X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/8f6c56a50524aa785f7e596d52dddfb331e18961..99c3a10404e5d1ef94397ab4df5a8b74711fc4d3:/bsd/netinet/udp_usrreq.c diff --git a/bsd/netinet/udp_usrreq.c b/bsd/netinet/udp_usrreq.c index c0376e448..e5462393d 100644 --- a/bsd/netinet/udp_usrreq.c +++ b/bsd/netinet/udp_usrreq.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -72,6 +72,10 @@ #include #include #include +#include +#include + +#include #include #include @@ -87,6 +91,7 @@ #include #include #if INET6 +#include #include #endif #include @@ -97,8 +102,8 @@ #if IPSEC #include +#include extern int ipsec_bypass; -extern lck_mtx_t *sadb_mutex; #endif /*IPSEC*/ @@ -118,15 +123,35 @@ static int udpcksum = 1; #else static int udpcksum = 0; /* XXX */ #endif -SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, +SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW | CTLFLAG_LOCKED, &udpcksum, 0, ""); +static u_int32_t udps_in_sw_cksum; +SYSCTL_UINT(_net_inet_udp, OID_AUTO, in_sw_cksum, CTLFLAG_RD | CTLFLAG_LOCKED, + &udps_in_sw_cksum, 0, + "Number of received packets checksummed in software"); + +static u_int64_t udps_in_sw_cksum_bytes; +SYSCTL_QUAD(_net_inet_udp, OID_AUTO, in_sw_cksum_bytes, CTLFLAG_RD | CTLFLAG_LOCKED, + &udps_in_sw_cksum_bytes, + "Amount of received data checksummed in software"); + +static u_int32_t udps_out_sw_cksum; +SYSCTL_UINT(_net_inet_udp, OID_AUTO, out_sw_cksum, CTLFLAG_RD | CTLFLAG_LOCKED, + &udps_out_sw_cksum, 0, + "Number of transmitted packets checksummed in software"); + +static u_int64_t udps_out_sw_cksum_bytes; +SYSCTL_QUAD(_net_inet_udp, OID_AUTO, out_sw_cksum_bytes, CTLFLAG_RD | CTLFLAG_LOCKED, + &udps_out_sw_cksum_bytes, + "Amount of transmitted data checksummed in software"); + int log_in_vain = 0; -SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, +SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW | CTLFLAG_LOCKED, &log_in_vain, 0, "Log all incoming UDP packets"); static int blackhole = 0; -SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, +SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW | CTLFLAG_LOCKED, &blackhole, 0, "Do not send port unreachables for refused connects"); struct inpcbhead udb; /* from udp_var.h */ @@ -137,46 +162,50 @@ struct inpcbinfo udbinfo; #define UDBHASHSIZE 16 #endif -extern int apple_hwcksum_rx; extern int esp_udp_encap_port; -extern u_long route_generation; -extern void ipfwsyslog( int level, char *format,...); +extern void ipfwsyslog( int level, const char *format,...); extern int fw_verbose; +static int udp_gc_done = FALSE; /* Garbage collection performed last slowtimo */ +#if IPFIREWALL #define log_in_vain_log( a ) { \ if ( (log_in_vain == 3 ) && (fw_verbose == 2)) { /* Apple logging, log to ipfw.log */ \ ipfwsyslog a ; \ } \ else log a ; \ } +#else +#define log_in_vain_log( a ) { log a; } +#endif struct udpstat udpstat; /* from udp_var.h */ -SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RD, +SYSCTL_STRUCT(_net_inet_udp, UDPCTL_STATS, stats, CTLFLAG_RD | CTLFLAG_LOCKED, &udpstat, udpstat, "UDP statistics (struct udpstat, netinet/udp_var.h)"); -SYSCTL_INT(_net_inet_udp, OID_AUTO, pcbcount, CTLFLAG_RD, +SYSCTL_INT(_net_inet_udp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED, &udbinfo.ipi_count, 0, "Number of active PCBs"); -static struct sockaddr_in udp_in = { sizeof(udp_in), AF_INET }; +__private_extern__ int udp_use_randomport = 1; +SYSCTL_INT(_net_inet_udp, OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED, + &udp_use_randomport, 0, "Randomize UDP port numbers"); + #if INET6 struct udp_in6 { struct sockaddr_in6 uin6_sin; u_char uin6_init_done : 1; -} udp_in6 = { - { sizeof(udp_in6.uin6_sin), AF_INET6 }, - 0 }; struct udp_ip6 { struct ip6_hdr uip6_ip6; u_char uip6_init_done : 1; -} udp_ip6; -#endif /* INET6 */ - -static void udp_append(struct inpcb *last, struct ip *ip, - struct mbuf *n, int off); -#if INET6 +}; static void ip_2_ip6_hdr(struct ip6_hdr *ip6, struct ip *ip); +static void udp_append(struct inpcb *last, struct ip *ip, + struct mbuf *n, int off, struct sockaddr_in *pudp_in, + struct udp_in6 *pudp_in6, struct udp_ip6 *pudp_ip6); +#else +static void udp_append(struct inpcb *last, struct ip *ip, + struct mbuf *n, int off, struct sockaddr_in *pudp_in); #endif static int udp_detach(struct socket *so); @@ -205,47 +234,17 @@ udp_init() * allocate lock group attribute and group for udp pcb mutexes */ pcbinfo->mtx_grp_attr = lck_grp_attr_alloc_init(); - lck_grp_attr_setdefault(pcbinfo->mtx_grp_attr); pcbinfo->mtx_grp = lck_grp_alloc_init("udppcb", pcbinfo->mtx_grp_attr); pcbinfo->mtx_attr = lck_attr_alloc_init(); - lck_attr_setdefault(pcbinfo->mtx_attr); if ((pcbinfo->mtx = lck_rw_alloc_init(pcbinfo->mtx_grp, pcbinfo->mtx_attr)) == NULL) return; /* pretty much dead if this fails... */ - - in_pcb_nat_init(&udbinfo, AF_INET, IPPROTO_UDP, SOCK_DGRAM); #else udbinfo.ipi_zone = zinit("udpcb", sizeof(struct inpcb), maxsockets, ZONE_INTERRUPT, 0); #endif - -#if 0 - /* for pcb sharing testing only */ - stat = in_pcb_new_share_client(&udbinfo, &fake_owner); - kprintf("udp_init in_pcb_new_share_client - stat = %d\n", stat); - - laddr.s_addr = 0x11646464; - faddr.s_addr = 0x11646465; - - lport = 1500; - in_pcb_grab_port(&udbinfo, 0, laddr, &lport, faddr, 1600, 0, fake_owner); - kprintf("udp_init in_pcb_grab_port - stat = %d\n", stat); - - stat = in_pcb_rem_share_client(&udbinfo, fake_owner); - kprintf("udp_init in_pcb_rem_share_client - stat = %d\n", stat); - - stat = in_pcb_new_share_client(&udbinfo, &fake_owner); - kprintf("udp_init in_pcb_new_share_client(2) - stat = %d\n", stat); - - laddr.s_addr = 0x11646464; - faddr.s_addr = 0x11646465; - - lport = 1500; - stat = in_pcb_grab_port(&udbinfo, 0, laddr, &lport, faddr, 1600, 0, fake_owner); - kprintf("udp_init in_pcb_grab_port(2) - stat = %d\n", stat); -#endif } void @@ -257,10 +256,24 @@ udp_input(m, iphlen) register struct udphdr *uh; register struct inpcb *inp; struct mbuf *opts = 0; - int len; + int len, isbroadcast; struct ip save_ip; struct sockaddr *append_sa; struct inpcbinfo *pcbinfo = &udbinfo; + struct sockaddr_in udp_in = { + sizeof (udp_in), AF_INET, 0, { 0 }, { 0, 0, 0, 0, 0, 0, 0, 0 } + }; + struct ip_moptions *imo = NULL; + int foundmembership = 0, ret = 0; +#if INET6 + struct udp_in6 udp_in6 = { + { sizeof (udp_in6.uin6_sin), AF_INET6, 0, 0, + IN6ADDR_ANY_INIT, 0 }, + 0 + }; + struct udp_ip6 udp_ip6; +#endif /* INET6 */ + struct ifnet *ifp = (m->m_pkthdr.rcvif != NULL) ? m->m_pkthdr.rcvif: NULL; udpstat.udps_ipackets++; @@ -268,6 +281,9 @@ udp_input(m, iphlen) if (m->m_pkthdr.csum_flags & CSUM_TCP_SUM16) m->m_pkthdr.csum_flags = 0; /* invalidate hwcksum for UDP */ + /* Expect 32-bit aligned data pointer on strict-align platforms */ + MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m); + /* * Strip IP options, if any; should skip this, * make available to user, and use on returned packets, @@ -291,11 +307,16 @@ udp_input(m, iphlen) } ip = mtod(m, struct ip *); } - uh = (struct udphdr *)((caddr_t)ip + iphlen); + uh = (struct udphdr *)(void *)((caddr_t)ip + iphlen); /* destination port of 0 is illegal, based on RFC768. */ - if (uh->uh_dport == 0) + if (uh->uh_dport == 0) { + + if (ifp->if_udp_stat != NULL) + atomic_add_64(&ifp->if_udp_stat->port0, 1); + goto bad; + } KERNEL_DEBUG(DBG_LAYER_IN_BEG, uh->uh_dport, uh->uh_sport, ip->ip_src.s_addr, ip->ip_dst.s_addr, uh->uh_ulen); @@ -308,6 +329,10 @@ udp_input(m, iphlen) if (ip->ip_len != len) { if (len > ip->ip_len || len < sizeof(struct udphdr)) { udpstat.udps_badlen++; + + if (ifp->if_udp_stat != NULL) + atomic_add_64(&ifp->if_udp_stat->badlength, 1); + goto bad; } m_adj(m, len - ip->ip_len); @@ -332,20 +357,23 @@ udp_input(m, iphlen) } else { char b[9]; doudpcksum: - *(uint32_t*)&b[0] = *(uint32_t*)&((struct ipovly *)ip)->ih_x1[0]; - *(uint32_t*)&b[4] = *(uint32_t*)&((struct ipovly *)ip)->ih_x1[4]; - *(uint8_t*)&b[8] = *(uint8_t*)&((struct ipovly *)ip)->ih_x1[8]; - - bzero(((struct ipovly *)ip)->ih_x1, 9); + bcopy(((struct ipovly *)ip)->ih_x1, b, + sizeof (((struct ipovly *)ip)->ih_x1)); + bzero(((struct ipovly *)ip)->ih_x1, + sizeof (((struct ipovly *)ip)->ih_x1)); ((struct ipovly *)ip)->ih_len = uh->uh_ulen; uh->uh_sum = in_cksum(m, len + sizeof (struct ip)); - - *(uint32_t*)&((struct ipovly *)ip)->ih_x1[0] = *(uint32_t*)&b[0]; - *(uint32_t*)&((struct ipovly *)ip)->ih_x1[4] = *(uint32_t*)&b[4]; - *(uint8_t*)&((struct ipovly *)ip)->ih_x1[8] = *(uint8_t*)&b[8]; + bcopy(b, ((struct ipovly *)ip)->ih_x1, + sizeof (((struct ipovly *)ip)->ih_x1)); + + udp_in_cksum_stats(len); } if (uh->uh_sum) { udpstat.udps_badsum++; + + if (ifp->if_udp_stat != NULL) + atomic_add_64(&ifp->if_udp_stat->badchksum, 1); + m_freem(m); KERNEL_DEBUG(DBG_FNC_UDP_INPUT | DBG_FUNC_END, 0,0,0,0,0); return; @@ -356,9 +384,12 @@ doudpcksum: udpstat.udps_nosum++; #endif - if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || - in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { - struct inpcb *last; + isbroadcast = in_broadcast(ip->ip_dst, ifp); + + if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || isbroadcast) { + + int reuse_sock = 0, mcast_delivered = 0; + lck_rw_lock_shared(pcbinfo->mtx); /* * Deliver a multicast or broadcast datagram to *all* sockets @@ -386,24 +417,29 @@ doudpcksum: * Locate pcb(s) for datagram. * (Algorithm copied from raw_intr().) */ - last = NULL; #if INET6 udp_in6.uin6_init_done = udp_ip6.uip6_init_done = 0; #endif LIST_FOREACH(inp, &udb, inp_list) { -#ifdef __APPLE__ - /* Ignore nat/SharedIP dummy pcbs */ - if (inp->inp_socket == &udbinfo.nat_dummy_socket) - continue; -#endif if (inp->inp_socket == NULL) continue; if (inp != sotoinpcb(inp->inp_socket)) - panic("udp_input: bad so back ptr inp=%x\n", inp); + panic("udp_input: bad so back ptr inp=%p\n", inp); #if INET6 if ((inp->inp_vflag & INP_IPV4) == 0) continue; #endif + if (ip_restrictrecvif && ifp != NULL && + (ifp->if_eflags & IFEF_RESTRICTED_RECV) && + !(inp->inp_flags & INP_RECV_ANYIF)) + continue; + + if ((inp->inp_moptions == NULL) && + (ntohl(ip->ip_dst.s_addr) != INADDR_ALLHOSTS_GROUP) && + (isbroadcast == 0) ) + continue; + + if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) { continue; } @@ -435,67 +471,110 @@ doudpcksum: } } - if (last != NULL) { - struct mbuf *n; + if (isbroadcast == 0 && (ntohl(ip->ip_dst.s_addr) != INADDR_ALLHOSTS_GROUP)) { + if((imo = inp->inp_moptions) == NULL) { + udp_unlock(inp->inp_socket, 1, 0); + continue; + } else { + struct sockaddr_in group; + int blocked; + + IMO_LOCK(imo); + + bzero(&group, sizeof(struct sockaddr_in)); + group.sin_len = sizeof(struct sockaddr_in); + group.sin_family = AF_INET; + group.sin_addr = ip->ip_dst; + + blocked = imo_multi_filter(imo, ifp, + (struct sockaddr *)&group, + (struct sockaddr *)&udp_in); + if (blocked == MCAST_PASS) + foundmembership = 1; + + IMO_UNLOCK(imo); + if (!foundmembership) { + udp_unlock(inp->inp_socket, 1, 0); + continue; + } + foundmembership = 0; + } + } + reuse_sock = inp->inp_socket->so_options& (SO_REUSEPORT|SO_REUSEADDR); + { #if IPSEC int skipit = 0; /* check AH/ESP integrity. */ if (ipsec_bypass == 0) { - lck_mtx_lock(sadb_mutex); - if (ipsec4_in_reject_so(m, last->inp_socket)) { - ipsecstat.in_polvio++; + if (ipsec4_in_reject_so(m, inp->inp_socket)) { + IPSEC_STAT_INCREMENT(ipsecstat.in_polvio); /* do not inject data to pcb */ skipit = 1; } - lck_mtx_unlock(sadb_mutex); } - if (skipit == 0) + if (skipit == 0) #endif /*IPSEC*/ - if ((n = m_copy(m, 0, M_COPYALL)) != NULL) { - udp_append(last, ip, n, - iphlen + - sizeof(struct udphdr)); + { + struct mbuf *n = NULL; + + if (reuse_sock) + n = m_copy(m, 0, M_COPYALL); +#if INET6 + udp_append(inp, ip, m, + iphlen + sizeof(struct udphdr), + &udp_in, &udp_in6, &udp_ip6); +#else + udp_append(inp, ip, m, + iphlen + sizeof(struct udphdr), + &udp_in); +#endif /* INET6 */ + mcast_delivered++; + + m = n; } - udp_unlock(last->inp_socket, 1, 0); + udp_unlock(inp->inp_socket, 1, 0); } - last = inp; /* * Don't look for additional matches if this one does * not have either the SO_REUSEPORT or SO_REUSEADDR * socket options set. This heuristic avoids searching * through all pcbs in the common case of a non-shared - * port. It * assumes that an application will never + * port. It assumes that an application will never * clear these options after setting them. */ - if ((last->inp_socket->so_options&(SO_REUSEPORT|SO_REUSEADDR)) == 0) + if (reuse_sock == 0 || m == NULL) break; + + /* + * Expect 32-bit aligned data pointer on strict-align + * platforms. + */ + MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m); + /* + * Recompute IP and UDP header pointers for new mbuf + */ + ip = mtod(m, struct ip *); + uh = (struct udphdr *)(void *)((caddr_t)ip + iphlen); } lck_rw_done(pcbinfo->mtx); - if (last == NULL) { + if (mcast_delivered == 0) { /* * No matching pcb found; discard datagram. * (No need to send an ICMP Port Unreachable * for a broadcast or multicast datgram.) */ udpstat.udps_noportbcast++; + + if (ifp->if_udp_stat != NULL) + atomic_add_64(&ifp->if_udp_stat->port_unreach, 1); + goto bad; } -#if IPSEC - /* check AH/ESP integrity. */ - if (ipsec_bypass == 0 && m) { - lck_mtx_lock(sadb_mutex); - if (ipsec4_in_reject_so(m, last->inp_socket)) { - ipsecstat.in_polvio++; - lck_mtx_unlock(sadb_mutex); - udp_unlock(last->inp_socket, 1, 0); - goto bad; - } - lck_mtx_unlock(sadb_mutex); - } -#endif /*IPSEC*/ - udp_append(last, ip, m, iphlen + sizeof(struct udphdr)); - udp_unlock(last->inp_socket, 1, 0); + + if (m != NULL) /* free the extra copy of mbuf or skipped by IPSec */ + m_freem(m); + KERNEL_DEBUG(DBG_FNC_UDP_INPUT | DBG_FUNC_END, 0,0,0,0,0); return; } @@ -516,8 +595,14 @@ doudpcksum: KERNEL_DEBUG(DBG_FNC_UDP_INPUT | DBG_FUNC_END, 0,0,0,0,0); return; } + /* + * Expect 32-bit aligned data pointer on strict-align + * platforms. + */ + MBUF_STRICT_DATA_ALIGNMENT_CHECK_32(m); + ip = mtod(m, struct ip *); - uh = (struct udphdr *)((caddr_t)ip + iphlen); + uh = (struct udphdr *)(void *)((caddr_t)ip + iphlen); } /* Check for NAT keepalive packet */ if (payload_len == 1 && *(u_int8_t*)((caddr_t)uh + sizeof(struct udphdr)) == 0xFF) { @@ -525,23 +610,11 @@ doudpcksum: KERNEL_DEBUG(DBG_FNC_UDP_INPUT | DBG_FUNC_END, 0,0,0,0,0); return; } - else if (payload_len == 4 && *(u_int32_t*)((caddr_t)uh + sizeof(struct udphdr)) != 0) { + else if (payload_len == 4 && *(u_int32_t*)(void *)((caddr_t)uh + sizeof(struct udphdr)) != 0) { /* UDP encapsulated IPSec packet to pass through NAT */ - size_t stripsiz; - - stripsiz = sizeof(struct udphdr); - - ip = mtod(m, struct ip *); - ovbcopy((caddr_t)ip, (caddr_t)(((u_char *)ip) + stripsiz), iphlen); - m->m_data += stripsiz; - m->m_len -= stripsiz; - m->m_pkthdr.len -= stripsiz; - ip = mtod(m, struct ip *); - ip->ip_len = ip->ip_len - stripsiz; - ip->ip_p = IPPROTO_ESP; - KERNEL_DEBUG(DBG_FNC_UDP_INPUT | DBG_FUNC_END, 0,0,0,0,0); - esp4_input(m, iphlen); + /* preserve the udp header */ + esp4_input(m, iphlen + sizeof(struct udphdr)); return; } } @@ -551,8 +624,12 @@ doudpcksum: * Locate pcb for datagram. */ inp = in_pcblookup_hash(&udbinfo, ip->ip_src, uh->uh_sport, - ip->ip_dst, uh->uh_dport, 1, m->m_pkthdr.rcvif); + ip->ip_dst, uh->uh_dport, 1, ifp); if (inp == NULL) { + + if (ifp->if_udp_stat != NULL) + atomic_add_64(&ifp->if_udp_stat->port_unreach, 1); + if (log_in_vain) { char buf[MAX_IPv4_STR_LEN]; char buf2[MAX_IPv4_STR_LEN]; @@ -584,7 +661,7 @@ doudpcksum: goto bad; #endif if (blackhole) - if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type != IFT_LOOP) + if (ifp && ifp->if_type != IFT_LOOP) goto bad; *ip = save_ip; ip->ip_len += iphlen; @@ -596,18 +673,23 @@ doudpcksum: if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { udp_unlock(inp->inp_socket, 1, 0); + + if (ifp->if_udp_stat != NULL) + atomic_add_64(&ifp->if_udp_stat->cleanup, 1); + goto bad; } #if IPSEC if (ipsec_bypass == 0 && inp != NULL) { - lck_mtx_lock(sadb_mutex); if (ipsec4_in_reject_so(m, inp->inp_socket)) { - ipsecstat.in_polvio++; - lck_mtx_unlock(sadb_mutex); + IPSEC_STAT_INCREMENT(ipsecstat.in_polvio); udp_unlock(inp->inp_socket, 1, 0); + + if (ifp->if_udp_stat != NULL) + atomic_add_64(&ifp->if_udp_stat->badipsec, 1); + goto bad; - } - lck_mtx_unlock(sadb_mutex); + } } #endif /*IPSEC*/ @@ -617,8 +699,9 @@ doudpcksum: */ udp_in.sin_port = uh->uh_sport; udp_in.sin_addr = ip->ip_src; - if (inp->inp_flags & INP_CONTROLOPTS - || inp->inp_socket->so_options & SO_TIMESTAMP) { + if ((inp->inp_flags & INP_CONTROLOPTS) != 0 + || (inp->inp_socket->so_options & SO_TIMESTAMP) != 0 + || (inp->inp_socket->so_options & SO_TIMESTAMP_MONOTONIC) != 0) { #if INET6 if (inp->inp_vflag & INP_IPV6) { int savedflags; @@ -626,11 +709,17 @@ doudpcksum: ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip); savedflags = inp->inp_flags; inp->inp_flags &= ~INP_UNMAPPABLEOPTS; - ip6_savecontrol(inp, &opts, &udp_ip6.uip6_ip6, m); + ret = ip6_savecontrol(inp, m, &opts); inp->inp_flags = savedflags; } else #endif - ip_savecontrol(inp, &opts, ip, m); + { + ret = ip_savecontrol(inp, &opts, ip, m); + } + if (ret != 0) { + udp_unlock(inp->inp_socket, 1, 0); + goto bad; + } } m_adj(m, iphlen + sizeof(struct udphdr)); @@ -640,14 +729,18 @@ doudpcksum: #if INET6 if (inp->inp_vflag & INP_IPV6) { in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin); - append_sa = (struct sockaddr *)&udp_in6; + append_sa = (struct sockaddr *)&udp_in6.uin6_sin; } else #endif append_sa = (struct sockaddr *)&udp_in; + if (nstat_collect) { + locked_add_64(&inp->inp_stat->rxpackets, 1); + locked_add_64(&inp->inp_stat->rxbytes, m->m_pkthdr.len); + } + so_recv_data_stat(inp->inp_socket, m, 0); if (sbappendaddr(&inp->inp_socket->so_rcv, append_sa, m, opts, NULL) == 0) { udpstat.udps_fullsock++; - } - else { + } else { sorwakeup(inp->inp_socket); } udp_unlock(inp->inp_socket, 1, 0); @@ -673,60 +766,94 @@ ip_2_ip6_hdr(ip6, ip) ip6->ip6_plen = ip->ip_len; ip6->ip6_nxt = ip->ip_p; ip6->ip6_hlim = ip->ip_ttl; - ip6->ip6_src.s6_addr32[2] = ip6->ip6_dst.s6_addr32[2] = - IPV6_ADDR_INT32_SMP; - ip6->ip6_src.s6_addr32[3] = ip->ip_src.s_addr; - ip6->ip6_dst.s6_addr32[3] = ip->ip_dst.s_addr; + if (ip->ip_src.s_addr) { + ip6->ip6_src.s6_addr32[2] = IPV6_ADDR_INT32_SMP; + ip6->ip6_src.s6_addr32[3] = ip->ip_src.s_addr; + } + if (ip->ip_dst.s_addr) { + ip6->ip6_dst.s6_addr32[2] = IPV6_ADDR_INT32_SMP; + ip6->ip6_dst.s6_addr32[3] = ip->ip_dst.s_addr; + } } #endif /* * subroutine of udp_input(), mainly for source code readability. - * caller must properly init udp_ip6 and udp_in6 beforehand. */ static void -udp_append(last, ip, n, off) - struct inpcb *last; - struct ip *ip; - struct mbuf *n; - int off; +#if INET6 +udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off, + struct sockaddr_in *pudp_in, struct udp_in6 *pudp_in6, + struct udp_ip6 *pudp_ip6) +#else +udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off, + struct sockaddr_in *pudp_in) +#endif { struct sockaddr *append_sa; struct mbuf *opts = 0; + int ret = 0; - if (last->inp_flags & INP_CONTROLOPTS || - last->inp_socket->so_options & SO_TIMESTAMP) { +#if CONFIG_MACF_NET + if (mac_inpcb_check_deliver(last, n, AF_INET, SOCK_DGRAM) != 0) { + m_freem(n); + return; + } +#endif + if ((last->inp_flags & INP_CONTROLOPTS) != 0 || + (last->inp_socket->so_options & SO_TIMESTAMP) != 0 || + (last->inp_socket->so_options & SO_TIMESTAMP_MONOTONIC) != 0) { #if INET6 if (last->inp_vflag & INP_IPV6) { int savedflags; - if (udp_ip6.uip6_init_done == 0) { - ip_2_ip6_hdr(&udp_ip6.uip6_ip6, ip); - udp_ip6.uip6_init_done = 1; + if (pudp_ip6->uip6_init_done == 0) { + ip_2_ip6_hdr(&pudp_ip6->uip6_ip6, ip); + pudp_ip6->uip6_init_done = 1; } savedflags = last->inp_flags; last->inp_flags &= ~INP_UNMAPPABLEOPTS; - ip6_savecontrol(last, &opts, &udp_ip6.uip6_ip6, n); + ret = ip6_savecontrol(last, n, &opts); + if (ret != 0) { + last->inp_flags = savedflags; + goto error; + } last->inp_flags = savedflags; } else #endif - ip_savecontrol(last, &opts, ip, n); + { + ret = ip_savecontrol(last, &opts, ip, n); + if (ret != 0) { + goto error; + } + } } #if INET6 if (last->inp_vflag & INP_IPV6) { - if (udp_in6.uin6_init_done == 0) { - in6_sin_2_v4mapsin6(&udp_in, &udp_in6.uin6_sin); - udp_in6.uin6_init_done = 1; + if (pudp_in6->uin6_init_done == 0) { + in6_sin_2_v4mapsin6(pudp_in, &pudp_in6->uin6_sin); + pudp_in6->uin6_init_done = 1; } - append_sa = (struct sockaddr *)&udp_in6.uin6_sin; + append_sa = (struct sockaddr *)&pudp_in6->uin6_sin; } else #endif - append_sa = (struct sockaddr *)&udp_in; + append_sa = (struct sockaddr *)pudp_in; + if (nstat_collect) { + locked_add_64(&last->inp_stat->rxpackets, 1); + locked_add_64(&last->inp_stat->rxbytes, n->m_pkthdr.len); + } + so_recv_data_stat(last->inp_socket, n, 0); m_adj(n, off); if (sbappendaddr(&last->inp_socket->so_rcv, append_sa, n, opts, NULL) == 0) { udpstat.udps_fullsock++; - } else + } else { sorwakeup(last->inp_socket); + } + return; +error: + m_freem(n); + m_freem(opts); + return; } /* @@ -750,14 +877,13 @@ udp_ctlinput(cmd, sa, vip) void *vip; { struct ip *ip = vip; - struct udphdr *uh; void (*notify)(struct inpcb *, int) = udp_notify; struct in_addr faddr; struct inpcb *inp; - faddr = ((struct sockaddr_in *)sa)->sin_addr; + faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) - return; + return; if (PRC_IS_REDIRECT(cmd)) { ip = 0; @@ -767,12 +893,15 @@ udp_ctlinput(cmd, sa, vip) else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip) { - uh = (struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); - inp = in_pcblookup_hash(&udbinfo, faddr, uh->uh_dport, - ip->ip_src, uh->uh_sport, 0, NULL); + struct udphdr uh; + + bcopy(((caddr_t)ip + (ip->ip_hl << 2)), &uh, sizeof (uh)); + inp = in_pcblookup_hash(&udbinfo, faddr, uh.uh_dport, + ip->ip_src, uh.uh_sport, 0, NULL); if (inp != NULL && inp->inp_socket != NULL) { udp_lock(inp->inp_socket, 1, 0); - if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) { + if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == + WNT_STOPUSING) { udp_unlock(inp->inp_socket, 1, 0); return; } @@ -783,9 +912,75 @@ udp_ctlinput(cmd, sa, vip) in_pcbnotifyall(&udbinfo, faddr, inetctlerrmap[cmd], notify); } +int +udp_ctloutput(struct socket *so, struct sockopt *sopt) +{ + int error, optval; + struct inpcb *inp; + + /* Allow at this level */ + if (sopt->sopt_level != IPPROTO_UDP && + !(sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_FLUSH)) + return (ip_ctloutput(so, sopt)); + + error = 0; + inp = sotoinpcb(so); + + switch (sopt->sopt_dir) { + case SOPT_SET: + switch (sopt->sopt_name) { + case UDP_NOCKSUM: + /* This option is settable only for UDP over IPv4 */ + if (!(inp->inp_vflag & INP_IPV4)) { + error = EINVAL; + break; + } + + if ((error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval))) != 0) + break; + + if (optval != 0) + inp->inp_flags |= INP_UDP_NOCKSUM; + else + inp->inp_flags &= ~INP_UDP_NOCKSUM; + break; + + case SO_FLUSH: + if ((error = sooptcopyin(sopt, &optval, sizeof (optval), + sizeof (optval))) != 0) + break; + + error = inp_flush(inp, optval); + break; + + default: + error = ENOPROTOOPT; + break; + } + break; + + case SOPT_GET: + switch (sopt->sopt_name) { + case UDP_NOCKSUM: + optval = inp->inp_flags & INP_UDP_NOCKSUM; + break; + + default: + error = ENOPROTOOPT; + break; + } + if (error == 0) + error = sooptcopyout(sopt, &optval, sizeof (optval)); + break; + } + return (error); +} + static int udp_pcblist SYSCTL_HANDLER_ARGS { +#pragma unused(oidp, arg1, arg2) int error, i, n; struct inpcb *inp, **inp_list; inp_gen_t gencnt; @@ -881,10 +1076,140 @@ udp_pcblist SYSCTL_HANDLER_ARGS return error; } -SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0, +SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, udp_pcblist, "S,xinpcb", "List of active UDP sockets"); +#if !CONFIG_EMBEDDED + +static int +udp_pcblist64 SYSCTL_HANDLER_ARGS +{ +#pragma unused(oidp, arg1, arg2) + int error, i, n; + struct inpcb *inp, **inp_list; + inp_gen_t gencnt; + struct xinpgen xig; + + /* + * The process of preparing the TCB list is too time-consuming and + * resource-intensive to repeat twice on every request. + */ + lck_rw_lock_shared(udbinfo.mtx); + if (req->oldptr == USER_ADDR_NULL) { + n = udbinfo.ipi_count; + req->oldidx = 2 * (sizeof xig) + + (n + n/8) * sizeof(struct xinpcb64); + lck_rw_done(udbinfo.mtx); + return 0; + } + + if (req->newptr != USER_ADDR_NULL) { + lck_rw_done(udbinfo.mtx); + return EPERM; + } + + /* + * OK, now we're committed to doing something. + */ + gencnt = udbinfo.ipi_gencnt; + n = udbinfo.ipi_count; + + bzero(&xig, sizeof(xig)); + xig.xig_len = sizeof xig; + xig.xig_count = n; + xig.xig_gen = gencnt; + xig.xig_sogen = so_gencnt; + error = SYSCTL_OUT(req, &xig, sizeof xig); + if (error) { + lck_rw_done(udbinfo.mtx); + return error; + } + /* + * We are done if there is no pcb + */ + if (n == 0) { + lck_rw_done(udbinfo.mtx); + return 0; + } + inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK); + if (inp_list == 0) { + lck_rw_done(udbinfo.mtx); + return ENOMEM; + } + + for (inp = LIST_FIRST(udbinfo.listhead), i = 0; inp && i < n; + inp = LIST_NEXT(inp, inp_list)) { + if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) + inp_list[i++] = inp; + } + n = i; + + error = 0; + for (i = 0; i < n; i++) { + inp = inp_list[i]; + if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) { + struct xinpcb64 xi; + + bzero(&xi, sizeof(xi)); + xi.xi_len = sizeof xi; + inpcb_to_xinpcb64(inp, &xi); + if (inp->inp_socket) + sotoxsocket64(inp->inp_socket, &xi.xi_socket); + error = SYSCTL_OUT(req, &xi, sizeof xi); + } + } + if (!error) { + /* + * Give the user an updated idea of our state. + * If the generation differs from what we told + * her before, she knows that something happened + * while we were processing this request, and it + * might be necessary to retry. + */ + bzero(&xig, sizeof(xig)); + xig.xig_len = sizeof xig; + xig.xig_gen = udbinfo.ipi_gencnt; + xig.xig_sogen = so_gencnt; + xig.xig_count = udbinfo.ipi_count; + error = SYSCTL_OUT(req, &xig, sizeof xig); + } + FREE(inp_list, M_TEMP); + lck_rw_done(udbinfo.mtx); + return error; +} + +SYSCTL_PROC(_net_inet_udp, OID_AUTO, pcblist64, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + udp_pcblist64, "S,xinpcb64", "List of active UDP sockets"); + +#endif /* !CONFIG_EMBEDDED */ + +static int +udp_pcblist_n SYSCTL_HANDLER_ARGS +{ +#pragma unused(oidp, arg1, arg2) + int error = 0; + + error = get_pcblist_n(IPPROTO_UDP, req, &udbinfo); + + return error; +} + +SYSCTL_PROC(_net_inet_udp, OID_AUTO, pcblist_n, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + udp_pcblist_n, "S,xinpcb_n", "List of active UDP sockets"); + + +__private_extern__ void +udp_get_ports_used(unsigned int ifindex, uint8_t *bitfield) +{ + inpcb_get_ports_used(ifindex, bitfield, &udbinfo); +} + +__private_extern__ uint32_t +udp_count_opportunistic(unsigned int ifindex, u_int32_t flags) +{ + return inpcb_count_opportunistic(ifindex, &udbinfo, flags); +} static __inline__ u_int16_t get_socket_id(struct socket * s) @@ -894,13 +1219,78 @@ get_socket_id(struct socket * s) if (s == NULL) { return (0); } - val = (u_int16_t)(((u_int32_t)s) / sizeof(struct socket)); + val = (u_int16_t)(((uintptr_t)s) / sizeof(struct socket)); if (val == 0) { val = 0xffff; } return (val); } +static int +udp_check_pktinfo(struct mbuf *control, struct ifnet **outif, struct in_addr *laddr) +{ + struct cmsghdr *cm = 0; + struct in_pktinfo *pktinfo; + struct ifnet *ifp; + + /* + * XXX: Currently, we assume all the optional information is stored + * in a single mbuf. + */ + if (control->m_next) + return (EINVAL); + + if (control->m_len < CMSG_LEN(0)) + return (EINVAL); + + for (cm = M_FIRST_CMSGHDR(control); cm; cm = M_NXT_CMSGHDR(control, cm)) { + if (cm->cmsg_len < sizeof(struct cmsghdr) || cm->cmsg_len > control->m_len) + return (EINVAL); + + if (cm->cmsg_level != IPPROTO_IP || cm->cmsg_type != IP_PKTINFO) + continue; + + if (cm->cmsg_len != CMSG_LEN(sizeof(struct in_pktinfo))) + return (EINVAL); + + pktinfo = (struct in_pktinfo *)(void *)CMSG_DATA(cm); + + /* Check for a valid ifindex in pktinfo */ + ifnet_head_lock_shared(); + + if (pktinfo->ipi_ifindex > if_index) { + ifnet_head_done(); + return (ENXIO); + } + + /* If ipi_ifindex is specified it takes precedence over ipi_spec_dst */ + + if (pktinfo->ipi_ifindex) { + ifp = ifindex2ifnet[pktinfo->ipi_ifindex]; + if (ifp == NULL) { + ifnet_head_done(); + return (ENXIO); + } + + ifnet_head_done(); + + if (outif != NULL) + *outif = ifp; + laddr->s_addr = INADDR_ANY; + break; + } + + ifnet_head_done(); + + /* Use the provided ipi_spec_dst address for temp source address */ + if (outif != NULL) + *outif = NULL; + *laddr = pktinfo->ipi_spec_dst; + break; + } + return (0); +} + static int udp_output(inp, m, addr, control, p) register struct inpcb *inp; @@ -911,17 +1301,43 @@ udp_output(inp, m, addr, control, p) { register struct udpiphdr *ui; register int len = m->m_pkthdr.len; - struct sockaddr_in *sin, src; - struct in_addr origladdr, laddr, faddr; + struct sockaddr_in *sin; + struct in_addr origladdr, laddr, faddr, pi_laddr; u_short lport, fport; - struct sockaddr_in *ifaddr; - int error = 0, udp_dodisconnect = 0; - + struct sockaddr_in ifaddr; + int error = 0, udp_dodisconnect = 0, pktinfo = 0; + struct socket *so = inp->inp_socket; + int soopts = 0; + struct mbuf *inpopts; + struct ip_moptions *mopts; + struct route ro; + struct ip_out_args ipoa = { IFSCOPE_NONE, { 0 }, IPOAF_SELECT_SRCIF }; + struct ifnet *outif = NULL; + struct flowadv *adv = &ipoa.ipoa_flowadv; + mbuf_svc_class_t msc = MBUF_SC_UNSPEC; + struct ifnet *origoutifp; + int flowadv = 0; + + /* Enable flow advisory only when connected */ + flowadv = (so->so_state & SS_ISCONNECTED) ? 1 : 0; + + pi_laddr.s_addr = INADDR_ANY; KERNEL_DEBUG(DBG_FNC_UDP_OUTPUT | DBG_FUNC_START, 0,0,0,0,0); - if (control) - m_freem(control); /* XXX */ + lck_mtx_assert(&inp->inpcb_mtx, LCK_MTX_ASSERT_OWNED); + if (control != NULL) { + msc = mbuf_service_class_from_control(control); + + error = udp_check_pktinfo(control, &outif, &pi_laddr); + + m_freem(control); + if (error) + goto release; + pktinfo++; + if (outif != NULL) + ipoa.ipoa_boundif = outif->if_index; + } KERNEL_DEBUG(DBG_LAYER_OUT_BEG, inp->inp_fport, inp->inp_lport, inp->inp_laddr.s_addr, inp->inp_faddr.s_addr, @@ -932,35 +1348,87 @@ udp_output(inp, m, addr, control, p) goto release; } + if (flowadv && INP_WAIT_FOR_IF_FEEDBACK(inp)) { + /* + * The socket is flow-controlled, drop the packets + * until the inp is not flow controlled + */ + error = ENOBUFS; + goto release; + } + /* + * If socket was bound to an ifindex, tell ip_output about it. + * If the ancillary IP_PKTINFO option contains an interface index, + * it takes precedence over the one specified by IP_BOUND_IF. + */ + if (ipoa.ipoa_boundif == IFSCOPE_NONE && + (inp->inp_flags & INP_BOUND_IF)) { + outif = inp->inp_boundifp; + ipoa.ipoa_boundif = outif->if_index; + } + if (inp->inp_flags & INP_NO_IFT_CELLULAR) + ipoa.ipoa_flags |= IPOAF_NO_CELLULAR; + soopts |= IP_OUTARGS; + /* If there was a routing change, discard cached route and check * that we have a valid source address. * Reacquire a new source address if INADDR_ANY was specified */ - -#if 1 - lck_mtx_assert(inp->inpcb_mtx, LCK_MTX_ASSERT_OWNED); -#endif - - if (inp->inp_route.ro_rt && inp->inp_route.ro_rt->generation_id != route_generation) { - if (ifa_foraddr(inp->inp_laddr.s_addr) == 0) { /* src address is gone */ - if (inp->inp_flags & INP_INADDR_ANY) - inp->inp_faddr.s_addr = INADDR_ANY; /* new src will be set later */ - else { + if (inp->inp_route.ro_rt != NULL && + inp->inp_route.ro_rt->generation_id != route_generation) { + struct in_ifaddr *ia; + + /* src address is gone? */ + if ((ia = ifa_foraddr(inp->inp_laddr.s_addr)) == NULL) { + if (((inp->inp_flags & INP_INADDR_ANY) == 0) || + (so->so_state & SS_ISCONNECTED)) { + /* Rdar://5448998 + * If the source address is gone, return an error if: + * - the source was specified + * - the socket was already connected + */ + soevent(so, + (SO_FILT_HINT_LOCKED | + SO_FILT_HINT_NOSRCADDR)); error = EADDRNOTAVAIL; goto release; + } else { + /* new src will be set later */ + inp->inp_laddr.s_addr = INADDR_ANY; + inp->inp_last_outifp = NULL; } } - rtfree(inp->inp_route.ro_rt); - inp->inp_route.ro_rt = (struct rtentry *)0; + if (ia != NULL) + IFA_REMREF(&ia->ia_ifa); + if (inp->inp_route.ro_rt != NULL) + rtfree(inp->inp_route.ro_rt); + inp->inp_route.ro_rt = NULL; + } + + origoutifp = inp->inp_last_outifp; + + /* IP_PKTINFO option check. + * If a temporary scope or src address is provided, use it for this packet only + * and make sure we forget it after sending this datagram. + */ + + if (pi_laddr.s_addr != INADDR_ANY || + (ipoa.ipoa_boundif != IFSCOPE_NONE && pktinfo)) { + laddr = pi_laddr; /* temp src address for this datagram only */ + origladdr.s_addr = INADDR_ANY; + udp_dodisconnect = 1; /* we don't want to keep the laddr or route */ + inp->inp_flags |= INP_INADDR_ANY; /* remember we don't care about src addr.*/ + } else { + origladdr = laddr = inp->inp_laddr; } - origladdr= laddr = inp->inp_laddr; + origoutifp = inp->inp_last_outifp; faddr = inp->inp_faddr; lport = inp->inp_lport; fport = inp->inp_fport; if (addr) { - sin = (struct sockaddr_in *)addr; + sin = (struct sockaddr_in *)(void *)addr; if (faddr.s_addr != INADDR_ANY) { error = EISCONN; goto release; @@ -970,8 +1438,11 @@ udp_output(inp, m, addr, control, p) * In case we don't have a local port set, go through the full connect. * We don't have a local port yet (ie, we can't be looked up), * so it's not an issue if the input runs at the same time we do this. - */ - error = in_pcbconnect(inp, addr, p); + */ + + if (pi_laddr.s_addr != INADDR_ANY) /* if we have a source address specified, use that */ + inp->inp_laddr = pi_laddr; + error = in_pcbconnect(inp, addr, p, &outif); /* if a scope is specified, use it */ if (error) { goto release; } @@ -980,20 +1451,26 @@ udp_output(inp, m, addr, control, p) faddr = inp->inp_faddr; fport = inp->inp_fport; udp_dodisconnect = 1; + ipoa.ipoa_boundif = (outif != NULL) ? + outif->if_index : IFSCOPE_NONE; } - else { + else { /* Fast path case * we have a full address and a local port. * use those info to build the packet without changing the pcb * and interfering with the input path. See 3851370 + * Note: if we may have a scope from IP_PKTINFO but the + * priority is always given to the scope provided by INP_BOUND_IF. */ if (laddr.s_addr == INADDR_ANY) { - if ((error = in_pcbladdr(inp, addr, &ifaddr)) != 0) - goto release; - laddr = ifaddr->sin_addr; - inp->inp_flags |= INP_INADDR_ANY; /* from pcbconnect: remember we don't care about src addr.*/ + if ((error = in_pcbladdr(inp, addr, &ifaddr, &outif)) != 0) + goto release; + laddr = ifaddr.sin_addr; + inp->inp_flags |= INP_INADDR_ANY; /* from pcbconnect: remember we don't care about src addr.*/ + ipoa.ipoa_boundif = (outif != NULL) ? + outif->if_index : IFSCOPE_NONE; } - + faddr = sin->sin_addr; fport = sin->sin_port; } @@ -1004,6 +1481,11 @@ udp_output(inp, m, addr, control, p) } } +#if CONFIG_MACF_NET + mac_mbuf_label_associate_inpcb(inp, m); +#endif + if (inp->inp_flowhash == 0) + inp->inp_flowhash = inp_calc_flowhash(inp); /* * Calculate data length and get a mbuf @@ -1031,7 +1513,7 @@ udp_output(inp, m, addr, control, p) /* * Set up checksum and output datagram. */ - if (udpcksum) { + if (udpcksum && !(inp->inp_flags & INP_UDP_NOCKSUM)) { ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr, htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP)); m->m_pkthdr.csum_flags = CSUM_UDP; @@ -1053,44 +1535,149 @@ udp_output(inp, m, addr, control, p) goto abort; } #endif /*IPSEC*/ + + inpopts = inp->inp_options; + soopts |= (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST)); + mopts = inp->inp_moptions; + if (mopts != NULL) { + IMO_LOCK(mopts); + IMO_ADDREF_LOCKED(mopts); + if (IN_MULTICAST(ntohl(ui->ui_dst.s_addr)) && + mopts->imo_multicast_ifp != NULL) { + inp->inp_last_outifp = mopts->imo_multicast_ifp; + } + IMO_UNLOCK(mopts); + } + + /* Copy the cached route and take an extra reference */ + inp_route_copyout(inp, &ro); + + set_packet_service_class(m, so, msc, 0); m->m_pkthdr.socket_id = get_socket_id(inp->inp_socket); - error = ip_output_list(m, 0, inp->inp_options, &inp->inp_route, - (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST)), - inp->inp_moptions); + m->m_pkthdr.m_flowhash = inp->inp_flowhash; + m->m_pkthdr.m_fhflags |= PF_TAG_FLOWHASH; + if (flowadv) + m->m_pkthdr.m_fhflags |= PF_TAG_FLOWADV; + + if (ipoa.ipoa_boundif != IFSCOPE_NONE) + ipoa.ipoa_flags |= IPOAF_BOUND_IF; + if (laddr.s_addr != INADDR_ANY) + ipoa.ipoa_flags |= IPOAF_BOUND_SRCADDR; + + inp->inp_sndinprog_cnt++; + + socket_unlock(so, 0); + error = ip_output_list(m, 0, inpopts, &ro, soopts, mopts, &ipoa); + m = NULL; + socket_lock(so, 0); + if (mopts != NULL) + IMO_REMREF(mopts); + + if (error == 0 && nstat_collect) { + locked_add_64(&inp->inp_stat->txpackets, 1); + locked_add_64(&inp->inp_stat->txbytes, len); + } + + if (flowadv && (adv->code == FADV_FLOW_CONTROLLED || + adv->code == FADV_SUSPENDED)) { + /* return a hint to the application that + * the packet has been dropped + */ + error = ENOBUFS; + inp_set_fc_state(inp, adv->code); + } + + VERIFY(inp->inp_sndinprog_cnt > 0); + if ( --inp->inp_sndinprog_cnt == 0) + inp->inp_flags &= ~(INP_FC_FEEDBACK); + + /* Synchronize PCB cached route */ + inp_route_copyin(inp, &ro); + +abort: if (udp_dodisconnect) { + /* Always discard the cached route for unconnected socket */ + if (inp->inp_route.ro_rt != NULL) { + rtfree(inp->inp_route.ro_rt); + inp->inp_route.ro_rt = NULL; + } in_pcbdisconnect(inp); inp->inp_laddr = origladdr; /* XXX rehash? */ - } - KERNEL_DEBUG(DBG_FNC_UDP_OUTPUT | DBG_FUNC_END, error, 0,0,0,0); - return (error); + inp->inp_last_outifp = origoutifp; + } else if (inp->inp_route.ro_rt != NULL) { + struct rtentry *rt = inp->inp_route.ro_rt; + struct ifnet *outifp; -abort: - if (udp_dodisconnect) { - in_pcbdisconnect(inp); - inp->inp_laddr = origladdr; /* XXX rehash? */ - } + if (rt->rt_flags & (RTF_MULTICAST|RTF_BROADCAST)) + rt = NULL; /* unusable */ + /* + * Always discard if it is a multicast or broadcast route. + */ + if (rt == NULL) { + rtfree(inp->inp_route.ro_rt); + inp->inp_route.ro_rt = NULL; + } + /* + * If the destination route is unicast, update outifp with + * that of the route interface used by IP. + */ + if (rt != NULL && (outifp = rt->rt_ifp) != inp->inp_last_outifp) + inp->inp_last_outifp = outifp; + } release: - m_freem(m); + if (m != NULL) + m_freem(m); KERNEL_DEBUG(DBG_FNC_UDP_OUTPUT | DBG_FUNC_END, error, 0,0,0,0); return (error); } -u_long udp_sendspace = 9216; /* really max datagram size */ - /* 40 1K datagrams */ -SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, - &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); - -u_long udp_recvspace = 40 * (1024 + +u_int32_t udp_sendspace = 9216; /* really max datagram size */ +/* 187 1K datagrams (approx 192 KB) */ +u_int32_t udp_recvspace = 187 * (1024 + #if INET6 sizeof(struct sockaddr_in6) #else sizeof(struct sockaddr_in) #endif ); -SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, - &udp_recvspace, 0, "Maximum incoming UDP datagram size"); + +/* Check that the values of udp send and recv space do not exceed sb_max */ +static int +sysctl_udp_sospace(struct sysctl_oid *oidp, __unused void *arg1, + __unused int arg2, struct sysctl_req *req) { + u_int32_t new_value = 0, *space_p = NULL; + int changed = 0, error = 0; + u_quad_t sb_effective_max = (sb_max/ (MSIZE+MCLBYTES)) * MCLBYTES; + + switch (oidp->oid_number) { + case UDPCTL_RECVSPACE: + space_p = &udp_recvspace; + break; + case UDPCTL_MAXDGRAM: + space_p = &udp_sendspace; + break; + default: + return EINVAL; + } + error = sysctl_io_number(req, *space_p, sizeof(u_int32_t), + &new_value, &changed); + if (changed) { + if (new_value > 0 && new_value <= sb_effective_max) { + *space_p = new_value; + } else { + error = ERANGE; + } + } + return error; +} + +SYSCTL_PROC(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, + &udp_recvspace, 0, &sysctl_udp_sospace, "IU", "Maximum incoming UDP datagram size"); + +SYSCTL_PROC(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, + &udp_sendspace, 0, &sysctl_udp_sospace, "IU", "Maximum outgoing UDP datagram size"); static int udp_abort(struct socket *so) @@ -1099,21 +1686,21 @@ udp_abort(struct socket *so) inp = sotoinpcb(so); if (inp == 0) - panic("udp_abort: so=%x null inp\n", so); /* ??? possible? panic instead? */ + panic("udp_abort: so=%p null inp\n", so); /* ??? possible? panic instead? */ soisdisconnected(so); in_pcbdetach(inp); return 0; } static int -udp_attach(struct socket *so, int proto, struct proc *p) +udp_attach(struct socket *so, __unused int proto, struct proc *p) { struct inpcb *inp; int error; inp = sotoinpcb(so); if (inp != 0) - panic ("udp_attach so=%x inp=%x\n", so, inp); + panic ("udp_attach so=%p inp=%p\n", so, inp); error = in_pcballoc(so, &udbinfo, p); if (error) @@ -1124,6 +1711,8 @@ udp_attach(struct socket *so, int proto, struct proc *p) inp = (struct inpcb *)so->so_pcb; inp->inp_vflag |= INP_IPV4; inp->inp_ip_ttl = ip_defttl; + if (nstat_collect) + nstat_udp_new_pcb(inp); return 0; } @@ -1133,6 +1722,10 @@ udp_bind(struct socket *so, struct sockaddr *nam, struct proc *p) struct inpcb *inp; int error; + if (nam->sa_family != 0 && nam->sa_family != AF_INET + && nam->sa_family != AF_INET6) { + return EAFNOSUPPORT; + } inp = sotoinpcb(so); if (inp == 0) return EINVAL; @@ -1151,9 +1744,12 @@ udp_connect(struct socket *so, struct sockaddr *nam, struct proc *p) return EINVAL; if (inp->inp_faddr.s_addr != INADDR_ANY) return EISCONN; - error = in_pcbconnect(inp, nam, p); - if (error == 0) + error = in_pcbconnect(inp, nam, p, NULL); + if (error == 0) { soisconnected(so); + if (inp->inp_flowhash == 0) + inp->inp_flowhash = inp_calc_flowhash(inp); + } return error; } @@ -1164,7 +1760,7 @@ udp_detach(struct socket *so) inp = sotoinpcb(so); if (inp == 0) - panic("udp_detach: so=%x null inp\n", so); /* ??? possible? panic instead? */ + panic("udp_detach: so=%p null inp\n", so); /* ??? possible? panic instead? */ in_pcbdetach(inp); inp->inp_state = INPCB_STATE_DEAD; return 0; @@ -1182,13 +1778,18 @@ udp_disconnect(struct socket *so) return ENOTCONN; in_pcbdisconnect(inp); + + /* reset flow controlled state, just in case */ + inp_reset_fc_state(inp); + inp->inp_laddr.s_addr = INADDR_ANY; so->so_state &= ~SS_ISCONNECTED; /* XXX */ + inp->inp_last_outifp = NULL; return 0; } static int -udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, +udp_send(struct socket *so, __unused int flags, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct proc *p) { struct inpcb *inp; @@ -1198,6 +1799,7 @@ udp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, m_freem(m); return EINVAL; } + return udp_output(inp, m, addr, control, p); } @@ -1223,89 +1825,72 @@ struct pr_usrreqs udp_usrreqs = { int -udp_lock(so, refcount, debug) - struct socket *so; - int refcount, debug; +udp_lock(struct socket *so, int refcount, void *debug) { - int lr_saved; -#ifdef __ppc__ - if (debug == 0) { - __asm__ volatile("mflr %0" : "=r" (lr_saved)); - } - else lr_saved = debug; -#endif + void *lr_saved; + + if (debug == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = debug; if (so->so_pcb) { - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_NOTOWNED); - lck_mtx_lock(((struct inpcb *)so->so_pcb)->inpcb_mtx); - } - else { - panic("udp_lock: so=%x NO PCB! lr=%x\n", so, lr_saved); - lck_mtx_assert(so->so_proto->pr_domain->dom_mtx, LCK_MTX_ASSERT_NOTOWNED); - lck_mtx_lock(so->so_proto->pr_domain->dom_mtx); + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, + LCK_MTX_ASSERT_NOTOWNED); + lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); + } else { + panic("udp_lock: so=%p NO PCB! lr=%p lrh= %s\n", + so, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ } - - if (refcount) + if (refcount) so->so_usecount++; - so->reserved3= lr_saved; + so->lock_lr[so->next_lock_lr] = lr_saved; + so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; return (0); } int -udp_unlock(so, refcount, debug) - struct socket *so; - int refcount; - int debug; +udp_unlock(struct socket *so, int refcount, void *debug) { - int lr_saved; - struct inpcb *inp = sotoinpcb(so); - struct inpcbinfo *pcbinfo = &udbinfo; -#ifdef __ppc__ - if (debug == 0) { - __asm__ volatile("mflr %0" : "=r" (lr_saved)); - } - else lr_saved = debug; -#endif - if (refcount) { + void *lr_saved; + + if (debug == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = debug; + + if (refcount) so->so_usecount--; -#if 0 - if (so->so_usecount == 0 && (inp->inp_wantcnt == WNT_STOPUSING)) { - if (lck_rw_try_lock_exclusive(pcbinfo->mtx)) { - in_pcbdispose(inp); - lck_rw_done(pcbinfo->mtx); - return(0); - } - } -#endif - } + if (so->so_pcb == NULL) { - panic("udp_unlock: so=%x NO PCB! lr=%x\n", so, lr_saved); - lck_mtx_assert(so->so_proto->pr_domain->dom_mtx, LCK_MTX_ASSERT_OWNED); - lck_mtx_unlock(so->so_proto->pr_domain->dom_mtx); - } - else { - lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED); - lck_mtx_unlock(((struct inpcb *)so->so_pcb)->inpcb_mtx); + panic("udp_unlock: so=%p NO PCB! lr=%p lrh= %s\n", + so, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ + } else { + lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx, + LCK_MTX_ASSERT_OWNED); + so->unlock_lr[so->next_unlock_lr] = lr_saved; + so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; + lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx); } - so->reserved4 = lr_saved; return (0); } lck_mtx_t * -udp_getlock(so, locktype) - struct socket *so; - int locktype; +udp_getlock(struct socket *so, __unused int locktype) { struct inpcb *inp = sotoinpcb(so); if (so->so_pcb) - return(inp->inpcb_mtx); + return(&inp->inpcb_mtx); else { - panic("udp_getlock: so=%x NULL so_pcb\n", so); + panic("udp_getlock: so=%p NULL so_pcb lrh= %s\n", + so, solockhistory_nr(so)); return (so->so_proto->pr_domain->dom_mtx); } } @@ -1317,26 +1902,39 @@ udp_slowtimo() struct socket *so; struct inpcbinfo *pcbinfo = &udbinfo; - lck_rw_lock_exclusive(pcbinfo->mtx); + if (lck_rw_try_lock_exclusive(pcbinfo->mtx) == FALSE) { + if (udp_gc_done == TRUE) { + udp_gc_done = FALSE; + return; /* couldn't get the lock, better lock next time */ + } + lck_rw_lock_exclusive(pcbinfo->mtx); + } + + udp_gc_done = TRUE; for (inp = udb.lh_first; inp != NULL; inp = inpnxt) { inpnxt = inp->inp_list.le_next; - - /* Ignore nat/SharedIP dummy pcbs */ - if (inp->inp_socket == &udbinfo.nat_dummy_socket) - continue; if (inp->inp_wantcnt != WNT_STOPUSING) continue; so = inp->inp_socket; - if (!lck_mtx_try_lock(inp->inpcb_mtx)) /* skip if busy, no hurry for cleanup... */ + if (!lck_mtx_try_lock(&inp->inpcb_mtx)) /* skip if busy, no hurry for cleanup... */ continue; - if (so->so_usecount == 0) + if (so->so_usecount == 0) { + if (inp->inp_state != INPCB_STATE_DEAD) { +#if INET6 + if (INP_CHECK_SOCKAF(so, AF_INET6)) + in6_pcbdetach(inp); + else +#endif /* INET6 */ + in_pcbdetach(inp); + } in_pcbdispose(inp); - else - lck_mtx_unlock(inp->inpcb_mtx); + } else { + lck_mtx_unlock(&inp->inpcb_mtx); + } } lck_rw_done(pcbinfo->mtx); } @@ -1350,3 +1948,16 @@ ChkAddressOK( __uint32_t dstaddr, __uint32_t srcaddr ) return 1; } +void +udp_in_cksum_stats(u_int32_t len) +{ + udps_in_sw_cksum++; + udps_in_sw_cksum_bytes += len; +} + +void +udp_out_cksum_stats(u_int32_t len) +{ + udps_out_sw_cksum++; + udps_out_sw_cksum_bytes += len; +}