| /* |
| * Copyright (C) 1993-2003 by Darren Reed. |
| * |
| * See the IPFILTER.LICENCE file for details on licencing. |
| * |
| * Copyright 2008 Sun Microsystems. |
| * |
| * $Id$ |
| * |
| */ |
| #if defined(KERNEL) || defined(_KERNEL) |
| # undef KERNEL |
| # undef _KERNEL |
| # define KERNEL 1 |
| # define _KERNEL 1 |
| #endif |
| #include <sys/errno.h> |
| #include <sys/types.h> |
| #include <sys/param.h> |
| #include <sys/time.h> |
| #if defined(_KERNEL) && defined(__FreeBSD_version) && \ |
| (__FreeBSD_version >= 220000) |
| # if (__FreeBSD_version >= 400000) |
| # if !defined(IPFILTER_LKM) |
| # include "opt_inet6.h" |
| # endif |
| # if (__FreeBSD_version == 400019) |
| # define CSUM_DELAY_DATA |
| # endif |
| # endif |
| # include <sys/filio.h> |
| #else |
| # include <sys/ioctl.h> |
| #endif |
| #if (defined(__SVR4) || defined(__svr4__)) && defined(sun) |
| # include <sys/filio.h> |
| #endif |
| #if !defined(_AIX51) |
| # include <sys/fcntl.h> |
| #endif |
| #if defined(_KERNEL) |
| # include <sys/systm.h> |
| # include <sys/file.h> |
| #else |
| # include <stdio.h> |
| # include <string.h> |
| # include <stdlib.h> |
| # include <stddef.h> |
| # include <sys/file.h> |
| # define _KERNEL |
| # ifdef __OpenBSD__ |
| struct file; |
| # endif |
| # include <sys/uio.h> |
| # undef _KERNEL |
| #endif |
| #if !defined(__SVR4) && !defined(__svr4__) && !defined(__hpux) && \ |
| !defined(linux) |
| # include <sys/mbuf.h> |
| #else |
| # if !defined(linux) |
| # include <sys/byteorder.h> |
| # endif |
| # if (SOLARIS2 < 5) && defined(sun) |
| # include <sys/dditypes.h> |
| # endif |
| #endif |
| #ifdef __hpux |
| # define _NET_ROUTE_INCLUDED |
| #endif |
| #if !defined(linux) |
| # include <sys/protosw.h> |
| #endif |
| #include <sys/socket.h> |
| #include <net/if.h> |
| #ifdef sun |
| # include <net/af.h> |
| #endif |
| #include <netinet/in.h> |
| #include <netinet/in_systm.h> |
| #include <netinet/ip.h> |
| #if defined(__sgi) && defined(IFF_DRVRLOCK) /* IRIX 6 */ |
| # include <sys/hashing.h> |
| # include <netinet/in_var.h> |
| #endif |
| #include <netinet/tcp.h> |
| #if (!defined(__sgi) && !defined(AIX)) || defined(_KERNEL) |
| # include <netinet/udp.h> |
| # include <netinet/ip_icmp.h> |
| #endif |
| #ifdef __hpux |
| # undef _NET_ROUTE_INCLUDED |
| #endif |
| #ifdef __osf__ |
| # undef _RADIX_H_ |
| #endif |
| #include "netinet/ip_compat.h" |
| #ifdef USE_INET6 |
| # include <netinet/icmp6.h> |
| # if !SOLARIS && defined(_KERNEL) && !defined(__osf__) && !defined(__hpux) |
| # include <netinet6/in6_var.h> |
| # endif |
| #endif |
| #include "netinet/ip_fil.h" |
| #include "netinet/ip_nat.h" |
| #include "netinet/ip_frag.h" |
| #include "netinet/ip_state.h" |
| #include "netinet/ip_proxy.h" |
| #include "netinet/ip_auth.h" |
| #ifdef IPFILTER_SCAN |
| # include "netinet/ip_scan.h" |
| #endif |
| #include "netinet/ip_sync.h" |
| #include "netinet/ip_pool.h" |
| #include "netinet/ip_htable.h" |
| #ifdef IPFILTER_COMPILED |
| # include "netinet/ip_rules.h" |
| #endif |
| #if defined(IPFILTER_BPF) && defined(_KERNEL) |
| # include <net/bpf.h> |
| #endif |
| #if defined(__FreeBSD_version) && (__FreeBSD_version >= 300000) |
| # include <sys/malloc.h> |
| #endif |
| #include "netinet/ipl.h" |
| |
| #if defined(__NetBSD__) && (__NetBSD_Version__ >= 104230000) |
| # include <sys/callout.h> |
| extern struct callout ipf_slowtimer_ch; |
| #endif |
| #if defined(__OpenBSD__) |
| # include <sys/timeout.h> |
| extern struct timeout ipf_slowtimer_ch; |
| #endif |
| /* END OF INCLUDES */ |
| |
| #if !defined(lint) |
| static const char sccsid[] = "@(#)fil.c 1.36 6/5/96 (C) 1993-2000 Darren Reed"; |
| static const char rcsid[] = "@(#)$Id$"; |
| #endif |
| |
| #ifndef _KERNEL |
| # include "ipf.h" |
| # include "ipt.h" |
| # include "bpf-ipf.h" |
| extern int opts; |
| extern int blockreason; |
| #endif /* _KERNEL */ |
| |
| static INLINE int ipf_check_ipf __P((fr_info_t *, frentry_t *, int)); |
| static u_32_t ipf_checkcipso __P((fr_info_t *, u_char *, int)); |
| static u_32_t ipf_checkripso __P((u_char *)); |
| static u_32_t ipf_decaps __P((fr_info_t *, u_32_t, int)); |
| static frentry_t *ipf_dolog __P((fr_info_t *, u_32_t *)); |
| static int ipf_flushlist __P((ipf_main_softc_t *, int, minor_t, |
| int *, frentry_t **)); |
| static int ipf_flush_groups __P((ipf_main_softc_t *, |
| int, int, int)); |
| static ipfunc_t ipf_findfunc __P((ipfunc_t)); |
| static void *ipf_findlookup __P((ipf_main_softc_t *, int, |
| frentry_t *, i6addr_t *)); |
| static frentry_t *ipf_firewall __P((fr_info_t *, u_32_t *)); |
| static int ipf_fr_matcharray __P((fr_info_t *, int *)); |
| static int ipf_frruleiter __P((ipf_main_softc_t *, void *, int, |
| void *)); |
| static int ipf_funcinit __P((ipf_main_softc_t *, frentry_t *fr)); |
| static int ipf_geniter __P((ipf_main_softc_t *, ipftoken_t *, |
| ipfgeniter_t *)); |
| static int ipf_grpmapinit __P((struct ipf_main_softc_s *, |
| frentry_t *)); |
| static int ipf_portcheck __P((frpcmp_t *, u_32_t)); |
| static INLINE int ipf_pr_ah __P((fr_info_t *)); |
| static INLINE void ipf_pr_esp __P((fr_info_t *)); |
| static INLINE void ipf_pr_gre __P((fr_info_t *)); |
| static INLINE void ipf_pr_udp __P((fr_info_t *)); |
| static INLINE void ipf_pr_tcp __P((fr_info_t *)); |
| static INLINE void ipf_pr_icmp __P((fr_info_t *)); |
| static INLINE void ipf_pr_ipv4hdr __P((fr_info_t *)); |
| static INLINE void ipf_pr_short __P((fr_info_t *, int)); |
| static INLINE int ipf_pr_tcpcommon __P((fr_info_t *)); |
| static INLINE int ipf_pr_udpcommon __P((fr_info_t *)); |
| static void ipf_rule_delete __P((ipf_main_softc_t *, frentry_t *f, |
| int, int)); |
| static void ipf_rule_expire_insert __P((ipf_main_softc_t *, |
| frentry_t *, int)); |
| static void ipf_synclist __P((ipf_main_softc_t *, frentry_t *, |
| void *)); |
| static ipftuneable_t *ipf_tune_findbyname __P((ipftuneable_t *, |
| const char *)); |
| static ipftuneable_t *ipf_tune_findbycookie __P((ipftuneable_t **, void *, |
| void **)); |
| static void ipf_token_unlink __P((ipf_main_softc_t *, |
| ipftoken_t *)); |
| static int ipf_updateipid __P((fr_info_t *)); |
| static int ipf_settimeout __P((struct ipf_main_softc_s *, |
| struct ipftuneable *, |
| ipftuneval_t *)); |
| |
| |
| /* |
| * bit values for identifying presence of individual IP options |
| * All of these tables should be ordered by increasing key value on the left |
| * hand side to allow for binary searching of the array and include a trailer |
| * with a 0 for the bitmask for linear searches to easily find the end with. |
| */ |
| static const struct optlist ipopts[20] = { |
| { IPOPT_NOP, 0x000001 }, |
| { IPOPT_RR, 0x000002 }, |
| { IPOPT_ZSU, 0x000004 }, |
| { IPOPT_MTUP, 0x000008 }, |
| { IPOPT_MTUR, 0x000010 }, |
| { IPOPT_ENCODE, 0x000020 }, |
| { IPOPT_TS, 0x000040 }, |
| { IPOPT_TR, 0x000080 }, |
| { IPOPT_SECURITY, 0x000100 }, |
| { IPOPT_LSRR, 0x000200 }, |
| { IPOPT_E_SEC, 0x000400 }, |
| { IPOPT_CIPSO, 0x000800 }, |
| { IPOPT_SATID, 0x001000 }, |
| { IPOPT_SSRR, 0x002000 }, |
| { IPOPT_ADDEXT, 0x004000 }, |
| { IPOPT_VISA, 0x008000 }, |
| { IPOPT_IMITD, 0x010000 }, |
| { IPOPT_EIP, 0x020000 }, |
| { IPOPT_FINN, 0x040000 }, |
| { 0, 0x000000 } |
| }; |
| |
| #ifdef USE_INET6 |
| static struct optlist ip6exthdr[] = { |
| { IPPROTO_HOPOPTS, 0x000001 }, |
| { IPPROTO_IPV6, 0x000002 }, |
| { IPPROTO_ROUTING, 0x000004 }, |
| { IPPROTO_FRAGMENT, 0x000008 }, |
| { IPPROTO_ESP, 0x000010 }, |
| { IPPROTO_AH, 0x000020 }, |
| { IPPROTO_NONE, 0x000040 }, |
| { IPPROTO_DSTOPTS, 0x000080 }, |
| { IPPROTO_MOBILITY, 0x000100 }, |
| { 0, 0 } |
| }; |
| #endif |
| |
| /* |
| * bit values for identifying presence of individual IP security options |
| */ |
| static const struct optlist secopt[8] = { |
| { IPSO_CLASS_RES4, 0x01 }, |
| { IPSO_CLASS_TOPS, 0x02 }, |
| { IPSO_CLASS_SECR, 0x04 }, |
| { IPSO_CLASS_RES3, 0x08 }, |
| { IPSO_CLASS_CONF, 0x10 }, |
| { IPSO_CLASS_UNCL, 0x20 }, |
| { IPSO_CLASS_RES2, 0x40 }, |
| { IPSO_CLASS_RES1, 0x80 } |
| }; |
| |
| char ipfilter_version[] = IPL_VERSION; |
| |
| int ipf_features = 0 |
| #ifdef IPFILTER_LKM |
| | IPF_FEAT_LKM |
| #endif |
| #ifdef IPFILTER_LOG |
| | IPF_FEAT_LOG |
| #endif |
| | IPF_FEAT_LOOKUP |
| #ifdef IPFILTER_BPF |
| | IPF_FEAT_BPF |
| #endif |
| #ifdef IPFILTER_COMPILED |
| | IPF_FEAT_COMPILED |
| #endif |
| #ifdef IPFILTER_CKSUM |
| | IPF_FEAT_CKSUM |
| #endif |
| | IPF_FEAT_SYNC |
| #ifdef IPFILTER_SCAN |
| | IPF_FEAT_SCAN |
| #endif |
| #ifdef USE_INET6 |
| | IPF_FEAT_IPV6 |
| #endif |
| ; |
| |
| |
| /* |
| * Table of functions available for use with call rules. |
| */ |
| static ipfunc_resolve_t ipf_availfuncs[] = { |
| { "srcgrpmap", ipf_srcgrpmap, ipf_grpmapinit }, |
| { "dstgrpmap", ipf_dstgrpmap, ipf_grpmapinit }, |
| { "", NULL, NULL } |
| }; |
| |
| static ipftuneable_t ipf_main_tuneables[] = { |
| { { (void *)offsetof(struct ipf_main_softc_s, ipf_flags) }, |
| "ipf_flags", 0, 0xffffffff, |
| stsizeof(ipf_main_softc_t, ipf_flags), |
| 0, NULL, NULL }, |
| { { (void *)offsetof(struct ipf_main_softc_s, ipf_active) }, |
| "active", 0, 0, |
| stsizeof(ipf_main_softc_t, ipf_active), |
| IPFT_RDONLY, NULL, NULL }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_control_forwarding) }, |
| "control_forwarding", 0, 1, |
| stsizeof(ipf_main_softc_t, ipf_control_forwarding), |
| 0, NULL, NULL }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_update_ipid) }, |
| "update_ipid", 0, 1, |
| stsizeof(ipf_main_softc_t, ipf_update_ipid), |
| 0, NULL, NULL }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_chksrc) }, |
| "chksrc", 0, 1, |
| stsizeof(ipf_main_softc_t, ipf_chksrc), |
| 0, NULL, NULL }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_minttl) }, |
| "min_ttl", 0, 1, |
| stsizeof(ipf_main_softc_t, ipf_minttl), |
| 0, NULL, NULL }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_icmpminfragmtu) }, |
| "icmp_minfragmtu", 0, 1, |
| stsizeof(ipf_main_softc_t, ipf_icmpminfragmtu), |
| 0, NULL, NULL }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_pass) }, |
| "default_pass", 0, 0xffffffff, |
| stsizeof(ipf_main_softc_t, ipf_pass), |
| 0, NULL, NULL }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcpidletimeout) }, |
| "tcp_idle_timeout", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcpidletimeout), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcpclosewait) }, |
| "tcp_close_wait", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcpclosewait), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcplastack) }, |
| "tcp_last_ack", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcplastack), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcptimeout) }, |
| "tcp_timeout", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcptimeout), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcpsynsent) }, |
| "tcp_syn_sent", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcpsynsent), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcpsynrecv) }, |
| "tcp_syn_received", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcpsynrecv), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcpclosed) }, |
| "tcp_closed", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcpclosed), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcphalfclosed) }, |
| "tcp_half_closed", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcphalfclosed), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_tcptimewait) }, |
| "tcp_time_wait", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_tcptimewait), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_udptimeout) }, |
| "udp_timeout", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_udptimeout), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_udpacktimeout) }, |
| "udp_ack_timeout", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_udpacktimeout), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_icmptimeout) }, |
| "icmp_timeout", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_icmptimeout), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_icmpacktimeout) }, |
| "icmp_ack_timeout", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_icmpacktimeout), |
| 0, NULL, ipf_settimeout }, |
| { { (void *)offsetof(ipf_main_softc_t, ipf_iptimeout) }, |
| "ip_timeout", 1, 0x7fffffff, |
| stsizeof(ipf_main_softc_t, ipf_iptimeout), |
| 0, NULL, ipf_settimeout }, |
| #if defined(INSTANCES) && defined(_KERNEL) |
| { { (void *)offsetof(ipf_main_softc_t, ipf_get_loopback) }, |
| "intercept_loopback", 0, 1, |
| stsizeof(ipf_main_softc_t, ipf_get_loopback), |
| 0, NULL, ipf_set_loopback }, |
| #endif |
| { { 0 }, |
| NULL, 0, 0, |
| 0, |
| 0, NULL, NULL } |
| }; |
| |
| |
| /* |
| * The next section of code is a a collection of small routines that set |
| * fields in the fr_info_t structure passed based on properties of the |
| * current packet. There are different routines for the same protocol |
| * for each of IPv4 and IPv6. Adding a new protocol, for which there |
| * will "special" inspection for setup, is now more easily done by adding |
| * a new routine and expanding the ipf_pr_ipinit*() function rather than by |
| * adding more code to a growing switch statement. |
| */ |
| #ifdef USE_INET6 |
| static INLINE int ipf_pr_ah6 __P((fr_info_t *)); |
| static INLINE void ipf_pr_esp6 __P((fr_info_t *)); |
| static INLINE void ipf_pr_gre6 __P((fr_info_t *)); |
| static INLINE void ipf_pr_udp6 __P((fr_info_t *)); |
| static INLINE void ipf_pr_tcp6 __P((fr_info_t *)); |
| static INLINE void ipf_pr_icmp6 __P((fr_info_t *)); |
| static INLINE void ipf_pr_ipv6hdr __P((fr_info_t *)); |
| static INLINE void ipf_pr_short6 __P((fr_info_t *, int)); |
| static INLINE int ipf_pr_hopopts6 __P((fr_info_t *)); |
| static INLINE int ipf_pr_mobility6 __P((fr_info_t *)); |
| static INLINE int ipf_pr_routing6 __P((fr_info_t *)); |
| static INLINE int ipf_pr_dstopts6 __P((fr_info_t *)); |
| static INLINE int ipf_pr_fragment6 __P((fr_info_t *)); |
| static INLINE struct ip6_ext *ipf_pr_ipv6exthdr __P((fr_info_t *, int, int)); |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_short6 */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* xmin(I) - minimum header size */ |
| /* */ |
| /* IPv6 Only */ |
| /* This is function enforces the 'is a packet too short to be legit' rule */ |
| /* for IPv6 and marks the packet with FI_SHORT if so. See function comment */ |
| /* for ipf_pr_short() for more details. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_short6(fin, xmin) |
| fr_info_t *fin; |
| int xmin; |
| { |
| |
| if (fin->fin_dlen < xmin) |
| fin->fin_flx |= FI_SHORT; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_ipv6hdr */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* Copy values from the IPv6 header into the fr_info_t struct and call the */ |
| /* per-protocol analyzer if it exists. In validating the packet, a protocol*/ |
| /* analyzer may pullup or free the packet itself so we need to be vigiliant */ |
| /* of that possibility arising. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_ipv6hdr(fin) |
| fr_info_t *fin; |
| { |
| ip6_t *ip6 = (ip6_t *)fin->fin_ip; |
| int p, go = 1, i, hdrcount; |
| fr_ip_t *fi = &fin->fin_fi; |
| |
| fin->fin_off = 0; |
| |
| fi->fi_tos = 0; |
| fi->fi_optmsk = 0; |
| fi->fi_secmsk = 0; |
| fi->fi_auth = 0; |
| |
| p = ip6->ip6_nxt; |
| fin->fin_crc = p; |
| fi->fi_ttl = ip6->ip6_hlim; |
| fi->fi_src.in6 = ip6->ip6_src; |
| fin->fin_crc += fi->fi_src.i6[0]; |
| fin->fin_crc += fi->fi_src.i6[1]; |
| fin->fin_crc += fi->fi_src.i6[2]; |
| fin->fin_crc += fi->fi_src.i6[3]; |
| fi->fi_dst.in6 = ip6->ip6_dst; |
| fin->fin_crc += fi->fi_dst.i6[0]; |
| fin->fin_crc += fi->fi_dst.i6[1]; |
| fin->fin_crc += fi->fi_dst.i6[2]; |
| fin->fin_crc += fi->fi_dst.i6[3]; |
| fin->fin_id = 0; |
| |
| hdrcount = 0; |
| while (go && !(fin->fin_flx & FI_SHORT)) { |
| switch (p) |
| { |
| case IPPROTO_UDP : |
| ipf_pr_udp6(fin); |
| go = 0; |
| break; |
| |
| case IPPROTO_TCP : |
| ipf_pr_tcp6(fin); |
| go = 0; |
| break; |
| |
| case IPPROTO_ICMPV6 : |
| ipf_pr_icmp6(fin); |
| go = 0; |
| break; |
| |
| case IPPROTO_GRE : |
| ipf_pr_gre6(fin); |
| go = 0; |
| break; |
| |
| case IPPROTO_HOPOPTS : |
| p = ipf_pr_hopopts6(fin); |
| break; |
| |
| case IPPROTO_MOBILITY : |
| p = ipf_pr_mobility6(fin); |
| break; |
| |
| case IPPROTO_DSTOPTS : |
| p = ipf_pr_dstopts6(fin); |
| break; |
| |
| case IPPROTO_ROUTING : |
| p = ipf_pr_routing6(fin); |
| break; |
| |
| case IPPROTO_AH : |
| p = ipf_pr_ah6(fin); |
| break; |
| |
| case IPPROTO_ESP : |
| ipf_pr_esp6(fin); |
| go = 0; |
| break; |
| |
| case IPPROTO_IPV6 : |
| for (i = 0; ip6exthdr[i].ol_bit != 0; i++) |
| if (ip6exthdr[i].ol_val == p) { |
| fin->fin_flx |= ip6exthdr[i].ol_bit; |
| break; |
| } |
| go = 0; |
| break; |
| |
| case IPPROTO_NONE : |
| go = 0; |
| break; |
| |
| case IPPROTO_FRAGMENT : |
| p = ipf_pr_fragment6(fin); |
| /* |
| * Given that the only fragments we want to let through |
| * (where fin_off != 0) are those where the non-first |
| * fragments only have data, we can safely stop looking |
| * at headers if this is a non-leading fragment. |
| */ |
| if (fin->fin_off != 0) |
| go = 0; |
| break; |
| |
| default : |
| go = 0; |
| break; |
| } |
| hdrcount++; |
| |
| /* |
| * It is important to note that at this point, for the |
| * extension headers (go != 0), the entire header may not have |
| * been pulled up when the code gets to this point. This is |
| * only done for "go != 0" because the other header handlers |
| * will all pullup their complete header. The other indicator |
| * of an incomplete packet is that this was just an extension |
| * header. |
| */ |
| if ((go != 0) && (p != IPPROTO_NONE) && |
| (ipf_pr_pullup(fin, 0) == -1)) { |
| p = IPPROTO_NONE; |
| break; |
| } |
| } |
| |
| /* |
| * Some of the above functions, like ipf_pr_esp6(), can call ipf_pullup |
| * and destroy whatever packet was here. The caller of this function |
| * expects us to return if there is a problem with ipf_pullup. |
| */ |
| if (fin->fin_m == NULL) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_bad); |
| return; |
| } |
| |
| fi->fi_p = p; |
| |
| /* |
| * IPv6 fragment case 1 - see comment for ipf_pr_fragment6(). |
| * "go != 0" imples the above loop hasn't arrived at a layer 4 header. |
| */ |
| if ((go != 0) && (fin->fin_flx & FI_FRAG) && (fin->fin_off == 0)) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| fin->fin_flx |= FI_BAD; |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_badfrag); |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_bad); |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_ipv6exthdr */ |
| /* Returns: struct ip6_ext * - pointer to the start of the next header */ |
| /* or NULL if there is a prolblem. */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* multiple(I) - flag indicating yes/no if multiple occurances */ |
| /* of this extension header are allowed. */ |
| /* proto(I) - protocol number for this extension header */ |
| /* */ |
| /* IPv6 Only */ |
| /* This function embodies a number of common checks that all IPv6 extension */ |
| /* headers must be subjected to. For example, making sure the packet is */ |
| /* big enough for it to be in, checking if it is repeated and setting a */ |
| /* flag to indicate its presence. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE struct ip6_ext * |
| ipf_pr_ipv6exthdr(fin, multiple, proto) |
| fr_info_t *fin; |
| int multiple, proto; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| struct ip6_ext *hdr; |
| u_short shift; |
| int i; |
| |
| fin->fin_flx |= FI_V6EXTHDR; |
| |
| /* 8 is default length of extension hdr */ |
| if ((fin->fin_dlen - 8) < 0) { |
| fin->fin_flx |= FI_SHORT; |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_ext_short); |
| return NULL; |
| } |
| |
| if (ipf_pr_pullup(fin, 8) == -1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_ext_pullup); |
| return NULL; |
| } |
| |
| hdr = fin->fin_dp; |
| switch (proto) |
| { |
| case IPPROTO_FRAGMENT : |
| shift = 8; |
| break; |
| default : |
| shift = 8 + (hdr->ip6e_len << 3); |
| break; |
| } |
| |
| if (shift > fin->fin_dlen) { /* Nasty extension header length? */ |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_ext_hlen); |
| fin->fin_flx |= FI_BAD; |
| return NULL; |
| } |
| |
| fin->fin_dp = (char *)fin->fin_dp + shift; |
| fin->fin_dlen -= shift; |
| |
| /* |
| * If we have seen a fragment header, do not set any flags to indicate |
| * the presence of this extension header as it has no impact on the |
| * end result until after it has been defragmented. |
| */ |
| if (fin->fin_flx & FI_FRAG) |
| return hdr; |
| |
| for (i = 0; ip6exthdr[i].ol_bit != 0; i++) |
| if (ip6exthdr[i].ol_val == proto) { |
| /* |
| * Most IPv6 extension headers are only allowed once. |
| */ |
| if ((multiple == 0) && |
| ((fin->fin_optmsk & ip6exthdr[i].ol_bit) != 0)) |
| fin->fin_flx |= FI_BAD; |
| else |
| fin->fin_optmsk |= ip6exthdr[i].ol_bit; |
| break; |
| } |
| |
| return hdr; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_hopopts6 */ |
| /* Returns: int - value of the next header or IPPROTO_NONE if error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* This is function checks pending hop by hop options extension header */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_hopopts6(fin) |
| fr_info_t *fin; |
| { |
| struct ip6_ext *hdr; |
| |
| hdr = ipf_pr_ipv6exthdr(fin, 0, IPPROTO_HOPOPTS); |
| if (hdr == NULL) |
| return IPPROTO_NONE; |
| return hdr->ip6e_nxt; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_mobility6 */ |
| /* Returns: int - value of the next header or IPPROTO_NONE if error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* This is function checks the IPv6 mobility extension header */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_mobility6(fin) |
| fr_info_t *fin; |
| { |
| struct ip6_ext *hdr; |
| |
| hdr = ipf_pr_ipv6exthdr(fin, 0, IPPROTO_MOBILITY); |
| if (hdr == NULL) |
| return IPPROTO_NONE; |
| return hdr->ip6e_nxt; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_routing6 */ |
| /* Returns: int - value of the next header or IPPROTO_NONE if error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* This is function checks pending routing extension header */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_routing6(fin) |
| fr_info_t *fin; |
| { |
| struct ip6_routing *hdr; |
| |
| hdr = (struct ip6_routing *)ipf_pr_ipv6exthdr(fin, 0, IPPROTO_ROUTING); |
| if (hdr == NULL) |
| return IPPROTO_NONE; |
| |
| switch (hdr->ip6r_type) |
| { |
| case 0 : |
| /* |
| * Nasty extension header length? |
| */ |
| if (((hdr->ip6r_len >> 1) < hdr->ip6r_segleft) || |
| (hdr->ip6r_segleft && (hdr->ip6r_len & 1))) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| fin->fin_flx |= FI_BAD; |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_rh_bad); |
| return IPPROTO_NONE; |
| } |
| break; |
| |
| default : |
| break; |
| } |
| |
| return hdr->ip6r_nxt; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_fragment6 */ |
| /* Returns: int - value of the next header or IPPROTO_NONE if error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* Examine the IPv6 fragment header and extract fragment offset information.*/ |
| /* */ |
| /* Fragments in IPv6 are extraordinarily difficult to deal with - much more */ |
| /* so than in IPv4. There are 5 cases of fragments with IPv6 that all */ |
| /* packets with a fragment header can fit into. They are as follows: */ |
| /* */ |
| /* 1. [IPv6][0-n EH][FH][0-n EH] (no L4HDR present) */ |
| /* 2. [IPV6][0-n EH][FH][0-n EH][L4HDR part] (short) */ |
| /* 3. [IPV6][0-n EH][FH][L4HDR part][0-n data] (short) */ |
| /* 4. [IPV6][0-n EH][FH][0-n EH][L4HDR][0-n data] */ |
| /* 5. [IPV6][0-n EH][FH][data] */ |
| /* */ |
| /* IPV6 = IPv6 header, FH = Fragment Header, */ |
| /* 0-n EH = 0 or more extension headers, 0-n data = 0 or more bytes of data */ |
| /* */ |
| /* Packets that match 1, 2, 3 will be dropped as the only reasonable */ |
| /* scenario in which they happen is in extreme circumstances that are most */ |
| /* likely to be an indication of an attack rather than normal traffic. */ |
| /* A type 3 packet may be sent by an attacked after a type 4 packet. There */ |
| /* are two rules that can be used to guard against type 3 packets: L4 */ |
| /* headers must always be in a packet that has the offset field set to 0 */ |
| /* and no packet is allowed to overlay that where offset = 0. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_fragment6(fin) |
| fr_info_t *fin; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| struct ip6_frag *frag; |
| |
| fin->fin_flx |= FI_FRAG; |
| |
| frag = (struct ip6_frag *)ipf_pr_ipv6exthdr(fin, 0, IPPROTO_FRAGMENT); |
| if (frag == NULL) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_frag_bad); |
| return IPPROTO_NONE; |
| } |
| |
| if ((frag->ip6f_offlg & IP6F_MORE_FRAG) != 0) { |
| /* |
| * Any fragment that isn't the last fragment must have its |
| * length as a multiple of 8. |
| */ |
| if ((fin->fin_plen & 7) != 0) |
| fin->fin_flx |= FI_BAD; |
| } |
| |
| fin->fin_fraghdr = frag; |
| fin->fin_id = frag->ip6f_ident; |
| fin->fin_off = ntohs(frag->ip6f_offlg & IP6F_OFF_MASK); |
| if (fin->fin_off != 0) |
| fin->fin_flx |= FI_FRAGBODY; |
| |
| /* |
| * Jumbograms aren't handled, so the max. length is 64k |
| */ |
| if ((fin->fin_off << 3) + fin->fin_dlen > 65535) |
| fin->fin_flx |= FI_BAD; |
| |
| /* |
| * We don't know where the transport layer header (or whatever is next |
| * is), as it could be behind destination options (amongst others) so |
| * return the fragment header as the type of packet this is. Note that |
| * this effectively disables the fragment cache for > 1 protocol at a |
| * time. |
| */ |
| return frag->ip6f_nxt; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_dstopts6 */ |
| /* Returns: int - value of the next header or IPPROTO_NONE if error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* This is function checks pending destination options extension header */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_dstopts6(fin) |
| fr_info_t *fin; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| struct ip6_ext *hdr; |
| |
| hdr = ipf_pr_ipv6exthdr(fin, 0, IPPROTO_DSTOPTS); |
| if (hdr == NULL) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_dst_bad); |
| return IPPROTO_NONE; |
| } |
| return hdr->ip6e_nxt; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_icmp6 */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* This routine is mainly concerned with determining the minimum valid size */ |
| /* for an ICMPv6 packet. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_icmp6(fin) |
| fr_info_t *fin; |
| { |
| int minicmpsz = sizeof(struct icmp6_hdr); |
| struct icmp6_hdr *icmp6; |
| |
| if (ipf_pr_pullup(fin, ICMP6ERR_MINPKTLEN - sizeof(ip6_t)) == -1) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_icmp6_pullup); |
| return; |
| } |
| |
| if (fin->fin_dlen > 1) { |
| ip6_t *ip6; |
| |
| icmp6 = fin->fin_dp; |
| |
| fin->fin_data[0] = *(u_short *)icmp6; |
| |
| if ((icmp6->icmp6_type & ICMP6_INFOMSG_MASK) != 0) |
| fin->fin_flx |= FI_ICMPQUERY; |
| |
| switch (icmp6->icmp6_type) |
| { |
| case ICMP6_ECHO_REPLY : |
| case ICMP6_ECHO_REQUEST : |
| if (fin->fin_dlen >= 6) |
| fin->fin_data[1] = icmp6->icmp6_id; |
| minicmpsz = ICMP6ERR_MINPKTLEN - sizeof(ip6_t); |
| break; |
| |
| case ICMP6_DST_UNREACH : |
| case ICMP6_PACKET_TOO_BIG : |
| case ICMP6_TIME_EXCEEDED : |
| case ICMP6_PARAM_PROB : |
| fin->fin_flx |= FI_ICMPERR; |
| minicmpsz = ICMP6ERR_IPICMPHLEN - sizeof(ip6_t); |
| if (fin->fin_plen < ICMP6ERR_IPICMPHLEN) |
| break; |
| |
| if (M_LEN(fin->fin_m) < fin->fin_plen) { |
| if (ipf_coalesce(fin) != 1) |
| return; |
| } |
| |
| if (ipf_pr_pullup(fin, ICMP6ERR_MINPKTLEN) == -1) |
| return; |
| |
| /* |
| * If the destination of this packet doesn't match the |
| * source of the original packet then this packet is |
| * not correct. |
| */ |
| icmp6 = fin->fin_dp; |
| ip6 = (ip6_t *)((char *)icmp6 + ICMPERR_ICMPHLEN); |
| if (IP6_NEQ(&fin->fin_fi.fi_dst, |
| (i6addr_t *)&ip6->ip6_src)) |
| fin->fin_flx |= FI_BAD; |
| |
| break; |
| default : |
| break; |
| } |
| } |
| |
| ipf_pr_short6(fin, minicmpsz); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_udp6 */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* Analyse the packet for IPv6/UDP properties. */ |
| /* Is not expected to be called for fragmented packets. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_udp6(fin) |
| fr_info_t *fin; |
| { |
| |
| if (ipf_pr_udpcommon(fin) == 0) { |
| u_char p = fin->fin_p; |
| |
| fin->fin_p = IPPROTO_UDP; |
| ipf_checkv6sum(fin); |
| fin->fin_p = p; |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_tcp6 */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* Analyse the packet for IPv6/TCP properties. */ |
| /* Is not expected to be called for fragmented packets. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_tcp6(fin) |
| fr_info_t *fin; |
| { |
| |
| if (ipf_pr_tcpcommon(fin) == 0) { |
| u_char p = fin->fin_p; |
| |
| fin->fin_p = IPPROTO_UDP; |
| ipf_checkv6sum(fin); |
| fin->fin_p = p; |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_esp6 */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* Analyse the packet for ESP properties. */ |
| /* The minimum length is taken to be the SPI (32bits) plus a tail (32bits) */ |
| /* even though the newer ESP packets must also have a sequence number that */ |
| /* is 32bits as well, it is not possible(?) to determine the version from a */ |
| /* simple packet header. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_esp6(fin) |
| fr_info_t *fin; |
| { |
| |
| if ((fin->fin_off == 0) && (ipf_pr_pullup(fin, 8) == -1)) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_esp_pullup); |
| return; |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_ah6 */ |
| /* Returns: int - value of the next header or IPPROTO_NONE if error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv6 Only */ |
| /* Analyse the packet for AH properties. */ |
| /* The minimum length is taken to be the combination of all fields in the */ |
| /* header being present and no authentication data (null algorithm used.) */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_ah6(fin) |
| fr_info_t *fin; |
| { |
| authhdr_t *ah; |
| |
| fin->fin_flx |= FI_AH; |
| |
| ah = (authhdr_t *)ipf_pr_ipv6exthdr(fin, 0, IPPROTO_HOPOPTS); |
| if (ah == NULL) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_ah_bad); |
| return IPPROTO_NONE; |
| } |
| |
| ipf_pr_short6(fin, sizeof(*ah)); |
| |
| /* |
| * No need for another pullup, ipf_pr_ipv6exthdr() will pullup |
| * enough data to satisfy ah_next (the very first one.) |
| */ |
| return ah->ah_next; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_gre6 */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* Analyse the packet for GRE properties. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_gre6(fin) |
| fr_info_t *fin; |
| { |
| grehdr_t *gre; |
| |
| if (ipf_pr_pullup(fin, sizeof(grehdr_t)) == -1) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| LBUMP(ipf_stats[fin->fin_out].fr_v6_gre_pullup); |
| return; |
| } |
| |
| gre = fin->fin_dp; |
| if (GRE_REV(gre->gr_flags) == 1) |
| fin->fin_data[0] = gre->gr_call; |
| } |
| #endif /* USE_INET6 */ |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_pullup */ |
| /* Returns: int - 0 == pullup succeeded, -1 == failure */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* plen(I) - length (excluding L3 header) to pullup */ |
| /* */ |
| /* Short inline function to cut down on code duplication to perform a call */ |
| /* to ipf_pullup to ensure there is the required amount of data, */ |
| /* consecutively in the packet buffer. */ |
| /* */ |
| /* This function pulls up 'extra' data at the location of fin_dp. fin_dp */ |
| /* points to the first byte after the complete layer 3 header, which will */ |
| /* include all of the known extension headers for IPv6 or options for IPv4. */ |
| /* */ |
| /* Since fr_pullup() expects the total length of bytes to be pulled up, it */ |
| /* is necessary to add those we can already assume to be pulled up (fin_dp */ |
| /* - fin_ip) to what is passed through. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_pr_pullup(fin, plen) |
| fr_info_t *fin; |
| int plen; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| if (fin->fin_m != NULL) { |
| if (fin->fin_dp != NULL) |
| plen += (char *)fin->fin_dp - |
| ((char *)fin->fin_ip + fin->fin_hlen); |
| plen += fin->fin_hlen; |
| if (M_LEN(fin->fin_m) < plen) { |
| #if defined(_KERNEL) |
| if (ipf_pullup(fin->fin_m, fin, plen) == NULL) { |
| LBUMP(ipf_stats[fin->fin_out].fr_pull[1]); |
| return -1; |
| } |
| #else |
| LBUMP(ipf_stats[fin->fin_out].fr_pull[1]); |
| /* |
| * Fake ipf_pullup failing |
| */ |
| *fin->fin_mp = NULL; |
| fin->fin_m = NULL; |
| fin->fin_ip = NULL; |
| return -1; |
| #endif |
| } |
| LBUMP(ipf_stats[fin->fin_out].fr_pull[0]); |
| } |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_short */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* xmin(I) - minimum header size */ |
| /* */ |
| /* Check if a packet is "short" as defined by xmin. The rule we are */ |
| /* applying here is that the packet must not be fragmented within the layer */ |
| /* 4 header. That is, it must not be a fragment that has its offset set to */ |
| /* start within the layer 4 header (hdrmin) or if it is at offset 0, the */ |
| /* entire layer 4 header must be present (min). */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_short(fin, xmin) |
| fr_info_t *fin; |
| int xmin; |
| { |
| |
| if (fin->fin_off == 0) { |
| if (fin->fin_dlen < xmin) |
| fin->fin_flx |= FI_SHORT; |
| } else if (fin->fin_off < xmin) { |
| fin->fin_flx |= FI_SHORT; |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_icmp */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv4 Only */ |
| /* Do a sanity check on the packet for ICMP (v4). In nearly all cases, */ |
| /* except extrememly bad packets, both type and code will be present. */ |
| /* The expected minimum size of an ICMP packet is very much dependent on */ |
| /* the type of it. */ |
| /* */ |
| /* XXX - other ICMP sanity checks? */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_icmp(fin) |
| fr_info_t *fin; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| int minicmpsz = sizeof(struct icmp); |
| icmphdr_t *icmp; |
| ip_t *oip; |
| |
| ipf_pr_short(fin, ICMPERR_ICMPHLEN); |
| |
| if (fin->fin_off != 0) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_icmp_frag); |
| return; |
| } |
| |
| if (ipf_pr_pullup(fin, ICMPERR_ICMPHLEN) == -1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_icmp_pullup); |
| return; |
| } |
| |
| icmp = fin->fin_dp; |
| |
| fin->fin_data[0] = *(u_short *)icmp; |
| fin->fin_data[1] = icmp->icmp_id; |
| |
| switch (icmp->icmp_type) |
| { |
| case ICMP_ECHOREPLY : |
| case ICMP_ECHO : |
| /* Router discovery messaes - RFC 1256 */ |
| case ICMP_ROUTERADVERT : |
| case ICMP_ROUTERSOLICIT : |
| fin->fin_flx |= FI_ICMPQUERY; |
| minicmpsz = ICMP_MINLEN; |
| break; |
| /* |
| * type(1) + code(1) + cksum(2) + id(2) seq(2) + |
| * 3 * timestamp(3 * 4) |
| */ |
| case ICMP_TSTAMP : |
| case ICMP_TSTAMPREPLY : |
| fin->fin_flx |= FI_ICMPQUERY; |
| minicmpsz = 20; |
| break; |
| /* |
| * type(1) + code(1) + cksum(2) + id(2) seq(2) + |
| * mask(4) |
| */ |
| case ICMP_MASKREQ : |
| case ICMP_MASKREPLY : |
| fin->fin_flx |= FI_ICMPQUERY; |
| minicmpsz = 12; |
| break; |
| /* |
| * type(1) + code(1) + cksum(2) + id(2) seq(2) + ip(20+) |
| */ |
| case ICMP_UNREACH : |
| #ifdef icmp_nextmtu |
| if (icmp->icmp_code == ICMP_UNREACH_NEEDFRAG) { |
| if (icmp->icmp_nextmtu < softc->ipf_icmpminfragmtu) |
| fin->fin_flx |= FI_BAD; |
| } |
| #endif |
| case ICMP_SOURCEQUENCH : |
| case ICMP_REDIRECT : |
| case ICMP_TIMXCEED : |
| case ICMP_PARAMPROB : |
| fin->fin_flx |= FI_ICMPERR; |
| if (ipf_coalesce(fin) != 1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_icmp_coalesce); |
| return; |
| } |
| |
| /* |
| * ICMP error packets should not be generated for IP |
| * packets that are a fragment that isn't the first |
| * fragment. |
| */ |
| oip = (ip_t *)((char *)fin->fin_dp + ICMPERR_ICMPHLEN); |
| if ((ntohs(oip->ip_off) & IP_OFFMASK) != 0) |
| fin->fin_flx |= FI_BAD; |
| |
| /* |
| * If the destination of this packet doesn't match the |
| * source of the original packet then this packet is |
| * not correct. |
| */ |
| if (oip->ip_src.s_addr != fin->fin_daddr) |
| fin->fin_flx |= FI_BAD; |
| break; |
| default : |
| break; |
| } |
| |
| ipf_pr_short(fin, minicmpsz); |
| |
| ipf_checkv4sum(fin); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_tcpcommon */ |
| /* Returns: int - 0 = header ok, 1 = bad packet, -1 = buffer error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* TCP header sanity checking. Look for bad combinations of TCP flags, */ |
| /* and make some checks with how they interact with other fields. */ |
| /* If compiled with IPFILTER_CKSUM, check to see if the TCP checksum is */ |
| /* valid and mark the packet as bad if not. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_tcpcommon(fin) |
| fr_info_t *fin; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| int flags, tlen; |
| tcphdr_t *tcp; |
| |
| fin->fin_flx |= FI_TCPUDP; |
| if (fin->fin_off != 0) { |
| LBUMP(ipf_stats[fin->fin_out].fr_tcp_frag); |
| return 0; |
| } |
| |
| if (ipf_pr_pullup(fin, sizeof(*tcp)) == -1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_tcp_pullup); |
| return -1; |
| } |
| |
| tcp = fin->fin_dp; |
| if (fin->fin_dlen > 3) { |
| fin->fin_sport = ntohs(tcp->th_sport); |
| fin->fin_dport = ntohs(tcp->th_dport); |
| } |
| |
| if ((fin->fin_flx & FI_SHORT) != 0) { |
| LBUMP(ipf_stats[fin->fin_out].fr_tcp_short); |
| return 1; |
| } |
| |
| /* |
| * Use of the TCP data offset *must* result in a value that is at |
| * least the same size as the TCP header. |
| */ |
| tlen = TCP_OFF(tcp) << 2; |
| if (tlen < sizeof(tcphdr_t)) { |
| LBUMP(ipf_stats[fin->fin_out].fr_tcp_small); |
| fin->fin_flx |= FI_BAD; |
| return 1; |
| } |
| |
| flags = tcp->th_flags; |
| fin->fin_tcpf = tcp->th_flags; |
| |
| /* |
| * If the urgent flag is set, then the urgent pointer must |
| * also be set and vice versa. Good TCP packets do not have |
| * just one of these set. |
| */ |
| if ((flags & TH_URG) != 0 && (tcp->th_urp == 0)) { |
| fin->fin_flx |= FI_BAD; |
| #if 0 |
| } else if ((flags & TH_URG) == 0 && (tcp->th_urp != 0)) { |
| /* |
| * Ignore this case (#if 0) as it shows up in "real" |
| * traffic with bogus values in the urgent pointer field. |
| */ |
| fin->fin_flx |= FI_BAD; |
| #endif |
| } else if (((flags & (TH_SYN|TH_FIN)) != 0) && |
| ((flags & (TH_RST|TH_ACK)) == TH_RST)) { |
| /* TH_FIN|TH_RST|TH_ACK seems to appear "naturally" */ |
| fin->fin_flx |= FI_BAD; |
| #if 1 |
| } else if (((flags & TH_SYN) != 0) && |
| ((flags & (TH_URG|TH_PUSH)) != 0)) { |
| /* |
| * SYN with URG and PUSH set is not for normal TCP but it is |
| * possible(?) with T/TCP...but who uses T/TCP? |
| */ |
| fin->fin_flx |= FI_BAD; |
| #endif |
| } else if (!(flags & TH_ACK)) { |
| /* |
| * If the ack bit isn't set, then either the SYN or |
| * RST bit must be set. If the SYN bit is set, then |
| * we expect the ACK field to be 0. If the ACK is |
| * not set and if URG, PSH or FIN are set, consdier |
| * that to indicate a bad TCP packet. |
| */ |
| if ((flags == TH_SYN) && (tcp->th_ack != 0)) { |
| /* |
| * Cisco PIX sets the ACK field to a random value. |
| * In light of this, do not set FI_BAD until a patch |
| * is available from Cisco to ensure that |
| * interoperability between existing systems is |
| * achieved. |
| */ |
| /*fin->fin_flx |= FI_BAD*/; |
| } else if (!(flags & (TH_RST|TH_SYN))) { |
| fin->fin_flx |= FI_BAD; |
| } else if ((flags & (TH_URG|TH_PUSH|TH_FIN)) != 0) { |
| fin->fin_flx |= FI_BAD; |
| } |
| } |
| if (fin->fin_flx & FI_BAD) { |
| LBUMP(ipf_stats[fin->fin_out].fr_tcp_bad_flags); |
| return 1; |
| } |
| |
| /* |
| * At this point, it's not exactly clear what is to be gained by |
| * marking up which TCP options are and are not present. The one we |
| * are most interested in is the TCP window scale. This is only in |
| * a SYN packet [RFC1323] so we don't need this here...? |
| * Now if we were to analyse the header for passive fingerprinting, |
| * then that might add some weight to adding this... |
| */ |
| if (tlen == sizeof(tcphdr_t)) { |
| return 0; |
| } |
| |
| if (ipf_pr_pullup(fin, tlen) == -1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_tcp_pullup); |
| return -1; |
| } |
| |
| #if 0 |
| tcp = fin->fin_dp; |
| ip = fin->fin_ip; |
| s = (u_char *)(tcp + 1); |
| off = IP_HL(ip) << 2; |
| # ifdef _KERNEL |
| if (fin->fin_mp != NULL) { |
| mb_t *m = *fin->fin_mp; |
| |
| if (off + tlen > M_LEN(m)) |
| return; |
| } |
| # endif |
| for (tlen -= (int)sizeof(*tcp); tlen > 0; ) { |
| opt = *s; |
| if (opt == '\0') |
| break; |
| else if (opt == TCPOPT_NOP) |
| ol = 1; |
| else { |
| if (tlen < 2) |
| break; |
| ol = (int)*(s + 1); |
| if (ol < 2 || ol > tlen) |
| break; |
| } |
| |
| for (i = 9, mv = 4; mv >= 0; ) { |
| op = ipopts + i; |
| if (opt == (u_char)op->ol_val) { |
| optmsk |= op->ol_bit; |
| break; |
| } |
| } |
| tlen -= ol; |
| s += ol; |
| } |
| #endif /* 0 */ |
| |
| return 0; |
| } |
| |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_udpcommon */ |
| /* Returns: int - 0 = header ok, 1 = bad packet */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* Extract the UDP source and destination ports, if present. If compiled */ |
| /* with IPFILTER_CKSUM, check to see if the UDP checksum is valid. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_udpcommon(fin) |
| fr_info_t *fin; |
| { |
| udphdr_t *udp; |
| |
| fin->fin_flx |= FI_TCPUDP; |
| |
| if (!fin->fin_off && (fin->fin_dlen > 3)) { |
| if (ipf_pr_pullup(fin, sizeof(*udp)) == -1) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| fin->fin_flx |= FI_SHORT; |
| LBUMP(ipf_stats[fin->fin_out].fr_udp_pullup); |
| return 1; |
| } |
| |
| udp = fin->fin_dp; |
| |
| fin->fin_sport = ntohs(udp->uh_sport); |
| fin->fin_dport = ntohs(udp->uh_dport); |
| } |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_tcp */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv4 Only */ |
| /* Analyse the packet for IPv4/TCP properties. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_tcp(fin) |
| fr_info_t *fin; |
| { |
| |
| ipf_pr_short(fin, sizeof(tcphdr_t)); |
| |
| if (ipf_pr_tcpcommon(fin) == 0) |
| ipf_checkv4sum(fin); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_udp */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv4 Only */ |
| /* Analyse the packet for IPv4/UDP properties. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_udp(fin) |
| fr_info_t *fin; |
| { |
| |
| ipf_pr_short(fin, sizeof(udphdr_t)); |
| |
| if (ipf_pr_udpcommon(fin) == 0) |
| ipf_checkv4sum(fin); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_esp */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* Analyse the packet for ESP properties. */ |
| /* The minimum length is taken to be the SPI (32bits) plus a tail (32bits) */ |
| /* even though the newer ESP packets must also have a sequence number that */ |
| /* is 32bits as well, it is not possible(?) to determine the version from a */ |
| /* simple packet header. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_esp(fin) |
| fr_info_t *fin; |
| { |
| |
| if (fin->fin_off == 0) { |
| ipf_pr_short(fin, 8); |
| if (ipf_pr_pullup(fin, 8) == -1) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_esp_pullup); |
| } |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_ah */ |
| /* Returns: int - value of the next header or IPPROTO_NONE if error */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* Analyse the packet for AH properties. */ |
| /* The minimum length is taken to be the combination of all fields in the */ |
| /* header being present and no authentication data (null algorithm used.) */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_pr_ah(fin) |
| fr_info_t *fin; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| authhdr_t *ah; |
| int len; |
| |
| fin->fin_flx |= FI_AH; |
| ipf_pr_short(fin, sizeof(*ah)); |
| |
| if (((fin->fin_flx & FI_SHORT) != 0) || (fin->fin_off != 0)) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_ah_bad); |
| return IPPROTO_NONE; |
| } |
| |
| if (ipf_pr_pullup(fin, sizeof(*ah)) == -1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_ah_pullup); |
| return IPPROTO_NONE; |
| } |
| |
| ah = (authhdr_t *)fin->fin_dp; |
| |
| len = (ah->ah_plen + 2) << 2; |
| ipf_pr_short(fin, len); |
| if (ipf_pr_pullup(fin, len) == -1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_ah_pullup); |
| return IPPROTO_NONE; |
| } |
| |
| /* |
| * Adjust fin_dp and fin_dlen for skipping over the authentication |
| * header. |
| */ |
| fin->fin_dp = (char *)fin->fin_dp + len; |
| fin->fin_dlen -= len; |
| return ah->ah_next; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_gre */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* Analyse the packet for GRE properties. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_gre(fin) |
| fr_info_t *fin; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| grehdr_t *gre; |
| |
| ipf_pr_short(fin, sizeof(grehdr_t)); |
| |
| if (fin->fin_off != 0) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_gre_frag); |
| return; |
| } |
| |
| if (ipf_pr_pullup(fin, sizeof(grehdr_t)) == -1) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_gre_pullup); |
| return; |
| } |
| |
| gre = fin->fin_dp; |
| if (GRE_REV(gre->gr_flags) == 1) |
| fin->fin_data[0] = gre->gr_call; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_pr_ipv4hdr */ |
| /* Returns: void */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* IPv4 Only */ |
| /* Analyze the IPv4 header and set fields in the fr_info_t structure. */ |
| /* Check all options present and flag their presence if any exist. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE void |
| ipf_pr_ipv4hdr(fin) |
| fr_info_t *fin; |
| { |
| u_short optmsk = 0, secmsk = 0, auth = 0; |
| int hlen, ol, mv, p, i; |
| const struct optlist *op; |
| u_char *s, opt; |
| u_short off; |
| fr_ip_t *fi; |
| ip_t *ip; |
| |
| fi = &fin->fin_fi; |
| hlen = fin->fin_hlen; |
| |
| ip = fin->fin_ip; |
| p = ip->ip_p; |
| fi->fi_p = p; |
| fin->fin_crc = p; |
| fi->fi_tos = ip->ip_tos; |
| fin->fin_id = ip->ip_id; |
| off = ntohs(ip->ip_off); |
| |
| /* Get both TTL and protocol */ |
| fi->fi_p = ip->ip_p; |
| fi->fi_ttl = ip->ip_ttl; |
| |
| /* Zero out bits not used in IPv6 address */ |
| fi->fi_src.i6[1] = 0; |
| fi->fi_src.i6[2] = 0; |
| fi->fi_src.i6[3] = 0; |
| fi->fi_dst.i6[1] = 0; |
| fi->fi_dst.i6[2] = 0; |
| fi->fi_dst.i6[3] = 0; |
| |
| fi->fi_saddr = ip->ip_src.s_addr; |
| fin->fin_crc += fi->fi_saddr; |
| fi->fi_daddr = ip->ip_dst.s_addr; |
| fin->fin_crc += fi->fi_daddr; |
| |
| /* |
| * set packet attribute flags based on the offset and |
| * calculate the byte offset that it represents. |
| */ |
| off &= IP_MF|IP_OFFMASK; |
| if (off != 0) { |
| int morefrag = off & IP_MF; |
| |
| fi->fi_flx |= FI_FRAG; |
| off &= IP_OFFMASK; |
| if (off != 0) { |
| fin->fin_flx |= FI_FRAGBODY; |
| off <<= 3; |
| if ((off + fin->fin_dlen > 65535) || |
| (fin->fin_dlen == 0) || |
| ((morefrag != 0) && ((fin->fin_dlen & 7) != 0))) { |
| /* |
| * The length of the packet, starting at its |
| * offset cannot exceed 65535 (0xffff) as the |
| * length of an IP packet is only 16 bits. |
| * |
| * Any fragment that isn't the last fragment |
| * must have a length greater than 0 and it |
| * must be an even multiple of 8. |
| */ |
| fi->fi_flx |= FI_BAD; |
| } |
| } |
| } |
| fin->fin_off = off; |
| |
| /* |
| * Call per-protocol setup and checking |
| */ |
| if (p == IPPROTO_AH) { |
| /* |
| * Treat AH differently because we expect there to be another |
| * layer 4 header after it. |
| */ |
| p = ipf_pr_ah(fin); |
| } |
| |
| switch (p) |
| { |
| case IPPROTO_UDP : |
| ipf_pr_udp(fin); |
| break; |
| case IPPROTO_TCP : |
| ipf_pr_tcp(fin); |
| break; |
| case IPPROTO_ICMP : |
| ipf_pr_icmp(fin); |
| break; |
| case IPPROTO_ESP : |
| ipf_pr_esp(fin); |
| break; |
| case IPPROTO_GRE : |
| ipf_pr_gre(fin); |
| break; |
| } |
| |
| ip = fin->fin_ip; |
| if (ip == NULL) |
| return; |
| |
| /* |
| * If it is a standard IP header (no options), set the flag fields |
| * which relate to options to 0. |
| */ |
| if (hlen == sizeof(*ip)) { |
| fi->fi_optmsk = 0; |
| fi->fi_secmsk = 0; |
| fi->fi_auth = 0; |
| return; |
| } |
| |
| /* |
| * So the IP header has some IP options attached. Walk the entire |
| * list of options present with this packet and set flags to indicate |
| * which ones are here and which ones are not. For the somewhat out |
| * of date and obscure security classification options, set a flag to |
| * represent which classification is present. |
| */ |
| fi->fi_flx |= FI_OPTIONS; |
| |
| for (s = (u_char *)(ip + 1), hlen -= (int)sizeof(*ip); hlen > 0; ) { |
| opt = *s; |
| if (opt == '\0') |
| break; |
| else if (opt == IPOPT_NOP) |
| ol = 1; |
| else { |
| if (hlen < 2) |
| break; |
| ol = (int)*(s + 1); |
| if (ol < 2 || ol > hlen) |
| break; |
| } |
| for (i = 9, mv = 4; mv >= 0; ) { |
| op = ipopts + i; |
| |
| if ((opt == (u_char)op->ol_val) && (ol > 4)) { |
| u_32_t doi; |
| |
| switch (opt) |
| { |
| case IPOPT_SECURITY : |
| if (optmsk & op->ol_bit) { |
| fin->fin_flx |= FI_BAD; |
| } else { |
| doi = ipf_checkripso(s); |
| secmsk = doi >> 16; |
| auth = doi & 0xffff; |
| } |
| break; |
| |
| case IPOPT_CIPSO : |
| |
| if (optmsk & op->ol_bit) { |
| fin->fin_flx |= FI_BAD; |
| } else { |
| doi = ipf_checkcipso(fin, |
| s, ol); |
| secmsk = doi >> 16; |
| auth = doi & 0xffff; |
| } |
| break; |
| } |
| optmsk |= op->ol_bit; |
| } |
| |
| if (opt < op->ol_val) |
| i -= mv; |
| else |
| i += mv; |
| mv--; |
| } |
| hlen -= ol; |
| s += ol; |
| } |
| |
| /* |
| * |
| */ |
| if (auth && !(auth & 0x0100)) |
| auth &= 0xff00; |
| fi->fi_optmsk = optmsk; |
| fi->fi_secmsk = secmsk; |
| fi->fi_auth = auth; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_checkripso */ |
| /* Returns: void */ |
| /* Parameters: s(I) - pointer to start of RIPSO option */ |
| /* */ |
| /* ------------------------------------------------------------------------ */ |
| static u_32_t |
| ipf_checkripso(s) |
| u_char *s; |
| { |
| const struct optlist *sp; |
| u_short secmsk = 0, auth = 0; |
| u_char sec; |
| int j, m; |
| |
| sec = *(s + 2); /* classification */ |
| for (j = 3, m = 2; m >= 0; ) { |
| sp = secopt + j; |
| if (sec == sp->ol_val) { |
| secmsk |= sp->ol_bit; |
| auth = *(s + 3); |
| auth *= 256; |
| auth += *(s + 4); |
| break; |
| } |
| if (sec < sp->ol_val) |
| j -= m; |
| else |
| j += m; |
| m--; |
| } |
| |
| return (secmsk << 16) | auth; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_checkcipso */ |
| /* Returns: u_32_t - 0 = failure, else the doi from the header */ |
| /* Parameters: fin(IO) - pointer to packet information */ |
| /* s(I) - pointer to start of CIPSO option */ |
| /* ol(I) - length of CIPSO option field */ |
| /* */ |
| /* This function returns the domain of integrity (DOI) field from the CIPSO */ |
| /* header and returns that whilst also storing the highest sensitivity */ |
| /* value found in the fr_info_t structure. */ |
| /* */ |
| /* No attempt is made to extract the category bitmaps as these are defined */ |
| /* by the user (rather than the protocol) and can be rather numerous on the */ |
| /* end nodes. */ |
| /* ------------------------------------------------------------------------ */ |
| static u_32_t |
| ipf_checkcipso(fin, s, ol) |
| fr_info_t *fin; |
| u_char *s; |
| int ol; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| fr_ip_t *fi; |
| u_32_t doi; |
| u_char *t, tag, tlen, sensitivity; |
| int len; |
| |
| if (ol < 6 || ol > 40) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_cipso_bad); |
| fin->fin_flx |= FI_BAD; |
| return 0; |
| } |
| |
| fi = &fin->fin_fi; |
| fi->fi_sensitivity = 0; |
| /* |
| * The DOI field MUST be there. |
| */ |
| bcopy(s + 2, &doi, sizeof(doi)); |
| |
| t = (u_char *)s + 6; |
| for (len = ol - 6; len >= 2; len -= tlen, t+= tlen) { |
| tag = *t; |
| tlen = *(t + 1); |
| if (tlen > len || tlen < 4 || tlen > 34) { |
| LBUMP(ipf_stats[fin->fin_out].fr_v4_cipso_tlen); |
| fin->fin_flx |= FI_BAD; |
| return 0; |
| } |
| |
| sensitivity = 0; |
| /* |
| * Tag numbers 0, 1, 2, 5 are laid out in the CIPSO Internet |
| * draft (16 July 1992) that has expired. |
| */ |
| if (tag == 0) { |
| fin->fin_flx |= FI_BAD; |
| continue; |
| } else if (tag == 1) { |
| if (*(t + 2) != 0) { |
| fin->fin_flx |= FI_BAD; |
| continue; |
| } |
| sensitivity = *(t + 3); |
| /* Category bitmap for categories 0-239 */ |
| |
| } else if (tag == 4) { |
| if (*(t + 2) != 0) { |
| fin->fin_flx |= FI_BAD; |
| continue; |
| } |
| sensitivity = *(t + 3); |
| /* Enumerated categories, 16bits each, upto 15 */ |
| |
| } else if (tag == 5) { |
| if (*(t + 2) != 0) { |
| fin->fin_flx |= FI_BAD; |
| continue; |
| } |
| sensitivity = *(t + 3); |
| /* Range of categories (2*16bits), up to 7 pairs */ |
| |
| } else if (tag > 127) { |
| /* Custom defined DOI */ |
| ; |
| } else { |
| fin->fin_flx |= FI_BAD; |
| continue; |
| } |
| |
| if (sensitivity > fi->fi_sensitivity) |
| fi->fi_sensitivity = sensitivity; |
| } |
| |
| return doi; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_makefrip */ |
| /* Returns: void */ |
| /* Parameters: hlen(I) - length of IP packet header */ |
| /* ip(I) - pointer to the IP header */ |
| /* fin(IO) - pointer to packet information */ |
| /* */ |
| /* Compact the IP header into a structure which contains just the info. */ |
| /* which is useful for comparing IP headers with and store this information */ |
| /* in the fr_info_t structure pointer to by fin. At present, it is assumed */ |
| /* this function will be called with either an IPv4 or IPv6 packet. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_makefrip(hlen, ip, fin) |
| int hlen; |
| ip_t *ip; |
| fr_info_t *fin; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| int v; |
| |
| fin->fin_depth = 0; |
| fin->fin_hlen = (u_short)hlen; |
| fin->fin_ip = ip; |
| fin->fin_rule = 0xffffffff; |
| fin->fin_group[0] = -1; |
| fin->fin_group[1] = '\0'; |
| fin->fin_dp = (char *)ip + hlen; |
| |
| v = fin->fin_v; |
| if (v == 4) { |
| fin->fin_plen = ntohs(ip->ip_len); |
| fin->fin_dlen = fin->fin_plen - hlen; |
| ipf_pr_ipv4hdr(fin); |
| #ifdef USE_INET6 |
| } else if (v == 6) { |
| fin->fin_plen = ntohs(((ip6_t *)ip)->ip6_plen); |
| fin->fin_dlen = fin->fin_plen; |
| fin->fin_plen += hlen; |
| |
| ipf_pr_ipv6hdr(fin); |
| #endif |
| } |
| if (fin->fin_ip == NULL) { |
| LBUMP(ipf_stats[fin->fin_out].fr_ip_freed); |
| return -1; |
| } |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_portcheck */ |
| /* Returns: int - 1 == port matched, 0 == port match failed */ |
| /* Parameters: frp(I) - pointer to port check `expression' */ |
| /* pop(I) - port number to evaluate */ |
| /* */ |
| /* Perform a comparison of a port number against some other(s), using a */ |
| /* structure with compare information stored in it. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_portcheck(frp, pop) |
| frpcmp_t *frp; |
| u_32_t pop; |
| { |
| int err = 1; |
| u_32_t po; |
| |
| po = frp->frp_port; |
| |
| /* |
| * Do opposite test to that required and continue if that succeeds. |
| */ |
| switch (frp->frp_cmp) |
| { |
| case FR_EQUAL : |
| if (pop != po) /* EQUAL */ |
| err = 0; |
| break; |
| case FR_NEQUAL : |
| if (pop == po) /* NOTEQUAL */ |
| err = 0; |
| break; |
| case FR_LESST : |
| if (pop >= po) /* LESSTHAN */ |
| err = 0; |
| break; |
| case FR_GREATERT : |
| if (pop <= po) /* GREATERTHAN */ |
| err = 0; |
| break; |
| case FR_LESSTE : |
| if (pop > po) /* LT or EQ */ |
| err = 0; |
| break; |
| case FR_GREATERTE : |
| if (pop < po) /* GT or EQ */ |
| err = 0; |
| break; |
| case FR_OUTRANGE : |
| if (pop >= po && pop <= frp->frp_top) /* Out of range */ |
| err = 0; |
| break; |
| case FR_INRANGE : |
| if (pop <= po || pop >= frp->frp_top) /* In range */ |
| err = 0; |
| break; |
| case FR_INCRANGE : |
| if (pop < po || pop > frp->frp_top) /* Inclusive range */ |
| err = 0; |
| break; |
| default : |
| break; |
| } |
| return err; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tcpudpchk */ |
| /* Returns: int - 1 == protocol matched, 0 == check failed */ |
| /* Parameters: fda(I) - pointer to packet information */ |
| /* ft(I) - pointer to structure with comparison data */ |
| /* */ |
| /* Compares the current pcket (assuming it is TCP/UDP) information with a */ |
| /* structure containing information that we want to match against. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_tcpudpchk(fi, ft) |
| fr_ip_t *fi; |
| frtuc_t *ft; |
| { |
| int err = 1; |
| |
| /* |
| * Both ports should *always* be in the first fragment. |
| * So far, I cannot find any cases where they can not be. |
| * |
| * compare destination ports |
| */ |
| if (ft->ftu_dcmp) |
| err = ipf_portcheck(&ft->ftu_dst, fi->fi_ports[1]); |
| |
| /* |
| * compare source ports |
| */ |
| if (err && ft->ftu_scmp) |
| err = ipf_portcheck(&ft->ftu_src, fi->fi_ports[0]); |
| |
| /* |
| * If we don't have all the TCP/UDP header, then how can we |
| * expect to do any sort of match on it ? If we were looking for |
| * TCP flags, then NO match. If not, then match (which should |
| * satisfy the "short" class too). |
| */ |
| if (err && (fi->fi_p == IPPROTO_TCP)) { |
| if (fi->fi_flx & FI_SHORT) |
| return !(ft->ftu_tcpf | ft->ftu_tcpfm); |
| /* |
| * Match the flags ? If not, abort this match. |
| */ |
| if (ft->ftu_tcpfm && |
| ft->ftu_tcpf != (fi->fi_tcpf & ft->ftu_tcpfm)) { |
| FR_DEBUG(("f. %#x & %#x != %#x\n", fi->fi_tcpf, |
| ft->ftu_tcpfm, ft->ftu_tcpf)); |
| err = 0; |
| } |
| } |
| return err; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_check_ipf */ |
| /* Returns: int - 0 == match, else no match */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* fr(I) - pointer to filter rule */ |
| /* portcmp(I) - flag indicating whether to attempt matching on */ |
| /* TCP/UDP port data. */ |
| /* */ |
| /* Check to see if a packet matches an IPFilter rule. Checks of addresses, */ |
| /* port numbers, etc, for "standard" IPFilter rules are all orchestrated in */ |
| /* this function. */ |
| /* ------------------------------------------------------------------------ */ |
| static INLINE int |
| ipf_check_ipf(fin, fr, portcmp) |
| fr_info_t *fin; |
| frentry_t *fr; |
| int portcmp; |
| { |
| u_32_t *ld, *lm, *lip; |
| fripf_t *fri; |
| fr_ip_t *fi; |
| int i; |
| |
| fi = &fin->fin_fi; |
| fri = fr->fr_ipf; |
| lip = (u_32_t *)fi; |
| lm = (u_32_t *)&fri->fri_mip; |
| ld = (u_32_t *)&fri->fri_ip; |
| |
| /* |
| * first 32 bits to check coversion: |
| * IP version, TOS, TTL, protocol |
| */ |
| i = ((*lip & *lm) != *ld); |
| FR_DEBUG(("0. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| if (i) |
| return 1; |
| |
| /* |
| * Next 32 bits is a constructed bitmask indicating which IP options |
| * are present (if any) in this packet. |
| */ |
| lip++, lm++, ld++; |
| i = ((*lip & *lm) != *ld); |
| FR_DEBUG(("1. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| if (i != 0) |
| return 1; |
| |
| lip++, lm++, ld++; |
| /* |
| * Unrolled loops (4 each, for 32 bits) for address checks. |
| */ |
| /* |
| * Check the source address. |
| */ |
| if (fr->fr_satype == FRI_LOOKUP) { |
| i = (*fr->fr_srcfunc)(fin->fin_main_soft, fr->fr_srcptr, |
| fi->fi_v, lip); |
| if (i == -1) |
| return 1; |
| lip += 3; |
| lm += 3; |
| ld += 3; |
| } else { |
| i = ((*lip & *lm) != *ld); |
| FR_DEBUG(("2a. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| if (fi->fi_v == 6) { |
| lip++, lm++, ld++; |
| i |= ((*lip & *lm) != *ld); |
| FR_DEBUG(("2b. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| lip++, lm++, ld++; |
| i |= ((*lip & *lm) != *ld); |
| FR_DEBUG(("2c. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| lip++, lm++, ld++; |
| i |= ((*lip & *lm) != *ld); |
| FR_DEBUG(("2d. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| } else { |
| lip += 3; |
| lm += 3; |
| ld += 3; |
| } |
| } |
| i ^= (fr->fr_flags & FR_NOTSRCIP) >> 6; |
| if (i != 0) |
| return 1; |
| |
| /* |
| * Check the destination address. |
| */ |
| lip++, lm++, ld++; |
| if (fr->fr_datype == FRI_LOOKUP) { |
| i = (*fr->fr_dstfunc)(fin->fin_main_soft, fr->fr_dstptr, |
| fi->fi_v, lip); |
| if (i == -1) |
| return 1; |
| lip += 3; |
| lm += 3; |
| ld += 3; |
| } else { |
| i = ((*lip & *lm) != *ld); |
| FR_DEBUG(("3a. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| if (fi->fi_v == 6) { |
| lip++, lm++, ld++; |
| i |= ((*lip & *lm) != *ld); |
| FR_DEBUG(("3b. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| lip++, lm++, ld++; |
| i |= ((*lip & *lm) != *ld); |
| FR_DEBUG(("3c. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| lip++, lm++, ld++; |
| i |= ((*lip & *lm) != *ld); |
| FR_DEBUG(("3d. %#08x & %#08x != %#08x\n", |
| ntohl(*lip), ntohl(*lm), ntohl(*ld))); |
| } else { |
| lip += 3; |
| lm += 3; |
| ld += 3; |
| } |
| } |
| i ^= (fr->fr_flags & FR_NOTDSTIP) >> 7; |
| if (i != 0) |
| return 1; |
| /* |
| * IP addresses matched. The next 32bits contains: |
| * mast of old IP header security & authentication bits. |
| */ |
| lip++, lm++, ld++; |
| i = (*ld - (*lip & *lm)); |
| FR_DEBUG(("4. %#08x & %#08x != %#08x\n", *lip, *lm, *ld)); |
| |
| /* |
| * Next we have 32 bits of packet flags. |
| */ |
| lip++, lm++, ld++; |
| i |= (*ld - (*lip & *lm)); |
| FR_DEBUG(("5. %#08x & %#08x != %#08x\n", *lip, *lm, *ld)); |
| |
| if (i == 0) { |
| /* |
| * If a fragment, then only the first has what we're |
| * looking for here... |
| */ |
| if (portcmp) { |
| if (!ipf_tcpudpchk(&fin->fin_fi, &fr->fr_tuc)) |
| i = 1; |
| } else { |
| if (fr->fr_dcmp || fr->fr_scmp || |
| fr->fr_tcpf || fr->fr_tcpfm) |
| i = 1; |
| if (fr->fr_icmpm || fr->fr_icmp) { |
| if (((fi->fi_p != IPPROTO_ICMP) && |
| (fi->fi_p != IPPROTO_ICMPV6)) || |
| fin->fin_off || (fin->fin_dlen < 2)) |
| i = 1; |
| else if ((fin->fin_data[0] & fr->fr_icmpm) != |
| fr->fr_icmp) { |
| FR_DEBUG(("i. %#x & %#x != %#x\n", |
| fin->fin_data[0], |
| fr->fr_icmpm, fr->fr_icmp)); |
| i = 1; |
| } |
| } |
| } |
| } |
| return i; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_scanlist */ |
| /* Returns: int - result flags of scanning filter list */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* pass(I) - default result to return for filtering */ |
| /* */ |
| /* Check the input/output list of rules for a match to the current packet. */ |
| /* If a match is found, the value of fr_flags from the rule becomes the */ |
| /* return value and fin->fin_fr points to the matched rule. */ |
| /* */ |
| /* This function may be called recusively upto 16 times (limit inbuilt.) */ |
| /* When unwinding, it should finish up with fin_depth as 0. */ |
| /* */ |
| /* Could be per interface, but this gets real nasty when you don't have, */ |
| /* or can't easily change, the kernel source code to . */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_scanlist(fin, pass) |
| fr_info_t *fin; |
| u_32_t pass; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| int rulen, portcmp, off, skip; |
| struct frentry *fr, *fnext; |
| u_32_t passt, passo; |
| |
| /* |
| * Do not allow nesting deeper than 16 levels. |
| */ |
| if (fin->fin_depth >= 16) |
| return pass; |
| |
| fr = fin->fin_fr; |
| |
| /* |
| * If there are no rules in this list, return now. |
| */ |
| if (fr == NULL) |
| return pass; |
| |
| skip = 0; |
| portcmp = 0; |
| fin->fin_depth++; |
| fin->fin_fr = NULL; |
| off = fin->fin_off; |
| |
| if ((fin->fin_flx & FI_TCPUDP) && (fin->fin_dlen > 3) && !off) |
| portcmp = 1; |
| |
| for (rulen = 0; fr; fr = fnext, rulen++) { |
| fnext = fr->fr_next; |
| if (skip != 0) { |
| FR_VERBOSE(("SKIP %d (%#x)\n", skip, fr->fr_flags)); |
| skip--; |
| continue; |
| } |
| |
| /* |
| * In all checks below, a null (zero) value in the |
| * filter struture is taken to mean a wildcard. |
| * |
| * check that we are working for the right interface |
| */ |
| #ifdef _KERNEL |
| if (fr->fr_ifa && fr->fr_ifa != fin->fin_ifp) |
| continue; |
| #else |
| if (opts & (OPT_VERBOSE|OPT_DEBUG)) |
| printf("\n"); |
| FR_VERBOSE(("%c", FR_ISSKIP(pass) ? 's' : |
| FR_ISPASS(pass) ? 'p' : |
| FR_ISACCOUNT(pass) ? 'A' : |
| FR_ISAUTH(pass) ? 'a' : |
| (pass & FR_NOMATCH) ? 'n' :'b')); |
| if (fr->fr_ifa && fr->fr_ifa != fin->fin_ifp) |
| continue; |
| FR_VERBOSE((":i")); |
| #endif |
| |
| switch (fr->fr_type) |
| { |
| case FR_T_IPF : |
| case FR_T_IPF|FR_T_BUILTIN : |
| if (ipf_check_ipf(fin, fr, portcmp)) |
| continue; |
| break; |
| #if defined(IPFILTER_BPF) |
| case FR_T_BPFOPC : |
| case FR_T_BPFOPC|FR_T_BUILTIN : |
| { |
| u_char *mc; |
| int wlen; |
| |
| if (*fin->fin_mp == NULL) |
| continue; |
| if (fin->fin_family != fr->fr_family) |
| continue; |
| mc = (u_char *)fin->fin_m; |
| wlen = fin->fin_dlen + fin->fin_hlen; |
| if (!bpf_filter(fr->fr_data, mc, wlen, 0)) |
| continue; |
| break; |
| } |
| #endif |
| case FR_T_CALLFUNC|FR_T_BUILTIN : |
| { |
| frentry_t *f; |
| |
| f = (*fr->fr_func)(fin, &pass); |
| if (f != NULL) |
| fr = f; |
| else |
| continue; |
| break; |
| } |
| |
| case FR_T_IPFEXPR : |
| case FR_T_IPFEXPR|FR_T_BUILTIN : |
| if (fin->fin_family != fr->fr_family) |
| continue; |
| if (ipf_fr_matcharray(fin, fr->fr_data) == 0) |
| continue; |
| break; |
| |
| default : |
| break; |
| } |
| |
| if ((fin->fin_out == 0) && (fr->fr_nattag.ipt_num[0] != 0)) { |
| if (fin->fin_nattag == NULL) |
| continue; |
| if (ipf_matchtag(&fr->fr_nattag, fin->fin_nattag) == 0) |
| continue; |
| } |
| FR_VERBOSE(("=%d/%d.%d *", fr->fr_grhead, fr->fr_group, rulen)); |
| |
| passt = fr->fr_flags; |
| |
| /* |
| * If the rule is a "call now" rule, then call the function |
| * in the rule, if it exists and use the results from that. |
| * If the function pointer is bad, just make like we ignore |
| * it, except for increasing the hit counter. |
| */ |
| if ((passt & FR_CALLNOW) != 0) { |
| frentry_t *frs; |
| |
| ATOMIC_INC64(fr->fr_hits); |
| if ((fr->fr_func == NULL) || |
| (fr->fr_func == (ipfunc_t)-1)) |
| continue; |
| |
| frs = fin->fin_fr; |
| fin->fin_fr = fr; |
| fr = (*fr->fr_func)(fin, &passt); |
| if (fr == NULL) { |
| fin->fin_fr = frs; |
| continue; |
| } |
| passt = fr->fr_flags; |
| } |
| fin->fin_fr = fr; |
| |
| #ifdef IPFILTER_LOG |
| /* |
| * Just log this packet... |
| */ |
| if ((passt & FR_LOGMASK) == FR_LOG) { |
| if (ipf_log_pkt(fin, passt) == -1) { |
| if (passt & FR_LOGORBLOCK) { |
| passt &= ~FR_CMDMASK; |
| passt |= FR_BLOCK|FR_QUICK; |
| fin->fin_reason = 1; |
| } |
| } |
| } |
| #endif /* IPFILTER_LOG */ |
| |
| MUTEX_ENTER(&fr->fr_lock); |
| fr->fr_bytes += (U_QUAD_T)fin->fin_plen; |
| fr->fr_hits++; |
| MUTEX_EXIT(&fr->fr_lock); |
| fin->fin_rule = rulen; |
| |
| passo = pass; |
| if (FR_ISSKIP(passt)) { |
| skip = fr->fr_arg; |
| continue; |
| } else if ((passt & FR_LOGMASK) != FR_LOG) { |
| pass = passt; |
| } |
| |
| if (passt & (FR_RETICMP|FR_FAKEICMP)) |
| fin->fin_icode = fr->fr_icode; |
| |
| if (fr->fr_group != -1) { |
| (void) strncpy(fin->fin_group, |
| FR_NAME(fr, fr_group), |
| strlen(FR_NAME(fr, fr_group))); |
| } else { |
| fin->fin_group[0] = '\0'; |
| } |
| |
| FR_DEBUG(("pass %#x\n", pass)); |
| |
| if (fr->fr_grp != NULL) { |
| |
| fin->fin_fr = *fr->fr_grp; |
| FR_VERBOSE(("group %s\n", FR_NAME(fr, fr_grhead))); |
| |
| if (FR_ISDECAPS(pass)) |
| passt = ipf_decaps(fin, pass, fr->fr_icode); |
| else |
| passt = ipf_scanlist(fin, pass); |
| |
| if (fin->fin_fr == NULL) { |
| fin->fin_rule = rulen; |
| if (fr->fr_group != -1) |
| (void) strncpy(fin->fin_group, |
| fr->fr_names + |
| fr->fr_group, |
| strlen(fr->fr_names + |
| fr->fr_group)); |
| fin->fin_fr = fr; |
| passt = pass; |
| } |
| pass = passt; |
| } |
| |
| if (pass & FR_QUICK) { |
| /* |
| * Finally, if we've asked to track state for this |
| * packet, set it up. Add state for "quick" rules |
| * here so that if the action fails we can consider |
| * the rule to "not match" and keep on processing |
| * filter rules. |
| */ |
| if ((pass & FR_KEEPSTATE) && !FR_ISAUTH(pass) && |
| !(fin->fin_flx & FI_STATE)) { |
| int out = fin->fin_out; |
| |
| fin->fin_fr = fr; |
| if (ipf_state_add(softc, fin, NULL, 0) == 0) { |
| LBUMP(ipf_stats[out].fr_ads); |
| } else { |
| LBUMP(ipf_stats[out].fr_bads); |
| pass = passo; |
| continue; |
| } |
| } |
| break; |
| } |
| } |
| fin->fin_depth--; |
| return pass; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_acctpkt */ |
| /* Returns: frentry_t* - always returns NULL */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* passp(IO) - pointer to current/new filter decision (unused) */ |
| /* */ |
| /* Checks a packet against accounting rules, if there are any for the given */ |
| /* IP protocol version. */ |
| /* */ |
| /* N.B.: this function returns NULL to match the prototype used by other */ |
| /* functions called from the IPFilter "mainline" in ipf_check(). */ |
| /* ------------------------------------------------------------------------ */ |
| frentry_t * |
| ipf_acctpkt(fin, passp) |
| fr_info_t *fin; |
| u_32_t *passp; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| char group[FR_GROUPLEN]; |
| frentry_t *fr, *frsave; |
| u_32_t pass, rulen; |
| |
| passp = passp; |
| fr = softc->ipf_acct[fin->fin_out][softc->ipf_active]; |
| |
| if (fr != NULL) { |
| frsave = fin->fin_fr; |
| bcopy(fin->fin_group, group, FR_GROUPLEN); |
| rulen = fin->fin_rule; |
| fin->fin_fr = fr; |
| pass = ipf_scanlist(fin, FR_NOMATCH); |
| if (FR_ISACCOUNT(pass)) { |
| LBUMP(ipf_stats[0].fr_acct); |
| } |
| fin->fin_fr = frsave; |
| bcopy(group, fin->fin_group, FR_GROUPLEN); |
| fin->fin_rule = rulen; |
| } |
| return NULL; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_firewall */ |
| /* Returns: frentry_t* - returns pointer to matched rule, if no matches */ |
| /* were found, returns NULL. */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* passp(IO) - pointer to current/new filter decision (unused) */ |
| /* */ |
| /* Applies an appropriate set of firewall rules to the packet, to see if */ |
| /* there are any matches. The first check is to see if a match can be seen */ |
| /* in the cache. If not, then search an appropriate list of rules. Once a */ |
| /* matching rule is found, take any appropriate actions as defined by the */ |
| /* rule - except logging. */ |
| /* ------------------------------------------------------------------------ */ |
| static frentry_t * |
| ipf_firewall(fin, passp) |
| fr_info_t *fin; |
| u_32_t *passp; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| frentry_t *fr; |
| u_32_t pass; |
| int out; |
| |
| out = fin->fin_out; |
| pass = *passp; |
| |
| /* |
| * This rule cache will only affect packets that are not being |
| * statefully filtered. |
| */ |
| fin->fin_fr = softc->ipf_rules[out][softc->ipf_active]; |
| if (fin->fin_fr != NULL) |
| pass = ipf_scanlist(fin, softc->ipf_pass); |
| |
| if ((pass & FR_NOMATCH)) { |
| LBUMP(ipf_stats[out].fr_nom); |
| } |
| fr = fin->fin_fr; |
| |
| /* |
| * Apply packets per second rate-limiting to a rule as required. |
| */ |
| if ((fr != NULL) && (fr->fr_pps != 0) && |
| !ppsratecheck(&fr->fr_lastpkt, &fr->fr_curpps, fr->fr_pps)) { |
| pass &= ~(FR_CMDMASK|FR_DUP|FR_RETICMP|FR_RETRST); |
| pass |= FR_BLOCK; |
| LBUMP(ipf_stats[out].fr_ppshit); |
| fin->fin_reason = 2; |
| } |
| |
| /* |
| * If we fail to add a packet to the authorization queue, then we |
| * drop the packet later. However, if it was added then pretend |
| * we've dropped it already. |
| */ |
| if (FR_ISAUTH(pass)) { |
| if (ipf_auth_new(fin->fin_m, fin) != 0) { |
| #ifdef _KERNEL |
| fin->fin_m = *fin->fin_mp = NULL; |
| #else |
| ; |
| #endif |
| fin->fin_error = 0; |
| } else { |
| softc->ipf_interror = 1; |
| fin->fin_error = ENOSPC; |
| } |
| } |
| |
| if ((fr != NULL) && (fr->fr_func != NULL) && |
| (fr->fr_func != (ipfunc_t)-1) && !(pass & FR_CALLNOW)) |
| (void) (*fr->fr_func)(fin, &pass); |
| |
| /* |
| * If a rule is a pre-auth rule, check again in the list of rules |
| * loaded for authenticated use. It does not particulary matter |
| * if this search fails because a "preauth" result, from a rule, |
| * is treated as "not a pass", hence the packet is blocked. |
| */ |
| if (FR_ISPREAUTH(pass)) { |
| pass = ipf_auth_pre_scanlist(softc, fin, pass); |
| } |
| |
| /* |
| * If the rule has "keep frag" and the packet is actually a fragment, |
| * then create a fragment state entry. |
| */ |
| if ((pass & (FR_KEEPFRAG|FR_KEEPSTATE)) == FR_KEEPFRAG) { |
| if (fin->fin_flx & FI_FRAG) { |
| if (ipf_frag_new(softc, fin, pass) == -1) { |
| LBUMP(ipf_stats[out].fr_bnfr); |
| } else { |
| LBUMP(ipf_stats[out].fr_nfr); |
| } |
| } else { |
| LBUMP(ipf_stats[out].fr_cfr); |
| } |
| } |
| |
| fr = fin->fin_fr; |
| |
| if (passp != NULL) |
| *passp = pass; |
| |
| return fr; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_check */ |
| /* Returns: int - 0 == packet allowed through, */ |
| /* User space: */ |
| /* -1 == packet blocked */ |
| /* 1 == packet not matched */ |
| /* -2 == requires authentication */ |
| /* Kernel: */ |
| /* > 0 == filter error # for packet */ |
| /* Parameters: ip(I) - pointer to start of IPv4/6 packet */ |
| /* hlen(I) - length of header */ |
| /* ifp(I) - pointer to interface this packet is on */ |
| /* out(I) - 0 == packet going in, 1 == packet going out */ |
| /* mp(IO) - pointer to caller's buffer pointer that holds this */ |
| /* IP packet. */ |
| /* Solaris & HP-UX ONLY : */ |
| /* qpi(I) - pointer to STREAMS queue information for this */ |
| /* interface & direction. */ |
| /* */ |
| /* ipf_check() is the master function for all IPFilter packet processing. */ |
| /* It orchestrates: Network Address Translation (NAT), checking for packet */ |
| /* authorisation (or pre-authorisation), presence of related state info., */ |
| /* generating log entries, IP packet accounting, routing of packets as */ |
| /* directed by firewall rules and of course whether or not to allow the */ |
| /* packet to be further processed by the kernel. */ |
| /* */ |
| /* For packets blocked, the contents of "mp" will be NULL'd and the buffer */ |
| /* freed. Packets passed may be returned with the pointer pointed to by */ |
| /* by "mp" changed to a new buffer. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_check(ctx, ip, hlen, ifp, out |
| #if defined(_KERNEL) && defined(MENTAT) |
| , qif, mp) |
| void *qif; |
| #else |
| , mp) |
| #endif |
| mb_t **mp; |
| ip_t *ip; |
| int hlen; |
| void *ifp; |
| int out; |
| void *ctx; |
| { |
| /* |
| * The above really sucks, but short of writing a diff |
| */ |
| ipf_main_softc_t *softc = ctx; |
| fr_info_t frinfo; |
| fr_info_t *fin = &frinfo; |
| u_32_t pass = softc->ipf_pass; |
| frentry_t *fr = NULL; |
| int v = IP_V(ip); |
| mb_t *mc = NULL; |
| mb_t *m; |
| /* |
| * The first part of ipf_check() deals with making sure that what goes |
| * into the filtering engine makes some sense. Information about the |
| * the packet is distilled, collected into a fr_info_t structure and |
| * the an attempt to ensure the buffer the packet is in is big enough |
| * to hold all the required packet headers. |
| */ |
| #ifdef _KERNEL |
| # ifdef MENTAT |
| qpktinfo_t *qpi = qif; |
| |
| # ifdef __sparc |
| if ((u_int)ip & 0x3) |
| return 2; |
| # endif |
| # else |
| SPL_INT(s); |
| # endif |
| |
| if (softc->ipf_running <= 0) { |
| return 0; |
| } |
| |
| bzero((char *)fin, sizeof(*fin)); |
| |
| # ifdef MENTAT |
| if (qpi->qpi_flags & QF_GROUP) |
| fin->fin_flx |= FI_MBCAST; |
| m = qpi->qpi_m; |
| fin->fin_qfm = m; |
| fin->fin_qpi = qpi; |
| # else /* MENTAT */ |
| |
| m = *mp; |
| |
| # if defined(M_MCAST) |
| if ((m->m_flags & M_MCAST) != 0) |
| fin->fin_flx |= FI_MBCAST|FI_MULTICAST; |
| # endif |
| # if defined(M_MLOOP) |
| if ((m->m_flags & M_MLOOP) != 0) |
| fin->fin_flx |= FI_MBCAST|FI_MULTICAST; |
| # endif |
| # if defined(M_BCAST) |
| if ((m->m_flags & M_BCAST) != 0) |
| fin->fin_flx |= FI_MBCAST|FI_BROADCAST; |
| # endif |
| # ifdef M_CANFASTFWD |
| /* |
| * XXX For now, IP Filter and fast-forwarding of cached flows |
| * XXX are mutually exclusive. Eventually, IP Filter should |
| * XXX get a "can-fast-forward" filter rule. |
| */ |
| m->m_flags &= ~M_CANFASTFWD; |
| # endif /* M_CANFASTFWD */ |
| # if defined(CSUM_DELAY_DATA) && (!defined(__FreeBSD_version) || \ |
| (__FreeBSD_version < 501108)) |
| /* |
| * disable delayed checksums. |
| */ |
| if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { |
| in_delayed_cksum(m); |
| m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; |
| } |
| # endif /* CSUM_DELAY_DATA */ |
| # endif /* MENTAT */ |
| #else |
| bzero((char *)fin, sizeof(*fin)); |
| m = *mp; |
| # if defined(M_MCAST) |
| if ((m->m_flags & M_MCAST) != 0) |
| fin->fin_flx |= FI_MBCAST|FI_MULTICAST; |
| # endif |
| # if defined(M_MLOOP) |
| if ((m->m_flags & M_MLOOP) != 0) |
| fin->fin_flx |= FI_MBCAST|FI_MULTICAST; |
| # endif |
| # if defined(M_BCAST) |
| if ((m->m_flags & M_BCAST) != 0) |
| fin->fin_flx |= FI_MBCAST|FI_BROADCAST; |
| # endif |
| #endif /* _KERNEL */ |
| |
| fin->fin_v = v; |
| fin->fin_m = m; |
| fin->fin_ip = ip; |
| fin->fin_mp = mp; |
| fin->fin_out = out; |
| fin->fin_ifp = ifp; |
| fin->fin_error = ENETUNREACH; |
| fin->fin_hlen = (u_short)hlen; |
| fin->fin_dp = (char *)ip + hlen; |
| fin->fin_main_soft = softc; |
| |
| fin->fin_ipoff = (char *)ip - MTOD(m, char *); |
| |
| SPL_NET(s); |
| |
| #ifdef USE_INET6 |
| if (v == 6) { |
| LBUMP(ipf_stats[out].fr_ipv6); |
| /* |
| * Jumbo grams are quite likely too big for internal buffer |
| * structures to handle comfortably, for now, so just drop |
| * them. |
| */ |
| if (((ip6_t *)ip)->ip6_plen == 0) { |
| pass = FR_BLOCK|FR_NOMATCH; |
| fin->fin_reason = 3; |
| goto finished; |
| } |
| fin->fin_family = AF_INET6; |
| } else |
| #endif |
| { |
| fin->fin_family = AF_INET; |
| } |
| |
| if (ipf_makefrip(hlen, ip, fin) == -1) { |
| pass = FR_BLOCK|FR_NOMATCH; |
| fin->fin_reason = 4; |
| goto finished; |
| } |
| |
| /* |
| * For at least IPv6 packets, if a m_pullup() fails then this pointer |
| * becomes NULL and so we have no packet to free. |
| */ |
| if (*fin->fin_mp == NULL) |
| goto finished; |
| |
| if (!out) { |
| if (v == 4) { |
| if (softc->ipf_chksrc && !ipf_verifysrc(fin)) { |
| LBUMP(ipf_stats[0].fr_badsrc); |
| fin->fin_flx |= FI_BADSRC; |
| } |
| if (fin->fin_ip->ip_ttl < softc->ipf_minttl) { |
| LBUMP(ipf_stats[0].fr_badttl); |
| fin->fin_flx |= FI_LOWTTL; |
| } |
| } |
| #ifdef USE_INET6 |
| else if (v == 6) { |
| if (((ip6_t *)ip)->ip6_hlim < softc->ipf_minttl) { |
| LBUMP(ipf_stats[0].fr_badttl); |
| fin->fin_flx |= FI_LOWTTL; |
| } |
| } |
| #endif |
| } |
| |
| if (fin->fin_flx & FI_SHORT) { |
| LBUMP(ipf_stats[out].fr_short); |
| } |
| |
| READ_ENTER(&softc->ipf_mutex); |
| |
| if (!out && (softc->ipf_specfuncref[1][softc->ipf_active] == 0)) { |
| switch (fin->fin_v) |
| { |
| case 4 : |
| if (ipf_nat_checkin(fin, &pass) == -1) { |
| goto filterdone; |
| } |
| break; |
| #ifdef USE_INET6 |
| case 6 : |
| if (ipf_nat6_checkin(fin, &pass) == -1) { |
| goto filterdone; |
| } |
| break; |
| #endif |
| default : |
| break; |
| } |
| } |
| /* |
| * Check auth now. This, combined with the check below to see if |
| * apass is 0 is to ensure that we don't count the packet twice, |
| * which can otherwise occur when we reprocess it. As it is, we |
| * only count it after it has no auth. table matchup. |
| * |
| * If a packet is found in the auth table, then skip checking |
| * the access lists for permission but we do need to consider |
| * the result as if it were from the ACL's. In addition, being |
| * found in the auth table means it has been seen before, so do |
| * not pass it through accounting (again), lest it be counted twice. |
| */ |
| fr = ipf_auth_check(fin, &pass); |
| if (!out && (fr == NULL)) |
| (void) ipf_acctpkt(fin, NULL); |
| |
| if (fr == NULL) { |
| if ((fin->fin_flx & FI_FRAG) != 0) |
| fr = ipf_frag_known(fin, &pass); |
| |
| if ((fr == NULL) && (softc->ipf_specfuncref[0] |
| [softc->ipf_active] == 0)) |
| fr = ipf_state_check(fin, &pass); |
| } |
| |
| if ((pass & FR_NOMATCH) || (fr == NULL)) |
| fr = ipf_firewall(fin, &pass); |
| |
| /* |
| * If we've asked to track state for this packet, set it up. |
| * Here rather than ipf_firewall because ipf_checkauth may decide |
| * to return a packet for "keep state" |
| */ |
| if ((pass & FR_KEEPSTATE) && (fin->fin_m != NULL) && |
| !(fin->fin_flx & FI_STATE)) { |
| if (ipf_state_add(softc, fin, NULL, 0) == 0) { |
| LBUMP(ipf_stats[out].fr_ads); |
| } else { |
| LBUMP(ipf_stats[out].fr_bads); |
| if (FR_ISPASS(pass)) { |
| pass &= ~FR_CMDMASK; |
| pass |= FR_BLOCK; |
| fin->fin_reason = 5; |
| } |
| } |
| } |
| |
| fin->fin_fr = fr; |
| |
| /* |
| * Only count/translate packets which will be passed on, out the |
| * interface. |
| */ |
| if (out && FR_ISPASS(pass)) { |
| (void) ipf_acctpkt(fin, NULL); |
| |
| switch (fin->fin_v) |
| { |
| case 4 : |
| if (!softc->ipf_specfuncref[2][softc->ipf_active] && |
| (ipf_nat_checkout(fin, &pass) == -1)) { |
| ; |
| } else if ((softc->ipf_update_ipid != 0) && (v == 4)) { |
| if (ipf_updateipid(fin) == -1) { |
| LBUMP(ipf_stats[1].fr_ipud); |
| pass &= ~FR_CMDMASK; |
| pass |= FR_BLOCK; |
| fin->fin_reason = 6; |
| } else { |
| LBUMP(ipf_stats[0].fr_ipud); |
| } |
| } |
| break; |
| #ifdef USE_INET6 |
| case 6 : |
| if (softc->ipf_specfuncref[2][softc->ipf_active] == 0) |
| (void) ipf_nat6_checkout(fin, &pass); |
| break; |
| #endif |
| default : |
| break; |
| } |
| } |
| |
| filterdone: |
| #ifdef IPFILTER_LOG |
| if ((softc->ipf_flags & FF_LOGGING) || (pass & FR_LOGMASK)) { |
| (void) ipf_dolog(fin, &pass); |
| } |
| #endif |
| |
| /* |
| * The FI_STATE flag is cleared here so that calling ipf_state_check |
| * will work when called from inside of fr_fastroute. Although |
| * there is a similar flag, FI_NATED, for NAT, it does have the same |
| * impact on code execution. |
| */ |
| fin->fin_flx &= ~FI_STATE; |
| |
| #if defined(FASTROUTE_RECURSION) |
| /* |
| * Up the reference on fr_lock and exit ipf_mutex. The generation of |
| * a packet below can sometimes cause a recursive call into IPFilter. |
| * On those platforms where that does happen, we need to hang onto |
| * the filter rule just in case someone decides to remove or flush it |
| * in the meantime. |
| */ |
| if (fr != NULL) { |
| MUTEX_ENTER(&fr->fr_lock); |
| fr->fr_ref++; |
| MUTEX_EXIT(&fr->fr_lock); |
| } |
| |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| #endif |
| |
| if ((pass & FR_RETMASK) != 0) { |
| /* |
| * Should we return an ICMP packet to indicate error |
| * status passing through the packet filter ? |
| * WARNING: ICMP error packets AND TCP RST packets should |
| * ONLY be sent in repsonse to incoming packets. Sending |
| * them in response to outbound packets can result in a |
| * panic on some operating systems. |
| */ |
| if (!out) { |
| if (pass & FR_RETICMP) { |
| int dst; |
| |
| if ((pass & FR_RETMASK) == FR_FAKEICMP) |
| dst = 1; |
| else |
| dst = 0; |
| (void) ipf_send_icmp_err(ICMP_UNREACH, fin, |
| dst); |
| LBUMP(ipf_stats[0].fr_ret); |
| } else if (((pass & FR_RETMASK) == FR_RETRST) && |
| !(fin->fin_flx & FI_SHORT)) { |
| if (((fin->fin_flx & FI_OOW) != 0) || |
| (ipf_send_reset(fin) == 0)) { |
| LBUMP(ipf_stats[1].fr_ret); |
| } |
| } |
| |
| /* |
| * When using return-* with auth rules, the auth code |
| * takes over disposing of this packet. |
| */ |
| if (FR_ISAUTH(pass) && (fin->fin_m != NULL)) { |
| fin->fin_m = *fin->fin_mp = NULL; |
| m = NULL; |
| } |
| } else { |
| if (pass & FR_RETRST) { |
| fin->fin_error = ECONNRESET; |
| } |
| } |
| } |
| |
| /* |
| * After the above so that ICMP unreachables and TCP RSTs get |
| * created properly. |
| */ |
| if (FR_ISBLOCK(pass) && (fin->fin_flx & FI_NEWNAT)) |
| ipf_nat_uncreate(fin); |
| |
| /* |
| * If we didn't drop off the bottom of the list of rules (and thus |
| * the 'current' rule fr is not NULL), then we may have some extra |
| * instructions about what to do with a packet. |
| * Once we're finished return to our caller, freeing the packet if |
| * we are dropping it. |
| */ |
| if (fr != NULL) { |
| frdest_t *fdp; |
| |
| /* |
| * Generate a duplicated packet first because ipf_fastroute |
| * can lead to fin_m being free'd... not good. |
| */ |
| if ((pass & FR_DUP) != 0) { |
| mc = M_COPY(fin->fin_m); |
| if (mc != NULL) |
| ipf_fastroute(mc, &mc, fin, &fr->fr_dif); |
| } |
| |
| fdp = &fr->fr_tifs[fin->fin_rev]; |
| |
| if (!out && (pass & FR_FASTROUTE)) { |
| /* |
| * For fastroute rule, no destination interface defined |
| * so pass NULL as the frdest_t parameter |
| */ |
| (void) ipf_fastroute(fin->fin_m, mp, fin, NULL); |
| m = *mp = NULL; |
| } else if ((fdp->fd_ptr != NULL) && |
| (fdp->fd_ptr != (struct ifnet *)-1)) { |
| /* this is for to rules: */ |
| ipf_fastroute(fin->fin_m, mp, fin, fdp); |
| m = *mp = NULL; |
| } |
| |
| #if defined(FASTROUTE_RECURSION) |
| (void) ipf_derefrule(softc, &fr); |
| #endif |
| } |
| #if !defined(FASTROUTE_RECURSION) |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| #endif |
| |
| finished: |
| if (!FR_ISPASS(pass)) { |
| LBUMP(ipf_stats[out].fr_block); |
| if (*mp != NULL) { |
| FREE_MB_T(*mp); |
| m = *mp = NULL; |
| } |
| } else { |
| LBUMP(ipf_stats[out].fr_pass); |
| #if defined(_KERNEL) && defined(__sgi) |
| if ((fin->fin_hbuf != NULL) && |
| (mtod(fin->fin_m, struct ip *) != fin->fin_ip)) { |
| COPYBACK(fin->fin_m, 0, fin->fin_plen, fin->fin_hbuf); |
| } |
| #endif |
| } |
| |
| SPL_X(s); |
| |
| #ifdef _KERNEL |
| return (FR_ISPASS(pass)) ? 0 : fin->fin_error; |
| #else /* _KERNEL */ |
| blockreason = fin->fin_reason; |
| FR_VERBOSE(("fin_flx %#x pass %#x ", fin->fin_flx, pass)); |
| /*if ((pass & FR_CMDMASK) == (softc->ipf_pass & FR_CMDMASK))*/ |
| if ((pass & FR_NOMATCH) != 0) |
| return 1; |
| |
| if ((pass & FR_RETMASK) != 0) |
| switch (pass & FR_RETMASK) |
| { |
| case FR_RETRST : |
| return 3; |
| case FR_RETICMP : |
| return 4; |
| case FR_FAKEICMP : |
| return 5; |
| } |
| |
| switch (pass & FR_CMDMASK) |
| { |
| case FR_PASS : |
| return 0; |
| case FR_BLOCK : |
| return -1; |
| case FR_AUTH : |
| return -2; |
| case FR_ACCOUNT : |
| return -3; |
| case FR_PREAUTH : |
| return -4; |
| } |
| return 2; |
| #endif /* _KERNEL */ |
| } |
| |
| |
| #ifdef IPFILTER_LOG |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_dolog */ |
| /* Returns: frentry_t* - returns contents of fin_fr (no change made) */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* passp(IO) - pointer to current/new filter decision (unused) */ |
| /* */ |
| /* Checks flags set to see how a packet should be logged, if it is to be */ |
| /* logged. Adjust statistics based on its success or not. */ |
| /* ------------------------------------------------------------------------ */ |
| frentry_t * |
| ipf_dolog(fin, passp) |
| fr_info_t *fin; |
| u_32_t *passp; |
| { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| u_32_t pass; |
| int out; |
| |
| out = fin->fin_out; |
| pass = *passp; |
| |
| if ((softc->ipf_flags & FF_LOGNOMATCH) && (pass & FR_NOMATCH)) { |
| pass |= FF_LOGNOMATCH; |
| LBUMP(ipf_stats[out].fr_npkl); |
| goto logit; |
| |
| } else if (((pass & FR_LOGMASK) == FR_LOGP) || |
| (FR_ISPASS(pass) && (softc->ipf_flags & FF_LOGPASS))) { |
| if ((pass & FR_LOGMASK) != FR_LOGP) |
| pass |= FF_LOGPASS; |
| LBUMP(ipf_stats[out].fr_ppkl); |
| goto logit; |
| |
| } else if (((pass & FR_LOGMASK) == FR_LOGB) || |
| (FR_ISBLOCK(pass) && (softc->ipf_flags & FF_LOGBLOCK))) { |
| if ((pass & FR_LOGMASK) != FR_LOGB) |
| pass |= FF_LOGBLOCK; |
| LBUMP(ipf_stats[out].fr_bpkl); |
| |
| logit: |
| if (ipf_log_pkt(fin, pass) == -1) { |
| /* |
| * If the "or-block" option has been used then |
| * block the packet if we failed to log it. |
| */ |
| if ((pass & FR_LOGORBLOCK) && |
| FR_ISPASS(pass)) { |
| pass &= ~FR_CMDMASK; |
| pass |= FR_BLOCK; |
| fin->fin_reason = 7; |
| } |
| } |
| *passp = pass; |
| } |
| |
| return fin->fin_fr; |
| } |
| #endif /* IPFILTER_LOG */ |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_cksum */ |
| /* Returns: u_short - IP header checksum */ |
| /* Parameters: addr(I) - pointer to start of buffer to checksum */ |
| /* len(I) - length of buffer in bytes */ |
| /* */ |
| /* Calculate the two's complement 16 bit checksum of the buffer passed. */ |
| /* */ |
| /* N.B.: addr should be 16bit aligned. */ |
| /* ------------------------------------------------------------------------ */ |
| u_short |
| ipf_cksum(addr, len) |
| u_short *addr; |
| int len; |
| { |
| u_32_t sum = 0; |
| |
| for (sum = 0; len > 1; len -= 2) |
| sum += *addr++; |
| |
| /* mop up an odd byte, if necessary */ |
| if (len == 1) |
| sum += *(u_char *)addr; |
| |
| /* |
| * add back carry outs from top 16 bits to low 16 bits |
| */ |
| sum = (sum >> 16) + (sum & 0xffff); /* add hi 16 to low 16 */ |
| sum += (sum >> 16); /* add carry */ |
| return (u_short)(~sum); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: fr_cksum */ |
| /* Returns: u_short - layer 4 checksum */ |
| /* Parameters: m(I ) - pointer to buffer holding packet */ |
| /* ip(I) - pointer to IP header */ |
| /* l4proto(I) - protocol to caclulate checksum for */ |
| /* l4hdr(I) - pointer to layer 4 header */ |
| /* l3len(I) - length of layer 4 data plus layer 3 header */ |
| /* */ |
| /* Calculates the TCP checksum for the packet held in "m", using the data */ |
| /* in the IP header "ip" to seed it. */ |
| /* */ |
| /* NB: This function assumes we've pullup'd enough for all of the IP header */ |
| /* and the TCP header. We also assume that data blocks aren't allocated in */ |
| /* odd sizes. */ |
| /* */ |
| /* For IPv6, l3len excludes extension header size. */ |
| /* */ |
| /* Expects ip_len and ip_off to be in network byte order when called. */ |
| /* ------------------------------------------------------------------------ */ |
| u_short |
| fr_cksum(m, ip, l4proto, l4hdr, l3len) |
| mb_t *m; |
| ip_t *ip; |
| int l4proto, l3len; |
| void *l4hdr; |
| { |
| u_short *sp, slen, sumsave, l4hlen, *csump; |
| u_int sum, sum2; |
| int hlen; |
| #ifdef USE_INET6 |
| ip6_t *ip6; |
| #endif |
| |
| csump = NULL; |
| sumsave = 0; |
| l4hlen = 0; |
| sp = NULL; |
| slen = 0; |
| hlen = 0; |
| sum = 0; |
| |
| sum = htons((u_short)l4proto); |
| /* |
| * Add up IP Header portion |
| */ |
| #ifdef USE_INET6 |
| if (IP_V(ip) == 4) { |
| #endif |
| hlen = IP_HL(ip) << 2; |
| slen = l3len - hlen; |
| sum += htons(slen); |
| sp = (u_short *)&ip->ip_src; |
| sum += *sp++; /* ip_src */ |
| sum += *sp++; |
| sum += *sp++; /* ip_dst */ |
| sum += *sp++; |
| #ifdef USE_INET6 |
| } else if (IP_V(ip) == 6) { |
| ip6 = (ip6_t *)ip; |
| hlen = sizeof(*ip6); |
| slen = l3len - hlen; |
| sum += htons(slen); |
| sp = (u_short *)&ip6->ip6_src; |
| sum += *sp++; /* ip6_src */ |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; /* ip6_dst */ |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| sum += *sp++; |
| } |
| #endif |
| |
| switch (l4proto) |
| { |
| case IPPROTO_UDP : |
| csump = &((udphdr_t *)l4hdr)->uh_sum; |
| l4hlen = sizeof(udphdr_t); |
| break; |
| |
| case IPPROTO_TCP : |
| csump = &((tcphdr_t *)l4hdr)->th_sum; |
| l4hlen = sizeof(tcphdr_t); |
| break; |
| case IPPROTO_ICMP : |
| csump = &((icmphdr_t *)l4hdr)->icmp_cksum; |
| l4hlen = 4; |
| sum = 0; |
| break; |
| default : |
| break; |
| } |
| |
| if (csump != NULL) { |
| sumsave = *csump; |
| *csump = 0; |
| } |
| |
| l4hlen = l4hlen; /* LINT */ |
| |
| #ifdef _KERNEL |
| # ifdef MENTAT |
| { |
| void *rp = m->b_rptr; |
| |
| if ((unsigned char *)ip > m->b_rptr && (unsigned char *)ip < m->b_wptr) |
| m->b_rptr = (u_char *)ip; |
| sum2 = ip_cksum(m, hlen, sum); /* hlen == offset */ |
| m->b_rptr = rp; |
| sum2 = (u_short)(~sum2 & 0xffff); |
| } |
| # else /* MENTAT */ |
| # if defined(BSD) || defined(sun) |
| # if BSD_GE_YEAR(199103) |
| m->m_data += hlen; |
| # else |
| m->m_off += hlen; |
| # endif |
| m->m_len -= hlen; |
| sum2 = in_cksum(m, slen); |
| m->m_len += hlen; |
| # if BSD >= 199103 |
| m->m_data -= hlen; |
| # else |
| m->m_off -= hlen; |
| # endif |
| /* |
| * Both sum and sum2 are partial sums, so combine them together. |
| */ |
| sum += ~sum2 & 0xffff; |
| while (sum > 0xffff) |
| sum = (sum & 0xffff) + (sum >> 16); |
| sum2 = ~sum & 0xffff; |
| # else /* defined(BSD) || defined(sun) */ |
| { |
| union { |
| u_char c[2]; |
| u_short s; |
| } bytes; |
| u_short len = ntohs(ip->ip_len); |
| # if defined(__sgi) |
| int add; |
| # endif |
| |
| /* |
| * Add up IP Header portion |
| */ |
| if (sp != (u_short *)l4hdr) |
| sp = (u_short *)l4hdr; |
| |
| switch (l4proto) |
| { |
| case IPPROTO_UDP : |
| sum += *sp++; /* sport */ |
| sum += *sp++; /* dport */ |
| sum += *sp++; /* udp length */ |
| sum += *sp++; /* checksum */ |
| break; |
| |
| case IPPROTO_TCP : |
| sum += *sp++; /* sport */ |
| sum += *sp++; /* dport */ |
| sum += *sp++; /* seq */ |
| sum += *sp++; |
| sum += *sp++; /* ack */ |
| sum += *sp++; |
| sum += *sp++; /* off */ |
| sum += *sp++; /* win */ |
| sum += *sp++; /* checksum */ |
| sum += *sp++; /* urp */ |
| break; |
| case IPPROTO_ICMP : |
| sum = *sp++; /* type/code */ |
| sum += *sp++; /* checksum */ |
| break; |
| } |
| |
| # ifdef __sgi |
| /* |
| * In case we had to copy the IP & TCP header out of mbufs, |
| * skip over the mbuf bits which are the header |
| */ |
| if ((char *)ip != mtod(m, char *)) { |
| hlen = (char *)sp - (char *)ip; |
| while (hlen) { |
| add = MIN(hlen, m->m_len); |
| sp = (u_short *)(mtod(m, char *) + add); |
| hlen -= add; |
| if (add == m->m_len) { |
| m = m->m_next; |
| if (!hlen) { |
| if (!m) |
| break; |
| sp = mtod(m, u_short *); |
| } |
| PANIC((!m),("fr_cksum(1): not enough data")); |
| } |
| } |
| } |
| # endif |
| |
| len -= (l4hlen + hlen); |
| if (len <= 0) |
| goto nodata; |
| |
| while (len > 1) { |
| if (((char *)sp - mtod(m, char *)) >= m->m_len) { |
| m = m->m_next; |
| PANIC((!m),("fr_cksum(2): not enough data")); |
| sp = mtod(m, u_short *); |
| } |
| if (((char *)(sp + 1) - mtod(m, char *)) > m->m_len) { |
| bytes.c[0] = *(u_char *)sp; |
| m = m->m_next; |
| PANIC((!m),("fr_cksum(3): not enough data")); |
| sp = mtod(m, u_short *); |
| bytes.c[1] = *(u_char *)sp; |
| sum += bytes.s; |
| sp = (u_short *)((u_char *)sp + 1); |
| } |
| if ((u_long)sp & 1) { |
| bcopy((char *)sp++, (char *)&bytes.s, sizeof(bytes.s)); |
| sum += bytes.s; |
| } else |
| sum += *sp++; |
| len -= 2; |
| } |
| |
| if (len != 0) |
| sum += ntohs(*(u_char *)sp << 8); |
| nodata: |
| while (sum > 0xffff) |
| sum = (sum & 0xffff) + (sum >> 16); |
| sum2 = (u_short)(~sum & 0xffff); |
| } |
| # endif /* defined(BSD) || defined(sun) */ |
| # endif /* MENTAT */ |
| #else /* _KERNEL */ |
| /* |
| * Add up IP Header portion |
| */ |
| if (sp != (u_short *)l4hdr) |
| sp = (u_short *)l4hdr; |
| |
| for (; slen > 1; slen -= 2) |
| sum += *sp++; |
| if (slen) |
| sum += ntohs(*(u_char *)sp << 8); |
| while (sum > 0xffff) |
| sum = (sum & 0xffff) + (sum >> 16); |
| sum2 = (u_short)(~sum & 0xffff); |
| #endif /* _KERNEL */ |
| if (csump != NULL) |
| *csump = sumsave; |
| return sum2; |
| } |
| |
| |
| #if defined(_KERNEL) && ( (BSD_LT_YEAR(199103) && !defined(MENTAT)) || \ |
| defined(__sgi) ) && !defined(linux) && !defined(_AIX51) |
| /* |
| * Copyright (c) 1982, 1986, 1988, 1991, 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. |
| * |
| * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 |
| * $Id$"; |
| */ |
| /* |
| * Copy data from an mbuf chain starting "off" bytes from the beginning, |
| * continuing for "len" bytes, into the indicated buffer. |
| */ |
| void |
| m_copydata(m, off, len, cp) |
| mb_t *m; |
| int off; |
| int len; |
| caddr_t cp; |
| { |
| unsigned count; |
| |
| if (off < 0 || len < 0) |
| panic("m_copydata"); |
| while (off > 0) { |
| if (m == 0) |
| panic("m_copydata"); |
| if (off < m->m_len) |
| break; |
| off -= m->m_len; |
| m = m->m_next; |
| } |
| while (len > 0) { |
| if (m == 0) |
| panic("m_copydata"); |
| count = MIN(m->m_len - off, len); |
| bcopy(mtod(m, caddr_t) + off, cp, count); |
| len -= count; |
| cp += count; |
| off = 0; |
| m = m->m_next; |
| } |
| } |
| |
| |
| /* |
| * Copy data from a buffer back into the indicated mbuf chain, |
| * starting "off" bytes from the beginning, extending the mbuf |
| * chain if necessary. |
| */ |
| void |
| m_copyback(m0, off, len, cp) |
| struct mbuf *m0; |
| int off; |
| int len; |
| caddr_t cp; |
| { |
| int mlen; |
| struct mbuf *m = m0, *n; |
| int totlen = 0; |
| |
| if (m0 == 0) |
| return; |
| while (off > (mlen = m->m_len)) { |
| off -= mlen; |
| totlen += mlen; |
| if (m->m_next == 0) { |
| n = m_getclr(M_DONTWAIT, m->m_type); |
| if (n == 0) |
| goto out; |
| n->m_len = min(MLEN, len + off); |
| m->m_next = n; |
| } |
| m = m->m_next; |
| } |
| while (len > 0) { |
| mlen = min(m->m_len - off, len); |
| bcopy(cp, off + mtod(m, char *), (unsigned)mlen); |
| cp += mlen; |
| len -= mlen; |
| mlen += off; |
| off = 0; |
| totlen += mlen; |
| if (len == 0) |
| break; |
| if (m->m_next == 0) { |
| n = m_get(M_DONTWAIT, m->m_type); |
| if (n == 0) |
| break; |
| n->m_len = min(MLEN, len); |
| m->m_next = n; |
| } |
| m = m->m_next; |
| } |
| out: |
| #if 0 |
| if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) |
| m->m_pkthdr.len = totlen; |
| #endif |
| return; |
| } |
| #endif /* (_KERNEL) && ( ((BSD < 199103) && !MENTAT) || __sgi) */ |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_findgroup */ |
| /* Returns: frgroup_t * - NULL = group not found, else pointer to group */ |
| /* Parameters: group(I) - group name to search for */ |
| /* unit(I) - device to which this group belongs */ |
| /* set(I) - which set of rules (inactive/inactive) this is */ |
| /* fgpp(O) - pointer to place to store pointer to the pointer */ |
| /* to where to add the next (last) group or where */ |
| /* to delete group from. */ |
| /* */ |
| /* Search amongst the defined groups for a particular group number. */ |
| /* ------------------------------------------------------------------------ */ |
| frgroup_t * |
| ipf_findgroup(softc, group, unit, set, fgpp) |
| ipf_main_softc_t *softc; |
| char *group; |
| minor_t unit; |
| int set; |
| frgroup_t ***fgpp; |
| { |
| frgroup_t *fg, **fgp; |
| |
| /* |
| * Which list of groups to search in is dependent on which list of |
| * rules are being operated on. |
| */ |
| fgp = &softc->ipf_groups[unit][set]; |
| |
| while ((fg = *fgp) != NULL) { |
| if (strncmp(group, fg->fg_name, FR_GROUPLEN) == 0) |
| break; |
| else |
| fgp = &fg->fg_next; |
| } |
| if (fgpp != NULL) |
| *fgpp = fgp; |
| return fg; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_group_add */ |
| /* Returns: frgroup_t * - NULL == did not create group, */ |
| /* != NULL == pointer to the group */ |
| /* Parameters: num(I) - group number to add */ |
| /* head(I) - rule pointer that is using this as the head */ |
| /* flags(I) - rule flags which describe the type of rule it is */ |
| /* unit(I) - device to which this group will belong to */ |
| /* set(I) - which set of rules (inactive/inactive) this is */ |
| /* Write Locks: ipf_mutex */ |
| /* */ |
| /* Add a new group head, or if it already exists, increase the reference */ |
| /* count to it. */ |
| /* ------------------------------------------------------------------------ */ |
| frgroup_t * |
| ipf_group_add(softc, group, head, flags, unit, set) |
| ipf_main_softc_t *softc; |
| char *group; |
| void *head; |
| u_32_t flags; |
| minor_t unit; |
| int set; |
| { |
| frgroup_t *fg, **fgp; |
| u_32_t gflags; |
| |
| if (group == NULL) |
| return NULL; |
| |
| if (unit == IPL_LOGIPF && *group == '\0') |
| return NULL; |
| |
| fgp = NULL; |
| gflags = flags & FR_INOUT; |
| |
| fg = ipf_findgroup(softc, group, unit, set, &fgp); |
| if (fg != NULL) { |
| if (fg->fg_flags == 0) |
| fg->fg_flags = gflags; |
| else if (gflags != fg->fg_flags) |
| return NULL; |
| fg->fg_ref++; |
| return fg; |
| } |
| |
| KMALLOC(fg, frgroup_t *); |
| if (fg != NULL) { |
| fg->fg_head = head; |
| fg->fg_start = NULL; |
| fg->fg_next = *fgp; |
| bcopy(group, fg->fg_name, strlen(group) + 1); |
| fg->fg_flags = gflags; |
| fg->fg_ref = 1; |
| *fgp = fg; |
| } |
| return fg; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_group_del */ |
| /* Returns: int - number of rules deleted */ |
| /* Parameters: group(I) - group name to delete */ |
| /* unit(I) - device to which this group belongs */ |
| /* set(I) - which set of rules (inactive/inactive) this is */ |
| /* Write Locks: ipf_mutex */ |
| /* */ |
| /* Attempt to delete a group head. */ |
| /* Only do this when its reference count reaches 0. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_group_del(softc, group, unit, set) |
| ipf_main_softc_t *softc; |
| char *group; |
| minor_t unit; |
| int set; |
| { |
| frgroup_t *fg, **fgp; |
| int gone = 0; |
| |
| fg = ipf_findgroup(softc, group, unit, set, &fgp); |
| if (fg == NULL) |
| return 0; |
| |
| fg->fg_ref--; |
| if (fg->fg_ref == 0) { |
| |
| (void) ipf_flushlist(softc, set, unit, &gone, &fg->fg_start); |
| *fgp = fg->fg_next; |
| KFREE(fg); |
| } |
| |
| return gone; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_getrulen */ |
| /* Returns: frentry_t * - NULL == not found, else pointer to rule n */ |
| /* Parameters: unit(I) - device for which to count the rule's number */ |
| /* flags(I) - which set of rules to find the rule in */ |
| /* group(I) - group name */ |
| /* n(I) - rule number to find */ |
| /* */ |
| /* Find rule # n in group # g and return a pointer to it. Return NULl if */ |
| /* group # g doesn't exist or there are less than n rules in the group. */ |
| /* ------------------------------------------------------------------------ */ |
| frentry_t * |
| ipf_getrulen(softc, unit, group, n) |
| ipf_main_softc_t *softc; |
| int unit; |
| char *group; |
| u_32_t n; |
| { |
| frentry_t *fr; |
| frgroup_t *fg; |
| |
| fg = ipf_findgroup(softc, group, unit, softc->ipf_active, NULL); |
| if (fg == NULL) |
| return NULL; |
| for (fr = fg->fg_start; fr && n; fr = fr->fr_next, n--) |
| ; |
| if (n != 0) |
| return NULL; |
| return fr; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_flushlist */ |
| /* Returns: int - >= 0 - number of flushed rules */ |
| /* Parameters: set(I) - which set of rules (inactive/inactive) this is */ |
| /* unit(I) - device for which to flush rules */ |
| /* flags(I) - which set of rules to flush */ |
| /* nfreedp(O) - pointer to int where flush count is stored */ |
| /* listp(I) - pointer to list to flush pointer */ |
| /* Write Locks: ipf_mutex */ |
| /* */ |
| /* Recursively flush rules from the list, descending groups as they are */ |
| /* encountered. if a rule is the head of a group and it has lost all its */ |
| /* group members, then also delete the group reference. nfreedp is needed */ |
| /* to store the accumulating count of rules removed, whereas the returned */ |
| /* value is just the number removed from the current list. The latter is */ |
| /* needed to correctly adjust reference counts on rules that define groups. */ |
| /* */ |
| /* NOTE: Rules not loaded from user space cannot be flushed. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_flushlist(softc, set, unit, nfreedp, listp) |
| ipf_main_softc_t *softc; |
| int set; |
| minor_t unit; |
| int *nfreedp; |
| frentry_t **listp; |
| { |
| int freed = 0; |
| frentry_t *fp; |
| |
| while ((fp = *listp) != NULL) { |
| if ((fp->fr_type & FR_T_BUILTIN) || |
| !(fp->fr_flags & FR_COPIED)) { |
| listp = &fp->fr_next; |
| continue; |
| } |
| *listp = fp->fr_next; |
| if (fp->fr_next != NULL) |
| fp->fr_next->fr_pnext = fp->fr_pnext; |
| fp->fr_pnext = NULL; |
| |
| if (fp->fr_grp != NULL) { |
| ipf_flushlist(softc, set, unit, nfreedp, fp->fr_grp); |
| } |
| if (fp->fr_icmpgrp != NULL) { |
| ipf_flushlist(softc, set, unit, nfreedp, |
| fp->fr_icmpgrp); |
| } |
| |
| if (fp->fr_grhead != -1) { |
| freed += ipf_group_del(softc, |
| FR_NAME(fp, fr_grhead), |
| unit, set); |
| fp->fr_names[fp->fr_grhead] = '\0'; |
| } |
| |
| if (fp->fr_icmphead != -1) { |
| freed += ipf_group_del(softc, |
| FR_NAME(fp, fr_icmphead), |
| unit, set); |
| fp->fr_names[fp->fr_icmphead] = '\0'; |
| } |
| |
| if (fp->fr_srctrack.ht_max_nodes) |
| ipf_rb_ht_init(&fp->fr_srctrack); |
| |
| fp->fr_next = NULL; |
| |
| ASSERT(fp->fr_ref > 0); |
| if (ipf_derefrule(softc, &fp) == 0) |
| freed++; |
| } |
| *nfreedp += freed; |
| return freed; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_flush */ |
| /* Returns: int - >= 0 - number of flushed rules */ |
| /* Parameters: unit(I) - device for which to flush rules */ |
| /* flags(I) - which set of rules to flush */ |
| /* */ |
| /* Calls flushlist() for all filter rules (accounting, firewall - both IPv4 */ |
| /* and IPv6) as defined by the value of flags. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_flush(softc, unit, flags) |
| ipf_main_softc_t *softc; |
| minor_t unit; |
| int flags; |
| { |
| int flushed = 0, set; |
| |
| WRITE_ENTER(&softc->ipf_mutex); |
| |
| set = softc->ipf_active; |
| if ((flags & FR_INACTIVE) == FR_INACTIVE) |
| set = 1 - set; |
| |
| if (flags & FR_OUTQUE) { |
| ipf_flushlist(softc, set, unit, &flushed, |
| &softc->ipf_rules[1][set]); |
| ipf_flushlist(softc, set, unit, &flushed, |
| &softc->ipf_acct[1][set]); |
| } |
| if (flags & FR_INQUE) { |
| ipf_flushlist(softc, set, unit, &flushed, |
| &softc->ipf_rules[0][set]); |
| ipf_flushlist(softc, set, unit, &flushed, |
| &softc->ipf_acct[0][set]); |
| } |
| |
| flushed += ipf_flush_groups(softc, unit, set, |
| flags & (FR_INQUE|FR_OUTQUE)); |
| |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| |
| if (unit == IPL_LOGIPF) { |
| int tmp; |
| |
| tmp = ipf_flush(softc, IPL_LOGCOUNT, flags); |
| if (tmp >= 0) |
| flushed += tmp; |
| } |
| return flushed; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_flush_groups */ |
| /* Returns: int - >= 0 - number of flushed rules */ |
| /* Parameters: softc(I) - soft context pointerto work with */ |
| /* unit(I) - device for which to flush rules */ |
| /* set(I) - 1 or 0 (filter set) */ |
| /* flags(I) - which set of rules to flush */ |
| /* */ |
| /* Walk through all of the groups for the given unit and remove those that */ |
| /* match the flags passed in the correct set. Which set (either 1 or 0) is */ |
| /* determined from a combination of softc->ipf_active and whether or not */ |
| /* we are flushing active/inactive. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_flush_groups(softc, unit, set, flags) |
| ipf_main_softc_t *softc; |
| int unit, set, flags; |
| { |
| frgroup_t *fg, **fgp; |
| frentry_t *fr, **frp; |
| int flushed = 0; |
| |
| fgp = &softc->ipf_groups[unit][set]; |
| while ((fg = *fgp) != NULL) { |
| frp = &fg->fg_start; |
| while ((fr = *frp) != NULL) { |
| if ((fr->fr_flags & flags) == 0) { |
| frp = &fr->fr_next; |
| continue; |
| } |
| if (fr->fr_next != NULL) |
| fr->fr_next->fr_pnext = fr->fr_pnext; |
| fr->fr_pnext = NULL; |
| *frp = fr->fr_next; |
| (void) ipf_derefrule(softc, &fr); |
| flushed++; |
| } |
| |
| if (fg->fg_head != NULL) { |
| if (fg->fg_head->fr_grp == &fg->fg_start) |
| fg->fg_head->fr_grp = NULL; |
| if (fg->fg_head->fr_icmpgrp == &fg->fg_start) |
| fg->fg_head->fr_icmpgrp = NULL; |
| } |
| *fgp = fg->fg_next; |
| KFREE(fg); |
| } |
| return flushed; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: memstr */ |
| /* Returns: char * - NULL if failed, != NULL pointer to matching bytes */ |
| /* Parameters: src(I) - pointer to byte sequence to match */ |
| /* dst(I) - pointer to byte sequence to search */ |
| /* slen(I) - match length */ |
| /* dlen(I) - length available to search in */ |
| /* */ |
| /* Search dst for a sequence of bytes matching those at src and extend for */ |
| /* slen bytes. */ |
| /* ------------------------------------------------------------------------ */ |
| char * |
| memstr(src, dst, slen, dlen) |
| const char *src; |
| char *dst; |
| size_t slen, dlen; |
| { |
| char *s = NULL; |
| |
| while (dlen >= slen) { |
| if (bcmp(src, dst, slen) == 0) { |
| s = dst; |
| break; |
| } |
| dst++; |
| dlen--; |
| } |
| return s; |
| } |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_fixskip */ |
| /* Returns: Nil */ |
| /* Parameters: listp(IO) - pointer to start of list with skip rule */ |
| /* rp(I) - rule added/removed with skip in it. */ |
| /* addremove(I) - adjustment (-1/+1) to make to skip count, */ |
| /* depending on whether a rule was just added */ |
| /* or removed. */ |
| /* */ |
| /* Adjust all the rules in a list which would have skip'd past the position */ |
| /* where we are inserting to skip to the right place given the change. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_fixskip(listp, rp, addremove) |
| frentry_t **listp, *rp; |
| int addremove; |
| { |
| int rules, rn; |
| frentry_t *fp; |
| |
| rules = 0; |
| for (fp = *listp; (fp != NULL) && (fp != rp); fp = fp->fr_next) |
| rules++; |
| |
| if (!fp) |
| return; |
| |
| for (rn = 0, fp = *listp; fp && (fp != rp); fp = fp->fr_next, rn++) |
| if (FR_ISSKIP(fp->fr_flags) && (rn + fp->fr_arg >= rules)) |
| fp->fr_arg += addremove; |
| } |
| |
| |
| #ifdef _KERNEL |
| /* ------------------------------------------------------------------------ */ |
| /* Function: count4bits */ |
| /* Returns: int - >= 0 - number of consecutive bits in input */ |
| /* Parameters: ip(I) - 32bit IP address */ |
| /* */ |
| /* IPv4 ONLY */ |
| /* count consecutive 1's in bit mask. If the mask generated by counting */ |
| /* consecutive 1's is different to that passed, return -1, else return # */ |
| /* of bits. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| count4bits(ip) |
| u_32_t ip; |
| { |
| u_32_t ipn; |
| int cnt = 0, i, j; |
| |
| ip = ipn = ntohl(ip); |
| for (i = 32; i; i--, ipn *= 2) |
| if (ipn & 0x80000000) |
| cnt++; |
| else |
| break; |
| ipn = 0; |
| for (i = 32, j = cnt; i; i--, j--) { |
| ipn *= 2; |
| if (j > 0) |
| ipn++; |
| } |
| if (ipn == ip) |
| return cnt; |
| return -1; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: count6bits */ |
| /* Returns: int - >= 0 - number of consecutive bits in input */ |
| /* Parameters: msk(I) - pointer to start of IPv6 bitmask */ |
| /* */ |
| /* IPv6 ONLY */ |
| /* count consecutive 1's in bit mask. */ |
| /* ------------------------------------------------------------------------ */ |
| # ifdef USE_INET6 |
| int |
| count6bits(msk) |
| u_32_t *msk; |
| { |
| int i = 0, k; |
| u_32_t j; |
| |
| for (k = 3; k >= 0; k--) |
| if (msk[k] == 0xffffffff) |
| i += 32; |
| else { |
| for (j = msk[k]; j; j <<= 1) |
| if (j & 0x80000000) |
| i++; |
| } |
| return i; |
| } |
| # endif |
| #endif /* _KERNEL */ |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_synclist */ |
| /* Returns: void */ |
| /* Parameters: fr(I) - start of filter list to sync interface names for */ |
| /* ifp(I) - interface pointer for limiting sync lookups */ |
| /* Write Locks: ipf_mutex */ |
| /* */ |
| /* Walk through a list of filter rules and resolve any interface names into */ |
| /* pointers. Where dynamic addresses are used, also update the IP address */ |
| /* used in the rule. The interface pointer is used to limit the lookups to */ |
| /* a specific set of matching names if it is non-NULL. */ |
| /* ------------------------------------------------------------------------ */ |
| static void |
| ipf_synclist(softc, fr, ifp) |
| ipf_main_softc_t *softc; |
| frentry_t *fr; |
| void *ifp; |
| { |
| frdest_t *fdp; |
| char *name; |
| void *ifa; |
| int v, i; |
| |
| for (; fr; fr = fr->fr_next) { |
| v = fr->fr_family; |
| |
| /* |
| * Lookup all the interface names that are part of the rule. |
| */ |
| for (i = 0; i < 4; i++) { |
| if ((ifp != NULL) && (fr->fr_ifas[i] != ifp)) |
| continue; |
| if (fr->fr_ifnames[i] == -1) |
| continue; |
| name = FR_NAME(fr, fr_ifnames[i]); |
| fr->fr_ifas[i] = ipf_resolvenic(softc, name, v); |
| } |
| |
| if ((fr->fr_type & ~FR_T_BUILTIN) == FR_T_IPF) { |
| if (fr->fr_satype != FRI_NORMAL && |
| fr->fr_satype != FRI_LOOKUP) { |
| ifa = ipf_resolvenic(softc, fr->fr_names + |
| fr->fr_sifpidx, v); |
| ipf_ifpaddr(softc, v, fr->fr_satype, ifa, |
| &fr->fr_src6, &fr->fr_smsk6); |
| } |
| if (fr->fr_datype != FRI_NORMAL && |
| fr->fr_datype != FRI_LOOKUP) { |
| ifa = ipf_resolvenic(softc, fr->fr_names + |
| fr->fr_sifpidx, v); |
| ipf_ifpaddr(softc, v, fr->fr_datype, ifa, |
| &fr->fr_dst6, &fr->fr_dmsk6); |
| } |
| } |
| |
| fdp = &fr->fr_tifs[0]; |
| if ((ifp == NULL) || (fdp->fd_ptr == ifp)) |
| ipf_resolvedest(softc, fr->fr_names, fdp, v); |
| |
| fdp = &fr->fr_tifs[1]; |
| if ((ifp == NULL) || (fdp->fd_ptr == ifp)) |
| ipf_resolvedest(softc, fr->fr_names, fdp, v); |
| |
| fdp = &fr->fr_dif; |
| if ((ifp == NULL) || (fdp->fd_ptr == ifp)) { |
| ipf_resolvedest(softc, fr->fr_names, fdp, v); |
| |
| fr->fr_flags &= ~FR_DUP; |
| if ((fdp->fd_ptr != (void *)-1) && |
| (fdp->fd_ptr != NULL)) |
| fr->fr_flags |= FR_DUP; |
| } |
| |
| if (((fr->fr_type & FR_T_BUILTIN) == FR_T_IPF) && |
| (fr->fr_satype == FRI_LOOKUP) && (fr->fr_srcptr == NULL)) { |
| fr->fr_srcptr = ipf_lookup_res_num(softc, |
| fr->fr_srctype, |
| IPL_LOGIPF, |
| fr->fr_srcnum, |
| &fr->fr_srcfunc); |
| } |
| if (((fr->fr_type & FR_T_BUILTIN) == FR_T_IPF) && |
| (fr->fr_datype == FRI_LOOKUP) && (fr->fr_dstptr == NULL)) { |
| fr->fr_dstptr = ipf_lookup_res_num(softc, |
| fr->fr_dsttype, |
| IPL_LOGIPF, |
| fr->fr_dstnum, |
| &fr->fr_dstfunc); |
| } |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_sync */ |
| /* Returns: void */ |
| /* Parameters: Nil */ |
| /* */ |
| /* ipf_sync() is called when we suspect that the interface list or */ |
| /* information about interfaces (like IP#) has changed. Go through all */ |
| /* filter rules, NAT entries and the state table and check if anything */ |
| /* needs to be changed/updated. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_sync(softc, ifp) |
| ipf_main_softc_t *softc; |
| void *ifp; |
| { |
| int i; |
| |
| # if !SOLARIS |
| ipf_nat_sync(softc, ifp); |
| ipf_state_sync(softc, ifp); |
| # endif |
| |
| WRITE_ENTER(&softc->ipf_mutex); |
| ipf_synclist(softc, softc->ipf_acct[0][softc->ipf_active], ifp); |
| ipf_synclist(softc, softc->ipf_acct[1][softc->ipf_active], ifp); |
| ipf_synclist(softc, softc->ipf_rules[0][softc->ipf_active], ifp); |
| ipf_synclist(softc, softc->ipf_rules[1][softc->ipf_active], ifp); |
| |
| for (i = 0; i < IPL_LOGSIZE; i++) { |
| frgroup_t *g; |
| |
| for (g = softc->ipf_groups[i][0]; g != NULL; g = g->fg_next) |
| ipf_synclist(softc, g->fg_start, ifp); |
| for (g = softc->ipf_groups[i][1]; g != NULL; g = g->fg_next) |
| ipf_synclist(softc, g->fg_start, ifp); |
| } |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| |
| return 0; |
| } |
| |
| |
| /* |
| * In the functions below, bcopy() is called because the pointer being |
| * copied _from_ in this instance is a pointer to a char buf (which could |
| * end up being unaligned) and on the kernel's local stack. |
| */ |
| /* ------------------------------------------------------------------------ */ |
| /* Function: copyinptr */ |
| /* Returns: int - 0 = success, else failure */ |
| /* Parameters: src(I) - pointer to the source address */ |
| /* dst(I) - destination address */ |
| /* size(I) - number of bytes to copy */ |
| /* */ |
| /* Copy a block of data in from user space, given a pointer to the pointer */ |
| /* to start copying from (src) and a pointer to where to store it (dst). */ |
| /* NB: src - pointer to user space pointer, dst - kernel space pointer */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| copyinptr(softc, src, dst, size) |
| ipf_main_softc_t *softc; |
| void *src, *dst; |
| size_t size; |
| { |
| caddr_t ca; |
| int error; |
| |
| # if SOLARIS |
| error = COPYIN(src, &ca, sizeof(ca)); |
| if (error != 0) |
| return error; |
| # else |
| bcopy(src, (caddr_t)&ca, sizeof(ca)); |
| # endif |
| error = COPYIN(ca, dst, size); |
| if (error != 0) { |
| softc->ipf_interror = 3; |
| error = EFAULT; |
| } |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: copyoutptr */ |
| /* Returns: int - 0 = success, else failure */ |
| /* Parameters: src(I) - pointer to the source address */ |
| /* dst(I) - destination address */ |
| /* size(I) - number of bytes to copy */ |
| /* */ |
| /* Copy a block of data out to user space, given a pointer to the pointer */ |
| /* to start copying from (src) and a pointer to where to store it (dst). */ |
| /* NB: src - kernel space pointer, dst - pointer to user space pointer. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| copyoutptr(softc, src, dst, size) |
| ipf_main_softc_t *softc; |
| void *src, *dst; |
| size_t size; |
| { |
| caddr_t ca; |
| int error; |
| |
| bcopy(dst, (caddr_t)&ca, sizeof(ca)); |
| error = COPYOUT(src, ca, size); |
| if (error != 0) { |
| softc->ipf_interror = 4; |
| error = EFAULT; |
| } |
| return error; |
| } |
| #ifdef _KERNEL |
| #endif |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_lock */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: data(I) - pointer to lock value to set */ |
| /* lockp(O) - pointer to location to store old lock value */ |
| /* */ |
| /* Get the new value for the lock integer, set it and return the old value */ |
| /* in *lockp. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_lock(data, lockp) |
| caddr_t data; |
| int *lockp; |
| { |
| int arg, err; |
| |
| err = BCOPYIN(data, &arg, sizeof(arg)); |
| if (err != 0) |
| return EFAULT; |
| err = BCOPYOUT(lockp, data, sizeof(*lockp)); |
| if (err != 0) |
| return EFAULT; |
| *lockp = arg; |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_getstat */ |
| /* Returns: Nil */ |
| /* Parameters: fiop(I) - pointer to ipfilter stats structure */ |
| /* */ |
| /* Stores a copy of current pointers, counters, etc, in the friostat */ |
| /* structure. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_getstat(softc, fiop) |
| ipf_main_softc_t *softc; |
| friostat_t *fiop; |
| { |
| int i; |
| |
| bcopy((char *)softc->ipf_stats, (char *)fiop->f_st, |
| sizeof(ipf_statistics_t) * 2); |
| fiop->f_locks[IPL_LOGSTATE] = -1; |
| fiop->f_locks[IPL_LOGNAT] = -1; |
| fiop->f_locks[IPL_LOGIPF] = -1; |
| fiop->f_locks[IPL_LOGAUTH] = -1; |
| |
| fiop->f_ipf[0][0] = softc->ipf_rules[0][0]; |
| fiop->f_acct[0][0] = softc->ipf_acct[0][0]; |
| fiop->f_ipf[0][1] = softc->ipf_rules[0][1]; |
| fiop->f_acct[0][1] = softc->ipf_acct[0][1]; |
| fiop->f_ipf[1][0] = softc->ipf_rules[1][0]; |
| fiop->f_acct[1][0] = softc->ipf_acct[1][0]; |
| fiop->f_ipf[1][1] = softc->ipf_rules[1][1]; |
| fiop->f_acct[1][1] = softc->ipf_acct[1][1]; |
| |
| fiop->f_ticks = softc->ipf_ticks; |
| fiop->f_active = softc->ipf_active; |
| fiop->f_froute[0] = softc->ipf_frouteok[0]; |
| fiop->f_froute[1] = softc->ipf_frouteok[1]; |
| fiop->f_rb_no_mem = softc->ipf_rb_no_mem; |
| fiop->f_rb_node_max = softc->ipf_rb_node_max; |
| |
| fiop->f_running = softc->ipf_running; |
| for (i = 0; i < IPL_LOGSIZE; i++) { |
| fiop->f_groups[i][0] = softc->ipf_groups[i][0]; |
| fiop->f_groups[i][1] = softc->ipf_groups[i][1]; |
| } |
| #ifdef IPFILTER_LOG |
| fiop->f_log_ok = ipf_log_logok(softc, IPL_LOGIPF); |
| fiop->f_log_fail = ipf_log_failures(softc, IPL_LOGIPF); |
| fiop->f_logging = 1; |
| #else |
| fiop->f_log_ok = 0; |
| fiop->f_log_fail = 0; |
| fiop->f_logging = 0; |
| #endif |
| fiop->f_defpass = softc->ipf_pass; |
| fiop->f_features = ipf_features; |
| (void) strncpy(fiop->f_version, ipfilter_version, |
| sizeof(fiop->f_version)); |
| } |
| |
| |
| #ifdef USE_INET6 |
| int icmptoicmp6types[ICMP_MAXTYPE+1] = { |
| ICMP6_ECHO_REPLY, /* 0: ICMP_ECHOREPLY */ |
| -1, /* 1: UNUSED */ |
| -1, /* 2: UNUSED */ |
| ICMP6_DST_UNREACH, /* 3: ICMP_UNREACH */ |
| -1, /* 4: ICMP_SOURCEQUENCH */ |
| ND_REDIRECT, /* 5: ICMP_REDIRECT */ |
| -1, /* 6: UNUSED */ |
| -1, /* 7: UNUSED */ |
| ICMP6_ECHO_REQUEST, /* 8: ICMP_ECHO */ |
| -1, /* 9: UNUSED */ |
| -1, /* 10: UNUSED */ |
| ICMP6_TIME_EXCEEDED, /* 11: ICMP_TIMXCEED */ |
| ICMP6_PARAM_PROB, /* 12: ICMP_PARAMPROB */ |
| -1, /* 13: ICMP_TSTAMP */ |
| -1, /* 14: ICMP_TSTAMPREPLY */ |
| -1, /* 15: ICMP_IREQ */ |
| -1, /* 16: ICMP_IREQREPLY */ |
| -1, /* 17: ICMP_MASKREQ */ |
| -1, /* 18: ICMP_MASKREPLY */ |
| }; |
| |
| |
| int icmptoicmp6unreach[ICMP_MAX_UNREACH] = { |
| ICMP6_DST_UNREACH_ADDR, /* 0: ICMP_UNREACH_NET */ |
| ICMP6_DST_UNREACH_ADDR, /* 1: ICMP_UNREACH_HOST */ |
| -1, /* 2: ICMP_UNREACH_PROTOCOL */ |
| ICMP6_DST_UNREACH_NOPORT, /* 3: ICMP_UNREACH_PORT */ |
| -1, /* 4: ICMP_UNREACH_NEEDFRAG */ |
| ICMP6_DST_UNREACH_NOTNEIGHBOR, /* 5: ICMP_UNREACH_SRCFAIL */ |
| ICMP6_DST_UNREACH_ADDR, /* 6: ICMP_UNREACH_NET_UNKNOWN */ |
| ICMP6_DST_UNREACH_ADDR, /* 7: ICMP_UNREACH_HOST_UNKNOWN */ |
| -1, /* 8: ICMP_UNREACH_ISOLATED */ |
| ICMP6_DST_UNREACH_ADMIN, /* 9: ICMP_UNREACH_NET_PROHIB */ |
| ICMP6_DST_UNREACH_ADMIN, /* 10: ICMP_UNREACH_HOST_PROHIB */ |
| -1, /* 11: ICMP_UNREACH_TOSNET */ |
| -1, /* 12: ICMP_UNREACH_TOSHOST */ |
| ICMP6_DST_UNREACH_ADMIN, /* 13: ICMP_UNREACH_ADMIN_PROHIBIT */ |
| }; |
| int icmpreplytype6[ICMP6_MAXTYPE + 1]; |
| #endif |
| |
| int icmpreplytype4[ICMP_MAXTYPE + 1]; |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_matchicmpqueryreply */ |
| /* Returns: int - 1 if "icmp" is a valid reply to "ic" else 0. */ |
| /* Parameters: v(I) - IP protocol version (4 or 6) */ |
| /* ic(I) - ICMP information */ |
| /* icmp(I) - ICMP packet header */ |
| /* rev(I) - direction (0 = forward/1 = reverse) of packet */ |
| /* */ |
| /* Check if the ICMP packet defined by the header pointed to by icmp is a */ |
| /* reply to one as described by what's in ic. If it is a match, return 1, */ |
| /* else return 0 for no match. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_matchicmpqueryreply(v, ic, icmp, rev) |
| int v; |
| icmpinfo_t *ic; |
| icmphdr_t *icmp; |
| int rev; |
| { |
| int ictype; |
| |
| ictype = ic->ici_type; |
| |
| if (v == 4) { |
| /* |
| * If we matched its type on the way in, then when going out |
| * it will still be the same type. |
| */ |
| if ((!rev && (icmp->icmp_type == ictype)) || |
| (rev && (icmpreplytype4[ictype] == icmp->icmp_type))) { |
| if (icmp->icmp_type != ICMP_ECHOREPLY) |
| return 1; |
| if (icmp->icmp_id == ic->ici_id) |
| return 1; |
| } |
| } |
| #ifdef USE_INET6 |
| else if (v == 6) { |
| if ((!rev && (icmp->icmp_type == ictype)) || |
| (rev && (icmpreplytype6[ictype] == icmp->icmp_type))) { |
| if (icmp->icmp_type != ICMP6_ECHO_REPLY) |
| return 1; |
| if (icmp->icmp_id == ic->ici_id) |
| return 1; |
| } |
| } |
| #endif |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: frrequest */ |
| /* Returns: int - 0 == success, > 0 == errno value */ |
| /* Parameters: unit(I) - device for which this is for */ |
| /* req(I) - ioctl command (SIOC*) */ |
| /* data(I) - pointr to ioctl data */ |
| /* set(I) - 1 or 0 (filter set) */ |
| /* makecopy(I) - flag indicating whether data points to a rule */ |
| /* in kernel space & hence doesn't need copying. */ |
| /* */ |
| /* This function handles all the requests which operate on the list of */ |
| /* filter rules. This includes adding, deleting, insertion. It is also */ |
| /* responsible for creating groups when a "head" rule is loaded. Interface */ |
| /* names are resolved here and other sanity checks are made on the content */ |
| /* of the rule structure being loaded. If a rule has user defined timeouts */ |
| /* then make sure they are created and initialised before exiting. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| frrequest(softc, unit, req, data, set, makecopy) |
| ipf_main_softc_t *softc; |
| int unit; |
| ioctlcmd_t req; |
| int set, makecopy; |
| caddr_t data; |
| { |
| int error = 0, in, family, addrem, need_free = 0; |
| frentry_t frd, *fp, *f, **fprev, **ftail; |
| void *ptr, *uptr, *cptr; |
| u_int *p, *pp; |
| frgroup_t *fg; |
| char *group; |
| |
| ptr = NULL; |
| cptr = NULL; |
| fg = NULL; |
| fp = &frd; |
| if (makecopy != 0) { |
| error = ipf_inobj(softc, data, fp, IPFOBJ_FRENTRY); |
| if (error) { |
| return error; |
| } |
| if ((fp->fr_type & FR_T_BUILTIN) != 0) { |
| softc->ipf_interror = 6; |
| return EINVAL; |
| } |
| KMALLOCS(f, frentry_t *, fp->fr_size); |
| if (f == NULL) { |
| softc->ipf_interror = 131; |
| return ENOMEM; |
| } |
| error = ipf_inobjsz(softc, data, f, IPFOBJ_FRENTRY, |
| fp->fr_size); |
| if (error) { |
| KFREES(f, fp->fr_size); |
| return error; |
| } |
| |
| fp = f; |
| f = NULL; |
| fp->fr_ref = 0; |
| fp->fr_flags |= FR_COPIED; |
| } else { |
| fp = (frentry_t *)data; |
| if ((fp->fr_type & FR_T_BUILTIN) == 0) { |
| softc->ipf_interror = 7; |
| return EINVAL; |
| } |
| fp->fr_flags &= ~FR_COPIED; |
| } |
| |
| if (((fp->fr_dsize == 0) && (fp->fr_data != NULL)) || |
| ((fp->fr_dsize != 0) && (fp->fr_data == NULL))) { |
| softc->ipf_interror = 8; |
| error = EINVAL; |
| goto donenolock; |
| } |
| |
| family = fp->fr_family; |
| uptr = fp->fr_data; |
| |
| if (req == (ioctlcmd_t)SIOCINAFR || req == (ioctlcmd_t)SIOCINIFR || |
| req == (ioctlcmd_t)SIOCADAFR || req == (ioctlcmd_t)SIOCADIFR) |
| addrem = 0; |
| else if (req == (ioctlcmd_t)SIOCRMAFR || req == (ioctlcmd_t)SIOCRMIFR) |
| addrem = 1; |
| else if (req == (ioctlcmd_t)SIOCZRLST) |
| addrem = 2; |
| else { |
| softc->ipf_interror = 9; |
| error = EINVAL; |
| goto donenolock; |
| } |
| |
| /* |
| * Only filter rules for IPv4 or IPv6 are accepted. |
| */ |
| if (family == AF_INET) { |
| /*EMPTY*/; |
| #ifdef USE_INET6 |
| } else if (family == AF_INET6) { |
| /*EMPTY*/; |
| #endif |
| } else if (family != 0) { |
| softc->ipf_interror = 10; |
| error = EINVAL; |
| goto donenolock; |
| } |
| |
| /* |
| * If the rule is being loaded from user space, i.e. we had to copy it |
| * into kernel space, then do not trust the function pointer in the |
| * rule. |
| */ |
| if ((makecopy == 1) && (fp->fr_func != NULL)) { |
| if (ipf_findfunc(fp->fr_func) == NULL) { |
| softc->ipf_interror = 11; |
| error = ESRCH; |
| goto donenolock; |
| } |
| error = ipf_funcinit(softc, fp); |
| if (error != 0) |
| goto donenolock; |
| } |
| if ((fp->fr_flags & FR_CALLNOW) && |
| ((fp->fr_func == NULL) || (fp->fr_func == (void *)-1))) { |
| softc->ipf_interror = 142; |
| error = ESRCH; |
| goto donenolock; |
| } |
| if (((fp->fr_flags & FR_CMDMASK) == FR_CALL) && |
| ((fp->fr_func == NULL) || (fp->fr_func == (void *)-1))) { |
| softc->ipf_interror = 143; |
| error = ESRCH; |
| goto donenolock; |
| } |
| |
| ptr = NULL; |
| cptr = NULL; |
| |
| if (FR_ISACCOUNT(fp->fr_flags)) |
| unit = IPL_LOGCOUNT; |
| |
| /* |
| * Check that each group name in the rule has a start index that |
| * is valid. |
| */ |
| if (fp->fr_icmphead != -1) { |
| if ((fp->fr_icmphead < 0) || |
| (fp->fr_icmphead >= fp->fr_namelen)) { |
| softc->ipf_interror = 136; |
| error = EINVAL; |
| goto donenolock; |
| } |
| if (!strcmp(FR_NAME(fp, fr_icmphead), "0")) |
| fp->fr_names[fp->fr_icmphead] = '\0'; |
| } |
| |
| if (fp->fr_grhead != -1) { |
| if ((fp->fr_grhead < 0) || |
| (fp->fr_grhead >= fp->fr_namelen)) { |
| softc->ipf_interror = 137; |
| error = EINVAL; |
| goto donenolock; |
| } |
| if (!strcmp(FR_NAME(fp, fr_grhead), "0")) |
| fp->fr_names[fp->fr_grhead] = '\0'; |
| } |
| |
| if (fp->fr_group != -1) { |
| if ((fp->fr_group < 0) || |
| (fp->fr_group >= fp->fr_namelen)) { |
| softc->ipf_interror = 138; |
| error = EINVAL; |
| goto donenolock; |
| } |
| if ((req != (int)SIOCZRLST) && (fp->fr_group != -1)) { |
| /* |
| * Allow loading rules that are in groups to cause |
| * them to be created if they don't already exit. |
| */ |
| group = FR_NAME(fp, fr_group); |
| fg = ipf_findgroup(softc, group, unit, set, NULL); |
| if (fg == NULL) { |
| if (addrem == 0) { |
| fg = ipf_group_add(softc, group, NULL, |
| fp->fr_flags, unit, |
| set); |
| } |
| if (fg == NULL) { |
| softc->ipf_interror = 12; |
| error = ESRCH; |
| goto donenolock; |
| } |
| } |
| if (fg->fg_flags == 0) |
| fg->fg_flags = fp->fr_flags & FR_INOUT; |
| else if (fg->fg_flags != (fp->fr_flags & FR_INOUT)) { |
| softc->ipf_interror = 13; |
| error = ESRCH; |
| goto donenolock; |
| } |
| } |
| } else { |
| /* |
| * If a rule is going to be part of a group then it does |
| * not matter whether it is an in or out rule, but if it |
| * isn't in a group, then it does... |
| */ |
| if ((fp->fr_flags & (FR_INQUE|FR_OUTQUE)) == 0) { |
| softc->ipf_interror = 14; |
| error = EINVAL; |
| goto donenolock; |
| } |
| } |
| in = (fp->fr_flags & FR_INQUE) ? 0 : 1; |
| |
| /* |
| * Work out which rule list this change is being applied to. |
| */ |
| ftail = NULL; |
| fprev = NULL; |
| if (unit == IPL_LOGAUTH) |
| fprev = ipf_auth_rulehead(softc); |
| else { |
| if (FR_ISACCOUNT(fp->fr_flags)) |
| fprev = &softc->ipf_acct[in][set]; |
| else if ((fp->fr_flags & (FR_OUTQUE|FR_INQUE)) != 0) |
| fprev = &softc->ipf_rules[in][set]; |
| } |
| if (fprev == NULL) { |
| softc->ipf_interror = 15; |
| error = ESRCH; |
| goto donenolock; |
| } |
| |
| if (fg != NULL) |
| fprev = &fg->fg_start; |
| |
| /* |
| * Copy in extra data for the rule. |
| */ |
| if (fp->fr_dsize != 0) { |
| if (makecopy != 0) { |
| KMALLOCS(ptr, void *, fp->fr_dsize); |
| if (ptr == NULL) { |
| softc->ipf_interror = 16; |
| error = ENOMEM; |
| goto donenolock; |
| } |
| error = COPYIN(uptr, ptr, fp->fr_dsize); |
| if (error != 0) { |
| softc->ipf_interror = 17; |
| KFREES(ptr, fp->fr_dsize); |
| error = EFAULT; |
| goto donenolock; |
| } |
| } else { |
| ptr = uptr; |
| } |
| fp->fr_data = ptr; |
| } else { |
| fp->fr_data = NULL; |
| } |
| |
| /* |
| * Perform per-rule type sanity checks of their members. |
| * All code after this needs to be aware that allocated memory |
| * may need to be free'd before exiting. |
| */ |
| switch (fp->fr_type & ~FR_T_BUILTIN) |
| { |
| #if defined(IPFILTER_BPF) |
| case FR_T_BPFOPC : |
| if (fp->fr_dsize == 0) { |
| softc->ipf_interror = 19; |
| error = EINVAL; |
| break; |
| } |
| if (!bpf_validate(ptr, fp->fr_dsize/sizeof(struct bpf_insn))) { |
| softc->ipf_interror = 20; |
| error = EINVAL; |
| break; |
| } |
| break; |
| #endif |
| case FR_T_IPF : |
| /* |
| * Preparation for error case at the bottom of this function. |
| */ |
| if (fp->fr_datype == FRI_LOOKUP) |
| fp->fr_dstptr = NULL; |
| if (fp->fr_satype == FRI_LOOKUP) |
| fp->fr_srcptr = NULL; |
| |
| if (fp->fr_dsize != sizeof(fripf_t)) { |
| softc->ipf_interror = 21; |
| error = EINVAL; |
| break; |
| } |
| |
| /* |
| * Allowing a rule with both "keep state" and "with oow" is |
| * pointless because adding a state entry to the table will |
| * fail with the out of window (oow) flag set. |
| */ |
| if ((fp->fr_flags & FR_KEEPSTATE) && (fp->fr_flx & FI_OOW)) { |
| softc->ipf_interror = 22; |
| error = EINVAL; |
| break; |
| } |
| |
| switch (fp->fr_satype) |
| { |
| case FRI_BROADCAST : |
| case FRI_DYNAMIC : |
| case FRI_NETWORK : |
| case FRI_NETMASKED : |
| case FRI_PEERADDR : |
| if (fp->fr_sifpidx < 0) { |
| softc->ipf_interror = 23; |
| error = EINVAL; |
| } |
| break; |
| case FRI_LOOKUP : |
| fp->fr_srcptr = ipf_findlookup(softc, unit, fp, |
| &fp->fr_src6); |
| if (fp->fr_srcfunc == NULL) { |
| softc->ipf_interror = 132; |
| error = ESRCH; |
| break; |
| } |
| break; |
| case FRI_NORMAL : |
| break; |
| default : |
| softc->ipf_interror = 133; |
| error = EINVAL; |
| break; |
| } |
| if (error != 0) |
| break; |
| |
| switch (fp->fr_datype) |
| { |
| case FRI_BROADCAST : |
| case FRI_DYNAMIC : |
| case FRI_NETWORK : |
| case FRI_NETMASKED : |
| case FRI_PEERADDR : |
| if (fp->fr_difpidx < 0) { |
| softc->ipf_interror = 24; |
| error = EINVAL; |
| } |
| break; |
| case FRI_LOOKUP : |
| fp->fr_dstptr = ipf_findlookup(softc, unit, fp, |
| &fp->fr_dst6); |
| if (fp->fr_dstfunc == NULL) { |
| softc->ipf_interror = 134; |
| error = ESRCH; |
| } |
| break; |
| case FRI_NORMAL : |
| break; |
| default : |
| softc->ipf_interror = 135; |
| error = EINVAL; |
| } |
| break; |
| |
| case FR_T_NONE : |
| case FR_T_CALLFUNC : |
| case FR_T_COMPIPF : |
| break; |
| |
| case FR_T_IPFEXPR : |
| if (ipf_matcharray_verify(fp->fr_data, fp->fr_dsize) == -1) { |
| softc->ipf_interror = 25; |
| error = EINVAL; |
| } |
| break; |
| |
| default : |
| softc->ipf_interror = 26; |
| error = EINVAL; |
| break; |
| } |
| if (error != 0) |
| goto donenolock; |
| |
| if (fp->fr_tif.fd_name != -1) { |
| if ((fp->fr_tif.fd_name < 0) || |
| (fp->fr_tif.fd_name >= fp->fr_namelen)) { |
| softc->ipf_interror = 139; |
| error = EINVAL; |
| goto donenolock; |
| } |
| } |
| |
| if (fp->fr_dif.fd_name != -1) { |
| if ((fp->fr_dif.fd_name < 0) || |
| (fp->fr_dif.fd_name >= fp->fr_namelen)) { |
| softc->ipf_interror = 140; |
| error = EINVAL; |
| goto donenolock; |
| } |
| } |
| |
| if (fp->fr_rif.fd_name != -1) { |
| if ((fp->fr_rif.fd_name < 0) || |
| (fp->fr_rif.fd_name >= fp->fr_namelen)) { |
| softc->ipf_interror = 141; |
| error = EINVAL; |
| goto donenolock; |
| } |
| } |
| |
| /* |
| * Lookup all the interface names that are part of the rule. |
| */ |
| ipf_synclist(softc, fp, NULL); |
| fp->fr_statecnt = 0; |
| |
| /* |
| * Look for an existing matching filter rule, but don't include the |
| * next or interface pointer in the comparison (fr_next, fr_ifa). |
| * This elminates rules which are indentical being loaded. Checksum |
| * the constant part of the filter rule to make comparisons quicker |
| * (this meaning no pointers are included). |
| */ |
| for (fp->fr_cksum = 0, p = (u_int *)&fp->fr_func, pp = &fp->fr_cksum; |
| p < pp; p++) |
| fp->fr_cksum += *p; |
| pp = (u_int *)(fp->fr_caddr + fp->fr_dsize); |
| for (p = (u_int *)fp->fr_data; p < pp; p++) |
| fp->fr_cksum += *p; |
| |
| WRITE_ENTER(&softc->ipf_mutex); |
| |
| /* |
| * Now that the filter rule lists are locked, we can walk the |
| * chain of them without fear. |
| */ |
| ftail = fprev; |
| for (f = *ftail; (f = *ftail) != NULL; ftail = &f->fr_next) { |
| if (fp->fr_collect <= f->fr_collect) { |
| ftail = fprev; |
| f = NULL; |
| break; |
| } |
| fprev = ftail; |
| } |
| |
| for (; (f = *ftail) != NULL; ftail = &f->fr_next) { |
| if ((fp->fr_cksum != f->fr_cksum) || |
| (f->fr_dsize != fp->fr_dsize)) |
| continue; |
| if (bcmp((char *)&f->fr_func, (char *)&fp->fr_func, FR_CMPSIZ)) |
| continue; |
| if ((!ptr && !f->fr_data) || |
| (ptr && f->fr_data && |
| !bcmp((char *)ptr, (char *)f->fr_data, f->fr_dsize))) |
| break; |
| } |
| |
| /* |
| * If zero'ing statistics, copy current to caller and zero. |
| */ |
| if (addrem == 2) { |
| if (f == NULL) { |
| softc->ipf_interror = 27; |
| error = ESRCH; |
| } else { |
| /* |
| * Copy and reduce lock because of impending copyout. |
| * Well we should, but if we do then the atomicity of |
| * this call and the correctness of fr_hits and |
| * fr_bytes cannot be guaranteed. As it is, this code |
| * only resets them to 0 if they are successfully |
| * copied out into user space. |
| */ |
| bcopy((char *)f, (char *)fp, f->fr_size); |
| /* MUTEX_DOWNGRADE(&softc->ipf_mutex); */ |
| |
| /* |
| * When we copy this rule back out, set the data |
| * pointer to be what it was in user space. |
| */ |
| fp->fr_data = uptr; |
| error = ipf_outobj(softc, data, fp, IPFOBJ_FRENTRY); |
| |
| if (error == 0) { |
| if ((f->fr_dsize != 0) && (uptr != NULL)) |
| error = COPYOUT(f->fr_data, uptr, |
| f->fr_dsize); |
| if (error != 0) { |
| softc->ipf_interror = 28; |
| error = EFAULT; |
| } |
| if (error == 0) { |
| f->fr_hits = 0; |
| f->fr_bytes = 0; |
| } |
| } |
| } |
| |
| if (makecopy != 0) { |
| if (ptr != NULL) { |
| KFREES(ptr, fp->fr_dsize); |
| } |
| KFREES(fp, fp->fr_size); |
| } |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| return error; |
| } |
| |
| if (!f) { |
| /* |
| * At the end of this, ftail must point to the place where the |
| * new rule is to be saved/inserted/added. |
| * For SIOCAD*FR, this should be the last rule in the group of |
| * rules that have equal fr_collect fields. |
| * For SIOCIN*FR, ... |
| */ |
| if (req == (ioctlcmd_t)SIOCADAFR || |
| req == (ioctlcmd_t)SIOCADIFR) { |
| |
| for (ftail = fprev; (f = *ftail) != NULL; ) { |
| if (f->fr_collect > fp->fr_collect) |
| break; |
| ftail = &f->fr_next; |
| } |
| f = NULL; |
| ptr = NULL; |
| } else if (req == (ioctlcmd_t)SIOCINAFR || |
| req == (ioctlcmd_t)SIOCINIFR) { |
| while ((f = *fprev) != NULL) { |
| if (f->fr_collect >= fp->fr_collect) |
| break; |
| fprev = &f->fr_next; |
| } |
| ftail = fprev; |
| if (fp->fr_hits != 0) { |
| while (fp->fr_hits && (f = *ftail)) { |
| if (f->fr_collect != fp->fr_collect) |
| break; |
| fprev = ftail; |
| ftail = &f->fr_next; |
| fp->fr_hits--; |
| } |
| } |
| f = NULL; |
| ptr = NULL; |
| } |
| } |
| |
| /* |
| * Request to remove a rule. |
| */ |
| if (addrem == 1) { |
| if (!f) { |
| softc->ipf_interror = 29; |
| error = ESRCH; |
| } else { |
| /* |
| * Do not allow activity from user space to interfere |
| * with rules not loaded that way. |
| */ |
| if ((makecopy == 1) && !(f->fr_flags & FR_COPIED)) { |
| softc->ipf_interror = 30; |
| error = EPERM; |
| goto done; |
| } |
| |
| /* |
| * Return EBUSY if the rule is being reference by |
| * something else (eg state information.) |
| */ |
| if (f->fr_ref > 1) { |
| softc->ipf_interror = 31; |
| error = EBUSY; |
| goto done; |
| } |
| #ifdef IPFILTER_SCAN |
| if (f->fr_isctag != -1 && |
| (f->fr_isc != (struct ipscan *)-1)) |
| ipf_scan_detachfr(f); |
| #endif |
| |
| need_free = 1; |
| if (unit == IPL_LOGAUTH) { |
| error = ipf_auth_precmd(softc, req, f, ftail); |
| goto done; |
| } |
| |
| ipf_rule_delete(softc, f, unit, set); |
| |
| if (makecopy != 0) { |
| if (ptr != NULL) { |
| KFREES(ptr, fp->fr_dsize); |
| } |
| KFREES(fp, fp->fr_size); |
| fp = NULL; |
| } |
| } |
| } else { |
| /* |
| * Not removing, so we must be adding/inserting a rule. |
| */ |
| if (f != NULL) { |
| softc->ipf_interror = 32; |
| error = EEXIST; |
| goto done; |
| } |
| if (unit == IPL_LOGAUTH) { |
| error = ipf_auth_precmd(softc, req, fp, ftail); |
| goto done; |
| } |
| |
| MUTEX_NUKE(&fp->fr_lock); |
| MUTEX_INIT(&fp->fr_lock, "filter rule lock"); |
| if (fp->fr_die != 0) |
| ipf_rule_expire_insert(softc, fp, set); |
| |
| fp->fr_hits = 0; |
| if (makecopy != 0) |
| fp->fr_ref = 1; |
| fp->fr_pnext = ftail; |
| fp->fr_next = *ftail; |
| *ftail = fp; |
| if (addrem == 0) |
| ipf_fixskip(ftail, fp, 1); |
| |
| fp->fr_icmpgrp = NULL; |
| if (fp->fr_icmphead != -1) { |
| group = FR_NAME(fp, fr_icmphead); |
| fg = ipf_group_add(softc, group, fp, 0, unit, set); |
| if (fg != NULL) |
| fp->fr_icmpgrp = &fg->fg_start; |
| } |
| |
| fp->fr_grp = NULL; |
| if (fp->fr_grhead != -1) { |
| group = FR_NAME(fp, fr_grhead); |
| fg = ipf_group_add(softc, group, fp, fp->fr_flags, |
| unit, set); |
| if (fg != NULL) |
| fp->fr_grp = &fg->fg_start; |
| } |
| } |
| done: |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| donenolock: |
| if (need_free || (error != 0)) { |
| if ((fp->fr_type & ~FR_T_BUILTIN) == FR_T_IPF) { |
| if ((fp->fr_satype == FRI_LOOKUP) && |
| (fp->fr_srcptr != NULL)) |
| ipf_lookup_deref(softc, fp->fr_srctype, |
| fp->fr_srcptr); |
| if ((fp->fr_datype == FRI_LOOKUP) && |
| (fp->fr_dstptr != NULL)) |
| ipf_lookup_deref(softc, fp->fr_dsttype, |
| fp->fr_dstptr); |
| } |
| if ((ptr != NULL) && (makecopy != 0)) { |
| KFREES(ptr, fp->fr_dsize); |
| } |
| KFREES(fp, fp->fr_size); |
| } |
| return (error); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_rule_delete */ |
| /* Returns: Nil */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* f(I) - pointer to the rule being deleted */ |
| /* ftail(I) - pointer to the pointer to f */ |
| /* unit(I) - device for which this is for */ |
| /* set(I) - 1 or 0 (filter set) */ |
| /* */ |
| /* This function attempts to do what it can to delete a filter rule: remove */ |
| /* it from any linked lists and remove any groups it is responsible for. */ |
| /* But in the end, removing a rule can only drop the reference count - we */ |
| /* must use that as the guide for whether or not it can be freed. */ |
| /* ------------------------------------------------------------------------ */ |
| static void |
| ipf_rule_delete(softc, f, unit, set) |
| ipf_main_softc_t *softc; |
| frentry_t *f; |
| int unit, set; |
| { |
| |
| if (f->fr_grhead != -1) |
| ipf_group_del(softc, FR_NAME(f, fr_grhead), unit, set); |
| |
| if (f->fr_icmphead != -1) |
| ipf_group_del(softc, FR_NAME(f, fr_icmphead), unit, set); |
| |
| /* |
| * If fr_pdnext is set, then the rule is on the expire list, so |
| * remove it from there. |
| */ |
| if (f->fr_pdnext != NULL) { |
| *f->fr_pdnext = f->fr_dnext; |
| if (f->fr_dnext != NULL) |
| f->fr_dnext->fr_pdnext = f->fr_pdnext; |
| f->fr_pdnext = NULL; |
| f->fr_dnext = NULL; |
| } |
| |
| ipf_fixskip(f->fr_pnext, f, -1); |
| if (f->fr_pnext != NULL) |
| *f->fr_pnext = f->fr_next; |
| if (f->fr_next != NULL) |
| f->fr_next->fr_pnext = f->fr_pnext; |
| f->fr_pnext = NULL; |
| f->fr_next = NULL; |
| |
| (void) ipf_derefrule(softc, &f); |
| } |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_rule_expire_insert */ |
| /* Returns: Nil */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* f(I) - pointer to rule to be added to expire list */ |
| /* set(I) - 1 or 0 (filter set) */ |
| /* */ |
| /* If the new rule has a given expiration time, insert it into the list of */ |
| /* expiring rules with the ones to be removed first added to the front of */ |
| /* the list. The insertion is O(n) but it is kept sorted for quick scans at */ |
| /* expiration interval checks. */ |
| /* ------------------------------------------------------------------------ */ |
| static void |
| ipf_rule_expire_insert(softc, f, set) |
| ipf_main_softc_t *softc; |
| frentry_t *f; |
| int set; |
| { |
| frentry_t *fr; |
| |
| /* |
| */ |
| |
| f->fr_die = softc->ipf_ticks + IPF_TTLVAL(f->fr_die); |
| for (fr = softc->ipf_rule_explist[set]; fr != NULL; |
| fr = fr->fr_dnext) { |
| if (f->fr_die < fr->fr_die) |
| break; |
| if (fr->fr_dnext == NULL) { |
| /* |
| * We've got to the last rule and everything |
| * wanted to be expired before this new node, |
| * so we have to tack it on the end... |
| */ |
| fr->fr_dnext = f; |
| f->fr_pdnext = &fr->fr_dnext; |
| fr = NULL; |
| break; |
| } |
| } |
| |
| if (softc->ipf_rule_explist[set] == NULL) { |
| softc->ipf_rule_explist[set] = f; |
| f->fr_pdnext = &softc->ipf_rule_explist[set]; |
| } else if (fr != NULL) { |
| f->fr_dnext = fr; |
| f->fr_pdnext = fr->fr_pdnext; |
| fr->fr_pdnext = &f->fr_dnext; |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_findlookup */ |
| /* Returns: NULL = failure, else success */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* unit(I) - ipf device we want to find match for */ |
| /* fp(I) - rule for which lookup is for */ |
| /* addrp(I) - pointer to lookup information in address struct */ |
| /* */ |
| /* When using pools and hash tables to store addresses for matching in */ |
| /* rules, it is necessary to resolve both the object referred to by the */ |
| /* name or address (and return that pointer) and also provide the means by */ |
| /* which to determine if an address belongs to that object to make the */ |
| /* packet matching quicker. */ |
| /* ------------------------------------------------------------------------ */ |
| static void * |
| ipf_findlookup(softc, unit, fr, addrp) |
| ipf_main_softc_t *softc; |
| int unit; |
| frentry_t *fr; |
| i6addr_t *addrp; |
| { |
| void *ptr = NULL; |
| |
| switch (addrp->iplookupsubtype) |
| { |
| case 0 : |
| ptr = ipf_lookup_res_num(softc, unit, addrp->iplookuptype, |
| addrp->iplookupnum, |
| &addrp->iplookupfunc); |
| break; |
| case 1 : |
| if (addrp->iplookupname < 0) |
| break; |
| if (addrp->iplookupname >= fr->fr_namelen) |
| break; |
| ptr = ipf_lookup_res_name(softc, unit, addrp->iplookuptype, |
| fr->fr_names + addrp->iplookupname, |
| &addrp->iplookupfunc); |
| break; |
| default : |
| break; |
| } |
| |
| return ptr; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_funcinit */ |
| /* Returns: int - 0 == success, else ESRCH: cannot resolve rule details */ |
| /* Parameters: fr(I) - pointer to filter rule */ |
| /* */ |
| /* If a rule is a call rule, then check if the function it points to needs */ |
| /* an init function to be called now the rule has been loaded. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_funcinit(softc, fr) |
| ipf_main_softc_t *softc; |
| frentry_t *fr; |
| { |
| ipfunc_resolve_t *ft; |
| int err; |
| |
| softc->ipf_interror = 34; |
| err = ESRCH; |
| |
| for (ft = ipf_availfuncs; ft->ipfu_addr != NULL; ft++) |
| if (ft->ipfu_addr == fr->fr_func) { |
| err = 0; |
| if (ft->ipfu_init != NULL) |
| err = (*ft->ipfu_init)(softc, fr); |
| break; |
| } |
| return err; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_findfunc */ |
| /* Returns: ipfunc_t - pointer to function if found, else NULL */ |
| /* Parameters: funcptr(I) - function pointer to lookup */ |
| /* */ |
| /* Look for a function in the table of known functions. */ |
| /* ------------------------------------------------------------------------ */ |
| static ipfunc_t |
| ipf_findfunc(funcptr) |
| ipfunc_t funcptr; |
| { |
| ipfunc_resolve_t *ft; |
| |
| for (ft = ipf_availfuncs; ft->ipfu_addr != NULL; ft++) |
| if (ft->ipfu_addr == funcptr) |
| return funcptr; |
| return NULL; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_resolvefunc */ |
| /* Returns: int - 0 == success, else error */ |
| /* Parameters: data(IO) - ioctl data pointer to ipfunc_resolve_t struct */ |
| /* */ |
| /* Copy in a ipfunc_resolve_t structure and then fill in the missing field. */ |
| /* This will either be the function name (if the pointer is set) or the */ |
| /* function pointer if the name is set. When found, fill in the other one */ |
| /* so that the entire, complete, structure can be copied back to user space.*/ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_resolvefunc(softc, data) |
| ipf_main_softc_t *softc; |
| void *data; |
| { |
| ipfunc_resolve_t res, *ft; |
| int error; |
| |
| error = BCOPYIN(data, &res, sizeof(res)); |
| if (error != 0) { |
| softc->ipf_interror = 123; |
| return EFAULT; |
| } |
| |
| if (res.ipfu_addr == NULL && res.ipfu_name[0] != '\0') { |
| for (ft = ipf_availfuncs; ft->ipfu_addr != NULL; ft++) |
| if (strncmp(res.ipfu_name, ft->ipfu_name, |
| sizeof(res.ipfu_name)) == 0) { |
| res.ipfu_addr = ft->ipfu_addr; |
| res.ipfu_init = ft->ipfu_init; |
| if (COPYOUT(&res, data, sizeof(res)) != 0) { |
| softc->ipf_interror = 35; |
| return EFAULT; |
| } |
| return 0; |
| } |
| } |
| if (res.ipfu_addr != NULL && res.ipfu_name[0] == '\0') { |
| for (ft = ipf_availfuncs; ft->ipfu_addr != NULL; ft++) |
| if (ft->ipfu_addr == res.ipfu_addr) { |
| (void) strncpy(res.ipfu_name, ft->ipfu_name, |
| sizeof(res.ipfu_name)); |
| res.ipfu_init = ft->ipfu_init; |
| if (COPYOUT(&res, data, sizeof(res)) != 0) { |
| softc->ipf_interror = 36; |
| return EFAULT; |
| } |
| return 0; |
| } |
| } |
| softc->ipf_interror = 37; |
| return ESRCH; |
| } |
| |
| |
| #if !defined(_KERNEL) || (!defined(__NetBSD__) && !defined(__OpenBSD__) && \ |
| !defined(__FreeBSD__)) || \ |
| FREEBSD_LT_REV(501000) || NETBSD_LT_REV(105000000) || \ |
| OPENBSD_LT_REV(200006) |
| /* |
| * From: NetBSD |
| * ppsratecheck(): packets (or events) per second limitation. |
| */ |
| int |
| ppsratecheck(lasttime, curpps, maxpps) |
| struct timeval *lasttime; |
| int *curpps; |
| int maxpps; /* maximum pps allowed */ |
| { |
| struct timeval tv, delta; |
| int rv; |
| |
| GETKTIME(&tv); |
| |
| delta.tv_sec = tv.tv_sec - lasttime->tv_sec; |
| delta.tv_usec = tv.tv_usec - lasttime->tv_usec; |
| if (delta.tv_usec < 0) { |
| delta.tv_sec--; |
| delta.tv_usec += 1000000; |
| } |
| |
| /* |
| * check for 0,0 is so that the message will be seen at least once. |
| * if more than one second have passed since the last update of |
| * lasttime, reset the counter. |
| * |
| * we do increment *curpps even in *curpps < maxpps case, as some may |
| * try to use *curpps for stat purposes as well. |
| */ |
| if ((lasttime->tv_sec == 0 && lasttime->tv_usec == 0) || |
| delta.tv_sec >= 1) { |
| *lasttime = tv; |
| *curpps = 0; |
| rv = 1; |
| } else if (maxpps < 0) |
| rv = 1; |
| else if (*curpps < maxpps) |
| rv = 1; |
| else |
| rv = 0; |
| *curpps = *curpps + 1; |
| |
| return (rv); |
| } |
| #endif |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_derefrule */ |
| /* Returns: int - 0 == rule freed up, else rule not freed */ |
| /* Parameters: fr(I) - pointer to filter rule */ |
| /* */ |
| /* Decrement the reference counter to a rule by one. If it reaches zero, */ |
| /* free it and any associated storage space being used by it. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_derefrule(softc, frp) |
| ipf_main_softc_t *softc; |
| frentry_t **frp; |
| { |
| frentry_t *fr; |
| |
| fr = *frp; |
| *frp = NULL; |
| |
| MUTEX_ENTER(&fr->fr_lock); |
| fr->fr_ref--; |
| if (fr->fr_ref == 0) { |
| MUTEX_EXIT(&fr->fr_lock); |
| MUTEX_DESTROY(&fr->fr_lock); |
| |
| if ((fr->fr_type & ~FR_T_BUILTIN) == FR_T_IPF && |
| fr->fr_satype == FRI_LOOKUP) |
| ipf_lookup_deref(softc, fr->fr_srctype, fr->fr_srcptr); |
| if ((fr->fr_type & ~FR_T_BUILTIN) == FR_T_IPF && |
| fr->fr_datype == FRI_LOOKUP) |
| ipf_lookup_deref(softc, fr->fr_dsttype, fr->fr_dstptr); |
| |
| if ((fr->fr_flags & FR_COPIED) != 0) { |
| if (fr->fr_dsize) { |
| KFREES(fr->fr_data, fr->fr_dsize); |
| } |
| KFREES(fr, fr->fr_size); |
| return 0; |
| } |
| return 1; |
| } else { |
| MUTEX_EXIT(&fr->fr_lock); |
| } |
| return -1; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_grpmapinit */ |
| /* Returns: int - 0 == success, else ESRCH because table entry not found*/ |
| /* Parameters: fr(I) - pointer to rule to find hash table for */ |
| /* */ |
| /* Looks for group hash table fr_arg and stores a pointer to it in fr_ptr. */ |
| /* fr_ptr is later used by ipf_srcgrpmap and ipf_dstgrpmap. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_grpmapinit(softc, fr) |
| ipf_main_softc_t *softc; |
| frentry_t *fr; |
| { |
| char name[FR_GROUPLEN]; |
| iphtable_t *iph; |
| |
| #if defined(SNPRINTF) && defined(_KERNEL) |
| SNPRINTF(name, sizeof(name), "%d", fr->fr_arg); |
| #else |
| (void) sprintf(name, "%d", fr->fr_arg); |
| #endif |
| iph = ipf_lookup_find_htable(softc, IPL_LOGIPF, name); |
| if (iph == NULL) { |
| softc->ipf_interror = 38; |
| return ESRCH; |
| } |
| if ((iph->iph_flags & FR_INOUT) != (fr->fr_flags & FR_INOUT)) { |
| softc->ipf_interror = 39; |
| return ESRCH; |
| } |
| iph->iph_ref++; |
| fr->fr_ptr = iph; |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_srcgrpmap */ |
| /* Returns: frentry_t * - pointer to "new last matching" rule or NULL */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* passp(IO) - pointer to current/new filter decision (unused) */ |
| /* */ |
| /* Look for a rule group head in a hash table, using the source address as */ |
| /* the key, and descend into that group and continue matching rules against */ |
| /* the packet. */ |
| /* ------------------------------------------------------------------------ */ |
| frentry_t * |
| ipf_srcgrpmap(fin, passp) |
| fr_info_t *fin; |
| u_32_t *passp; |
| { |
| frgroup_t *fg; |
| void *rval; |
| |
| rval = ipf_iphmfindgroup(fin->fin_main_soft, fin->fin_fr->fr_ptr, |
| &fin->fin_src); |
| if (rval == NULL) |
| return NULL; |
| |
| fg = rval; |
| fin->fin_fr = fg->fg_start; |
| (void) ipf_scanlist(fin, *passp); |
| return fin->fin_fr; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_dstgrpmap */ |
| /* Returns: frentry_t * - pointer to "new last matching" rule or NULL */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* passp(IO) - pointer to current/new filter decision (unused) */ |
| /* */ |
| /* Look for a rule group head in a hash table, using the destination */ |
| /* address as the key, and descend into that group and continue matching */ |
| /* rules against the packet. */ |
| /* ------------------------------------------------------------------------ */ |
| frentry_t * |
| ipf_dstgrpmap(fin, passp) |
| fr_info_t *fin; |
| u_32_t *passp; |
| { |
| frgroup_t *fg; |
| void *rval; |
| |
| rval = ipf_iphmfindgroup(fin->fin_main_soft, fin->fin_fr->fr_ptr, |
| &fin->fin_dst); |
| if (rval == NULL) |
| return NULL; |
| |
| fg = rval; |
| fin->fin_fr = fg->fg_start; |
| (void) ipf_scanlist(fin, *passp); |
| return fin->fin_fr; |
| } |
| |
| /* |
| * Queue functions |
| * =============== |
| * These functions manage objects on queues for efficient timeouts. There |
| * are a number of system defined queues as well as user defined timeouts. |
| * It is expected that a lock is held in the domain in which the queue |
| * belongs (i.e. either state or NAT) when calling any of these functions |
| * that prevents ipf_freetimeoutqueue() from being called at the same time |
| * as any other. |
| */ |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_addtimeoutqueue */ |
| /* Returns: struct ifqtq * - NULL if malloc fails, else pointer to */ |
| /* timeout queue with given interval. */ |
| /* Parameters: parent(I) - pointer to pointer to parent node of this list */ |
| /* of interface queues. */ |
| /* seconds(I) - timeout value in seconds for this queue. */ |
| /* */ |
| /* This routine first looks for a timeout queue that matches the interval */ |
| /* being requested. If it finds one, increments the reference counter and */ |
| /* returns a pointer to it. If none are found, it allocates a new one and */ |
| /* inserts it at the top of the list. */ |
| /* */ |
| /* Locking. */ |
| /* It is assumed that the caller of this function has an appropriate lock */ |
| /* held (exclusively) in the domain that encompases 'parent'. */ |
| /* ------------------------------------------------------------------------ */ |
| ipftq_t * |
| ipf_addtimeoutqueue(softc, parent, seconds) |
| ipf_main_softc_t *softc; |
| ipftq_t **parent; |
| u_int seconds; |
| { |
| ipftq_t *ifq; |
| u_int period; |
| |
| period = seconds * IPF_HZ_DIVIDE; |
| |
| MUTEX_ENTER(&softc->ipf_timeoutlock); |
| for (ifq = *parent; ifq != NULL; ifq = ifq->ifq_next) { |
| if (ifq->ifq_ttl == period) { |
| /* |
| * Reset the delete flag, if set, so the structure |
| * gets reused rather than freed and reallocated. |
| */ |
| MUTEX_ENTER(&ifq->ifq_lock); |
| ifq->ifq_flags &= ~IFQF_DELETE; |
| ifq->ifq_ref++; |
| MUTEX_EXIT(&ifq->ifq_lock); |
| MUTEX_EXIT(&softc->ipf_timeoutlock); |
| |
| return ifq; |
| } |
| } |
| |
| KMALLOC(ifq, ipftq_t *); |
| if (ifq != NULL) { |
| MUTEX_NUKE(&ifq->ifq_lock); |
| IPFTQ_INIT(ifq, period, "ipftq mutex"); |
| ifq->ifq_next = *parent; |
| ifq->ifq_pnext = parent; |
| ifq->ifq_flags = IFQF_USER; |
| ifq->ifq_ref++; |
| *parent = ifq; |
| softc->ipf_userifqs++; |
| } |
| MUTEX_EXIT(&softc->ipf_timeoutlock); |
| return ifq; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_deletetimeoutqueue */ |
| /* Returns: int - new reference count value of the timeout queue */ |
| /* Parameters: ifq(I) - timeout queue which is losing a reference. */ |
| /* Locks: ifq->ifq_lock */ |
| /* */ |
| /* This routine must be called when we're discarding a pointer to a timeout */ |
| /* queue object, taking care of the reference counter. */ |
| /* */ |
| /* Now that this just sets a DELETE flag, it requires the expire code to */ |
| /* check the list of user defined timeout queues and call the free function */ |
| /* below (currently commented out) to stop memory leaking. It is done this */ |
| /* way because the locking may not be sufficient to safely do a free when */ |
| /* this function is called. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_deletetimeoutqueue(ifq) |
| ipftq_t *ifq; |
| { |
| |
| ifq->ifq_ref--; |
| if ((ifq->ifq_ref == 0) && ((ifq->ifq_flags & IFQF_USER) != 0)) { |
| ifq->ifq_flags |= IFQF_DELETE; |
| } |
| |
| return ifq->ifq_ref; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_freetimeoutqueue */ |
| /* Parameters: ifq(I) - timeout queue which is losing a reference. */ |
| /* Returns: Nil */ |
| /* */ |
| /* Locking: */ |
| /* It is assumed that the caller of this function has an appropriate lock */ |
| /* held (exclusively) in the domain that encompases the callers "domain". */ |
| /* The ifq_lock for this structure should not be held. */ |
| /* */ |
| /* Remove a user defined timeout queue from the list of queues it is in and */ |
| /* tidy up after this is done. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_freetimeoutqueue(softc, ifq) |
| ipf_main_softc_t *softc; |
| ipftq_t *ifq; |
| { |
| |
| if (((ifq->ifq_flags & IFQF_DELETE) == 0) || (ifq->ifq_ref != 0) || |
| ((ifq->ifq_flags & IFQF_USER) == 0)) { |
| printf("ipf_freetimeoutqueue(%lx) flags 0x%x ttl %d ref %d\n", |
| (u_long)ifq, ifq->ifq_flags, ifq->ifq_ttl, |
| ifq->ifq_ref); |
| return; |
| } |
| |
| /* |
| * Remove from its position in the list. |
| */ |
| *ifq->ifq_pnext = ifq->ifq_next; |
| if (ifq->ifq_next != NULL) |
| ifq->ifq_next->ifq_pnext = ifq->ifq_pnext; |
| ifq->ifq_next = NULL; |
| ifq->ifq_pnext = NULL; |
| |
| MUTEX_DESTROY(&ifq->ifq_lock); |
| ATOMIC_DEC(softc->ipf_userifqs); |
| KFREE(ifq); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_deletequeueentry */ |
| /* Returns: Nil */ |
| /* Parameters: tqe(I) - timeout queue entry to delete */ |
| /* */ |
| /* Remove a tail queue entry from its queue and make it an orphan. */ |
| /* ipf_deletetimeoutqueue is called to make sure the reference count on the */ |
| /* queue is correct. We can't, however, call ipf_freetimeoutqueue because */ |
| /* the correct lock(s) may not be held that would make it safe to do so. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_deletequeueentry(tqe) |
| ipftqent_t *tqe; |
| { |
| ipftq_t *ifq; |
| |
| ifq = tqe->tqe_ifq; |
| |
| MUTEX_ENTER(&ifq->ifq_lock); |
| |
| if (tqe->tqe_pnext != NULL) { |
| *tqe->tqe_pnext = tqe->tqe_next; |
| if (tqe->tqe_next != NULL) |
| tqe->tqe_next->tqe_pnext = tqe->tqe_pnext; |
| else /* we must be the tail anyway */ |
| ifq->ifq_tail = tqe->tqe_pnext; |
| |
| tqe->tqe_pnext = NULL; |
| tqe->tqe_ifq = NULL; |
| } |
| |
| (void) ipf_deletetimeoutqueue(ifq); |
| ASSERT(ifq->ifq_ref > 0); |
| |
| MUTEX_EXIT(&ifq->ifq_lock); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_queuefront */ |
| /* Returns: Nil */ |
| /* Parameters: tqe(I) - pointer to timeout queue entry */ |
| /* */ |
| /* Move a queue entry to the front of the queue, if it isn't already there. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_queuefront(tqe) |
| ipftqent_t *tqe; |
| { |
| ipftq_t *ifq; |
| |
| ifq = tqe->tqe_ifq; |
| if (ifq == NULL) |
| return; |
| |
| MUTEX_ENTER(&ifq->ifq_lock); |
| if (ifq->ifq_head != tqe) { |
| *tqe->tqe_pnext = tqe->tqe_next; |
| if (tqe->tqe_next) |
| tqe->tqe_next->tqe_pnext = tqe->tqe_pnext; |
| else |
| ifq->ifq_tail = tqe->tqe_pnext; |
| |
| tqe->tqe_next = ifq->ifq_head; |
| ifq->ifq_head->tqe_pnext = &tqe->tqe_next; |
| ifq->ifq_head = tqe; |
| tqe->tqe_pnext = &ifq->ifq_head; |
| } |
| MUTEX_EXIT(&ifq->ifq_lock); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_queueback */ |
| /* Returns: Nil */ |
| /* Parameters: ticks(I) - ipf tick time to use with this call */ |
| /* tqe(I) - pointer to timeout queue entry */ |
| /* */ |
| /* Move a queue entry to the back of the queue, if it isn't already there. */ |
| /* We use use ticks to calculate the expiration and mark for when we last */ |
| /* touched the structure. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_queueback(ticks, tqe) |
| u_long ticks; |
| ipftqent_t *tqe; |
| { |
| ipftq_t *ifq; |
| |
| ifq = tqe->tqe_ifq; |
| if (ifq == NULL) |
| return; |
| tqe->tqe_die = ticks + ifq->ifq_ttl; |
| tqe->tqe_touched = ticks; |
| |
| MUTEX_ENTER(&ifq->ifq_lock); |
| if (tqe->tqe_next != NULL) { /* at the end already ? */ |
| /* |
| * Remove from list |
| */ |
| *tqe->tqe_pnext = tqe->tqe_next; |
| tqe->tqe_next->tqe_pnext = tqe->tqe_pnext; |
| |
| /* |
| * Make it the last entry. |
| */ |
| tqe->tqe_next = NULL; |
| tqe->tqe_pnext = ifq->ifq_tail; |
| *ifq->ifq_tail = tqe; |
| ifq->ifq_tail = &tqe->tqe_next; |
| } |
| MUTEX_EXIT(&ifq->ifq_lock); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_queueappend */ |
| /* Returns: Nil */ |
| /* Parameters: ticks(I) - ipf tick time to use with this call */ |
| /* tqe(I) - pointer to timeout queue entry */ |
| /* ifq(I) - pointer to timeout queue */ |
| /* parent(I) - owing object pointer */ |
| /* */ |
| /* Add a new item to this queue and put it on the very end. */ |
| /* We use use ticks to calculate the expiration and mark for when we last */ |
| /* touched the structure. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_queueappend(ticks, tqe, ifq, parent) |
| u_long ticks; |
| ipftqent_t *tqe; |
| ipftq_t *ifq; |
| void *parent; |
| { |
| |
| MUTEX_ENTER(&ifq->ifq_lock); |
| tqe->tqe_parent = parent; |
| tqe->tqe_pnext = ifq->ifq_tail; |
| *ifq->ifq_tail = tqe; |
| ifq->ifq_tail = &tqe->tqe_next; |
| tqe->tqe_next = NULL; |
| tqe->tqe_ifq = ifq; |
| tqe->tqe_die = ticks + ifq->ifq_ttl; |
| tqe->tqe_touched = ticks; |
| ifq->ifq_ref++; |
| MUTEX_EXIT(&ifq->ifq_lock); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_movequeue */ |
| /* Returns: Nil */ |
| /* Parameters: tq(I) - pointer to timeout queue information */ |
| /* oifp(I) - old timeout queue entry was on */ |
| /* nifp(I) - new timeout queue to put entry on */ |
| /* */ |
| /* Move a queue entry from one timeout queue to another timeout queue. */ |
| /* If it notices that the current entry is already last and does not need */ |
| /* to move queue, the return. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_movequeue(ticks, tqe, oifq, nifq) |
| u_long ticks; |
| ipftqent_t *tqe; |
| ipftq_t *oifq, *nifq; |
| { |
| |
| /* |
| * If the queue hasn't changed and we last touched this entry at the |
| * same ipf time, then we're not going to achieve anything by either |
| * changing the ttl or moving it on the queue. |
| */ |
| if (oifq == nifq && tqe->tqe_touched == ticks) |
| return; |
| |
| /* |
| * For any of this to be outside the lock, there is a risk that two |
| * packets entering simultaneously, with one changing to a different |
| * queue and one not, could end up with things in a bizarre state. |
| */ |
| MUTEX_ENTER(&oifq->ifq_lock); |
| |
| tqe->tqe_touched = ticks; |
| tqe->tqe_die = ticks + nifq->ifq_ttl; |
| /* |
| * Is the operation here going to be a no-op ? |
| */ |
| if (oifq == nifq) { |
| if ((tqe->tqe_next == NULL) || |
| (tqe->tqe_next->tqe_die == tqe->tqe_die)) { |
| MUTEX_EXIT(&oifq->ifq_lock); |
| return; |
| } |
| } |
| |
| /* |
| * Remove from the old queue |
| */ |
| *tqe->tqe_pnext = tqe->tqe_next; |
| if (tqe->tqe_next) |
| tqe->tqe_next->tqe_pnext = tqe->tqe_pnext; |
| else |
| oifq->ifq_tail = tqe->tqe_pnext; |
| tqe->tqe_next = NULL; |
| |
| /* |
| * If we're moving from one queue to another, release the |
| * lock on the old queue and get a lock on the new queue. |
| * For user defined queues, if we're moving off it, call |
| * delete in case it can now be freed. |
| */ |
| if (oifq != nifq) { |
| tqe->tqe_ifq = NULL; |
| |
| (void) ipf_deletetimeoutqueue(oifq); |
| |
| MUTEX_EXIT(&oifq->ifq_lock); |
| |
| MUTEX_ENTER(&nifq->ifq_lock); |
| |
| tqe->tqe_ifq = nifq; |
| nifq->ifq_ref++; |
| } |
| |
| /* |
| * Add to the bottom of the new queue |
| */ |
| tqe->tqe_pnext = nifq->ifq_tail; |
| *nifq->ifq_tail = tqe; |
| nifq->ifq_tail = &tqe->tqe_next; |
| MUTEX_EXIT(&nifq->ifq_lock); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_updateipid */ |
| /* Returns: int - 0 == success, -1 == error (packet should be droppped) */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* When we are doing NAT, change the IP of every packet to represent a */ |
| /* single sequence of packets coming from the host, hiding any host */ |
| /* specific sequencing that might otherwise be revealed. If the packet is */ |
| /* a fragment, then store the 'new' IPid in the fragment cache and look up */ |
| /* the fragment cache for non-leading fragments. If a non-leading fragment */ |
| /* has no match in the cache, return an error. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_updateipid(fin) |
| fr_info_t *fin; |
| { |
| u_short id, ido, sums; |
| u_32_t sumd, sum; |
| ip_t *ip; |
| |
| if (fin->fin_off != 0) { |
| sum = ipf_frag_ipidknown(fin); |
| if (sum == 0xffffffff) |
| return -1; |
| sum &= 0xffff; |
| id = (u_short)sum; |
| } else { |
| id = ipf_nextipid(fin); |
| if (fin->fin_off == 0 && (fin->fin_flx & FI_FRAG) != 0) |
| (void) ipf_frag_ipidnew(fin, (u_32_t)id); |
| } |
| |
| ip = fin->fin_ip; |
| ido = ntohs(ip->ip_id); |
| if (id == ido) |
| return 0; |
| ip->ip_id = htons(id); |
| CALC_SUMD(ido, id, sumd); /* DESTRUCTIVE MACRO! id,ido change */ |
| sum = (~ntohs(ip->ip_sum)) & 0xffff; |
| sum += sumd; |
| sum = (sum >> 16) + (sum & 0xffff); |
| sum = (sum >> 16) + (sum & 0xffff); |
| sums = ~(u_short)sum; |
| ip->ip_sum = htons(sums); |
| return 0; |
| } |
| |
| |
| #ifdef NEED_FRGETIFNAME |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_getifname */ |
| /* Returns: char * - pointer to interface name */ |
| /* Parameters: ifp(I) - pointer to network interface */ |
| /* buffer(O) - pointer to where to store interface name */ |
| /* */ |
| /* Constructs an interface name in the buffer passed. The buffer passed is */ |
| /* expected to be at least LIFNAMSIZ in bytes big. If buffer is passed in */ |
| /* as a NULL pointer then return a pointer to a static array. */ |
| /* ------------------------------------------------------------------------ */ |
| char * |
| ipf_getifname(ifp, buffer) |
| struct ifnet *ifp; |
| char *buffer; |
| { |
| static char namebuf[LIFNAMSIZ]; |
| # if defined(MENTAT) || defined(__FreeBSD__) || defined(__osf__) || \ |
| defined(__sgi) || defined(linux) || defined(_AIX51) || \ |
| (defined(sun) && !defined(__SVR4) && !defined(__svr4__)) |
| int unit, space; |
| char temp[20]; |
| char *s; |
| # endif |
| |
| if (buffer == NULL) |
| buffer = namebuf; |
| (void) strncpy(buffer, ifp->if_name, LIFNAMSIZ); |
| buffer[LIFNAMSIZ - 1] = '\0'; |
| # if defined(MENTAT) || defined(__FreeBSD__) || defined(__osf__) || \ |
| defined(__sgi) || defined(_AIX51) || \ |
| (defined(sun) && !defined(__SVR4) && !defined(__svr4__)) |
| for (s = buffer; *s; s++) |
| ; |
| unit = ifp->if_unit; |
| space = LIFNAMSIZ - (s - buffer); |
| if (space > 0) { |
| # if defined(SNPRINTF) && defined(_KERNEL) |
| SNPRINTF(temp, sizeof(temp), "%d", unit); |
| # else |
| (void) sprintf(temp, "%d", unit); |
| # endif |
| (void) strncpy(s, temp, space); |
| } |
| # endif |
| return buffer; |
| } |
| #endif |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ioctlswitch */ |
| /* Returns: int - -1 continue processing, else ioctl return value */ |
| /* Parameters: unit(I) - device unit opened */ |
| /* data(I) - pointer to ioctl data */ |
| /* cmd(I) - ioctl command */ |
| /* mode(I) - mode value */ |
| /* uid(I) - uid making the ioctl call */ |
| /* ctx(I) - pointer to context data */ |
| /* */ |
| /* Based on the value of unit, call the appropriate ioctl handler or return */ |
| /* EIO if ipfilter is not running. Also checks if write perms are req'd */ |
| /* for the device in order to execute the ioctl. A special case is made */ |
| /* SIOCIPFINTERROR so that the same code isn't required in every handler. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_ioctlswitch(softc, unit, data, cmd, mode, uid, ctx) |
| ipf_main_softc_t *softc; |
| int unit, mode, uid; |
| ioctlcmd_t cmd; |
| void *data, *ctx; |
| { |
| int error = 0; |
| |
| switch (cmd) |
| { |
| case SIOCIPFINTERROR : |
| error = BCOPYOUT(&softc->ipf_interror, data, |
| sizeof(softc->ipf_interror)); |
| if (error != 0) { |
| softc->ipf_interror = 40; |
| error = EFAULT; |
| } |
| return error; |
| default : |
| break; |
| } |
| |
| switch (unit) |
| { |
| case IPL_LOGIPF : |
| error = ipf_ipf_ioctl(softc, data, cmd, mode, uid, ctx); |
| break; |
| case IPL_LOGNAT : |
| if (softc->ipf_running > 0) { |
| error = ipf_nat_ioctl(softc, data, cmd, mode, |
| uid, ctx); |
| } else { |
| softc->ipf_interror = 42; |
| error = EIO; |
| } |
| break; |
| case IPL_LOGSTATE : |
| if (softc->ipf_running > 0) { |
| error = ipf_state_ioctl(softc, data, cmd, mode, |
| uid, ctx); |
| } else { |
| softc->ipf_interror = 43; |
| error = EIO; |
| } |
| break; |
| case IPL_LOGAUTH : |
| if (softc->ipf_running > 0) { |
| error = ipf_auth_ioctl(softc, data, cmd, mode, |
| uid, ctx); |
| } else { |
| softc->ipf_interror = 44; |
| error = EIO; |
| } |
| break; |
| case IPL_LOGSYNC : |
| if (softc->ipf_running > 0) { |
| error = ipf_sync_ioctl(softc, data, cmd, mode, |
| uid, ctx); |
| } else { |
| error = EIO; |
| softc->ipf_interror = 45; |
| } |
| break; |
| case IPL_LOGSCAN : |
| #ifdef IPFILTER_SCAN |
| if (softc->ipf_running > 0) |
| error = ipf_scan_ioctl(softc, data, cmd, mode, |
| uid, ctx); |
| else |
| #endif |
| { |
| error = EIO; |
| softc->ipf_interror = 46; |
| } |
| break; |
| case IPL_LOGLOOKUP : |
| if (softc->ipf_running > 0) { |
| error = ipf_lookup_ioctl(softc, data, cmd, mode, |
| uid, ctx); |
| } else { |
| error = EIO; |
| softc->ipf_interror = 47; |
| } |
| break; |
| default : |
| softc->ipf_interror = 48; |
| error = EIO; |
| break; |
| } |
| |
| return error; |
| } |
| |
| |
| /* |
| * This array defines the expected size of objects coming into the kernel |
| * for the various recognised object types. |
| */ |
| static int ipf_objbytes[IPFOBJ_COUNT][2] = { |
| { 1, sizeof(struct frentry) }, /* frentry */ |
| { 1, sizeof(struct friostat) }, |
| { 0, sizeof(struct fr_info) }, |
| { 0, sizeof(struct ipf_authstat) }, |
| { 0, sizeof(struct ipfrstat) }, |
| { 1, sizeof(struct ipnat) }, |
| { 0, sizeof(struct natstat) }, |
| { 0, sizeof(struct ipstate_save) }, |
| { 1, sizeof(struct nat_save) }, /* nat_save */ |
| { 0, sizeof(struct natlookup) }, |
| { 1, sizeof(struct ipstate) }, /* ipstate */ |
| { 0, sizeof(struct ips_stat) }, |
| { 0, sizeof(struct frauth) }, |
| { 0, sizeof(struct ipftune) }, |
| { 0, sizeof(struct nat) }, /* nat_t */ |
| { 0, sizeof(struct ipfruleiter) }, |
| { 0, sizeof(struct ipfgeniter) }, |
| { 0, sizeof(struct ipftable) }, |
| { 0, sizeof(struct ipflookupiter) }, |
| { 0, sizeof(struct ipftq) * IPF_TCP_NSTATES }, |
| }; |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_inobj */ |
| /* Returns: int - 0 = success, else failure */ |
| /* Parameters: data(I) - pointer to ioctl data */ |
| /* ptr(I) - pointer to store real data in */ |
| /* type(I) - type of structure being moved */ |
| /* */ |
| /* Copy in the contents of what the ipfobj_t points to. In future, we */ |
| /* add things to check for version numbers, sizes, etc, to make it backward */ |
| /* compatible at the ABI for user land. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_inobj(softc, data, ptr, type) |
| ipf_main_softc_t *softc; |
| void *data; |
| void *ptr; |
| int type; |
| { |
| ipfobj_t obj; |
| int error; |
| int size; |
| |
| if ((type < 0) || (type >= IPFOBJ_COUNT)) { |
| softc->ipf_interror = 49; |
| return EINVAL; |
| } |
| |
| error = BCOPYIN(data, &obj, sizeof(obj)); |
| if (error != 0) { |
| softc->ipf_interror = 124; |
| return EFAULT; |
| } |
| |
| if (obj.ipfo_type != type) { |
| softc->ipf_interror = 50; |
| return EINVAL; |
| } |
| |
| if (obj.ipfo_rev == IPFILTER_VERSION) { |
| if ((ipf_objbytes[type][0] & 1) != 0) { |
| if (obj.ipfo_size < ipf_objbytes[type][1]) { |
| softc->ipf_interror = 51; |
| return EINVAL; |
| } |
| size = ipf_objbytes[type][1]; |
| } else if (obj.ipfo_size == ipf_objbytes[type][1]) { |
| size = ipf_objbytes[type][1]; |
| } else { |
| softc->ipf_interror = 52; |
| return EINVAL; |
| } |
| error = COPYIN(obj.ipfo_ptr, ptr, size); |
| } else { |
| #ifdef IPFILTER_COMPAT |
| /* XXX compatibility hook here */ |
| /* error = hook(&obj, ptr) */ |
| softc->ipf_interror = 53; |
| return EINVAL; |
| #else |
| softc->ipf_interror = 54; |
| return EINVAL; |
| #endif |
| } |
| |
| if (error != 0) { |
| softc->ipf_interror = 55; |
| error = EFAULT; |
| } |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_inobjsz */ |
| /* Returns: int - 0 = success, else failure */ |
| /* Parameters: data(I) - pointer to ioctl data */ |
| /* ptr(I) - pointer to store real data in */ |
| /* type(I) - type of structure being moved */ |
| /* sz(I) - size of data to copy */ |
| /* */ |
| /* As per ipf_inobj, except the size of the object to copy in is passed in */ |
| /* but it must not be smaller than the size defined for the type and the */ |
| /* type must allow for varied sized objects. The extra requirement here is */ |
| /* that sz must match the size of the object being passed in - this is not */ |
| /* not possible nor required in ipf_inobj(). */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_inobjsz(softc, data, ptr, type, sz) |
| ipf_main_softc_t *softc; |
| void *data; |
| void *ptr; |
| int type, sz; |
| { |
| ipfobj_t obj; |
| int error; |
| |
| if ((type < 0) || (type >= IPFOBJ_COUNT)) { |
| softc->ipf_interror = 56; |
| return EINVAL; |
| } |
| |
| error = BCOPYIN(data, &obj, sizeof(obj)); |
| if (error != 0) { |
| softc->ipf_interror = 125; |
| return EFAULT; |
| } |
| |
| if (obj.ipfo_type != type) { |
| softc->ipf_interror = 58; |
| return EINVAL; |
| } |
| |
| if (obj.ipfo_rev == IPFILTER_VERSION) { |
| if (((ipf_objbytes[type][0] & 1) == 0) || |
| (sz < ipf_objbytes[type][1])) { |
| softc->ipf_interror = 57; |
| return EINVAL; |
| } |
| error = COPYIN(obj.ipfo_ptr, ptr, sz); |
| } else { |
| #ifdef IPFILTER_COMPAT |
| /* XXX compatibility hook here */ |
| /* error = hook(&obj, ptr) */ |
| softc->ipf_interror = 59; |
| return EINVAL; |
| #else |
| softc->ipf_interror = 60; |
| return EINVAL; |
| #endif |
| } |
| |
| if (error != 0) { |
| softc->ipf_interror = 61; |
| error = EFAULT; |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_outobjsz */ |
| /* Returns: int - 0 = success, else failure */ |
| /* Parameters: data(I) - pointer to ioctl data */ |
| /* ptr(I) - pointer to store real data in */ |
| /* type(I) - type of structure being moved */ |
| /* sz(I) - size of data to copy */ |
| /* */ |
| /* As per ipf_outobj, except the size of the object to copy out is passed in*/ |
| /* but it must not be smaller than the size defined for the type and the */ |
| /* type must allow for varied sized objects. The extra requirement here is */ |
| /* that sz must match the size of the object being passed in - this is not */ |
| /* not possible nor required in ipf_outobj(). */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_outobjsz(softc, data, ptr, type, sz) |
| ipf_main_softc_t *softc; |
| void *data; |
| void *ptr; |
| int type, sz; |
| { |
| ipfobj_t obj; |
| int error; |
| |
| if ((type < 0) || (type >= IPFOBJ_COUNT)) { |
| softc->ipf_interror = 62; |
| return EINVAL; |
| } |
| |
| error = BCOPYIN(data, &obj, sizeof(obj)); |
| if (error != 0) { |
| softc->ipf_interror = 127; |
| return EFAULT; |
| } |
| |
| if (obj.ipfo_type != type) { |
| softc->ipf_interror = 63; |
| return EINVAL; |
| } |
| |
| if (obj.ipfo_rev == IPFILTER_VERSION) { |
| if (((ipf_objbytes[type][0] & 1) == 0) || |
| (sz < ipf_objbytes[type][1])) { |
| softc->ipf_interror = 61; |
| return EINVAL; |
| } |
| error = COPYOUT(ptr, obj.ipfo_ptr, sz); |
| } else { |
| #ifdef IPFILTER_COMPAT |
| /* XXX compatibility hook here */ |
| /* error = hook(&obj, ptr) */ |
| softc->ipf_interror = 64; |
| return EINVAL; |
| #else |
| softc->ipf_interror = 65; |
| return EINVAL; |
| #endif |
| } |
| |
| if (error != 0) { |
| softc->ipf_interror = 66; |
| error = EFAULT; |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_outobj */ |
| /* Returns: int - 0 = success, else failure */ |
| /* Parameters: data(I) - pointer to ioctl data */ |
| /* ptr(I) - pointer to store real data in */ |
| /* type(I) - type of structure being moved */ |
| /* */ |
| /* Copy out the contents of what ptr is to where ipfobj points to. In */ |
| /* future, we add things to check for version numbers, sizes, etc, to make */ |
| /* it backward compatible at the ABI for user land. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_outobj(softc, data, ptr, type) |
| ipf_main_softc_t *softc; |
| void *data; |
| void *ptr; |
| int type; |
| { |
| ipfobj_t obj; |
| int error; |
| |
| if ((type < 0) || (type >= IPFOBJ_COUNT)) { |
| softc->ipf_interror = 67; |
| return EINVAL; |
| } |
| |
| error = BCOPYIN(data, &obj, sizeof(obj)); |
| if (error != 0) { |
| softc->ipf_interror = 126; |
| return EFAULT; |
| } |
| |
| if (obj.ipfo_type != type) { |
| softc->ipf_interror = 68; |
| return EINVAL; |
| } |
| |
| if (obj.ipfo_rev == IPFILTER_VERSION) { |
| if ((ipf_objbytes[type][0] & 1) != 0) { |
| if (obj.ipfo_size < ipf_objbytes[type][1]) { |
| softc->ipf_interror = 69; |
| return EINVAL; |
| } |
| } else if (obj.ipfo_size != ipf_objbytes[type][1]) { |
| softc->ipf_interror = 70; |
| return EINVAL; |
| } |
| |
| error = COPYOUT(ptr, obj.ipfo_ptr, obj.ipfo_size); |
| } else { |
| #ifdef IPFILTER_COMPAT |
| /* XXX compatibility hook here */ |
| /* error = hook(&obj, ptr, ipf_objbytes[type]); */ |
| softc->ipf_interror = 71; |
| return EINVAL; |
| #else |
| softc->ipf_interror = 72; |
| return EINVAL; |
| #endif |
| } |
| |
| if (error != 0) { |
| softc->ipf_interror = 73; |
| error = EFAULT; |
| } |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_checkl4sum */ |
| /* Returns: int - 0 = good, -1 = bad, 1 = cannot check */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* If possible, calculate the layer 4 checksum for the packet. If this is */ |
| /* not possible, return without indicating a failure or success but in a */ |
| /* way that is ditinguishable. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_checkl4sum(fin) |
| fr_info_t *fin; |
| { |
| u_short sum, hdrsum, *csump; |
| udphdr_t *udp; |
| int dosum; |
| |
| if ((fin->fin_flx & FI_NOCKSUM) != 0) |
| return 0; |
| |
| if (fin->fin_cksum == -1) |
| return -1; |
| |
| if (fin->fin_cksum == 1) |
| return 0; |
| |
| /* |
| * If the TCP packet isn't a fragment, isn't too short and otherwise |
| * isn't already considered "bad", then validate the checksum. If |
| * this check fails then considered the packet to be "bad". |
| */ |
| if ((fin->fin_flx & (FI_FRAG|FI_SHORT|FI_BAD)) != 0) |
| return 1; |
| |
| csump = NULL; |
| hdrsum = 0; |
| dosum = 0; |
| sum = 0; |
| |
| #if SOLARIS && defined(_KERNEL) && (SOLARIS2 >= 6) && defined(ICK_VALID) |
| if (dohwcksum && ((*fin->fin_mp)->b_ick_flag == ICK_VALID)) { |
| hdrsum = 0; |
| sum = 0; |
| } else { |
| #endif |
| switch (fin->fin_p) |
| { |
| case IPPROTO_TCP : |
| csump = &((tcphdr_t *)fin->fin_dp)->th_sum; |
| dosum = 1; |
| break; |
| |
| case IPPROTO_UDP : |
| udp = fin->fin_dp; |
| if (udp->uh_sum != 0) { |
| csump = &udp->uh_sum; |
| dosum = 1; |
| } |
| break; |
| |
| case IPPROTO_ICMP : |
| csump = &((struct icmp *)fin->fin_dp)->icmp_cksum; |
| dosum = 1; |
| break; |
| |
| default : |
| return 1; |
| /*NOTREACHED*/ |
| } |
| |
| if (csump != NULL) |
| hdrsum = *csump; |
| |
| if (dosum) { |
| sum = fr_cksum(fin->fin_m, fin->fin_ip, |
| fin->fin_p, fin->fin_dp, |
| fin->fin_dlen + fin->fin_hlen); |
| } |
| #if SOLARIS && defined(_KERNEL) && (SOLARIS2 >= 6) && defined(ICK_VALID) |
| } |
| #endif |
| #if !defined(_KERNEL) |
| if (sum == hdrsum) { |
| FR_DEBUG(("checkl4sum: %hx == %hx\n", sum, hdrsum)); |
| } else { |
| FR_DEBUG(("checkl4sum: %hx != %hx\n", sum, hdrsum)); |
| } |
| #endif |
| if (hdrsum == sum) { |
| fin->fin_cksum = 1; |
| return 0; |
| } |
| fin->fin_cksum = -1; |
| return -1; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ifpfillv4addr */ |
| /* Returns: int - 0 = address update, -1 = address not updated */ |
| /* Parameters: atype(I) - type of network address update to perform */ |
| /* sin(I) - pointer to source of address information */ |
| /* mask(I) - pointer to source of netmask information */ |
| /* inp(I) - pointer to destination address store */ |
| /* inpmask(I) - pointer to destination netmask store */ |
| /* */ |
| /* Given a type of network address update (atype) to perform, copy */ |
| /* information from sin/mask into inp/inpmask. If ipnmask is NULL then no */ |
| /* netmask update is performed unless FRI_NETMASKED is passed as atype, in */ |
| /* which case the operation fails. For all values of atype other than */ |
| /* FRI_NETMASKED, if inpmask is non-NULL then the mask is set to an all 1s */ |
| /* value. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_ifpfillv4addr(atype, sin, mask, inp, inpmask) |
| int atype; |
| struct sockaddr_in *sin, *mask; |
| struct in_addr *inp, *inpmask; |
| { |
| if (inpmask != NULL && atype != FRI_NETMASKED) |
| inpmask->s_addr = 0xffffffff; |
| |
| if (atype == FRI_NETWORK || atype == FRI_NETMASKED) { |
| if (atype == FRI_NETMASKED) { |
| if (inpmask == NULL) |
| return -1; |
| inpmask->s_addr = mask->sin_addr.s_addr; |
| } |
| inp->s_addr = sin->sin_addr.s_addr & mask->sin_addr.s_addr; |
| } else { |
| inp->s_addr = sin->sin_addr.s_addr; |
| } |
| return 0; |
| } |
| |
| |
| #ifdef USE_INET6 |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ifpfillv6addr */ |
| /* Returns: int - 0 = address update, -1 = address not updated */ |
| /* Parameters: atype(I) - type of network address update to perform */ |
| /* sin(I) - pointer to source of address information */ |
| /* mask(I) - pointer to source of netmask information */ |
| /* inp(I) - pointer to destination address store */ |
| /* inpmask(I) - pointer to destination netmask store */ |
| /* */ |
| /* Given a type of network address update (atype) to perform, copy */ |
| /* information from sin/mask into inp/inpmask. If ipnmask is NULL then no */ |
| /* netmask update is performed unless FRI_NETMASKED is passed as atype, in */ |
| /* which case the operation fails. For all values of atype other than */ |
| /* FRI_NETMASKED, if inpmask is non-NULL then the mask is set to an all 1s */ |
| /* value. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_ifpfillv6addr(atype, sin, mask, inp, inpmask) |
| int atype; |
| struct sockaddr_in6 *sin, *mask; |
| i6addr_t *inp, *inpmask; |
| { |
| i6addr_t *src, *and; |
| |
| src = (i6addr_t *)&sin->sin6_addr; |
| and = (i6addr_t *)&mask->sin6_addr; |
| |
| if (inpmask != NULL && atype != FRI_NETMASKED) { |
| inpmask->i6[0] = 0xffffffff; |
| inpmask->i6[1] = 0xffffffff; |
| inpmask->i6[2] = 0xffffffff; |
| inpmask->i6[3] = 0xffffffff; |
| } |
| |
| if (atype == FRI_NETWORK || atype == FRI_NETMASKED) { |
| if (atype == FRI_NETMASKED) { |
| if (inpmask == NULL) |
| return -1; |
| inpmask->i6[0] = and->i6[0]; |
| inpmask->i6[1] = and->i6[1]; |
| inpmask->i6[2] = and->i6[2]; |
| inpmask->i6[3] = and->i6[3]; |
| } |
| |
| inp->i6[0] = src->i6[0] & and->i6[0]; |
| inp->i6[1] = src->i6[1] & and->i6[1]; |
| inp->i6[2] = src->i6[2] & and->i6[2]; |
| inp->i6[3] = src->i6[3] & and->i6[3]; |
| } else { |
| inp->i6[0] = src->i6[0]; |
| inp->i6[1] = src->i6[1]; |
| inp->i6[2] = src->i6[2]; |
| inp->i6[3] = src->i6[3]; |
| } |
| return 0; |
| } |
| #endif |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_matchtag */ |
| /* Returns: 0 == mismatch, 1 == match. */ |
| /* Parameters: tag1(I) - pointer to first tag to compare */ |
| /* tag2(I) - pointer to second tag to compare */ |
| /* */ |
| /* Returns true (non-zero) or false(0) if the two tag structures can be */ |
| /* considered to be a match or not match, respectively. The tag is 16 */ |
| /* bytes long (16 characters) but that is overlayed with 4 32bit ints so */ |
| /* compare the ints instead, for speed. tag1 is the master of the */ |
| /* comparison. This function should only be called with both tag1 and tag2 */ |
| /* as non-NULL pointers. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_matchtag(tag1, tag2) |
| ipftag_t *tag1, *tag2; |
| { |
| if (tag1 == tag2) |
| return 1; |
| |
| if ((tag1->ipt_num[0] == 0) && (tag2->ipt_num[0] == 0)) |
| return 1; |
| |
| if ((tag1->ipt_num[0] == tag2->ipt_num[0]) && |
| (tag1->ipt_num[1] == tag2->ipt_num[1]) && |
| (tag1->ipt_num[2] == tag2->ipt_num[2]) && |
| (tag1->ipt_num[3] == tag2->ipt_num[3])) |
| return 1; |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_coalesce */ |
| /* Returns: 1 == success, -1 == failure, 0 == no change */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* */ |
| /* Attempt to get all of the packet data into a single, contiguous buffer. */ |
| /* If this call returns a failure then the buffers have also been freed. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_coalesce(fin) |
| fr_info_t *fin; |
| { |
| |
| if ((fin->fin_flx & FI_COALESCE) != 0) |
| return 1; |
| |
| /* |
| * If the mbuf pointers indicate that there is no mbuf to work with, |
| * return but do not indicate success or failure. |
| */ |
| if (fin->fin_m == NULL || fin->fin_mp == NULL) |
| return 0; |
| |
| #if defined(_KERNEL) |
| if (ipf_pullup(fin->fin_m, fin, fin->fin_plen) == NULL) { |
| ipf_main_softc_t *softc = fin->fin_main_soft; |
| |
| LBUMP(ipf_stats[fin->fin_out].fr_badcoalesces); |
| # ifdef MENTAT |
| FREE_MB_T(*fin->fin_mp); |
| # endif |
| *fin->fin_mp = NULL; |
| fin->fin_m = NULL; |
| return -1; |
| } |
| #else |
| fin = fin; /* LINT */ |
| #endif |
| return 1; |
| } |
| |
| |
| /* |
| * The following table lists all of the tunable variables that can be |
| * accessed via SIOCIPFGET/SIOCIPFSET/SIOCIPFGETNEXt. The format of each row |
| * in the table below is as follows: |
| * |
| * pointer to value, name of value, minimum, maximum, size of the value's |
| * container, value attribute flags |
| * |
| * For convienience, IPFT_RDONLY means the value is read-only, IPFT_WRDISABLED |
| * means the value can only be written to when IPFilter is loaded but disabled. |
| * The obvious implication is if neither of these are set then the value can be |
| * changed at any time without harm. |
| */ |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_findbycookie */ |
| /* Returns: NULL = search failed, else pointer to tune struct */ |
| /* Parameters: cookie(I) - cookie value to search for amongst tuneables */ |
| /* next(O) - pointer to place to store the cookie for the */ |
| /* "next" tuneable, if it is desired. */ |
| /* */ |
| /* This function is used to walk through all of the existing tunables with */ |
| /* successive calls. It searches the known tunables for the one which has */ |
| /* a matching value for "cookie" - ie its address. When returning a match, */ |
| /* the next one to be found may be returned inside next. */ |
| /* ------------------------------------------------------------------------ */ |
| static ipftuneable_t * |
| ipf_tune_findbycookie(ptop, cookie, next) |
| ipftuneable_t **ptop; |
| void *cookie, **next; |
| { |
| ipftuneable_t *ta, **tap; |
| |
| for (ta = *ptop; ta->ipft_name != NULL; ta++) |
| if (ta == cookie) { |
| if (next != NULL) { |
| /* |
| * If the next entry in the array has a name |
| * present, then return a pointer to it for |
| * where to go next, else return a pointer to |
| * the dynaminc list as a key to search there |
| * next. This facilitates a weak linking of |
| * the two "lists" together. |
| */ |
| if ((ta + 1)->ipft_name != NULL) |
| *next = ta + 1; |
| else |
| *next = ptop; |
| } |
| return ta; |
| } |
| |
| for (tap = ptop; (ta = *tap) != NULL; tap = &ta->ipft_next) |
| if (tap == cookie) { |
| if (next != NULL) |
| *next = &ta->ipft_next; |
| return ta; |
| } |
| |
| if (next != NULL) |
| *next = NULL; |
| return NULL; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_findbyname */ |
| /* Returns: NULL = search failed, else pointer to tune struct */ |
| /* Parameters: name(I) - name of the tuneable entry to find. */ |
| /* */ |
| /* Search the static array of tuneables and the list of dynamic tuneables */ |
| /* for an entry with a matching name. If we can find one, return a pointer */ |
| /* to the matching structure. */ |
| /* ------------------------------------------------------------------------ */ |
| static ipftuneable_t * |
| ipf_tune_findbyname(top, name) |
| ipftuneable_t *top; |
| const char *name; |
| { |
| ipftuneable_t *ta; |
| |
| for (ta = top; ta != NULL; ta = ta->ipft_next) |
| if (!strcmp(ta->ipft_name, name)) { |
| return ta; |
| } |
| |
| return NULL; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_add_array */ |
| /* Returns: int - 0 == success, else failure */ |
| /* Parameters: newtune - pointer to new tune array to add to tuneables */ |
| /* */ |
| /* Appends tune structures from the array passed in (newtune) to the end of */ |
| /* the current list of "dynamic" tuneable parameters. */ |
| /* If any entry to be added is already present (by name) then the operation */ |
| /* is aborted - entries that have been added are removed before returning. */ |
| /* An entry with no name (NULL) is used as the indication that the end of */ |
| /* the array has been reached. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_tune_add_array(softc, newtune) |
| ipf_main_softc_t *softc; |
| ipftuneable_t *newtune; |
| { |
| ipftuneable_t *nt, *dt; |
| int error = 0; |
| |
| for (nt = newtune; nt->ipft_name != NULL; nt++) { |
| error = ipf_tune_add(softc, nt); |
| if (error != 0) { |
| for (dt = newtune; dt != nt; dt++) { |
| (void) ipf_tune_del(softc, dt); |
| } |
| } |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_array_link */ |
| /* Returns: 0 == success, -1 == failure */ |
| /* Parameters: softc(I) - soft context pointerto work with */ |
| /* array(I) - pointer to an array of tuneables */ |
| /* */ |
| /* Given an array of tunables (array), append them to the current list of */ |
| /* tuneables for this context (softc->ipf_tuners.) To properly prepare the */ |
| /* the array for being appended to the list, initialise all of the next */ |
| /* pointers so we don't need to walk parts of it with ++ and others with */ |
| /* next. The array is expected to have an entry with a NULL name as the */ |
| /* terminator. Trying to add an array with no non-NULL names will return as */ |
| /* a failure. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_tune_array_link(softc, array) |
| ipf_main_softc_t *softc; |
| ipftuneable_t *array; |
| { |
| ipftuneable_t *t, **p; |
| |
| t = array; |
| if (t->ipft_name == NULL) |
| return -1; |
| |
| for (; t[1].ipft_name != NULL; t++) |
| t[0].ipft_next = &t[1]; |
| t->ipft_next = NULL; |
| |
| /* |
| * Since a pointer to the last entry isn't kept, we need to find it |
| * each time we want to add new variables to the list. |
| */ |
| for (p = &softc->ipf_tuners; (t = *p) != NULL; p = &t->ipft_next) |
| if (t->ipft_name == NULL) |
| break; |
| *p = array; |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_array_unlink */ |
| /* Returns: 0 == success, -1 == failure */ |
| /* Parameters: softc(I) - soft context pointerto work with */ |
| /* array(I) - pointer to an array of tuneables */ |
| /* */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_tune_array_unlink(softc, array) |
| ipf_main_softc_t *softc; |
| ipftuneable_t *array; |
| { |
| ipftuneable_t *t, **p; |
| |
| for (p = &softc->ipf_tuners; (t = *p) != NULL; p = &t->ipft_next) |
| if (t == array) |
| break; |
| if (t == NULL) |
| return -1; |
| |
| for (; t[1].ipft_name != NULL; t++) |
| ; |
| |
| *p = t->ipft_next; |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_array_copy */ |
| /* Returns: NULL = failure, else pointer to new array */ |
| /* Parameters: base(I) - pointer to structure base */ |
| /* size(I) - size of the array at template */ |
| /* template(I) - original array to copy */ |
| /* */ |
| /* Allocate memory for a new set of tuneable values and copy everything */ |
| /* from template into the new region of memory. The new region is full of */ |
| /* uninitialised pointers (ipft_next) so set them up. Now, ipftp_offset... */ |
| /* */ |
| /* NOTE: the following assumes that sizeof(long) == sizeof(void *) */ |
| /* In the array template, ipftp_offset is the offset (in bytes) of the */ |
| /* location of the tuneable value inside the structure pointed to by base. */ |
| /* As ipftp_offset is a union over the pointers to the tuneable values, if */ |
| /* we add base to the copy's ipftp_offset, copy ends up with a pointer in */ |
| /* ipftp_void that points to the stored value. */ |
| /* ------------------------------------------------------------------------ */ |
| ipftuneable_t * |
| ipf_tune_array_copy(base, size, template) |
| void *base; |
| size_t size; |
| ipftuneable_t *template; |
| { |
| ipftuneable_t *copy; |
| int i; |
| |
| |
| KMALLOCS(copy, ipftuneable_t *, size); |
| if (copy == NULL) { |
| return NULL; |
| } |
| bcopy(template, copy, size); |
| |
| for (i = 0; copy[i].ipft_name; i++) { |
| copy[i].ipft_una.ipftp_offset += (u_long)base; |
| copy[i].ipft_next = copy + i + 1; |
| } |
| |
| return copy; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_add */ |
| /* Returns: int - 0 == success, else failure */ |
| /* Parameters: newtune - pointer to new tune entry to add to tuneables */ |
| /* */ |
| /* Appends tune structures from the array passed in (newtune) to the end of */ |
| /* the current list of "dynamic" tuneable parameters. Once added, the */ |
| /* owner of the object is not expected to ever change "ipft_next". */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_tune_add(softc, newtune) |
| ipf_main_softc_t *softc; |
| ipftuneable_t *newtune; |
| { |
| ipftuneable_t *ta, **tap; |
| |
| ta = ipf_tune_findbyname(softc->ipf_tuners, newtune->ipft_name); |
| if (ta != NULL) { |
| softc->ipf_interror = 74; |
| return EEXIST; |
| } |
| |
| for (tap = &softc->ipf_tuners; *tap != NULL; tap = &(*tap)->ipft_next) |
| ; |
| |
| newtune->ipft_next = NULL; |
| *tap = newtune; |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_del */ |
| /* Returns: int - 0 == success, else failure */ |
| /* Parameters: oldtune - pointer to tune entry to remove from the list of */ |
| /* current dynamic tuneables */ |
| /* */ |
| /* Search for the tune structure, by pointer, in the list of those that are */ |
| /* dynamically added at run time. If found, adjust the list so that this */ |
| /* structure is no longer part of it. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_tune_del(softc, oldtune) |
| ipf_main_softc_t *softc; |
| ipftuneable_t *oldtune; |
| { |
| ipftuneable_t *ta, **tap; |
| int error = 0; |
| |
| for (tap = &softc->ipf_tuners; (ta = *tap) != NULL; |
| tap = &ta->ipft_next) { |
| if (ta == oldtune) { |
| *tap = oldtune->ipft_next; |
| oldtune->ipft_next = NULL; |
| break; |
| } |
| } |
| |
| if (ta == NULL) { |
| error = ESRCH; |
| softc->ipf_interror = 75; |
| } |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune_del_array */ |
| /* Returns: int - 0 == success, else failure */ |
| /* Parameters: oldtune - pointer to tuneables array */ |
| /* */ |
| /* Remove each tuneable entry in the array from the list of "dynamic" */ |
| /* tunables. If one entry should fail to be found, an error will be */ |
| /* returned and no further ones removed. */ |
| /* An entry with a NULL name is used as the indicator of the last entry in */ |
| /* the array. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_tune_del_array(softc, oldtune) |
| ipf_main_softc_t *softc; |
| ipftuneable_t *oldtune; |
| { |
| ipftuneable_t *ot; |
| int error = 0; |
| |
| for (ot = oldtune; ot->ipft_name != NULL; ot++) { |
| error = ipf_tune_del(softc, ot); |
| if (error != 0) |
| break; |
| } |
| |
| return error; |
| |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_tune */ |
| /* Returns: int - 0 == success, else failure */ |
| /* Parameters: cmd(I) - ioctl command number */ |
| /* data(I) - pointer to ioctl data structure */ |
| /* */ |
| /* Implement handling of SIOCIPFGETNEXT, SIOCIPFGET and SIOCIPFSET. These */ |
| /* three ioctls provide the means to access and control global variables */ |
| /* within IPFilter, allowing (for example) timeouts and table sizes to be */ |
| /* changed without rebooting, reloading or recompiling. The initialisation */ |
| /* and 'destruction' routines of the various components of ipfilter are all */ |
| /* each responsible for handling their own values being too big. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_ipftune(softc, cmd, data) |
| ipf_main_softc_t *softc; |
| ioctlcmd_t cmd; |
| void *data; |
| { |
| ipftuneable_t *ta; |
| ipftune_t tu; |
| void *cookie; |
| int error; |
| |
| error = ipf_inobj(softc, data, &tu, IPFOBJ_TUNEABLE); |
| if (error != 0) |
| return error; |
| |
| tu.ipft_name[sizeof(tu.ipft_name) - 1] = '\0'; |
| cookie = tu.ipft_cookie; |
| ta = NULL; |
| |
| switch (cmd) |
| { |
| case SIOCIPFGETNEXT : |
| /* |
| * If cookie is non-NULL, assume it to be a pointer to the last |
| * entry we looked at, so find it (if possible) and return a |
| * pointer to the next one after it. The last entry in the |
| * the table is a NULL entry, so when we get to it, set cookie |
| * to NULL and return that, indicating end of list, erstwhile |
| * if we come in with cookie set to NULL, we are starting anew |
| * at the front of the list. |
| */ |
| if (cookie != NULL) { |
| ta = ipf_tune_findbycookie(&softc->ipf_tuners, |
| cookie, &tu.ipft_cookie); |
| } else { |
| ta = softc->ipf_tuners; |
| tu.ipft_cookie = ta + 1; |
| } |
| if (ta != NULL) { |
| /* |
| * Entry found, but does the data pointed to by that |
| * row fit in what we can return? |
| */ |
| if (ta->ipft_sz > sizeof(tu.ipft_un)) { |
| softc->ipf_interror = 76; |
| return EINVAL; |
| } |
| |
| tu.ipft_vlong = 0; |
| if (ta->ipft_sz == sizeof(u_long)) |
| tu.ipft_vlong = *ta->ipft_plong; |
| else if (ta->ipft_sz == sizeof(u_int)) |
| tu.ipft_vint = *ta->ipft_pint; |
| else if (ta->ipft_sz == sizeof(u_short)) |
| tu.ipft_vshort = *ta->ipft_pshort; |
| else if (ta->ipft_sz == sizeof(u_char)) |
| tu.ipft_vchar = *ta->ipft_pchar; |
| |
| tu.ipft_sz = ta->ipft_sz; |
| tu.ipft_min = ta->ipft_min; |
| tu.ipft_max = ta->ipft_max; |
| tu.ipft_flags = ta->ipft_flags; |
| bcopy(ta->ipft_name, tu.ipft_name, |
| MIN(sizeof(tu.ipft_name), |
| strlen(ta->ipft_name) + 1)); |
| } |
| error = ipf_outobj(softc, data, &tu, IPFOBJ_TUNEABLE); |
| break; |
| |
| case SIOCIPFGET : |
| case SIOCIPFSET : |
| /* |
| * Search by name or by cookie value for a particular entry |
| * in the tuning paramter table. |
| */ |
| softc->ipf_interror = 77; |
| error = ESRCH; |
| if (cookie != NULL) { |
| ta = ipf_tune_findbycookie(&softc->ipf_tuners, |
| cookie, NULL); |
| if (ta != NULL) |
| error = 0; |
| } else if (tu.ipft_name[0] != '\0') { |
| ta = ipf_tune_findbyname(softc->ipf_tuners, |
| tu.ipft_name); |
| if (ta != NULL) |
| error = 0; |
| } |
| if (error != 0) |
| break; |
| |
| if (cmd == (ioctlcmd_t)SIOCIPFGET) { |
| /* |
| * Fetch the tuning parameters for a particular value |
| */ |
| tu.ipft_vlong = 0; |
| if (ta->ipft_sz == sizeof(u_long)) |
| tu.ipft_vlong = *ta->ipft_plong; |
| else if (ta->ipft_sz == sizeof(u_int)) |
| tu.ipft_vint = *ta->ipft_pint; |
| else if (ta->ipft_sz == sizeof(u_short)) |
| tu.ipft_vshort = *ta->ipft_pshort; |
| else if (ta->ipft_sz == sizeof(u_char)) |
| tu.ipft_vchar = *ta->ipft_pchar; |
| tu.ipft_cookie = ta; |
| tu.ipft_sz = ta->ipft_sz; |
| tu.ipft_min = ta->ipft_min; |
| tu.ipft_max = ta->ipft_max; |
| tu.ipft_flags = ta->ipft_flags; |
| error = ipf_outobj(softc, data, &tu, IPFOBJ_TUNEABLE); |
| |
| } else if (cmd == (ioctlcmd_t)SIOCIPFSET) { |
| /* |
| * Set an internal parameter. The hard part here is |
| * getting the new value safely and correctly out of |
| * the kernel (given we only know its size, not type.) |
| */ |
| u_long in; |
| |
| if (((ta->ipft_flags & IPFT_WRDISABLED) != 0) && |
| (softc->ipf_running > 0)) { |
| softc->ipf_interror = 78; |
| error = EBUSY; |
| break; |
| } |
| |
| in = tu.ipft_vlong; |
| if (in < ta->ipft_min || in > ta->ipft_max) { |
| softc->ipf_interror = 79; |
| error = EINVAL; |
| break; |
| } |
| |
| if (ta->ipft_func != NULL) { |
| SPL_INT(s); |
| |
| SPL_NET(s); |
| error = (*ta->ipft_func)(softc, ta, |
| &tu.ipft_un); |
| SPL_X(s); |
| |
| } else if (ta->ipft_sz == sizeof(u_long)) { |
| tu.ipft_vlong = *ta->ipft_plong; |
| *ta->ipft_plong = in; |
| |
| } else if (ta->ipft_sz == sizeof(u_int)) { |
| tu.ipft_vint = *ta->ipft_pint; |
| *ta->ipft_pint = (u_int)(in & 0xffffffff); |
| |
| } else if (ta->ipft_sz == sizeof(u_short)) { |
| tu.ipft_vshort = *ta->ipft_pshort; |
| *ta->ipft_pshort = (u_short)(in & 0xffff); |
| |
| } else if (ta->ipft_sz == sizeof(u_char)) { |
| tu.ipft_vchar = *ta->ipft_pchar; |
| *ta->ipft_pchar = (u_char)(in & 0xff); |
| } |
| error = ipf_outobj(softc, data, &tu, IPFOBJ_TUNEABLE); |
| } |
| break; |
| |
| default : |
| softc->ipf_interror = 80; |
| error = EINVAL; |
| break; |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_zerostats */ |
| /* Returns: int - 0 = success, else failure */ |
| /* Parameters: data(O) - pointer to pointer for copying data back to */ |
| /* */ |
| /* Copies the current statistics out to userspace and then zero's the */ |
| /* current ones in the kernel. The lock is only held across the bzero() as */ |
| /* the copyout may result in paging (ie network activity.) */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_zerostats(softc, data) |
| ipf_main_softc_t *softc; |
| caddr_t data; |
| { |
| friostat_t fio; |
| int error; |
| |
| ipf_getstat(softc, &fio); |
| error = ipf_outobj(softc, data, &fio, IPFOBJ_IPFSTAT); |
| if (error != 0) |
| return error; |
| |
| WRITE_ENTER(&softc->ipf_mutex); |
| bzero(&softc->ipf_stats, sizeof(softc->ipf_stats)); |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_resolvedest */ |
| /* Returns: Nil */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* base(I) - where strings are stored */ |
| /* fdp(IO) - pointer to destination information to resolve */ |
| /* v(I) - IP protocol version to match */ |
| /* */ |
| /* Looks up an interface name in the frdest structure pointed to by fdp and */ |
| /* if a matching name can be found for the particular IP protocol version */ |
| /* then store the interface pointer in the frdest struct. If no match is */ |
| /* found, then set the interface pointer to be -1 as NULL is considered to */ |
| /* indicate there is no information at all in the structure. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_resolvedest(softc, base, fdp, v) |
| ipf_main_softc_t *softc; |
| char *base; |
| frdest_t *fdp; |
| int v; |
| { |
| void *ifp; |
| |
| ifp = NULL; |
| |
| if (fdp->fd_name != -1) { |
| ifp = GETIFP(base + fdp->fd_name, v); |
| if (ifp == NULL) |
| ifp = (void *)-1; |
| } |
| fdp->fd_ptr = ifp; |
| |
| if ((ifp != NULL) && (ifp != (void *)-1)) { |
| fdp->fd_local = ipf_deliverlocal(softc, v, ifp, &fdp->fd_ip6); |
| } |
| |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_resolvenic */ |
| /* Returns: void* - NULL = wildcard name, -1 = failed to find NIC, else */ |
| /* pointer to interface structure for NIC */ |
| /* Parameters: name(I) - complete interface name */ |
| /* v(I) - IP protocol version */ |
| /* */ |
| /* Look for a network interface structure that firstly has a matching name */ |
| /* to that passed in and that is also being used for that IP protocol */ |
| /* version (necessary on some platforms where there are separate listings */ |
| /* for both IPv4 and IPv6 on the same physical NIC. */ |
| /* ------------------------------------------------------------------------ */ |
| void * |
| ipf_resolvenic(softc, name, v) |
| ipf_main_softc_t *softc; |
| char *name; |
| int v; |
| { |
| void *nic; |
| |
| if (name[0] == '\0') |
| return NULL; |
| |
| if ((name[1] == '\0') && ((name[0] == '-') || (name[0] == '*'))) { |
| return NULL; |
| } |
| |
| nic = GETIFP(name, v); |
| if (nic == NULL) |
| nic = (void *)-1; |
| return nic; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_token_expire */ |
| /* Returns: None. */ |
| /* Parameters: None. */ |
| /* */ |
| /* This function is run every ipf tick to see if there are any tokens that */ |
| /* have been held for too long and need to be freed up. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_token_expire(softc) |
| ipf_main_softc_t *softc; |
| { |
| ipftoken_t *it; |
| |
| WRITE_ENTER(&softc->ipf_tokens); |
| while ((it = softc->ipf_token_head) != NULL) { |
| if (it->ipt_die > softc->ipf_ticks) |
| break; |
| |
| ipf_token_free(softc, it); |
| } |
| RWLOCK_EXIT(&softc->ipf_tokens); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_token_del */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: type(I) - the token type to match */ |
| /* uid(I) - uid owning the token */ |
| /* ptr(I) - context pointer for the token */ |
| /* */ |
| /* This function looks for a a token in the current list that matches up */ |
| /* the fields (type, uid, ptr). If none is found, ESRCH is returned, else */ |
| /* call ipf_token_free() to remove it from the list. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_token_del(softc, type, uid, ptr) |
| ipf_main_softc_t *softc; |
| int type, uid; |
| void *ptr; |
| { |
| ipftoken_t *it; |
| int error; |
| |
| softc->ipf_interror = 82; |
| error = ESRCH; |
| |
| WRITE_ENTER(&softc->ipf_tokens); |
| for (it = softc->ipf_token_head; it != NULL; it = it->ipt_next) |
| if (ptr == it->ipt_ctx && type == it->ipt_type && |
| uid == it->ipt_uid) { |
| ipf_token_free(softc, it); |
| softc->ipf_interror = 83; |
| error = 0; |
| break; |
| } |
| RWLOCK_EXIT(&softc->ipf_tokens); |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_token_find */ |
| /* Returns: ipftoken_t * - NULL if no memory, else pointer to token */ |
| /* Parameters: type(I) - the token type to match */ |
| /* uid(I) - uid owning the token */ |
| /* ptr(I) - context pointer for the token */ |
| /* */ |
| /* This function looks for a live token in the list of current tokens that */ |
| /* matches the tuple (type, uid, ptr). If one cannot be found then one is */ |
| /* allocated. If one is found then it is moved to the top of the list of */ |
| /* currently active tokens. */ |
| /* ------------------------------------------------------------------------ */ |
| ipftoken_t * |
| ipf_token_find(softc, type, uid, ptr) |
| ipf_main_softc_t *softc; |
| int type, uid; |
| void *ptr; |
| { |
| ipftoken_t *it, *new; |
| |
| KMALLOC(new, ipftoken_t *); |
| |
| WRITE_ENTER(&softc->ipf_tokens); |
| for (it = softc->ipf_token_head; it != NULL; it = it->ipt_next) { |
| if (ptr == it->ipt_ctx && type == it->ipt_type && |
| uid == it->ipt_uid) |
| break; |
| } |
| |
| if (it == NULL) { |
| it = new; |
| new = NULL; |
| if (it == NULL) |
| return NULL; |
| it->ipt_data = NULL; |
| it->ipt_ctx = ptr; |
| it->ipt_uid = uid; |
| it->ipt_type = type; |
| it->ipt_next = NULL; |
| it->ipt_ref = 2; |
| } else { |
| if (new != NULL) { |
| KFREE(new); |
| new = NULL; |
| } |
| |
| ipf_token_unlink(softc, it); |
| it->ipt_ref++; |
| } |
| it->ipt_pnext = softc->ipf_token_tail; |
| *softc->ipf_token_tail = it; |
| softc->ipf_token_tail = &it->ipt_next; |
| it->ipt_next = NULL; |
| |
| it->ipt_die = softc->ipf_ticks + 2; |
| |
| RWLOCK_EXIT(&softc->ipf_tokens); |
| |
| return it; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_token_unlink */ |
| /* Returns: None. */ |
| /* Parameters: token(I) - pointer to token structure */ |
| /* Write Locks: ipf_tokens */ |
| /* */ |
| /* This function unlinks a token structure from the linked list of tokens */ |
| /* that "own" it. The head pointer never needs to be explicitly adjusted */ |
| /* but the tail does due to the linked list implementation. */ |
| /* ------------------------------------------------------------------------ */ |
| static void |
| ipf_token_unlink(softc, token) |
| ipf_main_softc_t *softc; |
| ipftoken_t *token; |
| { |
| |
| if (softc->ipf_token_tail == &token->ipt_next) |
| softc->ipf_token_tail = token->ipt_pnext; |
| |
| *token->ipt_pnext = token->ipt_next; |
| if (token->ipt_next != NULL) |
| token->ipt_next->ipt_pnext = token->ipt_pnext; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_token_deref */ |
| /* Returns: None. */ |
| /* Parameters: token(I) - pointer to token structure */ |
| /* Write Locks: ipf_tokens */ |
| /* */ |
| /* Drop the reference count on the token structure and if it drops to zero, */ |
| /* call the dereference function for the token type because it is then */ |
| /* possible to free the token data structure. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_token_deref(softc, token) |
| ipf_main_softc_t *softc; |
| ipftoken_t *token; |
| { |
| void *data, **datap; |
| |
| token->ipt_ref--; |
| if (token->ipt_ref > 0) |
| return; |
| |
| data = token->ipt_data; |
| datap = &data; |
| |
| if ((data != NULL) && (data != (void *)-1)) { |
| switch (token->ipt_type) |
| { |
| case IPFGENITER_IPF : |
| (void) ipf_derefrule(softc, (frentry_t **)datap); |
| break; |
| case IPFGENITER_IPNAT : |
| WRITE_ENTER(&softc->ipf_nat); |
| ipf_nat_rulederef(softc, (ipnat_t **)datap); |
| RWLOCK_EXIT(&softc->ipf_nat); |
| break; |
| case IPFGENITER_NAT : |
| ipf_nat_deref(softc, (nat_t **)datap); |
| break; |
| case IPFGENITER_STATE : |
| ipf_state_deref(softc, (ipstate_t **)datap); |
| break; |
| case IPFGENITER_FRAG : |
| ipf_frag_pkt_deref(softc, (ipfr_t **)datap); |
| break; |
| case IPFGENITER_NATFRAG : |
| ipf_frag_nat_deref(softc, (ipfr_t **)datap); |
| break; |
| case IPFGENITER_HOSTMAP : |
| WRITE_ENTER(&softc->ipf_nat); |
| ipf_nat_hostmapdel((hostmap_t **)datap); |
| RWLOCK_EXIT(&softc->ipf_nat); |
| break; |
| default : |
| ipf_lookup_iterderef(softc, token->ipt_type, data); |
| break; |
| } |
| } |
| |
| KFREE(token); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_token_free */ |
| /* Returns: None. */ |
| /* Parameters: token(I) - pointer to token structure */ |
| /* Write Locks: ipf_tokens */ |
| /* */ |
| /* This function unlinks a token from the linked list and does a dereference*/ |
| /* on it to encourage it to be freed. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_token_free(softc, token) |
| ipf_main_softc_t *softc; |
| ipftoken_t *token; |
| { |
| |
| ipf_token_unlink(softc, token); |
| |
| ipf_token_deref(softc, token); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_getnextrule */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: t(I) - pointer to destination information to resolve */ |
| /* ptr(I) - pointer to ipfobj_t to copyin from user space */ |
| /* */ |
| /* This function's first job is to bring in the ipfruleiter_t structure via */ |
| /* the ipfobj_t structure to determine what should be the next rule to */ |
| /* return. Once the ipfruleiter_t has been brought in, it then tries to */ |
| /* find the 'next rule'. This may include searching rule group lists or */ |
| /* just be as simple as looking at the 'next' field in the rule structure. */ |
| /* When we have found the rule to return, increase its reference count and */ |
| /* if we used an existing rule to get here, decrease its reference count. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_getnextrule(softc, t, ptr) |
| ipf_main_softc_t *softc; |
| ipftoken_t *t; |
| void *ptr; |
| { |
| frentry_t *fr, *next, zero; |
| int error, count, out; |
| ipfruleiter_t it; |
| frgroup_t *fg; |
| char *dst; |
| |
| if (t == NULL || ptr == NULL) { |
| softc->ipf_interror = 84; |
| return EFAULT; |
| } |
| |
| error = ipf_inobj(softc, ptr, &it, IPFOBJ_IPFITER); |
| if (error != 0) |
| return error; |
| |
| if ((it.iri_inout < 0) || (it.iri_inout > 3)) { |
| softc->ipf_interror = 85; |
| return EINVAL; |
| } |
| if ((it.iri_active != 0) && (it.iri_active != 1)) { |
| softc->ipf_interror = 86; |
| return EINVAL; |
| } |
| if (it.iri_nrules == 0) { |
| softc->ipf_interror = 87; |
| return ENOSPC; |
| } |
| if (it.iri_rule == NULL) { |
| softc->ipf_interror = 88; |
| return EFAULT; |
| } |
| |
| fg = NULL; |
| out = it.iri_inout & F_OUT; |
| |
| READ_ENTER(&softc->ipf_mutex); |
| /* |
| * Retrieve "previous" entry from token and find the next entry. |
| */ |
| fr = t->ipt_data; |
| if (fr == NULL) { |
| if (*it.iri_group == '\0') { |
| if ((it.iri_inout & F_ACIN) != 0) |
| next = softc->ipf_acct[out][it.iri_active]; |
| else |
| next = softc->ipf_rules[out][it.iri_active]; |
| } else { |
| fg = ipf_findgroup(softc, it.iri_group, IPL_LOGIPF, |
| it.iri_active, NULL); |
| if (fg != NULL) |
| next = fg->fg_start; |
| else |
| next = NULL; |
| } |
| } else { |
| next = fr->fr_next; |
| } |
| |
| dst = (char *)it.iri_rule; |
| /* |
| * The ipfruleiter may ask for more than 1 rule at a time to be |
| * copied out, so long as that many exist in the list to start with! |
| */ |
| for (count = it.iri_nrules; count > 0; count--) { |
| /* |
| * If we found an entry, add reference to it and update token. |
| * Otherwise, zero out data to be returned and NULL out token. |
| */ |
| if (next != NULL) { |
| MUTEX_ENTER(&next->fr_lock); |
| next->fr_ref++; |
| MUTEX_EXIT(&next->fr_lock); |
| t->ipt_data = next; |
| } else { |
| bzero(&zero, sizeof(zero)); |
| next = &zero; |
| t->ipt_data = NULL; |
| } |
| |
| /* |
| * Now that we have ref, it's save to give up lock. |
| */ |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| |
| /* |
| * Copy out data and clean up references and token as needed. |
| */ |
| error = COPYOUT(next, dst, next->fr_size); |
| if (error != 0) { |
| softc->ipf_interror = 89; |
| return EFAULT; |
| } |
| if (t->ipt_data == NULL) { |
| break; |
| } else { |
| if (fr != NULL) |
| (void) ipf_derefrule(softc, &fr); |
| dst += next->fr_size; |
| if (next->fr_data != NULL) { |
| error = COPYOUT(next->fr_data, dst, |
| next->fr_dsize); |
| if (error != 0) { |
| softc->ipf_interror = 90; |
| error = EFAULT; |
| } else { |
| dst += next->fr_dsize; |
| } |
| } |
| if (next->fr_next == NULL) { |
| t->ipt_data = NULL; |
| break; |
| } |
| } |
| |
| if ((count == 1) || (error != 0)) |
| break; |
| |
| READ_ENTER(&softc->ipf_mutex); |
| fr = next; |
| next = next->fr_next; |
| } |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_frruleiter */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: data(I) - the token type to match */ |
| /* uid(I) - uid owning the token */ |
| /* ptr(I) - context pointer for the token */ |
| /* */ |
| /* This function serves as a stepping stone between ipf_ipf_ioctl and */ |
| /* ipf_getnextrule. It's role is to find the right token in the kernel for */ |
| /* the process doing the ioctl and use that to ask for the next rule. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_frruleiter(softc, data, uid, ctx) |
| ipf_main_softc_t *softc; |
| void *data, *ctx; |
| int uid; |
| { |
| ipftoken_t *token; |
| int error; |
| |
| token = ipf_token_find(softc, IPFGENITER_IPF, uid, ctx); |
| if (token != NULL) { |
| error = ipf_getnextrule(softc, token, data); |
| } else { |
| WRITE_ENTER(&softc->ipf_tokens); |
| if (token->ipt_data == NULL) |
| ipf_token_free(softc, token); |
| else |
| ipf_token_deref(softc, token); |
| RWLOCK_EXIT(&softc->ipf_tokens); |
| softc->ipf_interror = 91; |
| error = EFAULT; |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_geniter */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: token(I) - pointer to ipftoken_t structure */ |
| /* itp(I) - */ |
| /* */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_geniter(softc, token, itp) |
| ipf_main_softc_t *softc; |
| ipftoken_t *token; |
| ipfgeniter_t *itp; |
| { |
| int error; |
| |
| switch (itp->igi_type) |
| { |
| case IPFGENITER_FRAG : |
| error = ipf_frag_pkt_next(softc, token, itp); |
| break; |
| default : |
| softc->ipf_interror = 92; |
| error = EINVAL; |
| break; |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_genericiter */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: data(I) - the token type to match */ |
| /* uid(I) - uid owning the token */ |
| /* ptr(I) - context pointer for the token */ |
| /* */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_genericiter(softc, data, uid, ctx) |
| ipf_main_softc_t *softc; |
| void *data, *ctx; |
| int uid; |
| { |
| ipftoken_t *token; |
| ipfgeniter_t iter; |
| int error; |
| |
| error = ipf_inobj(softc, data, &iter, IPFOBJ_GENITER); |
| if (error != 0) |
| return error; |
| |
| token = ipf_token_find(softc, iter.igi_type, uid, ctx); |
| if (token != NULL) { |
| token->ipt_subtype = iter.igi_type; |
| error = ipf_geniter(softc, token, &iter); |
| } else { |
| WRITE_ENTER(&softc->ipf_tokens); |
| if (token->ipt_data == NULL) |
| ipf_token_free(softc, token); |
| else |
| ipf_token_deref(softc, token); |
| RWLOCK_EXIT(&softc->ipf_tokens); |
| softc->ipf_interror = 93; |
| error = EFAULT; |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ipf_ioctl */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: data(I) - the token type to match */ |
| /* cmd(I) - the ioctl command number */ |
| /* mode(I) - mode flags for the ioctl */ |
| /* uid(I) - uid owning the token */ |
| /* ptr(I) - context pointer for the token */ |
| /* */ |
| /* This function handles all of the ioctl command that are actually isssued */ |
| /* to the /dev/ipl device. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_ipf_ioctl(softc, data, cmd, mode, uid, ctx) |
| ipf_main_softc_t *softc; |
| caddr_t data; |
| ioctlcmd_t cmd; |
| int mode, uid; |
| void *ctx; |
| { |
| friostat_t fio; |
| int error, tmp; |
| SPL_INT(s); |
| |
| switch (cmd) |
| { |
| case SIOCFRENB : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 94; |
| error = EPERM; |
| } else { |
| error = BCOPYIN(data, &tmp, sizeof(tmp)); |
| if (error != 0) { |
| softc->ipf_interror = 95; |
| error = EFAULT; |
| break; |
| } |
| |
| WRITE_ENTER(&softc->ipf_global); |
| if (tmp) { |
| if (softc->ipf_running > 0) |
| error = 0; |
| else |
| error = ipfattach(softc); |
| if (error == 0) |
| softc->ipf_running = 1; |
| else |
| (void) ipfdetach(softc); |
| } else { |
| error = ipfdetach(softc); |
| if (error == 0) |
| softc->ipf_running = -1; |
| } |
| RWLOCK_EXIT(&softc->ipf_global); |
| } |
| break; |
| |
| case SIOCIPFSET : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 96; |
| error = EPERM; |
| break; |
| } |
| /* FALLTHRU */ |
| case SIOCIPFGETNEXT : |
| case SIOCIPFGET : |
| error = ipf_ipftune(softc, cmd, (void *)data); |
| break; |
| |
| case SIOCSETFF : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 97; |
| error = EPERM; |
| } else { |
| error = BCOPYIN(data, &softc->ipf_flags, |
| sizeof(softc->ipf_flags)); |
| if (error != 0) { |
| softc->ipf_interror = 98; |
| error = EFAULT; |
| } |
| } |
| break; |
| |
| case SIOCGETFF : |
| error = BCOPYOUT(&softc->ipf_flags, data, |
| sizeof(softc->ipf_flags)); |
| if (error != 0) { |
| softc->ipf_interror = 99; |
| error = EFAULT; |
| } |
| break; |
| |
| case SIOCFUNCL : |
| error = ipf_resolvefunc(softc, (void *)data); |
| break; |
| |
| case SIOCINAFR : |
| case SIOCRMAFR : |
| case SIOCADAFR : |
| case SIOCZRLST : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 100; |
| error = EPERM; |
| } else { |
| error = frrequest(softc, IPL_LOGIPF, cmd, (caddr_t)data, |
| softc->ipf_active, 1); |
| } |
| break; |
| |
| case SIOCINIFR : |
| case SIOCRMIFR : |
| case SIOCADIFR : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 101; |
| error = EPERM; |
| } else { |
| error = frrequest(softc, IPL_LOGIPF, cmd, (caddr_t)data, |
| 1 - softc->ipf_active, 1); |
| } |
| break; |
| |
| case SIOCSWAPA : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 102; |
| error = EPERM; |
| } else { |
| WRITE_ENTER(&softc->ipf_mutex); |
| error = BCOPYOUT(&softc->ipf_active, data, |
| sizeof(softc->ipf_active)); |
| if (error != 0) { |
| softc->ipf_interror = 103; |
| error = EFAULT; |
| } else { |
| softc->ipf_active = 1 - softc->ipf_active; |
| } |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| } |
| break; |
| |
| case SIOCGETFS : |
| ipf_getstat(softc, &fio); |
| error = ipf_outobj(softc, (void *)data, &fio, IPFOBJ_IPFSTAT); |
| break; |
| |
| case SIOCFRZST : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 104; |
| error = EPERM; |
| } else |
| error = ipf_zerostats(softc, (caddr_t)data); |
| break; |
| |
| case SIOCIPFFL : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 105; |
| error = EPERM; |
| } else { |
| error = BCOPYIN(data, &tmp, sizeof(tmp)); |
| if (!error) { |
| tmp = ipf_flush(softc, IPL_LOGIPF, tmp); |
| error = BCOPYOUT(&tmp, data, sizeof(tmp)); |
| if (error != 0) { |
| softc->ipf_interror = 106; |
| error = EFAULT; |
| } |
| } else { |
| softc->ipf_interror = 107; |
| error = EFAULT; |
| } |
| } |
| break; |
| |
| #ifdef USE_INET6 |
| case SIOCIPFL6 : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 108; |
| error = EPERM; |
| } else { |
| error = BCOPYIN(data, &tmp, sizeof(tmp)); |
| if (!error) { |
| tmp = ipf_flush(softc, IPL_LOGIPF, tmp); |
| error = BCOPYOUT(&tmp, data, sizeof(tmp)); |
| if (error != 0) { |
| softc->ipf_interror = 109; |
| error = EFAULT; |
| } |
| } else { |
| softc->ipf_interror = 110; |
| error = EFAULT; |
| } |
| } |
| break; |
| #endif |
| |
| case SIOCSTLCK : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 122; |
| error = EPERM; |
| } else { |
| error = BCOPYIN(data, &tmp, sizeof(tmp)); |
| if (error == 0) { |
| ipf_state_setlock(softc->ipf_state_soft, tmp); |
| ipf_nat_setlock(softc->ipf_nat_soft, tmp); |
| ipf_frag_setlock(softc->ipf_frag_soft, tmp); |
| ipf_auth_setlock(softc->ipf_auth_soft, tmp); |
| } else { |
| softc->ipf_interror = 111; |
| error = EFAULT; |
| } |
| } |
| break; |
| |
| #ifdef IPFILTER_LOG |
| case SIOCIPFFB : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 112; |
| error = EPERM; |
| } else { |
| tmp = ipf_log_clear(softc, IPL_LOGIPF); |
| error = BCOPYOUT(&tmp, data, sizeof(tmp)); |
| if (error) { |
| softc->ipf_interror = 113; |
| error = EFAULT; |
| } |
| } |
| break; |
| #endif /* IPFILTER_LOG */ |
| |
| case SIOCFRSYN : |
| if (!(mode & FWRITE)) { |
| softc->ipf_interror = 114; |
| error = EPERM; |
| } else { |
| WRITE_ENTER(&softc->ipf_global); |
| #if (defined(MENTAT) && defined(_KERNEL)) && !defined(INSTANCES) |
| error = ipfsync(); |
| #else |
| ipf_sync(softc, NULL); |
| error = 0; |
| #endif |
| RWLOCK_EXIT(&softc->ipf_global); |
| |
| } |
| break; |
| |
| case SIOCGFRST : |
| error = ipf_outobj(softc, (void *)data, ipf_frag_stats(softc), |
| IPFOBJ_FRAGSTAT); |
| break; |
| |
| #ifdef IPFILTER_LOG |
| case FIONREAD : |
| tmp = ipf_log_bytesused(softc, IPL_LOGIPF); |
| error = BCOPYOUT(&tmp, data, sizeof(tmp)); |
| break; |
| #endif |
| |
| case SIOCIPFITER : |
| SPL_SCHED(s); |
| error = ipf_frruleiter(softc, data, uid, ctx); |
| SPL_X(s); |
| break; |
| |
| case SIOCGENITER : |
| SPL_SCHED(s); |
| error = ipf_genericiter(softc, data, uid, ctx); |
| SPL_X(s); |
| break; |
| |
| case SIOCIPFDELTOK : |
| error = BCOPYIN(data, &tmp, sizeof(tmp)); |
| if (error == 0) { |
| SPL_SCHED(s); |
| error = ipf_token_del(softc, tmp, uid, ctx); |
| SPL_X(s); |
| } |
| break; |
| |
| default : |
| softc->ipf_interror = 115; |
| error = EINVAL; |
| break; |
| } |
| |
| return error; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_decaps */ |
| /* Returns: int - -1 == decapsulation failed, else bit mask of */ |
| /* flags indicating packet filtering decision. */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* pass(I) - IP protocol version to match */ |
| /* l5proto(I) - layer 5 protocol to decode UDP data as. */ |
| /* */ |
| /* This function is called for packets that are wrapt up in other packets, */ |
| /* for example, an IP packet that is the entire data segment for another IP */ |
| /* packet. If the basic constraints for this are satisfied, change the */ |
| /* buffer to point to the start of the inner packet and start processing */ |
| /* rules belonging to the head group this rule specifies. */ |
| /* ------------------------------------------------------------------------ */ |
| u_32_t |
| ipf_decaps(fin, pass, l5proto) |
| fr_info_t *fin; |
| u_32_t pass; |
| int l5proto; |
| { |
| fr_info_t fin2, *fino = NULL; |
| int elen, hlen, nh; |
| grehdr_t gre; |
| ip_t *ip; |
| mb_t *m; |
| |
| if ((fin->fin_flx & FI_COALESCE) == 0) |
| if (ipf_coalesce(fin) == -1) |
| goto cantdecaps; |
| |
| m = fin->fin_m; |
| hlen = fin->fin_hlen; |
| |
| switch (fin->fin_p) |
| { |
| case IPPROTO_UDP : |
| /* |
| * In this case, the specific protocol being decapsulated |
| * inside UDP frames comes from the rule. |
| */ |
| nh = fin->fin_fr->fr_icode; |
| break; |
| |
| case IPPROTO_GRE : /* 47 */ |
| bcopy(fin->fin_dp, (char *)&gre, sizeof(gre)); |
| hlen += sizeof(grehdr_t); |
| if (gre.gr_R|gre.gr_s) |
| goto cantdecaps; |
| if (gre.gr_C) |
| hlen += 4; |
| if (gre.gr_K) |
| hlen += 4; |
| if (gre.gr_S) |
| hlen += 4; |
| |
| nh = IPPROTO_IP; |
| |
| /* |
| * If the routing options flag is set, validate that it is |
| * there and bounce over it. |
| */ |
| #if 0 |
| /* This is really heavy weight and lots of room for error, */ |
| /* so for now, put it off and get the simple stuff right. */ |
| if (gre.gr_R) { |
| u_char off, len, *s; |
| u_short af; |
| int end; |
| |
| end = 0; |
| s = fin->fin_dp; |
| s += hlen; |
| aplen = fin->fin_plen - hlen; |
| while (aplen > 3) { |
| af = (s[0] << 8) | s[1]; |
| off = s[2]; |
| len = s[3]; |
| aplen -= 4; |
| s += 4; |
| if (af == 0 && len == 0) { |
| end = 1; |
| break; |
| } |
| if (aplen < len) |
| break; |
| s += len; |
| aplen -= len; |
| } |
| if (end != 1) |
| goto cantdecaps; |
| hlen = s - (u_char *)fin->fin_dp; |
| } |
| #endif |
| break; |
| |
| #ifdef IPPROTO_IPIP |
| case IPPROTO_IPIP : /* 4 */ |
| #endif |
| nh = IPPROTO_IP; |
| break; |
| |
| default : /* Includes ESP, AH is special for IPv4 */ |
| goto cantdecaps; |
| } |
| |
| switch (nh) |
| { |
| case IPPROTO_IP : |
| case IPPROTO_IPV6 : |
| break; |
| default : |
| goto cantdecaps; |
| } |
| |
| bcopy((char *)fin, (char *)&fin2, sizeof(fin2)); |
| fino = fin; |
| fin = &fin2; |
| elen = hlen; |
| #if defined(MENTAT) && defined(_KERNEL) |
| m->b_rptr += elen; |
| #else |
| m->m_data += elen; |
| m->m_len -= elen; |
| #endif |
| fin->fin_plen -= elen; |
| fin->fin_ipoff += elen; |
| |
| ip = (ip_t *)((char *)fin->fin_ip + elen); |
| |
| /* |
| * Make sure we have at least enough data for the network layer |
| * header. |
| */ |
| if (IP_V(ip) == 4) |
| hlen = IP_HL(ip) << 2; |
| #ifdef USE_INET6 |
| else if (IP_V(ip) == 6) |
| hlen = sizeof(ip6_t); |
| #endif |
| else |
| goto cantdecaps; |
| |
| if (fin->fin_plen < hlen) |
| goto cantdecaps; |
| |
| fin->fin_dp = (char *)ip + hlen; |
| |
| if (IP_V(ip) == 4) { |
| /* |
| * Perform IPv4 header checksum validation. |
| */ |
| if (ipf_cksum((u_short *)ip, hlen)) |
| goto cantdecaps; |
| } |
| |
| if (ipf_makefrip(hlen, ip, fin) == -1) { |
| cantdecaps: |
| pass &= ~FR_CMDMASK; |
| pass |= FR_BLOCK|FR_QUICK; |
| fin->fin_reason = 8; |
| return -1; |
| } |
| |
| /*fin->fin_fr = *fr->fr_grp;*/ |
| pass = ipf_scanlist(fin, pass); |
| |
| /* |
| * Copy the packet filter "result" fields out of the fr_info_t struct |
| * that is local to the decapsulation processing and back into the |
| * one we were called with. |
| */ |
| fino->fin_flx = fin->fin_flx; |
| fino->fin_rev = fin->fin_rev; |
| fino->fin_icode = fin->fin_icode; |
| fino->fin_rule = fin->fin_rule; |
| (void) strncpy(fino->fin_group, fin->fin_group, FR_GROUPLEN); |
| fino->fin_fr = fin->fin_fr; |
| fino->fin_error = fin->fin_error; |
| fino->fin_mp = fin->fin_mp; |
| fino->fin_m = fin->fin_m; |
| m = fin->fin_m; |
| if (m != NULL) { |
| #if defined(MENTAT) && defined(_KERNEL) |
| m->b_rptr -= elen; |
| #else |
| m->m_data -= elen; |
| m->m_len += elen; |
| #endif |
| } |
| return pass; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_matcharray_load */ |
| /* Returns: int - 0 = success, else error */ |
| /* Parameters: data(I) - pointer to ioctl data */ |
| /* objp(I) - ipfobj_t structure to load data into */ |
| /* arrayptr(I) - pointer to location to store array pointer */ |
| /* */ |
| /* This function loads in a mathing array through the ipfobj_t struct that */ |
| /* describes it. Sanity checking and array size limitations are enforced */ |
| /* in this function to prevent userspace from trying to load in something */ |
| /* that is insanely big. Once the size of the array is known, the memory */ |
| /* required is malloc'd and returned through changing *arrayptr. The */ |
| /* contents of the array are verified before returning. Only in the event */ |
| /* of a successful call is the caller required to free up the malloc area. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_matcharray_load(softc, data, objp, arrayptr) |
| ipf_main_softc_t *softc; |
| caddr_t data; |
| ipfobj_t *objp; |
| int **arrayptr; |
| { |
| int arraysize, *array, error; |
| |
| *arrayptr = NULL; |
| |
| error = BCOPYIN(data, objp, sizeof(*objp)); |
| if (error != 0) { |
| softc->ipf_interror = 116; |
| return EFAULT; |
| } |
| |
| if (objp->ipfo_type != IPFOBJ_IPFEXPR) { |
| softc->ipf_interror = 117; |
| return EINVAL; |
| } |
| |
| if (((objp->ipfo_size & 3) != 0) || (objp->ipfo_size == 0) || |
| (objp->ipfo_size > 1024)) { |
| softc->ipf_interror = 118; |
| return EINVAL; |
| } |
| |
| arraysize = objp->ipfo_size * sizeof(*array); |
| KMALLOCS(array, int *, arraysize); |
| if (array == NULL) { |
| softc->ipf_interror = 119; |
| return ENOMEM; |
| } |
| |
| error = COPYIN(objp->ipfo_ptr, array, arraysize); |
| if (error != 0) { |
| KFREES(array, arraysize); |
| softc->ipf_interror = 120; |
| return EFAULT; |
| } |
| |
| if (ipf_matcharray_verify(array, arraysize) != 0) { |
| KFREES(array, arraysize); |
| softc->ipf_interror = 121; |
| return EINVAL; |
| } |
| |
| *arrayptr = array; |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_matcharray_verify */ |
| /* Returns: Nil */ |
| /* Parameters: array(I) - pointer to matching array */ |
| /* arraysize(I) - number of elements in the array */ |
| /* */ |
| /* Verify the contents of a matching array by stepping through each element */ |
| /* in it. The actual commands in the array are not verified for */ |
| /* correctness, only that all of the sizes are correctly within limits. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_matcharray_verify(array, arraysize) |
| int *array, arraysize; |
| { |
| int i, nelem, maxidx, len; |
| |
| nelem = arraysize / sizeof(*array); |
| |
| /* |
| * Currently, it makes no sense to have an array less than 6 |
| * elements long - the initial size at the from, a single operation |
| * (minimum 4 in length) and a trailer, for a total of 6. |
| */ |
| if ((array[0] < 6) || (arraysize < 24) || (arraysize > 4096)) { |
| return -1; |
| } |
| |
| /* |
| * Verify the size of data pointed to by array with how long |
| * the array claims to be itself. |
| */ |
| if (array[0] * sizeof(*array) != arraysize) { |
| return -1; |
| } |
| |
| maxidx = nelem - 1; |
| /* |
| * The last opcode in this array should be an IPF_EXP_END. |
| */ |
| if (array[maxidx] != IPF_EXP_END) { |
| return -1; |
| } |
| |
| for (i = 1; i < maxidx; ) { |
| len = array[i + 2]; |
| |
| /* |
| * The length of the bits to check must be at least 1 |
| * (or else there is nothing to comapre with!) and it |
| * cannot exceed the length of the data present. |
| */ |
| if ((len < 1) || (i + 3 + len > maxidx)) { |
| return -1; |
| } |
| i += 3 + len; |
| } |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_fr_matcharray */ |
| /* Returns: int - 0 = match failed, else positive match */ |
| /* Parameters: fin(I) - pointer to packet information */ |
| /* array(I) - pointer to matching array */ |
| /* */ |
| /* This function is used to apply a matching array against a packet and */ |
| /* return an indication of whether or not the packet successfully matches */ |
| /* all of the commands in it. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_fr_matcharray(fin, array) |
| fr_info_t *fin; |
| int *array; |
| { |
| int i, n, *x, e, p; |
| |
| e = 0; |
| n = array[0]; |
| x = array + 1; |
| |
| for (; n > 0; x += 3 + x[3], e = 0) { |
| n -= x[3] + 3; |
| |
| /* |
| * The upper 16 bits currently store the protocol value. |
| * This is currently used with TCP and UDP port compares and |
| * allows "tcp.port = 80" without requiring an explicit |
| " "ip.pr = tcp" first. |
| */ |
| p = x[0] >> 16; |
| if ((p != 0) && (p != fin->fin_p)) |
| break; |
| |
| switch (x[0]) |
| { |
| case IPF_EXP_IP_PR : |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= (fin->fin_p == x[i + 3]); |
| } |
| break; |
| |
| case IPF_EXP_IP_SRCADDR : |
| if (fin->fin_v != 4) |
| break; |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= ((fin->fin_saddr & x[i + 4]) == |
| x[i + 3]); |
| } |
| break; |
| |
| case IPF_EXP_IP_DSTADDR : |
| if (fin->fin_v != 4) |
| break; |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= ((fin->fin_daddr & x[i + 4]) == |
| x[i + 3]); |
| } |
| break; |
| |
| case IPF_EXP_IP_ADDR : |
| if (fin->fin_v != 4) |
| break; |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= ((fin->fin_saddr & x[i + 4]) == |
| x[i + 3]) || |
| ((fin->fin_daddr & x[i + 4]) == |
| x[i + 3]); |
| } |
| break; |
| |
| #ifdef USE_INET6 |
| case IPF_EXP_IP6_SRCADDR : |
| if (fin->fin_v != 6) |
| break; |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= IP6_MASKEQ(&fin->fin_src6, x + i + 7, |
| x + i + 3); |
| } |
| break; |
| |
| case IPF_EXP_IP6_DSTADDR : |
| if (fin->fin_v != 6) |
| break; |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= IP6_MASKEQ(&fin->fin_dst6, x + i + 7, |
| x + i + 3); |
| } |
| break; |
| |
| case IPF_EXP_IP6_ADDR : |
| if (fin->fin_v != 6) |
| break; |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= IP6_MASKEQ(&fin->fin_src6, x + i + 7, |
| x + i + 3) || |
| IP6_MASKEQ(&fin->fin_dst6, x + i + 7, |
| x + i + 3); |
| } |
| break; |
| #endif |
| |
| case IPF_EXP_UDP_PORT : |
| case IPF_EXP_TCP_PORT : |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= (fin->fin_sport == x[i + 3]) || |
| (fin->fin_dport == x[i + 3]); |
| } |
| break; |
| |
| case IPF_EXP_UDP_SPORT : |
| case IPF_EXP_TCP_SPORT : |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= (fin->fin_sport == x[i + 3]); |
| } |
| break; |
| |
| case IPF_EXP_UDP_DPORT : |
| case IPF_EXP_TCP_DPORT : |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= (fin->fin_dport == x[i + 3]); |
| } |
| break; |
| |
| case IPF_EXP_TCP_FLAGS : |
| for (i = 0; !e && i < x[3]; i++) { |
| e |= ((fin->fin_tcpf & x[i + 4]) == x[i + 3]); |
| } |
| break; |
| } |
| e ^= x[1]; |
| |
| if (!e) |
| break; |
| } |
| |
| return e; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_queueflush */ |
| /* Returns: int - number of entries flushed (0 = none) */ |
| /* Parameters: deletefn(I) - function to call to delete entry */ |
| /* ipfqs(I) - top of the list of ipf internal queues */ |
| /* userqs(I) - top of the list of user defined timeouts */ |
| /* */ |
| /* This fucntion gets called when the state/NAT hash tables fill up and we */ |
| /* need to try a bit harder to free up some space. The algorithm used here */ |
| /* split into two parts but both halves have the same goal: to reduce the */ |
| /* number of connections considered to be "active" to the low watermark. */ |
| /* There are two steps in doing this: */ |
| /* 1) Remove any TCP connections that are already considered to be "closed" */ |
| /* but have not yet been removed from the state table. The two states */ |
| /* TCPS_TIME_WAIT and TCPS_CLOSED are considered to be the perfect */ |
| /* candidates for this style of removal. If freeing up entries in */ |
| /* CLOSED or both CLOSED and TIME_WAIT brings us to the low watermark, */ |
| /* we do not go on to step 2. */ |
| /* */ |
| /* 2) Look for the oldest entries on each timeout queue and free them if */ |
| /* they are within the given window we are considering. Where the */ |
| /* window starts and the steps taken to increase its size depend upon */ |
| /* how long ipf has been running (ipf_ticks.) Anything modified in the */ |
| /* last 30 seconds is not touched. */ |
| /* touched */ |
| /* die ipf_ticks 30*1.5 1800*1.5 | 43200*1.5 */ |
| /* | | | | | | */ |
| /* future <--+----------+--------+-----------+-----+-----+-----------> past */ |
| /* now \_int=30s_/ \_int=1hr_/ \_int=12hr */ |
| /* */ |
| /* Points to note: */ |
| /* - tqe_die is the time, in the future, when entries die. */ |
| /* - tqe_die - ipf_ticks is how long left the connection has to live in ipf */ |
| /* ticks. */ |
| /* - tqe_touched is when the entry was last used by NAT/state */ |
| /* - the closer tqe_touched is to ipf_ticks, the further tqe_die will be */ |
| /* ipf_ticks any given timeout queue and vice versa. */ |
| /* - both tqe_die and tqe_touched increase over time */ |
| /* - timeout queues are sorted with the highest value of tqe_die at the */ |
| /* bottom and therefore the smallest values of each are at the top */ |
| /* - the pointer passed in as ipfqs should point to an array of timeout */ |
| /* queues representing each of the TCP states */ |
| /* */ |
| /* We start by setting up a maximum range to scan for things to move of */ |
| /* iend (newest) to istart (oldest) in chunks of "interval". If nothing is */ |
| /* found in that range, "interval" is adjusted (so long as it isn't 30) and */ |
| /* we start again with a new value for "iend" and "istart". This is */ |
| /* continued until we either finish the scan of 30 second intervals or the */ |
| /* low water mark is reached. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_queueflush(softc, deletefn, ipfqs, userqs, activep, size, low) |
| ipf_main_softc_t *softc; |
| ipftq_delete_fn_t deletefn; |
| ipftq_t *ipfqs, *userqs; |
| u_int *activep; |
| int size, low; |
| { |
| u_long interval, istart, iend; |
| ipftq_t *ifq, *ifqnext; |
| ipftqent_t *tqe, *tqn; |
| int removed = 0; |
| |
| for (tqn = ipfqs[IPF_TCPS_CLOSED].ifq_head; ((tqe = tqn) != NULL); ) { |
| tqn = tqe->tqe_next; |
| if ((*deletefn)(softc, tqe->tqe_parent) == 0) |
| removed++; |
| } |
| if ((*activep * 100 / size) > low) { |
| for (tqn = ipfqs[IPF_TCPS_TIME_WAIT].ifq_head; |
| ((tqe = tqn) != NULL); ) { |
| tqn = tqe->tqe_next; |
| if ((*deletefn)(softc, tqe->tqe_parent) == 0) |
| removed++; |
| } |
| } |
| |
| if ((*activep * 100 / size) <= low) { |
| return removed; |
| } |
| |
| /* |
| * NOTE: Use of "* 15 / 10" is required here because if "* 1.5" is |
| * used then the operations are upgraded to floating point |
| * and kernels don't like floating point... |
| */ |
| if (softc->ipf_ticks > IPF_TTLVAL(43200 * 15 / 10)) { |
| istart = IPF_TTLVAL(86400 * 4); |
| interval = IPF_TTLVAL(43200); |
| } else if (softc->ipf_ticks > IPF_TTLVAL(1800 * 15 / 10)) { |
| istart = IPF_TTLVAL(43200); |
| interval = IPF_TTLVAL(1800); |
| } else if (softc->ipf_ticks > IPF_TTLVAL(30 * 15 / 10)) { |
| istart = IPF_TTLVAL(1800); |
| interval = IPF_TTLVAL(30); |
| } else { |
| return 0; |
| } |
| if (istart > softc->ipf_ticks) { |
| if (softc->ipf_ticks - interval < interval) |
| istart = interval; |
| else |
| istart = (softc->ipf_ticks / interval) * interval; |
| } |
| |
| iend = softc->ipf_ticks - interval; |
| |
| while ((*activep * 100 / size) > low) { |
| u_long try; |
| |
| try = softc->ipf_ticks - istart; |
| |
| for (ifq = ipfqs; ifq != NULL; ifq = ifq->ifq_next) { |
| for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { |
| if (try < tqe->tqe_touched) |
| break; |
| tqn = tqe->tqe_next; |
| if ((*deletefn)(softc, tqe->tqe_parent) == 0) |
| removed++; |
| } |
| } |
| |
| for (ifq = userqs; ifq != NULL; ifq = ifqnext) { |
| ifqnext = ifq->ifq_next; |
| |
| for (tqn = ifq->ifq_head; ((tqe = tqn) != NULL); ) { |
| if (try < tqe->tqe_touched) |
| break; |
| tqn = tqe->tqe_next; |
| if ((*deletefn)(softc, tqe->tqe_parent) == 0) |
| removed++; |
| } |
| } |
| |
| if (try >= iend) { |
| if (interval == IPF_TTLVAL(43200)) { |
| interval = IPF_TTLVAL(1800); |
| } else if (interval == IPF_TTLVAL(1800)) { |
| interval = IPF_TTLVAL(30); |
| } else { |
| break; |
| } |
| if (interval >= softc->ipf_ticks) |
| break; |
| |
| iend = softc->ipf_ticks - interval; |
| } |
| istart -= interval; |
| } |
| |
| return removed; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_deliverlocal */ |
| /* Returns: int - 1 = local address, 0 = non-local address */ |
| /* Parameters: ipversion(I) - IP protocol version (4 or 6) */ |
| /* ifp(I) - network interface pointer */ |
| /* ipaddr(I) - IPv4/6 destination address */ |
| /* */ |
| /* This fucntion is used to determine in the address "ipaddr" belongs to */ |
| /* the network interface represented by ifp. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_deliverlocal(softc, ipversion, ifp, ipaddr) |
| ipf_main_softc_t *softc; |
| int ipversion; |
| void *ifp; |
| i6addr_t *ipaddr; |
| { |
| i6addr_t addr; |
| int islocal = 0; |
| |
| if (ipversion == 4) { |
| if (ipf_ifpaddr(softc, 4, FRI_NORMAL, ifp, &addr, NULL) == 0) { |
| if (addr.in4.s_addr == ipaddr->in4.s_addr) |
| islocal = 1; |
| } |
| |
| #ifdef USE_INET6 |
| } else if (ipversion == 6) { |
| if (ipf_ifpaddr(softc, 6, FRI_NORMAL, ifp, &addr, NULL) == 0) { |
| if (IP6_EQ(&addr, ipaddr)) |
| islocal = 1; |
| } |
| #endif |
| } |
| |
| return islocal; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_settimeout */ |
| /* Returns: int - 0 = success, -1 = failure */ |
| /* Parameters: t(I) - pointer to tuneable array entry */ |
| /* p(I) - pointer to values passed in to apply */ |
| /* */ |
| /* This function is called to set the timeout values for each distinct */ |
| /* queue timeout that is available. When called, it calls into both the */ |
| /* state and NAT code, telling them to update their timeout queues. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_settimeout(softc, t, p) |
| struct ipf_main_softc_s *softc; |
| ipftuneable_t *t; |
| ipftuneval_t *p; |
| { |
| |
| /* |
| * ipf_interror should be set by the functions called here, not |
| * by this function - it's just a middle man. |
| */ |
| if (ipf_state_settimeout(softc, t, p) == -1) |
| return -1; |
| if (ipf_nat_settimeout(softc, t, p) == -1) |
| return -1; |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_apply_timeout */ |
| /* Returns: int - 0 = success, -1 = failure */ |
| /* Parameters: head(I) - pointer to tuneable array entry */ |
| /* seconds(I) - pointer to values passed in to apply */ |
| /* */ |
| /* This function applies a timeout of "seconds" to the timeout queue that */ |
| /* is pointed to by "head". All entries on this list have an expiration */ |
| /* set to be the current tick value of ipf plus the ttl. Given that this */ |
| /* function should only be called when the delta is non-zero, the task is */ |
| /* to walk the entire list and apply the change. The sort order will not */ |
| /* change. The only catch is that this is O(n) across the list, so if the */ |
| /* queue has lots of entries (10s of thousands or 100s of thousands), it */ |
| /* could take a relatively long time to work through them all. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_apply_timeout(head, seconds) |
| ipftq_t *head; |
| u_int seconds; |
| { |
| u_int oldtimeout, newtimeout; |
| ipftqent_t *tqe; |
| int delta; |
| |
| MUTEX_ENTER(&head->ifq_lock); |
| oldtimeout = head->ifq_ttl; |
| newtimeout = IPF_TTLVAL(seconds); |
| delta = oldtimeout - newtimeout; |
| |
| head->ifq_ttl = newtimeout; |
| |
| for (tqe = head->ifq_head; tqe != NULL; tqe = tqe->tqe_next) { |
| tqe->tqe_die += delta; |
| } |
| MUTEX_EXIT(&head->ifq_lock); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_settimeout_tcp */ |
| /* Returns: int - 0 = successfully applied, -1 = failed */ |
| /* Parameters: t(I) - pointer to tuneable to change */ |
| /* p(I) - pointer to new timeout information */ |
| /* tab(I) - pointer to table of TCP queues */ |
| /* */ |
| /* This function applies the new timeout (p) to the TCP tunable (t) and */ |
| /* updates all of the entries on the relevant timeout queue by calling */ |
| /* ipf_apply_timeout(). */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_settimeout_tcp(t, p, tab) |
| ipftuneable_t *t; |
| ipftuneval_t *p; |
| ipftq_t *tab; |
| { |
| if (!strcmp(t->ipft_name, "tcp_idle_timeout") || |
| !strcmp(t->ipft_name, "tcp_established")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_ESTABLISHED], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_close_wait")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_CLOSE_WAIT], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_last_ack")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_LAST_ACK], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_timeout")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_LISTEN], p->ipftu_int); |
| ipf_apply_timeout(&tab[IPF_TCPS_HALF_ESTAB], p->ipftu_int); |
| ipf_apply_timeout(&tab[IPF_TCPS_CLOSING], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_listen")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_LISTEN], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_half_established")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_HALF_ESTAB], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_closing")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_CLOSING], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_syn_received")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_SYN_RECEIVED], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_syn_sent")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_SYN_SENT], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_closed")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_CLOSED], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_half_closed")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_CLOSED], p->ipftu_int); |
| } else if (!strcmp(t->ipft_name, "tcp_time_wait")) { |
| ipf_apply_timeout(&tab[IPF_TCPS_TIME_WAIT], p->ipftu_int); |
| } else { |
| /* |
| * ipf_interror isn't set here because it should be set |
| * by whatever called this function. |
| */ |
| return -1; |
| } |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_main_soft_create */ |
| /* Returns: NULL = failure, else success */ |
| /* Parameters: arg(I) - pointer to soft context structure if already allocd */ |
| /* */ |
| /* Create the foundation soft context structure. In circumstances where it */ |
| /* is not required to dynamically allocate the context, a pointer can be */ |
| /* passed in (rather than NULL) to a structure to be initialised. */ |
| /* The main thing of interest* is that a number of locks are initialised */ |
| /* here instead of in the where might be expected - in the relevant create */ |
| /* function elsewhere. This is done because the current locking design has */ |
| /* some areas where these locks are used outside of their module. */ |
| /* Possibly the most important exercise that is done here is setting of all */ |
| /* the timeout values, allowing them to be changed before init(). */ |
| /* ------------------------------------------------------------------------ */ |
| void * |
| ipf_main_soft_create(arg) |
| void *arg; |
| { |
| ipf_main_softc_t *softc; |
| |
| if (arg == NULL) { |
| KMALLOC(softc, ipf_main_softc_t *); |
| if (softc == NULL) |
| return NULL; |
| } else { |
| softc = arg; |
| } |
| |
| bzero((char *)softc, sizeof(*softc)); |
| |
| /* |
| * This serves as a flag as to whether or not the softc should be |
| * free'd when _destroy is called. |
| */ |
| softc->ipf_dynamic_softc = (arg == NULL) ? 1 : 0; |
| |
| softc->ipf_tuners = ipf_tune_array_copy(softc, |
| sizeof(ipf_main_tuneables), |
| ipf_main_tuneables); |
| if (softc->ipf_tuners == NULL) { |
| ipf_main_soft_destroy(softc, NULL); |
| return NULL; |
| } |
| |
| MUTEX_INIT(&softc->ipf_rw, "ipf rw mutex"); |
| MUTEX_INIT(&softc->ipf_timeoutlock, "ipf timeout lock"); |
| RWLOCK_INIT(&softc->ipf_global, "ipf filter load/unload mutex"); |
| RWLOCK_INIT(&softc->ipf_mutex, "ipf filter rwlock"); |
| RWLOCK_INIT(&softc->ipf_tokens, "ipf token rwlock"); |
| RWLOCK_INIT(&softc->ipf_state, "ipf state rwlock"); |
| RWLOCK_INIT(&softc->ipf_nat, "ipf IP NAT rwlock"); |
| RWLOCK_INIT(&softc->ipf_poolrw, "ipf pool rwlock"); |
| RWLOCK_INIT(&softc->ipf_frag, "ipf frag rwlock"); |
| |
| softc->ipf_token_head = NULL; |
| softc->ipf_token_tail = &softc->ipf_token_head; |
| |
| softc->ipf_tcpidletimeout = FIVE_DAYS; |
| softc->ipf_tcpclosewait = IPF_TTLVAL(2 * TCP_MSL); |
| softc->ipf_tcplastack = IPF_TTLVAL(30); |
| softc->ipf_tcptimewait = IPF_TTLVAL(2 * TCP_MSL); |
| softc->ipf_tcptimeout = IPF_TTLVAL(2 * TCP_MSL); |
| softc->ipf_tcpsynsent = IPF_TTLVAL(2 * TCP_MSL); |
| softc->ipf_tcpsynrecv = IPF_TTLVAL(2 * TCP_MSL); |
| softc->ipf_tcpclosed = IPF_TTLVAL(30); |
| softc->ipf_tcphalfclosed = IPF_TTLVAL(2 * 3600); |
| softc->ipf_udptimeout = IPF_TTLVAL(120); |
| softc->ipf_udpacktimeout = IPF_TTLVAL(12); |
| softc->ipf_icmptimeout = IPF_TTLVAL(60); |
| softc->ipf_icmpacktimeout = IPF_TTLVAL(6); |
| softc->ipf_iptimeout = IPF_TTLVAL(60); |
| |
| #if defined(IPFILTER_DEFAULT_BLOCK) |
| softc->ipf_pass = FR_BLOCK|FR_NOMATCH; |
| #else |
| softc->ipf_pass = (IPF_DEFAULT_PASS)|FR_NOMATCH; |
| #endif |
| softc->ipf_minttl = 4; |
| softc->ipf_icmpminfragmtu = 68; |
| softc->ipf_flags = IPF_LOGGING; |
| |
| return softc; |
| } |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_main_soft_init */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* */ |
| /* A null-op function that exists as a placeholder so that the flow in */ |
| /* other functions is obvious. */ |
| /* ------------------------------------------------------------------------ */ |
| /*ARGSUSED*/ |
| int |
| ipf_main_soft_init(softc) |
| ipf_main_softc_t *softc; |
| { |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_main_soft_destroy */ |
| /* Returns: void */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* arg(I) - not used (present for symmetry.) */ |
| /* */ |
| /* Undo everything that we did in ipf_main_soft_create. */ |
| /* */ |
| /* The most important check that needs to be made here is whether or not */ |
| /* the structure was allocated by ipf_main_soft_create() by checking what */ |
| /* value is stored in ipf_dynamic_main. */ |
| /* ------------------------------------------------------------------------ */ |
| /*ARGSUSED*/ |
| void |
| ipf_main_soft_destroy(softc, arg) |
| ipf_main_softc_t *softc; |
| void *arg; |
| { |
| |
| RW_DESTROY(&softc->ipf_frag); |
| RW_DESTROY(&softc->ipf_poolrw); |
| RW_DESTROY(&softc->ipf_nat); |
| RW_DESTROY(&softc->ipf_state); |
| RW_DESTROY(&softc->ipf_tokens); |
| RW_DESTROY(&softc->ipf_mutex); |
| RW_DESTROY(&softc->ipf_global); |
| MUTEX_DESTROY(&softc->ipf_timeoutlock); |
| MUTEX_DESTROY(&softc->ipf_rw); |
| |
| if (softc->ipf_tuners != NULL) { |
| KFREES(softc->ipf_tuners, sizeof(ipf_main_tuneables)); |
| } |
| if (softc->ipf_dynamic_softc == 1) { |
| KFREE(softc); |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_main_soft_fini */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* */ |
| /* Clean out the rules which have been added since _init was last called, */ |
| /* the only dynamic part of the mainline. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_main_soft_fini(softc) |
| ipf_main_softc_t *softc; |
| { |
| (void) ipf_flush(softc, IPL_LOGIPF, FR_INQUE|FR_OUTQUE|FR_INACTIVE); |
| (void) ipf_flush(softc, IPL_LOGIPF, FR_INQUE|FR_OUTQUE); |
| (void) ipf_flush(softc, IPL_LOGCOUNT, FR_INQUE|FR_OUTQUE|FR_INACTIVE); |
| (void) ipf_flush(softc, IPL_LOGCOUNT, FR_INQUE|FR_OUTQUE); |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_main_load */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: none */ |
| /* */ |
| /* Handle global initialisation that needs to be done for the base part of */ |
| /* IPFilter. At present this just amounts to initialising some ICMP lookup */ |
| /* arrays that get used by the state/NAT code. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_main_load() |
| { |
| int i; |
| |
| /* fill icmp reply type table */ |
| for (i = 0; i <= ICMP_MAXTYPE; i++) |
| icmpreplytype4[i] = -1; |
| icmpreplytype4[ICMP_ECHO] = ICMP_ECHOREPLY; |
| icmpreplytype4[ICMP_TSTAMP] = ICMP_TSTAMPREPLY; |
| icmpreplytype4[ICMP_IREQ] = ICMP_IREQREPLY; |
| icmpreplytype4[ICMP_MASKREQ] = ICMP_MASKREPLY; |
| |
| #ifdef USE_INET6 |
| /* fill icmp reply type table */ |
| for (i = 0; i <= ICMP6_MAXTYPE; i++) |
| icmpreplytype6[i] = -1; |
| icmpreplytype6[ICMP6_ECHO_REQUEST] = ICMP6_ECHO_REPLY; |
| icmpreplytype6[ICMP6_MEMBERSHIP_QUERY] = ICMP6_MEMBERSHIP_REPORT; |
| icmpreplytype6[ICMP6_NI_QUERY] = ICMP6_NI_REPLY; |
| icmpreplytype6[ND_ROUTER_SOLICIT] = ND_ROUTER_ADVERT; |
| icmpreplytype6[ND_NEIGHBOR_SOLICIT] = ND_NEIGHBOR_ADVERT; |
| #endif |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_main_unload */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: none */ |
| /* */ |
| /* A null-op function that exists as a placeholder so that the flow in */ |
| /* other functions is obvious. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_main_unload() |
| { |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_load_all */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: none */ |
| /* */ |
| /* Work through all of the subsystems inside IPFilter and call the load */ |
| /* function for each in an order that won't lead to a crash :) */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_load_all() |
| { |
| if (ipf_main_load() == -1) |
| return -1; |
| |
| if (ipf_state_main_load() == -1) |
| return -1; |
| |
| if (ipf_nat_main_load() == -1) |
| return -1; |
| |
| if (ipf_frag_main_load() == -1) |
| return -1; |
| |
| if (ipf_auth_main_load() == -1) |
| return -1; |
| |
| if (ipf_proxy_main_load() == -1) |
| return -1; |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_unload_all */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: none */ |
| /* */ |
| /* Work through all of the subsystems inside IPFilter and call the unload */ |
| /* function for each in an order that won't lead to a crash :) */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_unload_all() |
| { |
| if (ipf_proxy_main_unload() == -1) |
| return -1; |
| |
| if (ipf_auth_main_unload() == -1) |
| return -1; |
| |
| if (ipf_frag_main_unload() == -1) |
| return -1; |
| |
| if (ipf_nat_main_unload() == -1) |
| return -1; |
| |
| if (ipf_state_main_unload() == -1) |
| return -1; |
| |
| if (ipf_main_unload() == -1) |
| return -1; |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_create_all */ |
| /* Returns: NULL = failure, else success */ |
| /* Parameters: arg(I) - pointer to soft context main structure */ |
| /* */ |
| /* Work through all of the subsystems inside IPFilter and call the create */ |
| /* function for each in an order that won't lead to a crash :) */ |
| /* ------------------------------------------------------------------------ */ |
| ipf_main_softc_t * |
| ipf_create_all(arg) |
| void *arg; |
| { |
| ipf_main_softc_t *softc; |
| |
| softc = ipf_main_soft_create(arg); |
| if (softc == NULL) |
| return NULL; |
| |
| softc->ipf_log_soft = ipf_log_soft_create(softc); |
| if (softc->ipf_log_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| softc->ipf_lookup_soft = ipf_lookup_soft_create(softc); |
| if (softc->ipf_lookup_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| softc->ipf_sync_soft = ipf_sync_soft_create(softc); |
| if (softc->ipf_sync_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| softc->ipf_state_soft = ipf_state_soft_create(softc); |
| if (softc->ipf_state_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| softc->ipf_nat_soft = ipf_nat_soft_create(softc); |
| if (softc->ipf_nat_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| softc->ipf_frag_soft = ipf_frag_soft_create(softc); |
| if (softc->ipf_frag_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| softc->ipf_auth_soft = ipf_auth_soft_create(softc); |
| if (softc->ipf_auth_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| softc->ipf_proxy_soft = ipf_proxy_soft_create(softc); |
| if (softc->ipf_proxy_soft == NULL) { |
| ipf_destroy_all(softc); |
| return NULL; |
| } |
| |
| return softc; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_destroy_all */ |
| /* Returns: void */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* */ |
| /* Work through all of the subsystems inside IPFilter and call the destroy */ |
| /* function for each in an order that won't lead to a crash :) */ |
| /* */ |
| /* Every one of these functions is expected to succeed, so there is no */ |
| /* checking of return values. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_destroy_all(softc) |
| ipf_main_softc_t *softc; |
| { |
| |
| if (softc->ipf_state_soft != NULL) { |
| ipf_state_soft_destroy(softc, softc->ipf_state_soft); |
| softc->ipf_state_soft = NULL; |
| } |
| |
| if (softc->ipf_nat_soft != NULL) { |
| ipf_nat_soft_destroy(softc, softc->ipf_nat_soft); |
| softc->ipf_nat_soft = NULL; |
| } |
| |
| if (softc->ipf_frag_soft != NULL) { |
| ipf_frag_soft_destroy(softc, softc->ipf_frag_soft); |
| softc->ipf_frag_soft = NULL; |
| } |
| |
| if (softc->ipf_auth_soft != NULL) { |
| ipf_auth_soft_destroy(softc, softc->ipf_auth_soft); |
| softc->ipf_auth_soft = NULL; |
| } |
| |
| if (softc->ipf_proxy_soft != NULL) { |
| ipf_proxy_soft_destroy(softc, softc->ipf_proxy_soft); |
| softc->ipf_proxy_soft = NULL; |
| } |
| |
| if (softc->ipf_sync_soft != NULL) { |
| ipf_sync_soft_destroy(softc, softc->ipf_sync_soft); |
| softc->ipf_sync_soft = NULL; |
| } |
| |
| if (softc->ipf_lookup_soft != NULL) { |
| ipf_lookup_soft_destroy(softc, softc->ipf_lookup_soft); |
| softc->ipf_lookup_soft = NULL; |
| } |
| |
| if (softc->ipf_log_soft != NULL) { |
| ipf_log_soft_destroy(softc, softc->ipf_log_soft); |
| softc->ipf_log_soft = NULL; |
| } |
| |
| ipf_main_soft_destroy(softc, NULL); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_init_all */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* */ |
| /* Work through all of the subsystems inside IPFilter and call the init */ |
| /* function for each in an order that won't lead to a crash :) */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_init_all(softc) |
| ipf_main_softc_t *softc; |
| { |
| |
| if (ipf_main_soft_init(softc) == -1) |
| return -1; |
| |
| if (ipf_log_soft_init(softc, softc->ipf_log_soft) == -1) |
| return -1; |
| |
| if (ipf_lookup_soft_init(softc, softc->ipf_lookup_soft) == -1) |
| return -1; |
| |
| if (ipf_sync_soft_init(softc, softc->ipf_sync_soft) == -1) |
| return -1; |
| |
| if (ipf_state_soft_init(softc, softc->ipf_state_soft) == -1) |
| return -1; |
| |
| if (ipf_nat_soft_init(softc, softc->ipf_nat_soft) == -1) |
| return -1; |
| |
| if (ipf_frag_soft_init(softc, softc->ipf_frag_soft) == -1) |
| return -1; |
| |
| if (ipf_auth_soft_init(softc, softc->ipf_auth_soft) == -1) |
| return -1; |
| |
| if (ipf_proxy_soft_init(softc, softc->ipf_proxy_soft) == -1) |
| return -1; |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_fini_all */ |
| /* Returns: 0 = success, -1 = failure */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* */ |
| /* Work through all of the subsystems inside IPFilter and call the fini */ |
| /* function for each in an order that won't lead to a crash :) */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_fini_all(softc) |
| ipf_main_softc_t *softc; |
| { |
| |
| if (ipf_proxy_soft_fini(softc, softc->ipf_proxy_soft) == -1) |
| return -1; |
| |
| if (ipf_auth_soft_fini(softc, softc->ipf_auth_soft) == -1) |
| return -1; |
| |
| if (ipf_frag_soft_fini(softc, softc->ipf_frag_soft) == -1) |
| return -1; |
| |
| if (ipf_nat_soft_fini(softc, softc->ipf_nat_soft) == -1) |
| return -1; |
| |
| if (ipf_state_soft_fini(softc, softc->ipf_state_soft) == -1) |
| return -1; |
| |
| if (ipf_sync_soft_fini(softc, softc->ipf_sync_soft) == -1) |
| return -1; |
| |
| if (ipf_lookup_soft_fini(softc, softc->ipf_lookup_soft) == -1) |
| return -1; |
| |
| if (ipf_log_soft_fini(softc, softc->ipf_log_soft) == -1) |
| return -1; |
| |
| if (ipf_main_soft_fini(softc) == -1) |
| return -1; |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_rule_expire */ |
| /* Returns: Nil */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* */ |
| /* At present this function exists just to support temporary addition of */ |
| /* firewall rules. Both inactive and active lists are scanned for items to */ |
| /* purge, as by rights, the expiration is computed as soon as the rule is */ |
| /* loaded in. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_rule_expire(softc) |
| ipf_main_softc_t *softc; |
| { |
| frentry_t *fr; |
| |
| if ((softc->ipf_rule_explist[0] == NULL) && |
| (softc->ipf_rule_explist[1] == NULL)) |
| return; |
| |
| WRITE_ENTER(&softc->ipf_mutex); |
| |
| while ((fr = softc->ipf_rule_explist[0]) != NULL) { |
| /* |
| * Because the list is kept sorted on insertion, the fist |
| * one that dies in the future means no more work to do. |
| */ |
| if (fr->fr_die > softc->ipf_ticks) |
| break; |
| ipf_rule_delete(softc, fr, IPL_LOGIPF, 0); |
| } |
| |
| while ((fr = softc->ipf_rule_explist[1]) != NULL) { |
| /* |
| * Because the list is kept sorted on insertion, the fist |
| * one that dies in the future means no more work to do. |
| */ |
| if (fr->fr_die > softc->ipf_ticks) |
| break; |
| ipf_rule_delete(softc, fr, IPL_LOGIPF, 1); |
| } |
| |
| RWLOCK_EXIT(&softc->ipf_mutex); |
| } |
| |
| |
| /* |
| * NOTE: |
| * |
| * As there is currently only 1 use of a red-black tree in IPFilter, I'm |
| * happy to use the widespread BSD interfaces in <sys/tree.h>. Was there |
| * more than one use, I'd be using a .c file implementation fof a red- |
| * -black tree. |
| */ |
| static int ipf_ht_node_cmp __P((struct host_node_s *, struct host_node_s *)); |
| static void ipf_ht_node_make_key __P((host_track_t *, host_node_t *, int, |
| i6addr_t *)); |
| |
| RB_PROTOTYPE(ipf_rb, host_node_s, hn_entry, ipf_ht_node_cmp); |
| RB_GENERATE(ipf_rb, host_node_s, hn_entry, ipf_ht_node_cmp); |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ht_node_cmp */ |
| /* Returns: int - 0 == nodes are the same, .. */ |
| /* Parameters: k1(I) - pointer to first key to compare */ |
| /* k2(I) - pointer to second key to compare */ |
| /* */ |
| /* The "key" for the node is a combination of two fields: the address */ |
| /* family and the address itself. */ |
| /* */ |
| /* Because we're not actually interpreting the address data, it isn't */ |
| /* necessary to convert them to/from network/host byte order. The mask is */ |
| /* just used to remove bits that aren't significant - it doesn't matter */ |
| /* where they are, as long as they're always in the same place. */ |
| /* */ |
| /* As with IP6_EQ, comparing IPv6 addresses starts at the bottom because */ |
| /* this is where individual ones will differ the most - but not true for */ |
| /* for /48's, etc. */ |
| /* ------------------------------------------------------------------------ */ |
| static int |
| ipf_ht_node_cmp(k1, k2) |
| struct host_node_s *k1, *k2; |
| { |
| int i; |
| |
| i = (k2->hn_family - k1->hn_family); |
| if (i != 0) |
| return i; |
| |
| if (k1->hn_family == AF_INET) |
| return (k2->hn_addr.in4.s_addr - k1->hn_addr.in4.s_addr); |
| |
| i = k2->hn_addr.i6[3] - k1->hn_addr.i6[3]; |
| if (i != 0) |
| return i; |
| i = k2->hn_addr.i6[2] - k1->hn_addr.i6[2]; |
| if (i != 0) |
| return i; |
| i = k2->hn_addr.i6[1] - k1->hn_addr.i6[1]; |
| if (i != 0) |
| return i; |
| i = k2->hn_addr.i6[0] - k1->hn_addr.i6[0]; |
| return i; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ht_node_make_key */ |
| /* Returns: Nil */ |
| /* parameters: htp(I) - pointer to address tracking structure */ |
| /* key(I) - where to store masked address for lookup */ |
| /* family(I) - protocol family of address */ |
| /* addr(I) - pointer to network address */ |
| /* */ |
| /* Using the "netmask" (number of bits) stored parent host tracking struct, */ |
| /* copy the address passed in into the key structure whilst masking out the */ |
| /* bits that we don't want. */ |
| /* */ |
| /* Because the parser will set ht_netmask to 128 if there is no protocol */ |
| /* specified (the parser doesn't know if it should be a v4 or v6 rule), we */ |
| /* have to be wary of that and not allow 32-128 to happen. */ |
| /* ------------------------------------------------------------------------ */ |
| static void |
| ipf_ht_node_make_key(htp, key, family, addr) |
| host_track_t *htp; |
| host_node_t *key; |
| int family; |
| i6addr_t *addr; |
| { |
| key->hn_family = family; |
| if (family == AF_INET) { |
| u_32_t mask; |
| int bits; |
| |
| bits = htp->ht_netmask; |
| if (bits >= 32) { |
| mask = 0xffffffff; |
| } else { |
| mask = htonl(0xffffffff << (32 - bits)); |
| } |
| key->hn_addr.in4.s_addr = addr->in4.s_addr & mask; |
| } else { |
| int bits = htp->ht_netmask; |
| |
| if (bits > 96) { |
| key->hn_addr.i6[3] = addr->i6[3] & |
| htonl(0xffffffff << (128 - bits)); |
| key->hn_addr.i6[2] = addr->i6[2]; |
| key->hn_addr.i6[1] = addr->i6[2]; |
| key->hn_addr.i6[0] = addr->i6[2]; |
| } else if (bits > 64) { |
| key->hn_addr.i6[3] = 0; |
| key->hn_addr.i6[2] = addr->i6[2] & |
| htonl(0xffffffff << (96 - bits)); |
| key->hn_addr.i6[1] = addr->i6[1]; |
| key->hn_addr.i6[0] = addr->i6[0]; |
| } else if (bits > 32) { |
| key->hn_addr.i6[3] = 0; |
| key->hn_addr.i6[2] = 0; |
| key->hn_addr.i6[1] = addr->i6[1] & |
| htonl(0xffffffff << (64 - bits)); |
| key->hn_addr.i6[0] = addr->i6[0]; |
| } else { |
| key->hn_addr.i6[3] = 0; |
| key->hn_addr.i6[2] = 0; |
| key->hn_addr.i6[1] = 0; |
| key->hn_addr.i6[0] = addr->i6[0] & |
| htonl(0xffffffff << (32 - bits)); |
| } |
| } |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ht_node_add */ |
| /* Returns: int - 0 == success, -1 == failure */ |
| /* Parameters: softc(I) - pointer to soft context main structure */ |
| /* htp(I) - pointer to address tracking structure */ |
| /* family(I) - protocol family of address */ |
| /* addr(I) - pointer to network address */ |
| /* */ |
| /* NOTE: THIS FUNCTION MUST BE CALLED WITH AN EXCLUSIVE LOCK THAT PREVENTS */ |
| /* ipf_ht_node_del FROM RUNNING CONCURRENTLY ON THE SAME htp. */ |
| /* */ |
| /* After preparing the key with the address information to find, look in */ |
| /* the red-black tree to see if the address is known. A successful call to */ |
| /* this function can mean one of two things: a new node was added to the */ |
| /* tree or a matching node exists and we're able to bump up its activity. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_ht_node_add(softc, htp, family, addr) |
| ipf_main_softc_t *softc; |
| host_track_t *htp; |
| int family; |
| i6addr_t *addr; |
| { |
| host_node_t *h; |
| host_node_t k; |
| |
| ipf_ht_node_make_key(htp, &k, family, addr); |
| |
| h = RB_FIND(ipf_rb, &htp->ht_root, &k); |
| if (h == NULL) { |
| if (htp->ht_cur_nodes >= htp->ht_max_nodes) |
| return -1; |
| KMALLOC(h, host_node_t *); |
| if (h == NULL) { |
| LBUMP(ipf_rb_no_mem); |
| return -1; |
| } |
| |
| /* |
| * If there was a macro to initialise the RB node then that |
| * would get used here, but there isn't... |
| */ |
| bzero((char *)h, sizeof(*h)); |
| h->hn_addr = k.hn_addr; |
| h->hn_family = k.hn_family; |
| RB_INSERT(ipf_rb, &htp->ht_root, h); |
| htp->ht_cur_nodes++; |
| } else { |
| if ((htp->ht_max_per_node != 0) && |
| (h->hn_active >= htp->ht_max_per_node)) { |
| LBUMP(ipf_rb_node_max); |
| return -1; |
| } |
| } |
| |
| h->hn_active++; |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_ht_node_del */ |
| /* Returns: int - 0 == success, -1 == failure */ |
| /* parameters: htp(I) - pointer to address tracking structure */ |
| /* family(I) - protocol family of address */ |
| /* addr(I) - pointer to network address */ |
| /* */ |
| /* NOTE: THIS FUNCTION MUST BE CALLED WITH AN EXCLUSIVE LOCK THAT PREVENTS */ |
| /* ipf_ht_node_add FROM RUNNING CONCURRENTLY ON THE SAME htp. */ |
| /* */ |
| /* Try and find the address passed in amongst the leavese on this tree to */ |
| /* be friend. If found then drop the active account for that node drops by */ |
| /* one. If that count reaches 0, it is time to free it all up. */ |
| /* ------------------------------------------------------------------------ */ |
| int |
| ipf_ht_node_del(htp, family, addr) |
| host_track_t *htp; |
| int family; |
| i6addr_t *addr; |
| { |
| host_node_t *h; |
| host_node_t k; |
| |
| ipf_ht_node_make_key(htp, &k, family, addr); |
| |
| h = RB_FIND(ipf_rb, &htp->ht_root, &k); |
| if (h == NULL) { |
| return -1; |
| } else { |
| h->hn_active--; |
| if (h->hn_active == 0) { |
| RB_REMOVE(ipf_rb, &htp->ht_root, h); |
| htp->ht_cur_nodes--; |
| KFREE(h); |
| } |
| } |
| |
| return 0; |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_rb_ht_init */ |
| /* Returns: Nil */ |
| /* Parameters: head(I) - pointer to host tracking structure */ |
| /* */ |
| /* Initialise the host tracking structure to be ready for use above. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_rb_ht_init(head) |
| host_track_t *head; |
| { |
| RB_INIT(&head->ht_root); |
| } |
| |
| |
| /* ------------------------------------------------------------------------ */ |
| /* Function: ipf_slowtimer */ |
| /* Returns: Nil */ |
| /* Parameters: ptr(I) - pointer to main ipf soft context structure */ |
| /* */ |
| /* Slowly expire held state for fragments. Timeouts are set * in */ |
| /* expectation of this being called twice per second. */ |
| /* ------------------------------------------------------------------------ */ |
| void |
| ipf_slowtimer(softc) |
| ipf_main_softc_t *softc; |
| { |
| |
| ipf_token_expire(softc); |
| ipf_frag_expire(softc); |
| ipf_state_expire(softc); |
| ipf_nat_expire(softc); |
| ipf_auth_expire(softc); |
| ipf_lookup_expire(softc); |
| ipf_rule_expire(softc); |
| ipf_sync_expire(softc); |
| softc->ipf_ticks++; |
| # if defined(__OpenBSD__) |
| timeout_add(&ipf_slowtimer_ch, hz/2); |
| # endif |
| } |