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 221c00464a2SBjoern A. Zeeb ip6_deletefraghdr(struct mbuf *m, int offset, int wait) 222c00464a2SBjoern A. Zeeb { 2235778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 224c00464a2SBjoern A. Zeeb struct mbuf *t; 225c00464a2SBjoern A. Zeeb 226c00464a2SBjoern A. Zeeb /* Delete frag6 header. */ 227c00464a2SBjoern A. Zeeb if (m->m_len >= offset + sizeof(struct ip6_frag)) { 2285778b399SBjoern A. Zeeb 229c00464a2SBjoern A. Zeeb /* This is the only possible case with !PULLDOWN_TEST. */ 2305778b399SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 231c00464a2SBjoern A. Zeeb bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), 232c00464a2SBjoern A. Zeeb offset); 233c00464a2SBjoern A. Zeeb m->m_data += sizeof(struct ip6_frag); 234c00464a2SBjoern A. Zeeb m->m_len -= sizeof(struct ip6_frag); 235c00464a2SBjoern A. Zeeb } else { 2365778b399SBjoern A. Zeeb 237c00464a2SBjoern A. Zeeb /* This comes with no copy if the boundary is on cluster. */ 238c00464a2SBjoern A. Zeeb if ((t = m_split(m, offset, wait)) == NULL) 239c00464a2SBjoern A. Zeeb return (ENOMEM); 240c00464a2SBjoern A. Zeeb m_adj(t, sizeof(struct ip6_frag)); 241c00464a2SBjoern A. Zeeb m_cat(m, t); 242c00464a2SBjoern A. Zeeb } 243c00464a2SBjoern A. Zeeb 244c00464a2SBjoern A. Zeeb m->m_flags |= M_FRAGMENTED; 245c00464a2SBjoern A. Zeeb return (0); 246c00464a2SBjoern A. Zeeb } 247c00464a2SBjoern A. Zeeb 248c00464a2SBjoern A. Zeeb /* 24923d374aaSBjoern A. Zeeb * Free a fragment reassembly header and all associated datagrams. 250757cb678SBjoern A. Zeeb */ 2514f590175SPaul Saab static void 252c00464a2SBjoern A. Zeeb frag6_freef(struct ip6q *q6, uint32_t bucket) 2534f590175SPaul Saab { 2545778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 25521f08a07SBjoern A. Zeeb struct ip6asfrag *af6; 2565778b399SBjoern A. Zeeb struct mbuf *m; 2574f590175SPaul Saab 2589cb1a47aSBjoern A. Zeeb IP6QB_LOCK_ASSERT(bucket); 259c00464a2SBjoern A. Zeeb 26021f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 261c00464a2SBjoern A. Zeeb 262da89a0feSBjoern A. Zeeb m = af6->ip6af_m; 26321f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 264c00464a2SBjoern A. Zeeb 265c00464a2SBjoern A. Zeeb /* 266c00464a2SBjoern A. Zeeb * Return ICMP time exceeded error for the 1st fragment. 267c00464a2SBjoern A. Zeeb * Just free other fragments. 268c00464a2SBjoern A. Zeeb */ 269a55383e7SHans Petter Selasky if (af6->ip6af_off == 0 && m->m_pkthdr.rcvif != NULL) { 270c00464a2SBjoern A. Zeeb 27123d374aaSBjoern A. Zeeb /* Adjust pointer. */ 272c00464a2SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 273c00464a2SBjoern A. Zeeb 27423d374aaSBjoern A. Zeeb /* Restore source and destination addresses. */ 275c00464a2SBjoern A. Zeeb ip6->ip6_src = q6->ip6q_src; 276c00464a2SBjoern A. Zeeb ip6->ip6_dst = q6->ip6q_dst; 277c00464a2SBjoern A. Zeeb 278c00464a2SBjoern A. Zeeb icmp6_error(m, ICMP6_TIME_EXCEEDED, 279c00464a2SBjoern A. Zeeb ICMP6_TIME_EXCEED_REASSEMBLY, 0); 280c00464a2SBjoern A. Zeeb } else 281c00464a2SBjoern A. Zeeb m_freem(m); 28223d374aaSBjoern A. Zeeb 283c00464a2SBjoern A. Zeeb free(af6, M_FRAG6); 2842adfd64fSJonathan T. Looney } 28521f08a07SBjoern A. Zeeb 28621f08a07SBjoern A. Zeeb TAILQ_REMOVE(IP6QB_HEAD(bucket), q6, ip6q_tq); 28721f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 288c00464a2SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 289c00464a2SBjoern A. Zeeb #ifdef MAC 290c00464a2SBjoern A. Zeeb mac_ip6q_destroy(q6); 291c00464a2SBjoern A. Zeeb #endif 292c00464a2SBjoern A. Zeeb free(q6, M_FRAG6); 293c00464a2SBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 29482cd038dSYoshinobu Inoue } 29582cd038dSYoshinobu Inoue 29682cd038dSYoshinobu Inoue /* 297a55383e7SHans Petter Selasky * Drain off all datagram fragments belonging to 298a55383e7SHans Petter Selasky * the given network interface. 299a55383e7SHans Petter Selasky */ 300a55383e7SHans Petter Selasky static void 301a55383e7SHans Petter Selasky frag6_cleanup(void *arg __unused, struct ifnet *ifp) 302a55383e7SHans Petter Selasky { 30321f08a07SBjoern A. Zeeb struct ip6qhead *head; 30421f08a07SBjoern A. Zeeb struct ip6q *q6; 305a55383e7SHans Petter Selasky struct ip6asfrag *af6; 30621f08a07SBjoern A. Zeeb uint32_t bucket; 307a55383e7SHans Petter Selasky 308a55383e7SHans Petter Selasky KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); 309a55383e7SHans Petter Selasky 31067a10c46SBjoern A. Zeeb #ifdef VIMAGE 31167a10c46SBjoern A. Zeeb /* 31267a10c46SBjoern A. Zeeb * Skip processing if IPv6 reassembly is not initialised or 31367a10c46SBjoern A. Zeeb * torn down by frag6_destroy(). 31467a10c46SBjoern A. Zeeb */ 31567a10c46SBjoern A. Zeeb if (!V_frag6_on) 31667a10c46SBjoern A. Zeeb return; 31767a10c46SBjoern A. Zeeb #endif 31867a10c46SBjoern A. Zeeb 319a55383e7SHans Petter Selasky CURVNET_SET_QUIET(ifp->if_vnet); 32021f08a07SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 32121f08a07SBjoern A. Zeeb IP6QB_LOCK(bucket); 32221f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 323a55383e7SHans Petter Selasky /* Scan fragment list. */ 32421f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) { 32521f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 326a55383e7SHans Petter Selasky 327f1664f32SBjoern A. Zeeb /* Clear no longer valid rcvif pointer. */ 328da89a0feSBjoern A. Zeeb if (af6->ip6af_m->m_pkthdr.rcvif == ifp) 329da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.rcvif = NULL; 330a55383e7SHans Petter Selasky } 331a55383e7SHans Petter Selasky } 33221f08a07SBjoern A. Zeeb IP6QB_UNLOCK(bucket); 333a55383e7SHans Petter Selasky } 334a55383e7SHans Petter Selasky CURVNET_RESTORE(); 335a55383e7SHans Petter Selasky } 336a55383e7SHans Petter Selasky EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0); 337a55383e7SHans Petter Selasky 338a55383e7SHans Petter Selasky /* 33923d374aaSBjoern A. Zeeb * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with 34023d374aaSBjoern A. Zeeb * each other, in terms of next header field handling in fragment header. 341686cdd19SJun-ichiro itojun Hagino * While the sender will use the same value for all of the fragmented packets, 34223d374aaSBjoern A. Zeeb * receiver is suggested not to check for consistency. 343686cdd19SJun-ichiro itojun Hagino * 34423d374aaSBjoern A. Zeeb * Fragment rules (p18,p19): 345686cdd19SJun-ichiro itojun Hagino * (2) A Fragment header containing: 34623d374aaSBjoern A. Zeeb * The Next Header value that identifies the first header 34723d374aaSBjoern A. Zeeb * after the Per-Fragment headers of the original packet. 348686cdd19SJun-ichiro itojun Hagino * -> next header field is same for all fragments 349686cdd19SJun-ichiro itojun Hagino * 35023d374aaSBjoern A. Zeeb * Reassembly rule (p20): 35123d374aaSBjoern A. Zeeb * The Next Header field of the last header of the Per-Fragment 35223d374aaSBjoern A. Zeeb * headers is obtained from the Next Header field of the first 353686cdd19SJun-ichiro itojun Hagino * fragment's Fragment header. 354686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 355686cdd19SJun-ichiro itojun Hagino * 356686cdd19SJun-ichiro itojun Hagino * The following note also contradicts with fragment rule - no one is going to 357686cdd19SJun-ichiro itojun Hagino * send different fragment with different next header field. 358686cdd19SJun-ichiro itojun Hagino * 35923d374aaSBjoern A. Zeeb * Additional note (p22) [not an error]: 360686cdd19SJun-ichiro itojun Hagino * The Next Header values in the Fragment headers of different 361686cdd19SJun-ichiro itojun Hagino * fragments of the same original packet may differ. Only the value 362686cdd19SJun-ichiro itojun Hagino * from the Offset zero fragment packet is used for reassembly. 363686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 364686cdd19SJun-ichiro itojun Hagino * 365686cdd19SJun-ichiro itojun Hagino * There is no explicit reason given in the RFC. Historical reason maybe? 366686cdd19SJun-ichiro itojun Hagino */ 367686cdd19SJun-ichiro itojun Hagino /* 36823d374aaSBjoern A. Zeeb * Fragment input. 36982cd038dSYoshinobu Inoue */ 37082cd038dSYoshinobu Inoue int 3711272577eSXin LI frag6_input(struct mbuf **mp, int *offp, int proto) 37282cd038dSYoshinobu Inoue { 37321f08a07SBjoern A. Zeeb struct mbuf *m, *t; 37482cd038dSYoshinobu Inoue struct ip6_hdr *ip6; 37582cd038dSYoshinobu Inoue struct ip6_frag *ip6f; 37621f08a07SBjoern A. Zeeb struct ip6qhead *head; 37721f08a07SBjoern A. Zeeb struct ip6q *q6; 37821f08a07SBjoern A. Zeeb struct ip6asfrag *af6, *ip6af, *af6tmp; 37921f08a07SBjoern A. Zeeb struct in6_ifaddr *ia6; 38021f08a07SBjoern A. Zeeb struct ifnet *dstifp, *srcifp; 381505e91f5SKristof Provost uint32_t hashkey[(sizeof(struct in6_addr) * 2 + 382505e91f5SKristof Provost sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)]; 3839cb1a47aSBjoern A. Zeeb uint32_t bucket, *hashkeyp; 3845778b399SBjoern A. Zeeb int fragoff, frgpartlen; /* Must be larger than uint16_t. */ 3855778b399SBjoern A. Zeeb int nxt, offset, plen; 3865778b399SBjoern A. Zeeb uint8_t ecn, ecn0; 3875778b399SBjoern A. Zeeb bool only_frag; 388aaa46574SAdrian Chadd #ifdef RSS 389aaa46574SAdrian Chadd struct ip6_direct_ctx *ip6dc; 3905778b399SBjoern A. Zeeb struct m_tag *mtag; 391aaa46574SAdrian Chadd #endif 392aaa46574SAdrian Chadd 3935778b399SBjoern A. Zeeb m = *mp; 3945778b399SBjoern A. Zeeb offset = *offp; 3955778b399SBjoern A. Zeeb 39682cd038dSYoshinobu Inoue ip6 = mtod(m, struct ip6_hdr *); 397686cdd19SJun-ichiro itojun Hagino #ifndef PULLDOWN_TEST 398686cdd19SJun-ichiro itojun Hagino IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE); 39982cd038dSYoshinobu Inoue ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); 400686cdd19SJun-ichiro itojun Hagino #else 401686cdd19SJun-ichiro itojun Hagino IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f)); 402686cdd19SJun-ichiro itojun Hagino if (ip6f == NULL) 40340e39bbbSHajimu UMEMOTO return (IPPROTO_DONE); 404686cdd19SJun-ichiro itojun Hagino #endif 40582cd038dSYoshinobu Inoue 40682cd038dSYoshinobu Inoue dstifp = NULL; 40723d374aaSBjoern A. Zeeb /* Find the destination interface of the packet. */ 4085778b399SBjoern A. Zeeb ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 4095778b399SBjoern A. Zeeb if (ia6 != NULL) { 4105778b399SBjoern A. Zeeb dstifp = ia6->ia_ifp; 4115778b399SBjoern A. Zeeb ifa_free(&ia6->ia_ifa); 4128c0fec80SRobert Watson } 41323d374aaSBjoern A. Zeeb 41423d374aaSBjoern A. Zeeb /* Jumbo payload cannot contain a fragment header. */ 41582cd038dSYoshinobu Inoue if (ip6->ip6_plen == 0) { 41682cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); 41782cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 4185778b399SBjoern A. Zeeb return (IPPROTO_DONE); 41982cd038dSYoshinobu Inoue } 42082cd038dSYoshinobu Inoue 42182cd038dSYoshinobu Inoue /* 42223d374aaSBjoern A. Zeeb * Check whether fragment packet's fragment length is a 42323d374aaSBjoern A. Zeeb * multiple of 8 octets (unless it is the last one). 42482cd038dSYoshinobu Inoue * sizeof(struct ip6_frag) == 8 42582cd038dSYoshinobu Inoue * sizeof(struct ip6_hdr) = 40 42682cd038dSYoshinobu Inoue */ 42782cd038dSYoshinobu Inoue if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && 42882cd038dSYoshinobu Inoue (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { 42906cd0a3fSHajimu UMEMOTO icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 430686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_hdr, ip6_plen)); 43182cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 4325778b399SBjoern A. Zeeb return (IPPROTO_DONE); 43382cd038dSYoshinobu Inoue } 43482cd038dSYoshinobu Inoue 4359cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragments); 43682cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_reqd); 43782cd038dSYoshinobu Inoue 43823d374aaSBjoern A. Zeeb /* Offset now points to data portion. */ 43982cd038dSYoshinobu Inoue offset += sizeof(struct ip6_frag); 44082cd038dSYoshinobu Inoue 4414018ea9aSBjoern A. Zeeb /* 4422946a941STom Jones * Handle "atomic" fragments (offset and m bit set to 0) upfront, 44323d374aaSBjoern A. Zeeb * unrelated to any reassembly. Still need to remove the frag hdr. 44423d374aaSBjoern A. Zeeb * See RFC 6946 and section 4.5 of RFC 8200. 4454018ea9aSBjoern A. Zeeb */ 4464018ea9aSBjoern A. Zeeb if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { 4472946a941STom Jones IP6STAT_INC(ip6s_atomicfrags); 44823d374aaSBjoern A. Zeeb /* XXX-BZ handle correctly. */ 4494018ea9aSBjoern A. Zeeb in6_ifstat_inc(dstifp, ifs6_reass_ok); 4504018ea9aSBjoern A. Zeeb *offp = offset; 451a4061289SAndrey V. Elsukov m->m_flags |= M_FRAGMENTED; 4524018ea9aSBjoern A. Zeeb return (ip6f->ip6f_nxt); 4534018ea9aSBjoern A. Zeeb } 4544018ea9aSBjoern A. Zeeb 4555f9f192dSJonathan T. Looney /* Get fragment length and discard 0-byte fragments. */ 4565f9f192dSJonathan T. Looney frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; 4575f9f192dSJonathan T. Looney if (frgpartlen == 0) { 4585f9f192dSJonathan T. Looney icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 4595f9f192dSJonathan T. Looney offsetof(struct ip6_hdr, ip6_plen)); 4605f9f192dSJonathan T. Looney in6_ifstat_inc(dstifp, ifs6_reass_fail); 4615f9f192dSJonathan T. Looney IP6STAT_INC(ip6s_fragdropped); 4625778b399SBjoern A. Zeeb return (IPPROTO_DONE); 4635f9f192dSJonathan T. Looney } 4645f9f192dSJonathan T. Looney 4657715d794SBjoern A. Zeeb /* 4667715d794SBjoern A. Zeeb * Enforce upper bound on number of fragments for the entire system. 4677715d794SBjoern A. Zeeb * If maxfrag is 0, never accept fragments. 4687715d794SBjoern A. Zeeb * If maxfrag is -1, accept all fragments without limitation. 4697715d794SBjoern A. Zeeb */ 4707715d794SBjoern A. Zeeb if (ip6_maxfrags < 0) 4717715d794SBjoern A. Zeeb ; 4727715d794SBjoern A. Zeeb else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags) 4737715d794SBjoern A. Zeeb goto dropfrag2; 4747715d794SBjoern A. Zeeb 47530809ba9SBjoern A. Zeeb /* 47630809ba9SBjoern A. Zeeb * Validate that a full header chain to the ULP is present in the 47730809ba9SBjoern A. Zeeb * packet containing the first fragment as per RFC RFC7112 and 47830809ba9SBjoern A. Zeeb * RFC 8200 pages 18,19: 47930809ba9SBjoern A. Zeeb * The first fragment packet is composed of: 48030809ba9SBjoern A. Zeeb * (3) Extension headers, if any, and the Upper-Layer header. These 48130809ba9SBjoern A. Zeeb * headers must be in the first fragment. ... 48230809ba9SBjoern A. Zeeb */ 48330809ba9SBjoern A. Zeeb fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); 48430809ba9SBjoern A. Zeeb /* XXX TODO. thj has D16851 open for this. */ 48530809ba9SBjoern A. Zeeb /* Send ICMPv6 4,3 in case of violation. */ 48630809ba9SBjoern A. Zeeb 487efdfee93SBjoern A. Zeeb /* Store receive network interface pointer for later. */ 488efdfee93SBjoern A. Zeeb srcifp = m->m_pkthdr.rcvif; 489efdfee93SBjoern A. Zeeb 49023d374aaSBjoern A. Zeeb /* Generate a hash value for fragment bucket selection. */ 49180d7a853SJonathan T. Looney hashkeyp = hashkey; 49280d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr)); 49380d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 49480d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr)); 49580d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 49680d7a853SJonathan T. Looney *hashkeyp = ip6f->ip6f_ident; 4979cb1a47aSBjoern A. Zeeb bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed); 4989cb1a47aSBjoern A. Zeeb bucket &= IP6REASS_HMASK; 4999cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 50021f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 5019888c401SHajimu UMEMOTO 50221f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) 50382cd038dSYoshinobu Inoue if (ip6f->ip6f_ident == q6->ip6q_ident && 50482cd038dSYoshinobu Inoue IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && 5054b908c8bSRobert Watson IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) 5064b908c8bSRobert Watson #ifdef MAC 5074b908c8bSRobert Watson && mac_ip6q_match(m, q6) 5084b908c8bSRobert Watson #endif 5094b908c8bSRobert Watson ) 51082cd038dSYoshinobu Inoue break; 51182cd038dSYoshinobu Inoue 5125778b399SBjoern A. Zeeb only_frag = false; 51321f08a07SBjoern A. Zeeb if (q6 == NULL) { 51423d374aaSBjoern A. Zeeb 5155778b399SBjoern A. Zeeb /* A first fragment to arrive creates a reassembly queue. */ 5165778b399SBjoern A. Zeeb only_frag = true; 51782cd038dSYoshinobu Inoue 51882cd038dSYoshinobu Inoue /* 51982cd038dSYoshinobu Inoue * Enforce upper bound on number of fragmented packets 52082cd038dSYoshinobu Inoue * for which we attempt reassembly; 5219888c401SHajimu UMEMOTO * If maxfragpackets is 0, never accept fragments. 5229888c401SHajimu UMEMOTO * If maxfragpackets is -1, accept all fragments without 5239888c401SHajimu UMEMOTO * limitation. 52482cd038dSYoshinobu Inoue */ 525603724d3SBjoern A. Zeeb if (V_ip6_maxfragpackets < 0) 52633841545SHajimu UMEMOTO ; 5279cb1a47aSBjoern A. Zeeb else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize || 5281e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) >= 52980d7a853SJonathan T. Looney (u_int)V_ip6_maxfragpackets) 53033841545SHajimu UMEMOTO goto dropfrag; 53123d374aaSBjoern A. Zeeb 53223d374aaSBjoern A. Zeeb /* Allocate IPv6 fragement packet queue entry. */ 533487a161cSBjoern A. Zeeb q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6, 534487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 53582cd038dSYoshinobu Inoue if (q6 == NULL) 53682cd038dSYoshinobu Inoue goto dropfrag; 5374b908c8bSRobert Watson #ifdef MAC 5384b908c8bSRobert Watson if (mac_ip6q_init(q6, M_NOWAIT) != 0) { 539487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 5404b908c8bSRobert Watson goto dropfrag; 5414b908c8bSRobert Watson } 5424b908c8bSRobert Watson mac_ip6q_create(m, q6); 5434b908c8bSRobert Watson #endif 544702828f6SBjoern A. Zeeb atomic_add_int(&V_frag6_nfragpackets, 1); 54582cd038dSYoshinobu Inoue 54623d374aaSBjoern A. Zeeb /* ip6q_nxt will be filled afterwards, from 1st fragment. */ 54721f08a07SBjoern A. Zeeb TAILQ_INIT(&q6->ip6q_frags); 54882cd038dSYoshinobu Inoue q6->ip6q_ident = ip6f->ip6f_ident; 54982cd038dSYoshinobu Inoue q6->ip6q_ttl = IPV6_FRAGTTL; 55082cd038dSYoshinobu Inoue q6->ip6q_src = ip6->ip6_src; 55182cd038dSYoshinobu Inoue q6->ip6q_dst = ip6->ip6_dst; 5525e9510e3SJINMEI Tatuya q6->ip6q_ecn = 5535e9510e3SJINMEI Tatuya (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 55482cd038dSYoshinobu Inoue q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ 5559888c401SHajimu UMEMOTO 55621f08a07SBjoern A. Zeeb /* Add the fragemented packet to the bucket. */ 55721f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(head, q6, ip6q_tq); 55821f08a07SBjoern A. Zeeb V_ip6qb[bucket].count++; 55982cd038dSYoshinobu Inoue } 56082cd038dSYoshinobu Inoue 56182cd038dSYoshinobu Inoue /* 56223d374aaSBjoern A. Zeeb * If it is the 1st fragment, record the length of the 56382cd038dSYoshinobu Inoue * unfragmentable part and the next header of the fragment header. 564619456bbSBjoern A. Zeeb * Assume the first 1st fragement to arrive will be correct. 565619456bbSBjoern A. Zeeb * We do not have any duplicate checks here yet so another packet 566619456bbSBjoern A. Zeeb * with fragoff == 0 could come and overwrite the ip6q_unfrglen 567619456bbSBjoern A. Zeeb * and worse, the next header, at any time. 56882cd038dSYoshinobu Inoue */ 569619456bbSBjoern A. Zeeb if (fragoff == 0 && q6->ip6q_unfrglen == -1) { 57006cd0a3fSHajimu UMEMOTO q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - 57106cd0a3fSHajimu UMEMOTO sizeof(struct ip6_frag); 57282cd038dSYoshinobu Inoue q6->ip6q_nxt = ip6f->ip6f_nxt; 573619456bbSBjoern A. Zeeb /* XXX ECN? */ 57482cd038dSYoshinobu Inoue } 57582cd038dSYoshinobu Inoue 57682cd038dSYoshinobu Inoue /* 57782cd038dSYoshinobu Inoue * Check that the reassembled packet would not exceed 65535 bytes 57882cd038dSYoshinobu Inoue * in size. 57982cd038dSYoshinobu Inoue * If it would exceed, discard the fragment and return an ICMP error. 58082cd038dSYoshinobu Inoue */ 58182cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen >= 0) { 58282cd038dSYoshinobu Inoue /* The 1st fragment has already arrived. */ 58382cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { 584e5fffe9aSBjoern A. Zeeb if (only_frag) { 585e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 586e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 587e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 588e5fffe9aSBjoern A. Zeeb #ifdef MAC 589e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 590e5fffe9aSBjoern A. Zeeb #endif 591e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 592e5fffe9aSBjoern A. Zeeb } 593e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 59482cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 595686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 596686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 59782cd038dSYoshinobu Inoue return (IPPROTO_DONE); 59882cd038dSYoshinobu Inoue } 59906cd0a3fSHajimu UMEMOTO } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { 600e5fffe9aSBjoern A. Zeeb if (only_frag) { 601e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 602e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 603e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 604e5fffe9aSBjoern A. Zeeb #ifdef MAC 605e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 606e5fffe9aSBjoern A. Zeeb #endif 607e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 608e5fffe9aSBjoern A. Zeeb } 609e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 61082cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 611686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 612686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 61382cd038dSYoshinobu Inoue return (IPPROTO_DONE); 61482cd038dSYoshinobu Inoue } 615f1664f32SBjoern A. Zeeb 61682cd038dSYoshinobu Inoue /* 61723d374aaSBjoern A. Zeeb * If it is the first fragment, do the above check for each 61882cd038dSYoshinobu Inoue * fragment already stored in the reassembly queue. 61982cd038dSYoshinobu Inoue */ 620dda02192SBjoern A. Zeeb if (fragoff == 0 && !only_frag) { 62121f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(af6, &q6->ip6q_frags, ip6af_tq, af6tmp) { 62282cd038dSYoshinobu Inoue 623dda02192SBjoern A. Zeeb if (q6->ip6q_unfrglen + af6->ip6af_off + 624dda02192SBjoern A. Zeeb af6->ip6af_frglen > IPV6_MAXPACKET) { 62582cd038dSYoshinobu Inoue struct ip6_hdr *ip6err; 6265778b399SBjoern A. Zeeb struct mbuf *merr; 6275778b399SBjoern A. Zeeb int erroff; 6285778b399SBjoern A. Zeeb 629da89a0feSBjoern A. Zeeb merr = af6->ip6af_m; 6305778b399SBjoern A. Zeeb erroff = af6->ip6af_offset; 63182cd038dSYoshinobu Inoue 63223d374aaSBjoern A. Zeeb /* Dequeue the fragment. */ 63321f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 634dda02192SBjoern A. Zeeb q6->ip6q_nfrag--; 635dda02192SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, 1); 636487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 63782cd038dSYoshinobu Inoue 638a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 639a55383e7SHans Petter Selasky merr->m_pkthdr.rcvif = srcifp; 640a55383e7SHans Petter Selasky 64123d374aaSBjoern A. Zeeb /* Adjust pointer. */ 64282cd038dSYoshinobu Inoue ip6err = mtod(merr, struct ip6_hdr *); 64382cd038dSYoshinobu Inoue 64482cd038dSYoshinobu Inoue /* 64582cd038dSYoshinobu Inoue * Restore source and destination addresses 64682cd038dSYoshinobu Inoue * in the erroneous IPv6 header. 64782cd038dSYoshinobu Inoue */ 64882cd038dSYoshinobu Inoue ip6err->ip6_src = q6->ip6q_src; 64982cd038dSYoshinobu Inoue ip6err->ip6_dst = q6->ip6q_dst; 65082cd038dSYoshinobu Inoue 65182cd038dSYoshinobu Inoue icmp6_error(merr, ICMP6_PARAM_PROB, 65282cd038dSYoshinobu Inoue ICMP6_PARAMPROB_HEADER, 653686cdd19SJun-ichiro itojun Hagino erroff - sizeof(struct ip6_frag) + 654686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 65582cd038dSYoshinobu Inoue } 65682cd038dSYoshinobu Inoue } 65782cd038dSYoshinobu Inoue } 65882cd038dSYoshinobu Inoue 65923d374aaSBjoern A. Zeeb /* Allocate an IPv6 fragement queue entry for this fragmented part. */ 660487a161cSBjoern A. Zeeb ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6, 661487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 662686cdd19SJun-ichiro itojun Hagino if (ip6af == NULL) 663686cdd19SJun-ichiro itojun Hagino goto dropfrag; 66421f08a07SBjoern A. Zeeb ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) ? true : false; 66582cd038dSYoshinobu Inoue ip6af->ip6af_off = fragoff; 66682cd038dSYoshinobu Inoue ip6af->ip6af_frglen = frgpartlen; 66782cd038dSYoshinobu Inoue ip6af->ip6af_offset = offset; 668da89a0feSBjoern A. Zeeb ip6af->ip6af_m = m; 66982cd038dSYoshinobu Inoue 6705778b399SBjoern A. Zeeb if (only_frag) { 67121f08a07SBjoern A. Zeeb /* 67221f08a07SBjoern A. Zeeb * Do a manual insert rather than a hard-to-understand cast 67321f08a07SBjoern A. Zeeb * to a different type relying on data structure order to work. 67421f08a07SBjoern A. Zeeb */ 67521f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(&q6->ip6q_frags, ip6af, ip6af_tq); 67621f08a07SBjoern A. Zeeb goto postinsert; 67782cd038dSYoshinobu Inoue } 67882cd038dSYoshinobu Inoue 67923d374aaSBjoern A. Zeeb /* Do duplicate, condition, and boundry checks. */ 68082cd038dSYoshinobu Inoue /* 68159dfcba4SHajimu UMEMOTO * Handle ECN by comparing this segment with the first one; 68259dfcba4SHajimu UMEMOTO * if CE is set, do not lose CE. 68323d374aaSBjoern A. Zeeb * Drop if CE and not-ECT are mixed for the same packet. 68459dfcba4SHajimu UMEMOTO */ 68559dfcba4SHajimu UMEMOTO ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 6865e9510e3SJINMEI Tatuya ecn0 = q6->ip6q_ecn; 68759dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_CE) { 68859dfcba4SHajimu UMEMOTO if (ecn0 == IPTOS_ECN_NOTECT) { 689487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 69059dfcba4SHajimu UMEMOTO goto dropfrag; 69159dfcba4SHajimu UMEMOTO } 69259dfcba4SHajimu UMEMOTO if (ecn0 != IPTOS_ECN_CE) 6935e9510e3SJINMEI Tatuya q6->ip6q_ecn = IPTOS_ECN_CE; 69459dfcba4SHajimu UMEMOTO } 69559dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { 696487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 69759dfcba4SHajimu UMEMOTO goto dropfrag; 69859dfcba4SHajimu UMEMOTO } 69959dfcba4SHajimu UMEMOTO 70023d374aaSBjoern A. Zeeb /* Find a fragmented part which begins after this one does. */ 70121f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) 70282cd038dSYoshinobu Inoue if (af6->ip6af_off > ip6af->ip6af_off) 70382cd038dSYoshinobu Inoue break; 70482cd038dSYoshinobu Inoue 70582cd038dSYoshinobu Inoue /* 70682cd038dSYoshinobu Inoue * If the incoming framgent overlaps some existing fragments in 70723d374aaSBjoern A. Zeeb * the reassembly queue, drop both the new fragment and the 70823d374aaSBjoern A. Zeeb * entire reassembly queue. However, if the new fragment 70923d374aaSBjoern A. Zeeb * is an exact duplicate of an existing fragment, only silently 71023d374aaSBjoern A. Zeeb * drop the existing fragment and leave the fragmentation queue 71123d374aaSBjoern A. Zeeb * unchanged, as allowed by the RFC. (RFC 8200, 4.5) 71282cd038dSYoshinobu Inoue */ 71321f08a07SBjoern A. Zeeb if (af6 != NULL) 71421f08a07SBjoern A. Zeeb af6tmp = TAILQ_PREV(af6, ip6fraghead, ip6af_tq); 71521f08a07SBjoern A. Zeeb else 71621f08a07SBjoern A. Zeeb af6tmp = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 71721f08a07SBjoern A. Zeeb if (af6tmp != NULL) { 71821f08a07SBjoern A. Zeeb if (af6tmp->ip6af_off + af6tmp->ip6af_frglen - 7195778b399SBjoern A. Zeeb ip6af->ip6af_off > 0) { 720cd188da2SBjoern A. Zeeb if (af6tmp->ip6af_off != ip6af->ip6af_off || 721cd188da2SBjoern A. Zeeb af6tmp->ip6af_frglen != ip6af->ip6af_frglen) 722cd188da2SBjoern A. Zeeb frag6_freef(q6, bucket); 723487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 72482cd038dSYoshinobu Inoue goto dropfrag; 72582cd038dSYoshinobu Inoue } 72682cd038dSYoshinobu Inoue } 72721f08a07SBjoern A. Zeeb if (af6 != NULL) { 7285778b399SBjoern A. Zeeb if (ip6af->ip6af_off + ip6af->ip6af_frglen - 7295778b399SBjoern A. Zeeb af6->ip6af_off > 0) { 730cd188da2SBjoern A. Zeeb if (af6->ip6af_off != ip6af->ip6af_off || 731cd188da2SBjoern A. Zeeb af6->ip6af_frglen != ip6af->ip6af_frglen) 732cd188da2SBjoern A. Zeeb frag6_freef(q6, bucket); 733487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 73482cd038dSYoshinobu Inoue goto dropfrag; 73582cd038dSYoshinobu Inoue } 73682cd038dSYoshinobu Inoue } 73782cd038dSYoshinobu Inoue 7384b908c8bSRobert Watson #ifdef MAC 7394b908c8bSRobert Watson mac_ip6q_update(m, q6); 7404b908c8bSRobert Watson #endif 74182cd038dSYoshinobu Inoue 74282cd038dSYoshinobu Inoue /* 74323d374aaSBjoern A. Zeeb * Stick new segment in its place; check for complete reassembly. 74423d374aaSBjoern A. Zeeb * If not complete, check fragment limit. Move to front of packet 74523d374aaSBjoern A. Zeeb * queue, as we are the most recently active fragmented packet. 74682cd038dSYoshinobu Inoue */ 74721f08a07SBjoern A. Zeeb if (af6 != NULL) 74821f08a07SBjoern A. Zeeb TAILQ_INSERT_BEFORE(af6, ip6af, ip6af_tq); 74921f08a07SBjoern A. Zeeb else 75021f08a07SBjoern A. Zeeb TAILQ_INSERT_TAIL(&q6->ip6q_frags, ip6af, ip6af_tq); 75121f08a07SBjoern A. Zeeb postinsert: 7522adfd64fSJonathan T. Looney atomic_add_int(&frag6_nfrags, 1); 7539888c401SHajimu UMEMOTO q6->ip6q_nfrag++; 7543c7165b3SBjoern A. Zeeb 7555778b399SBjoern A. Zeeb plen = 0; 75621f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 7575778b399SBjoern A. Zeeb if (af6->ip6af_off != plen) { 75803c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 759198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7609cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 76103c99d76SJonathan T. Looney } 7629cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 7635778b399SBjoern A. Zeeb return (IPPROTO_DONE); 76482cd038dSYoshinobu Inoue } 7655778b399SBjoern A. Zeeb plen += af6->ip6af_frglen; 76682cd038dSYoshinobu Inoue } 76721f08a07SBjoern A. Zeeb af6 = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 76821f08a07SBjoern A. Zeeb if (af6->ip6af_mff) { 76903c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 770198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7719cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 77203c99d76SJonathan T. Looney } 7739cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 7745778b399SBjoern A. Zeeb return (IPPROTO_DONE); 77582cd038dSYoshinobu Inoue } 77682cd038dSYoshinobu Inoue 77723d374aaSBjoern A. Zeeb /* Reassembly is complete; concatenate fragments. */ 77821f08a07SBjoern A. Zeeb ip6af = TAILQ_FIRST(&q6->ip6q_frags); 779da89a0feSBjoern A. Zeeb t = m = ip6af->ip6af_m; 78021f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, ip6af, ip6af_tq); 78121f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 7829907aba3SAndrey V. Elsukov m->m_pkthdr.csum_flags &= 783da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_flags; 7849907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data += 785da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_data; 7869907aba3SAndrey V. Elsukov 78721f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 788efdfee93SBjoern A. Zeeb t = m_last(t); 789da89a0feSBjoern A. Zeeb m_adj(af6->ip6af_m, af6->ip6af_offset); 790da89a0feSBjoern A. Zeeb m_demote_pkthdr(af6->ip6af_m); 791da89a0feSBjoern A. Zeeb m_cat(t, af6->ip6af_m); 792487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 79382cd038dSYoshinobu Inoue } 79482cd038dSYoshinobu Inoue 7959907aba3SAndrey V. Elsukov while (m->m_pkthdr.csum_data & 0xffff0000) 7969907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 7979907aba3SAndrey V. Elsukov (m->m_pkthdr.csum_data >> 16); 7989907aba3SAndrey V. Elsukov 79923d374aaSBjoern A. Zeeb /* Adjust offset to point where the original next header starts. */ 80082cd038dSYoshinobu Inoue offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); 801487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 802686cdd19SJun-ichiro itojun Hagino ip6 = mtod(m, struct ip6_hdr *); 8035778b399SBjoern A. Zeeb ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr)); 8045e9510e3SJINMEI Tatuya if (q6->ip6q_ecn == IPTOS_ECN_CE) 8055e9510e3SJINMEI Tatuya ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); 80682cd038dSYoshinobu Inoue nxt = q6->ip6q_nxt; 80782cd038dSYoshinobu Inoue 80821f08a07SBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 80921f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 8102adfd64fSJonathan T. Looney atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 81121f08a07SBjoern A. Zeeb 81221f08a07SBjoern A. Zeeb if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) { 8134b908c8bSRobert Watson #ifdef MAC 8144b908c8bSRobert Watson mac_ip6q_destroy(q6); 8154b908c8bSRobert Watson #endif 816487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 81780d7a853SJonathan T. Looney atomic_subtract_int(&V_frag6_nfragpackets, 1); 8180b438b0fSGleb Smirnoff 819686cdd19SJun-ichiro itojun Hagino goto dropfrag; 82082cd038dSYoshinobu Inoue } 82182cd038dSYoshinobu Inoue 82223d374aaSBjoern A. Zeeb /* Set nxt(-hdr field value) to the original value. */ 82368e0e5a6SAndrey V. Elsukov m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 82468e0e5a6SAndrey V. Elsukov (caddr_t)&nxt); 82582cd038dSYoshinobu Inoue 8264b908c8bSRobert Watson #ifdef MAC 8274b908c8bSRobert Watson mac_ip6q_reassemble(q6, m); 8284b908c8bSRobert Watson mac_ip6q_destroy(q6); 8294b908c8bSRobert Watson #endif 830487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 83180d7a853SJonathan T. Looney atomic_subtract_int(&V_frag6_nfragpackets, 1); 83282cd038dSYoshinobu Inoue 83382cd038dSYoshinobu Inoue if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ 8345778b399SBjoern A. Zeeb 8355778b399SBjoern A. Zeeb plen = 0; 83682cd038dSYoshinobu Inoue for (t = m; t; t = t->m_next) 83782cd038dSYoshinobu Inoue plen += t->m_len; 83882cd038dSYoshinobu Inoue m->m_pkthdr.len = plen; 839a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 840a55383e7SHans Petter Selasky m->m_pkthdr.rcvif = srcifp; 84182cd038dSYoshinobu Inoue } 84282cd038dSYoshinobu Inoue 843aaa46574SAdrian Chadd #ifdef RSS 844aaa46574SAdrian Chadd mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), 845aaa46574SAdrian Chadd M_NOWAIT); 846aaa46574SAdrian Chadd if (mtag == NULL) 847aaa46574SAdrian Chadd goto dropfrag; 848aaa46574SAdrian Chadd 849aaa46574SAdrian Chadd ip6dc = (struct ip6_direct_ctx *)(mtag + 1); 850aaa46574SAdrian Chadd ip6dc->ip6dc_nxt = nxt; 851aaa46574SAdrian Chadd ip6dc->ip6dc_off = offset; 852aaa46574SAdrian Chadd 853aaa46574SAdrian Chadd m_tag_prepend(m, mtag); 854aaa46574SAdrian Chadd #endif 855aaa46574SAdrian Chadd 8569cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 8579cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_reassembled); 85882cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_ok); 85982cd038dSYoshinobu Inoue 860aaa46574SAdrian Chadd #ifdef RSS 86123d374aaSBjoern A. Zeeb /* Queue/dispatch for reprocessing. */ 862aaa46574SAdrian Chadd netisr_dispatch(NETISR_IPV6_DIRECT, m); 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); 8785778b399SBjoern A. Zeeb return (IPPROTO_DONE); 87982cd038dSYoshinobu Inoue } 88082cd038dSYoshinobu Inoue 88182cd038dSYoshinobu Inoue /* 88233841545SHajimu UMEMOTO * IPv6 reassembling timer processing; 88323d374aaSBjoern A. Zeeb * if a timer expires on a reassembly queue, discard it. 88482cd038dSYoshinobu Inoue */ 88582cd038dSYoshinobu Inoue void 8861272577eSXin LI frag6_slowtimo(void) 88782cd038dSYoshinobu Inoue { 8888b615593SMarko Zec VNET_ITERATOR_DECL(vnet_iter); 88921f08a07SBjoern A. Zeeb struct ip6qhead *head; 89021f08a07SBjoern A. Zeeb struct ip6q *q6, *q6tmp; 8919cb1a47aSBjoern A. Zeeb uint32_t bucket; 89282cd038dSYoshinobu Inoue 8935ee847d3SRobert Watson VNET_LIST_RLOCK_NOSLEEP(); 8948b615593SMarko Zec VNET_FOREACH(vnet_iter) { 8958b615593SMarko Zec CURVNET_SET(vnet_iter); 8969cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 8979cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 8989cb1a47aSBjoern A. Zeeb head = IP6QB_HEAD(bucket); 89921f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(q6, head, ip6q_tq, q6tmp) 90021f08a07SBjoern A. Zeeb if (--q6->ip6q_ttl == 0) { 901198fdaedSTom Jones IP6STAT_ADD(ip6s_fragtimeout, 90221f08a07SBjoern A. Zeeb q6->ip6q_nfrag); 90382cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 90421f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 90582cd038dSYoshinobu Inoue } 90682cd038dSYoshinobu Inoue /* 90782cd038dSYoshinobu Inoue * If we are over the maximum number of fragments 90882cd038dSYoshinobu Inoue * (due to the limit being lowered), drain off 90982cd038dSYoshinobu Inoue * enough to get down to the new limit. 9101e9f3b73SJonathan T. Looney * Note that we drain all reassembly queues if 9111e9f3b73SJonathan T. Looney * maxfragpackets is 0 (fragmentation is disabled), 91223d374aaSBjoern A. Zeeb * and do not enforce a limit when maxfragpackets 9131e9f3b73SJonathan T. Looney * is negative. 91482cd038dSYoshinobu Inoue */ 9151e9f3b73SJonathan T. Looney while ((V_ip6_maxfragpackets == 0 || 9161e9f3b73SJonathan T. Looney (V_ip6_maxfragpackets > 0 && 9179cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) && 91821f08a07SBjoern A. Zeeb (q6 = TAILQ_LAST(head, ip6qhead)) != NULL) { 91921f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 92082cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 92121f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 92280d7a853SJonathan T. Looney } 9239cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 92482cd038dSYoshinobu Inoue } 9251e9f3b73SJonathan T. Looney /* 9261e9f3b73SJonathan T. Looney * If we are still over the maximum number of fragmented 9271e9f3b73SJonathan T. Looney * packets, drain off enough to get down to the new limit. 9281e9f3b73SJonathan T. Looney */ 9299cb1a47aSBjoern A. Zeeb bucket = 0; 9301e9f3b73SJonathan T. Looney while (V_ip6_maxfragpackets >= 0 && 9311e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) > 9321e9f3b73SJonathan T. Looney (u_int)V_ip6_maxfragpackets) { 9339cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 93421f08a07SBjoern A. Zeeb q6 = TAILQ_LAST(IP6QB_HEAD(bucket), ip6qhead); 93521f08a07SBjoern A. Zeeb if (q6 != NULL) { 93621f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 9371e9f3b73SJonathan T. Looney /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 93821f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 9391e9f3b73SJonathan T. Looney } 9409cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 9419cb1a47aSBjoern A. Zeeb bucket = (bucket + 1) % IP6REASS_NHASH; 9421e9f3b73SJonathan T. Looney } 9438b615593SMarko Zec CURVNET_RESTORE(); 9448b615593SMarko Zec } 9455ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 94682cd038dSYoshinobu Inoue } 94782cd038dSYoshinobu Inoue 94823d374aaSBjoern A. Zeeb /* 94923d374aaSBjoern A. Zeeb * Eventhandler to adjust limits in case nmbclusters change. 95023d374aaSBjoern A. Zeeb */ 951c00464a2SBjoern A. Zeeb static void 952c00464a2SBjoern A. Zeeb frag6_change(void *tag) 953c00464a2SBjoern A. Zeeb { 954c00464a2SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 955c00464a2SBjoern A. Zeeb 956c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 957c00464a2SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 958c00464a2SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 959c00464a2SBjoern A. Zeeb CURVNET_SET(vnet_iter); 960c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 961c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 962c00464a2SBjoern A. Zeeb CURVNET_RESTORE(); 963c00464a2SBjoern A. Zeeb } 964c00464a2SBjoern A. Zeeb VNET_LIST_RUNLOCK_NOSLEEP(); 965c00464a2SBjoern A. Zeeb } 966c00464a2SBjoern A. Zeeb 967c00464a2SBjoern A. Zeeb /* 968c00464a2SBjoern A. Zeeb * Initialise reassembly queue and fragment identifier. 969c00464a2SBjoern A. Zeeb */ 970c00464a2SBjoern A. Zeeb void 971c00464a2SBjoern A. Zeeb frag6_init(void) 972c00464a2SBjoern A. Zeeb { 9739cb1a47aSBjoern A. Zeeb uint32_t bucket; 974c00464a2SBjoern A. Zeeb 975c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 976c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 9779cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 97821f08a07SBjoern A. Zeeb TAILQ_INIT(IP6QB_HEAD(bucket)); 979efdfee93SBjoern A. Zeeb mtx_init(&V_ip6qb[bucket].lock, "ip6qb", NULL, MTX_DEF); 9809cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count = 0; 981c00464a2SBjoern A. Zeeb } 9829cb1a47aSBjoern A. Zeeb V_ip6qb_hashseed = arc4random(); 983c00464a2SBjoern A. Zeeb V_ip6_maxfragsperpacket = 64; 98467a10c46SBjoern A. Zeeb #ifdef VIMAGE 98567a10c46SBjoern A. Zeeb V_frag6_on = true; 98667a10c46SBjoern A. Zeeb #endif 987c00464a2SBjoern A. Zeeb if (!IS_DEFAULT_VNET(curvnet)) 988c00464a2SBjoern A. Zeeb return; 989c00464a2SBjoern A. Zeeb 990c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 991c00464a2SBjoern A. Zeeb EVENTHANDLER_REGISTER(nmbclusters_change, 992c00464a2SBjoern A. Zeeb frag6_change, NULL, EVENTHANDLER_PRI_ANY); 993c00464a2SBjoern A. Zeeb } 994c00464a2SBjoern A. Zeeb 99582cd038dSYoshinobu Inoue /* 99682cd038dSYoshinobu Inoue * Drain off all datagram fragments. 99782cd038dSYoshinobu Inoue */ 99867a10c46SBjoern A. Zeeb static void 99967a10c46SBjoern A. Zeeb frag6_drain_one(void) 100082cd038dSYoshinobu Inoue { 100121f08a07SBjoern A. Zeeb struct ip6q *q6; 10029cb1a47aSBjoern A. Zeeb uint32_t bucket; 10039888c401SHajimu UMEMOTO 10049cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 100567a10c46SBjoern A. Zeeb IP6QB_LOCK(bucket); 100621f08a07SBjoern A. Zeeb while ((q6 = TAILQ_FIRST(IP6QB_HEAD(bucket))) != NULL) { 10079cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragdropped); 100882cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 100921f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 101080d7a853SJonathan T. Looney } 10119cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 101282cd038dSYoshinobu Inoue } 101367a10c46SBjoern A. Zeeb } 101467a10c46SBjoern A. Zeeb 101567a10c46SBjoern A. Zeeb void 101667a10c46SBjoern A. Zeeb frag6_drain(void) 101767a10c46SBjoern A. Zeeb { 101867a10c46SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 101967a10c46SBjoern A. Zeeb 102067a10c46SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 102167a10c46SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 102267a10c46SBjoern A. Zeeb CURVNET_SET(vnet_iter); 102367a10c46SBjoern A. Zeeb frag6_drain_one(); 10248b615593SMarko Zec CURVNET_RESTORE(); 10258b615593SMarko Zec } 10265ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 102782cd038dSYoshinobu Inoue } 1028e5ee7060SGleb Smirnoff 102967a10c46SBjoern A. Zeeb #ifdef VIMAGE 103067a10c46SBjoern A. Zeeb /* 103167a10c46SBjoern A. Zeeb * Clear up IPv6 reassembly structures. 103267a10c46SBjoern A. Zeeb */ 103367a10c46SBjoern A. Zeeb void 103467a10c46SBjoern A. Zeeb frag6_destroy(void) 103567a10c46SBjoern A. Zeeb { 103667a10c46SBjoern A. Zeeb uint32_t bucket; 103767a10c46SBjoern A. Zeeb 103867a10c46SBjoern A. Zeeb frag6_drain_one(); 103967a10c46SBjoern A. Zeeb V_frag6_on = false; 104067a10c46SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 104167a10c46SBjoern A. Zeeb KASSERT(V_ip6qb[bucket].count == 0, 104267a10c46SBjoern A. Zeeb ("%s: V_ip6qb[%d] (%p) count not 0 (%d)", __func__, 104367a10c46SBjoern A. Zeeb bucket, &V_ip6qb[bucket], V_ip6qb[bucket].count)); 104467a10c46SBjoern A. Zeeb mtx_destroy(&V_ip6qb[bucket].lock); 104567a10c46SBjoern A. Zeeb } 104667a10c46SBjoern A. Zeeb } 104767a10c46SBjoern A. Zeeb #endif 1048