Please note that diffs are not public domain; they are subject to the copyright notices on the relevant files. =================================================================== RCS file: /ftp/cvs/cvsroot/src/sys/netinet/ip_input.c,v rcsdiff: /ftp/cvs/cvsroot/src/sys/netinet/ip_input.c,v: warning: Unknown phrases like `commitid ...;' are present. retrieving revision 1.343 retrieving revision 1.347.2.1 diff -u -p -r1.343 -r1.347.2.1 --- src/sys/netinet/ip_input.c 2016/10/18 01:15:20 1.343 +++ src/sys/netinet/ip_input.c 2017/04/21 16:54:06 1.347.2.1 @@ -1,4 +1,4 @@ -/* $NetBSD: ip_input.c,v 1.343 2016/10/18 01:15:20 ozaki-r Exp $ */ +/* $NetBSD: ip_input.c,v 1.347.2.1 2017/04/21 16:54:06 bouyer Exp $ */ /* * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. @@ -91,7 +91,7 @@ */ #include -__KERNEL_RCSID(0, "$NetBSD: ip_input.c,v 1.343 2016/10/18 01:15:20 ozaki-r Exp $"); +__KERNEL_RCSID(0, "$NetBSD: ip_input.c,v 1.347.2.1 2017/04/21 16:54:06 bouyer Exp $"); #ifdef _KERNEL_OPT #include "opt_inet.h" @@ -101,6 +101,7 @@ __KERNEL_RCSID(0, "$NetBSD: ip_input.c,v #include "opt_mrouting.h" #include "opt_mbuftrace.h" #include "opt_inet_csum.h" +#include "opt_net_mpsafe.h" #endif #include "arp.h" @@ -252,30 +253,23 @@ EVCNT_ATTACH_STATIC(ip_swcsum); #endif /* INET_CSUM_COUNTERS */ /* - * We need to save the IP options in case a protocol wants to respond + * Used to save the IP options in case a protocol wants to respond * to an incoming packet over the same route if the packet got here * using IP source routing. This allows connection establishment and * maintenance when the remote end is on a network that is not known * to us. */ - -static int ip_nhops = 0; - -static struct ip_srcrt { - struct in_addr dst; /* final destination */ - char nop; /* one NOP to align */ - char srcopt[IPOPT_OFFSET + 1]; /* OPTVAL, OLEN and OFFSET */ - struct in_addr route[MAX_IPOPTLEN/sizeof(struct in_addr)]; -} ip_srcrt; +struct ip_srcrt { + int isr_nhops; /* number of hops */ + struct in_addr isr_dst; /* final destination */ + char isr_nop; /* one NOP to align */ + char isr_hdr[IPOPT_OFFSET + 1]; /* OPTVAL, OLEN & OFFSET */ + struct in_addr isr_routes[MAX_IPOPTLEN/sizeof(struct in_addr)]; +}; static int ip_drainwanted; -struct sockaddr_in ipaddr = { - .sin_len = sizeof(ipaddr), - .sin_family = AF_INET, -}; - -static void save_rte(u_char *, struct in_addr); +static void save_rte(struct mbuf *, u_char *, struct in_addr); #ifdef MBUFTRACE struct mowner ip_rx_mowner = MOWNER_INIT("internet", "rx"); @@ -286,7 +280,7 @@ static void ipintr(void *); static void ip_input(struct mbuf *); static void ip_forward(struct mbuf *, int, struct ifnet *); static bool ip_dooptions(struct mbuf *); -static struct in_ifaddr *ip_rtaddr(struct in_addr); +static struct in_ifaddr *ip_rtaddr(struct in_addr, struct psref *); static void sysctl_net_inet_ip_setup(struct sysctllog **); static struct in_ifaddr *ip_match_our_address(struct ifnet *, struct ip *, @@ -294,9 +288,13 @@ static struct in_ifaddr *ip_match_our_ad static struct in_ifaddr *ip_match_our_address_broadcast(struct ifnet *, struct ip *); -/* XXX: Not yet enabled. */ +#ifdef NET_MPSAFE +#define SOFTNET_LOCK() mutex_enter(softnet_lock) +#define SOFTNET_UNLOCK() mutex_exit(softnet_lock) +#else #define SOFTNET_LOCK() KASSERT(mutex_owned(softnet_lock)) #define SOFTNET_UNLOCK() KASSERT(mutex_owned(softnet_lock)) +#endif /* * IP initialization: fill in IP protocol switch table. @@ -330,7 +328,6 @@ ip_init(void) ip_ids = ip_id_init(); ip_id = time_uptime & 0xfffff; - ip_mtudisc_timeout_q = rt_timer_queue_create(ip_mtudisc_timeout); #ifdef GATEWAY ipflow_init(); #endif @@ -349,6 +346,8 @@ ip_init(void) ipforward_rt_percpu = percpu_alloc(sizeof(struct route)); if (ipforward_rt_percpu == NULL) panic("failed to allocate ipforward_rt_percpu"); + + ip_mtudisc_timeout_q = rt_timer_queue_create(ip_mtudisc_timeout); } static struct in_ifaddr * @@ -437,11 +436,15 @@ ipintr(void *arg __unused) KASSERT(cpu_softintr_p()); +#ifndef NET_MPSAFE mutex_enter(softnet_lock); +#endif while ((m = pktq_dequeue(ip_pktq)) != NULL) { ip_input(m); } +#ifndef NET_MPSAFE mutex_exit(softnet_lock); +#endif } /* @@ -614,9 +617,7 @@ ip_input(struct mbuf *m) struct in_addr odst = ip->ip_dst; bool freed; - SOFTNET_LOCK(); freed = pfil_run_hooks(inet_pfil_hook, &m, ifp, PFIL_IN) != 0; - SOFTNET_UNLOCK(); if (freed || m == NULL) { m = NULL; goto out; @@ -661,7 +662,6 @@ ip_input(struct mbuf *m) * error was detected (causing an icmp message * to be sent and the original packet to be freed). */ - ip_nhops = 0; /* for source routed packets */ if (hlen > sizeof (struct ip) && ip_dooptions(m)) { m = NULL; goto out; @@ -860,13 +860,17 @@ void ip_slowtimo(void) { +#ifndef NET_MPSAFE mutex_enter(softnet_lock); KERNEL_LOCK(1, NULL); +#endif ip_reass_slowtimo(); +#ifndef NET_MPSAFE KERNEL_UNLOCK_ONE(NULL); mutex_exit(softnet_lock); +#endif } /* @@ -898,7 +902,7 @@ ip_dooptions(struct mbuf *m) int opt, optlen, cnt, off, code, type = ICMP_PARAMPROB, forward = 0; struct in_addr dst; n_time ntime; - struct ifaddr *ifa; + struct ifaddr *ifa = NULL; int s; dst = ip->ip_dst; @@ -936,7 +940,13 @@ ip_dooptions(struct mbuf *m) * address is on directly accessible net. */ case IPOPT_LSRR: - case IPOPT_SSRR: + case IPOPT_SSRR: { + struct psref psref; + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; + if (ip_allowsrcrt == 0) { type = ICMP_UNREACH; code = ICMP_UNREACH_NET_PROHIB; @@ -974,7 +984,7 @@ ip_dooptions(struct mbuf *m) /* * End of source route. Should be for us. */ - save_rte(cp, ip->ip_src); + save_rte(m, cp, ip->ip_src); break; } /* @@ -982,34 +992,40 @@ ip_dooptions(struct mbuf *m) */ memcpy((void *)&ipaddr.sin_addr, (void *)(cp + off), sizeof(ipaddr.sin_addr)); - s = pserialize_read_enter(); if (opt == IPOPT_SSRR) { - ifa = ifa_ifwithladdr(sintosa(&ipaddr)); + ifa = ifa_ifwithladdr_psref(sintosa(&ipaddr), + &psref); if (ifa != NULL) ia = ifatoia(ifa); else ia = NULL; } else { - ia = ip_rtaddr(ipaddr.sin_addr); + ia = ip_rtaddr(ipaddr.sin_addr, &psref); } if (ia == NULL) { type = ICMP_UNREACH; code = ICMP_UNREACH_SRCFAIL; - pserialize_read_exit(s); goto bad; } ip->ip_dst = ipaddr.sin_addr; bcopy((void *)&ia->ia_addr.sin_addr, (void *)(cp + off), sizeof(struct in_addr)); - pserialize_read_exit(s); + ia4_release(ia, &psref); cp[IPOPT_OFFSET] += sizeof(struct in_addr); /* * Let ip_intr's mcast routing check handle mcast pkts */ forward = !IN_MULTICAST(ip->ip_dst.s_addr); break; + } + + case IPOPT_RR: { + struct psref psref; + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; - case IPOPT_RR: if (optlen < IPOPT_OFFSET + sizeof(*cp)) { code = &cp[IPOPT_OLEN] - (u_char *)ip; goto bad; @@ -1030,12 +1046,10 @@ ip_dooptions(struct mbuf *m) * locate outgoing interface; if we're the destination, * use the incoming interface (should be same). */ - s = pserialize_read_enter(); - ifa = ifa_ifwithaddr(sintosa(&ipaddr)); + ifa = ifa_ifwithaddr_psref(sintosa(&ipaddr), &psref); if (ifa == NULL) { - ia = ip_rtaddr(ipaddr.sin_addr); + ia = ip_rtaddr(ipaddr.sin_addr, &psref); if (ia == NULL) { - pserialize_read_exit(s); type = ICMP_UNREACH; code = ICMP_UNREACH_HOST; goto bad; @@ -1045,9 +1059,10 @@ ip_dooptions(struct mbuf *m) } bcopy((void *)&ia->ia_addr.sin_addr, (void *)(cp + off), sizeof(struct in_addr)); - pserialize_read_exit(s); + ia4_release(ia, &psref); cp[IPOPT_OFFSET] += sizeof(struct in_addr); break; + } case IPOPT_TS: code = cp - (u_char *)ip; @@ -1077,6 +1092,10 @@ ip_dooptions(struct mbuf *m) case IPOPT_TS_TSANDADDR: { struct ifnet *rcvif; int _s, _ss; + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; if (ipt->ipt_ptr - 1 + sizeof(n_time) + sizeof(struct in_addr) > ipt->ipt_len) { @@ -1087,7 +1106,10 @@ ip_dooptions(struct mbuf *m) ipaddr.sin_addr = dst; _ss = pserialize_read_enter(); rcvif = m_get_rcvif(m, &_s); - ifa = ifaof_ifpforaddr(sintosa(&ipaddr), rcvif); + if (__predict_true(rcvif != NULL)) { + ifa = ifaof_ifpforaddr(sintosa(&ipaddr), + rcvif); + } m_put_rcvif(rcvif, &_s); if (ifa == NULL) { pserialize_read_exit(_ss); @@ -1101,7 +1123,12 @@ ip_dooptions(struct mbuf *m) break; } - case IPOPT_TS_PRESPEC: + case IPOPT_TS_PRESPEC: { + struct sockaddr_in ipaddr = { + .sin_len = sizeof(ipaddr), + .sin_family = AF_INET, + }; + if (ipt->ipt_ptr - 1 + sizeof(n_time) + sizeof(struct in_addr) > ipt->ipt_len) { code = (u_char *)&ipt->ipt_ptr - @@ -1119,6 +1146,7 @@ ip_dooptions(struct mbuf *m) pserialize_read_exit(s); ipt->ipt_ptr += sizeof(struct in_addr); break; + } default: /* XXX can't take &ipt->ipt_flg */ @@ -1165,7 +1193,7 @@ bad: * return internet address info of interface to be used to get there. */ static struct in_ifaddr * -ip_rtaddr(struct in_addr dst) +ip_rtaddr(struct in_addr dst, struct psref *psref) { struct rtentry *rt; union { @@ -1176,13 +1204,16 @@ ip_rtaddr(struct in_addr dst) sockaddr_in_init(&u.dst4, &dst, 0); - SOFTNET_LOCK(); ro = percpu_getref(ipforward_rt_percpu); rt = rtcache_lookup(ro, &u.dst); - percpu_putref(ipforward_rt_percpu); - SOFTNET_UNLOCK(); - if (rt == NULL) + if (rt == NULL) { + percpu_putref(ipforward_rt_percpu); return NULL; + } + + ia4_acquire(ifatoia(rt->rt_ifa), psref); + rtcache_unref(rt, ro); + percpu_putref(ipforward_rt_percpu); return ifatoia(rt->rt_ifa); } @@ -1192,16 +1223,25 @@ ip_rtaddr(struct in_addr dst) * up later by ip_srcroute if the receiver is interested. */ static void -save_rte(u_char *option, struct in_addr dst) +save_rte(struct mbuf *m, u_char *option, struct in_addr dst) { + struct ip_srcrt *isr; + struct m_tag *mtag; unsigned olen; olen = option[IPOPT_OLEN]; - if (olen > sizeof(ip_srcrt) - (1 + sizeof(dst))) + if (olen > sizeof(isr->isr_hdr) + sizeof(isr->isr_routes)) + return; + + mtag = m_tag_get(PACKET_TAG_SRCROUTE, sizeof(*isr), M_NOWAIT); + if (mtag == NULL) return; - memcpy((void *)ip_srcrt.srcopt, (void *)option, olen); - ip_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr); - ip_srcrt.dst = dst; + isr = (struct ip_srcrt *)(mtag + 1); + + memcpy(isr->isr_hdr, option, olen); + isr->isr_nhops = (olen - IPOPT_OFFSET - 1) / sizeof(struct in_addr); + isr->isr_dst = dst; + m_tag_prepend(m, mtag); } /* @@ -1210,36 +1250,43 @@ save_rte(u_char *option, struct in_addr * The first hop is placed before the options, will be removed later. */ struct mbuf * -ip_srcroute(void) +ip_srcroute(struct mbuf *m0) { struct in_addr *p, *q; struct mbuf *m; + struct ip_srcrt *isr; + struct m_tag *mtag; + + mtag = m_tag_find(m0, PACKET_TAG_SRCROUTE, NULL); + if (mtag == NULL) + return NULL; + isr = (struct ip_srcrt *)(mtag + 1); - if (ip_nhops == 0) + if (isr->isr_nhops == 0) return NULL; + m = m_get(M_DONTWAIT, MT_SOOPTS); - if (m == 0) + if (m == NULL) return NULL; MCLAIM(m, &inetdomain.dom_mowner); -#define OPTSIZ (sizeof(ip_srcrt.nop) + sizeof(ip_srcrt.srcopt)) +#define OPTSIZ (sizeof(isr->isr_nop) + sizeof(isr->isr_hdr)) - /* length is (nhops+1)*sizeof(addr) + sizeof(nop + srcrt header) */ - m->m_len = ip_nhops * sizeof(struct in_addr) + sizeof(struct in_addr) + - OPTSIZ; + /* length is (nhops+1)*sizeof(addr) + sizeof(nop + header) */ + m->m_len = (isr->isr_nhops + 1) * sizeof(struct in_addr) + OPTSIZ; /* * First save first hop for return route */ - p = &ip_srcrt.route[ip_nhops - 1]; + p = &(isr->isr_routes[isr->isr_nhops - 1]); *(mtod(m, struct in_addr *)) = *p--; /* * Copy option fields and padding (nop) to mbuf. */ - ip_srcrt.nop = IPOPT_NOP; - ip_srcrt.srcopt[IPOPT_OFFSET] = IPOPT_MINOFF; - memmove(mtod(m, char *) + sizeof(struct in_addr), &ip_srcrt.nop, + isr->isr_nop = IPOPT_NOP; + isr->isr_hdr[IPOPT_OFFSET] = IPOPT_MINOFF; + memmove(mtod(m, char *) + sizeof(struct in_addr), &isr->isr_nop, OPTSIZ); q = (struct in_addr *)(mtod(m, char *) + sizeof(struct in_addr) + OPTSIZ); @@ -1248,14 +1295,15 @@ ip_srcroute(void) * Record return path as an IP source route, * reversing the path (pointers are now aligned). */ - while (p >= ip_srcrt.route) { + while (p >= isr->isr_routes) { *q++ = *p--; } /* * Last hop goes to final destination. */ - *q = ip_srcrt.dst; - return (m); + *q = isr->isr_dst; + m_tag_delete(m0, mtag); + return m; } const int inetctlerrmap[PRC_NCMDS] = { @@ -1334,21 +1382,18 @@ ip_forward(struct mbuf *m, int srcrt, st return; } - SOFTNET_LOCK(); - if (ip->ip_ttl <= IPTTLDEC) { icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, dest, 0); - SOFTNET_UNLOCK(); return; } sockaddr_in_init(&u.dst4, &ip->ip_dst, 0); ro = percpu_getref(ipforward_rt_percpu); - if ((rt = rtcache_lookup(ro, &u.dst)) == NULL) { + rt = rtcache_lookup(ro, &u.dst); + if (rt == NULL) { percpu_putref(ipforward_rt_percpu); icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NET, dest, 0); - SOFTNET_UNLOCK(); return; } @@ -1390,6 +1435,7 @@ ip_forward(struct mbuf *m, int srcrt, st code = ICMP_REDIRECT_HOST; } } + rtcache_unref(rt, ro); error = ip_output(m, NULL, ro, (IP_FORWARDING | (ip_directedbcast ? IP_ALLOWBROADCAST : 0)), @@ -1419,14 +1465,12 @@ ip_forward(struct mbuf *m, int srcrt, st } percpu_putref(ipforward_rt_percpu); - SOFTNET_UNLOCK(); return; redirect: error: if (mcopy == NULL) { percpu_putref(ipforward_rt_percpu); - SOFTNET_UNLOCK(); return; } @@ -1449,8 +1493,10 @@ error: type = ICMP_UNREACH; code = ICMP_UNREACH_NEEDFRAG; - if ((rt = rtcache_validate(ro)) != NULL) + if ((rt = rtcache_validate(ro)) != NULL) { destmtu = rt->rt_ifp->if_mtu; + rtcache_unref(rt, ro); + } #ifdef IPSEC if (ipsec_used) (void)ipsec4_forward(mcopy, &destmtu); @@ -1468,12 +1514,10 @@ error: if (mcopy) m_freem(mcopy); percpu_putref(ipforward_rt_percpu); - SOFTNET_UNLOCK(); return; } icmp_error(mcopy, type, code, dest, destmtu); percpu_putref(ipforward_rt_percpu); - SOFTNET_UNLOCK(); } void @@ -1588,23 +1632,25 @@ sysctl_net_inet_ip_pmtudto(SYSCTLFN_ARGS int error, tmp; struct sysctlnode node; + icmp_mtudisc_lock(); + node = *rnode; tmp = ip_mtudisc_timeout; node.sysctl_data = &tmp; error = sysctl_lookup(SYSCTLFN_CALL(&node)); if (error || newp == NULL) - return (error); - if (tmp < 0) - return (EINVAL); - - mutex_enter(softnet_lock); + goto out; + if (tmp < 0) { + error = EINVAL; + goto out; + } ip_mtudisc_timeout = tmp; rt_timer_queue_change(ip_mtudisc_timeout_q, ip_mtudisc_timeout); - - mutex_exit(softnet_lock); - - return (0); + error = 0; +out: + icmp_mtudisc_unlock(); + return error; } static int