Index: head/sys/netinet/ip_reass.c =================================================================== --- head/sys/netinet/ip_reass.c (revision 346529) +++ head/sys/netinet/ip_reass.c (revision 346530) @@ -1,801 +1,835 @@ /*- * Copyright (c) 2015 Gleb Smirnoff * Copyright (c) 2015 Adrian Chadd * Copyright (c) 1982, 1986, 1988, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ip_input.c 8.2 (Berkeley) 1/4/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include +#include +#include +#include #include #include #include #include #include #include #include #ifdef MAC #include #endif SYSCTL_DECL(_net_inet_ip); /* * Reassembly headers are stored in hash buckets. */ #define IPREASS_NHASH_LOG2 10 #define IPREASS_NHASH (1 << IPREASS_NHASH_LOG2) #define IPREASS_HMASK (IPREASS_NHASH - 1) struct ipqbucket { TAILQ_HEAD(ipqhead, ipq) head; struct mtx lock; int count; }; VNET_DEFINE_STATIC(struct ipqbucket, ipq[IPREASS_NHASH]); #define V_ipq VNET(ipq) VNET_DEFINE_STATIC(uint32_t, ipq_hashseed); #define V_ipq_hashseed VNET(ipq_hashseed) #define IPQ_LOCK(i) mtx_lock(&V_ipq[i].lock) #define IPQ_TRYLOCK(i) mtx_trylock(&V_ipq[i].lock) #define IPQ_UNLOCK(i) mtx_unlock(&V_ipq[i].lock) #define IPQ_LOCK_ASSERT(i) mtx_assert(&V_ipq[i].lock, MA_OWNED) VNET_DEFINE_STATIC(int, ipreass_maxbucketsize); #define V_ipreass_maxbucketsize VNET(ipreass_maxbucketsize) void ipreass_init(void); void ipreass_drain(void); void ipreass_slowtimo(void); #ifdef VIMAGE void ipreass_destroy(void); #endif static int sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS); static int sysctl_maxfragbucketsize(SYSCTL_HANDLER_ARGS); static void ipreass_zone_change(void *); static void ipreass_drain_tomax(void); static void ipq_free(struct ipqbucket *, struct ipq *); static struct ipq * ipq_reuse(int); static inline void ipq_timeout(struct ipqbucket *bucket, struct ipq *fp) { IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); ipq_free(bucket, fp); } static inline void ipq_drop(struct ipqbucket *bucket, struct ipq *fp) { IPSTAT_ADD(ips_fragdropped, fp->ipq_nfrags); ipq_free(bucket, fp); } /* * By default, limit the number of IP fragments across all reassembly * queues to 1/32 of the total number of mbuf clusters. * * Limit the total number of reassembly queues per VNET to the * IP fragment limit, but ensure the limit will not allow any bucket * to grow above 100 items. (The bucket limit is * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct * multiplier to reach a 100-item limit.) * The 100-item limit was chosen as brief testing seems to show that * this produces "reasonable" performance on some subset of systems * under DoS attack. */ #define IP_MAXFRAGS (nmbclusters / 32) #define IP_MAXFRAGPACKETS (imin(IP_MAXFRAGS, IPREASS_NHASH * 50)) static int maxfrags; static volatile u_int nfrags; SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfrags, CTLFLAG_RW, &maxfrags, 0, "Maximum number of IPv4 fragments allowed across all reassembly queues"); SYSCTL_UINT(_net_inet_ip, OID_AUTO, curfrags, CTLFLAG_RD, __DEVOLATILE(u_int *, &nfrags), 0, "Current number of IPv4 fragments across all reassembly queues"); VNET_DEFINE_STATIC(uma_zone_t, ipq_zone); #define V_ipq_zone VNET(ipq_zone) SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragpackets, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, NULL, 0, sysctl_maxfragpackets, "I", "Maximum number of IPv4 fragment reassembly queue entries"); SYSCTL_UMA_CUR(_net_inet_ip, OID_AUTO, fragpackets, CTLFLAG_VNET, &VNET_NAME(ipq_zone), "Current number of IPv4 fragment reassembly queue entries"); VNET_DEFINE_STATIC(int, noreass); #define V_noreass VNET(noreass) VNET_DEFINE_STATIC(int, maxfragsperpacket); #define V_maxfragsperpacket VNET(maxfragsperpacket) SYSCTL_INT(_net_inet_ip, OID_AUTO, maxfragsperpacket, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(maxfragsperpacket), 0, "Maximum number of IPv4 fragments allowed per packet"); SYSCTL_PROC(_net_inet_ip, OID_AUTO, maxfragbucketsize, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_maxfragbucketsize, "I", "Maximum number of IPv4 fragment reassembly queue entries per bucket"); /* * Take incoming datagram fragment and try to reassemble it into * whole datagram. If the argument is the first fragment or one * in between the function will return NULL and store the mbuf * in the fragment chain. If the argument is the last fragment * the packet will be reassembled and the pointer to the new * mbuf returned for further processing. Only m_tags attached * to the first packet/fragment are preserved. * The IP header is *NOT* adjusted out of iplen. */ #define M_IP_FRAG M_PROTO9 struct mbuf * ip_reass(struct mbuf *m) { struct ip *ip; struct mbuf *p, *q, *nq, *t; struct ipq *fp; struct ipqhead *head; int i, hlen, next, tmpmax; u_int8_t ecn, ecn0; uint32_t hash, hashkey[3]; #ifdef RSS uint32_t rss_hash, rss_type; #endif /* * If no reassembling or maxfragsperpacket are 0, * never accept fragments. * Also, drop packet if it would exceed the maximum * number of fragments. */ tmpmax = maxfrags; if (V_noreass == 1 || V_maxfragsperpacket == 0 || (tmpmax >= 0 && atomic_load_int(&nfrags) >= (u_int)tmpmax)) { IPSTAT_INC(ips_fragments); IPSTAT_INC(ips_fragdropped); m_freem(m); return (NULL); } ip = mtod(m, struct ip *); hlen = ip->ip_hl << 2; /* * Adjust ip_len to not reflect header, * convert offset of this to bytes. */ ip->ip_len = htons(ntohs(ip->ip_len) - hlen); /* * Make sure that fragments have a data length * that's a non-zero multiple of 8 bytes, unless * this is the last fragment. */ if (ip->ip_len == htons(0) || ((ip->ip_off & htons(IP_MF)) && (ntohs(ip->ip_len) & 0x7) != 0)) { IPSTAT_INC(ips_toosmall); /* XXX */ IPSTAT_INC(ips_fragdropped); m_freem(m); return (NULL); } if (ip->ip_off & htons(IP_MF)) m->m_flags |= M_IP_FRAG; else m->m_flags &= ~M_IP_FRAG; ip->ip_off = htons(ntohs(ip->ip_off) << 3); /* * Make sure the fragment lies within a packet of valid size. */ if (ntohs(ip->ip_len) + ntohs(ip->ip_off) > IP_MAXPACKET) { IPSTAT_INC(ips_toolong); IPSTAT_INC(ips_fragdropped); m_freem(m); return (NULL); } /* * Attempt reassembly; if it succeeds, proceed. * ip_reass() will return a different mbuf. */ IPSTAT_INC(ips_fragments); m->m_pkthdr.PH_loc.ptr = ip; /* * Presence of header sizes in mbufs * would confuse code below. */ m->m_data += hlen; m->m_len -= hlen; hashkey[0] = ip->ip_src.s_addr; hashkey[1] = ip->ip_dst.s_addr; hashkey[2] = (uint32_t)ip->ip_p << 16; hashkey[2] += ip->ip_id; hash = jenkins_hash32(hashkey, nitems(hashkey), V_ipq_hashseed); hash &= IPREASS_HMASK; head = &V_ipq[hash].head; IPQ_LOCK(hash); /* * Look for queue of fragments * of this datagram. */ TAILQ_FOREACH(fp, head, ipq_list) if (ip->ip_id == fp->ipq_id && ip->ip_src.s_addr == fp->ipq_src.s_addr && ip->ip_dst.s_addr == fp->ipq_dst.s_addr && #ifdef MAC mac_ipq_match(m, fp) && #endif ip->ip_p == fp->ipq_p) break; /* * If first fragment to arrive, create a reassembly queue. */ if (fp == NULL) { if (V_ipq[hash].count < V_ipreass_maxbucketsize) fp = uma_zalloc(V_ipq_zone, M_NOWAIT); if (fp == NULL) fp = ipq_reuse(hash); if (fp == NULL) goto dropfrag; #ifdef MAC if (mac_ipq_init(fp, M_NOWAIT) != 0) { uma_zfree(V_ipq_zone, fp); fp = NULL; goto dropfrag; } mac_ipq_create(m, fp); #endif TAILQ_INSERT_HEAD(head, fp, ipq_list); V_ipq[hash].count++; fp->ipq_nfrags = 1; atomic_add_int(&nfrags, 1); fp->ipq_ttl = IPFRAGTTL; fp->ipq_p = ip->ip_p; fp->ipq_id = ip->ip_id; fp->ipq_src = ip->ip_src; fp->ipq_dst = ip->ip_dst; fp->ipq_frags = m; if (m->m_flags & M_IP_FRAG) fp->ipq_maxoff = -1; else fp->ipq_maxoff = ntohs(ip->ip_off) + ntohs(ip->ip_len); m->m_nextpkt = NULL; goto done; } else { /* * If we already saw the last fragment, make sure * this fragment's offset looks sane. Otherwise, if * this is the last fragment, record its endpoint. */ if (fp->ipq_maxoff > 0) { i = ntohs(ip->ip_off) + ntohs(ip->ip_len); if (((m->m_flags & M_IP_FRAG) && i >= fp->ipq_maxoff) || ((m->m_flags & M_IP_FRAG) == 0 && i != fp->ipq_maxoff)) { fp = NULL; goto dropfrag; } } else if ((m->m_flags & M_IP_FRAG) == 0) fp->ipq_maxoff = ntohs(ip->ip_off) + ntohs(ip->ip_len); fp->ipq_nfrags++; atomic_add_int(&nfrags, 1); #ifdef MAC mac_ipq_update(m, fp); #endif } #define GETIP(m) ((struct ip*)((m)->m_pkthdr.PH_loc.ptr)) /* * Handle ECN by comparing this segment with the first one; * if CE is set, do not lose CE. * drop if CE and not-ECT are mixed for the same packet. */ ecn = ip->ip_tos & IPTOS_ECN_MASK; ecn0 = GETIP(fp->ipq_frags)->ip_tos & IPTOS_ECN_MASK; if (ecn == IPTOS_ECN_CE) { if (ecn0 == IPTOS_ECN_NOTECT) goto dropfrag; if (ecn0 != IPTOS_ECN_CE) GETIP(fp->ipq_frags)->ip_tos |= IPTOS_ECN_CE; } if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) goto dropfrag; /* * Find a segment which begins after this one does. */ for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) if (ntohs(GETIP(q)->ip_off) > ntohs(ip->ip_off)) break; /* * If there is a preceding segment, it may provide some of * our data already. If so, drop the data from the incoming * segment. If it provides all of our data, drop us, otherwise * stick new segment in the proper place. * * If some of the data is dropped from the preceding * segment, then it's checksum is invalidated. */ if (p) { i = ntohs(GETIP(p)->ip_off) + ntohs(GETIP(p)->ip_len) - ntohs(ip->ip_off); if (i > 0) { if (i >= ntohs(ip->ip_len)) goto dropfrag; m_adj(m, i); m->m_pkthdr.csum_flags = 0; ip->ip_off = htons(ntohs(ip->ip_off) + i); ip->ip_len = htons(ntohs(ip->ip_len) - i); } m->m_nextpkt = p->m_nextpkt; p->m_nextpkt = m; } else { m->m_nextpkt = fp->ipq_frags; fp->ipq_frags = m; } /* * While we overlap succeeding segments trim them or, * if they are completely covered, dequeue them. */ for (; q != NULL && ntohs(ip->ip_off) + ntohs(ip->ip_len) > ntohs(GETIP(q)->ip_off); q = nq) { i = (ntohs(ip->ip_off) + ntohs(ip->ip_len)) - ntohs(GETIP(q)->ip_off); if (i < ntohs(GETIP(q)->ip_len)) { GETIP(q)->ip_len = htons(ntohs(GETIP(q)->ip_len) - i); GETIP(q)->ip_off = htons(ntohs(GETIP(q)->ip_off) + i); m_adj(q, i); q->m_pkthdr.csum_flags = 0; break; } nq = q->m_nextpkt; m->m_nextpkt = nq; IPSTAT_INC(ips_fragdropped); fp->ipq_nfrags--; atomic_subtract_int(&nfrags, 1); m_freem(q); } /* * Check for complete reassembly and perform frag per packet * limiting. * * Frag limiting is performed here so that the nth frag has * a chance to complete the packet before we drop the packet. * As a result, n+1 frags are actually allowed per packet, but * only n will ever be stored. (n = maxfragsperpacket.) * */ next = 0; for (p = NULL, q = fp->ipq_frags; q; p = q, q = q->m_nextpkt) { if (ntohs(GETIP(q)->ip_off) != next) { if (fp->ipq_nfrags > V_maxfragsperpacket) ipq_drop(&V_ipq[hash], fp); goto done; } next += ntohs(GETIP(q)->ip_len); } /* Make sure the last packet didn't have the IP_MF flag */ if (p->m_flags & M_IP_FRAG) { if (fp->ipq_nfrags > V_maxfragsperpacket) ipq_drop(&V_ipq[hash], fp); goto done; } /* * Reassembly is complete. Make sure the packet is a sane size. */ q = fp->ipq_frags; ip = GETIP(q); if (next + (ip->ip_hl << 2) > IP_MAXPACKET) { IPSTAT_INC(ips_toolong); ipq_drop(&V_ipq[hash], fp); goto done; } /* * Concatenate fragments. */ m = q; t = m->m_next; m->m_next = NULL; m_cat(m, t); nq = q->m_nextpkt; q->m_nextpkt = NULL; for (q = nq; q != NULL; q = nq) { nq = q->m_nextpkt; q->m_nextpkt = NULL; m->m_pkthdr.csum_flags &= q->m_pkthdr.csum_flags; m->m_pkthdr.csum_data += q->m_pkthdr.csum_data; m_demote_pkthdr(q); m_cat(m, q); } /* * In order to do checksumming faster we do 'end-around carry' here * (and not in for{} loop), though it implies we are not going to * reassemble more than 64k fragments. */ while (m->m_pkthdr.csum_data & 0xffff0000) m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16); atomic_subtract_int(&nfrags, fp->ipq_nfrags); #ifdef MAC mac_ipq_reassemble(fp, m); mac_ipq_destroy(fp); #endif /* * Create header for new ip packet by modifying header of first * packet; dequeue and discard fragment reassembly header. * Make header visible. */ ip->ip_len = htons((ip->ip_hl << 2) + next); ip->ip_src = fp->ipq_src; ip->ip_dst = fp->ipq_dst; TAILQ_REMOVE(head, fp, ipq_list); V_ipq[hash].count--; uma_zfree(V_ipq_zone, fp); m->m_len += (ip->ip_hl << 2); m->m_data -= (ip->ip_hl << 2); /* some debugging cruft by sklower, below, will go away soon */ if (m->m_flags & M_PKTHDR) /* XXX this should be done elsewhere */ m_fixhdr(m); IPSTAT_INC(ips_reassembled); IPQ_UNLOCK(hash); #ifdef RSS /* * Query the RSS layer for the flowid / flowtype for the * mbuf payload. * * For now, just assume we have to calculate a new one. * Later on we should check to see if the assigned flowid matches * what RSS wants for the given IP protocol and if so, just keep it. * * We then queue into the relevant netisr so it can be dispatched * to the correct CPU. * * Note - this may return 1, which means the flowid in the mbuf * is correct for the configured RSS hash types and can be used. */ if (rss_mbuf_software_hash_v4(m, 0, &rss_hash, &rss_type) == 0) { m->m_pkthdr.flowid = rss_hash; M_HASHTYPE_SET(m, rss_type); } /* * Queue/dispatch for reprocessing. * * Note: this is much slower than just handling the frame in the * current receive context. It's likely worth investigating * why this is. */ netisr_dispatch(NETISR_IP_DIRECT, m); return (NULL); #endif /* Handle in-line */ return (m); dropfrag: IPSTAT_INC(ips_fragdropped); if (fp != NULL) { fp->ipq_nfrags--; atomic_subtract_int(&nfrags, 1); } m_freem(m); done: IPQ_UNLOCK(hash); return (NULL); #undef GETIP } /* * Initialize IP reassembly structures. */ void ipreass_init(void) { int max; for (int i = 0; i < IPREASS_NHASH; i++) { TAILQ_INIT(&V_ipq[i].head); mtx_init(&V_ipq[i].lock, "IP reassembly", NULL, MTX_DEF | MTX_DUPOK); V_ipq[i].count = 0; } V_ipq_hashseed = arc4random(); V_maxfragsperpacket = 16; V_ipq_zone = uma_zcreate("ipq", sizeof(struct ipq), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); max = IP_MAXFRAGPACKETS; max = uma_zone_set_max(V_ipq_zone, max); V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); if (IS_DEFAULT_VNET(curvnet)) { maxfrags = IP_MAXFRAGS; EVENTHANDLER_REGISTER(nmbclusters_change, ipreass_zone_change, NULL, EVENTHANDLER_PRI_ANY); } } /* * If a timer expires on a reassembly queue, discard it. */ void ipreass_slowtimo(void) { struct ipq *fp, *tmp; for (int i = 0; i < IPREASS_NHASH; i++) { IPQ_LOCK(i); TAILQ_FOREACH_SAFE(fp, &V_ipq[i].head, ipq_list, tmp) if (--fp->ipq_ttl == 0) ipq_timeout(&V_ipq[i], fp); IPQ_UNLOCK(i); } } /* * Drain off all datagram fragments. */ void ipreass_drain(void) { for (int i = 0; i < IPREASS_NHASH; i++) { IPQ_LOCK(i); while(!TAILQ_EMPTY(&V_ipq[i].head)) ipq_drop(&V_ipq[i], TAILQ_FIRST(&V_ipq[i].head)); KASSERT(V_ipq[i].count == 0, ("%s: V_ipq[%d] count %d (V_ipq=%p)", __func__, i, V_ipq[i].count, V_ipq)); IPQ_UNLOCK(i); } } + +/* + * Drain off all datagram fragments belonging to + * the given network interface. + */ +static void +ipreass_cleanup(void *arg __unused, struct ifnet *ifp) +{ + struct ipq *fp, *temp; + struct mbuf *m; + int i; + + KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); + + CURVNET_SET_QUIET(ifp->if_vnet); + for (i = 0; i < IPREASS_NHASH; i++) { + IPQ_LOCK(i); + /* Scan fragment list. */ + TAILQ_FOREACH_SAFE(fp, &V_ipq[i].head, ipq_list, temp) { + for (m = fp->ipq_frags; m != NULL; m = m->m_nextpkt) { + if (m->m_pkthdr.rcvif == ifp) { + ipq_drop(&V_ipq[i], fp); + break; + } + } + } + IPQ_UNLOCK(i); + } + CURVNET_RESTORE(); +} +EVENTHANDLER_DEFINE(ifnet_departure_event, ipreass_cleanup, NULL, 0); #ifdef VIMAGE /* * Destroy IP reassembly structures. */ void ipreass_destroy(void) { ipreass_drain(); uma_zdestroy(V_ipq_zone); for (int i = 0; i < IPREASS_NHASH; i++) mtx_destroy(&V_ipq[i].lock); } #endif /* * After maxnipq has been updated, propagate the change to UMA. The UMA zone * max has slightly different semantics than the sysctl, for historical * reasons. */ static void ipreass_drain_tomax(void) { struct ipq *fp; int target; /* * Make sure each bucket is under the new limit. If * necessary, drop enough of the oldest elements from * each bucket to get under the new limit. */ for (int i = 0; i < IPREASS_NHASH; i++) { IPQ_LOCK(i); while (V_ipq[i].count > V_ipreass_maxbucketsize && (fp = TAILQ_LAST(&V_ipq[i].head, ipqhead)) != NULL) ipq_timeout(&V_ipq[i], fp); IPQ_UNLOCK(i); } /* * If we are over the maximum number of fragments, * drain off enough to get down to the new limit, * stripping off last elements on queues. Every * run we strip the oldest element from each bucket. */ target = uma_zone_get_max(V_ipq_zone); while (uma_zone_get_cur(V_ipq_zone) > target) { for (int i = 0; i < IPREASS_NHASH; i++) { IPQ_LOCK(i); fp = TAILQ_LAST(&V_ipq[i].head, ipqhead); if (fp != NULL) ipq_timeout(&V_ipq[i], fp); IPQ_UNLOCK(i); } } } static void ipreass_zone_change(void *tag) { VNET_ITERATOR_DECL(vnet_iter); int max; maxfrags = IP_MAXFRAGS; max = IP_MAXFRAGPACKETS; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); max = uma_zone_set_max(V_ipq_zone, max); V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); ipreass_drain_tomax(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Change the limit on the UMA zone, or disable the fragment allocation * at all. Since 0 and -1 is a special values here, we need our own handler, * instead of sysctl_handle_uma_zone_max(). */ static int sysctl_maxfragpackets(SYSCTL_HANDLER_ARGS) { int error, max; if (V_noreass == 0) { max = uma_zone_get_max(V_ipq_zone); if (max == 0) max = -1; } else max = 0; error = sysctl_handle_int(oidp, &max, 0, req); if (error || !req->newptr) return (error); if (max > 0) { /* * XXXRW: Might be a good idea to sanity check the argument * and place an extreme upper bound. */ max = uma_zone_set_max(V_ipq_zone, max); V_ipreass_maxbucketsize = imax(max / (IPREASS_NHASH / 2), 1); ipreass_drain_tomax(); V_noreass = 0; } else if (max == 0) { V_noreass = 1; ipreass_drain(); } else if (max == -1) { V_noreass = 0; uma_zone_set_max(V_ipq_zone, 0); V_ipreass_maxbucketsize = INT_MAX; } else return (EINVAL); return (0); } /* * Seek for old fragment queue header that can be reused. Try to * reuse a header from currently locked hash bucket. */ static struct ipq * ipq_reuse(int start) { struct ipq *fp; int bucket, i; IPQ_LOCK_ASSERT(start); for (i = 0; i < IPREASS_NHASH; i++) { bucket = (start + i) % IPREASS_NHASH; if (bucket != start && IPQ_TRYLOCK(bucket) == 0) continue; fp = TAILQ_LAST(&V_ipq[bucket].head, ipqhead); if (fp) { struct mbuf *m; IPSTAT_ADD(ips_fragtimeout, fp->ipq_nfrags); atomic_subtract_int(&nfrags, fp->ipq_nfrags); while (fp->ipq_frags) { m = fp->ipq_frags; fp->ipq_frags = m->m_nextpkt; m_freem(m); } TAILQ_REMOVE(&V_ipq[bucket].head, fp, ipq_list); V_ipq[bucket].count--; if (bucket != start) IPQ_UNLOCK(bucket); break; } if (bucket != start) IPQ_UNLOCK(bucket); } IPQ_LOCK_ASSERT(start); return (fp); } /* * Free a fragment reassembly header and all associated datagrams. */ static void ipq_free(struct ipqbucket *bucket, struct ipq *fp) { struct mbuf *q; atomic_subtract_int(&nfrags, fp->ipq_nfrags); while (fp->ipq_frags) { q = fp->ipq_frags; fp->ipq_frags = q->m_nextpkt; m_freem(q); } TAILQ_REMOVE(&bucket->head, fp, ipq_list); bucket->count--; uma_zfree(V_ipq_zone, fp); } /* * Get or set the maximum number of reassembly queues per bucket. */ static int sysctl_maxfragbucketsize(SYSCTL_HANDLER_ARGS) { int error, max; max = V_ipreass_maxbucketsize; error = sysctl_handle_int(oidp, &max, 0, req); if (error || !req->newptr) return (error); if (max <= 0) return (EINVAL); V_ipreass_maxbucketsize = max; ipreass_drain_tomax(); return (0); } Index: head/sys/netinet6/frag6.c =================================================================== --- head/sys/netinet6/frag6.c (revision 346529) +++ head/sys/netinet6/frag6.c (revision 346530) @@ -1,966 +1,1005 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the project nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: frag6.c,v 1.33 2002/01/07 11:34:48 kjc Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_rss.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* for ECN definitions */ #include /* for ECN definitions */ #include /* * Reassembly headers are stored in hash buckets. */ #define IP6REASS_NHASH_LOG2 10 #define IP6REASS_NHASH (1 << IP6REASS_NHASH_LOG2) #define IP6REASS_HMASK (IP6REASS_NHASH - 1) static void frag6_enq(struct ip6asfrag *, struct ip6asfrag *, uint32_t bucket __unused); static void frag6_deq(struct ip6asfrag *, uint32_t bucket __unused); static void frag6_insque_head(struct ip6q *, struct ip6q *, uint32_t bucket); static void frag6_remque(struct ip6q *, uint32_t bucket); -static void frag6_freef(struct ip6q *, uint32_t bucket); +static void frag6_freef(struct ip6q *, uint32_t bucket, bool send_icmp); struct ip6qbucket { struct ip6q ip6q; struct mtx lock; int count; }; VNET_DEFINE_STATIC(volatile u_int, frag6_nfragpackets); volatile u_int frag6_nfrags = 0; VNET_DEFINE_STATIC(struct ip6qbucket, ip6q[IP6REASS_NHASH]); VNET_DEFINE_STATIC(uint32_t, ip6q_hashseed); #define V_frag6_nfragpackets VNET(frag6_nfragpackets) #define V_ip6q VNET(ip6q) #define V_ip6q_hashseed VNET(ip6q_hashseed) #define IP6Q_LOCK(i) mtx_lock(&V_ip6q[(i)].lock) #define IP6Q_TRYLOCK(i) mtx_trylock(&V_ip6q[(i)].lock) #define IP6Q_LOCK_ASSERT(i) mtx_assert(&V_ip6q[(i)].lock, MA_OWNED) #define IP6Q_UNLOCK(i) mtx_unlock(&V_ip6q[(i)].lock) #define IP6Q_HEAD(i) (&V_ip6q[(i)].ip6q) static MALLOC_DEFINE(M_FTABLE, "fragment", "fragment reassembly header"); /* * By default, limit the number of IP6 fragments across all reassembly * queues to 1/32 of the total number of mbuf clusters. * * Limit the total number of reassembly queues per VNET to the * IP6 fragment limit, but ensure the limit will not allow any bucket * to grow above 100 items. (The bucket limit is * IP_MAXFRAGPACKETS / (IPREASS_NHASH / 2), so the 50 is the correct * multiplier to reach a 100-item limit.) * The 100-item limit was chosen as brief testing seems to show that * this produces "reasonable" performance on some subset of systems * under DoS attack. */ #define IP6_MAXFRAGS (nmbclusters / 32) #define IP6_MAXFRAGPACKETS (imin(IP6_MAXFRAGS, IP6REASS_NHASH * 50)) /* * Initialise reassembly queue and fragment identifier. */ void frag6_set_bucketsize() { int i; if ((i = V_ip6_maxfragpackets) > 0) V_ip6_maxfragbucketsize = imax(i / (IP6REASS_NHASH / 2), 1); } static void frag6_change(void *tag) { VNET_ITERATOR_DECL(vnet_iter); ip6_maxfrags = IP6_MAXFRAGS; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; frag6_set_bucketsize(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } void frag6_init(void) { struct ip6q *q6; int i; V_ip6_maxfragpackets = IP6_MAXFRAGPACKETS; frag6_set_bucketsize(); for (i = 0; i < IP6REASS_NHASH; i++) { q6 = IP6Q_HEAD(i); q6->ip6q_next = q6->ip6q_prev = q6; mtx_init(&V_ip6q[i].lock, "ip6qlock", NULL, MTX_DEF); V_ip6q[i].count = 0; } V_ip6q_hashseed = arc4random(); V_ip6_maxfragsperpacket = 64; if (!IS_DEFAULT_VNET(curvnet)) return; ip6_maxfrags = IP6_MAXFRAGS; EVENTHANDLER_REGISTER(nmbclusters_change, frag6_change, NULL, EVENTHANDLER_PRI_ANY); } /* * In RFC2460, fragment and reassembly rule do not agree with each other, * in terms of next header field handling in fragment header. * While the sender will use the same value for all of the fragmented packets, * receiver is suggested not to check the consistency. * * fragment rule (p20): * (2) A Fragment header containing: * The Next Header value that identifies the first header of * the Fragmentable Part of the original packet. * -> next header field is same for all fragments * * reassembly rule (p21): * The Next Header field of the last header of the Unfragmentable * Part is obtained from the Next Header field of the first * fragment's Fragment header. * -> should grab it from the first fragment only * * The following note also contradicts with fragment rule - no one is going to * send different fragment with different next header field. * * additional note (p22): * The Next Header values in the Fragment headers of different * fragments of the same original packet may differ. Only the value * from the Offset zero fragment packet is used for reassembly. * -> should grab it from the first fragment only * * There is no explicit reason given in the RFC. Historical reason maybe? */ /* * Fragment input */ int frag6_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp, *t; struct ip6_hdr *ip6; struct ip6_frag *ip6f; struct ip6q *head, *q6; struct ip6asfrag *af6, *ip6af, *af6dwn; struct in6_ifaddr *ia; int offset = *offp, nxt, i, next; int first_frag = 0; int fragoff, frgpartlen; /* must be larger than u_int16_t */ uint32_t hashkey[(sizeof(struct in6_addr) * 2 + sizeof(ip6f->ip6f_ident)) / sizeof(uint32_t)]; uint32_t hash, *hashkeyp; struct ifnet *dstifp; u_int8_t ecn, ecn0; #ifdef RSS struct m_tag *mtag; struct ip6_direct_ctx *ip6dc; #endif #if 0 char ip6buf[INET6_ADDRSTRLEN]; #endif ip6 = mtod(m, struct ip6_hdr *); #ifndef PULLDOWN_TEST IP6_EXTHDR_CHECK(m, offset, sizeof(struct ip6_frag), IPPROTO_DONE); ip6f = (struct ip6_frag *)((caddr_t)ip6 + offset); #else IP6_EXTHDR_GET(ip6f, struct ip6_frag *, m, offset, sizeof(*ip6f)); if (ip6f == NULL) return (IPPROTO_DONE); #endif dstifp = NULL; /* find the destination interface of the packet. */ ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia != NULL) { dstifp = ia->ia_ifp; ifa_free(&ia->ia_ifa); } /* jumbo payload can't contain a fragment header */ if (ip6->ip6_plen == 0) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset); in6_ifstat_inc(dstifp, ifs6_reass_fail); return IPPROTO_DONE; } /* * check whether fragment packet's fragment length is * multiple of 8 octets. * sizeof(struct ip6_frag) == 8 * sizeof(struct ip6_hdr) = 40 */ if ((ip6f->ip6f_offlg & IP6F_MORE_FRAG) && (((ntohs(ip6->ip6_plen) - offset) & 0x7) != 0)) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offsetof(struct ip6_hdr, ip6_plen)); in6_ifstat_inc(dstifp, ifs6_reass_fail); return IPPROTO_DONE; } IP6STAT_INC(ip6s_fragments); in6_ifstat_inc(dstifp, ifs6_reass_reqd); /* offset now points to data portion */ offset += sizeof(struct ip6_frag); /* * Handle "atomic" fragments (offset and m bit set to 0) upfront, * unrelated to any reassembly (see RFC 6946 and section 4.5 of RFC * 8200). Just skip the fragment header. */ if ((ip6f->ip6f_offlg & ~IP6F_RESERVED_MASK) == 0) { IP6STAT_INC(ip6s_atomicfrags); in6_ifstat_inc(dstifp, ifs6_reass_ok); *offp = offset; m->m_flags |= M_FRAGMENTED; return (ip6f->ip6f_nxt); } /* Get fragment length and discard 0-byte fragments. */ frgpartlen = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen) - offset; if (frgpartlen == 0) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offsetof(struct ip6_hdr, ip6_plen)); in6_ifstat_inc(dstifp, ifs6_reass_fail); IP6STAT_INC(ip6s_fragdropped); return IPPROTO_DONE; } hashkeyp = hashkey; memcpy(hashkeyp, &ip6->ip6_src, sizeof(struct in6_addr)); hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); memcpy(hashkeyp, &ip6->ip6_dst, sizeof(struct in6_addr)); hashkeyp += sizeof(struct in6_addr) / sizeof(*hashkeyp); *hashkeyp = ip6f->ip6f_ident; hash = jenkins_hash32(hashkey, nitems(hashkey), V_ip6q_hashseed); hash &= IP6REASS_HMASK; head = IP6Q_HEAD(hash); IP6Q_LOCK(hash); /* * Enforce upper bound on number of fragments. * If maxfrag is 0, never accept fragments. * If maxfrag is -1, accept all fragments without limitation. */ if (ip6_maxfrags < 0) ; else if (atomic_load_int(&frag6_nfrags) >= (u_int)ip6_maxfrags) goto dropfrag; for (q6 = head->ip6q_next; q6 != head; q6 = q6->ip6q_next) if (ip6f->ip6f_ident == q6->ip6q_ident && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &q6->ip6q_src) && IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &q6->ip6q_dst) #ifdef MAC && mac_ip6q_match(m, q6) #endif ) break; if (q6 == head) { /* * the first fragment to arrive, create a reassembly queue. */ first_frag = 1; /* * Enforce upper bound on number of fragmented packets * for which we attempt reassembly; * If maxfragpackets is 0, never accept fragments. * If maxfragpackets is -1, accept all fragments without * limitation. */ if (V_ip6_maxfragpackets < 0) ; else if (V_ip6q[hash].count >= V_ip6_maxfragbucketsize || atomic_load_int(&V_frag6_nfragpackets) >= (u_int)V_ip6_maxfragpackets) goto dropfrag; atomic_add_int(&V_frag6_nfragpackets, 1); q6 = (struct ip6q *)malloc(sizeof(struct ip6q), M_FTABLE, M_NOWAIT); if (q6 == NULL) goto dropfrag; bzero(q6, sizeof(*q6)); #ifdef MAC if (mac_ip6q_init(q6, M_NOWAIT) != 0) { free(q6, M_FTABLE); goto dropfrag; } mac_ip6q_create(m, q6); #endif frag6_insque_head(q6, head, hash); /* ip6q_nxt will be filled afterwards, from 1st fragment */ q6->ip6q_down = q6->ip6q_up = (struct ip6asfrag *)q6; #ifdef notyet q6->ip6q_nxtp = (u_char *)nxtp; #endif q6->ip6q_ident = ip6f->ip6f_ident; q6->ip6q_ttl = IPV6_FRAGTTL; q6->ip6q_src = ip6->ip6_src; q6->ip6q_dst = ip6->ip6_dst; q6->ip6q_ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; q6->ip6q_unfrglen = -1; /* The 1st fragment has not arrived. */ q6->ip6q_nfrag = 0; } /* * If it's the 1st fragment, record the length of the * unfragmentable part and the next header of the fragment header. */ fragoff = ntohs(ip6f->ip6f_offlg & IP6F_OFF_MASK); if (fragoff == 0) { q6->ip6q_unfrglen = offset - sizeof(struct ip6_hdr) - sizeof(struct ip6_frag); q6->ip6q_nxt = ip6f->ip6f_nxt; } /* * Check that the reassembled packet would not exceed 65535 bytes * in size. * If it would exceed, discard the fragment and return an ICMP error. */ if (q6->ip6q_unfrglen >= 0) { /* The 1st fragment has already arrived. */ if (q6->ip6q_unfrglen + fragoff + frgpartlen > IPV6_MAXPACKET) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset - sizeof(struct ip6_frag) + offsetof(struct ip6_frag, ip6f_offlg)); IP6Q_UNLOCK(hash); return (IPPROTO_DONE); } } else if (fragoff + frgpartlen > IPV6_MAXPACKET) { icmp6_error(m, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, offset - sizeof(struct ip6_frag) + offsetof(struct ip6_frag, ip6f_offlg)); IP6Q_UNLOCK(hash); return (IPPROTO_DONE); } /* * If it's the first fragment, do the above check for each * fragment already stored in the reassembly queue. */ if (fragoff == 0) { for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = af6dwn) { af6dwn = af6->ip6af_down; if (q6->ip6q_unfrglen + af6->ip6af_off + af6->ip6af_frglen > IPV6_MAXPACKET) { struct mbuf *merr = IP6_REASS_MBUF(af6); struct ip6_hdr *ip6err; int erroff = af6->ip6af_offset; /* dequeue the fragment. */ frag6_deq(af6, hash); free(af6, M_FTABLE); /* adjust pointer. */ ip6err = mtod(merr, struct ip6_hdr *); /* * Restore source and destination addresses * in the erroneous IPv6 header. */ ip6err->ip6_src = q6->ip6q_src; ip6err->ip6_dst = q6->ip6q_dst; icmp6_error(merr, ICMP6_PARAM_PROB, ICMP6_PARAMPROB_HEADER, erroff - sizeof(struct ip6_frag) + offsetof(struct ip6_frag, ip6f_offlg)); } } } ip6af = (struct ip6asfrag *)malloc(sizeof(struct ip6asfrag), M_FTABLE, M_NOWAIT); if (ip6af == NULL) goto dropfrag; bzero(ip6af, sizeof(*ip6af)); ip6af->ip6af_mff = ip6f->ip6f_offlg & IP6F_MORE_FRAG; ip6af->ip6af_off = fragoff; ip6af->ip6af_frglen = frgpartlen; ip6af->ip6af_offset = offset; IP6_REASS_MBUF(ip6af) = m; if (first_frag) { af6 = (struct ip6asfrag *)q6; goto insert; } /* * Handle ECN by comparing this segment with the first one; * if CE is set, do not lose CE. * drop if CE and not-ECT are mixed for the same packet. */ ecn = (ntohl(ip6->ip6_flow) >> 20) & IPTOS_ECN_MASK; ecn0 = q6->ip6q_ecn; if (ecn == IPTOS_ECN_CE) { if (ecn0 == IPTOS_ECN_NOTECT) { free(ip6af, M_FTABLE); goto dropfrag; } if (ecn0 != IPTOS_ECN_CE) q6->ip6q_ecn = IPTOS_ECN_CE; } if (ecn == IPTOS_ECN_NOTECT && ecn0 != IPTOS_ECN_NOTECT) { free(ip6af, M_FTABLE); goto dropfrag; } /* * Find a segment which begins after this one does. */ for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = af6->ip6af_down) if (af6->ip6af_off > ip6af->ip6af_off) break; #if 0 /* * If there is a preceding segment, it may provide some of * our data already. If so, drop the data from the incoming * segment. If it provides all of our data, drop us. */ if (af6->ip6af_up != (struct ip6asfrag *)q6) { i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen - ip6af->ip6af_off; if (i > 0) { if (i >= ip6af->ip6af_frglen) goto dropfrag; m_adj(IP6_REASS_MBUF(ip6af), i); ip6af->ip6af_off += i; ip6af->ip6af_frglen -= i; } } /* * While we overlap succeeding segments trim them or, * if they are completely covered, dequeue them. */ while (af6 != (struct ip6asfrag *)q6 && ip6af->ip6af_off + ip6af->ip6af_frglen > af6->ip6af_off) { i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off; if (i < af6->ip6af_frglen) { af6->ip6af_frglen -= i; af6->ip6af_off += i; m_adj(IP6_REASS_MBUF(af6), i); break; } af6 = af6->ip6af_down; m_freem(IP6_REASS_MBUF(af6->ip6af_up)); frag6_deq(af6->ip6af_up, hash); } #else /* * If the incoming framgent overlaps some existing fragments in * the reassembly queue, drop it, since it is dangerous to override * existing fragments from a security point of view. * We don't know which fragment is the bad guy - here we trust * fragment that came in earlier, with no real reason. * * Note: due to changes after disabling this part, mbuf passed to * m_adj() below now does not meet the requirement. */ if (af6->ip6af_up != (struct ip6asfrag *)q6) { i = af6->ip6af_up->ip6af_off + af6->ip6af_up->ip6af_frglen - ip6af->ip6af_off; if (i > 0) { #if 0 /* suppress the noisy log */ log(LOG_ERR, "%d bytes of a fragment from %s " "overlaps the previous fragment\n", i, ip6_sprintf(ip6buf, &q6->ip6q_src)); #endif free(ip6af, M_FTABLE); goto dropfrag; } } if (af6 != (struct ip6asfrag *)q6) { i = (ip6af->ip6af_off + ip6af->ip6af_frglen) - af6->ip6af_off; if (i > 0) { #if 0 /* suppress the noisy log */ log(LOG_ERR, "%d bytes of a fragment from %s " "overlaps the succeeding fragment", i, ip6_sprintf(ip6buf, &q6->ip6q_src)); #endif free(ip6af, M_FTABLE); goto dropfrag; } } #endif insert: #ifdef MAC if (!first_frag) mac_ip6q_update(m, q6); #endif /* * Stick new segment in its place; * check for complete reassembly. * If not complete, check fragment limit. * Move to front of packet queue, as we are * the most recently active fragmented packet. */ frag6_enq(ip6af, af6->ip6af_up, hash); atomic_add_int(&frag6_nfrags, 1); q6->ip6q_nfrag++; #if 0 /* xxx */ if (q6 != head->ip6q_next) { frag6_remque(q6, hash); frag6_insque_head(q6, head, hash); } #endif next = 0; for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = af6->ip6af_down) { if (af6->ip6af_off != next) { if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); - frag6_freef(q6, hash); + frag6_freef(q6, hash, true); } IP6Q_UNLOCK(hash); return IPPROTO_DONE; } next += af6->ip6af_frglen; } if (af6->ip6af_up->ip6af_mff) { if (q6->ip6q_nfrag > V_ip6_maxfragsperpacket) { IP6STAT_ADD(ip6s_fragdropped, q6->ip6q_nfrag); - frag6_freef(q6, hash); + frag6_freef(q6, hash, true); } IP6Q_UNLOCK(hash); return IPPROTO_DONE; } /* * Reassembly is complete; concatenate fragments. */ ip6af = q6->ip6q_down; t = m = IP6_REASS_MBUF(ip6af); af6 = ip6af->ip6af_down; frag6_deq(ip6af, hash); while (af6 != (struct ip6asfrag *)q6) { m->m_pkthdr.csum_flags &= IP6_REASS_MBUF(af6)->m_pkthdr.csum_flags; m->m_pkthdr.csum_data += IP6_REASS_MBUF(af6)->m_pkthdr.csum_data; af6dwn = af6->ip6af_down; frag6_deq(af6, hash); while (t->m_next) t = t->m_next; m_adj(IP6_REASS_MBUF(af6), af6->ip6af_offset); m_demote_pkthdr(IP6_REASS_MBUF(af6)); m_cat(t, IP6_REASS_MBUF(af6)); free(af6, M_FTABLE); af6 = af6dwn; } while (m->m_pkthdr.csum_data & 0xffff0000) m->m_pkthdr.csum_data = (m->m_pkthdr.csum_data & 0xffff) + (m->m_pkthdr.csum_data >> 16); /* adjust offset to point where the original next header starts */ offset = ip6af->ip6af_offset - sizeof(struct ip6_frag); free(ip6af, M_FTABLE); ip6 = mtod(m, struct ip6_hdr *); ip6->ip6_plen = htons((u_short)next + offset - sizeof(struct ip6_hdr)); if (q6->ip6q_ecn == IPTOS_ECN_CE) ip6->ip6_flow |= htonl(IPTOS_ECN_CE << 20); nxt = q6->ip6q_nxt; #ifdef notyet *q6->ip6q_nxtp = (u_char)(nxt & 0xff); #endif if (ip6_deletefraghdr(m, offset, M_NOWAIT) != 0) { frag6_remque(q6, hash); atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); #ifdef MAC mac_ip6q_destroy(q6); #endif free(q6, M_FTABLE); atomic_subtract_int(&V_frag6_nfragpackets, 1); goto dropfrag; } /* * Store NXT to the original. */ m_copyback(m, ip6_get_prevhdr(m, offset), sizeof(uint8_t), (caddr_t)&nxt); frag6_remque(q6, hash); atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); #ifdef MAC mac_ip6q_reassemble(q6, m); mac_ip6q_destroy(q6); #endif free(q6, M_FTABLE); atomic_subtract_int(&V_frag6_nfragpackets, 1); if (m->m_flags & M_PKTHDR) { /* Isn't it always true? */ int plen = 0; for (t = m; t; t = t->m_next) plen += t->m_len; m->m_pkthdr.len = plen; } #ifdef RSS mtag = m_tag_alloc(MTAG_ABI_IPV6, IPV6_TAG_DIRECT, sizeof(*ip6dc), M_NOWAIT); if (mtag == NULL) goto dropfrag; ip6dc = (struct ip6_direct_ctx *)(mtag + 1); ip6dc->ip6dc_nxt = nxt; ip6dc->ip6dc_off = offset; m_tag_prepend(m, mtag); #endif IP6Q_UNLOCK(hash); IP6STAT_INC(ip6s_reassembled); in6_ifstat_inc(dstifp, ifs6_reass_ok); #ifdef RSS /* * Queue/dispatch for reprocessing. */ netisr_dispatch(NETISR_IPV6_DIRECT, m); return IPPROTO_DONE; #endif /* * Tell launch routine the next header */ *mp = m; *offp = offset; return nxt; dropfrag: IP6Q_UNLOCK(hash); in6_ifstat_inc(dstifp, ifs6_reass_fail); IP6STAT_INC(ip6s_fragdropped); m_freem(m); return IPPROTO_DONE; } /* * Free a fragment reassembly header and all * associated datagrams. */ static void -frag6_freef(struct ip6q *q6, uint32_t bucket) +frag6_freef(struct ip6q *q6, uint32_t bucket, bool send_icmp) { struct ip6asfrag *af6, *down6; IP6Q_LOCK_ASSERT(bucket); for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; af6 = down6) { struct mbuf *m = IP6_REASS_MBUF(af6); down6 = af6->ip6af_down; frag6_deq(af6, bucket); /* * Return ICMP time exceeded error for the 1st fragment. * Just free other fragments. */ - if (af6->ip6af_off == 0) { + if (af6->ip6af_off == 0 && send_icmp != false) { struct ip6_hdr *ip6; /* adjust pointer */ ip6 = mtod(m, struct ip6_hdr *); /* restore source and destination addresses */ ip6->ip6_src = q6->ip6q_src; ip6->ip6_dst = q6->ip6q_dst; icmp6_error(m, ICMP6_TIME_EXCEEDED, ICMP6_TIME_EXCEED_REASSEMBLY, 0); } else m_freem(m); free(af6, M_FTABLE); } frag6_remque(q6, bucket); atomic_subtract_int(&frag6_nfrags, q6->ip6q_nfrag); #ifdef MAC mac_ip6q_destroy(q6); #endif free(q6, M_FTABLE); atomic_subtract_int(&V_frag6_nfragpackets, 1); } /* * Put an ip fragment on a reassembly chain. * Like insque, but pointers in middle of structure. */ static void frag6_enq(struct ip6asfrag *af6, struct ip6asfrag *up6, uint32_t bucket __unused) { IP6Q_LOCK_ASSERT(bucket); af6->ip6af_up = up6; af6->ip6af_down = up6->ip6af_down; up6->ip6af_down->ip6af_up = af6; up6->ip6af_down = af6; } /* * To frag6_enq as remque is to insque. */ static void frag6_deq(struct ip6asfrag *af6, uint32_t bucket __unused) { IP6Q_LOCK_ASSERT(bucket); af6->ip6af_up->ip6af_down = af6->ip6af_down; af6->ip6af_down->ip6af_up = af6->ip6af_up; } static void frag6_insque_head(struct ip6q *new, struct ip6q *old, uint32_t bucket) { IP6Q_LOCK_ASSERT(bucket); KASSERT(IP6Q_HEAD(bucket) == old, ("%s: attempt to insert at head of wrong bucket" " (bucket=%u, old=%p)", __func__, bucket, old)); new->ip6q_prev = old; new->ip6q_next = old->ip6q_next; old->ip6q_next->ip6q_prev= new; old->ip6q_next = new; V_ip6q[bucket].count++; } static void frag6_remque(struct ip6q *p6, uint32_t bucket) { IP6Q_LOCK_ASSERT(bucket); p6->ip6q_prev->ip6q_next = p6->ip6q_next; p6->ip6q_next->ip6q_prev = p6->ip6q_prev; V_ip6q[bucket].count--; } /* * IPv6 reassembling timer processing; * if a timer expires on a reassembly * queue, discard it. */ void frag6_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); struct ip6q *head, *q6; int i; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); for (i = 0; i < IP6REASS_NHASH; i++) { IP6Q_LOCK(i); head = IP6Q_HEAD(i); q6 = head->ip6q_next; if (q6 == NULL) { /* * XXXJTL: This should never happen. This * should turn into an assertion. */ IP6Q_UNLOCK(i); continue; } while (q6 != head) { --q6->ip6q_ttl; q6 = q6->ip6q_next; if (q6->ip6q_prev->ip6q_ttl == 0) { IP6STAT_ADD(ip6s_fragtimeout, q6->ip6q_prev->ip6q_nfrag); /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ - frag6_freef(q6->ip6q_prev, i); + frag6_freef(q6->ip6q_prev, i, true); } } /* * If we are over the maximum number of fragments * (due to the limit being lowered), drain off * enough to get down to the new limit. * Note that we drain all reassembly queues if * maxfragpackets is 0 (fragmentation is disabled), * and don't enforce a limit when maxfragpackets * is negative. */ while ((V_ip6_maxfragpackets == 0 || (V_ip6_maxfragpackets > 0 && V_ip6q[i].count > V_ip6_maxfragbucketsize)) && head->ip6q_prev != head) { IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_prev->ip6q_nfrag); /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ - frag6_freef(head->ip6q_prev, i); + frag6_freef(head->ip6q_prev, i, true); } IP6Q_UNLOCK(i); } /* * If we are still over the maximum number of fragmented * packets, drain off enough to get down to the new limit. */ i = 0; while (V_ip6_maxfragpackets >= 0 && atomic_load_int(&V_frag6_nfragpackets) > (u_int)V_ip6_maxfragpackets) { IP6Q_LOCK(i); head = IP6Q_HEAD(i); if (head->ip6q_prev != head) { IP6STAT_ADD(ip6s_fragoverflow, q6->ip6q_prev->ip6q_nfrag); /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ - frag6_freef(head->ip6q_prev, i); + frag6_freef(head->ip6q_prev, i, true); } IP6Q_UNLOCK(i); i = (i + 1) % IP6REASS_NHASH; } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Drain off all datagram fragments. */ void frag6_drain(void) { VNET_ITERATOR_DECL(vnet_iter); struct ip6q *head; int i; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); for (i = 0; i < IP6REASS_NHASH; i++) { if (IP6Q_TRYLOCK(i) == 0) continue; head = IP6Q_HEAD(i); while (head->ip6q_next != head) { IP6STAT_INC(ip6s_fragdropped); /* XXX in6_ifstat_inc(ifp, ifs6_reass_fail) */ - frag6_freef(head->ip6q_next, i); + frag6_freef(head->ip6q_next, i, true); } IP6Q_UNLOCK(i); } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } + +/* + * Drain off all datagram fragments belonging to + * the given network interface. + */ +static void +frag6_cleanup(void *arg __unused, struct ifnet *ifp) +{ + struct ip6q *q6, *q6n, *head; + struct ip6asfrag *af6; + struct mbuf *m; + int i; + + KASSERT(ifp != NULL, ("%s: ifp is NULL", __func__)); + + CURVNET_SET_QUIET(ifp->if_vnet); + for (i = 0; i < IP6REASS_NHASH; i++) { + IP6Q_LOCK(i); + head = IP6Q_HEAD(i); + /* Scan fragment list. */ + for (q6 = head->ip6q_next; q6 != head; q6 = q6n) { + q6n = q6->ip6q_next; + + for (af6 = q6->ip6q_down; af6 != (struct ip6asfrag *)q6; + af6 = af6->ip6af_down) { + m = IP6_REASS_MBUF(af6); + + if (m->m_pkthdr.rcvif == ifp) { + IP6STAT_INC(ip6s_fragdropped); + frag6_freef(q6, i, false); + break; + } + } + } + IP6Q_UNLOCK(i); + } + CURVNET_RESTORE(); +} +EVENTHANDLER_DEFINE(ifnet_departure_event, frag6_cleanup, NULL, 0); int ip6_deletefraghdr(struct mbuf *m, int offset, int wait) { struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); struct mbuf *t; /* Delete frag6 header. */ if (m->m_len >= offset + sizeof(struct ip6_frag)) { /* This is the only possible case with !PULLDOWN_TEST. */ bcopy(ip6, (char *)ip6 + sizeof(struct ip6_frag), offset); m->m_data += sizeof(struct ip6_frag); m->m_len -= sizeof(struct ip6_frag); } else { /* This comes with no copy if the boundary is on cluster. */ if ((t = m_split(m, offset, wait)) == NULL) return (ENOMEM); m_adj(t, sizeof(struct ip6_frag)); m_cat(m, t); } m->m_flags |= M_FRAGMENTED; return (0); }