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"); 205757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGSPERPACKET, maxfragsperpacket, 206757cb678SBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragsperpacket), 0, 207757cb678SBjoern A. Zeeb "Maximum allowed number of fragments per packet"); 208757cb678SBjoern A. Zeeb SYSCTL_INT(_net_inet6_ip6, IPV6CTL_MAXFRAGBUCKETSIZE, maxfragbucketsize, 209757cb678SBjoern A. Zeeb CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_maxfragbucketsize), 0, 210757cb678SBjoern A. Zeeb "Maximum number of reassembly queues per hash bucket"); 211757cb678SBjoern A. Zeeb 212757cb678SBjoern A. Zeeb 213757cb678SBjoern A. Zeeb /* 214c00464a2SBjoern A. Zeeb * Remove the IPv6 fragmentation header from the mbuf. 215c00464a2SBjoern A. Zeeb */ 216c00464a2SBjoern A. Zeeb int 217c00464a2SBjoern A. Zeeb ip6_deletefraghdr(struct mbuf *m, int offset, int wait) 218c00464a2SBjoern A. Zeeb { 2195778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 220c00464a2SBjoern A. Zeeb struct mbuf *t; 221c00464a2SBjoern A. Zeeb 222c00464a2SBjoern A. Zeeb /* Delete frag6 header. */ 223c00464a2SBjoern A. Zeeb if (m->m_len >= offset + sizeof(struct ip6_frag)) { 2245778b399SBjoern A. Zeeb 225c00464a2SBjoern A. Zeeb /* This is the only possible case with !PULLDOWN_TEST. */ 2265778b399SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 227c00464a2SBjoern A. Zeeb bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), 228c00464a2SBjoern A. Zeeb offset); 229c00464a2SBjoern A. Zeeb m->m_data += sizeof(struct ip6_frag); 230c00464a2SBjoern A. Zeeb m->m_len -= sizeof(struct ip6_frag); 231c00464a2SBjoern A. Zeeb } else { 2325778b399SBjoern A. Zeeb 233c00464a2SBjoern A. Zeeb /* This comes with no copy if the boundary is on cluster. */ 234c00464a2SBjoern A. Zeeb if ((t = m_split(m, offset, wait)) == NULL) 235c00464a2SBjoern A. Zeeb return (ENOMEM); 236c00464a2SBjoern A. Zeeb m_adj(t, sizeof(struct ip6_frag)); 237c00464a2SBjoern A. Zeeb m_cat(m, t); 238c00464a2SBjoern A. Zeeb } 239c00464a2SBjoern A. Zeeb 240c00464a2SBjoern A. Zeeb m->m_flags |= M_FRAGMENTED; 241c00464a2SBjoern A. Zeeb return (0); 242c00464a2SBjoern A. Zeeb } 243c00464a2SBjoern A. Zeeb 244c00464a2SBjoern A. Zeeb /* 24523d374aaSBjoern A. Zeeb * Free a fragment reassembly header and all associated datagrams. 246757cb678SBjoern A. Zeeb */ 2474f590175SPaul Saab static void 248c00464a2SBjoern A. Zeeb frag6_freef(struct ip6q *q6, uint32_t bucket) 2494f590175SPaul Saab { 2505778b399SBjoern A. Zeeb struct ip6_hdr *ip6; 25121f08a07SBjoern A. Zeeb struct ip6asfrag *af6; 2525778b399SBjoern A. Zeeb struct mbuf *m; 2534f590175SPaul Saab 2549cb1a47aSBjoern A. Zeeb IP6QB_LOCK_ASSERT(bucket); 255c00464a2SBjoern A. Zeeb 25621f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 257c00464a2SBjoern A. Zeeb 258da89a0feSBjoern A. Zeeb m = af6->ip6af_m; 25921f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 260c00464a2SBjoern A. Zeeb 261c00464a2SBjoern A. Zeeb /* 262c00464a2SBjoern A. Zeeb * Return ICMP time exceeded error for the 1st fragment. 263c00464a2SBjoern A. Zeeb * Just free other fragments. 264c00464a2SBjoern A. Zeeb */ 265a55383e7SHans Petter Selasky if (af6->ip6af_off == 0 && m->m_pkthdr.rcvif != NULL) { 266c00464a2SBjoern A. Zeeb 26723d374aaSBjoern A. Zeeb /* Adjust pointer. */ 268c00464a2SBjoern A. Zeeb ip6 = mtod(m, struct ip6_hdr *); 269c00464a2SBjoern A. Zeeb 27023d374aaSBjoern A. Zeeb /* Restore source and destination addresses. */ 271c00464a2SBjoern A. Zeeb ip6->ip6_src = q6->ip6q_src; 272c00464a2SBjoern A. Zeeb ip6->ip6_dst = q6->ip6q_dst; 273c00464a2SBjoern A. Zeeb 274c00464a2SBjoern A. Zeeb icmp6_error(m, ICMP6_TIME_EXCEEDED, 275c00464a2SBjoern A. Zeeb ICMP6_TIME_EXCEED_REASSEMBLY, 0); 276c00464a2SBjoern A. Zeeb } else 277c00464a2SBjoern A. Zeeb m_freem(m); 27823d374aaSBjoern A. Zeeb 279c00464a2SBjoern A. Zeeb free(af6, M_FRAG6); 2802adfd64fSJonathan T. Looney } 28121f08a07SBjoern A. Zeeb 28221f08a07SBjoern A. Zeeb TAILQ_REMOVE(IP6QB_HEAD(bucket), q6, ip6q_tq); 28321f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 284c00464a2SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 285c00464a2SBjoern A. Zeeb #ifdef MAC 286c00464a2SBjoern A. Zeeb mac_ip6q_destroy(q6); 287c00464a2SBjoern A. Zeeb #endif 288c00464a2SBjoern A. Zeeb free(q6, M_FRAG6); 289c00464a2SBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 29082cd038dSYoshinobu Inoue } 29182cd038dSYoshinobu Inoue 29282cd038dSYoshinobu Inoue /* 293a55383e7SHans Petter Selasky * Drain off all datagram fragments belonging to 294a55383e7SHans Petter Selasky * the given network interface. 295a55383e7SHans Petter Selasky */ 296a55383e7SHans Petter Selasky static void 297a55383e7SHans Petter Selasky frag6_cleanup(void *arg __unused, struct ifnet *ifp) 298a55383e7SHans Petter Selasky { 29921f08a07SBjoern A. Zeeb struct ip6qhead *head; 30021f08a07SBjoern A. Zeeb struct ip6q *q6; 301a55383e7SHans Petter Selasky struct ip6asfrag *af6; 30221f08a07SBjoern A. Zeeb uint32_t bucket; 303a55383e7SHans Petter Selasky 304a55383e7SHans Petter Selasky KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); 305a55383e7SHans Petter Selasky 30667a10c46SBjoern A. Zeeb #ifdef VIMAGE 30767a10c46SBjoern A. Zeeb /* 30867a10c46SBjoern A. Zeeb * Skip processing if IPv6 reassembly is not initialised or 30967a10c46SBjoern A. Zeeb * torn down by frag6_destroy(). 31067a10c46SBjoern A. Zeeb */ 31167a10c46SBjoern A. Zeeb if (!V_frag6_on) 31267a10c46SBjoern A. Zeeb return; 31367a10c46SBjoern A. Zeeb #endif 31467a10c46SBjoern A. Zeeb 315a55383e7SHans Petter Selasky CURVNET_SET_QUIET(ifp->if_vnet); 31621f08a07SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 31721f08a07SBjoern A. Zeeb IP6QB_LOCK(bucket); 31821f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 319a55383e7SHans Petter Selasky /* Scan fragment list. */ 32021f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) { 32121f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 322a55383e7SHans Petter Selasky 323f1664f32SBjoern A. Zeeb /* Clear no longer valid rcvif pointer. */ 324da89a0feSBjoern A. Zeeb if (af6->ip6af_m->m_pkthdr.rcvif == ifp) 325da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.rcvif = NULL; 326a55383e7SHans Petter Selasky } 327a55383e7SHans Petter Selasky } 32821f08a07SBjoern A. Zeeb IP6QB_UNLOCK(bucket); 329a55383e7SHans Petter Selasky } 330a55383e7SHans Petter Selasky CURVNET_RESTORE(); 331a55383e7SHans Petter Selasky } 332a55383e7SHans Petter Selasky EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0); 333a55383e7SHans Petter Selasky 334a55383e7SHans Petter Selasky /* 33523d374aaSBjoern A. Zeeb * Like in RFC2460, in RFC8200, fragment and reassembly rules do not agree with 33623d374aaSBjoern A. Zeeb * each other, in terms of next header field handling in fragment header. 337686cdd19SJun-ichiro itojun Hagino * While the sender will use the same value for all of the fragmented packets, 33823d374aaSBjoern A. Zeeb * receiver is suggested not to check for consistency. 339686cdd19SJun-ichiro itojun Hagino * 34023d374aaSBjoern A. Zeeb * Fragment rules (p18,p19): 341686cdd19SJun-ichiro itojun Hagino * (2) A Fragment header containing: 34223d374aaSBjoern A. Zeeb * The Next Header value that identifies the first header 34323d374aaSBjoern A. Zeeb * after the Per-Fragment headers of the original packet. 344686cdd19SJun-ichiro itojun Hagino * -> next header field is same for all fragments 345686cdd19SJun-ichiro itojun Hagino * 34623d374aaSBjoern A. Zeeb * Reassembly rule (p20): 34723d374aaSBjoern A. Zeeb * The Next Header field of the last header of the Per-Fragment 34823d374aaSBjoern A. Zeeb * headers is obtained from the Next Header field of the first 349686cdd19SJun-ichiro itojun Hagino * fragment's Fragment header. 350686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 351686cdd19SJun-ichiro itojun Hagino * 352686cdd19SJun-ichiro itojun Hagino * The following note also contradicts with fragment rule - no one is going to 353686cdd19SJun-ichiro itojun Hagino * send different fragment with different next header field. 354686cdd19SJun-ichiro itojun Hagino * 35523d374aaSBjoern A. Zeeb * Additional note (p22) [not an error]: 356686cdd19SJun-ichiro itojun Hagino * The Next Header values in the Fragment headers of different 357686cdd19SJun-ichiro itojun Hagino * fragments of the same original packet may differ. Only the value 358686cdd19SJun-ichiro itojun Hagino * from the Offset zero fragment packet is used for reassembly. 359686cdd19SJun-ichiro itojun Hagino * -> should grab it from the first fragment only 360686cdd19SJun-ichiro itojun Hagino * 361686cdd19SJun-ichiro itojun Hagino * There is no explicit reason given in the RFC. Historical reason maybe? 362686cdd19SJun-ichiro itojun Hagino */ 363686cdd19SJun-ichiro itojun Hagino /* 36423d374aaSBjoern A. Zeeb * Fragment input. 36582cd038dSYoshinobu Inoue */ 36682cd038dSYoshinobu Inoue int 3671272577eSXin LI frag6_input(struct mbuf **mp, int *offp, int proto) 36882cd038dSYoshinobu Inoue { 36921f08a07SBjoern A. Zeeb struct mbuf *m, *t; 37082cd038dSYoshinobu Inoue struct ip6_hdr *ip6; 37182cd038dSYoshinobu Inoue struct ip6_frag *ip6f; 37221f08a07SBjoern A. Zeeb struct ip6qhead *head; 37321f08a07SBjoern A. Zeeb struct ip6q *q6; 37421f08a07SBjoern A. Zeeb struct ip6asfrag *af6, *ip6af, *af6tmp; 37521f08a07SBjoern A. Zeeb struct in6_ifaddr *ia6; 37621f08a07SBjoern A. Zeeb struct ifnet *dstifp, *srcifp; 377505e91f5SKristof Provost uint32_t hashkey[(sizeof(struct in6_addr) * 2 + 378505e91f5SKristof Provost sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)]; 3799cb1a47aSBjoern A. Zeeb uint32_t bucket, *hashkeyp; 3805778b399SBjoern A. Zeeb int fragoff, frgpartlen; /* Must be larger than uint16_t. */ 3815778b399SBjoern A. Zeeb int nxt, offset, plen; 3825778b399SBjoern A. Zeeb uint8_t ecn, ecn0; 3835778b399SBjoern A. Zeeb bool only_frag; 384aaa46574SAdrian Chadd #ifdef RSS 385aaa46574SAdrian Chadd struct ip6_direct_ctx *ip6dc; 3865778b399SBjoern A. Zeeb struct m_tag *mtag; 387aaa46574SAdrian Chadd #endif 388aaa46574SAdrian Chadd 3895778b399SBjoern A. Zeeb m = *mp; 3905778b399SBjoern A. Zeeb offset = *offp; 3915778b399SBjoern A. Zeeb 39282cd038dSYoshinobu Inoue ip6 = mtod(m, struct ip6_hdr *); 393686cdd19SJun-ichiro itojun Hagino #ifndef PULLDOWN_TEST 394686cdd19SJun-ichiro itojun Hagino IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE); 39582cd038dSYoshinobu Inoue ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); 396686cdd19SJun-ichiro itojun Hagino #else 397686cdd19SJun-ichiro itojun Hagino IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f)); 398686cdd19SJun-ichiro itojun Hagino if (ip6f == NULL) 39940e39bbbSHajimu UMEMOTO return (IPPROTO_DONE); 400686cdd19SJun-ichiro itojun Hagino #endif 40182cd038dSYoshinobu Inoue 40282cd038dSYoshinobu Inoue dstifp = NULL; 40323d374aaSBjoern A. Zeeb /* Find the destination interface of the packet. */ 4045778b399SBjoern A. Zeeb ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 4055778b399SBjoern A. Zeeb if (ia6 != NULL) { 4065778b399SBjoern A. Zeeb dstifp = ia6->ia_ifp; 4075778b399SBjoern A. Zeeb ifa_free(&ia6->ia_ifa); 4088c0fec80SRobert Watson } 40923d374aaSBjoern A. Zeeb 41023d374aaSBjoern A. Zeeb /* Jumbo payload cannot contain a fragment header. */ 41182cd038dSYoshinobu Inoue if (ip6->ip6_plen == 0) { 41282cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); 41382cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 4145778b399SBjoern A. Zeeb return (IPPROTO_DONE); 41582cd038dSYoshinobu Inoue } 41682cd038dSYoshinobu Inoue 41782cd038dSYoshinobu Inoue /* 41823d374aaSBjoern A. Zeeb * Check whether fragment packet's fragment length is a 41923d374aaSBjoern A. Zeeb * multiple of 8 octets (unless it is the last one). 42082cd038dSYoshinobu Inoue * sizeof(struct ip6_frag) == 8 42182cd038dSYoshinobu Inoue * sizeof(struct ip6_hdr) = 40 42282cd038dSYoshinobu Inoue */ 42382cd038dSYoshinobu Inoue if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && 42482cd038dSYoshinobu Inoue (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { 42506cd0a3fSHajimu UMEMOTO icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 426686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_hdr, ip6_plen)); 42782cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 4285778b399SBjoern A. Zeeb return (IPPROTO_DONE); 42982cd038dSYoshinobu Inoue } 43082cd038dSYoshinobu Inoue 4319cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragments); 43282cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_reqd); 43382cd038dSYoshinobu Inoue 43423d374aaSBjoern A. Zeeb /* Offset now points to data portion. */ 43582cd038dSYoshinobu Inoue offset += sizeof(struct ip6_frag); 43682cd038dSYoshinobu Inoue 4374018ea9aSBjoern A. Zeeb /* 4382946a941STom Jones * Handle "atomic" fragments (offset and m bit set to 0) upfront, 43923d374aaSBjoern A. Zeeb * unrelated to any reassembly. Still need to remove the frag hdr. 44023d374aaSBjoern A. Zeeb * See RFC 6946 and section 4.5 of RFC 8200. 4414018ea9aSBjoern A. Zeeb */ 4424018ea9aSBjoern A. Zeeb if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { 4432946a941STom Jones IP6STAT_INC(ip6s_atomicfrags); 44423d374aaSBjoern A. Zeeb /* XXX-BZ handle correctly. */ 4454018ea9aSBjoern A. Zeeb in6_ifstat_inc(dstifp, ifs6_reass_ok); 4464018ea9aSBjoern A. Zeeb *offp = offset; 447a4061289SAndrey V. Elsukov m->m_flags |= M_FRAGMENTED; 4484018ea9aSBjoern A. Zeeb return (ip6f->ip6f_nxt); 4494018ea9aSBjoern A. Zeeb } 4504018ea9aSBjoern A. Zeeb 4515f9f192dSJonathan T. Looney /* Get fragment length and discard 0-byte fragments. */ 4525f9f192dSJonathan T. Looney frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; 4535f9f192dSJonathan T. Looney if (frgpartlen == 0) { 4545f9f192dSJonathan T. Looney icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 4555f9f192dSJonathan T. Looney offsetof(struct ip6_hdr, ip6_plen)); 4565f9f192dSJonathan T. Looney in6_ifstat_inc(dstifp, ifs6_reass_fail); 4575f9f192dSJonathan T. Looney IP6STAT_INC(ip6s_fragdropped); 4585778b399SBjoern A. Zeeb return (IPPROTO_DONE); 4595f9f192dSJonathan T. Looney } 4605f9f192dSJonathan T. Looney 4617715d794SBjoern A. Zeeb /* 4627715d794SBjoern A. Zeeb * Enforce upper bound on number of fragments for the entire system. 4637715d794SBjoern A. Zeeb * If maxfrag is 0, never accept fragments. 4647715d794SBjoern A. Zeeb * If maxfrag is -1, accept all fragments without limitation. 4657715d794SBjoern A. Zeeb */ 4667715d794SBjoern A. Zeeb if (ip6_maxfrags < 0) 4677715d794SBjoern A. Zeeb ; 4687715d794SBjoern A. Zeeb else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags) 4697715d794SBjoern A. Zeeb goto dropfrag2; 4707715d794SBjoern A. Zeeb 47130809ba9SBjoern A. Zeeb /* 47230809ba9SBjoern A. Zeeb * Validate that a full header chain to the ULP is present in the 47330809ba9SBjoern A. Zeeb * packet containing the first fragment as per RFC RFC7112 and 47430809ba9SBjoern A. Zeeb * RFC 8200 pages 18,19: 47530809ba9SBjoern A. Zeeb * The first fragment packet is composed of: 47630809ba9SBjoern A. Zeeb * (3) Extension headers, if any, and the Upper-Layer header. These 47730809ba9SBjoern A. Zeeb * headers must be in the first fragment. ... 47830809ba9SBjoern A. Zeeb */ 47930809ba9SBjoern A. Zeeb fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); 48030809ba9SBjoern A. Zeeb /* XXX TODO. thj has D16851 open for this. */ 48130809ba9SBjoern A. Zeeb /* Send ICMPv6 4,3 in case of violation. */ 48230809ba9SBjoern A. Zeeb 483efdfee93SBjoern A. Zeeb /* Store receive network interface pointer for later. */ 484efdfee93SBjoern A. Zeeb srcifp = m->m_pkthdr.rcvif; 485efdfee93SBjoern A. Zeeb 48623d374aaSBjoern A. Zeeb /* Generate a hash value for fragment bucket selection. */ 48780d7a853SJonathan T. Looney hashkeyp = hashkey; 48880d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr)); 48980d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 49080d7a853SJonathan T. Looney memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr)); 49180d7a853SJonathan T. Looney hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); 49280d7a853SJonathan T. Looney *hashkeyp = ip6f->ip6f_ident; 4939cb1a47aSBjoern A. Zeeb bucket = jenkins_hash32(hashkey, nitems(hashkey), V_ip6qb_hashseed); 4949cb1a47aSBjoern A. Zeeb bucket &= IP6REASS_HMASK; 4959cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 49621f08a07SBjoern A. Zeeb head = IP6QB_HEAD(bucket); 4979888c401SHajimu UMEMOTO 49821f08a07SBjoern A. Zeeb TAILQ_FOREACH(q6, head, ip6q_tq) 49982cd038dSYoshinobu Inoue if (ip6f->ip6f_ident == q6->ip6q_ident && 50082cd038dSYoshinobu Inoue IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && 5014b908c8bSRobert Watson IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) 5024b908c8bSRobert Watson #ifdef MAC 5034b908c8bSRobert Watson && mac_ip6q_match(m, q6) 5044b908c8bSRobert Watson #endif 5054b908c8bSRobert Watson ) 50682cd038dSYoshinobu Inoue break; 50782cd038dSYoshinobu Inoue 5085778b399SBjoern A. Zeeb only_frag = false; 50921f08a07SBjoern A. Zeeb if (q6 == NULL) { 51023d374aaSBjoern A. Zeeb 5115778b399SBjoern A. Zeeb /* A first fragment to arrive creates a reassembly queue. */ 5125778b399SBjoern A. Zeeb only_frag = true; 51382cd038dSYoshinobu Inoue 51482cd038dSYoshinobu Inoue /* 51582cd038dSYoshinobu Inoue * Enforce upper bound on number of fragmented packets 51682cd038dSYoshinobu Inoue * for which we attempt reassembly; 5179888c401SHajimu UMEMOTO * If maxfragpackets is 0, never accept fragments. 5189888c401SHajimu UMEMOTO * If maxfragpackets is -1, accept all fragments without 5199888c401SHajimu UMEMOTO * limitation. 52082cd038dSYoshinobu Inoue */ 521603724d3SBjoern A. Zeeb if (V_ip6_maxfragpackets < 0) 52233841545SHajimu UMEMOTO ; 5239cb1a47aSBjoern A. Zeeb else if (V_ip6qb[bucket].count >= V_ip6_maxfragbucketsize || 5241e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) >= 52580d7a853SJonathan T. Looney (u_int)V_ip6_maxfragpackets) 52633841545SHajimu UMEMOTO goto dropfrag; 52780d7a853SJonathan T. Looney atomic_add_int(&V_frag6_nfragpackets, 1); 52823d374aaSBjoern A. Zeeb 52923d374aaSBjoern A. Zeeb /* Allocate IPv6 fragement packet queue entry. */ 530487a161cSBjoern A. Zeeb q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FRAG6, 531487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 53282cd038dSYoshinobu Inoue if (q6 == NULL) 53382cd038dSYoshinobu Inoue goto dropfrag; 5344b908c8bSRobert Watson #ifdef MAC 5354b908c8bSRobert Watson if (mac_ip6q_init(q6, M_NOWAIT) != 0) { 536487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 5374b908c8bSRobert Watson goto dropfrag; 5384b908c8bSRobert Watson } 5394b908c8bSRobert Watson mac_ip6q_create(m, q6); 5404b908c8bSRobert Watson #endif 54182cd038dSYoshinobu Inoue 54223d374aaSBjoern A. Zeeb /* ip6q_nxt will be filled afterwards, from 1st fragment. */ 54321f08a07SBjoern A. Zeeb TAILQ_INIT(&q6->ip6q_frags); 54482cd038dSYoshinobu Inoue q6->ip6q_ident = ip6f->ip6f_ident; 54582cd038dSYoshinobu Inoue q6->ip6q_ttl = IPV6_FRAGTTL; 54682cd038dSYoshinobu Inoue q6->ip6q_src = ip6->ip6_src; 54782cd038dSYoshinobu Inoue q6->ip6q_dst = ip6->ip6_dst; 5485e9510e3SJINMEI Tatuya q6->ip6q_ecn = 5495e9510e3SJINMEI Tatuya (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 55082cd038dSYoshinobu Inoue q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ 5519888c401SHajimu UMEMOTO 55221f08a07SBjoern A. Zeeb /* Add the fragemented packet to the bucket. */ 55321f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(head, q6, ip6q_tq); 55421f08a07SBjoern A. Zeeb V_ip6qb[bucket].count++; 55582cd038dSYoshinobu Inoue } 55682cd038dSYoshinobu Inoue 55782cd038dSYoshinobu Inoue /* 55823d374aaSBjoern A. Zeeb * If it is the 1st fragment, record the length of the 55982cd038dSYoshinobu Inoue * unfragmentable part and the next header of the fragment header. 56082cd038dSYoshinobu Inoue */ 56182cd038dSYoshinobu Inoue if (fragoff == 0) { 56206cd0a3fSHajimu UMEMOTO q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - 56306cd0a3fSHajimu UMEMOTO sizeof(struct ip6_frag); 56482cd038dSYoshinobu Inoue q6->ip6q_nxt = ip6f->ip6f_nxt; 56582cd038dSYoshinobu Inoue } 56682cd038dSYoshinobu Inoue 56782cd038dSYoshinobu Inoue /* 56882cd038dSYoshinobu Inoue * Check that the reassembled packet would not exceed 65535 bytes 56982cd038dSYoshinobu Inoue * in size. 57082cd038dSYoshinobu Inoue * If it would exceed, discard the fragment and return an ICMP error. 57182cd038dSYoshinobu Inoue */ 57282cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen >= 0) { 57382cd038dSYoshinobu Inoue /* The 1st fragment has already arrived. */ 57482cd038dSYoshinobu Inoue if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { 575e5fffe9aSBjoern A. Zeeb if (only_frag) { 576e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 577e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 578e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 579e5fffe9aSBjoern A. Zeeb #ifdef MAC 580e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 581e5fffe9aSBjoern A. Zeeb #endif 582e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 583e5fffe9aSBjoern A. Zeeb } 584e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 58582cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 586686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 587686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 58882cd038dSYoshinobu Inoue return (IPPROTO_DONE); 58982cd038dSYoshinobu Inoue } 59006cd0a3fSHajimu UMEMOTO } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { 591e5fffe9aSBjoern A. Zeeb if (only_frag) { 592e5fffe9aSBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 593e5fffe9aSBjoern A. Zeeb V_ip6qb[bucket].count--; 594e5fffe9aSBjoern A. Zeeb atomic_subtract_int(&V_frag6_nfragpackets, 1); 595e5fffe9aSBjoern A. Zeeb #ifdef MAC 596e5fffe9aSBjoern A. Zeeb mac_ip6q_destroy(q6); 597e5fffe9aSBjoern A. Zeeb #endif 598e5fffe9aSBjoern A. Zeeb free(q6, M_FRAG6); 599e5fffe9aSBjoern A. Zeeb } 600e5fffe9aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 60182cd038dSYoshinobu Inoue icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, 602686cdd19SJun-ichiro itojun Hagino offset - sizeof(struct ip6_frag) + 603686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 60482cd038dSYoshinobu Inoue return (IPPROTO_DONE); 60582cd038dSYoshinobu Inoue } 606f1664f32SBjoern A. Zeeb 60782cd038dSYoshinobu Inoue /* 60823d374aaSBjoern A. Zeeb * If it is the first fragment, do the above check for each 60982cd038dSYoshinobu Inoue * fragment already stored in the reassembly queue. 61082cd038dSYoshinobu Inoue */ 611dda02192SBjoern A. Zeeb if (fragoff == 0 && !only_frag) { 61221f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(af6, &q6->ip6q_frags, ip6af_tq, af6tmp) { 61382cd038dSYoshinobu Inoue 614dda02192SBjoern A. Zeeb if (q6->ip6q_unfrglen + af6->ip6af_off + 615dda02192SBjoern A. Zeeb af6->ip6af_frglen > IPV6_MAXPACKET) { 61682cd038dSYoshinobu Inoue struct ip6_hdr *ip6err; 6175778b399SBjoern A. Zeeb struct mbuf *merr; 6185778b399SBjoern A. Zeeb int erroff; 6195778b399SBjoern A. Zeeb 620da89a0feSBjoern A. Zeeb merr = af6->ip6af_m; 6215778b399SBjoern A. Zeeb erroff = af6->ip6af_offset; 62282cd038dSYoshinobu Inoue 62323d374aaSBjoern A. Zeeb /* Dequeue the fragment. */ 62421f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 625dda02192SBjoern A. Zeeb q6->ip6q_nfrag--; 626dda02192SBjoern A. Zeeb atomic_subtract_int(&frag6_nfrags, 1); 627487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 62882cd038dSYoshinobu Inoue 629a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 630a55383e7SHans Petter Selasky merr->m_pkthdr.rcvif = srcifp; 631a55383e7SHans Petter Selasky 63223d374aaSBjoern A. Zeeb /* Adjust pointer. */ 63382cd038dSYoshinobu Inoue ip6err = mtod(merr, struct ip6_hdr *); 63482cd038dSYoshinobu Inoue 63582cd038dSYoshinobu Inoue /* 63682cd038dSYoshinobu Inoue * Restore source and destination addresses 63782cd038dSYoshinobu Inoue * in the erroneous IPv6 header. 63882cd038dSYoshinobu Inoue */ 63982cd038dSYoshinobu Inoue ip6err->ip6_src = q6->ip6q_src; 64082cd038dSYoshinobu Inoue ip6err->ip6_dst = q6->ip6q_dst; 64182cd038dSYoshinobu Inoue 64282cd038dSYoshinobu Inoue icmp6_error(merr, ICMP6_PARAM_PROB, 64382cd038dSYoshinobu Inoue ICMP6_PARAMPROB_HEADER, 644686cdd19SJun-ichiro itojun Hagino erroff - sizeof(struct ip6_frag) + 645686cdd19SJun-ichiro itojun Hagino offsetof(struct ip6_frag, ip6f_offlg)); 64682cd038dSYoshinobu Inoue } 64782cd038dSYoshinobu Inoue } 64882cd038dSYoshinobu Inoue } 64982cd038dSYoshinobu Inoue 65023d374aaSBjoern A. Zeeb /* Allocate an IPv6 fragement queue entry for this fragmented part. */ 651487a161cSBjoern A. Zeeb ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FRAG6, 652487a161cSBjoern A. Zeeb M_NOWAIT | M_ZERO); 653686cdd19SJun-ichiro itojun Hagino if (ip6af == NULL) 654686cdd19SJun-ichiro itojun Hagino goto dropfrag; 65521f08a07SBjoern A. Zeeb ip6af->ip6af_mff = (ip6f->ip6f_offlg & IP6F_MORE_FRAG) ? true : false; 65682cd038dSYoshinobu Inoue ip6af->ip6af_off = fragoff; 65782cd038dSYoshinobu Inoue ip6af->ip6af_frglen = frgpartlen; 65882cd038dSYoshinobu Inoue ip6af->ip6af_offset = offset; 659da89a0feSBjoern A. Zeeb ip6af->ip6af_m = m; 66082cd038dSYoshinobu Inoue 6615778b399SBjoern A. Zeeb if (only_frag) { 66221f08a07SBjoern A. Zeeb /* 66321f08a07SBjoern A. Zeeb * Do a manual insert rather than a hard-to-understand cast 66421f08a07SBjoern A. Zeeb * to a different type relying on data structure order to work. 66521f08a07SBjoern A. Zeeb */ 66621f08a07SBjoern A. Zeeb TAILQ_INSERT_HEAD(&q6->ip6q_frags, ip6af, ip6af_tq); 66721f08a07SBjoern A. Zeeb goto postinsert; 66882cd038dSYoshinobu Inoue } 66982cd038dSYoshinobu Inoue 67023d374aaSBjoern A. Zeeb /* Do duplicate, condition, and boundry checks. */ 67182cd038dSYoshinobu Inoue /* 67259dfcba4SHajimu UMEMOTO * Handle ECN by comparing this segment with the first one; 67359dfcba4SHajimu UMEMOTO * if CE is set, do not lose CE. 67423d374aaSBjoern A. Zeeb * Drop if CE and not-ECT are mixed for the same packet. 67559dfcba4SHajimu UMEMOTO */ 67659dfcba4SHajimu UMEMOTO ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; 6775e9510e3SJINMEI Tatuya ecn0 = q6->ip6q_ecn; 67859dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_CE) { 67959dfcba4SHajimu UMEMOTO if (ecn0 == IPTOS_ECN_NOTECT) { 680487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 68159dfcba4SHajimu UMEMOTO goto dropfrag; 68259dfcba4SHajimu UMEMOTO } 68359dfcba4SHajimu UMEMOTO if (ecn0 != IPTOS_ECN_CE) 6845e9510e3SJINMEI Tatuya q6->ip6q_ecn = IPTOS_ECN_CE; 68559dfcba4SHajimu UMEMOTO } 68659dfcba4SHajimu UMEMOTO if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { 687487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 68859dfcba4SHajimu UMEMOTO goto dropfrag; 68959dfcba4SHajimu UMEMOTO } 69059dfcba4SHajimu UMEMOTO 69123d374aaSBjoern A. Zeeb /* Find a fragmented part which begins after this one does. */ 69221f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) 69382cd038dSYoshinobu Inoue if (af6->ip6af_off > ip6af->ip6af_off) 69482cd038dSYoshinobu Inoue break; 69582cd038dSYoshinobu Inoue 69682cd038dSYoshinobu Inoue /* 69782cd038dSYoshinobu Inoue * If the incoming framgent overlaps some existing fragments in 69823d374aaSBjoern A. Zeeb * the reassembly queue, drop both the new fragment and the 69923d374aaSBjoern A. Zeeb * entire reassembly queue. However, if the new fragment 70023d374aaSBjoern A. Zeeb * is an exact duplicate of an existing fragment, only silently 70123d374aaSBjoern A. Zeeb * drop the existing fragment and leave the fragmentation queue 70223d374aaSBjoern A. Zeeb * unchanged, as allowed by the RFC. (RFC 8200, 4.5) 70382cd038dSYoshinobu Inoue */ 70421f08a07SBjoern A. Zeeb if (af6 != NULL) 70521f08a07SBjoern A. Zeeb af6tmp = TAILQ_PREV(af6, ip6fraghead, ip6af_tq); 70621f08a07SBjoern A. Zeeb else 70721f08a07SBjoern A. Zeeb af6tmp = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 70821f08a07SBjoern A. Zeeb if (af6tmp != NULL) { 70921f08a07SBjoern A. Zeeb if (af6tmp->ip6af_off + af6tmp->ip6af_frglen - 7105778b399SBjoern A. Zeeb ip6af->ip6af_off > 0) { 711487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 71282cd038dSYoshinobu Inoue goto dropfrag; 71382cd038dSYoshinobu Inoue } 71482cd038dSYoshinobu Inoue } 71521f08a07SBjoern A. Zeeb if (af6 != NULL) { 7165778b399SBjoern A. Zeeb if (ip6af->ip6af_off + ip6af->ip6af_frglen - 7175778b399SBjoern A. Zeeb af6->ip6af_off > 0) { 718487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 71982cd038dSYoshinobu Inoue goto dropfrag; 72082cd038dSYoshinobu Inoue } 72182cd038dSYoshinobu Inoue } 72282cd038dSYoshinobu Inoue 7234b908c8bSRobert Watson #ifdef MAC 7244b908c8bSRobert Watson mac_ip6q_update(m, q6); 7254b908c8bSRobert Watson #endif 72682cd038dSYoshinobu Inoue 72782cd038dSYoshinobu Inoue /* 72823d374aaSBjoern A. Zeeb * Stick new segment in its place; check for complete reassembly. 72923d374aaSBjoern A. Zeeb * If not complete, check fragment limit. Move to front of packet 73023d374aaSBjoern A. Zeeb * queue, as we are the most recently active fragmented packet. 73182cd038dSYoshinobu Inoue */ 73221f08a07SBjoern A. Zeeb if (af6 != NULL) 73321f08a07SBjoern A. Zeeb TAILQ_INSERT_BEFORE(af6, ip6af, ip6af_tq); 73421f08a07SBjoern A. Zeeb else 73521f08a07SBjoern A. Zeeb TAILQ_INSERT_TAIL(&q6->ip6q_frags, ip6af, ip6af_tq); 73621f08a07SBjoern A. Zeeb postinsert: 7372adfd64fSJonathan T. Looney atomic_add_int(&frag6_nfrags, 1); 7389888c401SHajimu UMEMOTO q6->ip6q_nfrag++; 7393c7165b3SBjoern A. Zeeb 7405778b399SBjoern A. Zeeb plen = 0; 74121f08a07SBjoern A. Zeeb TAILQ_FOREACH(af6, &q6->ip6q_frags, ip6af_tq) { 7425778b399SBjoern A. Zeeb if (af6->ip6af_off != plen) { 74303c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 744198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7459cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 74603c99d76SJonathan T. Looney } 7479cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 7485778b399SBjoern A. Zeeb return (IPPROTO_DONE); 74982cd038dSYoshinobu Inoue } 7505778b399SBjoern A. Zeeb plen += af6->ip6af_frglen; 75182cd038dSYoshinobu Inoue } 75221f08a07SBjoern A. Zeeb af6 = TAILQ_LAST(&q6->ip6q_frags, ip6fraghead); 75321f08a07SBjoern A. Zeeb if (af6->ip6af_mff) { 75403c99d76SJonathan T. Looney if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { 755198fdaedSTom Jones IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); 7569cb1a47aSBjoern A. Zeeb frag6_freef(q6, bucket); 75703c99d76SJonathan T. Looney } 7589cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 7595778b399SBjoern A. Zeeb return (IPPROTO_DONE); 76082cd038dSYoshinobu Inoue } 76182cd038dSYoshinobu Inoue 76223d374aaSBjoern A. Zeeb /* Reassembly is complete; concatenate fragments. */ 76321f08a07SBjoern A. Zeeb ip6af = TAILQ_FIRST(&q6->ip6q_frags); 764da89a0feSBjoern A. Zeeb t = m = ip6af->ip6af_m; 76521f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, ip6af, ip6af_tq); 76621f08a07SBjoern A. Zeeb while ((af6 = TAILQ_FIRST(&q6->ip6q_frags)) != NULL) { 7679907aba3SAndrey V. Elsukov m->m_pkthdr.csum_flags &= 768da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_flags; 7699907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data += 770da89a0feSBjoern A. Zeeb af6->ip6af_m->m_pkthdr.csum_data; 7719907aba3SAndrey V. Elsukov 77221f08a07SBjoern A. Zeeb TAILQ_REMOVE(&q6->ip6q_frags, af6, ip6af_tq); 773efdfee93SBjoern A. Zeeb t = m_last(t); 774da89a0feSBjoern A. Zeeb m_adj(af6->ip6af_m, af6->ip6af_offset); 775da89a0feSBjoern A. Zeeb m_demote_pkthdr(af6->ip6af_m); 776da89a0feSBjoern A. Zeeb m_cat(t, af6->ip6af_m); 777487a161cSBjoern A. Zeeb free(af6, M_FRAG6); 77882cd038dSYoshinobu Inoue } 77982cd038dSYoshinobu Inoue 7809907aba3SAndrey V. Elsukov while (m->m_pkthdr.csum_data & 0xffff0000) 7819907aba3SAndrey V. Elsukov m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 7829907aba3SAndrey V. Elsukov (m->m_pkthdr.csum_data >> 16); 7839907aba3SAndrey V. Elsukov 78423d374aaSBjoern A. Zeeb /* Adjust offset to point where the original next header starts. */ 78582cd038dSYoshinobu Inoue offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); 786487a161cSBjoern A. Zeeb free(ip6af, M_FRAG6); 787686cdd19SJun-ichiro itojun Hagino ip6 = mtod(m, struct ip6_hdr *); 7885778b399SBjoern A. Zeeb ip6->ip6_plen = htons((u_short)plen + offset - sizeof(struct ip6_hdr)); 7895e9510e3SJINMEI Tatuya if (q6->ip6q_ecn == IPTOS_ECN_CE) 7905e9510e3SJINMEI Tatuya ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); 79182cd038dSYoshinobu Inoue nxt = q6->ip6q_nxt; 79282cd038dSYoshinobu Inoue 79321f08a07SBjoern A. Zeeb TAILQ_REMOVE(head, q6, ip6q_tq); 79421f08a07SBjoern A. Zeeb V_ip6qb[bucket].count--; 7952adfd64fSJonathan T. Looney atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); 79621f08a07SBjoern A. Zeeb 79721f08a07SBjoern A. Zeeb if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) { 7984b908c8bSRobert Watson #ifdef MAC 7994b908c8bSRobert Watson mac_ip6q_destroy(q6); 8004b908c8bSRobert Watson #endif 801487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 80280d7a853SJonathan T. Looney atomic_subtract_int(&V_frag6_nfragpackets, 1); 8030b438b0fSGleb Smirnoff 804686cdd19SJun-ichiro itojun Hagino goto dropfrag; 80582cd038dSYoshinobu Inoue } 80682cd038dSYoshinobu Inoue 80723d374aaSBjoern A. Zeeb /* Set nxt(-hdr field value) to the original value. */ 80868e0e5a6SAndrey V. Elsukov m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), 80968e0e5a6SAndrey V. Elsukov (caddr_t)&nxt); 81082cd038dSYoshinobu Inoue 8114b908c8bSRobert Watson #ifdef MAC 8124b908c8bSRobert Watson mac_ip6q_reassemble(q6, m); 8134b908c8bSRobert Watson mac_ip6q_destroy(q6); 8144b908c8bSRobert Watson #endif 815487a161cSBjoern A. Zeeb free(q6, M_FRAG6); 81680d7a853SJonathan T. Looney atomic_subtract_int(&V_frag6_nfragpackets, 1); 81782cd038dSYoshinobu Inoue 81882cd038dSYoshinobu Inoue if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ 8195778b399SBjoern A. Zeeb 8205778b399SBjoern A. Zeeb plen = 0; 82182cd038dSYoshinobu Inoue for (t = m; t; t = t->m_next) 82282cd038dSYoshinobu Inoue plen += t->m_len; 82382cd038dSYoshinobu Inoue m->m_pkthdr.len = plen; 824a55383e7SHans Petter Selasky /* Set a valid receive interface pointer. */ 825a55383e7SHans Petter Selasky m->m_pkthdr.rcvif = srcifp; 82682cd038dSYoshinobu Inoue } 82782cd038dSYoshinobu Inoue 828aaa46574SAdrian Chadd #ifdef RSS 829aaa46574SAdrian Chadd mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), 830aaa46574SAdrian Chadd M_NOWAIT); 831aaa46574SAdrian Chadd if (mtag == NULL) 832aaa46574SAdrian Chadd goto dropfrag; 833aaa46574SAdrian Chadd 834aaa46574SAdrian Chadd ip6dc = (struct ip6_direct_ctx *)(mtag + 1); 835aaa46574SAdrian Chadd ip6dc->ip6dc_nxt = nxt; 836aaa46574SAdrian Chadd ip6dc->ip6dc_off = offset; 837aaa46574SAdrian Chadd 838aaa46574SAdrian Chadd m_tag_prepend(m, mtag); 839aaa46574SAdrian Chadd #endif 840aaa46574SAdrian Chadd 8419cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 8429cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_reassembled); 84382cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_ok); 84482cd038dSYoshinobu Inoue 845aaa46574SAdrian Chadd #ifdef RSS 84623d374aaSBjoern A. Zeeb /* Queue/dispatch for reprocessing. */ 847aaa46574SAdrian Chadd netisr_dispatch(NETISR_IPV6_DIRECT, m); 8485778b399SBjoern A. Zeeb return (IPPROTO_DONE); 849aaa46574SAdrian Chadd #endif 850aaa46574SAdrian Chadd 85123d374aaSBjoern A. Zeeb /* Tell launch routine the next header. */ 85282cd038dSYoshinobu Inoue *mp = m; 85382cd038dSYoshinobu Inoue *offp = offset; 85482cd038dSYoshinobu Inoue 8555778b399SBjoern A. Zeeb return (nxt); 85682cd038dSYoshinobu Inoue 85782cd038dSYoshinobu Inoue dropfrag: 8589cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 8597715d794SBjoern A. Zeeb dropfrag2: 86082cd038dSYoshinobu Inoue in6_ifstat_inc(dstifp, ifs6_reass_fail); 8619cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragdropped); 86282cd038dSYoshinobu Inoue m_freem(m); 8635778b399SBjoern A. Zeeb return (IPPROTO_DONE); 86482cd038dSYoshinobu Inoue } 86582cd038dSYoshinobu Inoue 86682cd038dSYoshinobu Inoue /* 86733841545SHajimu UMEMOTO * IPv6 reassembling timer processing; 86823d374aaSBjoern A. Zeeb * if a timer expires on a reassembly queue, discard it. 86982cd038dSYoshinobu Inoue */ 87082cd038dSYoshinobu Inoue void 8711272577eSXin LI frag6_slowtimo(void) 87282cd038dSYoshinobu Inoue { 8738b615593SMarko Zec VNET_ITERATOR_DECL(vnet_iter); 87421f08a07SBjoern A. Zeeb struct ip6qhead *head; 87521f08a07SBjoern A. Zeeb struct ip6q *q6, *q6tmp; 8769cb1a47aSBjoern A. Zeeb uint32_t bucket; 87782cd038dSYoshinobu Inoue 8785ee847d3SRobert Watson VNET_LIST_RLOCK_NOSLEEP(); 8798b615593SMarko Zec VNET_FOREACH(vnet_iter) { 8808b615593SMarko Zec CURVNET_SET(vnet_iter); 8819cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 8829cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 8839cb1a47aSBjoern A. Zeeb head = IP6QB_HEAD(bucket); 88421f08a07SBjoern A. Zeeb TAILQ_FOREACH_SAFE(q6, head, ip6q_tq, q6tmp) 88521f08a07SBjoern A. Zeeb if (--q6->ip6q_ttl == 0) { 886198fdaedSTom Jones IP6STAT_ADD(ip6s_fragtimeout, 88721f08a07SBjoern A. Zeeb q6->ip6q_nfrag); 88882cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 88921f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 89082cd038dSYoshinobu Inoue } 89182cd038dSYoshinobu Inoue /* 89282cd038dSYoshinobu Inoue * If we are over the maximum number of fragments 89382cd038dSYoshinobu Inoue * (due to the limit being lowered), drain off 89482cd038dSYoshinobu Inoue * enough to get down to the new limit. 8951e9f3b73SJonathan T. Looney * Note that we drain all reassembly queues if 8961e9f3b73SJonathan T. Looney * maxfragpackets is 0 (fragmentation is disabled), 89723d374aaSBjoern A. Zeeb * and do not enforce a limit when maxfragpackets 8981e9f3b73SJonathan T. Looney * is negative. 89982cd038dSYoshinobu Inoue */ 9001e9f3b73SJonathan T. Looney while ((V_ip6_maxfragpackets == 0 || 9011e9f3b73SJonathan T. Looney (V_ip6_maxfragpackets > 0 && 9029cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count > V_ip6_maxfragbucketsize)) && 90321f08a07SBjoern A. Zeeb (q6 = TAILQ_LAST(head, ip6qhead)) != NULL) { 90421f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 90582cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 90621f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 90780d7a853SJonathan T. Looney } 9089cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 90982cd038dSYoshinobu Inoue } 9101e9f3b73SJonathan T. Looney /* 9111e9f3b73SJonathan T. Looney * If we are still over the maximum number of fragmented 9121e9f3b73SJonathan T. Looney * packets, drain off enough to get down to the new limit. 9131e9f3b73SJonathan T. Looney */ 9149cb1a47aSBjoern A. Zeeb bucket = 0; 9151e9f3b73SJonathan T. Looney while (V_ip6_maxfragpackets >= 0 && 9161e9f3b73SJonathan T. Looney atomic_load_int(&V_frag6_nfragpackets) > 9171e9f3b73SJonathan T. Looney (u_int)V_ip6_maxfragpackets) { 9189cb1a47aSBjoern A. Zeeb IP6QB_LOCK(bucket); 91921f08a07SBjoern A. Zeeb q6 = TAILQ_LAST(IP6QB_HEAD(bucket), ip6qhead); 92021f08a07SBjoern A. Zeeb if (q6 != NULL) { 92121f08a07SBjoern A. Zeeb IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_nfrag); 9221e9f3b73SJonathan T. Looney /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 92321f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 9241e9f3b73SJonathan T. Looney } 9259cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 9269cb1a47aSBjoern A. Zeeb bucket = (bucket + 1) % IP6REASS_NHASH; 9271e9f3b73SJonathan T. Looney } 9288b615593SMarko Zec CURVNET_RESTORE(); 9298b615593SMarko Zec } 9305ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 93182cd038dSYoshinobu Inoue } 93282cd038dSYoshinobu Inoue 93323d374aaSBjoern A. Zeeb /* 93423d374aaSBjoern A. Zeeb * Eventhandler to adjust limits in case nmbclusters change. 93523d374aaSBjoern A. Zeeb */ 936c00464a2SBjoern A. Zeeb static void 937c00464a2SBjoern A. Zeeb frag6_change(void *tag) 938c00464a2SBjoern A. Zeeb { 939c00464a2SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 940c00464a2SBjoern A. Zeeb 941c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 942c00464a2SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 943c00464a2SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 944c00464a2SBjoern A. Zeeb CURVNET_SET(vnet_iter); 945c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 946c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 947c00464a2SBjoern A. Zeeb CURVNET_RESTORE(); 948c00464a2SBjoern A. Zeeb } 949c00464a2SBjoern A. Zeeb VNET_LIST_RUNLOCK_NOSLEEP(); 950c00464a2SBjoern A. Zeeb } 951c00464a2SBjoern A. Zeeb 952c00464a2SBjoern A. Zeeb /* 953c00464a2SBjoern A. Zeeb * Initialise reassembly queue and fragment identifier. 954c00464a2SBjoern A. Zeeb */ 955c00464a2SBjoern A. Zeeb void 956c00464a2SBjoern A. Zeeb frag6_init(void) 957c00464a2SBjoern A. Zeeb { 9589cb1a47aSBjoern A. Zeeb uint32_t bucket; 959c00464a2SBjoern A. Zeeb 960c00464a2SBjoern A. Zeeb V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; 961c00464a2SBjoern A. Zeeb frag6_set_bucketsize(); 9629cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 96321f08a07SBjoern A. Zeeb TAILQ_INIT(IP6QB_HEAD(bucket)); 964efdfee93SBjoern A. Zeeb mtx_init(&V_ip6qb[bucket].lock, "ip6qb", NULL, MTX_DEF); 9659cb1a47aSBjoern A. Zeeb V_ip6qb[bucket].count = 0; 966c00464a2SBjoern A. Zeeb } 9679cb1a47aSBjoern A. Zeeb V_ip6qb_hashseed = arc4random(); 968c00464a2SBjoern A. Zeeb V_ip6_maxfragsperpacket = 64; 96967a10c46SBjoern A. Zeeb #ifdef VIMAGE 97067a10c46SBjoern A. Zeeb V_frag6_on = true; 97167a10c46SBjoern A. Zeeb #endif 972c00464a2SBjoern A. Zeeb if (!IS_DEFAULT_VNET(curvnet)) 973c00464a2SBjoern A. Zeeb return; 974c00464a2SBjoern A. Zeeb 975c00464a2SBjoern A. Zeeb ip6_maxfrags = IP6_MAXFRAGS; 976c00464a2SBjoern A. Zeeb EVENTHANDLER_REGISTER(nmbclusters_change, 977c00464a2SBjoern A. Zeeb frag6_change, NULL, EVENTHANDLER_PRI_ANY); 978c00464a2SBjoern A. Zeeb } 979c00464a2SBjoern A. Zeeb 98082cd038dSYoshinobu Inoue /* 98182cd038dSYoshinobu Inoue * Drain off all datagram fragments. 98282cd038dSYoshinobu Inoue */ 98367a10c46SBjoern A. Zeeb static void 98467a10c46SBjoern A. Zeeb frag6_drain_one(void) 98582cd038dSYoshinobu Inoue { 98621f08a07SBjoern A. Zeeb struct ip6q *q6; 9879cb1a47aSBjoern A. Zeeb uint32_t bucket; 9889888c401SHajimu UMEMOTO 9899cb1a47aSBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 99067a10c46SBjoern A. Zeeb IP6QB_LOCK(bucket); 99121f08a07SBjoern A. Zeeb while ((q6 = TAILQ_FIRST(IP6QB_HEAD(bucket))) != NULL) { 9929cb8d207SAndrey V. Elsukov IP6STAT_INC(ip6s_fragdropped); 99382cd038dSYoshinobu Inoue /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ 99421f08a07SBjoern A. Zeeb frag6_freef(q6, bucket); 99580d7a853SJonathan T. Looney } 9969cb1a47aSBjoern A. Zeeb IP6QB_UNLOCK(bucket); 99782cd038dSYoshinobu Inoue } 99867a10c46SBjoern A. Zeeb } 99967a10c46SBjoern A. Zeeb 100067a10c46SBjoern A. Zeeb void 100167a10c46SBjoern A. Zeeb frag6_drain(void) 100267a10c46SBjoern A. Zeeb { 100367a10c46SBjoern A. Zeeb VNET_ITERATOR_DECL(vnet_iter); 100467a10c46SBjoern A. Zeeb 100567a10c46SBjoern A. Zeeb VNET_LIST_RLOCK_NOSLEEP(); 100667a10c46SBjoern A. Zeeb VNET_FOREACH(vnet_iter) { 100767a10c46SBjoern A. Zeeb CURVNET_SET(vnet_iter); 100867a10c46SBjoern A. Zeeb frag6_drain_one(); 10098b615593SMarko Zec CURVNET_RESTORE(); 10108b615593SMarko Zec } 10115ee847d3SRobert Watson VNET_LIST_RUNLOCK_NOSLEEP(); 101282cd038dSYoshinobu Inoue } 1013e5ee7060SGleb Smirnoff 101467a10c46SBjoern A. Zeeb #ifdef VIMAGE 101567a10c46SBjoern A. Zeeb /* 101667a10c46SBjoern A. Zeeb * Clear up IPv6 reassembly structures. 101767a10c46SBjoern A. Zeeb */ 101867a10c46SBjoern A. Zeeb void 101967a10c46SBjoern A. Zeeb frag6_destroy(void) 102067a10c46SBjoern A. Zeeb { 102167a10c46SBjoern A. Zeeb uint32_t bucket; 102267a10c46SBjoern A. Zeeb 102367a10c46SBjoern A. Zeeb frag6_drain_one(); 102467a10c46SBjoern A. Zeeb V_frag6_on = false; 102567a10c46SBjoern A. Zeeb for (bucket = 0; bucket < IP6REASS_NHASH; bucket++) { 102667a10c46SBjoern A. Zeeb KASSERT(V_ip6qb[bucket].count == 0, 102767a10c46SBjoern A. Zeeb ("%s: V_ip6qb[%d] (%p) count not 0 (%d)", __func__, 102867a10c46SBjoern A. Zeeb bucket, &V_ip6qb[bucket], V_ip6qb[bucket].count)); 102967a10c46SBjoern A. Zeeb mtx_destroy(&V_ip6qb[bucket].lock); 103067a10c46SBjoern A. Zeeb } 103167a10c46SBjoern A. Zeeb } 103267a10c46SBjoern A. Zeeb #endif 1033