_flow || flush)) 247 pp = head; 248 249 out: 250 NAPI_GRO_CB(skb)->flush |= (flush != 0); 251 252 return pp; 253 } 254 255 int tcp_gro_complete(struct sk_buff *skb) 256 { 257 struct tcphdr *th = tcp_hdr(skb); 258 259 skb->csum_start = (unsigned char *)th - skb->head; 260 skb->csum_offset = offsetof(struct tcphdr, check); 261 skb->ip_summed = CHECKSUM_PARTIAL; 262 263 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count; 264 265 if (th->cwr) 266 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; 267 268 return 0; 269 } 270 EXPORT_SYMBOL(tcp_gro_complete); 271 272 static int tcp_v4_gso_send_check(struct sk_buff *skb) 273 { 274 const struct iphdr *iph; 275 struct tcphdr *th; 276 277 if (!pskb_may_pull(skb, sizeof(*th))) 278 return -EINVAL; 279 280 iph = ip_hdr(skb); 281 th = tcp_hdr(skb); 282 283 th->check = 0; 284 skb->ip_summed = CHECKSUM_PARTIAL; 285 __tcp_v4_send_check(skb, iph->saddr, iph->daddr); 286 return 0; 287 } 288 289 static struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb) 290 { 291 /* Use the IP hdr immediately proceeding for this transport */ 292 const struct iphdr *iph = skb_gro_network_header(skb); 293 __wsum wsum; 294 295 /* Don't bother verifying checksum if we're going to flush anyway. */ 296 if (NAPI_GRO_CB(skb)->flush) 297 goto skip_csum; 298 299 wsum = NAPI_GRO_CB(skb)->csum; 300 301 switch (skb->ip_summed) { 302 case CHECKSUM_NONE: 303 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 304 0); 305 306 /* fall through */ 307 308 case CHECKSUM_COMPLETE: 309 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr, 310 wsum)) { 311 skb->ip_summed = CHECKSUM_UNNECESSARY; 312 break; 313 } 314 315 NAPI_GRO_CB(skb)->flush = 1; 316 return NULL; 317 } 318 319 skip_csum: 320 return tcp_gro_receive(head, skb); 321 } 322 323 static int tcp4_gro_complete(struct sk_buff *skb, int thoff) 324 { 325 const struct iphdr *iph = ip_hdr(skb); 326 struct tcphdr *th = tcp_hdr(skb); 327 328 th->check = ~tcp_v4_check(skb->len - thoff, iph->saddr, 329 iph->daddr, 0); 330 skb_shinfo(skb)->gso_type |= SKB_GSO_TCPV4; 331 332 return tcp_gro_complete(skb); 333 } 334 335 static const struct net_offload tcpv4_offload = { 336 .callbacks = { 337 .gso_send_check = tcp_v4_gso_send_check, 338 .gso_segment = tcp_gso_segment, 339 .gro_receive = tcp4_gro_receive, 340 .gro_complete = tcp4_gro_complete, 341 }, 342 }; 343 344 int __init tcpv4_offload_init(void) 345 { 346 return inet_add_offload(&tcpv4_offload, IPPROTO_TCP); 347 } 348