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> 42c047fd1bSGleb Smirnoff #include <sys/hash.h> 431dbefcc0SGleb Smirnoff #include <sys/mbuf.h> 441dbefcc0SGleb Smirnoff #include <sys/malloc.h> 45ff790bbaSJonathan T. Looney #include <sys/limits.h> 461dbefcc0SGleb Smirnoff #include <sys/lock.h> 471dbefcc0SGleb Smirnoff #include <sys/mutex.h> 481dbefcc0SGleb Smirnoff #include <sys/sysctl.h> 491dbefcc0SGleb Smirnoff 501dbefcc0SGleb Smirnoff #include <net/rss_config.h> 513e217461SAdrian Chadd #include <net/netisr.h> 521dbefcc0SGleb Smirnoff #include <net/vnet.h> 531dbefcc0SGleb Smirnoff 541dbefcc0SGleb Smirnoff #include <netinet/in.h> 551dbefcc0SGleb Smirnoff #include <netinet/ip.h> 561dbefcc0SGleb Smirnoff #include <netinet/ip_var.h> 571dbefcc0SGleb Smirnoff #include <netinet/in_rss.h> 581dbefcc0SGleb Smirnoff #ifdef MAC 591dbefcc0SGleb Smirnoff #include <security/mac/mac_framework.h> 601dbefcc0SGleb Smirnoff #endif 611dbefcc0SGleb Smirnoff 621dbefcc0SGleb Smirnoff SYSCTL_DECL(_net_inet_ip); 631dbefcc0SGleb Smirnoff 641dbefcc0SGleb Smirnoff /* 651dbefcc0SGleb Smirnoff * Reassembly headers are stored in hash buckets. 661dbefcc0SGleb Smirnoff */ 67a967df1cSJonathan T. Looney #define IPREASS_NHASH_LOG2 10 681dbefcc0SGleb Smirnoff #define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2) 691dbefcc0SGleb Smirnoff #define IPREASS_HMASK (IPREASS_NHASH - 1) 701dbefcc0SGleb Smirnoff 711dbefcc0SGleb Smirnoff struct ipqbucket { 721dbefcc0SGleb Smirnoff TAILQ_HEAD(ipqhead, ipq) head; 731dbefcc0SGleb Smirnoff struct mtx lock; 74ff790bbaSJonathan T. Looney int count; 751dbefcc0SGleb Smirnoff }; 761dbefcc0SGleb Smirnoff 775f901c92SAndrew Turner VNET_DEFINE_STATIC(struct ipqbucket, ipq[IPREASS_NHASH]); 781dbefcc0SGleb Smirnoff #define V_ipq VNET(ipq) 795f901c92SAndrew Turner VNET_DEFINE_STATIC(uint32_t, ipq_hashseed); 80c047fd1bSGleb Smirnoff #define V_ipq_hashseed VNET(ipq_hashseed) 811dbefcc0SGleb Smirnoff 821dbefcc0SGleb Smirnoff #define IPQ_LOCK(i) mtx_lock(&V_ipq[i].lock) 831dbefcc0SGleb Smirnoff #define IPQ_TRYLOCK(i) mtx_trylock(&V_ipq[i].lock) 841dbefcc0SGleb Smirnoff #define IPQ_UNLOCK(i) mtx_unlock(&V_ipq[i].lock) 851dbefcc0SGleb Smirnoff #define IPQ_LOCK_ASSERT(i) mtx_assert(&V_ipq[i].lock, MA_OWNED) 861dbefcc0SGleb Smirnoff 87ff790bbaSJonathan T. Looney VNET_DEFINE_STATIC(int, ipreass_maxbucketsize); 88ff790bbaSJonathan T. Looney #define V_ipreass_maxbucketsize VNET(ipreass_maxbucketsize) 89ff790bbaSJonathan T. Looney 901dbefcc0SGleb Smirnoff void ipreass_init(void); 911dbefcc0SGleb Smirnoff void ipreass_drain(void); 921dbefcc0SGleb Smirnoff void ipreass_slowtimo(void); 931dbefcc0SGleb Smirnoff #ifdef VIMAGE 941dbefcc0SGleb Smirnoff void ipreass_destroy(void); 951dbefcc0SGleb Smirnoff #endif 961dbefcc0SGleb Smirnoff static int sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS); 97ff790bbaSJonathan T. Looney static int sysctl_maxfragbucketsize(SYSCTL_HANDLER_ARGS); 981dbefcc0SGleb Smirnoff static void ipreass_zone_change(void *); 991dbefcc0SGleb Smirnoff static void ipreass_drain_tomax(void); 100ff790bbaSJonathan T. Looney static void ipq_free(struct ipqbucket *, struct ipq *); 1011dbefcc0SGleb Smirnoff static struct ipq * ipq_reuse(int); 1021dbefcc0SGleb Smirnoff 1031dbefcc0SGleb Smirnoff static inline void 104ff790bbaSJonathan T. Looney ipq_timeout(struct ipqbucket *bucket, struct ipq *fp) 1051dbefcc0SGleb Smirnoff { 1061dbefcc0SGleb Smirnoff 1071dbefcc0SGleb Smirnoff IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); 108ff790bbaSJonathan T. Looney ipq_free(bucket, fp); 1091dbefcc0SGleb Smirnoff } 1101dbefcc0SGleb Smirnoff 1111dbefcc0SGleb Smirnoff static inline void 112ff790bbaSJonathan T. Looney ipq_drop(struct ipqbucket *bucket, struct ipq *fp) 1131dbefcc0SGleb Smirnoff { 1141dbefcc0SGleb Smirnoff 1151dbefcc0SGleb Smirnoff IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); 116ff790bbaSJonathan T. Looney ipq_free(bucket, fp); 1171dbefcc0SGleb Smirnoff } 1181dbefcc0SGleb Smirnoff 119a967df1cSJonathan T. Looney /* 120a967df1cSJonathan T. Looney * By default, limit the number of IP fragments across all reassembly 121a967df1cSJonathan T. Looney * queues to 1/32 of the total number of mbuf clusters. 122a967df1cSJonathan T. Looney * 123a967df1cSJonathan T. Looney * Limit the total number of reassembly queues per VNET to the 124a967df1cSJonathan T. Looney * IP fragment limit, but ensure the limit will not allow any bucket 125a967df1cSJonathan T. Looney * to grow above 100 items. (The bucket limit is 126a967df1cSJonathan T. Looney * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct 127a967df1cSJonathan T. Looney * multiplier to reach a 100-item limit.) 128a967df1cSJonathan T. Looney * The 100-item limit was chosen as brief testing seems to show that 129a967df1cSJonathan T. Looney * this produces "reasonable" performance on some subset of systems 130a967df1cSJonathan T. Looney * under DoS attack. 131a967df1cSJonathan T. Looney */ 132a967df1cSJonathan T. Looney #define IP_MAXFRAGS (nmbclusters / 32) 133a967df1cSJonathan T. Looney #define IP_MAXFRAGPACKETS (imin(IP_MAXFRAGS, IPREASS_NHASH * 50)) 134a967df1cSJonathan T. Looney 1357b9c5eb0SJonathan T. Looney static int maxfrags; 1367b9c5eb0SJonathan T. Looney static volatile u_int nfrags; 1377b9c5eb0SJonathan T. Looney SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfrags, CTLFLAG_RW, 1387b9c5eb0SJonathan T. Looney &maxfrags, 0, 1397b9c5eb0SJonathan T. Looney "Maximum number of IPv4 fragments allowed across all reassembly queues"); 1407b9c5eb0SJonathan T. Looney SYSCTL_UINT(_net_inet_ip, OID_AUTO, curfrags, CTLFLAG_RD, 1417b9c5eb0SJonathan T. Looney __DEVOLATILE(u_int *, &nfrags), 0, 1427b9c5eb0SJonathan T. Looney "Current number of IPv4 fragments across all reassembly queues"); 1437b9c5eb0SJonathan T. Looney 1445f901c92SAndrew Turner VNET_DEFINE_STATIC(uma_zone_t, ipq_zone); 1451dbefcc0SGleb Smirnoff #define V_ipq_zone VNET(ipq_zone) 1461dbefcc0SGleb Smirnoff SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_VNET | 1471dbefcc0SGleb Smirnoff CTLTYPE_INT | CTLFLAG_RW, NULL, 0, sysctl_maxfragpackets, "I", 1481dbefcc0SGleb Smirnoff "Maximum number of IPv4 fragment reassembly queue entries"); 1491dbefcc0SGleb Smirnoff SYSCTL_UMA_CUR(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_VNET, 1501dbefcc0SGleb Smirnoff &VNET_NAME(ipq_zone), 1511dbefcc0SGleb Smirnoff "Current number of IPv4 fragment reassembly queue entries"); 1521dbefcc0SGleb Smirnoff 1535f901c92SAndrew Turner VNET_DEFINE_STATIC(int, noreass); 1541dbefcc0SGleb Smirnoff #define V_noreass VNET(noreass) 1551dbefcc0SGleb Smirnoff 1565f901c92SAndrew Turner VNET_DEFINE_STATIC(int, maxfragsperpacket); 1571dbefcc0SGleb Smirnoff #define V_maxfragsperpacket VNET(maxfragsperpacket) 1581dbefcc0SGleb Smirnoff SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_VNET | CTLFLAG_RW, 1591dbefcc0SGleb Smirnoff &VNET_NAME(maxfragsperpacket), 0, 1601dbefcc0SGleb Smirnoff "Maximum number of IPv4 fragments allowed per packet"); 161ff790bbaSJonathan T. Looney SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragbucketsize, 162ff790bbaSJonathan T. Looney CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, 163ff790bbaSJonathan T. Looney sysctl_maxfragbucketsize, "I", 164ff790bbaSJonathan T. Looney "Maximum number of IPv4 fragment reassembly queue entries per bucket"); 1651dbefcc0SGleb Smirnoff 1661dbefcc0SGleb Smirnoff /* 1671dbefcc0SGleb Smirnoff * Take incoming datagram fragment and try to reassemble it into 1681dbefcc0SGleb Smirnoff * whole datagram. If the argument is the first fragment or one 1691dbefcc0SGleb Smirnoff * in between the function will return NULL and store the mbuf 1701dbefcc0SGleb Smirnoff * in the fragment chain. If the argument is the last fragment 1711dbefcc0SGleb Smirnoff * the packet will be reassembled and the pointer to the new 1721dbefcc0SGleb Smirnoff * mbuf returned for further processing. Only m_tags attached 1731dbefcc0SGleb Smirnoff * to the first packet/fragment are preserved. 1741dbefcc0SGleb Smirnoff * The IP header is *NOT* adjusted out of iplen. 1751dbefcc0SGleb Smirnoff */ 1761dbefcc0SGleb Smirnoff #define M_IP_FRAG M_PROTO9 1771dbefcc0SGleb Smirnoff struct mbuf * 1781dbefcc0SGleb Smirnoff ip_reass(struct mbuf *m) 1791dbefcc0SGleb Smirnoff { 1801dbefcc0SGleb Smirnoff struct ip *ip; 1811dbefcc0SGleb Smirnoff struct mbuf *p, *q, *nq, *t; 1821dbefcc0SGleb Smirnoff struct ipq *fp; 1831dbefcc0SGleb Smirnoff struct ipqhead *head; 1847b9c5eb0SJonathan T. Looney int i, hlen, next, tmpmax; 1851dbefcc0SGleb Smirnoff u_int8_t ecn, ecn0; 1865d9bd455SJonathan T. Looney uint32_t hash, hashkey[3]; 1871dbefcc0SGleb Smirnoff #ifdef RSS 1881dbefcc0SGleb Smirnoff uint32_t rss_hash, rss_type; 1891dbefcc0SGleb Smirnoff #endif 1901dbefcc0SGleb Smirnoff 1911dbefcc0SGleb Smirnoff /* 1921dbefcc0SGleb Smirnoff * If no reassembling or maxfragsperpacket are 0, 1931dbefcc0SGleb Smirnoff * never accept fragments. 1947b9c5eb0SJonathan T. Looney * Also, drop packet if it would exceed the maximum 1957b9c5eb0SJonathan T. Looney * number of fragments. 1961dbefcc0SGleb Smirnoff */ 1977b9c5eb0SJonathan T. Looney tmpmax = maxfrags; 1987b9c5eb0SJonathan T. Looney if (V_noreass == 1 || V_maxfragsperpacket == 0 || 1997b9c5eb0SJonathan T. Looney (tmpmax >= 0 && atomic_load_int(&nfrags) >= (u_int)tmpmax)) { 2001dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragments); 2011dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 2021dbefcc0SGleb Smirnoff m_freem(m); 2031dbefcc0SGleb Smirnoff return (NULL); 2041dbefcc0SGleb Smirnoff } 2051dbefcc0SGleb Smirnoff 2061dbefcc0SGleb Smirnoff ip = mtod(m, struct ip *); 2071dbefcc0SGleb Smirnoff hlen = ip->ip_hl << 2; 2081dbefcc0SGleb Smirnoff 2091dbefcc0SGleb Smirnoff /* 2101dbefcc0SGleb Smirnoff * Adjust ip_len to not reflect header, 2111dbefcc0SGleb Smirnoff * convert offset of this to bytes. 2121dbefcc0SGleb Smirnoff */ 2131dbefcc0SGleb Smirnoff ip->ip_len = htons(ntohs(ip->ip_len) - hlen); 2141dbefcc0SGleb Smirnoff if (ip->ip_off & htons(IP_MF)) { 2151dbefcc0SGleb Smirnoff /* 2161dbefcc0SGleb Smirnoff * Make sure that fragments have a data length 2171dbefcc0SGleb Smirnoff * that's a non-zero multiple of 8 bytes. 2181dbefcc0SGleb Smirnoff */ 2191dbefcc0SGleb Smirnoff if (ip->ip_len == htons(0) || (ntohs(ip->ip_len) & 0x7) != 0) { 2201dbefcc0SGleb Smirnoff IPSTAT_INC(ips_toosmall); /* XXX */ 2211dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 2221dbefcc0SGleb Smirnoff m_freem(m); 2231dbefcc0SGleb Smirnoff return (NULL); 2241dbefcc0SGleb Smirnoff } 2251dbefcc0SGleb Smirnoff m->m_flags |= M_IP_FRAG; 2261dbefcc0SGleb Smirnoff } else 2271dbefcc0SGleb Smirnoff m->m_flags &= ~M_IP_FRAG; 2281dbefcc0SGleb Smirnoff ip->ip_off = htons(ntohs(ip->ip_off) << 3); 2291dbefcc0SGleb Smirnoff 2301dbefcc0SGleb Smirnoff /* 2311dbefcc0SGleb Smirnoff * Attempt reassembly; if it succeeds, proceed. 2321dbefcc0SGleb Smirnoff * ip_reass() will return a different mbuf. 2331dbefcc0SGleb Smirnoff */ 2341dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragments); 2351dbefcc0SGleb Smirnoff m->m_pkthdr.PH_loc.ptr = ip; 2361dbefcc0SGleb Smirnoff 2371dbefcc0SGleb Smirnoff /* 2381dbefcc0SGleb Smirnoff * Presence of header sizes in mbufs 2391dbefcc0SGleb Smirnoff * would confuse code below. 2401dbefcc0SGleb Smirnoff */ 2411dbefcc0SGleb Smirnoff m->m_data += hlen; 2421dbefcc0SGleb Smirnoff m->m_len -= hlen; 2431dbefcc0SGleb Smirnoff 2445d9bd455SJonathan T. Looney hashkey[0] = ip->ip_src.s_addr; 2455d9bd455SJonathan T. Looney hashkey[1] = ip->ip_dst.s_addr; 2465d9bd455SJonathan T. Looney hashkey[2] = (uint32_t)ip->ip_p << 16; 2475d9bd455SJonathan T. Looney hashkey[2] += ip->ip_id; 2485d9bd455SJonathan T. Looney hash = jenkins_hash32(hashkey, nitems(hashkey), V_ipq_hashseed); 2495d9bd455SJonathan T. Looney hash &= IPREASS_HMASK; 2501dbefcc0SGleb Smirnoff head = &V_ipq[hash].head; 2511dbefcc0SGleb Smirnoff IPQ_LOCK(hash); 2521dbefcc0SGleb Smirnoff 2531dbefcc0SGleb Smirnoff /* 2541dbefcc0SGleb Smirnoff * Look for queue of fragments 2551dbefcc0SGleb Smirnoff * of this datagram. 2561dbefcc0SGleb Smirnoff */ 2571dbefcc0SGleb Smirnoff TAILQ_FOREACH(fp, head, ipq_list) 2581dbefcc0SGleb Smirnoff if (ip->ip_id == fp->ipq_id && 2591dbefcc0SGleb Smirnoff ip->ip_src.s_addr == fp->ipq_src.s_addr && 2601dbefcc0SGleb Smirnoff ip->ip_dst.s_addr == fp->ipq_dst.s_addr && 2611dbefcc0SGleb Smirnoff #ifdef MAC 2621dbefcc0SGleb Smirnoff mac_ipq_match(m, fp) && 2631dbefcc0SGleb Smirnoff #endif 2641dbefcc0SGleb Smirnoff ip->ip_p == fp->ipq_p) 2651dbefcc0SGleb Smirnoff break; 2661dbefcc0SGleb Smirnoff /* 2671dbefcc0SGleb Smirnoff * If first fragment to arrive, create a reassembly queue. 2681dbefcc0SGleb Smirnoff */ 2691dbefcc0SGleb Smirnoff if (fp == NULL) { 270ff790bbaSJonathan T. Looney if (V_ipq[hash].count < V_ipreass_maxbucketsize) 2711dbefcc0SGleb Smirnoff fp = uma_zalloc(V_ipq_zone, M_NOWAIT); 2721dbefcc0SGleb Smirnoff if (fp == NULL) 2731dbefcc0SGleb Smirnoff fp = ipq_reuse(hash); 274ff790bbaSJonathan T. Looney if (fp == NULL) 275ff790bbaSJonathan T. Looney goto dropfrag; 2761dbefcc0SGleb Smirnoff #ifdef MAC 2771dbefcc0SGleb Smirnoff if (mac_ipq_init(fp, M_NOWAIT) != 0) { 2781dbefcc0SGleb Smirnoff uma_zfree(V_ipq_zone, fp); 2791dbefcc0SGleb Smirnoff fp = NULL; 2801dbefcc0SGleb Smirnoff goto dropfrag; 2811dbefcc0SGleb Smirnoff } 2821dbefcc0SGleb Smirnoff mac_ipq_create(m, fp); 2831dbefcc0SGleb Smirnoff #endif 2841dbefcc0SGleb Smirnoff TAILQ_INSERT_HEAD(head, fp, ipq_list); 285ff790bbaSJonathan T. Looney V_ipq[hash].count++; 2861dbefcc0SGleb Smirnoff fp->ipq_nfrags = 1; 2877b9c5eb0SJonathan T. Looney atomic_add_int(&nfrags, 1); 2881dbefcc0SGleb Smirnoff fp->ipq_ttl = IPFRAGTTL; 2891dbefcc0SGleb Smirnoff fp->ipq_p = ip->ip_p; 2901dbefcc0SGleb Smirnoff fp->ipq_id = ip->ip_id; 2911dbefcc0SGleb Smirnoff fp->ipq_src = ip->ip_src; 2921dbefcc0SGleb Smirnoff fp->ipq_dst = ip->ip_dst; 2931dbefcc0SGleb Smirnoff fp->ipq_frags = m; 2941dbefcc0SGleb Smirnoff m->m_nextpkt = NULL; 2951dbefcc0SGleb Smirnoff goto done; 2961dbefcc0SGleb Smirnoff } else { 2971dbefcc0SGleb Smirnoff fp->ipq_nfrags++; 2987b9c5eb0SJonathan T. Looney atomic_add_int(&nfrags, 1); 2991dbefcc0SGleb Smirnoff #ifdef MAC 3001dbefcc0SGleb Smirnoff mac_ipq_update(m, fp); 3011dbefcc0SGleb Smirnoff #endif 3021dbefcc0SGleb Smirnoff } 3031dbefcc0SGleb Smirnoff 3041dbefcc0SGleb Smirnoff #define GETIP(m) ((struct ip*)((m)->m_pkthdr.PH_loc.ptr)) 3051dbefcc0SGleb Smirnoff 3061dbefcc0SGleb Smirnoff /* 3071dbefcc0SGleb Smirnoff * Handle ECN by comparing this segment with the first one; 3081dbefcc0SGleb Smirnoff * if CE is set, do not lose CE. 3091dbefcc0SGleb Smirnoff * drop if CE and not-ECT are mixed for the same packet. 3101dbefcc0SGleb Smirnoff */ 3111dbefcc0SGleb Smirnoff ecn = ip->ip_tos & IPTOS_ECN_MASK; 3121dbefcc0SGleb Smirnoff ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; 3131dbefcc0SGleb Smirnoff if (ecn == IPTOS_ECN_CE) { 3141dbefcc0SGleb Smirnoff if (ecn0 == IPTOS_ECN_NOTECT) 3151dbefcc0SGleb Smirnoff goto dropfrag; 3161dbefcc0SGleb Smirnoff if (ecn0 != IPTOS_ECN_CE) 3171dbefcc0SGleb Smirnoff GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; 3181dbefcc0SGleb Smirnoff } 3191dbefcc0SGleb Smirnoff if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) 3201dbefcc0SGleb Smirnoff goto dropfrag; 3211dbefcc0SGleb Smirnoff 3221dbefcc0SGleb Smirnoff /* 3231dbefcc0SGleb Smirnoff * Find a segment which begins after this one does. 3241dbefcc0SGleb Smirnoff */ 3251dbefcc0SGleb Smirnoff for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) 3261dbefcc0SGleb Smirnoff if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off)) 3271dbefcc0SGleb Smirnoff break; 3281dbefcc0SGleb Smirnoff 3291dbefcc0SGleb Smirnoff /* 3301dbefcc0SGleb Smirnoff * If there is a preceding segment, it may provide some of 3311dbefcc0SGleb Smirnoff * our data already. If so, drop the data from the incoming 3321dbefcc0SGleb Smirnoff * segment. If it provides all of our data, drop us, otherwise 3331dbefcc0SGleb Smirnoff * stick new segment in the proper place. 3341dbefcc0SGleb Smirnoff * 3351dbefcc0SGleb Smirnoff * If some of the data is dropped from the preceding 3361dbefcc0SGleb Smirnoff * segment, then it's checksum is invalidated. 3371dbefcc0SGleb Smirnoff */ 3381dbefcc0SGleb Smirnoff if (p) { 3391dbefcc0SGleb Smirnoff i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) - 3401dbefcc0SGleb Smirnoff ntohs(ip->ip_off); 3411dbefcc0SGleb Smirnoff if (i > 0) { 3421dbefcc0SGleb Smirnoff if (i >= ntohs(ip->ip_len)) 3431dbefcc0SGleb Smirnoff goto dropfrag; 3441dbefcc0SGleb Smirnoff m_adj(m, i); 3451dbefcc0SGleb Smirnoff m->m_pkthdr.csum_flags = 0; 3461dbefcc0SGleb Smirnoff ip->ip_off = htons(ntohs(ip->ip_off) + i); 3471dbefcc0SGleb Smirnoff ip->ip_len = htons(ntohs(ip->ip_len) - i); 3481dbefcc0SGleb Smirnoff } 3491dbefcc0SGleb Smirnoff m->m_nextpkt = p->m_nextpkt; 3501dbefcc0SGleb Smirnoff p->m_nextpkt = m; 3511dbefcc0SGleb Smirnoff } else { 3521dbefcc0SGleb Smirnoff m->m_nextpkt = fp->ipq_frags; 3531dbefcc0SGleb Smirnoff fp->ipq_frags = m; 3541dbefcc0SGleb Smirnoff } 3551dbefcc0SGleb Smirnoff 3561dbefcc0SGleb Smirnoff /* 3571dbefcc0SGleb Smirnoff * While we overlap succeeding segments trim them or, 3581dbefcc0SGleb Smirnoff * if they are completely covered, dequeue them. 3591dbefcc0SGleb Smirnoff */ 3601dbefcc0SGleb Smirnoff for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) > 3611dbefcc0SGleb Smirnoff ntohs(GETIP(q)->ip_off); q = nq) { 3621dbefcc0SGleb Smirnoff i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) - 3631dbefcc0SGleb Smirnoff ntohs(GETIP(q)->ip_off); 3641dbefcc0SGleb Smirnoff if (i < ntohs(GETIP(q)->ip_len)) { 3651dbefcc0SGleb Smirnoff GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i); 3661dbefcc0SGleb Smirnoff GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i); 3671dbefcc0SGleb Smirnoff m_adj(q, i); 3681dbefcc0SGleb Smirnoff q->m_pkthdr.csum_flags = 0; 3691dbefcc0SGleb Smirnoff break; 3701dbefcc0SGleb Smirnoff } 3711dbefcc0SGleb Smirnoff nq = q->m_nextpkt; 3721dbefcc0SGleb Smirnoff m->m_nextpkt = nq; 3731dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 3741dbefcc0SGleb Smirnoff fp->ipq_nfrags--; 3757b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, 1); 3761dbefcc0SGleb Smirnoff m_freem(q); 3771dbefcc0SGleb Smirnoff } 3781dbefcc0SGleb Smirnoff 3791dbefcc0SGleb Smirnoff /* 3801dbefcc0SGleb Smirnoff * Check for complete reassembly and perform frag per packet 3811dbefcc0SGleb Smirnoff * limiting. 3821dbefcc0SGleb Smirnoff * 3831dbefcc0SGleb Smirnoff * Frag limiting is performed here so that the nth frag has 3841dbefcc0SGleb Smirnoff * a chance to complete the packet before we drop the packet. 3851dbefcc0SGleb Smirnoff * As a result, n+1 frags are actually allowed per packet, but 3861dbefcc0SGleb Smirnoff * only n will ever be stored. (n = maxfragsperpacket.) 3871dbefcc0SGleb Smirnoff * 3881dbefcc0SGleb Smirnoff */ 3891dbefcc0SGleb Smirnoff next = 0; 3901dbefcc0SGleb Smirnoff for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { 3911dbefcc0SGleb Smirnoff if (ntohs(GETIP(q)->ip_off) != next) { 3921dbefcc0SGleb Smirnoff if (fp->ipq_nfrags > V_maxfragsperpacket) 393ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[hash], fp); 3941dbefcc0SGleb Smirnoff goto done; 3951dbefcc0SGleb Smirnoff } 3961dbefcc0SGleb Smirnoff next += ntohs(GETIP(q)->ip_len); 3971dbefcc0SGleb Smirnoff } 3981dbefcc0SGleb Smirnoff /* Make sure the last packet didn't have the IP_MF flag */ 3991dbefcc0SGleb Smirnoff if (p->m_flags & M_IP_FRAG) { 4001dbefcc0SGleb Smirnoff if (fp->ipq_nfrags > V_maxfragsperpacket) 401ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[hash], fp); 4021dbefcc0SGleb Smirnoff goto done; 4031dbefcc0SGleb Smirnoff } 4041dbefcc0SGleb Smirnoff 4051dbefcc0SGleb Smirnoff /* 4061dbefcc0SGleb Smirnoff * Reassembly is complete. Make sure the packet is a sane size. 4071dbefcc0SGleb Smirnoff */ 4081dbefcc0SGleb Smirnoff q = fp->ipq_frags; 4091dbefcc0SGleb Smirnoff ip = GETIP(q); 4101dbefcc0SGleb Smirnoff if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { 4111dbefcc0SGleb Smirnoff IPSTAT_INC(ips_toolong); 412ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[hash], fp); 4131dbefcc0SGleb Smirnoff goto done; 4141dbefcc0SGleb Smirnoff } 4151dbefcc0SGleb Smirnoff 4161dbefcc0SGleb Smirnoff /* 4171dbefcc0SGleb Smirnoff * Concatenate fragments. 4181dbefcc0SGleb Smirnoff */ 4191dbefcc0SGleb Smirnoff m = q; 4201dbefcc0SGleb Smirnoff t = m->m_next; 4211dbefcc0SGleb Smirnoff m->m_next = NULL; 4221dbefcc0SGleb Smirnoff m_cat(m, t); 4231dbefcc0SGleb Smirnoff nq = q->m_nextpkt; 4241dbefcc0SGleb Smirnoff q->m_nextpkt = NULL; 4251dbefcc0SGleb Smirnoff for (q = nq; q != NULL; q = nq) { 4261dbefcc0SGleb Smirnoff nq = q->m_nextpkt; 4271dbefcc0SGleb Smirnoff q->m_nextpkt = NULL; 4281dbefcc0SGleb Smirnoff m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; 4291dbefcc0SGleb Smirnoff m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; 43009b0b8c0SNavdeep Parhar m_demote_pkthdr(q); 4311dbefcc0SGleb Smirnoff m_cat(m, q); 4321dbefcc0SGleb Smirnoff } 4331dbefcc0SGleb Smirnoff /* 4341dbefcc0SGleb Smirnoff * In order to do checksumming faster we do 'end-around carry' here 4351dbefcc0SGleb Smirnoff * (and not in for{} loop), though it implies we are not going to 4361dbefcc0SGleb Smirnoff * reassemble more than 64k fragments. 4371dbefcc0SGleb Smirnoff */ 4381dbefcc0SGleb Smirnoff while (m->m_pkthdr.csum_data & 0xffff0000) 4391dbefcc0SGleb Smirnoff m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + 4401dbefcc0SGleb Smirnoff (m->m_pkthdr.csum_data >> 16); 4417b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, fp->ipq_nfrags); 4421dbefcc0SGleb Smirnoff #ifdef MAC 4431dbefcc0SGleb Smirnoff mac_ipq_reassemble(fp, m); 4441dbefcc0SGleb Smirnoff mac_ipq_destroy(fp); 4451dbefcc0SGleb Smirnoff #endif 4461dbefcc0SGleb Smirnoff 4471dbefcc0SGleb Smirnoff /* 4481dbefcc0SGleb Smirnoff * Create header for new ip packet by modifying header of first 4491dbefcc0SGleb Smirnoff * packet; dequeue and discard fragment reassembly header. 4501dbefcc0SGleb Smirnoff * Make header visible. 4511dbefcc0SGleb Smirnoff */ 4521dbefcc0SGleb Smirnoff ip->ip_len = htons((ip->ip_hl << 2) + next); 4531dbefcc0SGleb Smirnoff ip->ip_src = fp->ipq_src; 4541dbefcc0SGleb Smirnoff ip->ip_dst = fp->ipq_dst; 4551dbefcc0SGleb Smirnoff TAILQ_REMOVE(head, fp, ipq_list); 456ff790bbaSJonathan T. Looney V_ipq[hash].count--; 4571dbefcc0SGleb Smirnoff uma_zfree(V_ipq_zone, fp); 4581dbefcc0SGleb Smirnoff m->m_len += (ip->ip_hl << 2); 4591dbefcc0SGleb Smirnoff m->m_data -= (ip->ip_hl << 2); 4601dbefcc0SGleb Smirnoff /* some debugging cruft by sklower, below, will go away soon */ 4611dbefcc0SGleb Smirnoff if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ 4621dbefcc0SGleb Smirnoff m_fixhdr(m); 4631dbefcc0SGleb Smirnoff IPSTAT_INC(ips_reassembled); 4641dbefcc0SGleb Smirnoff IPQ_UNLOCK(hash); 4651dbefcc0SGleb Smirnoff 4661dbefcc0SGleb Smirnoff #ifdef RSS 4671dbefcc0SGleb Smirnoff /* 4681dbefcc0SGleb Smirnoff * Query the RSS layer for the flowid / flowtype for the 4691dbefcc0SGleb Smirnoff * mbuf payload. 4701dbefcc0SGleb Smirnoff * 4711dbefcc0SGleb Smirnoff * For now, just assume we have to calculate a new one. 4721dbefcc0SGleb Smirnoff * Later on we should check to see if the assigned flowid matches 4731dbefcc0SGleb Smirnoff * what RSS wants for the given IP protocol and if so, just keep it. 4741dbefcc0SGleb Smirnoff * 4751dbefcc0SGleb Smirnoff * We then queue into the relevant netisr so it can be dispatched 4761dbefcc0SGleb Smirnoff * to the correct CPU. 4771dbefcc0SGleb Smirnoff * 4781dbefcc0SGleb Smirnoff * Note - this may return 1, which means the flowid in the mbuf 4791dbefcc0SGleb Smirnoff * is correct for the configured RSS hash types and can be used. 4801dbefcc0SGleb Smirnoff */ 4811dbefcc0SGleb Smirnoff if (rss_mbuf_software_hash_v4(m, 0, &rss_hash, &rss_type) == 0) { 4821dbefcc0SGleb Smirnoff m->m_pkthdr.flowid = rss_hash; 4831dbefcc0SGleb Smirnoff M_HASHTYPE_SET(m, rss_type); 4841dbefcc0SGleb Smirnoff } 4851dbefcc0SGleb Smirnoff 4861dbefcc0SGleb Smirnoff /* 4871dbefcc0SGleb Smirnoff * Queue/dispatch for reprocessing. 4881dbefcc0SGleb Smirnoff * 4891dbefcc0SGleb Smirnoff * Note: this is much slower than just handling the frame in the 4901dbefcc0SGleb Smirnoff * current receive context. It's likely worth investigating 4911dbefcc0SGleb Smirnoff * why this is. 4921dbefcc0SGleb Smirnoff */ 4931dbefcc0SGleb Smirnoff netisr_dispatch(NETISR_IP_DIRECT, m); 4941dbefcc0SGleb Smirnoff return (NULL); 4951dbefcc0SGleb Smirnoff #endif 4961dbefcc0SGleb Smirnoff 4971dbefcc0SGleb Smirnoff /* Handle in-line */ 4981dbefcc0SGleb Smirnoff return (m); 4991dbefcc0SGleb Smirnoff 5001dbefcc0SGleb Smirnoff dropfrag: 5011dbefcc0SGleb Smirnoff IPSTAT_INC(ips_fragdropped); 5027b9c5eb0SJonathan T. Looney if (fp != NULL) { 5031dbefcc0SGleb Smirnoff fp->ipq_nfrags--; 5047b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, 1); 5057b9c5eb0SJonathan T. Looney } 5061dbefcc0SGleb Smirnoff m_freem(m); 5071dbefcc0SGleb Smirnoff done: 5081dbefcc0SGleb Smirnoff IPQ_UNLOCK(hash); 5091dbefcc0SGleb Smirnoff return (NULL); 5101dbefcc0SGleb Smirnoff 5111dbefcc0SGleb Smirnoff #undef GETIP 5121dbefcc0SGleb Smirnoff } 5131dbefcc0SGleb Smirnoff 5141dbefcc0SGleb Smirnoff /* 5151dbefcc0SGleb Smirnoff * Initialize IP reassembly structures. 5161dbefcc0SGleb Smirnoff */ 5171dbefcc0SGleb Smirnoff void 5181dbefcc0SGleb Smirnoff ipreass_init(void) 5191dbefcc0SGleb Smirnoff { 520ff790bbaSJonathan T. Looney int max; 5211dbefcc0SGleb Smirnoff 5221dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 5231dbefcc0SGleb Smirnoff TAILQ_INIT(&V_ipq[i].head); 5241dbefcc0SGleb Smirnoff mtx_init(&V_ipq[i].lock, "IP reassembly", NULL, 5251dbefcc0SGleb Smirnoff MTX_DEF | MTX_DUPOK); 526ff790bbaSJonathan T. Looney V_ipq[i].count = 0; 5271dbefcc0SGleb Smirnoff } 528c047fd1bSGleb Smirnoff V_ipq_hashseed = arc4random(); 5291dbefcc0SGleb Smirnoff V_maxfragsperpacket = 16; 5301dbefcc0SGleb Smirnoff V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, 5311dbefcc0SGleb Smirnoff NULL, UMA_ALIGN_PTR, 0); 532a967df1cSJonathan T. Looney max = IP_MAXFRAGPACKETS; 533ff790bbaSJonathan T. Looney max = uma_zone_set_max(V_ipq_zone, max); 534ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); 5351dbefcc0SGleb Smirnoff 5367b9c5eb0SJonathan T. Looney if (IS_DEFAULT_VNET(curvnet)) { 537a967df1cSJonathan T. Looney maxfrags = IP_MAXFRAGS; 5381dbefcc0SGleb Smirnoff EVENTHANDLER_REGISTER(nmbclusters_change, ipreass_zone_change, 5391dbefcc0SGleb Smirnoff NULL, EVENTHANDLER_PRI_ANY); 5401dbefcc0SGleb Smirnoff } 5417b9c5eb0SJonathan T. Looney } 5421dbefcc0SGleb Smirnoff 5431dbefcc0SGleb Smirnoff /* 5441dbefcc0SGleb Smirnoff * If a timer expires on a reassembly queue, discard it. 5451dbefcc0SGleb Smirnoff */ 5461dbefcc0SGleb Smirnoff void 5471dbefcc0SGleb Smirnoff ipreass_slowtimo(void) 5481dbefcc0SGleb Smirnoff { 5491dbefcc0SGleb Smirnoff struct ipq *fp, *tmp; 5501dbefcc0SGleb Smirnoff 5511dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 5521dbefcc0SGleb Smirnoff IPQ_LOCK(i); 5531dbefcc0SGleb Smirnoff TAILQ_FOREACH_SAFE(fp, &V_ipq[i].head, ipq_list, tmp) 5541dbefcc0SGleb Smirnoff if (--fp->ipq_ttl == 0) 555ff790bbaSJonathan T. Looney ipq_timeout(&V_ipq[i], fp); 5561dbefcc0SGleb Smirnoff IPQ_UNLOCK(i); 5571dbefcc0SGleb Smirnoff } 5581dbefcc0SGleb Smirnoff } 5591dbefcc0SGleb Smirnoff 5601dbefcc0SGleb Smirnoff /* 5611dbefcc0SGleb Smirnoff * Drain off all datagram fragments. 5621dbefcc0SGleb Smirnoff */ 5631dbefcc0SGleb Smirnoff void 5641dbefcc0SGleb Smirnoff ipreass_drain(void) 5651dbefcc0SGleb Smirnoff { 5661dbefcc0SGleb Smirnoff 5671dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 5681dbefcc0SGleb Smirnoff IPQ_LOCK(i); 5691dbefcc0SGleb Smirnoff while(!TAILQ_EMPTY(&V_ipq[i].head)) 570ff790bbaSJonathan T. Looney ipq_drop(&V_ipq[i], TAILQ_FIRST(&V_ipq[i].head)); 571ff790bbaSJonathan T. Looney KASSERT(V_ipq[i].count == 0, 572ff790bbaSJonathan T. Looney ("%s: V_ipq[%d] count %d (V_ipq=%p)", __func__, i, 573ff790bbaSJonathan T. Looney V_ipq[i].count, V_ipq)); 5741dbefcc0SGleb Smirnoff IPQ_UNLOCK(i); 5751dbefcc0SGleb Smirnoff } 5761dbefcc0SGleb Smirnoff } 5771dbefcc0SGleb Smirnoff 5781dbefcc0SGleb Smirnoff #ifdef VIMAGE 5791dbefcc0SGleb Smirnoff /* 5801dbefcc0SGleb Smirnoff * Destroy IP reassembly structures. 5811dbefcc0SGleb Smirnoff */ 5821dbefcc0SGleb Smirnoff void 5831dbefcc0SGleb Smirnoff ipreass_destroy(void) 5841dbefcc0SGleb Smirnoff { 5851dbefcc0SGleb Smirnoff 5861dbefcc0SGleb Smirnoff ipreass_drain(); 5871dbefcc0SGleb Smirnoff uma_zdestroy(V_ipq_zone); 5881dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) 5891dbefcc0SGleb Smirnoff mtx_destroy(&V_ipq[i].lock); 5901dbefcc0SGleb Smirnoff } 5911dbefcc0SGleb Smirnoff #endif 5921dbefcc0SGleb Smirnoff 5931dbefcc0SGleb Smirnoff /* 5941dbefcc0SGleb Smirnoff * After maxnipq has been updated, propagate the change to UMA. The UMA zone 5951dbefcc0SGleb Smirnoff * max has slightly different semantics than the sysctl, for historical 5961dbefcc0SGleb Smirnoff * reasons. 5971dbefcc0SGleb Smirnoff */ 5981dbefcc0SGleb Smirnoff static void 5991dbefcc0SGleb Smirnoff ipreass_drain_tomax(void) 6001dbefcc0SGleb Smirnoff { 601ff790bbaSJonathan T. Looney struct ipq *fp; 6021dbefcc0SGleb Smirnoff int target; 6031dbefcc0SGleb Smirnoff 6041dbefcc0SGleb Smirnoff /* 605ff790bbaSJonathan T. Looney * Make sure each bucket is under the new limit. If 606ff790bbaSJonathan T. Looney * necessary, drop enough of the oldest elements from 607ff790bbaSJonathan T. Looney * each bucket to get under the new limit. 608ff790bbaSJonathan T. Looney */ 609ff790bbaSJonathan T. Looney for (int i = 0; i < IPREASS_NHASH; i++) { 610ff790bbaSJonathan T. Looney IPQ_LOCK(i); 611ff790bbaSJonathan T. Looney while (V_ipq[i].count > V_ipreass_maxbucketsize && 612ff790bbaSJonathan T. Looney (fp = TAILQ_LAST(&V_ipq[i].head, ipqhead)) != NULL) 613ff790bbaSJonathan T. Looney ipq_timeout(&V_ipq[i], fp); 614ff790bbaSJonathan T. Looney IPQ_UNLOCK(i); 615ff790bbaSJonathan T. Looney } 616ff790bbaSJonathan T. Looney 617ff790bbaSJonathan T. Looney /* 6181dbefcc0SGleb Smirnoff * If we are over the maximum number of fragments, 6191dbefcc0SGleb Smirnoff * drain off enough to get down to the new limit, 6201dbefcc0SGleb Smirnoff * stripping off last elements on queues. Every 6211dbefcc0SGleb Smirnoff * run we strip the oldest element from each bucket. 6221dbefcc0SGleb Smirnoff */ 6231dbefcc0SGleb Smirnoff target = uma_zone_get_max(V_ipq_zone); 6241dbefcc0SGleb Smirnoff while (uma_zone_get_cur(V_ipq_zone) > target) { 6251dbefcc0SGleb Smirnoff for (int i = 0; i < IPREASS_NHASH; i++) { 6261dbefcc0SGleb Smirnoff IPQ_LOCK(i); 6271dbefcc0SGleb Smirnoff fp = TAILQ_LAST(&V_ipq[i].head, ipqhead); 6281dbefcc0SGleb Smirnoff if (fp != NULL) 629ff790bbaSJonathan T. Looney ipq_timeout(&V_ipq[i], fp); 6301dbefcc0SGleb Smirnoff IPQ_UNLOCK(i); 6311dbefcc0SGleb Smirnoff } 6321dbefcc0SGleb Smirnoff } 6331dbefcc0SGleb Smirnoff } 6341dbefcc0SGleb Smirnoff 6351dbefcc0SGleb Smirnoff static void 6361dbefcc0SGleb Smirnoff ipreass_zone_change(void *tag) 6371dbefcc0SGleb Smirnoff { 6387b9c5eb0SJonathan T. Looney VNET_ITERATOR_DECL(vnet_iter); 6397b9c5eb0SJonathan T. Looney int max; 6401dbefcc0SGleb Smirnoff 641a967df1cSJonathan T. Looney maxfrags = IP_MAXFRAGS; 642a967df1cSJonathan T. Looney max = IP_MAXFRAGPACKETS; 6437b9c5eb0SJonathan T. Looney VNET_LIST_RLOCK_NOSLEEP(); 6447b9c5eb0SJonathan T. Looney VNET_FOREACH(vnet_iter) { 6457b9c5eb0SJonathan T. Looney CURVNET_SET(vnet_iter); 646ff790bbaSJonathan T. Looney max = uma_zone_set_max(V_ipq_zone, max); 647ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); 6481dbefcc0SGleb Smirnoff ipreass_drain_tomax(); 6497b9c5eb0SJonathan T. Looney CURVNET_RESTORE(); 6507b9c5eb0SJonathan T. Looney } 6517b9c5eb0SJonathan T. Looney VNET_LIST_RUNLOCK_NOSLEEP(); 6521dbefcc0SGleb Smirnoff } 6531dbefcc0SGleb Smirnoff 6541dbefcc0SGleb Smirnoff /* 6551dbefcc0SGleb Smirnoff * Change the limit on the UMA zone, or disable the fragment allocation 6561dbefcc0SGleb Smirnoff * at all. Since 0 and -1 is a special values here, we need our own handler, 6571dbefcc0SGleb Smirnoff * instead of sysctl_handle_uma_zone_max(). 6581dbefcc0SGleb Smirnoff */ 6591dbefcc0SGleb Smirnoff static int 6601dbefcc0SGleb Smirnoff sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS) 6611dbefcc0SGleb Smirnoff { 6621dbefcc0SGleb Smirnoff int error, max; 6631dbefcc0SGleb Smirnoff 6641dbefcc0SGleb Smirnoff if (V_noreass == 0) { 6651dbefcc0SGleb Smirnoff max = uma_zone_get_max(V_ipq_zone); 6661dbefcc0SGleb Smirnoff if (max == 0) 6671dbefcc0SGleb Smirnoff max = -1; 6681dbefcc0SGleb Smirnoff } else 6691dbefcc0SGleb Smirnoff max = 0; 6701dbefcc0SGleb Smirnoff error = sysctl_handle_int(oidp, &max, 0, req); 6711dbefcc0SGleb Smirnoff if (error || !req->newptr) 6721dbefcc0SGleb Smirnoff return (error); 6731dbefcc0SGleb Smirnoff if (max > 0) { 6741dbefcc0SGleb Smirnoff /* 6751dbefcc0SGleb Smirnoff * XXXRW: Might be a good idea to sanity check the argument 6761dbefcc0SGleb Smirnoff * and place an extreme upper bound. 6771dbefcc0SGleb Smirnoff */ 6781dbefcc0SGleb Smirnoff max = uma_zone_set_max(V_ipq_zone, max); 679ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); 6801dbefcc0SGleb Smirnoff ipreass_drain_tomax(); 6811dbefcc0SGleb Smirnoff V_noreass = 0; 6821dbefcc0SGleb Smirnoff } else if (max == 0) { 6831dbefcc0SGleb Smirnoff V_noreass = 1; 6841dbefcc0SGleb Smirnoff ipreass_drain(); 6851dbefcc0SGleb Smirnoff } else if (max == -1) { 6861dbefcc0SGleb Smirnoff V_noreass = 0; 6871dbefcc0SGleb Smirnoff uma_zone_set_max(V_ipq_zone, 0); 688ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = INT_MAX; 6891dbefcc0SGleb Smirnoff } else 6901dbefcc0SGleb Smirnoff return (EINVAL); 6911dbefcc0SGleb Smirnoff return (0); 6921dbefcc0SGleb Smirnoff } 6931dbefcc0SGleb Smirnoff 6941dbefcc0SGleb Smirnoff /* 6951dbefcc0SGleb Smirnoff * Seek for old fragment queue header that can be reused. Try to 6961dbefcc0SGleb Smirnoff * reuse a header from currently locked hash bucket. 6971dbefcc0SGleb Smirnoff */ 6981dbefcc0SGleb Smirnoff static struct ipq * 6991dbefcc0SGleb Smirnoff ipq_reuse(int start) 7001dbefcc0SGleb Smirnoff { 7011dbefcc0SGleb Smirnoff struct ipq *fp; 702ff790bbaSJonathan T. Looney int bucket, i; 7031dbefcc0SGleb Smirnoff 7041dbefcc0SGleb Smirnoff IPQ_LOCK_ASSERT(start); 7051dbefcc0SGleb Smirnoff 706ff790bbaSJonathan T. Looney for (i = 0; i < IPREASS_NHASH; i++) { 707ff790bbaSJonathan T. Looney bucket = (start + i) % IPREASS_NHASH; 708ff790bbaSJonathan T. Looney if (bucket != start && IPQ_TRYLOCK(bucket) == 0) 7091dbefcc0SGleb Smirnoff continue; 710ff790bbaSJonathan T. Looney fp = TAILQ_LAST(&V_ipq[bucket].head, ipqhead); 7111dbefcc0SGleb Smirnoff if (fp) { 7121dbefcc0SGleb Smirnoff struct mbuf *m; 7131dbefcc0SGleb Smirnoff 7141dbefcc0SGleb Smirnoff IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); 7157b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, fp->ipq_nfrags); 7161dbefcc0SGleb Smirnoff while (fp->ipq_frags) { 7171dbefcc0SGleb Smirnoff m = fp->ipq_frags; 7181dbefcc0SGleb Smirnoff fp->ipq_frags = m->m_nextpkt; 7191dbefcc0SGleb Smirnoff m_freem(m); 7201dbefcc0SGleb Smirnoff } 721ff790bbaSJonathan T. Looney TAILQ_REMOVE(&V_ipq[bucket].head, fp, ipq_list); 722ff790bbaSJonathan T. Looney V_ipq[bucket].count--; 723ff790bbaSJonathan T. Looney if (bucket != start) 724ff790bbaSJonathan T. Looney IPQ_UNLOCK(bucket); 725ff790bbaSJonathan T. Looney break; 726ff790bbaSJonathan T. Looney } 727ff790bbaSJonathan T. Looney if (bucket != start) 728ff790bbaSJonathan T. Looney IPQ_UNLOCK(bucket); 729ff790bbaSJonathan T. Looney } 7301dbefcc0SGleb Smirnoff IPQ_LOCK_ASSERT(start); 7311dbefcc0SGleb Smirnoff return (fp); 7321dbefcc0SGleb Smirnoff } 7331dbefcc0SGleb Smirnoff 7341dbefcc0SGleb Smirnoff /* 7351dbefcc0SGleb Smirnoff * Free a fragment reassembly header and all associated datagrams. 7361dbefcc0SGleb Smirnoff */ 7371dbefcc0SGleb Smirnoff static void 738ff790bbaSJonathan T. Looney ipq_free(struct ipqbucket *bucket, struct ipq *fp) 7391dbefcc0SGleb Smirnoff { 7401dbefcc0SGleb Smirnoff struct mbuf *q; 7411dbefcc0SGleb Smirnoff 7427b9c5eb0SJonathan T. Looney atomic_subtract_int(&nfrags, fp->ipq_nfrags); 7431dbefcc0SGleb Smirnoff while (fp->ipq_frags) { 7441dbefcc0SGleb Smirnoff q = fp->ipq_frags; 7451dbefcc0SGleb Smirnoff fp->ipq_frags = q->m_nextpkt; 7461dbefcc0SGleb Smirnoff m_freem(q); 7471dbefcc0SGleb Smirnoff } 748ff790bbaSJonathan T. Looney TAILQ_REMOVE(&bucket->head, fp, ipq_list); 749ff790bbaSJonathan T. Looney bucket->count--; 7501dbefcc0SGleb Smirnoff uma_zfree(V_ipq_zone, fp); 7511dbefcc0SGleb Smirnoff } 752ff790bbaSJonathan T. Looney 753ff790bbaSJonathan T. Looney /* 754ff790bbaSJonathan T. Looney * Get or set the maximum number of reassembly queues per bucket. 755ff790bbaSJonathan T. Looney */ 756ff790bbaSJonathan T. Looney static int 757ff790bbaSJonathan T. Looney sysctl_maxfragbucketsize(SYSCTL_HANDLER_ARGS) 758ff790bbaSJonathan T. Looney { 759ff790bbaSJonathan T. Looney int error, max; 760ff790bbaSJonathan T. Looney 761ff790bbaSJonathan T. Looney max = V_ipreass_maxbucketsize; 762ff790bbaSJonathan T. Looney error = sysctl_handle_int(oidp, &max, 0, req); 763ff790bbaSJonathan T. Looney if (error || !req->newptr) 764ff790bbaSJonathan T. Looney return (error); 765ff790bbaSJonathan T. Looney if (max <= 0) 766ff790bbaSJonathan T. Looney return (EINVAL); 767ff790bbaSJonathan T. Looney V_ipreass_maxbucketsize = max; 768ff790bbaSJonathan T. Looney ipreass_drain_tomax(); 769ff790bbaSJonathan T. Looney return (0); 770ff790bbaSJonathan T. Looney } 771