datapath/linux/compat/stt.c

   1 /*
   2  * Stateless TCP Tunnel (STT) vport.
   3  *
   4  * Copyright (c) 2015 Nicira, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the License, or (at your option) any later version.
  10  */
  11
  12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  13 #include <asm/unaligned.h>
  14
  15 #include <linux/delay.h>
  16 #include <linux/flex_array.h>
  17 #include <linux/if.h>
  18 #include <linux/if_vlan.h>
  19 #include <linux/ip.h>
  20 #include <linux/ipv6.h>
  21 #include <linux/jhash.h>
  22 #include <linux/list.h>
  23 #include <linux/log2.h>
  24 #include <linux/module.h>
  25 #include <linux/net.h>
  26 #include <linux/netfilter.h>
  27 #include <linux/percpu.h>
  28 #include <linux/skbuff.h>
  29 #include <linux/tcp.h>
  30 #include <linux/workqueue.h>
  31
  32 #include <net/dst_metadata.h>
  33 #include <net/icmp.h>
  34 #include <net/inet_ecn.h>
  35 #include <net/ip.h>
  36 #include <net/ip_tunnels.h>
  37 #include <net/ip6_checksum.h>
  38 #include <net/net_namespace.h>
  39 #include <net/netns/generic.h>
  40 #include <net/sock.h>
  41 #include <net/stt.h>
  42 #include <net/tcp.h>
  43 #include <net/udp.h>
  44
  45 #include "gso.h"
  46 #include "compat.h"
  47
  48 #define STT_NETDEV_VER  "0.1"
  49 #define STT_DST_PORT 7471
  50
  51 #ifdef OVS_STT
  52 #define STT_VER 0
  53
  54 /* @list: Per-net list of STT ports.
  55  * @rcv: The callback is called on STT packet recv, STT reassembly can generate
  56  * multiple packets, in this case first packet has tunnel outer header, rest
  57  * of the packets are inner packet segments with no stt header.
  58  * @rcv_data: user data.
  59  * @sock: Fake TCP socket for the STT port.
  60  */
  61 struct stt_dev {
  62         struct net_device       *dev;
  63         struct net              *net;
  64         struct list_head        next;
  65         struct socket           *sock;
  66         __be16                  dst_port;
  67 };
  68
  69 #define STT_CSUM_VERIFIED       BIT(0)
  70 #define STT_CSUM_PARTIAL        BIT(1)
  71 #define STT_PROTO_IPV4          BIT(2)
  72 #define STT_PROTO_TCP           BIT(3)
  73 #define STT_PROTO_TYPES         (STT_PROTO_IPV4 | STT_PROTO_TCP)
  74
  75 #define SUPPORTED_GSO_TYPES (SKB_GSO_TCPV4 | SKB_GSO_UDP | SKB_GSO_DODGY | \
  76                              SKB_GSO_TCPV6)
  77
  78 /* The length and offset of a fragment are encoded in the sequence number.
  79  * STT_SEQ_LEN_SHIFT is the left shift needed to store the length.
  80  * STT_SEQ_OFFSET_MASK is the mask to extract the offset.
  81  */
  82 #define STT_SEQ_LEN_SHIFT 16
  83 #define STT_SEQ_OFFSET_MASK (BIT(STT_SEQ_LEN_SHIFT) - 1)
  84
  85 /* The maximum amount of memory used to store packets waiting to be reassembled
  86  * on a given CPU.  Once this threshold is exceeded we will begin freeing the
  87  * least recently used fragments.
  88  */
  89 #define REASM_HI_THRESH (4 * 1024 * 1024)
  90 /* The target for the high memory evictor.  Once we have exceeded
  91  * REASM_HI_THRESH, we will continue freeing fragments until we hit
  92  * this limit.
  93  */
  94 #define REASM_LO_THRESH (3 * 1024 * 1024)
  95 /* The length of time a given packet has to be reassembled from the time the
  96  * first fragment arrives.  Once this limit is exceeded it becomes available
  97  * for cleaning.
  98  */
  99 #define FRAG_EXP_TIME (30 * HZ)
 100 /* Number of hash entries.  Each entry has only a single slot to hold a packet
 101  * so if there are collisions, we will drop packets.  This is allocated
 102  * per-cpu and each entry consists of struct pkt_frag.
 103  */
 104 #define FRAG_HASH_SHIFT         8
 105 #define FRAG_HASH_ENTRIES       BIT(FRAG_HASH_SHIFT)
 106 #define FRAG_HASH_SEGS          ((sizeof(u32) * 8) / FRAG_HASH_SHIFT)
 107
 108 #define CLEAN_PERCPU_INTERVAL (30 * HZ)
 109
 110 struct pkt_key {
 111         __be32 saddr;
 112         __be32 daddr;
 113         __be32 pkt_seq;
 114         u32 mark;
 115 };
 116
 117 struct pkt_frag {
 118         struct sk_buff *skbs;
 119         unsigned long timestamp;
 120         struct list_head lru_node;
 121         struct pkt_key key;
 122 };
 123
 124 struct stt_percpu {
 125         struct flex_array *frag_hash;
 126         struct list_head frag_lru;
 127         unsigned int frag_mem_used;
 128
 129         /* Protect frags table. */
 130         spinlock_t lock;
 131 };
 132
 133 struct first_frag {
 134         struct sk_buff *last_skb;
 135         unsigned int mem_used;
 136         u16 tot_len;
 137         u16 rcvd_len;
 138         bool set_ecn_ce;
 139 };
 140
 141 struct frag_skb_cb {
 142         u16 offset;
 143
 144         /* Only valid for the first skb in the chain. */
 145         struct first_frag first;
 146 };
 147
 148 #define FRAG_CB(skb) ((struct frag_skb_cb *)(skb)->cb)
 149
 150 /* per-network namespace private data for this module */
 151 struct stt_net {
 152         struct list_head stt_list;
 153         int n_tunnels;
 154 };
 155
 156 static int stt_net_id;
 157
 158 static struct stt_percpu __percpu *stt_percpu_data __read_mostly;
 159 static u32 frag_hash_seed __read_mostly;
 160
 161 /* Protects sock-hash and refcounts. */
 162 static DEFINE_MUTEX(stt_mutex);
 163
 164 static int n_tunnels;
 165 static DEFINE_PER_CPU(u32, pkt_seq_counter);
 166
 167 static void clean_percpu(struct work_struct *work);
 168 static DECLARE_DELAYED_WORK(clean_percpu_wq, clean_percpu);
 169
 170 static struct stt_dev *stt_find_sock(struct net *net, __be16 port)
 171 {
 172         struct stt_net *sn = net_generic(net, stt_net_id);
 173         struct stt_dev *stt_dev;
 174
 175         list_for_each_entry_rcu(stt_dev, &sn->stt_list, next) {
 176                 if (inet_sk(stt_dev->sock->sk)->inet_sport == port)
 177                         return stt_dev;
 178         }
 179         return NULL;
 180 }
 181
 182 static __be32 ack_seq(void)
 183 {
 184 #if NR_CPUS <= 65536
 185         u32 pkt_seq, ack;
 186
 187         pkt_seq = this_cpu_read(pkt_seq_counter);
 188         ack = pkt_seq << ilog2(NR_CPUS) | smp_processor_id();
 189         this_cpu_inc(pkt_seq_counter);
 190
 191         return (__force __be32)ack;
 192 #else
 193 #error "Support for greater than 64k CPUs not implemented"
 194 #endif
 195 }
 196
 197 static int clear_gso(struct sk_buff *skb)
 198 {
 199         struct skb_shared_info *shinfo = skb_shinfo(skb);
 200         int err;
 201
 202         if (shinfo->gso_type == 0 && shinfo->gso_size == 0 &&
 203             shinfo->gso_segs == 0)
 204                 return 0;
 205
 206         err = skb_unclone(skb, GFP_ATOMIC);
 207         if (unlikely(err))
 208                 return err;
 209
 210         shinfo = skb_shinfo(skb);
 211         shinfo->gso_type = 0;
 212         shinfo->gso_size = 0;
 213         shinfo->gso_segs = 0;
 214         return 0;
 215 }
 216
 217 static struct sk_buff *normalize_frag_list(struct sk_buff *head,
 218                                            struct sk_buff **skbp)
 219 {
 220         struct sk_buff *skb = *skbp;
 221         struct sk_buff *last;
 222
 223         do {
 224                 struct sk_buff *frags;
 225
 226                 if (skb_shared(skb)) {
 227                         struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
 228
 229                         if (unlikely(!nskb))
 230                                 return ERR_PTR(-ENOMEM);
 231
 232                         nskb->next = skb->next;
 233                         consume_skb(skb);
 234                         skb = nskb;
 235                         *skbp = skb;
 236                 }
 237
 238                 if (head) {
 239                         head->len -= skb->len;
 240                         head->data_len -= skb->len;
 241                         head->truesize -= skb->truesize;
 242                 }
 243
 244                 frags = skb_shinfo(skb)->frag_list;
 245                 if (frags) {
 246                         int err;
 247
 248                         err = skb_unclone(skb, GFP_ATOMIC);
 249                         if (unlikely(err))
 250                                 return ERR_PTR(err);
 251
 252                         last = normalize_frag_list(skb, &frags);
 253                         if (IS_ERR(last))
 254                                 return last;
 255
 256                         skb_shinfo(skb)->frag_list = NULL;
 257                         last->next = skb->next;
 258                         skb->next = frags;
 259                 } else {
 260                         last = skb;
 261                 }
 262
 263                 skbp = &skb->next;
 264         } while ((skb = skb->next));
 265
 266         return last;
 267 }
 268
 269 /* Takes a linked list of skbs, which potentially contain frag_list
 270  * (whose members in turn potentially contain frag_lists, etc.) and
 271  * converts them into a single linear linked list.
 272  */
 273 static int straighten_frag_list(struct sk_buff **skbp)
 274 {
 275         struct sk_buff *err_skb;
 276
 277         err_skb = normalize_frag_list(NULL, skbp);
 278         if (IS_ERR(err_skb))
 279                 return PTR_ERR(err_skb);
 280
 281         return 0;
 282 }
 283
 284 static void copy_skb_metadata(struct sk_buff *to, struct sk_buff *from)
 285 {
 286         to->protocol = from->protocol;
 287         to->tstamp = from->tstamp;
 288         to->priority = from->priority;
 289         to->mark = from->mark;
 290         to->vlan_tci = from->vlan_tci;
 291 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,10,0)
 292         to->vlan_proto = from->vlan_proto;
 293 #endif
 294         skb_copy_secmark(to, from);
 295 }
 296
 297 static void update_headers(struct sk_buff *skb, bool head,
 298                                unsigned int l4_offset, unsigned int hdr_len,
 299                                bool ipv4, u32 tcp_seq)
 300 {
 301         u16 old_len, new_len;
 302         __be32 delta;
 303         struct tcphdr *tcph;
 304         int gso_size;
 305
 306         if (ipv4) {
 307                 struct iphdr *iph = (struct iphdr *)(skb->data + ETH_HLEN);
 308
 309                 old_len = ntohs(iph->tot_len);
 310                 new_len = skb->len - ETH_HLEN;
 311                 iph->tot_len = htons(new_len);
 312
 313                 ip_send_check(iph);
 314         } else {
 315                 struct ipv6hdr *ip6h = (struct ipv6hdr *)(skb->data + ETH_HLEN);
 316
 317                 old_len = ntohs(ip6h->payload_len);
 318                 new_len = skb->len - ETH_HLEN - sizeof(struct ipv6hdr);
 319                 ip6h->payload_len = htons(new_len);
 320         }
 321
 322         tcph = (struct tcphdr *)(skb->data + l4_offset);
 323         if (!head) {
 324                 tcph->seq = htonl(tcp_seq);
 325                 tcph->cwr = 0;
 326         }
 327
 328         if (skb->next) {
 329                 tcph->fin = 0;
 330                 tcph->psh = 0;
 331         }
 332
 333         delta = htonl(~old_len + new_len);
 334         tcph->check = ~csum_fold((__force __wsum)((__force u32)tcph->check +
 335                                  (__force u32)delta));
 336
 337         gso_size = skb_shinfo(skb)->gso_size;
 338         if (gso_size && skb->len - hdr_len <= gso_size)
 339                 BUG_ON(clear_gso(skb));
 340 }
 341
 342 static bool can_segment(struct sk_buff *head, bool ipv4, bool tcp, bool csum_partial)
 343 {
 344         /* If no offloading is in use then we don't have enough information
 345          * to process the headers.
 346          */
 347         if (!csum_partial)
 348                 goto linearize;
 349
 350         /* Handling UDP packets requires IP fragmentation, which means that
 351          * the L4 checksum can no longer be calculated by hardware (since the
 352          * fragments are in different packets.  If we have to compute the
 353          * checksum it's faster just to linearize and large UDP packets are
 354          * pretty uncommon anyways, so it's not worth dealing with for now.
 355          */
 356         if (!tcp)
 357                 goto linearize;
 358
 359         if (ipv4) {
 360                 struct iphdr *iph = (struct iphdr *)(head->data + ETH_HLEN);
 361
 362                 /* It's difficult to get the IP IDs exactly right here due to
 363                  * varying segment sizes and potentially multiple layers of
 364                  * segmentation.  IP ID isn't important when DF is set and DF
 365                  * is generally set for TCP packets, so just linearize if it's
 366                  * not.
 367                  */
 368                 if (!(iph->frag_off & htons(IP_DF)))
 369                         goto linearize;
 370         } else {
 371                 struct ipv6hdr *ip6h = (struct ipv6hdr *)(head->data + ETH_HLEN);
 372
 373                 /* Jumbograms require more processing to update and we'll
 374                  * probably never see them, so just linearize.
 375                  */
 376                 if (ip6h->payload_len == 0)
 377                         goto linearize;
 378         }
 379         return true;
 380
 381 linearize:
 382         return false;
 383 }
 384
 385 static int copy_headers(struct sk_buff *head, struct sk_buff *frag,
 386                             int hdr_len)
 387 {
 388         u16 csum_start;
 389
 390         if (skb_cloned(frag) || skb_headroom(frag) < hdr_len) {
 391                 int extra_head = hdr_len - skb_headroom(frag);
 392
 393                 extra_head = extra_head > 0 ? extra_head : 0;
 394                 if (unlikely(pskb_expand_head(frag, extra_head, 0,
 395                                               GFP_ATOMIC)))
 396                         return -ENOMEM;
 397         }
 398
 399         memcpy(__skb_push(frag, hdr_len), head->data, hdr_len);
 400
 401         csum_start = head->csum_start - skb_headroom(head);
 402         frag->csum_start = skb_headroom(frag) + csum_start;
 403         frag->csum_offset = head->csum_offset;
 404         frag->ip_summed = head->ip_summed;
 405
 406         skb_shinfo(frag)->gso_size = skb_shinfo(head)->gso_size;
 407         skb_shinfo(frag)->gso_type = skb_shinfo(head)->gso_type;
 408         skb_shinfo(frag)->gso_segs = 0;
 409
 410         copy_skb_metadata(frag, head);
 411         return 0;
 412 }
 413
 414 static int skb_list_segment(struct sk_buff *head, bool ipv4, int l4_offset)
 415 {
 416         struct sk_buff *skb;
 417         struct tcphdr *tcph;
 418         int seg_len;
 419         int hdr_len;
 420         int tcp_len;
 421         u32 seq;
 422
 423         if (unlikely(!pskb_may_pull(head, l4_offset + sizeof(*tcph))))
 424                 return -ENOMEM;
 425
 426         tcph = (struct tcphdr *)(head->data + l4_offset);
 427         tcp_len = tcph->doff * 4;
 428         hdr_len = l4_offset + tcp_len;
 429
 430         if (unlikely((tcp_len < sizeof(struct tcphdr)) ||
 431                      (head->len < hdr_len)))
 432                 return -EINVAL;
 433
 434         if (unlikely(!pskb_may_pull(head, hdr_len)))
 435                 return -ENOMEM;
 436
 437         tcph = (struct tcphdr *)(head->data + l4_offset);
 438         /* Update header of each segment. */
 439         seq = ntohl(tcph->seq);
 440         seg_len = skb_pagelen(head) - hdr_len;
 441
 442         skb = skb_shinfo(head)->frag_list;
 443         skb_shinfo(head)->frag_list = NULL;
 444         head->next = skb;
 445         for (; skb; skb = skb->next) {
 446                 int err;
 447
 448                 head->len -= skb->len;
 449                 head->data_len -= skb->len;
 450                 head->truesize -= skb->truesize;
 451
 452                 seq += seg_len;
 453                 seg_len = skb->len;
 454                 err = copy_headers(head, skb, hdr_len);
 455                 if (err)
 456                         return err;
 457                 update_headers(skb, false, l4_offset, hdr_len, ipv4, seq);
 458         }
 459         update_headers(head, true, l4_offset, hdr_len, ipv4, 0);
 460         return 0;
 461 }
 462
 463 static int coalesce_skb(struct sk_buff **headp)
 464 {
 465         struct sk_buff *frag, *head, *prev;
 466         int err;
 467
 468         err = straighten_frag_list(headp);
 469         if (unlikely(err))
 470                 return err;
 471         head = *headp;
 472
 473         /* Coalesce frag list. */
 474         prev = head;
 475         for (frag = head->next; frag; frag = frag->next) {
 476                 bool headstolen;
 477                 int delta;
 478
 479                 if (unlikely(skb_unclone(prev, GFP_ATOMIC)))
 480                         return -ENOMEM;
 481
 482                 if (!skb_try_coalesce(prev, frag, &headstolen, &delta)) {
 483                         prev = frag;
 484                         continue;
 485                 }
 486
 487                 prev->next = frag->next;
 488                 frag->len = 0;
 489                 frag->data_len = 0;
 490                 frag->truesize -= delta;
 491                 kfree_skb_partial(frag, headstolen);
 492                 frag = prev;
 493         }
 494
 495         if (!head->next)
 496                 return 0;
 497
 498         for (frag = head->next; frag; frag = frag->next) {
 499                 head->len += frag->len;
 500                 head->data_len += frag->len;
 501                 head->truesize += frag->truesize;
 502         }
 503
 504         skb_shinfo(head)->frag_list = head->next;
 505         head->next = NULL;
 506         return 0;
 507 }
 508
 509 static int __try_to_segment(struct sk_buff *skb, bool csum_partial,
 510                             bool ipv4, bool tcp, int l4_offset)
 511 {
 512         if (can_segment(skb, ipv4, tcp, csum_partial))
 513                 return skb_list_segment(skb, ipv4, l4_offset);
 514         else
 515                 return skb_linearize(skb);
 516 }
 517
 518 static int try_to_segment(struct sk_buff *skb)
 519 {
 520         struct stthdr *stth = stt_hdr(skb);
 521         bool csum_partial = !!(stth->flags & STT_CSUM_PARTIAL);
 522         bool ipv4 = !!(stth->flags & STT_PROTO_IPV4);
 523         bool tcp = !!(stth->flags & STT_PROTO_TCP);
 524         int l4_offset = stth->l4_offset;
 525
 526         return __try_to_segment(skb, csum_partial, ipv4, tcp, l4_offset);
 527 }
 528
 529 static int segment_skb(struct sk_buff **headp, bool csum_partial,
 530                        bool ipv4, bool tcp, int l4_offset)
 531 {
 532         int err;
 533
 534         err = coalesce_skb(headp);
 535         if (err)
 536                 return err;
 537
 538         if (skb_shinfo(*headp)->frag_list)
 539                 return __try_to_segment(*headp, csum_partial,
 540                                         ipv4, tcp, l4_offset);
 541         return 0;
 542 }
 543
 544 static int __push_stt_header(struct sk_buff *skb, __be64 tun_id,
 545                              __be16 s_port, __be16 d_port,
 546                              __be32 saddr, __be32 dst,
 547                              __be16 l3_proto, u8 l4_proto,
 548                              int dst_mtu)
 549 {
 550         int data_len = skb->len + sizeof(struct stthdr) + STT_ETH_PAD;
 551         unsigned short encap_mss;
 552         struct tcphdr *tcph;
 553         struct stthdr *stth;
 554
 555         skb_push(skb, STT_HEADER_LEN);
 556         skb_reset_transport_header(skb);
 557         tcph = tcp_hdr(skb);
 558         memset(tcph, 0, STT_HEADER_LEN);
 559         stth = stt_hdr(skb);
 560
 561         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 562                 stth->flags |= STT_CSUM_PARTIAL;
 563
 564                 stth->l4_offset = skb->csum_start -
 565                                         (skb_headroom(skb) +
 566                                         STT_HEADER_LEN);
 567
 568                 if (l3_proto == htons(ETH_P_IP))
 569                         stth->flags |= STT_PROTO_IPV4;
 570
 571                 if (l4_proto == IPPROTO_TCP)
 572                         stth->flags |= STT_PROTO_TCP;
 573
 574                 stth->mss = htons(skb_shinfo(skb)->gso_size);
 575         } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
 576                 stth->flags |= STT_CSUM_VERIFIED;
 577         }
 578
 579         stth->vlan_tci = htons(skb->vlan_tci);
 580         skb->vlan_tci = 0;
 581         put_unaligned(tun_id, &stth->key);
 582
 583         tcph->source    = s_port;
 584         tcph->dest      = d_port;
 585         tcph->doff      = sizeof(struct tcphdr) / 4;
 586         tcph->ack       = 1;
 587         tcph->psh       = 1;
 588         tcph->window    = htons(USHRT_MAX);
 589         tcph->seq       = htonl(data_len << STT_SEQ_LEN_SHIFT);
 590         tcph->ack_seq   = ack_seq();
 591         tcph->check     = ~tcp_v4_check(skb->len, saddr, dst, 0);
 592
 593         skb->csum_start = skb_transport_header(skb) - skb->head;
 594         skb->csum_offset = offsetof(struct tcphdr, check);
 595         skb->ip_summed = CHECKSUM_PARTIAL;
 596
 597         encap_mss = dst_mtu - sizeof(struct iphdr) - sizeof(struct tcphdr);
 598         if (data_len > encap_mss) {
 599                 if (unlikely(skb_unclone(skb, GFP_ATOMIC)))
 600                         return -EINVAL;
 601
 602                 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 603                 skb_shinfo(skb)->gso_size = encap_mss;
 604                 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(data_len, encap_mss);
 605         } else {
 606                 if (unlikely(clear_gso(skb)))
 607                         return -EINVAL;
 608         }
 609         return 0;
 610 }
 611
 612 static struct sk_buff *push_stt_header(struct sk_buff *head, __be64 tun_id,
 613                                        __be16 s_port, __be16 d_port,
 614                                        __be32 saddr, __be32 dst,
 615                                        __be16 l3_proto, u8 l4_proto,
 616                                        int dst_mtu)
 617 {
 618         struct sk_buff *skb;
 619
 620         if (skb_shinfo(head)->frag_list) {
 621                 bool ipv4 = (l3_proto == htons(ETH_P_IP));
 622                 bool tcp = (l4_proto == IPPROTO_TCP);
 623                 bool csum_partial = (head->ip_summed == CHECKSUM_PARTIAL);
 624                 int l4_offset = skb_transport_offset(head);
 625
 626                 /* Need to call skb_orphan() to report currect true-size.
 627                  * calling skb_orphan() in this layer is odd but SKB with
 628                  * frag-list should not be associated with any socket, so
 629                  * skb-orphan should be no-op. */
 630                 skb_orphan(head);
 631                 if (unlikely(segment_skb(&head, csum_partial,
 632                                          ipv4, tcp, l4_offset)))
 633                         goto error;
 634         }
 635
 636         for (skb = head; skb; skb = skb->next) {
 637                 if (__push_stt_header(skb, tun_id, s_port, d_port, saddr, dst,
 638                                       l3_proto, l4_proto, dst_mtu))
 639                         goto error;
 640         }
 641
 642         return head;
 643 error:
 644         kfree_skb_list(head);
 645         return NULL;
 646 }
 647
 648 static int stt_can_offload(struct sk_buff *skb, __be16 l3_proto, u8 l4_proto)
 649 {
 650         if (skb_is_gso(skb) && skb->ip_summed != CHECKSUM_PARTIAL) {
 651                 int csum_offset;
 652                 __sum16 *csum;
 653                 int len;
 654
 655                 if (l4_proto == IPPROTO_TCP)
 656                         csum_offset = offsetof(struct tcphdr, check);
 657                 else if (l4_proto == IPPROTO_UDP)
 658                         csum_offset = offsetof(struct udphdr, check);
 659                 else
 660                         return 0;
 661
 662                 len = skb->len - skb_transport_offset(skb);
 663                 csum = (__sum16 *)(skb_transport_header(skb) + csum_offset);
 664
 665                 if (unlikely(!pskb_may_pull(skb, skb_transport_offset(skb) +
 666                                                  csum_offset + sizeof(*csum))))
 667                         return -EINVAL;
 668
 669                 if (l3_proto == htons(ETH_P_IP)) {
 670                         struct iphdr *iph = ip_hdr(skb);
 671
 672                         *csum = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 673                                                    len, l4_proto, 0);
 674                 } else if (l3_proto == htons(ETH_P_IPV6)) {
 675                         struct ipv6hdr *ip6h = ipv6_hdr(skb);
 676
 677                         *csum = ~csum_ipv6_magic(&ip6h->saddr, &ip6h->daddr,
 678                                                  len, l4_proto, 0);
 679                 } else {
 680                         return 0;
 681                 }
 682                 skb->csum_start = skb_transport_header(skb) - skb->head;
 683                 skb->csum_offset = csum_offset;
 684                 skb->ip_summed = CHECKSUM_PARTIAL;
 685         }
 686
 687         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 688                 /* Assume receiver can only offload TCP/UDP over IPv4/6,
 689                  * and require 802.1Q VLANs to be accelerated.
 690                  */
 691                 if (l3_proto != htons(ETH_P_IP) &&
 692                     l3_proto != htons(ETH_P_IPV6))
 693                         return 0;
 694
 695                 if (l4_proto != IPPROTO_TCP && l4_proto != IPPROTO_UDP)
 696                         return 0;
 697
 698                 /* L4 offset must fit in a 1-byte field. */
 699                 if (skb->csum_start - skb_headroom(skb) > 255)
 700                         return 0;
 701
 702                 if (skb_shinfo(skb)->gso_type & ~SUPPORTED_GSO_TYPES)
 703                         return 0;
 704         }
 705         /* Total size of encapsulated packet must fit in 16 bits. */
 706         if (skb->len + STT_HEADER_LEN + sizeof(struct iphdr) > 65535)
 707                 return 0;
 708
 709 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,10,0)
 710         if (skb_vlan_tag_present(skb) && skb->vlan_proto != htons(ETH_P_8021Q))
 711                 return 0;
 712 #endif
 713         return 1;
 714 }
 715
 716 static bool need_linearize(const struct sk_buff *skb)
 717 {
 718         struct skb_shared_info *shinfo = skb_shinfo(skb);
 719         int i;
 720
 721         if (unlikely(shinfo->frag_list))
 722                 return true;
 723
 724         /* Generally speaking we should linearize if there are paged frags.
 725          * However, if all of the refcounts are 1 we know nobody else can
 726          * change them from underneath us and we can skip the linearization.
 727          */
 728         for (i = 0; i < shinfo->nr_frags; i++)
 729                 if (unlikely(page_count(skb_frag_page(&shinfo->frags[i])) > 1))
 730                         return true;
 731
 732         return false;
 733 }
 734
 735 static struct sk_buff *handle_offloads(struct sk_buff *skb, int min_headroom)
 736 {
 737         int err;
 738
 739 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,10,0)
 740         if (skb_vlan_tag_present(skb) && skb->vlan_proto != htons(ETH_P_8021Q)) {
 741
 742                 min_headroom += VLAN_HLEN;
 743                 if (skb_headroom(skb) < min_headroom) {
 744                         int head_delta = SKB_DATA_ALIGN(min_headroom -
 745                                                         skb_headroom(skb) + 16);
 746
 747                         err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 748                                                0, GFP_ATOMIC);
 749                         if (unlikely(err))
 750                                 goto error;
 751                 }
 752
 753                 skb = __vlan_hwaccel_push_inside(skb);
 754                 if (!skb) {
 755                         err = -ENOMEM;
 756                         goto error;
 757                 }
 758         }
 759 #endif
 760
 761         if (skb_is_gso(skb)) {
 762                 struct sk_buff *nskb;
 763                 char cb[sizeof(skb->cb)];
 764
 765                 memcpy(cb, skb->cb, sizeof(cb));
 766
 767                 nskb = __skb_gso_segment(skb, 0, false);
 768                 if (IS_ERR(nskb)) {
 769                         err = PTR_ERR(nskb);
 770                         goto error;
 771                 }
 772
 773                 consume_skb(skb);
 774                 skb = nskb;
 775                 while (nskb) {
 776                         memcpy(nskb->cb, cb, sizeof(cb));
 777                         nskb = nskb->next;
 778                 }
 779         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 780                 /* Pages aren't locked and could change at any time.
 781                  * If this happens after we compute the checksum, the
 782                  * checksum will be wrong.  We linearize now to avoid
 783                  * this problem.
 784                  */
 785                 if (unlikely(need_linearize(skb))) {
 786                         err = __skb_linearize(skb);
 787                         if (unlikely(err))
 788                                 goto error;
 789                 }
 790
 791                 err = skb_checksum_help(skb);
 792                 if (unlikely(err))
 793                         goto error;
 794         }
 795         skb->ip_summed = CHECKSUM_NONE;
 796
 797         return skb;
 798 error:
 799         kfree_skb(skb);
 800         return ERR_PTR(err);
 801 }
 802
 803 static int skb_list_xmit(struct rtable *rt, struct sk_buff *skb, __be32 src,
 804                          __be32 dst, __u8 tos, __u8 ttl, __be16 df)
 805 {
 806         int len = 0;
 807
 808         while (skb) {
 809                 struct sk_buff *next = skb->next;
 810
 811                 if (next)
 812                         dst_clone(&rt->dst);
 813
 814                 skb->next = NULL;
 815                 len += iptunnel_xmit(NULL, rt, skb, src, dst, IPPROTO_TCP,
 816                                      tos, ttl, df, false);
 817
 818                 skb = next;
 819         }
 820         return len;
 821 }
 822
 823 static u8 parse_ipv6_l4_proto(struct sk_buff *skb)
 824 {
 825         unsigned int nh_ofs = skb_network_offset(skb);
 826         int payload_ofs;
 827         struct ipv6hdr *nh;
 828         uint8_t nexthdr;
 829         __be16 frag_off;
 830
 831         if (unlikely(!pskb_may_pull(skb, nh_ofs + sizeof(struct ipv6hdr))))
 832                 return 0;
 833
 834         nh = ipv6_hdr(skb);
 835         nexthdr = nh->nexthdr;
 836         payload_ofs = (u8 *)(nh + 1) - skb->data;
 837
 838         payload_ofs = ipv6_skip_exthdr(skb, payload_ofs, &nexthdr, &frag_off);
 839         if (unlikely(payload_ofs < 0))
 840                 return 0;
 841
 842         return nexthdr;
 843 }
 844
 845 static u8 skb_get_l4_proto(struct sk_buff *skb, __be16 l3_proto)
 846 {
 847         if (l3_proto == htons(ETH_P_IP)) {
 848                 unsigned int nh_ofs = skb_network_offset(skb);
 849
 850                 if (unlikely(!pskb_may_pull(skb, nh_ofs + sizeof(struct iphdr))))
 851                         return 0;
 852
 853                 return ip_hdr(skb)->protocol;
 854         } else if (l3_proto == htons(ETH_P_IPV6)) {
 855                 return parse_ipv6_l4_proto(skb);
 856         }
 857         return 0;
 858 }
 859
 860 static int stt_xmit_skb(struct sk_buff *skb, struct rtable *rt,
 861                  __be32 src, __be32 dst, __u8 tos,
 862                  __u8 ttl, __be16 df, __be16 src_port, __be16 dst_port,
 863                  __be64 tun_id)
 864 {
 865         struct ethhdr *eh = eth_hdr(skb);
 866         int ret = 0, min_headroom;
 867         __be16 inner_l3_proto;
 868          u8 inner_l4_proto;
 869
 870         inner_l3_proto = eh->h_proto;
 871         inner_l4_proto = skb_get_l4_proto(skb, inner_l3_proto);
 872
 873         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 874                         + STT_HEADER_LEN + sizeof(struct iphdr);
 875
 876         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 877                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 878                                                 skb_headroom(skb) +
 879                                                 16);
 880
 881                 ret = pskb_expand_head(skb, max_t(int, head_delta, 0),
 882                                        0, GFP_ATOMIC);
 883                 if (unlikely(ret))
 884                         goto err_free_rt;
 885         }
 886
 887         ret = stt_can_offload(skb, inner_l3_proto, inner_l4_proto);
 888         if (ret < 0)
 889                 goto err_free_rt;
 890         if (!ret) {
 891                 skb = handle_offloads(skb, min_headroom);
 892                 if (IS_ERR(skb)) {
 893                         ret = PTR_ERR(skb);
 894                         skb = NULL;
 895                         goto err_free_rt;
 896                 }
 897         }
 898
 899         ret = 0;
 900         while (skb) {
 901                 struct sk_buff *next_skb = skb->next;
 902
 903                 skb->next = NULL;
 904
 905                 if (next_skb)
 906                         dst_clone(&rt->dst);
 907
 908                 /* Push STT and TCP header. */
 909                 skb = push_stt_header(skb, tun_id, src_port, dst_port, src,
 910                                       dst, inner_l3_proto, inner_l4_proto,
 911                                       dst_mtu(&rt->dst));
 912                 if (unlikely(!skb)) {
 913                         ip_rt_put(rt);
 914                         goto next;
 915                 }
 916
 917                 /* Push IP header. */
 918                 ret += skb_list_xmit(rt, skb, src, dst, tos, ttl, df);
 919
 920 next:
 921                 skb = next_skb;
 922         }
 923
 924         return ret;
 925
 926 err_free_rt:
 927         ip_rt_put(rt);
 928         kfree_skb(skb);
 929         return ret;
 930 }
 931
 932 netdev_tx_t ovs_stt_xmit(struct sk_buff *skb)
 933 {
 934         struct net_device *dev = skb->dev;
 935         struct stt_dev *stt_dev = netdev_priv(dev);
 936         struct net *net = stt_dev->net;
 937         __be16 dport = inet_sk(stt_dev->sock->sk)->inet_sport;
 938         struct ip_tunnel_key *tun_key;
 939         struct ip_tunnel_info *tun_info;
 940         struct rtable *rt;
 941         struct flowi4 fl;
 942         __be16 sport;
 943         __be16 df;
 944         int err;
 945
 946         tun_info = skb_tunnel_info(skb);
 947         if (unlikely(!tun_info)) {
 948                 err = -EINVAL;
 949                 goto error;
 950         }
 951
 952         tun_key = &tun_info->key;
 953
 954         /* Route lookup */
 955         memset(&fl, 0, sizeof(fl));
 956         fl.daddr = tun_key->u.ipv4.dst;
 957         fl.saddr = tun_key->u.ipv4.src;
 958         fl.flowi4_tos = RT_TOS(tun_key->tos);
 959         fl.flowi4_mark = skb->mark;
 960         fl.flowi4_proto = IPPROTO_TCP;
 961         rt = ip_route_output_key(net, &fl);
 962         if (IS_ERR(rt)) {
 963                 err = PTR_ERR(rt);
 964                 goto error;
 965         }
 966
 967         df = tun_key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0;
 968         sport = udp_flow_src_port(net, skb, 1, USHRT_MAX, true);
 969         skb->ignore_df = 1;
 970
 971         err = stt_xmit_skb(skb, rt, fl.saddr, tun_key->u.ipv4.dst,
 972                             tun_key->tos, tun_key->ttl,
 973                             df, sport, dport, tun_key->tun_id);
 974         iptunnel_xmit_stats(err, &dev->stats, (struct pcpu_sw_netstats __percpu *)dev->tstats);
 975         return NETDEV_TX_OK;
 976 error:
 977         kfree_skb(skb);
 978         dev->stats.tx_errors++;
 979         return NETDEV_TX_OK;
 980 }
 981 EXPORT_SYMBOL(ovs_stt_xmit);
 982
 983 static void free_frag(struct stt_percpu *stt_percpu,
 984                       struct pkt_frag *frag)
 985 {
 986         stt_percpu->frag_mem_used -= FRAG_CB(frag->skbs)->first.mem_used;
 987         kfree_skb_list(frag->skbs);
 988         list_del(&frag->lru_node);
 989         frag->skbs = NULL;
 990 }
 991
 992 static void evict_frags(struct stt_percpu *stt_percpu)
 993 {
 994         while (!list_empty(&stt_percpu->frag_lru) &&
 995                stt_percpu->frag_mem_used > REASM_LO_THRESH) {
 996                 struct pkt_frag *frag;
 997
 998                 frag = list_first_entry(&stt_percpu->frag_lru,
 999                                         struct pkt_frag,
1000                                         lru_node);
1001                 free_frag(stt_percpu, frag);
1002         }
1003 }
1004
1005 static bool pkt_key_match(struct net *net,
1006                           const struct pkt_frag *a, const struct pkt_key *b)
1007 {
1008         return a->key.saddr == b->saddr && a->key.daddr == b->daddr &&
1009                a->key.pkt_seq == b->pkt_seq && a->key.mark == b->mark &&
1010                net_eq(dev_net(a->skbs->dev), net);
1011 }
1012
1013 static u32 pkt_key_hash(const struct net *net, const struct pkt_key *key)
1014 {
1015         u32 initval = frag_hash_seed ^ (u32)(unsigned long)net ^ key->mark;
1016
1017         return jhash_3words((__force u32)key->saddr, (__force u32)key->daddr,
1018                             (__force u32)key->pkt_seq, initval);
1019 }
1020
1021 static struct pkt_frag *lookup_frag(struct net *net,
1022                                     struct stt_percpu *stt_percpu,
1023                                     const struct pkt_key *key, u32 hash)
1024 {
1025         struct pkt_frag *frag, *victim_frag = NULL;
1026         int i;
1027
1028         for (i = 0; i < FRAG_HASH_SEGS; i++) {
1029                 frag = flex_array_get(stt_percpu->frag_hash,
1030                                       hash & (FRAG_HASH_ENTRIES - 1));
1031
1032                 if (frag->skbs &&
1033                     time_before(jiffies, frag->timestamp + FRAG_EXP_TIME) &&
1034                     pkt_key_match(net, frag, key))
1035                         return frag;
1036
1037                 if (!victim_frag ||
1038                     (victim_frag->skbs &&
1039                      (!frag->skbs ||
1040                       time_before(frag->timestamp, victim_frag->timestamp))))
1041                         victim_frag = frag;
1042
1043                 hash >>= FRAG_HASH_SHIFT;
1044         }
1045
1046         if (victim_frag->skbs)
1047                 free_frag(stt_percpu, victim_frag);
1048
1049         return victim_frag;
1050 }
1051
1052 static struct sk_buff *reassemble(struct sk_buff *skb)
1053 {
1054         struct iphdr *iph = ip_hdr(skb);
1055         struct tcphdr *tcph = tcp_hdr(skb);
1056         u32 seq = ntohl(tcph->seq);
1057         struct stt_percpu *stt_percpu;
1058         struct sk_buff *last_skb;
1059         struct pkt_frag *frag;
1060         struct pkt_key key;
1061         int tot_len;
1062         u32 hash;
1063
1064         tot_len = seq >> STT_SEQ_LEN_SHIFT;
1065         FRAG_CB(skb)->offset = seq & STT_SEQ_OFFSET_MASK;
1066
1067         if (unlikely(skb->len == 0))
1068                 goto out_free;
1069
1070         if (unlikely(FRAG_CB(skb)->offset + skb->len > tot_len))
1071                 goto out_free;
1072
1073         if (tot_len == skb->len)
1074                 goto out;
1075
1076         key.saddr = iph->saddr;
1077         key.daddr = iph->daddr;
1078         key.pkt_seq = tcph->ack_seq;
1079         key.mark = skb->mark;
1080         hash = pkt_key_hash(dev_net(skb->dev), &key);
1081
1082         stt_percpu = per_cpu_ptr(stt_percpu_data, smp_processor_id());
1083
1084         spin_lock(&stt_percpu->lock);
1085
1086         if (unlikely(stt_percpu->frag_mem_used + skb->truesize > REASM_HI_THRESH))
1087                 evict_frags(stt_percpu);
1088
1089         frag = lookup_frag(dev_net(skb->dev), stt_percpu, &key, hash);
1090         if (!frag->skbs) {
1091                 frag->skbs = skb;
1092                 frag->key = key;
1093                 frag->timestamp = jiffies;
1094                 FRAG_CB(skb)->first.last_skb = skb;
1095                 FRAG_CB(skb)->first.mem_used = skb->truesize;
1096                 FRAG_CB(skb)->first.tot_len = tot_len;
1097                 FRAG_CB(skb)->first.rcvd_len = skb->len;
1098                 FRAG_CB(skb)->first.set_ecn_ce = false;
1099                 list_add_tail(&frag->lru_node, &stt_percpu->frag_lru);
1100                 stt_percpu->frag_mem_used += skb->truesize;
1101
1102                 skb = NULL;
1103                 goto unlock;
1104         }
1105
1106         /* Optimize for the common case where fragments are received in-order
1107          * and not overlapping.
1108          */
1109         last_skb = FRAG_CB(frag->skbs)->first.last_skb;
1110         if (likely(FRAG_CB(last_skb)->offset + last_skb->len ==
1111                    FRAG_CB(skb)->offset)) {
1112                 last_skb->next = skb;
1113                 FRAG_CB(frag->skbs)->first.last_skb = skb;
1114         } else {
1115                 struct sk_buff *prev = NULL, *next;
1116
1117                 for (next = frag->skbs; next; next = next->next) {
1118                         if (FRAG_CB(next)->offset >= FRAG_CB(skb)->offset)
1119                                 break;
1120                         prev = next;
1121                 }
1122
1123                 /* Overlapping fragments aren't allowed.  We shouldn't start
1124                  * before the end of the previous fragment.
1125                  */
1126                 if (prev &&
1127                     FRAG_CB(prev)->offset + prev->len > FRAG_CB(skb)->offset)
1128                         goto unlock_free;
1129
1130                 /* We also shouldn't end after the beginning of the next
1131                  * fragment.
1132                  */
1133                 if (next &&
1134                     FRAG_CB(skb)->offset + skb->len > FRAG_CB(next)->offset)
1135                         goto unlock_free;
1136
1137                 if (prev) {
1138                         prev->next = skb;
1139                 } else {
1140                         FRAG_CB(skb)->first = FRAG_CB(frag->skbs)->first;
1141                         frag->skbs = skb;
1142                 }
1143
1144                 if (next)
1145                         skb->next = next;
1146                 else
1147                         FRAG_CB(frag->skbs)->first.last_skb = skb;
1148         }
1149
1150         FRAG_CB(frag->skbs)->first.set_ecn_ce |= INET_ECN_is_ce(iph->tos);
1151         FRAG_CB(frag->skbs)->first.rcvd_len += skb->len;
1152         FRAG_CB(frag->skbs)->first.mem_used += skb->truesize;
1153         stt_percpu->frag_mem_used += skb->truesize;
1154
1155         if (FRAG_CB(frag->skbs)->first.tot_len ==
1156             FRAG_CB(frag->skbs)->first.rcvd_len) {
1157                 struct sk_buff *frag_head = frag->skbs;
1158
1159                 frag_head->tstamp = skb->tstamp;
1160                 if (FRAG_CB(frag_head)->first.set_ecn_ce)
1161                         INET_ECN_set_ce(frag_head);
1162
1163                 list_del(&frag->lru_node);
1164                 stt_percpu->frag_mem_used -= FRAG_CB(frag_head)->first.mem_used;
1165                 frag->skbs = NULL;
1166                 skb = frag_head;
1167         } else {
1168                 list_move_tail(&frag->lru_node, &stt_percpu->frag_lru);
1169                 skb = NULL;
1170         }
1171
1172         goto unlock;
1173
1174 unlock_free:
1175         kfree_skb(skb);
1176         skb = NULL;
1177 unlock:
1178         spin_unlock(&stt_percpu->lock);
1179         return skb;
1180 out_free:
1181         kfree_skb(skb);
1182         skb = NULL;
1183 out:
1184         return skb;
1185 }
1186
1187 static bool validate_checksum(struct sk_buff *skb)
1188 {
1189         struct iphdr *iph = ip_hdr(skb);
1190
1191         if (skb_csum_unnecessary(skb))
1192                 return true;
1193
1194         if (skb->ip_summed == CHECKSUM_COMPLETE &&
1195             !tcp_v4_check(skb->len, iph->saddr, iph->daddr, skb->csum))
1196                 return true;
1197
1198         skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, skb->len,
1199                                        IPPROTO_TCP, 0);
1200
1201         return __tcp_checksum_complete(skb) == 0;
1202 }
1203
1204 static bool set_offloads(struct sk_buff *skb)
1205 {
1206         struct stthdr *stth = stt_hdr(skb);
1207         unsigned short gso_type;
1208         int l3_header_size;
1209         int l4_header_size;
1210         u16 csum_offset;
1211         u8 proto_type;
1212
1213         if (stth->vlan_tci)
1214                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1215                                        ntohs(stth->vlan_tci));
1216
1217         if (!(stth->flags & STT_CSUM_PARTIAL)) {
1218                 if (stth->flags & STT_CSUM_VERIFIED)
1219                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1220                 else
1221                         skb->ip_summed = CHECKSUM_NONE;
1222
1223                 return clear_gso(skb) == 0;
1224         }
1225
1226         proto_type = stth->flags & STT_PROTO_TYPES;
1227
1228         switch (proto_type) {
1229         case (STT_PROTO_IPV4 | STT_PROTO_TCP):
1230                 /* TCP/IPv4 */
1231                 csum_offset = offsetof(struct tcphdr, check);
1232                 gso_type = SKB_GSO_TCPV4;
1233                 l3_header_size = sizeof(struct iphdr);
1234                 l4_header_size = sizeof(struct tcphdr);
1235                 skb->protocol = htons(ETH_P_IP);
1236                 break;
1237         case STT_PROTO_TCP:
1238                 /* TCP/IPv6 */
1239                 csum_offset = offsetof(struct tcphdr, check);
1240                 gso_type = SKB_GSO_TCPV6;
1241                 l3_header_size = sizeof(struct ipv6hdr);
1242                 l4_header_size = sizeof(struct tcphdr);
1243                 skb->protocol = htons(ETH_P_IPV6);
1244                 break;
1245         case STT_PROTO_IPV4:
1246                 /* UDP/IPv4 */
1247                 csum_offset = offsetof(struct udphdr, check);
1248                 gso_type = SKB_GSO_UDP;
1249                 l3_header_size = sizeof(struct iphdr);
1250                 l4_header_size = sizeof(struct udphdr);
1251                 skb->protocol = htons(ETH_P_IP);
1252                 break;
1253         default:
1254                 /* UDP/IPv6 */
1255                 csum_offset = offsetof(struct udphdr, check);
1256                 gso_type = SKB_GSO_UDP;
1257                 l3_header_size = sizeof(struct ipv6hdr);
1258                 l4_header_size = sizeof(struct udphdr);
1259                 skb->protocol = htons(ETH_P_IPV6);
1260         }
1261
1262         if (unlikely(stth->l4_offset < ETH_HLEN + l3_header_size))
1263                 return false;
1264
1265         if (unlikely(!pskb_may_pull(skb, stth->l4_offset + l4_header_size)))
1266                 return false;
1267
1268         stth = stt_hdr(skb);
1269
1270         skb->csum_start = skb_headroom(skb) + stth->l4_offset;
1271         skb->csum_offset = csum_offset;
1272         skb->ip_summed = CHECKSUM_PARTIAL;
1273
1274         if (stth->mss) {
1275                 if (unlikely(skb_unclone(skb, GFP_ATOMIC)))
1276                         return false;
1277
1278                 skb_shinfo(skb)->gso_type = gso_type | SKB_GSO_DODGY;
1279                 skb_shinfo(skb)->gso_size = ntohs(stth->mss);
1280                 skb_shinfo(skb)->gso_segs = 0;
1281         } else {
1282                 if (unlikely(clear_gso(skb)))
1283                         return false;
1284         }
1285
1286         return true;
1287 }
1288
1289 #ifndef HAVE_METADATA_DST
1290 static int __rcv(struct stt_dev *stt_dev, struct sk_buff *skb)
1291 {
1292         struct metadata_dst tun_dst;
1293
1294         ovs_ip_tun_rx_dst(&tun_dst.u.tun_info, skb, TUNNEL_KEY | TUNNEL_CSUM,
1295                           get_unaligned(&stt_hdr(skb)->key), 0);
1296         tun_dst.u.tun_info.key.tp_src = tcp_hdr(skb)->source;
1297         tun_dst.u.tun_info.key.tp_dst = tcp_hdr(skb)->dest;
1298
1299         ovs_ip_tunnel_rcv(stt_dev->dev, skb, &tun_dst);
1300         return 0;
1301 }
1302 #else
1303 static int __rcv(struct stt_dev *stt_dev, struct sk_buff *skb)
1304 {
1305         struct metadata_dst *tun_dst;
1306         __be16 flags;
1307         __be64 tun_id;
1308
1309         flags = TUNNEL_KEY | TUNNEL_CSUM;
1310         tun_id = get_unaligned(&stt_hdr(skb)->key);
1311         tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
1312         if (!tun_dst)
1313                 return -ENOMEM;
1314         tun_dst->u.tun_info.key.tp_src = tcp_hdr(skb)->source;
1315         tun_dst->u.tun_info.key.tp_dst = tcp_hdr(skb)->dest;
1316
1317         ovs_ip_tunnel_rcv(stt_dev->dev, skb, tun_dst);
1318         return 0;
1319 }
1320
1321 #endif
1322 static void stt_rcv(struct stt_dev *stt_dev, struct sk_buff *skb)
1323 {
1324         int err;
1325
1326         if (unlikely(!validate_checksum(skb)))
1327                 goto drop;
1328
1329         skb = reassemble(skb);
1330         if (!skb)
1331                 return;
1332
1333         if (skb->next && coalesce_skb(&skb))
1334                 goto drop;
1335
1336         err = iptunnel_pull_header(skb,
1337                                    sizeof(struct stthdr) + STT_ETH_PAD,
1338                                    htons(ETH_P_TEB));
1339         if (unlikely(err))
1340                 goto drop;
1341
1342         if (unlikely(stt_hdr(skb)->version != 0))
1343                 goto drop;
1344
1345         if (unlikely(!set_offloads(skb)))
1346                 goto drop;
1347
1348         if (skb_shinfo(skb)->frag_list && try_to_segment(skb))
1349                 goto drop;
1350
1351         err = __rcv(stt_dev, skb);
1352         if (err)
1353                 goto drop;
1354         return;
1355 drop:
1356         /* Consume bad packet */
1357         kfree_skb_list(skb);
1358         stt_dev->dev->stats.rx_errors++;
1359 }
1360
1361 static void tcp_sock_release(struct socket *sock)
1362 {
1363         kernel_sock_shutdown(sock, SHUT_RDWR);
1364         sock_release(sock);
1365 }
1366
1367 static int tcp_sock_create4(struct net *net, __be16 port,
1368                             struct socket **sockp)
1369 {
1370         struct sockaddr_in tcp_addr;
1371         struct socket *sock = NULL;
1372         int err;
1373
1374         err = sock_create_kern(net, AF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1375         if (err < 0)
1376                 goto error;
1377
1378         memset(&tcp_addr, 0, sizeof(tcp_addr));
1379         tcp_addr.sin_family = AF_INET;
1380         tcp_addr.sin_addr.s_addr = htonl(INADDR_ANY);
1381         tcp_addr.sin_port = port;
1382         err = kernel_bind(sock, (struct sockaddr *)&tcp_addr,
1383                           sizeof(tcp_addr));
1384         if (err < 0)
1385                 goto error;
1386
1387         *sockp = sock;
1388         return 0;
1389
1390 error:
1391         if (sock)
1392                 tcp_sock_release(sock);
1393         *sockp = NULL;
1394         return err;
1395 }
1396
1397 static void schedule_clean_percpu(void)
1398 {
1399         schedule_delayed_work(&clean_percpu_wq, CLEAN_PERCPU_INTERVAL);
1400 }
1401
1402 static void clean_percpu(struct work_struct *work)
1403 {
1404         int i;
1405
1406         for_each_possible_cpu(i) {
1407                 struct stt_percpu *stt_percpu = per_cpu_ptr(stt_percpu_data, i);
1408                 int j;
1409
1410                 for (j = 0; j < FRAG_HASH_ENTRIES; j++) {
1411                         struct pkt_frag *frag;
1412
1413                         frag = flex_array_get(stt_percpu->frag_hash, j);
1414                         if (!frag->skbs ||
1415                             time_before(jiffies, frag->timestamp + FRAG_EXP_TIME))
1416                                 continue;
1417
1418                         spin_lock_bh(&stt_percpu->lock);
1419
1420                         if (frag->skbs &&
1421                             time_after(jiffies, frag->timestamp + FRAG_EXP_TIME))
1422                                 free_frag(stt_percpu, frag);
1423
1424                         spin_unlock_bh(&stt_percpu->lock);
1425                 }
1426         }
1427         schedule_clean_percpu();
1428 }
1429
1430 #ifdef HAVE_NF_HOOKFN_ARG_OPS
1431 #define FIRST_PARAM const struct nf_hook_ops *ops
1432 #else
1433 #define FIRST_PARAM unsigned int hooknum
1434 #endif
1435
1436 #ifdef HAVE_NF_HOOK_STATE
1437 #if RHEL_RELEASE_CODE > RHEL_RELEASE_VERSION(7,0)
1438 /* RHEL nfhook hacks. */
1439 #ifndef __GENKSYMS__
1440 #define LAST_PARAM const struct net_device *in, const struct net_device *out, \
1441                    const struct nf_hook_state *state
1442 #else
1443 #define LAST_PARAM const struct net_device *in, const struct net_device *out, \
1444                    int (*okfn)(struct sk_buff *)
1445 #endif
1446 #else
1447 #define LAST_PARAM const struct nf_hook_state *state
1448 #endif
1449 #else
1450 #define LAST_PARAM const struct net_device *in, const struct net_device *out, \
1451                    int (*okfn)(struct sk_buff *)
1452 #endif
1453
1454 static unsigned int nf_ip_hook(FIRST_PARAM, struct sk_buff *skb, LAST_PARAM)
1455 {
1456         struct stt_dev *stt_dev;
1457         int ip_hdr_len;
1458
1459         if (ip_hdr(skb)->protocol != IPPROTO_TCP)
1460                 return NF_ACCEPT;
1461
1462         ip_hdr_len = ip_hdrlen(skb);
1463         if (unlikely(!pskb_may_pull(skb, ip_hdr_len + sizeof(struct tcphdr))))
1464                 return NF_ACCEPT;
1465
1466         skb_set_transport_header(skb, ip_hdr_len);
1467
1468         stt_dev = stt_find_sock(dev_net(skb->dev), tcp_hdr(skb)->dest);
1469         if (!stt_dev)
1470                 return NF_ACCEPT;
1471
1472         __skb_pull(skb, ip_hdr_len + sizeof(struct tcphdr));
1473         stt_rcv(stt_dev, skb);
1474         return NF_STOLEN;
1475 }
1476
1477 static struct nf_hook_ops nf_hook_ops __read_mostly = {
1478         .hook           = nf_ip_hook,
1479         .owner          = THIS_MODULE,
1480         .pf             = NFPROTO_IPV4,
1481         .hooknum        = NF_INET_LOCAL_IN,
1482         .priority       = INT_MAX,
1483 };
1484
1485 static int stt_start(struct net *net)
1486 {
1487         struct stt_net *sn = net_generic(net, stt_net_id);
1488         int err;
1489         int i;
1490
1491         if (n_tunnels) {
1492                 n_tunnels++;
1493                 return 0;
1494         }
1495         get_random_bytes(&frag_hash_seed, sizeof(u32));
1496
1497         stt_percpu_data = alloc_percpu(struct stt_percpu);
1498         if (!stt_percpu_data) {
1499                 err = -ENOMEM;
1500                 goto error;
1501         }
1502
1503         for_each_possible_cpu(i) {
1504                 struct stt_percpu *stt_percpu = per_cpu_ptr(stt_percpu_data, i);
1505                 struct flex_array *frag_hash;
1506
1507                 spin_lock_init(&stt_percpu->lock);
1508                 INIT_LIST_HEAD(&stt_percpu->frag_lru);
1509                 get_random_bytes(&per_cpu(pkt_seq_counter, i), sizeof(u32));
1510
1511                 frag_hash = flex_array_alloc(sizeof(struct pkt_frag),
1512                                              FRAG_HASH_ENTRIES,
1513                                              GFP_KERNEL | __GFP_ZERO);
1514                 if (!frag_hash) {
1515                         err = -ENOMEM;
1516                         goto free_percpu;
1517                 }
1518                 stt_percpu->frag_hash = frag_hash;
1519
1520                 err = flex_array_prealloc(stt_percpu->frag_hash, 0,
1521                                           FRAG_HASH_ENTRIES,
1522                                           GFP_KERNEL | __GFP_ZERO);
1523                 if (err)
1524                         goto free_percpu;
1525         }
1526         schedule_clean_percpu();
1527         n_tunnels++;
1528
1529         if (sn->n_tunnels) {
1530                 sn->n_tunnels++;
1531                 return 0;
1532         }
1533 #ifdef HAVE_NF_REGISTER_NET_HOOK
1534         /* On kernel which support per net nf-hook, nf_register_hook() takes
1535          * rtnl-lock, which results in dead lock in stt-dev-create. Therefore
1536          * use this new API.
1537          */
1538         err = nf_register_net_hook(net, &nf_hook_ops);
1539 #else
1540         err = nf_register_hook(&nf_hook_ops);
1541 #endif
1542         if (err)
1543                 goto free_percpu;
1544         sn->n_tunnels++;
1545         return 0;
1546
1547 free_percpu:
1548         for_each_possible_cpu(i) {
1549                 struct stt_percpu *stt_percpu = per_cpu_ptr(stt_percpu_data, i);
1550
1551                 if (stt_percpu->frag_hash)
1552                         flex_array_free(stt_percpu->frag_hash);
1553         }
1554
1555         free_percpu(stt_percpu_data);
1556
1557 error:
1558         return err;
1559 }
1560
1561 static void stt_cleanup(struct net *net)
1562 {
1563         struct stt_net *sn = net_generic(net, stt_net_id);
1564         int i;
1565
1566         sn->n_tunnels--;
1567         if (sn->n_tunnels)
1568                 goto out;
1569 #ifdef HAVE_NF_REGISTER_NET_HOOK
1570         nf_unregister_net_hook(net, &nf_hook_ops);
1571 #else
1572         nf_unregister_hook(&nf_hook_ops);
1573 #endif
1574
1575 out:
1576         n_tunnels--;
1577         if (n_tunnels)
1578                 return;
1579
1580         cancel_delayed_work_sync(&clean_percpu_wq);
1581         for_each_possible_cpu(i) {
1582                 struct stt_percpu *stt_percpu = per_cpu_ptr(stt_percpu_data, i);
1583                 int j;
1584
1585                 for (j = 0; j < FRAG_HASH_ENTRIES; j++) {
1586                         struct pkt_frag *frag;
1587
1588                         frag = flex_array_get(stt_percpu->frag_hash, j);
1589                         kfree_skb_list(frag->skbs);
1590                 }
1591
1592                 flex_array_free(stt_percpu->frag_hash);
1593         }
1594
1595         free_percpu(stt_percpu_data);
1596 }
1597
1598 static netdev_tx_t stt_dev_xmit(struct sk_buff *skb, struct net_device *dev)
1599 {
1600 #ifdef HAVE_METADATA_DST
1601         return ovs_stt_xmit(skb);
1602 #else
1603         /* Drop All packets coming from networking stack. OVS-CB is
1604          * not initialized for these packets.
1605          */
1606         dev_kfree_skb(skb);
1607         dev->stats.tx_dropped++;
1608         return NETDEV_TX_OK;
1609 #endif
1610 }
1611
1612 /* Setup stats when device is created */
1613 static int stt_init(struct net_device *dev)
1614 {
1615         dev->tstats = (typeof(dev->tstats)) netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1616         if (!dev->tstats)
1617                 return -ENOMEM;
1618
1619         return 0;
1620 }
1621
1622 static void stt_uninit(struct net_device *dev)
1623 {
1624         free_percpu(dev->tstats);
1625 }
1626
1627 static int stt_open(struct net_device *dev)
1628 {
1629         struct stt_dev *stt = netdev_priv(dev);
1630         struct net *net = stt->net;
1631         int err;
1632
1633         err = stt_start(net);
1634         if (err)
1635                 return err;
1636
1637         err = tcp_sock_create4(net, stt->dst_port, &stt->sock);
1638         if (err)
1639                 return err;
1640         return 0;
1641 }
1642
1643 static int stt_stop(struct net_device *dev)
1644 {
1645         struct stt_dev *stt_dev = netdev_priv(dev);
1646         struct net *net = stt_dev->net;
1647
1648         tcp_sock_release(stt_dev->sock);
1649         stt_dev->sock = NULL;
1650         stt_cleanup(net);
1651         return 0;
1652 }
1653
1654 static const struct net_device_ops stt_netdev_ops = {
1655         .ndo_init               = stt_init,
1656         .ndo_uninit             = stt_uninit,
1657         .ndo_open               = stt_open,
1658         .ndo_stop               = stt_stop,
1659         .ndo_start_xmit         = stt_dev_xmit,
1660         .ndo_get_stats64        = ip_tunnel_get_stats64,
1661         .ndo_change_mtu         = eth_change_mtu,
1662         .ndo_validate_addr      = eth_validate_addr,
1663         .ndo_set_mac_address    = eth_mac_addr,
1664 };
1665
1666 static void stt_get_drvinfo(struct net_device *dev,
1667                 struct ethtool_drvinfo *drvinfo)
1668 {
1669         strlcpy(drvinfo->version, STT_NETDEV_VER, sizeof(drvinfo->version));
1670         strlcpy(drvinfo->driver, "stt", sizeof(drvinfo->driver));
1671 }
1672
1673 static const struct ethtool_ops stt_ethtool_ops = {
1674         .get_drvinfo    = stt_get_drvinfo,
1675         .get_link       = ethtool_op_get_link,
1676 };
1677
1678 /* Info for udev, that this is a virtual tunnel endpoint */
1679 static struct device_type stt_type = {
1680         .name = "stt",
1681 };
1682
1683 /* Initialize the device structure. */
1684 static void stt_setup(struct net_device *dev)
1685 {
1686         ether_setup(dev);
1687
1688         dev->netdev_ops = &stt_netdev_ops;
1689         dev->ethtool_ops = &stt_ethtool_ops;
1690         dev->destructor = free_netdev;
1691
1692         SET_NETDEV_DEVTYPE(dev, &stt_type);
1693
1694         dev->features    |= NETIF_F_LLTX | NETIF_F_NETNS_LOCAL;
1695         dev->features    |= NETIF_F_SG | NETIF_F_HW_CSUM;
1696         dev->features    |= NETIF_F_RXCSUM;
1697         dev->features    |= NETIF_F_GSO_SOFTWARE;
1698
1699         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
1700         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1701
1702 #ifdef HAVE_METADATA_DST
1703         netif_keep_dst(dev);
1704 #endif
1705         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE;
1706         eth_hw_addr_random(dev);
1707 }
1708
1709 static const struct nla_policy stt_policy[IFLA_STT_MAX + 1] = {
1710         [IFLA_STT_PORT]              = { .type = NLA_U16 },
1711 };
1712
1713 static int stt_validate(struct nlattr *tb[], struct nlattr *data[])
1714 {
1715         if (tb[IFLA_ADDRESS]) {
1716                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1717                         return -EINVAL;
1718
1719                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1720                         return -EADDRNOTAVAIL;
1721         }
1722
1723         return 0;
1724 }
1725
1726 static struct stt_dev *find_dev(struct net *net, __be16 dst_port)
1727 {
1728         struct stt_net *sn = net_generic(net, stt_net_id);
1729         struct stt_dev *dev;
1730
1731         list_for_each_entry(dev, &sn->stt_list, next) {
1732                 if (dev->dst_port == dst_port)
1733                         return dev;
1734         }
1735         return NULL;
1736 }
1737
1738 static int stt_configure(struct net *net, struct net_device *dev,
1739                           __be16 dst_port)
1740 {
1741         struct stt_net *sn = net_generic(net, stt_net_id);
1742         struct stt_dev *stt = netdev_priv(dev);
1743         int err;
1744
1745         stt->net = net;
1746         stt->dev = dev;
1747
1748         stt->dst_port = dst_port;
1749
1750         if (find_dev(net, dst_port))
1751                 return -EBUSY;
1752
1753         err = register_netdevice(dev);
1754         if (err)
1755                 return err;
1756
1757         list_add(&stt->next, &sn->stt_list);
1758         return 0;
1759 }
1760
1761 static int stt_newlink(struct net *net, struct net_device *dev,
1762                 struct nlattr *tb[], struct nlattr *data[])
1763 {
1764         __be16 dst_port = htons(STT_DST_PORT);
1765
1766         if (data[IFLA_STT_PORT])
1767                 dst_port = nla_get_be16(data[IFLA_STT_PORT]);
1768
1769         return stt_configure(net, dev, dst_port);
1770 }
1771
1772 static void stt_dellink(struct net_device *dev, struct list_head *head)
1773 {
1774         struct stt_dev *stt = netdev_priv(dev);
1775
1776         list_del(&stt->next);
1777         unregister_netdevice_queue(dev, head);
1778 }
1779
1780 static size_t stt_get_size(const struct net_device *dev)
1781 {
1782         return nla_total_size(sizeof(__be32));  /* IFLA_STT_PORT */
1783 }
1784
1785 static int stt_fill_info(struct sk_buff *skb, const struct net_device *dev)
1786 {
1787         struct stt_dev *stt = netdev_priv(dev);
1788
1789         if (nla_put_be16(skb, IFLA_STT_PORT, stt->dst_port))
1790                 goto nla_put_failure;
1791
1792         return 0;
1793
1794 nla_put_failure:
1795         return -EMSGSIZE;
1796 }
1797
1798 static struct rtnl_link_ops stt_link_ops __read_mostly = {
1799         .kind           = "stt",
1800         .maxtype        = IFLA_STT_MAX,
1801         .policy         = stt_policy,
1802         .priv_size      = sizeof(struct stt_dev),
1803         .setup          = stt_setup,
1804         .validate       = stt_validate,
1805         .newlink        = stt_newlink,
1806         .dellink        = stt_dellink,
1807         .get_size       = stt_get_size,
1808         .fill_info      = stt_fill_info,
1809 };
1810
1811 struct net_device *ovs_stt_dev_create_fb(struct net *net, const char *name,
1812                                       u8 name_assign_type, u16 dst_port)
1813 {
1814         struct nlattr *tb[IFLA_MAX + 1];
1815         struct net_device *dev;
1816         int err;
1817
1818         memset(tb, 0, sizeof(tb));
1819         dev = rtnl_create_link(net, (char *) name, name_assign_type,
1820                         &stt_link_ops, tb);
1821         if (IS_ERR(dev))
1822                 return dev;
1823
1824         err = stt_configure(net, dev, htons(dst_port));
1825         if (err) {
1826                 free_netdev(dev);
1827                 return ERR_PTR(err);
1828         }
1829         return dev;
1830 }
1831 EXPORT_SYMBOL_GPL(ovs_stt_dev_create_fb);
1832
1833 static int stt_init_net(struct net *net)
1834 {
1835         struct stt_net *sn = net_generic(net, stt_net_id);
1836
1837         INIT_LIST_HEAD(&sn->stt_list);
1838         return 0;
1839 }
1840
1841 static void stt_exit_net(struct net *net)
1842 {
1843         struct stt_net *sn = net_generic(net, stt_net_id);
1844         struct stt_dev *stt, *next;
1845         struct net_device *dev, *aux;
1846         LIST_HEAD(list);
1847
1848         rtnl_lock();
1849
1850         /* gather any stt devices that were moved into this ns */
1851         for_each_netdev_safe(net, dev, aux)
1852                 if (dev->rtnl_link_ops == &stt_link_ops)
1853                         unregister_netdevice_queue(dev, &list);
1854
1855         list_for_each_entry_safe(stt, next, &sn->stt_list, next) {
1856                 /* If stt->dev is in the same netns, it was already added
1857                  * to the stt by the previous loop.
1858                  */
1859                 if (!net_eq(dev_net(stt->dev), net))
1860                         unregister_netdevice_queue(stt->dev, &list);
1861         }
1862
1863         /* unregister the devices gathered above */
1864         unregister_netdevice_many(&list);
1865         rtnl_unlock();
1866 }
1867
1868 static struct pernet_operations stt_net_ops = {
1869         .init = stt_init_net,
1870         .exit = stt_exit_net,
1871         .id   = &stt_net_id,
1872         .size = sizeof(struct stt_net),
1873 };
1874
1875 int stt_init_module(void)
1876 {
1877         int rc;
1878
1879         rc = register_pernet_subsys(&stt_net_ops);
1880         if (rc)
1881                 goto out1;
1882
1883         rc = rtnl_link_register(&stt_link_ops);
1884         if (rc)
1885                 goto out2;
1886
1887         pr_info("STT tunneling driver\n");
1888         return 0;
1889 out2:
1890         unregister_pernet_subsys(&stt_net_ops);
1891 out1:
1892         return rc;
1893 }
1894
1895 void stt_cleanup_module(void)
1896 {
1897         rtnl_link_unregister(&stt_link_ops);
1898         unregister_pernet_subsys(&stt_net_ops);
1899 }
1900 #endif