11dbefcc0SGleb Smirnoff /*- 21dbefcc0SGleb Smirnoff * Copyright (c) 2015 Gleb Smirnoff <glebius@FreeBSD.org> 31dbefcc0SGleb Smirnoff * Copyright (c) 2015 Adrian Chadd <adrian@FreeBSD.org> 41dbefcc0SGleb Smirnoff * Copyright (c) 1982, 1986, 1988, 1993 51dbefcc0SGleb Smirnoff * The Regents of the University of California. All rights reserved. 61dbefcc0SGleb Smirnoff * 71dbefcc0SGleb Smirnoff * Redistribution and use in source and binary forms, with or without 81dbefcc0SGleb Smirnoff * modification, are permitted provided that the following conditions 91dbefcc0SGleb Smirnoff * are met: 101dbefcc0SGleb Smirnoff * 1. Redistributions of source code must retain the above copyright 111dbefcc0SGleb Smirnoff * notice, this list of conditions and the following disclaimer. 121dbefcc0SGleb Smirnoff * 2. Redistributions in binary form must reproduce the above copyright 131dbefcc0SGleb Smirnoff * notice, this list of conditions and the following disclaimer in the 141dbefcc0SGleb Smirnoff * documentation and/or other materials provided with the distribution. 15fbbd9655SWarner Losh * 3. Neither the name of the University nor the names of its contributors 161dbefcc0SGleb Smirnoff * may be used to endorse or promote products derived from this software 171dbefcc0SGleb Smirnoff * without specific prior written permission. 181dbefcc0SGleb Smirnoff * 191dbefcc0SGleb Smirnoff * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 201dbefcc0SGleb Smirnoff * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 211dbefcc0SGleb Smirnoff * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 221dbefcc0SGleb Smirnoff * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 231dbefcc0SGleb Smirnoff * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 241dbefcc0SGleb Smirnoff * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 251dbefcc0SGleb Smirnoff * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 261dbefcc0SGleb Smirnoff * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 271dbefcc0SGleb Smirnoff * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 281dbefcc0SGleb Smirnoff * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 291dbefcc0SGleb Smirnoff * SUCH DAMAGE. 301dbefcc0SGleb Smirnoff * 311dbefcc0SGleb Smirnoff * @(#)ip_input.c 8.2 (Berkeley) 1/4/94 321dbefcc0SGleb Smirnoff */ 331dbefcc0SGleb Smirnoff 341dbefcc0SGleb Smirnoff #include <sys/cdefs.h> 351dbefcc0SGleb Smirnoff __FBSDID("$FreeBSD$"); 361dbefcc0SGleb Smirnoff 371dbefcc0SGleb Smirnoff #include "opt_rss.h" 381dbefcc0SGleb Smirnoff 391dbefcc0SGleb Smirnoff #include <sys/param.h> 401dbefcc0SGleb Smirnoff #include <sys/systm.h> 411dbefcc0SGleb Smirnoff #include <sys/eventhandler.h> 4204f44499SHans Petter Selasky #include <sys/kernel.h> 43c047fd1bSGleb Smirnoff #include <sys/hash.h> 441dbefcc0SGleb Smirnoff #include <sys/mbuf.h> 451dbefcc0SGleb Smirnoff #include <sys/malloc.h> 46ff790bbaSJonathan T. Looney #include <sys/limits.h> 471dbefcc0SGleb Smirnoff #include <sys/lock.h> 481dbefcc0SGleb Smirnoff #include <sys/mutex.h> 491dbefcc0SGleb Smirnoff #include <sys/sysctl.h> 5040eb3896SHans Petter Selasky #include <sys/socket.h> 511dbefcc0SGleb Smirnoff 5240eb3896SHans Petter Selasky #include <net/if.h> 5340eb3896SHans Petter Selasky #include <net/if_var.h> 541dbefcc0SGleb Smirnoff #include <net/rss_config.h> 553e217461SAdrian Chadd #include <net/netisr.h> 561dbefcc0SGleb Smirnoff #include <net/vnet.h> 571dbefcc0SGleb Smirnoff 581dbefcc0SGleb Smirnoff #include <netinet/in.h> 591dbefcc0SGleb Smirnoff #include <netinet/ip.h> 601dbefcc0SGleb Smirnoff #include <netinet/ip_var.h> 611dbefcc0SGleb Smirnoff #include <netinet/in_rss.h> 621dbefcc0SGleb Smirnoff #ifdef MAC 631dbefcc0SGleb Smirnoff #include <security/mac/mac_framework.h> 641dbefcc0SGleb Smirnoff #endif 651dbefcc0SGleb Smirnoff 661dbefcc0SGleb Smirnoff SYSCTL_DECL(_net_inet_ip); 671dbefcc0SGleb Smirnoff 681dbefcc0SGleb Smirnoff /* 691dbefcc0SGleb Smirnoff * Reassembly headers are stored in hash buckets. 701dbefcc0SGleb Smirnoff */ 71a967df1cSJonathan T. Looney #define IPREASS_NHASH_LOG2 10 721dbefcc0SGleb Smirnoff #define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2) 731dbefcc0SGleb Smirnoff #define IPREASS_HMASK (IPREASS_NHASH - 1) 741dbefcc0SGleb Smirnoff 751dbefcc0SGleb Smirnoff struct ipqbucket { 761dbefcc0SGleb Smirnoff TAILQ_HEAD(ipqhead, ipq) head; 771dbefcc0SGleb Smirnoff struct mtx lock; 78ff790bbaSJonathan T. Looney int count; 791dbefcc0SGleb Smirnoff }; 801dbefcc0SGleb Smirnoff 815f901c92SAndrew Turner VNET_DEFINE_STATIC(struct ipqbucket, ipq[IPREASS_NHASH]); 821dbefcc0SGleb Smirnoff #define V_ipq VNET(ipq) 835f901c92SAndrew Turner VNET_DEFINE_STATIC(uint32_t, ipq_hashseed); 84c047fd1bSGleb Smirnoff #define V_ipq_hashseed VNET(ipq_hashseed) 851dbefcc0SGleb Smirnoff 861dbefcc0SGleb Smirnoff #define IPQ_LOCK(i) mtx_lock(&V_ipq[i].lock) 871dbefcc0SGleb Smirnoff #define IPQ_TRYLOCK(i) mtx_trylock(&V_ipq[i].lock) 881dbefcc0SGleb Smirnoff #define IPQ_UNLOCK(i) mtx_unlock(&V_ipq[i].lock) 891dbefcc0SGleb Smirnoff #define IPQ_LOCK_ASSERT(i) mtx_assert(&V_ipq[i].lock, MA_OWNED) 901dbefcc0SGleb Smirnoff 91ff790bbaSJonathan T. Looney VNET_DEFINE_STATIC(int, ipreass_maxbucketsize); 92ff790bbaSJonathan T. Looney #define V_ipreass_maxbucketsize VNET(ipreass_maxbucketsize) 93ff790bbaSJonathan T. Looney 941dbefcc0SGleb Smirnoff void ipreass_init(void); 951dbefcc0SGleb Smirnoff void ipreass_drain(void); 961dbefcc0SGleb Smirnoff void ipreass_slowtimo(void); 971dbefcc0SGleb Smirnoff #ifdef VIMAGE 981dbefcc0SGleb Smirnoff void ipreass_destroy(void); 991dbefcc0SGleb Smirnoff #endif 1001dbefcc0SGleb Smirnoff static int sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS); 101ff790bbaSJonathan T. Looney static int sysctl_maxfragbucketsize(SYSCTL_HANDLER_ARGS); 1021dbefcc0SGleb Smirnoff static void ipreass_zone_change(void *); 1031dbefcc0SGleb Smirnoff static void ipreass_drain_tomax(void); 104ff790bbaSJonathan T. Looney static void ipq_free(struct ipqbucket *, struct ipq *); 1051dbefcc0SGleb Smirnoff static struct ipq * ipq_reuse(int); 1061dbefcc0SGleb Smirnoff 1071dbefcc0SGleb Smirnoff static inline void 108ff790bbaSJonathan T. Looney ipq_timeout(struct ipqbucket *bucket, struct ipq *fp) 1091dbefcc0SGleb Smirnoff { 1101dbefcc0SGleb Smirnoff 1111dbefcc0SGleb Smirnoff IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); 112ff790bbaSJonathan T. Looney ipq_free(bucket, fp); 1131dbefcc0SGleb Smirnoff } 1141dbefcc0SGleb Smirnoff 1151dbefcc0SGleb Smirnoff static inline void 116ff790bbaSJonathan T. Looney ipq_drop(struct ipqbucket *bucket, struct ipq *fp) 1171dbefcc0SGleb Smirnoff { 1181dbefcc0SGleb Smirnoff 1191dbefcc0SGleb Smirnoff IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 120ff790bbaSJonathan T. Looney ipq_free(bucket, fp); 1211dbefcc0SGleb Smirnoff } 1221dbefcc0SGleb Smirnoff 123a967df1cSJonathan T. Looney /* 124a967df1cSJonathan T. Looney * By default, limit the number of IP fragments across all reassembly 125a967df1cSJonathan T. Looney * queues to 1/32 of the total number of mbuf clusters. 126a967df1cSJonathan T. Looney * 127a967df1cSJonathan T. Looney * Limit the total number of reassembly queues per VNET to the 128a967df1cSJonathan T. Looney * IP fragment limit, but ensure the limit will not allow any bucket 129a967df1cSJonathan T. Looney * to grow above 100 items. (The bucket limit is 130a967df1cSJonathan T. Looney * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct 131a967df1cSJonathan T. Looney * multiplier to reach a 100-item limit.) 132a967df1cSJonathan T. Looney * The 100-item limit was chosen as brief testing seems to show that 133a967df1cSJonathan T. Looney * this produces "reasonable" performance on some subset of systems 134a967df1cSJonathan T. Looney * under DoS attack. 135a967df1cSJonathan T. Looney */ 136a967df1cSJonathan T. Looney #define IP_MAXFRAGS (nmbclusters / 32) 137a967df1cSJonathan T. Looney #define IP_MAXFRAGPACKETS (imin(IP_MAXFRAGS, IPREASS_NHASH * 50)) 138a967df1cSJonathan T. Looney 1397b9c5eb0SJonathan T. Looney static int maxfrags; 1407b9c5eb0SJonathan T. Looney static volatile u_int nfrags; 1417b9c5eb0SJonathan T. Looney SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfrags, CTLFLAG_RW, 1427b9c5eb0SJonathan T. Looney &maxfrags, 0, 1437b9c5eb0SJonathan T. Looney "Maximum number of IPv4 fragments allowed across all reassembly queues"); 1447b9c5eb0SJonathan T. Looney SYSCTL_UINT(_net_inet_ip, OID_AUTO, curfrags, CTLFLAG_RD, 1457b9c5eb0SJonathan T. Looney __DEVOLATILE(u_int *, &nfrags), 0, 1467b9c5eb0SJonathan T. Looney "Current number of IPv4 fragments across all reassembly queues"); 1477b9c5eb0SJonathan T. Looney 1485f901c92SAndrew Turner VNET_DEFINE_STATIC(uma_zone_t, ipq_zone); 1491dbefcc0SGleb Smirnoff #define V_ipq_zone VNET(ipq_zone) 1501dbefcc0SGleb Smirnoff SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_VNET | 1511dbefcc0SGleb Smirnoff CTLTYPE_INT | CTLFLAG_RW, NULL, 0, sysctl_maxfragpackets, "I", 1521dbefcc0SGleb Smirnoff "Maximum number of IPv4 fragment reassembly queue entries"); 1531dbefcc0SGleb Smirnoff SYSCTL_UMA_CUR(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_VNET, 1541dbefcc0SGleb Smirnoff &VNET_NAME(ipq_zone), 1551dbefcc0SGleb Smirnoff "Current number of IPv4 fragment reassembly queue entries"); 1561dbefcc0SGleb Smirnoff 1575f901c92SAndrew Turner VNET_DEFINE_STATIC(int, noreass); 1581dbefcc0SGleb Smirnoff #define V_noreass VNET(noreass) 1591dbefcc0SGleb Smirnoff 1605f901c92SAndrew Turner VNET_DEFINE_STATIC(int, maxfragsperpacket); 1611dbefcc0SGleb Smirnoff #define V_maxfragsperpacket VNET(maxfragsperpacket) 1621dbefcc0SGleb Smirnoff SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_VNET | CTLFLAG_RW, 1631dbefcc0SGleb Smirnoff &VNET_NAME(maxfragsperpacket), 0, 1641dbefcc0SGleb Smirnoff "Maximum number of IPv4 fragments allowed per packet"); 165ff790bbaSJonathan T. Looney SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragbucketsize, 166ff790bbaSJonathan T. Looney CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 167ff790bbaSJonathan T. Looney sysctl_maxfragbucketsize, "I", 168ff790bbaSJonathan T. Looney "Maximum number of IPv4 fragment reassembly queue entries per bucket"); 1691dbefcc0SGleb Smirnoff 1701dbefcc0SGleb Smirnoff /* 1711dbefcc0SGleb Smirnoff * Take incoming datagram fragment and try to reassemble it into 1721dbefcc0SGleb Smirnoff * whole datagram. If the argument is the first fragment or one 1731dbefcc0SGleb Smirnoff * in between the function will return NULL and store the mbuf 1741dbefcc0SGleb Smirnoff * in the fragment chain. If the argument is the last fragment 1751dbefcc0SGleb Smirnoff * the packet will be reassembled and the pointer to the new 1761dbefcc0SGleb Smirnoff * mbuf returned for further processing. Only m_tags attached 1771dbefcc0SGleb Smirnoff * to the first packet/fragment are preserved. 1781dbefcc0SGleb Smirnoff * The IP header is *NOT* adjusted out of iplen. 1791dbefcc0SGleb Smirnoff */ 1801dbefcc0SGleb Smirnoff #define M_IP_FRAG M_PROTO9 1811dbefcc0SGleb Smirnoff struct mbuf * 1821dbefcc0SGleb Smirnoff ip_reass(struct mbuf *m) 1831dbefcc0SGleb Smirnoff { 1841dbefcc0SGleb Smirnoff struct ip *ip; 1851dbefcc0SGleb Smirnoff struct mbuf *p, *q, *nq, *t; 1861dbefcc0SGleb Smirnoff struct ipq *fp; 1871dbefcc0SGleb Smirnoff struct ipqhead *head; 1887b9c5eb0SJonathan T. Looney int i, hlen, next, tmpmax; 1891dbefcc0SGleb Smirnoff u_int8_t ecn, ecn0; 1905d9bd455SJonathan T. Looney uint32_t hash, hashkey[3]; 1911dbefcc0SGleb Smirnoff #ifdef RSS 1921dbefcc0SGleb Smirnoff uint32_t rss_hash, rss_type; 1931dbefcc0SGleb Smirnoff #endif 1941dbefcc0SGleb Smirnoff 1951dbefcc0SGleb Smirnoff /* 1961dbefcc0SGleb Smirnoff * If no reassembling or maxfragsperpacket are 0, 1971dbefcc0SGleb Smirnoff * never accept fragments. 1987b9c5eb0SJonathan T. Looney * Also, drop packet if it would exceed the maximum 1997b9c5eb0SJonathan T. Looney * number of fragments. 2001dbefcc0SGleb Smirnoff */ 2017b9c5eb0SJonathan T. Looney tmpmax = maxfrags; 2027b9c5eb0SJonathan T. Looney if (V_noreass == 1 || V_maxfragsperpacket == 0 || 2037b9c5eb0SJonathan T. Looney (tmpmax >= 0 && atomic_load_int(&nfrags) >= (u_int)tmpmax)) { 2041dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragments); 2051dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 2061dbefcc0SGleb Smirnoff m_freem(m); 2071dbefcc0SGleb Smirnoff return (NULL); 2081dbefcc0SGleb Smirnoff } 2091dbefcc0SGleb Smirnoff 2101dbefcc0SGleb Smirnoff ip = mtod(m, struct ip *); 2111dbefcc0SGleb Smirnoff hlen = ip->ip_hl << 2; 2121dbefcc0SGleb Smirnoff 2131dbefcc0SGleb Smirnoff /* 2141dbefcc0SGleb Smirnoff * Adjust ip_len to not reflect header, 2151dbefcc0SGleb Smirnoff * convert offset of this to bytes. 2161dbefcc0SGleb Smirnoff */ 2171dbefcc0SGleb Smirnoff ip->ip_len = htons(ntohs(ip->ip_len) - hlen); 2181dbefcc0SGleb Smirnoff /* 2191dbefcc0SGleb Smirnoff * Make sure that fragments have a data length 2202157f3c3SJonathan T. Looney * that's a non-zero multiple of 8 bytes, unless 2212157f3c3SJonathan T. Looney * this is the last fragment. 2221dbefcc0SGleb Smirnoff */ 2232157f3c3SJonathan T. Looney if (ip->ip_len == htons(0) || 2242157f3c3SJonathan T. Looney ((ip->ip_off & htons(IP_MF)) && (ntohs(ip->ip_len) & 0x7) != 0)) { 2251dbefcc0SGleb Smirnoff IPSTAT_INC(ips_toosmall); /* XXX */ 2261dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 2271dbefcc0SGleb Smirnoff m_freem(m); 2281dbefcc0SGleb Smirnoff return (NULL); 2291dbefcc0SGleb Smirnoff } 2302157f3c3SJonathan T. Looney if (ip->ip_off & htons(IP_MF)) 2311dbefcc0SGleb Smirnoff m->m_flags |= M_IP_FRAG; 2322157f3c3SJonathan T. Looney else 2331dbefcc0SGleb Smirnoff m->m_flags &= ~M_IP_FRAG; 2341dbefcc0SGleb Smirnoff ip->ip_off = htons(ntohs(ip->ip_off) << 3); 2351dbefcc0SGleb Smirnoff 2361dbefcc0SGleb Smirnoff /* 23786af1d02SMark Johnston * Make sure the fragment lies within a packet of valid size. 23886af1d02SMark Johnston */ 23986af1d02SMark Johnston if (ntohs(ip->ip_len) + ntohs(ip->ip_off) > IP_MAXPACKET) { 24086af1d02SMark Johnston IPSTAT_INC(ips_toolong); 24186af1d02SMark Johnston IPSTAT_INC(ips_fragdropped); 24286af1d02SMark Johnston m_freem(m); 24386af1d02SMark Johnston return (NULL); 24486af1d02SMark Johnston } 24586af1d02SMark Johnston 24686af1d02SMark Johnston /* 2471dbefcc0SGleb Smirnoff * Attempt reassembly; if it succeeds, proceed. 2481dbefcc0SGleb Smirnoff * ip_reass() will return a different mbuf. 2491dbefcc0SGleb Smirnoff */ 2501dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragments); 2511dbefcc0SGleb Smirnoff m->m_pkthdr.PH_loc.ptr = ip; 2521dbefcc0SGleb Smirnoff 2531dbefcc0SGleb Smirnoff /* 2541dbefcc0SGleb Smirnoff * Presence of header sizes in mbufs 2551dbefcc0SGleb Smirnoff * would confuse code below. 2561dbefcc0SGleb Smirnoff */ 2571dbefcc0SGleb Smirnoff m->m_data += hlen; 2581dbefcc0SGleb Smirnoff m->m_len -= hlen; 2591dbefcc0SGleb Smirnoff 2605d9bd455SJonathan T. Looney hashkey[0] = ip->ip_src.s_addr; 2615d9bd455SJonathan T. Looney hashkey[1] = ip->ip_dst.s_addr; 2625d9bd455SJonathan T. Looney hashkey[2] = (uint32_t)ip->ip_p << 16; 2635d9bd455SJonathan T. Looney hashkey[2] += ip->ip_id; 2645d9bd455SJonathan T. Looney hash = jenkins_hash32(hashkey, nitems(hashkey), V_ipq_hashseed); 2655d9bd455SJonathan T. Looney hash &= IPREASS_HMASK; 2661dbefcc0SGleb Smirnoff head = &V_ipq[hash].head; 2671dbefcc0SGleb Smirnoff IPQ_LOCK(hash); 2681dbefcc0SGleb Smirnoff 2691dbefcc0SGleb Smirnoff /* 2701dbefcc0SGleb Smirnoff * Look for queue of fragments 2711dbefcc0SGleb Smirnoff * of this datagram. 2721dbefcc0SGleb Smirnoff */ 2731dbefcc0SGleb Smirnoff TAILQ_FOREACH(fp, head, ipq_list) 2741dbefcc0SGleb Smirnoff if (ip->ip_id == fp->ipq_id && 2751dbefcc0SGleb Smirnoff ip->ip_src.s_addr == fp->ipq_src.s_addr && 2761dbefcc0SGleb Smirnoff ip->ip_dst.s_addr == fp->ipq_dst.s_addr && 2771dbefcc0SGleb Smirnoff #ifdef MAC 2781dbefcc0SGleb Smirnoff mac_ipq_match(m, fp) && 2791dbefcc0SGleb Smirnoff #endif 2801dbefcc0SGleb Smirnoff ip->ip_p == fp->ipq_p) 2811dbefcc0SGleb Smirnoff break; 2821dbefcc0SGleb Smirnoff /* 2831dbefcc0SGleb Smirnoff * If first fragment to arrive, create a reassembly queue. 2841dbefcc0SGleb Smirnoff */ 2851dbefcc0SGleb Smirnoff if (fp == NULL) { 286ff790bbaSJonathan T. Looney if (V_ipq[hash].count < V_ipreass_maxbucketsize) 2871dbefcc0SGleb Smirnoff fp = uma_zalloc(V_ipq_zone, M_NOWAIT); 2881dbefcc0SGleb Smirnoff if (fp == NULL) 2891dbefcc0SGleb Smirnoff fp = ipq_reuse(hash); 290ff790bbaSJonathan T. Looney if (fp == NULL) 291ff790bbaSJonathan T. Looney goto dropfrag; 2921dbefcc0SGleb Smirnoff #ifdef MAC 2931dbefcc0SGleb Smirnoff if (mac_ipq_init(fp, M_NOWAIT) != 0) { 2941dbefcc0SGleb Smirnoff uma_zfree(V_ipq_zone, fp); 2951dbefcc0SGleb Smirnoff fp = NULL; 2961dbefcc0SGleb Smirnoff goto dropfrag; 2971dbefcc0SGleb Smirnoff } 2981dbefcc0SGleb Smirnoff mac_ipq_create(m, fp); 2991dbefcc0SGleb Smirnoff #endif 3001dbefcc0SGleb Smirnoff TAILQ_INSERT_HEAD(head, fp, ipq_list); 301ff790bbaSJonathan T. Looney V_ipq[hash].count++; 3021dbefcc0SGleb Smirnoff fp->ipq_nfrags = 1; 3037b9c5eb0SJonathan T. Looney atomic_add_int(&nfrags, 1); 3041dbefcc0SGleb Smirnoff fp->ipq_ttl = IPFRAGTTL; 3051dbefcc0SGleb Smirnoff fp->ipq_p = ip->ip_p; 3061dbefcc0SGleb Smirnoff fp->ipq_id = ip->ip_id; 3071dbefcc0SGleb Smirnoff fp->ipq_src = ip->ip_src; 3081dbefcc0SGleb Smirnoff fp->ipq_dst = ip->ip_dst; 3091dbefcc0SGleb Smirnoff fp->ipq_frags = m; 3102157f3c3SJonathan T. Looney if (m->m_flags & M_IP_FRAG) 3112157f3c3SJonathan T. Looney fp->ipq_maxoff = -1; 3122157f3c3SJonathan T. Looney else 3132157f3c3SJonathan T. Looney fp->ipq_maxoff = ntohs(ip->ip_off) + ntohs(ip->ip_len); 3141dbefcc0SGleb Smirnoff m->m_nextpkt = NULL; 3151dbefcc0SGleb Smirnoff goto done; 3161dbefcc0SGleb Smirnoff } else { 3172157f3c3SJonathan T. Looney /* 3182157f3c3SJonathan T. Looney * If we already saw the last fragment, make sure 3192157f3c3SJonathan T. Looney * this fragment's offset looks sane. Otherwise, if 3202157f3c3SJonathan T. Looney * this is the last fragment, record its endpoint. 3212157f3c3SJonathan T. Looney */ 3222157f3c3SJonathan T. Looney if (fp->ipq_maxoff > 0) { 3232157f3c3SJonathan T. Looney i = ntohs(ip->ip_off) + ntohs(ip->ip_len); 3242157f3c3SJonathan T. Looney if (((m->m_flags & M_IP_FRAG) && i >= fp->ipq_maxoff) || 3252157f3c3SJonathan T. Looney ((m->m_flags & M_IP_FRAG) == 0 && 3262157f3c3SJonathan T. Looney i != fp->ipq_maxoff)) { 3272157f3c3SJonathan T. Looney fp = NULL; 3282157f3c3SJonathan T. Looney goto dropfrag; 3292157f3c3SJonathan T. Looney } 3302157f3c3SJonathan T. Looney } else if ((m->m_flags & M_IP_FRAG) == 0) 3312157f3c3SJonathan T. Looney fp->ipq_maxoff = ntohs(ip->ip_off) + ntohs(ip->ip_len); 3321dbefcc0SGleb Smirnoff fp->ipq_nfrags++; 3337b9c5eb0SJonathan T. Looney atomic_add_int(&nfrags, 1); 3341dbefcc0SGleb Smirnoff #ifdef MAC 3351dbefcc0SGleb Smirnoff mac_ipq_update(m, fp); 3361dbefcc0SGleb Smirnoff #endif 3371dbefcc0SGleb Smirnoff } 3381dbefcc0SGleb Smirnoff 3391dbefcc0SGleb Smirnoff #define GETIP(m) ((struct ip*)((m)->m_pkthdr.PH_loc.ptr)) 3401dbefcc0SGleb Smirnoff 3411dbefcc0SGleb Smirnoff /* 3421dbefcc0SGleb Smirnoff * Handle ECN by comparing this segment with the first one; 3431dbefcc0SGleb Smirnoff * if CE is set, do not lose CE. 3441dbefcc0SGleb Smirnoff * drop if CE and not-ECT are mixed for the same packet. 3451dbefcc0SGleb Smirnoff */ 3461dbefcc0SGleb Smirnoff ecn = ip->ip_tos & IPTOS_ECN_MASK; 3471dbefcc0SGleb Smirnoff ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; 3481dbefcc0SGleb Smirnoff if (ecn == IPTOS_ECN_CE) { 3491dbefcc0SGleb Smirnoff if (ecn0 == IPTOS_ECN_NOTECT) 3501dbefcc0SGleb Smirnoff goto dropfrag; 3511dbefcc0SGleb Smirnoff if (ecn0 != IPTOS_ECN_CE) 3521dbefcc0SGleb Smirnoff GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; 3531dbefcc0SGleb Smirnoff } 3541dbefcc0SGleb Smirnoff if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) 3551dbefcc0SGleb Smirnoff goto dropfrag; 3561dbefcc0SGleb Smirnoff 3571dbefcc0SGleb Smirnoff /* 3581dbefcc0SGleb Smirnoff * Find a segment which begins after this one does. 3591dbefcc0SGleb Smirnoff */ 3601dbefcc0SGleb Smirnoff for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) 3611dbefcc0SGleb Smirnoff if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off)) 3621dbefcc0SGleb Smirnoff break; 3631dbefcc0SGleb Smirnoff 3641dbefcc0SGleb Smirnoff /* 3651dbefcc0SGleb Smirnoff * If there is a preceding segment, it may provide some of 3661dbefcc0SGleb Smirnoff * our data already. If so, drop the data from the incoming 3671dbefcc0SGleb Smirnoff * segment. If it provides all of our data, drop us, otherwise 3681dbefcc0SGleb Smirnoff * stick new segment in the proper place. 3691dbefcc0SGleb Smirnoff * 3701dbefcc0SGleb Smirnoff * If some of the data is dropped from the preceding 3711dbefcc0SGleb Smirnoff * segment, then it's checksum is invalidated. 3721dbefcc0SGleb Smirnoff */ 3731dbefcc0SGleb Smirnoff if (p) { 3741dbefcc0SGleb Smirnoff i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) - 3751dbefcc0SGleb Smirnoff ntohs(ip->ip_off); 3761dbefcc0SGleb Smirnoff if (i > 0) { 3771dbefcc0SGleb Smirnoff if (i >= ntohs(ip->ip_len)) 3781dbefcc0SGleb Smirnoff goto dropfrag; 3791dbefcc0SGleb Smirnoff m_adj(m, i); 3801dbefcc0SGleb Smirnoff m->m_pkthdr.csum_flags = 0; 3811dbefcc0SGleb Smirnoff ip->ip_off = htons(ntohs(ip->ip_off) + i); 3821dbefcc0SGleb Smirnoff ip->ip_len = htons(ntohs(ip->ip_len) - i); 3831dbefcc0SGleb Smirnoff } 3841dbefcc0SGleb Smirnoff m->m_nextpkt = p->m_nextpkt; 3851dbefcc0SGleb Smirnoff p->m_nextpkt = m; 3861dbefcc0SGleb Smirnoff } else { 3871dbefcc0SGleb Smirnoff m->m_nextpkt = fp->ipq_frags; 3881dbefcc0SGleb Smirnoff fp->ipq_frags = m; 3891dbefcc0SGleb Smirnoff } 3901dbefcc0SGleb Smirnoff 3911dbefcc0SGleb Smirnoff /* 3921dbefcc0SGleb Smirnoff * While we overlap succeeding segments trim them or, 3931dbefcc0SGleb Smirnoff * if they are completely covered, dequeue them. 3941dbefcc0SGleb Smirnoff */ 3951dbefcc0SGleb Smirnoff for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) > 3961dbefcc0SGleb Smirnoff ntohs(GETIP(q)->ip_off); q = nq) { 3971dbefcc0SGleb Smirnoff i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) - 3981dbefcc0SGleb Smirnoff ntohs(GETIP(q)->ip_off); 3991dbefcc0SGleb Smirnoff if (i < ntohs(GETIP(q)->ip_len)) { 4001dbefcc0SGleb Smirnoff GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i); 4011dbefcc0SGleb Smirnoff GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i); 4021dbefcc0SGleb Smirnoff m_adj(q, i); 4031dbefcc0SGleb Smirnoff q->m_pkthdr.csum_flags = 0; 4041dbefcc0SGleb Smirnoff break; 4051dbefcc0SGleb Smirnoff } 4061dbefcc0SGleb Smirnoff nq = q->m_nextpkt; 4071dbefcc0SGleb Smirnoff m->m_nextpkt = nq; 4081dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 4091dbefcc0SGleb Smirnoff fp->ipq_nfrags--; 4107b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, 1); 4111dbefcc0SGleb Smirnoff m_freem(q); 4121dbefcc0SGleb Smirnoff } 4131dbefcc0SGleb Smirnoff 4141dbefcc0SGleb Smirnoff /* 4151dbefcc0SGleb Smirnoff * Check for complete reassembly and perform frag per packet 4161dbefcc0SGleb Smirnoff * limiting. 4171dbefcc0SGleb Smirnoff * 4181dbefcc0SGleb Smirnoff * Frag limiting is performed here so that the nth frag has 4191dbefcc0SGleb Smirnoff * a chance to complete the packet before we drop the packet. 4201dbefcc0SGleb Smirnoff * As a result, n+1 frags are actually allowed per packet, but 4211dbefcc0SGleb Smirnoff * only n will ever be stored. (n = maxfragsperpacket.) 4221dbefcc0SGleb Smirnoff * 4231dbefcc0SGleb Smirnoff */ 4241dbefcc0SGleb Smirnoff next = 0; 4251dbefcc0SGleb Smirnoff for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { 4261dbefcc0SGleb Smirnoff if (ntohs(GETIP(q)->ip_off) != next) { 4271dbefcc0SGleb Smirnoff if (fp->ipq_nfrags > V_maxfragsperpacket) 428ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[hash], fp); 4291dbefcc0SGleb Smirnoff goto done; 4301dbefcc0SGleb Smirnoff } 4311dbefcc0SGleb Smirnoff next += ntohs(GETIP(q)->ip_len); 4321dbefcc0SGleb Smirnoff } 4331dbefcc0SGleb Smirnoff /* Make sure the last packet didn't have the IP_MF flag */ 4341dbefcc0SGleb Smirnoff if (p->m_flags & M_IP_FRAG) { 4351dbefcc0SGleb Smirnoff if (fp->ipq_nfrags > V_maxfragsperpacket) 436ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[hash], fp); 4371dbefcc0SGleb Smirnoff goto done; 4381dbefcc0SGleb Smirnoff } 4391dbefcc0SGleb Smirnoff 4401dbefcc0SGleb Smirnoff /* 4411dbefcc0SGleb Smirnoff * Reassembly is complete. Make sure the packet is a sane size. 4421dbefcc0SGleb Smirnoff */ 4431dbefcc0SGleb Smirnoff q = fp->ipq_frags; 4441dbefcc0SGleb Smirnoff ip = GETIP(q); 4451dbefcc0SGleb Smirnoff if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { 4461dbefcc0SGleb Smirnoff IPSTAT_INC(ips_toolong); 447ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[hash], fp); 4481dbefcc0SGleb Smirnoff goto done; 4491dbefcc0SGleb Smirnoff } 4501dbefcc0SGleb Smirnoff 4511dbefcc0SGleb Smirnoff /* 4521dbefcc0SGleb Smirnoff * Concatenate fragments. 4531dbefcc0SGleb Smirnoff */ 4541dbefcc0SGleb Smirnoff m = q; 4551dbefcc0SGleb Smirnoff t = m->m_next; 4561dbefcc0SGleb Smirnoff m->m_next = NULL; 4571dbefcc0SGleb Smirnoff m_cat(m, t); 4581dbefcc0SGleb Smirnoff nq = q->m_nextpkt; 4591dbefcc0SGleb Smirnoff q->m_nextpkt = NULL; 4601dbefcc0SGleb Smirnoff for (q = nq; q != NULL; q = nq) { 4611dbefcc0SGleb Smirnoff nq = q->m_nextpkt; 4621dbefcc0SGleb Smirnoff q->m_nextpkt = NULL; 4631dbefcc0SGleb Smirnoff m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; 4641dbefcc0SGleb Smirnoff m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; 46509b0b8c0SNavdeep Parhar m_demote_pkthdr(q); 4661dbefcc0SGleb Smirnoff m_cat(m, q); 4671dbefcc0SGleb Smirnoff } 4681dbefcc0SGleb Smirnoff /* 4691dbefcc0SGleb Smirnoff * In order to do checksumming faster we do 'end-around carry' here 4701dbefcc0SGleb Smirnoff * (and not in for{} loop), though it implies we are not going to 4711dbefcc0SGleb Smirnoff * reassemble more than 64k fragments. 4721dbefcc0SGleb Smirnoff */ 4731dbefcc0SGleb Smirnoff while (m->m_pkthdr.csum_data & 0xffff0000) 4741dbefcc0SGleb Smirnoff m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 4751dbefcc0SGleb Smirnoff (m->m_pkthdr.csum_data >> 16); 4767b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, fp->ipq_nfrags); 4771dbefcc0SGleb Smirnoff #ifdef MAC 4781dbefcc0SGleb Smirnoff mac_ipq_reassemble(fp, m); 4791dbefcc0SGleb Smirnoff mac_ipq_destroy(fp); 4801dbefcc0SGleb Smirnoff #endif 4811dbefcc0SGleb Smirnoff 4821dbefcc0SGleb Smirnoff /* 4831dbefcc0SGleb Smirnoff * Create header for new ip packet by modifying header of first 4841dbefcc0SGleb Smirnoff * packet; dequeue and discard fragment reassembly header. 4851dbefcc0SGleb Smirnoff * Make header visible. 4861dbefcc0SGleb Smirnoff */ 4871dbefcc0SGleb Smirnoff ip->ip_len = htons((ip->ip_hl << 2) + next); 4881dbefcc0SGleb Smirnoff ip->ip_src = fp->ipq_src; 4891dbefcc0SGleb Smirnoff ip->ip_dst = fp->ipq_dst; 4901dbefcc0SGleb Smirnoff TAILQ_REMOVE(head, fp, ipq_list); 491ff790bbaSJonathan T. Looney V_ipq[hash].count--; 4921dbefcc0SGleb Smirnoff uma_zfree(V_ipq_zone, fp); 4931dbefcc0SGleb Smirnoff m->m_len += (ip->ip_hl << 2); 4941dbefcc0SGleb Smirnoff m->m_data -= (ip->ip_hl << 2); 4951dbefcc0SGleb Smirnoff /* some debugging cruft by sklower, below, will go away soon */ 4961dbefcc0SGleb Smirnoff if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ 4971dbefcc0SGleb Smirnoff m_fixhdr(m); 4981dbefcc0SGleb Smirnoff IPSTAT_INC(ips_reassembled); 4991dbefcc0SGleb Smirnoff IPQ_UNLOCK(hash); 5001dbefcc0SGleb Smirnoff 5011dbefcc0SGleb Smirnoff #ifdef RSS 5021dbefcc0SGleb Smirnoff /* 5031dbefcc0SGleb Smirnoff * Query the RSS layer for the flowid / flowtype for the 5041dbefcc0SGleb Smirnoff * mbuf payload. 5051dbefcc0SGleb Smirnoff * 5061dbefcc0SGleb Smirnoff * For now, just assume we have to calculate a new one. 5071dbefcc0SGleb Smirnoff * Later on we should check to see if the assigned flowid matches 5081dbefcc0SGleb Smirnoff * what RSS wants for the given IP protocol and if so, just keep it. 5091dbefcc0SGleb Smirnoff * 5101dbefcc0SGleb Smirnoff * We then queue into the relevant netisr so it can be dispatched 5111dbefcc0SGleb Smirnoff * to the correct CPU. 5121dbefcc0SGleb Smirnoff * 5131dbefcc0SGleb Smirnoff * Note - this may return 1, which means the flowid in the mbuf 5141dbefcc0SGleb Smirnoff * is correct for the configured RSS hash types and can be used. 5151dbefcc0SGleb Smirnoff */ 5161dbefcc0SGleb Smirnoff if (rss_mbuf_software_hash_v4(m, 0, &rss_hash, &rss_type) == 0) { 5171dbefcc0SGleb Smirnoff m->m_pkthdr.flowid = rss_hash; 5181dbefcc0SGleb Smirnoff M_HASHTYPE_SET(m, rss_type); 5191dbefcc0SGleb Smirnoff } 5201dbefcc0SGleb Smirnoff 5211dbefcc0SGleb Smirnoff /* 5221dbefcc0SGleb Smirnoff * Queue/dispatch for reprocessing. 5231dbefcc0SGleb Smirnoff * 5241dbefcc0SGleb Smirnoff * Note: this is much slower than just handling the frame in the 5251dbefcc0SGleb Smirnoff * current receive context. It's likely worth investigating 5261dbefcc0SGleb Smirnoff * why this is. 5271dbefcc0SGleb Smirnoff */ 5281dbefcc0SGleb Smirnoff netisr_dispatch(NETISR_IP_DIRECT, m); 5291dbefcc0SGleb Smirnoff return (NULL); 5301dbefcc0SGleb Smirnoff #endif 5311dbefcc0SGleb Smirnoff 5321dbefcc0SGleb Smirnoff /* Handle in-line */ 5331dbefcc0SGleb Smirnoff return (m); 5341dbefcc0SGleb Smirnoff 5351dbefcc0SGleb Smirnoff dropfrag: 5361dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 5377b9c5eb0SJonathan T. Looney if (fp != NULL) { 5381dbefcc0SGleb Smirnoff fp->ipq_nfrags--; 5397b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, 1); 5407b9c5eb0SJonathan T. Looney } 5411dbefcc0SGleb Smirnoff m_freem(m); 5421dbefcc0SGleb Smirnoff done: 5431dbefcc0SGleb Smirnoff IPQ_UNLOCK(hash); 5441dbefcc0SGleb Smirnoff return (NULL); 5451dbefcc0SGleb Smirnoff 5461dbefcc0SGleb Smirnoff #undef GETIP 5471dbefcc0SGleb Smirnoff } 5481dbefcc0SGleb Smirnoff 5491dbefcc0SGleb Smirnoff /* 5501dbefcc0SGleb Smirnoff * Initialize IP reassembly structures. 5511dbefcc0SGleb Smirnoff */ 5521dbefcc0SGleb Smirnoff void 5531dbefcc0SGleb Smirnoff ipreass_init(void) 5541dbefcc0SGleb Smirnoff { 555ff790bbaSJonathan T. Looney int max; 5561dbefcc0SGleb Smirnoff 5571dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 5581dbefcc0SGleb Smirnoff TAILQ_INIT(&V_ipq[i].head); 5591dbefcc0SGleb Smirnoff mtx_init(&V_ipq[i].lock, "IP reassembly", NULL, 5601dbefcc0SGleb Smirnoff MTX_DEF | MTX_DUPOK); 561ff790bbaSJonathan T. Looney V_ipq[i].count = 0; 5621dbefcc0SGleb Smirnoff } 563c047fd1bSGleb Smirnoff V_ipq_hashseed = arc4random(); 5641dbefcc0SGleb Smirnoff V_maxfragsperpacket = 16; 5651dbefcc0SGleb Smirnoff V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, 5661dbefcc0SGleb Smirnoff NULL, UMA_ALIGN_PTR, 0); 567a967df1cSJonathan T. Looney max = IP_MAXFRAGPACKETS; 568ff790bbaSJonathan T. Looney max = uma_zone_set_max(V_ipq_zone, max); 569ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); 5701dbefcc0SGleb Smirnoff 5717b9c5eb0SJonathan T. Looney if (IS_DEFAULT_VNET(curvnet)) { 572a967df1cSJonathan T. Looney maxfrags = IP_MAXFRAGS; 5731dbefcc0SGleb Smirnoff EVENTHANDLER_REGISTER(nmbclusters_change, ipreass_zone_change, 5741dbefcc0SGleb Smirnoff NULL, EVENTHANDLER_PRI_ANY); 5751dbefcc0SGleb Smirnoff } 5767b9c5eb0SJonathan T. Looney } 5771dbefcc0SGleb Smirnoff 5781dbefcc0SGleb Smirnoff /* 5791dbefcc0SGleb Smirnoff * If a timer expires on a reassembly queue, discard it. 5801dbefcc0SGleb Smirnoff */ 5811dbefcc0SGleb Smirnoff void 5821dbefcc0SGleb Smirnoff ipreass_slowtimo(void) 5831dbefcc0SGleb Smirnoff { 5841dbefcc0SGleb Smirnoff struct ipq *fp, *tmp; 5851dbefcc0SGleb Smirnoff 5861dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 5871dbefcc0SGleb Smirnoff IPQ_LOCK(i); 5881dbefcc0SGleb Smirnoff TAILQ_FOREACH_SAFE(fp, &V_ipq[i].head, ipq_list, tmp) 5891dbefcc0SGleb Smirnoff if (--fp->ipq_ttl == 0) 590ff790bbaSJonathan T. Looney ipq_timeout(&V_ipq[i], fp); 5911dbefcc0SGleb Smirnoff IPQ_UNLOCK(i); 5921dbefcc0SGleb Smirnoff } 5931dbefcc0SGleb Smirnoff } 5941dbefcc0SGleb Smirnoff 5951dbefcc0SGleb Smirnoff /* 5961dbefcc0SGleb Smirnoff * Drain off all datagram fragments. 5971dbefcc0SGleb Smirnoff */ 5981dbefcc0SGleb Smirnoff void 5991dbefcc0SGleb Smirnoff ipreass_drain(void) 6001dbefcc0SGleb Smirnoff { 6011dbefcc0SGleb Smirnoff 6021dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 6031dbefcc0SGleb Smirnoff IPQ_LOCK(i); 6041dbefcc0SGleb Smirnoff while(!TAILQ_EMPTY(&V_ipq[i].head)) 605ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[i], TAILQ_FIRST(&V_ipq[i].head)); 606ff790bbaSJonathan T. Looney KASSERT(V_ipq[i].count == 0, 607ff790bbaSJonathan T. Looney ("%s: V_ipq[%d] count %d (V_ipq=%p)", __func__, i, 608ff790bbaSJonathan T. Looney V_ipq[i].count, V_ipq)); 6091dbefcc0SGleb Smirnoff IPQ_UNLOCK(i); 6101dbefcc0SGleb Smirnoff } 6111dbefcc0SGleb Smirnoff } 6121dbefcc0SGleb Smirnoff 61340eb3896SHans Petter Selasky /* 61440eb3896SHans Petter Selasky * Drain off all datagram fragments belonging to 61540eb3896SHans Petter Selasky * the given network interface. 61640eb3896SHans Petter Selasky */ 61740eb3896SHans Petter Selasky static void 61840eb3896SHans Petter Selasky ipreass_cleanup(void *arg __unused, struct ifnet *ifp) 61940eb3896SHans Petter Selasky { 62040eb3896SHans Petter Selasky struct ipq *fp, *temp; 62140eb3896SHans Petter Selasky struct mbuf *m; 62240eb3896SHans Petter Selasky int i; 62340eb3896SHans Petter Selasky 62440eb3896SHans Petter Selasky KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); 62540eb3896SHans Petter Selasky 62640eb3896SHans Petter Selasky CURVNET_SET_QUIET(ifp->if_vnet); 62740eb3896SHans Petter Selasky for (i = 0; i < IPREASS_NHASH; i++) { 62840eb3896SHans Petter Selasky IPQ_LOCK(i); 62940eb3896SHans Petter Selasky /* Scan fragment list. */ 63040eb3896SHans Petter Selasky TAILQ_FOREACH_SAFE(fp, &V_ipq[i].head, ipq_list, temp) { 63140eb3896SHans Petter Selasky for (m = fp->ipq_frags; m != NULL; m = m->m_nextpkt) { 63240eb3896SHans Petter Selasky if (m->m_pkthdr.rcvif == ifp) { 63340eb3896SHans Petter Selasky ipq_drop(&V_ipq[i], fp); 63440eb3896SHans Petter Selasky break; 63540eb3896SHans Petter Selasky } 63640eb3896SHans Petter Selasky } 63740eb3896SHans Petter Selasky } 63840eb3896SHans Petter Selasky IPQ_UNLOCK(i); 63940eb3896SHans Petter Selasky } 64040eb3896SHans Petter Selasky CURVNET_RESTORE(); 64140eb3896SHans Petter Selasky } 64240eb3896SHans Petter Selasky EVENTHANDLER_DEFINE(ifnet_departure_event, ipreass_cleanup, NULL, 0); 64340eb3896SHans Petter Selasky 6441dbefcc0SGleb Smirnoff #ifdef VIMAGE 6451dbefcc0SGleb Smirnoff /* 6461dbefcc0SGleb Smirnoff * Destroy IP reassembly structures. 6471dbefcc0SGleb Smirnoff */ 6481dbefcc0SGleb Smirnoff void 6491dbefcc0SGleb Smirnoff ipreass_destroy(void) 6501dbefcc0SGleb Smirnoff { 6511dbefcc0SGleb Smirnoff 6521dbefcc0SGleb Smirnoff ipreass_drain(); 6531dbefcc0SGleb Smirnoff uma_zdestroy(V_ipq_zone); 6541dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) 6551dbefcc0SGleb Smirnoff mtx_destroy(&V_ipq[i].lock); 6561dbefcc0SGleb Smirnoff } 6571dbefcc0SGleb Smirnoff #endif 6581dbefcc0SGleb Smirnoff 6591dbefcc0SGleb Smirnoff /* 6601dbefcc0SGleb Smirnoff * After maxnipq has been updated, propagate the change to UMA. The UMA zone 6611dbefcc0SGleb Smirnoff * max has slightly different semantics than the sysctl, for historical 6621dbefcc0SGleb Smirnoff * reasons. 6631dbefcc0SGleb Smirnoff */ 6641dbefcc0SGleb Smirnoff static void 6651dbefcc0SGleb Smirnoff ipreass_drain_tomax(void) 6661dbefcc0SGleb Smirnoff { 667ff790bbaSJonathan T. Looney struct ipq *fp; 6681dbefcc0SGleb Smirnoff int target; 6691dbefcc0SGleb Smirnoff 6701dbefcc0SGleb Smirnoff /* 671ff790bbaSJonathan T. Looney * Make sure each bucket is under the new limit. If 672ff790bbaSJonathan T. Looney * necessary, drop enough of the oldest elements from 673ff790bbaSJonathan T. Looney * each bucket to get under the new limit. 674ff790bbaSJonathan T. Looney */ 675ff790bbaSJonathan T. Looney for (int i = 0; i < IPREASS_NHASH; i++) { 676ff790bbaSJonathan T. Looney IPQ_LOCK(i); 677ff790bbaSJonathan T. Looney while (V_ipq[i].count > V_ipreass_maxbucketsize && 678ff790bbaSJonathan T. Looney (fp = TAILQ_LAST(&V_ipq[i].head, ipqhead)) != NULL) 679ff790bbaSJonathan T. Looney ipq_timeout(&V_ipq[i], fp); 680ff790bbaSJonathan T. Looney IPQ_UNLOCK(i); 681ff790bbaSJonathan T. Looney } 682ff790bbaSJonathan T. Looney 683ff790bbaSJonathan T. Looney /* 6841dbefcc0SGleb Smirnoff * If we are over the maximum number of fragments, 6851dbefcc0SGleb Smirnoff * drain off enough to get down to the new limit, 6861dbefcc0SGleb Smirnoff * stripping off last elements on queues. Every 6871dbefcc0SGleb Smirnoff * run we strip the oldest element from each bucket. 6881dbefcc0SGleb Smirnoff */ 6891dbefcc0SGleb Smirnoff target = uma_zone_get_max(V_ipq_zone); 6901dbefcc0SGleb Smirnoff while (uma_zone_get_cur(V_ipq_zone) > target) { 6911dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 6921dbefcc0SGleb Smirnoff IPQ_LOCK(i); 6931dbefcc0SGleb Smirnoff fp = TAILQ_LAST(&V_ipq[i].head, ipqhead); 6941dbefcc0SGleb Smirnoff if (fp != NULL) 695ff790bbaSJonathan T. Looney ipq_timeout(&V_ipq[i], fp); 6961dbefcc0SGleb Smirnoff IPQ_UNLOCK(i); 6971dbefcc0SGleb Smirnoff } 6981dbefcc0SGleb Smirnoff } 6991dbefcc0SGleb Smirnoff } 7001dbefcc0SGleb Smirnoff 7011dbefcc0SGleb Smirnoff static void 7021dbefcc0SGleb Smirnoff ipreass_zone_change(void *tag) 7031dbefcc0SGleb Smirnoff { 7047b9c5eb0SJonathan T. Looney VNET_ITERATOR_DECL(vnet_iter); 7057b9c5eb0SJonathan T. Looney int max; 7061dbefcc0SGleb Smirnoff 707a967df1cSJonathan T. Looney maxfrags = IP_MAXFRAGS; 708a967df1cSJonathan T. Looney max = IP_MAXFRAGPACKETS; 7097b9c5eb0SJonathan T. Looney VNET_LIST_RLOCK_NOSLEEP(); 7107b9c5eb0SJonathan T. Looney VNET_FOREACH(vnet_iter) { 7117b9c5eb0SJonathan T. Looney CURVNET_SET(vnet_iter); 712ff790bbaSJonathan T. Looney max = uma_zone_set_max(V_ipq_zone, max); 713ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); 7141dbefcc0SGleb Smirnoff ipreass_drain_tomax(); 7157b9c5eb0SJonathan T. Looney CURVNET_RESTORE(); 7167b9c5eb0SJonathan T. Looney } 7177b9c5eb0SJonathan T. Looney VNET_LIST_RUNLOCK_NOSLEEP(); 7181dbefcc0SGleb Smirnoff } 7191dbefcc0SGleb Smirnoff 7201dbefcc0SGleb Smirnoff /* 7211dbefcc0SGleb Smirnoff * Change the limit on the UMA zone, or disable the fragment allocation 7221dbefcc0SGleb Smirnoff * at all. Since 0 and -1 is a special values here, we need our own handler, 7231dbefcc0SGleb Smirnoff * instead of sysctl_handle_uma_zone_max(). 7241dbefcc0SGleb Smirnoff */ 7251dbefcc0SGleb Smirnoff static int 7261dbefcc0SGleb Smirnoff sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS) 7271dbefcc0SGleb Smirnoff { 7281dbefcc0SGleb Smirnoff int error, max; 7291dbefcc0SGleb Smirnoff 7301dbefcc0SGleb Smirnoff if (V_noreass == 0) { 7311dbefcc0SGleb Smirnoff max = uma_zone_get_max(V_ipq_zone); 7321dbefcc0SGleb Smirnoff if (max == 0) 7331dbefcc0SGleb Smirnoff max = -1; 7341dbefcc0SGleb Smirnoff } else 7351dbefcc0SGleb Smirnoff max = 0; 7361dbefcc0SGleb Smirnoff error = sysctl_handle_int(oidp, &max, 0, req); 7371dbefcc0SGleb Smirnoff if (error || !req->newptr) 7381dbefcc0SGleb Smirnoff return (error); 7391dbefcc0SGleb Smirnoff if (max > 0) { 7401dbefcc0SGleb Smirnoff /* 7411dbefcc0SGleb Smirnoff * XXXRW: Might be a good idea to sanity check the argument 7421dbefcc0SGleb Smirnoff * and place an extreme upper bound. 7431dbefcc0SGleb Smirnoff */ 7441dbefcc0SGleb Smirnoff max = uma_zone_set_max(V_ipq_zone, max); 745ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); 7461dbefcc0SGleb Smirnoff ipreass_drain_tomax(); 7471dbefcc0SGleb Smirnoff V_noreass = 0; 7481dbefcc0SGleb Smirnoff } else if (max == 0) { 7491dbefcc0SGleb Smirnoff V_noreass = 1; 7501dbefcc0SGleb Smirnoff ipreass_drain(); 7511dbefcc0SGleb Smirnoff } else if (max == -1) { 7521dbefcc0SGleb Smirnoff V_noreass = 0; 7531dbefcc0SGleb Smirnoff uma_zone_set_max(V_ipq_zone, 0); 754ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = INT_MAX; 7551dbefcc0SGleb Smirnoff } else 7561dbefcc0SGleb Smirnoff return (EINVAL); 7571dbefcc0SGleb Smirnoff return (0); 7581dbefcc0SGleb Smirnoff } 7591dbefcc0SGleb Smirnoff 7601dbefcc0SGleb Smirnoff /* 7611dbefcc0SGleb Smirnoff * Seek for old fragment queue header that can be reused. Try to 7621dbefcc0SGleb Smirnoff * reuse a header from currently locked hash bucket. 7631dbefcc0SGleb Smirnoff */ 7641dbefcc0SGleb Smirnoff static struct ipq * 7651dbefcc0SGleb Smirnoff ipq_reuse(int start) 7661dbefcc0SGleb Smirnoff { 7671dbefcc0SGleb Smirnoff struct ipq *fp; 768ff790bbaSJonathan T. Looney int bucket, i; 7691dbefcc0SGleb Smirnoff 7701dbefcc0SGleb Smirnoff IPQ_LOCK_ASSERT(start); 7711dbefcc0SGleb Smirnoff 772ff790bbaSJonathan T. Looney for (i = 0; i < IPREASS_NHASH; i++) { 773ff790bbaSJonathan T. Looney bucket = (start + i) % IPREASS_NHASH; 774ff790bbaSJonathan T. Looney if (bucket != start && IPQ_TRYLOCK(bucket) == 0) 7751dbefcc0SGleb Smirnoff continue; 776ff790bbaSJonathan T. Looney fp = TAILQ_LAST(&V_ipq[bucket].head, ipqhead); 7771dbefcc0SGleb Smirnoff if (fp) { 7781dbefcc0SGleb Smirnoff struct mbuf *m; 7791dbefcc0SGleb Smirnoff 7801dbefcc0SGleb Smirnoff IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); 7817b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, fp->ipq_nfrags); 7821dbefcc0SGleb Smirnoff while (fp->ipq_frags) { 7831dbefcc0SGleb Smirnoff m = fp->ipq_frags; 7841dbefcc0SGleb Smirnoff fp->ipq_frags = m->m_nextpkt; 7851dbefcc0SGleb Smirnoff m_freem(m); 7861dbefcc0SGleb Smirnoff } 787ff790bbaSJonathan T. Looney TAILQ_REMOVE(&V_ipq[bucket].head, fp, ipq_list); 788ff790bbaSJonathan T. Looney V_ipq[bucket].count--; 789ff790bbaSJonathan T. Looney if (bucket != start) 790ff790bbaSJonathan T. Looney IPQ_UNLOCK(bucket); 791ff790bbaSJonathan T. Looney break; 792ff790bbaSJonathan T. Looney } 793ff790bbaSJonathan T. Looney if (bucket != start) 794ff790bbaSJonathan T. Looney IPQ_UNLOCK(bucket); 795ff790bbaSJonathan T. Looney } 7961dbefcc0SGleb Smirnoff IPQ_LOCK_ASSERT(start); 7971dbefcc0SGleb Smirnoff return (fp); 7981dbefcc0SGleb Smirnoff } 7991dbefcc0SGleb Smirnoff 8001dbefcc0SGleb Smirnoff /* 8011dbefcc0SGleb Smirnoff * Free a fragment reassembly header and all associated datagrams. 8021dbefcc0SGleb Smirnoff */ 8031dbefcc0SGleb Smirnoff static void 804ff790bbaSJonathan T. Looney ipq_free(struct ipqbucket *bucket, struct ipq *fp) 8051dbefcc0SGleb Smirnoff { 8061dbefcc0SGleb Smirnoff struct mbuf *q; 8071dbefcc0SGleb Smirnoff 8087b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, fp->ipq_nfrags); 8091dbefcc0SGleb Smirnoff while (fp->ipq_frags) { 8101dbefcc0SGleb Smirnoff q = fp->ipq_frags; 8111dbefcc0SGleb Smirnoff fp->ipq_frags = q->m_nextpkt; 8121dbefcc0SGleb Smirnoff m_freem(q); 8131dbefcc0SGleb Smirnoff } 814ff790bbaSJonathan T. Looney TAILQ_REMOVE(&bucket->head, fp, ipq_list); 815ff790bbaSJonathan T. Looney bucket->count--; 8161dbefcc0SGleb Smirnoff uma_zfree(V_ipq_zone, fp); 8171dbefcc0SGleb Smirnoff } 818ff790bbaSJonathan T. Looney 819ff790bbaSJonathan T. Looney /* 820ff790bbaSJonathan T. Looney * Get or set the maximum number of reassembly queues per bucket. 821ff790bbaSJonathan T. Looney */ 822ff790bbaSJonathan T. Looney static int 823ff790bbaSJonathan T. Looney sysctl_maxfragbucketsize(SYSCTL_HANDLER_ARGS) 824ff790bbaSJonathan T. Looney { 825ff790bbaSJonathan T. Looney int error, max; 826ff790bbaSJonathan T. Looney 827ff790bbaSJonathan T. Looney max = V_ipreass_maxbucketsize; 828ff790bbaSJonathan T. Looney error = sysctl_handle_int(oidp, &max, 0, req); 829ff790bbaSJonathan T. Looney if (error || !req->newptr) 830ff790bbaSJonathan T. Looney return (error); 831ff790bbaSJonathan T. Looney if (max <= 0) 832ff790bbaSJonathan T. Looney return (EINVAL); 833ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = max; 834ff790bbaSJonathan T. Looney ipreass_drain_tomax(); 835ff790bbaSJonathan T. Looney return (0); 836ff790bbaSJonathan T. Looney } 837