1caf43b02SWarner Losh /*- 251369649SPedro F. Giffuni * SPDX-License-Identifier: BSD-3-Clause 351369649SPedro F. Giffuni * 482cd038dSYoshinobu Inoue * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 582cd038dSYoshinobu Inoue * All rights reserved. 621f08a07SBjoern A. Zeeb * Copyright (c) 2019 Netflix, Inc. 782cd038dSYoshinobu Inoue * 882cd038dSYoshinobu Inoue * Redistribution and use in source and binary forms, with or without 982cd038dSYoshinobu Inoue * modification, are permitted provided that the following conditions 1082cd038dSYoshinobu Inoue * are met: 1182cd038dSYoshinobu Inoue * 1. Redistributions of source code must retain the above copyright 1282cd038dSYoshinobu Inoue * notice, this list of conditions and the following disclaimer. 1382cd038dSYoshinobu Inoue * 2. Redistributions in binary form must reproduce the above copyright 1482cd038dSYoshinobu Inoue * notice, this list of conditions and the following disclaimer in the 1582cd038dSYoshinobu Inoue * documentation and/or other materials provided with the distribution. 1682cd038dSYoshinobu Inoue * 3. Neither the name of the project nor the names of its contributors 1782cd038dSYoshinobu Inoue * may be used to endorse or promote products derived from this software 1882cd038dSYoshinobu Inoue * without specific prior written permission. 1982cd038dSYoshinobu Inoue * 2082cd038dSYoshinobu Inoue * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 2182cd038dSYoshinobu Inoue * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2282cd038dSYoshinobu Inoue * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2382cd038dSYoshinobu Inoue * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 2482cd038dSYoshinobu Inoue * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2582cd038dSYoshinobu Inoue * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2682cd038dSYoshinobu Inoue * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2782cd038dSYoshinobu Inoue * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2882cd038dSYoshinobu Inoue * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2982cd038dSYoshinobu Inoue * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3082cd038dSYoshinobu Inoue * SUCH DAMAGE. 31b48287a3SDavid E. O'Brien * 32b48287a3SDavid E. O'Brien * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $ 3382cd038dSYoshinobu Inoue */ 3482cd038dSYoshinobu Inoue 35b48287a3SDavid E. O'Brien #include <sys/cdefs.h> 36b48287a3SDavid E. O'Brien __FBSDID("$FreeBSD$"); 37b48287a3SDavid E. O'Brien 38aaa46574SAdrian Chadd #include "opt_rss.h" 39aaa46574SAdrian Chadd 4082cd038dSYoshinobu Inoue #include <sys/param.h> 41f349c821SBjoern A. Zeeb #include <sys/systm.h> 421a3044faSBjoern A. Zeeb #include <sys/domain.h> 431a3044faSBjoern A. Zeeb #include <sys/eventhandler.h> 4480d7a853SJonathan T. Looney #include <sys/hash.h> 451a3044faSBjoern A. Zeeb #include <sys/kernel.h> 4682cd038dSYoshinobu Inoue #include <sys/malloc.h> 4782cd038dSYoshinobu Inoue #include <sys/mbuf.h> 4882cd038dSYoshinobu Inoue #include <sys/protosw.h> 4921f08a07SBjoern A. Zeeb #include <sys/queue.h> 5082cd038dSYoshinobu Inoue #include <sys/socket.h> 51757cb678SBjoern A. Zeeb #include <sys/sysctl.h> 5282cd038dSYoshinobu Inoue #include <sys/syslog.h> 5382cd038dSYoshinobu Inoue 5482cd038dSYoshinobu Inoue #include <net/if.h> 5576039bc8SGleb Smirnoff #include <net/if_var.h> 56aaa46574SAdrian Chadd #include <net/netisr.h> 5782cd038dSYoshinobu Inoue #include <net/route.h> 58eddfbb76SRobert Watson #include <net/vnet.h> 5982cd038dSYoshinobu Inoue 6082cd038dSYoshinobu Inoue #include <netinet/in.h> 6182cd038dSYoshinobu Inoue #include <netinet/in_var.h> 62686cdd19SJun-ichiro itojun Hagino #include <netinet/ip6.h> 6382cd038dSYoshinobu Inoue #include <netinet6/ip6_var.h> 64686cdd19SJun-ichiro itojun Hagino #include <netinet/icmp6.h> 6523d374aaSBjoern A. Zeeb #include <netinet/in_systm.h> /* For ECN definitions. */ 6623d374aaSBjoern A. Zeeb #include <netinet/ip.h> /* For ECN definitions. */ 6782cd038dSYoshinobu Inoue 681a3044faSBjoern A. Zeeb #ifdef MAC 694b908c8bSRobert Watson #include <security/mac/mac_framework.h> 701a3044faSBjoern A. Zeeb #endif 714b908c8bSRobert Watson 72f1664f32SBjoern A. Zeeb /* 73f1664f32SBjoern A. Zeeb * A "big picture" of how IPv6 fragment queues are all linked together. 74f1664f32SBjoern A. Zeeb * 75f1664f32SBjoern A. Zeeb * struct ip6qbucket ip6qb[...]; hashed buckets 76f1664f32SBjoern A. Zeeb * |||||||| 77f1664f32SBjoern A. Zeeb * | 78f1664f32SBjoern A. Zeeb * +--- TAILQ(struct ip6q, packets) *q6; tailq entries holding 79f1664f32SBjoern A. Zeeb * |||||||| fragmented packets 80f1664f32SBjoern A. Zeeb * | (1 per original packet) 81f1664f32SBjoern A. Zeeb * | 82f1664f32SBjoern A. Zeeb * +--- TAILQ(struct ip6asfrag, ip6q_frags) *af6; tailq entries of IPv6 83f1664f32SBjoern A. Zeeb * | *ip6af;fragment packets 84f1664f32SBjoern A. Zeeb * | for one original packet 85f1664f32SBjoern A. Zeeb * + *mbuf 86f1664f32SBjoern A. Zeeb */ 87f1664f32SBjoern A. Zeeb 8823d374aaSBjoern A. Zeeb /* Reassembly headers are stored in hash buckets. */ 892ceeacbeSJonathan T. Looney #define IP6REASS_NHASH_LOG2 10 9080d7a853SJonathan T. Looney #define IP6REASS_NHASH (1 << IP6REASS_NHASH_LOG2) 9180d7a853SJonathan T. Looney #define IP6REASS_HMASK (IP6REASS_NHASH - 1) 9280d7a853SJonathan T. Looney 9321f08a07SBjoern A. Zeeb TAILQ_HEAD(ip6qhead, ip6q); 9480d7a853SJonathan T. Looney struct ip6qbucket { 9521f08a07SBjoern A. Zeeb struct ip6qhead packets; 9680d7a853SJonathan T. Looney struct mtx lock; 971e9f3b73SJonathan T. Looney int count; 9880d7a853SJonathan T. Looney }; 9980d7a853SJonathan T. Looney 1001540a98eSBjoern A. Zeeb struct ip6asfrag { 10121f08a07SBjoern A. Zeeb TAILQ_ENTRY(ip6asfrag) ip6af_tq; 1021540a98eSBjoern A. Zeeb struct mbuf *ip6af_m; 103f1664f32SBjoern A. Zeeb int ip6af_offset; /* Offset in ip6af_m to next header. */ 104f1664f32SBjoern A. Zeeb int ip6af_frglen; /* Fragmentable part length. */ 105f1664f32SBjoern A. Zeeb int ip6af_off; /* Fragment offset. */ 106f1664f32SBjoern A. Zeeb bool ip6af_mff; /* More fragment bit in frag off. */ 1071540a98eSBjoern A. Zeeb }; 1081540a98eSBjoern A. Zeeb 109487a161cSBjoern A. Zeeb static MALLOC_DEFINE(M_FRAG6, "frag6", "IPv6 fragment reassembly header"); 110487a161cSBjoern A. Zeeb 11167a10c46SBjoern A. Zeeb #ifdef VIMAGE 11267a10c46SBjoern A. Zeeb /* A flag to indicate if IPv6 fragmentation is initialized. */ 11367a10c46SBjoern A. Zeeb VNET_DEFINE_STATIC(bool, frag6_on); 11467a10c46SBjoern A. Zeeb #define V_frag6_on VNET(frag6_on) 11567a10c46SBjoern A. Zeeb #endif 11667a10c46SBjoern A. Zeeb 117757cb678SBjoern A. Zeeb /* System wide (global) maximum and count of packets in reassembly queues. */ 118757cb678SBjoern A. Zeeb static int ip6_maxfrags; 119757cb678SBjoern A. Zeeb static volatile u_int frag6_nfrags = 0; 120757cb678SBjoern A. Zeeb 121757cb678SBjoern A. Zeeb /* Maximum and current packets in per-VNET reassembly queue. */ 122757cb678SBjoern A. Zeeb VNET_DEFINE_STATIC(int, ip6_maxfragpackets); 12380d7a853SJonathan T. Looney VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets); 124757cb678SBjoern A. Zeeb #define V_ip6_maxfragpackets VNET(ip6_maxfragpackets) 125757cb678SBjoern A. Zeeb #define V_frag6_nfragpackets VNET(frag6_nfragpackets) 126757cb678SBjoern A. Zeeb 127757cb678SBjoern A. Zeeb /* Maximum per-VNET reassembly queues per bucket and fragments per packet. */ 128757cb678SBjoern A. Zeeb VNET_DEFINE_STATIC(int, ip6_maxfragbucketsize); 129757cb678SBjoern A. Zeeb VNET_DEFINE_STATIC(int, ip6_maxfragsperpacket); 130757cb678SBjoern A. Zeeb #define V_ip6_maxfragbucketsize VNET(ip6_maxfragbucketsize) 131757cb678SBjoern A. Zeeb #define V_ip6_maxfragsperpacket VNET(ip6_maxfragsperpacket) 132757cb678SBjoern A. Zeeb 133757cb678SBjoern A. Zeeb /* Per-VNET reassembly queue buckets. */ 1349cb1a47aSBjoern A. Zeeb VNET_DEFINE_STATIC(struct ip6qbucket, ip6qb[IP6REASS_NHASH]); 1359cb1a47aSBjoern A. Zeeb VNET_DEFINE_STATIC(uint32_t, ip6qb_hashseed); 1369cb1a47aSBjoern A. Zeeb #define V_ip6qb VNET(ip6qb) 1379cb1a47aSBjoern A. Zeeb #define V_ip6qb_hashseed VNET(ip6qb_hashseed) 13882cd038dSYoshinobu Inoue 1399cb1a47aSBjoern A. Zeeb #define IP6QB_LOCK(_b) mtx_lock(&V_ip6qb[(_b)].lock) 1409cb1a47aSBjoern A. Zeeb #define IP6QB_TRYLOCK(_b) mtx_trylock(&V_ip6qb[(_b)].lock) 1419cb1a47aSBjoern A. Zeeb #define IP6QB_LOCK_ASSERT(_b) mtx_assert(&V_ip6qb[(_b)].lock, MA_OWNED) 1429cb1a47aSBjoern A. Zeeb #define IP6QB_UNLOCK(_b) mtx_unlock(&V_ip6qb[(_b)].lock) 14321f08a07SBjoern A. Zeeb #define IP6QB_HEAD(_b) (&V_ip6qb[(_b)].packets) 1449888c401SHajimu UMEMOTO 14582cd038dSYoshinobu Inoue /* 1462ceeacbeSJonathan T. Looney * By default, limit the number of IP6 fragments across all reassembly 1472ceeacbeSJonathan T. Looney * queues to 1/32 of the total number of mbuf clusters. 1482ceeacbeSJonathan T. Looney * 1492ceeacbeSJonathan T. Looney * Limit the total number of reassembly queues per VNET to the 1502ceeacbeSJonathan T. Looney * IP6 fragment limit, but ensure the limit will not allow any bucket 1512ceeacbeSJonathan T. Looney * to grow above 100 items. (The bucket limit is 1522ceeacbeSJonathan T. Looney * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct 1532ceeacbeSJonathan T. Looney * multiplier to reach a 100-item limit.) 1542ceeacbeSJonathan T. Looney * The 100-item limit was chosen as brief testing seems to show that 1552ceeacbeSJonathan T. Looney * this produces "reasonable" performance on some subset of systems 1562ceeacbeSJonathan T. Looney * under DoS attack. 1572ceeacbeSJonathan T. Looney */ 1582ceeacbeSJonathan T. Looney #define IP6_MAXFRAGS (nmbclusters / 32) 1592ceeacbeSJonathan T. Looney #define IP6_MAXFRAGPACKETS (imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50)) 1602ceeacbeSJonathan T. Looney 161757cb678SBjoern A. Zeeb 1622ceeacbeSJonathan T. Looney /* 163757cb678SBjoern A. Zeeb * Sysctls and helper function. 16482cd038dSYoshinobu Inoue */ 165757cb678SBjoern A. Zeeb SYSCTL_DECL(_net_inet6_ip6); 166757cb678SBjoern A. Zeeb 16765456706SBjoern A. Zeeb SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfrags, 16865456706SBjoern A. Zeeb CTLFLAG_RD, __DEVOLATILE(u_int *, &frag6_nfrags), 0, 16965456706SBjoern A. Zeeb "Global number of IPv6 fragments across all reassembly queues."); 17065456706SBjoern A. Zeeb 171757cb678SBjoern A. Zeeb static void 17209b361c7SBjoern A. Zeeb frag6_set_bucketsize(void) 1731e9f3b73SJonathan T. Looney { 1741e9f3b73SJonathan T. Looney int i; 1751e9f3b73SJonathan T. Looney 1761e9f3b73SJonathan T. Looney if ((i = V_ip6_maxfragpackets) > 0) 1771e9f3b73SJonathan T. Looney V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1); 1781e9f3b73SJonathan T. Looney } 1791e9f3b73SJonathan T. Looney 180757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags, 181757cb678SBjoern A. Zeeb CTLFLAG_RW, &ip6_maxfrags, 0, 182757cb678SBjoern A. Zeeb "Maximum allowed number of outstanding IPv6 packet fragments. " 183757cb678SBjoern A. Zeeb "A value of 0 means no fragmented packets will be accepted, while a " 184757cb678SBjoern A. Zeeb "a value of -1 means no limit"); 185757cb678SBjoern A. Zeeb 186757cb678SBjoern A. Zeeb static int 187757cb678SBjoern A. Zeeb sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS) 188757cb678SBjoern A. Zeeb { 189757cb678SBjoern A. Zeeb int error, val; 190757cb678SBjoern A. Zeeb 191757cb678SBjoern A. Zeeb val = V_ip6_maxfragpackets; 192757cb678SBjoern A. Zeeb error = sysctl_handle_int(oidp, &val, 0, req); 193757cb678SBjoern A. Zeeb if (error != 0 || !req->newptr) 194757cb678SBjoern A. Zeeb return (error); 195757cb678SBjoern A. Zeeb V_ip6_maxfragpackets = val; 196757cb678SBjoern A. Zeeb frag6_set_bucketsize(); 197757cb678SBjoern A. Zeeb return (0); 198757cb678SBjoern A. Zeeb } 199757cb678SBjoern A. Zeeb SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets, 200757cb678SBjoern A. Zeeb CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, NULL, 0, 201757cb678SBjoern A. Zeeb sysctl_ip6_maxfragpackets, "I", 202757cb678SBjoern A. Zeeb "Default maximum number of outstanding fragmented IPv6 packets. " 203757cb678SBjoern A. Zeeb "A value of 0 means no fragmented packets will be accepted, while a " 204757cb678SBjoern A. Zeeb "a value of -1 means no limit"); 20553707abdSBjoern A. Zeeb SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfragpackets, 20653707abdSBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RD, 20753707abdSBjoern A. Zeeb __DEVOLATILE(u_int *, &VNET_NAME(frag6_nfragpackets)), 0, 20853707abdSBjoern A. Zeeb "Per-VNET number of IPv6 fragments across all reassembly queues."); 209757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket, 210757cb678SBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0, 211757cb678SBjoern A. Zeeb "Maximum allowed number of fragments per packet"); 212757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize, 213757cb678SBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0, 214757cb678SBjoern A. Zeeb "Maximum number of reassembly queues per hash bucket"); 215757cb678SBjoern A. Zeeb 216757cb678SBjoern A. Zeeb 217757cb678SBjoern A. Zeeb /* 218c00464a2SBjoern A. Zeeb * Remove the IPv6 fragmentation header from the mbuf. 219c00464a2SBjoern A. Zeeb */ 220c00464a2SBjoern A. Zeeb int 221a61b5cfbSBjoern A. Zeeb ip6_deletefraghdr(struct mbuf *m, int offset, int wait __unused) 222c00464a2SBjoern A. Zeeb { 2235778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 224a61b5cfbSBjoern A. Zeeb 225a61b5cfbSBjoern A. Zeeb KASSERT(m->m_len >= offset + sizeof(struct ip6_frag), 226a61b5cfbSBjoern A. Zeeb ("%s: ext headers not contigous in mbuf %p m_len %d >= " 227a61b5cfbSBjoern A. Zeeb "offset %d + %zu\n", __func__, m, m->m_len, offset, 228a61b5cfbSBjoern A. Zeeb sizeof(struct ip6_frag))); 229c00464a2SBjoern A. Zeeb 230c00464a2SBjoern A. Zeeb /* Delete frag6 header. */ 2315778b399SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 232a61b5cfbSBjoern A. Zeeb bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), offset); 233c00464a2SBjoern A. Zeeb m->m_data += sizeof(struct ip6_frag); 234c00464a2SBjoern A. Zeeb m->m_len -= sizeof(struct ip6_frag); 235c00464a2SBjoern A. Zeeb m->m_flags |= M_FRAGMENTED; 236a61b5cfbSBjoern A. Zeeb 237c00464a2SBjoern A. Zeeb return (0); 238c00464a2SBjoern A. Zeeb } 239c00464a2SBjoern A. Zeeb 240c00464a2SBjoern A. Zeeb /* 24123d374aaSBjoern A. Zeeb * Free a fragment reassembly header and all associated datagrams. 242757cb678SBjoern A. Zeeb */ 2434f590175SPaul Saab static void 244c00464a2SBjoern A. Zeeb frag6_freef(struct ip6q *q6, uint32_t bucket) 2454f590175SPaul Saab { 2465778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 24721f08a07SBjoern A. Zeeb struct ip6asfrag *af6; 2485778b399SBjoern A. Zeeb struct mbuf *m; 2494f590175SPaul Saab 2509cb1a47aSBjoern A. Zeeb IP6QB_LOCK_ASSERT(bucket); 251c00464a2SBjoern A. Zeeb 25221f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 253c00464a2SBjoern A. Zeeb 254da89a0feSBjoern A. Zeeb m = af6->ip6af_m; 25521f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 256c00464a2SBjoern A. Zeeb 257c00464a2SBjoern A. Zeeb /* 258c00464a2SBjoern A. Zeeb * Return ICMP time exceeded error for the 1st fragment. 259c00464a2SBjoern A. Zeeb * Just free other fragments. 260c00464a2SBjoern A. Zeeb */ 261a55383e7SHans Petter Selasky if (af6->ip6af_off == 0 && m->m_pkthdr.rcvif != NULL) { 262c00464a2SBjoern A. Zeeb 26323d374aaSBjoern A. Zeeb /* Adjust pointer. */ 264c00464a2SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 265c00464a2SBjoern A. Zeeb 26623d374aaSBjoern A. Zeeb /* Restore source and destination addresses. */ 267c00464a2SBjoern A. Zeeb ip6->ip6_src = q6->ip6q_src; 268c00464a2SBjoern A. Zeeb ip6->ip6_dst = q6->ip6q_dst; 269c00464a2SBjoern A. Zeeb 270c00464a2SBjoern A. Zeeb icmp6_error(m, ICMP6_TIME_EXCEEDED, 271c00464a2SBjoern A. Zeeb ICMP6_TIME_EXCEED_REASSEMBLY, 0); 272c00464a2SBjoern A. Zeeb } else 273c00464a2SBjoern A. Zeeb m_freem(m); 27423d374aaSBjoern A. Zeeb 275c00464a2SBjoern A. Zeeb free(af6, M_FRAG6); 2762adfd64fSJonathan T. Looney } 27721f08a07SBjoern A. Zeeb 27821f08a07SBjoern A. Zeeb TAILQ_REMOVE(IP6QB_HEAD(bucket), q6, ip6q_tq); 27921f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 280c00464a2SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 281c00464a2SBjoern A. Zeeb #ifdef MAC 282c00464a2SBjoern A. Zeeb mac_ip6q_destroy(q6); 283c00464a2SBjoern A. Zeeb #endif 284c00464a2SBjoern A. Zeeb free(q6, M_FRAG6); 285c00464a2SBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 28682cd038dSYoshinobu Inoue } 28782cd038dSYoshinobu Inoue 28882cd038dSYoshinobu Inoue /* 289a55383e7SHans Petter Selasky * Drain off all datagram fragments belonging to 290a55383e7SHans Petter Selasky * the given network interface. 291a55383e7SHans Petter Selasky */ 292a55383e7SHans Petter Selasky static void 293a55383e7SHans Petter Selasky frag6_cleanup(void *arg __unused, struct ifnet *ifp) 294a55383e7SHans Petter Selasky { 29521f08a07SBjoern A. Zeeb struct ip6qhead *head; 29621f08a07SBjoern A. Zeeb struct ip6q *q6; 297a55383e7SHans Petter Selasky struct ip6asfrag *af6; 29821f08a07SBjoern A. Zeeb uint32_t bucket; 299a55383e7SHans Petter Selasky 300a55383e7SHans Petter Selasky KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); 301a55383e7SHans Petter Selasky 3026e6b5143SBjoern A. Zeeb CURVNET_SET_QUIET(ifp->if_vnet); 30367a10c46SBjoern A. Zeeb #ifdef VIMAGE 30467a10c46SBjoern A. Zeeb /* 30567a10c46SBjoern A. Zeeb * Skip processing if IPv6 reassembly is not initialised or 30667a10c46SBjoern A. Zeeb * torn down by frag6_destroy(). 30767a10c46SBjoern A. Zeeb */ 3086e6b5143SBjoern A. Zeeb if (!V_frag6_on) { 3096e6b5143SBjoern A. Zeeb CURVNET_RESTORE(); 31067a10c46SBjoern A. Zeeb return; 3116e6b5143SBjoern A. Zeeb } 31267a10c46SBjoern A. Zeeb #endif 31367a10c46SBjoern A. Zeeb 31421f08a07SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 31521f08a07SBjoern A. Zeeb IP6QB_LOCK(bucket); 31621f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 317a55383e7SHans Petter Selasky /* Scan fragment list. */ 31821f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) { 31921f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 320a55383e7SHans Petter Selasky 321f1664f32SBjoern A. Zeeb /* Clear no longer valid rcvif pointer. */ 322da89a0feSBjoern A. Zeeb if (af6->ip6af_m->m_pkthdr.rcvif == ifp) 323da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.rcvif = NULL; 324a55383e7SHans Petter Selasky } 325a55383e7SHans Petter Selasky } 32621f08a07SBjoern A. Zeeb IP6QB_UNLOCK(bucket); 327a55383e7SHans Petter Selasky } 328a55383e7SHans Petter Selasky CURVNET_RESTORE(); 329a55383e7SHans Petter Selasky } 330a55383e7SHans Petter Selasky EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0); 331a55383e7SHans Petter Selasky 332a55383e7SHans Petter Selasky /* 33323d374aaSBjoern A. Zeeb * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with 33423d374aaSBjoern A. Zeeb * each other, in terms of next header field handling in fragment header. 335686cdd19SJun-ichiro itojun Hagino * While the sender will use the same value for all of the fragmented packets, 33623d374aaSBjoern A. Zeeb * receiver is suggested not to check for consistency. 337686cdd19SJun-ichiro itojun Hagino * 33823d374aaSBjoern A. Zeeb * Fragment rules (p18,p19): 339686cdd19SJun-ichiro itojun Hagino * (2) A Fragment header containing: 34023d374aaSBjoern A. Zeeb * The Next Header value that identifies the first header 34123d374aaSBjoern A. Zeeb * after the Per-Fragment headers of the original packet. 342686cdd19SJun-ichiro itojun Hagino * -> next header field is same for all fragments 343686cdd19SJun-ichiro itojun Hagino * 34423d374aaSBjoern A. Zeeb * Reassembly rule (p20): 34523d374aaSBjoern A. Zeeb * The Next Header field of the last header of the Per-Fragment 34623d374aaSBjoern A. Zeeb * headers is obtained from the Next Header field of the first 347686cdd19SJun-ichiro itojun Hagino * fragment's Fragment header. 348686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 349686cdd19SJun-ichiro itojun Hagino * 350686cdd19SJun-ichiro itojun Hagino * The following note also contradicts with fragment rule - no one is going to 351686cdd19SJun-ichiro itojun Hagino * send different fragment with different next header field. 352686cdd19SJun-ichiro itojun Hagino * 35323d374aaSBjoern A. Zeeb * Additional note (p22) [not an error]: 354686cdd19SJun-ichiro itojun Hagino * The Next Header values in the Fragment headers of different 355686cdd19SJun-ichiro itojun Hagino * fragments of the same original packet may differ. Only the value 356686cdd19SJun-ichiro itojun Hagino * from the Offset zero fragment packet is used for reassembly. 357686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 358686cdd19SJun-ichiro itojun Hagino * 359686cdd19SJun-ichiro itojun Hagino * There is no explicit reason given in the RFC. Historical reason maybe? 360686cdd19SJun-ichiro itojun Hagino */ 361686cdd19SJun-ichiro itojun Hagino /* 36223d374aaSBjoern A. Zeeb * Fragment input. 36382cd038dSYoshinobu Inoue */ 36482cd038dSYoshinobu Inoue int 3651272577eSXin LI frag6_input(struct mbuf **mp, int *offp, int proto) 36682cd038dSYoshinobu Inoue { 36721f08a07SBjoern A. Zeeb struct mbuf *m, *t; 36882cd038dSYoshinobu Inoue struct ip6_hdr *ip6; 36982cd038dSYoshinobu Inoue struct ip6_frag *ip6f; 37021f08a07SBjoern A. Zeeb struct ip6qhead *head; 37121f08a07SBjoern A. Zeeb struct ip6q *q6; 37221f08a07SBjoern A. Zeeb struct ip6asfrag *af6, *ip6af, *af6tmp; 37321f08a07SBjoern A. Zeeb struct in6_ifaddr *ia6; 37421f08a07SBjoern A. Zeeb struct ifnet *dstifp, *srcifp; 375505e91f5SKristof Provost uint32_t hashkey[(sizeof(struct in6_addr) * 2 + 376505e91f5SKristof Provost sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)]; 3779cb1a47aSBjoern A. Zeeb uint32_t bucket, *hashkeyp; 3785778b399SBjoern A. Zeeb int fragoff, frgpartlen; /* Must be larger than uint16_t. */ 3795778b399SBjoern A. Zeeb int nxt, offset, plen; 3805778b399SBjoern A. Zeeb uint8_t ecn, ecn0; 3815778b399SBjoern A. Zeeb bool only_frag; 382aaa46574SAdrian Chadd #ifdef RSS 383aaa46574SAdrian Chadd struct ip6_direct_ctx *ip6dc; 3845778b399SBjoern A. Zeeb struct m_tag *mtag; 385aaa46574SAdrian Chadd #endif 386aaa46574SAdrian Chadd 3875778b399SBjoern A. Zeeb m = *mp; 3885778b399SBjoern A. Zeeb offset = *offp; 3895778b399SBjoern A. Zeeb 390c1131de6SBjoern A. Zeeb M_ASSERTPKTHDR(m); 391c1131de6SBjoern A. Zeeb 392a4adf6ccSBjoern A. Zeeb if (m->m_len < offset + sizeof(struct ip6_frag)) { 393a61b5cfbSBjoern A. Zeeb m = m_pullup(m, offset + sizeof(struct ip6_frag)); 394a61b5cfbSBjoern A. Zeeb if (m == NULL) { 395a61b5cfbSBjoern A. Zeeb IP6STAT_INC(ip6s_exthdrtoolong); 396a61b5cfbSBjoern A. Zeeb *mp = NULL; 39740e39bbbSHajimu UMEMOTO return (IPPROTO_DONE); 398a61b5cfbSBjoern A. Zeeb } 399a4adf6ccSBjoern A. Zeeb } 400a61b5cfbSBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 40182cd038dSYoshinobu Inoue 40282cd038dSYoshinobu Inoue dstifp = NULL; 40323d374aaSBjoern A. Zeeb /* Find the destination interface of the packet. */ 4045778b399SBjoern A. Zeeb ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 4055778b399SBjoern A. Zeeb if (ia6 != NULL) { 4065778b399SBjoern A. Zeeb dstifp = ia6->ia_ifp; 4075778b399SBjoern A. Zeeb ifa_free(&ia6->ia_ifa); 4088c0fec80SRobert Watson } 40923d374aaSBjoern A. Zeeb 41023d374aaSBjoern A. Zeeb /* Jumbo payload cannot contain a fragment header. */ 41182cd038dSYoshinobu Inoue if (ip6->ip6_plen == 0) { 41282cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); 41382cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 414a8fe77d8SBjoern A. Zeeb *mp = NULL; 4155778b399SBjoern A. Zeeb return (IPPROTO_DONE); 41682cd038dSYoshinobu Inoue } 41782cd038dSYoshinobu Inoue 41882cd038dSYoshinobu Inoue /* 41923d374aaSBjoern A. Zeeb * Check whether fragment packet's fragment length is a 42023d374aaSBjoern A. Zeeb * multiple of 8 octets (unless it is the last one). 42182cd038dSYoshinobu Inoue * sizeof(struct ip6_frag) == 8 42282cd038dSYoshinobu Inoue * sizeof(struct ip6_hdr) = 40 42382cd038dSYoshinobu Inoue */ 424a61b5cfbSBjoern A. Zeeb ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); 42582cd038dSYoshinobu Inoue if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && 42682cd038dSYoshinobu Inoue (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { 42706cd0a3fSHajimu UMEMOTO icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 428686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_hdr, ip6_plen)); 42982cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 430a8fe77d8SBjoern A. Zeeb *mp = NULL; 4315778b399SBjoern A. Zeeb return (IPPROTO_DONE); 43282cd038dSYoshinobu Inoue } 43382cd038dSYoshinobu Inoue 4349cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragments); 43582cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_reqd); 43682cd038dSYoshinobu Inoue 4374018ea9aSBjoern A. Zeeb /* 4382946a941STom Jones * Handle "atomic" fragments (offset and m bit set to 0) upfront, 439c1131de6SBjoern A. Zeeb * unrelated to any reassembly. We need to remove the frag hdr 440c1131de6SBjoern A. Zeeb * which is ugly. 44123d374aaSBjoern A. Zeeb * See RFC 6946 and section 4.5 of RFC 8200. 4424018ea9aSBjoern A. Zeeb */ 4434018ea9aSBjoern A. Zeeb if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { 4442946a941STom Jones IP6STAT_INC(ip6s_atomicfrags); 445c1131de6SBjoern A. Zeeb nxt = ip6f->ip6f_nxt; 446c1131de6SBjoern A. Zeeb /* 447c1131de6SBjoern A. Zeeb * Set nxt(-hdr field value) to the original value. 448c1131de6SBjoern A. Zeeb * We cannot just set ip6->ip6_nxt as there might be 449c1131de6SBjoern A. Zeeb * an unfragmentable part with extension headers and 450c1131de6SBjoern A. Zeeb * we must update the last one. 451c1131de6SBjoern A. Zeeb */ 452c1131de6SBjoern A. Zeeb m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 453c1131de6SBjoern A. Zeeb (caddr_t)&nxt); 454c1131de6SBjoern A. Zeeb ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - 455c1131de6SBjoern A. Zeeb sizeof(struct ip6_frag)); 456c1131de6SBjoern A. Zeeb if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) 457c1131de6SBjoern A. Zeeb goto dropfrag2; 458c1131de6SBjoern A. Zeeb m->m_pkthdr.len -= sizeof(struct ip6_frag); 4594018ea9aSBjoern A. Zeeb in6_ifstat_inc(dstifp, ifs6_reass_ok); 460c1131de6SBjoern A. Zeeb *mp = m; 461c1131de6SBjoern A. Zeeb return (nxt); 4624018ea9aSBjoern A. Zeeb } 4634018ea9aSBjoern A. Zeeb 464c1131de6SBjoern A. Zeeb /* Offset now points to data portion. */ 465c1131de6SBjoern A. Zeeb offset += sizeof(struct ip6_frag); 466c1131de6SBjoern A. Zeeb 4675f9f192dSJonathan T. Looney /* Get fragment length and discard 0-byte fragments. */ 4685f9f192dSJonathan T. Looney frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; 4695f9f192dSJonathan T. Looney if (frgpartlen == 0) { 4705f9f192dSJonathan T. Looney icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 4715f9f192dSJonathan T. Looney offsetof(struct ip6_hdr, ip6_plen)); 4725f9f192dSJonathan T. Looney in6_ifstat_inc(dstifp, ifs6_reass_fail); 4735f9f192dSJonathan T. Looney IP6STAT_INC(ip6s_fragdropped); 474a8fe77d8SBjoern A. Zeeb *mp = NULL; 4755778b399SBjoern A. Zeeb return (IPPROTO_DONE); 4765f9f192dSJonathan T. Looney } 4775f9f192dSJonathan T. Looney 4787715d794SBjoern A. Zeeb /* 4797715d794SBjoern A. Zeeb * Enforce upper bound on number of fragments for the entire system. 4807715d794SBjoern A. Zeeb * If maxfrag is 0, never accept fragments. 4817715d794SBjoern A. Zeeb * If maxfrag is -1, accept all fragments without limitation. 4827715d794SBjoern A. Zeeb */ 4837715d794SBjoern A. Zeeb if (ip6_maxfrags < 0) 4847715d794SBjoern A. Zeeb ; 4857715d794SBjoern A. Zeeb else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags) 4867715d794SBjoern A. Zeeb goto dropfrag2; 4877715d794SBjoern A. Zeeb 48830809ba9SBjoern A. Zeeb /* 48930809ba9SBjoern A. Zeeb * Validate that a full header chain to the ULP is present in the 49030809ba9SBjoern A. Zeeb * packet containing the first fragment as per RFC RFC7112 and 49130809ba9SBjoern A. Zeeb * RFC 8200 pages 18,19: 49230809ba9SBjoern A. Zeeb * The first fragment packet is composed of: 49330809ba9SBjoern A. Zeeb * (3) Extension headers, if any, and the Upper-Layer header. These 49430809ba9SBjoern A. Zeeb * headers must be in the first fragment. ... 49530809ba9SBjoern A. Zeeb */ 49630809ba9SBjoern A. Zeeb fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); 49730809ba9SBjoern A. Zeeb /* XXX TODO. thj has D16851 open for this. */ 49830809ba9SBjoern A. Zeeb /* Send ICMPv6 4,3 in case of violation. */ 49930809ba9SBjoern A. Zeeb 500efdfee93SBjoern A. Zeeb /* Store receive network interface pointer for later. */ 501efdfee93SBjoern A. Zeeb srcifp = m->m_pkthdr.rcvif; 502efdfee93SBjoern A. Zeeb 50323d374aaSBjoern A. Zeeb /* Generate a hash value for fragment bucket selection. */ 50480d7a853SJonathan T. Looney hashkeyp = hashkey; 50580d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr)); 50680d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 50780d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr)); 50880d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 50980d7a853SJonathan T. Looney *hashkeyp = ip6f->ip6f_ident; 5109cb1a47aSBjoern A. Zeeb bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed); 5119cb1a47aSBjoern A. Zeeb bucket &= IP6REASS_HMASK; 5129cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 51321f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 5149888c401SHajimu UMEMOTO 51521f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) 51682cd038dSYoshinobu Inoue if (ip6f->ip6f_ident == q6->ip6q_ident && 51782cd038dSYoshinobu Inoue IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && 5184b908c8bSRobert Watson IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) 5194b908c8bSRobert Watson #ifdef MAC 5204b908c8bSRobert Watson && mac_ip6q_match(m, q6) 5214b908c8bSRobert Watson #endif 5224b908c8bSRobert Watson ) 52382cd038dSYoshinobu Inoue break; 52482cd038dSYoshinobu Inoue 5255778b399SBjoern A. Zeeb only_frag = false; 52621f08a07SBjoern A. Zeeb if (q6 == NULL) { 52723d374aaSBjoern A. Zeeb 5285778b399SBjoern A. Zeeb /* A first fragment to arrive creates a reassembly queue. */ 5295778b399SBjoern A. Zeeb only_frag = true; 53082cd038dSYoshinobu Inoue 53182cd038dSYoshinobu Inoue /* 53282cd038dSYoshinobu Inoue * Enforce upper bound on number of fragmented packets 53382cd038dSYoshinobu Inoue * for which we attempt reassembly; 5349888c401SHajimu UMEMOTO * If maxfragpackets is 0, never accept fragments. 5359888c401SHajimu UMEMOTO * If maxfragpackets is -1, accept all fragments without 5369888c401SHajimu UMEMOTO * limitation. 53782cd038dSYoshinobu Inoue */ 538603724d3SBjoern A. Zeeb if (V_ip6_maxfragpackets < 0) 53933841545SHajimu UMEMOTO ; 5409cb1a47aSBjoern A. Zeeb else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize || 5411e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) >= 54280d7a853SJonathan T. Looney (u_int)V_ip6_maxfragpackets) 54333841545SHajimu UMEMOTO goto dropfrag; 54423d374aaSBjoern A. Zeeb 54523d374aaSBjoern A. Zeeb /* Allocate IPv6 fragement packet queue entry. */ 546487a161cSBjoern A. Zeeb q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6, 547487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 54882cd038dSYoshinobu Inoue if (q6 == NULL) 54982cd038dSYoshinobu Inoue goto dropfrag; 5504b908c8bSRobert Watson #ifdef MAC 5514b908c8bSRobert Watson if (mac_ip6q_init(q6, M_NOWAIT) != 0) { 552487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 5534b908c8bSRobert Watson goto dropfrag; 5544b908c8bSRobert Watson } 5554b908c8bSRobert Watson mac_ip6q_create(m, q6); 5564b908c8bSRobert Watson #endif 557702828f6SBjoern A. Zeeb atomic_add_int(&V_frag6_nfragpackets, 1); 55882cd038dSYoshinobu Inoue 55923d374aaSBjoern A. Zeeb /* ip6q_nxt will be filled afterwards, from 1st fragment. */ 56021f08a07SBjoern A. Zeeb TAILQ_INIT(&q6->ip6q_frags); 56182cd038dSYoshinobu Inoue q6->ip6q_ident = ip6f->ip6f_ident; 56282cd038dSYoshinobu Inoue q6->ip6q_ttl = IPV6_FRAGTTL; 56382cd038dSYoshinobu Inoue q6->ip6q_src = ip6->ip6_src; 56482cd038dSYoshinobu Inoue q6->ip6q_dst = ip6->ip6_dst; 5655e9510e3SJINMEI Tatuya q6->ip6q_ecn = 5665e9510e3SJINMEI Tatuya (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 56782cd038dSYoshinobu Inoue q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ 5689888c401SHajimu UMEMOTO 56921f08a07SBjoern A. Zeeb /* Add the fragemented packet to the bucket. */ 57021f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(head, q6, ip6q_tq); 57121f08a07SBjoern A. Zeeb V_ip6qb[bucket].count++; 57282cd038dSYoshinobu Inoue } 57382cd038dSYoshinobu Inoue 57482cd038dSYoshinobu Inoue /* 57523d374aaSBjoern A. Zeeb * If it is the 1st fragment, record the length of the 57682cd038dSYoshinobu Inoue * unfragmentable part and the next header of the fragment header. 577619456bbSBjoern A. Zeeb * Assume the first 1st fragement to arrive will be correct. 578619456bbSBjoern A. Zeeb * We do not have any duplicate checks here yet so another packet 579619456bbSBjoern A. Zeeb * with fragoff == 0 could come and overwrite the ip6q_unfrglen 580619456bbSBjoern A. Zeeb * and worse, the next header, at any time. 58182cd038dSYoshinobu Inoue */ 582619456bbSBjoern A. Zeeb if (fragoff == 0 && q6->ip6q_unfrglen == -1) { 58306cd0a3fSHajimu UMEMOTO q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - 58406cd0a3fSHajimu UMEMOTO sizeof(struct ip6_frag); 58582cd038dSYoshinobu Inoue q6->ip6q_nxt = ip6f->ip6f_nxt; 586619456bbSBjoern A. Zeeb /* XXX ECN? */ 58782cd038dSYoshinobu Inoue } 58882cd038dSYoshinobu Inoue 58982cd038dSYoshinobu Inoue /* 59082cd038dSYoshinobu Inoue * Check that the reassembled packet would not exceed 65535 bytes 59182cd038dSYoshinobu Inoue * in size. 59282cd038dSYoshinobu Inoue * If it would exceed, discard the fragment and return an ICMP error. 59382cd038dSYoshinobu Inoue */ 59482cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen >= 0) { 59582cd038dSYoshinobu Inoue /* The 1st fragment has already arrived. */ 59682cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { 597e5fffe9aSBjoern A. Zeeb if (only_frag) { 598e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 599e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 600e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 601e5fffe9aSBjoern A. Zeeb #ifdef MAC 602e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 603e5fffe9aSBjoern A. Zeeb #endif 604e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 605e5fffe9aSBjoern A. Zeeb } 606e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 60782cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 608686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 609686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 610a8fe77d8SBjoern A. Zeeb *mp = NULL; 61182cd038dSYoshinobu Inoue return (IPPROTO_DONE); 61282cd038dSYoshinobu Inoue } 61306cd0a3fSHajimu UMEMOTO } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { 614e5fffe9aSBjoern A. Zeeb if (only_frag) { 615e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 616e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 617e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 618e5fffe9aSBjoern A. Zeeb #ifdef MAC 619e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 620e5fffe9aSBjoern A. Zeeb #endif 621e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 622e5fffe9aSBjoern A. Zeeb } 623e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 62482cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 625686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 626686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 627a8fe77d8SBjoern A. Zeeb *mp = NULL; 62882cd038dSYoshinobu Inoue return (IPPROTO_DONE); 62982cd038dSYoshinobu Inoue } 630f1664f32SBjoern A. Zeeb 63182cd038dSYoshinobu Inoue /* 63223d374aaSBjoern A. Zeeb * If it is the first fragment, do the above check for each 63382cd038dSYoshinobu Inoue * fragment already stored in the reassembly queue. 63482cd038dSYoshinobu Inoue */ 635dda02192SBjoern A. Zeeb if (fragoff == 0 && !only_frag) { 63621f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(af6, &q6->ip6q_frags, ip6af_tq, af6tmp) { 63782cd038dSYoshinobu Inoue 638dda02192SBjoern A. Zeeb if (q6->ip6q_unfrglen + af6->ip6af_off + 639dda02192SBjoern A. Zeeb af6->ip6af_frglen > IPV6_MAXPACKET) { 64082cd038dSYoshinobu Inoue struct ip6_hdr *ip6err; 6415778b399SBjoern A. Zeeb struct mbuf *merr; 6425778b399SBjoern A. Zeeb int erroff; 6435778b399SBjoern A. Zeeb 644da89a0feSBjoern A. Zeeb merr = af6->ip6af_m; 6455778b399SBjoern A. Zeeb erroff = af6->ip6af_offset; 64682cd038dSYoshinobu Inoue 64723d374aaSBjoern A. Zeeb /* Dequeue the fragment. */ 64821f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 649dda02192SBjoern A. Zeeb q6->ip6q_nfrag--; 650dda02192SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, 1); 651487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 65282cd038dSYoshinobu Inoue 653a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 654a55383e7SHans Petter Selasky merr->m_pkthdr.rcvif = srcifp; 655a55383e7SHans Petter Selasky 65623d374aaSBjoern A. Zeeb /* Adjust pointer. */ 65782cd038dSYoshinobu Inoue ip6err = mtod(merr, struct ip6_hdr *); 65882cd038dSYoshinobu Inoue 65982cd038dSYoshinobu Inoue /* 66082cd038dSYoshinobu Inoue * Restore source and destination addresses 66182cd038dSYoshinobu Inoue * in the erroneous IPv6 header. 66282cd038dSYoshinobu Inoue */ 66382cd038dSYoshinobu Inoue ip6err->ip6_src = q6->ip6q_src; 66482cd038dSYoshinobu Inoue ip6err->ip6_dst = q6->ip6q_dst; 66582cd038dSYoshinobu Inoue 66682cd038dSYoshinobu Inoue icmp6_error(merr, ICMP6_PARAM_PROB, 66782cd038dSYoshinobu Inoue ICMP6_PARAMPROB_HEADER, 668686cdd19SJun-ichiro itojun Hagino erroff - sizeof(struct ip6_frag) + 669686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 67082cd038dSYoshinobu Inoue } 67182cd038dSYoshinobu Inoue } 67282cd038dSYoshinobu Inoue } 67382cd038dSYoshinobu Inoue 67423d374aaSBjoern A. Zeeb /* Allocate an IPv6 fragement queue entry for this fragmented part. */ 675487a161cSBjoern A. Zeeb ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6, 676487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 677686cdd19SJun-ichiro itojun Hagino if (ip6af == NULL) 678686cdd19SJun-ichiro itojun Hagino goto dropfrag; 67921f08a07SBjoern A. Zeeb ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) ? true : false; 68082cd038dSYoshinobu Inoue ip6af->ip6af_off = fragoff; 68182cd038dSYoshinobu Inoue ip6af->ip6af_frglen = frgpartlen; 68282cd038dSYoshinobu Inoue ip6af->ip6af_offset = offset; 683da89a0feSBjoern A. Zeeb ip6af->ip6af_m = m; 68482cd038dSYoshinobu Inoue 6855778b399SBjoern A. Zeeb if (only_frag) { 68621f08a07SBjoern A. Zeeb /* 68721f08a07SBjoern A. Zeeb * Do a manual insert rather than a hard-to-understand cast 68821f08a07SBjoern A. Zeeb * to a different type relying on data structure order to work. 68921f08a07SBjoern A. Zeeb */ 69021f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(&q6->ip6q_frags, ip6af, ip6af_tq); 69121f08a07SBjoern A. Zeeb goto postinsert; 69282cd038dSYoshinobu Inoue } 69382cd038dSYoshinobu Inoue 69423d374aaSBjoern A. Zeeb /* Do duplicate, condition, and boundry checks. */ 69582cd038dSYoshinobu Inoue /* 69659dfcba4SHajimu UMEMOTO * Handle ECN by comparing this segment with the first one; 69759dfcba4SHajimu UMEMOTO * if CE is set, do not lose CE. 69823d374aaSBjoern A. Zeeb * Drop if CE and not-ECT are mixed for the same packet. 69959dfcba4SHajimu UMEMOTO */ 70059dfcba4SHajimu UMEMOTO ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 7015e9510e3SJINMEI Tatuya ecn0 = q6->ip6q_ecn; 70259dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_CE) { 70359dfcba4SHajimu UMEMOTO if (ecn0 == IPTOS_ECN_NOTECT) { 704487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 70559dfcba4SHajimu UMEMOTO goto dropfrag; 70659dfcba4SHajimu UMEMOTO } 70759dfcba4SHajimu UMEMOTO if (ecn0 != IPTOS_ECN_CE) 7085e9510e3SJINMEI Tatuya q6->ip6q_ecn = IPTOS_ECN_CE; 70959dfcba4SHajimu UMEMOTO } 71059dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { 711487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 71259dfcba4SHajimu UMEMOTO goto dropfrag; 71359dfcba4SHajimu UMEMOTO } 71459dfcba4SHajimu UMEMOTO 71523d374aaSBjoern A. Zeeb /* Find a fragmented part which begins after this one does. */ 71621f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) 71782cd038dSYoshinobu Inoue if (af6->ip6af_off > ip6af->ip6af_off) 71882cd038dSYoshinobu Inoue break; 71982cd038dSYoshinobu Inoue 72082cd038dSYoshinobu Inoue /* 72182cd038dSYoshinobu Inoue * If the incoming framgent overlaps some existing fragments in 72223d374aaSBjoern A. Zeeb * the reassembly queue, drop both the new fragment and the 72323d374aaSBjoern A. Zeeb * entire reassembly queue. However, if the new fragment 72423d374aaSBjoern A. Zeeb * is an exact duplicate of an existing fragment, only silently 72523d374aaSBjoern A. Zeeb * drop the existing fragment and leave the fragmentation queue 72623d374aaSBjoern A. Zeeb * unchanged, as allowed by the RFC. (RFC 8200, 4.5) 72782cd038dSYoshinobu Inoue */ 72821f08a07SBjoern A. Zeeb if (af6 != NULL) 72921f08a07SBjoern A. Zeeb af6tmp = TAILQ_PREV(af6, ip6fraghead, ip6af_tq); 73021f08a07SBjoern A. Zeeb else 73121f08a07SBjoern A. Zeeb af6tmp = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 73221f08a07SBjoern A. Zeeb if (af6tmp != NULL) { 73321f08a07SBjoern A. Zeeb if (af6tmp->ip6af_off + af6tmp->ip6af_frglen - 7345778b399SBjoern A. Zeeb ip6af->ip6af_off > 0) { 735cd188da2SBjoern A. Zeeb if (af6tmp->ip6af_off != ip6af->ip6af_off || 736cd188da2SBjoern A. Zeeb af6tmp->ip6af_frglen != ip6af->ip6af_frglen) 737cd188da2SBjoern A. Zeeb frag6_freef(q6, bucket); 738487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 73982cd038dSYoshinobu Inoue goto dropfrag; 74082cd038dSYoshinobu Inoue } 74182cd038dSYoshinobu Inoue } 74221f08a07SBjoern A. Zeeb if (af6 != NULL) { 7435778b399SBjoern A. Zeeb if (ip6af->ip6af_off + ip6af->ip6af_frglen - 7445778b399SBjoern A. Zeeb af6->ip6af_off > 0) { 745cd188da2SBjoern A. Zeeb if (af6->ip6af_off != ip6af->ip6af_off || 746cd188da2SBjoern A. Zeeb af6->ip6af_frglen != ip6af->ip6af_frglen) 747cd188da2SBjoern A. Zeeb frag6_freef(q6, bucket); 748487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 74982cd038dSYoshinobu Inoue goto dropfrag; 75082cd038dSYoshinobu Inoue } 75182cd038dSYoshinobu Inoue } 75282cd038dSYoshinobu Inoue 7534b908c8bSRobert Watson #ifdef MAC 7544b908c8bSRobert Watson mac_ip6q_update(m, q6); 7554b908c8bSRobert Watson #endif 75682cd038dSYoshinobu Inoue 75782cd038dSYoshinobu Inoue /* 75823d374aaSBjoern A. Zeeb * Stick new segment in its place; check for complete reassembly. 75923d374aaSBjoern A. Zeeb * If not complete, check fragment limit. Move to front of packet 76023d374aaSBjoern A. Zeeb * queue, as we are the most recently active fragmented packet. 76182cd038dSYoshinobu Inoue */ 76221f08a07SBjoern A. Zeeb if (af6 != NULL) 76321f08a07SBjoern A. Zeeb TAILQ_INSERT_BEFORE(af6, ip6af, ip6af_tq); 76421f08a07SBjoern A. Zeeb else 76521f08a07SBjoern A. Zeeb TAILQ_INSERT_TAIL(&q6->ip6q_frags, ip6af, ip6af_tq); 76621f08a07SBjoern A. Zeeb postinsert: 7672adfd64fSJonathan T. Looney atomic_add_int(&frag6_nfrags, 1); 7689888c401SHajimu UMEMOTO q6->ip6q_nfrag++; 7693c7165b3SBjoern A. Zeeb 7705778b399SBjoern A. Zeeb plen = 0; 77121f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 7725778b399SBjoern A. Zeeb if (af6->ip6af_off != plen) { 77303c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 774198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7759cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 77603c99d76SJonathan T. Looney } 7779cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 778a8fe77d8SBjoern A. Zeeb *mp = NULL; 7795778b399SBjoern A. Zeeb return (IPPROTO_DONE); 78082cd038dSYoshinobu Inoue } 7815778b399SBjoern A. Zeeb plen += af6->ip6af_frglen; 78282cd038dSYoshinobu Inoue } 78321f08a07SBjoern A. Zeeb af6 = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 78421f08a07SBjoern A. Zeeb if (af6->ip6af_mff) { 78503c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 786198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7879cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 78803c99d76SJonathan T. Looney } 7899cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 790a8fe77d8SBjoern A. Zeeb *mp = NULL; 7915778b399SBjoern A. Zeeb return (IPPROTO_DONE); 79282cd038dSYoshinobu Inoue } 79382cd038dSYoshinobu Inoue 79423d374aaSBjoern A. Zeeb /* Reassembly is complete; concatenate fragments. */ 79521f08a07SBjoern A. Zeeb ip6af = TAILQ_FIRST(&q6->ip6q_frags); 796da89a0feSBjoern A. Zeeb t = m = ip6af->ip6af_m; 79721f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, ip6af, ip6af_tq); 79821f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 7999907aba3SAndrey V. Elsukov m->m_pkthdr.csum_flags &= 800da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_flags; 8019907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data += 802da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_data; 8039907aba3SAndrey V. Elsukov 80421f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 805efdfee93SBjoern A. Zeeb t = m_last(t); 806da89a0feSBjoern A. Zeeb m_adj(af6->ip6af_m, af6->ip6af_offset); 807da89a0feSBjoern A. Zeeb m_demote_pkthdr(af6->ip6af_m); 808da89a0feSBjoern A. Zeeb m_cat(t, af6->ip6af_m); 809487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 81082cd038dSYoshinobu Inoue } 81182cd038dSYoshinobu Inoue 8129907aba3SAndrey V. Elsukov while (m->m_pkthdr.csum_data & 0xffff0000) 8139907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 8149907aba3SAndrey V. Elsukov (m->m_pkthdr.csum_data >> 16); 8159907aba3SAndrey V. Elsukov 81623d374aaSBjoern A. Zeeb /* Adjust offset to point where the original next header starts. */ 81782cd038dSYoshinobu Inoue offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); 818487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 819686cdd19SJun-ichiro itojun Hagino ip6 = mtod(m, struct ip6_hdr *); 8205778b399SBjoern A. Zeeb ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr)); 8215e9510e3SJINMEI Tatuya if (q6->ip6q_ecn == IPTOS_ECN_CE) 8225e9510e3SJINMEI Tatuya ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); 82382cd038dSYoshinobu Inoue nxt = q6->ip6q_nxt; 82482cd038dSYoshinobu Inoue 82521f08a07SBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 82621f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 8272adfd64fSJonathan T. Looney atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 82821f08a07SBjoern A. Zeeb 829a61b5cfbSBjoern A. Zeeb ip6_deletefraghdr(m, offset, M_NOWAIT); 83082cd038dSYoshinobu Inoue 83123d374aaSBjoern A. Zeeb /* Set nxt(-hdr field value) to the original value. */ 83268e0e5a6SAndrey V. Elsukov m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 83368e0e5a6SAndrey V. Elsukov (caddr_t)&nxt); 83482cd038dSYoshinobu Inoue 8354b908c8bSRobert Watson #ifdef MAC 8364b908c8bSRobert Watson mac_ip6q_reassemble(q6, m); 8374b908c8bSRobert Watson mac_ip6q_destroy(q6); 8384b908c8bSRobert Watson #endif 839487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 84080d7a853SJonathan T. Looney atomic_subtract_int(&V_frag6_nfragpackets, 1); 84182cd038dSYoshinobu Inoue 84282cd038dSYoshinobu Inoue if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ 8435778b399SBjoern A. Zeeb 8445778b399SBjoern A. Zeeb plen = 0; 84582cd038dSYoshinobu Inoue for (t = m; t; t = t->m_next) 84682cd038dSYoshinobu Inoue plen += t->m_len; 84782cd038dSYoshinobu Inoue m->m_pkthdr.len = plen; 848a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 849a55383e7SHans Petter Selasky m->m_pkthdr.rcvif = srcifp; 85082cd038dSYoshinobu Inoue } 85182cd038dSYoshinobu Inoue 852aaa46574SAdrian Chadd #ifdef RSS 853aaa46574SAdrian Chadd mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), 854aaa46574SAdrian Chadd M_NOWAIT); 855aaa46574SAdrian Chadd if (mtag == NULL) 856aaa46574SAdrian Chadd goto dropfrag; 857aaa46574SAdrian Chadd 858aaa46574SAdrian Chadd ip6dc = (struct ip6_direct_ctx *)(mtag + 1); 859aaa46574SAdrian Chadd ip6dc->ip6dc_nxt = nxt; 860aaa46574SAdrian Chadd ip6dc->ip6dc_off = offset; 861aaa46574SAdrian Chadd 862aaa46574SAdrian Chadd m_tag_prepend(m, mtag); 863aaa46574SAdrian Chadd #endif 864aaa46574SAdrian Chadd 8659cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 8669cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_reassembled); 86782cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_ok); 86882cd038dSYoshinobu Inoue 869aaa46574SAdrian Chadd #ifdef RSS 87023d374aaSBjoern A. Zeeb /* Queue/dispatch for reprocessing. */ 871aaa46574SAdrian Chadd netisr_dispatch(NETISR_IPV6_DIRECT, m); 872a8fe77d8SBjoern A. Zeeb *mp = NULL; 8735778b399SBjoern A. Zeeb return (IPPROTO_DONE); 874aaa46574SAdrian Chadd #endif 875aaa46574SAdrian Chadd 87623d374aaSBjoern A. Zeeb /* Tell launch routine the next header. */ 87782cd038dSYoshinobu Inoue *mp = m; 87882cd038dSYoshinobu Inoue *offp = offset; 87982cd038dSYoshinobu Inoue 8805778b399SBjoern A. Zeeb return (nxt); 88182cd038dSYoshinobu Inoue 88282cd038dSYoshinobu Inoue dropfrag: 8839cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 8847715d794SBjoern A. Zeeb dropfrag2: 88582cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 8869cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragdropped); 88782cd038dSYoshinobu Inoue m_freem(m); 888a8fe77d8SBjoern A. Zeeb *mp = NULL; 8895778b399SBjoern A. Zeeb return (IPPROTO_DONE); 89082cd038dSYoshinobu Inoue } 89182cd038dSYoshinobu Inoue 89282cd038dSYoshinobu Inoue /* 89333841545SHajimu UMEMOTO * IPv6 reassembling timer processing; 89423d374aaSBjoern A. Zeeb * if a timer expires on a reassembly queue, discard it. 89582cd038dSYoshinobu Inoue */ 89682cd038dSYoshinobu Inoue void 8971272577eSXin LI frag6_slowtimo(void) 89882cd038dSYoshinobu Inoue { 8998b615593SMarko Zec VNET_ITERATOR_DECL(vnet_iter); 90021f08a07SBjoern A. Zeeb struct ip6qhead *head; 90121f08a07SBjoern A. Zeeb struct ip6q *q6, *q6tmp; 9029cb1a47aSBjoern A. Zeeb uint32_t bucket; 90382cd038dSYoshinobu Inoue 9045ee847d3SRobert Watson VNET_LIST_RLOCK_NOSLEEP(); 9058b615593SMarko Zec VNET_FOREACH(vnet_iter) { 9068b615593SMarko Zec CURVNET_SET(vnet_iter); 9079cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 9089cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 9099cb1a47aSBjoern A. Zeeb head = IP6QB_HEAD(bucket); 91021f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(q6, head, ip6q_tq, q6tmp) 91121f08a07SBjoern A. Zeeb if (--q6->ip6q_ttl == 0) { 912198fdaedSTom Jones IP6STAT_ADD(ip6s_fragtimeout, 91321f08a07SBjoern A. Zeeb q6->ip6q_nfrag); 91482cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 91521f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 91682cd038dSYoshinobu Inoue } 91782cd038dSYoshinobu Inoue /* 91882cd038dSYoshinobu Inoue * If we are over the maximum number of fragments 91982cd038dSYoshinobu Inoue * (due to the limit being lowered), drain off 92082cd038dSYoshinobu Inoue * enough to get down to the new limit. 9211e9f3b73SJonathan T. Looney * Note that we drain all reassembly queues if 9221e9f3b73SJonathan T. Looney * maxfragpackets is 0 (fragmentation is disabled), 92323d374aaSBjoern A. Zeeb * and do not enforce a limit when maxfragpackets 9241e9f3b73SJonathan T. Looney * is negative. 92582cd038dSYoshinobu Inoue */ 9261e9f3b73SJonathan T. Looney while ((V_ip6_maxfragpackets == 0 || 9271e9f3b73SJonathan T. Looney (V_ip6_maxfragpackets > 0 && 9289cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) && 92921f08a07SBjoern A. Zeeb (q6 = TAILQ_LAST(head, ip6qhead)) != NULL) { 93021f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 93182cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 93221f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 93380d7a853SJonathan T. Looney } 9349cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 93582cd038dSYoshinobu Inoue } 9361e9f3b73SJonathan T. Looney /* 9371e9f3b73SJonathan T. Looney * If we are still over the maximum number of fragmented 9381e9f3b73SJonathan T. Looney * packets, drain off enough to get down to the new limit. 9391e9f3b73SJonathan T. Looney */ 9409cb1a47aSBjoern A. Zeeb bucket = 0; 9411e9f3b73SJonathan T. Looney while (V_ip6_maxfragpackets >= 0 && 9421e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) > 9431e9f3b73SJonathan T. Looney (u_int)V_ip6_maxfragpackets) { 9449cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 94521f08a07SBjoern A. Zeeb q6 = TAILQ_LAST(IP6QB_HEAD(bucket), ip6qhead); 94621f08a07SBjoern A. Zeeb if (q6 != NULL) { 94721f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 9481e9f3b73SJonathan T. Looney /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 94921f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 9501e9f3b73SJonathan T. Looney } 9519cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 9529cb1a47aSBjoern A. Zeeb bucket = (bucket + 1) % IP6REASS_NHASH; 9531e9f3b73SJonathan T. Looney } 9548b615593SMarko Zec CURVNET_RESTORE(); 9558b615593SMarko Zec } 9565ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 95782cd038dSYoshinobu Inoue } 95882cd038dSYoshinobu Inoue 95923d374aaSBjoern A. Zeeb /* 96023d374aaSBjoern A. Zeeb * Eventhandler to adjust limits in case nmbclusters change. 96123d374aaSBjoern A. Zeeb */ 962c00464a2SBjoern A. Zeeb static void 963c00464a2SBjoern A. Zeeb frag6_change(void *tag) 964c00464a2SBjoern A. Zeeb { 965c00464a2SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 966c00464a2SBjoern A. Zeeb 967c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 968c00464a2SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 969c00464a2SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 970c00464a2SBjoern A. Zeeb CURVNET_SET(vnet_iter); 971c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 972c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 973c00464a2SBjoern A. Zeeb CURVNET_RESTORE(); 974c00464a2SBjoern A. Zeeb } 975c00464a2SBjoern A. Zeeb VNET_LIST_RUNLOCK_NOSLEEP(); 976c00464a2SBjoern A. Zeeb } 977c00464a2SBjoern A. Zeeb 978c00464a2SBjoern A. Zeeb /* 979c00464a2SBjoern A. Zeeb * Initialise reassembly queue and fragment identifier. 980c00464a2SBjoern A. Zeeb */ 981c00464a2SBjoern A. Zeeb void 982c00464a2SBjoern A. Zeeb frag6_init(void) 983c00464a2SBjoern A. Zeeb { 9849cb1a47aSBjoern A. Zeeb uint32_t bucket; 985c00464a2SBjoern A. Zeeb 986c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 987c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 9889cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 98921f08a07SBjoern A. Zeeb TAILQ_INIT(IP6QB_HEAD(bucket)); 990efdfee93SBjoern A. Zeeb mtx_init(&V_ip6qb[bucket].lock, "ip6qb", NULL, MTX_DEF); 9919cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count = 0; 992c00464a2SBjoern A. Zeeb } 9939cb1a47aSBjoern A. Zeeb V_ip6qb_hashseed = arc4random(); 994c00464a2SBjoern A. Zeeb V_ip6_maxfragsperpacket = 64; 99567a10c46SBjoern A. Zeeb #ifdef VIMAGE 99667a10c46SBjoern A. Zeeb V_frag6_on = true; 99767a10c46SBjoern A. Zeeb #endif 998c00464a2SBjoern A. Zeeb if (!IS_DEFAULT_VNET(curvnet)) 999c00464a2SBjoern A. Zeeb return; 1000c00464a2SBjoern A. Zeeb 1001c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 1002c00464a2SBjoern A. Zeeb EVENTHANDLER_REGISTER(nmbclusters_change, 1003c00464a2SBjoern A. Zeeb frag6_change, NULL, EVENTHANDLER_PRI_ANY); 1004c00464a2SBjoern A. Zeeb } 1005c00464a2SBjoern A. Zeeb 100682cd038dSYoshinobu Inoue /* 100782cd038dSYoshinobu Inoue * Drain off all datagram fragments. 100882cd038dSYoshinobu Inoue */ 100967a10c46SBjoern A. Zeeb static void 101067a10c46SBjoern A. Zeeb frag6_drain_one(void) 101182cd038dSYoshinobu Inoue { 101221f08a07SBjoern A. Zeeb struct ip6q *q6; 10139cb1a47aSBjoern A. Zeeb uint32_t bucket; 10149888c401SHajimu UMEMOTO 10159cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 101667a10c46SBjoern A. Zeeb IP6QB_LOCK(bucket); 101721f08a07SBjoern A. Zeeb while ((q6 = TAILQ_FIRST(IP6QB_HEAD(bucket))) != NULL) { 10189cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragdropped); 101982cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 102021f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 102180d7a853SJonathan T. Looney } 10229cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 102382cd038dSYoshinobu Inoue } 102467a10c46SBjoern A. Zeeb } 102567a10c46SBjoern A. Zeeb 102667a10c46SBjoern A. Zeeb void 102767a10c46SBjoern A. Zeeb frag6_drain(void) 102867a10c46SBjoern A. Zeeb { 102967a10c46SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 103067a10c46SBjoern A. Zeeb 103167a10c46SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 103267a10c46SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 103367a10c46SBjoern A. Zeeb CURVNET_SET(vnet_iter); 103467a10c46SBjoern A. Zeeb frag6_drain_one(); 10358b615593SMarko Zec CURVNET_RESTORE(); 10368b615593SMarko Zec } 10375ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 103882cd038dSYoshinobu Inoue } 1039e5ee7060SGleb Smirnoff 104067a10c46SBjoern A. Zeeb #ifdef VIMAGE 104167a10c46SBjoern A. Zeeb /* 104267a10c46SBjoern A. Zeeb * Clear up IPv6 reassembly structures. 104367a10c46SBjoern A. Zeeb */ 104467a10c46SBjoern A. Zeeb void 104567a10c46SBjoern A. Zeeb frag6_destroy(void) 104667a10c46SBjoern A. Zeeb { 104767a10c46SBjoern A. Zeeb uint32_t bucket; 104867a10c46SBjoern A. Zeeb 104967a10c46SBjoern A. Zeeb frag6_drain_one(); 105067a10c46SBjoern A. Zeeb V_frag6_on = false; 105167a10c46SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 105267a10c46SBjoern A. Zeeb KASSERT(V_ip6qb[bucket].count == 0, 105367a10c46SBjoern A. Zeeb ("%s: V_ip6qb[%d] (%p) count not 0 (%d)", __func__, 105467a10c46SBjoern A. Zeeb bucket, &V_ip6qb[bucket], V_ip6qb[bucket].count)); 105567a10c46SBjoern A. Zeeb mtx_destroy(&V_ip6qb[bucket].lock); 105667a10c46SBjoern A. Zeeb } 105767a10c46SBjoern A. Zeeb } 105867a10c46SBjoern A. Zeeb #endif 1059