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; 119c17ae180SMateusz Guzik static u_int __exclusive_cache_line frag6_nfrags; 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 1612ceeacbeSJonathan T. Looney /* 162757cb678SBjoern A. Zeeb * Sysctls and helper function. 16382cd038dSYoshinobu Inoue */ 164757cb678SBjoern A. Zeeb SYSCTL_DECL(_net_inet6_ip6); 165757cb678SBjoern A. Zeeb 16665456706SBjoern A. Zeeb SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfrags, 167c17ae180SMateusz Guzik CTLFLAG_RD, &frag6_nfrags, 0, 16865456706SBjoern A. Zeeb "Global number of IPv6 fragments across all reassembly queues."); 16965456706SBjoern A. Zeeb 170757cb678SBjoern A. Zeeb static void 17109b361c7SBjoern A. Zeeb frag6_set_bucketsize(void) 1721e9f3b73SJonathan T. Looney { 1731e9f3b73SJonathan T. Looney int i; 1741e9f3b73SJonathan T. Looney 1751e9f3b73SJonathan T. Looney if ((i = V_ip6_maxfragpackets) > 0) 1761e9f3b73SJonathan T. Looney V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1); 1771e9f3b73SJonathan T. Looney } 1781e9f3b73SJonathan T. Looney 179757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGS, maxfrags, 180757cb678SBjoern A. Zeeb CTLFLAG_RW, &ip6_maxfrags, 0, 181757cb678SBjoern A. Zeeb "Maximum allowed number of outstanding IPv6 packet fragments. " 182757cb678SBjoern A. Zeeb "A value of 0 means no fragmented packets will be accepted, while a " 183757cb678SBjoern A. Zeeb "a value of -1 means no limit"); 184757cb678SBjoern A. Zeeb 185757cb678SBjoern A. Zeeb static int 186757cb678SBjoern A. Zeeb sysctl_ip6_maxfragpackets(SYSCTL_HANDLER_ARGS) 187757cb678SBjoern A. Zeeb { 188757cb678SBjoern A. Zeeb int error, val; 189757cb678SBjoern A. Zeeb 190757cb678SBjoern A. Zeeb val = V_ip6_maxfragpackets; 191757cb678SBjoern A. Zeeb error = sysctl_handle_int(oidp, &val, 0, req); 192757cb678SBjoern A. Zeeb if (error != 0 || !req->newptr) 193757cb678SBjoern A. Zeeb return (error); 194757cb678SBjoern A. Zeeb V_ip6_maxfragpackets = val; 195757cb678SBjoern A. Zeeb frag6_set_bucketsize(); 196757cb678SBjoern A. Zeeb return (0); 197757cb678SBjoern A. Zeeb } 198757cb678SBjoern A. Zeeb SYSCTL_PROC(_net_inet6_ip6, IPV6CTL_MAXFRAGPACKETS, maxfragpackets, 1997029da5cSPawel Biernacki CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 2007029da5cSPawel Biernacki NULL, 0, sysctl_ip6_maxfragpackets, "I", 201757cb678SBjoern A. Zeeb "Default maximum number of outstanding fragmented IPv6 packets. " 202757cb678SBjoern A. Zeeb "A value of 0 means no fragmented packets will be accepted, while a " 203757cb678SBjoern A. Zeeb "a value of -1 means no limit"); 20453707abdSBjoern A. Zeeb SYSCTL_UINT(_net_inet6_ip6, OID_AUTO, frag6_nfragpackets, 20553707abdSBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RD, 20653707abdSBjoern A. Zeeb __DEVOLATILE(u_int *, &VNET_NAME(frag6_nfragpackets)), 0, 20753707abdSBjoern A. Zeeb "Per-VNET number of IPv6 fragments across all reassembly queues."); 208757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket, 209757cb678SBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0, 210757cb678SBjoern A. Zeeb "Maximum allowed number of fragments per packet"); 211757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize, 212757cb678SBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0, 213757cb678SBjoern A. Zeeb "Maximum number of reassembly queues per hash bucket"); 214757cb678SBjoern A. Zeeb 215757cb678SBjoern A. Zeeb /* 216c00464a2SBjoern A. Zeeb * Remove the IPv6 fragmentation header from the mbuf. 217c00464a2SBjoern A. Zeeb */ 218c00464a2SBjoern A. Zeeb int 219a61b5cfbSBjoern A. Zeeb ip6_deletefraghdr(struct mbuf *m, int offset, int wait __unused) 220c00464a2SBjoern A. Zeeb { 2215778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 222a61b5cfbSBjoern A. Zeeb 223a61b5cfbSBjoern A. Zeeb KASSERT(m->m_len >= offset + sizeof(struct ip6_frag), 224a61b5cfbSBjoern A. Zeeb ("%s: ext headers not contigous in mbuf %p m_len %d >= " 225a61b5cfbSBjoern A. Zeeb "offset %d + %zu\n", __func__, m, m->m_len, offset, 226a61b5cfbSBjoern A. Zeeb sizeof(struct ip6_frag))); 227c00464a2SBjoern A. Zeeb 228c00464a2SBjoern A. Zeeb /* Delete frag6 header. */ 2295778b399SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 230a61b5cfbSBjoern A. Zeeb bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), offset); 231c00464a2SBjoern A. Zeeb m->m_data += sizeof(struct ip6_frag); 232c00464a2SBjoern A. Zeeb m->m_len -= sizeof(struct ip6_frag); 233c00464a2SBjoern A. Zeeb m->m_flags |= M_FRAGMENTED; 234a61b5cfbSBjoern A. Zeeb 235c00464a2SBjoern A. Zeeb return (0); 236c00464a2SBjoern A. Zeeb } 237c00464a2SBjoern A. Zeeb 238c00464a2SBjoern A. Zeeb /* 23923d374aaSBjoern A. Zeeb * Free a fragment reassembly header and all associated datagrams. 240757cb678SBjoern A. Zeeb */ 2414f590175SPaul Saab static void 242c00464a2SBjoern A. Zeeb frag6_freef(struct ip6q *q6, uint32_t bucket) 2434f590175SPaul Saab { 2445778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 24521f08a07SBjoern A. Zeeb struct ip6asfrag *af6; 2465778b399SBjoern A. Zeeb struct mbuf *m; 2474f590175SPaul Saab 2489cb1a47aSBjoern A. Zeeb IP6QB_LOCK_ASSERT(bucket); 249c00464a2SBjoern A. Zeeb 25021f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 251da89a0feSBjoern A. Zeeb m = af6->ip6af_m; 25221f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 253c00464a2SBjoern A. Zeeb 254c00464a2SBjoern A. Zeeb /* 255c00464a2SBjoern A. Zeeb * Return ICMP time exceeded error for the 1st fragment. 256c00464a2SBjoern A. Zeeb * Just free other fragments. 257c00464a2SBjoern A. Zeeb */ 258a55383e7SHans Petter Selasky if (af6->ip6af_off == 0 && m->m_pkthdr.rcvif != NULL) { 25923d374aaSBjoern A. Zeeb /* Adjust pointer. */ 260c00464a2SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 261c00464a2SBjoern A. Zeeb 26223d374aaSBjoern A. Zeeb /* Restore source and destination addresses. */ 263c00464a2SBjoern A. Zeeb ip6->ip6_src = q6->ip6q_src; 264c00464a2SBjoern A. Zeeb ip6->ip6_dst = q6->ip6q_dst; 265c00464a2SBjoern A. Zeeb 266c00464a2SBjoern A. Zeeb icmp6_error(m, ICMP6_TIME_EXCEEDED, 267c00464a2SBjoern A. Zeeb ICMP6_TIME_EXCEED_REASSEMBLY, 0); 268c00464a2SBjoern A. Zeeb } else 269c00464a2SBjoern A. Zeeb m_freem(m); 27023d374aaSBjoern A. Zeeb 271c00464a2SBjoern A. Zeeb free(af6, M_FRAG6); 2722adfd64fSJonathan T. Looney } 27321f08a07SBjoern A. Zeeb 27421f08a07SBjoern A. Zeeb TAILQ_REMOVE(IP6QB_HEAD(bucket), q6, ip6q_tq); 27521f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 276c00464a2SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 277c00464a2SBjoern A. Zeeb #ifdef MAC 278c00464a2SBjoern A. Zeeb mac_ip6q_destroy(q6); 279c00464a2SBjoern A. Zeeb #endif 280c00464a2SBjoern A. Zeeb free(q6, M_FRAG6); 281c00464a2SBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 28282cd038dSYoshinobu Inoue } 28382cd038dSYoshinobu Inoue 28482cd038dSYoshinobu Inoue /* 285a55383e7SHans Petter Selasky * Drain off all datagram fragments belonging to 286a55383e7SHans Petter Selasky * the given network interface. 287a55383e7SHans Petter Selasky */ 288a55383e7SHans Petter Selasky static void 289a55383e7SHans Petter Selasky frag6_cleanup(void *arg __unused, struct ifnet *ifp) 290a55383e7SHans Petter Selasky { 29121f08a07SBjoern A. Zeeb struct ip6qhead *head; 29221f08a07SBjoern A. Zeeb struct ip6q *q6; 293a55383e7SHans Petter Selasky struct ip6asfrag *af6; 29421f08a07SBjoern A. Zeeb uint32_t bucket; 295a55383e7SHans Petter Selasky 296a55383e7SHans Petter Selasky KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); 297a55383e7SHans Petter Selasky 2986e6b5143SBjoern A. Zeeb CURVNET_SET_QUIET(ifp->if_vnet); 29967a10c46SBjoern A. Zeeb #ifdef VIMAGE 30067a10c46SBjoern A. Zeeb /* 30167a10c46SBjoern A. Zeeb * Skip processing if IPv6 reassembly is not initialised or 30267a10c46SBjoern A. Zeeb * torn down by frag6_destroy(). 30367a10c46SBjoern A. Zeeb */ 3046e6b5143SBjoern A. Zeeb if (!V_frag6_on) { 3056e6b5143SBjoern A. Zeeb CURVNET_RESTORE(); 30667a10c46SBjoern A. Zeeb return; 3076e6b5143SBjoern A. Zeeb } 30867a10c46SBjoern A. Zeeb #endif 30967a10c46SBjoern A. Zeeb 31021f08a07SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 31121f08a07SBjoern A. Zeeb IP6QB_LOCK(bucket); 31221f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 313a55383e7SHans Petter Selasky /* Scan fragment list. */ 31421f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) { 31521f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 316f1664f32SBjoern A. Zeeb /* Clear no longer valid rcvif pointer. */ 317da89a0feSBjoern A. Zeeb if (af6->ip6af_m->m_pkthdr.rcvif == ifp) 318da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.rcvif = NULL; 319a55383e7SHans Petter Selasky } 320a55383e7SHans Petter Selasky } 32121f08a07SBjoern A. Zeeb IP6QB_UNLOCK(bucket); 322a55383e7SHans Petter Selasky } 323a55383e7SHans Petter Selasky CURVNET_RESTORE(); 324a55383e7SHans Petter Selasky } 325a55383e7SHans Petter Selasky EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0); 326a55383e7SHans Petter Selasky 327a55383e7SHans Petter Selasky /* 32823d374aaSBjoern A. Zeeb * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with 32923d374aaSBjoern A. Zeeb * each other, in terms of next header field handling in fragment header. 330686cdd19SJun-ichiro itojun Hagino * While the sender will use the same value for all of the fragmented packets, 33123d374aaSBjoern A. Zeeb * receiver is suggested not to check for consistency. 332686cdd19SJun-ichiro itojun Hagino * 33323d374aaSBjoern A. Zeeb * Fragment rules (p18,p19): 334686cdd19SJun-ichiro itojun Hagino * (2) A Fragment header containing: 33523d374aaSBjoern A. Zeeb * The Next Header value that identifies the first header 33623d374aaSBjoern A. Zeeb * after the Per-Fragment headers of the original packet. 337686cdd19SJun-ichiro itojun Hagino * -> next header field is same for all fragments 338686cdd19SJun-ichiro itojun Hagino * 33923d374aaSBjoern A. Zeeb * Reassembly rule (p20): 34023d374aaSBjoern A. Zeeb * The Next Header field of the last header of the Per-Fragment 34123d374aaSBjoern A. Zeeb * headers is obtained from the Next Header field of the first 342686cdd19SJun-ichiro itojun Hagino * fragment's Fragment header. 343686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 344686cdd19SJun-ichiro itojun Hagino * 345686cdd19SJun-ichiro itojun Hagino * The following note also contradicts with fragment rule - no one is going to 346686cdd19SJun-ichiro itojun Hagino * send different fragment with different next header field. 347686cdd19SJun-ichiro itojun Hagino * 34823d374aaSBjoern A. Zeeb * Additional note (p22) [not an error]: 349686cdd19SJun-ichiro itojun Hagino * The Next Header values in the Fragment headers of different 350686cdd19SJun-ichiro itojun Hagino * fragments of the same original packet may differ. Only the value 351686cdd19SJun-ichiro itojun Hagino * from the Offset zero fragment packet is used for reassembly. 352686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 353686cdd19SJun-ichiro itojun Hagino * 354686cdd19SJun-ichiro itojun Hagino * There is no explicit reason given in the RFC. Historical reason maybe? 355686cdd19SJun-ichiro itojun Hagino */ 356686cdd19SJun-ichiro itojun Hagino /* 35723d374aaSBjoern A. Zeeb * Fragment input. 35882cd038dSYoshinobu Inoue */ 35982cd038dSYoshinobu Inoue int 3601272577eSXin LI frag6_input(struct mbuf **mp, int *offp, int proto) 36182cd038dSYoshinobu Inoue { 36221f08a07SBjoern A. Zeeb struct mbuf *m, *t; 36382cd038dSYoshinobu Inoue struct ip6_hdr *ip6; 36482cd038dSYoshinobu Inoue struct ip6_frag *ip6f; 36521f08a07SBjoern A. Zeeb struct ip6qhead *head; 36621f08a07SBjoern A. Zeeb struct ip6q *q6; 36721f08a07SBjoern A. Zeeb struct ip6asfrag *af6, *ip6af, *af6tmp; 36821f08a07SBjoern A. Zeeb struct in6_ifaddr *ia6; 36921f08a07SBjoern A. Zeeb struct ifnet *dstifp, *srcifp; 370505e91f5SKristof Provost uint32_t hashkey[(sizeof(struct in6_addr) * 2 + 371505e91f5SKristof Provost sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)]; 3729cb1a47aSBjoern A. Zeeb uint32_t bucket, *hashkeyp; 3735778b399SBjoern A. Zeeb int fragoff, frgpartlen; /* Must be larger than uint16_t. */ 3745778b399SBjoern A. Zeeb int nxt, offset, plen; 3755778b399SBjoern A. Zeeb uint8_t ecn, ecn0; 3765778b399SBjoern A. Zeeb bool only_frag; 377aaa46574SAdrian Chadd #ifdef RSS 378aaa46574SAdrian Chadd struct ip6_direct_ctx *ip6dc; 3795778b399SBjoern A. Zeeb struct m_tag *mtag; 380aaa46574SAdrian Chadd #endif 381aaa46574SAdrian Chadd 3825778b399SBjoern A. Zeeb m = *mp; 3835778b399SBjoern A. Zeeb offset = *offp; 3845778b399SBjoern A. Zeeb 385c1131de6SBjoern A. Zeeb M_ASSERTPKTHDR(m); 386c1131de6SBjoern A. Zeeb 387a4adf6ccSBjoern A. Zeeb if (m->m_len < offset + sizeof(struct ip6_frag)) { 388a61b5cfbSBjoern A. Zeeb m = m_pullup(m, offset + sizeof(struct ip6_frag)); 389a61b5cfbSBjoern A. Zeeb if (m == NULL) { 390a61b5cfbSBjoern A. Zeeb IP6STAT_INC(ip6s_exthdrtoolong); 391a61b5cfbSBjoern A. Zeeb *mp = NULL; 39240e39bbbSHajimu UMEMOTO return (IPPROTO_DONE); 393a61b5cfbSBjoern A. Zeeb } 394a4adf6ccSBjoern A. Zeeb } 395a61b5cfbSBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 39682cd038dSYoshinobu Inoue 39782cd038dSYoshinobu Inoue dstifp = NULL; 39823d374aaSBjoern A. Zeeb /* Find the destination interface of the packet. */ 3998268d82cSAlexander V. Chernikov ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 4008268d82cSAlexander V. Chernikov if (ia6 != NULL) 4015778b399SBjoern A. Zeeb dstifp = ia6->ia_ifp; 40223d374aaSBjoern A. Zeeb 40323d374aaSBjoern A. Zeeb /* Jumbo payload cannot contain a fragment header. */ 40482cd038dSYoshinobu Inoue if (ip6->ip6_plen == 0) { 40582cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); 40682cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 407a8fe77d8SBjoern A. Zeeb *mp = NULL; 4085778b399SBjoern A. Zeeb return (IPPROTO_DONE); 40982cd038dSYoshinobu Inoue } 41082cd038dSYoshinobu Inoue 41182cd038dSYoshinobu Inoue /* 41223d374aaSBjoern A. Zeeb * Check whether fragment packet's fragment length is a 41323d374aaSBjoern A. Zeeb * multiple of 8 octets (unless it is the last one). 41482cd038dSYoshinobu Inoue * sizeof(struct ip6_frag) == 8 41582cd038dSYoshinobu Inoue * sizeof(struct ip6_hdr) = 40 41682cd038dSYoshinobu Inoue */ 417a61b5cfbSBjoern A. Zeeb ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); 41882cd038dSYoshinobu Inoue if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && 41982cd038dSYoshinobu Inoue (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { 42006cd0a3fSHajimu UMEMOTO icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 421686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_hdr, ip6_plen)); 42282cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 423a8fe77d8SBjoern A. Zeeb *mp = NULL; 4245778b399SBjoern A. Zeeb return (IPPROTO_DONE); 42582cd038dSYoshinobu Inoue } 42682cd038dSYoshinobu Inoue 4279cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragments); 42882cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_reqd); 42982cd038dSYoshinobu Inoue 4304018ea9aSBjoern A. Zeeb /* 4312946a941STom Jones * Handle "atomic" fragments (offset and m bit set to 0) upfront, 432c1131de6SBjoern A. Zeeb * unrelated to any reassembly. We need to remove the frag hdr 433c1131de6SBjoern A. Zeeb * which is ugly. 43423d374aaSBjoern A. Zeeb * See RFC 6946 and section 4.5 of RFC 8200. 4354018ea9aSBjoern A. Zeeb */ 4364018ea9aSBjoern A. Zeeb if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { 4372946a941STom Jones IP6STAT_INC(ip6s_atomicfrags); 438c1131de6SBjoern A. Zeeb nxt = ip6f->ip6f_nxt; 439c1131de6SBjoern A. Zeeb /* 440c1131de6SBjoern A. Zeeb * Set nxt(-hdr field value) to the original value. 441c1131de6SBjoern A. Zeeb * We cannot just set ip6->ip6_nxt as there might be 442c1131de6SBjoern A. Zeeb * an unfragmentable part with extension headers and 443c1131de6SBjoern A. Zeeb * we must update the last one. 444c1131de6SBjoern A. Zeeb */ 445c1131de6SBjoern A. Zeeb m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 446c1131de6SBjoern A. Zeeb (caddr_t)&nxt); 447c1131de6SBjoern A. Zeeb ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - 448c1131de6SBjoern A. Zeeb sizeof(struct ip6_frag)); 449c1131de6SBjoern A. Zeeb if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) 450c1131de6SBjoern A. Zeeb goto dropfrag2; 451c1131de6SBjoern A. Zeeb m->m_pkthdr.len -= sizeof(struct ip6_frag); 4524018ea9aSBjoern A. Zeeb in6_ifstat_inc(dstifp, ifs6_reass_ok); 453c1131de6SBjoern A. Zeeb *mp = m; 454c1131de6SBjoern A. Zeeb return (nxt); 4554018ea9aSBjoern A. Zeeb } 4564018ea9aSBjoern A. Zeeb 457c1131de6SBjoern A. Zeeb /* Offset now points to data portion. */ 458c1131de6SBjoern A. Zeeb offset += sizeof(struct ip6_frag); 459c1131de6SBjoern A. Zeeb 4605f9f192dSJonathan T. Looney /* Get fragment length and discard 0-byte fragments. */ 4615f9f192dSJonathan T. Looney frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; 4625f9f192dSJonathan T. Looney if (frgpartlen == 0) { 4635f9f192dSJonathan T. Looney icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 4645f9f192dSJonathan T. Looney offsetof(struct ip6_hdr, ip6_plen)); 4655f9f192dSJonathan T. Looney in6_ifstat_inc(dstifp, ifs6_reass_fail); 4665f9f192dSJonathan T. Looney IP6STAT_INC(ip6s_fragdropped); 467a8fe77d8SBjoern A. Zeeb *mp = NULL; 4685778b399SBjoern A. Zeeb return (IPPROTO_DONE); 4695f9f192dSJonathan T. Looney } 4705f9f192dSJonathan T. Looney 4717715d794SBjoern A. Zeeb /* 4727715d794SBjoern A. Zeeb * Enforce upper bound on number of fragments for the entire system. 4737715d794SBjoern A. Zeeb * If maxfrag is 0, never accept fragments. 4747715d794SBjoern A. Zeeb * If maxfrag is -1, accept all fragments without limitation. 4757715d794SBjoern A. Zeeb */ 4767715d794SBjoern A. Zeeb if (ip6_maxfrags < 0) 4777715d794SBjoern A. Zeeb ; 4787715d794SBjoern A. Zeeb else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags) 4797715d794SBjoern A. Zeeb goto dropfrag2; 4807715d794SBjoern A. Zeeb 48130809ba9SBjoern A. Zeeb /* 48230809ba9SBjoern A. Zeeb * Validate that a full header chain to the ULP is present in the 48330809ba9SBjoern A. Zeeb * packet containing the first fragment as per RFC RFC7112 and 48430809ba9SBjoern A. Zeeb * RFC 8200 pages 18,19: 48530809ba9SBjoern A. Zeeb * The first fragment packet is composed of: 48630809ba9SBjoern A. Zeeb * (3) Extension headers, if any, and the Upper-Layer header. These 48730809ba9SBjoern A. Zeeb * headers must be in the first fragment. ... 48830809ba9SBjoern A. Zeeb */ 48930809ba9SBjoern A. Zeeb fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); 49030809ba9SBjoern A. Zeeb /* XXX TODO. thj has D16851 open for this. */ 49130809ba9SBjoern A. Zeeb /* Send ICMPv6 4,3 in case of violation. */ 49230809ba9SBjoern A. Zeeb 493efdfee93SBjoern A. Zeeb /* Store receive network interface pointer for later. */ 494efdfee93SBjoern A. Zeeb srcifp = m->m_pkthdr.rcvif; 495efdfee93SBjoern A. Zeeb 49623d374aaSBjoern A. Zeeb /* Generate a hash value for fragment bucket selection. */ 49780d7a853SJonathan T. Looney hashkeyp = hashkey; 49880d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr)); 49980d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 50080d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr)); 50180d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 50280d7a853SJonathan T. Looney *hashkeyp = ip6f->ip6f_ident; 5039cb1a47aSBjoern A. Zeeb bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed); 5049cb1a47aSBjoern A. Zeeb bucket &= IP6REASS_HMASK; 5059cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 50621f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 5079888c401SHajimu UMEMOTO 50821f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) 50982cd038dSYoshinobu Inoue if (ip6f->ip6f_ident == q6->ip6q_ident && 51082cd038dSYoshinobu Inoue IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && 5114b908c8bSRobert Watson IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) 5124b908c8bSRobert Watson #ifdef MAC 5134b908c8bSRobert Watson && mac_ip6q_match(m, q6) 5144b908c8bSRobert Watson #endif 5154b908c8bSRobert Watson ) 51682cd038dSYoshinobu Inoue break; 51782cd038dSYoshinobu Inoue 5185778b399SBjoern A. Zeeb only_frag = false; 51921f08a07SBjoern A. Zeeb if (q6 == NULL) { 5205778b399SBjoern A. Zeeb /* A first fragment to arrive creates a reassembly queue. */ 5215778b399SBjoern A. Zeeb only_frag = true; 52282cd038dSYoshinobu Inoue 52382cd038dSYoshinobu Inoue /* 52482cd038dSYoshinobu Inoue * Enforce upper bound on number of fragmented packets 52582cd038dSYoshinobu Inoue * for which we attempt reassembly; 5269888c401SHajimu UMEMOTO * If maxfragpackets is 0, never accept fragments. 5279888c401SHajimu UMEMOTO * If maxfragpackets is -1, accept all fragments without 5289888c401SHajimu UMEMOTO * limitation. 52982cd038dSYoshinobu Inoue */ 530603724d3SBjoern A. Zeeb if (V_ip6_maxfragpackets < 0) 53133841545SHajimu UMEMOTO ; 5329cb1a47aSBjoern A. Zeeb else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize || 5331e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) >= 53480d7a853SJonathan T. Looney (u_int)V_ip6_maxfragpackets) 53533841545SHajimu UMEMOTO goto dropfrag; 53623d374aaSBjoern A. Zeeb 53723d374aaSBjoern A. Zeeb /* Allocate IPv6 fragement packet queue entry. */ 538487a161cSBjoern A. Zeeb q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6, 539487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 54082cd038dSYoshinobu Inoue if (q6 == NULL) 54182cd038dSYoshinobu Inoue goto dropfrag; 5424b908c8bSRobert Watson #ifdef MAC 5434b908c8bSRobert Watson if (mac_ip6q_init(q6, M_NOWAIT) != 0) { 544487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 5454b908c8bSRobert Watson goto dropfrag; 5464b908c8bSRobert Watson } 5474b908c8bSRobert Watson mac_ip6q_create(m, q6); 5484b908c8bSRobert Watson #endif 549702828f6SBjoern A. Zeeb atomic_add_int(&V_frag6_nfragpackets, 1); 55082cd038dSYoshinobu Inoue 55123d374aaSBjoern A. Zeeb /* ip6q_nxt will be filled afterwards, from 1st fragment. */ 55221f08a07SBjoern A. Zeeb TAILQ_INIT(&q6->ip6q_frags); 55382cd038dSYoshinobu Inoue q6->ip6q_ident = ip6f->ip6f_ident; 55482cd038dSYoshinobu Inoue q6->ip6q_ttl = IPV6_FRAGTTL; 55582cd038dSYoshinobu Inoue q6->ip6q_src = ip6->ip6_src; 55682cd038dSYoshinobu Inoue q6->ip6q_dst = ip6->ip6_dst; 557bb4a7d94SKristof Provost q6->ip6q_ecn = IPV6_ECN(ip6); 55882cd038dSYoshinobu Inoue q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ 5599888c401SHajimu UMEMOTO 56021f08a07SBjoern A. Zeeb /* Add the fragemented packet to the bucket. */ 56121f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(head, q6, ip6q_tq); 56221f08a07SBjoern A. Zeeb V_ip6qb[bucket].count++; 56382cd038dSYoshinobu Inoue } 56482cd038dSYoshinobu Inoue 56582cd038dSYoshinobu Inoue /* 56623d374aaSBjoern A. Zeeb * If it is the 1st fragment, record the length of the 56782cd038dSYoshinobu Inoue * unfragmentable part and the next header of the fragment header. 568619456bbSBjoern A. Zeeb * Assume the first 1st fragement to arrive will be correct. 569619456bbSBjoern A. Zeeb * We do not have any duplicate checks here yet so another packet 570619456bbSBjoern A. Zeeb * with fragoff == 0 could come and overwrite the ip6q_unfrglen 571619456bbSBjoern A. Zeeb * and worse, the next header, at any time. 57282cd038dSYoshinobu Inoue */ 573619456bbSBjoern A. Zeeb if (fragoff == 0 && q6->ip6q_unfrglen == -1) { 57406cd0a3fSHajimu UMEMOTO q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - 57506cd0a3fSHajimu UMEMOTO sizeof(struct ip6_frag); 57682cd038dSYoshinobu Inoue q6->ip6q_nxt = ip6f->ip6f_nxt; 577619456bbSBjoern A. Zeeb /* XXX ECN? */ 57882cd038dSYoshinobu Inoue } 57982cd038dSYoshinobu Inoue 58082cd038dSYoshinobu Inoue /* 58182cd038dSYoshinobu Inoue * Check that the reassembled packet would not exceed 65535 bytes 58282cd038dSYoshinobu Inoue * in size. 58382cd038dSYoshinobu Inoue * If it would exceed, discard the fragment and return an ICMP error. 58482cd038dSYoshinobu Inoue */ 58582cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen >= 0) { 58682cd038dSYoshinobu Inoue /* The 1st fragment has already arrived. */ 58782cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { 588e5fffe9aSBjoern A. Zeeb if (only_frag) { 589e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 590e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 591e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 592e5fffe9aSBjoern A. Zeeb #ifdef MAC 593e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 594e5fffe9aSBjoern A. Zeeb #endif 595e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 596e5fffe9aSBjoern A. Zeeb } 597e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 59882cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 599686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 600686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 601a8fe77d8SBjoern A. Zeeb *mp = NULL; 60282cd038dSYoshinobu Inoue return (IPPROTO_DONE); 60382cd038dSYoshinobu Inoue } 60406cd0a3fSHajimu UMEMOTO } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { 605e5fffe9aSBjoern A. Zeeb if (only_frag) { 606e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 607e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 608e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 609e5fffe9aSBjoern A. Zeeb #ifdef MAC 610e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 611e5fffe9aSBjoern A. Zeeb #endif 612e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 613e5fffe9aSBjoern A. Zeeb } 614e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 61582cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 616686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 617686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 618a8fe77d8SBjoern A. Zeeb *mp = NULL; 61982cd038dSYoshinobu Inoue return (IPPROTO_DONE); 62082cd038dSYoshinobu Inoue } 621f1664f32SBjoern A. Zeeb 62282cd038dSYoshinobu Inoue /* 62323d374aaSBjoern A. Zeeb * If it is the first fragment, do the above check for each 62482cd038dSYoshinobu Inoue * fragment already stored in the reassembly queue. 62582cd038dSYoshinobu Inoue */ 626dda02192SBjoern A. Zeeb if (fragoff == 0 && !only_frag) { 62721f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(af6, &q6->ip6q_frags, ip6af_tq, af6tmp) { 628dda02192SBjoern A. Zeeb if (q6->ip6q_unfrglen + af6->ip6af_off + 629dda02192SBjoern A. Zeeb af6->ip6af_frglen > IPV6_MAXPACKET) { 63082cd038dSYoshinobu Inoue struct ip6_hdr *ip6err; 6315778b399SBjoern A. Zeeb struct mbuf *merr; 6325778b399SBjoern A. Zeeb int erroff; 6335778b399SBjoern A. Zeeb 634da89a0feSBjoern A. Zeeb merr = af6->ip6af_m; 6355778b399SBjoern A. Zeeb erroff = af6->ip6af_offset; 63682cd038dSYoshinobu Inoue 63723d374aaSBjoern A. Zeeb /* Dequeue the fragment. */ 63821f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 639dda02192SBjoern A. Zeeb q6->ip6q_nfrag--; 640dda02192SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, 1); 641487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 64282cd038dSYoshinobu Inoue 643a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 644a55383e7SHans Petter Selasky merr->m_pkthdr.rcvif = srcifp; 645a55383e7SHans Petter Selasky 64623d374aaSBjoern A. Zeeb /* Adjust pointer. */ 64782cd038dSYoshinobu Inoue ip6err = mtod(merr, struct ip6_hdr *); 64882cd038dSYoshinobu Inoue 64982cd038dSYoshinobu Inoue /* 65082cd038dSYoshinobu Inoue * Restore source and destination addresses 65182cd038dSYoshinobu Inoue * in the erroneous IPv6 header. 65282cd038dSYoshinobu Inoue */ 65382cd038dSYoshinobu Inoue ip6err->ip6_src = q6->ip6q_src; 65482cd038dSYoshinobu Inoue ip6err->ip6_dst = q6->ip6q_dst; 65582cd038dSYoshinobu Inoue 65682cd038dSYoshinobu Inoue icmp6_error(merr, ICMP6_PARAM_PROB, 65782cd038dSYoshinobu Inoue ICMP6_PARAMPROB_HEADER, 658686cdd19SJun-ichiro itojun Hagino erroff - sizeof(struct ip6_frag) + 659686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 66082cd038dSYoshinobu Inoue } 66182cd038dSYoshinobu Inoue } 66282cd038dSYoshinobu Inoue } 66382cd038dSYoshinobu Inoue 66423d374aaSBjoern A. Zeeb /* Allocate an IPv6 fragement queue entry for this fragmented part. */ 665487a161cSBjoern A. Zeeb ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6, 666487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 667686cdd19SJun-ichiro itojun Hagino if (ip6af == NULL) 668686cdd19SJun-ichiro itojun Hagino goto dropfrag; 66921f08a07SBjoern A. Zeeb ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) ? true : false; 67082cd038dSYoshinobu Inoue ip6af->ip6af_off = fragoff; 67182cd038dSYoshinobu Inoue ip6af->ip6af_frglen = frgpartlen; 67282cd038dSYoshinobu Inoue ip6af->ip6af_offset = offset; 673da89a0feSBjoern A. Zeeb ip6af->ip6af_m = m; 67482cd038dSYoshinobu Inoue 6755778b399SBjoern A. Zeeb if (only_frag) { 67621f08a07SBjoern A. Zeeb /* 67721f08a07SBjoern A. Zeeb * Do a manual insert rather than a hard-to-understand cast 67821f08a07SBjoern A. Zeeb * to a different type relying on data structure order to work. 67921f08a07SBjoern A. Zeeb */ 68021f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(&q6->ip6q_frags, ip6af, ip6af_tq); 68121f08a07SBjoern A. Zeeb goto postinsert; 68282cd038dSYoshinobu Inoue } 68382cd038dSYoshinobu Inoue 68423d374aaSBjoern A. Zeeb /* Do duplicate, condition, and boundry checks. */ 68582cd038dSYoshinobu Inoue /* 68659dfcba4SHajimu UMEMOTO * Handle ECN by comparing this segment with the first one; 68759dfcba4SHajimu UMEMOTO * if CE is set, do not lose CE. 68823d374aaSBjoern A. Zeeb * Drop if CE and not-ECT are mixed for the same packet. 68959dfcba4SHajimu UMEMOTO */ 690bb4a7d94SKristof Provost ecn = IPV6_ECN(ip6); 6915e9510e3SJINMEI Tatuya ecn0 = q6->ip6q_ecn; 69259dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_CE) { 69359dfcba4SHajimu UMEMOTO if (ecn0 == IPTOS_ECN_NOTECT) { 694487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 69559dfcba4SHajimu UMEMOTO goto dropfrag; 69659dfcba4SHajimu UMEMOTO } 69759dfcba4SHajimu UMEMOTO if (ecn0 != IPTOS_ECN_CE) 6985e9510e3SJINMEI Tatuya q6->ip6q_ecn = IPTOS_ECN_CE; 69959dfcba4SHajimu UMEMOTO } 70059dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { 701487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 70259dfcba4SHajimu UMEMOTO goto dropfrag; 70359dfcba4SHajimu UMEMOTO } 70459dfcba4SHajimu UMEMOTO 70523d374aaSBjoern A. Zeeb /* Find a fragmented part which begins after this one does. */ 70621f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) 70782cd038dSYoshinobu Inoue if (af6->ip6af_off > ip6af->ip6af_off) 70882cd038dSYoshinobu Inoue break; 70982cd038dSYoshinobu Inoue 71082cd038dSYoshinobu Inoue /* 71182cd038dSYoshinobu Inoue * If the incoming framgent overlaps some existing fragments in 71223d374aaSBjoern A. Zeeb * the reassembly queue, drop both the new fragment and the 71323d374aaSBjoern A. Zeeb * entire reassembly queue. However, if the new fragment 71423d374aaSBjoern A. Zeeb * is an exact duplicate of an existing fragment, only silently 71523d374aaSBjoern A. Zeeb * drop the existing fragment and leave the fragmentation queue 71623d374aaSBjoern A. Zeeb * unchanged, as allowed by the RFC. (RFC 8200, 4.5) 71782cd038dSYoshinobu Inoue */ 71821f08a07SBjoern A. Zeeb if (af6 != NULL) 71921f08a07SBjoern A. Zeeb af6tmp = TAILQ_PREV(af6, ip6fraghead, ip6af_tq); 72021f08a07SBjoern A. Zeeb else 72121f08a07SBjoern A. Zeeb af6tmp = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 72221f08a07SBjoern A. Zeeb if (af6tmp != NULL) { 72321f08a07SBjoern A. Zeeb if (af6tmp->ip6af_off + af6tmp->ip6af_frglen - 7245778b399SBjoern A. Zeeb ip6af->ip6af_off > 0) { 725cd188da2SBjoern A. Zeeb if (af6tmp->ip6af_off != ip6af->ip6af_off || 726cd188da2SBjoern A. Zeeb af6tmp->ip6af_frglen != ip6af->ip6af_frglen) 727cd188da2SBjoern A. Zeeb frag6_freef(q6, bucket); 728487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 72982cd038dSYoshinobu Inoue goto dropfrag; 73082cd038dSYoshinobu Inoue } 73182cd038dSYoshinobu Inoue } 73221f08a07SBjoern A. Zeeb if (af6 != NULL) { 7335778b399SBjoern A. Zeeb if (ip6af->ip6af_off + ip6af->ip6af_frglen - 7345778b399SBjoern A. Zeeb af6->ip6af_off > 0) { 735cd188da2SBjoern A. Zeeb if (af6->ip6af_off != ip6af->ip6af_off || 736cd188da2SBjoern A. Zeeb af6->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 } 74282cd038dSYoshinobu Inoue 7434b908c8bSRobert Watson #ifdef MAC 7444b908c8bSRobert Watson mac_ip6q_update(m, q6); 7454b908c8bSRobert Watson #endif 74682cd038dSYoshinobu Inoue 74782cd038dSYoshinobu Inoue /* 74823d374aaSBjoern A. Zeeb * Stick new segment in its place; check for complete reassembly. 74923d374aaSBjoern A. Zeeb * If not complete, check fragment limit. Move to front of packet 75023d374aaSBjoern A. Zeeb * queue, as we are the most recently active fragmented packet. 75182cd038dSYoshinobu Inoue */ 75221f08a07SBjoern A. Zeeb if (af6 != NULL) 75321f08a07SBjoern A. Zeeb TAILQ_INSERT_BEFORE(af6, ip6af, ip6af_tq); 75421f08a07SBjoern A. Zeeb else 75521f08a07SBjoern A. Zeeb TAILQ_INSERT_TAIL(&q6->ip6q_frags, ip6af, ip6af_tq); 75621f08a07SBjoern A. Zeeb postinsert: 7572adfd64fSJonathan T. Looney atomic_add_int(&frag6_nfrags, 1); 7589888c401SHajimu UMEMOTO q6->ip6q_nfrag++; 7593c7165b3SBjoern A. Zeeb 7605778b399SBjoern A. Zeeb plen = 0; 76121f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 7625778b399SBjoern A. Zeeb if (af6->ip6af_off != plen) { 76303c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 764198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7659cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 76603c99d76SJonathan T. Looney } 7679cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 768a8fe77d8SBjoern A. Zeeb *mp = NULL; 7695778b399SBjoern A. Zeeb return (IPPROTO_DONE); 77082cd038dSYoshinobu Inoue } 7715778b399SBjoern A. Zeeb plen += af6->ip6af_frglen; 77282cd038dSYoshinobu Inoue } 77321f08a07SBjoern A. Zeeb af6 = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 77421f08a07SBjoern A. Zeeb if (af6->ip6af_mff) { 77503c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 776198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7779cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 77803c99d76SJonathan T. Looney } 7799cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 780a8fe77d8SBjoern A. Zeeb *mp = NULL; 7815778b399SBjoern A. Zeeb return (IPPROTO_DONE); 78282cd038dSYoshinobu Inoue } 78382cd038dSYoshinobu Inoue 78423d374aaSBjoern A. Zeeb /* Reassembly is complete; concatenate fragments. */ 78521f08a07SBjoern A. Zeeb ip6af = TAILQ_FIRST(&q6->ip6q_frags); 786da89a0feSBjoern A. Zeeb t = m = ip6af->ip6af_m; 78721f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, ip6af, ip6af_tq); 78821f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 7899907aba3SAndrey V. Elsukov m->m_pkthdr.csum_flags &= 790da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_flags; 7919907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data += 792da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_data; 7939907aba3SAndrey V. Elsukov 79421f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 795efdfee93SBjoern A. Zeeb t = m_last(t); 796da89a0feSBjoern A. Zeeb m_adj(af6->ip6af_m, af6->ip6af_offset); 797da89a0feSBjoern A. Zeeb m_demote_pkthdr(af6->ip6af_m); 798da89a0feSBjoern A. Zeeb m_cat(t, af6->ip6af_m); 799487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 80082cd038dSYoshinobu Inoue } 80182cd038dSYoshinobu Inoue 8029907aba3SAndrey V. Elsukov while (m->m_pkthdr.csum_data & 0xffff0000) 8039907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 8049907aba3SAndrey V. Elsukov (m->m_pkthdr.csum_data >> 16); 8059907aba3SAndrey V. Elsukov 80623d374aaSBjoern A. Zeeb /* Adjust offset to point where the original next header starts. */ 80782cd038dSYoshinobu Inoue offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); 808487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 809686cdd19SJun-ichiro itojun Hagino ip6 = mtod(m, struct ip6_hdr *); 8105778b399SBjoern A. Zeeb ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr)); 8115e9510e3SJINMEI Tatuya if (q6->ip6q_ecn == IPTOS_ECN_CE) 8125e9510e3SJINMEI Tatuya ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); 81382cd038dSYoshinobu Inoue nxt = q6->ip6q_nxt; 81482cd038dSYoshinobu Inoue 81521f08a07SBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 81621f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 8172adfd64fSJonathan T. Looney atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 81821f08a07SBjoern A. Zeeb 819a61b5cfbSBjoern A. Zeeb ip6_deletefraghdr(m, offset, M_NOWAIT); 82082cd038dSYoshinobu Inoue 82123d374aaSBjoern A. Zeeb /* Set nxt(-hdr field value) to the original value. */ 82268e0e5a6SAndrey V. Elsukov m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 82368e0e5a6SAndrey V. Elsukov (caddr_t)&nxt); 82482cd038dSYoshinobu Inoue 8254b908c8bSRobert Watson #ifdef MAC 8264b908c8bSRobert Watson mac_ip6q_reassemble(q6, m); 8274b908c8bSRobert Watson mac_ip6q_destroy(q6); 8284b908c8bSRobert Watson #endif 829487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 83080d7a853SJonathan T. Looney atomic_subtract_int(&V_frag6_nfragpackets, 1); 83182cd038dSYoshinobu Inoue 83282cd038dSYoshinobu Inoue if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ 8335778b399SBjoern A. Zeeb 8345778b399SBjoern A. Zeeb plen = 0; 83582cd038dSYoshinobu Inoue for (t = m; t; t = t->m_next) 83682cd038dSYoshinobu Inoue plen += t->m_len; 83782cd038dSYoshinobu Inoue m->m_pkthdr.len = plen; 838a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 839a55383e7SHans Petter Selasky m->m_pkthdr.rcvif = srcifp; 84082cd038dSYoshinobu Inoue } 84182cd038dSYoshinobu Inoue 842aaa46574SAdrian Chadd #ifdef RSS 843aaa46574SAdrian Chadd mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), 844aaa46574SAdrian Chadd M_NOWAIT); 845aaa46574SAdrian Chadd if (mtag == NULL) 846aaa46574SAdrian Chadd goto dropfrag; 847aaa46574SAdrian Chadd 848aaa46574SAdrian Chadd ip6dc = (struct ip6_direct_ctx *)(mtag + 1); 849aaa46574SAdrian Chadd ip6dc->ip6dc_nxt = nxt; 850aaa46574SAdrian Chadd ip6dc->ip6dc_off = offset; 851aaa46574SAdrian Chadd 852aaa46574SAdrian Chadd m_tag_prepend(m, mtag); 853aaa46574SAdrian Chadd #endif 854aaa46574SAdrian Chadd 8559cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 8569cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_reassembled); 85782cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_ok); 85882cd038dSYoshinobu Inoue 859aaa46574SAdrian Chadd #ifdef RSS 86023d374aaSBjoern A. Zeeb /* Queue/dispatch for reprocessing. */ 861aaa46574SAdrian Chadd netisr_dispatch(NETISR_IPV6_DIRECT, m); 862a8fe77d8SBjoern A. Zeeb *mp = NULL; 8635778b399SBjoern A. Zeeb return (IPPROTO_DONE); 864aaa46574SAdrian Chadd #endif 865aaa46574SAdrian Chadd 86623d374aaSBjoern A. Zeeb /* Tell launch routine the next header. */ 86782cd038dSYoshinobu Inoue *mp = m; 86882cd038dSYoshinobu Inoue *offp = offset; 86982cd038dSYoshinobu Inoue 8705778b399SBjoern A. Zeeb return (nxt); 87182cd038dSYoshinobu Inoue 87282cd038dSYoshinobu Inoue dropfrag: 8739cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 8747715d794SBjoern A. Zeeb dropfrag2: 87582cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 8769cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragdropped); 87782cd038dSYoshinobu Inoue m_freem(m); 878a8fe77d8SBjoern A. Zeeb *mp = NULL; 8795778b399SBjoern A. Zeeb return (IPPROTO_DONE); 88082cd038dSYoshinobu Inoue } 88182cd038dSYoshinobu Inoue 88282cd038dSYoshinobu Inoue /* 88333841545SHajimu UMEMOTO * IPv6 reassembling timer processing; 88423d374aaSBjoern A. Zeeb * if a timer expires on a reassembly queue, discard it. 88582cd038dSYoshinobu Inoue */ 88682cd038dSYoshinobu Inoue void 8871272577eSXin LI frag6_slowtimo(void) 88882cd038dSYoshinobu Inoue { 8898b615593SMarko Zec VNET_ITERATOR_DECL(vnet_iter); 89021f08a07SBjoern A. Zeeb struct ip6qhead *head; 89121f08a07SBjoern A. Zeeb struct ip6q *q6, *q6tmp; 8929cb1a47aSBjoern A. Zeeb uint32_t bucket; 89382cd038dSYoshinobu Inoue 8945ee847d3SRobert Watson VNET_LIST_RLOCK_NOSLEEP(); 8958b615593SMarko Zec VNET_FOREACH(vnet_iter) { 8968b615593SMarko Zec CURVNET_SET(vnet_iter); 8979cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 8989cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 8999cb1a47aSBjoern A. Zeeb head = IP6QB_HEAD(bucket); 90021f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(q6, head, ip6q_tq, q6tmp) 90121f08a07SBjoern A. Zeeb if (--q6->ip6q_ttl == 0) { 902198fdaedSTom Jones IP6STAT_ADD(ip6s_fragtimeout, 90321f08a07SBjoern A. Zeeb q6->ip6q_nfrag); 90482cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 90521f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 90682cd038dSYoshinobu Inoue } 90782cd038dSYoshinobu Inoue /* 90882cd038dSYoshinobu Inoue * If we are over the maximum number of fragments 90982cd038dSYoshinobu Inoue * (due to the limit being lowered), drain off 91082cd038dSYoshinobu Inoue * enough to get down to the new limit. 9111e9f3b73SJonathan T. Looney * Note that we drain all reassembly queues if 9121e9f3b73SJonathan T. Looney * maxfragpackets is 0 (fragmentation is disabled), 91323d374aaSBjoern A. Zeeb * and do not enforce a limit when maxfragpackets 9141e9f3b73SJonathan T. Looney * is negative. 91582cd038dSYoshinobu Inoue */ 9161e9f3b73SJonathan T. Looney while ((V_ip6_maxfragpackets == 0 || 9171e9f3b73SJonathan T. Looney (V_ip6_maxfragpackets > 0 && 9189cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) && 91921f08a07SBjoern A. Zeeb (q6 = TAILQ_LAST(head, ip6qhead)) != NULL) { 92021f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 92182cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 92221f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 92380d7a853SJonathan T. Looney } 9249cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 92582cd038dSYoshinobu Inoue } 9261e9f3b73SJonathan T. Looney /* 9271e9f3b73SJonathan T. Looney * If we are still over the maximum number of fragmented 9281e9f3b73SJonathan T. Looney * packets, drain off enough to get down to the new limit. 9291e9f3b73SJonathan T. Looney */ 9309cb1a47aSBjoern A. Zeeb bucket = 0; 9311e9f3b73SJonathan T. Looney while (V_ip6_maxfragpackets >= 0 && 9321e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) > 9331e9f3b73SJonathan T. Looney (u_int)V_ip6_maxfragpackets) { 9349cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 93521f08a07SBjoern A. Zeeb q6 = TAILQ_LAST(IP6QB_HEAD(bucket), ip6qhead); 93621f08a07SBjoern A. Zeeb if (q6 != NULL) { 93721f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 9381e9f3b73SJonathan T. Looney /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 93921f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 9401e9f3b73SJonathan T. Looney } 9419cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 9429cb1a47aSBjoern A. Zeeb bucket = (bucket + 1) % IP6REASS_NHASH; 9431e9f3b73SJonathan T. Looney } 9448b615593SMarko Zec CURVNET_RESTORE(); 9458b615593SMarko Zec } 9465ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 94782cd038dSYoshinobu Inoue } 94882cd038dSYoshinobu Inoue 94923d374aaSBjoern A. Zeeb /* 95023d374aaSBjoern A. Zeeb * Eventhandler to adjust limits in case nmbclusters change. 95123d374aaSBjoern A. Zeeb */ 952c00464a2SBjoern A. Zeeb static void 953c00464a2SBjoern A. Zeeb frag6_change(void *tag) 954c00464a2SBjoern A. Zeeb { 955c00464a2SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 956c00464a2SBjoern A. Zeeb 957c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 958c00464a2SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 959c00464a2SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 960c00464a2SBjoern A. Zeeb CURVNET_SET(vnet_iter); 961c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 962c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 963c00464a2SBjoern A. Zeeb CURVNET_RESTORE(); 964c00464a2SBjoern A. Zeeb } 965c00464a2SBjoern A. Zeeb VNET_LIST_RUNLOCK_NOSLEEP(); 966c00464a2SBjoern A. Zeeb } 967c00464a2SBjoern A. Zeeb 968c00464a2SBjoern A. Zeeb /* 969c00464a2SBjoern A. Zeeb * Initialise reassembly queue and fragment identifier. 970c00464a2SBjoern A. Zeeb */ 971c00464a2SBjoern A. Zeeb void 972c00464a2SBjoern A. Zeeb frag6_init(void) 973c00464a2SBjoern A. Zeeb { 9749cb1a47aSBjoern A. Zeeb uint32_t bucket; 975c00464a2SBjoern A. Zeeb 976c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 977c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 9789cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 97921f08a07SBjoern A. Zeeb TAILQ_INIT(IP6QB_HEAD(bucket)); 980efdfee93SBjoern A. Zeeb mtx_init(&V_ip6qb[bucket].lock, "ip6qb", NULL, MTX_DEF); 9819cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count = 0; 982c00464a2SBjoern A. Zeeb } 9839cb1a47aSBjoern A. Zeeb V_ip6qb_hashseed = arc4random(); 984c00464a2SBjoern A. Zeeb V_ip6_maxfragsperpacket = 64; 98567a10c46SBjoern A. Zeeb #ifdef VIMAGE 98667a10c46SBjoern A. Zeeb V_frag6_on = true; 98767a10c46SBjoern A. Zeeb #endif 988c00464a2SBjoern A. Zeeb if (!IS_DEFAULT_VNET(curvnet)) 989c00464a2SBjoern A. Zeeb return; 990c00464a2SBjoern A. Zeeb 991c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 992c00464a2SBjoern A. Zeeb EVENTHANDLER_REGISTER(nmbclusters_change, 993c00464a2SBjoern A. Zeeb frag6_change, NULL, EVENTHANDLER_PRI_ANY); 994c00464a2SBjoern A. Zeeb } 995c00464a2SBjoern A. Zeeb 99682cd038dSYoshinobu Inoue /* 99782cd038dSYoshinobu Inoue * Drain off all datagram fragments. 99882cd038dSYoshinobu Inoue */ 99967a10c46SBjoern A. Zeeb static void 100067a10c46SBjoern A. Zeeb frag6_drain_one(void) 100182cd038dSYoshinobu Inoue { 100221f08a07SBjoern A. Zeeb struct ip6q *q6; 10039cb1a47aSBjoern A. Zeeb uint32_t bucket; 10049888c401SHajimu UMEMOTO 10059cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 100667a10c46SBjoern A. Zeeb IP6QB_LOCK(bucket); 100721f08a07SBjoern A. Zeeb while ((q6 = TAILQ_FIRST(IP6QB_HEAD(bucket))) != NULL) { 10089cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragdropped); 100982cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 101021f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 101180d7a853SJonathan T. Looney } 10129cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 101382cd038dSYoshinobu Inoue } 101467a10c46SBjoern A. Zeeb } 101567a10c46SBjoern A. Zeeb 101667a10c46SBjoern A. Zeeb void 101767a10c46SBjoern A. Zeeb frag6_drain(void) 101867a10c46SBjoern A. Zeeb { 101967a10c46SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 102067a10c46SBjoern A. Zeeb 102167a10c46SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 102267a10c46SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 102367a10c46SBjoern A. Zeeb CURVNET_SET(vnet_iter); 102467a10c46SBjoern A. Zeeb frag6_drain_one(); 10258b615593SMarko Zec CURVNET_RESTORE(); 10268b615593SMarko Zec } 10275ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 102882cd038dSYoshinobu Inoue } 1029e5ee7060SGleb Smirnoff 103067a10c46SBjoern A. Zeeb #ifdef VIMAGE 103167a10c46SBjoern A. Zeeb /* 103267a10c46SBjoern A. Zeeb * Clear up IPv6 reassembly structures. 103367a10c46SBjoern A. Zeeb */ 103467a10c46SBjoern A. Zeeb void 103567a10c46SBjoern A. Zeeb frag6_destroy(void) 103667a10c46SBjoern A. Zeeb { 103767a10c46SBjoern A. Zeeb uint32_t bucket; 103867a10c46SBjoern A. Zeeb 103967a10c46SBjoern A. Zeeb frag6_drain_one(); 104067a10c46SBjoern A. Zeeb V_frag6_on = false; 104167a10c46SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 104267a10c46SBjoern A. Zeeb KASSERT(V_ip6qb[bucket].count == 0, 104367a10c46SBjoern A. Zeeb ("%s: V_ip6qb[%d] (%p) count not 0 (%d)", __func__, 104467a10c46SBjoern A. Zeeb bucket, &V_ip6qb[bucket], V_ip6qb[bucket].count)); 104567a10c46SBjoern A. Zeeb mtx_destroy(&V_ip6qb[bucket].lock); 104667a10c46SBjoern A. Zeeb } 104767a10c46SBjoern A. Zeeb } 104867a10c46SBjoern A. Zeeb #endif 1049